This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Promote v5.36 usage and feature bundles doc
[perl5.git] / hv.c
CommitLineData
a0d0e21e 1/* hv.c
79072805 2 *
1129b882
NC
3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
79072805
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
a0d0e21e
LW
9 */
10
11/*
4ac71550
TC
12 * I sit beside the fire and think
13 * of all that I have seen.
14 * --Bilbo
15 *
16 * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
79072805
LW
17 */
18
d5afce77 19/*
51b56f5c 20=head1 HV Handling
db4fbf16
FC
21A HV structure represents a Perl hash. It consists mainly of an array
22of pointers, each of which points to a linked list of HE structures. The
166f8a29 23array is indexed by the hash function of the key, so each linked list
db4fbf16 24represents all the hash entries with the same hash value. Each HE contains
166f8a29
DM
25a pointer to the actual value, plus a pointer to a HEK structure which
26holds the key and hash value.
27
28=cut
29
d5afce77
RB
30*/
31
79072805 32#include "EXTERN.h"
864dbfa3 33#define PERL_IN_HV_C
3d78eb94 34#define PERL_HASH_INTERNAL_ACCESS
79072805
LW
35#include "perl.h"
36
6f019ba7
YO
37/* we split when we collide and we have a load factor over 0.667.
38 * NOTE if you change this formula so we split earlier than previously
39 * you MUST change the logic in hv_ksplit()
40 */
aae087f7
YO
41
42/* MAX_BUCKET_MAX is the maximum max bucket index, at which point we stop growing the
43 * number of buckets,
44 */
45#define MAX_BUCKET_MAX ((1<<26)-1)
46#define DO_HSPLIT(xhv) ( ( ((xhv)->xhv_keys + ((xhv)->xhv_keys >> 1)) > (xhv)->xhv_max ) && \
47 ((xhv)->xhv_max < MAX_BUCKET_MAX) )
fdcd69b6 48
d75ce684 49static const char S_strtab_error[]
5d2b1485
NC
50 = "Cannot modify shared string table in hv_%s";
51
c3c9d6b1
YO
52#define DEBUG_HASH_RAND_BITS (DEBUG_h_TEST)
53
54/* Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs"
55 * See also https://en.wikipedia.org/wiki/Xorshift
56 */
57#if IVSIZE == 8
58/* 64 bit version */
bf2a3dae 59#define XORSHIFT_RAND_BITS(x) PERL_XORSHIFT64_A(x)
c3c9d6b1
YO
60#else
61/* 32 bit version */
bf2a3dae 62#define XORSHIFT_RAND_BITS(x) PERL_XORSHIFT32_A(x)
c3c9d6b1
YO
63#endif
64
65#define UPDATE_HASH_RAND_BITS_KEY(key,klen) \
66STMT_START { \
67 XORSHIFT_RAND_BITS(PL_hash_rand_bits); \
68 if (DEBUG_HASH_RAND_BITS) { \
69 PerlIO_printf( Perl_debug_log, \
034c827a 70 "PL_hash_rand_bits=%016" UVxf" @ %s:%-4d", \
c3c9d6b1
YO
71 (UV)PL_hash_rand_bits, __FILE__, __LINE__ \
72 ); \
73 if (DEBUG_v_TEST && key) { \
034c827a 74 PerlIO_printf( Perl_debug_log, " key:'%.*s' %" UVuf"\n", \
c3c9d6b1
YO
75 (int)klen, \
76 key ? key : "", /* silence warning */ \
77 (UV)klen \
78 ); \
79 } else { \
80 PerlIO_printf( Perl_debug_log, "\n"); \
81 } \
82 } \
83} STMT_END
84
85#define MAYBE_UPDATE_HASH_RAND_BITS_KEY(key,klen) \
86STMT_START { \
87 if (PL_HASH_RAND_BITS_ENABLED) \
88 UPDATE_HASH_RAND_BITS_KEY(key,klen); \
89} STMT_END
90
91
92#define UPDATE_HASH_RAND_BITS() \
93 UPDATE_HASH_RAND_BITS_KEY(NULL,0)
94
95#define MAYBE_UPDATE_HASH_RAND_BITS() \
96 MAYBE_UPDATE_HASH_RAND_BITS_KEY(NULL,0)
97
38b26de3
NC
98/* HeKFLAGS(entry) is a single U8, so only provides 8 flags bits.
99 We currently use 3. All 3 we have behave differently, so if we find a use for
100 more flags it's hard to predict which they group with.
101
102 Hash keys are stored as flat octet sequences, not SVs. Hence we need a flag
103 bit to say whether those octet sequences represent ISO-8859-1 or UTF-8 -
104 HVhek_UTF8. The value of this flag bit matters for (regular) hash key
105 lookups.
106
107 To speed up comparisons, keys are normalised to octets. But we (also)
108 preserve whether the key was supplied, so we need another flag bit to say
109 whether to reverse the normalisation when iterating the keys (converting them
110 back to SVs) - HVhek_WASUTF8. The value of this flag bit must be ignored for
111 (regular) hash key lookups.
112
113 But for the shared string table (the private "hash" that manages shared hash
114 keys and their reference counts), we need to be able to store both variants
115 (HVhek_WASUTF8 set and clear), so the code performing lookups in this hash
116 must be different and consider both keys.
117
118 However, regular hashes (now) can have a mix of shared and unshared keys.
119 (This avoids the need to reallocate all the keys into unshared storage at
120 the point where hash passes the "large" hash threshold, and no longer uses
121 the shared string table - existing keys remain shared, to avoid makework.)
122
123 Meaning that HVhek_NOTSHARED *may* be set in regular hashes (but should be
124 ignored for hash lookups) but must always be clear in the keys in the shared
125 string table (because the pointers to these keys are directly copied into
126 regular hashes - this is how shared keys work.)
127
128 Hence all 3 are different, and it's hard to predict the best way to future
129 proof what is needed next.
130
131 We also have HVhek_ENABLEHVKFLAGS, which is used as a mask within the code
132 below to determine whether to set HvHASKFLAGS() true on the hash as a whole.
133 This is a public "optimisation" flag provided to serealisers, to indicate
134 (up front) that a hash contains non-8-bit keys, if they want to use different
135 storage formats for hashes where all keys are simple octet sequences
136 (avoiding needing to store an extra byte per hash key), and they need to know
137 that this holds *before* iterating the hash keys. Only Storable seems to use
138 this. (For this use case, HVhek_NOTSHARED doesn't matter)
139
140 For now, we assume that any future flag bits will need to be distinguished
141 in the shared string table, hence we create this mask for the shared string
142 table code. It happens to be the same as HVhek_ENABLEHVKFLAGS, but that might
143 change if we add a flag bit that matters to the shared string table but not
144 to Storable (or similar). */
145
146#define HVhek_STORAGE_MASK (0xFF & ~HVhek_NOTSHARED)
1db404fc 147
c941fb51
NC
148#ifdef PURIFY
149
150#define new_HE() (HE*)safemalloc(sizeof(HE))
151#define del_HE(p) safefree((char*)p)
152
153#else
154
76e3520e 155STATIC HE*
cea2e8a9 156S_new_he(pTHX)
4633a7c4
LW
157{
158 HE* he;
caf0b9e5 159 void ** const root = &PL_body_roots[HE_ARENA_ROOT_IX];
6a93a7e5 160
6a93a7e5 161 if (!*root)
caf0b9e5 162 Perl_more_bodies(aTHX_ HE_ARENA_ROOT_IX, sizeof(HE), PERL_ARENA_SIZE);
10edeb5d 163 he = (HE*) *root;
ce3e5c45 164 assert(he);
6a93a7e5 165 *root = HeNEXT(he);
333f433b 166 return he;
4633a7c4
LW
167}
168
c941fb51
NC
169#define new_HE() new_he()
170#define del_HE(p) \
171 STMT_START { \
caf0b9e5
NC
172 HeNEXT(p) = (HE*)(PL_body_roots[HE_ARENA_ROOT_IX]); \
173 PL_body_roots[HE_ARENA_ROOT_IX] = p; \
c941fb51 174 } STMT_END
d33b2eba 175
d33b2eba 176
d33b2eba
GS
177
178#endif
179
76e3520e 180STATIC HEK *
5f66b61c 181S_save_hek_flags(const char *str, I32 len, U32 hash, int flags)
bbce6d69 182{
183 char *k;
eb578fdb 184 HEK *hek;
1c846c1f 185
7918f24d 186 PERL_ARGS_ASSERT_SAVE_HEK_FLAGS;
ea3ffa52
TC
187
188 Newx(k, HEK_BASESIZE + len + 2, char);
bbce6d69 189 hek = (HEK*)k;
ff68c719 190 Copy(str, HEK_KEY(hek), len, char);
e05949c7 191 HEK_KEY(hek)[len] = 0;
ff68c719 192 HEK_LEN(hek) = len;
193 HEK_HASH(hek) = hash;
38b26de3 194 HEK_FLAGS(hek) = HVhek_NOTSHARED | (flags & HVhek_STORAGE_MASK);
dcf933a4
NC
195
196 if (flags & HVhek_FREEKEY)
1604cfb0 197 Safefree(str);
bbce6d69 198 return hek;
199}
200
4a31713e 201/* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent
dd28f7bb
DM
202 * for tied hashes */
203
204void
205Perl_free_tied_hv_pool(pTHX)
206{
dd28f7bb
DM
207 HE *he = PL_hv_fetch_ent_mh;
208 while (he) {
1604cfb0
MS
209 HE * const ohe = he;
210 Safefree(HeKEY_hek(he));
211 he = HeNEXT(he);
212 del_HE(ohe);
dd28f7bb 213 }
4608196e 214 PL_hv_fetch_ent_mh = NULL;
dd28f7bb
DM
215}
216
d18c6117 217#if defined(USE_ITHREADS)
0bff533c
NC
218HEK *
219Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param)
220{
566771cc 221 HEK *shared;
9d4ba2ae 222
7918f24d 223 PERL_ARGS_ASSERT_HEK_DUP;
9d4ba2ae 224 PERL_UNUSED_ARG(param);
0bff533c 225
566771cc 226 if (!source)
1604cfb0 227 return NULL;
566771cc
NC
228
229 shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
0bff533c 230 if (shared) {
1604cfb0
MS
231 /* We already shared this hash key. */
232 (void)share_hek_hek(shared);
0bff533c
NC
233 }
234 else {
1604cfb0
MS
235 shared
236 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
237 HEK_HASH(source), HEK_FLAGS(source));
238 ptr_table_store(PL_ptr_table, source, shared);
0bff533c 239 }
658b4a4a 240 return shared;
0bff533c
NC
241}
242
d18c6117 243HE *
5c4138a0 244Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param)
d18c6117
GS
245{
246 HE *ret;
247
7918f24d
NC
248 PERL_ARGS_ASSERT_HE_DUP;
249
2c94601f
NC
250 /* All the *_dup functions are deemed to be API, despite most being deeply
251 tied to the internals. Hence we can't simply remove the parameter
252 "shared" from this function. */
253 /* sv_dup and sv_dup_inc seem to be the only two that are used by XS code.
254 Probably the others should be dropped from the API. See #19409 */
255 PERL_UNUSED_ARG(shared);
256
d18c6117 257 if (!e)
1604cfb0 258 return NULL;
7766f137
GS
259 /* look for it in the table first */
260 ret = (HE*)ptr_table_fetch(PL_ptr_table, e);
261 if (ret)
1604cfb0 262 return ret;
7766f137
GS
263
264 /* create anew and remember what it is */
d33b2eba 265 ret = new_HE();
7766f137
GS
266 ptr_table_store(PL_ptr_table, e, ret);
267
dd28f7bb 268 if (HeKLEN(e) == HEf_SVKEY) {
1604cfb0
MS
269 char *k;
270 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
271 HeKEY_hek(ret) = (HEK*)k;
272 HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param);
dd28f7bb 273 }
1db404fc 274 else if (!(HeKFLAGS(e) & HVhek_NOTSHARED)) {
1604cfb0
MS
275 /* This is hek_dup inlined, which seems to be important for speed
276 reasons. */
277 HEK * const source = HeKEY_hek(e);
278 HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source);
279
280 if (shared) {
281 /* We already shared this hash key. */
282 (void)share_hek_hek(shared);
283 }
284 else {
285 shared
286 = share_hek_flags(HEK_KEY(source), HEK_LEN(source),
287 HEK_HASH(source), HEK_FLAGS(source));
288 ptr_table_store(PL_ptr_table, source, shared);
289 }
290 HeKEY_hek(ret) = shared;
c21d1a0f 291 }
d18c6117 292 else
1604cfb0 293 HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e),
19692e8d 294 HeKFLAGS(e));
a09252eb 295 HeVAL(ret) = sv_dup_inc(HeVAL(e), param);
2c94601f
NC
296
297 HeNEXT(ret) = he_dup(HeNEXT(e), FALSE, param);
d18c6117
GS
298 return ret;
299}
300#endif /* USE_ITHREADS */
301
1b1f1335 302static void
2393f1b9 303S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen,
1604cfb0 304 const char *msg)
1b1f1335 305{
7ea8b04b
RL
306 /* Straight to SVt_PVN here, as needed by sv_setpvn_fresh and
307 * sv_usepvn would otherwise call it */
308 SV * const sv = newSV_type_mortal(SVt_PV);
7918f24d
NC
309
310 PERL_ARGS_ASSERT_HV_NOTALLOWED;
311
19692e8d 312 if (!(flags & HVhek_FREEKEY)) {
f7f919a0 313 sv_setpvn_fresh(sv, key, klen);
1b1f1335
NIS
314 }
315 else {
1604cfb0
MS
316 /* Need to free saved eventually assign to mortal SV */
317 /* XXX is this line an error ???: SV *sv = sv_newmortal(); */
318 sv_usepvn(sv, (char *) key, klen);
1b1f1335 319 }
19692e8d 320 if (flags & HVhek_UTF8) {
1604cfb0 321 SvUTF8_on(sv);
1b1f1335 322 }
be2597df 323 Perl_croak(aTHX_ msg, SVfARG(sv));
1b1f1335
NIS
324}
325
fde52b5c 326/* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot
327 * contains an SV* */
328
34a6f7b4 329/*
704ee149
KW
330=for apidoc hv_store
331=for apidoc_item hv_stores
332
333These each store SV C<val> with the specified key in hash C<hv>, returning NULL
334if the operation failed or if the value did not need to be actually stored
335within the hash (as in the case of tied hashes). Otherwise it can be
336dereferenced to get the original C<SV*>.
337
338They differ only in how the hash key is specified.
339
340In C<hv_stores>, the key is a C language string literal, enclosed in double
341quotes. It is never treated as being in UTF-8.
342
343In C<hv_store>, C<key> is either NULL or points to the first byte of the string
344specifying the key, and its length in bytes is given by the absolute value of
345an additional parameter, C<klen>. A NULL key indicates the key is to be
346treated as C<undef>, and C<klen> is ignored; otherwise the key string may
347contain embedded-NUL bytes. If C<klen> is negative, the string is treated as
348being encoded in UTF-8; otherwise not.
349
350C<hv_store> has another extra parameter, C<hash>, a precomputed hash of the key
351string, or zero if it has not been precomputed. This parameter is omitted from
352C<hv_stores>, as it is computed automatically at compile time.
353
354If <hv> is NULL, NULL is returned and no action is taken.
355
356If C<val> is NULL, it is treated as being C<undef>; otherwise the caller is
34a6f7b4 357responsible for suitably incrementing the reference count of C<val> before
796b6530
KW
358the call, and decrementing it if the function returned C<NULL>. Effectively
359a successful C<hv_store> takes ownership of one reference to C<val>. This is
34a6f7b4 360usually what you want; a newly created SV has a reference count of one, so
796b6530 361if all your code does is create SVs then store them in a hash, C<hv_store>
34a6f7b4 362will own the only reference to the new SV, and your code doesn't need to do
704ee149
KW
363anything further to tidy up.
364
365C<hv_store> is not implemented as a call to L</C<hv_store_ent>>, and does not
366create a temporary SV for the key, so if your key data is not already in SV
367form then use C<hv_store> in preference to C<hv_store_ent>.
34a6f7b4
NC
368
369See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
370information on how to use this function on tied hashes.
371
34a6f7b4
NC
372=for apidoc hv_store_ent
373
374Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash>
375parameter is the precomputed hash value; if it is zero then Perl will
376compute it. The return value is the new hash entry so created. It will be
796b6530 377C<NULL> if the operation failed or if the value did not need to be actually
34a6f7b4
NC
378stored within the hash (as in the case of tied hashes). Otherwise the
379contents of the return value can be accessed using the C<He?> macros
380described here. Note that the caller is responsible for suitably
381incrementing the reference count of C<val> before the call, and
382decrementing it if the function returned NULL. Effectively a successful
796b6530 383C<hv_store_ent> takes ownership of one reference to C<val>. This is
34a6f7b4 384usually what you want; a newly created SV has a reference count of one, so
796b6530 385if all your code does is create SVs then store them in a hash, C<hv_store>
34a6f7b4 386will own the only reference to the new SV, and your code doesn't need to do
796b6530 387anything further to tidy up. Note that C<hv_store_ent> only reads the C<key>;
34a6f7b4 388unlike C<val> it does not take ownership of it, so maintaining the correct
83082a19
KW
389reference count on C<key> is entirely the caller's responsibility. The reason
390it does not take ownership, is that C<key> is not used after this function
391returns, and so can be freed immediately. C<hv_store>
796b6530 392is not implemented as a call to C<hv_store_ent>, and does not create a temporary
34a6f7b4 393SV for the key, so if your key data is not already in SV form then use
796b6530 394C<hv_store> in preference to C<hv_store_ent>.
34a6f7b4
NC
395
396See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
397information on how to use this function on tied hashes.
398
34a6f7b4
NC
399=for apidoc hv_exists
400
401Returns a boolean indicating whether the specified hash key exists. The
a05d6c5d
TC
402absolute value of C<klen> is the length of the key. If C<klen> is
403negative the key is assumed to be in UTF-8-encoded Unicode.
34a6f7b4 404
954c1994
GS
405=for apidoc hv_fetch
406
a05d6c5d
TC
407Returns the SV which corresponds to the specified key in the hash.
408The absolute value of C<klen> is the length of the key. If C<klen> is
409negative the key is assumed to be in UTF-8-encoded Unicode. If
43d3b06a
KW
410C<lval> is set then the fetch will be part of a store. This means that if
411there is no value in the hash associated with the given key, then one is
412created and a pointer to it is returned. The C<SV*> it points to can be
413assigned to. But always check that the
a05d6c5d 414return value is non-null before dereferencing it to an C<SV*>.
954c1994 415
96f1132b 416See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
954c1994
GS
417information on how to use this function on tied hashes.
418
34a6f7b4
NC
419=for apidoc hv_exists_ent
420
db4fbf16
FC
421Returns a boolean indicating whether
422the specified hash key exists. C<hash>
34a6f7b4
NC
423can be a valid precomputed hash value, or 0 to ask for it to be
424computed.
425
426=cut
427*/
428
d1be9408 429/* returns an HE * structure with the all fields set */
fde52b5c 430/* note that hent_val will be a mortal sv for MAGICAL hashes */
954c1994
GS
431/*
432=for apidoc hv_fetch_ent
433
434Returns the hash entry which corresponds to the specified key in the hash.
435C<hash> must be a valid precomputed hash number for the given C<key>, or 0
436if you want the function to compute it. IF C<lval> is set then the fetch
437will be part of a store. Make sure the return value is non-null before
b24b84ef 438accessing it. The return value when C<hv> is a tied hash is a pointer to a
954c1994 439static location, so be sure to make a copy of the structure if you need to
1c846c1f 440store it somewhere.
954c1994 441
96f1132b 442See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more
954c1994
GS
443information on how to use this function on tied hashes.
444
445=cut
446*/
447
a038e571
NC
448/* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */
449void *
450Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32,
1604cfb0 451 const int action, SV *val, const U32 hash)
a038e571
NC
452{
453 STRLEN klen;
454 int flags;
455
7918f24d
NC
456 PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN;
457
a038e571 458 if (klen_i32 < 0) {
1604cfb0
MS
459 klen = -klen_i32;
460 flags = HVhek_UTF8;
a038e571 461 } else {
1604cfb0
MS
462 klen = klen_i32;
463 flags = 0;
a038e571
NC
464 }
465 return hv_common(hv, NULL, key, klen, flags, action, val, hash);
466}
467
63c89345 468void *
d3ba3f5c 469Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
1604cfb0 470 int flags, int action, SV *val, U32 hash)
113738bb 471{
b2c64049 472 XPVHV* xhv;
b2c64049
NC
473 HE *entry;
474 HE **oentry;
fde52b5c 475 SV *sv;
da58a35d 476 bool is_utf8;
6f019ba7 477 bool in_collision;
3c84c864 478 const int return_svp = action & HV_FETCH_JUST_SV;
34dadc62 479 HEK *keysv_hek = NULL;
fde52b5c 480
481 if (!hv)
1604cfb0 482 return NULL;
e4787c0c 483 if (SvTYPE(hv) == (svtype)SVTYPEMASK)
1604cfb0 484 return NULL;
8265e3d1
NC
485
486 assert(SvTYPE(hv) == SVt_PVHV);
fde52b5c 487
bdee33e4 488 if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) {
1604cfb0
MS
489 MAGIC* mg;
490 if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) {
491 struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr;
492 if (uf->uf_set == NULL) {
493 SV* obj = mg->mg_obj;
494
495 if (!keysv) {
496 keysv = newSVpvn_flags(key, klen, SVs_TEMP |
497 ((flags & HVhek_UTF8)
498 ? SVf_UTF8 : 0));
499 }
500
501 mg->mg_obj = keysv; /* pass key */
502 uf->uf_index = action; /* pass action */
503 magic_getuvar(MUTABLE_SV(hv), mg);
504 keysv = mg->mg_obj; /* may have changed */
505 mg->mg_obj = obj;
506
507 /* If the key may have changed, then we need to invalidate
508 any passed-in computed hash value. */
509 hash = 0;
510 }
511 }
bdee33e4 512 }
abb96f57
NC
513
514 /* flags might have HVhek_NOTSHARED set. If so, we need to ignore that.
515 Some callers to hv_common() pass the flags value from an existing HEK,
516 and if that HEK is not shared, then it has the relevant flag bit set,
517 which must not be passed into share_hek_flags().
518
519 It would be "purer" to insist that all callers clear it, but we'll end up
520 with subtle bugs if we leave it to them, or runtime assertion failures if
521 we try to enforce our documentation with landmines.
