3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * I sit beside the fire and think
13 * of all that I have seen.
16 * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
20 =head1 Hash Manipulation Functions
22 A HV structure represents a Perl hash. It consists mainly of an array
23 of pointers, each of which points to a linked list of HE structures. The
24 array is indexed by the hash function of the key, so each linked list
25 represents all the hash entries with the same hash value. Each HE contains
26 a pointer to the actual value, plus a pointer to a HEK structure which
27 holds the key and hash value.
35 #define PERL_HASH_INTERNAL_ACCESS
38 #define HV_MAX_LENGTH_BEFORE_SPLIT 14
40 static const char S_strtab_error[]
41 = "Cannot modify shared string table in hv_%s";
45 #define new_HE() (HE*)safemalloc(sizeof(HE))
46 #define del_HE(p) safefree((char*)p)
55 void ** const root = &PL_body_roots[HE_SVSLOT];
58 Perl_more_bodies(aTHX_ HE_SVSLOT, sizeof(HE), PERL_ARENA_SIZE);
65 #define new_HE() new_he()
68 HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \
69 PL_body_roots[HE_SVSLOT] = p; \
77 S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
79 const int flags_masked = flags & HVhek_MASK;
83 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
85 Newx(k, HEK_BASESIZE + len + 2, char);
87 Copy(str, HEK_KEY(hek), len, char);
88 HEK_KEY(hek)[len] = 0;
91 HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED;
93 if (flags & HVhek_FREEKEY)
98 /* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
102 Perl_free_tied_hv_pool(pTHX)
105 HE *he = PL_hv_fetch_ent_mh;
108 Safefree(HeKEY_hek(he));
112 PL_hv_fetch_ent_mh = NULL;
115 #if defined(USE_ITHREADS)
117 Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
121 PERL_ARGS_ASSERT_HEK_DUP;
122 PERL_UNUSED_ARG(param);
127 shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
129 /* We already shared this hash key. */
130 (void)share_hek_hek(shared);
134 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
135 HEK_HASH(source), HEK_FLAGS(source));
136 ptr_table_store(PL_ptr_table, source, shared);
142 Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
146 PERL_ARGS_ASSERT_HE_DUP;
150 /* look for it in the table first */
151 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
155 /* create anew and remember what it is */
157 ptr_table_store(PL_ptr_table, e, ret);
159 HeNEXT(ret) = he_dup(HeNEXT(e),shared, param);
160 if (HeKLEN(e) == HEf_SVKEY) {
162 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
163 HeKEY_hek(ret) = (HEK*)k;
164 HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param);
167 /* This is hek_dup inlined, which seems to be important for speed
169 HEK * const source = HeKEY_hek(e);
170 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
173 /* We already shared this hash key. */
174 (void)share_hek_hek(shared);
178 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
179 HEK_HASH(source), HEK_FLAGS(source));
180 ptr_table_store(PL_ptr_table, source, shared);
182 HeKEY_hek(ret) = shared;
185 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
187 HeVAL(ret) = sv_dup_inc(HeVAL(e), param);
190 #endif /* USE_ITHREADS */
193 S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
196 SV * const sv = sv_newmortal();
198 PERL_ARGS_ASSERT_HV_NOTALLOWED;
200 if (!(flags & HVhek_FREEKEY)) {
201 sv_setpvn(sv, key, klen);
204 /* Need to free saved eventually assign to mortal SV */
205 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
206 sv_usepvn(sv, (char *) key, klen);
208 if (flags & HVhek_UTF8) {
211 Perl_croak(aTHX_ msg, SVfARG(sv));
214 /* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
220 Stores an SV in a hash. The hash key is specified as C<key> and C<klen> is
221 the length of the key. The C<hash> parameter is the precomputed hash
222 value; if it is zero then Perl will compute it. The return value will be
223 NULL if the operation failed or if the value did not need to be actually
224 stored within the hash (as in the case of tied hashes). Otherwise it can
225 be dereferenced to get the original C<SV*>. Note that the caller is
226 responsible for suitably incrementing the reference count of C<val> before
227 the call, and decrementing it if the function returned NULL. Effectively
228 a successful hv_store takes ownership of one reference to C<val>. This is
229 usually what you want; a newly created SV has a reference count of one, so
230 if all your code does is create SVs then store them in a hash, hv_store
231 will own the only reference to the new SV, and your code doesn't need to do
232 anything further to tidy up. hv_store is not implemented as a call to
233 hv_store_ent, and does not create a temporary SV for the key, so if your
234 key data is not already in SV form then use hv_store in preference to
237 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
238 information on how to use this function on tied hashes.
240 =for apidoc hv_store_ent
242 Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
243 parameter is the precomputed hash value; if it is zero then Perl will
244 compute it. The return value is the new hash entry so created. It will be
245 NULL if the operation failed or if the value did not need to be actually
246 stored within the hash (as in the case of tied hashes). Otherwise the
247 contents of the return value can be accessed using the C<He?> macros
248 described here. Note that the caller is responsible for suitably
249 incrementing the reference count of C<val> before the call, and
250 decrementing it if the function returned NULL. Effectively a successful
251 hv_store_ent takes ownership of one reference to C<val>. This is
252 usually what you want; a newly created SV has a reference count of one, so
253 if all your code does is create SVs then store them in a hash, hv_store
254 will own the only reference to the new SV, and your code doesn't need to do
255 anything further to tidy up. Note that hv_store_ent only reads the C<key>;
256 unlike C<val> it does not take ownership of it, so maintaining the correct
257 reference count on C<key> is entirely the caller's responsibility. hv_store
258 is not implemented as a call to hv_store_ent, and does not create a temporary
259 SV for the key, so if your key data is not already in SV form then use
260 hv_store in preference to hv_store_ent.
262 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
263 information on how to use this function on tied hashes.
265 =for apidoc hv_exists
267 Returns a boolean indicating whether the specified hash key exists. The
268 C<klen> is the length of the key.
272 Returns the SV which corresponds to the specified key in the hash. The
273 C<klen> is the length of the key. If C<lval> is set then the fetch will be
274 part of a store. Check that the return value is non-null before
275 dereferencing it to an C<SV*>.
277 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
278 information on how to use this function on tied hashes.
280 =for apidoc hv_exists_ent
282 Returns a boolean indicating whether the specified hash key exists. C<hash>
283 can be a valid precomputed hash value, or 0 to ask for it to be
289 /* returns an HE * structure with the all fields set */
290 /* note that hent_val will be a mortal sv for MAGICAL hashes */
292 =for apidoc hv_fetch_ent
294 Returns the hash entry which corresponds to the specified key in the hash.
295 C<hash> must be a valid precomputed hash number for the given C<key>, or 0
296 if you want the function to compute it. IF C<lval> is set then the fetch
297 will be part of a store. Make sure the return value is non-null before
298 accessing it. The return value when C<tb> is a tied hash is a pointer to a
299 static location, so be sure to make a copy of the structure if you need to
302 See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
303 information on how to use this function on tied hashes.
308 /* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
310 Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
311 const int action, SV *val, const U32 hash)
316 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
325 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
329 Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
330 int flags, int action, SV *val, register U32 hash)
339 const int return_svp = action & HV_FETCH_JUST_SV;
343 if (SvTYPE(hv) == SVTYPEMASK)
346 assert(SvTYPE(hv) == SVt_PVHV);
348 if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) {
350 if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) {
351 struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr;
352 if (uf->uf_set == NULL) {
353 SV* obj = mg->mg_obj;
356 keysv = newSVpvn_flags(key, klen, SVs_TEMP |
357 ((flags & HVhek_UTF8)
361 mg->mg_obj = keysv; /* pass key */
362 uf->uf_index = action; /* pass action */
363 magic_getuvar(MUTABLE_SV(hv), mg);
364 keysv = mg->mg_obj; /* may have changed */
367 /* If the key may have changed, then we need to invalidate
368 any passed-in computed hash value. */
374 if (flags & HVhek_FREEKEY)
376 key = SvPV_const(keysv, klen);
377 is_utf8 = (SvUTF8(keysv) != 0);
378 if (SvIsCOW_shared_hash(keysv)) {
379 flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0);
384 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
387 if (action & HV_DELETE) {
388 return (void *) hv_delete_common(hv, keysv, key, klen,
389 flags | (is_utf8 ? HVhek_UTF8 : 0),
393 xhv = (XPVHV*)SvANY(hv);
395 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
396 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
397 || SvGMAGICAL((const SV *)hv))
399 /* FIXME should be able to skimp on the HE/HEK here when
400 HV_FETCH_JUST_SV is true. */
402 keysv = newSVpvn_utf8(key, klen, is_utf8);
404 keysv = newSVsv(keysv);
407 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
409 /* grab a fake HE/HEK pair from the pool or make a new one */
410 entry = PL_hv_fetch_ent_mh;
412 PL_hv_fetch_ent_mh = HeNEXT(entry);
416 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
417 HeKEY_hek(entry) = (HEK*)k;
419 HeNEXT(entry) = NULL;
420 HeSVKEY_set(entry, keysv);
422 sv_upgrade(sv, SVt_PVLV);
424 /* so we can free entry when freeing sv */
425 LvTARG(sv) = MUTABLE_SV(entry);
427 /* XXX remove at some point? */
428 if (flags & HVhek_FREEKEY)
432 return entry ? (void *) &HeVAL(entry) : NULL;
434 return (void *) entry;
436 #ifdef ENV_IS_CASELESS
437 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
439 for (i = 0; i < klen; ++i)
440 if (isLOWER(key[i])) {
441 /* Would be nice if we had a routine to do the
442 copy and upercase in a single pass through. */
443 const char * const nkey = strupr(savepvn(key,klen));
444 /* Note that this fetch is for nkey (the uppercased
445 key) whereas the store is for key (the original) */
446 void *result = hv_common(hv, NULL, nkey, klen,
447 HVhek_FREEKEY, /* free nkey */
448 0 /* non-LVAL fetch */
449 | HV_DISABLE_UVAR_XKEY
452 0 /* compute hash */);
453 if (!result && (action & HV_FETCH_LVALUE)) {
454 /* This call will free key if necessary.
