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
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.
34 #define PERL_HASH_INTERNAL_ACCESS
37 /* we split when we collide and we have a load factor over 0.667.
38 * NOTE if you change this formula so we split earlier than previously
39 * you MUST change the logic in hv_ksplit()
41 #define DO_HSPLIT(xhv) ( ((xhv)->xhv_keys + ((xhv)->xhv_keys >> 1)) > (xhv)->xhv_max )
42 #define HV_FILL_THRESHOLD 31
44 static const char S_strtab_error[]
45 = "Cannot modify shared string table in hv_%s";
49 #define new_HE() (HE*)safemalloc(sizeof(HE))
50 #define del_HE(p) safefree((char*)p)
58 void ** const root = &PL_body_roots[HE_SVSLOT];
61 Perl_more_bodies(aTHX_ HE_SVSLOT, sizeof(HE), PERL_ARENA_SIZE);
68 #define new_HE() new_he()
71 HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \
72 PL_body_roots[HE_SVSLOT] = p; \
80 S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
82 const int flags_masked = flags & HVhek_MASK;
86 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
88 Newx(k, HEK_BASESIZE + len + 2, char);
90 Copy(str, HEK_KEY(hek), len, char);
91 HEK_KEY(hek)[len] = 0;
94 HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED;
96 if (flags & HVhek_FREEKEY)
101 /* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
105 Perl_free_tied_hv_pool(pTHX)
107 HE *he = PL_hv_fetch_ent_mh;
110 Safefree(HeKEY_hek(he));
114 PL_hv_fetch_ent_mh = NULL;
117 #if defined(USE_ITHREADS)
119 Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
123 PERL_ARGS_ASSERT_HEK_DUP;
124 PERL_UNUSED_ARG(param);
129 shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
131 /* We already shared this hash key. */
132 (void)share_hek_hek(shared);
136 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
137 HEK_HASH(source), HEK_FLAGS(source));
138 ptr_table_store(PL_ptr_table, source, shared);
144 Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
148 PERL_ARGS_ASSERT_HE_DUP;
152 /* look for it in the table first */
153 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
157 /* create anew and remember what it is */
159 ptr_table_store(PL_ptr_table, e, ret);
161 HeNEXT(ret) = he_dup(HeNEXT(e),shared, param);
162 if (HeKLEN(e) == HEf_SVKEY) {
164 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
165 HeKEY_hek(ret) = (HEK*)k;
166 HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param);
169 /* This is hek_dup inlined, which seems to be important for speed
171 HEK * const source = HeKEY_hek(e);
172 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
175 /* We already shared this hash key. */
176 (void)share_hek_hek(shared);
180 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
181 HEK_HASH(source), HEK_FLAGS(source));
182 ptr_table_store(PL_ptr_table, source, shared);
184 HeKEY_hek(ret) = shared;
187 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
189 HeVAL(ret) = sv_dup_inc(HeVAL(e), param);
192 #endif /* USE_ITHREADS */
195 S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
198 SV * const sv = sv_newmortal();
200 PERL_ARGS_ASSERT_HV_NOTALLOWED;
202 if (!(flags & HVhek_FREEKEY)) {
203 sv_setpvn(sv, key, klen);
206 /* Need to free saved eventually assign to mortal SV */
207 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
208 sv_usepvn(sv, (char *) key, klen);
210 if (flags & HVhek_UTF8) {
213 Perl_croak(aTHX_ msg, SVfARG(sv));
216 /* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
222 Stores an SV in a hash. The hash key is specified as C<key> and the
223 absolute value of C<klen> is the length of the key. If C<klen> is
224 negative the key is assumed to be in UTF-8-encoded Unicode. The
225 C<hash> parameter is the precomputed hash value; if it is zero then
226 Perl will compute it.
228 The return value will be
229 C<NULL> if the operation failed or if the value did not need to be actually
230 stored within the hash (as in the case of tied hashes). Otherwise it can
231 be dereferenced to get the original C<SV*>. Note that the caller is
232 responsible for suitably incrementing the reference count of C<val> before
233 the call, and decrementing it if the function returned C<NULL>. Effectively
234 a successful C<hv_store> takes ownership of one reference to C<val>. This is
235 usually what you want; a newly created SV has a reference count of one, so
236 if all your code does is create SVs then store them in a hash, C<hv_store>
237 will own the only reference to the new SV, and your code doesn't need to do
238 anything further to tidy up. C<hv_store> is not implemented as a call to
239 C<hv_store_ent>, and does not create a temporary SV for the key, so if your
240 key data is not already in SV form then use C<hv_store> in preference to
243 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
244 information on how to use this function on tied hashes.
246 =for apidoc hv_store_ent
248 Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
249 parameter is the precomputed hash value; if it is zero then Perl will
250 compute it. The return value is the new hash entry so created. It will be
251 C<NULL> if the operation failed or if the value did not need to be actually
252 stored within the hash (as in the case of tied hashes). Otherwise the
253 contents of the return value can be accessed using the C<He?> macros
254 described here. Note that the caller is responsible for suitably
255 incrementing the reference count of C<val> before the call, and
256 decrementing it if the function returned NULL. Effectively a successful
257 C<hv_store_ent> takes ownership of one reference to C<val>. This is
258 usually what you want; a newly created SV has a reference count of one, so
259 if all your code does is create SVs then store them in a hash, C<hv_store>
260 will own the only reference to the new SV, and your code doesn't need to do
261 anything further to tidy up. Note that C<hv_store_ent> only reads the C<key>;
262 unlike C<val> it does not take ownership of it, so maintaining the correct
263 reference count on C<key> is entirely the caller's responsibility. C<hv_store>
264 is not implemented as a call to C<hv_store_ent>, and does not create a temporary
265 SV for the key, so if your key data is not already in SV form then use
266 C<hv_store> in preference to C<hv_store_ent>.
268 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
269 information on how to use this function on tied hashes.
271 =for apidoc hv_exists
273 Returns a boolean indicating whether the specified hash key exists. The
274 absolute value of C<klen> is the length of the key. If C<klen> is
275 negative the key is assumed to be in UTF-8-encoded Unicode.
279 Returns the SV which corresponds to the specified key in the hash.
280 The absolute value of C<klen> is the length of the key. If C<klen> is
281 negative the key is assumed to be in UTF-8-encoded Unicode. If
282 C<lval> is set then the fetch will be part of a store. This means that if
283 there is no value in the hash associated with the given key, then one is
284 created and a pointer to it is returned. The C<SV*> it points to can be
285 assigned to. But always check that the
286 return value is non-null before dereferencing it to an C<SV*>.
288 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
289 information on how to use this function on tied hashes.
291 =for apidoc hv_exists_ent
293 Returns a boolean indicating whether
294 the specified hash key exists. C<hash>
295 can be a valid precomputed hash value, or 0 to ask for it to be
301 /* returns an HE * structure with the all fields set */
302 /* note that hent_val will be a mortal sv for MAGICAL hashes */
304 =for apidoc hv_fetch_ent
306 Returns the hash entry which corresponds to the specified key in the hash.
307 C<hash> must be a valid precomputed hash number for the given C<key>, or 0
308 if you want the function to compute it. IF C<lval> is set then the fetch
309 will be part of a store. Make sure the return value is non-null before
310 accessing it. The return value when C<hv> is a tied hash is a pointer to a
311 static location, so be sure to make a copy of the structure if you need to
314 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
315 information on how to use this function on tied hashes.
320 /* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
322 Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
323 const int action, SV *val, const U32 hash)
328 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
337 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
341 Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
342 int flags, int action, SV *val, U32 hash)
352 const int return_svp = action & HV_FETCH_JUST_SV;
353 HEK *keysv_hek = NULL;
357 if (SvTYPE(hv) == (svtype)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);
391 is_utf8 = (SvUTF8(keysv) != 0);
392 if (SvIsCOW_shared_hash(keysv)) {
393 flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0);
395 flags = is_utf8 ? HVhek_UTF8 : 0;
398 is_utf8 = cBOOL(flags & HVhek_UTF8);
401 if (action & HV_DELETE) {
402 return (void *) hv_delete_common(hv, keysv, key, klen,
403 flags, action, hash);
406 xhv = (XPVHV*)SvANY(hv);
408 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
409 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
410 || SvGMAGICAL((const SV *)hv))
412 /* FIXME should be able to skimp on the HE/HEK here when
413 HV_FETCH_JUST_SV is true. */
415 keysv = newSVpvn_utf8(key, klen, is_utf8);
417 keysv = newSVsv(keysv);
420 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
422 /* grab a fake HE/HEK pair from the pool or make a new one */
423 entry = PL_hv_fetch_ent_mh;
425 PL_hv_fetch_ent_mh = HeNEXT(entry);
429 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
430 HeKEY_hek(entry) = (HEK*)k;
432 HeNEXT(entry) = NULL;
433 HeSVKEY_set(entry, keysv);
435 sv_upgrade(sv, SVt_PVLV);
437 /* so we can free entry when freeing sv */
438 LvTARG(sv) = MUTABLE_SV(entry);
440 /* XXX remove at some point? */
441 if (flags & HVhek_FREEKEY)
445 return entry ? (void *) &HeVAL(entry) : NULL;
447 return (void *) entry;
449 #ifdef ENV_IS_CASELESS
450 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
452 for (i = 0; i < klen; ++i)
453 if (isLOWER(key[i])) {
454 /* Would be nice if we had a routine to do the
455 copy and upercase in a single pass through. */
456 const char * const nkey = strupr(savepvn(key,klen));
457 /* Note that this fetch is for nkey (the uppercased
458 key) whereas the store is for key (the original) */
459 void *result = hv_common(hv, NULL, nkey, klen,
460 HVhek_FREEKEY, /* free nkey */
461 0 /* non-LVAL fetch */
462 | HV_DISABLE_UVAR_XKEY
465 0 /* compute hash */);
466 if (!result && (action & HV_FETCH_LVALUE)) {
467 /* This call will free key if necessary.
468 Do it this way to encourage compiler to tail
470 result = hv_common(hv, keysv, key, klen, flags,
472 | HV_DISABLE_UVAR_XKEY
476 if (flags & HVhek_FREEKEY)
484 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
485 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
486 || SvGMAGICAL((const SV *)hv)) {
487 /* I don't understand why hv_exists_ent has svret and sv,
488 whereas hv_exists only had one. */
489 SV * const svret = sv_newmortal();
492 if (keysv || is_utf8) {
494 keysv = newSVpvn_utf8(key, klen, TRUE);
496 keysv = newSVsv(keysv);
498 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
500 mg_copy(MUTABLE_SV(hv), sv, key, klen);
502 if (flags & HVhek_FREEKEY)
505 MAGIC * const mg = mg_find(sv, PERL_MAGIC_tiedelem);
507 magic_existspack(svret, mg);
509 /* This cast somewhat evil, but I'm merely using NULL/
510 not NULL to return the boolean exists.