522
523 If keysv is true, all code paths assign a new value to flags with that
524 bit clear, so we're always "good". Hence we only need to explicitly clear
525 this bit in the else block. */
113738bb 526 if (keysv) {
1604cfb0
MS
527 if (flags & HVhek_FREEKEY)
528 Safefree(key);
529 key = SvPV_const(keysv, klen);
530 is_utf8 = (SvUTF8(keysv) != 0);
531 if (SvIsCOW_shared_hash(keysv)) {
532 flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0);
533 } else {
534 flags = 0;
535 }
113738bb 536 } else {
1604cfb0 537 is_utf8 = cBOOL(flags & HVhek_UTF8);
abb96f57 538 flags &= ~HVhek_NOTSHARED;
113738bb 539 }
113738bb 540
9dbc5603 541 if (action & HV_DELETE) {
1604cfb0
MS
542 return (void *) hv_delete_common(hv, keysv, key, klen,
543 flags | (is_utf8 ? HVhek_UTF8 : 0),
544 action, hash);
9dbc5603
NC
545 }
546
b2c64049 547 xhv = (XPVHV*)SvANY(hv);
7f66fda2 548 if (SvMAGICAL(hv)) {
1604cfb0
MS
549 if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) {
550 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
551 || SvGMAGICAL((const SV *)hv))
552 {
553 /* FIXME should be able to skimp on the HE/HEK here when
554 HV_FETCH_JUST_SV is true. */
555 if (!keysv) {
556 keysv = newSVpvn_utf8(key, klen, is_utf8);
557 } else {
558 keysv = newSVsv(keysv);
559 }
44a2ac75 560 sv = sv_newmortal();
ad64d0ec 561 mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY);
7f66fda2 562
1604cfb0
MS
563 /* grab a fake HE/HEK pair from the pool or make a new one */
564 entry = PL_hv_fetch_ent_mh;
565 if (entry)
566 PL_hv_fetch_ent_mh = HeNEXT(entry);
567 else {
568 char *k;
569 entry = new_HE();
570 Newx(k, HEK_BASESIZE + sizeof(const SV *), char);
571 HeKEY_hek(entry) = (HEK*)k;
572 }
573 HeNEXT(entry) = NULL;
574 HeSVKEY_set(entry, keysv);
575 HeVAL(entry) = sv;
576 sv_upgrade(sv, SVt_PVLV);
577 LvTYPE(sv) = 'T';
578 /* so we can free entry when freeing sv */
579 LvTARG(sv) = MUTABLE_SV(entry);
580
581 /* XXX remove at some point? */
582 if (flags & HVhek_FREEKEY)
583 Safefree(key);
584
585 if (return_svp) {
586 return entry ? (void *) &HeVAL(entry) : NULL;
587 }
588 return (void *) entry;
589 }
7f66fda2 590#ifdef ENV_IS_CASELESS
1604cfb0
MS
591 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
592 U32 i;
593 for (i = 0; i < klen; ++i)
594 if (isLOWER(key[i])) {
595 /* Would be nice if we had a routine to do the
596 copy and upercase in a single pass through. */
597 const char * const nkey = strupr(savepvn(key,klen));
598 /* Note that this fetch is for nkey (the uppercased
599 key) whereas the store is for key (the original) */
600 void *result = hv_common(hv, NULL, nkey, klen,
601 HVhek_FREEKEY, /* free nkey */
602 0 /* non-LVAL fetch */
603 | HV_DISABLE_UVAR_XKEY
604 | return_svp,
605 NULL /* no value */,
606 0 /* compute hash */);
607 if (!result && (action & HV_FETCH_LVALUE)) {
608 /* This call will free key if necessary.
609 Do it this way to encourage compiler to tail
610 call optimise. */
611 result = hv_common(hv, keysv, key, klen, flags,
612 HV_FETCH_ISSTORE
613 | HV_DISABLE_UVAR_XKEY
614 | return_svp,
8fcb2425 615 newSV_type(SVt_NULL), hash);
1604cfb0
MS
616 } else {
617 if (flags & HVhek_FREEKEY)
618 Safefree(key);
619 }
620 return result;
621 }
622 }
7f66fda2 623#endif
1604cfb0
MS
624 } /* ISFETCH */
625 else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) {
626 if (mg_find((const SV *)hv, PERL_MAGIC_tied)
627 || SvGMAGICAL((const SV *)hv)) {
628 /* I don't understand why hv_exists_ent has svret and sv,
629 whereas hv_exists only had one. */
630 SV * const svret = sv_newmortal();
631 sv = sv_newmortal();
632
633 if (keysv || is_utf8) {
634 if (!keysv) {
635 keysv = newSVpvn_utf8(key, klen, TRUE);
636 } else {
637 keysv = newSVsv(keysv);
638 }
639 mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY);
640 } else {
641 mg_copy(MUTABLE_SV(hv), sv, key, klen);
642 }
643 if (flags & HVhek_FREEKEY)
644 Safefree(key);
645 {
c818886e
JH
646 MAGIC * const mg = mg_find(sv, PERL_MAGIC_tiedelem);
647 if (mg)
648 magic_existspack(svret, mg);
1604cfb0
MS
649 }
650 /* This cast somewhat evil, but I'm merely using NULL/
651 not NULL to return the boolean exists.
652 And I know hv is not NULL. */
653 return SvTRUE_NN(svret) ? (void *)hv : NULL;
654 }
7f66fda2 655#ifdef ENV_IS_CASELESS
1604cfb0
MS
656 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
657 /* XXX This code isn't UTF8 clean. */
658 char * const keysave = (char * const)key;
659 /* Will need to free this, so set FREEKEY flag. */
660 key = savepvn(key,klen);
661 key = (const char*)strupr((char*)key);
662 is_utf8 = FALSE;
663 hash = 0;
664 keysv = 0;
665
666 if (flags & HVhek_FREEKEY) {
667 Safefree(keysave);
668 }
669 flags |= HVhek_FREEKEY;
670 }
902173a3 671#endif
1604cfb0
MS
672 } /* ISEXISTS */
673 else if (action & HV_FETCH_ISSTORE) {
674 bool needs_copy;
675 bool needs_store;
676 hv_magic_check (hv, &needs_copy, &needs_store);
677 if (needs_copy) {
678 const bool save_taint = TAINT_get;
679 if (keysv || is_utf8) {
680 if (!keysv) {
681 keysv = newSVpvn_utf8(key, klen, TRUE);
682 }
683 if (TAINTING_get)
684 TAINT_set(SvTAINTED(keysv));
685 keysv = sv_2mortal(newSVsv(keysv));
686 mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY);
687 } else {
688 mg_copy(MUTABLE_SV(hv), val, key, klen);
689 }
690
691 TAINT_IF(save_taint);
9a9b5ec9
DM
692#ifdef NO_TAINT_SUPPORT
693 PERL_UNUSED_VAR(save_taint);
694#endif
1604cfb0
MS
695 if (!needs_store) {
696 if (flags & HVhek_FREEKEY)
697 Safefree(key);
698 return NULL;
699 }
b2c64049 700#ifdef ENV_IS_CASELESS
1604cfb0
MS
701 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
702 /* XXX This code isn't UTF8 clean. */
703 const char *keysave = key;
704 /* Will need to free this, so set FREEKEY flag. */
705 key = savepvn(key,klen);
706 key = (const char*)strupr((char*)key);
707 is_utf8 = FALSE;
708 hash = 0;
709 keysv = 0;
710
711 if (flags & HVhek_FREEKEY) {
712 Safefree(keysave);
713 }
714 flags |= HVhek_FREEKEY;
715 }
b2c64049 716#endif
1604cfb0
MS
717 }
718 } /* ISSTORE */
7f66fda2 719 } /* SvMAGICAL */
fde52b5c 720
7b2c381c 721 if (!HvARRAY(hv)) {
1604cfb0 722 if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE))
fde52b5c 723#ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */
1604cfb0
MS
724 || (SvRMAGICAL((const SV *)hv)
725 && mg_find((const SV *)hv, PERL_MAGIC_env))
fde52b5c 726#endif
1604cfb0
MS
727 ) {
728 char *array;
729 Newxz(array,
730 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
731 char);
732 HvARRAY(hv) = (HE**)array;
733 }
7f66fda2 734#ifdef DYNAMIC_ENV_FETCH
1604cfb0
MS
735 else if (action & HV_FETCH_ISEXISTS) {
736 /* for an %ENV exists, if we do an insert it's by a recursive
737 store call, so avoid creating HvARRAY(hv) right now. */
738 }
7f66fda2 739#endif
1604cfb0
MS
740 else {
741 /* XXX remove at some point? */
113738bb
NC
742 if (flags & HVhek_FREEKEY)
743 Safefree(key);
744
1604cfb0
MS
745 return NULL;
746 }
fde52b5c 747 }
748
37ae23ff 749 if (is_utf8 && !(flags & HVhek_KEYCANONICAL)) {
1604cfb0
MS
750 char * const keysave = (char *)key;
751 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
19692e8d 752 if (is_utf8)
1604cfb0
MS
753 flags |= HVhek_UTF8;
754 else
755 flags &= ~HVhek_UTF8;
7f66fda2 756 if (key != keysave) {
1604cfb0
MS
757 if (flags & HVhek_FREEKEY)
758 Safefree(keysave);
19692e8d 759 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
1604cfb0
MS
760 /* If the caller calculated a hash, it was on the sequence of
761 octets that are the UTF-8 form. We've now changed the sequence
762 of octets stored to that of the equivalent byte representation,
763 so the hash we need is different. */
764 hash = 0;
765 }
19692e8d 766 }
f9a63242 767
34dadc62
DM
768 if (keysv && (SvIsCOW_shared_hash(keysv))) {
769 if (HvSHAREKEYS(hv))
770 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
771 hash = SvSHARED_HASH(keysv);
7dc86639 772 }
34dadc62
DM
773 else if (!hash)
774 PERL_HASH(hash, key, klen);
effa1e2d 775
7f66fda2 776#ifdef DYNAMIC_ENV_FETCH
4608196e 777 if (!HvARRAY(hv)) entry = NULL;
7f66fda2
NC
778 else
779#endif
b2c64049 780 {
1604cfb0 781 entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)];
b2c64049 782 }
34dadc62
DM
783
784 if (!entry)
785 goto not_found;
786
787 if (keysv_hek) {
788 /* keysv is actually a HEK in disguise, so we can match just by
789 * comparing the HEK pointers in the HE chain. There is a slight
790 * caveat: on something like "\x80", which has both plain and utf8
791 * representations, perl's hashes do encoding-insensitive lookups,
792 * but preserve the encoding of the stored key. Thus a particular
793 * key could map to two different HEKs in PL_strtab. We only
794 * conclude 'not found' if all the flags are the same; otherwise
795 * we fall back to a full search (this should only happen in rare
796 * cases).
797 */
798 int keysv_flags = HEK_FLAGS(keysv_hek);
799 HE *orig_entry = entry;
800
801 for (; entry; entry = HeNEXT(entry)) {
802 HEK *hek = HeKEY_hek(entry);
803 if (hek == keysv_hek)
804 goto found;
805 if (HEK_FLAGS(hek) != keysv_flags)
806 break; /* need to do full match */
807 }
808 if (!entry)
809 goto not_found;
810 /* failed on shortcut - do full search loop */
811 entry = orig_entry;
812 }
813
0298d7b9 814 for (; entry; entry = HeNEXT(entry)) {
1604cfb0
MS
815 if (HeHASH(entry) != hash) /* strings can't be equal */
816 continue;
817 if (HeKLEN(entry) != (I32)klen)
818 continue;
819 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
820 continue;
38b26de3 821 if ((HeKFLAGS(entry) ^ flags) & HVhek_UTF8)
1604cfb0 822 continue;
b2c64049 823
34dadc62 824 found:
b2c64049 825 if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) {
38b26de3 826 if ((HeKFLAGS(entry) ^ flags) & HVhek_WASUTF8) {
1604cfb0
MS
827 /* We match if HVhek_UTF8 bit in our flags and hash key's
828 match. But if entry was set previously with HVhek_WASUTF8
829 and key now doesn't (or vice versa) then we should change
830 the key's flag, as this is assignment. */
1db404fc 831 if ((HeKFLAGS(entry) & HVhek_NOTSHARED) == 0) {
1604cfb0
MS
832 /* Need to swap the key we have for a key with the flags we
833 need. As keys are shared we can't just write to the
834 flag, so we share the new one, unshare the old one. */
38b26de3
NC
835 HEK * const new_hek
836 = share_hek_flags(key, klen, hash, flags & ~HVhek_FREEKEY);
1604cfb0
MS
837 unshare_hek (HeKEY_hek(entry));
838 HeKEY_hek(entry) = new_hek;
839 }
840 else if (hv == PL_strtab) {
841 /* PL_strtab is usually the only hash without HvSHAREKEYS,
842 so putting this test here is cheap */
843 if (flags & HVhek_FREEKEY)
844 Safefree(key);
845 Perl_croak(aTHX_ S_strtab_error,
846 action & HV_FETCH_LVALUE ? "fetch" : "store");
847 }
2c94601f
NC
848 else {
849 /* Effectively this is save_hek_flags() for a new version
38b26de3
NC
850 of the HEK and Safefree() of the old rolled together. */
851 HeKFLAGS(entry) ^= HVhek_WASUTF8;
2c94601f 852 }
38b26de3 853 if (flags & HVhek_ENABLEHVKFLAGS)
1604cfb0
MS
854 HvHASKFLAGS_on(hv);
855 }
856 if (HeVAL(entry) == &PL_sv_placeholder) {
857 /* yes, can store into placeholder slot */
858 if (action & HV_FETCH_LVALUE) {
859 if (SvMAGICAL(hv)) {
860 /* This preserves behaviour with the old hv_fetch
861 implementation which at this point would bail out
862 with a break; (at "if we find a placeholder, we
863 pretend we haven't found anything")
864
865 That break mean that if a placeholder were found, it
866 caused a call into hv_store, which in turn would
867 check magic, and if there is no magic end up pretty
868 much back at this point (in hv_store's code). */
869 break;
870 }
871 /* LVAL fetch which actually needs a store. */
8fcb2425 872 val = newSV_type(SVt_NULL);
1604cfb0
MS
873 HvPLACEHOLDERS(hv)--;
874 } else {
875 /* store */
876 if (val != &PL_sv_placeholder)
877 HvPLACEHOLDERS(hv)--;
878 }
879 HeVAL(entry) = val;
880 } else if (action & HV_FETCH_ISSTORE) {
881 SvREFCNT_dec(HeVAL(entry));
882 HeVAL(entry) = val;
883 }
884 } else if (HeVAL(entry) == &PL_sv_placeholder) {
885 /* if we find a placeholder, we pretend we haven't found
886 anything */
887 break;
888 }
889 if (flags & HVhek_FREEKEY)
890 Safefree(key);
891 if (return_svp) {
b66176f0 892 return (void *) &HeVAL(entry);
1604cfb0
MS
893 }
894 return entry;
fde52b5c 895 }
34dadc62
DM
896
897 not_found:
fde52b5c 898#ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */
0ed29950 899 if (!(action & HV_FETCH_ISSTORE)
1604cfb0
MS
900 && SvRMAGICAL((const SV *)hv)
901 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
902 unsigned long len;
903 const char * const env = PerlEnv_ENVgetenv_len(key,&len);
904 if (env) {
905 sv = newSVpvn(env,len);
906 SvTAINTED_on(sv);
907 return hv_common(hv, keysv, key, klen, flags,
908 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
909 sv, hash);
910 }
fde52b5c 911 }
912#endif
7f66fda2
NC
913
914 if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) {
1604cfb0
MS
915 hv_notallowed(flags, key, klen,
916 "Attempt to access disallowed key '%" SVf "' in"
917 " a restricted hash");
1b1f1335 918 }
b2c64049 919 if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) {
1604cfb0
MS
920 /* Not doing some form of store, so return failure. */
921 if (flags & HVhek_FREEKEY)
922 Safefree(key);
923 return NULL;
b2c64049 924 }
113738bb 925 if (action & HV_FETCH_LVALUE) {
8fcb2425 926 val = action & HV_FETCH_EMPTY_HE ? NULL : newSV_type(SVt_NULL);
1604cfb0
MS
927 if (SvMAGICAL(hv)) {
928 /* At this point the old hv_fetch code would call to hv_store,
929 which in turn might do some tied magic. So we need to make that
930 magic check happen. */
931 /* gonna assign to this, so it better be there */
932 /* If a fetch-as-store fails on the fetch, then the action is to
933 recurse once into "hv_store". If we didn't do this, then that
934 recursive call would call the key conversion routine again.
935 However, as we replace the original key with the converted
936 key, this would result in a double conversion, which would show
937 up as a bug if the conversion routine is not idempotent.
938 Hence the use of HV_DISABLE_UVAR_XKEY. */
939 return hv_common(hv, keysv, key, klen, flags,
940 HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp,
941 val, hash);
942 /* XXX Surely that could leak if the fetch-was-store fails?
943 Just like the hv_fetch. */
944 }
113738bb
NC
945 }
946
b2c64049
NC
947 /* Welcome to hv_store... */
948
7b2c381c 949 if (!HvARRAY(hv)) {
1604cfb0
MS
950 /* Not sure if we can get here. I think the only case of oentry being
951 NULL is for %ENV with dynamic env fetch. But that should disappear
952 with magic in the previous code. */
953 char *array;
954 Newxz(array,
955 PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */),
956 char);
957 HvARRAY(hv) = (HE**)array;
b2c64049
NC
958 }
959
7b2c381c 960 oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max];
ab4af705 961
b2c64049
NC
962 /* share_hek_flags will do the free for us. This might be considered
963 bad API design. */
d978f069
NC
964 if (LIKELY(HvSHAREKEYS(hv))) {
965 entry = new_HE();
1604cfb0 966 HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags);
d978f069
NC
967 }
968 else if (UNLIKELY(hv == PL_strtab)) {
1604cfb0
MS
969 /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting
970 this test here is cheap */
971 if (flags & HVhek_FREEKEY)
972 Safefree(key);
973 Perl_croak(aTHX_ S_strtab_error,
974 action & HV_FETCH_LVALUE ? "fetch" : "store");
5d2b1485 975 }
d978f069
NC
976 else {
977 /* gotta do the real thing */
978 entry = new_HE();
1604cfb0 979 HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags);
d978f069 980 }
b2c64049 981 HeVAL(entry) = val;
a977878d 982 in_collision = cBOOL(*oentry != NULL);
3078e109 983
c3c9d6b1 984
6a5b4183 985#ifdef PERL_HASH_RANDOMIZE_KEYS
3078e109
YO
986 /* This logic semi-randomizes the insert order in a bucket.
987 * Either we insert into the top, or the slot below the top,
d5fc06cb
YO
988 * making it harder to see if there is a collision. We also
989 * reset the iterator randomizer if there is one.
3078e109 990 */
c3c9d6b1
YO
991
992
6a5b4183 993 if ( *oentry && PL_HASH_RAND_BITS_ENABLED) {
c3c9d6b1 994 UPDATE_HASH_RAND_BITS_KEY(key,klen);
6a5b4183
YO
995 if ( PL_hash_rand_bits & 1 ) {
996 HeNEXT(entry) = HeNEXT(*oentry);
997 HeNEXT(*oentry) = entry;
998 } else {
999 HeNEXT(entry) = *oentry;
1000 *oentry = entry;
1001 }
1002 } else
1003#endif
1004 {
3078e109
YO
1005 HeNEXT(entry) = *oentry;
1006 *oentry = entry;
3078e109 1007 }
6a5b4183 1008#ifdef PERL_HASH_RANDOMIZE_KEYS
53083cad 1009 if (HvHasAUX(hv)) {
ff20b672
YO
1010 /* Currently this makes various tests warn in annoying ways.
1011 * So Silenced for now. - Yves | bogus end of comment =>* /
1012 if (HvAUX(hv)->xhv_riter != -1) {
1013 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
1014 "[TESTING] Inserting into a hash during each() traversal results in undefined behavior"
1015 pTHX__FORMAT
1016 pTHX__VALUE);
1017 }
1018 */
c3c9d6b1 1019 MAYBE_UPDATE_HASH_RAND_BITS_KEY(key,klen);
3a714294
YO
1020 HvAUX(hv)->xhv_rand= (U32)PL_hash_rand_bits;
1021 }
6a5b4183 1022#endif
b2c64049
NC
1023
1024 if (val == &PL_sv_placeholder)
1604cfb0 1025 HvPLACEHOLDERS(hv)++;
38b26de3 1026 if (flags & HVhek_ENABLEHVKFLAGS)
1604cfb0 1027 HvHASKFLAGS_on(hv);
b2c64049 1028
8e317198 1029 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
6f019ba7 1030 if ( in_collision && DO_HSPLIT(xhv) ) {
adf6906b 1031 const STRLEN oldsize = xhv->xhv_max + 1;
81a3ba35 1032 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
adf6906b 1033
81a3ba35 1034 if (items /* hash has placeholders */
1eaee784
NC
1035 && !SvREADONLY(hv) /* but is not a restricted hash */) {
1036 /* If this hash previously was a "restricted hash" and had
1037 placeholders, but the "restricted" flag has been turned off,
1038 then the placeholders no longer serve any useful purpose.
1039 However, they have the downsides of taking up RAM, and adding
1040 extra steps when finding used values. It's safe to clear them
1041 at this point, even though Storable rebuilds restricted hashes by
0ca1b5c3 1042 putting in all the placeholders (first) before turning on the
1eaee784
NC
1043 readonly flag, because Storable always pre-splits the hash.