455 Do it this way to encourage compiler to tail
457 result = hv_common(hv, keysv, key, klen, flags,
459 | HV_DISABLE_UVAR_XKEY
463 if (flags & HVhek_FREEKEY)
471 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
472 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
473 || SvGMAGICAL((const SV *)hv)) {
474 /* I don't understand why hv_exists_ent has svret and sv,
475 whereas hv_exists only had one. */
476 SV * const svret = sv_newmortal();
479 if (keysv || is_utf8) {
481 keysv = newSVpvn_utf8(key, klen, TRUE);
483 keysv = newSVsv(keysv);
485 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
487 mg_copy(MUTABLE_SV(hv), sv, key, klen);
489 if (flags & HVhek_FREEKEY)
491 magic_existspack(svret, mg_find(sv, PERL_MAGIC_tiedelem));
492 /* This cast somewhat evil, but I'm merely using NULL/
493 not NULL to return the boolean exists.
494 And I know hv is not NULL. */
495 return SvTRUE(svret) ? (void *)hv : NULL;
497 #ifdef ENV_IS_CASELESS
498 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
499 /* XXX This code isn't UTF8 clean. */
500 char * const keysave = (char * const)key;
501 /* Will need to free this, so set FREEKEY flag. */
502 key = savepvn(key,klen);
503 key = (const char*)strupr((char*)key);
508 if (flags & HVhek_FREEKEY) {
511 flags |= HVhek_FREEKEY;
515 else if (action & HV_FETCH_ISSTORE) {
518 hv_magic_check (hv, &needs_copy, &needs_store);
520 const bool save_taint = PL_tainted;
521 if (keysv || is_utf8) {
523 keysv = newSVpvn_utf8(key, klen, TRUE);
526 PL_tainted = SvTAINTED(keysv);
527 keysv = sv_2mortal(newSVsv(keysv));
528 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
530 mg_copy(MUTABLE_SV(hv), val, key, klen);
533 TAINT_IF(save_taint);
535 if (flags & HVhek_FREEKEY)
539 #ifdef ENV_IS_CASELESS
540 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
541 /* XXX This code isn't UTF8 clean. */
542 const char *keysave = key;
543 /* Will need to free this, so set FREEKEY flag. */
544 key = savepvn(key,klen);
545 key = (const char*)strupr((char*)key);
550 if (flags & HVhek_FREEKEY) {
553 flags |= HVhek_FREEKEY;
561 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
562 #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
563 || (SvRMAGICAL((const SV *)hv)
564 && mg_find((const SV *)hv, PERL_MAGIC_env))
569 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
571 HvARRAY(hv) = (HE**)array;
573 #ifdef DYNAMIC_ENV_FETCH
574 else if (action & HV_FETCH_ISEXISTS) {
575 /* for an %ENV exists, if we do an insert it's by a recursive
576 store call, so avoid creating HvARRAY(hv) right now. */
580 /* XXX remove at some point? */
581 if (flags & HVhek_FREEKEY)
588 if (is_utf8 & !(flags & HVhek_KEYCANONICAL)) {
589 char * const keysave = (char *)key;
590 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
594 flags &= ~HVhek_UTF8;
595 if (key != keysave) {
596 if (flags & HVhek_FREEKEY)
598 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
599 /* If the caller calculated a hash, it was on the sequence of
600 octets that are the UTF-8 form. We've now changed the sequence
601 of octets stored to that of the equivalent byte representation,
602 so the hash we need is different. */
608 PERL_HASH_INTERNAL(hash, key, klen);
609 /* We don't have a pointer to the hv, so we have to replicate the
610 flag into every HEK, so that hv_iterkeysv can see it. */
611 /* And yes, you do need this even though you are not "storing" because
612 you can flip the flags below if doing an lval lookup. (And that
613 was put in to give the semantics Andreas was expecting.) */
614 flags |= HVhek_REHASH;
616 if (keysv && (SvIsCOW_shared_hash(keysv))) {
617 hash = SvSHARED_HASH(keysv);
619 PERL_HASH(hash, key, klen);
623 masked_flags = (flags & HVhek_MASK);
625 #ifdef DYNAMIC_ENV_FETCH
626 if (!HvARRAY(hv)) entry = NULL;
630 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
632 for (; entry; entry = HeNEXT(entry)) {
633 if (HeHASH(entry) != hash) /* strings can't be equal */
635 if (HeKLEN(entry) != (I32)klen)
637 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
639 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
642 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
643 if (HeKFLAGS(entry) != masked_flags) {
644 /* We match if HVhek_UTF8 bit in our flags and hash key's
645 match. But if entry was set previously with HVhek_WASUTF8
646 and key now doesn't (or vice versa) then we should change
647 the key's flag, as this is assignment. */
648 if (HvSHAREKEYS(hv)) {
649 /* Need to swap the key we have for a key with the flags we
650 need. As keys are shared we can't just write to the
651 flag, so we share the new one, unshare the old one. */
652 HEK * const new_hek = share_hek_flags(key, klen, hash,
654 unshare_hek (HeKEY_hek(entry));
655 HeKEY_hek(entry) = new_hek;
657 else if (hv == PL_strtab) {
658 /* PL_strtab is usually the only hash without HvSHAREKEYS,
659 so putting this test here is cheap */
660 if (flags & HVhek_FREEKEY)
662 Perl_croak(aTHX_ S_strtab_error,
663 action & HV_FETCH_LVALUE ? "fetch" : "store");
666 HeKFLAGS(entry) = masked_flags;
667 if (masked_flags & HVhek_ENABLEHVKFLAGS)
670 if (HeVAL(entry) == &PL_sv_placeholder) {
671 /* yes, can store into placeholder slot */
672 if (action & HV_FETCH_LVALUE) {
674 /* This preserves behaviour with the old hv_fetch
675 implementation which at this point would bail out
676 with a break; (at "if we find a placeholder, we
677 pretend we haven't found anything")
679 That break mean that if a placeholder were found, it
680 caused a call into hv_store, which in turn would
681 check magic, and if there is no magic end up pretty
682 much back at this point (in hv_store's code). */
685 /* LVAL fetch which actaully needs a store. */
687 HvPLACEHOLDERS(hv)--;
690 if (val != &PL_sv_placeholder)
691 HvPLACEHOLDERS(hv)--;
694 } else if (action & HV_FETCH_ISSTORE) {
695 SvREFCNT_dec(HeVAL(entry));
698 } else if (HeVAL(entry) == &PL_sv_placeholder) {
699 /* if we find a placeholder, we pretend we haven't found
703 if (flags & HVhek_FREEKEY)
706 return entry ? (void *) &HeVAL(entry) : NULL;
710 #ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
711 if (!(action & HV_FETCH_ISSTORE)
712 && SvRMAGICAL((const SV *)hv)
713 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
715 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
717 sv = newSVpvn(env,len);
719 return hv_common(hv, keysv, key, klen, flags,
720 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
726 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
727 hv_notallowed(flags, key, klen,
728 "Attempt to access disallowed key '%"SVf"' in"
729 " a restricted hash");
731 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
732 /* Not doing some form of store, so return failure. */
733 if (flags & HVhek_FREEKEY)
737 if (action & HV_FETCH_LVALUE) {
740 /* At this point the old hv_fetch code would call to hv_store,
741 which in turn might do some tied magic. So we need to make that
742 magic check happen. */
743 /* gonna assign to this, so it better be there */
744 /* If a fetch-as-store fails on the fetch, then the action is to
745 recurse once into "hv_store". If we didn't do this, then that
746 recursive call would call the key conversion routine again.
