This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Current ppport.h forcibly overrides older buggy versions of utf8_to_uvchr_buf
[perl5.git] / hv.c
CommitLineData
a0d0e21e 1/* hv.c
79072805 2 *
1129b882
NC
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
79072805
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
a0d0e21e
LW
9 */
10
11/*
4ac71550
TC
12 * I sit beside the fire and think
13 * of all that I have seen.
14 * --Bilbo
15 *
16 * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
79072805
LW
17 */
18
d5afce77 19/*
51b56f5c 20=head1 HV Handling
db4fbf16
FC
21A HV structure represents a Perl hash. It consists mainly of an array
22of pointers, each of which points to a linked list of HE structures. The
166f8a29 23array is indexed by the hash function of the key, so each linked list
db4fbf16 24represents all the hash entries with the same hash value. Each HE contains
166f8a29
DM
25a pointer to the actual value, plus a pointer to a HEK structure which
26holds the key and hash value.
27
28=cut
29
d5afce77
RB
30*/
31
79072805 32#include "EXTERN.h"
864dbfa3 33#define PERL_IN_HV_C
3d78eb94 34#define PERL_HASH_INTERNAL_ACCESS
79072805
LW
35#include "perl.h"
36
6f019ba7
YO
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()
40 */
aae087f7
YO
41
42/* MAX_BUCKET_MAX is the maximum max bucket index, at which point we stop growing the
43 * number of buckets,
44 */
45#define MAX_BUCKET_MAX ((1<<26)-1)
46#define DO_HSPLIT(xhv) ( ( ((xhv)->xhv_keys + ((xhv)->xhv_keys >> 1)) > (xhv)->xhv_max ) && \
47 ((xhv)->xhv_max < MAX_BUCKET_MAX) )
fdcd69b6 48
d75ce684 49static const char S_strtab_error[]
5d2b1485
NC
50 = "Cannot modify shared string table in hv_%s";
51
c941fb51
NC
52#ifdef PURIFY
53
54#define new_HE() (HE*)safemalloc(sizeof(HE))
55#define del_HE(p) safefree((char*)p)
56
57#else
58
76e3520e 59STATIC HE*
cea2e8a9 60S_new_he(pTHX)
4633a7c4
LW
61{
62 HE* he;
0bd48802 63 void ** const root = &PL_body_roots[HE_SVSLOT];
6a93a7e5 64
6a93a7e5 65 if (!*root)
1604cfb0 66 Perl_more_bodies(aTHX_ HE_SVSLOT, sizeof(HE), PERL_ARENA_SIZE);
10edeb5d 67 he = (HE*) *root;
ce3e5c45 68 assert(he);
6a93a7e5 69 *root = HeNEXT(he);
333f433b 70 return he;
4633a7c4
LW
71}
72
c941fb51
NC
73#define new_HE() new_he()
74#define del_HE(p) \
75 STMT_START { \
1604cfb0
MS
76 HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \
77 PL_body_roots[HE_SVSLOT] = p; \
c941fb51 78 } STMT_END
d33b2eba 79
d33b2eba 80
d33b2eba
GS
81
82#endif
83
76e3520e 84STATIC HEK *
5f66b61c 85S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
bbce6d69 86{
35a4481c 87 const int flags_masked = flags & HVhek_MASK;
bbce6d69 88 char *k;
eb578fdb 89 HEK *hek;
1c846c1f 90
7918f24d 91 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
ea3ffa52
TC
92
93 Newx(k, HEK_BASESIZE + len + 2, char);
bbce6d69 94 hek = (HEK*)k;
ff68c719 95 Copy(str, HEK_KEY(hek), len, char);
e05949c7 96 HEK_KEY(hek)[len] = 0;
ff68c719 97 HEK_LEN(hek) = len;
98 HEK_HASH(hek) = hash;
45e34800 99 HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED;
dcf933a4
NC
100
101 if (flags & HVhek_FREEKEY)
1604cfb0 102 Safefree(str);
bbce6d69 103 return hek;
104}
105
4a31713e 106/* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
dd28f7bb
DM
107 * for tied hashes */
108
109void
110Perl_free_tied_hv_pool(pTHX)
111{
dd28f7bb
DM
112 HE *he = PL_hv_fetch_ent_mh;
113 while (he) {
1604cfb0
MS
114 HE * const ohe = he;
115 Safefree(HeKEY_hek(he));
116 he = HeNEXT(he);
117 del_HE(ohe);
dd28f7bb 118 }
4608196e 119 PL_hv_fetch_ent_mh = NULL;
dd28f7bb
DM
120}
121
d18c6117 122#if defined(USE_ITHREADS)
0bff533c
NC
123HEK *
124Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
125{
566771cc 126 HEK *shared;
9d4ba2ae 127
7918f24d 128 PERL_ARGS_ASSERT_HEK_DUP;
9d4ba2ae 129 PERL_UNUSED_ARG(param);
0bff533c 130
566771cc 131 if (!source)
1604cfb0 132 return NULL;
566771cc
NC
133
134 shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
0bff533c 135 if (shared) {
1604cfb0
MS
136 /* We already shared this hash key. */
137 (void)share_hek_hek(shared);
0bff533c
NC
138 }
139 else {
1604cfb0
MS
140 shared
141 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
142 HEK_HASH(source), HEK_FLAGS(source));
143 ptr_table_store(PL_ptr_table, source, shared);
0bff533c 144 }
658b4a4a 145 return shared;
0bff533c
NC
146}
147
d18c6117 148HE *
5c4138a0 149Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
d18c6117
GS
150{
151 HE *ret;
152
7918f24d
NC
153 PERL_ARGS_ASSERT_HE_DUP;
154
d18c6117 155 if (!e)
1604cfb0 156 return NULL;
7766f137
GS
157 /* look for it in the table first */
158 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
159 if (ret)
1604cfb0 160 return ret;
7766f137
GS
161
162 /* create anew and remember what it is */
d33b2eba 163 ret = new_HE();
7766f137
GS
164 ptr_table_store(PL_ptr_table, e, ret);
165
d2d73c3e 166 HeNEXT(ret) = he_dup(HeNEXT(e),shared, param);
dd28f7bb 167 if (HeKLEN(e) == HEf_SVKEY) {
1604cfb0
MS
168 char *k;
169 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
170 HeKEY_hek(ret) = (HEK*)k;
171 HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param);
dd28f7bb 172 }
c21d1a0f 173 else if (shared) {
1604cfb0
MS
174 /* This is hek_dup inlined, which seems to be important for speed
175 reasons. */
176 HEK * const source = HeKEY_hek(e);
177 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
178
179 if (shared) {
180 /* We already shared this hash key. */
181 (void)share_hek_hek(shared);
182 }
183 else {
184 shared
185 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
186 HEK_HASH(source), HEK_FLAGS(source));
187 ptr_table_store(PL_ptr_table, source, shared);
188 }
189 HeKEY_hek(ret) = shared;
c21d1a0f 190 }
d18c6117 191 else
1604cfb0 192 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
19692e8d 193 HeKFLAGS(e));
a09252eb 194 HeVAL(ret) = sv_dup_inc(HeVAL(e), param);
d18c6117
GS
195 return ret;
196}
197#endif /* USE_ITHREADS */
198
1b1f1335 199static void
2393f1b9 200S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
1604cfb0 201 const char *msg)
1b1f1335 202{
1b6737cc 203 SV * const sv = sv_newmortal();
7918f24d
NC
204
205 PERL_ARGS_ASSERT_HV_NOTALLOWED;
206
19692e8d 207 if (!(flags & HVhek_FREEKEY)) {
1604cfb0 208 sv_setpvn(sv, key, klen);
1b1f1335
NIS
209 }
210 else {
1604cfb0
MS
211 /* Need to free saved eventually assign to mortal SV */
212 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
213 sv_usepvn(sv, (char *) key, klen);
1b1f1335 214 }
19692e8d 215 if (flags & HVhek_UTF8) {
1604cfb0 216 SvUTF8_on(sv);
1b1f1335 217 }
be2597df 218 Perl_croak(aTHX_ msg, SVfARG(sv));
1b1f1335
NIS
219}
220
fde52b5c 221/* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
222 * contains an SV* */
223
34a6f7b4
NC
224/*
225=for apidoc hv_store
226
a05d6c5d
TC
227Stores an SV in a hash. The hash key is specified as C<key> and the
228absolute value of C<klen> is the length of the key. If C<klen> is
229negative the key is assumed to be in UTF-8-encoded Unicode. The
230C<hash> parameter is the precomputed hash value; if it is zero then
231Perl will compute it.
232
233The return value will be
796b6530 234C<NULL> if the operation failed or if the value did not need to be actually
34a6f7b4
NC
235stored within the hash (as in the case of tied hashes). Otherwise it can
236be dereferenced to get the original C<SV*>. Note that the caller is
237responsible for suitably incrementing the reference count of C<val> before
796b6530
KW
238the call, and decrementing it if the function returned C<NULL>. Effectively
239a successful C<hv_store> takes ownership of one reference to C<val>. This is
34a6f7b4 240usually what you want; a newly created SV has a reference count of one, so
796b6530 241if all your code does is create SVs then store them in a hash, C<hv_store>
34a6f7b4 242will own the only reference to the new SV, and your code doesn't need to do
796b6530
KW
243anything further to tidy up. C<hv_store> is not implemented as a call to
244C<hv_store_ent>, and does not create a temporary SV for the key, so if your
245key data is not already in SV form then use C<hv_store> in preference to
246C<hv_store_ent>.
34a6f7b4
NC
247
248See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
249information on how to use this function on tied hashes.
250
34a6f7b4
NC
251=for apidoc hv_store_ent
252
253Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
254parameter is the precomputed hash value; if it is zero then Perl will
255compute it. The return value is the new hash entry so created. It will be
796b6530 256C<NULL> if the operation failed or if the value did not need to be actually
34a6f7b4
NC
257stored within the hash (as in the case of tied hashes). Otherwise the
258contents of the return value can be accessed using the C<He?> macros
259described here. Note that the caller is responsible for suitably
260incrementing the reference count of C<val> before the call, and
261decrementing it if the function returned NULL. Effectively a successful
796b6530 262C<hv_store_ent> takes ownership of one reference to C<val>. This is
34a6f7b4 263usually what you want; a newly created SV has a reference count of one, so
796b6530 264if all your code does is create SVs then store them in a hash, C<hv_store>
34a6f7b4 265will own the only reference to the new SV, and your code doesn't need to do
796b6530 266anything further to tidy up. Note that C<hv_store_ent> only reads the C<key>;
34a6f7b4 267unlike C<val> it does not take ownership of it, so maintaining the correct
83082a19
KW
268reference count on C<key> is entirely the caller's responsibility. The reason
269it does not take ownership, is that C<key> is not used after this function
270returns, and so can be freed immediately. C<hv_store>
796b6530 271is not implemented as a call to C<hv_store_ent>, and does not create a temporary
34a6f7b4 272SV for the key, so if your key data is not already in SV form then use
796b6530 273C<hv_store> in preference to C<hv_store_ent>.
34a6f7b4
NC
274
275See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
276information on how to use this function on tied hashes.
277
34a6f7b4
NC
278=for apidoc hv_exists
279
280Returns a boolean indicating whether the specified hash key exists. The
a05d6c5d
TC
281absolute value of C<klen> is the length of the key. If C<klen> is
282negative the key is assumed to be in UTF-8-encoded Unicode.
34a6f7b4 283
954c1994
GS
284=for apidoc hv_fetch
285
a05d6c5d
TC
286Returns the SV which corresponds to the specified key in the hash.
287The absolute value of C<klen> is the length of the key. If C<klen> is
288negative the key is assumed to be in UTF-8-encoded Unicode. If
43d3b06a
KW
289C<lval> is set then the fetch will be part of a store. This means that if
290there is no value in the hash associated with the given key, then one is
291created and a pointer to it is returned. The C<SV*> it points to can be
292assigned to. But always check that the
a05d6c5d 293return value is non-null before dereferencing it to an C<SV*>.
954c1994 294
96f1132b 295See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
954c1994
GS
296information on how to use this function on tied hashes.
297
34a6f7b4
NC
298=for apidoc hv_exists_ent
299
db4fbf16
FC
300Returns a boolean indicating whether
301the specified hash key exists. C<hash>
34a6f7b4
NC
302can be a valid precomputed hash value, or 0 to ask for it to be
303computed.
304
305=cut
306*/
307
d1be9408 308/* returns an HE * structure with the all fields set */
fde52b5c 309/* note that hent_val will be a mortal sv for MAGICAL hashes */
954c1994
GS
310/*
311=for apidoc hv_fetch_ent
312
313Returns the hash entry which corresponds to the specified key in the hash.
314C<hash> must be a valid precomputed hash number for the given C<key>, or 0
315if you want the function to compute it. IF C<lval> is set then the fetch
316will be part of a store. Make sure the return value is non-null before
b24b84ef 317accessing it. The return value when C<hv> is a tied hash is a pointer to a
954c1994 318static location, so be sure to make a copy of the structure if you need to
1c846c1f 319store it somewhere.
954c1994 320
96f1132b 321See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
954c1994
GS
322information on how to use this function on tied hashes.
323
324=cut
325*/
326
a038e571
NC
327/* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
328void *
329Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
1604cfb0 330 const int action, SV *val, const U32 hash)
a038e571
NC
331{
332 STRLEN klen;
333 int flags;
334
7918f24d
NC
335 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
336
a038e571 337 if (klen_i32 < 0) {
1604cfb0
MS
338 klen = -klen_i32;
339 flags = HVhek_UTF8;
a038e571 340 } else {
1604cfb0
MS
341 klen = klen_i32;
342 flags = 0;
a038e571
NC
343 }
344 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
345}
346
63c89345 347void *
d3ba3f5c 348Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
1604cfb0 349 int flags, int action, SV *val, U32 hash)
113738bb 350{
b2c64049 351 XPVHV* xhv;
b2c64049
NC
352 HE *entry;
353 HE **oentry;
fde52b5c 354 SV *sv;
da58a35d 355 bool is_utf8;
6f019ba7 356 bool in_collision;
113738bb 357 int masked_flags;
3c84c864 358 const int return_svp = action & HV_FETCH_JUST_SV;
34dadc62 359 HEK *keysv_hek = NULL;
fde52b5c 360
361 if (!hv)
1604cfb0 362 return NULL;
e4787c0c 363 if (SvTYPE(hv) == (svtype)SVTYPEMASK)
1604cfb0 364 return NULL;
8265e3d1
NC
365
366 assert(SvTYPE(hv) == SVt_PVHV);
fde52b5c 367
bdee33e4 368 if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) {
1604cfb0
MS
369 MAGIC* mg;
370 if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) {
371 struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr;
372 if (uf->uf_set == NULL) {
373 SV* obj = mg->mg_obj;
374
375 if (!keysv) {
376 keysv = newSVpvn_flags(key, klen, SVs_TEMP |
377 ((flags & HVhek_UTF8)
378 ? SVf_UTF8 : 0));
379 }
380
381 mg->mg_obj = keysv; /* pass key */
382 uf->uf_index = action; /* pass action */
383 magic_getuvar(MUTABLE_SV(hv), mg);
384 keysv = mg->mg_obj; /* may have changed */
385 mg->mg_obj = obj;
386
387 /* If the key may have changed, then we need to invalidate
388 any passed-in computed hash value. */
389 hash = 0;
390 }
391 }
bdee33e4 392 }
113738bb 393 if (keysv) {
1604cfb0
MS
394 if (flags & HVhek_FREEKEY)
395 Safefree(key);
396 key = SvPV_const(keysv, klen);
397 is_utf8 = (SvUTF8(keysv) != 0);
398 if (SvIsCOW_shared_hash(keysv)) {
399 flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0);
400 } else {
401 flags = 0;
402 }
113738bb 403 } else {
1604cfb0 404 is_utf8 = cBOOL(flags & HVhek_UTF8);
113738bb 405 }
113738bb 406
9dbc5603 407 if (action & HV_DELETE) {
1604cfb0
MS
408 return (void *) hv_delete_common(hv, keysv, key, klen,
409 flags | (is_utf8 ? HVhek_UTF8 : 0),
410 action, hash);
9dbc5603
NC
411 }
412
b2c64049 413 xhv = (XPVHV*)SvANY(hv);
7f66fda2 414 if (SvMAGICAL(hv)) {
1604cfb0
MS
415 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
416 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
417 || SvGMAGICAL((const SV *)hv))
418 {
419 /* FIXME should be able to skimp on the HE/HEK here when
420 HV_FETCH_JUST_SV is true. */
421 if (!keysv) {
422 keysv = newSVpvn_utf8(key, klen, is_utf8);
423 } else {
424 keysv = newSVsv(keysv);
425 }
44a2ac75 426 sv = sv_newmortal();
ad64d0ec 427 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
7f66fda2 428
1604cfb0
MS
429 /* grab a fake HE/HEK pair from the pool or make a new one */
430 entry = PL_hv_fetch_ent_mh;
431 if (entry)
432 PL_hv_fetch_ent_mh = HeNEXT(entry);
433 else {
434 char *k;
435 entry = new_HE();
436 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
437 HeKEY_hek(entry) = (HEK*)k;
438 }
439 HeNEXT(entry) = NULL;
440 HeSVKEY_set(entry, keysv);
441 HeVAL(entry) = sv;
442 sv_upgrade(sv, SVt_PVLV);
443 LvTYPE(sv) = 'T';
444 /* so we can free entry when freeing sv */
445 LvTARG(sv) = MUTABLE_SV(entry);
446
447 /* XXX remove at some point? */
448 if (flags & HVhek_FREEKEY)
449 Safefree(key);
450
451 if (return_svp) {
452 return entry ? (void *) &HeVAL(entry) : NULL;
453 }
454 return (void *) entry;
455 }
7f66fda2 456#ifdef ENV_IS_CASELESS
1604cfb0
MS
457 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
458 U32 i;
459 for (i = 0; i < klen; ++i)
460 if (isLOWER(key[i])) {
461 /* Would be nice if we had a routine to do the
462 copy and upercase in a single pass through. */
463 const char * const nkey = strupr(savepvn(key,klen));
464 /* Note that this fetch is for nkey (the uppercased
465 key) whereas the store is for key (the original) */
466 void *result = hv_common(hv, NULL, nkey, klen,
467 HVhek_FREEKEY, /* free nkey */
468 0 /* non-LVAL fetch */
469 | HV_DISABLE_UVAR_XKEY
470 | return_svp,
471 NULL /* no value */,
472 0 /* compute hash */);
473 if (!result && (action & HV_FETCH_LVALUE)) {
474 /* This call will free key if necessary.
475 Do it this way to encourage compiler to tail
476 call optimise. */
477 result = hv_common(hv, keysv, key, klen, flags,
478 HV_FETCH_ISSTORE
479 | HV_DISABLE_UVAR_XKEY
480 | return_svp,
481 newSV(0), hash);
482 } else {
483 if (flags & HVhek_FREEKEY)
484 Safefree(key);
485 }
486 return result;
487 }
488 }
7f66fda2 489#endif
1604cfb0
MS
490 } /* ISFETCH */
491 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
492 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
493 || SvGMAGICAL((const SV *)hv)) {
494 /* I don't understand why hv_exists_ent has svret and sv,
495 whereas hv_exists only had one. */
496 SV * const svret = sv_newmortal();
497 sv = sv_newmortal();
498
499 if (keysv || is_utf8) {
500 if (!keysv) {
501 keysv = newSVpvn_utf8(key, klen, TRUE);
502 } else {
503 keysv = newSVsv(keysv);
504 }
505 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
506 } else {
507 mg_copy(MUTABLE_SV(hv), sv, key, klen);
508 }
509 if (flags & HVhek_FREEKEY)
510 Safefree(key);
511 {
c818886e
JH
512 MAGIC * const mg = mg_find(sv, PERL_MAGIC_tiedelem);
513 if (mg)
514 magic_existspack(svret, mg);
1604cfb0
MS
515 }
516 /* This cast somewhat evil, but I'm merely using NULL/
517 not NULL to return the boolean exists.