1044 If we're lucky, then we may clear sufficient placeholders to
1045 avoid needing to split the hash at all. */
81a3ba35 1046 clear_placeholders(hv, items);
1eaee784
NC
1047 if (DO_HSPLIT(xhv))
1048 hsplit(hv, oldsize, oldsize * 2);
1049 } else
1050 hsplit(hv, oldsize, oldsize * 2);
fde52b5c 1051 }
b2c64049 1052
3c84c864 1053 if (return_svp) {
1604cfb0 1054 return entry ? (void *) &HeVAL(entry) : NULL;
3c84c864
NC
1055 }
1056 return (void *) entry;
fde52b5c 1057}
1058
864dbfa3 1059STATIC void
b0e6ae5b 1060S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store)
d0066dc7 1061{
a3b680e6 1062 const MAGIC *mg = SvMAGIC(hv);
7918f24d
NC
1063
1064 PERL_ARGS_ASSERT_HV_MAGIC_CHECK;
1065
d0066dc7
OT
1066 *needs_copy = FALSE;
1067 *needs_store = TRUE;
1068 while (mg) {
1604cfb0
MS
1069 if (isUPPER(mg->mg_type)) {
1070 *needs_copy = TRUE;
1071 if (mg->mg_type == PERL_MAGIC_tied) {
1072 *needs_store = FALSE;
1073 return; /* We've set all there is to set. */
1074 }
1075 }
1076 mg = mg->mg_moremagic;
d0066dc7
OT
1077 }
1078}
1079
954c1994 1080/*
a3bcc51e
TP
1081=for apidoc hv_scalar
1082
8bf4c401
YO
1083Evaluates the hash in scalar context and returns the result.
1084
1085When the hash is tied dispatches through to the SCALAR method,
1086otherwise returns a mortal SV containing the number of keys
1087in the hash.
1088
1089Note, prior to 5.25 this function returned what is now
1090returned by the hv_bucket_ratio() function.
a3bcc51e
TP
1091
1092=cut
1093*/
1094
1095SV *
1096Perl_hv_scalar(pTHX_ HV *hv)
1097{
a3bcc51e 1098 SV *sv;
7ea8b04b 1099 UV u;
823a54a3 1100
7918f24d
NC
1101 PERL_ARGS_ASSERT_HV_SCALAR;
1102
823a54a3 1103 if (SvRMAGICAL(hv)) {
1604cfb0
MS
1104 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
1105 if (mg)
1106 return magic_scalarpack(hv, mg);
823a54a3 1107 }
a3bcc51e 1108
7ea8b04b
RL
1109 sv = newSV_type_mortal(SVt_IV);
1110
1111 /* Inlined sv_setuv(sv, HvUSEDKEYS(hv)) follows:*/
1112 u = HvUSEDKEYS(hv);
1113
1114 if (u <= (UV)IV_MAX) {
1115 SvIV_set(sv, (IV)u);
1116 (void)SvIOK_only(sv);
1117 SvTAINT(sv);
1118 } else {
1119 SvIV_set(sv, 0);
1120 SvUV_set(sv, u);
1121 (void)SvIOK_only_UV(sv);
1122 SvTAINT(sv);
1123 }
8bf4c401
YO
1124
1125 return sv;
1126}
1127
af3b1cba
DM
1128
1129/*
8dc9003f
DM
1130hv_pushkv(): push all the keys and/or values of a hash onto the stack.
1131The rough Perl equivalents:
1132 () = %hash;
1133 () = keys %hash;
1134 () = values %hash;
1135
af3b1cba 1136Resets the hash's iterator.
8dc9003f
DM
1137
1138flags : 1 = push keys
1139 2 = push values
1140 1|2 = push keys and values
1141 XXX use symbolic flag constants at some point?
1142I might unroll the non-tied hv_iternext() in here at some point - DAPM
af3b1cba
DM
1143*/
1144
1145void
8dc9003f 1146Perl_hv_pushkv(pTHX_ HV *hv, U32 flags)
af3b1cba
DM
1147{
1148 HE *entry;
61084eef
CB
1149 bool tied = SvRMAGICAL(hv) && (mg_find(MUTABLE_SV(hv), PERL_MAGIC_tied)
1150#ifdef DYNAMIC_ENV_FETCH /* might not know number of keys yet */
1151 || mg_find(MUTABLE_SV(hv), PERL_MAGIC_env)
1152#endif
1153 );
af3b1cba
DM
1154 dSP;
1155
1156 PERL_ARGS_ASSERT_HV_PUSHKV;
8dc9003f 1157 assert(flags); /* must be pushing at least one of keys and values */
af3b1cba
DM
1158
1159 (void)hv_iterinit(hv);
1160
1161 if (tied) {
8dc9003f 1162 SSize_t ext = (flags == 3) ? 2 : 1;
af3b1cba 1163 while ((entry = hv_iternext(hv))) {
8dc9003f
DM
1164 EXTEND(SP, ext);
1165 if (flags & 1)
1166 PUSHs(hv_iterkeysv(entry));
1167 if (flags & 2)
1168 PUSHs(hv_iterval(hv, entry));
af3b1cba
DM
1169 }
1170 }
1171 else {
87c9dcef 1172 Size_t nkeys = HvUSEDKEYS(hv);
8dc9003f
DM
1173 SSize_t ext;
1174
1175 if (!nkeys)
1176 return;
1177
87c9dcef
DM
1178 /* 2*nkeys() should never be big enough to truncate or wrap */
1179 assert(nkeys <= (SSize_t_MAX >> 1));
8dc9003f 1180 ext = nkeys * ((flags == 3) ? 2 : 1);
87c9dcef
DM
1181
1182 EXTEND_MORTAL(nkeys);
8dc9003f 1183 EXTEND(SP, ext);
af3b1cba
DM
1184
1185 while ((entry = hv_iternext(hv))) {
8dc9003f
DM
1186 if (flags & 1) {
1187 SV *keysv = newSVhek(HeKEY_hek(entry));
1188 SvTEMP_on(keysv);
1189 PL_tmps_stack[++PL_tmps_ix] = keysv;
1190 PUSHs(keysv);
1191 }
1192 if (flags & 2)
1193 PUSHs(HeVAL(entry));
af3b1cba
DM
1194 }
1195 }
1196
1197 PUTBACK;
1198}
1199
1200
8bf4c401 1201/*
7d7345cf 1202=for apidoc hv_bucket_ratio
8bf4c401
YO
1203
1204If the hash is tied dispatches through to the SCALAR tied method,
1205otherwise if the hash contains no keys returns 0, otherwise returns
1206a mortal sv containing a string specifying the number of used buckets,
1207followed by a slash, followed by the number of available buckets.
1208
1209This function is expensive, it must scan all of the buckets
1210to determine which are used, and the count is NOT cached.
1211In a large hash this could be a lot of buckets.
1212
1213=cut
1214*/
1215
1216SV *
1217Perl_hv_bucket_ratio(pTHX_ HV *hv)
1218{
1219 SV *sv;
1220
1221 PERL_ARGS_ASSERT_HV_BUCKET_RATIO;
1222
1223 if (SvRMAGICAL(hv)) {
1224 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied);
1225 if (mg)
1226 return magic_scalarpack(hv, mg);
1227 }
1228
725c44f9
DM
1229 if (HvUSEDKEYS((HV *)hv)) {
1230 sv = sv_newmortal();
a3bcc51e
TP
1231 Perl_sv_setpvf(aTHX_ sv, "%ld/%ld",
1232 (long)HvFILL(hv), (long)HvMAX(hv) + 1);
725c44f9 1233 }
a3bcc51e 1234 else
725c44f9 1235 sv = &PL_sv_zero;
a3bcc51e
TP
1236
1237 return sv;
1238}
1239
1240/*
954c1994
GS
1241=for apidoc hv_delete
1242
a05d6c5d
TC
1243Deletes a key/value pair in the hash. The value's SV is removed from
1244the hash, made mortal, and returned to the caller. The absolute
1245value of C<klen> is the length of the key. If C<klen> is negative the
1246key is assumed to be in UTF-8-encoded Unicode. The C<flags> value
796b6530
KW
1247will normally be zero; if set to C<G_DISCARD> then C<NULL> will be returned.
1248C<NULL> will also be returned if the key is not found.
954c1994 1249
954c1994
GS
1250=for apidoc hv_delete_ent
1251
3025a2e4
CS
1252Deletes a key/value pair in the hash. The value SV is removed from the hash,
1253made mortal, and returned to the caller. The C<flags> value will normally be
796b6530
KW
1254zero; if set to C<G_DISCARD> then C<NULL> will be returned. C<NULL> will also
1255be returned if the key is not found. C<hash> can be a valid precomputed hash
3025a2e4 1256value, or 0 to ask for it to be computed.
954c1994
GS
1257
1258=cut
1259*/
1260
8f8d40ab 1261STATIC SV *
cd6d36ac 1262S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
1604cfb0 1263 int k_flags, I32 d_flags, U32 hash)
f1317c8d 1264{
eb578fdb
KW
1265 XPVHV* xhv;
1266 HE *entry;
1267 HE **oentry;
34dadc62 1268 HE **first_entry;
8298454c 1269 bool is_utf8 = cBOOL(k_flags & HVhek_UTF8);
34dadc62
DM
1270 HEK *keysv_hek = NULL;
1271 U8 mro_changes = 0; /* 1 = isa; 2 = package moved */
1272 SV *sv;
1273 GV *gv = NULL;
1274 HV *stash = NULL;
1c846c1f 1275
307a07c2 1276 if (SvMAGICAL(hv)) {
1604cfb0
MS
1277 bool needs_copy;
1278 bool needs_store;
1279 hv_magic_check (hv, &needs_copy, &needs_store);
1280
1281 if (needs_copy) {
1282 SV *sv;
1283 entry = (HE *) hv_common(hv, keysv, key, klen,
1284 k_flags & ~HVhek_FREEKEY,
1285 HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY,
1286 NULL, hash);
1287 sv = entry ? HeVAL(entry) : NULL;
1288 if (sv) {
1289 if (SvMAGICAL(sv)) {
1290 mg_clear(sv);
1291 }
1292 if (!needs_store) {
1293 if (mg_find(sv, PERL_MAGIC_tiedelem)) {
1294 /* No longer an element */
1295 sv_unmagic(sv, PERL_MAGIC_tiedelem);
1296 return sv;
1297 }
1298 return NULL; /* element cannot be deleted */
1299 }
902173a3 1300#ifdef ENV_IS_CASELESS
1604cfb0
MS
1301 else if (mg_find((const SV *)hv, PERL_MAGIC_env)) {
1302 /* XXX This code isn't UTF8 clean. */
1303 keysv = newSVpvn_flags(key, klen, SVs_TEMP);
1304 if (k_flags & HVhek_FREEKEY) {
1305 Safefree(key);
1306 }
1307 key = strupr(SvPVX(keysv));
1308 is_utf8 = 0;
1309 k_flags = 0;
1310 hash = 0;
1311 }
510ac311 1312#endif
1604cfb0
MS
1313 }
1314 }
fde52b5c 1315 }
cbec9347 1316 xhv = (XPVHV*)SvANY(hv);
ec7598c6 1317 if (!HvTOTALKEYS(hv))
1604cfb0 1318 return NULL;
fde52b5c 1319
6b230254 1320 if (is_utf8 && !(k_flags & HVhek_KEYCANONICAL)) {
1604cfb0
MS
1321 const char * const keysave = key;
1322 key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8);
cd6d36ac 1323
19692e8d 1324 if (is_utf8)
cd6d36ac 1325 k_flags |= HVhek_UTF8;
1604cfb0 1326 else
cd6d36ac 1327 k_flags &= ~HVhek_UTF8;
7f66fda2 1328 if (key != keysave) {
1604cfb0
MS
1329 if (k_flags & HVhek_FREEKEY) {
1330 /* This shouldn't happen if our caller does what we expect,
1331 but strictly the API allows it. */
1332 Safefree(keysave);
1333 }
1334 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
1335 }
19692e8d 1336 }
f9a63242 1337
34dadc62
DM
1338 if (keysv && (SvIsCOW_shared_hash(keysv))) {
1339 if (HvSHAREKEYS(hv))
1340 keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv));
1341 hash = SvSHARED_HASH(keysv);
7dc86639 1342 }
34dadc62
DM
1343 else if (!hash)
1344 PERL_HASH(hash, key, klen);
fde52b5c 1345
9faf471a 1346 first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)];
fde52b5c 1347 entry = *oentry;
0c3bb3c2 1348
34dadc62
DM
1349 if (!entry)
1350 goto not_found;
1351
1352 if (keysv_hek) {
1353 /* keysv is actually a HEK in disguise, so we can match just by
1354 * comparing the HEK pointers in the HE chain. There is a slight
1355 * caveat: on something like "\x80", which has both plain and utf8
1356 * representations, perl's hashes do encoding-insensitive lookups,
1357 * but preserve the encoding of the stored key. Thus a particular
1358 * key could map to two different HEKs in PL_strtab. We only
1359 * conclude 'not found' if all the flags are the same; otherwise
1360 * we fall back to a full search (this should only happen in rare
1361 * cases).
1362 */
1363 int keysv_flags = HEK_FLAGS(keysv_hek);
1364
1365 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1366 HEK *hek = HeKEY_hek(entry);
1367 if (hek == keysv_hek)
1368 goto found;
1369 if (HEK_FLAGS(hek) != keysv_flags)
1370 break; /* need to do full match */
1371 }
1372 if (!entry)
1373 goto not_found;
1374 /* failed on shortcut - do full search loop */
1375 oentry = first_entry;
1376 entry = *oentry;
1377 }
1378
1379 for (; entry; oentry = &HeNEXT(entry), entry = *oentry) {
1604cfb0
MS
1380 if (HeHASH(entry) != hash) /* strings can't be equal */
1381 continue;
1382 if (HeKLEN(entry) != (I32)klen)
1383 continue;
1384 if (memNE(HeKEY(entry),key,klen)) /* is this it? */
1385 continue;
38b26de3 1386 if ((HeKFLAGS(entry) ^ k_flags) & HVhek_UTF8)
1604cfb0 1387 continue;
8aacddc1 1388
34dadc62 1389 found:
1604cfb0
MS
1390 if (hv == PL_strtab) {
1391 if (k_flags & HVhek_FREEKEY)
1392 Safefree(key);
1393 Perl_croak(aTHX_ S_strtab_error, "delete");
1394 }
1395
d15612fe
NC
1396 sv = HeVAL(entry);
1397
1604cfb0 1398 /* if placeholder is here, it's already been deleted.... */
d15612fe 1399 if (sv == &PL_sv_placeholder) {
1604cfb0
MS
1400 if (k_flags & HVhek_FREEKEY)
1401 Safefree(key);
1402 return NULL;
1403 }
d15612fe 1404 if (SvREADONLY(hv) && sv && SvREADONLY(sv)) {
1604cfb0
MS
1405 hv_notallowed(k_flags, key, klen,
1406 "Attempt to delete readonly key '%" SVf "' from"
1407 " a restricted hash");
1408 }
8aacddc1 1409
d61c34dd
NC
1410 /*
1411 * If a restricted hash, rather than really deleting the entry, put
1412 * a placeholder there. This marks the key as being "approved", so
1413 * we can still access via not-really-existing key without raising
1414 * an error.
1415 */
1416 if (SvREADONLY(hv)) {
1417 /* We'll be saving this slot, so the number of allocated keys
1418 * doesn't go down, but the number placeholders goes up */
1419 HeVAL(entry) = &PL_sv_placeholder;
1420 HvPLACEHOLDERS(hv)++;
1421 }
1422 else {
1423 HeVAL(entry) = NULL;
1424 *oentry = HeNEXT(entry);
53083cad 1425 if (HvHasAUX(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */) {
d61c34dd
NC
1426 HvLAZYDEL_on(hv);
1427 }
1428 else {
53083cad 1429 if (HvHasAUX(hv) && HvLAZYDEL(hv) &&
d61c34dd
NC
1430 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
1431 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
59830266 1432 hv_free_ent(NULL, entry);
d61c34dd
NC
1433 }
1434 xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */
1435 if (xhv->xhv_keys == 0)
1436 HvHASKFLAGS_off(hv);
1437 }
1438
1604cfb0
MS
1439 /* If this is a stash and the key ends with ::, then someone is
1440 * deleting a package.
1441 */
638b4d90 1442 if (sv && SvTYPE(sv) == SVt_PVGV && HvENAME_get(hv)) {
d15612fe 1443 gv = (GV *)sv;
1604cfb0
MS
1444 if ((
1445 (klen > 1 && key[klen-2] == ':' && key[klen-1] == ':')
1446 ||
1447 (klen == 1 && key[0] == ':')
1448 )
1449 && (klen != 6 || hv!=PL_defstash || memNE(key,"main::",6))
638b4d90 1450 && (stash = GvHV((GV *)gv))
1604cfb0
MS
1451 && HvENAME_get(stash)) {
1452 /* A previous version of this code checked that the
1453 * GV was still in the symbol table by fetching the
1454 * GV with its name. That is not necessary (and
1455 * sometimes incorrect), as HvENAME cannot be set
1456 * on hv if it is not in the symtab. */
1457 mro_changes = 2;
1458 /* Hang on to it for a bit. */
1459 SvREFCNT_inc_simple_void_NN(
1460 sv_2mortal((SV *)gv)
1461 );
1462 }
1463 else if (memEQs(key, klen, "ISA") && GvAV(gv)) {
6146d9e1
TC
1464 AV *isa = GvAV(gv);
1465 MAGIC *mg = mg_find((SV*)isa, PERL_MAGIC_isa);
1466
1604cfb0 1467 mro_changes = 1;
6146d9e1
TC
1468 if (mg) {
1469 if (mg->mg_obj == (SV*)gv) {
1470 /* This is the only stash this ISA was used for.
1471 * The isaelem magic asserts if there's no
1472 * isa magic on the array, so explicitly
1473 * remove the magic on both the array and its
1474 * elements. @ISA shouldn't be /too/ large.
1475 */
1476 SV **svp, **end;
1477 strip_magic:
1478 svp = AvARRAY(isa);
f67cdee3
TC
1479 if (svp) {
1480 end = svp + (AvFILLp(isa)+1);
1481 while (svp < end) {
1482 if (*svp)
1483 mg_free_type(*svp, PERL_MAGIC_isaelem);
1484 ++svp;
1485 }
6146d9e1
TC
1486 }
1487 mg_free_type((SV*)GvAV(gv), PERL_MAGIC_isa);
1488 }
1489 else {
1490 /* mg_obj is an array of stashes
1491 Note that the array doesn't keep a reference
1492 count on the stashes.
1493 */
1494 AV *av = (AV*)mg->mg_obj;
1495 SV **svp, **arrayp;
1496 SSize_t index;
1497 SSize_t items;
1498
1499 assert(SvTYPE(mg->mg_obj) == SVt_PVAV);
1500
1501 /* remove the stash from the magic array */
1502 arrayp = svp = AvARRAY(av);
1503 items = AvFILLp(av) + 1;
1504 if (items == 1) {
1505 assert(*arrayp == (SV *)gv);
1506 mg->mg_obj = NULL;
1507 /* avoid a double free on the last stash */
1508 AvFILLp(av) = -1;
1509 /* The magic isn't MGf_REFCOUNTED, so release
1510 * the array manually.
1511 */
1512 SvREFCNT_dec_NN(av);
1513 goto strip_magic;
1514 }
1515 else {
1516 while (items--) {
1517 if (*svp == (SV*)gv)
1518 break;
1519 ++svp;
1520 }
1521 index = svp - arrayp;
1522 assert(index >= 0 && index <= AvFILLp(av));
1523 if (index < AvFILLp(av)) {
1524 arrayp[index] = arrayp[AvFILLp(av)];
1525 }
1526 arrayp[AvFILLp(av)] = NULL;
1527 --AvFILLp(av);
1528 }
1529 }
1530 }
1531 }
1604cfb0
MS
1532 }
1533
f15a20bc
NC
1534 if (k_flags & HVhek_FREEKEY)
1535 Safefree(key);
1536
73ace1cb
NC
1537 if (sv) {
1538 /* deletion of method from stash */
1539 if (isGV(sv) && isGV_with_GP(sv) && GvCVu(sv)
1540 && HvENAME_get(hv))
1541 mro_method_changed_in(hv);
73ace1cb 1542
d15612fe
NC
1543 if (d_flags & G_DISCARD) {
1544 SvREFCNT_dec(sv);
1545 sv = NULL;
1546 }
1547 else {
1548 sv_2mortal(sv);
1549 }
1604cfb0
MS
1550 }
1551
1552 if (mro_changes == 1) mro_isa_changed_in(hv);
1553 else if (mro_changes == 2)
1554 mro_package_moved(NULL, stash, gv, 1);
1555
1556 return sv;
79072805 1557 }
34dadc62
DM
1558
1559 not_found:
8aacddc1 1560 if (SvREADONLY(hv)) {
1604cfb0
MS
1561 hv_notallowed(k_flags, key, klen,
1562 "Attempt to delete disallowed key '%" SVf "' from"
1563 " a restricted hash");
8aacddc1
NIS
1564 }
1565
19692e8d 1566 if (k_flags & HVhek_FREEKEY)
1604cfb0 1567 Safefree(key);
a0714e2c 1568 return NULL;
79072805
LW
1569}
1570
f9c625b2
NC
1571/* HVs are used for (at least) three things
1572 1) objects
1573 2) symbol tables
1574 3) associative arrays
1575
1576 shared hash keys benefit the first two greatly, because keys are likely
1577 to be re-used between objects, or for constants in the optree
1578
1579 However, for large associative arrays (lookup tables, "seen" hashes) keys are
1580 unlikely to be re-used. Hence having those keys in the shared string table as
1581 well as the hash is a memory hit, if they are never actually shared with a
1582 second hash. Hence we turn off shared hash keys if a (regular) hash gets
1583 large.
1584
1585 This is a heuristic. There might be a better answer than 42, but for now
1586 we'll use it.
d4c01272
YO
1587
1588 NOTE: Configure with -Accflags='-DPERL_USE_UNSHARED_KEYS_IN_LARGE_HASHES'
1589 to enable this new functionality.
f9c625b2 1590*/
d4c01272
YO
1591
1592#ifdef PERL_USE_UNSHARED_KEYS_IN_LARGE_HASHES
f9c625b2
NC
1593static bool
1594S_large_hash_heuristic(pTHX_ HV *hv, STRLEN size) {
1595 if (size > 42
1596 && !SvOBJECT(hv)
53083cad 1597 && !(HvHasAUX(hv) && HvENAME_get(hv))) {
f9c625b2
NC
1598 /* This hash appears to be growing quite large.