747 However, as we replace the original key with the converted
748 key, this would result in a double conversion, which would show
749 up as a bug if the conversion routine is not idempotent. */
750 return hv_common(hv, keysv, key, klen, flags,
751 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
753 /* XXX Surely that could leak if the fetch-was-store fails?
754 Just like the hv_fetch. */
758 /* Welcome to hv_store... */
761 /* Not sure if we can get here. I think the only case of oentry being
762 NULL is for %ENV with dynamic env fetch. But that should disappear
763 with magic in the previous code. */
766 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
768 HvARRAY(hv) = (HE**)array;
771 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
774 /* share_hek_flags will do the free for us. This might be considered
777 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
778 else if (hv == PL_strtab) {
779 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
780 this test here is cheap */
781 if (flags & HVhek_FREEKEY)
783 Perl_croak(aTHX_ S_strtab_error,
784 action & HV_FETCH_LVALUE ? "fetch" : "store");
786 else /* gotta do the real thing */
787 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
789 HeNEXT(entry) = *oentry;
792 if (val == &PL_sv_placeholder)
793 HvPLACEHOLDERS(hv)++;
794 if (masked_flags & HVhek_ENABLEHVKFLAGS)
798 const HE *counter = HeNEXT(entry);
800 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
801 if (!counter) { /* initial entry? */
802 } else if (xhv->xhv_keys > xhv->xhv_max) {
804 } else if(!HvREHASH(hv)) {
807 while ((counter = HeNEXT(counter)))
810 if (n_links > HV_MAX_LENGTH_BEFORE_SPLIT) {
811 /* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit
812 bucket splits on a rehashed hash, as we're not going to
813 split it again, and if someone is lucky (evil) enough to
814 get all the keys in one list they could exhaust our memory
815 as we repeatedly double the number of buckets on every
816 entry. Linear search feels a less worse thing to do. */
823 return entry ? (void *) &HeVAL(entry) : NULL;
825 return (void *) entry;
829 S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
831 const MAGIC *mg = SvMAGIC(hv);
833 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
838 if (isUPPER(mg->mg_type)) {
840 if (mg->mg_type == PERL_MAGIC_tied) {
841 *needs_store = FALSE;
842 return; /* We've set all there is to set. */
845 mg = mg->mg_moremagic;
850 =for apidoc hv_scalar
852 Evaluates the hash in scalar context and returns the result. Handles magic when the hash is tied.
858 Perl_hv_scalar(pTHX_ HV *hv)
862 PERL_ARGS_ASSERT_HV_SCALAR;
864 if (SvRMAGICAL(hv)) {
865 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
867 return magic_scalarpack(hv, mg);
871 if (HvTOTALKEYS((const HV *)hv))
872 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
873 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
881 =for apidoc hv_delete
883 Deletes a key/value pair in the hash. The value SV is removed from the
884 hash and returned to the caller. The C<klen> is the length of the key.
885 The C<flags> value will normally be zero; if set to G_DISCARD then NULL
888 =for apidoc hv_delete_ent
890 Deletes a key/value pair in the hash. The value SV is removed from the
891 hash and returned to the caller. The C<flags> value will normally be zero;
892 if set to G_DISCARD then NULL will be returned. C<hash> can be a valid
893 precomputed hash value, or 0 to ask for it to be computed.
899 S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
900 int k_flags, I32 d_flags, U32 hash)
905 register HE **oentry;
906 HE *const *first_entry;
907 bool is_utf8 = (k_flags & HVhek_UTF8) ? TRUE : FALSE;
910 if (SvRMAGICAL(hv)) {
913 hv_magic_check (hv, &needs_copy, &needs_store);
917 entry = (HE *) hv_common(hv, keysv, key, klen,
918 k_flags & ~HVhek_FREEKEY,
919 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
921 sv = entry ? HeVAL(entry) : NULL;
927 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
928 /* No longer an element */
929 sv_unmagic(sv, PERL_MAGIC_tiedelem);
932 return NULL; /* element cannot be deleted */
934 #ifdef ENV_IS_CASELESS
935 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
936 /* XXX This code isn't UTF8 clean. */
937 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
938 if (k_flags & HVhek_FREEKEY) {
941 key = strupr(SvPVX(keysv));
950 xhv = (XPVHV*)SvANY(hv);
955 const char * const keysave = key;
956 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
959 k_flags |= HVhek_UTF8;
961 k_flags &= ~HVhek_UTF8;
962 if (key != keysave) {
963 if (k_flags & HVhek_FREEKEY) {
964 /* This shouldn't happen if our caller does what we expect,
965 but strictly the API allows it. */
968 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
970 HvHASKFLAGS_on(MUTABLE_SV(hv));
974 PERL_HASH_INTERNAL(hash, key, klen);
976 if (keysv && (SvIsCOW_shared_hash(keysv))) {
977 hash = SvSHARED_HASH(keysv);
979 PERL_HASH(hash, key, klen);
983 masked_flags = (k_flags & HVhek_MASK);
985 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
987 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
989 if (HeHASH(entry) != hash) /* strings can't be equal */
991 if (HeKLEN(entry) != (I32)klen)
993 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
995 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
998 if (hv == PL_strtab) {
999 if (k_flags & HVhek_FREEKEY)
1001 Perl_croak(aTHX_ S_strtab_error, "delete");
1004 /* if placeholder is here, it's already been deleted.... */
1005 if (HeVAL(entry) == &PL_sv_placeholder) {
1006 if (k_flags & HVhek_FREEKEY)
1010 if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1011 hv_notallowed(k_flags, key, klen,
1012 "Attempt to delete readonly key '%"SVf"' from"
1013 " a restricted hash");
1015 if (k_flags & HVhek_FREEKEY)
1018 if (d_flags & G_DISCARD)
1021 sv = sv_2mortal(HeVAL(entry));
1022 HeVAL(entry) = &PL_sv_placeholder;
1026 * If a restricted hash, rather than really deleting the entry, put
1027 * a placeholder there. This marks the key as being "approved", so
1028 * we can still access via not-really-existing key without raising
1031 if (SvREADONLY(hv)) {
1032 SvREFCNT_dec(HeVAL(entry));
1033 HeVAL(entry) = &PL_sv_placeholder;
1034 /* We'll be saving this slot, so the number of allocated keys
1035 * doesn't go down, but the number placeholders goes up */
1036 HvPLACEHOLDERS(hv)++;
1038 *oentry = HeNEXT(entry);
1039 if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */)
1042 hv_free_ent(hv, entry);
1043 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
1044 if (xhv->xhv_keys == 0)
1045 HvHASKFLAGS_off(hv);
1049 if (SvREADONLY(hv)) {
1050 hv_notallowed(k_flags, key, klen,
1051 "Attempt to delete disallowed key '%"SVf"' from"
1052 " a restricted hash");
1055 if (k_flags & HVhek_FREEKEY)
1061 S_hsplit(pTHX_ HV *hv)
1064 register XPVHV* const xhv = (XPVHV*)SvANY(hv);
1065 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
1066 register I32 newsize = oldsize * 2;
1068 char *a = (char*) HvARRAY(hv);
1070 int longest_chain = 0;
1073 PERL_ARGS_ASSERT_HSPLIT;
1075 /*PerlIO_printf(PerlIO_stderr(), "hsplit called for %p which had %d\n",
1076 (void*)hv, (int) oldsize);*/
1078 if (HvPLACEHOLDERS_get(hv) && !SvREADONLY(hv)) {
1079 /* Can make this clear any placeholders first for non-restricted hashes,
1080 even though Storable rebuilds restricted hashes by putting in all the
1081 placeholders (first) before turning on the readonly flag, because
1082 Storable always pre-splits the hash. */
1083 hv_clear_placeholders(hv);
1087 #if defined(STRANGE_MALLOC) || defined(MYMALLOC)
1088 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1089 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1095 Move(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1098 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1099 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1104 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
1106 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1108 if (oldsize >= 64) {
1109 offer_nice_chunk(HvARRAY(hv),
1110 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1111 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
1114 Safefree(HvARRAY(hv));
1118 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
1119 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
1120 HvARRAY(hv) = (HE**) a;
1123 for (i=0; i<oldsize; i++,aep++) {
1124 int left_length = 0;
1125 int right_length = 0;
1130 if (!entry) /* non-existent */
1134 if ((HeHASH(entry) & newsize) != (U32)i) {
1135 *oentry = HeNEXT(entry);
1136 HeNEXT(entry) = *bep;
1141 oentry = &HeNEXT(entry);
1146 /* I think we don't actually need to keep track of the longest length,
1147 merely flag if anything is too long. But for the moment while
1148 developing this code I'll track it. */
1149 if (left_length > longest_chain)
1150 longest_chain = left_length;
1151 if (right_length > longest_chain)
1152 longest_chain = right_length;
1156 /* Pick your policy for "hashing isn't working" here: */
1157 if (longest_chain <= HV_MAX_LENGTH_BEFORE_SPLIT /* split worked? */
1162 if (hv == PL_strtab) {
1163 /* Urg. Someone is doing something nasty to the string table.