518 And I know hv is not NULL. */
519 return SvTRUE_NN(svret) ? (void *)hv : NULL;
520 }
7f66fda2 521#ifdef ENV_IS_CASELESS
1604cfb0
MS
522 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
523 /* XXX This code isn't UTF8 clean. */
524 char * const keysave = (char * const)key;
525 /* Will need to free this, so set FREEKEY flag. */
526 key = savepvn(key,klen);
527 key = (const char*)strupr((char*)key);
528 is_utf8 = FALSE;
529 hash = 0;
530 keysv = 0;
531
532 if (flags & HVhek_FREEKEY) {
533 Safefree(keysave);
534 }
535 flags |= HVhek_FREEKEY;
536 }
902173a3 537#endif
1604cfb0
MS
538 } /* ISEXISTS */
539 else if (action & HV_FETCH_ISSTORE) {
540 bool needs_copy;
541 bool needs_store;
542 hv_magic_check (hv, &needs_copy, &needs_store);
543 if (needs_copy) {
544 const bool save_taint = TAINT_get;
545 if (keysv || is_utf8) {
546 if (!keysv) {
547 keysv = newSVpvn_utf8(key, klen, TRUE);
548 }
549 if (TAINTING_get)
550 TAINT_set(SvTAINTED(keysv));
551 keysv = sv_2mortal(newSVsv(keysv));
552 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
553 } else {
554 mg_copy(MUTABLE_SV(hv), val, key, klen);
555 }
556
557 TAINT_IF(save_taint);
9a9b5ec9
DM
558#ifdef NO_TAINT_SUPPORT
559 PERL_UNUSED_VAR(save_taint);
560#endif
1604cfb0
MS
561 if (!needs_store) {
562 if (flags & HVhek_FREEKEY)
563 Safefree(key);
564 return NULL;
565 }
b2c64049 566#ifdef ENV_IS_CASELESS
1604cfb0
MS
567 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
568 /* XXX This code isn't UTF8 clean. */
569 const char *keysave = key;
570 /* Will need to free this, so set FREEKEY flag. */
571 key = savepvn(key,klen);
572 key = (const char*)strupr((char*)key);
573 is_utf8 = FALSE;
574 hash = 0;
575 keysv = 0;
576
577 if (flags & HVhek_FREEKEY) {
578 Safefree(keysave);
579 }
580 flags |= HVhek_FREEKEY;
581 }
b2c64049 582#endif
1604cfb0
MS
583 }
584 } /* ISSTORE */
7f66fda2 585 } /* SvMAGICAL */
fde52b5c 586
7b2c381c 587 if (!HvARRAY(hv)) {
1604cfb0 588 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
fde52b5c 589#ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
1604cfb0
MS
590 || (SvRMAGICAL((const SV *)hv)
591 && mg_find((const SV *)hv, PERL_MAGIC_env))
fde52b5c 592#endif
1604cfb0
MS
593 ) {
594 char *array;
595 Newxz(array,
596 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
597 char);
598 HvARRAY(hv) = (HE**)array;
599 }
7f66fda2 600#ifdef DYNAMIC_ENV_FETCH
1604cfb0
MS
601 else if (action & HV_FETCH_ISEXISTS) {
602 /* for an %ENV exists, if we do an insert it's by a recursive
603 store call, so avoid creating HvARRAY(hv) right now. */
604 }
7f66fda2 605#endif
1604cfb0
MS
606 else {
607 /* XXX remove at some point? */
113738bb
NC
608 if (flags & HVhek_FREEKEY)
609 Safefree(key);
610
1604cfb0
MS
611 return NULL;
612 }
fde52b5c 613 }
614
37ae23ff 615 if (is_utf8 && !(flags & HVhek_KEYCANONICAL)) {
1604cfb0
MS
616 char * const keysave = (char *)key;
617 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
19692e8d 618 if (is_utf8)
1604cfb0
MS
619 flags |= HVhek_UTF8;
620 else
621 flags &= ~HVhek_UTF8;
7f66fda2 622 if (key != keysave) {
1604cfb0
MS
623 if (flags & HVhek_FREEKEY)
624 Safefree(keysave);
19692e8d 625 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
1604cfb0
MS
626 /* If the caller calculated a hash, it was on the sequence of
627 octets that are the UTF-8 form. We've now changed the sequence
628 of octets stored to that of the equivalent byte representation,
629 so the hash we need is different. */
630 hash = 0;
631 }
19692e8d 632 }
f9a63242 633
34dadc62
DM
634 if (keysv && (SvIsCOW_shared_hash(keysv))) {
635 if (HvSHAREKEYS(hv))
636 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
637 hash = SvSHARED_HASH(keysv);
7dc86639 638 }
34dadc62
DM
639 else if (!hash)
640 PERL_HASH(hash, key, klen);
effa1e2d 641
113738bb
NC
642 masked_flags = (flags & HVhek_MASK);
643
7f66fda2 644#ifdef DYNAMIC_ENV_FETCH
4608196e 645 if (!HvARRAY(hv)) entry = NULL;
7f66fda2
NC
646 else
647#endif
b2c64049 648 {
1604cfb0 649 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
b2c64049 650 }
34dadc62
DM
651
652 if (!entry)
653 goto not_found;
654
655 if (keysv_hek) {
656 /* keysv is actually a HEK in disguise, so we can match just by
657 * comparing the HEK pointers in the HE chain. There is a slight
658 * caveat: on something like "\x80", which has both plain and utf8
659 * representations, perl's hashes do encoding-insensitive lookups,
660 * but preserve the encoding of the stored key. Thus a particular
661 * key could map to two different HEKs in PL_strtab. We only
662 * conclude 'not found' if all the flags are the same; otherwise
663 * we fall back to a full search (this should only happen in rare
664 * cases).
665 */
666 int keysv_flags = HEK_FLAGS(keysv_hek);
667 HE *orig_entry = entry;
668
669 for (; entry; entry = HeNEXT(entry)) {
670 HEK *hek = HeKEY_hek(entry);
671 if (hek == keysv_hek)
672 goto found;
673 if (HEK_FLAGS(hek) != keysv_flags)
674 break; /* need to do full match */
675 }
676 if (!entry)
677 goto not_found;
678 /* failed on shortcut - do full search loop */
679 entry = orig_entry;
680 }
681
0298d7b9 682 for (; entry; entry = HeNEXT(entry)) {
1604cfb0
MS
683 if (HeHASH(entry) != hash) /* strings can't be equal */
684 continue;
685 if (HeKLEN(entry) != (I32)klen)
686 continue;
687 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
688 continue;
689 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
690 continue;
b2c64049 691
34dadc62 692 found:
b2c64049 693 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
1604cfb0
MS
694 if (HeKFLAGS(entry) != masked_flags) {
695 /* We match if HVhek_UTF8 bit in our flags and hash key's
696 match. But if entry was set previously with HVhek_WASUTF8
697 and key now doesn't (or vice versa) then we should change
698 the key's flag, as this is assignment. */
699 if (HvSHAREKEYS(hv)) {
700 /* Need to swap the key we have for a key with the flags we
701 need. As keys are shared we can't just write to the
702 flag, so we share the new one, unshare the old one. */
703 HEK * const new_hek = share_hek_flags(key, klen, hash,
704 masked_flags);
705 unshare_hek (HeKEY_hek(entry));
706 HeKEY_hek(entry) = new_hek;
707 }
708 else if (hv == PL_strtab) {
709 /* PL_strtab is usually the only hash without HvSHAREKEYS,
710 so putting this test here is cheap */
711 if (flags & HVhek_FREEKEY)
712 Safefree(key);
713 Perl_croak(aTHX_ S_strtab_error,
714 action & HV_FETCH_LVALUE ? "fetch" : "store");
715 }
716 else
717 HeKFLAGS(entry) = masked_flags;
718 if (masked_flags & HVhek_ENABLEHVKFLAGS)
719 HvHASKFLAGS_on(hv);
720 }
721 if (HeVAL(entry) == &PL_sv_placeholder) {
722 /* yes, can store into placeholder slot */
723 if (action & HV_FETCH_LVALUE) {
724 if (SvMAGICAL(hv)) {
725 /* This preserves behaviour with the old hv_fetch
726 implementation which at this point would bail out
727 with a break; (at "if we find a placeholder, we
728 pretend we haven't found anything")
729
730 That break mean that if a placeholder were found, it
731 caused a call into hv_store, which in turn would
732 check magic, and if there is no magic end up pretty
733 much back at this point (in hv_store's code). */
734 break;
735 }
736 /* LVAL fetch which actually needs a store. */
737 val = newSV(0);
738 HvPLACEHOLDERS(hv)--;
739 } else {
740 /* store */
741 if (val != &PL_sv_placeholder)
742 HvPLACEHOLDERS(hv)--;
743 }
744 HeVAL(entry) = val;
745 } else if (action & HV_FETCH_ISSTORE) {
746 SvREFCNT_dec(HeVAL(entry));
747 HeVAL(entry) = val;
748 }
749 } else if (HeVAL(entry) == &PL_sv_placeholder) {
750 /* if we find a placeholder, we pretend we haven't found
751 anything */
752 break;
753 }
754 if (flags & HVhek_FREEKEY)
755 Safefree(key);
756 if (return_svp) {
b66176f0 757 return (void *) &HeVAL(entry);
1604cfb0
MS
758 }
759 return entry;
fde52b5c 760 }
34dadc62
DM
761
762 not_found:
fde52b5c 763#ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
0ed29950 764 if (!(action & HV_FETCH_ISSTORE)
1604cfb0
MS
765 && SvRMAGICAL((const SV *)hv)
766 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
767 unsigned long len;
768 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
769 if (env) {
770 sv = newSVpvn(env,len);
771 SvTAINTED_on(sv);
772 return hv_common(hv, keysv, key, klen, flags,
773 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
774 sv, hash);
775 }
fde52b5c 776 }
777#endif
7f66fda2
NC
778
779 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
1604cfb0
MS
780 hv_notallowed(flags, key, klen,
781 "Attempt to access disallowed key '%" SVf "' in"
782 " a restricted hash");
1b1f1335 783 }
b2c64049 784 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
1604cfb0
MS
785 /* Not doing some form of store, so return failure. */
786 if (flags & HVhek_FREEKEY)
787 Safefree(key);
788 return NULL;
b2c64049 789 }
113738bb 790 if (action & HV_FETCH_LVALUE) {
1604cfb0
MS
791 val = action & HV_FETCH_EMPTY_HE ? NULL : newSV(0);
792 if (SvMAGICAL(hv)) {
793 /* At this point the old hv_fetch code would call to hv_store,
794 which in turn might do some tied magic. So we need to make that
795 magic check happen. */
796 /* gonna assign to this, so it better be there */
797 /* If a fetch-as-store fails on the fetch, then the action is to
798 recurse once into "hv_store". If we didn't do this, then that
799 recursive call would call the key conversion routine again.
800 However, as we replace the original key with the converted
801 key, this would result in a double conversion, which would show
802 up as a bug if the conversion routine is not idempotent.
803 Hence the use of HV_DISABLE_UVAR_XKEY. */
804 return hv_common(hv, keysv, key, klen, flags,
805 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
806 val, hash);
807 /* XXX Surely that could leak if the fetch-was-store fails?
808 Just like the hv_fetch. */
809 }
113738bb
NC
810 }
811
b2c64049
NC
812 /* Welcome to hv_store... */
813
7b2c381c 814 if (!HvARRAY(hv)) {
1604cfb0
MS
815 /* Not sure if we can get here. I think the only case of oentry being
816 NULL is for %ENV with dynamic env fetch. But that should disappear
817 with magic in the previous code. */
818 char *array;
819 Newxz(array,
820 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
821 char);
822 HvARRAY(hv) = (HE**)array;
b2c64049
NC
823 }
824
7b2c381c 825 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
ab4af705 826
b2c64049
NC
827 entry = new_HE();
828 /* share_hek_flags will do the free for us. This might be considered
829 bad API design. */
830 if (HvSHAREKEYS(hv))
1604cfb0 831 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
5d2b1485 832 else if (hv == PL_strtab) {
1604cfb0
MS
833 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
834 this test here is cheap */
835 if (flags & HVhek_FREEKEY)
836 Safefree(key);
837 Perl_croak(aTHX_ S_strtab_error,
838 action & HV_FETCH_LVALUE ? "fetch" : "store");
5d2b1485 839 }
b2c64049 840 else /* gotta do the real thing */
1604cfb0 841 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
b2c64049 842 HeVAL(entry) = val;
3078e109 843
6a5b4183 844#ifdef PERL_HASH_RANDOMIZE_KEYS
3078e109
YO
845 /* This logic semi-randomizes the insert order in a bucket.
846 * Either we insert into the top, or the slot below the top,
d5fc06cb
YO
847 * making it harder to see if there is a collision. We also
848 * reset the iterator randomizer if there is one.
3078e109 849 */
6f019ba7 850 in_collision = *oentry != NULL;
6a5b4183
YO
851 if ( *oentry && PL_HASH_RAND_BITS_ENABLED) {
852 PL_hash_rand_bits++;
853 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
854 if ( PL_hash_rand_bits & 1 ) {
855 HeNEXT(entry) = HeNEXT(*oentry);
856 HeNEXT(*oentry) = entry;
857 } else {
858 HeNEXT(entry) = *oentry;
859 *oentry = entry;
860 }
861 } else
862#endif
863 {
3078e109
YO
864 HeNEXT(entry) = *oentry;
865 *oentry = entry;
3078e109 866 }
6a5b4183 867#ifdef PERL_HASH_RANDOMIZE_KEYS
3a714294 868 if (SvOOK(hv)) {
ff20b672
YO
869 /* Currently this makes various tests warn in annoying ways.
870 * So Silenced for now. - Yves | bogus end of comment =>* /
871 if (HvAUX(hv)->xhv_riter != -1) {
872 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
873 "[TESTING] Inserting into a hash during each() traversal results in undefined behavior"
874 pTHX__FORMAT
875 pTHX__VALUE);
876 }
877 */
6a5b4183
YO
878 if (PL_HASH_RAND_BITS_ENABLED) {
879 if (PL_HASH_RAND_BITS_ENABLED == 1)
880 PL_hash_rand_bits += (PTRV)entry + 1; /* we don't bother to use ptr_hash here */
881 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
882 }
3a714294
YO
883 HvAUX(hv)->xhv_rand= (U32)PL_hash_rand_bits;
884 }
6a5b4183 885#endif
b2c64049
NC
886
887 if (val == &PL_sv_placeholder)
1604cfb0 888 HvPLACEHOLDERS(hv)++;
b2c64049 889 if (masked_flags & HVhek_ENABLEHVKFLAGS)
1604cfb0 890 HvHASKFLAGS_on(hv);
b2c64049 891
8e317198 892 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
6f019ba7 893 if ( in_collision && DO_HSPLIT(xhv) ) {
adf6906b 894 const STRLEN oldsize = xhv->xhv_max + 1;
81a3ba35 895 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
adf6906b 896
81a3ba35 897 if (items /* hash has placeholders */
1eaee784
NC
898 && !SvREADONLY(hv) /* but is not a restricted hash */) {
899 /* If this hash previously was a "restricted hash" and had
900 placeholders, but the "restricted" flag has been turned off,
901 then the placeholders no longer serve any useful purpose.
902 However, they have the downsides of taking up RAM, and adding
903 extra steps when finding used values. It's safe to clear them
904 at this point, even though Storable rebuilds restricted hashes by
0ca1b5c3 905 putting in all the placeholders (first) before turning on the
1eaee784
NC
906 readonly flag, because Storable always pre-splits the hash.
907 If we're lucky, then we may clear sufficient placeholders to
908 avoid needing to split the hash at all. */
81a3ba35 909 clear_placeholders(hv, items);
1eaee784
NC
910 if (DO_HSPLIT(xhv))
911 hsplit(hv, oldsize, oldsize * 2);
912 } else
913 hsplit(hv, oldsize, oldsize * 2);
fde52b5c 914 }
b2c64049 915
3c84c864 916 if (return_svp) {
1604cfb0 917 return entry ? (void *) &HeVAL(entry) : NULL;
3c84c864
NC
918 }
919 return (void *) entry;
fde52b5c 920}
921
864dbfa3 922STATIC void
b0e6ae5b 923S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
d0066dc7 924{
a3b680e6 925 const MAGIC *mg = SvMAGIC(hv);
7918f24d
NC
926
927 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
928
d0066dc7
OT
929 *needs_copy = FALSE;
930 *needs_store = TRUE;
931 while (mg) {
1604cfb0
MS
932 if (isUPPER(mg->mg_type)) {
933 *needs_copy = TRUE;
934 if (mg->mg_type == PERL_MAGIC_tied) {
935 *needs_store = FALSE;
936 return; /* We've set all there is to set. */
937 }
938 }
939 mg = mg->mg_moremagic;
d0066dc7
OT
940 }
941}
942
954c1994 943/*
a3bcc51e
TP
944=for apidoc hv_scalar
945
8bf4c401
YO
946Evaluates the hash in scalar context and returns the result.
947
948When the hash is tied dispatches through to the SCALAR method,
949otherwise returns a mortal SV containing the number of keys
950in the hash.
951
952Note, prior to 5.25 this function returned what is now
953returned by the hv_bucket_ratio() function.
a3bcc51e
TP
954
955=cut
956*/
957
958SV *
959Perl_hv_scalar(pTHX_ HV *hv)
960{
a3bcc51e 961 SV *sv;
823a54a3 962
7918f24d
NC
963 PERL_ARGS_ASSERT_HV_SCALAR;
964
823a54a3 965 if (SvRMAGICAL(hv)) {
1604cfb0
MS
966 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
967 if (mg)
968 return magic_scalarpack(hv, mg);
823a54a3 969 }
a3bcc51e
TP
970
971 sv = sv_newmortal();
8bf4c401
YO
972 sv_setuv(sv, HvUSEDKEYS(hv));
973
974 return sv;
975}
976
af3b1cba
DM
977
978/*
8dc9003f
DM
979hv_pushkv(): push all the keys and/or values of a hash onto the stack.
980The rough Perl equivalents:
981 () = %hash;
982 () = keys %hash;
983 () = values %hash;
984
af3b1cba 985Resets the hash's iterator.
8dc9003f
DM
986
987flags : 1 = push keys
988 2 = push values
989 1|2 = push keys and values
990 XXX use symbolic flag constants at some point?
991I might unroll the non-tied hv_iternext() in here at some point - DAPM
af3b1cba
DM
992*/
993
994void
8dc9003f 995Perl_hv_pushkv(pTHX_ HV *hv, U32 flags)
af3b1cba
DM
996{
997 HE *entry;
61084eef
CB
998 bool tied = SvRMAGICAL(hv) && (mg_find(MUTABLE_SV(hv), PERL_MAGIC_tied)
999#ifdef DYNAMIC_ENV_FETCH /* might not know number of keys yet */
1000 || mg_find(MUTABLE_SV(hv), PERL_MAGIC_env)
1001#endif
1002 );
af3b1cba
DM
1003 dSP;
1004
1005 PERL_ARGS_ASSERT_HV_PUSHKV;
8dc9003f 1006 assert(flags); /* must be pushing at least one of keys and values */
af3b1cba
DM
1007
1008 (void)hv_iterinit(hv);
1009
1010 if (tied) {
8dc9003f 1011 SSize_t ext = (flags == 3) ? 2 : 1;
af3b1cba 1012 while ((entry = hv_iternext(hv))) {
8dc9003f
DM
1013 EXTEND(SP, ext);
1014 if (flags & 1)
1015 PUSHs(hv_iterkeysv(entry));
1016 if (flags & 2)
1017 PUSHs(hv_iterval(hv, entry));
af3b1cba
DM
1018 }
1019 }
1020 else {
87c9dcef 1021 Size_t nkeys = HvUSEDKEYS(hv);
8dc9003f
DM
1022 SSize_t ext;
1023
1024 if (!nkeys)
1025 return;
1026
87c9dcef
DM
1027 /* 2*nkeys() should never be big enough to truncate or wrap */
1028 assert(nkeys <= (SSize_t_MAX >> 1));
8dc9003f 1029 ext = nkeys * ((flags == 3) ? 2 : 1);
87c9dcef
DM
1030
1031 EXTEND_MORTAL(nkeys);
8dc9003f 1032 EXTEND(SP, ext);
af3b1cba
DM
1033
1034 while ((entry = hv_iternext(hv))) {
8dc9003f
DM
1035 if (flags & 1) {
1036 SV *keysv = newSVhek(HeKEY_hek(entry));
1037 SvTEMP_on(keysv);
1038 PL_tmps_stack[++PL_tmps_ix] = keysv;
1039 PUSHs(keysv);
1040 }
1041 if (flags & 2)
1042 PUSHs(HeVAL(entry));
af3b1cba
DM
1043 }
1044 }
1045
1046 PUTBACK;
1047}
1048
1049
8bf4c401 1050/*
7d7345cf 1051=for apidoc hv_bucket_ratio
8bf4c401
YO
1052
1053If the hash is tied dispatches through to the SCALAR tied method,
1054otherwise if the hash contains no keys returns 0, otherwise returns
1055a mortal sv containing a string specifying the number of used buckets,
1056followed by a slash, followed by the number of available buckets.
