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
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.
12 * I sit beside the fire and think
13 * of all that I have seen.
16 * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
20 =head1 Hash Manipulation Functions
22 A HV structure represents a Perl hash. It consists mainly of an array
23 of pointers, each of which points to a linked list of HE structures. The
24 array is indexed by the hash function of the key, so each linked list
25 represents all the hash entries with the same hash value. Each HE contains
26 a pointer to the actual value, plus a pointer to a HEK structure which
27 holds the key and hash value.
35 #define PERL_HASH_INTERNAL_ACCESS
38 #define HV_MAX_LENGTH_BEFORE_SPLIT 14
40 static const char S_strtab_error[]
41 = "Cannot modify shared string table in hv_%s";
47 /* We could generate this at compile time via (another) auxiliary C
49 const size_t arena_size = Perl_malloc_good_size(PERL_ARENA_SIZE);
50 HE* he = (HE*) Perl_get_arena(aTHX_ arena_size, HE_SVSLOT);
51 HE * const heend = &he[arena_size / sizeof(HE) - 1];
53 PL_body_roots[HE_SVSLOT] = he;
55 HeNEXT(he) = (HE*)(he + 1);
63 #define new_HE() (HE*)safemalloc(sizeof(HE))
64 #define del_HE(p) safefree((char*)p)
73 void ** const root = &PL_body_roots[HE_SVSLOT];
83 #define new_HE() new_he()
86 HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \
87 PL_body_roots[HE_SVSLOT] = p; \
95 S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
97 const int flags_masked = flags & HVhek_MASK;
101 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
103 Newx(k, HEK_BASESIZE + len + 2, char);
105 Copy(str, HEK_KEY(hek), len, char);
106 HEK_KEY(hek)[len] = 0;
108 HEK_HASH(hek) = hash;
109 HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED;
111 if (flags & HVhek_FREEKEY)
116 /* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
120 Perl_free_tied_hv_pool(pTHX)
123 HE *he = PL_hv_fetch_ent_mh;
126 Safefree(HeKEY_hek(he));
130 PL_hv_fetch_ent_mh = NULL;
133 #if defined(USE_ITHREADS)
135 Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
137 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
139 PERL_ARGS_ASSERT_HEK_DUP;
140 PERL_UNUSED_ARG(param);
143 /* We already shared this hash key. */
144 (void)share_hek_hek(shared);
148 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
149 HEK_HASH(source), HEK_FLAGS(source));
150 ptr_table_store(PL_ptr_table, source, shared);
156 Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
160 PERL_ARGS_ASSERT_HE_DUP;
164 /* look for it in the table first */
165 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
169 /* create anew and remember what it is */
171 ptr_table_store(PL_ptr_table, e, ret);
173 HeNEXT(ret) = he_dup(HeNEXT(e),shared, param);
174 if (HeKLEN(e) == HEf_SVKEY) {
176 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
177 HeKEY_hek(ret) = (HEK*)k;
178 HeKEY_sv(ret) = SvREFCNT_inc(sv_dup(HeKEY_sv(e), param));
181 /* This is hek_dup inlined, which seems to be important for speed
183 HEK * const source = HeKEY_hek(e);
184 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
187 /* We already shared this hash key. */
188 (void)share_hek_hek(shared);
192 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
193 HEK_HASH(source), HEK_FLAGS(source));
194 ptr_table_store(PL_ptr_table, source, shared);
196 HeKEY_hek(ret) = shared;
199 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
201 HeVAL(ret) = SvREFCNT_inc(sv_dup(HeVAL(e), param));
204 #endif /* USE_ITHREADS */
207 S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
210 SV * const sv = sv_newmortal();
212 PERL_ARGS_ASSERT_HV_NOTALLOWED;
214 if (!(flags & HVhek_FREEKEY)) {
215 sv_setpvn(sv, key, klen);
218 /* Need to free saved eventually assign to mortal SV */
219 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
220 sv_usepvn(sv, (char *) key, klen);
222 if (flags & HVhek_UTF8) {
225 Perl_croak(aTHX_ msg, SVfARG(sv));
228 /* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
234 Stores an SV in a hash. The hash key is specified as C<key> and C<klen> is
235 the length of the key. The C<hash> parameter is the precomputed hash
236 value; if it is zero then Perl will compute it. The return value will be
237 NULL if the operation failed or if the value did not need to be actually
238 stored within the hash (as in the case of tied hashes). Otherwise it can
239 be dereferenced to get the original C<SV*>. Note that the caller is
240 responsible for suitably incrementing the reference count of C<val> before
241 the call, and decrementing it if the function returned NULL. Effectively
242 a successful hv_store takes ownership of one reference to C<val>. This is
243 usually what you want; a newly created SV has a reference count of one, so
244 if all your code does is create SVs then store them in a hash, hv_store
245 will own the only reference to the new SV, and your code doesn't need to do
246 anything further to tidy up. hv_store is not implemented as a call to
247 hv_store_ent, and does not create a temporary SV for the key, so if your
248 key data is not already in SV form then use hv_store in preference to
251 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
252 information on how to use this function on tied hashes.
254 =for apidoc hv_store_ent
256 Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
257 parameter is the precomputed hash value; if it is zero then Perl will
258 compute it. The return value is the new hash entry so created. It will be
259 NULL if the operation failed or if the value did not need to be actually
260 stored within the hash (as in the case of tied hashes). Otherwise the
261 contents of the return value can be accessed using the C<He?> macros
262 described here. Note that the caller is responsible for suitably
263 incrementing the reference count of C<val> before the call, and
264 decrementing it if the function returned NULL. Effectively a successful
265 hv_store_ent takes ownership of one reference to C<val>. This is
266 usually what you want; a newly created SV has a reference count of one, so
267 if all your code does is create SVs then store them in a hash, hv_store
268 will own the only reference to the new SV, and your code doesn't need to do
269 anything further to tidy up. Note that hv_store_ent only reads the C<key>;
270 unlike C<val> it does not take ownership of it, so maintaining the correct
271 reference count on C<key> is entirely the caller's responsibility. hv_store
272 is not implemented as a call to hv_store_ent, and does not create a temporary
273 SV for the key, so if your key data is not already in SV form then use
274 hv_store in preference to hv_store_ent.
276 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
277 information on how to use this function on tied hashes.
279 =for apidoc hv_exists
281 Returns a boolean indicating whether the specified hash key exists. The
282 C<klen> is the length of the key.
286 Returns the SV which corresponds to the specified key in the hash. The
287 C<klen> is the length of the key. If C<lval> is set then the fetch will be
288 part of a store. Check that the return value is non-null before
289 dereferencing it to an C<SV*>.
291 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
292 information on how to use this function on tied hashes.
294 =for apidoc hv_exists_ent
296 Returns a boolean indicating whether the specified hash key exists. C<hash>
297 can be a valid precomputed hash value, or 0 to ask for it to be
303 /* returns an HE * structure with the all fields set */
304 /* note that hent_val will be a mortal sv for MAGICAL hashes */
306 =for apidoc hv_fetch_ent
308 Returns the hash entry which corresponds to the specified key in the hash.
309 C<hash> must be a valid precomputed hash number for the given C<key>, or 0
310 if you want the function to compute it. IF C<lval> is set then the fetch
311 will be part of a store. Make sure the return value is non-null before
312 accessing it. The return value when C<tb> is a tied hash is a pointer to a
313 static location, so be sure to make a copy of the structure if you need to
316 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
317 information on how to use this function on tied hashes.