511 And I know hv is not NULL. */
512 return SvTRUE_NN(svret) ? (void *)hv : NULL;
514 #ifdef ENV_IS_CASELESS
515 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
516 /* XXX This code isn't UTF8 clean. */
517 char * const keysave = (char * const)key;
518 /* Will need to free this, so set FREEKEY flag. */
519 key = savepvn(key,klen);
520 key = (const char*)strupr((char*)key);
525 if (flags & HVhek_FREEKEY) {
528 flags |= HVhek_FREEKEY;
532 else if (action & HV_FETCH_ISSTORE) {
535 hv_magic_check (hv, &needs_copy, &needs_store);
537 const bool save_taint = TAINT_get;
538 if (keysv || is_utf8) {
540 keysv = newSVpvn_utf8(key, klen, TRUE);
543 TAINT_set(SvTAINTED(keysv));
544 keysv = sv_2mortal(newSVsv(keysv));
545 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
547 mg_copy(MUTABLE_SV(hv), val, key, klen);
550 TAINT_IF(save_taint);
551 #ifdef NO_TAINT_SUPPORT
552 PERL_UNUSED_VAR(save_taint);
555 if (flags & HVhek_FREEKEY)
559 #ifdef ENV_IS_CASELESS
560 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
561 /* XXX This code isn't UTF8 clean. */
562 const char *keysave = key;
563 /* Will need to free this, so set FREEKEY flag. */
564 key = savepvn(key,klen);
565 key = (const char*)strupr((char*)key);
570 if (flags & HVhek_FREEKEY) {
573 flags |= HVhek_FREEKEY;
581 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
582 #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
583 || (SvRMAGICAL((const SV *)hv)
584 && mg_find((const SV *)hv, PERL_MAGIC_env))
589 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
591 HvARRAY(hv) = (HE**)array;
593 #ifdef DYNAMIC_ENV_FETCH
594 else if (action & HV_FETCH_ISEXISTS) {
595 /* for an %ENV exists, if we do an insert it's by a recursive
596 store call, so avoid creating HvARRAY(hv) right now. */
600 /* XXX remove at some point? */
601 if (flags & HVhek_FREEKEY)
608 if (is_utf8 && !(flags & HVhek_KEYCANONICAL)) {
609 char * const keysave = (char *)key;
610 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
614 flags &= ~HVhek_UTF8;
615 if (key != keysave) {
616 if (flags & HVhek_FREEKEY)
618 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
619 /* If the caller calculated a hash, it was on the sequence of
620 octets that are the UTF-8 form. We've now changed the sequence
621 of octets stored to that of the equivalent byte representation,
622 so the hash we need is different. */
627 if (keysv && (SvIsCOW_shared_hash(keysv))) {
629 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
630 hash = SvSHARED_HASH(keysv);
633 PERL_HASH(hash, key, klen);
635 masked_flags = (flags & HVhek_MASK);
637 #ifdef DYNAMIC_ENV_FETCH
638 if (!HvARRAY(hv)) entry = NULL;
642 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
649 /* keysv is actually a HEK in disguise, so we can match just by
650 * comparing the HEK pointers in the HE chain. There is a slight
651 * caveat: on something like "\x80", which has both plain and utf8
652 * representations, perl's hashes do encoding-insensitive lookups,
653 * but preserve the encoding of the stored key. Thus a particular
654 * key could map to two different HEKs in PL_strtab. We only
655 * conclude 'not found' if all the flags are the same; otherwise
656 * we fall back to a full search (this should only happen in rare
659 int keysv_flags = HEK_FLAGS(keysv_hek);
660 HE *orig_entry = entry;
662 for (; entry; entry = HeNEXT(entry)) {
663 HEK *hek = HeKEY_hek(entry);
664 if (hek == keysv_hek)
666 if (HEK_FLAGS(hek) != keysv_flags)
667 break; /* need to do full match */
671 /* failed on shortcut - do full search loop */
675 for (; entry; entry = HeNEXT(entry)) {
676 if (HeHASH(entry) != hash) /* strings can't be equal */
678 if (HeKLEN(entry) != (I32)klen)
680 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
682 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
686 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
687 if (HeKFLAGS(entry) != masked_flags) {
688 /* We match if HVhek_UTF8 bit in our flags and hash key's
689 match. But if entry was set previously with HVhek_WASUTF8
690 and key now doesn't (or vice versa) then we should change
691 the key's flag, as this is assignment. */
692 if (HvSHAREKEYS(hv)) {
693 /* Need to swap the key we have for a key with the flags we
694 need. As keys are shared we can't just write to the
695 flag, so we share the new one, unshare the old one. */
696 HEK * const new_hek = share_hek_flags(key, klen, hash,
698 unshare_hek (HeKEY_hek(entry));
699 HeKEY_hek(entry) = new_hek;
701 else if (hv == PL_strtab) {
702 /* PL_strtab is usually the only hash without HvSHAREKEYS,
703 so putting this test here is cheap */
704 if (flags & HVhek_FREEKEY)
706 Perl_croak(aTHX_ S_strtab_error,
707 action & HV_FETCH_LVALUE ? "fetch" : "store");
710 HeKFLAGS(entry) = masked_flags;
711 if (masked_flags & HVhek_ENABLEHVKFLAGS)
714 if (HeVAL(entry) == &PL_sv_placeholder) {
715 /* yes, can store into placeholder slot */
716 if (action & HV_FETCH_LVALUE) {
718 /* This preserves behaviour with the old hv_fetch
719 implementation which at this point would bail out
720 with a break; (at "if we find a placeholder, we
721 pretend we haven't found anything")
723 That break mean that if a placeholder were found, it
724 caused a call into hv_store, which in turn would
725 check magic, and if there is no magic end up pretty
726 much back at this point (in hv_store's code). */
729 /* LVAL fetch which actually needs a store. */
731 HvPLACEHOLDERS(hv)--;
734 if (val != &PL_sv_placeholder)
735 HvPLACEHOLDERS(hv)--;
738 } else if (action & HV_FETCH_ISSTORE) {
739 SvREFCNT_dec(HeVAL(entry));
742 } else if (HeVAL(entry) == &PL_sv_placeholder) {
743 /* if we find a placeholder, we pretend we haven't found
747 if (flags & HVhek_FREEKEY)
750 return (void *) &HeVAL(entry);
756 #ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
757 if (!(action & HV_FETCH_ISSTORE)
758 && SvRMAGICAL((const SV *)hv)
759 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
761 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
763 sv = newSVpvn(env,len);
765 return hv_common(hv, keysv, key, klen, flags,
766 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
772 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
773 hv_notallowed(flags, key, klen,
774 "Attempt to access disallowed key '%" SVf "' in"
775 " a restricted hash");
777 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
778 /* Not doing some form of store, so return failure. */
779 if (flags & HVhek_FREEKEY)
783 if (action & HV_FETCH_LVALUE) {
784 val = action & HV_FETCH_EMPTY_HE ? NULL : newSV(0);
786 /* At this point the old hv_fetch code would call to hv_store,
787 which in turn might do some tied magic. So we need to make that
788 magic check happen. */
789 /* gonna assign to this, so it better be there */
790 /* If a fetch-as-store fails on the fetch, then the action is to
791 recurse once into "hv_store". If we didn't do this, then that
792 recursive call would call the key conversion routine again.
793 However, as we replace the original key with the converted
794 key, this would result in a double conversion, which would show
795 up as a bug if the conversion routine is not idempotent.
796 Hence the use of HV_DISABLE_UVAR_XKEY. */
797 return hv_common(hv, keysv, key, klen, flags,
798 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
800 /* XXX Surely that could leak if the fetch-was-store fails?
801 Just like the hv_fetch. */
805 /* Welcome to hv_store... */
808 /* Not sure if we can get here. I think the only case of oentry being
809 NULL is for %ENV with dynamic env fetch. But that should disappear
810 with magic in the previous code. */
813 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
815 HvARRAY(hv) = (HE**)array;
818 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
821 /* share_hek_flags will do the free for us. This might be considered
824 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
825 else if (hv == PL_strtab) {
826 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
827 this test here is cheap */
828 if (flags & HVhek_FREEKEY)
830 Perl_croak(aTHX_ S_strtab_error,
831 action & HV_FETCH_LVALUE ? "fetch" : "store");
833 else /* gotta do the real thing */
834 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
837 #ifdef PERL_HASH_RANDOMIZE_KEYS
838 /* This logic semi-randomizes the insert order in a bucket.
839 * Either we insert into the top, or the slot below the top,
840 * making it harder to see if there is a collision. We also
841 * reset the iterator randomizer if there is one.
843 in_collision = *oentry != NULL;
844 if ( *oentry && PL_HASH_RAND_BITS_ENABLED) {
846 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
847 if ( PL_hash_rand_bits & 1 ) {
848 HeNEXT(entry) = HeNEXT(*oentry);
849 HeNEXT(*oentry) = entry;
851 HeNEXT(entry) = *oentry;
857 HeNEXT(entry) = *oentry;
860 #ifdef PERL_HASH_RANDOMIZE_KEYS
862 /* Currently this makes various tests warn in annoying ways.
863 * So Silenced for now. - Yves | bogus end of comment =>* /
864 if (HvAUX(hv)->xhv_riter != -1) {
865 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
866 "[TESTING] Inserting into a hash during each() traversal results in undefined behavior"
871 if (PL_HASH_RAND_BITS_ENABLED) {
872 if (PL_HASH_RAND_BITS_ENABLED == 1)
873 PL_hash_rand_bits += (PTRV)entry + 1; /* we don't bother to use ptr_hash here */
874 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
876 HvAUX(hv)->xhv_rand= (U32)PL_hash_rand_bits;
880 if (val == &PL_sv_placeholder)
881 HvPLACEHOLDERS(hv)++;
882 if (masked_flags & HVhek_ENABLEHVKFLAGS)
885 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
886 if ( in_collision && DO_HSPLIT(xhv) ) {
887 const STRLEN oldsize = xhv->xhv_max + 1;
888 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
890 if (items /* hash has placeholders */
891 && !SvREADONLY(hv) /* but is not a restricted hash */) {
892 /* If this hash previously was a "restricted hash" and had
893 placeholders, but the "restricted" flag has been turned off,
894 then the placeholders no longer serve any useful purpose.
895 However, they have the downsides of taking up RAM, and adding
896 extra steps when finding used values. It's safe to clear them
897 at this point, even though Storable rebuilds restricted hashes by
898 putting in all the placeholders (first) before turning on the
899 readonly flag, because Storable always pre-splits the hash.
900 If we're lucky, then we may clear sufficient placeholders to
901 avoid needing to split the hash at all. */
902 clear_placeholders(hv, items);
904 hsplit(hv, oldsize, oldsize * 2);
906 hsplit(hv, oldsize, oldsize * 2);
910 return entry ? (void *) &HeVAL(entry) : NULL;
912 return (void *) entry;
916 S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
918 const MAGIC *mg = SvMAGIC(hv);
920 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
925 if (isUPPER(mg->mg_type)) {
927 if (mg->mg_type == PERL_MAGIC_tied) {
928 *needs_store = FALSE;
929 return; /* We've set all there is to set. */
932 mg = mg->mg_moremagic;
937 =for apidoc hv_scalar
939 Evaluates the hash in scalar context and returns the result.
941 When the hash is tied dispatches through to the SCALAR method,
942 otherwise returns a mortal SV containing the number of keys
945 Note, prior to 5.25 this function returned what is now
946 returned by the hv_bucket_ratio() function.
952 Perl_hv_scalar(pTHX_ HV *hv)
956 PERL_ARGS_ASSERT_HV_SCALAR;
958 if (SvRMAGICAL(hv)) {
959 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
961 return magic_scalarpack(hv, mg);
965 sv_setuv(sv, HvUSEDKEYS(hv));
972 hv_pushkv(): push all the keys and/or values of a hash onto the stack.
973 The rough Perl equivalents:
978 Resets the hash's iterator.
980 flags : 1 = push keys
982 1|2 = push keys and values
983 XXX use symbolic flag constants at some point?
984 I might unroll the non-tied hv_iternext() in here at some point - DAPM
988 Perl_hv_pushkv(pTHX_ HV *hv, U32 flags)
991 bool tied = SvRMAGICAL(hv) && (mg_find(MUTABLE_SV(hv), PERL_MAGIC_tied)
992 #ifdef DYNAMIC_ENV_FETCH /* might not know number of keys yet */
993 || mg_find(MUTABLE_SV(hv), PERL_MAGIC_env)
998 PERL_ARGS_ASSERT_HV_PUSHKV;
999 assert(flags); /* must be pushing at least one of keys and values */
1001 (void)hv_iterinit(hv);
1004 SSize_t ext = (flags == 3) ? 2 : 1;
1005 while ((entry = hv_iternext(hv))) {
1008 PUSHs(hv_iterkeysv(entry));
1010 PUSHs(hv_iterval(hv, entry));
1014 Size_t nkeys = HvUSEDKEYS(hv);
1020 /* 2*nkeys() should never be big enough to truncate or wrap */
1021 assert(nkeys <= (SSize_t_MAX >> 1));
1022 ext = nkeys * ((flags == 3) ? 2 : 1);
1024 EXTEND_MORTAL(nkeys);
1027 while ((entry = hv_iternext(hv))) {
1029 SV *keysv = newSVhek(HeKEY_hek(entry));
1031 PL_tmps_stack[++PL_tmps_ix] = keysv;
1035 PUSHs(HeVAL(entry));
1044 =for apidoc hv_bucket_ratio
1046 If the hash is tied dispatches through to the SCALAR tied method,
1047 otherwise if the hash contains no keys returns 0, otherwise returns
1048 a mortal sv containing a string specifying the number of used buckets,
1049 followed by a slash, followed by the number of available buckets.
1051 This function is expensive, it must scan all of the buckets
1052 to determine which are used, and the count is NOT cached.