1057
1058This function is expensive, it must scan all of the buckets
1059to determine which are used, and the count is NOT cached.
1060In a large hash this could be a lot of buckets.
1061
1062=cut
1063*/
1064
1065SV *
1066Perl_hv_bucket_ratio(pTHX_ HV *hv)
1067{
1068 SV *sv;
1069
1070 PERL_ARGS_ASSERT_HV_BUCKET_RATIO;
1071
1072 if (SvRMAGICAL(hv)) {
1073 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
1074 if (mg)
1075 return magic_scalarpack(hv, mg);
1076 }
1077
725c44f9
DM
1078 if (HvUSEDKEYS((HV *)hv)) {
1079 sv = sv_newmortal();
a3bcc51e
TP
1080 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
1081 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
725c44f9 1082 }
a3bcc51e 1083 else
725c44f9 1084 sv = &PL_sv_zero;
a3bcc51e
TP
1085
1086 return sv;
1087}
1088
1089/*
954c1994
GS
1090=for apidoc hv_delete
1091
a05d6c5d
TC
1092Deletes a key/value pair in the hash. The value's SV is removed from
1093the hash, made mortal, and returned to the caller. The absolute
1094value of C<klen> is the length of the key. If C<klen> is negative the
1095key is assumed to be in UTF-8-encoded Unicode. The C<flags> value
796b6530
KW
1096will normally be zero; if set to C<G_DISCARD> then C<NULL> will be returned.
1097C<NULL> will also be returned if the key is not found.
954c1994 1098
954c1994
GS
1099=for apidoc hv_delete_ent
1100
3025a2e4
CS
1101Deletes a key/value pair in the hash. The value SV is removed from the hash,
1102made mortal, and returned to the caller. The C<flags> value will normally be
796b6530
KW
1103zero; if set to C<G_DISCARD> then C<NULL> will be returned. C<NULL> will also
1104be returned if the key is not found. C<hash> can be a valid precomputed hash
3025a2e4 1105value, or 0 to ask for it to be computed.
954c1994
GS
1106
1107=cut
1108*/
1109
8f8d40ab 1110STATIC SV *
cd6d36ac 1111S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
1604cfb0 1112 int k_flags, I32 d_flags, U32 hash)
f1317c8d 1113{
eb578fdb
KW
1114 XPVHV* xhv;
1115 HE *entry;
1116 HE **oentry;
34dadc62 1117 HE **first_entry;
8298454c 1118 bool is_utf8 = cBOOL(k_flags & HVhek_UTF8);
7a9669ca 1119 int masked_flags;
34dadc62
DM
1120 HEK *keysv_hek = NULL;
1121 U8 mro_changes = 0; /* 1 = isa; 2 = package moved */
1122 SV *sv;
1123 GV *gv = NULL;
1124 HV *stash = NULL;
1c846c1f 1125
fde52b5c 1126 if (SvRMAGICAL(hv)) {
1604cfb0
MS
1127 bool needs_copy;
1128 bool needs_store;
1129 hv_magic_check (hv, &needs_copy, &needs_store);
1130
1131 if (needs_copy) {
1132 SV *sv;
1133 entry = (HE *) hv_common(hv, keysv, key, klen,
1134 k_flags & ~HVhek_FREEKEY,
1135 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
1136 NULL, hash);
1137 sv = entry ? HeVAL(entry) : NULL;
1138 if (sv) {
1139 if (SvMAGICAL(sv)) {
1140 mg_clear(sv);
1141 }
1142 if (!needs_store) {
1143 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
1144 /* No longer an element */
1145 sv_unmagic(sv, PERL_MAGIC_tiedelem);
1146 return sv;
1147 }
1148 return NULL; /* element cannot be deleted */
1149 }
902173a3 1150#ifdef ENV_IS_CASELESS
1604cfb0
MS
1151 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
1152 /* XXX This code isn't UTF8 clean. */
1153 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
1154 if (k_flags & HVhek_FREEKEY) {
1155 Safefree(key);
1156 }
1157 key = strupr(SvPVX(keysv));
1158 is_utf8 = 0;
1159 k_flags = 0;
1160 hash = 0;
1161 }
510ac311 1162#endif
1604cfb0
MS
1163 }
1164 }
fde52b5c 1165 }
cbec9347 1166 xhv = (XPVHV*)SvANY(hv);
7b2c381c 1167 if (!HvARRAY(hv))
1604cfb0 1168 return NULL;
fde52b5c 1169
6b230254 1170 if (is_utf8 && !(k_flags & HVhek_KEYCANONICAL)) {
1604cfb0
MS
1171 const char * const keysave = key;
1172 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
cd6d36ac 1173
19692e8d 1174 if (is_utf8)
cd6d36ac 1175 k_flags |= HVhek_UTF8;
1604cfb0 1176 else
cd6d36ac 1177 k_flags &= ~HVhek_UTF8;
7f66fda2 1178 if (key != keysave) {
1604cfb0
MS
1179 if (k_flags & HVhek_FREEKEY) {
1180 /* This shouldn't happen if our caller does what we expect,
1181 but strictly the API allows it. */
1182 Safefree(keysave);
1183 }
1184 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
1185 }
ad64d0ec 1186 HvHASKFLAGS_on(MUTABLE_SV(hv));
19692e8d 1187 }
f9a63242 1188
34dadc62
DM
1189 if (keysv && (SvIsCOW_shared_hash(keysv))) {
1190 if (HvSHAREKEYS(hv))
1191 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
1192 hash = SvSHARED_HASH(keysv);
7dc86639 1193 }
34dadc62
DM
1194 else if (!hash)
1195 PERL_HASH(hash, key, klen);
fde52b5c 1196
7a9669ca
NC
1197 masked_flags = (k_flags & HVhek_MASK);
1198
9faf471a 1199 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
fde52b5c 1200 entry = *oentry;
0c3bb3c2 1201
34dadc62
DM
1202 if (!entry)
1203 goto not_found;
1204
1205 if (keysv_hek) {
1206 /* keysv is actually a HEK in disguise, so we can match just by
1207 * comparing the HEK pointers in the HE chain. There is a slight
1208 * caveat: on something like "\x80", which has both plain and utf8
1209 * representations, perl's hashes do encoding-insensitive lookups,
1210 * but preserve the encoding of the stored key. Thus a particular
1211 * key could map to two different HEKs in PL_strtab. We only
1212 * conclude 'not found' if all the flags are the same; otherwise
1213 * we fall back to a full search (this should only happen in rare
1214 * cases).
1215 */
1216 int keysv_flags = HEK_FLAGS(keysv_hek);
1217
1218 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1219 HEK *hek = HeKEY_hek(entry);
1220 if (hek == keysv_hek)
1221 goto found;
1222 if (HEK_FLAGS(hek) != keysv_flags)
1223 break; /* need to do full match */
1224 }
1225 if (!entry)
1226 goto not_found;
1227 /* failed on shortcut - do full search loop */
1228 oentry = first_entry;
1229 entry = *oentry;
1230 }
1231
1232 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1604cfb0
MS
1233 if (HeHASH(entry) != hash) /* strings can't be equal */
1234 continue;
1235 if (HeKLEN(entry) != (I32)klen)
1236 continue;
1237 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
1238 continue;
1239 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
1240 continue;
8aacddc1 1241
34dadc62 1242 found:
1604cfb0
MS
1243 if (hv == PL_strtab) {
1244 if (k_flags & HVhek_FREEKEY)
1245 Safefree(key);
1246 Perl_croak(aTHX_ S_strtab_error, "delete");
1247 }
1248
1249 /* if placeholder is here, it's already been deleted.... */
1250 if (HeVAL(entry) == &PL_sv_placeholder) {
1251 if (k_flags & HVhek_FREEKEY)
1252 Safefree(key);
1253 return NULL;
1254 }
1255 if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
1256 hv_notallowed(k_flags, key, klen,
1257 "Attempt to delete readonly key '%" SVf "' from"
1258 " a restricted hash");
1259 }
b84d0860
NC
1260 if (k_flags & HVhek_FREEKEY)
1261 Safefree(key);
8aacddc1 1262
1604cfb0
MS
1263 /* If this is a stash and the key ends with ::, then someone is
1264 * deleting a package.
1265 */
1266 if (HeVAL(entry) && HvENAME_get(hv)) {
1267 gv = (GV *)HeVAL(entry);
1268 if (keysv) key = SvPV(keysv, klen);
1269 if ((
1270 (klen > 1 && key[klen-2] == ':' && key[klen-1] == ':')
1271 ||
1272 (klen == 1 && key[0] == ':')
1273 )
1274 && (klen != 6 || hv!=PL_defstash || memNE(key,"main::",6))
1275 && SvTYPE(gv) == SVt_PVGV && (stash = GvHV((GV *)gv))
1276 && HvENAME_get(stash)) {
1277 /* A previous version of this code checked that the
1278 * GV was still in the symbol table by fetching the
1279 * GV with its name. That is not necessary (and
1280 * sometimes incorrect), as HvENAME cannot be set
1281 * on hv if it is not in the symtab. */
1282 mro_changes = 2;
1283 /* Hang on to it for a bit. */
1284 SvREFCNT_inc_simple_void_NN(
1285 sv_2mortal((SV *)gv)
1286 );
1287 }
1288 else if (memEQs(key, klen, "ISA") && GvAV(gv)) {
6146d9e1
TC
1289 AV *isa = GvAV(gv);
1290 MAGIC *mg = mg_find((SV*)isa, PERL_MAGIC_isa);
1291
1604cfb0 1292 mro_changes = 1;
6146d9e1
TC
1293 if (mg) {
1294 if (mg->mg_obj == (SV*)gv) {
1295 /* This is the only stash this ISA was used for.
1296 * The isaelem magic asserts if there's no
1297 * isa magic on the array, so explicitly
1298 * remove the magic on both the array and its
1299 * elements. @ISA shouldn't be /too/ large.
1300 */
1301 SV **svp, **end;
1302 strip_magic:
1303 svp = AvARRAY(isa);
9ba9a28a 1304 end = svp + (AvFILLp(isa)+1);
6146d9e1
TC
1305 while (svp < end) {
1306 if (*svp)
1307 mg_free_type(*svp, PERL_MAGIC_isaelem);
1308 ++svp;
1309 }
1310 mg_free_type((SV*)GvAV(gv), PERL_MAGIC_isa);
1311 }
1312 else {
1313 /* mg_obj is an array of stashes
1314 Note that the array doesn't keep a reference
1315 count on the stashes.
1316 */
1317 AV *av = (AV*)mg->mg_obj;
1318 SV **svp, **arrayp;
1319 SSize_t index;
1320 SSize_t items;
1321
1322 assert(SvTYPE(mg->mg_obj) == SVt_PVAV);
1323
1324 /* remove the stash from the magic array */
1325 arrayp = svp = AvARRAY(av);
1326 items = AvFILLp(av) + 1;
1327 if (items == 1) {
1328 assert(*arrayp == (SV *)gv);
1329 mg->mg_obj = NULL;
1330 /* avoid a double free on the last stash */
1331 AvFILLp(av) = -1;
1332 /* The magic isn't MGf_REFCOUNTED, so release
1333 * the array manually.
1334 */
1335 SvREFCNT_dec_NN(av);
1336 goto strip_magic;
1337 }
1338 else {
1339 while (items--) {
1340 if (*svp == (SV*)gv)
1341 break;
1342 ++svp;
1343 }
1344 index = svp - arrayp;
1345 assert(index >= 0 && index <= AvFILLp(av));
1346 if (index < AvFILLp(av)) {
1347 arrayp[index] = arrayp[AvFILLp(av)];
1348 }
1349 arrayp[AvFILLp(av)] = NULL;
1350 --AvFILLp(av);
1351 }
1352 }
1353 }
1354 }
1604cfb0
MS
1355 }
1356
1357 sv = d_flags & G_DISCARD ? HeVAL(entry) : sv_2mortal(HeVAL(entry));
1358 HeVAL(entry) = &PL_sv_placeholder;
1359 if (sv) {
1360 /* deletion of method from stash */
1361 if (isGV(sv) && isGV_with_GP(sv) && GvCVu(sv)
1362 && HvENAME_get(hv))
1363 mro_method_changed_in(hv);
1364 }
1365
1366 /*
1367 * If a restricted hash, rather than really deleting the entry, put
1368 * a placeholder there. This marks the key as being "approved", so
1369 * we can still access via not-really-existing key without raising
1370 * an error.
1371 */
1372 if (SvREADONLY(hv))
1373 /* We'll be saving this slot, so the number of allocated keys
1374 * doesn't go down, but the number placeholders goes up */
1375 HvPLACEHOLDERS(hv)++;
1376 else {
1377 *oentry = HeNEXT(entry);
1378 if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */)
1379 HvLAZYDEL_on(hv);
1380 else {
1381 if (SvOOK(hv) && HvLAZYDEL(hv) &&
1382 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
1383 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
1384 hv_free_ent(hv, entry);
1385 }
1386 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
1387 if (xhv->xhv_keys == 0)
1388 HvHASKFLAGS_off(hv);
1389 }
1390
1391 if (d_flags & G_DISCARD) {
1392 SvREFCNT_dec(sv);
1393 sv = NULL;
1394 }
1395
1396 if (mro_changes == 1) mro_isa_changed_in(hv);
1397 else if (mro_changes == 2)
1398 mro_package_moved(NULL, stash, gv, 1);
1399
1400 return sv;
79072805 1401 }
34dadc62
DM
1402
1403 not_found:
8aacddc1 1404 if (SvREADONLY(hv)) {
1604cfb0
MS
1405 hv_notallowed(k_flags, key, klen,
1406 "Attempt to delete disallowed key '%" SVf "' from"
1407 " a restricted hash");
8aacddc1
NIS
1408 }
1409
19692e8d 1410 if (k_flags & HVhek_FREEKEY)
1604cfb0 1411 Safefree(key);
a0714e2c 1412 return NULL;
79072805
LW
1413}
1414
32dfa2a7 1415
76e3520e 1416STATIC void
adf6906b 1417S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize)
79072805 1418{
7663aa67 1419 STRLEN i = 0;
7b2c381c 1420 char *a = (char*) HvARRAY(hv);
eb578fdb 1421 HE **aep;
79072805 1422
32dfa2a7
YO
1423 bool do_aux= (
1424 /* already have an HvAUX(hv) so we have to move it */
1425 SvOOK(hv) ||
1426 /* no HvAUX() but array we are going to allocate is large enough
1427 * there is no point in saving the space for the iterator, and
1428 * speeds up later traversals. */
1429 ( ( hv != PL_strtab ) && ( newsize >= PERL_HV_ALLOC_AUX_SIZE ) )
1430 );
7918f24d 1431
32dfa2a7 1432 PERL_ARGS_ASSERT_HSPLIT;
aae087f7
YO
1433 if (newsize > MAX_BUCKET_MAX+1)
1434 return;
18026298 1435
3280af22 1436 PL_nomemok = TRUE;
b79f7545 1437 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
32dfa2a7
YO
1438 + (do_aux ? sizeof(struct xpvhv_aux) : 0), char);
1439 PL_nomemok = FALSE;
422a93e5 1440 if (!a) {
422a93e5
GA
1441 return;
1442 }
32dfa2a7 1443
6a5b4183 1444#ifdef PERL_HASH_RANDOMIZE_KEYS
3078e109
YO
1445 /* the idea of this is that we create a "random" value by hashing the address of
1446 * the array, we then use the low bit to decide if we insert at the top, or insert
1447 * second from top. After each such insert we rotate the hashed value. So we can
1448 * use the same hashed value over and over, and in normal build environments use
1449 * very few ops to do so. ROTL32() should produce a single machine operation. */
6a5b4183
YO
1450 if (PL_HASH_RAND_BITS_ENABLED) {
1451 if (PL_HASH_RAND_BITS_ENABLED == 1)
1452 PL_hash_rand_bits += ptr_hash((PTRV)a);
1453 PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1);
1454 }
1455#endif
32dfa2a7
YO
1456 HvARRAY(hv) = (HE**) a;
1457 HvMAX(hv) = newsize - 1;
1458 /* before we zero the newly added memory, we
1459 * need to deal with the aux struct that may be there
1460 * or have been allocated by us*/
1461 if (do_aux) {
3078e109
YO
1462 struct xpvhv_aux *const dest
1463 = (struct xpvhv_aux*) &a[newsize * sizeof(HE*)];
32dfa2a7
YO
1464 if (SvOOK(hv)) {
1465 /* alread have an aux, copy the old one in place. */
1466 Move(&a[oldsize * sizeof(HE*)], dest, 1, struct xpvhv_aux);
1467 /* we reset the iterator's xhv_rand as well, so they get a totally new ordering */
6a5b4183 1468#ifdef PERL_HASH_RANDOMIZE_KEYS
32dfa2a7 1469 dest->xhv_rand = (U32)PL_hash_rand_bits;
6a5b4183 1470#endif
32dfa2a7
YO
1471 } else {
1472 /* no existing aux structure, but we allocated space for one
f6bab5f6 1473 * so initialize it properly. This unrolls hv_auxinit() a bit,
32dfa2a7
YO
1474 * since we have to do the realloc anyway. */
1475 /* first we set the iterator's xhv_rand so it can be copied into lastrand below */
1476#ifdef PERL_HASH_RANDOMIZE_KEYS
1477 dest->xhv_rand = (U32)PL_hash_rand_bits;
1478#endif
1479 /* this is the "non realloc" part of the hv_auxinit() */
1480 (void)hv_auxinit_internal(dest);
1481 /* Turn on the OOK flag */
1482 SvOOK_on(hv);
1483 }
b79f7545 1484 }
32dfa2a7 1485 /* now we can safely clear the second half */
72311751 1486 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
79072805 1487
68303b5c
NC
1488 if (!HvTOTALKEYS(hv)) /* skip rest if no entries */
1489 return;
1490
32dfa2a7 1491 newsize--;
68303b5c 1492 aep = (HE**)a;
7663aa67 1493 do {
1604cfb0
MS
1494 HE **oentry = aep + i;
1495 HE *entry = aep[i];
4b5190b5 1496
1604cfb0
MS
1497 if (!entry) /* non-existent */
1498 continue;
1499 do {
c23dc12b 1500 U32 j = (HeHASH(entry) & newsize);
1604cfb0
MS
1501 if (j != (U32)i) {
1502 *oentry = HeNEXT(entry);
6a5b4183
YO
1503#ifdef PERL_HASH_RANDOMIZE_KEYS
1504 /* if the target cell is empty or PL_HASH_RAND_BITS_ENABLED is false
1505 * insert to top, otherwise rotate the bucket rand 1 bit,
1506 * and use the new low bit to decide if we insert at top,
1507 * or next from top. IOW, we only rotate on a collision.*/
1508 if (aep[j] && PL_HASH_RAND_BITS_ENABLED) {
3f49e765 1509 PL_hash_rand_bits+= ROTL32(HeHASH(entry), 17);
6a5b4183
YO
1510 PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1);
1511 if (PL_hash_rand_bits & 1) {
1512 HeNEXT(entry)= HeNEXT(aep[j]);
1513 HeNEXT(aep[j])= entry;
1514 } else {
1515 /* Note, this is structured in such a way as the optimizer
1516 * should eliminate the duplicated code here and below without
1517 * us needing to explicitly use a goto. */
1518 HeNEXT(entry) = aep[j];
1519 aep[j] = entry;
1520 }
1521 } else
1522#endif
1523 {
1524 /* see comment above about duplicated code */
3078e109
YO
1525 HeNEXT(entry) = aep[j];
1526 aep[j] = entry;
3078e109 1527 }
1604cfb0
MS
1528 }
1529 else {
1530 oentry = &HeNEXT(entry);
1531 }
1532 entry = *oentry;
1533 } while (entry);
7663aa67 1534 } while (i++ < oldsize);
79072805
LW
1535}
1536
72940dca 1537void
864dbfa3 1538Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
72940dca 1539{
eb578fdb 1540 XPVHV* xhv = (XPVHV*)SvANY(hv);
6f019ba7 1541 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 */
eb578fdb 1542 I32 newsize;
6f019ba7
YO
1543 I32 wantsize;
1544 I32 trysize;
eb578fdb 1545 char *a;
72940dca 1546
7918f24d
NC
1547 PERL_ARGS_ASSERT_HV_KSPLIT;
1548
6f019ba7
YO
1549 wantsize = (I32) newmax; /* possible truncation here */
1550 if (wantsize != newmax)
1604cfb0 1551 return;
6f019ba7
YO
1552
1553 wantsize= wantsize + (wantsize >> 1); /* wantsize *= 1.5 */
1554 if (wantsize < newmax) /* overflow detection */
1555 return;
1556
1557 newsize = oldsize;
1558 while (wantsize > newsize) {
1559 trysize = newsize << 1;
1560 if (trysize > newsize) {
1561 newsize = trysize;
1562 } else {
1563 /* we overflowed */
1564 return;
1565 }
72940dca 1566 }
6f019ba7
YO
1567
1568 if (newsize <= oldsize)
1569 return; /* overflow detection */
72940dca 1570
7b2c381c 1571 a = (char *) HvARRAY(hv);
e8c10cf3
NC
1572 if (a) {
1573 hsplit(hv, oldsize, newsize);
1574 } else {
0df05616 1575 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
6f019ba7 1576 xhv->xhv_max = newsize - 1;
0df05616 1577 HvARRAY(hv) = (HE **) a;
72940dca 1578 }
1579}
1580
f6bb1c88
YO
1581/* IMO this should also handle cases where hv_max is smaller than hv_keys
1582 * as tied hashes could play silly buggers and mess us around. We will
1583 * do the right thing during hv_store() afterwards, but still - Yves */
1584#define HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys) STMT_START {\
1585 /* Can we use fewer buckets? (hv_max is always 2^n-1) */ \
1586 if (hv_max < PERL_HASH_DEFAULT_HvMAX) { \
1587 hv_max = PERL_HASH_DEFAULT_HvMAX; \
1588 } else { \
1589 while (hv_max > PERL_HASH_DEFAULT_HvMAX && hv_max + 1 >= hv_keys * 2) \
1590 hv_max = hv_max / 2; \
1591 } \
1592 HvMAX(hv) = hv_max; \
1593} STMT_END
1594
1595
b3ac6de7 1596HV *
864dbfa3 1597Perl_newHVhv(pTHX_ HV *ohv)
b3ac6de7 1598{
9d4ba2ae 1599 HV * const hv = newHV();
f4431c56 1600 STRLEN hv_max;
4beac62f 1601
3f4d1d78 1602 if (!ohv || (!HvTOTALKEYS(ohv) && !SvMAGICAL((const SV *)ohv)))
1604cfb0 1603 return hv;
4beac62f 1604 hv_max = HvMAX(ohv);
b3ac6de7 1605
ad64d0ec 1606 if (!SvMAGICAL((const SV *)ohv)) {
1604cfb0
MS
1607 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
1608 STRLEN i;
1609 const bool shared = !!HvSHAREKEYS(ohv);
1610 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
1611 char *a;
1612 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
1613 ents = (HE**)a;
1614
1615 /* In each bucket... */
1616 for (i = 0; i <= hv_max; i++) {
1617 HE *prev = NULL;
1618 HE *oent = oents[i];
1619
1620 if (!oent) {
1621 ents[i] = NULL;
1622 continue;
1623 }
1624
1625 /* Copy the linked list of entries. */
1626 for (; oent; oent = HeNEXT(oent)) {
1627 const U32 hash = HeHASH(oent);
1628 const char * const key = HeKEY(oent);
1629 const STRLEN len = HeKLEN(oent);
1630 const int flags = HeKFLAGS(oent);
1631 HE * const ent = new_HE();
1632 SV *const val = HeVAL(oent);
1633
1634 HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val);
1635 HeKEY_hek(ent)
6e838c70 1636 = shared ? share_hek_flags(key, len, hash, flags)
19692e8d 1637 : save_hek_flags(key, len, hash, flags);
1604cfb0
MS
1638 if (prev)
1639 HeNEXT(prev) = ent;
1640 else
1641 ents[i] = ent;
1642 prev = ent;
1643 HeNEXT(ent) = NULL;
1644 }
1645 }
1646
1647 HvMAX(hv) = hv_max;
1648 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
1649 HvARRAY(hv) = ents;
aec46f14 1650 } /* not magical */
b56ba0bf 1651 else {
1604cfb0
MS
1652 /* Iterate over ohv, copying keys and values one at a time. */
1653 HE *entry;
1654 const I32 riter = HvRITER_get(ohv);
1655 HE * const eiter = HvEITER_get(ohv);
f6bb1c88 1656 STRLEN hv_keys = HvTOTALKEYS(ohv);
b56ba0bf 1657
f6bb1c88 1658 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
b56ba0bf 1659
1604cfb0
MS
1660 hv_iterinit(ohv);
1661 while ((entry = hv_iternext_flags(ohv, 0))) {
1662 SV *val = hv_iterval(ohv,entry);
1663 SV * const keysv = HeSVKEY(entry);
1664 val = SvIMMORTAL(val) ? val : newSVsv(val);
1665 if (keysv)
1666 (void)hv_store_ent(hv, keysv, val, 0);
1667 else
1668 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry), val,
1669 HeHASH(entry), HeKFLAGS(entry));
1670 }
1671 HvRITER_set(ohv, riter);
1672 HvEITER_set(ohv, eiter);
b3ac6de7 1673 }
1c846c1f 1674
b3ac6de7
IZ
1675 return hv;
1676}
1677
defdfed5 1678/*
44170c9a 1679=for apidoc hv_copy_hints_hv
defdfed5 1680
2d7f6611 1681A specialised version of L</newHVhv> for copying C<%^H>. C<ohv> must be
defdfed5
Z
1682a pointer to a hash (which may have C<%^H> magic, but should be generally
1683non-magical), or C<NULL> (interpreted as an empty hash). The content
2d7f6611 1684of C<ohv> is copied to a new hash, which has the C<%^H>-specific magic
defdfed5
Z
1685added to it. A pointer to the new hash is returned.
1686
1687=cut
1688*/
1689
5b9c0671
NC
1690HV *
1691Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1692{
1693 HV * const hv = newHV();
5b9c0671 1694
cb1f05e8 1695 if (ohv) {
1604cfb0 1696 STRLEN hv_max = HvMAX(ohv);
f6bb1c88 1697 STRLEN hv_keys = HvTOTALKEYS(ohv);
1604cfb0
MS
1698 HE *entry;
1699 const I32 riter = HvRITER_get(ohv);
1700 HE * const eiter = HvEITER_get(ohv);
5b9c0671 1701
1604cfb0
MS
1702 ENTER;
1703 SAVEFREESV(hv);
0db511c0 1704
f6bb1c88 1705 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
5b9c0671 1706
1604cfb0
MS
1707 hv_iterinit(ohv);
1708 while ((entry = hv_iternext_flags(ohv, 0))) {
1709 SV *const sv = newSVsv(hv_iterval(ohv,entry));
1710 SV *heksv = HeSVKEY(entry);
1711 if (!heksv && sv) heksv = newSVhek(HeKEY_hek(entry));
1712 if (sv) sv_magic(sv, NULL, PERL_MAGIC_hintselem,
1713 (char *)heksv, HEf_SVKEY);
1714 if (heksv == HeSVKEY(entry))
1715 (void)hv_store_ent(hv, heksv, sv, 0);
1716 else {
1717 (void)hv_common(hv, heksv, HeKEY(entry), HeKLEN(entry),
1718 HeKFLAGS(entry), HV_FETCH_ISSTORE|HV_FETCH_JUST_SV, sv, HeHASH(entry));
1719 SvREFCNT_dec_NN(heksv);
1720 }
1721 }
1722 HvRITER_set(ohv, riter);
1723 HvEITER_set(ohv, eiter);
1724
1725 SvREFCNT_inc_simple_void_NN(hv);
1726 LEAVE;
5b9c0671
NC
1727 }
1728 hv_magic(hv, NULL, PERL_MAGIC_hints);
1729 return hv;
1730}
f6bb1c88 1731#undef HV_SET_MAX_ADJUSTED_FOR_KEYS
5b9c0671 1732
e0171a1a
DM
1733/* like hv_free_ent, but returns the SV rather than freeing it */
1734STATIC SV*
5aaab254 1735S_hv_free_ent_ret(pTHX_ HV *hv, HE *entry)
79072805 1736{
16bdeea2
GS
1737 SV *val;
1738
e0171a1a 1739 PERL_ARGS_ASSERT_HV_FREE_ENT_RET;
7918f24d 1740
16bdeea2 1741 val = HeVAL(entry);
68dc0745 1742 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0
MS
1743 SvREFCNT_dec(HeKEY_sv(entry));
1744 Safefree(HeKEY_hek(entry));
44a8e56a 1745 }
1746 else if (HvSHAREKEYS(hv))
1604cfb0 1747 unshare_hek(HeKEY_hek(entry));
fde52b5c 1748 else
1604cfb0 1749 Safefree(HeKEY_hek(entry));
d33b2eba 1750 del_HE(entry);
e0171a1a
DM
1751 return val;
1752}
1753
1754
1755void
5aaab254 1756Perl_hv_free_ent(pTHX_ HV *hv, HE *entry)
e0171a1a 1757{
e0171a1a
DM
1758 SV *val;
1759
1760 PERL_ARGS_ASSERT_HV_FREE_ENT;
1761
1762 if (!entry)
1604cfb0 1763 return;
e0171a1a 1764 val = hv_free_ent_ret(hv, entry);
272e8453 1765 SvREFCNT_dec(val);
79072805
LW
1766}
1767
f1c32fec 1768
79072805 1769void
5aaab254 1770Perl_hv_delayfree_ent(pTHX_ HV *hv, HE *entry)
79072805 1771{
7918f24d
NC
1772 PERL_ARGS_ASSERT_HV_DELAYFREE_ENT;
1773
68dc0745 1774 if (!entry)
1604cfb0 1775 return;
bc4947fc
NC
1776 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1777 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
68dc0745 1778 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0 1779 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
44a8e56a 1780 }
bc4947fc 1781 hv_free_ent(hv, entry);
79072805
LW
1782}
1783
954c1994
GS
1784/*
1785=for apidoc hv_clear
1786
bb8005f7 1787Frees all the elements of a hash, leaving it empty.
8b9a1153
FC
1788The XS equivalent of C<%hash = ()>. See also L</hv_undef>.
1789
a4395eba
DM
1790See L</av_clear> for a note about the hash possibly being invalid on
1791return.
954c1994
GS
1792
1793=cut
1794*/
1795
79072805 1796void
864dbfa3 1797Perl_hv_clear(pTHX_ HV *hv)
79072805 1798{
be988557
DM
1799 SSize_t orig_ix;
1800
eb578fdb 1801 XPVHV* xhv;
79072805 1802 if (!hv)
1604cfb0 1803 return;
49293501 1804
ecae49c0
NC
1805 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1806
34c3c4e3
DM
1807 xhv = (XPVHV*)SvANY(hv);
1808
be988557
DM
1809 /* avoid hv being freed when calling destructors below */
1810 EXTEND_MORTAL(1);
1811 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
1812 orig_ix = PL_tmps_ix;
7b2c381c 1813 if (SvREADONLY(hv) && HvARRAY(hv) != NULL) {
1604cfb0
MS
1814 /* restricted hash: convert all keys to placeholders */
1815 STRLEN i;
1816 for (i = 0; i <= xhv->xhv_max; i++) {
1817 HE *entry = (HvARRAY(hv))[i];
1818 for (; entry; entry = HeNEXT(entry)) {
1819 /* not already placeholder */
1820 if (HeVAL(entry) != &PL_sv_placeholder) {
1821 if (HeVAL(entry)) {
1822 if (SvREADONLY(HeVAL(entry))) {
1823 SV* const keysv = hv_iterkeysv(entry);
1824 Perl_croak_nocontext(
1825 "Attempt to delete readonly key '%" SVf "' from a restricted hash",
1826 (void*)keysv);
1827 }
1828 SvREFCNT_dec_NN(HeVAL(entry));
1829 }
1830 HeVAL(entry) = &PL_sv_placeholder;
1831 HvPLACEHOLDERS(hv)++;
1832 }
1833 }
1834 }
49293501 1835 }
afbbf215 1836 else {
1604cfb0
MS
1837 hv_free_entries(hv);
1838 HvPLACEHOLDERS_set(hv, 0);
49293501 1839
1604cfb0
MS
1840 if (SvRMAGICAL(hv))
1841 mg_clear(MUTABLE_SV(hv));
574c8022 1842
1604cfb0 1843 HvHASKFLAGS_off(hv);
afbbf215 1844 }
b79f7545 1845 if (SvOOK(hv)) {
00169e2c 1846 if(HvENAME_get(hv))
dd69841b 1847 mro_isa_changed_in(hv);
1604cfb0 1848 HvEITER_set(hv, NULL);
bfcb3514 1849 }
be988557
DM
1850 /* disarm hv's premature free guard */
1851 if (LIKELY(PL_tmps_ix == orig_ix))
1852 PL_tmps_ix--;
1853 else
1854 PL_tmps_stack[orig_ix] = &PL_sv_undef;
1855 SvREFCNT_dec_NN(hv);
79072805
LW
1856}
1857
3540d4ce
AB
1858/*
1859=for apidoc hv_clear_placeholders
1860
1861Clears any placeholders from a hash. If a restricted hash has any of its keys
1862marked as readonly and the key is subsequently deleted, the key is not actually
796b6530 1863deleted but is marked by assigning it a value of C<&PL_sv_placeholder>. This tags
3540d4ce 1864it so it will be ignored by future operations such as iterating over the hash,
4cdaeff7 1865but will still allow the hash to have a value reassigned to the key at some
3540d4ce 1866future point. This function clears any such placeholder keys from the hash.
796b6530
KW
1867See C<L<Hash::Util::lock_keys()|Hash::Util/lock_keys>> for an example of its
1868use.
3540d4ce
AB
1869
1870=cut
1871*/
1872
1873void
1874Perl_hv_clear_placeholders(pTHX_ HV *hv)
1875{
b3ca2e83
NC
1876 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
1877
7918f24d
NC
1878 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
1879
b3ca2e83 1880 if (items)
1604cfb0 1881 clear_placeholders(hv, items);
b3ca2e83
NC
1882}
1883
1884static void
1885S_clear_placeholders(pTHX_ HV *hv, U32 items)
1886{
b464bac0 1887 I32 i;
d3677389 1888
7918f24d
NC
1889 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
1890
d3677389 1891 if (items == 0)
1604cfb0 1892 return;
d3677389 1893
b464bac0 1894 i = HvMAX(hv);
d3677389 1895 do {
1604cfb0
MS
1896 /* Loop down the linked list heads */
1897 HE **oentry = &(HvARRAY(hv))[i];
1898 HE *entry;
1899
1900 while ((entry = *oentry)) {
1901 if (HeVAL(entry) == &PL_sv_placeholder) {
1902 *oentry = HeNEXT(entry);
1903 if (entry == HvEITER_get(hv))
1904 HvLAZYDEL_on(hv);
1905 else {
1906 if (SvOOK(hv) && HvLAZYDEL(hv) &&
1907 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
1908 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
1909 hv_free_ent(hv, entry);
1910 }
1911
1912 if (--items == 0) {
1913 /* Finished. */
1914 I32 placeholders = HvPLACEHOLDERS_get(hv);
1915 HvTOTALKEYS(hv) -= (IV)placeholders;
1916 /* HvUSEDKEYS expanded */
1917 if ((HvTOTALKEYS(hv) - placeholders) == 0)
1918 HvHASKFLAGS_off(hv);
1919 HvPLACEHOLDERS_set(hv, 0);
1920 return;
1921 }
1922 } else {
1923 oentry = &HeNEXT(entry);
1924 }
1925 }
d3677389
NC
1926 } while (--i >= 0);
1927 /* You can't get here, hence assertion should always fail. */
1928 assert (items == 0);
661d43c4 1929 NOT_REACHED; /* NOTREACHED */
3540d4ce
AB
1930}
1931
76e3520e 1932STATIC void
b6bb1ecb 1933S_hv_free_entries(pTHX_ HV *hv)
79072805 1934{
e0171a1a 1935 STRLEN index = 0;
7d6175ef 1936 XPVHV * const xhv = (XPVHV*)SvANY(hv);
6d1c68e6 1937 SV *sv;
3abe233e 1938
367fba63 1939 PERL_ARGS_ASSERT_HV_FREE_ENTRIES;
7918f24d 1940
6d1c68e6 1941 while ((sv = Perl_hfree_next_entry(aTHX_ hv, &index))||xhv->xhv_keys) {
1604cfb0 1942 SvREFCNT_dec(sv);
e0171a1a
DM
1943 }
1944}
23976bdd 1945
b79f7545 1946
e0171a1a 1947/* hfree_next_entry()
b6bb1ecb 1948 * For use only by S_hv_free_entries() and sv_clear().
e0171a1a 1949 * Delete the next available HE from hv and return the associated SV.