1599 We gamble that it is not sharing keys with other hashes. */
1600 return TRUE;
1601 }
1602 return FALSE;
1603}
d4c01272 1604#endif
32dfa2a7 1605
76e3520e 1606STATIC void
adf6906b 1607S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize)
79072805 1608{
7663aa67 1609 STRLEN i = 0;
7b2c381c 1610 char *a = (char*) HvARRAY(hv);
eb578fdb 1611 HE **aep;
79072805 1612
32dfa2a7 1613 PERL_ARGS_ASSERT_HSPLIT;
aae087f7
YO
1614 if (newsize > MAX_BUCKET_MAX+1)
1615 return;
18026298 1616
3280af22 1617 PL_nomemok = TRUE;
53a41f9c 1618 Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
32dfa2a7 1619 PL_nomemok = FALSE;
422a93e5 1620 if (!a) {
422a93e5
GA
1621 return;
1622 }
32dfa2a7 1623
6a5b4183 1624#ifdef PERL_HASH_RANDOMIZE_KEYS
3078e109
YO
1625 /* the idea of this is that we create a "random" value by hashing the address of
1626 * the array, we then use the low bit to decide if we insert at the top, or insert
1627 * second from top. After each such insert we rotate the hashed value. So we can
1628 * use the same hashed value over and over, and in normal build environments use
1629 * very few ops to do so. ROTL32() should produce a single machine operation. */
c3c9d6b1 1630 MAYBE_UPDATE_HASH_RAND_BITS();
6a5b4183 1631#endif
32dfa2a7
YO
1632 HvARRAY(hv) = (HE**) a;
1633 HvMAX(hv) = newsize - 1;
32dfa2a7 1634 /* now we can safely clear the second half */
72311751 1635 Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/
79072805 1636
68303b5c
NC
1637 if (!HvTOTALKEYS(hv)) /* skip rest if no entries */
1638 return;
1639
f9c625b2 1640 /* don't share keys in large simple hashes */
d4c01272 1641 if (LARGE_HASH_HEURISTIC(hv, HvTOTALKEYS(hv)))
f9c625b2
NC
1642 HvSHAREKEYS_off(hv);
1643
1644
32dfa2a7 1645 newsize--;
68303b5c 1646 aep = (HE**)a;
7663aa67 1647 do {
1604cfb0
MS
1648 HE **oentry = aep + i;
1649 HE *entry = aep[i];
4b5190b5 1650
1604cfb0
MS
1651 if (!entry) /* non-existent */
1652 continue;
1653 do {
c23dc12b 1654 U32 j = (HeHASH(entry) & newsize);
1604cfb0
MS
1655 if (j != (U32)i) {
1656 *oentry = HeNEXT(entry);
6a5b4183
YO
1657#ifdef PERL_HASH_RANDOMIZE_KEYS
1658 /* if the target cell is empty or PL_HASH_RAND_BITS_ENABLED is false
1659 * insert to top, otherwise rotate the bucket rand 1 bit,
1660 * and use the new low bit to decide if we insert at top,
1661 * or next from top. IOW, we only rotate on a collision.*/
1662 if (aep[j] && PL_HASH_RAND_BITS_ENABLED) {
c3c9d6b1 1663 UPDATE_HASH_RAND_BITS();
6a5b4183
YO
1664 if (PL_hash_rand_bits & 1) {
1665 HeNEXT(entry)= HeNEXT(aep[j]);
1666 HeNEXT(aep[j])= entry;
1667 } else {
1668 /* Note, this is structured in such a way as the optimizer
1669 * should eliminate the duplicated code here and below without
1670 * us needing to explicitly use a goto. */
1671 HeNEXT(entry) = aep[j];
1672 aep[j] = entry;
1673 }
1674 } else
1675#endif
1676 {
1677 /* see comment above about duplicated code */
3078e109
YO
1678 HeNEXT(entry) = aep[j];
1679 aep[j] = entry;
3078e109 1680 }
1604cfb0
MS
1681 }
1682 else {
1683 oentry = &HeNEXT(entry);
1684 }
1685 entry = *oentry;
1686 } while (entry);
7663aa67 1687 } while (i++ < oldsize);
79072805
LW
1688}
1689
06c12f4e
KW
1690/*
1691=for apidoc hv_ksplit
1692
1693Attempt to grow the hash C<hv> so it has at least C<newmax> buckets available.
1694Perl chooses the actual number for its convenience.
1695
1696This is the same as doing the following in Perl code:
1697
1698 keys %hv = newmax;
1699
1700=cut
1701*/
1702
72940dca 1703void
864dbfa3 1704Perl_hv_ksplit(pTHX_ HV *hv, IV newmax)
72940dca 1705{
eb578fdb 1706 XPVHV* xhv = (XPVHV*)SvANY(hv);
6f019ba7 1707 const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 */
eb578fdb 1708 I32 newsize;
6f019ba7
YO
1709 I32 wantsize;
1710 I32 trysize;
eb578fdb 1711 char *a;
72940dca 1712
7918f24d
NC
1713 PERL_ARGS_ASSERT_HV_KSPLIT;
1714
6f019ba7
YO
1715 wantsize = (I32) newmax; /* possible truncation here */
1716 if (wantsize != newmax)
1604cfb0 1717 return;
6f019ba7
YO
1718
1719 wantsize= wantsize + (wantsize >> 1); /* wantsize *= 1.5 */
1720 if (wantsize < newmax) /* overflow detection */
1721 return;
1722
1723 newsize = oldsize;
1724 while (wantsize > newsize) {
1725 trysize = newsize << 1;
1726 if (trysize > newsize) {
1727 newsize = trysize;
1728 } else {
1729 /* we overflowed */
1730 return;
1731 }
72940dca 1732 }
6f019ba7
YO
1733
1734 if (newsize <= oldsize)
1735 return; /* overflow detection */
72940dca 1736
7b2c381c 1737 a = (char *) HvARRAY(hv);
e8c10cf3 1738 if (a) {
0095ebd8 1739#ifdef PERL_HASH_RANDOMIZE_KEYS
53083cad 1740 U32 was_ook = HvHasAUX(hv);
0095ebd8 1741#endif
e8c10cf3 1742 hsplit(hv, oldsize, newsize);
0095ebd8 1743#ifdef PERL_HASH_RANDOMIZE_KEYS
53083cad 1744 if (was_ook && HvHasAUX(hv) && HvTOTALKEYS(hv)) {
c3c9d6b1 1745 MAYBE_UPDATE_HASH_RAND_BITS();
0095ebd8
NC
1746 HvAUX(hv)->xhv_rand = (U32)PL_hash_rand_bits;
1747 }
1748#endif
e8c10cf3 1749 } else {
d4c01272 1750 if (LARGE_HASH_HEURISTIC(hv, newmax))
f9c625b2 1751 HvSHAREKEYS_off(hv);
0df05616 1752 Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char);
6f019ba7 1753 xhv->xhv_max = newsize - 1;
0df05616 1754 HvARRAY(hv) = (HE **) a;
72940dca 1755 }
1756}
1757
f6bb1c88
YO
1758/* IMO this should also handle cases where hv_max is smaller than hv_keys
1759 * as tied hashes could play silly buggers and mess us around. We will
1760 * do the right thing during hv_store() afterwards, but still - Yves */
1761#define HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys) STMT_START {\
1762 /* Can we use fewer buckets? (hv_max is always 2^n-1) */ \
1763 if (hv_max < PERL_HASH_DEFAULT_HvMAX) { \
1764 hv_max = PERL_HASH_DEFAULT_HvMAX; \
1765 } else { \
1766 while (hv_max > PERL_HASH_DEFAULT_HvMAX && hv_max + 1 >= hv_keys * 2) \
1767 hv_max = hv_max / 2; \
1768 } \
1769 HvMAX(hv) = hv_max; \
1770} STMT_END
1771
1772
62f3b341
KW
1773/*
1774=for apidoc newHVhv
1775
1776The content of C<ohv> is copied to a new hash. A pointer to the new hash is
1777returned.
1778
1779=cut
1780*/
1781
b3ac6de7 1782HV *
864dbfa3 1783Perl_newHVhv(pTHX_ HV *ohv)
b3ac6de7 1784{
9d4ba2ae 1785 HV * const hv = newHV();
f4431c56 1786 STRLEN hv_max;
4beac62f 1787
3f4d1d78 1788 if (!ohv || (!HvTOTALKEYS(ohv) && !SvMAGICAL((const SV *)ohv)))
1604cfb0 1789 return hv;
4beac62f 1790 hv_max = HvMAX(ohv);
b3ac6de7 1791
ad64d0ec 1792 if (!SvMAGICAL((const SV *)ohv)) {
1604cfb0
MS
1793 /* It's an ordinary hash, so copy it fast. AMS 20010804 */
1794 STRLEN i;
1604cfb0
MS
1795 HE **ents, ** const oents = (HE **)HvARRAY(ohv);
1796 char *a;
1797 Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char);
1798 ents = (HE**)a;
1799
2c94601f 1800 if (HvSHAREKEYS(ohv)) {
33042aaf
NC
1801#ifdef NODEFAULT_SHAREKEYS
1802 HvSHAREKEYS_on(hv);
1803#else
7c4cc034
NC
1804 /* Shared is the default - it should have been set by newHV(). */
1805 assert(HvSHAREKEYS(hv));
33042aaf 1806#endif
7c4cc034
NC
1807 }
1808 else {
1809 HvSHAREKEYS_off(hv);
1810 }
1811
1604cfb0
MS
1812 /* In each bucket... */
1813 for (i = 0; i <= hv_max; i++) {
1814 HE *prev = NULL;
1815 HE *oent = oents[i];
1816
1817 if (!oent) {
1818 ents[i] = NULL;
1819 continue;
1820 }
1821
1822 /* Copy the linked list of entries. */
1823 for (; oent; oent = HeNEXT(oent)) {
1604cfb0
MS
1824 HE * const ent = new_HE();
1825 SV *const val = HeVAL(oent);
2c94601f 1826 const int flags = HeKFLAGS(oent);
1604cfb0
MS
1827
1828 HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val);
1db404fc 1829 if ((flags & HVhek_NOTSHARED) == 0) {
a31d0367
NC
1830 HeKEY_hek(ent) = share_hek_hek(HeKEY_hek(oent));
1831 }
1832 else {
1833 const U32 hash = HeHASH(oent);
1834 const char * const key = HeKEY(oent);
1835 const STRLEN len = HeKLEN(oent);
a31d0367
NC
1836 HeKEY_hek(ent) = save_hek_flags(key, len, hash, flags);
1837 }
1604cfb0
MS
1838 if (prev)
1839 HeNEXT(prev) = ent;
1840 else
1841 ents[i] = ent;
1842 prev = ent;
1843 HeNEXT(ent) = NULL;
1844 }
1845 }
1846
1847 HvMAX(hv) = hv_max;
1848 HvTOTALKEYS(hv) = HvTOTALKEYS(ohv);
1849 HvARRAY(hv) = ents;
aec46f14 1850 } /* not magical */
b56ba0bf 1851 else {
1604cfb0
MS
1852 /* Iterate over ohv, copying keys and values one at a time. */
1853 HE *entry;
1854 const I32 riter = HvRITER_get(ohv);
1855 HE * const eiter = HvEITER_get(ohv);
f6bb1c88 1856 STRLEN hv_keys = HvTOTALKEYS(ohv);
b56ba0bf 1857
f6bb1c88 1858 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
b56ba0bf 1859
1604cfb0
MS
1860 hv_iterinit(ohv);
1861 while ((entry = hv_iternext_flags(ohv, 0))) {
1862 SV *val = hv_iterval(ohv,entry);
1863 SV * const keysv = HeSVKEY(entry);
1864 val = SvIMMORTAL(val) ? val : newSVsv(val);
1865 if (keysv)
1866 (void)hv_store_ent(hv, keysv, val, 0);
1867 else
1868 (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry), val,
1869 HeHASH(entry), HeKFLAGS(entry));
1870 }
1871 HvRITER_set(ohv, riter);
1872 HvEITER_set(ohv, eiter);
b3ac6de7 1873 }
1c846c1f 1874
b3ac6de7
IZ
1875 return hv;
1876}
1877
defdfed5 1878/*
44170c9a 1879=for apidoc hv_copy_hints_hv
defdfed5 1880
2d7f6611 1881A specialised version of L</newHVhv> for copying C<%^H>. C<ohv> must be
defdfed5
Z
1882a pointer to a hash (which may have C<%^H> magic, but should be generally
1883non-magical), or C<NULL> (interpreted as an empty hash). The content
2d7f6611 1884of C<ohv> is copied to a new hash, which has the C<%^H>-specific magic
defdfed5
Z
1885added to it. A pointer to the new hash is returned.
1886
1887=cut
1888*/
1889
5b9c0671
NC
1890HV *
1891Perl_hv_copy_hints_hv(pTHX_ HV *const ohv)
1892{
1893 HV * const hv = newHV();
5b9c0671 1894
cb1f05e8 1895 if (ohv) {
1604cfb0 1896 STRLEN hv_max = HvMAX(ohv);
f6bb1c88 1897 STRLEN hv_keys = HvTOTALKEYS(ohv);
1604cfb0
MS
1898 HE *entry;
1899 const I32 riter = HvRITER_get(ohv);
1900 HE * const eiter = HvEITER_get(ohv);
5b9c0671 1901
1604cfb0
MS
1902 ENTER;
1903 SAVEFREESV(hv);
0db511c0 1904
f6bb1c88 1905 HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys);
5b9c0671 1906
1604cfb0
MS
1907 hv_iterinit(ohv);
1908 while ((entry = hv_iternext_flags(ohv, 0))) {
1909 SV *const sv = newSVsv(hv_iterval(ohv,entry));
1910 SV *heksv = HeSVKEY(entry);
1911 if (!heksv && sv) heksv = newSVhek(HeKEY_hek(entry));
1912 if (sv) sv_magic(sv, NULL, PERL_MAGIC_hintselem,
1913 (char *)heksv, HEf_SVKEY);
1914 if (heksv == HeSVKEY(entry))
1915 (void)hv_store_ent(hv, heksv, sv, 0);
1916 else {
1917 (void)hv_common(hv, heksv, HeKEY(entry), HeKLEN(entry),
1918 HeKFLAGS(entry), HV_FETCH_ISSTORE|HV_FETCH_JUST_SV, sv, HeHASH(entry));
1919 SvREFCNT_dec_NN(heksv);
1920 }
1921 }
1922 HvRITER_set(ohv, riter);
1923 HvEITER_set(ohv, eiter);
1924
1925 SvREFCNT_inc_simple_void_NN(hv);
1926 LEAVE;
5b9c0671
NC
1927 }
1928 hv_magic(hv, NULL, PERL_MAGIC_hints);
1929 return hv;
1930}
f6bb1c88 1931#undef HV_SET_MAX_ADJUSTED_FOR_KEYS
5b9c0671 1932
e0171a1a
DM
1933/* like hv_free_ent, but returns the SV rather than freeing it */
1934STATIC SV*
59830266 1935S_hv_free_ent_ret(pTHX_ HE *entry)
79072805 1936{
e0171a1a 1937 PERL_ARGS_ASSERT_HV_FREE_ENT_RET;
7918f24d 1938
59830266 1939 SV *val = HeVAL(entry);
68dc0745 1940 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0
MS
1941 SvREFCNT_dec(HeKEY_sv(entry));
1942 Safefree(HeKEY_hek(entry));
44a8e56a 1943 }
1db404fc 1944 else if ((HeKFLAGS(entry) & HVhek_NOTSHARED) == 0) {
1604cfb0 1945 unshare_hek(HeKEY_hek(entry));
33042aaf
NC
1946 }
1947 else {
1604cfb0 1948 Safefree(HeKEY_hek(entry));
33042aaf 1949 }
d33b2eba 1950 del_HE(entry);
e0171a1a
DM
1951 return val;
1952}
1953
1954
1955void
59830266 1956Perl_hv_free_ent(pTHX_ HV *notused, HE *entry)
e0171a1a 1957{
59830266 1958 PERL_UNUSED_ARG(notused);
e0171a1a
DM
1959
1960 if (!entry)
1604cfb0 1961 return;
59830266
NC
1962
1963 SV *val = hv_free_ent_ret(entry);
272e8453 1964 SvREFCNT_dec(val);
79072805
LW
1965}
1966
f1c32fec 1967
79072805 1968void
59830266 1969Perl_hv_delayfree_ent(pTHX_ HV *notused, HE *entry)
79072805 1970{
59830266 1971 PERL_UNUSED_ARG(notused);
7918f24d 1972
68dc0745 1973 if (!entry)
1604cfb0 1974 return;
bc4947fc
NC
1975 /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */
1976 sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */
68dc0745 1977 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0 1978 sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry)));
44a8e56a 1979 }
59830266 1980 hv_free_ent(NULL, entry);
79072805
LW
1981}
1982
954c1994
GS
1983/*
1984=for apidoc hv_clear
1985
bb8005f7 1986Frees all the elements of a hash, leaving it empty.
8b9a1153
FC
1987The XS equivalent of C<%hash = ()>. See also L</hv_undef>.
1988
a4395eba
DM
1989See L</av_clear> for a note about the hash possibly being invalid on
1990return.
954c1994
GS
1991
1992=cut
1993*/
1994
79072805 1995void
864dbfa3 1996Perl_hv_clear(pTHX_ HV *hv)
79072805 1997{
be988557
DM
1998 SSize_t orig_ix;
1999
79072805 2000 if (!hv)
1604cfb0 2001 return;
49293501 2002
ecae49c0
NC
2003 DEBUG_A(Perl_hv_assert(aTHX_ hv));
2004
be988557
DM
2005 /* avoid hv being freed when calling destructors below */
2006 EXTEND_MORTAL(1);
2007 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
2008 orig_ix = PL_tmps_ix;
ec7598c6 2009 if (SvREADONLY(hv) && HvTOTALKEYS(hv)) {
1604cfb0 2010 /* restricted hash: convert all keys to placeholders */
09cd3e1c 2011 STRLEN max = HvMAX(hv);
1604cfb0 2012 STRLEN i;
09cd3e1c 2013 for (i = 0; i <= max; i++) {
1604cfb0
MS
2014 HE *entry = (HvARRAY(hv))[i];
2015 for (; entry; entry = HeNEXT(entry)) {
2016 /* not already placeholder */
2017 if (HeVAL(entry) != &PL_sv_placeholder) {
2018 if (HeVAL(entry)) {
2019 if (SvREADONLY(HeVAL(entry))) {
2020 SV* const keysv = hv_iterkeysv(entry);
2021 Perl_croak_nocontext(
2022 "Attempt to delete readonly key '%" SVf "' from a restricted hash",
2023 (void*)keysv);
2024 }
2025 SvREFCNT_dec_NN(HeVAL(entry));
2026 }
2027 HeVAL(entry) = &PL_sv_placeholder;
2028 HvPLACEHOLDERS(hv)++;
2029 }
2030 }
2031 }
49293501 2032 }
afbbf215 2033 else {
1604cfb0
MS
2034 hv_free_entries(hv);
2035 HvPLACEHOLDERS_set(hv, 0);
49293501 2036
1604cfb0
MS
2037 if (SvRMAGICAL(hv))
2038 mg_clear(MUTABLE_SV(hv));
574c8022 2039
1604cfb0 2040 HvHASKFLAGS_off(hv);
afbbf215 2041 }
53083cad 2042 if (HvHasAUX(hv)) {
00169e2c 2043 if(HvENAME_get(hv))
dd69841b 2044 mro_isa_changed_in(hv);
1604cfb0 2045 HvEITER_set(hv, NULL);
bfcb3514 2046 }
be988557
DM
2047 /* disarm hv's premature free guard */
2048 if (LIKELY(PL_tmps_ix == orig_ix))
2049 PL_tmps_ix--;
2050 else
2051 PL_tmps_stack[orig_ix] = &PL_sv_undef;
2052 SvREFCNT_dec_NN(hv);
79072805
LW
2053}
2054
3540d4ce
AB
2055/*
2056=for apidoc hv_clear_placeholders
2057
2058Clears any placeholders from a hash. If a restricted hash has any of its keys
2059marked as readonly and the key is subsequently deleted, the key is not actually
796b6530 2060deleted but is marked by assigning it a value of C<&PL_sv_placeholder>. This tags
3540d4ce 2061it so it will be ignored by future operations such as iterating over the hash,
4cdaeff7 2062but will still allow the hash to have a value reassigned to the key at some
3540d4ce 2063future point. This function clears any such placeholder keys from the hash.
796b6530
KW
2064See C<L<Hash::Util::lock_keys()|Hash::Util/lock_keys>> for an example of its
2065use.
3540d4ce
AB
2066
2067=cut
2068*/
2069
2070void
2071Perl_hv_clear_placeholders(pTHX_ HV *hv)
2072{
b3ca2e83
NC
2073 const U32 items = (U32)HvPLACEHOLDERS_get(hv);
2074
7918f24d
NC
2075 PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS;
2076
b3ca2e83 2077 if (items)
1604cfb0 2078 clear_placeholders(hv, items);
b3ca2e83
NC
2079}
2080
2081static void
c23e25b4 2082S_clear_placeholders(pTHX_ HV *hv, const U32 placeholders)
b3ca2e83 2083{
b464bac0 2084 I32 i;
c23e25b4 2085 U32 to_find = placeholders;
d3677389 2086
7918f24d
NC
2087 PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS;
2088
c23e25b4 2089 assert(to_find);
d3677389 2090
b464bac0 2091 i = HvMAX(hv);
d3677389 2092 do {
1604cfb0
MS
2093 /* Loop down the linked list heads */
2094 HE **oentry = &(HvARRAY(hv))[i];
2095 HE *entry;
2096
2097 while ((entry = *oentry)) {
2098 if (HeVAL(entry) == &PL_sv_placeholder) {
2099 *oentry = HeNEXT(entry);
2100 if (entry == HvEITER_get(hv))
2101 HvLAZYDEL_on(hv);
2102 else {
53083cad 2103 if (HvHasAUX(hv) && HvLAZYDEL(hv) &&
1604cfb0
MS
2104 entry == HeNEXT(HvAUX(hv)->xhv_eiter))
2105 HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry);
59830266 2106 hv_free_ent(NULL, entry);
1604cfb0
MS
2107 }
2108
c23e25b4 2109 if (--to_find == 0) {
1604cfb0 2110 /* Finished. */
1604cfb0 2111 HvTOTALKEYS(hv) -= (IV)placeholders;
c23e25b4 2112 if (HvTOTALKEYS(hv) == 0)
1604cfb0
MS
2113 HvHASKFLAGS_off(hv);
2114 HvPLACEHOLDERS_set(hv, 0);
2115 return;
2116 }
2117 } else {
2118 oentry = &HeNEXT(entry);
2119 }
2120 }
d3677389
NC
2121 } while (--i >= 0);
2122 /* You can't get here, hence assertion should always fail. */
c23e25b4 2123 assert (to_find == 0);
661d43c4 2124 NOT_REACHED; /* NOTREACHED */
3540d4ce
AB
2125}
2126
76e3520e 2127STATIC void
b6bb1ecb 2128S_hv_free_entries(pTHX_ HV *hv)
79072805 2129{
e0171a1a 2130 STRLEN index = 0;
6d1c68e6 2131 SV *sv;
3abe233e 2132
367fba63 2133 PERL_ARGS_ASSERT_HV_FREE_ENTRIES;
7918f24d 2134
09cd3e1c 2135 while ((sv = Perl_hfree_next_entry(aTHX_ hv, &index)) || HvTOTALKEYS(hv)) {
1604cfb0 2136 SvREFCNT_dec(sv);
e0171a1a
DM
2137 }
2138}
23976bdd 2139
b79f7545 2140
e0171a1a 2141/* hfree_next_entry()
b6bb1ecb 2142 * For use only by S_hv_free_entries() and sv_clear().
e0171a1a 2143 * Delete the next available HE from hv and return the associated SV.