322 /* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
324 Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
325 const int action, SV *val, const U32 hash)
330 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
339 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
343 Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
344 int flags, int action, SV *val, register U32 hash)
353 const int return_svp = action & HV_FETCH_JUST_SV;
357 if (SvTYPE(hv) == SVTYPEMASK)
360 assert(SvTYPE(hv) == SVt_PVHV);
362 if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) {
364 if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) {
365 struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr;
366 if (uf->uf_set == NULL) {
367 SV* obj = mg->mg_obj;
370 keysv = newSVpvn_flags(key, klen, SVs_TEMP |
371 ((flags & HVhek_UTF8)
375 mg->mg_obj = keysv; /* pass key */
376 uf->uf_index = action; /* pass action */
377 magic_getuvar(MUTABLE_SV(hv), mg);
378 keysv = mg->mg_obj; /* may have changed */
381 /* If the key may have changed, then we need to invalidate
382 any passed-in computed hash value. */
388 if (flags & HVhek_FREEKEY)
390 key = SvPV_const(keysv, klen);
392 is_utf8 = (SvUTF8(keysv) != 0);
394 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
397 if (action & HV_DELETE) {
398 return (void *) hv_delete_common(hv, keysv, key, klen,
399 flags | (is_utf8 ? HVhek_UTF8 : 0),
403 xhv = (XPVHV*)SvANY(hv);
405 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
406 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
407 || SvGMAGICAL((const SV *)hv))
409 /* FIXME should be able to skimp on the HE/HEK here when
410 HV_FETCH_JUST_SV is true. */
412 keysv = newSVpvn_utf8(key, klen, is_utf8);
414 keysv = newSVsv(keysv);
417 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
419 /* grab a fake HE/HEK pair from the pool or make a new one */
420 entry = PL_hv_fetch_ent_mh;
422 PL_hv_fetch_ent_mh = HeNEXT(entry);
426 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
427 HeKEY_hek(entry) = (HEK*)k;
429 HeNEXT(entry) = NULL;
430 HeSVKEY_set(entry, keysv);
432 sv_upgrade(sv, SVt_PVLV);
434 /* so we can free entry when freeing sv */
435 LvTARG(sv) = MUTABLE_SV(entry);
437 /* XXX remove at some point? */
438 if (flags & HVhek_FREEKEY)
442 return entry ? (void *) &HeVAL(entry) : NULL;
444 return (void *) entry;
446 #ifdef ENV_IS_CASELESS
447 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
449 for (i = 0; i < klen; ++i)
450 if (isLOWER(key[i])) {
451 /* Would be nice if we had a routine to do the
452 copy and upercase in a single pass through. */
453 const char * const nkey = strupr(savepvn(key,klen));
454 /* Note that this fetch is for nkey (the uppercased
455 key) whereas the store is for key (the original) */
456 void *result = hv_common(hv, NULL, nkey, klen,
457 HVhek_FREEKEY, /* free nkey */
458 0 /* non-LVAL fetch */
459 | HV_DISABLE_UVAR_XKEY
462 0 /* compute hash */);
463 if (!result && (action & HV_FETCH_LVALUE)) {
464 /* This call will free key if necessary.
465 Do it this way to encourage compiler to tail
467 result = hv_common(hv, keysv, key, klen, flags,
469 | HV_DISABLE_UVAR_XKEY
473 if (flags & HVhek_FREEKEY)
481 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
482 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
483 || SvGMAGICAL((const SV *)hv)) {
484 /* I don't understand why hv_exists_ent has svret and sv,
485 whereas hv_exists only had one. */
486 SV * const svret = sv_newmortal();
489 if (keysv || is_utf8) {
491 keysv = newSVpvn_utf8(key, klen, TRUE);
493 keysv = newSVsv(keysv);
495 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
497 mg_copy(MUTABLE_SV(hv), sv, key, klen);
499 if (flags & HVhek_FREEKEY)
501 magic_existspack(svret, mg_find(sv, PERL_MAGIC_tiedelem));
502 /* This cast somewhat evil, but I'm merely using NULL/
503 not NULL to return the boolean exists.
504 And I know hv is not NULL. */
505 return SvTRUE(svret) ? (void *)hv : NULL;
507 #ifdef ENV_IS_CASELESS
508 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
509 /* XXX This code isn't UTF8 clean. */
510 char * const keysave = (char * const)key;
511 /* Will need to free this, so set FREEKEY flag. */
512 key = savepvn(key,klen);
513 key = (const char*)strupr((char*)key);
518 if (flags & HVhek_FREEKEY) {
521 flags |= HVhek_FREEKEY;
525 else if (action & HV_FETCH_ISSTORE) {
528 hv_magic_check (hv, &needs_copy, &needs_store);
530 const bool save_taint = PL_tainted;
531 if (keysv || is_utf8) {
533 keysv = newSVpvn_utf8(key, klen, TRUE);
536 PL_tainted = SvTAINTED(keysv);
537 keysv = sv_2mortal(newSVsv(keysv));
538 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
540 mg_copy(MUTABLE_SV(hv), val, key, klen);
543 TAINT_IF(save_taint);
545 if (flags & HVhek_FREEKEY)
549 #ifdef ENV_IS_CASELESS
550 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
551 /* XXX This code isn't UTF8 clean. */
552 const char *keysave = key;
553 /* Will need to free this, so set FREEKEY flag. */
554 key = savepvn(key,klen);
555 key = (const char*)strupr((char*)key);
560 if (flags & HVhek_FREEKEY) {
563 flags |= HVhek_FREEKEY;
571 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
572 #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
573 || (SvRMAGICAL((const SV *)hv)
574 && mg_find((const SV *)hv, PERL_MAGIC_env))
579 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
581 HvARRAY(hv) = (HE**)array;
583 #ifdef DYNAMIC_ENV_FETCH
584 else if (action & HV_FETCH_ISEXISTS) {
585 /* for an %ENV exists, if we do an insert it's by a recursive
586 store call, so avoid creating HvARRAY(hv) right now. */
590 /* XXX remove at some point? */
591 if (flags & HVhek_FREEKEY)
599 char * const keysave = (char *)key;
600 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
604 flags &= ~HVhek_UTF8;
605 if (key != keysave) {
606 if (flags & HVhek_FREEKEY)
608 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
613 PERL_HASH_INTERNAL(hash, key, klen);
614 /* We don't have a pointer to the hv, so we have to replicate the
615 flag into every HEK, so that hv_iterkeysv can see it. */
616 /* And yes, you do need this even though you are not "storing" because
617 you can flip the flags below if doing an lval lookup. (And that
618 was put in to give the semantics Andreas was expecting.) */
619 flags |= HVhek_REHASH;
621 if (keysv && (SvIsCOW_shared_hash(keysv))) {
622 hash = SvSHARED_HASH(keysv);
624 PERL_HASH(hash, key, klen);
628 masked_flags = (flags & HVhek_MASK);
630 #ifdef DYNAMIC_ENV_FETCH
631 if (!HvARRAY(hv)) entry = NULL;
635 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
637 for (; entry; entry = HeNEXT(entry)) {
638 if (HeHASH(entry) != hash) /* strings can't be equal */
640 if (HeKLEN(entry) != (I32)klen)
642 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
644 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
647 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
648 if (HeKFLAGS(entry) != masked_flags) {
649 /* We match if HVhek_UTF8 bit in our flags and hash key's
650 match. But if entry was set previously with HVhek_WASUTF8
651 and key now doesn't (or vice versa) then we should change
652 the key's flag, as this is assignment. */
653 if (HvSHAREKEYS(hv)) {
654 /* Need to swap the key we have for a key with the flags we
655 need. As keys are shared we can't just write to the
656 flag, so we share the new one, unshare the old one. */
657 HEK * const new_hek = share_hek_flags(key, klen, hash,
659 unshare_hek (HeKEY_hek(entry));
660 HeKEY_hek(entry) = new_hek;
662 else if (hv == PL_strtab) {
663 /* PL_strtab is usually the only hash without HvSHAREKEYS,
664 so putting this test here is cheap */
665 if (flags & HVhek_FREEKEY)
667 Perl_croak(aTHX_ S_strtab_error,
668 action & HV_FETCH_LVALUE ? "fetch" : "store");
671 HeKFLAGS(entry) = masked_flags;
672 if (masked_flags & HVhek_ENABLEHVKFLAGS)
675 if (HeVAL(entry) == &PL_sv_placeholder) {
676 /* yes, can store into placeholder slot */
677 if (action & HV_FETCH_LVALUE) {
679 /* This preserves behaviour with the old hv_fetch
680 implementation which at this point would bail out
681 with a break; (at "if we find a placeholder, we
682 pretend we haven't found anything")
684 That break mean that if a placeholder were found, it
685 caused a call into hv_store, which in turn would
686 check magic, and if there is no magic end up pretty
687 much back at this point (in hv_store's code). */
690 /* LVAL fetch which actaully needs a store. */
692 HvPLACEHOLDERS(hv)--;
695 if (val != &PL_sv_placeholder)
696 HvPLACEHOLDERS(hv)--;
699 } else if (action & HV_FETCH_ISSTORE) {
700 SvREFCNT_dec(HeVAL(entry));
703 } else if (HeVAL(entry) == &PL_sv_placeholder) {
704 /* if we find a placeholder, we pretend we haven't found
708 if (flags & HVhek_FREEKEY)
711 return entry ? (void *) &HeVAL(entry) : NULL;
715 #ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
716 if (!(action & HV_FETCH_ISSTORE)
717 && SvRMAGICAL((const SV *)hv)
718 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
720 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
722 sv = newSVpvn(env,len);
724 return hv_common(hv, keysv, key, klen, flags,
725 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
731 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
732 hv_notallowed(flags, key, klen,
733 "Attempt to access disallowed key '%"SVf"' in"
734 " a restricted hash");
736 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
737 /* Not doing some form of store, so return failure. */
738 if (flags & HVhek_FREEKEY)
742 if (action & HV_FETCH_LVALUE) {
745 /* At this point the old hv_fetch code would call to hv_store,
746 which in turn might do some tied magic. So we need to make that
747 magic check happen. */
748 /* gonna assign to this, so it better be there */
749 /* If a fetch-as-store fails on the fetch, then the action is to
750 recurse once into "hv_store". If we didn't do this, then that
751 recursive call would call the key conversion routine again.
