This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
update PERL_MEM_LOG in perlhack.pod
[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
RB
19/*
20=head1 Hash Manipulation Functions
166f8a29
DM
21
22A HV structure represents a Perl hash. It consists mainly of an array
23of pointers, each of which points to a linked list of HE structures. The
24array is indexed by the hash function of the key, so each linked list
25represents all the hash entries with the same hash value. Each HE contains
26a pointer to the actual value, plus a pointer to a HEK structure which
27holds the key and hash value.
28
29=cut
30
d5afce77
RB
31*/
32
79072805 33#include "EXTERN.h"
864dbfa3 34#define PERL_IN_HV_C
3d78eb94 35#define PERL_HASH_INTERNAL_ACCESS
79072805
LW
36#include "perl.h"
37
d8012aaf 38#define HV_MAX_LENGTH_BEFORE_SPLIT 14
fdcd69b6 39
d75ce684 40static const char S_strtab_error[]
5d2b1485
NC
41 = "Cannot modify shared string table in hv_%s";
42
cac9b346
NC
43STATIC void
44S_more_he(pTHX)
45{
97aff369 46 dVAR;
d8fca402
NC
47 /* We could generate this at compile time via (another) auxiliary C
48 program? */
49 const size_t arena_size = Perl_malloc_good_size(PERL_ARENA_SIZE);
50 HE* he = (HE*) Perl_get_arena(aTHX_ arena_size, HE_SVSLOT);
51 HE * const heend = &he[arena_size / sizeof(HE) - 1];
cac9b346 52
d2a0f284 53 PL_body_roots[HE_SVSLOT] = he;
cac9b346
NC
54 while (he < heend) {
55 HeNEXT(he) = (HE*)(he + 1);
56 he++;
57 }
58 HeNEXT(he) = 0;
59}
60
c941fb51
NC
61#ifdef PURIFY
62
63#define new_HE() (HE*)safemalloc(sizeof(HE))
64#define del_HE(p) safefree((char*)p)
65
66#else
67
76e3520e 68STATIC HE*
cea2e8a9 69S_new_he(pTHX)
4633a7c4 70{
97aff369 71 dVAR;
4633a7c4 72 HE* he;
0bd48802 73 void ** const root = &PL_body_roots[HE_SVSLOT];
6a93a7e5 74
6a93a7e5 75 if (!*root)
cac9b346 76 S_more_he(aTHX);
10edeb5d 77 he = (HE*) *root;
ce3e5c45 78 assert(he);
6a93a7e5 79 *root = HeNEXT(he);
333f433b 80 return he;
4633a7c4
LW
81}
82
c941fb51
NC
83#define new_HE() new_he()
84#define del_HE(p) \
85 STMT_START { \
6a93a7e5
NC
86 HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \
87 PL_body_roots[HE_SVSLOT] = p; \
c941fb51 88 } STMT_END
d33b2eba 89
d33b2eba 90
d33b2eba
GS
91
92#endif
93
76e3520e 94STATIC HEK *
5f66b61c 95S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
bbce6d69 96{
35a4481c 97 const int flags_masked = flags & HVhek_MASK;
bbce6d69 98 char *k;
99 register HEK *hek;
1c846c1f 100
7918f24d
NC
101 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
102
a02a5408 103 Newx(k, HEK_BASESIZE + len + 2, char);
bbce6d69 104 hek = (HEK*)k;
ff68c719 105 Copy(str, HEK_KEY(hek), len, char);
e05949c7 106 HEK_KEY(hek)[len] = 0;
ff68c719 107 HEK_LEN(hek) = len;
108 HEK_HASH(hek) = hash;
45e34800 109 HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED;
dcf933a4
NC
110
111 if (flags & HVhek_FREEKEY)
112 Safefree(str);
bbce6d69 113 return hek;
114}
115
4a31713e 116/* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
dd28f7bb
DM
117 * for tied hashes */
118
119void
120Perl_free_tied_hv_pool(pTHX)
121{
97aff369 122 dVAR;
dd28f7bb
DM
123 HE *he = PL_hv_fetch_ent_mh;
124 while (he) {
9d4ba2ae 125 HE * const ohe = he;
dd28f7bb 126 Safefree(HeKEY_hek(he));
dd28f7bb
DM
127 he = HeNEXT(he);
128 del_HE(ohe);
129 }
4608196e 130 PL_hv_fetch_ent_mh = NULL;
dd28f7bb
DM
131}
132
d18c6117 133#if defined(USE_ITHREADS)
0bff533c
NC
134HEK *
135Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
136{
566771cc 137 HEK *shared;
9d4ba2ae 138
7918f24d 139 PERL_ARGS_ASSERT_HEK_DUP;
9d4ba2ae 140 PERL_UNUSED_ARG(param);
0bff533c 141
566771cc
NC
142 if (!source)
143 return NULL;
144
145 shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
0bff533c
NC
146 if (shared) {
147 /* We already shared this hash key. */
454f1e26 148 (void)share_hek_hek(shared);
0bff533c
NC
149 }
150 else {
658b4a4a 151 shared
6e838c70
NC
152 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
153 HEK_HASH(source), HEK_FLAGS(source));
658b4a4a 154 ptr_table_store(PL_ptr_table, source, shared);
0bff533c 155 }
658b4a4a 156 return shared;
0bff533c
NC
157}
158
d18c6117 159HE *
5c4138a0 160Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
d18c6117
GS
161{
162 HE *ret;
163
7918f24d
NC
164 PERL_ARGS_ASSERT_HE_DUP;
165
d18c6117 166 if (!e)
4608196e 167 return NULL;
7766f137
GS
168 /* look for it in the table first */
169 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
170 if (ret)
171 return ret;
172
173 /* create anew and remember what it is */
d33b2eba 174 ret = new_HE();
7766f137
GS
175 ptr_table_store(PL_ptr_table, e, ret);
176
d2d73c3e 177 HeNEXT(ret) = he_dup(HeNEXT(e),shared, param);
dd28f7bb
DM
178 if (HeKLEN(e) == HEf_SVKEY) {
179 char *k;
ad64d0ec 180 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
dd28f7bb 181 HeKEY_hek(ret) = (HEK*)k;
d2d73c3e 182 HeKEY_sv(ret) = SvREFCNT_inc(sv_dup(HeKEY_sv(e), param));
dd28f7bb 183 }
c21d1a0f 184 else if (shared) {
0bff533c
NC
185 /* This is hek_dup inlined, which seems to be important for speed
186 reasons. */
1b6737cc 187 HEK * const source = HeKEY_hek(e);
658b4a4a 188 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
c21d1a0f
NC
189
190 if (shared) {
191 /* We already shared this hash key. */
454f1e26 192 (void)share_hek_hek(shared);
c21d1a0f
NC
193 }
194 else {
658b4a4a 195 shared
6e838c70
NC
196 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
197 HEK_HASH(source), HEK_FLAGS(source));
658b4a4a 198 ptr_table_store(PL_ptr_table, source, shared);
c21d1a0f 199 }
658b4a4a 200 HeKEY_hek(ret) = shared;
c21d1a0f 201 }
d18c6117 202 else
19692e8d
NC
203 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
204 HeKFLAGS(e));
d2d73c3e 205 HeVAL(ret) = SvREFCNT_inc(sv_dup(HeVAL(e), param));
d18c6117
GS
206 return ret;
207}
208#endif /* USE_ITHREADS */
209
1b1f1335 210static void
2393f1b9
JH
211S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
212 const char *msg)
1b1f1335 213{
1b6737cc 214 SV * const sv = sv_newmortal();
7918f24d
NC
215
216 PERL_ARGS_ASSERT_HV_NOTALLOWED;
217
19692e8d 218 if (!(flags & HVhek_FREEKEY)) {
1b1f1335
NIS
219 sv_setpvn(sv, key, klen);
220 }
221 else {
222 /* Need to free saved eventually assign to mortal SV */
34c3c4e3 223 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
1b1f1335
NIS
224 sv_usepvn(sv, (char *) key, klen);
225 }
19692e8d 226 if (flags & HVhek_UTF8) {
1b1f1335
NIS
227 SvUTF8_on(sv);
228 }
be2597df 229 Perl_croak(aTHX_ msg, SVfARG(sv));
1b1f1335
NIS
230}
231
fde52b5c 232/* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
233 * contains an SV* */
234
34a6f7b4
NC
235/*
236=for apidoc hv_store
237
238Stores an SV in a hash. The hash key is specified as C<key> and C<klen> is
239the length of the key. The C<hash> parameter is the precomputed hash
240value; if it is zero then Perl will compute it. The return value will be
241NULL if the operation failed or if the value did not need to be actually
242stored within the hash (as in the case of tied hashes). Otherwise it can
243be dereferenced to get the original C<SV*>. Note that the caller is
244responsible for suitably incrementing the reference count of C<val> before
245the call, and decrementing it if the function returned NULL. Effectively
246a successful hv_store takes ownership of one reference to C<val>. This is
247usually what you want; a newly created SV has a reference count of one, so
248if all your code does is create SVs then store them in a hash, hv_store
249will own the only reference to the new SV, and your code doesn't need to do
250anything further to tidy up. hv_store is not implemented as a call to
251hv_store_ent, and does not create a temporary SV for the key, so if your
252key data is not already in SV form then use hv_store in preference to
253hv_store_ent.
254
255See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
256information on how to use this function on tied hashes.
257
34a6f7b4
NC
258=for apidoc hv_store_ent
259
260Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
261parameter is the precomputed hash value; if it is zero then Perl will
262compute it. The return value is the new hash entry so created. It will be
263NULL if the operation failed or if the value did not need to be actually
264stored within the hash (as in the case of tied hashes). Otherwise the
265contents of the return value can be accessed using the C<He?> macros
266described here. Note that the caller is responsible for suitably
267incrementing the reference count of C<val> before the call, and
268decrementing it if the function returned NULL. Effectively a successful
269hv_store_ent takes ownership of one reference to C<val>. This is
270usually what you want; a newly created SV has a reference count of one, so
271if all your code does is create SVs then store them in a hash, hv_store
272will own the only reference to the new SV, and your code doesn't need to do
273anything further to tidy up. Note that hv_store_ent only reads the C<key>;
274unlike C<val> it does not take ownership of it, so maintaining the correct
275reference count on C<key> is entirely the caller's responsibility. hv_store
276is not implemented as a call to hv_store_ent, and does not create a temporary
277SV for the key, so if your key data is not already in SV form then use
278hv_store in preference to hv_store_ent.
279
280See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
281information on how to use this function on tied hashes.
282
34a6f7b4
NC
283=for apidoc hv_exists
284
285Returns a boolean indicating whether the specified hash key exists. The
286C<klen> is the length of the key.
287
954c1994
GS
288=for apidoc hv_fetch
289
290Returns the SV which corresponds to the specified key in the hash. The
291C<klen> is the length of the key. If C<lval> is set then the fetch will be
292part of a store. Check that the return value is non-null before
d1be9408 293dereferencing 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
300Returns a boolean indicating whether the specified hash key exists. C<hash>
301can be a valid precomputed hash value, or 0 to ask for it to be
302computed.
303
304=cut
305*/
306
d1be9408 307/* returns an HE * structure with the all fields set */
fde52b5c 308/* note that hent_val will be a mortal sv for MAGICAL hashes */
954c1994
GS
309/*
310=for apidoc hv_fetch_ent
311
312Returns the hash entry which corresponds to the specified key in the hash.
313C<hash> must be a valid precomputed hash number for the given C<key>, or 0
314if you want the function to compute it. IF C<lval> is set then the fetch
315will be part of a store. Make sure the return value is non-null before
316accessing it. The return value when C<tb> is a tied hash is a pointer to a
317static location, so be sure to make a copy of the structure if you need to
1c846c1f 318store it somewhere.
954c1994 319
96f1132b 320See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
954c1994
GS
321information on how to use this function on tied hashes.
322
323=cut
324*/
325
a038e571
NC
326/* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
327void *
328Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
329 const int action, SV *val, const U32 hash)
330{
331 STRLEN klen;
332 int flags;
333
7918f24d
NC
334 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
335
a038e571
NC
336 if (klen_i32 < 0) {
337 klen = -klen_i32;
338 flags = HVhek_UTF8;
339 } else {
340 klen = klen_i32;
341 flags = 0;
342 }
343 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
344}
345
63c89345 346void *
d3ba3f5c
NC
347Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
348 int flags, int action, SV *val, register U32 hash)
113738bb 349{
27da23d5 350 dVAR;
b2c64049 351 XPVHV* xhv;
b2c64049
NC
352 HE *entry;
353 HE **oentry;
fde52b5c 354 SV *sv;
da58a35d 355 bool is_utf8;
113738bb 356 int masked_flags;
3c84c864 357 const int return_svp = action & HV_FETCH_JUST_SV;
fde52b5c 358
359 if (!hv)
a4fc7abc 360 return NULL;
8265e3d1
NC
361 if (SvTYPE(hv) == SVTYPEMASK)
362 return NULL;
363
364 assert(SvTYPE(hv) == SVt_PVHV);
fde52b5c 365
bdee33e4 366 if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) {
fda2d18a 367 MAGIC* mg;
ad64d0ec 368 if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) {
fda2d18a
NC
369 struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr;
370 if (uf->uf_set == NULL) {
371 SV* obj = mg->mg_obj;
372
373 if (!keysv) {
59cd0e26
NC
374 keysv = newSVpvn_flags(key, klen, SVs_TEMP |
375 ((flags & HVhek_UTF8)
376 ? SVf_UTF8 : 0));
fda2d18a
NC
377 }
378
379 mg->mg_obj = keysv; /* pass key */
380 uf->uf_index = action; /* pass action */
ad64d0ec 381 magic_getuvar(MUTABLE_SV(hv), mg);
fda2d18a
NC
382 keysv = mg->mg_obj; /* may have changed */
383 mg->mg_obj = obj;
384
385 /* If the key may have changed, then we need to invalidate
386 any passed-in computed hash value. */
387 hash = 0;
388 }
389 }
bdee33e4 390 }
113738bb 391 if (keysv) {
e593d2fe
AE
392 if (flags & HVhek_FREEKEY)
393 Safefree(key);
5c144d81 394 key = SvPV_const(keysv, klen);
113738bb 395 is_utf8 = (SvUTF8(keysv) != 0);
44b87b50
NC
396 if (SvIsCOW_shared_hash(keysv)) {
397 flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0);
398 } else {
399 flags = 0;
400 }
113738bb 401 } else {
c1fe5510 402 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
113738bb 403 }
113738bb 404
9dbc5603 405 if (action & HV_DELETE) {
3c84c864
NC
406 return (void *) hv_delete_common(hv, keysv, key, klen,
407 flags | (is_utf8 ? HVhek_UTF8 : 0),
408 action, hash);
9dbc5603
NC
409 }
410
b2c64049 411 xhv = (XPVHV*)SvANY(hv);
7f66fda2 412 if (SvMAGICAL(hv)) {
6136c704 413 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
ad64d0ec
NC
414 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
415 || SvGMAGICAL((const SV *)hv))
e62cc96a 416 {
3c84c864 417 /* FIXME should be able to skimp on the HE/HEK here when
7f66fda2 418 HV_FETCH_JUST_SV is true. */
7f66fda2 419 if (!keysv) {
740cce10
NC
420 keysv = newSVpvn_utf8(key, klen, is_utf8);
421 } else {
7f66fda2 422 keysv = newSVsv(keysv);
113738bb 423 }
44a2ac75 424 sv = sv_newmortal();
ad64d0ec 425 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
7f66fda2
NC
426
427 /* grab a fake HE/HEK pair from the pool or make a new one */
428 entry = PL_hv_fetch_ent_mh;
429 if (entry)
430 PL_hv_fetch_ent_mh = HeNEXT(entry);
431 else {
432 char *k;
433 entry = new_HE();
ad64d0ec 434 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
7f66fda2
NC
435 HeKEY_hek(entry) = (HEK*)k;
436 }
4608196e 437 HeNEXT(entry) = NULL;
7f66fda2
NC
438 HeSVKEY_set(entry, keysv);
439 HeVAL(entry) = sv;
440 sv_upgrade(sv, SVt_PVLV);
441 LvTYPE(sv) = 'T';
442 /* so we can free entry when freeing sv */
ad64d0ec 443 LvTARG(sv) = MUTABLE_SV(entry);
7f66fda2
NC
444
445 /* XXX remove at some point? */
446 if (flags & HVhek_FREEKEY)
447 Safefree(key);
448
3c84c864
NC
449 if (return_svp) {
450 return entry ? (void *) &HeVAL(entry) : NULL;
451 }
452 return (void *) entry;
113738bb 453 }
7f66fda2 454#ifdef ENV_IS_CASELESS
ad64d0ec 455 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
7f66fda2
NC
456 U32 i;
457 for (i = 0; i < klen; ++i)
458 if (isLOWER(key[i])) {
086cb327
NC
459 /* Would be nice if we had a routine to do the
460 copy and upercase in a single pass through. */
0bd48802 461 const char * const nkey = strupr(savepvn(key,klen));
086cb327
NC
462 /* Note that this fetch is for nkey (the uppercased
463 key) whereas the store is for key (the original) */
63c89345
NC
464 void *result = hv_common(hv, NULL, nkey, klen,
465 HVhek_FREEKEY, /* free nkey */
466 0 /* non-LVAL fetch */
3c84c864
NC
467 | HV_DISABLE_UVAR_XKEY
468 | return_svp,
63c89345
NC
469 NULL /* no value */,
470 0 /* compute hash */);
26488bcf 471 if (!result && (action & HV_FETCH_LVALUE)) {
086cb327
NC
472 /* This call will free key if necessary.