1053 In a large hash this could be a lot of buckets.
1059 Perl_hv_bucket_ratio(pTHX_ HV *hv)
1063 PERL_ARGS_ASSERT_HV_BUCKET_RATIO;
1065 if (SvRMAGICAL(hv)) {
1066 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
1068 return magic_scalarpack(hv, mg);
1071 if (HvUSEDKEYS((HV *)hv)) {
1072 sv = sv_newmortal();
1073 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
1074 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
1083 =for apidoc hv_delete
1085 Deletes a key/value pair in the hash. The value's SV is removed from
1086 the hash, made mortal, and returned to the caller. The absolute
1087 value of C<klen> is the length of the key. If C<klen> is negative the
1088 key is assumed to be in UTF-8-encoded Unicode. The C<flags> value
1089 will normally be zero; if set to C<G_DISCARD> then C<NULL> will be returned.
1090 C<NULL> will also be returned if the key is not found.
1092 =for apidoc hv_delete_ent
1094 Deletes a key/value pair in the hash. The value SV is removed from the hash,
1095 made mortal, and returned to the caller. The C<flags> value will normally be
1096 zero; if set to C<G_DISCARD> then C<NULL> will be returned. C<NULL> will also
1097 be returned if the key is not found. C<hash> can be a valid precomputed hash
1098 value, or 0 to ask for it to be computed.
1104 S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
1105 int k_flags, I32 d_flags, U32 hash)
1112 bool is_utf8 = cBOOL(k_flags & HVhek_UTF8);
1114 HEK *keysv_hek = NULL;
1115 U8 mro_changes = 0; /* 1 = isa; 2 = package moved */
1120 if (SvRMAGICAL(hv)) {
1123 hv_magic_check (hv, &needs_copy, &needs_store);
1127 entry = (HE *) hv_common(hv, keysv, key, klen,
1128 k_flags & ~HVhek_FREEKEY,
1129 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
1131 sv = entry ? HeVAL(entry) : NULL;
1133 if (SvMAGICAL(sv)) {
1137 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
1138 /* No longer an element */
1139 sv_unmagic(sv, PERL_MAGIC_tiedelem);
1142 return NULL; /* element cannot be deleted */
1144 #ifdef ENV_IS_CASELESS
1145 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
1146 /* XXX This code isn't UTF8 clean. */
1147 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
1148 if (k_flags & HVhek_FREEKEY) {
1151 key = strupr(SvPVX(keysv));
1160 xhv = (XPVHV*)SvANY(hv);
1164 if (is_utf8 && !(k_flags & HVhek_KEYCANONICAL)) {
1165 const char * const keysave = key;
1166 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
1169 k_flags |= HVhek_UTF8;
1171 k_flags &= ~HVhek_UTF8;
1172 if (key != keysave) {
1173 if (k_flags & HVhek_FREEKEY) {
1174 /* This shouldn't happen if our caller does what we expect,
1175 but strictly the API allows it. */
1178 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
1180 HvHASKFLAGS_on(MUTABLE_SV(hv));
1183 if (keysv && (SvIsCOW_shared_hash(keysv))) {
1184 if (HvSHAREKEYS(hv))
1185 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
1186 hash = SvSHARED_HASH(keysv);
1189 PERL_HASH(hash, key, klen);
1191 masked_flags = (k_flags & HVhek_MASK);
1193 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
1200 /* keysv is actually a HEK in disguise, so we can match just by
1201 * comparing the HEK pointers in the HE chain. There is a slight
1202 * caveat: on something like "\x80", which has both plain and utf8
1203 * representations, perl's hashes do encoding-insensitive lookups,
1204 * but preserve the encoding of the stored key. Thus a particular
1205 * key could map to two different HEKs in PL_strtab. We only
1206 * conclude 'not found' if all the flags are the same; otherwise
1207 * we fall back to a full search (this should only happen in rare
1210 int keysv_flags = HEK_FLAGS(keysv_hek);
1212 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1213 HEK *hek = HeKEY_hek(entry);
1214 if (hek == keysv_hek)
1216 if (HEK_FLAGS(hek) != keysv_flags)
1217 break; /* need to do full match */
1221 /* failed on shortcut - do full search loop */
1222 oentry = first_entry;
1226 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1227 if (HeHASH(entry) != hash) /* strings can't be equal */
1229 if (HeKLEN(entry) != (I32)klen)
1231 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
1233 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
1237 if (hv == PL_strtab) {
1238 if (k_flags & HVhek_FREEKEY)
1240 Perl_croak(aTHX_ S_strtab_error, "delete");
1243 /* if placeholder is here, it's already been deleted.... */
1244 if (HeVAL(entry) == &PL_sv_placeholder) {
1245 if (k_flags & HVhek_FREEKEY)
1249 if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1250 hv_notallowed(k_flags, key, klen,
1251 "Attempt to delete readonly key '%" SVf "' from"
1252 " a restricted hash");
1254 if (k_flags & HVhek_FREEKEY)
1257 /* If this is a stash and the key ends with ::, then someone is
1258 * deleting a package.
1260 if (HeVAL(entry) && HvENAME_get(hv)) {
1261 gv = (GV *)HeVAL(entry);
1262 if (keysv) key = SvPV(keysv, klen);
1264 (klen > 1 && key[klen-2] == ':' && key[klen-1] == ':')
1266 (klen == 1 && key[0] == ':')
1268 && (klen != 6 || hv!=PL_defstash || memNE(key,"main::",6))
1269 && SvTYPE(gv) == SVt_PVGV && (stash = GvHV((GV *)gv))
1270 && HvENAME_get(stash)) {
1271 /* A previous version of this code checked that the
1272 * GV was still in the symbol table by fetching the
1273 * GV with its name. That is not necessary (and
1274 * sometimes incorrect), as HvENAME cannot be set
1275 * on hv if it is not in the symtab. */
1277 /* Hang on to it for a bit. */
1278 SvREFCNT_inc_simple_void_NN(
1279 sv_2mortal((SV *)gv)
1282 else if (memEQs(key, klen, "ISA") && GvAV(gv)) {
1284 MAGIC *mg = mg_find((SV*)isa, PERL_MAGIC_isa);
1288 if (mg->mg_obj == (SV*)gv) {
1289 /* This is the only stash this ISA was used for.
1290 * The isaelem magic asserts if there's no
1291 * isa magic on the array, so explicitly
1292 * remove the magic on both the array and its
1293 * elements. @ISA shouldn't be /too/ large.
1298 end = svp + AvFILLp(isa)+1;
1301 mg_free_type(*svp, PERL_MAGIC_isaelem);
1304 mg_free_type((SV*)GvAV(gv), PERL_MAGIC_isa);
1307 /* mg_obj is an array of stashes
1308 Note that the array doesn't keep a reference
1309 count on the stashes.
1311 AV *av = (AV*)mg->mg_obj;
1316 assert(SvTYPE(mg->mg_obj) == SVt_PVAV);
1318 /* remove the stash from the magic array */
1319 arrayp = svp = AvARRAY(av);
1320 items = AvFILLp(av) + 1;
1322 assert(*arrayp == (SV *)gv);
1324 /* avoid a double free on the last stash */
1326 /* The magic isn't MGf_REFCOUNTED, so release
1327 * the array manually.
1329 SvREFCNT_dec_NN(av);
1334 if (*svp == (SV*)gv)
1338 index = svp - arrayp;
1339 assert(index >= 0 && index <= AvFILLp(av));
1340 if (index < AvFILLp(av)) {
1341 arrayp[index] = arrayp[AvFILLp(av)];
1343 arrayp[AvFILLp(av)] = NULL;
1351 sv = d_flags & G_DISCARD ? HeVAL(entry) : sv_2mortal(HeVAL(entry));
1352 HeVAL(entry) = &PL_sv_placeholder;
1354 /* deletion of method from stash */
1355 if (isGV(sv) && isGV_with_GP(sv) && GvCVu(sv)
1357 mro_method_changed_in(hv);
1361 * If a restricted hash, rather than really deleting the entry, put
1362 * a placeholder there. This marks the key as being "approved", so
1363 * we can still access via not-really-existing key without raising
1367 /* We'll be saving this slot, so the number of allocated keys
1368 * doesn't go down, but the number placeholders goes up */
1369 HvPLACEHOLDERS(hv)++;
1371 *oentry = HeNEXT(entry);
1372 if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */)
1375 if (SvOOK(hv) && HvLAZYDEL(hv) &&
1376 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
1377 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
1378 hv_free_ent(hv, entry);
1380 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
1381 if (xhv->xhv_keys == 0)
1382 HvHASKFLAGS_off(hv);
1385 if (d_flags & G_DISCARD) {
1390 if (mro_changes == 1) mro_isa_changed_in(hv);
1391 else if (mro_changes == 2)
1392 mro_package_moved(NULL, stash, gv, 1);
1398 if (SvREADONLY(hv)) {
1399 hv_notallowed(k_flags, key, klen,
1400 "Attempt to delete disallowed key '%" SVf "' from"
1401 " a restricted hash");
1404 if (k_flags & HVhek_FREEKEY)
1411 S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize)
1414 char *a = (char*) HvARRAY(hv);
1418 /* already have an HvAUX(hv) so we have to move it */
1420 /* no HvAUX() but array we are going to allocate is large enough
1421 * there is no point in saving the space for the iterator, and
1422 * speeds up later traversals. */
1423 ( ( hv != PL_strtab ) && ( newsize >= PERL_HV_ALLOC_AUX_SIZE ) )
1426 PERL_ARGS_ASSERT_HSPLIT;
1429 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1430 + (do_aux ? sizeof(struct xpvhv_aux) : 0), char);
1436 #ifdef PERL_HASH_RANDOMIZE_KEYS
1437 /* the idea of this is that we create a "random" value by hashing the address of
1438 * the array, we then use the low bit to decide if we insert at the top, or insert
1439 * second from top. After each such insert we rotate the hashed value. So we can
1440 * use the same hashed value over and over, and in normal build environments use
1441 * very few ops to do so. ROTL32() should produce a single machine operation. */
1442 if (PL_HASH_RAND_BITS_ENABLED) {
1443 if (PL_HASH_RAND_BITS_ENABLED == 1)
1444 PL_hash_rand_bits += ptr_hash((PTRV)a);
1445 PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1);
1448 HvARRAY(hv) = (HE**) a;
1449 HvMAX(hv) = newsize - 1;
1450 /* before we zero the newly added memory, we
1451 * need to deal with the aux struct that may be there
1452 * or have been allocated by us*/
1454 struct xpvhv_aux *const dest
1455 = (struct xpvhv_aux*) &a[newsize * sizeof(HE*)];
1457 /* alread have an aux, copy the old one in place. */
1458 Move(&a[oldsize * sizeof(HE*)], dest, 1, struct xpvhv_aux);
1459 /* we reset the iterator's xhv_rand as well, so they get a totally new ordering */
1460 #ifdef PERL_HASH_RANDOMIZE_KEYS
1461 dest->xhv_rand = (U32)PL_hash_rand_bits;
1464 /* no existing aux structure, but we allocated space for one
1465 * so initialize it properly. This unrolls hv_auxinit() a bit,
1466 * since we have to do the realloc anyway. */
1467 /* first we set the iterator's xhv_rand so it can be copied into lastrand below */
1468 #ifdef PERL_HASH_RANDOMIZE_KEYS
1469 dest->xhv_rand = (U32)PL_hash_rand_bits;
1471 /* this is the "non realloc" part of the hv_auxinit() */
1472 (void)hv_auxinit_internal(dest);
1473 /* Turn on the OOK flag */
1477 /* now we can safely clear the second half */
1478 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
1480 if (!HvTOTALKEYS(hv)) /* skip rest if no entries */
1486 HE **oentry = aep + i;
1489 if (!entry) /* non-existent */
1492 U32 j = (HeHASH(entry) & newsize);
1494 *oentry = HeNEXT(entry);
1495 #ifdef PERL_HASH_RANDOMIZE_KEYS
1496 /* if the target cell is empty or PL_HASH_RAND_BITS_ENABLED is false
1497 * insert to top, otherwise rotate the bucket rand 1 bit,
1498 * and use the new low bit to decide if we insert at top,
1499 * or next from top. IOW, we only rotate on a collision.*/
1500 if (aep[j] && PL_HASH_RAND_BITS_ENABLED) {
1501 PL_hash_rand_bits+= ROTL32(HeHASH(entry), 17);
1502 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
1503 if (PL_hash_rand_bits & 1) {