752 However, as we replace the original key with the converted
753 key, this would result in a double conversion, which would show
754 up as a bug if the conversion routine is not idempotent. */
755 return hv_common(hv, keysv, key, klen, flags,
756 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
758 /* XXX Surely that could leak if the fetch-was-store fails?
759 Just like the hv_fetch. */
763 /* Welcome to hv_store... */
766 /* Not sure if we can get here. I think the only case of oentry being
767 NULL is for %ENV with dynamic env fetch. But that should disappear
768 with magic in the previous code. */
771 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
773 HvARRAY(hv) = (HE**)array;
776 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
779 /* share_hek_flags will do the free for us. This might be considered
782 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
783 else if (hv == PL_strtab) {
784 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
785 this test here is cheap */
786 if (flags & HVhek_FREEKEY)
788 Perl_croak(aTHX_ S_strtab_error,
789 action & HV_FETCH_LVALUE ? "fetch" : "store");
791 else /* gotta do the real thing */
792 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
794 HeNEXT(entry) = *oentry;
797 if (val == &PL_sv_placeholder)
798 HvPLACEHOLDERS(hv)++;
799 if (masked_flags & HVhek_ENABLEHVKFLAGS)
803 const HE *counter = HeNEXT(entry);
805 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
806 if (!counter) { /* initial entry? */
807 xhv->xhv_fill++; /* HvFILL(hv)++ */
808 } else if (xhv->xhv_keys > (IV)xhv->xhv_max) {
810 } else if(!HvREHASH(hv)) {
813 while ((counter = HeNEXT(counter)))
816 if (n_links > HV_MAX_LENGTH_BEFORE_SPLIT) {
817 /* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit
818 bucket splits on a rehashed hash, as we're not going to
819 split it again, and if someone is lucky (evil) enough to
820 get all the keys in one list they could exhaust our memory
821 as we repeatedly double the number of buckets on every
822 entry. Linear search feels a less worse thing to do. */
829 return entry ? (void *) &HeVAL(entry) : NULL;
831 return (void *) entry;
835 S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
837 const MAGIC *mg = SvMAGIC(hv);
839 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
844 if (isUPPER(mg->mg_type)) {
846 if (mg->mg_type == PERL_MAGIC_tied) {
847 *needs_store = FALSE;
848 return; /* We've set all there is to set. */
851 mg = mg->mg_moremagic;
856 =for apidoc hv_scalar
858 Evaluates the hash in scalar context and returns the result. Handles magic when the hash is tied.
864 Perl_hv_scalar(pTHX_ HV *hv)
868 PERL_ARGS_ASSERT_HV_SCALAR;
870 if (SvRMAGICAL(hv)) {
871 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
873 return magic_scalarpack(hv, mg);
877 if (HvFILL((const HV *)hv))
878 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
879 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
887 =for apidoc hv_delete
889 Deletes a key/value pair in the hash. The value SV is removed from the
890 hash and returned to the caller. The C<klen> is the length of the key.
891 The C<flags> value will normally be zero; if set to G_DISCARD then NULL
894 =for apidoc hv_delete_ent
896 Deletes a key/value pair in the hash. The value SV is removed from the
897 hash and returned to the caller. The C<flags> value will normally be zero;
898 if set to G_DISCARD then NULL will be returned. C<hash> can be a valid
899 precomputed hash value, or 0 to ask for it to be computed.
905 S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
906 int k_flags, I32 d_flags, U32 hash)
911 register HE **oentry;
912 HE *const *first_entry;
913 bool is_utf8 = (k_flags & HVhek_UTF8) ? TRUE : FALSE;
916 if (SvRMAGICAL(hv)) {
919 hv_magic_check (hv, &needs_copy, &needs_store);
923 entry = (HE *) hv_common(hv, keysv, key, klen,
924 k_flags & ~HVhek_FREEKEY,
925 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
927 sv = entry ? HeVAL(entry) : NULL;
933 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
934 /* No longer an element */
935 sv_unmagic(sv, PERL_MAGIC_tiedelem);
938 return NULL; /* element cannot be deleted */
940 #ifdef ENV_IS_CASELESS
941 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
942 /* XXX This code isn't UTF8 clean. */
943 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
944 if (k_flags & HVhek_FREEKEY) {
947 key = strupr(SvPVX(keysv));
956 xhv = (XPVHV*)SvANY(hv);
961 const char * const keysave = key;
962 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
965 k_flags |= HVhek_UTF8;
967 k_flags &= ~HVhek_UTF8;
968 if (key != keysave) {
969 if (k_flags & HVhek_FREEKEY) {
970 /* This shouldn't happen if our caller does what we expect,
971 but strictly the API allows it. */
974 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
976 HvHASKFLAGS_on(MUTABLE_SV(hv));
980 PERL_HASH_INTERNAL(hash, key, klen);
982 if (keysv && (SvIsCOW_shared_hash(keysv))) {
983 hash = SvSHARED_HASH(keysv);
985 PERL_HASH(hash, key, klen);
989 masked_flags = (k_flags & HVhek_MASK);
991 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
993 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
995 if (HeHASH(entry) != hash) /* strings can't be equal */
997 if (HeKLEN(entry) != (I32)klen)
999 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
1001 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
1004 if (hv == PL_strtab) {
1005 if (k_flags & HVhek_FREEKEY)
1007 Perl_croak(aTHX_ S_strtab_error, "delete");
1010 /* if placeholder is here, it's already been deleted.... */
1011 if (HeVAL(entry) == &PL_sv_placeholder) {
1012 if (k_flags & HVhek_FREEKEY)
1016 if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1017 hv_notallowed(k_flags, key, klen,
1018 "Attempt to delete readonly key '%"SVf"' from"
1019 " a restricted hash");
1021 if (k_flags & HVhek_FREEKEY)
1024 if (d_flags & G_DISCARD)
1027 sv = sv_2mortal(HeVAL(entry));
1028 HeVAL(entry) = &PL_sv_placeholder;
1032 * If a restricted hash, rather than really deleting the entry, put
1033 * a placeholder there. This marks the key as being "approved", so
1034 * we can still access via not-really-existing key without raising
1037 if (SvREADONLY(hv)) {
1038 SvREFCNT_dec(HeVAL(entry));
1039 HeVAL(entry) = &PL_sv_placeholder;
1040 /* We'll be saving this slot, so the number of allocated keys
1041 * doesn't go down, but the number placeholders goes up */
1042 HvPLACEHOLDERS(hv)++;
1044 *oentry = HeNEXT(entry);
1046 xhv->xhv_fill--; /* HvFILL(hv)-- */
1048 if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */)
1051 hv_free_ent(hv, entry);
1052 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
1053 if (xhv->xhv_keys == 0)
1054 HvHASKFLAGS_off(hv);
1058 if (SvREADONLY(hv)) {
1059 hv_notallowed(k_flags, key, klen,
1060 "Attempt to delete disallowed key '%"SVf"' from"
1061 " a restricted hash");
1064 if (k_flags & HVhek_FREEKEY)
1070 S_hsplit(pTHX_ HV *hv)
1073 register XPVHV* const xhv = (XPVHV*)SvANY(hv);
1074 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
1075 register I32 newsize = oldsize * 2;
1077 char *a = (char*) HvARRAY(hv);
1079 register HE **oentry;
1080 int longest_chain = 0;
1083 PERL_ARGS_ASSERT_HSPLIT;
1085 /*PerlIO_printf(PerlIO_stderr(), "hsplit called for %p which had %d\n",
1086 (void*)hv, (int) oldsize);*/
1088 if (HvPLACEHOLDERS_get(hv) && !SvREADONLY(hv)) {
1089 /* Can make this clear any placeholders first for non-restricted hashes,
1090 even though Storable rebuilds restricted hashes by putting in all the
1091 placeholders (first) before turning on the readonly flag, because
1092 Storable always pre-splits the hash. */
1093 hv_clear_placeholders(hv);
1097 #if defined(STRANGE_MALLOC) || defined(MYMALLOC)
1098 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1099 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1105 Move(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1108 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1109 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1114 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
1116 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1118 if (oldsize >= 64) {
1119 offer_nice_chunk(HvARRAY(hv),
1120 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1121 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
1124 Safefree(HvARRAY(hv));
1128 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
1129 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
1130 HvARRAY(hv) = (HE**) a;
1133 for (i=0; i<oldsize; i++,aep++) {
1134 int left_length = 0;
1135 int right_length = 0;
1139 if (!*aep) /* non-existent */
1142 for (oentry = aep, entry = *aep; entry; entry = *oentry) {
1143 if ((HeHASH(entry) & newsize) != (U32)i) {
1144 *oentry = HeNEXT(entry);
1145 HeNEXT(entry) = *bep;
1147 xhv->xhv_fill++; /* HvFILL(hv)++ */
1153 oentry = &HeNEXT(entry);
1157 if (!*aep) /* everything moved */
1158 xhv->xhv_fill--; /* HvFILL(hv)-- */
1159 /* I think we don't actually need to keep track of the longest length,
1160 merely flag if anything is too long. But for the moment while
1161 developing this code I'll track it. */
1162 if (left_length > longest_chain)
1163 longest_chain = left_length;
1164 if (right_length > longest_chain)
1165 longest_chain = right_length;
1169 /* Pick your policy for "hashing isn't working" here: */
1170 if (longest_chain <= HV_MAX_LENGTH_BEFORE_SPLIT /* split worked? */
1175 if (hv == PL_strtab) {
1176 /* Urg. Someone is doing something nasty to the string table.