1168 /* Awooga. Awooga. Pathological data. */
1169 /*PerlIO_printf(PerlIO_stderr(), "%p %d of %d with %d/%d buckets\n", (void*)hv,
1170 longest_chain, HvTOTALKEYS(hv), HvFILL(hv), 1+HvMAX(hv));*/
1173 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1174 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1176 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1179 was_shared = HvSHAREKEYS(hv);
1181 HvSHAREKEYS_off(hv);
1186 for (i=0; i<newsize; i++,aep++) {
1187 register HE *entry = *aep;
1189 /* We're going to trash this HE's next pointer when we chain it
1190 into the new hash below, so store where we go next. */
1191 HE * const next = HeNEXT(entry);
1196 PERL_HASH_INTERNAL(hash, HeKEY(entry), HeKLEN(entry));
1201 = save_hek_flags(HeKEY(entry), HeKLEN(entry),
1202 hash, HeKFLAGS(entry));
1203 unshare_hek (HeKEY_hek(entry));
1204 HeKEY_hek(entry) = new_hek;
1206 /* Not shared, so simply write the new hash in. */
1207 HeHASH(entry) = hash;
1209 /*PerlIO_printf(PerlIO_stderr(), "%d ", HeKFLAGS(entry));*/
1210 HEK_REHASH_on(HeKEY_hek(entry));
1211 /*PerlIO_printf(PerlIO_stderr(), "%d\n", HeKFLAGS(entry));*/
1213 /* Copy oentry to the correct new chain. */
1214 bep = ((HE**)a) + (hash & (I32) xhv->xhv_max);
1215 HeNEXT(entry) = *bep;
1221 Safefree (HvARRAY(hv));
1222 HvARRAY(hv) = (HE **)a;
1226 Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
1229 register XPVHV* xhv = (XPVHV*)SvANY(hv);
1230 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
1231 register I32 newsize;
1236 PERL_ARGS_ASSERT_HV_KSPLIT;
1238 newsize = (I32) newmax; /* possible truncation here */
1239 if (newsize != newmax || newmax <= oldsize)
1241 while ((newsize & (1 + ~newsize)) != newsize) {
1242 newsize &= ~(newsize & (1 + ~newsize)); /* get proper power of 2 */
1244 if (newsize < newmax)
1246 if (newsize < newmax)
1247 return; /* overflow detection */
1249 a = (char *) HvARRAY(hv);
1252 #if defined(STRANGE_MALLOC) || defined(MYMALLOC)
1253 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1254 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1260 Copy(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1263 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1264 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1269 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
1271 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1273 if (oldsize >= 64) {
1274 offer_nice_chunk(HvARRAY(hv),
1275 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1276 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
1279 Safefree(HvARRAY(hv));
1282 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
1285 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
1287 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
1288 HvARRAY(hv) = (HE **) a;
1289 if (!xhv->xhv_keys /* !HvTOTALKEYS(hv) */) /* skip rest if no entries */
1293 for (i=0; i<oldsize; i++,aep++) {
1297 if (!entry) /* non-existent */
1300 register I32 j = (HeHASH(entry) & newsize);
1304 *oentry = HeNEXT(entry);
1305 HeNEXT(entry) = aep[j];
1309 oentry = &HeNEXT(entry);
1316 Perl_newHVhv(pTHX_ HV *ohv)
1319 HV * const hv = newHV();
1322 if (!ohv || !HvTOTALKEYS(ohv))
1324 hv_max = HvMAX(ohv);
1326 if (!SvMAGICAL((const SV *)ohv)) {
1327 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
1329 const bool shared = !!HvSHAREKEYS(ohv);
1330 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
1332 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
1335 /* In each bucket... */
1336 for (i = 0; i <= hv_max; i++) {
1338 HE *oent = oents[i];
1345 /* Copy the linked list of entries. */
1346 for (; oent; oent = HeNEXT(oent)) {
1347 const U32 hash = HeHASH(oent);
1348 const char * const key = HeKEY(oent);
1349 const STRLEN len = HeKLEN(oent);
1350 const int flags = HeKFLAGS(oent);
1351 HE * const ent = new_HE();
1352 SV *const val = HeVAL(oent);
1354 HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val);
1356 = shared ? share_hek_flags(key, len, hash, flags)
1357 : save_hek_flags(key, len, hash, flags);
1368 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
1372 /* Iterate over ohv, copying keys and values one at a time. */
1374 const I32 riter = HvRITER_get(ohv);
1375 HE * const eiter = HvEITER_get(ohv);
1376 STRLEN hv_fill = HvFILL(ohv);
1378 /* Can we use fewer buckets? (hv_max is always 2^n-1) */
1379 while (hv_max && hv_max + 1 >= hv_fill * 2)
1380 hv_max = hv_max / 2;
1384 while ((entry = hv_iternext_flags(ohv, 0))) {
1385 SV *const val = HeVAL(entry);
1386 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1387 SvIMMORTAL(val) ? val : newSVsv(val),
1388 HeHASH(entry), HeKFLAGS(entry));
1390 HvRITER_set(ohv, riter);
1391 HvEITER_set(ohv, eiter);
1397 /* A rather specialised version of newHVhv for copying %^H, ensuring all the
1398 magic stays on it. */
1400 Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1402 HV * const hv = newHV();
1404 if (ohv && HvTOTALKEYS(ohv)) {
1405 STRLEN hv_max = HvMAX(ohv);
1406 STRLEN hv_fill = HvFILL(ohv);
1408 const I32 riter = HvRITER_get(ohv);
1409 HE * const eiter = HvEITER_get(ohv);
1411 while (hv_max && hv_max + 1 >= hv_fill * 2)
1412 hv_max = hv_max / 2;
1416 while ((entry = hv_iternext_flags(ohv, 0))) {
1417 SV *const sv = newSVsv(HeVAL(entry));
1418 SV *heksv = newSVhek(HeKEY_hek(entry));
1419 sv_magic(sv, NULL, PERL_MAGIC_hintselem,
1420 (char *)heksv, HEf_SVKEY);
1421 SvREFCNT_dec(heksv);
1422 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1423 sv, HeHASH(entry), HeKFLAGS(entry));
1425 HvRITER_set(ohv, riter);
1426 HvEITER_set(ohv, eiter);
1428 hv_magic(hv, NULL, PERL_MAGIC_hints);
1433 Perl_hv_free_ent(pTHX_ HV *hv, register HE *entry)
1438 PERL_ARGS_ASSERT_HV_FREE_ENT;
1443 if (val && isGV(val) && isGV_with_GP(val) && GvCVu(val) && HvNAME_get(hv))
1444 mro_method_changed_in(hv); /* deletion of method from stash */
1446 if (HeKLEN(entry) == HEf_SVKEY) {
1447 SvREFCNT_dec(HeKEY_sv(entry));
1448 Safefree(HeKEY_hek(entry));
1450 else if (HvSHAREKEYS(hv))
1451 unshare_hek(HeKEY_hek(entry));
1453 Safefree(HeKEY_hek(entry));
1459 Perl_hv_delayfree_ent(pTHX_ HV *hv, register HE *entry)
1463 PERL_ARGS_ASSERT_HV_DELAYFREE_ENT;
1467 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1468 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
1469 if (HeKLEN(entry) == HEf_SVKEY) {
1470 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
1472 hv_free_ent(hv, entry);
1476 =for apidoc hv_clear
1478 Clears a hash, making it empty.
1484 Perl_hv_clear(pTHX_ HV *hv)
1487 register XPVHV* xhv;
1491 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1493 xhv = (XPVHV*)SvANY(hv);
1495 if (SvREADONLY(hv) && HvARRAY(hv) != NULL) {
1496 /* restricted hash: convert all keys to placeholders */
1498 for (i = 0; i <= xhv->xhv_max; i++) {
1499 HE *entry = (HvARRAY(hv))[i];
1500 for (; entry; entry = HeNEXT(entry)) {
1501 /* not already placeholder */
1502 if (HeVAL(entry) != &PL_sv_placeholder) {
1503 if (HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1504 SV* const keysv = hv_iterkeysv(entry);
1506 "Attempt to delete readonly key '%"SVf"' from a restricted hash",
1509 SvREFCNT_dec(HeVAL(entry));
1510 HeVAL(entry) = &PL_sv_placeholder;
1511 HvPLACEHOLDERS(hv)++;
1519 HvPLACEHOLDERS_set(hv, 0);
1521 Zero(HvARRAY(hv), xhv->xhv_max+1 /* HvMAX(hv)+1 */, HE*);
1524 mg_clear(MUTABLE_SV(hv));
1526 HvHASKFLAGS_off(hv);
1531 mro_isa_changed_in(hv);
1532 HvEITER_set(hv, NULL);
1537 =for apidoc hv_clear_placeholders
1539 Clears any placeholders from a hash. If a restricted hash has any of its keys
1540 marked as readonly and the key is subsequently deleted, the key is not actually
1541 deleted but is marked by assigning it a value of &PL_sv_placeholder. This tags
1542 it so it will be ignored by future operations such as iterating over the hash,
1543 but will still allow the hash to have a value reassigned to the key at some
1544 future point. This function clears any such placeholder keys from the hash.