473 Do it this way to encourage compiler to tail
474 call optimise. */
63c89345
NC
475 result = hv_common(hv, keysv, key, klen, flags,
476 HV_FETCH_ISSTORE
3c84c864
NC
477 | HV_DISABLE_UVAR_XKEY
478 | return_svp,
63c89345 479 newSV(0), hash);
086cb327
NC
480 } else {
481 if (flags & HVhek_FREEKEY)
482 Safefree(key);
483 }
63c89345 484 return result;
7f66fda2 485 }
902173a3 486 }
7f66fda2
NC
487#endif
488 } /* ISFETCH */
489 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
ad64d0ec
NC
490 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
491 || SvGMAGICAL((const SV *)hv)) {
b2c64049
NC
492 /* I don't understand why hv_exists_ent has svret and sv,
493 whereas hv_exists only had one. */
9d4ba2ae 494 SV * const svret = sv_newmortal();
b2c64049 495 sv = sv_newmortal();
7f66fda2
NC
496
497 if (keysv || is_utf8) {
498 if (!keysv) {
740cce10 499 keysv = newSVpvn_utf8(key, klen, TRUE);
7f66fda2
NC
500 } else {
501 keysv = newSVsv(keysv);
502 }
ad64d0ec 503 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
b2c64049 504 } else {
ad64d0ec 505 mg_copy(MUTABLE_SV(hv), sv, key, klen);
7f66fda2 506 }
b2c64049
NC
507 if (flags & HVhek_FREEKEY)
508 Safefree(key);
7f66fda2
NC
509 magic_existspack(svret, mg_find(sv, PERL_MAGIC_tiedelem));
510 /* This cast somewhat evil, but I'm merely using NULL/
511 not NULL to return the boolean exists.
512 And I know hv is not NULL. */
3c84c864 513 return SvTRUE(svret) ? (void *)hv : NULL;
e7152ba2 514 }
7f66fda2 515#ifdef ENV_IS_CASELESS
ad64d0ec 516 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
7f66fda2 517 /* XXX This code isn't UTF8 clean. */
a15d23f8 518 char * const keysave = (char * const)key;
b2c64049
NC
519 /* Will need to free this, so set FREEKEY flag. */
520 key = savepvn(key,klen);
521 key = (const char*)strupr((char*)key);
6136c704 522 is_utf8 = FALSE;
7f66fda2 523 hash = 0;
8b4f7dd5 524 keysv = 0;
b2c64049
NC
525
526 if (flags & HVhek_FREEKEY) {
527 Safefree(keysave);
528 }
529 flags |= HVhek_FREEKEY;
7f66fda2 530 }
902173a3 531#endif
7f66fda2 532 } /* ISEXISTS */
b2c64049
NC
533 else if (action & HV_FETCH_ISSTORE) {
534 bool needs_copy;
535 bool needs_store;
536 hv_magic_check (hv, &needs_copy, &needs_store);
537 if (needs_copy) {
a3b680e6 538 const bool save_taint = PL_tainted;
b2c64049
NC
539 if (keysv || is_utf8) {
540 if (!keysv) {
740cce10 541 keysv = newSVpvn_utf8(key, klen, TRUE);
b2c64049
NC
542 }
543 if (PL_tainting)
544 PL_tainted = SvTAINTED(keysv);
545 keysv = sv_2mortal(newSVsv(keysv));
ad64d0ec 546 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
b2c64049 547 } else {
ad64d0ec 548 mg_copy(MUTABLE_SV(hv), val, key, klen);
b2c64049
NC
549 }
550
551 TAINT_IF(save_taint);
1baaf5d7 552 if (!needs_store) {
b2c64049
NC
553 if (flags & HVhek_FREEKEY)
554 Safefree(key);
4608196e 555 return NULL;
b2c64049
NC
556 }
557#ifdef ENV_IS_CASELESS
ad64d0ec 558 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
b2c64049
NC
559 /* XXX This code isn't UTF8 clean. */
560 const char *keysave = key;
561 /* Will need to free this, so set FREEKEY flag. */
562 key = savepvn(key,klen);
563 key = (const char*)strupr((char*)key);
6136c704 564 is_utf8 = FALSE;
b2c64049 565 hash = 0;
8b4f7dd5 566 keysv = 0;
b2c64049
NC
567
568 if (flags & HVhek_FREEKEY) {
569 Safefree(keysave);
570 }
571 flags |= HVhek_FREEKEY;
572 }
573#endif
574 }
575 } /* ISSTORE */
7f66fda2 576 } /* SvMAGICAL */
fde52b5c 577
7b2c381c 578 if (!HvARRAY(hv)) {
b2c64049 579 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
fde52b5c 580#ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
ad64d0ec
NC
581 || (SvRMAGICAL((const SV *)hv)
582 && mg_find((const SV *)hv, PERL_MAGIC_env))
fde52b5c 583#endif
d58e6666
NC
584 ) {
585 char *array;
a02a5408 586 Newxz(array,
cbec9347 587 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
d58e6666
NC
588 char);
589 HvARRAY(hv) = (HE**)array;
590 }
7f66fda2
NC
591#ifdef DYNAMIC_ENV_FETCH
592 else if (action & HV_FETCH_ISEXISTS) {
593 /* for an %ENV exists, if we do an insert it's by a recursive
594 store call, so avoid creating HvARRAY(hv) right now. */
595 }
596#endif
113738bb
NC
597 else {
598 /* XXX remove at some point? */
599 if (flags & HVhek_FREEKEY)
600 Safefree(key);
601
3c84c864 602 return NULL;
113738bb 603 }
fde52b5c 604 }
605
44b87b50 606 if (is_utf8 & !(flags & HVhek_KEYCANONICAL)) {
41d88b63 607 char * const keysave = (char *)key;
f9a63242 608 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
19692e8d 609 if (is_utf8)
c1fe5510
NC
610 flags |= HVhek_UTF8;
611 else
612 flags &= ~HVhek_UTF8;
7f66fda2
NC
613 if (key != keysave) {
614 if (flags & HVhek_FREEKEY)
615 Safefree(keysave);
19692e8d 616 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
527df579
NC
617 /* If the caller calculated a hash, it was on the sequence of
618 octets that are the UTF-8 form. We've now changed the sequence
619 of octets stored to that of the equivalent byte representation,
620 so the hash we need is different. */
621 hash = 0;
7f66fda2 622 }
19692e8d 623 }
f9a63242 624
4b5190b5
NC
625 if (HvREHASH(hv)) {
626 PERL_HASH_INTERNAL(hash, key, klen);
b2c64049
NC
627 /* We don't have a pointer to the hv, so we have to replicate the
628 flag into every HEK, so that hv_iterkeysv can see it. */
629 /* And yes, you do need this even though you are not "storing" because
fdcd69b6
NC
630 you can flip the flags below if doing an lval lookup. (And that
631 was put in to give the semantics Andreas was expecting.) */
632 flags |= HVhek_REHASH;
4b5190b5 633 } else if (!hash) {
113738bb 634 if (keysv && (SvIsCOW_shared_hash(keysv))) {
c158a4fd 635 hash = SvSHARED_HASH(keysv);
46187eeb
NC
636 } else {
637 PERL_HASH(hash, key, klen);
638 }
639 }
effa1e2d 640
113738bb
NC
641 masked_flags = (flags & HVhek_MASK);
642
7f66fda2 643#ifdef DYNAMIC_ENV_FETCH
4608196e 644 if (!HvARRAY(hv)) entry = NULL;
7f66fda2
NC
645 else
646#endif
b2c64049 647 {
7b2c381c 648 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
b2c64049 649 }
0298d7b9 650 for (; entry; entry = HeNEXT(entry)) {
fde52b5c 651 if (HeHASH(entry) != hash) /* strings can't be equal */
652 continue;
eb160463 653 if (HeKLEN(entry) != (I32)klen)
fde52b5c 654 continue;
1c846c1f 655 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
fde52b5c 656 continue;
113738bb 657 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
c3654f1a 658 continue;
b2c64049
NC
659
660 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
661 if (HeKFLAGS(entry) != masked_flags) {
662 /* We match if HVhek_UTF8 bit in our flags and hash key's
663 match. But if entry was set previously with HVhek_WASUTF8
664 and key now doesn't (or vice versa) then we should change
665 the key's flag, as this is assignment. */
666 if (HvSHAREKEYS(hv)) {
667 /* Need to swap the key we have for a key with the flags we
668 need. As keys are shared we can't just write to the
669 flag, so we share the new one, unshare the old one. */
6136c704 670 HEK * const new_hek = share_hek_flags(key, klen, hash,
6e838c70 671 masked_flags);
b2c64049
NC
672 unshare_hek (HeKEY_hek(entry));
673 HeKEY_hek(entry) = new_hek;
674 }
5d2b1485
NC
675 else if (hv == PL_strtab) {
676 /* PL_strtab is usually the only hash without HvSHAREKEYS,
677 so putting this test here is cheap */
678 if (flags & HVhek_FREEKEY)
679 Safefree(key);
680 Perl_croak(aTHX_ S_strtab_error,
681 action & HV_FETCH_LVALUE ? "fetch" : "store");
682 }
b2c64049
NC
683 else
684 HeKFLAGS(entry) = masked_flags;
685 if (masked_flags & HVhek_ENABLEHVKFLAGS)
686 HvHASKFLAGS_on(hv);
687 }
688 if (HeVAL(entry) == &PL_sv_placeholder) {
689 /* yes, can store into placeholder slot */
690 if (action & HV_FETCH_LVALUE) {
691 if (SvMAGICAL(hv)) {
692 /* This preserves behaviour with the old hv_fetch
693 implementation which at this point would bail out
694 with a break; (at "if we find a placeholder, we
695 pretend we haven't found anything")
696
697 That break mean that if a placeholder were found, it
698 caused a call into hv_store, which in turn would
699 check magic, and if there is no magic end up pretty
700 much back at this point (in hv_store's code). */
701 break;
702 }
703 /* LVAL fetch which actaully needs a store. */
561b68a9 704 val = newSV(0);
ca732855 705 HvPLACEHOLDERS(hv)--;
b2c64049
NC
706 } else {
707 /* store */
708 if (val != &PL_sv_placeholder)
ca732855 709 HvPLACEHOLDERS(hv)--;
b2c64049
NC
710 }
711 HeVAL(entry) = val;
712 } else if (action & HV_FETCH_ISSTORE) {
713 SvREFCNT_dec(HeVAL(entry));
714 HeVAL(entry) = val;
715 }
27bcc0a7 716 } else if (HeVAL(entry) == &PL_sv_placeholder) {
b2c64049
NC
717 /* if we find a placeholder, we pretend we haven't found
718 anything */
8aacddc1 719 break;
b2c64049 720 }
113738bb
NC
721 if (flags & HVhek_FREEKEY)
722 Safefree(key);
3c84c864
NC
723 if (return_svp) {
724 return entry ? (void *) &HeVAL(entry) : NULL;
725 }
fde52b5c 726 return entry;
727 }
728#ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
0ed29950 729 if (!(action & HV_FETCH_ISSTORE)
ad64d0ec
NC
730 && SvRMAGICAL((const SV *)hv)
731 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
a6c40364 732 unsigned long len;
9d4ba2ae 733 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
a6c40364
GS
734 if (env) {
735 sv = newSVpvn(env,len);
736 SvTAINTED_on(sv);
d3ba3f5c 737 return hv_common(hv, keysv, key, klen, flags,
3c84c864
NC
738 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
739 sv, hash);
a6c40364 740 }
fde52b5c 741 }
742#endif
7f66fda2
NC
743
744 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
c445ea15 745 hv_notallowed(flags, key, klen,
c8cd6465
NC
746 "Attempt to access disallowed key '%"SVf"' in"
747 " a restricted hash");
1b1f1335 748 }
b2c64049
NC
749 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
750 /* Not doing some form of store, so return failure. */
751 if (flags & HVhek_FREEKEY)
752 Safefree(key);
3c84c864 753 return NULL;
b2c64049 754 }
113738bb 755 if (action & HV_FETCH_LVALUE) {
561b68a9 756 val = newSV(0);
b2c64049
NC
757 if (SvMAGICAL(hv)) {
758 /* At this point the old hv_fetch code would call to hv_store,
759 which in turn might do some tied magic. So we need to make that
760 magic check happen. */
761 /* gonna assign to this, so it better be there */
fda2d18a
NC
762 /* If a fetch-as-store fails on the fetch, then the action is to
763 recurse once into "hv_store". If we didn't do this, then that
764 recursive call would call the key conversion routine again.
765 However, as we replace the original key with the converted
766 key, this would result in a double conversion, which would show
767 up as a bug if the conversion routine is not idempotent. */
d3ba3f5c 768 return hv_common(hv, keysv, key, klen, flags,
3c84c864
NC
769 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
770 val, hash);
b2c64049
NC
771 /* XXX Surely that could leak if the fetch-was-store fails?