1504 HeNEXT(entry)= HeNEXT(aep[j]);
1505 HeNEXT(aep[j])= entry;
1507 /* Note, this is structured in such a way as the optimizer
1508 * should eliminate the duplicated code here and below without
1509 * us needing to explicitly use a goto. */
1510 HeNEXT(entry) = aep[j];
1516 /* see comment above about duplicated code */
1517 HeNEXT(entry) = aep[j];
1522 oentry = &HeNEXT(entry);
1526 } while (i++ < oldsize);
1530 Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
1532 XPVHV* xhv = (XPVHV*)SvANY(hv);
1533 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 */
1539 PERL_ARGS_ASSERT_HV_KSPLIT;
1541 wantsize = (I32) newmax; /* possible truncation here */
1542 if (wantsize != newmax)
1545 wantsize= wantsize + (wantsize >> 1); /* wantsize *= 1.5 */
1546 if (wantsize < newmax) /* overflow detection */
1550 while (wantsize > newsize) {
1551 trysize = newsize << 1;
1552 if (trysize > newsize) {
1560 if (newsize <= oldsize)
1561 return; /* overflow detection */
1563 a = (char *) HvARRAY(hv);
1565 hsplit(hv, oldsize, newsize);
1567 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
1568 xhv->xhv_max = newsize - 1;
1569 HvARRAY(hv) = (HE **) a;
1573 /* IMO this should also handle cases where hv_max is smaller than hv_keys
1574 * as tied hashes could play silly buggers and mess us around. We will
1575 * do the right thing during hv_store() afterwards, but still - Yves */
1576 #define HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys) STMT_START {\
1577 /* Can we use fewer buckets? (hv_max is always 2^n-1) */ \
1578 if (hv_max < PERL_HASH_DEFAULT_HvMAX) { \
1579 hv_max = PERL_HASH_DEFAULT_HvMAX; \
1581 while (hv_max > PERL_HASH_DEFAULT_HvMAX && hv_max + 1 >= hv_keys * 2) \
1582 hv_max = hv_max / 2; \
1584 HvMAX(hv) = hv_max; \
1589 Perl_newHVhv(pTHX_ HV *ohv)
1592 HV * const hv = newHV();
1595 if (!ohv || (!HvTOTALKEYS(ohv) && !SvMAGICAL((const SV *)ohv)))
1597 hv_max = HvMAX(ohv);
1599 if (!SvMAGICAL((const SV *)ohv)) {
1600 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
1602 const bool shared = !!HvSHAREKEYS(ohv);
1603 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
1605 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
1608 /* In each bucket... */
1609 for (i = 0; i <= hv_max; i++) {
1611 HE *oent = oents[i];
1618 /* Copy the linked list of entries. */
1619 for (; oent; oent = HeNEXT(oent)) {
1620 const U32 hash = HeHASH(oent);
1621 const char * const key = HeKEY(oent);
1622 const STRLEN len = HeKLEN(oent);
1623 const int flags = HeKFLAGS(oent);
1624 HE * const ent = new_HE();
1625 SV *const val = HeVAL(oent);
1627 HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val);
1629 = shared ? share_hek_flags(key, len, hash, flags)
1630 : save_hek_flags(key, len, hash, flags);
1641 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
1645 /* Iterate over ohv, copying keys and values one at a time. */
1647 const I32 riter = HvRITER_get(ohv);
1648 HE * const eiter = HvEITER_get(ohv);
1649 STRLEN hv_keys = HvTOTALKEYS(ohv);
1651 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
1654 while ((entry = hv_iternext_flags(ohv, 0))) {
1655 SV *val = hv_iterval(ohv,entry);
1656 SV * const keysv = HeSVKEY(entry);
1657 val = SvIMMORTAL(val) ? val : newSVsv(val);
1659 (void)hv_store_ent(hv, keysv, val, 0);
1661 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry), val,
1662 HeHASH(entry), HeKFLAGS(entry));
1664 HvRITER_set(ohv, riter);
1665 HvEITER_set(ohv, eiter);
1672 =for apidoc Am|HV *|hv_copy_hints_hv|HV *ohv
1674 A specialised version of L</newHVhv> for copying C<%^H>. C<ohv> must be
1675 a pointer to a hash (which may have C<%^H> magic, but should be generally
1676 non-magical), or C<NULL> (interpreted as an empty hash). The content
1677 of C<ohv> is copied to a new hash, which has the C<%^H>-specific magic
1678 added to it. A pointer to the new hash is returned.
1684 Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1686 HV * const hv = newHV();
1689 STRLEN hv_max = HvMAX(ohv);
1690 STRLEN hv_keys = HvTOTALKEYS(ohv);
1692 const I32 riter = HvRITER_get(ohv);
1693 HE * const eiter = HvEITER_get(ohv);
1698 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
1701 while ((entry = hv_iternext_flags(ohv, 0))) {
1702 SV *const sv = newSVsv(hv_iterval(ohv,entry));
1703 SV *heksv = HeSVKEY(entry);
1704 if (!heksv && sv) heksv = newSVhek(HeKEY_hek(entry));
1705 if (sv) sv_magic(sv, NULL, PERL_MAGIC_hintselem,
1706 (char *)heksv, HEf_SVKEY);
1707 if (heksv == HeSVKEY(entry))
1708 (void)hv_store_ent(hv, heksv, sv, 0);
1710 (void)hv_common(hv, heksv, HeKEY(entry), HeKLEN(entry),
1711 HeKFLAGS(entry), HV_FETCH_ISSTORE|HV_FETCH_JUST_SV, sv, HeHASH(entry));
1712 SvREFCNT_dec_NN(heksv);
1715 HvRITER_set(ohv, riter);
1716 HvEITER_set(ohv, eiter);
1718 SvREFCNT_inc_simple_void_NN(hv);
1721 hv_magic(hv, NULL, PERL_MAGIC_hints);
1724 #undef HV_SET_MAX_ADJUSTED_FOR_KEYS
1726 /* like hv_free_ent, but returns the SV rather than freeing it */
1728 S_hv_free_ent_ret(pTHX_ HV *hv, HE *entry)
1732 PERL_ARGS_ASSERT_HV_FREE_ENT_RET;
1735 if (HeKLEN(entry) == HEf_SVKEY) {
1736 SvREFCNT_dec(HeKEY_sv(entry));
1737 Safefree(HeKEY_hek(entry));
1739 else if (HvSHAREKEYS(hv))
1740 unshare_hek(HeKEY_hek(entry));
1742 Safefree(HeKEY_hek(entry));
1749 Perl_hv_free_ent(pTHX_ HV *hv, HE *entry)
1753 PERL_ARGS_ASSERT_HV_FREE_ENT;
1757 val = hv_free_ent_ret(hv, entry);
1763 Perl_hv_delayfree_ent(pTHX_ HV *hv, HE *entry)
1765 PERL_ARGS_ASSERT_HV_DELAYFREE_ENT;
1769 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1770 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
1771 if (HeKLEN(entry) == HEf_SVKEY) {
1772 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
1774 hv_free_ent(hv, entry);
1778 =for apidoc hv_clear
1780 Frees the all the elements of a hash, leaving it empty.
1781 The XS equivalent of C<%hash = ()>. See also L</hv_undef>.
1783 See L</av_clear> for a note about the hash possibly being invalid on
1790 Perl_hv_clear(pTHX_ HV *hv)
1799 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1801 xhv = (XPVHV*)SvANY(hv);
1803 /* avoid hv being freed when calling destructors below */
1805 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
1806 orig_ix = PL_tmps_ix;
1807 if (SvREADONLY(hv) && HvARRAY(hv) != NULL) {
1808 /* restricted hash: convert all keys to placeholders */
1810 for (i = 0; i <= xhv->xhv_max; i++) {
1811 HE *entry = (HvARRAY(hv))[i];
1812 for (; entry; entry = HeNEXT(entry)) {
1813 /* not already placeholder */
1814 if (HeVAL(entry) != &PL_sv_placeholder) {
1816 if (SvREADONLY(HeVAL(entry))) {
1817 SV* const keysv = hv_iterkeysv(entry);
1818 Perl_croak_nocontext(
1819 "Attempt to delete readonly key '%" SVf "' from a restricted hash",
1822 SvREFCNT_dec_NN(HeVAL(entry));
1824 HeVAL(entry) = &PL_sv_placeholder;
1825 HvPLACEHOLDERS(hv)++;
1831 hv_free_entries(hv);
1832 HvPLACEHOLDERS_set(hv, 0);
1835 mg_clear(MUTABLE_SV(hv));
1837 HvHASKFLAGS_off(hv);
1841 mro_isa_changed_in(hv);
1842 HvEITER_set(hv, NULL);
1844 /* disarm hv's premature free guard */
1845 if (LIKELY(PL_tmps_ix == orig_ix))
1848 PL_tmps_stack[orig_ix] = &PL_sv_undef;
1849 SvREFCNT_dec_NN(hv);
1853 =for apidoc hv_clear_placeholders
1855 Clears any placeholders from a hash. If a restricted hash has any of its keys
1856 marked as readonly and the key is subsequently deleted, the key is not actually
1857 deleted but is marked by assigning it a value of C<&PL_sv_placeholder>. This tags
1858 it so it will be ignored by future operations such as iterating over the hash,
1859 but will still allow the hash to have a value reassigned to the key at some
1860 future point. This function clears any such placeholder keys from the hash.
1861 See C<L<Hash::Util::lock_keys()|Hash::Util/lock_keys>> for an example of its
1868 Perl_hv_clear_placeholders(pTHX_ HV *hv)
1870 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
1872 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
1875 clear_placeholders(hv, items);
1879 S_clear_placeholders(pTHX_ HV *hv, U32 items)
1884 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
1891 /* Loop down the linked list heads */
1892 HE **oentry = &(HvARRAY(hv))[i];
1895 while ((entry = *oentry)) {
1896 if (HeVAL(entry) == &PL_sv_placeholder) {
1897 *oentry = HeNEXT(entry);
1898 if (entry == HvEITER_get(hv))
1901 if (SvOOK(hv) && HvLAZYDEL(hv) &&
1902 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
1903 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
1904 hv_free_ent(hv, entry);
1909 I32 placeholders = HvPLACEHOLDERS_get(hv);
1910 HvTOTALKEYS(hv) -= (IV)placeholders;
1911 /* HvUSEDKEYS expanded */
1912 if ((HvTOTALKEYS(hv) - placeholders) == 0)
1913 HvHASKFLAGS_off(hv);
1914 HvPLACEHOLDERS_set(hv, 0);
1918 oentry = &HeNEXT(entry);
1922 /* You can't get here, hence assertion should always fail. */
1923 assert (items == 0);
1924 NOT_REACHED; /* NOTREACHED */
1928 S_hv_free_entries(pTHX_ HV *hv)
1931 XPVHV * const xhv = (XPVHV*)SvANY(hv);
1934 PERL_ARGS_ASSERT_HV_FREE_ENTRIES;
1936 while ((sv = Perl_hfree_next_entry(aTHX_ hv, &index))||xhv->xhv_keys) {
1942 /* hfree_next_entry()
1943 * For use only by S_hv_free_entries() and sv_clear().
1944 * Delete the next available HE from hv and return the associated SV.
1945 * Returns null on empty hash. Nevertheless null is not a reliable
1946 * indicator that the hash is empty, as the deleted entry may have a
1948 * indexp is a pointer to the current index into HvARRAY. The index should
1949 * initially be set to 0. hfree_next_entry() may update it. */
1952 Perl_hfree_next_entry(pTHX_ HV *hv, STRLEN *indexp)
1954 struct xpvhv_aux *iter;
1958 STRLEN orig_index = *indexp;
1961 PERL_ARGS_ASSERT_HFREE_NEXT_ENTRY;
1963 if (SvOOK(hv) && ((iter = HvAUX(hv)))) {
1964 if ((entry = iter->xhv_eiter)) {
1965 /* the iterator may get resurrected after each
1966 * destructor call, so check each time */
1967 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1969 hv_free_ent(hv, entry);
1970 /* warning: at this point HvARRAY may have been
1971 * re-allocated, HvMAX changed etc */
1973 iter = HvAUX(hv); /* may have been realloced */
1974 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1975 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1976 #ifdef PERL_HASH_RANDOMIZE_KEYS
1977 iter->xhv_last_rand = iter->xhv_rand;
1982 if (!((XPVHV*)SvANY(hv))->xhv_keys)
1985 array = HvARRAY(hv);
1987 while ( ! ((entry = array[*indexp])) ) {
1988 if ((*indexp)++ >= HvMAX(hv))
1990 assert(*indexp != orig_index);
1992 array[*indexp] = HeNEXT(entry);
1993 ((XPVHV*) SvANY(hv))->xhv_keys--;
1995 if ( PL_phase != PERL_PHASE_DESTRUCT && HvENAME(hv)
1996 && HeVAL(entry) && isGV(HeVAL(entry))
1997 && GvHV(HeVAL(entry)) && HvENAME(GvHV(HeVAL(entry)))
2000 const char * const key = HePV(entry,klen);
2001 if ((klen > 1 && key[klen-1]==':' && key[klen-2]==':')
2002 || (klen == 1 && key[0] == ':')) {
2004 NULL, GvHV(HeVAL(entry)),
2005 (GV *)HeVAL(entry), 0
2009 return hv_free_ent_ret(hv, entry);
2014 =for apidoc hv_undef
2016 Undefines the hash. The XS equivalent of C<undef(%hash)>.
2018 As well as freeing all the elements of the hash (like C<hv_clear()>), this
2019 also frees any auxiliary data and storage associated with the hash.