7d6175ef
FC
1950 * Returns null on empty hash. Nevertheless null is not a reliable
1951 * indicator that the hash is empty, as the deleted entry may have a
1952 * null value.
e0171a1a
DM
1953 * indexp is a pointer to the current index into HvARRAY. The index should
1954 * initially be set to 0. hfree_next_entry() may update it. */
1955
1956SV*
1957Perl_hfree_next_entry(pTHX_ HV *hv, STRLEN *indexp)
1958{
1959 struct xpvhv_aux *iter;
1960 HE *entry;
1961 HE ** array;
1962#ifdef DEBUGGING
1963 STRLEN orig_index = *indexp;
1964#endif
1965
1966 PERL_ARGS_ASSERT_HFREE_NEXT_ENTRY;
1967
9faf471a 1968 if (SvOOK(hv) && ((iter = HvAUX(hv)))) {
1604cfb0 1969 if ((entry = iter->xhv_eiter)) {
9faf471a
NC
1970 /* the iterator may get resurrected after each
1971 * destructor call, so check each time */
1972 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1973 HvLAZYDEL_off(hv);
1974 hv_free_ent(hv, entry);
1975 /* warning: at this point HvARRAY may have been
1976 * re-allocated, HvMAX changed etc */
1977 }
339441ef 1978 iter = HvAUX(hv); /* may have been realloced */
9faf471a
NC
1979 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
1980 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
6a5b4183 1981#ifdef PERL_HASH_RANDOMIZE_KEYS
9faf471a 1982 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 1983#endif
9faf471a 1984 }
e0171a1a
DM
1985 }
1986
00a1a643 1987 if (!((XPVHV*)SvANY(hv))->xhv_keys)
1604cfb0 1988 return NULL;
00a1a643 1989
e0171a1a
DM
1990 array = HvARRAY(hv);
1991 assert(array);
1992 while ( ! ((entry = array[*indexp])) ) {
1604cfb0
MS
1993 if ((*indexp)++ >= HvMAX(hv))
1994 *indexp = 0;
1995 assert(*indexp != orig_index);
e0171a1a
DM
1996 }
1997 array[*indexp] = HeNEXT(entry);
1998 ((XPVHV*) SvANY(hv))->xhv_keys--;
1999
2000 if ( PL_phase != PERL_PHASE_DESTRUCT && HvENAME(hv)
1604cfb0
MS
2001 && HeVAL(entry) && isGV(HeVAL(entry))
2002 && GvHV(HeVAL(entry)) && HvENAME(GvHV(HeVAL(entry)))
e0171a1a 2003 ) {
1604cfb0
MS
2004 STRLEN klen;
2005 const char * const key = HePV(entry,klen);
2006 if ((klen > 1 && key[klen-1]==':' && key[klen-2]==':')
2007 || (klen == 1 && key[0] == ':')) {
2008 mro_package_moved(
2009 NULL, GvHV(HeVAL(entry)),
2010 (GV *)HeVAL(entry), 0
2011 );
2012 }
e0171a1a
DM
2013 }
2014 return hv_free_ent_ret(hv, entry);
79072805
LW
2015}
2016
e0171a1a 2017
954c1994
GS
2018/*
2019=for apidoc hv_undef
2020
8b9a1153 2021Undefines the hash. The XS equivalent of C<undef(%hash)>.
c2217cd3 2022
796b6530 2023As well as freeing all the elements of the hash (like C<hv_clear()>), this
c2217cd3 2024also frees any auxiliary data and storage associated with the hash.
8b9a1153 2025
a4395eba
DM
2026See L</av_clear> for a note about the hash possibly being invalid on
2027return.
954c1994
GS
2028
2029=cut
2030*/
2031
79072805 2032void
8581adba 2033Perl_hv_undef_flags(pTHX_ HV *hv, U32 flags)
79072805 2034{
eb578fdb 2035 XPVHV* xhv;
8a50cd03 2036 bool save;
6f15e0a5 2037 SSize_t orig_ix = PL_tmps_ix; /* silence compiler warning about unitialized vars */
86f55936 2038
79072805 2039 if (!hv)
1604cfb0 2040 return;
be988557 2041 save = cBOOL(SvREFCNT(hv));
ecae49c0 2042 DEBUG_A(Perl_hv_assert(aTHX_ hv));
cbec9347 2043 xhv = (XPVHV*)SvANY(hv);
dd69841b 2044
b6bb1ecb 2045 /* The name must be deleted before the call to hv_free_entries so that
745edda6
FC
2046 CVs are anonymised properly. But the effective name must be pre-
2047 served until after that call (and only deleted afterwards if the
2048 call originated from sv_clear). For stashes with one name that is
2049 both the canonical name and the effective name, hv_name_set has to
2050 allocate an array for storing the effective name. We can skip that
2051 during global destruction, as it does not matter where the CVs point
2052 if they will be freed anyway. */
b6bb1ecb 2053 /* note that the code following prior to hv_free_entries is duplicated
104d7b69 2054 * in sv_clear(), and changes here should be done there too */
0ca9877d 2055 if (PL_phase != PERL_PHASE_DESTRUCT && HvNAME(hv)) {
103f5a36
NC
2056 if (PL_stashcache) {
2057 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for '%"
147e3846 2058 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
1604cfb0 2059 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
103f5a36 2060 }
1604cfb0 2061 hv_name_set(hv, NULL, 0, 0);
85e6fe83 2062 }
8505eec0 2063 if (save) {
be988557
DM
2064 /* avoid hv being freed when calling destructors below */
2065 EXTEND_MORTAL(1);
2066 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
2067 orig_ix = PL_tmps_ix;
8505eec0 2068 }
b6bb1ecb 2069 hv_free_entries(hv);
47f1cf77 2070 if (SvOOK(hv)) {
47f1cf77 2071 struct mro_meta *meta;
0ca9877d 2072 const char *name;
745edda6 2073
0ca9877d 2074 if (HvENAME_get(hv)) {
1604cfb0
MS
2075 if (PL_phase != PERL_PHASE_DESTRUCT)
2076 mro_isa_changed_in(hv);
103f5a36
NC
2077 if (PL_stashcache) {
2078 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for effective name '%"
147e3846 2079 HEKf "'\n", HEKfARG(HvENAME_HEK(hv))));
1604cfb0 2080 (void)hv_deletehek(PL_stashcache, HvENAME_HEK(hv), G_DISCARD);
103f5a36 2081 }
745edda6
FC
2082 }
2083
2084 /* If this call originated from sv_clear, then we must check for
2085 * effective names that need freeing, as well as the usual name. */
2086 name = HvNAME(hv);
339441ef 2087 if (flags & HV_NAME_SETALL ? !!HvAUX(hv)->xhv_name_u.xhvnameu_name : !!name) {
103f5a36
NC
2088 if (name && PL_stashcache) {
2089 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for name '%"
147e3846 2090 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
1604cfb0 2091 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
103f5a36 2092 }
1604cfb0 2093 hv_name_set(hv, NULL, 0, flags);
47f1cf77 2094 }
339441ef 2095 if((meta = HvAUX(hv)->xhv_mro_meta)) {
1604cfb0
MS
2096 if (meta->mro_linear_all) {
2097 SvREFCNT_dec_NN(meta->mro_linear_all);
2098 /* mro_linear_current is just acting as a shortcut pointer,
2099 hence the else. */
2100 }
2101 else
2102 /* Only the current MRO is stored, so this owns the data.
2103 */
2104 SvREFCNT_dec(meta->mro_linear_current);
2105 SvREFCNT_dec(meta->mro_nextmethod);
2106 SvREFCNT_dec(meta->isa);
2107 SvREFCNT_dec(meta->super);
2108 Safefree(meta);
2109 HvAUX(hv)->xhv_mro_meta = NULL;
47f1cf77 2110 }
339441ef 2111 if (!HvAUX(hv)->xhv_name_u.xhvnameu_name && ! HvAUX(hv)->xhv_backreferences)
1604cfb0 2112 SvFLAGS(hv) &= ~SVf_OOK;
745edda6
FC
2113 }
2114 if (!SvOOK(hv)) {
1604cfb0 2115 Safefree(HvARRAY(hv));
f6bb1c88 2116 xhv->xhv_max = PERL_HASH_DEFAULT_HvMAX; /* HvMAX(hv) = 7 (it's a normal hash) */
1604cfb0 2117 HvARRAY(hv) = 0;
2d0d1ecc 2118 }
5bec93be
DM
2119 /* if we're freeing the HV, the SvMAGIC field has been reused for
2120 * other purposes, and so there can't be any placeholder magic */
2121 if (SvREFCNT(hv))
1604cfb0 2122 HvPLACEHOLDERS_set(hv, 0);
a0d0e21e
LW
2123
2124 if (SvRMAGICAL(hv))
1604cfb0 2125 mg_clear(MUTABLE_SV(hv));
be988557
DM
2126
2127 if (save) {
2128 /* disarm hv's premature free guard */
2129 if (LIKELY(PL_tmps_ix == orig_ix))
2130 PL_tmps_ix--;
2131 else
2132 PL_tmps_stack[orig_ix] = &PL_sv_undef;
2133 SvREFCNT_dec_NN(hv);
2134 }
79072805
LW
2135}
2136
4d0fbddd
NC
2137/*
2138=for apidoc hv_fill
2139
8bf4c401
YO
2140Returns the number of hash buckets that happen to be in use.
2141
2142This function is wrapped by the macro C<HvFILL>.
4d0fbddd 2143
8bf4c401
YO
2144As of perl 5.25 this function is used only for debugging
2145purposes, and the number of used hash buckets is not
2146in any way cached, thus this function can be costly
2147to execute as it must iterate over all the buckets in the
2148hash.
4d0fbddd
NC
2149
2150=cut
2151*/
2152
2153STRLEN
9faf471a 2154Perl_hv_fill(pTHX_ HV *const hv)
4d0fbddd
NC
2155{
2156 STRLEN count = 0;
2157 HE **ents = HvARRAY(hv);
2158
e9b8343f 2159 PERL_UNUSED_CONTEXT;
4d0fbddd
NC
2160 PERL_ARGS_ASSERT_HV_FILL;
2161
553215cc
NC
2162 /* No keys implies no buckets used.
2163 One key can only possibly mean one bucket used. */
2164 if (HvTOTALKEYS(hv) < 2)
2165 return HvTOTALKEYS(hv);
2166
4d0fbddd 2167 if (ents) {
8bf4c401
YO
2168 /* I wonder why we count down here...
2169 * Is it some micro-optimisation?
2170 * I would have thought counting up was better.
2171 * - Yves
2172 */
1604cfb0
MS
2173 HE *const *const last = ents + HvMAX(hv);
2174 count = last + 1 - ents;
4d0fbddd 2175
1604cfb0
MS
2176 do {
2177 if (!*ents)
2178 --count;
2179 } while (++ents <= last);
4d0fbddd
NC
2180 }
2181 return count;
2182}
2183
0e0ab621
YO
2184/* hash a pointer to a U32 - Used in the hash traversal randomization
2185 * and bucket order randomization code
2186 *
2187 * this code was derived from Sereal, which was derived from autobox.
2188 */
2189
2190PERL_STATIC_INLINE U32 S_ptr_hash(PTRV u) {
2191#if PTRSIZE == 8
2192 /*
2193 * This is one of Thomas Wang's hash functions for 64-bit integers from:
2194 * http://www.concentric.net/~Ttwang/tech/inthash.htm
2195 */
2196 u = (~u) + (u << 18);
2197 u = u ^ (u >> 31);
2198 u = u * 21;
2199 u = u ^ (u >> 11);
2200 u = u + (u << 6);
2201 u = u ^ (u >> 22);
2202#else
2203 /*
2204 * This is one of Bob Jenkins' hash functions for 32-bit integers
2205 * from: http://burtleburtle.net/bob/hash/integer.html
2206 */
2207 u = (u + 0x7ed55d16) + (u << 12);
2208 u = (u ^ 0xc761c23c) ^ (u >> 19);
2209 u = (u + 0x165667b1) + (u << 5);
2210 u = (u + 0xd3a2646c) ^ (u << 9);
2211 u = (u + 0xfd7046c5) + (u << 3);
2212 u = (u ^ 0xb55a4f09) ^ (u >> 16);
2213#endif
2214 return (U32)u;
2215}
2216
bea177f3
YO
2217static struct xpvhv_aux*
2218S_hv_auxinit_internal(struct xpvhv_aux *iter) {
2219 PERL_ARGS_ASSERT_HV_AUXINIT_INTERNAL;
2220 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2221 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2222#ifdef PERL_HASH_RANDOMIZE_KEYS
2223 iter->xhv_last_rand = iter->xhv_rand;
2224#endif
bea177f3
YO
2225 iter->xhv_name_u.xhvnameu_name = 0;
2226 iter->xhv_name_count = 0;
2227 iter->xhv_backreferences = 0;
2228 iter->xhv_mro_meta = NULL;
2229 iter->xhv_aux_flags = 0;
2230 return iter;
2231}
2232
0e0ab621 2233
b464bac0 2234static struct xpvhv_aux*
0e0ab621 2235S_hv_auxinit(pTHX_ HV *hv) {
bfcb3514 2236 struct xpvhv_aux *iter;
b79f7545 2237 char *array;
bfcb3514 2238
7918f24d
NC
2239 PERL_ARGS_ASSERT_HV_AUXINIT;
2240
0e0ab621
YO
2241 if (!SvOOK(hv)) {
2242 if (!HvARRAY(hv)) {
2243 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
2244 + sizeof(struct xpvhv_aux), char);
2245 } else {
2246 array = (char *) HvARRAY(hv);
2247 Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
2248 + sizeof(struct xpvhv_aux), char);
2249 }
2250 HvARRAY(hv) = (HE**)array;
2251 SvOOK_on(hv);
a7b39f85 2252 iter = HvAUX(hv);
6a5b4183
YO
2253#ifdef PERL_HASH_RANDOMIZE_KEYS
2254 if (PL_HASH_RAND_BITS_ENABLED) {
2255 /* mix in some new state to PL_hash_rand_bits to "randomize" the traversal order*/
2256 if (PL_HASH_RAND_BITS_ENABLED == 1)
2257 PL_hash_rand_bits += ptr_hash((PTRV)array);
2258 PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1);
2259 }
a7b39f85 2260 iter->xhv_rand = (U32)PL_hash_rand_bits;
6a5b4183 2261#endif
a7b39f85
YO
2262 } else {
2263 iter = HvAUX(hv);
b79f7545 2264 }
bfcb3514 2265
bea177f3 2266 return hv_auxinit_internal(iter);
bfcb3514
NC
2267}
2268
954c1994
GS
2269/*
2270=for apidoc hv_iterinit
2271
2272Prepares a starting point to traverse a hash table. Returns the number of
fe7d7ed3
MH
2273keys in the hash, including placeholders (i.e. the same as C<HvTOTALKEYS(hv)>).
2274The return value is currently only meaningful for hashes without tie magic.
954c1994
GS
2275
2276NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
2277hash buckets that happen to be in use. If you still need that esoteric
b24b84ef 2278value, you can get it through the macro C<HvFILL(hv)>.
954c1994 2279
e16e2ff8 2280
954c1994
GS
2281=cut
2282*/
2283
79072805 2284I32
864dbfa3 2285Perl_hv_iterinit(pTHX_ HV *hv)
79072805 2286{
7918f24d
NC
2287 PERL_ARGS_ASSERT_HV_ITERINIT;
2288
b79f7545 2289 if (SvOOK(hv)) {
1604cfb0
MS
2290 struct xpvhv_aux * iter = HvAUX(hv);
2291 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
2292 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2293 HvLAZYDEL_off(hv);
2294 hv_free_ent(hv, entry);
2295 }
2296 iter = HvAUX(hv); /* may have been reallocated */
2297 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2298 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
6a5b4183 2299#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85 2300 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 2301#endif
bfcb3514 2302 } else {
1604cfb0 2303 hv_auxinit(hv);
72940dca 2304 }
44a2ac75 2305
8bf4c401 2306 /* note this includes placeholders! */
5d88ecd7 2307 return HvTOTALKEYS(hv);
79072805 2308}
bfcb3514
NC
2309
2310I32 *
2311Perl_hv_riter_p(pTHX_ HV *hv) {
2312 struct xpvhv_aux *iter;
2313
7918f24d
NC
2314 PERL_ARGS_ASSERT_HV_RITER_P;
2315
6136c704 2316 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
2317 return &(iter->xhv_riter);
2318}
2319
2320HE **
2321Perl_hv_eiter_p(pTHX_ HV *hv) {
2322 struct xpvhv_aux *iter;
2323
7918f24d
NC
2324 PERL_ARGS_ASSERT_HV_EITER_P;
2325
6136c704 2326 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
2327 return &(iter->xhv_eiter);
2328}
2329
2330void
2331Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
2332 struct xpvhv_aux *iter;
2333
7918f24d
NC
2334 PERL_ARGS_ASSERT_HV_RITER_SET;
2335
b79f7545 2336 if (SvOOK(hv)) {
1604cfb0 2337 iter = HvAUX(hv);
b79f7545 2338 } else {
1604cfb0
MS
2339 if (riter == -1)
2340 return;
bfcb3514 2341
1604cfb0 2342 iter = hv_auxinit(hv);
bfcb3514
NC
2343 }
2344 iter->xhv_riter = riter;
2345}
2346
2347void
6a5b4183
YO
2348Perl_hv_rand_set(pTHX_ HV *hv, U32 new_xhv_rand) {
2349 struct xpvhv_aux *iter;
2350
2351 PERL_ARGS_ASSERT_HV_RAND_SET;
2352
2353#ifdef PERL_HASH_RANDOMIZE_KEYS
6a5b4183
YO
2354 if (SvOOK(hv)) {
2355 iter = HvAUX(hv);
2356 } else {
2357 iter = hv_auxinit(hv);
2358 }
2359 iter->xhv_rand = new_xhv_rand;
2360#else
2361 Perl_croak(aTHX_ "This Perl has not been built with support for randomized hash key traversal but something called Perl_hv_rand_set().");
2362#endif
2363}
2364
2365void
bfcb3514
NC
2366Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
2367 struct xpvhv_aux *iter;
2368
7918f24d
NC
2369 PERL_ARGS_ASSERT_HV_EITER_SET;
2370
b79f7545 2371 if (SvOOK(hv)) {
1604cfb0 2372 iter = HvAUX(hv);
b79f7545 2373 } else {
1604cfb0
MS
2374 /* 0 is the default so don't go malloc()ing a new structure just to
2375 hold 0. */
2376 if (!eiter)
2377 return;
bfcb3514 2378
1604cfb0 2379 iter = hv_auxinit(hv);
bfcb3514
NC
2380 }
2381 iter->xhv_eiter = eiter;
2382}
2383
bfcb3514 2384void
4164be69 2385Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
bfcb3514 2386{
b79f7545 2387 struct xpvhv_aux *iter;
7423f6db 2388 U32 hash;
78b79c77 2389 HEK **spot;
46c461b5 2390
7918f24d 2391 PERL_ARGS_ASSERT_HV_NAME_SET;
bfcb3514 2392
4164be69 2393 if (len > I32_MAX)
1604cfb0 2394 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
4164be69 2395
b79f7545 2396 if (SvOOK(hv)) {
1604cfb0
MS
2397 iter = HvAUX(hv);
2398 if (iter->xhv_name_u.xhvnameu_name) {
2399 if(iter->xhv_name_count) {
2400 if(flags & HV_NAME_SETALL) {
2401 HEK ** const this_name = HvAUX(hv)->xhv_name_u.xhvnameu_names;
2402 HEK **hekp = this_name + (
2403 iter->xhv_name_count < 0
2404 ? -iter->xhv_name_count
2405 : iter->xhv_name_count
2406 );
2407 while(hekp-- > this_name+1)
2408 unshare_hek_or_pvn(*hekp, 0, 0, 0);
2409 /* The first elem may be null. */
2410 if(*this_name) unshare_hek_or_pvn(*this_name, 0, 0, 0);
2411 Safefree(this_name);
339441ef 2412 iter = HvAUX(hv); /* may been realloced */
1604cfb0
MS
2413 spot = &iter->xhv_name_u.xhvnameu_name;
2414 iter->xhv_name_count = 0;
2415 }
2416 else {
2417 if(iter->xhv_name_count > 0) {
2418 /* shift some things over */
2419 Renew(
2420 iter->xhv_name_u.xhvnameu_names, iter->xhv_name_count + 1, HEK *
2421 );
2422 spot = iter->xhv_name_u.xhvnameu_names;
2423 spot[iter->xhv_name_count] = spot[1];
2424 spot[1] = spot[0];
2425 iter->xhv_name_count = -(iter->xhv_name_count + 1);
2426 }
2427 else if(*(spot = iter->xhv_name_u.xhvnameu_names)) {
2428 unshare_hek_or_pvn(*spot, 0, 0, 0);
2429 }
2430 }
2431 }
2432 else if (flags & HV_NAME_SETALL) {
2433 unshare_hek_or_pvn(iter->xhv_name_u.xhvnameu_name, 0, 0, 0);
339441ef 2434 iter = HvAUX(hv); /* may been realloced */
1604cfb0
MS
2435 spot = &iter->xhv_name_u.xhvnameu_name;
2436 }
2437 else {
2438 HEK * const existing_name = iter->xhv_name_u.xhvnameu_name;
2439 Newx(iter->xhv_name_u.xhvnameu_names, 2, HEK *);
2440 iter->xhv_name_count = -2;
2441 spot = iter->xhv_name_u.xhvnameu_names;
2442 spot[1] = existing_name;
2443 }
2444 }
2445 else { spot = &iter->xhv_name_u.xhvnameu_name; iter->xhv_name_count = 0; }
16580ff5 2446 } else {
1604cfb0
MS
2447 if (name == 0)
2448 return;
bfcb3514 2449
1604cfb0
MS
2450 iter = hv_auxinit(hv);
2451 spot = &iter->xhv_name_u.xhvnameu_name;
bfcb3514 2452 }
7423f6db 2453 PERL_HASH(hash, name, len);
c60dbbc3 2454 *spot = name ? share_hek(name, flags & SVf_UTF8 ? -(I32)len : (I32)len, hash) : NULL;
4643eb69
BF
2455}
2456
2457/*
2458This is basically sv_eq_flags() in sv.c, but we avoid the magic
2459and bytes checking.