772 Just like the hv_fetch. */
113738bb
NC
773 }
774 }
775
b2c64049
NC
776 /* Welcome to hv_store... */
777
7b2c381c 778 if (!HvARRAY(hv)) {
b2c64049
NC
779 /* Not sure if we can get here. I think the only case of oentry being
780 NULL is for %ENV with dynamic env fetch. But that should disappear
781 with magic in the previous code. */
d58e6666 782 char *array;
a02a5408 783 Newxz(array,
b2c64049 784 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
d58e6666
NC
785 char);
786 HvARRAY(hv) = (HE**)array;
b2c64049
NC
787 }
788
7b2c381c 789 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
ab4af705 790
b2c64049
NC
791 entry = new_HE();
792 /* share_hek_flags will do the free for us. This might be considered
793 bad API design. */
794 if (HvSHAREKEYS(hv))
6e838c70 795 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
5d2b1485
NC
796 else if (hv == PL_strtab) {
797 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
798 this test here is cheap */
799 if (flags & HVhek_FREEKEY)
800 Safefree(key);
801 Perl_croak(aTHX_ S_strtab_error,
802 action & HV_FETCH_LVALUE ? "fetch" : "store");
803 }
b2c64049
NC
804 else /* gotta do the real thing */
805 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
806 HeVAL(entry) = val;
807 HeNEXT(entry) = *oentry;
808 *oentry = entry;
809
810 if (val == &PL_sv_placeholder)
ca732855 811 HvPLACEHOLDERS(hv)++;
b2c64049
NC
812 if (masked_flags & HVhek_ENABLEHVKFLAGS)
813 HvHASKFLAGS_on(hv);
814
0298d7b9
NC
815 {
816 const HE *counter = HeNEXT(entry);
817
4c7185a0 818 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
0298d7b9
NC
819 if (!counter) { /* initial entry? */
820 xhv->xhv_fill++; /* HvFILL(hv)++ */
821 } else if (xhv->xhv_keys > (IV)xhv->xhv_max) {
822 hsplit(hv);
823 } else if(!HvREHASH(hv)) {
824 U32 n_links = 1;
825
826 while ((counter = HeNEXT(counter)))
827 n_links++;
828
829 if (n_links > HV_MAX_LENGTH_BEFORE_SPLIT) {
830 /* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit
831 bucket splits on a rehashed hash, as we're not going to
832 split it again, and if someone is lucky (evil) enough to
833 get all the keys in one list they could exhaust our memory
834 as we repeatedly double the number of buckets on every
835 entry. Linear search feels a less worse thing to do. */
836 hsplit(hv);
837 }
838 }
fde52b5c 839 }
b2c64049 840
3c84c864
NC
841 if (return_svp) {
842 return entry ? (void *) &HeVAL(entry) : NULL;
843 }
844 return (void *) entry;
fde52b5c 845}
846
864dbfa3 847STATIC void
b0e6ae5b 848S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
d0066dc7 849{
a3b680e6 850 const MAGIC *mg = SvMAGIC(hv);
7918f24d
NC
851
852 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
853
d0066dc7
OT
854 *needs_copy = FALSE;
855 *needs_store = TRUE;
856 while (mg) {
857 if (isUPPER(mg->mg_type)) {
858 *needs_copy = TRUE;
d60c5a05 859 if (mg->mg_type == PERL_MAGIC_tied) {
d0066dc7 860 *needs_store = FALSE;
4ab2a30b 861 return; /* We've set all there is to set. */
d0066dc7
OT
862 }
863 }
864 mg = mg->mg_moremagic;
865 }
866}
867
954c1994 868/*
a3bcc51e
TP
869=for apidoc hv_scalar
870
871Evaluates the hash in scalar context and returns the result. Handles magic when the hash is tied.
872
873=cut
874*/
875
876SV *
877Perl_hv_scalar(pTHX_ HV *hv)
878{
a3bcc51e 879 SV *sv;
823a54a3 880
7918f24d
NC
881 PERL_ARGS_ASSERT_HV_SCALAR;
882
823a54a3 883 if (SvRMAGICAL(hv)) {
ad64d0ec 884 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
823a54a3
AL
885 if (mg)
886 return magic_scalarpack(hv, mg);
887 }
a3bcc51e
TP
888
889 sv = sv_newmortal();
85fbaab2 890 if (HvFILL((const HV *)hv))
a3bcc51e
TP
891 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
892 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
893 else
894 sv_setiv(sv, 0);
895
896 return sv;
897}
898
899/*
954c1994
GS
900=for apidoc hv_delete
901
902Deletes a key/value pair in the hash. The value SV is removed from the
1c846c1f 903hash and returned to the caller. The C<klen> is the length of the key.
954c1994
GS
904The C<flags> value will normally be zero; if set to G_DISCARD then NULL
905will be returned.
906
954c1994
GS
907=for apidoc hv_delete_ent
908
909Deletes a key/value pair in the hash. The value SV is removed from the
910hash and returned to the caller. The C<flags> value will normally be zero;
911if set to G_DISCARD then NULL will be returned. C<hash> can be a valid
912precomputed hash value, or 0 to ask for it to be computed.
913
914=cut
915*/
916
8f8d40ab 917STATIC SV *
cd6d36ac
NC
918S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
919 int k_flags, I32 d_flags, U32 hash)
f1317c8d 920{
27da23d5 921 dVAR;
cbec9347 922 register XPVHV* xhv;
fde52b5c 923 register HE *entry;
924 register HE **oentry;
9e720f71 925 HE *const *first_entry;
9dbc5603 926 bool is_utf8 = (k_flags & HVhek_UTF8) ? TRUE : FALSE;
7a9669ca 927 int masked_flags;
1c846c1f 928
fde52b5c 929 if (SvRMAGICAL(hv)) {
0a0bb7c7
OT
930 bool needs_copy;
931 bool needs_store;
932 hv_magic_check (hv, &needs_copy, &needs_store);
933
f1317c8d 934 if (needs_copy) {
6136c704 935 SV *sv;
63c89345
NC
936 entry = (HE *) hv_common(hv, keysv, key, klen,
937 k_flags & ~HVhek_FREEKEY,
938 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
939 NULL, hash);
7a9669ca 940 sv = entry ? HeVAL(entry) : NULL;
f1317c8d
NC
941 if (sv) {
942 if (SvMAGICAL(sv)) {
943 mg_clear(sv);
944 }
945 if (!needs_store) {
946 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
947 /* No longer an element */
948 sv_unmagic(sv, PERL_MAGIC_tiedelem);
949 return sv;
950 }
a0714e2c 951 return NULL; /* element cannot be deleted */
f1317c8d 952 }
902173a3 953#ifdef ENV_IS_CASELESS
ad64d0ec 954 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
8167a60a 955 /* XXX This code isn't UTF8 clean. */
59cd0e26 956 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
8167a60a
NC
957 if (k_flags & HVhek_FREEKEY) {
958 Safefree(key);
959 }
960 key = strupr(SvPVX(keysv));
961 is_utf8 = 0;
962 k_flags = 0;
963 hash = 0;
7f66fda2 964 }
510ac311 965#endif
2fd1c6b8 966 }
2fd1c6b8 967 }
fde52b5c 968 }
cbec9347 969 xhv = (XPVHV*)SvANY(hv);
7b2c381c 970 if (!HvARRAY(hv))
a0714e2c 971 return NULL;
fde52b5c 972
19692e8d 973 if (is_utf8) {
c445ea15 974 const char * const keysave = key;
b464bac0 975 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
cd6d36ac 976
19692e8d 977 if (is_utf8)
cd6d36ac
NC
978 k_flags |= HVhek_UTF8;
979 else
980 k_flags &= ~HVhek_UTF8;
7f66fda2
NC
981 if (key != keysave) {
982 if (k_flags & HVhek_FREEKEY) {
983 /* This shouldn't happen if our caller does what we expect,
984 but strictly the API allows it. */
985 Safefree(keysave);
986 }
987 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
988 }
ad64d0ec 989 HvHASKFLAGS_on(MUTABLE_SV(hv));
19692e8d 990 }
f9a63242 991
4b5190b5
NC
992 if (HvREHASH(hv)) {
993 PERL_HASH_INTERNAL(hash, key, klen);
994 } else if (!hash) {
7a9669ca 995 if (keysv && (SvIsCOW_shared_hash(keysv))) {
c158a4fd 996 hash = SvSHARED_HASH(keysv);
7a9669ca
NC
997 } else {
998 PERL_HASH(hash, key, klen);
999 }
4b5190b5 1000 }
fde52b5c 1001
7a9669ca
NC
1002 masked_flags = (k_flags & HVhek_MASK);
1003
9e720f71 1004 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
fde52b5c 1005 entry = *oentry;
9e720f71 1006 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
6136c704 1007 SV *sv;
fde52b5c 1008 if (HeHASH(entry) != hash) /* strings can't be equal */
1009 continue;
eb160463 1010 if (HeKLEN(entry) != (I32)klen)
fde52b5c 1011 continue;
1c846c1f 1012 if (HeKEY(entry) != key && memNE(HeKEY(entry),key,klen)) /* is this it? */
fde52b5c 1013 continue;
7a9669ca 1014 if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8)
c3654f1a 1015 continue;
8aacddc1 1016
5d2b1485
NC
1017 if (hv == PL_strtab) {
1018 if (k_flags & HVhek_FREEKEY)
1019 Safefree(key);
1020 Perl_croak(aTHX_ S_strtab_error, "delete");
1021 }
1022
8aacddc1 1023 /* if placeholder is here, it's already been deleted.... */
6136c704
AL
1024 if (HeVAL(entry) == &PL_sv_placeholder) {
1025 if (k_flags & HVhek_FREEKEY)
1026 Safefree(key);
1027 return NULL;
8aacddc1 1028 }
6136c704 1029 if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
d4c19fe8 1030 hv_notallowed(k_flags, key, klen,
c8cd6465
NC
1031 "Attempt to delete readonly key '%"SVf"' from"
1032 " a restricted hash");
8aacddc1 1033 }
b84d0860
NC
1034 if (k_flags & HVhek_FREEKEY)
1035 Safefree(key);
8aacddc1 1036
cd6d36ac 1037 if (d_flags & G_DISCARD)
a0714e2c 1038 sv = NULL;
94f7643d 1039 else {
79d01fbf 1040 sv = sv_2mortal(HeVAL(entry));
7996736c 1041 HeVAL(entry) = &PL_sv_placeholder;
94f7643d 1042 }
8aacddc1
NIS
1043
1044 /*
1045 * If a restricted hash, rather than really deleting the entry, put
1046 * a placeholder there. This marks the key as being "approved", so
1047 * we can still access via not-really-existing key without raising
1048 * an error.
1049 */
1050 if (SvREADONLY(hv)) {
754604c4 1051 SvREFCNT_dec(HeVAL(entry));
7996736c 1052 HeVAL(entry) = &PL_sv_placeholder;
8aacddc1
NIS
1053 /* We'll be saving this slot, so the number of allocated keys
1054 * doesn't go down, but the number placeholders goes up */
ca732855 1055 HvPLACEHOLDERS(hv)++;
8aacddc1 1056 } else {
a26e96df 1057 *oentry = HeNEXT(entry);
9e720f71 1058 if(!*first_entry) {
a26e96df 1059 xhv->xhv_fill--; /* HvFILL(hv)-- */
9e720f71 1060 }
b79f7545 1061 if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */)
8aacddc1
NIS
1062 HvLAZYDEL_on(hv);
1063 else
1064 hv_free_ent(hv, entry);
4c7185a0 1065 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
574c8022 1066 if (xhv->xhv_keys == 0)
19692e8d 1067 HvHASKFLAGS_off(hv);
8aacddc1 1068 }
79072805
LW
1069 return sv;
1070 }
8aacddc1 1071 if (SvREADONLY(hv)) {
d4c19fe8 1072 hv_notallowed(k_flags, key, klen,
c8cd6465
NC
1073 "Attempt to delete disallowed key '%"SVf"' from"
1074 " a restricted hash");
8aacddc1
NIS
1075 }
1076
19692e8d 1077 if (k_flags & HVhek_FREEKEY)
f9a63242 1078 Safefree(key);
a0714e2c 1079 return NULL;
79072805
LW
1080}
1081
76e3520e 1082STATIC void
cea2e8a9 1083S_hsplit(pTHX_ HV *hv)
79072805 1084{
97aff369 1085 dVAR;
1e05feb3 1086 register XPVHV* const xhv = (XPVHV*)SvANY(hv);
a3b680e6 1087 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
79072805
LW
1088 register I32 newsize = oldsize * 2;
1089 register I32 i;
7b2c381c 1090 char *a = (char*) HvARRAY(hv);
72311751 1091 register HE **aep;
79072805 1092 register HE **oentry;
4b5190b5
NC
1093 int longest_chain = 0;
1094 int was_shared;
79072805 1095
7918f24d
NC
1096 PERL_ARGS_ASSERT_HSPLIT;
1097
18026298 1098 /*PerlIO_printf(PerlIO_stderr(), "hsplit called for %p which had %d\n",
6c9570dc 1099 (void*)hv, (int) oldsize);*/
18026298 1100
5d88ecd7 1101 if (HvPLACEHOLDERS_get(hv) && !SvREADONLY(hv)) {
18026298
NC
1102 /* Can make this clear any placeholders first for non-restricted hashes,
1103 even though Storable rebuilds restricted hashes by putting in all the
1104 placeholders (first) before turning on the readonly flag, because
1105 Storable always pre-splits the hash. */
1106 hv_clear_placeholders(hv);
1107 }
1108
3280af22 1109 PL_nomemok = TRUE;
8d6dde3e 1110#if defined(STRANGE_MALLOC) || defined(MYMALLOC)
b79f7545
NC
1111 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1112 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
422a93e5 1113 if (!a) {
4a33f861 1114 PL_nomemok = FALSE;
422a93e5
GA
1115 return;
1116 }
b79f7545 1117 if (SvOOK(hv)) {
31f0e52e 1118 Move(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
b79f7545 1119 }
4633a7c4 1120#else
a02a5408 1121 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
b79f7545 1122 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
422a93e5 1123 if (!a) {
3280af22 1124 PL_nomemok = FALSE;
422a93e5
GA
1125 return;
1126 }
7b2c381c 1127 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
b79f7545
NC
1128 if (SvOOK(hv)) {
1129 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1130 }
fba3b22e 1131 if (oldsize >= 64) {
7b2c381c 1132 offer_nice_chunk(HvARRAY(hv),
b79f7545
NC
1133 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1134 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
4633a7c4
LW
1135 }
1136 else
7b2c381c 1137 Safefree(HvARRAY(hv));
4633a7c4
LW
1138#endif
1139
3280af22 1140 PL_nomemok = FALSE;
72311751 1141 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
cbec9347 1142 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
7b2c381c 1143 HvARRAY(hv) = (HE**) a;
72311751 1144 aep = (HE**)a;
79072805 1145
72311751 1146 for (i=0; i<oldsize; i++,aep++) {
4b5190b5
NC
1147 int left_length = 0;
1148 int right_length = 0;
a3b680e6
AL
1149 register HE *entry;
1150 register HE **bep;
4b5190b5 1151
72311751 1152 if (!*aep) /* non-existent */
79072805 1153 continue;
72311751
GS
1154 bep = aep+oldsize;
1155 for (oentry = aep, entry = *aep; entry; entry = *oentry) {
eb160463 1156 if ((HeHASH(entry) & newsize) != (U32)i) {
fde52b5c 1157 *oentry = HeNEXT(entry);
72311751
GS
1158 HeNEXT(entry) = *bep;
1159 if (!*bep)
cbec9347 1160 xhv->xhv_fill++; /* HvFILL(hv)++ */
72311751 1161 *bep = entry;
4b5190b5 1162 right_length++;
79072805
LW
1163 continue;
1164 }
4b5190b5 1165 else {
fde52b5c 1166 oentry = &HeNEXT(entry);
4b5190b5
NC
1167 left_length++;
1168 }
79072805 1169 }
72311751 1170 if (!*aep) /* everything moved */
cbec9347 1171 xhv->xhv_fill--; /* HvFILL(hv)-- */
4b5190b5
NC
1172 /* I think we don't actually need to keep track of the longest length,
1173 merely flag if anything is too long. But for the moment while
1174 developing this code I'll track it. */
1175 if (left_length > longest_chain)
1176 longest_chain = left_length;
1177 if (right_length > longest_chain)
1178 longest_chain = right_length;
1179 }
1180
1181
1182 /* Pick your policy for "hashing isn't working" here: */
fdcd69b6 1183 if (longest_chain <= HV_MAX_LENGTH_BEFORE_SPLIT /* split worked? */
4b5190b5
NC
1184 || HvREHASH(hv)) {
1185 return;
79072805 1186 }
4b5190b5
NC
1187
1188 if (hv == PL_strtab) {
1189 /* Urg. Someone is doing something nasty to the string table.