2021 See L</av_clear> for a note about the hash possibly being invalid on
2028 Perl_hv_undef_flags(pTHX_ HV *hv, U32 flags)
2032 SSize_t orig_ix = PL_tmps_ix; /* silence compiler warning about unitialized vars */
2036 save = cBOOL(SvREFCNT(hv));
2037 DEBUG_A(Perl_hv_assert(aTHX_ hv));
2038 xhv = (XPVHV*)SvANY(hv);
2040 /* The name must be deleted before the call to hv_free_entries so that
2041 CVs are anonymised properly. But the effective name must be pre-
2042 served until after that call (and only deleted afterwards if the
2043 call originated from sv_clear). For stashes with one name that is
2044 both the canonical name and the effective name, hv_name_set has to
2045 allocate an array for storing the effective name. We can skip that
2046 during global destruction, as it does not matter where the CVs point
2047 if they will be freed anyway. */
2048 /* note that the code following prior to hv_free_entries is duplicated
2049 * in sv_clear(), and changes here should be done there too */
2050 if (PL_phase != PERL_PHASE_DESTRUCT && HvNAME(hv)) {
2051 if (PL_stashcache) {
2052 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for '%"
2053 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
2054 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
2056 hv_name_set(hv, NULL, 0, 0);
2059 /* avoid hv being freed when calling destructors below */
2061 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
2062 orig_ix = PL_tmps_ix;
2064 hv_free_entries(hv);
2066 struct mro_meta *meta;
2069 if (HvENAME_get(hv)) {
2070 if (PL_phase != PERL_PHASE_DESTRUCT)
2071 mro_isa_changed_in(hv);
2072 if (PL_stashcache) {
2073 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for effective name '%"
2074 HEKf "'\n", HEKfARG(HvENAME_HEK(hv))));
2075 (void)hv_deletehek(PL_stashcache, HvENAME_HEK(hv), G_DISCARD);
2079 /* If this call originated from sv_clear, then we must check for
2080 * effective names that need freeing, as well as the usual name. */
2082 if (flags & HV_NAME_SETALL ? !!HvAUX(hv)->xhv_name_u.xhvnameu_name : !!name) {
2083 if (name && PL_stashcache) {
2084 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for name '%"
2085 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
2086 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
2088 hv_name_set(hv, NULL, 0, flags);
2090 if((meta = HvAUX(hv)->xhv_mro_meta)) {
2091 if (meta->mro_linear_all) {
2092 SvREFCNT_dec_NN(meta->mro_linear_all);
2093 /* mro_linear_current is just acting as a shortcut pointer,
2097 /* Only the current MRO is stored, so this owns the data.
2099 SvREFCNT_dec(meta->mro_linear_current);
2100 SvREFCNT_dec(meta->mro_nextmethod);
2101 SvREFCNT_dec(meta->isa);
2102 SvREFCNT_dec(meta->super);
2104 HvAUX(hv)->xhv_mro_meta = NULL;
2106 if (!HvAUX(hv)->xhv_name_u.xhvnameu_name && ! HvAUX(hv)->xhv_backreferences)
2107 SvFLAGS(hv) &= ~SVf_OOK;
2110 Safefree(HvARRAY(hv));
2111 xhv->xhv_max = PERL_HASH_DEFAULT_HvMAX; /* HvMAX(hv) = 7 (it's a normal hash) */
2114 /* if we're freeing the HV, the SvMAGIC field has been reused for
2115 * other purposes, and so there can't be any placeholder magic */
2117 HvPLACEHOLDERS_set(hv, 0);
2120 mg_clear(MUTABLE_SV(hv));
2123 /* disarm hv's premature free guard */
2124 if (LIKELY(PL_tmps_ix == orig_ix))
2127 PL_tmps_stack[orig_ix] = &PL_sv_undef;
2128 SvREFCNT_dec_NN(hv);
2135 Returns the number of hash buckets that happen to be in use.
2137 This function is wrapped by the macro C<HvFILL>.
2139 As of perl 5.25 this function is used only for debugging
2140 purposes, and the number of used hash buckets is not
2141 in any way cached, thus this function can be costly
2142 to execute as it must iterate over all the buckets in the
2149 Perl_hv_fill(pTHX_ HV *const hv)
2152 HE **ents = HvARRAY(hv);
2154 PERL_UNUSED_CONTEXT;
2155 PERL_ARGS_ASSERT_HV_FILL;
2157 /* No keys implies no buckets used.
2158 One key can only possibly mean one bucket used. */
2159 if (HvTOTALKEYS(hv) < 2)
2160 return HvTOTALKEYS(hv);
2163 /* I wonder why we count down here...
2164 * Is it some micro-optimisation?
2165 * I would have thought counting up was better.
2168 HE *const *const last = ents + HvMAX(hv);
2169 count = last + 1 - ents;
2174 } while (++ents <= last);
2179 /* hash a pointer to a U32 - Used in the hash traversal randomization
2180 * and bucket order randomization code
2182 * this code was derived from Sereal, which was derived from autobox.
2185 PERL_STATIC_INLINE U32 S_ptr_hash(PTRV u) {
2188 * This is one of Thomas Wang's hash functions for 64-bit integers from:
2189 * http://www.concentric.net/~Ttwang/tech/inthash.htm
2191 u = (~u) + (u << 18);
2199 * This is one of Bob Jenkins' hash functions for 32-bit integers
2200 * from: http://burtleburtle.net/bob/hash/integer.html
2202 u = (u + 0x7ed55d16) + (u << 12);
2203 u = (u ^ 0xc761c23c) ^ (u >> 19);
2204 u = (u + 0x165667b1) + (u << 5);
2205 u = (u + 0xd3a2646c) ^ (u << 9);
2206 u = (u + 0xfd7046c5) + (u << 3);
2207 u = (u ^ 0xb55a4f09) ^ (u >> 16);
2212 static struct xpvhv_aux*
2213 S_hv_auxinit_internal(struct xpvhv_aux *iter) {
2214 PERL_ARGS_ASSERT_HV_AUXINIT_INTERNAL;
2215 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2216 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2217 #ifdef PERL_HASH_RANDOMIZE_KEYS
2218 iter->xhv_last_rand = iter->xhv_rand;
2220 iter->xhv_name_u.xhvnameu_name = 0;
2221 iter->xhv_name_count = 0;
2222 iter->xhv_backreferences = 0;
2223 iter->xhv_mro_meta = NULL;
2224 iter->xhv_aux_flags = 0;
2229 static struct xpvhv_aux*
2230 S_hv_auxinit(pTHX_ HV *hv) {
2231 struct xpvhv_aux *iter;
2234 PERL_ARGS_ASSERT_HV_AUXINIT;
2238 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
2239 + sizeof(struct xpvhv_aux), char);
2241 array = (char *) HvARRAY(hv);
2242 Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
2243 + sizeof(struct xpvhv_aux), char);
2245 HvARRAY(hv) = (HE**)array;
2248 #ifdef PERL_HASH_RANDOMIZE_KEYS
2249 if (PL_HASH_RAND_BITS_ENABLED) {
2250 /* mix in some new state to PL_hash_rand_bits to "randomize" the traversal order*/
2251 if (PL_HASH_RAND_BITS_ENABLED == 1)
2252 PL_hash_rand_bits += ptr_hash((PTRV)array);
2253 PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1);
2255 iter->xhv_rand = (U32)PL_hash_rand_bits;
2261 return hv_auxinit_internal(iter);
2265 =for apidoc hv_iterinit
2267 Prepares a starting point to traverse a hash table. Returns the number of
2268 keys in the hash, including placeholders (i.e. the same as C<HvTOTALKEYS(hv)>).
2269 The return value is currently only meaningful for hashes without tie magic.
2271 NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
2272 hash buckets that happen to be in use. If you still need that esoteric
2273 value, you can get it through the macro C<HvFILL(hv)>.
2280 Perl_hv_iterinit(pTHX_ HV *hv)
2282 PERL_ARGS_ASSERT_HV_ITERINIT;
2285 struct xpvhv_aux * iter = HvAUX(hv);
2286 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
2287 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2289 hv_free_ent(hv, entry);
2291 iter = HvAUX(hv); /* may have been reallocated */
2292 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2293 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2294 #ifdef PERL_HASH_RANDOMIZE_KEYS
2295 iter->xhv_last_rand = iter->xhv_rand;
2301 /* note this includes placeholders! */
2302 return HvTOTALKEYS(hv);
2306 Perl_hv_riter_p(pTHX_ HV *hv) {
2307 struct xpvhv_aux *iter;
2309 PERL_ARGS_ASSERT_HV_RITER_P;
2311 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2312 return &(iter->xhv_riter);
2316 Perl_hv_eiter_p(pTHX_ HV *hv) {
2317 struct xpvhv_aux *iter;
2319 PERL_ARGS_ASSERT_HV_EITER_P;
2321 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2322 return &(iter->xhv_eiter);
2326 Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
2327 struct xpvhv_aux *iter;
2329 PERL_ARGS_ASSERT_HV_RITER_SET;
2337 iter = hv_auxinit(hv);
2339 iter->xhv_riter = riter;
2343 Perl_hv_rand_set(pTHX_ HV *hv, U32 new_xhv_rand) {
2344 struct xpvhv_aux *iter;
2346 PERL_ARGS_ASSERT_HV_RAND_SET;
2348 #ifdef PERL_HASH_RANDOMIZE_KEYS
2352 iter = hv_auxinit(hv);
2354 iter->xhv_rand = new_xhv_rand;
2356 Perl_croak(aTHX_ "This Perl has not been built with support for randomized hash key traversal but something called Perl_hv_rand_set().");
2361 Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
2362 struct xpvhv_aux *iter;
2364 PERL_ARGS_ASSERT_HV_EITER_SET;
2369 /* 0 is the default so don't go malloc()ing a new structure just to
2374 iter = hv_auxinit(hv);
2376 iter->xhv_eiter = eiter;
2380 Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
2383 struct xpvhv_aux *iter;
2387 PERL_ARGS_ASSERT_HV_NAME_SET;
2390 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
2394 if (iter->xhv_name_u.xhvnameu_name) {
2395 if(iter->xhv_name_count) {
2396 if(flags & HV_NAME_SETALL) {
2397 HEK ** const name = HvAUX(hv)->xhv_name_u.xhvnameu_names;
2398 HEK **hekp = name + (
2399 iter->xhv_name_count < 0
2400 ? -iter->xhv_name_count
2401 : iter->xhv_name_count
2403 while(hekp-- > name+1)
2404 unshare_hek_or_pvn(*hekp, 0, 0, 0);
2405 /* The first elem may be null. */
2406 if(*name) unshare_hek_or_pvn(*name, 0, 0, 0);
2408 iter = HvAUX(hv); /* may been realloced */
2409 spot = &iter->xhv_name_u.xhvnameu_name;
2410 iter->xhv_name_count = 0;
2413 if(iter->xhv_name_count > 0) {
2414 /* shift some things over */
2416 iter->xhv_name_u.xhvnameu_names, iter->xhv_name_count + 1, HEK *
2418 spot = iter->xhv_name_u.xhvnameu_names;
2419 spot[iter->xhv_name_count] = spot[1];
2421 iter->xhv_name_count = -(iter->xhv_name_count + 1);
2423 else if(*(spot = iter->xhv_name_u.xhvnameu_names)) {
2424 unshare_hek_or_pvn(*spot, 0, 0, 0);
2428 else if (flags & HV_NAME_SETALL) {
2429 unshare_hek_or_pvn(iter->xhv_name_u.xhvnameu_name, 0, 0, 0);
2430 iter = HvAUX(hv); /* may been realloced */
2431 spot = &iter->xhv_name_u.xhvnameu_name;
2434 HEK * const existing_name = iter->xhv_name_u.xhvnameu_name;
2435 Newx(iter->xhv_name_u.xhvnameu_names, 2, HEK *);
2436 iter->xhv_name_count = -2;
2437 spot = iter->xhv_name_u.xhvnameu_names;
2438 spot[1] = existing_name;
2441 else { spot = &iter->xhv_name_u.xhvnameu_name; iter->xhv_name_count = 0; }
2446 iter = hv_auxinit(hv);
2447 spot = &iter->xhv_name_u.xhvnameu_name;
2449 PERL_HASH(hash, name, len);
2450 *spot = name ? share_hek(name, flags & SVf_UTF8 ? -(I32)len : (I32)len, hash) : NULL;
2454 This is basically sv_eq_flags() in sv.c, but we avoid the magic
2459 hek_eq_pvn_flags(pTHX_ const HEK *hek, const char* pv, const I32 pvlen, const U32 flags) {
2460 if ( (HEK_UTF8(hek) ? 1 : 0) != (flags & SVf_UTF8 ? 1 : 0) ) {
2461 if (flags & SVf_UTF8)
2462 return (bytes_cmp_utf8(
2463 (const U8*)HEK_KEY(hek), HEK_LEN(hek),
2464 (const U8*)pv, pvlen) == 0);
2466 return (bytes_cmp_utf8(
2467 (const U8*)pv, pvlen,
2468 (const U8*)HEK_KEY(hek), HEK_LEN(hek)) == 0);
2471 return HEK_LEN(hek) == pvlen && ((HEK_KEY(hek) == pv)
2472 || memEQ(HEK_KEY(hek), pv, pvlen));