1545 See Hash::Util::lock_keys() for an example of its use.
1551 Perl_hv_clear_placeholders(pTHX_ HV *hv)
1554 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
1556 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
1559 clear_placeholders(hv, items);
1563 S_clear_placeholders(pTHX_ HV *hv, U32 items)
1568 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
1575 /* Loop down the linked list heads */
1577 HE **oentry = &(HvARRAY(hv))[i];
1580 while ((entry = *oentry)) {
1581 if (HeVAL(entry) == &PL_sv_placeholder) {
1582 *oentry = HeNEXT(entry);
1583 if (entry == HvEITER_get(hv))
1586 hv_free_ent(hv, entry);
1590 HvTOTALKEYS(hv) -= (IV)HvPLACEHOLDERS_get(hv);
1591 if (HvKEYS(hv) == 0)
1592 HvHASKFLAGS_off(hv);
1593 HvPLACEHOLDERS_set(hv, 0);
1597 oentry = &HeNEXT(entry);
1602 /* You can't get here, hence assertion should always fail. */
1603 assert (items == 0);
1608 S_hfreeentries(pTHX_ HV *hv)
1610 /* This is the array that we're going to restore */
1611 HE **const orig_array = HvARRAY(hv);
1615 PERL_ARGS_ASSERT_HFREEENTRIES;
1621 /* If the hash is actually a symbol table with a name, look after the
1623 struct xpvhv_aux *iter = HvAUX(hv);
1625 name = iter->xhv_name;
1626 iter->xhv_name = NULL;
1631 /* orig_array remains unchanged throughout the loop. If after freeing all
1632 the entries it turns out that one of the little blighters has triggered
1633 an action that has caused HvARRAY to be re-allocated, then we set
1634 array to the new HvARRAY, and try again. */
1637 /* This is the one we're going to try to empty. First time round
1638 it's the original array. (Hopefully there will only be 1 time
1640 HE ** const array = HvARRAY(hv);
1643 /* Because we have taken xhv_name out, the only allocated pointer
1644 in the aux structure that might exist is the backreference array.
1649 struct mro_meta *meta;
1650 struct xpvhv_aux *iter = HvAUX(hv);
1651 /* weak references: if called from sv_clear(), the backrefs
1652 * should already have been killed; if there are any left, its
1653 * because we're doing hv_clear() or hv_undef(), and the HV
1654 * will continue to live.
1655 * Because while freeing the entries we fake up a NULL HvARRAY
1656 * (and hence HvAUX), we need to store the backref array
1657 * somewhere else; but it still needs to be visible in case
1658 * any the things we free happen to call sv_del_backref().
1659 * We do this by storing it in magic instead.
1660 * If, during the entry freeing, a destructor happens to add
1661 * a new weak backref, then sv_add_backref will look in both
1662 * places (magic in HvAUX) for the AV, but will create a new
1663 * AV in HvAUX if it can't find one (if it finds it in magic,
1664 * it moves it back into HvAUX. So at the end of the iteration
1665 * we have to allow for this. */
1668 if (iter->xhv_backreferences) {
1669 if (SvTYPE(iter->xhv_backreferences) == SVt_PVAV) {
1670 /* The sv_magic will increase the reference count of the AV,
1671 so we need to drop it first. */
1672 SvREFCNT_dec(iter->xhv_backreferences);
1673 if (AvFILLp(iter->xhv_backreferences) == -1) {
1674 /* Turns out that the array is empty. Just free it. */
1675 SvREFCNT_dec(iter->xhv_backreferences);
1678 sv_magic(MUTABLE_SV(hv),
1679 MUTABLE_SV(iter->xhv_backreferences),
1680 PERL_MAGIC_backref, NULL, 0);
1685 sv_magic(MUTABLE_SV(hv), NULL, PERL_MAGIC_backref, NULL, 0);
1686 mg = mg_find(MUTABLE_SV(hv), PERL_MAGIC_backref);
1687 mg->mg_obj = (SV*)iter->xhv_backreferences;
1689 iter->xhv_backreferences = NULL;
1692 entry = iter->xhv_eiter; /* HvEITER(hv) */
1693 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1695 hv_free_ent(hv, entry);
1697 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1698 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1700 if((meta = iter->xhv_mro_meta)) {
1701 if (meta->mro_linear_all) {
1702 SvREFCNT_dec(MUTABLE_SV(meta->mro_linear_all));
1703 meta->mro_linear_all = NULL;
1704 /* This is just acting as a shortcut pointer. */
1705 meta->mro_linear_current = NULL;
1706 } else if (meta->mro_linear_current) {
1707 /* Only the current MRO is stored, so this owns the data.
1709 SvREFCNT_dec(meta->mro_linear_current);
1710 meta->mro_linear_current = NULL;
1712 if(meta->mro_nextmethod) SvREFCNT_dec(meta->mro_nextmethod);
1713 SvREFCNT_dec(meta->isa);
1715 iter->xhv_mro_meta = NULL;
1718 /* There are now no allocated pointers in the aux structure. */
1720 SvFLAGS(hv) &= ~SVf_OOK; /* Goodbye, aux structure. */
1721 /* What aux structure? */
1724 /* make everyone else think the array is empty, so that the destructors
1725 * called for freed entries can't recursively mess with us */
1727 ((XPVHV*) SvANY(hv))->xhv_keys = 0;
1731 /* Loop down the linked list heads */
1732 HE *entry = array[i];
1735 register HE * const oentry = entry;
1736 entry = HeNEXT(entry);
1737 hv_free_ent(hv, oentry);
1741 /* As there are no allocated pointers in the aux structure, it's now
1742 safe to free the array we just cleaned up, if it's not the one we're
1743 going to put back. */
1744 if (array != orig_array) {
1749 /* Good. No-one added anything this time round. */
1754 /* Someone attempted to iterate or set the hash name while we had
1755 the array set to 0. We'll catch backferences on the next time
1756 round the while loop. */
1757 assert(HvARRAY(hv));
1759 if (HvAUX(hv)->xhv_name) {
1760 unshare_hek_or_pvn(HvAUX(hv)->xhv_name, 0, 0, 0);
1764 if (--attempts == 0) {
1765 Perl_die(aTHX_ "panic: hfreeentries failed to free hash - something is repeatedly re-creating entries");
1769 HvARRAY(hv) = orig_array;
1771 /* If the hash was actually a symbol table, put the name back. */
1773 /* We have restored the original array. If name is non-NULL, then
1774 the original array had an aux structure at the end. So this is
1776 SvFLAGS(hv) |= SVf_OOK;
1777 HvAUX(hv)->xhv_name = name;
1782 =for apidoc hv_undef
1790 Perl_hv_undef(pTHX_ HV *hv)
1793 register XPVHV* xhv;
1798 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1799 xhv = (XPVHV*)SvANY(hv);
1801 if ((name = HvNAME_get(hv)) && !PL_dirty)
1802 mro_isa_changed_in(hv);
1807 (void)hv_delete(PL_stashcache, name, HvNAMELEN_get(hv), G_DISCARD);
1808 hv_name_set(hv, NULL, 0, 0);
1810 SvFLAGS(hv) &= ~SVf_OOK;
1811 Safefree(HvARRAY(hv));
1812 xhv->xhv_max = 7; /* HvMAX(hv) = 7 (it's a normal hash) */
1814 HvPLACEHOLDERS_set(hv, 0);
1817 mg_clear(MUTABLE_SV(hv));
1823 Returns the number of hash buckets that happen to be in use. This function is
1824 wrapped by the macro C<HvFILL>.
1826 Previously this value was stored in the HV structure, rather than being
1827 calculated on demand.