2460*/
2461
2462STATIC I32
2463hek_eq_pvn_flags(pTHX_ const HEK *hek, const char* pv, const I32 pvlen, const U32 flags) {
2464 if ( (HEK_UTF8(hek) ? 1 : 0) != (flags & SVf_UTF8 ? 1 : 0) ) {
2465 if (flags & SVf_UTF8)
2466 return (bytes_cmp_utf8(
2467 (const U8*)HEK_KEY(hek), HEK_LEN(hek),
1604cfb0 2468 (const U8*)pv, pvlen) == 0);
4643eb69
BF
2469 else
2470 return (bytes_cmp_utf8(
2471 (const U8*)pv, pvlen,
1604cfb0 2472 (const U8*)HEK_KEY(hek), HEK_LEN(hek)) == 0);
4643eb69
BF
2473 }
2474 else
d35fec6c 2475 return HEK_LEN(hek) == pvlen && ((HEK_KEY(hek) == pv)
4643eb69 2476 || memEQ(HEK_KEY(hek), pv, pvlen));
bfcb3514
NC
2477}
2478
99206677
FC
2479/*
2480=for apidoc hv_ename_add
2481
db4fbf16 2482Adds a name to a stash's internal list of effective names. See
fbe13c60 2483C<L</hv_ename_delete>>.
99206677
FC
2484
2485This is called when a stash is assigned to a new location in the symbol
2486table.
2487
2488=cut
2489*/
2490
ee72b38d 2491void
27a1175b 2492Perl_hv_ename_add(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
ee72b38d 2493{
ee72b38d
FC
2494 struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2495 U32 hash;
2496
78b79c77 2497 PERL_ARGS_ASSERT_HV_ENAME_ADD;
ee72b38d
FC
2498
2499 if (len > I32_MAX)
1604cfb0 2500 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
ee72b38d
FC
2501
2502 PERL_HASH(hash, name, len);
2503
ee72b38d 2504 if (aux->xhv_name_count) {
1604cfb0
MS
2505 I32 count = aux->xhv_name_count;
2506 HEK ** const xhv_name = aux->xhv_name_u.xhvnameu_names + (count<0);
2507 HEK **hekp = xhv_name + (count < 0 ? -count - 1 : count);
2508 while (hekp-- > xhv_name)
2509 {
2510 assert(*hekp);
2511 if (
4643eb69
BF
2512 (HEK_UTF8(*hekp) || (flags & SVf_UTF8))
2513 ? hek_eq_pvn_flags(aTHX_ *hekp, name, (I32)len, flags)
1604cfb0 2514 : (HEK_LEN(*hekp) == (I32)len && memEQ(HEK_KEY(*hekp), name, len))
4643eb69 2515 ) {
1604cfb0
MS
2516 if (hekp == xhv_name && count < 0)
2517 aux->xhv_name_count = -count;
2518 return;
2519 }
2520 }
2521 if (count < 0) aux->xhv_name_count--, count = -count;
2522 else aux->xhv_name_count++;
2523 Renew(aux->xhv_name_u.xhvnameu_names, count + 1, HEK *);
2524 (aux->xhv_name_u.xhvnameu_names)[count] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
ee72b38d
FC
2525 }
2526 else {
1604cfb0
MS
2527 HEK *existing_name = aux->xhv_name_u.xhvnameu_name;
2528 if (
2529 existing_name && (
4643eb69
BF
2530 (HEK_UTF8(existing_name) || (flags & SVf_UTF8))
2531 ? hek_eq_pvn_flags(aTHX_ existing_name, name, (I32)len, flags)
1604cfb0
MS
2532 : (HEK_LEN(existing_name) == (I32)len && memEQ(HEK_KEY(existing_name), name, len))
2533 )
2534 ) return;
2535 Newx(aux->xhv_name_u.xhvnameu_names, 2, HEK *);
2536 aux->xhv_name_count = existing_name ? 2 : -2;
2537 *aux->xhv_name_u.xhvnameu_names = existing_name;
2538 (aux->xhv_name_u.xhvnameu_names)[1] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
ee72b38d
FC
2539 }
2540}
2541
99206677
FC
2542/*
2543=for apidoc hv_ename_delete
2544
db4fbf16 2545Removes a name from a stash's internal list of effective names. If this is
99206677
FC
2546the name returned by C<HvENAME>, then another name in the list will take
2547its place (C<HvENAME> will use it).
2548
2549This is called when a stash is deleted from the symbol table.
2550
2551=cut
2552*/
2553
ee72b38d 2554void
27a1175b 2555Perl_hv_ename_delete(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
ee72b38d 2556{
ee72b38d
FC
2557 struct xpvhv_aux *aux;
2558
78b79c77 2559 PERL_ARGS_ASSERT_HV_ENAME_DELETE;
ee72b38d
FC
2560
2561 if (len > I32_MAX)
1604cfb0 2562 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
ee72b38d
FC
2563
2564 if (!SvOOK(hv)) return;
2565
2566 aux = HvAUX(hv);
15d9236d 2567 if (!aux->xhv_name_u.xhvnameu_name) return;
ee72b38d
FC
2568
2569 if (aux->xhv_name_count) {
1604cfb0
MS
2570 HEK ** const namep = aux->xhv_name_u.xhvnameu_names;
2571 I32 const count = aux->xhv_name_count;
2572 HEK **victim = namep + (count < 0 ? -count : count);
2573 while (victim-- > namep + 1)
2574 if (
4643eb69
BF
2575 (HEK_UTF8(*victim) || (flags & SVf_UTF8))
2576 ? hek_eq_pvn_flags(aTHX_ *victim, name, (I32)len, flags)
1604cfb0
MS
2577 : (HEK_LEN(*victim) == (I32)len && memEQ(HEK_KEY(*victim), name, len))
2578 ) {
2579 unshare_hek_or_pvn(*victim, 0, 0, 0);
339441ef 2580 aux = HvAUX(hv); /* may been realloced */
1604cfb0
MS
2581 if (count < 0) ++aux->xhv_name_count;
2582 else --aux->xhv_name_count;
2583 if (
2584 (aux->xhv_name_count == 1 || aux->xhv_name_count == -1)
2585 && !*namep
2586 ) { /* if there are none left */
2587 Safefree(namep);
2588 aux->xhv_name_u.xhvnameu_names = NULL;
2589 aux->xhv_name_count = 0;
2590 }
2591 else {
2592 /* Move the last one back to fill the empty slot. It
2593 does not matter what order they are in. */
2594 *victim = *(namep + (count < 0 ? -count : count) - 1);
2595 }
2596 return;
2597 }
2598 if (
2599 count > 0 && ((HEK_UTF8(*namep) || (flags & SVf_UTF8))
4643eb69 2600 ? hek_eq_pvn_flags(aTHX_ *namep, name, (I32)len, flags)
1604cfb0 2601 : (HEK_LEN(*namep) == (I32)len && memEQ(HEK_KEY(*namep), name, len))
60a26c79 2602 )
1604cfb0
MS
2603 ) {
2604 aux->xhv_name_count = -count;
2605 }
ee72b38d
FC
2606 }
2607 else if(
4643eb69
BF
2608 (HEK_UTF8(aux->xhv_name_u.xhvnameu_name) || (flags & SVf_UTF8))
2609 ? hek_eq_pvn_flags(aTHX_ aux->xhv_name_u.xhvnameu_name, name, (I32)len, flags)
1604cfb0 2610 : (HEK_LEN(aux->xhv_name_u.xhvnameu_name) == (I32)len &&
4643eb69 2611 memEQ(HEK_KEY(aux->xhv_name_u.xhvnameu_name), name, len))
ee72b38d 2612 ) {
1604cfb0
MS
2613 HEK * const namehek = aux->xhv_name_u.xhvnameu_name;
2614 Newx(aux->xhv_name_u.xhvnameu_names, 1, HEK *);
2615 *aux->xhv_name_u.xhvnameu_names = namehek;
2616 aux->xhv_name_count = -1;
ee72b38d
FC
2617 }
2618}
2619
86f55936
NC
2620AV **
2621Perl_hv_backreferences_p(pTHX_ HV *hv) {
7918f24d 2622 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
8fbcb657 2623 /* See also Perl_sv_get_backrefs in sv.c where this logic is unrolled */
34f2dd85
YO
2624 {
2625 struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
2626 return &(iter->xhv_backreferences);
2627 }
86f55936
NC
2628}
2629
09aad8f0
DM
2630void
2631Perl_hv_kill_backrefs(pTHX_ HV *hv) {
2632 AV *av;
2633
2634 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2635
2636 if (!SvOOK(hv))
1604cfb0 2637 return;
09aad8f0
DM
2638
2639 av = HvAUX(hv)->xhv_backreferences;
2640
2641 if (av) {
1604cfb0
MS
2642 HvAUX(hv)->xhv_backreferences = 0;
2643 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
2644 if (SvTYPE(av) == SVt_PVAV)
2645 SvREFCNT_dec_NN(av);
09aad8f0
DM
2646 }
2647}
2648
954c1994 2649/*
7a7b9979
NC
2650hv_iternext is implemented as a macro in hv.h
2651
954c1994
GS
2652=for apidoc hv_iternext
2653
fbe13c60 2654Returns entries from a hash iterator. See C<L</hv_iterinit>>.
954c1994 2655
fe7bca90
NC
2656You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2657iterator currently points to, without losing your place or invalidating your
2658iterator. Note that in this case the current entry is deleted from the hash
2659with your iterator holding the last reference to it. Your iterator is flagged
2660to free the entry on the next call to C<hv_iternext>, so you must not discard
2661your iterator immediately else the entry will leak - call C<hv_iternext> to
2662trigger the resource deallocation.
2663
fe7bca90
NC
2664=for apidoc hv_iternext_flags
2665
fbe13c60
KW
2666Returns entries from a hash iterator. See C<L</hv_iterinit>> and
2667C<L</hv_iternext>>.
796b6530 2668The C<flags> value will normally be zero; if C<HV_ITERNEXT_WANTPLACEHOLDERS> is
fe7bca90 2669set the placeholders keys (for restricted hashes) will be returned in addition
72d33970 2670to normal keys. By default placeholders are automatically skipped over.
7996736c 2671Currently a placeholder is implemented with a value that is
990c89d7 2672C<&PL_sv_placeholder>. Note that the implementation of placeholders and
fe7bca90
NC
2673restricted hashes may change, and the implementation currently is
2674insufficiently abstracted for any change to be tidy.
e16e2ff8 2675
5af38e47
KW
2676=for apidoc Amnh||HV_ITERNEXT_WANTPLACEHOLDERS
2677
fe7bca90 2678=cut
e16e2ff8
NC
2679*/
2680
2681HE *
2682Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2683{
eb578fdb
KW
2684 XPVHV* xhv;
2685 HE *entry;
a0d0e21e 2686 HE *oldentry;
463ee0b2 2687 MAGIC* mg;
bfcb3514 2688 struct xpvhv_aux *iter;
79072805 2689
7918f24d
NC
2690 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2691
cbec9347 2692 xhv = (XPVHV*)SvANY(hv);
bfcb3514 2693
b79f7545 2694 if (!SvOOK(hv)) {
1604cfb0
MS
2695 /* Too many things (well, pp_each at least) merrily assume that you can
2696 call hv_iternext without calling hv_iterinit, so we'll have to deal
2697 with it. */
2698 hv_iterinit(hv);
bfcb3514 2699 }
b79f7545 2700 iter = HvAUX(hv);
bfcb3514
NC
2701
2702 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
e62cc96a 2703 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
1604cfb0 2704 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
e62cc96a
YO
2705 SV * const key = sv_newmortal();
2706 if (entry) {
2707 sv_setsv(key, HeSVKEY_force(entry));
2708 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
1604cfb0 2709 HeSVKEY_set(entry, NULL);
e62cc96a
YO
2710 }
2711 else {
2712 char *k;
2713 HEK *hek;
2714
2715 /* one HE per MAGICAL hash */
2716 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
1604cfb0 2717 HvLAZYDEL_on(hv); /* make sure entry gets freed */
e62cc96a 2718 Zero(entry, 1, HE);
ad64d0ec 2719 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
e62cc96a
YO
2720 hek = (HEK*)k;
2721 HeKEY_hek(entry) = hek;
2722 HeKLEN(entry) = HEf_SVKEY;
2723 }
ad64d0ec 2724 magic_nextpack(MUTABLE_SV(hv),mg,key);
e62cc96a
YO
2725 if (SvOK(key)) {
2726 /* force key to stay around until next time */
2727 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2728 return entry; /* beware, hent_val is not set */
2729 }
ef8d46e8 2730 SvREFCNT_dec(HeVAL(entry));
e62cc96a
YO
2731 Safefree(HeKEY_hek(entry));
2732 del_HE(entry);
339441ef 2733 iter = HvAUX(hv); /* may been realloced */
e62cc96a 2734 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1604cfb0 2735 HvLAZYDEL_off(hv);
e62cc96a 2736 return NULL;
81714fb9 2737 }
79072805 2738 }
7ee146b1 2739#if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */
ad64d0ec 2740 if (!entry && SvRMAGICAL((const SV *)hv)
1604cfb0
MS
2741 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
2742 prime_env_iter();
03026e68 2743#ifdef VMS
1604cfb0
MS
2744 /* The prime_env_iter() on VMS just loaded up new hash values
2745 * so the iteration count needs to be reset back to the beginning
2746 */
2747 hv_iterinit(hv);
2748 iter = HvAUX(hv);
2749 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
03026e68
JM
2750#endif
2751 }
f675dbe5 2752#endif
463ee0b2 2753
bfaf5b52 2754 /* hv_iterinit now ensures this. */
b79f7545
NC
2755 assert (HvARRAY(hv));
2756
015a5f36 2757 /* At start of hash, entry is NULL. */
fde52b5c 2758 if (entry)
8aacddc1 2759 {
1604cfb0 2760 entry = HeNEXT(entry);
e16e2ff8
NC
2761 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2762 /*
2763 * Skip past any placeholders -- don't want to include them in
2764 * any iteration.
2765 */
7996736c 2766 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
e16e2ff8
NC
2767 entry = HeNEXT(entry);
2768 }
1604cfb0 2769 }
8aacddc1 2770 }
6a5b4183
YO
2771
2772#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85
YO
2773 if (iter->xhv_last_rand != iter->xhv_rand) {
2774 if (iter->xhv_riter != -1) {
2775 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
2776 "Use of each() on hash after insertion without resetting hash iterator results in undefined behavior"
2777 pTHX__FORMAT
2778 pTHX__VALUE);
2779 }
339441ef 2780 iter = HvAUX(hv); /* may been realloced */
a7b39f85
YO
2781 iter->xhv_last_rand = iter->xhv_rand;
2782 }
6a5b4183 2783#endif
015a5f36 2784
9eb4ebd1
NC
2785 /* Skip the entire loop if the hash is empty. */
2786 if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS)
1604cfb0
MS
2787 ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) {
2788 while (!entry) {
2789 /* OK. Come to the end of the current list. Grab the next one. */
2790
2791 iter->xhv_riter++; /* HvRITER(hv)++ */
2792 if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) {
2793 /* There is no next one. End of the hash. */
2794 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
6a5b4183
YO
2795#ifdef PERL_HASH_RANDOMIZE_KEYS
2796 iter->xhv_last_rand = iter->xhv_rand; /* reset xhv_last_rand so we can detect inserts during traversal */
2797#endif
1604cfb0
MS
2798 break;
2799 }
6a5b4183 2800 entry = (HvARRAY(hv))[ PERL_HASH_ITER_BUCKET(iter) & xhv->xhv_max ];
8aacddc1 2801
1604cfb0
MS
2802 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2803 /* If we have an entry, but it's a placeholder, don't count it.
2804 Try the next. */
2805 while (entry && HeVAL(entry) == &PL_sv_placeholder)
2806 entry = HeNEXT(entry);
2807 }
2808 /* Will loop again if this linked list starts NULL
2809 (for HV_ITERNEXT_WANTPLACEHOLDERS)
2810 or if we run through it and find only placeholders. */
2811 }
fde52b5c 2812 }
a7b39f85
YO
2813 else {
2814 iter->xhv_riter = -1;
6a5b4183 2815#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85 2816 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 2817#endif
a7b39f85 2818 }
79072805 2819
72940dca 2820 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1604cfb0
MS
2821 HvLAZYDEL_off(hv);
2822 hv_free_ent(hv, oldentry);
72940dca 2823 }
a0d0e21e 2824
339441ef 2825 iter = HvAUX(hv); /* may been realloced */
bfcb3514 2826 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
79072805
LW
2827 return entry;
2828}
2829
954c1994
GS
2830/*
2831=for apidoc hv_iterkey
2832
2833Returns the key from the current position of the hash iterator. See
fbe13c60 2834C<L</hv_iterinit>>.
954c1994
GS
2835
2836=cut
2837*/
2838
79072805 2839char *
5aaab254 2840Perl_hv_iterkey(pTHX_ HE *entry, I32 *retlen)
79072805 2841{
7918f24d
NC
2842 PERL_ARGS_ASSERT_HV_ITERKEY;
2843
fde52b5c 2844 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0
MS
2845 STRLEN len;
2846 char * const p = SvPV(HeKEY_sv(entry), len);
2847 *retlen = len;
2848 return p;
fde52b5c 2849 }
2850 else {
1604cfb0
MS
2851 *retlen = HeKLEN(entry);
2852 return HeKEY(entry);
fde52b5c 2853 }
2854}
2855
2856/* unlike hv_iterval(), this always returns a mortal copy of the key */
954c1994
GS
2857/*
2858=for apidoc hv_iterkeysv
2859
2860Returns the key as an C<SV*> from the current position of the hash
2861iterator. The return value will always be a mortal copy of the key. Also
fbe13c60 2862see C<L</hv_iterinit>>.
954c1994
GS
2863
2864=cut
2865*/
2866
fde52b5c 2867SV *
5aaab254 2868Perl_hv_iterkeysv(pTHX_ HE *entry)
fde52b5c 2869{
7918f24d
NC
2870 PERL_ARGS_ASSERT_HV_ITERKEYSV;
2871
c1b02ed8 2872 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
79072805
LW
2873}
2874
954c1994
GS
2875/*
2876=for apidoc hv_iterval
2877
2878Returns the value from the current position of the hash iterator. See
fbe13c60 2879C<L</hv_iterkey>>.
954c1994
GS
2880
2881=cut
2882*/
2883
79072805 2884SV *
5aaab254 2885Perl_hv_iterval(pTHX_ HV *hv, HE *entry)
79072805 2886{
7918f24d
NC
2887 PERL_ARGS_ASSERT_HV_ITERVAL;
2888
8990e307 2889 if (SvRMAGICAL(hv)) {
1604cfb0
MS
2890 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
2891 SV* const sv = sv_newmortal();
2892 if (HeKLEN(entry) == HEf_SVKEY)
2893 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
2894 else
2895 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
2896 return sv;
2897 }
79072805 2898 }
fde52b5c 2899 return HeVAL(entry);
79072805
LW
2900}
2901
954c1994
GS
2902/*
2903=for apidoc hv_iternextsv
2904
2905Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
2906operation.