2476 =for apidoc hv_ename_add
2478 Adds a name to a stash's internal list of effective names. See
2479 C<L</hv_ename_delete>>.
2481 This is called when a stash is assigned to a new location in the symbol
2488 Perl_hv_ename_add(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
2491 struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2494 PERL_ARGS_ASSERT_HV_ENAME_ADD;
2497 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
2499 PERL_HASH(hash, name, len);
2501 if (aux->xhv_name_count) {
2502 I32 count = aux->xhv_name_count;
2503 HEK ** const xhv_name = aux->xhv_name_u.xhvnameu_names + (count<0);
2504 HEK **hekp = xhv_name + (count < 0 ? -count - 1 : count);
2505 while (hekp-- > xhv_name)
2509 (HEK_UTF8(*hekp) || (flags & SVf_UTF8))
2510 ? hek_eq_pvn_flags(aTHX_ *hekp, name, (I32)len, flags)
2511 : (HEK_LEN(*hekp) == (I32)len && memEQ(HEK_KEY(*hekp), name, len))
2513 if (hekp == xhv_name && count < 0)
2514 aux->xhv_name_count = -count;
2518 if (count < 0) aux->xhv_name_count--, count = -count;
2519 else aux->xhv_name_count++;
2520 Renew(aux->xhv_name_u.xhvnameu_names, count + 1, HEK *);
2521 (aux->xhv_name_u.xhvnameu_names)[count] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
2524 HEK *existing_name = aux->xhv_name_u.xhvnameu_name;
2527 (HEK_UTF8(existing_name) || (flags & SVf_UTF8))
2528 ? hek_eq_pvn_flags(aTHX_ existing_name, name, (I32)len, flags)
2529 : (HEK_LEN(existing_name) == (I32)len && memEQ(HEK_KEY(existing_name), name, len))
2532 Newx(aux->xhv_name_u.xhvnameu_names, 2, HEK *);
2533 aux->xhv_name_count = existing_name ? 2 : -2;
2534 *aux->xhv_name_u.xhvnameu_names = existing_name;
2535 (aux->xhv_name_u.xhvnameu_names)[1] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
2540 =for apidoc hv_ename_delete
2542 Removes a name from a stash's internal list of effective names. If this is
2543 the name returned by C<HvENAME>, then another name in the list will take
2544 its place (C<HvENAME> will use it).
2546 This is called when a stash is deleted from the symbol table.
2552 Perl_hv_ename_delete(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
2554 struct xpvhv_aux *aux;
2556 PERL_ARGS_ASSERT_HV_ENAME_DELETE;
2559 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
2561 if (!SvOOK(hv)) return;
2564 if (!aux->xhv_name_u.xhvnameu_name) return;
2566 if (aux->xhv_name_count) {
2567 HEK ** const namep = aux->xhv_name_u.xhvnameu_names;
2568 I32 const count = aux->xhv_name_count;
2569 HEK **victim = namep + (count < 0 ? -count : count);
2570 while (victim-- > namep + 1)
2572 (HEK_UTF8(*victim) || (flags & SVf_UTF8))
2573 ? hek_eq_pvn_flags(aTHX_ *victim, name, (I32)len, flags)
2574 : (HEK_LEN(*victim) == (I32)len && memEQ(HEK_KEY(*victim), name, len))
2576 unshare_hek_or_pvn(*victim, 0, 0, 0);
2577 aux = HvAUX(hv); /* may been realloced */
2578 if (count < 0) ++aux->xhv_name_count;
2579 else --aux->xhv_name_count;
2581 (aux->xhv_name_count == 1 || aux->xhv_name_count == -1)
2583 ) { /* if there are none left */
2585 aux->xhv_name_u.xhvnameu_names = NULL;
2586 aux->xhv_name_count = 0;
2589 /* Move the last one back to fill the empty slot. It
2590 does not matter what order they are in. */
2591 *victim = *(namep + (count < 0 ? -count : count) - 1);
2596 count > 0 && ((HEK_UTF8(*namep) || (flags & SVf_UTF8))
2597 ? hek_eq_pvn_flags(aTHX_ *namep, name, (I32)len, flags)
2598 : (HEK_LEN(*namep) == (I32)len && memEQ(HEK_KEY(*namep), name, len))
2601 aux->xhv_name_count = -count;
2605 (HEK_UTF8(aux->xhv_name_u.xhvnameu_name) || (flags & SVf_UTF8))
2606 ? hek_eq_pvn_flags(aTHX_ aux->xhv_name_u.xhvnameu_name, name, (I32)len, flags)
2607 : (HEK_LEN(aux->xhv_name_u.xhvnameu_name) == (I32)len &&
2608 memEQ(HEK_KEY(aux->xhv_name_u.xhvnameu_name), name, len))
2610 HEK * const namehek = aux->xhv_name_u.xhvnameu_name;
2611 Newx(aux->xhv_name_u.xhvnameu_names, 1, HEK *);
2612 *aux->xhv_name_u.xhvnameu_names = namehek;
2613 aux->xhv_name_count = -1;
2618 Perl_hv_backreferences_p(pTHX_ HV *hv) {
2619 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
2620 /* See also Perl_sv_get_backrefs in sv.c where this logic is unrolled */
2622 struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2623 return &(iter->xhv_backreferences);
2628 Perl_hv_kill_backrefs(pTHX_ HV *hv) {
2631 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2636 av = HvAUX(hv)->xhv_backreferences;
2639 HvAUX(hv)->xhv_backreferences = 0;
2640 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
2641 if (SvTYPE(av) == SVt_PVAV)
2642 SvREFCNT_dec_NN(av);
2647 hv_iternext is implemented as a macro in hv.h
2649 =for apidoc hv_iternext
2651 Returns entries from a hash iterator. See C<L</hv_iterinit>>.
2653 You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2654 iterator currently points to, without losing your place or invalidating your
2655 iterator. Note that in this case the current entry is deleted from the hash
2656 with your iterator holding the last reference to it. Your iterator is flagged
2657 to free the entry on the next call to C<hv_iternext>, so you must not discard
2658 your iterator immediately else the entry will leak - call C<hv_iternext> to
2659 trigger the resource deallocation.
2661 =for apidoc hv_iternext_flags
2663 Returns entries from a hash iterator. See C<L</hv_iterinit>> and
2665 The C<flags> value will normally be zero; if C<HV_ITERNEXT_WANTPLACEHOLDERS> is
2666 set the placeholders keys (for restricted hashes) will be returned in addition
2667 to normal keys. By default placeholders are automatically skipped over.
2668 Currently a placeholder is implemented with a value that is
2669 C<&PL_sv_placeholder>. Note that the implementation of placeholders and
2670 restricted hashes may change, and the implementation currently is
2671 insufficiently abstracted for any change to be tidy.
2677 Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2684 struct xpvhv_aux *iter;
2686 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2688 xhv = (XPVHV*)SvANY(hv);
2691 /* Too many things (well, pp_each at least) merrily assume that you can
2692 call hv_iternext without calling hv_iterinit, so we'll have to deal
2698 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2699 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
2700 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
2701 SV * const key = sv_newmortal();
2703 sv_setsv(key, HeSVKEY_force(entry));
2704 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
2705 HeSVKEY_set(entry, NULL);
2711 /* one HE per MAGICAL hash */
2712 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
2713 HvLAZYDEL_on(hv); /* make sure entry gets freed */
2715 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
2717 HeKEY_hek(entry) = hek;
2718 HeKLEN(entry) = HEf_SVKEY;
2720 magic_nextpack(MUTABLE_SV(hv),mg,key);
2722 /* force key to stay around until next time */
2723 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2724 return entry; /* beware, hent_val is not set */
2726 SvREFCNT_dec(HeVAL(entry));
2727 Safefree(HeKEY_hek(entry));
2729 iter = HvAUX(hv); /* may been realloced */
2730 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2735 #if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */
2736 if (!entry && SvRMAGICAL((const SV *)hv)
2737 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
2740 /* The prime_env_iter() on VMS just loaded up new hash values
2741 * so the iteration count needs to be reset back to the beginning
2745 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2750 /* hv_iterinit now ensures this. */
2751 assert (HvARRAY(hv));
2753 /* At start of hash, entry is NULL. */
2756 entry = HeNEXT(entry);
2757 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2759 * Skip past any placeholders -- don't want to include them in
2762 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
2763 entry = HeNEXT(entry);
2768 #ifdef PERL_HASH_RANDOMIZE_KEYS
2769 if (iter->xhv_last_rand != iter->xhv_rand) {
2770 if (iter->xhv_riter != -1) {
2771 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
2772 "Use of each() on hash after insertion without resetting hash iterator results in undefined behavior"
2776 iter = HvAUX(hv); /* may been realloced */
2777 iter->xhv_last_rand = iter->xhv_rand;
2781 /* Skip the entire loop if the hash is empty. */
2782 if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS)
2783 ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) {
2785 /* OK. Come to the end of the current list. Grab the next one. */
2787 iter->xhv_riter++; /* HvRITER(hv)++ */
2788 if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) {
2789 /* There is no next one. End of the hash. */
2790 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2791 #ifdef PERL_HASH_RANDOMIZE_KEYS
2792 iter->xhv_last_rand = iter->xhv_rand; /* reset xhv_last_rand so we can detect inserts during traversal */
2796 entry = (HvARRAY(hv))[ PERL_HASH_ITER_BUCKET(iter) & xhv->xhv_max ];
2798 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2799 /* If we have an entry, but it's a placeholder, don't count it.
2801 while (entry && HeVAL(entry) == &PL_sv_placeholder)
2802 entry = HeNEXT(entry);
2804 /* Will loop again if this linked list starts NULL
2805 (for HV_ITERNEXT_WANTPLACEHOLDERS)
2806 or if we run through it and find only placeholders. */
2810 iter->xhv_riter = -1;
2811 #ifdef PERL_HASH_RANDOMIZE_KEYS
2812 iter->xhv_last_rand = iter->xhv_rand;
2816 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2818 hv_free_ent(hv, oldentry);
2821 iter = HvAUX(hv); /* may been realloced */
2822 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
2827 =for apidoc hv_iterkey
2829 Returns the key from the current position of the hash iterator. See
2836 Perl_hv_iterkey(pTHX_ HE *entry, I32 *retlen)
2838 PERL_ARGS_ASSERT_HV_ITERKEY;
2840 if (HeKLEN(entry) == HEf_SVKEY) {
2842 char * const p = SvPV(HeKEY_sv(entry), len);
2847 *retlen = HeKLEN(entry);
2848 return HeKEY(entry);
2852 /* unlike hv_iterval(), this always returns a mortal copy of the key */
2854 =for apidoc hv_iterkeysv
2856 Returns the key as an C<SV*> from the current position of the hash
2857 iterator. The return value will always be a mortal copy of the key. Also
2858 see C<L</hv_iterinit>>.