1181 /* Awooga. Awooga. Pathological data. */
1182 /*PerlIO_printf(PerlIO_stderr(), "%p %d of %d with %d/%d buckets\n", (void*)hv,
1183 longest_chain, HvTOTALKEYS(hv), HvFILL(hv), 1+HvMAX(hv));*/
1186 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1187 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1189 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1192 was_shared = HvSHAREKEYS(hv);
1195 HvSHAREKEYS_off(hv);
1200 for (i=0; i<newsize; i++,aep++) {
1201 register HE *entry = *aep;
1203 /* We're going to trash this HE's next pointer when we chain it
1204 into the new hash below, so store where we go next. */
1205 HE * const next = HeNEXT(entry);
1210 PERL_HASH_INTERNAL(hash, HeKEY(entry), HeKLEN(entry));
1215 = save_hek_flags(HeKEY(entry), HeKLEN(entry),
1216 hash, HeKFLAGS(entry));
1217 unshare_hek (HeKEY_hek(entry));
1218 HeKEY_hek(entry) = new_hek;
1220 /* Not shared, so simply write the new hash in. */
1221 HeHASH(entry) = hash;
1223 /*PerlIO_printf(PerlIO_stderr(), "%d ", HeKFLAGS(entry));*/
1224 HEK_REHASH_on(HeKEY_hek(entry));
1225 /*PerlIO_printf(PerlIO_stderr(), "%d\n", HeKFLAGS(entry));*/
1227 /* Copy oentry to the correct new chain. */
1228 bep = ((HE**)a) + (hash & (I32) xhv->xhv_max);
1230 xhv->xhv_fill++; /* HvFILL(hv)++ */
1231 HeNEXT(entry) = *bep;
1237 Safefree (HvARRAY(hv));
1238 HvARRAY(hv) = (HE **)a;
1242 Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
1245 register XPVHV* xhv = (XPVHV*)SvANY(hv);
1246 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
1247 register I32 newsize;
1252 register HE **oentry;
1254 PERL_ARGS_ASSERT_HV_KSPLIT;
1256 newsize = (I32) newmax; /* possible truncation here */
1257 if (newsize != newmax || newmax <= oldsize)
1259 while ((newsize & (1 + ~newsize)) != newsize) {
1260 newsize &= ~(newsize & (1 + ~newsize)); /* get proper power of 2 */
1262 if (newsize < newmax)
1264 if (newsize < newmax)
1265 return; /* overflow detection */
1267 a = (char *) HvARRAY(hv);
1270 #if defined(STRANGE_MALLOC) || defined(MYMALLOC)
1271 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1272 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1278 Copy(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1281 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1282 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1287 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
1289 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1291 if (oldsize >= 64) {
1292 offer_nice_chunk(HvARRAY(hv),
1293 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1294 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
1297 Safefree(HvARRAY(hv));
1300 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
1303 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
1305 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
1306 HvARRAY(hv) = (HE **) a;
1307 if (!xhv->xhv_fill /* !HvFILL(hv) */) /* skip rest if no entries */
1311 for (i=0; i<oldsize; i++,aep++) {
1312 if (!*aep) /* non-existent */
1314 for (oentry = aep, entry = *aep; entry; entry = *oentry) {
1315 register I32 j = (HeHASH(entry) & newsize);
1319 *oentry = HeNEXT(entry);
1320 if (!(HeNEXT(entry) = aep[j]))
1321 xhv->xhv_fill++; /* HvFILL(hv)++ */
1326 oentry = &HeNEXT(entry);
1328 if (!*aep) /* everything moved */
1329 xhv->xhv_fill--; /* HvFILL(hv)-- */
1334 Perl_newHVhv(pTHX_ HV *ohv)
1336 HV * const hv = newHV();
1337 STRLEN hv_max, hv_fill;
1339 if (!ohv || (hv_fill = HvFILL(ohv)) == 0)
1341 hv_max = HvMAX(ohv);
1343 if (!SvMAGICAL((const SV *)ohv)) {
1344 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
1346 const bool shared = !!HvSHAREKEYS(ohv);
1347 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
1349 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
1352 /* In each bucket... */
1353 for (i = 0; i <= hv_max; i++) {
1355 HE *oent = oents[i];
1362 /* Copy the linked list of entries. */
1363 for (; oent; oent = HeNEXT(oent)) {
1364 const U32 hash = HeHASH(oent);
1365 const char * const key = HeKEY(oent);
1366 const STRLEN len = HeKLEN(oent);
1367 const int flags = HeKFLAGS(oent);
1368 HE * const ent = new_HE();
1370 HeVAL(ent) = newSVsv(HeVAL(oent));
1372 = shared ? share_hek_flags(key, len, hash, flags)
1373 : save_hek_flags(key, len, hash, flags);
1384 HvFILL(hv) = hv_fill;
1385 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
1389 /* Iterate over ohv, copying keys and values one at a time. */
1391 const I32 riter = HvRITER_get(ohv);
1392 HE * const eiter = HvEITER_get(ohv);
1394 /* Can we use fewer buckets? (hv_max is always 2^n-1) */
1395 while (hv_max && hv_max + 1 >= hv_fill * 2)
1396 hv_max = hv_max / 2;
1400 while ((entry = hv_iternext_flags(ohv, 0))) {
1401 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1402 newSVsv(HeVAL(entry)), HeHASH(entry),
1405 HvRITER_set(ohv, riter);
1406 HvEITER_set(ohv, eiter);
1412 /* A rather specialised version of newHVhv for copying %^H, ensuring all the
1413 magic stays on it. */
1415 Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1417 HV * const hv = newHV();
1420 if (ohv && (hv_fill = HvFILL(ohv))) {
1421 STRLEN hv_max = HvMAX(ohv);
1423 const I32 riter = HvRITER_get(ohv);
1424 HE * const eiter = HvEITER_get(ohv);
1426 while (hv_max && hv_max + 1 >= hv_fill * 2)
1427 hv_max = hv_max / 2;
1431 while ((entry = hv_iternext_flags(ohv, 0))) {
1432 SV *const sv = newSVsv(HeVAL(entry));
1433 sv_magic(sv, NULL, PERL_MAGIC_hintselem,
1434 (char *)newSVhek (HeKEY_hek(entry)), HEf_SVKEY);
1435 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1436 sv, HeHASH(entry), HeKFLAGS(entry));
1438 HvRITER_set(ohv, riter);
1439 HvEITER_set(ohv, eiter);
1441 hv_magic(hv, NULL, PERL_MAGIC_hints);
1446 Perl_hv_free_ent(pTHX_ HV *hv, register HE *entry)
1451 PERL_ARGS_ASSERT_HV_FREE_ENT;
1456 if (val && isGV(val) && isGV_with_GP(val) && GvCVu(val) && HvNAME_get(hv))
1457 mro_method_changed_in(hv); /* deletion of method from stash */
1459 if (HeKLEN(entry) == HEf_SVKEY) {
1460 SvREFCNT_dec(HeKEY_sv(entry));
1461 Safefree(HeKEY_hek(entry));
1463 else if (HvSHAREKEYS(hv))
1464 unshare_hek(HeKEY_hek(entry));
1466 Safefree(HeKEY_hek(entry));
1471 Perl_hv_delayfree_ent(pTHX_ HV *hv, register HE *entry)
1475 PERL_ARGS_ASSERT_HV_DELAYFREE_ENT;
1479 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1480 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
1481 if (HeKLEN(entry) == HEf_SVKEY) {
1482 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
1484 hv_free_ent(hv, entry);