2907
2908=cut
2909*/
2910
a0d0e21e 2911SV *
864dbfa3 2912Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
a0d0e21e 2913{
0bd48802
AL
2914 HE * const he = hv_iternext_flags(hv, 0);
2915
7918f24d
NC
2916 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
2917
0bd48802 2918 if (!he)
1604cfb0 2919 return NULL;
a0d0e21e
LW
2920 *key = hv_iterkey(he, retlen);
2921 return hv_iterval(hv, he);
2922}
2923
954c1994 2924/*
bc5cdc23
NC
2925
2926Now a macro in hv.h
2927
954c1994
GS
2928=for apidoc hv_magic
2929
fbe13c60 2930Adds magic to a hash. See C<L</sv_magic>>.
954c1994
GS
2931
2932=cut
2933*/
2934
bbce6d69 2935/* possibly free a shared string if no one has access to it
fde52b5c 2936 * len and hash must both be valid for str.
2937 */
bbce6d69 2938void
864dbfa3 2939Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
fde52b5c 2940{
19692e8d
NC
2941 unshare_hek_or_pvn (NULL, str, len, hash);
2942}
2943
2944
2945void
2946Perl_unshare_hek(pTHX_ HEK *hek)
2947{
bf11fd37 2948 assert(hek);
19692e8d
NC
2949 unshare_hek_or_pvn(hek, NULL, 0, 0);
2950}
2951
2952/* possibly free a shared string if no one has access to it
2953 hek if non-NULL takes priority over the other 3, else str, len and hash
2954 are used. If so, len and hash must both be valid for str.
2955 */
df132699 2956STATIC void
97ddebaf 2957S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
19692e8d 2958{
eb578fdb 2959 XPVHV* xhv;
20454177 2960 HE *entry;
eb578fdb 2961 HE **oentry;
c3654f1a 2962 bool is_utf8 = FALSE;
19692e8d 2963 int k_flags = 0;
aec46f14 2964 const char * const save = str;
cbbf8932 2965 struct shared_he *he = NULL;
c3654f1a 2966
19692e8d 2967 if (hek) {
1604cfb0
MS
2968 /* Find the shared he which is just before us in memory. */
2969 he = (struct shared_he *)(((char *)hek)
2970 - STRUCT_OFFSET(struct shared_he,
2971 shared_he_hek));
cbae3960 2972
1604cfb0
MS
2973 /* Assert that the caller passed us a genuine (or at least consistent)
2974 shared hek */
2975 assert (he->shared_he_he.hent_hek == hek);
29404ae0 2976
1604cfb0
MS
2977 if (he->shared_he_he.he_valu.hent_refcount - 1) {
2978 --he->shared_he_he.he_valu.hent_refcount;
2979 return;
2980 }
29404ae0 2981
19692e8d
NC
2982 hash = HEK_HASH(hek);
2983 } else if (len < 0) {
2984 STRLEN tmplen = -len;
2985 is_utf8 = TRUE;
2986 /* See the note in hv_fetch(). --jhi */
2987 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2988 len = tmplen;
2989 if (is_utf8)
2990 k_flags = HVhek_UTF8;
2991 if (str != save)
2992 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
c3654f1a 2993 }
1c846c1f 2994
de616631 2995 /* what follows was the moral equivalent of:
6b88bc9c 2996 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
1604cfb0
MS
2997 if (--*Svp == NULL)
2998 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
bbce6d69 2999 } */
cbec9347 3000 xhv = (XPVHV*)SvANY(PL_strtab);
fde52b5c 3001 /* assert(xhv_array != 0) */
9de10d5c 3002 oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
6c1b96a1 3003 if (he) {
1604cfb0 3004 const HE *const he_he = &(he->shared_he_he);
45d1cc86 3005 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
35ab5632
NC
3006 if (entry == he_he)
3007 break;
19692e8d
NC
3008 }
3009 } else {
35a4481c 3010 const int flags_masked = k_flags & HVhek_MASK;
45d1cc86 3011 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
19692e8d
NC
3012 if (HeHASH(entry) != hash) /* strings can't be equal */
3013 continue;
3014 if (HeKLEN(entry) != len)
3015 continue;
3016 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3017 continue;
3018 if (HeKFLAGS(entry) != flags_masked)
3019 continue;
19692e8d
NC
3020 break;
3021 }
3022 }
3023
35ab5632
NC
3024 if (entry) {
3025 if (--entry->he_valu.hent_refcount == 0) {
19692e8d 3026 *oentry = HeNEXT(entry);
cbae3960 3027 Safefree(entry);
4c7185a0 3028 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
19692e8d 3029 }
fde52b5c 3030 }
19692e8d 3031
9b387841 3032 if (!entry)
1604cfb0
MS
3033 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
3034 "Attempt to free nonexistent shared string '%s'%s"
3035 pTHX__FORMAT,
3036 hek ? HEK_KEY(hek) : str,
3037 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
19692e8d 3038 if (k_flags & HVhek_FREEKEY)
1604cfb0 3039 Safefree(str);
fde52b5c 3040}
3041
bbce6d69 3042/* get a (constant) string ptr from the global string table
3043 * string will get added if it is not already there.
fde52b5c 3044 * len and hash must both be valid for str.
3045 */
bbce6d69 3046HEK *
b02f3645 3047Perl_share_hek(pTHX_ const char *str, SSize_t len, U32 hash)
fde52b5c 3048{
da58a35d 3049 bool is_utf8 = FALSE;
19692e8d 3050 int flags = 0;
aec46f14 3051 const char * const save = str;
da58a35d 3052
7918f24d
NC
3053 PERL_ARGS_ASSERT_SHARE_HEK;
3054
da58a35d 3055 if (len < 0) {
77caf834 3056 STRLEN tmplen = -len;
da58a35d 3057 is_utf8 = TRUE;
77caf834
JH
3058 /* See the note in hv_fetch(). --jhi */
3059 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
3060 len = tmplen;
19692e8d
NC
3061 /* If we were able to downgrade here, then than means that we were passed
3062 in a key which only had chars 0-255, but was utf8 encoded. */
3063 if (is_utf8)
3064 flags = HVhek_UTF8;
3065 /* If we found we were able to downgrade the string to bytes, then
3066 we should flag that it needs upgrading on keys or each. Also flag
3067 that we need share_hek_flags to free the string. */
4643eb69
BF
3068 if (str != save) {
3069 PERL_HASH(hash, str, len);
19692e8d 3070 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
4643eb69 3071 }
19692e8d
NC
3072 }
3073
6e838c70 3074 return share_hek_flags (str, len, hash, flags);
19692e8d
NC
3075}
3076
6e838c70 3077STATIC HEK *
b02f3645 3078S_share_hek_flags(pTHX_ const char *str, STRLEN len, U32 hash, int flags)
19692e8d 3079{
eb578fdb 3080 HE *entry;
35a4481c 3081 const int flags_masked = flags & HVhek_MASK;
263cb4a6 3082 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
eb578fdb 3083 XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
7918f24d
NC
3084
3085 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
bbce6d69 3086
b02f3645
AC
3087 if (UNLIKELY(len > (STRLEN) I32_MAX)) {
3088 Perl_croak_nocontext("Sorry, hash keys must be smaller than 2**31 bytes");
3089 }
3090
fde52b5c 3091 /* what follows is the moral equivalent of:
1c846c1f 3092
6b88bc9c 3093 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
1604cfb0 3094 hv_store(PL_strtab, str, len, NULL, hash);
fdcd69b6 3095
1604cfb0
MS
3096 Can't rehash the shared string table, so not sure if it's worth
3097 counting the number of entries in the linked list
bbce6d69 3098 */
7918f24d 3099
fde52b5c 3100 /* assert(xhv_array != 0) */
263cb4a6
NC
3101 entry = (HvARRAY(PL_strtab))[hindex];
3102 for (;entry; entry = HeNEXT(entry)) {
1604cfb0
MS
3103 if (HeHASH(entry) != hash) /* strings can't be equal */
3104 continue;
3105 if (HeKLEN(entry) != (SSize_t) len)
3106 continue;
3107 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3108 continue;
3109 if (HeKFLAGS(entry) != flags_masked)
3110 continue;
3111 break;
fde52b5c 3112 }
263cb4a6
NC
3113
3114 if (!entry) {
1604cfb0
MS
3115 /* What used to be head of the list.
3116 If this is NULL, then we're the first entry for this slot, which
3117 means we need to increate fill. */
3118 struct shared_he *new_entry;
3119 HEK *hek;
3120 char *k;
3121 HE **const head = &HvARRAY(PL_strtab)[hindex];
3122 HE *const next = *head;
3123
3124 /* We don't actually store a HE from the arena and a regular HEK.
3125 Instead we allocate one chunk of memory big enough for both,
3126 and put the HEK straight after the HE. This way we can find the
3127 HE directly from the HEK.
3128 */
3129
3130 Newx(k, STRUCT_OFFSET(struct shared_he,
3131 shared_he_hek.hek_key[0]) + len + 2, char);
3132 new_entry = (struct shared_he *)k;
3133 entry = &(new_entry->shared_he_he);
3134 hek = &(new_entry->shared_he_hek);
3135
3136 Copy(str, HEK_KEY(hek), len, char);
3137 HEK_KEY(hek)[len] = 0;
3138 HEK_LEN(hek) = len;
3139 HEK_HASH(hek) = hash;
3140 HEK_FLAGS(hek) = (unsigned char)flags_masked;
3141
3142 /* Still "point" to the HEK, so that other code need not know what
3143 we're up to. */
3144 HeKEY_hek(entry) = hek;
3145 entry->he_valu.hent_refcount = 0;
3146 HeNEXT(entry) = next;
3147 *head = entry;
3148
3149 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
3150 if (!next) { /* initial entry? */
3151 } else if ( DO_HSPLIT(xhv) ) {
adf6906b
NC
3152 const STRLEN oldsize = xhv->xhv_max + 1;
3153 hsplit(PL_strtab, oldsize, oldsize * 2);
1604cfb0 3154 }
bbce6d69 3155 }
3156
de616631 3157 ++entry->he_valu.hent_refcount;
19692e8d
NC
3158
3159 if (flags & HVhek_FREEKEY)
1604cfb0 3160 Safefree(str);
19692e8d 3161
6e838c70 3162 return HeKEY_hek(entry);
fde52b5c 3163}
ecae49c0 3164
6174b39a 3165SSize_t *
ca732855
NC
3166Perl_hv_placeholders_p(pTHX_ HV *hv)
3167{
ad64d0ec 3168 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3169
7918f24d
NC
3170 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
3171
ca732855 3172 if (!mg) {
1604cfb0 3173 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
ca732855 3174
1604cfb0
MS
3175 if (!mg) {
3176 Perl_die(aTHX_ "panic: hv_placeholders_p");
3177 }
ca732855
NC
3178 }
3179 return &(mg->mg_len);
3180}
3181
3182
3183I32
0c289d13 3184Perl_hv_placeholders_get(pTHX_ const HV *hv)
ca732855 3185{
0c289d13 3186 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3187
7918f24d 3188 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
23491f1d 3189 PERL_UNUSED_CONTEXT;
7918f24d 3190
ca732855
NC
3191 return mg ? mg->mg_len : 0;
3192}
3193
3194void
ac1e784a 3195Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
ca732855 3196{
ad64d0ec 3197 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3198
7918f24d
NC
3199 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
3200
ca732855 3201 if (mg) {
1604cfb0 3202 mg->mg_len = ph;
ca732855 3203 } else if (ph) {
1604cfb0
MS
3204 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
3205 Perl_die(aTHX_ "panic: hv_placeholders_set");
ca732855
NC
3206 }
3207 /* else we don't need to add magic to record 0 placeholders. */
3208}
ecae49c0 3209
2a49f0f5 3210STATIC SV *
7b0bddfa
NC
3211S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
3212{
3213 SV *value;
7918f24d
NC
3214
3215 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
3216
7b0bddfa
NC
3217 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
3218 case HVrhek_undef:
1604cfb0
MS
3219 value = newSV(0);
3220 break;
7b0bddfa 3221 case HVrhek_delete:
1604cfb0
MS
3222 value = &PL_sv_placeholder;
3223 break;
7b0bddfa 3224 case HVrhek_IV:
1604cfb0
MS
3225 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
3226 break;
44ebaf21 3227 case HVrhek_UV:
1604cfb0
MS
3228 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
3229 break;
7b0bddfa 3230 case HVrhek_PV:
44ebaf21 3231 case HVrhek_PV_UTF8:
1604cfb0
MS
3232 /* Create a string SV that directly points to the bytes in our
3233 structure. */
3234 value = newSV_type(SVt_PV);
3235 SvPV_set(value, (char *) he->refcounted_he_data + 1);
3236 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
3237 /* This stops anything trying to free it */
3238 SvLEN_set(value, 0);
3239 SvPOK_on(value);
3240 SvREADONLY_on(value);
3241 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
3242 SvUTF8_on(value);
3243 break;
7b0bddfa 3244 default:
1604cfb0
MS
3245 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %" UVxf,
3246 (UV)he->refcounted_he_data[0]);
7b0bddfa
NC
3247 }
3248 return value;
3249}
3250
ecae49c0 3251/*
44170c9a 3252=for apidoc refcounted_he_chain_2hv
8dff4fc5 3253
20439bc7
Z
3254Generates and returns a C<HV *> representing the content of a
3255C<refcounted_he> chain.
2d7f6611 3256C<flags> is currently unused and must be zero.
8dff4fc5
BM
3257
3258=cut
3259*/
3260HV *
20439bc7 3261Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain, U32 flags)
8dff4fc5 3262{
20439bc7
Z
3263 HV *hv;
3264 U32 placeholders, max;
b3ca2e83 3265
20439bc7 3266 if (flags)
1604cfb0
MS
3267 Perl_croak(aTHX_ "panic: refcounted_he_chain_2hv bad flags %" UVxf,
3268 (UV)flags);
b3ca2e83 3269
b3ca2e83
NC
3270 /* We could chase the chain once to get an idea of the number of keys,
3271 and call ksplit. But for now we'll make a potentially inefficient
3272 hash with only 8 entries in its array. */
20439bc7
Z
3273 hv = newHV();
3274 max = HvMAX(hv);
b3ca2e83 3275 if (!HvARRAY(hv)) {
1604cfb0
MS
3276 char *array;
3277 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
3278 HvARRAY(hv) = (HE**)array;
b3ca2e83
NC
3279 }
3280
20439bc7 3281 placeholders = 0;
b3ca2e83 3282 while (chain) {
cbb1fbea 3283#ifdef USE_ITHREADS
1604cfb0 3284 U32 hash = chain->refcounted_he_hash;
cbb1fbea 3285#else
1604cfb0 3286 U32 hash = HEK_HASH(chain->refcounted_he_hek);
cbb1fbea 3287#endif
1604cfb0
MS
3288 HE **oentry = &((HvARRAY(hv))[hash & max]);
3289 HE *entry = *oentry;
3290 SV *value;
3291
3292 for (; entry; entry = HeNEXT(entry)) {
3293 if (HeHASH(entry) == hash) {
3294 /* We might have a duplicate key here. If so, entry is older
3295 than the key we've already put in the hash, so if they are
3296 the same, skip adding entry. */
9f769845 3297#ifdef USE_ITHREADS
1604cfb0
MS
3298 const STRLEN klen = HeKLEN(entry);
3299 const char *const key = HeKEY(entry);
3300 if (klen == chain->refcounted_he_keylen
3301 && (!!HeKUTF8(entry)
3302 == !!(chain->refcounted_he_data[0] & HVhek_UTF8))
3303 && memEQ(key, REF_HE_KEY(chain), klen))
3304 goto next_please;
9f769845 3305#else
1604cfb0
MS
3306 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
3307 goto next_please;
3308 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
3309 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
3310 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
3311 HeKLEN(entry)))
3312 goto next_please;
9f769845 3313#endif
1604cfb0
MS
3314 }
3315 }
3316 assert (!entry);
3317 entry = new_HE();
b3ca2e83 3318
cbb1fbea 3319#ifdef USE_ITHREADS
1604cfb0
MS
3320 HeKEY_hek(entry)
3321 = share_hek_flags(REF_HE_KEY(chain),
3322 chain->refcounted_he_keylen,
3323 chain->refcounted_he_hash,
3324 (chain->refcounted_he_data[0]
3325 & (HVhek_UTF8|HVhek_WASUTF8)));
cbb1fbea 3326#else
1604cfb0 3327 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
cbb1fbea 3328#endif
1604cfb0
MS
3329 value = refcounted_he_value(chain);
3330 if (value == &PL_sv_placeholder)
3331 placeholders++;
3332 HeVAL(entry) = value;
b3ca2e83 3333
1604cfb0
MS
3334 /* Link it into the chain. */
3335 HeNEXT(entry) = *oentry;
3336 *oentry = entry;
b3ca2e83 3337
1604cfb0 3338 HvTOTALKEYS(hv)++;
b3ca2e83
NC
3339
3340 next_please:
1604cfb0 3341 chain = chain->refcounted_he_next;
b3ca2e83
NC
3342 }
3343
3344 if (placeholders) {
1604cfb0
MS
3345 clear_placeholders(hv, placeholders);
3346 HvTOTALKEYS(hv) -= placeholders;
b3ca2e83
NC
3347 }
3348
3349 /* We could check in the loop to see if we encounter any keys with key
3350 flags, but it's probably not worth it, as this per-hash flag is only
3351 really meant as an optimisation for things like Storable. */
3352 HvHASKFLAGS_on(hv);
def9038f 3353 DEBUG_A(Perl_hv_assert(aTHX_ hv));
b3ca2e83
NC
3354
3355 return hv;
3356}
3357
20439bc7 3358/*
44170c9a 3359=for apidoc refcounted_he_fetch_pvn
20439bc7
Z
3360
3361Search along a C<refcounted_he> chain for an entry with the key specified
2d7f6611 3362by C<keypv> and C<keylen>. If C<flags> has the C<REFCOUNTED_HE_KEY_UTF8>
20439bc7 3363bit set, the key octets are interpreted as UTF-8, otherwise they
2d7f6611 3364are interpreted as Latin-1. C<hash> is a precomputed hash of the key
20439bc7
Z
3365string, or zero if it has not been precomputed. Returns a mortal scalar
3366representing the value associated with the key, or C<&PL_sv_placeholder>
3367if there is no value associated with the key.