1833 Perl_hv_fill(pTHX_ HV const *const hv)
1836 HE **ents = HvARRAY(hv);
1838 PERL_ARGS_ASSERT_HV_FILL;
1841 HE *const *const last = ents + HvMAX(hv);
1842 count = last + 1 - ents;
1847 } while (++ents <= last);
1852 static struct xpvhv_aux*
1853 S_hv_auxinit(HV *hv) {
1854 struct xpvhv_aux *iter;
1857 PERL_ARGS_ASSERT_HV_AUXINIT;
1860 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
1861 + sizeof(struct xpvhv_aux), char);
1863 array = (char *) HvARRAY(hv);
1864 Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
1865 + sizeof(struct xpvhv_aux), char);
1867 HvARRAY(hv) = (HE**) array;
1868 /* SvOOK_on(hv) attacks the IV flags. */
1869 SvFLAGS(hv) |= SVf_OOK;
1872 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1873 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1875 iter->xhv_backreferences = 0;
1876 iter->xhv_mro_meta = NULL;
1881 =for apidoc hv_iterinit
1883 Prepares a starting point to traverse a hash table. Returns the number of
1884 keys in the hash (i.e. the same as C<HvKEYS(tb)>). The return value is
1885 currently only meaningful for hashes without tie magic.
1887 NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
1888 hash buckets that happen to be in use. If you still need that esoteric
1889 value, you can get it through the macro C<HvFILL(tb)>.
1896 Perl_hv_iterinit(pTHX_ HV *hv)
1898 PERL_ARGS_ASSERT_HV_ITERINIT;
1900 /* FIXME: Are we not NULL, or do we croak? Place bets now! */
1903 Perl_croak(aTHX_ "Bad hash");
1906 struct xpvhv_aux * const iter = HvAUX(hv);
1907 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
1908 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1910 hv_free_ent(hv, entry);
1912 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1913 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1918 /* used to be xhv->xhv_fill before 5.004_65 */
1919 return HvTOTALKEYS(hv);
1923 Perl_hv_riter_p(pTHX_ HV *hv) {
1924 struct xpvhv_aux *iter;
1926 PERL_ARGS_ASSERT_HV_RITER_P;
1929 Perl_croak(aTHX_ "Bad hash");
1931 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
1932 return &(iter->xhv_riter);
1936 Perl_hv_eiter_p(pTHX_ HV *hv) {
1937 struct xpvhv_aux *iter;
1939 PERL_ARGS_ASSERT_HV_EITER_P;
1942 Perl_croak(aTHX_ "Bad hash");
1944 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
1945 return &(iter->xhv_eiter);
1949 Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
1950 struct xpvhv_aux *iter;
1952 PERL_ARGS_ASSERT_HV_RITER_SET;
1955 Perl_croak(aTHX_ "Bad hash");
1963 iter = hv_auxinit(hv);
1965 iter->xhv_riter = riter;
1969 Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
1970 struct xpvhv_aux *iter;
1972 PERL_ARGS_ASSERT_HV_EITER_SET;
1975 Perl_croak(aTHX_ "Bad hash");
1980 /* 0 is the default so don't go malloc()ing a new structure just to
1985 iter = hv_auxinit(hv);
1987 iter->xhv_eiter = eiter;
1991 Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
1994 struct xpvhv_aux *iter;
1997 PERL_ARGS_ASSERT_HV_NAME_SET;
1998 PERL_UNUSED_ARG(flags);
2001 Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len);
2005 if (iter->xhv_name) {
2006 unshare_hek_or_pvn(iter->xhv_name, 0, 0, 0);
2012 iter = hv_auxinit(hv);
2014 PERL_HASH(hash, name, len);
2015 iter->xhv_name = name ? share_hek(name, len, hash) : NULL;
2019 Perl_hv_backreferences_p(pTHX_ HV *hv) {
2020 struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2022 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
2023 PERL_UNUSED_CONTEXT;
2025 return &(iter->xhv_backreferences);
2029 Perl_hv_kill_backrefs(pTHX_ HV *hv) {
2032 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2037 av = HvAUX(hv)->xhv_backreferences;
2040 HvAUX(hv)->xhv_backreferences = 0;
2041 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
2042 if (SvTYPE(av) == SVt_PVAV)
2048 hv_iternext is implemented as a macro in hv.h
2050 =for apidoc hv_iternext
2052 Returns entries from a hash iterator. See C<hv_iterinit>.
2054 You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2055 iterator currently points to, without losing your place or invalidating your
2056 iterator. Note that in this case the current entry is deleted from the hash
2057 with your iterator holding the last reference to it. Your iterator is flagged
2058 to free the entry on the next call to C<hv_iternext>, so you must not discard
2059 your iterator immediately else the entry will leak - call C<hv_iternext> to
2060 trigger the resource deallocation.
2062 =for apidoc hv_iternext_flags
2064 Returns entries from a hash iterator. See C<hv_iterinit> and C<hv_iternext>.
2065 The C<flags> value will normally be zero; if HV_ITERNEXT_WANTPLACEHOLDERS is
2066 set the placeholders keys (for restricted hashes) will be returned in addition
2067 to normal keys. By default placeholders are automatically skipped over.
2068 Currently a placeholder is implemented with a value that is
2069 C<&Perl_sv_placeholder>. Note that the implementation of placeholders and
2070 restricted hashes may change, and the implementation currently is
2071 insufficiently abstracted for any change to be tidy.
2077 Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2080 register XPVHV* xhv;
2084 struct xpvhv_aux *iter;
2086 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2089 Perl_croak(aTHX_ "Bad hash");
2091 xhv = (XPVHV*)SvANY(hv);
2094 /* Too many things (well, pp_each at least) merrily assume that you can
2095 call iv_iternext without calling hv_iterinit, so we'll have to deal
2101 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2102 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
2103 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
2104 SV * const key = sv_newmortal();
2106 sv_setsv(key, HeSVKEY_force(entry));
2107 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
2113 /* one HE per MAGICAL hash */
2114 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
2116 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
2118 HeKEY_hek(entry) = hek;
2119 HeKLEN(entry) = HEf_SVKEY;
2121 magic_nextpack(MUTABLE_SV(hv),mg,key);
2123 /* force key to stay around until next time */
2124 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2125 return entry; /* beware, hent_val is not set */
2127 SvREFCNT_dec(HeVAL(entry));
2128 Safefree(HeKEY_hek(entry));
2130 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2134 #if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */
2135 if (!entry && SvRMAGICAL((const SV *)hv)
2136 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
2139 /* The prime_env_iter() on VMS just loaded up new hash values
2140 * so the iteration count needs to be reset back to the beginning
2144 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2149 /* hv_iterint now ensures this. */
2150 assert (HvARRAY(hv));
2152 /* At start of hash, entry is NULL. */
2155 entry = HeNEXT(entry);
2156 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2158 * Skip past any placeholders -- don't want to include them in
2161 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
2162 entry = HeNEXT(entry);
2167 /* Skip the entire loop if the hash is empty. */
2168 if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS)
2169 ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) {
2171 /* OK. Come to the end of the current list. Grab the next one. */
2173 iter->xhv_riter++; /* HvRITER(hv)++ */
2174 if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) {
2175 /* There is no next one. End of the hash. */
2176 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2179 entry = (HvARRAY(hv))[iter->xhv_riter];
2181 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2182 /* If we have an entry, but it's a placeholder, don't count it.
2184 while (entry && HeVAL(entry) == &PL_sv_placeholder)
2185 entry = HeNEXT(entry);
2187 /* Will loop again if this linked list starts NULL
2188 (for HV_ITERNEXT_WANTPLACEHOLDERS)
2189 or if we run through it and find only placeholders. */
2193 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2195 hv_free_ent(hv, oldentry);
2198 /*if (HvREHASH(hv) && entry && !HeKREHASH(entry))
2199 PerlIO_printf(PerlIO_stderr(), "Awooga %p %p\n", (void*)hv, (void*)entry);*/
2201 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
2206 =for apidoc hv_iterkey
2208 Returns the key from the current position of the hash iterator. See
2215 Perl_hv_iterkey(pTHX_ register HE *entry, I32 *retlen)
2217 PERL_ARGS_ASSERT_HV_ITERKEY;
2219 if (HeKLEN(entry) == HEf_SVKEY) {
2221 char * const p = SvPV(HeKEY_sv(entry), len);
2226 *retlen = HeKLEN(entry);
2227 return HeKEY(entry);
2231 /* unlike hv_iterval(), this always returns a mortal copy of the key */
2233 =for apidoc hv_iterkeysv
2235 Returns the key as an C<SV*> from the current position of the hash
2236 iterator. The return value will always be a mortal copy of the key. Also
2243 Perl_hv_iterkeysv(pTHX_ register HE *entry)
2245 PERL_ARGS_ASSERT_HV_ITERKEYSV;
2247 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
2251 =for apidoc hv_iterval
2253 Returns the value from the current position of the hash iterator. See
2260 Perl_hv_iterval(pTHX_ HV *hv, register HE *entry)
2262 PERL_ARGS_ASSERT_HV_ITERVAL;
2264 if (SvRMAGICAL(hv)) {
2265 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
2266 SV* const sv = sv_newmortal();
2267 if (HeKLEN(entry) == HEf_SVKEY)
2268 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
2270 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
2274 return HeVAL(entry);
2278 =for apidoc hv_iternextsv
2280 Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
2287 Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
2289 HE * const he = hv_iternext_flags(hv, 0);
2291 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
2295 *key = hv_iterkey(he, retlen);
2296 return hv_iterval(hv, he);
2303 =for apidoc hv_magic
2305 Adds magic to a hash. See C<sv_magic>.
2310 /* possibly free a shared string if no one has access to it
2311 * len and hash must both be valid for str.