1190 Can't win. */
1191 return;
1192 }
1193
1194 /* Awooga. Awooga. Pathological data. */
6c9570dc 1195 /*PerlIO_printf(PerlIO_stderr(), "%p %d of %d with %d/%d buckets\n", (void*)hv,
4b5190b5
NC
1196 longest_chain, HvTOTALKEYS(hv), HvFILL(hv), 1+HvMAX(hv));*/
1197
1198 ++newsize;
a02a5408 1199 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
b79f7545
NC
1200 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
1201 if (SvOOK(hv)) {
1202 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1203 }
1204
4b5190b5
NC
1205 was_shared = HvSHAREKEYS(hv);
1206
1207 xhv->xhv_fill = 0;
1208 HvSHAREKEYS_off(hv);
1209 HvREHASH_on(hv);
1210
7b2c381c 1211 aep = HvARRAY(hv);
4b5190b5
NC
1212
1213 for (i=0; i<newsize; i++,aep++) {
a3b680e6 1214 register HE *entry = *aep;
4b5190b5
NC
1215 while (entry) {
1216 /* We're going to trash this HE's next pointer when we chain it
1217 into the new hash below, so store where we go next. */
9d4ba2ae 1218 HE * const next = HeNEXT(entry);
4b5190b5 1219 UV hash;
a3b680e6 1220 HE **bep;
4b5190b5
NC
1221
1222 /* Rehash it */
1223 PERL_HASH_INTERNAL(hash, HeKEY(entry), HeKLEN(entry));
1224
1225 if (was_shared) {
1226 /* Unshare it. */
aec46f14 1227 HEK * const new_hek
4b5190b5
NC
1228 = save_hek_flags(HeKEY(entry), HeKLEN(entry),
1229 hash, HeKFLAGS(entry));
1230 unshare_hek (HeKEY_hek(entry));
1231 HeKEY_hek(entry) = new_hek;
1232 } else {
1233 /* Not shared, so simply write the new hash in. */
1234 HeHASH(entry) = hash;
1235 }
1236 /*PerlIO_printf(PerlIO_stderr(), "%d ", HeKFLAGS(entry));*/
1237 HEK_REHASH_on(HeKEY_hek(entry));
1238 /*PerlIO_printf(PerlIO_stderr(), "%d\n", HeKFLAGS(entry));*/
1239
1240 /* Copy oentry to the correct new chain. */
1241 bep = ((HE**)a) + (hash & (I32) xhv->xhv_max);
1242 if (!*bep)
1243 xhv->xhv_fill++; /* HvFILL(hv)++ */
1244 HeNEXT(entry) = *bep;
1245 *bep = entry;
1246
1247 entry = next;
1248 }
1249 }
7b2c381c
NC
1250 Safefree (HvARRAY(hv));
1251 HvARRAY(hv) = (HE **)a;
79072805
LW
1252}
1253
72940dca 1254void
864dbfa3 1255Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
72940dca 1256{
97aff369 1257 dVAR;
cbec9347 1258 register XPVHV* xhv = (XPVHV*)SvANY(hv);
a3b680e6 1259 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */
72940dca 1260 register I32 newsize;
1261 register I32 i;
72311751
GS
1262 register char *a;
1263 register HE **aep;
72940dca 1264 register HE *entry;
1265 register HE **oentry;
1266
7918f24d
NC
1267 PERL_ARGS_ASSERT_HV_KSPLIT;
1268
72940dca 1269 newsize = (I32) newmax; /* possible truncation here */
1270 if (newsize != newmax || newmax <= oldsize)
1271 return;
1272 while ((newsize & (1 + ~newsize)) != newsize) {
1273 newsize &= ~(newsize & (1 + ~newsize)); /* get proper power of 2 */
1274 }
1275 if (newsize < newmax)
1276 newsize *= 2;
1277 if (newsize < newmax)
1278 return; /* overflow detection */
1279
7b2c381c 1280 a = (char *) HvARRAY(hv);
72940dca 1281 if (a) {
3280af22 1282 PL_nomemok = TRUE;
8d6dde3e 1283#if defined(STRANGE_MALLOC) || defined(MYMALLOC)
b79f7545
NC
1284 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
1285 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
8aacddc1 1286 if (!a) {
4a33f861 1287 PL_nomemok = FALSE;
422a93e5
GA
1288 return;
1289 }
b79f7545 1290 if (SvOOK(hv)) {
7a9b70e9 1291 Copy(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
b79f7545 1292 }
72940dca 1293#else
a02a5408 1294 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize)
b79f7545 1295 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char);
8aacddc1 1296 if (!a) {
3280af22 1297 PL_nomemok = FALSE;
422a93e5
GA
1298 return;
1299 }
7b2c381c 1300 Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char);
b79f7545
NC
1301 if (SvOOK(hv)) {
1302 Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux);
1303 }
fba3b22e 1304 if (oldsize >= 64) {
7b2c381c 1305 offer_nice_chunk(HvARRAY(hv),
b79f7545
NC
1306 PERL_HV_ARRAY_ALLOC_BYTES(oldsize)
1307 + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0));
72940dca 1308 }
1309 else
7b2c381c 1310 Safefree(HvARRAY(hv));
72940dca 1311#endif
3280af22 1312 PL_nomemok = FALSE;
72311751 1313 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
72940dca 1314 }
1315 else {
a02a5408 1316 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
72940dca 1317 }
cbec9347 1318 xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */
7b2c381c 1319 HvARRAY(hv) = (HE **) a;
cbec9347 1320 if (!xhv->xhv_fill /* !HvFILL(hv) */) /* skip rest if no entries */
72940dca 1321 return;
1322
72311751
GS
1323 aep = (HE**)a;
1324 for (i=0; i<oldsize; i++,aep++) {
1325 if (!*aep) /* non-existent */
72940dca 1326 continue;
72311751 1327 for (oentry = aep, entry = *aep; entry; entry = *oentry) {
6136c704
AL
1328 register I32 j = (HeHASH(entry) & newsize);
1329
1330 if (j != i) {
72940dca 1331 j -= i;
1332 *oentry = HeNEXT(entry);
72311751 1333 if (!(HeNEXT(entry) = aep[j]))
cbec9347 1334 xhv->xhv_fill++; /* HvFILL(hv)++ */
72311751 1335 aep[j] = entry;
72940dca 1336 continue;
1337 }
1338 else
1339 oentry = &HeNEXT(entry);
1340 }
72311751 1341 if (!*aep) /* everything moved */
cbec9347 1342 xhv->xhv_fill--; /* HvFILL(hv)-- */
72940dca 1343 }
1344}
1345
b3ac6de7 1346HV *
864dbfa3 1347Perl_newHVhv(pTHX_ HV *ohv)
b3ac6de7 1348{
9d4ba2ae 1349 HV * const hv = newHV();
4beac62f 1350 STRLEN hv_max, hv_fill;
4beac62f
AMS
1351
1352 if (!ohv || (hv_fill = HvFILL(ohv)) == 0)
1353 return hv;
4beac62f 1354 hv_max = HvMAX(ohv);
b3ac6de7 1355
ad64d0ec 1356 if (!SvMAGICAL((const SV *)ohv)) {
b56ba0bf 1357 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
eb160463 1358 STRLEN i;
a3b680e6 1359 const bool shared = !!HvSHAREKEYS(ohv);
aec46f14 1360 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
ff875642 1361 char *a;
a02a5408 1362 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
ff875642 1363 ents = (HE**)a;
b56ba0bf
AMS
1364
1365 /* In each bucket... */
1366 for (i = 0; i <= hv_max; i++) {
6136c704 1367 HE *prev = NULL;
aec46f14 1368 HE *oent = oents[i];
b56ba0bf
AMS
1369
1370 if (!oent) {
1371 ents[i] = NULL;
1372 continue;
1373 }
1374
1375 /* Copy the linked list of entries. */
aec46f14 1376 for (; oent; oent = HeNEXT(oent)) {
a3b680e6
AL
1377 const U32 hash = HeHASH(oent);
1378 const char * const key = HeKEY(oent);
1379 const STRLEN len = HeKLEN(oent);
1380 const int flags = HeKFLAGS(oent);
6136c704 1381 HE * const ent = new_HE();
b56ba0bf 1382
45dea987 1383 HeVAL(ent) = newSVsv(HeVAL(oent));
19692e8d 1384 HeKEY_hek(ent)
6e838c70 1385 = shared ? share_hek_flags(key, len, hash, flags)
19692e8d 1386 : save_hek_flags(key, len, hash, flags);
b56ba0bf
AMS
1387 if (prev)
1388 HeNEXT(prev) = ent;
1389 else
1390 ents[i] = ent;
1391 prev = ent;
1392 HeNEXT(ent) = NULL;
1393 }
1394 }
1395
1396 HvMAX(hv) = hv_max;
1397 HvFILL(hv) = hv_fill;
8aacddc1 1398 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
b56ba0bf 1399 HvARRAY(hv) = ents;
aec46f14 1400 } /* not magical */
b56ba0bf
AMS
1401 else {
1402 /* Iterate over ohv, copying keys and values one at a time. */
b3ac6de7 1403 HE *entry;
bfcb3514
NC
1404 const I32 riter = HvRITER_get(ohv);
1405 HE * const eiter = HvEITER_get(ohv);
b56ba0bf
AMS
1406
1407 /* Can we use fewer buckets? (hv_max is always 2^n-1) */
1408 while (hv_max && hv_max + 1 >= hv_fill * 2)
1409 hv_max = hv_max / 2;
1410 HvMAX(hv) = hv_max;
1411
4a76a316 1412 hv_iterinit(ohv);
e16e2ff8 1413 while ((entry = hv_iternext_flags(ohv, 0))) {
04fe65b0
RGS
1414 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1415 newSVsv(HeVAL(entry)), HeHASH(entry),
1416 HeKFLAGS(entry));
b3ac6de7 1417 }
bfcb3514
NC
1418 HvRITER_set(ohv, riter);
1419 HvEITER_set(ohv, eiter);
b3ac6de7 1420 }
1c846c1f 1421
b3ac6de7
IZ
1422 return hv;
1423}
1424
5b9c0671
NC
1425/* A rather specialised version of newHVhv for copying %^H, ensuring all the
1426 magic stays on it. */
1427HV *
1428Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1429{
1430 HV * const hv = newHV();
1431 STRLEN hv_fill;
1432
1433 if (ohv && (hv_fill = HvFILL(ohv))) {
1434 STRLEN hv_max = HvMAX(ohv);
1435 HE *entry;
1436 const I32 riter = HvRITER_get(ohv);
1437 HE * const eiter = HvEITER_get(ohv);
1438
1439 while (hv_max && hv_max + 1 >= hv_fill * 2)
1440 hv_max = hv_max / 2;
1441 HvMAX(hv) = hv_max;
1442
1443 hv_iterinit(ohv);
1444 while ((entry = hv_iternext_flags(ohv, 0))) {
1445 SV *const sv = newSVsv(HeVAL(entry));
e3b1b6b1 1446 SV *heksv = newSVhek(HeKEY_hek(entry));
5b9c0671 1447 sv_magic(sv, NULL, PERL_MAGIC_hintselem,
e3b1b6b1
RGS
1448 (char *)heksv, HEf_SVKEY);
1449 SvREFCNT_dec(heksv);
04fe65b0
RGS
1450 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry),
1451 sv, HeHASH(entry), HeKFLAGS(entry));
5b9c0671
NC
1452 }
1453 HvRITER_set(ohv, riter);
1454 HvEITER_set(ohv, eiter);
1455 }
1456 hv_magic(hv, NULL, PERL_MAGIC_hints);
1457 return hv;
1458}
1459
79072805 1460void
864dbfa3 1461Perl_hv_free_ent(pTHX_ HV *hv, register HE *entry)
79072805 1462{
97aff369 1463 dVAR;
16bdeea2
GS
1464 SV *val;
1465
7918f24d
NC
1466 PERL_ARGS_ASSERT_HV_FREE_ENT;
1467
68dc0745 1468 if (!entry)
79072805 1469 return;
16bdeea2 1470 val = HeVAL(entry);
f1c32fec
BM
1471 if (HvNAME(hv) && anonymise_cv(HvNAME(hv), val) && GvCVu(val))
1472 mro_method_changed_in(hv);
16bdeea2 1473 SvREFCNT_dec(val);
68dc0745 1474 if (HeKLEN(entry) == HEf_SVKEY) {
1475 SvREFCNT_dec(HeKEY_sv(entry));
8aacddc1 1476 Safefree(HeKEY_hek(entry));
44a8e56a 1477 }
1478 else if (HvSHAREKEYS(hv))
68dc0745 1479 unshare_hek(HeKEY_hek(entry));
fde52b5c 1480 else
68dc0745 1481 Safefree(HeKEY_hek(entry));
d33b2eba 1482 del_HE(entry);
79072805
LW
1483}
1484
f1c32fec 1485static I32
4fec3216 1486S_anonymise_cv(pTHX_ const char *stash, SV *val)
f1c32fec
BM
1487{
1488 CV *cv;
1489
1490 PERL_ARGS_ASSERT_ANONYMISE_CV;
1491
1492 if (val && isGV(val) && isGV_with_GP(val) && (cv = GvCV(val))) {
1493 if ((SV *)CvGV(cv) == val) {
1494 SV *gvname;
1495 GV *anongv;
1496
4fec3216 1497 gvname = Perl_newSVpvf(aTHX_ "%s::__ANON__", stash ? stash : "__ANON__");
f1c32fec
BM
1498 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
1499 SvREFCNT_dec(gvname);
1500 CvGV(cv) = anongv;
1501 CvANON_on(cv);
1502 return 1;
1503 }
1504 }
1505 return 0;
1506}
1507
79072805 1508void
864dbfa3 1509Perl_hv_delayfree_ent(pTHX_ HV *hv, register HE *entry)
79072805 1510{
97aff369 1511 dVAR;
7918f24d
NC
1512
1513 PERL_ARGS_ASSERT_HV_DELAYFREE_ENT;
1514
68dc0745 1515 if (!entry)
79072805 1516 return;
bc4947fc
NC
1517 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1518 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
68dc0745 1519 if (HeKLEN(entry) == HEf_SVKEY) {
bc4947fc 1520 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
44a8e56a 1521 }
bc4947fc 1522 hv_free_ent(hv, entry);
79072805
LW
1523}
1524
954c1994
GS
1525/*
1526=for apidoc hv_clear
1527
1528Clears a hash, making it empty.