3368
3369=cut
3370*/
3371
7b0bddfa 3372SV *
20439bc7 3373Perl_refcounted_he_fetch_pvn(pTHX_ const struct refcounted_he *chain,
1604cfb0 3374 const char *keypv, STRLEN keylen, U32 hash, U32 flags)
7b0bddfa 3375{
20439bc7
Z
3376 U8 utf8_flag;
3377 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PVN;
7b0bddfa 3378
94250aee 3379 if (flags & ~(REFCOUNTED_HE_KEY_UTF8|REFCOUNTED_HE_EXISTS))
1604cfb0
MS
3380 Perl_croak(aTHX_ "panic: refcounted_he_fetch_pvn bad flags %" UVxf,
3381 (UV)flags);
20439bc7 3382 if (!chain)
1604cfb0 3383 goto ret;
20439bc7 3384 if (flags & REFCOUNTED_HE_KEY_UTF8) {
1604cfb0
MS
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;
e8e5e5b3 3392 }
1604cfb0 3393 nonascii_count++;
e8e5e5b3 3394 p++;
1604cfb0
MS
3395 }
3396 }
3397 if (nonascii_count) {
3398 char *q;
3399 const char *p = keypv, *keyend = keypv + keylen;
3400 keylen -= nonascii_count;
3401 Newx(q, keylen, char);
3402 SAVEFREEPV(q);
3403 keypv = q;
3404 for (; p != keyend; p++, q++) {
3405 U8 c = (U8)*p;
e8e5e5b3
KW
3406 if (UTF8_IS_INVARIANT(c)) {
3407 *q = (char) c;
3408 }
3409 else {
3410 p++;
a62b247b 3411 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
e8e5e5b3 3412 }
1604cfb0
MS
3413 }
3414 }
3415 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3416 canonicalised_key: ;
20439bc7
Z
3417 }
3418 utf8_flag = (flags & REFCOUNTED_HE_KEY_UTF8) ? HVhek_UTF8 : 0;
3419 if (!hash)
1604cfb0 3420 PERL_HASH(hash, keypv, keylen);
7b0bddfa 3421
20439bc7 3422 for (; chain; chain = chain->refcounted_he_next) {
1604cfb0 3423 if (
7b0bddfa 3424#ifdef USE_ITHREADS
1604cfb0
MS
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)
7b0bddfa 3429#else
1604cfb0
MS
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)
7b0bddfa 3434#endif
1604cfb0
MS
3435 ) {
3436 if (flags & REFCOUNTED_HE_EXISTS)
3437 return (chain->refcounted_he_data[0] & HVrhek_typemask)
3438 == HVrhek_delete
3439 ? NULL : &PL_sv_yes;
3440 return sv_2mortal(refcounted_he_value(chain));
3441 }
94250aee 3442 }
71622e40 3443 ret:
94250aee 3444 return flags & REFCOUNTED_HE_EXISTS ? NULL : &PL_sv_placeholder;
20439bc7 3445}
7b0bddfa 3446
20439bc7 3447/*
44170c9a 3448=for apidoc refcounted_he_fetch_pv
7b0bddfa 3449
20439bc7
Z
3450Like L</refcounted_he_fetch_pvn>, but takes a nul-terminated string
3451instead of a string/length pair.
3452
3453=cut
3454*/
3455
3456SV *
3457Perl_refcounted_he_fetch_pv(pTHX_ const struct refcounted_he *chain,
1604cfb0 3458 const char *key, U32 hash, U32 flags)
20439bc7
Z
3459{
3460 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PV;
3461 return refcounted_he_fetch_pvn(chain, key, strlen(key), hash, flags);
7b0bddfa
NC
3462}
3463
b3ca2e83 3464/*
44170c9a 3465=for apidoc refcounted_he_fetch_sv
20439bc7
Z
3466
3467Like L</refcounted_he_fetch_pvn>, but takes a Perl scalar instead of a
3468string/length pair.
3469
3470=cut
3471*/
b3ca2e83 3472
20439bc7
Z
3473SV *
3474Perl_refcounted_he_fetch_sv(pTHX_ const struct refcounted_he *chain,
1604cfb0 3475 SV *key, U32 hash, U32 flags)
20439bc7
Z
3476{
3477 const char *keypv;
3478 STRLEN keylen;
3479 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_SV;
3480 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0
MS
3481 Perl_croak(aTHX_ "panic: refcounted_he_fetch_sv bad flags %" UVxf,
3482 (UV)flags);
20439bc7
Z
3483 keypv = SvPV_const(key, keylen);
3484 if (SvUTF8(key))
1604cfb0 3485 flags |= REFCOUNTED_HE_KEY_UTF8;
20439bc7 3486 if (!hash && SvIsCOW_shared_hash(key))
1604cfb0 3487 hash = SvSHARED_HASH(key);
20439bc7
Z
3488 return refcounted_he_fetch_pvn(chain, keypv, keylen, hash, flags);
3489}
3490
3491/*
44170c9a 3492=for apidoc refcounted_he_new_pvn
20439bc7
Z
3493
3494Creates a new C<refcounted_he>. This consists of a single key/value
3495pair and a reference to an existing C<refcounted_he> chain (which may
3496be empty), and thus forms a longer chain. When using the longer chain,
3497the new key/value pair takes precedence over any entry for the same key
3498further along the chain.
3499
2d7f6611 3500The new key is specified by C<keypv> and C<keylen>. If C<flags> has
20439bc7 3501the C<REFCOUNTED_HE_KEY_UTF8> bit set, the key octets are interpreted
2d7f6611 3502as UTF-8, otherwise they are interpreted as Latin-1. C<hash> is
20439bc7
Z
3503a precomputed hash of the key string, or zero if it has not been
3504precomputed.
3505
2d7f6611 3506C<value> is the scalar value to store for this key. C<value> is copied
20439bc7
Z
3507by this function, which thus does not take ownership of any reference
3508to it, and later changes to the scalar will not be reflected in the
3509value visible in the C<refcounted_he>. Complex types of scalar will not
3510be stored with referential integrity, but will be coerced to strings.
2d7f6611 3511C<value> may be either null or C<&PL_sv_placeholder> to indicate that no
20439bc7
Z
3512value is to be associated with the key; this, as with any non-null value,
3513takes precedence over the existence of a value for the key further along
3514the chain.
3515
2d7f6611 3516C<parent> points to the rest of the C<refcounted_he> chain to be
20439bc7 3517attached to the new C<refcounted_he>. This function takes ownership
2d7f6611 3518of one reference to C<parent>, and returns one reference to the new
20439bc7 3519C<refcounted_he>.
b3ca2e83
NC
3520
3521=cut
3522*/
3523
3524struct refcounted_he *
20439bc7 3525Perl_refcounted_he_new_pvn(pTHX_ struct refcounted_he *parent,
1604cfb0 3526 const char *keypv, STRLEN keylen, U32 hash, SV *value, U32 flags)
20439bc7 3527{
b6bbf3fa 3528 STRLEN value_len = 0;
95b63a38 3529 const char *value_p = NULL;
20439bc7 3530 bool is_pv;
b6bbf3fa 3531 char value_type;
20439bc7
Z
3532 char hekflags;
3533 STRLEN key_offset = 1;
3534 struct refcounted_he *he;
3535 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PVN;
b6bbf3fa 3536
20439bc7 3537 if (!value || value == &PL_sv_placeholder) {
1604cfb0 3538 value_type = HVrhek_delete;
20439bc7 3539 } else if (SvPOK(value)) {
1604cfb0 3540 value_type = HVrhek_PV;
b6bbf3fa 3541 } else if (SvIOK(value)) {
1604cfb0 3542 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
b6bbf3fa 3543 } else if (!SvOK(value)) {
1604cfb0 3544 value_type = HVrhek_undef;
b6bbf3fa 3545 } else {
1604cfb0 3546 value_type = HVrhek_PV;
b6bbf3fa 3547 }
20439bc7
Z
3548 is_pv = value_type == HVrhek_PV;
3549 if (is_pv) {
1604cfb0
MS
3550 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
3551 the value is overloaded, and doesn't yet have the UTF-8flag set. */
3552 value_p = SvPV_const(value, value_len);
3553 if (SvUTF8(value))
3554 value_type = HVrhek_PV_UTF8;
3555 key_offset = value_len + 2;
20439bc7
Z
3556 }
3557 hekflags = value_type;
3558
3559 if (flags & REFCOUNTED_HE_KEY_UTF8) {
1604cfb0
MS
3560 /* Canonicalise to Latin-1 where possible. */
3561 const char *keyend = keypv + keylen, *p;
3562 STRLEN nonascii_count = 0;
3563 for (p = keypv; p != keyend; p++) {
3564 if (! UTF8_IS_INVARIANT(*p)) {
3565 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) {
3566 goto canonicalised_key;
e8e5e5b3 3567 }
1604cfb0 3568 nonascii_count++;
e8e5e5b3 3569 p++;
1604cfb0
MS
3570 }
3571 }
3572 if (nonascii_count) {
3573 char *q;
3574 const char *p = keypv, *keyend = keypv + keylen;
3575 keylen -= nonascii_count;
3576 Newx(q, keylen, char);
3577 SAVEFREEPV(q);
3578 keypv = q;
3579 for (; p != keyend; p++, q++) {
3580 U8 c = (U8)*p;
e8e5e5b3
KW
3581 if (UTF8_IS_INVARIANT(c)) {
3582 *q = (char) c;
3583 }
3584 else {
3585 p++;
a62b247b 3586 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
e8e5e5b3 3587 }
1604cfb0
MS
3588 }
3589 }
3590 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3591 canonicalised_key: ;
b6bbf3fa 3592 }
20439bc7 3593 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0 3594 hekflags |= HVhek_UTF8;
20439bc7 3595 if (!hash)
1604cfb0 3596 PERL_HASH(hash, keypv, keylen);
012da8e5 3597
0de694c5 3598#ifdef USE_ITHREADS
10edeb5d 3599 he = (struct refcounted_he*)
1604cfb0
MS
3600 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3601 + keylen
3602 + key_offset);
0de694c5
NC
3603#else
3604 he = (struct refcounted_he*)
1604cfb0
MS
3605 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3606 + key_offset);
0de694c5 3607#endif
b3ca2e83 3608
71ad1b0c 3609 he->refcounted_he_next = parent;
b6bbf3fa 3610
012da8e5 3611 if (is_pv) {
1604cfb0
MS
3612 Copy(value_p, he->refcounted_he_data + 1, value_len + 1, char);
3613 he->refcounted_he_val.refcounted_he_u_len = value_len;
b6bbf3fa 3614 } else if (value_type == HVrhek_IV) {
1604cfb0 3615 he->refcounted_he_val.refcounted_he_u_iv = SvIVX(value);
012da8e5 3616 } else if (value_type == HVrhek_UV) {
1604cfb0 3617 he->refcounted_he_val.refcounted_he_u_uv = SvUVX(value);
b6bbf3fa
NC
3618 }
3619
cbb1fbea 3620#ifdef USE_ITHREADS
b6bbf3fa 3621 he->refcounted_he_hash = hash;
20439bc7
Z
3622 he->refcounted_he_keylen = keylen;
3623 Copy(keypv, he->refcounted_he_data + key_offset, keylen, char);
cbb1fbea 3624#else
20439bc7 3625 he->refcounted_he_hek = share_hek_flags(keypv, keylen, hash, hekflags);
cbb1fbea 3626#endif
b6bbf3fa 3627
20439bc7 3628 he->refcounted_he_data[0] = hekflags;
b3ca2e83
NC
3629 he->refcounted_he_refcnt = 1;
3630
3631 return he;
3632}
3633
3634/*
44170c9a 3635=for apidoc refcounted_he_new_pv
b3ca2e83 3636
20439bc7
Z
3637Like L</refcounted_he_new_pvn>, but takes a nul-terminated string instead
3638of a string/length pair.
3639
3640=cut
3641*/
3642
3643struct refcounted_he *
3644Perl_refcounted_he_new_pv(pTHX_ struct refcounted_he *parent,
1604cfb0 3645 const char *key, U32 hash, SV *value, U32 flags)
20439bc7
Z
3646{
3647 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PV;
3648 return refcounted_he_new_pvn(parent, key, strlen(key), hash, value, flags);
3649}
3650
3651/*
44170c9a 3652=for apidoc refcounted_he_new_sv
20439bc7
Z
3653
3654Like L</refcounted_he_new_pvn>, but takes a Perl scalar instead of a
3655string/length pair.
3656
3657=cut
3658*/
3659
3660struct refcounted_he *
3661Perl_refcounted_he_new_sv(pTHX_ struct refcounted_he *parent,
1604cfb0 3662 SV *key, U32 hash, SV *value, U32 flags)
20439bc7
Z
3663{
3664 const char *keypv;
3665 STRLEN keylen;
3666 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_SV;
3667 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0
MS
3668 Perl_croak(aTHX_ "panic: refcounted_he_new_sv bad flags %" UVxf,
3669 (UV)flags);
20439bc7
Z
3670 keypv = SvPV_const(key, keylen);
3671 if (SvUTF8(key))
1604cfb0 3672 flags |= REFCOUNTED_HE_KEY_UTF8;
20439bc7 3673 if (!hash && SvIsCOW_shared_hash(key))
1604cfb0 3674 hash = SvSHARED_HASH(key);
20439bc7
Z
3675 return refcounted_he_new_pvn(parent, keypv, keylen, hash, value, flags);
3676}
3677
3678/*
44170c9a 3679=for apidoc refcounted_he_free
20439bc7
Z
3680
3681Decrements the reference count of a C<refcounted_he> by one. If the
3682reference count reaches zero the structure's memory is freed, which
3683(recursively) causes a reduction of its parent C<refcounted_he>'s
3684reference count. It is safe to pass a null pointer to this function:
3685no action occurs in this case.
b3ca2e83
NC
3686
3687=cut
3688*/
3689
3690void
3691Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
57ca3b03
AL
3692 PERL_UNUSED_CONTEXT;
3693
b3ca2e83 3694 while (he) {
1604cfb0
MS
3695 struct refcounted_he *copy;
3696 U32 new_count;
3697
3698 HINTS_REFCNT_LOCK;
3699 new_count = --he->refcounted_he_refcnt;
3700 HINTS_REFCNT_UNLOCK;
3701
3702 if (new_count) {
3703 return;
3704 }
b3ca2e83 3705
b6bbf3fa 3706#ifndef USE_ITHREADS
1604cfb0 3707 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
cbb1fbea 3708#endif
1604cfb0
MS
3709 copy = he;
3710 he = he->refcounted_he_next;
3711 PerlMemShared_free(copy);
b3ca2e83
NC
3712 }
3713}
3714
20439bc7 3715/*
44170c9a 3716=for apidoc refcounted_he_inc
20439bc7
Z
3717
3718Increment the reference count of a C<refcounted_he>. The pointer to the
3719C<refcounted_he> is also returned. It is safe to pass a null pointer
3720to this function: no action occurs and a null pointer is returned.
3721
3722=cut
3723*/
3724
3725struct refcounted_he *
3726Perl_refcounted_he_inc(pTHX_ struct refcounted_he *he)
3727{
dc3bf405 3728 PERL_UNUSED_CONTEXT;
20439bc7 3729 if (he) {
1604cfb0
MS
3730 HINTS_REFCNT_LOCK;
3731 he->refcounted_he_refcnt++;
3732 HINTS_REFCNT_UNLOCK;
20439bc7
Z
3733 }
3734 return he;
3735}
3736
8375c93e 3737/*
3f620621 3738=for apidoc_section $COP
aebc0cbe 3739=for apidoc cop_fetch_label
8375c93e 3740
7df56744
KW
3741Returns the label attached to a cop, and stores its length in bytes into
3742C<*len>.
3743Upon return, C<*flags> will be set to either C<SVf_UTF8> or 0.
3744
eb992c6f 3745Alternatively, use the macro C<L</CopLABEL_len_flags>>;
7df56744 3746or if you don't need to know if the label is UTF-8 or not, the macro
eb992c6f
KW
3747C<L</CopLABEL_len>>;
3748or if you additionally dont need to know the length, C<L</CopLABEL>>.
8375c93e
RU
3749
3750=cut
3751*/
3752
47550813
NC
3753/* pp_entereval is aware that labels are stored with a key ':' at the top of
3754 the linked list. */
dca6062a 3755const char *
aebc0cbe 3756Perl_cop_fetch_label(pTHX_ COP *const cop, STRLEN *len, U32 *flags) {
d6747b7a
NC
3757 struct refcounted_he *const chain = cop->cop_hints_hash;
3758
aebc0cbe 3759 PERL_ARGS_ASSERT_COP_FETCH_LABEL;
dc3bf405 3760 PERL_UNUSED_CONTEXT;
d6747b7a 3761
dca6062a 3762 if (!chain)
1604cfb0 3763 return NULL;
dca6062a
NC
3764#ifdef USE_ITHREADS
3765 if (chain->refcounted_he_keylen != 1)
1604cfb0 3766 return NULL;
dca6062a 3767 if (*REF_HE_KEY(chain) != ':')
1604cfb0 3768 return NULL;
dca6062a
NC
3769#else
3770 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
1604cfb0 3771 return NULL;
dca6062a 3772 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
1604cfb0 3773 return NULL;
dca6062a 3774#endif
012da8e5
NC
3775 /* Stop anyone trying to really mess us up by adding their own value for
3776 ':' into %^H */
3777 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
1604cfb0
MS
3778 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
3779 return NULL;
012da8e5 3780
dca6062a 3781 if (len)
1604cfb0 3782 *len = chain->refcounted_he_val.refcounted_he_u_len;
dca6062a 3783 if (flags) {
1604cfb0
MS
3784 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
3785 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
dca6062a
NC
3786 }
3787 return chain->refcounted_he_data + 1;
3788}
3789
8375c93e 3790/*
aebc0cbe 3791=for apidoc cop_store_label
8375c93e 3792
72d33970
FC
3793Save a label into a C<cop_hints_hash>.
3794You need to set flags to C<SVf_UTF8>
5f608e5f 3795for a UTF-8 label. Any other flag is ignored.
8375c93e
RU
3796
3797=cut
3798*/
3799
a77ac40c 3800void
aebc0cbe 3801Perl_cop_store_label(pTHX_ COP *const cop, const char *label, STRLEN len,
1604cfb0 3802 U32 flags)
012da8e5 3803{
20439bc7 3804 SV *labelsv;
aebc0cbe 3805 PERL_ARGS_ASSERT_COP_STORE_LABEL;
547bb267 3806
a77ac40c 3807 if (flags & ~(SVf_UTF8))
1604cfb0
MS
3808 Perl_croak(aTHX_ "panic: cop_store_label illegal flag bits 0x%" UVxf,
3809 (UV)flags);
a3179684 3810 labelsv = newSVpvn_flags(label, len, SVs_TEMP);
20439bc7 3811 if (flags & SVf_UTF8)
1604cfb0 3812 SvUTF8_on(labelsv);
a77ac40c 3813 cop->cop_hints_hash
1604cfb0 3814 = refcounted_he_new_pvs(cop->cop_hints_hash, ":", labelsv, 0);
012da8e5
NC
3815}
3816
b3ca2e83 3817/*
3f620621 3818=for apidoc_section $HV
ecae49c0
NC
3819=for apidoc hv_assert
3820
3821Check that a hash is in an internally consistent state.
3822
3823=cut
3824*/
3825
943795c2
NC
3826#ifdef DEBUGGING
3827
ecae49c0
NC
3828void
3829Perl_hv_assert(pTHX_ HV *hv)
3830{
57ca3b03
AL
3831 HE* entry;
3832 int withflags = 0;
3833 int placeholders = 0;
3834 int real = 0;
3835 int bad = 0;
3836 const I32 riter = HvRITER_get(hv);
3837 HE *eiter = HvEITER_get(hv);
3838
7918f24d
NC
3839 PERL_ARGS_ASSERT_HV_ASSERT;
3840
57ca3b03
AL
3841 (void)hv_iterinit(hv);
3842
3843 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
1604cfb0
MS
3844 /* sanity check the values */
3845 if (HeVAL(entry) == &PL_sv_placeholder)
3846 placeholders++;
3847 else
3848 real++;
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)) {
3853 withflags++;
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));
3858 bad = 1;
3859 }
3860 } else if (HeKWASUTF8(entry))
3861 withflags++;
57ca3b03 3862 }
ad64d0ec 3863 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
1604cfb0
MS
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);
3867
3868 if (nhashkeys != real) {
3869 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
3870 bad = 1;
3871 }
3872 if (nhashplaceholders != placeholders) {
3873 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
3874 bad = 1;
3875 }
57ca3b03
AL
3876 }
3877 if (withflags && ! HvHASKFLAGS(hv)) {
1604cfb0
MS
3878 PerlIO_printf(Perl_debug_log,
3879 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
3880 withflags);
3881 bad = 1;
57ca3b03
AL
3882 }
3883 if (bad) {
1604cfb0 3884 sv_dump(MUTABLE_SV(hv));
57ca3b03
AL
3885 }
3886 HvRITER_set(hv, riter); /* Restore hash iterator state */
3887 HvEITER_set(hv, eiter);
ecae49c0 3888}
af3babe4 3889
943795c2
NC
3890#endif
3891
af3babe4 3892/*
14d04a33 3893 * ex: set ts=8 sts=4 sw=4 et:
37442d52 3894 */