7d6175ef
FC
2144 * Returns null on empty hash. Nevertheless null is not a reliable
2145 * indicator that the hash is empty, as the deleted entry may have a
2146 * null value.
e0171a1a
DM
2147 * indexp is a pointer to the current index into HvARRAY. The index should
2148 * initially be set to 0. hfree_next_entry() may update it. */
2149
2150SV*
2151Perl_hfree_next_entry(pTHX_ HV *hv, STRLEN *indexp)
2152{
2153 struct xpvhv_aux *iter;
2154 HE *entry;
2155 HE ** array;
2156#ifdef DEBUGGING
2157 STRLEN orig_index = *indexp;
2158#endif
2159
2160 PERL_ARGS_ASSERT_HFREE_NEXT_ENTRY;
2161
53083cad 2162 if (HvHasAUX(hv) && ((iter = HvAUX(hv)))) {
1604cfb0 2163 if ((entry = iter->xhv_eiter)) {
9faf471a
NC
2164 /* the iterator may get resurrected after each
2165 * destructor call, so check each time */
2166 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2167 HvLAZYDEL_off(hv);
59830266 2168 hv_free_ent(NULL, entry);
9faf471a
NC
2169 /* warning: at this point HvARRAY may have been
2170 * re-allocated, HvMAX changed etc */
2171 }
2172 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2173 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
6a5b4183 2174#ifdef PERL_HASH_RANDOMIZE_KEYS
9faf471a 2175 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 2176#endif
9faf471a 2177 }
e0171a1a
DM
2178 }
2179
00a1a643 2180 if (!((XPVHV*)SvANY(hv))->xhv_keys)
1604cfb0 2181 return NULL;
00a1a643 2182
e0171a1a
DM
2183 array = HvARRAY(hv);
2184 assert(array);
2185 while ( ! ((entry = array[*indexp])) ) {
1604cfb0
MS
2186 if ((*indexp)++ >= HvMAX(hv))
2187 *indexp = 0;
2188 assert(*indexp != orig_index);
e0171a1a
DM
2189 }
2190 array[*indexp] = HeNEXT(entry);
2191 ((XPVHV*) SvANY(hv))->xhv_keys--;
2192
2193 if ( PL_phase != PERL_PHASE_DESTRUCT && HvENAME(hv)
1604cfb0
MS
2194 && HeVAL(entry) && isGV(HeVAL(entry))
2195 && GvHV(HeVAL(entry)) && HvENAME(GvHV(HeVAL(entry)))
e0171a1a 2196 ) {
1604cfb0
MS
2197 STRLEN klen;
2198 const char * const key = HePV(entry,klen);
2199 if ((klen > 1 && key[klen-1]==':' && key[klen-2]==':')
2200 || (klen == 1 && key[0] == ':')) {
2201 mro_package_moved(
2202 NULL, GvHV(HeVAL(entry)),
2203 (GV *)HeVAL(entry), 0
2204 );
2205 }
e0171a1a 2206 }
59830266 2207 return hv_free_ent_ret(entry);
79072805
LW
2208}
2209
e0171a1a 2210
954c1994
GS
2211/*
2212=for apidoc hv_undef
2213
8b9a1153 2214Undefines the hash. The XS equivalent of C<undef(%hash)>.
c2217cd3 2215
796b6530 2216As well as freeing all the elements of the hash (like C<hv_clear()>), this
c2217cd3 2217also frees any auxiliary data and storage associated with the hash.
8b9a1153 2218
a4395eba
DM
2219See L</av_clear> for a note about the hash possibly being invalid on
2220return.
954c1994
GS
2221
2222=cut
2223*/
2224
79072805 2225void
8581adba 2226Perl_hv_undef_flags(pTHX_ HV *hv, U32 flags)
79072805 2227{
8a50cd03 2228 bool save;
6f15e0a5 2229 SSize_t orig_ix = PL_tmps_ix; /* silence compiler warning about unitialized vars */
86f55936 2230
79072805 2231 if (!hv)
1604cfb0 2232 return;
be988557 2233 save = cBOOL(SvREFCNT(hv));
ecae49c0 2234 DEBUG_A(Perl_hv_assert(aTHX_ hv));
dd69841b 2235
b6bb1ecb 2236 /* The name must be deleted before the call to hv_free_entries so that
745edda6
FC
2237 CVs are anonymised properly. But the effective name must be pre-
2238 served until after that call (and only deleted afterwards if the
2239 call originated from sv_clear). For stashes with one name that is
2240 both the canonical name and the effective name, hv_name_set has to
2241 allocate an array for storing the effective name. We can skip that
2242 during global destruction, as it does not matter where the CVs point
2243 if they will be freed anyway. */
b6bb1ecb 2244 /* note that the code following prior to hv_free_entries is duplicated
104d7b69 2245 * in sv_clear(), and changes here should be done there too */
0ca9877d 2246 if (PL_phase != PERL_PHASE_DESTRUCT && HvNAME(hv)) {
103f5a36
NC
2247 if (PL_stashcache) {
2248 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for '%"
147e3846 2249 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
1604cfb0 2250 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
103f5a36 2251 }
1604cfb0 2252 hv_name_set(hv, NULL, 0, 0);
85e6fe83 2253 }
8505eec0 2254 if (save) {
be988557
DM
2255 /* avoid hv being freed when calling destructors below */
2256 EXTEND_MORTAL(1);
2257 PL_tmps_stack[++PL_tmps_ix] = SvREFCNT_inc_simple_NN(hv);
2258 orig_ix = PL_tmps_ix;
8505eec0 2259 }
37d796a5
NC
2260
2261 /* As well as any/all HE*s in HvARRAY(), this call also ensures that
2262 xhv_eiter is NULL, including handling the case of a tied hash partway
2263 through iteration where HvLAZYDEL() is true and xhv_eiter points to an
2264 HE* that needs to be explicitly freed. */
b6bb1ecb 2265 hv_free_entries(hv);
37d796a5 2266
53083cad 2267 /* HvHasAUX() is true for a hash if it has struct xpvhv_aux allocated. That
37d796a5
NC
2268 structure has several other pieces of allocated memory - hence those must
2269 be freed before the structure itself can be freed. Some can be freed when
d9e6229a
NC
2270 a hash is "undefined" (this function), but some must persist until it is
2271 destroyed (which might be this function's immediate caller).
2272
2273 Hence the code in this block frees what it is logical to free (and NULLs
2274 out anything freed) so that the structure is left in a logically
2275 consistent state - pointers are NULL or point to valid memory, and
2276 non-pointer values are correct for an empty hash. The structure state
2277 must remain consistent, because this code can no longer clear SVf_OOK,
2278 meaning that this structure might be read again at any point in the
37d796a5 2279 future without further checks or reinitialisation. */
53083cad 2280 if (HvHasAUX(hv)) {
47f1cf77 2281 struct mro_meta *meta;
0ca9877d 2282 const char *name;
745edda6 2283
0ca9877d 2284 if (HvENAME_get(hv)) {
1604cfb0
MS
2285 if (PL_phase != PERL_PHASE_DESTRUCT)
2286 mro_isa_changed_in(hv);
103f5a36
NC
2287 if (PL_stashcache) {
2288 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for effective name '%"
147e3846 2289 HEKf "'\n", HEKfARG(HvENAME_HEK(hv))));
1604cfb0 2290 (void)hv_deletehek(PL_stashcache, HvENAME_HEK(hv), G_DISCARD);
103f5a36 2291 }
745edda6
FC
2292 }
2293
2294 /* If this call originated from sv_clear, then we must check for
2295 * effective names that need freeing, as well as the usual name. */
2296 name = HvNAME(hv);
82943faa
KW
2297 if (flags & HV_NAME_SETALL
2298 ? cBOOL(HvAUX(hv)->xhv_name_u.xhvnameu_name)
2299 : cBOOL(name))
2300 {
103f5a36
NC
2301 if (name && PL_stashcache) {
2302 DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for name '%"
147e3846 2303 HEKf "'\n", HEKfARG(HvNAME_HEK(hv))));
1604cfb0 2304 (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD);
103f5a36 2305 }
1604cfb0 2306 hv_name_set(hv, NULL, 0, flags);
47f1cf77 2307 }
339441ef 2308 if((meta = HvAUX(hv)->xhv_mro_meta)) {
1604cfb0
MS
2309 if (meta->mro_linear_all) {
2310 SvREFCNT_dec_NN(meta->mro_linear_all);
2311 /* mro_linear_current is just acting as a shortcut pointer,
2312 hence the else. */
2313 }
2314 else
2315 /* Only the current MRO is stored, so this owns the data.
2316 */
2317 SvREFCNT_dec(meta->mro_linear_current);
2318 SvREFCNT_dec(meta->mro_nextmethod);
2319 SvREFCNT_dec(meta->isa);
2320 SvREFCNT_dec(meta->super);
2321 Safefree(meta);
2322 HvAUX(hv)->xhv_mro_meta = NULL;
47f1cf77 2323 }
2d0d1ecc 2324 }
d9e6229a
NC
2325
2326 Safefree(HvARRAY(hv));
2327 HvMAX(hv) = PERL_HASH_DEFAULT_HvMAX; /* 7 (it's a normal hash) */
2328 HvARRAY(hv) = 0;
2329
5bec93be
DM
2330 /* if we're freeing the HV, the SvMAGIC field has been reused for
2331 * other purposes, and so there can't be any placeholder magic */
2332 if (SvREFCNT(hv))
1604cfb0 2333 HvPLACEHOLDERS_set(hv, 0);
a0d0e21e
LW
2334
2335 if (SvRMAGICAL(hv))
1604cfb0 2336 mg_clear(MUTABLE_SV(hv));
be988557
DM
2337
2338 if (save) {
2339 /* disarm hv's premature free guard */
2340 if (LIKELY(PL_tmps_ix == orig_ix))
2341 PL_tmps_ix--;
2342 else
2343 PL_tmps_stack[orig_ix] = &PL_sv_undef;
2344 SvREFCNT_dec_NN(hv);
2345 }
79072805
LW
2346}
2347
4d0fbddd
NC
2348/*
2349=for apidoc hv_fill
2350
8bf4c401
YO
2351Returns the number of hash buckets that happen to be in use.
2352
6a0ca5a0
KW
2353This function implements the L<C<HvFILL> macro|perlapi/HvFILL> which you should
2354use instead.
4d0fbddd 2355
8bf4c401
YO
2356As of perl 5.25 this function is used only for debugging
2357purposes, and the number of used hash buckets is not
2358in any way cached, thus this function can be costly
2359to execute as it must iterate over all the buckets in the
2360hash.
4d0fbddd
NC
2361
2362=cut
2363*/
2364
2365STRLEN
9faf471a 2366Perl_hv_fill(pTHX_ HV *const hv)
4d0fbddd
NC
2367{
2368 STRLEN count = 0;
2369 HE **ents = HvARRAY(hv);
2370
e9b8343f 2371 PERL_UNUSED_CONTEXT;
4d0fbddd
NC
2372 PERL_ARGS_ASSERT_HV_FILL;
2373
553215cc
NC
2374 /* No keys implies no buckets used.
2375 One key can only possibly mean one bucket used. */
2376 if (HvTOTALKEYS(hv) < 2)
2377 return HvTOTALKEYS(hv);
2378
4d0fbddd 2379 if (ents) {
8bf4c401
YO
2380 /* I wonder why we count down here...
2381 * Is it some micro-optimisation?
2382 * I would have thought counting up was better.
2383 * - Yves
2384 */
1604cfb0
MS
2385 HE *const *const last = ents + HvMAX(hv);
2386 count = last + 1 - ents;
4d0fbddd 2387
1604cfb0
MS
2388 do {
2389 if (!*ents)
2390 --count;
2391 } while (++ents <= last);
4d0fbddd
NC
2392 }
2393 return count;
2394}
2395
bea177f3 2396static struct xpvhv_aux*
0e0ab621 2397S_hv_auxinit(pTHX_ HV *hv) {
bfcb3514
NC
2398 struct xpvhv_aux *iter;
2399
7918f24d
NC
2400 PERL_ARGS_ASSERT_HV_AUXINIT;
2401
53083cad 2402 if (!HvHasAUX(hv)) {
5d73aca1
NC
2403 char *array = (char *) HvARRAY(hv);
2404 if (!array) {
53a41f9c 2405 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1), char);
5d73aca1 2406 HvARRAY(hv) = (HE**)array;
0e0ab621 2407 }
94ee6ed7 2408 iter = Perl_hv_auxalloc(aTHX_ hv);
6a5b4183 2409#ifdef PERL_HASH_RANDOMIZE_KEYS
c3c9d6b1 2410 MAYBE_UPDATE_HASH_RAND_BITS();
a7b39f85 2411 iter->xhv_rand = (U32)PL_hash_rand_bits;
6a5b4183 2412#endif
a7b39f85
YO
2413 } else {
2414 iter = HvAUX(hv);
b79f7545 2415 }
bfcb3514 2416
9710057b
NC
2417 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2418 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
2419#ifdef PERL_HASH_RANDOMIZE_KEYS
2420 iter->xhv_last_rand = iter->xhv_rand;
2421#endif
2422 iter->xhv_name_u.xhvnameu_name = 0;
2423 iter->xhv_name_count = 0;
2424 iter->xhv_backreferences = 0;
2425 iter->xhv_mro_meta = NULL;
2426 iter->xhv_aux_flags = 0;
2427 return iter;
bfcb3514
NC
2428}
2429
954c1994
GS
2430/*
2431=for apidoc hv_iterinit
2432
2433Prepares a starting point to traverse a hash table. Returns the number of
fe7d7ed3
MH
2434keys in the hash, including placeholders (i.e. the same as C<HvTOTALKEYS(hv)>).
2435The return value is currently only meaningful for hashes without tie magic.
954c1994
GS
2436
2437NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of
2438hash buckets that happen to be in use. If you still need that esoteric
b24b84ef 2439value, you can get it through the macro C<HvFILL(hv)>.
954c1994 2440
e16e2ff8 2441
954c1994
GS
2442=cut
2443*/
2444
79072805 2445I32
864dbfa3 2446Perl_hv_iterinit(pTHX_ HV *hv)
79072805 2447{
7918f24d
NC
2448 PERL_ARGS_ASSERT_HV_ITERINIT;
2449
53083cad 2450 if (HvHasAUX(hv)) {
1604cfb0
MS
2451 struct xpvhv_aux * iter = HvAUX(hv);
2452 HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */
2453 if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */
2454 HvLAZYDEL_off(hv);
59830266 2455 hv_free_ent(NULL, entry);
1604cfb0 2456 }
1604cfb0
MS
2457 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
2458 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
6a5b4183 2459#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85 2460 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 2461#endif
bfcb3514 2462 } else {
1604cfb0 2463 hv_auxinit(hv);
72940dca 2464 }
44a2ac75 2465
8bf4c401 2466 /* note this includes placeholders! */
5d88ecd7 2467 return HvTOTALKEYS(hv);
79072805 2468}
bfcb3514 2469
991f482c
KW
2470/*
2471=for apidoc hv_riter_p
2472
2473Implements C<HvRITER> which you should use instead.
2474
2475=cut
2476*/
2477
bfcb3514
NC
2478I32 *
2479Perl_hv_riter_p(pTHX_ HV *hv) {
2480 struct xpvhv_aux *iter;
2481
7918f24d
NC
2482 PERL_ARGS_ASSERT_HV_RITER_P;
2483
53083cad 2484 iter = HvHasAUX(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
2485 return &(iter->xhv_riter);
2486}
2487
991f482c
KW
2488/*
2489=for apidoc hv_eiter_p
2490
2491Implements C<HvEITER> which you should use instead.
2492
2493=cut
2494*/
2495
bfcb3514
NC
2496HE **
2497Perl_hv_eiter_p(pTHX_ HV *hv) {
2498 struct xpvhv_aux *iter;
2499
7918f24d
NC
2500 PERL_ARGS_ASSERT_HV_EITER_P;
2501
53083cad 2502 iter = HvHasAUX(hv) ? HvAUX(hv) : hv_auxinit(hv);
bfcb3514
NC
2503 return &(iter->xhv_eiter);
2504}
2505
991f482c
KW
2506/*
2507=for apidoc hv_riter_set
2508
2509Implements C<HvRITER_set> which you should use instead.
2510
2511=cut
2512*/
2513
bfcb3514
NC
2514void
2515Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) {
2516 struct xpvhv_aux *iter;
2517
7918f24d
NC
2518 PERL_ARGS_ASSERT_HV_RITER_SET;
2519
53083cad 2520 if (HvHasAUX(hv)) {
1604cfb0 2521 iter = HvAUX(hv);
b79f7545 2522 } else {
1604cfb0
MS
2523 if (riter == -1)
2524 return;
bfcb3514 2525
1604cfb0 2526 iter = hv_auxinit(hv);
bfcb3514
NC
2527 }
2528 iter->xhv_riter = riter;
2529}
2530
2531void
6a5b4183
YO
2532Perl_hv_rand_set(pTHX_ HV *hv, U32 new_xhv_rand) {
2533 struct xpvhv_aux *iter;
2534
2535 PERL_ARGS_ASSERT_HV_RAND_SET;
2536
2537#ifdef PERL_HASH_RANDOMIZE_KEYS
53083cad 2538 if (HvHasAUX(hv)) {
6a5b4183
YO
2539 iter = HvAUX(hv);
2540 } else {
2541 iter = hv_auxinit(hv);
2542 }
2543 iter->xhv_rand = new_xhv_rand;
2544#else
2545 Perl_croak(aTHX_ "This Perl has not been built with support for randomized hash key traversal but something called Perl_hv_rand_set().");
2546#endif
2547}
2548
991f482c
KW
2549/*
2550=for apidoc hv_eiter_set
2551
2552Implements C<HvEITER_set> which you should use instead.
2553
2554=cut
2555*/
2556
6a5b4183 2557void
bfcb3514
NC
2558Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) {
2559 struct xpvhv_aux *iter;
2560
7918f24d
NC
2561 PERL_ARGS_ASSERT_HV_EITER_SET;
2562
53083cad 2563 if (HvHasAUX(hv)) {
1604cfb0 2564 iter = HvAUX(hv);
b79f7545 2565 } else {
1604cfb0
MS
2566 /* 0 is the default so don't go malloc()ing a new structure just to
2567 hold 0. */
2568 if (!eiter)
2569 return;
bfcb3514 2570
1604cfb0 2571 iter = hv_auxinit(hv);
bfcb3514
NC
2572 }
2573 iter->xhv_eiter = eiter;
2574}
2575
ca4a6e05
KW
2576/*
2577=for apidoc hv_name_set
2578=for apidoc_item ||hv_name_sets|HV *hv|"name"|U32 flags
2579
2580These each set the name of stash C<hv> to the specified name.
2581
2582They differ only in how the name is specified.
2583
2584In C<hv_name_sets>, the name is a literal C string, enclosed in double quotes.
2585
2586In C<hv_name_set>, C<name> points to the first byte of the name, and an
2587additional parameter, C<len>, specifies its length in bytes. Hence, the name
2588may contain embedded-NUL characters.
2589
2590If C<SVf_UTF8> is set in C<flags>, the name is treated as being in UTF-8;
2591otherwise not.
2592
2593If C<HV_NAME_SETALL> is set in C<flags>, both the name and the effective name
2594are set.
2595
2596=for apidoc Amnh||HV_NAME_SETALL
2597
2598=cut
2599*/
2600
bfcb3514 2601void
4164be69 2602Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
bfcb3514 2603{
b79f7545 2604 struct xpvhv_aux *iter;
7423f6db 2605 U32 hash;
78b79c77 2606 HEK **spot;
46c461b5 2607
7918f24d 2608 PERL_ARGS_ASSERT_HV_NAME_SET;
bfcb3514 2609
4164be69 2610 if (len > I32_MAX)
1604cfb0 2611 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
4164be69 2612
53083cad 2613 if (HvHasAUX(hv)) {
1604cfb0
MS
2614 iter = HvAUX(hv);
2615 if (iter->xhv_name_u.xhvnameu_name) {
2616 if(iter->xhv_name_count) {
2617 if(flags & HV_NAME_SETALL) {
2618 HEK ** const this_name = HvAUX(hv)->xhv_name_u.xhvnameu_names;
2619 HEK **hekp = this_name + (
2620 iter->xhv_name_count < 0
2621 ? -iter->xhv_name_count
2622 : iter->xhv_name_count
2623 );
2624 while(hekp-- > this_name+1)
2625 unshare_hek_or_pvn(*hekp, 0, 0, 0);
2626 /* The first elem may be null. */
2627 if(*this_name) unshare_hek_or_pvn(*this_name, 0, 0, 0);
2628 Safefree(this_name);
1604cfb0
MS
2629 spot = &iter->xhv_name_u.xhvnameu_name;
2630 iter->xhv_name_count = 0;
2631 }
2632 else {
2633 if(iter->xhv_name_count > 0) {
2634 /* shift some things over */
2635 Renew(
2636 iter->xhv_name_u.xhvnameu_names, iter->xhv_name_count + 1, HEK *
2637 );
2638 spot = iter->xhv_name_u.xhvnameu_names;
2639 spot[iter->xhv_name_count] = spot[1];
2640 spot[1] = spot[0];
2641 iter->xhv_name_count = -(iter->xhv_name_count + 1);
2642 }
2643 else if(*(spot = iter->xhv_name_u.xhvnameu_names)) {
2644 unshare_hek_or_pvn(*spot, 0, 0, 0);
2645 }
2646 }
2647 }
2648 else if (flags & HV_NAME_SETALL) {
2649 unshare_hek_or_pvn(iter->xhv_name_u.xhvnameu_name, 0, 0, 0);
1604cfb0
MS
2650 spot = &iter->xhv_name_u.xhvnameu_name;
2651 }
2652 else {
2653 HEK * const existing_name = iter->xhv_name_u.xhvnameu_name;
2654 Newx(iter->xhv_name_u.xhvnameu_names, 2, HEK *);
2655 iter->xhv_name_count = -2;
2656 spot = iter->xhv_name_u.xhvnameu_names;
2657 spot[1] = existing_name;
2658 }
2659 }
2660 else { spot = &iter->xhv_name_u.xhvnameu_name; iter->xhv_name_count = 0; }
16580ff5 2661 } else {
1604cfb0
MS
2662 if (name == 0)
2663 return;
bfcb3514 2664
1604cfb0
MS
2665 iter = hv_auxinit(hv);
2666 spot = &iter->xhv_name_u.xhvnameu_name;
bfcb3514 2667 }
7423f6db 2668 PERL_HASH(hash, name, len);
c60dbbc3 2669 *spot = name ? share_hek(name, flags & SVf_UTF8 ? -(I32)len : (I32)len, hash) : NULL;
4643eb69
BF
2670}
2671
2672/*
2673This is basically sv_eq_flags() in sv.c, but we avoid the magic
2674and bytes checking.
2675*/
2676
2677STATIC I32
2678hek_eq_pvn_flags(pTHX_ const HEK *hek, const char* pv, const I32 pvlen, const U32 flags) {
2679 if ( (HEK_UTF8(hek) ? 1 : 0) != (flags & SVf_UTF8 ? 1 : 0) ) {
2680 if (flags & SVf_UTF8)
2681 return (bytes_cmp_utf8(
2682 (const U8*)HEK_KEY(hek), HEK_LEN(hek),
1604cfb0 2683 (const U8*)pv, pvlen) == 0);
4643eb69
BF
2684 else
2685 return (bytes_cmp_utf8(
2686 (const U8*)pv, pvlen,
1604cfb0 2687 (const U8*)HEK_KEY(hek), HEK_LEN(hek)) == 0);
4643eb69
BF
2688 }
2689 else
d35fec6c 2690 return HEK_LEN(hek) == pvlen && ((HEK_KEY(hek) == pv)
4643eb69 2691 || memEQ(HEK_KEY(hek), pv, pvlen));
bfcb3514
NC
2692}
2693
99206677
FC
2694/*
2695=for apidoc hv_ename_add
2696
db4fbf16 2697Adds a name to a stash's internal list of effective names. See
fbe13c60 2698C<L</hv_ename_delete>>.
99206677
FC
2699
2700This is called when a stash is assigned to a new location in the symbol
2701table.
2702
2703=cut
2704*/
2705
ee72b38d 2706void
27a1175b 2707Perl_hv_ename_add(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
ee72b38d 2708{
53083cad 2709 struct xpvhv_aux *aux = HvHasAUX(hv) ? HvAUX(hv) : hv_auxinit(hv);
ee72b38d
FC
2710 U32 hash;
2711
78b79c77 2712 PERL_ARGS_ASSERT_HV_ENAME_ADD;
ee72b38d
FC
2713
2714 if (len > I32_MAX)
1604cfb0 2715 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
ee72b38d
FC
2716
2717 PERL_HASH(hash, name, len);
2718
ee72b38d 2719 if (aux->xhv_name_count) {
1604cfb0
MS
2720 I32 count = aux->xhv_name_count;
2721 HEK ** const xhv_name = aux->xhv_name_u.xhvnameu_names + (count<0);
2722 HEK **hekp = xhv_name + (count < 0 ? -count - 1 : count);
2723 while (hekp-- > xhv_name)
2724 {
2725 assert(*hekp);
2726 if (
4643eb69
BF
2727 (HEK_UTF8(*hekp) || (flags & SVf_UTF8))
2728 ? hek_eq_pvn_flags(aTHX_ *hekp, name, (I32)len, flags)
1604cfb0 2729 : (HEK_LEN(*hekp) == (I32)len && memEQ(HEK_KEY(*hekp), name, len))
4643eb69 2730 ) {
1604cfb0
MS
2731 if (hekp == xhv_name && count < 0)
2732 aux->xhv_name_count = -count;
2733 return;
2734 }
2735 }
2736 if (count < 0) aux->xhv_name_count--, count = -count;
2737 else aux->xhv_name_count++;
2738 Renew(aux->xhv_name_u.xhvnameu_names, count + 1, HEK *);
2739 (aux->xhv_name_u.xhvnameu_names)[count] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
ee72b38d
FC
2740 }
2741 else {
1604cfb0
MS
2742 HEK *existing_name = aux->xhv_name_u.xhvnameu_name;
2743 if (
2744 existing_name && (
4643eb69
BF
2745 (HEK_UTF8(existing_name) || (flags & SVf_UTF8))
2746 ? hek_eq_pvn_flags(aTHX_ existing_name, name, (I32)len, flags)
1604cfb0
MS
2747 : (HEK_LEN(existing_name) == (I32)len && memEQ(HEK_KEY(existing_name), name, len))
2748 )
2749 ) return;
2750 Newx(aux->xhv_name_u.xhvnameu_names, 2, HEK *);
2751 aux->xhv_name_count = existing_name ? 2 : -2;
2752 *aux->xhv_name_u.xhvnameu_names = existing_name;
2753 (aux->xhv_name_u.xhvnameu_names)[1] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash);
ee72b38d
FC
2754 }
2755}
2756
99206677
FC
2757/*
2758=for apidoc hv_ename_delete
2759
db4fbf16 2760Removes a name from a stash's internal list of effective names. If this is
99206677
FC
2761the name returned by C<HvENAME>, then another name in the list will take
2762its place (C<HvENAME> will use it).
2763
2764This is called when a stash is deleted from the symbol table.
2765
2766=cut
2767*/
2768
ee72b38d 2769void
27a1175b 2770Perl_hv_ename_delete(pTHX_ HV *hv, const char *name, U32 len, U32 flags)
ee72b38d 2771{
ee72b38d
FC
2772 struct xpvhv_aux *aux;
2773
78b79c77 2774 PERL_ARGS_ASSERT_HV_ENAME_DELETE;
ee72b38d
FC
2775
2776 if (len > I32_MAX)
1604cfb0 2777 Perl_croak(aTHX_ "panic: hv name too long (%" UVuf ")", (UV) len);
ee72b38d 2778
53083cad 2779 if (!HvHasAUX(hv)) return;
ee72b38d
FC
2780
2781 aux = HvAUX(hv);
15d9236d 2782 if (!aux->xhv_name_u.xhvnameu_name) return;
ee72b38d
FC
2783
2784 if (aux->xhv_name_count) {
1604cfb0
MS
2785 HEK ** const namep = aux->xhv_name_u.xhvnameu_names;
2786 I32 const count = aux->xhv_name_count;
2787 HEK **victim = namep + (count < 0 ? -count : count);
2788 while (victim-- > namep + 1)
2789 if (
4643eb69
BF
2790 (HEK_UTF8(*victim) || (flags & SVf_UTF8))
2791 ? hek_eq_pvn_flags(aTHX_ *victim, name, (I32)len, flags)
1604cfb0
MS
2792 : (HEK_LEN(*victim) == (I32)len && memEQ(HEK_KEY(*victim), name, len))
2793 ) {
2794 unshare_hek_or_pvn(*victim, 0, 0, 0);
1604cfb0
MS
2795 if (count < 0) ++aux->xhv_name_count;
2796 else --aux->xhv_name_count;
2797 if (
2798 (aux->xhv_name_count == 1 || aux->xhv_name_count == -1)
2799 && !*namep
2800 ) { /* if there are none left */
2801 Safefree(namep);
2802 aux->xhv_name_u.xhvnameu_names = NULL;
2803 aux->xhv_name_count = 0;
2804 }
2805 else {
2806 /* Move the last one back to fill the empty slot. It
2807 does not matter what order they are in. */
2808 *victim = *(namep + (count < 0 ? -count : count) - 1);
2809 }
2810 return;
2811 }
2812 if (
2813 count > 0 && ((HEK_UTF8(*namep) || (flags & SVf_UTF8))
4643eb69 2814 ? hek_eq_pvn_flags(aTHX_ *namep, name, (I32)len, flags)
1604cfb0 2815 : (HEK_LEN(*namep) == (I32)len && memEQ(HEK_KEY(*namep), name, len))
60a26c79 2816 )
1604cfb0
MS
2817 ) {
2818 aux->xhv_name_count = -count;
2819 }
ee72b38d
FC
2820 }
2821 else if(
4643eb69
BF
2822 (HEK_UTF8(aux->xhv_name_u.xhvnameu_name) || (flags & SVf_UTF8))
2823 ? hek_eq_pvn_flags(aTHX_ aux->xhv_name_u.xhvnameu_name, name, (I32)len, flags)
1604cfb0 2824 : (HEK_LEN(aux->xhv_name_u.xhvnameu_name) == (I32)len &&
4643eb69 2825 memEQ(HEK_KEY(aux->xhv_name_u.xhvnameu_name), name, len))
ee72b38d 2826 ) {
1604cfb0
MS
2827 HEK * const namehek = aux->xhv_name_u.xhvnameu_name;
2828 Newx(aux->xhv_name_u.xhvnameu_names, 1, HEK *);
2829 *aux->xhv_name_u.xhvnameu_names = namehek;
2830 aux->xhv_name_count = -1;
ee72b38d
FC
2831 }
2832}
2833
86f55936
NC
2834AV **
2835Perl_hv_backreferences_p(pTHX_ HV *hv) {
7918f24d 2836 PERL_ARGS_ASSERT_HV_BACKREFERENCES_P;
8fbcb657 2837 /* See also Perl_sv_get_backrefs in sv.c where this logic is unrolled */
34f2dd85 2838 {
53083cad 2839 struct xpvhv_aux * const iter = HvHasAUX(hv) ? HvAUX(hv) : hv_auxinit(hv);
34f2dd85
YO
2840 return &(iter->xhv_backreferences);
2841 }
86f55936
NC
2842}
2843
09aad8f0
DM
2844void
2845Perl_hv_kill_backrefs(pTHX_ HV *hv) {
2846 AV *av;
2847
2848 PERL_ARGS_ASSERT_HV_KILL_BACKREFS;
2849
53083cad 2850 if (!HvHasAUX(hv))
1604cfb0 2851 return;
09aad8f0
DM
2852
2853 av = HvAUX(hv)->xhv_backreferences;
2854
2855 if (av) {
1604cfb0
MS
2856 HvAUX(hv)->xhv_backreferences = 0;
2857 Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av);
2858 if (SvTYPE(av) == SVt_PVAV)
2859 SvREFCNT_dec_NN(av);
09aad8f0
DM
2860 }
2861}
2862
954c1994 2863/*
7a7b9979
NC
2864hv_iternext is implemented as a macro in hv.h
2865
954c1994
GS
2866=for apidoc hv_iternext
2867
fbe13c60 2868Returns entries from a hash iterator. See C<L</hv_iterinit>>.
954c1994 2869
fe7bca90
NC
2870You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the
2871iterator currently points to, without losing your place or invalidating your
2872iterator. Note that in this case the current entry is deleted from the hash
2873with your iterator holding the last reference to it. Your iterator is flagged
2874to free the entry on the next call to C<hv_iternext>, so you must not discard
2875your iterator immediately else the entry will leak - call C<hv_iternext> to
2876trigger the resource deallocation.
2877
fe7bca90
NC
2878=for apidoc hv_iternext_flags
2879
fbe13c60
KW
2880Returns entries from a hash iterator. See C<L</hv_iterinit>> and
2881C<L</hv_iternext>>.
796b6530 2882The C<flags> value will normally be zero; if C<HV_ITERNEXT_WANTPLACEHOLDERS> is
fe7bca90 2883set the placeholders keys (for restricted hashes) will be returned in addition
72d33970 2884to normal keys. By default placeholders are automatically skipped over.
7996736c 2885Currently a placeholder is implemented with a value that is
990c89d7 2886C<&PL_sv_placeholder>. Note that the implementation of placeholders and
fe7bca90
NC
2887restricted hashes may change, and the implementation currently is
2888insufficiently abstracted for any change to be tidy.
e16e2ff8 2889
5af38e47
KW
2890=for apidoc Amnh||HV_ITERNEXT_WANTPLACEHOLDERS
2891
fe7bca90 2892=cut
e16e2ff8
NC
2893*/
2894
2895HE *
2896Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags)
2897{
eb578fdb 2898 HE *entry;
a0d0e21e 2899 HE *oldentry;
463ee0b2 2900 MAGIC* mg;
bfcb3514 2901 struct xpvhv_aux *iter;
79072805 2902
7918f24d
NC
2903 PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS;
2904
53083cad 2905 if (!HvHasAUX(hv)) {
1604cfb0
MS
2906 /* Too many things (well, pp_each at least) merrily assume that you can
2907 call hv_iternext without calling hv_iterinit, so we'll have to deal
2908 with it. */
2909 hv_iterinit(hv);
bfcb3514 2910 }
d9e6229a
NC
2911 else if (!HvARRAY(hv)) {
2912 /* Since 5.002 calling hv_iternext() has ensured that HvARRAY() is
2913 non-NULL. There was explicit code for this added as part of commit
2914 4633a7c4bad06b47, without any explicit comment as to why, but from
2915 code inspection it seems to be a fix to ensure that the later line
2916 entry = ((HE**)xhv->xhv_array)[xhv->xhv_riter];
2917 was accessing a valid address, because that lookup in the loop was
2918 always reached even if the hash had no keys.
2919
2920 That explicit code was removed in 2005 as part of b79f7545f218479c:
2921 Store the xhv_aux structure after the main array.
2922 This reduces the size of HV bodies from 24 to 20 bytes on a 32 bit
2923 build. It has the side effect of defined %symbol_table:: now always
2924 being true. defined %hash is already deprecated.
2925
2926 with a comment and assertion added to note that after the call to
2927 hv_iterinit() HvARRAY() will now always be non-NULL.
2928
2929 In turn, that potential NULL-pointer access within the loop was made
2930 unreachable in 2009 by commit 9eb4ebd1619c0362
2931 In Perl_hv_iternext_flags(), clarify and generalise the empty hash bailout code.
2932
2933 which skipped the entire while loop if the hash had no keys.
2934 (If the hash has any keys, HvARRAY() cannot be NULL.)
2935 Hence the code in hv_iternext_flags() has long been able to handle
2936 HvARRAY() being NULL because no keys are allocated.
2937
2938 Now that we have decoupled the aux structure from HvARRAY(),
2939 HvARRAY() can now be NULL even when SVf_OOK is true (and the aux
2940 struct is allocated and correction initialised).
2941
2942 Is this actually a guarantee that we need to make? We should check
2943 whether anything is actually relying on this, or if we are simply
2944 making work for ourselves.
2945
2946 For now, keep the behaviour as-was - after calling hv_iternext_flags
2947 ensure that HvARRAY() is non-NULL. Many (other) things are changing -
2948 no need to add risk by changing this too. But in the future we should
2949 consider changing hv_iternext_flags() to avoid allocating HvARRAY()
2950 here, and potentially also we avoid allocating HvARRAY()
2951 automatically in hv_auxinit() */
2952
2953 char *array;
2954 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1), char);
2955 HvARRAY(hv) = (HE**)array;
2956 }
2957
b79f7545 2958 iter = HvAUX(hv);
bfcb3514
NC
2959
2960 oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */
e62cc96a 2961 if (SvMAGICAL(hv) && SvRMAGICAL(hv)) {
1604cfb0 2962 if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) {
e62cc96a
YO
2963 SV * const key = sv_newmortal();
2964 if (entry) {
2965 sv_setsv(key, HeSVKEY_force(entry));
2966 SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */
1604cfb0 2967 HeSVKEY_set(entry, NULL);
e62cc96a
YO
2968 }
2969 else {
2970 char *k;
2971 HEK *hek;
2972
2973 /* one HE per MAGICAL hash */
2974 iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */
1604cfb0 2975 HvLAZYDEL_on(hv); /* make sure entry gets freed */
e62cc96a 2976 Zero(entry, 1, HE);
ad64d0ec 2977 Newxz(k, HEK_BASESIZE + sizeof(const SV *), char);
e62cc96a
YO
2978 hek = (HEK*)k;
2979 HeKEY_hek(entry) = hek;
2980 HeKLEN(entry) = HEf_SVKEY;
2981 }
ad64d0ec 2982 magic_nextpack(MUTABLE_SV(hv),mg,key);
e62cc96a
YO
2983 if (SvOK(key)) {
2984 /* force key to stay around until next time */
2985 HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key));
2986 return entry; /* beware, hent_val is not set */
2987 }
ef8d46e8 2988 SvREFCNT_dec(HeVAL(entry));
e62cc96a
YO
2989 Safefree(HeKEY_hek(entry));
2990 del_HE(entry);
2991 iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */
1604cfb0 2992 HvLAZYDEL_off(hv);
e62cc96a 2993 return NULL;
81714fb9 2994 }
79072805 2995 }
d5a0a5dd 2996#if defined(DYNAMIC_ENV_FETCH) && defined(VMS) /* set up %ENV for iteration */
ad64d0ec 2997 if (!entry && SvRMAGICAL((const SV *)hv)
1604cfb0
MS
2998 && mg_find((const SV *)hv, PERL_MAGIC_env)) {
2999 prime_env_iter();
03026e68 3000 }
f675dbe5 3001#endif
463ee0b2 3002
bfaf5b52 3003 /* hv_iterinit now ensures this. */
b79f7545
NC
3004 assert (HvARRAY(hv));
3005
015a5f36 3006 /* At start of hash, entry is NULL. */
fde52b5c 3007 if (entry)
8aacddc1 3008 {
1604cfb0 3009 entry = HeNEXT(entry);
e16e2ff8
NC
3010 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
3011 /*
3012 * Skip past any placeholders -- don't want to include them in
3013 * any iteration.
3014 */
7996736c 3015 while (entry && HeVAL(entry) == &PL_sv_placeholder) {
e16e2ff8
NC
3016 entry = HeNEXT(entry);
3017 }
1604cfb0 3018 }
8aacddc1 3019 }
6a5b4183
YO
3020
3021#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85
YO
3022 if (iter->xhv_last_rand != iter->xhv_rand) {
3023 if (iter->xhv_riter != -1) {
3024 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
3025 "Use of each() on hash after insertion without resetting hash iterator results in undefined behavior"
3026 pTHX__FORMAT
3027 pTHX__VALUE);
3028 }
3029 iter->xhv_last_rand = iter->xhv_rand;
3030 }
6a5b4183 3031#endif
015a5f36 3032
9eb4ebd1
NC
3033 /* Skip the entire loop if the hash is empty. */
3034 if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS)
1604cfb0 3035 ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) {
4b6b6165 3036 STRLEN max = HvMAX(hv);
1604cfb0
MS
3037 while (!entry) {
3038 /* OK. Come to the end of the current list. Grab the next one. */
3039
3040 iter->xhv_riter++; /* HvRITER(hv)++ */
4b6b6165 3041 if (iter->xhv_riter > (I32)max /* HvRITER(hv) > HvMAX(hv) */) {
1604cfb0
MS
3042 /* There is no next one. End of the hash. */
3043 iter->xhv_riter = -1; /* HvRITER(hv) = -1 */
6a5b4183
YO
3044#ifdef PERL_HASH_RANDOMIZE_KEYS
3045 iter->xhv_last_rand = iter->xhv_rand; /* reset xhv_last_rand so we can detect inserts during traversal */
3046#endif
1604cfb0
MS
3047 break;
3048 }
4b6b6165 3049 entry = (HvARRAY(hv))[ PERL_HASH_ITER_BUCKET(iter) & max ];
8aacddc1 3050
1604cfb0
MS
3051 if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) {
3052 /* If we have an entry, but it's a placeholder, don't count it.
3053 Try the next. */
3054 while (entry && HeVAL(entry) == &PL_sv_placeholder)
3055 entry = HeNEXT(entry);
3056 }
3057 /* Will loop again if this linked list starts NULL
3058 (for HV_ITERNEXT_WANTPLACEHOLDERS)
3059 or if we run through it and find only placeholders. */
3060 }
fde52b5c 3061 }
a7b39f85
YO
3062 else {
3063 iter->xhv_riter = -1;
6a5b4183 3064#ifdef PERL_HASH_RANDOMIZE_KEYS
a7b39f85 3065 iter->xhv_last_rand = iter->xhv_rand;
6a5b4183 3066#endif
a7b39f85 3067 }
79072805 3068
72940dca 3069 if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */
1604cfb0 3070 HvLAZYDEL_off(hv);
59830266 3071 hv_free_ent(NULL, oldentry);
72940dca 3072 }
a0d0e21e 3073
bfcb3514 3074 iter->xhv_eiter = entry; /* HvEITER(hv) = entry */
79072805
LW
3075 return entry;
3076}
3077
954c1994
GS
3078/*
3079=for apidoc hv_iterkey
3080
3081Returns the key from the current position of the hash iterator. See
fbe13c60 3082C<L</hv_iterinit>>.
954c1994
GS
3083
3084=cut
3085*/
3086
79072805 3087char *
5aaab254 3088Perl_hv_iterkey(pTHX_ HE *entry, I32 *retlen)
79072805 3089{
7918f24d
NC
3090 PERL_ARGS_ASSERT_HV_ITERKEY;
3091
fde52b5c 3092 if (HeKLEN(entry) == HEf_SVKEY) {
1604cfb0
MS
3093 STRLEN len;
3094 char * const p = SvPV(HeKEY_sv(entry), len);
3095 *retlen = len;
3096 return p;
fde52b5c 3097 }
3098 else {
1604cfb0
MS
3099 *retlen = HeKLEN(entry);
3100 return HeKEY(entry);
fde52b5c 3101 }
3102}
3103
3104/* unlike hv_iterval(), this always returns a mortal copy of the key */
954c1994
GS
3105/*
3106=for apidoc hv_iterkeysv
3107
3108Returns the key as an C<SV*> from the current position of the hash
3109iterator. The return value will always be a mortal copy of the key. Also
fbe13c60 3110see C<L</hv_iterinit>>.