1529
1530=cut
1531*/
1532
79072805 1533void
864dbfa3 1534Perl_hv_clear(pTHX_ HV *hv)
79072805 1535{
27da23d5 1536 dVAR;
cbec9347 1537 register XPVHV* xhv;
79072805
LW
1538 if (!hv)
1539 return;
49293501 1540
ecae49c0
NC
1541 DEBUG_A(Perl_hv_assert(aTHX_ hv));
1542
34c3c4e3
DM
1543 xhv = (XPVHV*)SvANY(hv);
1544
7b2c381c 1545 if (SvREADONLY(hv) && HvARRAY(hv) != NULL) {
34c3c4e3 1546 /* restricted hash: convert all keys to placeholders */
b464bac0
AL
1547 STRLEN i;
1548 for (i = 0; i <= xhv->xhv_max; i++) {
7b2c381c 1549 HE *entry = (HvARRAY(hv))[i];
3a676441
JH
1550 for (; entry; entry = HeNEXT(entry)) {
1551 /* not already placeholder */
7996736c 1552 if (HeVAL(entry) != &PL_sv_placeholder) {
3a676441 1553 if (HeVAL(entry) && SvREADONLY(HeVAL(entry))) {
6136c704 1554 SV* const keysv = hv_iterkeysv(entry);
3a676441 1555 Perl_croak(aTHX_
95b63a38
JH
1556 "Attempt to delete readonly key '%"SVf"' from a restricted hash",
1557 (void*)keysv);
3a676441
JH
1558 }
1559 SvREFCNT_dec(HeVAL(entry));
7996736c 1560 HeVAL(entry) = &PL_sv_placeholder;
ca732855 1561 HvPLACEHOLDERS(hv)++;
3a676441 1562 }
34c3c4e3
DM
1563 }
1564 }
df8c6964 1565 goto reset;
49293501
MS
1566 }
1567
463ee0b2 1568 hfreeentries(hv);
ca732855 1569 HvPLACEHOLDERS_set(hv, 0);
7b2c381c 1570 if (HvARRAY(hv))
41f62432 1571 Zero(HvARRAY(hv), xhv->xhv_max+1 /* HvMAX(hv)+1 */, HE*);
a0d0e21e
LW
1572
1573 if (SvRMAGICAL(hv))
ad64d0ec 1574 mg_clear(MUTABLE_SV(hv));
574c8022 1575
19692e8d 1576 HvHASKFLAGS_off(hv);
bb443f97 1577 HvREHASH_off(hv);
df8c6964 1578 reset:
b79f7545 1579 if (SvOOK(hv)) {
dd69841b
BB
1580 if(HvNAME_get(hv))
1581 mro_isa_changed_in(hv);
bfcb3514
NC
1582 HvEITER_set(hv, NULL);
1583 }
79072805
LW
1584}
1585
3540d4ce
AB
1586/*
1587=for apidoc hv_clear_placeholders
1588
1589Clears any placeholders from a hash. If a restricted hash has any of its keys
1590marked as readonly and the key is subsequently deleted, the key is not actually
1591deleted but is marked by assigning it a value of &PL_sv_placeholder. This tags
1592it so it will be ignored by future operations such as iterating over the hash,
4cdaeff7 1593but will still allow the hash to have a value reassigned to the key at some
3540d4ce
AB
1594future point. This function clears any such placeholder keys from the hash.
1595See Hash::Util::lock_keys() for an example of its use.
1596
1597=cut
1598*/
1599
1600void
1601Perl_hv_clear_placeholders(pTHX_ HV *hv)
1602{
27da23d5 1603 dVAR;
b3ca2e83
NC
1604 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
1605
7918f24d
NC
1606 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
1607
b3ca2e83
NC
1608 if (items)
1609 clear_placeholders(hv, items);
1610}
1611
1612static void
1613S_clear_placeholders(pTHX_ HV *hv, U32 items)
1614{
1615 dVAR;
b464bac0 1616 I32 i;
d3677389 1617
7918f24d
NC
1618 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
1619
d3677389
NC
1620 if (items == 0)
1621 return;
1622
b464bac0 1623 i = HvMAX(hv);
d3677389
NC
1624 do {
1625 /* Loop down the linked list heads */
6136c704 1626 bool first = TRUE;
d3677389 1627 HE **oentry = &(HvARRAY(hv))[i];
cf6db12b 1628 HE *entry;
d3677389 1629
cf6db12b 1630 while ((entry = *oentry)) {
d3677389
NC
1631 if (HeVAL(entry) == &PL_sv_placeholder) {
1632 *oentry = HeNEXT(entry);
1633 if (first && !*oentry)
1634 HvFILL(hv)--; /* This linked list is now empty. */
2e58978b 1635 if (entry == HvEITER_get(hv))
d3677389
NC
1636 HvLAZYDEL_on(hv);
1637 else
1638 hv_free_ent(hv, entry);
1639
1640 if (--items == 0) {
1641 /* Finished. */
5d88ecd7 1642 HvTOTALKEYS(hv) -= (IV)HvPLACEHOLDERS_get(hv);
d3677389
NC
1643 if (HvKEYS(hv) == 0)
1644 HvHASKFLAGS_off(hv);
5d88ecd7 1645 HvPLACEHOLDERS_set(hv, 0);
d3677389
NC
1646 return;
1647 }
213ce8b3
NC
1648 } else {
1649 oentry = &HeNEXT(entry);
6136c704 1650 first = FALSE;
d3677389
NC
1651 }
1652 }
1653 } while (--i >= 0);
1654 /* You can't get here, hence assertion should always fail. */
1655 assert (items == 0);
1656 assert (0);
3540d4ce
AB
1657}
1658
76e3520e 1659STATIC void
cea2e8a9 1660S_hfreeentries(pTHX_ HV *hv)
79072805 1661{
23976bdd 1662 /* This is the array that we're going to restore */
fd7de8a8 1663 HE **const orig_array = HvARRAY(hv);
23976bdd
NC
1664 HEK *name;
1665 int attempts = 100;
3abe233e 1666
7918f24d
NC
1667 PERL_ARGS_ASSERT_HFREEENTRIES;
1668
fd7de8a8 1669 if (!orig_array)
79072805 1670 return;
a0d0e21e 1671
f1c32fec
BM
1672 if (HvNAME(hv) && orig_array != NULL) {
1673 /* symbol table: make all the contained subs ANON */
1674 STRLEN i;
1675 XPVHV *xhv = (XPVHV*)SvANY(hv);
1676
1677 for (i = 0; i <= xhv->xhv_max; i++) {
1678 HE *entry = (HvARRAY(hv))[i];
1679 for (; entry; entry = HeNEXT(entry)) {
1680 SV *val = HeVAL(entry);
1681 /* we need to put the subs in the __ANON__ symtable, as
1682 * this one is being cleared. */
1683 anonymise_cv(NULL, val);
1684 }
1685 }
1686 }
1687
23976bdd
NC
1688 if (SvOOK(hv)) {
1689 /* If the hash is actually a symbol table with a name, look after the
1690 name. */
1691 struct xpvhv_aux *iter = HvAUX(hv);
1692
1693 name = iter->xhv_name;
1694 iter->xhv_name = NULL;
1695 } else {
1696 name = NULL;
1697 }
1698
23976bdd
NC
1699 /* orig_array remains unchanged throughout the loop. If after freeing all
1700 the entries it turns out that one of the little blighters has triggered
1701 an action that has caused HvARRAY to be re-allocated, then we set
1702 array to the new HvARRAY, and try again. */
1703
1704 while (1) {
1705 /* This is the one we're going to try to empty. First time round
1706 it's the original array. (Hopefully there will only be 1 time
1707 round) */
6136c704 1708 HE ** const array = HvARRAY(hv);
7440661e 1709 I32 i = HvMAX(hv);
23976bdd
NC
1710
1711 /* Because we have taken xhv_name out, the only allocated pointer
1712 in the aux structure that might exist is the backreference array.
1713 */
1714
1715 if (SvOOK(hv)) {
7440661e 1716 HE *entry;
e1a479c5 1717 struct mro_meta *meta;
23976bdd
NC
1718 struct xpvhv_aux *iter = HvAUX(hv);
1719 /* If there are weak references to this HV, we need to avoid
1720 freeing them up here. In particular we need to keep the AV
1721 visible as what we're deleting might well have weak references
1722 back to this HV, so the for loop below may well trigger
1723 the removal of backreferences from this array. */
1724
1725 if (iter->xhv_backreferences) {
1726 /* So donate them to regular backref magic to keep them safe.
1727 The sv_magic will increase the reference count of the AV,
1728 so we need to drop it first. */
5b285ea4 1729 SvREFCNT_dec(iter->xhv_backreferences);
23976bdd
NC
1730 if (AvFILLp(iter->xhv_backreferences) == -1) {
1731 /* Turns out that the array is empty. Just free it. */
1732 SvREFCNT_dec(iter->xhv_backreferences);
1b8791d1 1733
23976bdd 1734 } else {
ad64d0ec
NC
1735 sv_magic(MUTABLE_SV(hv),
1736 MUTABLE_SV(iter->xhv_backreferences),
23976bdd
NC
1737 PERL_MAGIC_backref, NULL, 0);
1738 }
1739 iter->xhv_backreferences = NULL;
5b285ea4 1740 }
86f55936 1741
23976bdd
NC
1742 entry = iter->xhv_eiter; /* HvEITER(hv) */
1743 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1744 HvLAZYDEL_off(hv);
1745 hv_free_ent(hv, entry);
1746 }
1747 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
4608196e 1748 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
b79f7545 1749
e1a479c5 1750 if((meta = iter->xhv_mro_meta)) {
9953ff72
NC
1751 if (meta->mro_linear_all) {
1752 SvREFCNT_dec(MUTABLE_SV(meta->mro_linear_all));
1753 meta->mro_linear_all = NULL;
553e831a 1754 /* This is just acting as a shortcut pointer. */
3a6fa573
NC
1755 meta->mro_linear_current = NULL;
1756 } else if (meta->mro_linear_current) {
553e831a
NC
1757 /* Only the current MRO is stored, so this owns the data.
1758 */
3a6fa573
NC
1759 SvREFCNT_dec(meta->mro_linear_current);
1760 meta->mro_linear_current = NULL;
553e831a 1761 }
e1a479c5 1762 if(meta->mro_nextmethod) SvREFCNT_dec(meta->mro_nextmethod);
a49ba3fc 1763 SvREFCNT_dec(meta->isa);
e1a479c5
BB
1764 Safefree(meta);
1765 iter->xhv_mro_meta = NULL;
1766 }
1767
23976bdd 1768 /* There are now no allocated pointers in the aux structure. */
2f86008e 1769
23976bdd
NC
1770 SvFLAGS(hv) &= ~SVf_OOK; /* Goodbye, aux structure. */
1771 /* What aux structure? */
a0d0e21e 1772 }
bfcb3514 1773
23976bdd
NC
1774 /* make everyone else think the array is empty, so that the destructors
1775 * called for freed entries can't recusively mess with us */
1776 HvARRAY(hv) = NULL;
1777 HvFILL(hv) = 0;
1778 ((XPVHV*) SvANY(hv))->xhv_keys = 0;
1779
7440661e
NC
1780
1781 do {
1782 /* Loop down the linked list heads */
1783 HE *entry = array[i];
1784
1785 while (entry) {
23976bdd
NC
1786 register HE * const oentry = entry;
1787 entry = HeNEXT(entry);
1788 hv_free_ent(hv, oentry);
1789 }
7440661e 1790 } while (--i >= 0);
b79f7545 1791
23976bdd
NC
1792 /* As there are no allocated pointers in the aux structure, it's now
1793 safe to free the array we just cleaned up, if it's not the one we're
1794 going to put back. */
1795 if (array != orig_array) {
1796 Safefree(array);
1797 }
b79f7545 1798
23976bdd
NC
1799 if (!HvARRAY(hv)) {
1800 /* Good. No-one added anything this time round. */
1801 break;
bfcb3514 1802 }
b79f7545 1803
23976bdd
NC
1804 if (SvOOK(hv)) {
1805 /* Someone attempted to iterate or set the hash name while we had
1806 the array set to 0. We'll catch backferences on the next time
1807 round the while loop. */
1808 assert(HvARRAY(hv));
1b8791d1 1809
23976bdd
NC
1810 if (HvAUX(hv)->xhv_name) {
1811 unshare_hek_or_pvn(HvAUX(hv)->xhv_name, 0, 0, 0);
1812 }
1813 }
1814
1815 if (--attempts == 0) {
1816 Perl_die(aTHX_ "panic: hfreeentries failed to free hash - something is repeatedly re-creating entries");
1817 }
6136c704 1818 }
23976bdd
NC
1819
1820 HvARRAY(hv) = orig_array;
1821
1822 /* If the hash was actually a symbol table, put the name back. */
1823 if (name) {
1824 /* We have restored the original array. If name is non-NULL, then
1825 the original array had an aux structure at the end. So this is
1826 valid: */
1827 SvFLAGS(hv) |= SVf_OOK;
1828 HvAUX(hv)->xhv_name = name;
1b8791d1 1829 }
79072805
LW
1830}
1831
954c1994
GS
1832/*
1833=for apidoc hv_undef
1834
1835Undefines the hash.
1836
1837=cut
1838*/
1839
79072805 1840void
864dbfa3 1841Perl_hv_undef(pTHX_ HV *hv)
79072805 1842{
97aff369 1843 dVAR;
cbec9347 1844 register XPVHV* xhv;
bfcb3514 1845 const char *name;
86f55936 1846
79072805
LW
1847 if (!hv)
1848 return;
ecae49c0 1849 DEBUG_A(Perl_hv_assert(aTHX_ hv));
cbec9347 1850 xhv = (XPVHV*)SvANY(hv);
dd69841b 1851
0fa56319 1852 if ((name = HvNAME_get(hv)) && !PL_dirty)
dd69841b
BB
1853 mro_isa_changed_in(hv);
1854
463ee0b2 1855 hfreeentries(hv);
dd69841b 1856 if (name) {
04fe65b0
RGS
1857 if (PL_stashcache)
1858 (void)hv_delete(PL_stashcache, name, HvNAMELEN_get(hv), G_DISCARD);
bd61b366 1859 hv_name_set(hv, NULL, 0, 0);
85e6fe83 1860 }
b79f7545
NC
1861 SvFLAGS(hv) &= ~SVf_OOK;
1862 Safefree(HvARRAY(hv));
cbec9347 1863 xhv->xhv_max = 7; /* HvMAX(hv) = 7 (it's a normal hash) */
7b2c381c 1864 HvARRAY(hv) = 0;
ca732855 1865 HvPLACEHOLDERS_set(hv, 0);
a0d0e21e
LW
1866
1867 if (SvRMAGICAL(hv))
ad64d0ec 1868 mg_clear(MUTABLE_SV(hv));
79072805
LW
1869}
1870
b464bac0 1871static struct xpvhv_aux*
5f66b61c 1872S_hv_auxinit(HV *hv) {
bfcb3514 1873 struct xpvhv_aux *iter;
b79f7545 1874 char *array;
bfcb3514 1875
7918f24d
NC
1876 PERL_ARGS_ASSERT_HV_AUXINIT;
1877
b79f7545 1878 if (!HvARRAY(hv)) {
a02a5408 1879 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
b79f7545
NC
1880 + sizeof(struct xpvhv_aux), char);
1881 } else {
1882 array = (char *) HvARRAY(hv);
1883 Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1)
1884 + sizeof(struct xpvhv_aux), char);
1885 }
1886 HvARRAY(hv) = (HE**) array;
1887 /* SvOOK_on(hv) attacks the IV flags. */
1888 SvFLAGS(hv) |= SVf_OOK;
1889 iter = HvAUX(hv);
bfcb3514
NC
1890
1891 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
4608196e 1892 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
bfcb3514 1893 iter->xhv_name = 0;
86f55936 1894 iter->xhv_backreferences = 0;
e1a479c5 1895 iter->xhv_mro_meta = NULL;
bfcb3514
NC
1896 return iter;
1897}
1898
954c1994
GS
1899/*
1900=for apidoc hv_iterinit
1901
1902Prepares a starting point to traverse a hash table. Returns the number of
1903keys in the hash (i.e. the same as C<HvKEYS(tb)>). The return value is
1c846c1f 1904currently only meaningful for hashes without tie magic.