2864 Perl_hv_iterkeysv(pTHX_ HE *entry)
2866 PERL_ARGS_ASSERT_HV_ITERKEYSV;
2868 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
2872 =for apidoc hv_iterval
2874 Returns the value from the current position of the hash iterator. See
2881 Perl_hv_iterval(pTHX_ HV *hv, HE *entry)
2883 PERL_ARGS_ASSERT_HV_ITERVAL;
2885 if (SvRMAGICAL(hv)) {
2886 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
2887 SV* const sv = sv_newmortal();
2888 if (HeKLEN(entry) == HEf_SVKEY)
2889 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
2891 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
2895 return HeVAL(entry);
2899 =for apidoc hv_iternextsv
2901 Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
2908 Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
2910 HE * const he = hv_iternext_flags(hv, 0);
2912 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
2916 *key = hv_iterkey(he, retlen);
2917 return hv_iterval(hv, he);
2924 =for apidoc hv_magic
2926 Adds magic to a hash. See C<L</sv_magic>>.
2931 /* possibly free a shared string if no one has access to it
2932 * len and hash must both be valid for str.
2935 Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
2937 unshare_hek_or_pvn (NULL, str, len, hash);
2942 Perl_unshare_hek(pTHX_ HEK *hek)
2945 unshare_hek_or_pvn(hek, NULL, 0, 0);
2948 /* possibly free a shared string if no one has access to it
2949 hek if non-NULL takes priority over the other 3, else str, len and hash
2950 are used. If so, len and hash must both be valid for str.
2953 S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
2958 bool is_utf8 = FALSE;
2960 const char * const save = str;
2961 struct shared_he *he = NULL;
2964 /* Find the shared he which is just before us in memory. */
2965 he = (struct shared_he *)(((char *)hek)
2966 - STRUCT_OFFSET(struct shared_he,
2969 /* Assert that the caller passed us a genuine (or at least consistent)
2971 assert (he->shared_he_he.hent_hek == hek);
2973 if (he->shared_he_he.he_valu.hent_refcount - 1) {
2974 --he->shared_he_he.he_valu.hent_refcount;
2978 hash = HEK_HASH(hek);
2979 } else if (len < 0) {
2980 STRLEN tmplen = -len;
2982 /* See the note in hv_fetch(). --jhi */
2983 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2986 k_flags = HVhek_UTF8;
2988 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2991 /* what follows was the moral equivalent of:
2992 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
2994 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
2996 xhv = (XPVHV*)SvANY(PL_strtab);
2997 /* assert(xhv_array != 0) */
2998 oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
3000 const HE *const he_he = &(he->shared_he_he);
3001 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
3006 const int flags_masked = k_flags & HVhek_MASK;
3007 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
3008 if (HeHASH(entry) != hash) /* strings can't be equal */
3010 if (HeKLEN(entry) != len)
3012 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3014 if (HeKFLAGS(entry) != flags_masked)
3021 if (--entry->he_valu.hent_refcount == 0) {
3022 *oentry = HeNEXT(entry);
3024 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
3029 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
3030 "Attempt to free nonexistent shared string '%s'%s"
3032 hek ? HEK_KEY(hek) : str,
3033 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
3034 if (k_flags & HVhek_FREEKEY)
3038 /* get a (constant) string ptr from the global string table
3039 * string will get added if it is not already there.
3040 * len and hash must both be valid for str.
3043 Perl_share_hek(pTHX_ const char *str, SSize_t len, U32 hash)
3045 bool is_utf8 = FALSE;
3047 const char * const save = str;
3049 PERL_ARGS_ASSERT_SHARE_HEK;
3052 STRLEN tmplen = -len;
3054 /* See the note in hv_fetch(). --jhi */
3055 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
3057 /* If we were able to downgrade here, then than means that we were passed
3058 in a key which only had chars 0-255, but was utf8 encoded. */
3061 /* If we found we were able to downgrade the string to bytes, then
3062 we should flag that it needs upgrading on keys or each. Also flag
3063 that we need share_hek_flags to free the string. */
3066 PERL_HASH(hash, str, len);
3067 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
3071 return share_hek_flags (str, len, hash, flags);
3075 S_share_hek_flags(pTHX_ const char *str, STRLEN len, U32 hash, int flags)
3078 const int flags_masked = flags & HVhek_MASK;
3079 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
3080 XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
3082 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
3084 if (UNLIKELY(len > (STRLEN) I32_MAX)) {
3085 Perl_croak_nocontext("Sorry, hash keys must be smaller than 2**31 bytes");
3088 /* what follows is the moral equivalent of:
3090 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
3091 hv_store(PL_strtab, str, len, NULL, hash);
3093 Can't rehash the shared string table, so not sure if it's worth
3094 counting the number of entries in the linked list
3097 /* assert(xhv_array != 0) */
3098 entry = (HvARRAY(PL_strtab))[hindex];
3099 for (;entry; entry = HeNEXT(entry)) {
3100 if (HeHASH(entry) != hash) /* strings can't be equal */
3102 if (HeKLEN(entry) != (SSize_t) len)
3104 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3106 if (HeKFLAGS(entry) != flags_masked)
3112 /* What used to be head of the list.
3113 If this is NULL, then we're the first entry for this slot, which
3114 means we need to increate fill. */
3115 struct shared_he *new_entry;
3118 HE **const head = &HvARRAY(PL_strtab)[hindex];
3119 HE *const next = *head;
3121 /* We don't actually store a HE from the arena and a regular HEK.
3122 Instead we allocate one chunk of memory big enough for both,
3123 and put the HEK straight after the HE. This way we can find the
3124 HE directly from the HEK.
3127 Newx(k, STRUCT_OFFSET(struct shared_he,
3128 shared_he_hek.hek_key[0]) + len + 2, char);
3129 new_entry = (struct shared_he *)k;
3130 entry = &(new_entry->shared_he_he);
3131 hek = &(new_entry->shared_he_hek);
3133 Copy(str, HEK_KEY(hek), len, char);
3134 HEK_KEY(hek)[len] = 0;
3136 HEK_HASH(hek) = hash;
3137 HEK_FLAGS(hek) = (unsigned char)flags_masked;
3139 /* Still "point" to the HEK, so that other code need not know what
3141 HeKEY_hek(entry) = hek;
3142 entry->he_valu.hent_refcount = 0;
3143 HeNEXT(entry) = next;
3146 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
3147 if (!next) { /* initial entry? */
3148 } else if ( DO_HSPLIT(xhv) ) {
3149 const STRLEN oldsize = xhv->xhv_max + 1;
3150 hsplit(PL_strtab, oldsize, oldsize * 2);
3154 ++entry->he_valu.hent_refcount;
3156 if (flags & HVhek_FREEKEY)
3159 return HeKEY_hek(entry);
3163 Perl_hv_placeholders_p(pTHX_ HV *hv)
3165 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
3167 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
3170 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
3173 Perl_die(aTHX_ "panic: hv_placeholders_p");
3176 return &(mg->mg_len);
3181 Perl_hv_placeholders_get(pTHX_ const HV *hv)
3183 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
3185 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
3186 PERL_UNUSED_CONTEXT;
3188 return mg ? mg->mg_len : 0;
3192 Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
3194 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
3196 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
3201 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
3202 Perl_die(aTHX_ "panic: hv_placeholders_set");
3204 /* else we don't need to add magic to record 0 placeholders. */
3208 S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
3213 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
3215 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
3220 value = &PL_sv_placeholder;
3223 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
3226 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
3229 case HVrhek_PV_UTF8:
3230 /* Create a string SV that directly points to the bytes in our
3232 value = newSV_type(SVt_PV);
3233 SvPV_set(value, (char *) he->refcounted_he_data + 1);
3234 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
3235 /* This stops anything trying to free it */
3236 SvLEN_set(value, 0);
3238 SvREADONLY_on(value);
3239 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
3243 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %" UVxf,
3244 (UV)he->refcounted_he_data[0]);
3250 =for apidoc m|HV *|refcounted_he_chain_2hv|const struct refcounted_he *c|U32 flags
3252 Generates and returns a C<HV *> representing the content of a
3253 C<refcounted_he> chain.
3254 C<flags> is currently unused and must be zero.
3259 Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain, U32 flags)
3263 U32 placeholders, max;
3266 Perl_croak(aTHX_ "panic: refcounted_he_chain_2hv bad flags %" UVxf,
3269 /* We could chase the chain once to get an idea of the number of keys,
3270 and call ksplit. But for now we'll make a potentially inefficient
3271 hash with only 8 entries in its array. */
3276 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
3277 HvARRAY(hv) = (HE**)array;
3283 U32 hash = chain->refcounted_he_hash;
3285 U32 hash = HEK_HASH(chain->refcounted_he_hek);
3287 HE **oentry = &((HvARRAY(hv))[hash & max]);
3288 HE *entry = *oentry;
3291 for (; entry; entry = HeNEXT(entry)) {
3292 if (HeHASH(entry) == hash) {
3293 /* We might have a duplicate key here. If so, entry is older
3294 than the key we've already put in the hash, so if they are
3295 the same, skip adding entry. */
3297 const STRLEN klen = HeKLEN(entry);
3298 const char *const key = HeKEY(entry);
3299 if (klen == chain->refcounted_he_keylen
3300 && (!!HeKUTF8(entry)
3301 == !!(chain->refcounted_he_data[0] & HVhek_UTF8))
3302 && memEQ(key, REF_HE_KEY(chain), klen))
3305 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
3307 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
3308 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
3309 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
3320 = share_hek_flags(REF_HE_KEY(chain),
3321 chain->refcounted_he_keylen,
3322 chain->refcounted_he_hash,
3323 (chain->refcounted_he_data[0]
3324 & (HVhek_UTF8|HVhek_WASUTF8)));
3326 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
3328 value = refcounted_he_value(chain);
3329 if (value == &PL_sv_placeholder)
3331 HeVAL(entry) = value;
3333 /* Link it into the chain. */
3334 HeNEXT(entry) = *oentry;
3340 chain = chain->refcounted_he_next;
3344 clear_placeholders(hv, placeholders);
3345 HvTOTALKEYS(hv) -= placeholders;
3348 /* We could check in the loop to see if we encounter any keys with key
3349 flags, but it's probably not worth it, as this per-hash flag is only
3350 really meant as an optimisation for things like Storable. */
3352 DEBUG_A(Perl_hv_assert(aTHX_ hv));
3358 =for apidoc m|SV *|refcounted_he_fetch_pvn|const struct refcounted_he *chain|const char *keypv|STRLEN keylen|U32 hash|U32 flags
3360 Search along a C<refcounted_he> chain for an entry with the key specified
3361 by C<keypv> and C<keylen>. If C<flags> has the C<REFCOUNTED_HE_KEY_UTF8>
3362 bit set, the key octets are interpreted as UTF-8, otherwise they
3363 are interpreted as Latin-1. C<hash> is a precomputed hash of the key
3364 string, or zero if it has not been precomputed. Returns a mortal scalar
3365 representing the value associated with the key, or C<&PL_sv_placeholder>
3366 if there is no value associated with the key.
3372 Perl_refcounted_he_fetch_pvn(pTHX_ const struct refcounted_he *chain,
3373 const char *keypv, STRLEN keylen, U32 hash, U32 flags)
3377 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PVN;
3379 if (flags & ~(REFCOUNTED_HE_KEY_UTF8|REFCOUNTED_HE_EXISTS))
3380 Perl_croak(aTHX_ "panic: refcounted_he_fetch_pvn bad flags %" UVxf,
3384 if (flags & REFCOUNTED_HE_KEY_UTF8) {
3385 /* For searching purposes, canonicalise to Latin-1 where possible. */
3386 const char *keyend = keypv + keylen, *p;
3387 STRLEN nonascii_count = 0;
3388 for (p = keypv; p != keyend; p++) {
3389 if (! UTF8_IS_INVARIANT(*p)) {
3390 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) {
3391 goto canonicalised_key;
3397 if (nonascii_count) {
3399 const char *p = keypv, *keyend = keypv + keylen;
3400 keylen -= nonascii_count;
3401 Newx(q, keylen, char);
3404 for (; p != keyend; p++, q++) {
3406 if (UTF8_IS_INVARIANT(c)) {
3411 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
3415 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3416 canonicalised_key: ;
3418 utf8_flag = (flags & REFCOUNTED_HE_KEY_UTF8) ? HVhek_UTF8 : 0;
3420 PERL_HASH(hash, keypv, keylen);
3422 for (; chain; chain = chain->refcounted_he_next) {
3425 hash == chain->refcounted_he_hash &&
3426 keylen == chain->refcounted_he_keylen &&
3427 memEQ(REF_HE_KEY(chain), keypv, keylen) &&
3428 utf8_flag == (chain->refcounted_he_data[0] & HVhek_UTF8)
3430 hash == HEK_HASH(chain->refcounted_he_hek) &&
3431 keylen == (STRLEN)HEK_LEN(chain->refcounted_he_hek) &&
3432 memEQ(HEK_KEY(chain->refcounted_he_hek), keypv, keylen) &&
3433 utf8_flag == (HEK_FLAGS(chain->refcounted_he_hek) & HVhek_UTF8)
3436 if (flags & REFCOUNTED_HE_EXISTS)
3437 return (chain->refcounted_he_data[0] & HVrhek_typemask)
3439 ? NULL : &PL_sv_yes;
3440 return sv_2mortal(refcounted_he_value(chain));
3444 return flags & REFCOUNTED_HE_EXISTS ? NULL : &PL_sv_placeholder;
3448 =for apidoc m|SV *|refcounted_he_fetch_pv|const struct refcounted_he *chain|const char *key|U32 hash|U32 flags
3450 Like L</refcounted_he_fetch_pvn>, but takes a nul-terminated string
3451 instead of a string/length pair.