1488 =for apidoc hv_clear
1490 Clears a hash, making it empty.
1496 Perl_hv_clear(pTHX_ HV *hv)
1499 register XPVHV* xhv;
1503 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1505 xhv = (XPVHV*)SvANY(hv);
1507 if (SvREADONLY(hv) && HvARRAY(hv) != NULL) {
1508 /* restricted hash: convert all keys to placeholders */
1510 for (i = 0; i <= xhv->xhv_max; i++) {
1511 HE *entry = (HvARRAY(hv))[i];
1512 for (; entry; entry = HeNEXT(entry)) {
1513 /* not already placeholder */
1514 if (HeVAL(entry) != &PL_sv_placeholder) {
1515 if (HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1516 SV* const keysv = hv_iterkeysv(entry);
1518 "Attempt to delete readonly key '%"SVf"' from a restricted hash",
1521 SvREFCNT_dec(HeVAL(entry));
1522 HeVAL(entry) = &PL_sv_placeholder;
1523 HvPLACEHOLDERS(hv)++;
1531 HvPLACEHOLDERS_set(hv, 0);
1533 Zero(HvARRAY(hv), xhv->xhv_max+1 /* HvMAX(hv)+1 */, HE*);
1536 mg_clear(MUTABLE_SV(hv));
1538 HvHASKFLAGS_off(hv);
1543 mro_isa_changed_in(hv);
1544 HvEITER_set(hv, NULL);
1549 =for apidoc hv_clear_placeholders
1551 Clears any placeholders from a hash. If a restricted hash has any of its keys
1552 marked as readonly and the key is subsequently deleted, the key is not actually
1553 deleted but is marked by assigning it a value of &PL_sv_placeholder. This tags
1554 it so it will be ignored by future operations such as iterating over the hash,
1555 but will still allow the hash to have a value reassigned to the key at some
1556 future point. This function clears any such placeholder keys from the hash.
1557 See Hash::Util::lock_keys() for an example of its use.
1563 Perl_hv_clear_placeholders(pTHX_ HV *hv)
1566 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
1568 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
1571 clear_placeholders(hv, items);
1575 S_clear_placeholders(pTHX_ HV *hv, U32 items)
1580 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
1587 /* Loop down the linked list heads */
1589 HE **oentry = &(HvARRAY(hv))[i];
1592 while ((entry = *oentry)) {
1593 if (HeVAL(entry) == &PL_sv_placeholder) {
1594 *oentry = HeNEXT(entry);
1595 if (first && !*oentry)
1596 HvFILL(hv)--; /* This linked list is now empty. */
1597 if (entry == HvEITER_get(hv))
1600 hv_free_ent(hv, entry);
1604 HvTOTALKEYS(hv) -= (IV)HvPLACEHOLDERS_get(hv);
1605 if (HvKEYS(hv) == 0)
1606 HvHASKFLAGS_off(hv);
1607 HvPLACEHOLDERS_set(hv, 0);
1611 oentry = &HeNEXT(entry);
1616 /* You can't get here, hence assertion should always fail. */
1617 assert (items == 0);
1622 S_hfreeentries(pTHX_ HV *hv)
1624 /* This is the array that we're going to restore */
1625 HE **const orig_array = HvARRAY(hv);
1629 PERL_ARGS_ASSERT_HFREEENTRIES;
1635 /* If the hash is actually a symbol table with a name, look after the
1637 struct xpvhv_aux *iter = HvAUX(hv);
1639 name = iter->xhv_name;
1640 iter->xhv_name = NULL;
1645 /* orig_array remains unchanged throughout the loop. If after freeing all
1646 the entries it turns out that one of the little blighters has triggered
1647 an action that has caused HvARRAY to be re-allocated, then we set
1648 array to the new HvARRAY, and try again. */
1651 /* This is the one we're going to try to empty. First time round
1652 it's the original array. (Hopefully there will only be 1 time
1654 HE ** const array = HvARRAY(hv);
1657 /* Because we have taken xhv_name out, the only allocated pointer
1658 in the aux structure that might exist is the backreference array.
1663 struct mro_meta *meta;
1664 struct xpvhv_aux *iter = HvAUX(hv);
1665 /* If there are weak references to this HV, we need to avoid
1666 freeing them up here. In particular we need to keep the AV
1667 visible as what we're deleting might well have weak references
1668 back to this HV, so the for loop below may well trigger
1669 the removal of backreferences from this array. */
1671 if (iter->xhv_backreferences) {
1672 /* So donate them to regular backref magic to keep them safe.
1673 The sv_magic will increase the reference count of the AV,
1674 so we need to drop it first. */
1675 SvREFCNT_dec(iter->xhv_backreferences);
1676 if (AvFILLp(iter->xhv_backreferences) == -1) {
1677 /* Turns out that the array is empty. Just free it. */
1678 SvREFCNT_dec(iter->xhv_backreferences);
1681 sv_magic(MUTABLE_SV(hv),
1682 MUTABLE_SV(iter->xhv_backreferences),
1683 PERL_MAGIC_backref, NULL, 0);
1685 iter->xhv_backreferences = NULL;
1688 entry = iter->xhv_eiter; /* HvEITER(hv) */
1689 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1691 hv_free_ent(hv, entry);
1693 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1694 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1696 if((meta = iter->xhv_mro_meta)) {
1697 if (meta->mro_linear_all) {
1698 SvREFCNT_dec(MUTABLE_SV(meta->mro_linear_all));
1699 meta->mro_linear_all = NULL;
1700 /* This is just acting as a shortcut pointer. */
1701 meta->mro_linear_current = NULL;
1702 } else if (meta->mro_linear_current) {
1703 /* Only the current MRO is stored, so this owns the data.
1705 SvREFCNT_dec(meta->mro_linear_current);
1706 meta->mro_linear_current = NULL;
1708 if(meta->mro_nextmethod) SvREFCNT_dec(meta->mro_nextmethod);
1709 SvREFCNT_dec(meta->isa);
1711 iter->xhv_mro_meta = NULL;
1714 /* There are now no allocated pointers in the aux structure. */
1716 SvFLAGS(hv) &= ~SVf_OOK; /* Goodbye, aux structure. */
1717 /* What aux structure? */
1720 /* make everyone else think the array is empty, so that the destructors
1721 * called for freed entries can't recusively mess with us */
1724 ((XPVHV*) SvANY(hv))->xhv_keys = 0;
1728 /* Loop down the linked list heads */
1729 HE *entry = array[i];
1732 register HE * const oentry = entry;
1733 entry = HeNEXT(entry);
1734 hv_free_ent(hv, oentry);
1738 /* As there are no allocated pointers in the aux structure, it's now
1739 safe to free the array we just cleaned up, if it's not the one we're
1740 going to put back. */
1741 if (array != orig_array) {
1746 /* Good. No-one added anything this time round. */
1751 /* Someone attempted to iterate or set the hash name while we had
1752 the array set to 0. We'll catch backferences on the next time
1753 round the while loop. */
1754 assert(HvARRAY(hv));
1756 if (HvAUX(hv)->xhv_name) {
1757 unshare_hek_or_pvn(HvAUX(hv)->xhv_name, 0, 0, 0);
1761 if (--attempts == 0) {
1762 Perl_die(aTHX_ "panic: hfreeentries failed to free hash - something is repeatedly re-creating entries");
1766 HvARRAY(hv) = orig_array;
1768 /* If the hash was actually a symbol table, put the name back. */
1770 /* We have restored the original array. If name is non-NULL, then
1771 the original array had an aux structure at the end. So this is
1773 SvFLAGS(hv) |= SVf_OOK;
1774 HvAUX(hv)->xhv_name = name;
1779 =for apidoc hv_undef
1787 Perl_hv_undef(pTHX_ HV *hv)
1790 register XPVHV* xhv;
1795 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1796 xhv = (XPVHV*)SvANY(hv);
1798 if ((name = HvNAME_get(hv)) && !PL_dirty)
1799 mro_isa_changed_in(hv);
1804 (void)hv_delete(PL_stashcache, name, HvNAMELEN_get(hv), G_DISCARD);
1805 hv_name_set(hv, NULL, 0, 0);
1807 SvFLAGS(hv) &= ~SVf_OOK;
1808 Safefree(HvARRAY(hv));
1809 xhv->xhv_max = 7; /* HvMAX(hv) = 7 (it's a normal hash) */
1811 HvPLACEHOLDERS_set(hv, 0);
1814 mg_clear(MUTABLE_SV(hv));
1817 static struct xpvhv_aux*
1818 S_hv_auxinit(HV *hv) {
1819 struct xpvhv_aux *iter;
1822 PERL_ARGS_ASSERT_HV_AUXINIT;
1825 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
1826 + sizeof(struct xpvhv_aux), char);
1828 array = (char *) HvARRAY(hv);
1829 Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
1830 + sizeof(struct xpvhv_aux), char);
1832 HvARRAY(hv) = (HE**) array;
1833 /* SvOOK_on(hv) attacks the IV flags. */
1834 SvFLAGS(hv) |= SVf_OOK;
1837 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1838 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1840 iter->xhv_backreferences = 0;
1841 iter->xhv_mro_meta = NULL;
1846 =for apidoc hv_iterinit
1848 Prepares a starting point to traverse a hash table. Returns the number of
1849 keys in the hash (i.e. the same as C<HvKEYS(tb)>). The return value is
1850 currently only meaningful for hashes without tie magic.