2314 Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
2316 unshare_hek_or_pvn (NULL, str, len, hash);
2321 Perl_unshare_hek(pTHX_ HEK *hek)
2324 unshare_hek_or_pvn(hek, NULL, 0, 0);
2327 /* possibly free a shared string if no one has access to it
2328 hek if non-NULL takes priority over the other 3, else str, len and hash
2329 are used. If so, len and hash must both be valid for str.
2332 S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
2335 register XPVHV* xhv;
2337 register HE **oentry;
2339 bool is_utf8 = FALSE;
2341 const char * const save = str;
2342 struct shared_he *he = NULL;
2345 /* Find the shared he which is just before us in memory. */
2346 he = (struct shared_he *)(((char *)hek)
2347 - STRUCT_OFFSET(struct shared_he,
2350 /* Assert that the caller passed us a genuine (or at least consistent)
2352 assert (he->shared_he_he.hent_hek == hek);
2354 if (he->shared_he_he.he_valu.hent_refcount - 1) {
2355 --he->shared_he_he.he_valu.hent_refcount;
2359 hash = HEK_HASH(hek);
2360 } else if (len < 0) {
2361 STRLEN tmplen = -len;
2363 /* See the note in hv_fetch(). --jhi */
2364 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2367 k_flags = HVhek_UTF8;
2369 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2372 /* what follows was the moral equivalent of:
2373 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
2375 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
2377 xhv = (XPVHV*)SvANY(PL_strtab);
2378 /* assert(xhv_array != 0) */
2379 first = oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
2381 const HE *const he_he = &(he->shared_he_he);
2382 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
2387 const int flags_masked = k_flags & HVhek_MASK;
2388 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
2389 if (HeHASH(entry) != hash) /* strings can't be equal */
2391 if (HeKLEN(entry) != len)
2393 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
2395 if (HeKFLAGS(entry) != flags_masked)
2402 if (--entry->he_valu.hent_refcount == 0) {
2403 *oentry = HeNEXT(entry);
2405 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
2410 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
2411 "Attempt to free non-existent shared string '%s'%s"
2413 hek ? HEK_KEY(hek) : str,
2414 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
2415 if (k_flags & HVhek_FREEKEY)
2419 /* get a (constant) string ptr from the global string table
2420 * string will get added if it is not already there.
2421 * len and hash must both be valid for str.
2424 Perl_share_hek(pTHX_ const char *str, I32 len, register U32 hash)
2426 bool is_utf8 = FALSE;
2428 const char * const save = str;
2430 PERL_ARGS_ASSERT_SHARE_HEK;
2433 STRLEN tmplen = -len;
2435 /* See the note in hv_fetch(). --jhi */
2436 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2438 /* If we were able to downgrade here, then than means that we were passed
2439 in a key which only had chars 0-255, but was utf8 encoded. */
2442 /* If we found we were able to downgrade the string to bytes, then
2443 we should flag that it needs upgrading on keys or each. Also flag
2444 that we need share_hek_flags to free the string. */
2446 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2449 return share_hek_flags (str, len, hash, flags);
2453 S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags)
2457 const int flags_masked = flags & HVhek_MASK;
2458 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
2459 register XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
2461 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
2463 /* what follows is the moral equivalent of:
2465 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
2466 hv_store(PL_strtab, str, len, NULL, hash);
2468 Can't rehash the shared string table, so not sure if it's worth
2469 counting the number of entries in the linked list
2472 /* assert(xhv_array != 0) */
2473 entry = (HvARRAY(PL_strtab))[hindex];
2474 for (;entry; entry = HeNEXT(entry)) {
2475 if (HeHASH(entry) != hash) /* strings can't be equal */
2477 if (HeKLEN(entry) != len)
2479 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
2481 if (HeKFLAGS(entry) != flags_masked)
2487 /* What used to be head of the list.
2488 If this is NULL, then we're the first entry for this slot, which
2489 means we need to increate fill. */
2490 struct shared_he *new_entry;
2493 HE **const head = &HvARRAY(PL_strtab)[hindex];
2494 HE *const next = *head;
2496 /* We don't actually store a HE from the arena and a regular HEK.
2497 Instead we allocate one chunk of memory big enough for both,
2498 and put the HEK straight after the HE. This way we can find the
2499 HEK directly from the HE.
2502 Newx(k, STRUCT_OFFSET(struct shared_he,
2503 shared_he_hek.hek_key[0]) + len + 2, char);
2504 new_entry = (struct shared_he *)k;
2505 entry = &(new_entry->shared_he_he);
2506 hek = &(new_entry->shared_he_hek);
2508 Copy(str, HEK_KEY(hek), len, char);
2509 HEK_KEY(hek)[len] = 0;
2511 HEK_HASH(hek) = hash;
2512 HEK_FLAGS(hek) = (unsigned char)flags_masked;
2514 /* Still "point" to the HEK, so that other code need not know what
2516 HeKEY_hek(entry) = hek;
2517 entry->he_valu.hent_refcount = 0;
2518 HeNEXT(entry) = next;
2521 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
2522 if (!next) { /* initial entry? */
2523 } else if (xhv->xhv_keys > xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
2528 ++entry->he_valu.hent_refcount;
2530 if (flags & HVhek_FREEKEY)
2533 return HeKEY_hek(entry);
2537 Perl_hv_placeholders_p(pTHX_ HV *hv)
2540 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2542 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
2545 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
2548 Perl_die(aTHX_ "panic: hv_placeholders_p");
2551 return &(mg->mg_len);
2556 Perl_hv_placeholders_get(pTHX_ const HV *hv)
2559 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2561 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
2563 return mg ? mg->mg_len : 0;
2567 Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
2570 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
2572 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
2577 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
2578 Perl_die(aTHX_ "panic: hv_placeholders_set");
2580 /* else we don't need to add magic to record 0 placeholders. */
2584 S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
2589 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
2591 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
2596 value = &PL_sv_placeholder;
2599 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
2602 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
2605 case HVrhek_PV_UTF8:
2606 /* Create a string SV that directly points to the bytes in our
2608 value = newSV_type(SVt_PV);
2609 SvPV_set(value, (char *) he->refcounted_he_data + 1);
2610 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
2611 /* This stops anything trying to free it */
2612 SvLEN_set(value, 0);
2614 SvREADONLY_on(value);
2615 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
2619 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %x",
2620 he->refcounted_he_data[0]);
2626 =for apidoc refcounted_he_chain_2hv
2628 Generates and returns a C<HV *> by walking up the tree starting at the passed
2629 in C<struct refcounted_he *>.
2634 Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain)
2638 U32 placeholders = 0;
2639 /* We could chase the chain once to get an idea of the number of keys,
2640 and call ksplit. But for now we'll make a potentially inefficient
2641 hash with only 8 entries in its array. */
2642 const U32 max = HvMAX(hv);
2646 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
2647 HvARRAY(hv) = (HE**)array;
2652 U32 hash = chain->refcounted_he_hash;
2654 U32 hash = HEK_HASH(chain->refcounted_he_hek);
2656 HE **oentry = &((HvARRAY(hv))[hash & max]);
2657 HE *entry = *oentry;
2660 for (; entry; entry = HeNEXT(entry)) {
2661 if (HeHASH(entry) == hash) {
2662 /* We might have a duplicate key here. If so, entry is older
2663 than the key we've already put in the hash, so if they are
2664 the same, skip adding entry. */
2666 const STRLEN klen = HeKLEN(entry);
2667 const char *const key = HeKEY(entry);
2668 if (klen == chain->refcounted_he_keylen
2669 && (!!HeKUTF8(entry)
2670 == !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2671 && memEQ(key, REF_HE_KEY(chain), klen))
2674 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
2676 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
2677 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
2678 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
2689 = share_hek_flags(REF_HE_KEY(chain),
2690 chain->refcounted_he_keylen,
2691 chain->refcounted_he_hash,
2692 (chain->refcounted_he_data[0]
2693 & (HVhek_UTF8|HVhek_WASUTF8)));
2695 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
2697 value = refcounted_he_value(chain);
2698 if (value == &PL_sv_placeholder)
2700 HeVAL(entry) = value;
2702 /* Link it into the chain. */
2703 HeNEXT(entry) = *oentry;
2709 chain = chain->refcounted_he_next;
2713 clear_placeholders(hv, placeholders);
2714 HvTOTALKEYS(hv) -= placeholders;
2717 /* We could check in the loop to see if we encounter any keys with key
2718 flags, but it's probably not worth it, as this per-hash flag is only
2719 really meant as an optimisation for things like Storable. */
2721 DEBUG_A(Perl_hv_assert(aTHX_ hv));
2727 Perl_refcounted_he_fetch(pTHX_ const struct refcounted_he *chain, SV *keysv,
2728 const char *key, STRLEN klen, int flags, U32 hash)
2731 /* Just to be awkward, if you're using this interface the UTF-8-or-not-ness
2732 of your key has to exactly match that which is stored. */
2733 SV *value = &PL_sv_placeholder;
2736 /* No point in doing any of this if there's nothing to find. */
2740 if (flags & HVhek_FREEKEY)
2742 key = SvPV_const(keysv, klen);
2744 is_utf8 = (SvUTF8(keysv) != 0);
2746 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
2750 if (keysv && (SvIsCOW_shared_hash(keysv))) {
2751 hash = SvSHARED_HASH(keysv);
2753 PERL_HASH(hash, key, klen);
2757 for (; chain; chain = chain->refcounted_he_next) {
2759 if (hash != chain->refcounted_he_hash)
2761 if (klen != chain->refcounted_he_keylen)
2763 if (memNE(REF_HE_KEY(chain),key,klen))
2765 if (!!is_utf8 != !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2768 if (hash != HEK_HASH(chain->refcounted_he_hek))
2770 if (klen != (STRLEN)HEK_LEN(chain->refcounted_he_hek))
2772 if (memNE(HEK_KEY(chain->refcounted_he_hek),key,klen))
2774 if (!!is_utf8 != !!HEK_UTF8(chain->refcounted_he_hek))
2778 value = sv_2mortal(refcounted_he_value(chain));
2783 if (flags & HVhek_FREEKEY)
2790 =for apidoc refcounted_he_new
2792 Creates a new C<struct refcounted_he>. As S<key> is copied, and value is
2793 stored in a compact form, all references remain the property of the caller.
