1/* sv.c 2 * 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall 5 * and others 6 * 7 * You may distribute under the terms of either the GNU General Public 8 * License or the Artistic License, as specified in the README file. 9 * 10 */ 11 12/* 13 * 'I wonder what the Entish is for "yes" and "no",' he thought. 14 * --Pippin 15 * 16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"] 17 */ 18 19/* 20 * 21 * 22 * This file contains the code that creates, manipulates and destroys 23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the 24 * structure of an SV, so their creation and destruction is handled 25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode 26 * level functions (eg. substr, split, join) for each of the types are 27 * in the pp*.c files. 28 */ 29 30#include "EXTERN.h" 31#define PERL_IN_SV_C 32#include "perl.h" 33#include "regcomp.h" 34#ifdef __VMS 35# include <rms.h> 36#endif 37 38#ifdef __Lynx__ 39/* Missing proto on LynxOS */ 40 char *gconvert(double, int, int, char *); 41#endif 42 43#ifdef USE_QUADMATH 44# define SNPRINTF_G(nv, buffer, size, ndig) \ 45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv)) 46#else 47# define SNPRINTF_G(nv, buffer, size, ndig) \ 48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer)) 49#endif 50 51#ifndef SV_COW_THRESHOLD 52# define SV_COW_THRESHOLD 0 /* COW iff len > K */ 53#endif 54#ifndef SV_COWBUF_THRESHOLD 55# define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */ 56#endif 57#ifndef SV_COW_MAX_WASTE_THRESHOLD 58# define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */ 59#endif 60#ifndef SV_COWBUF_WASTE_THRESHOLD 61# define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */ 62#endif 63#ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD 64# define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */ 65#endif 66#ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD 67# define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */ 68#endif 69/* Work around compiler warnings about unsigned >= THRESHOLD when thres- 70 hold is 0. */ 71#if SV_COW_THRESHOLD 72# define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD) 73#else 74# define GE_COW_THRESHOLD(cur) 1 75#endif 76#if SV_COWBUF_THRESHOLD 77# define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD) 78#else 79# define GE_COWBUF_THRESHOLD(cur) 1 80#endif 81#if SV_COW_MAX_WASTE_THRESHOLD 82# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD) 83#else 84# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1 85#endif 86#if SV_COWBUF_WASTE_THRESHOLD 87# define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD) 88#else 89# define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1 90#endif 91#if SV_COW_MAX_WASTE_FACTOR_THRESHOLD 92# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur)) 93#else 94# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1 95#endif 96#if SV_COWBUF_WASTE_FACTOR_THRESHOLD 97# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur)) 98#else 99# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1 100#endif 101 102#define CHECK_COW_THRESHOLD(cur,len) (\ 103 GE_COW_THRESHOLD((cur)) && \ 104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \ 105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \ 106) 107#define CHECK_COWBUF_THRESHOLD(cur,len) (\ 108 GE_COWBUF_THRESHOLD((cur)) && \ 109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \ 110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \ 111) 112 113#ifdef PERL_UTF8_CACHE_ASSERT 114/* if adding more checks watch out for the following tests: 115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t 116 * lib/utf8.t lib/Unicode/Collate/t/index.t 117 * --jhi 118 */ 119# define ASSERT_UTF8_CACHE(cache) \ 120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \ 121 assert((cache)[2] <= (cache)[3]); \ 122 assert((cache)[3] <= (cache)[1]);} \ 123 } STMT_END 124#else 125# define ASSERT_UTF8_CACHE(cache) NOOP 126#endif 127 128static const char S_destroy[] = "DESTROY"; 129#define S_destroy_len (sizeof(S_destroy)-1) 130 131/* ============================================================================ 132 133An SV (or AV, HV, etc.) is allocated in two parts: the head (struct 134sv, av, hv...) contains type and reference count information, and for 135many types, a pointer to the body (struct xrv, xpv, xpviv...), which 136contains fields specific to each type. Some types store all they need 137in the head, so don't have a body. 138 139In all but the most memory-paranoid configurations (ex: PURIFY), heads 140and bodies are allocated out of arenas, which by default are 141approximately 4K chunks of memory parcelled up into N heads or bodies. 142Sv-bodies are allocated by their sv-type, guaranteeing size 143consistency needed to allocate safely from arrays. 144 145For SV-heads, the first slot in each arena is reserved, and holds a 146link to the next arena, some flags, and a note of the number of slots. 147Snaked through each arena chain is a linked list of free items; when 148this becomes empty, an extra arena is allocated and divided up into N 149items which are threaded into the free list. 150 151SV-bodies are similar, but they use arena-sets by default, which 152separate the link and info from the arena itself, and reclaim the 1st 153slot in the arena. SV-bodies are further described later. 154 155The following global variables are associated with arenas: 156 157 PL_sv_arenaroot pointer to list of SV arenas 158 PL_sv_root pointer to list of free SV structures 159 160 PL_body_arenas head of linked-list of body arenas 161 PL_body_roots[] array of pointers to list of free bodies of svtype 162 arrays are indexed by the svtype needed 163 164A few special SV heads are not allocated from an arena, but are 165instead directly created in the interpreter structure, eg PL_sv_undef. 166The size of arenas can be changed from the default by setting 167PERL_ARENA_SIZE appropriately at compile time. 168 169The SV arena serves the secondary purpose of allowing still-live SVs 170to be located and destroyed during final cleanup. 171 172At the lowest level, the macros new_SV() and del_SV() grab and free 173an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv() 174to return the SV to the free list with error checking.) new_SV() calls 175more_sv() / sv_add_arena() to add an extra arena if the free list is empty. 176SVs in the free list have their SvTYPE field set to all ones. 177 178At the time of very final cleanup, sv_free_arenas() is called from 179perl_destruct() to physically free all the arenas allocated since the 180start of the interpreter. 181 182The internal function visit() scans the SV arenas list, and calls a specified 183function for each SV it finds which is still live, I<i.e.> which has an SvTYPE 184other than all 1's, and a non-zero SvREFCNT. visit() is used by the 185following functions (specified as [function that calls visit()] / [function 186called by visit() for each SV]): 187 188 sv_report_used() / do_report_used() 189 dump all remaining SVs (debugging aid) 190 191 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(), 192 do_clean_named_io_objs(),do_curse() 193 Attempt to free all objects pointed to by RVs, 194 try to do the same for all objects indir- 195 ectly referenced by typeglobs too, and 196 then do a final sweep, cursing any 197 objects that remain. Called once from 198 perl_destruct(), prior to calling sv_clean_all() 199 below. 200 201 sv_clean_all() / do_clean_all() 202 SvREFCNT_dec(sv) each remaining SV, possibly 203 triggering an sv_free(). It also sets the 204 SVf_BREAK flag on the SV to indicate that the 205 refcnt has been artificially lowered, and thus 206 stopping sv_free() from giving spurious warnings 207 about SVs which unexpectedly have a refcnt 208 of zero. called repeatedly from perl_destruct() 209 until there are no SVs left. 210 211=head2 Arena allocator API Summary 212 213Private API to rest of sv.c 214 215 new_SV(), del_SV(), 216 217 new_XPVNV(), del_body() 218 etc 219 220Public API: 221 222 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas() 223 224=cut 225 226 * ========================================================================= */ 227 228/* 229 * "A time to plant, and a time to uproot what was planted..." 230 */ 231 232#ifdef DEBUG_LEAKING_SCALARS 233# define FREE_SV_DEBUG_FILE(sv) STMT_START { \ 234 if ((sv)->sv_debug_file) { \ 235 PerlMemShared_free((sv)->sv_debug_file); \ 236 sv->sv_debug_file = NULL; \ 237 } \ 238 } STMT_END 239# define DEBUG_SV_SERIAL(sv) \ 240 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \ 241 PTR2UV(sv), (long)(sv)->sv_debug_serial)) 242#else 243# define FREE_SV_DEBUG_FILE(sv) 244# define DEBUG_SV_SERIAL(sv) NOOP 245#endif 246 247/* Mark an SV head as unused, and add to free list. 248 * 249 * If SVf_BREAK is set, skip adding it to the free list, as this SV had 250 * its refcount artificially decremented during global destruction, so 251 * there may be dangling pointers to it. The last thing we want in that 252 * case is for it to be reused. */ 253 254#define plant_SV(p) \ 255 STMT_START { \ 256 const U32 old_flags = SvFLAGS(p); \ 257 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \ 258 DEBUG_SV_SERIAL(p); \ 259 FREE_SV_DEBUG_FILE(p); \ 260 POISON_SV_HEAD(p); \ 261 SvFLAGS(p) = SVTYPEMASK; \ 262 if (!(old_flags & SVf_BREAK)) { \ 263 SvARENA_CHAIN_SET(p, PL_sv_root); \ 264 PL_sv_root = (p); \ 265 } \ 266 --PL_sv_count; \ 267 } STMT_END 268 269 270/* make some more SVs by adding another arena */ 271 272SV* 273Perl_more_sv(pTHX) 274{ 275 SV* sv; 276 char *chunk; /* must use New here to match call to */ 277 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */ 278 sv_add_arena(chunk, PERL_ARENA_SIZE, 0); 279 uproot_SV(sv); 280 return sv; 281} 282 283/* del_SV(): return an empty SV head to the free list */ 284 285#ifdef DEBUGGING 286 287#define del_SV(p) \ 288 STMT_START { \ 289 if (DEBUG_D_TEST) \ 290 del_sv(p); \ 291 else \ 292 plant_SV(p); \ 293 } STMT_END 294 295STATIC void 296S_del_sv(pTHX_ SV *p) 297{ 298 PERL_ARGS_ASSERT_DEL_SV; 299 300 if (DEBUG_D_TEST) { 301 SV* sva; 302 bool ok = 0; 303 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) { 304 const SV * const sv = sva + 1; 305 const SV * const svend = &sva[SvREFCNT(sva)]; 306 if (p >= sv && p < svend) { 307 ok = 1; 308 break; 309 } 310 } 311 if (!ok) { 312 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), 313 "Attempt to free non-arena SV: 0x%" UVxf 314 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE); 315 return; 316 } 317 } 318 plant_SV(p); 319} 320 321#else /* ! DEBUGGING */ 322 323#define del_SV(p) plant_SV(p) 324 325#endif /* DEBUGGING */ 326 327 328/* 329=for apidoc_section $SV 330 331=for apidoc sv_add_arena 332 333Given a chunk of memory, link it to the head of the list of arenas, 334and split it into a list of free SVs. 335 336=cut 337*/ 338 339static void 340S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags) 341{ 342 SV *const sva = MUTABLE_SV(ptr); 343 SV* sv; 344 SV* svend; 345 346 PERL_ARGS_ASSERT_SV_ADD_ARENA; 347 348 /* The first SV in an arena isn't an SV. */ 349 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */ 350 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */ 351 SvFLAGS(sva) = flags; /* FAKE if not to be freed */ 352 353 PL_sv_arenaroot = sva; 354 PL_sv_root = sva + 1; 355 356 svend = &sva[SvREFCNT(sva) - 1]; 357 sv = sva + 1; 358 while (sv < svend) { 359 SvARENA_CHAIN_SET(sv, (sv + 1)); 360#ifdef DEBUGGING 361 SvREFCNT(sv) = 0; 362#endif 363 /* Must always set typemask because it's always checked in on cleanup 364 when the arenas are walked looking for objects. */ 365 SvFLAGS(sv) = SVTYPEMASK; 366 sv++; 367 } 368 SvARENA_CHAIN_SET(sv, 0); 369#ifdef DEBUGGING 370 SvREFCNT(sv) = 0; 371#endif 372 SvFLAGS(sv) = SVTYPEMASK; 373} 374 375/* visit(): call the named function for each non-free SV in the arenas 376 * whose flags field matches the flags/mask args. */ 377 378STATIC I32 379S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask) 380{ 381 SV* sva; 382 I32 visited = 0; 383 384 PERL_ARGS_ASSERT_VISIT; 385 386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) { 387 const SV * const svend = &sva[SvREFCNT(sva)]; 388 SV* sv; 389 for (sv = sva + 1; sv < svend; ++sv) { 390 if (!SvIS_FREED(sv) 391 && (sv->sv_flags & mask) == flags 392 && SvREFCNT(sv)) 393 { 394 (*f)(aTHX_ sv); 395 ++visited; 396 } 397 } 398 } 399 return visited; 400} 401 402#ifdef DEBUGGING 403 404/* called by sv_report_used() for each live SV */ 405 406static void 407do_report_used(pTHX_ SV *const sv) 408{ 409 if (!SvIS_FREED(sv)) { 410 PerlIO_printf(Perl_debug_log, "****\n"); 411 sv_dump(sv); 412 } 413} 414#endif 415 416/* 417=for apidoc sv_report_used 418 419Dump the contents of all SVs not yet freed (debugging aid). 420 421=cut 422*/ 423 424void 425Perl_sv_report_used(pTHX) 426{ 427#ifdef DEBUGGING 428 visit(do_report_used, 0, 0); 429#else 430 PERL_UNUSED_CONTEXT; 431#endif 432} 433 434/* called by sv_clean_objs() for each live SV */ 435 436static void 437do_clean_objs(pTHX_ SV *const ref) 438{ 439 assert (SvROK(ref)); 440 { 441 SV * const target = SvRV(ref); 442 if (SvOBJECT(target)) { 443 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref))); 444 if (SvWEAKREF(ref)) { 445 sv_del_backref(target, ref); 446 SvWEAKREF_off(ref); 447 SvRV_set(ref, NULL); 448 } else { 449 SvROK_off(ref); 450 SvRV_set(ref, NULL); 451 SvREFCNT_dec_NN(target); 452 } 453 } 454 } 455} 456 457 458/* clear any slots in a GV which hold objects - except IO; 459 * called by sv_clean_objs() for each live GV */ 460 461static void 462do_clean_named_objs(pTHX_ SV *const sv) 463{ 464 SV *obj; 465 assert(SvTYPE(sv) == SVt_PVGV); 466 assert(isGV_with_GP(sv)); 467 if (!GvGP(sv)) 468 return; 469 470 /* freeing GP entries may indirectly free the current GV; 471 * hold onto it while we mess with the GP slots */ 472 SvREFCNT_inc(sv); 473 474 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) { 475 DEBUG_D((PerlIO_printf(Perl_debug_log, 476 "Cleaning named glob SV object:\n "), sv_dump(obj))); 477 GvSV(sv) = NULL; 478 SvREFCNT_dec_NN(obj); 479 } 480 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) { 481 DEBUG_D((PerlIO_printf(Perl_debug_log, 482 "Cleaning named glob AV object:\n "), sv_dump(obj))); 483 GvAV(sv) = NULL; 484 SvREFCNT_dec_NN(obj); 485 } 486 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) { 487 DEBUG_D((PerlIO_printf(Perl_debug_log, 488 "Cleaning named glob HV object:\n "), sv_dump(obj))); 489 GvHV(sv) = NULL; 490 SvREFCNT_dec_NN(obj); 491 } 492 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) { 493 DEBUG_D((PerlIO_printf(Perl_debug_log, 494 "Cleaning named glob CV object:\n "), sv_dump(obj))); 495 GvCV_set(sv, NULL); 496 SvREFCNT_dec_NN(obj); 497 } 498 SvREFCNT_dec_NN(sv); /* undo the inc above */ 499} 500 501/* clear any IO slots in a GV which hold objects (except stderr, defout); 502 * called by sv_clean_objs() for each live GV */ 503 504static void 505do_clean_named_io_objs(pTHX_ SV *const sv) 506{ 507 SV *obj; 508 assert(SvTYPE(sv) == SVt_PVGV); 509 assert(isGV_with_GP(sv)); 510 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv) 511 return; 512 513 SvREFCNT_inc(sv); 514 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) { 515 DEBUG_D((PerlIO_printf(Perl_debug_log, 516 "Cleaning named glob IO object:\n "), sv_dump(obj))); 517 GvIOp(sv) = NULL; 518 SvREFCNT_dec_NN(obj); 519 } 520 SvREFCNT_dec_NN(sv); /* undo the inc above */ 521} 522 523/* Void wrapper to pass to visit() */ 524static void 525do_curse(pTHX_ SV * const sv) { 526 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv) 527 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv)) 528 return; 529 (void)curse(sv, 0); 530} 531 532/* 533=for apidoc sv_clean_objs 534 535Attempt to destroy all objects not yet freed. 536 537=cut 538*/ 539 540void 541Perl_sv_clean_objs(pTHX) 542{ 543 GV *olddef, *olderr; 544 PL_in_clean_objs = TRUE; 545 visit(do_clean_objs, SVf_ROK, SVf_ROK); 546 /* Some barnacles may yet remain, clinging to typeglobs. 547 * Run the non-IO destructors first: they may want to output 548 * error messages, close files etc */ 549 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP); 550 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP); 551 /* And if there are some very tenacious barnacles clinging to arrays, 552 closures, or what have you.... */ 553 visit(do_curse, SVs_OBJECT, SVs_OBJECT); 554 olddef = PL_defoutgv; 555 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */ 556 if (olddef && isGV_with_GP(olddef)) 557 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef)); 558 olderr = PL_stderrgv; 559 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */ 560 if (olderr && isGV_with_GP(olderr)) 561 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr)); 562 SvREFCNT_dec(olddef); 563 PL_in_clean_objs = FALSE; 564} 565 566/* called by sv_clean_all() for each live SV */ 567 568static void 569do_clean_all(pTHX_ SV *const sv) 570{ 571 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) { 572 /* don't clean pid table and strtab */ 573 return; 574 } 575 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) )); 576 SvFLAGS(sv) |= SVf_BREAK; 577 SvREFCNT_dec_NN(sv); 578} 579 580/* 581=for apidoc sv_clean_all 582 583Decrement the refcnt of each remaining SV, possibly triggering a 584cleanup. This function may have to be called multiple times to free 585SVs which are in complex self-referential hierarchies. 586 587=cut 588*/ 589 590I32 591Perl_sv_clean_all(pTHX) 592{ 593 I32 cleaned; 594 PL_in_clean_all = TRUE; 595 cleaned = visit(do_clean_all, 0,0); 596 return cleaned; 597} 598 599/* 600 ARENASETS: a meta-arena implementation which separates arena-info 601 into struct arena_set, which contains an array of struct 602 arena_descs, each holding info for a single arena. By separating 603 the meta-info from the arena, we recover the 1st slot, formerly 604 borrowed for list management. The arena_set is about the size of an 605 arena, avoiding the needless malloc overhead of a naive linked-list. 606 607 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused 608 memory in the last arena-set (1/2 on average). In trade, we get 609 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for 610 smaller types). The recovery of the wasted space allows use of 611 small arenas for large, rare body types, by changing array* fields 612 in body_details_by_type[] below. 613*/ 614struct arena_desc { 615 char *arena; /* the raw storage, allocated aligned */ 616 size_t size; /* its size ~4k typ */ 617 svtype utype; /* bodytype stored in arena */ 618}; 619 620struct arena_set; 621 622/* Get the maximum number of elements in set[] such that struct arena_set 623 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and 624 therefore likely to be 1 aligned memory page. */ 625 626#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \ 627 - 2 * sizeof(int)) / sizeof (struct arena_desc)) 628 629struct arena_set { 630 struct arena_set* next; 631 unsigned int set_size; /* ie ARENAS_PER_SET */ 632 unsigned int curr; /* index of next available arena-desc */ 633 struct arena_desc set[ARENAS_PER_SET]; 634}; 635 636/* 637=for apidoc sv_free_arenas 638 639Deallocate the memory used by all arenas. Note that all the individual SV 640heads and bodies within the arenas must already have been freed. 641 642=cut 643 644*/ 645void 646Perl_sv_free_arenas(pTHX) 647{ 648 SV* sva; 649 SV* svanext; 650 unsigned int i; 651 652 /* Free arenas here, but be careful about fake ones. (We assume 653 contiguity of the fake ones with the corresponding real ones.) */ 654 655 for (sva = PL_sv_arenaroot; sva; sva = svanext) { 656 svanext = MUTABLE_SV(SvANY(sva)); 657 while (svanext && SvFAKE(svanext)) 658 svanext = MUTABLE_SV(SvANY(svanext)); 659 660 if (!SvFAKE(sva)) 661 Safefree(sva); 662 } 663 664 { 665 struct arena_set *aroot = (struct arena_set*) PL_body_arenas; 666 667 while (aroot) { 668 struct arena_set *current = aroot; 669 i = aroot->curr; 670 while (i--) { 671 assert(aroot->set[i].arena); 672 Safefree(aroot->set[i].arena); 673 } 674 aroot = aroot->next; 675 Safefree(current); 676 } 677 } 678 PL_body_arenas = 0; 679 680 i = PERL_ARENA_ROOTS_SIZE; 681 while (i--) 682 PL_body_roots[i] = 0; 683 684 PL_sv_arenaroot = 0; 685 PL_sv_root = 0; 686} 687 688/* 689 Historically, here were mid-level routines that manage the 690 allocation of bodies out of the various arenas. Some of these 691 routines and related definitions remain here, but others were 692 moved into sv_inline.h to facilitate inlining of newSV_type(). 693 694 There are 4 kinds of arenas: 695 696 1. SV-head arenas, which are discussed and handled above 697 2. regular body arenas 698 3. arenas for reduced-size bodies 699 4. Hash-Entry arenas 700 701 Arena types 2 & 3 are chained by body-type off an array of 702 arena-root pointers, which is indexed by svtype. Some of the 703 larger/less used body types are malloced singly, since a large 704 unused block of them is wasteful. Also, several svtypes don't have 705 bodies; the data fits into the sv-head itself. The arena-root 706 pointer thus has a few unused root-pointers (which may be hijacked 707 later for arena type 4) 708 709 3 differs from 2 as an optimization; some body types have several 710 unused fields in the front of the structure (which are kept in-place 711 for consistency). These bodies can be allocated in smaller chunks, 712 because the leading fields arent accessed. Pointers to such bodies 713 are decremented to point at the unused 'ghost' memory, knowing that 714 the pointers are used with offsets to the real memory. 715 716Allocation of SV-bodies is similar to SV-heads, differing as follows; 717the allocation mechanism is used for many body types, so is somewhat 718more complicated, it uses arena-sets, and has no need for still-live 719SV detection. 720 721At the outermost level, (new|del)_X*V macros return bodies of the 722appropriate type. These macros call either (new|del)_body_type or 723(new|del)_body_allocated macro pairs, depending on specifics of the 724type. Most body types use the former pair, the latter pair is used to 725allocate body types with "ghost fields". 726 727"ghost fields" are fields that are unused in certain types, and 728consequently don't need to actually exist. They are declared because 729they're part of a "base type", which allows use of functions as 730methods. The simplest examples are AVs and HVs, 2 aggregate types 731which don't use the fields which support SCALAR semantics. 732 733For these types, the arenas are carved up into appropriately sized 734chunks, we thus avoid wasted memory for those unaccessed members. 735When bodies are allocated, we adjust the pointer back in memory by the 736size of the part not allocated, so it's as if we allocated the full 737structure. (But things will all go boom if you write to the part that 738is "not there", because you'll be overwriting the last members of the 739preceding structure in memory.) 740 741We calculate the correction using the STRUCT_OFFSET macro on the first 742member present. If the allocated structure is smaller (no initial NV 743actually allocated) then the net effect is to subtract the size of the NV 744from the pointer, to return a new pointer as if an initial NV were actually 745allocated. (We were using structures named *_allocated for this, but 746this turned out to be a subtle bug, because a structure without an NV 747could have a lower alignment constraint, but the compiler is allowed to 748optimised accesses based on the alignment constraint of the actual pointer 749to the full structure, for example, using a single 64 bit load instruction 750because it "knows" that two adjacent 32 bit members will be 8-byte aligned.) 751 752This is the same trick as was used for NV and IV bodies. Ironically it 753doesn't need to be used for NV bodies any more, because NV is now at 754the start of the structure. IV bodies, and also in some builds NV bodies, 755don't need it either, because they are no longer allocated. 756 757In turn, the new_body_* allocators call S_new_body(), which invokes 758new_body_from_arena macro, which takes a lock, and takes a body off the 759linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if 760necessary to refresh an empty list. Then the lock is released, and 761the body is returned. 762 763Perl_more_bodies allocates a new arena, and carves it up into an array of N 764bodies, which it strings into a linked list. It looks up arena-size 765and body-size from the body_details table described below, thus 766supporting the multiple body-types. 767 768If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and 769the (new|del)_X*V macros are mapped directly to malloc/free. 770 771For each sv-type, struct body_details bodies_by_type[] carries 772parameters which control these aspects of SV handling: 773 774Arena_size determines whether arenas are used for this body type, and if 775so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to 776zero, forcing individual mallocs and frees. 777 778Body_size determines how big a body is, and therefore how many fit into 779each arena. Offset carries the body-pointer adjustment needed for 780"ghost fields", and is used in *_allocated macros. 781 782But its main purpose is to parameterize info needed in 783Perl_sv_upgrade(). The info here dramatically simplifies the function 784vs the implementation in 5.8.8, making it table-driven. All fields 785are used for this, except for arena_size. 786 787For the sv-types that have no bodies, arenas are not used, so those 788PL_body_roots[sv_type] are unused, and can be overloaded. In 789something of a special case, SVt_NULL is borrowed for HE arenas; 790PL_body_roots[HE_ARENA_ROOT_IX=SVt_NULL] is filled by S_more_he, but the 791bodies_by_type[SVt_NULL] slot is not used, as the table is not 792available in hv.c. Similarly SVt_IV is re-used for HVAUX_ARENA_ROOT_IX. 793 794*/ 795 796/* return a thing to the free list */ 797 798#define del_body(thing, root) \ 799 STMT_START { \ 800 void ** const thing_copy = (void **)thing; \ 801 *thing_copy = *root; \ 802 *root = (void*)thing_copy; \ 803 } STMT_END 804 805 806void * 807Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size, 808 const size_t arena_size) 809{ 810 void ** const root = &PL_body_roots[sv_type]; 811 struct arena_desc *adesc; 812 struct arena_set *aroot = (struct arena_set *) PL_body_arenas; 813 unsigned int curr; 814 char *start; 815 const char *end; 816 const size_t good_arena_size = Perl_malloc_good_size(arena_size); 817#if defined(DEBUGGING) 818 static bool done_sanity_check; 819 820 if (!done_sanity_check) { 821 unsigned int i = SVt_LAST; 822 823 done_sanity_check = TRUE; 824 825 while (i--) 826 assert (bodies_by_type[i].type == i); 827 } 828#endif 829 830 assert(arena_size); 831 832 /* may need new arena-set to hold new arena */ 833 if (!aroot || aroot->curr >= aroot->set_size) { 834 struct arena_set *newroot; 835 Newxz(newroot, 1, struct arena_set); 836 newroot->set_size = ARENAS_PER_SET; 837 newroot->next = aroot; 838 aroot = newroot; 839 PL_body_arenas = (void *) newroot; 840 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot)); 841 } 842 843 /* ok, now have arena-set with at least 1 empty/available arena-desc */ 844 curr = aroot->curr++; 845 adesc = &(aroot->set[curr]); 846 assert(!adesc->arena); 847 848 Newx(adesc->arena, good_arena_size, char); 849 adesc->size = good_arena_size; 850 adesc->utype = sv_type; 851 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n", 852 curr, (void*)adesc->arena, (UV)good_arena_size)); 853 854 start = (char *) adesc->arena; 855 856 /* Get the address of the byte after the end of the last body we can fit. 857 Remember, this is integer division: */ 858 end = start + good_arena_size / body_size * body_size; 859 860 /* computed count doesn't reflect the 1st slot reservation */ 861#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE) 862 DEBUG_m(PerlIO_printf(Perl_debug_log, 863 "arena %p end %p arena-size %d (from %d) type %d " 864 "size %d ct %d\n", 865 (void*)start, (void*)end, (int)good_arena_size, 866 (int)arena_size, sv_type, (int)body_size, 867 (int)good_arena_size / (int)body_size)); 868#else 869 DEBUG_m(PerlIO_printf(Perl_debug_log, 870 "arena %p end %p arena-size %d type %d size %d ct %d\n", 871 (void*)start, (void*)end, 872 (int)arena_size, sv_type, (int)body_size, 873 (int)good_arena_size / (int)body_size)); 874#endif 875 *root = (void *)start; 876 877 while (1) { 878 /* Where the next body would start: */ 879 char * const next = start + body_size; 880 881 if (next >= end) { 882 /* This is the last body: */ 883 assert(next == end); 884 885 *(void **)start = 0; 886 return *root; 887 } 888 889 *(void**) start = (void *)next; 890 start = next; 891 } 892} 893 894/* 895=for apidoc sv_upgrade 896 897Upgrade an SV to a more complex form. Generally adds a new body type to the 898SV, then copies across as much information as possible from the old body. 899It croaks if the SV is already in a more complex form than requested. You 900generally want to use the C<SvUPGRADE> macro wrapper, which checks the type 901before calling C<sv_upgrade>, and hence does not croak. See also 902C<L</svtype>>. 903 904=cut 905*/ 906 907void 908Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type) 909{ 910 void* old_body; 911 void* new_body; 912 const svtype old_type = SvTYPE(sv); 913 const struct body_details *new_type_details; 914 const struct body_details *old_type_details 915 = bodies_by_type + old_type; 916 SV *referent = NULL; 917 918 PERL_ARGS_ASSERT_SV_UPGRADE; 919 920 if (old_type == new_type) 921 return; 922 923 /* This clause was purposefully added ahead of the early return above to 924 the shared string hackery for (sort {$a <=> $b} keys %hash), with the 925 inference by Nick I-S that it would fix other troublesome cases. See 926 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent) 927 928 Given that shared hash key scalars are no longer PVIV, but PV, there is 929 no longer need to unshare so as to free up the IVX slot for its proper 930 purpose. So it's safe to move the early return earlier. */ 931 932 if (new_type > SVt_PVMG && SvIsCOW(sv)) { 933 sv_force_normal_flags(sv, 0); 934 } 935 936 old_body = SvANY(sv); 937 938 /* Copying structures onto other structures that have been neatly zeroed 939 has a subtle gotcha. Consider XPVMG 940 941 +------+------+------+------+------+-------+-------+ 942 | NV | CUR | LEN | IV | MAGIC | STASH | 943 +------+------+------+------+------+-------+-------+ 944 0 4 8 12 16 20 24 28 945 946 where NVs are aligned to 8 bytes, so that sizeof that structure is 947 actually 32 bytes long, with 4 bytes of padding at the end: 948 949 +------+------+------+------+------+-------+-------+------+ 950 | NV | CUR | LEN | IV | MAGIC | STASH | ??? | 951 +------+------+------+------+------+-------+-------+------+ 952 0 4 8 12 16 20 24 28 32 953 954 so what happens if you allocate memory for this structure: 955 956 +------+------+------+------+------+-------+-------+------+------+... 957 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME | 958 +------+------+------+------+------+-------+-------+------+------+... 959 0 4 8 12 16 20 24 28 32 36 960 961 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you 962 expect, because you copy the area marked ??? onto GP. Now, ??? may have 963 started out as zero once, but it's quite possible that it isn't. So now, 964 rather than a nicely zeroed GP, you have it pointing somewhere random. 965 Bugs ensue. 966 967 (In fact, GP ends up pointing at a previous GP structure, because the 968 principle cause of the padding in XPVMG getting garbage is a copy of 969 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now 970 this happens to be moot because XPVGV has been re-ordered, with GP 971 no longer after STASH) 972 973 So we are careful and work out the size of used parts of all the 974 structures. */ 975 976 switch (old_type) { 977 case SVt_NULL: 978 break; 979 case SVt_IV: 980 if (SvROK(sv)) { 981 referent = SvRV(sv); 982 old_type_details = &fake_rv; 983 if (new_type == SVt_NV) 984 new_type = SVt_PVNV; 985 } else { 986 if (new_type < SVt_PVIV) { 987 new_type = (new_type == SVt_NV) 988 ? SVt_PVNV : SVt_PVIV; 989 } 990 } 991 break; 992 case SVt_NV: 993 if (new_type < SVt_PVNV) { 994 new_type = SVt_PVNV; 995 } 996 break; 997 case SVt_PV: 998 assert(new_type > SVt_PV); 999 STATIC_ASSERT_STMT(SVt_IV < SVt_PV); 1000 STATIC_ASSERT_STMT(SVt_NV < SVt_PV); 1001 break; 1002 case SVt_PVIV: 1003 break; 1004 case SVt_PVNV: 1005 break; 1006 case SVt_PVMG: 1007 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena, 1008 there's no way that it can be safely upgraded, because perl.c 1009 expects to Safefree(SvANY(PL_mess_sv)) */ 1010 assert(sv != PL_mess_sv); 1011 break; 1012 default: 1013 if (UNLIKELY(old_type_details->cant_upgrade)) 1014 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf, 1015 sv_reftype(sv, 0), (UV) old_type, (UV) new_type); 1016 } 1017 1018 if (UNLIKELY(old_type > new_type)) 1019 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d", 1020 (int)old_type, (int)new_type); 1021 1022 new_type_details = bodies_by_type + new_type; 1023 1024 SvFLAGS(sv) &= ~SVTYPEMASK; 1025 SvFLAGS(sv) |= new_type; 1026 1027 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of 1028 the return statements above will have triggered. */ 1029 assert (new_type != SVt_NULL); 1030 switch (new_type) { 1031 case SVt_IV: 1032 assert(old_type == SVt_NULL); 1033 SET_SVANY_FOR_BODYLESS_IV(sv); 1034 SvIV_set(sv, 0); 1035 return; 1036 case SVt_NV: 1037 assert(old_type == SVt_NULL); 1038#if NVSIZE <= IVSIZE 1039 SET_SVANY_FOR_BODYLESS_NV(sv); 1040#else 1041 SvANY(sv) = new_XNV(); 1042#endif 1043 SvNV_set(sv, 0); 1044 return; 1045 case SVt_PVHV: 1046 case SVt_PVAV: 1047 case SVt_PVOBJ: 1048 assert(new_type_details->body_size); 1049 1050#ifndef PURIFY 1051 assert(new_type_details->arena); 1052 assert(new_type_details->arena_size); 1053 /* This points to the start of the allocated area. */ 1054 new_body = S_new_body(aTHX_ new_type); 1055 /* xpvav and xpvhv have no offset, so no need to adjust new_body */ 1056 assert(!(new_type_details->offset)); 1057#else 1058 /* We always allocated the full length item with PURIFY. To do this 1059 we fake things so that arena is false for all 16 types.. */ 1060 new_body = new_NOARENAZ(new_type_details); 1061#endif 1062 SvANY(sv) = new_body; 1063 switch(new_type) { 1064 case SVt_PVAV: 1065 { 1066 XPVAV pvav = { 1067 .xmg_stash = NULL, 1068 .xmg_u = {.xmg_magic = NULL}, 1069 .xav_fill = -1, .xav_max = -1, .xav_alloc = 0 1070 }; 1071 *((XPVAV*) SvANY(sv)) = pvav; 1072 } 1073 1074 AvREAL_only(sv); 1075 break; 1076 case SVt_PVHV: 1077 { 1078 XPVHV pvhv = { 1079 .xmg_stash = NULL, 1080 .xmg_u = {.xmg_magic = NULL}, 1081 .xhv_keys = 0, 1082 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */ 1083 .xhv_max = PERL_HASH_DEFAULT_HvMAX 1084 }; 1085 *((XPVHV*) SvANY(sv)) = pvhv; 1086 } 1087 1088 assert(!SvOK(sv)); 1089 SvOK_off(sv); 1090#ifndef NODEFAULT_SHAREKEYS 1091 HvSHAREKEYS_on(sv); /* key-sharing on by default */ 1092#endif 1093 break; 1094 case SVt_PVOBJ: 1095 { 1096 XPVOBJ pvo = { 1097 .xmg_stash = NULL, .xmg_u = {.xmg_magic = NULL}, 1098 .xobject_maxfield = -1, 1099 .xobject_iter_sv_at = 0, 1100 .xobject_fields = NULL, 1101 }; 1102 *((XPVOBJ*) SvANY(sv)) = pvo; 1103 } 1104 break; 1105 default: 1106 NOT_REACHED; 1107 } 1108 1109 /* SVt_NULL isn't the only thing upgraded to AV or HV. 1110 The target created by newSVrv also is, and it can have magic. 1111 However, it never has SvPVX set. 1112 */ 1113 if (old_type == SVt_IV) { 1114 assert(!SvROK(sv)); 1115 } else if (old_type >= SVt_PV) { 1116 assert(SvPVX_const(sv) == 0); 1117 } 1118 1119 if (old_type >= SVt_PVMG) { 1120 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic); 1121 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash); 1122 } else { 1123 sv->sv_u.svu_array = NULL; /* or svu_hash */ 1124 } 1125 break; 1126 1127 case SVt_PVIV: 1128 /* XXX Is this still needed? Was it ever needed? Surely as there is 1129 no route from NV to PVIV, NOK can never be true */ 1130 assert(!SvNOKp(sv)); 1131 assert(!SvNOK(sv)); 1132 /* FALLTHROUGH */ 1133 case SVt_PVIO: 1134 case SVt_PVFM: 1135 case SVt_PVGV: 1136 case SVt_PVCV: 1137 case SVt_PVLV: 1138 case SVt_INVLIST: 1139 case SVt_REGEXP: 1140 case SVt_PVMG: 1141 case SVt_PVNV: 1142 case SVt_PV: 1143 1144 assert(new_type_details->body_size); 1145 /* We always allocated the full length item with PURIFY. To do this 1146 we fake things so that arena is false for all 16 types.. */ 1147#ifndef PURIFY 1148 if(new_type_details->arena) { 1149 /* This points to the start of the allocated area. */ 1150 new_body = S_new_body(aTHX_ new_type); 1151 Zero(new_body, new_type_details->body_size, char); 1152 new_body = ((char *)new_body) - new_type_details->offset; 1153 } else 1154#endif 1155 { 1156 new_body = new_NOARENAZ(new_type_details); 1157 } 1158 SvANY(sv) = new_body; 1159 1160 if (old_type_details->copy) { 1161 /* There is now the potential for an upgrade from something without 1162 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */ 1163 int offset = old_type_details->offset; 1164 int length = old_type_details->copy; 1165 1166 if (new_type_details->offset > old_type_details->offset) { 1167 const int difference 1168 = new_type_details->offset - old_type_details->offset; 1169 offset += difference; 1170 length -= difference; 1171 } 1172 assert (length >= 0); 1173 1174 Copy((char *)old_body + offset, (char *)new_body + offset, length, 1175 char); 1176 } 1177 1178#ifndef NV_ZERO_IS_ALLBITS_ZERO 1179 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a 1180 * correct 0.0 for us. Otherwise, if the old body didn't have an 1181 * NV slot, but the new one does, then we need to initialise the 1182 * freshly created NV slot with whatever the correct bit pattern is 1183 * for 0.0 */ 1184 if (old_type_details->zero_nv && !new_type_details->zero_nv 1185 && !isGV_with_GP(sv)) 1186 SvNV_set(sv, 0); 1187#endif 1188 1189 if (UNLIKELY(new_type == SVt_PVIO)) { 1190 IO * const io = MUTABLE_IO(sv); 1191 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV); 1192 1193 SvOBJECT_on(io); 1194 /* Clear the stashcache because a new IO could overrule a package 1195 name */ 1196 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n")); 1197 hv_clear(PL_stashcache); 1198 1199 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv)))); 1200 IoPAGE_LEN(sv) = 60; 1201 } 1202 if (old_type < SVt_PV) { 1203 /* referent will be NULL unless the old type was SVt_IV emulating 1204 SVt_RV */ 1205 sv->sv_u.svu_rv = referent; 1206 } 1207 break; 1208 default: 1209 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu", 1210 (unsigned long)new_type); 1211 } 1212 1213 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV, 1214 and sometimes SVt_NV */ 1215 if (old_type_details->body_size) { 1216#ifdef PURIFY 1217 safefree(old_body); 1218#else 1219 /* Note that there is an assumption that all bodies of types that 1220 can be upgraded came from arenas. Only the more complex non- 1221 upgradable types are allowed to be directly malloc()ed. */ 1222 assert(old_type_details->arena); 1223 del_body((void*)((char*)old_body + old_type_details->offset), 1224 &PL_body_roots[old_type]); 1225#endif 1226 } 1227} 1228 1229struct xpvhv_aux* 1230Perl_hv_auxalloc(pTHX_ HV *hv) { 1231 const struct body_details *old_type_details = bodies_by_type + SVt_PVHV; 1232 void *old_body; 1233 void *new_body; 1234 1235 PERL_ARGS_ASSERT_HV_AUXALLOC; 1236 assert(SvTYPE(hv) == SVt_PVHV); 1237 assert(!HvHasAUX(hv)); 1238 1239#ifdef PURIFY 1240 new_body = new_NOARENAZ(&fake_hv_with_aux); 1241#else 1242 new_body_from_arena(new_body, HVAUX_ARENA_ROOT_IX, fake_hv_with_aux); 1243#endif 1244 1245 old_body = SvANY(hv); 1246 1247 Copy((char *)old_body + old_type_details->offset, 1248 (char *)new_body + fake_hv_with_aux.offset, 1249 old_type_details->copy, 1250 char); 1251 1252#ifdef PURIFY 1253 safefree(old_body); 1254#else 1255 assert(old_type_details->arena); 1256 del_body((void*)((char*)old_body + old_type_details->offset), 1257 &PL_body_roots[SVt_PVHV]); 1258#endif 1259 1260 SvANY(hv) = (XPVHV *) new_body; 1261 SvFLAGS(hv) |= SVphv_HasAUX; 1262 return HvAUX(hv); 1263} 1264 1265/* 1266=for apidoc sv_backoff 1267 1268Remove any string offset. You should normally use the C<SvOOK_off> macro 1269wrapper instead. 1270 1271=cut 1272*/ 1273 1274/* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS 1275 prior to 5.23.4 this function always returned 0 1276*/ 1277 1278void 1279Perl_sv_backoff(SV *const sv) 1280{ 1281 STRLEN delta; 1282 const char * const s = SvPVX_const(sv); 1283 1284 PERL_ARGS_ASSERT_SV_BACKOFF; 1285 1286 assert(SvOOK(sv)); 1287 assert(SvTYPE(sv) != SVt_PVHV); 1288 assert(SvTYPE(sv) != SVt_PVAV); 1289 1290 SvOOK_offset(sv, delta); 1291 1292 SvLEN_set(sv, SvLEN(sv) + delta); 1293 SvPV_set(sv, SvPVX(sv) - delta); 1294 SvFLAGS(sv) &= ~SVf_OOK; 1295 Move(s, SvPVX(sv), SvCUR(sv)+1, char); 1296 return; 1297} 1298 1299 1300/* forward declaration */ 1301static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags); 1302 1303 1304/* 1305=for apidoc sv_grow 1306 1307Expands the character buffer in the SV. If necessary, uses C<sv_unref> and 1308upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer. 1309Use the C<SvGROW> wrapper instead. 1310 1311=cut 1312*/ 1313 1314 1315char * 1316Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen) 1317{ 1318 char *s; 1319 1320 PERL_ARGS_ASSERT_SV_GROW; 1321 1322 if (SvROK(sv)) 1323 sv_unref(sv); 1324 if (SvTYPE(sv) < SVt_PV) { 1325 sv_upgrade(sv, SVt_PV); 1326 s = SvPVX_mutable(sv); 1327 } 1328 else if (SvOOK(sv)) { /* pv is offset? */ 1329 sv_backoff(sv); 1330 s = SvPVX_mutable(sv); 1331 if (newlen > SvLEN(sv)) 1332 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */ 1333 } 1334 else 1335 { 1336 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0); 1337 s = SvPVX_mutable(sv); 1338 } 1339 1340#ifdef PERL_COPY_ON_WRITE 1341 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare) 1342 * to store the COW count. So in general, allocate one more byte than 1343 * asked for, to make it likely this byte is always spare: and thus 1344 * make more strings COW-able. 1345 * 1346 * Only increment if the allocation isn't MEM_SIZE_MAX, 1347 * otherwise it will wrap to 0. 1348 */ 1349 if ( newlen != MEM_SIZE_MAX ) 1350 newlen++; 1351#endif 1352 1353#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size) 1354#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC 1355#endif 1356 1357 if (newlen > SvLEN(sv)) { /* need more room? */ 1358 STRLEN minlen = SvCUR(sv); 1359 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + PERL_STRLEN_NEW_MIN; 1360 if (newlen < minlen) 1361 newlen = minlen; 1362#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC 1363 1364 /* Don't round up on the first allocation, as odds are pretty good that 1365 * the initial request is accurate as to what is really needed */ 1366 if (SvLEN(sv)) { 1367 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen); 1368 if (rounded > newlen) 1369 newlen = rounded; 1370 } 1371#endif 1372 if (SvLEN(sv) && s) { 1373 s = (char*)saferealloc(s, newlen); 1374 } 1375 else { 1376 s = (char*)safemalloc(newlen); 1377 if (SvPVX_const(sv) && SvCUR(sv)) { 1378 Move(SvPVX_const(sv), s, SvCUR(sv), char); 1379 } 1380 } 1381 SvPV_set(sv, s); 1382#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC 1383 /* Do this here, do it once, do it right, and then we will never get 1384 called back into sv_grow() unless there really is some growing 1385 needed. */ 1386 SvLEN_set(sv, Perl_safesysmalloc_size(s)); 1387#else 1388 SvLEN_set(sv, newlen); 1389#endif 1390 } 1391 return s; 1392} 1393 1394/* 1395=for apidoc sv_grow_fresh 1396 1397A cut-down version of sv_grow intended only for when sv is a freshly-minted 1398SVt_PV, SVt_PVIV, SVt_PVNV, or SVt_PVMG. i.e. sv has the default flags, has 1399never been any other type, and does not have an existing string. Basically, 1400just assigns a char buffer and returns a pointer to it. 1401 1402=cut 1403*/ 1404 1405 1406char * 1407Perl_sv_grow_fresh(pTHX_ SV *const sv, STRLEN newlen) 1408{ 1409 char *s; 1410 1411 PERL_ARGS_ASSERT_SV_GROW_FRESH; 1412 1413 assert(SvTYPE(sv) >= SVt_PV && SvTYPE(sv) <= SVt_PVMG); 1414 assert(!SvROK(sv)); 1415 assert(!SvOOK(sv)); 1416 assert(!SvIsCOW(sv)); 1417 assert(!SvLEN(sv)); 1418 assert(!SvCUR(sv)); 1419 1420#ifdef PERL_COPY_ON_WRITE 1421 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare) 1422 * to store the COW count. So in general, allocate one more byte than 1423 * asked for, to make it likely this byte is always spare: and thus 1424 * make more strings COW-able. 1425 * 1426 * Only increment if the allocation isn't MEM_SIZE_MAX, 1427 * otherwise it will wrap to 0. 1428 */ 1429 if ( newlen != MEM_SIZE_MAX ) 1430 newlen++; 1431#endif 1432 1433 if (newlen < PERL_STRLEN_NEW_MIN) 1434 newlen = PERL_STRLEN_NEW_MIN; 1435 1436 s = (char*)safemalloc(newlen); 1437 SvPV_set(sv, s); 1438 1439 /* No PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC here, since many strings */ 1440 /* will never be grown once set. Let the real sv_grow worry about that. */ 1441 SvLEN_set(sv, newlen); 1442 return s; 1443} 1444 1445/* 1446=for apidoc sv_setiv 1447=for apidoc_item sv_setiv_mg 1448 1449These copy an integer into the given SV, upgrading first if necessary. 1450 1451They differ only in that C<sv_setiv_mg> handles 'set' magic; C<sv_setiv> does 1452not. 1453 1454=cut 1455*/ 1456 1457void 1458Perl_sv_setiv(pTHX_ SV *const sv, const IV i) 1459{ 1460 PERL_ARGS_ASSERT_SV_SETIV; 1461 1462 SV_CHECK_THINKFIRST_COW_DROP(sv); 1463 switch (SvTYPE(sv)) { 1464#if NVSIZE <= IVSIZE 1465 case SVt_NULL: 1466 case SVt_NV: 1467 SET_SVANY_FOR_BODYLESS_IV(sv); 1468 SvFLAGS(sv) &= ~SVTYPEMASK; 1469 SvFLAGS(sv) |= SVt_IV; 1470 break; 1471#else 1472 case SVt_NULL: 1473 SET_SVANY_FOR_BODYLESS_IV(sv); 1474 SvFLAGS(sv) &= ~SVTYPEMASK; 1475 SvFLAGS(sv) |= SVt_IV; 1476 break; 1477 case SVt_NV: 1478 sv_upgrade(sv, SVt_IV); 1479 break; 1480#endif 1481 case SVt_PV: 1482 sv_upgrade(sv, SVt_PVIV); 1483 break; 1484 1485 case SVt_PVGV: 1486 if (!isGV_with_GP(sv)) 1487 break; 1488 /* FALLTHROUGH */ 1489 case SVt_PVAV: 1490 case SVt_PVHV: 1491 case SVt_PVCV: 1492 case SVt_PVFM: 1493 case SVt_PVIO: 1494 /* diag_listed_as: Can't coerce %s to %s in %s */ 1495 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0), 1496 OP_DESC(PL_op)); 1497 NOT_REACHED; /* NOTREACHED */ 1498 break; 1499 default: NOOP; 1500 } 1501 (void)SvIOK_only(sv); /* validate number */ 1502 SvIV_set(sv, i); 1503 SvTAINT(sv); 1504} 1505 1506void 1507Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i) 1508{ 1509 PERL_ARGS_ASSERT_SV_SETIV_MG; 1510 1511 sv_setiv(sv,i); 1512 SvSETMAGIC(sv); 1513} 1514 1515/* 1516=for apidoc sv_setuv 1517=for apidoc_item sv_setuv_mg 1518 1519These copy an unsigned integer into the given SV, upgrading first if necessary. 1520 1521 1522They differ only in that C<sv_setuv_mg> handles 'set' magic; C<sv_setuv> does 1523not. 1524 1525=cut 1526*/ 1527 1528void 1529Perl_sv_setuv(pTHX_ SV *const sv, const UV u) 1530{ 1531 PERL_ARGS_ASSERT_SV_SETUV; 1532 1533 /* With the if statement to ensure that integers are stored as IVs whenever 1534 possible: 1535 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865 1536 1537 without 1538 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865 1539 1540 If you wish to remove the following if statement, so that this routine 1541 (and its callers) always return UVs, please benchmark to see what the 1542 effect is. Modern CPUs may be different. Or may not :-) 1543 */ 1544 if (u <= (UV)IV_MAX) { 1545 sv_setiv(sv, (IV)u); 1546 return; 1547 } 1548 sv_setiv(sv, 0); 1549 SvIsUV_on(sv); 1550 SvUV_set(sv, u); 1551} 1552 1553void 1554Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u) 1555{ 1556 PERL_ARGS_ASSERT_SV_SETUV_MG; 1557 1558 sv_setuv(sv,u); 1559 SvSETMAGIC(sv); 1560} 1561 1562/* 1563=for apidoc sv_setnv 1564=for apidoc_item sv_setnv_mg 1565 1566These copy a double into the given SV, upgrading first if necessary. 1567 1568They differ only in that C<sv_setnv_mg> handles 'set' magic; C<sv_setnv> does 1569not. 1570 1571=cut 1572*/ 1573 1574void 1575Perl_sv_setnv(pTHX_ SV *const sv, const NV num) 1576{ 1577 PERL_ARGS_ASSERT_SV_SETNV; 1578 1579 SV_CHECK_THINKFIRST_COW_DROP(sv); 1580 switch (SvTYPE(sv)) { 1581 case SVt_NULL: 1582 case SVt_IV: 1583#if NVSIZE <= IVSIZE 1584 SET_SVANY_FOR_BODYLESS_NV(sv); 1585 SvFLAGS(sv) &= ~SVTYPEMASK; 1586 SvFLAGS(sv) |= SVt_NV; 1587 break; 1588#else 1589 sv_upgrade(sv, SVt_NV); 1590 break; 1591#endif 1592 case SVt_PV: 1593 case SVt_PVIV: 1594 sv_upgrade(sv, SVt_PVNV); 1595 break; 1596 1597 case SVt_PVGV: 1598 if (!isGV_with_GP(sv)) 1599 break; 1600 /* FALLTHROUGH */ 1601 case SVt_PVAV: 1602 case SVt_PVHV: 1603 case SVt_PVCV: 1604 case SVt_PVFM: 1605 case SVt_PVIO: 1606 /* diag_listed_as: Can't coerce %s to %s in %s */ 1607 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0), 1608 OP_DESC(PL_op)); 1609 NOT_REACHED; /* NOTREACHED */ 1610 break; 1611 default: NOOP; 1612 } 1613 SvNV_set(sv, num); 1614 (void)SvNOK_only(sv); /* validate number */ 1615 SvTAINT(sv); 1616} 1617 1618void 1619Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num) 1620{ 1621 PERL_ARGS_ASSERT_SV_SETNV_MG; 1622 1623 sv_setnv(sv,num); 1624 SvSETMAGIC(sv); 1625} 1626 1627/* 1628=for apidoc sv_setrv_noinc 1629=for apidoc_item sv_setrv_noinc_mg 1630 1631Copies an SV pointer into the given SV as an SV reference, upgrading it if 1632necessary. After this, C<SvRV(sv)> is equal to I<ref>. This does not adjust 1633the reference count of I<ref>. The reference I<ref> must not be NULL. 1634 1635C<sv_setrv_noinc_mg> will invoke 'set' magic on the SV; C<sv_setrv_noinc> will 1636not. 1637 1638=cut 1639*/ 1640 1641void 1642Perl_sv_setrv_noinc(pTHX_ SV *const sv, SV *const ref) 1643{ 1644 PERL_ARGS_ASSERT_SV_SETRV_NOINC; 1645 1646 SV_CHECK_THINKFIRST_COW_DROP(sv); 1647 prepare_SV_for_RV(sv); 1648 1649 SvOK_off(sv); 1650 SvRV_set(sv, ref); 1651 SvROK_on(sv); 1652} 1653 1654void 1655Perl_sv_setrv_noinc_mg(pTHX_ SV *const sv, SV *const ref) 1656{ 1657 PERL_ARGS_ASSERT_SV_SETRV_NOINC_MG; 1658 1659 sv_setrv_noinc(sv, ref); 1660 SvSETMAGIC(sv); 1661} 1662 1663/* 1664=for apidoc sv_setrv_inc 1665=for apidoc_item sv_setrv_inc_mg 1666 1667As C<sv_setrv_noinc> but increments the reference count of I<ref>. 1668 1669C<sv_setrv_inc_mg> will invoke 'set' magic on the SV; C<sv_setrv_inc> will 1670not. 1671 1672=cut 1673*/ 1674 1675void 1676Perl_sv_setrv_inc(pTHX_ SV *const sv, SV *const ref) 1677{ 1678 PERL_ARGS_ASSERT_SV_SETRV_INC; 1679 1680 sv_setrv_noinc(sv, SvREFCNT_inc_simple_NN(ref)); 1681} 1682 1683void 1684Perl_sv_setrv_inc_mg(pTHX_ SV *const sv, SV *const ref) 1685{ 1686 PERL_ARGS_ASSERT_SV_SETRV_INC_MG; 1687 1688 sv_setrv_noinc(sv, SvREFCNT_inc_simple_NN(ref)); 1689 SvSETMAGIC(sv); 1690} 1691 1692/* Return a cleaned-up, printable version of sv, for non-numeric, or 1693 * not incrementable warning display. 1694 * Originally part of S_not_a_number(). 1695 * The return value may be != tmpbuf. 1696 */ 1697 1698STATIC const char * 1699S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) { 1700 const char *pv; 1701 1702 PERL_ARGS_ASSERT_SV_DISPLAY; 1703 1704 if (DO_UTF8(sv)) { 1705 SV *dsv = newSVpvs_flags("", SVs_TEMP); 1706 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT); 1707 } else { 1708 char *d = tmpbuf; 1709 const char * const limit = tmpbuf + tmpbuf_size - 8; 1710 /* each *s can expand to 4 chars + "...\0", 1711 i.e. need room for 8 chars */ 1712 1713 const char *s = SvPVX_const(sv); 1714 const char * const end = s + SvCUR(sv); 1715 for ( ; s < end && d < limit; s++ ) { 1716 int ch = (U8) *s; 1717 if (! isASCII(ch) && !isPRINT_LC(ch)) { 1718 *d++ = 'M'; 1719 *d++ = '-'; 1720 1721 /* Map to ASCII "equivalent" of Latin1 */ 1722 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127); 1723 } 1724 if (ch == '\n') { 1725 *d++ = '\\'; 1726 *d++ = 'n'; 1727 } 1728 else if (ch == '\r') { 1729 *d++ = '\\'; 1730 *d++ = 'r'; 1731 } 1732 else if (ch == '\f') { 1733 *d++ = '\\'; 1734 *d++ = 'f'; 1735 } 1736 else if (ch == '\\') { 1737 *d++ = '\\'; 1738 *d++ = '\\'; 1739 } 1740 else if (ch == '\0') { 1741 *d++ = '\\'; 1742 *d++ = '0'; 1743 } 1744 else if (isPRINT_LC(ch)) 1745 *d++ = ch; 1746 else { 1747 *d++ = '^'; 1748 *d++ = toCTRL(ch); 1749 } 1750 } 1751 if (s < end) { 1752 *d++ = '.'; 1753 *d++ = '.'; 1754 *d++ = '.'; 1755 } 1756 *d = '\0'; 1757 pv = tmpbuf; 1758 } 1759 1760 return pv; 1761} 1762 1763/* Print an "isn't numeric" warning, using a cleaned-up, 1764 * printable version of the offending string 1765 */ 1766 1767STATIC void 1768S_not_a_number(pTHX_ SV *const sv) 1769{ 1770 char tmpbuf[64]; 1771 const char *pv; 1772 1773 PERL_ARGS_ASSERT_NOT_A_NUMBER; 1774 1775 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf)); 1776 1777 if (PL_op) 1778 Perl_warner(aTHX_ packWARN(WARN_NUMERIC), 1779 /* diag_listed_as: Argument "%s" isn't numeric%s */ 1780 "Argument \"%s\" isn't numeric in %s", pv, 1781 OP_DESC(PL_op)); 1782 else 1783 Perl_warner(aTHX_ packWARN(WARN_NUMERIC), 1784 /* diag_listed_as: Argument "%s" isn't numeric%s */ 1785 "Argument \"%s\" isn't numeric", pv); 1786} 1787 1788STATIC void 1789S_not_incrementable(pTHX_ SV *const sv) { 1790 char tmpbuf[64]; 1791 const char *pv; 1792 1793 PERL_ARGS_ASSERT_NOT_INCREMENTABLE; 1794 1795 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf)); 1796 1797 Perl_warner(aTHX_ packWARN(WARN_NUMERIC), 1798 "Argument \"%s\" treated as 0 in increment (++)", pv); 1799} 1800 1801/* 1802=for apidoc looks_like_number 1803 1804Test if the content of an SV looks like a number (or is a number). 1805C<Inf> and C<Infinity> are treated as numbers (so will not issue a 1806non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is 1807ignored. 1808 1809=cut 1810*/ 1811 1812I32 1813Perl_looks_like_number(pTHX_ SV *const sv) 1814{ 1815 const char *sbegin; 1816 STRLEN len; 1817 int numtype; 1818 1819 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER; 1820 1821 if (SvPOK(sv) || SvPOKp(sv)) { 1822 sbegin = SvPV_nomg_const(sv, len); 1823 } 1824 else 1825 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK); 1826 numtype = grok_number(sbegin, len, NULL); 1827 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype; 1828} 1829 1830STATIC bool 1831S_glob_2number(pTHX_ GV * const gv) 1832{ 1833 PERL_ARGS_ASSERT_GLOB_2NUMBER; 1834 1835 /* We know that all GVs stringify to something that is not-a-number, 1836 so no need to test that. */ 1837 if (ckWARN(WARN_NUMERIC)) 1838 { 1839 SV *const buffer = sv_newmortal(); 1840 gv_efullname3(buffer, gv, "*"); 1841 not_a_number(buffer); 1842 } 1843 /* We just want something true to return, so that S_sv_2iuv_common 1844 can tail call us and return true. */ 1845 return TRUE; 1846} 1847 1848/* Actually, ISO C leaves conversion of UV to IV undefined, but 1849 until proven guilty, assume that things are not that bad... */ 1850 1851/* 1852 NV_PRESERVES_UV: 1853 1854 As 64 bit platforms often have an NV that doesn't preserve all bits of 1855 an IV (an assumption perl has been based on to date) it becomes necessary 1856 to remove the assumption that the NV always carries enough precision to 1857 recreate the IV whenever needed, and that the NV is the canonical form. 1858 Instead, IV/UV and NV need to be given equal rights. So as to not lose 1859 precision as a side effect of conversion (which would lead to insanity 1860 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is 1861 1) to distinguish between IV/UV/NV slots that have a valid conversion cached 1862 where precision was lost, and IV/UV/NV slots that have a valid conversion 1863 which has lost no precision 1864 2) to ensure that if a numeric conversion to one form is requested that 1865 would lose precision, the precise conversion (or differently 1866 imprecise conversion) is also performed and cached, to prevent 1867 requests for different numeric formats on the same SV causing 1868 lossy conversion chains. (lossless conversion chains are perfectly 1869 acceptable (still)) 1870 1871 1872 flags are used: 1873 SvIOKp is true if the IV slot contains a valid value 1874 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true) 1875 SvNOKp is true if the NV slot contains a valid value 1876 SvNOK is true only if the NV value is accurate 1877 1878 so 1879 while converting from PV to NV, check to see if converting that NV to an 1880 IV(or UV) would lose accuracy over a direct conversion from PV to 1881 IV(or UV). If it would, cache both conversions, return NV, but mark 1882 SV as IOK NOKp (ie not NOK). 1883 1884 While converting from PV to IV, check to see if converting that IV to an 1885 NV would lose accuracy over a direct conversion from PV to NV. If it 1886 would, cache both conversions, flag similarly. 1887 1888 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite 1889 correctly because if IV & NV were set NV *always* overruled. 1890 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning 1891 changes - now IV and NV together means that the two are interchangeable: 1892 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX; 1893 1894 The benefit of this is that operations such as pp_add know that if 1895 SvIOK is true for both left and right operands, then integer addition 1896 can be used instead of floating point (for cases where the result won't 1897 overflow). Before, floating point was always used, which could lead to 1898 loss of precision compared with integer addition. 1899 1900 * making IV and NV equal status should make maths accurate on 64 bit 1901 platforms 1902 * may speed up maths somewhat if pp_add and friends start to use 1903 integers when possible instead of fp. (Hopefully the overhead in 1904 looking for SvIOK and checking for overflow will not outweigh the 1905 fp to integer speedup) 1906 * will slow down integer operations (callers of SvIV) on "inaccurate" 1907 values, as the change from SvIOK to SvIOKp will cause a call into 1908 sv_2iv each time rather than a macro access direct to the IV slot 1909 * should speed up number->string conversion on integers as IV is 1910 favoured when IV and NV are equally accurate 1911 1912 #################################################################### 1913 You had better be using SvIOK_notUV if you want an IV for arithmetic: 1914 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV. 1915 On the other hand, SvUOK is true iff UV. 1916 #################################################################### 1917 1918 Your mileage will vary depending your CPU's relative fp to integer 1919 performance ratio. 1920*/ 1921 1922#ifndef NV_PRESERVES_UV 1923# define IS_NUMBER_UNDERFLOW_IV 1 1924# define IS_NUMBER_UNDERFLOW_UV 2 1925# define IS_NUMBER_IV_AND_UV 2 1926# define IS_NUMBER_OVERFLOW_IV 4 1927# define IS_NUMBER_OVERFLOW_UV 5 1928 1929/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */ 1930 1931/* For sv_2nv these three cases are "SvNOK and don't bother casting" */ 1932STATIC int 1933S_sv_2iuv_non_preserve(pTHX_ SV *const sv 1934# ifdef DEBUGGING 1935 , I32 numtype 1936# endif 1937 ) 1938{ 1939 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE; 1940 PERL_UNUSED_CONTEXT; 1941 1942 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%" UVxf " NV=%" NVgf " inttype=%" UVXf "\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype)); 1943 if (SvNVX(sv) < (NV)IV_MIN) { 1944 (void)SvIOKp_on(sv); 1945 (void)SvNOK_on(sv); 1946 SvIV_set(sv, IV_MIN); 1947 return IS_NUMBER_UNDERFLOW_IV; 1948 } 1949 if (SvNVX(sv) > (NV)UV_MAX) { 1950 (void)SvIOKp_on(sv); 1951 (void)SvNOK_on(sv); 1952 SvIsUV_on(sv); 1953 SvUV_set(sv, UV_MAX); 1954 return IS_NUMBER_OVERFLOW_UV; 1955 } 1956 (void)SvIOKp_on(sv); 1957 (void)SvNOK_on(sv); 1958 /* Can't use strtol etc to convert this string. (See truth table in 1959 sv_2iv */ 1960 if (SvNVX(sv) < IV_MAX_P1) { 1961 SvIV_set(sv, I_V(SvNVX(sv))); 1962 if ((NV)(SvIVX(sv)) == SvNVX(sv)) { 1963 SvIOK_on(sv); /* Integer is precise. NOK, IOK */ 1964 } else { 1965 /* Integer is imprecise. NOK, IOKp */ 1966 } 1967 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV; 1968 } 1969 SvIsUV_on(sv); 1970 SvUV_set(sv, U_V(SvNVX(sv))); 1971 if ((NV)(SvUVX(sv)) == SvNVX(sv)) { 1972 if (SvUVX(sv) == UV_MAX) { 1973 /* As we know that NVs don't preserve UVs, UV_MAX cannot 1974 possibly be preserved by NV. Hence, it must be overflow. 1975 NOK, IOKp */ 1976 return IS_NUMBER_OVERFLOW_UV; 1977 } 1978 SvIOK_on(sv); /* Integer is precise. NOK, UOK */ 1979 } else { 1980 /* Integer is imprecise. NOK, IOKp */ 1981 } 1982 return IS_NUMBER_OVERFLOW_IV; 1983} 1984#endif /* !NV_PRESERVES_UV*/ 1985 1986/* If numtype is infnan, set the NV of the sv accordingly. 1987 * If numtype is anything else, try setting the NV using Atof(PV). */ 1988static void 1989S_sv_setnv(pTHX_ SV* sv, int numtype) 1990{ 1991 bool pok = cBOOL(SvPOK(sv)); 1992 bool nok = FALSE; 1993#ifdef NV_INF 1994 if ((numtype & IS_NUMBER_INFINITY)) { 1995 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF); 1996 nok = TRUE; 1997 } else 1998#endif 1999#ifdef NV_NAN 2000 if ((numtype & IS_NUMBER_NAN)) { 2001 SvNV_set(sv, NV_NAN); 2002 nok = TRUE; 2003 } else 2004#endif 2005 if (pok) { 2006 SvNV_set(sv, Atof(SvPVX_const(sv))); 2007 /* Purposefully no true nok here, since we don't want to blow 2008 * away the possible IOK/UV of an existing sv. */ 2009 } 2010 if (nok) { 2011 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */ 2012 if (pok) 2013 SvPOK_on(sv); /* PV is okay, though. */ 2014 } 2015} 2016 2017STATIC bool 2018S_sv_2iuv_common(pTHX_ SV *const sv) 2019{ 2020 PERL_ARGS_ASSERT_SV_2IUV_COMMON; 2021 2022 if (SvNOKp(sv)) { 2023 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv 2024 * without also getting a cached IV/UV from it at the same time 2025 * (ie PV->NV conversion should detect loss of accuracy and cache 2026 * IV or UV at same time to avoid this. */ 2027 /* IV-over-UV optimisation - choose to cache IV if possible */ 2028 2029 if (SvTYPE(sv) == SVt_NV) 2030 sv_upgrade(sv, SVt_PVNV); 2031 2032 got_nv: 2033 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */ 2034 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost 2035 certainly cast into the IV range at IV_MAX, whereas the correct 2036 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary 2037 cases go to UV */ 2038#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan) 2039 if (Perl_isnan(SvNVX(sv))) { 2040 SvUV_set(sv, 0); 2041 SvIsUV_on(sv); 2042 return FALSE; 2043 } 2044#endif 2045 if (SvNVX(sv) < (NV)IV_MAX + 0.5) { 2046 SvIV_set(sv, I_V(SvNVX(sv))); 2047 if (SvNVX(sv) == (NV) SvIVX(sv) 2048#ifndef NV_PRESERVES_UV 2049 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */ 2050 && (((UV)1 << NV_PRESERVES_UV_BITS) > 2051 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv))) 2052 /* Don't flag it as "accurately an integer" if the number 2053 came from a (by definition imprecise) NV operation, and 2054 we're outside the range of NV integer precision */ 2055#endif 2056 ) { 2057 if (SvNOK(sv)) 2058 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */ 2059 else { 2060 /* scalar has trailing garbage, eg "42a" */ 2061 } 2062 DEBUG_c(PerlIO_printf(Perl_debug_log, 2063 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n", 2064 PTR2UV(sv), 2065 SvNVX(sv), 2066 SvIVX(sv))); 2067 2068 } else { 2069 /* IV not precise. No need to convert from PV, as NV 2070 conversion would already have cached IV if it detected 2071 that PV->IV would be better than PV->NV->IV 2072 flags already correct - don't set public IOK. */ 2073 DEBUG_c(PerlIO_printf(Perl_debug_log, 2074 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n", 2075 PTR2UV(sv), 2076 SvNVX(sv), 2077 SvIVX(sv))); 2078 } 2079 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN, 2080 but the cast (NV)IV_MIN rounds to a the value less (more 2081 negative) than IV_MIN which happens to be equal to SvNVX ?? 2082 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and 2083 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and 2084 (NV)UVX == NVX are both true, but the values differ. :-( 2085 Hopefully for 2s complement IV_MIN is something like 2086 0x8000000000000000 which will be exact. NWC */ 2087 } 2088 else { 2089 SvUV_set(sv, U_V(SvNVX(sv))); 2090 if ( 2091 (SvNVX(sv) == (NV) SvUVX(sv)) 2092#ifndef NV_PRESERVES_UV 2093 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */ 2094 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */ 2095 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv)) 2096 /* Don't flag it as "accurately an integer" if the number 2097 came from a (by definition imprecise) NV operation, and 2098 we're outside the range of NV integer precision */ 2099#endif 2100 && SvNOK(sv) 2101 ) 2102 SvIOK_on(sv); 2103 SvIsUV_on(sv); 2104 DEBUG_c(PerlIO_printf(Perl_debug_log, 2105 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n", 2106 PTR2UV(sv), 2107 SvUVX(sv), 2108 SvUVX(sv))); 2109 } 2110 } 2111 else if (SvPOKp(sv)) { 2112 UV value; 2113 int numtype; 2114 const char *s = SvPVX_const(sv); 2115 const STRLEN cur = SvCUR(sv); 2116 2117 /* short-cut for a single digit string like "1" */ 2118 2119 if (cur == 1) { 2120 char c = *s; 2121 if (isDIGIT(c)) { 2122 if (SvTYPE(sv) < SVt_PVIV) 2123 sv_upgrade(sv, SVt_PVIV); 2124 (void)SvIOK_on(sv); 2125 SvIV_set(sv, (IV)(c - '0')); 2126 return FALSE; 2127 } 2128 } 2129 2130 numtype = grok_number(s, cur, &value); 2131 /* We want to avoid a possible problem when we cache an IV/ a UV which 2132 may be later translated to an NV, and the resulting NV is not 2133 the same as the direct translation of the initial string 2134 (eg 123.456 can shortcut to the IV 123 with atol(), but we must 2135 be careful to ensure that the value with the .456 is around if the 2136 NV value is requested in the future). 2137 2138 This means that if we cache such an IV/a UV, we need to cache the 2139 NV as well. Moreover, we trade speed for space, and do not 2140 cache the NV if we are sure it's not needed. 2141 */ 2142 2143 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */ 2144 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2145 == IS_NUMBER_IN_UV) { 2146 /* It's definitely an integer, only upgrade to PVIV */ 2147 if (SvTYPE(sv) < SVt_PVIV) 2148 sv_upgrade(sv, SVt_PVIV); 2149 (void)SvIOK_on(sv); 2150 } else if (SvTYPE(sv) < SVt_PVNV) 2151 sv_upgrade(sv, SVt_PVNV); 2152 2153 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) { 2154 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING))) 2155 not_a_number(sv); 2156 S_sv_setnv(aTHX_ sv, numtype); 2157 goto got_nv; /* Fill IV/UV slot and set IOKp */ 2158 } 2159 2160 /* If NVs preserve UVs then we only use the UV value if we know that 2161 we aren't going to call atof() below. If NVs don't preserve UVs 2162 then the value returned may have more precision than atof() will 2163 return, even though value isn't perfectly accurate. */ 2164 if ((numtype & (IS_NUMBER_IN_UV 2165#ifdef NV_PRESERVES_UV 2166 | IS_NUMBER_NOT_INT 2167#endif 2168 )) == IS_NUMBER_IN_UV) { 2169 /* This won't turn off the public IOK flag if it was set above */ 2170 (void)SvIOKp_on(sv); 2171 2172 if (!(numtype & IS_NUMBER_NEG)) { 2173 /* positive */; 2174 if (value <= (UV)IV_MAX) { 2175 SvIV_set(sv, (IV)value); 2176 } else { 2177 /* it didn't overflow, and it was positive. */ 2178 SvUV_set(sv, value); 2179 SvIsUV_on(sv); 2180 } 2181 } else { 2182 /* 2s complement assumption */ 2183 if (value <= (UV)IV_MIN) { 2184 SvIV_set(sv, value == (UV)IV_MIN 2185 ? IV_MIN : -(IV)value); 2186 } else { 2187 /* Too negative for an IV. This is a double upgrade, but 2188 I'm assuming it will be rare. */ 2189 if (SvTYPE(sv) < SVt_PVNV) 2190 sv_upgrade(sv, SVt_PVNV); 2191 SvNOK_on(sv); 2192 SvIOK_off(sv); 2193 SvIOKp_on(sv); 2194 SvNV_set(sv, -(NV)value); 2195 SvIV_set(sv, IV_MIN); 2196 } 2197 } 2198 } 2199 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we 2200 will be in the previous block to set the IV slot, and the next 2201 block to set the NV slot. So no else here. */ 2202 2203 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2204 != IS_NUMBER_IN_UV) { 2205 /* It wasn't an (integer that doesn't overflow the UV). */ 2206 S_sv_setnv(aTHX_ sv, numtype); 2207 2208 if (! numtype && ckWARN(WARN_NUMERIC)) 2209 not_a_number(sv); 2210 2211 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n", 2212 PTR2UV(sv), SvNVX(sv))); 2213 2214#ifdef NV_PRESERVES_UV 2215 SvNOKp_on(sv); 2216 if (numtype) 2217 SvNOK_on(sv); 2218 goto got_nv; /* Fill IV/UV slot and set IOKp, maybe IOK */ 2219#else /* NV_PRESERVES_UV */ 2220 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2221 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) { 2222 /* The IV/UV slot will have been set from value returned by 2223 grok_number above. The NV slot has just been set using 2224 Atof. */ 2225 SvNOK_on(sv); 2226 assert (SvIOKp(sv)); 2227 } else { 2228 if (((UV)1 << NV_PRESERVES_UV_BITS) > 2229 U_V(Perl_fabs(SvNVX(sv)))) { 2230 /* Small enough to preserve all bits. */ 2231 (void)SvIOKp_on(sv); 2232 SvNOK_on(sv); 2233 SvIV_set(sv, I_V(SvNVX(sv))); 2234 if ((NV)(SvIVX(sv)) == SvNVX(sv)) 2235 SvIOK_on(sv); 2236 /* There had been runtime checking for 2237 "U_V(Perl_fabs(SvNVX(sv))) < (UV)IV_MAX" here to ensure 2238 that this NV is in the preserved range, but this should 2239 be always true if the following assertion is true: */ 2240 STATIC_ASSERT_STMT(((UV)1 << NV_PRESERVES_UV_BITS) <= 2241 (UV)IV_MAX); 2242 } else { 2243 /* IN_UV NOT_INT 2244 0 0 already failed to read UV. 2245 0 1 already failed to read UV. 2246 1 0 you won't get here in this case. IV/UV 2247 slot set, public IOK, Atof() unneeded. 2248 1 1 already read UV. 2249 so there's no point in sv_2iuv_non_preserve() attempting 2250 to use atol, strtol, strtoul etc. */ 2251# ifdef DEBUGGING 2252 sv_2iuv_non_preserve (sv, numtype); 2253# else 2254 sv_2iuv_non_preserve (sv); 2255# endif 2256 } 2257 } 2258 /* It might be more code efficient to go through the entire logic above 2259 and conditionally set with SvIOKp_on() rather than SvIOK(), but it 2260 gets complex and potentially buggy, so more programmer efficient 2261 to do it this way, by turning off the public flags: */ 2262 if (!numtype) 2263 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK); 2264#endif /* NV_PRESERVES_UV */ 2265 } 2266 } 2267 else { 2268 if (isGV_with_GP(sv)) 2269 return glob_2number(MUTABLE_GV(sv)); 2270 2271 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) 2272 report_uninit(sv); 2273 if (SvTYPE(sv) < SVt_IV) 2274 /* Typically the caller expects that sv_any is not NULL now. */ 2275 sv_upgrade(sv, SVt_IV); 2276 /* Return 0 from the caller. */ 2277 return TRUE; 2278 } 2279 return FALSE; 2280} 2281 2282/* 2283=for apidoc sv_2iv_flags 2284 2285Return the integer value of an SV, doing any necessary string 2286conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. 2287Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros. 2288 2289=cut 2290*/ 2291 2292IV 2293Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags) 2294{ 2295 PERL_ARGS_ASSERT_SV_2IV_FLAGS; 2296 2297 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV 2298 && SvTYPE(sv) != SVt_PVFM); 2299 2300 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) 2301 mg_get(sv); 2302 2303 if (SvROK(sv)) { 2304 if (SvAMAGIC(sv)) { 2305 SV * tmpstr; 2306 if (flags & SV_SKIP_OVERLOAD) 2307 return 0; 2308 tmpstr = AMG_CALLunary(sv, numer_amg); 2309 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { 2310 return SvIV(tmpstr); 2311 } 2312 } 2313 return PTR2IV(SvRV(sv)); 2314 } 2315 2316 if (SvVALID(sv) || isREGEXP(sv)) { 2317 /* FBMs use the space for SvIVX and SvNVX for other purposes, so 2318 must not let them cache IVs. 2319 In practice they are extremely unlikely to actually get anywhere 2320 accessible by user Perl code - the only way that I'm aware of is when 2321 a constant subroutine which is used as the second argument to index. 2322 2323 Regexps have no SvIVX and SvNVX fields. 2324 */ 2325 assert(SvPOKp(sv)); 2326 { 2327 UV value; 2328 const char * const ptr = 2329 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv); 2330 const int numtype 2331 = grok_number(ptr, SvCUR(sv), &value); 2332 2333 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2334 == IS_NUMBER_IN_UV) { 2335 /* It's definitely an integer */ 2336 if (numtype & IS_NUMBER_NEG) { 2337 if (value < (UV)IV_MIN) 2338 return -(IV)value; 2339 } else { 2340 if (value < (UV)IV_MAX) 2341 return (IV)value; 2342 } 2343 } 2344 2345 /* Quite wrong but no good choices. */ 2346 if ((numtype & IS_NUMBER_INFINITY)) { 2347 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX; 2348 } else if ((numtype & IS_NUMBER_NAN)) { 2349 return 0; /* So wrong. */ 2350 } 2351 2352 if (!numtype) { 2353 if (ckWARN(WARN_NUMERIC)) 2354 not_a_number(sv); 2355 } 2356 return I_V(Atof(ptr)); 2357 } 2358 } 2359 2360 if (SvTHINKFIRST(sv)) { 2361 if (SvREADONLY(sv) && !SvOK(sv)) { 2362 if (ckWARN(WARN_UNINITIALIZED)) 2363 report_uninit(sv); 2364 return 0; 2365 } 2366 } 2367 2368 if (!SvIOKp(sv)) { 2369 if (S_sv_2iuv_common(aTHX_ sv)) 2370 return 0; 2371 } 2372 2373 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n", 2374 PTR2UV(sv),SvIVX(sv))); 2375 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv); 2376} 2377 2378/* 2379=for apidoc sv_2uv_flags 2380 2381Return the unsigned integer value of an SV, doing any necessary string 2382conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. 2383Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros. 2384 2385=for apidoc Amnh||SV_GMAGIC 2386 2387=cut 2388*/ 2389 2390UV 2391Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags) 2392{ 2393 PERL_ARGS_ASSERT_SV_2UV_FLAGS; 2394 2395 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) 2396 mg_get(sv); 2397 2398 if (SvROK(sv)) { 2399 if (SvAMAGIC(sv)) { 2400 SV *tmpstr; 2401 if (flags & SV_SKIP_OVERLOAD) 2402 return 0; 2403 tmpstr = AMG_CALLunary(sv, numer_amg); 2404 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { 2405 return SvUV(tmpstr); 2406 } 2407 } 2408 return PTR2UV(SvRV(sv)); 2409 } 2410 2411 if (SvVALID(sv) || isREGEXP(sv)) { 2412 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use 2413 the same flag bit as SVf_IVisUV, so must not let them cache IVs. 2414 Regexps have no SvIVX and SvNVX fields. */ 2415 assert(SvPOKp(sv)); 2416 { 2417 UV value; 2418 const char * const ptr = 2419 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv); 2420 const int numtype 2421 = grok_number(ptr, SvCUR(sv), &value); 2422 2423 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2424 == IS_NUMBER_IN_UV) { 2425 /* It's definitely an integer */ 2426 if (!(numtype & IS_NUMBER_NEG)) 2427 return value; 2428 } 2429 2430 /* Quite wrong but no good choices. */ 2431 if ((numtype & IS_NUMBER_INFINITY)) { 2432 return UV_MAX; /* So wrong. */ 2433 } else if ((numtype & IS_NUMBER_NAN)) { 2434 return 0; /* So wrong. */ 2435 } 2436 2437 if (!numtype) { 2438 if (ckWARN(WARN_NUMERIC)) 2439 not_a_number(sv); 2440 } 2441 return U_V(Atof(ptr)); 2442 } 2443 } 2444 2445 if (SvTHINKFIRST(sv)) { 2446 if (SvREADONLY(sv) && !SvOK(sv)) { 2447 if (ckWARN(WARN_UNINITIALIZED)) 2448 report_uninit(sv); 2449 return 0; 2450 } 2451 } 2452 2453 if (!SvIOKp(sv)) { 2454 if (S_sv_2iuv_common(aTHX_ sv)) 2455 return 0; 2456 } 2457 2458 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n", 2459 PTR2UV(sv),SvUVX(sv))); 2460 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv); 2461} 2462 2463/* 2464=for apidoc sv_2nv_flags 2465 2466Return the num value of an SV, doing any necessary string or integer 2467conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. 2468Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros. 2469 2470=cut 2471*/ 2472 2473NV 2474Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags) 2475{ 2476 PERL_ARGS_ASSERT_SV_2NV_FLAGS; 2477 2478 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV 2479 && SvTYPE(sv) != SVt_PVFM); 2480 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) { 2481 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use 2482 the same flag bit as SVf_IVisUV, so must not let them cache NVs. 2483 Regexps have no SvIVX and SvNVX fields. */ 2484 const char *ptr; 2485 if (flags & SV_GMAGIC) 2486 mg_get(sv); 2487 if (SvNOKp(sv)) 2488 return SvNVX(sv); 2489 if (SvPOKp(sv) && !SvIOKp(sv)) { 2490 ptr = SvPVX_const(sv); 2491 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) && 2492 !grok_number(ptr, SvCUR(sv), NULL)) 2493 not_a_number(sv); 2494 return Atof(ptr); 2495 } 2496 if (SvIOKp(sv)) { 2497 if (SvIsUV(sv)) 2498 return (NV)SvUVX(sv); 2499 else 2500 return (NV)SvIVX(sv); 2501 } 2502 if (SvROK(sv)) { 2503 goto return_rok; 2504 } 2505 assert(SvTYPE(sv) >= SVt_PVMG); 2506 /* This falls through to the report_uninit near the end of the 2507 function. */ 2508 } else if (SvTHINKFIRST(sv)) { 2509 if (SvROK(sv)) { 2510 return_rok: 2511 if (SvAMAGIC(sv)) { 2512 SV *tmpstr; 2513 if (flags & SV_SKIP_OVERLOAD) 2514 return 0; 2515 tmpstr = AMG_CALLunary(sv, numer_amg); 2516 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { 2517 return SvNV(tmpstr); 2518 } 2519 } 2520 return PTR2NV(SvRV(sv)); 2521 } 2522 if (SvREADONLY(sv) && !SvOK(sv)) { 2523 if (ckWARN(WARN_UNINITIALIZED)) 2524 report_uninit(sv); 2525 return 0.0; 2526 } 2527 } 2528 if (SvTYPE(sv) < SVt_NV) { 2529 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */ 2530 sv_upgrade(sv, SVt_NV); 2531 CLANG_DIAG_IGNORE_STMT(-Wthread-safety); 2532 DEBUG_c({ 2533 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 2534 STORE_LC_NUMERIC_SET_STANDARD(); 2535 PerlIO_printf(Perl_debug_log, 2536 "0x%" UVxf " num(%" NVgf ")\n", 2537 PTR2UV(sv), SvNVX(sv)); 2538 RESTORE_LC_NUMERIC(); 2539 }); 2540 CLANG_DIAG_RESTORE_STMT; 2541 2542 } 2543 else if (SvTYPE(sv) < SVt_PVNV) 2544 sv_upgrade(sv, SVt_PVNV); 2545 if (SvNOKp(sv)) { 2546 return SvNVX(sv); 2547 } 2548 if (SvIOKp(sv)) { 2549 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv)); 2550#ifdef NV_PRESERVES_UV 2551 if (SvIOK(sv)) 2552 SvNOK_on(sv); 2553 else 2554 SvNOKp_on(sv); 2555#else 2556 /* Only set the public NV OK flag if this NV preserves the IV */ 2557 /* Check it's not 0xFFFFFFFFFFFFFFFF */ 2558 if (SvIOK(sv) && 2559 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv)))) 2560 : (SvIVX(sv) == I_V(SvNVX(sv)))) 2561 SvNOK_on(sv); 2562 else 2563 SvNOKp_on(sv); 2564#endif 2565 } 2566 else if (SvPOKp(sv)) { 2567 UV value; 2568 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value); 2569 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC)) 2570 not_a_number(sv); 2571#ifdef NV_PRESERVES_UV 2572 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) 2573 == IS_NUMBER_IN_UV) { 2574 /* It's definitely an integer */ 2575 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value); 2576 } else { 2577 S_sv_setnv(aTHX_ sv, numtype); 2578 } 2579 if (numtype) 2580 SvNOK_on(sv); 2581 else 2582 SvNOKp_on(sv); 2583#else 2584 SvNV_set(sv, Atof(SvPVX_const(sv))); 2585 /* Only set the public NV OK flag if this NV preserves the value in 2586 the PV at least as well as an IV/UV would. 2587 Not sure how to do this 100% reliably. */ 2588 /* if that shift count is out of range then Configure's test is 2589 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS == 2590 UV_BITS */ 2591 if (((UV)1 << NV_PRESERVES_UV_BITS) > U_V(Perl_fabs(SvNVX(sv)))) { 2592 SvNOK_on(sv); /* Definitely small enough to preserve all bits */ 2593 } else if (!(numtype & IS_NUMBER_IN_UV)) { 2594 /* Can't use strtol etc to convert this string, so don't try. 2595 sv_2iv and sv_2uv will use the NV to convert, not the PV. */ 2596 SvNOK_on(sv); 2597 } else { 2598 /* value has been set. It may not be precise. */ 2599 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) { 2600 /* 2s complement assumption for (UV)IV_MIN */ 2601 SvNOK_on(sv); /* Integer is too negative. */ 2602 } else { 2603 SvNOKp_on(sv); 2604 SvIOKp_on(sv); 2605 2606 if (numtype & IS_NUMBER_NEG) { 2607 /* -IV_MIN is undefined, but we should never reach 2608 * this point with both IS_NUMBER_NEG and value == 2609 * (UV)IV_MIN */ 2610 assert(value != (UV)IV_MIN); 2611 SvIV_set(sv, -(IV)value); 2612 } else if (value <= (UV)IV_MAX) { 2613 SvIV_set(sv, (IV)value); 2614 } else { 2615 SvUV_set(sv, value); 2616 SvIsUV_on(sv); 2617 } 2618 2619 if (numtype & IS_NUMBER_NOT_INT) { 2620 /* I believe that even if the original PV had decimals, 2621 they are lost beyond the limit of the FP precision. 2622 However, neither is canonical, so both only get p 2623 flags. NWC, 2000/11/25 */ 2624 /* Both already have p flags, so do nothing */ 2625 } else { 2626 const NV nv = SvNVX(sv); 2627 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */ 2628 if (SvNVX(sv) < (NV)IV_MAX + 0.5) { 2629 if (SvIVX(sv) == I_V(nv)) { 2630 SvNOK_on(sv); 2631 } else { 2632 /* It had no "." so it must be integer. */ 2633 } 2634 SvIOK_on(sv); 2635 } else { 2636 /* between IV_MAX and NV(UV_MAX). 2637 Could be slightly > UV_MAX */ 2638 2639 if (numtype & IS_NUMBER_NOT_INT) { 2640 /* UV and NV both imprecise. */ 2641 } else { 2642 const UV nv_as_uv = U_V(nv); 2643 2644 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) { 2645 SvNOK_on(sv); 2646 } 2647 SvIOK_on(sv); 2648 } 2649 } 2650 } 2651 } 2652 } 2653 /* It might be more code efficient to go through the entire logic above 2654 and conditionally set with SvNOKp_on() rather than SvNOK(), but it 2655 gets complex and potentially buggy, so more programmer efficient 2656 to do it this way, by turning off the public flags: */ 2657 if (!numtype) 2658 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK); 2659#endif /* NV_PRESERVES_UV */ 2660 } 2661 else { 2662 if (isGV_with_GP(sv)) { 2663 glob_2number(MUTABLE_GV(sv)); 2664 return 0.0; 2665 } 2666 2667 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) 2668 report_uninit(sv); 2669 assert (SvTYPE(sv) >= SVt_NV); 2670 /* Typically the caller expects that sv_any is not NULL now. */ 2671 /* XXX Ilya implies that this is a bug in callers that assume this 2672 and ideally should be fixed. */ 2673 return 0.0; 2674 } 2675 CLANG_DIAG_IGNORE_STMT(-Wthread-safety); 2676 DEBUG_c({ 2677 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 2678 STORE_LC_NUMERIC_SET_STANDARD(); 2679 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n", 2680 PTR2UV(sv), SvNVX(sv)); 2681 RESTORE_LC_NUMERIC(); 2682 }); 2683 CLANG_DIAG_RESTORE_STMT; 2684 return SvNVX(sv); 2685} 2686 2687/* 2688=for apidoc sv_2num 2689 2690Return an SV with the numeric value of the source SV, doing any necessary 2691reference or overload conversion. The caller is expected to have handled 2692get-magic already. 2693 2694=cut 2695*/ 2696 2697SV * 2698Perl_sv_2num(pTHX_ SV *const sv) 2699{ 2700 PERL_ARGS_ASSERT_SV_2NUM; 2701 2702 if (!SvROK(sv)) 2703 return sv; 2704 if (SvAMAGIC(sv)) { 2705 SV * const tmpsv = AMG_CALLunary(sv, numer_amg); 2706 TAINT_IF(tmpsv && SvTAINTED(tmpsv)); 2707 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) 2708 return sv_2num(tmpsv); 2709 } 2710 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv)))); 2711} 2712 2713/* int2str_table: lookup table containing string representations of all 2714 * two digit numbers. For example, int2str_table.arr[0] is "00" and 2715 * int2str_table.arr[12*2] is "12". 2716 * 2717 * We are going to read two bytes at a time, so we have to ensure that 2718 * the array is aligned to a 2 byte boundary. That's why it was made a 2719 * union with a dummy U16 member. */ 2720static const union { 2721 char arr[200]; 2722 U16 dummy; 2723} int2str_table = {{ 2724 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6', 2725 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3', 2726 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0', 2727 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7', 2728 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4', 2729 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1', 2730 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8', 2731 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5', 2732 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2', 2733 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9', 2734 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6', 2735 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3', 2736 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0', 2737 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7', 2738 '9', '8', '9', '9' 2739}}; 2740 2741/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or 2742 * UV as a string towards the end of buf, and return pointers to start and 2743 * end of it. 2744 * 2745 * We assume that buf is at least TYPE_CHARS(UV) long. 2746 */ 2747 2748PERL_STATIC_INLINE char * 2749S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob) 2750{ 2751 char *ptr = buf + TYPE_CHARS(UV); 2752 char * const ebuf = ptr; 2753 int sign; 2754 U16 *word_ptr, *word_table; 2755 2756 PERL_ARGS_ASSERT_UIV_2BUF; 2757 2758 /* ptr has to be properly aligned, because we will cast it to U16* */ 2759 assert(PTR2nat(ptr) % 2 == 0); 2760 /* we are going to read/write two bytes at a time */ 2761 word_ptr = (U16*)ptr; 2762 word_table = (U16*)int2str_table.arr; 2763 2764 if (UNLIKELY(is_uv)) 2765 sign = 0; 2766 else if (iv >= 0) { 2767 uv = iv; 2768 sign = 0; 2769 } else { 2770 /* Using 0- here to silence bogus warning from MS VC */ 2771 uv = (UV) (0 - (UV) iv); 2772 sign = 1; 2773 } 2774 2775 while (uv > 99) { 2776 *--word_ptr = word_table[uv % 100]; 2777 uv /= 100; 2778 } 2779 ptr = (char*)word_ptr; 2780 2781 if (uv < 10) 2782 *--ptr = (char)uv + '0'; 2783 else { 2784 *--word_ptr = word_table[uv]; 2785 ptr = (char*)word_ptr; 2786 } 2787 2788 if (sign) 2789 *--ptr = '-'; 2790 2791 *peob = ebuf; 2792 return ptr; 2793} 2794 2795/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an 2796 * infinity or a not-a-number, writes the appropriate strings to the 2797 * buffer, including a zero byte. On success returns the written length, 2798 * excluding the zero byte, on failure (not an infinity, not a nan) 2799 * returns zero, assert-fails on maxlen being too short. 2800 * 2801 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only 2802 * shared string constants we point to, instead of generating a new 2803 * string for each instance. */ 2804STATIC size_t 2805S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) { 2806 char* s = buffer; 2807 assert(maxlen >= 4); 2808 if (Perl_isinf(nv)) { 2809 if (nv < 0) { 2810 if (maxlen < 5) /* "-Inf\0" */ 2811 return 0; 2812 *s++ = '-'; 2813 } else if (plus) { 2814 *s++ = '+'; 2815 } 2816 *s++ = 'I'; 2817 *s++ = 'n'; 2818 *s++ = 'f'; 2819 } 2820 else if (Perl_isnan(nv)) { 2821 *s++ = 'N'; 2822 *s++ = 'a'; 2823 *s++ = 'N'; 2824 /* XXX optionally output the payload mantissa bits as 2825 * "(unsigned)" (to match the nan("...") C99 function, 2826 * or maybe as "(0xhhh...)" would make more sense... 2827 * provide a format string so that the user can decide? 2828 * NOTE: would affect the maxlen and assert() logic.*/ 2829 } 2830 else { 2831 return 0; 2832 } 2833 assert((s == buffer + 3) || (s == buffer + 4)); 2834 *s = 0; 2835 return s - buffer; 2836} 2837 2838/* 2839=for apidoc sv_2pv 2840=for apidoc_item sv_2pv_flags 2841 2842These implement the various forms of the L<perlapi/C<SvPV>> macros. 2843The macros are the preferred interface. 2844 2845These return a pointer to the string value of an SV (coercing it to a string if 2846necessary), and set C<*lp> to its length in bytes. 2847 2848The forms differ in that plain C<sv_2pvbyte> always processes 'get' magic; and 2849C<sv_2pvbyte_flags> processes 'get' magic if and only if C<flags> contains 2850C<SV_GMAGIC>. 2851 2852=for apidoc Amnh||SV_GMAGIC 2853 2854=cut 2855*/ 2856 2857char * 2858Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const U32 flags) 2859{ 2860 char *s; 2861 bool done_gmagic = FALSE; 2862 2863 PERL_ARGS_ASSERT_SV_2PV_FLAGS; 2864 2865 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV 2866 && SvTYPE(sv) != SVt_PVFM); 2867 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) { 2868 mg_get(sv); 2869 done_gmagic = TRUE; 2870 } 2871 2872 if (SvROK(sv)) { 2873 if (SvAMAGIC(sv)) { 2874 SV *tmpstr; 2875 SV *nsv= (SV *)sv; 2876 if (flags & SV_SKIP_OVERLOAD) 2877 return NULL; 2878 if (done_gmagic) 2879 nsv = sv_mortalcopy_flags(sv,0); 2880 tmpstr = AMG_CALLunary(nsv, string_amg); 2881 TAINT_IF(tmpstr && SvTAINTED(tmpstr)); 2882 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(nsv)))) { 2883 /* Unwrap this: */ 2884 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr); 2885 */ 2886 2887 char *pv; 2888 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) { 2889 if (flags & SV_CONST_RETURN) { 2890 pv = (char *) SvPVX_const(tmpstr); 2891 } else { 2892 pv = (flags & SV_MUTABLE_RETURN) 2893 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr); 2894 } 2895 if (lp) 2896 *lp = SvCUR(tmpstr); 2897 } else { 2898 pv = sv_2pv_flags(tmpstr, lp, flags); 2899 } 2900 if (SvUTF8(tmpstr)) 2901 SvUTF8_on(sv); 2902 else 2903 SvUTF8_off(sv); 2904 return pv; 2905 } 2906 } 2907 { 2908 STRLEN len; 2909 char *retval; 2910 char *buffer; 2911 SV *const referent = SvRV(sv); 2912 2913 if (!referent) { 2914 len = 7; 2915 retval = buffer = savepvn("NULLREF", len); 2916 } else if (SvTYPE(referent) == SVt_REGEXP && 2917 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) || 2918 amagic_is_enabled(string_amg))) { 2919 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent); 2920 2921 assert(re); 2922 2923 /* If the regex is UTF-8 we want the containing scalar to 2924 have an UTF-8 flag too */ 2925 if (RX_UTF8(re)) 2926 SvUTF8_on(sv); 2927 else 2928 SvUTF8_off(sv); 2929 2930 if (lp) 2931 *lp = RX_WRAPLEN(re); 2932 2933 return RX_WRAPPED(re); 2934 } else { 2935 const char *const typestring = sv_reftype(referent, 0); 2936 const STRLEN typelen = strlen(typestring); 2937 UV addr = PTR2UV(referent); 2938 const char *stashname = NULL; 2939 STRLEN stashnamelen = 0; /* hush, gcc */ 2940 const char *buffer_end; 2941 2942 if (SvOBJECT(referent)) { 2943 const HEK *const name = HvNAME_HEK(SvSTASH(referent)); 2944 2945 if (name) { 2946 stashname = HEK_KEY(name); 2947 stashnamelen = HEK_LEN(name); 2948 2949 if (HEK_UTF8(name)) { 2950 SvUTF8_on(sv); 2951 } else { 2952 SvUTF8_off(sv); 2953 } 2954 } else { 2955 stashname = "__ANON__"; 2956 stashnamelen = 8; 2957 } 2958 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */ 2959 + 2 * sizeof(UV) + 2 /* )\0 */; 2960 } else { 2961 len = typelen + 3 /* (0x */ 2962 + 2 * sizeof(UV) + 2 /* )\0 */; 2963 } 2964 2965 Newx(buffer, len, char); 2966 buffer_end = retval = buffer + len; 2967 2968 /* Working backwards */ 2969 *--retval = '\0'; 2970 *--retval = ')'; 2971 do { 2972 *--retval = PL_hexdigit[addr & 15]; 2973 } while (addr >>= 4); 2974 *--retval = 'x'; 2975 *--retval = '0'; 2976 *--retval = '('; 2977 2978 retval -= typelen; 2979 memcpy(retval, typestring, typelen); 2980 2981 if (stashname) { 2982 *--retval = '='; 2983 retval -= stashnamelen; 2984 memcpy(retval, stashname, stashnamelen); 2985 } 2986 /* retval may not necessarily have reached the start of the 2987 buffer here. */ 2988 assert (retval >= buffer); 2989 2990 len = buffer_end - retval - 1; /* -1 for that \0 */ 2991 } 2992 if (lp) 2993 *lp = len; 2994 SAVEFREEPV(buffer); 2995 return retval; 2996 } 2997 } 2998 2999 if (SvPOKp(sv)) { 3000 if (lp) 3001 *lp = SvCUR(sv); 3002 if (flags & SV_MUTABLE_RETURN) 3003 return SvPVX_mutable(sv); 3004 if (flags & SV_CONST_RETURN) 3005 return (char *)SvPVX_const(sv); 3006 return SvPVX(sv); 3007 } 3008 3009 if (SvIOK(sv)) { 3010 /* I'm assuming that if both IV and NV are equally valid then 3011 converting the IV is going to be more efficient */ 3012 const U32 isUIOK = SvIsUV(sv); 3013 /* The purpose of this union is to ensure that arr is aligned on 3014 a 2 byte boundary, because that is what uiv_2buf() requires */ 3015 union { 3016 char arr[TYPE_CHARS(UV)]; 3017 U16 dummy; 3018 } buf; 3019 char *ebuf, *ptr; 3020 STRLEN len; 3021 3022 if (SvTYPE(sv) < SVt_PVIV) 3023 sv_upgrade(sv, SVt_PVIV); 3024 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf); 3025 len = ebuf - ptr; 3026 /* inlined from sv_setpvn */ 3027 s = SvGROW_mutable(sv, len + 1); 3028 Move(ptr, s, len, char); 3029 s += len; 3030 *s = '\0'; 3031 /* We used to call SvPOK_on(). Whilst this is fine for (most) Perl code, 3032 it means that after this stringification is cached, there is no way 3033 to distinguish between values originally assigned as $a = 42; and 3034 $a = "42"; (or results of string operators vs numeric operators) 3035 where the value has subsequently been used in the other sense 3036 and had a value cached. 3037 This (somewhat) hack means that we retain the cached stringification, 3038 but don't set SVf_POK. Hence if a value is SVf_IOK|SVf_POK then it 3039 originated as "42", whereas if it's SVf_IOK then it originated as 42. 3040 (ignore SVp_IOK and SVp_POK) 3041 The SvPV macros are now updated to recognise this specific case 3042 (and that there isn't overloading or magic that could alter the 3043 cached value) and so return the cached value immediately without 3044 re-entering this function, getting back here to this block of code, 3045 and repeating the same conversion. */ 3046 SvPOKp_on(sv); 3047 } 3048 else if (SvNOK(sv)) { 3049 if (SvTYPE(sv) < SVt_PVNV) 3050 sv_upgrade(sv, SVt_PVNV); 3051 if (SvNVX(sv) == 0.0 3052#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan) 3053 && !Perl_isnan(SvNVX(sv)) 3054#endif 3055 ) { 3056 s = SvGROW_mutable(sv, 2); 3057 *s++ = '0'; 3058 *s = '\0'; 3059 } else { 3060 STRLEN len; 3061 STRLEN size = 5; /* "-Inf\0" */ 3062 3063 s = SvGROW_mutable(sv, size); 3064 len = S_infnan_2pv(SvNVX(sv), s, size, 0); 3065 if (len > 0) { 3066 s += len; 3067 SvPOKp_on(sv); 3068 } 3069 else { 3070 /* some Xenix systems wipe out errno here */ 3071 dSAVE_ERRNO; 3072 3073 size = 3074 1 + /* sign */ 3075 1 + /* "." */ 3076 NV_DIG + 3077 1 + /* "e" */ 3078 1 + /* sign */ 3079 5 + /* exponent digits */ 3080 1 + /* \0 */ 3081 2; /* paranoia */ 3082 3083 s = SvGROW_mutable(sv, size); 3084#ifndef USE_LOCALE_NUMERIC 3085 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG); 3086 3087 SvPOKp_on(sv); 3088#else 3089 { 3090 bool local_radix; 3091 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 3092 STORE_LC_NUMERIC_SET_TO_NEEDED(); 3093 3094 local_radix = NOT_IN_NUMERIC_STANDARD_; 3095 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) { 3096 size += SvCUR(PL_numeric_radix_sv) - 1; 3097 s = SvGROW_mutable(sv, size); 3098 } 3099 3100 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG); 3101 3102 /* If the radix character is UTF-8, and actually is in the 3103 * output, turn on the UTF-8 flag for the scalar */ 3104 if ( local_radix 3105 && SvUTF8(PL_numeric_radix_sv) 3106 && instr(s, SvPVX_const(PL_numeric_radix_sv))) 3107 { 3108 SvUTF8_on(sv); 3109 } 3110 3111 RESTORE_LC_NUMERIC(); 3112 } 3113 3114 /* We don't call SvPOK_on(), because it may come to 3115 * pass that the locale changes so that the 3116 * stringification we just did is no longer correct. We 3117 * will have to re-stringify every time it is needed */ 3118#endif 3119 RESTORE_ERRNO; 3120 } 3121 while (*s) s++; 3122 } 3123 } 3124 else if (isGV_with_GP(sv)) { 3125 GV *const gv = MUTABLE_GV(sv); 3126 SV *const buffer = sv_newmortal(); 3127 3128 gv_efullname3(buffer, gv, "*"); 3129 3130 assert(SvPOK(buffer)); 3131 if (SvUTF8(buffer)) 3132 SvUTF8_on(sv); 3133 else 3134 SvUTF8_off(sv); 3135 if (lp) 3136 *lp = SvCUR(buffer); 3137 return SvPVX(buffer); 3138 } 3139 else { 3140 if (lp) 3141 *lp = 0; 3142 if (flags & SV_UNDEF_RETURNS_NULL) 3143 return NULL; 3144 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) 3145 report_uninit(sv); 3146 /* Typically the caller expects that sv_any is not NULL now. */ 3147 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV) 3148 sv_upgrade(sv, SVt_PV); 3149 return (char *)""; 3150 } 3151 3152 { 3153 const STRLEN len = s - SvPVX_const(sv); 3154 if (lp) 3155 *lp = len; 3156 SvCUR_set(sv, len); 3157 } 3158 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n", 3159 PTR2UV(sv),SvPVX_const(sv))); 3160 if (flags & SV_CONST_RETURN) 3161 return (char *)SvPVX_const(sv); 3162 if (flags & SV_MUTABLE_RETURN) 3163 return SvPVX_mutable(sv); 3164 return SvPVX(sv); 3165} 3166 3167/* 3168=for apidoc sv_copypv 3169=for apidoc_item sv_copypv_flags 3170=for apidoc_item sv_copypv_nomg 3171 3172These copy a stringified representation of the source SV into the 3173destination SV. They automatically perform coercion of numeric values into 3174strings. Guaranteed to preserve the C<UTF8> flag even from overloaded objects. 3175Similar in nature to C<sv_2pv[_flags]> but they operate directly on an SV 3176instead of just the string. Mostly they use L</C<sv_2pv_flags>> to 3177do the work, except when that would lose the UTF-8'ness of the PV. 3178 3179The three forms differ only in whether or not they perform 'get magic' on 3180C<sv>. C<sv_copypv_nomg> skips 'get magic'; C<sv_copypv> performs it; and 3181C<sv_copypv_flags> either performs it (if the C<SV_GMAGIC> bit is set in 3182C<flags>) or doesn't (if that bit is cleared). 3183 3184=cut 3185*/ 3186 3187void 3188Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags) 3189{ 3190 STRLEN len; 3191 const char *s; 3192 3193 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS; 3194 3195 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC)); 3196 sv_setpvn(dsv,s,len); 3197 if (SvUTF8(ssv)) 3198 SvUTF8_on(dsv); 3199 else 3200 SvUTF8_off(dsv); 3201} 3202 3203/* 3204=for apidoc sv_2pvbyte 3205=for apidoc_item sv_2pvbyte_flags 3206 3207These implement the various forms of the L<perlapi/C<SvPVbyte>> macros. 3208The macros are the preferred interface. 3209 3210These return a pointer to the byte-encoded representation of the SV, and set 3211C<*lp> to its length. If the SV is marked as being encoded as UTF-8, it will 3212be downgraded, if possible, to a byte string. If the SV cannot be downgraded, 3213they croak. 3214 3215The forms differ in that plain C<sv_2pvbyte> always processes 'get' magic; and 3216C<sv_2pvbyte_flags> processes 'get' magic if and only if C<flags> contains 3217C<SV_GMAGIC>. 3218 3219=for apidoc Amnh||SV_GMAGIC 3220 3221=cut 3222*/ 3223 3224char * 3225Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags) 3226{ 3227 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS; 3228 3229 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) 3230 mg_get(sv); 3231 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv)) 3232 || isGV_with_GP(sv) || SvROK(sv)) { 3233 SV *sv2 = sv_newmortal(); 3234 sv_copypv_nomg(sv2,sv); 3235 sv = sv2; 3236 } 3237 sv_utf8_downgrade_nomg(sv,0); 3238 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv); 3239} 3240 3241/* 3242=for apidoc sv_2pvutf8 3243=for apidoc_item sv_2pvutf8_flags 3244 3245These implement the various forms of the L<perlapi/C<SvPVutf8>> macros. 3246The macros are the preferred interface. 3247 3248These return a pointer to the UTF-8-encoded representation of the SV, and set 3249C<*lp> to its length in bytes. They may cause the SV to be upgraded to UTF-8 3250as a side-effect. 3251 3252The forms differ in that plain C<sv_2pvutf8> always processes 'get' magic; and 3253C<sv_2pvutf8_flags> processes 'get' magic if and only if C<flags> contains 3254C<SV_GMAGIC>. 3255 3256=cut 3257*/ 3258 3259char * 3260Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags) 3261{ 3262 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS; 3263 3264 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) 3265 mg_get(sv); 3266 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv)) 3267 || isGV_with_GP(sv) || SvROK(sv)) { 3268 SV *sv2 = sv_newmortal(); 3269 sv_copypv_nomg(sv2,sv); 3270 sv = sv2; 3271 } 3272 sv_utf8_upgrade_nomg(sv); 3273 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv); 3274} 3275 3276 3277/* 3278=for apidoc sv_2bool 3279 3280This macro is only used by C<sv_true()> or its macro equivalent, and only if 3281the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. 3282It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag. 3283 3284=for apidoc sv_2bool_flags 3285 3286This function is only used by C<sv_true()> and friends, and only if 3287the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags 3288contain C<SV_GMAGIC>, then it does an C<mg_get()> first. 3289 3290 3291=cut 3292*/ 3293 3294bool 3295Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags) 3296{ 3297 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS; 3298 3299 restart: 3300 if(flags & SV_GMAGIC) SvGETMAGIC(sv); 3301 3302 if (!SvOK(sv)) 3303 return 0; 3304 if (SvROK(sv)) { 3305 if (SvAMAGIC(sv)) { 3306 SV * const tmpsv = AMG_CALLunary(sv, bool__amg); 3307 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) { 3308 bool svb; 3309 sv = tmpsv; 3310 if(SvGMAGICAL(sv)) { 3311 flags = SV_GMAGIC; 3312 goto restart; /* call sv_2bool */ 3313 } 3314 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */ 3315 else if(!SvOK(sv)) { 3316 svb = 0; 3317 } 3318 else if(SvPOK(sv)) { 3319 svb = SvPVXtrue(sv); 3320 } 3321 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) { 3322 svb = (SvIOK(sv) && SvIVX(sv) != 0) 3323 || (SvNOK(sv) && SvNVX(sv) != 0.0); 3324 } 3325 else { 3326 flags = 0; 3327 goto restart; /* call sv_2bool_nomg */ 3328 } 3329 return cBOOL(svb); 3330 } 3331 } 3332 assert(SvRV(sv)); 3333 return TRUE; 3334 } 3335 if (isREGEXP(sv)) 3336 return 3337 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0'); 3338 3339 if (SvNOK(sv) && !SvPOK(sv)) 3340 return SvNVX(sv) != 0.0; 3341 3342 return SvTRUE_common(sv, 0); 3343} 3344 3345/* 3346=for apidoc sv_utf8_upgrade 3347=for apidoc_item sv_utf8_upgrade_flags 3348=for apidoc_item sv_utf8_upgrade_flags_grow 3349=for apidoc_item sv_utf8_upgrade_nomg 3350 3351These convert the PV of an SV to its UTF-8-encoded form. 3352The SV is forced to string form if it is not already. 3353They always set the C<SvUTF8> flag to avoid future validity checks even if the 3354whole string is the same in UTF-8 as not. 3355They return the number of bytes in the converted string 3356 3357The forms differ in just two ways. The main difference is whether or not they 3358perform 'get magic' on C<sv>. C<sv_utf8_upgrade_nomg> skips 'get magic'; 3359C<sv_utf8_upgrade> performs it; and C<sv_utf8_upgrade_flags> and 3360C<sv_utf8_upgrade_flags_grow> either perform it (if the C<SV_GMAGIC> bit is set 3361in C<flags>) or don't (if that bit is cleared). 3362 3363The other difference is that C<sv_utf8_upgrade_flags_grow> has an additional 3364parameter, C<extra>, which allows the caller to specify an amount of space to 3365be reserved as spare beyond what is needed for the actual conversion. This is 3366used when the caller knows it will soon be needing yet more space, and it is 3367more efficient to request space from the system in a single call. 3368This form is otherwise identical to C<sv_utf8_upgrade_flags>. 3369 3370These are not a general purpose byte encoding to Unicode interface: use the 3371Encode extension for that. 3372 3373The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored. 3374 3375=for apidoc Amnh||SV_GMAGIC| 3376=for apidoc Amnh||SV_FORCE_UTF8_UPGRADE| 3377 3378=cut 3379 3380If the routine itself changes the string, it adds a trailing C<NUL>. Such a 3381C<NUL> isn't guaranteed due to having other routines do the work in some input 3382cases, or if the input is already flagged as being in utf8. 3383 3384*/ 3385 3386STRLEN 3387Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra) 3388{ 3389 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW; 3390 3391 if (sv == &PL_sv_undef) 3392 return 0; 3393 if (!SvPOK_nog(sv)) { 3394 STRLEN len = 0; 3395 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) { 3396 (void) sv_2pv_flags(sv,&len, flags); 3397 if (SvUTF8(sv)) { 3398 if (extra) SvGROW(sv, SvCUR(sv) + extra); 3399 return len; 3400 } 3401 } else { 3402 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC); 3403 } 3404 } 3405 3406 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already 3407 * compiled and individual nodes will remain non-utf8 even if the 3408 * stringified version of the pattern gets upgraded. Whether the 3409 * PVX of a REGEXP should be grown or we should just croak, I don't 3410 * know - DAPM */ 3411 if (SvUTF8(sv) || isREGEXP(sv)) { 3412 if (extra) SvGROW(sv, SvCUR(sv) + extra); 3413 return SvCUR(sv); 3414 } 3415 3416 if (SvIsCOW(sv)) { 3417 S_sv_uncow(aTHX_ sv, 0); 3418 } 3419 3420 if (SvCUR(sv) == 0) { 3421 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing 3422 byte */ 3423 } else { /* Assume Latin-1/EBCDIC */ 3424 /* This function could be much more efficient if we 3425 * had a FLAG in SVs to signal if there are any variant 3426 * chars in the PV. Given that there isn't such a flag 3427 * make the loop as fast as possible. */ 3428 U8 * s = (U8 *) SvPVX_const(sv); 3429 U8 *t = s; 3430 3431 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) { 3432 3433 /* utf8 conversion not needed because all are invariants. Mark 3434 * as UTF-8 even if no variant - saves scanning loop */ 3435 SvUTF8_on(sv); 3436 if (extra) SvGROW(sv, SvCUR(sv) + extra); 3437 return SvCUR(sv); 3438 } 3439 3440 /* Here, there is at least one variant (t points to the first one), so 3441 * the string should be converted to utf8. Everything from 's' to 3442 * 't - 1' will occupy only 1 byte each on output. 3443 * 3444 * Note that the incoming SV may not have a trailing '\0', as certain 3445 * code in pp_formline can send us partially built SVs. 3446 * 3447 * There are two main ways to convert. One is to create a new string 3448 * and go through the input starting from the beginning, appending each 3449 * converted value onto the new string as we go along. Going this 3450 * route, it's probably best to initially allocate enough space in the 3451 * string rather than possibly running out of space and having to 3452 * reallocate and then copy what we've done so far. Since everything 3453 * from 's' to 't - 1' is invariant, the destination can be initialized 3454 * with these using a fast memory copy. To be sure to allocate enough 3455 * space, one could use the worst case scenario, where every remaining 3456 * byte expands to two under UTF-8, or one could parse it and count 3457 * exactly how many do expand. 3458 * 3459 * The other way is to unconditionally parse the remainder of the 3460 * string to figure out exactly how big the expanded string will be, 3461 * growing if needed. Then start at the end of the string and place 3462 * the character there at the end of the unfilled space in the expanded 3463 * one, working backwards until reaching 't'. 3464 * 3465 * The problem with assuming the worst case scenario is that for very 3466 * long strings, we could allocate much more memory than actually 3467 * needed, which can create performance problems. If we have to parse 3468 * anyway, the second method is the winner as it may avoid an extra 3469 * copy. The code used to use the first method under some 3470 * circumstances, but now that there is faster variant counting on 3471 * ASCII platforms, the second method is used exclusively, eliminating 3472 * some code that no longer has to be maintained. */ 3473 3474 { 3475 /* Count the total number of variants there are. We can start 3476 * just beyond the first one, which is known to be at 't' */ 3477 const Size_t invariant_length = t - s; 3478 U8 * e = (U8 *) SvEND(sv); 3479 3480 /* The length of the left overs, plus 1. */ 3481 const Size_t remaining_length_p1 = e - t; 3482 3483 /* We expand by 1 for the variant at 't' and one for each remaining 3484 * variant (we start looking at 't+1') */ 3485 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e); 3486 3487 /* +1 = trailing NUL */ 3488 Size_t need = SvCUR(sv) + expansion + extra + 1; 3489 U8 * d; 3490 3491 /* Grow if needed */ 3492 if (SvLEN(sv) < need) { 3493 t = invariant_length + (U8*) SvGROW(sv, need); 3494 e = t + remaining_length_p1; 3495 } 3496 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion); 3497 3498 /* Set the NUL at the end */ 3499 d = (U8 *) SvEND(sv); 3500 *d-- = '\0'; 3501 3502 /* Having decremented d, it points to the position to put the 3503 * very last byte of the expanded string. Go backwards through 3504 * the string, copying and expanding as we go, stopping when we 3505 * get to the part that is invariant the rest of the way down */ 3506 3507 e--; 3508 while (e >= t) { 3509 if (NATIVE_BYTE_IS_INVARIANT(*e)) { 3510 *d-- = *e; 3511 } else { 3512 *d-- = UTF8_EIGHT_BIT_LO(*e); 3513 *d-- = UTF8_EIGHT_BIT_HI(*e); 3514 } 3515 e--; 3516 } 3517 3518 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 3519 /* Update pos. We do it at the end rather than during 3520 * the upgrade, to avoid slowing down the common case 3521 * (upgrade without pos). 3522 * pos can be stored as either bytes or characters. Since 3523 * this was previously a byte string we can just turn off 3524 * the bytes flag. */ 3525 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); 3526 if (mg) { 3527 mg->mg_flags &= ~MGf_BYTES; 3528 } 3529 if ((mg = mg_find(sv, PERL_MAGIC_utf8))) 3530 magic_setutf8(sv,mg); /* clear UTF8 cache */ 3531 } 3532 } 3533 } 3534 3535 SvUTF8_on(sv); 3536 return SvCUR(sv); 3537} 3538 3539/* 3540=for apidoc sv_utf8_downgrade 3541=for apidoc_item sv_utf8_downgrade_flags 3542=for apidoc_item sv_utf8_downgrade_nomg 3543 3544These attempt to convert the PV of an SV from characters to bytes. If the PV 3545contains a character that cannot fit in a byte, this conversion will fail; in 3546this case, C<FALSE> is returned if C<fail_ok> is true; otherwise they croak. 3547 3548They are not a general purpose Unicode to byte encoding interface: 3549use the C<Encode> extension for that. 3550 3551They differ only in that: 3552 3553C<sv_utf8_downgrade> processes 'get' magic on C<sv>. 3554 3555C<sv_utf8_downgrade_nomg> does not. 3556 3557C<sv_utf8_downgrade_flags> has an additional C<flags> parameter in which you can specify 3558C<SV_GMAGIC> to process 'get' magic, or leave it cleared to not process 'get' magic. 3559 3560=cut 3561*/ 3562 3563bool 3564Perl_sv_utf8_downgrade_flags(pTHX_ SV *const sv, const bool fail_ok, const U32 flags) 3565{ 3566 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE_FLAGS; 3567 3568 if (SvPOKp(sv) && SvUTF8(sv)) { 3569 if (SvCUR(sv)) { 3570 U8 *s; 3571 STRLEN len; 3572 U32 mg_flags = flags & SV_GMAGIC; 3573 3574 if (SvIsCOW(sv)) { 3575 S_sv_uncow(aTHX_ sv, 0); 3576 } 3577 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 3578 /* update pos */ 3579 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); 3580 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) { 3581 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len, 3582 mg_flags|SV_CONST_RETURN); 3583 mg_flags = 0; /* sv_pos_b2u does get magic */ 3584 } 3585 if ((mg = mg_find(sv, PERL_MAGIC_utf8))) 3586 magic_setutf8(sv,mg); /* clear UTF8 cache */ 3587 3588 } 3589 s = (U8 *) SvPV_flags(sv, len, mg_flags); 3590 3591 if (!utf8_to_bytes(s, &len)) { 3592 if (fail_ok) 3593 return FALSE; 3594 else { 3595 if (PL_op) 3596 Perl_croak(aTHX_ "Wide character in %s", 3597 OP_DESC(PL_op)); 3598 else 3599 Perl_croak(aTHX_ "Wide character"); 3600 } 3601 } 3602 SvCUR_set(sv, len); 3603 } 3604 } 3605 SvUTF8_off(sv); 3606 return TRUE; 3607} 3608 3609/* 3610=for apidoc sv_utf8_encode 3611 3612Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8> 3613flag off so that it looks like octets again. 3614 3615=cut 3616*/ 3617 3618void 3619Perl_sv_utf8_encode(pTHX_ SV *const sv) 3620{ 3621 PERL_ARGS_ASSERT_SV_UTF8_ENCODE; 3622 3623 if (SvREADONLY(sv)) { 3624 sv_force_normal_flags(sv, 0); 3625 } 3626 (void) sv_utf8_upgrade(sv); 3627 SvUTF8_off(sv); 3628} 3629 3630/* 3631=for apidoc sv_utf8_decode 3632 3633If the PV of the SV is an octet sequence in Perl's extended UTF-8 3634and contains a multiple-byte character, the C<SvUTF8> flag is turned on 3635so that it looks like a character. If the PV contains only single-byte 3636characters, the C<SvUTF8> flag stays off. 3637Scans PV for validity and returns FALSE if the PV is invalid UTF-8. 3638 3639=cut 3640*/ 3641 3642bool 3643Perl_sv_utf8_decode(pTHX_ SV *const sv) 3644{ 3645 PERL_ARGS_ASSERT_SV_UTF8_DECODE; 3646 3647 if (SvPOKp(sv)) { 3648 const U8 *start, *c, *first_variant; 3649 3650 /* The octets may have got themselves encoded - get them back as 3651 * bytes 3652 */ 3653 if (!sv_utf8_downgrade(sv, TRUE)) 3654 return FALSE; 3655 3656 /* it is actually just a matter of turning the utf8 flag on, but 3657 * we want to make sure everything inside is valid utf8 first. 3658 */ 3659 c = start = (const U8 *) SvPVX_const(sv); 3660 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) { 3661 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c))) 3662 return FALSE; 3663 SvUTF8_on(sv); 3664 } 3665 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 3666 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC 3667 after this, clearing pos. Does anything on CPAN 3668 need this? */ 3669 /* adjust pos to the start of a UTF8 char sequence */ 3670 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); 3671 if (mg) { 3672 I32 pos = mg->mg_len; 3673 if (pos > 0) { 3674 for (c = start + pos; c > start; c--) { 3675 if (UTF8_IS_START(*c)) 3676 break; 3677 } 3678 mg->mg_len = c - start; 3679 } 3680 } 3681 if ((mg = mg_find(sv, PERL_MAGIC_utf8))) 3682 magic_setutf8(sv,mg); /* clear UTF8 cache */ 3683 } 3684 } 3685 return TRUE; 3686} 3687 3688/* 3689=for apidoc sv_setsv 3690=for apidoc_item sv_setsv_flags 3691=for apidoc_item sv_setsv_mg 3692=for apidoc_item sv_setsv_nomg 3693 3694These copy the contents of the source SV C<ssv> into the destination SV C<dsv>. 3695C<ssv> may be destroyed if it is mortal, so don't use these functions if 3696the source SV needs to be reused. 3697Loosely speaking, they perform a copy-by-value, obliterating any previous 3698content of the destination. 3699 3700They differ only in that: 3701 3702C<sv_setsv> calls 'get' magic on C<ssv>, but skips 'set' magic on C<dsv>. 3703 3704C<sv_setsv_mg> calls both 'get' magic on C<ssv> and 'set' magic on C<dsv>. 3705 3706C<sv_setsv_nomg> skips all magic. 3707 3708C<sv_setsv_flags> has a C<flags> parameter which you can use to specify any 3709combination of magic handling, and also you can specify C<SV_NOSTEAL> so that 3710the buffers of temps will not be stolen. 3711 3712You probably want to instead use one of the assortment of wrappers, such as 3713C<L</SvSetSV>>, C<L</SvSetSV_nosteal>>, C<L</SvSetMagicSV>> and 3714C<L</SvSetMagicSV_nosteal>>. 3715 3716C<sv_setsv_flags> is the primary function for copying scalars, and most other 3717copy-ish functions and macros use it underneath. 3718 3719=for apidoc Amnh||SV_NOSTEAL 3720 3721=cut 3722*/ 3723 3724static void 3725S_glob_assign_glob(pTHX_ SV *const dsv, SV *const ssv, const int dtype) 3726{ 3727 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */ 3728 HV *old_stash = NULL; 3729 3730 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB; 3731 3732 if (dtype != SVt_PVGV && !isGV_with_GP(dsv)) { 3733 const char * const name = GvNAME(ssv); 3734 const STRLEN len = GvNAMELEN(ssv); 3735 { 3736 if (dtype >= SVt_PV) { 3737 SvPV_free(dsv); 3738 SvPV_set(dsv, 0); 3739 SvLEN_set(dsv, 0); 3740 SvCUR_set(dsv, 0); 3741 } 3742 SvUPGRADE(dsv, SVt_PVGV); 3743 (void)SvOK_off(dsv); 3744 isGV_with_GP_on(dsv); 3745 } 3746 GvSTASH(dsv) = GvSTASH(ssv); 3747 if (GvSTASH(dsv)) 3748 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dsv)), dsv); 3749 gv_name_set(MUTABLE_GV(dsv), name, len, 3750 GV_ADD | (GvNAMEUTF8(ssv) ? SVf_UTF8 : 0 )); 3751 SvFAKE_on(dsv); /* can coerce to non-glob */ 3752 } 3753 3754 if(GvGP(MUTABLE_GV(ssv))) { 3755 /* If source has method cache entry, clear it */ 3756 if(GvCVGEN(ssv)) { 3757 SvREFCNT_dec(GvCV(ssv)); 3758 GvCV_set(ssv, NULL); 3759 GvCVGEN(ssv) = 0; 3760 } 3761 /* If source has a real method, then a method is 3762 going to change */ 3763 else if( 3764 GvCV((const GV *)ssv) && GvSTASH(dsv) && HvHasENAME(GvSTASH(dsv)) 3765 ) { 3766 mro_changes = 1; 3767 } 3768 } 3769 3770 /* If dest already had a real method, that's a change as well */ 3771 if( 3772 !mro_changes && GvGP(MUTABLE_GV(dsv)) && GvCVu((const GV *)dsv) 3773 && GvSTASH(dsv) && HvHasENAME(GvSTASH(dsv)) 3774 ) { 3775 mro_changes = 1; 3776 } 3777 3778 /* We don't need to check the name of the destination if it was not a 3779 glob to begin with. */ 3780 if(dtype == SVt_PVGV) { 3781 const char * const name = GvNAME((const GV *)dsv); 3782 const STRLEN len = GvNAMELEN(dsv); 3783 if(memEQs(name, len, "ISA") 3784 /* The stash may have been detached from the symbol table, so 3785 check its name. */ 3786 && GvSTASH(dsv) && HvHasENAME(GvSTASH(dsv)) 3787 ) 3788 mro_changes = 2; 3789 else { 3790 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':') 3791 || (len == 1 && name[0] == ':')) { 3792 mro_changes = 3; 3793 3794 /* Set aside the old stash, so we can reset isa caches on 3795 its subclasses. */ 3796 if((old_stash = GvHV(dsv))) 3797 /* Make sure we do not lose it early. */ 3798 SvREFCNT_inc_simple_void_NN( 3799 sv_2mortal((SV *)old_stash) 3800 ); 3801 } 3802 } 3803 3804 SvREFCNT_inc_simple_void_NN(sv_2mortal(dsv)); 3805 } 3806 3807 /* freeing dsv's GP might free ssv (e.g. *x = $x), 3808 * so temporarily protect it */ 3809 ENTER; 3810 SAVEFREESV(SvREFCNT_inc_simple_NN(ssv)); 3811 gp_free(MUTABLE_GV(dsv)); 3812 GvINTRO_off(dsv); /* one-shot flag */ 3813 GvGP_set(dsv, gp_ref(GvGP(ssv))); 3814 LEAVE; 3815 3816 if (SvTAINTED(ssv)) 3817 SvTAINT(dsv); 3818 if (GvIMPORTED(dsv) != GVf_IMPORTED 3819 && CopSTASH_ne(PL_curcop, GvSTASH(dsv))) 3820 { 3821 GvIMPORTED_on(dsv); 3822 } 3823 GvMULTI_on(dsv); 3824 if(mro_changes == 2) { 3825 if (GvAV((const GV *)ssv)) { 3826 MAGIC *mg; 3827 SV * const sref = (SV *)GvAV((const GV *)dsv); 3828 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) { 3829 if (SvTYPE(mg->mg_obj) != SVt_PVAV) { 3830 AV * const ary = newAV_alloc_x(2); 3831 av_push_simple(ary, mg->mg_obj); /* takes the refcount */ 3832 av_push_simple(ary, SvREFCNT_inc_simple_NN(dsv)); 3833 mg->mg_obj = (SV *)ary; 3834 } else { 3835 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dsv)); 3836 } 3837 } 3838 else sv_magic(sref, dsv, PERL_MAGIC_isa, NULL, 0); 3839 } 3840 mro_isa_changed_in(GvSTASH(dsv)); 3841 } 3842 else if(mro_changes == 3) { 3843 HV * const stash = GvHV(dsv); 3844 if(old_stash ? HvHasENAME(old_stash) : cBOOL(stash)) 3845 mro_package_moved( 3846 stash, old_stash, 3847 (GV *)dsv, 0 3848 ); 3849 } 3850 else if(mro_changes) mro_method_changed_in(GvSTASH(dsv)); 3851 if (GvIO(dsv) && dtype == SVt_PVGV) { 3852 DEBUG_o(Perl_deb(aTHX_ 3853 "glob_assign_glob clearing PL_stashcache\n")); 3854 /* It's a cache. It will rebuild itself quite happily. 3855 It's a lot of effort to work out exactly which key (or keys) 3856 might be invalidated by the creation of the this file handle. 3857 */ 3858 hv_clear(PL_stashcache); 3859 } 3860 return; 3861} 3862 3863void 3864Perl_gv_setref(pTHX_ SV *const dsv, SV *const ssv) 3865{ 3866 SV * const sref = SvRV(ssv); 3867 SV *dref; 3868 const int intro = GvINTRO(dsv); 3869 SV **location; 3870 U8 import_flag = 0; 3871 const U32 stype = SvTYPE(sref); 3872 3873 PERL_ARGS_ASSERT_GV_SETREF; 3874 3875 if (intro) { 3876 GvINTRO_off(dsv); /* one-shot flag */ 3877 GvLINE(dsv) = CopLINE(PL_curcop); 3878 GvEGV(dsv) = MUTABLE_GV(dsv); 3879 } 3880 GvMULTI_on(dsv); 3881 switch (stype) { 3882 case SVt_PVCV: 3883 location = (SV **) &(GvGP(dsv)->gp_cv); /* XXX bypassing GvCV_set */ 3884 import_flag = GVf_IMPORTED_CV; 3885 goto common; 3886 case SVt_PVHV: 3887 location = (SV **) &GvHV(dsv); 3888 import_flag = GVf_IMPORTED_HV; 3889 goto common; 3890 case SVt_PVAV: 3891 location = (SV **) &GvAV(dsv); 3892 import_flag = GVf_IMPORTED_AV; 3893 goto common; 3894 case SVt_PVIO: 3895 location = (SV **) &GvIOp(dsv); 3896 goto common; 3897 case SVt_PVFM: 3898 location = (SV **) &GvFORM(dsv); 3899 goto common; 3900 default: 3901 location = &GvSV(dsv); 3902 import_flag = GVf_IMPORTED_SV; 3903 common: 3904 if (intro) { 3905 if (stype == SVt_PVCV) { 3906 /*if (GvCVGEN(dsv) && (GvCV(dsv) != (const CV *)sref || GvCVGEN(dsv))) {*/ 3907 if (GvCVGEN(dsv)) { 3908 SvREFCNT_dec(GvCV(dsv)); 3909 GvCV_set(dsv, NULL); 3910 GvCVGEN(dsv) = 0; /* Switch off cacheness. */ 3911 } 3912 } 3913 /* SAVEt_GVSLOT takes more room on the savestack and has more 3914 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs 3915 leave_scope needs access to the GV so it can reset method 3916 caches. We must use SAVEt_GVSLOT whenever the type is 3917 SVt_PVCV, even if the stash is anonymous, as the stash may 3918 gain a name somehow before leave_scope. */ 3919 if (stype == SVt_PVCV) { 3920 /* There is no save_pushptrptrptr. Creating it for this 3921 one call site would be overkill. So inline the ss add 3922 routines here. */ 3923 dSS_ADD; 3924 SS_ADD_PTR(dsv); 3925 SS_ADD_PTR(location); 3926 SS_ADD_PTR(SvREFCNT_inc(*location)); 3927 SS_ADD_UV(SAVEt_GVSLOT); 3928 SS_ADD_END(4); 3929 } 3930 else SAVEGENERICSV(*location); 3931 } 3932 dref = *location; 3933 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dsv))) { 3934 CV* const cv = MUTABLE_CV(*location); 3935 if (cv) { 3936 if (!GvCVGEN((const GV *)dsv) && 3937 (CvROOT(cv) || CvXSUB(cv)) && 3938 /* redundant check that avoids creating the extra SV 3939 most of the time: */ 3940 (CvCONST(cv) || (ckWARN(WARN_REDEFINE) && !intro))) 3941 { 3942 SV * const new_const_sv = 3943 CvCONST((const CV *)sref) 3944 ? cv_const_sv_or_av((const CV *)sref) 3945 : NULL; 3946 HV * const stash = GvSTASH((const GV *)dsv); 3947 report_redefined_cv( 3948 sv_2mortal( 3949 stash 3950 ? Perl_newSVpvf(aTHX_ 3951 "%" HEKf "::%" HEKf, 3952 HEKfARG(HvNAME_HEK(stash)), 3953 HEKfARG(GvENAME_HEK(MUTABLE_GV(dsv)))) 3954 : Perl_newSVpvf(aTHX_ 3955 "%" HEKf, 3956 HEKfARG(GvENAME_HEK(MUTABLE_GV(dsv)))) 3957 ), 3958 cv, 3959 CvCONST((const CV *)sref) ? &new_const_sv : NULL 3960 ); 3961 } 3962 if (!intro) 3963 cv_ckproto_len_flags(cv, (const GV *)dsv, 3964 SvPOK(sref) ? CvPROTO(sref) : NULL, 3965 SvPOK(sref) ? CvPROTOLEN(sref) : 0, 3966 SvPOK(sref) ? SvUTF8(sref) : 0); 3967 } 3968 GvCVGEN(dsv) = 0; /* Switch off cacheness. */ 3969 GvASSUMECV_on(dsv); 3970 if(GvSTASH(dsv)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */ 3971 if (intro && GvREFCNT(dsv) > 1) { 3972 /* temporary remove extra savestack's ref */ 3973 --GvREFCNT(dsv); 3974 gv_method_changed(dsv); 3975 ++GvREFCNT(dsv); 3976 } 3977 else gv_method_changed(dsv); 3978 } 3979 } 3980 *location = SvREFCNT_inc_simple_NN(sref); 3981 if (import_flag && !(GvFLAGS(dsv) & import_flag) 3982 && CopSTASH_ne(PL_curcop, GvSTASH(dsv))) { 3983 GvFLAGS(dsv) |= import_flag; 3984 } 3985 3986 if (stype == SVt_PVHV) { 3987 const char * const name = GvNAME((GV*)dsv); 3988 const STRLEN len = GvNAMELEN(dsv); 3989 if ( 3990 ( 3991 (len > 1 && name[len-2] == ':' && name[len-1] == ':') 3992 || (len == 1 && name[0] == ':') 3993 ) 3994 && (!dref || HvHasENAME(dref)) 3995 ) { 3996 mro_package_moved( 3997 (HV *)sref, (HV *)dref, 3998 (GV *)dsv, 0 3999 ); 4000 } 4001 } 4002 else if ( 4003 stype == SVt_PVAV && sref != dref 4004 && memEQs(GvNAME((GV*)dsv), GvNAMELEN((GV*)dsv), "ISA") 4005 /* The stash may have been detached from the symbol table, so 4006 check its name before doing anything. */ 4007 && GvSTASH(dsv) && HvHasENAME(GvSTASH(dsv)) 4008 ) { 4009 MAGIC *mg; 4010 MAGIC * const omg = dref && SvSMAGICAL(dref) 4011 ? mg_find(dref, PERL_MAGIC_isa) 4012 : NULL; 4013 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) { 4014 if (SvTYPE(mg->mg_obj) != SVt_PVAV) { 4015 AV * const ary = newAV_alloc_xz(4); 4016 av_push_simple(ary, mg->mg_obj); /* takes the refcount */ 4017 mg->mg_obj = (SV *)ary; 4018 } 4019 if (omg) { 4020 if (SvTYPE(omg->mg_obj) == SVt_PVAV) { 4021 SV **svp = AvARRAY((AV *)omg->mg_obj); 4022 I32 items = AvFILLp((AV *)omg->mg_obj) + 1; 4023 while (items--) 4024 av_push( 4025 (AV *)mg->mg_obj, 4026 SvREFCNT_inc_simple_NN(*svp++) 4027 ); 4028 } 4029 else 4030 av_push( 4031 (AV *)mg->mg_obj, 4032 SvREFCNT_inc_simple_NN(omg->mg_obj) 4033 ); 4034 } 4035 else 4036 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dsv)); 4037 } 4038 else 4039 { 4040 SSize_t i; 4041 sv_magic( 4042 sref, omg ? omg->mg_obj : dsv, PERL_MAGIC_isa, NULL, 0 4043 ); 4044 for (i = 0; i <= AvFILL(sref); ++i) { 4045 SV **elem = av_fetch ((AV*)sref, i, 0); 4046 if (elem) { 4047 sv_magic( 4048 *elem, sref, PERL_MAGIC_isaelem, NULL, i 4049 ); 4050 } 4051 } 4052 mg = mg_find(sref, PERL_MAGIC_isa); 4053 } 4054 /* Since the *ISA assignment could have affected more than 4055 one stash, don't call mro_isa_changed_in directly, but let 4056 magic_clearisa do it for us, as it already has the logic for 4057 dealing with globs vs arrays of globs. */ 4058 assert(mg); 4059 Perl_magic_clearisa(aTHX_ NULL, mg); 4060 } 4061 else if (stype == SVt_PVIO) { 4062 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n")); 4063 /* It's a cache. It will rebuild itself quite happily. 4064 It's a lot of effort to work out exactly which key (or keys) 4065 might be invalidated by the creation of the this file handle. 4066 */ 4067 hv_clear(PL_stashcache); 4068 } 4069 break; 4070 } 4071 if (!intro) SvREFCNT_dec(dref); 4072 if (SvTAINTED(ssv)) 4073 SvTAINT(dsv); 4074 return; 4075} 4076 4077 4078 4079 4080#ifdef PERL_DEBUG_READONLY_COW 4081# include <sys/mman.h> 4082 4083# ifndef PERL_MEMORY_DEBUG_HEADER_SIZE 4084# define PERL_MEMORY_DEBUG_HEADER_SIZE 0 4085# endif 4086 4087void 4088Perl_sv_buf_to_ro(pTHX_ SV *sv) 4089{ 4090 struct perl_memory_debug_header * const header = 4091 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE); 4092 const MEM_SIZE len = header->size; 4093 PERL_ARGS_ASSERT_SV_BUF_TO_RO; 4094# ifdef PERL_TRACK_MEMPOOL 4095 if (!header->readonly) header->readonly = 1; 4096# endif 4097 if (mprotect(header, len, PROT_READ)) 4098 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d", 4099 header, len, errno); 4100} 4101 4102static void 4103S_sv_buf_to_rw(pTHX_ SV *sv) 4104{ 4105 struct perl_memory_debug_header * const header = 4106 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE); 4107 const MEM_SIZE len = header->size; 4108 PERL_ARGS_ASSERT_SV_BUF_TO_RW; 4109 if (mprotect(header, len, PROT_READ|PROT_WRITE)) 4110 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d", 4111 header, len, errno); 4112# ifdef PERL_TRACK_MEMPOOL 4113 header->readonly = 0; 4114# endif 4115} 4116 4117#else 4118# define sv_buf_to_ro(sv) NOOP 4119# define sv_buf_to_rw(sv) NOOP 4120#endif 4121 4122void 4123Perl_sv_setsv_flags(pTHX_ SV *dsv, SV* ssv, const I32 flags) 4124{ 4125 U32 sflags; 4126 int dtype; 4127 svtype stype; 4128 unsigned int both_type; 4129 4130 PERL_ARGS_ASSERT_SV_SETSV_FLAGS; 4131 4132 if (UNLIKELY( ssv == dsv )) 4133 return; 4134 4135 if (UNLIKELY( !ssv )) 4136 ssv = &PL_sv_undef; 4137 4138 stype = SvTYPE(ssv); 4139 dtype = SvTYPE(dsv); 4140 both_type = (stype | dtype); 4141 4142 /* with these values, we can check that both SVs are NULL/IV (and not 4143 * freed) just by testing the or'ed types */ 4144 STATIC_ASSERT_STMT(SVt_NULL == 0); 4145 STATIC_ASSERT_STMT(SVt_IV == 1); 4146 STATIC_ASSERT_STMT(SVt_NV == 2); 4147#if NVSIZE <= IVSIZE 4148 if (both_type <= 2) { 4149#else 4150 if (both_type <= 1) { 4151#endif 4152 /* both src and dst are UNDEF/IV/RV - maybe NV depending on config, 4153 * so we can do a lot of special-casing */ 4154 U32 sflags; 4155 U32 new_dflags; 4156 SV *old_rv = NULL; 4157 4158 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dsv) */ 4159 if (SvREADONLY(dsv)) 4160 Perl_croak_no_modify(); 4161 if (SvROK(dsv)) { 4162 if (SvWEAKREF(dsv)) 4163 sv_unref_flags(dsv, 0); 4164 else 4165 old_rv = SvRV(dsv); 4166 SvROK_off(dsv); 4167 } 4168 4169 assert(!SvGMAGICAL(ssv)); 4170 assert(!SvGMAGICAL(dsv)); 4171 4172 sflags = SvFLAGS(ssv); 4173 if (sflags & (SVf_IOK|SVf_ROK)) { 4174 SET_SVANY_FOR_BODYLESS_IV(dsv); 4175 new_dflags = SVt_IV; 4176 4177 if (sflags & SVf_ROK) { 4178 dsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(ssv)); 4179 new_dflags |= SVf_ROK; 4180 } 4181 else { 4182 /* both src and dst are <= SVt_IV, so sv_any points to the 4183 * head; so access the head directly 4184 */ 4185 assert( &(ssv->sv_u.svu_iv) 4186 == &(((XPVIV*) SvANY(ssv))->xiv_iv)); 4187 assert( &(dsv->sv_u.svu_iv) 4188 == &(((XPVIV*) SvANY(dsv))->xiv_iv)); 4189 dsv->sv_u.svu_iv = ssv->sv_u.svu_iv; 4190 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV)); 4191 } 4192 } 4193#if NVSIZE <= IVSIZE 4194 else if (sflags & SVf_NOK) { 4195 SET_SVANY_FOR_BODYLESS_NV(dsv); 4196 new_dflags = (SVt_NV|SVf_NOK|SVp_NOK); 4197 4198 /* both src and dst are <= SVt_MV, so sv_any points to the 4199 * head; so access the head directly 4200 */ 4201 assert( &(ssv->sv_u.svu_nv) 4202 == &(((XPVNV*) SvANY(ssv))->xnv_u.xnv_nv)); 4203 assert( &(dsv->sv_u.svu_nv) 4204 == &(((XPVNV*) SvANY(dsv))->xnv_u.xnv_nv)); 4205 dsv->sv_u.svu_nv = ssv->sv_u.svu_nv; 4206 } 4207#endif 4208 else { 4209 new_dflags = dtype; /* turn off everything except the type */ 4210 } 4211 /* Should preserve some dsv flags - at least SVs_TEMP, */ 4212 /* so cannot just set SvFLAGS(dsv) = new_dflags */ 4213 /* First clear the flags that we do want to clobber */ 4214 (void)SvOK_off(dsv); 4215 SvFLAGS(dsv) &= ~SVTYPEMASK; 4216 /* Now set the new flags */ 4217 SvFLAGS(dsv) |= new_dflags; 4218 4219 SvREFCNT_dec(old_rv); 4220 return; 4221 } 4222 4223 if (UNLIKELY(both_type == SVTYPEMASK)) { 4224 if (SvIS_FREED(dsv)) { 4225 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf 4226 " to a freed scalar %p", SVfARG(ssv), (void *)dsv); 4227 } 4228 if (SvIS_FREED(ssv)) { 4229 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p", 4230 (void*)ssv, (void*)dsv); 4231 } 4232 } 4233 4234 4235 4236 SV_CHECK_THINKFIRST_COW_DROP(dsv); 4237 dtype = SvTYPE(dsv); /* THINKFIRST may have changed type */ 4238 4239 /* There's a lot of redundancy below but we're going for speed here 4240 * Note: some of the cases below do return; rather than break; so the 4241 * if-elseif-else logic below this switch does not see all cases. */ 4242 4243 switch (stype) { 4244 case SVt_NULL: 4245 undef_sstr: 4246 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) { 4247 (void)SvOK_off(dsv); 4248 return; 4249 } 4250 break; 4251 case SVt_IV: 4252 if (SvIOK(ssv)) { 4253 switch (dtype) { 4254 case SVt_NULL: 4255 /* For performance, we inline promoting to type SVt_IV. */ 4256 /* We're starting from SVt_NULL, so provided that define is 4257 * actual 0, we don't have to unset any SV type flags 4258 * to promote to SVt_IV. */ 4259 STATIC_ASSERT_STMT(SVt_NULL == 0); 4260 SET_SVANY_FOR_BODYLESS_IV(dsv); 4261 SvFLAGS(dsv) |= SVt_IV; 4262 break; 4263 case SVt_NV: 4264 case SVt_PV: 4265 sv_upgrade(dsv, SVt_PVIV); 4266 break; 4267 case SVt_PVGV: 4268 case SVt_PVLV: 4269 goto end_of_first_switch; 4270 } 4271 (void)SvIOK_only(dsv); 4272 SvIV_set(dsv, SvIVX(ssv)); 4273 if (SvIsUV(ssv)) 4274 SvIsUV_on(dsv); 4275 /* SvTAINTED can only be true if the SV has taint magic, which in 4276 turn means that the SV type is PVMG (or greater). This is the 4277 case statement for SVt_IV, so this cannot be true (whatever gcov 4278 may say). */ 4279 assert(!SvTAINTED(ssv)); 4280 return; 4281 } 4282 if (!SvROK(ssv)) 4283 goto undef_sstr; 4284 if (dtype < SVt_PV && dtype != SVt_IV) 4285 sv_upgrade(dsv, SVt_IV); 4286 break; 4287 4288 case SVt_NV: 4289 if (LIKELY( SvNOK(ssv) )) { 4290 switch (dtype) { 4291 case SVt_NULL: 4292 case SVt_IV: 4293 sv_upgrade(dsv, SVt_NV); 4294 break; 4295 case SVt_PV: 4296 case SVt_PVIV: 4297 sv_upgrade(dsv, SVt_PVNV); 4298 break; 4299 case SVt_PVGV: 4300 case SVt_PVLV: 4301 goto end_of_first_switch; 4302 } 4303 SvNV_set(dsv, SvNVX(ssv)); 4304 (void)SvNOK_only(dsv); 4305 /* SvTAINTED can only be true if the SV has taint magic, which in 4306 turn means that the SV type is PVMG (or greater). This is the 4307 case statement for SVt_NV, so this cannot be true (whatever gcov 4308 may say). */ 4309 assert(!SvTAINTED(ssv)); 4310 return; 4311 } 4312 goto undef_sstr; 4313 4314 case SVt_PV: 4315 if (dtype < SVt_PV) 4316 sv_upgrade(dsv, SVt_PV); 4317 break; 4318 case SVt_PVIV: 4319 if (dtype < SVt_PVIV) 4320 sv_upgrade(dsv, SVt_PVIV); 4321 break; 4322 case SVt_PVNV: 4323 if (dtype < SVt_PVNV) 4324 sv_upgrade(dsv, SVt_PVNV); 4325 break; 4326 4327 case SVt_INVLIST: 4328 invlist_clone(ssv, dsv); 4329 return; 4330 default: 4331 { 4332 const char * const type = sv_reftype(ssv,0); 4333 if (PL_op) 4334 /* diag_listed_as: Bizarre copy of %s */ 4335 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op)); 4336 else 4337 Perl_croak(aTHX_ "Bizarre copy of %s", type); 4338 } 4339 NOT_REACHED; /* NOTREACHED */ 4340 4341 case SVt_REGEXP: 4342 upgregexp: 4343 if (dtype < SVt_REGEXP) 4344 sv_upgrade(dsv, SVt_REGEXP); 4345 break; 4346 4347 case SVt_PVLV: 4348 case SVt_PVGV: 4349 case SVt_PVMG: 4350 if (SvGMAGICAL(ssv) && (flags & SV_GMAGIC)) { 4351 mg_get(ssv); 4352 if (SvTYPE(ssv) != stype) 4353 stype = SvTYPE(ssv); 4354 } 4355 if (isGV_with_GP(ssv) && dtype <= SVt_PVLV) { 4356 glob_assign_glob(dsv, ssv, dtype); 4357 return; 4358 } 4359 if (stype == SVt_PVLV) 4360 { 4361 if (isREGEXP(ssv)) goto upgregexp; 4362 SvUPGRADE(dsv, SVt_PVNV); 4363 } 4364 else 4365 SvUPGRADE(dsv, (svtype)stype); 4366 } 4367 end_of_first_switch: 4368 4369 /* dsv may have been upgraded. */ 4370 dtype = SvTYPE(dsv); 4371 sflags = SvFLAGS(ssv); 4372 4373 if (UNLIKELY( dtype == SVt_PVCV )) { 4374 /* Assigning to a subroutine sets the prototype. */ 4375 if (SvOK(ssv)) { 4376 STRLEN len; 4377 const char *const ptr = SvPV_const(ssv, len); 4378 4379 SvGROW(dsv, len + 1); 4380 Copy(ptr, SvPVX(dsv), len + 1, char); 4381 SvCUR_set(dsv, len); 4382 SvPOK_only(dsv); 4383 SvFLAGS(dsv) |= sflags & SVf_UTF8; 4384 CvAUTOLOAD_off(dsv); 4385 } else { 4386 SvOK_off(dsv); 4387 } 4388 } 4389 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV 4390 || dtype == SVt_PVFM)) 4391 { 4392 const char * const type = sv_reftype(dsv,0); 4393 if (PL_op) 4394 /* diag_listed_as: Cannot copy to %s */ 4395 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op)); 4396 else 4397 Perl_croak(aTHX_ "Cannot copy to %s", type); 4398 } else if (sflags & SVf_ROK) { 4399 if (isGV_with_GP(dsv) 4400 && SvTYPE(SvRV(ssv)) == SVt_PVGV && isGV_with_GP(SvRV(ssv))) { 4401 ssv = SvRV(ssv); 4402 if (ssv == dsv) { 4403 if (GvIMPORTED(dsv) != GVf_IMPORTED 4404 && CopSTASH_ne(PL_curcop, GvSTASH(dsv))) 4405 { 4406 GvIMPORTED_on(dsv); 4407 } 4408 GvMULTI_on(dsv); 4409 return; 4410 } 4411 glob_assign_glob(dsv, ssv, dtype); 4412 return; 4413 } 4414 4415 if (dtype >= SVt_PV) { 4416 if (isGV_with_GP(dsv)) { 4417 gv_setref(dsv, ssv); 4418 return; 4419 } 4420 if (SvPVX_const(dsv)) { 4421 SvPV_free(dsv); 4422 SvLEN_set(dsv, 0); 4423 SvCUR_set(dsv, 0); 4424 } 4425 } 4426 (void)SvOK_off(dsv); 4427 SvRV_set(dsv, SvREFCNT_inc(SvRV(ssv))); 4428 SvFLAGS(dsv) |= sflags & SVf_ROK; 4429 assert(!(sflags & SVp_NOK)); 4430 assert(!(sflags & SVp_IOK)); 4431 assert(!(sflags & SVf_NOK)); 4432 assert(!(sflags & SVf_IOK)); 4433 } 4434 else if (isGV_with_GP(dsv)) { 4435 if (!(sflags & SVf_OK)) { 4436 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), 4437 "Undefined value assigned to typeglob"); 4438 } 4439 else { 4440 GV *gv = gv_fetchsv_nomg(ssv, GV_ADD, SVt_PVGV); 4441 if (dsv != (const SV *)gv) { 4442 const char * const name = GvNAME((const GV *)dsv); 4443 const STRLEN len = GvNAMELEN(dsv); 4444 HV *old_stash = NULL; 4445 bool reset_isa = FALSE; 4446 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':') 4447 || (len == 1 && name[0] == ':')) { 4448 /* Set aside the old stash, so we can reset isa caches 4449 on its subclasses. */ 4450 if((old_stash = GvHV(dsv))) { 4451 /* Make sure we do not lose it early. */ 4452 SvREFCNT_inc_simple_void_NN( 4453 sv_2mortal((SV *)old_stash) 4454 ); 4455 } 4456 reset_isa = TRUE; 4457 } 4458 4459 if (GvGP(dsv)) { 4460 SvREFCNT_inc_simple_void_NN(sv_2mortal(dsv)); 4461 gp_free(MUTABLE_GV(dsv)); 4462 } 4463 GvGP_set(dsv, gp_ref(GvGP(gv))); 4464 4465 if (reset_isa) { 4466 HV * const stash = GvHV(dsv); 4467 if( 4468 old_stash ? HvHasENAME(old_stash) : cBOOL(stash) 4469 ) 4470 mro_package_moved( 4471 stash, old_stash, 4472 (GV *)dsv, 0 4473 ); 4474 } 4475 } 4476 } 4477 } 4478 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV) 4479 && (stype == SVt_REGEXP || isREGEXP(ssv))) { 4480 reg_temp_copy((REGEXP*)dsv, (REGEXP*)ssv); 4481 } 4482 else if (sflags & SVp_POK) { 4483 const STRLEN cur = SvCUR(ssv); 4484 const STRLEN len = SvLEN(ssv); 4485 4486 /* 4487 * We have three basic ways to copy the string: 4488 * 4489 * 1. Swipe 4490 * 2. Copy-on-write 4491 * 3. Actual copy 4492 * 4493 * Which we choose is based on various factors. The following 4494 * things are listed in order of speed, fastest to slowest: 4495 * - Swipe 4496 * - Copying a short string 4497 * - Copy-on-write bookkeeping 4498 * - malloc 4499 * - Copying a long string 4500 * 4501 * We swipe the string (steal the string buffer) if the SV on the 4502 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a 4503 * big win on long strings. It should be a win on short strings if 4504 * SvPVX_const(dsv) has to be allocated. If not, it should not 4505 * slow things down, as SvPVX_const(ssv) would have been freed 4506 * soon anyway. 4507 * 4508 * We also steal the buffer from a PADTMP (operator target) if it 4509 * is ���long enough���. For short strings, a swipe does not help 4510 * here, as it causes more malloc calls the next time the target 4511 * is used. Benchmarks show that even if SvPVX_const(dsv) has to 4512 * be allocated it is still not worth swiping PADTMPs for short 4513 * strings, as the savings here are small. 4514 * 4515 * If swiping is not an option, then we see whether it is worth using 4516 * copy-on-write. If the lhs already has a buffer big enough and the 4517 * string is short, we skip it and fall back to method 3, since memcpy 4518 * is faster for short strings than the later bookkeeping overhead that 4519 * copy-on-write entails. 4520 4521 * If the rhs is not a copy-on-write string yet, then we also 4522 * consider whether the buffer is too large relative to the string 4523 * it holds. Some operations such as readline allocate a large 4524 * buffer in the expectation of reusing it. But turning such into 4525 * a COW buffer is counter-productive because it increases memory 4526 * usage by making readline allocate a new large buffer the sec- 4527 * ond time round. So, if the buffer is too large, again, we use 4528 * method 3 (copy). 4529 * 4530 * Finally, if there is no buffer on the left, or the buffer is too 4531 * small, then we use copy-on-write and make both SVs share the 4532 * string buffer. 4533 * 4534 */ 4535 4536 /* Whichever path we take through the next code, we want this true, 4537 and doing it now facilitates the COW check. */ 4538 (void)SvPOK_only(dsv); 4539 4540 if ( 4541 ( /* Either ... */ 4542 /* slated for free anyway (and not COW)? */ 4543 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP 4544 /* or a swipable TARG */ 4545 || ((sflags & 4546 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW)) 4547 == SVs_PADTMP 4548 /* whose buffer is worth stealing */ 4549 && CHECK_COWBUF_THRESHOLD(cur,len) 4550 ) 4551 ) && 4552 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */ 4553 (!(flags & SV_NOSTEAL)) && 4554 /* and we're allowed to steal temps */ 4555 SvREFCNT(ssv) == 1 && /* and no other references to it? */ 4556 len) /* and really is a string */ 4557 { /* Passes the swipe test. */ 4558 if (SvPVX_const(dsv)) /* we know that dtype >= SVt_PV */ 4559 SvPV_free(dsv); 4560 SvPV_set(dsv, SvPVX_mutable(ssv)); 4561 SvLEN_set(dsv, SvLEN(ssv)); 4562 SvCUR_set(dsv, SvCUR(ssv)); 4563 4564 SvTEMP_off(dsv); 4565 (void)SvOK_off(ssv); /* NOTE: nukes most SvFLAGS on ssv */ 4566 SvPV_set(ssv, NULL); 4567 SvLEN_set(ssv, 0); 4568 SvCUR_set(ssv, 0); 4569 SvTEMP_off(ssv); 4570 } 4571 /* We must check for SvIsCOW_static() even without 4572 * SV_COW_SHARED_HASH_KEYS being set or else we'll break SvIsBOOL() 4573 */ 4574 else if (SvIsCOW_static(ssv)) { 4575 if (SvPVX_const(dsv)) { /* we know that dtype >= SVt_PV */ 4576 SvPV_free(dsv); 4577 } 4578 SvPV_set(dsv, SvPVX(ssv)); 4579 SvLEN_set(dsv, 0); 4580 SvCUR_set(dsv, cur); 4581 SvFLAGS(dsv) |= (SVf_IsCOW|SVppv_STATIC); 4582 } 4583 else if (flags & SV_COW_SHARED_HASH_KEYS 4584 && 4585#ifdef PERL_COPY_ON_WRITE 4586 (sflags & SVf_IsCOW 4587 ? (!len || 4588 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dsv) < cur+1) 4589 /* If this is a regular (non-hek) COW, only so 4590 many COW "copies" are possible. */ 4591 && CowREFCNT(ssv) != SV_COW_REFCNT_MAX )) 4592 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS 4593 && !(SvFLAGS(dsv) & SVf_BREAK) 4594 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len 4595 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dsv) < cur+1) 4596 )) 4597#else 4598 sflags & SVf_IsCOW 4599 && !(SvFLAGS(dsv) & SVf_BREAK) 4600#endif 4601 ) { 4602 /* Either it's a shared hash key, or it's suitable for 4603 copy-on-write. */ 4604#ifdef DEBUGGING 4605 if (DEBUG_C_TEST) { 4606 PerlIO_printf(Perl_debug_log, "Copy on write: ssv --> dsv\n"); 4607 sv_dump(ssv); 4608 sv_dump(dsv); 4609 } 4610#endif 4611#ifdef PERL_ANY_COW 4612 if (!(sflags & SVf_IsCOW)) { 4613 SvIsCOW_on(ssv); 4614 CowREFCNT(ssv) = 0; 4615 } 4616#endif 4617 if (SvPVX_const(dsv)) { /* we know that dtype >= SVt_PV */ 4618 SvPV_free(dsv); 4619 } 4620 4621#ifdef PERL_ANY_COW 4622 if (len) { 4623 if (sflags & SVf_IsCOW) { 4624 sv_buf_to_rw(ssv); 4625 } 4626 CowREFCNT(ssv)++; 4627 SvPV_set(dsv, SvPVX_mutable(ssv)); 4628 sv_buf_to_ro(ssv); 4629 } else 4630#endif 4631 { 4632 /* SvIsCOW_shared_hash */ 4633 DEBUG_C(PerlIO_printf(Perl_debug_log, 4634 "Copy on write: Sharing hash\n")); 4635 4636 assert (SvTYPE(dsv) >= SVt_PV); 4637 SvPV_set(dsv, 4638 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(ssv))))); 4639 } 4640 SvLEN_set(dsv, len); 4641 SvCUR_set(dsv, cur); 4642 SvIsCOW_on(dsv); 4643 } else { 4644 /* Failed the swipe test, and we cannot do copy-on-write either. 4645 Have to copy the string. */ 4646 SvGROW(dsv, cur + 1); /* inlined from sv_setpvn */ 4647 Move(SvPVX_const(ssv),SvPVX(dsv),cur,char); 4648 SvCUR_set(dsv, cur); 4649 *SvEND(dsv) = '\0'; 4650 } 4651 if (sflags & SVp_NOK) { 4652 SvNV_set(dsv, SvNVX(ssv)); 4653 if ((sflags & SVf_NOK) && !(sflags & SVf_POK)) { 4654 /* Source was SVf_NOK|SVp_NOK|SVp_POK but not SVf_POK, meaning 4655 a value set as floating point and later stringified, where 4656 the value happens to be one of the few that we know aren't 4657 affected by the numeric locale, hence we can cache the 4658 stringification. Currently that's +Inf, -Inf and NaN, but 4659 conceivably we might extend this to -9 .. +9 (excluding -0). 4660 So mark destination the same: */ 4661 SvFLAGS(dsv) &= ~SVf_POK; 4662 } 4663 } 4664 if (sflags & SVp_IOK) { 4665 SvIV_set(dsv, SvIVX(ssv)); 4666 if (sflags & SVf_IVisUV) 4667 SvIsUV_on(dsv); 4668 if ((sflags & SVf_IOK) && !(sflags & SVf_POK)) { 4669 /* Source was SVf_IOK|SVp_IOK|SVp_POK but not SVf_POK, meaning 4670 a value set as an integer and later stringified. So mark 4671 destination the same: */ 4672 SvFLAGS(dsv) &= ~SVf_POK; 4673 } 4674 } 4675 SvFLAGS(dsv) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8); 4676 { 4677 const MAGIC * const smg = SvVSTRING_mg(ssv); 4678 if (smg) { 4679 sv_magic(dsv, NULL, PERL_MAGIC_vstring, 4680 smg->mg_ptr, smg->mg_len); 4681 SvRMAGICAL_on(dsv); 4682 } 4683 } 4684 } 4685 else if (sflags & (SVp_IOK|SVp_NOK)) { 4686 (void)SvOK_off(dsv); 4687 SvFLAGS(dsv) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK); 4688 if (sflags & SVp_IOK) { 4689 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */ 4690 SvIV_set(dsv, SvIVX(ssv)); 4691 } 4692 if (sflags & SVp_NOK) { 4693 SvNV_set(dsv, SvNVX(ssv)); 4694 } 4695 } 4696 else { 4697 if (isGV_with_GP(ssv)) { 4698 gv_efullname3(dsv, MUTABLE_GV(ssv), "*"); 4699 } 4700 else 4701 (void)SvOK_off(dsv); 4702 } 4703 if (SvTAINTED(ssv)) 4704 SvTAINT(dsv); 4705} 4706 4707 4708/* 4709=for apidoc sv_set_undef 4710 4711Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient. 4712Doesn't handle set magic. 4713 4714The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string 4715buffer, unlike C<undef $sv>. 4716 4717Introduced in perl 5.25.12. 4718 4719=cut 4720*/ 4721 4722void 4723Perl_sv_set_undef(pTHX_ SV *sv) 4724{ 4725 U32 type = SvTYPE(sv); 4726 4727 PERL_ARGS_ASSERT_SV_SET_UNDEF; 4728 4729 /* shortcut, NULL, IV, RV */ 4730 4731 if (type <= SVt_IV) { 4732 assert(!SvGMAGICAL(sv)); 4733 if (SvREADONLY(sv)) { 4734 /* does undeffing PL_sv_undef count as modifying a read-only 4735 * variable? Some XS code does this */ 4736 if (sv == &PL_sv_undef) 4737 return; 4738 Perl_croak_no_modify(); 4739 } 4740 4741 if (SvROK(sv)) { 4742 if (SvWEAKREF(sv)) 4743 sv_unref_flags(sv, 0); 4744 else { 4745 SV *rv = SvRV(sv); 4746 SvFLAGS(sv) = type; /* quickly turn off all flags */ 4747 SvREFCNT_dec_NN(rv); 4748 return; 4749 } 4750 } 4751 SvFLAGS(sv) = type; /* quickly turn off all flags */ 4752 return; 4753 } 4754 4755 if (SvIS_FREED(sv)) 4756 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p", 4757 (void *)sv); 4758 4759 SV_CHECK_THINKFIRST_COW_DROP(sv); 4760 4761 if (isGV_with_GP(sv)) 4762 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), 4763 "Undefined value assigned to typeglob"); 4764 else 4765 SvOK_off(sv); 4766} 4767 4768/* 4769=for apidoc sv_set_true 4770 4771Equivalent to C<sv_setsv(sv, &PL_sv_yes)>, but may be made more 4772efficient in the future. Doesn't handle set magic. 4773 4774The perl equivalent is C<$sv = !0;>. 4775 4776Introduced in perl 5.35.11. 4777 4778=cut 4779*/ 4780 4781void 4782Perl_sv_set_true(pTHX_ SV *sv) 4783{ 4784 PERL_ARGS_ASSERT_SV_SET_TRUE; 4785 sv_setsv(sv, &PL_sv_yes); 4786} 4787 4788/* 4789=for apidoc sv_set_false 4790 4791Equivalent to C<sv_setsv(sv, &PL_sv_no)>, but may be made more 4792efficient in the future. Doesn't handle set magic. 4793 4794The perl equivalent is C<$sv = !1;>. 4795 4796Introduced in perl 5.35.11. 4797 4798=cut 4799*/ 4800 4801void 4802Perl_sv_set_false(pTHX_ SV *sv) 4803{ 4804 PERL_ARGS_ASSERT_SV_SET_FALSE; 4805 sv_setsv(sv, &PL_sv_no); 4806} 4807 4808/* 4809=for apidoc sv_set_bool 4810 4811Equivalent to C<sv_setsv(sv, bool_val ? &Pl_sv_yes : &PL_sv_no)>, but 4812may be made more efficient in the future. Doesn't handle set magic. 4813 4814The perl equivalent is C<$sv = !!$expr;>. 4815 4816Introduced in perl 5.35.11. 4817 4818=cut 4819*/ 4820 4821void 4822Perl_sv_set_bool(pTHX_ SV *sv, const bool bool_val) 4823{ 4824 PERL_ARGS_ASSERT_SV_SET_BOOL; 4825 sv_setsv(sv, bool_val ? &PL_sv_yes : &PL_sv_no); 4826} 4827 4828 4829void 4830Perl_sv_setsv_mg(pTHX_ SV *const dsv, SV *const ssv) 4831{ 4832 PERL_ARGS_ASSERT_SV_SETSV_MG; 4833 4834 sv_setsv(dsv,ssv); 4835 SvSETMAGIC(dsv); 4836} 4837 4838#ifdef PERL_ANY_COW 4839# define SVt_COW SVt_PV 4840SV * 4841Perl_sv_setsv_cow(pTHX_ SV *dsv, SV *ssv) 4842{ 4843 STRLEN cur = SvCUR(ssv); 4844 STRLEN len = SvLEN(ssv); 4845 char *new_pv; 4846 U32 new_flags = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW); 4847#if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE) 4848 const bool already = cBOOL(SvIsCOW(ssv)); 4849#endif 4850 4851 PERL_ARGS_ASSERT_SV_SETSV_COW; 4852#ifdef DEBUGGING 4853 if (DEBUG_C_TEST) { 4854 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n", 4855 (void*)ssv, (void*)dsv); 4856 sv_dump(ssv); 4857 if (dsv) 4858 sv_dump(dsv); 4859 } 4860#endif 4861 if (dsv) { 4862 if (SvTHINKFIRST(dsv)) 4863 sv_force_normal_flags(dsv, SV_COW_DROP_PV); 4864 else if (SvPVX_const(dsv)) 4865 Safefree(SvPVX_mutable(dsv)); 4866 SvUPGRADE(dsv, SVt_COW); 4867 } 4868 else 4869 dsv = newSV_type(SVt_COW); 4870 4871 assert (SvPOK(ssv)); 4872 assert (SvPOKp(ssv)); 4873 4874 if (SvIsCOW(ssv)) { 4875 if (SvIsCOW_shared_hash(ssv)) { 4876 /* source is a COW shared hash key. */ 4877 DEBUG_C(PerlIO_printf(Perl_debug_log, 4878 "Fast copy on write: Sharing hash\n")); 4879 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(ssv)))); 4880 goto common_exit; 4881 } 4882 else if (SvIsCOW_static(ssv)) { 4883 /* source is static constant; preserve this */ 4884 new_pv = SvPVX(ssv); 4885 new_flags |= SVppv_STATIC; 4886 goto common_exit; 4887 } 4888 assert(SvCUR(ssv)+1 < SvLEN(ssv)); 4889 assert(CowREFCNT(ssv) < SV_COW_REFCNT_MAX); 4890 } else { 4891 assert ((SvFLAGS(ssv) & CAN_COW_MASK) == CAN_COW_FLAGS); 4892 SvUPGRADE(ssv, SVt_COW); 4893 SvIsCOW_on(ssv); 4894 DEBUG_C(PerlIO_printf(Perl_debug_log, 4895 "Fast copy on write: Converting ssv to COW\n")); 4896 CowREFCNT(ssv) = 0; 4897 } 4898# ifdef PERL_DEBUG_READONLY_COW 4899 if (already) sv_buf_to_rw(ssv); 4900# endif 4901 CowREFCNT(ssv)++; 4902 new_pv = SvPVX_mutable(ssv); 4903 sv_buf_to_ro(ssv); 4904 4905 common_exit: 4906 SvPV_set(dsv, new_pv); 4907 SvFLAGS(dsv) = new_flags; 4908 if (SvUTF8(ssv)) 4909 SvUTF8_on(dsv); 4910 SvLEN_set(dsv, len); 4911 SvCUR_set(dsv, cur); 4912#ifdef DEBUGGING 4913 if (DEBUG_C_TEST) 4914 sv_dump(dsv); 4915#endif 4916 return dsv; 4917} 4918#endif 4919 4920/* 4921=for apidoc sv_setpv_bufsize 4922 4923Sets the SV to be a string of cur bytes length, with at least 4924len bytes available. Ensures that there is a null byte at SvEND. 4925Returns a char * pointer to the SvPV buffer. 4926 4927=cut 4928*/ 4929 4930char * 4931Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len) 4932{ 4933 char *pv; 4934 4935 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE; 4936 4937 SV_CHECK_THINKFIRST_COW_DROP(sv); 4938 SvUPGRADE(sv, SVt_PV); 4939 pv = SvGROW(sv, len + 1); 4940 SvCUR_set(sv, cur); 4941 *(SvEND(sv))= '\0'; 4942 (void)SvPOK_only_UTF8(sv); /* validate pointer */ 4943 4944 SvTAINT(sv); 4945 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv); 4946 return pv; 4947} 4948 4949/* 4950=for apidoc sv_setpv 4951=for apidoc_item sv_setpv_mg 4952=for apidoc_item sv_setpvn 4953=for apidoc_item sv_setpvn_fresh 4954=for apidoc_item sv_setpvn_mg 4955=for apidoc_item |void|sv_setpvs|SV* sv|"literal string" 4956=for apidoc_item |void|sv_setpvs_mg|SV* sv|"literal string" 4957 4958These copy a string into the SV C<sv>, making sure it is C<L</SvPOK_only>>. 4959 4960In the C<pvs> forms, the string must be a C literal string, enclosed in double 4961quotes. 4962 4963In the C<pvn> forms, the first byte of the string is pointed to by C<ptr>, and 4964C<len> indicates the number of bytes to be copied, potentially including 4965embedded C<NUL> characters. 4966 4967In the plain C<pv> forms, C<ptr> points to a NUL-terminated C string. That is, 4968it points to the first byte of the string, and the copy proceeds up through the 4969first encountered C<NUL> byte. 4970 4971In the forms that take a C<ptr> argument, if it is NULL, the SV will become 4972undefined. 4973 4974The UTF-8 flag is not changed by these functions. A terminating NUL byte is 4975guaranteed in the result. 4976 4977The C<_mg> forms handle 'set' magic; the other forms skip all magic. 4978 4979C<sv_setpvn_fresh> is a cut-down alternative to C<sv_setpvn>, intended ONLY 4980to be used with a fresh sv that has been upgraded to a SVt_PV, SVt_PVIV, 4981SVt_PVNV, or SVt_PVMG. 4982 4983=cut 4984*/ 4985 4986void 4987Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len) 4988{ 4989 char *dptr; 4990 4991 PERL_ARGS_ASSERT_SV_SETPVN; 4992 4993 SV_CHECK_THINKFIRST_COW_DROP(sv); 4994 if (isGV_with_GP(sv)) 4995 Perl_croak_no_modify(); 4996 if (!ptr) { 4997 (void)SvOK_off(sv); 4998 return; 4999 } 5000 else { 5001 /* len is STRLEN which is unsigned, need to copy to signed */ 5002 const IV iv = len; 5003 if (iv < 0) 5004 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %" 5005 IVdf, iv); 5006 } 5007 SvUPGRADE(sv, SVt_PV); 5008 5009 dptr = SvGROW(sv, len + 1); 5010 Move(ptr,dptr,len,char); 5011 dptr[len] = '\0'; 5012 SvCUR_set(sv, len); 5013 (void)SvPOK_only_UTF8(sv); /* validate pointer */ 5014 SvTAINT(sv); 5015 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv); 5016} 5017 5018void 5019Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len) 5020{ 5021 PERL_ARGS_ASSERT_SV_SETPVN_MG; 5022 5023 sv_setpvn(sv,ptr,len); 5024 SvSETMAGIC(sv); 5025} 5026 5027void 5028Perl_sv_setpvn_fresh(pTHX_ SV *const sv, const char *const ptr, const STRLEN len) 5029{ 5030 char *dptr; 5031 5032 PERL_ARGS_ASSERT_SV_SETPVN_FRESH; 5033 assert(SvTYPE(sv) >= SVt_PV && SvTYPE(sv) <= SVt_PVMG); 5034 assert(!SvTHINKFIRST(sv)); 5035 assert(!isGV_with_GP(sv)); 5036 5037 if (ptr) { 5038 const IV iv = len; 5039 /* len is STRLEN which is unsigned, need to copy to signed */ 5040 if (iv < 0) 5041 Perl_croak(aTHX_ "panic: sv_setpvn_fresh called with negative strlen %" 5042 IVdf, iv); 5043 5044 dptr = sv_grow_fresh(sv, len + 1); 5045 Move(ptr,dptr,len,char); 5046 dptr[len] = '\0'; 5047 SvCUR_set(sv, len); 5048 SvPOK_on(sv); 5049 SvTAINT(sv); 5050 } 5051} 5052 5053void 5054Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr) 5055{ 5056 STRLEN len; 5057 5058 PERL_ARGS_ASSERT_SV_SETPV; 5059 5060 SV_CHECK_THINKFIRST_COW_DROP(sv); 5061 if (!ptr) { 5062 (void)SvOK_off(sv); 5063 return; 5064 } 5065 len = strlen(ptr); 5066 SvUPGRADE(sv, SVt_PV); 5067 5068 SvGROW(sv, len + 1); 5069 Move(ptr,SvPVX(sv),len+1,char); 5070 SvCUR_set(sv, len); 5071 (void)SvPOK_only_UTF8(sv); /* validate pointer */ 5072 SvTAINT(sv); 5073 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv); 5074} 5075 5076void 5077Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr) 5078{ 5079 PERL_ARGS_ASSERT_SV_SETPV_MG; 5080 5081 sv_setpv(sv,ptr); 5082 SvSETMAGIC(sv); 5083} 5084 5085void 5086Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek) 5087{ 5088 PERL_ARGS_ASSERT_SV_SETHEK; 5089 5090 if (!hek) { 5091 return; 5092 } 5093 5094 if (HEK_LEN(hek) == HEf_SVKEY) { 5095 sv_setsv(sv, *(SV**)HEK_KEY(hek)); 5096 return; 5097 } else { 5098 const int flags = HEK_FLAGS(hek); 5099 if (flags & HVhek_WASUTF8) { 5100 STRLEN utf8_len = HEK_LEN(hek); 5101 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len); 5102 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL); 5103 SvUTF8_on(sv); 5104 return; 5105 } else if (flags & HVhek_NOTSHARED) { 5106 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek)); 5107 if (HEK_UTF8(hek)) 5108 SvUTF8_on(sv); 5109 else SvUTF8_off(sv); 5110 return; 5111 } 5112 { 5113 SV_CHECK_THINKFIRST_COW_DROP(sv); 5114 SvUPGRADE(sv, SVt_PV); 5115 SvPV_free(sv); 5116 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek))); 5117 SvCUR_set(sv, HEK_LEN(hek)); 5118 SvLEN_set(sv, 0); 5119 SvIsCOW_on(sv); 5120 SvPOK_on(sv); 5121 if (HEK_UTF8(hek)) 5122 SvUTF8_on(sv); 5123 else SvUTF8_off(sv); 5124 return; 5125 } 5126 } 5127} 5128 5129 5130/* 5131=for apidoc sv_usepvn 5132=for apidoc_item sv_usepvn_flags 5133=for apidoc_item sv_usepvn_mg 5134 5135These tell an SV to use C<ptr> for its string value. Normally SVs have 5136their string stored inside the SV, but these tell the SV to use an 5137external string instead. 5138 5139C<ptr> should point to memory that was allocated 5140by L</C<Newx>>. It must be 5141the start of a C<Newx>-ed block of memory, and not a pointer to the 5142middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write), 5143and not be from a non-C<Newx> memory allocator like C<malloc>. The 5144string length, C<len>, must be supplied. By default this function 5145will L</C<Renew>> (i.e. realloc, move) the memory pointed to by C<ptr>, 5146so that the pointer should not be freed or used by the programmer after giving 5147it to C<sv_usepvn>, and neither should any pointers from "behind" that pointer 5148(I<e.g.>, S<C<ptr> + 1>) be used. 5149 5150In the C<sv_usepvn_flags> form, if S<C<flags & SV_SMAGIC>> is true, 5151C<SvSETMAGIC> is called before returning. 5152And if S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be 5153C<NUL>, and the realloc will be skipped (I<i.e.>, the buffer is actually at 5154least 1 byte longer than C<len>, and already meets the requirements for storing 5155in C<SvPVX>). 5156 5157C<sv_usepvn> is merely C<sv_usepvn_flags> with C<flags> set to 0, so 'set' 5158magic is skipped. 5159 5160C<sv_usepvn_mg> is merely C<sv_usepvn_flags> with C<flags> set to C<SV_SMAGIC>, 5161so 'set' magic is performed. 5162 5163=for apidoc Amnh||SV_SMAGIC 5164=for apidoc Amnh||SV_HAS_TRAILING_NUL 5165 5166=cut 5167*/ 5168 5169void 5170Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags) 5171{ 5172 STRLEN allocate; 5173 5174 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS; 5175 5176 SV_CHECK_THINKFIRST_COW_DROP(sv); 5177 SvUPGRADE(sv, SVt_PV); 5178 if (!ptr) { 5179 (void)SvOK_off(sv); 5180 if (flags & SV_SMAGIC) 5181 SvSETMAGIC(sv); 5182 return; 5183 } 5184 if (SvPVX_const(sv)) 5185 SvPV_free(sv); 5186 5187#ifdef DEBUGGING 5188 if (flags & SV_HAS_TRAILING_NUL) 5189 assert(ptr[len] == '\0'); 5190#endif 5191 5192 allocate = (flags & SV_HAS_TRAILING_NUL) 5193 ? len + 1 : 5194#ifdef Perl_safesysmalloc_size 5195 len + 1; 5196#else 5197 PERL_STRLEN_ROUNDUP(len + 1); 5198#endif 5199 if (flags & SV_HAS_TRAILING_NUL) { 5200 /* It's long enough - do nothing. 5201 Specifically Perl_newCONSTSUB is relying on this. */ 5202 } else { 5203#ifdef DEBUGGING 5204 /* Force a move to shake out bugs in callers. */ 5205 char *new_ptr = (char*)safemalloc(allocate); 5206 Copy(ptr, new_ptr, len, char); 5207 PoisonFree(ptr,len,char); 5208 Safefree(ptr); 5209 ptr = new_ptr; 5210#else 5211 ptr = (char*) saferealloc (ptr, allocate); 5212#endif 5213 } 5214#ifdef Perl_safesysmalloc_size 5215 SvLEN_set(sv, Perl_safesysmalloc_size(ptr)); 5216#else 5217 SvLEN_set(sv, allocate); 5218#endif 5219 SvCUR_set(sv, len); 5220 SvPV_set(sv, ptr); 5221 if (!(flags & SV_HAS_TRAILING_NUL)) { 5222 ptr[len] = '\0'; 5223 } 5224 (void)SvPOK_only_UTF8(sv); /* validate pointer */ 5225 SvTAINT(sv); 5226 if (flags & SV_SMAGIC) 5227 SvSETMAGIC(sv); 5228} 5229 5230 5231static void 5232S_sv_uncow(pTHX_ SV * const sv, const U32 flags) 5233{ 5234 assert(SvIsCOW(sv)); 5235 { 5236#ifdef PERL_ANY_COW 5237 const char * const pvx = SvPVX_const(sv); 5238 const STRLEN len = SvLEN(sv); 5239 const STRLEN cur = SvCUR(sv); 5240 const bool was_shared_hek = SvIsCOW_shared_hash(sv); 5241 5242#ifdef DEBUGGING 5243 if (DEBUG_C_TEST) { 5244 PerlIO_printf(Perl_debug_log, 5245 "Copy on write: Force normal %ld\n", 5246 (long) flags); 5247 sv_dump(sv); 5248 } 5249#endif 5250 SvIsCOW_off(sv); 5251# ifdef PERL_COPY_ON_WRITE 5252 if (len) { 5253 /* Must do this first, since the CowREFCNT uses SvPVX and 5254 we need to write to CowREFCNT, or de-RO the whole buffer if we are 5255 the only owner left of the buffer. */ 5256 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */ 5257 { 5258 U8 cowrefcnt = CowREFCNT(sv); 5259 if(cowrefcnt != 0) { 5260 cowrefcnt--; 5261 CowREFCNT(sv) = cowrefcnt; 5262 sv_buf_to_ro(sv); 5263 goto copy_over; 5264 } 5265 } 5266 /* Else we are the only owner of the buffer. */ 5267 } 5268 else 5269# endif 5270 { 5271 /* This SV doesn't own the buffer, so need to Newx() a new one: */ 5272 copy_over: 5273 SvPV_set(sv, NULL); 5274 SvCUR_set(sv, 0); 5275 SvLEN_set(sv, 0); 5276 if (flags & SV_COW_DROP_PV) { 5277 /* OK, so we don't need to copy our buffer. */ 5278 SvPOK_off(sv); 5279 } else { 5280 SvGROW(sv, cur + 1); 5281 Move(pvx,SvPVX(sv),cur,char); 5282 SvCUR_set(sv, cur); 5283 *SvEND(sv) = '\0'; 5284 } 5285 if (was_shared_hek) { 5286 unshare_hek(SvSHARED_HEK_FROM_PV(pvx)); 5287 } 5288#ifdef DEBUGGING 5289 if (DEBUG_C_TEST) 5290 sv_dump(sv); 5291#endif 5292 } 5293#else 5294 const char * const pvx = SvPVX_const(sv); 5295 const STRLEN len = SvCUR(sv); 5296 SvIsCOW_off(sv); 5297 SvPV_set(sv, NULL); 5298 SvLEN_set(sv, 0); 5299 if (flags & SV_COW_DROP_PV) { 5300 /* OK, so we don't need to copy our buffer. */ 5301 SvPOK_off(sv); 5302 } else { 5303 SvGROW(sv, len + 1); 5304 Move(pvx,SvPVX(sv),len,char); 5305 *SvEND(sv) = '\0'; 5306 } 5307 unshare_hek(SvSHARED_HEK_FROM_PV(pvx)); 5308#endif 5309 } 5310} 5311 5312 5313/* 5314=for apidoc sv_force_normal_flags 5315 5316Undo various types of fakery on an SV, where fakery means 5317"more than" a string: if the PV is a shared string, make 5318a private copy; if we're a ref, stop refing; if we're a glob, downgrade to 5319an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when 5320we do the copy, and is also used locally; if this is a 5321vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set 5322then a copy-on-write scalar drops its PV buffer (if any) and becomes 5323C<SvPOK_off> rather than making a copy. (Used where this 5324scalar is about to be set to some other value.) In addition, 5325the C<flags> parameter gets passed to C<sv_unref_flags()> 5326when unreffing. C<sv_force_normal> calls this function 5327with flags set to 0. 5328 5329This function is expected to be used to signal to perl that this SV is 5330about to be written to, and any extra book-keeping needs to be taken care 5331of. Hence, it croaks on read-only values. 5332 5333=for apidoc Amnh||SV_COW_DROP_PV 5334 5335=cut 5336*/ 5337 5338void 5339Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags) 5340{ 5341 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS; 5342 5343 if (SvREADONLY(sv)) 5344 Perl_croak_no_modify(); 5345 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV)) 5346 S_sv_uncow(aTHX_ sv, flags); 5347 if (SvROK(sv)) 5348 sv_unref_flags(sv, flags); 5349 else if (SvFAKE(sv) && isGV_with_GP(sv)) 5350 sv_unglob(sv, flags); 5351 else if (SvFAKE(sv) && isREGEXP(sv)) { 5352 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous 5353 to sv_unglob. We only need it here, so inline it. */ 5354 const bool islv = SvTYPE(sv) == SVt_PVLV; 5355 const svtype new_type = 5356 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV; 5357 SV *const temp = newSV_type(new_type); 5358 regexp *old_rx_body; 5359 5360 if (new_type == SVt_PVMG) { 5361 SvMAGIC_set(temp, SvMAGIC(sv)); 5362 SvMAGIC_set(sv, NULL); 5363 SvSTASH_set(temp, SvSTASH(sv)); 5364 SvSTASH_set(sv, NULL); 5365 } 5366 if (!islv) 5367 SvCUR_set(temp, SvCUR(sv)); 5368 /* Remember that SvPVX is in the head, not the body. */ 5369 assert(ReANY((REGEXP *)sv)->mother_re); 5370 5371 if (islv) { 5372 /* LV-as-regex has sv->sv_any pointing to an XPVLV body, 5373 * whose xpvlenu_rx field points to the regex body */ 5374 XPV *xpv = (XPV*)(SvANY(sv)); 5375 old_rx_body = xpv->xpv_len_u.xpvlenu_rx; 5376 xpv->xpv_len_u.xpvlenu_rx = NULL; 5377 } 5378 else 5379 old_rx_body = ReANY((REGEXP *)sv); 5380 5381 /* Their buffer is already owned by someone else. */ 5382 if (flags & SV_COW_DROP_PV) { 5383 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new 5384 zeroed body. For SVt_PVLV, we zeroed it above (len field 5385 a union with xpvlenu_rx) */ 5386 assert(!SvLEN(islv ? sv : temp)); 5387 sv->sv_u.svu_pv = 0; 5388 } 5389 else { 5390 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv)); 5391 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1); 5392 SvPOK_on(sv); 5393 } 5394 5395 /* Now swap the rest of the bodies. */ 5396 5397 SvFAKE_off(sv); 5398 if (!islv) { 5399 SvFLAGS(sv) &= ~SVTYPEMASK; 5400 SvFLAGS(sv) |= new_type; 5401 SvANY(sv) = SvANY(temp); 5402 } 5403 5404 SvFLAGS(temp) &= ~(SVTYPEMASK); 5405 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE; 5406 SvANY(temp) = old_rx_body; 5407 5408 /* temp is now rebuilt as a correctly structured SVt_REGEXP, so this 5409 * will trigger a call to sv_clear() which will correctly free the 5410 * body. */ 5411 SvREFCNT_dec_NN(temp); 5412 } 5413 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring); 5414} 5415 5416/* 5417=for apidoc sv_chop 5418 5419Efficient removal of characters from the beginning of the string buffer. 5420C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a 5421pointer to somewhere inside the string buffer. C<ptr> becomes the first 5422character of the adjusted string. Uses the C<OOK> hack. On return, only 5423C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true. 5424 5425Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer 5426refer to the same chunk of data. 5427 5428The unfortunate similarity of this function's name to that of Perl's C<chop> 5429operator is strictly coincidental. This function works from the left; 5430C<chop> works from the right. 5431 5432=cut 5433*/ 5434 5435void 5436Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr) 5437{ 5438 STRLEN delta; 5439 STRLEN old_delta; 5440 U8 *p; 5441#ifdef DEBUGGING 5442 const U8 *evacp; 5443 STRLEN evacn; 5444#endif 5445 STRLEN max_delta; 5446 5447 PERL_ARGS_ASSERT_SV_CHOP; 5448 5449 if (!ptr || !SvPOKp(sv)) 5450 return; 5451 delta = ptr - SvPVX_const(sv); 5452 if (!delta) { 5453 /* Nothing to do. */ 5454 return; 5455 } 5456 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv); 5457 if (delta > max_delta) 5458 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p", 5459 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta); 5460 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */ 5461 SV_CHECK_THINKFIRST(sv); 5462 SvPOK_only_UTF8(sv); 5463 5464 if (!SvOOK(sv)) { 5465 if (!SvLEN(sv)) { /* make copy of shared string */ 5466 const char *pvx = SvPVX_const(sv); 5467 const STRLEN len = SvCUR(sv); 5468 SvGROW(sv, len + 1); 5469 Move(pvx,SvPVX(sv),len,char); 5470 *SvEND(sv) = '\0'; 5471 } 5472 SvOOK_on(sv); 5473 old_delta = 0; 5474 } else { 5475 SvOOK_offset(sv, old_delta); 5476 } 5477 SvLEN_set(sv, SvLEN(sv) - delta); 5478 SvCUR_set(sv, SvCUR(sv) - delta); 5479 SvPV_set(sv, SvPVX(sv) + delta); 5480 5481 p = (U8 *)SvPVX_const(sv); 5482 5483#ifdef DEBUGGING 5484 /* how many bytes were evacuated? we will fill them with sentinel 5485 bytes, except for the part holding the new offset of course. */ 5486 evacn = delta; 5487 if (old_delta) 5488 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN)); 5489 assert(evacn); 5490 assert(evacn <= delta + old_delta); 5491 evacp = p - evacn; 5492#endif 5493 5494 /* This sets 'delta' to the accumulated value of all deltas so far */ 5495 delta += old_delta; 5496 assert(delta); 5497 5498 /* If 'delta' fits in a byte, store it just prior to the new beginning of 5499 * the string; otherwise store a 0 byte there and store 'delta' just prior 5500 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a 5501 * portion of the chopped part of the string */ 5502 if (delta < 0x100) { 5503 *--p = (U8) delta; 5504 } else { 5505 *--p = 0; 5506 p -= sizeof(STRLEN); 5507 Copy((U8*)&delta, p, sizeof(STRLEN), U8); 5508 } 5509 5510#ifdef DEBUGGING 5511 /* Fill the preceding buffer with sentinals to verify that no-one is 5512 using it. */ 5513 while (p > evacp) { 5514 --p; 5515 *p = (U8)PTR2UV(p); 5516 } 5517#endif 5518} 5519 5520/* 5521=for apidoc sv_catpvn 5522=for apidoc_item sv_catpvn_flags 5523=for apidoc_item sv_catpvn_mg 5524=for apidoc_item sv_catpvn_nomg 5525 5526These concatenate the C<len> bytes of the string beginning at C<ptr> onto the 5527end of the string which is in C<dsv>. The caller must make sure C<ptr> 5528contains at least C<len> bytes. 5529 5530For all but C<sv_catpvn_flags>, the string appended is assumed to be valid 5531UTF-8 if the SV has the UTF-8 status set, and a string of bytes otherwise. 5532 5533They differ in that: 5534 5535C<sv_catpvn_mg> performs both 'get' and 'set' magic on C<dsv>. 5536 5537C<sv_catpvn> performs only 'get' magic. 5538 5539C<sv_catpvn_nomg> skips all magic. 5540 5541C<sv_catpvn_flags> has an extra C<flags> parameter which allows you to specify 5542any combination of magic handling (using C<SV_GMAGIC> and/or C<SV_SMAGIC>) and 5543to also override the UTF-8 handling. By supplying the C<SV_CATBYTES> flag, the 5544appended string is interpreted as plain bytes; by supplying instead the 5545C<SV_CATUTF8> flag, it will be interpreted as UTF-8, and the C<dsv> will be 5546upgraded to UTF-8 if necessary. 5547 5548C<sv_catpvn>, C<sv_catpvn_mg>, and C<sv_catpvn_nomg> are implemented 5549in terms of C<sv_catpvn_flags>. 5550 5551=for apidoc Amnh||SV_CATUTF8 5552=for apidoc Amnh||SV_CATBYTES 5553 5554=cut 5555*/ 5556 5557void 5558Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags) 5559{ 5560 STRLEN dlen; 5561 const char * const dstr = SvPV_force_flags(dsv, dlen, flags); 5562 5563 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS; 5564 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8)); 5565 5566 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) { 5567 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) { 5568 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1); 5569 dlen = SvCUR(dsv); 5570 } 5571 else SvGROW(dsv, dlen + slen + 3); 5572 if (sstr == dstr) 5573 sstr = SvPVX_const(dsv); 5574 Move(sstr, SvPVX(dsv) + dlen, slen, char); 5575 SvCUR_set(dsv, SvCUR(dsv) + slen); 5576 } 5577 else { 5578 /* We inline bytes_to_utf8, to avoid an extra malloc. */ 5579 const char * const send = sstr + slen; 5580 U8 *d; 5581 5582 /* Something this code does not account for, which I think is 5583 impossible; it would require the same pv to be treated as 5584 bytes *and* utf8, which would indicate a bug elsewhere. */ 5585 assert(sstr != dstr); 5586 5587 SvGROW(dsv, dlen + slen * 2 + 3); 5588 d = (U8 *)SvPVX(dsv) + dlen; 5589 5590 while (sstr < send) { 5591 append_utf8_from_native_byte(*sstr, &d); 5592 sstr++; 5593 } 5594 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv)); 5595 } 5596 *SvEND(dsv) = '\0'; 5597 (void)SvPOK_only_UTF8(dsv); /* validate pointer */ 5598 SvTAINT(dsv); 5599 if (flags & SV_SMAGIC) 5600 SvSETMAGIC(dsv); 5601} 5602 5603/* 5604=for apidoc sv_catsv 5605=for apidoc_item sv_catsv_flags 5606=for apidoc_item sv_catsv_mg 5607=for apidoc_item sv_catsv_nomg 5608 5609These concatenate the string from SV C<sstr> onto the end of the string in SV 5610C<dsv>. If C<sstr> is null, these are no-ops; otherwise only C<dsv> is 5611modified. 5612 5613They differ only in what magic they perform: 5614 5615C<sv_catsv_mg> performs 'get' magic on both SVs before the copy, and 'set' magic 5616on C<dsv> afterwards. 5617 5618C<sv_catsv> performs just 'get' magic, on both SVs. 5619 5620C<sv_catsv_nomg> skips all magic. 5621 5622C<sv_catsv_flags> has an extra C<flags> parameter which allows you to use 5623C<SV_GMAGIC> and/or C<SV_SMAGIC> to specify any combination of magic handling 5624(although either both or neither SV will have 'get' magic applied to it.) 5625 5626C<sv_catsv>, C<sv_catsv_mg>, and C<sv_catsv_nomg> are implemented 5627in terms of C<sv_catsv_flags>. 5628 5629=cut */ 5630 5631void 5632Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const sstr, const I32 flags) 5633{ 5634 PERL_ARGS_ASSERT_SV_CATSV_FLAGS; 5635 5636 if (sstr) { 5637 STRLEN slen; 5638 const char *spv = SvPV_flags_const(sstr, slen, flags); 5639 if (flags & SV_GMAGIC) 5640 SvGETMAGIC(dsv); 5641 sv_catpvn_flags(dsv, spv, slen, 5642 DO_UTF8(sstr) ? SV_CATUTF8 : SV_CATBYTES); 5643 if (flags & SV_SMAGIC) 5644 SvSETMAGIC(dsv); 5645 } 5646} 5647 5648/* 5649=for apidoc sv_catpv 5650=for apidoc_item sv_catpv_flags 5651=for apidoc_item sv_catpv_mg 5652=for apidoc_item sv_catpv_nomg 5653 5654These concatenate the C<NUL>-terminated string C<sstr> onto the end of the 5655string which is in the SV. 5656If the SV has the UTF-8 status set, then the bytes appended should be 5657valid UTF-8. 5658 5659They differ only in how they handle magic: 5660 5661C<sv_catpv_mg> performs both 'get' and 'set' magic. 5662 5663C<sv_catpv> performs only 'get' magic. 5664 5665C<sv_catpv_nomg> skips all magic. 5666 5667C<sv_catpv_flags> has an extra C<flags> parameter which allows you to specify 5668any combination of magic handling (using C<SV_GMAGIC> and/or C<SV_SMAGIC>), and 5669to also override the UTF-8 handling. By supplying the C<SV_CATUTF8> flag, the 5670appended string is forced to be interpreted as UTF-8; by supplying instead the 5671C<SV_CATBYTES> flag, it will be interpreted as just bytes. Either the SV or 5672the string appended will be upgraded to UTF-8 if necessary. 5673 5674=cut 5675*/ 5676 5677void 5678Perl_sv_catpv(pTHX_ SV *const dsv, const char *sstr) 5679{ 5680 STRLEN len; 5681 STRLEN tlen; 5682 char *junk; 5683 5684 PERL_ARGS_ASSERT_SV_CATPV; 5685 5686 if (!sstr) 5687 return; 5688 junk = SvPV_force(dsv, tlen); 5689 len = strlen(sstr); 5690 SvGROW(dsv, tlen + len + 1); 5691 if (sstr == junk) 5692 sstr = SvPVX_const(dsv); 5693 Move(sstr,SvPVX(dsv)+tlen,len+1,char); 5694 SvCUR_set(dsv, SvCUR(dsv) + len); 5695 (void)SvPOK_only_UTF8(dsv); /* validate pointer */ 5696 SvTAINT(dsv); 5697} 5698 5699void 5700Perl_sv_catpv_flags(pTHX_ SV *dsv, const char *sstr, const I32 flags) 5701{ 5702 PERL_ARGS_ASSERT_SV_CATPV_FLAGS; 5703 sv_catpvn_flags(dsv, sstr, strlen(sstr), flags); 5704} 5705 5706void 5707Perl_sv_catpv_mg(pTHX_ SV *const dsv, const char *const sstr) 5708{ 5709 PERL_ARGS_ASSERT_SV_CATPV_MG; 5710 5711 sv_catpv(dsv,sstr); 5712 SvSETMAGIC(dsv); 5713} 5714 5715/* 5716=for apidoc newSV 5717 5718Creates a new SV. A non-zero C<len> parameter indicates the number of 5719bytes of preallocated string space the SV should have. An extra byte for a 5720trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string 5721space is allocated.) The reference count for the new SV is set to 1. 5722 5723In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first 5724parameter, I<x>, a debug aid which allowed callers to identify themselves. 5725This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see 5726L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS 5727modules supporting older perls. 5728 5729=cut 5730*/ 5731 5732SV * 5733Perl_newSV(pTHX_ const STRLEN len) 5734{ 5735 SV *sv; 5736 5737 if (!len) 5738 new_SV(sv); 5739 else { 5740 sv = newSV_type(SVt_PV); 5741 sv_grow_fresh(sv, len + 1); 5742 } 5743 return sv; 5744} 5745/* 5746=for apidoc sv_magicext 5747 5748Adds magic to an SV, upgrading it if necessary. Applies the 5749supplied C<vtable> and returns a pointer to the magic added. 5750 5751Note that C<sv_magicext> will allow things that C<sv_magic> will not. 5752In particular, you can add magic to C<SvREADONLY> SVs, and add more than 5753one instance of the same C<how>. 5754 5755If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is 5756stored, if C<namlen> is zero then C<name> is stored as-is and - as another 5757special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed 5758to contain an SV* and is stored as-is with its C<REFCNT> incremented. 5759 5760(This is now used as a subroutine by C<sv_magic>.) 5761 5762=cut 5763*/ 5764MAGIC * 5765Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how, 5766 const MGVTBL *const vtable, const char *const name, const I32 namlen) 5767{ 5768 MAGIC* mg; 5769 5770 PERL_ARGS_ASSERT_SV_MAGICEXT; 5771 5772 SvUPGRADE(sv, SVt_PVMG); 5773 Newxz(mg, 1, MAGIC); 5774 mg->mg_moremagic = SvMAGIC(sv); 5775 SvMAGIC_set(sv, mg); 5776 5777 /* Sometimes a magic contains a reference loop, where the sv and 5778 object refer to each other. To prevent a reference loop that 5779 would prevent such objects being freed, we look for such loops 5780 and if we find one we avoid incrementing the object refcount. 5781 5782 Note we cannot do this to avoid self-tie loops as intervening RV must 5783 have its REFCNT incremented to keep it in existence. 5784 5785 */ 5786 if (!obj || obj == sv || 5787 how == PERL_MAGIC_arylen || 5788 how == PERL_MAGIC_regdata || 5789 how == PERL_MAGIC_regdatum || 5790 how == PERL_MAGIC_symtab || 5791 (SvTYPE(obj) == SVt_PVGV && 5792 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv 5793 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv 5794 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv))) 5795 { 5796 mg->mg_obj = obj; 5797 } 5798 else { 5799 mg->mg_obj = SvREFCNT_inc_simple(obj); 5800 mg->mg_flags |= MGf_REFCOUNTED; 5801 } 5802 5803 /* Normal self-ties simply pass a null object, and instead of 5804 using mg_obj directly, use the SvTIED_obj macro to produce a 5805 new RV as needed. For glob "self-ties", we are tieing the PVIO 5806 with an RV obj pointing to the glob containing the PVIO. In 5807 this case, to avoid a reference loop, we need to weaken the 5808 reference. 5809 */ 5810 5811 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO && 5812 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv) 5813 { 5814 sv_rvweaken(obj); 5815 } 5816 5817 mg->mg_type = how; 5818 mg->mg_len = namlen; 5819 if (name) { 5820 if (namlen > 0) 5821 mg->mg_ptr = savepvn(name, namlen); 5822 else if (namlen == HEf_SVKEY) { 5823 /* Yes, this is casting away const. This is only for the case of 5824 HEf_SVKEY. I think we need to document this aberration of the 5825 constness of the API, rather than making name non-const, as 5826 that change propagating outwards a long way. */ 5827 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name); 5828 } else 5829 mg->mg_ptr = (char *) name; 5830 } 5831 mg->mg_virtual = (MGVTBL *) vtable; 5832 5833 mg_magical(sv); 5834 return mg; 5835} 5836 5837MAGIC * 5838Perl_sv_magicext_mglob(pTHX_ SV *sv) 5839{ 5840 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB; 5841 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') { 5842 /* This sv is only a delegate. //g magic must be attached to 5843 its target. */ 5844 vivify_defelem(sv); 5845 sv = LvTARG(sv); 5846 } 5847 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global, 5848 &PL_vtbl_mglob, 0, 0); 5849} 5850 5851/* 5852=for apidoc sv_magic 5853 5854Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if 5855necessary, then adds a new magic item of type C<how> to the head of the 5856magic list. 5857 5858See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the 5859handling of the C<name> and C<namlen> arguments. 5860 5861You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also 5862to add more than one instance of the same C<how>. 5863 5864=cut 5865*/ 5866 5867void 5868Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how, 5869 const char *const name, const I32 namlen) 5870{ 5871 const MGVTBL *vtable; 5872 MAGIC* mg; 5873 unsigned int flags; 5874 unsigned int vtable_index; 5875 5876 PERL_ARGS_ASSERT_SV_MAGIC; 5877 5878 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data) 5879 || ((flags = PL_magic_data[how]), 5880 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK) 5881 > magic_vtable_max)) 5882 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how); 5883 5884 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals. 5885 Useful for attaching extension internal data to perl vars. 5886 Note that multiple extensions may clash if magical scalars 5887 etc holding private data from one are passed to another. */ 5888 5889 vtable = (vtable_index == magic_vtable_max) 5890 ? NULL : PL_magic_vtables + vtable_index; 5891 5892 if (SvREADONLY(sv)) { 5893 if ( 5894 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how) 5895 ) 5896 { 5897 Perl_croak_no_modify(); 5898 } 5899 } 5900 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) { 5901 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) { 5902 /* sv_magic() refuses to add a magic of the same 'how' as an 5903 existing one 5904 */ 5905 if (how == PERL_MAGIC_taint) 5906 mg->mg_len |= 1; 5907 return; 5908 } 5909 } 5910 5911 /* Rest of work is done else where */ 5912 mg = sv_magicext(sv,obj,how,vtable,name,namlen); 5913 5914 switch (how) { 5915 case PERL_MAGIC_taint: 5916 mg->mg_len = 1; 5917 break; 5918 case PERL_MAGIC_ext: 5919 case PERL_MAGIC_dbfile: 5920 SvRMAGICAL_on(sv); 5921 break; 5922 } 5923} 5924 5925static int 5926S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, const MGVTBL *vtbl, const U32 flags) 5927{ 5928 MAGIC* mg; 5929 MAGIC** mgp; 5930 5931 assert(flags <= 1); 5932 5933 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv)) 5934 return 0; 5935 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic); 5936 for (mg = *mgp; mg; mg = *mgp) { 5937 const MGVTBL* const virt = mg->mg_virtual; 5938 if (mg->mg_type == type && (!flags || virt == vtbl)) { 5939 *mgp = mg->mg_moremagic; 5940 if (virt && virt->svt_free) 5941 virt->svt_free(aTHX_ sv, mg); 5942 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) { 5943 if (mg->mg_len > 0) 5944 Safefree(mg->mg_ptr); 5945 else if (mg->mg_len == HEf_SVKEY) 5946 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr)); 5947 else if (mg->mg_type == PERL_MAGIC_utf8) 5948 Safefree(mg->mg_ptr); 5949 } 5950 if (mg->mg_flags & MGf_REFCOUNTED) 5951 SvREFCNT_dec(mg->mg_obj); 5952 Safefree(mg); 5953 } 5954 else 5955 mgp = &mg->mg_moremagic; 5956 } 5957 if (SvMAGIC(sv)) { 5958 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */ 5959 mg_magical(sv); /* else fix the flags now */ 5960 } 5961 else 5962 SvMAGICAL_off(sv); 5963 5964 return 0; 5965} 5966 5967/* 5968=for apidoc sv_unmagic 5969 5970Removes all magic of type C<type> from an SV. 5971 5972=cut 5973*/ 5974 5975int 5976Perl_sv_unmagic(pTHX_ SV *const sv, const int type) 5977{ 5978 PERL_ARGS_ASSERT_SV_UNMAGIC; 5979 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0); 5980} 5981 5982/* 5983=for apidoc sv_unmagicext 5984 5985Removes all magic of type C<type> with the specified C<vtbl> from an SV. 5986 5987=cut 5988*/ 5989 5990int 5991Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, const MGVTBL *vtbl) 5992{ 5993 PERL_ARGS_ASSERT_SV_UNMAGICEXT; 5994 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1); 5995} 5996 5997/* 5998=for apidoc sv_rvweaken 5999 6000Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the 6001referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and 6002push a back-reference to this RV onto the array of backreferences 6003associated with that magic. If the RV is magical, set magic will be 6004called after the RV is cleared. Silently ignores C<undef> and warns 6005on already-weak references. 6006 6007=cut 6008*/ 6009 6010SV * 6011Perl_sv_rvweaken(pTHX_ SV *const sv) 6012{ 6013 SV *tsv; 6014 6015 PERL_ARGS_ASSERT_SV_RVWEAKEN; 6016 6017 if (!SvOK(sv)) /* let undefs pass */ 6018 return sv; 6019 if (!SvROK(sv)) 6020 Perl_croak(aTHX_ "Can't weaken a nonreference"); 6021 else if (SvWEAKREF(sv)) { 6022 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak"); 6023 return sv; 6024 } 6025 else if (SvREADONLY(sv)) croak_no_modify(); 6026 tsv = SvRV(sv); 6027 Perl_sv_add_backref(aTHX_ tsv, sv); 6028 SvWEAKREF_on(sv); 6029 SvREFCNT_dec_NN(tsv); 6030 return sv; 6031} 6032 6033/* 6034=for apidoc sv_rvunweaken 6035 6036Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove 6037the backreference to this RV from the array of backreferences 6038associated with the target SV, increment the refcount of the target. 6039Silently ignores C<undef> and warns on non-weak references. 6040 6041=cut 6042*/ 6043 6044SV * 6045Perl_sv_rvunweaken(pTHX_ SV *const sv) 6046{ 6047 SV *tsv; 6048 6049 PERL_ARGS_ASSERT_SV_RVUNWEAKEN; 6050 6051 if (!SvOK(sv)) /* let undefs pass */ 6052 return sv; 6053 if (!SvROK(sv)) 6054 Perl_croak(aTHX_ "Can't unweaken a nonreference"); 6055 else if (!SvWEAKREF(sv)) { 6056 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak"); 6057 return sv; 6058 } 6059 else if (SvREADONLY(sv)) croak_no_modify(); 6060 6061 tsv = SvRV(sv); 6062 SvWEAKREF_off(sv); 6063 SvROK_on(sv); 6064 SvREFCNT_inc_NN(tsv); 6065 Perl_sv_del_backref(aTHX_ tsv, sv); 6066 return sv; 6067} 6068 6069/* 6070=for apidoc sv_get_backrefs 6071 6072If C<sv> is the target of a weak reference then it returns the back 6073references structure associated with the sv; otherwise return C<NULL>. 6074 6075When returning a non-null result the type of the return is relevant. If it 6076is an AV then the elements of the AV are the weak reference RVs which 6077point at this item. If it is any other type then the item itself is the 6078weak reference. 6079 6080See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>, 6081C<Perl_sv_kill_backrefs()> 6082 6083=cut 6084*/ 6085 6086SV * 6087Perl_sv_get_backrefs(SV *const sv) 6088{ 6089 SV *backrefs= NULL; 6090 6091 PERL_ARGS_ASSERT_SV_GET_BACKREFS; 6092 6093 /* find slot to store array or singleton backref */ 6094 6095 if (SvTYPE(sv) == SVt_PVHV) { 6096 if (HvHasAUX(sv)) { 6097 struct xpvhv_aux * const iter = HvAUX((HV *)sv); 6098 backrefs = (SV *)iter->xhv_backreferences; 6099 } 6100 } else if (SvMAGICAL(sv)) { 6101 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref); 6102 if (mg) 6103 backrefs = mg->mg_obj; 6104 } 6105 return backrefs; 6106} 6107 6108/* Give tsv backref magic if it hasn't already got it, then push a 6109 * back-reference to sv onto the array associated with the backref magic. 6110 * 6111 * As an optimisation, if there's only one backref and it's not an AV, 6112 * store it directly in the HvAUX or mg_obj slot, avoiding the need to 6113 * allocate an AV. (Whether the slot holds an AV tells us whether this is 6114 * active.) 6115 */ 6116 6117/* A discussion about the backreferences array and its refcount: 6118 * 6119 * The AV holding the backreferences is pointed to either as the mg_obj of 6120 * PERL_MAGIC_backref, or in the specific case of a HV, from the 6121 * xhv_backreferences field. The array is created with a refcount 6122 * of 2. This means that if during global destruction the array gets 6123 * picked on before its parent to have its refcount decremented by the 6124 * random zapper, it won't actually be freed, meaning it's still there for 6125 * when its parent gets freed. 6126 * 6127 * When the parent SV is freed, the extra ref is killed by 6128 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic, 6129 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs. 6130 * 6131 * When a single backref SV is stored directly, it is not reference 6132 * counted. 6133 */ 6134 6135void 6136Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv) 6137{ 6138 SV **svp; 6139 AV *av = NULL; 6140 MAGIC *mg = NULL; 6141 6142 PERL_ARGS_ASSERT_SV_ADD_BACKREF; 6143 6144 /* find slot to store array or singleton backref */ 6145 6146 if (SvTYPE(tsv) == SVt_PVHV) { 6147 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv)); 6148 } else { 6149 if (SvMAGICAL(tsv)) 6150 mg = mg_find(tsv, PERL_MAGIC_backref); 6151 if (!mg) 6152 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0); 6153 svp = &(mg->mg_obj); 6154 } 6155 6156 /* create or retrieve the array */ 6157 6158 if ( (!*svp && SvTYPE(sv) == SVt_PVAV) 6159 || (*svp && SvTYPE(*svp) != SVt_PVAV) 6160 ) { 6161 /* create array */ 6162 if (mg) 6163 mg->mg_flags |= MGf_REFCOUNTED; 6164 av = newAV(); 6165 AvREAL_off(av); 6166 SvREFCNT_inc_simple_void_NN(av); 6167 /* av now has a refcnt of 2; see discussion above */ 6168 av_extend(av, *svp ? 2 : 1); 6169 if (*svp) { 6170 /* move single existing backref to the array */ 6171 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */ 6172 } 6173 *svp = (SV*)av; 6174 } 6175 else { 6176 av = MUTABLE_AV(*svp); 6177 if (!av) { 6178 /* optimisation: store single backref directly in HvAUX or mg_obj */ 6179 *svp = sv; 6180 return; 6181 } 6182 assert(SvTYPE(av) == SVt_PVAV); 6183 if (AvFILLp(av) >= AvMAX(av)) { 6184 av_extend(av, AvFILLp(av)+1); 6185 } 6186 } 6187 /* push new backref */ 6188 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */ 6189} 6190 6191/* delete a back-reference to ourselves from the backref magic associated 6192 * with the SV we point to. 6193 */ 6194 6195void 6196Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv) 6197{ 6198 SV **svp = NULL; 6199 6200 PERL_ARGS_ASSERT_SV_DEL_BACKREF; 6201 6202 if (SvTYPE(tsv) == SVt_PVHV) { 6203 if (HvHasAUX(tsv)) 6204 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv)); 6205 } 6206 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) { 6207 /* It's possible for the last (strong) reference to tsv to have 6208 become freed *before* the last thing holding a weak reference. 6209 If both survive longer than the backreferences array, then when 6210 the referent's reference count drops to 0 and it is freed, it's 6211 not able to chase the backreferences, so they aren't NULLed. 6212 6213 For example, a CV holds a weak reference to its stash. If both the 6214 CV and the stash survive longer than the backreferences array, 6215 and the CV gets picked for the SvBREAK() treatment first, 6216 *and* it turns out that the stash is only being kept alive because 6217 of an our variable in the pad of the CV, then midway during CV 6218 destruction the stash gets freed, but CvSTASH() isn't set to NULL. 6219 It ends up pointing to the freed HV. Hence it's chased in here, and 6220 if this block wasn't here, it would hit the !svp panic just below. 6221 6222 I don't believe that "better" destruction ordering is going to help 6223 here - during global destruction there's always going to be the 6224 chance that something goes out of order. We've tried to make it 6225 foolproof before, and it only resulted in evolutionary pressure on 6226 fools. Which made us look foolish for our hubris. :-( 6227 */ 6228 return; 6229 } 6230 else { 6231 MAGIC *const mg 6232 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL; 6233 svp = mg ? &(mg->mg_obj) : NULL; 6234 } 6235 6236 if (!svp) 6237 Perl_croak(aTHX_ "panic: del_backref, svp=0"); 6238 if (!*svp) { 6239 /* It's possible that sv is being freed recursively part way through the 6240 freeing of tsv. If this happens, the backreferences array of tsv has 6241 already been freed, and so svp will be NULL. If this is the case, 6242 we should not panic. Instead, nothing needs doing, so return. */ 6243 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0) 6244 return; 6245 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf, 6246 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv)); 6247 } 6248 6249 if (SvTYPE(*svp) == SVt_PVAV) { 6250#ifdef DEBUGGING 6251 int count = 1; 6252#endif 6253 AV * const av = (AV*)*svp; 6254 SSize_t fill; 6255 assert(!SvIS_FREED(av)); 6256 fill = AvFILLp(av); 6257 assert(fill > -1); 6258 svp = AvARRAY(av); 6259 /* for an SV with N weak references to it, if all those 6260 * weak refs are deleted, then sv_del_backref will be called 6261 * N times and O(N^2) compares will be done within the backref 6262 * array. To ameliorate this potential slowness, we: 6263 * 1) make sure this code is as tight as possible; 6264 * 2) when looking for SV, look for it at both the head and tail of the 6265 * array first before searching the rest, since some create/destroy 6266 * patterns will cause the backrefs to be freed in order. 6267 */ 6268 if (*svp == sv) { 6269 AvARRAY(av)++; 6270 AvMAX(av)--; 6271 } 6272 else { 6273 SV **p = &svp[fill]; 6274 SV *const topsv = *p; 6275 if (topsv != sv) { 6276#ifdef DEBUGGING 6277 count = 0; 6278#endif 6279 while (--p > svp) { 6280 if (*p == sv) { 6281 /* We weren't the last entry. 6282 An unordered list has this property that you 6283 can take the last element off the end to fill 6284 the hole, and it's still an unordered list :-) 6285 */ 6286 *p = topsv; 6287#ifdef DEBUGGING 6288 count++; 6289#else 6290 break; /* should only be one */ 6291#endif 6292 } 6293 } 6294 } 6295 } 6296 assert(count ==1); 6297 AvFILLp(av) = fill-1; 6298 } 6299 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) { 6300 /* freed AV; skip */ 6301 } 6302 else { 6303 /* optimisation: only a single backref, stored directly */ 6304 if (*svp != sv) 6305 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p", 6306 (void*)*svp, (void*)sv); 6307 *svp = NULL; 6308 } 6309 6310} 6311 6312void 6313Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av) 6314{ 6315 SV **svp; 6316 SV **last; 6317 bool is_array; 6318 6319 PERL_ARGS_ASSERT_SV_KILL_BACKREFS; 6320 6321 if (!av) 6322 return; 6323 6324 /* after multiple passes through Perl_sv_clean_all() for a thingy 6325 * that has badly leaked, the backref array may have gotten freed, 6326 * since we only protect it against 1 round of cleanup */ 6327 if (SvIS_FREED(av)) { 6328 if (PL_in_clean_all) /* All is fair */ 6329 return; 6330 Perl_croak(aTHX_ 6331 "panic: magic_killbackrefs (freed backref AV/SV)"); 6332 } 6333 6334 6335 is_array = (SvTYPE(av) == SVt_PVAV); 6336 if (is_array) { 6337 assert(!SvIS_FREED(av)); 6338 svp = AvARRAY(av); 6339 if (svp) 6340 last = svp + AvFILLp(av); 6341 } 6342 else { 6343 /* optimisation: only a single backref, stored directly */ 6344 svp = (SV**)&av; 6345 last = svp; 6346 } 6347 6348 if (svp) { 6349 while (svp <= last) { 6350 if (*svp) { 6351 SV *const referrer = *svp; 6352 if (SvWEAKREF(referrer)) { 6353 /* XXX Should we check that it hasn't changed? */ 6354 assert(SvROK(referrer)); 6355 SvRV_set(referrer, 0); 6356 SvOK_off(referrer); 6357 SvWEAKREF_off(referrer); 6358 SvSETMAGIC(referrer); 6359 } else if (SvTYPE(referrer) == SVt_PVGV || 6360 SvTYPE(referrer) == SVt_PVLV) { 6361 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */ 6362 /* You lookin' at me? */ 6363 assert(GvSTASH(referrer)); 6364 assert(GvSTASH(referrer) == (const HV *)sv); 6365 GvSTASH(referrer) = 0; 6366 } else if (SvTYPE(referrer) == SVt_PVCV || 6367 SvTYPE(referrer) == SVt_PVFM) { 6368 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */ 6369 /* You lookin' at me? */ 6370 assert(CvSTASH(referrer)); 6371 assert(CvSTASH(referrer) == (const HV *)sv); 6372 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0; 6373 } 6374 else { 6375 assert(SvTYPE(sv) == SVt_PVGV); 6376 /* You lookin' at me? */ 6377 assert(CvGV(referrer)); 6378 assert(CvGV(referrer) == (const GV *)sv); 6379 anonymise_cv_maybe(MUTABLE_GV(sv), 6380 MUTABLE_CV(referrer)); 6381 } 6382 6383 } else { 6384 Perl_croak(aTHX_ 6385 "panic: magic_killbackrefs (flags=%" UVxf ")", 6386 (UV)SvFLAGS(referrer)); 6387 } 6388 6389 if (is_array) 6390 *svp = NULL; 6391 } 6392 svp++; 6393 } 6394 } 6395 if (is_array) { 6396 AvFILLp(av) = -1; 6397 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */ 6398 } 6399 return; 6400} 6401 6402/* 6403=for apidoc sv_insert 6404 6405Inserts and/or replaces a string at the specified offset/length within the SV. 6406Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at 6407C<little> replacing C<len> bytes of the string in C<bigstr> starting at 6408C<offset>. Handles get magic. 6409 6410=for apidoc sv_insert_flags 6411 6412Same as C<sv_insert>, but the extra C<flags> are passed to the 6413C<SvPV_force_flags> that applies to C<bigstr>. 6414 6415=cut 6416*/ 6417 6418void 6419Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags) 6420{ 6421 char *big; 6422 char *mid; 6423 char *midend; 6424 char *bigend; 6425 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */ 6426 STRLEN curlen; 6427 6428 PERL_ARGS_ASSERT_SV_INSERT_FLAGS; 6429 6430 SvPV_force_flags(bigstr, curlen, flags); 6431 (void)SvPOK_only_UTF8(bigstr); 6432 6433 if (little >= SvPVX(bigstr) && 6434 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) { 6435 /* little is a pointer to within bigstr, since we can reallocate bigstr, 6436 or little...little+littlelen might overlap offset...offset+len we make a copy 6437 */ 6438 little = savepvn(little, littlelen); 6439 SAVEFREEPV(little); 6440 } 6441 6442 if (offset + len > curlen) { 6443 SvGROW(bigstr, offset+len+1); 6444 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char); 6445 SvCUR_set(bigstr, offset+len); 6446 } 6447 6448 SvTAINT(bigstr); 6449 i = littlelen - len; 6450 if (i > 0) { /* string might grow */ 6451 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1); 6452 mid = big + offset + len; 6453 midend = bigend = big + SvCUR(bigstr); 6454 bigend += i; 6455 *bigend = '\0'; 6456 while (midend > mid) /* shove everything down */ 6457 *--bigend = *--midend; 6458 Move(little,big+offset,littlelen,char); 6459 SvCUR_set(bigstr, SvCUR(bigstr) + i); 6460 SvSETMAGIC(bigstr); 6461 return; 6462 } 6463 else if (i == 0) { 6464 Move(little,SvPVX(bigstr)+offset,len,char); 6465 SvSETMAGIC(bigstr); 6466 return; 6467 } 6468 6469 big = SvPVX(bigstr); 6470 mid = big + offset; 6471 midend = mid + len; 6472 bigend = big + SvCUR(bigstr); 6473 6474 if (midend > bigend) 6475 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p", 6476 midend, bigend); 6477 6478 if (mid - big > bigend - midend) { /* faster to shorten from end */ 6479 if (littlelen) { 6480 Move(little, mid, littlelen,char); 6481 mid += littlelen; 6482 } 6483 i = bigend - midend; 6484 if (i > 0) { 6485 Move(midend, mid, i,char); 6486 mid += i; 6487 } 6488 *mid = '\0'; 6489 SvCUR_set(bigstr, mid - big); 6490 } 6491 else if ((i = mid - big)) { /* faster from front */ 6492 midend -= littlelen; 6493 mid = midend; 6494 Move(big, midend - i, i, char); 6495 sv_chop(bigstr,midend-i); 6496 if (littlelen) 6497 Move(little, mid, littlelen,char); 6498 } 6499 else if (littlelen) { 6500 midend -= littlelen; 6501 sv_chop(bigstr,midend); 6502 Move(little,midend,littlelen,char); 6503 } 6504 else { 6505 sv_chop(bigstr,midend); 6506 } 6507 SvSETMAGIC(bigstr); 6508} 6509 6510/* 6511=for apidoc sv_replace 6512 6513Make the first argument a copy of the second, then delete the original. 6514The target SV physically takes over ownership of the body of the source SV 6515and inherits its flags; however, the target keeps any magic it owns, 6516and any magic in the source is discarded. 6517Note that this is a rather specialist SV copying operation; most of the 6518time you'll want to use C<sv_setsv> or one of its many macro front-ends. 6519 6520=cut 6521*/ 6522 6523void 6524Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv) 6525{ 6526 const U32 refcnt = SvREFCNT(sv); 6527 6528 PERL_ARGS_ASSERT_SV_REPLACE; 6529 6530 SV_CHECK_THINKFIRST_COW_DROP(sv); 6531 if (SvREFCNT(nsv) != 1) { 6532 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()" 6533 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv)); 6534 } 6535 if (SvMAGICAL(sv)) { 6536 if (SvMAGICAL(nsv)) 6537 mg_free(nsv); 6538 else 6539 sv_upgrade(nsv, SVt_PVMG); 6540 SvMAGIC_set(nsv, SvMAGIC(sv)); 6541 SvFLAGS(nsv) |= SvMAGICAL(sv); 6542 SvMAGICAL_off(sv); 6543 SvMAGIC_set(sv, NULL); 6544 } 6545 SvREFCNT(sv) = 0; 6546 sv_clear(sv); 6547 assert(!SvREFCNT(sv)); 6548#ifdef DEBUG_LEAKING_SCALARS 6549 sv->sv_flags = nsv->sv_flags; 6550 sv->sv_any = nsv->sv_any; 6551 sv->sv_refcnt = nsv->sv_refcnt; 6552 sv->sv_u = nsv->sv_u; 6553#else 6554 StructCopy(nsv,sv,SV); 6555#endif 6556 if(SvTYPE(sv) == SVt_IV) { 6557 SET_SVANY_FOR_BODYLESS_IV(sv); 6558 } 6559 6560 6561 SvREFCNT(sv) = refcnt; 6562 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */ 6563 SvREFCNT(nsv) = 0; 6564 del_SV(nsv); 6565} 6566 6567/* We're about to free a GV which has a CV that refers back to us. 6568 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV 6569 * field) */ 6570 6571STATIC void 6572S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv) 6573{ 6574 SV *gvname; 6575 GV *anongv; 6576 6577 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE; 6578 6579 /* be assertive! */ 6580 assert(SvREFCNT(gv) == 0); 6581 assert(isGV(gv) && isGV_with_GP(gv)); 6582 assert(GvGP(gv)); 6583 assert(!CvANON(cv)); 6584 assert(CvGV(cv) == gv); 6585 assert(!CvNAMED(cv)); 6586 6587 /* will the CV shortly be freed by gp_free() ? */ 6588 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) { 6589 SvANY(cv)->xcv_gv_u.xcv_gv = NULL; 6590 return; 6591 } 6592 6593 /* if not, anonymise: */ 6594 gvname = (GvSTASH(gv) && HvHasNAME(GvSTASH(gv)) && HvHasENAME(GvSTASH(gv))) 6595 ? newSVhek(HvENAME_HEK(GvSTASH(gv))) 6596 : newSVpvn_flags( "__ANON__", 8, 0 ); 6597 sv_catpvs(gvname, "::__ANON__"); 6598 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV); 6599 SvREFCNT_dec_NN(gvname); 6600 6601 CvANON_on(cv); 6602 CvCVGV_RC_on(cv); 6603 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv)); 6604} 6605 6606 6607/* 6608=for apidoc sv_clear 6609 6610Clear an SV: call any destructors, free up any memory used by the body, 6611and free the body itself. The SV's head is I<not> freed, although 6612its type is set to all 1's so that it won't inadvertently be assumed 6613to be live during global destruction etc. 6614This function should only be called when C<REFCNT> is zero. Most of the time 6615you'll want to call C<SvREFCNT_dec> instead. 