1852 NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
1853 hash buckets that happen to be in use. If you still need that esoteric
1854 value, you can get it through the macro C<HvFILL(tb)>.
1861 Perl_hv_iterinit(pTHX_ HV *hv)
1863 PERL_ARGS_ASSERT_HV_ITERINIT;
1865 /* FIXME: Are we not NULL, or do we croak? Place bets now! */
1868 Perl_croak(aTHX_ "Bad hash");
1871 struct xpvhv_aux * const iter = HvAUX(hv);
1872 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
1873 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1875 hv_free_ent(hv, entry);
1877 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1878 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1883 /* used to be xhv->xhv_fill before 5.004_65 */
1884 return HvTOTALKEYS(hv);
1888 Perl_hv_riter_p(pTHX_ HV *hv) {
1889 struct xpvhv_aux *iter;
1891 PERL_ARGS_ASSERT_HV_RITER_P;
1894 Perl_croak(aTHX_ "Bad hash");
1896 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
1897 return &(iter->xhv_riter);
1901 Perl_hv_eiter_p(pTHX_ HV *hv) {
1902 struct xpvhv_aux *iter;
1904 PERL_ARGS_ASSERT_HV_EITER_P;
1907 Perl_croak(aTHX_ "Bad hash");
1909 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
1910 return &(iter->xhv_eiter);
1914 Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
1915 struct xpvhv_aux *iter;
1917 PERL_ARGS_ASSERT_HV_RITER_SET;
1920 Perl_croak(aTHX_ "Bad hash");
1928 iter = hv_auxinit(hv);
1930 iter->xhv_riter = riter;
1934 Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
1935 struct xpvhv_aux *iter;
1937 PERL_ARGS_ASSERT_HV_EITER_SET;
1940 Perl_croak(aTHX_ "Bad hash");
1945 /* 0 is the default so don't go malloc()ing a new structure just to
1950 iter = hv_auxinit(hv);
1952 iter->xhv_eiter = eiter;
1956 Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
1959 struct xpvhv_aux *iter;
1962 PERL_ARGS_ASSERT_HV_NAME_SET;
1963 PERL_UNUSED_ARG(flags);
1966 Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len);
1970 if (iter->xhv_name) {
1971 unshare_hek_or_pvn(iter->xhv_name, 0, 0, 0);
1977 iter = hv_auxinit(hv);
1979 PERL_HASH(hash, name, len);
1980 iter->xhv_name = name ? share_hek(name, len, hash) : NULL;
1984 Perl_hv_backreferences_p(pTHX_ HV *hv) {
1985 struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
1987 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
1988 PERL_UNUSED_CONTEXT;
1990 return &(iter->xhv_backreferences);
1994 Perl_hv_kill_backrefs(pTHX_ HV *hv) {
1997 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2002 av = HvAUX(hv)->xhv_backreferences;
2005 HvAUX(hv)->xhv_backreferences = 0;
2006 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
2012 hv_iternext is implemented as a macro in hv.h
2014 =for apidoc hv_iternext
2016 Returns entries from a hash iterator. See C<hv_iterinit>.
2018 You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2019 iterator currently points to, without losing your place or invalidating your
2020 iterator. Note that in this case the current entry is deleted from the hash
2021 with your iterator holding the last reference to it. Your iterator is flagged
2022 to free the entry on the next call to C<hv_iternext>, so you must not discard
2023 your iterator immediately else the entry will leak - call C<hv_iternext> to
2024 trigger the resource deallocation.
2026 =for apidoc hv_iternext_flags
2028 Returns entries from a hash iterator. See C<hv_iterinit> and C<hv_iternext>.
2029 The C<flags> value will normally be zero; if HV_ITERNEXT_WANTPLACEHOLDERS is
2030 set the placeholders keys (for restricted hashes) will be returned in addition
2031 to normal keys. By default placeholders are automatically skipped over.
2032 Currently a placeholder is implemented with a value that is
2033 C<&Perl_sv_placeholder>. Note that the implementation of placeholders and
2034 restricted hashes may change, and the implementation currently is
2035 insufficiently abstracted for any change to be tidy.
2041 Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2044 register XPVHV* xhv;
2048 struct xpvhv_aux *iter;
2050 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2053 Perl_croak(aTHX_ "Bad hash");
2055 xhv = (XPVHV*)SvANY(hv);
2058 /* Too many things (well, pp_each at least) merrily assume that you can
2059 call iv_iternext without calling hv_iterinit, so we'll have to deal
2065 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2066 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
2067 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
2068 SV * const key = sv_newmortal();
2070 sv_setsv(key, HeSVKEY_force(entry));
2071 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
2077 /* one HE per MAGICAL hash */
2078 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
2080 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
2082 HeKEY_hek(entry) = hek;
2083 HeKLEN(entry) = HEf_SVKEY;
2085 magic_nextpack(MUTABLE_SV(hv),mg,key);
2087 /* force key to stay around until next time */
2088 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2089 return entry; /* beware, hent_val is not set */
2092 SvREFCNT_dec(HeVAL(entry));
2093 Safefree(HeKEY_hek(entry));
2095 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2099 #if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */
2100 if (!entry && SvRMAGICAL((const SV *)hv)
2101 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
2104 /* The prime_env_iter() on VMS just loaded up new hash values
2105 * so the iteration count needs to be reset back to the beginning
2109 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2114 /* hv_iterint now ensures this. */
2115 assert (HvARRAY(hv));
2117 /* At start of hash, entry is NULL. */
2120 entry = HeNEXT(entry);
2121 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2123 * Skip past any placeholders -- don't want to include them in
2126 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
2127 entry = HeNEXT(entry);
2132 /* OK. Come to the end of the current list. Grab the next one. */
2134 iter->xhv_riter++; /* HvRITER(hv)++ */
2135 if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) {
2136 /* There is no next one. End of the hash. */
2137 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2140 entry = (HvARRAY(hv))[iter->xhv_riter];
2142 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2143 /* If we have an entry, but it's a placeholder, don't count it.
2145 while (entry && HeVAL(entry) == &PL_sv_placeholder)
2146 entry = HeNEXT(entry);
2148 /* Will loop again if this linked list starts NULL
2149 (for HV_ITERNEXT_WANTPLACEHOLDERS)
2150 or if we run through it and find only placeholders. */
2153 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2155 hv_free_ent(hv, oldentry);
2158 /*if (HvREHASH(hv) && entry && !HeKREHASH(entry))
2159 PerlIO_printf(PerlIO_stderr(), "Awooga %p %p\n", (void*)hv, (void*)entry);*/
2161 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
2166 =for apidoc hv_iterkey
2168 Returns the key from the current position of the hash iterator. See
2175 Perl_hv_iterkey(pTHX_ register HE *entry, I32 *retlen)
2177 PERL_ARGS_ASSERT_HV_ITERKEY;
2179 if (HeKLEN(entry) == HEf_SVKEY) {
2181 char * const p = SvPV(HeKEY_sv(entry), len);
2186 *retlen = HeKLEN(entry);
2187 return HeKEY(entry);
2191 /* unlike hv_iterval(), this always returns a mortal copy of the key */
2193 =for apidoc hv_iterkeysv
2195 Returns the key as an C<SV*> from the current position of the hash
2196 iterator. The return value will always be a mortal copy of the key. Also
2203 Perl_hv_iterkeysv(pTHX_ register HE *entry)
2205 PERL_ARGS_ASSERT_HV_ITERKEYSV;
2207 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
2211 =for apidoc hv_iterval
2213 Returns the value from the current position of the hash iterator. See
2220 Perl_hv_iterval(pTHX_ HV *hv, register HE *entry)
2222 PERL_ARGS_ASSERT_HV_ITERVAL;
2224 if (SvRMAGICAL(hv)) {
2225 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
2226 SV* const sv = sv_newmortal();
2227 if (HeKLEN(entry) == HEf_SVKEY)
2228 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
2230 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
2234 return HeVAL(entry);
2238 =for apidoc hv_iternextsv
2240 Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
2247 Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
2249 HE * const he = hv_iternext_flags(hv, 0);
2251 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
2255 *key = hv_iterkey(he, retlen);
2256 return hv_iterval(hv, he);
2263 =for apidoc hv_magic
2265 Adds magic to a hash. See C<sv_magic>.
2270 /* possibly free a shared string if no one has access to it
2271 * len and hash must both be valid for str.