954c1994
GS
3111
3112=cut
3113*/
3114
fde52b5c 3115SV *
5aaab254 3116Perl_hv_iterkeysv(pTHX_ HE *entry)
fde52b5c 3117{
7918f24d
NC
3118 PERL_ARGS_ASSERT_HV_ITERKEYSV;
3119
518db96a 3120 return newSVhek_mortal(HeKEY_hek(entry));
79072805
LW
3121}
3122
954c1994
GS
3123/*
3124=for apidoc hv_iterval
3125
3126Returns the value from the current position of the hash iterator. See
fbe13c60 3127C<L</hv_iterkey>>.
954c1994
GS
3128
3129=cut
3130*/
3131
79072805 3132SV *
5aaab254 3133Perl_hv_iterval(pTHX_ HV *hv, HE *entry)
79072805 3134{
7918f24d
NC
3135 PERL_ARGS_ASSERT_HV_ITERVAL;
3136
8990e307 3137 if (SvRMAGICAL(hv)) {
1604cfb0
MS
3138 if (mg_find((const SV *)hv, PERL_MAGIC_tied)) {
3139 SV* const sv = sv_newmortal();
3140 if (HeKLEN(entry) == HEf_SVKEY)
3141 mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY);
3142 else
3143 mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry));
3144 return sv;
3145 }
79072805 3146 }
fde52b5c 3147 return HeVAL(entry);
79072805
LW
3148}
3149
954c1994
GS
3150/*
3151=for apidoc hv_iternextsv
3152
3153Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one
3154operation.
3155
3156=cut
3157*/
3158
a0d0e21e 3159SV *
864dbfa3 3160Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen)
a0d0e21e 3161{
0bd48802
AL
3162 HE * const he = hv_iternext_flags(hv, 0);
3163
7918f24d
NC
3164 PERL_ARGS_ASSERT_HV_ITERNEXTSV;
3165
0bd48802 3166 if (!he)
1604cfb0 3167 return NULL;
a0d0e21e
LW
3168 *key = hv_iterkey(he, retlen);
3169 return hv_iterval(hv, he);
3170}
3171
954c1994 3172/*
bc5cdc23
NC
3173
3174Now a macro in hv.h
3175
954c1994
GS
3176=for apidoc hv_magic
3177
fbe13c60 3178Adds magic to a hash. See C<L</sv_magic>>.
954c1994 3179
78d7bba4
KW
3180=for apidoc unsharepvn
3181
3182If no one has access to shared string C<str> with length C<len>, free it.
3183
3184C<len> and C<hash> must both be valid for C<str>.
3185
954c1994
GS
3186=cut
3187*/
3188
bbce6d69 3189void
864dbfa3 3190Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash)
fde52b5c 3191{
19692e8d
NC
3192 unshare_hek_or_pvn (NULL, str, len, hash);
3193}
3194
3195
3196void
3197Perl_unshare_hek(pTHX_ HEK *hek)
3198{
bf11fd37 3199 assert(hek);
19692e8d
NC
3200 unshare_hek_or_pvn(hek, NULL, 0, 0);
3201}
3202
3203/* possibly free a shared string if no one has access to it
3204 hek if non-NULL takes priority over the other 3, else str, len and hash
3205 are used. If so, len and hash must both be valid for str.
3206 */
df132699 3207STATIC void
97ddebaf 3208S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash)
19692e8d 3209{
20454177 3210 HE *entry;
eb578fdb 3211 HE **oentry;
c3654f1a 3212 bool is_utf8 = FALSE;
19692e8d 3213 int k_flags = 0;
aec46f14 3214 const char * const save = str;
cbbf8932 3215 struct shared_he *he = NULL;
c3654f1a 3216
19692e8d 3217 if (hek) {
1db404fc 3218 assert((HEK_FLAGS(hek) & HVhek_NOTSHARED) == 0);
1604cfb0
MS
3219 /* Find the shared he which is just before us in memory. */
3220 he = (struct shared_he *)(((char *)hek)
3221 - STRUCT_OFFSET(struct shared_he,
3222 shared_he_hek));
cbae3960 3223
1604cfb0
MS
3224 /* Assert that the caller passed us a genuine (or at least consistent)
3225 shared hek */
3226 assert (he->shared_he_he.hent_hek == hek);
29404ae0 3227
1604cfb0
MS
3228 if (he->shared_he_he.he_valu.hent_refcount - 1) {
3229 --he->shared_he_he.he_valu.hent_refcount;
3230 return;
3231 }
29404ae0 3232
19692e8d
NC
3233 hash = HEK_HASH(hek);
3234 } else if (len < 0) {
3235 STRLEN tmplen = -len;
3236 is_utf8 = TRUE;
3237 /* See the note in hv_fetch(). --jhi */
3238 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
3239 len = tmplen;
3240 if (is_utf8)
3241 k_flags = HVhek_UTF8;
3242 if (str != save)
3243 k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
c3654f1a 3244 }
1c846c1f 3245
de616631 3246 /* what follows was the moral equivalent of:
6b88bc9c 3247 if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) {
1604cfb0
MS
3248 if (--*Svp == NULL)
3249 hv_delete(PL_strtab, str, len, G_DISCARD, hash);
bbce6d69 3250 } */
09cd3e1c 3251
fde52b5c 3252 /* assert(xhv_array != 0) */
9de10d5c 3253 oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)];
6c1b96a1 3254 if (he) {
1604cfb0 3255 const HE *const he_he = &(he->shared_he_he);
45d1cc86 3256 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
35ab5632
NC
3257 if (entry == he_he)
3258 break;
19692e8d
NC
3259 }
3260 } else {
38b26de3 3261 const U8 flags_masked = k_flags & HVhek_STORAGE_MASK;
45d1cc86 3262 for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) {
19692e8d
NC
3263 if (HeHASH(entry) != hash) /* strings can't be equal */
3264 continue;
3265 if (HeKLEN(entry) != len)
3266 continue;
3267 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3268 continue;
3269 if (HeKFLAGS(entry) != flags_masked)
3270 continue;
19692e8d
NC
3271 break;
3272 }
3273 }
3274
35ab5632
NC
3275 if (entry) {
3276 if (--entry->he_valu.hent_refcount == 0) {
19692e8d 3277 *oentry = HeNEXT(entry);
cbae3960 3278 Safefree(entry);
09cd3e1c 3279 HvTOTALKEYS(PL_strtab)--;
19692e8d 3280 }
fde52b5c 3281 }
19692e8d 3282
9b387841 3283 if (!entry)
1604cfb0
MS
3284 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
3285 "Attempt to free nonexistent shared string '%s'%s"
3286 pTHX__FORMAT,
3287 hek ? HEK_KEY(hek) : str,
3288 ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE);
19692e8d 3289 if (k_flags & HVhek_FREEKEY)
1604cfb0 3290 Safefree(str);
fde52b5c 3291}
3292
bbce6d69 3293/* get a (constant) string ptr from the global string table
3294 * string will get added if it is not already there.
fde52b5c 3295 * len and hash must both be valid for str.
3296 */
bbce6d69 3297HEK *
b02f3645 3298Perl_share_hek(pTHX_ const char *str, SSize_t len, U32 hash)
fde52b5c 3299{
da58a35d 3300 bool is_utf8 = FALSE;
19692e8d 3301 int flags = 0;
aec46f14 3302 const char * const save = str;
da58a35d 3303
7918f24d
NC
3304 PERL_ARGS_ASSERT_SHARE_HEK;
3305
da58a35d 3306 if (len < 0) {
77caf834 3307 STRLEN tmplen = -len;
da58a35d 3308 is_utf8 = TRUE;
77caf834
JH
3309 /* See the note in hv_fetch(). --jhi */
3310 str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8);
3311 len = tmplen;
19692e8d
NC
3312 /* If we were able to downgrade here, then than means that we were passed
3313 in a key which only had chars 0-255, but was utf8 encoded. */
3314 if (is_utf8)
3315 flags = HVhek_UTF8;
3316 /* If we found we were able to downgrade the string to bytes, then
3317 we should flag that it needs upgrading on keys or each. Also flag
3318 that we need share_hek_flags to free the string. */
4643eb69
BF
3319 if (str != save) {
3320 PERL_HASH(hash, str, len);
19692e8d 3321 flags |= HVhek_WASUTF8 | HVhek_FREEKEY;
4643eb69 3322 }
19692e8d
NC
3323 }
3324
6e838c70 3325 return share_hek_flags (str, len, hash, flags);
19692e8d
NC
3326}
3327
6e838c70 3328STATIC HEK *
b02f3645 3329S_share_hek_flags(pTHX_ const char *str, STRLEN len, U32 hash, int flags)
19692e8d 3330{
eb578fdb 3331 HE *entry;
38b26de3 3332 const U8 flags_masked = flags & HVhek_STORAGE_MASK;
263cb4a6 3333 const U32 hindex = hash & (I32) HvMAX(PL_strtab);
7918f24d
NC
3334
3335 PERL_ARGS_ASSERT_SHARE_HEK_FLAGS;
38b26de3 3336 assert(!(flags & HVhek_NOTSHARED));
bbce6d69 3337
b02f3645
AC
3338 if (UNLIKELY(len > (STRLEN) I32_MAX)) {
3339 Perl_croak_nocontext("Sorry, hash keys must be smaller than 2**31 bytes");
3340 }
3341
fde52b5c 3342 /* what follows is the moral equivalent of:
1c846c1f 3343
6b88bc9c 3344 if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
1604cfb0 3345 hv_store(PL_strtab, str, len, NULL, hash);
fdcd69b6 3346
1604cfb0
MS
3347 Can't rehash the shared string table, so not sure if it's worth
3348 counting the number of entries in the linked list
bbce6d69 3349 */
7918f24d 3350
fde52b5c 3351 /* assert(xhv_array != 0) */
263cb4a6
NC
3352 entry = (HvARRAY(PL_strtab))[hindex];
3353 for (;entry; entry = HeNEXT(entry)) {
1604cfb0
MS
3354 if (HeHASH(entry) != hash) /* strings can't be equal */
3355 continue;
3356 if (HeKLEN(entry) != (SSize_t) len)
3357 continue;
3358 if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */
3359 continue;
3360 if (HeKFLAGS(entry) != flags_masked)
3361 continue;
3362 break;
fde52b5c 3363 }
263cb4a6
NC
3364
3365 if (!entry) {
1604cfb0
MS
3366 /* What used to be head of the list.
3367 If this is NULL, then we're the first entry for this slot, which
3368 means we need to increate fill. */
3369 struct shared_he *new_entry;
3370 HEK *hek;
3371 char *k;
3372 HE **const head = &HvARRAY(PL_strtab)[hindex];
3373 HE *const next = *head;
09cd3e1c 3374 XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab);
1604cfb0
MS
3375
3376 /* We don't actually store a HE from the arena and a regular HEK.
3377 Instead we allocate one chunk of memory big enough for both,
3378 and put the HEK straight after the HE. This way we can find the
3379 HE directly from the HEK.
3380 */
3381
3382 Newx(k, STRUCT_OFFSET(struct shared_he,
3383 shared_he_hek.hek_key[0]) + len + 2, char);
3384 new_entry = (struct shared_he *)k;
3385 entry = &(new_entry->shared_he_he);
3386 hek = &(new_entry->shared_he_hek);
3387
3388 Copy(str, HEK_KEY(hek), len, char);
3389 HEK_KEY(hek)[len] = 0;
3390 HEK_LEN(hek) = len;
3391 HEK_HASH(hek) = hash;
3392 HEK_FLAGS(hek) = (unsigned char)flags_masked;
3393
3394 /* Still "point" to the HEK, so that other code need not know what
3395 we're up to. */
3396 HeKEY_hek(entry) = hek;
3397 entry->he_valu.hent_refcount = 0;
3398 HeNEXT(entry) = next;
3399 *head = entry;
3400
3401 xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
3402 if (!next) { /* initial entry? */
3403 } else if ( DO_HSPLIT(xhv) ) {
adf6906b
NC
3404 const STRLEN oldsize = xhv->xhv_max + 1;
3405 hsplit(PL_strtab, oldsize, oldsize * 2);
1604cfb0 3406 }
bbce6d69 3407 }
3408
de616631 3409 ++entry->he_valu.hent_refcount;
19692e8d
NC
3410
3411 if (flags & HVhek_FREEKEY)
1604cfb0 3412 Safefree(str);
19692e8d 3413
6e838c70 3414 return HeKEY_hek(entry);
fde52b5c 3415}
ecae49c0 3416
6174b39a 3417SSize_t *
ca732855
NC
3418Perl_hv_placeholders_p(pTHX_ HV *hv)
3419{
ad64d0ec 3420 MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3421
7918f24d
NC
3422 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P;
3423
ca732855 3424 if (!mg) {
1604cfb0 3425 mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0);
ca732855 3426
1604cfb0
MS
3427 if (!mg) {
3428 Perl_die(aTHX_ "panic: hv_placeholders_p");
3429 }
ca732855
NC
3430 }
3431 return &(mg->mg_len);
3432}
3433
991f482c
KW
3434/*
3435=for apidoc hv_placeholders_get
3436
3437Implements C<HvPLACEHOLDERS_get>, which you should use instead.
3438
3439=cut
3440*/
ca732855
NC
3441
3442I32
0c289d13 3443Perl_hv_placeholders_get(pTHX_ const HV *hv)
ca732855 3444{
0c289d13 3445 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3446
7918f24d 3447 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET;
23491f1d 3448 PERL_UNUSED_CONTEXT;
7918f24d 3449
ca732855
NC
3450 return mg ? mg->mg_len : 0;
3451}
3452
991f482c
KW
3453/*
3454=for apidoc hv_placeholders_set
3455
3456Implements C<HvPLACEHOLDERS_set>, which you should use instead.
3457
3458=cut
3459*/
3460
ca732855 3461void
ac1e784a 3462Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph)
ca732855 3463{
ad64d0ec 3464 MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash);
ca732855 3465
7918f24d
NC
3466 PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET;
3467
ca732855 3468 if (mg) {
1604cfb0 3469 mg->mg_len = ph;
ca732855 3470 } else if (ph) {
1604cfb0
MS
3471 if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph))
3472 Perl_die(aTHX_ "panic: hv_placeholders_set");
ca732855
NC
3473 }
3474 /* else we don't need to add magic to record 0 placeholders. */
3475}
ecae49c0 3476
2a49f0f5 3477STATIC SV *
7b0bddfa
NC
3478S_refcounted_he_value(pTHX_ const struct refcounted_he *he)
3479{
3480 SV *value;
7918f24d
NC
3481
3482 PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE;
3483
7b0bddfa
NC
3484 switch(he->refcounted_he_data[0] & HVrhek_typemask) {
3485 case HVrhek_undef:
8fcb2425 3486 value = newSV_type(SVt_NULL);
1604cfb0 3487 break;
7b0bddfa 3488 case HVrhek_delete:
1604cfb0
MS
3489 value = &PL_sv_placeholder;
3490 break;
7b0bddfa 3491 case HVrhek_IV:
1604cfb0
MS
3492 value = newSViv(he->refcounted_he_val.refcounted_he_u_iv);
3493 break;
44ebaf21 3494 case HVrhek_UV:
1604cfb0
MS
3495 value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv);
3496 break;
7b0bddfa 3497 case HVrhek_PV:
44ebaf21 3498 case HVrhek_PV_UTF8:
1604cfb0
MS
3499 /* Create a string SV that directly points to the bytes in our
3500 structure. */
3501 value = newSV_type(SVt_PV);
3502 SvPV_set(value, (char *) he->refcounted_he_data + 1);
3503 SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len);
3504 /* This stops anything trying to free it */
3505 SvLEN_set(value, 0);
3506 SvPOK_on(value);
3507 SvREADONLY_on(value);
3508 if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8)
3509 SvUTF8_on(value);
3510 break;
7b0bddfa 3511 default:
1604cfb0
MS
3512 Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %" UVxf,
3513 (UV)he->refcounted_he_data[0]);
7b0bddfa
NC
3514 }
3515 return value;
3516}
3517
ecae49c0 3518/*
44170c9a 3519=for apidoc refcounted_he_chain_2hv
8dff4fc5 3520
20439bc7
Z
3521Generates and returns a C<HV *> representing the content of a
3522C<refcounted_he> chain.
2d7f6611 3523C<flags> is currently unused and must be zero.
8dff4fc5
BM
3524
3525=cut
3526*/
3527HV *
20439bc7 3528Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain, U32 flags)
8dff4fc5 3529{
20439bc7
Z
3530 HV *hv;
3531 U32 placeholders, max;
b3ca2e83 3532
20439bc7 3533 if (flags)
1604cfb0
MS
3534 Perl_croak(aTHX_ "panic: refcounted_he_chain_2hv bad flags %" UVxf,
3535 (UV)flags);
b3ca2e83 3536
b3ca2e83
NC
3537 /* We could chase the chain once to get an idea of the number of keys,
3538 and call ksplit. But for now we'll make a potentially inefficient
3539 hash with only 8 entries in its array. */
20439bc7 3540 hv = newHV();
33042aaf
NC
3541#ifdef NODEFAULT_SHAREKEYS
3542 /* We share keys in the COP, so it's much easier to keep sharing keys in
3543 the hash we build from it. */
3544 HvSHAREKEYS_on(hv);
3545#endif
20439bc7 3546 max = HvMAX(hv);
b3ca2e83 3547 if (!HvARRAY(hv)) {
1604cfb0
MS
3548 char *array;
3549 Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char);
3550 HvARRAY(hv) = (HE**)array;
b3ca2e83
NC
3551 }
3552
20439bc7 3553 placeholders = 0;
b3ca2e83 3554 while (chain) {
cbb1fbea 3555#ifdef USE_ITHREADS
1604cfb0 3556 U32 hash = chain->refcounted_he_hash;
cbb1fbea 3557#else
1604cfb0 3558 U32 hash = HEK_HASH(chain->refcounted_he_hek);
cbb1fbea 3559#endif
1604cfb0
MS
3560 HE **oentry = &((HvARRAY(hv))[hash & max]);
3561 HE *entry = *oentry;
3562 SV *value;
3563
3564 for (; entry; entry = HeNEXT(entry)) {
3565 if (HeHASH(entry) == hash) {
3566 /* We might have a duplicate key here. If so, entry is older
3567 than the key we've already put in the hash, so if they are
3568 the same, skip adding entry. */
9f769845 3569#ifdef USE_ITHREADS
1604cfb0
MS
3570 const STRLEN klen = HeKLEN(entry);
3571 const char *const key = HeKEY(entry);
3572 if (klen == chain->refcounted_he_keylen
82943faa
KW
3573 && (cBOOL(HeKUTF8(entry))
3574 == cBOOL((chain->refcounted_he_data[0] & HVhek_UTF8)))
1604cfb0
MS
3575 && memEQ(key, REF_HE_KEY(chain), klen))
3576 goto next_please;
9f769845 3577#else
1604cfb0
MS
3578 if (HeKEY_hek(entry) == chain->refcounted_he_hek)
3579 goto next_please;
3580 if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek)
3581 && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek)
3582 && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek),
3583 HeKLEN(entry)))
3584 goto next_please;
9f769845 3585#endif
1604cfb0
MS
3586 }
3587 }
3588 assert (!entry);
3589 entry = new_HE();
b3ca2e83 3590
cbb1fbea 3591#ifdef USE_ITHREADS
1604cfb0
MS
3592 HeKEY_hek(entry)
3593 = share_hek_flags(REF_HE_KEY(chain),
3594 chain->refcounted_he_keylen,
3595 chain->refcounted_he_hash,
3596 (chain->refcounted_he_data[0]
3597 & (HVhek_UTF8|HVhek_WASUTF8)));
cbb1fbea 3598#else
1604cfb0 3599 HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek);
cbb1fbea 3600#endif
1604cfb0
MS
3601 value = refcounted_he_value(chain);
3602 if (value == &PL_sv_placeholder)
3603 placeholders++;
3604 HeVAL(entry) = value;
b3ca2e83 3605
1604cfb0
MS
3606 /* Link it into the chain. */
3607 HeNEXT(entry) = *oentry;
3608 *oentry = entry;
b3ca2e83 3609
1604cfb0 3610 HvTOTALKEYS(hv)++;
b3ca2e83
NC
3611
3612 next_please:
1604cfb0 3613 chain = chain->refcounted_he_next;
b3ca2e83
NC
3614 }
3615
3616 if (placeholders) {
1604cfb0 3617 clear_placeholders(hv, placeholders);
b3ca2e83
NC
3618 }
3619
3620 /* We could check in the loop to see if we encounter any keys with key
3621 flags, but it's probably not worth it, as this per-hash flag is only
3622 really meant as an optimisation for things like Storable. */
3623 HvHASKFLAGS_on(hv);
def9038f 3624 DEBUG_A(Perl_hv_assert(aTHX_ hv));
b3ca2e83
NC
3625
3626 return hv;
3627}
3628
20439bc7 3629/*
44170c9a 3630=for apidoc refcounted_he_fetch_pvn
20439bc7
Z
3631
3632Search along a C<refcounted_he> chain for an entry with the key specified
2d7f6611 3633by C<keypv> and C<keylen>. If C<flags> has the C<REFCOUNTED_HE_KEY_UTF8>
20439bc7 3634bit set, the key octets are interpreted as UTF-8, otherwise they
2d7f6611 3635are interpreted as Latin-1. C<hash> is a precomputed hash of the key
20439bc7
Z
3636string, or zero if it has not been precomputed. Returns a mortal scalar
3637representing the value associated with the key, or C<&PL_sv_placeholder>
3638if there is no value associated with the key.