2794 The C<struct refcounted_he> is returned with a reference count of 1.
2799 struct refcounted_he *
2800 Perl_refcounted_he_new(pTHX_ struct refcounted_he *const parent,
2801 SV *const key, SV *const value) {
2804 const char *key_p = SvPV_const(key, key_len);
2805 STRLEN value_len = 0;
2806 const char *value_p = NULL;
2809 bool is_utf8 = SvUTF8(key) ? TRUE : FALSE;
2812 value_type = HVrhek_PV;
2813 } else if (SvIOK(value)) {
2814 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
2815 } else if (value == &PL_sv_placeholder) {
2816 value_type = HVrhek_delete;
2817 } else if (!SvOK(value)) {
2818 value_type = HVrhek_undef;
2820 value_type = HVrhek_PV;
2823 if (value_type == HVrhek_PV) {
2824 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
2825 the value is overloaded, and doesn't yet have the UTF-8flag set. */
2826 value_p = SvPV_const(value, value_len);
2828 value_type = HVrhek_PV_UTF8;
2833 /* Hash keys are always stored normalised to (yes) ISO-8859-1.
2834 As we're going to be building hash keys from this value in future,
2835 normalise it now. */
2836 key_p = (char*)bytes_from_utf8((const U8*)key_p, &key_len, &is_utf8);
2837 flags |= is_utf8 ? HVhek_UTF8 : HVhek_WASUTF8;
2840 return refcounted_he_new_common(parent, key_p, key_len, flags, value_type,
2841 ((value_type == HVrhek_PV
2842 || value_type == HVrhek_PV_UTF8) ?
2843 (void *)value_p : (void *)value),
2847 static struct refcounted_he *
2848 S_refcounted_he_new_common(pTHX_ struct refcounted_he *const parent,
2849 const char *const key_p, const STRLEN key_len,
2850 const char flags, char value_type,
2851 const void *value, const STRLEN value_len) {
2853 struct refcounted_he *he;
2855 const bool is_pv = value_type == HVrhek_PV || value_type == HVrhek_PV_UTF8;
2856 STRLEN key_offset = is_pv ? value_len + 2 : 1;
2858 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_COMMON;
2861 he = (struct refcounted_he*)
2862 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2866 he = (struct refcounted_he*)
2867 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2871 he->refcounted_he_next = parent;
2874 Copy((char *)value, he->refcounted_he_data + 1, value_len + 1, char);
2875 he->refcounted_he_val.refcounted_he_u_len = value_len;
2876 } else if (value_type == HVrhek_IV) {
2877 he->refcounted_he_val.refcounted_he_u_iv = SvIVX((const SV *)value);
2878 } else if (value_type == HVrhek_UV) {
2879 he->refcounted_he_val.refcounted_he_u_uv = SvUVX((const SV *)value);
2882 PERL_HASH(hash, key_p, key_len);
2885 he->refcounted_he_hash = hash;
2886 he->refcounted_he_keylen = key_len;
2887 Copy(key_p, he->refcounted_he_data + key_offset, key_len, char);
2889 he->refcounted_he_hek = share_hek_flags(key_p, key_len, hash, flags);
2892 if (flags & HVhek_WASUTF8) {
2893 /* If it was downgraded from UTF-8, then the pointer returned from
2894 bytes_from_utf8 is an allocated pointer that we must free. */
2898 he->refcounted_he_data[0] = flags;
2899 he->refcounted_he_refcnt = 1;
2905 =for apidoc refcounted_he_free
2907 Decrements the reference count of the passed in C<struct refcounted_he *>
2908 by one. If the reference count reaches zero the structure's memory is freed,
2909 and C<refcounted_he_free> iterates onto the parent node.
2915 Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
2917 PERL_UNUSED_CONTEXT;
2920 struct refcounted_he *copy;
2924 new_count = --he->refcounted_he_refcnt;
2925 HINTS_REFCNT_UNLOCK;
2931 #ifndef USE_ITHREADS
2932 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
2935 he = he->refcounted_he_next;
2936 PerlMemShared_free(copy);
2940 /* pp_entereval is aware that labels are stored with a key ':' at the top of
2943 Perl_fetch_cop_label(pTHX_ struct refcounted_he *const chain, STRLEN *len,
2948 if (chain->refcounted_he_keylen != 1)
2950 if (*REF_HE_KEY(chain) != ':')
2953 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
2955 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
2958 /* Stop anyone trying to really mess us up by adding their own value for
2960 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
2961 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
2965 *len = chain->refcounted_he_val.refcounted_he_u_len;
2967 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
2968 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
2970 return chain->refcounted_he_data + 1;
2973 /* As newSTATEOP currently gets passed plain char* labels, we will only provide
2974 that interface. Once it works out how to pass in length and UTF-8 ness, this
2975 function will need superseding. */
2976 struct refcounted_he *
2977 Perl_store_cop_label(pTHX_ struct refcounted_he *const chain, const char *label)
2979 PERL_ARGS_ASSERT_STORE_COP_LABEL;
2981 return refcounted_he_new_common(chain, ":", 1, HVrhek_PV, HVrhek_PV,
2982 label, strlen(label));
2986 =for apidoc hv_assert
2988 Check that a hash is in an internally consistent state.
2996 Perl_hv_assert(pTHX_ HV *hv)
3001 int placeholders = 0;
3004 const I32 riter = HvRITER_get(hv);
3005 HE *eiter = HvEITER_get(hv);
3007 PERL_ARGS_ASSERT_HV_ASSERT;
3009 (void)hv_iterinit(hv);
3011 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
3012 /* sanity check the values */
3013 if (HeVAL(entry) == &PL_sv_placeholder)
3017 /* sanity check the keys */
3018 if (HeSVKEY(entry)) {
3019 NOOP; /* Don't know what to check on SV keys. */
3020 } else if (HeKUTF8(entry)) {
3022 if (HeKWASUTF8(entry)) {
3023 PerlIO_printf(Perl_debug_log,
3024 "hash key has both WASUTF8 and UTF8: '%.*s'\n",
3025 (int) HeKLEN(entry), HeKEY(entry));
3028 } else if (HeKWASUTF8(entry))
3031 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
3032 static const char bad_count[] = "Count %d %s(s), but hash reports %d\n";
3033 const int nhashkeys = HvUSEDKEYS(hv);
3034 const int nhashplaceholders = HvPLACEHOLDERS_get(hv);
3036 if (nhashkeys != real) {
3037 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
3040 if (nhashplaceholders != placeholders) {
3041 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
3045 if (withflags && ! HvHASKFLAGS(hv)) {
3046 PerlIO_printf(Perl_debug_log,
3047 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
3052 sv_dump(MUTABLE_SV(hv));
3054 HvRITER_set(hv, riter); /* Restore hash iterator state */
3055 HvEITER_set(hv, eiter);
3062 * c-indentation-style: bsd
3064 * indent-tabs-mode: t
3067 * ex: set ts=8 sts=4 sw=4 noet: