| 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 | |
| 128 | /* ============================================================================ |
| 129 | |
| 130 | =head1 Allocation and deallocation of SVs. |
| 131 | An SV (or AV, HV, etc.) is allocated in two parts: the head (struct |
| 132 | sv, av, hv...) contains type and reference count information, and for |
| 133 | many types, a pointer to the body (struct xrv, xpv, xpviv...), which |
| 134 | contains fields specific to each type. Some types store all they need |
| 135 | in the head, so don't have a body. |
| 136 | |
| 137 | In all but the most memory-paranoid configurations (ex: PURIFY), heads |
| 138 | and bodies are allocated out of arenas, which by default are |
| 139 | approximately 4K chunks of memory parcelled up into N heads or bodies. |
| 140 | Sv-bodies are allocated by their sv-type, guaranteeing size |
| 141 | consistency needed to allocate safely from arrays. |
| 142 | |
| 143 | For SV-heads, the first slot in each arena is reserved, and holds a |
| 144 | link to the next arena, some flags, and a note of the number of slots. |
| 145 | Snaked through each arena chain is a linked list of free items; when |
| 146 | this becomes empty, an extra arena is allocated and divided up into N |
| 147 | items which are threaded into the free list. |
| 148 | |
| 149 | SV-bodies are similar, but they use arena-sets by default, which |
| 150 | separate the link and info from the arena itself, and reclaim the 1st |
| 151 | slot in the arena. SV-bodies are further described later. |
| 152 | |
| 153 | The following global variables are associated with arenas: |
| 154 | |
| 155 | PL_sv_arenaroot pointer to list of SV arenas |
| 156 | PL_sv_root pointer to list of free SV structures |
| 157 | |
| 158 | PL_body_arenas head of linked-list of body arenas |
| 159 | PL_body_roots[] array of pointers to list of free bodies of svtype |
| 160 | arrays are indexed by the svtype needed |
| 161 | |
| 162 | A few special SV heads are not allocated from an arena, but are |
| 163 | instead directly created in the interpreter structure, eg PL_sv_undef. |
| 164 | The size of arenas can be changed from the default by setting |
| 165 | PERL_ARENA_SIZE appropriately at compile time. |
| 166 | |
| 167 | The SV arena serves the secondary purpose of allowing still-live SVs |
| 168 | to be located and destroyed during final cleanup. |
| 169 | |
| 170 | At the lowest level, the macros new_SV() and del_SV() grab and free |
| 171 | an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv() |
| 172 | to return the SV to the free list with error checking.) new_SV() calls |
| 173 | more_sv() / sv_add_arena() to add an extra arena if the free list is empty. |
| 174 | SVs in the free list have their SvTYPE field set to all ones. |
| 175 | |
| 176 | At the time of very final cleanup, sv_free_arenas() is called from |
| 177 | perl_destruct() to physically free all the arenas allocated since the |
| 178 | start of the interpreter. |
| 179 | |
| 180 | The function visit() scans the SV arenas list, and calls a specified |
| 181 | function for each SV it finds which is still live - ie which has an SvTYPE |
| 182 | other than all 1's, and a non-zero SvREFCNT. visit() is used by the |
| 183 | following functions (specified as [function that calls visit()] / [function |
| 184 | called by visit() for each SV]): |
| 185 | |
| 186 | sv_report_used() / do_report_used() |
| 187 | dump all remaining SVs (debugging aid) |
| 188 | |
| 189 | sv_clean_objs() / do_clean_objs(),do_clean_named_objs(), |
| 190 | do_clean_named_io_objs(),do_curse() |
| 191 | Attempt to free all objects pointed to by RVs, |
| 192 | try to do the same for all objects indir- |
| 193 | ectly referenced by typeglobs too, and |
| 194 | then do a final sweep, cursing any |
| 195 | objects that remain. Called once from |
| 196 | perl_destruct(), prior to calling sv_clean_all() |
| 197 | below. |
| 198 | |
| 199 | sv_clean_all() / do_clean_all() |
| 200 | SvREFCNT_dec(sv) each remaining SV, possibly |
| 201 | triggering an sv_free(). It also sets the |
| 202 | SVf_BREAK flag on the SV to indicate that the |
| 203 | refcnt has been artificially lowered, and thus |
| 204 | stopping sv_free() from giving spurious warnings |
| 205 | about SVs which unexpectedly have a refcnt |
| 206 | of zero. called repeatedly from perl_destruct() |
| 207 | until there are no SVs left. |
| 208 | |
| 209 | =head2 Arena allocator API Summary |
| 210 | |
| 211 | Private API to rest of sv.c |
| 212 | |
| 213 | new_SV(), del_SV(), |
| 214 | |
| 215 | new_XPVNV(), del_XPVGV(), |
| 216 | etc |
| 217 | |
| 218 | Public API: |
| 219 | |
| 220 | sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas() |
| 221 | |
| 222 | =cut |
| 223 | |
| 224 | * ========================================================================= */ |
| 225 | |
| 226 | /* |
| 227 | * "A time to plant, and a time to uproot what was planted..." |
| 228 | */ |
| 229 | |
| 230 | #ifdef PERL_MEM_LOG |
| 231 | # define MEM_LOG_NEW_SV(sv, file, line, func) \ |
| 232 | Perl_mem_log_new_sv(sv, file, line, func) |
| 233 | # define MEM_LOG_DEL_SV(sv, file, line, func) \ |
| 234 | Perl_mem_log_del_sv(sv, file, line, func) |
| 235 | #else |
| 236 | # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP |
| 237 | # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP |
| 238 | #endif |
| 239 | |
| 240 | #ifdef DEBUG_LEAKING_SCALARS |
| 241 | # define FREE_SV_DEBUG_FILE(sv) STMT_START { \ |
| 242 | if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \ |
| 243 | } STMT_END |
| 244 | # define DEBUG_SV_SERIAL(sv) \ |
| 245 | DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \ |
| 246 | PTR2UV(sv), (long)(sv)->sv_debug_serial)) |
| 247 | #else |
| 248 | # define FREE_SV_DEBUG_FILE(sv) |
| 249 | # define DEBUG_SV_SERIAL(sv) NOOP |
| 250 | #endif |
| 251 | |
| 252 | #ifdef PERL_POISON |
| 253 | # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv) |
| 254 | # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val)) |
| 255 | /* Whilst I'd love to do this, it seems that things like to check on |
| 256 | unreferenced scalars |
| 257 | # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV) |
| 258 | */ |
| 259 | # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \ |
| 260 | PoisonNew(&SvREFCNT(sv), 1, U32) |
| 261 | #else |
| 262 | # define SvARENA_CHAIN(sv) SvANY(sv) |
| 263 | # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val) |
| 264 | # define POISON_SV_HEAD(sv) |
| 265 | #endif |
| 266 | |
| 267 | /* Mark an SV head as unused, and add to free list. |
| 268 | * |
| 269 | * If SVf_BREAK is set, skip adding it to the free list, as this SV had |
| 270 | * its refcount artificially decremented during global destruction, so |
| 271 | * there may be dangling pointers to it. The last thing we want in that |
| 272 | * case is for it to be reused. */ |
| 273 | |
| 274 | #define plant_SV(p) \ |
| 275 | STMT_START { \ |
| 276 | const U32 old_flags = SvFLAGS(p); \ |
| 277 | MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \ |
| 278 | DEBUG_SV_SERIAL(p); \ |
| 279 | FREE_SV_DEBUG_FILE(p); \ |
| 280 | POISON_SV_HEAD(p); \ |
| 281 | SvFLAGS(p) = SVTYPEMASK; \ |
| 282 | if (!(old_flags & SVf_BREAK)) { \ |
| 283 | SvARENA_CHAIN_SET(p, PL_sv_root); \ |
| 284 | PL_sv_root = (p); \ |
| 285 | } \ |
| 286 | --PL_sv_count; \ |
| 287 | } STMT_END |
| 288 | |
| 289 | #define uproot_SV(p) \ |
| 290 | STMT_START { \ |
| 291 | (p) = PL_sv_root; \ |
| 292 | PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \ |
| 293 | ++PL_sv_count; \ |
| 294 | } STMT_END |
| 295 | |
| 296 | |
| 297 | /* make some more SVs by adding another arena */ |
| 298 | |
| 299 | STATIC SV* |
| 300 | S_more_sv(pTHX) |
| 301 | { |
| 302 | SV* sv; |
| 303 | char *chunk; /* must use New here to match call to */ |
| 304 | Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */ |
| 305 | sv_add_arena(chunk, PERL_ARENA_SIZE, 0); |
| 306 | uproot_SV(sv); |
| 307 | return sv; |
| 308 | } |
| 309 | |
| 310 | /* new_SV(): return a new, empty SV head */ |
| 311 | |
| 312 | #ifdef DEBUG_LEAKING_SCALARS |
| 313 | /* provide a real function for a debugger to play with */ |
| 314 | STATIC SV* |
| 315 | S_new_SV(pTHX_ const char *file, int line, const char *func) |
| 316 | { |
| 317 | SV* sv; |
| 318 | |
| 319 | if (PL_sv_root) |
| 320 | uproot_SV(sv); |
| 321 | else |
| 322 | sv = S_more_sv(aTHX); |
| 323 | SvANY(sv) = 0; |
| 324 | SvREFCNT(sv) = 1; |
| 325 | SvFLAGS(sv) = 0; |
| 326 | sv->sv_debug_optype = PL_op ? PL_op->op_type : 0; |
| 327 | sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE |
| 328 | ? PL_parser->copline |
| 329 | : PL_curcop |
| 330 | ? CopLINE(PL_curcop) |
| 331 | : 0 |
| 332 | ); |
| 333 | sv->sv_debug_inpad = 0; |
| 334 | sv->sv_debug_parent = NULL; |
| 335 | sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL; |
| 336 | |
| 337 | sv->sv_debug_serial = PL_sv_serial++; |
| 338 | |
| 339 | MEM_LOG_NEW_SV(sv, file, line, func); |
| 340 | DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n", |
| 341 | PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func)); |
| 342 | |
| 343 | return sv; |
| 344 | } |
| 345 | # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__) |
| 346 | |
| 347 | #else |
| 348 | # define new_SV(p) \ |
| 349 | STMT_START { \ |
| 350 | if (PL_sv_root) \ |
| 351 | uproot_SV(p); \ |
| 352 | else \ |
| 353 | (p) = S_more_sv(aTHX); \ |
| 354 | SvANY(p) = 0; \ |
| 355 | SvREFCNT(p) = 1; \ |
| 356 | SvFLAGS(p) = 0; \ |
| 357 | MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \ |
| 358 | } STMT_END |
| 359 | #endif |
| 360 | |
| 361 | |
| 362 | /* del_SV(): return an empty SV head to the free list */ |
| 363 | |
| 364 | #ifdef DEBUGGING |
| 365 | |
| 366 | #define del_SV(p) \ |
| 367 | STMT_START { \ |
| 368 | if (DEBUG_D_TEST) \ |
| 369 | del_sv(p); \ |
| 370 | else \ |
| 371 | plant_SV(p); \ |
| 372 | } STMT_END |
| 373 | |
| 374 | STATIC void |
| 375 | S_del_sv(pTHX_ SV *p) |
| 376 | { |
| 377 | PERL_ARGS_ASSERT_DEL_SV; |
| 378 | |
| 379 | if (DEBUG_D_TEST) { |
| 380 | SV* sva; |
| 381 | bool ok = 0; |
| 382 | for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) { |
| 383 | const SV * const sv = sva + 1; |
| 384 | const SV * const svend = &sva[SvREFCNT(sva)]; |
| 385 | if (p >= sv && p < svend) { |
| 386 | ok = 1; |
| 387 | break; |
| 388 | } |
| 389 | } |
| 390 | if (!ok) { |
| 391 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), |
| 392 | "Attempt to free non-arena SV: 0x%"UVxf |
| 393 | pTHX__FORMAT, PTR2UV(p) pTHX__VALUE); |
| 394 | return; |
| 395 | } |
| 396 | } |
| 397 | plant_SV(p); |
| 398 | } |
| 399 | |
| 400 | #else /* ! DEBUGGING */ |
| 401 | |
| 402 | #define del_SV(p) plant_SV(p) |
| 403 | |
| 404 | #endif /* DEBUGGING */ |
| 405 | |
| 406 | |
| 407 | /* |
| 408 | =head1 SV Manipulation Functions |
| 409 | |
| 410 | =for apidoc sv_add_arena |
| 411 | |
| 412 | Given a chunk of memory, link it to the head of the list of arenas, |
| 413 | and split it into a list of free SVs. |
| 414 | |
| 415 | =cut |
| 416 | */ |
| 417 | |
| 418 | static void |
| 419 | S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags) |
| 420 | { |
| 421 | SV *const sva = MUTABLE_SV(ptr); |
| 422 | SV* sv; |
| 423 | SV* svend; |
| 424 | |
| 425 | PERL_ARGS_ASSERT_SV_ADD_ARENA; |
| 426 | |
| 427 | /* The first SV in an arena isn't an SV. */ |
| 428 | SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */ |
| 429 | SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */ |
| 430 | SvFLAGS(sva) = flags; /* FAKE if not to be freed */ |
| 431 | |
| 432 | PL_sv_arenaroot = sva; |
| 433 | PL_sv_root = sva + 1; |
| 434 | |
| 435 | svend = &sva[SvREFCNT(sva) - 1]; |
| 436 | sv = sva + 1; |
| 437 | while (sv < svend) { |
| 438 | SvARENA_CHAIN_SET(sv, (sv + 1)); |
| 439 | #ifdef DEBUGGING |
| 440 | SvREFCNT(sv) = 0; |
| 441 | #endif |
| 442 | /* Must always set typemask because it's always checked in on cleanup |
| 443 | when the arenas are walked looking for objects. */ |
| 444 | SvFLAGS(sv) = SVTYPEMASK; |
| 445 | sv++; |
| 446 | } |
| 447 | SvARENA_CHAIN_SET(sv, 0); |
| 448 | #ifdef DEBUGGING |
| 449 | SvREFCNT(sv) = 0; |
| 450 | #endif |
| 451 | SvFLAGS(sv) = SVTYPEMASK; |
| 452 | } |
| 453 | |
| 454 | /* visit(): call the named function for each non-free SV in the arenas |
| 455 | * whose flags field matches the flags/mask args. */ |
| 456 | |
| 457 | STATIC I32 |
| 458 | S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask) |
| 459 | { |
| 460 | SV* sva; |
| 461 | I32 visited = 0; |
| 462 | |
| 463 | PERL_ARGS_ASSERT_VISIT; |
| 464 | |
| 465 | for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) { |
| 466 | const SV * const svend = &sva[SvREFCNT(sva)]; |
| 467 | SV* sv; |
| 468 | for (sv = sva + 1; sv < svend; ++sv) { |
| 469 | if (SvTYPE(sv) != (svtype)SVTYPEMASK |
| 470 | && (sv->sv_flags & mask) == flags |
| 471 | && SvREFCNT(sv)) |
| 472 | { |
| 473 | (*f)(aTHX_ sv); |
| 474 | ++visited; |
| 475 | } |
| 476 | } |
| 477 | } |
| 478 | return visited; |
| 479 | } |
| 480 | |
| 481 | #ifdef DEBUGGING |
| 482 | |
| 483 | /* called by sv_report_used() for each live SV */ |
| 484 | |
| 485 | static void |
| 486 | do_report_used(pTHX_ SV *const sv) |
| 487 | { |
| 488 | if (SvTYPE(sv) != (svtype)SVTYPEMASK) { |
| 489 | PerlIO_printf(Perl_debug_log, "****\n"); |
| 490 | sv_dump(sv); |
| 491 | } |
| 492 | } |
| 493 | #endif |
| 494 | |
| 495 | /* |
| 496 | =for apidoc sv_report_used |
| 497 | |
| 498 | Dump the contents of all SVs not yet freed (debugging aid). |
| 499 | |
| 500 | =cut |
| 501 | */ |
| 502 | |
| 503 | void |
| 504 | Perl_sv_report_used(pTHX) |
| 505 | { |
| 506 | #ifdef DEBUGGING |
| 507 | visit(do_report_used, 0, 0); |
| 508 | #else |
| 509 | PERL_UNUSED_CONTEXT; |
| 510 | #endif |
| 511 | } |
| 512 | |
| 513 | /* called by sv_clean_objs() for each live SV */ |
| 514 | |
| 515 | static void |
| 516 | do_clean_objs(pTHX_ SV *const ref) |
| 517 | { |
| 518 | assert (SvROK(ref)); |
| 519 | { |
| 520 | SV * const target = SvRV(ref); |
| 521 | if (SvOBJECT(target)) { |
| 522 | DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref))); |
| 523 | if (SvWEAKREF(ref)) { |
| 524 | sv_del_backref(target, ref); |
| 525 | SvWEAKREF_off(ref); |
| 526 | SvRV_set(ref, NULL); |
| 527 | } else { |
| 528 | SvROK_off(ref); |
| 529 | SvRV_set(ref, NULL); |
| 530 | SvREFCNT_dec_NN(target); |
| 531 | } |
| 532 | } |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | |
| 537 | /* clear any slots in a GV which hold objects - except IO; |
| 538 | * called by sv_clean_objs() for each live GV */ |
| 539 | |
| 540 | static void |
| 541 | do_clean_named_objs(pTHX_ SV *const sv) |
| 542 | { |
| 543 | SV *obj; |
| 544 | assert(SvTYPE(sv) == SVt_PVGV); |
| 545 | assert(isGV_with_GP(sv)); |
| 546 | if (!GvGP(sv)) |
| 547 | return; |
| 548 | |
| 549 | /* freeing GP entries may indirectly free the current GV; |
| 550 | * hold onto it while we mess with the GP slots */ |
| 551 | SvREFCNT_inc(sv); |
| 552 | |
| 553 | if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) { |
| 554 | DEBUG_D((PerlIO_printf(Perl_debug_log, |
| 555 | "Cleaning named glob SV object:\n "), sv_dump(obj))); |
| 556 | GvSV(sv) = NULL; |
| 557 | SvREFCNT_dec_NN(obj); |
| 558 | } |
| 559 | if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) { |
| 560 | DEBUG_D((PerlIO_printf(Perl_debug_log, |
| 561 | "Cleaning named glob AV object:\n "), sv_dump(obj))); |
| 562 | GvAV(sv) = NULL; |
| 563 | SvREFCNT_dec_NN(obj); |
| 564 | } |
| 565 | if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) { |
| 566 | DEBUG_D((PerlIO_printf(Perl_debug_log, |
| 567 | "Cleaning named glob HV object:\n "), sv_dump(obj))); |
| 568 | GvHV(sv) = NULL; |
| 569 | SvREFCNT_dec_NN(obj); |
| 570 | } |
| 571 | if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) { |
| 572 | DEBUG_D((PerlIO_printf(Perl_debug_log, |
| 573 | "Cleaning named glob CV object:\n "), sv_dump(obj))); |
| 574 | GvCV_set(sv, NULL); |
| 575 | SvREFCNT_dec_NN(obj); |
| 576 | } |
| 577 | SvREFCNT_dec_NN(sv); /* undo the inc above */ |
| 578 | } |
| 579 | |
| 580 | /* clear any IO slots in a GV which hold objects (except stderr, defout); |
| 581 | * called by sv_clean_objs() for each live GV */ |
| 582 | |
| 583 | static void |
| 584 | do_clean_named_io_objs(pTHX_ SV *const sv) |
| 585 | { |
| 586 | SV *obj; |
| 587 | assert(SvTYPE(sv) == SVt_PVGV); |
| 588 | assert(isGV_with_GP(sv)); |
| 589 | if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv) |
| 590 | return; |
| 591 | |
| 592 | SvREFCNT_inc(sv); |
| 593 | if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) { |
| 594 | DEBUG_D((PerlIO_printf(Perl_debug_log, |
| 595 | "Cleaning named glob IO object:\n "), sv_dump(obj))); |
| 596 | GvIOp(sv) = NULL; |
| 597 | SvREFCNT_dec_NN(obj); |
| 598 | } |
| 599 | SvREFCNT_dec_NN(sv); /* undo the inc above */ |
| 600 | } |
| 601 | |
| 602 | /* Void wrapper to pass to visit() */ |
| 603 | static void |
| 604 | do_curse(pTHX_ SV * const sv) { |
| 605 | if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv) |
| 606 | || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv)) |
| 607 | return; |
| 608 | (void)curse(sv, 0); |
| 609 | } |
| 610 | |
| 611 | /* |
| 612 | =for apidoc sv_clean_objs |
| 613 | |
| 614 | Attempt to destroy all objects not yet freed. |
| 615 | |
| 616 | =cut |
| 617 | */ |
| 618 | |
| 619 | void |
| 620 | Perl_sv_clean_objs(pTHX) |
| 621 | { |
| 622 | GV *olddef, *olderr; |
| 623 | PL_in_clean_objs = TRUE; |
| 624 | visit(do_clean_objs, SVf_ROK, SVf_ROK); |
| 625 | /* Some barnacles may yet remain, clinging to typeglobs. |
| 626 | * Run the non-IO destructors first: they may want to output |
| 627 | * error messages, close files etc */ |
| 628 | visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP); |
| 629 | visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP); |
| 630 | /* And if there are some very tenacious barnacles clinging to arrays, |
| 631 | closures, or what have you.... */ |
| 632 | visit(do_curse, SVs_OBJECT, SVs_OBJECT); |
| 633 | olddef = PL_defoutgv; |
| 634 | PL_defoutgv = NULL; /* disable skip of PL_defoutgv */ |
| 635 | if (olddef && isGV_with_GP(olddef)) |
| 636 | do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef)); |
| 637 | olderr = PL_stderrgv; |
| 638 | PL_stderrgv = NULL; /* disable skip of PL_stderrgv */ |
| 639 | if (olderr && isGV_with_GP(olderr)) |
| 640 | do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr)); |
| 641 | SvREFCNT_dec(olddef); |
| 642 | PL_in_clean_objs = FALSE; |
| 643 | } |
| 644 | |
| 645 | /* called by sv_clean_all() for each live SV */ |
| 646 | |
| 647 | static void |
| 648 | do_clean_all(pTHX_ SV *const sv) |
| 649 | { |
| 650 | if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) { |
| 651 | /* don't clean pid table and strtab */ |
| 652 | return; |
| 653 | } |
| 654 | DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) )); |
| 655 | SvFLAGS(sv) |= SVf_BREAK; |
| 656 | SvREFCNT_dec_NN(sv); |
| 657 | } |
| 658 | |
| 659 | /* |
| 660 | =for apidoc sv_clean_all |
| 661 | |
| 662 | Decrement the refcnt of each remaining SV, possibly triggering a |
| 663 | cleanup. This function may have to be called multiple times to free |
| 664 | SVs which are in complex self-referential hierarchies. |
| 665 | |
| 666 | =cut |
| 667 | */ |
| 668 | |
| 669 | I32 |
| 670 | Perl_sv_clean_all(pTHX) |
| 671 | { |
| 672 | I32 cleaned; |
| 673 | PL_in_clean_all = TRUE; |
| 674 | cleaned = visit(do_clean_all, 0,0); |
| 675 | return cleaned; |
| 676 | } |
| 677 | |
| 678 | /* |
| 679 | ARENASETS: a meta-arena implementation which separates arena-info |
| 680 | into struct arena_set, which contains an array of struct |
| 681 | arena_descs, each holding info for a single arena. By separating |
| 682 | the meta-info from the arena, we recover the 1st slot, formerly |
| 683 | borrowed for list management. The arena_set is about the size of an |
| 684 | arena, avoiding the needless malloc overhead of a naive linked-list. |
| 685 | |
| 686 | The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused |
| 687 | memory in the last arena-set (1/2 on average). In trade, we get |
| 688 | back the 1st slot in each arena (ie 1.7% of a CV-arena, less for |
| 689 | smaller types). The recovery of the wasted space allows use of |
| 690 | small arenas for large, rare body types, by changing array* fields |
| 691 | in body_details_by_type[] below. |
| 692 | */ |
| 693 | struct arena_desc { |
| 694 | char *arena; /* the raw storage, allocated aligned */ |
| 695 | size_t size; /* its size ~4k typ */ |
| 696 | svtype utype; /* bodytype stored in arena */ |
| 697 | }; |
| 698 | |
| 699 | struct arena_set; |
| 700 | |
| 701 | /* Get the maximum number of elements in set[] such that struct arena_set |
| 702 | will fit within PERL_ARENA_SIZE, which is probably just under 4K, and |
| 703 | therefore likely to be 1 aligned memory page. */ |
| 704 | |
| 705 | #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \ |
| 706 | - 2 * sizeof(int)) / sizeof (struct arena_desc)) |
| 707 | |
| 708 | struct arena_set { |
| 709 | struct arena_set* next; |
| 710 | unsigned int set_size; /* ie ARENAS_PER_SET */ |
| 711 | unsigned int curr; /* index of next available arena-desc */ |
| 712 | struct arena_desc set[ARENAS_PER_SET]; |
| 713 | }; |
| 714 | |
| 715 | /* |
| 716 | =for apidoc sv_free_arenas |
| 717 | |
| 718 | Deallocate the memory used by all arenas. Note that all the individual SV |
| 719 | heads and bodies within the arenas must already have been freed. |
| 720 | |
| 721 | =cut |
| 722 | |
| 723 | */ |
| 724 | void |
| 725 | Perl_sv_free_arenas(pTHX) |
| 726 | { |
| 727 | SV* sva; |
| 728 | SV* svanext; |
| 729 | unsigned int i; |
| 730 | |
| 731 | /* Free arenas here, but be careful about fake ones. (We assume |
| 732 | contiguity of the fake ones with the corresponding real ones.) */ |
| 733 | |
| 734 | for (sva = PL_sv_arenaroot; sva; sva = svanext) { |
| 735 | svanext = MUTABLE_SV(SvANY(sva)); |
| 736 | while (svanext && SvFAKE(svanext)) |
| 737 | svanext = MUTABLE_SV(SvANY(svanext)); |
| 738 | |
| 739 | if (!SvFAKE(sva)) |
| 740 | Safefree(sva); |
| 741 | } |
| 742 | |
| 743 | { |
| 744 | struct arena_set *aroot = (struct arena_set*) PL_body_arenas; |
| 745 | |
| 746 | while (aroot) { |
| 747 | struct arena_set *current = aroot; |
| 748 | i = aroot->curr; |
| 749 | while (i--) { |
| 750 | assert(aroot->set[i].arena); |
| 751 | Safefree(aroot->set[i].arena); |
| 752 | } |
| 753 | aroot = aroot->next; |
| 754 | Safefree(current); |
| 755 | } |
| 756 | } |
| 757 | PL_body_arenas = 0; |
| 758 | |
| 759 | i = PERL_ARENA_ROOTS_SIZE; |
| 760 | while (i--) |
| 761 | PL_body_roots[i] = 0; |
| 762 | |
| 763 | PL_sv_arenaroot = 0; |
| 764 | PL_sv_root = 0; |
| 765 | } |
| 766 | |
| 767 | /* |
| 768 | Here are mid-level routines that manage the allocation of bodies out |
| 769 | of the various arenas. There are 5 kinds of arenas: |
| 770 | |
| 771 | 1. SV-head arenas, which are discussed and handled above |
| 772 | 2. regular body arenas |
| 773 | 3. arenas for reduced-size bodies |
| 774 | 4. Hash-Entry arenas |
| 775 | |
| 776 | Arena types 2 & 3 are chained by body-type off an array of |
| 777 | arena-root pointers, which is indexed by svtype. Some of the |
| 778 | larger/less used body types are malloced singly, since a large |
| 779 | unused block of them is wasteful. Also, several svtypes dont have |
| 780 | bodies; the data fits into the sv-head itself. The arena-root |
| 781 | pointer thus has a few unused root-pointers (which may be hijacked |
| 782 | later for arena types 4,5) |
| 783 | |
| 784 | 3 differs from 2 as an optimization; some body types have several |
| 785 | unused fields in the front of the structure (which are kept in-place |
| 786 | for consistency). These bodies can be allocated in smaller chunks, |
| 787 | because the leading fields arent accessed. Pointers to such bodies |
| 788 | are decremented to point at the unused 'ghost' memory, knowing that |
| 789 | the pointers are used with offsets to the real memory. |
| 790 | |
| 791 | |
| 792 | =head1 SV-Body Allocation |
| 793 | |
| 794 | =cut |
| 795 | |
| 796 | Allocation of SV-bodies is similar to SV-heads, differing as follows; |
| 797 | the allocation mechanism is used for many body types, so is somewhat |
| 798 | more complicated, it uses arena-sets, and has no need for still-live |
| 799 | SV detection. |
| 800 | |
| 801 | At the outermost level, (new|del)_X*V macros return bodies of the |
| 802 | appropriate type. These macros call either (new|del)_body_type or |
| 803 | (new|del)_body_allocated macro pairs, depending on specifics of the |
| 804 | type. Most body types use the former pair, the latter pair is used to |
| 805 | allocate body types with "ghost fields". |
| 806 | |
| 807 | "ghost fields" are fields that are unused in certain types, and |
| 808 | consequently don't need to actually exist. They are declared because |
| 809 | they're part of a "base type", which allows use of functions as |
| 810 | methods. The simplest examples are AVs and HVs, 2 aggregate types |
| 811 | which don't use the fields which support SCALAR semantics. |
| 812 | |
| 813 | For these types, the arenas are carved up into appropriately sized |
| 814 | chunks, we thus avoid wasted memory for those unaccessed members. |
| 815 | When bodies are allocated, we adjust the pointer back in memory by the |
| 816 | size of the part not allocated, so it's as if we allocated the full |
| 817 | structure. (But things will all go boom if you write to the part that |
| 818 | is "not there", because you'll be overwriting the last members of the |
| 819 | preceding structure in memory.) |
| 820 | |
| 821 | We calculate the correction using the STRUCT_OFFSET macro on the first |
| 822 | member present. If the allocated structure is smaller (no initial NV |
| 823 | actually allocated) then the net effect is to subtract the size of the NV |
| 824 | from the pointer, to return a new pointer as if an initial NV were actually |
| 825 | allocated. (We were using structures named *_allocated for this, but |
| 826 | this turned out to be a subtle bug, because a structure without an NV |
| 827 | could have a lower alignment constraint, but the compiler is allowed to |
| 828 | optimised accesses based on the alignment constraint of the actual pointer |
| 829 | to the full structure, for example, using a single 64 bit load instruction |
| 830 | because it "knows" that two adjacent 32 bit members will be 8-byte aligned.) |
| 831 | |
| 832 | This is the same trick as was used for NV and IV bodies. Ironically it |
| 833 | doesn't need to be used for NV bodies any more, because NV is now at |
| 834 | the start of the structure. IV bodies, and also in some builds NV bodies, |
| 835 | don't need it either, because they are no longer allocated. |
| 836 | |
| 837 | In turn, the new_body_* allocators call S_new_body(), which invokes |
| 838 | new_body_inline macro, which takes a lock, and takes a body off the |
| 839 | linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if |
| 840 | necessary to refresh an empty list. Then the lock is released, and |
| 841 | the body is returned. |
| 842 | |
| 843 | Perl_more_bodies allocates a new arena, and carves it up into an array of N |
| 844 | bodies, which it strings into a linked list. It looks up arena-size |
| 845 | and body-size from the body_details table described below, thus |
| 846 | supporting the multiple body-types. |
| 847 | |
| 848 | If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and |
| 849 | the (new|del)_X*V macros are mapped directly to malloc/free. |
| 850 | |
| 851 | For each sv-type, struct body_details bodies_by_type[] carries |
| 852 | parameters which control these aspects of SV handling: |
| 853 | |
| 854 | Arena_size determines whether arenas are used for this body type, and if |
| 855 | so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to |
| 856 | zero, forcing individual mallocs and frees. |
| 857 | |
| 858 | Body_size determines how big a body is, and therefore how many fit into |
| 859 | each arena. Offset carries the body-pointer adjustment needed for |
| 860 | "ghost fields", and is used in *_allocated macros. |
| 861 | |
| 862 | But its main purpose is to parameterize info needed in |
| 863 | Perl_sv_upgrade(). The info here dramatically simplifies the function |
| 864 | vs the implementation in 5.8.8, making it table-driven. All fields |
| 865 | are used for this, except for arena_size. |
| 866 | |
| 867 | For the sv-types that have no bodies, arenas are not used, so those |
| 868 | PL_body_roots[sv_type] are unused, and can be overloaded. In |
| 869 | something of a special case, SVt_NULL is borrowed for HE arenas; |
| 870 | PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the |
| 871 | bodies_by_type[SVt_NULL] slot is not used, as the table is not |
| 872 | available in hv.c. |
| 873 | |
| 874 | */ |
| 875 | |
| 876 | struct body_details { |
| 877 | U8 body_size; /* Size to allocate */ |
| 878 | U8 copy; /* Size of structure to copy (may be shorter) */ |
| 879 | U8 offset; /* Size of unalloced ghost fields to first alloced field*/ |
| 880 | PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */ |
| 881 | PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */ |
| 882 | PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */ |
| 883 | PERL_BITFIELD8 arena : 1; /* Allocated from an arena */ |
| 884 | U32 arena_size; /* Size of arena to allocate */ |
| 885 | }; |
| 886 | |
| 887 | #define HADNV FALSE |
| 888 | #define NONV TRUE |
| 889 | |
| 890 | |
| 891 | #ifdef PURIFY |
| 892 | /* With -DPURFIY we allocate everything directly, and don't use arenas. |
| 893 | This seems a rather elegant way to simplify some of the code below. */ |
| 894 | #define HASARENA FALSE |
| 895 | #else |
| 896 | #define HASARENA TRUE |
| 897 | #endif |
| 898 | #define NOARENA FALSE |
| 899 | |
| 900 | /* Size the arenas to exactly fit a given number of bodies. A count |
| 901 | of 0 fits the max number bodies into a PERL_ARENA_SIZE.block, |
| 902 | simplifying the default. If count > 0, the arena is sized to fit |
| 903 | only that many bodies, allowing arenas to be used for large, rare |
| 904 | bodies (XPVFM, XPVIO) without undue waste. The arena size is |
| 905 | limited by PERL_ARENA_SIZE, so we can safely oversize the |
| 906 | declarations. |
| 907 | */ |
| 908 | #define FIT_ARENA0(body_size) \ |
| 909 | ((size_t)(PERL_ARENA_SIZE / body_size) * body_size) |
| 910 | #define FIT_ARENAn(count,body_size) \ |
| 911 | ( count * body_size <= PERL_ARENA_SIZE) \ |
| 912 | ? count * body_size \ |
| 913 | : FIT_ARENA0 (body_size) |
| 914 | #define FIT_ARENA(count,body_size) \ |
| 915 | (U32)(count \ |
| 916 | ? FIT_ARENAn (count, body_size) \ |
| 917 | : FIT_ARENA0 (body_size)) |
| 918 | |
| 919 | /* Calculate the length to copy. Specifically work out the length less any |
| 920 | final padding the compiler needed to add. See the comment in sv_upgrade |
| 921 | for why copying the padding proved to be a bug. */ |
| 922 | |
| 923 | #define copy_length(type, last_member) \ |
| 924 | STRUCT_OFFSET(type, last_member) \ |
| 925 | + sizeof (((type*)SvANY((const SV *)0))->last_member) |
| 926 | |
| 927 | static const struct body_details bodies_by_type[] = { |
| 928 | /* HEs use this offset for their arena. */ |
| 929 | { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 }, |
| 930 | |
| 931 | /* IVs are in the head, so the allocation size is 0. */ |
| 932 | { 0, |
| 933 | sizeof(IV), /* This is used to copy out the IV body. */ |
| 934 | STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV, |
| 935 | NOARENA /* IVS don't need an arena */, 0 |
| 936 | }, |
| 937 | |
| 938 | #if NVSIZE <= IVSIZE |
| 939 | { 0, sizeof(NV), |
| 940 | STRUCT_OFFSET(XPVNV, xnv_u), |
| 941 | SVt_NV, FALSE, HADNV, NOARENA, 0 }, |
| 942 | #else |
| 943 | { sizeof(NV), sizeof(NV), |
| 944 | STRUCT_OFFSET(XPVNV, xnv_u), |
| 945 | SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) }, |
| 946 | #endif |
| 947 | |
| 948 | { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur), |
| 949 | copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur), |
| 950 | + STRUCT_OFFSET(XPV, xpv_cur), |
| 951 | SVt_PV, FALSE, NONV, HASARENA, |
| 952 | FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) }, |
| 953 | |
| 954 | { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur), |
| 955 | copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur), |
| 956 | + STRUCT_OFFSET(XPV, xpv_cur), |
| 957 | SVt_INVLIST, TRUE, NONV, HASARENA, |
| 958 | FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) }, |
| 959 | |
| 960 | { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur), |
| 961 | copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur), |
| 962 | + STRUCT_OFFSET(XPV, xpv_cur), |
| 963 | SVt_PVIV, FALSE, NONV, HASARENA, |
| 964 | FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) }, |
| 965 | |
| 966 | { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur), |
| 967 | copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur), |
| 968 | + STRUCT_OFFSET(XPV, xpv_cur), |
| 969 | SVt_PVNV, FALSE, HADNV, HASARENA, |
| 970 | FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) }, |
| 971 | |
| 972 | { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV, |
| 973 | HASARENA, FIT_ARENA(0, sizeof(XPVMG)) }, |
| 974 | |
| 975 | { sizeof(regexp), |
| 976 | sizeof(regexp), |
| 977 | 0, |
| 978 | SVt_REGEXP, TRUE, NONV, HASARENA, |
| 979 | FIT_ARENA(0, sizeof(regexp)) |
| 980 | }, |
| 981 | |
| 982 | { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV, |
| 983 | HASARENA, FIT_ARENA(0, sizeof(XPVGV)) }, |
| 984 | |
| 985 | { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV, |
| 986 | HASARENA, FIT_ARENA(0, sizeof(XPVLV)) }, |
| 987 | |
| 988 | { sizeof(XPVAV), |
| 989 | copy_length(XPVAV, xav_alloc), |
| 990 | 0, |
| 991 | SVt_PVAV, TRUE, NONV, HASARENA, |
| 992 | FIT_ARENA(0, sizeof(XPVAV)) }, |
| 993 | |
| 994 | { sizeof(XPVHV), |
| 995 | copy_length(XPVHV, xhv_max), |
| 996 | 0, |
| 997 | SVt_PVHV, TRUE, NONV, HASARENA, |
| 998 | FIT_ARENA(0, sizeof(XPVHV)) }, |
| 999 | |
| 1000 | { sizeof(XPVCV), |
| 1001 | sizeof(XPVCV), |
| 1002 | 0, |
| 1003 | SVt_PVCV, TRUE, NONV, HASARENA, |
| 1004 | FIT_ARENA(0, sizeof(XPVCV)) }, |
| 1005 | |
| 1006 | { sizeof(XPVFM), |
| 1007 | sizeof(XPVFM), |
| 1008 | 0, |
| 1009 | SVt_PVFM, TRUE, NONV, NOARENA, |
| 1010 | FIT_ARENA(20, sizeof(XPVFM)) }, |
| 1011 | |
| 1012 | { sizeof(XPVIO), |
| 1013 | sizeof(XPVIO), |
| 1014 | 0, |
| 1015 | SVt_PVIO, TRUE, NONV, HASARENA, |
| 1016 | FIT_ARENA(24, sizeof(XPVIO)) }, |
| 1017 | }; |
| 1018 | |
| 1019 | #define new_body_allocated(sv_type) \ |
| 1020 | (void *)((char *)S_new_body(aTHX_ sv_type) \ |
| 1021 | - bodies_by_type[sv_type].offset) |
| 1022 | |
| 1023 | /* return a thing to the free list */ |
| 1024 | |
| 1025 | #define del_body(thing, root) \ |
| 1026 | STMT_START { \ |
| 1027 | void ** const thing_copy = (void **)thing; \ |
| 1028 | *thing_copy = *root; \ |
| 1029 | *root = (void*)thing_copy; \ |
| 1030 | } STMT_END |
| 1031 | |
| 1032 | #ifdef PURIFY |
| 1033 | #if !(NVSIZE <= IVSIZE) |
| 1034 | # define new_XNV() safemalloc(sizeof(XPVNV)) |
| 1035 | #endif |
| 1036 | #define new_XPVNV() safemalloc(sizeof(XPVNV)) |
| 1037 | #define new_XPVMG() safemalloc(sizeof(XPVMG)) |
| 1038 | |
| 1039 | #define del_XPVGV(p) safefree(p) |
| 1040 | |
| 1041 | #else /* !PURIFY */ |
| 1042 | |
| 1043 | #if !(NVSIZE <= IVSIZE) |
| 1044 | # define new_XNV() new_body_allocated(SVt_NV) |
| 1045 | #endif |
| 1046 | #define new_XPVNV() new_body_allocated(SVt_PVNV) |
| 1047 | #define new_XPVMG() new_body_allocated(SVt_PVMG) |
| 1048 | |
| 1049 | #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \ |
| 1050 | &PL_body_roots[SVt_PVGV]) |
| 1051 | |
| 1052 | #endif /* PURIFY */ |
| 1053 | |
| 1054 | /* no arena for you! */ |
| 1055 | |
| 1056 | #define new_NOARENA(details) \ |
| 1057 | safemalloc((details)->body_size + (details)->offset) |
| 1058 | #define new_NOARENAZ(details) \ |
| 1059 | safecalloc((details)->body_size + (details)->offset, 1) |
| 1060 | |
| 1061 | void * |
| 1062 | Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size, |
| 1063 | const size_t arena_size) |
| 1064 | { |
| 1065 | void ** const root = &PL_body_roots[sv_type]; |
| 1066 | struct arena_desc *adesc; |
| 1067 | struct arena_set *aroot = (struct arena_set *) PL_body_arenas; |
| 1068 | unsigned int curr; |
| 1069 | char *start; |
| 1070 | const char *end; |
| 1071 | const size_t good_arena_size = Perl_malloc_good_size(arena_size); |
| 1072 | #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT) |
| 1073 | dVAR; |
| 1074 | #endif |
| 1075 | #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE) |
| 1076 | static bool done_sanity_check; |
| 1077 | |
| 1078 | /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global |
| 1079 | * variables like done_sanity_check. */ |
| 1080 | if (!done_sanity_check) { |
| 1081 | unsigned int i = SVt_LAST; |
| 1082 | |
| 1083 | done_sanity_check = TRUE; |
| 1084 | |
| 1085 | while (i--) |
| 1086 | assert (bodies_by_type[i].type == i); |
| 1087 | } |
| 1088 | #endif |
| 1089 | |
| 1090 | assert(arena_size); |
| 1091 | |
| 1092 | /* may need new arena-set to hold new arena */ |
| 1093 | if (!aroot || aroot->curr >= aroot->set_size) { |
| 1094 | struct arena_set *newroot; |
| 1095 | Newxz(newroot, 1, struct arena_set); |
| 1096 | newroot->set_size = ARENAS_PER_SET; |
| 1097 | newroot->next = aroot; |
| 1098 | aroot = newroot; |
| 1099 | PL_body_arenas = (void *) newroot; |
| 1100 | DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot)); |
| 1101 | } |
| 1102 | |
| 1103 | /* ok, now have arena-set with at least 1 empty/available arena-desc */ |
| 1104 | curr = aroot->curr++; |
| 1105 | adesc = &(aroot->set[curr]); |
| 1106 | assert(!adesc->arena); |
| 1107 | |
| 1108 | Newx(adesc->arena, good_arena_size, char); |
| 1109 | adesc->size = good_arena_size; |
| 1110 | adesc->utype = sv_type; |
| 1111 | DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n", |
| 1112 | curr, (void*)adesc->arena, (UV)good_arena_size)); |
| 1113 | |
| 1114 | start = (char *) adesc->arena; |
| 1115 | |
| 1116 | /* Get the address of the byte after the end of the last body we can fit. |
| 1117 | Remember, this is integer division: */ |
| 1118 | end = start + good_arena_size / body_size * body_size; |
| 1119 | |
| 1120 | /* computed count doesn't reflect the 1st slot reservation */ |
| 1121 | #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE) |
| 1122 | DEBUG_m(PerlIO_printf(Perl_debug_log, |
| 1123 | "arena %p end %p arena-size %d (from %d) type %d " |
| 1124 | "size %d ct %d\n", |
| 1125 | (void*)start, (void*)end, (int)good_arena_size, |
| 1126 | (int)arena_size, sv_type, (int)body_size, |
| 1127 | (int)good_arena_size / (int)body_size)); |
| 1128 | #else |
| 1129 | DEBUG_m(PerlIO_printf(Perl_debug_log, |
| 1130 | "arena %p end %p arena-size %d type %d size %d ct %d\n", |
| 1131 | (void*)start, (void*)end, |
| 1132 | (int)arena_size, sv_type, (int)body_size, |
| 1133 | (int)good_arena_size / (int)body_size)); |
| 1134 | #endif |
| 1135 | *root = (void *)start; |
| 1136 | |
| 1137 | while (1) { |
| 1138 | /* Where the next body would start: */ |
| 1139 | char * const next = start + body_size; |
| 1140 | |
| 1141 | if (next >= end) { |
| 1142 | /* This is the last body: */ |
| 1143 | assert(next == end); |
| 1144 | |
| 1145 | *(void **)start = 0; |
| 1146 | return *root; |
| 1147 | } |
| 1148 | |
| 1149 | *(void**) start = (void *)next; |
| 1150 | start = next; |
| 1151 | } |
| 1152 | } |
| 1153 | |
| 1154 | /* grab a new thing from the free list, allocating more if necessary. |
| 1155 | The inline version is used for speed in hot routines, and the |
| 1156 | function using it serves the rest (unless PURIFY). |
| 1157 | */ |
| 1158 | #define new_body_inline(xpv, sv_type) \ |
| 1159 | STMT_START { \ |
| 1160 | void ** const r3wt = &PL_body_roots[sv_type]; \ |
| 1161 | xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \ |
| 1162 | ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \ |
| 1163 | bodies_by_type[sv_type].body_size,\ |
| 1164 | bodies_by_type[sv_type].arena_size)); \ |
| 1165 | *(r3wt) = *(void**)(xpv); \ |
| 1166 | } STMT_END |
| 1167 | |
| 1168 | #ifndef PURIFY |
| 1169 | |
| 1170 | STATIC void * |
| 1171 | S_new_body(pTHX_ const svtype sv_type) |
| 1172 | { |
| 1173 | void *xpv; |
| 1174 | new_body_inline(xpv, sv_type); |
| 1175 | return xpv; |
| 1176 | } |
| 1177 | |
| 1178 | #endif |
| 1179 | |
| 1180 | static const struct body_details fake_rv = |
| 1181 | { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 }; |
| 1182 | |
| 1183 | /* |
| 1184 | =for apidoc sv_upgrade |
| 1185 | |
| 1186 | Upgrade an SV to a more complex form. Generally adds a new body type to the |
| 1187 | SV, then copies across as much information as possible from the old body. |
| 1188 | It croaks if the SV is already in a more complex form than requested. You |
| 1189 | generally want to use the C<SvUPGRADE> macro wrapper, which checks the type |
| 1190 | before calling C<sv_upgrade>, and hence does not croak. See also |
| 1191 | C<L</svtype>>. |
| 1192 | |
| 1193 | =cut |
| 1194 | */ |
| 1195 | |
| 1196 | void |
| 1197 | Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type) |
| 1198 | { |
| 1199 | void* old_body; |
| 1200 | void* new_body; |
| 1201 | const svtype old_type = SvTYPE(sv); |
| 1202 | const struct body_details *new_type_details; |
| 1203 | const struct body_details *old_type_details |
| 1204 | = bodies_by_type + old_type; |
| 1205 | SV *referant = NULL; |
| 1206 | |
| 1207 | PERL_ARGS_ASSERT_SV_UPGRADE; |
| 1208 | |
| 1209 | if (old_type == new_type) |
| 1210 | return; |
| 1211 | |
| 1212 | /* This clause was purposefully added ahead of the early return above to |
| 1213 | the shared string hackery for (sort {$a <=> $b} keys %hash), with the |
| 1214 | inference by Nick I-S that it would fix other troublesome cases. See |
| 1215 | changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent) |
| 1216 | |
| 1217 | Given that shared hash key scalars are no longer PVIV, but PV, there is |
| 1218 | no longer need to unshare so as to free up the IVX slot for its proper |
| 1219 | purpose. So it's safe to move the early return earlier. */ |
| 1220 | |
| 1221 | if (new_type > SVt_PVMG && SvIsCOW(sv)) { |
| 1222 | sv_force_normal_flags(sv, 0); |
| 1223 | } |
| 1224 | |
| 1225 | old_body = SvANY(sv); |
| 1226 | |
| 1227 | /* Copying structures onto other structures that have been neatly zeroed |
| 1228 | has a subtle gotcha. Consider XPVMG |
| 1229 | |
| 1230 | +------+------+------+------+------+-------+-------+ |
| 1231 | | NV | CUR | LEN | IV | MAGIC | STASH | |
| 1232 | +------+------+------+------+------+-------+-------+ |
| 1233 | 0 4 8 12 16 20 24 28 |
| 1234 | |
| 1235 | where NVs are aligned to 8 bytes, so that sizeof that structure is |
| 1236 | actually 32 bytes long, with 4 bytes of padding at the end: |
| 1237 | |
| 1238 | +------+------+------+------+------+-------+-------+------+ |
| 1239 | | NV | CUR | LEN | IV | MAGIC | STASH | ??? | |
| 1240 | +------+------+------+------+------+-------+-------+------+ |
| 1241 | 0 4 8 12 16 20 24 28 32 |
| 1242 | |
| 1243 | so what happens if you allocate memory for this structure: |
| 1244 | |
| 1245 | +------+------+------+------+------+-------+-------+------+------+... |
| 1246 | | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME | |
| 1247 | +------+------+------+------+------+-------+-------+------+------+... |
| 1248 | 0 4 8 12 16 20 24 28 32 36 |
| 1249 | |
| 1250 | zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you |
| 1251 | expect, because you copy the area marked ??? onto GP. Now, ??? may have |
| 1252 | started out as zero once, but it's quite possible that it isn't. So now, |
| 1253 | rather than a nicely zeroed GP, you have it pointing somewhere random. |
| 1254 | Bugs ensue. |
| 1255 | |
| 1256 | (In fact, GP ends up pointing at a previous GP structure, because the |
| 1257 | principle cause of the padding in XPVMG getting garbage is a copy of |
| 1258 | sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now |
| 1259 | this happens to be moot because XPVGV has been re-ordered, with GP |
| 1260 | no longer after STASH) |
| 1261 | |
| 1262 | So we are careful and work out the size of used parts of all the |
| 1263 | structures. */ |
| 1264 | |
| 1265 | switch (old_type) { |
| 1266 | case SVt_NULL: |
| 1267 | break; |
| 1268 | case SVt_IV: |
| 1269 | if (SvROK(sv)) { |
| 1270 | referant = SvRV(sv); |
| 1271 | old_type_details = &fake_rv; |
| 1272 | if (new_type == SVt_NV) |
| 1273 | new_type = SVt_PVNV; |
| 1274 | } else { |
| 1275 | if (new_type < SVt_PVIV) { |
| 1276 | new_type = (new_type == SVt_NV) |
| 1277 | ? SVt_PVNV : SVt_PVIV; |
| 1278 | } |
| 1279 | } |
| 1280 | break; |
| 1281 | case SVt_NV: |
| 1282 | if (new_type < SVt_PVNV) { |
| 1283 | new_type = SVt_PVNV; |
| 1284 | } |
| 1285 | break; |
| 1286 | case SVt_PV: |
| 1287 | assert(new_type > SVt_PV); |
| 1288 | STATIC_ASSERT_STMT(SVt_IV < SVt_PV); |
| 1289 | STATIC_ASSERT_STMT(SVt_NV < SVt_PV); |
| 1290 | break; |
| 1291 | case SVt_PVIV: |
| 1292 | break; |
| 1293 | case SVt_PVNV: |
| 1294 | break; |
| 1295 | case SVt_PVMG: |
| 1296 | /* Because the XPVMG of PL_mess_sv isn't allocated from the arena, |
| 1297 | there's no way that it can be safely upgraded, because perl.c |
| 1298 | expects to Safefree(SvANY(PL_mess_sv)) */ |
| 1299 | assert(sv != PL_mess_sv); |
| 1300 | break; |
| 1301 | default: |
| 1302 | if (UNLIKELY(old_type_details->cant_upgrade)) |
| 1303 | Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf, |
| 1304 | sv_reftype(sv, 0), (UV) old_type, (UV) new_type); |
| 1305 | } |
| 1306 | |
| 1307 | if (UNLIKELY(old_type > new_type)) |
| 1308 | Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d", |
| 1309 | (int)old_type, (int)new_type); |
| 1310 | |
| 1311 | new_type_details = bodies_by_type + new_type; |
| 1312 | |
| 1313 | SvFLAGS(sv) &= ~SVTYPEMASK; |
| 1314 | SvFLAGS(sv) |= new_type; |
| 1315 | |
| 1316 | /* This can't happen, as SVt_NULL is <= all values of new_type, so one of |
| 1317 | the return statements above will have triggered. */ |
| 1318 | assert (new_type != SVt_NULL); |
| 1319 | switch (new_type) { |
| 1320 | case SVt_IV: |
| 1321 | assert(old_type == SVt_NULL); |
| 1322 | SET_SVANY_FOR_BODYLESS_IV(sv); |
| 1323 | SvIV_set(sv, 0); |
| 1324 | return; |
| 1325 | case SVt_NV: |
| 1326 | assert(old_type == SVt_NULL); |
| 1327 | #if NVSIZE <= IVSIZE |
| 1328 | SET_SVANY_FOR_BODYLESS_NV(sv); |
| 1329 | #else |
| 1330 | SvANY(sv) = new_XNV(); |
| 1331 | #endif |
| 1332 | SvNV_set(sv, 0); |
| 1333 | return; |
| 1334 | case SVt_PVHV: |
| 1335 | case SVt_PVAV: |
| 1336 | assert(new_type_details->body_size); |
| 1337 | |
| 1338 | #ifndef PURIFY |
| 1339 | assert(new_type_details->arena); |
| 1340 | assert(new_type_details->arena_size); |
| 1341 | /* This points to the start of the allocated area. */ |
| 1342 | new_body_inline(new_body, new_type); |
| 1343 | Zero(new_body, new_type_details->body_size, char); |
| 1344 | new_body = ((char *)new_body) - new_type_details->offset; |
| 1345 | #else |
| 1346 | /* We always allocated the full length item with PURIFY. To do this |
| 1347 | we fake things so that arena is false for all 16 types.. */ |
| 1348 | new_body = new_NOARENAZ(new_type_details); |
| 1349 | #endif |
| 1350 | SvANY(sv) = new_body; |
| 1351 | if (new_type == SVt_PVAV) { |
| 1352 | AvMAX(sv) = -1; |
| 1353 | AvFILLp(sv) = -1; |
| 1354 | AvREAL_only(sv); |
| 1355 | if (old_type_details->body_size) { |
| 1356 | AvALLOC(sv) = 0; |
| 1357 | } else { |
| 1358 | /* It will have been zeroed when the new body was allocated. |
| 1359 | Lets not write to it, in case it confuses a write-back |
| 1360 | cache. */ |
| 1361 | } |
| 1362 | } else { |
| 1363 | assert(!SvOK(sv)); |
| 1364 | SvOK_off(sv); |
| 1365 | #ifndef NODEFAULT_SHAREKEYS |
| 1366 | HvSHAREKEYS_on(sv); /* key-sharing on by default */ |
| 1367 | #endif |
| 1368 | /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */ |
| 1369 | HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX; |
| 1370 | } |
| 1371 | |
| 1372 | /* SVt_NULL isn't the only thing upgraded to AV or HV. |
| 1373 | The target created by newSVrv also is, and it can have magic. |
| 1374 | However, it never has SvPVX set. |
| 1375 | */ |
| 1376 | if (old_type == SVt_IV) { |
| 1377 | assert(!SvROK(sv)); |
| 1378 | } else if (old_type >= SVt_PV) { |
| 1379 | assert(SvPVX_const(sv) == 0); |
| 1380 | } |
| 1381 | |
| 1382 | if (old_type >= SVt_PVMG) { |
| 1383 | SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic); |
| 1384 | SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash); |
| 1385 | } else { |
| 1386 | sv->sv_u.svu_array = NULL; /* or svu_hash */ |
| 1387 | } |
| 1388 | break; |
| 1389 | |
| 1390 | case SVt_PVIV: |
| 1391 | /* XXX Is this still needed? Was it ever needed? Surely as there is |
| 1392 | no route from NV to PVIV, NOK can never be true */ |
| 1393 | assert(!SvNOKp(sv)); |
| 1394 | assert(!SvNOK(sv)); |
| 1395 | /* FALLTHROUGH */ |
| 1396 | case SVt_PVIO: |
| 1397 | case SVt_PVFM: |
| 1398 | case SVt_PVGV: |
| 1399 | case SVt_PVCV: |
| 1400 | case SVt_PVLV: |
| 1401 | case SVt_INVLIST: |
| 1402 | case SVt_REGEXP: |
| 1403 | case SVt_PVMG: |
| 1404 | case SVt_PVNV: |
| 1405 | case SVt_PV: |
| 1406 | |
| 1407 | assert(new_type_details->body_size); |
| 1408 | /* We always allocated the full length item with PURIFY. To do this |
| 1409 | we fake things so that arena is false for all 16 types.. */ |
| 1410 | if(new_type_details->arena) { |
| 1411 | /* This points to the start of the allocated area. */ |
| 1412 | new_body_inline(new_body, new_type); |
| 1413 | Zero(new_body, new_type_details->body_size, char); |
| 1414 | new_body = ((char *)new_body) - new_type_details->offset; |
| 1415 | } else { |
| 1416 | new_body = new_NOARENAZ(new_type_details); |
| 1417 | } |
| 1418 | SvANY(sv) = new_body; |
| 1419 | |
| 1420 | if (old_type_details->copy) { |
| 1421 | /* There is now the potential for an upgrade from something without |
| 1422 | an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */ |
| 1423 | int offset = old_type_details->offset; |
| 1424 | int length = old_type_details->copy; |
| 1425 | |
| 1426 | if (new_type_details->offset > old_type_details->offset) { |
| 1427 | const int difference |
| 1428 | = new_type_details->offset - old_type_details->offset; |
| 1429 | offset += difference; |
| 1430 | length -= difference; |
| 1431 | } |
| 1432 | assert (length >= 0); |
| 1433 | |
| 1434 | Copy((char *)old_body + offset, (char *)new_body + offset, length, |
| 1435 | char); |
| 1436 | } |
| 1437 | |
| 1438 | #ifndef NV_ZERO_IS_ALLBITS_ZERO |
| 1439 | /* If NV 0.0 is stores as all bits 0 then Zero() already creates a |
| 1440 | * correct 0.0 for us. Otherwise, if the old body didn't have an |
| 1441 | * NV slot, but the new one does, then we need to initialise the |
| 1442 | * freshly created NV slot with whatever the correct bit pattern is |
| 1443 | * for 0.0 */ |
| 1444 | if (old_type_details->zero_nv && !new_type_details->zero_nv |
| 1445 | && !isGV_with_GP(sv)) |
| 1446 | SvNV_set(sv, 0); |
| 1447 | #endif |
| 1448 | |
| 1449 | if (UNLIKELY(new_type == SVt_PVIO)) { |
| 1450 | IO * const io = MUTABLE_IO(sv); |
| 1451 | GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV); |
| 1452 | |
| 1453 | SvOBJECT_on(io); |
| 1454 | /* Clear the stashcache because a new IO could overrule a package |
| 1455 | name */ |
| 1456 | DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n")); |
| 1457 | hv_clear(PL_stashcache); |
| 1458 | |
| 1459 | SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv)))); |
| 1460 | IoPAGE_LEN(sv) = 60; |
| 1461 | } |
| 1462 | if (UNLIKELY(new_type == SVt_REGEXP)) |
| 1463 | sv->sv_u.svu_rx = (regexp *)new_body; |
| 1464 | else if (old_type < SVt_PV) { |
| 1465 | /* referant will be NULL unless the old type was SVt_IV emulating |
| 1466 | SVt_RV */ |
| 1467 | sv->sv_u.svu_rv = referant; |
| 1468 | } |
| 1469 | break; |
| 1470 | default: |
| 1471 | Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu", |
| 1472 | (unsigned long)new_type); |
| 1473 | } |
| 1474 | |
| 1475 | /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV, |
| 1476 | and sometimes SVt_NV */ |
| 1477 | if (old_type_details->body_size) { |
| 1478 | #ifdef PURIFY |
| 1479 | safefree(old_body); |
| 1480 | #else |
| 1481 | /* Note that there is an assumption that all bodies of types that |
| 1482 | can be upgraded came from arenas. Only the more complex non- |
| 1483 | upgradable types are allowed to be directly malloc()ed. */ |
| 1484 | assert(old_type_details->arena); |
| 1485 | del_body((void*)((char*)old_body + old_type_details->offset), |
| 1486 | &PL_body_roots[old_type]); |
| 1487 | #endif |
| 1488 | } |
| 1489 | } |
| 1490 | |
| 1491 | /* |
| 1492 | =for apidoc sv_backoff |
| 1493 | |
| 1494 | Remove any string offset. You should normally use the C<SvOOK_off> macro |
| 1495 | wrapper instead. |
| 1496 | |
| 1497 | =cut |
| 1498 | */ |
| 1499 | |
| 1500 | /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS |
| 1501 | prior to 5.23.4 this function always returned 0 |
| 1502 | */ |
| 1503 | |
| 1504 | void |
| 1505 | Perl_sv_backoff(SV *const sv) |
| 1506 | { |
| 1507 | STRLEN delta; |
| 1508 | const char * const s = SvPVX_const(sv); |
| 1509 | |
| 1510 | PERL_ARGS_ASSERT_SV_BACKOFF; |
| 1511 | |
| 1512 | assert(SvOOK(sv)); |
| 1513 | assert(SvTYPE(sv) != SVt_PVHV); |
| 1514 | assert(SvTYPE(sv) != SVt_PVAV); |
| 1515 | |
| 1516 | SvOOK_offset(sv, delta); |
| 1517 | |
| 1518 | SvLEN_set(sv, SvLEN(sv) + delta); |
| 1519 | SvPV_set(sv, SvPVX(sv) - delta); |
| 1520 | SvFLAGS(sv) &= ~SVf_OOK; |
| 1521 | Move(s, SvPVX(sv), SvCUR(sv)+1, char); |
| 1522 | return; |
| 1523 | } |
| 1524 | |
| 1525 | /* |
| 1526 | =for apidoc sv_grow |
| 1527 | |
| 1528 | Expands the character buffer in the SV. If necessary, uses C<sv_unref> and |
| 1529 | upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer. |
| 1530 | Use the C<SvGROW> wrapper instead. |
| 1531 | |
| 1532 | =cut |
| 1533 | */ |
| 1534 | |
| 1535 | static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags); |
| 1536 | |
| 1537 | char * |
| 1538 | Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen) |
| 1539 | { |
| 1540 | char *s; |
| 1541 | |
| 1542 | PERL_ARGS_ASSERT_SV_GROW; |
| 1543 | |
| 1544 | if (SvROK(sv)) |
| 1545 | sv_unref(sv); |
| 1546 | if (SvTYPE(sv) < SVt_PV) { |
| 1547 | sv_upgrade(sv, SVt_PV); |
| 1548 | s = SvPVX_mutable(sv); |
| 1549 | } |
| 1550 | else if (SvOOK(sv)) { /* pv is offset? */ |
| 1551 | sv_backoff(sv); |
| 1552 | s = SvPVX_mutable(sv); |
| 1553 | if (newlen > SvLEN(sv)) |
| 1554 | newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */ |
| 1555 | } |
| 1556 | else |
| 1557 | { |
| 1558 | if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0); |
| 1559 | s = SvPVX_mutable(sv); |
| 1560 | } |
| 1561 | |
| 1562 | #ifdef PERL_COPY_ON_WRITE |
| 1563 | /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare) |
| 1564 | * to store the COW count. So in general, allocate one more byte than |
| 1565 | * asked for, to make it likely this byte is always spare: and thus |
| 1566 | * make more strings COW-able. |
| 1567 | * If the new size is a big power of two, don't bother: we assume the |
| 1568 | * caller wanted a nice 2^N sized block and will be annoyed at getting |
| 1569 | * 2^N+1. |
| 1570 | * Only increment if the allocation isn't MEM_SIZE_MAX, |
| 1571 | * otherwise it will wrap to 0. |
| 1572 | */ |
| 1573 | if (newlen & 0xff && newlen != MEM_SIZE_MAX) |
| 1574 | newlen++; |
| 1575 | #endif |
| 1576 | |
| 1577 | #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size) |
| 1578 | #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC |
| 1579 | #endif |
| 1580 | |
| 1581 | if (newlen > SvLEN(sv)) { /* need more room? */ |
| 1582 | STRLEN minlen = SvCUR(sv); |
| 1583 | minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10; |
| 1584 | if (newlen < minlen) |
| 1585 | newlen = minlen; |
| 1586 | #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC |
| 1587 | |
| 1588 | /* Don't round up on the first allocation, as odds are pretty good that |
| 1589 | * the initial request is accurate as to what is really needed */ |
| 1590 | if (SvLEN(sv)) { |
| 1591 | STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen); |
| 1592 | if (rounded > newlen) |
| 1593 | newlen = rounded; |
| 1594 | } |
| 1595 | #endif |
| 1596 | if (SvLEN(sv) && s) { |
| 1597 | s = (char*)saferealloc(s, newlen); |
| 1598 | } |
| 1599 | else { |
| 1600 | s = (char*)safemalloc(newlen); |
| 1601 | if (SvPVX_const(sv) && SvCUR(sv)) { |
| 1602 | Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char); |
| 1603 | } |
| 1604 | } |
| 1605 | SvPV_set(sv, s); |
| 1606 | #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC |
| 1607 | /* Do this here, do it once, do it right, and then we will never get |
| 1608 | called back into sv_grow() unless there really is some growing |
| 1609 | needed. */ |
| 1610 | SvLEN_set(sv, Perl_safesysmalloc_size(s)); |
| 1611 | #else |
| 1612 | SvLEN_set(sv, newlen); |
| 1613 | #endif |
| 1614 | } |
| 1615 | return s; |
| 1616 | } |
| 1617 | |
| 1618 | /* |
| 1619 | =for apidoc sv_setiv |
| 1620 | |
| 1621 | Copies an integer into the given SV, upgrading first if necessary. |
| 1622 | Does not handle 'set' magic. See also C<L</sv_setiv_mg>>. |
| 1623 | |
| 1624 | =cut |
| 1625 | */ |
| 1626 | |
| 1627 | void |
| 1628 | Perl_sv_setiv(pTHX_ SV *const sv, const IV i) |
| 1629 | { |
| 1630 | PERL_ARGS_ASSERT_SV_SETIV; |
| 1631 | |
| 1632 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 1633 | switch (SvTYPE(sv)) { |
| 1634 | case SVt_NULL: |
| 1635 | case SVt_NV: |
| 1636 | sv_upgrade(sv, SVt_IV); |
| 1637 | break; |
| 1638 | case SVt_PV: |
| 1639 | sv_upgrade(sv, SVt_PVIV); |
| 1640 | break; |
| 1641 | |
| 1642 | case SVt_PVGV: |
| 1643 | if (!isGV_with_GP(sv)) |
| 1644 | break; |
| 1645 | case SVt_PVAV: |
| 1646 | case SVt_PVHV: |
| 1647 | case SVt_PVCV: |
| 1648 | case SVt_PVFM: |
| 1649 | case SVt_PVIO: |
| 1650 | /* diag_listed_as: Can't coerce %s to %s in %s */ |
| 1651 | Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0), |
| 1652 | OP_DESC(PL_op)); |
| 1653 | break; |
| 1654 | default: NOOP; |
| 1655 | } |
| 1656 | (void)SvIOK_only(sv); /* validate number */ |
| 1657 | SvIV_set(sv, i); |
| 1658 | SvTAINT(sv); |
| 1659 | } |
| 1660 | |
| 1661 | /* |
| 1662 | =for apidoc sv_setiv_mg |
| 1663 | |
| 1664 | Like C<sv_setiv>, but also handles 'set' magic. |
| 1665 | |
| 1666 | =cut |
| 1667 | */ |
| 1668 | |
| 1669 | void |
| 1670 | Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i) |
| 1671 | { |
| 1672 | PERL_ARGS_ASSERT_SV_SETIV_MG; |
| 1673 | |
| 1674 | sv_setiv(sv,i); |
| 1675 | SvSETMAGIC(sv); |
| 1676 | } |
| 1677 | |
| 1678 | /* |
| 1679 | =for apidoc sv_setuv |
| 1680 | |
| 1681 | Copies an unsigned integer into the given SV, upgrading first if necessary. |
| 1682 | Does not handle 'set' magic. See also C<L</sv_setuv_mg>>. |
| 1683 | |
| 1684 | =cut |
| 1685 | */ |
| 1686 | |
| 1687 | void |
| 1688 | Perl_sv_setuv(pTHX_ SV *const sv, const UV u) |
| 1689 | { |
| 1690 | PERL_ARGS_ASSERT_SV_SETUV; |
| 1691 | |
| 1692 | /* With the if statement to ensure that integers are stored as IVs whenever |
| 1693 | possible: |
| 1694 | u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865 |
| 1695 | |
| 1696 | without |
| 1697 | u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865 |
| 1698 | |
| 1699 | If you wish to remove the following if statement, so that this routine |
| 1700 | (and its callers) always return UVs, please benchmark to see what the |
| 1701 | effect is. Modern CPUs may be different. Or may not :-) |
| 1702 | */ |
| 1703 | if (u <= (UV)IV_MAX) { |
| 1704 | sv_setiv(sv, (IV)u); |
| 1705 | return; |
| 1706 | } |
| 1707 | sv_setiv(sv, 0); |
| 1708 | SvIsUV_on(sv); |
| 1709 | SvUV_set(sv, u); |
| 1710 | } |
| 1711 | |
| 1712 | /* |
| 1713 | =for apidoc sv_setuv_mg |
| 1714 | |
| 1715 | Like C<sv_setuv>, but also handles 'set' magic. |
| 1716 | |
| 1717 | =cut |
| 1718 | */ |
| 1719 | |
| 1720 | void |
| 1721 | Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u) |
| 1722 | { |
| 1723 | PERL_ARGS_ASSERT_SV_SETUV_MG; |
| 1724 | |
| 1725 | sv_setuv(sv,u); |
| 1726 | SvSETMAGIC(sv); |
| 1727 | } |
| 1728 | |
| 1729 | /* |
| 1730 | =for apidoc sv_setnv |
| 1731 | |
| 1732 | Copies a double into the given SV, upgrading first if necessary. |
| 1733 | Does not handle 'set' magic. See also C<L</sv_setnv_mg>>. |
| 1734 | |
| 1735 | =cut |
| 1736 | */ |
| 1737 | |
| 1738 | void |
| 1739 | Perl_sv_setnv(pTHX_ SV *const sv, const NV num) |
| 1740 | { |
| 1741 | PERL_ARGS_ASSERT_SV_SETNV; |
| 1742 | |
| 1743 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 1744 | switch (SvTYPE(sv)) { |
| 1745 | case SVt_NULL: |
| 1746 | case SVt_IV: |
| 1747 | sv_upgrade(sv, SVt_NV); |
| 1748 | break; |
| 1749 | case SVt_PV: |
| 1750 | case SVt_PVIV: |
| 1751 | sv_upgrade(sv, SVt_PVNV); |
| 1752 | break; |
| 1753 | |
| 1754 | case SVt_PVGV: |
| 1755 | if (!isGV_with_GP(sv)) |
| 1756 | break; |
| 1757 | case SVt_PVAV: |
| 1758 | case SVt_PVHV: |
| 1759 | case SVt_PVCV: |
| 1760 | case SVt_PVFM: |
| 1761 | case SVt_PVIO: |
| 1762 | /* diag_listed_as: Can't coerce %s to %s in %s */ |
| 1763 | Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0), |
| 1764 | OP_DESC(PL_op)); |
| 1765 | break; |
| 1766 | default: NOOP; |
| 1767 | } |
| 1768 | SvNV_set(sv, num); |
| 1769 | (void)SvNOK_only(sv); /* validate number */ |
| 1770 | SvTAINT(sv); |
| 1771 | } |
| 1772 | |
| 1773 | /* |
| 1774 | =for apidoc sv_setnv_mg |
| 1775 | |
| 1776 | Like C<sv_setnv>, but also handles 'set' magic. |
| 1777 | |
| 1778 | =cut |
| 1779 | */ |
| 1780 | |
| 1781 | void |
| 1782 | Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num) |
| 1783 | { |
| 1784 | PERL_ARGS_ASSERT_SV_SETNV_MG; |
| 1785 | |
| 1786 | sv_setnv(sv,num); |
| 1787 | SvSETMAGIC(sv); |
| 1788 | } |
| 1789 | |
| 1790 | /* Return a cleaned-up, printable version of sv, for non-numeric, or |
| 1791 | * not incrementable warning display. |
| 1792 | * Originally part of S_not_a_number(). |
| 1793 | * The return value may be != tmpbuf. |
| 1794 | */ |
| 1795 | |
| 1796 | STATIC const char * |
| 1797 | S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) { |
| 1798 | const char *pv; |
| 1799 | |
| 1800 | PERL_ARGS_ASSERT_SV_DISPLAY; |
| 1801 | |
| 1802 | if (DO_UTF8(sv)) { |
| 1803 | SV *dsv = newSVpvs_flags("", SVs_TEMP); |
| 1804 | pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT); |
| 1805 | } else { |
| 1806 | char *d = tmpbuf; |
| 1807 | const char * const limit = tmpbuf + tmpbuf_size - 8; |
| 1808 | /* each *s can expand to 4 chars + "...\0", |
| 1809 | i.e. need room for 8 chars */ |
| 1810 | |
| 1811 | const char *s = SvPVX_const(sv); |
| 1812 | const char * const end = s + SvCUR(sv); |
| 1813 | for ( ; s < end && d < limit; s++ ) { |
| 1814 | int ch = *s & 0xFF; |
| 1815 | if (! isASCII(ch) && !isPRINT_LC(ch)) { |
| 1816 | *d++ = 'M'; |
| 1817 | *d++ = '-'; |
| 1818 | |
| 1819 | /* Map to ASCII "equivalent" of Latin1 */ |
| 1820 | ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127); |
| 1821 | } |
| 1822 | if (ch == '\n') { |
| 1823 | *d++ = '\\'; |
| 1824 | *d++ = 'n'; |
| 1825 | } |
| 1826 | else if (ch == '\r') { |
| 1827 | *d++ = '\\'; |
| 1828 | *d++ = 'r'; |
| 1829 | } |
| 1830 | else if (ch == '\f') { |
| 1831 | *d++ = '\\'; |
| 1832 | *d++ = 'f'; |
| 1833 | } |
| 1834 | else if (ch == '\\') { |
| 1835 | *d++ = '\\'; |
| 1836 | *d++ = '\\'; |
| 1837 | } |
| 1838 | else if (ch == '\0') { |
| 1839 | *d++ = '\\'; |
| 1840 | *d++ = '0'; |
| 1841 | } |
| 1842 | else if (isPRINT_LC(ch)) |
| 1843 | *d++ = ch; |
| 1844 | else { |
| 1845 | *d++ = '^'; |
| 1846 | *d++ = toCTRL(ch); |
| 1847 | } |
| 1848 | } |
| 1849 | if (s < end) { |
| 1850 | *d++ = '.'; |
| 1851 | *d++ = '.'; |
| 1852 | *d++ = '.'; |
| 1853 | } |
| 1854 | *d = '\0'; |
| 1855 | pv = tmpbuf; |
| 1856 | } |
| 1857 | |
| 1858 | return pv; |
| 1859 | } |
| 1860 | |
| 1861 | /* Print an "isn't numeric" warning, using a cleaned-up, |
| 1862 | * printable version of the offending string |
| 1863 | */ |
| 1864 | |
| 1865 | STATIC void |
| 1866 | S_not_a_number(pTHX_ SV *const sv) |
| 1867 | { |
| 1868 | char tmpbuf[64]; |
| 1869 | const char *pv; |
| 1870 | |
| 1871 | PERL_ARGS_ASSERT_NOT_A_NUMBER; |
| 1872 | |
| 1873 | pv = sv_display(sv, tmpbuf, sizeof(tmpbuf)); |
| 1874 | |
| 1875 | if (PL_op) |
| 1876 | Perl_warner(aTHX_ packWARN(WARN_NUMERIC), |
| 1877 | /* diag_listed_as: Argument "%s" isn't numeric%s */ |
| 1878 | "Argument \"%s\" isn't numeric in %s", pv, |
| 1879 | OP_DESC(PL_op)); |
| 1880 | else |
| 1881 | Perl_warner(aTHX_ packWARN(WARN_NUMERIC), |
| 1882 | /* diag_listed_as: Argument "%s" isn't numeric%s */ |
| 1883 | "Argument \"%s\" isn't numeric", pv); |
| 1884 | } |
| 1885 | |
| 1886 | STATIC void |
| 1887 | S_not_incrementable(pTHX_ SV *const sv) { |
| 1888 | char tmpbuf[64]; |
| 1889 | const char *pv; |
| 1890 | |
| 1891 | PERL_ARGS_ASSERT_NOT_INCREMENTABLE; |
| 1892 | |
| 1893 | pv = sv_display(sv, tmpbuf, sizeof(tmpbuf)); |
| 1894 | |
| 1895 | Perl_warner(aTHX_ packWARN(WARN_NUMERIC), |
| 1896 | "Argument \"%s\" treated as 0 in increment (++)", pv); |
| 1897 | } |
| 1898 | |
| 1899 | /* |
| 1900 | =for apidoc looks_like_number |
| 1901 | |
| 1902 | Test if the content of an SV looks like a number (or is a number). |
| 1903 | C<Inf> and C<Infinity> are treated as numbers (so will not issue a |
| 1904 | non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is |
| 1905 | ignored. |
| 1906 | |
| 1907 | =cut |
| 1908 | */ |
| 1909 | |
| 1910 | I32 |
| 1911 | Perl_looks_like_number(pTHX_ SV *const sv) |
| 1912 | { |
| 1913 | const char *sbegin; |
| 1914 | STRLEN len; |
| 1915 | int numtype; |
| 1916 | |
| 1917 | PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER; |
| 1918 | |
| 1919 | if (SvPOK(sv) || SvPOKp(sv)) { |
| 1920 | sbegin = SvPV_nomg_const(sv, len); |
| 1921 | } |
| 1922 | else |
| 1923 | return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK); |
| 1924 | numtype = grok_number(sbegin, len, NULL); |
| 1925 | return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype; |
| 1926 | } |
| 1927 | |
| 1928 | STATIC bool |
| 1929 | S_glob_2number(pTHX_ GV * const gv) |
| 1930 | { |
| 1931 | PERL_ARGS_ASSERT_GLOB_2NUMBER; |
| 1932 | |
| 1933 | /* We know that all GVs stringify to something that is not-a-number, |
| 1934 | so no need to test that. */ |
| 1935 | if (ckWARN(WARN_NUMERIC)) |
| 1936 | { |
| 1937 | SV *const buffer = sv_newmortal(); |
| 1938 | gv_efullname3(buffer, gv, "*"); |
| 1939 | not_a_number(buffer); |
| 1940 | } |
| 1941 | /* We just want something true to return, so that S_sv_2iuv_common |
| 1942 | can tail call us and return true. */ |
| 1943 | return TRUE; |
| 1944 | } |
| 1945 | |
| 1946 | /* Actually, ISO C leaves conversion of UV to IV undefined, but |
| 1947 | until proven guilty, assume that things are not that bad... */ |
| 1948 | |
| 1949 | /* |
| 1950 | NV_PRESERVES_UV: |
| 1951 | |
| 1952 | As 64 bit platforms often have an NV that doesn't preserve all bits of |
| 1953 | an IV (an assumption perl has been based on to date) it becomes necessary |
| 1954 | to remove the assumption that the NV always carries enough precision to |
| 1955 | recreate the IV whenever needed, and that the NV is the canonical form. |
| 1956 | Instead, IV/UV and NV need to be given equal rights. So as to not lose |
| 1957 | precision as a side effect of conversion (which would lead to insanity |
| 1958 | and the dragon(s) in t/op/numconvert.t getting very angry) the intent is |
| 1959 | 1) to distinguish between IV/UV/NV slots that have a valid conversion cached |
| 1960 | where precision was lost, and IV/UV/NV slots that have a valid conversion |
| 1961 | which has lost no precision |
| 1962 | 2) to ensure that if a numeric conversion to one form is requested that |
| 1963 | would lose precision, the precise conversion (or differently |
| 1964 | imprecise conversion) is also performed and cached, to prevent |
| 1965 | requests for different numeric formats on the same SV causing |
| 1966 | lossy conversion chains. (lossless conversion chains are perfectly |
| 1967 | acceptable (still)) |
| 1968 | |
| 1969 | |
| 1970 | flags are used: |
| 1971 | SvIOKp is true if the IV slot contains a valid value |
| 1972 | SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true) |
| 1973 | SvNOKp is true if the NV slot contains a valid value |
| 1974 | SvNOK is true only if the NV value is accurate |
| 1975 | |
| 1976 | so |
| 1977 | while converting from PV to NV, check to see if converting that NV to an |
| 1978 | IV(or UV) would lose accuracy over a direct conversion from PV to |
| 1979 | IV(or UV). If it would, cache both conversions, return NV, but mark |
| 1980 | SV as IOK NOKp (ie not NOK). |
| 1981 | |
| 1982 | While converting from PV to IV, check to see if converting that IV to an |
| 1983 | NV would lose accuracy over a direct conversion from PV to NV. If it |
| 1984 | would, cache both conversions, flag similarly. |
| 1985 | |
| 1986 | Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite |
| 1987 | correctly because if IV & NV were set NV *always* overruled. |
| 1988 | Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning |
| 1989 | changes - now IV and NV together means that the two are interchangeable: |
| 1990 | SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX; |
| 1991 | |
| 1992 | The benefit of this is that operations such as pp_add know that if |
| 1993 | SvIOK is true for both left and right operands, then integer addition |
| 1994 | can be used instead of floating point (for cases where the result won't |
| 1995 | overflow). Before, floating point was always used, which could lead to |
| 1996 | loss of precision compared with integer addition. |
| 1997 | |
| 1998 | * making IV and NV equal status should make maths accurate on 64 bit |
| 1999 | platforms |
| 2000 | * may speed up maths somewhat if pp_add and friends start to use |
| 2001 | integers when possible instead of fp. (Hopefully the overhead in |
| 2002 | looking for SvIOK and checking for overflow will not outweigh the |
| 2003 | fp to integer speedup) |
| 2004 | * will slow down integer operations (callers of SvIV) on "inaccurate" |
| 2005 | values, as the change from SvIOK to SvIOKp will cause a call into |
| 2006 | sv_2iv each time rather than a macro access direct to the IV slot |
| 2007 | * should speed up number->string conversion on integers as IV is |
| 2008 | favoured when IV and NV are equally accurate |
| 2009 | |
| 2010 | #################################################################### |
| 2011 | You had better be using SvIOK_notUV if you want an IV for arithmetic: |
| 2012 | SvIOK is true if (IV or UV), so you might be getting (IV)SvUV. |
| 2013 | On the other hand, SvUOK is true iff UV. |
| 2014 | #################################################################### |
| 2015 | |
| 2016 | Your mileage will vary depending your CPU's relative fp to integer |
| 2017 | performance ratio. |
| 2018 | */ |
| 2019 | |
| 2020 | #ifndef NV_PRESERVES_UV |
| 2021 | # define IS_NUMBER_UNDERFLOW_IV 1 |
| 2022 | # define IS_NUMBER_UNDERFLOW_UV 2 |
| 2023 | # define IS_NUMBER_IV_AND_UV 2 |
| 2024 | # define IS_NUMBER_OVERFLOW_IV 4 |
| 2025 | # define IS_NUMBER_OVERFLOW_UV 5 |
| 2026 | |
| 2027 | /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */ |
| 2028 | |
| 2029 | /* For sv_2nv these three cases are "SvNOK and don't bother casting" */ |
| 2030 | STATIC int |
| 2031 | S_sv_2iuv_non_preserve(pTHX_ SV *const sv |
| 2032 | # ifdef DEBUGGING |
| 2033 | , I32 numtype |
| 2034 | # endif |
| 2035 | ) |
| 2036 | { |
| 2037 | PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE; |
| 2038 | PERL_UNUSED_CONTEXT; |
| 2039 | |
| 2040 | 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)); |
| 2041 | if (SvNVX(sv) < (NV)IV_MIN) { |
| 2042 | (void)SvIOKp_on(sv); |
| 2043 | (void)SvNOK_on(sv); |
| 2044 | SvIV_set(sv, IV_MIN); |
| 2045 | return IS_NUMBER_UNDERFLOW_IV; |
| 2046 | } |
| 2047 | if (SvNVX(sv) > (NV)UV_MAX) { |
| 2048 | (void)SvIOKp_on(sv); |
| 2049 | (void)SvNOK_on(sv); |
| 2050 | SvIsUV_on(sv); |
| 2051 | SvUV_set(sv, UV_MAX); |
| 2052 | return IS_NUMBER_OVERFLOW_UV; |
| 2053 | } |
| 2054 | (void)SvIOKp_on(sv); |
| 2055 | (void)SvNOK_on(sv); |
| 2056 | /* Can't use strtol etc to convert this string. (See truth table in |
| 2057 | sv_2iv */ |
| 2058 | if (SvNVX(sv) <= (UV)IV_MAX) { |
| 2059 | SvIV_set(sv, I_V(SvNVX(sv))); |
| 2060 | if ((NV)(SvIVX(sv)) == SvNVX(sv)) { |
| 2061 | SvIOK_on(sv); /* Integer is precise. NOK, IOK */ |
| 2062 | } else { |
| 2063 | /* Integer is imprecise. NOK, IOKp */ |
| 2064 | } |
| 2065 | return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV; |
| 2066 | } |
| 2067 | SvIsUV_on(sv); |
| 2068 | SvUV_set(sv, U_V(SvNVX(sv))); |
| 2069 | if ((NV)(SvUVX(sv)) == SvNVX(sv)) { |
| 2070 | if (SvUVX(sv) == UV_MAX) { |
| 2071 | /* As we know that NVs don't preserve UVs, UV_MAX cannot |
| 2072 | possibly be preserved by NV. Hence, it must be overflow. |
| 2073 | NOK, IOKp */ |
| 2074 | return IS_NUMBER_OVERFLOW_UV; |
| 2075 | } |
| 2076 | SvIOK_on(sv); /* Integer is precise. NOK, UOK */ |
| 2077 | } else { |
| 2078 | /* Integer is imprecise. NOK, IOKp */ |
| 2079 | } |
| 2080 | return IS_NUMBER_OVERFLOW_IV; |
| 2081 | } |
| 2082 | #endif /* !NV_PRESERVES_UV*/ |
| 2083 | |
| 2084 | /* If numtype is infnan, set the NV of the sv accordingly. |
| 2085 | * If numtype is anything else, try setting the NV using Atof(PV). */ |
| 2086 | #ifdef USING_MSVC6 |
| 2087 | # pragma warning(push) |
| 2088 | # pragma warning(disable:4756;disable:4056) |
| 2089 | #endif |
| 2090 | static void |
| 2091 | S_sv_setnv(pTHX_ SV* sv, int numtype) |
| 2092 | { |
| 2093 | bool pok = cBOOL(SvPOK(sv)); |
| 2094 | bool nok = FALSE; |
| 2095 | if ((numtype & IS_NUMBER_INFINITY)) { |
| 2096 | SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF); |
| 2097 | nok = TRUE; |
| 2098 | } |
| 2099 | else if ((numtype & IS_NUMBER_NAN)) { |
| 2100 | SvNV_set(sv, NV_NAN); |
| 2101 | nok = TRUE; |
| 2102 | } |
| 2103 | else if (pok) { |
| 2104 | SvNV_set(sv, Atof(SvPVX_const(sv))); |
| 2105 | /* Purposefully no true nok here, since we don't want to blow |
| 2106 | * away the possible IOK/UV of an existing sv. */ |
| 2107 | } |
| 2108 | if (nok) { |
| 2109 | SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */ |
| 2110 | if (pok) |
| 2111 | SvPOK_on(sv); /* PV is okay, though. */ |
| 2112 | } |
| 2113 | } |
| 2114 | #ifdef USING_MSVC6 |
| 2115 | # pragma warning(pop) |
| 2116 | #endif |
| 2117 | |
| 2118 | STATIC bool |
| 2119 | S_sv_2iuv_common(pTHX_ SV *const sv) |
| 2120 | { |
| 2121 | PERL_ARGS_ASSERT_SV_2IUV_COMMON; |
| 2122 | |
| 2123 | if (SvNOKp(sv)) { |
| 2124 | /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv |
| 2125 | * without also getting a cached IV/UV from it at the same time |
| 2126 | * (ie PV->NV conversion should detect loss of accuracy and cache |
| 2127 | * IV or UV at same time to avoid this. */ |
| 2128 | /* IV-over-UV optimisation - choose to cache IV if possible */ |
| 2129 | |
| 2130 | if (SvTYPE(sv) == SVt_NV) |
| 2131 | sv_upgrade(sv, SVt_PVNV); |
| 2132 | |
| 2133 | (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */ |
| 2134 | /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost |
| 2135 | certainly cast into the IV range at IV_MAX, whereas the correct |
| 2136 | answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary |
| 2137 | cases go to UV */ |
| 2138 | #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan) |
| 2139 | if (Perl_isnan(SvNVX(sv))) { |
| 2140 | SvUV_set(sv, 0); |
| 2141 | SvIsUV_on(sv); |
| 2142 | return FALSE; |
| 2143 | } |
| 2144 | #endif |
| 2145 | if (SvNVX(sv) < (NV)IV_MAX + 0.5) { |
| 2146 | SvIV_set(sv, I_V(SvNVX(sv))); |
| 2147 | if (SvNVX(sv) == (NV) SvIVX(sv) |
| 2148 | #ifndef NV_PRESERVES_UV |
| 2149 | && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */ |
| 2150 | && (((UV)1 << NV_PRESERVES_UV_BITS) > |
| 2151 | (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv))) |
| 2152 | /* Don't flag it as "accurately an integer" if the number |
| 2153 | came from a (by definition imprecise) NV operation, and |
| 2154 | we're outside the range of NV integer precision */ |
| 2155 | #endif |
| 2156 | ) { |
| 2157 | if (SvNOK(sv)) |
| 2158 | SvIOK_on(sv); /* Can this go wrong with rounding? NWC */ |
| 2159 | else { |
| 2160 | /* scalar has trailing garbage, eg "42a" */ |
| 2161 | } |
| 2162 | DEBUG_c(PerlIO_printf(Perl_debug_log, |
| 2163 | "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n", |
| 2164 | PTR2UV(sv), |
| 2165 | SvNVX(sv), |
| 2166 | SvIVX(sv))); |
| 2167 | |
| 2168 | } else { |
| 2169 | /* IV not precise. No need to convert from PV, as NV |
| 2170 | conversion would already have cached IV if it detected |
| 2171 | that PV->IV would be better than PV->NV->IV |
| 2172 | flags already correct - don't set public IOK. */ |
| 2173 | DEBUG_c(PerlIO_printf(Perl_debug_log, |
| 2174 | "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n", |
| 2175 | PTR2UV(sv), |
| 2176 | SvNVX(sv), |
| 2177 | SvIVX(sv))); |
| 2178 | } |
| 2179 | /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN, |
| 2180 | but the cast (NV)IV_MIN rounds to a the value less (more |
| 2181 | negative) than IV_MIN which happens to be equal to SvNVX ?? |
| 2182 | Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and |
| 2183 | NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and |
| 2184 | (NV)UVX == NVX are both true, but the values differ. :-( |
| 2185 | Hopefully for 2s complement IV_MIN is something like |
| 2186 | 0x8000000000000000 which will be exact. NWC */ |
| 2187 | } |
| 2188 | else { |
| 2189 | SvUV_set(sv, U_V(SvNVX(sv))); |
| 2190 | if ( |
| 2191 | (SvNVX(sv) == (NV) SvUVX(sv)) |
| 2192 | #ifndef NV_PRESERVES_UV |
| 2193 | /* Make sure it's not 0xFFFFFFFFFFFFFFFF */ |
| 2194 | /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */ |
| 2195 | && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv)) |
| 2196 | /* Don't flag it as "accurately an integer" if the number |
| 2197 | came from a (by definition imprecise) NV operation, and |
| 2198 | we're outside the range of NV integer precision */ |
| 2199 | #endif |
| 2200 | && SvNOK(sv) |
| 2201 | ) |
| 2202 | SvIOK_on(sv); |
| 2203 | SvIsUV_on(sv); |
| 2204 | DEBUG_c(PerlIO_printf(Perl_debug_log, |
| 2205 | "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n", |
| 2206 | PTR2UV(sv), |
| 2207 | SvUVX(sv), |
| 2208 | SvUVX(sv))); |
| 2209 | } |
| 2210 | } |
| 2211 | else if (SvPOKp(sv)) { |
| 2212 | UV value; |
| 2213 | const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value); |
| 2214 | /* We want to avoid a possible problem when we cache an IV/ a UV which |
| 2215 | may be later translated to an NV, and the resulting NV is not |
| 2216 | the same as the direct translation of the initial string |
| 2217 | (eg 123.456 can shortcut to the IV 123 with atol(), but we must |
| 2218 | be careful to ensure that the value with the .456 is around if the |
| 2219 | NV value is requested in the future). |
| 2220 | |
| 2221 | This means that if we cache such an IV/a UV, we need to cache the |
| 2222 | NV as well. Moreover, we trade speed for space, and do not |
| 2223 | cache the NV if we are sure it's not needed. |
| 2224 | */ |
| 2225 | |
| 2226 | /* SVt_PVNV is one higher than SVt_PVIV, hence this order */ |
| 2227 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2228 | == IS_NUMBER_IN_UV) { |
| 2229 | /* It's definitely an integer, only upgrade to PVIV */ |
| 2230 | if (SvTYPE(sv) < SVt_PVIV) |
| 2231 | sv_upgrade(sv, SVt_PVIV); |
| 2232 | (void)SvIOK_on(sv); |
| 2233 | } else if (SvTYPE(sv) < SVt_PVNV) |
| 2234 | sv_upgrade(sv, SVt_PVNV); |
| 2235 | |
| 2236 | if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) { |
| 2237 | if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING))) |
| 2238 | not_a_number(sv); |
| 2239 | S_sv_setnv(aTHX_ sv, numtype); |
| 2240 | return FALSE; |
| 2241 | } |
| 2242 | |
| 2243 | /* If NVs preserve UVs then we only use the UV value if we know that |
| 2244 | we aren't going to call atof() below. If NVs don't preserve UVs |
| 2245 | then the value returned may have more precision than atof() will |
| 2246 | return, even though value isn't perfectly accurate. */ |
| 2247 | if ((numtype & (IS_NUMBER_IN_UV |
| 2248 | #ifdef NV_PRESERVES_UV |
| 2249 | | IS_NUMBER_NOT_INT |
| 2250 | #endif |
| 2251 | )) == IS_NUMBER_IN_UV) { |
| 2252 | /* This won't turn off the public IOK flag if it was set above */ |
| 2253 | (void)SvIOKp_on(sv); |
| 2254 | |
| 2255 | if (!(numtype & IS_NUMBER_NEG)) { |
| 2256 | /* positive */; |
| 2257 | if (value <= (UV)IV_MAX) { |
| 2258 | SvIV_set(sv, (IV)value); |
| 2259 | } else { |
| 2260 | /* it didn't overflow, and it was positive. */ |
| 2261 | SvUV_set(sv, value); |
| 2262 | SvIsUV_on(sv); |
| 2263 | } |
| 2264 | } else { |
| 2265 | /* 2s complement assumption */ |
| 2266 | if (value <= (UV)IV_MIN) { |
| 2267 | SvIV_set(sv, value == (UV)IV_MIN |
| 2268 | ? IV_MIN : -(IV)value); |
| 2269 | } else { |
| 2270 | /* Too negative for an IV. This is a double upgrade, but |
| 2271 | I'm assuming it will be rare. */ |
| 2272 | if (SvTYPE(sv) < SVt_PVNV) |
| 2273 | sv_upgrade(sv, SVt_PVNV); |
| 2274 | SvNOK_on(sv); |
| 2275 | SvIOK_off(sv); |
| 2276 | SvIOKp_on(sv); |
| 2277 | SvNV_set(sv, -(NV)value); |
| 2278 | SvIV_set(sv, IV_MIN); |
| 2279 | } |
| 2280 | } |
| 2281 | } |
| 2282 | /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we |
| 2283 | will be in the previous block to set the IV slot, and the next |
| 2284 | block to set the NV slot. So no else here. */ |
| 2285 | |
| 2286 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2287 | != IS_NUMBER_IN_UV) { |
| 2288 | /* It wasn't an (integer that doesn't overflow the UV). */ |
| 2289 | S_sv_setnv(aTHX_ sv, numtype); |
| 2290 | |
| 2291 | if (! numtype && ckWARN(WARN_NUMERIC)) |
| 2292 | not_a_number(sv); |
| 2293 | |
| 2294 | DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n", |
| 2295 | PTR2UV(sv), SvNVX(sv))); |
| 2296 | |
| 2297 | #ifdef NV_PRESERVES_UV |
| 2298 | (void)SvIOKp_on(sv); |
| 2299 | (void)SvNOK_on(sv); |
| 2300 | #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan) |
| 2301 | if (Perl_isnan(SvNVX(sv))) { |
| 2302 | SvUV_set(sv, 0); |
| 2303 | SvIsUV_on(sv); |
| 2304 | return FALSE; |
| 2305 | } |
| 2306 | #endif |
| 2307 | if (SvNVX(sv) < (NV)IV_MAX + 0.5) { |
| 2308 | SvIV_set(sv, I_V(SvNVX(sv))); |
| 2309 | if ((NV)(SvIVX(sv)) == SvNVX(sv)) { |
| 2310 | SvIOK_on(sv); |
| 2311 | } else { |
| 2312 | NOOP; /* Integer is imprecise. NOK, IOKp */ |
| 2313 | } |
| 2314 | /* UV will not work better than IV */ |
| 2315 | } else { |
| 2316 | if (SvNVX(sv) > (NV)UV_MAX) { |
| 2317 | SvIsUV_on(sv); |
| 2318 | /* Integer is inaccurate. NOK, IOKp, is UV */ |
| 2319 | SvUV_set(sv, UV_MAX); |
| 2320 | } else { |
| 2321 | SvUV_set(sv, U_V(SvNVX(sv))); |
| 2322 | /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs |
| 2323 | NV preservse UV so can do correct comparison. */ |
| 2324 | if ((NV)(SvUVX(sv)) == SvNVX(sv)) { |
| 2325 | SvIOK_on(sv); |
| 2326 | } else { |
| 2327 | NOOP; /* Integer is imprecise. NOK, IOKp, is UV */ |
| 2328 | } |
| 2329 | } |
| 2330 | SvIsUV_on(sv); |
| 2331 | } |
| 2332 | #else /* NV_PRESERVES_UV */ |
| 2333 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2334 | == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) { |
| 2335 | /* The IV/UV slot will have been set from value returned by |
| 2336 | grok_number above. The NV slot has just been set using |
| 2337 | Atof. */ |
| 2338 | SvNOK_on(sv); |
| 2339 | assert (SvIOKp(sv)); |
| 2340 | } else { |
| 2341 | if (((UV)1 << NV_PRESERVES_UV_BITS) > |
| 2342 | U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) { |
| 2343 | /* Small enough to preserve all bits. */ |
| 2344 | (void)SvIOKp_on(sv); |
| 2345 | SvNOK_on(sv); |
| 2346 | SvIV_set(sv, I_V(SvNVX(sv))); |
| 2347 | if ((NV)(SvIVX(sv)) == SvNVX(sv)) |
| 2348 | SvIOK_on(sv); |
| 2349 | /* Assumption: first non-preserved integer is < IV_MAX, |
| 2350 | this NV is in the preserved range, therefore: */ |
| 2351 | if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv)) |
| 2352 | < (UV)IV_MAX)) { |
| 2353 | Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX); |
| 2354 | } |
| 2355 | } else { |
| 2356 | /* IN_UV NOT_INT |
| 2357 | 0 0 already failed to read UV. |
| 2358 | 0 1 already failed to read UV. |
| 2359 | 1 0 you won't get here in this case. IV/UV |
| 2360 | slot set, public IOK, Atof() unneeded. |
| 2361 | 1 1 already read UV. |
| 2362 | so there's no point in sv_2iuv_non_preserve() attempting |
| 2363 | to use atol, strtol, strtoul etc. */ |
| 2364 | # ifdef DEBUGGING |
| 2365 | sv_2iuv_non_preserve (sv, numtype); |
| 2366 | # else |
| 2367 | sv_2iuv_non_preserve (sv); |
| 2368 | # endif |
| 2369 | } |
| 2370 | } |
| 2371 | #endif /* NV_PRESERVES_UV */ |
| 2372 | /* It might be more code efficient to go through the entire logic above |
| 2373 | and conditionally set with SvIOKp_on() rather than SvIOK(), but it |
| 2374 | gets complex and potentially buggy, so more programmer efficient |
| 2375 | to do it this way, by turning off the public flags: */ |
| 2376 | if (!numtype) |
| 2377 | SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK); |
| 2378 | } |
| 2379 | } |
| 2380 | else { |
| 2381 | if (isGV_with_GP(sv)) |
| 2382 | return glob_2number(MUTABLE_GV(sv)); |
| 2383 | |
| 2384 | if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) |
| 2385 | report_uninit(sv); |
| 2386 | if (SvTYPE(sv) < SVt_IV) |
| 2387 | /* Typically the caller expects that sv_any is not NULL now. */ |
| 2388 | sv_upgrade(sv, SVt_IV); |
| 2389 | /* Return 0 from the caller. */ |
| 2390 | return TRUE; |
| 2391 | } |
| 2392 | return FALSE; |
| 2393 | } |
| 2394 | |
| 2395 | /* |
| 2396 | =for apidoc sv_2iv_flags |
| 2397 | |
| 2398 | Return the integer value of an SV, doing any necessary string |
| 2399 | conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. |
| 2400 | Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros. |
| 2401 | |
| 2402 | =cut |
| 2403 | */ |
| 2404 | |
| 2405 | IV |
| 2406 | Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags) |
| 2407 | { |
| 2408 | PERL_ARGS_ASSERT_SV_2IV_FLAGS; |
| 2409 | |
| 2410 | assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV |
| 2411 | && SvTYPE(sv) != SVt_PVFM); |
| 2412 | |
| 2413 | if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) |
| 2414 | mg_get(sv); |
| 2415 | |
| 2416 | if (SvROK(sv)) { |
| 2417 | if (SvAMAGIC(sv)) { |
| 2418 | SV * tmpstr; |
| 2419 | if (flags & SV_SKIP_OVERLOAD) |
| 2420 | return 0; |
| 2421 | tmpstr = AMG_CALLunary(sv, numer_amg); |
| 2422 | if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { |
| 2423 | return SvIV(tmpstr); |
| 2424 | } |
| 2425 | } |
| 2426 | return PTR2IV(SvRV(sv)); |
| 2427 | } |
| 2428 | |
| 2429 | if (SvVALID(sv) || isREGEXP(sv)) { |
| 2430 | /* FBMs use the space for SvIVX and SvNVX for other purposes, and use |
| 2431 | the same flag bit as SVf_IVisUV, so must not let them cache IVs. |
| 2432 | In practice they are extremely unlikely to actually get anywhere |
| 2433 | accessible by user Perl code - the only way that I'm aware of is when |
| 2434 | a constant subroutine which is used as the second argument to index. |
| 2435 | |
| 2436 | Regexps have no SvIVX and SvNVX fields. |
| 2437 | */ |
| 2438 | assert(isREGEXP(sv) || SvPOKp(sv)); |
| 2439 | { |
| 2440 | UV value; |
| 2441 | const char * const ptr = |
| 2442 | isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv); |
| 2443 | const int numtype |
| 2444 | = grok_number(ptr, SvCUR(sv), &value); |
| 2445 | |
| 2446 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2447 | == IS_NUMBER_IN_UV) { |
| 2448 | /* It's definitely an integer */ |
| 2449 | if (numtype & IS_NUMBER_NEG) { |
| 2450 | if (value < (UV)IV_MIN) |
| 2451 | return -(IV)value; |
| 2452 | } else { |
| 2453 | if (value < (UV)IV_MAX) |
| 2454 | return (IV)value; |
| 2455 | } |
| 2456 | } |
| 2457 | |
| 2458 | /* Quite wrong but no good choices. */ |
| 2459 | if ((numtype & IS_NUMBER_INFINITY)) { |
| 2460 | return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX; |
| 2461 | } else if ((numtype & IS_NUMBER_NAN)) { |
| 2462 | return 0; /* So wrong. */ |
| 2463 | } |
| 2464 | |
| 2465 | if (!numtype) { |
| 2466 | if (ckWARN(WARN_NUMERIC)) |
| 2467 | not_a_number(sv); |
| 2468 | } |
| 2469 | return I_V(Atof(ptr)); |
| 2470 | } |
| 2471 | } |
| 2472 | |
| 2473 | if (SvTHINKFIRST(sv)) { |
| 2474 | if (SvREADONLY(sv) && !SvOK(sv)) { |
| 2475 | if (ckWARN(WARN_UNINITIALIZED)) |
| 2476 | report_uninit(sv); |
| 2477 | return 0; |
| 2478 | } |
| 2479 | } |
| 2480 | |
| 2481 | if (!SvIOKp(sv)) { |
| 2482 | if (S_sv_2iuv_common(aTHX_ sv)) |
| 2483 | return 0; |
| 2484 | } |
| 2485 | |
| 2486 | DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n", |
| 2487 | PTR2UV(sv),SvIVX(sv))); |
| 2488 | return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv); |
| 2489 | } |
| 2490 | |
| 2491 | /* |
| 2492 | =for apidoc sv_2uv_flags |
| 2493 | |
| 2494 | Return the unsigned integer value of an SV, doing any necessary string |
| 2495 | conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. |
| 2496 | Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros. |
| 2497 | |
| 2498 | =cut |
| 2499 | */ |
| 2500 | |
| 2501 | UV |
| 2502 | Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags) |
| 2503 | { |
| 2504 | PERL_ARGS_ASSERT_SV_2UV_FLAGS; |
| 2505 | |
| 2506 | if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) |
| 2507 | mg_get(sv); |
| 2508 | |
| 2509 | if (SvROK(sv)) { |
| 2510 | if (SvAMAGIC(sv)) { |
| 2511 | SV *tmpstr; |
| 2512 | if (flags & SV_SKIP_OVERLOAD) |
| 2513 | return 0; |
| 2514 | tmpstr = AMG_CALLunary(sv, numer_amg); |
| 2515 | if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { |
| 2516 | return SvUV(tmpstr); |
| 2517 | } |
| 2518 | } |
| 2519 | return PTR2UV(SvRV(sv)); |
| 2520 | } |
| 2521 | |
| 2522 | if (SvVALID(sv) || isREGEXP(sv)) { |
| 2523 | /* FBMs use the space for SvIVX and SvNVX for other purposes, and use |
| 2524 | the same flag bit as SVf_IVisUV, so must not let them cache IVs. |
| 2525 | Regexps have no SvIVX and SvNVX fields. */ |
| 2526 | assert(isREGEXP(sv) || SvPOKp(sv)); |
| 2527 | { |
| 2528 | UV value; |
| 2529 | const char * const ptr = |
| 2530 | isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv); |
| 2531 | const int numtype |
| 2532 | = grok_number(ptr, SvCUR(sv), &value); |
| 2533 | |
| 2534 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2535 | == IS_NUMBER_IN_UV) { |
| 2536 | /* It's definitely an integer */ |
| 2537 | if (!(numtype & IS_NUMBER_NEG)) |
| 2538 | return value; |
| 2539 | } |
| 2540 | |
| 2541 | /* Quite wrong but no good choices. */ |
| 2542 | if ((numtype & IS_NUMBER_INFINITY)) { |
| 2543 | return UV_MAX; /* So wrong. */ |
| 2544 | } else if ((numtype & IS_NUMBER_NAN)) { |
| 2545 | return 0; /* So wrong. */ |
| 2546 | } |
| 2547 | |
| 2548 | if (!numtype) { |
| 2549 | if (ckWARN(WARN_NUMERIC)) |
| 2550 | not_a_number(sv); |
| 2551 | } |
| 2552 | return U_V(Atof(ptr)); |
| 2553 | } |
| 2554 | } |
| 2555 | |
| 2556 | if (SvTHINKFIRST(sv)) { |
| 2557 | if (SvREADONLY(sv) && !SvOK(sv)) { |
| 2558 | if (ckWARN(WARN_UNINITIALIZED)) |
| 2559 | report_uninit(sv); |
| 2560 | return 0; |
| 2561 | } |
| 2562 | } |
| 2563 | |
| 2564 | if (!SvIOKp(sv)) { |
| 2565 | if (S_sv_2iuv_common(aTHX_ sv)) |
| 2566 | return 0; |
| 2567 | } |
| 2568 | |
| 2569 | DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n", |
| 2570 | PTR2UV(sv),SvUVX(sv))); |
| 2571 | return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv); |
| 2572 | } |
| 2573 | |
| 2574 | /* |
| 2575 | =for apidoc sv_2nv_flags |
| 2576 | |
| 2577 | Return the num value of an SV, doing any necessary string or integer |
| 2578 | conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. |
| 2579 | Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros. |
| 2580 | |
| 2581 | =cut |
| 2582 | */ |
| 2583 | |
| 2584 | NV |
| 2585 | Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags) |
| 2586 | { |
| 2587 | PERL_ARGS_ASSERT_SV_2NV_FLAGS; |
| 2588 | |
| 2589 | assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV |
| 2590 | && SvTYPE(sv) != SVt_PVFM); |
| 2591 | if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) { |
| 2592 | /* FBMs use the space for SvIVX and SvNVX for other purposes, and use |
| 2593 | the same flag bit as SVf_IVisUV, so must not let them cache NVs. |
| 2594 | Regexps have no SvIVX and SvNVX fields. */ |
| 2595 | const char *ptr; |
| 2596 | if (flags & SV_GMAGIC) |
| 2597 | mg_get(sv); |
| 2598 | if (SvNOKp(sv)) |
| 2599 | return SvNVX(sv); |
| 2600 | if (SvPOKp(sv) && !SvIOKp(sv)) { |
| 2601 | ptr = SvPVX_const(sv); |
| 2602 | grokpv: |
| 2603 | if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) && |
| 2604 | !grok_number(ptr, SvCUR(sv), NULL)) |
| 2605 | not_a_number(sv); |
| 2606 | return Atof(ptr); |
| 2607 | } |
| 2608 | if (SvIOKp(sv)) { |
| 2609 | if (SvIsUV(sv)) |
| 2610 | return (NV)SvUVX(sv); |
| 2611 | else |
| 2612 | return (NV)SvIVX(sv); |
| 2613 | } |
| 2614 | if (SvROK(sv)) { |
| 2615 | goto return_rok; |
| 2616 | } |
| 2617 | if (isREGEXP(sv)) { |
| 2618 | ptr = RX_WRAPPED((REGEXP *)sv); |
| 2619 | goto grokpv; |
| 2620 | } |
| 2621 | assert(SvTYPE(sv) >= SVt_PVMG); |
| 2622 | /* This falls through to the report_uninit near the end of the |
| 2623 | function. */ |
| 2624 | } else if (SvTHINKFIRST(sv)) { |
| 2625 | if (SvROK(sv)) { |
| 2626 | return_rok: |
| 2627 | if (SvAMAGIC(sv)) { |
| 2628 | SV *tmpstr; |
| 2629 | if (flags & SV_SKIP_OVERLOAD) |
| 2630 | return 0; |
| 2631 | tmpstr = AMG_CALLunary(sv, numer_amg); |
| 2632 | if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { |
| 2633 | return SvNV(tmpstr); |
| 2634 | } |
| 2635 | } |
| 2636 | return PTR2NV(SvRV(sv)); |
| 2637 | } |
| 2638 | if (SvREADONLY(sv) && !SvOK(sv)) { |
| 2639 | if (ckWARN(WARN_UNINITIALIZED)) |
| 2640 | report_uninit(sv); |
| 2641 | return 0.0; |
| 2642 | } |
| 2643 | } |
| 2644 | if (SvTYPE(sv) < SVt_NV) { |
| 2645 | /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */ |
| 2646 | sv_upgrade(sv, SVt_NV); |
| 2647 | DEBUG_c({ |
| 2648 | STORE_NUMERIC_LOCAL_SET_STANDARD(); |
| 2649 | PerlIO_printf(Perl_debug_log, |
| 2650 | "0x%"UVxf" num(%" NVgf ")\n", |
| 2651 | PTR2UV(sv), SvNVX(sv)); |
| 2652 | RESTORE_NUMERIC_LOCAL(); |
| 2653 | }); |
| 2654 | } |
| 2655 | else if (SvTYPE(sv) < SVt_PVNV) |
| 2656 | sv_upgrade(sv, SVt_PVNV); |
| 2657 | if (SvNOKp(sv)) { |
| 2658 | return SvNVX(sv); |
| 2659 | } |
| 2660 | if (SvIOKp(sv)) { |
| 2661 | SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv)); |
| 2662 | #ifdef NV_PRESERVES_UV |
| 2663 | if (SvIOK(sv)) |
| 2664 | SvNOK_on(sv); |
| 2665 | else |
| 2666 | SvNOKp_on(sv); |
| 2667 | #else |
| 2668 | /* Only set the public NV OK flag if this NV preserves the IV */ |
| 2669 | /* Check it's not 0xFFFFFFFFFFFFFFFF */ |
| 2670 | if (SvIOK(sv) && |
| 2671 | SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv)))) |
| 2672 | : (SvIVX(sv) == I_V(SvNVX(sv)))) |
| 2673 | SvNOK_on(sv); |
| 2674 | else |
| 2675 | SvNOKp_on(sv); |
| 2676 | #endif |
| 2677 | } |
| 2678 | else if (SvPOKp(sv)) { |
| 2679 | UV value; |
| 2680 | const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value); |
| 2681 | if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC)) |
| 2682 | not_a_number(sv); |
| 2683 | #ifdef NV_PRESERVES_UV |
| 2684 | if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) |
| 2685 | == IS_NUMBER_IN_UV) { |
| 2686 | /* It's definitely an integer */ |
| 2687 | SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value); |
| 2688 | } else { |
| 2689 | S_sv_setnv(aTHX_ sv, numtype); |
| 2690 | } |
| 2691 | if (numtype) |
| 2692 | SvNOK_on(sv); |
| 2693 | else |
| 2694 | SvNOKp_on(sv); |
| 2695 | #else |
| 2696 | SvNV_set(sv, Atof(SvPVX_const(sv))); |
| 2697 | /* Only set the public NV OK flag if this NV preserves the value in |
| 2698 | the PV at least as well as an IV/UV would. |
| 2699 | Not sure how to do this 100% reliably. */ |
| 2700 | /* if that shift count is out of range then Configure's test is |
| 2701 | wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS == |
| 2702 | UV_BITS */ |
| 2703 | if (((UV)1 << NV_PRESERVES_UV_BITS) > |
| 2704 | U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) { |
| 2705 | SvNOK_on(sv); /* Definitely small enough to preserve all bits */ |
| 2706 | } else if (!(numtype & IS_NUMBER_IN_UV)) { |
| 2707 | /* Can't use strtol etc to convert this string, so don't try. |
| 2708 | sv_2iv and sv_2uv will use the NV to convert, not the PV. */ |
| 2709 | SvNOK_on(sv); |
| 2710 | } else { |
| 2711 | /* value has been set. It may not be precise. */ |
| 2712 | if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) { |
| 2713 | /* 2s complement assumption for (UV)IV_MIN */ |
| 2714 | SvNOK_on(sv); /* Integer is too negative. */ |
| 2715 | } else { |
| 2716 | SvNOKp_on(sv); |
| 2717 | SvIOKp_on(sv); |
| 2718 | |
| 2719 | if (numtype & IS_NUMBER_NEG) { |
| 2720 | /* -IV_MIN is undefined, but we should never reach |
| 2721 | * this point with both IS_NUMBER_NEG and value == |
| 2722 | * (UV)IV_MIN */ |
| 2723 | assert(value != (UV)IV_MIN); |
| 2724 | SvIV_set(sv, -(IV)value); |
| 2725 | } else if (value <= (UV)IV_MAX) { |
| 2726 | SvIV_set(sv, (IV)value); |
| 2727 | } else { |
| 2728 | SvUV_set(sv, value); |
| 2729 | SvIsUV_on(sv); |
| 2730 | } |
| 2731 | |
| 2732 | if (numtype & IS_NUMBER_NOT_INT) { |
| 2733 | /* I believe that even if the original PV had decimals, |
| 2734 | they are lost beyond the limit of the FP precision. |
| 2735 | However, neither is canonical, so both only get p |
| 2736 | flags. NWC, 2000/11/25 */ |
| 2737 | /* Both already have p flags, so do nothing */ |
| 2738 | } else { |
| 2739 | const NV nv = SvNVX(sv); |
| 2740 | /* XXX should this spot have NAN_COMPARE_BROKEN, too? */ |
| 2741 | if (SvNVX(sv) < (NV)IV_MAX + 0.5) { |
| 2742 | if (SvIVX(sv) == I_V(nv)) { |
| 2743 | SvNOK_on(sv); |
| 2744 | } else { |
| 2745 | /* It had no "." so it must be integer. */ |
| 2746 | } |
| 2747 | SvIOK_on(sv); |
| 2748 | } else { |
| 2749 | /* between IV_MAX and NV(UV_MAX). |
| 2750 | Could be slightly > UV_MAX */ |
| 2751 | |
| 2752 | if (numtype & IS_NUMBER_NOT_INT) { |
| 2753 | /* UV and NV both imprecise. */ |
| 2754 | } else { |
| 2755 | const UV nv_as_uv = U_V(nv); |
| 2756 | |
| 2757 | if (value == nv_as_uv && SvUVX(sv) != UV_MAX) { |
| 2758 | SvNOK_on(sv); |
| 2759 | } |
| 2760 | SvIOK_on(sv); |
| 2761 | } |
| 2762 | } |
| 2763 | } |
| 2764 | } |
| 2765 | } |
| 2766 | /* It might be more code efficient to go through the entire logic above |
| 2767 | and conditionally set with SvNOKp_on() rather than SvNOK(), but it |
| 2768 | gets complex and potentially buggy, so more programmer efficient |
| 2769 | to do it this way, by turning off the public flags: */ |
| 2770 | if (!numtype) |
| 2771 | SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK); |
| 2772 | #endif /* NV_PRESERVES_UV */ |
| 2773 | } |
| 2774 | else { |
| 2775 | if (isGV_with_GP(sv)) { |
| 2776 | glob_2number(MUTABLE_GV(sv)); |
| 2777 | return 0.0; |
| 2778 | } |
| 2779 | |
| 2780 | if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) |
| 2781 | report_uninit(sv); |
| 2782 | assert (SvTYPE(sv) >= SVt_NV); |
| 2783 | /* Typically the caller expects that sv_any is not NULL now. */ |
| 2784 | /* XXX Ilya implies that this is a bug in callers that assume this |
| 2785 | and ideally should be fixed. */ |
| 2786 | return 0.0; |
| 2787 | } |
| 2788 | DEBUG_c({ |
| 2789 | STORE_NUMERIC_LOCAL_SET_STANDARD(); |
| 2790 | PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n", |
| 2791 | PTR2UV(sv), SvNVX(sv)); |
| 2792 | RESTORE_NUMERIC_LOCAL(); |
| 2793 | }); |
| 2794 | return SvNVX(sv); |
| 2795 | } |
| 2796 | |
| 2797 | /* |
| 2798 | =for apidoc sv_2num |
| 2799 | |
| 2800 | Return an SV with the numeric value of the source SV, doing any necessary |
| 2801 | reference or overload conversion. The caller is expected to have handled |
| 2802 | get-magic already. |
| 2803 | |
| 2804 | =cut |
| 2805 | */ |
| 2806 | |
| 2807 | SV * |
| 2808 | Perl_sv_2num(pTHX_ SV *const sv) |
| 2809 | { |
| 2810 | PERL_ARGS_ASSERT_SV_2NUM; |
| 2811 | |
| 2812 | if (!SvROK(sv)) |
| 2813 | return sv; |
| 2814 | if (SvAMAGIC(sv)) { |
| 2815 | SV * const tmpsv = AMG_CALLunary(sv, numer_amg); |
| 2816 | TAINT_IF(tmpsv && SvTAINTED(tmpsv)); |
| 2817 | if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) |
| 2818 | return sv_2num(tmpsv); |
| 2819 | } |
| 2820 | return sv_2mortal(newSVuv(PTR2UV(SvRV(sv)))); |
| 2821 | } |
| 2822 | |
| 2823 | /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or |
| 2824 | * UV as a string towards the end of buf, and return pointers to start and |
| 2825 | * end of it. |
| 2826 | * |
| 2827 | * We assume that buf is at least TYPE_CHARS(UV) long. |
| 2828 | */ |
| 2829 | |
| 2830 | static char * |
| 2831 | S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob) |
| 2832 | { |
| 2833 | char *ptr = buf + TYPE_CHARS(UV); |
| 2834 | char * const ebuf = ptr; |
| 2835 | int sign; |
| 2836 | |
| 2837 | PERL_ARGS_ASSERT_UIV_2BUF; |
| 2838 | |
| 2839 | if (is_uv) |
| 2840 | sign = 0; |
| 2841 | else if (iv >= 0) { |
| 2842 | uv = iv; |
| 2843 | sign = 0; |
| 2844 | } else { |
| 2845 | uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv); |
| 2846 | sign = 1; |
| 2847 | } |
| 2848 | do { |
| 2849 | *--ptr = '0' + (char)(uv % 10); |
| 2850 | } while (uv /= 10); |
| 2851 | if (sign) |
| 2852 | *--ptr = '-'; |
| 2853 | *peob = ebuf; |
| 2854 | return ptr; |
| 2855 | } |
| 2856 | |
| 2857 | /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an |
| 2858 | * infinity or a not-a-number, writes the appropriate strings to the |
| 2859 | * buffer, including a zero byte. On success returns the written length, |
| 2860 | * excluding the zero byte, on failure (not an infinity, not a nan) |
| 2861 | * returns zero, assert-fails on maxlen being too short. |
| 2862 | * |
| 2863 | * XXX for "Inf", "-Inf", and "NaN", we could have three read-only |
| 2864 | * shared string constants we point to, instead of generating a new |
| 2865 | * string for each instance. */ |
| 2866 | STATIC size_t |
| 2867 | S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) { |
| 2868 | char* s = buffer; |
| 2869 | assert(maxlen >= 4); |
| 2870 | if (Perl_isinf(nv)) { |
| 2871 | if (nv < 0) { |
| 2872 | if (maxlen < 5) /* "-Inf\0" */ |
| 2873 | return 0; |
| 2874 | *s++ = '-'; |
| 2875 | } else if (plus) { |
| 2876 | *s++ = '+'; |
| 2877 | } |
| 2878 | *s++ = 'I'; |
| 2879 | *s++ = 'n'; |
| 2880 | *s++ = 'f'; |
| 2881 | } |
| 2882 | else if (Perl_isnan(nv)) { |
| 2883 | *s++ = 'N'; |
| 2884 | *s++ = 'a'; |
| 2885 | *s++ = 'N'; |
| 2886 | /* XXX optionally output the payload mantissa bits as |
| 2887 | * "(unsigned)" (to match the nan("...") C99 function, |
| 2888 | * or maybe as "(0xhhh...)" would make more sense... |
| 2889 | * provide a format string so that the user can decide? |
| 2890 | * NOTE: would affect the maxlen and assert() logic.*/ |
| 2891 | } |
| 2892 | else { |
| 2893 | return 0; |
| 2894 | } |
| 2895 | assert((s == buffer + 3) || (s == buffer + 4)); |
| 2896 | *s++ = 0; |
| 2897 | return s - buffer - 1; /* -1: excluding the zero byte */ |
| 2898 | } |
| 2899 | |
| 2900 | /* |
| 2901 | =for apidoc sv_2pv_flags |
| 2902 | |
| 2903 | Returns a pointer to the string value of an SV, and sets C<*lp> to its length. |
| 2904 | If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a |
| 2905 | string if necessary. Normally invoked via the C<SvPV_flags> macro. |
| 2906 | C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too. |
| 2907 | |
| 2908 | =cut |
| 2909 | */ |
| 2910 | |
| 2911 | char * |
| 2912 | Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags) |
| 2913 | { |
| 2914 | char *s; |
| 2915 | |
| 2916 | PERL_ARGS_ASSERT_SV_2PV_FLAGS; |
| 2917 | |
| 2918 | assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV |
| 2919 | && SvTYPE(sv) != SVt_PVFM); |
| 2920 | if (SvGMAGICAL(sv) && (flags & SV_GMAGIC)) |
| 2921 | mg_get(sv); |
| 2922 | if (SvROK(sv)) { |
| 2923 | if (SvAMAGIC(sv)) { |
| 2924 | SV *tmpstr; |
| 2925 | if (flags & SV_SKIP_OVERLOAD) |
| 2926 | return NULL; |
| 2927 | tmpstr = AMG_CALLunary(sv, string_amg); |
| 2928 | TAINT_IF(tmpstr && SvTAINTED(tmpstr)); |
| 2929 | if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) { |
| 2930 | /* Unwrap this: */ |
| 2931 | /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr); |
| 2932 | */ |
| 2933 | |
| 2934 | char *pv; |
| 2935 | if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) { |
| 2936 | if (flags & SV_CONST_RETURN) { |
| 2937 | pv = (char *) SvPVX_const(tmpstr); |
| 2938 | } else { |
| 2939 | pv = (flags & SV_MUTABLE_RETURN) |
| 2940 | ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr); |
| 2941 | } |
| 2942 | if (lp) |
| 2943 | *lp = SvCUR(tmpstr); |
| 2944 | } else { |
| 2945 | pv = sv_2pv_flags(tmpstr, lp, flags); |
| 2946 | } |
| 2947 | if (SvUTF8(tmpstr)) |
| 2948 | SvUTF8_on(sv); |
| 2949 | else |
| 2950 | SvUTF8_off(sv); |
| 2951 | return pv; |
| 2952 | } |
| 2953 | } |
| 2954 | { |
| 2955 | STRLEN len; |
| 2956 | char *retval; |
| 2957 | char *buffer; |
| 2958 | SV *const referent = SvRV(sv); |
| 2959 | |
| 2960 | if (!referent) { |
| 2961 | len = 7; |
| 2962 | retval = buffer = savepvn("NULLREF", len); |
| 2963 | } else if (SvTYPE(referent) == SVt_REGEXP && |
| 2964 | (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) || |
| 2965 | amagic_is_enabled(string_amg))) { |
| 2966 | REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent); |
| 2967 | |
| 2968 | assert(re); |
| 2969 | |
| 2970 | /* If the regex is UTF-8 we want the containing scalar to |
| 2971 | have an UTF-8 flag too */ |
| 2972 | if (RX_UTF8(re)) |
| 2973 | SvUTF8_on(sv); |
| 2974 | else |
| 2975 | SvUTF8_off(sv); |
| 2976 | |
| 2977 | if (lp) |
| 2978 | *lp = RX_WRAPLEN(re); |
| 2979 | |
| 2980 | return RX_WRAPPED(re); |
| 2981 | } else { |
| 2982 | const char *const typestr = sv_reftype(referent, 0); |
| 2983 | const STRLEN typelen = strlen(typestr); |
| 2984 | UV addr = PTR2UV(referent); |
| 2985 | const char *stashname = NULL; |
| 2986 | STRLEN stashnamelen = 0; /* hush, gcc */ |
| 2987 | const char *buffer_end; |
| 2988 | |
| 2989 | if (SvOBJECT(referent)) { |
| 2990 | const HEK *const name = HvNAME_HEK(SvSTASH(referent)); |
| 2991 | |
| 2992 | if (name) { |
| 2993 | stashname = HEK_KEY(name); |
| 2994 | stashnamelen = HEK_LEN(name); |
| 2995 | |
| 2996 | if (HEK_UTF8(name)) { |
| 2997 | SvUTF8_on(sv); |
| 2998 | } else { |
| 2999 | SvUTF8_off(sv); |
| 3000 | } |
| 3001 | } else { |
| 3002 | stashname = "__ANON__"; |
| 3003 | stashnamelen = 8; |
| 3004 | } |
| 3005 | len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */ |
| 3006 | + 2 * sizeof(UV) + 2 /* )\0 */; |
| 3007 | } else { |
| 3008 | len = typelen + 3 /* (0x */ |
| 3009 | + 2 * sizeof(UV) + 2 /* )\0 */; |
| 3010 | } |
| 3011 | |
| 3012 | Newx(buffer, len, char); |
| 3013 | buffer_end = retval = buffer + len; |
| 3014 | |
| 3015 | /* Working backwards */ |
| 3016 | *--retval = '\0'; |
| 3017 | *--retval = ')'; |
| 3018 | do { |
| 3019 | *--retval = PL_hexdigit[addr & 15]; |
| 3020 | } while (addr >>= 4); |
| 3021 | *--retval = 'x'; |
| 3022 | *--retval = '0'; |
| 3023 | *--retval = '('; |
| 3024 | |
| 3025 | retval -= typelen; |
| 3026 | memcpy(retval, typestr, typelen); |
| 3027 | |
| 3028 | if (stashname) { |
| 3029 | *--retval = '='; |
| 3030 | retval -= stashnamelen; |
| 3031 | memcpy(retval, stashname, stashnamelen); |
| 3032 | } |
| 3033 | /* retval may not necessarily have reached the start of the |
| 3034 | buffer here. */ |
| 3035 | assert (retval >= buffer); |
| 3036 | |
| 3037 | len = buffer_end - retval - 1; /* -1 for that \0 */ |
| 3038 | } |
| 3039 | if (lp) |
| 3040 | *lp = len; |
| 3041 | SAVEFREEPV(buffer); |
| 3042 | return retval; |
| 3043 | } |
| 3044 | } |
| 3045 | |
| 3046 | if (SvPOKp(sv)) { |
| 3047 | if (lp) |
| 3048 | *lp = SvCUR(sv); |
| 3049 | if (flags & SV_MUTABLE_RETURN) |
| 3050 | return SvPVX_mutable(sv); |
| 3051 | if (flags & SV_CONST_RETURN) |
| 3052 | return (char *)SvPVX_const(sv); |
| 3053 | return SvPVX(sv); |
| 3054 | } |
| 3055 | |
| 3056 | if (SvIOK(sv)) { |
| 3057 | /* I'm assuming that if both IV and NV are equally valid then |
| 3058 | converting the IV is going to be more efficient */ |
| 3059 | const U32 isUIOK = SvIsUV(sv); |
| 3060 | char buf[TYPE_CHARS(UV)]; |
| 3061 | char *ebuf, *ptr; |
| 3062 | STRLEN len; |
| 3063 | |
| 3064 | if (SvTYPE(sv) < SVt_PVIV) |
| 3065 | sv_upgrade(sv, SVt_PVIV); |
| 3066 | ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf); |
| 3067 | len = ebuf - ptr; |
| 3068 | /* inlined from sv_setpvn */ |
| 3069 | s = SvGROW_mutable(sv, len + 1); |
| 3070 | Move(ptr, s, len, char); |
| 3071 | s += len; |
| 3072 | *s = '\0'; |
| 3073 | SvPOK_on(sv); |
| 3074 | } |
| 3075 | else if (SvNOK(sv)) { |
| 3076 | if (SvTYPE(sv) < SVt_PVNV) |
| 3077 | sv_upgrade(sv, SVt_PVNV); |
| 3078 | if (SvNVX(sv) == 0.0 |
| 3079 | #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan) |
| 3080 | && !Perl_isnan(SvNVX(sv)) |
| 3081 | #endif |
| 3082 | ) { |
| 3083 | s = SvGROW_mutable(sv, 2); |
| 3084 | *s++ = '0'; |
| 3085 | *s = '\0'; |
| 3086 | } else { |
| 3087 | STRLEN len; |
| 3088 | STRLEN size = 5; /* "-Inf\0" */ |
| 3089 | |
| 3090 | s = SvGROW_mutable(sv, size); |
| 3091 | len = S_infnan_2pv(SvNVX(sv), s, size, 0); |
| 3092 | if (len > 0) { |
| 3093 | s += len; |
| 3094 | SvPOK_on(sv); |
| 3095 | } |
| 3096 | else { |
| 3097 | /* some Xenix systems wipe out errno here */ |
| 3098 | dSAVE_ERRNO; |
| 3099 | |
| 3100 | size = |
| 3101 | 1 + /* sign */ |
| 3102 | 1 + /* "." */ |
| 3103 | NV_DIG + |
| 3104 | 1 + /* "e" */ |
| 3105 | 1 + /* sign */ |
| 3106 | 5 + /* exponent digits */ |
| 3107 | 1 + /* \0 */ |
| 3108 | 2; /* paranoia */ |
| 3109 | |
| 3110 | s = SvGROW_mutable(sv, size); |
| 3111 | #ifndef USE_LOCALE_NUMERIC |
| 3112 | SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG); |
| 3113 | |
| 3114 | SvPOK_on(sv); |
| 3115 | #else |
| 3116 | { |
| 3117 | bool local_radix; |
| 3118 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
| 3119 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 3120 | |
| 3121 | local_radix = |
| 3122 | PL_numeric_local && |
| 3123 | PL_numeric_radix_sv && |
| 3124 | SvUTF8(PL_numeric_radix_sv); |
| 3125 | if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) { |
| 3126 | size += SvLEN(PL_numeric_radix_sv) - 1; |
| 3127 | s = SvGROW_mutable(sv, size); |
| 3128 | } |
| 3129 | |
| 3130 | SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG); |
| 3131 | |
| 3132 | /* If the radix character is UTF-8, and actually is in the |
| 3133 | * output, turn on the UTF-8 flag for the scalar */ |
| 3134 | if (local_radix && |
| 3135 | instr(s, SvPVX_const(PL_numeric_radix_sv))) { |
| 3136 | SvUTF8_on(sv); |
| 3137 | } |
| 3138 | |
| 3139 | RESTORE_LC_NUMERIC(); |
| 3140 | } |
| 3141 | |
| 3142 | /* We don't call SvPOK_on(), because it may come to |
| 3143 | * pass that the locale changes so that the |
| 3144 | * stringification we just did is no longer correct. We |
| 3145 | * will have to re-stringify every time it is needed */ |
| 3146 | #endif |
| 3147 | RESTORE_ERRNO; |
| 3148 | } |
| 3149 | while (*s) s++; |
| 3150 | } |
| 3151 | } |
| 3152 | else if (isGV_with_GP(sv)) { |
| 3153 | GV *const gv = MUTABLE_GV(sv); |
| 3154 | SV *const buffer = sv_newmortal(); |
| 3155 | |
| 3156 | gv_efullname3(buffer, gv, "*"); |
| 3157 | |
| 3158 | assert(SvPOK(buffer)); |
| 3159 | if (SvUTF8(buffer)) |
| 3160 | SvUTF8_on(sv); |
| 3161 | if (lp) |
| 3162 | *lp = SvCUR(buffer); |
| 3163 | return SvPVX(buffer); |
| 3164 | } |
| 3165 | else if (isREGEXP(sv)) { |
| 3166 | if (lp) *lp = RX_WRAPLEN((REGEXP *)sv); |
| 3167 | return RX_WRAPPED((REGEXP *)sv); |
| 3168 | } |
| 3169 | else { |
| 3170 | if (lp) |
| 3171 | *lp = 0; |
| 3172 | if (flags & SV_UNDEF_RETURNS_NULL) |
| 3173 | return NULL; |
| 3174 | if (!PL_localizing && ckWARN(WARN_UNINITIALIZED)) |
| 3175 | report_uninit(sv); |
| 3176 | /* Typically the caller expects that sv_any is not NULL now. */ |
| 3177 | if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV) |
| 3178 | sv_upgrade(sv, SVt_PV); |
| 3179 | return (char *)""; |
| 3180 | } |
| 3181 | |
| 3182 | { |
| 3183 | const STRLEN len = s - SvPVX_const(sv); |
| 3184 | if (lp) |
| 3185 | *lp = len; |
| 3186 | SvCUR_set(sv, len); |
| 3187 | } |
| 3188 | DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n", |
| 3189 | PTR2UV(sv),SvPVX_const(sv))); |
| 3190 | if (flags & SV_CONST_RETURN) |
| 3191 | return (char *)SvPVX_const(sv); |
| 3192 | if (flags & SV_MUTABLE_RETURN) |
| 3193 | return SvPVX_mutable(sv); |
| 3194 | return SvPVX(sv); |
| 3195 | } |
| 3196 | |
| 3197 | /* |
| 3198 | =for apidoc sv_copypv |
| 3199 | |
| 3200 | Copies a stringified representation of the source SV into the |
| 3201 | destination SV. Automatically performs any necessary C<mg_get> and |
| 3202 | coercion of numeric values into strings. Guaranteed to preserve |
| 3203 | C<UTF8> flag even from overloaded objects. Similar in nature to |
| 3204 | C<sv_2pv[_flags]> but operates directly on an SV instead of just the |
| 3205 | string. Mostly uses C<sv_2pv_flags> to do its work, except when that |
| 3206 | would lose the UTF-8'ness of the PV. |
| 3207 | |
| 3208 | =for apidoc sv_copypv_nomg |
| 3209 | |
| 3210 | Like C<sv_copypv>, but doesn't invoke get magic first. |
| 3211 | |
| 3212 | =for apidoc sv_copypv_flags |
| 3213 | |
| 3214 | Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags |
| 3215 | has the C<SV_GMAGIC> bit set. |
| 3216 | |
| 3217 | =cut |
| 3218 | */ |
| 3219 | |
| 3220 | void |
| 3221 | Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags) |
| 3222 | { |
| 3223 | STRLEN len; |
| 3224 | const char *s; |
| 3225 | |
| 3226 | PERL_ARGS_ASSERT_SV_COPYPV_FLAGS; |
| 3227 | |
| 3228 | s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC)); |
| 3229 | sv_setpvn(dsv,s,len); |
| 3230 | if (SvUTF8(ssv)) |
| 3231 | SvUTF8_on(dsv); |
| 3232 | else |
| 3233 | SvUTF8_off(dsv); |
| 3234 | } |
| 3235 | |
| 3236 | /* |
| 3237 | =for apidoc sv_2pvbyte |
| 3238 | |
| 3239 | Return a pointer to the byte-encoded representation of the SV, and set C<*lp> |
| 3240 | to its length. May cause the SV to be downgraded from UTF-8 as a |
| 3241 | side-effect. |
| 3242 | |
| 3243 | Usually accessed via the C<SvPVbyte> macro. |
| 3244 | |
| 3245 | =cut |
| 3246 | */ |
| 3247 | |
| 3248 | char * |
| 3249 | Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp) |
| 3250 | { |
| 3251 | PERL_ARGS_ASSERT_SV_2PVBYTE; |
| 3252 | |
| 3253 | SvGETMAGIC(sv); |
| 3254 | if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv)) |
| 3255 | || isGV_with_GP(sv) || SvROK(sv)) { |
| 3256 | SV *sv2 = sv_newmortal(); |
| 3257 | sv_copypv_nomg(sv2,sv); |
| 3258 | sv = sv2; |
| 3259 | } |
| 3260 | sv_utf8_downgrade(sv,0); |
| 3261 | return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv); |
| 3262 | } |
| 3263 | |
| 3264 | /* |
| 3265 | =for apidoc sv_2pvutf8 |
| 3266 | |
| 3267 | Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp> |
| 3268 | to its length. May cause the SV to be upgraded to UTF-8 as a side-effect. |
| 3269 | |
| 3270 | Usually accessed via the C<SvPVutf8> macro. |
| 3271 | |
| 3272 | =cut |
| 3273 | */ |
| 3274 | |
| 3275 | char * |
| 3276 | Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp) |
| 3277 | { |
| 3278 | PERL_ARGS_ASSERT_SV_2PVUTF8; |
| 3279 | |
| 3280 | if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv)) |
| 3281 | || isGV_with_GP(sv) || SvROK(sv)) |
| 3282 | sv = sv_mortalcopy(sv); |
| 3283 | else |
| 3284 | SvGETMAGIC(sv); |
| 3285 | sv_utf8_upgrade_nomg(sv); |
| 3286 | return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv); |
| 3287 | } |
| 3288 | |
| 3289 | |
| 3290 | /* |
| 3291 | =for apidoc sv_2bool |
| 3292 | |
| 3293 | This macro is only used by C<sv_true()> or its macro equivalent, and only if |
| 3294 | the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. |
| 3295 | It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag. |
| 3296 | |
| 3297 | =for apidoc sv_2bool_flags |
| 3298 | |
| 3299 | This function is only used by C<sv_true()> and friends, and only if |
| 3300 | the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags |
| 3301 | contain C<SV_GMAGIC>, then it does an C<mg_get()> first. |
| 3302 | |
| 3303 | |
| 3304 | =cut |
| 3305 | */ |
| 3306 | |
| 3307 | bool |
| 3308 | Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags) |
| 3309 | { |
| 3310 | PERL_ARGS_ASSERT_SV_2BOOL_FLAGS; |
| 3311 | |
| 3312 | restart: |
| 3313 | if(flags & SV_GMAGIC) SvGETMAGIC(sv); |
| 3314 | |
| 3315 | if (!SvOK(sv)) |
| 3316 | return 0; |
| 3317 | if (SvROK(sv)) { |
| 3318 | if (SvAMAGIC(sv)) { |
| 3319 | SV * const tmpsv = AMG_CALLunary(sv, bool__amg); |
| 3320 | if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) { |
| 3321 | bool svb; |
| 3322 | sv = tmpsv; |
| 3323 | if(SvGMAGICAL(sv)) { |
| 3324 | flags = SV_GMAGIC; |
| 3325 | goto restart; /* call sv_2bool */ |
| 3326 | } |
| 3327 | /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */ |
| 3328 | else if(!SvOK(sv)) { |
| 3329 | svb = 0; |
| 3330 | } |
| 3331 | else if(SvPOK(sv)) { |
| 3332 | svb = SvPVXtrue(sv); |
| 3333 | } |
| 3334 | else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) { |
| 3335 | svb = (SvIOK(sv) && SvIVX(sv) != 0) |
| 3336 | || (SvNOK(sv) && SvNVX(sv) != 0.0); |
| 3337 | } |
| 3338 | else { |
| 3339 | flags = 0; |
| 3340 | goto restart; /* call sv_2bool_nomg */ |
| 3341 | } |
| 3342 | return cBOOL(svb); |
| 3343 | } |
| 3344 | } |
| 3345 | return SvRV(sv) != 0; |
| 3346 | } |
| 3347 | if (isREGEXP(sv)) |
| 3348 | return |
| 3349 | RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0'); |
| 3350 | return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0); |
| 3351 | } |
| 3352 | |
| 3353 | /* |
| 3354 | =for apidoc sv_utf8_upgrade |
| 3355 | |
| 3356 | Converts the PV of an SV to its UTF-8-encoded form. |
| 3357 | Forces the SV to string form if it is not already. |
| 3358 | Will C<mg_get> on C<sv> if appropriate. |
| 3359 | Always sets the C<SvUTF8> flag to avoid future validity checks even |
| 3360 | if the whole string is the same in UTF-8 as not. |
| 3361 | Returns the number of bytes in the converted string |
| 3362 | |
| 3363 | This is not a general purpose byte encoding to Unicode interface: |
| 3364 | use the Encode extension for that. |
| 3365 | |
| 3366 | =for apidoc sv_utf8_upgrade_nomg |
| 3367 | |
| 3368 | Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>. |
| 3369 | |
| 3370 | =for apidoc sv_utf8_upgrade_flags |
| 3371 | |
| 3372 | Converts the PV of an SV to its UTF-8-encoded form. |
| 3373 | Forces the SV to string form if it is not already. |
| 3374 | Always sets the SvUTF8 flag to avoid future validity checks even |
| 3375 | if all the bytes are invariant in UTF-8. |
| 3376 | If C<flags> has C<SV_GMAGIC> bit set, |
| 3377 | will C<mg_get> on C<sv> if appropriate, else not. |
| 3378 | |
| 3379 | If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV |
| 3380 | will expand when converted to UTF-8, and skips the extra work of checking for |
| 3381 | that. Typically this flag is used by a routine that has already parsed the |
| 3382 | string and found such characters, and passes this information on so that the |
| 3383 | work doesn't have to be repeated. |
| 3384 | |
| 3385 | Returns the number of bytes in the converted string. |
| 3386 | |
| 3387 | This is not a general purpose byte encoding to Unicode interface: |
| 3388 | use the Encode extension for that. |
| 3389 | |
| 3390 | =for apidoc sv_utf8_upgrade_flags_grow |
| 3391 | |
| 3392 | Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is |
| 3393 | the number of unused bytes the string of C<sv> is guaranteed to have free after |
| 3394 | it upon return. This allows the caller to reserve extra space that it intends |
| 3395 | to fill, to avoid extra grows. |
| 3396 | |
| 3397 | C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags> |
| 3398 | are implemented in terms of this function. |
| 3399 | |
| 3400 | Returns the number of bytes in the converted string (not including the spares). |
| 3401 | |
| 3402 | =cut |
| 3403 | |
| 3404 | (One might think that the calling routine could pass in the position of the |
| 3405 | first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't |
| 3406 | have to be found again. But that is not the case, because typically when the |
| 3407 | caller is likely to use this flag, it won't be calling this routine unless it |
| 3408 | finds something that won't fit into a byte. Otherwise it tries to not upgrade |
| 3409 | and just use bytes. But some things that do fit into a byte are variants in |
| 3410 | utf8, and the caller may not have been keeping track of these.) |
| 3411 | |
| 3412 | If the routine itself changes the string, it adds a trailing C<NUL>. Such a |
| 3413 | C<NUL> isn't guaranteed due to having other routines do the work in some input |
| 3414 | cases, or if the input is already flagged as being in utf8. |
| 3415 | |
| 3416 | The speed of this could perhaps be improved for many cases if someone wanted to |
| 3417 | write a fast function that counts the number of variant characters in a string, |
| 3418 | especially if it could return the position of the first one. |
| 3419 | |
| 3420 | */ |
| 3421 | |
| 3422 | STRLEN |
| 3423 | Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra) |
| 3424 | { |
| 3425 | PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW; |
| 3426 | |
| 3427 | if (sv == &PL_sv_undef) |
| 3428 | return 0; |
| 3429 | if (!SvPOK_nog(sv)) { |
| 3430 | STRLEN len = 0; |
| 3431 | if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) { |
| 3432 | (void) sv_2pv_flags(sv,&len, flags); |
| 3433 | if (SvUTF8(sv)) { |
| 3434 | if (extra) SvGROW(sv, SvCUR(sv) + extra); |
| 3435 | return len; |
| 3436 | } |
| 3437 | } else { |
| 3438 | (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC); |
| 3439 | } |
| 3440 | } |
| 3441 | |
| 3442 | if (SvUTF8(sv)) { |
| 3443 | if (extra) SvGROW(sv, SvCUR(sv) + extra); |
| 3444 | return SvCUR(sv); |
| 3445 | } |
| 3446 | |
| 3447 | if (SvIsCOW(sv)) { |
| 3448 | S_sv_uncow(aTHX_ sv, 0); |
| 3449 | } |
| 3450 | |
| 3451 | if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) { |
| 3452 | sv_recode_to_utf8(sv, _get_encoding()); |
| 3453 | if (extra) SvGROW(sv, SvCUR(sv) + extra); |
| 3454 | return SvCUR(sv); |
| 3455 | } |
| 3456 | |
| 3457 | if (SvCUR(sv) == 0) { |
| 3458 | if (extra) SvGROW(sv, extra); |
| 3459 | } else { /* Assume Latin-1/EBCDIC */ |
| 3460 | /* This function could be much more efficient if we |
| 3461 | * had a FLAG in SVs to signal if there are any variant |
| 3462 | * chars in the PV. Given that there isn't such a flag |
| 3463 | * make the loop as fast as possible (although there are certainly ways |
| 3464 | * to speed this up, eg. through vectorization) */ |
| 3465 | U8 * s = (U8 *) SvPVX_const(sv); |
| 3466 | U8 * e = (U8 *) SvEND(sv); |
| 3467 | U8 *t = s; |
| 3468 | STRLEN two_byte_count = 0; |
| 3469 | |
| 3470 | if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8; |
| 3471 | |
| 3472 | /* See if really will need to convert to utf8. We mustn't rely on our |
| 3473 | * incoming SV being well formed and having a trailing '\0', as certain |
| 3474 | * code in pp_formline can send us partially built SVs. */ |
| 3475 | |
| 3476 | while (t < e) { |
| 3477 | const U8 ch = *t++; |
| 3478 | if (NATIVE_BYTE_IS_INVARIANT(ch)) continue; |
| 3479 | |
| 3480 | t--; /* t already incremented; re-point to first variant */ |
| 3481 | two_byte_count = 1; |
| 3482 | goto must_be_utf8; |
| 3483 | } |
| 3484 | |
| 3485 | /* utf8 conversion not needed because all are invariants. Mark as |
| 3486 | * UTF-8 even if no variant - saves scanning loop */ |
| 3487 | SvUTF8_on(sv); |
| 3488 | if (extra) SvGROW(sv, SvCUR(sv) + extra); |
| 3489 | return SvCUR(sv); |
| 3490 | |
| 3491 | must_be_utf8: |
| 3492 | |
| 3493 | /* Here, the string should be converted to utf8, either because of an |
| 3494 | * input flag (two_byte_count = 0), or because a character that |
| 3495 | * requires 2 bytes was found (two_byte_count = 1). t points either to |
| 3496 | * the beginning of the string (if we didn't examine anything), or to |
| 3497 | * the first variant. In either case, everything from s to t - 1 will |
| 3498 | * occupy only 1 byte each on output. |
| 3499 | * |
| 3500 | * There are two main ways to convert. One is to create a new string |
| 3501 | * and go through the input starting from the beginning, appending each |
| 3502 | * converted value onto the new string as we go along. It's probably |
| 3503 | * best to allocate enough space in the string for the worst possible |
| 3504 | * case rather than possibly running out of space and having to |
| 3505 | * reallocate and then copy what we've done so far. Since everything |
| 3506 | * from s to t - 1 is invariant, the destination can be initialized |
| 3507 | * with these using a fast memory copy |
| 3508 | * |
| 3509 | * The other way is to figure out exactly how big the string should be |
| 3510 | * by parsing the entire input. Then you don't have to make it big |
| 3511 | * enough to handle the worst possible case, and more importantly, if |
| 3512 | * the string you already have is large enough, you don't have to |
| 3513 | * allocate a new string, you can copy the last character in the input |
| 3514 | * string to the final position(s) that will be occupied by the |
| 3515 | * converted string and go backwards, stopping at t, since everything |
| 3516 | * before that is invariant. |
| 3517 | * |
| 3518 | * There are advantages and disadvantages to each method. |
| 3519 | * |
| 3520 | * In the first method, we can allocate a new string, do the memory |
| 3521 | * copy from the s to t - 1, and then proceed through the rest of the |
| 3522 | * string byte-by-byte. |
| 3523 | * |
| 3524 | * In the second method, we proceed through the rest of the input |
| 3525 | * string just calculating how big the converted string will be. Then |
| 3526 | * there are two cases: |
| 3527 | * 1) if the string has enough extra space to handle the converted |
| 3528 | * value. We go backwards through the string, converting until we |
| 3529 | * get to the position we are at now, and then stop. If this |
| 3530 | * position is far enough along in the string, this method is |
| 3531 | * faster than the other method. If the memory copy were the same |
| 3532 | * speed as the byte-by-byte loop, that position would be about |
| 3533 | * half-way, as at the half-way mark, parsing to the end and back |
| 3534 | * is one complete string's parse, the same amount as starting |
| 3535 | * over and going all the way through. Actually, it would be |
| 3536 | * somewhat less than half-way, as it's faster to just count bytes |
| 3537 | * than to also copy, and we don't have the overhead of allocating |
| 3538 | * a new string, changing the scalar to use it, and freeing the |
| 3539 | * existing one. But if the memory copy is fast, the break-even |
| 3540 | * point is somewhere after half way. The counting loop could be |
| 3541 | * sped up by vectorization, etc, to move the break-even point |
| 3542 | * further towards the beginning. |
| 3543 | * 2) if the string doesn't have enough space to handle the converted |
| 3544 | * value. A new string will have to be allocated, and one might |
| 3545 | * as well, given that, start from the beginning doing the first |
| 3546 | * method. We've spent extra time parsing the string and in |
| 3547 | * exchange all we've gotten is that we know precisely how big to |
| 3548 | * make the new one. Perl is more optimized for time than space, |
| 3549 | * so this case is a loser. |
| 3550 | * So what I've decided to do is not use the 2nd method unless it is |
| 3551 | * guaranteed that a new string won't have to be allocated, assuming |
| 3552 | * the worst case. I also decided not to put any more conditions on it |
| 3553 | * than this, for now. It seems likely that, since the worst case is |
| 3554 | * twice as big as the unknown portion of the string (plus 1), we won't |
| 3555 | * be guaranteed enough space, causing us to go to the first method, |
| 3556 | * unless the string is short, or the first variant character is near |
| 3557 | * the end of it. In either of these cases, it seems best to use the |
| 3558 | * 2nd method. The only circumstance I can think of where this would |
| 3559 | * be really slower is if the string had once had much more data in it |
| 3560 | * than it does now, but there is still a substantial amount in it */ |
| 3561 | |
| 3562 | { |
| 3563 | STRLEN invariant_head = t - s; |
| 3564 | STRLEN size = invariant_head + (e - t) * 2 + 1 + extra; |
| 3565 | if (SvLEN(sv) < size) { |
| 3566 | |
| 3567 | /* Here, have decided to allocate a new string */ |
| 3568 | |
| 3569 | U8 *dst; |
| 3570 | U8 *d; |
| 3571 | |
| 3572 | Newx(dst, size, U8); |
| 3573 | |
| 3574 | /* If no known invariants at the beginning of the input string, |
| 3575 | * set so starts from there. Otherwise, can use memory copy to |
| 3576 | * get up to where we are now, and then start from here */ |
| 3577 | |
| 3578 | if (invariant_head == 0) { |
| 3579 | d = dst; |
| 3580 | } else { |
| 3581 | Copy(s, dst, invariant_head, char); |
| 3582 | d = dst + invariant_head; |
| 3583 | } |
| 3584 | |
| 3585 | while (t < e) { |
| 3586 | append_utf8_from_native_byte(*t, &d); |
| 3587 | t++; |
| 3588 | } |
| 3589 | *d = '\0'; |
| 3590 | SvPV_free(sv); /* No longer using pre-existing string */ |
| 3591 | SvPV_set(sv, (char*)dst); |
| 3592 | SvCUR_set(sv, d - dst); |
| 3593 | SvLEN_set(sv, size); |
| 3594 | } else { |
| 3595 | |
| 3596 | /* Here, have decided to get the exact size of the string. |
| 3597 | * Currently this happens only when we know that there is |
| 3598 | * guaranteed enough space to fit the converted string, so |
| 3599 | * don't have to worry about growing. If two_byte_count is 0, |
| 3600 | * then t points to the first byte of the string which hasn't |
| 3601 | * been examined yet. Otherwise two_byte_count is 1, and t |
| 3602 | * points to the first byte in the string that will expand to |
| 3603 | * two. Depending on this, start examining at t or 1 after t. |
| 3604 | * */ |
| 3605 | |
| 3606 | U8 *d = t + two_byte_count; |
| 3607 | |
| 3608 | |
| 3609 | /* Count up the remaining bytes that expand to two */ |
| 3610 | |
| 3611 | while (d < e) { |
| 3612 | const U8 chr = *d++; |
| 3613 | if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++; |
| 3614 | } |
| 3615 | |
| 3616 | /* The string will expand by just the number of bytes that |
| 3617 | * occupy two positions. But we are one afterwards because of |
| 3618 | * the increment just above. This is the place to put the |
| 3619 | * trailing NUL, and to set the length before we decrement */ |
| 3620 | |
| 3621 | d += two_byte_count; |
| 3622 | SvCUR_set(sv, d - s); |
| 3623 | *d-- = '\0'; |
| 3624 | |
| 3625 | |
| 3626 | /* Having decremented d, it points to the position to put the |
| 3627 | * very last byte of the expanded string. Go backwards through |
| 3628 | * the string, copying and expanding as we go, stopping when we |
| 3629 | * get to the part that is invariant the rest of the way down */ |
| 3630 | |
| 3631 | e--; |
| 3632 | while (e >= t) { |
| 3633 | if (NATIVE_BYTE_IS_INVARIANT(*e)) { |
| 3634 | *d-- = *e; |
| 3635 | } else { |
| 3636 | *d-- = UTF8_EIGHT_BIT_LO(*e); |
| 3637 | *d-- = UTF8_EIGHT_BIT_HI(*e); |
| 3638 | } |
| 3639 | e--; |
| 3640 | } |
| 3641 | } |
| 3642 | |
| 3643 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 3644 | /* Update pos. We do it at the end rather than during |
| 3645 | * the upgrade, to avoid slowing down the common case |
| 3646 | * (upgrade without pos). |
| 3647 | * pos can be stored as either bytes or characters. Since |
| 3648 | * this was previously a byte string we can just turn off |
| 3649 | * the bytes flag. */ |
| 3650 | MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); |
| 3651 | if (mg) { |
| 3652 | mg->mg_flags &= ~MGf_BYTES; |
| 3653 | } |
| 3654 | if ((mg = mg_find(sv, PERL_MAGIC_utf8))) |
| 3655 | magic_setutf8(sv,mg); /* clear UTF8 cache */ |
| 3656 | } |
| 3657 | } |
| 3658 | } |
| 3659 | |
| 3660 | /* Mark as UTF-8 even if no variant - saves scanning loop */ |
| 3661 | SvUTF8_on(sv); |
| 3662 | return SvCUR(sv); |
| 3663 | } |
| 3664 | |
| 3665 | /* |
| 3666 | =for apidoc sv_utf8_downgrade |
| 3667 | |
| 3668 | Attempts to convert the PV of an SV from characters to bytes. |
| 3669 | If the PV contains a character that cannot fit |
| 3670 | in a byte, this conversion will fail; |
| 3671 | in this case, either returns false or, if C<fail_ok> is not |
| 3672 | true, croaks. |
| 3673 | |
| 3674 | This is not a general purpose Unicode to byte encoding interface: |
| 3675 | use the C<Encode> extension for that. |
| 3676 | |
| 3677 | =cut |
| 3678 | */ |
| 3679 | |
| 3680 | bool |
| 3681 | Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok) |
| 3682 | { |
| 3683 | PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE; |
| 3684 | |
| 3685 | if (SvPOKp(sv) && SvUTF8(sv)) { |
| 3686 | if (SvCUR(sv)) { |
| 3687 | U8 *s; |
| 3688 | STRLEN len; |
| 3689 | int mg_flags = SV_GMAGIC; |
| 3690 | |
| 3691 | if (SvIsCOW(sv)) { |
| 3692 | S_sv_uncow(aTHX_ sv, 0); |
| 3693 | } |
| 3694 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 3695 | /* update pos */ |
| 3696 | MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); |
| 3697 | if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) { |
| 3698 | mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len, |
| 3699 | SV_GMAGIC|SV_CONST_RETURN); |
| 3700 | mg_flags = 0; /* sv_pos_b2u does get magic */ |
| 3701 | } |
| 3702 | if ((mg = mg_find(sv, PERL_MAGIC_utf8))) |
| 3703 | magic_setutf8(sv,mg); /* clear UTF8 cache */ |
| 3704 | |
| 3705 | } |
| 3706 | s = (U8 *) SvPV_flags(sv, len, mg_flags); |
| 3707 | |
| 3708 | if (!utf8_to_bytes(s, &len)) { |
| 3709 | if (fail_ok) |
| 3710 | return FALSE; |
| 3711 | else { |
| 3712 | if (PL_op) |
| 3713 | Perl_croak(aTHX_ "Wide character in %s", |
| 3714 | OP_DESC(PL_op)); |
| 3715 | else |
| 3716 | Perl_croak(aTHX_ "Wide character"); |
| 3717 | } |
| 3718 | } |
| 3719 | SvCUR_set(sv, len); |
| 3720 | } |
| 3721 | } |
| 3722 | SvUTF8_off(sv); |
| 3723 | return TRUE; |
| 3724 | } |
| 3725 | |
| 3726 | /* |
| 3727 | =for apidoc sv_utf8_encode |
| 3728 | |
| 3729 | Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8> |
| 3730 | flag off so that it looks like octets again. |
| 3731 | |
| 3732 | =cut |
| 3733 | */ |
| 3734 | |
| 3735 | void |
| 3736 | Perl_sv_utf8_encode(pTHX_ SV *const sv) |
| 3737 | { |
| 3738 | PERL_ARGS_ASSERT_SV_UTF8_ENCODE; |
| 3739 | |
| 3740 | if (SvREADONLY(sv)) { |
| 3741 | sv_force_normal_flags(sv, 0); |
| 3742 | } |
| 3743 | (void) sv_utf8_upgrade(sv); |
| 3744 | SvUTF8_off(sv); |
| 3745 | } |
| 3746 | |
| 3747 | /* |
| 3748 | =for apidoc sv_utf8_decode |
| 3749 | |
| 3750 | If the PV of the SV is an octet sequence in UTF-8 |
| 3751 | and contains a multiple-byte character, the C<SvUTF8> flag is turned on |
| 3752 | so that it looks like a character. If the PV contains only single-byte |
| 3753 | characters, the C<SvUTF8> flag stays off. |
| 3754 | Scans PV for validity and returns false if the PV is invalid UTF-8. |
| 3755 | |
| 3756 | =cut |
| 3757 | */ |
| 3758 | |
| 3759 | bool |
| 3760 | Perl_sv_utf8_decode(pTHX_ SV *const sv) |
| 3761 | { |
| 3762 | PERL_ARGS_ASSERT_SV_UTF8_DECODE; |
| 3763 | |
| 3764 | if (SvPOKp(sv)) { |
| 3765 | const U8 *start, *c; |
| 3766 | const U8 *e; |
| 3767 | |
| 3768 | /* The octets may have got themselves encoded - get them back as |
| 3769 | * bytes |
| 3770 | */ |
| 3771 | if (!sv_utf8_downgrade(sv, TRUE)) |
| 3772 | return FALSE; |
| 3773 | |
| 3774 | /* it is actually just a matter of turning the utf8 flag on, but |
| 3775 | * we want to make sure everything inside is valid utf8 first. |
| 3776 | */ |
| 3777 | c = start = (const U8 *) SvPVX_const(sv); |
| 3778 | if (!is_utf8_string(c, SvCUR(sv))) |
| 3779 | return FALSE; |
| 3780 | e = (const U8 *) SvEND(sv); |
| 3781 | while (c < e) { |
| 3782 | const U8 ch = *c++; |
| 3783 | if (!UTF8_IS_INVARIANT(ch)) { |
| 3784 | SvUTF8_on(sv); |
| 3785 | break; |
| 3786 | } |
| 3787 | } |
| 3788 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 3789 | /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC |
| 3790 | after this, clearing pos. Does anything on CPAN |
| 3791 | need this? */ |
| 3792 | /* adjust pos to the start of a UTF8 char sequence */ |
| 3793 | MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); |
| 3794 | if (mg) { |
| 3795 | I32 pos = mg->mg_len; |
| 3796 | if (pos > 0) { |
| 3797 | for (c = start + pos; c > start; c--) { |
| 3798 | if (UTF8_IS_START(*c)) |
| 3799 | break; |
| 3800 | } |
| 3801 | mg->mg_len = c - start; |
| 3802 | } |
| 3803 | } |
| 3804 | if ((mg = mg_find(sv, PERL_MAGIC_utf8))) |
| 3805 | magic_setutf8(sv,mg); /* clear UTF8 cache */ |
| 3806 | } |
| 3807 | } |
| 3808 | return TRUE; |
| 3809 | } |
| 3810 | |
| 3811 | /* |
| 3812 | =for apidoc sv_setsv |
| 3813 | |
| 3814 | Copies the contents of the source SV C<ssv> into the destination SV |
| 3815 | C<dsv>. The source SV may be destroyed if it is mortal, so don't use this |
| 3816 | function if the source SV needs to be reused. Does not handle 'set' magic on |
| 3817 | destination SV. Calls 'get' magic on source SV. Loosely speaking, it |
| 3818 | performs a copy-by-value, obliterating any previous content of the |
| 3819 | destination. |
| 3820 | |
| 3821 | You probably want to use one of the assortment of wrappers, such as |
| 3822 | C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and |
| 3823 | C<SvSetMagicSV_nosteal>. |
| 3824 | |
| 3825 | =for apidoc sv_setsv_flags |
| 3826 | |
| 3827 | Copies the contents of the source SV C<ssv> into the destination SV |
| 3828 | C<dsv>. The source SV may be destroyed if it is mortal, so don't use this |
| 3829 | function if the source SV needs to be reused. Does not handle 'set' magic. |
| 3830 | Loosely speaking, it performs a copy-by-value, obliterating any previous |
| 3831 | content of the destination. |
| 3832 | If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on |
| 3833 | C<ssv> if appropriate, else not. If the C<flags> |
| 3834 | parameter has the C<SV_NOSTEAL> bit set then the |
| 3835 | buffers of temps will not be stolen. C<sv_setsv> |
| 3836 | and C<sv_setsv_nomg> are implemented in terms of this function. |
| 3837 | |
| 3838 | You probably want to use one of the assortment of wrappers, such as |
| 3839 | C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and |
| 3840 | C<SvSetMagicSV_nosteal>. |
| 3841 | |
| 3842 | This is the primary function for copying scalars, and most other |
| 3843 | copy-ish functions and macros use this underneath. |
| 3844 | |
| 3845 | =cut |
| 3846 | */ |
| 3847 | |
| 3848 | static void |
| 3849 | S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype) |
| 3850 | { |
| 3851 | I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */ |
| 3852 | HV *old_stash = NULL; |
| 3853 | |
| 3854 | PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB; |
| 3855 | |
| 3856 | if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) { |
| 3857 | const char * const name = GvNAME(sstr); |
| 3858 | const STRLEN len = GvNAMELEN(sstr); |
| 3859 | { |
| 3860 | if (dtype >= SVt_PV) { |
| 3861 | SvPV_free(dstr); |
| 3862 | SvPV_set(dstr, 0); |
| 3863 | SvLEN_set(dstr, 0); |
| 3864 | SvCUR_set(dstr, 0); |
| 3865 | } |
| 3866 | SvUPGRADE(dstr, SVt_PVGV); |
| 3867 | (void)SvOK_off(dstr); |
| 3868 | isGV_with_GP_on(dstr); |
| 3869 | } |
| 3870 | GvSTASH(dstr) = GvSTASH(sstr); |
| 3871 | if (GvSTASH(dstr)) |
| 3872 | Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr); |
| 3873 | gv_name_set(MUTABLE_GV(dstr), name, len, |
| 3874 | GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 )); |
| 3875 | SvFAKE_on(dstr); /* can coerce to non-glob */ |
| 3876 | } |
| 3877 | |
| 3878 | if(GvGP(MUTABLE_GV(sstr))) { |
| 3879 | /* If source has method cache entry, clear it */ |
| 3880 | if(GvCVGEN(sstr)) { |
| 3881 | SvREFCNT_dec(GvCV(sstr)); |
| 3882 | GvCV_set(sstr, NULL); |
| 3883 | GvCVGEN(sstr) = 0; |
| 3884 | } |
| 3885 | /* If source has a real method, then a method is |
| 3886 | going to change */ |
| 3887 | else if( |
| 3888 | GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr)) |
| 3889 | ) { |
| 3890 | mro_changes = 1; |
| 3891 | } |
| 3892 | } |
| 3893 | |
| 3894 | /* If dest already had a real method, that's a change as well */ |
| 3895 | if( |
| 3896 | !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr) |
| 3897 | && GvSTASH(dstr) && HvENAME(GvSTASH(dstr)) |
| 3898 | ) { |
| 3899 | mro_changes = 1; |
| 3900 | } |
| 3901 | |
| 3902 | /* We don't need to check the name of the destination if it was not a |
| 3903 | glob to begin with. */ |
| 3904 | if(dtype == SVt_PVGV) { |
| 3905 | const char * const name = GvNAME((const GV *)dstr); |
| 3906 | if( |
| 3907 | strEQ(name,"ISA") |
| 3908 | /* The stash may have been detached from the symbol table, so |
| 3909 | check its name. */ |
| 3910 | && GvSTASH(dstr) && HvENAME(GvSTASH(dstr)) |
| 3911 | ) |
| 3912 | mro_changes = 2; |
| 3913 | else { |
| 3914 | const STRLEN len = GvNAMELEN(dstr); |
| 3915 | if ((len > 1 && name[len-2] == ':' && name[len-1] == ':') |
| 3916 | || (len == 1 && name[0] == ':')) { |
| 3917 | mro_changes = 3; |
| 3918 | |
| 3919 | /* Set aside the old stash, so we can reset isa caches on |
| 3920 | its subclasses. */ |
| 3921 | if((old_stash = GvHV(dstr))) |
| 3922 | /* Make sure we do not lose it early. */ |
| 3923 | SvREFCNT_inc_simple_void_NN( |
| 3924 | sv_2mortal((SV *)old_stash) |
| 3925 | ); |
| 3926 | } |
| 3927 | } |
| 3928 | |
| 3929 | SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr)); |
| 3930 | } |
| 3931 | |
| 3932 | /* freeing dstr's GP might free sstr (e.g. *x = $x), |
| 3933 | * so temporarily protect it */ |
| 3934 | ENTER; |
| 3935 | SAVEFREESV(SvREFCNT_inc_simple_NN(sstr)); |
| 3936 | gp_free(MUTABLE_GV(dstr)); |
| 3937 | GvINTRO_off(dstr); /* one-shot flag */ |
| 3938 | GvGP_set(dstr, gp_ref(GvGP(sstr))); |
| 3939 | LEAVE; |
| 3940 | |
| 3941 | if (SvTAINTED(sstr)) |
| 3942 | SvTAINT(dstr); |
| 3943 | if (GvIMPORTED(dstr) != GVf_IMPORTED |
| 3944 | && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) |
| 3945 | { |
| 3946 | GvIMPORTED_on(dstr); |
| 3947 | } |
| 3948 | GvMULTI_on(dstr); |
| 3949 | if(mro_changes == 2) { |
| 3950 | if (GvAV((const GV *)sstr)) { |
| 3951 | MAGIC *mg; |
| 3952 | SV * const sref = (SV *)GvAV((const GV *)dstr); |
| 3953 | if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) { |
| 3954 | if (SvTYPE(mg->mg_obj) != SVt_PVAV) { |
| 3955 | AV * const ary = newAV(); |
| 3956 | av_push(ary, mg->mg_obj); /* takes the refcount */ |
| 3957 | mg->mg_obj = (SV *)ary; |
| 3958 | } |
| 3959 | av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr)); |
| 3960 | } |
| 3961 | else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0); |
| 3962 | } |
| 3963 | mro_isa_changed_in(GvSTASH(dstr)); |
| 3964 | } |
| 3965 | else if(mro_changes == 3) { |
| 3966 | HV * const stash = GvHV(dstr); |
| 3967 | if(old_stash ? (HV *)HvENAME_get(old_stash) : stash) |
| 3968 | mro_package_moved( |
| 3969 | stash, old_stash, |
| 3970 | (GV *)dstr, 0 |
| 3971 | ); |
| 3972 | } |
| 3973 | else if(mro_changes) mro_method_changed_in(GvSTASH(dstr)); |
| 3974 | if (GvIO(dstr) && dtype == SVt_PVGV) { |
| 3975 | DEBUG_o(Perl_deb(aTHX_ |
| 3976 | "glob_assign_glob clearing PL_stashcache\n")); |
| 3977 | /* It's a cache. It will rebuild itself quite happily. |
| 3978 | It's a lot of effort to work out exactly which key (or keys) |
| 3979 | might be invalidated by the creation of the this file handle. |
| 3980 | */ |
| 3981 | hv_clear(PL_stashcache); |
| 3982 | } |
| 3983 | return; |
| 3984 | } |
| 3985 | |
| 3986 | void |
| 3987 | Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr) |
| 3988 | { |
| 3989 | SV * const sref = SvRV(sstr); |
| 3990 | SV *dref; |
| 3991 | const int intro = GvINTRO(dstr); |
| 3992 | SV **location; |
| 3993 | U8 import_flag = 0; |
| 3994 | const U32 stype = SvTYPE(sref); |
| 3995 | |
| 3996 | PERL_ARGS_ASSERT_GV_SETREF; |
| 3997 | |
| 3998 | if (intro) { |
| 3999 | GvINTRO_off(dstr); /* one-shot flag */ |
| 4000 | GvLINE(dstr) = CopLINE(PL_curcop); |
| 4001 | GvEGV(dstr) = MUTABLE_GV(dstr); |
| 4002 | } |
| 4003 | GvMULTI_on(dstr); |
| 4004 | switch (stype) { |
| 4005 | case SVt_PVCV: |
| 4006 | location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */ |
| 4007 | import_flag = GVf_IMPORTED_CV; |
| 4008 | goto common; |
| 4009 | case SVt_PVHV: |
| 4010 | location = (SV **) &GvHV(dstr); |
| 4011 | import_flag = GVf_IMPORTED_HV; |
| 4012 | goto common; |
| 4013 | case SVt_PVAV: |
| 4014 | location = (SV **) &GvAV(dstr); |
| 4015 | import_flag = GVf_IMPORTED_AV; |
| 4016 | goto common; |
| 4017 | case SVt_PVIO: |
| 4018 | location = (SV **) &GvIOp(dstr); |
| 4019 | goto common; |
| 4020 | case SVt_PVFM: |
| 4021 | location = (SV **) &GvFORM(dstr); |
| 4022 | goto common; |
| 4023 | default: |
| 4024 | location = &GvSV(dstr); |
| 4025 | import_flag = GVf_IMPORTED_SV; |
| 4026 | common: |
| 4027 | if (intro) { |
| 4028 | if (stype == SVt_PVCV) { |
| 4029 | /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/ |
| 4030 | if (GvCVGEN(dstr)) { |
| 4031 | SvREFCNT_dec(GvCV(dstr)); |
| 4032 | GvCV_set(dstr, NULL); |
| 4033 | GvCVGEN(dstr) = 0; /* Switch off cacheness. */ |
| 4034 | } |
| 4035 | } |
| 4036 | /* SAVEt_GVSLOT takes more room on the savestack and has more |
| 4037 | overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs |
| 4038 | leave_scope needs access to the GV so it can reset method |
| 4039 | caches. We must use SAVEt_GVSLOT whenever the type is |
| 4040 | SVt_PVCV, even if the stash is anonymous, as the stash may |
| 4041 | gain a name somehow before leave_scope. */ |
| 4042 | if (stype == SVt_PVCV) { |
| 4043 | /* There is no save_pushptrptrptr. Creating it for this |
| 4044 | one call site would be overkill. So inline the ss add |
| 4045 | routines here. */ |
| 4046 | dSS_ADD; |
| 4047 | SS_ADD_PTR(dstr); |
| 4048 | SS_ADD_PTR(location); |
| 4049 | SS_ADD_PTR(SvREFCNT_inc(*location)); |
| 4050 | SS_ADD_UV(SAVEt_GVSLOT); |
| 4051 | SS_ADD_END(4); |
| 4052 | } |
| 4053 | else SAVEGENERICSV(*location); |
| 4054 | } |
| 4055 | dref = *location; |
| 4056 | if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) { |
| 4057 | CV* const cv = MUTABLE_CV(*location); |
| 4058 | if (cv) { |
| 4059 | if (!GvCVGEN((const GV *)dstr) && |
| 4060 | (CvROOT(cv) || CvXSUB(cv)) && |
| 4061 | /* redundant check that avoids creating the extra SV |
| 4062 | most of the time: */ |
| 4063 | (CvCONST(cv) || ckWARN(WARN_REDEFINE))) |
| 4064 | { |
| 4065 | SV * const new_const_sv = |
| 4066 | CvCONST((const CV *)sref) |
| 4067 | ? cv_const_sv((const CV *)sref) |
| 4068 | : NULL; |
| 4069 | report_redefined_cv( |
| 4070 | sv_2mortal(Perl_newSVpvf(aTHX_ |
| 4071 | "%"HEKf"::%"HEKf, |
| 4072 | HEKfARG( |
| 4073 | HvNAME_HEK(GvSTASH((const GV *)dstr)) |
| 4074 | ), |
| 4075 | HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))) |
| 4076 | )), |
| 4077 | cv, |
| 4078 | CvCONST((const CV *)sref) ? &new_const_sv : NULL |
| 4079 | ); |
| 4080 | } |
| 4081 | if (!intro) |
| 4082 | cv_ckproto_len_flags(cv, (const GV *)dstr, |
| 4083 | SvPOK(sref) ? CvPROTO(sref) : NULL, |
| 4084 | SvPOK(sref) ? CvPROTOLEN(sref) : 0, |
| 4085 | SvPOK(sref) ? SvUTF8(sref) : 0); |
| 4086 | } |
| 4087 | GvCVGEN(dstr) = 0; /* Switch off cacheness. */ |
| 4088 | GvASSUMECV_on(dstr); |
| 4089 | if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */ |
| 4090 | if (intro && GvREFCNT(dstr) > 1) { |
| 4091 | /* temporary remove extra savestack's ref */ |
| 4092 | --GvREFCNT(dstr); |
| 4093 | gv_method_changed(dstr); |
| 4094 | ++GvREFCNT(dstr); |
| 4095 | } |
| 4096 | else gv_method_changed(dstr); |
| 4097 | } |
| 4098 | } |
| 4099 | *location = SvREFCNT_inc_simple_NN(sref); |
| 4100 | if (import_flag && !(GvFLAGS(dstr) & import_flag) |
| 4101 | && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) { |
| 4102 | GvFLAGS(dstr) |= import_flag; |
| 4103 | } |
| 4104 | |
| 4105 | if (stype == SVt_PVHV) { |
| 4106 | const char * const name = GvNAME((GV*)dstr); |
| 4107 | const STRLEN len = GvNAMELEN(dstr); |
| 4108 | if ( |
| 4109 | ( |
| 4110 | (len > 1 && name[len-2] == ':' && name[len-1] == ':') |
| 4111 | || (len == 1 && name[0] == ':') |
| 4112 | ) |
| 4113 | && (!dref || HvENAME_get(dref)) |
| 4114 | ) { |
| 4115 | mro_package_moved( |
| 4116 | (HV *)sref, (HV *)dref, |
| 4117 | (GV *)dstr, 0 |
| 4118 | ); |
| 4119 | } |
| 4120 | } |
| 4121 | else if ( |
| 4122 | stype == SVt_PVAV && sref != dref |
| 4123 | && strEQ(GvNAME((GV*)dstr), "ISA") |
| 4124 | /* The stash may have been detached from the symbol table, so |
| 4125 | check its name before doing anything. */ |
| 4126 | && GvSTASH(dstr) && HvENAME(GvSTASH(dstr)) |
| 4127 | ) { |
| 4128 | MAGIC *mg; |
| 4129 | MAGIC * const omg = dref && SvSMAGICAL(dref) |
| 4130 | ? mg_find(dref, PERL_MAGIC_isa) |
| 4131 | : NULL; |
| 4132 | if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) { |
| 4133 | if (SvTYPE(mg->mg_obj) != SVt_PVAV) { |
| 4134 | AV * const ary = newAV(); |
| 4135 | av_push(ary, mg->mg_obj); /* takes the refcount */ |
| 4136 | mg->mg_obj = (SV *)ary; |
| 4137 | } |
| 4138 | if (omg) { |
| 4139 | if (SvTYPE(omg->mg_obj) == SVt_PVAV) { |
| 4140 | SV **svp = AvARRAY((AV *)omg->mg_obj); |
| 4141 | I32 items = AvFILLp((AV *)omg->mg_obj) + 1; |
| 4142 | while (items--) |
| 4143 | av_push( |
| 4144 | (AV *)mg->mg_obj, |
| 4145 | SvREFCNT_inc_simple_NN(*svp++) |
| 4146 | ); |
| 4147 | } |
| 4148 | else |
| 4149 | av_push( |
| 4150 | (AV *)mg->mg_obj, |
| 4151 | SvREFCNT_inc_simple_NN(omg->mg_obj) |
| 4152 | ); |
| 4153 | } |
| 4154 | else |
| 4155 | av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr)); |
| 4156 | } |
| 4157 | else |
| 4158 | { |
| 4159 | SSize_t i; |
| 4160 | sv_magic( |
| 4161 | sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0 |
| 4162 | ); |
| 4163 | for (i = 0; i <= AvFILL(sref); ++i) { |
| 4164 | SV **elem = av_fetch ((AV*)sref, i, 0); |
| 4165 | if (elem) { |
| 4166 | sv_magic( |
| 4167 | *elem, sref, PERL_MAGIC_isaelem, NULL, i |
| 4168 | ); |
| 4169 | } |
| 4170 | } |
| 4171 | mg = mg_find(sref, PERL_MAGIC_isa); |
| 4172 | } |
| 4173 | /* Since the *ISA assignment could have affected more than |
| 4174 | one stash, don't call mro_isa_changed_in directly, but let |
| 4175 | magic_clearisa do it for us, as it already has the logic for |
| 4176 | dealing with globs vs arrays of globs. */ |
| 4177 | assert(mg); |
| 4178 | Perl_magic_clearisa(aTHX_ NULL, mg); |
| 4179 | } |
| 4180 | else if (stype == SVt_PVIO) { |
| 4181 | DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n")); |
| 4182 | /* It's a cache. It will rebuild itself quite happily. |
| 4183 | It's a lot of effort to work out exactly which key (or keys) |
| 4184 | might be invalidated by the creation of the this file handle. |
| 4185 | */ |
| 4186 | hv_clear(PL_stashcache); |
| 4187 | } |
| 4188 | break; |
| 4189 | } |
| 4190 | if (!intro) SvREFCNT_dec(dref); |
| 4191 | if (SvTAINTED(sstr)) |
| 4192 | SvTAINT(dstr); |
| 4193 | return; |
| 4194 | } |
| 4195 | |
| 4196 | |
| 4197 | |
| 4198 | |
| 4199 | #ifdef PERL_DEBUG_READONLY_COW |
| 4200 | # include <sys/mman.h> |
| 4201 | |
| 4202 | # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE |
| 4203 | # define PERL_MEMORY_DEBUG_HEADER_SIZE 0 |
| 4204 | # endif |
| 4205 | |
| 4206 | void |
| 4207 | Perl_sv_buf_to_ro(pTHX_ SV *sv) |
| 4208 | { |
| 4209 | struct perl_memory_debug_header * const header = |
| 4210 | (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE); |
| 4211 | const MEM_SIZE len = header->size; |
| 4212 | PERL_ARGS_ASSERT_SV_BUF_TO_RO; |
| 4213 | # ifdef PERL_TRACK_MEMPOOL |
| 4214 | if (!header->readonly) header->readonly = 1; |
| 4215 | # endif |
| 4216 | if (mprotect(header, len, PROT_READ)) |
| 4217 | Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d", |
| 4218 | header, len, errno); |
| 4219 | } |
| 4220 | |
| 4221 | static void |
| 4222 | S_sv_buf_to_rw(pTHX_ SV *sv) |
| 4223 | { |
| 4224 | struct perl_memory_debug_header * const header = |
| 4225 | (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE); |
| 4226 | const MEM_SIZE len = header->size; |
| 4227 | PERL_ARGS_ASSERT_SV_BUF_TO_RW; |
| 4228 | if (mprotect(header, len, PROT_READ|PROT_WRITE)) |
| 4229 | Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d", |
| 4230 | header, len, errno); |
| 4231 | # ifdef PERL_TRACK_MEMPOOL |
| 4232 | header->readonly = 0; |
| 4233 | # endif |
| 4234 | } |
| 4235 | |
| 4236 | #else |
| 4237 | # define sv_buf_to_ro(sv) NOOP |
| 4238 | # define sv_buf_to_rw(sv) NOOP |
| 4239 | #endif |
| 4240 | |
| 4241 | void |
| 4242 | Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags) |
| 4243 | { |
| 4244 | U32 sflags; |
| 4245 | int dtype; |
| 4246 | svtype stype; |
| 4247 | unsigned int both_type; |
| 4248 | |
| 4249 | PERL_ARGS_ASSERT_SV_SETSV_FLAGS; |
| 4250 | |
| 4251 | if (UNLIKELY( sstr == dstr )) |
| 4252 | return; |
| 4253 | |
| 4254 | if (UNLIKELY( !sstr )) |
| 4255 | sstr = &PL_sv_undef; |
| 4256 | |
| 4257 | stype = SvTYPE(sstr); |
| 4258 | dtype = SvTYPE(dstr); |
| 4259 | both_type = (stype | dtype); |
| 4260 | |
| 4261 | /* with these values, we can check that both SVs are NULL/IV (and not |
| 4262 | * freed) just by testing the or'ed types */ |
| 4263 | STATIC_ASSERT_STMT(SVt_NULL == 0); |
| 4264 | STATIC_ASSERT_STMT(SVt_IV == 1); |
| 4265 | if (both_type <= 1) { |
| 4266 | /* both src and dst are UNDEF/IV/RV, so we can do a lot of |
| 4267 | * special-casing */ |
| 4268 | U32 sflags; |
| 4269 | U32 new_dflags; |
| 4270 | |
| 4271 | /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */ |
| 4272 | if (SvREADONLY(dstr)) |
| 4273 | Perl_croak_no_modify(); |
| 4274 | if (SvROK(dstr)) |
| 4275 | sv_unref_flags(dstr, 0); |
| 4276 | |
| 4277 | assert(!SvGMAGICAL(sstr)); |
| 4278 | assert(!SvGMAGICAL(dstr)); |
| 4279 | |
| 4280 | sflags = SvFLAGS(sstr); |
| 4281 | if (sflags & (SVf_IOK|SVf_ROK)) { |
| 4282 | SET_SVANY_FOR_BODYLESS_IV(dstr); |
| 4283 | new_dflags = SVt_IV; |
| 4284 | |
| 4285 | if (sflags & SVf_ROK) { |
| 4286 | dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr)); |
| 4287 | new_dflags |= SVf_ROK; |
| 4288 | } |
| 4289 | else { |
| 4290 | /* both src and dst are <= SVt_IV, so sv_any points to the |
| 4291 | * head; so access the head directly |
| 4292 | */ |
| 4293 | assert( &(sstr->sv_u.svu_iv) |
| 4294 | == &(((XPVIV*) SvANY(sstr))->xiv_iv)); |
| 4295 | assert( &(dstr->sv_u.svu_iv) |
| 4296 | == &(((XPVIV*) SvANY(dstr))->xiv_iv)); |
| 4297 | dstr->sv_u.svu_iv = sstr->sv_u.svu_iv; |
| 4298 | new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV)); |
| 4299 | } |
| 4300 | } |
| 4301 | else { |
| 4302 | new_dflags = dtype; /* turn off everything except the type */ |
| 4303 | } |
| 4304 | SvFLAGS(dstr) = new_dflags; |
| 4305 | |
| 4306 | return; |
| 4307 | } |
| 4308 | |
| 4309 | if (UNLIKELY(both_type == SVTYPEMASK)) { |
| 4310 | if (SvIS_FREED(dstr)) { |
| 4311 | Perl_croak(aTHX_ "panic: attempt to copy value %" SVf |
| 4312 | " to a freed scalar %p", SVfARG(sstr), (void *)dstr); |
| 4313 | } |
| 4314 | if (SvIS_FREED(sstr)) { |
| 4315 | Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p", |
| 4316 | (void*)sstr, (void*)dstr); |
| 4317 | } |
| 4318 | } |
| 4319 | |
| 4320 | |
| 4321 | |
| 4322 | SV_CHECK_THINKFIRST_COW_DROP(dstr); |
| 4323 | dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */ |
| 4324 | |
| 4325 | /* There's a lot of redundancy below but we're going for speed here */ |
| 4326 | |
| 4327 | switch (stype) { |
| 4328 | case SVt_NULL: |
| 4329 | undef_sstr: |
| 4330 | if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) { |
| 4331 | (void)SvOK_off(dstr); |
| 4332 | return; |
| 4333 | } |
| 4334 | break; |
| 4335 | case SVt_IV: |
| 4336 | if (SvIOK(sstr)) { |
| 4337 | switch (dtype) { |
| 4338 | case SVt_NULL: |
| 4339 | /* For performance, we inline promoting to type SVt_IV. */ |
| 4340 | /* We're starting from SVt_NULL, so provided that define is |
| 4341 | * actual 0, we don't have to unset any SV type flags |
| 4342 | * to promote to SVt_IV. */ |
| 4343 | STATIC_ASSERT_STMT(SVt_NULL == 0); |
| 4344 | SET_SVANY_FOR_BODYLESS_IV(dstr); |
| 4345 | SvFLAGS(dstr) |= SVt_IV; |
| 4346 | break; |
| 4347 | case SVt_NV: |
| 4348 | case SVt_PV: |
| 4349 | sv_upgrade(dstr, SVt_PVIV); |
| 4350 | break; |
| 4351 | case SVt_PVGV: |
| 4352 | case SVt_PVLV: |
| 4353 | goto end_of_first_switch; |
| 4354 | } |
| 4355 | (void)SvIOK_only(dstr); |
| 4356 | SvIV_set(dstr, SvIVX(sstr)); |
| 4357 | if (SvIsUV(sstr)) |
| 4358 | SvIsUV_on(dstr); |
| 4359 | /* SvTAINTED can only be true if the SV has taint magic, which in |
| 4360 | turn means that the SV type is PVMG (or greater). This is the |
| 4361 | case statement for SVt_IV, so this cannot be true (whatever gcov |
| 4362 | may say). */ |
| 4363 | assert(!SvTAINTED(sstr)); |
| 4364 | return; |
| 4365 | } |
| 4366 | if (!SvROK(sstr)) |
| 4367 | goto undef_sstr; |
| 4368 | if (dtype < SVt_PV && dtype != SVt_IV) |
| 4369 | sv_upgrade(dstr, SVt_IV); |
| 4370 | break; |
| 4371 | |
| 4372 | case SVt_NV: |
| 4373 | if (LIKELY( SvNOK(sstr) )) { |
| 4374 | switch (dtype) { |
| 4375 | case SVt_NULL: |
| 4376 | case SVt_IV: |
| 4377 | sv_upgrade(dstr, SVt_NV); |
| 4378 | break; |
| 4379 | case SVt_PV: |
| 4380 | case SVt_PVIV: |
| 4381 | sv_upgrade(dstr, SVt_PVNV); |
| 4382 | break; |
| 4383 | case SVt_PVGV: |
| 4384 | case SVt_PVLV: |
| 4385 | goto end_of_first_switch; |
| 4386 | } |
| 4387 | SvNV_set(dstr, SvNVX(sstr)); |
| 4388 | (void)SvNOK_only(dstr); |
| 4389 | /* SvTAINTED can only be true if the SV has taint magic, which in |
| 4390 | turn means that the SV type is PVMG (or greater). This is the |
| 4391 | case statement for SVt_NV, so this cannot be true (whatever gcov |
| 4392 | may say). */ |
| 4393 | assert(!SvTAINTED(sstr)); |
| 4394 | return; |
| 4395 | } |
| 4396 | goto undef_sstr; |
| 4397 | |
| 4398 | case SVt_PV: |
| 4399 | if (dtype < SVt_PV) |
| 4400 | sv_upgrade(dstr, SVt_PV); |
| 4401 | break; |
| 4402 | case SVt_PVIV: |
| 4403 | if (dtype < SVt_PVIV) |
| 4404 | sv_upgrade(dstr, SVt_PVIV); |
| 4405 | break; |
| 4406 | case SVt_PVNV: |
| 4407 | if (dtype < SVt_PVNV) |
| 4408 | sv_upgrade(dstr, SVt_PVNV); |
| 4409 | break; |
| 4410 | default: |
| 4411 | { |
| 4412 | const char * const type = sv_reftype(sstr,0); |
| 4413 | if (PL_op) |
| 4414 | /* diag_listed_as: Bizarre copy of %s */ |
| 4415 | Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op)); |
| 4416 | else |
| 4417 | Perl_croak(aTHX_ "Bizarre copy of %s", type); |
| 4418 | } |
| 4419 | NOT_REACHED; /* NOTREACHED */ |
| 4420 | |
| 4421 | case SVt_REGEXP: |
| 4422 | upgregexp: |
| 4423 | if (dtype < SVt_REGEXP) |
| 4424 | { |
| 4425 | if (dtype >= SVt_PV) { |
| 4426 | SvPV_free(dstr); |
| 4427 | SvPV_set(dstr, 0); |
| 4428 | SvLEN_set(dstr, 0); |
| 4429 | SvCUR_set(dstr, 0); |
| 4430 | } |
| 4431 | sv_upgrade(dstr, SVt_REGEXP); |
| 4432 | } |
| 4433 | break; |
| 4434 | |
| 4435 | case SVt_INVLIST: |
| 4436 | case SVt_PVLV: |
| 4437 | case SVt_PVGV: |
| 4438 | case SVt_PVMG: |
| 4439 | if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) { |
| 4440 | mg_get(sstr); |
| 4441 | if (SvTYPE(sstr) != stype) |
| 4442 | stype = SvTYPE(sstr); |
| 4443 | } |
| 4444 | if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) { |
| 4445 | glob_assign_glob(dstr, sstr, dtype); |
| 4446 | return; |
| 4447 | } |
| 4448 | if (stype == SVt_PVLV) |
| 4449 | { |
| 4450 | if (isREGEXP(sstr)) goto upgregexp; |
| 4451 | SvUPGRADE(dstr, SVt_PVNV); |
| 4452 | } |
| 4453 | else |
| 4454 | SvUPGRADE(dstr, (svtype)stype); |
| 4455 | } |
| 4456 | end_of_first_switch: |
| 4457 | |
| 4458 | /* dstr may have been upgraded. */ |
| 4459 | dtype = SvTYPE(dstr); |
| 4460 | sflags = SvFLAGS(sstr); |
| 4461 | |
| 4462 | if (UNLIKELY( dtype == SVt_PVCV )) { |
| 4463 | /* Assigning to a subroutine sets the prototype. */ |
| 4464 | if (SvOK(sstr)) { |
| 4465 | STRLEN len; |
| 4466 | const char *const ptr = SvPV_const(sstr, len); |
| 4467 | |
| 4468 | SvGROW(dstr, len + 1); |
| 4469 | Copy(ptr, SvPVX(dstr), len + 1, char); |
| 4470 | SvCUR_set(dstr, len); |
| 4471 | SvPOK_only(dstr); |
| 4472 | SvFLAGS(dstr) |= sflags & SVf_UTF8; |
| 4473 | CvAUTOLOAD_off(dstr); |
| 4474 | } else { |
| 4475 | SvOK_off(dstr); |
| 4476 | } |
| 4477 | } |
| 4478 | else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV |
| 4479 | || dtype == SVt_PVFM)) |
| 4480 | { |
| 4481 | const char * const type = sv_reftype(dstr,0); |
| 4482 | if (PL_op) |
| 4483 | /* diag_listed_as: Cannot copy to %s */ |
| 4484 | Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op)); |
| 4485 | else |
| 4486 | Perl_croak(aTHX_ "Cannot copy to %s", type); |
| 4487 | } else if (sflags & SVf_ROK) { |
| 4488 | if (isGV_with_GP(dstr) |
| 4489 | && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) { |
| 4490 | sstr = SvRV(sstr); |
| 4491 | if (sstr == dstr) { |
| 4492 | if (GvIMPORTED(dstr) != GVf_IMPORTED |
| 4493 | && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) |
| 4494 | { |
| 4495 | GvIMPORTED_on(dstr); |
| 4496 | } |
| 4497 | GvMULTI_on(dstr); |
| 4498 | return; |
| 4499 | } |
| 4500 | glob_assign_glob(dstr, sstr, dtype); |
| 4501 | return; |
| 4502 | } |
| 4503 | |
| 4504 | if (dtype >= SVt_PV) { |
| 4505 | if (isGV_with_GP(dstr)) { |
| 4506 | gv_setref(dstr, sstr); |
| 4507 | return; |
| 4508 | } |
| 4509 | if (SvPVX_const(dstr)) { |
| 4510 | SvPV_free(dstr); |
| 4511 | SvLEN_set(dstr, 0); |
| 4512 | SvCUR_set(dstr, 0); |
| 4513 | } |
| 4514 | } |
| 4515 | (void)SvOK_off(dstr); |
| 4516 | SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr))); |
| 4517 | SvFLAGS(dstr) |= sflags & SVf_ROK; |
| 4518 | assert(!(sflags & SVp_NOK)); |
| 4519 | assert(!(sflags & SVp_IOK)); |
| 4520 | assert(!(sflags & SVf_NOK)); |
| 4521 | assert(!(sflags & SVf_IOK)); |
| 4522 | } |
| 4523 | else if (isGV_with_GP(dstr)) { |
| 4524 | if (!(sflags & SVf_OK)) { |
| 4525 | Perl_ck_warner(aTHX_ packWARN(WARN_MISC), |
| 4526 | "Undefined value assigned to typeglob"); |
| 4527 | } |
| 4528 | else { |
| 4529 | GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV); |
| 4530 | if (dstr != (const SV *)gv) { |
| 4531 | const char * const name = GvNAME((const GV *)dstr); |
| 4532 | const STRLEN len = GvNAMELEN(dstr); |
| 4533 | HV *old_stash = NULL; |
| 4534 | bool reset_isa = FALSE; |
| 4535 | if ((len > 1 && name[len-2] == ':' && name[len-1] == ':') |
| 4536 | || (len == 1 && name[0] == ':')) { |
| 4537 | /* Set aside the old stash, so we can reset isa caches |
| 4538 | on its subclasses. */ |
| 4539 | if((old_stash = GvHV(dstr))) { |
| 4540 | /* Make sure we do not lose it early. */ |
| 4541 | SvREFCNT_inc_simple_void_NN( |
| 4542 | sv_2mortal((SV *)old_stash) |
| 4543 | ); |
| 4544 | } |
| 4545 | reset_isa = TRUE; |
| 4546 | } |
| 4547 | |
| 4548 | if (GvGP(dstr)) { |
| 4549 | SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr)); |
| 4550 | gp_free(MUTABLE_GV(dstr)); |
| 4551 | } |
| 4552 | GvGP_set(dstr, gp_ref(GvGP(gv))); |
| 4553 | |
| 4554 | if (reset_isa) { |
| 4555 | HV * const stash = GvHV(dstr); |
| 4556 | if( |
| 4557 | old_stash ? (HV *)HvENAME_get(old_stash) : stash |
| 4558 | ) |
| 4559 | mro_package_moved( |
| 4560 | stash, old_stash, |
| 4561 | (GV *)dstr, 0 |
| 4562 | ); |
| 4563 | } |
| 4564 | } |
| 4565 | } |
| 4566 | } |
| 4567 | else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV) |
| 4568 | && (stype == SVt_REGEXP || isREGEXP(sstr))) { |
| 4569 | reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr); |
| 4570 | } |
| 4571 | else if (sflags & SVp_POK) { |
| 4572 | const STRLEN cur = SvCUR(sstr); |
| 4573 | const STRLEN len = SvLEN(sstr); |
| 4574 | |
| 4575 | /* |
| 4576 | * We have three basic ways to copy the string: |
| 4577 | * |
| 4578 | * 1. Swipe |
| 4579 | * 2. Copy-on-write |
| 4580 | * 3. Actual copy |
| 4581 | * |
| 4582 | * Which we choose is based on various factors. The following |
| 4583 | * things are listed in order of speed, fastest to slowest: |
| 4584 | * - Swipe |
| 4585 | * - Copying a short string |
| 4586 | * - Copy-on-write bookkeeping |
| 4587 | * - malloc |
| 4588 | * - Copying a long string |
| 4589 | * |
| 4590 | * We swipe the string (steal the string buffer) if the SV on the |
| 4591 | * rhs is about to be freed anyway (TEMP and refcnt==1). This is a |
| 4592 | * big win on long strings. It should be a win on short strings if |
| 4593 | * SvPVX_const(dstr) has to be allocated. If not, it should not |
| 4594 | * slow things down, as SvPVX_const(sstr) would have been freed |
| 4595 | * soon anyway. |
| 4596 | * |
| 4597 | * We also steal the buffer from a PADTMP (operator target) if it |
| 4598 | * is ‘long enough’. For short strings, a swipe does not help |
| 4599 | * here, as it causes more malloc calls the next time the target |
| 4600 | * is used. Benchmarks show that even if SvPVX_const(dstr) has to |
| 4601 | * be allocated it is still not worth swiping PADTMPs for short |
| 4602 | * strings, as the savings here are small. |
| 4603 | * |
| 4604 | * If swiping is not an option, then we see whether it is |
| 4605 | * worth using copy-on-write. If the lhs already has a buf- |
| 4606 | * fer big enough and the string is short, we skip it and fall back |
| 4607 | * to method 3, since memcpy is faster for short strings than the |
| 4608 | * later bookkeeping overhead that copy-on-write entails. |
| 4609 | |
| 4610 | * If the rhs is not a copy-on-write string yet, then we also |
| 4611 | * consider whether the buffer is too large relative to the string |
| 4612 | * it holds. Some operations such as readline allocate a large |
| 4613 | * buffer in the expectation of reusing it. But turning such into |
| 4614 | * a COW buffer is counter-productive because it increases memory |
| 4615 | * usage by making readline allocate a new large buffer the sec- |
| 4616 | * ond time round. So, if the buffer is too large, again, we use |
| 4617 | * method 3 (copy). |
| 4618 | * |
| 4619 | * Finally, if there is no buffer on the left, or the buffer is too |
| 4620 | * small, then we use copy-on-write and make both SVs share the |
| 4621 | * string buffer. |
| 4622 | * |
| 4623 | */ |
| 4624 | |
| 4625 | /* Whichever path we take through the next code, we want this true, |
| 4626 | and doing it now facilitates the COW check. */ |
| 4627 | (void)SvPOK_only(dstr); |
| 4628 | |
| 4629 | if ( |
| 4630 | ( /* Either ... */ |
| 4631 | /* slated for free anyway (and not COW)? */ |
| 4632 | (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP |
| 4633 | /* or a swipable TARG */ |
| 4634 | || ((sflags & |
| 4635 | (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW)) |
| 4636 | == SVs_PADTMP |
| 4637 | /* whose buffer is worth stealing */ |
| 4638 | && CHECK_COWBUF_THRESHOLD(cur,len) |
| 4639 | ) |
| 4640 | ) && |
| 4641 | !(sflags & SVf_OOK) && /* and not involved in OOK hack? */ |
| 4642 | (!(flags & SV_NOSTEAL)) && |
| 4643 | /* and we're allowed to steal temps */ |
| 4644 | SvREFCNT(sstr) == 1 && /* and no other references to it? */ |
| 4645 | len) /* and really is a string */ |
| 4646 | { /* Passes the swipe test. */ |
| 4647 | if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */ |
| 4648 | SvPV_free(dstr); |
| 4649 | SvPV_set(dstr, SvPVX_mutable(sstr)); |
| 4650 | SvLEN_set(dstr, SvLEN(sstr)); |
| 4651 | SvCUR_set(dstr, SvCUR(sstr)); |
| 4652 | |
| 4653 | SvTEMP_off(dstr); |
| 4654 | (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */ |
| 4655 | SvPV_set(sstr, NULL); |
| 4656 | SvLEN_set(sstr, 0); |
| 4657 | SvCUR_set(sstr, 0); |
| 4658 | SvTEMP_off(sstr); |
| 4659 | } |
| 4660 | else if (flags & SV_COW_SHARED_HASH_KEYS |
| 4661 | && |
| 4662 | #ifdef PERL_COPY_ON_WRITE |
| 4663 | (sflags & SVf_IsCOW |
| 4664 | ? (!len || |
| 4665 | ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1) |
| 4666 | /* If this is a regular (non-hek) COW, only so |
| 4667 | many COW "copies" are possible. */ |
| 4668 | && CowREFCNT(sstr) != SV_COW_REFCNT_MAX )) |
| 4669 | : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS |
| 4670 | && !(SvFLAGS(dstr) & SVf_BREAK) |
| 4671 | && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len |
| 4672 | && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1) |
| 4673 | )) |
| 4674 | #else |
| 4675 | sflags & SVf_IsCOW |
| 4676 | && !(SvFLAGS(dstr) & SVf_BREAK) |
| 4677 | #endif |
| 4678 | ) { |
| 4679 | /* Either it's a shared hash key, or it's suitable for |
| 4680 | copy-on-write. */ |
| 4681 | if (DEBUG_C_TEST) { |
| 4682 | PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n"); |
| 4683 | sv_dump(sstr); |
| 4684 | sv_dump(dstr); |
| 4685 | } |
| 4686 | #ifdef PERL_ANY_COW |
| 4687 | if (!(sflags & SVf_IsCOW)) { |
| 4688 | SvIsCOW_on(sstr); |
| 4689 | CowREFCNT(sstr) = 0; |
| 4690 | } |
| 4691 | #endif |
| 4692 | if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */ |
| 4693 | SvPV_free(dstr); |
| 4694 | } |
| 4695 | |
| 4696 | #ifdef PERL_ANY_COW |
| 4697 | if (len) { |
| 4698 | if (sflags & SVf_IsCOW) { |
| 4699 | sv_buf_to_rw(sstr); |
| 4700 | } |
| 4701 | CowREFCNT(sstr)++; |
| 4702 | SvPV_set(dstr, SvPVX_mutable(sstr)); |
| 4703 | sv_buf_to_ro(sstr); |
| 4704 | } else |
| 4705 | #endif |
| 4706 | { |
| 4707 | /* SvIsCOW_shared_hash */ |
| 4708 | DEBUG_C(PerlIO_printf(Perl_debug_log, |
| 4709 | "Copy on write: Sharing hash\n")); |
| 4710 | |
| 4711 | assert (SvTYPE(dstr) >= SVt_PV); |
| 4712 | SvPV_set(dstr, |
| 4713 | HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))))); |
| 4714 | } |
| 4715 | SvLEN_set(dstr, len); |
| 4716 | SvCUR_set(dstr, cur); |
| 4717 | SvIsCOW_on(dstr); |
| 4718 | } else { |
| 4719 | /* Failed the swipe test, and we cannot do copy-on-write either. |
| 4720 | Have to copy the string. */ |
| 4721 | SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */ |
| 4722 | Move(SvPVX_const(sstr),SvPVX(dstr),cur,char); |
| 4723 | SvCUR_set(dstr, cur); |
| 4724 | *SvEND(dstr) = '\0'; |
| 4725 | } |
| 4726 | if (sflags & SVp_NOK) { |
| 4727 | SvNV_set(dstr, SvNVX(sstr)); |
| 4728 | } |
| 4729 | if (sflags & SVp_IOK) { |
| 4730 | SvIV_set(dstr, SvIVX(sstr)); |
| 4731 | /* Must do this otherwise some other overloaded use of 0x80000000 |
| 4732 | gets confused. I guess SVpbm_VALID */ |
| 4733 | if (sflags & SVf_IVisUV) |
| 4734 | SvIsUV_on(dstr); |
| 4735 | } |
| 4736 | SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8); |
| 4737 | { |
| 4738 | const MAGIC * const smg = SvVSTRING_mg(sstr); |
| 4739 | if (smg) { |
| 4740 | sv_magic(dstr, NULL, PERL_MAGIC_vstring, |
| 4741 | smg->mg_ptr, smg->mg_len); |
| 4742 | SvRMAGICAL_on(dstr); |
| 4743 | } |
| 4744 | } |
| 4745 | } |
| 4746 | else if (sflags & (SVp_IOK|SVp_NOK)) { |
| 4747 | (void)SvOK_off(dstr); |
| 4748 | SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK); |
| 4749 | if (sflags & SVp_IOK) { |
| 4750 | /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */ |
| 4751 | SvIV_set(dstr, SvIVX(sstr)); |
| 4752 | } |
| 4753 | if (sflags & SVp_NOK) { |
| 4754 | SvNV_set(dstr, SvNVX(sstr)); |
| 4755 | } |
| 4756 | } |
| 4757 | else { |
| 4758 | if (isGV_with_GP(sstr)) { |
| 4759 | gv_efullname3(dstr, MUTABLE_GV(sstr), "*"); |
| 4760 | } |
| 4761 | else |
| 4762 | (void)SvOK_off(dstr); |
| 4763 | } |
| 4764 | if (SvTAINTED(sstr)) |
| 4765 | SvTAINT(dstr); |
| 4766 | } |
| 4767 | |
| 4768 | /* |
| 4769 | =for apidoc sv_setsv_mg |
| 4770 | |
| 4771 | Like C<sv_setsv>, but also handles 'set' magic. |
| 4772 | |
| 4773 | =cut |
| 4774 | */ |
| 4775 | |
| 4776 | void |
| 4777 | Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr) |
| 4778 | { |
| 4779 | PERL_ARGS_ASSERT_SV_SETSV_MG; |
| 4780 | |
| 4781 | sv_setsv(dstr,sstr); |
| 4782 | SvSETMAGIC(dstr); |
| 4783 | } |
| 4784 | |
| 4785 | #ifdef PERL_ANY_COW |
| 4786 | # define SVt_COW SVt_PV |
| 4787 | SV * |
| 4788 | Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr) |
| 4789 | { |
| 4790 | STRLEN cur = SvCUR(sstr); |
| 4791 | STRLEN len = SvLEN(sstr); |
| 4792 | char *new_pv; |
| 4793 | #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE) |
| 4794 | const bool already = cBOOL(SvIsCOW(sstr)); |
| 4795 | #endif |
| 4796 | |
| 4797 | PERL_ARGS_ASSERT_SV_SETSV_COW; |
| 4798 | |
| 4799 | if (DEBUG_C_TEST) { |
| 4800 | PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n", |
| 4801 | (void*)sstr, (void*)dstr); |
| 4802 | sv_dump(sstr); |
| 4803 | if (dstr) |
| 4804 | sv_dump(dstr); |
| 4805 | } |
| 4806 | |
| 4807 | if (dstr) { |
| 4808 | if (SvTHINKFIRST(dstr)) |
| 4809 | sv_force_normal_flags(dstr, SV_COW_DROP_PV); |
| 4810 | else if (SvPVX_const(dstr)) |
| 4811 | Safefree(SvPVX_mutable(dstr)); |
| 4812 | } |
| 4813 | else |
| 4814 | new_SV(dstr); |
| 4815 | SvUPGRADE(dstr, SVt_COW); |
| 4816 | |
| 4817 | assert (SvPOK(sstr)); |
| 4818 | assert (SvPOKp(sstr)); |
| 4819 | |
| 4820 | if (SvIsCOW(sstr)) { |
| 4821 | |
| 4822 | if (SvLEN(sstr) == 0) { |
| 4823 | /* source is a COW shared hash key. */ |
| 4824 | DEBUG_C(PerlIO_printf(Perl_debug_log, |
| 4825 | "Fast copy on write: Sharing hash\n")); |
| 4826 | new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))); |
| 4827 | goto common_exit; |
| 4828 | } |
| 4829 | assert(SvCUR(sstr)+1 < SvLEN(sstr)); |
| 4830 | assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX); |
| 4831 | } else { |
| 4832 | assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS); |
| 4833 | SvUPGRADE(sstr, SVt_COW); |
| 4834 | SvIsCOW_on(sstr); |
| 4835 | DEBUG_C(PerlIO_printf(Perl_debug_log, |
| 4836 | "Fast copy on write: Converting sstr to COW\n")); |
| 4837 | CowREFCNT(sstr) = 0; |
| 4838 | } |
| 4839 | # ifdef PERL_DEBUG_READONLY_COW |
| 4840 | if (already) sv_buf_to_rw(sstr); |
| 4841 | # endif |
| 4842 | CowREFCNT(sstr)++; |
| 4843 | new_pv = SvPVX_mutable(sstr); |
| 4844 | sv_buf_to_ro(sstr); |
| 4845 | |
| 4846 | common_exit: |
| 4847 | SvPV_set(dstr, new_pv); |
| 4848 | SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW); |
| 4849 | if (SvUTF8(sstr)) |
| 4850 | SvUTF8_on(dstr); |
| 4851 | SvLEN_set(dstr, len); |
| 4852 | SvCUR_set(dstr, cur); |
| 4853 | if (DEBUG_C_TEST) { |
| 4854 | sv_dump(dstr); |
| 4855 | } |
| 4856 | return dstr; |
| 4857 | } |
| 4858 | #endif |
| 4859 | |
| 4860 | /* |
| 4861 | =for apidoc sv_setpvn |
| 4862 | |
| 4863 | Copies a string (possibly containing embedded C<NUL> characters) into an SV. |
| 4864 | The C<len> parameter indicates the number of |
| 4865 | bytes to be copied. If the C<ptr> argument is NULL the SV will become |
| 4866 | undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>. |
| 4867 | |
| 4868 | =cut |
| 4869 | */ |
| 4870 | |
| 4871 | void |
| 4872 | Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len) |
| 4873 | { |
| 4874 | char *dptr; |
| 4875 | |
| 4876 | PERL_ARGS_ASSERT_SV_SETPVN; |
| 4877 | |
| 4878 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 4879 | if (!ptr) { |
| 4880 | (void)SvOK_off(sv); |
| 4881 | return; |
| 4882 | } |
| 4883 | else { |
| 4884 | /* len is STRLEN which is unsigned, need to copy to signed */ |
| 4885 | const IV iv = len; |
| 4886 | if (iv < 0) |
| 4887 | Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %" |
| 4888 | IVdf, iv); |
| 4889 | } |
| 4890 | SvUPGRADE(sv, SVt_PV); |
| 4891 | |
| 4892 | dptr = SvGROW(sv, len + 1); |
| 4893 | Move(ptr,dptr,len,char); |
| 4894 | dptr[len] = '\0'; |
| 4895 | SvCUR_set(sv, len); |
| 4896 | (void)SvPOK_only_UTF8(sv); /* validate pointer */ |
| 4897 | SvTAINT(sv); |
| 4898 | if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv); |
| 4899 | } |
| 4900 | |
| 4901 | /* |
| 4902 | =for apidoc sv_setpvn_mg |
| 4903 | |
| 4904 | Like C<sv_setpvn>, but also handles 'set' magic. |
| 4905 | |
| 4906 | =cut |
| 4907 | */ |
| 4908 | |
| 4909 | void |
| 4910 | Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len) |
| 4911 | { |
| 4912 | PERL_ARGS_ASSERT_SV_SETPVN_MG; |
| 4913 | |
| 4914 | sv_setpvn(sv,ptr,len); |
| 4915 | SvSETMAGIC(sv); |
| 4916 | } |
| 4917 | |
| 4918 | /* |
| 4919 | =for apidoc sv_setpv |
| 4920 | |
| 4921 | Copies a string into an SV. The string must be terminated with a C<NUL> |
| 4922 | character. |
| 4923 | Does not handle 'set' magic. See C<L</sv_setpv_mg>>. |
| 4924 | |
| 4925 | =cut |
| 4926 | */ |
| 4927 | |
| 4928 | void |
| 4929 | Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr) |
| 4930 | { |
| 4931 | STRLEN len; |
| 4932 | |
| 4933 | PERL_ARGS_ASSERT_SV_SETPV; |
| 4934 | |
| 4935 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 4936 | if (!ptr) { |
| 4937 | (void)SvOK_off(sv); |
| 4938 | return; |
| 4939 | } |
| 4940 | len = strlen(ptr); |
| 4941 | SvUPGRADE(sv, SVt_PV); |
| 4942 | |
| 4943 | SvGROW(sv, len + 1); |
| 4944 | Move(ptr,SvPVX(sv),len+1,char); |
| 4945 | SvCUR_set(sv, len); |
| 4946 | (void)SvPOK_only_UTF8(sv); /* validate pointer */ |
| 4947 | SvTAINT(sv); |
| 4948 | if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv); |
| 4949 | } |
| 4950 | |
| 4951 | /* |
| 4952 | =for apidoc sv_setpv_mg |
| 4953 | |
| 4954 | Like C<sv_setpv>, but also handles 'set' magic. |
| 4955 | |
| 4956 | =cut |
| 4957 | */ |
| 4958 | |
| 4959 | void |
| 4960 | Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr) |
| 4961 | { |
| 4962 | PERL_ARGS_ASSERT_SV_SETPV_MG; |
| 4963 | |
| 4964 | sv_setpv(sv,ptr); |
| 4965 | SvSETMAGIC(sv); |
| 4966 | } |
| 4967 | |
| 4968 | void |
| 4969 | Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek) |
| 4970 | { |
| 4971 | PERL_ARGS_ASSERT_SV_SETHEK; |
| 4972 | |
| 4973 | if (!hek) { |
| 4974 | return; |
| 4975 | } |
| 4976 | |
| 4977 | if (HEK_LEN(hek) == HEf_SVKEY) { |
| 4978 | sv_setsv(sv, *(SV**)HEK_KEY(hek)); |
| 4979 | return; |
| 4980 | } else { |
| 4981 | const int flags = HEK_FLAGS(hek); |
| 4982 | if (flags & HVhek_WASUTF8) { |
| 4983 | STRLEN utf8_len = HEK_LEN(hek); |
| 4984 | char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len); |
| 4985 | sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL); |
| 4986 | SvUTF8_on(sv); |
| 4987 | return; |
| 4988 | } else if (flags & HVhek_UNSHARED) { |
| 4989 | sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek)); |
| 4990 | if (HEK_UTF8(hek)) |
| 4991 | SvUTF8_on(sv); |
| 4992 | else SvUTF8_off(sv); |
| 4993 | return; |
| 4994 | } |
| 4995 | { |
| 4996 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 4997 | SvUPGRADE(sv, SVt_PV); |
| 4998 | SvPV_free(sv); |
| 4999 | SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek))); |
| 5000 | SvCUR_set(sv, HEK_LEN(hek)); |
| 5001 | SvLEN_set(sv, 0); |
| 5002 | SvIsCOW_on(sv); |
| 5003 | SvPOK_on(sv); |
| 5004 | if (HEK_UTF8(hek)) |
| 5005 | SvUTF8_on(sv); |
| 5006 | else SvUTF8_off(sv); |
| 5007 | return; |
| 5008 | } |
| 5009 | } |
| 5010 | } |
| 5011 | |
| 5012 | |
| 5013 | /* |
| 5014 | =for apidoc sv_usepvn_flags |
| 5015 | |
| 5016 | Tells an SV to use C<ptr> to find its string value. Normally the |
| 5017 | string is stored inside the SV, but sv_usepvn allows the SV to use an |
| 5018 | outside string. C<ptr> should point to memory that was allocated |
| 5019 | by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be |
| 5020 | the start of a C<Newx>-ed block of memory, and not a pointer to the |
| 5021 | middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write), |
| 5022 | and not be from a non-C<Newx> memory allocator like C<malloc>. The |
| 5023 | string length, C<len>, must be supplied. By default this function |
| 5024 | will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>, |
| 5025 | so that pointer should not be freed or used by the programmer after |
| 5026 | giving it to C<sv_usepvn>, and neither should any pointers from "behind" |
| 5027 | that pointer (e.g. ptr + 1) be used. |
| 5028 | |
| 5029 | If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If |
| 5030 | S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>, |
| 5031 | and the realloc |
| 5032 | will be skipped (i.e. the buffer is actually at least 1 byte longer than |
| 5033 | C<len>, and already meets the requirements for storing in C<SvPVX>). |
| 5034 | |
| 5035 | =cut |
| 5036 | */ |
| 5037 | |
| 5038 | void |
| 5039 | Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags) |
| 5040 | { |
| 5041 | STRLEN allocate; |
| 5042 | |
| 5043 | PERL_ARGS_ASSERT_SV_USEPVN_FLAGS; |
| 5044 | |
| 5045 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 5046 | SvUPGRADE(sv, SVt_PV); |
| 5047 | if (!ptr) { |
| 5048 | (void)SvOK_off(sv); |
| 5049 | if (flags & SV_SMAGIC) |
| 5050 | SvSETMAGIC(sv); |
| 5051 | return; |
| 5052 | } |
| 5053 | if (SvPVX_const(sv)) |
| 5054 | SvPV_free(sv); |
| 5055 | |
| 5056 | #ifdef DEBUGGING |
| 5057 | if (flags & SV_HAS_TRAILING_NUL) |
| 5058 | assert(ptr[len] == '\0'); |
| 5059 | #endif |
| 5060 | |
| 5061 | allocate = (flags & SV_HAS_TRAILING_NUL) |
| 5062 | ? len + 1 : |
| 5063 | #ifdef Perl_safesysmalloc_size |
| 5064 | len + 1; |
| 5065 | #else |
| 5066 | PERL_STRLEN_ROUNDUP(len + 1); |
| 5067 | #endif |
| 5068 | if (flags & SV_HAS_TRAILING_NUL) { |
| 5069 | /* It's long enough - do nothing. |
| 5070 | Specifically Perl_newCONSTSUB is relying on this. */ |
| 5071 | } else { |
| 5072 | #ifdef DEBUGGING |
| 5073 | /* Force a move to shake out bugs in callers. */ |
| 5074 | char *new_ptr = (char*)safemalloc(allocate); |
| 5075 | Copy(ptr, new_ptr, len, char); |
| 5076 | PoisonFree(ptr,len,char); |
| 5077 | Safefree(ptr); |
| 5078 | ptr = new_ptr; |
| 5079 | #else |
| 5080 | ptr = (char*) saferealloc (ptr, allocate); |
| 5081 | #endif |
| 5082 | } |
| 5083 | #ifdef Perl_safesysmalloc_size |
| 5084 | SvLEN_set(sv, Perl_safesysmalloc_size(ptr)); |
| 5085 | #else |
| 5086 | SvLEN_set(sv, allocate); |
| 5087 | #endif |
| 5088 | SvCUR_set(sv, len); |
| 5089 | SvPV_set(sv, ptr); |
| 5090 | if (!(flags & SV_HAS_TRAILING_NUL)) { |
| 5091 | ptr[len] = '\0'; |
| 5092 | } |
| 5093 | (void)SvPOK_only_UTF8(sv); /* validate pointer */ |
| 5094 | SvTAINT(sv); |
| 5095 | if (flags & SV_SMAGIC) |
| 5096 | SvSETMAGIC(sv); |
| 5097 | } |
| 5098 | |
| 5099 | /* |
| 5100 | =for apidoc sv_force_normal_flags |
| 5101 | |
| 5102 | Undo various types of fakery on an SV, where fakery means |
| 5103 | "more than" a string: if the PV is a shared string, make |
| 5104 | a private copy; if we're a ref, stop refing; if we're a glob, downgrade to |
| 5105 | an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when |
| 5106 | we do the copy, and is also used locally; if this is a |
| 5107 | vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set |
| 5108 | then a copy-on-write scalar drops its PV buffer (if any) and becomes |
| 5109 | C<SvPOK_off> rather than making a copy. (Used where this |
| 5110 | scalar is about to be set to some other value.) In addition, |
| 5111 | the C<flags> parameter gets passed to C<sv_unref_flags()> |
| 5112 | when unreffing. C<sv_force_normal> calls this function |
| 5113 | with flags set to 0. |
| 5114 | |
| 5115 | This function is expected to be used to signal to perl that this SV is |
| 5116 | about to be written to, and any extra book-keeping needs to be taken care |
| 5117 | of. Hence, it croaks on read-only values. |
| 5118 | |
| 5119 | =cut |
| 5120 | */ |
| 5121 | |
| 5122 | static void |
| 5123 | S_sv_uncow(pTHX_ SV * const sv, const U32 flags) |
| 5124 | { |
| 5125 | assert(SvIsCOW(sv)); |
| 5126 | { |
| 5127 | #ifdef PERL_ANY_COW |
| 5128 | const char * const pvx = SvPVX_const(sv); |
| 5129 | const STRLEN len = SvLEN(sv); |
| 5130 | const STRLEN cur = SvCUR(sv); |
| 5131 | |
| 5132 | if (DEBUG_C_TEST) { |
| 5133 | PerlIO_printf(Perl_debug_log, |
| 5134 | "Copy on write: Force normal %ld\n", |
| 5135 | (long) flags); |
| 5136 | sv_dump(sv); |
| 5137 | } |
| 5138 | SvIsCOW_off(sv); |
| 5139 | # ifdef PERL_COPY_ON_WRITE |
| 5140 | if (len) { |
| 5141 | /* Must do this first, since the CowREFCNT uses SvPVX and |
| 5142 | we need to write to CowREFCNT, or de-RO the whole buffer if we are |
| 5143 | the only owner left of the buffer. */ |
| 5144 | sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */ |
| 5145 | { |
| 5146 | U8 cowrefcnt = CowREFCNT(sv); |
| 5147 | if(cowrefcnt != 0) { |
| 5148 | cowrefcnt--; |
| 5149 | CowREFCNT(sv) = cowrefcnt; |
| 5150 | sv_buf_to_ro(sv); |
| 5151 | goto copy_over; |
| 5152 | } |
| 5153 | } |
| 5154 | /* Else we are the only owner of the buffer. */ |
| 5155 | } |
| 5156 | else |
| 5157 | # endif |
| 5158 | { |
| 5159 | /* This SV doesn't own the buffer, so need to Newx() a new one: */ |
| 5160 | copy_over: |
| 5161 | SvPV_set(sv, NULL); |
| 5162 | SvCUR_set(sv, 0); |
| 5163 | SvLEN_set(sv, 0); |
| 5164 | if (flags & SV_COW_DROP_PV) { |
| 5165 | /* OK, so we don't need to copy our buffer. */ |
| 5166 | SvPOK_off(sv); |
| 5167 | } else { |
| 5168 | SvGROW(sv, cur + 1); |
| 5169 | Move(pvx,SvPVX(sv),cur,char); |
| 5170 | SvCUR_set(sv, cur); |
| 5171 | *SvEND(sv) = '\0'; |
| 5172 | } |
| 5173 | if (len) { |
| 5174 | } else { |
| 5175 | unshare_hek(SvSHARED_HEK_FROM_PV(pvx)); |
| 5176 | } |
| 5177 | if (DEBUG_C_TEST) { |
| 5178 | sv_dump(sv); |
| 5179 | } |
| 5180 | } |
| 5181 | #else |
| 5182 | const char * const pvx = SvPVX_const(sv); |
| 5183 | const STRLEN len = SvCUR(sv); |
| 5184 | SvIsCOW_off(sv); |
| 5185 | SvPV_set(sv, NULL); |
| 5186 | SvLEN_set(sv, 0); |
| 5187 | if (flags & SV_COW_DROP_PV) { |
| 5188 | /* OK, so we don't need to copy our buffer. */ |
| 5189 | SvPOK_off(sv); |
| 5190 | } else { |
| 5191 | SvGROW(sv, len + 1); |
| 5192 | Move(pvx,SvPVX(sv),len,char); |
| 5193 | *SvEND(sv) = '\0'; |
| 5194 | } |
| 5195 | unshare_hek(SvSHARED_HEK_FROM_PV(pvx)); |
| 5196 | #endif |
| 5197 | } |
| 5198 | } |
| 5199 | |
| 5200 | void |
| 5201 | Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags) |
| 5202 | { |
| 5203 | PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS; |
| 5204 | |
| 5205 | if (SvREADONLY(sv)) |
| 5206 | Perl_croak_no_modify(); |
| 5207 | else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV)) |
| 5208 | S_sv_uncow(aTHX_ sv, flags); |
| 5209 | if (SvROK(sv)) |
| 5210 | sv_unref_flags(sv, flags); |
| 5211 | else if (SvFAKE(sv) && isGV_with_GP(sv)) |
| 5212 | sv_unglob(sv, flags); |
| 5213 | else if (SvFAKE(sv) && isREGEXP(sv)) { |
| 5214 | /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous |
| 5215 | to sv_unglob. We only need it here, so inline it. */ |
| 5216 | const bool islv = SvTYPE(sv) == SVt_PVLV; |
| 5217 | const svtype new_type = |
| 5218 | islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV; |
| 5219 | SV *const temp = newSV_type(new_type); |
| 5220 | regexp *const temp_p = ReANY((REGEXP *)sv); |
| 5221 | |
| 5222 | if (new_type == SVt_PVMG) { |
| 5223 | SvMAGIC_set(temp, SvMAGIC(sv)); |
| 5224 | SvMAGIC_set(sv, NULL); |
| 5225 | SvSTASH_set(temp, SvSTASH(sv)); |
| 5226 | SvSTASH_set(sv, NULL); |
| 5227 | } |
| 5228 | if (!islv) SvCUR_set(temp, SvCUR(sv)); |
| 5229 | /* Remember that SvPVX is in the head, not the body. But |
| 5230 | RX_WRAPPED is in the body. */ |
| 5231 | assert(ReANY((REGEXP *)sv)->mother_re); |
| 5232 | /* Their buffer is already owned by someone else. */ |
| 5233 | if (flags & SV_COW_DROP_PV) { |
| 5234 | /* SvLEN is already 0. For SVt_REGEXP, we have a brand new |
| 5235 | zeroed body. For SVt_PVLV, it should have been set to 0 |
| 5236 | before turning into a regexp. */ |
| 5237 | assert(!SvLEN(islv ? sv : temp)); |
| 5238 | sv->sv_u.svu_pv = 0; |
| 5239 | } |
| 5240 | else { |
| 5241 | sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv)); |
| 5242 | SvLEN_set(islv ? sv : temp, SvCUR(sv)+1); |
| 5243 | SvPOK_on(sv); |
| 5244 | } |
| 5245 | |
| 5246 | /* Now swap the rest of the bodies. */ |
| 5247 | |
| 5248 | SvFAKE_off(sv); |
| 5249 | if (!islv) { |
| 5250 | SvFLAGS(sv) &= ~SVTYPEMASK; |
| 5251 | SvFLAGS(sv) |= new_type; |
| 5252 | SvANY(sv) = SvANY(temp); |
| 5253 | } |
| 5254 | |
| 5255 | SvFLAGS(temp) &= ~(SVTYPEMASK); |
| 5256 | SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE; |
| 5257 | SvANY(temp) = temp_p; |
| 5258 | temp->sv_u.svu_rx = (regexp *)temp_p; |
| 5259 | |
| 5260 | SvREFCNT_dec_NN(temp); |
| 5261 | } |
| 5262 | else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring); |
| 5263 | } |
| 5264 | |
| 5265 | /* |
| 5266 | =for apidoc sv_chop |
| 5267 | |
| 5268 | Efficient removal of characters from the beginning of the string buffer. |
| 5269 | C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a |
| 5270 | pointer to somewhere inside the string buffer. C<ptr> becomes the first |
| 5271 | character of the adjusted string. Uses the C<OOK> hack. On return, only |
| 5272 | C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true. |
| 5273 | |
| 5274 | Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer |
| 5275 | refer to the same chunk of data. |
| 5276 | |
| 5277 | The unfortunate similarity of this function's name to that of Perl's C<chop> |
| 5278 | operator is strictly coincidental. This function works from the left; |
| 5279 | C<chop> works from the right. |
| 5280 | |
| 5281 | =cut |
| 5282 | */ |
| 5283 | |
| 5284 | void |
| 5285 | Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr) |
| 5286 | { |
| 5287 | STRLEN delta; |
| 5288 | STRLEN old_delta; |
| 5289 | U8 *p; |
| 5290 | #ifdef DEBUGGING |
| 5291 | const U8 *evacp; |
| 5292 | STRLEN evacn; |
| 5293 | #endif |
| 5294 | STRLEN max_delta; |
| 5295 | |
| 5296 | PERL_ARGS_ASSERT_SV_CHOP; |
| 5297 | |
| 5298 | if (!ptr || !SvPOKp(sv)) |
| 5299 | return; |
| 5300 | delta = ptr - SvPVX_const(sv); |
| 5301 | if (!delta) { |
| 5302 | /* Nothing to do. */ |
| 5303 | return; |
| 5304 | } |
| 5305 | max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv); |
| 5306 | if (delta > max_delta) |
| 5307 | Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p", |
| 5308 | ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta); |
| 5309 | /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */ |
| 5310 | SV_CHECK_THINKFIRST(sv); |
| 5311 | SvPOK_only_UTF8(sv); |
| 5312 | |
| 5313 | if (!SvOOK(sv)) { |
| 5314 | if (!SvLEN(sv)) { /* make copy of shared string */ |
| 5315 | const char *pvx = SvPVX_const(sv); |
| 5316 | const STRLEN len = SvCUR(sv); |
| 5317 | SvGROW(sv, len + 1); |
| 5318 | Move(pvx,SvPVX(sv),len,char); |
| 5319 | *SvEND(sv) = '\0'; |
| 5320 | } |
| 5321 | SvOOK_on(sv); |
| 5322 | old_delta = 0; |
| 5323 | } else { |
| 5324 | SvOOK_offset(sv, old_delta); |
| 5325 | } |
| 5326 | SvLEN_set(sv, SvLEN(sv) - delta); |
| 5327 | SvCUR_set(sv, SvCUR(sv) - delta); |
| 5328 | SvPV_set(sv, SvPVX(sv) + delta); |
| 5329 | |
| 5330 | p = (U8 *)SvPVX_const(sv); |
| 5331 | |
| 5332 | #ifdef DEBUGGING |
| 5333 | /* how many bytes were evacuated? we will fill them with sentinel |
| 5334 | bytes, except for the part holding the new offset of course. */ |
| 5335 | evacn = delta; |
| 5336 | if (old_delta) |
| 5337 | evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN)); |
| 5338 | assert(evacn); |
| 5339 | assert(evacn <= delta + old_delta); |
| 5340 | evacp = p - evacn; |
| 5341 | #endif |
| 5342 | |
| 5343 | /* This sets 'delta' to the accumulated value of all deltas so far */ |
| 5344 | delta += old_delta; |
| 5345 | assert(delta); |
| 5346 | |
| 5347 | /* If 'delta' fits in a byte, store it just prior to the new beginning of |
| 5348 | * the string; otherwise store a 0 byte there and store 'delta' just prior |
| 5349 | * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a |
| 5350 | * portion of the chopped part of the string */ |
| 5351 | if (delta < 0x100) { |
| 5352 | *--p = (U8) delta; |
| 5353 | } else { |
| 5354 | *--p = 0; |
| 5355 | p -= sizeof(STRLEN); |
| 5356 | Copy((U8*)&delta, p, sizeof(STRLEN), U8); |
| 5357 | } |
| 5358 | |
| 5359 | #ifdef DEBUGGING |
| 5360 | /* Fill the preceding buffer with sentinals to verify that no-one is |
| 5361 | using it. */ |
| 5362 | while (p > evacp) { |
| 5363 | --p; |
| 5364 | *p = (U8)PTR2UV(p); |
| 5365 | } |
| 5366 | #endif |
| 5367 | } |
| 5368 | |
| 5369 | /* |
| 5370 | =for apidoc sv_catpvn |
| 5371 | |
| 5372 | Concatenates the string onto the end of the string which is in the SV. |
| 5373 | C<len> indicates number of bytes to copy. If the SV has the UTF-8 |
| 5374 | status set, then the bytes appended should be valid UTF-8. |
| 5375 | Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>. |
| 5376 | |
| 5377 | =for apidoc sv_catpvn_flags |
| 5378 | |
| 5379 | Concatenates the string onto the end of the string which is in the SV. The |
| 5380 | C<len> indicates number of bytes to copy. |
| 5381 | |
| 5382 | By default, the string appended is assumed to be valid UTF-8 if the SV has |
| 5383 | the UTF-8 status set, and a string of bytes otherwise. One can force the |
| 5384 | appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8> |
| 5385 | flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the |
| 5386 | string appended will be upgraded to UTF-8 if necessary. |
| 5387 | |
| 5388 | If C<flags> has the C<SV_SMAGIC> bit set, will |
| 5389 | C<mg_set> on C<dsv> afterwards if appropriate. |
| 5390 | C<sv_catpvn> and C<sv_catpvn_nomg> are implemented |
| 5391 | in terms of this function. |
| 5392 | |
| 5393 | =cut |
| 5394 | */ |
| 5395 | |
| 5396 | void |
| 5397 | Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags) |
| 5398 | { |
| 5399 | STRLEN dlen; |
| 5400 | const char * const dstr = SvPV_force_flags(dsv, dlen, flags); |
| 5401 | |
| 5402 | PERL_ARGS_ASSERT_SV_CATPVN_FLAGS; |
| 5403 | assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8)); |
| 5404 | |
| 5405 | if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) { |
| 5406 | if (flags & SV_CATUTF8 && !SvUTF8(dsv)) { |
| 5407 | sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1); |
| 5408 | dlen = SvCUR(dsv); |
| 5409 | } |
| 5410 | else SvGROW(dsv, dlen + slen + 1); |
| 5411 | if (sstr == dstr) |
| 5412 | sstr = SvPVX_const(dsv); |
| 5413 | Move(sstr, SvPVX(dsv) + dlen, slen, char); |
| 5414 | SvCUR_set(dsv, SvCUR(dsv) + slen); |
| 5415 | } |
| 5416 | else { |
| 5417 | /* We inline bytes_to_utf8, to avoid an extra malloc. */ |
| 5418 | const char * const send = sstr + slen; |
| 5419 | U8 *d; |
| 5420 | |
| 5421 | /* Something this code does not account for, which I think is |
| 5422 | impossible; it would require the same pv to be treated as |
| 5423 | bytes *and* utf8, which would indicate a bug elsewhere. */ |
| 5424 | assert(sstr != dstr); |
| 5425 | |
| 5426 | SvGROW(dsv, dlen + slen * 2 + 1); |
| 5427 | d = (U8 *)SvPVX(dsv) + dlen; |
| 5428 | |
| 5429 | while (sstr < send) { |
| 5430 | append_utf8_from_native_byte(*sstr, &d); |
| 5431 | sstr++; |
| 5432 | } |
| 5433 | SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv)); |
| 5434 | } |
| 5435 | *SvEND(dsv) = '\0'; |
| 5436 | (void)SvPOK_only_UTF8(dsv); /* validate pointer */ |
| 5437 | SvTAINT(dsv); |
| 5438 | if (flags & SV_SMAGIC) |
| 5439 | SvSETMAGIC(dsv); |
| 5440 | } |
| 5441 | |
| 5442 | /* |
| 5443 | =for apidoc sv_catsv |
| 5444 | |
| 5445 | Concatenates the string from SV C<ssv> onto the end of the string in SV |
| 5446 | C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>. |
| 5447 | Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>> |
| 5448 | and C<L</sv_catsv_nomg>>. |
| 5449 | |
| 5450 | =for apidoc sv_catsv_flags |
| 5451 | |
| 5452 | Concatenates the string from SV C<ssv> onto the end of the string in SV |
| 5453 | C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>. |
| 5454 | If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if |
| 5455 | appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on |
| 5456 | the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>, |
| 5457 | and C<sv_catsv_mg> are implemented in terms of this function. |
| 5458 | |
| 5459 | =cut */ |
| 5460 | |
| 5461 | void |
| 5462 | Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags) |
| 5463 | { |
| 5464 | PERL_ARGS_ASSERT_SV_CATSV_FLAGS; |
| 5465 | |
| 5466 | if (ssv) { |
| 5467 | STRLEN slen; |
| 5468 | const char *spv = SvPV_flags_const(ssv, slen, flags); |
| 5469 | if (flags & SV_GMAGIC) |
| 5470 | SvGETMAGIC(dsv); |
| 5471 | sv_catpvn_flags(dsv, spv, slen, |
| 5472 | DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES); |
| 5473 | if (flags & SV_SMAGIC) |
| 5474 | SvSETMAGIC(dsv); |
| 5475 | } |
| 5476 | } |
| 5477 | |
| 5478 | /* |
| 5479 | =for apidoc sv_catpv |
| 5480 | |
| 5481 | Concatenates the C<NUL>-terminated string onto the end of the string which is |
| 5482 | in the SV. |
| 5483 | If the SV has the UTF-8 status set, then the bytes appended should be |
| 5484 | valid UTF-8. Handles 'get' magic, but not 'set' magic. See |
| 5485 | C<L</sv_catpv_mg>>. |
| 5486 | |
| 5487 | =cut */ |
| 5488 | |
| 5489 | void |
| 5490 | Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr) |
| 5491 | { |
| 5492 | STRLEN len; |
| 5493 | STRLEN tlen; |
| 5494 | char *junk; |
| 5495 | |
| 5496 | PERL_ARGS_ASSERT_SV_CATPV; |
| 5497 | |
| 5498 | if (!ptr) |
| 5499 | return; |
| 5500 | junk = SvPV_force(sv, tlen); |
| 5501 | len = strlen(ptr); |
| 5502 | SvGROW(sv, tlen + len + 1); |
| 5503 | if (ptr == junk) |
| 5504 | ptr = SvPVX_const(sv); |
| 5505 | Move(ptr,SvPVX(sv)+tlen,len+1,char); |
| 5506 | SvCUR_set(sv, SvCUR(sv) + len); |
| 5507 | (void)SvPOK_only_UTF8(sv); /* validate pointer */ |
| 5508 | SvTAINT(sv); |
| 5509 | } |
| 5510 | |
| 5511 | /* |
| 5512 | =for apidoc sv_catpv_flags |
| 5513 | |
| 5514 | Concatenates the C<NUL>-terminated string onto the end of the string which is |
| 5515 | in the SV. |
| 5516 | If the SV has the UTF-8 status set, then the bytes appended should |
| 5517 | be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set> |
| 5518 | on the modified SV if appropriate. |
| 5519 | |
| 5520 | =cut |
| 5521 | */ |
| 5522 | |
| 5523 | void |
| 5524 | Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags) |
| 5525 | { |
| 5526 | PERL_ARGS_ASSERT_SV_CATPV_FLAGS; |
| 5527 | sv_catpvn_flags(dstr, sstr, strlen(sstr), flags); |
| 5528 | } |
| 5529 | |
| 5530 | /* |
| 5531 | =for apidoc sv_catpv_mg |
| 5532 | |
| 5533 | Like C<sv_catpv>, but also handles 'set' magic. |
| 5534 | |
| 5535 | =cut |
| 5536 | */ |
| 5537 | |
| 5538 | void |
| 5539 | Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr) |
| 5540 | { |
| 5541 | PERL_ARGS_ASSERT_SV_CATPV_MG; |
| 5542 | |
| 5543 | sv_catpv(sv,ptr); |
| 5544 | SvSETMAGIC(sv); |
| 5545 | } |
| 5546 | |
| 5547 | /* |
| 5548 | =for apidoc newSV |
| 5549 | |
| 5550 | Creates a new SV. A non-zero C<len> parameter indicates the number of |
| 5551 | bytes of preallocated string space the SV should have. An extra byte for a |
| 5552 | trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string |
| 5553 | space is allocated.) The reference count for the new SV is set to 1. |
| 5554 | |
| 5555 | In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first |
| 5556 | parameter, I<x>, a debug aid which allowed callers to identify themselves. |
| 5557 | This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see |
| 5558 | L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS |
| 5559 | modules supporting older perls. |
| 5560 | |
| 5561 | =cut |
| 5562 | */ |
| 5563 | |
| 5564 | SV * |
| 5565 | Perl_newSV(pTHX_ const STRLEN len) |
| 5566 | { |
| 5567 | SV *sv; |
| 5568 | |
| 5569 | new_SV(sv); |
| 5570 | if (len) { |
| 5571 | sv_grow(sv, len + 1); |
| 5572 | } |
| 5573 | return sv; |
| 5574 | } |
| 5575 | /* |
| 5576 | =for apidoc sv_magicext |
| 5577 | |
| 5578 | Adds magic to an SV, upgrading it if necessary. Applies the |
| 5579 | supplied C<vtable> and returns a pointer to the magic added. |
| 5580 | |
| 5581 | Note that C<sv_magicext> will allow things that C<sv_magic> will not. |
| 5582 | In particular, you can add magic to C<SvREADONLY> SVs, and add more than |
| 5583 | one instance of the same C<how>. |
| 5584 | |
| 5585 | If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is |
| 5586 | stored, if C<namlen> is zero then C<name> is stored as-is and - as another |
| 5587 | special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed |
| 5588 | to contain an SV* and is stored as-is with its C<REFCNT> incremented. |
| 5589 | |
| 5590 | (This is now used as a subroutine by C<sv_magic>.) |
| 5591 | |
| 5592 | =cut |
| 5593 | */ |
| 5594 | MAGIC * |
| 5595 | Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how, |
| 5596 | const MGVTBL *const vtable, const char *const name, const I32 namlen) |
| 5597 | { |
| 5598 | MAGIC* mg; |
| 5599 | |
| 5600 | PERL_ARGS_ASSERT_SV_MAGICEXT; |
| 5601 | |
| 5602 | SvUPGRADE(sv, SVt_PVMG); |
| 5603 | Newxz(mg, 1, MAGIC); |
| 5604 | mg->mg_moremagic = SvMAGIC(sv); |
| 5605 | SvMAGIC_set(sv, mg); |
| 5606 | |
| 5607 | /* Sometimes a magic contains a reference loop, where the sv and |
| 5608 | object refer to each other. To prevent a reference loop that |
| 5609 | would prevent such objects being freed, we look for such loops |
| 5610 | and if we find one we avoid incrementing the object refcount. |
| 5611 | |
| 5612 | Note we cannot do this to avoid self-tie loops as intervening RV must |
| 5613 | have its REFCNT incremented to keep it in existence. |
| 5614 | |
| 5615 | */ |
| 5616 | if (!obj || obj == sv || |
| 5617 | how == PERL_MAGIC_arylen || |
| 5618 | how == PERL_MAGIC_symtab || |
| 5619 | (SvTYPE(obj) == SVt_PVGV && |
| 5620 | (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv |
| 5621 | || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv |
| 5622 | || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv))) |
| 5623 | { |
| 5624 | mg->mg_obj = obj; |
| 5625 | } |
| 5626 | else { |
| 5627 | mg->mg_obj = SvREFCNT_inc_simple(obj); |
| 5628 | mg->mg_flags |= MGf_REFCOUNTED; |
| 5629 | } |
| 5630 | |
| 5631 | /* Normal self-ties simply pass a null object, and instead of |
| 5632 | using mg_obj directly, use the SvTIED_obj macro to produce a |
| 5633 | new RV as needed. For glob "self-ties", we are tieing the PVIO |
| 5634 | with an RV obj pointing to the glob containing the PVIO. In |
| 5635 | this case, to avoid a reference loop, we need to weaken the |
| 5636 | reference. |
| 5637 | */ |
| 5638 | |
| 5639 | if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO && |
| 5640 | obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv) |
| 5641 | { |
| 5642 | sv_rvweaken(obj); |
| 5643 | } |
| 5644 | |
| 5645 | mg->mg_type = how; |
| 5646 | mg->mg_len = namlen; |
| 5647 | if (name) { |
| 5648 | if (namlen > 0) |
| 5649 | mg->mg_ptr = savepvn(name, namlen); |
| 5650 | else if (namlen == HEf_SVKEY) { |
| 5651 | /* Yes, this is casting away const. This is only for the case of |
| 5652 | HEf_SVKEY. I think we need to document this aberation of the |
| 5653 | constness of the API, rather than making name non-const, as |
| 5654 | that change propagating outwards a long way. */ |
| 5655 | mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name); |
| 5656 | } else |
| 5657 | mg->mg_ptr = (char *) name; |
| 5658 | } |
| 5659 | mg->mg_virtual = (MGVTBL *) vtable; |
| 5660 | |
| 5661 | mg_magical(sv); |
| 5662 | return mg; |
| 5663 | } |
| 5664 | |
| 5665 | MAGIC * |
| 5666 | Perl_sv_magicext_mglob(pTHX_ SV *sv) |
| 5667 | { |
| 5668 | PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB; |
| 5669 | if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') { |
| 5670 | /* This sv is only a delegate. //g magic must be attached to |
| 5671 | its target. */ |
| 5672 | vivify_defelem(sv); |
| 5673 | sv = LvTARG(sv); |
| 5674 | } |
| 5675 | return sv_magicext(sv, NULL, PERL_MAGIC_regex_global, |
| 5676 | &PL_vtbl_mglob, 0, 0); |
| 5677 | } |
| 5678 | |
| 5679 | /* |
| 5680 | =for apidoc sv_magic |
| 5681 | |
| 5682 | Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if |
| 5683 | necessary, then adds a new magic item of type C<how> to the head of the |
| 5684 | magic list. |
| 5685 | |
| 5686 | See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the |
| 5687 | handling of the C<name> and C<namlen> arguments. |
| 5688 | |
| 5689 | You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also |
| 5690 | to add more than one instance of the same C<how>. |
| 5691 | |
| 5692 | =cut |
| 5693 | */ |
| 5694 | |
| 5695 | void |
| 5696 | Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how, |
| 5697 | const char *const name, const I32 namlen) |
| 5698 | { |
| 5699 | const MGVTBL *vtable; |
| 5700 | MAGIC* mg; |
| 5701 | unsigned int flags; |
| 5702 | unsigned int vtable_index; |
| 5703 | |
| 5704 | PERL_ARGS_ASSERT_SV_MAGIC; |
| 5705 | |
| 5706 | if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data) |
| 5707 | || ((flags = PL_magic_data[how]), |
| 5708 | (vtable_index = flags & PERL_MAGIC_VTABLE_MASK) |
| 5709 | > magic_vtable_max)) |
| 5710 | Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how); |
| 5711 | |
| 5712 | /* PERL_MAGIC_ext is reserved for use by extensions not perl internals. |
| 5713 | Useful for attaching extension internal data to perl vars. |
| 5714 | Note that multiple extensions may clash if magical scalars |
| 5715 | etc holding private data from one are passed to another. */ |
| 5716 | |
| 5717 | vtable = (vtable_index == magic_vtable_max) |
| 5718 | ? NULL : PL_magic_vtables + vtable_index; |
| 5719 | |
| 5720 | if (SvREADONLY(sv)) { |
| 5721 | if ( |
| 5722 | !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how) |
| 5723 | ) |
| 5724 | { |
| 5725 | Perl_croak_no_modify(); |
| 5726 | } |
| 5727 | } |
| 5728 | if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) { |
| 5729 | if (SvMAGIC(sv) && (mg = mg_find(sv, how))) { |
| 5730 | /* sv_magic() refuses to add a magic of the same 'how' as an |
| 5731 | existing one |
| 5732 | */ |
| 5733 | if (how == PERL_MAGIC_taint) |
| 5734 | mg->mg_len |= 1; |
| 5735 | return; |
| 5736 | } |
| 5737 | } |
| 5738 | |
| 5739 | /* Force pos to be stored as characters, not bytes. */ |
| 5740 | if (SvMAGICAL(sv) && DO_UTF8(sv) |
| 5741 | && (mg = mg_find(sv, PERL_MAGIC_regex_global)) |
| 5742 | && mg->mg_len != -1 |
| 5743 | && mg->mg_flags & MGf_BYTES) { |
| 5744 | mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len, |
| 5745 | SV_CONST_RETURN); |
| 5746 | mg->mg_flags &= ~MGf_BYTES; |
| 5747 | } |
| 5748 | |
| 5749 | /* Rest of work is done else where */ |
| 5750 | mg = sv_magicext(sv,obj,how,vtable,name,namlen); |
| 5751 | |
| 5752 | switch (how) { |
| 5753 | case PERL_MAGIC_taint: |
| 5754 | mg->mg_len = 1; |
| 5755 | break; |
| 5756 | case PERL_MAGIC_ext: |
| 5757 | case PERL_MAGIC_dbfile: |
| 5758 | SvRMAGICAL_on(sv); |
| 5759 | break; |
| 5760 | } |
| 5761 | } |
| 5762 | |
| 5763 | static int |
| 5764 | S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags) |
| 5765 | { |
| 5766 | MAGIC* mg; |
| 5767 | MAGIC** mgp; |
| 5768 | |
| 5769 | assert(flags <= 1); |
| 5770 | |
| 5771 | if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv)) |
| 5772 | return 0; |
| 5773 | mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic); |
| 5774 | for (mg = *mgp; mg; mg = *mgp) { |
| 5775 | const MGVTBL* const virt = mg->mg_virtual; |
| 5776 | if (mg->mg_type == type && (!flags || virt == vtbl)) { |
| 5777 | *mgp = mg->mg_moremagic; |
| 5778 | if (virt && virt->svt_free) |
| 5779 | virt->svt_free(aTHX_ sv, mg); |
| 5780 | if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) { |
| 5781 | if (mg->mg_len > 0) |
| 5782 | Safefree(mg->mg_ptr); |
| 5783 | else if (mg->mg_len == HEf_SVKEY) |
| 5784 | SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr)); |
| 5785 | else if (mg->mg_type == PERL_MAGIC_utf8) |
| 5786 | Safefree(mg->mg_ptr); |
| 5787 | } |
| 5788 | if (mg->mg_flags & MGf_REFCOUNTED) |
| 5789 | SvREFCNT_dec(mg->mg_obj); |
| 5790 | Safefree(mg); |
| 5791 | } |
| 5792 | else |
| 5793 | mgp = &mg->mg_moremagic; |
| 5794 | } |
| 5795 | if (SvMAGIC(sv)) { |
| 5796 | if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */ |
| 5797 | mg_magical(sv); /* else fix the flags now */ |
| 5798 | } |
| 5799 | else |
| 5800 | SvMAGICAL_off(sv); |
| 5801 | |
| 5802 | return 0; |
| 5803 | } |
| 5804 | |
| 5805 | /* |
| 5806 | =for apidoc sv_unmagic |
| 5807 | |
| 5808 | Removes all magic of type C<type> from an SV. |
| 5809 | |
| 5810 | =cut |
| 5811 | */ |
| 5812 | |
| 5813 | int |
| 5814 | Perl_sv_unmagic(pTHX_ SV *const sv, const int type) |
| 5815 | { |
| 5816 | PERL_ARGS_ASSERT_SV_UNMAGIC; |
| 5817 | return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0); |
| 5818 | } |
| 5819 | |
| 5820 | /* |
| 5821 | =for apidoc sv_unmagicext |
| 5822 | |
| 5823 | Removes all magic of type C<type> with the specified C<vtbl> from an SV. |
| 5824 | |
| 5825 | =cut |
| 5826 | */ |
| 5827 | |
| 5828 | int |
| 5829 | Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl) |
| 5830 | { |
| 5831 | PERL_ARGS_ASSERT_SV_UNMAGICEXT; |
| 5832 | return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1); |
| 5833 | } |
| 5834 | |
| 5835 | /* |
| 5836 | =for apidoc sv_rvweaken |
| 5837 | |
| 5838 | Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the |
| 5839 | referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and |
| 5840 | push a back-reference to this RV onto the array of backreferences |
| 5841 | associated with that magic. If the RV is magical, set magic will be |
| 5842 | called after the RV is cleared. |
| 5843 | |
| 5844 | =cut |
| 5845 | */ |
| 5846 | |
| 5847 | SV * |
| 5848 | Perl_sv_rvweaken(pTHX_ SV *const sv) |
| 5849 | { |
| 5850 | SV *tsv; |
| 5851 | |
| 5852 | PERL_ARGS_ASSERT_SV_RVWEAKEN; |
| 5853 | |
| 5854 | if (!SvOK(sv)) /* let undefs pass */ |
| 5855 | return sv; |
| 5856 | if (!SvROK(sv)) |
| 5857 | Perl_croak(aTHX_ "Can't weaken a nonreference"); |
| 5858 | else if (SvWEAKREF(sv)) { |
| 5859 | Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak"); |
| 5860 | return sv; |
| 5861 | } |
| 5862 | else if (SvREADONLY(sv)) croak_no_modify(); |
| 5863 | tsv = SvRV(sv); |
| 5864 | Perl_sv_add_backref(aTHX_ tsv, sv); |
| 5865 | SvWEAKREF_on(sv); |
| 5866 | SvREFCNT_dec_NN(tsv); |
| 5867 | return sv; |
| 5868 | } |
| 5869 | |
| 5870 | /* |
| 5871 | =for apidoc sv_get_backrefs |
| 5872 | |
| 5873 | If C<sv> is the target of a weak reference then it returns the back |
| 5874 | references structure associated with the sv; otherwise return C<NULL>. |
| 5875 | |
| 5876 | When returning a non-null result the type of the return is relevant. If it |
| 5877 | is an AV then the elements of the AV are the weak reference RVs which |
| 5878 | point at this item. If it is any other type then the item itself is the |
| 5879 | weak reference. |
| 5880 | |
| 5881 | See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>, |
| 5882 | C<Perl_sv_kill_backrefs()> |
| 5883 | |
| 5884 | =cut |
| 5885 | */ |
| 5886 | |
| 5887 | SV * |
| 5888 | Perl_sv_get_backrefs(SV *const sv) |
| 5889 | { |
| 5890 | SV *backrefs= NULL; |
| 5891 | |
| 5892 | PERL_ARGS_ASSERT_SV_GET_BACKREFS; |
| 5893 | |
| 5894 | /* find slot to store array or singleton backref */ |
| 5895 | |
| 5896 | if (SvTYPE(sv) == SVt_PVHV) { |
| 5897 | if (SvOOK(sv)) { |
| 5898 | struct xpvhv_aux * const iter = HvAUX((HV *)sv); |
| 5899 | backrefs = (SV *)iter->xhv_backreferences; |
| 5900 | } |
| 5901 | } else if (SvMAGICAL(sv)) { |
| 5902 | MAGIC *mg = mg_find(sv, PERL_MAGIC_backref); |
| 5903 | if (mg) |
| 5904 | backrefs = mg->mg_obj; |
| 5905 | } |
| 5906 | return backrefs; |
| 5907 | } |
| 5908 | |
| 5909 | /* Give tsv backref magic if it hasn't already got it, then push a |
| 5910 | * back-reference to sv onto the array associated with the backref magic. |
| 5911 | * |
| 5912 | * As an optimisation, if there's only one backref and it's not an AV, |
| 5913 | * store it directly in the HvAUX or mg_obj slot, avoiding the need to |
| 5914 | * allocate an AV. (Whether the slot holds an AV tells us whether this is |
| 5915 | * active.) |
| 5916 | */ |
| 5917 | |
| 5918 | /* A discussion about the backreferences array and its refcount: |
| 5919 | * |
| 5920 | * The AV holding the backreferences is pointed to either as the mg_obj of |
| 5921 | * PERL_MAGIC_backref, or in the specific case of a HV, from the |
| 5922 | * xhv_backreferences field. The array is created with a refcount |
| 5923 | * of 2. This means that if during global destruction the array gets |
| 5924 | * picked on before its parent to have its refcount decremented by the |
| 5925 | * random zapper, it won't actually be freed, meaning it's still there for |
| 5926 | * when its parent gets freed. |
| 5927 | * |
| 5928 | * When the parent SV is freed, the extra ref is killed by |
| 5929 | * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic, |
| 5930 | * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs. |
| 5931 | * |
| 5932 | * When a single backref SV is stored directly, it is not reference |
| 5933 | * counted. |
| 5934 | */ |
| 5935 | |
| 5936 | void |
| 5937 | Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv) |
| 5938 | { |
| 5939 | SV **svp; |
| 5940 | AV *av = NULL; |
| 5941 | MAGIC *mg = NULL; |
| 5942 | |
| 5943 | PERL_ARGS_ASSERT_SV_ADD_BACKREF; |
| 5944 | |
| 5945 | /* find slot to store array or singleton backref */ |
| 5946 | |
| 5947 | if (SvTYPE(tsv) == SVt_PVHV) { |
| 5948 | svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv)); |
| 5949 | } else { |
| 5950 | if (SvMAGICAL(tsv)) |
| 5951 | mg = mg_find(tsv, PERL_MAGIC_backref); |
| 5952 | if (!mg) |
| 5953 | mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0); |
| 5954 | svp = &(mg->mg_obj); |
| 5955 | } |
| 5956 | |
| 5957 | /* create or retrieve the array */ |
| 5958 | |
| 5959 | if ( (!*svp && SvTYPE(sv) == SVt_PVAV) |
| 5960 | || (*svp && SvTYPE(*svp) != SVt_PVAV) |
| 5961 | ) { |
| 5962 | /* create array */ |
| 5963 | if (mg) |
| 5964 | mg->mg_flags |= MGf_REFCOUNTED; |
| 5965 | av = newAV(); |
| 5966 | AvREAL_off(av); |
| 5967 | SvREFCNT_inc_simple_void_NN(av); |
| 5968 | /* av now has a refcnt of 2; see discussion above */ |
| 5969 | av_extend(av, *svp ? 2 : 1); |
| 5970 | if (*svp) { |
| 5971 | /* move single existing backref to the array */ |
| 5972 | AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */ |
| 5973 | } |
| 5974 | *svp = (SV*)av; |
| 5975 | } |
| 5976 | else { |
| 5977 | av = MUTABLE_AV(*svp); |
| 5978 | if (!av) { |
| 5979 | /* optimisation: store single backref directly in HvAUX or mg_obj */ |
| 5980 | *svp = sv; |
| 5981 | return; |
| 5982 | } |
| 5983 | assert(SvTYPE(av) == SVt_PVAV); |
| 5984 | if (AvFILLp(av) >= AvMAX(av)) { |
| 5985 | av_extend(av, AvFILLp(av)+1); |
| 5986 | } |
| 5987 | } |
| 5988 | /* push new backref */ |
| 5989 | AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */ |
| 5990 | } |
| 5991 | |
| 5992 | /* delete a back-reference to ourselves from the backref magic associated |
| 5993 | * with the SV we point to. |
| 5994 | */ |
| 5995 | |
| 5996 | void |
| 5997 | Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv) |
| 5998 | { |
| 5999 | SV **svp = NULL; |
| 6000 | |
| 6001 | PERL_ARGS_ASSERT_SV_DEL_BACKREF; |
| 6002 | |
| 6003 | if (SvTYPE(tsv) == SVt_PVHV) { |
| 6004 | if (SvOOK(tsv)) |
| 6005 | svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv)); |
| 6006 | } |
| 6007 | else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) { |
| 6008 | /* It's possible for the the last (strong) reference to tsv to have |
| 6009 | become freed *before* the last thing holding a weak reference. |
| 6010 | If both survive longer than the backreferences array, then when |
| 6011 | the referent's reference count drops to 0 and it is freed, it's |
| 6012 | not able to chase the backreferences, so they aren't NULLed. |
| 6013 | |
| 6014 | For example, a CV holds a weak reference to its stash. If both the |
| 6015 | CV and the stash survive longer than the backreferences array, |
| 6016 | and the CV gets picked for the SvBREAK() treatment first, |
| 6017 | *and* it turns out that the stash is only being kept alive because |
| 6018 | of an our variable in the pad of the CV, then midway during CV |
| 6019 | destruction the stash gets freed, but CvSTASH() isn't set to NULL. |
| 6020 | It ends up pointing to the freed HV. Hence it's chased in here, and |
| 6021 | if this block wasn't here, it would hit the !svp panic just below. |
| 6022 | |
| 6023 | I don't believe that "better" destruction ordering is going to help |
| 6024 | here - during global destruction there's always going to be the |
| 6025 | chance that something goes out of order. We've tried to make it |
| 6026 | foolproof before, and it only resulted in evolutionary pressure on |
| 6027 | fools. Which made us look foolish for our hubris. :-( |
| 6028 | */ |
| 6029 | return; |
| 6030 | } |
| 6031 | else { |
| 6032 | MAGIC *const mg |
| 6033 | = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL; |
| 6034 | svp = mg ? &(mg->mg_obj) : NULL; |
| 6035 | } |
| 6036 | |
| 6037 | if (!svp) |
| 6038 | Perl_croak(aTHX_ "panic: del_backref, svp=0"); |
| 6039 | if (!*svp) { |
| 6040 | /* It's possible that sv is being freed recursively part way through the |
| 6041 | freeing of tsv. If this happens, the backreferences array of tsv has |
| 6042 | already been freed, and so svp will be NULL. If this is the case, |
| 6043 | we should not panic. Instead, nothing needs doing, so return. */ |
| 6044 | if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0) |
| 6045 | return; |
| 6046 | Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf, |
| 6047 | (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv)); |
| 6048 | } |
| 6049 | |
| 6050 | if (SvTYPE(*svp) == SVt_PVAV) { |
| 6051 | #ifdef DEBUGGING |
| 6052 | int count = 1; |
| 6053 | #endif |
| 6054 | AV * const av = (AV*)*svp; |
| 6055 | SSize_t fill; |
| 6056 | assert(!SvIS_FREED(av)); |
| 6057 | fill = AvFILLp(av); |
| 6058 | assert(fill > -1); |
| 6059 | svp = AvARRAY(av); |
| 6060 | /* for an SV with N weak references to it, if all those |
| 6061 | * weak refs are deleted, then sv_del_backref will be called |
| 6062 | * N times and O(N^2) compares will be done within the backref |
| 6063 | * array. To ameliorate this potential slowness, we: |
| 6064 | * 1) make sure this code is as tight as possible; |
| 6065 | * 2) when looking for SV, look for it at both the head and tail of the |
| 6066 | * array first before searching the rest, since some create/destroy |
| 6067 | * patterns will cause the backrefs to be freed in order. |
| 6068 | */ |
| 6069 | if (*svp == sv) { |
| 6070 | AvARRAY(av)++; |
| 6071 | AvMAX(av)--; |
| 6072 | } |
| 6073 | else { |
| 6074 | SV **p = &svp[fill]; |
| 6075 | SV *const topsv = *p; |
| 6076 | if (topsv != sv) { |
| 6077 | #ifdef DEBUGGING |
| 6078 | count = 0; |
| 6079 | #endif |
| 6080 | while (--p > svp) { |
| 6081 | if (*p == sv) { |
| 6082 | /* We weren't the last entry. |
| 6083 | An unordered list has this property that you |
| 6084 | can take the last element off the end to fill |
| 6085 | the hole, and it's still an unordered list :-) |
| 6086 | */ |
| 6087 | *p = topsv; |
| 6088 | #ifdef DEBUGGING |
| 6089 | count++; |
| 6090 | #else |
| 6091 | break; /* should only be one */ |
| 6092 | #endif |
| 6093 | } |
| 6094 | } |
| 6095 | } |
| 6096 | } |
| 6097 | assert(count ==1); |
| 6098 | AvFILLp(av) = fill-1; |
| 6099 | } |
| 6100 | else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) { |
| 6101 | /* freed AV; skip */ |
| 6102 | } |
| 6103 | else { |
| 6104 | /* optimisation: only a single backref, stored directly */ |
| 6105 | if (*svp != sv) |
| 6106 | Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p", |
| 6107 | (void*)*svp, (void*)sv); |
| 6108 | *svp = NULL; |
| 6109 | } |
| 6110 | |
| 6111 | } |
| 6112 | |
| 6113 | void |
| 6114 | Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av) |
| 6115 | { |
| 6116 | SV **svp; |
| 6117 | SV **last; |
| 6118 | bool is_array; |
| 6119 | |
| 6120 | PERL_ARGS_ASSERT_SV_KILL_BACKREFS; |
| 6121 | |
| 6122 | if (!av) |
| 6123 | return; |
| 6124 | |
| 6125 | /* after multiple passes through Perl_sv_clean_all() for a thingy |
| 6126 | * that has badly leaked, the backref array may have gotten freed, |
| 6127 | * since we only protect it against 1 round of cleanup */ |
| 6128 | if (SvIS_FREED(av)) { |
| 6129 | if (PL_in_clean_all) /* All is fair */ |
| 6130 | return; |
| 6131 | Perl_croak(aTHX_ |
| 6132 | "panic: magic_killbackrefs (freed backref AV/SV)"); |
| 6133 | } |
| 6134 | |
| 6135 | |
| 6136 | is_array = (SvTYPE(av) == SVt_PVAV); |
| 6137 | if (is_array) { |
| 6138 | assert(!SvIS_FREED(av)); |
| 6139 | svp = AvARRAY(av); |
| 6140 | if (svp) |
| 6141 | last = svp + AvFILLp(av); |
| 6142 | } |
| 6143 | else { |
| 6144 | /* optimisation: only a single backref, stored directly */ |
| 6145 | svp = (SV**)&av; |
| 6146 | last = svp; |
| 6147 | } |
| 6148 | |
| 6149 | if (svp) { |
| 6150 | while (svp <= last) { |
| 6151 | if (*svp) { |
| 6152 | SV *const referrer = *svp; |
| 6153 | if (SvWEAKREF(referrer)) { |
| 6154 | /* XXX Should we check that it hasn't changed? */ |
| 6155 | assert(SvROK(referrer)); |
| 6156 | SvRV_set(referrer, 0); |
| 6157 | SvOK_off(referrer); |
| 6158 | SvWEAKREF_off(referrer); |
| 6159 | SvSETMAGIC(referrer); |
| 6160 | } else if (SvTYPE(referrer) == SVt_PVGV || |
| 6161 | SvTYPE(referrer) == SVt_PVLV) { |
| 6162 | assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */ |
| 6163 | /* You lookin' at me? */ |
| 6164 | assert(GvSTASH(referrer)); |
| 6165 | assert(GvSTASH(referrer) == (const HV *)sv); |
| 6166 | GvSTASH(referrer) = 0; |
| 6167 | } else if (SvTYPE(referrer) == SVt_PVCV || |
| 6168 | SvTYPE(referrer) == SVt_PVFM) { |
| 6169 | if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */ |
| 6170 | /* You lookin' at me? */ |
| 6171 | assert(CvSTASH(referrer)); |
| 6172 | assert(CvSTASH(referrer) == (const HV *)sv); |
| 6173 | SvANY(MUTABLE_CV(referrer))->xcv_stash = 0; |
| 6174 | } |
| 6175 | else { |
| 6176 | assert(SvTYPE(sv) == SVt_PVGV); |
| 6177 | /* You lookin' at me? */ |
| 6178 | assert(CvGV(referrer)); |
| 6179 | assert(CvGV(referrer) == (const GV *)sv); |
| 6180 | anonymise_cv_maybe(MUTABLE_GV(sv), |
| 6181 | MUTABLE_CV(referrer)); |
| 6182 | } |
| 6183 | |
| 6184 | } else { |
| 6185 | Perl_croak(aTHX_ |
| 6186 | "panic: magic_killbackrefs (flags=%"UVxf")", |
| 6187 | (UV)SvFLAGS(referrer)); |
| 6188 | } |
| 6189 | |
| 6190 | if (is_array) |
| 6191 | *svp = NULL; |
| 6192 | } |
| 6193 | svp++; |
| 6194 | } |
| 6195 | } |
| 6196 | if (is_array) { |
| 6197 | AvFILLp(av) = -1; |
| 6198 | SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */ |
| 6199 | } |
| 6200 | return; |
| 6201 | } |
| 6202 | |
| 6203 | /* |
| 6204 | =for apidoc sv_insert |
| 6205 | |
| 6206 | Inserts a string at the specified offset/length within the SV. Similar to |
| 6207 | the Perl C<substr()> function. Handles get magic. |
| 6208 | |
| 6209 | =for apidoc sv_insert_flags |
| 6210 | |
| 6211 | Same as C<sv_insert>, but the extra C<flags> are passed to the |
| 6212 | C<SvPV_force_flags> that applies to C<bigstr>. |
| 6213 | |
| 6214 | =cut |
| 6215 | */ |
| 6216 | |
| 6217 | void |
| 6218 | Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags) |
| 6219 | { |
| 6220 | char *big; |
| 6221 | char *mid; |
| 6222 | char *midend; |
| 6223 | char *bigend; |
| 6224 | SSize_t i; /* better be sizeof(STRLEN) or bad things happen */ |
| 6225 | STRLEN curlen; |
| 6226 | |
| 6227 | PERL_ARGS_ASSERT_SV_INSERT_FLAGS; |
| 6228 | |
| 6229 | SvPV_force_flags(bigstr, curlen, flags); |
| 6230 | (void)SvPOK_only_UTF8(bigstr); |
| 6231 | if (offset + len > curlen) { |
| 6232 | SvGROW(bigstr, offset+len+1); |
| 6233 | Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char); |
| 6234 | SvCUR_set(bigstr, offset+len); |
| 6235 | } |
| 6236 | |
| 6237 | SvTAINT(bigstr); |
| 6238 | i = littlelen - len; |
| 6239 | if (i > 0) { /* string might grow */ |
| 6240 | big = SvGROW(bigstr, SvCUR(bigstr) + i + 1); |
| 6241 | mid = big + offset + len; |
| 6242 | midend = bigend = big + SvCUR(bigstr); |
| 6243 | bigend += i; |
| 6244 | *bigend = '\0'; |
| 6245 | while (midend > mid) /* shove everything down */ |
| 6246 | *--bigend = *--midend; |
| 6247 | Move(little,big+offset,littlelen,char); |
| 6248 | SvCUR_set(bigstr, SvCUR(bigstr) + i); |
| 6249 | SvSETMAGIC(bigstr); |
| 6250 | return; |
| 6251 | } |
| 6252 | else if (i == 0) { |
| 6253 | Move(little,SvPVX(bigstr)+offset,len,char); |
| 6254 | SvSETMAGIC(bigstr); |
| 6255 | return; |
| 6256 | } |
| 6257 | |
| 6258 | big = SvPVX(bigstr); |
| 6259 | mid = big + offset; |
| 6260 | midend = mid + len; |
| 6261 | bigend = big + SvCUR(bigstr); |
| 6262 | |
| 6263 | if (midend > bigend) |
| 6264 | Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p", |
| 6265 | midend, bigend); |
| 6266 | |
| 6267 | if (mid - big > bigend - midend) { /* faster to shorten from end */ |
| 6268 | if (littlelen) { |
| 6269 | Move(little, mid, littlelen,char); |
| 6270 | mid += littlelen; |
| 6271 | } |
| 6272 | i = bigend - midend; |
| 6273 | if (i > 0) { |
| 6274 | Move(midend, mid, i,char); |
| 6275 | mid += i; |
| 6276 | } |
| 6277 | *mid = '\0'; |
| 6278 | SvCUR_set(bigstr, mid - big); |
| 6279 | } |
| 6280 | else if ((i = mid - big)) { /* faster from front */ |
| 6281 | midend -= littlelen; |
| 6282 | mid = midend; |
| 6283 | Move(big, midend - i, i, char); |
| 6284 | sv_chop(bigstr,midend-i); |
| 6285 | if (littlelen) |
| 6286 | Move(little, mid, littlelen,char); |
| 6287 | } |
| 6288 | else if (littlelen) { |
| 6289 | midend -= littlelen; |
| 6290 | sv_chop(bigstr,midend); |
| 6291 | Move(little,midend,littlelen,char); |
| 6292 | } |
| 6293 | else { |
| 6294 | sv_chop(bigstr,midend); |
| 6295 | } |
| 6296 | SvSETMAGIC(bigstr); |
| 6297 | } |
| 6298 | |
| 6299 | /* |
| 6300 | =for apidoc sv_replace |
| 6301 | |
| 6302 | Make the first argument a copy of the second, then delete the original. |
| 6303 | The target SV physically takes over ownership of the body of the source SV |
| 6304 | and inherits its flags; however, the target keeps any magic it owns, |
| 6305 | and any magic in the source is discarded. |
| 6306 | Note that this is a rather specialist SV copying operation; most of the |
| 6307 | time you'll want to use C<sv_setsv> or one of its many macro front-ends. |
| 6308 | |
| 6309 | =cut |
| 6310 | */ |
| 6311 | |
| 6312 | void |
| 6313 | Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv) |
| 6314 | { |
| 6315 | const U32 refcnt = SvREFCNT(sv); |
| 6316 | |
| 6317 | PERL_ARGS_ASSERT_SV_REPLACE; |
| 6318 | |
| 6319 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 6320 | if (SvREFCNT(nsv) != 1) { |
| 6321 | Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()" |
| 6322 | " (%" UVuf " != 1)", (UV) SvREFCNT(nsv)); |
| 6323 | } |
| 6324 | if (SvMAGICAL(sv)) { |
| 6325 | if (SvMAGICAL(nsv)) |
| 6326 | mg_free(nsv); |
| 6327 | else |
| 6328 | sv_upgrade(nsv, SVt_PVMG); |
| 6329 | SvMAGIC_set(nsv, SvMAGIC(sv)); |
| 6330 | SvFLAGS(nsv) |= SvMAGICAL(sv); |
| 6331 | SvMAGICAL_off(sv); |
| 6332 | SvMAGIC_set(sv, NULL); |
| 6333 | } |
| 6334 | SvREFCNT(sv) = 0; |
| 6335 | sv_clear(sv); |
| 6336 | assert(!SvREFCNT(sv)); |
| 6337 | #ifdef DEBUG_LEAKING_SCALARS |
| 6338 | sv->sv_flags = nsv->sv_flags; |
| 6339 | sv->sv_any = nsv->sv_any; |
| 6340 | sv->sv_refcnt = nsv->sv_refcnt; |
| 6341 | sv->sv_u = nsv->sv_u; |
| 6342 | #else |
| 6343 | StructCopy(nsv,sv,SV); |
| 6344 | #endif |
| 6345 | if(SvTYPE(sv) == SVt_IV) { |
| 6346 | SET_SVANY_FOR_BODYLESS_IV(sv); |
| 6347 | } |
| 6348 | |
| 6349 | |
| 6350 | SvREFCNT(sv) = refcnt; |
| 6351 | SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */ |
| 6352 | SvREFCNT(nsv) = 0; |
| 6353 | del_SV(nsv); |
| 6354 | } |
| 6355 | |
| 6356 | /* We're about to free a GV which has a CV that refers back to us. |
| 6357 | * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV |
| 6358 | * field) */ |
| 6359 | |
| 6360 | STATIC void |
| 6361 | S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv) |
| 6362 | { |
| 6363 | SV *gvname; |
| 6364 | GV *anongv; |
| 6365 | |
| 6366 | PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE; |
| 6367 | |
| 6368 | /* be assertive! */ |
| 6369 | assert(SvREFCNT(gv) == 0); |
| 6370 | assert(isGV(gv) && isGV_with_GP(gv)); |
| 6371 | assert(GvGP(gv)); |
| 6372 | assert(!CvANON(cv)); |
| 6373 | assert(CvGV(cv) == gv); |
| 6374 | assert(!CvNAMED(cv)); |
| 6375 | |
| 6376 | /* will the CV shortly be freed by gp_free() ? */ |
| 6377 | if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) { |
| 6378 | SvANY(cv)->xcv_gv_u.xcv_gv = NULL; |
| 6379 | return; |
| 6380 | } |
| 6381 | |
| 6382 | /* if not, anonymise: */ |
| 6383 | gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv))) |
| 6384 | ? newSVhek(HvENAME_HEK(GvSTASH(gv))) |
| 6385 | : newSVpvn_flags( "__ANON__", 8, 0 ); |
| 6386 | sv_catpvs(gvname, "::__ANON__"); |
| 6387 | anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV); |
| 6388 | SvREFCNT_dec_NN(gvname); |
| 6389 | |
| 6390 | CvANON_on(cv); |
| 6391 | CvCVGV_RC_on(cv); |
| 6392 | SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv)); |
| 6393 | } |
| 6394 | |
| 6395 | |
| 6396 | /* |
| 6397 | =for apidoc sv_clear |
| 6398 | |
| 6399 | Clear an SV: call any destructors, free up any memory used by the body, |
| 6400 | and free the body itself. The SV's head is I<not> freed, although |
| 6401 | its type is set to all 1's so that it won't inadvertently be assumed |
| 6402 | to be live during global destruction etc. |
| 6403 | This function should only be called when C<REFCNT> is zero. Most of the time |
| 6404 | you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>) |
| 6405 | instead. |
| 6406 | |
| 6407 | =cut |
| 6408 | */ |
| 6409 | |
| 6410 | void |
| 6411 | Perl_sv_clear(pTHX_ SV *const orig_sv) |
| 6412 | { |
| 6413 | dVAR; |
| 6414 | HV *stash; |
| 6415 | U32 type; |
| 6416 | const struct body_details *sv_type_details; |
| 6417 | SV* iter_sv = NULL; |
| 6418 | SV* next_sv = NULL; |
| 6419 | SV *sv = orig_sv; |
| 6420 | STRLEN hash_index = 0; /* initialise to make Coverity et al happy. |
| 6421 | Not strictly necessary */ |
| 6422 | |
| 6423 | PERL_ARGS_ASSERT_SV_CLEAR; |
| 6424 | |
| 6425 | /* within this loop, sv is the SV currently being freed, and |
| 6426 | * iter_sv is the most recent AV or whatever that's being iterated |
| 6427 | * over to provide more SVs */ |
| 6428 | |
| 6429 | while (sv) { |
| 6430 | |
| 6431 | type = SvTYPE(sv); |
| 6432 | |
| 6433 | assert(SvREFCNT(sv) == 0); |
| 6434 | assert(SvTYPE(sv) != (svtype)SVTYPEMASK); |
| 6435 | |
| 6436 | if (type <= SVt_IV) { |
| 6437 | /* See the comment in sv.h about the collusion between this |
| 6438 | * early return and the overloading of the NULL slots in the |
| 6439 | * size table. */ |
| 6440 | if (SvROK(sv)) |
| 6441 | goto free_rv; |
| 6442 | SvFLAGS(sv) &= SVf_BREAK; |
| 6443 | SvFLAGS(sv) |= SVTYPEMASK; |
| 6444 | goto free_head; |
| 6445 | } |
| 6446 | |
| 6447 | /* objs are always >= MG, but pad names use the SVs_OBJECT flag |
| 6448 | for another purpose */ |
| 6449 | assert(!SvOBJECT(sv) || type >= SVt_PVMG); |
| 6450 | |
| 6451 | if (type >= SVt_PVMG) { |
| 6452 | if (SvOBJECT(sv)) { |
| 6453 | if (!curse(sv, 1)) goto get_next_sv; |
| 6454 | type = SvTYPE(sv); /* destructor may have changed it */ |
| 6455 | } |
| 6456 | /* Free back-references before magic, in case the magic calls |
| 6457 | * Perl code that has weak references to sv. */ |
| 6458 | if (type == SVt_PVHV) { |
| 6459 | Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv)); |
| 6460 | if (SvMAGIC(sv)) |
| 6461 | mg_free(sv); |
| 6462 | } |
| 6463 | else if (SvMAGIC(sv)) { |
| 6464 | /* Free back-references before other types of magic. */ |
| 6465 | sv_unmagic(sv, PERL_MAGIC_backref); |
| 6466 | mg_free(sv); |
| 6467 | } |
| 6468 | SvMAGICAL_off(sv); |
| 6469 | } |
| 6470 | switch (type) { |
| 6471 | /* case SVt_INVLIST: */ |
| 6472 | case SVt_PVIO: |
| 6473 | if (IoIFP(sv) && |
| 6474 | IoIFP(sv) != PerlIO_stdin() && |
| 6475 | IoIFP(sv) != PerlIO_stdout() && |
| 6476 | IoIFP(sv) != PerlIO_stderr() && |
| 6477 | !(IoFLAGS(sv) & IOf_FAKE_DIRP)) |
| 6478 | { |
| 6479 | io_close(MUTABLE_IO(sv), NULL, FALSE, |
| 6480 | (IoTYPE(sv) == IoTYPE_WRONLY || |
| 6481 | IoTYPE(sv) == IoTYPE_RDWR || |
| 6482 | IoTYPE(sv) == IoTYPE_APPEND)); |
| 6483 | } |
| 6484 | if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP)) |
| 6485 | PerlDir_close(IoDIRP(sv)); |
| 6486 | IoDIRP(sv) = (DIR*)NULL; |
| 6487 | Safefree(IoTOP_NAME(sv)); |
| 6488 | Safefree(IoFMT_NAME(sv)); |
| 6489 | Safefree(IoBOTTOM_NAME(sv)); |
| 6490 | if ((const GV *)sv == PL_statgv) |
| 6491 | PL_statgv = NULL; |
| 6492 | goto freescalar; |
| 6493 | case SVt_REGEXP: |
| 6494 | /* FIXME for plugins */ |
| 6495 | freeregexp: |
| 6496 | pregfree2((REGEXP*) sv); |
| 6497 | goto freescalar; |
| 6498 | case SVt_PVCV: |
| 6499 | case SVt_PVFM: |
| 6500 | cv_undef(MUTABLE_CV(sv)); |
| 6501 | /* If we're in a stash, we don't own a reference to it. |
| 6502 | * However it does have a back reference to us, which needs to |
| 6503 | * be cleared. */ |
| 6504 | if ((stash = CvSTASH(sv))) |
| 6505 | sv_del_backref(MUTABLE_SV(stash), sv); |
| 6506 | goto freescalar; |
| 6507 | case SVt_PVHV: |
| 6508 | if (PL_last_swash_hv == (const HV *)sv) { |
| 6509 | PL_last_swash_hv = NULL; |
| 6510 | } |
| 6511 | if (HvTOTALKEYS((HV*)sv) > 0) { |
| 6512 | const HEK *hek; |
| 6513 | /* this statement should match the one at the beginning of |
| 6514 | * hv_undef_flags() */ |
| 6515 | if ( PL_phase != PERL_PHASE_DESTRUCT |
| 6516 | && (hek = HvNAME_HEK((HV*)sv))) |
| 6517 | { |
| 6518 | if (PL_stashcache) { |
| 6519 | DEBUG_o(Perl_deb(aTHX_ |
| 6520 | "sv_clear clearing PL_stashcache for '%"HEKf |
| 6521 | "'\n", |
| 6522 | HEKfARG(hek))); |
| 6523 | (void)hv_deletehek(PL_stashcache, |
| 6524 | hek, G_DISCARD); |
| 6525 | } |
| 6526 | hv_name_set((HV*)sv, NULL, 0, 0); |
| 6527 | } |
| 6528 | |
| 6529 | /* save old iter_sv in unused SvSTASH field */ |
| 6530 | assert(!SvOBJECT(sv)); |
| 6531 | SvSTASH(sv) = (HV*)iter_sv; |
| 6532 | iter_sv = sv; |
| 6533 | |
| 6534 | /* save old hash_index in unused SvMAGIC field */ |
| 6535 | assert(!SvMAGICAL(sv)); |
| 6536 | assert(!SvMAGIC(sv)); |
| 6537 | ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index; |
| 6538 | hash_index = 0; |
| 6539 | |
| 6540 | next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index); |
| 6541 | goto get_next_sv; /* process this new sv */ |
| 6542 | } |
| 6543 | /* free empty hash */ |
| 6544 | Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL); |
| 6545 | assert(!HvARRAY((HV*)sv)); |
| 6546 | break; |
| 6547 | case SVt_PVAV: |
| 6548 | { |
| 6549 | AV* av = MUTABLE_AV(sv); |
| 6550 | if (PL_comppad == av) { |
| 6551 | PL_comppad = NULL; |
| 6552 | PL_curpad = NULL; |
| 6553 | } |
| 6554 | if (AvREAL(av) && AvFILLp(av) > -1) { |
| 6555 | next_sv = AvARRAY(av)[AvFILLp(av)--]; |
| 6556 | /* save old iter_sv in top-most slot of AV, |
| 6557 | * and pray that it doesn't get wiped in the meantime */ |
| 6558 | AvARRAY(av)[AvMAX(av)] = iter_sv; |
| 6559 | iter_sv = sv; |
| 6560 | goto get_next_sv; /* process this new sv */ |
| 6561 | } |
| 6562 | Safefree(AvALLOC(av)); |
| 6563 | } |
| 6564 | |
| 6565 | break; |
| 6566 | case SVt_PVLV: |
| 6567 | if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */ |
| 6568 | SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv))); |
| 6569 | HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh; |
| 6570 | PL_hv_fetch_ent_mh = (HE*)LvTARG(sv); |
| 6571 | } |
| 6572 | else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */ |
| 6573 | SvREFCNT_dec(LvTARG(sv)); |
| 6574 | if (isREGEXP(sv)) goto freeregexp; |
| 6575 | /* FALLTHROUGH */ |
| 6576 | case SVt_PVGV: |
| 6577 | if (isGV_with_GP(sv)) { |
| 6578 | if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv))) |
| 6579 | && HvENAME_get(stash)) |
| 6580 | mro_method_changed_in(stash); |
| 6581 | gp_free(MUTABLE_GV(sv)); |
| 6582 | if (GvNAME_HEK(sv)) |
| 6583 | unshare_hek(GvNAME_HEK(sv)); |
| 6584 | /* If we're in a stash, we don't own a reference to it. |
| 6585 | * However it does have a back reference to us, which |
| 6586 | * needs to be cleared. */ |
| 6587 | if (!SvVALID(sv) && (stash = GvSTASH(sv))) |
| 6588 | sv_del_backref(MUTABLE_SV(stash), sv); |
| 6589 | } |
| 6590 | /* FIXME. There are probably more unreferenced pointers to SVs |
| 6591 | * in the interpreter struct that we should check and tidy in |
| 6592 | * a similar fashion to this: */ |
| 6593 | /* See also S_sv_unglob, which does the same thing. */ |
| 6594 | if ((const GV *)sv == PL_last_in_gv) |
| 6595 | PL_last_in_gv = NULL; |
| 6596 | else if ((const GV *)sv == PL_statgv) |
| 6597 | PL_statgv = NULL; |
| 6598 | else if ((const GV *)sv == PL_stderrgv) |
| 6599 | PL_stderrgv = NULL; |
| 6600 | /* FALLTHROUGH */ |
| 6601 | case SVt_PVMG: |
| 6602 | case SVt_PVNV: |
| 6603 | case SVt_PVIV: |
| 6604 | case SVt_INVLIST: |
| 6605 | case SVt_PV: |
| 6606 | freescalar: |
| 6607 | /* Don't bother with SvOOK_off(sv); as we're only going to |
| 6608 | * free it. */ |
| 6609 | if (SvOOK(sv)) { |
| 6610 | STRLEN offset; |
| 6611 | SvOOK_offset(sv, offset); |
| 6612 | SvPV_set(sv, SvPVX_mutable(sv) - offset); |
| 6613 | /* Don't even bother with turning off the OOK flag. */ |
| 6614 | } |
| 6615 | if (SvROK(sv)) { |
| 6616 | free_rv: |
| 6617 | { |
| 6618 | SV * const target = SvRV(sv); |
| 6619 | if (SvWEAKREF(sv)) |
| 6620 | sv_del_backref(target, sv); |
| 6621 | else |
| 6622 | next_sv = target; |
| 6623 | } |
| 6624 | } |
| 6625 | #ifdef PERL_ANY_COW |
| 6626 | else if (SvPVX_const(sv) |
| 6627 | && !(SvTYPE(sv) == SVt_PVIO |
| 6628 | && !(IoFLAGS(sv) & IOf_FAKE_DIRP))) |
| 6629 | { |
| 6630 | if (SvIsCOW(sv)) { |
| 6631 | if (DEBUG_C_TEST) { |
| 6632 | PerlIO_printf(Perl_debug_log, "Copy on write: clear\n"); |
| 6633 | sv_dump(sv); |
| 6634 | } |
| 6635 | if (SvLEN(sv)) { |
| 6636 | if (CowREFCNT(sv)) { |
| 6637 | sv_buf_to_rw(sv); |
| 6638 | CowREFCNT(sv)--; |
| 6639 | sv_buf_to_ro(sv); |
| 6640 | SvLEN_set(sv, 0); |
| 6641 | } |
| 6642 | } else { |
| 6643 | unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv))); |
| 6644 | } |
| 6645 | |
| 6646 | } |
| 6647 | if (SvLEN(sv)) { |
| 6648 | Safefree(SvPVX_mutable(sv)); |
| 6649 | } |
| 6650 | } |
| 6651 | #else |
| 6652 | else if (SvPVX_const(sv) && SvLEN(sv) |
| 6653 | && !(SvTYPE(sv) == SVt_PVIO |
| 6654 | && !(IoFLAGS(sv) & IOf_FAKE_DIRP))) |
| 6655 | Safefree(SvPVX_mutable(sv)); |
| 6656 | else if (SvPVX_const(sv) && SvIsCOW(sv)) { |
| 6657 | unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv))); |
| 6658 | } |
| 6659 | #endif |
| 6660 | break; |
| 6661 | case SVt_NV: |
| 6662 | break; |
| 6663 | } |
| 6664 | |
| 6665 | free_body: |
| 6666 | |
| 6667 | SvFLAGS(sv) &= SVf_BREAK; |
| 6668 | SvFLAGS(sv) |= SVTYPEMASK; |
| 6669 | |
| 6670 | sv_type_details = bodies_by_type + type; |
| 6671 | if (sv_type_details->arena) { |
| 6672 | del_body(((char *)SvANY(sv) + sv_type_details->offset), |
| 6673 | &PL_body_roots[type]); |
| 6674 | } |
| 6675 | else if (sv_type_details->body_size) { |
| 6676 | safefree(SvANY(sv)); |
| 6677 | } |
| 6678 | |
| 6679 | free_head: |
| 6680 | /* caller is responsible for freeing the head of the original sv */ |
| 6681 | if (sv != orig_sv && !SvREFCNT(sv)) |
| 6682 | del_SV(sv); |
| 6683 | |
| 6684 | /* grab and free next sv, if any */ |
| 6685 | get_next_sv: |
| 6686 | while (1) { |
| 6687 | sv = NULL; |
| 6688 | if (next_sv) { |
| 6689 | sv = next_sv; |
| 6690 | next_sv = NULL; |
| 6691 | } |
| 6692 | else if (!iter_sv) { |
| 6693 | break; |
| 6694 | } else if (SvTYPE(iter_sv) == SVt_PVAV) { |
| 6695 | AV *const av = (AV*)iter_sv; |
| 6696 | if (AvFILLp(av) > -1) { |
| 6697 | sv = AvARRAY(av)[AvFILLp(av)--]; |
| 6698 | } |
| 6699 | else { /* no more elements of current AV to free */ |
| 6700 | sv = iter_sv; |
| 6701 | type = SvTYPE(sv); |
| 6702 | /* restore previous value, squirrelled away */ |
| 6703 | iter_sv = AvARRAY(av)[AvMAX(av)]; |
| 6704 | Safefree(AvALLOC(av)); |
| 6705 | goto free_body; |
| 6706 | } |
| 6707 | } else if (SvTYPE(iter_sv) == SVt_PVHV) { |
| 6708 | sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index); |
| 6709 | if (!sv && !HvTOTALKEYS((HV *)iter_sv)) { |
| 6710 | /* no more elements of current HV to free */ |
| 6711 | sv = iter_sv; |
| 6712 | type = SvTYPE(sv); |
| 6713 | /* Restore previous values of iter_sv and hash_index, |
| 6714 | * squirrelled away */ |
| 6715 | assert(!SvOBJECT(sv)); |
| 6716 | iter_sv = (SV*)SvSTASH(sv); |
| 6717 | assert(!SvMAGICAL(sv)); |
| 6718 | hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index; |
| 6719 | #ifdef DEBUGGING |
| 6720 | /* perl -DA does not like rubbish in SvMAGIC. */ |
| 6721 | SvMAGIC_set(sv, 0); |
| 6722 | #endif |
| 6723 | |
| 6724 | /* free any remaining detritus from the hash struct */ |
| 6725 | Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL); |
| 6726 | assert(!HvARRAY((HV*)sv)); |
| 6727 | goto free_body; |
| 6728 | } |
| 6729 | } |
| 6730 | |
| 6731 | /* unrolled SvREFCNT_dec and sv_free2 follows: */ |
| 6732 | |
| 6733 | if (!sv) |
| 6734 | continue; |
| 6735 | if (!SvREFCNT(sv)) { |
| 6736 | sv_free(sv); |
| 6737 | continue; |
| 6738 | } |
| 6739 | if (--(SvREFCNT(sv))) |
| 6740 | continue; |
| 6741 | #ifdef DEBUGGING |
| 6742 | if (SvTEMP(sv)) { |
| 6743 | Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING), |
| 6744 | "Attempt to free temp prematurely: SV 0x%"UVxf |
| 6745 | pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); |
| 6746 | continue; |
| 6747 | } |
| 6748 | #endif |
| 6749 | if (SvIMMORTAL(sv)) { |
| 6750 | /* make sure SvREFCNT(sv)==0 happens very seldom */ |
| 6751 | SvREFCNT(sv) = SvREFCNT_IMMORTAL; |
| 6752 | continue; |
| 6753 | } |
| 6754 | break; |
| 6755 | } /* while 1 */ |
| 6756 | |
| 6757 | } /* while sv */ |
| 6758 | } |
| 6759 | |
| 6760 | /* This routine curses the sv itself, not the object referenced by sv. So |
| 6761 | sv does not have to be ROK. */ |
| 6762 | |
| 6763 | static bool |
| 6764 | S_curse(pTHX_ SV * const sv, const bool check_refcnt) { |
| 6765 | PERL_ARGS_ASSERT_CURSE; |
| 6766 | assert(SvOBJECT(sv)); |
| 6767 | |
| 6768 | if (PL_defstash && /* Still have a symbol table? */ |
| 6769 | SvDESTROYABLE(sv)) |
| 6770 | { |
| 6771 | dSP; |
| 6772 | HV* stash; |
| 6773 | do { |
| 6774 | stash = SvSTASH(sv); |
| 6775 | assert(SvTYPE(stash) == SVt_PVHV); |
| 6776 | if (HvNAME(stash)) { |
| 6777 | CV* destructor = NULL; |
| 6778 | assert (SvOOK(stash)); |
| 6779 | if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash); |
| 6780 | if (!destructor || HvMROMETA(stash)->destroy_gen |
| 6781 | != PL_sub_generation) |
| 6782 | { |
| 6783 | GV * const gv = |
| 6784 | gv_fetchmeth_autoload(stash, "DESTROY", 7, 0); |
| 6785 | if (gv) destructor = GvCV(gv); |
| 6786 | if (!SvOBJECT(stash)) |
| 6787 | { |
| 6788 | SvSTASH(stash) = |
| 6789 | destructor ? (HV *)destructor : ((HV *)0)+1; |
| 6790 | HvAUX(stash)->xhv_mro_meta->destroy_gen = |
| 6791 | PL_sub_generation; |
| 6792 | } |
| 6793 | } |
| 6794 | assert(!destructor || destructor == ((CV *)0)+1 |
| 6795 | || SvTYPE(destructor) == SVt_PVCV); |
| 6796 | if (destructor && destructor != ((CV *)0)+1 |
| 6797 | /* A constant subroutine can have no side effects, so |
| 6798 | don't bother calling it. */ |
| 6799 | && !CvCONST(destructor) |
| 6800 | /* Don't bother calling an empty destructor or one that |
| 6801 | returns immediately. */ |
| 6802 | && (CvISXSUB(destructor) |
| 6803 | || (CvSTART(destructor) |
| 6804 | && (CvSTART(destructor)->op_next->op_type |
| 6805 | != OP_LEAVESUB) |
| 6806 | && (CvSTART(destructor)->op_next->op_type |
| 6807 | != OP_PUSHMARK |
| 6808 | || CvSTART(destructor)->op_next->op_next->op_type |
| 6809 | != OP_RETURN |
| 6810 | ) |
| 6811 | )) |
| 6812 | ) |
| 6813 | { |
| 6814 | SV* const tmpref = newRV(sv); |
| 6815 | SvREADONLY_on(tmpref); /* DESTROY() could be naughty */ |
| 6816 | ENTER; |
| 6817 | PUSHSTACKi(PERLSI_DESTROY); |
| 6818 | EXTEND(SP, 2); |
| 6819 | PUSHMARK(SP); |
| 6820 | PUSHs(tmpref); |
| 6821 | PUTBACK; |
| 6822 | call_sv(MUTABLE_SV(destructor), |
| 6823 | G_DISCARD|G_EVAL|G_KEEPERR|G_VOID); |
| 6824 | POPSTACK; |
| 6825 | SPAGAIN; |
| 6826 | LEAVE; |
| 6827 | if(SvREFCNT(tmpref) < 2) { |
| 6828 | /* tmpref is not kept alive! */ |
| 6829 | SvREFCNT(sv)--; |
| 6830 | SvRV_set(tmpref, NULL); |
| 6831 | SvROK_off(tmpref); |
| 6832 | } |
| 6833 | SvREFCNT_dec_NN(tmpref); |
| 6834 | } |
| 6835 | } |
| 6836 | } while (SvOBJECT(sv) && SvSTASH(sv) != stash); |
| 6837 | |
| 6838 | |
| 6839 | if (check_refcnt && SvREFCNT(sv)) { |
| 6840 | if (PL_in_clean_objs) |
| 6841 | Perl_croak(aTHX_ |
| 6842 | "DESTROY created new reference to dead object '%"HEKf"'", |
| 6843 | HEKfARG(HvNAME_HEK(stash))); |
| 6844 | /* DESTROY gave object new lease on life */ |
| 6845 | return FALSE; |
| 6846 | } |
| 6847 | } |
| 6848 | |
| 6849 | if (SvOBJECT(sv)) { |
| 6850 | HV * const stash = SvSTASH(sv); |
| 6851 | /* Curse before freeing the stash, as freeing the stash could cause |
| 6852 | a recursive call into S_curse. */ |
| 6853 | SvOBJECT_off(sv); /* Curse the object. */ |
| 6854 | SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */ |
| 6855 | SvREFCNT_dec(stash); /* possibly of changed persuasion */ |
| 6856 | } |
| 6857 | return TRUE; |
| 6858 | } |
| 6859 | |
| 6860 | /* |
| 6861 | =for apidoc sv_newref |
| 6862 | |
| 6863 | Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper |
| 6864 | instead. |
| 6865 | |
| 6866 | =cut |
| 6867 | */ |
| 6868 | |
| 6869 | SV * |
| 6870 | Perl_sv_newref(pTHX_ SV *const sv) |
| 6871 | { |
| 6872 | PERL_UNUSED_CONTEXT; |
| 6873 | if (sv) |
| 6874 | (SvREFCNT(sv))++; |
| 6875 | return sv; |
| 6876 | } |
| 6877 | |
| 6878 | /* |
| 6879 | =for apidoc sv_free |
| 6880 | |
| 6881 | Decrement an SV's reference count, and if it drops to zero, call |
| 6882 | C<sv_clear> to invoke destructors and free up any memory used by |
| 6883 | the body; finally, deallocating the SV's head itself. |
| 6884 | Normally called via a wrapper macro C<SvREFCNT_dec>. |
| 6885 | |
| 6886 | =cut |
| 6887 | */ |
| 6888 | |
| 6889 | void |
| 6890 | Perl_sv_free(pTHX_ SV *const sv) |
| 6891 | { |
| 6892 | SvREFCNT_dec(sv); |
| 6893 | } |
| 6894 | |
| 6895 | |
| 6896 | /* Private helper function for SvREFCNT_dec(). |
| 6897 | * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */ |
| 6898 | |
| 6899 | void |
| 6900 | Perl_sv_free2(pTHX_ SV *const sv, const U32 rc) |
| 6901 | { |
| 6902 | dVAR; |
| 6903 | |
| 6904 | PERL_ARGS_ASSERT_SV_FREE2; |
| 6905 | |
| 6906 | if (LIKELY( rc == 1 )) { |
| 6907 | /* normal case */ |
| 6908 | SvREFCNT(sv) = 0; |
| 6909 | |
| 6910 | #ifdef DEBUGGING |
| 6911 | if (SvTEMP(sv)) { |
| 6912 | Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING), |
| 6913 | "Attempt to free temp prematurely: SV 0x%"UVxf |
| 6914 | pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); |
| 6915 | return; |
| 6916 | } |
| 6917 | #endif |
| 6918 | if (SvIMMORTAL(sv)) { |
| 6919 | /* make sure SvREFCNT(sv)==0 happens very seldom */ |
| 6920 | SvREFCNT(sv) = SvREFCNT_IMMORTAL; |
| 6921 | return; |
| 6922 | } |
| 6923 | sv_clear(sv); |
| 6924 | if (! SvREFCNT(sv)) /* may have have been resurrected */ |
| 6925 | del_SV(sv); |
| 6926 | return; |
| 6927 | } |
| 6928 | |
| 6929 | /* handle exceptional cases */ |
| 6930 | |
| 6931 | assert(rc == 0); |
| 6932 | |
| 6933 | if (SvFLAGS(sv) & SVf_BREAK) |
| 6934 | /* this SV's refcnt has been artificially decremented to |
| 6935 | * trigger cleanup */ |
| 6936 | return; |
| 6937 | if (PL_in_clean_all) /* All is fair */ |
| 6938 | return; |
| 6939 | if (SvIMMORTAL(sv)) { |
| 6940 | /* make sure SvREFCNT(sv)==0 happens very seldom */ |
| 6941 | SvREFCNT(sv) = SvREFCNT_IMMORTAL; |
| 6942 | return; |
| 6943 | } |
| 6944 | if (ckWARN_d(WARN_INTERNAL)) { |
| 6945 | #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP |
| 6946 | Perl_dump_sv_child(aTHX_ sv); |
| 6947 | #else |
| 6948 | #ifdef DEBUG_LEAKING_SCALARS |
| 6949 | sv_dump(sv); |
| 6950 | #endif |
| 6951 | #ifdef DEBUG_LEAKING_SCALARS_ABORT |
| 6952 | if (PL_warnhook == PERL_WARNHOOK_FATAL |
| 6953 | || ckDEAD(packWARN(WARN_INTERNAL))) { |
| 6954 | /* Don't let Perl_warner cause us to escape our fate: */ |
| 6955 | abort(); |
| 6956 | } |
| 6957 | #endif |
| 6958 | /* This may not return: */ |
| 6959 | Perl_warner(aTHX_ packWARN(WARN_INTERNAL), |
| 6960 | "Attempt to free unreferenced scalar: SV 0x%"UVxf |
| 6961 | pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE); |
| 6962 | #endif |
| 6963 | } |
| 6964 | #ifdef DEBUG_LEAKING_SCALARS_ABORT |
| 6965 | abort(); |
| 6966 | #endif |
| 6967 | |
| 6968 | } |
| 6969 | |
| 6970 | |
| 6971 | /* |
| 6972 | =for apidoc sv_len |
| 6973 | |
| 6974 | Returns the length of the string in the SV. Handles magic and type |
| 6975 | coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which |
| 6976 | gives raw access to the C<xpv_cur> slot. |
| 6977 | |
| 6978 | =cut |
| 6979 | */ |
| 6980 | |
| 6981 | STRLEN |
| 6982 | Perl_sv_len(pTHX_ SV *const sv) |
| 6983 | { |
| 6984 | STRLEN len; |
| 6985 | |
| 6986 | if (!sv) |
| 6987 | return 0; |
| 6988 | |
| 6989 | (void)SvPV_const(sv, len); |
| 6990 | return len; |
| 6991 | } |
| 6992 | |
| 6993 | /* |
| 6994 | =for apidoc sv_len_utf8 |
| 6995 | |
| 6996 | Returns the number of characters in the string in an SV, counting wide |
| 6997 | UTF-8 bytes as a single character. Handles magic and type coercion. |
| 6998 | |
| 6999 | =cut |
| 7000 | */ |
| 7001 | |
| 7002 | /* |
| 7003 | * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the |
| 7004 | * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below. |
| 7005 | * (Note that the mg_len is not the length of the mg_ptr field. |
| 7006 | * This allows the cache to store the character length of the string without |
| 7007 | * needing to malloc() extra storage to attach to the mg_ptr.) |
| 7008 | * |
| 7009 | */ |
| 7010 | |
| 7011 | STRLEN |
| 7012 | Perl_sv_len_utf8(pTHX_ SV *const sv) |
| 7013 | { |
| 7014 | if (!sv) |
| 7015 | return 0; |
| 7016 | |
| 7017 | SvGETMAGIC(sv); |
| 7018 | return sv_len_utf8_nomg(sv); |
| 7019 | } |
| 7020 | |
| 7021 | STRLEN |
| 7022 | Perl_sv_len_utf8_nomg(pTHX_ SV * const sv) |
| 7023 | { |
| 7024 | STRLEN len; |
| 7025 | const U8 *s = (U8*)SvPV_nomg_const(sv, len); |
| 7026 | |
| 7027 | PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG; |
| 7028 | |
| 7029 | if (PL_utf8cache && SvUTF8(sv)) { |
| 7030 | STRLEN ulen; |
| 7031 | MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 7032 | |
| 7033 | if (mg && (mg->mg_len != -1 || mg->mg_ptr)) { |
| 7034 | if (mg->mg_len != -1) |
| 7035 | ulen = mg->mg_len; |
| 7036 | else { |
| 7037 | /* We can use the offset cache for a headstart. |
| 7038 | The longer value is stored in the first pair. */ |
| 7039 | STRLEN *cache = (STRLEN *) mg->mg_ptr; |
| 7040 | |
| 7041 | ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1], |
| 7042 | s + len); |
| 7043 | } |
| 7044 | |
| 7045 | if (PL_utf8cache < 0) { |
| 7046 | const STRLEN real = Perl_utf8_length(aTHX_ s, s + len); |
| 7047 | assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv); |
| 7048 | } |
| 7049 | } |
| 7050 | else { |
| 7051 | ulen = Perl_utf8_length(aTHX_ s, s + len); |
| 7052 | utf8_mg_len_cache_update(sv, &mg, ulen); |
| 7053 | } |
| 7054 | return ulen; |
| 7055 | } |
| 7056 | return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len; |
| 7057 | } |
| 7058 | |
| 7059 | /* Walk forwards to find the byte corresponding to the passed in UTF-8 |
| 7060 | offset. */ |
| 7061 | static STRLEN |
| 7062 | S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send, |
| 7063 | STRLEN *const uoffset_p, bool *const at_end) |
| 7064 | { |
| 7065 | const U8 *s = start; |
| 7066 | STRLEN uoffset = *uoffset_p; |
| 7067 | |
| 7068 | PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS; |
| 7069 | |
| 7070 | while (s < send && uoffset) { |
| 7071 | --uoffset; |
| 7072 | s += UTF8SKIP(s); |
| 7073 | } |
| 7074 | if (s == send) { |
| 7075 | *at_end = TRUE; |
| 7076 | } |
| 7077 | else if (s > send) { |
| 7078 | *at_end = TRUE; |
| 7079 | /* This is the existing behaviour. Possibly it should be a croak, as |
| 7080 | it's actually a bounds error */ |
| 7081 | s = send; |
| 7082 | } |
| 7083 | *uoffset_p -= uoffset; |
| 7084 | return s - start; |
| 7085 | } |
| 7086 | |
| 7087 | /* Given the length of the string in both bytes and UTF-8 characters, decide |
| 7088 | whether to walk forwards or backwards to find the byte corresponding to |
| 7089 | the passed in UTF-8 offset. */ |
| 7090 | static STRLEN |
| 7091 | S_sv_pos_u2b_midway(const U8 *const start, const U8 *send, |
| 7092 | STRLEN uoffset, const STRLEN uend) |
| 7093 | { |
| 7094 | STRLEN backw = uend - uoffset; |
| 7095 | |
| 7096 | PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY; |
| 7097 | |
| 7098 | if (uoffset < 2 * backw) { |
| 7099 | /* The assumption is that going forwards is twice the speed of going |
| 7100 | forward (that's where the 2 * backw comes from). |
| 7101 | (The real figure of course depends on the UTF-8 data.) */ |
| 7102 | const U8 *s = start; |
| 7103 | |
| 7104 | while (s < send && uoffset--) |
| 7105 | s += UTF8SKIP(s); |
| 7106 | assert (s <= send); |
| 7107 | if (s > send) |
| 7108 | s = send; |
| 7109 | return s - start; |
| 7110 | } |
| 7111 | |
| 7112 | while (backw--) { |
| 7113 | send--; |
| 7114 | while (UTF8_IS_CONTINUATION(*send)) |
| 7115 | send--; |
| 7116 | } |
| 7117 | return send - start; |
| 7118 | } |
| 7119 | |
| 7120 | /* For the string representation of the given scalar, find the byte |
| 7121 | corresponding to the passed in UTF-8 offset. uoffset0 and boffset0 |
| 7122 | give another position in the string, *before* the sought offset, which |
| 7123 | (which is always true, as 0, 0 is a valid pair of positions), which should |
| 7124 | help reduce the amount of linear searching. |
| 7125 | If *mgp is non-NULL, it should point to the UTF-8 cache magic, which |
| 7126 | will be used to reduce the amount of linear searching. The cache will be |
| 7127 | created if necessary, and the found value offered to it for update. */ |
| 7128 | static STRLEN |
| 7129 | S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start, |
| 7130 | const U8 *const send, STRLEN uoffset, |
| 7131 | STRLEN uoffset0, STRLEN boffset0) |
| 7132 | { |
| 7133 | STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */ |
| 7134 | bool found = FALSE; |
| 7135 | bool at_end = FALSE; |
| 7136 | |
| 7137 | PERL_ARGS_ASSERT_SV_POS_U2B_CACHED; |
| 7138 | |
| 7139 | assert (uoffset >= uoffset0); |
| 7140 | |
| 7141 | if (!uoffset) |
| 7142 | return 0; |
| 7143 | |
| 7144 | if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv) |
| 7145 | && PL_utf8cache |
| 7146 | && (*mgp || (SvTYPE(sv) >= SVt_PVMG && |
| 7147 | (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) { |
| 7148 | if ((*mgp)->mg_ptr) { |
| 7149 | STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr; |
| 7150 | if (cache[0] == uoffset) { |
| 7151 | /* An exact match. */ |
| 7152 | return cache[1]; |
| 7153 | } |
| 7154 | if (cache[2] == uoffset) { |
| 7155 | /* An exact match. */ |
| 7156 | return cache[3]; |
| 7157 | } |
| 7158 | |
| 7159 | if (cache[0] < uoffset) { |
| 7160 | /* The cache already knows part of the way. */ |
| 7161 | if (cache[0] > uoffset0) { |
| 7162 | /* The cache knows more than the passed in pair */ |
| 7163 | uoffset0 = cache[0]; |
| 7164 | boffset0 = cache[1]; |
| 7165 | } |
| 7166 | if ((*mgp)->mg_len != -1) { |
| 7167 | /* And we know the end too. */ |
| 7168 | boffset = boffset0 |
| 7169 | + sv_pos_u2b_midway(start + boffset0, send, |
| 7170 | uoffset - uoffset0, |
| 7171 | (*mgp)->mg_len - uoffset0); |
| 7172 | } else { |
| 7173 | uoffset -= uoffset0; |
| 7174 | boffset = boffset0 |
| 7175 | + sv_pos_u2b_forwards(start + boffset0, |
| 7176 | send, &uoffset, &at_end); |
| 7177 | uoffset += uoffset0; |
| 7178 | } |
| 7179 | } |
| 7180 | else if (cache[2] < uoffset) { |
| 7181 | /* We're between the two cache entries. */ |
| 7182 | if (cache[2] > uoffset0) { |
| 7183 | /* and the cache knows more than the passed in pair */ |
| 7184 | uoffset0 = cache[2]; |
| 7185 | boffset0 = cache[3]; |
| 7186 | } |
| 7187 | |
| 7188 | boffset = boffset0 |
| 7189 | + sv_pos_u2b_midway(start + boffset0, |
| 7190 | start + cache[1], |
| 7191 | uoffset - uoffset0, |
| 7192 | cache[0] - uoffset0); |
| 7193 | } else { |
| 7194 | boffset = boffset0 |
| 7195 | + sv_pos_u2b_midway(start + boffset0, |
| 7196 | start + cache[3], |
| 7197 | uoffset - uoffset0, |
| 7198 | cache[2] - uoffset0); |
| 7199 | } |
| 7200 | found = TRUE; |
| 7201 | } |
| 7202 | else if ((*mgp)->mg_len != -1) { |
| 7203 | /* If we can take advantage of a passed in offset, do so. */ |
| 7204 | /* In fact, offset0 is either 0, or less than offset, so don't |
| 7205 | need to worry about the other possibility. */ |
| 7206 | boffset = boffset0 |
| 7207 | + sv_pos_u2b_midway(start + boffset0, send, |
| 7208 | uoffset - uoffset0, |
| 7209 | (*mgp)->mg_len - uoffset0); |
| 7210 | found = TRUE; |
| 7211 | } |
| 7212 | } |
| 7213 | |
| 7214 | if (!found || PL_utf8cache < 0) { |
| 7215 | STRLEN real_boffset; |
| 7216 | uoffset -= uoffset0; |
| 7217 | real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0, |
| 7218 | send, &uoffset, &at_end); |
| 7219 | uoffset += uoffset0; |
| 7220 | |
| 7221 | if (found && PL_utf8cache < 0) |
| 7222 | assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset, |
| 7223 | real_boffset, sv); |
| 7224 | boffset = real_boffset; |
| 7225 | } |
| 7226 | |
| 7227 | if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) { |
| 7228 | if (at_end) |
| 7229 | utf8_mg_len_cache_update(sv, mgp, uoffset); |
| 7230 | else |
| 7231 | utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start); |
| 7232 | } |
| 7233 | return boffset; |
| 7234 | } |
| 7235 | |
| 7236 | |
| 7237 | /* |
| 7238 | =for apidoc sv_pos_u2b_flags |
| 7239 | |
| 7240 | Converts the offset from a count of UTF-8 chars from |
| 7241 | the start of the string, to a count of the equivalent number of bytes; if |
| 7242 | C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from |
| 7243 | C<offset>, rather than from the start |
| 7244 | of the string. Handles type coercion. |
| 7245 | C<flags> is passed to C<SvPV_flags>, and usually should be |
| 7246 | C<SV_GMAGIC|SV_CONST_RETURN> to handle magic. |
| 7247 | |
| 7248 | =cut |
| 7249 | */ |
| 7250 | |
| 7251 | /* |
| 7252 | * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential |
| 7253 | * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and |
| 7254 | * byte offsets. See also the comments of S_utf8_mg_pos_cache_update(). |
| 7255 | * |
| 7256 | */ |
| 7257 | |
| 7258 | STRLEN |
| 7259 | Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp, |
| 7260 | U32 flags) |
| 7261 | { |
| 7262 | const U8 *start; |
| 7263 | STRLEN len; |
| 7264 | STRLEN boffset; |
| 7265 | |
| 7266 | PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS; |
| 7267 | |
| 7268 | start = (U8*)SvPV_flags(sv, len, flags); |
| 7269 | if (len) { |
| 7270 | const U8 * const send = start + len; |
| 7271 | MAGIC *mg = NULL; |
| 7272 | boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0); |
| 7273 | |
| 7274 | if (lenp |
| 7275 | && *lenp /* don't bother doing work for 0, as its bytes equivalent |
| 7276 | is 0, and *lenp is already set to that. */) { |
| 7277 | /* Convert the relative offset to absolute. */ |
| 7278 | const STRLEN uoffset2 = uoffset + *lenp; |
| 7279 | const STRLEN boffset2 |
| 7280 | = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2, |
| 7281 | uoffset, boffset) - boffset; |
| 7282 | |
| 7283 | *lenp = boffset2; |
| 7284 | } |
| 7285 | } else { |
| 7286 | if (lenp) |
| 7287 | *lenp = 0; |
| 7288 | boffset = 0; |
| 7289 | } |
| 7290 | |
| 7291 | return boffset; |
| 7292 | } |
| 7293 | |
| 7294 | /* |
| 7295 | =for apidoc sv_pos_u2b |
| 7296 | |
| 7297 | Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from |
| 7298 | the start of the string, to a count of the equivalent number of bytes; if |
| 7299 | C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from |
| 7300 | the offset, rather than from the start of the string. Handles magic and |
| 7301 | type coercion. |
| 7302 | |
| 7303 | Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer |
| 7304 | than 2Gb. |
| 7305 | |
| 7306 | =cut |
| 7307 | */ |
| 7308 | |
| 7309 | /* |
| 7310 | * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential |
| 7311 | * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and |
| 7312 | * byte offsets. See also the comments of S_utf8_mg_pos_cache_update(). |
| 7313 | * |
| 7314 | */ |
| 7315 | |
| 7316 | /* This function is subject to size and sign problems */ |
| 7317 | |
| 7318 | void |
| 7319 | Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp) |
| 7320 | { |
| 7321 | PERL_ARGS_ASSERT_SV_POS_U2B; |
| 7322 | |
| 7323 | if (lenp) { |
| 7324 | STRLEN ulen = (STRLEN)*lenp; |
| 7325 | *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen, |
| 7326 | SV_GMAGIC|SV_CONST_RETURN); |
| 7327 | *lenp = (I32)ulen; |
| 7328 | } else { |
| 7329 | *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL, |
| 7330 | SV_GMAGIC|SV_CONST_RETURN); |
| 7331 | } |
| 7332 | } |
| 7333 | |
| 7334 | static void |
| 7335 | S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, |
| 7336 | const STRLEN ulen) |
| 7337 | { |
| 7338 | PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE; |
| 7339 | if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv)) |
| 7340 | return; |
| 7341 | |
| 7342 | if (!*mgp && (SvTYPE(sv) < SVt_PVMG || |
| 7343 | !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) { |
| 7344 | *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0); |
| 7345 | } |
| 7346 | assert(*mgp); |
| 7347 | |
| 7348 | (*mgp)->mg_len = ulen; |
| 7349 | } |
| 7350 | |
| 7351 | /* Create and update the UTF8 magic offset cache, with the proffered utf8/ |
| 7352 | byte length pairing. The (byte) length of the total SV is passed in too, |
| 7353 | as blen, because for some (more esoteric) SVs, the call to SvPV_const() |
| 7354 | may not have updated SvCUR, so we can't rely on reading it directly. |
| 7355 | |
| 7356 | The proffered utf8/byte length pairing isn't used if the cache already has |
| 7357 | two pairs, and swapping either for the proffered pair would increase the |
| 7358 | RMS of the intervals between known byte offsets. |
| 7359 | |
| 7360 | The cache itself consists of 4 STRLEN values |
| 7361 | 0: larger UTF-8 offset |
| 7362 | 1: corresponding byte offset |
| 7363 | 2: smaller UTF-8 offset |
| 7364 | 3: corresponding byte offset |
| 7365 | |
| 7366 | Unused cache pairs have the value 0, 0. |
| 7367 | Keeping the cache "backwards" means that the invariant of |
| 7368 | cache[0] >= cache[2] is maintained even with empty slots, which means that |
| 7369 | the code that uses it doesn't need to worry if only 1 entry has actually |
| 7370 | been set to non-zero. It also makes the "position beyond the end of the |
| 7371 | cache" logic much simpler, as the first slot is always the one to start |
| 7372 | from. |
| 7373 | */ |
| 7374 | static void |
| 7375 | S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte, |
| 7376 | const STRLEN utf8, const STRLEN blen) |
| 7377 | { |
| 7378 | STRLEN *cache; |
| 7379 | |
| 7380 | PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE; |
| 7381 | |
| 7382 | if (SvREADONLY(sv)) |
| 7383 | return; |
| 7384 | |
| 7385 | if (!*mgp && (SvTYPE(sv) < SVt_PVMG || |
| 7386 | !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) { |
| 7387 | *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0, |
| 7388 | 0); |
| 7389 | (*mgp)->mg_len = -1; |
| 7390 | } |
| 7391 | assert(*mgp); |
| 7392 | |
| 7393 | if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) { |
| 7394 | Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN); |
| 7395 | (*mgp)->mg_ptr = (char *) cache; |
| 7396 | } |
| 7397 | assert(cache); |
| 7398 | |
| 7399 | if (PL_utf8cache < 0 && SvPOKp(sv)) { |
| 7400 | /* SvPOKp() because, if sv is a reference, then SvPVX() is actually |
| 7401 | a pointer. Note that we no longer cache utf8 offsets on refer- |
| 7402 | ences, but this check is still a good idea, for robustness. */ |
| 7403 | const U8 *start = (const U8 *) SvPVX_const(sv); |
| 7404 | const STRLEN realutf8 = utf8_length(start, start + byte); |
| 7405 | |
| 7406 | assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8, |
| 7407 | sv); |
| 7408 | } |
| 7409 | |
| 7410 | /* Cache is held with the later position first, to simplify the code |
| 7411 | that deals with unbounded ends. */ |
| 7412 | |
| 7413 | ASSERT_UTF8_CACHE(cache); |
| 7414 | if (cache[1] == 0) { |
| 7415 | /* Cache is totally empty */ |
| 7416 | cache[0] = utf8; |
| 7417 | cache[1] = byte; |
| 7418 | } else if (cache[3] == 0) { |
| 7419 | if (byte > cache[1]) { |
| 7420 | /* New one is larger, so goes first. */ |
| 7421 | cache[2] = cache[0]; |
| 7422 | cache[3] = cache[1]; |
| 7423 | cache[0] = utf8; |
| 7424 | cache[1] = byte; |
| 7425 | } else { |
| 7426 | cache[2] = utf8; |
| 7427 | cache[3] = byte; |
| 7428 | } |
| 7429 | } else { |
| 7430 | /* float casts necessary? XXX */ |
| 7431 | #define THREEWAY_SQUARE(a,b,c,d) \ |
| 7432 | ((float)((d) - (c))) * ((float)((d) - (c))) \ |
| 7433 | + ((float)((c) - (b))) * ((float)((c) - (b))) \ |
| 7434 | + ((float)((b) - (a))) * ((float)((b) - (a))) |
| 7435 | |
| 7436 | /* Cache has 2 slots in use, and we know three potential pairs. |
| 7437 | Keep the two that give the lowest RMS distance. Do the |
| 7438 | calculation in bytes simply because we always know the byte |
| 7439 | length. squareroot has the same ordering as the positive value, |
| 7440 | so don't bother with the actual square root. */ |
| 7441 | if (byte > cache[1]) { |
| 7442 | /* New position is after the existing pair of pairs. */ |
| 7443 | const float keep_earlier |
| 7444 | = THREEWAY_SQUARE(0, cache[3], byte, blen); |
| 7445 | const float keep_later |
| 7446 | = THREEWAY_SQUARE(0, cache[1], byte, blen); |
| 7447 | |
| 7448 | if (keep_later < keep_earlier) { |
| 7449 | cache[2] = cache[0]; |
| 7450 | cache[3] = cache[1]; |
| 7451 | } |
| 7452 | cache[0] = utf8; |
| 7453 | cache[1] = byte; |
| 7454 | } |
| 7455 | else { |
| 7456 | const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen); |
| 7457 | float b, c, keep_earlier; |
| 7458 | if (byte > cache[3]) { |
| 7459 | /* New position is between the existing pair of pairs. */ |
| 7460 | b = (float)cache[3]; |
| 7461 | c = (float)byte; |
| 7462 | } else { |
| 7463 | /* New position is before the existing pair of pairs. */ |
| 7464 | b = (float)byte; |
| 7465 | c = (float)cache[3]; |
| 7466 | } |
| 7467 | keep_earlier = THREEWAY_SQUARE(0, b, c, blen); |
| 7468 | if (byte > cache[3]) { |
| 7469 | if (keep_later < keep_earlier) { |
| 7470 | cache[2] = utf8; |
| 7471 | cache[3] = byte; |
| 7472 | } |
| 7473 | else { |
| 7474 | cache[0] = utf8; |
| 7475 | cache[1] = byte; |
| 7476 | } |
| 7477 | } |
| 7478 | else { |
| 7479 | if (! (keep_later < keep_earlier)) { |
| 7480 | cache[0] = cache[2]; |
| 7481 | cache[1] = cache[3]; |
| 7482 | } |
| 7483 | cache[2] = utf8; |
| 7484 | cache[3] = byte; |
| 7485 | } |
| 7486 | } |
| 7487 | } |
| 7488 | ASSERT_UTF8_CACHE(cache); |
| 7489 | } |
| 7490 | |
| 7491 | /* We already know all of the way, now we may be able to walk back. The same |
| 7492 | assumption is made as in S_sv_pos_u2b_midway(), namely that walking |
| 7493 | backward is half the speed of walking forward. */ |
| 7494 | static STRLEN |
| 7495 | S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target, |
| 7496 | const U8 *end, STRLEN endu) |
| 7497 | { |
| 7498 | const STRLEN forw = target - s; |
| 7499 | STRLEN backw = end - target; |
| 7500 | |
| 7501 | PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY; |
| 7502 | |
| 7503 | if (forw < 2 * backw) { |
| 7504 | return utf8_length(s, target); |
| 7505 | } |
| 7506 | |
| 7507 | while (end > target) { |
| 7508 | end--; |
| 7509 | while (UTF8_IS_CONTINUATION(*end)) { |
| 7510 | end--; |
| 7511 | } |
| 7512 | endu--; |
| 7513 | } |
| 7514 | return endu; |
| 7515 | } |
| 7516 | |
| 7517 | /* |
| 7518 | =for apidoc sv_pos_b2u_flags |
| 7519 | |
| 7520 | Converts C<offset> from a count of bytes from the start of the string, to |
| 7521 | a count of the equivalent number of UTF-8 chars. Handles type coercion. |
| 7522 | C<flags> is passed to C<SvPV_flags>, and usually should be |
| 7523 | C<SV_GMAGIC|SV_CONST_RETURN> to handle magic. |
| 7524 | |
| 7525 | =cut |
| 7526 | */ |
| 7527 | |
| 7528 | /* |
| 7529 | * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the |
| 7530 | * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 |
| 7531 | * and byte offsets. |
| 7532 | * |
| 7533 | */ |
| 7534 | STRLEN |
| 7535 | Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags) |
| 7536 | { |
| 7537 | const U8* s; |
| 7538 | STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */ |
| 7539 | STRLEN blen; |
| 7540 | MAGIC* mg = NULL; |
| 7541 | const U8* send; |
| 7542 | bool found = FALSE; |
| 7543 | |
| 7544 | PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS; |
| 7545 | |
| 7546 | s = (const U8*)SvPV_flags(sv, blen, flags); |
| 7547 | |
| 7548 | if (blen < offset) |
| 7549 | Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf |
| 7550 | ", byte=%"UVuf, (UV)blen, (UV)offset); |
| 7551 | |
| 7552 | send = s + offset; |
| 7553 | |
| 7554 | if (!SvREADONLY(sv) |
| 7555 | && PL_utf8cache |
| 7556 | && SvTYPE(sv) >= SVt_PVMG |
| 7557 | && (mg = mg_find(sv, PERL_MAGIC_utf8))) |
| 7558 | { |
| 7559 | if (mg->mg_ptr) { |
| 7560 | STRLEN * const cache = (STRLEN *) mg->mg_ptr; |
| 7561 | if (cache[1] == offset) { |
| 7562 | /* An exact match. */ |
| 7563 | return cache[0]; |
| 7564 | } |
| 7565 | if (cache[3] == offset) { |
| 7566 | /* An exact match. */ |
| 7567 | return cache[2]; |
| 7568 | } |
| 7569 | |
| 7570 | if (cache[1] < offset) { |
| 7571 | /* We already know part of the way. */ |
| 7572 | if (mg->mg_len != -1) { |
| 7573 | /* Actually, we know the end too. */ |
| 7574 | len = cache[0] |
| 7575 | + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send, |
| 7576 | s + blen, mg->mg_len - cache[0]); |
| 7577 | } else { |
| 7578 | len = cache[0] + utf8_length(s + cache[1], send); |
| 7579 | } |
| 7580 | } |
| 7581 | else if (cache[3] < offset) { |
| 7582 | /* We're between the two cached pairs, so we do the calculation |
| 7583 | offset by the byte/utf-8 positions for the earlier pair, |
| 7584 | then add the utf-8 characters from the string start to |
| 7585 | there. */ |
| 7586 | len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send, |
| 7587 | s + cache[1], cache[0] - cache[2]) |
| 7588 | + cache[2]; |
| 7589 | |
| 7590 | } |
| 7591 | else { /* cache[3] > offset */ |
| 7592 | len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3], |
| 7593 | cache[2]); |
| 7594 | |
| 7595 | } |
| 7596 | ASSERT_UTF8_CACHE(cache); |
| 7597 | found = TRUE; |
| 7598 | } else if (mg->mg_len != -1) { |
| 7599 | len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len); |
| 7600 | found = TRUE; |
| 7601 | } |
| 7602 | } |
| 7603 | if (!found || PL_utf8cache < 0) { |
| 7604 | const STRLEN real_len = utf8_length(s, send); |
| 7605 | |
| 7606 | if (found && PL_utf8cache < 0) |
| 7607 | assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv); |
| 7608 | len = real_len; |
| 7609 | } |
| 7610 | |
| 7611 | if (PL_utf8cache) { |
| 7612 | if (blen == offset) |
| 7613 | utf8_mg_len_cache_update(sv, &mg, len); |
| 7614 | else |
| 7615 | utf8_mg_pos_cache_update(sv, &mg, offset, len, blen); |
| 7616 | } |
| 7617 | |
| 7618 | return len; |
| 7619 | } |
| 7620 | |
| 7621 | /* |
| 7622 | =for apidoc sv_pos_b2u |
| 7623 | |
| 7624 | Converts the value pointed to by C<offsetp> from a count of bytes from the |
| 7625 | start of the string, to a count of the equivalent number of UTF-8 chars. |
| 7626 | Handles magic and type coercion. |
| 7627 | |
| 7628 | Use C<sv_pos_b2u_flags> in preference, which correctly handles strings |
| 7629 | longer than 2Gb. |
| 7630 | |
| 7631 | =cut |
| 7632 | */ |
| 7633 | |
| 7634 | /* |
| 7635 | * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential |
| 7636 | * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and |
| 7637 | * byte offsets. |
| 7638 | * |
| 7639 | */ |
| 7640 | void |
| 7641 | Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp) |
| 7642 | { |
| 7643 | PERL_ARGS_ASSERT_SV_POS_B2U; |
| 7644 | |
| 7645 | if (!sv) |
| 7646 | return; |
| 7647 | |
| 7648 | *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp, |
| 7649 | SV_GMAGIC|SV_CONST_RETURN); |
| 7650 | } |
| 7651 | |
| 7652 | static void |
| 7653 | S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache, |
| 7654 | STRLEN real, SV *const sv) |
| 7655 | { |
| 7656 | PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT; |
| 7657 | |
| 7658 | /* As this is debugging only code, save space by keeping this test here, |
| 7659 | rather than inlining it in all the callers. */ |
| 7660 | if (from_cache == real) |
| 7661 | return; |
| 7662 | |
| 7663 | /* Need to turn the assertions off otherwise we may recurse infinitely |
| 7664 | while printing error messages. */ |
| 7665 | SAVEI8(PL_utf8cache); |
| 7666 | PL_utf8cache = 0; |
| 7667 | Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf, |
| 7668 | func, (UV) from_cache, (UV) real, SVfARG(sv)); |
| 7669 | } |
| 7670 | |
| 7671 | /* |
| 7672 | =for apidoc sv_eq |
| 7673 | |
| 7674 | Returns a boolean indicating whether the strings in the two SVs are |
| 7675 | identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will |
| 7676 | coerce its args to strings if necessary. |
| 7677 | |
| 7678 | =for apidoc sv_eq_flags |
| 7679 | |
| 7680 | Returns a boolean indicating whether the strings in the two SVs are |
| 7681 | identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings |
| 7682 | if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too. |
| 7683 | |
| 7684 | =cut |
| 7685 | */ |
| 7686 | |
| 7687 | I32 |
| 7688 | Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags) |
| 7689 | { |
| 7690 | const char *pv1; |
| 7691 | STRLEN cur1; |
| 7692 | const char *pv2; |
| 7693 | STRLEN cur2; |
| 7694 | I32 eq = 0; |
| 7695 | SV* svrecode = NULL; |
| 7696 | |
| 7697 | if (!sv1) { |
| 7698 | pv1 = ""; |
| 7699 | cur1 = 0; |
| 7700 | } |
| 7701 | else { |
| 7702 | /* if pv1 and pv2 are the same, second SvPV_const call may |
| 7703 | * invalidate pv1 (if we are handling magic), so we may need to |
| 7704 | * make a copy */ |
| 7705 | if (sv1 == sv2 && flags & SV_GMAGIC |
| 7706 | && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) { |
| 7707 | pv1 = SvPV_const(sv1, cur1); |
| 7708 | sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2)); |
| 7709 | } |
| 7710 | pv1 = SvPV_flags_const(sv1, cur1, flags); |
| 7711 | } |
| 7712 | |
| 7713 | if (!sv2){ |
| 7714 | pv2 = ""; |
| 7715 | cur2 = 0; |
| 7716 | } |
| 7717 | else |
| 7718 | pv2 = SvPV_flags_const(sv2, cur2, flags); |
| 7719 | |
| 7720 | if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) { |
| 7721 | /* Differing utf8ness. |
| 7722 | * Do not UTF8size the comparands as a side-effect. */ |
| 7723 | if (IN_ENCODING) { |
| 7724 | if (SvUTF8(sv1)) { |
| 7725 | svrecode = newSVpvn(pv2, cur2); |
| 7726 | sv_recode_to_utf8(svrecode, _get_encoding()); |
| 7727 | pv2 = SvPV_const(svrecode, cur2); |
| 7728 | } |
| 7729 | else { |
| 7730 | svrecode = newSVpvn(pv1, cur1); |
| 7731 | sv_recode_to_utf8(svrecode, _get_encoding()); |
| 7732 | pv1 = SvPV_const(svrecode, cur1); |
| 7733 | } |
| 7734 | /* Now both are in UTF-8. */ |
| 7735 | if (cur1 != cur2) { |
| 7736 | SvREFCNT_dec_NN(svrecode); |
| 7737 | return FALSE; |
| 7738 | } |
| 7739 | } |
| 7740 | else { |
| 7741 | if (SvUTF8(sv1)) { |
| 7742 | /* sv1 is the UTF-8 one */ |
| 7743 | return bytes_cmp_utf8((const U8*)pv2, cur2, |
| 7744 | (const U8*)pv1, cur1) == 0; |
| 7745 | } |
| 7746 | else { |
| 7747 | /* sv2 is the UTF-8 one */ |
| 7748 | return bytes_cmp_utf8((const U8*)pv1, cur1, |
| 7749 | (const U8*)pv2, cur2) == 0; |
| 7750 | } |
| 7751 | } |
| 7752 | } |
| 7753 | |
| 7754 | if (cur1 == cur2) |
| 7755 | eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1); |
| 7756 | |
| 7757 | SvREFCNT_dec(svrecode); |
| 7758 | |
| 7759 | return eq; |
| 7760 | } |
| 7761 | |
| 7762 | /* |
| 7763 | =for apidoc sv_cmp |
| 7764 | |
| 7765 | Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the |
| 7766 | string in C<sv1> is less than, equal to, or greater than the string in |
| 7767 | C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will |
| 7768 | coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>. |
| 7769 | |
| 7770 | =for apidoc sv_cmp_flags |
| 7771 | |
| 7772 | Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the |
| 7773 | string in C<sv1> is less than, equal to, or greater than the string in |
| 7774 | C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings |
| 7775 | if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See |
| 7776 | also C<L</sv_cmp_locale_flags>>. |
| 7777 | |
| 7778 | =cut |
| 7779 | */ |
| 7780 | |
| 7781 | I32 |
| 7782 | Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2) |
| 7783 | { |
| 7784 | return sv_cmp_flags(sv1, sv2, SV_GMAGIC); |
| 7785 | } |
| 7786 | |
| 7787 | I32 |
| 7788 | Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2, |
| 7789 | const U32 flags) |
| 7790 | { |
| 7791 | STRLEN cur1, cur2; |
| 7792 | const char *pv1, *pv2; |
| 7793 | I32 cmp; |
| 7794 | SV *svrecode = NULL; |
| 7795 | |
| 7796 | if (!sv1) { |
| 7797 | pv1 = ""; |
| 7798 | cur1 = 0; |
| 7799 | } |
| 7800 | else |
| 7801 | pv1 = SvPV_flags_const(sv1, cur1, flags); |
| 7802 | |
| 7803 | if (!sv2) { |
| 7804 | pv2 = ""; |
| 7805 | cur2 = 0; |
| 7806 | } |
| 7807 | else |
| 7808 | pv2 = SvPV_flags_const(sv2, cur2, flags); |
| 7809 | |
| 7810 | if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) { |
| 7811 | /* Differing utf8ness. |
| 7812 | * Do not UTF8size the comparands as a side-effect. */ |
| 7813 | if (SvUTF8(sv1)) { |
| 7814 | if (IN_ENCODING) { |
| 7815 | svrecode = newSVpvn(pv2, cur2); |
| 7816 | sv_recode_to_utf8(svrecode, _get_encoding()); |
| 7817 | pv2 = SvPV_const(svrecode, cur2); |
| 7818 | } |
| 7819 | else { |
| 7820 | const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2, |
| 7821 | (const U8*)pv1, cur1); |
| 7822 | return retval ? retval < 0 ? -1 : +1 : 0; |
| 7823 | } |
| 7824 | } |
| 7825 | else { |
| 7826 | if (IN_ENCODING) { |
| 7827 | svrecode = newSVpvn(pv1, cur1); |
| 7828 | sv_recode_to_utf8(svrecode, _get_encoding()); |
| 7829 | pv1 = SvPV_const(svrecode, cur1); |
| 7830 | } |
| 7831 | else { |
| 7832 | const int retval = bytes_cmp_utf8((const U8*)pv1, cur1, |
| 7833 | (const U8*)pv2, cur2); |
| 7834 | return retval ? retval < 0 ? -1 : +1 : 0; |
| 7835 | } |
| 7836 | } |
| 7837 | } |
| 7838 | |
| 7839 | /* Here, if both are non-NULL, then they have the same UTF8ness. */ |
| 7840 | |
| 7841 | if (!cur1) { |
| 7842 | cmp = cur2 ? -1 : 0; |
| 7843 | } else if (!cur2) { |
| 7844 | cmp = 1; |
| 7845 | } else { |
| 7846 | STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2; |
| 7847 | |
| 7848 | #ifdef EBCDIC |
| 7849 | if (! DO_UTF8(sv1)) { |
| 7850 | #endif |
| 7851 | const I32 retval = memcmp((const void*)pv1, |
| 7852 | (const void*)pv2, |
| 7853 | shortest_len); |
| 7854 | if (retval) { |
| 7855 | cmp = retval < 0 ? -1 : 1; |
| 7856 | } else if (cur1 == cur2) { |
| 7857 | cmp = 0; |
| 7858 | } else { |
| 7859 | cmp = cur1 < cur2 ? -1 : 1; |
| 7860 | } |
| 7861 | #ifdef EBCDIC |
| 7862 | } |
| 7863 | else { /* Both are to be treated as UTF-EBCDIC */ |
| 7864 | |
| 7865 | /* EBCDIC UTF-8 is complicated by the fact that it is based on I8 |
| 7866 | * which remaps code points 0-255. We therefore generally have to |
| 7867 | * unmap back to the original values to get an accurate comparison. |
| 7868 | * But we don't have to do that for UTF-8 invariants, as by |
| 7869 | * definition, they aren't remapped, nor do we have to do it for |
| 7870 | * above-latin1 code points, as they also aren't remapped. (This |
| 7871 | * code also works on ASCII platforms, but the memcmp() above is |
| 7872 | * much faster). */ |
| 7873 | |
| 7874 | const char *e = pv1 + shortest_len; |
| 7875 | |
| 7876 | /* Find the first bytes that differ between the two strings */ |
| 7877 | while (pv1 < e && *pv1 == *pv2) { |
| 7878 | pv1++; |
| 7879 | pv2++; |
| 7880 | } |
| 7881 | |
| 7882 | |
| 7883 | if (pv1 == e) { /* Are the same all the way to the end */ |
| 7884 | if (cur1 == cur2) { |
| 7885 | cmp = 0; |
| 7886 | } else { |
| 7887 | cmp = cur1 < cur2 ? -1 : 1; |
| 7888 | } |
| 7889 | } |
| 7890 | else /* Here *pv1 and *pv2 are not equal, but all bytes earlier |
| 7891 | * in the strings were. The current bytes may or may not be |
| 7892 | * at the beginning of a character. But neither or both are |
| 7893 | * (or else earlier bytes would have been different). And |
| 7894 | * if we are in the middle of a character, the two |
| 7895 | * characters are comprised of the same number of bytes |
| 7896 | * (because in this case the start bytes are the same, and |
| 7897 | * the start bytes encode the character's length). */ |
| 7898 | if (UTF8_IS_INVARIANT(*pv1)) |
| 7899 | { |
| 7900 | /* If both are invariants; can just compare directly */ |
| 7901 | if (UTF8_IS_INVARIANT(*pv2)) { |
| 7902 | cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; |
| 7903 | } |
| 7904 | else /* Since *pv1 is invariant, it is the whole character, |
| 7905 | which means it is at the beginning of a character. |
| 7906 | That means pv2 is also at the beginning of a |
| 7907 | character (see earlier comment). Since it isn't |
| 7908 | invariant, it must be a start byte. If it starts a |
| 7909 | character whose code point is above 255, that |
| 7910 | character is greater than any single-byte char, which |
| 7911 | *pv1 is */ |
| 7912 | if (UTF8_IS_ABOVE_LATIN1_START(*pv2)) |
| 7913 | { |
| 7914 | cmp = -1; |
| 7915 | } |
| 7916 | else { |
| 7917 | /* Here, pv2 points to a character composed of 2 bytes |
| 7918 | * whose code point is < 256. Get its code point and |
| 7919 | * compare with *pv1 */ |
| 7920 | cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1))) |
| 7921 | ? -1 |
| 7922 | : 1; |
| 7923 | } |
| 7924 | } |
| 7925 | else /* The code point starting at pv1 isn't a single byte */ |
| 7926 | if (UTF8_IS_INVARIANT(*pv2)) |
| 7927 | { |
| 7928 | /* But here, the code point starting at *pv2 is a single byte, |
| 7929 | * and so *pv1 must begin a character, hence is a start byte. |
| 7930 | * If that character is above 255, it is larger than any |
| 7931 | * single-byte char, which *pv2 is */ |
| 7932 | if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) { |
| 7933 | cmp = 1; |
| 7934 | } |
| 7935 | else { |
| 7936 | /* Here, pv1 points to a character composed of 2 bytes |
| 7937 | * whose code point is < 256. Get its code point and |
| 7938 | * compare with the single byte character *pv2 */ |
| 7939 | cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2) |
| 7940 | ? -1 |
| 7941 | : 1; |
| 7942 | } |
| 7943 | } |
| 7944 | else /* Here, we've ruled out either *pv1 and *pv2 being |
| 7945 | invariant. That means both are part of variants, but not |
| 7946 | necessarily at the start of a character */ |
| 7947 | if ( UTF8_IS_ABOVE_LATIN1_START(*pv1) |
| 7948 | || UTF8_IS_ABOVE_LATIN1_START(*pv2)) |
| 7949 | { |
| 7950 | /* Here, at least one is the start of a character, which means |
| 7951 | * the other is also a start byte. And the code point of at |
| 7952 | * least one of the characters is above 255. It is a |
| 7953 | * characteristic of UTF-EBCDIC that all start bytes for |
| 7954 | * above-latin1 code points are well behaved as far as code |
| 7955 | * point comparisons go, and all are larger than all other |
| 7956 | * start bytes, so the comparison with those is also well |
| 7957 | * behaved */ |
| 7958 | cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; |
| 7959 | } |
| 7960 | else { |
| 7961 | /* Here both *pv1 and *pv2 are part of variant characters. |
| 7962 | * They could be both continuations, or both start characters. |
| 7963 | * (One or both could even be an illegal start character (for |
| 7964 | * an overlong) which for the purposes of sorting we treat as |
| 7965 | * legal. */ |
| 7966 | if (UTF8_IS_CONTINUATION(*pv1)) { |
| 7967 | |
| 7968 | /* If they are continuations for code points above 255, |
| 7969 | * then comparing the current byte is sufficient, as there |
| 7970 | * is no remapping of these and so the comparison is |
| 7971 | * well-behaved. We determine if they are such |
| 7972 | * continuations by looking at the preceding byte. It |
| 7973 | * could be a start byte, from which we can tell if it is |
| 7974 | * for an above 255 code point. Or it could be a |
| 7975 | * continuation, which means the character occupies at |
| 7976 | * least 3 bytes, so must be above 255. */ |
| 7977 | if ( UTF8_IS_CONTINUATION(*(pv2 - 1)) |
| 7978 | || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1))) |
| 7979 | { |
| 7980 | cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1; |
| 7981 | goto cmp_done; |
| 7982 | } |
| 7983 | |
| 7984 | /* Here, the continuations are for code points below 256; |
| 7985 | * back up one to get to the start byte */ |
| 7986 | pv1--; |
| 7987 | pv2--; |
| 7988 | } |
| 7989 | |
| 7990 | /* We need to get the actual native code point of each of these |
| 7991 | * variants in order to compare them */ |
| 7992 | cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) |
| 7993 | < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1))) |
| 7994 | ? -1 |
| 7995 | : 1; |
| 7996 | } |
| 7997 | } |
| 7998 | cmp_done: ; |
| 7999 | #endif |
| 8000 | } |
| 8001 | |
| 8002 | SvREFCNT_dec(svrecode); |
| 8003 | |
| 8004 | return cmp; |
| 8005 | } |
| 8006 | |
| 8007 | /* |
| 8008 | =for apidoc sv_cmp_locale |
| 8009 | |
| 8010 | Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and |
| 8011 | S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings |
| 8012 | if necessary. See also C<L</sv_cmp>>. |
| 8013 | |
| 8014 | =for apidoc sv_cmp_locale_flags |
| 8015 | |
| 8016 | Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and |
| 8017 | S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If |
| 8018 | the flags contain C<SV_GMAGIC>, it handles get magic. See also |
| 8019 | C<L</sv_cmp_flags>>. |
| 8020 | |
| 8021 | =cut |
| 8022 | */ |
| 8023 | |
| 8024 | I32 |
| 8025 | Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2) |
| 8026 | { |
| 8027 | return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC); |
| 8028 | } |
| 8029 | |
| 8030 | I32 |
| 8031 | Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2, |
| 8032 | const U32 flags) |
| 8033 | { |
| 8034 | #ifdef USE_LOCALE_COLLATE |
| 8035 | |
| 8036 | char *pv1, *pv2; |
| 8037 | STRLEN len1, len2; |
| 8038 | I32 retval; |
| 8039 | |
| 8040 | if (PL_collation_standard) |
| 8041 | goto raw_compare; |
| 8042 | |
| 8043 | len1 = 0; |
| 8044 | pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL; |
| 8045 | len2 = 0; |
| 8046 | pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL; |
| 8047 | |
| 8048 | if (!pv1 || !len1) { |
| 8049 | if (pv2 && len2) |
| 8050 | return -1; |
| 8051 | else |
| 8052 | goto raw_compare; |
| 8053 | } |
| 8054 | else { |
| 8055 | if (!pv2 || !len2) |
| 8056 | return 1; |
| 8057 | } |
| 8058 | |
| 8059 | retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2); |
| 8060 | |
| 8061 | if (retval) |
| 8062 | return retval < 0 ? -1 : 1; |
| 8063 | |
| 8064 | /* |
| 8065 | * When the result of collation is equality, that doesn't mean |
| 8066 | * that there are no differences -- some locales exclude some |
| 8067 | * characters from consideration. So to avoid false equalities, |
| 8068 | * we use the raw string as a tiebreaker. |
| 8069 | */ |
| 8070 | |
| 8071 | raw_compare: |
| 8072 | /* FALLTHROUGH */ |
| 8073 | |
| 8074 | #else |
| 8075 | PERL_UNUSED_ARG(flags); |
| 8076 | #endif /* USE_LOCALE_COLLATE */ |
| 8077 | |
| 8078 | return sv_cmp(sv1, sv2); |
| 8079 | } |
| 8080 | |
| 8081 | |
| 8082 | #ifdef USE_LOCALE_COLLATE |
| 8083 | |
| 8084 | /* |
| 8085 | =for apidoc sv_collxfrm |
| 8086 | |
| 8087 | This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See |
| 8088 | C<L</sv_collxfrm_flags>>. |
| 8089 | |
| 8090 | =for apidoc sv_collxfrm_flags |
| 8091 | |
| 8092 | Add Collate Transform magic to an SV if it doesn't already have it. If the |
| 8093 | flags contain C<SV_GMAGIC>, it handles get-magic. |
| 8094 | |
| 8095 | Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the |
| 8096 | scalar data of the variable, but transformed to such a format that a normal |
| 8097 | memory comparison can be used to compare the data according to the locale |
| 8098 | settings. |
| 8099 | |
| 8100 | =cut |
| 8101 | */ |
| 8102 | |
| 8103 | char * |
| 8104 | Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags) |
| 8105 | { |
| 8106 | MAGIC *mg; |
| 8107 | |
| 8108 | PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS; |
| 8109 | |
| 8110 | mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL; |
| 8111 | if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) { |
| 8112 | const char *s; |
| 8113 | char *xf; |
| 8114 | STRLEN len, xlen; |
| 8115 | |
| 8116 | if (mg) |
| 8117 | Safefree(mg->mg_ptr); |
| 8118 | s = SvPV_flags_const(sv, len, flags); |
| 8119 | if ((xf = mem_collxfrm(s, len, &xlen))) { |
| 8120 | if (! mg) { |
| 8121 | mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm, |
| 8122 | 0, 0); |
| 8123 | assert(mg); |
| 8124 | } |
| 8125 | mg->mg_ptr = xf; |
| 8126 | mg->mg_len = xlen; |
| 8127 | } |
| 8128 | else { |
| 8129 | if (mg) { |
| 8130 | mg->mg_ptr = NULL; |
| 8131 | mg->mg_len = -1; |
| 8132 | } |
| 8133 | } |
| 8134 | } |
| 8135 | if (mg && mg->mg_ptr) { |
| 8136 | *nxp = mg->mg_len; |
| 8137 | return mg->mg_ptr + sizeof(PL_collation_ix); |
| 8138 | } |
| 8139 | else { |
| 8140 | *nxp = 0; |
| 8141 | return NULL; |
| 8142 | } |
| 8143 | } |
| 8144 | |
| 8145 | #endif /* USE_LOCALE_COLLATE */ |
| 8146 | |
| 8147 | static char * |
| 8148 | S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append) |
| 8149 | { |
| 8150 | SV * const tsv = newSV(0); |
| 8151 | ENTER; |
| 8152 | SAVEFREESV(tsv); |
| 8153 | sv_gets(tsv, fp, 0); |
| 8154 | sv_utf8_upgrade_nomg(tsv); |
| 8155 | SvCUR_set(sv,append); |
| 8156 | sv_catsv(sv,tsv); |
| 8157 | LEAVE; |
| 8158 | return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; |
| 8159 | } |
| 8160 | |
| 8161 | static char * |
| 8162 | S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append) |
| 8163 | { |
| 8164 | SSize_t bytesread; |
| 8165 | const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */ |
| 8166 | /* Grab the size of the record we're getting */ |
| 8167 | char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append; |
| 8168 | |
| 8169 | /* Go yank in */ |
| 8170 | #ifdef __VMS |
| 8171 | int fd; |
| 8172 | Stat_t st; |
| 8173 | |
| 8174 | /* With a true, record-oriented file on VMS, we need to use read directly |
| 8175 | * to ensure that we respect RMS record boundaries. The user is responsible |
| 8176 | * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum |
| 8177 | * record size) field. N.B. This is likely to produce invalid results on |
| 8178 | * varying-width character data when a record ends mid-character. |
| 8179 | */ |
| 8180 | fd = PerlIO_fileno(fp); |
| 8181 | if (fd != -1 |
| 8182 | && PerlLIO_fstat(fd, &st) == 0 |
| 8183 | && (st.st_fab_rfm == FAB$C_VAR |
| 8184 | || st.st_fab_rfm == FAB$C_VFC |
| 8185 | || st.st_fab_rfm == FAB$C_FIX)) { |
| 8186 | |
| 8187 | bytesread = PerlLIO_read(fd, buffer, recsize); |
| 8188 | } |
| 8189 | else /* in-memory file from PerlIO::Scalar |
| 8190 | * or not a record-oriented file |
| 8191 | */ |
| 8192 | #endif |
| 8193 | { |
| 8194 | bytesread = PerlIO_read(fp, buffer, recsize); |
| 8195 | |
| 8196 | /* At this point, the logic in sv_get() means that sv will |
| 8197 | be treated as utf-8 if the handle is utf8. |
| 8198 | */ |
| 8199 | if (PerlIO_isutf8(fp) && bytesread > 0) { |
| 8200 | char *bend = buffer + bytesread; |
| 8201 | char *bufp = buffer; |
| 8202 | size_t charcount = 0; |
| 8203 | bool charstart = TRUE; |
| 8204 | STRLEN skip = 0; |
| 8205 | |
| 8206 | while (charcount < recsize) { |
| 8207 | /* count accumulated characters */ |
| 8208 | while (bufp < bend) { |
| 8209 | if (charstart) { |
| 8210 | skip = UTF8SKIP(bufp); |
| 8211 | } |
| 8212 | if (bufp + skip > bend) { |
| 8213 | /* partial at the end */ |
| 8214 | charstart = FALSE; |
| 8215 | break; |
| 8216 | } |
| 8217 | else { |
| 8218 | ++charcount; |
| 8219 | bufp += skip; |
| 8220 | charstart = TRUE; |
| 8221 | } |
| 8222 | } |
| 8223 | |
| 8224 | if (charcount < recsize) { |
| 8225 | STRLEN readsize; |
| 8226 | STRLEN bufp_offset = bufp - buffer; |
| 8227 | SSize_t morebytesread; |
| 8228 | |
| 8229 | /* originally I read enough to fill any incomplete |
| 8230 | character and the first byte of the next |
| 8231 | character if needed, but if there's many |
| 8232 | multi-byte encoded characters we're going to be |
| 8233 | making a read call for every character beyond |
| 8234 | the original read size. |
| 8235 | |
| 8236 | So instead, read the rest of the character if |
| 8237 | any, and enough bytes to match at least the |
| 8238 | start bytes for each character we're going to |
| 8239 | read. |
| 8240 | */ |
| 8241 | if (charstart) |
| 8242 | readsize = recsize - charcount; |
| 8243 | else |
| 8244 | readsize = skip - (bend - bufp) + recsize - charcount - 1; |
| 8245 | buffer = SvGROW(sv, append + bytesread + readsize + 1) + append; |
| 8246 | bend = buffer + bytesread; |
| 8247 | morebytesread = PerlIO_read(fp, bend, readsize); |
| 8248 | if (morebytesread <= 0) { |
| 8249 | /* we're done, if we still have incomplete |
| 8250 | characters the check code in sv_gets() will |
| 8251 | warn about them. |
| 8252 | |
| 8253 | I'd originally considered doing |
| 8254 | PerlIO_ungetc() on all but the lead |
| 8255 | character of the incomplete character, but |
| 8256 | read() doesn't do that, so I don't. |
| 8257 | */ |
| 8258 | break; |
| 8259 | } |
| 8260 | |
| 8261 | /* prepare to scan some more */ |
| 8262 | bytesread += morebytesread; |
| 8263 | bend = buffer + bytesread; |
| 8264 | bufp = buffer + bufp_offset; |
| 8265 | } |
| 8266 | } |
| 8267 | } |
| 8268 | } |
| 8269 | |
| 8270 | if (bytesread < 0) |
| 8271 | bytesread = 0; |
| 8272 | SvCUR_set(sv, bytesread + append); |
| 8273 | buffer[bytesread] = '\0'; |
| 8274 | return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; |
| 8275 | } |
| 8276 | |
| 8277 | /* |
| 8278 | =for apidoc sv_gets |
| 8279 | |
| 8280 | Get a line from the filehandle and store it into the SV, optionally |
| 8281 | appending to the currently-stored string. If C<append> is not 0, the |
| 8282 | line is appended to the SV instead of overwriting it. C<append> should |
| 8283 | be set to the byte offset that the appended string should start at |
| 8284 | in the SV (typically, C<SvCUR(sv)> is a suitable choice). |
| 8285 | |
| 8286 | =cut |
| 8287 | */ |
| 8288 | |
| 8289 | char * |
| 8290 | Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append) |
| 8291 | { |
| 8292 | const char *rsptr; |
| 8293 | STRLEN rslen; |
| 8294 | STDCHAR rslast; |
| 8295 | STDCHAR *bp; |
| 8296 | SSize_t cnt; |
| 8297 | int i = 0; |
| 8298 | int rspara = 0; |
| 8299 | |
| 8300 | PERL_ARGS_ASSERT_SV_GETS; |
| 8301 | |
| 8302 | if (SvTHINKFIRST(sv)) |
| 8303 | sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV); |
| 8304 | /* XXX. If you make this PVIV, then copy on write can copy scalars read |
| 8305 | from <>. |
| 8306 | However, perlbench says it's slower, because the existing swipe code |
| 8307 | is faster than copy on write. |
| 8308 | Swings and roundabouts. */ |
| 8309 | SvUPGRADE(sv, SVt_PV); |
| 8310 | |
| 8311 | if (append) { |
| 8312 | /* line is going to be appended to the existing buffer in the sv */ |
| 8313 | if (PerlIO_isutf8(fp)) { |
| 8314 | if (!SvUTF8(sv)) { |
| 8315 | sv_utf8_upgrade_nomg(sv); |
| 8316 | sv_pos_u2b(sv,&append,0); |
| 8317 | } |
| 8318 | } else if (SvUTF8(sv)) { |
| 8319 | return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append); |
| 8320 | } |
| 8321 | } |
| 8322 | |
| 8323 | SvPOK_only(sv); |
| 8324 | if (!append) { |
| 8325 | /* not appending - "clear" the string by setting SvCUR to 0, |
| 8326 | * the pv is still avaiable. */ |
| 8327 | SvCUR_set(sv,0); |
| 8328 | } |
| 8329 | if (PerlIO_isutf8(fp)) |
| 8330 | SvUTF8_on(sv); |
| 8331 | |
| 8332 | if (IN_PERL_COMPILETIME) { |
| 8333 | /* we always read code in line mode */ |
| 8334 | rsptr = "\n"; |
| 8335 | rslen = 1; |
| 8336 | } |
| 8337 | else if (RsSNARF(PL_rs)) { |
| 8338 | /* If it is a regular disk file use size from stat() as estimate |
| 8339 | of amount we are going to read -- may result in mallocing |
| 8340 | more memory than we really need if the layers below reduce |
| 8341 | the size we read (e.g. CRLF or a gzip layer). |
| 8342 | */ |
| 8343 | Stat_t st; |
| 8344 | int fd = PerlIO_fileno(fp); |
| 8345 | if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) { |
| 8346 | const Off_t offset = PerlIO_tell(fp); |
| 8347 | if (offset != (Off_t) -1 && st.st_size + append > offset) { |
| 8348 | #ifdef PERL_COPY_ON_WRITE |
| 8349 | /* Add an extra byte for the sake of copy-on-write's |
| 8350 | * buffer reference count. */ |
| 8351 | (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2)); |
| 8352 | #else |
| 8353 | (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1)); |
| 8354 | #endif |
| 8355 | } |
| 8356 | } |
| 8357 | rsptr = NULL; |
| 8358 | rslen = 0; |
| 8359 | } |
| 8360 | else if (RsRECORD(PL_rs)) { |
| 8361 | return S_sv_gets_read_record(aTHX_ sv, fp, append); |
| 8362 | } |
| 8363 | else if (RsPARA(PL_rs)) { |
| 8364 | rsptr = "\n\n"; |
| 8365 | rslen = 2; |
| 8366 | rspara = 1; |
| 8367 | } |
| 8368 | else { |
| 8369 | /* Get $/ i.e. PL_rs into same encoding as stream wants */ |
| 8370 | if (PerlIO_isutf8(fp)) { |
| 8371 | rsptr = SvPVutf8(PL_rs, rslen); |
| 8372 | } |
| 8373 | else { |
| 8374 | if (SvUTF8(PL_rs)) { |
| 8375 | if (!sv_utf8_downgrade(PL_rs, TRUE)) { |
| 8376 | Perl_croak(aTHX_ "Wide character in $/"); |
| 8377 | } |
| 8378 | } |
| 8379 | /* extract the raw pointer to the record separator */ |
| 8380 | rsptr = SvPV_const(PL_rs, rslen); |
| 8381 | } |
| 8382 | } |
| 8383 | |
| 8384 | /* rslast is the last character in the record separator |
| 8385 | * note we don't use rslast except when rslen is true, so the |
| 8386 | * null assign is a placeholder. */ |
| 8387 | rslast = rslen ? rsptr[rslen - 1] : '\0'; |
| 8388 | |
| 8389 | if (rspara) { /* have to do this both before and after */ |
| 8390 | do { /* to make sure file boundaries work right */ |
| 8391 | if (PerlIO_eof(fp)) |
| 8392 | return 0; |
| 8393 | i = PerlIO_getc(fp); |
| 8394 | if (i != '\n') { |
| 8395 | if (i == -1) |
| 8396 | return 0; |
| 8397 | PerlIO_ungetc(fp,i); |
| 8398 | break; |
| 8399 | } |
| 8400 | } while (i != EOF); |
| 8401 | } |
| 8402 | |
| 8403 | /* See if we know enough about I/O mechanism to cheat it ! */ |
| 8404 | |
| 8405 | /* This used to be #ifdef test - it is made run-time test for ease |
| 8406 | of abstracting out stdio interface. One call should be cheap |
| 8407 | enough here - and may even be a macro allowing compile |
| 8408 | time optimization. |
| 8409 | */ |
| 8410 | |
| 8411 | if (PerlIO_fast_gets(fp)) { |
| 8412 | /* |
| 8413 | * We can do buffer based IO operations on this filehandle. |
| 8414 | * |
| 8415 | * This means we can bypass a lot of subcalls and process |
| 8416 | * the buffer directly, it also means we know the upper bound |
| 8417 | * on the amount of data we might read of the current buffer |
| 8418 | * into our sv. Knowing this allows us to preallocate the pv |
| 8419 | * to be able to hold that maximum, which allows us to simplify |
| 8420 | * a lot of logic. */ |
| 8421 | |
| 8422 | /* |
| 8423 | * We're going to steal some values from the stdio struct |
| 8424 | * and put EVERYTHING in the innermost loop into registers. |
| 8425 | */ |
| 8426 | STDCHAR *ptr; /* pointer into fp's read-ahead buffer */ |
| 8427 | STRLEN bpx; /* length of the data in the target sv |
| 8428 | used to fix pointers after a SvGROW */ |
| 8429 | I32 shortbuffered; /* If the pv buffer is shorter than the amount |
| 8430 | of data left in the read-ahead buffer. |
| 8431 | If 0 then the pv buffer can hold the full |
| 8432 | amount left, otherwise this is the amount it |
| 8433 | can hold. */ |
| 8434 | |
| 8435 | /* Here is some breathtakingly efficient cheating */ |
| 8436 | |
| 8437 | /* When you read the following logic resist the urge to think |
| 8438 | * of record separators that are 1 byte long. They are an |
| 8439 | * uninteresting special (simple) case. |
| 8440 | * |
| 8441 | * Instead think of record separators which are at least 2 bytes |
| 8442 | * long, and keep in mind that we need to deal with such |
| 8443 | * separators when they cross a read-ahead buffer boundary. |
| 8444 | * |
| 8445 | * Also consider that we need to gracefully deal with separators |
| 8446 | * that may be longer than a single read ahead buffer. |
| 8447 | * |
| 8448 | * Lastly do not forget we want to copy the delimiter as well. We |
| 8449 | * are copying all data in the file _up_to_and_including_ the separator |
| 8450 | * itself. |
| 8451 | * |
| 8452 | * Now that you have all that in mind here is what is happening below: |
| 8453 | * |
| 8454 | * 1. When we first enter the loop we do some memory book keeping to see |
| 8455 | * how much free space there is in the target SV. (This sub assumes that |
| 8456 | * it is operating on the same SV most of the time via $_ and that it is |
| 8457 | * going to be able to reuse the same pv buffer each call.) If there is |
| 8458 | * "enough" room then we set "shortbuffered" to how much space there is |
| 8459 | * and start reading forward. |
| 8460 | * |
| 8461 | * 2. When we scan forward we copy from the read-ahead buffer to the target |
| 8462 | * SV's pv buffer. While we go we watch for the end of the read-ahead buffer, |
| 8463 | * and the end of the of pv, as well as for the "rslast", which is the last |
| 8464 | * char of the separator. |
| 8465 | * |
| 8466 | * 3. When scanning forward if we see rslast then we jump backwards in *pv* |
| 8467 | * (which has a "complete" record up to the point we saw rslast) and check |
| 8468 | * it to see if it matches the separator. If it does we are done. If it doesn't |
| 8469 | * we continue on with the scan/copy. |
| 8470 | * |
| 8471 | * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get |
| 8472 | * the IO system to read the next buffer. We do this by doing a getc(), which |
| 8473 | * returns a single char read (or EOF), and prefills the buffer, and also |
| 8474 | * allows us to find out how full the buffer is. We use this information to |
| 8475 | * SvGROW() the sv to the size remaining in the buffer, after which we copy |
| 8476 | * the returned single char into the target sv, and then go back into scan |
| 8477 | * forward mode. |
| 8478 | * |
| 8479 | * 5. If we run out of write-buffer then we SvGROW() it by the size of the |
| 8480 | * remaining space in the read-buffer. |
| 8481 | * |
| 8482 | * Note that this code despite its twisty-turny nature is pretty darn slick. |
| 8483 | * It manages single byte separators, multi-byte cross boundary separators, |
| 8484 | * and cross-read-buffer separators cleanly and efficiently at the cost |
| 8485 | * of potentially greatly overallocating the target SV. |
| 8486 | * |
| 8487 | * Yves |
| 8488 | */ |
| 8489 | |
| 8490 | |
| 8491 | /* get the number of bytes remaining in the read-ahead buffer |
| 8492 | * on first call on a given fp this will return 0.*/ |
| 8493 | cnt = PerlIO_get_cnt(fp); |
| 8494 | |
| 8495 | /* make sure we have the room */ |
| 8496 | if ((I32)(SvLEN(sv) - append) <= cnt + 1) { |
| 8497 | /* Not room for all of it |
| 8498 | if we are looking for a separator and room for some |
| 8499 | */ |
| 8500 | if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) { |
| 8501 | /* just process what we have room for */ |
| 8502 | shortbuffered = cnt - SvLEN(sv) + append + 1; |
| 8503 | cnt -= shortbuffered; |
| 8504 | } |
| 8505 | else { |
| 8506 | /* ensure that the target sv has enough room to hold |
| 8507 | * the rest of the read-ahead buffer */ |
| 8508 | shortbuffered = 0; |
| 8509 | /* remember that cnt can be negative */ |
| 8510 | SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1)))); |
| 8511 | } |
| 8512 | } |
| 8513 | else { |
| 8514 | /* we have enough room to hold the full buffer, lets scream */ |
| 8515 | shortbuffered = 0; |
| 8516 | } |
| 8517 | |
| 8518 | /* extract the pointer to sv's string buffer, offset by append as necessary */ |
| 8519 | bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */ |
| 8520 | /* extract the point to the read-ahead buffer */ |
| 8521 | ptr = (STDCHAR*)PerlIO_get_ptr(fp); |
| 8522 | |
| 8523 | /* some trace debug output */ |
| 8524 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8525 | "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt)); |
| 8526 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8527 | "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%" |
| 8528 | UVuf"\n", |
| 8529 | PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), |
| 8530 | PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0))); |
| 8531 | |
| 8532 | for (;;) { |
| 8533 | screamer: |
| 8534 | /* if there is stuff left in the read-ahead buffer */ |
| 8535 | if (cnt > 0) { |
| 8536 | /* if there is a separator */ |
| 8537 | if (rslen) { |
| 8538 | /* loop until we hit the end of the read-ahead buffer */ |
| 8539 | while (cnt > 0) { /* this | eat */ |
| 8540 | /* scan forward copying and searching for rslast as we go */ |
| 8541 | cnt--; |
| 8542 | if ((*bp++ = *ptr++) == rslast) /* really | dust */ |
| 8543 | goto thats_all_folks; /* screams | sed :-) */ |
| 8544 | } |
| 8545 | } |
| 8546 | else { |
| 8547 | /* no separator, slurp the full buffer */ |
| 8548 | Copy(ptr, bp, cnt, char); /* this | eat */ |
| 8549 | bp += cnt; /* screams | dust */ |
| 8550 | ptr += cnt; /* louder | sed :-) */ |
| 8551 | cnt = 0; |
| 8552 | assert (!shortbuffered); |
| 8553 | goto cannot_be_shortbuffered; |
| 8554 | } |
| 8555 | } |
| 8556 | |
| 8557 | if (shortbuffered) { /* oh well, must extend */ |
| 8558 | /* we didnt have enough room to fit the line into the target buffer |
| 8559 | * so we must extend the target buffer and keep going */ |
| 8560 | cnt = shortbuffered; |
| 8561 | shortbuffered = 0; |
| 8562 | bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */ |
| 8563 | SvCUR_set(sv, bpx); |
| 8564 | /* extned the target sv's buffer so it can hold the full read-ahead buffer */ |
| 8565 | SvGROW(sv, SvLEN(sv) + append + cnt + 2); |
| 8566 | bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */ |
| 8567 | continue; |
| 8568 | } |
| 8569 | |
| 8570 | cannot_be_shortbuffered: |
| 8571 | /* we need to refill the read-ahead buffer if possible */ |
| 8572 | |
| 8573 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8574 | "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n", |
| 8575 | PTR2UV(ptr),(IV)cnt)); |
| 8576 | PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */ |
| 8577 | |
| 8578 | DEBUG_Pv(PerlIO_printf(Perl_debug_log, |
| 8579 | "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n", |
| 8580 | PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), |
| 8581 | PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); |
| 8582 | |
| 8583 | /* |
| 8584 | call PerlIO_getc() to let it prefill the lookahead buffer |
| 8585 | |
| 8586 | This used to call 'filbuf' in stdio form, but as that behaves like |
| 8587 | getc when cnt <= 0 we use PerlIO_getc here to avoid introducing |
| 8588 | another abstraction. |
| 8589 | |
| 8590 | Note we have to deal with the char in 'i' if we are not at EOF |
| 8591 | */ |
| 8592 | i = PerlIO_getc(fp); /* get more characters */ |
| 8593 | |
| 8594 | DEBUG_Pv(PerlIO_printf(Perl_debug_log, |
| 8595 | "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n", |
| 8596 | PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), |
| 8597 | PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); |
| 8598 | |
| 8599 | /* find out how much is left in the read-ahead buffer, and rextract its pointer */ |
| 8600 | cnt = PerlIO_get_cnt(fp); |
| 8601 | ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */ |
| 8602 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8603 | "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n", |
| 8604 | PTR2UV(ptr),(IV)cnt)); |
| 8605 | |
| 8606 | if (i == EOF) /* all done for ever? */ |
| 8607 | goto thats_really_all_folks; |
| 8608 | |
| 8609 | /* make sure we have enough space in the target sv */ |
| 8610 | bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */ |
| 8611 | SvCUR_set(sv, bpx); |
| 8612 | SvGROW(sv, bpx + cnt + 2); |
| 8613 | bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */ |
| 8614 | |
| 8615 | /* copy of the char we got from getc() */ |
| 8616 | *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */ |
| 8617 | |
| 8618 | /* make sure we deal with the i being the last character of a separator */ |
| 8619 | if (rslen && (STDCHAR)i == rslast) /* all done for now? */ |
| 8620 | goto thats_all_folks; |
| 8621 | } |
| 8622 | |
| 8623 | thats_all_folks: |
| 8624 | /* check if we have actually found the separator - only really applies |
| 8625 | * when rslen > 1 */ |
| 8626 | if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) || |
| 8627 | memNE((char*)bp - rslen, rsptr, rslen)) |
| 8628 | goto screamer; /* go back to the fray */ |
| 8629 | thats_really_all_folks: |
| 8630 | if (shortbuffered) |
| 8631 | cnt += shortbuffered; |
| 8632 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8633 | "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt)); |
| 8634 | PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */ |
| 8635 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8636 | "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf |
| 8637 | "\n", |
| 8638 | PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp), |
| 8639 | PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0))); |
| 8640 | *bp = '\0'; |
| 8641 | SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */ |
| 8642 | DEBUG_P(PerlIO_printf(Perl_debug_log, |
| 8643 | "Screamer: done, len=%ld, string=|%.*s|\n", |
| 8644 | (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv))); |
| 8645 | } |
| 8646 | else |
| 8647 | { |
| 8648 | /*The big, slow, and stupid way. */ |
| 8649 | #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */ |
| 8650 | STDCHAR *buf = NULL; |
| 8651 | Newx(buf, 8192, STDCHAR); |
| 8652 | assert(buf); |
| 8653 | #else |
| 8654 | STDCHAR buf[8192]; |
| 8655 | #endif |
| 8656 | |
| 8657 | screamer2: |
| 8658 | if (rslen) { |
| 8659 | const STDCHAR * const bpe = buf + sizeof(buf); |
| 8660 | bp = buf; |
| 8661 | while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe) |
| 8662 | ; /* keep reading */ |
| 8663 | cnt = bp - buf; |
| 8664 | } |
| 8665 | else { |
| 8666 | cnt = PerlIO_read(fp,(char*)buf, sizeof(buf)); |
| 8667 | /* Accommodate broken VAXC compiler, which applies U8 cast to |
| 8668 | * both args of ?: operator, causing EOF to change into 255 |
| 8669 | */ |
| 8670 | if (cnt > 0) |
| 8671 | i = (U8)buf[cnt - 1]; |
| 8672 | else |
| 8673 | i = EOF; |
| 8674 | } |
| 8675 | |
| 8676 | if (cnt < 0) |
| 8677 | cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */ |
| 8678 | if (append) |
| 8679 | sv_catpvn_nomg(sv, (char *) buf, cnt); |
| 8680 | else |
| 8681 | sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */ |
| 8682 | |
| 8683 | if (i != EOF && /* joy */ |
| 8684 | (!rslen || |
| 8685 | SvCUR(sv) < rslen || |
| 8686 | memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen))) |
| 8687 | { |
| 8688 | append = -1; |
| 8689 | /* |
| 8690 | * If we're reading from a TTY and we get a short read, |
| 8691 | * indicating that the user hit his EOF character, we need |
| 8692 | * to notice it now, because if we try to read from the TTY |
| 8693 | * again, the EOF condition will disappear. |
| 8694 | * |
| 8695 | * The comparison of cnt to sizeof(buf) is an optimization |
| 8696 | * that prevents unnecessary calls to feof(). |
| 8697 | * |
| 8698 | * - jik 9/25/96 |
| 8699 | */ |
| 8700 | if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp))) |
| 8701 | goto screamer2; |
| 8702 | } |
| 8703 | |
| 8704 | #ifdef USE_HEAP_INSTEAD_OF_STACK |
| 8705 | Safefree(buf); |
| 8706 | #endif |
| 8707 | } |
| 8708 | |
| 8709 | if (rspara) { /* have to do this both before and after */ |
| 8710 | while (i != EOF) { /* to make sure file boundaries work right */ |
| 8711 | i = PerlIO_getc(fp); |
| 8712 | if (i != '\n') { |
| 8713 | PerlIO_ungetc(fp,i); |
| 8714 | break; |
| 8715 | } |
| 8716 | } |
| 8717 | } |
| 8718 | |
| 8719 | return (SvCUR(sv) - append) ? SvPVX(sv) : NULL; |
| 8720 | } |
| 8721 | |
| 8722 | /* |
| 8723 | =for apidoc sv_inc |
| 8724 | |
| 8725 | Auto-increment of the value in the SV, doing string to numeric conversion |
| 8726 | if necessary. Handles 'get' magic and operator overloading. |
| 8727 | |
| 8728 | =cut |
| 8729 | */ |
| 8730 | |
| 8731 | void |
| 8732 | Perl_sv_inc(pTHX_ SV *const sv) |
| 8733 | { |
| 8734 | if (!sv) |
| 8735 | return; |
| 8736 | SvGETMAGIC(sv); |
| 8737 | sv_inc_nomg(sv); |
| 8738 | } |
| 8739 | |
| 8740 | /* |
| 8741 | =for apidoc sv_inc_nomg |
| 8742 | |
| 8743 | Auto-increment of the value in the SV, doing string to numeric conversion |
| 8744 | if necessary. Handles operator overloading. Skips handling 'get' magic. |
| 8745 | |
| 8746 | =cut |
| 8747 | */ |
| 8748 | |
| 8749 | void |
| 8750 | Perl_sv_inc_nomg(pTHX_ SV *const sv) |
| 8751 | { |
| 8752 | char *d; |
| 8753 | int flags; |
| 8754 | |
| 8755 | if (!sv) |
| 8756 | return; |
| 8757 | if (SvTHINKFIRST(sv)) { |
| 8758 | if (SvREADONLY(sv)) { |
| 8759 | Perl_croak_no_modify(); |
| 8760 | } |
| 8761 | if (SvROK(sv)) { |
| 8762 | IV i; |
| 8763 | if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg)) |
| 8764 | return; |
| 8765 | i = PTR2IV(SvRV(sv)); |
| 8766 | sv_unref(sv); |
| 8767 | sv_setiv(sv, i); |
| 8768 | } |
| 8769 | else sv_force_normal_flags(sv, 0); |
| 8770 | } |
| 8771 | flags = SvFLAGS(sv); |
| 8772 | if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) { |
| 8773 | /* It's (privately or publicly) a float, but not tested as an |
| 8774 | integer, so test it to see. */ |
| 8775 | (void) SvIV(sv); |
| 8776 | flags = SvFLAGS(sv); |
| 8777 | } |
| 8778 | if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) { |
| 8779 | /* It's publicly an integer, or privately an integer-not-float */ |
| 8780 | #ifdef PERL_PRESERVE_IVUV |
| 8781 | oops_its_int: |
| 8782 | #endif |
| 8783 | if (SvIsUV(sv)) { |
| 8784 | if (SvUVX(sv) == UV_MAX) |
| 8785 | sv_setnv(sv, UV_MAX_P1); |
| 8786 | else |
| 8787 | (void)SvIOK_only_UV(sv); |
| 8788 | SvUV_set(sv, SvUVX(sv) + 1); |
| 8789 | } else { |
| 8790 | if (SvIVX(sv) == IV_MAX) |
| 8791 | sv_setuv(sv, (UV)IV_MAX + 1); |
| 8792 | else { |
| 8793 | (void)SvIOK_only(sv); |
| 8794 | SvIV_set(sv, SvIVX(sv) + 1); |
| 8795 | } |
| 8796 | } |
| 8797 | return; |
| 8798 | } |
| 8799 | if (flags & SVp_NOK) { |
| 8800 | const NV was = SvNVX(sv); |
| 8801 | if (LIKELY(!Perl_isinfnan(was)) && |
| 8802 | NV_OVERFLOWS_INTEGERS_AT && |
| 8803 | was >= NV_OVERFLOWS_INTEGERS_AT) { |
| 8804 | /* diag_listed_as: Lost precision when %s %f by 1 */ |
| 8805 | Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION), |
| 8806 | "Lost precision when incrementing %" NVff " by 1", |
| 8807 | was); |
| 8808 | } |
| 8809 | (void)SvNOK_only(sv); |
| 8810 | SvNV_set(sv, was + 1.0); |
| 8811 | return; |
| 8812 | } |
| 8813 | |
| 8814 | /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */ |
| 8815 | if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv))) |
| 8816 | Perl_croak_no_modify(); |
| 8817 | |
| 8818 | if (!(flags & SVp_POK) || !*SvPVX_const(sv)) { |
| 8819 | if ((flags & SVTYPEMASK) < SVt_PVIV) |
| 8820 | sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV)); |
| 8821 | (void)SvIOK_only(sv); |
| 8822 | SvIV_set(sv, 1); |
| 8823 | return; |
| 8824 | } |
| 8825 | d = SvPVX(sv); |
| 8826 | while (isALPHA(*d)) d++; |
| 8827 | while (isDIGIT(*d)) d++; |
| 8828 | if (d < SvEND(sv)) { |
| 8829 | const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING); |
| 8830 | #ifdef PERL_PRESERVE_IVUV |
| 8831 | /* Got to punt this as an integer if needs be, but we don't issue |
| 8832 | warnings. Probably ought to make the sv_iv_please() that does |
| 8833 | the conversion if possible, and silently. */ |
| 8834 | if (numtype && !(numtype & IS_NUMBER_INFINITY)) { |
| 8835 | /* Need to try really hard to see if it's an integer. |
| 8836 | 9.22337203685478e+18 is an integer. |
| 8837 | but "9.22337203685478e+18" + 0 is UV=9223372036854779904 |
| 8838 | so $a="9.22337203685478e+18"; $a+0; $a++ |
| 8839 | needs to be the same as $a="9.22337203685478e+18"; $a++ |
| 8840 | or we go insane. */ |
| 8841 | |
| 8842 | (void) sv_2iv(sv); |
| 8843 | if (SvIOK(sv)) |
| 8844 | goto oops_its_int; |
| 8845 | |
| 8846 | /* sv_2iv *should* have made this an NV */ |
| 8847 | if (flags & SVp_NOK) { |
| 8848 | (void)SvNOK_only(sv); |
| 8849 | SvNV_set(sv, SvNVX(sv) + 1.0); |
| 8850 | return; |
| 8851 | } |
| 8852 | /* I don't think we can get here. Maybe I should assert this |
| 8853 | And if we do get here I suspect that sv_setnv will croak. NWC |
| 8854 | Fall through. */ |
| 8855 | DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n", |
| 8856 | SvPVX_const(sv), SvIVX(sv), SvNVX(sv))); |
| 8857 | } |
| 8858 | #endif /* PERL_PRESERVE_IVUV */ |
| 8859 | if (!numtype && ckWARN(WARN_NUMERIC)) |
| 8860 | not_incrementable(sv); |
| 8861 | sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0); |
| 8862 | return; |
| 8863 | } |
| 8864 | d--; |
| 8865 | while (d >= SvPVX_const(sv)) { |
| 8866 | if (isDIGIT(*d)) { |
| 8867 | if (++*d <= '9') |
| 8868 | return; |
| 8869 | *(d--) = '0'; |
| 8870 | } |
| 8871 | else { |
| 8872 | #ifdef EBCDIC |
| 8873 | /* MKS: The original code here died if letters weren't consecutive. |
| 8874 | * at least it didn't have to worry about non-C locales. The |
| 8875 | * new code assumes that ('z'-'a')==('Z'-'A'), letters are |
| 8876 | * arranged in order (although not consecutively) and that only |
| 8877 | * [A-Za-z] are accepted by isALPHA in the C locale. |
| 8878 | */ |
| 8879 | if (isALPHA_FOLD_NE(*d, 'z')) { |
| 8880 | do { ++*d; } while (!isALPHA(*d)); |
| 8881 | return; |
| 8882 | } |
| 8883 | *(d--) -= 'z' - 'a'; |
| 8884 | #else |
| 8885 | ++*d; |
| 8886 | if (isALPHA(*d)) |
| 8887 | return; |
| 8888 | *(d--) -= 'z' - 'a' + 1; |
| 8889 | #endif |
| 8890 | } |
| 8891 | } |
| 8892 | /* oh,oh, the number grew */ |
| 8893 | SvGROW(sv, SvCUR(sv) + 2); |
| 8894 | SvCUR_set(sv, SvCUR(sv) + 1); |
| 8895 | for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--) |
| 8896 | *d = d[-1]; |
| 8897 | if (isDIGIT(d[1])) |
| 8898 | *d = '1'; |
| 8899 | else |
| 8900 | *d = d[1]; |
| 8901 | } |
| 8902 | |
| 8903 | /* |
| 8904 | =for apidoc sv_dec |
| 8905 | |
| 8906 | Auto-decrement of the value in the SV, doing string to numeric conversion |
| 8907 | if necessary. Handles 'get' magic and operator overloading. |
| 8908 | |
| 8909 | =cut |
| 8910 | */ |
| 8911 | |
| 8912 | void |
| 8913 | Perl_sv_dec(pTHX_ SV *const sv) |
| 8914 | { |
| 8915 | if (!sv) |
| 8916 | return; |
| 8917 | SvGETMAGIC(sv); |
| 8918 | sv_dec_nomg(sv); |
| 8919 | } |
| 8920 | |
| 8921 | /* |
| 8922 | =for apidoc sv_dec_nomg |
| 8923 | |
| 8924 | Auto-decrement of the value in the SV, doing string to numeric conversion |
| 8925 | if necessary. Handles operator overloading. Skips handling 'get' magic. |
| 8926 | |
| 8927 | =cut |
| 8928 | */ |
| 8929 | |
| 8930 | void |
| 8931 | Perl_sv_dec_nomg(pTHX_ SV *const sv) |
| 8932 | { |
| 8933 | int flags; |
| 8934 | |
| 8935 | if (!sv) |
| 8936 | return; |
| 8937 | if (SvTHINKFIRST(sv)) { |
| 8938 | if (SvREADONLY(sv)) { |
| 8939 | Perl_croak_no_modify(); |
| 8940 | } |
| 8941 | if (SvROK(sv)) { |
| 8942 | IV i; |
| 8943 | if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg)) |
| 8944 | return; |
| 8945 | i = PTR2IV(SvRV(sv)); |
| 8946 | sv_unref(sv); |
| 8947 | sv_setiv(sv, i); |
| 8948 | } |
| 8949 | else sv_force_normal_flags(sv, 0); |
| 8950 | } |
| 8951 | /* Unlike sv_inc we don't have to worry about string-never-numbers |
| 8952 | and keeping them magic. But we mustn't warn on punting */ |
| 8953 | flags = SvFLAGS(sv); |
| 8954 | if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) { |
| 8955 | /* It's publicly an integer, or privately an integer-not-float */ |
| 8956 | #ifdef PERL_PRESERVE_IVUV |
| 8957 | oops_its_int: |
| 8958 | #endif |
| 8959 | if (SvIsUV(sv)) { |
| 8960 | if (SvUVX(sv) == 0) { |
| 8961 | (void)SvIOK_only(sv); |
| 8962 | SvIV_set(sv, -1); |
| 8963 | } |
| 8964 | else { |
| 8965 | (void)SvIOK_only_UV(sv); |
| 8966 | SvUV_set(sv, SvUVX(sv) - 1); |
| 8967 | } |
| 8968 | } else { |
| 8969 | if (SvIVX(sv) == IV_MIN) { |
| 8970 | sv_setnv(sv, (NV)IV_MIN); |
| 8971 | goto oops_its_num; |
| 8972 | } |
| 8973 | else { |
| 8974 | (void)SvIOK_only(sv); |
| 8975 | SvIV_set(sv, SvIVX(sv) - 1); |
| 8976 | } |
| 8977 | } |
| 8978 | return; |
| 8979 | } |
| 8980 | if (flags & SVp_NOK) { |
| 8981 | oops_its_num: |
| 8982 | { |
| 8983 | const NV was = SvNVX(sv); |
| 8984 | if (LIKELY(!Perl_isinfnan(was)) && |
| 8985 | NV_OVERFLOWS_INTEGERS_AT && |
| 8986 | was <= -NV_OVERFLOWS_INTEGERS_AT) { |
| 8987 | /* diag_listed_as: Lost precision when %s %f by 1 */ |
| 8988 | Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION), |
| 8989 | "Lost precision when decrementing %" NVff " by 1", |
| 8990 | was); |
| 8991 | } |
| 8992 | (void)SvNOK_only(sv); |
| 8993 | SvNV_set(sv, was - 1.0); |
| 8994 | return; |
| 8995 | } |
| 8996 | } |
| 8997 | |
| 8998 | /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */ |
| 8999 | if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv))) |
| 9000 | Perl_croak_no_modify(); |
| 9001 | |
| 9002 | if (!(flags & SVp_POK)) { |
| 9003 | if ((flags & SVTYPEMASK) < SVt_PVIV) |
| 9004 | sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV); |
| 9005 | SvIV_set(sv, -1); |
| 9006 | (void)SvIOK_only(sv); |
| 9007 | return; |
| 9008 | } |
| 9009 | #ifdef PERL_PRESERVE_IVUV |
| 9010 | { |
| 9011 | const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL); |
| 9012 | if (numtype && !(numtype & IS_NUMBER_INFINITY)) { |
| 9013 | /* Need to try really hard to see if it's an integer. |
| 9014 | 9.22337203685478e+18 is an integer. |
| 9015 | but "9.22337203685478e+18" + 0 is UV=9223372036854779904 |
| 9016 | so $a="9.22337203685478e+18"; $a+0; $a-- |
| 9017 | needs to be the same as $a="9.22337203685478e+18"; $a-- |
| 9018 | or we go insane. */ |
| 9019 | |
| 9020 | (void) sv_2iv(sv); |
| 9021 | if (SvIOK(sv)) |
| 9022 | goto oops_its_int; |
| 9023 | |
| 9024 | /* sv_2iv *should* have made this an NV */ |
| 9025 | if (flags & SVp_NOK) { |
| 9026 | (void)SvNOK_only(sv); |
| 9027 | SvNV_set(sv, SvNVX(sv) - 1.0); |
| 9028 | return; |
| 9029 | } |
| 9030 | /* I don't think we can get here. Maybe I should assert this |
| 9031 | And if we do get here I suspect that sv_setnv will croak. NWC |
| 9032 | Fall through. */ |
| 9033 | DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n", |
| 9034 | SvPVX_const(sv), SvIVX(sv), SvNVX(sv))); |
| 9035 | } |
| 9036 | } |
| 9037 | #endif /* PERL_PRESERVE_IVUV */ |
| 9038 | sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */ |
| 9039 | } |
| 9040 | |
| 9041 | /* this define is used to eliminate a chunk of duplicated but shared logic |
| 9042 | * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be |
| 9043 | * used anywhere but here - yves |
| 9044 | */ |
| 9045 | #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \ |
| 9046 | STMT_START { \ |
| 9047 | SSize_t ix = ++PL_tmps_ix; \ |
| 9048 | if (UNLIKELY(ix >= PL_tmps_max)) \ |
| 9049 | ix = tmps_grow_p(ix); \ |
| 9050 | PL_tmps_stack[ix] = (AnSv); \ |
| 9051 | } STMT_END |
| 9052 | |
| 9053 | /* |
| 9054 | =for apidoc sv_mortalcopy |
| 9055 | |
| 9056 | Creates a new SV which is a copy of the original SV (using C<sv_setsv>). |
| 9057 | The new SV is marked as mortal. It will be destroyed "soon", either by an |
| 9058 | explicit call to C<FREETMPS>, or by an implicit call at places such as |
| 9059 | statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>. |
| 9060 | |
| 9061 | =cut |
| 9062 | */ |
| 9063 | |
| 9064 | /* Make a string that will exist for the duration of the expression |
| 9065 | * evaluation. Actually, it may have to last longer than that, but |
| 9066 | * hopefully we won't free it until it has been assigned to a |
| 9067 | * permanent location. */ |
| 9068 | |
| 9069 | SV * |
| 9070 | Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags) |
| 9071 | { |
| 9072 | SV *sv; |
| 9073 | |
| 9074 | if (flags & SV_GMAGIC) |
| 9075 | SvGETMAGIC(oldstr); /* before new_SV, in case it dies */ |
| 9076 | new_SV(sv); |
| 9077 | sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC); |
| 9078 | PUSH_EXTEND_MORTAL__SV_C(sv); |
| 9079 | SvTEMP_on(sv); |
| 9080 | return sv; |
| 9081 | } |
| 9082 | |
| 9083 | /* |
| 9084 | =for apidoc sv_newmortal |
| 9085 | |
| 9086 | Creates a new null SV which is mortal. The reference count of the SV is |
| 9087 | set to 1. It will be destroyed "soon", either by an explicit call to |
| 9088 | C<FREETMPS>, or by an implicit call at places such as statement boundaries. |
| 9089 | See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>. |
| 9090 | |
| 9091 | =cut |
| 9092 | */ |
| 9093 | |
| 9094 | SV * |
| 9095 | Perl_sv_newmortal(pTHX) |
| 9096 | { |
| 9097 | SV *sv; |
| 9098 | |
| 9099 | new_SV(sv); |
| 9100 | SvFLAGS(sv) = SVs_TEMP; |
| 9101 | PUSH_EXTEND_MORTAL__SV_C(sv); |
| 9102 | return sv; |
| 9103 | } |
| 9104 | |
| 9105 | |
| 9106 | /* |
| 9107 | =for apidoc newSVpvn_flags |
| 9108 | |
| 9109 | Creates a new SV and copies a string (which may contain C<NUL> (C<\0>) |
| 9110 | characters) into it. The reference count for the |
| 9111 | SV is set to 1. Note that if C<len> is zero, Perl will create a zero length |
| 9112 | string. You are responsible for ensuring that the source string is at least |
| 9113 | C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined. |
| 9114 | Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>. |
| 9115 | If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before |
| 9116 | returning. If C<SVf_UTF8> is set, C<s> |
| 9117 | is considered to be in UTF-8 and the |
| 9118 | C<SVf_UTF8> flag will be set on the new SV. |
| 9119 | C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as |
| 9120 | |
| 9121 | #define newSVpvn_utf8(s, len, u) \ |
| 9122 | newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0) |
| 9123 | |
| 9124 | =cut |
| 9125 | */ |
| 9126 | |
| 9127 | SV * |
| 9128 | Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags) |
| 9129 | { |
| 9130 | SV *sv; |
| 9131 | |
| 9132 | /* All the flags we don't support must be zero. |
| 9133 | And we're new code so I'm going to assert this from the start. */ |
| 9134 | assert(!(flags & ~(SVf_UTF8|SVs_TEMP))); |
| 9135 | new_SV(sv); |
| 9136 | sv_setpvn(sv,s,len); |
| 9137 | |
| 9138 | /* This code used to do a sv_2mortal(), however we now unroll the call to |
| 9139 | * sv_2mortal() and do what it does ourselves here. Since we have asserted |
| 9140 | * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we |
| 9141 | * can use it to enable the sv flags directly (bypassing SvTEMP_on), which |
| 9142 | * in turn means we dont need to mask out the SVf_UTF8 flag below, which |
| 9143 | * means that we eliminate quite a few steps than it looks - Yves |
| 9144 | * (explaining patch by gfx) */ |
| 9145 | |
| 9146 | SvFLAGS(sv) |= flags; |
| 9147 | |
| 9148 | if(flags & SVs_TEMP){ |
| 9149 | PUSH_EXTEND_MORTAL__SV_C(sv); |
| 9150 | } |
| 9151 | |
| 9152 | return sv; |
| 9153 | } |
| 9154 | |
| 9155 | /* |
| 9156 | =for apidoc sv_2mortal |
| 9157 | |
| 9158 | Marks an existing SV as mortal. The SV will be destroyed "soon", either |
| 9159 | by an explicit call to C<FREETMPS>, or by an implicit call at places such as |
| 9160 | statement boundaries. C<SvTEMP()> is turned on which means that the SV's |
| 9161 | string buffer can be "stolen" if this SV is copied. See also |
| 9162 | C<L</sv_newmortal>> and C<L</sv_mortalcopy>>. |
| 9163 | |
| 9164 | =cut |
| 9165 | */ |
| 9166 | |
| 9167 | SV * |
| 9168 | Perl_sv_2mortal(pTHX_ SV *const sv) |
| 9169 | { |
| 9170 | dVAR; |
| 9171 | if (!sv) |
| 9172 | return sv; |
| 9173 | if (SvIMMORTAL(sv)) |
| 9174 | return sv; |
| 9175 | PUSH_EXTEND_MORTAL__SV_C(sv); |
| 9176 | SvTEMP_on(sv); |
| 9177 | return sv; |
| 9178 | } |
| 9179 | |
| 9180 | /* |
| 9181 | =for apidoc newSVpv |
| 9182 | |
| 9183 | Creates a new SV and copies a string (which may contain C<NUL> (C<\0>) |
| 9184 | characters) into it. The reference count for the |
| 9185 | SV is set to 1. If C<len> is zero, Perl will compute the length using |
| 9186 | C<strlen()>, (which means if you use this option, that C<s> can't have embedded |
| 9187 | C<NUL> characters and has to have a terminating C<NUL> byte). |
| 9188 | |
| 9189 | For efficiency, consider using C<newSVpvn> instead. |
| 9190 | |
| 9191 | =cut |
| 9192 | */ |
| 9193 | |
| 9194 | SV * |
| 9195 | Perl_newSVpv(pTHX_ const char *const s, const STRLEN len) |
| 9196 | { |
| 9197 | SV *sv; |
| 9198 | |
| 9199 | new_SV(sv); |
| 9200 | sv_setpvn(sv, s, len || s == NULL ? len : strlen(s)); |
| 9201 | return sv; |
| 9202 | } |
| 9203 | |
| 9204 | /* |
| 9205 | =for apidoc newSVpvn |
| 9206 | |
| 9207 | Creates a new SV and copies a string into it, which may contain C<NUL> characters |
| 9208 | (C<\0>) and other binary data. The reference count for the SV is set to 1. |
| 9209 | Note that if C<len> is zero, Perl will create a zero length (Perl) string. You |
| 9210 | are responsible for ensuring that the source buffer is at least |
| 9211 | C<len> bytes long. If the C<buffer> argument is NULL the new SV will be |
| 9212 | undefined. |
| 9213 | |
| 9214 | =cut |
| 9215 | */ |
| 9216 | |
| 9217 | SV * |
| 9218 | Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len) |
| 9219 | { |
| 9220 | SV *sv; |
| 9221 | new_SV(sv); |
| 9222 | sv_setpvn(sv,buffer,len); |
| 9223 | return sv; |
| 9224 | } |
| 9225 | |
| 9226 | /* |
| 9227 | =for apidoc newSVhek |
| 9228 | |
| 9229 | Creates a new SV from the hash key structure. It will generate scalars that |
| 9230 | point to the shared string table where possible. Returns a new (undefined) |
| 9231 | SV if C<hek> is NULL. |
| 9232 | |
| 9233 | =cut |
| 9234 | */ |
| 9235 | |
| 9236 | SV * |
| 9237 | Perl_newSVhek(pTHX_ const HEK *const hek) |
| 9238 | { |
| 9239 | if (!hek) { |
| 9240 | SV *sv; |
| 9241 | |
| 9242 | new_SV(sv); |
| 9243 | return sv; |
| 9244 | } |
| 9245 | |
| 9246 | if (HEK_LEN(hek) == HEf_SVKEY) { |
| 9247 | return newSVsv(*(SV**)HEK_KEY(hek)); |
| 9248 | } else { |
| 9249 | const int flags = HEK_FLAGS(hek); |
| 9250 | if (flags & HVhek_WASUTF8) { |
| 9251 | /* Trouble :-) |
| 9252 | Andreas would like keys he put in as utf8 to come back as utf8 |
| 9253 | */ |
| 9254 | STRLEN utf8_len = HEK_LEN(hek); |
| 9255 | SV * const sv = newSV_type(SVt_PV); |
| 9256 | char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len); |
| 9257 | /* bytes_to_utf8() allocates a new string, which we can repurpose: */ |
| 9258 | sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL); |
| 9259 | SvUTF8_on (sv); |
| 9260 | return sv; |
| 9261 | } else if (flags & HVhek_UNSHARED) { |
| 9262 | /* A hash that isn't using shared hash keys has to have |
| 9263 | the flag in every key so that we know not to try to call |
| 9264 | share_hek_hek on it. */ |
| 9265 | |
| 9266 | SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek)); |
| 9267 | if (HEK_UTF8(hek)) |
| 9268 | SvUTF8_on (sv); |
| 9269 | return sv; |
| 9270 | } |
| 9271 | /* This will be overwhelminly the most common case. */ |
| 9272 | { |
| 9273 | /* Inline most of newSVpvn_share(), because share_hek_hek() is far |
| 9274 | more efficient than sharepvn(). */ |
| 9275 | SV *sv; |
| 9276 | |
| 9277 | new_SV(sv); |
| 9278 | sv_upgrade(sv, SVt_PV); |
| 9279 | SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek))); |
| 9280 | SvCUR_set(sv, HEK_LEN(hek)); |
| 9281 | SvLEN_set(sv, 0); |
| 9282 | SvIsCOW_on(sv); |
| 9283 | SvPOK_on(sv); |
| 9284 | if (HEK_UTF8(hek)) |
| 9285 | SvUTF8_on(sv); |
| 9286 | return sv; |
| 9287 | } |
| 9288 | } |
| 9289 | } |
| 9290 | |
| 9291 | /* |
| 9292 | =for apidoc newSVpvn_share |
| 9293 | |
| 9294 | Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string |
| 9295 | table. If the string does not already exist in the table, it is |
| 9296 | created first. Turns on the C<SvIsCOW> flag (or C<READONLY> |
| 9297 | and C<FAKE> in 5.16 and earlier). If the C<hash> parameter |
| 9298 | is non-zero, that value is used; otherwise the hash is computed. |
| 9299 | The string's hash can later be retrieved from the SV |
| 9300 | with the C<SvSHARED_HASH()> macro. The idea here is |
| 9301 | that as the string table is used for shared hash keys these strings will have |
| 9302 | C<SvPVX_const == HeKEY> and hash lookup will avoid string compare. |
| 9303 | |
| 9304 | =cut |
| 9305 | */ |
| 9306 | |
| 9307 | SV * |
| 9308 | Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash) |
| 9309 | { |
| 9310 | dVAR; |
| 9311 | SV *sv; |
| 9312 | bool is_utf8 = FALSE; |
| 9313 | const char *const orig_src = src; |
| 9314 | |
| 9315 | if (len < 0) { |
| 9316 | STRLEN tmplen = -len; |
| 9317 | is_utf8 = TRUE; |
| 9318 | /* See the note in hv.c:hv_fetch() --jhi */ |
| 9319 | src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8); |
| 9320 | len = tmplen; |
| 9321 | } |
| 9322 | if (!hash) |
| 9323 | PERL_HASH(hash, src, len); |
| 9324 | new_SV(sv); |
| 9325 | /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it |
| 9326 | changes here, update it there too. */ |
| 9327 | sv_upgrade(sv, SVt_PV); |
| 9328 | SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash)); |
| 9329 | SvCUR_set(sv, len); |
| 9330 | SvLEN_set(sv, 0); |
| 9331 | SvIsCOW_on(sv); |
| 9332 | SvPOK_on(sv); |
| 9333 | if (is_utf8) |
| 9334 | SvUTF8_on(sv); |
| 9335 | if (src != orig_src) |
| 9336 | Safefree(src); |
| 9337 | return sv; |
| 9338 | } |
| 9339 | |
| 9340 | /* |
| 9341 | =for apidoc newSVpv_share |
| 9342 | |
| 9343 | Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a |
| 9344 | string/length pair. |
| 9345 | |
| 9346 | =cut |
| 9347 | */ |
| 9348 | |
| 9349 | SV * |
| 9350 | Perl_newSVpv_share(pTHX_ const char *src, U32 hash) |
| 9351 | { |
| 9352 | return newSVpvn_share(src, strlen(src), hash); |
| 9353 | } |
| 9354 | |
| 9355 | #if defined(PERL_IMPLICIT_CONTEXT) |
| 9356 | |
| 9357 | /* pTHX_ magic can't cope with varargs, so this is a no-context |
| 9358 | * version of the main function, (which may itself be aliased to us). |
| 9359 | * Don't access this version directly. |
| 9360 | */ |
| 9361 | |
| 9362 | SV * |
| 9363 | Perl_newSVpvf_nocontext(const char *const pat, ...) |
| 9364 | { |
| 9365 | dTHX; |
| 9366 | SV *sv; |
| 9367 | va_list args; |
| 9368 | |
| 9369 | PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT; |
| 9370 | |
| 9371 | va_start(args, pat); |
| 9372 | sv = vnewSVpvf(pat, &args); |
| 9373 | va_end(args); |
| 9374 | return sv; |
| 9375 | } |
| 9376 | #endif |
| 9377 | |
| 9378 | /* |
| 9379 | =for apidoc newSVpvf |
| 9380 | |
| 9381 | Creates a new SV and initializes it with the string formatted like |
| 9382 | C<sv_catpvf>. |
| 9383 | |
| 9384 | =cut |
| 9385 | */ |
| 9386 | |
| 9387 | SV * |
| 9388 | Perl_newSVpvf(pTHX_ const char *const pat, ...) |
| 9389 | { |
| 9390 | SV *sv; |
| 9391 | va_list args; |
| 9392 | |
| 9393 | PERL_ARGS_ASSERT_NEWSVPVF; |
| 9394 | |
| 9395 | va_start(args, pat); |
| 9396 | sv = vnewSVpvf(pat, &args); |
| 9397 | va_end(args); |
| 9398 | return sv; |
| 9399 | } |
| 9400 | |
| 9401 | /* backend for newSVpvf() and newSVpvf_nocontext() */ |
| 9402 | |
| 9403 | SV * |
| 9404 | Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args) |
| 9405 | { |
| 9406 | SV *sv; |
| 9407 | |
| 9408 | PERL_ARGS_ASSERT_VNEWSVPVF; |
| 9409 | |
| 9410 | new_SV(sv); |
| 9411 | sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); |
| 9412 | return sv; |
| 9413 | } |
| 9414 | |
| 9415 | /* |
| 9416 | =for apidoc newSVnv |
| 9417 | |
| 9418 | Creates a new SV and copies a floating point value into it. |
| 9419 | The reference count for the SV is set to 1. |
| 9420 | |
| 9421 | =cut |
| 9422 | */ |
| 9423 | |
| 9424 | SV * |
| 9425 | Perl_newSVnv(pTHX_ const NV n) |
| 9426 | { |
| 9427 | SV *sv; |
| 9428 | |
| 9429 | new_SV(sv); |
| 9430 | sv_setnv(sv,n); |
| 9431 | return sv; |
| 9432 | } |
| 9433 | |
| 9434 | /* |
| 9435 | =for apidoc newSViv |
| 9436 | |
| 9437 | Creates a new SV and copies an integer into it. The reference count for the |
| 9438 | SV is set to 1. |
| 9439 | |
| 9440 | =cut |
| 9441 | */ |
| 9442 | |
| 9443 | SV * |
| 9444 | Perl_newSViv(pTHX_ const IV i) |
| 9445 | { |
| 9446 | SV *sv; |
| 9447 | |
| 9448 | new_SV(sv); |
| 9449 | |
| 9450 | /* Inlining ONLY the small relevant subset of sv_setiv here |
| 9451 | * for performance. Makes a significant difference. */ |
| 9452 | |
| 9453 | /* We're starting from SVt_FIRST, so provided that's |
| 9454 | * actual 0, we don't have to unset any SV type flags |
| 9455 | * to promote to SVt_IV. */ |
| 9456 | STATIC_ASSERT_STMT(SVt_FIRST == 0); |
| 9457 | |
| 9458 | SET_SVANY_FOR_BODYLESS_IV(sv); |
| 9459 | SvFLAGS(sv) |= SVt_IV; |
| 9460 | (void)SvIOK_on(sv); |
| 9461 | |
| 9462 | SvIV_set(sv, i); |
| 9463 | SvTAINT(sv); |
| 9464 | |
| 9465 | return sv; |
| 9466 | } |
| 9467 | |
| 9468 | /* |
| 9469 | =for apidoc newSVuv |
| 9470 | |
| 9471 | Creates a new SV and copies an unsigned integer into it. |
| 9472 | The reference count for the SV is set to 1. |
| 9473 | |
| 9474 | =cut |
| 9475 | */ |
| 9476 | |
| 9477 | SV * |
| 9478 | Perl_newSVuv(pTHX_ const UV u) |
| 9479 | { |
| 9480 | SV *sv; |
| 9481 | |
| 9482 | /* Inlining ONLY the small relevant subset of sv_setuv here |
| 9483 | * for performance. Makes a significant difference. */ |
| 9484 | |
| 9485 | /* Using ivs is more efficient than using uvs - see sv_setuv */ |
| 9486 | if (u <= (UV)IV_MAX) { |
| 9487 | return newSViv((IV)u); |
| 9488 | } |
| 9489 | |
| 9490 | new_SV(sv); |
| 9491 | |
| 9492 | /* We're starting from SVt_FIRST, so provided that's |
| 9493 | * actual 0, we don't have to unset any SV type flags |
| 9494 | * to promote to SVt_IV. */ |
| 9495 | STATIC_ASSERT_STMT(SVt_FIRST == 0); |
| 9496 | |
| 9497 | SET_SVANY_FOR_BODYLESS_IV(sv); |
| 9498 | SvFLAGS(sv) |= SVt_IV; |
| 9499 | (void)SvIOK_on(sv); |
| 9500 | (void)SvIsUV_on(sv); |
| 9501 | |
| 9502 | SvUV_set(sv, u); |
| 9503 | SvTAINT(sv); |
| 9504 | |
| 9505 | return sv; |
| 9506 | } |
| 9507 | |
| 9508 | /* |
| 9509 | =for apidoc newSV_type |
| 9510 | |
| 9511 | Creates a new SV, of the type specified. The reference count for the new SV |
| 9512 | is set to 1. |
| 9513 | |
| 9514 | =cut |
| 9515 | */ |
| 9516 | |
| 9517 | SV * |
| 9518 | Perl_newSV_type(pTHX_ const svtype type) |
| 9519 | { |
| 9520 | SV *sv; |
| 9521 | |
| 9522 | new_SV(sv); |
| 9523 | ASSUME(SvTYPE(sv) == SVt_FIRST); |
| 9524 | if(type != SVt_FIRST) |
| 9525 | sv_upgrade(sv, type); |
| 9526 | return sv; |
| 9527 | } |
| 9528 | |
| 9529 | /* |
| 9530 | =for apidoc newRV_noinc |
| 9531 | |
| 9532 | Creates an RV wrapper for an SV. The reference count for the original |
| 9533 | SV is B<not> incremented. |
| 9534 | |
| 9535 | =cut |
| 9536 | */ |
| 9537 | |
| 9538 | SV * |
| 9539 | Perl_newRV_noinc(pTHX_ SV *const tmpRef) |
| 9540 | { |
| 9541 | SV *sv; |
| 9542 | |
| 9543 | PERL_ARGS_ASSERT_NEWRV_NOINC; |
| 9544 | |
| 9545 | new_SV(sv); |
| 9546 | |
| 9547 | /* We're starting from SVt_FIRST, so provided that's |
| 9548 | * actual 0, we don't have to unset any SV type flags |
| 9549 | * to promote to SVt_IV. */ |
| 9550 | STATIC_ASSERT_STMT(SVt_FIRST == 0); |
| 9551 | |
| 9552 | SET_SVANY_FOR_BODYLESS_IV(sv); |
| 9553 | SvFLAGS(sv) |= SVt_IV; |
| 9554 | SvROK_on(sv); |
| 9555 | SvIV_set(sv, 0); |
| 9556 | |
| 9557 | SvTEMP_off(tmpRef); |
| 9558 | SvRV_set(sv, tmpRef); |
| 9559 | |
| 9560 | return sv; |
| 9561 | } |
| 9562 | |
| 9563 | /* newRV_inc is the official function name to use now. |
| 9564 | * newRV_inc is in fact #defined to newRV in sv.h |
| 9565 | */ |
| 9566 | |
| 9567 | SV * |
| 9568 | Perl_newRV(pTHX_ SV *const sv) |
| 9569 | { |
| 9570 | PERL_ARGS_ASSERT_NEWRV; |
| 9571 | |
| 9572 | return newRV_noinc(SvREFCNT_inc_simple_NN(sv)); |
| 9573 | } |
| 9574 | |
| 9575 | /* |
| 9576 | =for apidoc newSVsv |
| 9577 | |
| 9578 | Creates a new SV which is an exact duplicate of the original SV. |
| 9579 | (Uses C<sv_setsv>.) |
| 9580 | |
| 9581 | =cut |
| 9582 | */ |
| 9583 | |
| 9584 | SV * |
| 9585 | Perl_newSVsv(pTHX_ SV *const old) |
| 9586 | { |
| 9587 | SV *sv; |
| 9588 | |
| 9589 | if (!old) |
| 9590 | return NULL; |
| 9591 | if (SvTYPE(old) == (svtype)SVTYPEMASK) { |
| 9592 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string"); |
| 9593 | return NULL; |
| 9594 | } |
| 9595 | /* Do this here, otherwise we leak the new SV if this croaks. */ |
| 9596 | SvGETMAGIC(old); |
| 9597 | new_SV(sv); |
| 9598 | /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games |
| 9599 | with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */ |
| 9600 | sv_setsv_flags(sv, old, SV_NOSTEAL); |
| 9601 | return sv; |
| 9602 | } |
| 9603 | |
| 9604 | /* |
| 9605 | =for apidoc sv_reset |
| 9606 | |
| 9607 | Underlying implementation for the C<reset> Perl function. |
| 9608 | Note that the perl-level function is vaguely deprecated. |
| 9609 | |
| 9610 | =cut |
| 9611 | */ |
| 9612 | |
| 9613 | void |
| 9614 | Perl_sv_reset(pTHX_ const char *s, HV *const stash) |
| 9615 | { |
| 9616 | PERL_ARGS_ASSERT_SV_RESET; |
| 9617 | |
| 9618 | sv_resetpvn(*s ? s : NULL, strlen(s), stash); |
| 9619 | } |
| 9620 | |
| 9621 | void |
| 9622 | Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash) |
| 9623 | { |
| 9624 | char todo[PERL_UCHAR_MAX+1]; |
| 9625 | const char *send; |
| 9626 | |
| 9627 | if (!stash || SvTYPE(stash) != SVt_PVHV) |
| 9628 | return; |
| 9629 | |
| 9630 | if (!s) { /* reset ?? searches */ |
| 9631 | MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab); |
| 9632 | if (mg) { |
| 9633 | const U32 count = mg->mg_len / sizeof(PMOP**); |
| 9634 | PMOP **pmp = (PMOP**) mg->mg_ptr; |
| 9635 | PMOP *const *const end = pmp + count; |
| 9636 | |
| 9637 | while (pmp < end) { |
| 9638 | #ifdef USE_ITHREADS |
| 9639 | SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]); |
| 9640 | #else |
| 9641 | (*pmp)->op_pmflags &= ~PMf_USED; |
| 9642 | #endif |
| 9643 | ++pmp; |
| 9644 | } |
| 9645 | } |
| 9646 | return; |
| 9647 | } |
| 9648 | |
| 9649 | /* reset variables */ |
| 9650 | |
| 9651 | if (!HvARRAY(stash)) |
| 9652 | return; |
| 9653 | |
| 9654 | Zero(todo, 256, char); |
| 9655 | send = s + len; |
| 9656 | while (s < send) { |
| 9657 | I32 max; |
| 9658 | I32 i = (unsigned char)*s; |
| 9659 | if (s[1] == '-') { |
| 9660 | s += 2; |
| 9661 | } |
| 9662 | max = (unsigned char)*s++; |
| 9663 | for ( ; i <= max; i++) { |
| 9664 | todo[i] = 1; |
| 9665 | } |
| 9666 | for (i = 0; i <= (I32) HvMAX(stash); i++) { |
| 9667 | HE *entry; |
| 9668 | for (entry = HvARRAY(stash)[i]; |
| 9669 | entry; |
| 9670 | entry = HeNEXT(entry)) |
| 9671 | { |
| 9672 | GV *gv; |
| 9673 | SV *sv; |
| 9674 | |
| 9675 | if (!todo[(U8)*HeKEY(entry)]) |
| 9676 | continue; |
| 9677 | gv = MUTABLE_GV(HeVAL(entry)); |
| 9678 | sv = GvSV(gv); |
| 9679 | if (sv && !SvREADONLY(sv)) { |
| 9680 | SV_CHECK_THINKFIRST_COW_DROP(sv); |
| 9681 | if (!isGV(sv)) SvOK_off(sv); |
| 9682 | } |
| 9683 | if (GvAV(gv)) { |
| 9684 | av_clear(GvAV(gv)); |
| 9685 | } |
| 9686 | if (GvHV(gv) && !HvNAME_get(GvHV(gv))) { |
| 9687 | hv_clear(GvHV(gv)); |
| 9688 | } |
| 9689 | } |
| 9690 | } |
| 9691 | } |
| 9692 | } |
| 9693 | |
| 9694 | /* |
| 9695 | =for apidoc sv_2io |
| 9696 | |
| 9697 | Using various gambits, try to get an IO from an SV: the IO slot if its a |
| 9698 | GV; or the recursive result if we're an RV; or the IO slot of the symbol |
| 9699 | named after the PV if we're a string. |
| 9700 | |
| 9701 | 'Get' magic is ignored on the C<sv> passed in, but will be called on |
| 9702 | C<SvRV(sv)> if C<sv> is an RV. |
| 9703 | |
| 9704 | =cut |
| 9705 | */ |
| 9706 | |
| 9707 | IO* |
| 9708 | Perl_sv_2io(pTHX_ SV *const sv) |
| 9709 | { |
| 9710 | IO* io; |
| 9711 | GV* gv; |
| 9712 | |
| 9713 | PERL_ARGS_ASSERT_SV_2IO; |
| 9714 | |
| 9715 | switch (SvTYPE(sv)) { |
| 9716 | case SVt_PVIO: |
| 9717 | io = MUTABLE_IO(sv); |
| 9718 | break; |
| 9719 | case SVt_PVGV: |
| 9720 | case SVt_PVLV: |
| 9721 | if (isGV_with_GP(sv)) { |
| 9722 | gv = MUTABLE_GV(sv); |
| 9723 | io = GvIO(gv); |
| 9724 | if (!io) |
| 9725 | Perl_croak(aTHX_ "Bad filehandle: %"HEKf, |
| 9726 | HEKfARG(GvNAME_HEK(gv))); |
| 9727 | break; |
| 9728 | } |
| 9729 | /* FALLTHROUGH */ |
| 9730 | default: |
| 9731 | if (!SvOK(sv)) |
| 9732 | Perl_croak(aTHX_ PL_no_usym, "filehandle"); |
| 9733 | if (SvROK(sv)) { |
| 9734 | SvGETMAGIC(SvRV(sv)); |
| 9735 | return sv_2io(SvRV(sv)); |
| 9736 | } |
| 9737 | gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO); |
| 9738 | if (gv) |
| 9739 | io = GvIO(gv); |
| 9740 | else |
| 9741 | io = 0; |
| 9742 | if (!io) { |
| 9743 | SV *newsv = sv; |
| 9744 | if (SvGMAGICAL(sv)) { |
| 9745 | newsv = sv_newmortal(); |
| 9746 | sv_setsv_nomg(newsv, sv); |
| 9747 | } |
| 9748 | Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv)); |
| 9749 | } |
| 9750 | break; |
| 9751 | } |
| 9752 | return io; |
| 9753 | } |
| 9754 | |
| 9755 | /* |
| 9756 | =for apidoc sv_2cv |
| 9757 | |
| 9758 | Using various gambits, try to get a CV from an SV; in addition, try if |
| 9759 | possible to set C<*st> and C<*gvp> to the stash and GV associated with it. |
| 9760 | The flags in C<lref> are passed to C<gv_fetchsv>. |
| 9761 | |
| 9762 | =cut |
| 9763 | */ |
| 9764 | |
| 9765 | CV * |
| 9766 | Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref) |
| 9767 | { |
| 9768 | GV *gv = NULL; |
| 9769 | CV *cv = NULL; |
| 9770 | |
| 9771 | PERL_ARGS_ASSERT_SV_2CV; |
| 9772 | |
| 9773 | if (!sv) { |
| 9774 | *st = NULL; |
| 9775 | *gvp = NULL; |
| 9776 | return NULL; |
| 9777 | } |
| 9778 | switch (SvTYPE(sv)) { |
| 9779 | case SVt_PVCV: |
| 9780 | *st = CvSTASH(sv); |
| 9781 | *gvp = NULL; |
| 9782 | return MUTABLE_CV(sv); |
| 9783 | case SVt_PVHV: |
| 9784 | case SVt_PVAV: |
| 9785 | *st = NULL; |
| 9786 | *gvp = NULL; |
| 9787 | return NULL; |
| 9788 | default: |
| 9789 | SvGETMAGIC(sv); |
| 9790 | if (SvROK(sv)) { |
| 9791 | if (SvAMAGIC(sv)) |
| 9792 | sv = amagic_deref_call(sv, to_cv_amg); |
| 9793 | |
| 9794 | sv = SvRV(sv); |
| 9795 | if (SvTYPE(sv) == SVt_PVCV) { |
| 9796 | cv = MUTABLE_CV(sv); |
| 9797 | *gvp = NULL; |
| 9798 | *st = CvSTASH(cv); |
| 9799 | return cv; |
| 9800 | } |
| 9801 | else if(SvGETMAGIC(sv), isGV_with_GP(sv)) |
| 9802 | gv = MUTABLE_GV(sv); |
| 9803 | else |
| 9804 | Perl_croak(aTHX_ "Not a subroutine reference"); |
| 9805 | } |
| 9806 | else if (isGV_with_GP(sv)) { |
| 9807 | gv = MUTABLE_GV(sv); |
| 9808 | } |
| 9809 | else { |
| 9810 | gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV); |
| 9811 | } |
| 9812 | *gvp = gv; |
| 9813 | if (!gv) { |
| 9814 | *st = NULL; |
| 9815 | return NULL; |
| 9816 | } |
| 9817 | /* Some flags to gv_fetchsv mean don't really create the GV */ |
| 9818 | if (!isGV_with_GP(gv)) { |
| 9819 | *st = NULL; |
| 9820 | return NULL; |
| 9821 | } |
| 9822 | *st = GvESTASH(gv); |
| 9823 | if (lref & ~GV_ADDMG && !GvCVu(gv)) { |
| 9824 | /* XXX this is probably not what they think they're getting. |
| 9825 | * It has the same effect as "sub name;", i.e. just a forward |
| 9826 | * declaration! */ |
| 9827 | newSTUB(gv,0); |
| 9828 | } |
| 9829 | return GvCVu(gv); |
| 9830 | } |
| 9831 | } |
| 9832 | |
| 9833 | /* |
| 9834 | =for apidoc sv_true |
| 9835 | |
| 9836 | Returns true if the SV has a true value by Perl's rules. |
| 9837 | Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may |
| 9838 | instead use an in-line version. |
| 9839 | |
| 9840 | =cut |
| 9841 | */ |
| 9842 | |
| 9843 | I32 |
| 9844 | Perl_sv_true(pTHX_ SV *const sv) |
| 9845 | { |
| 9846 | if (!sv) |
| 9847 | return 0; |
| 9848 | if (SvPOK(sv)) { |
| 9849 | const XPV* const tXpv = (XPV*)SvANY(sv); |
| 9850 | if (tXpv && |
| 9851 | (tXpv->xpv_cur > 1 || |
| 9852 | (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0'))) |
| 9853 | return 1; |
| 9854 | else |
| 9855 | return 0; |
| 9856 | } |
| 9857 | else { |
| 9858 | if (SvIOK(sv)) |
| 9859 | return SvIVX(sv) != 0; |
| 9860 | else { |
| 9861 | if (SvNOK(sv)) |
| 9862 | return SvNVX(sv) != 0.0; |
| 9863 | else |
| 9864 | return sv_2bool(sv); |
| 9865 | } |
| 9866 | } |
| 9867 | } |
| 9868 | |
| 9869 | /* |
| 9870 | =for apidoc sv_pvn_force |
| 9871 | |
| 9872 | Get a sensible string out of the SV somehow. |
| 9873 | A private implementation of the C<SvPV_force> macro for compilers which |
| 9874 | can't cope with complex macro expressions. Always use the macro instead. |
| 9875 | |
| 9876 | =for apidoc sv_pvn_force_flags |
| 9877 | |
| 9878 | Get a sensible string out of the SV somehow. |
| 9879 | If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if |
| 9880 | appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are |
| 9881 | implemented in terms of this function. |
| 9882 | You normally want to use the various wrapper macros instead: see |
| 9883 | C<L</SvPV_force>> and C<L</SvPV_force_nomg>>. |
| 9884 | |
| 9885 | =cut |
| 9886 | */ |
| 9887 | |
| 9888 | char * |
| 9889 | Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags) |
| 9890 | { |
| 9891 | PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS; |
| 9892 | |
| 9893 | if (flags & SV_GMAGIC) SvGETMAGIC(sv); |
| 9894 | if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv))) |
| 9895 | sv_force_normal_flags(sv, 0); |
| 9896 | |
| 9897 | if (SvPOK(sv)) { |
| 9898 | if (lp) |
| 9899 | *lp = SvCUR(sv); |
| 9900 | } |
| 9901 | else { |
| 9902 | char *s; |
| 9903 | STRLEN len; |
| 9904 | |
| 9905 | if (SvTYPE(sv) > SVt_PVLV |
| 9906 | || isGV_with_GP(sv)) |
| 9907 | /* diag_listed_as: Can't coerce %s to %s in %s */ |
| 9908 | Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0), |
| 9909 | OP_DESC(PL_op)); |
| 9910 | s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC); |
| 9911 | if (!s) { |
| 9912 | s = (char *)""; |
| 9913 | } |
| 9914 | if (lp) |
| 9915 | *lp = len; |
| 9916 | |
| 9917 | if (SvTYPE(sv) < SVt_PV || |
| 9918 | s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */ |
| 9919 | if (SvROK(sv)) |
| 9920 | sv_unref(sv); |
| 9921 | SvUPGRADE(sv, SVt_PV); /* Never FALSE */ |
| 9922 | SvGROW(sv, len + 1); |
| 9923 | Move(s,SvPVX(sv),len,char); |
| 9924 | SvCUR_set(sv, len); |
| 9925 | SvPVX(sv)[len] = '\0'; |
| 9926 | } |
| 9927 | if (!SvPOK(sv)) { |
| 9928 | SvPOK_on(sv); /* validate pointer */ |
| 9929 | SvTAINT(sv); |
| 9930 | DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n", |
| 9931 | PTR2UV(sv),SvPVX_const(sv))); |
| 9932 | } |
| 9933 | } |
| 9934 | (void)SvPOK_only_UTF8(sv); |
| 9935 | return SvPVX_mutable(sv); |
| 9936 | } |
| 9937 | |
| 9938 | /* |
| 9939 | =for apidoc sv_pvbyten_force |
| 9940 | |
| 9941 | The backend for the C<SvPVbytex_force> macro. Always use the macro |
| 9942 | instead. |
| 9943 | |
| 9944 | =cut |
| 9945 | */ |
| 9946 | |
| 9947 | char * |
| 9948 | Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp) |
| 9949 | { |
| 9950 | PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE; |
| 9951 | |
| 9952 | sv_pvn_force(sv,lp); |
| 9953 | sv_utf8_downgrade(sv,0); |
| 9954 | *lp = SvCUR(sv); |
| 9955 | return SvPVX(sv); |
| 9956 | } |
| 9957 | |
| 9958 | /* |
| 9959 | =for apidoc sv_pvutf8n_force |
| 9960 | |
| 9961 | The backend for the C<SvPVutf8x_force> macro. Always use the macro |
| 9962 | instead. |
| 9963 | |
| 9964 | =cut |
| 9965 | */ |
| 9966 | |
| 9967 | char * |
| 9968 | Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp) |
| 9969 | { |
| 9970 | PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE; |
| 9971 | |
| 9972 | sv_pvn_force(sv,0); |
| 9973 | sv_utf8_upgrade_nomg(sv); |
| 9974 | *lp = SvCUR(sv); |
| 9975 | return SvPVX(sv); |
| 9976 | } |
| 9977 | |
| 9978 | /* |
| 9979 | =for apidoc sv_reftype |
| 9980 | |
| 9981 | Returns a string describing what the SV is a reference to. |
| 9982 | |
| 9983 | If ob is true and the SV is blessed, the string is the class name, |
| 9984 | otherwise it is the type of the SV, "SCALAR", "ARRAY" etc. |
| 9985 | |
| 9986 | =cut |
| 9987 | */ |
| 9988 | |
| 9989 | const char * |
| 9990 | Perl_sv_reftype(pTHX_ const SV *const sv, const int ob) |
| 9991 | { |
| 9992 | PERL_ARGS_ASSERT_SV_REFTYPE; |
| 9993 | if (ob && SvOBJECT(sv)) { |
| 9994 | return SvPV_nolen_const(sv_ref(NULL, sv, ob)); |
| 9995 | } |
| 9996 | else { |
| 9997 | /* WARNING - There is code, for instance in mg.c, that assumes that |
| 9998 | * the only reason that sv_reftype(sv,0) would return a string starting |
| 9999 | * with 'L' or 'S' is that it is a LVALUE or a SCALAR. |
| 10000 | * Yes this a dodgy way to do type checking, but it saves practically reimplementing |
| 10001 | * this routine inside other subs, and it saves time. |
| 10002 | * Do not change this assumption without searching for "dodgy type check" in |
| 10003 | * the code. |
| 10004 | * - Yves */ |
| 10005 | switch (SvTYPE(sv)) { |
| 10006 | case SVt_NULL: |
| 10007 | case SVt_IV: |
| 10008 | case SVt_NV: |
| 10009 | case SVt_PV: |
| 10010 | case SVt_PVIV: |
| 10011 | case SVt_PVNV: |
| 10012 | case SVt_PVMG: |
| 10013 | if (SvVOK(sv)) |
| 10014 | return "VSTRING"; |
| 10015 | if (SvROK(sv)) |
| 10016 | return "REF"; |
| 10017 | else |
| 10018 | return "SCALAR"; |
| 10019 | |
| 10020 | case SVt_PVLV: return (char *) (SvROK(sv) ? "REF" |
| 10021 | /* tied lvalues should appear to be |
| 10022 | * scalars for backwards compatibility */ |
| 10023 | : (isALPHA_FOLD_EQ(LvTYPE(sv), 't')) |
| 10024 | ? "SCALAR" : "LVALUE"); |
| 10025 | case SVt_PVAV: return "ARRAY"; |
| 10026 | case SVt_PVHV: return "HASH"; |
| 10027 | case SVt_PVCV: return "CODE"; |
| 10028 | case SVt_PVGV: return (char *) (isGV_with_GP(sv) |
| 10029 | ? "GLOB" : "SCALAR"); |
| 10030 | case SVt_PVFM: return "FORMAT"; |
| 10031 | case SVt_PVIO: return "IO"; |
| 10032 | case SVt_INVLIST: return "INVLIST"; |
| 10033 | case SVt_REGEXP: return "REGEXP"; |
| 10034 | default: return "UNKNOWN"; |
| 10035 | } |
| 10036 | } |
| 10037 | } |
| 10038 | |
| 10039 | /* |
| 10040 | =for apidoc sv_ref |
| 10041 | |
| 10042 | Returns a SV describing what the SV passed in is a reference to. |
| 10043 | |
| 10044 | dst can be a SV to be set to the description or NULL, in which case a |
| 10045 | mortal SV is returned. |
| 10046 | |
| 10047 | If ob is true and the SV is blessed, the description is the class |
| 10048 | name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc. |
| 10049 | |
| 10050 | =cut |
| 10051 | */ |
| 10052 | |
| 10053 | SV * |
| 10054 | Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob) |
| 10055 | { |
| 10056 | PERL_ARGS_ASSERT_SV_REF; |
| 10057 | |
| 10058 | if (!dst) |
| 10059 | dst = sv_newmortal(); |
| 10060 | |
| 10061 | if (ob && SvOBJECT(sv)) { |
| 10062 | HvNAME_get(SvSTASH(sv)) |
| 10063 | ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv))) |
| 10064 | : sv_setpvn(dst, "__ANON__", 8); |
| 10065 | } |
| 10066 | else { |
| 10067 | const char * reftype = sv_reftype(sv, 0); |
| 10068 | sv_setpv(dst, reftype); |
| 10069 | } |
| 10070 | return dst; |
| 10071 | } |
| 10072 | |
| 10073 | /* |
| 10074 | =for apidoc sv_isobject |
| 10075 | |
| 10076 | Returns a boolean indicating whether the SV is an RV pointing to a blessed |
| 10077 | object. If the SV is not an RV, or if the object is not blessed, then this |
| 10078 | will return false. |
| 10079 | |
| 10080 | =cut |
| 10081 | */ |
| 10082 | |
| 10083 | int |
| 10084 | Perl_sv_isobject(pTHX_ SV *sv) |
| 10085 | { |
| 10086 | if (!sv) |
| 10087 | return 0; |
| 10088 | SvGETMAGIC(sv); |
| 10089 | if (!SvROK(sv)) |
| 10090 | return 0; |
| 10091 | sv = SvRV(sv); |
| 10092 | if (!SvOBJECT(sv)) |
| 10093 | return 0; |
| 10094 | return 1; |
| 10095 | } |
| 10096 | |
| 10097 | /* |
| 10098 | =for apidoc sv_isa |
| 10099 | |
| 10100 | Returns a boolean indicating whether the SV is blessed into the specified |
| 10101 | class. This does not check for subtypes; use C<sv_derived_from> to verify |
| 10102 | an inheritance relationship. |
| 10103 | |
| 10104 | =cut |
| 10105 | */ |
| 10106 | |
| 10107 | int |
| 10108 | Perl_sv_isa(pTHX_ SV *sv, const char *const name) |
| 10109 | { |
| 10110 | const char *hvname; |
| 10111 | |
| 10112 | PERL_ARGS_ASSERT_SV_ISA; |
| 10113 | |
| 10114 | if (!sv) |
| 10115 | return 0; |
| 10116 | SvGETMAGIC(sv); |
| 10117 | if (!SvROK(sv)) |
| 10118 | return 0; |
| 10119 | sv = SvRV(sv); |
| 10120 | if (!SvOBJECT(sv)) |
| 10121 | return 0; |
| 10122 | hvname = HvNAME_get(SvSTASH(sv)); |
| 10123 | if (!hvname) |
| 10124 | return 0; |
| 10125 | |
| 10126 | return strEQ(hvname, name); |
| 10127 | } |
| 10128 | |
| 10129 | /* |
| 10130 | =for apidoc newSVrv |
| 10131 | |
| 10132 | Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an |
| 10133 | RV then it will be upgraded to one. If C<classname> is non-null then the new |
| 10134 | SV will be blessed in the specified package. The new SV is returned and its |
| 10135 | reference count is 1. The reference count 1 is owned by C<rv>. |
| 10136 | |
| 10137 | =cut |
| 10138 | */ |
| 10139 | |
| 10140 | SV* |
| 10141 | Perl_newSVrv(pTHX_ SV *const rv, const char *const classname) |
| 10142 | { |
| 10143 | SV *sv; |
| 10144 | |
| 10145 | PERL_ARGS_ASSERT_NEWSVRV; |
| 10146 | |
| 10147 | new_SV(sv); |
| 10148 | |
| 10149 | SV_CHECK_THINKFIRST_COW_DROP(rv); |
| 10150 | |
| 10151 | if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) { |
| 10152 | const U32 refcnt = SvREFCNT(rv); |
| 10153 | SvREFCNT(rv) = 0; |
| 10154 | sv_clear(rv); |
| 10155 | SvFLAGS(rv) = 0; |
| 10156 | SvREFCNT(rv) = refcnt; |
| 10157 | |
| 10158 | sv_upgrade(rv, SVt_IV); |
| 10159 | } else if (SvROK(rv)) { |
| 10160 | SvREFCNT_dec(SvRV(rv)); |
| 10161 | } else { |
| 10162 | prepare_SV_for_RV(rv); |
| 10163 | } |
| 10164 | |
| 10165 | SvOK_off(rv); |
| 10166 | SvRV_set(rv, sv); |
| 10167 | SvROK_on(rv); |
| 10168 | |
| 10169 | if (classname) { |
| 10170 | HV* const stash = gv_stashpv(classname, GV_ADD); |
| 10171 | (void)sv_bless(rv, stash); |
| 10172 | } |
| 10173 | return sv; |
| 10174 | } |
| 10175 | |
| 10176 | SV * |
| 10177 | Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible) |
| 10178 | { |
| 10179 | SV * const lv = newSV_type(SVt_PVLV); |
| 10180 | PERL_ARGS_ASSERT_NEWSVAVDEFELEM; |
| 10181 | LvTYPE(lv) = 'y'; |
| 10182 | sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0); |
| 10183 | LvTARG(lv) = SvREFCNT_inc_simple_NN(av); |
| 10184 | LvSTARGOFF(lv) = ix; |
| 10185 | LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX; |
| 10186 | return lv; |
| 10187 | } |
| 10188 | |
| 10189 | /* |
| 10190 | =for apidoc sv_setref_pv |
| 10191 | |
| 10192 | Copies a pointer into a new SV, optionally blessing the SV. The C<rv> |
| 10193 | argument will be upgraded to an RV. That RV will be modified to point to |
| 10194 | the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed |
| 10195 | into the SV. The C<classname> argument indicates the package for the |
| 10196 | blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV |
| 10197 | will have a reference count of 1, and the RV will be returned. |
| 10198 | |
| 10199 | Do not use with other Perl types such as HV, AV, SV, CV, because those |
| 10200 | objects will become corrupted by the pointer copy process. |
| 10201 | |
| 10202 | Note that C<sv_setref_pvn> copies the string while this copies the pointer. |
| 10203 | |
| 10204 | =cut |
| 10205 | */ |
| 10206 | |
| 10207 | SV* |
| 10208 | Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv) |
| 10209 | { |
| 10210 | PERL_ARGS_ASSERT_SV_SETREF_PV; |
| 10211 | |
| 10212 | if (!pv) { |
| 10213 | sv_setsv(rv, &PL_sv_undef); |
| 10214 | SvSETMAGIC(rv); |
| 10215 | } |
| 10216 | else |
| 10217 | sv_setiv(newSVrv(rv,classname), PTR2IV(pv)); |
| 10218 | return rv; |
| 10219 | } |
| 10220 | |
| 10221 | /* |
| 10222 | =for apidoc sv_setref_iv |
| 10223 | |
| 10224 | Copies an integer into a new SV, optionally blessing the SV. The C<rv> |
| 10225 | argument will be upgraded to an RV. That RV will be modified to point to |
| 10226 | the new SV. The C<classname> argument indicates the package for the |
| 10227 | blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV |
| 10228 | will have a reference count of 1, and the RV will be returned. |
| 10229 | |
| 10230 | =cut |
| 10231 | */ |
| 10232 | |
| 10233 | SV* |
| 10234 | Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv) |
| 10235 | { |
| 10236 | PERL_ARGS_ASSERT_SV_SETREF_IV; |
| 10237 | |
| 10238 | sv_setiv(newSVrv(rv,classname), iv); |
| 10239 | return rv; |
| 10240 | } |
| 10241 | |
| 10242 | /* |
| 10243 | =for apidoc sv_setref_uv |
| 10244 | |
| 10245 | Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv> |
| 10246 | argument will be upgraded to an RV. That RV will be modified to point to |
| 10247 | the new SV. The C<classname> argument indicates the package for the |
| 10248 | blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV |
| 10249 | will have a reference count of 1, and the RV will be returned. |
| 10250 | |
| 10251 | =cut |
| 10252 | */ |
| 10253 | |
| 10254 | SV* |
| 10255 | Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv) |
| 10256 | { |
| 10257 | PERL_ARGS_ASSERT_SV_SETREF_UV; |
| 10258 | |
| 10259 | sv_setuv(newSVrv(rv,classname), uv); |
| 10260 | return rv; |
| 10261 | } |
| 10262 | |
| 10263 | /* |
| 10264 | =for apidoc sv_setref_nv |
| 10265 | |
| 10266 | Copies a double into a new SV, optionally blessing the SV. The C<rv> |
| 10267 | argument will be upgraded to an RV. That RV will be modified to point to |
| 10268 | the new SV. The C<classname> argument indicates the package for the |
| 10269 | blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV |
| 10270 | will have a reference count of 1, and the RV will be returned. |
| 10271 | |
| 10272 | =cut |
| 10273 | */ |
| 10274 | |
| 10275 | SV* |
| 10276 | Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv) |
| 10277 | { |
| 10278 | PERL_ARGS_ASSERT_SV_SETREF_NV; |
| 10279 | |
| 10280 | sv_setnv(newSVrv(rv,classname), nv); |
| 10281 | return rv; |
| 10282 | } |
| 10283 | |
| 10284 | /* |
| 10285 | =for apidoc sv_setref_pvn |
| 10286 | |
| 10287 | Copies a string into a new SV, optionally blessing the SV. The length of the |
| 10288 | string must be specified with C<n>. The C<rv> argument will be upgraded to |
| 10289 | an RV. That RV will be modified to point to the new SV. The C<classname> |
| 10290 | argument indicates the package for the blessing. Set C<classname> to |
| 10291 | C<NULL> to avoid the blessing. The new SV will have a reference count |
| 10292 | of 1, and the RV will be returned. |
| 10293 | |
| 10294 | Note that C<sv_setref_pv> copies the pointer while this copies the string. |
| 10295 | |
| 10296 | =cut |
| 10297 | */ |
| 10298 | |
| 10299 | SV* |
| 10300 | Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname, |
| 10301 | const char *const pv, const STRLEN n) |
| 10302 | { |
| 10303 | PERL_ARGS_ASSERT_SV_SETREF_PVN; |
| 10304 | |
| 10305 | sv_setpvn(newSVrv(rv,classname), pv, n); |
| 10306 | return rv; |
| 10307 | } |
| 10308 | |
| 10309 | /* |
| 10310 | =for apidoc sv_bless |
| 10311 | |
| 10312 | Blesses an SV into a specified package. The SV must be an RV. The package |
| 10313 | must be designated by its stash (see C<L</gv_stashpv>>). The reference count |
| 10314 | of the SV is unaffected. |
| 10315 | |
| 10316 | =cut |
| 10317 | */ |
| 10318 | |
| 10319 | SV* |
| 10320 | Perl_sv_bless(pTHX_ SV *const sv, HV *const stash) |
| 10321 | { |
| 10322 | SV *tmpRef; |
| 10323 | HV *oldstash = NULL; |
| 10324 | |
| 10325 | PERL_ARGS_ASSERT_SV_BLESS; |
| 10326 | |
| 10327 | SvGETMAGIC(sv); |
| 10328 | if (!SvROK(sv)) |
| 10329 | Perl_croak(aTHX_ "Can't bless non-reference value"); |
| 10330 | tmpRef = SvRV(sv); |
| 10331 | if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) { |
| 10332 | if (SvREADONLY(tmpRef)) |
| 10333 | Perl_croak_no_modify(); |
| 10334 | if (SvOBJECT(tmpRef)) { |
| 10335 | oldstash = SvSTASH(tmpRef); |
| 10336 | } |
| 10337 | } |
| 10338 | SvOBJECT_on(tmpRef); |
| 10339 | SvUPGRADE(tmpRef, SVt_PVMG); |
| 10340 | SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash))); |
| 10341 | SvREFCNT_dec(oldstash); |
| 10342 | |
| 10343 | if(SvSMAGICAL(tmpRef)) |
| 10344 | if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar)) |
| 10345 | mg_set(tmpRef); |
| 10346 | |
| 10347 | |
| 10348 | |
| 10349 | return sv; |
| 10350 | } |
| 10351 | |
| 10352 | /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type |
| 10353 | * as it is after unglobbing it. |
| 10354 | */ |
| 10355 | |
| 10356 | PERL_STATIC_INLINE void |
| 10357 | S_sv_unglob(pTHX_ SV *const sv, U32 flags) |
| 10358 | { |
| 10359 | void *xpvmg; |
| 10360 | HV *stash; |
| 10361 | SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal(); |
| 10362 | |
| 10363 | PERL_ARGS_ASSERT_SV_UNGLOB; |
| 10364 | |
| 10365 | assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV); |
| 10366 | SvFAKE_off(sv); |
| 10367 | if (!(flags & SV_COW_DROP_PV)) |
| 10368 | gv_efullname3(temp, MUTABLE_GV(sv), "*"); |
| 10369 | |
| 10370 | SvREFCNT_inc_simple_void_NN(sv_2mortal(sv)); |
| 10371 | if (GvGP(sv)) { |
| 10372 | if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv))) |
| 10373 | && HvNAME_get(stash)) |
| 10374 | mro_method_changed_in(stash); |
| 10375 | gp_free(MUTABLE_GV(sv)); |
| 10376 | } |
| 10377 | if (GvSTASH(sv)) { |
| 10378 | sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv); |
| 10379 | GvSTASH(sv) = NULL; |
| 10380 | } |
| 10381 | GvMULTI_off(sv); |
| 10382 | if (GvNAME_HEK(sv)) { |
| 10383 | unshare_hek(GvNAME_HEK(sv)); |
| 10384 | } |
| 10385 | isGV_with_GP_off(sv); |
| 10386 | |
| 10387 | if(SvTYPE(sv) == SVt_PVGV) { |
| 10388 | /* need to keep SvANY(sv) in the right arena */ |
| 10389 | xpvmg = new_XPVMG(); |
| 10390 | StructCopy(SvANY(sv), xpvmg, XPVMG); |
| 10391 | del_XPVGV(SvANY(sv)); |
| 10392 | SvANY(sv) = xpvmg; |
| 10393 | |
| 10394 | SvFLAGS(sv) &= ~SVTYPEMASK; |
| 10395 | SvFLAGS(sv) |= SVt_PVMG; |
| 10396 | } |
| 10397 | |
| 10398 | /* Intentionally not calling any local SET magic, as this isn't so much a |
| 10399 | set operation as merely an internal storage change. */ |
| 10400 | if (flags & SV_COW_DROP_PV) SvOK_off(sv); |
| 10401 | else sv_setsv_flags(sv, temp, 0); |
| 10402 | |
| 10403 | if ((const GV *)sv == PL_last_in_gv) |
| 10404 | PL_last_in_gv = NULL; |
| 10405 | else if ((const GV *)sv == PL_statgv) |
| 10406 | PL_statgv = NULL; |
| 10407 | } |
| 10408 | |
| 10409 | /* |
| 10410 | =for apidoc sv_unref_flags |
| 10411 | |
| 10412 | Unsets the RV status of the SV, and decrements the reference count of |
| 10413 | whatever was being referenced by the RV. This can almost be thought of |
| 10414 | as a reversal of C<newSVrv>. The C<cflags> argument can contain |
| 10415 | C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented |
| 10416 | (otherwise the decrementing is conditional on the reference count being |
| 10417 | different from one or the reference being a readonly SV). |
| 10418 | See C<L</SvROK_off>>. |
| 10419 | |
| 10420 | =cut |
| 10421 | */ |
| 10422 | |
| 10423 | void |
| 10424 | Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags) |
| 10425 | { |
| 10426 | SV* const target = SvRV(ref); |
| 10427 | |
| 10428 | PERL_ARGS_ASSERT_SV_UNREF_FLAGS; |
| 10429 | |
| 10430 | if (SvWEAKREF(ref)) { |
| 10431 | sv_del_backref(target, ref); |
| 10432 | SvWEAKREF_off(ref); |
| 10433 | SvRV_set(ref, NULL); |
| 10434 | return; |
| 10435 | } |
| 10436 | SvRV_set(ref, NULL); |
| 10437 | SvROK_off(ref); |
| 10438 | /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was |
| 10439 | assigned to as BEGIN {$a = \"Foo"} will fail. */ |
| 10440 | if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF)) |
| 10441 | SvREFCNT_dec_NN(target); |
| 10442 | else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */ |
| 10443 | sv_2mortal(target); /* Schedule for freeing later */ |
| 10444 | } |
| 10445 | |
| 10446 | /* |
| 10447 | =for apidoc sv_untaint |
| 10448 | |
| 10449 | Untaint an SV. Use C<SvTAINTED_off> instead. |
| 10450 | |
| 10451 | =cut |
| 10452 | */ |
| 10453 | |
| 10454 | void |
| 10455 | Perl_sv_untaint(pTHX_ SV *const sv) |
| 10456 | { |
| 10457 | PERL_ARGS_ASSERT_SV_UNTAINT; |
| 10458 | PERL_UNUSED_CONTEXT; |
| 10459 | |
| 10460 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 10461 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint); |
| 10462 | if (mg) |
| 10463 | mg->mg_len &= ~1; |
| 10464 | } |
| 10465 | } |
| 10466 | |
| 10467 | /* |
| 10468 | =for apidoc sv_tainted |
| 10469 | |
| 10470 | Test an SV for taintedness. Use C<SvTAINTED> instead. |
| 10471 | |
| 10472 | =cut |
| 10473 | */ |
| 10474 | |
| 10475 | bool |
| 10476 | Perl_sv_tainted(pTHX_ SV *const sv) |
| 10477 | { |
| 10478 | PERL_ARGS_ASSERT_SV_TAINTED; |
| 10479 | PERL_UNUSED_CONTEXT; |
| 10480 | |
| 10481 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 10482 | const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint); |
| 10483 | if (mg && (mg->mg_len & 1) ) |
| 10484 | return TRUE; |
| 10485 | } |
| 10486 | return FALSE; |
| 10487 | } |
| 10488 | |
| 10489 | /* |
| 10490 | =for apidoc sv_setpviv |
| 10491 | |
| 10492 | Copies an integer into the given SV, also updating its string value. |
| 10493 | Does not handle 'set' magic. See C<L</sv_setpviv_mg>>. |
| 10494 | |
| 10495 | =cut |
| 10496 | */ |
| 10497 | |
| 10498 | void |
| 10499 | Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv) |
| 10500 | { |
| 10501 | char buf[TYPE_CHARS(UV)]; |
| 10502 | char *ebuf; |
| 10503 | char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf); |
| 10504 | |
| 10505 | PERL_ARGS_ASSERT_SV_SETPVIV; |
| 10506 | |
| 10507 | sv_setpvn(sv, ptr, ebuf - ptr); |
| 10508 | } |
| 10509 | |
| 10510 | /* |
| 10511 | =for apidoc sv_setpviv_mg |
| 10512 | |
| 10513 | Like C<sv_setpviv>, but also handles 'set' magic. |
| 10514 | |
| 10515 | =cut |
| 10516 | */ |
| 10517 | |
| 10518 | void |
| 10519 | Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv) |
| 10520 | { |
| 10521 | PERL_ARGS_ASSERT_SV_SETPVIV_MG; |
| 10522 | |
| 10523 | sv_setpviv(sv, iv); |
| 10524 | SvSETMAGIC(sv); |
| 10525 | } |
| 10526 | |
| 10527 | #if defined(PERL_IMPLICIT_CONTEXT) |
| 10528 | |
| 10529 | /* pTHX_ magic can't cope with varargs, so this is a no-context |
| 10530 | * version of the main function, (which may itself be aliased to us). |
| 10531 | * Don't access this version directly. |
| 10532 | */ |
| 10533 | |
| 10534 | void |
| 10535 | Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...) |
| 10536 | { |
| 10537 | dTHX; |
| 10538 | va_list args; |
| 10539 | |
| 10540 | PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT; |
| 10541 | |
| 10542 | va_start(args, pat); |
| 10543 | sv_vsetpvf(sv, pat, &args); |
| 10544 | va_end(args); |
| 10545 | } |
| 10546 | |
| 10547 | /* pTHX_ magic can't cope with varargs, so this is a no-context |
| 10548 | * version of the main function, (which may itself be aliased to us). |
| 10549 | * Don't access this version directly. |
| 10550 | */ |
| 10551 | |
| 10552 | void |
| 10553 | Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...) |
| 10554 | { |
| 10555 | dTHX; |
| 10556 | va_list args; |
| 10557 | |
| 10558 | PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT; |
| 10559 | |
| 10560 | va_start(args, pat); |
| 10561 | sv_vsetpvf_mg(sv, pat, &args); |
| 10562 | va_end(args); |
| 10563 | } |
| 10564 | #endif |
| 10565 | |
| 10566 | /* |
| 10567 | =for apidoc sv_setpvf |
| 10568 | |
| 10569 | Works like C<sv_catpvf> but copies the text into the SV instead of |
| 10570 | appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>. |
| 10571 | |
| 10572 | =cut |
| 10573 | */ |
| 10574 | |
| 10575 | void |
| 10576 | Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...) |
| 10577 | { |
| 10578 | va_list args; |
| 10579 | |
| 10580 | PERL_ARGS_ASSERT_SV_SETPVF; |
| 10581 | |
| 10582 | va_start(args, pat); |
| 10583 | sv_vsetpvf(sv, pat, &args); |
| 10584 | va_end(args); |
| 10585 | } |
| 10586 | |
| 10587 | /* |
| 10588 | =for apidoc sv_vsetpvf |
| 10589 | |
| 10590 | Works like C<sv_vcatpvf> but copies the text into the SV instead of |
| 10591 | appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>. |
| 10592 | |
| 10593 | Usually used via its frontend C<sv_setpvf>. |
| 10594 | |
| 10595 | =cut |
| 10596 | */ |
| 10597 | |
| 10598 | void |
| 10599 | Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args) |
| 10600 | { |
| 10601 | PERL_ARGS_ASSERT_SV_VSETPVF; |
| 10602 | |
| 10603 | sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); |
| 10604 | } |
| 10605 | |
| 10606 | /* |
| 10607 | =for apidoc sv_setpvf_mg |
| 10608 | |
| 10609 | Like C<sv_setpvf>, but also handles 'set' magic. |
| 10610 | |
| 10611 | =cut |
| 10612 | */ |
| 10613 | |
| 10614 | void |
| 10615 | Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...) |
| 10616 | { |
| 10617 | va_list args; |
| 10618 | |
| 10619 | PERL_ARGS_ASSERT_SV_SETPVF_MG; |
| 10620 | |
| 10621 | va_start(args, pat); |
| 10622 | sv_vsetpvf_mg(sv, pat, &args); |
| 10623 | va_end(args); |
| 10624 | } |
| 10625 | |
| 10626 | /* |
| 10627 | =for apidoc sv_vsetpvf_mg |
| 10628 | |
| 10629 | Like C<sv_vsetpvf>, but also handles 'set' magic. |
| 10630 | |
| 10631 | Usually used via its frontend C<sv_setpvf_mg>. |
| 10632 | |
| 10633 | =cut |
| 10634 | */ |
| 10635 | |
| 10636 | void |
| 10637 | Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args) |
| 10638 | { |
| 10639 | PERL_ARGS_ASSERT_SV_VSETPVF_MG; |
| 10640 | |
| 10641 | sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); |
| 10642 | SvSETMAGIC(sv); |
| 10643 | } |
| 10644 | |
| 10645 | #if defined(PERL_IMPLICIT_CONTEXT) |
| 10646 | |
| 10647 | /* pTHX_ magic can't cope with varargs, so this is a no-context |
| 10648 | * version of the main function, (which may itself be aliased to us). |
| 10649 | * Don't access this version directly. |
| 10650 | */ |
| 10651 | |
| 10652 | void |
| 10653 | Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...) |
| 10654 | { |
| 10655 | dTHX; |
| 10656 | va_list args; |
| 10657 | |
| 10658 | PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT; |
| 10659 | |
| 10660 | va_start(args, pat); |
| 10661 | sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); |
| 10662 | va_end(args); |
| 10663 | } |
| 10664 | |
| 10665 | /* pTHX_ magic can't cope with varargs, so this is a no-context |
| 10666 | * version of the main function, (which may itself be aliased to us). |
| 10667 | * Don't access this version directly. |
| 10668 | */ |
| 10669 | |
| 10670 | void |
| 10671 | Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...) |
| 10672 | { |
| 10673 | dTHX; |
| 10674 | va_list args; |
| 10675 | |
| 10676 | PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT; |
| 10677 | |
| 10678 | va_start(args, pat); |
| 10679 | sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); |
| 10680 | SvSETMAGIC(sv); |
| 10681 | va_end(args); |
| 10682 | } |
| 10683 | #endif |
| 10684 | |
| 10685 | /* |
| 10686 | =for apidoc sv_catpvf |
| 10687 | |
| 10688 | Processes its arguments like C<sv_catpvfn>, and appends the formatted |
| 10689 | output to an SV. As with C<sv_catpvfn> called with a non-null C-style |
| 10690 | variable argument list, argument reordering is not supported. |
| 10691 | If the appended data contains "wide" characters |
| 10692 | (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>, |
| 10693 | and characters >255 formatted with C<%c>), the original SV might get |
| 10694 | upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See |
| 10695 | C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be |
| 10696 | valid UTF-8; if the original SV was bytes, the pattern should be too. |
| 10697 | |
| 10698 | =cut */ |
| 10699 | |
| 10700 | void |
| 10701 | Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...) |
| 10702 | { |
| 10703 | va_list args; |
| 10704 | |
| 10705 | PERL_ARGS_ASSERT_SV_CATPVF; |
| 10706 | |
| 10707 | va_start(args, pat); |
| 10708 | sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); |
| 10709 | va_end(args); |
| 10710 | } |
| 10711 | |
| 10712 | /* |
| 10713 | =for apidoc sv_vcatpvf |
| 10714 | |
| 10715 | Processes its arguments like C<sv_catpvfn> called with a non-null C-style |
| 10716 | variable argument list, and appends the formatted |
| 10717 | to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>. |
| 10718 | |
| 10719 | Usually used via its frontend C<sv_catpvf>. |
| 10720 | |
| 10721 | =cut |
| 10722 | */ |
| 10723 | |
| 10724 | void |
| 10725 | Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args) |
| 10726 | { |
| 10727 | PERL_ARGS_ASSERT_SV_VCATPVF; |
| 10728 | |
| 10729 | sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); |
| 10730 | } |
| 10731 | |
| 10732 | /* |
| 10733 | =for apidoc sv_catpvf_mg |
| 10734 | |
| 10735 | Like C<sv_catpvf>, but also handles 'set' magic. |
| 10736 | |
| 10737 | =cut |
| 10738 | */ |
| 10739 | |
| 10740 | void |
| 10741 | Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...) |
| 10742 | { |
| 10743 | va_list args; |
| 10744 | |
| 10745 | PERL_ARGS_ASSERT_SV_CATPVF_MG; |
| 10746 | |
| 10747 | va_start(args, pat); |
| 10748 | sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC); |
| 10749 | SvSETMAGIC(sv); |
| 10750 | va_end(args); |
| 10751 | } |
| 10752 | |
| 10753 | /* |
| 10754 | =for apidoc sv_vcatpvf_mg |
| 10755 | |
| 10756 | Like C<sv_vcatpvf>, but also handles 'set' magic. |
| 10757 | |
| 10758 | Usually used via its frontend C<sv_catpvf_mg>. |
| 10759 | |
| 10760 | =cut |
| 10761 | */ |
| 10762 | |
| 10763 | void |
| 10764 | Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args) |
| 10765 | { |
| 10766 | PERL_ARGS_ASSERT_SV_VCATPVF_MG; |
| 10767 | |
| 10768 | sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL); |
| 10769 | SvSETMAGIC(sv); |
| 10770 | } |
| 10771 | |
| 10772 | /* |
| 10773 | =for apidoc sv_vsetpvfn |
| 10774 | |
| 10775 | Works like C<sv_vcatpvfn> but copies the text into the SV instead of |
| 10776 | appending it. |
| 10777 | |
| 10778 | Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>. |
| 10779 | |
| 10780 | =cut |
| 10781 | */ |
| 10782 | |
| 10783 | void |
| 10784 | Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, |
| 10785 | va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted) |
| 10786 | { |
| 10787 | PERL_ARGS_ASSERT_SV_VSETPVFN; |
| 10788 | |
| 10789 | sv_setpvs(sv, ""); |
| 10790 | sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0); |
| 10791 | } |
| 10792 | |
| 10793 | |
| 10794 | /* |
| 10795 | * Warn of missing argument to sprintf. The value used in place of such |
| 10796 | * arguments should be &PL_sv_no; an undefined value would yield |
| 10797 | * inappropriate "use of uninit" warnings [perl #71000]. |
| 10798 | */ |
| 10799 | STATIC void |
| 10800 | S_warn_vcatpvfn_missing_argument(pTHX) { |
| 10801 | if (ckWARN(WARN_MISSING)) { |
| 10802 | Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s", |
| 10803 | PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); |
| 10804 | } |
| 10805 | } |
| 10806 | |
| 10807 | |
| 10808 | STATIC I32 |
| 10809 | S_expect_number(pTHX_ char **const pattern) |
| 10810 | { |
| 10811 | I32 var = 0; |
| 10812 | |
| 10813 | PERL_ARGS_ASSERT_EXPECT_NUMBER; |
| 10814 | |
| 10815 | switch (**pattern) { |
| 10816 | case '1': case '2': case '3': |
| 10817 | case '4': case '5': case '6': |
| 10818 | case '7': case '8': case '9': |
| 10819 | var = *(*pattern)++ - '0'; |
| 10820 | while (isDIGIT(**pattern)) { |
| 10821 | const I32 tmp = var * 10 + (*(*pattern)++ - '0'); |
| 10822 | if (tmp < var) |
| 10823 | Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn")); |
| 10824 | var = tmp; |
| 10825 | } |
| 10826 | } |
| 10827 | return var; |
| 10828 | } |
| 10829 | |
| 10830 | STATIC char * |
| 10831 | S_F0convert(NV nv, char *const endbuf, STRLEN *const len) |
| 10832 | { |
| 10833 | const int neg = nv < 0; |
| 10834 | UV uv; |
| 10835 | |
| 10836 | PERL_ARGS_ASSERT_F0CONVERT; |
| 10837 | |
| 10838 | if (UNLIKELY(Perl_isinfnan(nv))) { |
| 10839 | STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0); |
| 10840 | *len = n; |
| 10841 | return endbuf - n; |
| 10842 | } |
| 10843 | if (neg) |
| 10844 | nv = -nv; |
| 10845 | if (nv < UV_MAX) { |
| 10846 | char *p = endbuf; |
| 10847 | nv += 0.5; |
| 10848 | uv = (UV)nv; |
| 10849 | if (uv & 1 && uv == nv) |
| 10850 | uv--; /* Round to even */ |
| 10851 | do { |
| 10852 | const unsigned dig = uv % 10; |
| 10853 | *--p = '0' + dig; |
| 10854 | } while (uv /= 10); |
| 10855 | if (neg) |
| 10856 | *--p = '-'; |
| 10857 | *len = endbuf - p; |
| 10858 | return p; |
| 10859 | } |
| 10860 | return NULL; |
| 10861 | } |
| 10862 | |
| 10863 | |
| 10864 | /* |
| 10865 | =for apidoc sv_vcatpvfn |
| 10866 | |
| 10867 | =for apidoc sv_vcatpvfn_flags |
| 10868 | |
| 10869 | Processes its arguments like C<vsprintf> and appends the formatted output |
| 10870 | to an SV. Uses an array of SVs if the C-style variable argument list is |
| 10871 | missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d> |
| 10872 | or C<%*2$d>) is supported only when using an array of SVs; using a C-style |
| 10873 | C<va_list> argument list with a format string that uses argument reordering |
| 10874 | will yield an exception. |
| 10875 | |
| 10876 | When running with taint checks enabled, indicates via |
| 10877 | C<maybe_tainted> if results are untrustworthy (often due to the use of |
| 10878 | locales). |
| 10879 | |
| 10880 | If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic. |
| 10881 | |
| 10882 | Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>. |
| 10883 | |
| 10884 | =cut |
| 10885 | */ |
| 10886 | |
| 10887 | #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\ |
| 10888 | vecstr = (U8*)SvPV_const(vecsv,veclen);\ |
| 10889 | vec_utf8 = DO_UTF8(vecsv); |
| 10890 | |
| 10891 | /* XXX maybe_tainted is never assigned to, so the doc above is lying. */ |
| 10892 | |
| 10893 | void |
| 10894 | Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, |
| 10895 | va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted) |
| 10896 | { |
| 10897 | PERL_ARGS_ASSERT_SV_VCATPVFN; |
| 10898 | |
| 10899 | sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC); |
| 10900 | } |
| 10901 | |
| 10902 | #ifdef LONGDOUBLE_DOUBLEDOUBLE |
| 10903 | /* The first double can be as large as 2**1023, or '1' x '0' x 1023. |
| 10904 | * The second double can be as small as 2**-1074, or '0' x 1073 . '1'. |
| 10905 | * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point |
| 10906 | * after the first 1023 zero bits. |
| 10907 | * |
| 10908 | * XXX The 2098 is quite large (262.25 bytes) and therefore some sort |
| 10909 | * of dynamically growing buffer might be better, start at just 16 bytes |
| 10910 | * (for example) and grow only when necessary. Or maybe just by looking |
| 10911 | * at the exponents of the two doubles? */ |
| 10912 | # define DOUBLEDOUBLE_MAXBITS 2098 |
| 10913 | #endif |
| 10914 | |
| 10915 | /* vhex will contain the values (0..15) of the hex digits ("nybbles" |
| 10916 | * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits |
| 10917 | * per xdigit. For the double-double case, this can be rather many. |
| 10918 | * The non-double-double-long-double overshoots since all bits of NV |
| 10919 | * are not mantissa bits, there are also exponent bits. */ |
| 10920 | #ifdef LONGDOUBLE_DOUBLEDOUBLE |
| 10921 | # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4) |
| 10922 | #else |
| 10923 | # define VHEX_SIZE (1+(NVSIZE * 8)/4) |
| 10924 | #endif |
| 10925 | |
| 10926 | /* If we do not have a known long double format, (including not using |
| 10927 | * long doubles, or long doubles being equal to doubles) then we will |
| 10928 | * fall back to the ldexp/frexp route, with which we can retrieve at |
| 10929 | * most as many bits as our widest unsigned integer type is. We try |
| 10930 | * to get a 64-bit unsigned integer even if we are not using a 64-bit UV. |
| 10931 | * |
| 10932 | * (If you want to test the case of UVSIZE == 4, NVSIZE == 8, |
| 10933 | * set the MANTISSATYPE to int and the MANTISSASIZE to 4.) |
| 10934 | */ |
| 10935 | #if defined(HAS_QUAD) && defined(Uquad_t) |
| 10936 | # define MANTISSATYPE Uquad_t |
| 10937 | # define MANTISSASIZE 8 |
| 10938 | #else |
| 10939 | # define MANTISSATYPE UV |
| 10940 | # define MANTISSASIZE UVSIZE |
| 10941 | #endif |
| 10942 | |
| 10943 | #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN) |
| 10944 | # define HEXTRACT_LITTLE_ENDIAN |
| 10945 | #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN) |
| 10946 | # define HEXTRACT_BIG_ENDIAN |
| 10947 | #else |
| 10948 | # define HEXTRACT_MIX_ENDIAN |
| 10949 | #endif |
| 10950 | |
| 10951 | /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting |
| 10952 | * the hexadecimal values (for %a/%A). The nv is the NV where the value |
| 10953 | * are being extracted from (either directly from the long double in-memory |
| 10954 | * presentation, or from the uquad computed via frexp+ldexp). frexp also |
| 10955 | * is used to update the exponent. vhex is the pointer to the beginning |
| 10956 | * of the output buffer (of VHEX_SIZE). |
| 10957 | * |
| 10958 | * The tricky part is that S_hextract() needs to be called twice: |
| 10959 | * the first time with vend as NULL, and the second time with vend as |
| 10960 | * the pointer returned by the first call. What happens is that on |
| 10961 | * the first round the output size is computed, and the intended |
| 10962 | * extraction sanity checked. On the second round the actual output |
| 10963 | * (the extraction of the hexadecimal values) takes place. |
| 10964 | * Sanity failures cause fatal failures during both rounds. */ |
| 10965 | STATIC U8* |
| 10966 | S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend) |
| 10967 | { |
| 10968 | U8* v = vhex; |
| 10969 | int ix; |
| 10970 | int ixmin = 0, ixmax = 0; |
| 10971 | |
| 10972 | /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT, |
| 10973 | * and elsewhere. */ |
| 10974 | |
| 10975 | /* These macros are just to reduce typos, they have multiple |
| 10976 | * repetitions below, but usually only one (or sometimes two) |
| 10977 | * of them is really being used. */ |
| 10978 | /* HEXTRACT_OUTPUT() extracts the high nybble first. */ |
| 10979 | #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4) |
| 10980 | #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF) |
| 10981 | #define HEXTRACT_OUTPUT(ix) \ |
| 10982 | STMT_START { \ |
| 10983 | HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \ |
| 10984 | } STMT_END |
| 10985 | #define HEXTRACT_COUNT(ix, c) \ |
| 10986 | STMT_START { \ |
| 10987 | v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \ |
| 10988 | } STMT_END |
| 10989 | #define HEXTRACT_BYTE(ix) \ |
| 10990 | STMT_START { \ |
| 10991 | if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \ |
| 10992 | } STMT_END |
| 10993 | #define HEXTRACT_LO_NYBBLE(ix) \ |
| 10994 | STMT_START { \ |
| 10995 | if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \ |
| 10996 | } STMT_END |
| 10997 | /* HEXTRACT_TOP_NYBBLE is just convenience disguise, |
| 10998 | * to make it look less odd when the top bits of a NV |
| 10999 | * are extracted using HEXTRACT_LO_NYBBLE: the highest |
| 11000 | * order bits can be in the "low nybble" of a byte. */ |
| 11001 | #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix) |
| 11002 | #define HEXTRACT_BYTES_LE(a, b) \ |
| 11003 | for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); } |
| 11004 | #define HEXTRACT_BYTES_BE(a, b) \ |
| 11005 | for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); } |
| 11006 | #define HEXTRACT_IMPLICIT_BIT(nv) \ |
| 11007 | STMT_START { \ |
| 11008 | if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \ |
| 11009 | } STMT_END |
| 11010 | |
| 11011 | /* Most formats do. Those which don't should undef this. */ |
| 11012 | #define HEXTRACT_HAS_IMPLICIT_BIT |
| 11013 | /* Many formats do. Those which don't should undef this. */ |
| 11014 | #define HEXTRACT_HAS_TOP_NYBBLE |
| 11015 | |
| 11016 | /* HEXTRACTSIZE is the maximum number of xdigits. */ |
| 11017 | #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE) |
| 11018 | # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4) |
| 11019 | #else |
| 11020 | # define HEXTRACTSIZE 2 * NVSIZE |
| 11021 | #endif |
| 11022 | |
| 11023 | const U8* vmaxend = vhex + HEXTRACTSIZE; |
| 11024 | PERL_UNUSED_VAR(ix); /* might happen */ |
| 11025 | (void)Perl_frexp(PERL_ABS(nv), exponent); |
| 11026 | if (vend && (vend <= vhex || vend > vmaxend)) { |
| 11027 | /* diag_listed_as: Hexadecimal float: internal error (%s) */ |
| 11028 | Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)"); |
| 11029 | } |
| 11030 | { |
| 11031 | /* First check if using long doubles. */ |
| 11032 | #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) |
| 11033 | # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN |
| 11034 | /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L: |
| 11035 | * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */ |
| 11036 | /* The bytes 13..0 are the mantissa/fraction, |
| 11037 | * the 15,14 are the sign+exponent. */ |
| 11038 | const U8* nvp = (const U8*)(&nv); |
| 11039 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11040 | # undef HEXTRACT_HAS_TOP_NYBBLE |
| 11041 | HEXTRACT_BYTES_LE(13, 0); |
| 11042 | # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN |
| 11043 | /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L: |
| 11044 | * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */ |
| 11045 | /* The bytes 2..15 are the mantissa/fraction, |
| 11046 | * the 0,1 are the sign+exponent. */ |
| 11047 | const U8* nvp = (const U8*)(&nv); |
| 11048 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11049 | # undef HEXTRACT_HAS_TOP_NYBBLE |
| 11050 | HEXTRACT_BYTES_BE(2, 15); |
| 11051 | # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN |
| 11052 | /* x86 80-bit "extended precision", 64 bits of mantissa / fraction / |
| 11053 | * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can |
| 11054 | * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X), |
| 11055 | * meaning that 2 or 6 bytes are empty padding. */ |
| 11056 | /* The bytes 7..0 are the mantissa/fraction */ |
| 11057 | const U8* nvp = (const U8*)(&nv); |
| 11058 | # undef HEXTRACT_HAS_IMPLICIT_BIT |
| 11059 | # undef HEXTRACT_HAS_TOP_NYBBLE |
| 11060 | HEXTRACT_BYTES_LE(7, 0); |
| 11061 | # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN |
| 11062 | /* Does this format ever happen? (Wikipedia says the Motorola |
| 11063 | * 6888x math coprocessors used format _like_ this but padded |
| 11064 | * to 96 bits with 16 unused bits between the exponent and the |
| 11065 | * mantissa.) */ |
| 11066 | const U8* nvp = (const U8*)(&nv); |
| 11067 | # undef HEXTRACT_HAS_IMPLICIT_BIT |
| 11068 | # undef HEXTRACT_HAS_TOP_NYBBLE |
| 11069 | HEXTRACT_BYTES_BE(0, 7); |
| 11070 | # else |
| 11071 | # define HEXTRACT_FALLBACK |
| 11072 | /* Double-double format: two doubles next to each other. |
| 11073 | * The first double is the high-order one, exactly like |
| 11074 | * it would be for a "lone" double. The second double |
| 11075 | * is shifted down using the exponent so that that there |
| 11076 | * are no common bits. The tricky part is that the value |
| 11077 | * of the double-double is the SUM of the two doubles and |
| 11078 | * the second one can be also NEGATIVE. |
| 11079 | * |
| 11080 | * Because of this tricky construction the bytewise extraction we |
| 11081 | * use for the other long double formats doesn't work, we must |
| 11082 | * extract the values bit by bit. |
| 11083 | * |
| 11084 | * The little-endian double-double is used .. somewhere? |
| 11085 | * |
| 11086 | * The big endian double-double is used in e.g. PPC/Power (AIX) |
| 11087 | * and MIPS (SGI). |
| 11088 | * |
| 11089 | * The mantissa bits are in two separate stretches, e.g. for -0.1L: |
| 11090 | * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE) |
| 11091 | * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE) |
| 11092 | */ |
| 11093 | # endif |
| 11094 | #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */ |
| 11095 | /* Using normal doubles, not long doubles. |
| 11096 | * |
| 11097 | * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit |
| 11098 | * bytes, since we might need to handle printf precision, and |
| 11099 | * also need to insert the radix. */ |
| 11100 | # if NVSIZE == 8 |
| 11101 | # ifdef HEXTRACT_LITTLE_ENDIAN |
| 11102 | /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */ |
| 11103 | const U8* nvp = (const U8*)(&nv); |
| 11104 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11105 | HEXTRACT_TOP_NYBBLE(6); |
| 11106 | HEXTRACT_BYTES_LE(5, 0); |
| 11107 | # elif defined(HEXTRACT_BIG_ENDIAN) |
| 11108 | /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */ |
| 11109 | const U8* nvp = (const U8*)(&nv); |
| 11110 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11111 | HEXTRACT_TOP_NYBBLE(1); |
| 11112 | HEXTRACT_BYTES_BE(2, 7); |
| 11113 | # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE |
| 11114 | /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */ |
| 11115 | const U8* nvp = (const U8*)(&nv); |
| 11116 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11117 | HEXTRACT_TOP_NYBBLE(2); /* 6 */ |
| 11118 | HEXTRACT_BYTE(1); /* 5 */ |
| 11119 | HEXTRACT_BYTE(0); /* 4 */ |
| 11120 | HEXTRACT_BYTE(7); /* 3 */ |
| 11121 | HEXTRACT_BYTE(6); /* 2 */ |
| 11122 | HEXTRACT_BYTE(5); /* 1 */ |
| 11123 | HEXTRACT_BYTE(4); /* 0 */ |
| 11124 | # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE |
| 11125 | /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */ |
| 11126 | const U8* nvp = (const U8*)(&nv); |
| 11127 | HEXTRACT_IMPLICIT_BIT(nv); |
| 11128 | HEXTRACT_TOP_NYBBLE(5); /* 6 */ |
| 11129 | HEXTRACT_BYTE(6); /* 5 */ |
| 11130 | HEXTRACT_BYTE(7); /* 4 */ |
| 11131 | HEXTRACT_BYTE(0); /* 3 */ |
| 11132 | HEXTRACT_BYTE(1); /* 2 */ |
| 11133 | HEXTRACT_BYTE(2); /* 1 */ |
| 11134 | HEXTRACT_BYTE(3); /* 0 */ |
| 11135 | # else |
| 11136 | # define HEXTRACT_FALLBACK |
| 11137 | # endif |
| 11138 | # else |
| 11139 | # define HEXTRACT_FALLBACK |
| 11140 | # endif |
| 11141 | #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */ |
| 11142 | # ifdef HEXTRACT_FALLBACK |
| 11143 | # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */ |
| 11144 | /* The fallback is used for the double-double format, and |
| 11145 | * for unknown long double formats, and for unknown double |
| 11146 | * formats, or in general unknown NV formats. */ |
| 11147 | if (nv == (NV)0.0) { |
| 11148 | if (vend) |
| 11149 | *v++ = 0; |
| 11150 | else |
| 11151 | v++; |
| 11152 | *exponent = 0; |
| 11153 | } |
| 11154 | else { |
| 11155 | NV d = nv < 0 ? -nv : nv; |
| 11156 | NV e = (NV)1.0; |
| 11157 | U8 ha = 0x0; /* hexvalue accumulator */ |
| 11158 | U8 hd = 0x8; /* hexvalue digit */ |
| 11159 | |
| 11160 | /* Shift d and e (and update exponent) so that e <= d < 2*e, |
| 11161 | * this is essentially manual frexp(). Multiplying by 0.5 and |
| 11162 | * doubling should be lossless in binary floating point. */ |
| 11163 | |
| 11164 | *exponent = 1; |
| 11165 | |
| 11166 | while (e > d) { |
| 11167 | e *= (NV)0.5; |
| 11168 | (*exponent)--; |
| 11169 | } |
| 11170 | /* Now d >= e */ |
| 11171 | |
| 11172 | while (d >= e + e) { |
| 11173 | e += e; |
| 11174 | (*exponent)++; |
| 11175 | } |
| 11176 | /* Now e <= d < 2*e */ |
| 11177 | |
| 11178 | /* First extract the leading hexdigit (the implicit bit). */ |
| 11179 | if (d >= e) { |
| 11180 | d -= e; |
| 11181 | if (vend) |
| 11182 | *v++ = 1; |
| 11183 | else |
| 11184 | v++; |
| 11185 | } |
| 11186 | else { |
| 11187 | if (vend) |
| 11188 | *v++ = 0; |
| 11189 | else |
| 11190 | v++; |
| 11191 | } |
| 11192 | e *= (NV)0.5; |
| 11193 | |
| 11194 | /* Then extract the remaining hexdigits. */ |
| 11195 | while (d > (NV)0.0) { |
| 11196 | if (d >= e) { |
| 11197 | ha |= hd; |
| 11198 | d -= e; |
| 11199 | } |
| 11200 | if (hd == 1) { |
| 11201 | /* Output or count in groups of four bits, |
| 11202 | * that is, when the hexdigit is down to one. */ |
| 11203 | if (vend) |
| 11204 | *v++ = ha; |
| 11205 | else |
| 11206 | v++; |
| 11207 | /* Reset the hexvalue. */ |
| 11208 | ha = 0x0; |
| 11209 | hd = 0x8; |
| 11210 | } |
| 11211 | else |
| 11212 | hd >>= 1; |
| 11213 | e *= (NV)0.5; |
| 11214 | } |
| 11215 | |
| 11216 | /* Flush possible pending hexvalue. */ |
| 11217 | if (ha) { |
| 11218 | if (vend) |
| 11219 | *v++ = ha; |
| 11220 | else |
| 11221 | v++; |
| 11222 | } |
| 11223 | } |
| 11224 | # endif |
| 11225 | } |
| 11226 | /* Croak for various reasons: if the output pointer escaped the |
| 11227 | * output buffer, if the extraction index escaped the extraction |
| 11228 | * buffer, or if the ending output pointer didn't match the |
| 11229 | * previously computed value. */ |
| 11230 | if (v <= vhex || v - vhex >= VHEX_SIZE || |
| 11231 | /* For double-double the ixmin and ixmax stay at zero, |
| 11232 | * which is convenient since the HEXTRACTSIZE is tricky |
| 11233 | * for double-double. */ |
| 11234 | ixmin < 0 || ixmax >= NVSIZE || |
| 11235 | (vend && v != vend)) { |
| 11236 | /* diag_listed_as: Hexadecimal float: internal error (%s) */ |
| 11237 | Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)"); |
| 11238 | } |
| 11239 | return v; |
| 11240 | } |
| 11241 | |
| 11242 | /* Helper for sv_vcatpvfn_flags(). */ |
| 11243 | #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \ |
| 11244 | STMT_START { \ |
| 11245 | if (in_range) \ |
| 11246 | (var) = (expr); \ |
| 11247 | else { \ |
| 11248 | (var) = &PL_sv_no; /* [perl #71000] */ \ |
| 11249 | arg_missing = TRUE; \ |
| 11250 | } \ |
| 11251 | } STMT_END |
| 11252 | |
| 11253 | void |
| 11254 | Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen, |
| 11255 | va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted, |
| 11256 | const U32 flags) |
| 11257 | { |
| 11258 | char *p; |
| 11259 | char *q; |
| 11260 | const char *patend; |
| 11261 | STRLEN origlen; |
| 11262 | I32 svix = 0; |
| 11263 | static const char nullstr[] = "(null)"; |
| 11264 | SV *argsv = NULL; |
| 11265 | bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */ |
| 11266 | const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */ |
| 11267 | SV *nsv = NULL; |
| 11268 | /* Times 4: a decimal digit takes more than 3 binary digits. |
| 11269 | * NV_DIG: mantissa takes than many decimal digits. |
| 11270 | * Plus 32: Playing safe. */ |
| 11271 | char ebuf[IV_DIG * 4 + NV_DIG + 32]; |
| 11272 | bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */ |
| 11273 | bool hexfp = FALSE; /* hexadecimal floating point? */ |
| 11274 | |
| 11275 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
| 11276 | |
| 11277 | PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS; |
| 11278 | PERL_UNUSED_ARG(maybe_tainted); |
| 11279 | |
| 11280 | if (flags & SV_GMAGIC) |
| 11281 | SvGETMAGIC(sv); |
| 11282 | |
| 11283 | /* no matter what, this is a string now */ |
| 11284 | (void)SvPV_force_nomg(sv, origlen); |
| 11285 | |
| 11286 | /* special-case "", "%s", and "%-p" (SVf - see below) */ |
| 11287 | if (patlen == 0) { |
| 11288 | if (svmax && ckWARN(WARN_REDUNDANT)) |
| 11289 | Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s", |
| 11290 | PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); |
| 11291 | return; |
| 11292 | } |
| 11293 | if (patlen == 2 && pat[0] == '%' && pat[1] == 's') { |
| 11294 | if (svmax > 1 && ckWARN(WARN_REDUNDANT)) |
| 11295 | Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s", |
| 11296 | PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); |
| 11297 | |
| 11298 | if (args) { |
| 11299 | const char * const s = va_arg(*args, char*); |
| 11300 | sv_catpv_nomg(sv, s ? s : nullstr); |
| 11301 | } |
| 11302 | else if (svix < svmax) { |
| 11303 | /* we want get magic on the source but not the target. sv_catsv can't do that, though */ |
| 11304 | SvGETMAGIC(*svargs); |
| 11305 | sv_catsv_nomg(sv, *svargs); |
| 11306 | } |
| 11307 | else |
| 11308 | S_warn_vcatpvfn_missing_argument(aTHX); |
| 11309 | return; |
| 11310 | } |
| 11311 | if (args && patlen == 3 && pat[0] == '%' && |
| 11312 | pat[1] == '-' && pat[2] == 'p') { |
| 11313 | if (svmax > 1 && ckWARN(WARN_REDUNDANT)) |
| 11314 | Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s", |
| 11315 | PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); |
| 11316 | argsv = MUTABLE_SV(va_arg(*args, void*)); |
| 11317 | sv_catsv_nomg(sv, argsv); |
| 11318 | return; |
| 11319 | } |
| 11320 | |
| 11321 | #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH) |
| 11322 | /* special-case "%.<number>[gf]" */ |
| 11323 | if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.' |
| 11324 | && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) { |
| 11325 | unsigned digits = 0; |
| 11326 | const char *pp; |
| 11327 | |
| 11328 | pp = pat + 2; |
| 11329 | while (*pp >= '0' && *pp <= '9') |
| 11330 | digits = 10 * digits + (*pp++ - '0'); |
| 11331 | |
| 11332 | /* XXX: Why do this `svix < svmax` test? Couldn't we just |
| 11333 | format the first argument and WARN_REDUNDANT if svmax > 1? |
| 11334 | Munged by Nicholas Clark in v5.13.0-209-g95ea86d */ |
| 11335 | if (pp - pat == (int)patlen - 1 && svix < svmax) { |
| 11336 | const NV nv = SvNV(*svargs); |
| 11337 | if (LIKELY(!Perl_isinfnan(nv))) { |
| 11338 | if (*pp == 'g') { |
| 11339 | /* Add check for digits != 0 because it seems that some |
| 11340 | gconverts are buggy in this case, and we don't yet have |
| 11341 | a Configure test for this. */ |
| 11342 | if (digits && digits < sizeof(ebuf) - NV_DIG - 10) { |
| 11343 | /* 0, point, slack */ |
| 11344 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 11345 | SNPRINTF_G(nv, ebuf, size, digits); |
| 11346 | sv_catpv_nomg(sv, ebuf); |
| 11347 | if (*ebuf) /* May return an empty string for digits==0 */ |
| 11348 | return; |
| 11349 | } |
| 11350 | } else if (!digits) { |
| 11351 | STRLEN l; |
| 11352 | |
| 11353 | if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) { |
| 11354 | sv_catpvn_nomg(sv, p, l); |
| 11355 | return; |
| 11356 | } |
| 11357 | } |
| 11358 | } |
| 11359 | } |
| 11360 | } |
| 11361 | #endif /* !USE_LONG_DOUBLE */ |
| 11362 | |
| 11363 | if (!args && svix < svmax && DO_UTF8(*svargs)) |
| 11364 | has_utf8 = TRUE; |
| 11365 | |
| 11366 | patend = (char*)pat + patlen; |
| 11367 | for (p = (char*)pat; p < patend; p = q) { |
| 11368 | bool alt = FALSE; |
| 11369 | bool left = FALSE; |
| 11370 | bool vectorize = FALSE; |
| 11371 | bool vectorarg = FALSE; |
| 11372 | bool vec_utf8 = FALSE; |
| 11373 | char fill = ' '; |
| 11374 | char plus = 0; |
| 11375 | char intsize = 0; |
| 11376 | STRLEN width = 0; |
| 11377 | STRLEN zeros = 0; |
| 11378 | bool has_precis = FALSE; |
| 11379 | STRLEN precis = 0; |
| 11380 | const I32 osvix = svix; |
| 11381 | bool is_utf8 = FALSE; /* is this item utf8? */ |
| 11382 | bool used_explicit_ix = FALSE; |
| 11383 | bool arg_missing = FALSE; |
| 11384 | #ifdef HAS_LDBL_SPRINTF_BUG |
| 11385 | /* This is to try to fix a bug with irix/nonstop-ux/powerux and |
| 11386 | with sfio - Allen <allens@cpan.org> */ |
| 11387 | bool fix_ldbl_sprintf_bug = FALSE; |
| 11388 | #endif |
| 11389 | |
| 11390 | char esignbuf[4]; |
| 11391 | U8 utf8buf[UTF8_MAXBYTES+1]; |
| 11392 | STRLEN esignlen = 0; |
| 11393 | |
| 11394 | const char *eptr = NULL; |
| 11395 | const char *fmtstart; |
| 11396 | STRLEN elen = 0; |
| 11397 | SV *vecsv = NULL; |
| 11398 | const U8 *vecstr = NULL; |
| 11399 | STRLEN veclen = 0; |
| 11400 | char c = 0; |
| 11401 | int i; |
| 11402 | unsigned base = 0; |
| 11403 | IV iv = 0; |
| 11404 | UV uv = 0; |
| 11405 | /* We need a long double target in case HAS_LONG_DOUBLE, |
| 11406 | * even without USE_LONG_DOUBLE, so that we can printf with |
| 11407 | * long double formats, even without NV being long double. |
| 11408 | * But we call the target 'fv' instead of 'nv', since most of |
| 11409 | * the time it is not (most compilers these days recognize |
| 11410 | * "long double", even if only as a synonym for "double"). |
| 11411 | */ |
| 11412 | #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \ |
| 11413 | defined(PERL_PRIgldbl) && !defined(USE_QUADMATH) |
| 11414 | long double fv; |
| 11415 | # ifdef Perl_isfinitel |
| 11416 | # define FV_ISFINITE(x) Perl_isfinitel(x) |
| 11417 | # endif |
| 11418 | # define FV_GF PERL_PRIgldbl |
| 11419 | # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT) |
| 11420 | /* Work around breakage in OTS$CVT_FLOAT_T_X */ |
| 11421 | # define NV_TO_FV(nv,fv) STMT_START { \ |
| 11422 | double _dv = nv; \ |
| 11423 | fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \ |
| 11424 | } STMT_END |
| 11425 | # else |
| 11426 | # define NV_TO_FV(nv,fv) (fv)=(nv) |
| 11427 | # endif |
| 11428 | #else |
| 11429 | NV fv; |
| 11430 | # define FV_GF NVgf |
| 11431 | # define NV_TO_FV(nv,fv) (fv)=(nv) |
| 11432 | #endif |
| 11433 | #ifndef FV_ISFINITE |
| 11434 | # define FV_ISFINITE(x) Perl_isfinite((NV)(x)) |
| 11435 | #endif |
| 11436 | NV nv; |
| 11437 | STRLEN have; |
| 11438 | STRLEN need; |
| 11439 | STRLEN gap; |
| 11440 | const char *dotstr = "."; |
| 11441 | STRLEN dotstrlen = 1; |
| 11442 | I32 efix = 0; /* explicit format parameter index */ |
| 11443 | I32 ewix = 0; /* explicit width index */ |
| 11444 | I32 epix = 0; /* explicit precision index */ |
| 11445 | I32 evix = 0; /* explicit vector index */ |
| 11446 | bool asterisk = FALSE; |
| 11447 | bool infnan = FALSE; |
| 11448 | |
| 11449 | /* echo everything up to the next format specification */ |
| 11450 | for (q = p; q < patend && *q != '%'; ++q) ; |
| 11451 | if (q > p) { |
| 11452 | if (has_utf8 && !pat_utf8) |
| 11453 | sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv); |
| 11454 | else |
| 11455 | sv_catpvn_nomg(sv, p, q - p); |
| 11456 | p = q; |
| 11457 | } |
| 11458 | if (q++ >= patend) |
| 11459 | break; |
| 11460 | |
| 11461 | fmtstart = q; |
| 11462 | |
| 11463 | /* |
| 11464 | We allow format specification elements in this order: |
| 11465 | \d+\$ explicit format parameter index |
| 11466 | [-+ 0#]+ flags |
| 11467 | v|\*(\d+\$)?v vector with optional (optionally specified) arg |
| 11468 | 0 flag (as above): repeated to allow "v02" |
| 11469 | \d+|\*(\d+\$)? width using optional (optionally specified) arg |
| 11470 | \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg |
| 11471 | [hlqLV] size |
| 11472 | [%bcdefginopsuxDFOUX] format (mandatory) |
| 11473 | */ |
| 11474 | |
| 11475 | if (args) { |
| 11476 | /* |
| 11477 | As of perl5.9.3, printf format checking is on by default. |
| 11478 | Internally, perl uses %p formats to provide an escape to |
| 11479 | some extended formatting. This block deals with those |
| 11480 | extensions: if it does not match, (char*)q is reset and |
| 11481 | the normal format processing code is used. |
| 11482 | |
| 11483 | Currently defined extensions are: |
| 11484 | %p include pointer address (standard) |
| 11485 | %-p (SVf) include an SV (previously %_) |
| 11486 | %-<num>p include an SV with precision <num> |
| 11487 | %2p include a HEK |
| 11488 | %3p include a HEK with precision of 256 |
| 11489 | %4p char* preceded by utf8 flag and length |
| 11490 | %<num>p (where num is 1 or > 4) reserved for future |
| 11491 | extensions |
| 11492 | |
| 11493 | Robin Barker 2005-07-14 (but modified since) |
| 11494 | |
| 11495 | %1p (VDf) removed. RMB 2007-10-19 |
| 11496 | */ |
| 11497 | char* r = q; |
| 11498 | bool sv = FALSE; |
| 11499 | STRLEN n = 0; |
| 11500 | if (*q == '-') |
| 11501 | sv = *q++; |
| 11502 | else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */ |
| 11503 | /* The argument has already gone through cBOOL, so the cast |
| 11504 | is safe. */ |
| 11505 | is_utf8 = (bool)va_arg(*args, int); |
| 11506 | elen = va_arg(*args, UV); |
| 11507 | /* if utf8 length is larger than 0x7ffff..., then it might |
| 11508 | * have been a signed value that wrapped */ |
| 11509 | if (elen > ((~(STRLEN)0) >> 1)) { |
| 11510 | assert(0); /* in DEBUGGING build we want to crash */ |
| 11511 | elen= 0; /* otherwise we want to treat this as an empty string */ |
| 11512 | } |
| 11513 | eptr = va_arg(*args, char *); |
| 11514 | q += sizeof(UTF8f)-1; |
| 11515 | goto string; |
| 11516 | } |
| 11517 | n = expect_number(&q); |
| 11518 | if (*q++ == 'p') { |
| 11519 | if (sv) { /* SVf */ |
| 11520 | if (n) { |
| 11521 | precis = n; |
| 11522 | has_precis = TRUE; |
| 11523 | } |
| 11524 | argsv = MUTABLE_SV(va_arg(*args, void*)); |
| 11525 | eptr = SvPV_const(argsv, elen); |
| 11526 | if (DO_UTF8(argsv)) |
| 11527 | is_utf8 = TRUE; |
| 11528 | goto string; |
| 11529 | } |
| 11530 | else if (n==2 || n==3) { /* HEKf */ |
| 11531 | HEK * const hek = va_arg(*args, HEK *); |
| 11532 | eptr = HEK_KEY(hek); |
| 11533 | elen = HEK_LEN(hek); |
| 11534 | if (HEK_UTF8(hek)) is_utf8 = TRUE; |
| 11535 | if (n==3) precis = 256, has_precis = TRUE; |
| 11536 | goto string; |
| 11537 | } |
| 11538 | else if (n) { |
| 11539 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), |
| 11540 | "internal %%<num>p might conflict with future printf extensions"); |
| 11541 | } |
| 11542 | } |
| 11543 | q = r; |
| 11544 | } |
| 11545 | |
| 11546 | if ( (width = expect_number(&q)) ) { |
| 11547 | if (*q == '$') { |
| 11548 | if (args) |
| 11549 | Perl_croak_nocontext( |
| 11550 | "Cannot yet reorder sv_catpvfn() arguments from va_list"); |
| 11551 | ++q; |
| 11552 | efix = width; |
| 11553 | used_explicit_ix = TRUE; |
| 11554 | } else { |
| 11555 | goto gotwidth; |
| 11556 | } |
| 11557 | } |
| 11558 | |
| 11559 | /* FLAGS */ |
| 11560 | |
| 11561 | while (*q) { |
| 11562 | switch (*q) { |
| 11563 | case ' ': |
| 11564 | case '+': |
| 11565 | if (plus == '+' && *q == ' ') /* '+' over ' ' */ |
| 11566 | q++; |
| 11567 | else |
| 11568 | plus = *q++; |
| 11569 | continue; |
| 11570 | |
| 11571 | case '-': |
| 11572 | left = TRUE; |
| 11573 | q++; |
| 11574 | continue; |
| 11575 | |
| 11576 | case '0': |
| 11577 | fill = *q++; |
| 11578 | continue; |
| 11579 | |
| 11580 | case '#': |
| 11581 | alt = TRUE; |
| 11582 | q++; |
| 11583 | continue; |
| 11584 | |
| 11585 | default: |
| 11586 | break; |
| 11587 | } |
| 11588 | break; |
| 11589 | } |
| 11590 | |
| 11591 | tryasterisk: |
| 11592 | if (*q == '*') { |
| 11593 | q++; |
| 11594 | if ( (ewix = expect_number(&q)) ) { |
| 11595 | if (*q++ == '$') { |
| 11596 | if (args) |
| 11597 | Perl_croak_nocontext( |
| 11598 | "Cannot yet reorder sv_catpvfn() arguments from va_list"); |
| 11599 | used_explicit_ix = TRUE; |
| 11600 | } else |
| 11601 | goto unknown; |
| 11602 | } |
| 11603 | asterisk = TRUE; |
| 11604 | } |
| 11605 | if (*q == 'v') { |
| 11606 | q++; |
| 11607 | if (vectorize) |
| 11608 | goto unknown; |
| 11609 | if ((vectorarg = asterisk)) { |
| 11610 | evix = ewix; |
| 11611 | ewix = 0; |
| 11612 | asterisk = FALSE; |
| 11613 | } |
| 11614 | vectorize = TRUE; |
| 11615 | goto tryasterisk; |
| 11616 | } |
| 11617 | |
| 11618 | if (!asterisk) |
| 11619 | { |
| 11620 | if( *q == '0' ) |
| 11621 | fill = *q++; |
| 11622 | width = expect_number(&q); |
| 11623 | } |
| 11624 | |
| 11625 | if (vectorize && vectorarg) { |
| 11626 | /* vectorizing, but not with the default "." */ |
| 11627 | if (args) |
| 11628 | vecsv = va_arg(*args, SV*); |
| 11629 | else if (evix) { |
| 11630 | FETCH_VCATPVFN_ARGUMENT( |
| 11631 | vecsv, evix > 0 && evix <= svmax, svargs[evix-1]); |
| 11632 | } else { |
| 11633 | FETCH_VCATPVFN_ARGUMENT( |
| 11634 | vecsv, svix < svmax, svargs[svix++]); |
| 11635 | } |
| 11636 | dotstr = SvPV_const(vecsv, dotstrlen); |
| 11637 | /* Keep the DO_UTF8 test *after* the SvPV call, else things go |
| 11638 | bad with tied or overloaded values that return UTF8. */ |
| 11639 | if (DO_UTF8(vecsv)) |
| 11640 | is_utf8 = TRUE; |
| 11641 | else if (has_utf8) { |
| 11642 | vecsv = sv_mortalcopy(vecsv); |
| 11643 | sv_utf8_upgrade(vecsv); |
| 11644 | dotstr = SvPV_const(vecsv, dotstrlen); |
| 11645 | is_utf8 = TRUE; |
| 11646 | } |
| 11647 | } |
| 11648 | |
| 11649 | if (asterisk) { |
| 11650 | if (args) |
| 11651 | i = va_arg(*args, int); |
| 11652 | else |
| 11653 | i = (ewix ? ewix <= svmax : svix < svmax) ? |
| 11654 | SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0; |
| 11655 | left |= (i < 0); |
| 11656 | width = (i < 0) ? -i : i; |
| 11657 | } |
| 11658 | gotwidth: |
| 11659 | |
| 11660 | /* PRECISION */ |
| 11661 | |
| 11662 | if (*q == '.') { |
| 11663 | q++; |
| 11664 | if (*q == '*') { |
| 11665 | q++; |
| 11666 | if ( (epix = expect_number(&q)) ) { |
| 11667 | if (*q++ == '$') { |
| 11668 | if (args) |
| 11669 | Perl_croak_nocontext( |
| 11670 | "Cannot yet reorder sv_catpvfn() arguments from va_list"); |
| 11671 | used_explicit_ix = TRUE; |
| 11672 | } else |
| 11673 | goto unknown; |
| 11674 | } |
| 11675 | if (args) |
| 11676 | i = va_arg(*args, int); |
| 11677 | else { |
| 11678 | SV *precsv; |
| 11679 | if (epix) |
| 11680 | FETCH_VCATPVFN_ARGUMENT( |
| 11681 | precsv, epix > 0 && epix <= svmax, svargs[epix-1]); |
| 11682 | else |
| 11683 | FETCH_VCATPVFN_ARGUMENT( |
| 11684 | precsv, svix < svmax, svargs[svix++]); |
| 11685 | i = precsv == &PL_sv_no ? 0 : SvIVx(precsv); |
| 11686 | } |
| 11687 | precis = i; |
| 11688 | has_precis = !(i < 0); |
| 11689 | } |
| 11690 | else { |
| 11691 | precis = 0; |
| 11692 | while (isDIGIT(*q)) |
| 11693 | precis = precis * 10 + (*q++ - '0'); |
| 11694 | has_precis = TRUE; |
| 11695 | } |
| 11696 | } |
| 11697 | |
| 11698 | if (vectorize) { |
| 11699 | if (args) { |
| 11700 | VECTORIZE_ARGS |
| 11701 | } |
| 11702 | else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) { |
| 11703 | vecsv = svargs[efix ? efix-1 : svix++]; |
| 11704 | vecstr = (U8*)SvPV_const(vecsv,veclen); |
| 11705 | vec_utf8 = DO_UTF8(vecsv); |
| 11706 | |
| 11707 | /* if this is a version object, we need to convert |
| 11708 | * back into v-string notation and then let the |
| 11709 | * vectorize happen normally |
| 11710 | */ |
| 11711 | if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) { |
| 11712 | if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) { |
| 11713 | Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF), |
| 11714 | "vector argument not supported with alpha versions"); |
| 11715 | goto vdblank; |
| 11716 | } |
| 11717 | vecsv = sv_newmortal(); |
| 11718 | scan_vstring((char *)vecstr, (char *)vecstr + veclen, |
| 11719 | vecsv); |
| 11720 | vecstr = (U8*)SvPV_const(vecsv, veclen); |
| 11721 | vec_utf8 = DO_UTF8(vecsv); |
| 11722 | } |
| 11723 | } |
| 11724 | else { |
| 11725 | vdblank: |
| 11726 | vecstr = (U8*)""; |
| 11727 | veclen = 0; |
| 11728 | } |
| 11729 | } |
| 11730 | |
| 11731 | /* SIZE */ |
| 11732 | |
| 11733 | switch (*q) { |
| 11734 | #ifdef WIN32 |
| 11735 | case 'I': /* Ix, I32x, and I64x */ |
| 11736 | # ifdef USE_64_BIT_INT |
| 11737 | if (q[1] == '6' && q[2] == '4') { |
| 11738 | q += 3; |
| 11739 | intsize = 'q'; |
| 11740 | break; |
| 11741 | } |
| 11742 | # endif |
| 11743 | if (q[1] == '3' && q[2] == '2') { |
| 11744 | q += 3; |
| 11745 | break; |
| 11746 | } |
| 11747 | # ifdef USE_64_BIT_INT |
| 11748 | intsize = 'q'; |
| 11749 | # endif |
| 11750 | q++; |
| 11751 | break; |
| 11752 | #endif |
| 11753 | #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \ |
| 11754 | (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE)) |
| 11755 | case 'L': /* Ld */ |
| 11756 | /* FALLTHROUGH */ |
| 11757 | # ifdef USE_QUADMATH |
| 11758 | case 'Q': |
| 11759 | /* FALLTHROUGH */ |
| 11760 | # endif |
| 11761 | # if IVSIZE >= 8 |
| 11762 | case 'q': /* qd */ |
| 11763 | # endif |
| 11764 | intsize = 'q'; |
| 11765 | q++; |
| 11766 | break; |
| 11767 | #endif |
| 11768 | case 'l': |
| 11769 | ++q; |
| 11770 | #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \ |
| 11771 | (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE)) |
| 11772 | if (*q == 'l') { /* lld, llf */ |
| 11773 | intsize = 'q'; |
| 11774 | ++q; |
| 11775 | } |
| 11776 | else |
| 11777 | #endif |
| 11778 | intsize = 'l'; |
| 11779 | break; |
| 11780 | case 'h': |
| 11781 | if (*++q == 'h') { /* hhd, hhu */ |
| 11782 | intsize = 'c'; |
| 11783 | ++q; |
| 11784 | } |
| 11785 | else |
| 11786 | intsize = 'h'; |
| 11787 | break; |
| 11788 | case 'V': |
| 11789 | case 'z': |
| 11790 | case 't': |
| 11791 | #ifdef I_STDINT |
| 11792 | case 'j': |
| 11793 | #endif |
| 11794 | intsize = *q++; |
| 11795 | break; |
| 11796 | } |
| 11797 | |
| 11798 | /* CONVERSION */ |
| 11799 | |
| 11800 | if (*q == '%') { |
| 11801 | eptr = q++; |
| 11802 | elen = 1; |
| 11803 | if (vectorize) { |
| 11804 | c = '%'; |
| 11805 | goto unknown; |
| 11806 | } |
| 11807 | goto string; |
| 11808 | } |
| 11809 | |
| 11810 | if (!vectorize && !args) { |
| 11811 | if (efix) { |
| 11812 | const I32 i = efix-1; |
| 11813 | FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]); |
| 11814 | } else { |
| 11815 | FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax, |
| 11816 | svargs[svix++]); |
| 11817 | } |
| 11818 | } |
| 11819 | |
| 11820 | if (argsv && strchr("BbcDdiOopuUXx",*q)) { |
| 11821 | /* XXX va_arg(*args) case? need peek, use va_copy? */ |
| 11822 | SvGETMAGIC(argsv); |
| 11823 | if (UNLIKELY(SvAMAGIC(argsv))) |
| 11824 | argsv = sv_2num(argsv); |
| 11825 | infnan = UNLIKELY(isinfnansv(argsv)); |
| 11826 | } |
| 11827 | |
| 11828 | switch (c = *q++) { |
| 11829 | |
| 11830 | /* STRINGS */ |
| 11831 | |
| 11832 | case 'c': |
| 11833 | if (vectorize) |
| 11834 | goto unknown; |
| 11835 | if (infnan) |
| 11836 | Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'", |
| 11837 | /* no va_arg() case */ |
| 11838 | SvNV_nomg(argsv), (int)c); |
| 11839 | uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv); |
| 11840 | if ((uv > 255 || |
| 11841 | (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv))) |
| 11842 | && !IN_BYTES) { |
| 11843 | eptr = (char*)utf8buf; |
| 11844 | elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf; |
| 11845 | is_utf8 = TRUE; |
| 11846 | } |
| 11847 | else { |
| 11848 | c = (char)uv; |
| 11849 | eptr = &c; |
| 11850 | elen = 1; |
| 11851 | } |
| 11852 | goto string; |
| 11853 | |
| 11854 | case 's': |
| 11855 | if (vectorize) |
| 11856 | goto unknown; |
| 11857 | if (args) { |
| 11858 | eptr = va_arg(*args, char*); |
| 11859 | if (eptr) |
| 11860 | elen = strlen(eptr); |
| 11861 | else { |
| 11862 | eptr = (char *)nullstr; |
| 11863 | elen = sizeof nullstr - 1; |
| 11864 | } |
| 11865 | } |
| 11866 | else { |
| 11867 | eptr = SvPV_const(argsv, elen); |
| 11868 | if (DO_UTF8(argsv)) { |
| 11869 | STRLEN old_precis = precis; |
| 11870 | if (has_precis && precis < elen) { |
| 11871 | STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen); |
| 11872 | STRLEN p = precis > ulen ? ulen : precis; |
| 11873 | precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0); |
| 11874 | /* sticks at end */ |
| 11875 | } |
| 11876 | if (width) { /* fudge width (can't fudge elen) */ |
| 11877 | if (has_precis && precis < elen) |
| 11878 | width += precis - old_precis; |
| 11879 | else |
| 11880 | width += |
| 11881 | elen - sv_or_pv_len_utf8(argsv,eptr,elen); |
| 11882 | } |
| 11883 | is_utf8 = TRUE; |
| 11884 | } |
| 11885 | } |
| 11886 | |
| 11887 | string: |
| 11888 | if (has_precis && precis < elen) |
| 11889 | elen = precis; |
| 11890 | break; |
| 11891 | |
| 11892 | /* INTEGERS */ |
| 11893 | |
| 11894 | case 'p': |
| 11895 | if (infnan) { |
| 11896 | goto floating_point; |
| 11897 | } |
| 11898 | if (alt || vectorize) |
| 11899 | goto unknown; |
| 11900 | uv = PTR2UV(args ? va_arg(*args, void*) : argsv); |
| 11901 | base = 16; |
| 11902 | goto integer; |
| 11903 | |
| 11904 | case 'D': |
| 11905 | #ifdef IV_IS_QUAD |
| 11906 | intsize = 'q'; |
| 11907 | #else |
| 11908 | intsize = 'l'; |
| 11909 | #endif |
| 11910 | /* FALLTHROUGH */ |
| 11911 | case 'd': |
| 11912 | case 'i': |
| 11913 | if (infnan) { |
| 11914 | goto floating_point; |
| 11915 | } |
| 11916 | if (vectorize) { |
| 11917 | STRLEN ulen; |
| 11918 | if (!veclen) |
| 11919 | goto donevalidconversion; |
| 11920 | if (vec_utf8) |
| 11921 | uv = utf8n_to_uvchr(vecstr, veclen, &ulen, |
| 11922 | UTF8_ALLOW_ANYUV); |
| 11923 | else { |
| 11924 | uv = *vecstr; |
| 11925 | ulen = 1; |
| 11926 | } |
| 11927 | vecstr += ulen; |
| 11928 | veclen -= ulen; |
| 11929 | if (plus) |
| 11930 | esignbuf[esignlen++] = plus; |
| 11931 | } |
| 11932 | else if (args) { |
| 11933 | switch (intsize) { |
| 11934 | case 'c': iv = (char)va_arg(*args, int); break; |
| 11935 | case 'h': iv = (short)va_arg(*args, int); break; |
| 11936 | case 'l': iv = va_arg(*args, long); break; |
| 11937 | case 'V': iv = va_arg(*args, IV); break; |
| 11938 | case 'z': iv = va_arg(*args, SSize_t); break; |
| 11939 | #ifdef HAS_PTRDIFF_T |
| 11940 | case 't': iv = va_arg(*args, ptrdiff_t); break; |
| 11941 | #endif |
| 11942 | default: iv = va_arg(*args, int); break; |
| 11943 | #ifdef I_STDINT |
| 11944 | case 'j': iv = va_arg(*args, intmax_t); break; |
| 11945 | #endif |
| 11946 | case 'q': |
| 11947 | #if IVSIZE >= 8 |
| 11948 | iv = va_arg(*args, Quad_t); break; |
| 11949 | #else |
| 11950 | goto unknown; |
| 11951 | #endif |
| 11952 | } |
| 11953 | } |
| 11954 | else { |
| 11955 | IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */ |
| 11956 | switch (intsize) { |
| 11957 | case 'c': iv = (char)tiv; break; |
| 11958 | case 'h': iv = (short)tiv; break; |
| 11959 | case 'l': iv = (long)tiv; break; |
| 11960 | case 'V': |
| 11961 | default: iv = tiv; break; |
| 11962 | case 'q': |
| 11963 | #if IVSIZE >= 8 |
| 11964 | iv = (Quad_t)tiv; break; |
| 11965 | #else |
| 11966 | goto unknown; |
| 11967 | #endif |
| 11968 | } |
| 11969 | } |
| 11970 | if ( !vectorize ) /* we already set uv above */ |
| 11971 | { |
| 11972 | if (iv >= 0) { |
| 11973 | uv = iv; |
| 11974 | if (plus) |
| 11975 | esignbuf[esignlen++] = plus; |
| 11976 | } |
| 11977 | else { |
| 11978 | uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv); |
| 11979 | esignbuf[esignlen++] = '-'; |
| 11980 | } |
| 11981 | } |
| 11982 | base = 10; |
| 11983 | goto integer; |
| 11984 | |
| 11985 | case 'U': |
| 11986 | #ifdef IV_IS_QUAD |
| 11987 | intsize = 'q'; |
| 11988 | #else |
| 11989 | intsize = 'l'; |
| 11990 | #endif |
| 11991 | /* FALLTHROUGH */ |
| 11992 | case 'u': |
| 11993 | base = 10; |
| 11994 | goto uns_integer; |
| 11995 | |
| 11996 | case 'B': |
| 11997 | case 'b': |
| 11998 | base = 2; |
| 11999 | goto uns_integer; |
| 12000 | |
| 12001 | case 'O': |
| 12002 | #ifdef IV_IS_QUAD |
| 12003 | intsize = 'q'; |
| 12004 | #else |
| 12005 | intsize = 'l'; |
| 12006 | #endif |
| 12007 | /* FALLTHROUGH */ |
| 12008 | case 'o': |
| 12009 | base = 8; |
| 12010 | goto uns_integer; |
| 12011 | |
| 12012 | case 'X': |
| 12013 | case 'x': |
| 12014 | base = 16; |
| 12015 | |
| 12016 | uns_integer: |
| 12017 | if (infnan) { |
| 12018 | goto floating_point; |
| 12019 | } |
| 12020 | if (vectorize) { |
| 12021 | STRLEN ulen; |
| 12022 | vector: |
| 12023 | if (!veclen) |
| 12024 | goto donevalidconversion; |
| 12025 | if (vec_utf8) |
| 12026 | uv = utf8n_to_uvchr(vecstr, veclen, &ulen, |
| 12027 | UTF8_ALLOW_ANYUV); |
| 12028 | else { |
| 12029 | uv = *vecstr; |
| 12030 | ulen = 1; |
| 12031 | } |
| 12032 | vecstr += ulen; |
| 12033 | veclen -= ulen; |
| 12034 | } |
| 12035 | else if (args) { |
| 12036 | switch (intsize) { |
| 12037 | case 'c': uv = (unsigned char)va_arg(*args, unsigned); break; |
| 12038 | case 'h': uv = (unsigned short)va_arg(*args, unsigned); break; |
| 12039 | case 'l': uv = va_arg(*args, unsigned long); break; |
| 12040 | case 'V': uv = va_arg(*args, UV); break; |
| 12041 | case 'z': uv = va_arg(*args, Size_t); break; |
| 12042 | #ifdef HAS_PTRDIFF_T |
| 12043 | case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */ |
| 12044 | #endif |
| 12045 | #ifdef I_STDINT |
| 12046 | case 'j': uv = va_arg(*args, uintmax_t); break; |
| 12047 | #endif |
| 12048 | default: uv = va_arg(*args, unsigned); break; |
| 12049 | case 'q': |
| 12050 | #if IVSIZE >= 8 |
| 12051 | uv = va_arg(*args, Uquad_t); break; |
| 12052 | #else |
| 12053 | goto unknown; |
| 12054 | #endif |
| 12055 | } |
| 12056 | } |
| 12057 | else { |
| 12058 | UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */ |
| 12059 | switch (intsize) { |
| 12060 | case 'c': uv = (unsigned char)tuv; break; |
| 12061 | case 'h': uv = (unsigned short)tuv; break; |
| 12062 | case 'l': uv = (unsigned long)tuv; break; |
| 12063 | case 'V': |
| 12064 | default: uv = tuv; break; |
| 12065 | case 'q': |
| 12066 | #if IVSIZE >= 8 |
| 12067 | uv = (Uquad_t)tuv; break; |
| 12068 | #else |
| 12069 | goto unknown; |
| 12070 | #endif |
| 12071 | } |
| 12072 | } |
| 12073 | |
| 12074 | integer: |
| 12075 | { |
| 12076 | char *ptr = ebuf + sizeof ebuf; |
| 12077 | bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */ |
| 12078 | unsigned dig; |
| 12079 | zeros = 0; |
| 12080 | |
| 12081 | switch (base) { |
| 12082 | case 16: |
| 12083 | p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit); |
| 12084 | do { |
| 12085 | dig = uv & 15; |
| 12086 | *--ptr = p[dig]; |
| 12087 | } while (uv >>= 4); |
| 12088 | if (tempalt) { |
| 12089 | esignbuf[esignlen++] = '0'; |
| 12090 | esignbuf[esignlen++] = c; /* 'x' or 'X' */ |
| 12091 | } |
| 12092 | break; |
| 12093 | case 8: |
| 12094 | do { |
| 12095 | dig = uv & 7; |
| 12096 | *--ptr = '0' + dig; |
| 12097 | } while (uv >>= 3); |
| 12098 | if (alt && *ptr != '0') |
| 12099 | *--ptr = '0'; |
| 12100 | break; |
| 12101 | case 2: |
| 12102 | do { |
| 12103 | dig = uv & 1; |
| 12104 | *--ptr = '0' + dig; |
| 12105 | } while (uv >>= 1); |
| 12106 | if (tempalt) { |
| 12107 | esignbuf[esignlen++] = '0'; |
| 12108 | esignbuf[esignlen++] = c; |
| 12109 | } |
| 12110 | break; |
| 12111 | default: /* it had better be ten or less */ |
| 12112 | do { |
| 12113 | dig = uv % base; |
| 12114 | *--ptr = '0' + dig; |
| 12115 | } while (uv /= base); |
| 12116 | break; |
| 12117 | } |
| 12118 | elen = (ebuf + sizeof ebuf) - ptr; |
| 12119 | eptr = ptr; |
| 12120 | if (has_precis) { |
| 12121 | if (precis > elen) |
| 12122 | zeros = precis - elen; |
| 12123 | else if (precis == 0 && elen == 1 && *eptr == '0' |
| 12124 | && !(base == 8 && alt)) /* "%#.0o" prints "0" */ |
| 12125 | elen = 0; |
| 12126 | |
| 12127 | /* a precision nullifies the 0 flag. */ |
| 12128 | if (fill == '0') |
| 12129 | fill = ' '; |
| 12130 | } |
| 12131 | } |
| 12132 | break; |
| 12133 | |
| 12134 | /* FLOATING POINT */ |
| 12135 | |
| 12136 | floating_point: |
| 12137 | |
| 12138 | case 'F': |
| 12139 | c = 'f'; /* maybe %F isn't supported here */ |
| 12140 | /* FALLTHROUGH */ |
| 12141 | case 'e': case 'E': |
| 12142 | case 'f': |
| 12143 | case 'g': case 'G': |
| 12144 | case 'a': case 'A': |
| 12145 | if (vectorize) |
| 12146 | goto unknown; |
| 12147 | |
| 12148 | /* This is evil, but floating point is even more evil */ |
| 12149 | |
| 12150 | /* for SV-style calling, we can only get NV |
| 12151 | for C-style calling, we assume %f is double; |
| 12152 | for simplicity we allow any of %Lf, %llf, %qf for long double |
| 12153 | */ |
| 12154 | switch (intsize) { |
| 12155 | case 'V': |
| 12156 | #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH) |
| 12157 | intsize = 'q'; |
| 12158 | #endif |
| 12159 | break; |
| 12160 | /* [perl #20339] - we should accept and ignore %lf rather than die */ |
| 12161 | case 'l': |
| 12162 | /* FALLTHROUGH */ |
| 12163 | default: |
| 12164 | #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH) |
| 12165 | intsize = args ? 0 : 'q'; |
| 12166 | #endif |
| 12167 | break; |
| 12168 | case 'q': |
| 12169 | #if defined(HAS_LONG_DOUBLE) |
| 12170 | break; |
| 12171 | #else |
| 12172 | /* FALLTHROUGH */ |
| 12173 | #endif |
| 12174 | case 'c': |
| 12175 | case 'h': |
| 12176 | case 'z': |
| 12177 | case 't': |
| 12178 | case 'j': |
| 12179 | goto unknown; |
| 12180 | } |
| 12181 | |
| 12182 | /* Now we need (long double) if intsize == 'q', else (double). */ |
| 12183 | if (args) { |
| 12184 | /* Note: do not pull NVs off the va_list with va_arg() |
| 12185 | * (pull doubles instead) because if you have a build |
| 12186 | * with long doubles, you would always be pulling long |
| 12187 | * doubles, which would badly break anyone using only |
| 12188 | * doubles (i.e. the majority of builds). In other |
| 12189 | * words, you cannot mix doubles and long doubles. |
| 12190 | * The only case where you can pull off long doubles |
| 12191 | * is when the format specifier explicitly asks so with |
| 12192 | * e.g. "%Lg". */ |
| 12193 | #ifdef USE_QUADMATH |
| 12194 | fv = intsize == 'q' ? |
| 12195 | va_arg(*args, NV) : va_arg(*args, double); |
| 12196 | nv = fv; |
| 12197 | #elif LONG_DOUBLESIZE > DOUBLESIZE |
| 12198 | if (intsize == 'q') { |
| 12199 | fv = va_arg(*args, long double); |
| 12200 | nv = fv; |
| 12201 | } else { |
| 12202 | nv = va_arg(*args, double); |
| 12203 | NV_TO_FV(nv, fv); |
| 12204 | } |
| 12205 | #else |
| 12206 | nv = va_arg(*args, double); |
| 12207 | fv = nv; |
| 12208 | #endif |
| 12209 | } |
| 12210 | else |
| 12211 | { |
| 12212 | if (!infnan) SvGETMAGIC(argsv); |
| 12213 | nv = SvNV_nomg(argsv); |
| 12214 | NV_TO_FV(nv, fv); |
| 12215 | } |
| 12216 | |
| 12217 | need = 0; |
| 12218 | /* frexp() (or frexpl) has some unspecified behaviour for |
| 12219 | * nan/inf/-inf, so let's avoid calling that on non-finites. */ |
| 12220 | if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) { |
| 12221 | i = PERL_INT_MIN; |
| 12222 | (void)Perl_frexp((NV)fv, &i); |
| 12223 | if (i == PERL_INT_MIN) |
| 12224 | Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv); |
| 12225 | /* Do not set hexfp earlier since we want to printf |
| 12226 | * Inf/NaN for Inf/NaN, not their hexfp. */ |
| 12227 | hexfp = isALPHA_FOLD_EQ(c, 'a'); |
| 12228 | if (UNLIKELY(hexfp)) { |
| 12229 | /* This seriously overshoots in most cases, but |
| 12230 | * better the undershooting. Firstly, all bytes |
| 12231 | * of the NV are not mantissa, some of them are |
| 12232 | * exponent. Secondly, for the reasonably common |
| 12233 | * long doubles case, the "80-bit extended", two |
| 12234 | * or six bytes of the NV are unused. */ |
| 12235 | need += |
| 12236 | (fv < 0) ? 1 : 0 + /* possible unary minus */ |
| 12237 | 2 + /* "0x" */ |
| 12238 | 1 + /* the very unlikely carry */ |
| 12239 | 1 + /* "1" */ |
| 12240 | 1 + /* "." */ |
| 12241 | 2 * NVSIZE + /* 2 hexdigits for each byte */ |
| 12242 | 2 + /* "p+" */ |
| 12243 | 6 + /* exponent: sign, plus up to 16383 (quad fp) */ |
| 12244 | 1; /* \0 */ |
| 12245 | #ifdef LONGDOUBLE_DOUBLEDOUBLE |
| 12246 | /* However, for the "double double", we need more. |
| 12247 | * Since each double has their own exponent, the |
| 12248 | * doubles may float (haha) rather far from each |
| 12249 | * other, and the number of required bits is much |
| 12250 | * larger, up to total of DOUBLEDOUBLE_MAXBITS bits. |
| 12251 | * See the definition of DOUBLEDOUBLE_MAXBITS. |
| 12252 | * |
| 12253 | * Need 2 hexdigits for each byte. */ |
| 12254 | need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2; |
| 12255 | /* the size for the exponent already added */ |
| 12256 | #endif |
| 12257 | #ifdef USE_LOCALE_NUMERIC |
| 12258 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 12259 | if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) |
| 12260 | need += SvLEN(PL_numeric_radix_sv); |
| 12261 | RESTORE_LC_NUMERIC(); |
| 12262 | #endif |
| 12263 | } |
| 12264 | else if (i > 0) { |
| 12265 | need = BIT_DIGITS(i); |
| 12266 | } /* if i < 0, the number of digits is hard to predict. */ |
| 12267 | } |
| 12268 | need += has_precis ? precis : 6; /* known default */ |
| 12269 | |
| 12270 | if (need < width) |
| 12271 | need = width; |
| 12272 | |
| 12273 | #ifdef HAS_LDBL_SPRINTF_BUG |
| 12274 | /* This is to try to fix a bug with irix/nonstop-ux/powerux and |
| 12275 | with sfio - Allen <allens@cpan.org> */ |
| 12276 | |
| 12277 | # ifdef DBL_MAX |
| 12278 | # define MY_DBL_MAX DBL_MAX |
| 12279 | # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */ |
| 12280 | # if DOUBLESIZE >= 8 |
| 12281 | # define MY_DBL_MAX 1.7976931348623157E+308L |
| 12282 | # else |
| 12283 | # define MY_DBL_MAX 3.40282347E+38L |
| 12284 | # endif |
| 12285 | # endif |
| 12286 | |
| 12287 | # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */ |
| 12288 | # define MY_DBL_MAX_BUG 1L |
| 12289 | # else |
| 12290 | # define MY_DBL_MAX_BUG MY_DBL_MAX |
| 12291 | # endif |
| 12292 | |
| 12293 | # ifdef DBL_MIN |
| 12294 | # define MY_DBL_MIN DBL_MIN |
| 12295 | # else /* XXX guessing! -Allen */ |
| 12296 | # if DOUBLESIZE >= 8 |
| 12297 | # define MY_DBL_MIN 2.2250738585072014E-308L |
| 12298 | # else |
| 12299 | # define MY_DBL_MIN 1.17549435E-38L |
| 12300 | # endif |
| 12301 | # endif |
| 12302 | |
| 12303 | if ((intsize == 'q') && (c == 'f') && |
| 12304 | ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) && |
| 12305 | (need < DBL_DIG)) { |
| 12306 | /* it's going to be short enough that |
| 12307 | * long double precision is not needed */ |
| 12308 | |
| 12309 | if ((fv <= 0L) && (fv >= -0L)) |
| 12310 | fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */ |
| 12311 | else { |
| 12312 | /* would use Perl_fp_class as a double-check but not |
| 12313 | * functional on IRIX - see perl.h comments */ |
| 12314 | |
| 12315 | if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) { |
| 12316 | /* It's within the range that a double can represent */ |
| 12317 | #if defined(DBL_MAX) && !defined(DBL_MIN) |
| 12318 | if ((fv >= ((long double)1/DBL_MAX)) || |
| 12319 | (fv <= (-(long double)1/DBL_MAX))) |
| 12320 | #endif |
| 12321 | fix_ldbl_sprintf_bug = TRUE; |
| 12322 | } |
| 12323 | } |
| 12324 | if (fix_ldbl_sprintf_bug == TRUE) { |
| 12325 | double temp; |
| 12326 | |
| 12327 | intsize = 0; |
| 12328 | temp = (double)fv; |
| 12329 | fv = (NV)temp; |
| 12330 | } |
| 12331 | } |
| 12332 | |
| 12333 | # undef MY_DBL_MAX |
| 12334 | # undef MY_DBL_MAX_BUG |
| 12335 | # undef MY_DBL_MIN |
| 12336 | |
| 12337 | #endif /* HAS_LDBL_SPRINTF_BUG */ |
| 12338 | |
| 12339 | need += 20; /* fudge factor */ |
| 12340 | if (PL_efloatsize < need) { |
| 12341 | Safefree(PL_efloatbuf); |
| 12342 | PL_efloatsize = need + 20; /* more fudge */ |
| 12343 | Newx(PL_efloatbuf, PL_efloatsize, char); |
| 12344 | PL_efloatbuf[0] = '\0'; |
| 12345 | } |
| 12346 | |
| 12347 | if ( !(width || left || plus || alt) && fill != '0' |
| 12348 | && has_precis && intsize != 'q' /* Shortcuts */ |
| 12349 | && LIKELY(!Perl_isinfnan((NV)fv)) ) { |
| 12350 | /* See earlier comment about buggy Gconvert when digits, |
| 12351 | aka precis is 0 */ |
| 12352 | if ( c == 'g' && precis ) { |
| 12353 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 12354 | SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis); |
| 12355 | /* May return an empty string for digits==0 */ |
| 12356 | if (*PL_efloatbuf) { |
| 12357 | elen = strlen(PL_efloatbuf); |
| 12358 | goto float_converted; |
| 12359 | } |
| 12360 | } else if ( c == 'f' && !precis ) { |
| 12361 | if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen))) |
| 12362 | break; |
| 12363 | } |
| 12364 | } |
| 12365 | |
| 12366 | if (UNLIKELY(hexfp)) { |
| 12367 | /* Hexadecimal floating point. */ |
| 12368 | char* p = PL_efloatbuf; |
| 12369 | U8 vhex[VHEX_SIZE]; |
| 12370 | U8* v = vhex; /* working pointer to vhex */ |
| 12371 | U8* vend; /* pointer to one beyond last digit of vhex */ |
| 12372 | U8* vfnz = NULL; /* first non-zero */ |
| 12373 | U8* vlnz = NULL; /* last non-zero */ |
| 12374 | const bool lower = (c == 'a'); |
| 12375 | /* At output the values of vhex (up to vend) will |
| 12376 | * be mapped through the xdig to get the actual |
| 12377 | * human-readable xdigits. */ |
| 12378 | const char* xdig = PL_hexdigit; |
| 12379 | int zerotail = 0; /* how many extra zeros to append */ |
| 12380 | int exponent = 0; /* exponent of the floating point input */ |
| 12381 | bool hexradix = FALSE; /* should we output the radix */ |
| 12382 | |
| 12383 | /* XXX: denormals, NaN, Inf. |
| 12384 | * |
| 12385 | * For example with denormals, (assuming the vanilla |
| 12386 | * 64-bit double): the exponent is zero. 1xp-1074 is |
| 12387 | * the smallest denormal and the smallest double, it |
| 12388 | * should be output as 0x0.0000000000001p-1022 to |
| 12389 | * match its internal structure. */ |
| 12390 | |
| 12391 | vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL); |
| 12392 | S_hextract(aTHX_ nv, &exponent, vhex, vend); |
| 12393 | |
| 12394 | #if NVSIZE > DOUBLESIZE |
| 12395 | # ifdef HEXTRACT_HAS_IMPLICIT_BIT |
| 12396 | /* In this case there is an implicit bit, |
| 12397 | * and therefore the exponent is shifted shift by one. */ |
| 12398 | exponent--; |
| 12399 | # else |
| 12400 | /* In this case there is no implicit bit, |
| 12401 | * and the exponent is shifted by the first xdigit. */ |
| 12402 | exponent -= 4; |
| 12403 | # endif |
| 12404 | #endif |
| 12405 | |
| 12406 | if (fv < 0 |
| 12407 | || Perl_signbit(nv) |
| 12408 | ) |
| 12409 | *p++ = '-'; |
| 12410 | else if (plus) |
| 12411 | *p++ = plus; |
| 12412 | *p++ = '0'; |
| 12413 | if (lower) { |
| 12414 | *p++ = 'x'; |
| 12415 | } |
| 12416 | else { |
| 12417 | *p++ = 'X'; |
| 12418 | xdig += 16; /* Use uppercase hex. */ |
| 12419 | } |
| 12420 | |
| 12421 | /* Find the first non-zero xdigit. */ |
| 12422 | for (v = vhex; v < vend; v++) { |
| 12423 | if (*v) { |
| 12424 | vfnz = v; |
| 12425 | break; |
| 12426 | } |
| 12427 | } |
| 12428 | |
| 12429 | if (vfnz) { |
| 12430 | /* Find the last non-zero xdigit. */ |
| 12431 | for (v = vend - 1; v >= vhex; v--) { |
| 12432 | if (*v) { |
| 12433 | vlnz = v; |
| 12434 | break; |
| 12435 | } |
| 12436 | } |
| 12437 | |
| 12438 | #if NVSIZE == DOUBLESIZE |
| 12439 | if (fv != 0.0) |
| 12440 | exponent--; |
| 12441 | #endif |
| 12442 | |
| 12443 | if (precis > 0) { |
| 12444 | if ((SSize_t)(precis + 1) < vend - vhex) { |
| 12445 | bool round; |
| 12446 | |
| 12447 | v = vhex + precis + 1; |
| 12448 | /* Round away from zero: if the tail |
| 12449 | * beyond the precis xdigits is equal to |
| 12450 | * or greater than 0x8000... */ |
| 12451 | round = *v > 0x8; |
| 12452 | if (!round && *v == 0x8) { |
| 12453 | for (v++; v < vend; v++) { |
| 12454 | if (*v) { |
| 12455 | round = TRUE; |
| 12456 | break; |
| 12457 | } |
| 12458 | } |
| 12459 | } |
| 12460 | if (round) { |
| 12461 | for (v = vhex + precis; v >= vhex; v--) { |
| 12462 | if (*v < 0xF) { |
| 12463 | (*v)++; |
| 12464 | break; |
| 12465 | } |
| 12466 | *v = 0; |
| 12467 | if (v == vhex) { |
| 12468 | /* If the carry goes all the way to |
| 12469 | * the front, we need to output |
| 12470 | * a single '1'. This goes against |
| 12471 | * the "xdigit and then radix" |
| 12472 | * but since this is "cannot happen" |
| 12473 | * category, that is probably good. */ |
| 12474 | *p++ = xdig[1]; |
| 12475 | } |
| 12476 | } |
| 12477 | } |
| 12478 | /* The new effective "last non zero". */ |
| 12479 | vlnz = vhex + precis; |
| 12480 | } |
| 12481 | else { |
| 12482 | zerotail = precis - (vlnz - vhex); |
| 12483 | } |
| 12484 | } |
| 12485 | |
| 12486 | v = vhex; |
| 12487 | *p++ = xdig[*v++]; |
| 12488 | |
| 12489 | /* If there are non-zero xdigits, the radix |
| 12490 | * is output after the first one. */ |
| 12491 | if (vfnz < vlnz) { |
| 12492 | hexradix = TRUE; |
| 12493 | } |
| 12494 | } |
| 12495 | else { |
| 12496 | *p++ = '0'; |
| 12497 | exponent = 0; |
| 12498 | zerotail = precis; |
| 12499 | } |
| 12500 | |
| 12501 | /* The radix is always output if precis, or if alt. */ |
| 12502 | if (precis > 0 || alt) { |
| 12503 | hexradix = TRUE; |
| 12504 | } |
| 12505 | |
| 12506 | if (hexradix) { |
| 12507 | #ifndef USE_LOCALE_NUMERIC |
| 12508 | *p++ = '.'; |
| 12509 | #else |
| 12510 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 12511 | if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) { |
| 12512 | STRLEN n; |
| 12513 | const char* r = SvPV(PL_numeric_radix_sv, n); |
| 12514 | Copy(r, p, n, char); |
| 12515 | p += n; |
| 12516 | } |
| 12517 | else { |
| 12518 | *p++ = '.'; |
| 12519 | } |
| 12520 | RESTORE_LC_NUMERIC(); |
| 12521 | #endif |
| 12522 | } |
| 12523 | |
| 12524 | if (vlnz) { |
| 12525 | while (v <= vlnz) |
| 12526 | *p++ = xdig[*v++]; |
| 12527 | } |
| 12528 | |
| 12529 | if (zerotail > 0) { |
| 12530 | while (zerotail--) { |
| 12531 | *p++ = '0'; |
| 12532 | } |
| 12533 | } |
| 12534 | |
| 12535 | elen = p - PL_efloatbuf; |
| 12536 | elen += my_snprintf(p, PL_efloatsize - elen, |
| 12537 | "%c%+d", lower ? 'p' : 'P', |
| 12538 | exponent); |
| 12539 | |
| 12540 | if (elen < width) { |
| 12541 | if (left) { |
| 12542 | /* Pad the back with spaces. */ |
| 12543 | memset(PL_efloatbuf + elen, ' ', width - elen); |
| 12544 | } |
| 12545 | else if (fill == '0') { |
| 12546 | /* Insert the zeros between the "0x" and |
| 12547 | * the digits, otherwise we end up with |
| 12548 | * "0000xHHH..." */ |
| 12549 | STRLEN nzero = width - elen; |
| 12550 | char* zerox = PL_efloatbuf + 2; |
| 12551 | Move(zerox, zerox + nzero, elen - 2, char); |
| 12552 | memset(zerox, fill, nzero); |
| 12553 | } |
| 12554 | else { |
| 12555 | /* Move it to the right. */ |
| 12556 | Move(PL_efloatbuf, PL_efloatbuf + width - elen, |
| 12557 | elen, char); |
| 12558 | /* Pad the front with spaces. */ |
| 12559 | memset(PL_efloatbuf, ' ', width - elen); |
| 12560 | } |
| 12561 | elen = width; |
| 12562 | } |
| 12563 | } |
| 12564 | else { |
| 12565 | elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus); |
| 12566 | if (elen) { |
| 12567 | /* Not affecting infnan output: precision, alt, fill. */ |
| 12568 | if (elen < width) { |
| 12569 | if (left) { |
| 12570 | /* Pack the back with spaces. */ |
| 12571 | memset(PL_efloatbuf + elen, ' ', width - elen); |
| 12572 | } else { |
| 12573 | /* Move it to the right. */ |
| 12574 | Move(PL_efloatbuf, PL_efloatbuf + width - elen, |
| 12575 | elen, char); |
| 12576 | /* Pad the front with spaces. */ |
| 12577 | memset(PL_efloatbuf, ' ', width - elen); |
| 12578 | } |
| 12579 | elen = width; |
| 12580 | } |
| 12581 | } |
| 12582 | } |
| 12583 | |
| 12584 | if (elen == 0) { |
| 12585 | char *ptr = ebuf + sizeof ebuf; |
| 12586 | *--ptr = '\0'; |
| 12587 | *--ptr = c; |
| 12588 | #if defined(USE_QUADMATH) |
| 12589 | if (intsize == 'q') { |
| 12590 | /* "g" -> "Qg" */ |
| 12591 | *--ptr = 'Q'; |
| 12592 | } |
| 12593 | /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */ |
| 12594 | #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl) |
| 12595 | /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl, |
| 12596 | * not USE_LONG_DOUBLE and NVff. In other words, |
| 12597 | * this needs to work without USE_LONG_DOUBLE. */ |
| 12598 | if (intsize == 'q') { |
| 12599 | /* Copy the one or more characters in a long double |
| 12600 | * format before the 'base' ([efgEFG]) character to |
| 12601 | * the format string. */ |
| 12602 | static char const ldblf[] = PERL_PRIfldbl; |
| 12603 | char const *p = ldblf + sizeof(ldblf) - 3; |
| 12604 | while (p >= ldblf) { *--ptr = *p--; } |
| 12605 | } |
| 12606 | #endif |
| 12607 | if (has_precis) { |
| 12608 | base = precis; |
| 12609 | do { *--ptr = '0' + (base % 10); } while (base /= 10); |
| 12610 | *--ptr = '.'; |
| 12611 | } |
| 12612 | if (width) { |
| 12613 | base = width; |
| 12614 | do { *--ptr = '0' + (base % 10); } while (base /= 10); |
| 12615 | } |
| 12616 | if (fill == '0') |
| 12617 | *--ptr = fill; |
| 12618 | if (left) |
| 12619 | *--ptr = '-'; |
| 12620 | if (plus) |
| 12621 | *--ptr = plus; |
| 12622 | if (alt) |
| 12623 | *--ptr = '#'; |
| 12624 | *--ptr = '%'; |
| 12625 | |
| 12626 | /* No taint. Otherwise we are in the strange situation |
| 12627 | * where printf() taints but print($float) doesn't. |
| 12628 | * --jhi */ |
| 12629 | |
| 12630 | STORE_LC_NUMERIC_SET_TO_NEEDED(); |
| 12631 | |
| 12632 | /* hopefully the above makes ptr a very constrained format |
| 12633 | * that is safe to use, even though it's not literal */ |
| 12634 | GCC_DIAG_IGNORE(-Wformat-nonliteral); |
| 12635 | #ifdef USE_QUADMATH |
| 12636 | { |
| 12637 | const char* qfmt = quadmath_format_single(ptr); |
| 12638 | if (!qfmt) |
| 12639 | Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr); |
| 12640 | elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize, |
| 12641 | qfmt, nv); |
| 12642 | if ((IV)elen == -1) |
| 12643 | Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt); |
| 12644 | if (qfmt != ptr) |
| 12645 | Safefree(qfmt); |
| 12646 | } |
| 12647 | #elif defined(HAS_LONG_DOUBLE) |
| 12648 | elen = ((intsize == 'q') |
| 12649 | ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv) |
| 12650 | : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv)); |
| 12651 | #else |
| 12652 | elen = my_sprintf(PL_efloatbuf, ptr, fv); |
| 12653 | #endif |
| 12654 | GCC_DIAG_RESTORE; |
| 12655 | } |
| 12656 | |
| 12657 | float_converted: |
| 12658 | eptr = PL_efloatbuf; |
| 12659 | assert((IV)elen > 0); /* here zero elen is bad */ |
| 12660 | |
| 12661 | #ifdef USE_LOCALE_NUMERIC |
| 12662 | /* If the decimal point character in the string is UTF-8, make the |
| 12663 | * output utf8 */ |
| 12664 | if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv) |
| 12665 | && instr(eptr, SvPVX_const(PL_numeric_radix_sv))) |
| 12666 | { |
| 12667 | is_utf8 = TRUE; |
| 12668 | } |
| 12669 | #endif |
| 12670 | |
| 12671 | break; |
| 12672 | |
| 12673 | /* SPECIAL */ |
| 12674 | |
| 12675 | case 'n': |
| 12676 | if (vectorize) |
| 12677 | goto unknown; |
| 12678 | i = SvCUR(sv) - origlen; |
| 12679 | if (args) { |
| 12680 | switch (intsize) { |
| 12681 | case 'c': *(va_arg(*args, char*)) = i; break; |
| 12682 | case 'h': *(va_arg(*args, short*)) = i; break; |
| 12683 | default: *(va_arg(*args, int*)) = i; break; |
| 12684 | case 'l': *(va_arg(*args, long*)) = i; break; |
| 12685 | case 'V': *(va_arg(*args, IV*)) = i; break; |
| 12686 | case 'z': *(va_arg(*args, SSize_t*)) = i; break; |
| 12687 | #ifdef HAS_PTRDIFF_T |
| 12688 | case 't': *(va_arg(*args, ptrdiff_t*)) = i; break; |
| 12689 | #endif |
| 12690 | #ifdef I_STDINT |
| 12691 | case 'j': *(va_arg(*args, intmax_t*)) = i; break; |
| 12692 | #endif |
| 12693 | case 'q': |
| 12694 | #if IVSIZE >= 8 |
| 12695 | *(va_arg(*args, Quad_t*)) = i; break; |
| 12696 | #else |
| 12697 | goto unknown; |
| 12698 | #endif |
| 12699 | } |
| 12700 | } |
| 12701 | else |
| 12702 | sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i); |
| 12703 | goto donevalidconversion; |
| 12704 | |
| 12705 | /* UNKNOWN */ |
| 12706 | |
| 12707 | default: |
| 12708 | unknown: |
| 12709 | if (!args |
| 12710 | && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF) |
| 12711 | && ckWARN(WARN_PRINTF)) |
| 12712 | { |
| 12713 | SV * const msg = sv_newmortal(); |
| 12714 | Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ", |
| 12715 | (PL_op->op_type == OP_PRTF) ? "" : "s"); |
| 12716 | if (fmtstart < patend) { |
| 12717 | const char * const fmtend = q < patend ? q : patend; |
| 12718 | const char * f; |
| 12719 | sv_catpvs(msg, "\"%"); |
| 12720 | for (f = fmtstart; f < fmtend; f++) { |
| 12721 | if (isPRINT(*f)) { |
| 12722 | sv_catpvn_nomg(msg, f, 1); |
| 12723 | } else { |
| 12724 | Perl_sv_catpvf(aTHX_ msg, |
| 12725 | "\\%03"UVof, (UV)*f & 0xFF); |
| 12726 | } |
| 12727 | } |
| 12728 | sv_catpvs(msg, "\""); |
| 12729 | } else { |
| 12730 | sv_catpvs(msg, "end of string"); |
| 12731 | } |
| 12732 | Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */ |
| 12733 | } |
| 12734 | |
| 12735 | /* output mangled stuff ... */ |
| 12736 | if (c == '\0') |
| 12737 | --q; |
| 12738 | eptr = p; |
| 12739 | elen = q - p; |
| 12740 | |
| 12741 | /* ... right here, because formatting flags should not apply */ |
| 12742 | SvGROW(sv, SvCUR(sv) + elen + 1); |
| 12743 | p = SvEND(sv); |
| 12744 | Copy(eptr, p, elen, char); |
| 12745 | p += elen; |
| 12746 | *p = '\0'; |
| 12747 | SvCUR_set(sv, p - SvPVX_const(sv)); |
| 12748 | svix = osvix; |
| 12749 | continue; /* not "break" */ |
| 12750 | } |
| 12751 | |
| 12752 | if (is_utf8 != has_utf8) { |
| 12753 | if (is_utf8) { |
| 12754 | if (SvCUR(sv)) |
| 12755 | sv_utf8_upgrade(sv); |
| 12756 | } |
| 12757 | else { |
| 12758 | const STRLEN old_elen = elen; |
| 12759 | SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP); |
| 12760 | sv_utf8_upgrade(nsv); |
| 12761 | eptr = SvPVX_const(nsv); |
| 12762 | elen = SvCUR(nsv); |
| 12763 | |
| 12764 | if (width) { /* fudge width (can't fudge elen) */ |
| 12765 | width += elen - old_elen; |
| 12766 | } |
| 12767 | is_utf8 = TRUE; |
| 12768 | } |
| 12769 | } |
| 12770 | |
| 12771 | /* signed value that's wrapped? */ |
| 12772 | assert(elen <= ((~(STRLEN)0) >> 1)); |
| 12773 | have = esignlen + zeros + elen; |
| 12774 | if (have < zeros) |
| 12775 | croak_memory_wrap(); |
| 12776 | |
| 12777 | need = (have > width ? have : width); |
| 12778 | gap = need - have; |
| 12779 | |
| 12780 | if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1)) |
| 12781 | croak_memory_wrap(); |
| 12782 | SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1); |
| 12783 | p = SvEND(sv); |
| 12784 | if (esignlen && fill == '0') { |
| 12785 | int i; |
| 12786 | for (i = 0; i < (int)esignlen; i++) |
| 12787 | *p++ = esignbuf[i]; |
| 12788 | } |
| 12789 | if (gap && !left) { |
| 12790 | memset(p, fill, gap); |
| 12791 | p += gap; |
| 12792 | } |
| 12793 | if (esignlen && fill != '0') { |
| 12794 | int i; |
| 12795 | for (i = 0; i < (int)esignlen; i++) |
| 12796 | *p++ = esignbuf[i]; |
| 12797 | } |
| 12798 | if (zeros) { |
| 12799 | int i; |
| 12800 | for (i = zeros; i; i--) |
| 12801 | *p++ = '0'; |
| 12802 | } |
| 12803 | if (elen) { |
| 12804 | Copy(eptr, p, elen, char); |
| 12805 | p += elen; |
| 12806 | } |
| 12807 | if (gap && left) { |
| 12808 | memset(p, ' ', gap); |
| 12809 | p += gap; |
| 12810 | } |
| 12811 | if (vectorize) { |
| 12812 | if (veclen) { |
| 12813 | Copy(dotstr, p, dotstrlen, char); |
| 12814 | p += dotstrlen; |
| 12815 | } |
| 12816 | else |
| 12817 | vectorize = FALSE; /* done iterating over vecstr */ |
| 12818 | } |
| 12819 | if (is_utf8) |
| 12820 | has_utf8 = TRUE; |
| 12821 | if (has_utf8) |
| 12822 | SvUTF8_on(sv); |
| 12823 | *p = '\0'; |
| 12824 | SvCUR_set(sv, p - SvPVX_const(sv)); |
| 12825 | if (vectorize) { |
| 12826 | esignlen = 0; |
| 12827 | goto vector; |
| 12828 | } |
| 12829 | |
| 12830 | donevalidconversion: |
| 12831 | if (used_explicit_ix) |
| 12832 | no_redundant_warning = TRUE; |
| 12833 | if (arg_missing) |
| 12834 | S_warn_vcatpvfn_missing_argument(aTHX); |
| 12835 | } |
| 12836 | |
| 12837 | /* Now that we've consumed all our printf format arguments (svix) |
| 12838 | * do we have things left on the stack that we didn't use? |
| 12839 | */ |
| 12840 | if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) { |
| 12841 | Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s", |
| 12842 | PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()"); |
| 12843 | } |
| 12844 | |
| 12845 | SvTAINT(sv); |
| 12846 | |
| 12847 | RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore |
| 12848 | each iteration. */ |
| 12849 | } |
| 12850 | |
| 12851 | /* ========================================================================= |
| 12852 | |
| 12853 | =head1 Cloning an interpreter |
| 12854 | |
| 12855 | =cut |
| 12856 | |
| 12857 | All the macros and functions in this section are for the private use of |
| 12858 | the main function, perl_clone(). |
| 12859 | |
| 12860 | The foo_dup() functions make an exact copy of an existing foo thingy. |
| 12861 | During the course of a cloning, a hash table is used to map old addresses |
| 12862 | to new addresses. The table is created and manipulated with the |
| 12863 | ptr_table_* functions. |
| 12864 | |
| 12865 | * =========================================================================*/ |
| 12866 | |
| 12867 | |
| 12868 | #if defined(USE_ITHREADS) |
| 12869 | |
| 12870 | /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */ |
| 12871 | #ifndef GpREFCNT_inc |
| 12872 | # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL) |
| 12873 | #endif |
| 12874 | |
| 12875 | |
| 12876 | /* Certain cases in Perl_ss_dup have been merged, by relying on the fact |
| 12877 | that currently av_dup, gv_dup and hv_dup are the same as sv_dup. |
| 12878 | If this changes, please unmerge ss_dup. |
| 12879 | Likewise, sv_dup_inc_multiple() relies on this fact. */ |
| 12880 | #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t)) |
| 12881 | #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t)) |
| 12882 | #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t)) |
| 12883 | #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t)) |
| 12884 | #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t)) |
| 12885 | #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t)) |
| 12886 | #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t)) |
| 12887 | #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t)) |
| 12888 | #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t)) |
| 12889 | #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t)) |
| 12890 | #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t)) |
| 12891 | #define SAVEPV(p) ((p) ? savepv(p) : NULL) |
| 12892 | #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL) |
| 12893 | |
| 12894 | /* clone a parser */ |
| 12895 | |
| 12896 | yy_parser * |
| 12897 | Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param) |
| 12898 | { |
| 12899 | yy_parser *parser; |
| 12900 | |
| 12901 | PERL_ARGS_ASSERT_PARSER_DUP; |
| 12902 | |
| 12903 | if (!proto) |
| 12904 | return NULL; |
| 12905 | |
| 12906 | /* look for it in the table first */ |
| 12907 | parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto); |
| 12908 | if (parser) |
| 12909 | return parser; |
| 12910 | |
| 12911 | /* create anew and remember what it is */ |
| 12912 | Newxz(parser, 1, yy_parser); |
| 12913 | ptr_table_store(PL_ptr_table, proto, parser); |
| 12914 | |
| 12915 | /* XXX these not yet duped */ |
| 12916 | parser->old_parser = NULL; |
| 12917 | parser->stack = NULL; |
| 12918 | parser->ps = NULL; |
| 12919 | parser->stack_size = 0; |
| 12920 | /* XXX parser->stack->state = 0; */ |
| 12921 | |
| 12922 | /* XXX eventually, just Copy() most of the parser struct ? */ |
| 12923 | |
| 12924 | parser->lex_brackets = proto->lex_brackets; |
| 12925 | parser->lex_casemods = proto->lex_casemods; |
| 12926 | parser->lex_brackstack = savepvn(proto->lex_brackstack, |
| 12927 | (proto->lex_brackets < 120 ? 120 : proto->lex_brackets)); |
| 12928 | parser->lex_casestack = savepvn(proto->lex_casestack, |
| 12929 | (proto->lex_casemods < 12 ? 12 : proto->lex_casemods)); |
| 12930 | parser->lex_defer = proto->lex_defer; |
| 12931 | parser->lex_dojoin = proto->lex_dojoin; |
| 12932 | parser->lex_formbrack = proto->lex_formbrack; |
| 12933 | parser->lex_inpat = proto->lex_inpat; |
| 12934 | parser->lex_inwhat = proto->lex_inwhat; |
| 12935 | parser->lex_op = proto->lex_op; |
| 12936 | parser->lex_repl = sv_dup_inc(proto->lex_repl, param); |
| 12937 | parser->lex_starts = proto->lex_starts; |
| 12938 | parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param); |
| 12939 | parser->multi_close = proto->multi_close; |
| 12940 | parser->multi_open = proto->multi_open; |
| 12941 | parser->multi_start = proto->multi_start; |
| 12942 | parser->multi_end = proto->multi_end; |
| 12943 | parser->preambled = proto->preambled; |
| 12944 | parser->sublex_info = proto->sublex_info; /* XXX not quite right */ |
| 12945 | parser->linestr = sv_dup_inc(proto->linestr, param); |
| 12946 | parser->expect = proto->expect; |
| 12947 | parser->copline = proto->copline; |
| 12948 | parser->last_lop_op = proto->last_lop_op; |
| 12949 | parser->lex_state = proto->lex_state; |
| 12950 | parser->rsfp = fp_dup(proto->rsfp, '<', param); |
| 12951 | /* rsfp_filters entries have fake IoDIRP() */ |
| 12952 | parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param); |
| 12953 | parser->in_my = proto->in_my; |
| 12954 | parser->in_my_stash = hv_dup(proto->in_my_stash, param); |
| 12955 | parser->error_count = proto->error_count; |
| 12956 | |
| 12957 | |
| 12958 | parser->linestr = sv_dup_inc(proto->linestr, param); |
| 12959 | |
| 12960 | { |
| 12961 | char * const ols = SvPVX(proto->linestr); |
| 12962 | char * const ls = SvPVX(parser->linestr); |
| 12963 | |
| 12964 | parser->bufptr = ls + (proto->bufptr >= ols ? |
| 12965 | proto->bufptr - ols : 0); |
| 12966 | parser->oldbufptr = ls + (proto->oldbufptr >= ols ? |
| 12967 | proto->oldbufptr - ols : 0); |
| 12968 | parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ? |
| 12969 | proto->oldoldbufptr - ols : 0); |
| 12970 | parser->linestart = ls + (proto->linestart >= ols ? |
| 12971 | proto->linestart - ols : 0); |
| 12972 | parser->last_uni = ls + (proto->last_uni >= ols ? |
| 12973 | proto->last_uni - ols : 0); |
| 12974 | parser->last_lop = ls + (proto->last_lop >= ols ? |
| 12975 | proto->last_lop - ols : 0); |
| 12976 | |
| 12977 | parser->bufend = ls + SvCUR(parser->linestr); |
| 12978 | } |
| 12979 | |
| 12980 | Copy(proto->tokenbuf, parser->tokenbuf, 256, char); |
| 12981 | |
| 12982 | |
| 12983 | Copy(proto->nextval, parser->nextval, 5, YYSTYPE); |
| 12984 | Copy(proto->nexttype, parser->nexttype, 5, I32); |
| 12985 | parser->nexttoke = proto->nexttoke; |
| 12986 | |
| 12987 | /* XXX should clone saved_curcop here, but we aren't passed |
| 12988 | * proto_perl; so do it in perl_clone_using instead */ |
| 12989 | |
| 12990 | return parser; |
| 12991 | } |
| 12992 | |
| 12993 | |
| 12994 | /* duplicate a file handle */ |
| 12995 | |
| 12996 | PerlIO * |
| 12997 | Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param) |
| 12998 | { |
| 12999 | PerlIO *ret; |
| 13000 | |
| 13001 | PERL_ARGS_ASSERT_FP_DUP; |
| 13002 | PERL_UNUSED_ARG(type); |
| 13003 | |
| 13004 | if (!fp) |
| 13005 | return (PerlIO*)NULL; |
| 13006 | |
| 13007 | /* look for it in the table first */ |
| 13008 | ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp); |
| 13009 | if (ret) |
| 13010 | return ret; |
| 13011 | |
| 13012 | /* create anew and remember what it is */ |
| 13013 | #ifdef __amigaos4__ |
| 13014 | ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD); |
| 13015 | #else |
| 13016 | ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE); |
| 13017 | #endif |
| 13018 | ptr_table_store(PL_ptr_table, fp, ret); |
| 13019 | return ret; |
| 13020 | } |
| 13021 | |
| 13022 | /* duplicate a directory handle */ |
| 13023 | |
| 13024 | DIR * |
| 13025 | Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param) |
| 13026 | { |
| 13027 | DIR *ret; |
| 13028 | |
| 13029 | #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR) |
| 13030 | DIR *pwd; |
| 13031 | const Direntry_t *dirent; |
| 13032 | char smallbuf[256]; |
| 13033 | char *name = NULL; |
| 13034 | STRLEN len = 0; |
| 13035 | long pos; |
| 13036 | #endif |
| 13037 | |
| 13038 | PERL_UNUSED_CONTEXT; |
| 13039 | PERL_ARGS_ASSERT_DIRP_DUP; |
| 13040 | |
| 13041 | if (!dp) |
| 13042 | return (DIR*)NULL; |
| 13043 | |
| 13044 | /* look for it in the table first */ |
| 13045 | ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp); |
| 13046 | if (ret) |
| 13047 | return ret; |
| 13048 | |
| 13049 | #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR) |
| 13050 | |
| 13051 | PERL_UNUSED_ARG(param); |
| 13052 | |
| 13053 | /* create anew */ |
| 13054 | |
| 13055 | /* open the current directory (so we can switch back) */ |
| 13056 | if (!(pwd = PerlDir_open("."))) return (DIR *)NULL; |
| 13057 | |
| 13058 | /* chdir to our dir handle and open the present working directory */ |
| 13059 | if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) { |
| 13060 | PerlDir_close(pwd); |
| 13061 | return (DIR *)NULL; |
| 13062 | } |
| 13063 | /* Now we should have two dir handles pointing to the same dir. */ |
| 13064 | |
| 13065 | /* Be nice to the calling code and chdir back to where we were. */ |
| 13066 | /* XXX If this fails, then what? */ |
| 13067 | PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd))); |
| 13068 | |
| 13069 | /* We have no need of the pwd handle any more. */ |
| 13070 | PerlDir_close(pwd); |
| 13071 | |
| 13072 | #ifdef DIRNAMLEN |
| 13073 | # define d_namlen(d) (d)->d_namlen |
| 13074 | #else |
| 13075 | # define d_namlen(d) strlen((d)->d_name) |
| 13076 | #endif |
| 13077 | /* Iterate once through dp, to get the file name at the current posi- |
| 13078 | tion. Then step back. */ |
| 13079 | pos = PerlDir_tell(dp); |
| 13080 | if ((dirent = PerlDir_read(dp))) { |
| 13081 | len = d_namlen(dirent); |
| 13082 | if (len <= sizeof smallbuf) name = smallbuf; |
| 13083 | else Newx(name, len, char); |
| 13084 | Move(dirent->d_name, name, len, char); |
| 13085 | } |
| 13086 | PerlDir_seek(dp, pos); |
| 13087 | |
| 13088 | /* Iterate through the new dir handle, till we find a file with the |
| 13089 | right name. */ |
| 13090 | if (!dirent) /* just before the end */ |
| 13091 | for(;;) { |
| 13092 | pos = PerlDir_tell(ret); |
| 13093 | if (PerlDir_read(ret)) continue; /* not there yet */ |
| 13094 | PerlDir_seek(ret, pos); /* step back */ |
| 13095 | break; |
| 13096 | } |
| 13097 | else { |
| 13098 | const long pos0 = PerlDir_tell(ret); |
| 13099 | for(;;) { |
| 13100 | pos = PerlDir_tell(ret); |
| 13101 | if ((dirent = PerlDir_read(ret))) { |
| 13102 | if (len == (STRLEN)d_namlen(dirent) |
| 13103 | && memEQ(name, dirent->d_name, len)) { |
| 13104 | /* found it */ |
| 13105 | PerlDir_seek(ret, pos); /* step back */ |
| 13106 | break; |
| 13107 | } |
| 13108 | /* else we are not there yet; keep iterating */ |
| 13109 | } |
| 13110 | else { /* This is not meant to happen. The best we can do is |
| 13111 | reset the iterator to the beginning. */ |
| 13112 | PerlDir_seek(ret, pos0); |
| 13113 | break; |
| 13114 | } |
| 13115 | } |
| 13116 | } |
| 13117 | #undef d_namlen |
| 13118 | |
| 13119 | if (name && name != smallbuf) |
| 13120 | Safefree(name); |
| 13121 | #endif |
| 13122 | |
| 13123 | #ifdef WIN32 |
| 13124 | ret = win32_dirp_dup(dp, param); |
| 13125 | #endif |
| 13126 | |
| 13127 | /* pop it in the pointer table */ |
| 13128 | if (ret) |
| 13129 | ptr_table_store(PL_ptr_table, dp, ret); |
| 13130 | |
| 13131 | return ret; |
| 13132 | } |
| 13133 | |
| 13134 | /* duplicate a typeglob */ |
| 13135 | |
| 13136 | GP * |
| 13137 | Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param) |
| 13138 | { |
| 13139 | GP *ret; |
| 13140 | |
| 13141 | PERL_ARGS_ASSERT_GP_DUP; |
| 13142 | |
| 13143 | if (!gp) |
| 13144 | return (GP*)NULL; |
| 13145 | /* look for it in the table first */ |
| 13146 | ret = (GP*)ptr_table_fetch(PL_ptr_table, gp); |
| 13147 | if (ret) |
| 13148 | return ret; |
| 13149 | |
| 13150 | /* create anew and remember what it is */ |
| 13151 | Newxz(ret, 1, GP); |
| 13152 | ptr_table_store(PL_ptr_table, gp, ret); |
| 13153 | |
| 13154 | /* clone */ |
| 13155 | /* ret->gp_refcnt must be 0 before any other dups are called. We're relying |
| 13156 | on Newxz() to do this for us. */ |
| 13157 | ret->gp_sv = sv_dup_inc(gp->gp_sv, param); |
| 13158 | ret->gp_io = io_dup_inc(gp->gp_io, param); |
| 13159 | ret->gp_form = cv_dup_inc(gp->gp_form, param); |
| 13160 | ret->gp_av = av_dup_inc(gp->gp_av, param); |
| 13161 | ret->gp_hv = hv_dup_inc(gp->gp_hv, param); |
| 13162 | ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */ |
| 13163 | ret->gp_cv = cv_dup_inc(gp->gp_cv, param); |
| 13164 | ret->gp_cvgen = gp->gp_cvgen; |
| 13165 | ret->gp_line = gp->gp_line; |
| 13166 | ret->gp_file_hek = hek_dup(gp->gp_file_hek, param); |
| 13167 | return ret; |
| 13168 | } |
| 13169 | |
| 13170 | /* duplicate a chain of magic */ |
| 13171 | |
| 13172 | MAGIC * |
| 13173 | Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param) |
| 13174 | { |
| 13175 | MAGIC *mgret = NULL; |
| 13176 | MAGIC **mgprev_p = &mgret; |
| 13177 | |
| 13178 | PERL_ARGS_ASSERT_MG_DUP; |
| 13179 | |
| 13180 | for (; mg; mg = mg->mg_moremagic) { |
| 13181 | MAGIC *nmg; |
| 13182 | |
| 13183 | if ((param->flags & CLONEf_JOIN_IN) |
| 13184 | && mg->mg_type == PERL_MAGIC_backref) |
| 13185 | /* when joining, we let the individual SVs add themselves to |
| 13186 | * backref as needed. */ |
| 13187 | continue; |
| 13188 | |
| 13189 | Newx(nmg, 1, MAGIC); |
| 13190 | *mgprev_p = nmg; |
| 13191 | mgprev_p = &(nmg->mg_moremagic); |
| 13192 | |
| 13193 | /* There was a comment "XXX copy dynamic vtable?" but as we don't have |
| 13194 | dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates |
| 13195 | from the original commit adding Perl_mg_dup() - revision 4538. |
| 13196 | Similarly there is the annotation "XXX random ptr?" next to the |
| 13197 | assignment to nmg->mg_ptr. */ |
| 13198 | *nmg = *mg; |
| 13199 | |
| 13200 | /* FIXME for plugins |
| 13201 | if (nmg->mg_type == PERL_MAGIC_qr) { |
| 13202 | nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param)); |
| 13203 | } |
| 13204 | else |
| 13205 | */ |
| 13206 | nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED) |
| 13207 | ? nmg->mg_type == PERL_MAGIC_backref |
| 13208 | /* The backref AV has its reference |
| 13209 | * count deliberately bumped by 1 */ |
| 13210 | ? SvREFCNT_inc(av_dup_inc((const AV *) |
| 13211 | nmg->mg_obj, param)) |
| 13212 | : sv_dup_inc(nmg->mg_obj, param) |
| 13213 | : sv_dup(nmg->mg_obj, param); |
| 13214 | |
| 13215 | if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) { |
| 13216 | if (nmg->mg_len > 0) { |
| 13217 | nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len); |
| 13218 | if (nmg->mg_type == PERL_MAGIC_overload_table && |
| 13219 | AMT_AMAGIC((AMT*)nmg->mg_ptr)) |
| 13220 | { |
| 13221 | AMT * const namtp = (AMT*)nmg->mg_ptr; |
| 13222 | sv_dup_inc_multiple((SV**)(namtp->table), |
| 13223 | (SV**)(namtp->table), NofAMmeth, param); |
| 13224 | } |
| 13225 | } |
| 13226 | else if (nmg->mg_len == HEf_SVKEY) |
| 13227 | nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param); |
| 13228 | } |
| 13229 | if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) { |
| 13230 | nmg->mg_virtual->svt_dup(aTHX_ nmg, param); |
| 13231 | } |
| 13232 | } |
| 13233 | return mgret; |
| 13234 | } |
| 13235 | |
| 13236 | #endif /* USE_ITHREADS */ |
| 13237 | |
| 13238 | struct ptr_tbl_arena { |
| 13239 | struct ptr_tbl_arena *next; |
| 13240 | struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */ |
| 13241 | }; |
| 13242 | |
| 13243 | /* create a new pointer-mapping table */ |
| 13244 | |
| 13245 | PTR_TBL_t * |
| 13246 | Perl_ptr_table_new(pTHX) |
| 13247 | { |
| 13248 | PTR_TBL_t *tbl; |
| 13249 | PERL_UNUSED_CONTEXT; |
| 13250 | |
| 13251 | Newx(tbl, 1, PTR_TBL_t); |
| 13252 | tbl->tbl_max = 511; |
| 13253 | tbl->tbl_items = 0; |
| 13254 | tbl->tbl_arena = NULL; |
| 13255 | tbl->tbl_arena_next = NULL; |
| 13256 | tbl->tbl_arena_end = NULL; |
| 13257 | Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*); |
| 13258 | return tbl; |
| 13259 | } |
| 13260 | |
| 13261 | #define PTR_TABLE_HASH(ptr) \ |
| 13262 | ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17))) |
| 13263 | |
| 13264 | /* map an existing pointer using a table */ |
| 13265 | |
| 13266 | STATIC PTR_TBL_ENT_t * |
| 13267 | S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv) |
| 13268 | { |
| 13269 | PTR_TBL_ENT_t *tblent; |
| 13270 | const UV hash = PTR_TABLE_HASH(sv); |
| 13271 | |
| 13272 | PERL_ARGS_ASSERT_PTR_TABLE_FIND; |
| 13273 | |
| 13274 | tblent = tbl->tbl_ary[hash & tbl->tbl_max]; |
| 13275 | for (; tblent; tblent = tblent->next) { |
| 13276 | if (tblent->oldval == sv) |
| 13277 | return tblent; |
| 13278 | } |
| 13279 | return NULL; |
| 13280 | } |
| 13281 | |
| 13282 | void * |
| 13283 | Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv) |
| 13284 | { |
| 13285 | PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv); |
| 13286 | |
| 13287 | PERL_ARGS_ASSERT_PTR_TABLE_FETCH; |
| 13288 | PERL_UNUSED_CONTEXT; |
| 13289 | |
| 13290 | return tblent ? tblent->newval : NULL; |
| 13291 | } |
| 13292 | |
| 13293 | /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is |
| 13294 | * the key; 'newsv' is the value. The names "old" and "new" are specific to |
| 13295 | * the core's typical use of ptr_tables in thread cloning. */ |
| 13296 | |
| 13297 | void |
| 13298 | Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv) |
| 13299 | { |
| 13300 | PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv); |
| 13301 | |
| 13302 | PERL_ARGS_ASSERT_PTR_TABLE_STORE; |
| 13303 | PERL_UNUSED_CONTEXT; |
| 13304 | |
| 13305 | if (tblent) { |
| 13306 | tblent->newval = newsv; |
| 13307 | } else { |
| 13308 | const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max; |
| 13309 | |
| 13310 | if (tbl->tbl_arena_next == tbl->tbl_arena_end) { |
| 13311 | struct ptr_tbl_arena *new_arena; |
| 13312 | |
| 13313 | Newx(new_arena, 1, struct ptr_tbl_arena); |
| 13314 | new_arena->next = tbl->tbl_arena; |
| 13315 | tbl->tbl_arena = new_arena; |
| 13316 | tbl->tbl_arena_next = new_arena->array; |
| 13317 | tbl->tbl_arena_end = C_ARRAY_END(new_arena->array); |
| 13318 | } |
| 13319 | |
| 13320 | tblent = tbl->tbl_arena_next++; |
| 13321 | |
| 13322 | tblent->oldval = oldsv; |
| 13323 | tblent->newval = newsv; |
| 13324 | tblent->next = tbl->tbl_ary[entry]; |
| 13325 | tbl->tbl_ary[entry] = tblent; |
| 13326 | tbl->tbl_items++; |
| 13327 | if (tblent->next && tbl->tbl_items > tbl->tbl_max) |
| 13328 | ptr_table_split(tbl); |
| 13329 | } |
| 13330 | } |
| 13331 | |
| 13332 | /* double the hash bucket size of an existing ptr table */ |
| 13333 | |
| 13334 | void |
| 13335 | Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl) |
| 13336 | { |
| 13337 | PTR_TBL_ENT_t **ary = tbl->tbl_ary; |
| 13338 | const UV oldsize = tbl->tbl_max + 1; |
| 13339 | UV newsize = oldsize * 2; |
| 13340 | UV i; |
| 13341 | |
| 13342 | PERL_ARGS_ASSERT_PTR_TABLE_SPLIT; |
| 13343 | PERL_UNUSED_CONTEXT; |
| 13344 | |
| 13345 | Renew(ary, newsize, PTR_TBL_ENT_t*); |
| 13346 | Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*); |
| 13347 | tbl->tbl_max = --newsize; |
| 13348 | tbl->tbl_ary = ary; |
| 13349 | for (i=0; i < oldsize; i++, ary++) { |
| 13350 | PTR_TBL_ENT_t **entp = ary; |
| 13351 | PTR_TBL_ENT_t *ent = *ary; |
| 13352 | PTR_TBL_ENT_t **curentp; |
| 13353 | if (!ent) |
| 13354 | continue; |
| 13355 | curentp = ary + oldsize; |
| 13356 | do { |
| 13357 | if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) { |
| 13358 | *entp = ent->next; |
| 13359 | ent->next = *curentp; |
| 13360 | *curentp = ent; |
| 13361 | } |
| 13362 | else |
| 13363 | entp = &ent->next; |
| 13364 | ent = *entp; |
| 13365 | } while (ent); |
| 13366 | } |
| 13367 | } |
| 13368 | |
| 13369 | /* remove all the entries from a ptr table */ |
| 13370 | /* Deprecated - will be removed post 5.14 */ |
| 13371 | |
| 13372 | void |
| 13373 | Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl) |
| 13374 | { |
| 13375 | PERL_UNUSED_CONTEXT; |
| 13376 | if (tbl && tbl->tbl_items) { |
| 13377 | struct ptr_tbl_arena *arena = tbl->tbl_arena; |
| 13378 | |
| 13379 | Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *); |
| 13380 | |
| 13381 | while (arena) { |
| 13382 | struct ptr_tbl_arena *next = arena->next; |
| 13383 | |
| 13384 | Safefree(arena); |
| 13385 | arena = next; |
| 13386 | }; |
| 13387 | |
| 13388 | tbl->tbl_items = 0; |
| 13389 | tbl->tbl_arena = NULL; |
| 13390 | tbl->tbl_arena_next = NULL; |
| 13391 | tbl->tbl_arena_end = NULL; |
| 13392 | } |
| 13393 | } |
| 13394 | |
| 13395 | /* clear and free a ptr table */ |
| 13396 | |
| 13397 | void |
| 13398 | Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl) |
| 13399 | { |
| 13400 | struct ptr_tbl_arena *arena; |
| 13401 | |
| 13402 | PERL_UNUSED_CONTEXT; |
| 13403 | |
| 13404 | if (!tbl) { |
| 13405 | return; |
| 13406 | } |
| 13407 | |
| 13408 | arena = tbl->tbl_arena; |
| 13409 | |
| 13410 | while (arena) { |
| 13411 | struct ptr_tbl_arena *next = arena->next; |
| 13412 | |
| 13413 | Safefree(arena); |
| 13414 | arena = next; |
| 13415 | } |
| 13416 | |
| 13417 | Safefree(tbl->tbl_ary); |
| 13418 | Safefree(tbl); |
| 13419 | } |
| 13420 | |
| 13421 | #if defined(USE_ITHREADS) |
| 13422 | |
| 13423 | void |
| 13424 | Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param) |
| 13425 | { |
| 13426 | PERL_ARGS_ASSERT_RVPV_DUP; |
| 13427 | |
| 13428 | assert(!isREGEXP(sstr)); |
| 13429 | if (SvROK(sstr)) { |
| 13430 | if (SvWEAKREF(sstr)) { |
| 13431 | SvRV_set(dstr, sv_dup(SvRV_const(sstr), param)); |
| 13432 | if (param->flags & CLONEf_JOIN_IN) { |
| 13433 | /* if joining, we add any back references individually rather |
| 13434 | * than copying the whole backref array */ |
| 13435 | Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr); |
| 13436 | } |
| 13437 | } |
| 13438 | else |
| 13439 | SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param)); |
| 13440 | } |
| 13441 | else if (SvPVX_const(sstr)) { |
| 13442 | /* Has something there */ |
| 13443 | if (SvLEN(sstr)) { |
| 13444 | /* Normal PV - clone whole allocated space */ |
| 13445 | SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1)); |
| 13446 | /* sstr may not be that normal, but actually copy on write. |
| 13447 | But we are a true, independent SV, so: */ |
| 13448 | SvIsCOW_off(dstr); |
| 13449 | } |
| 13450 | else { |
| 13451 | /* Special case - not normally malloced for some reason */ |
| 13452 | if (isGV_with_GP(sstr)) { |
| 13453 | /* Don't need to do anything here. */ |
| 13454 | } |
| 13455 | else if ((SvIsCOW(sstr))) { |
| 13456 | /* A "shared" PV - clone it as "shared" PV */ |
| 13457 | SvPV_set(dstr, |
| 13458 | HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)), |
| 13459 | param))); |
| 13460 | } |
| 13461 | else { |
| 13462 | /* Some other special case - random pointer */ |
| 13463 | SvPV_set(dstr, (char *) SvPVX_const(sstr)); |
| 13464 | } |
| 13465 | } |
| 13466 | } |
| 13467 | else { |
| 13468 | /* Copy the NULL */ |
| 13469 | SvPV_set(dstr, NULL); |
| 13470 | } |
| 13471 | } |
| 13472 | |
| 13473 | /* duplicate a list of SVs. source and dest may point to the same memory. */ |
| 13474 | static SV ** |
| 13475 | S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest, |
| 13476 | SSize_t items, CLONE_PARAMS *const param) |
| 13477 | { |
| 13478 | PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE; |
| 13479 | |
| 13480 | while (items-- > 0) { |
| 13481 | *dest++ = sv_dup_inc(*source++, param); |
| 13482 | } |
| 13483 | |
| 13484 | return dest; |
| 13485 | } |
| 13486 | |
| 13487 | /* duplicate an SV of any type (including AV, HV etc) */ |
| 13488 | |
| 13489 | static SV * |
| 13490 | S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param) |
| 13491 | { |
| 13492 | dVAR; |
| 13493 | SV *dstr; |
| 13494 | |
| 13495 | PERL_ARGS_ASSERT_SV_DUP_COMMON; |
| 13496 | |
| 13497 | if (SvTYPE(sstr) == (svtype)SVTYPEMASK) { |
| 13498 | #ifdef DEBUG_LEAKING_SCALARS_ABORT |
| 13499 | abort(); |
| 13500 | #endif |
| 13501 | return NULL; |
| 13502 | } |
| 13503 | /* look for it in the table first */ |
| 13504 | dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr)); |
| 13505 | if (dstr) |
| 13506 | return dstr; |
| 13507 | |
| 13508 | if(param->flags & CLONEf_JOIN_IN) { |
| 13509 | /** We are joining here so we don't want do clone |
| 13510 | something that is bad **/ |
| 13511 | if (SvTYPE(sstr) == SVt_PVHV) { |
| 13512 | const HEK * const hvname = HvNAME_HEK(sstr); |
| 13513 | if (hvname) { |
| 13514 | /** don't clone stashes if they already exist **/ |
| 13515 | dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname), |
| 13516 | HEK_UTF8(hvname) ? SVf_UTF8 : 0)); |
| 13517 | ptr_table_store(PL_ptr_table, sstr, dstr); |
| 13518 | return dstr; |
| 13519 | } |
| 13520 | } |
| 13521 | else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) { |
| 13522 | HV *stash = GvSTASH(sstr); |
| 13523 | const HEK * hvname; |
| 13524 | if (stash && (hvname = HvNAME_HEK(stash))) { |
| 13525 | /** don't clone GVs if they already exist **/ |
| 13526 | SV **svp; |
| 13527 | stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname), |
| 13528 | HEK_UTF8(hvname) ? SVf_UTF8 : 0); |
| 13529 | svp = hv_fetch( |
| 13530 | stash, GvNAME(sstr), |
| 13531 | GvNAMEUTF8(sstr) |
| 13532 | ? -GvNAMELEN(sstr) |
| 13533 | : GvNAMELEN(sstr), |
| 13534 | 0 |
| 13535 | ); |
| 13536 | if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) { |
| 13537 | ptr_table_store(PL_ptr_table, sstr, *svp); |
| 13538 | return *svp; |
| 13539 | } |
| 13540 | } |
| 13541 | } |
| 13542 | } |
| 13543 | |
| 13544 | /* create anew and remember what it is */ |
| 13545 | new_SV(dstr); |
| 13546 | |
| 13547 | #ifdef DEBUG_LEAKING_SCALARS |
| 13548 | dstr->sv_debug_optype = sstr->sv_debug_optype; |
| 13549 | dstr->sv_debug_line = sstr->sv_debug_line; |
| 13550 | dstr->sv_debug_inpad = sstr->sv_debug_inpad; |
| 13551 | dstr->sv_debug_parent = (SV*)sstr; |
| 13552 | FREE_SV_DEBUG_FILE(dstr); |
| 13553 | dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file); |
| 13554 | #endif |
| 13555 | |
| 13556 | ptr_table_store(PL_ptr_table, sstr, dstr); |
| 13557 | |
| 13558 | /* clone */ |
| 13559 | SvFLAGS(dstr) = SvFLAGS(sstr); |
| 13560 | SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */ |
| 13561 | SvREFCNT(dstr) = 0; /* must be before any other dups! */ |
| 13562 | |
| 13563 | #ifdef DEBUGGING |
| 13564 | if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx) |
| 13565 | PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n", |
| 13566 | (void*)PL_watch_pvx, SvPVX_const(sstr)); |
| 13567 | #endif |
| 13568 | |
| 13569 | /* don't clone objects whose class has asked us not to */ |
| 13570 | if (SvOBJECT(sstr) |
| 13571 | && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) |
| 13572 | { |
| 13573 | SvFLAGS(dstr) = 0; |
| 13574 | return dstr; |
| 13575 | } |
| 13576 | |
| 13577 | switch (SvTYPE(sstr)) { |
| 13578 | case SVt_NULL: |
| 13579 | SvANY(dstr) = NULL; |
| 13580 | break; |
| 13581 | case SVt_IV: |
| 13582 | SET_SVANY_FOR_BODYLESS_IV(dstr); |
| 13583 | if(SvROK(sstr)) { |
| 13584 | Perl_rvpv_dup(aTHX_ dstr, sstr, param); |
| 13585 | } else { |
| 13586 | SvIV_set(dstr, SvIVX(sstr)); |
| 13587 | } |
| 13588 | break; |
| 13589 | case SVt_NV: |
| 13590 | #if NVSIZE <= IVSIZE |
| 13591 | SET_SVANY_FOR_BODYLESS_NV(dstr); |
| 13592 | #else |
| 13593 | SvANY(dstr) = new_XNV(); |
| 13594 | #endif |
| 13595 | SvNV_set(dstr, SvNVX(sstr)); |
| 13596 | break; |
| 13597 | default: |
| 13598 | { |
| 13599 | /* These are all the types that need complex bodies allocating. */ |
| 13600 | void *new_body; |
| 13601 | const svtype sv_type = SvTYPE(sstr); |
| 13602 | const struct body_details *const sv_type_details |
| 13603 | = bodies_by_type + sv_type; |
| 13604 | |
| 13605 | switch (sv_type) { |
| 13606 | default: |
| 13607 | Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr)); |
| 13608 | break; |
| 13609 | |
| 13610 | case SVt_PVGV: |
| 13611 | case SVt_PVIO: |
| 13612 | case SVt_PVFM: |
| 13613 | case SVt_PVHV: |
| 13614 | case SVt_PVAV: |
| 13615 | case SVt_PVCV: |
| 13616 | case SVt_PVLV: |
| 13617 | case SVt_REGEXP: |
| 13618 | case SVt_PVMG: |
| 13619 | case SVt_PVNV: |
| 13620 | case SVt_PVIV: |
| 13621 | case SVt_INVLIST: |
| 13622 | case SVt_PV: |
| 13623 | assert(sv_type_details->body_size); |
| 13624 | if (sv_type_details->arena) { |
| 13625 | new_body_inline(new_body, sv_type); |
| 13626 | new_body |
| 13627 | = (void*)((char*)new_body - sv_type_details->offset); |
| 13628 | } else { |
| 13629 | new_body = new_NOARENA(sv_type_details); |
| 13630 | } |
| 13631 | } |
| 13632 | assert(new_body); |
| 13633 | SvANY(dstr) = new_body; |
| 13634 | |
| 13635 | #ifndef PURIFY |
| 13636 | Copy(((char*)SvANY(sstr)) + sv_type_details->offset, |
| 13637 | ((char*)SvANY(dstr)) + sv_type_details->offset, |
| 13638 | sv_type_details->copy, char); |
| 13639 | #else |
| 13640 | Copy(((char*)SvANY(sstr)), |
| 13641 | ((char*)SvANY(dstr)), |
| 13642 | sv_type_details->body_size + sv_type_details->offset, char); |
| 13643 | #endif |
| 13644 | |
| 13645 | if (sv_type != SVt_PVAV && sv_type != SVt_PVHV |
| 13646 | && !isGV_with_GP(dstr) |
| 13647 | && !isREGEXP(dstr) |
| 13648 | && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP))) |
| 13649 | Perl_rvpv_dup(aTHX_ dstr, sstr, param); |
| 13650 | |
| 13651 | /* The Copy above means that all the source (unduplicated) pointers |
| 13652 | are now in the destination. We can check the flags and the |
| 13653 | pointers in either, but it's possible that there's less cache |
| 13654 | missing by always going for the destination. |
| 13655 | FIXME - instrument and check that assumption */ |
| 13656 | if (sv_type >= SVt_PVMG) { |
| 13657 | if (SvMAGIC(dstr)) |
| 13658 | SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param)); |
| 13659 | if (SvOBJECT(dstr) && SvSTASH(dstr)) |
| 13660 | SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param)); |
| 13661 | else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */ |
| 13662 | } |
| 13663 | |
| 13664 | /* The cast silences a GCC warning about unhandled types. */ |
| 13665 | switch ((int)sv_type) { |
| 13666 | case SVt_PV: |
| 13667 | break; |
| 13668 | case SVt_PVIV: |
| 13669 | break; |
| 13670 | case SVt_PVNV: |
| 13671 | break; |
| 13672 | case SVt_PVMG: |
| 13673 | break; |
| 13674 | case SVt_REGEXP: |
| 13675 | duprex: |
| 13676 | /* FIXME for plugins */ |
| 13677 | dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any; |
| 13678 | re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param); |
| 13679 | break; |
| 13680 | case SVt_PVLV: |
| 13681 | /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */ |
| 13682 | if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */ |
| 13683 | LvTARG(dstr) = dstr; |
| 13684 | else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */ |
| 13685 | LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param)); |
| 13686 | else |
| 13687 | LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param); |
| 13688 | if (isREGEXP(sstr)) goto duprex; |
| 13689 | case SVt_PVGV: |
| 13690 | /* non-GP case already handled above */ |
| 13691 | if(isGV_with_GP(sstr)) { |
| 13692 | GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param); |
| 13693 | /* Don't call sv_add_backref here as it's going to be |
| 13694 | created as part of the magic cloning of the symbol |
| 13695 | table--unless this is during a join and the stash |
| 13696 | is not actually being cloned. */ |
| 13697 | /* Danger Will Robinson - GvGP(dstr) isn't initialised |
| 13698 | at the point of this comment. */ |
| 13699 | GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param); |
| 13700 | if (param->flags & CLONEf_JOIN_IN) |
| 13701 | Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr); |
| 13702 | GvGP_set(dstr, gp_dup(GvGP(sstr), param)); |
| 13703 | (void)GpREFCNT_inc(GvGP(dstr)); |
| 13704 | } |
| 13705 | break; |
| 13706 | case SVt_PVIO: |
| 13707 | /* PL_parser->rsfp_filters entries have fake IoDIRP() */ |
| 13708 | if(IoFLAGS(dstr) & IOf_FAKE_DIRP) { |
| 13709 | /* I have no idea why fake dirp (rsfps) |
| 13710 | should be treated differently but otherwise |
| 13711 | we end up with leaks -- sky*/ |
| 13712 | IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param); |
| 13713 | IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param); |
| 13714 | IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param); |
| 13715 | } else { |
| 13716 | IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param); |
| 13717 | IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param); |
| 13718 | IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param); |
| 13719 | if (IoDIRP(dstr)) { |
| 13720 | IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param); |
| 13721 | } else { |
| 13722 | NOOP; |
| 13723 | /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */ |
| 13724 | } |
| 13725 | IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param); |
| 13726 | } |
| 13727 | if (IoOFP(dstr) == IoIFP(sstr)) |
| 13728 | IoOFP(dstr) = IoIFP(dstr); |
| 13729 | else |
| 13730 | IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param); |
| 13731 | IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr)); |
| 13732 | IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr)); |
| 13733 | IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr)); |
| 13734 | break; |
| 13735 | case SVt_PVAV: |
| 13736 | /* avoid cloning an empty array */ |
| 13737 | if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) { |
| 13738 | SV **dst_ary, **src_ary; |
| 13739 | SSize_t items = AvFILLp((const AV *)sstr) + 1; |
| 13740 | |
| 13741 | src_ary = AvARRAY((const AV *)sstr); |
| 13742 | Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*); |
| 13743 | ptr_table_store(PL_ptr_table, src_ary, dst_ary); |
| 13744 | AvARRAY(MUTABLE_AV(dstr)) = dst_ary; |
| 13745 | AvALLOC((const AV *)dstr) = dst_ary; |
| 13746 | if (AvREAL((const AV *)sstr)) { |
| 13747 | dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items, |
| 13748 | param); |
| 13749 | } |
| 13750 | else { |
| 13751 | while (items-- > 0) |
| 13752 | *dst_ary++ = sv_dup(*src_ary++, param); |
| 13753 | } |
| 13754 | items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr); |
| 13755 | while (items-- > 0) { |
| 13756 | *dst_ary++ = NULL; |
| 13757 | } |
| 13758 | } |
| 13759 | else { |
| 13760 | AvARRAY(MUTABLE_AV(dstr)) = NULL; |
| 13761 | AvALLOC((const AV *)dstr) = (SV**)NULL; |
| 13762 | AvMAX( (const AV *)dstr) = -1; |
| 13763 | AvFILLp((const AV *)dstr) = -1; |
| 13764 | } |
| 13765 | break; |
| 13766 | case SVt_PVHV: |
| 13767 | if (HvARRAY((const HV *)sstr)) { |
| 13768 | STRLEN i = 0; |
| 13769 | const bool sharekeys = !!HvSHAREKEYS(sstr); |
| 13770 | XPVHV * const dxhv = (XPVHV*)SvANY(dstr); |
| 13771 | XPVHV * const sxhv = (XPVHV*)SvANY(sstr); |
| 13772 | char *darray; |
| 13773 | Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1) |
| 13774 | + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0), |
| 13775 | char); |
| 13776 | HvARRAY(dstr) = (HE**)darray; |
| 13777 | while (i <= sxhv->xhv_max) { |
| 13778 | const HE * const source = HvARRAY(sstr)[i]; |
| 13779 | HvARRAY(dstr)[i] = source |
| 13780 | ? he_dup(source, sharekeys, param) : 0; |
| 13781 | ++i; |
| 13782 | } |
| 13783 | if (SvOOK(sstr)) { |
| 13784 | const struct xpvhv_aux * const saux = HvAUX(sstr); |
| 13785 | struct xpvhv_aux * const daux = HvAUX(dstr); |
| 13786 | /* This flag isn't copied. */ |
| 13787 | SvOOK_on(dstr); |
| 13788 | |
| 13789 | if (saux->xhv_name_count) { |
| 13790 | HEK ** const sname = saux->xhv_name_u.xhvnameu_names; |
| 13791 | const I32 count |
| 13792 | = saux->xhv_name_count < 0 |
| 13793 | ? -saux->xhv_name_count |
| 13794 | : saux->xhv_name_count; |
| 13795 | HEK **shekp = sname + count; |
| 13796 | HEK **dhekp; |
| 13797 | Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *); |
| 13798 | dhekp = daux->xhv_name_u.xhvnameu_names + count; |
| 13799 | while (shekp-- > sname) { |
| 13800 | dhekp--; |
| 13801 | *dhekp = hek_dup(*shekp, param); |
| 13802 | } |
| 13803 | } |
| 13804 | else { |
| 13805 | daux->xhv_name_u.xhvnameu_name |
| 13806 | = hek_dup(saux->xhv_name_u.xhvnameu_name, |
| 13807 | param); |
| 13808 | } |
| 13809 | daux->xhv_name_count = saux->xhv_name_count; |
| 13810 | |
| 13811 | daux->xhv_fill_lazy = saux->xhv_fill_lazy; |
| 13812 | daux->xhv_aux_flags = saux->xhv_aux_flags; |
| 13813 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 13814 | daux->xhv_rand = saux->xhv_rand; |
| 13815 | daux->xhv_last_rand = saux->xhv_last_rand; |
| 13816 | #endif |
| 13817 | daux->xhv_riter = saux->xhv_riter; |
| 13818 | daux->xhv_eiter = saux->xhv_eiter |
| 13819 | ? he_dup(saux->xhv_eiter, |
| 13820 | cBOOL(HvSHAREKEYS(sstr)), param) : 0; |
| 13821 | /* backref array needs refcnt=2; see sv_add_backref */ |
| 13822 | daux->xhv_backreferences = |
| 13823 | (param->flags & CLONEf_JOIN_IN) |
| 13824 | /* when joining, we let the individual GVs and |
| 13825 | * CVs add themselves to backref as |
| 13826 | * needed. This avoids pulling in stuff |
| 13827 | * that isn't required, and simplifies the |
| 13828 | * case where stashes aren't cloned back |
| 13829 | * if they already exist in the parent |
| 13830 | * thread */ |
| 13831 | ? NULL |
| 13832 | : saux->xhv_backreferences |
| 13833 | ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV) |
| 13834 | ? MUTABLE_AV(SvREFCNT_inc( |
| 13835 | sv_dup_inc((const SV *) |
| 13836 | saux->xhv_backreferences, param))) |
| 13837 | : MUTABLE_AV(sv_dup((const SV *) |
| 13838 | saux->xhv_backreferences, param)) |
| 13839 | : 0; |
| 13840 | |
| 13841 | daux->xhv_mro_meta = saux->xhv_mro_meta |
| 13842 | ? mro_meta_dup(saux->xhv_mro_meta, param) |
| 13843 | : 0; |
| 13844 | |
| 13845 | /* Record stashes for possible cloning in Perl_clone(). */ |
| 13846 | if (HvNAME(sstr)) |
| 13847 | av_push(param->stashes, dstr); |
| 13848 | } |
| 13849 | } |
| 13850 | else |
| 13851 | HvARRAY(MUTABLE_HV(dstr)) = NULL; |
| 13852 | break; |
| 13853 | case SVt_PVCV: |
| 13854 | if (!(param->flags & CLONEf_COPY_STACKS)) { |
| 13855 | CvDEPTH(dstr) = 0; |
| 13856 | } |
| 13857 | /* FALLTHROUGH */ |
| 13858 | case SVt_PVFM: |
| 13859 | /* NOTE: not refcounted */ |
| 13860 | SvANY(MUTABLE_CV(dstr))->xcv_stash = |
| 13861 | hv_dup(CvSTASH(dstr), param); |
| 13862 | if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr)) |
| 13863 | Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr); |
| 13864 | if (!CvISXSUB(dstr)) { |
| 13865 | OP_REFCNT_LOCK; |
| 13866 | CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr)); |
| 13867 | OP_REFCNT_UNLOCK; |
| 13868 | CvSLABBED_off(dstr); |
| 13869 | } else if (CvCONST(dstr)) { |
| 13870 | CvXSUBANY(dstr).any_ptr = |
| 13871 | sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param); |
| 13872 | } |
| 13873 | assert(!CvSLABBED(dstr)); |
| 13874 | if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr)); |
| 13875 | if (CvNAMED(dstr)) |
| 13876 | SvANY((CV *)dstr)->xcv_gv_u.xcv_hek = |
| 13877 | hek_dup(CvNAME_HEK((CV *)sstr), param); |
| 13878 | /* don't dup if copying back - CvGV isn't refcounted, so the |
| 13879 | * duped GV may never be freed. A bit of a hack! DAPM */ |
| 13880 | else |
| 13881 | SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv = |
| 13882 | CvCVGV_RC(dstr) |
| 13883 | ? gv_dup_inc(CvGV(sstr), param) |
| 13884 | : (param->flags & CLONEf_JOIN_IN) |
| 13885 | ? NULL |
| 13886 | : gv_dup(CvGV(sstr), param); |
| 13887 | |
| 13888 | if (!CvISXSUB(sstr)) { |
| 13889 | PADLIST * padlist = CvPADLIST(sstr); |
| 13890 | if(padlist) |
| 13891 | padlist = padlist_dup(padlist, param); |
| 13892 | CvPADLIST_set(dstr, padlist); |
| 13893 | } else |
| 13894 | /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */ |
| 13895 | PoisonPADLIST(dstr); |
| 13896 | |
| 13897 | CvOUTSIDE(dstr) = |
| 13898 | CvWEAKOUTSIDE(sstr) |
| 13899 | ? cv_dup( CvOUTSIDE(dstr), param) |
| 13900 | : cv_dup_inc(CvOUTSIDE(dstr), param); |
| 13901 | break; |
| 13902 | } |
| 13903 | } |
| 13904 | } |
| 13905 | |
| 13906 | return dstr; |
| 13907 | } |
| 13908 | |
| 13909 | SV * |
| 13910 | Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param) |
| 13911 | { |
| 13912 | PERL_ARGS_ASSERT_SV_DUP_INC; |
| 13913 | return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL; |
| 13914 | } |
| 13915 | |
| 13916 | SV * |
| 13917 | Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param) |
| 13918 | { |
| 13919 | SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL; |
| 13920 | PERL_ARGS_ASSERT_SV_DUP; |
| 13921 | |
| 13922 | /* Track every SV that (at least initially) had a reference count of 0. |
| 13923 | We need to do this by holding an actual reference to it in this array. |
| 13924 | If we attempt to cheat, turn AvREAL_off(), and store only pointers |
| 13925 | (akin to the stashes hash, and the perl stack), we come unstuck if |
| 13926 | a weak reference (or other SV legitimately SvREFCNT() == 0 for this |
| 13927 | thread) is manipulated in a CLONE method, because CLONE runs before the |
| 13928 | unreferenced array is walked to find SVs still with SvREFCNT() == 0 |
| 13929 | (and fix things up by giving each a reference via the temps stack). |
| 13930 | Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and |
| 13931 | then SvREFCNT_dec(), it will be cleaned up (and added to the free list) |
| 13932 | before the walk of unreferenced happens and a reference to that is SV |
| 13933 | added to the temps stack. At which point we have the same SV considered |
| 13934 | to be in use, and free to be re-used. Not good. |
| 13935 | */ |
| 13936 | if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) { |
| 13937 | assert(param->unreferenced); |
| 13938 | av_push(param->unreferenced, SvREFCNT_inc(dstr)); |
| 13939 | } |
| 13940 | |
| 13941 | return dstr; |
| 13942 | } |
| 13943 | |
| 13944 | /* duplicate a context */ |
| 13945 | |
| 13946 | PERL_CONTEXT * |
| 13947 | Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param) |
| 13948 | { |
| 13949 | PERL_CONTEXT *ncxs; |
| 13950 | |
| 13951 | PERL_ARGS_ASSERT_CX_DUP; |
| 13952 | |
| 13953 | if (!cxs) |
| 13954 | return (PERL_CONTEXT*)NULL; |
| 13955 | |
| 13956 | /* look for it in the table first */ |
| 13957 | ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs); |
| 13958 | if (ncxs) |
| 13959 | return ncxs; |
| 13960 | |
| 13961 | /* create anew and remember what it is */ |
| 13962 | Newx(ncxs, max + 1, PERL_CONTEXT); |
| 13963 | ptr_table_store(PL_ptr_table, cxs, ncxs); |
| 13964 | Copy(cxs, ncxs, max + 1, PERL_CONTEXT); |
| 13965 | |
| 13966 | while (ix >= 0) { |
| 13967 | PERL_CONTEXT * const ncx = &ncxs[ix]; |
| 13968 | if (CxTYPE(ncx) == CXt_SUBST) { |
| 13969 | Perl_croak(aTHX_ "Cloning substitution context is unimplemented"); |
| 13970 | } |
| 13971 | else { |
| 13972 | ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl); |
| 13973 | switch (CxTYPE(ncx)) { |
| 13974 | case CXt_SUB: |
| 13975 | ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param); |
| 13976 | if(CxHASARGS(ncx)){ |
| 13977 | ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param); |
| 13978 | } else { |
| 13979 | ncx->blk_sub.savearray = NULL; |
| 13980 | } |
| 13981 | ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table, |
| 13982 | ncx->blk_sub.prevcomppad); |
| 13983 | break; |
| 13984 | case CXt_EVAL: |
| 13985 | ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv, |
| 13986 | param); |
| 13987 | ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param); |
| 13988 | ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param); |
| 13989 | break; |
| 13990 | case CXt_LOOP_LAZYSV: |
| 13991 | ncx->blk_loop.state_u.lazysv.end |
| 13992 | = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param); |
| 13993 | /* Fallthrough: duplicate lazysv.cur by using the ary.ary |
| 13994 | duplication code instead. |
| 13995 | We are taking advantage of (1) av_dup_inc and sv_dup_inc |
| 13996 | actually being the same function, and (2) order |
| 13997 | equivalence of the two unions. |
| 13998 | We can assert the later [but only at run time :-(] */ |
| 13999 | assert ((void *) &ncx->blk_loop.state_u.ary.ary == |
| 14000 | (void *) &ncx->blk_loop.state_u.lazysv.cur); |
| 14001 | /* FALLTHROUGH */ |
| 14002 | case CXt_LOOP_ARY: |
| 14003 | ncx->blk_loop.state_u.ary.ary |
| 14004 | = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param); |
| 14005 | /* FALLTHROUGH */ |
| 14006 | case CXt_LOOP_LIST: |
| 14007 | case CXt_LOOP_LAZYIV: |
| 14008 | case CXt_LOOP_PLAIN: |
| 14009 | /* code common to all CXt_LOOP_* types */ |
| 14010 | if (CxPADLOOP(ncx)) { |
| 14011 | PADOFFSET off = ncx->blk_loop.itervar_u.svp |
| 14012 | - &CX_CURPAD_SV(ncx->blk_loop, 0); |
| 14013 | ncx->blk_loop.oldcomppad = |
| 14014 | (PAD*)ptr_table_fetch(PL_ptr_table, |
| 14015 | ncx->blk_loop.oldcomppad); |
| 14016 | ncx->blk_loop.itervar_u.svp = |
| 14017 | &CX_CURPAD_SV(ncx->blk_loop, off); |
| 14018 | |
| 14019 | } |
| 14020 | else { |
| 14021 | ncx->blk_loop.itervar_u.gv |
| 14022 | = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv, |
| 14023 | param); |
| 14024 | } |
| 14025 | break; |
| 14026 | case CXt_FORMAT: |
| 14027 | ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param); |
| 14028 | ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param); |
| 14029 | ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv, |
| 14030 | param); |
| 14031 | break; |
| 14032 | case CXt_BLOCK: |
| 14033 | case CXt_NULL: |
| 14034 | case CXt_WHEN: |
| 14035 | case CXt_GIVEN: |
| 14036 | break; |
| 14037 | } |
| 14038 | } |
| 14039 | --ix; |
| 14040 | } |
| 14041 | return ncxs; |
| 14042 | } |
| 14043 | |
| 14044 | /* duplicate a stack info structure */ |
| 14045 | |
| 14046 | PERL_SI * |
| 14047 | Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param) |
| 14048 | { |
| 14049 | PERL_SI *nsi; |
| 14050 | |
| 14051 | PERL_ARGS_ASSERT_SI_DUP; |
| 14052 | |
| 14053 | if (!si) |
| 14054 | return (PERL_SI*)NULL; |
| 14055 | |
| 14056 | /* look for it in the table first */ |
| 14057 | nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si); |
| 14058 | if (nsi) |
| 14059 | return nsi; |
| 14060 | |
| 14061 | /* create anew and remember what it is */ |
| 14062 | Newxz(nsi, 1, PERL_SI); |
| 14063 | ptr_table_store(PL_ptr_table, si, nsi); |
| 14064 | |
| 14065 | nsi->si_stack = av_dup_inc(si->si_stack, param); |
| 14066 | nsi->si_cxix = si->si_cxix; |
| 14067 | nsi->si_cxmax = si->si_cxmax; |
| 14068 | nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param); |
| 14069 | nsi->si_type = si->si_type; |
| 14070 | nsi->si_prev = si_dup(si->si_prev, param); |
| 14071 | nsi->si_next = si_dup(si->si_next, param); |
| 14072 | nsi->si_markoff = si->si_markoff; |
| 14073 | |
| 14074 | return nsi; |
| 14075 | } |
| 14076 | |
| 14077 | #define POPINT(ss,ix) ((ss)[--(ix)].any_i32) |
| 14078 | #define TOPINT(ss,ix) ((ss)[ix].any_i32) |
| 14079 | #define POPLONG(ss,ix) ((ss)[--(ix)].any_long) |
| 14080 | #define TOPLONG(ss,ix) ((ss)[ix].any_long) |
| 14081 | #define POPIV(ss,ix) ((ss)[--(ix)].any_iv) |
| 14082 | #define TOPIV(ss,ix) ((ss)[ix].any_iv) |
| 14083 | #define POPUV(ss,ix) ((ss)[--(ix)].any_uv) |
| 14084 | #define TOPUV(ss,ix) ((ss)[ix].any_uv) |
| 14085 | #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool) |
| 14086 | #define TOPBOOL(ss,ix) ((ss)[ix].any_bool) |
| 14087 | #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr) |
| 14088 | #define TOPPTR(ss,ix) ((ss)[ix].any_ptr) |
| 14089 | #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr) |
| 14090 | #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr) |
| 14091 | #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr) |
| 14092 | #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr) |
| 14093 | |
| 14094 | /* XXXXX todo */ |
| 14095 | #define pv_dup_inc(p) SAVEPV(p) |
| 14096 | #define pv_dup(p) SAVEPV(p) |
| 14097 | #define svp_dup_inc(p,pp) any_dup(p,pp) |
| 14098 | |
| 14099 | /* map any object to the new equivent - either something in the |
| 14100 | * ptr table, or something in the interpreter structure |
| 14101 | */ |
| 14102 | |
| 14103 | void * |
| 14104 | Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl) |
| 14105 | { |
| 14106 | void *ret; |
| 14107 | |
| 14108 | PERL_ARGS_ASSERT_ANY_DUP; |
| 14109 | |
| 14110 | if (!v) |
| 14111 | return (void*)NULL; |
| 14112 | |
| 14113 | /* look for it in the table first */ |
| 14114 | ret = ptr_table_fetch(PL_ptr_table, v); |
| 14115 | if (ret) |
| 14116 | return ret; |
| 14117 | |
| 14118 | /* see if it is part of the interpreter structure */ |
| 14119 | if (v >= (void*)proto_perl && v < (void*)(proto_perl+1)) |
| 14120 | ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl)); |
| 14121 | else { |
| 14122 | ret = v; |
| 14123 | } |
| 14124 | |
| 14125 | return ret; |
| 14126 | } |
| 14127 | |
| 14128 | /* duplicate the save stack */ |
| 14129 | |
| 14130 | ANY * |
| 14131 | Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param) |
| 14132 | { |
| 14133 | dVAR; |
| 14134 | ANY * const ss = proto_perl->Isavestack; |
| 14135 | const I32 max = proto_perl->Isavestack_max; |
| 14136 | I32 ix = proto_perl->Isavestack_ix; |
| 14137 | ANY *nss; |
| 14138 | const SV *sv; |
| 14139 | const GV *gv; |
| 14140 | const AV *av; |
| 14141 | const HV *hv; |
| 14142 | void* ptr; |
| 14143 | int intval; |
| 14144 | long longval; |
| 14145 | GP *gp; |
| 14146 | IV iv; |
| 14147 | I32 i; |
| 14148 | char *c = NULL; |
| 14149 | void (*dptr) (void*); |
| 14150 | void (*dxptr) (pTHX_ void*); |
| 14151 | |
| 14152 | PERL_ARGS_ASSERT_SS_DUP; |
| 14153 | |
| 14154 | Newxz(nss, max, ANY); |
| 14155 | |
| 14156 | while (ix > 0) { |
| 14157 | const UV uv = POPUV(ss,ix); |
| 14158 | const U8 type = (U8)uv & SAVE_MASK; |
| 14159 | |
| 14160 | TOPUV(nss,ix) = uv; |
| 14161 | switch (type) { |
| 14162 | case SAVEt_CLEARSV: |
| 14163 | case SAVEt_CLEARPADRANGE: |
| 14164 | break; |
| 14165 | case SAVEt_HELEM: /* hash element */ |
| 14166 | case SAVEt_SV: /* scalar reference */ |
| 14167 | sv = (const SV *)POPPTR(ss,ix); |
| 14168 | TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param)); |
| 14169 | /* FALLTHROUGH */ |
| 14170 | case SAVEt_ITEM: /* normal string */ |
| 14171 | case SAVEt_GVSV: /* scalar slot in GV */ |
| 14172 | sv = (const SV *)POPPTR(ss,ix); |
| 14173 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14174 | if (type == SAVEt_SV) |
| 14175 | break; |
| 14176 | /* FALLTHROUGH */ |
| 14177 | case SAVEt_FREESV: |
| 14178 | case SAVEt_MORTALIZESV: |
| 14179 | case SAVEt_READONLY_OFF: |
| 14180 | sv = (const SV *)POPPTR(ss,ix); |
| 14181 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14182 | break; |
| 14183 | case SAVEt_FREEPADNAME: |
| 14184 | ptr = POPPTR(ss,ix); |
| 14185 | TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param); |
| 14186 | PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++; |
| 14187 | break; |
| 14188 | case SAVEt_SHARED_PVREF: /* char* in shared space */ |
| 14189 | c = (char*)POPPTR(ss,ix); |
| 14190 | TOPPTR(nss,ix) = savesharedpv(c); |
| 14191 | ptr = POPPTR(ss,ix); |
| 14192 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14193 | break; |
| 14194 | case SAVEt_GENERIC_SVREF: /* generic sv */ |
| 14195 | case SAVEt_SVREF: /* scalar reference */ |
| 14196 | sv = (const SV *)POPPTR(ss,ix); |
| 14197 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14198 | if (type == SAVEt_SVREF) |
| 14199 | SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix)); |
| 14200 | ptr = POPPTR(ss,ix); |
| 14201 | TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */ |
| 14202 | break; |
| 14203 | case SAVEt_GVSLOT: /* any slot in GV */ |
| 14204 | sv = (const SV *)POPPTR(ss,ix); |
| 14205 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14206 | ptr = POPPTR(ss,ix); |
| 14207 | TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */ |
| 14208 | sv = (const SV *)POPPTR(ss,ix); |
| 14209 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14210 | break; |
| 14211 | case SAVEt_HV: /* hash reference */ |
| 14212 | case SAVEt_AV: /* array reference */ |
| 14213 | sv = (const SV *) POPPTR(ss,ix); |
| 14214 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14215 | /* FALLTHROUGH */ |
| 14216 | case SAVEt_COMPPAD: |
| 14217 | case SAVEt_NSTAB: |
| 14218 | sv = (const SV *) POPPTR(ss,ix); |
| 14219 | TOPPTR(nss,ix) = sv_dup(sv, param); |
| 14220 | break; |
| 14221 | case SAVEt_INT: /* int reference */ |
| 14222 | ptr = POPPTR(ss,ix); |
| 14223 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14224 | intval = (int)POPINT(ss,ix); |
| 14225 | TOPINT(nss,ix) = intval; |
| 14226 | break; |
| 14227 | case SAVEt_LONG: /* long reference */ |
| 14228 | ptr = POPPTR(ss,ix); |
| 14229 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14230 | longval = (long)POPLONG(ss,ix); |
| 14231 | TOPLONG(nss,ix) = longval; |
| 14232 | break; |
| 14233 | case SAVEt_I32: /* I32 reference */ |
| 14234 | ptr = POPPTR(ss,ix); |
| 14235 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14236 | i = POPINT(ss,ix); |
| 14237 | TOPINT(nss,ix) = i; |
| 14238 | break; |
| 14239 | case SAVEt_IV: /* IV reference */ |
| 14240 | case SAVEt_STRLEN: /* STRLEN/size_t ref */ |
| 14241 | ptr = POPPTR(ss,ix); |
| 14242 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14243 | iv = POPIV(ss,ix); |
| 14244 | TOPIV(nss,ix) = iv; |
| 14245 | break; |
| 14246 | case SAVEt_HPTR: /* HV* reference */ |
| 14247 | case SAVEt_APTR: /* AV* reference */ |
| 14248 | case SAVEt_SPTR: /* SV* reference */ |
| 14249 | ptr = POPPTR(ss,ix); |
| 14250 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14251 | sv = (const SV *)POPPTR(ss,ix); |
| 14252 | TOPPTR(nss,ix) = sv_dup(sv, param); |
| 14253 | break; |
| 14254 | case SAVEt_VPTR: /* random* reference */ |
| 14255 | ptr = POPPTR(ss,ix); |
| 14256 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14257 | /* FALLTHROUGH */ |
| 14258 | case SAVEt_INT_SMALL: |
| 14259 | case SAVEt_I32_SMALL: |
| 14260 | case SAVEt_I16: /* I16 reference */ |
| 14261 | case SAVEt_I8: /* I8 reference */ |
| 14262 | case SAVEt_BOOL: |
| 14263 | ptr = POPPTR(ss,ix); |
| 14264 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14265 | break; |
| 14266 | case SAVEt_GENERIC_PVREF: /* generic char* */ |
| 14267 | case SAVEt_PPTR: /* char* reference */ |
| 14268 | ptr = POPPTR(ss,ix); |
| 14269 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14270 | c = (char*)POPPTR(ss,ix); |
| 14271 | TOPPTR(nss,ix) = pv_dup(c); |
| 14272 | break; |
| 14273 | case SAVEt_GP: /* scalar reference */ |
| 14274 | gp = (GP*)POPPTR(ss,ix); |
| 14275 | TOPPTR(nss,ix) = gp = gp_dup(gp, param); |
| 14276 | (void)GpREFCNT_inc(gp); |
| 14277 | gv = (const GV *)POPPTR(ss,ix); |
| 14278 | TOPPTR(nss,ix) = gv_dup_inc(gv, param); |
| 14279 | break; |
| 14280 | case SAVEt_FREEOP: |
| 14281 | ptr = POPPTR(ss,ix); |
| 14282 | if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) { |
| 14283 | /* these are assumed to be refcounted properly */ |
| 14284 | OP *o; |
| 14285 | switch (((OP*)ptr)->op_type) { |
| 14286 | case OP_LEAVESUB: |
| 14287 | case OP_LEAVESUBLV: |
| 14288 | case OP_LEAVEEVAL: |
| 14289 | case OP_LEAVE: |
| 14290 | case OP_SCOPE: |
| 14291 | case OP_LEAVEWRITE: |
| 14292 | TOPPTR(nss,ix) = ptr; |
| 14293 | o = (OP*)ptr; |
| 14294 | OP_REFCNT_LOCK; |
| 14295 | (void) OpREFCNT_inc(o); |
| 14296 | OP_REFCNT_UNLOCK; |
| 14297 | break; |
| 14298 | default: |
| 14299 | TOPPTR(nss,ix) = NULL; |
| 14300 | break; |
| 14301 | } |
| 14302 | } |
| 14303 | else |
| 14304 | TOPPTR(nss,ix) = NULL; |
| 14305 | break; |
| 14306 | case SAVEt_FREECOPHH: |
| 14307 | ptr = POPPTR(ss,ix); |
| 14308 | TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr); |
| 14309 | break; |
| 14310 | case SAVEt_ADELETE: |
| 14311 | av = (const AV *)POPPTR(ss,ix); |
| 14312 | TOPPTR(nss,ix) = av_dup_inc(av, param); |
| 14313 | i = POPINT(ss,ix); |
| 14314 | TOPINT(nss,ix) = i; |
| 14315 | break; |
| 14316 | case SAVEt_DELETE: |
| 14317 | hv = (const HV *)POPPTR(ss,ix); |
| 14318 | TOPPTR(nss,ix) = hv_dup_inc(hv, param); |
| 14319 | i = POPINT(ss,ix); |
| 14320 | TOPINT(nss,ix) = i; |
| 14321 | /* FALLTHROUGH */ |
| 14322 | case SAVEt_FREEPV: |
| 14323 | c = (char*)POPPTR(ss,ix); |
| 14324 | TOPPTR(nss,ix) = pv_dup_inc(c); |
| 14325 | break; |
| 14326 | case SAVEt_STACK_POS: /* Position on Perl stack */ |
| 14327 | i = POPINT(ss,ix); |
| 14328 | TOPINT(nss,ix) = i; |
| 14329 | break; |
| 14330 | case SAVEt_DESTRUCTOR: |
| 14331 | ptr = POPPTR(ss,ix); |
| 14332 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */ |
| 14333 | dptr = POPDPTR(ss,ix); |
| 14334 | TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*), |
| 14335 | any_dup(FPTR2DPTR(void *, dptr), |
| 14336 | proto_perl)); |
| 14337 | break; |
| 14338 | case SAVEt_DESTRUCTOR_X: |
| 14339 | ptr = POPPTR(ss,ix); |
| 14340 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */ |
| 14341 | dxptr = POPDXPTR(ss,ix); |
| 14342 | TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*), |
| 14343 | any_dup(FPTR2DPTR(void *, dxptr), |
| 14344 | proto_perl)); |
| 14345 | break; |
| 14346 | case SAVEt_REGCONTEXT: |
| 14347 | case SAVEt_ALLOC: |
| 14348 | ix -= uv >> SAVE_TIGHT_SHIFT; |
| 14349 | break; |
| 14350 | case SAVEt_AELEM: /* array element */ |
| 14351 | sv = (const SV *)POPPTR(ss,ix); |
| 14352 | TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param)); |
| 14353 | i = POPINT(ss,ix); |
| 14354 | TOPINT(nss,ix) = i; |
| 14355 | av = (const AV *)POPPTR(ss,ix); |
| 14356 | TOPPTR(nss,ix) = av_dup_inc(av, param); |
| 14357 | break; |
| 14358 | case SAVEt_OP: |
| 14359 | ptr = POPPTR(ss,ix); |
| 14360 | TOPPTR(nss,ix) = ptr; |
| 14361 | break; |
| 14362 | case SAVEt_HINTS: |
| 14363 | ptr = POPPTR(ss,ix); |
| 14364 | ptr = cophh_copy((COPHH*)ptr); |
| 14365 | TOPPTR(nss,ix) = ptr; |
| 14366 | i = POPINT(ss,ix); |
| 14367 | TOPINT(nss,ix) = i; |
| 14368 | if (i & HINT_LOCALIZE_HH) { |
| 14369 | hv = (const HV *)POPPTR(ss,ix); |
| 14370 | TOPPTR(nss,ix) = hv_dup_inc(hv, param); |
| 14371 | } |
| 14372 | break; |
| 14373 | case SAVEt_PADSV_AND_MORTALIZE: |
| 14374 | longval = (long)POPLONG(ss,ix); |
| 14375 | TOPLONG(nss,ix) = longval; |
| 14376 | ptr = POPPTR(ss,ix); |
| 14377 | TOPPTR(nss,ix) = any_dup(ptr, proto_perl); |
| 14378 | sv = (const SV *)POPPTR(ss,ix); |
| 14379 | TOPPTR(nss,ix) = sv_dup_inc(sv, param); |
| 14380 | break; |
| 14381 | case SAVEt_SET_SVFLAGS: |
| 14382 | i = POPINT(ss,ix); |
| 14383 | TOPINT(nss,ix) = i; |
| 14384 | i = POPINT(ss,ix); |
| 14385 | TOPINT(nss,ix) = i; |
| 14386 | sv = (const SV *)POPPTR(ss,ix); |
| 14387 | TOPPTR(nss,ix) = sv_dup(sv, param); |
| 14388 | break; |
| 14389 | case SAVEt_COMPILE_WARNINGS: |
| 14390 | ptr = POPPTR(ss,ix); |
| 14391 | TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr); |
| 14392 | break; |
| 14393 | case SAVEt_PARSER: |
| 14394 | ptr = POPPTR(ss,ix); |
| 14395 | TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param); |
| 14396 | break; |
| 14397 | default: |
| 14398 | Perl_croak(aTHX_ |
| 14399 | "panic: ss_dup inconsistency (%"IVdf")", (IV) type); |
| 14400 | } |
| 14401 | } |
| 14402 | |
| 14403 | return nss; |
| 14404 | } |
| 14405 | |
| 14406 | |
| 14407 | /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE |
| 14408 | * flag to the result. This is done for each stash before cloning starts, |
| 14409 | * so we know which stashes want their objects cloned */ |
| 14410 | |
| 14411 | static void |
| 14412 | do_mark_cloneable_stash(pTHX_ SV *const sv) |
| 14413 | { |
| 14414 | const HEK * const hvname = HvNAME_HEK((const HV *)sv); |
| 14415 | if (hvname) { |
| 14416 | GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0); |
| 14417 | SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */ |
| 14418 | if (cloner && GvCV(cloner)) { |
| 14419 | dSP; |
| 14420 | UV status; |
| 14421 | |
| 14422 | ENTER; |
| 14423 | SAVETMPS; |
| 14424 | PUSHMARK(SP); |
| 14425 | mXPUSHs(newSVhek(hvname)); |
| 14426 | PUTBACK; |
| 14427 | call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR); |
| 14428 | SPAGAIN; |
| 14429 | status = POPu; |
| 14430 | PUTBACK; |
| 14431 | FREETMPS; |
| 14432 | LEAVE; |
| 14433 | if (status) |
| 14434 | SvFLAGS(sv) &= ~SVphv_CLONEABLE; |
| 14435 | } |
| 14436 | } |
| 14437 | } |
| 14438 | |
| 14439 | |
| 14440 | |
| 14441 | /* |
| 14442 | =for apidoc perl_clone |
| 14443 | |
| 14444 | Create and return a new interpreter by cloning the current one. |
| 14445 | |
| 14446 | C<perl_clone> takes these flags as parameters: |
| 14447 | |
| 14448 | C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also, |
| 14449 | without it we only clone the data and zero the stacks, |
| 14450 | with it we copy the stacks and the new perl interpreter is |
| 14451 | ready to run at the exact same point as the previous one. |
| 14452 | The pseudo-fork code uses C<COPY_STACKS> while the |
| 14453 | threads->create doesn't. |
| 14454 | |
| 14455 | C<CLONEf_KEEP_PTR_TABLE> - |
| 14456 | C<perl_clone> keeps a ptr_table with the pointer of the old |
| 14457 | variable as a key and the new variable as a value, |
| 14458 | this allows it to check if something has been cloned and not |
| 14459 | clone it again but rather just use the value and increase the |
| 14460 | refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill |
| 14461 | the ptr_table using the function |
| 14462 | C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>, |
| 14463 | reason to keep it around is if you want to dup some of your own |
| 14464 | variable who are outside the graph perl scans, an example of this |
| 14465 | code is in F<threads.xs> create. |
| 14466 | |
| 14467 | C<CLONEf_CLONE_HOST> - |
| 14468 | This is a win32 thing, it is ignored on unix, it tells perls |
| 14469 | win32host code (which is c++) to clone itself, this is needed on |
| 14470 | win32 if you want to run two threads at the same time, |
| 14471 | if you just want to do some stuff in a separate perl interpreter |
| 14472 | and then throw it away and return to the original one, |
| 14473 | you don't need to do anything. |
| 14474 | |
| 14475 | =cut |
| 14476 | */ |
| 14477 | |
| 14478 | /* XXX the above needs expanding by someone who actually understands it ! */ |
| 14479 | EXTERN_C PerlInterpreter * |
| 14480 | perl_clone_host(PerlInterpreter* proto_perl, UV flags); |
| 14481 | |
| 14482 | PerlInterpreter * |
| 14483 | perl_clone(PerlInterpreter *proto_perl, UV flags) |
| 14484 | { |
| 14485 | dVAR; |
| 14486 | #ifdef PERL_IMPLICIT_SYS |
| 14487 | |
| 14488 | PERL_ARGS_ASSERT_PERL_CLONE; |
| 14489 | |
| 14490 | /* perlhost.h so we need to call into it |
| 14491 | to clone the host, CPerlHost should have a c interface, sky */ |
| 14492 | |
| 14493 | #ifndef __amigaos4__ |
| 14494 | if (flags & CLONEf_CLONE_HOST) { |
| 14495 | return perl_clone_host(proto_perl,flags); |
| 14496 | } |
| 14497 | #endif |
| 14498 | return perl_clone_using(proto_perl, flags, |
| 14499 | proto_perl->IMem, |
| 14500 | proto_perl->IMemShared, |
| 14501 | proto_perl->IMemParse, |
| 14502 | proto_perl->IEnv, |
| 14503 | proto_perl->IStdIO, |
| 14504 | proto_perl->ILIO, |
| 14505 | proto_perl->IDir, |
| 14506 | proto_perl->ISock, |
| 14507 | proto_perl->IProc); |
| 14508 | } |
| 14509 | |
| 14510 | PerlInterpreter * |
| 14511 | perl_clone_using(PerlInterpreter *proto_perl, UV flags, |
| 14512 | struct IPerlMem* ipM, struct IPerlMem* ipMS, |
| 14513 | struct IPerlMem* ipMP, struct IPerlEnv* ipE, |
| 14514 | struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO, |
| 14515 | struct IPerlDir* ipD, struct IPerlSock* ipS, |
| 14516 | struct IPerlProc* ipP) |
| 14517 | { |
| 14518 | /* XXX many of the string copies here can be optimized if they're |
| 14519 | * constants; they need to be allocated as common memory and just |
| 14520 | * their pointers copied. */ |
| 14521 | |
| 14522 | IV i; |
| 14523 | CLONE_PARAMS clone_params; |
| 14524 | CLONE_PARAMS* const param = &clone_params; |
| 14525 | |
| 14526 | PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter)); |
| 14527 | |
| 14528 | PERL_ARGS_ASSERT_PERL_CLONE_USING; |
| 14529 | #else /* !PERL_IMPLICIT_SYS */ |
| 14530 | IV i; |
| 14531 | CLONE_PARAMS clone_params; |
| 14532 | CLONE_PARAMS* param = &clone_params; |
| 14533 | PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter)); |
| 14534 | |
| 14535 | PERL_ARGS_ASSERT_PERL_CLONE; |
| 14536 | #endif /* PERL_IMPLICIT_SYS */ |
| 14537 | |
| 14538 | /* for each stash, determine whether its objects should be cloned */ |
| 14539 | S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK); |
| 14540 | PERL_SET_THX(my_perl); |
| 14541 | |
| 14542 | #ifdef DEBUGGING |
| 14543 | PoisonNew(my_perl, 1, PerlInterpreter); |
| 14544 | PL_op = NULL; |
| 14545 | PL_curcop = NULL; |
| 14546 | PL_defstash = NULL; /* may be used by perl malloc() */ |
| 14547 | PL_markstack = 0; |
| 14548 | PL_scopestack = 0; |
| 14549 | PL_scopestack_name = 0; |
| 14550 | PL_savestack = 0; |
| 14551 | PL_savestack_ix = 0; |
| 14552 | PL_savestack_max = -1; |
| 14553 | PL_sig_pending = 0; |
| 14554 | PL_parser = NULL; |
| 14555 | Zero(&PL_debug_pad, 1, struct perl_debug_pad); |
| 14556 | Zero(&PL_padname_undef, 1, PADNAME); |
| 14557 | Zero(&PL_padname_const, 1, PADNAME); |
| 14558 | # ifdef DEBUG_LEAKING_SCALARS |
| 14559 | PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000; |
| 14560 | # endif |
| 14561 | # ifdef PERL_TRACE_OPS |
| 14562 | Zero(PL_op_exec_cnt, OP_max+2, UV); |
| 14563 | # endif |
| 14564 | #else /* !DEBUGGING */ |
| 14565 | Zero(my_perl, 1, PerlInterpreter); |
| 14566 | #endif /* DEBUGGING */ |
| 14567 | |
| 14568 | #ifdef PERL_IMPLICIT_SYS |
| 14569 | /* host pointers */ |
| 14570 | PL_Mem = ipM; |
| 14571 | PL_MemShared = ipMS; |
| 14572 | PL_MemParse = ipMP; |
| 14573 | PL_Env = ipE; |
| 14574 | PL_StdIO = ipStd; |
| 14575 | PL_LIO = ipLIO; |
| 14576 | PL_Dir = ipD; |
| 14577 | PL_Sock = ipS; |
| 14578 | PL_Proc = ipP; |
| 14579 | #endif /* PERL_IMPLICIT_SYS */ |
| 14580 | |
| 14581 | |
| 14582 | param->flags = flags; |
| 14583 | /* Nothing in the core code uses this, but we make it available to |
| 14584 | extensions (using mg_dup). */ |
| 14585 | param->proto_perl = proto_perl; |
| 14586 | /* Likely nothing will use this, but it is initialised to be consistent |
| 14587 | with Perl_clone_params_new(). */ |
| 14588 | param->new_perl = my_perl; |
| 14589 | param->unreferenced = NULL; |
| 14590 | |
| 14591 | |
| 14592 | INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl); |
| 14593 | |
| 14594 | PL_body_arenas = NULL; |
| 14595 | Zero(&PL_body_roots, 1, PL_body_roots); |
| 14596 | |
| 14597 | PL_sv_count = 0; |
| 14598 | PL_sv_root = NULL; |
| 14599 | PL_sv_arenaroot = NULL; |
| 14600 | |
| 14601 | PL_debug = proto_perl->Idebug; |
| 14602 | |
| 14603 | /* dbargs array probably holds garbage */ |
| 14604 | PL_dbargs = NULL; |
| 14605 | |
| 14606 | PL_compiling = proto_perl->Icompiling; |
| 14607 | |
| 14608 | /* pseudo environmental stuff */ |
| 14609 | PL_origargc = proto_perl->Iorigargc; |
| 14610 | PL_origargv = proto_perl->Iorigargv; |
| 14611 | |
| 14612 | #ifndef NO_TAINT_SUPPORT |
| 14613 | /* Set tainting stuff before PerlIO_debug can possibly get called */ |
| 14614 | PL_tainting = proto_perl->Itainting; |
| 14615 | PL_taint_warn = proto_perl->Itaint_warn; |
| 14616 | #else |
| 14617 | PL_tainting = FALSE; |
| 14618 | PL_taint_warn = FALSE; |
| 14619 | #endif |
| 14620 | |
| 14621 | PL_minus_c = proto_perl->Iminus_c; |
| 14622 | |
| 14623 | PL_localpatches = proto_perl->Ilocalpatches; |
| 14624 | PL_splitstr = proto_perl->Isplitstr; |
| 14625 | PL_minus_n = proto_perl->Iminus_n; |
| 14626 | PL_minus_p = proto_perl->Iminus_p; |
| 14627 | PL_minus_l = proto_perl->Iminus_l; |
| 14628 | PL_minus_a = proto_perl->Iminus_a; |
| 14629 | PL_minus_E = proto_perl->Iminus_E; |
| 14630 | PL_minus_F = proto_perl->Iminus_F; |
| 14631 | PL_doswitches = proto_perl->Idoswitches; |
| 14632 | PL_dowarn = proto_perl->Idowarn; |
| 14633 | #ifdef PERL_SAWAMPERSAND |
| 14634 | PL_sawampersand = proto_perl->Isawampersand; |
| 14635 | #endif |
| 14636 | PL_unsafe = proto_perl->Iunsafe; |
| 14637 | PL_perldb = proto_perl->Iperldb; |
| 14638 | PL_perl_destruct_level = proto_perl->Iperl_destruct_level; |
| 14639 | PL_exit_flags = proto_perl->Iexit_flags; |
| 14640 | |
| 14641 | /* XXX time(&PL_basetime) when asked for? */ |
| 14642 | PL_basetime = proto_perl->Ibasetime; |
| 14643 | |
| 14644 | PL_maxsysfd = proto_perl->Imaxsysfd; |
| 14645 | PL_statusvalue = proto_perl->Istatusvalue; |
| 14646 | #ifdef __VMS |
| 14647 | PL_statusvalue_vms = proto_perl->Istatusvalue_vms; |
| 14648 | #else |
| 14649 | PL_statusvalue_posix = proto_perl->Istatusvalue_posix; |
| 14650 | #endif |
| 14651 | |
| 14652 | /* RE engine related */ |
| 14653 | PL_regmatch_slab = NULL; |
| 14654 | PL_reg_curpm = NULL; |
| 14655 | |
| 14656 | PL_sub_generation = proto_perl->Isub_generation; |
| 14657 | |
| 14658 | /* funky return mechanisms */ |
| 14659 | PL_forkprocess = proto_perl->Iforkprocess; |
| 14660 | |
| 14661 | /* internal state */ |
| 14662 | PL_maxo = proto_perl->Imaxo; |
| 14663 | |
| 14664 | PL_main_start = proto_perl->Imain_start; |
| 14665 | PL_eval_root = proto_perl->Ieval_root; |
| 14666 | PL_eval_start = proto_perl->Ieval_start; |
| 14667 | |
| 14668 | PL_filemode = proto_perl->Ifilemode; |
| 14669 | PL_lastfd = proto_perl->Ilastfd; |
| 14670 | PL_oldname = proto_perl->Ioldname; /* XXX not quite right */ |
| 14671 | PL_Argv = NULL; |
| 14672 | PL_Cmd = NULL; |
| 14673 | PL_gensym = proto_perl->Igensym; |
| 14674 | |
| 14675 | PL_laststatval = proto_perl->Ilaststatval; |
| 14676 | PL_laststype = proto_perl->Ilaststype; |
| 14677 | PL_mess_sv = NULL; |
| 14678 | |
| 14679 | PL_profiledata = NULL; |
| 14680 | |
| 14681 | PL_generation = proto_perl->Igeneration; |
| 14682 | |
| 14683 | PL_in_clean_objs = proto_perl->Iin_clean_objs; |
| 14684 | PL_in_clean_all = proto_perl->Iin_clean_all; |
| 14685 | |
| 14686 | PL_delaymagic_uid = proto_perl->Idelaymagic_uid; |
| 14687 | PL_delaymagic_euid = proto_perl->Idelaymagic_euid; |
| 14688 | PL_delaymagic_gid = proto_perl->Idelaymagic_gid; |
| 14689 | PL_delaymagic_egid = proto_perl->Idelaymagic_egid; |
| 14690 | PL_nomemok = proto_perl->Inomemok; |
| 14691 | PL_an = proto_perl->Ian; |
| 14692 | PL_evalseq = proto_perl->Ievalseq; |
| 14693 | PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */ |
| 14694 | PL_origalen = proto_perl->Iorigalen; |
| 14695 | |
| 14696 | PL_sighandlerp = proto_perl->Isighandlerp; |
| 14697 | |
| 14698 | PL_runops = proto_perl->Irunops; |
| 14699 | |
| 14700 | PL_subline = proto_perl->Isubline; |
| 14701 | |
| 14702 | PL_cv_has_eval = proto_perl->Icv_has_eval; |
| 14703 | |
| 14704 | #ifdef FCRYPT |
| 14705 | PL_cryptseen = proto_perl->Icryptseen; |
| 14706 | #endif |
| 14707 | |
| 14708 | #ifdef USE_LOCALE_COLLATE |
| 14709 | PL_collation_ix = proto_perl->Icollation_ix; |
| 14710 | PL_collation_standard = proto_perl->Icollation_standard; |
| 14711 | PL_collxfrm_base = proto_perl->Icollxfrm_base; |
| 14712 | PL_collxfrm_mult = proto_perl->Icollxfrm_mult; |
| 14713 | #endif /* USE_LOCALE_COLLATE */ |
| 14714 | |
| 14715 | #ifdef USE_LOCALE_NUMERIC |
| 14716 | PL_numeric_standard = proto_perl->Inumeric_standard; |
| 14717 | PL_numeric_local = proto_perl->Inumeric_local; |
| 14718 | #endif /* !USE_LOCALE_NUMERIC */ |
| 14719 | |
| 14720 | /* Did the locale setup indicate UTF-8? */ |
| 14721 | PL_utf8locale = proto_perl->Iutf8locale; |
| 14722 | PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale; |
| 14723 | /* Unicode features (see perlrun/-C) */ |
| 14724 | PL_unicode = proto_perl->Iunicode; |
| 14725 | |
| 14726 | /* Pre-5.8 signals control */ |
| 14727 | PL_signals = proto_perl->Isignals; |
| 14728 | |
| 14729 | /* times() ticks per second */ |
| 14730 | PL_clocktick = proto_perl->Iclocktick; |
| 14731 | |
| 14732 | /* Recursion stopper for PerlIO_find_layer */ |
| 14733 | PL_in_load_module = proto_perl->Iin_load_module; |
| 14734 | |
| 14735 | /* sort() routine */ |
| 14736 | PL_sort_RealCmp = proto_perl->Isort_RealCmp; |
| 14737 | |
| 14738 | /* Not really needed/useful since the reenrant_retint is "volatile", |
| 14739 | * but do it for consistency's sake. */ |
| 14740 | PL_reentrant_retint = proto_perl->Ireentrant_retint; |
| 14741 | |
| 14742 | /* Hooks to shared SVs and locks. */ |
| 14743 | PL_sharehook = proto_perl->Isharehook; |
| 14744 | PL_lockhook = proto_perl->Ilockhook; |
| 14745 | PL_unlockhook = proto_perl->Iunlockhook; |
| 14746 | PL_threadhook = proto_perl->Ithreadhook; |
| 14747 | PL_destroyhook = proto_perl->Idestroyhook; |
| 14748 | PL_signalhook = proto_perl->Isignalhook; |
| 14749 | |
| 14750 | PL_globhook = proto_perl->Iglobhook; |
| 14751 | |
| 14752 | /* swatch cache */ |
| 14753 | PL_last_swash_hv = NULL; /* reinits on demand */ |
| 14754 | PL_last_swash_klen = 0; |
| 14755 | PL_last_swash_key[0]= '\0'; |
| 14756 | PL_last_swash_tmps = (U8*)NULL; |
| 14757 | PL_last_swash_slen = 0; |
| 14758 | |
| 14759 | PL_srand_called = proto_perl->Isrand_called; |
| 14760 | Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE); |
| 14761 | |
| 14762 | if (flags & CLONEf_COPY_STACKS) { |
| 14763 | /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */ |
| 14764 | PL_tmps_ix = proto_perl->Itmps_ix; |
| 14765 | PL_tmps_max = proto_perl->Itmps_max; |
| 14766 | PL_tmps_floor = proto_perl->Itmps_floor; |
| 14767 | |
| 14768 | /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix] |
| 14769 | * NOTE: unlike the others! */ |
| 14770 | PL_scopestack_ix = proto_perl->Iscopestack_ix; |
| 14771 | PL_scopestack_max = proto_perl->Iscopestack_max; |
| 14772 | |
| 14773 | /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix] |
| 14774 | * NOTE: unlike the others! */ |
| 14775 | PL_savestack_ix = proto_perl->Isavestack_ix; |
| 14776 | PL_savestack_max = proto_perl->Isavestack_max; |
| 14777 | } |
| 14778 | |
| 14779 | PL_start_env = proto_perl->Istart_env; /* XXXXXX */ |
| 14780 | PL_top_env = &PL_start_env; |
| 14781 | |
| 14782 | PL_op = proto_perl->Iop; |
| 14783 | |
| 14784 | PL_Sv = NULL; |
| 14785 | PL_Xpv = (XPV*)NULL; |
| 14786 | my_perl->Ina = proto_perl->Ina; |
| 14787 | |
| 14788 | PL_statbuf = proto_perl->Istatbuf; |
| 14789 | PL_statcache = proto_perl->Istatcache; |
| 14790 | |
| 14791 | #ifndef NO_TAINT_SUPPORT |
| 14792 | PL_tainted = proto_perl->Itainted; |
| 14793 | #else |
| 14794 | PL_tainted = FALSE; |
| 14795 | #endif |
| 14796 | PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */ |
| 14797 | |
| 14798 | PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */ |
| 14799 | |
| 14800 | PL_restartjmpenv = proto_perl->Irestartjmpenv; |
| 14801 | PL_restartop = proto_perl->Irestartop; |
| 14802 | PL_in_eval = proto_perl->Iin_eval; |
| 14803 | PL_delaymagic = proto_perl->Idelaymagic; |
| 14804 | PL_phase = proto_perl->Iphase; |
| 14805 | PL_localizing = proto_perl->Ilocalizing; |
| 14806 | |
| 14807 | PL_hv_fetch_ent_mh = NULL; |
| 14808 | PL_modcount = proto_perl->Imodcount; |
| 14809 | PL_lastgotoprobe = NULL; |
| 14810 | PL_dumpindent = proto_perl->Idumpindent; |
| 14811 | |
| 14812 | PL_efloatbuf = NULL; /* reinits on demand */ |
| 14813 | PL_efloatsize = 0; /* reinits on demand */ |
| 14814 | |
| 14815 | /* regex stuff */ |
| 14816 | |
| 14817 | PL_colorset = 0; /* reinits PL_colors[] */ |
| 14818 | /*PL_colors[6] = {0,0,0,0,0,0};*/ |
| 14819 | |
| 14820 | /* Pluggable optimizer */ |
| 14821 | PL_peepp = proto_perl->Ipeepp; |
| 14822 | PL_rpeepp = proto_perl->Irpeepp; |
| 14823 | /* op_free() hook */ |
| 14824 | PL_opfreehook = proto_perl->Iopfreehook; |
| 14825 | |
| 14826 | #ifdef USE_REENTRANT_API |
| 14827 | /* XXX: things like -Dm will segfault here in perlio, but doing |
| 14828 | * PERL_SET_CONTEXT(proto_perl); |
| 14829 | * breaks too many other things |
| 14830 | */ |
| 14831 | Perl_reentrant_init(aTHX); |
| 14832 | #endif |
| 14833 | |
| 14834 | /* create SV map for pointer relocation */ |
| 14835 | PL_ptr_table = ptr_table_new(); |
| 14836 | |
| 14837 | /* initialize these special pointers as early as possible */ |
| 14838 | init_constants(); |
| 14839 | ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef); |
| 14840 | ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no); |
| 14841 | ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes); |
| 14842 | ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const, |
| 14843 | &PL_padname_const); |
| 14844 | |
| 14845 | /* create (a non-shared!) shared string table */ |
| 14846 | PL_strtab = newHV(); |
| 14847 | HvSHAREKEYS_off(PL_strtab); |
| 14848 | hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab)); |
| 14849 | ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab); |
| 14850 | |
| 14851 | Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*); |
| 14852 | |
| 14853 | /* This PV will be free'd special way so must set it same way op.c does */ |
| 14854 | PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file); |
| 14855 | ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file); |
| 14856 | |
| 14857 | ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling); |
| 14858 | PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings); |
| 14859 | CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling))); |
| 14860 | PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl); |
| 14861 | |
| 14862 | param->stashes = newAV(); /* Setup array of objects to call clone on */ |
| 14863 | /* This makes no difference to the implementation, as it always pushes |
| 14864 | and shifts pointers to other SVs without changing their reference |
| 14865 | count, with the array becoming empty before it is freed. However, it |
| 14866 | makes it conceptually clear what is going on, and will avoid some |
| 14867 | work inside av.c, filling slots between AvFILL() and AvMAX() with |
| 14868 | &PL_sv_undef, and SvREFCNT_dec()ing those. */ |
| 14869 | AvREAL_off(param->stashes); |
| 14870 | |
| 14871 | if (!(flags & CLONEf_COPY_STACKS)) { |
| 14872 | param->unreferenced = newAV(); |
| 14873 | } |
| 14874 | |
| 14875 | #ifdef PERLIO_LAYERS |
| 14876 | /* Clone PerlIO tables as soon as we can handle general xx_dup() */ |
| 14877 | PerlIO_clone(aTHX_ proto_perl, param); |
| 14878 | #endif |
| 14879 | |
| 14880 | PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param); |
| 14881 | PL_incgv = gv_dup_inc(proto_perl->Iincgv, param); |
| 14882 | PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param); |
| 14883 | PL_origfilename = SAVEPV(proto_perl->Iorigfilename); |
| 14884 | PL_xsubfilename = proto_perl->Ixsubfilename; |
| 14885 | PL_diehook = sv_dup_inc(proto_perl->Idiehook, param); |
| 14886 | PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param); |
| 14887 | |
| 14888 | /* switches */ |
| 14889 | PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param); |
| 14890 | PL_inplace = SAVEPV(proto_perl->Iinplace); |
| 14891 | PL_e_script = sv_dup_inc(proto_perl->Ie_script, param); |
| 14892 | |
| 14893 | /* magical thingies */ |
| 14894 | |
| 14895 | PL_encoding = sv_dup(proto_perl->Iencoding, param); |
| 14896 | PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param); |
| 14897 | |
| 14898 | sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */ |
| 14899 | sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */ |
| 14900 | sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */ |
| 14901 | |
| 14902 | |
| 14903 | /* Clone the regex array */ |
| 14904 | /* ORANGE FIXME for plugins, probably in the SV dup code. |
| 14905 | newSViv(PTR2IV(CALLREGDUPE( |
| 14906 | INT2PTR(REGEXP *, SvIVX(regex)), param)))) |
| 14907 | */ |
| 14908 | PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param); |
| 14909 | PL_regex_pad = AvARRAY(PL_regex_padav); |
| 14910 | |
| 14911 | PL_stashpadmax = proto_perl->Istashpadmax; |
| 14912 | PL_stashpadix = proto_perl->Istashpadix ; |
| 14913 | Newx(PL_stashpad, PL_stashpadmax, HV *); |
| 14914 | { |
| 14915 | PADOFFSET o = 0; |
| 14916 | for (; o < PL_stashpadmax; ++o) |
| 14917 | PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param); |
| 14918 | } |
| 14919 | |
| 14920 | /* shortcuts to various I/O objects */ |
| 14921 | PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param); |
| 14922 | PL_stdingv = gv_dup(proto_perl->Istdingv, param); |
| 14923 | PL_stderrgv = gv_dup(proto_perl->Istderrgv, param); |
| 14924 | PL_defgv = gv_dup(proto_perl->Idefgv, param); |
| 14925 | PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param); |
| 14926 | PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param); |
| 14927 | PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param); |
| 14928 | |
| 14929 | /* shortcuts to regexp stuff */ |
| 14930 | PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param); |
| 14931 | |
| 14932 | /* shortcuts to misc objects */ |
| 14933 | PL_errgv = gv_dup(proto_perl->Ierrgv, param); |
| 14934 | |
| 14935 | /* shortcuts to debugging objects */ |
| 14936 | PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param); |
| 14937 | PL_DBline = gv_dup_inc(proto_perl->IDBline, param); |
| 14938 | PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param); |
| 14939 | PL_DBsingle = sv_dup(proto_perl->IDBsingle, param); |
| 14940 | PL_DBtrace = sv_dup(proto_perl->IDBtrace, param); |
| 14941 | PL_DBsignal = sv_dup(proto_perl->IDBsignal, param); |
| 14942 | Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV); |
| 14943 | |
| 14944 | /* symbol tables */ |
| 14945 | PL_defstash = hv_dup_inc(proto_perl->Idefstash, param); |
| 14946 | PL_curstash = hv_dup_inc(proto_perl->Icurstash, param); |
| 14947 | PL_debstash = hv_dup(proto_perl->Idebstash, param); |
| 14948 | PL_globalstash = hv_dup(proto_perl->Iglobalstash, param); |
| 14949 | PL_curstname = sv_dup_inc(proto_perl->Icurstname, param); |
| 14950 | |
| 14951 | PL_beginav = av_dup_inc(proto_perl->Ibeginav, param); |
| 14952 | PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param); |
| 14953 | PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param); |
| 14954 | PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param); |
| 14955 | PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param); |
| 14956 | PL_endav = av_dup_inc(proto_perl->Iendav, param); |
| 14957 | PL_checkav = av_dup_inc(proto_perl->Icheckav, param); |
| 14958 | PL_initav = av_dup_inc(proto_perl->Iinitav, param); |
| 14959 | PL_savebegin = proto_perl->Isavebegin; |
| 14960 | |
| 14961 | PL_isarev = hv_dup_inc(proto_perl->Iisarev, param); |
| 14962 | |
| 14963 | /* subprocess state */ |
| 14964 | PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param); |
| 14965 | |
| 14966 | if (proto_perl->Iop_mask) |
| 14967 | PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo); |
| 14968 | else |
| 14969 | PL_op_mask = NULL; |
| 14970 | /* PL_asserting = proto_perl->Iasserting; */ |
| 14971 | |
| 14972 | /* current interpreter roots */ |
| 14973 | PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param); |
| 14974 | OP_REFCNT_LOCK; |
| 14975 | PL_main_root = OpREFCNT_inc(proto_perl->Imain_root); |
| 14976 | OP_REFCNT_UNLOCK; |
| 14977 | |
| 14978 | /* runtime control stuff */ |
| 14979 | PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl); |
| 14980 | |
| 14981 | PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param); |
| 14982 | |
| 14983 | PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param); |
| 14984 | |
| 14985 | /* interpreter atexit processing */ |
| 14986 | PL_exitlistlen = proto_perl->Iexitlistlen; |
| 14987 | if (PL_exitlistlen) { |
| 14988 | Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry); |
| 14989 | Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry); |
| 14990 | } |
| 14991 | else |
| 14992 | PL_exitlist = (PerlExitListEntry*)NULL; |
| 14993 | |
| 14994 | PL_my_cxt_size = proto_perl->Imy_cxt_size; |
| 14995 | if (PL_my_cxt_size) { |
| 14996 | Newx(PL_my_cxt_list, PL_my_cxt_size, void *); |
| 14997 | Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *); |
| 14998 | #ifdef PERL_GLOBAL_STRUCT_PRIVATE |
| 14999 | Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *); |
| 15000 | Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *); |
| 15001 | #endif |
| 15002 | } |
| 15003 | else { |
| 15004 | PL_my_cxt_list = (void**)NULL; |
| 15005 | #ifdef PERL_GLOBAL_STRUCT_PRIVATE |
| 15006 | PL_my_cxt_keys = (const char**)NULL; |
| 15007 | #endif |
| 15008 | } |
| 15009 | PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param); |
| 15010 | PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param); |
| 15011 | PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param); |
| 15012 | PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param); |
| 15013 | |
| 15014 | PL_compcv = cv_dup(proto_perl->Icompcv, param); |
| 15015 | |
| 15016 | PAD_CLONE_VARS(proto_perl, param); |
| 15017 | |
| 15018 | #ifdef HAVE_INTERP_INTERN |
| 15019 | sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern); |
| 15020 | #endif |
| 15021 | |
| 15022 | PL_DBcv = cv_dup(proto_perl->IDBcv, param); |
| 15023 | |
| 15024 | #ifdef PERL_USES_PL_PIDSTATUS |
| 15025 | PL_pidstatus = newHV(); /* XXX flag for cloning? */ |
| 15026 | #endif |
| 15027 | PL_osname = SAVEPV(proto_perl->Iosname); |
| 15028 | PL_parser = parser_dup(proto_perl->Iparser, param); |
| 15029 | |
| 15030 | /* XXX this only works if the saved cop has already been cloned */ |
| 15031 | if (proto_perl->Iparser) { |
| 15032 | PL_parser->saved_curcop = (COP*)any_dup( |
| 15033 | proto_perl->Iparser->saved_curcop, |
| 15034 | proto_perl); |
| 15035 | } |
| 15036 | |
| 15037 | PL_subname = sv_dup_inc(proto_perl->Isubname, param); |
| 15038 | |
| 15039 | #ifdef USE_LOCALE_CTYPE |
| 15040 | /* Should we warn if uses locale? */ |
| 15041 | PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param); |
| 15042 | #endif |
| 15043 | |
| 15044 | #ifdef USE_LOCALE_COLLATE |
| 15045 | PL_collation_name = SAVEPV(proto_perl->Icollation_name); |
| 15046 | #endif /* USE_LOCALE_COLLATE */ |
| 15047 | |
| 15048 | #ifdef USE_LOCALE_NUMERIC |
| 15049 | PL_numeric_name = SAVEPV(proto_perl->Inumeric_name); |
| 15050 | PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param); |
| 15051 | #endif /* !USE_LOCALE_NUMERIC */ |
| 15052 | |
| 15053 | /* Unicode inversion lists */ |
| 15054 | PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param); |
| 15055 | PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param); |
| 15056 | PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param); |
| 15057 | PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param); |
| 15058 | |
| 15059 | PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param); |
| 15060 | PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param); |
| 15061 | |
| 15062 | /* utf8 character class swashes */ |
| 15063 | for (i = 0; i < POSIX_SWASH_COUNT; i++) { |
| 15064 | PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param); |
| 15065 | } |
| 15066 | for (i = 0; i < POSIX_CC_COUNT; i++) { |
| 15067 | PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param); |
| 15068 | } |
| 15069 | PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param); |
| 15070 | PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param); |
| 15071 | PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param); |
| 15072 | PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param); |
| 15073 | PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param); |
| 15074 | PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param); |
| 15075 | PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param); |
| 15076 | PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param); |
| 15077 | PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param); |
| 15078 | PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param); |
| 15079 | PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param); |
| 15080 | PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param); |
| 15081 | PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param); |
| 15082 | PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param); |
| 15083 | PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param); |
| 15084 | PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param); |
| 15085 | PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param); |
| 15086 | |
| 15087 | if (proto_perl->Ipsig_pend) { |
| 15088 | Newxz(PL_psig_pend, SIG_SIZE, int); |
| 15089 | } |
| 15090 | else { |
| 15091 | PL_psig_pend = (int*)NULL; |
| 15092 | } |
| 15093 | |
| 15094 | if (proto_perl->Ipsig_name) { |
| 15095 | Newx(PL_psig_name, 2 * SIG_SIZE, SV*); |
| 15096 | sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE, |
| 15097 | param); |
| 15098 | PL_psig_ptr = PL_psig_name + SIG_SIZE; |
| 15099 | } |
| 15100 | else { |
| 15101 | PL_psig_ptr = (SV**)NULL; |
| 15102 | PL_psig_name = (SV**)NULL; |
| 15103 | } |
| 15104 | |
| 15105 | if (flags & CLONEf_COPY_STACKS) { |
| 15106 | Newx(PL_tmps_stack, PL_tmps_max, SV*); |
| 15107 | sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack, |
| 15108 | PL_tmps_ix+1, param); |
| 15109 | |
| 15110 | /* next PUSHMARK() sets *(PL_markstack_ptr+1) */ |
| 15111 | i = proto_perl->Imarkstack_max - proto_perl->Imarkstack; |
| 15112 | Newxz(PL_markstack, i, I32); |
| 15113 | PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max |
| 15114 | - proto_perl->Imarkstack); |
| 15115 | PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr |
| 15116 | - proto_perl->Imarkstack); |
| 15117 | Copy(proto_perl->Imarkstack, PL_markstack, |
| 15118 | PL_markstack_ptr - PL_markstack + 1, I32); |
| 15119 | |
| 15120 | /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix] |
| 15121 | * NOTE: unlike the others! */ |
| 15122 | Newxz(PL_scopestack, PL_scopestack_max, I32); |
| 15123 | Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32); |
| 15124 | |
| 15125 | #ifdef DEBUGGING |
| 15126 | Newxz(PL_scopestack_name, PL_scopestack_max, const char *); |
| 15127 | Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *); |
| 15128 | #endif |
| 15129 | /* reset stack AV to correct length before its duped via |
| 15130 | * PL_curstackinfo */ |
| 15131 | AvFILLp(proto_perl->Icurstack) = |
| 15132 | proto_perl->Istack_sp - proto_perl->Istack_base; |
| 15133 | |
| 15134 | /* NOTE: si_dup() looks at PL_markstack */ |
| 15135 | PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param); |
| 15136 | |
| 15137 | /* PL_curstack = PL_curstackinfo->si_stack; */ |
| 15138 | PL_curstack = av_dup(proto_perl->Icurstack, param); |
| 15139 | PL_mainstack = av_dup(proto_perl->Imainstack, param); |
| 15140 | |
| 15141 | /* next PUSHs() etc. set *(PL_stack_sp+1) */ |
| 15142 | PL_stack_base = AvARRAY(PL_curstack); |
| 15143 | PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp |
| 15144 | - proto_perl->Istack_base); |
| 15145 | PL_stack_max = PL_stack_base + AvMAX(PL_curstack); |
| 15146 | |
| 15147 | /*Newxz(PL_savestack, PL_savestack_max, ANY);*/ |
| 15148 | PL_savestack = ss_dup(proto_perl, param); |
| 15149 | } |
| 15150 | else { |
| 15151 | init_stacks(); |
| 15152 | ENTER; /* perl_destruct() wants to LEAVE; */ |
| 15153 | } |
| 15154 | |
| 15155 | PL_statgv = gv_dup(proto_perl->Istatgv, param); |
| 15156 | PL_statname = sv_dup_inc(proto_perl->Istatname, param); |
| 15157 | |
| 15158 | PL_rs = sv_dup_inc(proto_perl->Irs, param); |
| 15159 | PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param); |
| 15160 | PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param); |
| 15161 | PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param); |
| 15162 | PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param); |
| 15163 | PL_formtarget = sv_dup(proto_perl->Iformtarget, param); |
| 15164 | |
| 15165 | PL_errors = sv_dup_inc(proto_perl->Ierrors, param); |
| 15166 | |
| 15167 | PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl); |
| 15168 | PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param); |
| 15169 | PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param); |
| 15170 | |
| 15171 | PL_stashcache = newHV(); |
| 15172 | |
| 15173 | PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table, |
| 15174 | proto_perl->Iwatchaddr); |
| 15175 | PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL; |
| 15176 | if (PL_debug && PL_watchaddr) { |
| 15177 | PerlIO_printf(Perl_debug_log, |
| 15178 | "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n", |
| 15179 | PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr), |
| 15180 | PTR2UV(PL_watchok)); |
| 15181 | } |
| 15182 | |
| 15183 | PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param); |
| 15184 | PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param); |
| 15185 | PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param); |
| 15186 | |
| 15187 | /* Call the ->CLONE method, if it exists, for each of the stashes |
| 15188 | identified by sv_dup() above. |
| 15189 | */ |
| 15190 | while(av_tindex(param->stashes) != -1) { |
| 15191 | HV* const stash = MUTABLE_HV(av_shift(param->stashes)); |
| 15192 | GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0); |
| 15193 | if (cloner && GvCV(cloner)) { |
| 15194 | dSP; |
| 15195 | ENTER; |
| 15196 | SAVETMPS; |
| 15197 | PUSHMARK(SP); |
| 15198 | mXPUSHs(newSVhek(HvNAME_HEK(stash))); |
| 15199 | PUTBACK; |
| 15200 | call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD); |
| 15201 | FREETMPS; |
| 15202 | LEAVE; |
| 15203 | } |
| 15204 | } |
| 15205 | |
| 15206 | if (!(flags & CLONEf_KEEP_PTR_TABLE)) { |
| 15207 | ptr_table_free(PL_ptr_table); |
| 15208 | PL_ptr_table = NULL; |
| 15209 | } |
| 15210 | |
| 15211 | if (!(flags & CLONEf_COPY_STACKS)) { |
| 15212 | unreferenced_to_tmp_stack(param->unreferenced); |
| 15213 | } |
| 15214 | |
| 15215 | SvREFCNT_dec(param->stashes); |
| 15216 | |
| 15217 | /* orphaned? eg threads->new inside BEGIN or use */ |
| 15218 | if (PL_compcv && ! SvREFCNT(PL_compcv)) { |
| 15219 | SvREFCNT_inc_simple_void(PL_compcv); |
| 15220 | SAVEFREESV(PL_compcv); |
| 15221 | } |
| 15222 | |
| 15223 | return my_perl; |
| 15224 | } |
| 15225 | |
| 15226 | static void |
| 15227 | S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced) |
| 15228 | { |
| 15229 | PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK; |
| 15230 | |
| 15231 | if (AvFILLp(unreferenced) > -1) { |
| 15232 | SV **svp = AvARRAY(unreferenced); |
| 15233 | SV **const last = svp + AvFILLp(unreferenced); |
| 15234 | SSize_t count = 0; |
| 15235 | |
| 15236 | do { |
| 15237 | if (SvREFCNT(*svp) == 1) |
| 15238 | ++count; |
| 15239 | } while (++svp <= last); |
| 15240 | |
| 15241 | EXTEND_MORTAL(count); |
| 15242 | svp = AvARRAY(unreferenced); |
| 15243 | |
| 15244 | do { |
| 15245 | if (SvREFCNT(*svp) == 1) { |
| 15246 | /* Our reference is the only one to this SV. This means that |
| 15247 | in this thread, the scalar effectively has a 0 reference. |
| 15248 | That doesn't work (cleanup never happens), so donate our |
| 15249 | reference to it onto the save stack. */ |
| 15250 | PL_tmps_stack[++PL_tmps_ix] = *svp; |
| 15251 | } else { |
| 15252 | /* As an optimisation, because we are already walking the |
| 15253 | entire array, instead of above doing either |
| 15254 | SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead |
| 15255 | release our reference to the scalar, so that at the end of |
| 15256 | the array owns zero references to the scalars it happens to |
| 15257 | point to. We are effectively converting the array from |
| 15258 | AvREAL() on to AvREAL() off. This saves the av_clear() |
| 15259 | (triggered by the SvREFCNT_dec(unreferenced) below) from |
| 15260 | walking the array a second time. */ |
| 15261 | SvREFCNT_dec(*svp); |
| 15262 | } |
| 15263 | |
| 15264 | } while (++svp <= last); |
| 15265 | AvREAL_off(unreferenced); |
| 15266 | } |
| 15267 | SvREFCNT_dec_NN(unreferenced); |
| 15268 | } |
| 15269 | |
| 15270 | void |
| 15271 | Perl_clone_params_del(CLONE_PARAMS *param) |
| 15272 | { |
| 15273 | /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT |
| 15274 | happy: */ |
| 15275 | PerlInterpreter *const to = param->new_perl; |
| 15276 | dTHXa(to); |
| 15277 | PerlInterpreter *const was = PERL_GET_THX; |
| 15278 | |
| 15279 | PERL_ARGS_ASSERT_CLONE_PARAMS_DEL; |
| 15280 | |
| 15281 | if (was != to) { |
| 15282 | PERL_SET_THX(to); |
| 15283 | } |
| 15284 | |
| 15285 | SvREFCNT_dec(param->stashes); |
| 15286 | if (param->unreferenced) |
| 15287 | unreferenced_to_tmp_stack(param->unreferenced); |
| 15288 | |
| 15289 | Safefree(param); |
| 15290 | |
| 15291 | if (was != to) { |
| 15292 | PERL_SET_THX(was); |
| 15293 | } |
| 15294 | } |
| 15295 | |
| 15296 | CLONE_PARAMS * |
| 15297 | Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to) |
| 15298 | { |
| 15299 | dVAR; |
| 15300 | /* Need to play this game, as newAV() can call safesysmalloc(), and that |
| 15301 | does a dTHX; to get the context from thread local storage. |
| 15302 | FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to |
| 15303 | a version that passes in my_perl. */ |
| 15304 | PerlInterpreter *const was = PERL_GET_THX; |
| 15305 | CLONE_PARAMS *param; |
| 15306 | |
| 15307 | PERL_ARGS_ASSERT_CLONE_PARAMS_NEW; |
| 15308 | |
| 15309 | if (was != to) { |
| 15310 | PERL_SET_THX(to); |
| 15311 | } |
| 15312 | |
| 15313 | /* Given that we've set the context, we can do this unshared. */ |
| 15314 | Newx(param, 1, CLONE_PARAMS); |
| 15315 | |
| 15316 | param->flags = 0; |
| 15317 | param->proto_perl = from; |
| 15318 | param->new_perl = to; |
| 15319 | param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV); |
| 15320 | AvREAL_off(param->stashes); |
| 15321 | param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV); |
| 15322 | |
| 15323 | if (was != to) { |
| 15324 | PERL_SET_THX(was); |
| 15325 | } |
| 15326 | return param; |
| 15327 | } |
| 15328 | |
| 15329 | #endif /* USE_ITHREADS */ |
| 15330 | |
| 15331 | void |
| 15332 | Perl_init_constants(pTHX) |
| 15333 | { |
| 15334 | SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL; |
| 15335 | SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL; |
| 15336 | SvANY(&PL_sv_undef) = NULL; |
| 15337 | |
| 15338 | SvANY(&PL_sv_no) = new_XPVNV(); |
| 15339 | SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL; |
| 15340 | SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT |
| 15341 | |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK |
| 15342 | |SVp_POK|SVf_POK; |
| 15343 | |
| 15344 | SvANY(&PL_sv_yes) = new_XPVNV(); |
| 15345 | SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL; |
| 15346 | SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT |
| 15347 | |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK |
| 15348 | |SVp_POK|SVf_POK; |
| 15349 | |
| 15350 | SvPV_set(&PL_sv_no, (char*)PL_No); |
| 15351 | SvCUR_set(&PL_sv_no, 0); |
| 15352 | SvLEN_set(&PL_sv_no, 0); |
| 15353 | SvIV_set(&PL_sv_no, 0); |
| 15354 | SvNV_set(&PL_sv_no, 0); |
| 15355 | |
| 15356 | SvPV_set(&PL_sv_yes, (char*)PL_Yes); |
| 15357 | SvCUR_set(&PL_sv_yes, 1); |
| 15358 | SvLEN_set(&PL_sv_yes, 0); |
| 15359 | SvIV_set(&PL_sv_yes, 1); |
| 15360 | SvNV_set(&PL_sv_yes, 1); |
| 15361 | |
| 15362 | PadnamePV(&PL_padname_const) = (char *)PL_No; |
| 15363 | } |
| 15364 | |
| 15365 | /* |
| 15366 | =head1 Unicode Support |
| 15367 | |
| 15368 | =for apidoc sv_recode_to_utf8 |
| 15369 | |
| 15370 | C<encoding> is assumed to be an C<Encode> object, on entry the PV |
| 15371 | of C<sv> is assumed to be octets in that encoding, and C<sv> |
| 15372 | will be converted into Unicode (and UTF-8). |
| 15373 | |
| 15374 | If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding> |
| 15375 | is not a reference, nothing is done to C<sv>. If C<encoding> is not |
| 15376 | an C<Encode::XS> Encoding object, bad things will happen. |
| 15377 | (See F<cpan/Encode/encoding.pm> and L<Encode>.) |
| 15378 | |
| 15379 | The PV of C<sv> is returned. |
| 15380 | |
| 15381 | =cut */ |
| 15382 | |
| 15383 | char * |
| 15384 | Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding) |
| 15385 | { |
| 15386 | PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8; |
| 15387 | |
| 15388 | if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) { |
| 15389 | SV *uni; |
| 15390 | STRLEN len; |
| 15391 | const char *s; |
| 15392 | dSP; |
| 15393 | SV *nsv = sv; |
| 15394 | ENTER; |
| 15395 | PUSHSTACK; |
| 15396 | SAVETMPS; |
| 15397 | if (SvPADTMP(nsv)) { |
| 15398 | nsv = sv_newmortal(); |
| 15399 | SvSetSV_nosteal(nsv, sv); |
| 15400 | } |
| 15401 | save_re_context(); |
| 15402 | PUSHMARK(sp); |
| 15403 | EXTEND(SP, 3); |
| 15404 | PUSHs(encoding); |
| 15405 | PUSHs(nsv); |
| 15406 | /* |
| 15407 | NI-S 2002/07/09 |
| 15408 | Passing sv_yes is wrong - it needs to be or'ed set of constants |
| 15409 | for Encode::XS, while UTf-8 decode (currently) assumes a true value means |
| 15410 | remove converted chars from source. |
| 15411 | |
| 15412 | Both will default the value - let them. |
| 15413 | |
| 15414 | XPUSHs(&PL_sv_yes); |
| 15415 | */ |
| 15416 | PUTBACK; |
| 15417 | call_method("decode", G_SCALAR); |
| 15418 | SPAGAIN; |
| 15419 | uni = POPs; |
| 15420 | PUTBACK; |
| 15421 | s = SvPV_const(uni, len); |
| 15422 | if (s != SvPVX_const(sv)) { |
| 15423 | SvGROW(sv, len + 1); |
| 15424 | Move(s, SvPVX(sv), len + 1, char); |
| 15425 | SvCUR_set(sv, len); |
| 15426 | } |
| 15427 | FREETMPS; |
| 15428 | POPSTACK; |
| 15429 | LEAVE; |
| 15430 | if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) { |
| 15431 | /* clear pos and any utf8 cache */ |
| 15432 | MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global); |
| 15433 | if (mg) |
| 15434 | mg->mg_len = -1; |
| 15435 | if ((mg = mg_find(sv, PERL_MAGIC_utf8))) |
| 15436 | magic_setutf8(sv,mg); /* clear UTF8 cache */ |
| 15437 | } |
| 15438 | SvUTF8_on(sv); |
| 15439 | return SvPVX(sv); |
| 15440 | } |
| 15441 | return SvPOKp(sv) ? SvPVX(sv) : NULL; |
| 15442 | } |
| 15443 | |
| 15444 | /* |
| 15445 | =for apidoc sv_cat_decode |
| 15446 | |
| 15447 | C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is |
| 15448 | assumed to be octets in that encoding and decoding the input starts |
| 15449 | from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be |
| 15450 | concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate |
| 15451 | when the string C<tstr> appears in decoding output or the input ends on |
| 15452 | the PV of C<ssv>. The value which C<offset> points will be modified |
| 15453 | to the last input position on C<ssv>. |
| 15454 | |
| 15455 | Returns TRUE if the terminator was found, else returns FALSE. |
| 15456 | |
| 15457 | =cut */ |
| 15458 | |
| 15459 | bool |
| 15460 | Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding, |
| 15461 | SV *ssv, int *offset, char *tstr, int tlen) |
| 15462 | { |
| 15463 | bool ret = FALSE; |
| 15464 | |
| 15465 | PERL_ARGS_ASSERT_SV_CAT_DECODE; |
| 15466 | |
| 15467 | if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) { |
| 15468 | SV *offsv; |
| 15469 | dSP; |
| 15470 | ENTER; |
| 15471 | SAVETMPS; |
| 15472 | save_re_context(); |
| 15473 | PUSHMARK(sp); |
| 15474 | EXTEND(SP, 6); |
| 15475 | PUSHs(encoding); |
| 15476 | PUSHs(dsv); |
| 15477 | PUSHs(ssv); |
| 15478 | offsv = newSViv(*offset); |
| 15479 | mPUSHs(offsv); |
| 15480 | mPUSHp(tstr, tlen); |
| 15481 | PUTBACK; |
| 15482 | call_method("cat_decode", G_SCALAR); |
| 15483 | SPAGAIN; |
| 15484 | ret = SvTRUE(TOPs); |
| 15485 | *offset = SvIV(offsv); |
| 15486 | PUTBACK; |
| 15487 | FREETMPS; |
| 15488 | LEAVE; |
| 15489 | } |
| 15490 | else |
| 15491 | Perl_croak(aTHX_ "Invalid argument to sv_cat_decode"); |
| 15492 | return ret; |
| 15493 | |
| 15494 | } |
| 15495 | |
| 15496 | /* --------------------------------------------------------------------- |
| 15497 | * |
| 15498 | * support functions for report_uninit() |
| 15499 | */ |
| 15500 | |
| 15501 | /* the maxiumum size of array or hash where we will scan looking |
| 15502 | * for the undefined element that triggered the warning */ |
| 15503 | |
| 15504 | #define FUV_MAX_SEARCH_SIZE 1000 |
| 15505 | |
| 15506 | /* Look for an entry in the hash whose value has the same SV as val; |
| 15507 | * If so, return a mortal copy of the key. */ |
| 15508 | |
| 15509 | STATIC SV* |
| 15510 | S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val) |
| 15511 | { |
| 15512 | dVAR; |
| 15513 | HE **array; |
| 15514 | I32 i; |
| 15515 | |
| 15516 | PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT; |
| 15517 | |
| 15518 | if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) || |
| 15519 | (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE)) |
| 15520 | return NULL; |
| 15521 | |
| 15522 | array = HvARRAY(hv); |
| 15523 | |
| 15524 | for (i=HvMAX(hv); i>=0; i--) { |
| 15525 | HE *entry; |
| 15526 | for (entry = array[i]; entry; entry = HeNEXT(entry)) { |
| 15527 | if (HeVAL(entry) != val) |
| 15528 | continue; |
| 15529 | if ( HeVAL(entry) == &PL_sv_undef || |
| 15530 | HeVAL(entry) == &PL_sv_placeholder) |
| 15531 | continue; |
| 15532 | if (!HeKEY(entry)) |
| 15533 | return NULL; |
| 15534 | if (HeKLEN(entry) == HEf_SVKEY) |
| 15535 | return sv_mortalcopy(HeKEY_sv(entry)); |
| 15536 | return sv_2mortal(newSVhek(HeKEY_hek(entry))); |
| 15537 | } |
| 15538 | } |
| 15539 | return NULL; |
| 15540 | } |
| 15541 | |
| 15542 | /* Look for an entry in the array whose value has the same SV as val; |
| 15543 | * If so, return the index, otherwise return -1. */ |
| 15544 | |
| 15545 | STATIC I32 |
| 15546 | S_find_array_subscript(pTHX_ const AV *const av, const SV *const val) |
| 15547 | { |
| 15548 | PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT; |
| 15549 | |
| 15550 | if (!av || SvMAGICAL(av) || !AvARRAY(av) || |
| 15551 | (AvFILLp(av) > FUV_MAX_SEARCH_SIZE)) |
| 15552 | return -1; |
| 15553 | |
| 15554 | if (val != &PL_sv_undef) { |
| 15555 | SV ** const svp = AvARRAY(av); |
| 15556 | I32 i; |
| 15557 | |
| 15558 | for (i=AvFILLp(av); i>=0; i--) |
| 15559 | if (svp[i] == val) |
| 15560 | return i; |
| 15561 | } |
| 15562 | return -1; |
| 15563 | } |
| 15564 | |
| 15565 | /* varname(): return the name of a variable, optionally with a subscript. |
| 15566 | * If gv is non-zero, use the name of that global, along with gvtype (one |
| 15567 | * of "$", "@", "%"); otherwise use the name of the lexical at pad offset |
| 15568 | * targ. Depending on the value of the subscript_type flag, return: |
| 15569 | */ |
| 15570 | |
| 15571 | #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */ |
| 15572 | #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */ |
| 15573 | #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */ |
| 15574 | #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */ |
| 15575 | |
| 15576 | SV* |
| 15577 | Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ, |
| 15578 | const SV *const keyname, I32 aindex, int subscript_type) |
| 15579 | { |
| 15580 | |
| 15581 | SV * const name = sv_newmortal(); |
| 15582 | if (gv && isGV(gv)) { |
| 15583 | char buffer[2]; |
| 15584 | buffer[0] = gvtype; |
| 15585 | buffer[1] = 0; |
| 15586 | |
| 15587 | /* as gv_fullname4(), but add literal '^' for $^FOO names */ |
| 15588 | |
| 15589 | gv_fullname4(name, gv, buffer, 0); |
| 15590 | |
| 15591 | if ((unsigned int)SvPVX(name)[1] <= 26) { |
| 15592 | buffer[0] = '^'; |
| 15593 | buffer[1] = SvPVX(name)[1] + 'A' - 1; |
| 15594 | |
| 15595 | /* Swap the 1 unprintable control character for the 2 byte pretty |
| 15596 | version - ie substr($name, 1, 1) = $buffer; */ |
| 15597 | sv_insert(name, 1, 1, buffer, 2); |
| 15598 | } |
| 15599 | } |
| 15600 | else { |
| 15601 | CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL); |
| 15602 | PADNAME *sv; |
| 15603 | |
| 15604 | assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM); |
| 15605 | |
| 15606 | if (!cv || !CvPADLIST(cv)) |
| 15607 | return NULL; |
| 15608 | sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ); |
| 15609 | sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv)); |
| 15610 | SvUTF8_on(name); |
| 15611 | } |
| 15612 | |
| 15613 | if (subscript_type == FUV_SUBSCRIPT_HASH) { |
| 15614 | SV * const sv = newSV(0); |
| 15615 | *SvPVX(name) = '$'; |
| 15616 | Perl_sv_catpvf(aTHX_ name, "{%s}", |
| 15617 | pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL, |
| 15618 | PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT )); |
| 15619 | SvREFCNT_dec_NN(sv); |
| 15620 | } |
| 15621 | else if (subscript_type == FUV_SUBSCRIPT_ARRAY) { |
| 15622 | *SvPVX(name) = '$'; |
| 15623 | Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex); |
| 15624 | } |
| 15625 | else if (subscript_type == FUV_SUBSCRIPT_WITHIN) { |
| 15626 | /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */ |
| 15627 | Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0); |
| 15628 | } |
| 15629 | |
| 15630 | return name; |
| 15631 | } |
| 15632 | |
| 15633 | |
| 15634 | /* |
| 15635 | =for apidoc find_uninit_var |
| 15636 | |
| 15637 | Find the name of the undefined variable (if any) that caused the operator |
| 15638 | to issue a "Use of uninitialized value" warning. |
| 15639 | If match is true, only return a name if its value matches C<uninit_sv>. |
| 15640 | So roughly speaking, if a unary operator (such as C<OP_COS>) generates a |
| 15641 | warning, then following the direct child of the op may yield an |
| 15642 | C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the |
| 15643 | other hand, with C<OP_ADD> there are two branches to follow, so we only print |
| 15644 | the variable name if we get an exact match. |
| 15645 | C<desc_p> points to a string pointer holding the description of the op. |
| 15646 | This may be updated if needed. |
| 15647 | |
| 15648 | The name is returned as a mortal SV. |
| 15649 | |
| 15650 | Assumes that C<PL_op> is the OP that originally triggered the error, and that |
| 15651 | C<PL_comppad>/C<PL_curpad> points to the currently executing pad. |
| 15652 | |
| 15653 | =cut |
| 15654 | */ |
| 15655 | |
| 15656 | STATIC SV * |
| 15657 | S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv, |
| 15658 | bool match, const char **desc_p) |
| 15659 | { |
| 15660 | dVAR; |
| 15661 | SV *sv; |
| 15662 | const GV *gv; |
| 15663 | const OP *o, *o2, *kid; |
| 15664 | |
| 15665 | PERL_ARGS_ASSERT_FIND_UNINIT_VAR; |
| 15666 | |
| 15667 | if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef || |
| 15668 | uninit_sv == &PL_sv_placeholder))) |
| 15669 | return NULL; |
| 15670 | |
| 15671 | switch (obase->op_type) { |
| 15672 | |
| 15673 | case OP_RV2AV: |
| 15674 | case OP_RV2HV: |
| 15675 | case OP_PADAV: |
| 15676 | case OP_PADHV: |
| 15677 | { |
| 15678 | const bool pad = ( obase->op_type == OP_PADAV |
| 15679 | || obase->op_type == OP_PADHV |
| 15680 | || obase->op_type == OP_PADRANGE |
| 15681 | ); |
| 15682 | |
| 15683 | const bool hash = ( obase->op_type == OP_PADHV |
| 15684 | || obase->op_type == OP_RV2HV |
| 15685 | || (obase->op_type == OP_PADRANGE |
| 15686 | && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV) |
| 15687 | ); |
| 15688 | I32 index = 0; |
| 15689 | SV *keysv = NULL; |
| 15690 | int subscript_type = FUV_SUBSCRIPT_WITHIN; |
| 15691 | |
| 15692 | if (pad) { /* @lex, %lex */ |
| 15693 | sv = PAD_SVl(obase->op_targ); |
| 15694 | gv = NULL; |
| 15695 | } |
| 15696 | else { |
| 15697 | if (cUNOPx(obase)->op_first->op_type == OP_GV) { |
| 15698 | /* @global, %global */ |
| 15699 | gv = cGVOPx_gv(cUNOPx(obase)->op_first); |
| 15700 | if (!gv) |
| 15701 | break; |
| 15702 | sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv)); |
| 15703 | } |
| 15704 | else if (obase == PL_op) /* @{expr}, %{expr} */ |
| 15705 | return find_uninit_var(cUNOPx(obase)->op_first, |
| 15706 | uninit_sv, match, desc_p); |
| 15707 | else /* @{expr}, %{expr} as a sub-expression */ |
| 15708 | return NULL; |
| 15709 | } |
| 15710 | |
| 15711 | /* attempt to find a match within the aggregate */ |
| 15712 | if (hash) { |
| 15713 | keysv = find_hash_subscript((const HV*)sv, uninit_sv); |
| 15714 | if (keysv) |
| 15715 | subscript_type = FUV_SUBSCRIPT_HASH; |
| 15716 | } |
| 15717 | else { |
| 15718 | index = find_array_subscript((const AV *)sv, uninit_sv); |
| 15719 | if (index >= 0) |
| 15720 | subscript_type = FUV_SUBSCRIPT_ARRAY; |
| 15721 | } |
| 15722 | |
| 15723 | if (match && subscript_type == FUV_SUBSCRIPT_WITHIN) |
| 15724 | break; |
| 15725 | |
| 15726 | return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ, |
| 15727 | keysv, index, subscript_type); |
| 15728 | } |
| 15729 | |
| 15730 | case OP_RV2SV: |
| 15731 | if (cUNOPx(obase)->op_first->op_type == OP_GV) { |
| 15732 | /* $global */ |
| 15733 | gv = cGVOPx_gv(cUNOPx(obase)->op_first); |
| 15734 | if (!gv || !GvSTASH(gv)) |
| 15735 | break; |
| 15736 | if (match && (GvSV(gv) != uninit_sv)) |
| 15737 | break; |
| 15738 | return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE); |
| 15739 | } |
| 15740 | /* ${expr} */ |
| 15741 | return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p); |
| 15742 | |
| 15743 | case OP_PADSV: |
| 15744 | if (match && PAD_SVl(obase->op_targ) != uninit_sv) |
| 15745 | break; |
| 15746 | return varname(NULL, '$', obase->op_targ, |
| 15747 | NULL, 0, FUV_SUBSCRIPT_NONE); |
| 15748 | |
| 15749 | case OP_GVSV: |
| 15750 | gv = cGVOPx_gv(obase); |
| 15751 | if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv)) |
| 15752 | break; |
| 15753 | return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE); |
| 15754 | |
| 15755 | case OP_AELEMFAST_LEX: |
| 15756 | if (match) { |
| 15757 | SV **svp; |
| 15758 | AV *av = MUTABLE_AV(PAD_SV(obase->op_targ)); |
| 15759 | if (!av || SvRMAGICAL(av)) |
| 15760 | break; |
| 15761 | svp = av_fetch(av, (I8)obase->op_private, FALSE); |
| 15762 | if (!svp || *svp != uninit_sv) |
| 15763 | break; |
| 15764 | } |
| 15765 | return varname(NULL, '$', obase->op_targ, |
| 15766 | NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY); |
| 15767 | case OP_AELEMFAST: |
| 15768 | { |
| 15769 | gv = cGVOPx_gv(obase); |
| 15770 | if (!gv) |
| 15771 | break; |
| 15772 | if (match) { |
| 15773 | SV **svp; |
| 15774 | AV *const av = GvAV(gv); |
| 15775 | if (!av || SvRMAGICAL(av)) |
| 15776 | break; |
| 15777 | svp = av_fetch(av, (I8)obase->op_private, FALSE); |
| 15778 | if (!svp || *svp != uninit_sv) |
| 15779 | break; |
| 15780 | } |
| 15781 | return varname(gv, '$', 0, |
| 15782 | NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY); |
| 15783 | } |
| 15784 | NOT_REACHED; /* NOTREACHED */ |
| 15785 | |
| 15786 | case OP_EXISTS: |
| 15787 | o = cUNOPx(obase)->op_first; |
| 15788 | if (!o || o->op_type != OP_NULL || |
| 15789 | ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM)) |
| 15790 | break; |
| 15791 | return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p); |
| 15792 | |
| 15793 | case OP_AELEM: |
| 15794 | case OP_HELEM: |
| 15795 | { |
| 15796 | bool negate = FALSE; |
| 15797 | |
| 15798 | if (PL_op == obase) |
| 15799 | /* $a[uninit_expr] or $h{uninit_expr} */ |
| 15800 | return find_uninit_var(cBINOPx(obase)->op_last, |
| 15801 | uninit_sv, match, desc_p); |
| 15802 | |
| 15803 | gv = NULL; |
| 15804 | o = cBINOPx(obase)->op_first; |
| 15805 | kid = cBINOPx(obase)->op_last; |
| 15806 | |
| 15807 | /* get the av or hv, and optionally the gv */ |
| 15808 | sv = NULL; |
| 15809 | if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) { |
| 15810 | sv = PAD_SV(o->op_targ); |
| 15811 | } |
| 15812 | else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV) |
| 15813 | && cUNOPo->op_first->op_type == OP_GV) |
| 15814 | { |
| 15815 | gv = cGVOPx_gv(cUNOPo->op_first); |
| 15816 | if (!gv) |
| 15817 | break; |
| 15818 | sv = o->op_type |
| 15819 | == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv)); |
| 15820 | } |
| 15821 | if (!sv) |
| 15822 | break; |
| 15823 | |
| 15824 | if (kid && kid->op_type == OP_NEGATE) { |
| 15825 | negate = TRUE; |
| 15826 | kid = cUNOPx(kid)->op_first; |
| 15827 | } |
| 15828 | |
| 15829 | if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) { |
| 15830 | /* index is constant */ |
| 15831 | SV* kidsv; |
| 15832 | if (negate) { |
| 15833 | kidsv = newSVpvs_flags("-", SVs_TEMP); |
| 15834 | sv_catsv(kidsv, cSVOPx_sv(kid)); |
| 15835 | } |
| 15836 | else |
| 15837 | kidsv = cSVOPx_sv(kid); |
| 15838 | if (match) { |
| 15839 | if (SvMAGICAL(sv)) |
| 15840 | break; |
| 15841 | if (obase->op_type == OP_HELEM) { |
| 15842 | HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0); |
| 15843 | if (!he || HeVAL(he) != uninit_sv) |
| 15844 | break; |
| 15845 | } |
| 15846 | else { |
| 15847 | SV * const opsv = cSVOPx_sv(kid); |
| 15848 | const IV opsviv = SvIV(opsv); |
| 15849 | SV * const * const svp = av_fetch(MUTABLE_AV(sv), |
| 15850 | negate ? - opsviv : opsviv, |
| 15851 | FALSE); |
| 15852 | if (!svp || *svp != uninit_sv) |
| 15853 | break; |
| 15854 | } |
| 15855 | } |
| 15856 | if (obase->op_type == OP_HELEM) |
| 15857 | return varname(gv, '%', o->op_targ, |
| 15858 | kidsv, 0, FUV_SUBSCRIPT_HASH); |
| 15859 | else |
| 15860 | return varname(gv, '@', o->op_targ, NULL, |
| 15861 | negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)), |
| 15862 | FUV_SUBSCRIPT_ARRAY); |
| 15863 | } |
| 15864 | else { |
| 15865 | /* index is an expression; |
| 15866 | * attempt to find a match within the aggregate */ |
| 15867 | if (obase->op_type == OP_HELEM) { |
| 15868 | SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv); |
| 15869 | if (keysv) |
| 15870 | return varname(gv, '%', o->op_targ, |
| 15871 | keysv, 0, FUV_SUBSCRIPT_HASH); |
| 15872 | } |
| 15873 | else { |
| 15874 | const I32 index |
| 15875 | = find_array_subscript((const AV *)sv, uninit_sv); |
| 15876 | if (index >= 0) |
| 15877 | return varname(gv, '@', o->op_targ, |
| 15878 | NULL, index, FUV_SUBSCRIPT_ARRAY); |
| 15879 | } |
| 15880 | if (match) |
| 15881 | break; |
| 15882 | return varname(gv, |
| 15883 | (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV) |
| 15884 | ? '@' : '%'), |
| 15885 | o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN); |
| 15886 | } |
| 15887 | NOT_REACHED; /* NOTREACHED */ |
| 15888 | } |
| 15889 | |
| 15890 | case OP_MULTIDEREF: { |
| 15891 | /* If we were executing OP_MULTIDEREF when the undef warning |
| 15892 | * triggered, then it must be one of the index values within |
| 15893 | * that triggered it. If not, then the only possibility is that |
| 15894 | * the value retrieved by the last aggregate lookup might be the |
| 15895 | * culprit. For the former, we set PL_multideref_pc each time before |
| 15896 | * using an index, so work though the item list until we reach |
| 15897 | * that point. For the latter, just work through the entire item |
| 15898 | * list; the last aggregate retrieved will be the candidate. |
| 15899 | */ |
| 15900 | |
| 15901 | /* the named aggregate, if any */ |
| 15902 | PADOFFSET agg_targ = 0; |
| 15903 | GV *agg_gv = NULL; |
| 15904 | /* the last-seen index */ |
| 15905 | UV index_type; |
| 15906 | PADOFFSET index_targ; |
| 15907 | GV *index_gv; |
| 15908 | IV index_const_iv = 0; /* init for spurious compiler warn */ |
| 15909 | SV *index_const_sv; |
| 15910 | int depth = 0; /* how many array/hash lookups we've done */ |
| 15911 | |
| 15912 | UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux; |
| 15913 | UNOP_AUX_item *last = NULL; |
| 15914 | UV actions = items->uv; |
| 15915 | bool is_hv; |
| 15916 | |
| 15917 | if (PL_op == obase) { |
| 15918 | last = PL_multideref_pc; |
| 15919 | assert(last >= items && last <= items + items[-1].uv); |
| 15920 | } |
| 15921 | |
| 15922 | assert(actions); |
| 15923 | |
| 15924 | while (1) { |
| 15925 | is_hv = FALSE; |
| 15926 | switch (actions & MDEREF_ACTION_MASK) { |
| 15927 | |
| 15928 | case MDEREF_reload: |
| 15929 | actions = (++items)->uv; |
| 15930 | continue; |
| 15931 | |
| 15932 | case MDEREF_HV_padhv_helem: /* $lex{...} */ |
| 15933 | is_hv = TRUE; |
| 15934 | /* FALLTHROUGH */ |
| 15935 | case MDEREF_AV_padav_aelem: /* $lex[...] */ |
| 15936 | agg_targ = (++items)->pad_offset; |
| 15937 | agg_gv = NULL; |
| 15938 | break; |
| 15939 | |
| 15940 | case MDEREF_HV_gvhv_helem: /* $pkg{...} */ |
| 15941 | is_hv = TRUE; |
| 15942 | /* FALLTHROUGH */ |
| 15943 | case MDEREF_AV_gvav_aelem: /* $pkg[...] */ |
| 15944 | agg_targ = 0; |
| 15945 | agg_gv = (GV*)UNOP_AUX_item_sv(++items); |
| 15946 | assert(isGV_with_GP(agg_gv)); |
| 15947 | break; |
| 15948 | |
| 15949 | case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */ |
| 15950 | case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */ |
| 15951 | ++items; |
| 15952 | /* FALLTHROUGH */ |
| 15953 | case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */ |
| 15954 | case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */ |
| 15955 | agg_targ = 0; |
| 15956 | agg_gv = NULL; |
| 15957 | is_hv = TRUE; |
| 15958 | break; |
| 15959 | |
| 15960 | case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */ |
| 15961 | case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */ |
| 15962 | ++items; |
| 15963 | /* FALLTHROUGH */ |
| 15964 | case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */ |
| 15965 | case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */ |
| 15966 | agg_targ = 0; |
| 15967 | agg_gv = NULL; |
| 15968 | } /* switch */ |
| 15969 | |
| 15970 | index_targ = 0; |
| 15971 | index_gv = NULL; |
| 15972 | index_const_sv = NULL; |
| 15973 | |
| 15974 | index_type = (actions & MDEREF_INDEX_MASK); |
| 15975 | switch (index_type) { |
| 15976 | case MDEREF_INDEX_none: |
| 15977 | break; |
| 15978 | case MDEREF_INDEX_const: |
| 15979 | if (is_hv) |
| 15980 | index_const_sv = UNOP_AUX_item_sv(++items) |
| 15981 | else |
| 15982 | index_const_iv = (++items)->iv; |
| 15983 | break; |
| 15984 | case MDEREF_INDEX_padsv: |
| 15985 | index_targ = (++items)->pad_offset; |
| 15986 | break; |
| 15987 | case MDEREF_INDEX_gvsv: |
| 15988 | index_gv = (GV*)UNOP_AUX_item_sv(++items); |
| 15989 | assert(isGV_with_GP(index_gv)); |
| 15990 | break; |
| 15991 | } |
| 15992 | |
| 15993 | if (index_type != MDEREF_INDEX_none) |
| 15994 | depth++; |
| 15995 | |
| 15996 | if ( index_type == MDEREF_INDEX_none |
| 15997 | || (actions & MDEREF_FLAG_last) |
| 15998 | || (last && items == last) |
| 15999 | ) |
| 16000 | break; |
| 16001 | |
| 16002 | actions >>= MDEREF_SHIFT; |
| 16003 | } /* while */ |
| 16004 | |
| 16005 | if (PL_op == obase) { |
| 16006 | /* index was undef */ |
| 16007 | |
| 16008 | *desc_p = ( (actions & MDEREF_FLAG_last) |
| 16009 | && (obase->op_private |
| 16010 | & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))) |
| 16011 | ? |
| 16012 | (obase->op_private & OPpMULTIDEREF_EXISTS) |
| 16013 | ? "exists" |
| 16014 | : "delete" |
| 16015 | : is_hv ? "hash element" : "array element"; |
| 16016 | assert(index_type != MDEREF_INDEX_none); |
| 16017 | if (index_gv) |
| 16018 | return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE); |
| 16019 | if (index_targ) |
| 16020 | return varname(NULL, '$', index_targ, |
| 16021 | NULL, 0, FUV_SUBSCRIPT_NONE); |
| 16022 | assert(is_hv); /* AV index is an IV and can't be undef */ |
| 16023 | /* can a const HV index ever be undef? */ |
| 16024 | return NULL; |
| 16025 | } |
| 16026 | |
| 16027 | /* the SV returned by pp_multideref() was undef, if anything was */ |
| 16028 | |
| 16029 | if (depth != 1) |
| 16030 | break; |
| 16031 | |
| 16032 | if (agg_targ) |
| 16033 | sv = PAD_SV(agg_targ); |
| 16034 | else if (agg_gv) |
| 16035 | sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv)); |
| 16036 | else |
| 16037 | break; |
| 16038 | |
| 16039 | if (index_type == MDEREF_INDEX_const) { |
| 16040 | if (match) { |
| 16041 | if (SvMAGICAL(sv)) |
| 16042 | break; |
| 16043 | if (is_hv) { |
| 16044 | HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0); |
| 16045 | if (!he || HeVAL(he) != uninit_sv) |
| 16046 | break; |
| 16047 | } |
| 16048 | else { |
| 16049 | SV * const * const svp = |
| 16050 | av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE); |
| 16051 | if (!svp || *svp != uninit_sv) |
| 16052 | break; |
| 16053 | } |
| 16054 | } |
| 16055 | return is_hv |
| 16056 | ? varname(agg_gv, '%', agg_targ, |
| 16057 | index_const_sv, 0, FUV_SUBSCRIPT_HASH) |
| 16058 | : varname(agg_gv, '@', agg_targ, |
| 16059 | NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY); |
| 16060 | } |
| 16061 | else { |
| 16062 | /* index is an var */ |
| 16063 | if (is_hv) { |
| 16064 | SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv); |
| 16065 | if (keysv) |
| 16066 | return varname(agg_gv, '%', agg_targ, |
| 16067 | keysv, 0, FUV_SUBSCRIPT_HASH); |
| 16068 | } |
| 16069 | else { |
| 16070 | const I32 index |
| 16071 | = find_array_subscript((const AV *)sv, uninit_sv); |
| 16072 | if (index >= 0) |
| 16073 | return varname(agg_gv, '@', agg_targ, |
| 16074 | NULL, index, FUV_SUBSCRIPT_ARRAY); |
| 16075 | } |
| 16076 | if (match) |
| 16077 | break; |
| 16078 | return varname(agg_gv, |
| 16079 | is_hv ? '%' : '@', |
| 16080 | agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN); |
| 16081 | } |
| 16082 | NOT_REACHED; /* NOTREACHED */ |
| 16083 | } |
| 16084 | |
| 16085 | case OP_AASSIGN: |
| 16086 | /* only examine RHS */ |
| 16087 | return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, |
| 16088 | match, desc_p); |
| 16089 | |
| 16090 | case OP_OPEN: |
| 16091 | o = cUNOPx(obase)->op_first; |
| 16092 | if ( o->op_type == OP_PUSHMARK |
| 16093 | || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK) |
| 16094 | ) |
| 16095 | o = OpSIBLING(o); |
| 16096 | |
| 16097 | if (!OpHAS_SIBLING(o)) { |
| 16098 | /* one-arg version of open is highly magical */ |
| 16099 | |
| 16100 | if (o->op_type == OP_GV) { /* open FOO; */ |
| 16101 | gv = cGVOPx_gv(o); |
| 16102 | if (match && GvSV(gv) != uninit_sv) |
| 16103 | break; |
| 16104 | return varname(gv, '$', 0, |
| 16105 | NULL, 0, FUV_SUBSCRIPT_NONE); |
| 16106 | } |
| 16107 | /* other possibilities not handled are: |
| 16108 | * open $x; or open my $x; should return '${*$x}' |
| 16109 | * open expr; should return '$'.expr ideally |
| 16110 | */ |
| 16111 | break; |
| 16112 | } |
| 16113 | goto do_op; |
| 16114 | |
| 16115 | /* ops where $_ may be an implicit arg */ |
| 16116 | case OP_TRANS: |
| 16117 | case OP_TRANSR: |
| 16118 | case OP_SUBST: |
| 16119 | case OP_MATCH: |
| 16120 | if ( !(obase->op_flags & OPf_STACKED)) { |
| 16121 | if (uninit_sv == DEFSV) |
| 16122 | return newSVpvs_flags("$_", SVs_TEMP); |
| 16123 | else if (obase->op_targ |
| 16124 | && uninit_sv == PAD_SVl(obase->op_targ)) |
| 16125 | return varname(NULL, '$', obase->op_targ, NULL, 0, |
| 16126 | FUV_SUBSCRIPT_NONE); |
| 16127 | } |
| 16128 | goto do_op; |
| 16129 | |
| 16130 | case OP_PRTF: |
| 16131 | case OP_PRINT: |
| 16132 | case OP_SAY: |
| 16133 | match = 1; /* print etc can return undef on defined args */ |
| 16134 | /* skip filehandle as it can't produce 'undef' warning */ |
| 16135 | o = cUNOPx(obase)->op_first; |
| 16136 | if ((obase->op_flags & OPf_STACKED) |
| 16137 | && |
| 16138 | ( o->op_type == OP_PUSHMARK |
| 16139 | || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK))) |
| 16140 | o = OpSIBLING(OpSIBLING(o)); |
| 16141 | goto do_op2; |
| 16142 | |
| 16143 | |
| 16144 | case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */ |
| 16145 | case OP_CUSTOM: /* XS or custom code could trigger random warnings */ |
| 16146 | |
| 16147 | /* the following ops are capable of returning PL_sv_undef even for |
| 16148 | * defined arg(s) */ |
| 16149 | |
| 16150 | case OP_BACKTICK: |
| 16151 | case OP_PIPE_OP: |
| 16152 | case OP_FILENO: |
| 16153 | case OP_BINMODE: |
| 16154 | case OP_TIED: |
| 16155 | case OP_GETC: |
| 16156 | case OP_SYSREAD: |
| 16157 | case OP_SEND: |
| 16158 | case OP_IOCTL: |
| 16159 | case OP_SOCKET: |
| 16160 | case OP_SOCKPAIR: |
| 16161 | case OP_BIND: |
| 16162 | case OP_CONNECT: |
| 16163 | case OP_LISTEN: |
| 16164 | case OP_ACCEPT: |
| 16165 | case OP_SHUTDOWN: |
| 16166 | case OP_SSOCKOPT: |
| 16167 | case OP_GETPEERNAME: |
| 16168 | case OP_FTRREAD: |
| 16169 | case OP_FTRWRITE: |
| 16170 | case OP_FTREXEC: |
| 16171 | case OP_FTROWNED: |
| 16172 | case OP_FTEREAD: |
| 16173 | case OP_FTEWRITE: |
| 16174 | case OP_FTEEXEC: |
| 16175 | case OP_FTEOWNED: |
| 16176 | case OP_FTIS: |
| 16177 | case OP_FTZERO: |
| 16178 | case OP_FTSIZE: |
| 16179 | case OP_FTFILE: |
| 16180 | case OP_FTDIR: |
| 16181 | case OP_FTLINK: |
| 16182 | case OP_FTPIPE: |
| 16183 | case OP_FTSOCK: |
| 16184 | case OP_FTBLK: |
| 16185 | case OP_FTCHR: |
| 16186 | case OP_FTTTY: |
| 16187 | case OP_FTSUID: |
| 16188 | case OP_FTSGID: |
| 16189 | case OP_FTSVTX: |
| 16190 | case OP_FTTEXT: |
| 16191 | case OP_FTBINARY: |
| 16192 | case OP_FTMTIME: |
| 16193 | case OP_FTATIME: |
| 16194 | case OP_FTCTIME: |
| 16195 | case OP_READLINK: |
| 16196 | case OP_OPEN_DIR: |
| 16197 | case OP_READDIR: |
| 16198 | case OP_TELLDIR: |
| 16199 | case OP_SEEKDIR: |
| 16200 | case OP_REWINDDIR: |
| 16201 | case OP_CLOSEDIR: |
| 16202 | case OP_GMTIME: |
| 16203 | case OP_ALARM: |
| 16204 | case OP_SEMGET: |
| 16205 | case OP_GETLOGIN: |
| 16206 | case OP_UNDEF: |
| 16207 | case OP_SUBSTR: |
| 16208 | case OP_AEACH: |
| 16209 | case OP_EACH: |
| 16210 | case OP_SORT: |
| 16211 | case OP_CALLER: |
| 16212 | case OP_DOFILE: |
| 16213 | case OP_PROTOTYPE: |
| 16214 | case OP_NCMP: |
| 16215 | case OP_SMARTMATCH: |
| 16216 | case OP_UNPACK: |
| 16217 | case OP_SYSOPEN: |
| 16218 | case OP_SYSSEEK: |
| 16219 | match = 1; |
| 16220 | goto do_op; |
| 16221 | |
| 16222 | case OP_ENTERSUB: |
| 16223 | case OP_GOTO: |
| 16224 | /* XXX tmp hack: these two may call an XS sub, and currently |
| 16225 | XS subs don't have a SUB entry on the context stack, so CV and |
| 16226 | pad determination goes wrong, and BAD things happen. So, just |
| 16227 | don't try to determine the value under those circumstances. |
| 16228 | Need a better fix at dome point. DAPM 11/2007 */ |
| 16229 | break; |
| 16230 | |
| 16231 | case OP_FLIP: |
| 16232 | case OP_FLOP: |
| 16233 | { |
| 16234 | GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV); |
| 16235 | if (gv && GvSV(gv) == uninit_sv) |
| 16236 | return newSVpvs_flags("$.", SVs_TEMP); |
| 16237 | goto do_op; |
| 16238 | } |
| 16239 | |
| 16240 | case OP_POS: |
| 16241 | /* def-ness of rval pos() is independent of the def-ness of its arg */ |
| 16242 | if ( !(obase->op_flags & OPf_MOD)) |
| 16243 | break; |
| 16244 | |
| 16245 | case OP_SCHOMP: |
| 16246 | case OP_CHOMP: |
| 16247 | if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs)) |
| 16248 | return newSVpvs_flags("${$/}", SVs_TEMP); |
| 16249 | /* FALLTHROUGH */ |
| 16250 | |
| 16251 | default: |
| 16252 | do_op: |
| 16253 | if (!(obase->op_flags & OPf_KIDS)) |
| 16254 | break; |
| 16255 | o = cUNOPx(obase)->op_first; |
| 16256 | |
| 16257 | do_op2: |
| 16258 | if (!o) |
| 16259 | break; |
| 16260 | |
| 16261 | /* This loop checks all the kid ops, skipping any that cannot pos- |
| 16262 | * sibly be responsible for the uninitialized value; i.e., defined |
| 16263 | * constants and ops that return nothing. If there is only one op |
| 16264 | * left that is not skipped, then we *know* it is responsible for |
| 16265 | * the uninitialized value. If there is more than one op left, we |
| 16266 | * have to look for an exact match in the while() loop below. |
| 16267 | * Note that we skip padrange, because the individual pad ops that |
| 16268 | * it replaced are still in the tree, so we work on them instead. |
| 16269 | */ |
| 16270 | o2 = NULL; |
| 16271 | for (kid=o; kid; kid = OpSIBLING(kid)) { |
| 16272 | const OPCODE type = kid->op_type; |
| 16273 | if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid))) |
| 16274 | || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS)) |
| 16275 | || (type == OP_PUSHMARK) |
| 16276 | || (type == OP_PADRANGE) |
| 16277 | ) |
| 16278 | continue; |
| 16279 | |
| 16280 | if (o2) { /* more than one found */ |
| 16281 | o2 = NULL; |
| 16282 | break; |
| 16283 | } |
| 16284 | o2 = kid; |
| 16285 | } |
| 16286 | if (o2) |
| 16287 | return find_uninit_var(o2, uninit_sv, match, desc_p); |
| 16288 | |
| 16289 | /* scan all args */ |
| 16290 | while (o) { |
| 16291 | sv = find_uninit_var(o, uninit_sv, 1, desc_p); |
| 16292 | if (sv) |
| 16293 | return sv; |
| 16294 | o = OpSIBLING(o); |
| 16295 | } |
| 16296 | break; |
| 16297 | } |
| 16298 | return NULL; |
| 16299 | } |
| 16300 | |
| 16301 | |
| 16302 | /* |
| 16303 | =for apidoc report_uninit |
| 16304 | |
| 16305 | Print appropriate "Use of uninitialized variable" warning. |
| 16306 | |
| 16307 | =cut |
| 16308 | */ |
| 16309 | |
| 16310 | void |
| 16311 | Perl_report_uninit(pTHX_ const SV *uninit_sv) |
| 16312 | { |
| 16313 | const char *desc = NULL; |
| 16314 | SV* varname = NULL; |
| 16315 | |
| 16316 | if (PL_op) { |
| 16317 | desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded |
| 16318 | ? "join or string" |
| 16319 | : OP_DESC(PL_op); |
| 16320 | if (uninit_sv && PL_curpad) { |
| 16321 | varname = find_uninit_var(PL_op, uninit_sv, 0, &desc); |
| 16322 | if (varname) |
| 16323 | sv_insert(varname, 0, 0, " ", 1); |
| 16324 | } |
| 16325 | } |
| 16326 | else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0) |
| 16327 | /* we've reached the end of a sort block or sub, |
| 16328 | * and the uninit value is probably what that code returned */ |
| 16329 | desc = "sort"; |
| 16330 | |
| 16331 | /* PL_warn_uninit_sv is constant */ |
| 16332 | GCC_DIAG_IGNORE(-Wformat-nonliteral); |
| 16333 | if (desc) |
| 16334 | /* diag_listed_as: Use of uninitialized value%s */ |
| 16335 | Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv, |
| 16336 | SVfARG(varname ? varname : &PL_sv_no), |
| 16337 | " in ", desc); |
| 16338 | else |
| 16339 | Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit, |
| 16340 | "", "", ""); |
| 16341 | GCC_DIAG_RESTORE; |
| 16342 | } |
| 16343 | |
| 16344 | /* |
| 16345 | * ex: set ts=8 sts=4 sw=4 et: |
| 16346 | */ |