954c1994
GS
1905
1906NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
1907hash buckets that happen to be in use. If you still need that esoteric
1908value, you can get it through the macro C<HvFILL(tb)>.
1909
e16e2ff8 1910
954c1994
GS
1911=cut
1912*/
1913
79072805 1914I32
864dbfa3 1915Perl_hv_iterinit(pTHX_ HV *hv)
79072805 1916{
7918f24d
NC
1917 PERL_ARGS_ASSERT_HV_ITERINIT;
1918
1919 /* FIXME: Are we not NULL, or do we croak? Place bets now! */
1920
aa689395 1921 if (!hv)
cea2e8a9 1922 Perl_croak(aTHX_ "Bad hash");
bfcb3514 1923
b79f7545 1924 if (SvOOK(hv)) {
6136c704 1925 struct xpvhv_aux * const iter = HvAUX(hv);
0bd48802 1926 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
bfcb3514
NC
1927 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1928 HvLAZYDEL_off(hv);
1929 hv_free_ent(hv, entry);
1930 }
1931 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
4608196e 1932 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
bfcb3514 1933 } else {
6136c704 1934 hv_auxinit(hv);
72940dca 1935 }
44a2ac75 1936
cbec9347 1937 /* used to be xhv->xhv_fill before 5.004_65 */
5d88ecd7 1938 return HvTOTALKEYS(hv);
79072805 1939}
bfcb3514
NC
1940
1941I32 *
1942Perl_hv_riter_p(pTHX_ HV *hv) {
1943 struct xpvhv_aux *iter;
1944
7918f24d
NC
1945 PERL_ARGS_ASSERT_HV_RITER_P;
1946
bfcb3514
NC
1947 if (!hv)
1948 Perl_croak(aTHX_ "Bad hash");
1949
6136c704 1950 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
1951 return &(iter->xhv_riter);
1952}
1953
1954HE **
1955Perl_hv_eiter_p(pTHX_ HV *hv) {
1956 struct xpvhv_aux *iter;
1957
7918f24d
NC
1958 PERL_ARGS_ASSERT_HV_EITER_P;
1959
bfcb3514
NC
1960 if (!hv)
1961 Perl_croak(aTHX_ "Bad hash");
1962
6136c704 1963 iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
1964 return &(iter->xhv_eiter);
1965}
1966
1967void
1968Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
1969 struct xpvhv_aux *iter;
1970
7918f24d
NC
1971 PERL_ARGS_ASSERT_HV_RITER_SET;
1972
bfcb3514
NC
1973 if (!hv)
1974 Perl_croak(aTHX_ "Bad hash");
1975
b79f7545
NC
1976 if (SvOOK(hv)) {
1977 iter = HvAUX(hv);
1978 } else {
bfcb3514
NC
1979 if (riter == -1)
1980 return;
1981
6136c704 1982 iter = hv_auxinit(hv);
bfcb3514
NC
1983 }
1984 iter->xhv_riter = riter;
1985}
1986
1987void
1988Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
1989 struct xpvhv_aux *iter;
1990
7918f24d
NC
1991 PERL_ARGS_ASSERT_HV_EITER_SET;
1992
bfcb3514
NC
1993 if (!hv)
1994 Perl_croak(aTHX_ "Bad hash");
1995
b79f7545
NC
1996 if (SvOOK(hv)) {
1997 iter = HvAUX(hv);
1998 } else {
bfcb3514
NC
1999 /* 0 is the default so don't go malloc()ing a new structure just to
2000 hold 0. */
2001 if (!eiter)
2002 return;
2003
6136c704 2004 iter = hv_auxinit(hv);
bfcb3514
NC
2005 }
2006 iter->xhv_eiter = eiter;
2007}
2008
bfcb3514 2009void
4164be69 2010Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
bfcb3514 2011{
97aff369 2012 dVAR;
b79f7545 2013 struct xpvhv_aux *iter;
7423f6db 2014 U32 hash;
46c461b5 2015
7918f24d 2016 PERL_ARGS_ASSERT_HV_NAME_SET;
46c461b5 2017 PERL_UNUSED_ARG(flags);
bfcb3514 2018
4164be69
NC
2019 if (len > I32_MAX)
2020 Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len);
2021
b79f7545
NC
2022 if (SvOOK(hv)) {
2023 iter = HvAUX(hv);
7423f6db
NC
2024 if (iter->xhv_name) {
2025 unshare_hek_or_pvn(iter->xhv_name, 0, 0, 0);
2026 }
16580ff5 2027 } else {
bfcb3514
NC
2028 if (name == 0)
2029 return;
2030
6136c704 2031 iter = hv_auxinit(hv);
bfcb3514 2032 }
7423f6db 2033 PERL_HASH(hash, name, len);
adf4e37a 2034 iter->xhv_name = name ? share_hek(name, len, hash) : NULL;
bfcb3514
NC
2035}
2036
86f55936
NC
2037AV **
2038Perl_hv_backreferences_p(pTHX_ HV *hv) {
6136c704 2039 struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv);
7918f24d
NC
2040
2041 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
96a5add6 2042 PERL_UNUSED_CONTEXT;
7918f24d 2043
86f55936
NC
2044 return &(iter->xhv_backreferences);
2045}
2046
2047void
2048Perl_hv_kill_backrefs(pTHX_ HV *hv) {
2049 AV *av;
2050
7918f24d
NC
2051 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2052
86f55936
NC
2053 if (!SvOOK(hv))
2054 return;
2055
2056 av = HvAUX(hv)->xhv_backreferences;
2057
2058 if (av) {
2059 HvAUX(hv)->xhv_backreferences = 0;
ad64d0ec 2060 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
b17f5ab7 2061 SvREFCNT_dec(av);
86f55936
NC
2062 }
2063}
2064
954c1994 2065/*
7a7b9979
NC
2066hv_iternext is implemented as a macro in hv.h
2067
954c1994
GS
2068=for apidoc hv_iternext
2069
2070Returns entries from a hash iterator. See C<hv_iterinit>.
2071
fe7bca90
NC
2072You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2073iterator currently points to, without losing your place or invalidating your
2074iterator. Note that in this case the current entry is deleted from the hash
2075with your iterator holding the last reference to it. Your iterator is flagged
2076to free the entry on the next call to C<hv_iternext>, so you must not discard
2077your iterator immediately else the entry will leak - call C<hv_iternext> to
2078trigger the resource deallocation.
2079
fe7bca90
NC
2080=for apidoc hv_iternext_flags
2081
2082Returns entries from a hash iterator. See C<hv_iterinit> and C<hv_iternext>.
2083The C<flags> value will normally be zero; if HV_ITERNEXT_WANTPLACEHOLDERS is
2084set the placeholders keys (for restricted hashes) will be returned in addition
2085to normal keys. By default placeholders are automatically skipped over.
7996736c
MHM
2086Currently a placeholder is implemented with a value that is
2087C<&Perl_sv_placeholder>. Note that the implementation of placeholders and
fe7bca90
NC
2088restricted hashes may change, and the implementation currently is
2089insufficiently abstracted for any change to be tidy.
e16e2ff8 2090
fe7bca90 2091=cut
e16e2ff8
NC
2092*/
2093
2094HE *
2095Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2096{
27da23d5 2097 dVAR;
cbec9347 2098 register XPVHV* xhv;
79072805 2099 register HE *entry;
a0d0e21e 2100 HE *oldentry;
463ee0b2 2101 MAGIC* mg;
bfcb3514 2102 struct xpvhv_aux *iter;
79072805 2103
7918f24d
NC
2104 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2105
79072805 2106 if (!hv)
cea2e8a9 2107 Perl_croak(aTHX_ "Bad hash");
81714fb9 2108
cbec9347 2109 xhv = (XPVHV*)SvANY(hv);
bfcb3514 2110
b79f7545 2111 if (!SvOOK(hv)) {
bfcb3514
NC
2112 /* Too many things (well, pp_each at least) merrily assume that you can
2113 call iv_iternext without calling hv_iterinit, so we'll have to deal
2114 with it. */
2115 hv_iterinit(hv);
bfcb3514 2116 }
b79f7545 2117 iter = HvAUX(hv);
bfcb3514
NC
2118
2119 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
e62cc96a 2120 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
ad64d0ec 2121 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
e62cc96a
YO
2122 SV * const key = sv_newmortal();
2123 if (entry) {
2124 sv_setsv(key, HeSVKEY_force(entry));
2125 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
2126 }
2127 else {
2128 char *k;
2129 HEK *hek;
2130
2131 /* one HE per MAGICAL hash */
2132 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
2133 Zero(entry, 1, HE);
ad64d0ec 2134 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
e62cc96a
YO
2135 hek = (HEK*)k;
2136 HeKEY_hek(entry) = hek;
2137 HeKLEN(entry) = HEf_SVKEY;
2138 }
ad64d0ec 2139 magic_nextpack(MUTABLE_SV(hv),mg,key);
e62cc96a
YO
2140 if (SvOK(key)) {
2141 /* force key to stay around until next time */
2142 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2143 return entry; /* beware, hent_val is not set */
2144 }
2145 if (HeVAL(entry))
2146 SvREFCNT_dec(HeVAL(entry));
2147 Safefree(HeKEY_hek(entry));
2148 del_HE(entry);
2149 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2150 return NULL;
81714fb9 2151 }
79072805 2152 }
7ee146b1 2153#if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */
ad64d0ec
NC
2154 if (!entry && SvRMAGICAL((const SV *)hv)
2155 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
f675dbe5 2156 prime_env_iter();
03026e68
JM
2157#ifdef VMS
2158 /* The prime_env_iter() on VMS just loaded up new hash values
2159 * so the iteration count needs to be reset back to the beginning
2160 */
2161 hv_iterinit(hv);
2162 iter = HvAUX(hv);
2163 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
2164#endif
2165 }
f675dbe5 2166#endif
463ee0b2 2167
b79f7545
NC
2168 /* hv_iterint now ensures this. */
2169 assert (HvARRAY(hv));
2170
015a5f36 2171 /* At start of hash, entry is NULL. */
fde52b5c 2172 if (entry)
8aacddc1 2173 {
fde52b5c 2174 entry = HeNEXT(entry);
e16e2ff8
NC
2175 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2176 /*
2177 * Skip past any placeholders -- don't want to include them in
2178 * any iteration.
2179 */
7996736c 2180 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
e16e2ff8
NC
2181 entry = HeNEXT(entry);
2182 }
8aacddc1
NIS
2183 }
2184 }
015a5f36 2185
9eb4ebd1
NC
2186 /* Skip the entire loop if the hash is empty. */
2187 if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS)
2188 ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) {
900ac051
MM
2189 while (!entry) {
2190 /* OK. Come to the end of the current list. Grab the next one. */
2191
2192 iter->xhv_riter++; /* HvRITER(hv)++ */
2193 if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) {
2194 /* There is no next one. End of the hash. */
2195 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2196 break;
2197 }
2198 entry = (HvARRAY(hv))[iter->xhv_riter];
8aacddc1 2199
900ac051
MM
2200 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
2201 /* If we have an entry, but it's a placeholder, don't count it.
2202 Try the next. */
2203 while (entry && HeVAL(entry) == &PL_sv_placeholder)
2204 entry = HeNEXT(entry);
2205 }
2206 /* Will loop again if this linked list starts NULL
2207 (for HV_ITERNEXT_WANTPLACEHOLDERS)
2208 or if we run through it and find only placeholders. */
015a5f36 2209 }
fde52b5c 2210 }
79072805 2211
72940dca 2212 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2213 HvLAZYDEL_off(hv);
68dc0745 2214 hv_free_ent(hv, oldentry);
72940dca 2215 }
a0d0e21e 2216
fdcd69b6 2217 /*if (HvREHASH(hv) && entry && !HeKREHASH(entry))
6c9570dc 2218 PerlIO_printf(PerlIO_stderr(), "Awooga %p %p\n", (void*)hv, (void*)entry);*/
fdcd69b6 2219
bfcb3514 2220 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
79072805
LW
2221 return entry;
2222}
2223
954c1994
GS
2224/*
2225=for apidoc hv_iterkey
2226
2227Returns the key from the current position of the hash iterator. See
2228C<hv_iterinit>.
2229
2230=cut
2231*/
2232
79072805 2233char *
864dbfa3 2234Perl_hv_iterkey(pTHX_ register HE *entry, I32 *retlen)
79072805 2235{
7918f24d
NC
2236 PERL_ARGS_ASSERT_HV_ITERKEY;
2237
fde52b5c 2238 if (HeKLEN(entry) == HEf_SVKEY) {
fb73857a 2239 STRLEN len;
0bd48802 2240 char * const p = SvPV(HeKEY_sv(entry), len);
fb73857a 2241 *retlen = len;
2242 return p;
fde52b5c 2243 }
2244 else {
2245 *retlen = HeKLEN(entry);
2246 return HeKEY(entry);
2247 }
2248}
2249
2250/* unlike hv_iterval(), this always returns a mortal copy of the key */
954c1994
GS
2251/*
2252=for apidoc hv_iterkeysv
2253
2254Returns the key as an C<SV*> from the current position of the hash
2255iterator. The return value will always be a mortal copy of the key. Also
2256see C<hv_iterinit>.
2257
2258=cut
2259*/
2260
fde52b5c 2261SV *
864dbfa3 2262Perl_hv_iterkeysv(pTHX_ register HE *entry)
fde52b5c 2263{
7918f24d
NC
2264 PERL_ARGS_ASSERT_HV_ITERKEYSV;
2265
c1b02ed8 2266 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
79072805
LW
2267}
2268
954c1994
GS
2269/*
2270=for apidoc hv_iterval
2271
2272Returns the value from the current position of the hash iterator. See
2273C<hv_iterkey>.
2274
2275=cut
2276*/
2277
79072805 2278SV *
864dbfa3 2279Perl_hv_iterval(pTHX_ HV *hv, register HE *entry)
79072805 2280{
7918f24d
NC
2281 PERL_ARGS_ASSERT_HV_ITERVAL;
2282
8990e307 2283 if (SvRMAGICAL(hv)) {
ad64d0ec 2284 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
c4420975 2285 SV* const sv = sv_newmortal();
bbce6d69 2286 if (HeKLEN(entry) == HEf_SVKEY)
ad64d0ec 2287 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
a3b680e6 2288 else
ad64d0ec 2289 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
463ee0b2
LW
2290 return sv;
2291 }
79072805 2292 }
fde52b5c 2293 return HeVAL(entry);
79072805
LW
2294}
2295
954c1994
GS
2296/*
2297=for apidoc hv_iternextsv
2298
2299Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
2300operation.