3457 Perl_refcounted_he_fetch_pv(pTHX_ const struct refcounted_he *chain,
3458 const char *key, U32 hash, U32 flags)
3460 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PV;
3461 return refcounted_he_fetch_pvn(chain, key, strlen(key), hash, flags);
3465 =for apidoc m|SV *|refcounted_he_fetch_sv|const struct refcounted_he *chain|SV *key|U32 hash|U32 flags
3467 Like L</refcounted_he_fetch_pvn>, but takes a Perl scalar instead of a
3474 Perl_refcounted_he_fetch_sv(pTHX_ const struct refcounted_he *chain,
3475 SV *key, U32 hash, U32 flags)
3479 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_SV;
3480 if (flags & REFCOUNTED_HE_KEY_UTF8)
3481 Perl_croak(aTHX_ "panic: refcounted_he_fetch_sv bad flags %" UVxf,
3483 keypv = SvPV_const(key, keylen);
3485 flags |= REFCOUNTED_HE_KEY_UTF8;
3486 if (!hash && SvIsCOW_shared_hash(key))
3487 hash = SvSHARED_HASH(key);
3488 return refcounted_he_fetch_pvn(chain, keypv, keylen, hash, flags);
3492 =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
3494 Creates a new C<refcounted_he>. This consists of a single key/value
3495 pair and a reference to an existing C<refcounted_he> chain (which may
3496 be empty), and thus forms a longer chain. When using the longer chain,
3497 the new key/value pair takes precedence over any entry for the same key
3498 further along the chain.
3500 The new key is specified by C<keypv> and C<keylen>. If C<flags> has
3501 the C<REFCOUNTED_HE_KEY_UTF8> bit set, the key octets are interpreted
3502 as UTF-8, otherwise they are interpreted as Latin-1. C<hash> is
3503 a precomputed hash of the key string, or zero if it has not been
3506 C<value> is the scalar value to store for this key. C<value> is copied
3507 by this function, which thus does not take ownership of any reference
3508 to it, and later changes to the scalar will not be reflected in the
3509 value visible in the C<refcounted_he>. Complex types of scalar will not
3510 be stored with referential integrity, but will be coerced to strings.
3511 C<value> may be either null or C<&PL_sv_placeholder> to indicate that no
3512 value is to be associated with the key; this, as with any non-null value,
3513 takes precedence over the existence of a value for the key further along
3516 C<parent> points to the rest of the C<refcounted_he> chain to be
3517 attached to the new C<refcounted_he>. This function takes ownership
3518 of one reference to C<parent>, and returns one reference to the new
3524 struct refcounted_he *
3525 Perl_refcounted_he_new_pvn(pTHX_ struct refcounted_he *parent,
3526 const char *keypv, STRLEN keylen, U32 hash, SV *value, U32 flags)
3529 STRLEN value_len = 0;
3530 const char *value_p = NULL;
3534 STRLEN key_offset = 1;
3535 struct refcounted_he *he;
3536 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PVN;
3538 if (!value || value == &PL_sv_placeholder) {
3539 value_type = HVrhek_delete;
3540 } else if (SvPOK(value)) {
3541 value_type = HVrhek_PV;
3542 } else if (SvIOK(value)) {
3543 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
3544 } else if (!SvOK(value)) {
3545 value_type = HVrhek_undef;
3547 value_type = HVrhek_PV;
3549 is_pv = value_type == HVrhek_PV;
3551 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
3552 the value is overloaded, and doesn't yet have the UTF-8flag set. */
3553 value_p = SvPV_const(value, value_len);
3555 value_type = HVrhek_PV_UTF8;
3556 key_offset = value_len + 2;
3558 hekflags = value_type;
3560 if (flags & REFCOUNTED_HE_KEY_UTF8) {
3561 /* Canonicalise to Latin-1 where possible. */
3562 const char *keyend = keypv + keylen, *p;
3563 STRLEN nonascii_count = 0;
3564 for (p = keypv; p != keyend; p++) {
3565 if (! UTF8_IS_INVARIANT(*p)) {
3566 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) {
3567 goto canonicalised_key;
3573 if (nonascii_count) {
3575 const char *p = keypv, *keyend = keypv + keylen;
3576 keylen -= nonascii_count;
3577 Newx(q, keylen, char);
3580 for (; p != keyend; p++, q++) {
3582 if (UTF8_IS_INVARIANT(c)) {
3587 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
3591 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3592 canonicalised_key: ;
3594 if (flags & REFCOUNTED_HE_KEY_UTF8)
3595 hekflags |= HVhek_UTF8;
3597 PERL_HASH(hash, keypv, keylen);
3600 he = (struct refcounted_he*)
3601 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3605 he = (struct refcounted_he*)
3606 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3610 he->refcounted_he_next = parent;
3613 Copy(value_p, he->refcounted_he_data + 1, value_len + 1, char);
3614 he->refcounted_he_val.refcounted_he_u_len = value_len;
3615 } else if (value_type == HVrhek_IV) {
3616 he->refcounted_he_val.refcounted_he_u_iv = SvIVX(value);
3617 } else if (value_type == HVrhek_UV) {
3618 he->refcounted_he_val.refcounted_he_u_uv = SvUVX(value);
3622 he->refcounted_he_hash = hash;
3623 he->refcounted_he_keylen = keylen;
3624 Copy(keypv, he->refcounted_he_data + key_offset, keylen, char);
3626 he->refcounted_he_hek = share_hek_flags(keypv, keylen, hash, hekflags);
3629 he->refcounted_he_data[0] = hekflags;
3630 he->refcounted_he_refcnt = 1;
3636 =for apidoc m|struct refcounted_he *|refcounted_he_new_pv|struct refcounted_he *parent|const char *key|U32 hash|SV *value|U32 flags
3638 Like L</refcounted_he_new_pvn>, but takes a nul-terminated string instead
3639 of a string/length pair.
3644 struct refcounted_he *
3645 Perl_refcounted_he_new_pv(pTHX_ struct refcounted_he *parent,
3646 const char *key, U32 hash, SV *value, U32 flags)
3648 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PV;
3649 return refcounted_he_new_pvn(parent, key, strlen(key), hash, value, flags);
3653 =for apidoc m|struct refcounted_he *|refcounted_he_new_sv|struct refcounted_he *parent|SV *key|U32 hash|SV *value|U32 flags
3655 Like L</refcounted_he_new_pvn>, but takes a Perl scalar instead of a
3661 struct refcounted_he *
3662 Perl_refcounted_he_new_sv(pTHX_ struct refcounted_he *parent,
3663 SV *key, U32 hash, SV *value, U32 flags)
3667 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_SV;
3668 if (flags & REFCOUNTED_HE_KEY_UTF8)
3669 Perl_croak(aTHX_ "panic: refcounted_he_new_sv bad flags %" UVxf,
3671 keypv = SvPV_const(key, keylen);
3673 flags |= REFCOUNTED_HE_KEY_UTF8;
3674 if (!hash && SvIsCOW_shared_hash(key))
3675 hash = SvSHARED_HASH(key);
3676 return refcounted_he_new_pvn(parent, keypv, keylen, hash, value, flags);
3680 =for apidoc m|void|refcounted_he_free|struct refcounted_he *he
3682 Decrements the reference count of a C<refcounted_he> by one. If the
3683 reference count reaches zero the structure's memory is freed, which
3684 (recursively) causes a reduction of its parent C<refcounted_he>'s
3685 reference count. It is safe to pass a null pointer to this function:
3686 no action occurs in this case.
3692 Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
3696 PERL_UNUSED_CONTEXT;
3699 struct refcounted_he *copy;
3703 new_count = --he->refcounted_he_refcnt;
3704 HINTS_REFCNT_UNLOCK;
3710 #ifndef USE_ITHREADS
3711 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
3714 he = he->refcounted_he_next;
3715 PerlMemShared_free(copy);
3720 =for apidoc m|struct refcounted_he *|refcounted_he_inc|struct refcounted_he *he
3722 Increment the reference count of a C<refcounted_he>. The pointer to the
3723 C<refcounted_he> is also returned. It is safe to pass a null pointer
3724 to this function: no action occurs and a null pointer is returned.
3729 struct refcounted_he *
3730 Perl_refcounted_he_inc(pTHX_ struct refcounted_he *he)
3735 PERL_UNUSED_CONTEXT;
3738 he->refcounted_he_refcnt++;
3739 HINTS_REFCNT_UNLOCK;
3745 =for apidoc cop_fetch_label
3747 Returns the label attached to a cop.
3748 The flags pointer may be set to C<SVf_UTF8> or 0.
3753 /* pp_entereval is aware that labels are stored with a key ':' at the top of
3756 Perl_cop_fetch_label(pTHX_ COP *const cop, STRLEN *len, U32 *flags) {
3757 struct refcounted_he *const chain = cop->cop_hints_hash;
3759 PERL_ARGS_ASSERT_COP_FETCH_LABEL;
3760 PERL_UNUSED_CONTEXT;
3765 if (chain->refcounted_he_keylen != 1)
3767 if (*REF_HE_KEY(chain) != ':')
3770 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
3772 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
3775 /* Stop anyone trying to really mess us up by adding their own value for
3777 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
3778 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
3782 *len = chain->refcounted_he_val.refcounted_he_u_len;
3784 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
3785 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
3787 return chain->refcounted_he_data + 1;
3791 =for apidoc cop_store_label
3793 Save a label into a C<cop_hints_hash>.
3794 You need to set flags to C<SVf_UTF8>
3801 Perl_cop_store_label(pTHX_ COP *const cop, const char *label, STRLEN len,
3805 PERL_ARGS_ASSERT_COP_STORE_LABEL;
3807 if (flags & ~(SVf_UTF8))
3808 Perl_croak(aTHX_ "panic: cop_store_label illegal flag bits 0x%" UVxf,
3810 labelsv = newSVpvn_flags(label, len, SVs_TEMP);
3811 if (flags & SVf_UTF8)
3814 = refcounted_he_new_pvs(cop->cop_hints_hash, ":", labelsv, 0);
3818 =for apidoc hv_assert
3820 Check that a hash is in an internally consistent state.
3828 Perl_hv_assert(pTHX_ HV *hv)
3833 int placeholders = 0;
3836 const I32 riter = HvRITER_get(hv);
3837 HE *eiter = HvEITER_get(hv);
3839 PERL_ARGS_ASSERT_HV_ASSERT;
3841 (void)hv_iterinit(hv);
3843 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
3844 /* sanity check the values */
3845 if (HeVAL(entry) == &PL_sv_placeholder)
3849 /* sanity check the keys */
3850 if (HeSVKEY(entry)) {
3851 NOOP; /* Don't know what to check on SV keys. */
3852 } else if (HeKUTF8(entry)) {
3854 if (HeKWASUTF8(entry)) {
3855 PerlIO_printf(Perl_debug_log,
3856 "hash key has both WASUTF8 and UTF8: '%.*s'\n",
3857 (int) HeKLEN(entry), HeKEY(entry));
3860 } else if (HeKWASUTF8(entry))
3863 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
3864 static const char bad_count[] = "Count %d %s(s), but hash reports %d\n";
3865 const int nhashkeys = HvUSEDKEYS(hv);
3866 const int nhashplaceholders = HvPLACEHOLDERS_get(hv);
3868 if (nhashkeys != real) {
3869 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
3872 if (nhashplaceholders != placeholders) {
3873 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
3877 if (withflags && ! HvHASKFLAGS(hv)) {
3878 PerlIO_printf(Perl_debug_log,
3879 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
3884 sv_dump(MUTABLE_SV(hv));
3886 HvRITER_set(hv, riter); /* Restore hash iterator state */
3887 HvEITER_set(hv, eiter);
3893 * ex: set ts=8 sts=4 sw=4 et: