| 1 | /* hv.c |
| 2 | * |
| 3 | * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, |
| 4 | * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others |
| 5 | * |
| 6 | * You may distribute under the terms of either the GNU General Public |
| 7 | * License or the Artistic License, as specified in the README file. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | /* |
| 12 | * I sit beside the fire and think |
| 13 | * of all that I have seen. |
| 14 | * --Bilbo |
| 15 | * |
| 16 | * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"] |
| 17 | */ |
| 18 | |
| 19 | /* |
| 20 | =head1 Hash Manipulation Functions |
| 21 | A HV structure represents a Perl hash. It consists mainly of an array |
| 22 | of pointers, each of which points to a linked list of HE structures. The |
| 23 | array is indexed by the hash function of the key, so each linked list |
| 24 | represents all the hash entries with the same hash value. Each HE contains |
| 25 | a pointer to the actual value, plus a pointer to a HEK structure which |
| 26 | holds the key and hash value. |
| 27 | |
| 28 | =cut |
| 29 | |
| 30 | */ |
| 31 | |
| 32 | #include "EXTERN.h" |
| 33 | #define PERL_IN_HV_C |
| 34 | #define PERL_HASH_INTERNAL_ACCESS |
| 35 | #include "perl.h" |
| 36 | |
| 37 | #define DO_HSPLIT(xhv) ((xhv)->xhv_keys > (xhv)->xhv_max) /* HvTOTALKEYS(hv) > HvMAX(hv) */ |
| 38 | #define HV_FILL_THRESHOLD 31 |
| 39 | |
| 40 | static const char S_strtab_error[] |
| 41 | = "Cannot modify shared string table in hv_%s"; |
| 42 | |
| 43 | #ifdef PURIFY |
| 44 | |
| 45 | #define new_HE() (HE*)safemalloc(sizeof(HE)) |
| 46 | #define del_HE(p) safefree((char*)p) |
| 47 | |
| 48 | #else |
| 49 | |
| 50 | STATIC HE* |
| 51 | S_new_he(pTHX) |
| 52 | { |
| 53 | HE* he; |
| 54 | void ** const root = &PL_body_roots[HE_SVSLOT]; |
| 55 | |
| 56 | if (!*root) |
| 57 | Perl_more_bodies(aTHX_ HE_SVSLOT, sizeof(HE), PERL_ARENA_SIZE); |
| 58 | he = (HE*) *root; |
| 59 | assert(he); |
| 60 | *root = HeNEXT(he); |
| 61 | return he; |
| 62 | } |
| 63 | |
| 64 | #define new_HE() new_he() |
| 65 | #define del_HE(p) \ |
| 66 | STMT_START { \ |
| 67 | HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \ |
| 68 | PL_body_roots[HE_SVSLOT] = p; \ |
| 69 | } STMT_END |
| 70 | |
| 71 | |
| 72 | |
| 73 | #endif |
| 74 | |
| 75 | STATIC HEK * |
| 76 | S_save_hek_flags(const char *str, I32 len, U32 hash, int flags) |
| 77 | { |
| 78 | const int flags_masked = flags & HVhek_MASK; |
| 79 | char *k; |
| 80 | HEK *hek; |
| 81 | |
| 82 | PERL_ARGS_ASSERT_SAVE_HEK_FLAGS; |
| 83 | |
| 84 | Newx(k, HEK_BASESIZE + len + 2, char); |
| 85 | hek = (HEK*)k; |
| 86 | Copy(str, HEK_KEY(hek), len, char); |
| 87 | HEK_KEY(hek)[len] = 0; |
| 88 | HEK_LEN(hek) = len; |
| 89 | HEK_HASH(hek) = hash; |
| 90 | HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED; |
| 91 | |
| 92 | if (flags & HVhek_FREEKEY) |
| 93 | Safefree(str); |
| 94 | return hek; |
| 95 | } |
| 96 | |
| 97 | /* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent |
| 98 | * for tied hashes */ |
| 99 | |
| 100 | void |
| 101 | Perl_free_tied_hv_pool(pTHX) |
| 102 | { |
| 103 | HE *he = PL_hv_fetch_ent_mh; |
| 104 | while (he) { |
| 105 | HE * const ohe = he; |
| 106 | Safefree(HeKEY_hek(he)); |
| 107 | he = HeNEXT(he); |
| 108 | del_HE(ohe); |
| 109 | } |
| 110 | PL_hv_fetch_ent_mh = NULL; |
| 111 | } |
| 112 | |
| 113 | #if defined(USE_ITHREADS) |
| 114 | HEK * |
| 115 | Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param) |
| 116 | { |
| 117 | HEK *shared; |
| 118 | |
| 119 | PERL_ARGS_ASSERT_HEK_DUP; |
| 120 | PERL_UNUSED_ARG(param); |
| 121 | |
| 122 | if (!source) |
| 123 | return NULL; |
| 124 | |
| 125 | shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); |
| 126 | if (shared) { |
| 127 | /* We already shared this hash key. */ |
| 128 | (void)share_hek_hek(shared); |
| 129 | } |
| 130 | else { |
| 131 | shared |
| 132 | = share_hek_flags(HEK_KEY(source), HEK_LEN(source), |
| 133 | HEK_HASH(source), HEK_FLAGS(source)); |
| 134 | ptr_table_store(PL_ptr_table, source, shared); |
| 135 | } |
| 136 | return shared; |
| 137 | } |
| 138 | |
| 139 | HE * |
| 140 | Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param) |
| 141 | { |
| 142 | HE *ret; |
| 143 | |
| 144 | PERL_ARGS_ASSERT_HE_DUP; |
| 145 | |
| 146 | if (!e) |
| 147 | return NULL; |
| 148 | /* look for it in the table first */ |
| 149 | ret = (HE*)ptr_table_fetch(PL_ptr_table, e); |
| 150 | if (ret) |
| 151 | return ret; |
| 152 | |
| 153 | /* create anew and remember what it is */ |
| 154 | ret = new_HE(); |
| 155 | ptr_table_store(PL_ptr_table, e, ret); |
| 156 | |
| 157 | HeNEXT(ret) = he_dup(HeNEXT(e),shared, param); |
| 158 | if (HeKLEN(e) == HEf_SVKEY) { |
| 159 | char *k; |
| 160 | Newx(k, HEK_BASESIZE + sizeof(const SV *), char); |
| 161 | HeKEY_hek(ret) = (HEK*)k; |
| 162 | HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param); |
| 163 | } |
| 164 | else if (shared) { |
| 165 | /* This is hek_dup inlined, which seems to be important for speed |
| 166 | reasons. */ |
| 167 | HEK * const source = HeKEY_hek(e); |
| 168 | HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); |
| 169 | |
| 170 | if (shared) { |
| 171 | /* We already shared this hash key. */ |
| 172 | (void)share_hek_hek(shared); |
| 173 | } |
| 174 | else { |
| 175 | shared |
| 176 | = share_hek_flags(HEK_KEY(source), HEK_LEN(source), |
| 177 | HEK_HASH(source), HEK_FLAGS(source)); |
| 178 | ptr_table_store(PL_ptr_table, source, shared); |
| 179 | } |
| 180 | HeKEY_hek(ret) = shared; |
| 181 | } |
| 182 | else |
| 183 | HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e), |
| 184 | HeKFLAGS(e)); |
| 185 | HeVAL(ret) = sv_dup_inc(HeVAL(e), param); |
| 186 | return ret; |
| 187 | } |
| 188 | #endif /* USE_ITHREADS */ |
| 189 | |
| 190 | static void |
| 191 | S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen, |
| 192 | const char *msg) |
| 193 | { |
| 194 | SV * const sv = sv_newmortal(); |
| 195 | |
| 196 | PERL_ARGS_ASSERT_HV_NOTALLOWED; |
| 197 | |
| 198 | if (!(flags & HVhek_FREEKEY)) { |
| 199 | sv_setpvn(sv, key, klen); |
| 200 | } |
| 201 | else { |
| 202 | /* Need to free saved eventually assign to mortal SV */ |
| 203 | /* XXX is this line an error ???: SV *sv = sv_newmortal(); */ |
| 204 | sv_usepvn(sv, (char *) key, klen); |
| 205 | } |
| 206 | if (flags & HVhek_UTF8) { |
| 207 | SvUTF8_on(sv); |
| 208 | } |
| 209 | Perl_croak(aTHX_ msg, SVfARG(sv)); |
| 210 | } |
| 211 | |
| 212 | /* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot |
| 213 | * contains an SV* */ |
| 214 | |
| 215 | /* |
| 216 | =for apidoc hv_store |
| 217 | |
| 218 | Stores an SV in a hash. The hash key is specified as C<key> and the |
| 219 | absolute value of C<klen> is the length of the key. If C<klen> is |
| 220 | negative the key is assumed to be in UTF-8-encoded Unicode. The |
| 221 | C<hash> parameter is the precomputed hash value; if it is zero then |
| 222 | Perl will compute it. |
| 223 | |
| 224 | The return value will be |
| 225 | C<NULL> if the operation failed or if the value did not need to be actually |
| 226 | stored within the hash (as in the case of tied hashes). Otherwise it can |
| 227 | be dereferenced to get the original C<SV*>. Note that the caller is |
| 228 | responsible for suitably incrementing the reference count of C<val> before |
| 229 | the call, and decrementing it if the function returned C<NULL>. Effectively |
| 230 | a successful C<hv_store> takes ownership of one reference to C<val>. This is |
| 231 | usually what you want; a newly created SV has a reference count of one, so |
| 232 | if all your code does is create SVs then store them in a hash, C<hv_store> |
| 233 | will own the only reference to the new SV, and your code doesn't need to do |
| 234 | anything further to tidy up. C<hv_store> is not implemented as a call to |
| 235 | C<hv_store_ent>, and does not create a temporary SV for the key, so if your |
| 236 | key data is not already in SV form then use C<hv_store> in preference to |
| 237 | C<hv_store_ent>. |
| 238 | |
| 239 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
| 240 | information on how to use this function on tied hashes. |
| 241 | |
| 242 | =for apidoc hv_store_ent |
| 243 | |
| 244 | Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash> |
| 245 | parameter is the precomputed hash value; if it is zero then Perl will |
| 246 | compute it. The return value is the new hash entry so created. It will be |
| 247 | C<NULL> if the operation failed or if the value did not need to be actually |
| 248 | stored within the hash (as in the case of tied hashes). Otherwise the |
| 249 | contents of the return value can be accessed using the C<He?> macros |
| 250 | described here. Note that the caller is responsible for suitably |
| 251 | incrementing the reference count of C<val> before the call, and |
| 252 | decrementing it if the function returned NULL. Effectively a successful |
| 253 | C<hv_store_ent> takes ownership of one reference to C<val>. This is |
| 254 | usually what you want; a newly created SV has a reference count of one, so |
| 255 | if all your code does is create SVs then store them in a hash, C<hv_store> |
| 256 | will own the only reference to the new SV, and your code doesn't need to do |
| 257 | anything further to tidy up. Note that C<hv_store_ent> only reads the C<key>; |
| 258 | unlike C<val> it does not take ownership of it, so maintaining the correct |
| 259 | reference count on C<key> is entirely the caller's responsibility. C<hv_store> |
| 260 | is not implemented as a call to C<hv_store_ent>, and does not create a temporary |
| 261 | SV for the key, so if your key data is not already in SV form then use |
| 262 | C<hv_store> in preference to C<hv_store_ent>. |
| 263 | |
| 264 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
| 265 | information on how to use this function on tied hashes. |
| 266 | |
| 267 | =for apidoc hv_exists |
| 268 | |
| 269 | Returns a boolean indicating whether the specified hash key exists. The |
| 270 | absolute value of C<klen> is the length of the key. If C<klen> is |
| 271 | negative the key is assumed to be in UTF-8-encoded Unicode. |
| 272 | |
| 273 | =for apidoc hv_fetch |
| 274 | |
| 275 | Returns the SV which corresponds to the specified key in the hash. |
| 276 | The absolute value of C<klen> is the length of the key. If C<klen> is |
| 277 | negative the key is assumed to be in UTF-8-encoded Unicode. If |
| 278 | C<lval> is set then the fetch will be part of a store. This means that if |
| 279 | there is no value in the hash associated with the given key, then one is |
| 280 | created and a pointer to it is returned. The C<SV*> it points to can be |
| 281 | assigned to. But always check that the |
| 282 | return value is non-null before dereferencing it to an C<SV*>. |
| 283 | |
| 284 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
| 285 | information on how to use this function on tied hashes. |
| 286 | |
| 287 | =for apidoc hv_exists_ent |
| 288 | |
| 289 | Returns a boolean indicating whether |
| 290 | the specified hash key exists. C<hash> |
| 291 | can be a valid precomputed hash value, or 0 to ask for it to be |
| 292 | computed. |
| 293 | |
| 294 | =cut |
| 295 | */ |
| 296 | |
| 297 | /* returns an HE * structure with the all fields set */ |
| 298 | /* note that hent_val will be a mortal sv for MAGICAL hashes */ |
| 299 | /* |
| 300 | =for apidoc hv_fetch_ent |
| 301 | |
| 302 | Returns the hash entry which corresponds to the specified key in the hash. |
| 303 | C<hash> must be a valid precomputed hash number for the given C<key>, or 0 |
| 304 | if you want the function to compute it. IF C<lval> is set then the fetch |
| 305 | will be part of a store. Make sure the return value is non-null before |
| 306 | accessing it. The return value when C<hv> is a tied hash is a pointer to a |
| 307 | static location, so be sure to make a copy of the structure if you need to |
| 308 | store it somewhere. |
| 309 | |
| 310 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
| 311 | information on how to use this function on tied hashes. |
| 312 | |
| 313 | =cut |
| 314 | */ |
| 315 | |
| 316 | /* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */ |
| 317 | void * |
| 318 | Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32, |
| 319 | const int action, SV *val, const U32 hash) |
| 320 | { |
| 321 | STRLEN klen; |
| 322 | int flags; |
| 323 | |
| 324 | PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN; |
| 325 | |
| 326 | if (klen_i32 < 0) { |
| 327 | klen = -klen_i32; |
| 328 | flags = HVhek_UTF8; |
| 329 | } else { |
| 330 | klen = klen_i32; |
| 331 | flags = 0; |
| 332 | } |
| 333 | return hv_common(hv, NULL, key, klen, flags, action, val, hash); |
| 334 | } |
| 335 | |
| 336 | void * |
| 337 | Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, |
| 338 | int flags, int action, SV *val, U32 hash) |
| 339 | { |
| 340 | dVAR; |
| 341 | XPVHV* xhv; |
| 342 | HE *entry; |
| 343 | HE **oentry; |
| 344 | SV *sv; |
| 345 | bool is_utf8; |
| 346 | int masked_flags; |
| 347 | const int return_svp = action & HV_FETCH_JUST_SV; |
| 348 | HEK *keysv_hek = NULL; |
| 349 | |
| 350 | if (!hv) |
| 351 | return NULL; |
| 352 | if (SvTYPE(hv) == (svtype)SVTYPEMASK) |
| 353 | return NULL; |
| 354 | |
| 355 | assert(SvTYPE(hv) == SVt_PVHV); |
| 356 | |
| 357 | if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) { |
| 358 | MAGIC* mg; |
| 359 | if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) { |
| 360 | struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr; |
| 361 | if (uf->uf_set == NULL) { |
| 362 | SV* obj = mg->mg_obj; |
| 363 | |
| 364 | if (!keysv) { |
| 365 | keysv = newSVpvn_flags(key, klen, SVs_TEMP | |
| 366 | ((flags & HVhek_UTF8) |
| 367 | ? SVf_UTF8 : 0)); |
| 368 | } |
| 369 | |
| 370 | mg->mg_obj = keysv; /* pass key */ |
| 371 | uf->uf_index = action; /* pass action */ |
| 372 | magic_getuvar(MUTABLE_SV(hv), mg); |
| 373 | keysv = mg->mg_obj; /* may have changed */ |
| 374 | mg->mg_obj = obj; |
| 375 | |
| 376 | /* If the key may have changed, then we need to invalidate |
| 377 | any passed-in computed hash value. */ |
| 378 | hash = 0; |
| 379 | } |
| 380 | } |
| 381 | } |
| 382 | if (keysv) { |
| 383 | if (flags & HVhek_FREEKEY) |
| 384 | Safefree(key); |
| 385 | key = SvPV_const(keysv, klen); |
| 386 | is_utf8 = (SvUTF8(keysv) != 0); |
| 387 | if (SvIsCOW_shared_hash(keysv)) { |
| 388 | flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0); |
| 389 | } else { |
| 390 | flags = is_utf8 ? HVhek_UTF8 : 0; |
| 391 | } |
| 392 | } else { |
| 393 | is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE); |
| 394 | } |
| 395 | |
| 396 | if (action & HV_DELETE) { |
| 397 | return (void *) hv_delete_common(hv, keysv, key, klen, |
| 398 | flags, action, hash); |
| 399 | } |
| 400 | |
| 401 | xhv = (XPVHV*)SvANY(hv); |
| 402 | if (SvMAGICAL(hv)) { |
| 403 | if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) { |
| 404 | if (mg_find((const SV *)hv, PERL_MAGIC_tied) |
| 405 | || SvGMAGICAL((const SV *)hv)) |
| 406 | { |
| 407 | /* FIXME should be able to skimp on the HE/HEK here when |
| 408 | HV_FETCH_JUST_SV is true. */ |
| 409 | if (!keysv) { |
| 410 | keysv = newSVpvn_utf8(key, klen, is_utf8); |
| 411 | } else { |
| 412 | keysv = newSVsv(keysv); |
| 413 | } |
| 414 | sv = sv_newmortal(); |
| 415 | mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY); |
| 416 | |
| 417 | /* grab a fake HE/HEK pair from the pool or make a new one */ |
| 418 | entry = PL_hv_fetch_ent_mh; |
| 419 | if (entry) |
| 420 | PL_hv_fetch_ent_mh = HeNEXT(entry); |
| 421 | else { |
| 422 | char *k; |
| 423 | entry = new_HE(); |
| 424 | Newx(k, HEK_BASESIZE + sizeof(const SV *), char); |
| 425 | HeKEY_hek(entry) = (HEK*)k; |
| 426 | } |
| 427 | HeNEXT(entry) = NULL; |
| 428 | HeSVKEY_set(entry, keysv); |
| 429 | HeVAL(entry) = sv; |
| 430 | sv_upgrade(sv, SVt_PVLV); |
| 431 | LvTYPE(sv) = 'T'; |
| 432 | /* so we can free entry when freeing sv */ |
| 433 | LvTARG(sv) = MUTABLE_SV(entry); |
| 434 | |
| 435 | /* XXX remove at some point? */ |
| 436 | if (flags & HVhek_FREEKEY) |
| 437 | Safefree(key); |
| 438 | |
| 439 | if (return_svp) { |
| 440 | return entry ? (void *) &HeVAL(entry) : NULL; |
| 441 | } |
| 442 | return (void *) entry; |
| 443 | } |
| 444 | #ifdef ENV_IS_CASELESS |
| 445 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 446 | U32 i; |
| 447 | for (i = 0; i < klen; ++i) |
| 448 | if (isLOWER(key[i])) { |
| 449 | /* Would be nice if we had a routine to do the |
| 450 | copy and upercase in a single pass through. */ |
| 451 | const char * const nkey = strupr(savepvn(key,klen)); |
| 452 | /* Note that this fetch is for nkey (the uppercased |
| 453 | key) whereas the store is for key (the original) */ |
| 454 | void *result = hv_common(hv, NULL, nkey, klen, |
| 455 | HVhek_FREEKEY, /* free nkey */ |
| 456 | 0 /* non-LVAL fetch */ |
| 457 | | HV_DISABLE_UVAR_XKEY |
| 458 | | return_svp, |
| 459 | NULL /* no value */, |
| 460 | 0 /* compute hash */); |
| 461 | if (!result && (action & HV_FETCH_LVALUE)) { |
| 462 | /* This call will free key if necessary. |
| 463 | Do it this way to encourage compiler to tail |
| 464 | call optimise. */ |
| 465 | result = hv_common(hv, keysv, key, klen, flags, |
| 466 | HV_FETCH_ISSTORE |
| 467 | | HV_DISABLE_UVAR_XKEY |
| 468 | | return_svp, |
| 469 | newSV(0), hash); |
| 470 | } else { |
| 471 | if (flags & HVhek_FREEKEY) |
| 472 | Safefree(key); |
| 473 | } |
| 474 | return result; |
| 475 | } |
| 476 | } |
| 477 | #endif |
| 478 | } /* ISFETCH */ |
| 479 | else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) { |
| 480 | if (mg_find((const SV *)hv, PERL_MAGIC_tied) |
| 481 | || SvGMAGICAL((const SV *)hv)) { |
| 482 | /* I don't understand why hv_exists_ent has svret and sv, |
| 483 | whereas hv_exists only had one. */ |
| 484 | SV * const svret = sv_newmortal(); |
| 485 | sv = sv_newmortal(); |
| 486 | |
| 487 | if (keysv || is_utf8) { |
| 488 | if (!keysv) { |
| 489 | keysv = newSVpvn_utf8(key, klen, TRUE); |
| 490 | } else { |
| 491 | keysv = newSVsv(keysv); |
| 492 | } |
| 493 | mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY); |
| 494 | } else { |
| 495 | mg_copy(MUTABLE_SV(hv), sv, key, klen); |
| 496 | } |
| 497 | if (flags & HVhek_FREEKEY) |
| 498 | Safefree(key); |
| 499 | { |
| 500 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_tiedelem); |
| 501 | if (mg) |
| 502 | magic_existspack(svret, mg); |
| 503 | } |
| 504 | /* This cast somewhat evil, but I'm merely using NULL/ |
| 505 | not NULL to return the boolean exists. |
| 506 | And I know hv is not NULL. */ |
| 507 | return SvTRUE(svret) ? (void *)hv : NULL; |
| 508 | } |
| 509 | #ifdef ENV_IS_CASELESS |
| 510 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 511 | /* XXX This code isn't UTF8 clean. */ |
| 512 | char * const keysave = (char * const)key; |
| 513 | /* Will need to free this, so set FREEKEY flag. */ |
| 514 | key = savepvn(key,klen); |
| 515 | key = (const char*)strupr((char*)key); |
| 516 | is_utf8 = FALSE; |
| 517 | hash = 0; |
| 518 | keysv = 0; |
| 519 | |
| 520 | if (flags & HVhek_FREEKEY) { |
| 521 | Safefree(keysave); |
| 522 | } |
| 523 | flags |= HVhek_FREEKEY; |
| 524 | } |
| 525 | #endif |
| 526 | } /* ISEXISTS */ |
| 527 | else if (action & HV_FETCH_ISSTORE) { |
| 528 | bool needs_copy; |
| 529 | bool needs_store; |
| 530 | hv_magic_check (hv, &needs_copy, &needs_store); |
| 531 | if (needs_copy) { |
| 532 | const bool save_taint = TAINT_get; |
| 533 | if (keysv || is_utf8) { |
| 534 | if (!keysv) { |
| 535 | keysv = newSVpvn_utf8(key, klen, TRUE); |
| 536 | } |
| 537 | if (TAINTING_get) |
| 538 | TAINT_set(SvTAINTED(keysv)); |
| 539 | keysv = sv_2mortal(newSVsv(keysv)); |
| 540 | mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY); |
| 541 | } else { |
| 542 | mg_copy(MUTABLE_SV(hv), val, key, klen); |
| 543 | } |
| 544 | |
| 545 | TAINT_IF(save_taint); |
| 546 | #ifdef NO_TAINT_SUPPORT |
| 547 | PERL_UNUSED_VAR(save_taint); |
| 548 | #endif |
| 549 | if (!needs_store) { |
| 550 | if (flags & HVhek_FREEKEY) |
| 551 | Safefree(key); |
| 552 | return NULL; |
| 553 | } |
| 554 | #ifdef ENV_IS_CASELESS |
| 555 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 556 | /* XXX This code isn't UTF8 clean. */ |
| 557 | const char *keysave = key; |
| 558 | /* Will need to free this, so set FREEKEY flag. */ |
| 559 | key = savepvn(key,klen); |
| 560 | key = (const char*)strupr((char*)key); |
| 561 | is_utf8 = FALSE; |
| 562 | hash = 0; |
| 563 | keysv = 0; |
| 564 | |
| 565 | if (flags & HVhek_FREEKEY) { |
| 566 | Safefree(keysave); |
| 567 | } |
| 568 | flags |= HVhek_FREEKEY; |
| 569 | } |
| 570 | #endif |
| 571 | } |
| 572 | } /* ISSTORE */ |
| 573 | } /* SvMAGICAL */ |
| 574 | |
| 575 | if (!HvARRAY(hv)) { |
| 576 | if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE)) |
| 577 | #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */ |
| 578 | || (SvRMAGICAL((const SV *)hv) |
| 579 | && mg_find((const SV *)hv, PERL_MAGIC_env)) |
| 580 | #endif |
| 581 | ) { |
| 582 | char *array; |
| 583 | Newxz(array, |
| 584 | PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), |
| 585 | char); |
| 586 | HvARRAY(hv) = (HE**)array; |
| 587 | } |
| 588 | #ifdef DYNAMIC_ENV_FETCH |
| 589 | else if (action & HV_FETCH_ISEXISTS) { |
| 590 | /* for an %ENV exists, if we do an insert it's by a recursive |
| 591 | store call, so avoid creating HvARRAY(hv) right now. */ |
| 592 | } |
| 593 | #endif |
| 594 | else { |
| 595 | /* XXX remove at some point? */ |
| 596 | if (flags & HVhek_FREEKEY) |
| 597 | Safefree(key); |
| 598 | |
| 599 | return NULL; |
| 600 | } |
| 601 | } |
| 602 | |
| 603 | if (is_utf8 && !(flags & HVhek_KEYCANONICAL)) { |
| 604 | char * const keysave = (char *)key; |
| 605 | key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); |
| 606 | if (is_utf8) |
| 607 | flags |= HVhek_UTF8; |
| 608 | else |
| 609 | flags &= ~HVhek_UTF8; |
| 610 | if (key != keysave) { |
| 611 | if (flags & HVhek_FREEKEY) |
| 612 | Safefree(keysave); |
| 613 | flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
| 614 | /* If the caller calculated a hash, it was on the sequence of |
| 615 | octets that are the UTF-8 form. We've now changed the sequence |
| 616 | of octets stored to that of the equivalent byte representation, |
| 617 | so the hash we need is different. */ |
| 618 | hash = 0; |
| 619 | } |
| 620 | } |
| 621 | |
| 622 | if (keysv && (SvIsCOW_shared_hash(keysv))) { |
| 623 | if (HvSHAREKEYS(hv)) |
| 624 | keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv)); |
| 625 | hash = SvSHARED_HASH(keysv); |
| 626 | } |
| 627 | else if (!hash) |
| 628 | PERL_HASH(hash, key, klen); |
| 629 | |
| 630 | masked_flags = (flags & HVhek_MASK); |
| 631 | |
| 632 | #ifdef DYNAMIC_ENV_FETCH |
| 633 | if (!HvARRAY(hv)) entry = NULL; |
| 634 | else |
| 635 | #endif |
| 636 | { |
| 637 | entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)]; |
| 638 | } |
| 639 | |
| 640 | if (!entry) |
| 641 | goto not_found; |
| 642 | |
| 643 | if (keysv_hek) { |
| 644 | /* keysv is actually a HEK in disguise, so we can match just by |
| 645 | * comparing the HEK pointers in the HE chain. There is a slight |
| 646 | * caveat: on something like "\x80", which has both plain and utf8 |
| 647 | * representations, perl's hashes do encoding-insensitive lookups, |
| 648 | * but preserve the encoding of the stored key. Thus a particular |
| 649 | * key could map to two different HEKs in PL_strtab. We only |
| 650 | * conclude 'not found' if all the flags are the same; otherwise |
| 651 | * we fall back to a full search (this should only happen in rare |
| 652 | * cases). |
| 653 | */ |
| 654 | int keysv_flags = HEK_FLAGS(keysv_hek); |
| 655 | HE *orig_entry = entry; |
| 656 | |
| 657 | for (; entry; entry = HeNEXT(entry)) { |
| 658 | HEK *hek = HeKEY_hek(entry); |
| 659 | if (hek == keysv_hek) |
| 660 | goto found; |
| 661 | if (HEK_FLAGS(hek) != keysv_flags) |
| 662 | break; /* need to do full match */ |
| 663 | } |
| 664 | if (!entry) |
| 665 | goto not_found; |
| 666 | /* failed on shortcut - do full search loop */ |
| 667 | entry = orig_entry; |
| 668 | } |
| 669 | |
| 670 | for (; entry; entry = HeNEXT(entry)) { |
| 671 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
| 672 | continue; |
| 673 | if (HeKLEN(entry) != (I32)klen) |
| 674 | continue; |
| 675 | if (memNE(HeKEY(entry),key,klen)) /* is this it? */ |
| 676 | continue; |
| 677 | if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8) |
| 678 | continue; |
| 679 | |
| 680 | found: |
| 681 | if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) { |
| 682 | if (HeKFLAGS(entry) != masked_flags) { |
| 683 | /* We match if HVhek_UTF8 bit in our flags and hash key's |
| 684 | match. But if entry was set previously with HVhek_WASUTF8 |
| 685 | and key now doesn't (or vice versa) then we should change |
| 686 | the key's flag, as this is assignment. */ |
| 687 | if (HvSHAREKEYS(hv)) { |
| 688 | /* Need to swap the key we have for a key with the flags we |
| 689 | need. As keys are shared we can't just write to the |
| 690 | flag, so we share the new one, unshare the old one. */ |
| 691 | HEK * const new_hek = share_hek_flags(key, klen, hash, |
| 692 | masked_flags); |
| 693 | unshare_hek (HeKEY_hek(entry)); |
| 694 | HeKEY_hek(entry) = new_hek; |
| 695 | } |
| 696 | else if (hv == PL_strtab) { |
| 697 | /* PL_strtab is usually the only hash without HvSHAREKEYS, |
| 698 | so putting this test here is cheap */ |
| 699 | if (flags & HVhek_FREEKEY) |
| 700 | Safefree(key); |
| 701 | Perl_croak(aTHX_ S_strtab_error, |
| 702 | action & HV_FETCH_LVALUE ? "fetch" : "store"); |
| 703 | } |
| 704 | else |
| 705 | HeKFLAGS(entry) = masked_flags; |
| 706 | if (masked_flags & HVhek_ENABLEHVKFLAGS) |
| 707 | HvHASKFLAGS_on(hv); |
| 708 | } |
| 709 | if (HeVAL(entry) == &PL_sv_placeholder) { |
| 710 | /* yes, can store into placeholder slot */ |
| 711 | if (action & HV_FETCH_LVALUE) { |
| 712 | if (SvMAGICAL(hv)) { |
| 713 | /* This preserves behaviour with the old hv_fetch |
| 714 | implementation which at this point would bail out |
| 715 | with a break; (at "if we find a placeholder, we |
| 716 | pretend we haven't found anything") |
| 717 | |
| 718 | That break mean that if a placeholder were found, it |
| 719 | caused a call into hv_store, which in turn would |
| 720 | check magic, and if there is no magic end up pretty |
| 721 | much back at this point (in hv_store's code). */ |
| 722 | break; |
| 723 | } |
| 724 | /* LVAL fetch which actually needs a store. */ |
| 725 | val = newSV(0); |
| 726 | HvPLACEHOLDERS(hv)--; |
| 727 | } else { |
| 728 | /* store */ |
| 729 | if (val != &PL_sv_placeholder) |
| 730 | HvPLACEHOLDERS(hv)--; |
| 731 | } |
| 732 | HeVAL(entry) = val; |
| 733 | } else if (action & HV_FETCH_ISSTORE) { |
| 734 | SvREFCNT_dec(HeVAL(entry)); |
| 735 | HeVAL(entry) = val; |
| 736 | } |
| 737 | } else if (HeVAL(entry) == &PL_sv_placeholder) { |
| 738 | /* if we find a placeholder, we pretend we haven't found |
| 739 | anything */ |
| 740 | break; |
| 741 | } |
| 742 | if (flags & HVhek_FREEKEY) |
| 743 | Safefree(key); |
| 744 | if (return_svp) { |
| 745 | return (void *) &HeVAL(entry); |
| 746 | } |
| 747 | return entry; |
| 748 | } |
| 749 | |
| 750 | not_found: |
| 751 | #ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */ |
| 752 | if (!(action & HV_FETCH_ISSTORE) |
| 753 | && SvRMAGICAL((const SV *)hv) |
| 754 | && mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 755 | unsigned long len; |
| 756 | const char * const env = PerlEnv_ENVgetenv_len(key,&len); |
| 757 | if (env) { |
| 758 | sv = newSVpvn(env,len); |
| 759 | SvTAINTED_on(sv); |
| 760 | return hv_common(hv, keysv, key, klen, flags, |
| 761 | HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp, |
| 762 | sv, hash); |
| 763 | } |
| 764 | } |
| 765 | #endif |
| 766 | |
| 767 | if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) { |
| 768 | hv_notallowed(flags, key, klen, |
| 769 | "Attempt to access disallowed key '%"SVf"' in" |
| 770 | " a restricted hash"); |
| 771 | } |
| 772 | if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) { |
| 773 | /* Not doing some form of store, so return failure. */ |
| 774 | if (flags & HVhek_FREEKEY) |
| 775 | Safefree(key); |
| 776 | return NULL; |
| 777 | } |
| 778 | if (action & HV_FETCH_LVALUE) { |
| 779 | val = action & HV_FETCH_EMPTY_HE ? NULL : newSV(0); |
| 780 | if (SvMAGICAL(hv)) { |
| 781 | /* At this point the old hv_fetch code would call to hv_store, |
| 782 | which in turn might do some tied magic. So we need to make that |
| 783 | magic check happen. */ |
| 784 | /* gonna assign to this, so it better be there */ |
| 785 | /* If a fetch-as-store fails on the fetch, then the action is to |
| 786 | recurse once into "hv_store". If we didn't do this, then that |
| 787 | recursive call would call the key conversion routine again. |
| 788 | However, as we replace the original key with the converted |
| 789 | key, this would result in a double conversion, which would show |
| 790 | up as a bug if the conversion routine is not idempotent. |
| 791 | Hence the use of HV_DISABLE_UVAR_XKEY. */ |
| 792 | return hv_common(hv, keysv, key, klen, flags, |
| 793 | HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp, |
| 794 | val, hash); |
| 795 | /* XXX Surely that could leak if the fetch-was-store fails? |
| 796 | Just like the hv_fetch. */ |
| 797 | } |
| 798 | } |
| 799 | |
| 800 | /* Welcome to hv_store... */ |
| 801 | |
| 802 | if (!HvARRAY(hv)) { |
| 803 | /* Not sure if we can get here. I think the only case of oentry being |
| 804 | NULL is for %ENV with dynamic env fetch. But that should disappear |
| 805 | with magic in the previous code. */ |
| 806 | char *array; |
| 807 | Newxz(array, |
| 808 | PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), |
| 809 | char); |
| 810 | HvARRAY(hv) = (HE**)array; |
| 811 | } |
| 812 | |
| 813 | oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max]; |
| 814 | |
| 815 | entry = new_HE(); |
| 816 | /* share_hek_flags will do the free for us. This might be considered |
| 817 | bad API design. */ |
| 818 | if (HvSHAREKEYS(hv)) |
| 819 | HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags); |
| 820 | else if (hv == PL_strtab) { |
| 821 | /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting |
| 822 | this test here is cheap */ |
| 823 | if (flags & HVhek_FREEKEY) |
| 824 | Safefree(key); |
| 825 | Perl_croak(aTHX_ S_strtab_error, |
| 826 | action & HV_FETCH_LVALUE ? "fetch" : "store"); |
| 827 | } |
| 828 | else /* gotta do the real thing */ |
| 829 | HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags); |
| 830 | HeVAL(entry) = val; |
| 831 | |
| 832 | if (!*oentry && SvOOK(hv)) { |
| 833 | /* initial entry, and aux struct present. */ |
| 834 | struct xpvhv_aux *const aux = HvAUX(hv); |
| 835 | if (aux->xhv_fill_lazy) |
| 836 | ++aux->xhv_fill_lazy; |
| 837 | } |
| 838 | |
| 839 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 840 | /* This logic semi-randomizes the insert order in a bucket. |
| 841 | * Either we insert into the top, or the slot below the top, |
| 842 | * making it harder to see if there is a collision. We also |
| 843 | * reset the iterator randomizer if there is one. |
| 844 | */ |
| 845 | if ( *oentry && PL_HASH_RAND_BITS_ENABLED) { |
| 846 | PL_hash_rand_bits++; |
| 847 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); |
| 848 | if ( PL_hash_rand_bits & 1 ) { |
| 849 | HeNEXT(entry) = HeNEXT(*oentry); |
| 850 | HeNEXT(*oentry) = entry; |
| 851 | } else { |
| 852 | HeNEXT(entry) = *oentry; |
| 853 | *oentry = entry; |
| 854 | } |
| 855 | } else |
| 856 | #endif |
| 857 | { |
| 858 | HeNEXT(entry) = *oentry; |
| 859 | *oentry = entry; |
| 860 | } |
| 861 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 862 | if (SvOOK(hv)) { |
| 863 | /* Currently this makes various tests warn in annoying ways. |
| 864 | * So Silenced for now. - Yves | bogus end of comment =>* / |
| 865 | if (HvAUX(hv)->xhv_riter != -1) { |
| 866 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), |
| 867 | "[TESTING] Inserting into a hash during each() traversal results in undefined behavior" |
| 868 | pTHX__FORMAT |
| 869 | pTHX__VALUE); |
| 870 | } |
| 871 | */ |
| 872 | if (PL_HASH_RAND_BITS_ENABLED) { |
| 873 | if (PL_HASH_RAND_BITS_ENABLED == 1) |
| 874 | PL_hash_rand_bits += (PTRV)entry + 1; /* we don't bother to use ptr_hash here */ |
| 875 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); |
| 876 | } |
| 877 | HvAUX(hv)->xhv_rand= (U32)PL_hash_rand_bits; |
| 878 | } |
| 879 | #endif |
| 880 | |
| 881 | if (val == &PL_sv_placeholder) |
| 882 | HvPLACEHOLDERS(hv)++; |
| 883 | if (masked_flags & HVhek_ENABLEHVKFLAGS) |
| 884 | HvHASKFLAGS_on(hv); |
| 885 | |
| 886 | xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */ |
| 887 | if ( DO_HSPLIT(xhv) ) { |
| 888 | const STRLEN oldsize = xhv->xhv_max + 1; |
| 889 | const U32 items = (U32)HvPLACEHOLDERS_get(hv); |
| 890 | |
| 891 | if (items /* hash has placeholders */ |
| 892 | && !SvREADONLY(hv) /* but is not a restricted hash */) { |
| 893 | /* If this hash previously was a "restricted hash" and had |
| 894 | placeholders, but the "restricted" flag has been turned off, |
| 895 | then the placeholders no longer serve any useful purpose. |
| 896 | However, they have the downsides of taking up RAM, and adding |
| 897 | extra steps when finding used values. It's safe to clear them |
| 898 | at this point, even though Storable rebuilds restricted hashes by |
| 899 | putting in all the placeholders (first) before turning on the |
| 900 | readonly flag, because Storable always pre-splits the hash. |
| 901 | If we're lucky, then we may clear sufficient placeholders to |
| 902 | avoid needing to split the hash at all. */ |
| 903 | clear_placeholders(hv, items); |
| 904 | if (DO_HSPLIT(xhv)) |
| 905 | hsplit(hv, oldsize, oldsize * 2); |
| 906 | } else |
| 907 | hsplit(hv, oldsize, oldsize * 2); |
| 908 | } |
| 909 | |
| 910 | if (return_svp) { |
| 911 | return entry ? (void *) &HeVAL(entry) : NULL; |
| 912 | } |
| 913 | return (void *) entry; |
| 914 | } |
| 915 | |
| 916 | STATIC void |
| 917 | S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store) |
| 918 | { |
| 919 | const MAGIC *mg = SvMAGIC(hv); |
| 920 | |
| 921 | PERL_ARGS_ASSERT_HV_MAGIC_CHECK; |
| 922 | |
| 923 | *needs_copy = FALSE; |
| 924 | *needs_store = TRUE; |
| 925 | while (mg) { |
| 926 | if (isUPPER(mg->mg_type)) { |
| 927 | *needs_copy = TRUE; |
| 928 | if (mg->mg_type == PERL_MAGIC_tied) { |
| 929 | *needs_store = FALSE; |
| 930 | return; /* We've set all there is to set. */ |
| 931 | } |
| 932 | } |
| 933 | mg = mg->mg_moremagic; |
| 934 | } |
| 935 | } |
| 936 | |
| 937 | /* |
| 938 | =for apidoc hv_scalar |
| 939 | |
| 940 | Evaluates the hash in scalar context and returns the result. Handles magic |
| 941 | when the hash is tied. |
| 942 | |
| 943 | =cut |
| 944 | */ |
| 945 | |
| 946 | SV * |
| 947 | Perl_hv_scalar(pTHX_ HV *hv) |
| 948 | { |
| 949 | SV *sv; |
| 950 | |
| 951 | PERL_ARGS_ASSERT_HV_SCALAR; |
| 952 | |
| 953 | if (SvRMAGICAL(hv)) { |
| 954 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied); |
| 955 | if (mg) |
| 956 | return magic_scalarpack(hv, mg); |
| 957 | } |
| 958 | |
| 959 | sv = sv_newmortal(); |
| 960 | if (HvTOTALKEYS((const HV *)hv)) |
| 961 | Perl_sv_setpvf(aTHX_ sv, "%ld/%ld", |
| 962 | (long)HvFILL(hv), (long)HvMAX(hv) + 1); |
| 963 | else |
| 964 | sv_setiv(sv, 0); |
| 965 | |
| 966 | return sv; |
| 967 | } |
| 968 | |
| 969 | /* |
| 970 | =for apidoc hv_delete |
| 971 | |
| 972 | Deletes a key/value pair in the hash. The value's SV is removed from |
| 973 | the hash, made mortal, and returned to the caller. The absolute |
| 974 | value of C<klen> is the length of the key. If C<klen> is negative the |
| 975 | key is assumed to be in UTF-8-encoded Unicode. The C<flags> value |
| 976 | will normally be zero; if set to C<G_DISCARD> then C<NULL> will be returned. |
| 977 | C<NULL> will also be returned if the key is not found. |
| 978 | |
| 979 | =for apidoc hv_delete_ent |
| 980 | |
| 981 | Deletes a key/value pair in the hash. The value SV is removed from the hash, |
| 982 | made mortal, and returned to the caller. The C<flags> value will normally be |
| 983 | zero; if set to C<G_DISCARD> then C<NULL> will be returned. C<NULL> will also |
| 984 | be returned if the key is not found. C<hash> can be a valid precomputed hash |
| 985 | value, or 0 to ask for it to be computed. |
| 986 | |
| 987 | =cut |
| 988 | */ |
| 989 | |
| 990 | STATIC SV * |
| 991 | S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, |
| 992 | int k_flags, I32 d_flags, U32 hash) |
| 993 | { |
| 994 | dVAR; |
| 995 | XPVHV* xhv; |
| 996 | HE *entry; |
| 997 | HE **oentry; |
| 998 | HE **first_entry; |
| 999 | bool is_utf8 = (k_flags & HVhek_UTF8) ? TRUE : FALSE; |
| 1000 | int masked_flags; |
| 1001 | HEK *keysv_hek = NULL; |
| 1002 | U8 mro_changes = 0; /* 1 = isa; 2 = package moved */ |
| 1003 | SV *sv; |
| 1004 | GV *gv = NULL; |
| 1005 | HV *stash = NULL; |
| 1006 | |
| 1007 | if (SvRMAGICAL(hv)) { |
| 1008 | bool needs_copy; |
| 1009 | bool needs_store; |
| 1010 | hv_magic_check (hv, &needs_copy, &needs_store); |
| 1011 | |
| 1012 | if (needs_copy) { |
| 1013 | SV *sv; |
| 1014 | entry = (HE *) hv_common(hv, keysv, key, klen, |
| 1015 | k_flags & ~HVhek_FREEKEY, |
| 1016 | HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY, |
| 1017 | NULL, hash); |
| 1018 | sv = entry ? HeVAL(entry) : NULL; |
| 1019 | if (sv) { |
| 1020 | if (SvMAGICAL(sv)) { |
| 1021 | mg_clear(sv); |
| 1022 | } |
| 1023 | if (!needs_store) { |
| 1024 | if (mg_find(sv, PERL_MAGIC_tiedelem)) { |
| 1025 | /* No longer an element */ |
| 1026 | sv_unmagic(sv, PERL_MAGIC_tiedelem); |
| 1027 | return sv; |
| 1028 | } |
| 1029 | return NULL; /* element cannot be deleted */ |
| 1030 | } |
| 1031 | #ifdef ENV_IS_CASELESS |
| 1032 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 1033 | /* XXX This code isn't UTF8 clean. */ |
| 1034 | keysv = newSVpvn_flags(key, klen, SVs_TEMP); |
| 1035 | if (k_flags & HVhek_FREEKEY) { |
| 1036 | Safefree(key); |
| 1037 | } |
| 1038 | key = strupr(SvPVX(keysv)); |
| 1039 | is_utf8 = 0; |
| 1040 | k_flags = 0; |
| 1041 | hash = 0; |
| 1042 | } |
| 1043 | #endif |
| 1044 | } |
| 1045 | } |
| 1046 | } |
| 1047 | xhv = (XPVHV*)SvANY(hv); |
| 1048 | if (!HvARRAY(hv)) |
| 1049 | return NULL; |
| 1050 | |
| 1051 | if (is_utf8 && !(k_flags & HVhek_KEYCANONICAL)) { |
| 1052 | const char * const keysave = key; |
| 1053 | key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); |
| 1054 | |
| 1055 | if (is_utf8) |
| 1056 | k_flags |= HVhek_UTF8; |
| 1057 | else |
| 1058 | k_flags &= ~HVhek_UTF8; |
| 1059 | if (key != keysave) { |
| 1060 | if (k_flags & HVhek_FREEKEY) { |
| 1061 | /* This shouldn't happen if our caller does what we expect, |
| 1062 | but strictly the API allows it. */ |
| 1063 | Safefree(keysave); |
| 1064 | } |
| 1065 | k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
| 1066 | } |
| 1067 | HvHASKFLAGS_on(MUTABLE_SV(hv)); |
| 1068 | } |
| 1069 | |
| 1070 | if (keysv && (SvIsCOW_shared_hash(keysv))) { |
| 1071 | if (HvSHAREKEYS(hv)) |
| 1072 | keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv)); |
| 1073 | hash = SvSHARED_HASH(keysv); |
| 1074 | } |
| 1075 | else if (!hash) |
| 1076 | PERL_HASH(hash, key, klen); |
| 1077 | |
| 1078 | masked_flags = (k_flags & HVhek_MASK); |
| 1079 | |
| 1080 | first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; |
| 1081 | entry = *oentry; |
| 1082 | |
| 1083 | if (!entry) |
| 1084 | goto not_found; |
| 1085 | |
| 1086 | if (keysv_hek) { |
| 1087 | /* keysv is actually a HEK in disguise, so we can match just by |
| 1088 | * comparing the HEK pointers in the HE chain. There is a slight |
| 1089 | * caveat: on something like "\x80", which has both plain and utf8 |
| 1090 | * representations, perl's hashes do encoding-insensitive lookups, |
| 1091 | * but preserve the encoding of the stored key. Thus a particular |
| 1092 | * key could map to two different HEKs in PL_strtab. We only |
| 1093 | * conclude 'not found' if all the flags are the same; otherwise |
| 1094 | * we fall back to a full search (this should only happen in rare |
| 1095 | * cases). |
| 1096 | */ |
| 1097 | int keysv_flags = HEK_FLAGS(keysv_hek); |
| 1098 | |
| 1099 | for (; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
| 1100 | HEK *hek = HeKEY_hek(entry); |
| 1101 | if (hek == keysv_hek) |
| 1102 | goto found; |
| 1103 | if (HEK_FLAGS(hek) != keysv_flags) |
| 1104 | break; /* need to do full match */ |
| 1105 | } |
| 1106 | if (!entry) |
| 1107 | goto not_found; |
| 1108 | /* failed on shortcut - do full search loop */ |
| 1109 | oentry = first_entry; |
| 1110 | entry = *oentry; |
| 1111 | } |
| 1112 | |
| 1113 | for (; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
| 1114 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
| 1115 | continue; |
| 1116 | if (HeKLEN(entry) != (I32)klen) |
| 1117 | continue; |
| 1118 | if (memNE(HeKEY(entry),key,klen)) /* is this it? */ |
| 1119 | continue; |
| 1120 | if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8) |
| 1121 | continue; |
| 1122 | |
| 1123 | found: |
| 1124 | if (hv == PL_strtab) { |
| 1125 | if (k_flags & HVhek_FREEKEY) |
| 1126 | Safefree(key); |
| 1127 | Perl_croak(aTHX_ S_strtab_error, "delete"); |
| 1128 | } |
| 1129 | |
| 1130 | /* if placeholder is here, it's already been deleted.... */ |
| 1131 | if (HeVAL(entry) == &PL_sv_placeholder) { |
| 1132 | if (k_flags & HVhek_FREEKEY) |
| 1133 | Safefree(key); |
| 1134 | return NULL; |
| 1135 | } |
| 1136 | if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) { |
| 1137 | hv_notallowed(k_flags, key, klen, |
| 1138 | "Attempt to delete readonly key '%"SVf"' from" |
| 1139 | " a restricted hash"); |
| 1140 | } |
| 1141 | if (k_flags & HVhek_FREEKEY) |
| 1142 | Safefree(key); |
| 1143 | |
| 1144 | /* If this is a stash and the key ends with ::, then someone is |
| 1145 | * deleting a package. |
| 1146 | */ |
| 1147 | if (HeVAL(entry) && HvENAME_get(hv)) { |
| 1148 | gv = (GV *)HeVAL(entry); |
| 1149 | if (keysv) key = SvPV(keysv, klen); |
| 1150 | if (( |
| 1151 | (klen > 1 && key[klen-2] == ':' && key[klen-1] == ':') |
| 1152 | || |
| 1153 | (klen == 1 && key[0] == ':') |
| 1154 | ) |
| 1155 | && (klen != 6 || hv!=PL_defstash || memNE(key,"main::",6)) |
| 1156 | && SvTYPE(gv) == SVt_PVGV && (stash = GvHV((GV *)gv)) |
| 1157 | && HvENAME_get(stash)) { |
| 1158 | /* A previous version of this code checked that the |
| 1159 | * GV was still in the symbol table by fetching the |
| 1160 | * GV with its name. That is not necessary (and |
| 1161 | * sometimes incorrect), as HvENAME cannot be set |
| 1162 | * on hv if it is not in the symtab. */ |
| 1163 | mro_changes = 2; |
| 1164 | /* Hang on to it for a bit. */ |
| 1165 | SvREFCNT_inc_simple_void_NN( |
| 1166 | sv_2mortal((SV *)gv) |
| 1167 | ); |
| 1168 | } |
| 1169 | else if (klen == 3 && strnEQ(key, "ISA", 3) && GvAV(gv)) { |
| 1170 | AV *isa = GvAV(gv); |
| 1171 | MAGIC *mg = mg_find((SV*)isa, PERL_MAGIC_isa); |
| 1172 | |
| 1173 | mro_changes = 1; |
| 1174 | if (mg) { |
| 1175 | if (mg->mg_obj == (SV*)gv) { |
| 1176 | /* This is the only stash this ISA was used for. |
| 1177 | * The isaelem magic asserts if there's no |
| 1178 | * isa magic on the array, so explicitly |
| 1179 | * remove the magic on both the array and its |
| 1180 | * elements. @ISA shouldn't be /too/ large. |
| 1181 | */ |
| 1182 | SV **svp, **end; |
| 1183 | strip_magic: |
| 1184 | svp = AvARRAY(isa); |
| 1185 | end = svp + AvFILLp(isa)+1; |
| 1186 | while (svp < end) { |
| 1187 | if (*svp) |
| 1188 | mg_free_type(*svp, PERL_MAGIC_isaelem); |
| 1189 | ++svp; |
| 1190 | } |
| 1191 | mg_free_type((SV*)GvAV(gv), PERL_MAGIC_isa); |
| 1192 | } |
| 1193 | else { |
| 1194 | /* mg_obj is an array of stashes |
| 1195 | Note that the array doesn't keep a reference |
| 1196 | count on the stashes. |
| 1197 | */ |
| 1198 | AV *av = (AV*)mg->mg_obj; |
| 1199 | SV **svp, **arrayp; |
| 1200 | SSize_t index; |
| 1201 | SSize_t items; |
| 1202 | |
| 1203 | assert(SvTYPE(mg->mg_obj) == SVt_PVAV); |
| 1204 | |
| 1205 | /* remove the stash from the magic array */ |
| 1206 | arrayp = svp = AvARRAY(av); |
| 1207 | items = AvFILLp(av) + 1; |
| 1208 | if (items == 1) { |
| 1209 | assert(*arrayp == (SV *)gv); |
| 1210 | mg->mg_obj = NULL; |
| 1211 | /* avoid a double free on the last stash */ |
| 1212 | AvFILLp(av) = -1; |
| 1213 | /* The magic isn't MGf_REFCOUNTED, so release |
| 1214 | * the array manually. |
| 1215 | */ |
| 1216 | SvREFCNT_dec_NN(av); |
| 1217 | goto strip_magic; |
| 1218 | } |
| 1219 | else { |
| 1220 | while (items--) { |
| 1221 | if (*svp == (SV*)gv) |
| 1222 | break; |
| 1223 | ++svp; |
| 1224 | } |
| 1225 | index = svp - arrayp; |
| 1226 | assert(index >= 0 && index <= AvFILLp(av)); |
| 1227 | if (index < AvFILLp(av)) { |
| 1228 | arrayp[index] = arrayp[AvFILLp(av)]; |
| 1229 | } |
| 1230 | arrayp[AvFILLp(av)] = NULL; |
| 1231 | --AvFILLp(av); |
| 1232 | } |
| 1233 | } |
| 1234 | } |
| 1235 | } |
| 1236 | } |
| 1237 | |
| 1238 | sv = d_flags & G_DISCARD ? HeVAL(entry) : sv_2mortal(HeVAL(entry)); |
| 1239 | HeVAL(entry) = &PL_sv_placeholder; |
| 1240 | if (sv) { |
| 1241 | /* deletion of method from stash */ |
| 1242 | if (isGV(sv) && isGV_with_GP(sv) && GvCVu(sv) |
| 1243 | && HvENAME_get(hv)) |
| 1244 | mro_method_changed_in(hv); |
| 1245 | } |
| 1246 | |
| 1247 | /* |
| 1248 | * If a restricted hash, rather than really deleting the entry, put |
| 1249 | * a placeholder there. This marks the key as being "approved", so |
| 1250 | * we can still access via not-really-existing key without raising |
| 1251 | * an error. |
| 1252 | */ |
| 1253 | if (SvREADONLY(hv)) |
| 1254 | /* We'll be saving this slot, so the number of allocated keys |
| 1255 | * doesn't go down, but the number placeholders goes up */ |
| 1256 | HvPLACEHOLDERS(hv)++; |
| 1257 | else { |
| 1258 | *oentry = HeNEXT(entry); |
| 1259 | if(!*first_entry && SvOOK(hv)) { |
| 1260 | /* removed last entry, and aux struct present. */ |
| 1261 | struct xpvhv_aux *const aux = HvAUX(hv); |
| 1262 | if (aux->xhv_fill_lazy) |
| 1263 | --aux->xhv_fill_lazy; |
| 1264 | } |
| 1265 | if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */) |
| 1266 | HvLAZYDEL_on(hv); |
| 1267 | else { |
| 1268 | if (SvOOK(hv) && HvLAZYDEL(hv) && |
| 1269 | entry == HeNEXT(HvAUX(hv)->xhv_eiter)) |
| 1270 | HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry); |
| 1271 | hv_free_ent(hv, entry); |
| 1272 | } |
| 1273 | xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */ |
| 1274 | if (xhv->xhv_keys == 0) |
| 1275 | HvHASKFLAGS_off(hv); |
| 1276 | } |
| 1277 | |
| 1278 | if (d_flags & G_DISCARD) { |
| 1279 | SvREFCNT_dec(sv); |
| 1280 | sv = NULL; |
| 1281 | } |
| 1282 | |
| 1283 | if (mro_changes == 1) mro_isa_changed_in(hv); |
| 1284 | else if (mro_changes == 2) |
| 1285 | mro_package_moved(NULL, stash, gv, 1); |
| 1286 | |
| 1287 | return sv; |
| 1288 | } |
| 1289 | |
| 1290 | not_found: |
| 1291 | if (SvREADONLY(hv)) { |
| 1292 | hv_notallowed(k_flags, key, klen, |
| 1293 | "Attempt to delete disallowed key '%"SVf"' from" |
| 1294 | " a restricted hash"); |
| 1295 | } |
| 1296 | |
| 1297 | if (k_flags & HVhek_FREEKEY) |
| 1298 | Safefree(key); |
| 1299 | return NULL; |
| 1300 | } |
| 1301 | |
| 1302 | |
| 1303 | STATIC void |
| 1304 | S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize) |
| 1305 | { |
| 1306 | STRLEN i = 0; |
| 1307 | char *a = (char*) HvARRAY(hv); |
| 1308 | HE **aep; |
| 1309 | |
| 1310 | bool do_aux= ( |
| 1311 | /* already have an HvAUX(hv) so we have to move it */ |
| 1312 | SvOOK(hv) || |
| 1313 | /* no HvAUX() but array we are going to allocate is large enough |
| 1314 | * there is no point in saving the space for the iterator, and |
| 1315 | * speeds up later traversals. */ |
| 1316 | ( ( hv != PL_strtab ) && ( newsize >= PERL_HV_ALLOC_AUX_SIZE ) ) |
| 1317 | ); |
| 1318 | |
| 1319 | PERL_ARGS_ASSERT_HSPLIT; |
| 1320 | |
| 1321 | PL_nomemok = TRUE; |
| 1322 | Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) |
| 1323 | + (do_aux ? sizeof(struct xpvhv_aux) : 0), char); |
| 1324 | PL_nomemok = FALSE; |
| 1325 | if (!a) { |
| 1326 | return; |
| 1327 | } |
| 1328 | |
| 1329 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 1330 | /* the idea of this is that we create a "random" value by hashing the address of |
| 1331 | * the array, we then use the low bit to decide if we insert at the top, or insert |
| 1332 | * second from top. After each such insert we rotate the hashed value. So we can |
| 1333 | * use the same hashed value over and over, and in normal build environments use |
| 1334 | * very few ops to do so. ROTL32() should produce a single machine operation. */ |
| 1335 | if (PL_HASH_RAND_BITS_ENABLED) { |
| 1336 | if (PL_HASH_RAND_BITS_ENABLED == 1) |
| 1337 | PL_hash_rand_bits += ptr_hash((PTRV)a); |
| 1338 | PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1); |
| 1339 | } |
| 1340 | #endif |
| 1341 | HvARRAY(hv) = (HE**) a; |
| 1342 | HvMAX(hv) = newsize - 1; |
| 1343 | /* before we zero the newly added memory, we |
| 1344 | * need to deal with the aux struct that may be there |
| 1345 | * or have been allocated by us*/ |
| 1346 | if (do_aux) { |
| 1347 | struct xpvhv_aux *const dest |
| 1348 | = (struct xpvhv_aux*) &a[newsize * sizeof(HE*)]; |
| 1349 | if (SvOOK(hv)) { |
| 1350 | /* alread have an aux, copy the old one in place. */ |
| 1351 | Move(&a[oldsize * sizeof(HE*)], dest, 1, struct xpvhv_aux); |
| 1352 | /* we reset the iterator's xhv_rand as well, so they get a totally new ordering */ |
| 1353 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 1354 | dest->xhv_rand = (U32)PL_hash_rand_bits; |
| 1355 | #endif |
| 1356 | /* For now, just reset the lazy fill counter. |
| 1357 | It would be possible to update the counter in the code below |
| 1358 | instead. */ |
| 1359 | dest->xhv_fill_lazy = 0; |
| 1360 | } else { |
| 1361 | /* no existing aux structure, but we allocated space for one |
| 1362 | * so initialize it properly. This unrolls hv_auxinit() a bit, |
| 1363 | * since we have to do the realloc anyway. */ |
| 1364 | /* first we set the iterator's xhv_rand so it can be copied into lastrand below */ |
| 1365 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 1366 | dest->xhv_rand = (U32)PL_hash_rand_bits; |
| 1367 | #endif |
| 1368 | /* this is the "non realloc" part of the hv_auxinit() */ |
| 1369 | (void)hv_auxinit_internal(dest); |
| 1370 | /* Turn on the OOK flag */ |
| 1371 | SvOOK_on(hv); |
| 1372 | } |
| 1373 | } |
| 1374 | /* now we can safely clear the second half */ |
| 1375 | Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/ |
| 1376 | |
| 1377 | if (!HvTOTALKEYS(hv)) /* skip rest if no entries */ |
| 1378 | return; |
| 1379 | |
| 1380 | newsize--; |
| 1381 | aep = (HE**)a; |
| 1382 | do { |
| 1383 | HE **oentry = aep + i; |
| 1384 | HE *entry = aep[i]; |
| 1385 | |
| 1386 | if (!entry) /* non-existent */ |
| 1387 | continue; |
| 1388 | do { |
| 1389 | U32 j = (HeHASH(entry) & newsize); |
| 1390 | if (j != (U32)i) { |
| 1391 | *oentry = HeNEXT(entry); |
| 1392 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 1393 | /* if the target cell is empty or PL_HASH_RAND_BITS_ENABLED is false |
| 1394 | * insert to top, otherwise rotate the bucket rand 1 bit, |
| 1395 | * and use the new low bit to decide if we insert at top, |
| 1396 | * or next from top. IOW, we only rotate on a collision.*/ |
| 1397 | if (aep[j] && PL_HASH_RAND_BITS_ENABLED) { |
| 1398 | PL_hash_rand_bits+= ROTL32(HeHASH(entry), 17); |
| 1399 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); |
| 1400 | if (PL_hash_rand_bits & 1) { |
| 1401 | HeNEXT(entry)= HeNEXT(aep[j]); |
| 1402 | HeNEXT(aep[j])= entry; |
| 1403 | } else { |
| 1404 | /* Note, this is structured in such a way as the optimizer |
| 1405 | * should eliminate the duplicated code here and below without |
| 1406 | * us needing to explicitly use a goto. */ |
| 1407 | HeNEXT(entry) = aep[j]; |
| 1408 | aep[j] = entry; |
| 1409 | } |
| 1410 | } else |
| 1411 | #endif |
| 1412 | { |
| 1413 | /* see comment above about duplicated code */ |
| 1414 | HeNEXT(entry) = aep[j]; |
| 1415 | aep[j] = entry; |
| 1416 | } |
| 1417 | } |
| 1418 | else { |
| 1419 | oentry = &HeNEXT(entry); |
| 1420 | } |
| 1421 | entry = *oentry; |
| 1422 | } while (entry); |
| 1423 | } while (i++ < oldsize); |
| 1424 | } |
| 1425 | |
| 1426 | void |
| 1427 | Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) |
| 1428 | { |
| 1429 | XPVHV* xhv = (XPVHV*)SvANY(hv); |
| 1430 | const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */ |
| 1431 | I32 newsize; |
| 1432 | char *a; |
| 1433 | |
| 1434 | PERL_ARGS_ASSERT_HV_KSPLIT; |
| 1435 | |
| 1436 | newsize = (I32) newmax; /* possible truncation here */ |
| 1437 | if (newsize != newmax || newmax <= oldsize) |
| 1438 | return; |
| 1439 | while ((newsize & (1 + ~newsize)) != newsize) { |
| 1440 | newsize &= ~(newsize & (1 + ~newsize)); /* get proper power of 2 */ |
| 1441 | } |
| 1442 | if (newsize < newmax) |
| 1443 | newsize *= 2; |
| 1444 | if (newsize < newmax) |
| 1445 | return; /* overflow detection */ |
| 1446 | |
| 1447 | a = (char *) HvARRAY(hv); |
| 1448 | if (a) { |
| 1449 | hsplit(hv, oldsize, newsize); |
| 1450 | } else { |
| 1451 | Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); |
| 1452 | xhv->xhv_max = --newsize; |
| 1453 | HvARRAY(hv) = (HE **) a; |
| 1454 | } |
| 1455 | } |
| 1456 | |
| 1457 | /* IMO this should also handle cases where hv_max is smaller than hv_keys |
| 1458 | * as tied hashes could play silly buggers and mess us around. We will |
| 1459 | * do the right thing during hv_store() afterwards, but still - Yves */ |
| 1460 | #define HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys) STMT_START {\ |
| 1461 | /* Can we use fewer buckets? (hv_max is always 2^n-1) */ \ |
| 1462 | if (hv_max < PERL_HASH_DEFAULT_HvMAX) { \ |
| 1463 | hv_max = PERL_HASH_DEFAULT_HvMAX; \ |
| 1464 | } else { \ |
| 1465 | while (hv_max > PERL_HASH_DEFAULT_HvMAX && hv_max + 1 >= hv_keys * 2) \ |
| 1466 | hv_max = hv_max / 2; \ |
| 1467 | } \ |
| 1468 | HvMAX(hv) = hv_max; \ |
| 1469 | } STMT_END |
| 1470 | |
| 1471 | |
| 1472 | HV * |
| 1473 | Perl_newHVhv(pTHX_ HV *ohv) |
| 1474 | { |
| 1475 | dVAR; |
| 1476 | HV * const hv = newHV(); |
| 1477 | STRLEN hv_max; |
| 1478 | |
| 1479 | if (!ohv || (!HvTOTALKEYS(ohv) && !SvMAGICAL((const SV *)ohv))) |
| 1480 | return hv; |
| 1481 | hv_max = HvMAX(ohv); |
| 1482 | |
| 1483 | if (!SvMAGICAL((const SV *)ohv)) { |
| 1484 | /* It's an ordinary hash, so copy it fast. AMS 20010804 */ |
| 1485 | STRLEN i; |
| 1486 | const bool shared = !!HvSHAREKEYS(ohv); |
| 1487 | HE **ents, ** const oents = (HE **)HvARRAY(ohv); |
| 1488 | char *a; |
| 1489 | Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char); |
| 1490 | ents = (HE**)a; |
| 1491 | |
| 1492 | /* In each bucket... */ |
| 1493 | for (i = 0; i <= hv_max; i++) { |
| 1494 | HE *prev = NULL; |
| 1495 | HE *oent = oents[i]; |
| 1496 | |
| 1497 | if (!oent) { |
| 1498 | ents[i] = NULL; |
| 1499 | continue; |
| 1500 | } |
| 1501 | |
| 1502 | /* Copy the linked list of entries. */ |
| 1503 | for (; oent; oent = HeNEXT(oent)) { |
| 1504 | const U32 hash = HeHASH(oent); |
| 1505 | const char * const key = HeKEY(oent); |
| 1506 | const STRLEN len = HeKLEN(oent); |
| 1507 | const int flags = HeKFLAGS(oent); |
| 1508 | HE * const ent = new_HE(); |
| 1509 | SV *const val = HeVAL(oent); |
| 1510 | |
| 1511 | HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val); |
| 1512 | HeKEY_hek(ent) |
| 1513 | = shared ? share_hek_flags(key, len, hash, flags) |
| 1514 | : save_hek_flags(key, len, hash, flags); |
| 1515 | if (prev) |
| 1516 | HeNEXT(prev) = ent; |
| 1517 | else |
| 1518 | ents[i] = ent; |
| 1519 | prev = ent; |
| 1520 | HeNEXT(ent) = NULL; |
| 1521 | } |
| 1522 | } |
| 1523 | |
| 1524 | HvMAX(hv) = hv_max; |
| 1525 | HvTOTALKEYS(hv) = HvTOTALKEYS(ohv); |
| 1526 | HvARRAY(hv) = ents; |
| 1527 | } /* not magical */ |
| 1528 | else { |
| 1529 | /* Iterate over ohv, copying keys and values one at a time. */ |
| 1530 | HE *entry; |
| 1531 | const I32 riter = HvRITER_get(ohv); |
| 1532 | HE * const eiter = HvEITER_get(ohv); |
| 1533 | STRLEN hv_keys = HvTOTALKEYS(ohv); |
| 1534 | |
| 1535 | HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys); |
| 1536 | |
| 1537 | hv_iterinit(ohv); |
| 1538 | while ((entry = hv_iternext_flags(ohv, 0))) { |
| 1539 | SV *val = hv_iterval(ohv,entry); |
| 1540 | SV * const keysv = HeSVKEY(entry); |
| 1541 | val = SvIMMORTAL(val) ? val : newSVsv(val); |
| 1542 | if (keysv) |
| 1543 | (void)hv_store_ent(hv, keysv, val, 0); |
| 1544 | else |
| 1545 | (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry), val, |
| 1546 | HeHASH(entry), HeKFLAGS(entry)); |
| 1547 | } |
| 1548 | HvRITER_set(ohv, riter); |
| 1549 | HvEITER_set(ohv, eiter); |
| 1550 | } |
| 1551 | |
| 1552 | return hv; |
| 1553 | } |
| 1554 | |
| 1555 | /* |
| 1556 | =for apidoc Am|HV *|hv_copy_hints_hv|HV *ohv |
| 1557 | |
| 1558 | A specialised version of L</newHVhv> for copying C<%^H>. C<ohv> must be |
| 1559 | a pointer to a hash (which may have C<%^H> magic, but should be generally |
| 1560 | non-magical), or C<NULL> (interpreted as an empty hash). The content |
| 1561 | of C<ohv> is copied to a new hash, which has the C<%^H>-specific magic |
| 1562 | added to it. A pointer to the new hash is returned. |
| 1563 | |
| 1564 | =cut |
| 1565 | */ |
| 1566 | |
| 1567 | HV * |
| 1568 | Perl_hv_copy_hints_hv(pTHX_ HV *const ohv) |
| 1569 | { |
| 1570 | HV * const hv = newHV(); |
| 1571 | |
| 1572 | if (ohv) { |
| 1573 | STRLEN hv_max = HvMAX(ohv); |
| 1574 | STRLEN hv_keys = HvTOTALKEYS(ohv); |
| 1575 | HE *entry; |
| 1576 | const I32 riter = HvRITER_get(ohv); |
| 1577 | HE * const eiter = HvEITER_get(ohv); |
| 1578 | |
| 1579 | ENTER; |
| 1580 | SAVEFREESV(hv); |
| 1581 | |
| 1582 | HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys); |
| 1583 | |
| 1584 | hv_iterinit(ohv); |
| 1585 | while ((entry = hv_iternext_flags(ohv, 0))) { |
| 1586 | SV *const sv = newSVsv(hv_iterval(ohv,entry)); |
| 1587 | SV *heksv = HeSVKEY(entry); |
| 1588 | if (!heksv && sv) heksv = newSVhek(HeKEY_hek(entry)); |
| 1589 | if (sv) sv_magic(sv, NULL, PERL_MAGIC_hintselem, |
| 1590 | (char *)heksv, HEf_SVKEY); |
| 1591 | if (heksv == HeSVKEY(entry)) |
| 1592 | (void)hv_store_ent(hv, heksv, sv, 0); |
| 1593 | else { |
| 1594 | (void)hv_common(hv, heksv, HeKEY(entry), HeKLEN(entry), |
| 1595 | HeKFLAGS(entry), HV_FETCH_ISSTORE|HV_FETCH_JUST_SV, sv, HeHASH(entry)); |
| 1596 | SvREFCNT_dec_NN(heksv); |
| 1597 | } |
| 1598 | } |
| 1599 | HvRITER_set(ohv, riter); |
| 1600 | HvEITER_set(ohv, eiter); |
| 1601 | |
| 1602 | SvREFCNT_inc_simple_void_NN(hv); |
| 1603 | LEAVE; |
| 1604 | } |
| 1605 | hv_magic(hv, NULL, PERL_MAGIC_hints); |
| 1606 | return hv; |
| 1607 | } |
| 1608 | #undef HV_SET_MAX_ADJUSTED_FOR_KEYS |
| 1609 | |
| 1610 | /* like hv_free_ent, but returns the SV rather than freeing it */ |
| 1611 | STATIC SV* |
| 1612 | S_hv_free_ent_ret(pTHX_ HV *hv, HE *entry) |
| 1613 | { |
| 1614 | SV *val; |
| 1615 | |
| 1616 | PERL_ARGS_ASSERT_HV_FREE_ENT_RET; |
| 1617 | |
| 1618 | val = HeVAL(entry); |
| 1619 | if (HeKLEN(entry) == HEf_SVKEY) { |
| 1620 | SvREFCNT_dec(HeKEY_sv(entry)); |
| 1621 | Safefree(HeKEY_hek(entry)); |
| 1622 | } |
| 1623 | else if (HvSHAREKEYS(hv)) |
| 1624 | unshare_hek(HeKEY_hek(entry)); |
| 1625 | else |
| 1626 | Safefree(HeKEY_hek(entry)); |
| 1627 | del_HE(entry); |
| 1628 | return val; |
| 1629 | } |
| 1630 | |
| 1631 | |
| 1632 | void |
| 1633 | Perl_hv_free_ent(pTHX_ HV *hv, HE *entry) |
| 1634 | { |
| 1635 | SV *val; |
| 1636 | |
| 1637 | PERL_ARGS_ASSERT_HV_FREE_ENT; |
| 1638 | |
| 1639 | if (!entry) |
| 1640 | return; |
| 1641 | val = hv_free_ent_ret(hv, entry); |
| 1642 | SvREFCNT_dec(val); |
| 1643 | } |
| 1644 | |
| 1645 | |
| 1646 | void |
| 1647 | Perl_hv_delayfree_ent(pTHX_ HV *hv, HE *entry) |
| 1648 | { |
| 1649 | PERL_ARGS_ASSERT_HV_DELAYFREE_ENT; |
| 1650 | |
| 1651 | if (!entry) |
| 1652 | return; |
| 1653 | /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */ |
| 1654 | sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */ |
| 1655 | if (HeKLEN(entry) == HEf_SVKEY) { |
| 1656 | sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry))); |
| 1657 | } |
| 1658 | hv_free_ent(hv, entry); |
| 1659 | } |
| 1660 | |
| 1661 | /* |
| 1662 | =for apidoc hv_clear |
| 1663 | |
| 1664 | Frees the all the elements of a hash, leaving it empty. |
| 1665 | The XS equivalent of C<%hash = ()>. See also L</hv_undef>. |
| 1666 | |
| 1667 | See L</av_clear> for a note about the hash possibly being invalid on |
| 1668 | return. |
| 1669 | |
| 1670 | =cut |
| 1671 | */ |
| 1672 | |
| 1673 | void |
| 1674 | Perl_hv_clear(pTHX_ HV *hv) |
| 1675 | { |
| 1676 | dVAR; |
| 1677 | XPVHV* xhv; |
| 1678 | if (!hv) |
| 1679 | return; |
| 1680 | |
| 1681 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
| 1682 | |
| 1683 | xhv = (XPVHV*)SvANY(hv); |
| 1684 | |
| 1685 | ENTER; |
| 1686 | SAVEFREESV(SvREFCNT_inc_simple_NN(hv)); |
| 1687 | if (SvREADONLY(hv) && HvARRAY(hv) != NULL) { |
| 1688 | /* restricted hash: convert all keys to placeholders */ |
| 1689 | STRLEN i; |
| 1690 | for (i = 0; i <= xhv->xhv_max; i++) { |
| 1691 | HE *entry = (HvARRAY(hv))[i]; |
| 1692 | for (; entry; entry = HeNEXT(entry)) { |
| 1693 | /* not already placeholder */ |
| 1694 | if (HeVAL(entry) != &PL_sv_placeholder) { |
| 1695 | if (HeVAL(entry)) { |
| 1696 | if (SvREADONLY(HeVAL(entry))) { |
| 1697 | SV* const keysv = hv_iterkeysv(entry); |
| 1698 | Perl_croak_nocontext( |
| 1699 | "Attempt to delete readonly key '%"SVf"' from a restricted hash", |
| 1700 | (void*)keysv); |
| 1701 | } |
| 1702 | SvREFCNT_dec_NN(HeVAL(entry)); |
| 1703 | } |
| 1704 | HeVAL(entry) = &PL_sv_placeholder; |
| 1705 | HvPLACEHOLDERS(hv)++; |
| 1706 | } |
| 1707 | } |
| 1708 | } |
| 1709 | } |
| 1710 | else { |
| 1711 | hfreeentries(hv); |
| 1712 | HvPLACEHOLDERS_set(hv, 0); |
| 1713 | |
| 1714 | if (SvRMAGICAL(hv)) |
| 1715 | mg_clear(MUTABLE_SV(hv)); |
| 1716 | |
| 1717 | HvHASKFLAGS_off(hv); |
| 1718 | } |
| 1719 | if (SvOOK(hv)) { |
| 1720 | if(HvENAME_get(hv)) |
| 1721 | mro_isa_changed_in(hv); |
| 1722 | HvEITER_set(hv, NULL); |
| 1723 | } |
| 1724 | LEAVE; |
| 1725 | } |
| 1726 | |
| 1727 | /* |
| 1728 | =for apidoc hv_clear_placeholders |
| 1729 | |
| 1730 | Clears any placeholders from a hash. If a restricted hash has any of its keys |
| 1731 | marked as readonly and the key is subsequently deleted, the key is not actually |
| 1732 | deleted but is marked by assigning it a value of C<&PL_sv_placeholder>. This tags |
| 1733 | it so it will be ignored by future operations such as iterating over the hash, |
| 1734 | but will still allow the hash to have a value reassigned to the key at some |
| 1735 | future point. This function clears any such placeholder keys from the hash. |
| 1736 | See C<L<Hash::Util::lock_keys()|Hash::Util/lock_keys>> for an example of its |
| 1737 | use. |
| 1738 | |
| 1739 | =cut |
| 1740 | */ |
| 1741 | |
| 1742 | void |
| 1743 | Perl_hv_clear_placeholders(pTHX_ HV *hv) |
| 1744 | { |
| 1745 | const U32 items = (U32)HvPLACEHOLDERS_get(hv); |
| 1746 | |
| 1747 | PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS; |
| 1748 | |
| 1749 | if (items) |
| 1750 | clear_placeholders(hv, items); |
| 1751 | } |
| 1752 | |
| 1753 | static void |
| 1754 | S_clear_placeholders(pTHX_ HV *hv, U32 items) |
| 1755 | { |
| 1756 | dVAR; |
| 1757 | I32 i; |
| 1758 | |
| 1759 | PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS; |
| 1760 | |
| 1761 | if (items == 0) |
| 1762 | return; |
| 1763 | |
| 1764 | i = HvMAX(hv); |
| 1765 | do { |
| 1766 | /* Loop down the linked list heads */ |
| 1767 | HE **oentry = &(HvARRAY(hv))[i]; |
| 1768 | HE *entry; |
| 1769 | |
| 1770 | while ((entry = *oentry)) { |
| 1771 | if (HeVAL(entry) == &PL_sv_placeholder) { |
| 1772 | *oentry = HeNEXT(entry); |
| 1773 | if (entry == HvEITER_get(hv)) |
| 1774 | HvLAZYDEL_on(hv); |
| 1775 | else { |
| 1776 | if (SvOOK(hv) && HvLAZYDEL(hv) && |
| 1777 | entry == HeNEXT(HvAUX(hv)->xhv_eiter)) |
| 1778 | HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry); |
| 1779 | hv_free_ent(hv, entry); |
| 1780 | } |
| 1781 | |
| 1782 | if (--items == 0) { |
| 1783 | /* Finished. */ |
| 1784 | I32 placeholders = HvPLACEHOLDERS_get(hv); |
| 1785 | HvTOTALKEYS(hv) -= (IV)placeholders; |
| 1786 | /* HvUSEDKEYS expanded */ |
| 1787 | if ((HvTOTALKEYS(hv) - placeholders) == 0) |
| 1788 | HvHASKFLAGS_off(hv); |
| 1789 | HvPLACEHOLDERS_set(hv, 0); |
| 1790 | return; |
| 1791 | } |
| 1792 | } else { |
| 1793 | oentry = &HeNEXT(entry); |
| 1794 | } |
| 1795 | } |
| 1796 | } while (--i >= 0); |
| 1797 | /* You can't get here, hence assertion should always fail. */ |
| 1798 | assert (items == 0); |
| 1799 | NOT_REACHED; /* NOTREACHED */ |
| 1800 | } |
| 1801 | |
| 1802 | STATIC void |
| 1803 | S_hfreeentries(pTHX_ HV *hv) |
| 1804 | { |
| 1805 | STRLEN index = 0; |
| 1806 | XPVHV * const xhv = (XPVHV*)SvANY(hv); |
| 1807 | SV *sv; |
| 1808 | |
| 1809 | PERL_ARGS_ASSERT_HFREEENTRIES; |
| 1810 | |
| 1811 | while ((sv = Perl_hfree_next_entry(aTHX_ hv, &index))||xhv->xhv_keys) { |
| 1812 | SvREFCNT_dec(sv); |
| 1813 | } |
| 1814 | } |
| 1815 | |
| 1816 | |
| 1817 | /* hfree_next_entry() |
| 1818 | * For use only by S_hfreeentries() and sv_clear(). |
| 1819 | * Delete the next available HE from hv and return the associated SV. |
| 1820 | * Returns null on empty hash. Nevertheless null is not a reliable |
| 1821 | * indicator that the hash is empty, as the deleted entry may have a |
| 1822 | * null value. |
| 1823 | * indexp is a pointer to the current index into HvARRAY. The index should |
| 1824 | * initially be set to 0. hfree_next_entry() may update it. */ |
| 1825 | |
| 1826 | SV* |
| 1827 | Perl_hfree_next_entry(pTHX_ HV *hv, STRLEN *indexp) |
| 1828 | { |
| 1829 | struct xpvhv_aux *iter; |
| 1830 | HE *entry; |
| 1831 | HE ** array; |
| 1832 | #ifdef DEBUGGING |
| 1833 | STRLEN orig_index = *indexp; |
| 1834 | #endif |
| 1835 | |
| 1836 | PERL_ARGS_ASSERT_HFREE_NEXT_ENTRY; |
| 1837 | |
| 1838 | if (SvOOK(hv) && ((iter = HvAUX(hv)))) { |
| 1839 | if ((entry = iter->xhv_eiter)) { |
| 1840 | /* the iterator may get resurrected after each |
| 1841 | * destructor call, so check each time */ |
| 1842 | if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ |
| 1843 | HvLAZYDEL_off(hv); |
| 1844 | hv_free_ent(hv, entry); |
| 1845 | /* warning: at this point HvARRAY may have been |
| 1846 | * re-allocated, HvMAX changed etc */ |
| 1847 | } |
| 1848 | iter = HvAUX(hv); /* may have been realloced */ |
| 1849 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
| 1850 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
| 1851 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 1852 | iter->xhv_last_rand = iter->xhv_rand; |
| 1853 | #endif |
| 1854 | } |
| 1855 | /* Reset any cached HvFILL() to "unknown". It's unlikely that anyone |
| 1856 | will actually call HvFILL() on a hash under destruction, so it |
| 1857 | seems pointless attempting to track the number of keys remaining. |
| 1858 | But if they do, we want to reset it again. */ |
| 1859 | if (iter->xhv_fill_lazy) |
| 1860 | iter->xhv_fill_lazy = 0; |
| 1861 | } |
| 1862 | |
| 1863 | if (!((XPVHV*)SvANY(hv))->xhv_keys) |
| 1864 | return NULL; |
| 1865 | |
| 1866 | array = HvARRAY(hv); |
| 1867 | assert(array); |
| 1868 | while ( ! ((entry = array[*indexp])) ) { |
| 1869 | if ((*indexp)++ >= HvMAX(hv)) |
| 1870 | *indexp = 0; |
| 1871 | assert(*indexp != orig_index); |
| 1872 | } |
| 1873 | array[*indexp] = HeNEXT(entry); |
| 1874 | ((XPVHV*) SvANY(hv))->xhv_keys--; |
| 1875 | |
| 1876 | if ( PL_phase != PERL_PHASE_DESTRUCT && HvENAME(hv) |
| 1877 | && HeVAL(entry) && isGV(HeVAL(entry)) |
| 1878 | && GvHV(HeVAL(entry)) && HvENAME(GvHV(HeVAL(entry))) |
| 1879 | ) { |
| 1880 | STRLEN klen; |
| 1881 | const char * const key = HePV(entry,klen); |
| 1882 | if ((klen > 1 && key[klen-1]==':' && key[klen-2]==':') |
| 1883 | || (klen == 1 && key[0] == ':')) { |
| 1884 | mro_package_moved( |
| 1885 | NULL, GvHV(HeVAL(entry)), |
| 1886 | (GV *)HeVAL(entry), 0 |
| 1887 | ); |
| 1888 | } |
| 1889 | } |
| 1890 | return hv_free_ent_ret(hv, entry); |
| 1891 | } |
| 1892 | |
| 1893 | |
| 1894 | /* |
| 1895 | =for apidoc hv_undef |
| 1896 | |
| 1897 | Undefines the hash. The XS equivalent of C<undef(%hash)>. |
| 1898 | |
| 1899 | As well as freeing all the elements of the hash (like C<hv_clear()>), this |
| 1900 | also frees any auxiliary data and storage associated with the hash. |
| 1901 | |
| 1902 | See L</av_clear> for a note about the hash possibly being invalid on |
| 1903 | return. |
| 1904 | |
| 1905 | =cut |
| 1906 | */ |
| 1907 | |
| 1908 | void |
| 1909 | Perl_hv_undef_flags(pTHX_ HV *hv, U32 flags) |
| 1910 | { |
| 1911 | XPVHV* xhv; |
| 1912 | bool save; |
| 1913 | |
| 1914 | if (!hv) |
| 1915 | return; |
| 1916 | save = !!SvREFCNT(hv); |
| 1917 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
| 1918 | xhv = (XPVHV*)SvANY(hv); |
| 1919 | |
| 1920 | /* The name must be deleted before the call to hfreeeeentries so that |
| 1921 | CVs are anonymised properly. But the effective name must be pre- |
| 1922 | served until after that call (and only deleted afterwards if the |
| 1923 | call originated from sv_clear). For stashes with one name that is |
| 1924 | both the canonical name and the effective name, hv_name_set has to |
| 1925 | allocate an array for storing the effective name. We can skip that |
| 1926 | during global destruction, as it does not matter where the CVs point |
| 1927 | if they will be freed anyway. */ |
| 1928 | /* note that the code following prior to hfreeentries is duplicated |
| 1929 | * in sv_clear(), and changes here should be done there too */ |
| 1930 | if (PL_phase != PERL_PHASE_DESTRUCT && HvNAME(hv)) { |
| 1931 | if (PL_stashcache) { |
| 1932 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for '%" |
| 1933 | HEKf"'\n", HEKfARG(HvNAME_HEK(hv)))); |
| 1934 | (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD); |
| 1935 | } |
| 1936 | hv_name_set(hv, NULL, 0, 0); |
| 1937 | } |
| 1938 | if (save) { |
| 1939 | ENTER; |
| 1940 | SAVEFREESV(SvREFCNT_inc_simple_NN(hv)); |
| 1941 | } |
| 1942 | hfreeentries(hv); |
| 1943 | if (SvOOK(hv)) { |
| 1944 | struct mro_meta *meta; |
| 1945 | const char *name; |
| 1946 | |
| 1947 | if (HvENAME_get(hv)) { |
| 1948 | if (PL_phase != PERL_PHASE_DESTRUCT) |
| 1949 | mro_isa_changed_in(hv); |
| 1950 | if (PL_stashcache) { |
| 1951 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for effective name '%" |
| 1952 | HEKf"'\n", HEKfARG(HvENAME_HEK(hv)))); |
| 1953 | (void)hv_deletehek(PL_stashcache, HvENAME_HEK(hv), G_DISCARD); |
| 1954 | } |
| 1955 | } |
| 1956 | |
| 1957 | /* If this call originated from sv_clear, then we must check for |
| 1958 | * effective names that need freeing, as well as the usual name. */ |
| 1959 | name = HvNAME(hv); |
| 1960 | if (flags & HV_NAME_SETALL ? !!HvAUX(hv)->xhv_name_u.xhvnameu_name : !!name) { |
| 1961 | if (name && PL_stashcache) { |
| 1962 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for name '%" |
| 1963 | HEKf"'\n", HEKfARG(HvNAME_HEK(hv)))); |
| 1964 | (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD); |
| 1965 | } |
| 1966 | hv_name_set(hv, NULL, 0, flags); |
| 1967 | } |
| 1968 | if((meta = HvAUX(hv)->xhv_mro_meta)) { |
| 1969 | if (meta->mro_linear_all) { |
| 1970 | SvREFCNT_dec_NN(meta->mro_linear_all); |
| 1971 | /* mro_linear_current is just acting as a shortcut pointer, |
| 1972 | hence the else. */ |
| 1973 | } |
| 1974 | else |
| 1975 | /* Only the current MRO is stored, so this owns the data. |
| 1976 | */ |
| 1977 | SvREFCNT_dec(meta->mro_linear_current); |
| 1978 | SvREFCNT_dec(meta->mro_nextmethod); |
| 1979 | SvREFCNT_dec(meta->isa); |
| 1980 | SvREFCNT_dec(meta->super); |
| 1981 | Safefree(meta); |
| 1982 | HvAUX(hv)->xhv_mro_meta = NULL; |
| 1983 | } |
| 1984 | if (!HvAUX(hv)->xhv_name_u.xhvnameu_name && ! HvAUX(hv)->xhv_backreferences) |
| 1985 | SvFLAGS(hv) &= ~SVf_OOK; |
| 1986 | } |
| 1987 | if (!SvOOK(hv)) { |
| 1988 | Safefree(HvARRAY(hv)); |
| 1989 | xhv->xhv_max = PERL_HASH_DEFAULT_HvMAX; /* HvMAX(hv) = 7 (it's a normal hash) */ |
| 1990 | HvARRAY(hv) = 0; |
| 1991 | } |
| 1992 | /* if we're freeing the HV, the SvMAGIC field has been reused for |
| 1993 | * other purposes, and so there can't be any placeholder magic */ |
| 1994 | if (SvREFCNT(hv)) |
| 1995 | HvPLACEHOLDERS_set(hv, 0); |
| 1996 | |
| 1997 | if (SvRMAGICAL(hv)) |
| 1998 | mg_clear(MUTABLE_SV(hv)); |
| 1999 | if (save) LEAVE; |
| 2000 | } |
| 2001 | |
| 2002 | /* |
| 2003 | =for apidoc hv_fill |
| 2004 | |
| 2005 | Returns the number of hash buckets that |
| 2006 | happen to be in use. This function is |
| 2007 | wrapped by the macro C<HvFILL>. |
| 2008 | |
| 2009 | Previously this value was always stored in the HV structure, which created an |
| 2010 | overhead on every hash (and pretty much every object) for something that was |
| 2011 | rarely used. Now we calculate it on demand the first |
| 2012 | time that it is needed, and cache it if that calculation |
| 2013 | is going to be costly to repeat. The cached |
| 2014 | value is updated by insertions and deletions, but (currently) discarded if |
| 2015 | the hash is split. |
| 2016 | |
| 2017 | =cut |
| 2018 | */ |
| 2019 | |
| 2020 | STRLEN |
| 2021 | Perl_hv_fill(pTHX_ HV *const hv) |
| 2022 | { |
| 2023 | STRLEN count = 0; |
| 2024 | HE **ents = HvARRAY(hv); |
| 2025 | struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : NULL; |
| 2026 | |
| 2027 | PERL_ARGS_ASSERT_HV_FILL; |
| 2028 | |
| 2029 | /* No keys implies no buckets used. |
| 2030 | One key can only possibly mean one bucket used. */ |
| 2031 | if (HvTOTALKEYS(hv) < 2) |
| 2032 | return HvTOTALKEYS(hv); |
| 2033 | |
| 2034 | #ifndef DEBUGGING |
| 2035 | if (aux && aux->xhv_fill_lazy) |
| 2036 | return aux->xhv_fill_lazy; |
| 2037 | #endif |
| 2038 | |
| 2039 | if (ents) { |
| 2040 | HE *const *const last = ents + HvMAX(hv); |
| 2041 | count = last + 1 - ents; |
| 2042 | |
| 2043 | do { |
| 2044 | if (!*ents) |
| 2045 | --count; |
| 2046 | } while (++ents <= last); |
| 2047 | } |
| 2048 | if (aux) { |
| 2049 | #ifdef DEBUGGING |
| 2050 | if (aux->xhv_fill_lazy) |
| 2051 | assert(aux->xhv_fill_lazy == count); |
| 2052 | #endif |
| 2053 | aux->xhv_fill_lazy = count; |
| 2054 | } else if (HvMAX(hv) >= HV_FILL_THRESHOLD) { |
| 2055 | aux = hv_auxinit(hv); |
| 2056 | aux->xhv_fill_lazy = count; |
| 2057 | } |
| 2058 | return count; |
| 2059 | } |
| 2060 | |
| 2061 | /* hash a pointer to a U32 - Used in the hash traversal randomization |
| 2062 | * and bucket order randomization code |
| 2063 | * |
| 2064 | * this code was derived from Sereal, which was derived from autobox. |
| 2065 | */ |
| 2066 | |
| 2067 | PERL_STATIC_INLINE U32 S_ptr_hash(PTRV u) { |
| 2068 | #if PTRSIZE == 8 |
| 2069 | /* |
| 2070 | * This is one of Thomas Wang's hash functions for 64-bit integers from: |
| 2071 | * http://www.concentric.net/~Ttwang/tech/inthash.htm |
| 2072 | */ |
| 2073 | u = (~u) + (u << 18); |
| 2074 | u = u ^ (u >> 31); |
| 2075 | u = u * 21; |
| 2076 | u = u ^ (u >> 11); |
| 2077 | u = u + (u << 6); |
| 2078 | u = u ^ (u >> 22); |
| 2079 | #else |
| 2080 | /* |
| 2081 | * This is one of Bob Jenkins' hash functions for 32-bit integers |
| 2082 | * from: http://burtleburtle.net/bob/hash/integer.html |
| 2083 | */ |
| 2084 | u = (u + 0x7ed55d16) + (u << 12); |
| 2085 | u = (u ^ 0xc761c23c) ^ (u >> 19); |
| 2086 | u = (u + 0x165667b1) + (u << 5); |
| 2087 | u = (u + 0xd3a2646c) ^ (u << 9); |
| 2088 | u = (u + 0xfd7046c5) + (u << 3); |
| 2089 | u = (u ^ 0xb55a4f09) ^ (u >> 16); |
| 2090 | #endif |
| 2091 | return (U32)u; |
| 2092 | } |
| 2093 | |
| 2094 | static struct xpvhv_aux* |
| 2095 | S_hv_auxinit_internal(struct xpvhv_aux *iter) { |
| 2096 | PERL_ARGS_ASSERT_HV_AUXINIT_INTERNAL; |
| 2097 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
| 2098 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
| 2099 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2100 | iter->xhv_last_rand = iter->xhv_rand; |
| 2101 | #endif |
| 2102 | iter->xhv_fill_lazy = 0; |
| 2103 | iter->xhv_name_u.xhvnameu_name = 0; |
| 2104 | iter->xhv_name_count = 0; |
| 2105 | iter->xhv_backreferences = 0; |
| 2106 | iter->xhv_mro_meta = NULL; |
| 2107 | iter->xhv_aux_flags = 0; |
| 2108 | return iter; |
| 2109 | } |
| 2110 | |
| 2111 | |
| 2112 | static struct xpvhv_aux* |
| 2113 | S_hv_auxinit(pTHX_ HV *hv) { |
| 2114 | struct xpvhv_aux *iter; |
| 2115 | char *array; |
| 2116 | |
| 2117 | PERL_ARGS_ASSERT_HV_AUXINIT; |
| 2118 | |
| 2119 | if (!SvOOK(hv)) { |
| 2120 | if (!HvARRAY(hv)) { |
| 2121 | Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) |
| 2122 | + sizeof(struct xpvhv_aux), char); |
| 2123 | } else { |
| 2124 | array = (char *) HvARRAY(hv); |
| 2125 | Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) |
| 2126 | + sizeof(struct xpvhv_aux), char); |
| 2127 | } |
| 2128 | HvARRAY(hv) = (HE**)array; |
| 2129 | SvOOK_on(hv); |
| 2130 | iter = HvAUX(hv); |
| 2131 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2132 | if (PL_HASH_RAND_BITS_ENABLED) { |
| 2133 | /* mix in some new state to PL_hash_rand_bits to "randomize" the traversal order*/ |
| 2134 | if (PL_HASH_RAND_BITS_ENABLED == 1) |
| 2135 | PL_hash_rand_bits += ptr_hash((PTRV)array); |
| 2136 | PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1); |
| 2137 | } |
| 2138 | iter->xhv_rand = (U32)PL_hash_rand_bits; |
| 2139 | #endif |
| 2140 | } else { |
| 2141 | iter = HvAUX(hv); |
| 2142 | } |
| 2143 | |
| 2144 | return hv_auxinit_internal(iter); |
| 2145 | } |
| 2146 | |
| 2147 | /* |
| 2148 | =for apidoc hv_iterinit |
| 2149 | |
| 2150 | Prepares a starting point to traverse a hash table. Returns the number of |
| 2151 | keys in the hash (i.e. the same as C<HvUSEDKEYS(hv)>). The return value is |
| 2152 | currently only meaningful for hashes without tie magic. |
| 2153 | |
| 2154 | NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of |
| 2155 | hash buckets that happen to be in use. If you still need that esoteric |
| 2156 | value, you can get it through the macro C<HvFILL(hv)>. |
| 2157 | |
| 2158 | |
| 2159 | =cut |
| 2160 | */ |
| 2161 | |
| 2162 | I32 |
| 2163 | Perl_hv_iterinit(pTHX_ HV *hv) |
| 2164 | { |
| 2165 | PERL_ARGS_ASSERT_HV_ITERINIT; |
| 2166 | |
| 2167 | if (SvOOK(hv)) { |
| 2168 | struct xpvhv_aux * iter = HvAUX(hv); |
| 2169 | HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */ |
| 2170 | if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ |
| 2171 | HvLAZYDEL_off(hv); |
| 2172 | hv_free_ent(hv, entry); |
| 2173 | } |
| 2174 | iter = HvAUX(hv); /* may have been reallocated */ |
| 2175 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
| 2176 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
| 2177 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2178 | iter->xhv_last_rand = iter->xhv_rand; |
| 2179 | #endif |
| 2180 | } else { |
| 2181 | hv_auxinit(hv); |
| 2182 | } |
| 2183 | |
| 2184 | /* used to be xhv->xhv_fill before 5.004_65 */ |
| 2185 | return HvTOTALKEYS(hv); |
| 2186 | } |
| 2187 | |
| 2188 | I32 * |
| 2189 | Perl_hv_riter_p(pTHX_ HV *hv) { |
| 2190 | struct xpvhv_aux *iter; |
| 2191 | |
| 2192 | PERL_ARGS_ASSERT_HV_RITER_P; |
| 2193 | |
| 2194 | iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
| 2195 | return &(iter->xhv_riter); |
| 2196 | } |
| 2197 | |
| 2198 | HE ** |
| 2199 | Perl_hv_eiter_p(pTHX_ HV *hv) { |
| 2200 | struct xpvhv_aux *iter; |
| 2201 | |
| 2202 | PERL_ARGS_ASSERT_HV_EITER_P; |
| 2203 | |
| 2204 | iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
| 2205 | return &(iter->xhv_eiter); |
| 2206 | } |
| 2207 | |
| 2208 | void |
| 2209 | Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) { |
| 2210 | struct xpvhv_aux *iter; |
| 2211 | |
| 2212 | PERL_ARGS_ASSERT_HV_RITER_SET; |
| 2213 | |
| 2214 | if (SvOOK(hv)) { |
| 2215 | iter = HvAUX(hv); |
| 2216 | } else { |
| 2217 | if (riter == -1) |
| 2218 | return; |
| 2219 | |
| 2220 | iter = hv_auxinit(hv); |
| 2221 | } |
| 2222 | iter->xhv_riter = riter; |
| 2223 | } |
| 2224 | |
| 2225 | void |
| 2226 | Perl_hv_rand_set(pTHX_ HV *hv, U32 new_xhv_rand) { |
| 2227 | struct xpvhv_aux *iter; |
| 2228 | |
| 2229 | PERL_ARGS_ASSERT_HV_RAND_SET; |
| 2230 | |
| 2231 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2232 | if (SvOOK(hv)) { |
| 2233 | iter = HvAUX(hv); |
| 2234 | } else { |
| 2235 | iter = hv_auxinit(hv); |
| 2236 | } |
| 2237 | iter->xhv_rand = new_xhv_rand; |
| 2238 | #else |
| 2239 | Perl_croak(aTHX_ "This Perl has not been built with support for randomized hash key traversal but something called Perl_hv_rand_set()."); |
| 2240 | #endif |
| 2241 | } |
| 2242 | |
| 2243 | void |
| 2244 | Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) { |
| 2245 | struct xpvhv_aux *iter; |
| 2246 | |
| 2247 | PERL_ARGS_ASSERT_HV_EITER_SET; |
| 2248 | |
| 2249 | if (SvOOK(hv)) { |
| 2250 | iter = HvAUX(hv); |
| 2251 | } else { |
| 2252 | /* 0 is the default so don't go malloc()ing a new structure just to |
| 2253 | hold 0. */ |
| 2254 | if (!eiter) |
| 2255 | return; |
| 2256 | |
| 2257 | iter = hv_auxinit(hv); |
| 2258 | } |
| 2259 | iter->xhv_eiter = eiter; |
| 2260 | } |
| 2261 | |
| 2262 | void |
| 2263 | Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
| 2264 | { |
| 2265 | dVAR; |
| 2266 | struct xpvhv_aux *iter; |
| 2267 | U32 hash; |
| 2268 | HEK **spot; |
| 2269 | |
| 2270 | PERL_ARGS_ASSERT_HV_NAME_SET; |
| 2271 | |
| 2272 | if (len > I32_MAX) |
| 2273 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); |
| 2274 | |
| 2275 | if (SvOOK(hv)) { |
| 2276 | iter = HvAUX(hv); |
| 2277 | if (iter->xhv_name_u.xhvnameu_name) { |
| 2278 | if(iter->xhv_name_count) { |
| 2279 | if(flags & HV_NAME_SETALL) { |
| 2280 | HEK ** const name = HvAUX(hv)->xhv_name_u.xhvnameu_names; |
| 2281 | HEK **hekp = name + ( |
| 2282 | iter->xhv_name_count < 0 |
| 2283 | ? -iter->xhv_name_count |
| 2284 | : iter->xhv_name_count |
| 2285 | ); |
| 2286 | while(hekp-- > name+1) |
| 2287 | unshare_hek_or_pvn(*hekp, 0, 0, 0); |
| 2288 | /* The first elem may be null. */ |
| 2289 | if(*name) unshare_hek_or_pvn(*name, 0, 0, 0); |
| 2290 | Safefree(name); |
| 2291 | iter = HvAUX(hv); /* may been realloced */ |
| 2292 | spot = &iter->xhv_name_u.xhvnameu_name; |
| 2293 | iter->xhv_name_count = 0; |
| 2294 | } |
| 2295 | else { |
| 2296 | if(iter->xhv_name_count > 0) { |
| 2297 | /* shift some things over */ |
| 2298 | Renew( |
| 2299 | iter->xhv_name_u.xhvnameu_names, iter->xhv_name_count + 1, HEK * |
| 2300 | ); |
| 2301 | spot = iter->xhv_name_u.xhvnameu_names; |
| 2302 | spot[iter->xhv_name_count] = spot[1]; |
| 2303 | spot[1] = spot[0]; |
| 2304 | iter->xhv_name_count = -(iter->xhv_name_count + 1); |
| 2305 | } |
| 2306 | else if(*(spot = iter->xhv_name_u.xhvnameu_names)) { |
| 2307 | unshare_hek_or_pvn(*spot, 0, 0, 0); |
| 2308 | } |
| 2309 | } |
| 2310 | } |
| 2311 | else if (flags & HV_NAME_SETALL) { |
| 2312 | unshare_hek_or_pvn(iter->xhv_name_u.xhvnameu_name, 0, 0, 0); |
| 2313 | iter = HvAUX(hv); /* may been realloced */ |
| 2314 | spot = &iter->xhv_name_u.xhvnameu_name; |
| 2315 | } |
| 2316 | else { |
| 2317 | HEK * const existing_name = iter->xhv_name_u.xhvnameu_name; |
| 2318 | Newx(iter->xhv_name_u.xhvnameu_names, 2, HEK *); |
| 2319 | iter->xhv_name_count = -2; |
| 2320 | spot = iter->xhv_name_u.xhvnameu_names; |
| 2321 | spot[1] = existing_name; |
| 2322 | } |
| 2323 | } |
| 2324 | else { spot = &iter->xhv_name_u.xhvnameu_name; iter->xhv_name_count = 0; } |
| 2325 | } else { |
| 2326 | if (name == 0) |
| 2327 | return; |
| 2328 | |
| 2329 | iter = hv_auxinit(hv); |
| 2330 | spot = &iter->xhv_name_u.xhvnameu_name; |
| 2331 | } |
| 2332 | PERL_HASH(hash, name, len); |
| 2333 | *spot = name ? share_hek(name, flags & SVf_UTF8 ? -(I32)len : (I32)len, hash) : NULL; |
| 2334 | } |
| 2335 | |
| 2336 | /* |
| 2337 | This is basically sv_eq_flags() in sv.c, but we avoid the magic |
| 2338 | and bytes checking. |
| 2339 | */ |
| 2340 | |
| 2341 | STATIC I32 |
| 2342 | hek_eq_pvn_flags(pTHX_ const HEK *hek, const char* pv, const I32 pvlen, const U32 flags) { |
| 2343 | if ( (HEK_UTF8(hek) ? 1 : 0) != (flags & SVf_UTF8 ? 1 : 0) ) { |
| 2344 | if (flags & SVf_UTF8) |
| 2345 | return (bytes_cmp_utf8( |
| 2346 | (const U8*)HEK_KEY(hek), HEK_LEN(hek), |
| 2347 | (const U8*)pv, pvlen) == 0); |
| 2348 | else |
| 2349 | return (bytes_cmp_utf8( |
| 2350 | (const U8*)pv, pvlen, |
| 2351 | (const U8*)HEK_KEY(hek), HEK_LEN(hek)) == 0); |
| 2352 | } |
| 2353 | else |
| 2354 | return HEK_LEN(hek) == pvlen && ((HEK_KEY(hek) == pv) |
| 2355 | || memEQ(HEK_KEY(hek), pv, pvlen)); |
| 2356 | } |
| 2357 | |
| 2358 | /* |
| 2359 | =for apidoc hv_ename_add |
| 2360 | |
| 2361 | Adds a name to a stash's internal list of effective names. See |
| 2362 | C<L</hv_ename_delete>>. |
| 2363 | |
| 2364 | This is called when a stash is assigned to a new location in the symbol |
| 2365 | table. |
| 2366 | |
| 2367 | =cut |
| 2368 | */ |
| 2369 | |
| 2370 | void |
| 2371 | Perl_hv_ename_add(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
| 2372 | { |
| 2373 | dVAR; |
| 2374 | struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
| 2375 | U32 hash; |
| 2376 | |
| 2377 | PERL_ARGS_ASSERT_HV_ENAME_ADD; |
| 2378 | |
| 2379 | if (len > I32_MAX) |
| 2380 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); |
| 2381 | |
| 2382 | PERL_HASH(hash, name, len); |
| 2383 | |
| 2384 | if (aux->xhv_name_count) { |
| 2385 | I32 count = aux->xhv_name_count; |
| 2386 | HEK ** const xhv_name = aux->xhv_name_u.xhvnameu_names + (count<0); |
| 2387 | HEK **hekp = xhv_name + (count < 0 ? -count - 1 : count); |
| 2388 | while (hekp-- > xhv_name) |
| 2389 | { |
| 2390 | assert(*hekp); |
| 2391 | if ( |
| 2392 | (HEK_UTF8(*hekp) || (flags & SVf_UTF8)) |
| 2393 | ? hek_eq_pvn_flags(aTHX_ *hekp, name, (I32)len, flags) |
| 2394 | : (HEK_LEN(*hekp) == (I32)len && memEQ(HEK_KEY(*hekp), name, len)) |
| 2395 | ) { |
| 2396 | if (hekp == xhv_name && count < 0) |
| 2397 | aux->xhv_name_count = -count; |
| 2398 | return; |
| 2399 | } |
| 2400 | } |
| 2401 | if (count < 0) aux->xhv_name_count--, count = -count; |
| 2402 | else aux->xhv_name_count++; |
| 2403 | Renew(aux->xhv_name_u.xhvnameu_names, count + 1, HEK *); |
| 2404 | (aux->xhv_name_u.xhvnameu_names)[count] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash); |
| 2405 | } |
| 2406 | else { |
| 2407 | HEK *existing_name = aux->xhv_name_u.xhvnameu_name; |
| 2408 | if ( |
| 2409 | existing_name && ( |
| 2410 | (HEK_UTF8(existing_name) || (flags & SVf_UTF8)) |
| 2411 | ? hek_eq_pvn_flags(aTHX_ existing_name, name, (I32)len, flags) |
| 2412 | : (HEK_LEN(existing_name) == (I32)len && memEQ(HEK_KEY(existing_name), name, len)) |
| 2413 | ) |
| 2414 | ) return; |
| 2415 | Newx(aux->xhv_name_u.xhvnameu_names, 2, HEK *); |
| 2416 | aux->xhv_name_count = existing_name ? 2 : -2; |
| 2417 | *aux->xhv_name_u.xhvnameu_names = existing_name; |
| 2418 | (aux->xhv_name_u.xhvnameu_names)[1] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash); |
| 2419 | } |
| 2420 | } |
| 2421 | |
| 2422 | /* |
| 2423 | =for apidoc hv_ename_delete |
| 2424 | |
| 2425 | Removes a name from a stash's internal list of effective names. If this is |
| 2426 | the name returned by C<HvENAME>, then another name in the list will take |
| 2427 | its place (C<HvENAME> will use it). |
| 2428 | |
| 2429 | This is called when a stash is deleted from the symbol table. |
| 2430 | |
| 2431 | =cut |
| 2432 | */ |
| 2433 | |
| 2434 | void |
| 2435 | Perl_hv_ename_delete(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
| 2436 | { |
| 2437 | struct xpvhv_aux *aux; |
| 2438 | |
| 2439 | PERL_ARGS_ASSERT_HV_ENAME_DELETE; |
| 2440 | |
| 2441 | if (len > I32_MAX) |
| 2442 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); |
| 2443 | |
| 2444 | if (!SvOOK(hv)) return; |
| 2445 | |
| 2446 | aux = HvAUX(hv); |
| 2447 | if (!aux->xhv_name_u.xhvnameu_name) return; |
| 2448 | |
| 2449 | if (aux->xhv_name_count) { |
| 2450 | HEK ** const namep = aux->xhv_name_u.xhvnameu_names; |
| 2451 | I32 const count = aux->xhv_name_count; |
| 2452 | HEK **victim = namep + (count < 0 ? -count : count); |
| 2453 | while (victim-- > namep + 1) |
| 2454 | if ( |
| 2455 | (HEK_UTF8(*victim) || (flags & SVf_UTF8)) |
| 2456 | ? hek_eq_pvn_flags(aTHX_ *victim, name, (I32)len, flags) |
| 2457 | : (HEK_LEN(*victim) == (I32)len && memEQ(HEK_KEY(*victim), name, len)) |
| 2458 | ) { |
| 2459 | unshare_hek_or_pvn(*victim, 0, 0, 0); |
| 2460 | aux = HvAUX(hv); /* may been realloced */ |
| 2461 | if (count < 0) ++aux->xhv_name_count; |
| 2462 | else --aux->xhv_name_count; |
| 2463 | if ( |
| 2464 | (aux->xhv_name_count == 1 || aux->xhv_name_count == -1) |
| 2465 | && !*namep |
| 2466 | ) { /* if there are none left */ |
| 2467 | Safefree(namep); |
| 2468 | aux->xhv_name_u.xhvnameu_names = NULL; |
| 2469 | aux->xhv_name_count = 0; |
| 2470 | } |
| 2471 | else { |
| 2472 | /* Move the last one back to fill the empty slot. It |
| 2473 | does not matter what order they are in. */ |
| 2474 | *victim = *(namep + (count < 0 ? -count : count) - 1); |
| 2475 | } |
| 2476 | return; |
| 2477 | } |
| 2478 | if ( |
| 2479 | count > 0 && (HEK_UTF8(*namep) || (flags & SVf_UTF8)) |
| 2480 | ? hek_eq_pvn_flags(aTHX_ *namep, name, (I32)len, flags) |
| 2481 | : (HEK_LEN(*namep) == (I32)len && memEQ(HEK_KEY(*namep), name, len)) |
| 2482 | ) { |
| 2483 | aux->xhv_name_count = -count; |
| 2484 | } |
| 2485 | } |
| 2486 | else if( |
| 2487 | (HEK_UTF8(aux->xhv_name_u.xhvnameu_name) || (flags & SVf_UTF8)) |
| 2488 | ? hek_eq_pvn_flags(aTHX_ aux->xhv_name_u.xhvnameu_name, name, (I32)len, flags) |
| 2489 | : (HEK_LEN(aux->xhv_name_u.xhvnameu_name) == (I32)len && |
| 2490 | memEQ(HEK_KEY(aux->xhv_name_u.xhvnameu_name), name, len)) |
| 2491 | ) { |
| 2492 | HEK * const namehek = aux->xhv_name_u.xhvnameu_name; |
| 2493 | Newx(aux->xhv_name_u.xhvnameu_names, 1, HEK *); |
| 2494 | *aux->xhv_name_u.xhvnameu_names = namehek; |
| 2495 | aux->xhv_name_count = -1; |
| 2496 | } |
| 2497 | } |
| 2498 | |
| 2499 | AV ** |
| 2500 | Perl_hv_backreferences_p(pTHX_ HV *hv) { |
| 2501 | PERL_ARGS_ASSERT_HV_BACKREFERENCES_P; |
| 2502 | /* See also Perl_sv_get_backrefs in sv.c where this logic is unrolled */ |
| 2503 | { |
| 2504 | struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
| 2505 | return &(iter->xhv_backreferences); |
| 2506 | } |
| 2507 | } |
| 2508 | |
| 2509 | void |
| 2510 | Perl_hv_kill_backrefs(pTHX_ HV *hv) { |
| 2511 | AV *av; |
| 2512 | |
| 2513 | PERL_ARGS_ASSERT_HV_KILL_BACKREFS; |
| 2514 | |
| 2515 | if (!SvOOK(hv)) |
| 2516 | return; |
| 2517 | |
| 2518 | av = HvAUX(hv)->xhv_backreferences; |
| 2519 | |
| 2520 | if (av) { |
| 2521 | HvAUX(hv)->xhv_backreferences = 0; |
| 2522 | Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av); |
| 2523 | if (SvTYPE(av) == SVt_PVAV) |
| 2524 | SvREFCNT_dec_NN(av); |
| 2525 | } |
| 2526 | } |
| 2527 | |
| 2528 | /* |
| 2529 | hv_iternext is implemented as a macro in hv.h |
| 2530 | |
| 2531 | =for apidoc hv_iternext |
| 2532 | |
| 2533 | Returns entries from a hash iterator. See C<L</hv_iterinit>>. |
| 2534 | |
| 2535 | You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the |
| 2536 | iterator currently points to, without losing your place or invalidating your |
| 2537 | iterator. Note that in this case the current entry is deleted from the hash |
| 2538 | with your iterator holding the last reference to it. Your iterator is flagged |
| 2539 | to free the entry on the next call to C<hv_iternext>, so you must not discard |
| 2540 | your iterator immediately else the entry will leak - call C<hv_iternext> to |
| 2541 | trigger the resource deallocation. |
| 2542 | |
| 2543 | =for apidoc hv_iternext_flags |
| 2544 | |
| 2545 | Returns entries from a hash iterator. See C<L</hv_iterinit>> and |
| 2546 | C<L</hv_iternext>>. |
| 2547 | The C<flags> value will normally be zero; if C<HV_ITERNEXT_WANTPLACEHOLDERS> is |
| 2548 | set the placeholders keys (for restricted hashes) will be returned in addition |
| 2549 | to normal keys. By default placeholders are automatically skipped over. |
| 2550 | Currently a placeholder is implemented with a value that is |
| 2551 | C<&PL_sv_placeholder>. Note that the implementation of placeholders and |
| 2552 | restricted hashes may change, and the implementation currently is |
| 2553 | insufficiently abstracted for any change to be tidy. |
| 2554 | |
| 2555 | =cut |
| 2556 | */ |
| 2557 | |
| 2558 | HE * |
| 2559 | Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags) |
| 2560 | { |
| 2561 | dVAR; |
| 2562 | XPVHV* xhv; |
| 2563 | HE *entry; |
| 2564 | HE *oldentry; |
| 2565 | MAGIC* mg; |
| 2566 | struct xpvhv_aux *iter; |
| 2567 | |
| 2568 | PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS; |
| 2569 | |
| 2570 | xhv = (XPVHV*)SvANY(hv); |
| 2571 | |
| 2572 | if (!SvOOK(hv)) { |
| 2573 | /* Too many things (well, pp_each at least) merrily assume that you can |
| 2574 | call hv_iternext without calling hv_iterinit, so we'll have to deal |
| 2575 | with it. */ |
| 2576 | hv_iterinit(hv); |
| 2577 | } |
| 2578 | iter = HvAUX(hv); |
| 2579 | |
| 2580 | oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ |
| 2581 | if (SvMAGICAL(hv) && SvRMAGICAL(hv)) { |
| 2582 | if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) { |
| 2583 | SV * const key = sv_newmortal(); |
| 2584 | if (entry) { |
| 2585 | sv_setsv(key, HeSVKEY_force(entry)); |
| 2586 | SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */ |
| 2587 | HeSVKEY_set(entry, NULL); |
| 2588 | } |
| 2589 | else { |
| 2590 | char *k; |
| 2591 | HEK *hek; |
| 2592 | |
| 2593 | /* one HE per MAGICAL hash */ |
| 2594 | iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */ |
| 2595 | HvLAZYDEL_on(hv); /* make sure entry gets freed */ |
| 2596 | Zero(entry, 1, HE); |
| 2597 | Newxz(k, HEK_BASESIZE + sizeof(const SV *), char); |
| 2598 | hek = (HEK*)k; |
| 2599 | HeKEY_hek(entry) = hek; |
| 2600 | HeKLEN(entry) = HEf_SVKEY; |
| 2601 | } |
| 2602 | magic_nextpack(MUTABLE_SV(hv),mg,key); |
| 2603 | if (SvOK(key)) { |
| 2604 | /* force key to stay around until next time */ |
| 2605 | HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key)); |
| 2606 | return entry; /* beware, hent_val is not set */ |
| 2607 | } |
| 2608 | SvREFCNT_dec(HeVAL(entry)); |
| 2609 | Safefree(HeKEY_hek(entry)); |
| 2610 | del_HE(entry); |
| 2611 | iter = HvAUX(hv); /* may been realloced */ |
| 2612 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
| 2613 | HvLAZYDEL_off(hv); |
| 2614 | return NULL; |
| 2615 | } |
| 2616 | } |
| 2617 | #if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */ |
| 2618 | if (!entry && SvRMAGICAL((const SV *)hv) |
| 2619 | && mg_find((const SV *)hv, PERL_MAGIC_env)) { |
| 2620 | prime_env_iter(); |
| 2621 | #ifdef VMS |
| 2622 | /* The prime_env_iter() on VMS just loaded up new hash values |
| 2623 | * so the iteration count needs to be reset back to the beginning |
| 2624 | */ |
| 2625 | hv_iterinit(hv); |
| 2626 | iter = HvAUX(hv); |
| 2627 | oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ |
| 2628 | #endif |
| 2629 | } |
| 2630 | #endif |
| 2631 | |
| 2632 | /* hv_iterinit now ensures this. */ |
| 2633 | assert (HvARRAY(hv)); |
| 2634 | |
| 2635 | /* At start of hash, entry is NULL. */ |
| 2636 | if (entry) |
| 2637 | { |
| 2638 | entry = HeNEXT(entry); |
| 2639 | if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) { |
| 2640 | /* |
| 2641 | * Skip past any placeholders -- don't want to include them in |
| 2642 | * any iteration. |
| 2643 | */ |
| 2644 | while (entry && HeVAL(entry) == &PL_sv_placeholder) { |
| 2645 | entry = HeNEXT(entry); |
| 2646 | } |
| 2647 | } |
| 2648 | } |
| 2649 | |
| 2650 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2651 | if (iter->xhv_last_rand != iter->xhv_rand) { |
| 2652 | if (iter->xhv_riter != -1) { |
| 2653 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), |
| 2654 | "Use of each() on hash after insertion without resetting hash iterator results in undefined behavior" |
| 2655 | pTHX__FORMAT |
| 2656 | pTHX__VALUE); |
| 2657 | } |
| 2658 | iter = HvAUX(hv); /* may been realloced */ |
| 2659 | iter->xhv_last_rand = iter->xhv_rand; |
| 2660 | } |
| 2661 | #endif |
| 2662 | |
| 2663 | /* Skip the entire loop if the hash is empty. */ |
| 2664 | if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS) |
| 2665 | ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) { |
| 2666 | while (!entry) { |
| 2667 | /* OK. Come to the end of the current list. Grab the next one. */ |
| 2668 | |
| 2669 | iter->xhv_riter++; /* HvRITER(hv)++ */ |
| 2670 | if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) { |
| 2671 | /* There is no next one. End of the hash. */ |
| 2672 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
| 2673 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2674 | iter->xhv_last_rand = iter->xhv_rand; /* reset xhv_last_rand so we can detect inserts during traversal */ |
| 2675 | #endif |
| 2676 | break; |
| 2677 | } |
| 2678 | entry = (HvARRAY(hv))[ PERL_HASH_ITER_BUCKET(iter) & xhv->xhv_max ]; |
| 2679 | |
| 2680 | if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) { |
| 2681 | /* If we have an entry, but it's a placeholder, don't count it. |
| 2682 | Try the next. */ |
| 2683 | while (entry && HeVAL(entry) == &PL_sv_placeholder) |
| 2684 | entry = HeNEXT(entry); |
| 2685 | } |
| 2686 | /* Will loop again if this linked list starts NULL |
| 2687 | (for HV_ITERNEXT_WANTPLACEHOLDERS) |
| 2688 | or if we run through it and find only placeholders. */ |
| 2689 | } |
| 2690 | } |
| 2691 | else { |
| 2692 | iter->xhv_riter = -1; |
| 2693 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
| 2694 | iter->xhv_last_rand = iter->xhv_rand; |
| 2695 | #endif |
| 2696 | } |
| 2697 | |
| 2698 | if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */ |
| 2699 | HvLAZYDEL_off(hv); |
| 2700 | hv_free_ent(hv, oldentry); |
| 2701 | } |
| 2702 | |
| 2703 | iter = HvAUX(hv); /* may been realloced */ |
| 2704 | iter->xhv_eiter = entry; /* HvEITER(hv) = entry */ |
| 2705 | return entry; |
| 2706 | } |
| 2707 | |
| 2708 | /* |
| 2709 | =for apidoc hv_iterkey |
| 2710 | |
| 2711 | Returns the key from the current position of the hash iterator. See |
| 2712 | C<L</hv_iterinit>>. |
| 2713 | |
| 2714 | =cut |
| 2715 | */ |
| 2716 | |
| 2717 | char * |
| 2718 | Perl_hv_iterkey(pTHX_ HE *entry, I32 *retlen) |
| 2719 | { |
| 2720 | PERL_ARGS_ASSERT_HV_ITERKEY; |
| 2721 | |
| 2722 | if (HeKLEN(entry) == HEf_SVKEY) { |
| 2723 | STRLEN len; |
| 2724 | char * const p = SvPV(HeKEY_sv(entry), len); |
| 2725 | *retlen = len; |
| 2726 | return p; |
| 2727 | } |
| 2728 | else { |
| 2729 | *retlen = HeKLEN(entry); |
| 2730 | return HeKEY(entry); |
| 2731 | } |
| 2732 | } |
| 2733 | |
| 2734 | /* unlike hv_iterval(), this always returns a mortal copy of the key */ |
| 2735 | /* |
| 2736 | =for apidoc hv_iterkeysv |
| 2737 | |
| 2738 | Returns the key as an C<SV*> from the current position of the hash |
| 2739 | iterator. The return value will always be a mortal copy of the key. Also |
| 2740 | see C<L</hv_iterinit>>. |
| 2741 | |
| 2742 | =cut |
| 2743 | */ |
| 2744 | |
| 2745 | SV * |
| 2746 | Perl_hv_iterkeysv(pTHX_ HE *entry) |
| 2747 | { |
| 2748 | PERL_ARGS_ASSERT_HV_ITERKEYSV; |
| 2749 | |
| 2750 | return sv_2mortal(newSVhek(HeKEY_hek(entry))); |
| 2751 | } |
| 2752 | |
| 2753 | /* |
| 2754 | =for apidoc hv_iterval |
| 2755 | |
| 2756 | Returns the value from the current position of the hash iterator. See |
| 2757 | C<L</hv_iterkey>>. |
| 2758 | |
| 2759 | =cut |
| 2760 | */ |
| 2761 | |
| 2762 | SV * |
| 2763 | Perl_hv_iterval(pTHX_ HV *hv, HE *entry) |
| 2764 | { |
| 2765 | PERL_ARGS_ASSERT_HV_ITERVAL; |
| 2766 | |
| 2767 | if (SvRMAGICAL(hv)) { |
| 2768 | if (mg_find((const SV *)hv, PERL_MAGIC_tied)) { |
| 2769 | SV* const sv = sv_newmortal(); |
| 2770 | if (HeKLEN(entry) == HEf_SVKEY) |
| 2771 | mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY); |
| 2772 | else |
| 2773 | mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry)); |
| 2774 | return sv; |
| 2775 | } |
| 2776 | } |
| 2777 | return HeVAL(entry); |
| 2778 | } |
| 2779 | |
| 2780 | /* |
| 2781 | =for apidoc hv_iternextsv |
| 2782 | |
| 2783 | Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one |
| 2784 | operation. |
| 2785 | |
| 2786 | =cut |
| 2787 | */ |
| 2788 | |
| 2789 | SV * |
| 2790 | Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen) |
| 2791 | { |
| 2792 | HE * const he = hv_iternext_flags(hv, 0); |
| 2793 | |
| 2794 | PERL_ARGS_ASSERT_HV_ITERNEXTSV; |
| 2795 | |
| 2796 | if (!he) |
| 2797 | return NULL; |
| 2798 | *key = hv_iterkey(he, retlen); |
| 2799 | return hv_iterval(hv, he); |
| 2800 | } |
| 2801 | |
| 2802 | /* |
| 2803 | |
| 2804 | Now a macro in hv.h |
| 2805 | |
| 2806 | =for apidoc hv_magic |
| 2807 | |
| 2808 | Adds magic to a hash. See C<L</sv_magic>>. |
| 2809 | |
| 2810 | =cut |
| 2811 | */ |
| 2812 | |
| 2813 | /* possibly free a shared string if no one has access to it |
| 2814 | * len and hash must both be valid for str. |
| 2815 | */ |
| 2816 | void |
| 2817 | Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash) |
| 2818 | { |
| 2819 | unshare_hek_or_pvn (NULL, str, len, hash); |
| 2820 | } |
| 2821 | |
| 2822 | |
| 2823 | void |
| 2824 | Perl_unshare_hek(pTHX_ HEK *hek) |
| 2825 | { |
| 2826 | assert(hek); |
| 2827 | unshare_hek_or_pvn(hek, NULL, 0, 0); |
| 2828 | } |
| 2829 | |
| 2830 | /* possibly free a shared string if no one has access to it |
| 2831 | hek if non-NULL takes priority over the other 3, else str, len and hash |
| 2832 | are used. If so, len and hash must both be valid for str. |
| 2833 | */ |
| 2834 | STATIC void |
| 2835 | S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash) |
| 2836 | { |
| 2837 | XPVHV* xhv; |
| 2838 | HE *entry; |
| 2839 | HE **oentry; |
| 2840 | bool is_utf8 = FALSE; |
| 2841 | int k_flags = 0; |
| 2842 | const char * const save = str; |
| 2843 | struct shared_he *he = NULL; |
| 2844 | |
| 2845 | if (hek) { |
| 2846 | /* Find the shared he which is just before us in memory. */ |
| 2847 | he = (struct shared_he *)(((char *)hek) |
| 2848 | - STRUCT_OFFSET(struct shared_he, |
| 2849 | shared_he_hek)); |
| 2850 | |
| 2851 | /* Assert that the caller passed us a genuine (or at least consistent) |
| 2852 | shared hek */ |
| 2853 | assert (he->shared_he_he.hent_hek == hek); |
| 2854 | |
| 2855 | if (he->shared_he_he.he_valu.hent_refcount - 1) { |
| 2856 | --he->shared_he_he.he_valu.hent_refcount; |
| 2857 | return; |
| 2858 | } |
| 2859 | |
| 2860 | hash = HEK_HASH(hek); |
| 2861 | } else if (len < 0) { |
| 2862 | STRLEN tmplen = -len; |
| 2863 | is_utf8 = TRUE; |
| 2864 | /* See the note in hv_fetch(). --jhi */ |
| 2865 | str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8); |
| 2866 | len = tmplen; |
| 2867 | if (is_utf8) |
| 2868 | k_flags = HVhek_UTF8; |
| 2869 | if (str != save) |
| 2870 | k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
| 2871 | } |
| 2872 | |
| 2873 | /* what follows was the moral equivalent of: |
| 2874 | if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) { |
| 2875 | if (--*Svp == NULL) |
| 2876 | hv_delete(PL_strtab, str, len, G_DISCARD, hash); |
| 2877 | } */ |
| 2878 | xhv = (XPVHV*)SvANY(PL_strtab); |
| 2879 | /* assert(xhv_array != 0) */ |
| 2880 | oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)]; |
| 2881 | if (he) { |
| 2882 | const HE *const he_he = &(he->shared_he_he); |
| 2883 | for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
| 2884 | if (entry == he_he) |
| 2885 | break; |
| 2886 | } |
| 2887 | } else { |
| 2888 | const int flags_masked = k_flags & HVhek_MASK; |
| 2889 | for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
| 2890 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
| 2891 | continue; |
| 2892 | if (HeKLEN(entry) != len) |
| 2893 | continue; |
| 2894 | if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */ |
| 2895 | continue; |
| 2896 | if (HeKFLAGS(entry) != flags_masked) |
| 2897 | continue; |
| 2898 | break; |
| 2899 | } |
| 2900 | } |
| 2901 | |
| 2902 | if (entry) { |
| 2903 | if (--entry->he_valu.hent_refcount == 0) { |
| 2904 | *oentry = HeNEXT(entry); |
| 2905 | Safefree(entry); |
| 2906 | xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */ |
| 2907 | } |
| 2908 | } |
| 2909 | |
| 2910 | if (!entry) |
| 2911 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), |
| 2912 | "Attempt to free nonexistent shared string '%s'%s" |
| 2913 | pTHX__FORMAT, |
| 2914 | hek ? HEK_KEY(hek) : str, |
| 2915 | ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE); |
| 2916 | if (k_flags & HVhek_FREEKEY) |
| 2917 | Safefree(str); |
| 2918 | } |
| 2919 | |
| 2920 | /* get a (constant) string ptr from the global string table |
| 2921 | * string will get added if it is not already there. |
| 2922 | * len and hash must both be valid for str. |
| 2923 | */ |
| 2924 | HEK * |
| 2925 | Perl_share_hek(pTHX_ const char *str, I32 len, U32 hash) |
| 2926 | { |
| 2927 | bool is_utf8 = FALSE; |
| 2928 | int flags = 0; |
| 2929 | const char * const save = str; |
| 2930 | |
| 2931 | PERL_ARGS_ASSERT_SHARE_HEK; |
| 2932 | |
| 2933 | if (len < 0) { |
| 2934 | STRLEN tmplen = -len; |
| 2935 | is_utf8 = TRUE; |
| 2936 | /* See the note in hv_fetch(). --jhi */ |
| 2937 | str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8); |
| 2938 | len = tmplen; |
| 2939 | /* If we were able to downgrade here, then than means that we were passed |
| 2940 | in a key which only had chars 0-255, but was utf8 encoded. */ |
| 2941 | if (is_utf8) |
| 2942 | flags = HVhek_UTF8; |
| 2943 | /* If we found we were able to downgrade the string to bytes, then |
| 2944 | we should flag that it needs upgrading on keys or each. Also flag |
| 2945 | that we need share_hek_flags to free the string. */ |
| 2946 | if (str != save) { |
| 2947 | dVAR; |
| 2948 | PERL_HASH(hash, str, len); |
| 2949 | flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
| 2950 | } |
| 2951 | } |
| 2952 | |
| 2953 | return share_hek_flags (str, len, hash, flags); |
| 2954 | } |
| 2955 | |
| 2956 | STATIC HEK * |
| 2957 | S_share_hek_flags(pTHX_ const char *str, I32 len, U32 hash, int flags) |
| 2958 | { |
| 2959 | HE *entry; |
| 2960 | const int flags_masked = flags & HVhek_MASK; |
| 2961 | const U32 hindex = hash & (I32) HvMAX(PL_strtab); |
| 2962 | XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab); |
| 2963 | |
| 2964 | PERL_ARGS_ASSERT_SHARE_HEK_FLAGS; |
| 2965 | |
| 2966 | /* what follows is the moral equivalent of: |
| 2967 | |
| 2968 | if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE))) |
| 2969 | hv_store(PL_strtab, str, len, NULL, hash); |
| 2970 | |
| 2971 | Can't rehash the shared string table, so not sure if it's worth |
| 2972 | counting the number of entries in the linked list |
| 2973 | */ |
| 2974 | |
| 2975 | /* assert(xhv_array != 0) */ |
| 2976 | entry = (HvARRAY(PL_strtab))[hindex]; |
| 2977 | for (;entry; entry = HeNEXT(entry)) { |
| 2978 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
| 2979 | continue; |
| 2980 | if (HeKLEN(entry) != len) |
| 2981 | continue; |
| 2982 | if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */ |
| 2983 | continue; |
| 2984 | if (HeKFLAGS(entry) != flags_masked) |
| 2985 | continue; |
| 2986 | break; |
| 2987 | } |
| 2988 | |
| 2989 | if (!entry) { |
| 2990 | /* What used to be head of the list. |
| 2991 | If this is NULL, then we're the first entry for this slot, which |
| 2992 | means we need to increate fill. */ |
| 2993 | struct shared_he *new_entry; |
| 2994 | HEK *hek; |
| 2995 | char *k; |
| 2996 | HE **const head = &HvARRAY(PL_strtab)[hindex]; |
| 2997 | HE *const next = *head; |
| 2998 | |
| 2999 | /* We don't actually store a HE from the arena and a regular HEK. |
| 3000 | Instead we allocate one chunk of memory big enough for both, |
| 3001 | and put the HEK straight after the HE. This way we can find the |
| 3002 | HE directly from the HEK. |
| 3003 | */ |
| 3004 | |
| 3005 | Newx(k, STRUCT_OFFSET(struct shared_he, |
| 3006 | shared_he_hek.hek_key[0]) + len + 2, char); |
| 3007 | new_entry = (struct shared_he *)k; |
| 3008 | entry = &(new_entry->shared_he_he); |
| 3009 | hek = &(new_entry->shared_he_hek); |
| 3010 | |
| 3011 | Copy(str, HEK_KEY(hek), len, char); |
| 3012 | HEK_KEY(hek)[len] = 0; |
| 3013 | HEK_LEN(hek) = len; |
| 3014 | HEK_HASH(hek) = hash; |
| 3015 | HEK_FLAGS(hek) = (unsigned char)flags_masked; |
| 3016 | |
| 3017 | /* Still "point" to the HEK, so that other code need not know what |
| 3018 | we're up to. */ |
| 3019 | HeKEY_hek(entry) = hek; |
| 3020 | entry->he_valu.hent_refcount = 0; |
| 3021 | HeNEXT(entry) = next; |
| 3022 | *head = entry; |
| 3023 | |
| 3024 | xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */ |
| 3025 | if (!next) { /* initial entry? */ |
| 3026 | } else if ( DO_HSPLIT(xhv) ) { |
| 3027 | const STRLEN oldsize = xhv->xhv_max + 1; |
| 3028 | hsplit(PL_strtab, oldsize, oldsize * 2); |
| 3029 | } |
| 3030 | } |
| 3031 | |
| 3032 | ++entry->he_valu.hent_refcount; |
| 3033 | |
| 3034 | if (flags & HVhek_FREEKEY) |
| 3035 | Safefree(str); |
| 3036 | |
| 3037 | return HeKEY_hek(entry); |
| 3038 | } |
| 3039 | |
| 3040 | SSize_t * |
| 3041 | Perl_hv_placeholders_p(pTHX_ HV *hv) |
| 3042 | { |
| 3043 | MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
| 3044 | |
| 3045 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P; |
| 3046 | |
| 3047 | if (!mg) { |
| 3048 | mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0); |
| 3049 | |
| 3050 | if (!mg) { |
| 3051 | Perl_die(aTHX_ "panic: hv_placeholders_p"); |
| 3052 | } |
| 3053 | } |
| 3054 | return &(mg->mg_len); |
| 3055 | } |
| 3056 | |
| 3057 | |
| 3058 | I32 |
| 3059 | Perl_hv_placeholders_get(pTHX_ const HV *hv) |
| 3060 | { |
| 3061 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
| 3062 | |
| 3063 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET; |
| 3064 | PERL_UNUSED_CONTEXT; |
| 3065 | |
| 3066 | return mg ? mg->mg_len : 0; |
| 3067 | } |
| 3068 | |
| 3069 | void |
| 3070 | Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph) |
| 3071 | { |
| 3072 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
| 3073 | |
| 3074 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET; |
| 3075 | |
| 3076 | if (mg) { |
| 3077 | mg->mg_len = ph; |
| 3078 | } else if (ph) { |
| 3079 | if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph)) |
| 3080 | Perl_die(aTHX_ "panic: hv_placeholders_set"); |
| 3081 | } |
| 3082 | /* else we don't need to add magic to record 0 placeholders. */ |
| 3083 | } |
| 3084 | |
| 3085 | STATIC SV * |
| 3086 | S_refcounted_he_value(pTHX_ const struct refcounted_he *he) |
| 3087 | { |
| 3088 | dVAR; |
| 3089 | SV *value; |
| 3090 | |
| 3091 | PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE; |
| 3092 | |
| 3093 | switch(he->refcounted_he_data[0] & HVrhek_typemask) { |
| 3094 | case HVrhek_undef: |
| 3095 | value = newSV(0); |
| 3096 | break; |
| 3097 | case HVrhek_delete: |
| 3098 | value = &PL_sv_placeholder; |
| 3099 | break; |
| 3100 | case HVrhek_IV: |
| 3101 | value = newSViv(he->refcounted_he_val.refcounted_he_u_iv); |
| 3102 | break; |
| 3103 | case HVrhek_UV: |
| 3104 | value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv); |
| 3105 | break; |
| 3106 | case HVrhek_PV: |
| 3107 | case HVrhek_PV_UTF8: |
| 3108 | /* Create a string SV that directly points to the bytes in our |
| 3109 | structure. */ |
| 3110 | value = newSV_type(SVt_PV); |
| 3111 | SvPV_set(value, (char *) he->refcounted_he_data + 1); |
| 3112 | SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len); |
| 3113 | /* This stops anything trying to free it */ |
| 3114 | SvLEN_set(value, 0); |
| 3115 | SvPOK_on(value); |
| 3116 | SvREADONLY_on(value); |
| 3117 | if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8) |
| 3118 | SvUTF8_on(value); |
| 3119 | break; |
| 3120 | default: |
| 3121 | Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %"UVxf, |
| 3122 | (UV)he->refcounted_he_data[0]); |
| 3123 | } |
| 3124 | return value; |
| 3125 | } |
| 3126 | |
| 3127 | /* |
| 3128 | =for apidoc m|HV *|refcounted_he_chain_2hv|const struct refcounted_he *c|U32 flags |
| 3129 | |
| 3130 | Generates and returns a C<HV *> representing the content of a |
| 3131 | C<refcounted_he> chain. |
| 3132 | C<flags> is currently unused and must be zero. |
| 3133 | |
| 3134 | =cut |
| 3135 | */ |
| 3136 | HV * |
| 3137 | Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain, U32 flags) |
| 3138 | { |
| 3139 | dVAR; |
| 3140 | HV *hv; |
| 3141 | U32 placeholders, max; |
| 3142 | |
| 3143 | if (flags) |
| 3144 | Perl_croak(aTHX_ "panic: refcounted_he_chain_2hv bad flags %"UVxf, |
| 3145 | (UV)flags); |
| 3146 | |
| 3147 | /* We could chase the chain once to get an idea of the number of keys, |
| 3148 | and call ksplit. But for now we'll make a potentially inefficient |
| 3149 | hash with only 8 entries in its array. */ |
| 3150 | hv = newHV(); |
| 3151 | max = HvMAX(hv); |
| 3152 | if (!HvARRAY(hv)) { |
| 3153 | char *array; |
| 3154 | Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char); |
| 3155 | HvARRAY(hv) = (HE**)array; |
| 3156 | } |
| 3157 | |
| 3158 | placeholders = 0; |
| 3159 | while (chain) { |
| 3160 | #ifdef USE_ITHREADS |
| 3161 | U32 hash = chain->refcounted_he_hash; |
| 3162 | #else |
| 3163 | U32 hash = HEK_HASH(chain->refcounted_he_hek); |
| 3164 | #endif |
| 3165 | HE **oentry = &((HvARRAY(hv))[hash & max]); |
| 3166 | HE *entry = *oentry; |
| 3167 | SV *value; |
| 3168 | |
| 3169 | for (; entry; entry = HeNEXT(entry)) { |
| 3170 | if (HeHASH(entry) == hash) { |
| 3171 | /* We might have a duplicate key here. If so, entry is older |
| 3172 | than the key we've already put in the hash, so if they are |
| 3173 | the same, skip adding entry. */ |
| 3174 | #ifdef USE_ITHREADS |
| 3175 | const STRLEN klen = HeKLEN(entry); |
| 3176 | const char *const key = HeKEY(entry); |
| 3177 | if (klen == chain->refcounted_he_keylen |
| 3178 | && (!!HeKUTF8(entry) |
| 3179 | == !!(chain->refcounted_he_data[0] & HVhek_UTF8)) |
| 3180 | && memEQ(key, REF_HE_KEY(chain), klen)) |
| 3181 | goto next_please; |
| 3182 | #else |
| 3183 | if (HeKEY_hek(entry) == chain->refcounted_he_hek) |
| 3184 | goto next_please; |
| 3185 | if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek) |
| 3186 | && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek) |
| 3187 | && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek), |
| 3188 | HeKLEN(entry))) |
| 3189 | goto next_please; |
| 3190 | #endif |
| 3191 | } |
| 3192 | } |
| 3193 | assert (!entry); |
| 3194 | entry = new_HE(); |
| 3195 | |
| 3196 | #ifdef USE_ITHREADS |
| 3197 | HeKEY_hek(entry) |
| 3198 | = share_hek_flags(REF_HE_KEY(chain), |
| 3199 | chain->refcounted_he_keylen, |
| 3200 | chain->refcounted_he_hash, |
| 3201 | (chain->refcounted_he_data[0] |
| 3202 | & (HVhek_UTF8|HVhek_WASUTF8))); |
| 3203 | #else |
| 3204 | HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek); |
| 3205 | #endif |
| 3206 | value = refcounted_he_value(chain); |
| 3207 | if (value == &PL_sv_placeholder) |
| 3208 | placeholders++; |
| 3209 | HeVAL(entry) = value; |
| 3210 | |
| 3211 | /* Link it into the chain. */ |
| 3212 | HeNEXT(entry) = *oentry; |
| 3213 | *oentry = entry; |
| 3214 | |
| 3215 | HvTOTALKEYS(hv)++; |
| 3216 | |
| 3217 | next_please: |
| 3218 | chain = chain->refcounted_he_next; |
| 3219 | } |
| 3220 | |
| 3221 | if (placeholders) { |
| 3222 | clear_placeholders(hv, placeholders); |
| 3223 | HvTOTALKEYS(hv) -= placeholders; |
| 3224 | } |
| 3225 | |
| 3226 | /* We could check in the loop to see if we encounter any keys with key |
| 3227 | flags, but it's probably not worth it, as this per-hash flag is only |
| 3228 | really meant as an optimisation for things like Storable. */ |
| 3229 | HvHASKFLAGS_on(hv); |
| 3230 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
| 3231 | |
| 3232 | return hv; |
| 3233 | } |
| 3234 | |
| 3235 | /* |
| 3236 | =for apidoc m|SV *|refcounted_he_fetch_pvn|const struct refcounted_he *chain|const char *keypv|STRLEN keylen|U32 hash|U32 flags |
| 3237 | |
| 3238 | Search along a C<refcounted_he> chain for an entry with the key specified |
| 3239 | by C<keypv> and C<keylen>. If C<flags> has the C<REFCOUNTED_HE_KEY_UTF8> |
| 3240 | bit set, the key octets are interpreted as UTF-8, otherwise they |
| 3241 | are interpreted as Latin-1. C<hash> is a precomputed hash of the key |
| 3242 | string, or zero if it has not been precomputed. Returns a mortal scalar |
| 3243 | representing the value associated with the key, or C<&PL_sv_placeholder> |
| 3244 | if there is no value associated with the key. |
| 3245 | |
| 3246 | =cut |
| 3247 | */ |
| 3248 | |
| 3249 | SV * |
| 3250 | Perl_refcounted_he_fetch_pvn(pTHX_ const struct refcounted_he *chain, |
| 3251 | const char *keypv, STRLEN keylen, U32 hash, U32 flags) |
| 3252 | { |
| 3253 | dVAR; |
| 3254 | U8 utf8_flag; |
| 3255 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PVN; |
| 3256 | |
| 3257 | if (flags & ~(REFCOUNTED_HE_KEY_UTF8|REFCOUNTED_HE_EXISTS)) |
| 3258 | Perl_croak(aTHX_ "panic: refcounted_he_fetch_pvn bad flags %"UVxf, |
| 3259 | (UV)flags); |
| 3260 | if (!chain) |
| 3261 | goto ret; |
| 3262 | if (flags & REFCOUNTED_HE_KEY_UTF8) { |
| 3263 | /* For searching purposes, canonicalise to Latin-1 where possible. */ |
| 3264 | const char *keyend = keypv + keylen, *p; |
| 3265 | STRLEN nonascii_count = 0; |
| 3266 | for (p = keypv; p != keyend; p++) { |
| 3267 | if (! UTF8_IS_INVARIANT(*p)) { |
| 3268 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) { |
| 3269 | goto canonicalised_key; |
| 3270 | } |
| 3271 | nonascii_count++; |
| 3272 | p++; |
| 3273 | } |
| 3274 | } |
| 3275 | if (nonascii_count) { |
| 3276 | char *q; |
| 3277 | const char *p = keypv, *keyend = keypv + keylen; |
| 3278 | keylen -= nonascii_count; |
| 3279 | Newx(q, keylen, char); |
| 3280 | SAVEFREEPV(q); |
| 3281 | keypv = q; |
| 3282 | for (; p != keyend; p++, q++) { |
| 3283 | U8 c = (U8)*p; |
| 3284 | if (UTF8_IS_INVARIANT(c)) { |
| 3285 | *q = (char) c; |
| 3286 | } |
| 3287 | else { |
| 3288 | p++; |
| 3289 | *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p); |
| 3290 | } |
| 3291 | } |
| 3292 | } |
| 3293 | flags &= ~REFCOUNTED_HE_KEY_UTF8; |
| 3294 | canonicalised_key: ; |
| 3295 | } |
| 3296 | utf8_flag = (flags & REFCOUNTED_HE_KEY_UTF8) ? HVhek_UTF8 : 0; |
| 3297 | if (!hash) |
| 3298 | PERL_HASH(hash, keypv, keylen); |
| 3299 | |
| 3300 | for (; chain; chain = chain->refcounted_he_next) { |
| 3301 | if ( |
| 3302 | #ifdef USE_ITHREADS |
| 3303 | hash == chain->refcounted_he_hash && |
| 3304 | keylen == chain->refcounted_he_keylen && |
| 3305 | memEQ(REF_HE_KEY(chain), keypv, keylen) && |
| 3306 | utf8_flag == (chain->refcounted_he_data[0] & HVhek_UTF8) |
| 3307 | #else |
| 3308 | hash == HEK_HASH(chain->refcounted_he_hek) && |
| 3309 | keylen == (STRLEN)HEK_LEN(chain->refcounted_he_hek) && |
| 3310 | memEQ(HEK_KEY(chain->refcounted_he_hek), keypv, keylen) && |
| 3311 | utf8_flag == (HEK_FLAGS(chain->refcounted_he_hek) & HVhek_UTF8) |
| 3312 | #endif |
| 3313 | ) { |
| 3314 | if (flags & REFCOUNTED_HE_EXISTS) |
| 3315 | return (chain->refcounted_he_data[0] & HVrhek_typemask) |
| 3316 | == HVrhek_delete |
| 3317 | ? NULL : &PL_sv_yes; |
| 3318 | return sv_2mortal(refcounted_he_value(chain)); |
| 3319 | } |
| 3320 | } |
| 3321 | ret: |
| 3322 | return flags & REFCOUNTED_HE_EXISTS ? NULL : &PL_sv_placeholder; |
| 3323 | } |
| 3324 | |
| 3325 | /* |
| 3326 | =for apidoc m|SV *|refcounted_he_fetch_pv|const struct refcounted_he *chain|const char *key|U32 hash|U32 flags |
| 3327 | |
| 3328 | Like L</refcounted_he_fetch_pvn>, but takes a nul-terminated string |
| 3329 | instead of a string/length pair. |
| 3330 | |
| 3331 | =cut |
| 3332 | */ |
| 3333 | |
| 3334 | SV * |
| 3335 | Perl_refcounted_he_fetch_pv(pTHX_ const struct refcounted_he *chain, |
| 3336 | const char *key, U32 hash, U32 flags) |
| 3337 | { |
| 3338 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PV; |
| 3339 | return refcounted_he_fetch_pvn(chain, key, strlen(key), hash, flags); |
| 3340 | } |
| 3341 | |
| 3342 | /* |
| 3343 | =for apidoc m|SV *|refcounted_he_fetch_sv|const struct refcounted_he *chain|SV *key|U32 hash|U32 flags |
| 3344 | |
| 3345 | Like L</refcounted_he_fetch_pvn>, but takes a Perl scalar instead of a |
| 3346 | string/length pair. |
| 3347 | |
| 3348 | =cut |
| 3349 | */ |
| 3350 | |
| 3351 | SV * |
| 3352 | Perl_refcounted_he_fetch_sv(pTHX_ const struct refcounted_he *chain, |
| 3353 | SV *key, U32 hash, U32 flags) |
| 3354 | { |
| 3355 | const char *keypv; |
| 3356 | STRLEN keylen; |
| 3357 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_SV; |
| 3358 | if (flags & REFCOUNTED_HE_KEY_UTF8) |
| 3359 | Perl_croak(aTHX_ "panic: refcounted_he_fetch_sv bad flags %"UVxf, |
| 3360 | (UV)flags); |
| 3361 | keypv = SvPV_const(key, keylen); |
| 3362 | if (SvUTF8(key)) |
| 3363 | flags |= REFCOUNTED_HE_KEY_UTF8; |
| 3364 | if (!hash && SvIsCOW_shared_hash(key)) |
| 3365 | hash = SvSHARED_HASH(key); |
| 3366 | return refcounted_he_fetch_pvn(chain, keypv, keylen, hash, flags); |
| 3367 | } |
| 3368 | |
| 3369 | /* |
| 3370 | =for apidoc m|struct refcounted_he *|refcounted_he_new_pvn|struct refcounted_he *parent|const char *keypv|STRLEN keylen|U32 hash|SV *value|U32 flags |
| 3371 | |
| 3372 | Creates a new C<refcounted_he>. This consists of a single key/value |
| 3373 | pair and a reference to an existing C<refcounted_he> chain (which may |
| 3374 | be empty), and thus forms a longer chain. When using the longer chain, |
| 3375 | the new key/value pair takes precedence over any entry for the same key |
| 3376 | further along the chain. |
| 3377 | |
| 3378 | The new key is specified by C<keypv> and C<keylen>. If C<flags> has |
| 3379 | the C<REFCOUNTED_HE_KEY_UTF8> bit set, the key octets are interpreted |
| 3380 | as UTF-8, otherwise they are interpreted as Latin-1. C<hash> is |
| 3381 | a precomputed hash of the key string, or zero if it has not been |
| 3382 | precomputed. |
| 3383 | |
| 3384 | C<value> is the scalar value to store for this key. C<value> is copied |
| 3385 | by this function, which thus does not take ownership of any reference |
| 3386 | to it, and later changes to the scalar will not be reflected in the |
| 3387 | value visible in the C<refcounted_he>. Complex types of scalar will not |
| 3388 | be stored with referential integrity, but will be coerced to strings. |
| 3389 | C<value> may be either null or C<&PL_sv_placeholder> to indicate that no |
| 3390 | value is to be associated with the key; this, as with any non-null value, |
| 3391 | takes precedence over the existence of a value for the key further along |
| 3392 | the chain. |
| 3393 | |
| 3394 | C<parent> points to the rest of the C<refcounted_he> chain to be |
| 3395 | attached to the new C<refcounted_he>. This function takes ownership |
| 3396 | of one reference to C<parent>, and returns one reference to the new |
| 3397 | C<refcounted_he>. |
| 3398 | |
| 3399 | =cut |
| 3400 | */ |
| 3401 | |
| 3402 | struct refcounted_he * |
| 3403 | Perl_refcounted_he_new_pvn(pTHX_ struct refcounted_he *parent, |
| 3404 | const char *keypv, STRLEN keylen, U32 hash, SV *value, U32 flags) |
| 3405 | { |
| 3406 | dVAR; |
| 3407 | STRLEN value_len = 0; |
| 3408 | const char *value_p = NULL; |
| 3409 | bool is_pv; |
| 3410 | char value_type; |
| 3411 | char hekflags; |
| 3412 | STRLEN key_offset = 1; |
| 3413 | struct refcounted_he *he; |
| 3414 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PVN; |
| 3415 | |
| 3416 | if (!value || value == &PL_sv_placeholder) { |
| 3417 | value_type = HVrhek_delete; |
| 3418 | } else if (SvPOK(value)) { |
| 3419 | value_type = HVrhek_PV; |
| 3420 | } else if (SvIOK(value)) { |
| 3421 | value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV; |
| 3422 | } else if (!SvOK(value)) { |
| 3423 | value_type = HVrhek_undef; |
| 3424 | } else { |
| 3425 | value_type = HVrhek_PV; |
| 3426 | } |
| 3427 | is_pv = value_type == HVrhek_PV; |
| 3428 | if (is_pv) { |
| 3429 | /* Do it this way so that the SvUTF8() test is after the SvPV, in case |
| 3430 | the value is overloaded, and doesn't yet have the UTF-8flag set. */ |
| 3431 | value_p = SvPV_const(value, value_len); |
| 3432 | if (SvUTF8(value)) |
| 3433 | value_type = HVrhek_PV_UTF8; |
| 3434 | key_offset = value_len + 2; |
| 3435 | } |
| 3436 | hekflags = value_type; |
| 3437 | |
| 3438 | if (flags & REFCOUNTED_HE_KEY_UTF8) { |
| 3439 | /* Canonicalise to Latin-1 where possible. */ |
| 3440 | const char *keyend = keypv + keylen, *p; |
| 3441 | STRLEN nonascii_count = 0; |
| 3442 | for (p = keypv; p != keyend; p++) { |
| 3443 | if (! UTF8_IS_INVARIANT(*p)) { |
| 3444 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) { |
| 3445 | goto canonicalised_key; |
| 3446 | } |
| 3447 | nonascii_count++; |
| 3448 | p++; |
| 3449 | } |
| 3450 | } |
| 3451 | if (nonascii_count) { |
| 3452 | char *q; |
| 3453 | const char *p = keypv, *keyend = keypv + keylen; |
| 3454 | keylen -= nonascii_count; |
| 3455 | Newx(q, keylen, char); |
| 3456 | SAVEFREEPV(q); |
| 3457 | keypv = q; |
| 3458 | for (; p != keyend; p++, q++) { |
| 3459 | U8 c = (U8)*p; |
| 3460 | if (UTF8_IS_INVARIANT(c)) { |
| 3461 | *q = (char) c; |
| 3462 | } |
| 3463 | else { |
| 3464 | p++; |
| 3465 | *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p); |
| 3466 | } |
| 3467 | } |
| 3468 | } |
| 3469 | flags &= ~REFCOUNTED_HE_KEY_UTF8; |
| 3470 | canonicalised_key: ; |
| 3471 | } |
| 3472 | if (flags & REFCOUNTED_HE_KEY_UTF8) |
| 3473 | hekflags |= HVhek_UTF8; |
| 3474 | if (!hash) |
| 3475 | PERL_HASH(hash, keypv, keylen); |
| 3476 | |
| 3477 | #ifdef USE_ITHREADS |
| 3478 | he = (struct refcounted_he*) |
| 3479 | PerlMemShared_malloc(sizeof(struct refcounted_he) - 1 |
| 3480 | + keylen |
| 3481 | + key_offset); |
| 3482 | #else |
| 3483 | he = (struct refcounted_he*) |
| 3484 | PerlMemShared_malloc(sizeof(struct refcounted_he) - 1 |
| 3485 | + key_offset); |
| 3486 | #endif |
| 3487 | |
| 3488 | he->refcounted_he_next = parent; |
| 3489 | |
| 3490 | if (is_pv) { |
| 3491 | Copy(value_p, he->refcounted_he_data + 1, value_len + 1, char); |
| 3492 | he->refcounted_he_val.refcounted_he_u_len = value_len; |
| 3493 | } else if (value_type == HVrhek_IV) { |
| 3494 | he->refcounted_he_val.refcounted_he_u_iv = SvIVX(value); |
| 3495 | } else if (value_type == HVrhek_UV) { |
| 3496 | he->refcounted_he_val.refcounted_he_u_uv = SvUVX(value); |
| 3497 | } |
| 3498 | |
| 3499 | #ifdef USE_ITHREADS |
| 3500 | he->refcounted_he_hash = hash; |
| 3501 | he->refcounted_he_keylen = keylen; |
| 3502 | Copy(keypv, he->refcounted_he_data + key_offset, keylen, char); |
| 3503 | #else |
| 3504 | he->refcounted_he_hek = share_hek_flags(keypv, keylen, hash, hekflags); |
| 3505 | #endif |
| 3506 | |
| 3507 | he->refcounted_he_data[0] = hekflags; |
| 3508 | he->refcounted_he_refcnt = 1; |
| 3509 | |
| 3510 | return he; |
| 3511 | } |
| 3512 | |
| 3513 | /* |
| 3514 | =for apidoc m|struct refcounted_he *|refcounted_he_new_pv|struct refcounted_he *parent|const char *key|U32 hash|SV *value|U32 flags |
| 3515 | |
| 3516 | Like L</refcounted_he_new_pvn>, but takes a nul-terminated string instead |
| 3517 | of a string/length pair. |
| 3518 | |
| 3519 | =cut |
| 3520 | */ |
| 3521 | |
| 3522 | struct refcounted_he * |
| 3523 | Perl_refcounted_he_new_pv(pTHX_ struct refcounted_he *parent, |
| 3524 | const char *key, U32 hash, SV *value, U32 flags) |
| 3525 | { |
| 3526 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PV; |
| 3527 | return refcounted_he_new_pvn(parent, key, strlen(key), hash, value, flags); |
| 3528 | } |
| 3529 | |
| 3530 | /* |
| 3531 | =for apidoc m|struct refcounted_he *|refcounted_he_new_sv|struct refcounted_he *parent|SV *key|U32 hash|SV *value|U32 flags |
| 3532 | |
| 3533 | Like L</refcounted_he_new_pvn>, but takes a Perl scalar instead of a |
| 3534 | string/length pair. |
| 3535 | |
| 3536 | =cut |
| 3537 | */ |
| 3538 | |
| 3539 | struct refcounted_he * |
| 3540 | Perl_refcounted_he_new_sv(pTHX_ struct refcounted_he *parent, |
| 3541 | SV *key, U32 hash, SV *value, U32 flags) |
| 3542 | { |
| 3543 | const char *keypv; |
| 3544 | STRLEN keylen; |
| 3545 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_SV; |
| 3546 | if (flags & REFCOUNTED_HE_KEY_UTF8) |
| 3547 | Perl_croak(aTHX_ "panic: refcounted_he_new_sv bad flags %"UVxf, |
| 3548 | (UV)flags); |
| 3549 | keypv = SvPV_const(key, keylen); |
| 3550 | if (SvUTF8(key)) |
| 3551 | flags |= REFCOUNTED_HE_KEY_UTF8; |
| 3552 | if (!hash && SvIsCOW_shared_hash(key)) |
| 3553 | hash = SvSHARED_HASH(key); |
| 3554 | return refcounted_he_new_pvn(parent, keypv, keylen, hash, value, flags); |
| 3555 | } |
| 3556 | |
| 3557 | /* |
| 3558 | =for apidoc m|void|refcounted_he_free|struct refcounted_he *he |
| 3559 | |
| 3560 | Decrements the reference count of a C<refcounted_he> by one. If the |
| 3561 | reference count reaches zero the structure's memory is freed, which |
| 3562 | (recursively) causes a reduction of its parent C<refcounted_he>'s |
| 3563 | reference count. It is safe to pass a null pointer to this function: |
| 3564 | no action occurs in this case. |
| 3565 | |
| 3566 | =cut |
| 3567 | */ |
| 3568 | |
| 3569 | void |
| 3570 | Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) { |
| 3571 | #ifdef USE_ITHREADS |
| 3572 | dVAR; |
| 3573 | #endif |
| 3574 | PERL_UNUSED_CONTEXT; |
| 3575 | |
| 3576 | while (he) { |
| 3577 | struct refcounted_he *copy; |
| 3578 | U32 new_count; |
| 3579 | |
| 3580 | HINTS_REFCNT_LOCK; |
| 3581 | new_count = --he->refcounted_he_refcnt; |
| 3582 | HINTS_REFCNT_UNLOCK; |
| 3583 | |
| 3584 | if (new_count) { |
| 3585 | return; |
| 3586 | } |
| 3587 | |
| 3588 | #ifndef USE_ITHREADS |
| 3589 | unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0); |
| 3590 | #endif |
| 3591 | copy = he; |
| 3592 | he = he->refcounted_he_next; |
| 3593 | PerlMemShared_free(copy); |
| 3594 | } |
| 3595 | } |
| 3596 | |
| 3597 | /* |
| 3598 | =for apidoc m|struct refcounted_he *|refcounted_he_inc|struct refcounted_he *he |
| 3599 | |
| 3600 | Increment the reference count of a C<refcounted_he>. The pointer to the |
| 3601 | C<refcounted_he> is also returned. It is safe to pass a null pointer |
| 3602 | to this function: no action occurs and a null pointer is returned. |
| 3603 | |
| 3604 | =cut |
| 3605 | */ |
| 3606 | |
| 3607 | struct refcounted_he * |
| 3608 | Perl_refcounted_he_inc(pTHX_ struct refcounted_he *he) |
| 3609 | { |
| 3610 | #ifdef USE_ITHREADS |
| 3611 | dVAR; |
| 3612 | #endif |
| 3613 | PERL_UNUSED_CONTEXT; |
| 3614 | if (he) { |
| 3615 | HINTS_REFCNT_LOCK; |
| 3616 | he->refcounted_he_refcnt++; |
| 3617 | HINTS_REFCNT_UNLOCK; |
| 3618 | } |
| 3619 | return he; |
| 3620 | } |
| 3621 | |
| 3622 | /* |
| 3623 | =for apidoc cop_fetch_label |
| 3624 | |
| 3625 | Returns the label attached to a cop. |
| 3626 | The flags pointer may be set to C<SVf_UTF8> or 0. |
| 3627 | |
| 3628 | =cut |
| 3629 | */ |
| 3630 | |
| 3631 | /* pp_entereval is aware that labels are stored with a key ':' at the top of |
| 3632 | the linked list. */ |
| 3633 | const char * |
| 3634 | Perl_cop_fetch_label(pTHX_ COP *const cop, STRLEN *len, U32 *flags) { |
| 3635 | struct refcounted_he *const chain = cop->cop_hints_hash; |
| 3636 | |
| 3637 | PERL_ARGS_ASSERT_COP_FETCH_LABEL; |
| 3638 | PERL_UNUSED_CONTEXT; |
| 3639 | |
| 3640 | if (!chain) |
| 3641 | return NULL; |
| 3642 | #ifdef USE_ITHREADS |
| 3643 | if (chain->refcounted_he_keylen != 1) |
| 3644 | return NULL; |
| 3645 | if (*REF_HE_KEY(chain) != ':') |
| 3646 | return NULL; |
| 3647 | #else |
| 3648 | if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1) |
| 3649 | return NULL; |
| 3650 | if (*HEK_KEY(chain->refcounted_he_hek) != ':') |
| 3651 | return NULL; |
| 3652 | #endif |
| 3653 | /* Stop anyone trying to really mess us up by adding their own value for |
| 3654 | ':' into %^H */ |
| 3655 | if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV |
| 3656 | && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8) |
| 3657 | return NULL; |
| 3658 | |
| 3659 | if (len) |
| 3660 | *len = chain->refcounted_he_val.refcounted_he_u_len; |
| 3661 | if (flags) { |
| 3662 | *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask) |
| 3663 | == HVrhek_PV_UTF8) ? SVf_UTF8 : 0; |
| 3664 | } |
| 3665 | return chain->refcounted_he_data + 1; |
| 3666 | } |
| 3667 | |
| 3668 | /* |
| 3669 | =for apidoc cop_store_label |
| 3670 | |
| 3671 | Save a label into a C<cop_hints_hash>. |
| 3672 | You need to set flags to C<SVf_UTF8> |
| 3673 | for a UTF-8 label. |
| 3674 | |
| 3675 | =cut |
| 3676 | */ |
| 3677 | |
| 3678 | void |
| 3679 | Perl_cop_store_label(pTHX_ COP *const cop, const char *label, STRLEN len, |
| 3680 | U32 flags) |
| 3681 | { |
| 3682 | SV *labelsv; |
| 3683 | PERL_ARGS_ASSERT_COP_STORE_LABEL; |
| 3684 | |
| 3685 | if (flags & ~(SVf_UTF8)) |
| 3686 | Perl_croak(aTHX_ "panic: cop_store_label illegal flag bits 0x%" UVxf, |
| 3687 | (UV)flags); |
| 3688 | labelsv = newSVpvn_flags(label, len, SVs_TEMP); |
| 3689 | if (flags & SVf_UTF8) |
| 3690 | SvUTF8_on(labelsv); |
| 3691 | cop->cop_hints_hash |
| 3692 | = refcounted_he_new_pvs(cop->cop_hints_hash, ":", labelsv, 0); |
| 3693 | } |
| 3694 | |
| 3695 | /* |
| 3696 | =for apidoc hv_assert |
| 3697 | |
| 3698 | Check that a hash is in an internally consistent state. |
| 3699 | |
| 3700 | =cut |
| 3701 | */ |
| 3702 | |
| 3703 | #ifdef DEBUGGING |
| 3704 | |
| 3705 | void |
| 3706 | Perl_hv_assert(pTHX_ HV *hv) |
| 3707 | { |
| 3708 | dVAR; |
| 3709 | HE* entry; |
| 3710 | int withflags = 0; |
| 3711 | int placeholders = 0; |
| 3712 | int real = 0; |
| 3713 | int bad = 0; |
| 3714 | const I32 riter = HvRITER_get(hv); |
| 3715 | HE *eiter = HvEITER_get(hv); |
| 3716 | |
| 3717 | PERL_ARGS_ASSERT_HV_ASSERT; |
| 3718 | |
| 3719 | (void)hv_iterinit(hv); |
| 3720 | |
| 3721 | while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) { |
| 3722 | /* sanity check the values */ |
| 3723 | if (HeVAL(entry) == &PL_sv_placeholder) |
| 3724 | placeholders++; |
| 3725 | else |
| 3726 | real++; |
| 3727 | /* sanity check the keys */ |
| 3728 | if (HeSVKEY(entry)) { |
| 3729 | NOOP; /* Don't know what to check on SV keys. */ |
| 3730 | } else if (HeKUTF8(entry)) { |
| 3731 | withflags++; |
| 3732 | if (HeKWASUTF8(entry)) { |
| 3733 | PerlIO_printf(Perl_debug_log, |
| 3734 | "hash key has both WASUTF8 and UTF8: '%.*s'\n", |
| 3735 | (int) HeKLEN(entry), HeKEY(entry)); |
| 3736 | bad = 1; |
| 3737 | } |
| 3738 | } else if (HeKWASUTF8(entry)) |
| 3739 | withflags++; |
| 3740 | } |
| 3741 | if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) { |
| 3742 | static const char bad_count[] = "Count %d %s(s), but hash reports %d\n"; |
| 3743 | const int nhashkeys = HvUSEDKEYS(hv); |
| 3744 | const int nhashplaceholders = HvPLACEHOLDERS_get(hv); |
| 3745 | |
| 3746 | if (nhashkeys != real) { |
| 3747 | PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys ); |
| 3748 | bad = 1; |
| 3749 | } |
| 3750 | if (nhashplaceholders != placeholders) { |
| 3751 | PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders ); |
| 3752 | bad = 1; |
| 3753 | } |
| 3754 | } |
| 3755 | if (withflags && ! HvHASKFLAGS(hv)) { |
| 3756 | PerlIO_printf(Perl_debug_log, |
| 3757 | "Hash has HASKFLAGS off but I count %d key(s) with flags\n", |
| 3758 | withflags); |
| 3759 | bad = 1; |
| 3760 | } |
| 3761 | if (bad) { |
| 3762 | sv_dump(MUTABLE_SV(hv)); |
| 3763 | } |
| 3764 | HvRITER_set(hv, riter); /* Restore hash iterator state */ |
| 3765 | HvEITER_set(hv, eiter); |
| 3766 | } |
| 3767 | |
| 3768 | #endif |
| 3769 | |
| 3770 | /* |
| 3771 | * ex: set ts=8 sts=4 sw=4 et: |
| 3772 | */ |