2274 Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
2276 unshare_hek_or_pvn (NULL, str, len, hash);
2281 Perl_unshare_hek(pTHX_ HEK *hek)
2284 unshare_hek_or_pvn(hek, NULL, 0, 0);
2287 /* possibly free a shared string if no one has access to it
2288 hek if non-NULL takes priority over the other 3, else str, len and hash
2289 are used. If so, len and hash must both be valid for str.
2292 S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
2295 register XPVHV* xhv;
2297 register HE **oentry;
2299 bool is_utf8 = FALSE;
2301 const char * const save = str;
2302 struct shared_he *he = NULL;
2305 /* Find the shared he which is just before us in memory. */
2306 he = (struct shared_he *)(((char *)hek)
2307 - STRUCT_OFFSET(struct shared_he,
2310 /* Assert that the caller passed us a genuine (or at least consistent)
2312 assert (he->shared_he_he.hent_hek == hek);
2315 if (he->shared_he_he.he_valu.hent_refcount - 1) {
2316 --he->shared_he_he.he_valu.hent_refcount;
2317 UNLOCK_STRTAB_MUTEX;
2320 UNLOCK_STRTAB_MUTEX;
2322 hash = HEK_HASH(hek);
2323 } else if (len < 0) {
2324 STRLEN tmplen = -len;
2326 /* See the note in hv_fetch(). --jhi */
2327 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2330 k_flags = HVhek_UTF8;
2332 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2335 /* what follows was the moral equivalent of:
2336 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
2338 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
2340 xhv = (XPVHV*)SvANY(PL_strtab);
2341 /* assert(xhv_array != 0) */
2343 first = oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
2345 const HE *const he_he = &(he->shared_he_he);
2346 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
2351 const int flags_masked = k_flags & HVhek_MASK;
2352 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
2353 if (HeHASH(entry) != hash) /* strings can't be equal */
2355 if (HeKLEN(entry) != len)
2357 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
2359 if (HeKFLAGS(entry) != flags_masked)
2366 if (--entry->he_valu.hent_refcount == 0) {
2367 *oentry = HeNEXT(entry);
2369 /* There are now no entries in our slot. */
2370 xhv->xhv_fill--; /* HvFILL(hv)-- */
2373 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
2377 UNLOCK_STRTAB_MUTEX;
2378 if (!entry && ckWARN_d(WARN_INTERNAL))
2379 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
2380 "Attempt to free non-existent shared string '%s'%s"
2382 hek ? HEK_KEY(hek) : str,
2383 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
2384 if (k_flags & HVhek_FREEKEY)
2388 /* get a (constant) string ptr from the global string table
2389 * string will get added if it is not already there.
2390 * len and hash must both be valid for str.
2393 Perl_share_hek(pTHX_ const char *str, I32 len, register U32 hash)
2395 bool is_utf8 = FALSE;
2397 const char * const save = str;
2399 PERL_ARGS_ASSERT_SHARE_HEK;
2402 STRLEN tmplen = -len;
2404 /* See the note in hv_fetch(). --jhi */
2405 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2407 /* If we were able to downgrade here, then than means that we were passed
2408 in a key which only had chars 0-255, but was utf8 encoded. */
2411 /* If we found we were able to downgrade the string to bytes, then
2412 we should flag that it needs upgrading on keys or each. Also flag
2413 that we need share_hek_flags to free the string. */
2415 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2418 return share_hek_flags (str, len, hash, flags);
2422 S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags)
2426 const int flags_masked = flags & HVhek_MASK;
2427 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
2428 register XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
2430 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
2432 /* what follows is the moral equivalent of:
2434 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
2435 hv_store(PL_strtab, str, len, NULL, hash);
2437 Can't rehash the shared string table, so not sure if it's worth
2438 counting the number of entries in the linked list
2441 /* assert(xhv_array != 0) */
2443 entry = (HvARRAY(PL_strtab))[hindex];
2444 for (;entry; entry = HeNEXT(entry)) {
2445 if (HeHASH(entry) != hash) /* strings can't be equal */
2447 if (HeKLEN(entry) != len)
2449 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
2451 if (HeKFLAGS(entry) != flags_masked)
2457 /* What used to be head of the list.
2458 If this is NULL, then we're the first entry for this slot, which
2459 means we need to increate fill. */
2460 struct shared_he *new_entry;
2463 HE **const head = &HvARRAY(PL_strtab)[hindex];
2464 HE *const next = *head;
2466 /* We don't actually store a HE from the arena and a regular HEK.
2467 Instead we allocate one chunk of memory big enough for both,
2468 and put the HEK straight after the HE. This way we can find the
2469 HEK directly from the HE.
2472 Newx(k, STRUCT_OFFSET(struct shared_he,
2473 shared_he_hek.hek_key[0]) + len + 2, char);
2474 new_entry = (struct shared_he *)k;
2475 entry = &(new_entry->shared_he_he);
2476 hek = &(new_entry->shared_he_hek);
2478 Copy(str, HEK_KEY(hek), len, char);
2479 HEK_KEY(hek)[len] = 0;
2481 HEK_HASH(hek) = hash;
2482 HEK_FLAGS(hek) = (unsigned char)flags_masked;
2484 /* Still "point" to the HEK, so that other code need not know what
2486 HeKEY_hek(entry) = hek;
2487 entry->he_valu.hent_refcount = 0;
2488 HeNEXT(entry) = next;
2491 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
2492 if (!next) { /* initial entry? */
2493 xhv->xhv_fill++; /* HvFILL(hv)++ */
2494 } else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
2499 ++entry->he_valu.hent_refcount;
2500 UNLOCK_STRTAB_MUTEX;
2502 if (flags & HVhek_FREEKEY)
2505 return HeKEY_hek(entry);
2509 Perl_hv_placeholders_p(pTHX_ HV *hv)
2512 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2514 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
2517 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
2520 Perl_die(aTHX_ "panic: hv_placeholders_p");
2523 return &(mg->mg_len);
2528 Perl_hv_placeholders_get(pTHX_ const HV *hv)
2531 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2533 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
2535 return mg ? mg->mg_len : 0;
2539 Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
2542 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2544 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
2549 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
2550 Perl_die(aTHX_ "panic: hv_placeholders_set");
2552 /* else we don't need to add magic to record 0 placeholders. */
2556 S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
2561 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
2563 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
2568 value = &PL_sv_placeholder;
2571 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
2574 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
2577 case HVrhek_PV_UTF8:
2578 /* Create a string SV that directly points to the bytes in our
2580 value = newSV_type(SVt_PV);
2581 SvPV_set(value, (char *) he->refcounted_he_data + 1);
2582 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
2583 /* This stops anything trying to free it */
2584 SvLEN_set(value, 0);
2586 SvREADONLY_on(value);
2587 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
2591 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %x",
2592 he->refcounted_he_data[0]);
2598 =for apidoc refcounted_he_chain_2hv
2600 Generates and returns a C<HV *> by walking up the tree starting at the passed
2601 in C<struct refcounted_he *>.
2606 Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain)
2610 U32 placeholders = 0;
2611 /* We could chase the chain once to get an idea of the number of keys,
2612 and call ksplit. But for now we'll make a potentially inefficient
2613 hash with only 8 entries in its array. */
2614 const U32 max = HvMAX(hv);
2618 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
2619 HvARRAY(hv) = (HE**)array;
2624 U32 hash = chain->refcounted_he_hash;
2626 U32 hash = HEK_HASH(chain->refcounted_he_hek);
2628 HE **oentry = &((HvARRAY(hv))[hash & max]);
2629 HE *entry = *oentry;
2632 for (; entry; entry = HeNEXT(entry)) {
2633 if (HeHASH(entry) == hash) {
2634 /* We might have a duplicate key here. If so, entry is older
2635 than the key we've already put in the hash, so if they are
2636 the same, skip adding entry. */
2638 const STRLEN klen = HeKLEN(entry);
2639 const char *const key = HeKEY(entry);
2640 if (klen == chain->refcounted_he_keylen
2641 && (!!HeKUTF8(entry)
2642 == !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2643 && memEQ(key, REF_HE_KEY(chain), klen))
2646 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
2648 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
2649 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
2650 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
2661 = share_hek_flags(REF_HE_KEY(chain),
2662 chain->refcounted_he_keylen,
2663 chain->refcounted_he_hash,
2664 (chain->refcounted_he_data[0]
2665 & (HVhek_UTF8|HVhek_WASUTF8)));
2667 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
2669 value = refcounted_he_value(chain);
2670 if (value == &PL_sv_placeholder)
2672 HeVAL(entry) = value;
2674 /* Link it into the chain. */
2675 HeNEXT(entry) = *oentry;
2676 if (!HeNEXT(entry)) {
2677 /* initial entry. */
2685 chain = chain->refcounted_he_next;
2689 clear_placeholders(hv, placeholders);
2690 HvTOTALKEYS(hv) -= placeholders;
2693 /* We could check in the loop to see if we encounter any keys with key
2694 flags, but it's probably not worth it, as this per-hash flag is only
2695 really meant as an optimisation for things like Storable. */
2697 DEBUG_A(Perl_hv_assert(aTHX_ hv));
2703 Perl_refcounted_he_fetch(pTHX_ const struct refcounted_he *chain, SV *keysv,
2704 const char *key, STRLEN klen, int flags, U32 hash)
2707 /* Just to be awkward, if you're using this interface the UTF-8-or-not-ness
2708 of your key has to exactly match that which is stored. */
2709 SV *value = &PL_sv_placeholder;
2712 /* No point in doing any of this if there's nothing to find. */
2716 if (flags & HVhek_FREEKEY)
2718 key = SvPV_const(keysv, klen);
2720 is_utf8 = (SvUTF8(keysv) != 0);
2722 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
2726 if (keysv && (SvIsCOW_shared_hash(keysv))) {
2727 hash = SvSHARED_HASH(keysv);
2729 PERL_HASH(hash, key, klen);
2733 for (; chain; chain = chain->refcounted_he_next) {
2735 if (hash != chain->refcounted_he_hash)
2737 if (klen != chain->refcounted_he_keylen)
2739 if (memNE(REF_HE_KEY(chain),key,klen))
2741 if (!!is_utf8 != !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2744 if (hash != HEK_HASH(chain->refcounted_he_hek))
2746 if (klen != (STRLEN)HEK_LEN(chain->refcounted_he_hek))
2748 if (memNE(HEK_KEY(chain->refcounted_he_hek),key,klen))
2750 if (!!is_utf8 != !!HEK_UTF8(chain->refcounted_he_hek))
2754 value = sv_2mortal(refcounted_he_value(chain));
2759 if (flags & HVhek_FREEKEY)
2766 =for apidoc refcounted_he_new
2768 Creates a new C<struct refcounted_he>. As S<key> is copied, and value is
2769 stored in a compact form, all references remain the property of the caller.