2301
2302=cut
2303*/
2304
a0d0e21e 2305SV *
864dbfa3 2306Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
a0d0e21e 2307{
0bd48802
AL
2308 HE * const he = hv_iternext_flags(hv, 0);
2309
7918f24d
NC
2310 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
2311
0bd48802 2312 if (!he)
a0d0e21e
LW
2313 return NULL;
2314 *key = hv_iterkey(he, retlen);
2315 return hv_iterval(hv, he);
2316}
2317
954c1994 2318/*
bc5cdc23
NC
2319
2320Now a macro in hv.h
2321
954c1994
GS
2322=for apidoc hv_magic
2323
2324Adds magic to a hash. See C<sv_magic>.
2325
2326=cut
2327*/
2328
bbce6d69 2329/* possibly free a shared string if no one has access to it
fde52b5c 2330 * len and hash must both be valid for str.
2331 */
bbce6d69 2332void
864dbfa3 2333Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
fde52b5c 2334{
19692e8d
NC
2335 unshare_hek_or_pvn (NULL, str, len, hash);
2336}
2337
2338
2339void
2340Perl_unshare_hek(pTHX_ HEK *hek)
2341{
bf11fd37 2342 assert(hek);
19692e8d
NC
2343 unshare_hek_or_pvn(hek, NULL, 0, 0);
2344}
2345
2346/* possibly free a shared string if no one has access to it
2347 hek if non-NULL takes priority over the other 3, else str, len and hash
2348 are used. If so, len and hash must both be valid for str.
2349 */
df132699 2350STATIC void
97ddebaf 2351S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
19692e8d 2352{
97aff369 2353 dVAR;
cbec9347 2354 register XPVHV* xhv;
20454177 2355 HE *entry;
fde52b5c 2356 register HE **oentry;
45d1cc86 2357 HE **first;
c3654f1a 2358 bool is_utf8 = FALSE;
19692e8d 2359 int k_flags = 0;
aec46f14 2360 const char * const save = str;
cbbf8932 2361 struct shared_he *he = NULL;
c3654f1a 2362
19692e8d 2363 if (hek) {
cbae3960
NC
2364 /* Find the shared he which is just before us in memory. */
2365 he = (struct shared_he *)(((char *)hek)
2366 - STRUCT_OFFSET(struct shared_he,
2367 shared_he_hek));
2368
2369 /* Assert that the caller passed us a genuine (or at least consistent)
2370 shared hek */
2371 assert (he->shared_he_he.hent_hek == hek);
29404ae0 2372
de616631
NC
2373 if (he->shared_he_he.he_valu.hent_refcount - 1) {
2374 --he->shared_he_he.he_valu.hent_refcount;
29404ae0
NC
2375 return;
2376 }
29404ae0 2377
19692e8d
NC
2378 hash = HEK_HASH(hek);
2379 } else if (len < 0) {
2380 STRLEN tmplen = -len;
2381 is_utf8 = TRUE;
2382 /* See the note in hv_fetch(). --jhi */
2383 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2384 len = tmplen;
2385 if (is_utf8)
2386 k_flags = HVhek_UTF8;
2387 if (str != save)
2388 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
c3654f1a 2389 }
1c846c1f 2390
de616631 2391 /* what follows was the moral equivalent of:
6b88bc9c 2392 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
a0714e2c 2393 if (--*Svp == NULL)
6b88bc9c 2394 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
bbce6d69 2395 } */
cbec9347 2396 xhv = (XPVHV*)SvANY(PL_strtab);
fde52b5c 2397 /* assert(xhv_array != 0) */
45d1cc86 2398 first = oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
6c1b96a1
NC
2399 if (he) {
2400 const HE *const he_he = &(he->shared_he_he);
45d1cc86 2401 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
35ab5632
NC
2402 if (entry == he_he)
2403 break;
19692e8d
NC
2404 }
2405 } else {
35a4481c 2406 const int flags_masked = k_flags & HVhek_MASK;
45d1cc86 2407 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
19692e8d
NC
2408 if (HeHASH(entry) != hash) /* strings can't be equal */
2409 continue;
2410 if (HeKLEN(entry) != len)
2411 continue;
2412 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
2413 continue;
2414 if (HeKFLAGS(entry) != flags_masked)
2415 continue;
19692e8d
NC
2416 break;
2417 }
2418 }
2419
35ab5632
NC
2420 if (entry) {
2421 if (--entry->he_valu.hent_refcount == 0) {
19692e8d 2422 *oentry = HeNEXT(entry);
45d1cc86
NC
2423 if (!*first) {
2424 /* There are now no entries in our slot. */
19692e8d 2425 xhv->xhv_fill--; /* HvFILL(hv)-- */
45d1cc86 2426 }
cbae3960 2427 Safefree(entry);
4c7185a0 2428 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
19692e8d 2429 }
fde52b5c 2430 }
19692e8d 2431
35ab5632 2432 if (!entry && ckWARN_d(WARN_INTERNAL))
19692e8d 2433 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
472d47bc
SB
2434 "Attempt to free non-existent shared string '%s'%s"
2435 pTHX__FORMAT,
19692e8d 2436 hek ? HEK_KEY(hek) : str,
472d47bc 2437 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
19692e8d
NC
2438 if (k_flags & HVhek_FREEKEY)
2439 Safefree(str);
fde52b5c 2440}
2441
bbce6d69 2442/* get a (constant) string ptr from the global string table
2443 * string will get added if it is not already there.
fde52b5c 2444 * len and hash must both be valid for str.
2445 */
bbce6d69 2446HEK *
864dbfa3 2447Perl_share_hek(pTHX_ const char *str, I32 len, register U32 hash)
fde52b5c 2448{
da58a35d 2449 bool is_utf8 = FALSE;
19692e8d 2450 int flags = 0;
aec46f14 2451 const char * const save = str;
da58a35d 2452
7918f24d
NC
2453 PERL_ARGS_ASSERT_SHARE_HEK;
2454
da58a35d 2455 if (len < 0) {
77caf834 2456 STRLEN tmplen = -len;
da58a35d 2457 is_utf8 = TRUE;
77caf834
JH
2458 /* See the note in hv_fetch(). --jhi */
2459 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
2460 len = tmplen;
19692e8d
NC
2461 /* If we were able to downgrade here, then than means that we were passed
2462 in a key which only had chars 0-255, but was utf8 encoded. */
2463 if (is_utf8)
2464 flags = HVhek_UTF8;
2465 /* If we found we were able to downgrade the string to bytes, then
2466 we should flag that it needs upgrading on keys or each. Also flag
2467 that we need share_hek_flags to free the string. */
2468 if (str != save)
2469 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
2470 }
2471
6e838c70 2472 return share_hek_flags (str, len, hash, flags);
19692e8d
NC
2473}
2474
6e838c70 2475STATIC HEK *
19692e8d
NC
2476S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags)
2477{
97aff369 2478 dVAR;
19692e8d 2479 register HE *entry;
35a4481c 2480 const int flags_masked = flags & HVhek_MASK;
263cb4a6 2481 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
7918f24d
NC
2482 register XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
2483
2484 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
bbce6d69 2485
fde52b5c 2486 /* what follows is the moral equivalent of:
1c846c1f 2487
6b88bc9c 2488 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
a0714e2c 2489 hv_store(PL_strtab, str, len, NULL, hash);
fdcd69b6
NC
2490
2491 Can't rehash the shared string table, so not sure if it's worth
2492 counting the number of entries in the linked list
bbce6d69 2493 */
7918f24d 2494
fde52b5c 2495 /* assert(xhv_array != 0) */
263cb4a6
NC
2496 entry = (HvARRAY(PL_strtab))[hindex];
2497 for (;entry; entry = HeNEXT(entry)) {
fde52b5c 2498 if (HeHASH(entry) != hash) /* strings can't be equal */
2499 continue;
2500 if (HeKLEN(entry) != len)
2501 continue;
1c846c1f 2502 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
fde52b5c 2503 continue;
19692e8d 2504 if (HeKFLAGS(entry) != flags_masked)
c3654f1a 2505 continue;
fde52b5c 2506 break;
2507 }
263cb4a6
NC
2508
2509 if (!entry) {
45d1cc86
NC
2510 /* What used to be head of the list.
2511 If this is NULL, then we're the first entry for this slot, which
2512 means we need to increate fill. */
cbae3960
NC
2513 struct shared_he *new_entry;
2514 HEK *hek;
2515 char *k;
263cb4a6
NC
2516 HE **const head = &HvARRAY(PL_strtab)[hindex];
2517 HE *const next = *head;
cbae3960
NC
2518
2519 /* We don't actually store a HE from the arena and a regular HEK.
2520 Instead we allocate one chunk of memory big enough for both,
2521 and put the HEK straight after the HE. This way we can find the
2522 HEK directly from the HE.
2523 */
2524
a02a5408 2525 Newx(k, STRUCT_OFFSET(struct shared_he,
cbae3960
NC
2526 shared_he_hek.hek_key[0]) + len + 2, char);
2527 new_entry = (struct shared_he *)k;
2528 entry = &(new_entry->shared_he_he);
2529 hek = &(new_entry->shared_he_hek);
2530
2531 Copy(str, HEK_KEY(hek), len, char);
2532 HEK_KEY(hek)[len] = 0;
2533 HEK_LEN(hek) = len;
2534 HEK_HASH(hek) = hash;
2535 HEK_FLAGS(hek) = (unsigned char)flags_masked;
2536
2537 /* Still "point" to the HEK, so that other code need not know what
2538 we're up to. */
2539 HeKEY_hek(entry) = hek;
de616631 2540 entry->he_valu.hent_refcount = 0;
263cb4a6
NC
2541 HeNEXT(entry) = next;
2542 *head = entry;
cbae3960 2543
4c7185a0 2544 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
263cb4a6 2545 if (!next) { /* initial entry? */
cbec9347 2546 xhv->xhv_fill++; /* HvFILL(hv)++ */
4c9cc595 2547 } else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
cbec9347 2548 hsplit(PL_strtab);
bbce6d69 2549 }
2550 }
2551
de616631 2552 ++entry->he_valu.hent_refcount;
19692e8d
NC
2553
2554 if (flags & HVhek_FREEKEY)
f9a63242 2555 Safefree(str);
19692e8d 2556
6e838c70 2557 return HeKEY_hek(entry);
fde52b5c 2558}
ecae49c0 2559
ca732855
NC
2560I32 *
2561Perl_hv_placeholders_p(pTHX_ HV *hv)
2562{
2563 dVAR;
ad64d0ec 2564 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 2565
7918f24d
NC
2566 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
2567
ca732855 2568 if (!mg) {
ad64d0ec 2569 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
ca732855
NC
2570
2571 if (!mg) {
2572 Perl_die(aTHX_ "panic: hv_placeholders_p");
2573 }
2574 }
2575 return &(mg->mg_len);
2576}
2577
2578
2579I32
0c289d13 2580Perl_hv_placeholders_get(pTHX_ const HV *hv)
ca732855
NC
2581{
2582 dVAR;
0c289d13 2583 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 2584
7918f24d
NC
2585 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
2586
ca732855
NC
2587 return mg ? mg->mg_len : 0;
2588}
2589
2590void
ac1e784a 2591Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
ca732855
NC
2592{
2593 dVAR;
ad64d0ec 2594 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 2595
7918f24d
NC
2596 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
2597
ca732855
NC
2598 if (mg) {
2599 mg->mg_len = ph;
2600 } else if (ph) {
ad64d0ec 2601 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
ca732855
NC
2602 Perl_die(aTHX_ "panic: hv_placeholders_set");
2603 }
2604 /* else we don't need to add magic to record 0 placeholders. */
2605}
ecae49c0 2606
2a49f0f5 2607STATIC SV *
7b0bddfa
NC
2608S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
2609{
0b2d3faa 2610 dVAR;
7b0bddfa 2611 SV *value;
7918f24d
NC
2612
2613 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
2614
7b0bddfa
NC
2615 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
2616 case HVrhek_undef:
2617 value = newSV(0);
2618 break;
2619 case HVrhek_delete:
2620 value = &PL_sv_placeholder;
2621 break;
2622 case HVrhek_IV:
44ebaf21
NC
2623 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
2624 break;
2625 case HVrhek_UV:
2626 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
7b0bddfa
NC
2627 break;
2628 case HVrhek_PV:
44ebaf21 2629 case HVrhek_PV_UTF8:
7b0bddfa
NC
2630 /* Create a string SV that directly points to the bytes in our
2631 structure. */
b9f83d2f 2632 value = newSV_type(SVt_PV);
7b0bddfa
NC
2633 SvPV_set(value, (char *) he->refcounted_he_data + 1);
2634 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
2635 /* This stops anything trying to free it */
2636 SvLEN_set(value, 0);
2637 SvPOK_on(value);
2638 SvREADONLY_on(value);
44ebaf21 2639 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
7b0bddfa
NC
2640 SvUTF8_on(value);
2641 break;
2642 default:
2643 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %x",
2644 he->refcounted_he_data[0]);
2645 }
2646 return value;
2647}
2648
ecae49c0 2649/*
b3ca2e83
NC
2650=for apidoc refcounted_he_chain_2hv
2651
abc25d8c 2652Generates and returns a C<HV *> by walking up the tree starting at the passed
b3ca2e83
NC
2653in C<struct refcounted_he *>.