3639
3640=cut
3641*/
3642
7b0bddfa 3643SV *
20439bc7 3644Perl_refcounted_he_fetch_pvn(pTHX_ const struct refcounted_he *chain,
1604cfb0 3645 const char *keypv, STRLEN keylen, U32 hash, U32 flags)
7b0bddfa 3646{
20439bc7
Z
3647 U8 utf8_flag;
3648 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PVN;
7b0bddfa 3649
94250aee 3650 if (flags & ~(REFCOUNTED_HE_KEY_UTF8|REFCOUNTED_HE_EXISTS))
1604cfb0
MS
3651 Perl_croak(aTHX_ "panic: refcounted_he_fetch_pvn bad flags %" UVxf,
3652 (UV)flags);
20439bc7 3653 if (!chain)
1604cfb0 3654 goto ret;
20439bc7 3655 if (flags & REFCOUNTED_HE_KEY_UTF8) {
1604cfb0
MS
3656 /* For searching purposes, canonicalise to Latin-1 where possible. */
3657 const char *keyend = keypv + keylen, *p;
3658 STRLEN nonascii_count = 0;
3659 for (p = keypv; p != keyend; p++) {
3660 if (! UTF8_IS_INVARIANT(*p)) {
3661 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) {
3662 goto canonicalised_key;
e8e5e5b3 3663 }
1604cfb0 3664 nonascii_count++;
e8e5e5b3 3665 p++;
1604cfb0
MS
3666 }
3667 }
3668 if (nonascii_count) {
3669 char *q;
3670 const char *p = keypv, *keyend = keypv + keylen;
3671 keylen -= nonascii_count;
3672 Newx(q, keylen, char);
3673 SAVEFREEPV(q);
3674 keypv = q;
3675 for (; p != keyend; p++, q++) {
3676 U8 c = (U8)*p;
e8e5e5b3
KW
3677 if (UTF8_IS_INVARIANT(c)) {
3678 *q = (char) c;
3679 }
3680 else {
3681 p++;
a62b247b 3682 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
e8e5e5b3 3683 }
1604cfb0
MS
3684 }
3685 }
3686 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3687 canonicalised_key: ;
20439bc7
Z
3688 }
3689 utf8_flag = (flags & REFCOUNTED_HE_KEY_UTF8) ? HVhek_UTF8 : 0;
3690 if (!hash)
1604cfb0 3691 PERL_HASH(hash, keypv, keylen);
7b0bddfa 3692
20439bc7 3693 for (; chain; chain = chain->refcounted_he_next) {
1604cfb0 3694 if (
7b0bddfa 3695#ifdef USE_ITHREADS
1604cfb0
MS
3696 hash == chain->refcounted_he_hash &&
3697 keylen == chain->refcounted_he_keylen &&
3698 memEQ(REF_HE_KEY(chain), keypv, keylen) &&
3699 utf8_flag == (chain->refcounted_he_data[0] & HVhek_UTF8)
7b0bddfa 3700#else
1604cfb0
MS
3701 hash == HEK_HASH(chain->refcounted_he_hek) &&
3702 keylen == (STRLEN)HEK_LEN(chain->refcounted_he_hek) &&
3703 memEQ(HEK_KEY(chain->refcounted_he_hek), keypv, keylen) &&
3704 utf8_flag == (HEK_FLAGS(chain->refcounted_he_hek) & HVhek_UTF8)
7b0bddfa 3705#endif
1604cfb0
MS
3706 ) {
3707 if (flags & REFCOUNTED_HE_EXISTS)
3708 return (chain->refcounted_he_data[0] & HVrhek_typemask)
3709 == HVrhek_delete
3710 ? NULL : &PL_sv_yes;
3711 return sv_2mortal(refcounted_he_value(chain));
3712 }
94250aee 3713 }
71622e40 3714 ret:
94250aee 3715 return flags & REFCOUNTED_HE_EXISTS ? NULL : &PL_sv_placeholder;
20439bc7 3716}
7b0bddfa 3717
20439bc7 3718/*
44170c9a 3719=for apidoc refcounted_he_fetch_pv
7b0bddfa 3720
20439bc7
Z
3721Like L</refcounted_he_fetch_pvn>, but takes a nul-terminated string
3722instead of a string/length pair.
3723
3724=cut
3725*/
3726
3727SV *
3728Perl_refcounted_he_fetch_pv(pTHX_ const struct refcounted_he *chain,
1604cfb0 3729 const char *key, U32 hash, U32 flags)
20439bc7
Z
3730{
3731 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PV;
3732 return refcounted_he_fetch_pvn(chain, key, strlen(key), hash, flags);
7b0bddfa
NC
3733}
3734
b3ca2e83 3735/*
44170c9a 3736=for apidoc refcounted_he_fetch_sv
20439bc7
Z
3737
3738Like L</refcounted_he_fetch_pvn>, but takes a Perl scalar instead of a
3739string/length pair.
3740
3741=cut
3742*/
b3ca2e83 3743
20439bc7
Z
3744SV *
3745Perl_refcounted_he_fetch_sv(pTHX_ const struct refcounted_he *chain,
1604cfb0 3746 SV *key, U32 hash, U32 flags)
20439bc7
Z
3747{
3748 const char *keypv;
3749 STRLEN keylen;
3750 PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_SV;
3751 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0
MS
3752 Perl_croak(aTHX_ "panic: refcounted_he_fetch_sv bad flags %" UVxf,
3753 (UV)flags);
20439bc7
Z
3754 keypv = SvPV_const(key, keylen);
3755 if (SvUTF8(key))
1604cfb0 3756 flags |= REFCOUNTED_HE_KEY_UTF8;
20439bc7 3757 if (!hash && SvIsCOW_shared_hash(key))
1604cfb0 3758 hash = SvSHARED_HASH(key);
20439bc7
Z
3759 return refcounted_he_fetch_pvn(chain, keypv, keylen, hash, flags);
3760}
3761
3762/*
44170c9a 3763=for apidoc refcounted_he_new_pvn
20439bc7
Z
3764
3765Creates a new C<refcounted_he>. This consists of a single key/value
3766pair and a reference to an existing C<refcounted_he> chain (which may
3767be empty), and thus forms a longer chain. When using the longer chain,
3768the new key/value pair takes precedence over any entry for the same key
3769further along the chain.
3770
2d7f6611 3771The new key is specified by C<keypv> and C<keylen>. If C<flags> has
20439bc7 3772the C<REFCOUNTED_HE_KEY_UTF8> bit set, the key octets are interpreted
2d7f6611 3773as UTF-8, otherwise they are interpreted as Latin-1. C<hash> is
20439bc7
Z
3774a precomputed hash of the key string, or zero if it has not been
3775precomputed.
3776
2d7f6611 3777C<value> is the scalar value to store for this key. C<value> is copied
20439bc7
Z
3778by this function, which thus does not take ownership of any reference
3779to it, and later changes to the scalar will not be reflected in the
3780value visible in the C<refcounted_he>. Complex types of scalar will not
3781be stored with referential integrity, but will be coerced to strings.
2d7f6611 3782C<value> may be either null or C<&PL_sv_placeholder> to indicate that no
20439bc7
Z
3783value is to be associated with the key; this, as with any non-null value,
3784takes precedence over the existence of a value for the key further along
3785the chain.
3786
2d7f6611 3787C<parent> points to the rest of the C<refcounted_he> chain to be
20439bc7 3788attached to the new C<refcounted_he>. This function takes ownership
2d7f6611 3789of one reference to C<parent>, and returns one reference to the new
20439bc7 3790C<refcounted_he>.
b3ca2e83
NC
3791
3792=cut
3793*/
3794
3795struct refcounted_he *
20439bc7 3796Perl_refcounted_he_new_pvn(pTHX_ struct refcounted_he *parent,
1604cfb0 3797 const char *keypv, STRLEN keylen, U32 hash, SV *value, U32 flags)
20439bc7 3798{
b6bbf3fa 3799 STRLEN value_len = 0;
95b63a38 3800 const char *value_p = NULL;
20439bc7 3801 bool is_pv;
b6bbf3fa 3802 char value_type;
20439bc7
Z
3803 char hekflags;
3804 STRLEN key_offset = 1;
3805 struct refcounted_he *he;
3806 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PVN;
b6bbf3fa 3807
20439bc7 3808 if (!value || value == &PL_sv_placeholder) {
1604cfb0 3809 value_type = HVrhek_delete;
20439bc7 3810 } else if (SvPOK(value)) {
1604cfb0 3811 value_type = HVrhek_PV;
b6bbf3fa 3812 } else if (SvIOK(value)) {
1604cfb0 3813 value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV;
b6bbf3fa 3814 } else if (!SvOK(value)) {
1604cfb0 3815 value_type = HVrhek_undef;
b6bbf3fa 3816 } else {
1604cfb0 3817 value_type = HVrhek_PV;
b6bbf3fa 3818 }
20439bc7
Z
3819 is_pv = value_type == HVrhek_PV;
3820 if (is_pv) {
1604cfb0
MS
3821 /* Do it this way so that the SvUTF8() test is after the SvPV, in case
3822 the value is overloaded, and doesn't yet have the UTF-8flag set. */
3823 value_p = SvPV_const(value, value_len);
3824 if (SvUTF8(value))
3825 value_type = HVrhek_PV_UTF8;
3826 key_offset = value_len + 2;
20439bc7
Z
3827 }
3828 hekflags = value_type;
3829
3830 if (flags & REFCOUNTED_HE_KEY_UTF8) {
1604cfb0
MS
3831 /* Canonicalise to Latin-1 where possible. */
3832 const char *keyend = keypv + keylen, *p;
3833 STRLEN nonascii_count = 0;
3834 for (p = keypv; p != keyend; p++) {
3835 if (! UTF8_IS_INVARIANT(*p)) {
3836 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) {
3837 goto canonicalised_key;
e8e5e5b3 3838 }
1604cfb0 3839 nonascii_count++;
e8e5e5b3 3840 p++;
1604cfb0
MS
3841 }
3842 }
3843 if (nonascii_count) {
3844 char *q;
3845 const char *p = keypv, *keyend = keypv + keylen;
3846 keylen -= nonascii_count;
3847 Newx(q, keylen, char);
3848 SAVEFREEPV(q);
3849 keypv = q;
3850 for (; p != keyend; p++, q++) {
3851 U8 c = (U8)*p;
e8e5e5b3
KW
3852 if (UTF8_IS_INVARIANT(c)) {
3853 *q = (char) c;
3854 }
3855 else {
3856 p++;
a62b247b 3857 *q = (char) EIGHT_BIT_UTF8_TO_NATIVE(c, *p);
e8e5e5b3 3858 }
1604cfb0
MS
3859 }
3860 }
3861 flags &= ~REFCOUNTED_HE_KEY_UTF8;
3862 canonicalised_key: ;
b6bbf3fa 3863 }
20439bc7 3864 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0 3865 hekflags |= HVhek_UTF8;
20439bc7 3866 if (!hash)
1604cfb0 3867 PERL_HASH(hash, keypv, keylen);
012da8e5 3868
0de694c5 3869#ifdef USE_ITHREADS
10edeb5d 3870 he = (struct refcounted_he*)
1604cfb0
MS
3871 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3872 + keylen
3873 + key_offset);
0de694c5
NC
3874#else
3875 he = (struct refcounted_he*)
1604cfb0
MS
3876 PerlMemShared_malloc(sizeof(struct refcounted_he) - 1
3877 + key_offset);
0de694c5 3878#endif
b3ca2e83 3879
71ad1b0c 3880 he->refcounted_he_next = parent;
b6bbf3fa 3881
012da8e5 3882 if (is_pv) {
1604cfb0
MS
3883 Copy(value_p, he->refcounted_he_data + 1, value_len + 1, char);
3884 he->refcounted_he_val.refcounted_he_u_len = value_len;
b6bbf3fa 3885 } else if (value_type == HVrhek_IV) {
1604cfb0 3886 he->refcounted_he_val.refcounted_he_u_iv = SvIVX(value);
012da8e5 3887 } else if (value_type == HVrhek_UV) {
1604cfb0 3888 he->refcounted_he_val.refcounted_he_u_uv = SvUVX(value);
b6bbf3fa
NC
3889 }
3890
cbb1fbea 3891#ifdef USE_ITHREADS
b6bbf3fa 3892 he->refcounted_he_hash = hash;
20439bc7
Z
3893 he->refcounted_he_keylen = keylen;
3894 Copy(keypv, he->refcounted_he_data + key_offset, keylen, char);
cbb1fbea 3895#else
20439bc7 3896 he->refcounted_he_hek = share_hek_flags(keypv, keylen, hash, hekflags);
cbb1fbea 3897#endif
b6bbf3fa 3898
20439bc7 3899 he->refcounted_he_data[0] = hekflags;
b3ca2e83
NC
3900 he->refcounted_he_refcnt = 1;
3901
3902 return he;
3903}
3904
3905/*
44170c9a 3906=for apidoc refcounted_he_new_pv
b3ca2e83 3907
20439bc7
Z
3908Like L</refcounted_he_new_pvn>, but takes a nul-terminated string instead
3909of a string/length pair.
3910
3911=cut
3912*/
3913
3914struct refcounted_he *
3915Perl_refcounted_he_new_pv(pTHX_ struct refcounted_he *parent,
1604cfb0 3916 const char *key, U32 hash, SV *value, U32 flags)
20439bc7
Z
3917{
3918 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PV;
3919 return refcounted_he_new_pvn(parent, key, strlen(key), hash, value, flags);
3920}
3921
3922/*
44170c9a 3923=for apidoc refcounted_he_new_sv
20439bc7
Z
3924
3925Like L</refcounted_he_new_pvn>, but takes a Perl scalar instead of a
3926string/length pair.
3927
3928=cut
3929*/
3930
3931struct refcounted_he *
3932Perl_refcounted_he_new_sv(pTHX_ struct refcounted_he *parent,
1604cfb0 3933 SV *key, U32 hash, SV *value, U32 flags)
20439bc7
Z
3934{
3935 const char *keypv;
3936 STRLEN keylen;
3937 PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_SV;
3938 if (flags & REFCOUNTED_HE_KEY_UTF8)
1604cfb0
MS
3939 Perl_croak(aTHX_ "panic: refcounted_he_new_sv bad flags %" UVxf,
3940 (UV)flags);
20439bc7
Z
3941 keypv = SvPV_const(key, keylen);
3942 if (SvUTF8(key))
1604cfb0 3943 flags |= REFCOUNTED_HE_KEY_UTF8;
20439bc7 3944 if (!hash && SvIsCOW_shared_hash(key))
1604cfb0 3945 hash = SvSHARED_HASH(key);
20439bc7
Z
3946 return refcounted_he_new_pvn(parent, keypv, keylen, hash, value, flags);
3947}
3948
3949/*
44170c9a 3950=for apidoc refcounted_he_free
20439bc7
Z
3951
3952Decrements the reference count of a C<refcounted_he> by one. If the
3953reference count reaches zero the structure's memory is freed, which
3954(recursively) causes a reduction of its parent C<refcounted_he>'s
3955reference count. It is safe to pass a null pointer to this function:
3956no action occurs in this case.
b3ca2e83
NC
3957
3958=cut
3959*/
3960
3961void
3962Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) {
57ca3b03
AL
3963 PERL_UNUSED_CONTEXT;
3964
b3ca2e83 3965 while (he) {
1604cfb0
MS
3966 struct refcounted_he *copy;
3967 U32 new_count;
3968
3969 HINTS_REFCNT_LOCK;
3970 new_count = --he->refcounted_he_refcnt;
3971 HINTS_REFCNT_UNLOCK;
3972
3973 if (new_count) {
3974 return;
3975 }
b3ca2e83 3976
b6bbf3fa 3977#ifndef USE_ITHREADS
1604cfb0 3978 unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0);
cbb1fbea 3979#endif
1604cfb0
MS
3980 copy = he;
3981 he = he->refcounted_he_next;
3982 PerlMemShared_free(copy);
b3ca2e83
NC
3983 }
3984}
3985
20439bc7 3986/*
44170c9a 3987=for apidoc refcounted_he_inc
20439bc7
Z
3988
3989Increment the reference count of a C<refcounted_he>. The pointer to the
3990C<refcounted_he> is also returned. It is safe to pass a null pointer
3991to this function: no action occurs and a null pointer is returned.
3992
3993=cut
3994*/
3995
3996struct refcounted_he *
3997Perl_refcounted_he_inc(pTHX_ struct refcounted_he *he)
3998{
dc3bf405 3999 PERL_UNUSED_CONTEXT;
20439bc7 4000 if (he) {
1604cfb0
MS
4001 HINTS_REFCNT_LOCK;
4002 he->refcounted_he_refcnt++;
4003 HINTS_REFCNT_UNLOCK;
20439bc7
Z
4004 }
4005 return he;
4006}
4007
8375c93e 4008/*
3f620621 4009=for apidoc_section $COP
aebc0cbe 4010=for apidoc cop_fetch_label
8375c93e 4011
7df56744
KW
4012Returns the label attached to a cop, and stores its length in bytes into
4013C<*len>.
4014Upon return, C<*flags> will be set to either C<SVf_UTF8> or 0.
4015
eb992c6f 4016Alternatively, use the macro C<L</CopLABEL_len_flags>>;
7df56744 4017or if you don't need to know if the label is UTF-8 or not, the macro
eb992c6f
KW
4018C<L</CopLABEL_len>>;
4019or if you additionally dont need to know the length, C<L</CopLABEL>>.
8375c93e
RU
4020
4021=cut
4022*/
4023
47550813
NC
4024/* pp_entereval is aware that labels are stored with a key ':' at the top of
4025 the linked list. */
dca6062a 4026const char *
aebc0cbe 4027Perl_cop_fetch_label(pTHX_ COP *const cop, STRLEN *len, U32 *flags) {
d6747b7a
NC
4028 struct refcounted_he *const chain = cop->cop_hints_hash;
4029
aebc0cbe 4030 PERL_ARGS_ASSERT_COP_FETCH_LABEL;
dc3bf405 4031 PERL_UNUSED_CONTEXT;
d6747b7a 4032
dca6062a 4033 if (!chain)
1604cfb0 4034 return NULL;
dca6062a
NC
4035#ifdef USE_ITHREADS
4036 if (chain->refcounted_he_keylen != 1)
1604cfb0 4037 return NULL;
dca6062a 4038 if (*REF_HE_KEY(chain) != ':')
1604cfb0 4039 return NULL;
dca6062a
NC
4040#else
4041 if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1)
1604cfb0 4042 return NULL;
dca6062a 4043 if (*HEK_KEY(chain->refcounted_he_hek) != ':')
1604cfb0 4044 return NULL;
dca6062a 4045#endif
012da8e5
NC
4046 /* Stop anyone trying to really mess us up by adding their own value for
4047 ':' into %^H */
4048 if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV
1604cfb0
MS
4049 && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8)
4050 return NULL;
012da8e5 4051
dca6062a 4052 if (len)
1604cfb0 4053 *len = chain->refcounted_he_val.refcounted_he_u_len;
dca6062a 4054 if (flags) {
1604cfb0
MS
4055 *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask)
4056 == HVrhek_PV_UTF8) ? SVf_UTF8 : 0;
dca6062a
NC
4057 }
4058 return chain->refcounted_he_data + 1;
4059}
4060
8375c93e 4061/*
aebc0cbe 4062=for apidoc cop_store_label
8375c93e 4063
72d33970
FC
4064Save a label into a C<cop_hints_hash>.
4065You need to set flags to C<SVf_UTF8>
5f608e5f 4066for a UTF-8 label. Any other flag is ignored.
8375c93e
RU
4067
4068=cut
4069*/
4070
a77ac40c 4071void
aebc0cbe 4072Perl_cop_store_label(pTHX_ COP *const cop, const char *label, STRLEN len,
1604cfb0 4073 U32 flags)
012da8e5 4074{
20439bc7 4075 SV *labelsv;
aebc0cbe 4076 PERL_ARGS_ASSERT_COP_STORE_LABEL;
547bb267 4077
a77ac40c 4078 if (flags & ~(SVf_UTF8))
1604cfb0
MS
4079 Perl_croak(aTHX_ "panic: cop_store_label illegal flag bits 0x%" UVxf,
4080 (UV)flags);
a3179684 4081 labelsv = newSVpvn_flags(label, len, SVs_TEMP);
20439bc7 4082 if (flags & SVf_UTF8)
1604cfb0 4083 SvUTF8_on(labelsv);
a77ac40c 4084 cop->cop_hints_hash
1604cfb0 4085 = refcounted_he_new_pvs(cop->cop_hints_hash, ":", labelsv, 0);
012da8e5
NC
4086}
4087
b3ca2e83 4088/*
3f620621 4089=for apidoc_section $HV
ecae49c0
NC
4090=for apidoc hv_assert
4091
4092Check that a hash is in an internally consistent state.
4093
4094=cut
4095*/
4096
943795c2
NC
4097#ifdef DEBUGGING
4098
ecae49c0
NC
4099void
4100Perl_hv_assert(pTHX_ HV *hv)
4101{
57ca3b03
AL
4102 HE* entry;
4103 int withflags = 0;
4104 int placeholders = 0;
4105 int real = 0;
4106 int bad = 0;
4107 const I32 riter = HvRITER_get(hv);
4108 HE *eiter = HvEITER_get(hv);
4109
7918f24d
NC
4110 PERL_ARGS_ASSERT_HV_ASSERT;
4111
57ca3b03
AL
4112 (void)hv_iterinit(hv);
4113
4114 while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) {
1604cfb0
MS
4115 /* sanity check the values */
4116 if (HeVAL(entry) == &PL_sv_placeholder)
4117 placeholders++;
4118 else
4119 real++;
4120 /* sanity check the keys */
4121 if (HeSVKEY(entry)) {
4122 NOOP; /* Don't know what to check on SV keys. */
4123 } else if (HeKUTF8(entry)) {
4124 withflags++;
4125 if (HeKWASUTF8(entry)) {
4126 PerlIO_printf(Perl_debug_log,
4127 "hash key has both WASUTF8 and UTF8: '%.*s'\n",
4128 (int) HeKLEN(entry), HeKEY(entry));
4129 bad = 1;
4130 }
4131 } else if (HeKWASUTF8(entry))
4132 withflags++;
57ca3b03 4133 }
ad64d0ec 4134 if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) {
1604cfb0
MS
4135 static const char bad_count[] = "Count %d %s(s), but hash reports %d\n";
4136 const int nhashkeys = HvUSEDKEYS(hv);
4137 const int nhashplaceholders = HvPLACEHOLDERS_get(hv);
4138
4139 if (nhashkeys != real) {
4140 PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys );
4141 bad = 1;
4142 }
4143 if (nhashplaceholders != placeholders) {
4144 PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders );
4145 bad = 1;
4146 }
57ca3b03
AL
4147 }
4148 if (withflags && ! HvHASKFLAGS(hv)) {
1604cfb0
MS
4149 PerlIO_printf(Perl_debug_log,
4150 "Hash has HASKFLAGS off but I count %d key(s) with flags\n",
4151 withflags);
4152 bad = 1;
57ca3b03
AL
4153 }
4154 if (bad) {
1604cfb0 4155 sv_dump(MUTABLE_SV(hv));
57ca3b03
AL
4156 }
4157 HvRITER_set(hv, riter); /* Restore hash iterator state */
4158 HvEITER_set(hv, eiter);
ecae49c0 4159}
af3babe4 4160
943795c2
NC
4161#endif
4162
af3babe4 4163/*
14d04a33 4164 * ex: set ts=8 sts=4 sw=4 et:
37442d52 4165 */