6616 6617=cut 6618*/ 6619 6620void 6621Perl_sv_clear(pTHX_ SV *const orig_sv) 6622{ 6623 SV* iter_sv = NULL; 6624 SV* next_sv = NULL; 6625 SV *sv = orig_sv; 6626 STRLEN hash_index = 0; /* initialise to make Coverity et al happy. 6627 Not strictly necessary */ 6628 6629 PERL_ARGS_ASSERT_SV_CLEAR; 6630 6631 /* within this loop, sv is the SV currently being freed, and 6632 * iter_sv is the most recent AV or whatever that's being iterated 6633 * over to provide more SVs */ 6634 6635 while (sv) { 6636 U32 type = SvTYPE(sv); 6637 HV *stash; 6638 6639 assert(SvREFCNT(sv) == 0); 6640 assert(!SvIS_FREED(sv)); 6641#if NVSIZE <= IVSIZE 6642 if (type <= SVt_NV) { 6643#else 6644 if (type <= SVt_IV) { 6645#endif 6646 /* Historically this check on type was needed so that the code to 6647 * free bodies wasn't reached for these types, because the arena 6648 * slots were re-used for HEs and pointer table entries. The 6649 * metadata table `bodies_by_type` had the information for the sizes 6650 * for HEs and PTEs, hence the code here had to have a special-case 6651 * check to ensure that the "regular" body freeing code wasn't 6652 * reached, and get confused by the "lies" in `bodies_by_type`. 6653 * 6654 * However, it hasn't actually been needed for that reason since 6655 * Aug 2010 (commit 829cd18aa7f45221), because `bodies_by_type` was 6656 * changed to always hold the accurate metadata for the SV types. 6657 * This was possible because PTEs were no longer allocated from the 6658 * "SVt_IV" arena, and the code to allocate HEs from the "SVt_NULL" 6659 * arena is entirely in hv.c, so doesn't access the table. 6660 * 6661 * Some sort of check is still needed to handle SVt_IVs - pure RVs 6662 * need to take one code path which is common with RVs stored in 6663 * SVt_PV (or larger), but pure IVs mustn't take the "PV but not RV" 6664 * path, as SvPVX() doesn't point to valid memory. 6665 * 6666 * Hence this code is still the most efficient way to handle this. 6667 * 6668 * Additionally, for bodyless NVs, riding this branch is more 6669 * efficient than stepping through the general logic. 6670 */ 6671 6672 if (SvROK(sv)) 6673 goto free_rv; 6674 SvFLAGS(sv) &= SVf_BREAK; 6675 SvFLAGS(sv) |= SVTYPEMASK; 6676 goto free_head; 6677 } 6678 6679 /* objs are always >= MG, but pad names use the SVs_OBJECT flag 6680 for another purpose */ 6681 assert(!SvOBJECT(sv) || type >= SVt_PVMG); 6682 6683 if (type >= SVt_PVMG) { 6684 if (SvOBJECT(sv)) { 6685 if (!curse(sv, 1)) goto get_next_sv; 6686 type = SvTYPE(sv); /* destructor may have changed it */ 6687 } 6688 /* Free back-references before magic, in case the magic calls 6689 * Perl code that has weak references to sv. */ 6690 if (type == SVt_PVHV) { 6691 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv)); 6692 if (SvMAGIC(sv)) 6693 mg_free(sv); 6694 } 6695 else if (SvMAGIC(sv)) { 6696 /* Free back-references before other types of magic. */ 6697 sv_unmagic(sv, PERL_MAGIC_backref); 6698 mg_free(sv); 6699 } 6700 SvMAGICAL_off(sv); 6701 } 6702 switch (type) { 6703 /* case SVt_INVLIST: */ 6704 case SVt_PVIO: 6705 if (IoIFP(sv) && 6706 IoIFP(sv) != PerlIO_stdin() && 6707 IoIFP(sv) != PerlIO_stdout() && 6708 IoIFP(sv) != PerlIO_stderr() && 6709 !(IoFLAGS(sv) & IOf_FAKE_DIRP)) 6710 { 6711 io_close(MUTABLE_IO(sv), NULL, FALSE, 6712 (IoTYPE(sv) == IoTYPE_WRONLY || 6713 IoTYPE(sv) == IoTYPE_RDWR || 6714 IoTYPE(sv) == IoTYPE_APPEND)); 6715 } 6716 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP)) 6717 PerlDir_close(IoDIRP(sv)); 6718 IoDIRP(sv) = (DIR*)NULL; 6719 Safefree(IoTOP_NAME(sv)); 6720 Safefree(IoFMT_NAME(sv)); 6721 Safefree(IoBOTTOM_NAME(sv)); 6722 if ((const GV *)sv == PL_statgv) 6723 PL_statgv = NULL; 6724 goto freescalar; 6725 case SVt_REGEXP: 6726 /* FIXME for plugins */ 6727 pregfree2((REGEXP*) sv); 6728 goto freescalar; 6729 case SVt_PVCV: 6730 case SVt_PVFM: 6731 cv_undef(MUTABLE_CV(sv)); 6732 /* If we're in a stash, we don't own a reference to it. 6733 * However it does have a back reference to us, which needs to 6734 * be cleared. */ 6735 if ((stash = CvSTASH(sv))) 6736 sv_del_backref(MUTABLE_SV(stash), sv); 6737 goto freescalar; 6738 case SVt_PVHV: 6739 if (HvTOTALKEYS((HV*)sv) > 0) { 6740 const HEK *hek; 6741 /* this statement should match the one at the beginning of 6742 * hv_undef_flags() */ 6743 if ( PL_phase != PERL_PHASE_DESTRUCT 6744 && (hek = HvNAME_HEK((HV*)sv))) 6745 { 6746 if (PL_stashcache) { 6747 DEBUG_o(Perl_deb(aTHX_ 6748 "sv_clear clearing PL_stashcache for '%" HEKf 6749 "'\n", 6750 HEKfARG(hek))); 6751 (void)hv_deletehek(PL_stashcache, 6752 hek, G_DISCARD); 6753 } 6754 hv_name_set((HV*)sv, NULL, 0, 0); 6755 } 6756 6757 /* save old iter_sv in unused SvSTASH field */ 6758 assert(!SvOBJECT(sv)); 6759 SvSTASH(sv) = (HV*)iter_sv; 6760 iter_sv = sv; 6761 6762 /* save old hash_index in unused SvMAGIC field */ 6763 assert(!SvMAGICAL(sv)); 6764 assert(!SvMAGIC(sv)); 6765 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index; 6766 hash_index = 0; 6767 6768 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index); 6769 goto get_next_sv; /* process this new sv */ 6770 } 6771 /* free empty hash */ 6772 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL); 6773 assert(!HvARRAY((HV*)sv)); 6774 break; 6775 case SVt_PVAV: 6776 { 6777 AV* av = MUTABLE_AV(sv); 6778 if (PL_comppad == av) { 6779 PL_comppad = NULL; 6780 PL_curpad = NULL; 6781 } 6782 if (AvREAL(av) && AvFILLp(av) > -1) { 6783 next_sv = AvARRAY(av)[AvFILLp(av)--]; 6784 /* save old iter_sv in top-most slot of AV, 6785 * and pray that it doesn't get wiped in the meantime */ 6786 AvARRAY(av)[AvMAX(av)] = iter_sv; 6787 iter_sv = sv; 6788 goto get_next_sv; /* process this new sv */ 6789 } 6790 Safefree(AvALLOC(av)); 6791 } 6792 6793 break; 6794 case SVt_PVOBJ: 6795 if(ObjectMAXFIELD(sv) > -1) { 6796 next_sv = ObjectFIELDS(sv)[ObjectMAXFIELD(sv)--]; 6797 /* save old iter_sv in top-most field, and pray that it 6798 * doesn't get wiped in the meantime */ 6799 ObjectFIELDS(sv)[(ObjectITERSVAT(sv) = ObjectMAXFIELD(sv) + 1)] = iter_sv; 6800 iter_sv = sv; 6801 goto get_next_sv; 6802 } 6803 Safefree(ObjectFIELDS(sv)); 6804 break; 6805 case SVt_PVLV: 6806 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */ 6807 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv))); 6808 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh; 6809 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv); 6810 } 6811 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */ 6812 SvREFCNT_dec(LvTARG(sv)); 6813 if (isREGEXP(sv)) { 6814 /* This PVLV has had a REGEXP assigned to it - the memory 6815 * normally used to store SvLEN instead points to a regex body. 6816 * Retrieving the pointer to the regex body from the correct 6817 * location is normally abstracted by ReANY(), which handles 6818 * both SVt_PVLV and SVt_REGEXP 6819 * 6820 * This code is unwinding the storage specific to SVt_PVLV. 6821 * We get the body pointer directly from the union, free it, 6822 * then set SvLEN to whatever value was in the now-freed regex 6823 * body. The PVX buffer is shared by multiple re's and only 6824 * freed once, by the re whose SvLEN is non-null. 6825 * 6826 * Perl_sv_force_normal_flags() also has code to free this 6827 * hidden body - it swaps the body into a temporary SV it has 6828 * just allocated, then frees that SV. That causes execution 6829 * to reach the SVt_REGEXP: case about 60 lines earlier in this 6830 * function. 6831 * 6832 * See Perl_reg_temp_copy() for the code that sets up this 6833 * REGEXP body referenced by the PVLV. */ 6834 struct regexp *r = ((XPV*)SvANY(sv))->xpv_len_u.xpvlenu_rx; 6835 STRLEN len = r->xpv_len; 6836 pregfree2((REGEXP*) sv); 6837 del_body_by_type(r, SVt_REGEXP); 6838 SvLEN_set((sv), len); 6839 goto freescalar; 6840 } 6841 /* FALLTHROUGH */ 6842 case SVt_PVGV: 6843 if (isGV_with_GP(sv)) { 6844 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv))) 6845 && HvHasENAME(stash)) 6846 mro_method_changed_in(stash); 6847 gp_free(MUTABLE_GV(sv)); 6848 if (GvNAME_HEK(sv)) 6849 unshare_hek(GvNAME_HEK(sv)); 6850 /* If we're in a stash, we don't own a reference to it. 6851 * However it does have a back reference to us, which 6852 * needs to be cleared. */ 6853 if ((stash = GvSTASH(sv))) 6854 sv_del_backref(MUTABLE_SV(stash), sv); 6855 } 6856 /* FIXME. There are probably more unreferenced pointers to SVs 6857 * in the interpreter struct that we should check and tidy in 6858 * a similar fashion to this: */ 6859 /* See also S_sv_unglob, which does the same thing. */ 6860 if ((const GV *)sv == PL_last_in_gv) 6861 PL_last_in_gv = NULL; 6862 else if ((const GV *)sv == PL_statgv) 6863 PL_statgv = NULL; 6864 else if ((const GV *)sv == PL_stderrgv) 6865 PL_stderrgv = NULL; 6866 /* FALLTHROUGH */ 6867 case SVt_PVMG: 6868 case SVt_PVNV: 6869 case SVt_PVIV: 6870 case SVt_INVLIST: 6871 case SVt_PV: 6872 freescalar: 6873 /* Don't bother with SvOOK_off(sv); as we're only going to 6874 * free it. */ 6875 if (SvOOK(sv)) { 6876 STRLEN offset; 6877 SvOOK_offset(sv, offset); 6878 SvPV_set(sv, SvPVX_mutable(sv) - offset); 6879 /* Don't even bother with turning off the OOK flag. */ 6880 } 6881 if (SvROK(sv)) { 6882 free_rv: 6883 { 6884 SV * const target = SvRV(sv); 6885 if (SvWEAKREF(sv)) 6886 sv_del_backref(target, sv); 6887 else 6888 next_sv = target; 6889 } 6890 } 6891#ifdef PERL_ANY_COW 6892 else if (SvPVX_const(sv) 6893 && !(SvTYPE(sv) == SVt_PVIO 6894 && !(IoFLAGS(sv) & IOf_FAKE_DIRP))) 6895 { 6896 if (SvIsCOW(sv)) { 6897#ifdef DEBUGGING 6898 if (DEBUG_C_TEST) { 6899 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n"); 6900 sv_dump(sv); 6901 } 6902#endif 6903 if (SvIsCOW_static(sv)) { 6904 SvLEN_set(sv, 0); 6905 } 6906 else if (SvIsCOW_shared_hash(sv)) { 6907 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv))); 6908 } 6909 else { 6910 if (CowREFCNT(sv)) { 6911 sv_buf_to_rw(sv); 6912 CowREFCNT(sv)--; 6913 sv_buf_to_ro(sv); 6914 SvLEN_set(sv, 0); 6915 } 6916 } 6917 } 6918 if (SvLEN(sv)) { 6919 Safefree(SvPVX_mutable(sv)); 6920 } 6921 } 6922#else 6923 else if (SvPVX_const(sv) && SvLEN(sv) 6924 && !(SvTYPE(sv) == SVt_PVIO 6925 && !(IoFLAGS(sv) & IOf_FAKE_DIRP))) 6926 Safefree(SvPVX_mutable(sv)); 6927 else if (SvPVX_const(sv) && SvIsCOW(sv)) { 6928 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv))); 6929 } 6930#endif 6931 break; 6932 case SVt_NV: 6933 break; 6934 } 6935 6936 free_body: 6937 6938 { 6939 U32 arena_index; 6940 const struct body_details *sv_type_details; 6941 6942 if (type == SVt_PVHV && HvHasAUX(sv)) { 6943 arena_index = HVAUX_ARENA_ROOT_IX; 6944 sv_type_details = &fake_hv_with_aux; 6945 } 6946 else { 6947 arena_index = type; 6948 sv_type_details = bodies_by_type + arena_index; 6949 } 6950 6951 SvFLAGS(sv) &= SVf_BREAK; 6952 SvFLAGS(sv) |= SVTYPEMASK; 6953 6954 if (sv_type_details->arena) { 6955 del_body(((char *)SvANY(sv) + sv_type_details->offset), 6956 &PL_body_roots[arena_index]); 6957 } 6958 else if (sv_type_details->body_size) { 6959 safefree(SvANY(sv)); 6960 } 6961 } 6962 6963 free_head: 6964 /* caller is responsible for freeing the head of the original sv */ 6965 if (sv != orig_sv && !SvREFCNT(sv)) 6966 del_SV(sv); 6967 6968 /* grab and free next sv, if any */ 6969 get_next_sv: 6970 while (1) { 6971 sv = NULL; 6972 if (next_sv) { 6973 sv = next_sv; 6974 next_sv = NULL; 6975 } 6976 else if (!iter_sv) { 6977 break; 6978 } else if (SvTYPE(iter_sv) == SVt_PVAV) { 6979 AV *const av = (AV*)iter_sv; 6980 if (AvFILLp(av) > -1) { 6981 sv = AvARRAY(av)[AvFILLp(av)--]; 6982 } 6983 else { /* no more elements of current AV to free */ 6984 sv = iter_sv; 6985 type = SvTYPE(sv); 6986 /* restore previous value, squirrelled away */ 6987 iter_sv = AvARRAY(av)[AvMAX(av)]; 6988 Safefree(AvALLOC(av)); 6989 goto free_body; 6990 } 6991 } else if (SvTYPE(iter_sv) == SVt_PVOBJ) { 6992 if (ObjectMAXFIELD(iter_sv) > -1) { 6993 sv = ObjectFIELDS(iter_sv)[ObjectMAXFIELD(iter_sv)--]; 6994 } 6995 else { /* no more fields in the current SV to free */ 6996 sv = iter_sv; 6997 type = SvTYPE(sv); 6998 iter_sv = ObjectFIELDS(sv)[ObjectITERSVAT(sv)]; 6999 Safefree(ObjectFIELDS(sv)); 7000 goto free_body; 7001 } 7002 } else if (SvTYPE(iter_sv) == SVt_PVHV) { 7003 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index); 7004 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) { 7005 /* no more elements of current HV to free */ 7006 sv = iter_sv; 7007 type = SvTYPE(sv); 7008 /* Restore previous values of iter_sv and hash_index, 7009 * squirrelled away */ 7010 assert(!SvOBJECT(sv)); 7011 iter_sv = (SV*)SvSTASH(sv); 7012 assert(!SvMAGICAL(sv)); 7013 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index; 7014#ifdef DEBUGGING 7015 /* perl -DA does not like rubbish in SvMAGIC. */ 7016 SvMAGIC_set(sv, 0); 7017#endif 7018 7019 /* free any remaining detritus from the hash struct */ 7020 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL); 7021 assert(!HvARRAY((HV*)sv)); 7022 goto free_body; 7023 } 7024 } 7025 7026 /* unrolled SvREFCNT_dec and sv_free2 follows: */ 7027 7028 if (!sv) 7029 continue; 7030 if (!SvREFCNT(sv)) { 7031 sv_free(sv); 7032 continue; 7033 } 7034 if (--(SvREFCNT(sv))) 7035 continue; 7036#ifdef DEBUGGING 7037 if (SvTEMP(sv)) { 7038 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING), 7039 "Attempt to free temp prematurely: SV 0x%" UVxf 7040 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); 7041 continue; 7042 } 7043#endif 7044 if (SvIMMORTAL(sv)) { 7045 /* make sure SvREFCNT(sv)==0 happens very seldom */ 7046 SvREFCNT(sv) = SvREFCNT_IMMORTAL; 7047 continue; 7048 } 7049 break; 7050 } /* while 1 */ 7051 7052 } /* while sv */ 7053} 7054 7055/* This routine curses the sv itself, not the object referenced by sv. So 7056 sv does not have to be ROK. */ 7057 7058static bool 7059S_curse(pTHX_ SV * const sv, const bool check_refcnt) { 7060 PERL_ARGS_ASSERT_CURSE; 7061 assert(SvOBJECT(sv)); 7062 7063 if (PL_defstash && /* Still have a symbol table? */ 7064 SvDESTROYABLE(sv)) 7065 { 7066 dSP; 7067 HV* stash; 7068 do { 7069 stash = SvSTASH(sv); 7070 assert(SvTYPE(stash) == SVt_PVHV); 7071 if (HvNAME(stash)) { 7072 CV* destructor = NULL; 7073 struct mro_meta *meta; 7074 7075 assert (HvHasAUX(stash)); 7076 7077 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n", 7078 HvNAME(stash)) ); 7079 7080 /* don't make this an initialization above the assert, since it needs 7081 an AUX structure */ 7082 meta = HvMROMETA(stash); 7083 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) { 7084 destructor = meta->destroy; 7085 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n", 7086 (void *)destructor, HvNAME(stash)) ); 7087 } 7088 else { 7089 bool autoload = FALSE; 7090 GV *gv = 7091 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0); 7092 if (gv) 7093 destructor = GvCV(gv); 7094 if (!destructor) { 7095 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len, 7096 GV_AUTOLOAD_ISMETHOD); 7097 if (gv) 7098 destructor = GvCV(gv); 7099 if (destructor) 7100 autoload = TRUE; 7101 } 7102 /* we don't cache AUTOLOAD for DESTROY, since this code 7103 would then need to set $__PACKAGE__::AUTOLOAD, or the 7104 equivalent for XS AUTOLOADs */ 7105 if (!autoload) { 7106 meta->destroy_gen = PL_sub_generation; 7107 meta->destroy = destructor; 7108 7109 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n", 7110 (void *)destructor, HvNAME(stash)) ); 7111 } 7112 else { 7113 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n", 7114 HvNAME(stash)) ); 7115 } 7116 } 7117 assert(!destructor || SvTYPE(destructor) == SVt_PVCV); 7118 if (destructor 7119 /* A constant subroutine can have no side effects, so 7120 don't bother calling it. */ 7121 && !CvCONST(destructor) 7122 /* Don't bother calling an empty destructor or one that 7123 returns immediately. */ 7124 && (CvISXSUB(destructor) 7125 || (CvSTART(destructor) 7126 && (CvSTART(destructor)->op_next->op_type 7127 != OP_LEAVESUB) 7128 && (CvSTART(destructor)->op_next->op_type 7129 != OP_PUSHMARK 7130 || CvSTART(destructor)->op_next->op_next->op_type 7131 != OP_RETURN 7132 ) 7133 )) 7134 ) 7135 { 7136 SV* const tmpref = newRV(sv); 7137 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */ 7138 ENTER; 7139 PUSHSTACKi(PERLSI_DESTROY); 7140 EXTEND(SP, 2); 7141 PUSHMARK(SP); 7142 PUSHs(tmpref); 7143 PUTBACK; 7144 call_sv(MUTABLE_SV(destructor), 7145 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID); 7146 POPSTACK; 7147 SPAGAIN; 7148 LEAVE; 7149 if(SvREFCNT(tmpref) < 2) { 7150 /* tmpref is not kept alive! */ 7151 SvREFCNT(sv)--; 7152 SvRV_set(tmpref, NULL); 7153 SvROK_off(tmpref); 7154 } 7155 SvREFCNT_dec_NN(tmpref); 7156 } 7157 } 7158 } while (SvOBJECT(sv) && SvSTASH(sv) != stash); 7159 7160 7161 if (check_refcnt && SvREFCNT(sv)) { 7162 if (PL_in_clean_objs) 7163 Perl_croak(aTHX_ 7164 "DESTROY created new reference to dead object '%" HEKf "'", 7165 HEKfARG(HvNAME_HEK(stash))); 7166 /* DESTROY gave object new lease on life */ 7167 return FALSE; 7168 } 7169 } 7170 7171 if (SvOBJECT(sv)) { 7172 HV * const stash = SvSTASH(sv); 7173 /* Curse before freeing the stash, as freeing the stash could cause 7174 a recursive call into S_curse. */ 7175 SvOBJECT_off(sv); /* Curse the object. */ 7176 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */ 7177 SvREFCNT_dec(stash); /* possibly of changed persuasion */ 7178 } 7179 return TRUE; 7180} 7181 7182/* 7183=for apidoc sv_newref 7184 7185Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper 7186instead. 7187 7188=cut 7189*/ 7190 7191SV * 7192Perl_sv_newref(pTHX_ SV *const sv) 7193{ 7194 PERL_UNUSED_CONTEXT; 7195 if (sv) 7196 (SvREFCNT(sv))++; 7197 return sv; 7198} 7199 7200/* 7201=for apidoc sv_free 7202 7203Decrement an SV's reference count, and if it drops to zero, call 7204C<sv_clear> to invoke destructors and free up any memory used by 7205the body; finally, deallocating the SV's head itself. 7206Normally called via a wrapper macro C<SvREFCNT_dec>. 7207 7208=cut 7209*/ 7210 7211void 7212Perl_sv_free(pTHX_ SV *const sv) 7213{ 7214 SvREFCNT_dec(sv); 7215} 7216 7217 7218/* Private helper function for SvREFCNT_dec(). 7219 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */ 7220 7221void 7222Perl_sv_free2(pTHX_ SV *const sv, const U32 rc) 7223{ 7224 7225 PERL_ARGS_ASSERT_SV_FREE2; 7226 7227 if (LIKELY( rc == 1 )) { 7228 /* normal case */ 7229 SvREFCNT(sv) = 0; 7230 7231#ifdef DEBUGGING 7232 if (SvTEMP(sv)) { 7233 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING), 7234 "Attempt to free temp prematurely: SV 0x%" UVxf 7235 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); 7236 return; 7237 } 7238#endif 7239 if (SvIMMORTAL(sv)) { 7240 /* make sure SvREFCNT(sv)==0 happens very seldom */ 7241 SvREFCNT(sv) = SvREFCNT_IMMORTAL; 7242 return; 7243 } 7244 sv_clear(sv); 7245 if (! SvREFCNT(sv)) /* may have have been resurrected */ 7246 del_SV(sv); 7247 return; 7248 } 7249 7250 /* handle exceptional cases */ 7251 7252 assert(rc == 0); 7253 7254 if (SvFLAGS(sv) & SVf_BREAK) 7255 /* this SV's refcnt has been artificially decremented to 7256 * trigger cleanup */ 7257 return; 7258 if (PL_in_clean_all) /* All is fair */ 7259 return; 7260 if (SvIMMORTAL(sv)) { 7261 /* make sure SvREFCNT(sv)==0 happens very seldom */ 7262 SvREFCNT(sv) = SvREFCNT_IMMORTAL; 7263 return; 7264 } 7265 if (ckWARN_d(WARN_INTERNAL)) { 7266#ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP 7267 Perl_dump_sv_child(aTHX_ sv); 7268#else 7269 #ifdef DEBUG_LEAKING_SCALARS 7270 sv_dump(sv); 7271 #endif 7272#ifdef DEBUG_LEAKING_SCALARS_ABORT 7273 if (PL_warnhook == PERL_WARNHOOK_FATAL 7274 || ckDEAD(packWARN(WARN_INTERNAL))) { 7275 /* Don't let Perl_warner cause us to escape our fate: */ 7276 abort(); 7277 } 7278#endif 7279 /* This may not return: */ 7280 Perl_warner(aTHX_ packWARN(WARN_INTERNAL), 7281 "Attempt to free unreferenced scalar: SV 0x%" UVxf 7282 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); 7283#endif 7284 } 7285#ifdef DEBUG_LEAKING_SCALARS_ABORT 7286 abort(); 7287#endif 7288 7289} 7290 7291 7292/* 7293=for apidoc sv_len 7294 7295Returns the length of the string in the SV. Handles magic and type 7296coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which 7297gives raw access to the C<xpv_cur> slot. 7298 7299=cut 7300*/ 7301 7302STRLEN 7303Perl_sv_len(pTHX_ SV *const sv) 7304{ 7305 STRLEN len; 7306 7307 if (!sv) 7308 return 0; 7309 7310 (void)SvPV_const(sv, len); 7311 return len; 7312} 7313 7314/* 7315=for apidoc sv_len_utf8 7316=for apidoc_item sv_len_utf8_nomg 7317 7318These return the number of characters in the string in an SV, counting wide 7319UTF-8 bytes as a single character. Both handle type coercion. 7320They differ only in that C<sv_len_utf8> performs 'get' magic; 7321C<sv_len_utf8_nomg> skips any magic. 7322 7323=cut 7324*/ 7325 7326/* 7327 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the 7328 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below. 7329 * (Note that the mg_len is not the length of the mg_ptr field. 7330 * This allows the cache to store the character length of the string without 7331 * needing to malloc() extra storage to attach to the mg_ptr.) 7332 * 7333 */ 7334 7335STRLEN 7336Perl_sv_len_utf8(pTHX_ SV *const sv) 7337{ 7338 if (!sv) 7339 return 0; 7340 7341 SvGETMAGIC(sv); 7342 return sv_len_utf8_nomg(sv); 7343} 7344 7345STRLEN 7346Perl_sv_len_utf8_nomg(pTHX_ SV * const sv) 7347{ 7348 STRLEN len; 7349 const U8 *s = (U8*)SvPV_nomg_const(sv, len); 7350 7351 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG; 7352 7353 if (PL_utf8cache && SvUTF8(sv)) { 7354 STRLEN ulen; 7355 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL; 7356 7357 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) { 7358 if (mg->mg_len != -1) 7359 ulen = mg->mg_len; 7360 else { 7361 /* We can use the offset cache for a headstart. 7362 The longer value is stored in the first pair. */ 7363 STRLEN *cache = (STRLEN *) mg->mg_ptr; 7364 7365 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1], 7366 s + len); 7367 } 7368 7369 if (PL_utf8cache < 0) { 7370 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len); 7371 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv); 7372 } 7373 } 7374 else { 7375 ulen = Perl_utf8_length(aTHX_ s, s + len); 7376 utf8_mg_len_cache_update(sv, &mg, ulen); 7377 } 7378 return ulen; 7379 } 7380 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len; 7381} 7382 7383/* Walk forwards to find the byte corresponding to the passed in UTF-8 7384 offset. */ 7385static STRLEN 7386S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send, 7387 STRLEN *const uoffset_p, bool *const at_end, 7388 bool* canonical_position) 7389{ 7390 const U8 *s = start; 7391 STRLEN uoffset = *uoffset_p; 7392 7393 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS; 7394 7395 while (s < send && uoffset) { 7396 --uoffset; 7397 s += UTF8SKIP(s); 7398 } 7399 if (s == send) { 7400 *at_end = TRUE; 7401 } 7402 else if (s > send) { 7403 *at_end = TRUE; 7404 /* This is the existing behaviour. Possibly it should be a croak, as 7405 it's actually a bounds error */ 7406 s = send; 7407 } 7408 /* If the unicode position is beyond the end, we return the end but 7409 shouldn't cache that position */ 7410 *canonical_position = (uoffset == 0); 7411 *uoffset_p -= uoffset; 7412 return s - start; 7413} 7414 7415/* Given the length of the string in both bytes and UTF-8 characters, decide 7416 whether to walk forwards or backwards to find the byte corresponding to 7417 the passed in UTF-8 offset. */ 7418static STRLEN 7419S_sv_pos_u2b_midway(const U8 *const start, const U8 *send, 7420 STRLEN uoffset, const STRLEN uend) 7421{ 7422 STRLEN backw = uend - uoffset; 7423 7424 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY; 7425 7426 if (uoffset < 2 * backw) { 7427 /* The assumption is that going forwards is twice the speed of going 7428 forward (that's where the 2 * backw comes from). 7429 (The real figure of course depends on the UTF-8 data.) */ 7430 const U8 *s = start; 7431 7432 while (s < send && uoffset--) 7433 s += UTF8SKIP(s); 7434 assert (s <= send); 7435 if (s > send) 7436 s = send; 7437 return s - start; 7438 } 7439 7440 while (backw--) { 7441 send--; 7442 while (UTF8_IS_CONTINUATION(*send)) 7443 send--; 7444 } 7445 return send - start; 7446} 7447 7448/* For the string representation of the given scalar, find the byte 7449 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0 7450 give another position in the string, *before* the sought offset, which 7451 (which is always true, as 0, 0 is a valid pair of positions), which should 7452 help reduce the amount of linear searching. 7453 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which 7454 will be used to reduce the amount of linear searching. The cache will be 7455 created if necessary, and the found value offered to it for update. */ 7456static STRLEN 7457S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start, 7458 const U8 *const send, STRLEN uoffset, 7459 STRLEN uoffset0, STRLEN boffset0) 7460{ 7461 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */ 7462 bool found = FALSE; 7463 bool at_end = FALSE; 7464 bool canonical_position = FALSE; 7465 7466 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED; 7467 7468 assert (uoffset >= uoffset0); 7469 7470 if (!uoffset) 7471 return 0; 7472 7473 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv) 7474 && PL_utf8cache 7475 && (*mgp || (SvTYPE(sv) >= SVt_PVMG && 7476 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) { 7477 if ((*mgp)->mg_ptr) { 7478 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr; 7479 if (cache[0] == uoffset) { 7480 /* An exact match. */ 7481 return cache[1]; 7482 } 7483 if (cache[2] == uoffset) { 7484 /* An exact match. */ 7485 return cache[3]; 7486 } 7487 7488 if (cache[0] < uoffset) { 7489 /* The cache already knows part of the way. */ 7490 if (cache[0] > uoffset0) { 7491 /* The cache knows more than the passed in pair */ 7492 uoffset0 = cache[0]; 7493 boffset0 = cache[1]; 7494 } 7495 if ((*mgp)->mg_len != -1) { 7496 /* And we know the end too. */ 7497 boffset = boffset0 7498 + sv_pos_u2b_midway(start + boffset0, send, 7499 uoffset - uoffset0, 7500 (*mgp)->mg_len - uoffset0); 7501 } else { 7502 uoffset -= uoffset0; 7503 boffset = boffset0 7504 + sv_pos_u2b_forwards(start + boffset0, 7505 send, &uoffset, &at_end, 7506 &canonical_position); 7507 uoffset += uoffset0; 7508 } 7509 } 7510 else if (cache[2] < uoffset) { 7511 /* We're between the two cache entries. */ 7512 if (cache[2] > uoffset0) { 7513 /* and the cache knows more than the passed in pair */ 7514 uoffset0 = cache[2]; 7515 boffset0 = cache[3]; 7516 } 7517 7518 boffset = boffset0 7519 + sv_pos_u2b_midway(start + boffset0, 7520 start + cache[1], 7521 uoffset - uoffset0, 7522 cache[0] - uoffset0); 7523 } else { 7524 boffset = boffset0 7525 + sv_pos_u2b_midway(start + boffset0, 7526 start + cache[3], 7527 uoffset - uoffset0, 7528 cache[2] - uoffset0); 7529 } 7530 found = TRUE; 7531 } 7532 else if ((*mgp)->mg_len != -1) { 7533 /* If we can take advantage of a passed in offset, do so. */ 7534 /* In fact, offset0 is either 0, or less than offset, so don't 7535 need to worry about the other possibility. */ 7536 boffset = boffset0 7537 + sv_pos_u2b_midway(start + boffset0, send, 7538 uoffset - uoffset0, 7539 (*mgp)->mg_len - uoffset0); 7540 found = TRUE; 7541 } 7542 } 7543 7544 if (!found || PL_utf8cache < 0) { 7545 STRLEN real_boffset; 7546 uoffset -= uoffset0; 7547 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0, 7548 send, &uoffset, &at_end, 7549 &canonical_position); 7550 uoffset += uoffset0; 7551 7552 if (found && PL_utf8cache < 0) 7553 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset, 7554 real_boffset, sv); 7555 boffset = real_boffset; 7556 } 7557 7558 if (PL_utf8cache && canonical_position && !SvGMAGICAL(sv) && SvPOK(sv)) { 7559 if (at_end) 7560 utf8_mg_len_cache_update(sv, mgp, uoffset); 7561 else 7562 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start); 7563 } 7564 return boffset; 7565} 7566 7567 7568/* 7569=for apidoc sv_pos_u2b_flags 7570 7571Converts the offset from a count of UTF-8 chars from 7572the start of the string, to a count of the equivalent number of bytes; if 7573C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from 7574C<offset>, rather than from the start 7575of the string. Handles type coercion. 7576C<flags> is passed to C<SvPV_flags>, and usually should be 7577C<SV_GMAGIC|SV_CONST_RETURN> to handle magic. 7578 7579=cut 7580*/ 7581 7582/* 7583 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential 7584 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and 7585 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update(). 7586 * 7587 */ 7588 7589STRLEN 7590Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp, 7591 U32 flags) 7592{ 7593 const U8 *start; 7594 STRLEN len; 7595 STRLEN boffset; 7596 7597 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS; 7598 7599 start = (U8*)SvPV_flags(sv, len, flags); 7600 if (len) { 7601 const U8 * const send = start + len; 7602 MAGIC *mg = NULL; 7603 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0); 7604 7605 if (lenp 7606 && *lenp /* don't bother doing work for 0, as its bytes equivalent 7607 is 0, and *lenp is already set to that. */) { 7608 /* Convert the relative offset to absolute. */ 7609 const STRLEN uoffset2 = uoffset + *lenp; 7610 const STRLEN boffset2 7611 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2, 7612 uoffset, boffset) - boffset; 7613 7614 *lenp = boffset2; 7615 } 7616 } else { 7617 if (lenp) 7618 *lenp = 0; 7619 boffset = 0; 7620 } 7621 7622 return boffset; 7623} 7624 7625/* 7626=for apidoc sv_pos_u2b 7627 7628Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from 7629the start of the string, to a count of the equivalent number of bytes; if 7630C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from 7631the offset, rather than from the start of the string. Handles magic and 7632type coercion. 7633 7634Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer 7635than 2Gb. 7636 7637=cut 7638*/ 7639 7640/* 7641 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential 7642 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and 7643 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update(). 7644 * 7645 */ 7646 7647/* This function is subject to size and sign problems */ 7648 7649void 7650Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp) 7651{ 7652 PERL_ARGS_ASSERT_SV_POS_U2B; 7653 7654 if (lenp) { 7655 STRLEN ulen = (STRLEN)*lenp; 7656 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen, 7657 SV_GMAGIC|SV_CONST_RETURN); 7658 *lenp = (I32)ulen; 7659 } else { 7660 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL, 7661 SV_GMAGIC|SV_CONST_RETURN); 7662 } 7663} 7664 7665static void 7666S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, 7667 const STRLEN ulen) 7668{ 7669 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE; 7670 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv)) 7671 return; 7672 7673 if (!*mgp && (SvTYPE(sv) < SVt_PVMG || 7674 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) { 7675 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0); 7676 } 7677 assert(*mgp); 7678 7679 (*mgp)->mg_len = ulen; 7680} 7681 7682/* Create and update the UTF8 magic offset cache, with the proffered utf8/ 7683 byte length pairing. The (byte) length of the total SV is passed in too, 7684 as blen, because for some (more esoteric) SVs, the call to SvPV_const() 7685 may not have updated SvCUR, so we can't rely on reading it directly. 7686 7687 The proffered utf8/byte length pairing isn't used if the cache already has 7688 two pairs, and swapping either for the proffered pair would increase the 7689 RMS of the intervals between known byte offsets. 7690 7691 The cache itself consists of 4 STRLEN values 7692 0: larger UTF-8 offset 7693 1: corresponding byte offset 7694 2: smaller UTF-8 offset 7695 3: corresponding byte offset 7696 7697 Unused cache pairs have the value 0, 0. 7698 Keeping the cache "backwards" means that the invariant of 7699 cache[0] >= cache[2] is maintained even with empty slots, which means that 7700 the code that uses it doesn't need to worry if only 1 entry has actually 7701 been set to non-zero. It also makes the "position beyond the end of the 7702 cache" logic much simpler, as the first slot is always the one to start 7703 from. 7704*/ 7705static void 7706S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte, 7707 const STRLEN utf8, const STRLEN blen) 7708{ 7709 STRLEN *cache; 7710 7711 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE; 7712 7713 if (SvREADONLY(sv)) 7714 return; 7715 7716 if (!*mgp && (SvTYPE(sv) < SVt_PVMG || 7717 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) { 7718 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0, 7719 0); 7720 (*mgp)->mg_len = -1; 7721 } 7722 assert(*mgp); 7723 7724 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) { 7725 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN); 7726 (*mgp)->mg_ptr = (char *) cache; 7727 } 7728 assert(cache); 7729 7730 if (PL_utf8cache < 0 && SvPOKp(sv)) { 7731 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually 7732 a pointer. Note that we no longer cache utf8 offsets on refer- 7733 ences, but this check is still a good idea, for robustness. */ 7734 const U8 *start = (const U8 *) SvPVX_const(sv); 7735 const STRLEN realutf8 = utf8_length(start, start + byte); 7736 7737 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8, 7738 sv); 7739 } 7740 7741 /* Cache is held with the later position first, to simplify the code 7742 that deals with unbounded ends. */ 7743 7744 ASSERT_UTF8_CACHE(cache); 7745 if (cache[1] == 0) { 7746 /* Cache is totally empty */ 7747 cache[0] = utf8; 7748 cache[1] = byte; 7749 } else if (cache[3] == 0) { 7750 if (byte > cache[1]) { 7751 /* New one is larger, so goes first. */ 7752 cache[2] = cache[0]; 7753 cache[3] = cache[1]; 7754 cache[0] = utf8; 7755 cache[1] = byte; 7756 } else { 7757 cache[2] = utf8; 7758 cache[3] = byte; 7759 } 7760 } else { 7761/* float casts necessary? XXX */ 7762#define THREEWAY_SQUARE(a,b,c,d) \ 7763 ((float)((d) - (c))) * ((float)((d) - (c))) \ 7764 + ((float)((c) - (b))) * ((float)((c) - (b))) \ 7765 + ((float)((b) - (a))) * ((float)((b) - (a))) 7766 7767 /* Cache has 2 slots in use, and we know three potential pairs. 7768 Keep the two that give the lowest RMS distance. Do the 7769 calculation in bytes simply because we always know the byte 7770 length. squareroot has the same ordering as the positive value, 7771 so don't bother with the actual square root. */ 7772 if (byte > cache[1]) { 7773 /* New position is after the existing pair of pairs. */ 7774 const float keep_earlier 7775 = THREEWAY_SQUARE(0, cache[3], byte, blen); 7776 const float keep_later 7777 = THREEWAY_SQUARE(0, cache[1], byte, blen); 7778 7779 if (keep_later < keep_earlier) { 7780 cache[2] = cache[0]; 7781 cache[3] = cache[1]; 7782 } 7783 cache[0] = utf8; 7784 cache[1] = byte; 7785 } 7786 else { 7787 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen); 7788 float b, c, keep_earlier; 7789 if (byte > cache[3]) { 7790 /* New position is between the existing pair of pairs. */ 7791 b = (float)cache[3]; 7792 c = (float)byte; 7793 } else { 7794 /* New position is before the existing pair of pairs. */ 7795 b = (float)byte; 7796 c = (float)cache[3]; 7797 } 7798 keep_earlier = THREEWAY_SQUARE(0, b, c, blen); 7799 if (byte > cache[3]) { 7800 if (keep_later < keep_earlier) { 7801 cache[2] = utf8; 7802 cache[3] = byte; 7803 } 7804 else { 7805 cache[0] = utf8; 7806 cache[1] = byte; 7807 } 7808 } 7809 else { 7810 if (! (keep_later < keep_earlier)) { 7811 cache[0] = cache[2]; 7812 cache[1] = cache[3]; 7813 } 7814 cache[2] = utf8; 7815 cache[3] = byte; 7816 } 7817 } 7818 } 7819 ASSERT_UTF8_CACHE(cache); 7820} 7821 7822/* We already know all of the way, now we may be able to walk back. The same 7823 assumption is made as in S_sv_pos_u2b_midway(), namely that walking 7824 backward is half the speed of walking forward. */ 7825static STRLEN 7826S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target, 7827 const U8 *end, STRLEN endu) 7828{ 7829 const STRLEN forw = target - s; 7830 STRLEN backw = end - target; 7831 7832 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY; 7833 7834 if (forw < 2 * backw) { 7835 return utf8_length(s, target); 7836 } 7837 7838 while (end > target) { 7839 end--; 7840 while (UTF8_IS_CONTINUATION(*end)) { 7841 end--; 7842 } 7843 endu--; 7844 } 7845 return endu; 7846} 7847 7848/* 7849=for apidoc sv_pos_b2u_flags 7850 7851Converts C<offset> from a count of bytes from the start of the string, to 7852a count of the equivalent number of UTF-8 chars. Handles type coercion. 7853C<flags> is passed to C<SvPV_flags>, and usually should be 7854C<SV_GMAGIC|SV_CONST_RETURN> to handle magic. 7855 7856=cut 7857*/ 7858 7859/* 7860 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the 7861 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 7862 * and byte offsets. 7863 * 7864 */ 7865STRLEN 7866Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags) 7867{ 7868 const U8* s; 7869 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */ 7870 STRLEN blen; 7871 MAGIC* mg = NULL; 7872 const U8* send; 7873 bool found = FALSE; 7874 7875 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS; 7876 7877 s = (const U8*)SvPV_flags(sv, blen, flags); 7878 7879 if (blen < offset) 7880 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf 7881 ", byte=%" UVuf, (UV)blen, (UV)offset); 7882 7883 send = s + offset; 7884 7885 if (!SvREADONLY(sv) 7886 && PL_utf8cache 7887 && SvTYPE(sv) >= SVt_PVMG 7888 && (mg = mg_find(sv, PERL_MAGIC_utf8))) 7889 { 7890 if (mg->mg_ptr) { 7891 STRLEN * const cache = (STRLEN *) mg->mg_ptr; 7892 if (cache[1] == offset) { 7893 /* An exact match. */ 7894 return cache[0]; 7895 } 7896 if (cache[3] == offset) { 7897 /* An exact match. */ 7898 return cache[2]; 7899 } 7900 7901 if (cache[1] < offset) { 7902 /* We already know part of the way. */ 7903 if (mg->mg_len != -1) { 7904 /* Actually, we know the end too. */ 7905 len = cache[0] 7906 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send, 7907 s + blen, mg->mg_len - cache[0]); 7908 } else { 7909 len = cache[0] + utf8_length(s + cache[1], send); 7910 } 7911 } 7912 else if (cache[3] < offset) { 7913 /* We're between the two cached pairs, so we do the calculation 7914 offset by the byte/utf-8 positions for the earlier pair, 7915 then add the utf-8 characters from the string start to 7916 there. */ 7917 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send, 7918 s + cache[1], cache[0] - cache[2]) 7919 + cache[2]; 7920 7921 } 7922 else { /* cache[3] > offset */ 7923 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3], 7924 cache[2]); 7925 7926 } 7927 ASSERT_UTF8_CACHE(cache); 7928 found = TRUE; 7929 } else if (mg->mg_len != -1) { 7930 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len); 7931 found = TRUE; 7932 } 7933 } 7934 if (!found || PL_utf8cache < 0) { 7935 const STRLEN real_len = utf8_length(s, send); 7936 7937 if (found && PL_utf8cache < 0) 7938 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv); 7939 len = real_len; 7940 } 7941 7942 if (PL_utf8cache) { 7943 if (blen == offset) 7944 utf8_mg_len_cache_update(sv, &mg, len); 7945 else 7946 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen); 7947 } 7948 7949 return len; 7950} 7951 7952/* 7953=for apidoc sv_pos_b2u 7954 7955Converts the value pointed to by C<offsetp> from a count of bytes from the 7956start of the string, to a count of the equivalent number of UTF-8 chars. 7957Handles magic and type coercion. 7958 7959Use C<sv_pos_b2u_flags> in preference, which correctly handles strings 7960longer than 2Gb. 7961 7962=cut 7963*/ 7964 7965/* 7966 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential 7967 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and 7968 * byte offsets. 7969 * 7970 */ 7971void 7972Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp) 7973{ 7974 PERL_ARGS_ASSERT_SV_POS_B2U; 7975 7976 if (!sv) 7977 return; 7978 7979 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp, 7980 SV_GMAGIC|SV_CONST_RETURN); 7981} 7982 7983static void 7984S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache, 7985 STRLEN real, SV *const sv) 7986{ 7987 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT; 7988 7989 /* As this is debugging only code, save space by keeping this test here, 7990 rather than inlining it in all the callers. */ 7991 if (from_cache == real) 7992 return; 7993 7994 /* Need to turn the assertions off otherwise we may recurse infinitely 7995 while printing error messages. */ 7996 SAVEI8(PL_utf8cache); 7997 PL_utf8cache = 0; 7998 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf, 7999 func, (UV) from_cache, (UV) real, SVfARG(sv)); 8000} 8001 8002/* 8003=for apidoc sv_eq 8004 8005Returns a boolean indicating whether the strings in the two SVs are 8006identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will 8007coerce its args to strings if necessary. 8008 8009This function does not handle operator overloading. For a version that does, 8010see instead C<sv_streq>. 8011 8012=for apidoc sv_eq_flags 8013 8014Returns a boolean indicating whether the strings in the two SVs are 8015identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings 8016if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too. 8017 8018This function does not handle operator overloading. For a version that does, 8019see instead C<sv_streq_flags>. 8020 8021=cut 8022*/ 8023 8024I32 8025Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags) 8026{ 8027 const char *pv1; 8028 STRLEN cur1; 8029 const char *pv2; 8030 STRLEN cur2; 8031 8032 if (!sv1) { 8033 pv1 = ""; 8034 cur1 = 0; 8035 } 8036 else { 8037 /* if pv1 and pv2 are the same, second SvPV_const call may 8038 * invalidate pv1 (if we are handling magic), so we may need to 8039 * make a copy */ 8040 if (sv1 == sv2 && flags & SV_GMAGIC 8041 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) { 8042 pv1 = SvPV_const(sv1, cur1); 8043 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2)); 8044 } 8045 pv1 = SvPV_flags_const(sv1, cur1, flags); 8046 } 8047 8048 if (!sv2){ 8049 pv2 = ""; 8050 cur2 = 0; 8051 } 8052 else 8053 pv2 = SvPV_flags_const(sv2, cur2, flags); 8054 8055 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) { 8056 /* Differing utf8ness. */ 8057 if (SvUTF8(sv1)) { 8058 /* sv1 is the UTF-8 one */ 8059 return bytes_cmp_utf8((const U8*)pv2, cur2, 8060 (const U8*)pv1, cur1) == 0; 8061 } 8062 else { 8063 /* sv2 is the UTF-8 one */ 8064 return bytes_cmp_utf8((const U8*)pv1, cur1, 8065 (const U8*)pv2, cur2) == 0; 8066 } 8067 } 8068 8069 if (cur1 == cur2) 8070 return (pv1 == pv2) || memEQ(pv1, pv2, cur1); 8071 else 8072 return 0; 8073} 8074 8075/* 8076=for apidoc sv_streq_flags 8077 8078Returns a boolean indicating whether the strings in the two SVs are 8079identical. If the flags argument has the C<SV_GMAGIC> bit set, it handles 8080get-magic too. Will coerce its args to strings if necessary. Treats 8081C<NULL> as undef. Correctly handles the UTF8 flag. 8082 8083If flags does not have the C<SV_SKIP_OVERLOAD> bit set, an attempt to use 8084C<eq> overloading will be made. If such overloading does not exist or the 8085flag is set, then regular string comparison will be used instead. 8086 8087=for apidoc sv_streq 8088 8089A convenient shortcut for calling C<sv_streq_flags> with the C<SV_GMAGIC> 8090flag. This function basically behaves like the Perl code C<$sv1 eq $sv2>. 8091 8092=cut 8093*/ 8094 8095bool 8096Perl_sv_streq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags) 8097{ 8098 PERL_ARGS_ASSERT_SV_STREQ_FLAGS; 8099 8100 if(flags & SV_GMAGIC) { 8101 if(sv1) 8102 SvGETMAGIC(sv1); 8103 if(sv2) 8104 SvGETMAGIC(sv2); 8105 } 8106 8107 /* Treat NULL as undef */ 8108 if(!sv1) 8109 sv1 = &PL_sv_undef; 8110 if(!sv2) 8111 sv2 = &PL_sv_undef; 8112 8113 if(!(flags & SV_SKIP_OVERLOAD) && 8114 (SvAMAGIC(sv1) || SvAMAGIC(sv2))) { 8115 SV *ret = amagic_call(sv1, sv2, seq_amg, 0); 8116 if(ret) 8117 return SvTRUE(ret); 8118 } 8119 8120 return sv_eq_flags(sv1, sv2, 0); 8121} 8122 8123/* 8124=for apidoc sv_numeq_flags 8125 8126Returns a boolean indicating whether the numbers in the two SVs are 8127identical. If the flags argument has the C<SV_GMAGIC> bit set, it handles 8128get-magic too. Will coerce its args to numbers if necessary. Treats 8129C<NULL> as undef. 8130 8131If flags does not have the C<SV_SKIP_OVERLOAD> bit set, an attempt to use 8132C<==> overloading will be made. If such overloading does not exist or the 8133flag is set, then regular numerical comparison will be used instead. 8134 8135=for apidoc sv_numeq 8136 8137A convenient shortcut for calling C<sv_numeq_flags> with the C<SV_GMAGIC> 8138flag. This function basically behaves like the Perl code C<$sv1 == $sv2>. 8139 8140=cut 8141*/ 8142 8143bool 8144Perl_sv_numeq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags) 8145{ 8146 PERL_ARGS_ASSERT_SV_NUMEQ_FLAGS; 8147 8148 if(flags & SV_GMAGIC) { 8149 if(sv1) 8150 SvGETMAGIC(sv1); 8151 if(sv2) 8152 SvGETMAGIC(sv2); 8153 } 8154 8155 /* Treat NULL as undef */ 8156 if(!sv1) 8157 sv1 = &PL_sv_undef; 8158 if(!sv2) 8159 sv2 = &PL_sv_undef; 8160 8161 if(!(flags & SV_SKIP_OVERLOAD) && 8162 (SvAMAGIC(sv1) || SvAMAGIC(sv2))) { 8163 SV *ret = amagic_call(sv1, sv2, eq_amg, 0); 8164 if(ret) 8165 return SvTRUE(ret); 8166 } 8167 8168 return do_ncmp(sv1, sv2) == 0; 8169} 8170 8171/* 8172=for apidoc sv_cmp 8173 8174Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the 8175string in C<sv1> is less than, equal to, or greater than the string in 8176C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will 8177coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>. 8178 8179=for apidoc sv_cmp_flags 8180 8181Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the 8182string in C<sv1> is less than, equal to, or greater than the string in 8183C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings 8184if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See 8185also C<L</sv_cmp_locale_flags>>. 8186 8187=cut 8188*/ 8189 8190I32 8191Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2) 8192{ 8193 return sv_cmp_flags(sv1, sv2, SV_GMAGIC); 8194} 8195 8196I32 8197Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2, 8198 const U32 flags) 8199{ 8200 STRLEN cur1, cur2; 8201 const char *pv1, *pv2; 8202 I32 cmp; 8203 SV *svrecode = NULL; 8204 8205 if (!sv1) { 8206 pv1 = ""; 8207 cur1 = 0; 8208 } 8209 else 8210 pv1 = SvPV_flags_const(sv1, cur1, flags); 8211 8212 if (!sv2) { 8213 pv2 = ""; 8214 cur2 = 0; 8215 } 8216 else 8217 pv2 = SvPV_flags_const(sv2, cur2, flags); 8218 8219 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) { 8220 /* Differing utf8ness. */ 8221 if (SvUTF8(sv1)) { 8222 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2, 8223 (const U8*)pv1, cur1); 8224 return retval ? retval < 0 ? -1 : +1 : 0; 8225 } 8226 else { 8227 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1, 8228 (const U8*)pv2, cur2); 8229 return retval ? retval < 0 ? -1 : +1 : 0; 8230 } 8231 } 8232 8233 /* Here, if both are non-NULL, then they have the same UTF8ness. */ 8234 8235 if (!cur1) { 8236 cmp = cur2 ? -1 : 0; 8237 } else if (!cur2) { 8238 cmp = 1; 8239 } else { 8240 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2; 8241 8242#ifdef EBCDIC 8243 if (! DO_UTF8(sv1)) { 8244#endif 8245 const I32 retval = memcmp((const void*)pv1, 8246 (const void*)pv2, 8247 shortest_len); 8248 if (retval) { 8249 cmp = retval < 0 ? -1 : 1; 8250 } else if (cur1 == cur2) { 8251 cmp = 0; 8252 } else { 8253 cmp = cur1 < cur2 ? -1 : 1; 8254 } 8255#ifdef EBCDIC 8256 } 8257 else { /* Both are to be treated as UTF-EBCDIC */ 8258 8259 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8 8260 * which remaps code points 0-255. We therefore generally have to 8261 * unmap back to the original values to get an accurate comparison. 8262 * But we don't have to do that for UTF-8 invariants, as by 8263 * definition, they aren't remapped, nor do we have to do it for 8264 * above-latin1 code points, as they also aren't remapped. (This 8265 * code also works on ASCII platforms, but the memcmp() above is 8266 * much faster). */ 8267 8268 const char *e = pv1 + shortest_len; 8269 8270 /* Find the first bytes that differ between the two strings */ 8271 while (pv1 < e && *pv1 == *pv2) { 8272 pv1++; 8273 pv2++; 8274 } 8275 8276 8277 if (pv1 == e) { /* Are the same all the way to the end */ 8278 if (cur1 == cur2) { 8279 cmp = 0; 8280 } else { 8281 cmp = cur1 < cur2 ? -1 : 1; 8282 } 8283 } 8284 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier 8285 * in the strings were. The current bytes may or may not be 8286 * at the beginning of a character. But neither or both are 8287 * (or else earlier bytes would have been different). And 8288 * if we are in the middle of a character, the two 8289 * characters are comprised of the same number of bytes 8290 * (because in this case the start bytes are the same, and 8291 * the start bytes encode the character's length). */ 8292 if (UTF8_IS_INVARIANT(*pv1)) 8293 { 8294 /* If both are invariants; can just compare directly */ 8295 if (UTF8_IS_INVARIANT(*pv2)) { 8296 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; 8297 } 8298 else /* Since *pv1 is invariant, it is the whole character, 8299 which means it is at the beginning of a character. 8300 That means pv2 is also at the beginning of a 8301 character (see earlier comment). Since it isn't 8302 invariant, it must be a start byte. If it starts a 8303 character whose code point is above 255, that 8304 character is greater than any single-byte char, which 8305 *pv1 is */ 8306 if (UTF8_IS_ABOVE_LATIN1_START(*pv2)) 8307 { 8308 cmp = -1; 8309 } 8310 else { 8311 /* Here, pv2 points to a character composed of 2 bytes 8312 * whose code point is < 256. Get its code point and 8313 * compare with *pv1 */ 8314 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1))) 8315 ? -1 8316 : 1; 8317 } 8318 } 8319 else /* The code point starting at pv1 isn't a single byte */ 8320 if (UTF8_IS_INVARIANT(*pv2)) 8321 { 8322 /* But here, the code point starting at *pv2 is a single byte, 8323 * and so *pv1 must begin a character, hence is a start byte. 8324 * If that character is above 255, it is larger than any 8325 * single-byte char, which *pv2 is */ 8326 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) { 8327 cmp = 1; 8328 } 8329 else { 8330 /* Here, pv1 points to a character composed of 2 bytes 8331 * whose code point is < 256. Get its code point and 8332 * compare with the single byte character *pv2 */ 8333 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2) 8334 ? -1 8335 : 1; 8336 } 8337 } 8338 else /* Here, we've ruled out either *pv1 and *pv2 being 8339 invariant. That means both are part of variants, but not 8340 necessarily at the start of a character */ 8341 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1) 8342 || UTF8_IS_ABOVE_LATIN1_START(*pv2)) 8343 { 8344 /* Here, at least one is the start of a character, which means 8345 * the other is also a start byte. And the code point of at 8346 * least one of the characters is above 255. It is a 8347 * characteristic of UTF-EBCDIC that all start bytes for 8348 * above-latin1 code points are well behaved as far as code 8349 * point comparisons go, and all are larger than all other 8350 * start bytes, so the comparison with those is also well 8351 * behaved */ 8352 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; 8353 } 8354 else { 8355 /* Here both *pv1 and *pv2 are part of variant characters. 8356 * They could be both continuations, or both start characters. 8357 * (One or both could even be an illegal start character (for 8358 * an overlong) which for the purposes of sorting we treat as 8359 * legal. */ 8360 if (UTF8_IS_CONTINUATION(*pv1)) { 8361 8362 /* If they are continuations for code points above 255, 8363 * then comparing the current byte is sufficient, as there 8364 * is no remapping of these and so the comparison is 8365 * well-behaved. We determine if they are such 8366 * continuations by looking at the preceding byte. It 8367 * could be a start byte, from which we can tell if it is 8368 * for an above 255 code point. Or it could be a 8369 * continuation, which means the character occupies at 8370 * least 3 bytes, so must be above 255. */ 8371 if ( UTF8_IS_CONTINUATION(*(pv2 - 1)) 8372 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1))) 8373 { 8374 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; 8375 goto cmp_done; 8376 } 8377 8378 /* Here, the continuations are for code points below 256; 8379 * back up one to get to the start byte */ 8380 pv1--; 8381 pv2--; 8382 } 8383 8384 /* We need to get the actual native code point of each of these 8385 * variants in order to compare them */ 8386 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) 8387 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1))) 8388 ? -1 8389 : 1; 8390 } 8391 } 8392 cmp_done: ; 8393#endif 8394 } 8395 8396 SvREFCNT_dec(svrecode); 8397 8398 return cmp; 8399} 8400 8401/* 8402=for apidoc sv_cmp_locale 8403 8404Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and 8405S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings 8406if necessary. See also C<L</sv_cmp>>. 8407 8408=for apidoc sv_cmp_locale_flags 8409 8410Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and 8411S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If 8412the flags contain C<SV_GMAGIC>, it handles get magic. See also 8413C<L</sv_cmp_flags>>. 8414 8415=cut 8416*/ 8417 8418I32 8419Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2) 8420{ 8421 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC); 8422} 8423 8424I32 8425Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2, 8426 const U32 flags) 8427{ 8428#ifdef USE_LOCALE_COLLATE 8429 8430 char *pv1, *pv2; 8431 STRLEN len1, len2; 8432 I32 retval; 8433 8434 if (PL_collation_standard) 8435 goto raw_compare; 8436 8437 len1 = len2 = 0; 8438 8439 /* Revert to using raw compare if both operands exist, but either one 8440 * doesn't transform properly for collation */ 8441 if (sv1 && sv2) { 8442 pv1 = sv_collxfrm_flags(sv1, &len1, flags); 8443 if (! pv1) { 8444 goto raw_compare; 8445 } 8446 pv2 = sv_collxfrm_flags(sv2, &len2, flags); 8447 if (! pv2) { 8448 goto raw_compare; 8449 } 8450 } 8451 else { 8452 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL; 8453 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL; 8454 } 8455 8456 if (!pv1 || !len1) { 8457 if (pv2 && len2) 8458 return -1; 8459 else 8460 goto raw_compare; 8461 } 8462 else { 8463 if (!pv2 || !len2) 8464 return 1; 8465 } 8466 8467 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2); 8468 8469 if (retval) 8470 return retval < 0 ? -1 : 1; 8471 8472 /* 8473 * When the result of collation is equality, that doesn't mean 8474 * that there are no differences -- some locales exclude some 8475 * characters from consideration. So to avoid false equalities, 8476 * we use the raw string as a tiebreaker. 8477 */ 8478 8479 raw_compare: 8480 /* FALLTHROUGH */ 8481 8482#else 8483 PERL_UNUSED_ARG(flags); 8484#endif /* USE_LOCALE_COLLATE */ 8485 8486 return sv_cmp(sv1, sv2); 8487} 8488 8489 8490#ifdef USE_LOCALE_COLLATE 8491 8492/* 8493=for apidoc sv_collxfrm 8494 8495This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See 8496C<L</sv_collxfrm_flags>>. 8497 8498=for apidoc sv_collxfrm_flags 8499 8500Add Collate Transform magic to an SV if it doesn't already have it. If the 8501flags contain C<SV_GMAGIC>, it handles get-magic. 8502 8503Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the 8504scalar data of the variable, but transformed to such a format that a normal 8505memory comparison can be used to compare the data according to the locale 8506settings. 8507 8508=cut 8509*/ 8510 8511char * 8512Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags) 8513{ 8514 MAGIC *mg; 8515 8516 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS; 8517 8518 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL; 8519 8520 /* If we don't have collation magic on 'sv', or the locale has changed 8521 * since the last time we calculated it, get it and save it now */ 8522 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) { 8523 const char *s; 8524 char *xf; 8525 STRLEN len, xlen; 8526 8527 /* Free the old space */ 8528 if (mg) 8529 Safefree(mg->mg_ptr); 8530 8531 s = SvPV_flags_const(sv, len, flags); 8532 if ((xf = mem_collxfrm_(s, len, &xlen, cBOOL(SvUTF8(sv))))) { 8533 if (! mg) { 8534 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm, 8535 0, 0); 8536 assert(mg); 8537 } 8538 mg->mg_ptr = xf; 8539 mg->mg_len = xlen; 8540 } 8541 else { 8542 if (mg) { 8543 mg->mg_ptr = NULL; 8544 mg->mg_len = -1; 8545 } 8546 } 8547 } 8548 8549 if (mg && mg->mg_ptr) { 8550 *nxp = mg->mg_len; 8551 return mg->mg_ptr + sizeof(PL_collation_ix); 8552 } 8553 else { 8554 *nxp = 0; 8555 return NULL; 8556 } 8557} 8558 8559#endif /* USE_LOCALE_COLLATE */ 8560 8561static char * 8562S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append) 8563{ 8564 SV * const tsv = newSV_type(SVt_NULL); 8565 ENTER; 8566 SAVEFREESV(tsv); 8567 sv_gets(tsv, fp, 0); 8568 sv_utf8_upgrade_nomg(tsv); 8569 SvCUR_set(sv,append); 8570 sv_catsv(sv,tsv); 8571 LEAVE; 8572 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; 8573} 8574 8575static char * 8576S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append) 8577{ 8578 SSize_t bytesread; 8579 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */ 8580 /* Grab the size of the record we're getting */ 8581 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append; 8582 8583 /* Go yank in */ 8584#ifdef __VMS 8585 int fd; 8586 Stat_t st; 8587 8588 /* With a true, record-oriented file on VMS, we need to use read directly 8589 * to ensure that we respect RMS record boundaries. The user is responsible 8590 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum 8591 * record size) field. N.B. This is likely to produce invalid results on 8592 * varying-width character data when a record ends mid-character. 8593 */ 8594 fd = PerlIO_fileno(fp); 8595 if (fd != -1 8596 && PerlLIO_fstat(fd, &st) == 0 8597 && (st.st_fab_rfm == FAB$C_VAR 8598 || st.st_fab_rfm == FAB$C_VFC 8599 || st.st_fab_rfm == FAB$C_FIX)) { 8600 8601 bytesread = PerlLIO_read(fd, buffer, recsize); 8602 } 8603 else /* in-memory file from PerlIO::Scalar 8604 * or not a record-oriented file 8605 */ 8606#endif 8607 { 8608 bytesread = PerlIO_read(fp, buffer, recsize); 8609 8610 /* At this point, the logic in sv_get() means that sv will 8611 be treated as utf-8 if the handle is utf8. 8612 */ 8613 if (PerlIO_isutf8(fp) && bytesread > 0) { 8614 char *bend = buffer + bytesread; 8615 char *bufp = buffer; 8616 size_t charcount = 0; 8617 bool charstart = TRUE; 8618 STRLEN skip = 0; 8619 8620 while (charcount < recsize) { 8621 /* count accumulated characters */ 8622 while (bufp < bend) { 8623 if (charstart) { 8624 skip = UTF8SKIP(bufp); 8625 } 8626 if (bufp + skip > bend) { 8627 /* partial at the end */ 8628 charstart = FALSE; 8629 break; 8630 } 8631 else { 8632 ++charcount; 8633 bufp += skip; 8634 charstart = TRUE; 8635 } 8636 } 8637 8638 if (charcount < recsize) { 8639 STRLEN readsize; 8640 STRLEN bufp_offset = bufp - buffer; 8641 SSize_t morebytesread; 8642 8643 /* originally I read enough to fill any incomplete 8644 character and the first byte of the next 8645 character if needed, but if there's many 8646 multi-byte encoded characters we're going to be 8647 making a read call for every character beyond 8648 the original read size. 8649 8650 So instead, read the rest of the character if 8651 any, and enough bytes to match at least the 8652 start bytes for each character we're going to 8653 read. 8654 */ 8655 if (charstart) 8656 readsize = recsize - charcount; 8657 else 8658 readsize = skip - (bend - bufp) + recsize - charcount - 1; 8659 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append; 8660 bend = buffer + bytesread; 8661 morebytesread = PerlIO_read(fp, bend, readsize); 8662 if (morebytesread <= 0) { 8663 /* we're done, if we still have incomplete 8664 characters the check code in sv_gets() will 8665 warn about them. 8666 8667 I'd originally considered doing 8668 PerlIO_ungetc() on all but the lead 8669 character of the incomplete character, but 8670 read() doesn't do that, so I don't. 8671 */ 8672 break; 8673 } 8674 8675 /* prepare to scan some more */ 8676 bytesread += morebytesread; 8677 bend = buffer + bytesread; 8678 bufp = buffer + bufp_offset; 8679 } 8680 } 8681 } 8682 } 8683 8684 if (bytesread < 0) 8685 bytesread = 0; 8686 SvCUR_set(sv, bytesread + append); 8687 buffer[bytesread] = '\0'; 8688 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; 8689} 8690 8691/* 8692=for apidoc sv_gets 8693 8694Get a line from the filehandle and store it into the SV, optionally 8695appending to the currently-stored string. If C<append> is not 0, the 8696line is appended to the SV instead of overwriting it. C<append> should 8697be set to the byte offset that the appended string should start at 8698in the SV (typically, C<SvCUR(sv)> is a suitable choice). 8699 8700=cut 8701*/ 8702 8703char * 8704Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append) 8705{ 8706 const char *rsptr; 8707 STRLEN rslen; 8708 STDCHAR rslast; 8709 STDCHAR *bp; 8710 SSize_t cnt; 8711 int i = 0; 8712 int rspara = 0; 8713 8714 PERL_ARGS_ASSERT_SV_GETS; 8715 8716 if (SvTHINKFIRST(sv)) 8717 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV); 8718 /* XXX. If you make this PVIV, then copy on write can copy scalars read 8719 from <>. 8720 However, perlbench says it's slower, because the existing swipe code 8721 is faster than copy on write. 8722 Swings and roundabouts. */ 8723 SvUPGRADE(sv, SVt_PV); 8724 8725 if (append) { 8726 /* line is going to be appended to the existing buffer in the sv */ 8727 if (PerlIO_isutf8(fp)) { 8728 if (!SvUTF8(sv)) { 8729 sv_utf8_upgrade_nomg(sv); 8730 sv_pos_u2b(sv,&append,0); 8731 } 8732 } else if (SvUTF8(sv)) { 8733 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append); 8734 } 8735 } 8736 8737 SvPOK_only(sv); 8738 if (!append) { 8739 /* not appending - "clear" the string by setting SvCUR to 0, 8740 * the pv is still available. */ 8741 SvCUR_set(sv,0); 8742 } 8743 if (PerlIO_isutf8(fp)) 8744 SvUTF8_on(sv); 8745 8746 if (IN_PERL_COMPILETIME) { 8747 /* we always read code in line mode */ 8748 rsptr = "\n"; 8749 rslen = 1; 8750 } 8751 else if (RsSNARF(PL_rs)) { 8752 /* If it is a regular disk file use size from stat() as estimate 8753 of amount we are going to read -- may result in mallocing 8754 more memory than we really need if the layers below reduce 8755 the size we read (e.g. CRLF or a gzip layer). 8756 */ 8757 Stat_t st; 8758 int fd = PerlIO_fileno(fp); 8759 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) { 8760 const Off_t offset = PerlIO_tell(fp); 8761 if (offset != (Off_t) -1 && st.st_size + append > offset) { 8762#ifdef PERL_COPY_ON_WRITE 8763 /* Add an extra byte for the sake of copy-on-write's 8764 * buffer reference count. */ 8765 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2)); 8766#else 8767 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1)); 8768#endif 8769 } 8770 } 8771 rsptr = NULL; 8772 rslen = 0; 8773 } 8774 else if (RsRECORD(PL_rs)) { 8775 return S_sv_gets_read_record(aTHX_ sv, fp, append); 8776 } 8777 else if (RsPARA(PL_rs)) { 8778 rsptr = "\n\n"; 8779 rslen = 2; 8780 rspara = 1; 8781 } 8782 else { 8783 /* Get $/ i.e. PL_rs into same encoding as stream wants */ 8784 if (PerlIO_isutf8(fp)) { 8785 rsptr = SvPVutf8(PL_rs, rslen); 8786 } 8787 else { 8788 if (SvUTF8(PL_rs)) { 8789 if (!sv_utf8_downgrade(PL_rs, TRUE)) { 8790 Perl_croak(aTHX_ "Wide character in $/"); 8791 } 8792 } 8793 /* extract the raw pointer to the record separator */ 8794 rsptr = SvPV_const(PL_rs, rslen); 8795 } 8796 } 8797 8798 /* rslast is the last character in the record separator 8799 * note we don't use rslast except when rslen is true, so the 8800 * null assign is a placeholder. */ 8801 rslast = rslen ? rsptr[rslen - 1] : '\0'; 8802 8803 if (rspara) { /* have to do this both before and after */ 8804 /* to make sure file boundaries work right */ 8805 while (1) { 8806 if (PerlIO_eof(fp)) 8807 return 0; 8808 i = PerlIO_getc(fp); 8809 if (i != '\n') { 8810 if (i == -1) 8811 return 0; 8812 PerlIO_ungetc(fp,i); 8813 break; 8814 } 8815 } 8816 } 8817 8818 /* See if we know enough about I/O mechanism to cheat it ! */ 8819 8820 /* This used to be #ifdef test - it is made run-time test for ease 8821 of abstracting out stdio interface. One call should be cheap 8822 enough here - and may even be a macro allowing compile 8823 time optimization. 8824 */ 8825 8826 if (PerlIO_fast_gets(fp)) { 8827 /* 8828 * We can do buffer based IO operations on this filehandle. 8829 * 8830 * This means we can bypass a lot of subcalls and process 8831 * the buffer directly, it also means we know the upper bound 8832 * on the amount of data we might read of the current buffer 8833 * into our sv. Knowing this allows us to preallocate the pv 8834 * to be able to hold that maximum, which allows us to simplify 8835 * a lot of logic. */ 8836 8837 /* 8838 * We're going to steal some values from the stdio struct 8839 * and put EVERYTHING in the innermost loop into registers. 8840 */ 8841 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */ 8842 STRLEN bpx; /* length of the data in the target sv 8843 used to fix pointers after a SvGROW */ 8844 I32 shortbuffered; /* If the pv buffer is shorter than the amount 8845 of data left in the read-ahead buffer. 8846 If 0 then the pv buffer can hold the full 8847 amount left, otherwise this is the amount it 8848 can hold. */ 8849 8850 /* Here is some breathtakingly efficient cheating */ 8851 8852 /* When you read the following logic resist the urge to think 8853 * of record separators that are 1 byte long. They are an 8854 * uninteresting special (simple) case. 8855 * 8856 * Instead think of record separators which are at least 2 bytes 8857 * long, and keep in mind that we need to deal with such 8858 * separators when they cross a read-ahead buffer boundary. 8859 * 8860 * Also consider that we need to gracefully deal with separators 8861 * that may be longer than a single read ahead buffer. 8862 * 8863 * Lastly do not forget we want to copy the delimiter as well. We 8864 * are copying all data in the file _up_to_and_including_ the separator 8865 * itself. 8866 * 8867 * Now that you have all that in mind here is what is happening below: 8868 * 8869 * 1. When we first enter the loop we do some memory book keeping to see 8870 * how much free space there is in the target SV. (This sub assumes that 8871 * it is operating on the same SV most of the time via $_ and that it is 8872 * going to be able to reuse the same pv buffer each call.) If there is 8873 * "enough" room then we set "shortbuffered" to how much space there is 8874 * and start reading forward. 8875 * 8876 * 2. When we scan forward we copy from the read-ahead buffer to the target 8877 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer, 8878 * and the end of the of pv, as well as for the "rslast", which is the last 8879 * char of the separator. 8880 * 8881 * 3. When scanning forward if we see rslast then we jump backwards in *pv* 8882 * (which has a "complete" record up to the point we saw rslast) and check 8883 * it to see if it matches the separator. If it does we are done. If it doesn't 8884 * we continue on with the scan/copy. 8885 * 8886 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get 8887 * the IO system to read the next buffer. We do this by doing a getc(), which 8888 * returns a single char read (or EOF), and prefills the buffer, and also 8889 * allows us to find out how full the buffer is. We use this information to 8890 * SvGROW() the sv to the size remaining in the buffer, after which we copy 8891 * the returned single char into the target sv, and then go back into scan 8892 * forward mode. 8893 * 8894 * 5. If we run out of write-buffer then we SvGROW() it by the size of the 8895 * remaining space in the read-buffer. 8896 * 8897 * Note that this code despite its twisty-turny nature is pretty darn slick. 8898 * It manages single byte separators, multi-byte cross boundary separators, 8899 * and cross-read-buffer separators cleanly and efficiently at the cost 8900 * of potentially greatly overallocating the target SV. 8901 * 8902 * Yves 8903 */ 8904 8905 8906 /* get the number of bytes remaining in the read-ahead buffer 8907 * on first call on a given fp this will return 0.*/ 8908 cnt = PerlIO_get_cnt(fp); 8909 8910 /* make sure we have the room */ 8911 if ((I32)(SvLEN(sv) - append) <= cnt + 1) { 8912 /* Not room for all of it 8913 if we are looking for a separator and room for some 8914 */ 8915 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) { 8916 /* just process what we have room for */ 8917 shortbuffered = cnt - SvLEN(sv) + append + 1; 8918 cnt -= shortbuffered; 8919 } 8920 else { 8921 /* ensure that the target sv has enough room to hold 8922 * the rest of the read-ahead buffer */ 8923 shortbuffered = 0; 8924 /* remember that cnt can be negative */ 8925 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1)))); 8926 } 8927 } 8928 else { 8929 /* we have enough room to hold the full buffer, lets scream */ 8930 shortbuffered = 0; 8931 } 8932 8933 /* extract the pointer to sv's string buffer, offset by append as necessary */ 8934 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */ 8935 /* extract the point to the read-ahead buffer */ 8936 ptr = (STDCHAR*)PerlIO_get_ptr(fp); 8937 8938 /* some trace debug output */ 8939 DEBUG_P(PerlIO_printf(Perl_debug_log, 8940 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt)); 8941 DEBUG_P(PerlIO_printf(Perl_debug_log, 8942 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" 8943 UVuf "\n", 8944 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), 8945 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0))); 8946 8947 for (;;) { 8948 screamer: 8949 /* if there is stuff left in the read-ahead buffer */ 8950 if (cnt > 0) { 8951 /* if there is a separator */ 8952 if (rslen) { 8953 /* find next rslast */ 8954 STDCHAR *p; 8955 8956 /* shortcut common case of blank line */ 8957 cnt--; 8958 if ((*bp++ = *ptr++) == rslast) 8959 goto thats_all_folks; 8960 8961 p = (STDCHAR *)memchr(ptr, rslast, cnt); 8962 if (p) { 8963 SSize_t got = p - ptr + 1; 8964 Copy(ptr, bp, got, STDCHAR); 8965 ptr += got; 8966 bp += got; 8967 cnt -= got; 8968 goto thats_all_folks; 8969 } 8970 Copy(ptr, bp, cnt, STDCHAR); 8971 ptr += cnt; 8972 bp += cnt; 8973 cnt = 0; 8974 } 8975 else { 8976 /* no separator, slurp the full buffer */ 8977 Copy(ptr, bp, cnt, char); /* this | eat */ 8978 bp += cnt; /* screams | dust */ 8979 ptr += cnt; /* louder | sed :-) */ 8980 cnt = 0; 8981 assert (!shortbuffered); 8982 goto cannot_be_shortbuffered; 8983 } 8984 } 8985 8986 if (shortbuffered) { /* oh well, must extend */ 8987 /* we didn't have enough room to fit the line into the target buffer 8988 * so we must extend the target buffer and keep going */ 8989 cnt = shortbuffered; 8990 shortbuffered = 0; 8991 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */ 8992 SvCUR_set(sv, bpx); 8993 /* extned the target sv's buffer so it can hold the full read-ahead buffer */ 8994 SvGROW(sv, SvLEN(sv) + append + cnt + 2); 8995 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */ 8996 continue; 8997 } 8998 8999 cannot_be_shortbuffered: 9000 /* we need to refill the read-ahead buffer if possible */ 9001 9002 DEBUG_P(PerlIO_printf(Perl_debug_log, 9003 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n", 9004 PTR2UV(ptr),(IV)cnt)); 9005 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */ 9006 9007 DEBUG_Pv(PerlIO_printf(Perl_debug_log, 9008 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n", 9009 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), 9010 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); 9011 9012 /* 9013 call PerlIO_getc() to let it prefill the lookahead buffer 9014 9015 This used to call 'filbuf' in stdio form, but as that behaves like 9016 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing 9017 another abstraction. 9018 9019 Note we have to deal with the char in 'i' if we are not at EOF 9020 */ 9021 bpx = bp - (STDCHAR*)SvPVX_const(sv); 9022 /* signals might be called here, possibly modifying sv */ 9023 i = PerlIO_getc(fp); /* get more characters */ 9024 bp = (STDCHAR*)SvPVX_const(sv) + bpx; 9025 9026 DEBUG_Pv(PerlIO_printf(Perl_debug_log, 9027 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n", 9028 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), 9029 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); 9030 9031 /* find out how much is left in the read-ahead buffer, and rextract its pointer */ 9032 cnt = PerlIO_get_cnt(fp); 9033 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */ 9034 DEBUG_P(PerlIO_printf(Perl_debug_log, 9035 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n", 9036 PTR2UV(ptr),(IV)cnt)); 9037 9038 if (i == EOF) /* all done for ever? */ 9039 goto thats_really_all_folks; 9040 9041 /* make sure we have enough space in the target sv */ 9042 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */ 9043 SvCUR_set(sv, bpx); 9044 SvGROW(sv, bpx + cnt + 2); 9045 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */ 9046 9047 /* copy of the char we got from getc() */ 9048 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */ 9049 9050 /* make sure we deal with the i being the last character of a separator */ 9051 if (rslen && (STDCHAR)i == rslast) /* all done for now? */ 9052 goto thats_all_folks; 9053 } 9054 9055 thats_all_folks: 9056 /* check if we have actually found the separator - only really applies 9057 * when rslen > 1 */ 9058 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) || 9059 memNE((char*)bp - rslen, rsptr, rslen)) 9060 goto screamer; /* go back to the fray */ 9061 thats_really_all_folks: 9062 if (shortbuffered) 9063 cnt += shortbuffered; 9064 DEBUG_P(PerlIO_printf(Perl_debug_log, 9065 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt)); 9066 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */ 9067 DEBUG_P(PerlIO_printf(Perl_debug_log, 9068 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf 9069 "\n", 9070 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), 9071 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); 9072 *bp = '\0'; 9073 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */ 9074 DEBUG_P(PerlIO_printf(Perl_debug_log, 9075 "Screamer: done, len=%ld, string=|%.*s|\n", 9076 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv))); 9077 } 9078 else 9079 { 9080 /*The big, slow, and stupid way. */ 9081 STDCHAR buf[8192]; 9082 9083 screamer2: 9084 if (rslen) { 9085 const STDCHAR * const bpe = buf + sizeof(buf); 9086 bp = buf; 9087 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe) 9088 ; /* keep reading */ 9089 cnt = bp - buf; 9090 } 9091 else { 9092 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf)); 9093 /* Accommodate broken VAXC compiler, which applies U8 cast to 9094 * both args of ?: operator, causing EOF to change into 255 9095 */ 9096 if (cnt > 0) 9097 i = (U8)buf[cnt - 1]; 9098 else 9099 i = EOF; 9100 } 9101 9102 if (cnt < 0) 9103 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */ 9104 if (append) 9105 sv_catpvn_nomg(sv, (char *) buf, cnt); 9106 else 9107 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */ 9108 9109 if (i != EOF && /* joy */ 9110 (!rslen || 9111 SvCUR(sv) < rslen || 9112 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen))) 9113 { 9114 append = -1; 9115 /* 9116 * If we're reading from a TTY and we get a short read, 9117 * indicating that the user hit his EOF character, we need 9118 * to notice it now, because if we try to read from the TTY 9119 * again, the EOF condition will disappear. 9120 * 9121 * The comparison of cnt to sizeof(buf) is an optimization 9122 * that prevents unnecessary calls to feof(). 9123 * 9124 * - jik 9/25/96 9125 */ 9126 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp))) 9127 goto screamer2; 9128 } 9129 9130 } 9131 9132 if (rspara) { /* have to do this both before and after */ 9133 while (i != EOF) { /* to make sure file boundaries work right */ 9134 i = PerlIO_getc(fp); 9135 if (i != '\n') { 9136 PerlIO_ungetc(fp,i); 9137 break; 9138 } 9139 } 9140 } 9141 9142 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; 9143} 9144 9145/* 9146=for apidoc sv_inc 9147=for apidoc_item sv_inc_nomg 9148 9149These auto-increment the value in the SV, doing string to numeric conversion 9150if necessary. They both handle operator overloading. 9151 9152They differ only in that C<sv_inc> performs 'get' magic; C<sv_inc_nomg> skips 9153any magic. 9154 9155=cut 9156*/ 9157 9158void 9159Perl_sv_inc(pTHX_ SV *const sv) 9160{ 9161 if (!sv) 9162 return; 9163 SvGETMAGIC(sv); 9164 sv_inc_nomg(sv); 9165} 9166 9167void 9168Perl_sv_inc_nomg(pTHX_ SV *const sv) 9169{ 9170 char *d; 9171 int flags; 9172 9173 if (!sv) 9174 return; 9175 if (SvTHINKFIRST(sv)) { 9176 if (SvREADONLY(sv)) { 9177 Perl_croak_no_modify(); 9178 } 9179 if (SvROK(sv)) { 9180 IV i; 9181 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg)) 9182 return; 9183 i = PTR2IV(SvRV(sv)); 9184 sv_unref(sv); 9185 sv_setiv(sv, i); 9186 } 9187 else sv_force_normal_flags(sv, 0); 9188 } 9189 flags = SvFLAGS(sv); 9190 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) { 9191 /* It's (privately or publicly) a float, but not tested as an 9192 integer, so test it to see. */ 9193 (void) SvIV(sv); 9194 flags = SvFLAGS(sv); 9195 } 9196 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) { 9197 /* It's publicly an integer, or privately an integer-not-float */ 9198#ifdef PERL_PRESERVE_IVUV 9199 oops_its_int: 9200#endif 9201 if (SvIsUV(sv)) { 9202 if (SvUVX(sv) == UV_MAX) 9203 sv_setnv(sv, UV_MAX_P1); 9204 else { 9205 (void)SvIOK_only_UV(sv); 9206 SvUV_set(sv, SvUVX(sv) + 1); 9207 } 9208 } else { 9209 if (SvIVX(sv) == IV_MAX) 9210 sv_setuv(sv, (UV)IV_MAX + 1); 9211 else { 9212 (void)SvIOK_only(sv); 9213 SvIV_set(sv, SvIVX(sv) + 1); 9214 } 9215 } 9216 return; 9217 } 9218 if (flags & SVp_NOK) { 9219 const NV was = SvNVX(sv); 9220 if (NV_OVERFLOWS_INTEGERS_AT != 0.0 && 9221 /* If NVX was NaN, the following comparisons return always false */ 9222 UNLIKELY(was >= NV_OVERFLOWS_INTEGERS_AT || 9223 was < -NV_OVERFLOWS_INTEGERS_AT) && 9224#if defined(NAN_COMPARE_BROKEN) 9225 LIKELY(!Perl_isinfnan(was)) 9226#else 9227 LIKELY(!Perl_isinf(was)) 9228#endif 9229 ) { 9230 /* diag_listed_as: Lost precision when %s %f by 1 */ 9231 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION), 9232 "Lost precision when incrementing %" NVff " by 1", 9233 was); 9234 } 9235 (void)SvNOK_only(sv); 9236 SvNV_set(sv, was + 1.0); 9237 return; 9238 } 9239 9240 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */ 9241 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv))) 9242 Perl_croak_no_modify(); 9243 9244 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) { 9245 if ((flags & SVTYPEMASK) < SVt_PVIV) 9246 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV)); 9247 (void)SvIOK_only(sv); 9248 SvIV_set(sv, 1); 9249 return; 9250 } 9251 d = SvPVX(sv); 9252 while (isALPHA(*d)) d++; 9253 while (isDIGIT(*d)) d++; 9254 if (d < SvEND(sv)) { 9255 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING); 9256#ifdef PERL_PRESERVE_IVUV 9257 /* Got to punt this as an integer if needs be, but we don't issue 9258 warnings. Probably ought to make the sv_iv_please() that does 9259 the conversion if possible, and silently. */ 9260 if (numtype && !(numtype & IS_NUMBER_INFINITY)) { 9261 /* Need to try really hard to see if it's an integer. 9262 9.22337203685478e+18 is an integer. 9263 but "9.22337203685478e+18" + 0 is UV=9223372036854779904 9264 so $a="9.22337203685478e+18"; $a+0; $a++ 9265 needs to be the same as $a="9.22337203685478e+18"; $a++ 9266 or we go insane. */ 9267 9268 (void) sv_2iv(sv); 9269 if (SvIOK(sv)) 9270 goto oops_its_int; 9271 9272 /* sv_2iv *should* have made this an NV */ 9273 if (flags & SVp_NOK) { 9274 (void)SvNOK_only(sv); 9275 SvNV_set(sv, SvNVX(sv) + 1.0); 9276 return; 9277 } 9278 /* I don't think we can get here. Maybe I should assert this 9279 And if we do get here I suspect that sv_setnv will croak. NWC 9280 Fall through. */ 9281 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n", 9282 SvPVX_const(sv), SvIVX(sv), SvNVX(sv))); 9283 } 9284#endif /* PERL_PRESERVE_IVUV */ 9285 if (!numtype && ckWARN(WARN_NUMERIC)) 9286 not_incrementable(sv); 9287 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0); 9288 return; 9289 } 9290 d--; 9291 while (d >= SvPVX_const(sv)) { 9292 if (isDIGIT(*d)) { 9293 if (++*d <= '9') 9294 return; 9295 *(d--) = '0'; 9296 } 9297 else { 9298#ifdef EBCDIC 9299 /* MKS: The original code here died if letters weren't consecutive. 9300 * at least it didn't have to worry about non-C locales. The 9301 * new code assumes that ('z'-'a')==('Z'-'A'), letters are 9302 * arranged in order (although not consecutively) and that only 9303 * [A-Za-z] are accepted by isALPHA in the C locale. 9304 */ 9305 if (isALPHA_FOLD_NE(*d, 'z')) { 9306 do { ++*d; } while (!isALPHA(*d)); 9307 return; 9308 } 9309 *(d--) -= 'z' - 'a'; 9310#else 9311 ++*d; 9312 if (isALPHA(*d)) 9313 return; 9314 *(d--) -= 'z' - 'a' + 1; 9315#endif 9316 } 9317 } 9318 /* oh,oh, the number grew */ 9319 SvGROW(sv, SvCUR(sv) + 2); 9320 SvCUR_set(sv, SvCUR(sv) + 1); 9321 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--) 9322 *d = d[-1]; 9323 if (isDIGIT(d[1])) 9324 *d = '1'; 9325 else 9326 *d = d[1]; 9327} 9328 9329/* 9330=for apidoc sv_dec 9331=for apidoc_item sv_dec_nomg 9332 9333These auto-decrement the value in the SV, doing string to numeric conversion 9334if necessary. They both handle operator overloading. 9335 9336They differ only in that: 9337 9338C<sv_dec> handles 'get' magic; C<sv_dec_nomg> skips 'get' magic. 9339 9340=cut 9341*/ 9342 9343void 9344Perl_sv_dec(pTHX_ SV *const sv) 9345{ 9346 if (!sv) 9347 return; 9348 SvGETMAGIC(sv); 9349 sv_dec_nomg(sv); 9350} 9351 9352void 9353Perl_sv_dec_nomg(pTHX_ SV *const sv) 9354{ 9355 int flags; 9356 9357 if (!sv) 9358 return; 9359 if (SvTHINKFIRST(sv)) { 9360 if (SvREADONLY(sv)) { 9361 Perl_croak_no_modify(); 9362 } 9363 if (SvROK(sv)) { 9364 IV i; 9365 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg)) 9366 return; 9367 i = PTR2IV(SvRV(sv)); 9368 sv_unref(sv); 9369 sv_setiv(sv, i); 9370 } 9371 else sv_force_normal_flags(sv, 0); 9372 } 9373 /* Unlike sv_inc we don't have to worry about string-never-numbers 9374 and keeping them magic. But we mustn't warn on punting */ 9375 flags = SvFLAGS(sv); 9376 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) { 9377 /* It's publicly an integer, or privately an integer-not-float */ 9378#ifdef PERL_PRESERVE_IVUV 9379 oops_its_int: 9380#endif 9381 if (SvIsUV(sv)) { 9382 if (SvUVX(sv) == 0) { 9383 (void)SvIOK_only(sv); 9384 SvIV_set(sv, -1); 9385 } 9386 else { 9387 (void)SvIOK_only_UV(sv); 9388 SvUV_set(sv, SvUVX(sv) - 1); 9389 } 9390 } else { 9391 if (SvIVX(sv) == IV_MIN) { 9392 sv_setnv(sv, (NV)IV_MIN); 9393 goto oops_its_num; 9394 } 9395 else { 9396 (void)SvIOK_only(sv); 9397 SvIV_set(sv, SvIVX(sv) - 1); 9398 } 9399 } 9400 return; 9401 } 9402 if (flags & SVp_NOK) { 9403 oops_its_num: 9404 { 9405 const NV was = SvNVX(sv); 9406 if (NV_OVERFLOWS_INTEGERS_AT != 0.0 && 9407 /* If NVX was NaN, these comparisons return always false */ 9408 UNLIKELY(was <= -NV_OVERFLOWS_INTEGERS_AT || 9409 was > NV_OVERFLOWS_INTEGERS_AT) && 9410#if defined(NAN_COMPARE_BROKEN) 9411 LIKELY(!Perl_isinfnan(was)) 9412#else 9413 LIKELY(!Perl_isinf(was)) 9414#endif 9415 ) { 9416 /* diag_listed_as: Lost precision when %s %f by 1 */ 9417 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION), 9418 "Lost precision when decrementing %" NVff " by 1", 9419 was); 9420 } 9421 (void)SvNOK_only(sv); 9422 SvNV_set(sv, was - 1.0); 9423 return; 9424 } 9425 } 9426 9427 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */ 9428 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv))) 9429 Perl_croak_no_modify(); 9430 9431 if (!(flags & SVp_POK)) { 9432 if ((flags & SVTYPEMASK) < SVt_PVIV) 9433 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV); 9434 SvIV_set(sv, -1); 9435 (void)SvIOK_only(sv); 9436 return; 9437 } 9438#ifdef PERL_PRESERVE_IVUV 9439 { 9440 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL); 9441 if (numtype && !(numtype & IS_NUMBER_INFINITY)) { 9442 /* Need to try really hard to see if it's an integer. 9443 9.22337203685478e+18 is an integer. 9444 but "9.22337203685478e+18" + 0 is UV=9223372036854779904 9445 so $a="9.22337203685478e+18"; $a+0; $a-- 9446 needs to be the same as $a="9.22337203685478e+18"; $a-- 9447 or we go insane. */ 9448 9449 (void) sv_2iv(sv); 9450 if (SvIOK(sv)) 9451 goto oops_its_int; 9452 9453 /* sv_2iv *should* have made this an NV */ 9454 if (flags & SVp_NOK) { 9455 (void)SvNOK_only(sv); 9456 SvNV_set(sv, SvNVX(sv) - 1.0); 9457 return; 9458 } 9459 /* I don't think we can get here. Maybe I should assert this 9460 And if we do get here I suspect that sv_setnv will croak. NWC 9461 Fall through. */ 9462 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n", 9463 SvPVX_const(sv), SvIVX(sv), SvNVX(sv))); 9464 } 9465 } 9466#endif /* PERL_PRESERVE_IVUV */ 9467 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */ 9468} 9469 9470/* this define is used to eliminate a chunk of duplicated but shared logic 9471 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be 9472 * used anywhere but here - yves 9473 */ 9474#define PUSH_EXTEND_MORTAL__SV_C(AnSv) \ 9475 STMT_START { \ 9476 SSize_t ix = ++PL_tmps_ix; \ 9477 if (UNLIKELY(ix >= PL_tmps_max)) \ 9478 ix = tmps_grow_p(ix); \ 9479 PL_tmps_stack[ix] = (AnSv); \ 9480 } STMT_END 9481 9482/* 9483=for apidoc sv_mortalcopy 9484 9485Creates a new SV which is a copy of the original SV (using C<sv_setsv>). 9486The new SV is marked as mortal. It will be destroyed "soon", either by an 9487explicit call to C<FREETMPS>, or by an implicit call at places such as 9488statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>. 9489 9490=for apidoc sv_mortalcopy_flags 9491 9492Like C<sv_mortalcopy>, but the extra C<flags> are passed to the 9493C<sv_setsv_flags>. 9494 9495=cut 9496*/ 9497 9498/* Make a string that will exist for the duration of the expression 9499 * evaluation. Actually, it may have to last longer than that, but 9500 * hopefully we won't free it until it has been assigned to a 9501 * permanent location. */ 9502 9503SV * 9504Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags) 9505{ 9506 SV *sv; 9507 9508 if (flags & SV_GMAGIC) 9509 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */ 9510 new_SV(sv); 9511 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC); 9512 PUSH_EXTEND_MORTAL__SV_C(sv); 9513 SvTEMP_on(sv); 9514 return sv; 9515} 9516 9517/* 9518=for apidoc sv_newmortal 9519 9520Creates a new null SV which is mortal. The reference count of the SV is 9521set to 1. It will be destroyed "soon", either by an explicit call to 9522C<FREETMPS>, or by an implicit call at places such as statement boundaries. 9523See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>. 9524 9525=cut 9526*/ 9527 9528SV * 9529Perl_sv_newmortal(pTHX) 9530{ 9531 SV *sv; 9532 9533 new_SV(sv); 9534 SvFLAGS(sv) = SVs_TEMP; 9535 PUSH_EXTEND_MORTAL__SV_C(sv); 9536 return sv; 9537} 9538 9539 9540/* 9541=for apidoc newSVpvn_flags 9542 9543Creates a new SV and copies a string (which may contain C<NUL> (C<\0>) 9544characters) into it. The reference count for the 9545SV is set to 1. Note that if C<len> is zero, Perl will create a zero length 9546string. You are responsible for ensuring that the source string is at least 9547C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined. 9548Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>. 9549If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before 9550returning. If C<SVf_UTF8> is set, C<s> 9551is considered to be in UTF-8 and the 9552C<SVf_UTF8> flag will be set on the new SV. 9553C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as 9554 9555 #define newSVpvn_utf8(s, len, u) \ 9556 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0) 9557 9558=for apidoc Amnh||SVs_TEMP 9559 9560=cut 9561*/ 9562 9563SV * 9564Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags) 9565{ 9566 SV *sv; 9567 9568 /* All the flags we don't support must be zero. 9569 And we're new code so I'm going to assert this from the start. */ 9570 assert(!(flags & ~(SVf_UTF8|SVs_TEMP))); 9571 sv = newSV_type(SVt_PV); 9572 sv_setpvn_fresh(sv,s,len); 9573 9574 /* This code used to do a sv_2mortal(), however we now unroll the call to 9575 * sv_2mortal() and do what it does ourselves here. Since we have asserted 9576 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we 9577 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which 9578 * in turn means we don't need to mask out the SVf_UTF8 flag below, which 9579 * means that we eliminate quite a few steps than it looks - Yves 9580 * (explaining patch by gfx) */ 9581 9582 SvFLAGS(sv) |= flags; 9583 9584 if(flags & SVs_TEMP){ 9585 PUSH_EXTEND_MORTAL__SV_C(sv); 9586 } 9587 9588 return sv; 9589} 9590 9591/* 9592=for apidoc sv_2mortal 9593 9594Marks an existing SV as mortal. The SV will be destroyed "soon", either 9595by an explicit call to C<FREETMPS>, or by an implicit call at places such as 9596statement boundaries. C<SvTEMP()> is turned on which means that the SV's 9597string buffer can be "stolen" if this SV is copied. See also 9598C<L</sv_newmortal>> and C<L</sv_mortalcopy>>. 9599 9600=cut 9601*/ 9602 9603SV * 9604Perl_sv_2mortal(pTHX_ SV *const sv) 9605{ 9606 if (!sv) 9607 return sv; 9608 if (SvIMMORTAL(sv)) 9609 return sv; 9610 PUSH_EXTEND_MORTAL__SV_C(sv); 9611 SvTEMP_on(sv); 9612 return sv; 9613} 9614 9615/* 9616=for apidoc newSVpv 9617 9618Creates a new SV and copies a string (which may contain C<NUL> (C<\0>) 9619characters) into it. The reference count for the 9620SV is set to 1. If C<len> is zero, Perl will compute the length using 9621C<strlen()>, (which means if you use this option, that C<s> can't have embedded 9622C<NUL> characters and has to have a terminating C<NUL> byte). 9623 9624This function can cause reliability issues if you are likely to pass in 9625empty strings that are not null terminated, because it will run 9626strlen on the string and potentially run past valid memory. 9627 9628Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings. 9629For string literals use L</newSVpvs> instead. This function will work fine for 9630C<NUL> terminated strings, but if you want to avoid the if statement on whether 9631to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself). 9632 9633=cut 9634*/ 9635 9636SV * 9637Perl_newSVpv(pTHX_ const char *const s, const STRLEN len) 9638{ 9639 SV *sv = newSV_type(SVt_PV); 9640 sv_setpvn_fresh(sv, s, len || s == NULL ? len : strlen(s)); 9641 return sv; 9642} 9643 9644/* 9645=for apidoc newSVpvn 9646 9647Creates a new SV and copies a string into it, which may contain C<NUL> characters 9648(C<\0>) and other binary data. The reference count for the SV is set to 1. 9649Note that if C<len> is zero, Perl will create a zero length (Perl) string. You 9650are responsible for ensuring that the source buffer is at least 9651C<len> bytes long. If the C<buffer> argument is NULL the new SV will be 9652undefined. 9653 9654=cut 9655*/ 9656 9657SV * 9658Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len) 9659{ 9660 SV *sv = newSV_type(SVt_PV); 9661 sv_setpvn_fresh(sv,buffer,len); 9662 return sv; 9663} 9664 9665/* 9666=for apidoc newSVhek_mortal 9667 9668Creates a new mortal SV from the hash key structure. It will generate 9669scalars that point to the shared string table where possible. Returns 9670a new (undefined) SV if C<hek> is NULL. 9671 9672This is more efficient than using sv_2mortal(newSVhek( ... )) 9673 9674=cut 9675*/ 9676 9677SV * 9678Perl_newSVhek_mortal(pTHX_ const HEK *const hek) 9679{ 9680 SV * const sv = newSVhek(hek); 9681 assert(sv); 9682 assert(!SvIMMORTAL(sv)); 9683 9684 PUSH_EXTEND_MORTAL__SV_C(sv); 9685 SvTEMP_on(sv); 9686 return sv; 9687} 9688 9689/* 9690=for apidoc newSVhek 9691 9692Creates a new SV from the hash key structure. It will generate scalars that 9693point to the shared string table where possible. Returns a new (undefined) 9694SV if C<hek> is NULL. 9695 9696=cut 9697*/ 9698 9699SV * 9700Perl_newSVhek(pTHX_ const HEK *const hek) 9701{ 9702 if (!hek) { 9703 SV *sv; 9704 9705 new_SV(sv); 9706 return sv; 9707 } 9708 9709 if (HEK_LEN(hek) == HEf_SVKEY) { 9710 return newSVsv(*(SV**)HEK_KEY(hek)); 9711 } else { 9712 const int flags = HEK_FLAGS(hek); 9713 if (flags & HVhek_WASUTF8) { 9714 /* Trouble :-) 9715 Andreas would like keys he put in as utf8 to come back as utf8 9716 */ 9717 STRLEN utf8_len = HEK_LEN(hek); 9718 SV * const sv = newSV_type(SVt_PV); 9719 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len); 9720 /* bytes_to_utf8() allocates a new string, which we can repurpose: */ 9721 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL); 9722 SvUTF8_on (sv); 9723 return sv; 9724 } else if (flags & HVhek_NOTSHARED) { 9725 /* A hash that isn't using shared hash keys has to have 9726 the flag in every key so that we know not to try to call 9727 share_hek_hek on it. */ 9728 9729 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek)); 9730 if (HEK_UTF8(hek)) 9731 SvUTF8_on (sv); 9732 return sv; 9733 } 9734 /* This will be overwhelmingly the most common case. */ 9735 { 9736 /* Inline most of newSVpvn_share(), because share_hek_hek() is far 9737 more efficient than sharepvn(). */ 9738 SV *sv = newSV_type(SVt_PV); 9739 9740 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek))); 9741 SvCUR_set(sv, HEK_LEN(hek)); 9742 SvLEN_set(sv, 0); 9743 SvIsCOW_on(sv); 9744 SvPOK_on(sv); 9745 if (HEK_UTF8(hek)) 9746 SvUTF8_on(sv); 9747 return sv; 9748 } 9749 } 9750} 9751 9752/* 9753=for apidoc newSVpvn_share 9754 9755Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string 9756table. If the string does not already exist in the table, it is 9757created first. Turns on the C<SvIsCOW> flag (or C<READONLY> 9758and C<FAKE> in 5.16 and earlier). If the C<hash> parameter 9759is non-zero, that value is used; otherwise the hash is computed. 9760The string's hash can later be retrieved from the SV 9761with the C<L</SvSHARED_HASH>> macro. The idea here is 9762that as the string table is used for shared hash keys these strings will have 9763C<SvPVX_const == HeKEY> and hash lookup will avoid string compare. 9764 9765=cut 9766*/ 9767 9768SV * 9769Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash) 9770{ 9771 SV *sv; 9772 bool is_utf8 = FALSE; 9773 const char *const orig_src = src; 9774 9775 if (len < 0) { 9776 STRLEN tmplen = -len; 9777 is_utf8 = TRUE; 9778 /* See the note in hv.c:hv_fetch() --jhi */ 9779 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8); 9780 len = tmplen; 9781 } 9782 if (!hash) 9783 PERL_HASH(hash, src, len); 9784 sv = newSV_type(SVt_PV); 9785 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it 9786 changes here, update it there too. */ 9787 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash)); 9788 SvCUR_set(sv, len); 9789 SvLEN_set(sv, 0); 9790 SvIsCOW_on(sv); 9791 SvPOK_on(sv); 9792 if (is_utf8) 9793 SvUTF8_on(sv); 9794 if (src != orig_src) 9795 Safefree(src); 9796 return sv; 9797} 9798 9799/* 9800=for apidoc newSVpv_share 9801 9802Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a 9803string/length pair. 9804 9805=cut 9806*/ 9807 9808SV * 9809Perl_newSVpv_share(pTHX_ const char *src, U32 hash) 9810{ 9811 return newSVpvn_share(src, strlen(src), hash); 9812} 9813 9814#if defined(MULTIPLICITY) 9815 9816/* pTHX_ magic can't cope with varargs, so this is a no-context 9817 * version of the main function, (which may itself be aliased to us). 9818 * Don't access this version directly. 9819 */ 9820 9821SV * 9822Perl_newSVpvf_nocontext(const char *const pat, ...) 9823{ 9824 dTHX; 9825 SV *sv; 9826 va_list args; 9827 9828 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT; 9829 9830 va_start(args, pat); 9831 sv = vnewSVpvf(pat, &args); 9832 va_end(args); 9833 return sv; 9834} 9835#endif 9836 9837/* 9838=for apidoc newSVpvf 9839 9840Creates a new SV and initializes it with the string formatted like 9841C<sv_catpvf>. 9842 9843=for apidoc newSVpvf_nocontext 9844Like C<L</newSVpvf>> but does not take a thread context (C<aTHX>) parameter, 9845so is used in situations where the caller doesn't already have the thread 9846context. 9847 9848=for apidoc vnewSVpvf 9849Like C<L</newSVpvf>> but the arguments are an encapsulated argument list. 9850 9851=cut 9852*/ 9853 9854SV * 9855Perl_newSVpvf(pTHX_ const char *const pat, ...) 9856{ 9857 SV *sv; 9858 va_list args; 9859 9860 PERL_ARGS_ASSERT_NEWSVPVF; 9861 9862 va_start(args, pat); 9863 sv = vnewSVpvf(pat, &args); 9864 va_end(args); 9865 return sv; 9866} 9867 9868/* backend for newSVpvf() and newSVpvf_nocontext() */ 9869 9870SV * 9871Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args) 9872{ 9873 SV *sv; 9874 9875 PERL_ARGS_ASSERT_VNEWSVPVF; 9876 9877 sv = newSV(1); 9878 SvPVCLEAR_FRESH(sv); 9879 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, 0); 9880 return sv; 9881} 9882 9883/* 9884=for apidoc newSVnv 9885 9886Creates a new SV and copies a floating point value into it. 9887The reference count for the SV is set to 1. 9888 9889=cut 9890*/ 9891 9892SV * 9893Perl_newSVnv(pTHX_ const NV n) 9894{ 9895 SV *sv = newSV_type(SVt_NV); 9896 (void)SvNOK_on(sv); 9897 9898 SvNV_set(sv, n); 9899 SvTAINT(sv); 9900 9901 return sv; 9902} 9903 9904/* 9905=for apidoc newSViv 9906 9907Creates a new SV and copies an integer into it. The reference count for the 9908SV is set to 1. 9909 9910=cut 9911*/ 9912 9913SV * 9914Perl_newSViv(pTHX_ const IV i) 9915{ 9916 SV *sv = newSV_type(SVt_IV); 9917 (void)SvIOK_on(sv); 9918 9919 SvIV_set(sv, i); 9920 SvTAINT(sv); 9921 9922 return sv; 9923} 9924 9925/* 9926=for apidoc newSVuv 9927 9928Creates a new SV and copies an unsigned integer into it. 9929The reference count for the SV is set to 1. 9930 9931=cut 9932*/ 9933 9934SV * 9935Perl_newSVuv(pTHX_ const UV u) 9936{ 9937 SV *sv; 9938 9939 /* Inlining ONLY the small relevant subset of sv_setuv here 9940 * for performance. Makes a significant difference. */ 9941 9942 /* Using ivs is more efficient than using uvs - see sv_setuv */ 9943 if (u <= (UV)IV_MAX) { 9944 return newSViv((IV)u); 9945 } 9946 9947 new_SV(sv); 9948 9949 /* We're starting from SVt_FIRST, so provided that's 9950 * actual 0, we don't have to unset any SV type flags 9951 * to promote to SVt_IV. */ 9952 STATIC_ASSERT_STMT(SVt_FIRST == 0); 9953 9954 SET_SVANY_FOR_BODYLESS_IV(sv); 9955 SvFLAGS(sv) |= SVt_IV; 9956 (void)SvIOK_on(sv); 9957 (void)SvIsUV_on(sv); 9958 9959 SvUV_set(sv, u); 9960 SvTAINT(sv); 9961 9962 return sv; 9963} 9964 9965/* 9966=for apidoc newSVbool 9967 9968Creates a new SV boolean. 9969 9970=cut 9971*/ 9972 9973SV * 9974Perl_newSVbool(pTHX_ bool bool_val) 9975{ 9976 PERL_ARGS_ASSERT_NEWSVBOOL; 9977 SV *sv = newSVsv(bool_val ? &PL_sv_yes : &PL_sv_no); 9978 9979 return sv; 9980} 9981 9982/* 9983=for apidoc newSV_true 9984 9985Creates a new SV that is a boolean true. 9986 9987=cut 9988*/ 9989SV * 9990Perl_newSV_true(pTHX) 9991{ 9992 PERL_ARGS_ASSERT_NEWSV_TRUE; 9993 SV *sv = newSVsv(&PL_sv_yes); 9994 9995 return sv; 9996} 9997 9998/* 9999=for apidoc newSV_false 10000 10001Creates a new SV that is a boolean false. 10002 10003=cut 10004*/ 10005 10006SV * 10007Perl_newSV_false(pTHX) 10008{ 10009 PERL_ARGS_ASSERT_NEWSV_FALSE; 10010 SV *sv = newSVsv(&PL_sv_no); 10011 10012 return sv; 10013} 10014 10015/* newRV_inc is the official function name to use now. 10016 * newRV_inc is in fact #defined to newRV in sv.h 10017 */ 10018 10019SV * 10020Perl_newRV(pTHX_ SV *const sv) 10021{ 10022 PERL_ARGS_ASSERT_NEWRV; 10023 10024 return newRV_noinc(SvREFCNT_inc_simple_NN(sv)); 10025} 10026 10027/* 10028=for apidoc newSVsv 10029=for apidoc_item newSVsv_flags 10030=for apidoc_item newSVsv_nomg 10031 10032These create a new SV which is an exact duplicate of the original SV 10033(using C<sv_setsv>.) 10034 10035They differ only in that C<newSVsv> performs 'get' magic; C<newSVsv_nomg> skips 10036any magic; and C<newSVsv_flags> allows you to explicitly set a C<flags> 10037parameter. 10038 10039=cut 10040*/ 10041 10042SV * 10043Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags) 10044{ 10045 SV *sv; 10046 10047 if (!old) 10048 return NULL; 10049 if (SvIS_FREED(old)) { 10050 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string"); 10051 return NULL; 10052 } 10053 /* Do this here, otherwise we leak the new SV if this croaks. */ 10054 if (flags & SV_GMAGIC) 10055 SvGETMAGIC(old); 10056 new_SV(sv); 10057 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC); 10058 return sv; 10059} 10060 10061/* 10062=for apidoc sv_reset 10063 10064Underlying implementation for the C<reset> Perl function. 10065Note that the perl-level function is vaguely deprecated. 10066 10067=cut 10068*/ 10069 10070void 10071Perl_sv_reset(pTHX_ const char *s, HV *const stash) 10072{ 10073 PERL_ARGS_ASSERT_SV_RESET; 10074 10075 sv_resetpvn(*s ? s : NULL, strlen(s), stash); 10076} 10077 10078void 10079Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash) 10080{ 10081 char todo[PERL_UCHAR_MAX+1]; 10082 const char *send; 10083 10084 if (!stash || SvTYPE(stash) != SVt_PVHV) 10085 return; 10086 10087 if (!s) { /* reset ?? searches */ 10088 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab); 10089 if (mg && mg->mg_len) { 10090 const U32 count = mg->mg_len / sizeof(PMOP**); 10091 PMOP **pmp = (PMOP**) mg->mg_ptr; 10092 PMOP *const *const end = pmp + count; 10093 10094 while (pmp < end) { 10095#ifdef USE_ITHREADS 10096 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]); 10097#else 10098 (*pmp)->op_pmflags &= ~PMf_USED; 10099#endif 10100 ++pmp; 10101 } 10102 } 10103 return; 10104 } 10105 10106 /* reset variables */ 10107 10108 if (!HvTOTALKEYS(stash)) 10109 return; 10110 10111 Zero(todo, 256, char); 10112 send = s + len; 10113 while (s < send) { 10114 I32 max; 10115 I32 i = (unsigned char)*s; 10116 if (s[1] == '-') { 10117 s += 2; 10118 } 10119 max = (unsigned char)*s++; 10120 for ( ; i <= max; i++) { 10121 todo[i] = 1; 10122 } 10123 for (i = 0; i <= (I32) HvMAX(stash); i++) { 10124 HE *entry; 10125 for (entry = HvARRAY(stash)[i]; 10126 entry; 10127 entry = HeNEXT(entry)) 10128 { 10129 GV *gv; 10130 SV *sv; 10131 10132 if (!todo[(U8)*HeKEY(entry)]) 10133 continue; 10134 gv = MUTABLE_GV(HeVAL(entry)); 10135 if (!isGV(gv)) 10136 continue; 10137 sv = GvSV(gv); 10138 if (sv && !SvREADONLY(sv)) { 10139 SV_CHECK_THINKFIRST_COW_DROP(sv); 10140 if (!isGV(sv)) SvOK_off(sv); 10141 } 10142 if (GvAV(gv)) { 10143 av_clear(GvAV(gv)); 10144 } 10145 if (GvHV(gv) && !HvHasNAME(GvHV(gv))) { 10146 hv_clear(GvHV(gv)); 10147 } 10148 } 10149 } 10150 } 10151} 10152 10153/* 10154=for apidoc sv_2io 10155 10156Using various gambits, try to get an IO from an SV: the IO slot if its a 10157GV; or the recursive result if we're an RV; or the IO slot of the symbol 10158named after the PV if we're a string. 10159 10160'Get' magic is ignored on the C<sv> passed in, but will be called on 10161C<SvRV(sv)> if C<sv> is an RV. 10162 10163=cut 10164*/ 10165 10166IO* 10167Perl_sv_2io(pTHX_ SV *const sv) 10168{ 10169 IO* io; 10170 GV* gv; 10171 10172 PERL_ARGS_ASSERT_SV_2IO; 10173 10174 switch (SvTYPE(sv)) { 10175 case SVt_PVIO: 10176 io = MUTABLE_IO(sv); 10177 break; 10178 case SVt_PVGV: 10179 case SVt_PVLV: 10180 if (isGV_with_GP(sv)) { 10181 gv = MUTABLE_GV(sv); 10182 io = GvIO(gv); 10183 if (!io) 10184 Perl_croak(aTHX_ "Bad filehandle: %" HEKf, 10185 HEKfARG(GvNAME_HEK(gv))); 10186 break; 10187 } 10188 /* FALLTHROUGH */ 10189 default: 10190 if (!SvOK(sv)) 10191 Perl_croak(aTHX_ PL_no_usym, "filehandle"); 10192 if (SvROK(sv)) { 10193 SvGETMAGIC(SvRV(sv)); 10194 return sv_2io(SvRV(sv)); 10195 } 10196 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO); 10197 if (gv) 10198 io = GvIO(gv); 10199 else 10200 io = 0; 10201 if (!io) { 10202 SV *newsv = sv; 10203 if (SvGMAGICAL(sv)) { 10204 newsv = sv_newmortal(); 10205 sv_setsv_nomg(newsv, sv); 10206 } 10207 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv)); 10208 } 10209 break; 10210 } 10211 return io; 10212} 10213 10214/* 10215=for apidoc sv_2cv 10216 10217Using various gambits, try to get a CV from an SV; in addition, try if 10218possible to set C<*st> and C<*gvp> to the stash and GV associated with it. 10219The flags in C<lref> are passed to C<gv_fetchsv>. 10220 10221=cut 10222*/ 10223 10224CV * 10225Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref) 10226{ 10227 GV *gv = NULL; 10228 CV *cv = NULL; 10229 10230 PERL_ARGS_ASSERT_SV_2CV; 10231 10232 if (!sv) { 10233 *st = NULL; 10234 *gvp = NULL; 10235 return NULL; 10236 } 10237 switch (SvTYPE(sv)) { 10238 case SVt_PVCV: 10239 *st = CvSTASH(sv); 10240 *gvp = NULL; 10241 return MUTABLE_CV(sv); 10242 case SVt_PVHV: 10243 case SVt_PVAV: 10244 *st = NULL; 10245 *gvp = NULL; 10246 return NULL; 10247 default: 10248 SvGETMAGIC(sv); 10249 if (SvROK(sv)) { 10250 if (SvAMAGIC(sv)) 10251 sv = amagic_deref_call(sv, to_cv_amg); 10252 10253 sv = SvRV(sv); 10254 if (SvTYPE(sv) == SVt_PVCV) { 10255 cv = MUTABLE_CV(sv); 10256 *gvp = NULL; 10257 *st = CvSTASH(cv); 10258 return cv; 10259 } 10260 else if(SvGETMAGIC(sv), isGV_with_GP(sv)) 10261 gv = MUTABLE_GV(sv); 10262 else 10263 Perl_croak(aTHX_ "Not a subroutine reference"); 10264 } 10265 else if (isGV_with_GP(sv)) { 10266 gv = MUTABLE_GV(sv); 10267 } 10268 else { 10269 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV); 10270 } 10271 *gvp = gv; 10272 if (!gv) { 10273 *st = NULL; 10274 return NULL; 10275 } 10276 /* Some flags to gv_fetchsv mean don't really create the GV */ 10277 if (!isGV_with_GP(gv)) { 10278 *st = NULL; 10279 return NULL; 10280 } 10281 *st = GvESTASH(gv); 10282 if (lref & ~GV_ADDMG && !GvCVu(gv)) { 10283 /* XXX this is probably not what they think they're getting. 10284 * It has the same effect as "sub name;", i.e. just a forward 10285 * declaration! */ 10286 newSTUB(gv,0); 10287 } 10288 return GvCVu(gv); 10289 } 10290} 10291 10292/* 10293=for apidoc sv_true 10294 10295Returns true if the SV has a true value by Perl's rules. 10296Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may 10297instead use an in-line version. 10298 10299=cut 10300*/ 10301 10302I32 10303Perl_sv_true(pTHX_ SV *const sv) 10304{ 10305 if (!sv) 10306 return 0; 10307 if (SvPOK(sv)) { 10308 const XPV* const tXpv = (XPV*)SvANY(sv); 10309 if (tXpv && 10310 (tXpv->xpv_cur > 1 || 10311 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0'))) 10312 return 1; 10313 else 10314 return 0; 10315 } 10316 else { 10317 if (SvIOK(sv)) 10318 return SvIVX(sv) != 0; 10319 else { 10320 if (SvNOK(sv)) 10321 return SvNVX(sv) != 0.0; 10322 else 10323 return sv_2bool(sv); 10324 } 10325 } 10326} 10327 10328/* 10329=for apidoc sv_pvn_force 10330 10331Get a sensible string out of the SV somehow. 10332A private implementation of the C<SvPV_force> macro for compilers which 10333can't cope with complex macro expressions. Always use the macro instead. 10334 10335=for apidoc sv_pvn_force_flags 10336 10337Get a sensible string out of the SV somehow. 10338If C<flags> has the C<SV_GMAGIC> bit set, will C<L</mg_get>> on C<sv> if 10339appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are 10340implemented in terms of this function. 10341You normally want to use the various wrapper macros instead: see 10342C<L</SvPV_force>> and C<L</SvPV_force_nomg>>. 10343 10344=cut 10345*/ 10346 10347char * 10348Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const U32 flags) 10349{ 10350 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS; 10351 10352 if (flags & SV_GMAGIC) SvGETMAGIC(sv); 10353 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv))) 10354 sv_force_normal_flags(sv, 0); 10355 10356 if (SvPOK(sv)) { 10357 if (lp) 10358 *lp = SvCUR(sv); 10359 } 10360 else { 10361 char *s; 10362 STRLEN len; 10363 10364 if (SvTYPE(sv) > SVt_PVLV 10365 || isGV_with_GP(sv)) 10366 /* diag_listed_as: Can't coerce %s to %s in %s */ 10367 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0), 10368 OP_DESC(PL_op)); 10369 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC); 10370 if (!s) { 10371 s = (char *)""; 10372 } 10373 if (lp) 10374 *lp = len; 10375 10376 if (SvTYPE(sv) < SVt_PV || 10377 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */ 10378 if (SvROK(sv)) 10379 sv_unref(sv); 10380 SvUPGRADE(sv, SVt_PV); /* Never FALSE */ 10381 SvGROW(sv, len + 1); 10382 Move(s,SvPVX(sv),len,char); 10383 SvCUR_set(sv, len); 10384 SvPVX(sv)[len] = '\0'; 10385 } 10386 if (!SvPOK(sv)) { 10387 SvPOK_on(sv); /* validate pointer */ 10388 SvTAINT(sv); 10389 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n", 10390 PTR2UV(sv),SvPVX_const(sv))); 10391 } 10392 } 10393 (void)SvPOK_only_UTF8(sv); 10394 return SvPVX_mutable(sv); 10395} 10396 10397/* 10398=for apidoc sv_pvbyten_force 10399 10400The backend for the C<SvPVbytex_force> macro. Always use the macro 10401instead. If the SV cannot be downgraded from UTF-8, this croaks. 10402 10403=cut 10404*/ 10405 10406char * 10407Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp) 10408{ 10409 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE; 10410 10411 sv_pvn_force(sv,lp); 10412 (void)sv_utf8_downgrade(sv,0); 10413 *lp = SvCUR(sv); 10414 return SvPVX(sv); 10415} 10416 10417/* 10418=for apidoc sv_pvutf8n_force 10419 10420The backend for the C<SvPVutf8x_force> macro. Always use the macro 10421instead. 10422 10423=cut 10424*/ 10425 10426char * 10427Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp) 10428{ 10429 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE; 10430 10431 sv_pvn_force(sv,0); 10432 sv_utf8_upgrade_nomg(sv); 10433 *lp = SvCUR(sv); 10434 return SvPVX(sv); 10435} 10436 10437/* 10438=for apidoc sv_reftype 10439 10440Returns a string describing what the SV is a reference to. 10441 10442If ob is true and the SV is blessed, the string is the class name, 10443otherwise it is the type of the SV, "SCALAR", "ARRAY" etc. 10444 10445=cut 10446*/ 10447 10448const char * 10449Perl_sv_reftype(pTHX_ const SV *const sv, const int ob) 10450{ 10451 PERL_ARGS_ASSERT_SV_REFTYPE; 10452 if (ob && SvOBJECT(sv)) { 10453 return SvPV_nolen_const(sv_ref(NULL, sv, ob)); 10454 } 10455 else { 10456 /* WARNING - There is code, for instance in mg.c, that assumes that 10457 * the only reason that sv_reftype(sv,0) would return a string starting 10458 * with 'L' or 'S' is that it is a LVALUE or a SCALAR. 10459 * Yes this a dodgy way to do type checking, but it saves practically reimplementing 10460 * this routine inside other subs, and it saves time. 10461 * Do not change this assumption without searching for "dodgy type check" in 10462 * the code. 10463 * - Yves */ 10464 switch (SvTYPE(sv)) { 10465 case SVt_NULL: 10466 case SVt_IV: 10467 case SVt_NV: 10468 case SVt_PV: 10469 case SVt_PVIV: 10470 case SVt_PVNV: 10471 case SVt_PVMG: 10472 if (SvVOK(sv)) 10473 return "VSTRING"; 10474 if (SvROK(sv)) 10475 return "REF"; 10476 else 10477 return "SCALAR"; 10478 10479 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF" 10480 /* tied lvalues should appear to be 10481 * scalars for backwards compatibility */ 10482 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't')) 10483 ? "SCALAR" : "LVALUE"); 10484 case SVt_PVAV: return "ARRAY"; 10485 case SVt_PVHV: return "HASH"; 10486 case SVt_PVCV: return "CODE"; 10487 case SVt_PVGV: return (char *) (isGV_with_GP(sv) 10488 ? "GLOB" : "SCALAR"); 10489 case SVt_PVFM: return "FORMAT"; 10490 case SVt_PVIO: return "IO"; 10491 case SVt_INVLIST: return "INVLIST"; 10492 case SVt_REGEXP: return "REGEXP"; 10493 case SVt_PVOBJ: return "OBJECT"; 10494 default: return "UNKNOWN"; 10495 } 10496 } 10497} 10498 10499/* 10500=for apidoc sv_ref 10501 10502Returns a SV describing what the SV passed in is a reference to. 10503 10504dst can be a SV to be set to the description or NULL, in which case a 10505mortal SV is returned. 10506 10507If ob is true and the SV is blessed, the description is the class 10508name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc. 10509 10510=cut 10511*/ 10512 10513SV * 10514Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob) 10515{ 10516 PERL_ARGS_ASSERT_SV_REF; 10517 10518 if (!dst) 10519 dst = sv_newmortal(); 10520 10521 if (ob && SvOBJECT(sv)) { 10522 if (HvHasNAME(SvSTASH(sv))) 10523 sv_sethek(dst, HvNAME_HEK(SvSTASH(sv))); 10524 else 10525 sv_setpvs(dst, "__ANON__"); 10526 } 10527 else { 10528 const char * reftype = sv_reftype(sv, 0); 10529 sv_setpv(dst, reftype); 10530 } 10531 return dst; 10532} 10533 10534/* 10535=for apidoc sv_isobject 10536 10537Returns a boolean indicating whether the SV is an RV pointing to a blessed 10538object. If the SV is not an RV, or if the object is not blessed, then this 10539will return false. 10540 10541=cut 10542*/ 10543 10544int 10545Perl_sv_isobject(pTHX_ SV *sv) 10546{ 10547 if (!sv) 10548 return 0; 10549 SvGETMAGIC(sv); 10550 if (!SvROK(sv)) 10551 return 0; 10552 sv = SvRV(sv); 10553 if (!SvOBJECT(sv)) 10554 return 0; 10555 return 1; 10556} 10557 10558/* 10559=for apidoc sv_isa 10560 10561Returns a boolean indicating whether the SV is blessed into the specified 10562class. 10563 10564This does not check for subtypes or method overloading. Use C<sv_isa_sv> to 10565verify an inheritance relationship in the same way as the C<isa> operator by 10566respecting any C<isa()> method overloading; or C<sv_derived_from_sv> to test 10567directly on the actual object type. 10568 10569=cut 10570*/ 10571 10572int 10573Perl_sv_isa(pTHX_ SV *sv, const char *const name) 10574{ 10575 const char *hvname; 10576 10577 PERL_ARGS_ASSERT_SV_ISA; 10578 10579 if (!sv) 10580 return 0; 10581 SvGETMAGIC(sv); 10582 if (!SvROK(sv)) 10583 return 0; 10584 sv = SvRV(sv); 10585 if (!SvOBJECT(sv)) 10586 return 0; 10587 hvname = HvNAME_get(SvSTASH(sv)); 10588 if (!hvname) 10589 return 0; 10590 10591 return strEQ(hvname, name); 10592} 10593 10594/* 10595=for apidoc newSVrv 10596 10597Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an 10598RV then it will be upgraded to one. If C<classname> is non-null then the new 10599SV will be blessed in the specified package. The new SV is returned and its 10600reference count is 1. The reference count 1 is owned by C<rv>. See also 10601newRV_inc() and newRV_noinc() for creating a new RV properly. 10602 10603=cut 10604*/ 10605 10606SV* 10607Perl_newSVrv(pTHX_ SV *const rv, const char *const classname) 10608{ 10609 SV *sv; 10610 10611 PERL_ARGS_ASSERT_NEWSVRV; 10612 10613 new_SV(sv); 10614 10615 SV_CHECK_THINKFIRST_COW_DROP(rv); 10616 10617 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) { 10618 const U32 refcnt = SvREFCNT(rv); 10619 SvREFCNT(rv) = 0; 10620 sv_clear(rv); 10621 SvFLAGS(rv) = 0; 10622 SvREFCNT(rv) = refcnt; 10623 10624 sv_upgrade(rv, SVt_IV); 10625 } else if (SvROK(rv)) { 10626 SvREFCNT_dec(SvRV(rv)); 10627 } else { 10628 prepare_SV_for_RV(rv); 10629 } 10630 10631 SvOK_off(rv); 10632 SvRV_set(rv, sv); 10633 SvROK_on(rv); 10634 10635 if (classname) { 10636 HV* const stash = gv_stashpv(classname, GV_ADD); 10637 (void)sv_bless(rv, stash); 10638 } 10639 return sv; 10640} 10641 10642SV * 10643Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible) 10644{ 10645 SV * const lv = newSV_type(SVt_PVLV); 10646 PERL_ARGS_ASSERT_NEWSVAVDEFELEM; 10647 LvTYPE(lv) = 'y'; 10648 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0); 10649 LvTARG(lv) = SvREFCNT_inc_simple_NN(av); 10650 LvSTARGOFF(lv) = ix; 10651 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX; 10652 return lv; 10653} 10654 10655/* 10656=for apidoc sv_setref_pv 10657 10658Copies a pointer into a new SV, optionally blessing the SV. The C<rv> 10659argument will be upgraded to an RV. That RV will be modified to point to 10660the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed 10661into the SV. The C<classname> argument indicates the package for the 10662blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV 10663will have a reference count of 1, and the RV will be returned. 10664 10665Do not use with other Perl types such as HV, AV, SV, CV, because those 10666objects will become corrupted by the pointer copy process. 10667 10668Note that C<sv_setref_pvn> copies the string while this copies the pointer. 10669 10670=cut 10671*/ 10672 10673SV* 10674Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv) 10675{ 10676 PERL_ARGS_ASSERT_SV_SETREF_PV; 10677 10678 if (!pv) { 10679 sv_set_undef(rv); 10680 SvSETMAGIC(rv); 10681 } 10682 else 10683 sv_setiv(newSVrv(rv,classname), PTR2IV(pv)); 10684 return rv; 10685} 10686 10687/* 10688=for apidoc sv_setref_iv 10689 10690Copies an integer into a new SV, optionally blessing the SV. The C<rv> 10691argument will be upgraded to an RV. That RV will be modified to point to 10692the new SV. The C<classname> argument indicates the package for the 10693blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV 10694will have a reference count of 1, and the RV will be returned. 10695 10696=cut 10697*/ 10698 10699SV* 10700Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv) 10701{ 10702 PERL_ARGS_ASSERT_SV_SETREF_IV; 10703 10704 sv_setiv(newSVrv(rv,classname), iv); 10705 return rv; 10706} 10707 10708/* 10709=for apidoc sv_setref_uv 10710 10711Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv> 10712argument will be upgraded to an RV. That RV will be modified to point to 10713the new SV. The C<classname> argument indicates the package for the 10714blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV 10715will have a reference count of 1, and the RV will be returned. 10716 10717=cut 10718*/ 10719 10720SV* 10721Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv) 10722{ 10723 PERL_ARGS_ASSERT_SV_SETREF_UV; 10724 10725 sv_setuv(newSVrv(rv,classname), uv); 10726 return rv; 10727} 10728 10729/* 10730=for apidoc sv_setref_nv 10731 10732Copies a double into a new SV, optionally blessing the SV. The C<rv> 10733argument will be upgraded to an RV. That RV will be modified to point to 10734the new SV. The C<classname> argument indicates the package for the 10735blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV 10736will have a reference count of 1, and the RV will be returned. 10737 10738=cut 10739*/ 10740 10741SV* 10742Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv) 10743{ 10744 PERL_ARGS_ASSERT_SV_SETREF_NV; 10745 10746 sv_setnv(newSVrv(rv,classname), nv); 10747 return rv; 10748} 10749 10750/* 10751=for apidoc sv_setref_pvn 10752 10753Copies a string into a new SV, optionally blessing the SV. The length of the 10754string must be specified with C<n>. The C<rv> argument will be upgraded to 10755an RV. That RV will be modified to point to the new SV. The C<classname> 10756argument indicates the package for the blessing. Set C<classname> to 10757C<NULL> to avoid the blessing. The new SV will have a reference count 10758of 1, and the RV will be returned. 10759 10760Note that C<sv_setref_pv> copies the pointer while this copies the string. 10761 10762=cut 10763*/ 10764 10765SV* 10766Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname, 10767 const char *const pv, const STRLEN n) 10768{ 10769 PERL_ARGS_ASSERT_SV_SETREF_PVN; 10770 10771 sv_setpvn(newSVrv(rv,classname), pv, n); 10772 return rv; 10773} 10774 10775/* 10776=for apidoc sv_bless 10777 10778Blesses an SV into a specified package. The SV must be an RV. The package 10779must be designated by its stash (see C<L</gv_stashpv>>). The reference count 10780of the SV is unaffected. 10781 10782=cut 10783*/ 10784 10785SV* 10786Perl_sv_bless(pTHX_ SV *const sv, HV *const stash) 10787{ 10788 SV *tmpRef; 10789 HV *oldstash = NULL; 10790 10791 PERL_ARGS_ASSERT_SV_BLESS; 10792 10793 SvGETMAGIC(sv); 10794 if (!SvROK(sv)) 10795 Perl_croak(aTHX_ "Can't bless non-reference value"); 10796 if (HvSTASH_IS_CLASS(stash)) 10797 Perl_croak(aTHX_ "Attempt to bless into a class"); 10798 10799 tmpRef = SvRV(sv); 10800 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) { 10801 if (SvREADONLY(tmpRef)) 10802 Perl_croak_no_modify(); 10803 if (SvTYPE(tmpRef) == SVt_PVOBJ) 10804 Perl_croak(aTHX_ "Can't bless an object reference"); 10805 if (SvOBJECT(tmpRef)) { 10806 oldstash = SvSTASH(tmpRef); 10807 } 10808 } 10809 SvOBJECT_on(tmpRef); 10810 SvUPGRADE(tmpRef, SVt_PVMG); 10811 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash))); 10812 SvREFCNT_dec(oldstash); 10813 10814 if(SvSMAGICAL(tmpRef)) 10815 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar)) 10816 mg_set(tmpRef); 10817 10818 10819 10820 return sv; 10821} 10822 10823/* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type 10824 * as it is after unglobbing it. 10825 */ 10826 10827PERL_STATIC_INLINE void 10828S_sv_unglob(pTHX_ SV *const sv, U32 flags) 10829{ 10830 void *xpvmg; 10831 HV *stash; 10832 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal(); 10833 10834 PERL_ARGS_ASSERT_SV_UNGLOB; 10835 10836 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV); 10837 SvFAKE_off(sv); 10838 if (!(flags & SV_COW_DROP_PV)) 10839 gv_efullname3(temp, MUTABLE_GV(sv), "*"); 10840 10841 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv)); 10842 if (GvGP(sv)) { 10843 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv))) 10844 && HvHasNAME(stash)) 10845 mro_method_changed_in(stash); 10846 gp_free(MUTABLE_GV(sv)); 10847 } 10848 if (GvSTASH(sv)) { 10849 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv); 10850 GvSTASH(sv) = NULL; 10851 } 10852 GvMULTI_off(sv); 10853 if (GvNAME_HEK(sv)) { 10854 unshare_hek(GvNAME_HEK(sv)); 10855 } 10856 isGV_with_GP_off(sv); 10857 10858 if(SvTYPE(sv) == SVt_PVGV) { 10859 /* need to keep SvANY(sv) in the right arena */ 10860 xpvmg = new_XPVMG(); 10861 StructCopy(SvANY(sv), xpvmg, XPVMG); 10862 del_body_by_type(SvANY(sv), SVt_PVGV); 10863 SvANY(sv) = xpvmg; 10864 10865 SvFLAGS(sv) &= ~SVTYPEMASK; 10866 SvFLAGS(sv) |= SVt_PVMG; 10867 } 10868 10869 /* Intentionally not calling any local SET magic, as this isn't so much a 10870 set operation as merely an internal storage change. */ 10871 if (flags & SV_COW_DROP_PV) SvOK_off(sv); 10872 else sv_setsv_flags(sv, temp, 0); 10873 10874 if ((const GV *)sv == PL_last_in_gv) 10875 PL_last_in_gv = NULL; 10876 else if ((const GV *)sv == PL_statgv) 10877 PL_statgv = NULL; 10878} 10879 10880/* 10881=for apidoc sv_unref_flags 10882 10883Unsets the RV status of the SV, and decrements the reference count of 10884whatever was being referenced by the RV. This can almost be thought of 10885as a reversal of C<newSVrv>. The C<cflags> argument can contain 10886C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented 10887(otherwise the decrementing is conditional on the reference count being 10888different from one or the reference being a readonly SV). 10889See C<L</SvROK_off>>. 10890 10891=for apidoc Amnh||SV_IMMEDIATE_UNREF 10892 10893=cut 10894*/ 10895 10896void 10897Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags) 10898{ 10899 SV* const target = SvRV(ref); 10900 10901 PERL_ARGS_ASSERT_SV_UNREF_FLAGS; 10902 10903 if (SvWEAKREF(ref)) { 10904 sv_del_backref(target, ref); 10905 SvWEAKREF_off(ref); 10906 SvRV_set(ref, NULL); 10907 return; 10908 } 10909 SvRV_set(ref, NULL); 10910 SvROK_off(ref); 10911 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was 10912 assigned to as BEGIN {$a = \"Foo"} will fail. */ 10913 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF)) 10914 SvREFCNT_dec_NN(target); 10915 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */ 10916 sv_2mortal(target); /* Schedule for freeing later */ 10917} 10918 10919/* 10920=for apidoc sv_untaint 10921 10922Untaint an SV. Use C<SvTAINTED_off> instead. 10923 10924=cut 10925*/ 10926 10927void 10928Perl_sv_untaint(pTHX_ SV *const sv) 10929{ 10930 PERL_ARGS_ASSERT_SV_UNTAINT; 10931 PERL_UNUSED_CONTEXT; 10932 10933 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 10934 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint); 10935 if (mg) 10936 mg->mg_len &= ~1; 10937 } 10938} 10939 10940/* 10941=for apidoc sv_tainted 10942 10943Test an SV for taintedness. Use C<SvTAINTED> instead. 10944 10945=cut 10946*/ 10947 10948bool 10949Perl_sv_tainted(pTHX_ SV *const sv) 10950{ 10951 PERL_ARGS_ASSERT_SV_TAINTED; 10952 PERL_UNUSED_CONTEXT; 10953 10954 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 10955 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint); 10956 if (mg && (mg->mg_len & 1) ) 10957 return TRUE; 10958 } 10959 return FALSE; 10960} 10961 10962#if defined(MULTIPLICITY) 10963 10964/* pTHX_ magic can't cope with varargs, so this is a no-context 10965 * version of the main function, (which may itself be aliased to us). 10966 * Don't access this version directly. 10967 */ 10968 10969void 10970Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...) 10971{ 10972 dTHX; 10973 va_list args; 10974 10975 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT; 10976 10977 va_start(args, pat); 10978 sv_vsetpvf(sv, pat, &args); 10979 va_end(args); 10980} 10981 10982/* pTHX_ magic can't cope with varargs, so this is a no-context 10983 * version of the main function, (which may itself be aliased to us). 10984 * Don't access this version directly. 10985 */ 10986 10987void 10988Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...) 10989{ 10990 dTHX; 10991 va_list args; 10992 10993 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT; 10994 10995 va_start(args, pat); 10996 sv_vsetpvf_mg(sv, pat, &args); 10997 va_end(args); 10998} 10999#endif 11000 11001/* 11002=for apidoc sv_setpvf 11003=for apidoc_item sv_setpvf_mg 11004=for apidoc_item sv_setpvf_mg_nocontext 11005=for apidoc_item sv_setpvf_nocontext 11006 11007These work like C<L</sv_catpvf>> but copy the text into the SV instead of 11008appending it. 11009 11010The differences between these are: 11011 11012C<sv_setpvf_mg> and C<sv_setpvf_mg_nocontext> perform 'set' magic; C<sv_setpvf> 11013and C<sv_setpvf_nocontext> skip all magic. 11014 11015C<sv_setpvf_nocontext> and C<sv_setpvf_mg_nocontext> do not take a thread 11016context (C<aTHX>) parameter, so are used in situations where the caller 11017doesn't already have the thread context. 11018 11019=cut 11020*/ 11021 11022void 11023Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...) 11024{ 11025 va_list args; 11026 11027 PERL_ARGS_ASSERT_SV_SETPVF; 11028 11029 va_start(args, pat); 11030 sv_vsetpvf(sv, pat, &args); 11031 va_end(args); 11032} 11033 11034/* 11035=for apidoc sv_vsetpvf 11036=for apidoc_item sv_vsetpvf_mg 11037 11038These work like C<L</sv_vcatpvf>> but copy the text into the SV instead of 11039appending it. 11040 11041They differ only in that C<sv_vsetpvf_mg> performs 'set' magic; 11042C<sv_vsetpvf> skips all magic. 11043 11044They are usually used via their frontends, C<L</sv_setpvf>> and 11045C<L</sv_setpvf_mg>>. 11046 11047=cut 11048*/ 11049 11050void 11051Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args) 11052{ 11053 PERL_ARGS_ASSERT_SV_VSETPVF; 11054 11055 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); 11056} 11057 11058void 11059Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...) 11060{ 11061 va_list args; 11062 11063 PERL_ARGS_ASSERT_SV_SETPVF_MG; 11064 11065 va_start(args, pat); 11066 sv_vsetpvf_mg(sv, pat, &args); 11067 va_end(args); 11068} 11069 11070void 11071Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args) 11072{ 11073 PERL_ARGS_ASSERT_SV_VSETPVF_MG; 11074 11075 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); 11076 SvSETMAGIC(sv); 11077} 11078 11079#if defined(MULTIPLICITY) 11080 11081/* pTHX_ magic can't cope with varargs, so this is a no-context 11082 * version of the main function, (which may itself be aliased to us). 11083 * Don't access this version directly. 11084 */ 11085 11086void 11087Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...) 11088{ 11089 dTHX; 11090 va_list args; 11091 11092 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT; 11093 11094 va_start(args, pat); 11095 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); 11096 va_end(args); 11097} 11098 11099/* pTHX_ magic can't cope with varargs, so this is a no-context 11100 * version of the main function, (which may itself be aliased to us). 11101 * Don't access this version directly. 11102 */ 11103 11104void 11105Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...) 11106{ 11107 dTHX; 11108 va_list args; 11109 11110 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT; 11111 11112 va_start(args, pat); 11113 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); 11114 SvSETMAGIC(sv); 11115 va_end(args); 11116} 11117#endif 11118 11119/* 11120=for apidoc sv_catpvf 11121=for apidoc_item sv_catpvf_mg 11122=for apidoc_item sv_catpvf_mg_nocontext 11123=for apidoc_item sv_catpvf_nocontext 11124 11125These process their arguments like C<sprintf>, and append the formatted 11126output to an SV. As with C<sv_vcatpvfn>, argument reordering is not supporte 11127when called with a non-null C-style variable argument list. 11128 11129If the appended data contains "wide" characters 11130(including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>, 11131and characters >255 formatted with C<%c>), the original SV might get 11132upgraded to UTF-8. 11133 11134If the original SV was UTF-8, the pattern should be 11135valid UTF-8; if the original SV was bytes, the pattern should be too. 11136 11137All perform 'get' magic, but only C<sv_catpvf_mg> and C<sv_catpvf_mg_nocontext> 11138perform 'set' magic. 11139 11140C<sv_catpvf_nocontext> and C<sv_catpvf_mg_nocontext> do not take a thread 11141context (C<aTHX>) parameter, so are used in situations where the caller 11142doesn't already have the thread context. 11143 11144=cut 11145*/ 11146 11147void 11148Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...) 11149{ 11150 va_list args; 11151 11152 PERL_ARGS_ASSERT_SV_CATPVF; 11153 11154 va_start(args, pat); 11155 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); 11156 va_end(args); 11157} 11158 11159/* 11160=for apidoc sv_vcatpvf 11161=for apidoc_item sv_vcatpvf_mg 11162 11163These process their arguments like C<sv_vcatpvfn> called with a non-null 11164C-style variable argument list, and append the formatted output to C<sv>. 11165 11166They differ only in that C<sv_vcatpvf_mg> performs 'set' magic; 11167C<sv_vcatpvf> skips 'set' magic. 11168 11169Both perform 'get' magic. 11170 11171They are usually accessed via their frontends C<L</sv_catpvf>> and 11172C<L</sv_catpvf_mg>>. 11173 11174=cut 11175*/ 11176 11177void 11178Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args) 11179{ 11180 PERL_ARGS_ASSERT_SV_VCATPVF; 11181 11182 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); 11183} 11184 11185void 11186Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...) 11187{ 11188 va_list args; 11189 11190 PERL_ARGS_ASSERT_SV_CATPVF_MG; 11191 11192 va_start(args, pat); 11193 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); 11194 SvSETMAGIC(sv); 11195 va_end(args); 11196} 11197 11198void 11199Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args) 11200{ 11201 PERL_ARGS_ASSERT_SV_VCATPVF_MG; 11202 11203 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); 11204 SvSETMAGIC(sv); 11205} 11206 11207/* 11208=for apidoc sv_vsetpvfn 11209 11210Works like C<sv_vcatpvfn> but copies the text into the SV instead of 11211appending it. 11212 11213Usually used via one of its frontends L</C<sv_vsetpvf>> and 11214L</C<sv_vsetpvf_mg>>. 11215 11216=cut 11217*/ 11218 11219void 11220Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, 11221 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted) 11222{ 11223 PERL_ARGS_ASSERT_SV_VSETPVFN; 11224 11225 SvPVCLEAR(sv); 11226 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0); 11227} 11228 11229 11230/* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */ 11231 11232PERL_STATIC_INLINE void 11233S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len) 11234{ 11235 STRLEN const need = len + SvCUR(sv) + 1; 11236 char *end; 11237 11238 /* can't wrap as both len and SvCUR() are allocated in 11239 * memory and together can't consume all the address space 11240 */ 11241 assert(need > len); 11242 11243 assert(SvPOK(sv)); 11244 SvGROW(sv, need); 11245 end = SvEND(sv); 11246 Copy(buf, end, len, char); 11247 end += len; 11248 *end = '\0'; 11249 SvCUR_set(sv, need - 1); 11250} 11251 11252 11253/* 11254 * Warn of missing argument to sprintf. The value used in place of such 11255 * arguments should be &PL_sv_no; an undefined value would yield 11256 * inappropriate "use of uninit" warnings [perl #71000]. 11257 */ 11258STATIC void 11259S_warn_vcatpvfn_missing_argument(pTHX) { 11260 if (ckWARN(WARN_MISSING)) { 11261 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s", 11262 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); 11263 } 11264} 11265 11266 11267static void 11268S_croak_overflow() 11269{ 11270 dTHX; 11271 Perl_croak(aTHX_ "Integer overflow in format string for %s", 11272 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn")); 11273} 11274 11275 11276/* Given an int i from the next arg (if args is true) or an sv from an arg 11277 * (if args is false), try to extract a STRLEN-ranged value from the arg, 11278 * with overflow checking. 11279 * Sets *neg to true if the value was negative (untouched otherwise. 11280 * Returns the absolute value. 11281 * As an extra margin of safety, it croaks if the returned value would 11282 * exceed the maximum value of a STRLEN / 4. 11283 */ 11284 11285static STRLEN 11286S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg) 11287{ 11288 IV iv; 11289 11290 if (args) { 11291 iv = i; 11292 goto do_iv; 11293 } 11294 11295 if (!sv) 11296 return 0; 11297 11298 SvGETMAGIC(sv); 11299 11300 if (UNLIKELY(SvIsUV(sv))) { 11301 UV uv = SvUV_nomg(sv); 11302 if (uv > IV_MAX) 11303 S_croak_overflow(); 11304 iv = uv; 11305 } 11306 else { 11307 iv = SvIV_nomg(sv); 11308 do_iv: 11309 if (iv < 0) { 11310 if (iv < -IV_MAX) 11311 S_croak_overflow(); 11312 iv = -iv; 11313 *neg = TRUE; 11314 } 11315 } 11316 11317 if (iv > (IV)(((STRLEN)~0) / 4)) 11318 S_croak_overflow(); 11319 11320 return (STRLEN)iv; 11321} 11322 11323/* Read in and return a number. Updates *pattern to point to the char 11324 * following the number. Expects the first char to 1..9. 11325 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN. 11326 * This is a belt-and-braces safety measure to complement any 11327 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags. 11328 * It means that e.g. on a 32-bit system the width/precision can't be more 11329 * than 1G, which seems reasonable. 11330 */ 11331 11332STATIC STRLEN 11333S_expect_number(pTHX_ const char **const pattern) 11334{ 11335 STRLEN var; 11336 11337 PERL_ARGS_ASSERT_EXPECT_NUMBER; 11338 11339 assert(inRANGE(**pattern, '1', '9')); 11340 11341 var = *(*pattern)++ - '0'; 11342 while (isDIGIT(**pattern)) { 11343 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */ 11344 if (var > ((((STRLEN)~0) / 4 - 9) / 10)) 11345 S_croak_overflow(); 11346 var = var * 10 + (*(*pattern)++ - '0'); 11347 } 11348 return var; 11349} 11350 11351/* Implement a fast "%.0f": given a pointer to the end of a buffer (caller 11352 * ensures it's big enough), back fill it with the rounded integer part of 11353 * nv. Returns ptr to start of string, and sets *len to its length. 11354 * Returns NULL if not convertible. 11355 */ 11356 11357STATIC char * 11358S_F0convert(NV nv, char *const endbuf, STRLEN *const len) 11359{ 11360 const int neg = nv < 0; 11361 UV uv; 11362 11363 PERL_ARGS_ASSERT_F0CONVERT; 11364 11365 assert(!Perl_isinfnan(nv)); 11366 if (neg) 11367 nv = -nv; 11368 if (nv != 0.0 && nv < (NV) UV_MAX) { 11369 char *p = endbuf; 11370 uv = (UV)nv; 11371 if (uv != nv) { 11372 nv += 0.5; 11373 uv = (UV)nv; 11374 if (uv & 1 && uv == nv) 11375 uv--; /* Round to even */ 11376 } 11377 do { 11378 const unsigned dig = uv % 10; 11379 *--p = '0' + dig; 11380 } while (uv /= 10); 11381 if (neg) 11382 *--p = '-'; 11383 *len = endbuf - p; 11384 return p; 11385 } 11386 return NULL; 11387} 11388 11389 11390/* XXX maybe_tainted is never assigned to, so the doc above is lying. */ 11391 11392void 11393Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, 11394 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted) 11395{ 11396 PERL_ARGS_ASSERT_SV_VCATPVFN; 11397 11398 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC); 11399} 11400 11401 11402/* For the vcatpvfn code, we need a long double target in case 11403 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf 11404 * with long double formats, even without NV being long double. But we 11405 * call the target 'fv' instead of 'nv', since most of the time it is not 11406 * (most compilers these days recognize "long double", even if only as a 11407 * synonym for "double"). 11408*/ 11409#if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \ 11410 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH) 11411# define VCATPVFN_FV_GF PERL_PRIgldbl 11412# if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT) 11413 /* Work around breakage in OTS$CVT_FLOAT_T_X */ 11414# define VCATPVFN_NV_TO_FV(nv,fv) \ 11415 STMT_START { \ 11416 double _dv = nv; \ 11417 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \ 11418 } STMT_END 11419# else 11420# define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv) 11421# endif 11422 typedef long double vcatpvfn_long_double_t; 11423#else 11424# define VCATPVFN_FV_GF NVgf 11425# define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv) 11426 typedef NV vcatpvfn_long_double_t; 11427#endif 11428 11429#ifdef LONGDOUBLE_DOUBLEDOUBLE 11430/* The first double can be as large as 2**1023, or '1' x '0' x 1023. 11431 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'. 11432 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point 11433 * after the first 1023 zero bits. 11434 * 11435 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort 11436 * of dynamically growing buffer might be better, start at just 16 bytes 11437 * (for example) and grow only when necessary. Or maybe just by looking 11438 * at the exponents of the two doubles? */ 11439# define DOUBLEDOUBLE_MAXBITS 2098 11440#endif 11441 11442/* vhex will contain the values (0..15) of the hex digits ("nybbles" 11443 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits 11444 * per xdigit. For the double-double case, this can be rather many. 11445 * The non-double-double-long-double overshoots since all bits of NV 11446 * are not mantissa bits, there are also exponent bits. */ 11447#ifdef LONGDOUBLE_DOUBLEDOUBLE 11448# define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4) 11449#else 11450# define VHEX_SIZE (1+(NVSIZE * 8)/4) 11451#endif 11452 11453/* If we do not have a known long double format, (including not using 11454 * long doubles, or long doubles being equal to doubles) then we will 11455 * fall back to the ldexp/frexp route, with which we can retrieve at 11456 * most as many bits as our widest unsigned integer type is. We try 11457 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV. 11458 * 11459 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8, 11460 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.) 11461 */ 11462#if defined(HAS_QUAD) && defined(Uquad_t) 11463# define MANTISSATYPE Uquad_t 11464# define MANTISSASIZE 8 11465#else 11466# define MANTISSATYPE UV 11467# define MANTISSASIZE UVSIZE 11468#endif 11469 11470#if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN) 11471# define HEXTRACT_LITTLE_ENDIAN 11472#elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN) 11473# define HEXTRACT_BIG_ENDIAN 11474#else 11475# define HEXTRACT_MIX_ENDIAN 11476#endif 11477 11478/* S_hextract() is a helper for S_format_hexfp, for extracting 11479 * the hexadecimal values (for %a/%A). The nv is the NV where the value 11480 * are being extracted from (either directly from the long double in-memory 11481 * presentation, or from the uquad computed via frexp+ldexp). frexp also 11482 * is used to update the exponent. The subnormal is set to true 11483 * for IEEE 754 subnormals/denormals (including the x86 80-bit format). 11484 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE. 11485 * 11486 * The tricky part is that S_hextract() needs to be called twice: 11487 * the first time with vend as NULL, and the second time with vend as 11488 * the pointer returned by the first call. What happens is that on 11489 * the first round the output size is computed, and the intended 11490 * extraction sanity checked. On the second round the actual output 11491 * (the extraction of the hexadecimal values) takes place. 11492 * Sanity failures cause fatal failures during both rounds. */ 11493STATIC U8* 11494S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal, 11495 U8* vhex, U8* vend) 11496{ 11497 U8* v = vhex; 11498 int ix; 11499 int ixmin = 0, ixmax = 0; 11500 11501 /* XXX Inf/NaN are not handled here, since it is 11502 * assumed they are to be output as "Inf" and "NaN". */ 11503 11504 /* These macros are just to reduce typos, they have multiple 11505 * repetitions below, but usually only one (or sometimes two) 11506 * of them is really being used. */ 11507 /* HEXTRACT_OUTPUT() extracts the high nybble first. */ 11508#define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4) 11509#define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF) 11510#define HEXTRACT_OUTPUT(ix) \ 11511 STMT_START { \ 11512 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \ 11513 } STMT_END 11514#define HEXTRACT_COUNT(ix, c) \ 11515 STMT_START { \ 11516 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \ 11517 } STMT_END 11518#define HEXTRACT_BYTE(ix) \ 11519 STMT_START { \ 11520 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \ 11521 } STMT_END 11522#define HEXTRACT_LO_NYBBLE(ix) \ 11523 STMT_START { \ 11524 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \ 11525 } STMT_END 11526 /* HEXTRACT_TOP_NYBBLE is just convenience disguise, 11527 * to make it look less odd when the top bits of a NV 11528 * are extracted using HEXTRACT_LO_NYBBLE: the highest 11529 * order bits can be in the "low nybble" of a byte. */ 11530#define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix) 11531#define HEXTRACT_BYTES_LE(a, b) \ 11532 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); } 11533#define HEXTRACT_BYTES_BE(a, b) \ 11534 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); } 11535#define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv) 11536#define HEXTRACT_IMPLICIT_BIT(nv) \ 11537 STMT_START { \ 11538 if (!*subnormal) { \ 11539 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \ 11540 } \ 11541 } STMT_END 11542 11543/* Most formats do. Those which don't should undef this. 11544 * 11545 * But also note that IEEE 754 subnormals do not have it, or, 11546 * expressed alternatively, their implicit bit is zero. */ 11547#define HEXTRACT_HAS_IMPLICIT_BIT 11548 11549/* Many formats do. Those which don't should undef this. */ 11550#define HEXTRACT_HAS_TOP_NYBBLE 11551 11552 /* HEXTRACTSIZE is the maximum number of xdigits. */ 11553#if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE) 11554# define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4) 11555#else 11556# define HEXTRACTSIZE 2 * NVSIZE 11557#endif 11558 11559 const U8* vmaxend = vhex + HEXTRACTSIZE; 11560 11561 assert(HEXTRACTSIZE <= VHEX_SIZE); 11562 11563 PERL_UNUSED_VAR(ix); /* might happen */ 11564 (void)Perl_frexp(PERL_ABS(nv), exponent); 11565 *subnormal = FALSE; 11566 if (vend && (vend <= vhex || vend > vmaxend)) { 11567 /* diag_listed_as: Hexadecimal float: internal error (%s) */ 11568 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)"); 11569 } 11570 { 11571 /* First check if using long doubles. */ 11572#if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) 11573# if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN 11574 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L: 11575 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */ 11576 /* The bytes 13..0 are the mantissa/fraction, 11577 * the 15,14 are the sign+exponent. */ 11578 const U8* nvp = (const U8*)(&nv); 11579 HEXTRACT_GET_SUBNORMAL(nv); 11580 HEXTRACT_IMPLICIT_BIT(nv); 11581# undef HEXTRACT_HAS_TOP_NYBBLE 11582 HEXTRACT_BYTES_LE(13, 0); 11583# elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN 11584 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L: 11585 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */ 11586 /* The bytes 2..15 are the mantissa/fraction, 11587 * the 0,1 are the sign+exponent. */ 11588 const U8* nvp = (const U8*)(&nv); 11589 HEXTRACT_GET_SUBNORMAL(nv); 11590 HEXTRACT_IMPLICIT_BIT(nv); 11591# undef HEXTRACT_HAS_TOP_NYBBLE 11592 HEXTRACT_BYTES_BE(2, 15); 11593# elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN 11594 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction / 11595 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit. 11596 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux 11597 * and OS X), meaning that 2 or 6 bytes are empty padding. */ 11598 /* The bytes 0..1 are the sign+exponent, 11599 * the bytes 2..9 are the mantissa/fraction. */ 11600 const U8* nvp = (const U8*)(&nv); 11601# undef HEXTRACT_HAS_IMPLICIT_BIT 11602# undef HEXTRACT_HAS_TOP_NYBBLE 11603 HEXTRACT_GET_SUBNORMAL(nv); 11604 HEXTRACT_BYTES_LE(7, 0); 11605# elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN 11606 /* Does this format ever happen? (Wikipedia says the Motorola 11607 * 6888x math coprocessors used format _like_ this but padded 11608 * to 96 bits with 16 unused bits between the exponent and the 11609 * mantissa.) */ 11610 const U8* nvp = (const U8*)(&nv); 11611# undef HEXTRACT_HAS_IMPLICIT_BIT 11612# undef HEXTRACT_HAS_TOP_NYBBLE 11613 HEXTRACT_GET_SUBNORMAL(nv); 11614 HEXTRACT_BYTES_BE(0, 7); 11615# else 11616# define HEXTRACT_FALLBACK 11617 /* Double-double format: two doubles next to each other. 11618 * The first double is the high-order one, exactly like 11619 * it would be for a "lone" double. The second double 11620 * is shifted down using the exponent so that that there 11621 * are no common bits. The tricky part is that the value 11622 * of the double-double is the SUM of the two doubles and 11623 * the second one can be also NEGATIVE. 11624 * 11625 * Because of this tricky construction the bytewise extraction we 11626 * use for the other long double formats doesn't work, we must 11627 * extract the values bit by bit. 11628 * 11629 * The little-endian double-double is used .. somewhere? 11630 * 11631 * The big endian double-double is used in e.g. PPC/Power (AIX) 11632 * and MIPS (SGI). 11633 * 11634 * The mantissa bits are in two separate stretches, e.g. for -0.1L: 11635 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE) 11636 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE) 11637 */ 11638# endif 11639#else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */ 11640 /* Using normal doubles, not long doubles. 11641 * 11642 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit 11643 * bytes, since we might need to handle printf precision, and 11644 * also need to insert the radix. */ 11645# if NVSIZE == 8 11646# ifdef HEXTRACT_LITTLE_ENDIAN 11647 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */ 11648 const U8* nvp = (const U8*)(&nv); 11649 HEXTRACT_GET_SUBNORMAL(nv); 11650 HEXTRACT_IMPLICIT_BIT(nv); 11651 HEXTRACT_TOP_NYBBLE(6); 11652 HEXTRACT_BYTES_LE(5, 0); 11653# elif defined(HEXTRACT_BIG_ENDIAN) 11654 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */ 11655 const U8* nvp = (const U8*)(&nv); 11656 HEXTRACT_GET_SUBNORMAL(nv); 11657 HEXTRACT_IMPLICIT_BIT(nv); 11658 HEXTRACT_TOP_NYBBLE(1); 11659 HEXTRACT_BYTES_BE(2, 7); 11660# elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE 11661 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */ 11662 const U8* nvp = (const U8*)(&nv); 11663 HEXTRACT_GET_SUBNORMAL(nv); 11664 HEXTRACT_IMPLICIT_BIT(nv); 11665 HEXTRACT_TOP_NYBBLE(2); /* 6 */ 11666 HEXTRACT_BYTE(1); /* 5 */ 11667 HEXTRACT_BYTE(0); /* 4 */ 11668 HEXTRACT_BYTE(7); /* 3 */ 11669 HEXTRACT_BYTE(6); /* 2 */ 11670 HEXTRACT_BYTE(5); /* 1 */ 11671 HEXTRACT_BYTE(4); /* 0 */ 11672# elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE 11673 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */ 11674 const U8* nvp = (const U8*)(&nv); 11675 HEXTRACT_GET_SUBNORMAL(nv); 11676 HEXTRACT_IMPLICIT_BIT(nv); 11677 HEXTRACT_TOP_NYBBLE(5); /* 6 */ 11678 HEXTRACT_BYTE(6); /* 5 */ 11679 HEXTRACT_BYTE(7); /* 4 */ 11680 HEXTRACT_BYTE(0); /* 3 */ 11681 HEXTRACT_BYTE(1); /* 2 */ 11682 HEXTRACT_BYTE(2); /* 1 */ 11683 HEXTRACT_BYTE(3); /* 0 */ 11684# else 11685# define HEXTRACT_FALLBACK 11686# endif 11687# else 11688# define HEXTRACT_FALLBACK 11689# endif 11690#endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */ 11691 11692#ifdef HEXTRACT_FALLBACK 11693 HEXTRACT_GET_SUBNORMAL(nv); 11694# undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */ 11695 /* The fallback is used for the double-double format, and 11696 * for unknown long double formats, and for unknown double 11697 * formats, or in general unknown NV formats. */ 11698 if (nv == (NV)0.0) { 11699 if (vend) 11700 *v++ = 0; 11701 else 11702 v++; 11703 *exponent = 0; 11704 } 11705 else { 11706 NV d = nv < 0 ? -nv : nv; 11707 NV e = (NV)1.0; 11708 U8 ha = 0x0; /* hexvalue accumulator */ 11709 U8 hd = 0x8; /* hexvalue digit */ 11710 11711 /* Shift d and e (and update exponent) so that e <= d < 2*e, 11712 * this is essentially manual frexp(). Multiplying by 0.5 and 11713 * doubling should be lossless in binary floating point. */ 11714 11715 *exponent = 1; 11716 11717 while (e > d) { 11718 e *= (NV)0.5; 11719 (*exponent)--; 11720 } 11721 /* Now d >= e */ 11722 11723 while (d >= e + e) { 11724 e += e; 11725 (*exponent)++; 11726 } 11727 /* Now e <= d < 2*e */ 11728 11729 /* First extract the leading hexdigit (the implicit bit). */ 11730 if (d >= e) { 11731 d -= e; 11732 if (vend) 11733 *v++ = 1; 11734 else 11735 v++; 11736 } 11737 else { 11738 if (vend) 11739 *v++ = 0; 11740 else 11741 v++; 11742 } 11743 e *= (NV)0.5; 11744 11745 /* Then extract the remaining hexdigits. */ 11746 while (d > (NV)0.0) { 11747 if (d >= e) { 11748 ha |= hd; 11749 d -= e; 11750 } 11751 if (hd == 1) { 11752 /* Output or count in groups of four bits, 11753 * that is, when the hexdigit is down to one. */ 11754 if (vend) 11755 *v++ = ha; 11756 else 11757 v++; 11758 /* Reset the hexvalue. */ 11759 ha = 0x0; 11760 hd = 0x8; 11761 } 11762 else 11763 hd >>= 1; 11764 e *= (NV)0.5; 11765 } 11766 11767 /* Flush possible pending hexvalue. */ 11768 if (ha) { 11769 if (vend) 11770 *v++ = ha; 11771 else 11772 v++; 11773 } 11774 } 11775#endif 11776 } 11777 /* Croak for various reasons: if the output pointer escaped the 11778 * output buffer, if the extraction index escaped the extraction 11779 * buffer, or if the ending output pointer didn't match the 11780 * previously computed value. */ 11781 if (v <= vhex || v - vhex >= VHEX_SIZE || 11782 /* For double-double the ixmin and ixmax stay at zero, 11783 * which is convenient since the HEXTRACTSIZE is tricky 11784 * for double-double. */ 11785 ixmin < 0 || ixmax >= NVSIZE || 11786 (vend && v != vend)) { 11787 /* diag_listed_as: Hexadecimal float: internal error (%s) */ 11788 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)"); 11789 } 11790 return v; 11791} 11792 11793 11794/* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags(). 11795 * 11796 * Processes the %a/%A hexadecimal floating-point format, since the 11797 * built-in snprintf()s which are used for most of the f/p formats, don't 11798 * universally handle %a/%A. 11799 * Populates buf of length bufsize, and returns the length of the created 11800 * string. 11801 * The rest of the args have the same meaning as the local vars of the 11802 * same name within Perl_sv_vcatpvfn_flags(). 11803 * 11804 * The caller's determination of IN_LC(LC_NUMERIC), passed as in_lc_numeric, 11805 * is used to ensure we do the right thing when we need to access the locale's 11806 * numeric radix. 11807 * 11808 * It requires the caller to make buf large enough. 11809 */ 11810 11811static STRLEN 11812S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c, 11813 const NV nv, const vcatpvfn_long_double_t fv, 11814 bool has_precis, STRLEN precis, STRLEN width, 11815 bool alt, char plus, bool left, bool fill, bool in_lc_numeric) 11816{ 11817 /* Hexadecimal floating point. */ 11818 char* p = buf; 11819 U8 vhex[VHEX_SIZE]; 11820 U8* v = vhex; /* working pointer to vhex */ 11821 U8* vend; /* pointer to one beyond last digit of vhex */ 11822 U8* vfnz = NULL; /* first non-zero */ 11823 U8* vlnz = NULL; /* last non-zero */ 11824 U8* v0 = NULL; /* first output */ 11825 const bool lower = (c == 'a'); 11826 /* At output the values of vhex (up to vend) will 11827 * be mapped through the xdig to get the actual 11828 * human-readable xdigits. */ 11829 const char* xdig = PL_hexdigit; 11830 STRLEN zerotail = 0; /* how many extra zeros to append */ 11831 int exponent = 0; /* exponent of the floating point input */ 11832 bool hexradix = FALSE; /* should we output the radix */ 11833 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */ 11834 bool negative = FALSE; 11835 STRLEN elen; 11836 11837 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf". 11838 * 11839 * For example with denormals, (assuming the vanilla 11840 * 64-bit double): the exponent is zero. 1xp-1074 is 11841 * the smallest denormal and the smallest double, it 11842 * could be output also as 0x0.0000000000001p-1022 to 11843 * match its internal structure. */ 11844 11845 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL); 11846 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend); 11847 11848#if NVSIZE > DOUBLESIZE 11849# ifdef HEXTRACT_HAS_IMPLICIT_BIT 11850 /* In this case there is an implicit bit, 11851 * and therefore the exponent is shifted by one. */ 11852 exponent--; 11853# elif defined(NV_X86_80_BIT) 11854 if (subnormal) { 11855 /* The subnormals of the x86-80 have a base exponent of -16382, 11856 * (while the physical exponent bits are zero) but the frexp() 11857 * returned the scientific-style floating exponent. We want 11858 * to map the last one as: 11859 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382) 11860 * -16835..-16388 -> -16384 11861 * since we want to keep the first hexdigit 11862 * as one of the [8421]. */ 11863 exponent = -4 * ( (exponent + 1) / -4) - 2; 11864 } else { 11865 exponent -= 4; 11866 } 11867 /* TBD: other non-implicit-bit platforms than the x86-80. */ 11868# endif 11869#endif 11870 11871 negative = fv < 0 || Perl_signbit(nv); 11872 if (negative) 11873 *p++ = '-'; 11874 else if (plus) 11875 *p++ = plus; 11876 *p++ = '0'; 11877 if (lower) { 11878 *p++ = 'x'; 11879 } 11880 else { 11881 *p++ = 'X'; 11882 xdig += 16; /* Use uppercase hex. */ 11883 } 11884 11885 /* Find the first non-zero xdigit. */ 11886 for (v = vhex; v < vend; v++) { 11887 if (*v) { 11888 vfnz = v; 11889 break; 11890 } 11891 } 11892 11893 if (vfnz) { 11894 /* Find the last non-zero xdigit. */ 11895 for (v = vend - 1; v >= vhex; v--) { 11896 if (*v) { 11897 vlnz = v; 11898 break; 11899 } 11900 } 11901 11902#if NVSIZE == DOUBLESIZE 11903 if (fv != 0.0) 11904 exponent--; 11905#endif 11906 11907 if (subnormal) { 11908#ifndef NV_X86_80_BIT 11909 if (vfnz[0] > 1) { 11910 /* IEEE 754 subnormals (but not the x86 80-bit): 11911 * we want "normalize" the subnormal, 11912 * so we need to right shift the hex nybbles 11913 * so that the output of the subnormal starts 11914 * from the first true bit. (Another, equally 11915 * valid, policy would be to dump the subnormal 11916 * nybbles as-is, to display the "physical" layout.) */ 11917 int i, n; 11918 U8 *vshr; 11919 /* Find the ceil(log2(v[0])) of 11920 * the top non-zero nybble. */ 11921 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { } 11922 assert(n < 4); 11923 assert(vlnz); 11924 vlnz[1] = 0; 11925 for (vshr = vlnz; vshr >= vfnz; vshr--) { 11926 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n); 11927 vshr[0] >>= n; 11928 } 11929 if (vlnz[1]) { 11930 vlnz++; 11931 } 11932 } 11933#endif 11934 v0 = vfnz; 11935 } else { 11936 v0 = vhex; 11937 } 11938 11939 if (has_precis) { 11940 U8* ve = (subnormal ? vlnz + 1 : vend); 11941 SSize_t vn = ve - v0; 11942 assert(vn >= 1); 11943 if (precis < (Size_t)(vn - 1)) { 11944 bool overflow = FALSE; 11945 if (v0[precis + 1] < 0x8) { 11946 /* Round down, nothing to do. */ 11947 } else if (v0[precis + 1] > 0x8) { 11948 /* Round up. */ 11949 v0[precis]++; 11950 overflow = v0[precis] > 0xF; 11951 v0[precis] &= 0xF; 11952 } else { /* v0[precis] == 0x8 */ 11953 /* Half-point: round towards the one 11954 * with the even least-significant digit: 11955 * 08 -> 0 88 -> 8 11956 * 18 -> 2 98 -> a 11957 * 28 -> 2 a8 -> a 11958 * 38 -> 4 b8 -> c 11959 * 48 -> 4 c8 -> c 11960 * 58 -> 6 d8 -> e 11961 * 68 -> 6 e8 -> e 11962 * 78 -> 8 f8 -> 10 */ 11963 if ((v0[precis] & 0x1)) { 11964 v0[precis]++; 11965 } 11966 overflow = v0[precis] > 0xF; 11967 v0[precis] &= 0xF; 11968 } 11969 11970 if (overflow) { 11971 for (v = v0 + precis - 1; v >= v0; v--) { 11972 (*v)++; 11973 overflow = *v > 0xF; 11974 (*v) &= 0xF; 11975 if (!overflow) { 11976 break; 11977 } 11978 } 11979 if (v == v0 - 1 && overflow) { 11980 /* If the overflow goes all the 11981 * way to the front, we need to 11982 * insert 0x1 in front, and adjust 11983 * the exponent. */ 11984 Move(v0, v0 + 1, vn - 1, char); 11985 *v0 = 0x1; 11986 exponent += 4; 11987 } 11988 } 11989 11990 /* The new effective "last non zero". */ 11991 vlnz = v0 + precis; 11992 } 11993 else { 11994 zerotail = 11995 subnormal ? precis - vn + 1 : 11996 precis - (vlnz - vhex); 11997 } 11998 } 11999 12000 v = v0; 12001 *p++ = xdig[*v++]; 12002 12003 /* If there are non-zero xdigits, the radix 12004 * is output after the first one. */ 12005 if (vfnz < vlnz) { 12006 hexradix = TRUE; 12007 } 12008 } 12009 else { 12010 *p++ = '0'; 12011 exponent = 0; 12012 zerotail = has_precis ? precis : 0; 12013 } 12014 12015 /* The radix is always output if precis, or if alt. */ 12016 if ((has_precis && precis > 0) || alt) { 12017 hexradix = TRUE; 12018 } 12019 12020 if (hexradix) { 12021#ifndef USE_LOCALE_NUMERIC 12022 PERL_UNUSED_ARG(in_lc_numeric); 12023 12024 *p++ = '.'; 12025#else 12026 if (in_lc_numeric) { 12027 STRLEN n; 12028 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, { 12029 const char* r = SvPV(PL_numeric_radix_sv, n); 12030 Copy(r, p, n, char); 12031 }); 12032 p += n; 12033 } 12034 else { 12035 *p++ = '.'; 12036 } 12037#endif 12038 } 12039 12040 if (vlnz) { 12041 while (v <= vlnz) 12042 *p++ = xdig[*v++]; 12043 } 12044 12045 if (zerotail > 0) { 12046 while (zerotail--) { 12047 *p++ = '0'; 12048 } 12049 } 12050 12051 elen = p - buf; 12052 12053 /* sanity checks */ 12054 if (elen >= bufsize || width >= bufsize) 12055 /* diag_listed_as: Hexadecimal float: internal error (%s) */ 12056 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)"); 12057 12058 elen += my_snprintf(p, bufsize - elen, 12059 "%c%+d", lower ? 'p' : 'P', 12060 exponent); 12061 12062 if (elen < width) { 12063 STRLEN gap = (STRLEN)(width - elen); 12064 if (left) { 12065 /* Pad the back with spaces. */ 12066 memset(buf + elen, ' ', gap); 12067 } 12068 else if (fill) { 12069 /* Insert the zeros after the "0x" and the 12070 * the potential sign, but before the digits, 12071 * otherwise we end up with "0000xH.HHH...", 12072 * when we want "0x000H.HHH..." */ 12073 STRLEN nzero = gap; 12074 char* zerox = buf + 2; 12075 STRLEN nmove = elen - 2; 12076 if (negative || plus) { 12077 zerox++; 12078 nmove--; 12079 } 12080 Move(zerox, zerox + nzero, nmove, char); 12081 memset(zerox, fill ? '0' : ' ', nzero); 12082 } 12083 else { 12084 /* Move it to the right. */ 12085 Move(buf, buf + gap, 12086 elen, char); 12087 /* Pad the front with spaces. */ 12088 memset(buf, ' ', gap); 12089 } 12090 elen = width; 12091 } 12092 return elen; 12093} 12094 12095/* 12096=for apidoc sv_vcatpvfn 12097=for apidoc_item sv_vcatpvfn_flags 12098 12099These process their arguments like C<L<vsprintf(3)>> and append the formatted output 12100to an SV. They use an array of SVs if the C-style variable argument list is 12101missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d> or 12102C<%*2$d>) is supported only when using an array of SVs; using a C-style 12103C<va_list> argument list with a format string that uses argument reordering 12104will yield an exception. 12105 12106When running with taint checks enabled, they indicate via C<maybe_tainted> if 12107results are untrustworthy (often due to the use of locales). 12108 12109They assume that C<pat> has the same utf8-ness as C<sv>. It's the caller's 12110responsibility to ensure that this is so. 12111 12112They differ in that C<sv_vcatpvfn_flags> has a C<flags> parameter in which you 12113can set or clear the C<SV_GMAGIC> and/or S<SV_SMAGIC> flags, to specify which 12114magic to handle or not handle; whereas plain C<sv_vcatpvfn> always specifies 12115both 'get' and 'set' magic. 12116 12117They are usually used via one of the frontends L</C<sv_vcatpvf>> and 12118L</C<sv_vcatpvf_mg>>. 12119 12120=cut 12121*/ 12122 12123 12124void 12125Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, 12126 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted, 12127 const U32 flags) 12128{ 12129 const char *fmtstart; /* character following the current '%' */ 12130 const char *q; /* current position within format */ 12131 const char *patend; 12132 STRLEN origlen; 12133 Size_t svix = 0; 12134 static const char nullstr[] = "(null)"; 12135 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */ 12136 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */ 12137 /* Times 4: a decimal digit takes more than 3 binary digits. 12138 * NV_DIG: mantissa takes that many decimal digits. 12139 * Plus 32: Playing safe. */ 12140 char ebuf[IV_DIG * 4 + NV_DIG + 32]; 12141 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */ 12142#ifdef USE_LOCALE_NUMERIC 12143 bool have_in_lc_numeric = FALSE; 12144#endif 12145 /* we never change this unless USE_LOCALE_NUMERIC */ 12146 bool in_lc_numeric = FALSE; 12147 SV *tmp_sv = NULL; 12148 12149 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS; 12150 PERL_UNUSED_ARG(maybe_tainted); 12151 12152 if (flags & SV_GMAGIC) 12153 SvGETMAGIC(sv); 12154 12155 /* no matter what, this is a string now */ 12156 (void)SvPV_force_nomg(sv, origlen); 12157 12158 /* the code that scans for flags etc following a % relies on 12159 * a '\0' being present to avoid falling off the end. Ideally that 12160 * should be fixed */ 12161 assert(pat[patlen] == '\0'); 12162 12163 12164 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f". 12165 * In each case, if there isn't the correct number of args, instead 12166 * fall through to the main code to handle the issuing of any 12167 * warnings etc. 12168 */ 12169 12170 if (patlen == 0 && (args || sv_count == 0)) 12171 return; 12172 12173 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) { 12174 12175 /* "%s" */ 12176 if (patlen == 2 && pat[1] == 's') { 12177 if (args) { 12178 const char * const s = va_arg(*args, char*); 12179 sv_catpv_nomg(sv, s ? s : nullstr); 12180 } 12181 else { 12182 /* we want get magic on the source but not the target. 12183 * sv_catsv can't do that, though */ 12184 SvGETMAGIC(*svargs); 12185 sv_catsv_nomg(sv, *svargs); 12186 } 12187 return; 12188 } 12189 12190 /* "%-p" */ 12191 if (args) { 12192 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') { 12193 SV *asv = MUTABLE_SV(va_arg(*args, void*)); 12194 sv_catsv_nomg(sv, asv); 12195 return; 12196 } 12197 } 12198#if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH) 12199 /* special-case "%.0f" */ 12200 else if ( patlen == 4 12201 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f') 12202 { 12203 const NV nv = SvNV(*svargs); 12204 if (LIKELY(!Perl_isinfnan(nv))) { 12205 STRLEN l; 12206 char *p; 12207 12208 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) { 12209 sv_catpvn_nomg(sv, p, l); 12210 return; 12211 } 12212 } 12213 } 12214#endif /* !USE_LONG_DOUBLE */ 12215 } 12216 12217 12218 patend = (char*)pat + patlen; 12219 for (fmtstart = pat; fmtstart < patend; fmtstart = q) { 12220 char intsize = 0; /* size qualifier in "%hi..." etc */ 12221 bool alt = FALSE; /* has "%#..." */ 12222 bool left = FALSE; /* has "%-..." */ 12223 bool fill = FALSE; /* has "%0..." */ 12224 char plus = 0; /* has "%+..." */ 12225 STRLEN width = 0; /* value of "%NNN..." */ 12226 bool has_precis = FALSE; /* has "%.NNN..." */ 12227 STRLEN precis = 0; /* value of "%.NNN..." */ 12228 int base = 0; /* base to print in, e.g. 8 for %o */ 12229 UV uv = 0; /* the value to print of int-ish args */ 12230 12231 bool vectorize = FALSE; /* has "%v..." */ 12232 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */ 12233 const U8 *vecstr = NULL; /* SvPVX(vec arg) */ 12234 STRLEN veclen = 0; /* SvCUR(vec arg) */ 12235 const char *dotstr = NULL; /* separator string for %v */ 12236 STRLEN dotstrlen; /* length of separator string for %v */ 12237 12238 Size_t efix = 0; /* explicit format parameter index */ 12239 const Size_t osvix = svix; /* original index in case of bad fmt */ 12240 12241 SV *argsv = NULL; 12242 bool is_utf8 = FALSE; /* is this item utf8? */ 12243 bool arg_missing = FALSE; /* give "Missing argument" warning */ 12244 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */ 12245 STRLEN esignlen = 0; /* length of e.g. "-0x" */ 12246 STRLEN zeros = 0; /* how many '0' to prepend */ 12247 12248 const char *eptr = NULL; /* the address of the element string */ 12249 STRLEN elen = 0; /* the length of the element string */ 12250 12251 char c; /* the actual format ('d', s' etc) */ 12252 12253 bool escape_it = FALSE; /* if this is a string should we quote and escape it? */ 12254 12255 12256 /* echo everything up to the next format specification */ 12257 for (q = fmtstart; q < patend && *q != '%'; ++q) 12258 {}; 12259 12260 if (q > fmtstart) { 12261 if (has_utf8 && !pat_utf8) { 12262 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on 12263 * the fly */ 12264 const char *p; 12265 char *dst; 12266 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1; 12267 12268 for (p = fmtstart; p < q; p++) 12269 if (!NATIVE_BYTE_IS_INVARIANT(*p)) 12270 need++; 12271 SvGROW(sv, need); 12272 12273 dst = SvEND(sv); 12274 for (p = fmtstart; p < q; p++) 12275 append_utf8_from_native_byte((U8)*p, (U8**)&dst); 12276 *dst = '\0'; 12277 SvCUR_set(sv, need - 1); 12278 } 12279 else 12280 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart); 12281 } 12282 if (q++ >= patend) 12283 break; 12284 12285 fmtstart = q; /* fmtstart is char following the '%' */ 12286 12287/* 12288 We allow format specification elements in this order: 12289 \d+\$ explicit format parameter index 12290 [-+ 0#]+ flags 12291 v|\*(\d+\$)?v vector with optional (optionally specified) arg 12292 0 flag (as above): repeated to allow "v02" 12293 \d+|\*(\d+\$)? width using optional (optionally specified) arg 12294 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg 12295 [hlqLV] size 12296 [%bcdefginopsuxDFOUX] format (mandatory) 12297*/ 12298 12299 if (inRANGE(*q, '1', '9')) { 12300 width = expect_number(&q); 12301 if (*q == '$') { 12302 if (args) 12303 Perl_croak_nocontext( 12304 "Cannot yet reorder sv_vcatpvfn() arguments from va_list"); 12305 ++q; 12306 efix = (Size_t)width; 12307 width = 0; 12308 no_redundant_warning = TRUE; 12309 } else { 12310 goto gotwidth; 12311 } 12312 } 12313 12314 /* FLAGS */ 12315 12316 while (*q) { 12317 switch (*q) { 12318 case ' ': 12319 case '+': 12320 if (plus == '+' && *q == ' ') /* '+' over ' ' */ 12321 q++; 12322 else 12323 plus = *q++; 12324 continue; 12325 12326 case '-': 12327 left = TRUE; 12328 q++; 12329 continue; 12330 12331 case '0': 12332 fill = TRUE; 12333 q++; 12334 continue; 12335 12336 case '#': 12337 alt = TRUE; 12338 q++; 12339 continue; 12340 12341 default: 12342 break; 12343 } 12344 break; 12345 } 12346 12347 /* at this point we can expect one of: 12348 * 12349 * 123 an explicit width 12350 * * width taken from next arg 12351 * *12$ width taken from 12th arg 12352 * or no width 12353 * 12354 * But any width specification may be preceded by a v, in one of its 12355 * forms: 12356 * v 12357 * *v 12358 * *12$v 12359 * So an asterisk may be either a width specifier or a vector 12360 * separator arg specifier, and we don't know which initially 12361 */ 12362 12363 tryasterisk: 12364 if (*q == '*') { 12365 STRLEN ix; /* explicit width/vector separator index */ 12366 q++; 12367 if (inRANGE(*q, '1', '9')) { 12368 ix = expect_number(&q); 12369 if (*q++ == '$') { 12370 if (args) 12371 Perl_croak_nocontext( 12372 "Cannot yet reorder sv_vcatpvfn() arguments from va_list"); 12373 no_redundant_warning = TRUE; 12374 } else 12375 goto unknown; 12376 } 12377 else 12378 ix = 0; 12379 12380 if (*q == 'v') { 12381 SV *vecsv; 12382 /* The asterisk was for *v, *NNN$v: vectorizing, but not 12383 * with the default "." */ 12384 q++; 12385 if (vectorize) 12386 goto unknown; 12387 if (args) 12388 vecsv = va_arg(*args, SV*); 12389 else { 12390 ix = ix ? ix - 1 : svix++; 12391 vecsv = ix < sv_count ? svargs[ix] 12392 : (arg_missing = TRUE, &PL_sv_no); 12393 } 12394 dotstr = SvPV_const(vecsv, dotstrlen); 12395 /* Keep the DO_UTF8 test *after* the SvPV call, else things go 12396 bad with tied or overloaded values that return UTF8. */ 12397 if (DO_UTF8(vecsv)) 12398 is_utf8 = TRUE; 12399 else if (has_utf8) { 12400 vecsv = sv_mortalcopy(vecsv); 12401 sv_utf8_upgrade(vecsv); 12402 dotstr = SvPV_const(vecsv, dotstrlen); 12403 is_utf8 = TRUE; 12404 } 12405 vectorize = TRUE; 12406 goto tryasterisk; 12407 } 12408 12409 /* the asterisk specified a width */ 12410 { 12411 int i = 0; 12412 SV *width_sv = NULL; 12413 if (args) 12414 i = va_arg(*args, int); 12415 else { 12416 ix = ix ? ix - 1 : svix++; 12417 width_sv = (ix < sv_count) ? svargs[ix] 12418 : (arg_missing = TRUE, (SV*)NULL); 12419 } 12420 width = S_sprintf_arg_num_val(aTHX_ args, i, width_sv, &left); 12421 } 12422 } 12423 else if (*q == 'v') { 12424 q++; 12425 if (vectorize) 12426 goto unknown; 12427 vectorize = TRUE; 12428 dotstr = "."; 12429 dotstrlen = 1; 12430 goto tryasterisk; 12431 12432 } 12433 else { 12434 /* explicit width? */ 12435 if(*q == '0') { 12436 fill = TRUE; 12437 q++; 12438 } 12439 if (inRANGE(*q, '1', '9')) 12440 width = expect_number(&q); 12441 } 12442 12443 gotwidth: 12444 12445 /* PRECISION */ 12446 12447 if (*q == '.') { 12448 q++; 12449 if (*q == '*') { 12450 STRLEN ix; /* explicit precision index */ 12451 q++; 12452 if (inRANGE(*q, '1', '9')) { 12453 ix = expect_number(&q); 12454 if (*q++ == '$') { 12455 if (args) 12456 Perl_croak_nocontext( 12457 "Cannot yet reorder sv_vcatpvfn() arguments from va_list"); 12458 no_redundant_warning = TRUE; 12459 } else 12460 goto unknown; 12461 } 12462 else 12463 ix = 0; 12464 12465 { 12466 int i = 0; 12467 SV *width_sv = NULL; 12468 bool neg = FALSE; 12469 12470 if (args) 12471 i = va_arg(*args, int); 12472 else { 12473 ix = ix ? ix - 1 : svix++; 12474 width_sv = (ix < sv_count) ? svargs[ix] 12475 : (arg_missing = TRUE, (SV*)NULL); 12476 } 12477 precis = S_sprintf_arg_num_val(aTHX_ args, i, width_sv, &neg); 12478 has_precis = !neg; 12479 /* ignore negative precision */ 12480 if (!has_precis) 12481 precis = 0; 12482 } 12483 } 12484 else { 12485 /* although it doesn't seem documented, this code has long 12486 * behaved so that: 12487 * no digits following the '.' is treated like '.0' 12488 * the number may be preceded by any number of zeroes, 12489 * e.g. "%.0001f", which is the same as "%.1f" 12490 * so I've kept that behaviour. DAPM May 2017 12491 */ 12492 while (*q == '0') 12493 q++; 12494 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0; 12495 has_precis = TRUE; 12496 } 12497 } 12498 12499 /* SIZE */ 12500 12501 switch (*q) { 12502#ifdef WIN32 12503 case 'I': /* Ix, I32x, and I64x */ 12504# ifdef USE_64_BIT_INT 12505 if (q[1] == '6' && q[2] == '4') { 12506 q += 3; 12507 intsize = 'q'; 12508 break; 12509 } 12510# endif 12511 if (q[1] == '3' && q[2] == '2') { 12512 q += 3; 12513 break; 12514 } 12515# ifdef USE_64_BIT_INT 12516 intsize = 'q'; 12517# endif 12518 q++; 12519 break; 12520#endif 12521#if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \ 12522 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE)) 12523 case 'L': /* Ld */ 12524 /* FALLTHROUGH */ 12525# if IVSIZE >= 8 12526 case 'q': /* qd */ 12527# endif 12528 intsize = 'q'; 12529 q++; 12530 break; 12531#endif 12532 case 'l': 12533 ++q; 12534#if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \ 12535 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE)) 12536 if (*q == 'l') { /* lld, llf */ 12537 intsize = 'q'; 12538 ++q; 12539 } 12540 else 12541#endif 12542 intsize = 'l'; 12543 break; 12544 case 'h': 12545 if (*++q == 'h') { /* hhd, hhu */ 12546 intsize = 'c'; 12547 ++q; 12548 } 12549 else 12550 intsize = 'h'; 12551 break; 12552#ifdef USE_QUADMATH 12553 case 'Q': 12554#endif 12555 case 'V': 12556 case 'z': 12557 case 't': 12558 case 'j': 12559 intsize = *q++; 12560 break; 12561 } 12562 12563 /* CONVERSION */ 12564 12565 c = *q++; /* c now holds the conversion type */ 12566 12567 /* '%' doesn't have an arg, so skip arg processing */ 12568 if (c == '%') { 12569 eptr = q - 1; 12570 elen = 1; 12571 if (vectorize) 12572 goto unknown; 12573 goto string; 12574 } 12575 12576 if (vectorize && !memCHRs("BbDdiOouUXx", c)) 12577 goto unknown; 12578 12579 /* get next arg (individual branches do their own va_arg() 12580 * handling for the args case) */ 12581 12582 if (!args) { 12583 efix = efix ? efix - 1 : svix++; 12584 argsv = efix < sv_count ? svargs[efix] 12585 : (arg_missing = TRUE, &PL_sv_no); 12586 } 12587 12588 12589 switch (c) { 12590 12591 /* STRINGS */ 12592 12593 case 's': 12594 if (args) { 12595 eptr = va_arg(*args, char*); 12596 if (eptr) 12597 if (has_precis) 12598 elen = my_strnlen(eptr, precis); 12599 else 12600 elen = strlen(eptr); 12601 else { 12602 eptr = (char *)nullstr; 12603 elen = sizeof nullstr - 1; 12604 } 12605 } 12606 else { 12607 eptr = SvPV_const(argsv, elen); 12608 if (DO_UTF8(argsv)) { 12609 STRLEN old_precis = precis; 12610 if (has_precis && precis < elen) { 12611 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen); 12612 STRLEN p = precis > ulen ? ulen : precis; 12613 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0); 12614 /* sticks at end */ 12615 } 12616 if (width) { /* fudge width (can't fudge elen) */ 12617 if (has_precis && precis < elen) 12618 width += precis - old_precis; 12619 else 12620 width += 12621 elen - sv_or_pv_len_utf8(argsv,eptr,elen); 12622 } 12623 is_utf8 = TRUE; 12624 } 12625 } 12626 12627 string: 12628 if (escape_it) { 12629 U32 flags = PERL_PV_PRETTY_QUOTEDPREFIX; 12630 if (is_utf8) 12631 flags |= PERL_PV_ESCAPE_UNI; 12632 12633 if (!tmp_sv) { 12634 /* "blah"... where blah might be made up 12635 * of characters like \x{1234} */ 12636 tmp_sv = newSV(1 + (PERL_QUOTEDPREFIX_LEN * 8) + 1 + 3); 12637 sv_2mortal(tmp_sv); 12638 } 12639 pv_pretty(tmp_sv, eptr, elen, PERL_QUOTEDPREFIX_LEN, 12640 NULL, NULL, flags); 12641 eptr = SvPV_const(tmp_sv, elen); 12642 } 12643 if (has_precis && precis < elen) 12644 elen = precis; 12645 break; 12646 12647 /* INTEGERS */ 12648 12649 case 'p': 12650 12651 /* BEGIN NOTE 12652 * 12653 * We want to extend the C level sprintf format API with 12654 * custom formats for specific types (eg SV*) and behavior. 12655 * However some C compilers are "sprintf aware" and will 12656 * throw compile time exceptions when an illegal sprintf is 12657 * encountered, so we can't just add new format letters. 12658 * 12659 * However it turns out the length argument to the %p format 12660 * is more or less useless (the size of a pointer does not 12661 * change over time) and is not really used in the C level 12662 * code. Accordingly we can map our special behavior to 12663 * specific "length" options to the %p format. We hide these 12664 * mappings behind defines anyway, so nobody needs to know 12665 * that HEKf is actually %2p. This keeps the C compiler 12666 * happy while allowing us to add new formats. 12667 * 12668 * Note the existing logic for which number is used for what 12669 * is torturous. All negative values are used for SVf, and 12670 * non-negative values have arbitrary meanings with no 12671 * structure to them. This may change in the future. 12672 * 12673 * NEVER use the raw %p values directly. Always use the define 12674 * as the underlying mapping may change in the future. 12675 * 12676 * END NOTE 12677 * 12678 * %p extensions: 12679 * 12680 * "%...p" is normally treated like "%...x", except that the 12681 * number to print is the SV's address (or a pointer address 12682 * for C-ish sprintf). 12683 * 12684 * However, the C-ish sprintf variant allows a few special 12685 * extensions. These are currently: 12686 * 12687 * %-p (SVf) Like %s, but gets the string from an SV* 12688 * arg rather than a char* arg. Use C<SVfARG()> 12689 * to set up the argument properly. 12690 * (This was previously %_). 12691 * 12692 * %-<num>p Ditto but like %.<num>s (i.e. num is max 12693 * width), there is no escaped and quoted version 12694 * of this. 12695 * 12696 * %1p (PVf_QUOTEDPREFIX). Like raw %s, but it is escaped 12697 * and quoted. 12698 * 12699 * %5p (SVf_QUOTEDPREFIX) Like SVf, but length restricted, 12700 * escaped and quoted with pv_pretty. Intended 12701 * for error messages. 12702 * 12703 * %2p (HEKf) Like %s, but using the key string in a HEK 12704 * %7p (HEKf_QUOTEDPREFIX) ... but escaped and quoted. 12705 * 12706 * %3p (HEKf256) Ditto but like %.256s 12707 * %8p (HEKf256_QUOTEDPREFIX) ... but escaped and quoted 12708 * 12709 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args: 12710 * (cBOOL(utf8), len, string_buf). 12711 * It's handled by the "case 'd'" branch 12712 * rather than here. 12713 * %d%lu%9p (UTF8f_QUOTEDPREFIX) .. but escaped and quoted. 12714 * 12715 * %6p (HvNAMEf) Like %s, but using the HvNAME() and HvNAMELEN() 12716 * %10p (HvNAMEf_QUOTEDPREFIX) ... but escaped and quoted 12717 * 12718 * %<num>p where num is > 9: reserved for future 12719 * extensions. Warns, but then is treated as a 12720 * general %p (print hex address) format. 12721 * 12722 * NOTE: If you add a new magic %p value you will 12723 * need to update F<t/porting/diag.t> to be aware of it 12724 * on top of adding the various defines and etc. Do not 12725 * forget to add it to F<pod/perlguts.pod> as well. 12726 */ 12727 12728 if ( args 12729 && !intsize 12730 && !fill 12731 && !plus 12732 && !has_precis 12733 /* not %*p or %*1$p - any width was explicit */ 12734 && q[-2] != '*' 12735 && q[-2] != '$' 12736 ) { 12737 if (left || width == 5) { /* %-p (SVf), %-NNNp, %5p */ 12738 if (left && width) { 12739 precis = width; 12740 has_precis = TRUE; 12741 } else if (width == 5) { 12742 escape_it = TRUE; 12743 } 12744 argsv = MUTABLE_SV(va_arg(*args, void*)); 12745 eptr = SvPV_const(argsv, elen); 12746 if (DO_UTF8(argsv)) 12747 is_utf8 = TRUE; 12748 width = 0; 12749 goto string; 12750 } 12751 else if (width == 2 || width == 3 || 12752 width == 7 || width == 8) 12753 { /* HEKf, HEKf256, HEKf_QUOTEDPREFIX, HEKf256_QUOTEDPREFIX */ 12754 HEK * const hek = va_arg(*args, HEK *); 12755 eptr = HEK_KEY(hek); 12756 elen = HEK_LEN(hek); 12757 if (HEK_UTF8(hek)) 12758 is_utf8 = TRUE; 12759 if (width == 3) { 12760 precis = 256; 12761 has_precis = TRUE; 12762 } 12763 if (width > 5) 12764 escape_it = TRUE; 12765 width = 0; 12766 goto string; 12767 } 12768 else if (width == 1) { 12769 eptr = va_arg(*args,char *); 12770 elen = strlen(eptr); 12771 escape_it = TRUE; 12772 width = 0; 12773 goto string; 12774 } 12775 else if (width == 6 || width == 10) { 12776 HV *hv = va_arg(*args, HV *); 12777 eptr = HvNAME(hv); 12778 elen = HvNAMELEN(hv); 12779 if (HvNAMEUTF8(hv)) 12780 is_utf8 = TRUE; 12781 if (width == 10) 12782 escape_it = TRUE; 12783 width = 0; 12784 goto string; 12785 } 12786 else if (width) { 12787 /* note width=4 or width=9 is handled under %d */ 12788 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), 12789 "internal %%<num>p might conflict with future printf extensions"); 12790 } 12791 } 12792 12793 /* treat as normal %...p */ 12794 12795 uv = PTR2UV(args ? va_arg(*args, void*) : argsv); 12796 base = 16; 12797 c = 'x'; /* in case the format string contains '#' */ 12798 goto do_integer; 12799 12800 case 'c': 12801 /* Ignore any size specifiers, since they're not documented as 12802 * being allowed for %c (ideally we should warn on e.g. '%hc'). 12803 * Setting a default intsize, along with a positive 12804 * (which signals unsigned) base, causes, for C-ish use, the 12805 * va_arg to be interpreted as an unsigned int, when it's 12806 * actually signed, which will convert -ve values to high +ve 12807 * values. Note that unlike the libc %c, values > 255 will 12808 * convert to high unicode points rather than being truncated 12809 * to 8 bits. For perlish use, it will do SvUV(argsv), which 12810 * will again convert -ve args to high -ve values. 12811 */ 12812 intsize = 0; 12813 base = 1; /* special value that indicates we're doing a 'c' */ 12814 goto get_int_arg_val; 12815 12816 case 'D': 12817#ifdef IV_IS_QUAD 12818 intsize = 'q'; 12819#else 12820 intsize = 'l'; 12821#endif 12822 base = -10; 12823 goto get_int_arg_val; 12824 12825 case 'd': 12826 /* probably just a plain %d, but it might be the start of the 12827 * special UTF8f format, which usually looks something like 12828 * "%d%lu%4p" (the lu may vary by platform) or 12829 * "%d%lu%9p" for an escaped version. 12830 */ 12831 assert((UTF8f)[0] == 'd'); 12832 assert((UTF8f)[1] == '%'); 12833 12834 if ( args /* UTF8f only valid for C-ish sprintf */ 12835 && q == fmtstart + 1 /* plain %d, not %....d */ 12836 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */ 12837 && *q == '%' 12838 && strnEQ(q + 1, (UTF8f) + 2, sizeof(UTF8f) - 5) 12839 && q[sizeof(UTF8f)-3] == 'p' 12840 && (q[sizeof(UTF8f)-4] == '4' || 12841 q[sizeof(UTF8f)-4] == '9')) 12842 { 12843 /* The argument has already gone through cBOOL, so the cast 12844 is safe. */ 12845 if (q[sizeof(UTF8f)-4] == '9') 12846 escape_it = TRUE; 12847 is_utf8 = (bool)va_arg(*args, int); 12848 elen = va_arg(*args, UV); 12849 /* if utf8 length is larger than 0x7ffff..., then it might 12850 * have been a signed value that wrapped */ 12851 if (elen > ((~(STRLEN)0) >> 1)) { 12852 assert(0); /* in DEBUGGING build we want to crash */ 12853 elen = 0; /* otherwise we want to treat this as an empty string */ 12854 } 12855 eptr = va_arg(*args, char *); 12856 q += sizeof(UTF8f) - 2; 12857 goto string; 12858 } 12859 12860 /* FALLTHROUGH */ 12861 case 'i': 12862 base = -10; 12863 goto get_int_arg_val; 12864 12865 case 'U': 12866#ifdef IV_IS_QUAD 12867 intsize = 'q'; 12868#else 12869 intsize = 'l'; 12870#endif 12871 /* FALLTHROUGH */ 12872 case 'u': 12873 base = 10; 12874 goto get_int_arg_val; 12875 12876 case 'B': 12877 case 'b': 12878 base = 2; 12879 goto get_int_arg_val; 12880 12881 case 'O': 12882#ifdef IV_IS_QUAD 12883 intsize = 'q'; 12884#else 12885 intsize = 'l'; 12886#endif 12887 /* FALLTHROUGH */ 12888 case 'o': 12889 base = 8; 12890 goto get_int_arg_val; 12891 12892 case 'X': 12893 case 'x': 12894 base = 16; 12895 12896 get_int_arg_val: 12897 12898 if (vectorize) { 12899 STRLEN ulen; 12900 SV *vecsv; 12901 12902 if (base < 0) { 12903 base = -base; 12904 if (plus) 12905 esignbuf[esignlen++] = plus; 12906 } 12907 12908 /* initialise the vector string to iterate over */ 12909 12910 vecsv = args ? va_arg(*args, SV*) : argsv; 12911 12912 /* if this is a version object, we need to convert 12913 * back into v-string notation and then let the 12914 * vectorize happen normally 12915 */ 12916 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) { 12917 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) { 12918 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF), 12919 "vector argument not supported with alpha versions"); 12920 vecsv = &PL_sv_no; 12921 } 12922 else { 12923 vecstr = (U8*)SvPV_const(vecsv,veclen); 12924 vecsv = sv_newmortal(); 12925 scan_vstring((char *)vecstr, (char *)vecstr + veclen, 12926 vecsv); 12927 } 12928 } 12929 vecstr = (U8*)SvPV_const(vecsv, veclen); 12930 vec_utf8 = DO_UTF8(vecsv); 12931 12932 /* This is the re-entry point for when we're iterating 12933 * over the individual characters of a vector arg */ 12934 vector: 12935 if (!veclen) 12936 goto done_valid_conversion; 12937 if (vec_utf8) 12938 uv = utf8n_to_uvchr(vecstr, veclen, &ulen, 12939 UTF8_ALLOW_ANYUV); 12940 else { 12941 uv = *vecstr; 12942 ulen = 1; 12943 } 12944 vecstr += ulen; 12945 veclen -= ulen; 12946 } 12947 else { 12948 /* test arg for inf/nan. This can trigger an unwanted 12949 * 'str' overload, so manually force 'num' overload first 12950 * if necessary */ 12951 if (argsv) { 12952 SvGETMAGIC(argsv); 12953 if (UNLIKELY(SvAMAGIC(argsv))) 12954 argsv = sv_2num(argsv); 12955 if (UNLIKELY(isinfnansv(argsv))) 12956 goto handle_infnan_argsv; 12957 } 12958 12959 if (base < 0) { 12960 /* signed int type */ 12961 IV iv; 12962 base = -base; 12963 if (args) { 12964 switch (intsize) { 12965 case 'c': iv = (char)va_arg(*args, int); break; 12966 case 'h': iv = (short)va_arg(*args, int); break; 12967 case 'l': iv = va_arg(*args, long); break; 12968 case 'V': iv = va_arg(*args, IV); break; 12969 case 'z': iv = va_arg(*args, SSize_t); break; 12970#ifdef HAS_PTRDIFF_T 12971 case 't': iv = va_arg(*args, ptrdiff_t); break; 12972#endif 12973 default: iv = va_arg(*args, int); break; 12974 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break; 12975 case 'q': 12976#if IVSIZE >= 8 12977 iv = va_arg(*args, Quad_t); break; 12978#else 12979 goto unknown; 12980#endif 12981 } 12982 } 12983 else { 12984 /* assign to tiv then cast to iv to work around 12985 * 2003 GCC cast bug (gnu.org bugzilla #13488) */ 12986 IV tiv = SvIV_nomg(argsv); 12987 switch (intsize) { 12988 case 'c': iv = (char)tiv; break; 12989 case 'h': iv = (short)tiv; break; 12990 case 'l': iv = (long)tiv; break; 12991 case 'V': 12992 default: iv = tiv; break; 12993 case 'q': 12994#if IVSIZE >= 8 12995 iv = (Quad_t)tiv; break; 12996#else 12997 goto unknown; 12998#endif 12999 } 13000 } 13001 13002 /* now convert iv to uv */ 13003 if (iv >= 0) { 13004 uv = iv; 13005 if (plus) 13006 esignbuf[esignlen++] = plus; 13007 } 13008 else { 13009 /* Using 0- here to silence bogus warning from MS VC */ 13010 uv = (UV) (0 - (UV) iv); 13011 esignbuf[esignlen++] = '-'; 13012 } 13013 } 13014 else { 13015 /* unsigned int type */ 13016 if (args) { 13017 switch (intsize) { 13018 case 'c': uv = (unsigned char)va_arg(*args, unsigned); 13019 break; 13020 case 'h': uv = (unsigned short)va_arg(*args, unsigned); 13021 break; 13022 case 'l': uv = va_arg(*args, unsigned long); break; 13023 case 'V': uv = va_arg(*args, UV); break; 13024 case 'z': uv = va_arg(*args, Size_t); break; 13025#ifdef HAS_PTRDIFF_T 13026 /* will sign extend, but there is no 13027 * uptrdiff_t, so oh well */ 13028 case 't': uv = va_arg(*args, ptrdiff_t); break; 13029#endif 13030 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break; 13031 default: uv = va_arg(*args, unsigned); break; 13032 case 'q': 13033#if IVSIZE >= 8 13034 uv = va_arg(*args, Uquad_t); break; 13035#else 13036 goto unknown; 13037#endif 13038 } 13039 } 13040 else { 13041 /* assign to tiv then cast to iv to work around 13042 * 2003 GCC cast bug (gnu.org bugzilla #13488) */ 13043 UV tuv = SvUV_nomg(argsv); 13044 switch (intsize) { 13045 case 'c': uv = (unsigned char)tuv; break; 13046 case 'h': uv = (unsigned short)tuv; break; 13047 case 'l': uv = (unsigned long)tuv; break; 13048 case 'V': 13049 default: uv = tuv; break; 13050 case 'q': 13051#if IVSIZE >= 8 13052 uv = (Uquad_t)tuv; break; 13053#else 13054 goto unknown; 13055#endif 13056 } 13057 } 13058 } 13059 } 13060 13061 do_integer: 13062 { 13063 char *ptr = ebuf + sizeof ebuf; 13064 unsigned dig; 13065 zeros = 0; 13066 13067 switch (base) { 13068 case 16: 13069 { 13070 const char * const p = 13071 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit; 13072 13073 do { 13074 dig = uv & 15; 13075 *--ptr = p[dig]; 13076 } while (uv >>= 4); 13077 if (alt && *ptr != '0') { 13078 esignbuf[esignlen++] = '0'; 13079 esignbuf[esignlen++] = c; /* 'x' or 'X' */ 13080 } 13081 break; 13082 } 13083 case 8: 13084 do { 13085 dig = uv & 7; 13086 *--ptr = '0' + dig; 13087 } while (uv >>= 3); 13088 if (alt && *ptr != '0') 13089 *--ptr = '0'; 13090 break; 13091 case 2: 13092 do { 13093 dig = uv & 1; 13094 *--ptr = '0' + dig; 13095 } while (uv >>= 1); 13096 if (alt && *ptr != '0') { 13097 esignbuf[esignlen++] = '0'; 13098 esignbuf[esignlen++] = c; /* 'b' or 'B' */ 13099 } 13100 break; 13101 13102 case 1: 13103 /* special-case: base 1 indicates a 'c' format: 13104 * we use the common code for extracting a uv, 13105 * but handle that value differently here than 13106 * all the other int types */ 13107 if ((uv > 255 || 13108 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv))) 13109 && !IN_BYTES) 13110 { 13111 STATIC_ASSERT_STMT(sizeof(ebuf) >= UTF8_MAXBYTES + 1); 13112 eptr = ebuf; 13113 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf; 13114 is_utf8 = TRUE; 13115 } 13116 else { 13117 eptr = ebuf; 13118 ebuf[0] = (char)uv; 13119 elen = 1; 13120 } 13121 goto string; 13122 13123 default: /* it had better be ten or less */ 13124 do { 13125 dig = uv % base; 13126 *--ptr = '0' + dig; 13127 } while (uv /= base); 13128 break; 13129 } 13130 elen = (ebuf + sizeof ebuf) - ptr; 13131 eptr = ptr; 13132 if (has_precis) { 13133 if (precis > elen) 13134 zeros = precis - elen; 13135 else if (precis == 0 && elen == 1 && *eptr == '0' 13136 && !(base == 8 && alt)) /* "%#.0o" prints "0" */ 13137 elen = 0; 13138 13139 /* a precision nullifies the 0 flag. */ 13140 fill = FALSE; 13141 } 13142 } 13143 break; 13144 13145 /* FLOATING POINT */ 13146 13147 case 'F': 13148 c = 'f'; /* maybe %F isn't supported here */ 13149 /* FALLTHROUGH */ 13150 case 'e': case 'E': 13151 case 'f': 13152 case 'g': case 'G': 13153 case 'a': case 'A': 13154 13155 { 13156 STRLEN float_need; /* what PL_efloatsize needs to become */ 13157 bool hexfp; /* hexadecimal floating point? */ 13158 13159 vcatpvfn_long_double_t fv; 13160 NV nv; 13161 13162 /* This is evil, but floating point is even more evil */ 13163 13164 /* for SV-style calling, we can only get NV 13165 for C-style calling, we assume %f is double; 13166 for simplicity we allow any of %Lf, %llf, %qf for long double 13167 */ 13168 switch (intsize) { 13169#if defined(USE_QUADMATH) 13170 case 'Q': 13171 break; 13172#endif 13173 case 'V': 13174#if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH) 13175 intsize = 'q'; 13176#endif 13177 break; 13178/* [perl #20339] - we should accept and ignore %lf rather than die */ 13179 case 'l': 13180 /* FALLTHROUGH */ 13181 default: 13182#if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH) 13183 intsize = args ? 0 : 'q'; 13184#endif 13185 break; 13186 case 'q': 13187#if defined(HAS_LONG_DOUBLE) 13188 break; 13189#else 13190 /* FALLTHROUGH */ 13191#endif 13192 case 'c': 13193 case 'h': 13194 case 'z': 13195 case 't': 13196 case 'j': 13197 goto unknown; 13198 } 13199 13200 /* Now we need (long double) if intsize == 'q', else (double). */ 13201 if (args) { 13202 /* Note: do not pull NVs off the va_list with va_arg() 13203 * (pull doubles instead) because if you have a build 13204 * with long doubles, you would always be pulling long 13205 * doubles, which would badly break anyone using only 13206 * doubles (i.e. the majority of builds). In other 13207 * words, you cannot mix doubles and long doubles. 13208 * The only case where you can pull off long doubles 13209 * is when the format specifier explicitly asks so with 13210 * e.g. "%Lg". */ 13211#ifdef USE_QUADMATH 13212 nv = intsize == 'Q' ? va_arg(*args, NV) : 13213 intsize == 'q' ? va_arg(*args, long double) : 13214 va_arg(*args, double); 13215 fv = nv; 13216#elif LONG_DOUBLESIZE > DOUBLESIZE 13217 if (intsize == 'q') { 13218 fv = va_arg(*args, long double); 13219 nv = fv; 13220 } else { 13221 nv = va_arg(*args, double); 13222 VCATPVFN_NV_TO_FV(nv, fv); 13223 } 13224#else 13225 nv = va_arg(*args, double); 13226 fv = nv; 13227#endif 13228 } 13229 else 13230 { 13231 SvGETMAGIC(argsv); 13232 /* we jump here if an int-ish format encountered an 13233 * infinite/Nan argsv. After setting nv/fv, it falls 13234 * into the isinfnan block which follows */ 13235 handle_infnan_argsv: 13236 nv = SvNV_nomg(argsv); 13237 VCATPVFN_NV_TO_FV(nv, fv); 13238 } 13239 13240 if (Perl_isinfnan(nv)) { 13241 if (c == 'c') 13242 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'", 13243 nv, (int)c); 13244 13245 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus); 13246 assert(elen); 13247 eptr = ebuf; 13248 zeros = 0; 13249 esignlen = 0; 13250 dotstrlen = 0; 13251 break; 13252 } 13253 13254 /* special-case "%.0f" */ 13255 if ( c == 'f' 13256 && !precis 13257 && has_precis 13258 && !(width || left || plus || alt) 13259 && !fill 13260 && intsize != 'q' 13261 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen))) 13262 ) 13263 goto float_concat; 13264 13265 /* Determine the buffer size needed for the various 13266 * floating-point formats. 13267 * 13268 * The basic possibilities are: 13269 * 13270 * <---P---> 13271 * %f 1111111.123456789 13272 * %e 1.111111123e+06 13273 * %a 0x1.0f4471f9bp+20 13274 * %g 1111111.12 13275 * %g 1.11111112e+15 13276 * 13277 * where P is the value of the precision in the format, or 6 13278 * if not specified. Note the two possible output formats of 13279 * %g; in both cases the number of significant digits is <= 13280 * precision. 13281 * 13282 * For most of the format types the maximum buffer size needed 13283 * is precision, plus: any leading 1 or 0x1, the radix 13284 * point, and an exponent. The difficult one is %f: for a 13285 * large positive exponent it can have many leading digits, 13286 * which needs to be calculated specially. Also %a is slightly 13287 * different in that in the absence of a specified precision, 13288 * it uses as many digits as necessary to distinguish 13289 * different values. 13290 * 13291 * First, here are the constant bits. For ease of calculation 13292 * we over-estimate the needed buffer size, for example by 13293 * assuming all formats have an exponent and a leading 0x1. 13294 * 13295 * Also for production use, add a little extra overhead for 13296 * safety's sake. Under debugging don't, as it means we're 13297 * more likely to quickly spot issues during development. 13298 */ 13299 13300 float_need = 1 /* possible unary minus */ 13301 + 4 /* "0x1" plus very unlikely carry */ 13302 + 1 /* default radix point '.' */ 13303 + 2 /* "e-", "p+" etc */ 13304 + 6 /* exponent: up to 16383 (quad fp) */ 13305#ifndef DEBUGGING 13306 + 20 /* safety net */ 13307#endif 13308 + 1; /* \0 */ 13309 13310 13311 /* determine the radix point len, e.g. length(".") in "1.2" */ 13312#ifdef USE_LOCALE_NUMERIC 13313 /* note that we may either explicitly use PL_numeric_radix_sv 13314 * below, or implicitly, via an snprintf() variant. 13315 * Note also things like ps_AF.utf8 which has 13316 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */ 13317 if (! have_in_lc_numeric) { 13318 in_lc_numeric = IN_LC(LC_NUMERIC); 13319 have_in_lc_numeric = TRUE; 13320 } 13321 13322 if (in_lc_numeric) { 13323 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, { 13324 /* this can't wrap unless PL_numeric_radix_sv is a string 13325 * consuming virtually all the 32-bit or 64-bit address 13326 * space 13327 */ 13328 float_need += (SvCUR(PL_numeric_radix_sv) - 1); 13329 13330 /* floating-point formats only get utf8 if the radix point 13331 * is utf8. All other characters in the string are < 128 13332 * and so can be safely appended to both a non-utf8 and utf8 13333 * string as-is. 13334 * Note that this will convert the output to utf8 even if 13335 * the radix point didn't get output. 13336 */ 13337 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) { 13338 sv_utf8_upgrade(sv); 13339 has_utf8 = TRUE; 13340 } 13341 }); 13342 } 13343#endif 13344 13345 hexfp = FALSE; 13346 13347 if (isALPHA_FOLD_EQ(c, 'f')) { 13348 /* Determine how many digits before the radix point 13349 * might be emitted. frexp() (or frexpl) has some 13350 * unspecified behaviour for nan/inf/-inf, so lucky we've 13351 * already handled them above */ 13352 STRLEN digits; 13353 int i = PERL_INT_MIN; 13354 (void)Perl_frexp((NV)fv, &i); 13355 if (i == PERL_INT_MIN) 13356 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv); 13357 13358 if (i > 0) { 13359 digits = BIT_DIGITS(i); 13360 /* this can't overflow. 'digits' will only be a few 13361 * thousand even for the largest floating-point types. 13362 * And up until now float_need is just some small 13363 * constants plus radix len, which can't be in 13364 * overflow territory unless the radix SV is consuming 13365 * over 1/2 the address space */ 13366 assert(float_need < ((STRLEN)~0) - digits); 13367 float_need += digits; 13368 } 13369 } 13370 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) { 13371 hexfp = TRUE; 13372 if (!has_precis) { 13373 /* %a in the absence of precision may print as many 13374 * digits as needed to represent the entire mantissa 13375 * bit pattern. 13376 * This estimate seriously overshoots in most cases, 13377 * but better the undershooting. Firstly, all bytes 13378 * of the NV are not mantissa, some of them are 13379 * exponent. Secondly, for the reasonably common 13380 * long doubles case, the "80-bit extended", two 13381 * or six bytes of the NV are unused. Also, we'll 13382 * still pick up an extra +6 from the default 13383 * precision calculation below. */ 13384 STRLEN digits = 13385#ifdef LONGDOUBLE_DOUBLEDOUBLE 13386 /* For the "double double", we need more. 13387 * Since each double has their own exponent, the 13388 * doubles may float (haha) rather far from each 13389 * other, and the number of required bits is much 13390 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits. 13391 * See the definition of DOUBLEDOUBLE_MAXBITS. 13392 * 13393 * Need 2 hexdigits for each byte. */ 13394 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2; 13395#else 13396 NVSIZE * 2; /* 2 hexdigits for each byte */ 13397#endif 13398 /* see "this can't overflow" comment above */ 13399 assert(float_need < ((STRLEN)~0) - digits); 13400 float_need += digits; 13401 } 13402 } 13403 /* special-case "%.<number>g" if it will fit in ebuf */ 13404 else if (c == 'g' 13405 && precis /* See earlier comment about buggy Gconvert 13406 when digits, aka precis, is 0 */ 13407 && has_precis 13408 /* check that "%.<number>g" formatting will fit in ebuf */ 13409 && sizeof(ebuf) - float_need > precis 13410 /* sizeof(ebuf) - float_need will have wrapped if float_need > sizeof(ebuf). * 13411 * Therefore we should check that float_need < sizeof(ebuf). Normally, we would * 13412 * have run this check first, but that triggers incorrect -Wformat-overflow * 13413 * compilation warnings with some versions of gcc if Gconvert invokes sprintf(). * 13414 * ( See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89161 ) * 13415 * So, instead, we check it next: */ 13416 && float_need < sizeof(ebuf) 13417 && !(width || left || plus || alt) 13418 && !fill 13419 && intsize != 'q' 13420 ) { 13421 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric, 13422 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis) 13423 ); 13424 elen = strlen(ebuf); 13425 eptr = ebuf; 13426 goto float_concat; 13427 } 13428 13429 13430 { 13431 STRLEN pr = has_precis ? precis : 6; /* known default */ 13432 /* this probably can't wrap, since precis is limited 13433 * to 1/4 address space size, but better safe than sorry 13434 */ 13435 if (float_need >= ((STRLEN)~0) - pr) 13436 croak_memory_wrap(); 13437 float_need += pr; 13438 } 13439 13440 if (float_need < width) 13441 float_need = width; 13442 13443 if (float_need > INT_MAX) { 13444 /* snprintf() returns an int, and we use that return value, 13445 so die horribly if the expected size is too large for int 13446 */ 13447 Perl_croak(aTHX_ "Numeric format result too large"); 13448 } 13449 13450 if (PL_efloatsize <= float_need) { 13451 /* PL_efloatbuf should be at least 1 greater than 13452 * float_need to allow a trailing \0 to be returned by 13453 * snprintf(). If we need to grow, overgrow for the 13454 * benefit of future generations */ 13455 const STRLEN extra = 0x20; 13456 if (float_need >= ((STRLEN)~0) - extra) 13457 croak_memory_wrap(); 13458 float_need += extra; 13459 Safefree(PL_efloatbuf); 13460 PL_efloatsize = float_need; 13461 Newx(PL_efloatbuf, PL_efloatsize, char); 13462 PL_efloatbuf[0] = '\0'; 13463 } 13464 13465 if (UNLIKELY(hexfp)) { 13466 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c, 13467 nv, fv, has_precis, precis, width, 13468 alt, plus, left, fill, in_lc_numeric); 13469 } 13470 else { 13471 char *ptr = ebuf + sizeof ebuf; 13472 *--ptr = '\0'; 13473 *--ptr = c; 13474#if defined(USE_QUADMATH) 13475 /* always use Q here. my_snprint() throws an exception if we 13476 fallthrough to the double/long double code, even when the 13477 format is correct, presumably to avoid any accidentally 13478 missing Q. 13479 */ 13480 *--ptr = 'Q'; 13481 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */ 13482#elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl) 13483 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl, 13484 * not USE_LONG_DOUBLE and NVff. In other words, 13485 * this needs to work without USE_LONG_DOUBLE. */ 13486 if (intsize == 'q') { 13487 /* Copy the one or more characters in a long double 13488 * format before the 'base' ([efgEFG]) character to 13489 * the format string. */ 13490 static char const ldblf[] = PERL_PRIfldbl; 13491 char const *p = ldblf + sizeof(ldblf) - 3; 13492 while (p >= ldblf) { *--ptr = *p--; } 13493 } 13494#endif 13495 if (has_precis) { 13496 base = precis; 13497 do { *--ptr = '0' + (base % 10); } while (base /= 10); 13498 *--ptr = '.'; 13499 } 13500 if (width) { 13501 base = width; 13502 do { *--ptr = '0' + (base % 10); } while (base /= 10); 13503 } 13504 if (fill) 13505 *--ptr = '0'; 13506 if (left) 13507 *--ptr = '-'; 13508 if (plus) 13509 *--ptr = plus; 13510 if (alt) 13511 *--ptr = '#'; 13512 *--ptr = '%'; 13513 13514 /* No taint. Otherwise we are in the strange situation 13515 * where printf() taints but print($float) doesn't. 13516 * --jhi */ 13517 13518 /* hopefully the above makes ptr a very constrained format 13519 * that is safe to use, even though it's not literal */ 13520 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral); 13521#ifdef USE_QUADMATH 13522 { 13523 if (!quadmath_format_valid(ptr)) 13524 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr); 13525 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric, 13526 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize, 13527 ptr, nv); 13528 ); 13529 if ((IV)elen == -1) { 13530 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", ptr); 13531 } 13532 } 13533#elif defined(HAS_LONG_DOUBLE) 13534 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric, 13535 elen = ((intsize == 'q') 13536 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv) 13537 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv)) 13538 ); 13539#else 13540 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric, 13541 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv) 13542 ); 13543#endif 13544 GCC_DIAG_RESTORE_STMT; 13545 } 13546 13547 eptr = PL_efloatbuf; 13548 13549 float_concat: 13550 13551 /* Since floating-point formats do their own formatting and 13552 * padding, we skip the main block of code at the end of this 13553 * loop which handles appending eptr to sv, and do our own 13554 * stripped-down version */ 13555 13556 assert(!zeros); 13557 assert(!esignlen); 13558 assert(elen); 13559 assert(elen >= width); 13560 13561 S_sv_catpvn_simple(aTHX_ sv, eptr, elen); 13562 13563 goto done_valid_conversion; 13564 } 13565 13566 /* SPECIAL */ 13567 13568 case 'n': 13569 { 13570 STRLEN len; 13571 /* XXX ideally we should warn if any flags etc have been 13572 * set, e.g. "%-4.5n" */ 13573 /* XXX if sv was originally non-utf8 with a char in the 13574 * range 0x80-0xff, then if it got upgraded, we should 13575 * calculate char len rather than byte len here */ 13576 len = SvCUR(sv) - origlen; 13577 if (args) { 13578 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len; 13579 13580 switch (intsize) { 13581 case 'c': *(va_arg(*args, char*)) = i; break; 13582 case 'h': *(va_arg(*args, short*)) = i; break; 13583 default: *(va_arg(*args, int*)) = i; break; 13584 case 'l': *(va_arg(*args, long*)) = i; break; 13585 case 'V': *(va_arg(*args, IV*)) = i; break; 13586 case 'z': *(va_arg(*args, SSize_t*)) = i; break; 13587#ifdef HAS_PTRDIFF_T 13588 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break; 13589#endif 13590 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break; 13591 case 'q': 13592#if IVSIZE >= 8 13593 *(va_arg(*args, Quad_t*)) = i; break; 13594#else 13595 goto unknown; 13596#endif 13597 } 13598 } 13599 else { 13600 if (arg_missing) 13601 Perl_croak_nocontext( 13602 "Missing argument for %%n in %s", 13603 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); 13604 sv_setuv_mg(argsv, has_utf8 13605 ? (UV)utf8_length((U8*)SvPVX(sv), (U8*)SvEND(sv)) 13606 : (UV)len); 13607 } 13608 goto done_valid_conversion; 13609 } 13610 13611 /* UNKNOWN */ 13612 13613 default: 13614 unknown: 13615 if (!args 13616 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF) 13617 && ckWARN(WARN_PRINTF)) 13618 { 13619 SV * const msg = sv_newmortal(); 13620 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ", 13621 (PL_op->op_type == OP_PRTF) ? "" : "s"); 13622 if (fmtstart < patend) { 13623 const char * const fmtend = q < patend ? q : patend; 13624 const char * f; 13625 sv_catpvs(msg, "\"%"); 13626 for (f = fmtstart; f < fmtend; f++) { 13627 if (isPRINT(*f)) { 13628 sv_catpvn_nomg(msg, f, 1); 13629 } else { 13630 Perl_sv_catpvf(aTHX_ msg, "\\%03o", (U8) *f); 13631 } 13632 } 13633 sv_catpvs(msg, "\""); 13634 } else { 13635 sv_catpvs(msg, "end of string"); 13636 } 13637 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */ 13638 } 13639 13640 /* mangled format: output the '%', then continue from the 13641 * character following that */ 13642 sv_catpvn_nomg(sv, fmtstart-1, 1); 13643 q = fmtstart; 13644 svix = osvix; 13645 /* Any "redundant arg" warning from now onwards will probably 13646 * just be misleading, so don't bother. */ 13647 no_redundant_warning = TRUE; 13648 continue; /* not "break" */ 13649 } 13650 13651 if (is_utf8 != has_utf8) { 13652 if (is_utf8) { 13653 if (SvCUR(sv)) 13654 sv_utf8_upgrade(sv); 13655 } 13656 else { 13657 const STRLEN old_elen = elen; 13658 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP); 13659 sv_utf8_upgrade(nsv); 13660 eptr = SvPVX_const(nsv); 13661 elen = SvCUR(nsv); 13662 13663 if (width) { /* fudge width (can't fudge elen) */ 13664 width += elen - old_elen; 13665 } 13666 is_utf8 = TRUE; 13667 } 13668 } 13669 13670 13671 /* append esignbuf, filler, zeros, eptr and dotstr to sv */ 13672 13673 { 13674 STRLEN need, have, gap; 13675 STRLEN i; 13676 char *s; 13677 13678 /* signed value that's wrapped? */ 13679 assert(elen <= ((~(STRLEN)0) >> 1)); 13680 13681 /* if zeros is non-zero, then it represents filler between 13682 * elen and precis. So adding elen and zeros together will 13683 * always be <= precis, and the addition can never wrap */ 13684 assert(!zeros || (precis > elen && precis - elen == zeros)); 13685 have = elen + zeros; 13686 13687 if (have >= (((STRLEN)~0) - esignlen)) 13688 croak_memory_wrap(); 13689 have += esignlen; 13690 13691 need = (have > width ? have : width); 13692 gap = need - have; 13693 13694 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1))) 13695 croak_memory_wrap(); 13696 need += (SvCUR(sv) + 1); 13697 13698 SvGROW(sv, need); 13699 13700 s = SvEND(sv); 13701 13702 if (left) { 13703 for (i = 0; i < esignlen; i++) 13704 *s++ = esignbuf[i]; 13705 for (i = zeros; i; i--) 13706 *s++ = '0'; 13707 Copy(eptr, s, elen, char); 13708 s += elen; 13709 for (i = gap; i; i--) 13710 *s++ = ' '; 13711 } 13712 else { 13713 if (fill) { 13714 for (i = 0; i < esignlen; i++) 13715 *s++ = esignbuf[i]; 13716 assert(!zeros); 13717 zeros = gap; 13718 } 13719 else { 13720 for (i = gap; i; i--) 13721 *s++ = ' '; 13722 for (i = 0; i < esignlen; i++) 13723 *s++ = esignbuf[i]; 13724 } 13725 13726 for (i = zeros; i; i--) 13727 *s++ = '0'; 13728 Copy(eptr, s, elen, char); 13729 s += elen; 13730 } 13731 13732 *s = '\0'; 13733 SvCUR_set(sv, s - SvPVX_const(sv)); 13734 13735 if (is_utf8) 13736 has_utf8 = TRUE; 13737 if (has_utf8) 13738 SvUTF8_on(sv); 13739 } 13740 13741 if (vectorize && veclen) { 13742 /* we append the vector separator separately since %v isn't 13743 * very common: don't slow down the general case by adding 13744 * dotstrlen to need etc */ 13745 sv_catpvn_nomg(sv, dotstr, dotstrlen); 13746 esignlen = 0; 13747 goto vector; /* do next iteration */ 13748 } 13749 13750 done_valid_conversion: 13751 13752 if (arg_missing) 13753 S_warn_vcatpvfn_missing_argument(aTHX); 13754 } 13755 13756 /* Now that we've consumed all our printf format arguments (svix) 13757 * do we have things left on the stack that we didn't use? 13758 */ 13759 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) { 13760 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s", 13761 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); 13762 } 13763 13764 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 13765 /* while we shouldn't set the cache, it may have been previously 13766 set in the caller, so clear it */ 13767 MAGIC *mg = mg_find(sv, PERL_MAGIC_utf8); 13768 if (mg) 13769 magic_setutf8(sv,mg); /* clear UTF8 cache */ 13770 } 13771 SvTAINT(sv); 13772} 13773 13774/* ========================================================================= 13775 13776=for apidoc_section $embedding 13777 13778=cut 13779 13780All the macros and functions in this section are for the private use of 13781the main function, perl_clone(). 13782 13783The foo_dup() functions make an exact copy of an existing foo thingy. 13784During the course of a cloning, a hash table is used to map old addresses 13785to new addresses. The table is created and manipulated with the 13786ptr_table_* functions. 13787 13788 * =========================================================================*/ 13789 13790 13791#if defined(USE_ITHREADS) 13792 13793/* XXX Remove this so it doesn't have to go thru the macro and return for nothing */ 13794#ifndef GpREFCNT_inc 13795# define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL) 13796#endif 13797 13798 13799#define SAVEPV(p) ((p) ? savepv(p) : NULL) 13800#define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL) 13801 13802/* clone a parser */ 13803 13804yy_parser * 13805Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param) 13806{ 13807 yy_parser *parser; 13808 13809 PERL_ARGS_ASSERT_PARSER_DUP; 13810 13811 if (!proto) 13812 return NULL; 13813 13814 /* look for it in the table first */ 13815 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto); 13816 if (parser) 13817 return parser; 13818 13819 /* create anew and remember what it is */ 13820 Newxz(parser, 1, yy_parser); 13821 ptr_table_store(PL_ptr_table, proto, parser); 13822 13823 /* XXX eventually, just Copy() most of the parser struct ? */ 13824 13825 parser->lex_brackets = proto->lex_brackets; 13826 parser->lex_casemods = proto->lex_casemods; 13827 parser->lex_brackstack = savepvn(proto->lex_brackstack, 13828 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets)); 13829 parser->lex_casestack = savepvn(proto->lex_casestack, 13830 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods)); 13831 parser->lex_defer = proto->lex_defer; 13832 parser->lex_dojoin = proto->lex_dojoin; 13833 parser->lex_formbrack = proto->lex_formbrack; 13834 parser->lex_inpat = proto->lex_inpat; 13835 parser->lex_inwhat = proto->lex_inwhat; 13836 parser->lex_op = proto->lex_op; 13837 parser->lex_repl = sv_dup_inc(proto->lex_repl, param); 13838 parser->lex_starts = proto->lex_starts; 13839 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param); 13840 parser->multi_close = proto->multi_close; 13841 parser->multi_open = proto->multi_open; 13842 parser->multi_start = proto->multi_start; 13843 parser->multi_end = proto->multi_end; 13844 parser->preambled = proto->preambled; 13845 parser->lex_super_state = proto->lex_super_state; 13846 parser->lex_sub_inwhat = proto->lex_sub_inwhat; 13847 parser->lex_sub_op = proto->lex_sub_op; 13848 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param); 13849 parser->linestr = sv_dup_inc(proto->linestr, param); 13850 parser->expect = proto->expect; 13851 parser->copline = proto->copline; 13852 parser->last_lop_op = proto->last_lop_op; 13853 parser->lex_state = proto->lex_state; 13854 parser->rsfp = fp_dup(proto->rsfp, '<', param); 13855 /* rsfp_filters entries have fake IoDIRP() */ 13856 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param); 13857 parser->in_my = proto->in_my; 13858 parser->in_my_stash = hv_dup(proto->in_my_stash, param); 13859 parser->error_count = proto->error_count; 13860 parser->sig_elems = proto->sig_elems; 13861 parser->sig_optelems= proto->sig_optelems; 13862 parser->sig_slurpy = proto->sig_slurpy; 13863 parser->recheck_utf8_validity = proto->recheck_utf8_validity; 13864 13865 { 13866 char * const ols = SvPVX(proto->linestr); 13867 char * const ls = SvPVX(parser->linestr); 13868 13869 parser->bufptr = ls + (proto->bufptr >= ols ? 13870 proto->bufptr - ols : 0); 13871 parser->oldbufptr = ls + (proto->oldbufptr >= ols ? 13872 proto->oldbufptr - ols : 0); 13873 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ? 13874 proto->oldoldbufptr - ols : 0); 13875 parser->linestart = ls + (proto->linestart >= ols ? 13876 proto->linestart - ols : 0); 13877 parser->last_uni = ls + (proto->last_uni >= ols ? 13878 proto->last_uni - ols : 0); 13879 parser->last_lop = ls + (proto->last_lop >= ols ? 13880 proto->last_lop - ols : 0); 13881 13882 parser->bufend = ls + SvCUR(parser->linestr); 13883 } 13884 13885 Copy(proto->tokenbuf, parser->tokenbuf, 256, char); 13886 13887 13888 Copy(proto->nextval, parser->nextval, 5, YYSTYPE); 13889 Copy(proto->nexttype, parser->nexttype, 5, I32); 13890 parser->nexttoke = proto->nexttoke; 13891 13892 /* XXX should clone saved_curcop here, but we aren't passed 13893 * proto_perl; so do it in perl_clone_using instead */ 13894 13895 return parser; 13896} 13897 13898/* 13899=for apidoc_section $io 13900=for apidoc fp_dup 13901 13902Duplicate a file handle, returning a pointer to the cloned object. 13903 13904=cut 13905*/ 13906 13907PerlIO * 13908Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param) 13909{ 13910 PerlIO *ret; 13911 13912 PERL_ARGS_ASSERT_FP_DUP; 13913 PERL_UNUSED_ARG(type); 13914 13915 if (!fp) 13916 return (PerlIO*)NULL; 13917 13918 /* look for it in the table first */ 13919 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp); 13920 if (ret) 13921 return ret; 13922 13923 /* create anew and remember what it is */ 13924#ifdef __amigaos4__ 13925 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD); 13926#else 13927 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE); 13928#endif 13929 ptr_table_store(PL_ptr_table, fp, ret); 13930 return ret; 13931} 13932 13933/* 13934=for apidoc_section $io 13935=for apidoc dirp_dup 13936 13937Duplicate a directory handle, returning a pointer to the cloned object. 13938 13939=cut 13940*/ 13941 13942DIR * 13943Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param) 13944{ 13945 DIR *ret; 13946 13947#if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR) 13948 DIR *pwd; 13949 const Direntry_t *dirent; 13950 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */ 13951 char *name = NULL; 13952 STRLEN len = 0; 13953 long pos; 13954#endif 13955 13956 PERL_UNUSED_CONTEXT; 13957 PERL_ARGS_ASSERT_DIRP_DUP; 13958 13959 if (!dp) 13960 return (DIR*)NULL; 13961 13962 /* look for it in the table first */ 13963 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp); 13964 if (ret) 13965 return ret; 13966 13967#if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR) 13968 13969 PERL_UNUSED_ARG(param); 13970 13971 /* create anew */ 13972 13973 /* open the current directory (so we can switch back) */ 13974 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL; 13975 13976 /* chdir to our dir handle and open the present working directory */ 13977 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) { 13978 PerlDir_close(pwd); 13979 return (DIR *)NULL; 13980 } 13981 /* Now we should have two dir handles pointing to the same dir. */ 13982 13983 /* Be nice to the calling code and chdir back to where we were. */ 13984 /* XXX If this fails, then what? */ 13985 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd))); 13986 13987 /* We have no need of the pwd handle any more. */ 13988 PerlDir_close(pwd); 13989 13990#ifdef DIRNAMLEN 13991# define d_namlen(d) (d)->d_namlen 13992#else 13993# define d_namlen(d) strlen((d)->d_name) 13994#endif 13995 /* Iterate once through dp, to get the file name at the current posi- 13996 tion. Then step back. */ 13997 pos = PerlDir_tell(dp); 13998 if ((dirent = PerlDir_read(dp))) { 13999 len = d_namlen(dirent); 14000 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) { 14001 /* If the len is somehow magically longer than the 14002 * maximum length of the directory entry, even though 14003 * we could fit it in a buffer, we could not copy it 14004 * from the dirent. Bail out. */ 14005 PerlDir_close(ret); 14006 return (DIR*)NULL; 14007 } 14008 if (len <= sizeof smallbuf) name = smallbuf; 14009 else Newx(name, len, char); 14010 Move(dirent->d_name, name, len, char); 14011 } 14012 PerlDir_seek(dp, pos); 14013 14014 /* Iterate through the new dir handle, till we find a file with the 14015 right name. */ 14016 if (!dirent) /* just before the end */ 14017 for(;;) { 14018 pos = PerlDir_tell(ret); 14019 if (PerlDir_read(ret)) continue; /* not there yet */ 14020 PerlDir_seek(ret, pos); /* step back */ 14021 break; 14022 } 14023 else { 14024 const long pos0 = PerlDir_tell(ret); 14025 for(;;) { 14026 pos = PerlDir_tell(ret); 14027 if ((dirent = PerlDir_read(ret))) { 14028 if (len == (STRLEN)d_namlen(dirent) 14029 && memEQ(name, dirent->d_name, len)) { 14030 /* found it */ 14031 PerlDir_seek(ret, pos); /* step back */ 14032 break; 14033 } 14034 /* else we are not there yet; keep iterating */ 14035 } 14036 else { /* This is not meant to happen. The best we can do is 14037 reset the iterator to the beginning. */ 14038 PerlDir_seek(ret, pos0); 14039 break; 14040 } 14041 } 14042 } 14043#undef d_namlen 14044 14045 if (name && name != smallbuf) 14046 Safefree(name); 14047#endif 14048 14049#ifdef WIN32 14050 ret = win32_dirp_dup(dp, param); 14051#endif 14052 14053 /* pop it in the pointer table */ 14054 if (ret) 14055 ptr_table_store(PL_ptr_table, dp, ret); 14056 14057 return ret; 14058} 14059 14060/* 14061=for apidoc_section $GV 14062=for apidoc gp_dup 14063 14064Duplicate a typeglob, returning a pointer to the cloned object. 14065 14066=cut 14067*/ 14068 14069GP * 14070Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param) 14071{ 14072 GP *ret; 14073 14074 PERL_ARGS_ASSERT_GP_DUP; 14075 14076 if (!gp) 14077 return (GP*)NULL; 14078 /* look for it in the table first */ 14079 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp); 14080 if (ret) 14081 return ret; 14082 14083 /* create anew and remember what it is */ 14084 Newxz(ret, 1, GP); 14085 ptr_table_store(PL_ptr_table, gp, ret); 14086 14087 /* clone */ 14088 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying 14089 on Newxz() to do this for us. */ 14090 ret->gp_sv = sv_dup_inc(gp->gp_sv, param); 14091 ret->gp_io = io_dup_inc(gp->gp_io, param); 14092 ret->gp_form = cv_dup_inc(gp->gp_form, param); 14093 ret->gp_av = av_dup_inc(gp->gp_av, param); 14094 ret->gp_hv = hv_dup_inc(gp->gp_hv, param); 14095 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */ 14096 ret->gp_cv = cv_dup_inc(gp->gp_cv, param); 14097 ret->gp_cvgen = gp->gp_cvgen; 14098 ret->gp_line = gp->gp_line; 14099 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param); 14100 return ret; 14101} 14102 14103 14104/* 14105=for apidoc_section $magic 14106=for apidoc mg_dup 14107 14108Duplicate a chain of magic, returning a pointer to the cloned object. 14109 14110=cut 14111*/ 14112 14113MAGIC * 14114Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param) 14115{ 14116 MAGIC *mgret = NULL; 14117 MAGIC **mgprev_p = &mgret; 14118 14119 PERL_ARGS_ASSERT_MG_DUP; 14120 14121 for (; mg; mg = mg->mg_moremagic) { 14122 MAGIC *nmg; 14123 14124 if ((param->flags & CLONEf_JOIN_IN) 14125 && mg->mg_type == PERL_MAGIC_backref) 14126 /* when joining, we let the individual SVs add themselves to 14127 * backref as needed. */ 14128 continue; 14129 14130 Newx(nmg, 1, MAGIC); 14131 *mgprev_p = nmg; 14132 mgprev_p = &(nmg->mg_moremagic); 14133 14134 /* There was a comment "XXX copy dynamic vtable?" but as we don't have 14135 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates 14136 from the original commit adding Perl_mg_dup() - revision 4538. 14137 Similarly there is the annotation "XXX random ptr?" next to the 14138 assignment to nmg->mg_ptr. */ 14139 *nmg = *mg; 14140 14141 /* FIXME for plugins 14142 if (nmg->mg_type == PERL_MAGIC_qr) { 14143 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param)); 14144 } 14145 else 14146 */ 14147 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED) 14148 ? nmg->mg_type == PERL_MAGIC_backref 14149 /* The backref AV has its reference 14150 * count deliberately bumped by 1 */ 14151 ? SvREFCNT_inc(av_dup_inc((const AV *) 14152 nmg->mg_obj, param)) 14153 : sv_dup_inc(nmg->mg_obj, param) 14154 : (nmg->mg_type == PERL_MAGIC_regdatum || 14155 nmg->mg_type == PERL_MAGIC_regdata) 14156 ? nmg->mg_obj 14157 : sv_dup(nmg->mg_obj, param); 14158 14159 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) { 14160 if (nmg->mg_len > 0) { 14161 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len); 14162 if (nmg->mg_type == PERL_MAGIC_overload_table && 14163 AMT_AMAGIC((AMT*)nmg->mg_ptr)) 14164 { 14165 AMT * const namtp = (AMT*)nmg->mg_ptr; 14166 sv_dup_inc_multiple((SV**)(namtp->table), 14167 (SV**)(namtp->table), NofAMmeth, param); 14168 } 14169 } 14170 else if (nmg->mg_len == HEf_SVKEY) 14171 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param); 14172 } 14173 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) { 14174 nmg->mg_virtual->svt_dup(aTHX_ nmg, param); 14175 } 14176 } 14177 return mgret; 14178} 14179 14180#endif /* USE_ITHREADS */ 14181 14182struct ptr_tbl_arena { 14183 struct ptr_tbl_arena *next; 14184 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */ 14185}; 14186 14187/* 14188=for apidoc_section $embedding 14189=for apidoc ptr_table_new 14190 14191Create a new pointer-mapping table 14192 14193=cut 14194*/ 14195 14196PTR_TBL_t * 14197Perl_ptr_table_new(pTHX) 14198{ 14199 PTR_TBL_t *tbl; 14200 PERL_UNUSED_CONTEXT; 14201 14202 Newx(tbl, 1, PTR_TBL_t); 14203 tbl->tbl_max = 511; 14204 tbl->tbl_items = 0; 14205 tbl->tbl_arena = NULL; 14206 tbl->tbl_arena_next = NULL; 14207 tbl->tbl_arena_end = NULL; 14208 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*); 14209 return tbl; 14210} 14211 14212#define PTR_TABLE_HASH(ptr) \ 14213 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17))) 14214 14215/* map an existing pointer using a table */ 14216 14217STATIC PTR_TBL_ENT_t * 14218S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv) 14219{ 14220 PTR_TBL_ENT_t *tblent; 14221 const UV hash = PTR_TABLE_HASH(sv); 14222 14223 PERL_ARGS_ASSERT_PTR_TABLE_FIND; 14224 14225 tblent = tbl->tbl_ary[hash & tbl->tbl_max]; 14226 for (; tblent; tblent = tblent->next) { 14227 if (tblent->oldval == sv) 14228 return tblent; 14229 } 14230 return NULL; 14231} 14232 14233/* 14234=for apidoc ptr_table_fetch 14235 14236Look for C<sv> in the pointer-mapping table C<tbl>, returning its value, or 14237NULL if not found. 14238 14239=cut 14240*/ 14241 14242void * 14243Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv) 14244{ 14245 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv); 14246 14247 PERL_ARGS_ASSERT_PTR_TABLE_FETCH; 14248 PERL_UNUSED_CONTEXT; 14249 14250 return tblent ? tblent->newval : NULL; 14251} 14252 14253/* 14254=for apidoc ptr_table_store 14255 14256Add a new entry to a pointer-mapping table C<tbl>. 14257In hash terms, C<oldsv> is the key; Cnewsv> is the value. 14258 14259The names "old" and "new" are specific to the core's typical use of ptr_tables 14260in thread cloning. 14261 14262=cut 14263*/ 14264 14265void 14266Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv) 14267{ 14268 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv); 14269 14270 PERL_ARGS_ASSERT_PTR_TABLE_STORE; 14271 PERL_UNUSED_CONTEXT; 14272 14273 if (tblent) { 14274 tblent->newval = newsv; 14275 } else { 14276 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max; 14277 14278 if (tbl->tbl_arena_next == tbl->tbl_arena_end) { 14279 struct ptr_tbl_arena *new_arena; 14280 14281 Newx(new_arena, 1, struct ptr_tbl_arena); 14282 new_arena->next = tbl->tbl_arena; 14283 tbl->tbl_arena = new_arena; 14284 tbl->tbl_arena_next = new_arena->array; 14285 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array); 14286 } 14287 14288 tblent = tbl->tbl_arena_next++; 14289 14290 tblent->oldval = oldsv; 14291 tblent->newval = newsv; 14292 tblent->next = tbl->tbl_ary[entry]; 14293 tbl->tbl_ary[entry] = tblent; 14294 tbl->tbl_items++; 14295 if (tblent->next && tbl->tbl_items > tbl->tbl_max) 14296 ptr_table_split(tbl); 14297 } 14298} 14299 14300/* 14301=for apidoc ptr_table_split 14302 14303Double the hash bucket size of an existing ptr table 14304 14305=cut 14306*/ 14307 14308void 14309Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl) 14310{ 14311 PTR_TBL_ENT_t **ary = tbl->tbl_ary; 14312 const UV oldsize = tbl->tbl_max + 1; 14313 UV newsize = oldsize * 2; 14314 UV i; 14315 14316 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT; 14317 PERL_UNUSED_CONTEXT; 14318 14319 Renew(ary, newsize, PTR_TBL_ENT_t*); 14320 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*); 14321 tbl->tbl_max = --newsize; 14322 tbl->tbl_ary = ary; 14323 for (i=0; i < oldsize; i++, ary++) { 14324 PTR_TBL_ENT_t **entp = ary; 14325 PTR_TBL_ENT_t *ent = *ary; 14326 PTR_TBL_ENT_t **curentp; 14327 if (!ent) 14328 continue; 14329 curentp = ary + oldsize; 14330 do { 14331 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) { 14332 *entp = ent->next; 14333 ent->next = *curentp; 14334 *curentp = ent; 14335 } 14336 else 14337 entp = &ent->next; 14338 ent = *entp; 14339 } while (ent); 14340 } 14341} 14342 14343/* 14344=for apidoc ptr_table_free 14345 14346Clear and free a ptr table 14347 14348=cut 14349*/ 14350 14351void 14352Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl) 14353{ 14354 struct ptr_tbl_arena *arena; 14355 14356 PERL_UNUSED_CONTEXT; 14357 14358 if (!tbl) { 14359 return; 14360 } 14361 14362 arena = tbl->tbl_arena; 14363 14364 while (arena) { 14365 struct ptr_tbl_arena *next = arena->next; 14366 14367 Safefree(arena); 14368 arena = next; 14369 } 14370 14371 Safefree(tbl->tbl_ary); 14372 Safefree(tbl); 14373} 14374 14375#if defined(USE_ITHREADS) 14376 14377void 14378Perl_rvpv_dup(pTHX_ SV *const dsv, const SV *const ssv, CLONE_PARAMS *const param) 14379{ 14380 PERL_ARGS_ASSERT_RVPV_DUP; 14381 14382 assert(!isREGEXP(ssv)); 14383 if (SvROK(ssv)) { 14384 if (SvWEAKREF(ssv)) { 14385 SvRV_set(dsv, sv_dup(SvRV_const(ssv), param)); 14386 if (param->flags & CLONEf_JOIN_IN) { 14387 /* if joining, we add any back references individually rather 14388 * than copying the whole backref array */ 14389 Perl_sv_add_backref(aTHX_ SvRV(dsv), dsv); 14390 } 14391 } 14392 else 14393 SvRV_set(dsv, sv_dup_inc(SvRV_const(ssv), param)); 14394 } 14395 else if (SvPVX_const(ssv)) { 14396 /* Has something there */ 14397 if (SvLEN(ssv)) { 14398 /* Normal PV - clone whole allocated space */ 14399 SvPV_set(dsv, SAVEPVN(SvPVX_const(ssv), SvLEN(ssv)-1)); 14400 /* ssv may not be that normal, but actually copy on write. 14401 But we are a true, independent SV, so: */ 14402 SvIsCOW_off(dsv); 14403 } 14404 else { 14405 /* Special case - not normally malloced for some reason */ 14406 if (isGV_with_GP(ssv)) { 14407 /* Don't need to do anything here. */ 14408 } 14409 else if ((SvIsCOW_shared_hash(ssv))) { 14410 /* A "shared" PV - clone it as "shared" PV */ 14411 SvPV_set(dsv, 14412 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(ssv)), 14413 param))); 14414 } 14415 else { 14416 /* Some other special case - random pointer */ 14417 SvPV_set(dsv, (char *) SvPVX_const(ssv)); 14418 } 14419 } 14420 } 14421 else { 14422 /* Copy the NULL */ 14423 SvPV_set(dsv, NULL); 14424 } 14425} 14426 14427/* duplicate a list of SVs. source and dest may point to the same memory. */ 14428static SV ** 14429S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest, 14430 SSize_t items, CLONE_PARAMS *const param) 14431{ 14432 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE; 14433 14434 while (items-- > 0) { 14435 *dest++ = sv_dup_inc(*source++, param); 14436 } 14437 14438 return dest; 14439} 14440 14441/* duplicate the HvAUX of an HV */ 14442static void 14443S_sv_dup_hvaux(pTHX_ const SV *const ssv, SV *dsv, CLONE_PARAMS *const param) 14444{ 14445 PERL_ARGS_ASSERT_SV_DUP_HVAUX; 14446 14447 const struct xpvhv_aux * const saux = HvAUX(ssv); 14448 struct xpvhv_aux * const daux = HvAUX(dsv); 14449 /* This flag isn't copied. */ 14450 SvFLAGS(dsv) |= SVphv_HasAUX; 14451 14452 if (saux->xhv_name_count) { 14453 HEK ** const sname = saux->xhv_name_u.xhvnameu_names; 14454 const I32 count = saux->xhv_name_count < 0 14455 ? -saux->xhv_name_count 14456 : saux->xhv_name_count; 14457 HEK **shekp = sname + count; 14458 HEK **dhekp; 14459 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *); 14460 dhekp = daux->xhv_name_u.xhvnameu_names + count; 14461 while (shekp-- > sname) { 14462 dhekp--; 14463 *dhekp = hek_dup(*shekp, param); 14464 } 14465 } 14466 else { 14467 daux->xhv_name_u.xhvnameu_name = hek_dup(saux->xhv_name_u.xhvnameu_name, param); 14468 } 14469 daux->xhv_name_count = saux->xhv_name_count; 14470 14471 daux->xhv_aux_flags = saux->xhv_aux_flags; 14472#ifdef PERL_HASH_RANDOMIZE_KEYS 14473 daux->xhv_rand = saux->xhv_rand; 14474 daux->xhv_last_rand = saux->xhv_last_rand; 14475#endif 14476 daux->xhv_riter = saux->xhv_riter; 14477 daux->xhv_eiter = saux->xhv_eiter ? he_dup(saux->xhv_eiter, FALSE, param) : 0; 14478 /* backref array needs refcnt=2; see sv_add_backref */ 14479 daux->xhv_backreferences = 14480 (param->flags & CLONEf_JOIN_IN) 14481 /* when joining, we let the individual GVs and 14482 * CVs add themselves to backref as 14483 * needed. This avoids pulling in stuff 14484 * that isn't required, and simplifies the 14485 * case where stashes aren't cloned back 14486 * if they already exist in the parent 14487 * thread */ 14488 ? NULL 14489 : saux->xhv_backreferences 14490 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV) 14491 ? MUTABLE_AV(SvREFCNT_inc( 14492 sv_dup_inc((const SV *) 14493 saux->xhv_backreferences, param))) 14494 : MUTABLE_AV(sv_dup((const SV *) 14495 saux->xhv_backreferences, param)) 14496 : 0; 14497 14498 daux->xhv_mro_meta = saux->xhv_mro_meta 14499 ? mro_meta_dup(saux->xhv_mro_meta, param) 14500 : 0; 14501 14502 /* Record stashes for possible cloning in Perl_clone(). */ 14503 if (HvNAME(ssv)) 14504 av_push(param->stashes, dsv); 14505 14506 if (HvSTASH_IS_CLASS(ssv)) { 14507 daux->xhv_class_superclass = hv_dup_inc(saux->xhv_class_superclass, param); 14508 daux->xhv_class_initfields_cv = cv_dup_inc(saux->xhv_class_initfields_cv, param); 14509 daux->xhv_class_adjust_blocks = av_dup_inc(saux->xhv_class_adjust_blocks, param); 14510 daux->xhv_class_fields = padnamelist_dup_inc(saux->xhv_class_fields, param); 14511 daux->xhv_class_next_fieldix = saux->xhv_class_next_fieldix; 14512 daux->xhv_class_param_map = hv_dup_inc(saux->xhv_class_param_map, param); 14513 14514 /* TODO: This does mean that we can't compile more `field` expressions 14515 * in the cloned thread, but surely we're done with compiletime now..? 14516 */ 14517 daux->xhv_class_suspended_initfields_compcv = NULL; 14518 } 14519} 14520 14521/* duplicate an SV of any type (including AV, HV etc) */ 14522 14523static SV * 14524S_sv_dup_common(pTHX_ const SV *const ssv, CLONE_PARAMS *const param) 14525{ 14526 SV *dsv; 14527 14528 PERL_ARGS_ASSERT_SV_DUP_COMMON; 14529 14530 if (SvIS_FREED(ssv)) { 14531#ifdef DEBUG_LEAKING_SCALARS_ABORT 14532 abort(); 14533#endif 14534 return NULL; 14535 } 14536 /* look for it in the table first */ 14537 dsv = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, ssv)); 14538 if (dsv) 14539 return dsv; 14540 14541 if(param->flags & CLONEf_JOIN_IN) { 14542 /** We are joining here so we don't want do clone 14543 something that is bad **/ 14544 if (SvTYPE(ssv) == SVt_PVHV) { 14545 const HEK * const hvname = HvNAME_HEK(ssv); 14546 if (hvname) { 14547 /** don't clone stashes if they already exist **/ 14548 dsv = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname), 14549 HEK_UTF8(hvname) ? SVf_UTF8 : 0)); 14550 ptr_table_store(PL_ptr_table, ssv, dsv); 14551 return dsv; 14552 } 14553 } 14554 else if (SvTYPE(ssv) == SVt_PVGV && !SvFAKE(ssv)) { 14555 HV *stash = GvSTASH(ssv); 14556 const HEK * hvname; 14557 if (stash && (hvname = HvNAME_HEK(stash))) { 14558 /** don't clone GVs if they already exist **/ 14559 SV **svp; 14560 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname), 14561 HEK_UTF8(hvname) ? SVf_UTF8 : 0); 14562 svp = hv_fetch( 14563 stash, GvNAME(ssv), 14564 GvNAMEUTF8(ssv) 14565 ? -GvNAMELEN(ssv) 14566 : GvNAMELEN(ssv), 14567 0 14568 ); 14569 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) { 14570 ptr_table_store(PL_ptr_table, ssv, *svp); 14571 return *svp; 14572 } 14573 } 14574 } 14575 } 14576 14577 /* create anew and remember what it is */ 14578 new_SV(dsv); 14579 14580#ifdef DEBUG_LEAKING_SCALARS 14581 dsv->sv_debug_optype = ssv->sv_debug_optype; 14582 dsv->sv_debug_line = ssv->sv_debug_line; 14583 dsv->sv_debug_inpad = ssv->sv_debug_inpad; 14584 dsv->sv_debug_parent = (SV*)ssv; 14585 FREE_SV_DEBUG_FILE(dsv); 14586 dsv->sv_debug_file = savesharedpv(ssv->sv_debug_file); 14587#endif 14588 14589 ptr_table_store(PL_ptr_table, ssv, dsv); 14590 14591 /* clone */ 14592 SvFLAGS(dsv) = SvFLAGS(ssv); 14593 SvFLAGS(dsv) &= ~SVf_OOK; /* don't propagate OOK hack */ 14594 SvREFCNT(dsv) = 0; /* must be before any other dups! */ 14595 14596#ifdef DEBUGGING 14597 if (SvANY(ssv) && PL_watch_pvx && SvPVX_const(ssv) == PL_watch_pvx) 14598 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n", 14599 (void*)PL_watch_pvx, SvPVX_const(ssv)); 14600#endif 14601 14602 /* don't clone objects whose class has asked us not to */ 14603 if (SvOBJECT(ssv) 14604 && ! (SvFLAGS(SvSTASH(ssv)) & SVphv_CLONEABLE)) 14605 { 14606 SvFLAGS(dsv) = 0; 14607 return dsv; 14608 } 14609 14610 switch (SvTYPE(ssv)) { 14611 case SVt_NULL: 14612 SvANY(dsv) = NULL; 14613 break; 14614 case SVt_IV: 14615 SET_SVANY_FOR_BODYLESS_IV(dsv); 14616 if(SvROK(ssv)) { 14617 Perl_rvpv_dup(aTHX_ dsv, ssv, param); 14618 } else { 14619 SvIV_set(dsv, SvIVX(ssv)); 14620 } 14621 break; 14622 case SVt_NV: 14623#if NVSIZE <= IVSIZE 14624 SET_SVANY_FOR_BODYLESS_NV(dsv); 14625#else 14626 SvANY(dsv) = new_XNV(); 14627#endif 14628 SvNV_set(dsv, SvNVX(ssv)); 14629 break; 14630 default: 14631 { 14632 /* These are all the types that need complex bodies allocating. */ 14633 void *new_body; 14634 const svtype sv_type = SvTYPE(ssv); 14635 const struct body_details *sv_type_details 14636 = bodies_by_type + sv_type; 14637 14638 switch (sv_type) { 14639 default: 14640 Perl_croak(param->proto_perl, "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(ssv)); 14641 NOT_REACHED; /* NOTREACHED */ 14642 break; 14643 14644 case SVt_PVHV: 14645 if (HvHasAUX(ssv)) { 14646 sv_type_details = &fake_hv_with_aux; 14647#ifdef PURIFY 14648 new_body = new_NOARENA(sv_type_details); 14649#else 14650 new_body_from_arena(new_body, HVAUX_ARENA_ROOT_IX, fake_hv_with_aux); 14651#endif 14652 goto have_body; 14653 } 14654 /* FALLTHROUGH */ 14655 case SVt_PVOBJ: 14656 case SVt_PVGV: 14657 case SVt_PVIO: 14658 case SVt_PVFM: 14659 case SVt_PVAV: 14660 case SVt_PVCV: 14661 case SVt_PVLV: 14662 case SVt_REGEXP: 14663 case SVt_PVMG: 14664 case SVt_PVNV: 14665 case SVt_PVIV: 14666 case SVt_INVLIST: 14667 case SVt_PV: 14668 assert(sv_type_details->body_size); 14669#ifndef PURIFY 14670 if (sv_type_details->arena) { 14671 new_body = S_new_body(aTHX_ sv_type); 14672 new_body 14673 = (void*)((char*)new_body - sv_type_details->offset); 14674 } else 14675#endif 14676 { 14677 new_body = new_NOARENA(sv_type_details); 14678 } 14679 } 14680 have_body: 14681 assert(new_body); 14682 SvANY(dsv) = new_body; 14683 14684#ifndef PURIFY 14685 Copy(((char*)SvANY(ssv)) + sv_type_details->offset, 14686 ((char*)SvANY(dsv)) + sv_type_details->offset, 14687 sv_type_details->copy, char); 14688#else 14689 Copy(((char*)SvANY(ssv)), 14690 ((char*)SvANY(dsv)), 14691 sv_type_details->body_size + sv_type_details->offset, char); 14692#endif 14693 14694 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV && sv_type != SVt_PVOBJ 14695 && !isGV_with_GP(dsv) 14696 && !isREGEXP(dsv) 14697 && !(sv_type == SVt_PVIO && !(IoFLAGS(dsv) & IOf_FAKE_DIRP))) 14698 Perl_rvpv_dup(aTHX_ dsv, ssv, param); 14699 14700 /* The Copy above means that all the source (unduplicated) pointers 14701 are now in the destination. We can check the flags and the 14702 pointers in either, but it's possible that there's less cache 14703 missing by always going for the destination. 14704 FIXME - instrument and check that assumption */ 14705 if (sv_type >= SVt_PVMG) { 14706 if (SvMAGIC(dsv)) 14707 SvMAGIC_set(dsv, mg_dup(SvMAGIC(dsv), param)); 14708 if (SvOBJECT(dsv) && SvSTASH(dsv)) 14709 SvSTASH_set(dsv, hv_dup_inc(SvSTASH(dsv), param)); 14710 else SvSTASH_set(dsv, 0); /* don't copy DESTROY cache */ 14711 } 14712 14713 /* The cast silences a GCC warning about unhandled types. */ 14714 switch ((int)sv_type) { 14715 case SVt_PV: 14716 break; 14717 case SVt_PVIV: 14718 break; 14719 case SVt_PVNV: 14720 break; 14721 case SVt_PVMG: 14722 break; 14723 case SVt_REGEXP: 14724 duprex: 14725 /* FIXME for plugins */ 14726 re_dup_guts((REGEXP*) ssv, (REGEXP*) dsv, param); 14727 break; 14728 case SVt_PVLV: 14729 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */ 14730 if (LvTYPE(dsv) == 't') /* for tie: unrefcnted fake (SV**) */ 14731 LvTARG(dsv) = dsv; 14732 else if (LvTYPE(dsv) == 'T') /* for tie: fake HE */ 14733 LvTARG(dsv) = MUTABLE_SV(he_dup((HE*)LvTARG(dsv), FALSE, param)); 14734 else 14735 LvTARG(dsv) = sv_dup_inc(LvTARG(dsv), param); 14736 if (isREGEXP(ssv)) goto duprex; 14737 /* FALLTHROUGH */ 14738 case SVt_PVGV: 14739 /* non-GP case already handled above */ 14740 if(isGV_with_GP(ssv)) { 14741 GvNAME_HEK(dsv) = hek_dup(GvNAME_HEK(dsv), param); 14742 /* Don't call sv_add_backref here as it's going to be 14743 created as part of the magic cloning of the symbol 14744 table--unless this is during a join and the stash 14745 is not actually being cloned. */ 14746 /* Danger Will Robinson - GvGP(dsv) isn't initialised 14747 at the point of this comment. */ 14748 GvSTASH(dsv) = hv_dup(GvSTASH(dsv), param); 14749 if (param->flags & CLONEf_JOIN_IN) 14750 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dsv)), dsv); 14751 GvGP_set(dsv, gp_dup(GvGP(ssv), param)); 14752 (void)GpREFCNT_inc(GvGP(dsv)); 14753 } 14754 break; 14755 case SVt_PVIO: 14756 /* PL_parser->rsfp_filters entries have fake IoDIRP() */ 14757 if(IoFLAGS(dsv) & IOf_FAKE_DIRP) { 14758 /* I have no idea why fake dirp (rsfps) 14759 should be treated differently but otherwise 14760 we end up with leaks -- sky*/ 14761 IoTOP_GV(dsv) = gv_dup_inc(IoTOP_GV(dsv), param); 14762 IoFMT_GV(dsv) = gv_dup_inc(IoFMT_GV(dsv), param); 14763 IoBOTTOM_GV(dsv) = gv_dup_inc(IoBOTTOM_GV(dsv), param); 14764 } else { 14765 IoTOP_GV(dsv) = gv_dup(IoTOP_GV(dsv), param); 14766 IoFMT_GV(dsv) = gv_dup(IoFMT_GV(dsv), param); 14767 IoBOTTOM_GV(dsv) = gv_dup(IoBOTTOM_GV(dsv), param); 14768 if (IoDIRP(dsv)) { 14769 IoDIRP(dsv) = dirp_dup(IoDIRP(dsv), param); 14770 } else { 14771 NOOP; 14772 /* IoDIRP(dsv) is already a copy of IoDIRP(ssv) */ 14773 } 14774 IoIFP(dsv) = fp_dup(IoIFP(ssv), IoTYPE(dsv), param); 14775 } 14776 if (IoOFP(dsv) == IoIFP(ssv)) 14777 IoOFP(dsv) = IoIFP(dsv); 14778 else 14779 IoOFP(dsv) = fp_dup(IoOFP(dsv), IoTYPE(dsv), param); 14780 IoTOP_NAME(dsv) = SAVEPV(IoTOP_NAME(dsv)); 14781 IoFMT_NAME(dsv) = SAVEPV(IoFMT_NAME(dsv)); 14782 IoBOTTOM_NAME(dsv) = SAVEPV(IoBOTTOM_NAME(dsv)); 14783 break; 14784 case SVt_PVAV: 14785 /* avoid cloning an empty array */ 14786 if (AvARRAY((const AV *)ssv) && AvFILLp((const AV *)ssv) >= 0) { 14787 SV **dst_ary, **src_ary; 14788 SSize_t items = AvFILLp((const AV *)ssv) + 1; 14789 14790 src_ary = AvARRAY((const AV *)ssv); 14791 Newx(dst_ary, AvMAX((const AV *)ssv)+1, SV*); 14792 ptr_table_store(PL_ptr_table, src_ary, dst_ary); 14793 AvARRAY(MUTABLE_AV(dsv)) = dst_ary; 14794 AvALLOC((const AV *)dsv) = dst_ary; 14795 if (AvREAL((const AV *)ssv)) { 14796 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items, 14797 param); 14798 } 14799 else { 14800 while (items-- > 0) 14801 *dst_ary++ = sv_dup(*src_ary++, param); 14802 } 14803 items = AvMAX((const AV *)ssv) - AvFILLp((const AV *)ssv); 14804 while (items-- > 0) { 14805 *dst_ary++ = NULL; 14806 } 14807 } 14808 else { 14809 AvARRAY(MUTABLE_AV(dsv)) = NULL; 14810 AvALLOC((const AV *)dsv) = (SV**)NULL; 14811 AvMAX( (const AV *)dsv) = -1; 14812 AvFILLp((const AV *)dsv) = -1; 14813 } 14814 break; 14815 case SVt_PVHV: 14816 if (HvARRAY((const HV *)ssv)) { 14817 STRLEN i = 0; 14818 XPVHV * const dxhv = (XPVHV*)SvANY(dsv); 14819 XPVHV * const sxhv = (XPVHV*)SvANY(ssv); 14820 char *darray; 14821 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1), 14822 char); 14823 HvARRAY(dsv) = (HE**)darray; 14824 while (i <= sxhv->xhv_max) { 14825 const HE * const source = HvARRAY(ssv)[i]; 14826 HvARRAY(dsv)[i] = source 14827 ? he_dup(source, FALSE, param) : 0; 14828 ++i; 14829 } 14830 if (HvHasAUX(ssv)) 14831 sv_dup_hvaux(ssv, dsv, param); 14832 } 14833 else 14834 HvARRAY(MUTABLE_HV(dsv)) = NULL; 14835 break; 14836 case SVt_PVCV: 14837 if (!(param->flags & CLONEf_COPY_STACKS)) { 14838 CvDEPTH(dsv) = 0; 14839 } 14840 /* FALLTHROUGH */ 14841 case SVt_PVFM: 14842 /* NOTE: not refcounted */ 14843 SvANY(MUTABLE_CV(dsv))->xcv_stash = 14844 hv_dup(CvSTASH(dsv), param); 14845 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dsv)) 14846 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dsv)), dsv); 14847 if (!CvISXSUB(dsv)) { 14848 OP_REFCNT_LOCK; 14849 CvROOT(dsv) = OpREFCNT_inc(CvROOT(dsv)); 14850 OP_REFCNT_UNLOCK; 14851 CvSLABBED_off(dsv); 14852 } else if (CvCONST(dsv)) { 14853 CvXSUBANY(dsv).any_ptr = 14854 sv_dup_inc((const SV *)CvXSUBANY(dsv).any_ptr, param); 14855 } else if (CvREFCOUNTED_ANYSV(dsv)) { 14856 CvXSUBANY(dsv).any_sv = 14857 sv_dup_inc((const SV *)CvXSUBANY(dsv).any_sv, param); 14858 } 14859 assert(!CvSLABBED(dsv)); 14860 if (CvDYNFILE(dsv)) CvFILE(dsv) = SAVEPV(CvFILE(dsv)); 14861 if (CvNAMED(dsv)) 14862 SvANY((CV *)dsv)->xcv_gv_u.xcv_hek = 14863 hek_dup(CvNAME_HEK((CV *)ssv), param); 14864 /* don't dup if copying back - CvGV isn't refcounted, so the 14865 * duped GV may never be freed. A bit of a hack! DAPM */ 14866 else 14867 SvANY(MUTABLE_CV(dsv))->xcv_gv_u.xcv_gv = 14868 CvCVGV_RC(dsv) 14869 ? gv_dup_inc(CvGV(ssv), param) 14870 : (param->flags & CLONEf_JOIN_IN) 14871 ? NULL 14872 : gv_dup(CvGV(ssv), param); 14873 14874 if (!CvISXSUB(ssv)) { 14875 PADLIST * padlist = CvPADLIST(ssv); 14876 if(padlist) 14877 padlist = padlist_dup(padlist, param); 14878 CvPADLIST_set(dsv, padlist); 14879 } else 14880/* unthreaded perl can't sv_dup so we don't support unthreaded's CvHSCXT */ 14881 PoisonPADLIST(dsv); 14882 14883 CvOUTSIDE(dsv) = 14884 CvWEAKOUTSIDE(ssv) 14885 ? cv_dup( CvOUTSIDE(dsv), param) 14886 : cv_dup_inc(CvOUTSIDE(dsv), param); 14887 break; 14888 case SVt_PVOBJ: 14889 { 14890 Size_t fieldcount = ObjectMAXFIELD(ssv) + 1; 14891 14892 Newx(ObjectFIELDS(dsv), fieldcount, SV *); 14893 ObjectMAXFIELD(dsv) = fieldcount - 1; 14894 14895 sv_dup_inc_multiple(ObjectFIELDS(ssv), ObjectFIELDS(dsv), fieldcount, param); 14896 } 14897 break; 14898 } 14899 } 14900 } 14901 14902 return dsv; 14903 } 14904 14905SV * 14906Perl_sv_dup_inc(pTHX_ const SV *const ssv, CLONE_PARAMS *const param) 14907{ 14908 PERL_ARGS_ASSERT_SV_DUP_INC; 14909 return ssv ? SvREFCNT_inc(sv_dup_common(ssv, param)) : NULL; 14910} 14911 14912SV * 14913Perl_sv_dup(pTHX_ const SV *const ssv, CLONE_PARAMS *const param) 14914{ 14915 SV *dsv = ssv ? sv_dup_common(ssv, param) : NULL; 14916 PERL_ARGS_ASSERT_SV_DUP; 14917 14918 /* Track every SV that (at least initially) had a reference count of 0. 14919 We need to do this by holding an actual reference to it in this array. 14920 If we attempt to cheat, turn AvREAL_off(), and store only pointers 14921 (akin to the stashes hash, and the perl stack), we come unstuck if 14922 a weak reference (or other SV legitimately SvREFCNT() == 0 for this 14923 thread) is manipulated in a CLONE method, because CLONE runs before the 14924 unreferenced array is walked to find SVs still with SvREFCNT() == 0 14925 (and fix things up by giving each a reference via the temps stack). 14926 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and 14927 then SvREFCNT_dec(), it will be cleaned up (and added to the free list) 14928 before the walk of unreferenced happens and a reference to that is SV 14929 added to the temps stack. At which point we have the same SV considered 14930 to be in use, and free to be re-used. Not good. 14931 */ 14932 if (dsv && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dsv)) { 14933 assert(param->unreferenced); 14934 av_push(param->unreferenced, SvREFCNT_inc(dsv)); 14935 } 14936 14937 return dsv; 14938} 14939 14940/* duplicate a context */ 14941 14942PERL_CONTEXT * 14943Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param) 14944{ 14945 PERL_CONTEXT *ncxs; 14946 14947 PERL_ARGS_ASSERT_CX_DUP; 14948 14949 if (!cxs) 14950 return (PERL_CONTEXT*)NULL; 14951 14952 /* look for it in the table first */ 14953 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs); 14954 if (ncxs) 14955 return ncxs; 14956 14957 /* create anew and remember what it is */ 14958 Newx(ncxs, max + 1, PERL_CONTEXT); 14959 ptr_table_store(PL_ptr_table, cxs, ncxs); 14960 Copy(cxs, ncxs, max + 1, PERL_CONTEXT); 14961 14962 while (ix >= 0) { 14963 PERL_CONTEXT * const ncx = &ncxs[ix]; 14964 if (CxTYPE(ncx) == CXt_SUBST) { 14965 Perl_croak(aTHX_ "Cloning substitution context is unimplemented"); 14966 } 14967 else { 14968 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl); 14969 switch (CxTYPE(ncx)) { 14970 case CXt_SUB: 14971 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param); 14972 if(CxHASARGS(ncx)){ 14973 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param); 14974 } else { 14975 ncx->blk_sub.savearray = NULL; 14976 } 14977 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table, 14978 ncx->blk_sub.prevcomppad); 14979 break; 14980 case CXt_EVAL: 14981 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv, 14982 param); 14983 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */ 14984 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param); 14985 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param); 14986 /* XXX what to do with cur_top_env ???? */ 14987 break; 14988 case CXt_LOOP_LAZYSV: 14989 ncx->blk_loop.state_u.lazysv.end 14990 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param); 14991 /* Fallthrough: duplicate lazysv.cur by using the ary.ary 14992 duplication code instead. 14993 We are taking advantage of (1) av_dup_inc and sv_dup_inc 14994 actually being the same function, and (2) order 14995 equivalence of the two unions. 14996 We can assert the later [but only at run time :-(] */ 14997 assert ((void *) &ncx->blk_loop.state_u.ary.ary == 14998 (void *) &ncx->blk_loop.state_u.lazysv.cur); 14999 /* FALLTHROUGH */ 15000 case CXt_LOOP_ARY: 15001 ncx->blk_loop.state_u.ary.ary 15002 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param); 15003 /* FALLTHROUGH */ 15004 case CXt_LOOP_LIST: 15005 case CXt_LOOP_LAZYIV: 15006 /* code common to all 'for' CXt_LOOP_* types */ 15007 ncx->blk_loop.itersave = 15008 sv_dup_inc(ncx->blk_loop.itersave, param); 15009 if (CxPADLOOP(ncx)) { 15010 PADOFFSET off = ncx->blk_loop.itervar_u.svp 15011 - &CX_CURPAD_SV(ncx->blk_loop, 0); 15012 ncx->blk_loop.oldcomppad = 15013 (PAD*)ptr_table_fetch(PL_ptr_table, 15014 ncx->blk_loop.oldcomppad); 15015 ncx->blk_loop.itervar_u.svp = 15016 &CX_CURPAD_SV(ncx->blk_loop, off); 15017 } 15018 else { 15019 /* this copies the GV if CXp_FOR_GV, or the SV for an 15020 * alias (for \$x (...)) - relies on gv_dup being the 15021 * same as sv_dup */ 15022 ncx->blk_loop.itervar_u.gv 15023 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv, 15024 param); 15025 } 15026 break; 15027 case CXt_LOOP_PLAIN: 15028 break; 15029 case CXt_FORMAT: 15030 ncx->blk_format.prevcomppad = 15031 (PAD*)ptr_table_fetch(PL_ptr_table, 15032 ncx->blk_format.prevcomppad); 15033 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param); 15034 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param); 15035 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv, 15036 param); 15037 break; 15038 case CXt_GIVEN: 15039 ncx->blk_givwhen.defsv_save = 15040 sv_dup_inc(ncx->blk_givwhen.defsv_save, param); 15041 break; 15042 case CXt_BLOCK: 15043 case CXt_NULL: 15044 case CXt_WHEN: 15045 case CXt_DEFER: 15046 break; 15047 } 15048 } 15049 --ix; 15050 } 15051 return ncxs; 15052} 15053 15054/* 15055=for apidoc si_dup 15056 15057Duplicate a stack info structure, returning a pointer to the cloned object. 15058 15059=cut 15060*/ 15061 15062PERL_SI * 15063Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param) 15064{ 15065 PERL_SI *nsi; 15066 15067 PERL_ARGS_ASSERT_SI_DUP; 15068 15069 if (!si) 15070 return (PERL_SI*)NULL; 15071 15072 /* look for it in the table first */ 15073 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si); 15074 if (nsi) 15075 return nsi; 15076 15077 /* create anew and remember what it is */ 15078 Newx(nsi, 1, PERL_SI); 15079 ptr_table_store(PL_ptr_table, si, nsi); 15080 15081 nsi->si_stack = av_dup_inc(si->si_stack, param); 15082 nsi->si_cxix = si->si_cxix; 15083 nsi->si_cxsubix = si->si_cxsubix; 15084 nsi->si_cxmax = si->si_cxmax; 15085 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param); 15086 nsi->si_type = si->si_type; 15087 nsi->si_prev = si_dup(si->si_prev, param); 15088 nsi->si_next = si_dup(si->si_next, param); 15089 nsi->si_markoff = si->si_markoff; 15090#if defined DEBUGGING && !defined DEBUGGING_RE_ONLY 15091 nsi->si_stack_hwm = 0; 15092#endif 15093 15094 return nsi; 15095} 15096 15097#define POPINT(ss,ix) ((ss)[--(ix)].any_i32) 15098#define TOPINT(ss,ix) ((ss)[ix].any_i32) 15099#define POPLONG(ss,ix) ((ss)[--(ix)].any_long) 15100#define TOPLONG(ss,ix) ((ss)[ix].any_long) 15101#define POPIV(ss,ix) ((ss)[--(ix)].any_iv) 15102#define TOPIV(ss,ix) ((ss)[ix].any_iv) 15103#define POPUV(ss,ix) ((ss)[--(ix)].any_uv) 15104#define TOPUV(ss,ix) ((ss)[ix].any_uv) 15105#define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool) 15106#define TOPBOOL(ss,ix) ((ss)[ix].any_bool) 15107#define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr) 15108#define TOPPTR(ss,ix) ((ss)[ix].any_ptr) 15109#define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr) 15110#define TOPDPTR(ss,ix) ((ss)[ix].any_dptr) 15111#define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr) 15112#define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr) 15113 15114/* XXXXX todo */ 15115#define pv_dup_inc(p) SAVEPV(p) 15116#define pv_dup(p) SAVEPV(p) 15117#define svp_dup_inc(p,pp) any_dup(p,pp) 15118 15119/* map any object to the new equivalent - either something in the 15120 * ptr table, or something in the interpreter structure 15121 */ 15122 15123void * 15124Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl) 15125{ 15126 void *ret; 15127 15128 PERL_ARGS_ASSERT_ANY_DUP; 15129 15130 if (!v) 15131 return (void*)NULL; 15132 15133 /* look for it in the table first */ 15134 ret = ptr_table_fetch(PL_ptr_table, v); 15135 if (ret) 15136 return ret; 15137 15138 /* see if it is part of the interpreter structure */ 15139 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1)) 15140 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl)); 15141 else { 15142 ret = v; 15143 } 15144 15145 return ret; 15146} 15147 15148/* 15149=for apidoc ss_dup 15150 15151Duplicate the save stack, returning a pointer to the cloned object. 15152 15153=cut 15154*/ 15155 15156ANY * 15157Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param) 15158{ 15159 ANY * const ss = proto_perl->Isavestack; 15160 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH; 15161 I32 ix = proto_perl->Isavestack_ix; 15162 ANY *nss; 15163 const SV *sv; 15164 const GV *gv; 15165 const AV *av; 15166 const HV *hv; 15167 char *pv; /* no const deliberately */ 15168 void* ptr; 15169 int intval; 15170 long longval; 15171 GP *gp; 15172 IV iv; 15173 I32 i; 15174 char *c = NULL; 15175 void (*dptr) (void*); 15176 void (*dxptr) (pTHX_ void*); 15177 15178 PERL_ARGS_ASSERT_SS_DUP; 15179 15180 Newx(nss, max, ANY); 15181 15182 while (ix > 0) { 15183 const UV uv = POPUV(ss,ix); 15184 const U8 type = (U8)uv & SAVE_MASK; 15185 15186 TOPUV(nss,ix) = uv; 15187 switch (type) { 15188 case SAVEt_CLEARSV: 15189 case SAVEt_CLEARPADRANGE: 15190 break; 15191 case SAVEt_HELEM: /* hash element */ 15192 case SAVEt_SV: /* scalar reference */ 15193 sv = (const SV *)POPPTR(ss,ix); 15194 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param)); 15195 /* FALLTHROUGH */ 15196 case SAVEt_ITEM: /* normal string */ 15197 case SAVEt_GVSV: /* scalar slot in GV */ 15198 sv = (const SV *)POPPTR(ss,ix); 15199 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15200 if (type == SAVEt_SV) 15201 break; 15202 /* FALLTHROUGH */ 15203 case SAVEt_FREESV: 15204 case SAVEt_MORTALIZESV: 15205 case SAVEt_READONLY_OFF: 15206 sv = (const SV *)POPPTR(ss,ix); 15207 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15208 break; 15209 case SAVEt_FREEPADNAME: 15210 ptr = POPPTR(ss,ix); 15211 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param); 15212 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++; 15213 break; 15214 case SAVEt_SHARED_PVREF: /* char* in shared space */ 15215 c = (char*)POPPTR(ss,ix); 15216 TOPPTR(nss,ix) = savesharedpv(c); 15217 ptr = POPPTR(ss,ix); 15218 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15219 break; 15220 case SAVEt_GENERIC_SVREF: /* generic sv */ 15221 case SAVEt_SVREF: /* scalar reference */ 15222 sv = (const SV *)POPPTR(ss,ix); 15223 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15224 if (type == SAVEt_SVREF) 15225 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix)); 15226 ptr = POPPTR(ss,ix); 15227 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */ 15228 /* this feels very strange, we have a **SV from one thread, 15229 * we copy the SV, but dont change the **SV. But in this thread 15230 * the target of the **SV could be something from the *other* thread. 15231 * So how can this possibly work correctly? */ 15232 break; 15233 case SAVEt_RCPV: 15234 pv = (char *)POPPTR(ss,ix); 15235 TOPPTR(nss,ix) = rcpv_copy(pv); 15236 ptr = POPPTR(ss,ix); 15237 (void)rcpv_copy(*((char **)ptr)); 15238 TOPPTR(nss,ix) = ptr; 15239 /* XXXXX: see comment above. */ 15240 break; 15241 case SAVEt_GVSLOT: /* any slot in GV */ 15242 sv = (const SV *)POPPTR(ss,ix); 15243 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15244 ptr = POPPTR(ss,ix); 15245 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */ 15246 sv = (const SV *)POPPTR(ss,ix); 15247 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15248 break; 15249 case SAVEt_HV: /* hash reference */ 15250 case SAVEt_AV: /* array reference */ 15251 sv = (const SV *) POPPTR(ss,ix); 15252 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15253 /* FALLTHROUGH */ 15254 case SAVEt_COMPPAD: 15255 case SAVEt_NSTAB: 15256 sv = (const SV *) POPPTR(ss,ix); 15257 TOPPTR(nss,ix) = sv_dup(sv, param); 15258 break; 15259 case SAVEt_INT: /* int reference */ 15260 ptr = POPPTR(ss,ix); 15261 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15262 intval = (int)POPINT(ss,ix); 15263 TOPINT(nss,ix) = intval; 15264 break; 15265 case SAVEt_LONG: /* long reference */ 15266 ptr = POPPTR(ss,ix); 15267 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15268 longval = (long)POPLONG(ss,ix); 15269 TOPLONG(nss,ix) = longval; 15270 break; 15271 case SAVEt_I32: /* I32 reference */ 15272 ptr = POPPTR(ss,ix); 15273 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15274 i = POPINT(ss,ix); 15275 TOPINT(nss,ix) = i; 15276 break; 15277 case SAVEt_IV: /* IV reference */ 15278 case SAVEt_STRLEN: /* STRLEN/size_t ref */ 15279 ptr = POPPTR(ss,ix); 15280 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15281 iv = POPIV(ss,ix); 15282 TOPIV(nss,ix) = iv; 15283 break; 15284 case SAVEt_TMPSFLOOR: 15285 iv = POPIV(ss,ix); 15286 TOPIV(nss,ix) = iv; 15287 break; 15288 case SAVEt_HPTR: /* HV* reference */ 15289 case SAVEt_APTR: /* AV* reference */ 15290 case SAVEt_SPTR: /* SV* reference */ 15291 ptr = POPPTR(ss,ix); 15292 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15293 sv = (const SV *)POPPTR(ss,ix); 15294 TOPPTR(nss,ix) = sv_dup(sv, param); 15295 break; 15296 case SAVEt_VPTR: /* random* reference */ 15297 ptr = POPPTR(ss,ix); 15298 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15299 /* FALLTHROUGH */ 15300 case SAVEt_STRLEN_SMALL: 15301 case SAVEt_INT_SMALL: 15302 case SAVEt_I32_SMALL: 15303 case SAVEt_I16: /* I16 reference */ 15304 case SAVEt_I8: /* I8 reference */ 15305 case SAVEt_BOOL: 15306 ptr = POPPTR(ss,ix); 15307 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15308 break; 15309 case SAVEt_GENERIC_PVREF: /* generic char* */ 15310 case SAVEt_PPTR: /* char* reference */ 15311 ptr = POPPTR(ss,ix); 15312 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15313 c = (char*)POPPTR(ss,ix); 15314 TOPPTR(nss,ix) = pv_dup(c); 15315 break; 15316 case SAVEt_GP: /* scalar reference */ 15317 gp = (GP*)POPPTR(ss,ix); 15318 TOPPTR(nss,ix) = gp = gp_dup(gp, param); 15319 (void)GpREFCNT_inc(gp); 15320 gv = (const GV *)POPPTR(ss,ix); 15321 TOPPTR(nss,ix) = gv_dup_inc(gv, param); 15322 break; 15323 case SAVEt_FREEOP: 15324 ptr = POPPTR(ss,ix); 15325 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) { 15326 /* these are assumed to be refcounted properly */ 15327 OP *o; 15328 switch (((OP*)ptr)->op_type) { 15329 case OP_LEAVESUB: 15330 case OP_LEAVESUBLV: 15331 case OP_LEAVEEVAL: 15332 case OP_LEAVE: 15333 case OP_SCOPE: 15334 case OP_LEAVEWRITE: 15335 TOPPTR(nss,ix) = ptr; 15336 o = (OP*)ptr; 15337 OP_REFCNT_LOCK; 15338 (void) OpREFCNT_inc(o); 15339 OP_REFCNT_UNLOCK; 15340 break; 15341 default: 15342 TOPPTR(nss,ix) = NULL; 15343 break; 15344 } 15345 } 15346 else 15347 TOPPTR(nss,ix) = NULL; 15348 break; 15349 case SAVEt_FREECOPHH: 15350 ptr = POPPTR(ss,ix); 15351 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr); 15352 break; 15353 case SAVEt_ADELETE: 15354 av = (const AV *)POPPTR(ss,ix); 15355 TOPPTR(nss,ix) = av_dup_inc(av, param); 15356 i = POPINT(ss,ix); 15357 TOPINT(nss,ix) = i; 15358 break; 15359 case SAVEt_DELETE: 15360 hv = (const HV *)POPPTR(ss,ix); 15361 TOPPTR(nss,ix) = hv_dup_inc(hv, param); 15362 i = POPINT(ss,ix); 15363 TOPINT(nss,ix) = i; 15364 /* FALLTHROUGH */ 15365 case SAVEt_FREEPV: 15366 c = (char*)POPPTR(ss,ix); 15367 TOPPTR(nss,ix) = pv_dup_inc(c); 15368 break; 15369 case SAVEt_FREERCPV: 15370 c = (char *)POPPTR(ss,ix); 15371 TOPPTR(nss,ix) = rcpv_copy(c); 15372 break; 15373 case SAVEt_STACK_POS: /* Position on Perl stack */ 15374 i = POPINT(ss,ix); 15375 TOPINT(nss,ix) = i; 15376 break; 15377 case SAVEt_DESTRUCTOR: 15378 ptr = POPPTR(ss,ix); 15379 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */ 15380 dptr = POPDPTR(ss,ix); 15381 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*), 15382 any_dup(FPTR2DPTR(void *, dptr), 15383 proto_perl)); 15384 break; 15385 case SAVEt_DESTRUCTOR_X: 15386 ptr = POPPTR(ss,ix); 15387 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */ 15388 dxptr = POPDXPTR(ss,ix); 15389 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*), 15390 any_dup(FPTR2DPTR(void *, dxptr), 15391 proto_perl)); 15392 break; 15393 case SAVEt_REGCONTEXT: 15394 case SAVEt_ALLOC: 15395 ix -= uv >> SAVE_TIGHT_SHIFT; 15396 break; 15397 case SAVEt_AELEM: /* array element */ 15398 sv = (const SV *)POPPTR(ss,ix); 15399 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param)); 15400 iv = POPIV(ss,ix); 15401 TOPIV(nss,ix) = iv; 15402 av = (const AV *)POPPTR(ss,ix); 15403 TOPPTR(nss,ix) = av_dup_inc(av, param); 15404 break; 15405 case SAVEt_OP: 15406 ptr = POPPTR(ss,ix); 15407 TOPPTR(nss,ix) = ptr; 15408 break; 15409 case SAVEt_HINTS_HH: 15410 hv = (const HV *)POPPTR(ss,ix); 15411 TOPPTR(nss,ix) = hv_dup_inc(hv, param); 15412 /* FALLTHROUGH */ 15413 case SAVEt_HINTS: 15414 ptr = POPPTR(ss,ix); 15415 ptr = cophh_copy((COPHH*)ptr); 15416 TOPPTR(nss,ix) = ptr; 15417 i = POPINT(ss,ix); 15418 TOPINT(nss,ix) = i; 15419 break; 15420 case SAVEt_PADSV_AND_MORTALIZE: 15421 longval = (long)POPLONG(ss,ix); 15422 TOPLONG(nss,ix) = longval; 15423 ptr = POPPTR(ss,ix); 15424 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); 15425 sv = (const SV *)POPPTR(ss,ix); 15426 TOPPTR(nss,ix) = sv_dup_inc(sv, param); 15427 break; 15428 case SAVEt_SET_SVFLAGS: 15429 i = POPINT(ss,ix); 15430 TOPINT(nss,ix) = i; 15431 i = POPINT(ss,ix); 15432 TOPINT(nss,ix) = i; 15433 sv = (const SV *)POPPTR(ss,ix); 15434 TOPPTR(nss,ix) = sv_dup(sv, param); 15435 break; 15436 case SAVEt_CURCOP_WARNINGS: 15437 /* FALLTHROUGH */ 15438 case SAVEt_COMPILE_WARNINGS: 15439 ptr = POPPTR(ss,ix); 15440 TOPPTR(nss,ix) = DUP_WARNINGS((char*)ptr); 15441 break; 15442 case SAVEt_PARSER: 15443 ptr = POPPTR(ss,ix); 15444 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param); 15445 break; 15446 default: 15447 Perl_croak(aTHX_ 15448 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type); 15449 } 15450 } 15451 15452 return nss; 15453} 15454 15455 15456/* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE 15457 * flag to the result. This is done for each stash before cloning starts, 15458 * so we know which stashes want their objects cloned */ 15459 15460static void 15461do_mark_cloneable_stash(pTHX_ SV *const sv) 15462{ 15463 const HEK * const hvname = HvNAME_HEK((const HV *)sv); 15464 if (hvname) { 15465 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0); 15466 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */ 15467 if (cloner && GvCV(cloner)) { 15468 dSP; 15469 UV status; 15470 15471 ENTER; 15472 SAVETMPS; 15473 PUSHMARK(SP); 15474 mXPUSHs(newSVhek(hvname)); 15475 PUTBACK; 15476 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR); 15477 SPAGAIN; 15478 status = POPu; 15479 PUTBACK; 15480 FREETMPS; 15481 LEAVE; 15482 if (status) 15483 SvFLAGS(sv) &= ~SVphv_CLONEABLE; 15484 } 15485 } 15486} 15487 15488 15489 15490/* 15491=for apidoc perl_clone 15492 15493Create and return a new interpreter by cloning the current one. 15494 15495C<perl_clone> takes these flags as parameters: 15496 15497C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also, 15498without it we only clone the data and zero the stacks, 15499with it we copy the stacks and the new perl interpreter is 15500ready to run at the exact same point as the previous one. 15501The pseudo-fork code uses C<COPY_STACKS> while the 15502threads->create doesn't. 15503 15504C<CLONEf_KEEP_PTR_TABLE> - 15505C<perl_clone> keeps a ptr_table with the pointer of the old 15506variable as a key and the new variable as a value, 15507this allows it to check if something has been cloned and not 15508clone it again, but rather just use the value and increase the 15509refcount. 15510If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill the ptr_table 15511using the function S<C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>>. 15512A reason to keep it around is if you want to dup some of your own 15513variables which are outside the graph that perl scans. 15514 15515C<CLONEf_CLONE_HOST> - 15516This is a win32 thing, it is ignored on unix, it tells perl's 15517win32host code (which is c++) to clone itself, this is needed on 15518win32 if you want to run two threads at the same time, 15519if you just want to do some stuff in a separate perl interpreter 15520and then throw it away and return to the original one, 15521you don't need to do anything. 15522 15523=cut 15524*/ 15525 15526/* XXX the above needs expanding by someone who actually understands it ! */ 15527EXTERN_C PerlInterpreter * 15528perl_clone_host(PerlInterpreter* proto_perl, UV flags); 15529 15530PerlInterpreter * 15531perl_clone(PerlInterpreter *proto_perl, UV flags) 15532{ 15533#ifdef PERL_IMPLICIT_SYS 15534 15535 PERL_ARGS_ASSERT_PERL_CLONE; 15536 15537 /* perlhost.h so we need to call into it 15538 to clone the host, CPerlHost should have a c interface, sky */ 15539 15540#ifndef __amigaos4__ 15541 if (flags & CLONEf_CLONE_HOST) { 15542 return perl_clone_host(proto_perl,flags); 15543 } 15544#endif 15545 return perl_clone_using(proto_perl, flags, 15546 proto_perl->IMem, 15547 proto_perl->IMemShared, 15548 proto_perl->IMemParse, 15549 proto_perl->IEnv, 15550 proto_perl->IStdIO, 15551 proto_perl->ILIO, 15552 proto_perl->IDir, 15553 proto_perl->ISock, 15554 proto_perl->IProc); 15555} 15556 15557PerlInterpreter * 15558perl_clone_using(PerlInterpreter *proto_perl, UV flags, 15559 struct IPerlMem* ipM, struct IPerlMem* ipMS, 15560 struct IPerlMem* ipMP, struct IPerlEnv* ipE, 15561 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO, 15562 struct IPerlDir* ipD, struct IPerlSock* ipS, 15563 struct IPerlProc* ipP) 15564{ 15565 /* XXX many of the string copies here can be optimized if they're 15566 * constants; they need to be allocated as common memory and just 15567 * their pointers copied. */ 15568 15569 IV i; 15570 CLONE_PARAMS clone_params; 15571 CLONE_PARAMS* const param = &clone_params; 15572 15573 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter)); 15574 15575 PERL_ARGS_ASSERT_PERL_CLONE_USING; 15576#else /* !PERL_IMPLICIT_SYS */ 15577 IV i; 15578 CLONE_PARAMS clone_params; 15579 CLONE_PARAMS* param = &clone_params; 15580 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter)); 15581 15582 PERL_ARGS_ASSERT_PERL_CLONE; 15583#endif /* PERL_IMPLICIT_SYS */ 15584 15585 /* for each stash, determine whether its objects should be cloned */ 15586 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK); 15587 my_perl->Iphase = PERL_PHASE_CONSTRUCT; 15588 PERL_SET_THX(my_perl); 15589 15590#ifdef DEBUGGING 15591 PoisonNew(my_perl, 1, PerlInterpreter); 15592 PL_op = NULL; 15593 PL_curcop = NULL; 15594 PL_defstash = NULL; /* may be used by perl malloc() */ 15595 PL_markstack = 0; 15596 PL_scopestack = 0; 15597 PL_scopestack_name = 0; 15598 PL_savestack = 0; 15599 PL_savestack_ix = 0; 15600 PL_savestack_max = -1; 15601 PL_sig_pending = 0; 15602 PL_parser = NULL; 15603 PL_eval_begin_nest_depth = proto_perl->Ieval_begin_nest_depth; 15604 Zero(&PL_debug_pad, 1, struct perl_debug_pad); 15605 Zero(&PL_padname_undef, 1, PADNAME); 15606 Zero(&PL_padname_const, 1, PADNAME); 15607# ifdef DEBUG_LEAKING_SCALARS 15608 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000; 15609# endif 15610# ifdef PERL_TRACE_OPS 15611 Zero(PL_op_exec_cnt, OP_max+2, UV); 15612# endif 15613#else /* !DEBUGGING */ 15614 Zero(my_perl, 1, PerlInterpreter); 15615#endif /* DEBUGGING */ 15616 15617#ifdef PERL_IMPLICIT_SYS 15618 /* host pointers */ 15619 PL_Mem = ipM; 15620 PL_MemShared = ipMS; 15621 PL_MemParse = ipMP; 15622 PL_Env = ipE; 15623 PL_StdIO = ipStd; 15624 PL_LIO = ipLIO; 15625 PL_Dir = ipD; 15626 PL_Sock = ipS; 15627 PL_Proc = ipP; 15628#endif /* PERL_IMPLICIT_SYS */ 15629 15630 15631 param->flags = flags; 15632 /* Nothing in the core code uses this, but we make it available to 15633 extensions (using mg_dup). */ 15634 param->proto_perl = proto_perl; 15635 /* Likely nothing will use this, but it is initialised to be consistent 15636 with Perl_clone_params_new(). */ 15637 param->new_perl = my_perl; 15638 param->unreferenced = NULL; 15639 15640 15641 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl); 15642 15643 PL_body_arenas = NULL; 15644 Zero(&PL_body_roots, 1, PL_body_roots); 15645 15646 PL_sv_count = 0; 15647 PL_sv_root = NULL; 15648 PL_sv_arenaroot = NULL; 15649 15650 PL_debug = proto_perl->Idebug; 15651 15652 /* dbargs array probably holds garbage */ 15653 PL_dbargs = NULL; 15654 15655 PL_compiling = proto_perl->Icompiling; 15656 15657 /* pseudo environmental stuff */ 15658 PL_origargc = proto_perl->Iorigargc; 15659 PL_origargv = proto_perl->Iorigargv; 15660 15661#ifndef NO_TAINT_SUPPORT 15662 /* Set tainting stuff before PerlIO_debug can possibly get called */ 15663 PL_tainting = proto_perl->Itainting; 15664 PL_taint_warn = proto_perl->Itaint_warn; 15665#else 15666 PL_tainting = FALSE; 15667 PL_taint_warn = FALSE; 15668#endif 15669 15670 PL_minus_c = proto_perl->Iminus_c; 15671 15672 PL_localpatches = proto_perl->Ilocalpatches; 15673 PL_splitstr = SAVEPV(proto_perl->Isplitstr); 15674 PL_minus_n = proto_perl->Iminus_n; 15675 PL_minus_p = proto_perl->Iminus_p; 15676 PL_minus_l = proto_perl->Iminus_l; 15677 PL_minus_a = proto_perl->Iminus_a; 15678 PL_minus_E = proto_perl->Iminus_E; 15679 PL_minus_F = proto_perl->Iminus_F; 15680 PL_doswitches = proto_perl->Idoswitches; 15681 PL_dowarn = proto_perl->Idowarn; 15682#ifdef PERL_SAWAMPERSAND 15683 PL_sawampersand = proto_perl->Isawampersand; 15684#endif 15685 PL_unsafe = proto_perl->Iunsafe; 15686 PL_perldb = proto_perl->Iperldb; 15687 PL_perl_destruct_level = proto_perl->Iperl_destruct_level; 15688 PL_exit_flags = proto_perl->Iexit_flags; 15689 15690 /* XXX time(&PL_basetime) when asked for? */ 15691 PL_basetime = proto_perl->Ibasetime; 15692 15693 PL_maxsysfd = proto_perl->Imaxsysfd; 15694 PL_statusvalue = proto_perl->Istatusvalue; 15695#ifdef __VMS 15696 PL_statusvalue_vms = proto_perl->Istatusvalue_vms; 15697#else 15698 PL_statusvalue_posix = proto_perl->Istatusvalue_posix; 15699#endif 15700 15701 /* RE engine related */ 15702 PL_regmatch_slab = NULL; 15703 PL_reg_curpm = NULL; 15704 15705 PL_sub_generation = proto_perl->Isub_generation; 15706 15707 /* funky return mechanisms */ 15708 PL_forkprocess = proto_perl->Iforkprocess; 15709 15710 /* internal state */ 15711 PL_main_start = proto_perl->Imain_start; 15712 PL_eval_root = proto_perl->Ieval_root; 15713 PL_eval_start = proto_perl->Ieval_start; 15714 15715 PL_filemode = proto_perl->Ifilemode; 15716 PL_lastfd = proto_perl->Ilastfd; 15717 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */ 15718 PL_gensym = proto_perl->Igensym; 15719 15720 PL_laststatval = proto_perl->Ilaststatval; 15721 PL_laststype = proto_perl->Ilaststype; 15722 PL_mess_sv = NULL; 15723 15724 PL_profiledata = NULL; 15725 15726 PL_generation = proto_perl->Igeneration; 15727 15728 PL_in_clean_objs = proto_perl->Iin_clean_objs; 15729 PL_in_clean_all = proto_perl->Iin_clean_all; 15730 15731 PL_delaymagic_uid = proto_perl->Idelaymagic_uid; 15732 PL_delaymagic_euid = proto_perl->Idelaymagic_euid; 15733 PL_delaymagic_gid = proto_perl->Idelaymagic_gid; 15734 PL_delaymagic_egid = proto_perl->Idelaymagic_egid; 15735 PL_nomemok = proto_perl->Inomemok; 15736 PL_an = proto_perl->Ian; 15737 PL_evalseq = proto_perl->Ievalseq; 15738 PL_origalen = proto_perl->Iorigalen; 15739 15740 PL_sighandlerp = proto_perl->Isighandlerp; 15741 PL_sighandler1p = proto_perl->Isighandler1p; 15742 PL_sighandler3p = proto_perl->Isighandler3p; 15743 15744 PL_runops = proto_perl->Irunops; 15745 15746 PL_subline = proto_perl->Isubline; 15747 15748 PL_cv_has_eval = proto_perl->Icv_has_eval; 15749 /* Unicode features (see perlrun/-C) */ 15750 PL_unicode = proto_perl->Iunicode; 15751 15752 /* Pre-5.8 signals control */ 15753 PL_signals = proto_perl->Isignals; 15754 15755 /* times() ticks per second */ 15756 PL_clocktick = proto_perl->Iclocktick; 15757 15758 /* Recursion stopper for PerlIO_find_layer */ 15759 PL_in_load_module = proto_perl->Iin_load_module; 15760 15761 /* Not really needed/useful since the reenrant_retint is "volatile", 15762 * but do it for consistency's sake. */ 15763 PL_reentrant_retint = proto_perl->Ireentrant_retint; 15764 15765 /* Hooks to shared SVs and locks. */ 15766 PL_sharehook = proto_perl->Isharehook; 15767 PL_lockhook = proto_perl->Ilockhook; 15768 PL_unlockhook = proto_perl->Iunlockhook; 15769 PL_threadhook = proto_perl->Ithreadhook; 15770 PL_destroyhook = proto_perl->Idestroyhook; 15771 PL_signalhook = proto_perl->Isignalhook; 15772 15773 PL_globhook = proto_perl->Iglobhook; 15774 15775 PL_srand_called = proto_perl->Isrand_called; 15776 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE); 15777 PL_srand_override = proto_perl->Isrand_override; 15778 PL_srand_override_next = proto_perl->Isrand_override_next; 15779 15780 if (flags & CLONEf_COPY_STACKS) { 15781 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */ 15782 PL_tmps_ix = proto_perl->Itmps_ix; 15783 PL_tmps_max = proto_perl->Itmps_max; 15784 PL_tmps_floor = proto_perl->Itmps_floor; 15785 15786 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix] 15787 * NOTE: unlike the others! */ 15788 PL_scopestack_ix = proto_perl->Iscopestack_ix; 15789 PL_scopestack_max = proto_perl->Iscopestack_max; 15790 15791 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix] 15792 * NOTE: unlike the others! */ 15793 PL_savestack_ix = proto_perl->Isavestack_ix; 15794 PL_savestack_max = proto_perl->Isavestack_max; 15795 } 15796 15797 PL_start_env = proto_perl->Istart_env; /* XXXXXX */ 15798 PL_top_env = &PL_start_env; 15799 15800 PL_op = proto_perl->Iop; 15801 15802 PL_Sv = NULL; 15803 PL_Xpv = (XPV*)NULL; 15804 my_perl->Ina = proto_perl->Ina; 15805 15806 PL_statcache = proto_perl->Istatcache; 15807 15808#ifndef NO_TAINT_SUPPORT 15809 PL_tainted = proto_perl->Itainted; 15810#else 15811 PL_tainted = FALSE; 15812#endif 15813 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */ 15814 15815 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */ 15816 15817 PL_restartjmpenv = proto_perl->Irestartjmpenv; 15818 PL_restartop = proto_perl->Irestartop; 15819 PL_in_eval = proto_perl->Iin_eval; 15820 PL_delaymagic = proto_perl->Idelaymagic; 15821 PL_phase = proto_perl->Iphase; 15822 PL_localizing = proto_perl->Ilocalizing; 15823 15824 PL_hv_fetch_ent_mh = NULL; 15825 PL_modcount = proto_perl->Imodcount; 15826 PL_lastgotoprobe = NULL; 15827 PL_dumpindent = proto_perl->Idumpindent; 15828 15829 PL_efloatbuf = NULL; /* reinits on demand */ 15830 PL_efloatsize = 0; /* reinits on demand */ 15831 15832 /* regex stuff */ 15833 15834 PL_colorset = 0; /* reinits PL_colors[] */ 15835 /*PL_colors[6] = {0,0,0,0,0,0};*/ 15836 15837 /* Pluggable optimizer */ 15838 PL_peepp = proto_perl->Ipeepp; 15839 PL_rpeepp = proto_perl->Irpeepp; 15840 /* op_free() hook */ 15841 PL_opfreehook = proto_perl->Iopfreehook; 15842 15843# ifdef PERL_MEM_LOG 15844 Zero(PL_mem_log, sizeof(PL_mem_log), char); 15845# endif 15846 15847#ifdef USE_REENTRANT_API 15848 /* XXX: things like -Dm will segfault here in perlio, but doing 15849 * PERL_SET_CONTEXT(proto_perl); 15850 * breaks too many other things 15851 */ 15852 Perl_reentrant_init(aTHX); 15853#endif 15854 15855 /* create SV map for pointer relocation */ 15856 PL_ptr_table = ptr_table_new(); 15857 15858 /* initialize these special pointers as early as possible */ 15859 init_constants(); 15860 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef); 15861 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no); 15862 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero); 15863 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes); 15864 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const, 15865 &PL_padname_const); 15866 15867 /* create (a non-shared!) shared string table */ 15868 PL_strtab = newHV(); 15869 HvSHAREKEYS_off(PL_strtab); 15870 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab)); 15871 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab); 15872 15873 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*); 15874 15875 PL_compiling.cop_file = rcpv_copy(proto_perl->Icompiling.cop_file); 15876 15877 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling); 15878 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings); 15879 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling))); 15880 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl); 15881 15882 param->stashes = newAV(); /* Setup array of objects to call clone on */ 15883 /* This makes no difference to the implementation, as it always pushes 15884 and shifts pointers to other SVs without changing their reference 15885 count, with the array becoming empty before it is freed. However, it 15886 makes it conceptually clear what is going on, and will avoid some 15887 work inside av.c, filling slots between AvFILL() and AvMAX() with 15888 &PL_sv_undef, and SvREFCNT_dec()ing those. */ 15889 AvREAL_off(param->stashes); 15890 15891 if (!(flags & CLONEf_COPY_STACKS)) { 15892 param->unreferenced = newAV(); 15893 } 15894 15895#ifdef PERLIO_LAYERS 15896 /* Clone PerlIO tables as soon as we can handle general xx_dup() */ 15897 PerlIO_clone(aTHX_ proto_perl, param); 15898#endif 15899 15900 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param); 15901 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param); 15902 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param); 15903 PL_origfilename = SAVEPV(proto_perl->Iorigfilename); 15904 PL_xsubfilename = proto_perl->Ixsubfilename; 15905 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param); 15906 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param); 15907 15908 PL_hook__require__before = sv_dup_inc(proto_perl->Ihook__require__before, param); 15909 PL_hook__require__after = sv_dup_inc(proto_perl->Ihook__require__after, param); 15910 15911 /* switches */ 15912 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param); 15913 PL_inplace = SAVEPV(proto_perl->Iinplace); 15914 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param); 15915 15916 /* magical thingies */ 15917 15918 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */ 15919 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */ 15920 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */ 15921 15922 15923 /* Clone the regex array */ 15924 /* ORANGE FIXME for plugins, probably in the SV dup code. 15925 newSViv(PTR2IV(CALLREGDUPE( 15926 INT2PTR(REGEXP *, SvIVX(regex)), param)))) 15927 */ 15928 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param); 15929 PL_regex_pad = AvARRAY(PL_regex_padav); 15930 15931 PL_stashpadmax = proto_perl->Istashpadmax; 15932 PL_stashpadix = proto_perl->Istashpadix ; 15933 Newx(PL_stashpad, PL_stashpadmax, HV *); 15934 { 15935 PADOFFSET o = 0; 15936 for (; o < PL_stashpadmax; ++o) 15937 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param); 15938 } 15939 15940 /* shortcuts to various I/O objects */ 15941 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param); 15942 PL_stdingv = gv_dup(proto_perl->Istdingv, param); 15943 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param); 15944 PL_defgv = gv_dup(proto_perl->Idefgv, param); 15945 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param); 15946 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param); 15947 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param); 15948 15949 /* shortcuts to regexp stuff */ 15950 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param); 15951 15952 /* shortcuts to misc objects */ 15953 PL_errgv = gv_dup(proto_perl->Ierrgv, param); 15954 15955 /* shortcuts to debugging objects */ 15956 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param); 15957 PL_DBline = gv_dup_inc(proto_perl->IDBline, param); 15958 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param); 15959 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param); 15960 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param); 15961 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param); 15962 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV); 15963 15964 /* symbol tables */ 15965 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param); 15966 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param); 15967 PL_debstash = hv_dup(proto_perl->Idebstash, param); 15968 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param); 15969 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param); 15970 15971 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param); 15972 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param); 15973 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param); 15974 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param); 15975 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param); 15976 PL_endav = av_dup_inc(proto_perl->Iendav, param); 15977 PL_checkav = av_dup_inc(proto_perl->Icheckav, param); 15978 PL_initav = av_dup_inc(proto_perl->Iinitav, param); 15979 PL_savebegin = proto_perl->Isavebegin; 15980 15981 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param); 15982 15983 /* subprocess state */ 15984 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param); 15985 15986 if (proto_perl->Iop_mask) 15987 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo); 15988 else 15989 PL_op_mask = NULL; 15990 /* PL_asserting = proto_perl->Iasserting; */ 15991 15992 /* current interpreter roots */ 15993 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param); 15994 OP_REFCNT_LOCK; 15995 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root); 15996 OP_REFCNT_UNLOCK; 15997 15998 /* runtime control stuff */ 15999 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl); 16000 16001 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param); 16002 16003 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param); 16004 16005 /* interpreter atexit processing */ 16006 PL_exitlistlen = proto_perl->Iexitlistlen; 16007 if (PL_exitlistlen) { 16008 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry); 16009 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry); 16010 } 16011 else 16012 PL_exitlist = (PerlExitListEntry*)NULL; 16013 16014 PL_my_cxt_size = proto_perl->Imy_cxt_size; 16015 if (PL_my_cxt_size) { 16016 Newx(PL_my_cxt_list, PL_my_cxt_size, void *); 16017 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *); 16018 } 16019 else { 16020 PL_my_cxt_list = (void**)NULL; 16021 } 16022 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param); 16023 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param); 16024 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param); 16025 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param); 16026 16027 PL_compcv = cv_dup(proto_perl->Icompcv, param); 16028 16029 PAD_CLONE_VARS(proto_perl, param); 16030 16031#ifdef HAVE_INTERP_INTERN 16032 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern); 16033#endif 16034 16035 PL_DBcv = cv_dup(proto_perl->IDBcv, param); 16036 16037#ifdef PERL_USES_PL_PIDSTATUS 16038 PL_pidstatus = newHV(); /* XXX flag for cloning? */ 16039#endif 16040 PL_osname = SAVEPV(proto_perl->Iosname); 16041 PL_parser = parser_dup(proto_perl->Iparser, param); 16042 16043 /* XXX this only works if the saved cop has already been cloned */ 16044 if (proto_perl->Iparser) { 16045 PL_parser->saved_curcop = (COP*)any_dup( 16046 proto_perl->Iparser->saved_curcop, 16047 proto_perl); 16048 } 16049 16050 PL_subname = sv_dup_inc(proto_perl->Isubname, param); 16051 16052#ifdef USE_PL_CURLOCALES 16053 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) { 16054 PL_curlocales[i] = SAVEPV("C"); 16055 } 16056#endif 16057#ifdef USE_PL_CUR_LC_ALL 16058 PL_cur_LC_ALL = SAVEPV("C"); 16059#endif 16060#ifdef USE_LOCALE_CTYPE 16061 Copy(PL_fold, PL_fold_locale, 256, U8); 16062 16063 /* Should we warn if uses locale? */ 16064 PL_ctype_name = SAVEPV("C"); 16065 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param); 16066 PL_in_utf8_CTYPE_locale = false; 16067 PL_in_utf8_turkic_locale = false; 16068#endif 16069 16070 /* Did the locale setup indicate UTF-8? */ 16071 PL_utf8locale = false; 16072 16073#ifdef USE_LOCALE_COLLATE 16074 PL_in_utf8_COLLATE_locale = false; 16075 PL_collation_name = SAVEPV("C"); 16076 PL_collation_ix = proto_perl->Icollation_ix; 16077 PL_collation_standard = true; 16078 PL_collxfrm_base = 0; 16079 PL_collxfrm_mult = 0; 16080 PL_strxfrm_max_cp = 0; 16081 PL_strxfrm_is_behaved = proto_perl->Istrxfrm_is_behaved; 16082 PL_strxfrm_NUL_replacement = '\0'; 16083#endif /* USE_LOCALE_COLLATE */ 16084 16085#ifdef USE_LOCALE_THREADS 16086 assert(PL_locale_mutex_depth <= 0); 16087 PL_locale_mutex_depth = 0; 16088#endif 16089 16090#ifdef USE_LOCALE_NUMERIC 16091 PL_numeric_name = SAVEPV("C"); 16092 PL_numeric_radix_sv = newSVpvs("."); 16093 PL_underlying_radix_sv = newSVpvs("."); 16094 PL_numeric_standard = true; 16095 PL_numeric_underlying = true; 16096 PL_numeric_underlying_is_standard = true; 16097 16098# if defined(USE_POSIX_2008_LOCALE) 16099 PL_underlying_numeric_obj = NULL; 16100# endif 16101#endif /* !USE_LOCALE_NUMERIC */ 16102#if defined(USE_POSIX_2008_LOCALE) 16103 PL_scratch_locale_obj = NULL; 16104 PL_cur_locale_obj = PL_C_locale_obj; 16105#endif 16106 16107#ifdef HAS_MBRLEN 16108 PL_mbrlen_ps = proto_perl->Imbrlen_ps; 16109#endif 16110#ifdef HAS_MBRTOWC 16111 PL_mbrtowc_ps = proto_perl->Imbrtowc_ps; 16112#endif 16113#ifdef HAS_WCRTOMB 16114 PL_wcrtomb_ps = proto_perl->Iwcrtomb_ps; 16115#endif 16116 16117 PL_langinfo_buf = NULL; 16118 PL_langinfo_bufsize = 0; 16119 16120 PL_setlocale_buf = NULL; 16121 PL_setlocale_bufsize = 0; 16122 16123 PL_stdize_locale_buf = NULL; 16124 PL_stdize_locale_bufsize = 0; 16125 16126 /* Unicode inversion lists */ 16127 16128 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param); 16129 PL_Assigned_invlist = sv_dup_inc(proto_perl->IAssigned_invlist, param); 16130 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param); 16131 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param); 16132 PL_InMultiCharFold = sv_dup_inc(proto_perl->IInMultiCharFold, param); 16133 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param); 16134 PL_LB_invlist = sv_dup_inc(proto_perl->ILB_invlist, param); 16135 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param); 16136 PL_SCX_invlist = sv_dup_inc(proto_perl->ISCX_invlist, param); 16137 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param); 16138 PL_in_some_fold = sv_dup_inc(proto_perl->Iin_some_fold, param); 16139 PL_utf8_foldclosures = sv_dup_inc(proto_perl->Iutf8_foldclosures, param); 16140 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param); 16141 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param); 16142 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param); 16143 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param); 16144 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param); 16145 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param); 16146 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param); 16147 for (i = 0; i < POSIX_CC_COUNT; i++) { 16148 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param); 16149 if (i != CC_CASED_ && i != CC_VERTSPACE_) { 16150 PL_Posix_ptrs[i] = sv_dup_inc(proto_perl->IPosix_ptrs[i], param); 16151 } 16152 } 16153 PL_Posix_ptrs[CC_CASED_] = PL_Posix_ptrs[CC_ALPHA_]; 16154 PL_Posix_ptrs[CC_VERTSPACE_] = NULL; 16155 16156 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param); 16157 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param); 16158 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param); 16159 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param); 16160 PL_utf8_tosimplefold = sv_dup_inc(proto_perl->Iutf8_tosimplefold, param); 16161 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param); 16162 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param); 16163 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param); 16164 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param); 16165 PL_CCC_non0_non230 = sv_dup_inc(proto_perl->ICCC_non0_non230, param); 16166 PL_Private_Use = sv_dup_inc(proto_perl->IPrivate_Use, param); 16167 16168#if 0 16169 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param); 16170#endif 16171 16172 if (proto_perl->Ipsig_pend) { 16173 Newxz(PL_psig_pend, SIG_SIZE, int); 16174 } 16175 else { 16176 PL_psig_pend = (int*)NULL; 16177 } 16178 16179 if (proto_perl->Ipsig_name) { 16180 Newx(PL_psig_name, 2 * SIG_SIZE, SV*); 16181 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE, 16182 param); 16183 PL_psig_ptr = PL_psig_name + SIG_SIZE; 16184 } 16185 else { 16186 PL_psig_ptr = (SV**)NULL; 16187 PL_psig_name = (SV**)NULL; 16188 } 16189 16190 if (flags & CLONEf_COPY_STACKS) { 16191 Newx(PL_tmps_stack, PL_tmps_max, SV*); 16192 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack, 16193 PL_tmps_ix+1, param); 16194 16195 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */ 16196 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack; 16197 Newx(PL_markstack, i, I32); 16198 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max 16199 - proto_perl->Imarkstack); 16200 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr 16201 - proto_perl->Imarkstack); 16202 Copy(proto_perl->Imarkstack, PL_markstack, 16203 PL_markstack_ptr - PL_markstack + 1, I32); 16204 16205 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix] 16206 * NOTE: unlike the others! */ 16207 Newx(PL_scopestack, PL_scopestack_max, I32); 16208 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32); 16209 16210#ifdef DEBUGGING 16211 Newx(PL_scopestack_name, PL_scopestack_max, const char *); 16212 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *); 16213#endif 16214 /* reset stack AV to correct length before its duped via 16215 * PL_curstackinfo */ 16216 AvFILLp(proto_perl->Icurstack) = 16217 proto_perl->Istack_sp - proto_perl->Istack_base; 16218 16219 /* NOTE: si_dup() looks at PL_markstack */ 16220 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param); 16221 16222 /* PL_curstack = PL_curstackinfo->si_stack; */ 16223 PL_curstack = av_dup(proto_perl->Icurstack, param); 16224 PL_mainstack = av_dup(proto_perl->Imainstack, param); 16225 16226 /* next PUSHs() etc. set *(PL_stack_sp+1) */ 16227 PL_stack_base = AvARRAY(PL_curstack); 16228 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp 16229 - proto_perl->Istack_base); 16230 PL_stack_max = PL_stack_base + AvMAX(PL_curstack); 16231 16232 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/ 16233 PL_savestack = ss_dup(proto_perl, param); 16234 } 16235 else { 16236 init_stacks(); 16237 ENTER; /* perl_destruct() wants to LEAVE; */ 16238 } 16239 16240 PL_statgv = gv_dup(proto_perl->Istatgv, param); 16241 PL_statname = sv_dup_inc(proto_perl->Istatname, param); 16242 16243 PL_rs = sv_dup_inc(proto_perl->Irs, param); 16244 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param); 16245 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param); 16246 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param); 16247 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param); 16248 PL_formtarget = sv_dup(proto_perl->Iformtarget, param); 16249 16250 PL_errors = sv_dup_inc(proto_perl->Ierrors, param); 16251 16252 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl); 16253 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param); 16254 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param); 16255 16256 PL_stashcache = newHV(); 16257 16258 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table, 16259 proto_perl->Iwatchaddr); 16260 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL; 16261 if (PL_debug && PL_watchaddr) { 16262 PerlIO_printf(Perl_debug_log, 16263 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n", 16264 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr), 16265 PTR2UV(PL_watchok)); 16266 } 16267 16268 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param); 16269 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param); 16270 16271 /* Call the ->CLONE method, if it exists, for each of the stashes 16272 identified by sv_dup() above. 16273 */ 16274 while(av_count(param->stashes) != 0) { 16275 HV* const stash = MUTABLE_HV(av_shift(param->stashes)); 16276 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0); 16277 if (cloner && GvCV(cloner)) { 16278 dSP; 16279 ENTER; 16280 SAVETMPS; 16281 PUSHMARK(SP); 16282 mXPUSHs(newSVhek(HvNAME_HEK(stash))); 16283 PUTBACK; 16284 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD); 16285 FREETMPS; 16286 LEAVE; 16287 } 16288 } 16289 16290 if (!(flags & CLONEf_KEEP_PTR_TABLE)) { 16291 ptr_table_free(PL_ptr_table); 16292 PL_ptr_table = NULL; 16293 } 16294 16295 if (!(flags & CLONEf_COPY_STACKS)) { 16296 unreferenced_to_tmp_stack(param->unreferenced); 16297 } 16298 16299 SvREFCNT_dec(param->stashes); 16300 16301 /* orphaned? eg threads->new inside BEGIN or use */ 16302 if (PL_compcv && ! SvREFCNT(PL_compcv)) { 16303 SvREFCNT_inc_simple_void(PL_compcv); 16304 SAVEFREESV(PL_compcv); 16305 } 16306 16307 return my_perl; 16308} 16309 16310static void 16311S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced) 16312{ 16313 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK; 16314 16315 if (AvFILLp(unreferenced) > -1) { 16316 SV **svp = AvARRAY(unreferenced); 16317 SV **const last = svp + AvFILLp(unreferenced); 16318 SSize_t count = 0; 16319 16320 do { 16321 if (SvREFCNT(*svp) == 1) 16322 ++count; 16323 } while (++svp <= last); 16324 16325 EXTEND_MORTAL(count); 16326 svp = AvARRAY(unreferenced); 16327 16328 do { 16329 if (SvREFCNT(*svp) == 1) { 16330 /* Our reference is the only one to this SV. This means that 16331 in this thread, the scalar effectively has a 0 reference. 16332 That doesn't work (cleanup never happens), so donate our 16333 reference to it onto the save stack. */ 16334 PL_tmps_stack[++PL_tmps_ix] = *svp; 16335 } else { 16336 /* As an optimisation, because we are already walking the 16337 entire array, instead of above doing either 16338 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead 16339 release our reference to the scalar, so that at the end of 16340 the array owns zero references to the scalars it happens to 16341 point to. We are effectively converting the array from 16342 AvREAL() on to AvREAL() off. This saves the av_clear() 16343 (triggered by the SvREFCNT_dec(unreferenced) below) from 16344 walking the array a second time. */ 16345 SvREFCNT_dec(*svp); 16346 } 16347 16348 } while (++svp <= last); 16349 AvREAL_off(unreferenced); 16350 } 16351 SvREFCNT_dec_NN(unreferenced); 16352} 16353 16354void 16355Perl_clone_params_del(CLONE_PARAMS *param) 16356{ 16357 PerlInterpreter *const was = PERL_GET_THX; 16358 PerlInterpreter *const to = param->new_perl; 16359 dTHXa(to); 16360 16361 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL; 16362 16363 if (was != to) { 16364 PERL_SET_THX(to); 16365 } 16366 16367 SvREFCNT_dec(param->stashes); 16368 if (param->unreferenced) 16369 unreferenced_to_tmp_stack(param->unreferenced); 16370 16371 Safefree(param); 16372 16373 if (was != to) { 16374 PERL_SET_THX(was); 16375 } 16376} 16377 16378CLONE_PARAMS * 16379Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to) 16380{ 16381 /* Need to play this game, as newAV() can call safesysmalloc(), and that 16382 does a dTHX; to get the context from thread local storage. 16383 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to 16384 a version that passes in my_perl. */ 16385 PerlInterpreter *const was = PERL_GET_THX; 16386 CLONE_PARAMS *param; 16387 16388 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW; 16389 16390 if (was != to) { 16391 PERL_SET_THX(to); 16392 } 16393 16394 /* Given that we've set the context, we can do this unshared. */ 16395 Newx(param, 1, CLONE_PARAMS); 16396 16397 param->flags = 0; 16398 param->proto_perl = from; 16399 param->new_perl = to; 16400 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV); 16401 AvREAL_off(param->stashes); 16402 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV); 16403 16404 if (was != to) { 16405 PERL_SET_THX(was); 16406 } 16407 return param; 16408} 16409 16410#endif /* USE_ITHREADS */ 16411 16412void 16413Perl_init_constants(pTHX) 16414{ 16415 16416 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL; 16417 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL; 16418 SvANY(&PL_sv_undef) = NULL; 16419 16420 SvANY(&PL_sv_no) = new_XPVNV(); 16421 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL; 16422 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT 16423 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK 16424 |SVp_POK|SVf_POK|SVf_IsCOW|SVppv_STATIC; 16425 16426 SvANY(&PL_sv_yes) = new_XPVNV(); 16427 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL; 16428 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT 16429 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK 16430 |SVp_POK|SVf_POK|SVf_IsCOW|SVppv_STATIC; 16431 16432 SvANY(&PL_sv_zero) = new_XPVNV(); 16433 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL; 16434 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT 16435 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK 16436 |SVp_POK|SVf_POK 16437 |SVs_PADTMP; 16438 16439 SvPV_set(&PL_sv_no, (char*)PL_No); 16440 SvCUR_set(&PL_sv_no, 0); 16441 SvLEN_set(&PL_sv_no, 0); 16442 SvIV_set(&PL_sv_no, 0); 16443 SvNV_set(&PL_sv_no, 0); 16444 16445 SvPV_set(&PL_sv_yes, (char*)PL_Yes); 16446 SvCUR_set(&PL_sv_yes, 1); 16447 SvLEN_set(&PL_sv_yes, 0); 16448 SvIV_set(&PL_sv_yes, 1); 16449 SvNV_set(&PL_sv_yes, 1); 16450 16451 SvPV_set(&PL_sv_zero, (char*)PL_Zero); 16452 SvCUR_set(&PL_sv_zero, 1); 16453 SvLEN_set(&PL_sv_zero, 0); 16454 SvIV_set(&PL_sv_zero, 0); 16455 SvNV_set(&PL_sv_zero, 0); 16456 16457 PadnamePV(&PL_padname_const) = (char *)PL_No; 16458 16459 assert(SvIMMORTAL_INTERP(&PL_sv_yes)); 16460 assert(SvIMMORTAL_INTERP(&PL_sv_undef)); 16461 assert(SvIMMORTAL_INTERP(&PL_sv_no)); 16462 assert(SvIMMORTAL_INTERP(&PL_sv_zero)); 16463 16464 assert(SvIMMORTAL(&PL_sv_yes)); 16465 assert(SvIMMORTAL(&PL_sv_undef)); 16466 assert(SvIMMORTAL(&PL_sv_no)); 16467 assert(SvIMMORTAL(&PL_sv_zero)); 16468 16469 assert( SvIMMORTAL_TRUE(&PL_sv_yes)); 16470 assert(!SvIMMORTAL_TRUE(&PL_sv_undef)); 16471 assert(!SvIMMORTAL_TRUE(&PL_sv_no)); 16472 assert(!SvIMMORTAL_TRUE(&PL_sv_zero)); 16473 16474 assert( SvTRUE_nomg_NN(&PL_sv_yes)); 16475 assert(!SvTRUE_nomg_NN(&PL_sv_undef)); 16476 assert(!SvTRUE_nomg_NN(&PL_sv_no)); 16477 assert(!SvTRUE_nomg_NN(&PL_sv_zero)); 16478} 16479 16480/* 16481=for apidoc_section $unicode 16482 16483=for apidoc sv_recode_to_utf8 16484 16485C<encoding> is assumed to be an C<Encode> object, on entry the PV 16486of C<sv> is assumed to be octets in that encoding, and C<sv> 16487will be converted into Unicode (and UTF-8). 16488 16489If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding> 16490is not a reference, nothing is done to C<sv>. If C<encoding> is not 16491an C<Encode::XS> Encoding object, bad things will happen. 16492(See L<encoding> and L<Encode>.) 16493 16494The PV of C<sv> is returned. 16495 16496=cut */ 16497 16498char * 16499Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding) 16500{ 16501 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8; 16502 16503 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) { 16504 SV *uni; 16505 STRLEN len; 16506 const char *s; 16507 dSP; 16508 SV *nsv = sv; 16509 ENTER; 16510 PUSHSTACK; 16511 SAVETMPS; 16512 if (SvPADTMP(nsv)) { 16513 nsv = sv_newmortal(); 16514 SvSetSV_nosteal(nsv, sv); 16515 } 16516 save_re_context(); 16517 PUSHMARK(sp); 16518 EXTEND(SP, 3); 16519 PUSHs(encoding); 16520 PUSHs(nsv); 16521/* 16522 NI-S 2002/07/09 16523 Passing sv_yes is wrong - it needs to be or'ed set of constants 16524 for Encode::XS, while UTf-8 decode (currently) assumes a true value means 16525 remove converted chars from source. 16526 16527 Both will default the value - let them. 16528 16529 XPUSHs(&PL_sv_yes); 16530*/ 16531 PUTBACK; 16532 call_method("decode", G_SCALAR); 16533 SPAGAIN; 16534 uni = POPs; 16535 PUTBACK; 16536 s = SvPV_const(uni, len); 16537 if (s != SvPVX_const(sv)) { 16538 SvGROW(sv, len + 1); 16539 Move(s, SvPVX(sv), len + 1, char); 16540 SvCUR_set(sv, len); 16541 } 16542 FREETMPS; 16543 POPSTACK; 16544 LEAVE; 16545 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { 16546 /* clear pos and any utf8 cache */ 16547 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); 16548 if (mg) 16549 mg->mg_len = -1; 16550 if ((mg = mg_find(sv, PERL_MAGIC_utf8))) 16551 magic_setutf8(sv,mg); /* clear UTF8 cache */ 16552 } 16553 SvUTF8_on(sv); 16554 return SvPVX(sv); 16555 } 16556 return SvPOKp(sv) ? SvPVX(sv) : NULL; 16557} 16558 16559/* 16560=for apidoc sv_cat_decode 16561 16562C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is 16563assumed to be octets in that encoding and decoding the input starts 16564from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be 16565concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate 16566when the string C<tstr> appears in decoding output or the input ends on 16567the PV of C<ssv>. The value which C<offset> points will be modified 16568to the last input position on C<ssv>. 16569 16570Returns TRUE if the terminator was found, else returns FALSE. 16571 16572=cut */ 16573 16574bool 16575Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding, 16576 SV *ssv, int *offset, char *tstr, int tlen) 16577{ 16578 bool ret = FALSE; 16579 16580 PERL_ARGS_ASSERT_SV_CAT_DECODE; 16581 16582 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) { 16583 SV *offsv; 16584 dSP; 16585 ENTER; 16586 SAVETMPS; 16587 save_re_context(); 16588 PUSHMARK(sp); 16589 EXTEND(SP, 6); 16590 PUSHs(encoding); 16591 PUSHs(dsv); 16592 PUSHs(ssv); 16593 offsv = newSViv(*offset); 16594 mPUSHs(offsv); 16595 mPUSHp(tstr, tlen); 16596 PUTBACK; 16597 call_method("cat_decode", G_SCALAR); 16598 SPAGAIN; 16599 ret = SvTRUE(TOPs); 16600 *offset = SvIV(offsv); 16601 PUTBACK; 16602 FREETMPS; 16603 LEAVE; 16604 } 16605 else 16606 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode"); 16607 return ret; 16608 16609} 16610 16611/* --------------------------------------------------------------------- 16612 * 16613 * support functions for report_uninit() 16614 */ 16615 16616/* the maxiumum size of array or hash where we will scan looking 16617 * for the undefined element that triggered the warning */ 16618 16619#define FUV_MAX_SEARCH_SIZE 1000 16620 16621/* Look for an entry in the hash whose value has the same SV as val; 16622 * If so, return a mortal copy of the key. */ 16623 16624STATIC SV* 16625S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val) 16626{ 16627 HE **array; 16628 I32 i; 16629 16630 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT; 16631 16632 if (!hv || SvMAGICAL(hv) || !HvTOTALKEYS(hv) || 16633 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE)) 16634 return NULL; 16635 16636 if (val == &PL_sv_undef || val == &PL_sv_placeholder) 16637 return NULL; 16638 16639 array = HvARRAY(hv); 16640 16641 for (i=HvMAX(hv); i>=0; i--) { 16642 HE *entry; 16643 for (entry = array[i]; entry; entry = HeNEXT(entry)) { 16644 if (HeVAL(entry) == val) 16645 return newSVhek_mortal(HeKEY_hek(entry)); 16646 } 16647 } 16648 return NULL; 16649} 16650 16651/* Look for an entry in the array whose value has the same SV as val; 16652 * If so, return the index, otherwise return -1. */ 16653 16654STATIC SSize_t 16655S_find_array_subscript(pTHX_ const AV *const av, const SV *const val) 16656{ 16657 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT; 16658 16659 if (!av || SvMAGICAL(av) || !AvARRAY(av) || 16660 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE)) 16661 return -1; 16662 16663 if (val != &PL_sv_undef) { 16664 SV ** const svp = AvARRAY(av); 16665 SSize_t i; 16666 16667 for (i=AvFILLp(av); i>=0; i--) 16668 if (svp[i] == val) 16669 return i; 16670 } 16671 return -1; 16672} 16673 16674/* varname(): return the name of a variable, optionally with a subscript. 16675 * If gv is non-zero, use the name of that global, along with gvtype (one 16676 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset 16677 * targ. Depending on the value of the subscript_type flag, return: 16678 */ 16679 16680#define FUV_SUBSCRIPT_NONE 1 /* "@foo" */ 16681#define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */ 16682#define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */ 16683#define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */ 16684 16685SV* 16686Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ, 16687 const SV *const keyname, SSize_t aindex, int subscript_type) 16688{ 16689 16690 SV * const name = sv_newmortal(); 16691 if (gv && isGV(gv)) { 16692 char buffer[2]; 16693 buffer[0] = gvtype; 16694 buffer[1] = 0; 16695 16696 /* as gv_fullname4(), but add literal '^' for $^FOO names */ 16697 16698 gv_fullname4(name, gv, buffer, 0); 16699 16700 if ((unsigned int)SvPVX(name)[1] <= 26) { 16701 buffer[0] = '^'; 16702 buffer[1] = SvPVX(name)[1] + 'A' - 1; 16703 16704 /* Swap the 1 unprintable control character for the 2 byte pretty 16705 version - ie substr($name, 1, 1) = $buffer; */ 16706 sv_insert(name, 1, 1, buffer, 2); 16707 } 16708 } 16709 else { 16710 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL); 16711 PADNAME *sv; 16712 16713 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM); 16714 16715 if (!cv || !CvPADLIST(cv)) 16716 return NULL; 16717 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ); 16718 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv)); 16719 SvUTF8_on(name); 16720 } 16721 16722 if (subscript_type == FUV_SUBSCRIPT_HASH) { 16723 SV * const sv = newSV_type(SVt_NULL); 16724 STRLEN len; 16725 const char * const pv = SvPV_nomg_const((SV*)keyname, len); 16726 16727 *SvPVX(name) = '$'; 16728 Perl_sv_catpvf(aTHX_ name, "{%s}", 16729 pv_pretty(sv, pv, len, 32, NULL, NULL, 16730 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT )); 16731 SvREFCNT_dec_NN(sv); 16732 } 16733 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) { 16734 *SvPVX(name) = '$'; 16735 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex); 16736 } 16737 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) { 16738 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */ 16739 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0); 16740 } 16741 16742 return name; 16743} 16744 16745 16746/* 16747=apidoc_section $warning 16748=for apidoc find_uninit_var 16749 16750Find the name of the undefined variable (if any) that caused the operator 16751to issue a "Use of uninitialized value" warning. 16752If match is true, only return a name if its value matches C<uninit_sv>. 16753So roughly speaking, if a unary operator (such as C<OP_COS>) generates a 16754warning, then following the direct child of the op may yield an 16755C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the 16756other hand, with C<OP_ADD> there are two branches to follow, so we only print 16757the variable name if we get an exact match. 16758C<desc_p> points to a string pointer holding the description of the op. 16759This may be updated if needed. 16760 16761The name is returned as a mortal SV. 16762 16763Assumes that C<PL_op> is the OP that originally triggered the error, and that 16764C<PL_comppad>/C<PL_curpad> points to the currently executing pad. 16765 16766=cut 16767*/ 16768 16769STATIC SV * 16770S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv, 16771 bool match, const char **desc_p) 16772{ 16773 SV *sv; 16774 const GV *gv; 16775 const OP *o, *o2, *kid; 16776 16777 PERL_ARGS_ASSERT_FIND_UNINIT_VAR; 16778 16779 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef || 16780 uninit_sv == &PL_sv_placeholder))) 16781 return NULL; 16782 16783 switch (obase->op_type) { 16784 16785 case OP_UNDEF: 16786 /* the optimizer rewrites '$x = undef' to 'undef $x' for lexical 16787 * variables, which can occur as the source of warnings: 16788 * ($x = undef) =~ s/a/b/; 16789 * The OPpUNDEF_KEEP_PV flag indicates that this used to be an 16790 * assignment op. 16791 * Otherwise undef should not care if its args are undef - any warnings 16792 * will be from tied/magic vars */ 16793 if ( 16794 (obase->op_private & (OPpTARGET_MY | OPpUNDEF_KEEP_PV)) == (OPpTARGET_MY | OPpUNDEF_KEEP_PV) 16795 && (!match || PAD_SVl(obase->op_targ) == uninit_sv) 16796 ) { 16797 return varname(NULL, '$', obase->op_targ, NULL, 0, FUV_SUBSCRIPT_NONE); 16798 } 16799 break; 16800 16801 case OP_RV2AV: 16802 case OP_RV2HV: 16803 case OP_PADAV: 16804 case OP_PADHV: 16805 { 16806 const bool pad = ( obase->op_type == OP_PADAV 16807 || obase->op_type == OP_PADHV 16808 || obase->op_type == OP_PADRANGE 16809 ); 16810 16811 const bool hash = ( obase->op_type == OP_PADHV 16812 || obase->op_type == OP_RV2HV 16813 || (obase->op_type == OP_PADRANGE 16814 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV) 16815 ); 16816 SSize_t index = 0; 16817 SV *keysv = NULL; 16818 int subscript_type = FUV_SUBSCRIPT_WITHIN; 16819 16820 if (pad) { /* @lex, %lex */ 16821 sv = PAD_SVl(obase->op_targ); 16822 gv = NULL; 16823 } 16824 else { 16825 if (cUNOPx(obase)->op_first->op_type == OP_GV) { 16826 /* @global, %global */ 16827 gv = cGVOPx_gv(cUNOPx(obase)->op_first); 16828 if (!gv) 16829 break; 16830 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv)); 16831 } 16832 else if (obase == PL_op) /* @{expr}, %{expr} */ 16833 return find_uninit_var(cUNOPx(obase)->op_first, 16834 uninit_sv, match, desc_p); 16835 else /* @{expr}, %{expr} as a sub-expression */ 16836 return NULL; 16837 } 16838 16839 /* attempt to find a match within the aggregate */ 16840 if (hash) { 16841 keysv = find_hash_subscript((const HV*)sv, uninit_sv); 16842 if (keysv) 16843 subscript_type = FUV_SUBSCRIPT_HASH; 16844 } 16845 else { 16846 index = find_array_subscript((const AV *)sv, uninit_sv); 16847 if (index >= 0) 16848 subscript_type = FUV_SUBSCRIPT_ARRAY; 16849 } 16850 16851 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN) 16852 break; 16853 16854 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ, 16855 keysv, index, subscript_type); 16856 } 16857 16858 case OP_RV2SV: 16859 if (cUNOPx(obase)->op_first->op_type == OP_GV) { 16860 /* $global */ 16861 gv = cGVOPx_gv(cUNOPx(obase)->op_first); 16862 if (!gv || !GvSTASH(gv)) 16863 break; 16864 if (match && (GvSV(gv) != uninit_sv)) 16865 break; 16866 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE); 16867 } 16868 /* ${expr} */ 16869 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p); 16870 16871 case OP_PADSV: 16872 if (match && PAD_SVl(obase->op_targ) != uninit_sv) 16873 break; 16874 return varname(NULL, '$', obase->op_targ, 16875 NULL, 0, FUV_SUBSCRIPT_NONE); 16876 16877 case OP_PADSV_STORE: 16878 if (match && PAD_SVl(obase->op_targ) != uninit_sv) 16879 goto do_op; 16880 return varname(NULL, '$', obase->op_targ, 16881 NULL, 0, FUV_SUBSCRIPT_NONE); 16882 16883 case OP_GVSV: 16884 gv = cGVOPx_gv(obase); 16885 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv)) 16886 break; 16887 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE); 16888 16889 case OP_AELEMFAST_LEX: 16890 if (match) { 16891 SV **svp; 16892 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ)); 16893 if (!av || SvRMAGICAL(av)) 16894 break; 16895 svp = av_fetch(av, (I8)obase->op_private, FALSE); 16896 if (!svp || *svp != uninit_sv) 16897 break; 16898 } 16899 return varname(NULL, '$', obase->op_targ, 16900 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY); 16901 16902 case OP_AELEMFASTLEX_STORE: 16903 if (match) { 16904 SV **svp; 16905 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ)); 16906 if (!av || SvRMAGICAL(av)) 16907 goto do_op; 16908 svp = av_fetch(av, (I8)obase->op_private, FALSE); 16909 if (!svp || *svp != uninit_sv) 16910 goto do_op; 16911 } 16912 return varname(NULL, '$', obase->op_targ, 16913 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY); 16914 16915 case OP_AELEMFAST: 16916 { 16917 gv = cGVOPx_gv(obase); 16918 if (!gv) 16919 break; 16920 if (match) { 16921 SV **svp; 16922 AV *const av = GvAV(gv); 16923 if (!av || SvRMAGICAL(av)) 16924 break; 16925 svp = av_fetch(av, (I8)obase->op_private, FALSE); 16926 if (!svp || *svp != uninit_sv) 16927 break; 16928 } 16929 return varname(gv, '$', 0, 16930 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY); 16931 } 16932 NOT_REACHED; /* NOTREACHED */ 16933 16934 case OP_EXISTS: 16935 o = cUNOPx(obase)->op_first; 16936 if (!o || o->op_type != OP_NULL || 16937 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM)) 16938 break; 16939 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p); 16940 16941 case OP_AELEM: 16942 case OP_HELEM: 16943 { 16944 bool negate = FALSE; 16945 16946 if (PL_op == obase) 16947 /* $a[uninit_expr] or $h{uninit_expr} */ 16948 return find_uninit_var(cBINOPx(obase)->op_last, 16949 uninit_sv, match, desc_p); 16950 16951 gv = NULL; 16952 o = cBINOPx(obase)->op_first; 16953 kid = cBINOPx(obase)->op_last; 16954 16955 /* get the av or hv, and optionally the gv */ 16956 sv = NULL; 16957 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) { 16958 sv = PAD_SV(o->op_targ); 16959 } 16960 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV) 16961 && cUNOPo->op_first->op_type == OP_GV) 16962 { 16963 gv = cGVOPx_gv(cUNOPo->op_first); 16964 if (!gv) 16965 break; 16966 sv = o->op_type 16967 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv)); 16968 } 16969 if (!sv) 16970 break; 16971 16972 if (kid && kid->op_type == OP_NEGATE) { 16973 negate = TRUE; 16974 kid = cUNOPx(kid)->op_first; 16975 } 16976 16977 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) { 16978 /* index is constant */ 16979 SV* kidsv; 16980 if (negate) { 16981 kidsv = newSVpvs_flags("-", SVs_TEMP); 16982 sv_catsv(kidsv, cSVOPx_sv(kid)); 16983 } 16984 else 16985 kidsv = cSVOPx_sv(kid); 16986 if (match) { 16987 if (SvMAGICAL(sv)) 16988 break; 16989 if (obase->op_type == OP_HELEM) { 16990 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0); 16991 if (!he || HeVAL(he) != uninit_sv) 16992 break; 16993 } 16994 else { 16995 SV * const opsv = cSVOPx_sv(kid); 16996 const IV opsviv = SvIV(opsv); 16997 SV * const * const svp = av_fetch(MUTABLE_AV(sv), 16998 negate ? - opsviv : opsviv, 16999 FALSE); 17000 if (!svp || *svp != uninit_sv) 17001 break; 17002 } 17003 } 17004 if (obase->op_type == OP_HELEM) 17005 return varname(gv, '%', o->op_targ, 17006 kidsv, 0, FUV_SUBSCRIPT_HASH); 17007 else 17008 return varname(gv, '@', o->op_targ, NULL, 17009 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)), 17010 FUV_SUBSCRIPT_ARRAY); 17011 } 17012 else { 17013 /* index is an expression; 17014 * attempt to find a match within the aggregate */ 17015 if (obase->op_type == OP_HELEM) { 17016 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv); 17017 if (keysv) 17018 return varname(gv, '%', o->op_targ, 17019 keysv, 0, FUV_SUBSCRIPT_HASH); 17020 } 17021 else { 17022 const SSize_t index 17023 = find_array_subscript((const AV *)sv, uninit_sv); 17024 if (index >= 0) 17025 return varname(gv, '@', o->op_targ, 17026 NULL, index, FUV_SUBSCRIPT_ARRAY); 17027 } 17028 if (match) 17029 break; 17030 return varname(gv, 17031 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV) 17032 ? '@' : '%'), 17033 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN); 17034 } 17035 NOT_REACHED; /* NOTREACHED */ 17036 } 17037 17038 case OP_MULTIDEREF: { 17039 /* If we were executing OP_MULTIDEREF when the undef warning 17040 * triggered, then it must be one of the index values within 17041 * that triggered it. If not, then the only possibility is that 17042 * the value retrieved by the last aggregate index might be the 17043 * culprit. For the former, we set PL_multideref_pc each time before 17044 * using an index, so work though the item list until we reach 17045 * that point. For the latter, just work through the entire item 17046 * list; the last aggregate retrieved will be the candidate. 17047 * There is a third rare possibility: something triggered 17048 * magic while fetching an array/hash element. Just display 17049 * nothing in this case. 17050 */ 17051 17052 /* the named aggregate, if any */ 17053 PADOFFSET agg_targ = 0; 17054 GV *agg_gv = NULL; 17055 /* the last-seen index */ 17056 UV index_type; 17057 PADOFFSET index_targ; 17058 GV *index_gv; 17059 IV index_const_iv = 0; /* init for spurious compiler warn */ 17060 SV *index_const_sv; 17061 int depth = 0; /* how many array/hash lookups we've done */ 17062 17063 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux; 17064 UNOP_AUX_item *last = NULL; 17065 UV actions = items->uv; 17066 bool is_hv; 17067 17068 if (PL_op == obase) { 17069 last = PL_multideref_pc; 17070 assert(last >= items && last <= items + items[-1].uv); 17071 } 17072 17073 assert(actions); 17074 17075 while (1) { 17076 is_hv = FALSE; 17077 switch (actions & MDEREF_ACTION_MASK) { 17078 17079 case MDEREF_reload: 17080 actions = (++items)->uv; 17081 continue; 17082 17083 case MDEREF_HV_padhv_helem: /* $lex{...} */ 17084 is_hv = TRUE; 17085 /* FALLTHROUGH */ 17086 case MDEREF_AV_padav_aelem: /* $lex[...] */ 17087 agg_targ = (++items)->pad_offset; 17088 agg_gv = NULL; 17089 break; 17090 17091 case MDEREF_HV_gvhv_helem: /* $pkg{...} */ 17092 is_hv = TRUE; 17093 /* FALLTHROUGH */ 17094 case MDEREF_AV_gvav_aelem: /* $pkg[...] */ 17095 agg_targ = 0; 17096 agg_gv = (GV*)UNOP_AUX_item_sv(++items); 17097 assert(isGV_with_GP(agg_gv)); 17098 break; 17099 17100 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */ 17101 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */ 17102 ++items; 17103 /* FALLTHROUGH */ 17104 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */ 17105 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */ 17106 agg_targ = 0; 17107 agg_gv = NULL; 17108 is_hv = TRUE; 17109 break; 17110 17111 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */ 17112 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */ 17113 ++items; 17114 /* FALLTHROUGH */ 17115 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */ 17116 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */ 17117 agg_targ = 0; 17118 agg_gv = NULL; 17119 } /* switch */ 17120 17121 index_targ = 0; 17122 index_gv = NULL; 17123 index_const_sv = NULL; 17124 17125 index_type = (actions & MDEREF_INDEX_MASK); 17126 switch (index_type) { 17127 case MDEREF_INDEX_none: 17128 break; 17129 case MDEREF_INDEX_const: 17130 if (is_hv) 17131 index_const_sv = UNOP_AUX_item_sv(++items) 17132 else 17133 index_const_iv = (++items)->iv; 17134 break; 17135 case MDEREF_INDEX_padsv: 17136 index_targ = (++items)->pad_offset; 17137 break; 17138 case MDEREF_INDEX_gvsv: 17139 index_gv = (GV*)UNOP_AUX_item_sv(++items); 17140 assert(isGV_with_GP(index_gv)); 17141 break; 17142 } 17143 17144 if (index_type != MDEREF_INDEX_none) 17145 depth++; 17146 17147 if ( index_type == MDEREF_INDEX_none 17148 || (actions & MDEREF_FLAG_last) 17149 || (last && items >= last) 17150 ) 17151 break; 17152 17153 actions >>= MDEREF_SHIFT; 17154 } /* while */ 17155 17156 if (PL_op == obase) { 17157 /* most likely index was undef */ 17158 17159 *desc_p = ( (actions & MDEREF_FLAG_last) 17160 && (obase->op_private 17161 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))) 17162 ? 17163 (obase->op_private & OPpMULTIDEREF_EXISTS) 17164 ? "exists" 17165 : "delete" 17166 : is_hv ? "hash element" : "array element"; 17167 assert(index_type != MDEREF_INDEX_none); 17168 if (index_gv) { 17169 if (GvSV(index_gv) == uninit_sv) 17170 return varname(index_gv, '$', 0, NULL, 0, 17171 FUV_SUBSCRIPT_NONE); 17172 else 17173 return NULL; 17174 } 17175 if (index_targ) { 17176 if (PL_curpad[index_targ] == uninit_sv) 17177 return varname(NULL, '$', index_targ, 17178 NULL, 0, FUV_SUBSCRIPT_NONE); 17179 else 17180 return NULL; 17181 } 17182 /* If we got to this point it was undef on a const subscript, 17183 * so magic probably involved, e.g. $ISA[0]. Give up. */ 17184 return NULL; 17185 } 17186 17187 /* the SV returned by pp_multideref() was undef, if anything was */ 17188 17189 if (depth != 1) 17190 break; 17191 17192 if (agg_targ) 17193 sv = PAD_SV(agg_targ); 17194 else if (agg_gv) { 17195 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv)); 17196 if (!sv) 17197 break; 17198 } 17199 else 17200 break; 17201 17202 if (index_type == MDEREF_INDEX_const) { 17203 if (match) { 17204 if (SvMAGICAL(sv)) 17205 break; 17206 if (is_hv) { 17207 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0); 17208 if (!he || HeVAL(he) != uninit_sv) 17209 break; 17210 } 17211 else { 17212 SV * const * const svp = 17213 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE); 17214 if (!svp || *svp != uninit_sv) 17215 break; 17216 } 17217 } 17218 return is_hv 17219 ? varname(agg_gv, '%', agg_targ, 17220 index_const_sv, 0, FUV_SUBSCRIPT_HASH) 17221 : varname(agg_gv, '@', agg_targ, 17222 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY); 17223 } 17224 else { 17225 /* index is an var */ 17226 if (is_hv) { 17227 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv); 17228 if (keysv) 17229 return varname(agg_gv, '%', agg_targ, 17230 keysv, 0, FUV_SUBSCRIPT_HASH); 17231 } 17232 else { 17233 const SSize_t index 17234 = find_array_subscript((const AV *)sv, uninit_sv); 17235 if (index >= 0) 17236 return varname(agg_gv, '@', agg_targ, 17237 NULL, index, FUV_SUBSCRIPT_ARRAY); 17238 } 17239 /* look for an element not found */ 17240 if (!SvMAGICAL(sv)) { 17241 SV *index_sv = NULL; 17242 if (index_targ) { 17243 index_sv = PL_curpad[index_targ]; 17244 } 17245 else if (index_gv) { 17246 index_sv = GvSV(index_gv); 17247 } 17248 if (index_sv && !SvMAGICAL(index_sv) && !SvROK(index_sv)) { 17249 if (is_hv) { 17250 SV *report_index_sv = SvOK(index_sv) ? index_sv : &PL_sv_no; 17251 HE *he = hv_fetch_ent(MUTABLE_HV(sv), report_index_sv, 0, 0); 17252 if (!he) { 17253 return varname(agg_gv, '%', agg_targ, 17254 report_index_sv, 0, FUV_SUBSCRIPT_HASH); 17255 } 17256 } 17257 else { 17258 SSize_t index = SvOK(index_sv) ? SvIV(index_sv) : 0; 17259 SV * const * const svp = 17260 av_fetch(MUTABLE_AV(sv), index, FALSE); 17261 if (!svp) { 17262 return varname(agg_gv, '@', agg_targ, 17263 NULL, index, FUV_SUBSCRIPT_ARRAY); 17264 } 17265 } 17266 } 17267 } 17268 if (match) 17269 break; 17270 return varname(agg_gv, 17271 is_hv ? '%' : '@', 17272 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN); 17273 } 17274 NOT_REACHED; /* NOTREACHED */ 17275 } 17276 17277 case OP_AASSIGN: 17278 /* only examine RHS */ 17279 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, 17280 match, desc_p); 17281 17282 case OP_OPEN: 17283 o = cUNOPx(obase)->op_first; 17284 if ( o->op_type == OP_PUSHMARK 17285 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK) 17286 ) 17287 o = OpSIBLING(o); 17288 17289 if (!OpHAS_SIBLING(o)) { 17290 /* one-arg version of open is highly magical */ 17291 17292 if (o->op_type == OP_GV) { /* open FOO; */ 17293 gv = cGVOPx_gv(o); 17294 if (match && GvSV(gv) != uninit_sv) 17295 break; 17296 return varname(gv, '$', 0, 17297 NULL, 0, FUV_SUBSCRIPT_NONE); 17298 } 17299 /* other possibilities not handled are: 17300 * open $x; or open my $x; should return '${*$x}' 17301 * open expr; should return '$'.expr ideally 17302 */ 17303 break; 17304 } 17305 match = 1; 17306 goto do_op; 17307 17308 /* ops where $_ may be an implicit arg */ 17309 case OP_TRANS: 17310 case OP_TRANSR: 17311 case OP_SUBST: 17312 case OP_MATCH: 17313 if ( !(obase->op_flags & OPf_STACKED)) { 17314 if (uninit_sv == DEFSV) 17315 return newSVpvs_flags("$_", SVs_TEMP); 17316 else if (obase->op_targ 17317 && uninit_sv == PAD_SVl(obase->op_targ)) 17318 return varname(NULL, '$', obase->op_targ, NULL, 0, 17319 FUV_SUBSCRIPT_NONE); 17320 } 17321 goto do_op; 17322 17323 case OP_PRTF: 17324 case OP_PRINT: 17325 case OP_SAY: 17326 match = 1; /* print etc can return undef on defined args */ 17327 /* skip filehandle as it can't produce 'undef' warning */ 17328 o = cUNOPx(obase)->op_first; 17329 if ((obase->op_flags & OPf_STACKED) 17330 && 17331 ( o->op_type == OP_PUSHMARK 17332 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK))) 17333 o = OpSIBLING(OpSIBLING(o)); 17334 goto do_op2; 17335 17336 17337 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */ 17338 case OP_CUSTOM: /* XS or custom code could trigger random warnings */ 17339 17340 /* the following ops are capable of returning PL_sv_undef even for 17341 * defined arg(s) */ 17342 17343 case OP_BACKTICK: 17344 case OP_PIPE_OP: 17345 case OP_FILENO: 17346 case OP_BINMODE: 17347 case OP_TIED: 17348 case OP_GETC: 17349 case OP_SYSREAD: 17350 case OP_SEND: 17351 case OP_IOCTL: 17352 case OP_SOCKET: 17353 case OP_SOCKPAIR: 17354 case OP_BIND: 17355 case OP_CONNECT: 17356 case OP_LISTEN: 17357 case OP_ACCEPT: 17358 case OP_SHUTDOWN: 17359 case OP_SSOCKOPT: 17360 case OP_GETPEERNAME: 17361 case OP_FTRREAD: 17362 case OP_FTRWRITE: 17363 case OP_FTREXEC: 17364 case OP_FTROWNED: 17365 case OP_FTEREAD: 17366 case OP_FTEWRITE: 17367 case OP_FTEEXEC: 17368 case OP_FTEOWNED: 17369 case OP_FTIS: 17370 case OP_FTZERO: 17371 case OP_FTSIZE: 17372 case OP_FTFILE: 17373 case OP_FTDIR: 17374 case OP_FTLINK: 17375 case OP_FTPIPE: 17376 case OP_FTSOCK: 17377 case OP_FTBLK: 17378 case OP_FTCHR: 17379 case OP_FTTTY: 17380 case OP_FTSUID: 17381 case OP_FTSGID: 17382 case OP_FTSVTX: 17383 case OP_FTTEXT: 17384 case OP_FTBINARY: 17385 case OP_FTMTIME: 17386 case OP_FTATIME: 17387 case OP_FTCTIME: 17388 case OP_READLINK: 17389 case OP_OPEN_DIR: 17390 case OP_READDIR: 17391 case OP_TELLDIR: 17392 case OP_SEEKDIR: 17393 case OP_REWINDDIR: 17394 case OP_CLOSEDIR: 17395 case OP_GMTIME: 17396 case OP_ALARM: 17397 case OP_SEMGET: 17398 case OP_GETLOGIN: 17399 case OP_SUBSTR: 17400 case OP_AEACH: 17401 case OP_EACH: 17402 case OP_SORT: 17403 case OP_CALLER: 17404 case OP_DOFILE: 17405 case OP_PROTOTYPE: 17406 case OP_NCMP: 17407 case OP_SMARTMATCH: 17408 case OP_UNPACK: 17409 case OP_SYSOPEN: 17410 case OP_SYSSEEK: 17411 match = 1; 17412 goto do_op; 17413 17414 case OP_ENTERSUB: 17415 case OP_GOTO: 17416 /* XXX tmp hack: these two may call an XS sub, and currently 17417 XS subs don't have a SUB entry on the context stack, so CV and 17418 pad determination goes wrong, and BAD things happen. So, just 17419 don't try to determine the value under those circumstances. 17420 Need a better fix at dome point. DAPM 11/2007 */ 17421 break; 17422 17423 case OP_FLIP: 17424 case OP_FLOP: 17425 { 17426 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV); 17427 if (gv && GvSV(gv) == uninit_sv) 17428 return newSVpvs_flags("$.", SVs_TEMP); 17429 goto do_op; 17430 } 17431 17432 case OP_POS: 17433 /* def-ness of rval pos() is independent of the def-ness of its arg */ 17434 if ( !(obase->op_flags & OPf_MOD)) 17435 break; 17436 /* FALLTHROUGH */ 17437 17438 case OP_SCHOMP: 17439 case OP_CHOMP: 17440 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs)) 17441 return newSVpvs_flags("${$/}", SVs_TEMP); 17442 /* FALLTHROUGH */ 17443 17444 default: 17445 do_op: 17446 if (!(obase->op_flags & OPf_KIDS)) 17447 break; 17448 o = cUNOPx(obase)->op_first; 17449 17450 do_op2: 17451 if (!o) 17452 break; 17453 17454 /* This loop checks all the kid ops, skipping any that cannot pos- 17455 * sibly be responsible for the uninitialized value; i.e., defined 17456 * constants and ops that return nothing. If there is only one op 17457 * left that is not skipped, then we *know* it is responsible for 17458 * the uninitialized value. If there is more than one op left, we 17459 * have to look for an exact match in the while() loop below. 17460 * Note that we skip padrange, because the individual pad ops that 17461 * it replaced are still in the tree, so we work on them instead. 17462 */ 17463 o2 = NULL; 17464 for (kid=o; kid; kid = OpSIBLING(kid)) { 17465 const OPCODE type = kid->op_type; 17466 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid))) 17467 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS)) 17468 || (type == OP_PUSHMARK) 17469 || (type == OP_PADRANGE) 17470 ) 17471 continue; 17472 17473 if (o2) { /* more than one found */ 17474 o2 = NULL; 17475 break; 17476 } 17477 o2 = kid; 17478 } 17479 if (o2) 17480 return find_uninit_var(o2, uninit_sv, match, desc_p); 17481 17482 /* scan all args */ 17483 while (o) { 17484 sv = find_uninit_var(o, uninit_sv, 1, desc_p); 17485 if (sv) 17486 return sv; 17487 o = OpSIBLING(o); 17488 } 17489 break; 17490 } 17491 return NULL; 17492} 17493 17494 17495/* 17496=for apidoc_section $warning 17497=for apidoc report_uninit 17498 17499Print appropriate "Use of uninitialized variable" warning. 17500 17501=cut 17502*/ 17503 17504void 17505Perl_report_uninit(pTHX_ const SV *uninit_sv) 17506{ 17507 const char *desc = NULL; 17508 SV* varname = NULL; 17509 17510 if (PL_op) { 17511 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded 17512 ? "join or string" 17513 : PL_op->op_type == OP_MULTICONCAT 17514 && (PL_op->op_private & OPpMULTICONCAT_FAKE) 17515 ? "sprintf" 17516 : OP_DESC(PL_op); 17517 if (uninit_sv && PL_curpad) { 17518 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc); 17519 if (varname) 17520 sv_insert(varname, 0, 0, " ", 1); 17521 } 17522 } 17523 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0) 17524 /* we've reached the end of a sort block or sub, 17525 * and the uninit value is probably what that code returned */ 17526 desc = "sort"; 17527 17528 /* PL_warn_uninit_sv is constant */ 17529 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral); 17530 if (desc) 17531 /* diag_listed_as: Use of uninitialized value%s */ 17532 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv, 17533 SVfARG(varname ? varname : &PL_sv_no), 17534 " in ", desc); 17535 else 17536 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit, 17537 "", "", ""); 17538 GCC_DIAG_RESTORE_STMT; 17539} 17540 17541/* 17542 * ex: set ts=8 sts=4 sw=4 et: 17543 */ 17544