2654
2655=cut
2656*/
2657HV *
2658Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain)
2659{
7a89be66 2660 dVAR;
b3ca2e83
NC
2661 HV *hv = newHV();
2662 U32 placeholders = 0;
2663 /* We could chase the chain once to get an idea of the number of keys,
2664 and call ksplit. But for now we'll make a potentially inefficient
2665 hash with only 8 entries in its array. */
2666 const U32 max = HvMAX(hv);
2667
2668 if (!HvARRAY(hv)) {
2669 char *array;
2670 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
2671 HvARRAY(hv) = (HE**)array;
2672 }
2673
2674 while (chain) {
cbb1fbea 2675#ifdef USE_ITHREADS
b6bbf3fa 2676 U32 hash = chain->refcounted_he_hash;
cbb1fbea
NC
2677#else
2678 U32 hash = HEK_HASH(chain->refcounted_he_hek);
2679#endif
b3ca2e83
NC
2680 HE **oentry = &((HvARRAY(hv))[hash & max]);
2681 HE *entry = *oentry;
b6bbf3fa 2682 SV *value;
cbb1fbea 2683
b3ca2e83
NC
2684 for (; entry; entry = HeNEXT(entry)) {
2685 if (HeHASH(entry) == hash) {
9f769845
NC
2686 /* We might have a duplicate key here. If so, entry is older
2687 than the key we've already put in the hash, so if they are
2688 the same, skip adding entry. */
2689#ifdef USE_ITHREADS
2690 const STRLEN klen = HeKLEN(entry);
2691 const char *const key = HeKEY(entry);
2692 if (klen == chain->refcounted_he_keylen
2693 && (!!HeKUTF8(entry)
2694 == !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2695 && memEQ(key, REF_HE_KEY(chain), klen))
2696 goto next_please;
2697#else
2698 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
2699 goto next_please;
2700 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
2701 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
2702 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
2703 HeKLEN(entry)))
2704 goto next_please;
2705#endif
b3ca2e83
NC
2706 }
2707 }
2708 assert (!entry);
2709 entry = new_HE();
2710
cbb1fbea
NC
2711#ifdef USE_ITHREADS
2712 HeKEY_hek(entry)
7b0bddfa 2713 = share_hek_flags(REF_HE_KEY(chain),
b6bbf3fa
NC
2714 chain->refcounted_he_keylen,
2715 chain->refcounted_he_hash,
2716 (chain->refcounted_he_data[0]
2717 & (HVhek_UTF8|HVhek_WASUTF8)));
cbb1fbea 2718#else
71ad1b0c 2719 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
cbb1fbea 2720#endif
7b0bddfa
NC
2721 value = refcounted_he_value(chain);
2722 if (value == &PL_sv_placeholder)
b3ca2e83 2723 placeholders++;
b6bbf3fa 2724 HeVAL(entry) = value;
b3ca2e83
NC
2725
2726 /* Link it into the chain. */
2727 HeNEXT(entry) = *oentry;
2728 if (!HeNEXT(entry)) {
2729 /* initial entry. */
2730 HvFILL(hv)++;
2731 }
2732 *oentry = entry;
2733
2734 HvTOTALKEYS(hv)++;
2735
2736 next_please:
71ad1b0c 2737 chain = chain->refcounted_he_next;
b3ca2e83
NC
2738 }
2739
2740 if (placeholders) {
2741 clear_placeholders(hv, placeholders);
2742 HvTOTALKEYS(hv) -= placeholders;
2743 }
2744
2745 /* We could check in the loop to see if we encounter any keys with key
2746 flags, but it's probably not worth it, as this per-hash flag is only
2747 really meant as an optimisation for things like Storable. */
2748 HvHASKFLAGS_on(hv);
def9038f 2749 DEBUG_A(Perl_hv_assert(aTHX_ hv));
b3ca2e83
NC
2750
2751 return hv;
2752}
2753
7b0bddfa
NC
2754SV *
2755Perl_refcounted_he_fetch(pTHX_ const struct refcounted_he *chain, SV *keysv,
2756 const char *key, STRLEN klen, int flags, U32 hash)
2757{
0b2d3faa 2758 dVAR;
7b0bddfa
NC
2759 /* Just to be awkward, if you're using this interface the UTF-8-or-not-ness
2760 of your key has to exactly match that which is stored. */
2761 SV *value = &PL_sv_placeholder;
7b0bddfa 2762
cd1d2f8a
NC
2763 if (chain) {
2764 /* No point in doing any of this if there's nothing to find. */
2765 bool is_utf8;
7b0bddfa 2766
cd1d2f8a
NC
2767 if (keysv) {
2768 if (flags & HVhek_FREEKEY)
2769 Safefree(key);
2770 key = SvPV_const(keysv, klen);
2771 flags = 0;
2772 is_utf8 = (SvUTF8(keysv) != 0);
2773 } else {
2774 is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE);
2775 }
2776
2777 if (!hash) {
2778 if (keysv && (SvIsCOW_shared_hash(keysv))) {
2779 hash = SvSHARED_HASH(keysv);
2780 } else {
2781 PERL_HASH(hash, key, klen);
2782 }
2783 }
7b0bddfa 2784
cd1d2f8a 2785 for (; chain; chain = chain->refcounted_he_next) {
7b0bddfa 2786#ifdef USE_ITHREADS
cd1d2f8a
NC
2787 if (hash != chain->refcounted_he_hash)
2788 continue;
2789 if (klen != chain->refcounted_he_keylen)
2790 continue;
2791 if (memNE(REF_HE_KEY(chain),key,klen))
2792 continue;
2793 if (!!is_utf8 != !!(chain->refcounted_he_data[0] & HVhek_UTF8))
2794 continue;
7b0bddfa 2795#else
cd1d2f8a
NC
2796 if (hash != HEK_HASH(chain->refcounted_he_hek))
2797 continue;
2798 if (klen != (STRLEN)HEK_LEN(chain->refcounted_he_hek))
2799 continue;
2800 if (memNE(HEK_KEY(chain->refcounted_he_hek),key,klen))
2801 continue;
2802 if (!!is_utf8 != !!HEK_UTF8(chain->refcounted_he_hek))
2803 continue;
7b0bddfa
NC
2804#endif
2805
cd1d2f8a
NC
2806 value = sv_2mortal(refcounted_he_value(chain));
2807 break;
2808 }
7b0bddfa
NC
2809 }
2810
2811 if (flags & HVhek_FREEKEY)
2812 Safefree(key);
2813
2814 return value;
2815}
2816
b3ca2e83
NC
2817/*
2818=for apidoc refcounted_he_new
2819
ec2a1de7
NC
2820Creates a new C<struct refcounted_he>. As S<key> is copied, and value is
2821stored in a compact form, all references remain the property of the caller.
2822The C<struct refcounted_he> is returned with a reference count of 1.
b3ca2e83
NC
2823
2824=cut
2825*/
2826
2827struct refcounted_he *
2828Perl_refcounted_he_new(pTHX_ struct refcounted_he *const parent,
2829 SV *const key, SV *const value) {
7a89be66 2830 dVAR;
b6bbf3fa
NC
2831 STRLEN key_len;
2832 const char *key_p = SvPV_const(key, key_len);
2833 STRLEN value_len = 0;
95b63a38 2834 const char *value_p = NULL;
b6bbf3fa
NC
2835 char value_type;
2836 char flags;
d8c5b3c5 2837 bool is_utf8 = SvUTF8(key) ? TRUE : FALSE;
b6bbf3fa
NC
2838
2839 if (SvPOK(value)) {
2840 value_type = HVrhek_PV;
2841 } else if (SvIOK(value)) {
ad64d0ec 2842 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
b6bbf3fa
NC
2843 } else if (value == &PL_sv_placeholder) {
2844 value_type = HVrhek_delete;
2845 } else if (!SvOK(value)) {
2846 value_type = HVrhek_undef;
2847 } else {
2848 value_type = HVrhek_PV;
2849 }
b3ca2e83 2850
b6bbf3fa 2851 if (value_type == HVrhek_PV) {
012da8e5
NC
2852 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
2853 the value is overloaded, and doesn't yet have the UTF-8flag set. */
b6bbf3fa 2854 value_p = SvPV_const(value, value_len);
012da8e5
NC
2855 if (SvUTF8(value))
2856 value_type = HVrhek_PV_UTF8;
b6bbf3fa 2857 }
012da8e5
NC
2858 flags = value_type;
2859
2860 if (is_utf8) {
2861 /* Hash keys are always stored normalised to (yes) ISO-8859-1.
2862 As we're going to be building hash keys from this value in future,
2863 normalise it now. */
2864 key_p = (char*)bytes_from_utf8((const U8*)key_p, &key_len, &is_utf8);
2865 flags |= is_utf8 ? HVhek_UTF8 : HVhek_WASUTF8;
2866 }
2867
2868 return refcounted_he_new_common(parent, key_p, key_len, flags, value_type,
2869 ((value_type == HVrhek_PV
2870 || value_type == HVrhek_PV_UTF8) ?
2871 (void *)value_p : (void *)value),
2872 value_len);
2873}
2874
1da83c39 2875static struct refcounted_he *
012da8e5
NC
2876S_refcounted_he_new_common(pTHX_ struct refcounted_he *const parent,
2877 const char *const key_p, const STRLEN key_len,
2878 const char flags, char value_type,
2879 const void *value, const STRLEN value_len) {
2880 dVAR;
2881 struct refcounted_he *he;
2882 U32 hash;
2883 const bool is_pv = value_type == HVrhek_PV || value_type == HVrhek_PV_UTF8;
2884 STRLEN key_offset = is_pv ? value_len + 2 : 1;
2885
2886 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_COMMON;
b6bbf3fa 2887
b6bbf3fa 2888#ifdef USE_ITHREADS
10edeb5d
JH
2889 he = (struct refcounted_he*)
2890 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2891 + key_len
2892 + key_offset);
6cef672b 2893#else
10edeb5d
JH
2894 he = (struct refcounted_he*)
2895 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
2896 + key_offset);
6cef672b 2897#endif
b3ca2e83 2898
71ad1b0c 2899 he->refcounted_he_next = parent;
b6bbf3fa 2900
012da8e5
NC
2901 if (is_pv) {
2902 Copy((char *)value, he->refcounted_he_data + 1, value_len + 1, char);
b6bbf3fa 2903 he->refcounted_he_val.refcounted_he_u_len = value_len;
b6bbf3fa 2904 } else if (value_type == HVrhek_IV) {
ad64d0ec 2905 he->refcounted_he_val.refcounted_he_u_iv = SvIVX((const SV *)value);
012da8e5 2906 } else if (value_type == HVrhek_UV) {
ad64d0ec 2907 he->refcounted_he_val.refcounted_he_u_uv = SvUVX((const SV *)value);
b6bbf3fa
NC
2908 }
2909
b6bbf3fa
NC
2910 PERL_HASH(hash, key_p, key_len);
2911
cbb1fbea 2912#ifdef USE_ITHREADS
b6bbf3fa
NC
2913 he->refcounted_he_hash = hash;
2914 he->refcounted_he_keylen = key_len;
2915 Copy(key_p, he->refcounted_he_data + key_offset, key_len, char);
cbb1fbea 2916#else
b6bbf3fa 2917 he->refcounted_he_hek = share_hek_flags(key_p, key_len, hash, flags);
cbb1fbea 2918#endif
b6bbf3fa
NC
2919
2920 if (flags & HVhek_WASUTF8) {
2921 /* If it was downgraded from UTF-8, then the pointer returned from
2922 bytes_from_utf8 is an allocated pointer that we must free. */
2923 Safefree(key_p);
2924 }
2925
2926 he->refcounted_he_data[0] = flags;
b3ca2e83
NC
2927 he->refcounted_he_refcnt = 1;
2928
2929 return he;
2930}
2931
2932/*
2933=for apidoc refcounted_he_free
2934
2935Decrements the reference count of the passed in C<struct refcounted_he *>
2936by one. If the reference count reaches zero the structure's memory is freed,
2937and C<refcounted_he_free> iterates onto the parent node.
2938
2939=cut
2940*/
2941
2942void
2943Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
53d44271 2944 dVAR;
57ca3b03
AL
2945 PERL_UNUSED_CONTEXT;
2946
b3ca2e83
NC
2947 while (he) {
2948 struct refcounted_he *copy;
cbb1fbea 2949 U32 new_count;
b3ca2e83 2950
cbb1fbea
NC
2951 HINTS_REFCNT_LOCK;
2952 new_count = --he->refcounted_he_refcnt;
2953 HINTS_REFCNT_UNLOCK;
2954
2955 if (new_count) {
b3ca2e83 2956 return;
cbb1fbea 2957 }
b3ca2e83 2958
b6bbf3fa 2959#ifndef USE_ITHREADS
71ad1b0c 2960 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
cbb1fbea 2961#endif
b3ca2e83 2962 copy = he;
71ad1b0c 2963 he = he->refcounted_he_next;
b6bbf3fa 2964 PerlMemShared_free(copy);
b3ca2e83
NC
2965 }
2966}
2967
dca6062a
NC
2968const char *
2969Perl_fetch_cop_label(pTHX_ struct refcounted_he *const chain, STRLEN *len,
2970 U32 *flags) {
2971 if (!chain)
2972 return NULL;
2973#ifdef USE_ITHREADS
2974 if (chain->refcounted_he_keylen != 1)
2975 return NULL;
2976 if (*REF_HE_KEY(chain) != ':')
2977 return NULL;
2978#else
2979 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
2980 return NULL;
2981 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
2982 return NULL;
2983#endif
012da8e5
NC
2984 /* Stop anyone trying to really mess us up by adding their own value for
2985 ':' into %^H */
2986 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
2987 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
2988 return NULL;
2989
dca6062a
NC
2990 if (len)
2991 *len = chain->refcounted_he_val.refcounted_he_u_len;
2992 if (flags) {
2993 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
2994 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
2995 }
2996 return chain->refcounted_he_data + 1;
2997}
2998
012da8e5
NC
2999/* As newSTATEOP currently gets passed plain char* labels, we will only provide
3000 that interface. Once it works out how to pass in length and UTF-8 ness, this
3001 function will need superseding. */
3002struct refcounted_he *
3003Perl_store_cop_label(pTHX_ struct refcounted_he *const chain, const char *label)
3004{
547bb267
NC
3005 PERL_ARGS_ASSERT_STORE_COP_LABEL;
3006
012da8e5
NC
3007 return refcounted_he_new_common(chain, ":", 1, HVrhek_PV, HVrhek_PV,
3008 label, strlen(label));
3009}
3010
b3ca2e83 3011/*
ecae49c0
NC
3012=for apidoc hv_assert
3013
3014Check that a hash is in an internally consistent state.
3015
3016=cut
3017*/
3018
943795c2
NC
3019#ifdef DEBUGGING
3020
ecae49c0
NC
3021void
3022Perl_hv_assert(pTHX_ HV *hv)
3023{
57ca3b03
AL
3024 dVAR;
3025 HE* entry;
3026 int withflags = 0;
3027 int placeholders = 0;
3028 int real = 0;
3029 int bad = 0;
3030 const I32 riter = HvRITER_get(hv);
3031 HE *eiter = HvEITER_get(hv);
3032
7918f24d
NC
3033 PERL_ARGS_ASSERT_HV_ASSERT;
3034
57ca3b03
AL
3035 (void)hv_iterinit(hv);
3036
3037 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
3038 /* sanity check the values */
3039 if (HeVAL(entry) == &PL_sv_placeholder)
3040 placeholders++;
3041 else
3042 real++;
3043 /* sanity check the keys */
3044 if (HeSVKEY(entry)) {
6f207bd3 3045 NOOP; /* Don't know what to check on SV keys. */
57ca3b03
AL
3046 } else if (HeKUTF8(entry)) {
3047 withflags++;
3048 if (HeKWASUTF8(entry)) {
3049 PerlIO_printf(Perl_debug_log,
d2a455e7 3050 "hash key has both WASUTF8 and UTF8: '%.*s'\n",
57ca3b03
AL
3051 (int) HeKLEN(entry), HeKEY(entry));
3052 bad = 1;
3053 }
3054 } else if (HeKWASUTF8(entry))
3055 withflags++;
3056 }
ad64d0ec 3057 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
57ca3b03
AL
3058 static const char bad_count[] = "Count %d %s(s), but hash reports %d\n";
3059 const int nhashkeys = HvUSEDKEYS(hv);
3060 const int nhashplaceholders = HvPLACEHOLDERS_get(hv);
3061
3062 if (nhashkeys != real) {
3063 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
3064 bad = 1;
3065 }
3066 if (nhashplaceholders != placeholders) {
3067 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
3068 bad = 1;
3069 }
3070 }
3071 if (withflags && ! HvHASKFLAGS(hv)) {
3072 PerlIO_printf(Perl_debug_log,
3073 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
3074 withflags);
3075 bad = 1;
3076 }
3077 if (bad) {
ad64d0ec 3078 sv_dump(MUTABLE_SV(hv));
57ca3b03
AL
3079 }
3080 HvRITER_set(hv, riter); /* Restore hash iterator state */
3081 HvEITER_set(hv, eiter);
ecae49c0 3082}
af3babe4 3083
943795c2
NC
3084#endif
3085
af3babe4
NC
3086/*
3087 * Local variables:
3088 * c-indentation-style: bsd
3089 * c-basic-offset: 4
3090 * indent-tabs-mode: t
3091 * End:
3092 *
37442d52
RGS
3093 * ex: set ts=8 sts=4 sw=4 noet:
3094 */