2770 The C<struct refcounted_he> is returned with a reference count of 1.
2775 struct refcounted_he *
2776 Perl_refcounted_he_new(pTHX_ struct refcounted_he *const parent,
2777 SV *const key, SV *const value) {
2780 const char *key_p = SvPV_const(key, key_len);
2781 STRLEN value_len = 0;
2782 const char *value_p = NULL;
2785 bool is_utf8 = SvUTF8(key) ? TRUE : FALSE;
2788 value_type = HVrhek_PV;
2789 } else if (SvIOK(value)) {
2790 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
2791 } else if (value == &PL_sv_placeholder) {
2792 value_type = HVrhek_delete;
2793 } else if (!SvOK(value)) {
2794 value_type = HVrhek_undef;
2796 value_type = HVrhek_PV;
2799 if (value_type == HVrhek_PV) {
2800 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
2801 the value is overloaded, and doesn't yet have the UTF-8flag set. */
2802 value_p = SvPV_const(value, value_len);
2804 value_type = HVrhek_PV_UTF8;
2809 /* Hash keys are always stored normalised to (yes) ISO-8859-1.
2810 As we're going to be building hash keys from this value in future,
2811 normalise it now. */
2812 key_p = (char*)bytes_from_utf8((const U8*)key_p, &key_len, &is_utf8);
2813 flags |= is_utf8 ? HVhek_UTF8 : HVhek_WASUTF8;
2816 return refcounted_he_new_common(parent, key_p, key_len, flags, value_type,
2817 ((value_type == HVrhek_PV
2818 || value_type == HVrhek_PV_UTF8) ?
2819 (void *)value_p : (void *)value),
2823 static struct refcounted_he *
2824 S_refcounted_he_new_common(pTHX_ struct refcounted_he *const parent,
2825 const char *const key_p, const STRLEN key_len,
2826 const char flags, char value_type,
2827 const void *value, const STRLEN value_len) {
2829 struct refcounted_he *he;
2831 const bool is_pv = value_type == HVrhek_PV || value_type == HVrhek_PV_UTF8;
2832 STRLEN key_offset = is_pv ? value_len + 2 : 1;
2834 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_COMMON;
2837 he = (struct refcounted_he*)
2838 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2842 he = (struct refcounted_he*)
2843 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2847 he->refcounted_he_next = parent;
2850 Copy((char *)value, he->refcounted_he_data + 1, value_len + 1, char);
2851 he->refcounted_he_val.refcounted_he_u_len = value_len;
2852 } else if (value_type == HVrhek_IV) {
2853 he->refcounted_he_val.refcounted_he_u_iv = SvIVX((const SV *)value);
2854 } else if (value_type == HVrhek_UV) {
2855 he->refcounted_he_val.refcounted_he_u_uv = SvUVX((const SV *)value);
2858 PERL_HASH(hash, key_p, key_len);
2861 he->refcounted_he_hash = hash;
2862 he->refcounted_he_keylen = key_len;
2863 Copy(key_p, he->refcounted_he_data + key_offset, key_len, char);
2865 he->refcounted_he_hek = share_hek_flags(key_p, key_len, hash, flags);
2868 if (flags & HVhek_WASUTF8) {
2869 /* If it was downgraded from UTF-8, then the pointer returned from
2870 bytes_from_utf8 is an allocated pointer that we must free. */
2874 he->refcounted_he_data[0] = flags;
2875 he->refcounted_he_refcnt = 1;
2881 =for apidoc refcounted_he_free
2883 Decrements the reference count of the passed in C<struct refcounted_he *>
2884 by one. If the reference count reaches zero the structure's memory is freed,
2885 and C<refcounted_he_free> iterates onto the parent node.
2891 Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
2893 PERL_UNUSED_CONTEXT;
2896 struct refcounted_he *copy;
2900 new_count = --he->refcounted_he_refcnt;
2901 HINTS_REFCNT_UNLOCK;
2907 #ifndef USE_ITHREADS
2908 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
2911 he = he->refcounted_he_next;
2912 PerlMemShared_free(copy);
2917 Perl_fetch_cop_label(pTHX_ struct refcounted_he *const chain, STRLEN *len,
2922 if (chain->refcounted_he_keylen != 1)
2924 if (*REF_HE_KEY(chain) != ':')
2927 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
2929 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
2932 /* Stop anyone trying to really mess us up by adding their own value for
2934 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
2935 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
2939 *len = chain->refcounted_he_val.refcounted_he_u_len;
2941 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
2942 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
2944 return chain->refcounted_he_data + 1;
2947 /* As newSTATEOP currently gets passed plain char* labels, we will only provide
2948 that interface. Once it works out how to pass in length and UTF-8 ness, this
2949 function will need superseding. */
2950 struct refcounted_he *
2951 Perl_store_cop_label(pTHX_ struct refcounted_he *const chain, const char *label)
2953 PERL_ARGS_ASSERT_STORE_COP_LABEL;
2955 return refcounted_he_new_common(chain, ":", 1, HVrhek_PV, HVrhek_PV,
2956 label, strlen(label));
2960 =for apidoc hv_assert
2962 Check that a hash is in an internally consistent state.
2970 Perl_hv_assert(pTHX_ HV *hv)
2975 int placeholders = 0;
2978 const I32 riter = HvRITER_get(hv);
2979 HE *eiter = HvEITER_get(hv);
2981 PERL_ARGS_ASSERT_HV_ASSERT;
2983 (void)hv_iterinit(hv);
2985 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
2986 /* sanity check the values */
2987 if (HeVAL(entry) == &PL_sv_placeholder)
2991 /* sanity check the keys */
2992 if (HeSVKEY(entry)) {
2993 NOOP; /* Don't know what to check on SV keys. */
2994 } else if (HeKUTF8(entry)) {
2996 if (HeKWASUTF8(entry)) {
2997 PerlIO_printf(Perl_debug_log,
2998 "hash key has both WASUTF8 and UTF8: '%.*s'\n",
2999 (int) HeKLEN(entry), HeKEY(entry));
3002 } else if (HeKWASUTF8(entry))
3005 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
3006 static const char bad_count[] = "Count %d %s(s), but hash reports %d\n";
3007 const int nhashkeys = HvUSEDKEYS(hv);
3008 const int nhashplaceholders = HvPLACEHOLDERS_get(hv);
3010 if (nhashkeys != real) {
3011 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
3014 if (nhashplaceholders != placeholders) {
3015 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
3019 if (withflags && ! HvHASKFLAGS(hv)) {
3020 PerlIO_printf(Perl_debug_log,
3021 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
3026 sv_dump(MUTABLE_SV(hv));
3028 HvRITER_set(hv, riter); /* Restore hash iterator state */
3029 HvEITER_set(hv, eiter);
3036 * c-indentation-style: bsd
3038 * indent-tabs-mode: t
3041 * ex: set ts=8 sts=4 sw=4 noet: