Commit | Line | Data |
---|---|---|
a0d0e21e | 1 | /* hv.c |
79072805 | 2 | * |
1129b882 NC |
3 | * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, |
4 | * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others | |
79072805 LW |
5 | * |
6 | * You may distribute under the terms of either the GNU General Public | |
7 | * License or the Artistic License, as specified in the README file. | |
8 | * | |
a0d0e21e LW |
9 | */ |
10 | ||
11 | /* | |
4ac71550 TC |
12 | * I sit beside the fire and think |
13 | * of all that I have seen. | |
14 | * --Bilbo | |
15 | * | |
16 | * [p.278 of _The Lord of the Rings_, II/iii: "The Ring Goes South"] | |
79072805 LW |
17 | */ |
18 | ||
d5afce77 RB |
19 | /* |
20 | =head1 Hash Manipulation Functions | |
db4fbf16 FC |
21 | A HV structure represents a Perl hash. It consists mainly of an array |
22 | of pointers, each of which points to a linked list of HE structures. The | |
166f8a29 | 23 | array is indexed by the hash function of the key, so each linked list |
db4fbf16 | 24 | represents all the hash entries with the same hash value. Each HE contains |
166f8a29 DM |
25 | a pointer to the actual value, plus a pointer to a HEK structure which |
26 | holds 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 | ||
8e317198 | 37 | #define DO_HSPLIT(xhv) ((xhv)->xhv_keys > (xhv)->xhv_max) /* HvTOTALKEYS(hv) > HvMAX(hv) */ |
9faf471a | 38 | #define HV_FILL_THRESHOLD 31 |
fdcd69b6 | 39 | |
d75ce684 | 40 | static const char S_strtab_error[] |
5d2b1485 NC |
41 | = "Cannot modify shared string table in hv_%s"; |
42 | ||
c941fb51 NC |
43 | #ifdef PURIFY |
44 | ||
45 | #define new_HE() (HE*)safemalloc(sizeof(HE)) | |
46 | #define del_HE(p) safefree((char*)p) | |
47 | ||
48 | #else | |
49 | ||
76e3520e | 50 | STATIC HE* |
cea2e8a9 | 51 | S_new_he(pTHX) |
4633a7c4 LW |
52 | { |
53 | HE* he; | |
0bd48802 | 54 | void ** const root = &PL_body_roots[HE_SVSLOT]; |
6a93a7e5 | 55 | |
6a93a7e5 | 56 | if (!*root) |
1e30fcd5 | 57 | Perl_more_bodies(aTHX_ HE_SVSLOT, sizeof(HE), PERL_ARENA_SIZE); |
10edeb5d | 58 | he = (HE*) *root; |
ce3e5c45 | 59 | assert(he); |
6a93a7e5 | 60 | *root = HeNEXT(he); |
333f433b | 61 | return he; |
4633a7c4 LW |
62 | } |
63 | ||
c941fb51 NC |
64 | #define new_HE() new_he() |
65 | #define del_HE(p) \ | |
66 | STMT_START { \ | |
6a93a7e5 NC |
67 | HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \ |
68 | PL_body_roots[HE_SVSLOT] = p; \ | |
c941fb51 | 69 | } STMT_END |
d33b2eba | 70 | |
d33b2eba | 71 | |
d33b2eba GS |
72 | |
73 | #endif | |
74 | ||
76e3520e | 75 | STATIC HEK * |
5f66b61c | 76 | S_save_hek_flags(const char *str, I32 len, U32 hash, int flags) |
bbce6d69 | 77 | { |
35a4481c | 78 | const int flags_masked = flags & HVhek_MASK; |
bbce6d69 | 79 | char *k; |
eb578fdb | 80 | HEK *hek; |
1c846c1f | 81 | |
7918f24d NC |
82 | PERL_ARGS_ASSERT_SAVE_HEK_FLAGS; |
83 | ||
a02a5408 | 84 | Newx(k, HEK_BASESIZE + len + 2, char); |
bbce6d69 | 85 | hek = (HEK*)k; |
ff68c719 | 86 | Copy(str, HEK_KEY(hek), len, char); |
e05949c7 | 87 | HEK_KEY(hek)[len] = 0; |
ff68c719 | 88 | HEK_LEN(hek) = len; |
89 | HEK_HASH(hek) = hash; | |
45e34800 | 90 | HEK_FLAGS(hek) = (unsigned char)flags_masked | HVhek_UNSHARED; |
dcf933a4 NC |
91 | |
92 | if (flags & HVhek_FREEKEY) | |
93 | Safefree(str); | |
bbce6d69 | 94 | return hek; |
95 | } | |
96 | ||
4a31713e | 97 | /* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent |
dd28f7bb DM |
98 | * for tied hashes */ |
99 | ||
100 | void | |
101 | Perl_free_tied_hv_pool(pTHX) | |
102 | { | |
dd28f7bb DM |
103 | HE *he = PL_hv_fetch_ent_mh; |
104 | while (he) { | |
9d4ba2ae | 105 | HE * const ohe = he; |
dd28f7bb | 106 | Safefree(HeKEY_hek(he)); |
dd28f7bb DM |
107 | he = HeNEXT(he); |
108 | del_HE(ohe); | |
109 | } | |
4608196e | 110 | PL_hv_fetch_ent_mh = NULL; |
dd28f7bb DM |
111 | } |
112 | ||
d18c6117 | 113 | #if defined(USE_ITHREADS) |
0bff533c NC |
114 | HEK * |
115 | Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param) | |
116 | { | |
566771cc | 117 | HEK *shared; |
9d4ba2ae | 118 | |
7918f24d | 119 | PERL_ARGS_ASSERT_HEK_DUP; |
9d4ba2ae | 120 | PERL_UNUSED_ARG(param); |
0bff533c | 121 | |
566771cc NC |
122 | if (!source) |
123 | return NULL; | |
124 | ||
125 | shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); | |
0bff533c NC |
126 | if (shared) { |
127 | /* We already shared this hash key. */ | |
454f1e26 | 128 | (void)share_hek_hek(shared); |
0bff533c NC |
129 | } |
130 | else { | |
658b4a4a | 131 | shared |
6e838c70 NC |
132 | = share_hek_flags(HEK_KEY(source), HEK_LEN(source), |
133 | HEK_HASH(source), HEK_FLAGS(source)); | |
658b4a4a | 134 | ptr_table_store(PL_ptr_table, source, shared); |
0bff533c | 135 | } |
658b4a4a | 136 | return shared; |
0bff533c NC |
137 | } |
138 | ||
d18c6117 | 139 | HE * |
5c4138a0 | 140 | Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param) |
d18c6117 GS |
141 | { |
142 | HE *ret; | |
143 | ||
7918f24d NC |
144 | PERL_ARGS_ASSERT_HE_DUP; |
145 | ||
d18c6117 | 146 | if (!e) |
4608196e | 147 | return NULL; |
7766f137 GS |
148 | /* look for it in the table first */ |
149 | ret = (HE*)ptr_table_fetch(PL_ptr_table, e); | |
150 | if (ret) | |
151 | return ret; | |
152 | ||
153 | /* create anew and remember what it is */ | |
d33b2eba | 154 | ret = new_HE(); |
7766f137 GS |
155 | ptr_table_store(PL_ptr_table, e, ret); |
156 | ||
d2d73c3e | 157 | HeNEXT(ret) = he_dup(HeNEXT(e),shared, param); |
dd28f7bb DM |
158 | if (HeKLEN(e) == HEf_SVKEY) { |
159 | char *k; | |
ad64d0ec | 160 | Newx(k, HEK_BASESIZE + sizeof(const SV *), char); |
dd28f7bb | 161 | HeKEY_hek(ret) = (HEK*)k; |
a09252eb | 162 | HeKEY_sv(ret) = sv_dup_inc(HeKEY_sv(e), param); |
dd28f7bb | 163 | } |
c21d1a0f | 164 | else if (shared) { |
0bff533c NC |
165 | /* This is hek_dup inlined, which seems to be important for speed |
166 | reasons. */ | |
1b6737cc | 167 | HEK * const source = HeKEY_hek(e); |
658b4a4a | 168 | HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); |
c21d1a0f NC |
169 | |
170 | if (shared) { | |
171 | /* We already shared this hash key. */ | |
454f1e26 | 172 | (void)share_hek_hek(shared); |
c21d1a0f NC |
173 | } |
174 | else { | |
658b4a4a | 175 | shared |
6e838c70 NC |
176 | = share_hek_flags(HEK_KEY(source), HEK_LEN(source), |
177 | HEK_HASH(source), HEK_FLAGS(source)); | |
658b4a4a | 178 | ptr_table_store(PL_ptr_table, source, shared); |
c21d1a0f | 179 | } |
658b4a4a | 180 | HeKEY_hek(ret) = shared; |
c21d1a0f | 181 | } |
d18c6117 | 182 | else |
19692e8d NC |
183 | HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e), |
184 | HeKFLAGS(e)); | |
a09252eb | 185 | HeVAL(ret) = sv_dup_inc(HeVAL(e), param); |
d18c6117 GS |
186 | return ret; |
187 | } | |
188 | #endif /* USE_ITHREADS */ | |
189 | ||
1b1f1335 | 190 | static void |
2393f1b9 JH |
191 | S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen, |
192 | const char *msg) | |
1b1f1335 | 193 | { |
1b6737cc | 194 | SV * const sv = sv_newmortal(); |
7918f24d NC |
195 | |
196 | PERL_ARGS_ASSERT_HV_NOTALLOWED; | |
197 | ||
19692e8d | 198 | if (!(flags & HVhek_FREEKEY)) { |
1b1f1335 NIS |
199 | sv_setpvn(sv, key, klen); |
200 | } | |
201 | else { | |
202 | /* Need to free saved eventually assign to mortal SV */ | |
34c3c4e3 | 203 | /* XXX is this line an error ???: SV *sv = sv_newmortal(); */ |
1b1f1335 NIS |
204 | sv_usepvn(sv, (char *) key, klen); |
205 | } | |
19692e8d | 206 | if (flags & HVhek_UTF8) { |
1b1f1335 NIS |
207 | SvUTF8_on(sv); |
208 | } | |
be2597df | 209 | Perl_croak(aTHX_ msg, SVfARG(sv)); |
1b1f1335 NIS |
210 | } |
211 | ||
fde52b5c | 212 | /* (klen == HEf_SVKEY) is special for MAGICAL hv entries, meaning key slot |
213 | * contains an SV* */ | |
214 | ||
34a6f7b4 NC |
215 | /* |
216 | =for apidoc hv_store | |
217 | ||
a05d6c5d TC |
218 | Stores an SV in a hash. The hash key is specified as C<key> and the |
219 | absolute value of C<klen> is the length of the key. If C<klen> is | |
220 | negative the key is assumed to be in UTF-8-encoded Unicode. The | |
221 | C<hash> parameter is the precomputed hash value; if it is zero then | |
222 | Perl will compute it. | |
223 | ||
224 | The return value will be | |
34a6f7b4 NC |
225 | NULL if the operation failed or if the value did not need to be actually |
226 | stored within the hash (as in the case of tied hashes). Otherwise it can | |
227 | be dereferenced to get the original C<SV*>. Note that the caller is | |
228 | responsible for suitably incrementing the reference count of C<val> before | |
229 | the call, and decrementing it if the function returned NULL. Effectively | |
230 | a successful hv_store takes ownership of one reference to C<val>. This is | |
231 | usually what you want; a newly created SV has a reference count of one, so | |
232 | if all your code does is create SVs then store them in a hash, hv_store | |
233 | will own the only reference to the new SV, and your code doesn't need to do | |
234 | anything further to tidy up. hv_store is not implemented as a call to | |
235 | hv_store_ent, and does not create a temporary SV for the key, so if your | |
236 | key data is not already in SV form then use hv_store in preference to | |
237 | hv_store_ent. | |
238 | ||
239 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more | |
240 | information on how to use this function on tied hashes. | |
241 | ||
34a6f7b4 NC |
242 | =for apidoc hv_store_ent |
243 | ||
244 | Stores C<val> in a hash. The hash key is specified as C<key>. The C<hash> | |
245 | parameter is the precomputed hash value; if it is zero then Perl will | |
246 | compute it. The return value is the new hash entry so created. It will be | |
247 | NULL if the operation failed or if the value did not need to be actually | |
248 | stored within the hash (as in the case of tied hashes). Otherwise the | |
249 | contents of the return value can be accessed using the C<He?> macros | |
250 | described here. Note that the caller is responsible for suitably | |
251 | incrementing the reference count of C<val> before the call, and | |
252 | decrementing it if the function returned NULL. Effectively a successful | |
253 | hv_store_ent takes ownership of one reference to C<val>. This is | |
254 | usually what you want; a newly created SV has a reference count of one, so | |
255 | if all your code does is create SVs then store them in a hash, hv_store | |
256 | will own the only reference to the new SV, and your code doesn't need to do | |
257 | anything further to tidy up. Note that hv_store_ent only reads the C<key>; | |
258 | unlike C<val> it does not take ownership of it, so maintaining the correct | |
259 | reference count on C<key> is entirely the caller's responsibility. hv_store | |
260 | is not implemented as a call to hv_store_ent, and does not create a temporary | |
261 | SV for the key, so if your key data is not already in SV form then use | |
262 | hv_store in preference to hv_store_ent. | |
263 | ||
264 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more | |
265 | information on how to use this function on tied hashes. | |
266 | ||
34a6f7b4 NC |
267 | =for apidoc hv_exists |
268 | ||
269 | Returns a boolean indicating whether the specified hash key exists. The | |
a05d6c5d TC |
270 | absolute value of C<klen> is the length of the key. If C<klen> is |
271 | negative the key is assumed to be in UTF-8-encoded Unicode. | |
34a6f7b4 | 272 | |
954c1994 GS |
273 | =for apidoc hv_fetch |
274 | ||
a05d6c5d TC |
275 | Returns the SV which corresponds to the specified key in the hash. |
276 | The absolute value of C<klen> is the length of the key. If C<klen> is | |
277 | negative the key is assumed to be in UTF-8-encoded Unicode. If | |
43d3b06a KW |
278 | C<lval> is set then the fetch will be part of a store. This means that if |
279 | there is no value in the hash associated with the given key, then one is | |
280 | created and a pointer to it is returned. The C<SV*> it points to can be | |
281 | assigned to. But always check that the | |
a05d6c5d | 282 | return value is non-null before dereferencing it to an C<SV*>. |
954c1994 | 283 | |
96f1132b | 284 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
954c1994 GS |
285 | information on how to use this function on tied hashes. |
286 | ||
34a6f7b4 NC |
287 | =for apidoc hv_exists_ent |
288 | ||
db4fbf16 FC |
289 | Returns a boolean indicating whether |
290 | the specified hash key exists. C<hash> | |
34a6f7b4 NC |
291 | can be a valid precomputed hash value, or 0 to ask for it to be |
292 | computed. | |
293 | ||
294 | =cut | |
295 | */ | |
296 | ||
d1be9408 | 297 | /* returns an HE * structure with the all fields set */ |
fde52b5c | 298 | /* note that hent_val will be a mortal sv for MAGICAL hashes */ |
954c1994 GS |
299 | /* |
300 | =for apidoc hv_fetch_ent | |
301 | ||
302 | Returns the hash entry which corresponds to the specified key in the hash. | |
303 | C<hash> must be a valid precomputed hash number for the given C<key>, or 0 | |
304 | if you want the function to compute it. IF C<lval> is set then the fetch | |
305 | will be part of a store. Make sure the return value is non-null before | |
b24b84ef | 306 | accessing it. The return value when C<hv> is a tied hash is a pointer to a |
954c1994 | 307 | static location, so be sure to make a copy of the structure if you need to |
1c846c1f | 308 | store it somewhere. |
954c1994 | 309 | |
96f1132b | 310 | See L<perlguts/"Understanding the Magic of Tied Hashes and Arrays"> for more |
954c1994 GS |
311 | information on how to use this function on tied hashes. |
312 | ||
313 | =cut | |
314 | */ | |
315 | ||
a038e571 NC |
316 | /* Common code for hv_delete()/hv_exists()/hv_fetch()/hv_store() */ |
317 | void * | |
318 | Perl_hv_common_key_len(pTHX_ HV *hv, const char *key, I32 klen_i32, | |
319 | const int action, SV *val, const U32 hash) | |
320 | { | |
321 | STRLEN klen; | |
322 | int flags; | |
323 | ||
7918f24d NC |
324 | PERL_ARGS_ASSERT_HV_COMMON_KEY_LEN; |
325 | ||
a038e571 NC |
326 | if (klen_i32 < 0) { |
327 | klen = -klen_i32; | |
328 | flags = HVhek_UTF8; | |
329 | } else { | |
330 | klen = klen_i32; | |
331 | flags = 0; | |
332 | } | |
333 | return hv_common(hv, NULL, key, klen, flags, action, val, hash); | |
334 | } | |
335 | ||
63c89345 | 336 | void * |
d3ba3f5c | 337 | Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, |
5aaab254 | 338 | int flags, int action, SV *val, U32 hash) |
113738bb | 339 | { |
27da23d5 | 340 | dVAR; |
b2c64049 | 341 | XPVHV* xhv; |
b2c64049 NC |
342 | HE *entry; |
343 | HE **oentry; | |
fde52b5c | 344 | SV *sv; |
da58a35d | 345 | bool is_utf8; |
113738bb | 346 | int masked_flags; |
3c84c864 | 347 | const int return_svp = action & HV_FETCH_JUST_SV; |
34dadc62 | 348 | HEK *keysv_hek = NULL; |
fde52b5c | 349 | |
350 | if (!hv) | |
a4fc7abc | 351 | return NULL; |
e4787c0c | 352 | if (SvTYPE(hv) == (svtype)SVTYPEMASK) |
8265e3d1 NC |
353 | return NULL; |
354 | ||
355 | assert(SvTYPE(hv) == SVt_PVHV); | |
fde52b5c | 356 | |
bdee33e4 | 357 | if (SvSMAGICAL(hv) && SvGMAGICAL(hv) && !(action & HV_DISABLE_UVAR_XKEY)) { |
fda2d18a | 358 | MAGIC* mg; |
ad64d0ec | 359 | if ((mg = mg_find((const SV *)hv, PERL_MAGIC_uvar))) { |
fda2d18a NC |
360 | struct ufuncs * const uf = (struct ufuncs *)mg->mg_ptr; |
361 | if (uf->uf_set == NULL) { | |
362 | SV* obj = mg->mg_obj; | |
363 | ||
364 | if (!keysv) { | |
59cd0e26 NC |
365 | keysv = newSVpvn_flags(key, klen, SVs_TEMP | |
366 | ((flags & HVhek_UTF8) | |
367 | ? SVf_UTF8 : 0)); | |
fda2d18a NC |
368 | } |
369 | ||
370 | mg->mg_obj = keysv; /* pass key */ | |
371 | uf->uf_index = action; /* pass action */ | |
ad64d0ec | 372 | magic_getuvar(MUTABLE_SV(hv), mg); |
fda2d18a NC |
373 | keysv = mg->mg_obj; /* may have changed */ |
374 | mg->mg_obj = obj; | |
375 | ||
376 | /* If the key may have changed, then we need to invalidate | |
377 | any passed-in computed hash value. */ | |
378 | hash = 0; | |
379 | } | |
380 | } | |
bdee33e4 | 381 | } |
113738bb | 382 | if (keysv) { |
e593d2fe AE |
383 | if (flags & HVhek_FREEKEY) |
384 | Safefree(key); | |
5c144d81 | 385 | key = SvPV_const(keysv, klen); |
113738bb | 386 | is_utf8 = (SvUTF8(keysv) != 0); |
44b87b50 NC |
387 | if (SvIsCOW_shared_hash(keysv)) { |
388 | flags = HVhek_KEYCANONICAL | (is_utf8 ? HVhek_UTF8 : 0); | |
389 | } else { | |
0ddecb91 | 390 | flags = is_utf8 ? HVhek_UTF8 : 0; |
44b87b50 | 391 | } |
113738bb | 392 | } else { |
c1fe5510 | 393 | is_utf8 = ((flags & HVhek_UTF8) ? TRUE : FALSE); |
113738bb | 394 | } |
113738bb | 395 | |
9dbc5603 | 396 | if (action & HV_DELETE) { |
3c84c864 | 397 | return (void *) hv_delete_common(hv, keysv, key, klen, |
0ddecb91 | 398 | flags, action, hash); |
9dbc5603 NC |
399 | } |
400 | ||
b2c64049 | 401 | xhv = (XPVHV*)SvANY(hv); |
7f66fda2 | 402 | if (SvMAGICAL(hv)) { |
6136c704 | 403 | if (SvRMAGICAL(hv) && !(action & (HV_FETCH_ISSTORE|HV_FETCH_ISEXISTS))) { |
ad64d0ec NC |
404 | if (mg_find((const SV *)hv, PERL_MAGIC_tied) |
405 | || SvGMAGICAL((const SV *)hv)) | |
e62cc96a | 406 | { |
3c84c864 | 407 | /* FIXME should be able to skimp on the HE/HEK here when |
7f66fda2 | 408 | HV_FETCH_JUST_SV is true. */ |
7f66fda2 | 409 | if (!keysv) { |
740cce10 NC |
410 | keysv = newSVpvn_utf8(key, klen, is_utf8); |
411 | } else { | |
7f66fda2 | 412 | keysv = newSVsv(keysv); |
113738bb | 413 | } |
44a2ac75 | 414 | sv = sv_newmortal(); |
ad64d0ec | 415 | mg_copy(MUTABLE_SV(hv), sv, (char *)keysv, HEf_SVKEY); |
7f66fda2 NC |
416 | |
417 | /* grab a fake HE/HEK pair from the pool or make a new one */ | |
418 | entry = PL_hv_fetch_ent_mh; | |
419 | if (entry) | |
420 | PL_hv_fetch_ent_mh = HeNEXT(entry); | |
421 | else { | |
422 | char *k; | |
423 | entry = new_HE(); | |
ad64d0ec | 424 | Newx(k, HEK_BASESIZE + sizeof(const SV *), char); |
7f66fda2 NC |
425 | HeKEY_hek(entry) = (HEK*)k; |
426 | } | |
4608196e | 427 | HeNEXT(entry) = NULL; |
7f66fda2 NC |
428 | HeSVKEY_set(entry, keysv); |
429 | HeVAL(entry) = sv; | |
430 | sv_upgrade(sv, SVt_PVLV); | |
431 | LvTYPE(sv) = 'T'; | |
432 | /* so we can free entry when freeing sv */ | |
ad64d0ec | 433 | LvTARG(sv) = MUTABLE_SV(entry); |
7f66fda2 NC |
434 | |
435 | /* XXX remove at some point? */ | |
436 | if (flags & HVhek_FREEKEY) | |
437 | Safefree(key); | |
438 | ||
3c84c864 NC |
439 | if (return_svp) { |
440 | return entry ? (void *) &HeVAL(entry) : NULL; | |
441 | } | |
442 | return (void *) entry; | |
113738bb | 443 | } |
7f66fda2 | 444 | #ifdef ENV_IS_CASELESS |
ad64d0ec | 445 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
7f66fda2 NC |
446 | U32 i; |
447 | for (i = 0; i < klen; ++i) | |
448 | if (isLOWER(key[i])) { | |
086cb327 NC |
449 | /* Would be nice if we had a routine to do the |
450 | copy and upercase in a single pass through. */ | |
0bd48802 | 451 | const char * const nkey = strupr(savepvn(key,klen)); |
086cb327 NC |
452 | /* Note that this fetch is for nkey (the uppercased |
453 | key) whereas the store is for key (the original) */ | |
63c89345 NC |
454 | void *result = hv_common(hv, NULL, nkey, klen, |
455 | HVhek_FREEKEY, /* free nkey */ | |
456 | 0 /* non-LVAL fetch */ | |
3c84c864 NC |
457 | | HV_DISABLE_UVAR_XKEY |
458 | | return_svp, | |
63c89345 NC |
459 | NULL /* no value */, |
460 | 0 /* compute hash */); | |
26488bcf | 461 | if (!result && (action & HV_FETCH_LVALUE)) { |
086cb327 NC |
462 | /* This call will free key if necessary. |
463 | Do it this way to encourage compiler to tail | |
464 | call optimise. */ | |
63c89345 NC |
465 | result = hv_common(hv, keysv, key, klen, flags, |
466 | HV_FETCH_ISSTORE | |
3c84c864 NC |
467 | | HV_DISABLE_UVAR_XKEY |
468 | | return_svp, | |
63c89345 | 469 | newSV(0), hash); |
086cb327 NC |
470 | } else { |
471 | if (flags & HVhek_FREEKEY) | |
472 | Safefree(key); | |
473 | } | |
63c89345 | 474 | return result; |
7f66fda2 | 475 | } |
902173a3 | 476 | } |
7f66fda2 NC |
477 | #endif |
478 | } /* ISFETCH */ | |
479 | else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) { | |
ad64d0ec NC |
480 | if (mg_find((const SV *)hv, PERL_MAGIC_tied) |
481 | || SvGMAGICAL((const SV *)hv)) { | |
b2c64049 NC |
482 | /* I don't understand why hv_exists_ent has svret and sv, |
483 | whereas hv_exists only had one. */ | |
9d4ba2ae | 484 | SV * const svret = sv_newmortal(); |
b2c64049 | 485 | sv = sv_newmortal(); |
7f66fda2 NC |
486 | |
487 | if (keysv || is_utf8) { | |
488 | if (!keysv) { | |
740cce10 | 489 | keysv = newSVpvn_utf8(key, klen, TRUE); |
7f66fda2 NC |
490 | } else { |
491 | keysv = newSVsv(keysv); | |
492 | } | |
ad64d0ec | 493 | mg_copy(MUTABLE_SV(hv), sv, (char *)sv_2mortal(keysv), HEf_SVKEY); |
b2c64049 | 494 | } else { |
ad64d0ec | 495 | mg_copy(MUTABLE_SV(hv), sv, key, klen); |
7f66fda2 | 496 | } |
b2c64049 NC |
497 | if (flags & HVhek_FREEKEY) |
498 | Safefree(key); | |
7f66fda2 NC |
499 | magic_existspack(svret, mg_find(sv, PERL_MAGIC_tiedelem)); |
500 | /* This cast somewhat evil, but I'm merely using NULL/ | |
501 | not NULL to return the boolean exists. | |
502 | And I know hv is not NULL. */ | |
3c84c864 | 503 | return SvTRUE(svret) ? (void *)hv : NULL; |
e7152ba2 | 504 | } |
7f66fda2 | 505 | #ifdef ENV_IS_CASELESS |
ad64d0ec | 506 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
7f66fda2 | 507 | /* XXX This code isn't UTF8 clean. */ |
a15d23f8 | 508 | char * const keysave = (char * const)key; |
b2c64049 NC |
509 | /* Will need to free this, so set FREEKEY flag. */ |
510 | key = savepvn(key,klen); | |
511 | key = (const char*)strupr((char*)key); | |
6136c704 | 512 | is_utf8 = FALSE; |
7f66fda2 | 513 | hash = 0; |
8b4f7dd5 | 514 | keysv = 0; |
b2c64049 NC |
515 | |
516 | if (flags & HVhek_FREEKEY) { | |
517 | Safefree(keysave); | |
518 | } | |
519 | flags |= HVhek_FREEKEY; | |
7f66fda2 | 520 | } |
902173a3 | 521 | #endif |
7f66fda2 | 522 | } /* ISEXISTS */ |
b2c64049 NC |
523 | else if (action & HV_FETCH_ISSTORE) { |
524 | bool needs_copy; | |
525 | bool needs_store; | |
526 | hv_magic_check (hv, &needs_copy, &needs_store); | |
527 | if (needs_copy) { | |
9a9b5ec9 | 528 | const bool save_taint = TAINT_get; |
b2c64049 NC |
529 | if (keysv || is_utf8) { |
530 | if (!keysv) { | |
740cce10 | 531 | keysv = newSVpvn_utf8(key, klen, TRUE); |
b2c64049 | 532 | } |
284167a5 S |
533 | if (TAINTING_get) |
534 | TAINT_set(SvTAINTED(keysv)); | |
b2c64049 | 535 | keysv = sv_2mortal(newSVsv(keysv)); |
ad64d0ec | 536 | mg_copy(MUTABLE_SV(hv), val, (char*)keysv, HEf_SVKEY); |
b2c64049 | 537 | } else { |
ad64d0ec | 538 | mg_copy(MUTABLE_SV(hv), val, key, klen); |
b2c64049 NC |
539 | } |
540 | ||
541 | TAINT_IF(save_taint); | |
9a9b5ec9 DM |
542 | #ifdef NO_TAINT_SUPPORT |
543 | PERL_UNUSED_VAR(save_taint); | |
544 | #endif | |
1baaf5d7 | 545 | if (!needs_store) { |
b2c64049 NC |
546 | if (flags & HVhek_FREEKEY) |
547 | Safefree(key); | |
4608196e | 548 | return NULL; |
b2c64049 NC |
549 | } |
550 | #ifdef ENV_IS_CASELESS | |
ad64d0ec | 551 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
b2c64049 NC |
552 | /* XXX This code isn't UTF8 clean. */ |
553 | const char *keysave = key; | |
554 | /* Will need to free this, so set FREEKEY flag. */ | |
555 | key = savepvn(key,klen); | |
556 | key = (const char*)strupr((char*)key); | |
6136c704 | 557 | is_utf8 = FALSE; |
b2c64049 | 558 | hash = 0; |
8b4f7dd5 | 559 | keysv = 0; |
b2c64049 NC |
560 | |
561 | if (flags & HVhek_FREEKEY) { | |
562 | Safefree(keysave); | |
563 | } | |
564 | flags |= HVhek_FREEKEY; | |
565 | } | |
566 | #endif | |
567 | } | |
568 | } /* ISSTORE */ | |
7f66fda2 | 569 | } /* SvMAGICAL */ |
fde52b5c | 570 | |
7b2c381c | 571 | if (!HvARRAY(hv)) { |
b2c64049 | 572 | if ((action & (HV_FETCH_LVALUE | HV_FETCH_ISSTORE)) |
fde52b5c | 573 | #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */ |
ad64d0ec NC |
574 | || (SvRMAGICAL((const SV *)hv) |
575 | && mg_find((const SV *)hv, PERL_MAGIC_env)) | |
fde52b5c | 576 | #endif |
d58e6666 NC |
577 | ) { |
578 | char *array; | |
a02a5408 | 579 | Newxz(array, |
cbec9347 | 580 | PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), |
d58e6666 NC |
581 | char); |
582 | HvARRAY(hv) = (HE**)array; | |
583 | } | |
7f66fda2 NC |
584 | #ifdef DYNAMIC_ENV_FETCH |
585 | else if (action & HV_FETCH_ISEXISTS) { | |
586 | /* for an %ENV exists, if we do an insert it's by a recursive | |
587 | store call, so avoid creating HvARRAY(hv) right now. */ | |
588 | } | |
589 | #endif | |
113738bb NC |
590 | else { |
591 | /* XXX remove at some point? */ | |
592 | if (flags & HVhek_FREEKEY) | |
593 | Safefree(key); | |
594 | ||
3c84c864 | 595 | return NULL; |
113738bb | 596 | } |
fde52b5c | 597 | } |
598 | ||
37ae23ff | 599 | if (is_utf8 && !(flags & HVhek_KEYCANONICAL)) { |
41d88b63 | 600 | char * const keysave = (char *)key; |
f9a63242 | 601 | key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); |
19692e8d | 602 | if (is_utf8) |
c1fe5510 NC |
603 | flags |= HVhek_UTF8; |
604 | else | |
605 | flags &= ~HVhek_UTF8; | |
7f66fda2 NC |
606 | if (key != keysave) { |
607 | if (flags & HVhek_FREEKEY) | |
608 | Safefree(keysave); | |
19692e8d | 609 | flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
527df579 NC |
610 | /* If the caller calculated a hash, it was on the sequence of |
611 | octets that are the UTF-8 form. We've now changed the sequence | |
612 | of octets stored to that of the equivalent byte representation, | |
613 | so the hash we need is different. */ | |
614 | hash = 0; | |
7f66fda2 | 615 | } |
19692e8d | 616 | } |
f9a63242 | 617 | |
34dadc62 DM |
618 | if (keysv && (SvIsCOW_shared_hash(keysv))) { |
619 | if (HvSHAREKEYS(hv)) | |
620 | keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv)); | |
621 | hash = SvSHARED_HASH(keysv); | |
7dc86639 | 622 | } |
34dadc62 DM |
623 | else if (!hash) |
624 | PERL_HASH(hash, key, klen); | |
effa1e2d | 625 | |
113738bb NC |
626 | masked_flags = (flags & HVhek_MASK); |
627 | ||
7f66fda2 | 628 | #ifdef DYNAMIC_ENV_FETCH |
4608196e | 629 | if (!HvARRAY(hv)) entry = NULL; |
7f66fda2 NC |
630 | else |
631 | #endif | |
b2c64049 | 632 | { |
7b2c381c | 633 | entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)]; |
b2c64049 | 634 | } |
34dadc62 DM |
635 | |
636 | if (!entry) | |
637 | goto not_found; | |
638 | ||
639 | if (keysv_hek) { | |
640 | /* keysv is actually a HEK in disguise, so we can match just by | |
641 | * comparing the HEK pointers in the HE chain. There is a slight | |
642 | * caveat: on something like "\x80", which has both plain and utf8 | |
643 | * representations, perl's hashes do encoding-insensitive lookups, | |
644 | * but preserve the encoding of the stored key. Thus a particular | |
645 | * key could map to two different HEKs in PL_strtab. We only | |
646 | * conclude 'not found' if all the flags are the same; otherwise | |
647 | * we fall back to a full search (this should only happen in rare | |
648 | * cases). | |
649 | */ | |
650 | int keysv_flags = HEK_FLAGS(keysv_hek); | |
651 | HE *orig_entry = entry; | |
652 | ||
653 | for (; entry; entry = HeNEXT(entry)) { | |
654 | HEK *hek = HeKEY_hek(entry); | |
655 | if (hek == keysv_hek) | |
656 | goto found; | |
657 | if (HEK_FLAGS(hek) != keysv_flags) | |
658 | break; /* need to do full match */ | |
659 | } | |
660 | if (!entry) | |
661 | goto not_found; | |
662 | /* failed on shortcut - do full search loop */ | |
663 | entry = orig_entry; | |
664 | } | |
665 | ||
0298d7b9 | 666 | for (; entry; entry = HeNEXT(entry)) { |
fde52b5c | 667 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
668 | continue; | |
eb160463 | 669 | if (HeKLEN(entry) != (I32)klen) |
fde52b5c | 670 | continue; |
34dadc62 | 671 | if (memNE(HeKEY(entry),key,klen)) /* is this it? */ |
fde52b5c | 672 | continue; |
113738bb | 673 | if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8) |
c3654f1a | 674 | continue; |
b2c64049 | 675 | |
34dadc62 | 676 | found: |
b2c64049 NC |
677 | if (action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE)) { |
678 | if (HeKFLAGS(entry) != masked_flags) { | |
679 | /* We match if HVhek_UTF8 bit in our flags and hash key's | |
680 | match. But if entry was set previously with HVhek_WASUTF8 | |
681 | and key now doesn't (or vice versa) then we should change | |
682 | the key's flag, as this is assignment. */ | |
683 | if (HvSHAREKEYS(hv)) { | |
684 | /* Need to swap the key we have for a key with the flags we | |
685 | need. As keys are shared we can't just write to the | |
686 | flag, so we share the new one, unshare the old one. */ | |
6136c704 | 687 | HEK * const new_hek = share_hek_flags(key, klen, hash, |
6e838c70 | 688 | masked_flags); |
b2c64049 NC |
689 | unshare_hek (HeKEY_hek(entry)); |
690 | HeKEY_hek(entry) = new_hek; | |
691 | } | |
5d2b1485 NC |
692 | else if (hv == PL_strtab) { |
693 | /* PL_strtab is usually the only hash without HvSHAREKEYS, | |
694 | so putting this test here is cheap */ | |
695 | if (flags & HVhek_FREEKEY) | |
696 | Safefree(key); | |
697 | Perl_croak(aTHX_ S_strtab_error, | |
698 | action & HV_FETCH_LVALUE ? "fetch" : "store"); | |
699 | } | |
b2c64049 NC |
700 | else |
701 | HeKFLAGS(entry) = masked_flags; | |
702 | if (masked_flags & HVhek_ENABLEHVKFLAGS) | |
703 | HvHASKFLAGS_on(hv); | |
704 | } | |
705 | if (HeVAL(entry) == &PL_sv_placeholder) { | |
706 | /* yes, can store into placeholder slot */ | |
707 | if (action & HV_FETCH_LVALUE) { | |
708 | if (SvMAGICAL(hv)) { | |
709 | /* This preserves behaviour with the old hv_fetch | |
710 | implementation which at this point would bail out | |
711 | with a break; (at "if we find a placeholder, we | |
712 | pretend we haven't found anything") | |
713 | ||
714 | That break mean that if a placeholder were found, it | |
715 | caused a call into hv_store, which in turn would | |
716 | check magic, and if there is no magic end up pretty | |
717 | much back at this point (in hv_store's code). */ | |
718 | break; | |
719 | } | |
486ec47a | 720 | /* LVAL fetch which actually needs a store. */ |
561b68a9 | 721 | val = newSV(0); |
ca732855 | 722 | HvPLACEHOLDERS(hv)--; |
b2c64049 NC |
723 | } else { |
724 | /* store */ | |
725 | if (val != &PL_sv_placeholder) | |
ca732855 | 726 | HvPLACEHOLDERS(hv)--; |
b2c64049 NC |
727 | } |
728 | HeVAL(entry) = val; | |
729 | } else if (action & HV_FETCH_ISSTORE) { | |
cefd5c7c | 730 | SvREFCNT_dec(HeVAL(entry)); |
b2c64049 NC |
731 | HeVAL(entry) = val; |
732 | } | |
27bcc0a7 | 733 | } else if (HeVAL(entry) == &PL_sv_placeholder) { |
b2c64049 NC |
734 | /* if we find a placeholder, we pretend we haven't found |
735 | anything */ | |
8aacddc1 | 736 | break; |
b2c64049 | 737 | } |
113738bb NC |
738 | if (flags & HVhek_FREEKEY) |
739 | Safefree(key); | |
3c84c864 NC |
740 | if (return_svp) { |
741 | return entry ? (void *) &HeVAL(entry) : NULL; | |
742 | } | |
fde52b5c | 743 | return entry; |
744 | } | |
34dadc62 DM |
745 | |
746 | not_found: | |
fde52b5c | 747 | #ifdef DYNAMIC_ENV_FETCH /* %ENV lookup? If so, try to fetch the value now */ |
0ed29950 | 748 | if (!(action & HV_FETCH_ISSTORE) |
ad64d0ec NC |
749 | && SvRMAGICAL((const SV *)hv) |
750 | && mg_find((const SV *)hv, PERL_MAGIC_env)) { | |
a6c40364 | 751 | unsigned long len; |
9d4ba2ae | 752 | const char * const env = PerlEnv_ENVgetenv_len(key,&len); |
a6c40364 GS |
753 | if (env) { |
754 | sv = newSVpvn(env,len); | |
755 | SvTAINTED_on(sv); | |
d3ba3f5c | 756 | return hv_common(hv, keysv, key, klen, flags, |
3c84c864 NC |
757 | HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp, |
758 | sv, hash); | |
a6c40364 | 759 | } |
fde52b5c | 760 | } |
761 | #endif | |
7f66fda2 NC |
762 | |
763 | if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) { | |
c445ea15 | 764 | hv_notallowed(flags, key, klen, |
c8cd6465 NC |
765 | "Attempt to access disallowed key '%"SVf"' in" |
766 | " a restricted hash"); | |
1b1f1335 | 767 | } |
b2c64049 NC |
768 | if (!(action & (HV_FETCH_LVALUE|HV_FETCH_ISSTORE))) { |
769 | /* Not doing some form of store, so return failure. */ | |
770 | if (flags & HVhek_FREEKEY) | |
771 | Safefree(key); | |
3c84c864 | 772 | return NULL; |
b2c64049 | 773 | } |
113738bb | 774 | if (action & HV_FETCH_LVALUE) { |
df5f182b | 775 | val = action & HV_FETCH_EMPTY_HE ? NULL : newSV(0); |
b2c64049 NC |
776 | if (SvMAGICAL(hv)) { |
777 | /* At this point the old hv_fetch code would call to hv_store, | |
778 | which in turn might do some tied magic. So we need to make that | |
779 | magic check happen. */ | |
780 | /* gonna assign to this, so it better be there */ | |
fda2d18a NC |
781 | /* If a fetch-as-store fails on the fetch, then the action is to |
782 | recurse once into "hv_store". If we didn't do this, then that | |
783 | recursive call would call the key conversion routine again. | |
784 | However, as we replace the original key with the converted | |
785 | key, this would result in a double conversion, which would show | |
e987ad1c FC |
786 | up as a bug if the conversion routine is not idempotent. |
787 | Hence the use of HV_DISABLE_UVAR_XKEY. */ | |
d3ba3f5c | 788 | return hv_common(hv, keysv, key, klen, flags, |
3c84c864 NC |
789 | HV_FETCH_ISSTORE|HV_DISABLE_UVAR_XKEY|return_svp, |
790 | val, hash); | |
b2c64049 NC |
791 | /* XXX Surely that could leak if the fetch-was-store fails? |
792 | Just like the hv_fetch. */ | |
113738bb NC |
793 | } |
794 | } | |
795 | ||
b2c64049 NC |
796 | /* Welcome to hv_store... */ |
797 | ||
7b2c381c | 798 | if (!HvARRAY(hv)) { |
b2c64049 NC |
799 | /* Not sure if we can get here. I think the only case of oentry being |
800 | NULL is for %ENV with dynamic env fetch. But that should disappear | |
801 | with magic in the previous code. */ | |
d58e6666 | 802 | char *array; |
a02a5408 | 803 | Newxz(array, |
b2c64049 | 804 | PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), |
d58e6666 NC |
805 | char); |
806 | HvARRAY(hv) = (HE**)array; | |
b2c64049 NC |
807 | } |
808 | ||
7b2c381c | 809 | oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max]; |
ab4af705 | 810 | |
b2c64049 NC |
811 | entry = new_HE(); |
812 | /* share_hek_flags will do the free for us. This might be considered | |
813 | bad API design. */ | |
814 | if (HvSHAREKEYS(hv)) | |
6e838c70 | 815 | HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags); |
5d2b1485 NC |
816 | else if (hv == PL_strtab) { |
817 | /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting | |
818 | this test here is cheap */ | |
819 | if (flags & HVhek_FREEKEY) | |
820 | Safefree(key); | |
821 | Perl_croak(aTHX_ S_strtab_error, | |
822 | action & HV_FETCH_LVALUE ? "fetch" : "store"); | |
823 | } | |
b2c64049 NC |
824 | else /* gotta do the real thing */ |
825 | HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags); | |
826 | HeVAL(entry) = val; | |
3078e109 | 827 | |
9faf471a NC |
828 | if (!*oentry && SvOOK(hv)) { |
829 | /* initial entry, and aux struct present. */ | |
830 | struct xpvhv_aux *const aux = HvAUX(hv); | |
831 | if (aux->xhv_fill_lazy) | |
832 | ++aux->xhv_fill_lazy; | |
833 | } | |
834 | ||
6a5b4183 | 835 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
3078e109 YO |
836 | /* This logic semi-randomizes the insert order in a bucket. |
837 | * Either we insert into the top, or the slot below the top, | |
d5fc06cb YO |
838 | * making it harder to see if there is a collision. We also |
839 | * reset the iterator randomizer if there is one. | |
3078e109 | 840 | */ |
6a5b4183 YO |
841 | if ( *oentry && PL_HASH_RAND_BITS_ENABLED) { |
842 | PL_hash_rand_bits++; | |
843 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); | |
844 | if ( PL_hash_rand_bits & 1 ) { | |
845 | HeNEXT(entry) = HeNEXT(*oentry); | |
846 | HeNEXT(*oentry) = entry; | |
847 | } else { | |
848 | HeNEXT(entry) = *oentry; | |
849 | *oentry = entry; | |
850 | } | |
851 | } else | |
852 | #endif | |
853 | { | |
3078e109 YO |
854 | HeNEXT(entry) = *oentry; |
855 | *oentry = entry; | |
3078e109 | 856 | } |
6a5b4183 | 857 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
3a714294 | 858 | if (SvOOK(hv)) { |
ff20b672 YO |
859 | /* Currently this makes various tests warn in annoying ways. |
860 | * So Silenced for now. - Yves | bogus end of comment =>* / | |
861 | if (HvAUX(hv)->xhv_riter != -1) { | |
862 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), | |
863 | "[TESTING] Inserting into a hash during each() traversal results in undefined behavior" | |
864 | pTHX__FORMAT | |
865 | pTHX__VALUE); | |
866 | } | |
867 | */ | |
6a5b4183 YO |
868 | if (PL_HASH_RAND_BITS_ENABLED) { |
869 | if (PL_HASH_RAND_BITS_ENABLED == 1) | |
870 | PL_hash_rand_bits += (PTRV)entry + 1; /* we don't bother to use ptr_hash here */ | |
871 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); | |
872 | } | |
3a714294 YO |
873 | HvAUX(hv)->xhv_rand= (U32)PL_hash_rand_bits; |
874 | } | |
6a5b4183 | 875 | #endif |
b2c64049 NC |
876 | |
877 | if (val == &PL_sv_placeholder) | |
ca732855 | 878 | HvPLACEHOLDERS(hv)++; |
b2c64049 NC |
879 | if (masked_flags & HVhek_ENABLEHVKFLAGS) |
880 | HvHASKFLAGS_on(hv); | |
881 | ||
8e317198 YO |
882 | xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */ |
883 | if ( DO_HSPLIT(xhv) ) { | |
adf6906b | 884 | const STRLEN oldsize = xhv->xhv_max + 1; |
81a3ba35 | 885 | const U32 items = (U32)HvPLACEHOLDERS_get(hv); |
adf6906b | 886 | |
81a3ba35 | 887 | if (items /* hash has placeholders */ |
1eaee784 NC |
888 | && !SvREADONLY(hv) /* but is not a restricted hash */) { |
889 | /* If this hash previously was a "restricted hash" and had | |
890 | placeholders, but the "restricted" flag has been turned off, | |
891 | then the placeholders no longer serve any useful purpose. | |
892 | However, they have the downsides of taking up RAM, and adding | |
893 | extra steps when finding used values. It's safe to clear them | |
894 | at this point, even though Storable rebuilds restricted hashes by | |
0ca1b5c3 | 895 | putting in all the placeholders (first) before turning on the |
1eaee784 NC |
896 | readonly flag, because Storable always pre-splits the hash. |
897 | If we're lucky, then we may clear sufficient placeholders to | |
898 | avoid needing to split the hash at all. */ | |
81a3ba35 | 899 | clear_placeholders(hv, items); |
1eaee784 NC |
900 | if (DO_HSPLIT(xhv)) |
901 | hsplit(hv, oldsize, oldsize * 2); | |
902 | } else | |
903 | hsplit(hv, oldsize, oldsize * 2); | |
fde52b5c | 904 | } |
b2c64049 | 905 | |
3c84c864 NC |
906 | if (return_svp) { |
907 | return entry ? (void *) &HeVAL(entry) : NULL; | |
908 | } | |
909 | return (void *) entry; | |
fde52b5c | 910 | } |
911 | ||
864dbfa3 | 912 | STATIC void |
b0e6ae5b | 913 | S_hv_magic_check(HV *hv, bool *needs_copy, bool *needs_store) |
d0066dc7 | 914 | { |
a3b680e6 | 915 | const MAGIC *mg = SvMAGIC(hv); |
7918f24d NC |
916 | |
917 | PERL_ARGS_ASSERT_HV_MAGIC_CHECK; | |
918 | ||
d0066dc7 OT |
919 | *needs_copy = FALSE; |
920 | *needs_store = TRUE; | |
921 | while (mg) { | |
922 | if (isUPPER(mg->mg_type)) { | |
923 | *needs_copy = TRUE; | |
d60c5a05 | 924 | if (mg->mg_type == PERL_MAGIC_tied) { |
d0066dc7 | 925 | *needs_store = FALSE; |
4ab2a30b | 926 | return; /* We've set all there is to set. */ |
d0066dc7 OT |
927 | } |
928 | } | |
929 | mg = mg->mg_moremagic; | |
930 | } | |
931 | } | |
932 | ||
954c1994 | 933 | /* |
a3bcc51e TP |
934 | =for apidoc hv_scalar |
935 | ||
72d33970 FC |
936 | Evaluates the hash in scalar context and returns the result. Handles magic |
937 | when the hash is tied. | |
a3bcc51e TP |
938 | |
939 | =cut | |
940 | */ | |
941 | ||
942 | SV * | |
943 | Perl_hv_scalar(pTHX_ HV *hv) | |
944 | { | |
a3bcc51e | 945 | SV *sv; |
823a54a3 | 946 | |
7918f24d NC |
947 | PERL_ARGS_ASSERT_HV_SCALAR; |
948 | ||
823a54a3 | 949 | if (SvRMAGICAL(hv)) { |
ad64d0ec | 950 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_tied); |
823a54a3 AL |
951 | if (mg) |
952 | return magic_scalarpack(hv, mg); | |
953 | } | |
a3bcc51e TP |
954 | |
955 | sv = sv_newmortal(); | |
f4431c56 | 956 | if (HvTOTALKEYS((const HV *)hv)) |
a3bcc51e TP |
957 | Perl_sv_setpvf(aTHX_ sv, "%ld/%ld", |
958 | (long)HvFILL(hv), (long)HvMAX(hv) + 1); | |
959 | else | |
960 | sv_setiv(sv, 0); | |
961 | ||
962 | return sv; | |
963 | } | |
964 | ||
965 | /* | |
954c1994 GS |
966 | =for apidoc hv_delete |
967 | ||
a05d6c5d TC |
968 | Deletes a key/value pair in the hash. The value's SV is removed from |
969 | the hash, made mortal, and returned to the caller. The absolute | |
970 | value of C<klen> is the length of the key. If C<klen> is negative the | |
971 | key is assumed to be in UTF-8-encoded Unicode. The C<flags> value | |
972 | will normally be zero; if set to G_DISCARD then NULL will be returned. | |
973 | NULL will also be returned if the key is not found. | |
954c1994 | 974 | |
954c1994 GS |
975 | =for apidoc hv_delete_ent |
976 | ||
3025a2e4 CS |
977 | Deletes a key/value pair in the hash. The value SV is removed from the hash, |
978 | made mortal, and returned to the caller. The C<flags> value will normally be | |
979 | zero; if set to G_DISCARD then NULL will be returned. NULL will also be | |
980 | returned if the key is not found. C<hash> can be a valid precomputed hash | |
981 | value, or 0 to ask for it to be computed. | |
954c1994 GS |
982 | |
983 | =cut | |
984 | */ | |
985 | ||
8f8d40ab | 986 | STATIC SV * |
cd6d36ac NC |
987 | S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, |
988 | int k_flags, I32 d_flags, U32 hash) | |
f1317c8d | 989 | { |
27da23d5 | 990 | dVAR; |
eb578fdb KW |
991 | XPVHV* xhv; |
992 | HE *entry; | |
993 | HE **oentry; | |
34dadc62 | 994 | HE **first_entry; |
9dbc5603 | 995 | bool is_utf8 = (k_flags & HVhek_UTF8) ? TRUE : FALSE; |
7a9669ca | 996 | int masked_flags; |
34dadc62 DM |
997 | HEK *keysv_hek = NULL; |
998 | U8 mro_changes = 0; /* 1 = isa; 2 = package moved */ | |
999 | SV *sv; | |
1000 | GV *gv = NULL; | |
1001 | HV *stash = NULL; | |
1c846c1f | 1002 | |
fde52b5c | 1003 | if (SvRMAGICAL(hv)) { |
0a0bb7c7 OT |
1004 | bool needs_copy; |
1005 | bool needs_store; | |
1006 | hv_magic_check (hv, &needs_copy, &needs_store); | |
1007 | ||
f1317c8d | 1008 | if (needs_copy) { |
6136c704 | 1009 | SV *sv; |
63c89345 NC |
1010 | entry = (HE *) hv_common(hv, keysv, key, klen, |
1011 | k_flags & ~HVhek_FREEKEY, | |
1012 | HV_FETCH_LVALUE|HV_DISABLE_UVAR_XKEY, | |
1013 | NULL, hash); | |
7a9669ca | 1014 | sv = entry ? HeVAL(entry) : NULL; |
f1317c8d NC |
1015 | if (sv) { |
1016 | if (SvMAGICAL(sv)) { | |
1017 | mg_clear(sv); | |
1018 | } | |
1019 | if (!needs_store) { | |
1020 | if (mg_find(sv, PERL_MAGIC_tiedelem)) { | |
1021 | /* No longer an element */ | |
1022 | sv_unmagic(sv, PERL_MAGIC_tiedelem); | |
1023 | return sv; | |
1024 | } | |
a0714e2c | 1025 | return NULL; /* element cannot be deleted */ |
f1317c8d | 1026 | } |
902173a3 | 1027 | #ifdef ENV_IS_CASELESS |
ad64d0ec | 1028 | else if (mg_find((const SV *)hv, PERL_MAGIC_env)) { |
8167a60a | 1029 | /* XXX This code isn't UTF8 clean. */ |
59cd0e26 | 1030 | keysv = newSVpvn_flags(key, klen, SVs_TEMP); |
8167a60a NC |
1031 | if (k_flags & HVhek_FREEKEY) { |
1032 | Safefree(key); | |
1033 | } | |
1034 | key = strupr(SvPVX(keysv)); | |
1035 | is_utf8 = 0; | |
1036 | k_flags = 0; | |
1037 | hash = 0; | |
7f66fda2 | 1038 | } |
510ac311 | 1039 | #endif |
2fd1c6b8 | 1040 | } |
2fd1c6b8 | 1041 | } |
fde52b5c | 1042 | } |
cbec9347 | 1043 | xhv = (XPVHV*)SvANY(hv); |
7b2c381c | 1044 | if (!HvARRAY(hv)) |
a0714e2c | 1045 | return NULL; |
fde52b5c | 1046 | |
6b230254 | 1047 | if (is_utf8 && !(k_flags & HVhek_KEYCANONICAL)) { |
c445ea15 | 1048 | const char * const keysave = key; |
b464bac0 | 1049 | key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); |
cd6d36ac | 1050 | |
19692e8d | 1051 | if (is_utf8) |
cd6d36ac NC |
1052 | k_flags |= HVhek_UTF8; |
1053 | else | |
1054 | k_flags &= ~HVhek_UTF8; | |
7f66fda2 NC |
1055 | if (key != keysave) { |
1056 | if (k_flags & HVhek_FREEKEY) { | |
1057 | /* This shouldn't happen if our caller does what we expect, | |
1058 | but strictly the API allows it. */ | |
1059 | Safefree(keysave); | |
1060 | } | |
1061 | k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY; | |
1062 | } | |
ad64d0ec | 1063 | HvHASKFLAGS_on(MUTABLE_SV(hv)); |
19692e8d | 1064 | } |
f9a63242 | 1065 | |
34dadc62 DM |
1066 | if (keysv && (SvIsCOW_shared_hash(keysv))) { |
1067 | if (HvSHAREKEYS(hv)) | |
1068 | keysv_hek = SvSHARED_HEK_FROM_PV(SvPVX_const(keysv)); | |
1069 | hash = SvSHARED_HASH(keysv); | |
7dc86639 | 1070 | } |
34dadc62 DM |
1071 | else if (!hash) |
1072 | PERL_HASH(hash, key, klen); | |
fde52b5c | 1073 | |
7a9669ca NC |
1074 | masked_flags = (k_flags & HVhek_MASK); |
1075 | ||
9faf471a | 1076 | first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; |
fde52b5c | 1077 | entry = *oentry; |
0c3bb3c2 | 1078 | |
34dadc62 DM |
1079 | if (!entry) |
1080 | goto not_found; | |
1081 | ||
1082 | if (keysv_hek) { | |
1083 | /* keysv is actually a HEK in disguise, so we can match just by | |
1084 | * comparing the HEK pointers in the HE chain. There is a slight | |
1085 | * caveat: on something like "\x80", which has both plain and utf8 | |
1086 | * representations, perl's hashes do encoding-insensitive lookups, | |
1087 | * but preserve the encoding of the stored key. Thus a particular | |
1088 | * key could map to two different HEKs in PL_strtab. We only | |
1089 | * conclude 'not found' if all the flags are the same; otherwise | |
1090 | * we fall back to a full search (this should only happen in rare | |
1091 | * cases). | |
1092 | */ | |
1093 | int keysv_flags = HEK_FLAGS(keysv_hek); | |
1094 | ||
1095 | for (; entry; oentry = &HeNEXT(entry), entry = *oentry) { | |
1096 | HEK *hek = HeKEY_hek(entry); | |
1097 | if (hek == keysv_hek) | |
1098 | goto found; | |
1099 | if (HEK_FLAGS(hek) != keysv_flags) | |
1100 | break; /* need to do full match */ | |
1101 | } | |
1102 | if (!entry) | |
1103 | goto not_found; | |
1104 | /* failed on shortcut - do full search loop */ | |
1105 | oentry = first_entry; | |
1106 | entry = *oentry; | |
1107 | } | |
1108 | ||
1109 | for (; entry; oentry = &HeNEXT(entry), entry = *oentry) { | |
fde52b5c | 1110 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
1111 | continue; | |
eb160463 | 1112 | if (HeKLEN(entry) != (I32)klen) |
fde52b5c | 1113 | continue; |
34dadc62 | 1114 | if (memNE(HeKEY(entry),key,klen)) /* is this it? */ |
fde52b5c | 1115 | continue; |
7a9669ca | 1116 | if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8) |
c3654f1a | 1117 | continue; |
8aacddc1 | 1118 | |
34dadc62 | 1119 | found: |
5d2b1485 NC |
1120 | if (hv == PL_strtab) { |
1121 | if (k_flags & HVhek_FREEKEY) | |
1122 | Safefree(key); | |
1123 | Perl_croak(aTHX_ S_strtab_error, "delete"); | |
1124 | } | |
1125 | ||
8aacddc1 | 1126 | /* if placeholder is here, it's already been deleted.... */ |
6136c704 AL |
1127 | if (HeVAL(entry) == &PL_sv_placeholder) { |
1128 | if (k_flags & HVhek_FREEKEY) | |
1129 | Safefree(key); | |
1130 | return NULL; | |
8aacddc1 | 1131 | } |
0ffdaf1a | 1132 | if (SvREADONLY(hv) && HeVAL(entry) && SvREADONLY(HeVAL(entry))) { |
d4c19fe8 | 1133 | hv_notallowed(k_flags, key, klen, |
c8cd6465 NC |
1134 | "Attempt to delete readonly key '%"SVf"' from" |
1135 | " a restricted hash"); | |
8aacddc1 | 1136 | } |
b84d0860 NC |
1137 | if (k_flags & HVhek_FREEKEY) |
1138 | Safefree(key); | |
8aacddc1 | 1139 | |
35759254 | 1140 | /* If this is a stash and the key ends with ::, then someone is |
0c3bb3c2 | 1141 | * deleting a package. |
0c3bb3c2 | 1142 | */ |
78b79c77 | 1143 | if (HeVAL(entry) && HvENAME_get(hv)) { |
0290c710 | 1144 | gv = (GV *)HeVAL(entry); |
35759254 | 1145 | if (keysv) key = SvPV(keysv, klen); |
1f656fcf FC |
1146 | if (( |
1147 | (klen > 1 && key[klen-2] == ':' && key[klen-1] == ':') | |
1148 | || | |
1149 | (klen == 1 && key[0] == ':') | |
1150 | ) | |
e0a52395 | 1151 | && (klen != 6 || hv!=PL_defstash || memNE(key,"main::",6)) |
0290c710 | 1152 | && SvTYPE(gv) == SVt_PVGV && (stash = GvHV((GV *)gv)) |
0c3bb3c2 | 1153 | && HvENAME_get(stash)) { |
0290c710 FC |
1154 | /* A previous version of this code checked that the |
1155 | * GV was still in the symbol table by fetching the | |
1156 | * GV with its name. That is not necessary (and | |
1157 | * sometimes incorrect), as HvENAME cannot be set | |
1158 | * on hv if it is not in the symtab. */ | |
f3d2f32d | 1159 | mro_changes = 2; |
0c3bb3c2 FC |
1160 | /* Hang on to it for a bit. */ |
1161 | SvREFCNT_inc_simple_void_NN( | |
0290c710 | 1162 | sv_2mortal((SV *)gv) |
35759254 FC |
1163 | ); |
1164 | } | |
f3d2f32d FC |
1165 | else if (klen == 3 && strnEQ(key, "ISA", 3)) |
1166 | mro_changes = 1; | |
35759254 FC |
1167 | } |
1168 | ||
8571a3cc FC |
1169 | sv = d_flags & G_DISCARD ? HeVAL(entry) : sv_2mortal(HeVAL(entry)); |
1170 | HeVAL(entry) = &PL_sv_placeholder; | |
5743f2a3 FC |
1171 | if (sv) { |
1172 | /* deletion of method from stash */ | |
1173 | if (isGV(sv) && isGV_with_GP(sv) && GvCVu(sv) | |
1174 | && HvENAME_get(hv)) | |
1175 | mro_method_changed_in(hv); | |
5743f2a3 | 1176 | } |
8aacddc1 NIS |
1177 | |
1178 | /* | |
1179 | * If a restricted hash, rather than really deleting the entry, put | |
1180 | * a placeholder there. This marks the key as being "approved", so | |
1181 | * we can still access via not-really-existing key without raising | |
1182 | * an error. | |
1183 | */ | |
f50383f5 | 1184 | if (SvREADONLY(hv)) |
8aacddc1 NIS |
1185 | /* We'll be saving this slot, so the number of allocated keys |
1186 | * doesn't go down, but the number placeholders goes up */ | |
ca732855 | 1187 | HvPLACEHOLDERS(hv)++; |
f50383f5 | 1188 | else { |
a26e96df | 1189 | *oentry = HeNEXT(entry); |
9faf471a NC |
1190 | if(!*first_entry && SvOOK(hv)) { |
1191 | /* removed last entry, and aux struct present. */ | |
1192 | struct xpvhv_aux *const aux = HvAUX(hv); | |
1193 | if (aux->xhv_fill_lazy) | |
1194 | --aux->xhv_fill_lazy; | |
1195 | } | |
b79f7545 | 1196 | if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */) |
8aacddc1 | 1197 | HvLAZYDEL_on(hv); |
ae199939 TH |
1198 | else { |
1199 | if (SvOOK(hv) && HvLAZYDEL(hv) && | |
1200 | entry == HeNEXT(HvAUX(hv)->xhv_eiter)) | |
1201 | HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry); | |
8aacddc1 | 1202 | hv_free_ent(hv, entry); |
ae199939 | 1203 | } |
4c7185a0 | 1204 | xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */ |
574c8022 | 1205 | if (xhv->xhv_keys == 0) |
19692e8d | 1206 | HvHASKFLAGS_off(hv); |
8aacddc1 | 1207 | } |
0c3bb3c2 | 1208 | |
3b2cd809 FC |
1209 | if (d_flags & G_DISCARD) { |
1210 | SvREFCNT_dec(sv); | |
1211 | sv = NULL; | |
1212 | } | |
1213 | ||
f3d2f32d FC |
1214 | if (mro_changes == 1) mro_isa_changed_in(hv); |
1215 | else if (mro_changes == 2) | |
afdbe55d | 1216 | mro_package_moved(NULL, stash, gv, 1); |
0c3bb3c2 | 1217 | |
79072805 LW |
1218 | return sv; |
1219 | } | |
34dadc62 DM |
1220 | |
1221 | not_found: | |
8aacddc1 | 1222 | if (SvREADONLY(hv)) { |
d4c19fe8 | 1223 | hv_notallowed(k_flags, key, klen, |
c8cd6465 NC |
1224 | "Attempt to delete disallowed key '%"SVf"' from" |
1225 | " a restricted hash"); | |
8aacddc1 NIS |
1226 | } |
1227 | ||
19692e8d | 1228 | if (k_flags & HVhek_FREEKEY) |
f9a63242 | 1229 | Safefree(key); |
a0714e2c | 1230 | return NULL; |
79072805 LW |
1231 | } |
1232 | ||
32dfa2a7 | 1233 | |
76e3520e | 1234 | STATIC void |
adf6906b | 1235 | S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize) |
79072805 | 1236 | { |
7663aa67 | 1237 | STRLEN i = 0; |
7b2c381c | 1238 | char *a = (char*) HvARRAY(hv); |
eb578fdb | 1239 | HE **aep; |
79072805 | 1240 | |
32dfa2a7 YO |
1241 | bool do_aux= ( |
1242 | /* already have an HvAUX(hv) so we have to move it */ | |
1243 | SvOOK(hv) || | |
1244 | /* no HvAUX() but array we are going to allocate is large enough | |
1245 | * there is no point in saving the space for the iterator, and | |
1246 | * speeds up later traversals. */ | |
1247 | ( ( hv != PL_strtab ) && ( newsize >= PERL_HV_ALLOC_AUX_SIZE ) ) | |
1248 | ); | |
7918f24d | 1249 | |
32dfa2a7 | 1250 | PERL_ARGS_ASSERT_HSPLIT; |
18026298 | 1251 | |
3280af22 | 1252 | PL_nomemok = TRUE; |
b79f7545 | 1253 | Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) |
32dfa2a7 YO |
1254 | + (do_aux ? sizeof(struct xpvhv_aux) : 0), char); |
1255 | PL_nomemok = FALSE; | |
422a93e5 | 1256 | if (!a) { |
422a93e5 GA |
1257 | return; |
1258 | } | |
32dfa2a7 | 1259 | |
6a5b4183 | 1260 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
3078e109 YO |
1261 | /* the idea of this is that we create a "random" value by hashing the address of |
1262 | * the array, we then use the low bit to decide if we insert at the top, or insert | |
1263 | * second from top. After each such insert we rotate the hashed value. So we can | |
1264 | * use the same hashed value over and over, and in normal build environments use | |
1265 | * very few ops to do so. ROTL32() should produce a single machine operation. */ | |
6a5b4183 YO |
1266 | if (PL_HASH_RAND_BITS_ENABLED) { |
1267 | if (PL_HASH_RAND_BITS_ENABLED == 1) | |
1268 | PL_hash_rand_bits += ptr_hash((PTRV)a); | |
1269 | PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1); | |
1270 | } | |
1271 | #endif | |
32dfa2a7 YO |
1272 | HvARRAY(hv) = (HE**) a; |
1273 | HvMAX(hv) = newsize - 1; | |
1274 | /* before we zero the newly added memory, we | |
1275 | * need to deal with the aux struct that may be there | |
1276 | * or have been allocated by us*/ | |
1277 | if (do_aux) { | |
3078e109 YO |
1278 | struct xpvhv_aux *const dest |
1279 | = (struct xpvhv_aux*) &a[newsize * sizeof(HE*)]; | |
32dfa2a7 YO |
1280 | if (SvOOK(hv)) { |
1281 | /* alread have an aux, copy the old one in place. */ | |
1282 | Move(&a[oldsize * sizeof(HE*)], dest, 1, struct xpvhv_aux); | |
1283 | /* we reset the iterator's xhv_rand as well, so they get a totally new ordering */ | |
6a5b4183 | 1284 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
32dfa2a7 | 1285 | dest->xhv_rand = (U32)PL_hash_rand_bits; |
6a5b4183 | 1286 | #endif |
32dfa2a7 YO |
1287 | /* For now, just reset the lazy fill counter. |
1288 | It would be possible to update the counter in the code below | |
1289 | instead. */ | |
1290 | dest->xhv_fill_lazy = 0; | |
1291 | } else { | |
1292 | /* no existing aux structure, but we allocated space for one | |
1293 | * so intialize it properly. This unrolls hv_auxinit() a bit, | |
1294 | * since we have to do the realloc anyway. */ | |
1295 | /* first we set the iterator's xhv_rand so it can be copied into lastrand below */ | |
1296 | #ifdef PERL_HASH_RANDOMIZE_KEYS | |
1297 | dest->xhv_rand = (U32)PL_hash_rand_bits; | |
1298 | #endif | |
1299 | /* this is the "non realloc" part of the hv_auxinit() */ | |
1300 | (void)hv_auxinit_internal(dest); | |
1301 | /* Turn on the OOK flag */ | |
1302 | SvOOK_on(hv); | |
1303 | } | |
b79f7545 | 1304 | } |
32dfa2a7 | 1305 | /* now we can safely clear the second half */ |
72311751 | 1306 | Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/ |
79072805 | 1307 | |
68303b5c NC |
1308 | if (!HvTOTALKEYS(hv)) /* skip rest if no entries */ |
1309 | return; | |
1310 | ||
32dfa2a7 | 1311 | newsize--; |
68303b5c | 1312 | aep = (HE**)a; |
7663aa67 | 1313 | do { |
c23dc12b NC |
1314 | HE **oentry = aep + i; |
1315 | HE *entry = aep[i]; | |
4b5190b5 | 1316 | |
a50a3493 | 1317 | if (!entry) /* non-existent */ |
79072805 | 1318 | continue; |
4c9d89c5 | 1319 | do { |
c23dc12b NC |
1320 | U32 j = (HeHASH(entry) & newsize); |
1321 | if (j != (U32)i) { | |
fde52b5c | 1322 | *oentry = HeNEXT(entry); |
6a5b4183 YO |
1323 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
1324 | /* if the target cell is empty or PL_HASH_RAND_BITS_ENABLED is false | |
1325 | * insert to top, otherwise rotate the bucket rand 1 bit, | |
1326 | * and use the new low bit to decide if we insert at top, | |
1327 | * or next from top. IOW, we only rotate on a collision.*/ | |
1328 | if (aep[j] && PL_HASH_RAND_BITS_ENABLED) { | |
3f49e765 | 1329 | PL_hash_rand_bits+= ROTL32(HeHASH(entry), 17); |
6a5b4183 YO |
1330 | PL_hash_rand_bits= ROTL_UV(PL_hash_rand_bits,1); |
1331 | if (PL_hash_rand_bits & 1) { | |
1332 | HeNEXT(entry)= HeNEXT(aep[j]); | |
1333 | HeNEXT(aep[j])= entry; | |
1334 | } else { | |
1335 | /* Note, this is structured in such a way as the optimizer | |
1336 | * should eliminate the duplicated code here and below without | |
1337 | * us needing to explicitly use a goto. */ | |
1338 | HeNEXT(entry) = aep[j]; | |
1339 | aep[j] = entry; | |
1340 | } | |
1341 | } else | |
1342 | #endif | |
1343 | { | |
1344 | /* see comment above about duplicated code */ | |
3078e109 YO |
1345 | HeNEXT(entry) = aep[j]; |
1346 | aep[j] = entry; | |
3078e109 | 1347 | } |
79072805 | 1348 | } |
4b5190b5 | 1349 | else { |
fde52b5c | 1350 | oentry = &HeNEXT(entry); |
4b5190b5 | 1351 | } |
4c9d89c5 NC |
1352 | entry = *oentry; |
1353 | } while (entry); | |
7663aa67 | 1354 | } while (i++ < oldsize); |
79072805 LW |
1355 | } |
1356 | ||
72940dca | 1357 | void |
864dbfa3 | 1358 | Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) |
72940dca | 1359 | { |
eb578fdb | 1360 | XPVHV* xhv = (XPVHV*)SvANY(hv); |
a3b680e6 | 1361 | const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */ |
eb578fdb | 1362 | I32 newsize; |
eb578fdb | 1363 | char *a; |
72940dca | 1364 | |
7918f24d NC |
1365 | PERL_ARGS_ASSERT_HV_KSPLIT; |
1366 | ||
72940dca | 1367 | newsize = (I32) newmax; /* possible truncation here */ |
1368 | if (newsize != newmax || newmax <= oldsize) | |
1369 | return; | |
1370 | while ((newsize & (1 + ~newsize)) != newsize) { | |
1371 | newsize &= ~(newsize & (1 + ~newsize)); /* get proper power of 2 */ | |
1372 | } | |
1373 | if (newsize < newmax) | |
1374 | newsize *= 2; | |
1375 | if (newsize < newmax) | |
1376 | return; /* overflow detection */ | |
1377 | ||
7b2c381c | 1378 | a = (char *) HvARRAY(hv); |
e8c10cf3 NC |
1379 | if (a) { |
1380 | hsplit(hv, oldsize, newsize); | |
1381 | } else { | |
0df05616 NC |
1382 | Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); |
1383 | xhv->xhv_max = --newsize; | |
1384 | HvARRAY(hv) = (HE **) a; | |
72940dca | 1385 | } |
1386 | } | |
1387 | ||
f6bb1c88 YO |
1388 | /* IMO this should also handle cases where hv_max is smaller than hv_keys |
1389 | * as tied hashes could play silly buggers and mess us around. We will | |
1390 | * do the right thing during hv_store() afterwards, but still - Yves */ | |
1391 | #define HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys) STMT_START {\ | |
1392 | /* Can we use fewer buckets? (hv_max is always 2^n-1) */ \ | |
1393 | if (hv_max < PERL_HASH_DEFAULT_HvMAX) { \ | |
1394 | hv_max = PERL_HASH_DEFAULT_HvMAX; \ | |
1395 | } else { \ | |
1396 | while (hv_max > PERL_HASH_DEFAULT_HvMAX && hv_max + 1 >= hv_keys * 2) \ | |
1397 | hv_max = hv_max / 2; \ | |
1398 | } \ | |
1399 | HvMAX(hv) = hv_max; \ | |
1400 | } STMT_END | |
1401 | ||
1402 | ||
b3ac6de7 | 1403 | HV * |
864dbfa3 | 1404 | Perl_newHVhv(pTHX_ HV *ohv) |
b3ac6de7 | 1405 | { |
749123ff | 1406 | dVAR; |
9d4ba2ae | 1407 | HV * const hv = newHV(); |
f4431c56 | 1408 | STRLEN hv_max; |
4beac62f | 1409 | |
3f4d1d78 | 1410 | if (!ohv || (!HvTOTALKEYS(ohv) && !SvMAGICAL((const SV *)ohv))) |
4beac62f | 1411 | return hv; |
4beac62f | 1412 | hv_max = HvMAX(ohv); |
b3ac6de7 | 1413 | |
ad64d0ec | 1414 | if (!SvMAGICAL((const SV *)ohv)) { |
b56ba0bf | 1415 | /* It's an ordinary hash, so copy it fast. AMS 20010804 */ |
eb160463 | 1416 | STRLEN i; |
a3b680e6 | 1417 | const bool shared = !!HvSHAREKEYS(ohv); |
aec46f14 | 1418 | HE **ents, ** const oents = (HE **)HvARRAY(ohv); |
ff875642 | 1419 | char *a; |
a02a5408 | 1420 | Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char); |
ff875642 | 1421 | ents = (HE**)a; |
b56ba0bf AMS |
1422 | |
1423 | /* In each bucket... */ | |
1424 | for (i = 0; i <= hv_max; i++) { | |
6136c704 | 1425 | HE *prev = NULL; |
aec46f14 | 1426 | HE *oent = oents[i]; |
b56ba0bf AMS |
1427 | |
1428 | if (!oent) { | |
1429 | ents[i] = NULL; | |
1430 | continue; | |
1431 | } | |
1432 | ||
1433 | /* Copy the linked list of entries. */ | |
aec46f14 | 1434 | for (; oent; oent = HeNEXT(oent)) { |
a3b680e6 AL |
1435 | const U32 hash = HeHASH(oent); |
1436 | const char * const key = HeKEY(oent); | |
1437 | const STRLEN len = HeKLEN(oent); | |
1438 | const int flags = HeKFLAGS(oent); | |
6136c704 | 1439 | HE * const ent = new_HE(); |
c3acb9e0 | 1440 | SV *const val = HeVAL(oent); |
b56ba0bf | 1441 | |
c3acb9e0 | 1442 | HeVAL(ent) = SvIMMORTAL(val) ? val : newSVsv(val); |
19692e8d | 1443 | HeKEY_hek(ent) |
6e838c70 | 1444 | = shared ? share_hek_flags(key, len, hash, flags) |
19692e8d | 1445 | : save_hek_flags(key, len, hash, flags); |
b56ba0bf AMS |
1446 | if (prev) |
1447 | HeNEXT(prev) = ent; | |
1448 | else | |
1449 | ents[i] = ent; | |
1450 | prev = ent; | |
1451 | HeNEXT(ent) = NULL; | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | HvMAX(hv) = hv_max; | |
8aacddc1 | 1456 | HvTOTALKEYS(hv) = HvTOTALKEYS(ohv); |
b56ba0bf | 1457 | HvARRAY(hv) = ents; |
aec46f14 | 1458 | } /* not magical */ |
b56ba0bf AMS |
1459 | else { |
1460 | /* Iterate over ohv, copying keys and values one at a time. */ | |
b3ac6de7 | 1461 | HE *entry; |
bfcb3514 NC |
1462 | const I32 riter = HvRITER_get(ohv); |
1463 | HE * const eiter = HvEITER_get(ohv); | |
f6bb1c88 | 1464 | STRLEN hv_keys = HvTOTALKEYS(ohv); |
b56ba0bf | 1465 | |
f6bb1c88 | 1466 | HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys); |
b56ba0bf | 1467 | |
4a76a316 | 1468 | hv_iterinit(ohv); |
e16e2ff8 | 1469 | while ((entry = hv_iternext_flags(ohv, 0))) { |
3f4d1d78 FC |
1470 | SV *val = hv_iterval(ohv,entry); |
1471 | SV * const keysv = HeSVKEY(entry); | |
1472 | val = SvIMMORTAL(val) ? val : newSVsv(val); | |
1473 | if (keysv) | |
1474 | (void)hv_store_ent(hv, keysv, val, 0); | |
1475 | else | |
1476 | (void)hv_store_flags(hv, HeKEY(entry), HeKLEN(entry), val, | |
c3acb9e0 | 1477 | HeHASH(entry), HeKFLAGS(entry)); |
b3ac6de7 | 1478 | } |
bfcb3514 NC |
1479 | HvRITER_set(ohv, riter); |
1480 | HvEITER_set(ohv, eiter); | |
b3ac6de7 | 1481 | } |
1c846c1f | 1482 | |
b3ac6de7 IZ |
1483 | return hv; |
1484 | } | |
1485 | ||
defdfed5 Z |
1486 | /* |
1487 | =for apidoc Am|HV *|hv_copy_hints_hv|HV *ohv | |
1488 | ||
1489 | A specialised version of L</newHVhv> for copying C<%^H>. I<ohv> must be | |
1490 | a pointer to a hash (which may have C<%^H> magic, but should be generally | |
1491 | non-magical), or C<NULL> (interpreted as an empty hash). The content | |
1492 | of I<ohv> is copied to a new hash, which has the C<%^H>-specific magic | |
1493 | added to it. A pointer to the new hash is returned. | |
1494 | ||
1495 | =cut | |
1496 | */ | |
1497 | ||
5b9c0671 NC |
1498 | HV * |
1499 | Perl_hv_copy_hints_hv(pTHX_ HV *const ohv) | |
1500 | { | |
1501 | HV * const hv = newHV(); | |
5b9c0671 | 1502 | |
cb1f05e8 | 1503 | if (ohv) { |
5b9c0671 | 1504 | STRLEN hv_max = HvMAX(ohv); |
f6bb1c88 | 1505 | STRLEN hv_keys = HvTOTALKEYS(ohv); |
5b9c0671 NC |
1506 | HE *entry; |
1507 | const I32 riter = HvRITER_get(ohv); | |
1508 | HE * const eiter = HvEITER_get(ohv); | |
1509 | ||
0db511c0 FC |
1510 | ENTER; |
1511 | SAVEFREESV(hv); | |
1512 | ||
f6bb1c88 | 1513 | HV_SET_MAX_ADJUSTED_FOR_KEYS(hv,hv_max,hv_keys); |
5b9c0671 NC |
1514 | |
1515 | hv_iterinit(ohv); | |
1516 | while ((entry = hv_iternext_flags(ohv, 0))) { | |
cb1f05e8 | 1517 | SV *const sv = newSVsv(hv_iterval(ohv,entry)); |
7ef9d42c FC |
1518 | SV *heksv = HeSVKEY(entry); |
1519 | if (!heksv && sv) heksv = newSVhek(HeKEY_hek(entry)); | |
95cf2368 | 1520 | if (sv) sv_magic(sv, NULL, PERL_MAGIC_hintselem, |
e3b1b6b1 | 1521 | (char *)heksv, HEf_SVKEY); |
7ef9d42c FC |
1522 | if (heksv == HeSVKEY(entry)) |
1523 | (void)hv_store_ent(hv, heksv, sv, 0); | |
1524 | else { | |
1525 | (void)hv_common(hv, heksv, HeKEY(entry), HeKLEN(entry), | |
1526 | HeKFLAGS(entry), HV_FETCH_ISSTORE|HV_FETCH_JUST_SV, sv, HeHASH(entry)); | |
a03199ea | 1527 | SvREFCNT_dec_NN(heksv); |
7ef9d42c | 1528 | } |
5b9c0671 NC |
1529 | } |
1530 | HvRITER_set(ohv, riter); | |
1531 | HvEITER_set(ohv, eiter); | |
0db511c0 FC |
1532 | |
1533 | SvREFCNT_inc_simple_void_NN(hv); | |
1534 | LEAVE; | |
5b9c0671 NC |
1535 | } |
1536 | hv_magic(hv, NULL, PERL_MAGIC_hints); | |
1537 | return hv; | |
1538 | } | |
f6bb1c88 | 1539 | #undef HV_SET_MAX_ADJUSTED_FOR_KEYS |
5b9c0671 | 1540 | |
e0171a1a DM |
1541 | /* like hv_free_ent, but returns the SV rather than freeing it */ |
1542 | STATIC SV* | |
5aaab254 | 1543 | S_hv_free_ent_ret(pTHX_ HV *hv, HE *entry) |
79072805 | 1544 | { |
16bdeea2 GS |
1545 | SV *val; |
1546 | ||
e0171a1a | 1547 | PERL_ARGS_ASSERT_HV_FREE_ENT_RET; |
7918f24d | 1548 | |
16bdeea2 | 1549 | val = HeVAL(entry); |
68dc0745 | 1550 | if (HeKLEN(entry) == HEf_SVKEY) { |
1551 | SvREFCNT_dec(HeKEY_sv(entry)); | |
8aacddc1 | 1552 | Safefree(HeKEY_hek(entry)); |
44a8e56a | 1553 | } |
1554 | else if (HvSHAREKEYS(hv)) | |
68dc0745 | 1555 | unshare_hek(HeKEY_hek(entry)); |
fde52b5c | 1556 | else |
68dc0745 | 1557 | Safefree(HeKEY_hek(entry)); |
d33b2eba | 1558 | del_HE(entry); |
e0171a1a DM |
1559 | return val; |
1560 | } | |
1561 | ||
1562 | ||
1563 | void | |
5aaab254 | 1564 | Perl_hv_free_ent(pTHX_ HV *hv, HE *entry) |
e0171a1a | 1565 | { |
e0171a1a DM |
1566 | SV *val; |
1567 | ||
1568 | PERL_ARGS_ASSERT_HV_FREE_ENT; | |
1569 | ||
1570 | if (!entry) | |
1571 | return; | |
1572 | val = hv_free_ent_ret(hv, entry); | |
272e8453 | 1573 | SvREFCNT_dec(val); |
79072805 LW |
1574 | } |
1575 | ||
f1c32fec | 1576 | |
79072805 | 1577 | void |
5aaab254 | 1578 | Perl_hv_delayfree_ent(pTHX_ HV *hv, HE *entry) |
79072805 | 1579 | { |
7918f24d NC |
1580 | PERL_ARGS_ASSERT_HV_DELAYFREE_ENT; |
1581 | ||
68dc0745 | 1582 | if (!entry) |
79072805 | 1583 | return; |
bc4947fc NC |
1584 | /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */ |
1585 | sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */ | |
68dc0745 | 1586 | if (HeKLEN(entry) == HEf_SVKEY) { |
bc4947fc | 1587 | sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry))); |
44a8e56a | 1588 | } |
bc4947fc | 1589 | hv_free_ent(hv, entry); |
79072805 LW |
1590 | } |
1591 | ||
954c1994 GS |
1592 | /* |
1593 | =for apidoc hv_clear | |
1594 | ||
c2217cd3 | 1595 | Frees the all the elements of a hash, leaving it empty. |
8b9a1153 FC |
1596 | The XS equivalent of C<%hash = ()>. See also L</hv_undef>. |
1597 | ||
1598 | If any destructors are triggered as a result, the hv itself may | |
1599 | be freed. | |
954c1994 GS |
1600 | |
1601 | =cut | |
1602 | */ | |
1603 | ||
79072805 | 1604 | void |
864dbfa3 | 1605 | Perl_hv_clear(pTHX_ HV *hv) |
79072805 | 1606 | { |
27da23d5 | 1607 | dVAR; |
eb578fdb | 1608 | XPVHV* xhv; |
79072805 LW |
1609 | if (!hv) |
1610 | return; | |
49293501 | 1611 | |
ecae49c0 NC |
1612 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
1613 | ||
34c3c4e3 DM |
1614 | xhv = (XPVHV*)SvANY(hv); |
1615 | ||
8505eec0 FC |
1616 | ENTER; |
1617 | SAVEFREESV(SvREFCNT_inc_simple_NN(hv)); | |
7b2c381c | 1618 | if (SvREADONLY(hv) && HvARRAY(hv) != NULL) { |
34c3c4e3 | 1619 | /* restricted hash: convert all keys to placeholders */ |
b464bac0 AL |
1620 | STRLEN i; |
1621 | for (i = 0; i <= xhv->xhv_max; i++) { | |
7b2c381c | 1622 | HE *entry = (HvARRAY(hv))[i]; |
3a676441 JH |
1623 | for (; entry; entry = HeNEXT(entry)) { |
1624 | /* not already placeholder */ | |
7996736c | 1625 | if (HeVAL(entry) != &PL_sv_placeholder) { |
a03199ea | 1626 | if (HeVAL(entry)) { |
0ffdaf1a | 1627 | if (SvREADONLY(HeVAL(entry))) { |
a03199ea DD |
1628 | SV* const keysv = hv_iterkeysv(entry); |
1629 | Perl_croak_nocontext( | |
1630 | "Attempt to delete readonly key '%"SVf"' from a restricted hash", | |
1631 | (void*)keysv); | |
1632 | } | |
1633 | SvREFCNT_dec_NN(HeVAL(entry)); | |
3a676441 | 1634 | } |
7996736c | 1635 | HeVAL(entry) = &PL_sv_placeholder; |
ca732855 | 1636 | HvPLACEHOLDERS(hv)++; |
3a676441 | 1637 | } |
34c3c4e3 DM |
1638 | } |
1639 | } | |
49293501 | 1640 | } |
afbbf215 DM |
1641 | else { |
1642 | hfreeentries(hv); | |
1643 | HvPLACEHOLDERS_set(hv, 0); | |
49293501 | 1644 | |
afbbf215 DM |
1645 | if (SvRMAGICAL(hv)) |
1646 | mg_clear(MUTABLE_SV(hv)); | |
574c8022 | 1647 | |
afbbf215 | 1648 | HvHASKFLAGS_off(hv); |
afbbf215 | 1649 | } |
b79f7545 | 1650 | if (SvOOK(hv)) { |
00169e2c | 1651 | if(HvENAME_get(hv)) |
dd69841b | 1652 | mro_isa_changed_in(hv); |
bfcb3514 NC |
1653 | HvEITER_set(hv, NULL); |
1654 | } | |
8505eec0 | 1655 | LEAVE; |
79072805 LW |
1656 | } |
1657 | ||
3540d4ce AB |
1658 | /* |
1659 | =for apidoc hv_clear_placeholders | |
1660 | ||
1661 | Clears any placeholders from a hash. If a restricted hash has any of its keys | |
1662 | marked as readonly and the key is subsequently deleted, the key is not actually | |
1663 | deleted but is marked by assigning it a value of &PL_sv_placeholder. This tags | |
1664 | it so it will be ignored by future operations such as iterating over the hash, | |
4cdaeff7 | 1665 | but will still allow the hash to have a value reassigned to the key at some |
3540d4ce AB |
1666 | future point. This function clears any such placeholder keys from the hash. |
1667 | See Hash::Util::lock_keys() for an example of its use. | |
1668 | ||
1669 | =cut | |
1670 | */ | |
1671 | ||
1672 | void | |
1673 | Perl_hv_clear_placeholders(pTHX_ HV *hv) | |
1674 | { | |
b3ca2e83 NC |
1675 | const U32 items = (U32)HvPLACEHOLDERS_get(hv); |
1676 | ||
7918f24d NC |
1677 | PERL_ARGS_ASSERT_HV_CLEAR_PLACEHOLDERS; |
1678 | ||
b3ca2e83 NC |
1679 | if (items) |
1680 | clear_placeholders(hv, items); | |
1681 | } | |
1682 | ||
1683 | static void | |
1684 | S_clear_placeholders(pTHX_ HV *hv, U32 items) | |
1685 | { | |
1686 | dVAR; | |
b464bac0 | 1687 | I32 i; |
d3677389 | 1688 | |
7918f24d NC |
1689 | PERL_ARGS_ASSERT_CLEAR_PLACEHOLDERS; |
1690 | ||
d3677389 NC |
1691 | if (items == 0) |
1692 | return; | |
1693 | ||
b464bac0 | 1694 | i = HvMAX(hv); |
d3677389 NC |
1695 | do { |
1696 | /* Loop down the linked list heads */ | |
d3677389 | 1697 | HE **oentry = &(HvARRAY(hv))[i]; |
cf6db12b | 1698 | HE *entry; |
d3677389 | 1699 | |
cf6db12b | 1700 | while ((entry = *oentry)) { |
d3677389 NC |
1701 | if (HeVAL(entry) == &PL_sv_placeholder) { |
1702 | *oentry = HeNEXT(entry); | |
2e58978b | 1703 | if (entry == HvEITER_get(hv)) |
d3677389 | 1704 | HvLAZYDEL_on(hv); |
ae199939 TH |
1705 | else { |
1706 | if (SvOOK(hv) && HvLAZYDEL(hv) && | |
1707 | entry == HeNEXT(HvAUX(hv)->xhv_eiter)) | |
1708 | HeNEXT(HvAUX(hv)->xhv_eiter) = HeNEXT(entry); | |
d3677389 | 1709 | hv_free_ent(hv, entry); |
ae199939 | 1710 | } |
d3677389 NC |
1711 | |
1712 | if (--items == 0) { | |
1713 | /* Finished. */ | |
5d27ee4a DD |
1714 | I32 placeholders = HvPLACEHOLDERS_get(hv); |
1715 | HvTOTALKEYS(hv) -= (IV)placeholders; | |
1716 | /* HvUSEDKEYS expanded */ | |
1717 | if ((HvTOTALKEYS(hv) - placeholders) == 0) | |
d3677389 | 1718 | HvHASKFLAGS_off(hv); |
5d88ecd7 | 1719 | HvPLACEHOLDERS_set(hv, 0); |
d3677389 NC |
1720 | return; |
1721 | } | |
213ce8b3 NC |
1722 | } else { |
1723 | oentry = &HeNEXT(entry); | |
d3677389 NC |
1724 | } |
1725 | } | |
1726 | } while (--i >= 0); | |
1727 | /* You can't get here, hence assertion should always fail. */ | |
1728 | assert (items == 0); | |
31cdb61f | 1729 | NOT_REACHED; |
3540d4ce AB |
1730 | } |
1731 | ||
76e3520e | 1732 | STATIC void |
cea2e8a9 | 1733 | S_hfreeentries(pTHX_ HV *hv) |
79072805 | 1734 | { |
e0171a1a | 1735 | STRLEN index = 0; |
7d6175ef | 1736 | XPVHV * const xhv = (XPVHV*)SvANY(hv); |
6d1c68e6 | 1737 | SV *sv; |
3abe233e | 1738 | |
7918f24d NC |
1739 | PERL_ARGS_ASSERT_HFREEENTRIES; |
1740 | ||
6d1c68e6 FC |
1741 | while ((sv = Perl_hfree_next_entry(aTHX_ hv, &index))||xhv->xhv_keys) { |
1742 | SvREFCNT_dec(sv); | |
e0171a1a DM |
1743 | } |
1744 | } | |
23976bdd | 1745 | |
b79f7545 | 1746 | |
e0171a1a DM |
1747 | /* hfree_next_entry() |
1748 | * For use only by S_hfreeentries() and sv_clear(). | |
1749 | * Delete the next available HE from hv and return the associated SV. | |
7d6175ef FC |
1750 | * Returns null on empty hash. Nevertheless null is not a reliable |
1751 | * indicator that the hash is empty, as the deleted entry may have a | |
1752 | * null value. | |
e0171a1a DM |
1753 | * indexp is a pointer to the current index into HvARRAY. The index should |
1754 | * initially be set to 0. hfree_next_entry() may update it. */ | |
1755 | ||
1756 | SV* | |
1757 | Perl_hfree_next_entry(pTHX_ HV *hv, STRLEN *indexp) | |
1758 | { | |
1759 | struct xpvhv_aux *iter; | |
1760 | HE *entry; | |
1761 | HE ** array; | |
1762 | #ifdef DEBUGGING | |
1763 | STRLEN orig_index = *indexp; | |
1764 | #endif | |
1765 | ||
1766 | PERL_ARGS_ASSERT_HFREE_NEXT_ENTRY; | |
1767 | ||
9faf471a NC |
1768 | if (SvOOK(hv) && ((iter = HvAUX(hv)))) { |
1769 | if ((entry = iter->xhv_eiter)) { | |
1770 | /* the iterator may get resurrected after each | |
1771 | * destructor call, so check each time */ | |
1772 | if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ | |
1773 | HvLAZYDEL_off(hv); | |
1774 | hv_free_ent(hv, entry); | |
1775 | /* warning: at this point HvARRAY may have been | |
1776 | * re-allocated, HvMAX changed etc */ | |
1777 | } | |
339441ef | 1778 | iter = HvAUX(hv); /* may have been realloced */ |
9faf471a NC |
1779 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
1780 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ | |
6a5b4183 | 1781 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
9faf471a | 1782 | iter->xhv_last_rand = iter->xhv_rand; |
6a5b4183 | 1783 | #endif |
9faf471a NC |
1784 | } |
1785 | /* Reset any cached HvFILL() to "unknown". It's unlikely that anyone | |
1786 | will actually call HvFILL() on a hash under destruction, so it | |
1787 | seems pointless attempting to track the number of keys remaining. | |
1788 | But if they do, we want to reset it again. */ | |
1789 | if (iter->xhv_fill_lazy) | |
1790 | iter->xhv_fill_lazy = 0; | |
e0171a1a DM |
1791 | } |
1792 | ||
00a1a643 DM |
1793 | if (!((XPVHV*)SvANY(hv))->xhv_keys) |
1794 | return NULL; | |
1795 | ||
e0171a1a DM |
1796 | array = HvARRAY(hv); |
1797 | assert(array); | |
1798 | while ( ! ((entry = array[*indexp])) ) { | |
1799 | if ((*indexp)++ >= HvMAX(hv)) | |
1800 | *indexp = 0; | |
1801 | assert(*indexp != orig_index); | |
1802 | } | |
1803 | array[*indexp] = HeNEXT(entry); | |
1804 | ((XPVHV*) SvANY(hv))->xhv_keys--; | |
1805 | ||
1806 | if ( PL_phase != PERL_PHASE_DESTRUCT && HvENAME(hv) | |
1807 | && HeVAL(entry) && isGV(HeVAL(entry)) | |
1808 | && GvHV(HeVAL(entry)) && HvENAME(GvHV(HeVAL(entry))) | |
1809 | ) { | |
1810 | STRLEN klen; | |
1811 | const char * const key = HePV(entry,klen); | |
1812 | if ((klen > 1 && key[klen-1]==':' && key[klen-2]==':') | |
1813 | || (klen == 1 && key[0] == ':')) { | |
1814 | mro_package_moved( | |
1815 | NULL, GvHV(HeVAL(entry)), | |
1816 | (GV *)HeVAL(entry), 0 | |
1817 | ); | |
1818 | } | |
1819 | } | |
1820 | return hv_free_ent_ret(hv, entry); | |
79072805 LW |
1821 | } |
1822 | ||
e0171a1a | 1823 | |
954c1994 GS |
1824 | /* |
1825 | =for apidoc hv_undef | |
1826 | ||
8b9a1153 | 1827 | Undefines the hash. The XS equivalent of C<undef(%hash)>. |
c2217cd3 DM |
1828 | |
1829 | As well as freeing all the elements of the hash (like hv_clear()), this | |
1830 | also frees any auxiliary data and storage associated with the hash. | |
8b9a1153 FC |
1831 | |
1832 | If any destructors are triggered as a result, the hv itself may | |
1833 | be freed. | |
1834 | ||
c2217cd3 | 1835 | See also L</hv_clear>. |
954c1994 GS |
1836 | |
1837 | =cut | |
1838 | */ | |
1839 | ||
79072805 | 1840 | void |
8581adba | 1841 | Perl_hv_undef_flags(pTHX_ HV *hv, U32 flags) |
79072805 | 1842 | { |
eb578fdb | 1843 | XPVHV* xhv; |
8a50cd03 | 1844 | bool save; |
86f55936 | 1845 | |
79072805 LW |
1846 | if (!hv) |
1847 | return; | |
8a50cd03 | 1848 | save = !!SvREFCNT(hv); |
ecae49c0 | 1849 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
cbec9347 | 1850 | xhv = (XPVHV*)SvANY(hv); |
dd69841b | 1851 | |
745edda6 FC |
1852 | /* The name must be deleted before the call to hfreeeeentries so that |
1853 | CVs are anonymised properly. But the effective name must be pre- | |
1854 | served until after that call (and only deleted afterwards if the | |
1855 | call originated from sv_clear). For stashes with one name that is | |
1856 | both the canonical name and the effective name, hv_name_set has to | |
1857 | allocate an array for storing the effective name. We can skip that | |
1858 | during global destruction, as it does not matter where the CVs point | |
1859 | if they will be freed anyway. */ | |
104d7b69 DM |
1860 | /* note that the code following prior to hfreeentries is duplicated |
1861 | * in sv_clear(), and changes here should be done there too */ | |
0ca9877d | 1862 | if (PL_phase != PERL_PHASE_DESTRUCT && HvNAME(hv)) { |
103f5a36 NC |
1863 | if (PL_stashcache) { |
1864 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for '%" | |
10bafe90 | 1865 | HEKf"'\n", HEKfARG(HvNAME_HEK(hv)))); |
0ca9877d | 1866 | (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD); |
103f5a36 | 1867 | } |
bd61b366 | 1868 | hv_name_set(hv, NULL, 0, 0); |
85e6fe83 | 1869 | } |
8505eec0 FC |
1870 | if (save) { |
1871 | ENTER; | |
1872 | SAVEFREESV(SvREFCNT_inc_simple_NN(hv)); | |
1873 | } | |
2d0d1ecc | 1874 | hfreeentries(hv); |
47f1cf77 | 1875 | if (SvOOK(hv)) { |
47f1cf77 | 1876 | struct mro_meta *meta; |
0ca9877d | 1877 | const char *name; |
745edda6 | 1878 | |
0ca9877d | 1879 | if (HvENAME_get(hv)) { |
5f243b5f | 1880 | if (PL_phase != PERL_PHASE_DESTRUCT) |
745edda6 | 1881 | mro_isa_changed_in(hv); |
103f5a36 NC |
1882 | if (PL_stashcache) { |
1883 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for effective name '%" | |
10bafe90 | 1884 | HEKf"'\n", HEKfARG(HvENAME_HEK(hv)))); |
0ca9877d | 1885 | (void)hv_deletehek(PL_stashcache, HvENAME_HEK(hv), G_DISCARD); |
103f5a36 | 1886 | } |
745edda6 FC |
1887 | } |
1888 | ||
1889 | /* If this call originated from sv_clear, then we must check for | |
1890 | * effective names that need freeing, as well as the usual name. */ | |
1891 | name = HvNAME(hv); | |
339441ef | 1892 | if (flags & HV_NAME_SETALL ? !!HvAUX(hv)->xhv_name_u.xhvnameu_name : !!name) { |
103f5a36 NC |
1893 | if (name && PL_stashcache) { |
1894 | DEBUG_o(Perl_deb(aTHX_ "hv_undef_flags clearing PL_stashcache for name '%" | |
10bafe90 | 1895 | HEKf"'\n", HEKfARG(HvNAME_HEK(hv)))); |
0ca9877d | 1896 | (void)hv_deletehek(PL_stashcache, HvNAME_HEK(hv), G_DISCARD); |
103f5a36 | 1897 | } |
745edda6 | 1898 | hv_name_set(hv, NULL, 0, flags); |
47f1cf77 | 1899 | } |
339441ef | 1900 | if((meta = HvAUX(hv)->xhv_mro_meta)) { |
47f1cf77 | 1901 | if (meta->mro_linear_all) { |
d4f87935 FC |
1902 | SvREFCNT_dec_NN(meta->mro_linear_all); |
1903 | /* mro_linear_current is just acting as a shortcut pointer, | |
1904 | hence the else. */ | |
1905 | } | |
1906 | else | |
47f1cf77 FC |
1907 | /* Only the current MRO is stored, so this owns the data. |
1908 | */ | |
1909 | SvREFCNT_dec(meta->mro_linear_current); | |
9bfbb681 | 1910 | SvREFCNT_dec(meta->mro_nextmethod); |
47f1cf77 | 1911 | SvREFCNT_dec(meta->isa); |
1a33a059 | 1912 | SvREFCNT_dec(meta->super); |
47f1cf77 | 1913 | Safefree(meta); |
339441ef | 1914 | HvAUX(hv)->xhv_mro_meta = NULL; |
47f1cf77 | 1915 | } |
339441ef | 1916 | if (!HvAUX(hv)->xhv_name_u.xhvnameu_name && ! HvAUX(hv)->xhv_backreferences) |
745edda6 | 1917 | SvFLAGS(hv) &= ~SVf_OOK; |
745edda6 FC |
1918 | } |
1919 | if (!SvOOK(hv)) { | |
1920 | Safefree(HvARRAY(hv)); | |
f6bb1c88 | 1921 | xhv->xhv_max = PERL_HASH_DEFAULT_HvMAX; /* HvMAX(hv) = 7 (it's a normal hash) */ |
745edda6 | 1922 | HvARRAY(hv) = 0; |
2d0d1ecc | 1923 | } |
5bec93be DM |
1924 | /* if we're freeing the HV, the SvMAGIC field has been reused for |
1925 | * other purposes, and so there can't be any placeholder magic */ | |
1926 | if (SvREFCNT(hv)) | |
1927 | HvPLACEHOLDERS_set(hv, 0); | |
a0d0e21e LW |
1928 | |
1929 | if (SvRMAGICAL(hv)) | |
ad64d0ec | 1930 | mg_clear(MUTABLE_SV(hv)); |
8505eec0 | 1931 | if (save) LEAVE; |
79072805 LW |
1932 | } |
1933 | ||
4d0fbddd NC |
1934 | /* |
1935 | =for apidoc hv_fill | |
1936 | ||
72d33970 FC |
1937 | Returns the number of hash buckets that |
1938 | happen to be in use. This function is | |
4d0fbddd NC |
1939 | wrapped by the macro C<HvFILL>. |
1940 | ||
9faf471a NC |
1941 | Previously this value was always stored in the HV structure, which created an |
1942 | overhead on every hash (and pretty much every object) for something that was | |
72d33970 FC |
1943 | rarely used. Now we calculate it on demand the first |
1944 | time that it is needed, and cache it if that calculation | |
1945 | is going to be costly to repeat. The cached | |
9faf471a NC |
1946 | value is updated by insertions and deletions, but (currently) discarded if |
1947 | the hash is split. | |
4d0fbddd NC |
1948 | |
1949 | =cut | |
1950 | */ | |
1951 | ||
1952 | STRLEN | |
9faf471a | 1953 | Perl_hv_fill(pTHX_ HV *const hv) |
4d0fbddd NC |
1954 | { |
1955 | STRLEN count = 0; | |
1956 | HE **ents = HvARRAY(hv); | |
9faf471a | 1957 | struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : NULL; |
4d0fbddd NC |
1958 | |
1959 | PERL_ARGS_ASSERT_HV_FILL; | |
1960 | ||
553215cc NC |
1961 | /* No keys implies no buckets used. |
1962 | One key can only possibly mean one bucket used. */ | |
1963 | if (HvTOTALKEYS(hv) < 2) | |
1964 | return HvTOTALKEYS(hv); | |
1965 | ||
9faf471a NC |
1966 | #ifndef DEBUGGING |
1967 | if (aux && aux->xhv_fill_lazy) | |
1968 | return aux->xhv_fill_lazy; | |
1969 | #endif | |
1970 | ||
4d0fbddd | 1971 | if (ents) { |
fcd24582 NC |
1972 | HE *const *const last = ents + HvMAX(hv); |
1973 | count = last + 1 - ents; | |
4d0fbddd NC |
1974 | |
1975 | do { | |
fcd24582 NC |
1976 | if (!*ents) |
1977 | --count; | |
1978 | } while (++ents <= last); | |
4d0fbddd | 1979 | } |
9faf471a NC |
1980 | if (aux) { |
1981 | #ifdef DEBUGGING | |
1982 | if (aux->xhv_fill_lazy) | |
1983 | assert(aux->xhv_fill_lazy == count); | |
1984 | #endif | |
1985 | aux->xhv_fill_lazy = count; | |
1986 | } else if (HvMAX(hv) >= HV_FILL_THRESHOLD) { | |
1987 | aux = hv_auxinit(hv); | |
1988 | aux->xhv_fill_lazy = count; | |
1989 | } | |
4d0fbddd NC |
1990 | return count; |
1991 | } | |
1992 | ||
0e0ab621 YO |
1993 | /* hash a pointer to a U32 - Used in the hash traversal randomization |
1994 | * and bucket order randomization code | |
1995 | * | |
1996 | * this code was derived from Sereal, which was derived from autobox. | |
1997 | */ | |
1998 | ||
1999 | PERL_STATIC_INLINE U32 S_ptr_hash(PTRV u) { | |
2000 | #if PTRSIZE == 8 | |
2001 | /* | |
2002 | * This is one of Thomas Wang's hash functions for 64-bit integers from: | |
2003 | * http://www.concentric.net/~Ttwang/tech/inthash.htm | |
2004 | */ | |
2005 | u = (~u) + (u << 18); | |
2006 | u = u ^ (u >> 31); | |
2007 | u = u * 21; | |
2008 | u = u ^ (u >> 11); | |
2009 | u = u + (u << 6); | |
2010 | u = u ^ (u >> 22); | |
2011 | #else | |
2012 | /* | |
2013 | * This is one of Bob Jenkins' hash functions for 32-bit integers | |
2014 | * from: http://burtleburtle.net/bob/hash/integer.html | |
2015 | */ | |
2016 | u = (u + 0x7ed55d16) + (u << 12); | |
2017 | u = (u ^ 0xc761c23c) ^ (u >> 19); | |
2018 | u = (u + 0x165667b1) + (u << 5); | |
2019 | u = (u + 0xd3a2646c) ^ (u << 9); | |
2020 | u = (u + 0xfd7046c5) + (u << 3); | |
2021 | u = (u ^ 0xb55a4f09) ^ (u >> 16); | |
2022 | #endif | |
2023 | return (U32)u; | |
2024 | } | |
2025 | ||
bea177f3 YO |
2026 | static struct xpvhv_aux* |
2027 | S_hv_auxinit_internal(struct xpvhv_aux *iter) { | |
2028 | PERL_ARGS_ASSERT_HV_AUXINIT_INTERNAL; | |
2029 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ | |
2030 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ | |
2031 | #ifdef PERL_HASH_RANDOMIZE_KEYS | |
2032 | iter->xhv_last_rand = iter->xhv_rand; | |
2033 | #endif | |
2034 | iter->xhv_fill_lazy = 0; | |
2035 | iter->xhv_name_u.xhvnameu_name = 0; | |
2036 | iter->xhv_name_count = 0; | |
2037 | iter->xhv_backreferences = 0; | |
2038 | iter->xhv_mro_meta = NULL; | |
2039 | iter->xhv_aux_flags = 0; | |
2040 | return iter; | |
2041 | } | |
2042 | ||
0e0ab621 | 2043 | |
b464bac0 | 2044 | static struct xpvhv_aux* |
0e0ab621 | 2045 | S_hv_auxinit(pTHX_ HV *hv) { |
bfcb3514 | 2046 | struct xpvhv_aux *iter; |
b79f7545 | 2047 | char *array; |
bfcb3514 | 2048 | |
7918f24d NC |
2049 | PERL_ARGS_ASSERT_HV_AUXINIT; |
2050 | ||
0e0ab621 YO |
2051 | if (!SvOOK(hv)) { |
2052 | if (!HvARRAY(hv)) { | |
2053 | Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) | |
2054 | + sizeof(struct xpvhv_aux), char); | |
2055 | } else { | |
2056 | array = (char *) HvARRAY(hv); | |
2057 | Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) | |
2058 | + sizeof(struct xpvhv_aux), char); | |
2059 | } | |
2060 | HvARRAY(hv) = (HE**)array; | |
2061 | SvOOK_on(hv); | |
a7b39f85 | 2062 | iter = HvAUX(hv); |
6a5b4183 YO |
2063 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
2064 | if (PL_HASH_RAND_BITS_ENABLED) { | |
2065 | /* mix in some new state to PL_hash_rand_bits to "randomize" the traversal order*/ | |
2066 | if (PL_HASH_RAND_BITS_ENABLED == 1) | |
2067 | PL_hash_rand_bits += ptr_hash((PTRV)array); | |
2068 | PL_hash_rand_bits = ROTL_UV(PL_hash_rand_bits,1); | |
2069 | } | |
a7b39f85 | 2070 | iter->xhv_rand = (U32)PL_hash_rand_bits; |
6a5b4183 | 2071 | #endif |
a7b39f85 YO |
2072 | } else { |
2073 | iter = HvAUX(hv); | |
b79f7545 | 2074 | } |
bfcb3514 | 2075 | |
bea177f3 | 2076 | return hv_auxinit_internal(iter); |
bfcb3514 NC |
2077 | } |
2078 | ||
954c1994 GS |
2079 | /* |
2080 | =for apidoc hv_iterinit | |
2081 | ||
2082 | Prepares a starting point to traverse a hash table. Returns the number of | |
1b95d04f | 2083 | keys in the hash (i.e. the same as C<HvUSEDKEYS(hv)>). The return value is |
1c846c1f | 2084 | currently only meaningful for hashes without tie magic. |
954c1994 GS |
2085 | |
2086 | NOTE: Before version 5.004_65, C<hv_iterinit> used to return the number of | |
2087 | hash buckets that happen to be in use. If you still need that esoteric | |
b24b84ef | 2088 | value, you can get it through the macro C<HvFILL(hv)>. |
954c1994 | 2089 | |
e16e2ff8 | 2090 | |
954c1994 GS |
2091 | =cut |
2092 | */ | |
2093 | ||
79072805 | 2094 | I32 |
864dbfa3 | 2095 | Perl_hv_iterinit(pTHX_ HV *hv) |
79072805 | 2096 | { |
7918f24d NC |
2097 | PERL_ARGS_ASSERT_HV_ITERINIT; |
2098 | ||
2099 | /* FIXME: Are we not NULL, or do we croak? Place bets now! */ | |
2100 | ||
aa689395 | 2101 | if (!hv) |
cea2e8a9 | 2102 | Perl_croak(aTHX_ "Bad hash"); |
bfcb3514 | 2103 | |
b79f7545 | 2104 | if (SvOOK(hv)) { |
339441ef | 2105 | struct xpvhv_aux * iter = HvAUX(hv); |
0bd48802 | 2106 | HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */ |
bfcb3514 NC |
2107 | if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ |
2108 | HvLAZYDEL_off(hv); | |
2109 | hv_free_ent(hv, entry); | |
2110 | } | |
339441ef | 2111 | iter = HvAUX(hv); /* may have been reallocated */ |
bfcb3514 | 2112 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ |
4608196e | 2113 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
6a5b4183 | 2114 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
a7b39f85 | 2115 | iter->xhv_last_rand = iter->xhv_rand; |
6a5b4183 | 2116 | #endif |
bfcb3514 | 2117 | } else { |
6136c704 | 2118 | hv_auxinit(hv); |
72940dca | 2119 | } |
44a2ac75 | 2120 | |
cbec9347 | 2121 | /* used to be xhv->xhv_fill before 5.004_65 */ |
5d88ecd7 | 2122 | return HvTOTALKEYS(hv); |
79072805 | 2123 | } |
bfcb3514 NC |
2124 | |
2125 | I32 * | |
2126 | Perl_hv_riter_p(pTHX_ HV *hv) { | |
2127 | struct xpvhv_aux *iter; | |
2128 | ||
7918f24d NC |
2129 | PERL_ARGS_ASSERT_HV_RITER_P; |
2130 | ||
bfcb3514 NC |
2131 | if (!hv) |
2132 | Perl_croak(aTHX_ "Bad hash"); | |
2133 | ||
6136c704 | 2134 | iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
bfcb3514 NC |
2135 | return &(iter->xhv_riter); |
2136 | } | |
2137 | ||
2138 | HE ** | |
2139 | Perl_hv_eiter_p(pTHX_ HV *hv) { | |
2140 | struct xpvhv_aux *iter; | |
2141 | ||
7918f24d NC |
2142 | PERL_ARGS_ASSERT_HV_EITER_P; |
2143 | ||
bfcb3514 NC |
2144 | if (!hv) |
2145 | Perl_croak(aTHX_ "Bad hash"); | |
2146 | ||
6136c704 | 2147 | iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
bfcb3514 NC |
2148 | return &(iter->xhv_eiter); |
2149 | } | |
2150 | ||
2151 | void | |
2152 | Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) { | |
2153 | struct xpvhv_aux *iter; | |
2154 | ||
7918f24d NC |
2155 | PERL_ARGS_ASSERT_HV_RITER_SET; |
2156 | ||
bfcb3514 NC |
2157 | if (!hv) |
2158 | Perl_croak(aTHX_ "Bad hash"); | |
2159 | ||
b79f7545 NC |
2160 | if (SvOOK(hv)) { |
2161 | iter = HvAUX(hv); | |
2162 | } else { | |
bfcb3514 NC |
2163 | if (riter == -1) |
2164 | return; | |
2165 | ||
6136c704 | 2166 | iter = hv_auxinit(hv); |
bfcb3514 NC |
2167 | } |
2168 | iter->xhv_riter = riter; | |
2169 | } | |
2170 | ||
2171 | void | |
6a5b4183 YO |
2172 | Perl_hv_rand_set(pTHX_ HV *hv, U32 new_xhv_rand) { |
2173 | struct xpvhv_aux *iter; | |
2174 | ||
2175 | PERL_ARGS_ASSERT_HV_RAND_SET; | |
2176 | ||
2177 | #ifdef PERL_HASH_RANDOMIZE_KEYS | |
2178 | if (!hv) | |
2179 | Perl_croak(aTHX_ "Bad hash"); | |
2180 | ||
2181 | if (SvOOK(hv)) { | |
2182 | iter = HvAUX(hv); | |
2183 | } else { | |
2184 | iter = hv_auxinit(hv); | |
2185 | } | |
2186 | iter->xhv_rand = new_xhv_rand; | |
2187 | #else | |
2188 | Perl_croak(aTHX_ "This Perl has not been built with support for randomized hash key traversal but something called Perl_hv_rand_set()."); | |
2189 | #endif | |
2190 | } | |
2191 | ||
2192 | void | |
bfcb3514 NC |
2193 | Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) { |
2194 | struct xpvhv_aux *iter; | |
2195 | ||
7918f24d NC |
2196 | PERL_ARGS_ASSERT_HV_EITER_SET; |
2197 | ||
bfcb3514 NC |
2198 | if (!hv) |
2199 | Perl_croak(aTHX_ "Bad hash"); | |
2200 | ||
b79f7545 NC |
2201 | if (SvOOK(hv)) { |
2202 | iter = HvAUX(hv); | |
2203 | } else { | |
bfcb3514 NC |
2204 | /* 0 is the default so don't go malloc()ing a new structure just to |
2205 | hold 0. */ | |
2206 | if (!eiter) | |
2207 | return; | |
2208 | ||
6136c704 | 2209 | iter = hv_auxinit(hv); |
bfcb3514 NC |
2210 | } |
2211 | iter->xhv_eiter = eiter; | |
2212 | } | |
2213 | ||
bfcb3514 | 2214 | void |
4164be69 | 2215 | Perl_hv_name_set(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
bfcb3514 | 2216 | { |
97aff369 | 2217 | dVAR; |
b79f7545 | 2218 | struct xpvhv_aux *iter; |
7423f6db | 2219 | U32 hash; |
78b79c77 | 2220 | HEK **spot; |
46c461b5 | 2221 | |
7918f24d | 2222 | PERL_ARGS_ASSERT_HV_NAME_SET; |
bfcb3514 | 2223 | |
4164be69 NC |
2224 | if (len > I32_MAX) |
2225 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); | |
2226 | ||
b79f7545 NC |
2227 | if (SvOOK(hv)) { |
2228 | iter = HvAUX(hv); | |
15d9236d | 2229 | if (iter->xhv_name_u.xhvnameu_name) { |
b7247a80 | 2230 | if(iter->xhv_name_count) { |
745edda6 | 2231 | if(flags & HV_NAME_SETALL) { |
15d9236d | 2232 | HEK ** const name = HvAUX(hv)->xhv_name_u.xhvnameu_names; |
78b79c77 FC |
2233 | HEK **hekp = name + ( |
2234 | iter->xhv_name_count < 0 | |
2235 | ? -iter->xhv_name_count | |
2236 | : iter->xhv_name_count | |
2237 | ); | |
2238 | while(hekp-- > name+1) | |
b7247a80 | 2239 | unshare_hek_or_pvn(*hekp, 0, 0, 0); |
78b79c77 FC |
2240 | /* The first elem may be null. */ |
2241 | if(*name) unshare_hek_or_pvn(*name, 0, 0, 0); | |
b7247a80 | 2242 | Safefree(name); |
339441ef | 2243 | iter = HvAUX(hv); /* may been realloced */ |
15d9236d | 2244 | spot = &iter->xhv_name_u.xhvnameu_name; |
78b79c77 FC |
2245 | iter->xhv_name_count = 0; |
2246 | } | |
2247 | else { | |
78b79c77 FC |
2248 | if(iter->xhv_name_count > 0) { |
2249 | /* shift some things over */ | |
15d9236d NC |
2250 | Renew( |
2251 | iter->xhv_name_u.xhvnameu_names, iter->xhv_name_count + 1, HEK * | |
4c2bfb4f | 2252 | ); |
15d9236d | 2253 | spot = iter->xhv_name_u.xhvnameu_names; |
4c2bfb4f | 2254 | spot[iter->xhv_name_count] = spot[1]; |
78b79c77 | 2255 | spot[1] = spot[0]; |
4c2bfb4f | 2256 | iter->xhv_name_count = -(iter->xhv_name_count + 1); |
78b79c77 | 2257 | } |
15d9236d | 2258 | else if(*(spot = iter->xhv_name_u.xhvnameu_names)) { |
78b79c77 FC |
2259 | unshare_hek_or_pvn(*spot, 0, 0, 0); |
2260 | } | |
2261 | } | |
2262 | } | |
745edda6 | 2263 | else if (flags & HV_NAME_SETALL) { |
15d9236d | 2264 | unshare_hek_or_pvn(iter->xhv_name_u.xhvnameu_name, 0, 0, 0); |
339441ef | 2265 | iter = HvAUX(hv); /* may been realloced */ |
15d9236d | 2266 | spot = &iter->xhv_name_u.xhvnameu_name; |
b7247a80 | 2267 | } |
745edda6 | 2268 | else { |
15d9236d NC |
2269 | HEK * const existing_name = iter->xhv_name_u.xhvnameu_name; |
2270 | Newx(iter->xhv_name_u.xhvnameu_names, 2, HEK *); | |
745edda6 | 2271 | iter->xhv_name_count = -2; |
15d9236d | 2272 | spot = iter->xhv_name_u.xhvnameu_names; |
745edda6 FC |
2273 | spot[1] = existing_name; |
2274 | } | |
7423f6db | 2275 | } |
15d9236d | 2276 | else { spot = &iter->xhv_name_u.xhvnameu_name; iter->xhv_name_count = 0; } |
16580ff5 | 2277 | } else { |
bfcb3514 NC |
2278 | if (name == 0) |
2279 | return; | |
2280 | ||
6136c704 | 2281 | iter = hv_auxinit(hv); |
15d9236d | 2282 | spot = &iter->xhv_name_u.xhvnameu_name; |
bfcb3514 | 2283 | } |
7423f6db | 2284 | PERL_HASH(hash, name, len); |
c60dbbc3 | 2285 | *spot = name ? share_hek(name, flags & SVf_UTF8 ? -(I32)len : (I32)len, hash) : NULL; |
4643eb69 BF |
2286 | } |
2287 | ||
2288 | /* | |
2289 | This is basically sv_eq_flags() in sv.c, but we avoid the magic | |
2290 | and bytes checking. | |
2291 | */ | |
2292 | ||
2293 | STATIC I32 | |
2294 | hek_eq_pvn_flags(pTHX_ const HEK *hek, const char* pv, const I32 pvlen, const U32 flags) { | |
2295 | if ( (HEK_UTF8(hek) ? 1 : 0) != (flags & SVf_UTF8 ? 1 : 0) ) { | |
2296 | if (flags & SVf_UTF8) | |
2297 | return (bytes_cmp_utf8( | |
2298 | (const U8*)HEK_KEY(hek), HEK_LEN(hek), | |
2299 | (const U8*)pv, pvlen) == 0); | |
2300 | else | |
2301 | return (bytes_cmp_utf8( | |
2302 | (const U8*)pv, pvlen, | |
2303 | (const U8*)HEK_KEY(hek), HEK_LEN(hek)) == 0); | |
2304 | } | |
2305 | else | |
d35fec6c | 2306 | return HEK_LEN(hek) == pvlen && ((HEK_KEY(hek) == pv) |
4643eb69 | 2307 | || memEQ(HEK_KEY(hek), pv, pvlen)); |
bfcb3514 NC |
2308 | } |
2309 | ||
99206677 FC |
2310 | /* |
2311 | =for apidoc hv_ename_add | |
2312 | ||
db4fbf16 | 2313 | Adds a name to a stash's internal list of effective names. See |
99206677 FC |
2314 | C<hv_ename_delete>. |
2315 | ||
2316 | This is called when a stash is assigned to a new location in the symbol | |
2317 | table. | |
2318 | ||
2319 | =cut | |
2320 | */ | |
2321 | ||
ee72b38d | 2322 | void |
27a1175b | 2323 | Perl_hv_ename_add(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
ee72b38d FC |
2324 | { |
2325 | dVAR; | |
2326 | struct xpvhv_aux *aux = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); | |
2327 | U32 hash; | |
2328 | ||
78b79c77 | 2329 | PERL_ARGS_ASSERT_HV_ENAME_ADD; |
ee72b38d FC |
2330 | |
2331 | if (len > I32_MAX) | |
2332 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); | |
2333 | ||
2334 | PERL_HASH(hash, name, len); | |
2335 | ||
ee72b38d | 2336 | if (aux->xhv_name_count) { |
15d9236d | 2337 | HEK ** const xhv_name = aux->xhv_name_u.xhvnameu_names; |
78b79c77 FC |
2338 | I32 count = aux->xhv_name_count; |
2339 | HEK **hekp = xhv_name + (count < 0 ? -count : count); | |
ee72b38d FC |
2340 | while (hekp-- > xhv_name) |
2341 | if ( | |
4643eb69 BF |
2342 | (HEK_UTF8(*hekp) || (flags & SVf_UTF8)) |
2343 | ? hek_eq_pvn_flags(aTHX_ *hekp, name, (I32)len, flags) | |
2344 | : (HEK_LEN(*hekp) == (I32)len && memEQ(HEK_KEY(*hekp), name, len)) | |
2345 | ) { | |
78b79c77 FC |
2346 | if (hekp == xhv_name && count < 0) |
2347 | aux->xhv_name_count = -count; | |
2348 | return; | |
2349 | } | |
2350 | if (count < 0) aux->xhv_name_count--, count = -count; | |
2351 | else aux->xhv_name_count++; | |
15d9236d | 2352 | Renew(aux->xhv_name_u.xhvnameu_names, count + 1, HEK *); |
c60dbbc3 | 2353 | (aux->xhv_name_u.xhvnameu_names)[count] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash); |
ee72b38d FC |
2354 | } |
2355 | else { | |
15d9236d | 2356 | HEK *existing_name = aux->xhv_name_u.xhvnameu_name; |
ee72b38d | 2357 | if ( |
4643eb69 BF |
2358 | existing_name && ( |
2359 | (HEK_UTF8(existing_name) || (flags & SVf_UTF8)) | |
2360 | ? hek_eq_pvn_flags(aTHX_ existing_name, name, (I32)len, flags) | |
2361 | : (HEK_LEN(existing_name) == (I32)len && memEQ(HEK_KEY(existing_name), name, len)) | |
2362 | ) | |
ee72b38d | 2363 | ) return; |
15d9236d | 2364 | Newx(aux->xhv_name_u.xhvnameu_names, 2, HEK *); |
78b79c77 | 2365 | aux->xhv_name_count = existing_name ? 2 : -2; |
15d9236d | 2366 | *aux->xhv_name_u.xhvnameu_names = existing_name; |
c60dbbc3 | 2367 | (aux->xhv_name_u.xhvnameu_names)[1] = share_hek(name, (flags & SVf_UTF8 ? -(I32)len : (I32)len), hash); |
ee72b38d FC |
2368 | } |
2369 | } | |
2370 | ||
99206677 FC |
2371 | /* |
2372 | =for apidoc hv_ename_delete | |
2373 | ||
db4fbf16 | 2374 | Removes a name from a stash's internal list of effective names. If this is |
99206677 FC |
2375 | the name returned by C<HvENAME>, then another name in the list will take |
2376 | its place (C<HvENAME> will use it). | |
2377 | ||
2378 | This is called when a stash is deleted from the symbol table. | |
2379 | ||
2380 | =cut | |
2381 | */ | |
2382 | ||
ee72b38d | 2383 | void |
27a1175b | 2384 | Perl_hv_ename_delete(pTHX_ HV *hv, const char *name, U32 len, U32 flags) |
ee72b38d | 2385 | { |
ee72b38d FC |
2386 | struct xpvhv_aux *aux; |
2387 | ||
78b79c77 | 2388 | PERL_ARGS_ASSERT_HV_ENAME_DELETE; |
ee72b38d FC |
2389 | |
2390 | if (len > I32_MAX) | |
2391 | Perl_croak(aTHX_ "panic: hv name too long (%"UVuf")", (UV) len); | |
2392 | ||
2393 | if (!SvOOK(hv)) return; | |
2394 | ||
2395 | aux = HvAUX(hv); | |
15d9236d | 2396 | if (!aux->xhv_name_u.xhvnameu_name) return; |
ee72b38d FC |
2397 | |
2398 | if (aux->xhv_name_count) { | |
15d9236d | 2399 | HEK ** const namep = aux->xhv_name_u.xhvnameu_names; |
78b79c77 FC |
2400 | I32 const count = aux->xhv_name_count; |
2401 | HEK **victim = namep + (count < 0 ? -count : count); | |
2402 | while (victim-- > namep + 1) | |
ee72b38d | 2403 | if ( |
4643eb69 BF |
2404 | (HEK_UTF8(*victim) || (flags & SVf_UTF8)) |
2405 | ? hek_eq_pvn_flags(aTHX_ *victim, name, (I32)len, flags) | |
2406 | : (HEK_LEN(*victim) == (I32)len && memEQ(HEK_KEY(*victim), name, len)) | |
ee72b38d FC |
2407 | ) { |
2408 | unshare_hek_or_pvn(*victim, 0, 0, 0); | |
339441ef | 2409 | aux = HvAUX(hv); /* may been realloced */ |
78b79c77 FC |
2410 | if (count < 0) ++aux->xhv_name_count; |
2411 | else --aux->xhv_name_count; | |
2412 | if ( | |
2413 | (aux->xhv_name_count == 1 || aux->xhv_name_count == -1) | |
2414 | && !*namep | |
2415 | ) { /* if there are none left */ | |
ee72b38d | 2416 | Safefree(namep); |
15d9236d | 2417 | aux->xhv_name_u.xhvnameu_names = NULL; |
78b79c77 | 2418 | aux->xhv_name_count = 0; |
ee72b38d FC |
2419 | } |
2420 | else { | |
2421 | /* Move the last one back to fill the empty slot. It | |
2422 | does not matter what order they are in. */ | |
78b79c77 | 2423 | *victim = *(namep + (count < 0 ? -count : count) - 1); |
ee72b38d FC |
2424 | } |
2425 | return; | |
2426 | } | |
78b79c77 | 2427 | if ( |
4643eb69 BF |
2428 | count > 0 && (HEK_UTF8(*namep) || (flags & SVf_UTF8)) |
2429 | ? hek_eq_pvn_flags(aTHX_ *namep, name, (I32)len, flags) | |
2430 | : (HEK_LEN(*namep) == (I32)len && memEQ(HEK_KEY(*namep), name, len)) | |
78b79c77 FC |
2431 | ) { |
2432 | aux->xhv_name_count = -count; | |
2433 | } | |
ee72b38d FC |
2434 | } |
2435 | else if( | |
4643eb69 BF |
2436 | (HEK_UTF8(aux->xhv_name_u.xhvnameu_name) || (flags & SVf_UTF8)) |
2437 | ? hek_eq_pvn_flags(aTHX_ aux->xhv_name_u.xhvnameu_name, name, (I32)len, flags) | |
2438 | : (HEK_LEN(aux->xhv_name_u.xhvnameu_name) == (I32)len && | |
2439 | memEQ(HEK_KEY(aux->xhv_name_u.xhvnameu_name), name, len)) | |
ee72b38d | 2440 | ) { |
15d9236d NC |
2441 | HEK * const namehek = aux->xhv_name_u.xhvnameu_name; |
2442 | Newx(aux->xhv_name_u.xhvnameu_names, 1, HEK *); | |
2443 | *aux->xhv_name_u.xhvnameu_names = namehek; | |
3f783763 | 2444 | aux->xhv_name_count = -1; |
ee72b38d FC |
2445 | } |
2446 | } | |
2447 | ||
86f55936 NC |
2448 | AV ** |
2449 | Perl_hv_backreferences_p(pTHX_ HV *hv) { | |
6136c704 | 2450 | struct xpvhv_aux * const iter = SvOOK(hv) ? HvAUX(hv) : hv_auxinit(hv); |
7918f24d NC |
2451 | |
2452 | PERL_ARGS_ASSERT_HV_BACKREFERENCES_P; | |
7918f24d | 2453 | |
86f55936 NC |
2454 | return &(iter->xhv_backreferences); |
2455 | } | |
2456 | ||
09aad8f0 DM |
2457 | void |
2458 | Perl_hv_kill_backrefs(pTHX_ HV *hv) { | |
2459 | AV *av; | |
2460 | ||
2461 | PERL_ARGS_ASSERT_HV_KILL_BACKREFS; | |
2462 | ||
2463 | if (!SvOOK(hv)) | |
2464 | return; | |
2465 | ||
2466 | av = HvAUX(hv)->xhv_backreferences; | |
2467 | ||
2468 | if (av) { | |
2469 | HvAUX(hv)->xhv_backreferences = 0; | |
2470 | Perl_sv_kill_backrefs(aTHX_ MUTABLE_SV(hv), av); | |
5648c0ae | 2471 | if (SvTYPE(av) == SVt_PVAV) |
0c920c9b | 2472 | SvREFCNT_dec_NN(av); |
09aad8f0 DM |
2473 | } |
2474 | } | |
2475 | ||
954c1994 | 2476 | /* |
7a7b9979 NC |
2477 | hv_iternext is implemented as a macro in hv.h |
2478 | ||
954c1994 GS |
2479 | =for apidoc hv_iternext |
2480 | ||
2481 | Returns entries from a hash iterator. See C<hv_iterinit>. | |
2482 | ||
fe7bca90 NC |
2483 | You may call C<hv_delete> or C<hv_delete_ent> on the hash entry that the |
2484 | iterator currently points to, without losing your place or invalidating your | |
2485 | iterator. Note that in this case the current entry is deleted from the hash | |
2486 | with your iterator holding the last reference to it. Your iterator is flagged | |
2487 | to free the entry on the next call to C<hv_iternext>, so you must not discard | |
2488 | your iterator immediately else the entry will leak - call C<hv_iternext> to | |
2489 | trigger the resource deallocation. | |
2490 | ||
fe7bca90 NC |
2491 | =for apidoc hv_iternext_flags |
2492 | ||
2493 | Returns entries from a hash iterator. See C<hv_iterinit> and C<hv_iternext>. | |
2494 | The C<flags> value will normally be zero; if HV_ITERNEXT_WANTPLACEHOLDERS is | |
2495 | set the placeholders keys (for restricted hashes) will be returned in addition | |
72d33970 | 2496 | to normal keys. By default placeholders are automatically skipped over. |
7996736c | 2497 | Currently a placeholder is implemented with a value that is |
990c89d7 | 2498 | C<&PL_sv_placeholder>. Note that the implementation of placeholders and |
fe7bca90 NC |
2499 | restricted hashes may change, and the implementation currently is |
2500 | insufficiently abstracted for any change to be tidy. | |
e16e2ff8 | 2501 | |
fe7bca90 | 2502 | =cut |
e16e2ff8 NC |
2503 | */ |
2504 | ||
2505 | HE * | |
2506 | Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags) | |
2507 | { | |
27da23d5 | 2508 | dVAR; |
eb578fdb KW |
2509 | XPVHV* xhv; |
2510 | HE *entry; | |
a0d0e21e | 2511 | HE *oldentry; |
463ee0b2 | 2512 | MAGIC* mg; |
bfcb3514 | 2513 | struct xpvhv_aux *iter; |
79072805 | 2514 | |
7918f24d NC |
2515 | PERL_ARGS_ASSERT_HV_ITERNEXT_FLAGS; |
2516 | ||
79072805 | 2517 | if (!hv) |
cea2e8a9 | 2518 | Perl_croak(aTHX_ "Bad hash"); |
81714fb9 | 2519 | |
cbec9347 | 2520 | xhv = (XPVHV*)SvANY(hv); |
bfcb3514 | 2521 | |
b79f7545 | 2522 | if (!SvOOK(hv)) { |
bfcb3514 | 2523 | /* Too many things (well, pp_each at least) merrily assume that you can |
caee4c53 | 2524 | call hv_iternext without calling hv_iterinit, so we'll have to deal |
bfcb3514 NC |
2525 | with it. */ |
2526 | hv_iterinit(hv); | |
bfcb3514 | 2527 | } |
b79f7545 | 2528 | iter = HvAUX(hv); |
bfcb3514 NC |
2529 | |
2530 | oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ | |
e62cc96a | 2531 | if (SvMAGICAL(hv) && SvRMAGICAL(hv)) { |
ad64d0ec | 2532 | if ( ( mg = mg_find((const SV *)hv, PERL_MAGIC_tied) ) ) { |
e62cc96a YO |
2533 | SV * const key = sv_newmortal(); |
2534 | if (entry) { | |
2535 | sv_setsv(key, HeSVKEY_force(entry)); | |
2536 | SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */ | |
895cdc83 | 2537 | HeSVKEY_set(entry, NULL); |
e62cc96a YO |
2538 | } |
2539 | else { | |
2540 | char *k; | |
2541 | HEK *hek; | |
2542 | ||
2543 | /* one HE per MAGICAL hash */ | |
2544 | iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */ | |
895cdc83 | 2545 | HvLAZYDEL_on(hv); /* make sure entry gets freed */ |
e62cc96a | 2546 | Zero(entry, 1, HE); |
ad64d0ec | 2547 | Newxz(k, HEK_BASESIZE + sizeof(const SV *), char); |
e62cc96a YO |
2548 | hek = (HEK*)k; |
2549 | HeKEY_hek(entry) = hek; | |
2550 | HeKLEN(entry) = HEf_SVKEY; | |
2551 | } | |
ad64d0ec | 2552 | magic_nextpack(MUTABLE_SV(hv),mg,key); |
e62cc96a YO |
2553 | if (SvOK(key)) { |
2554 | /* force key to stay around until next time */ | |
2555 | HeSVKEY_set(entry, SvREFCNT_inc_simple_NN(key)); | |
2556 | return entry; /* beware, hent_val is not set */ | |
2557 | } | |
ef8d46e8 | 2558 | SvREFCNT_dec(HeVAL(entry)); |
e62cc96a YO |
2559 | Safefree(HeKEY_hek(entry)); |
2560 | del_HE(entry); | |
339441ef | 2561 | iter = HvAUX(hv); /* may been realloced */ |
e62cc96a | 2562 | iter->xhv_eiter = NULL; /* HvEITER(hv) = NULL */ |
895cdc83 | 2563 | HvLAZYDEL_off(hv); |
e62cc96a | 2564 | return NULL; |
81714fb9 | 2565 | } |
79072805 | 2566 | } |
7ee146b1 | 2567 | #if defined(DYNAMIC_ENV_FETCH) && !defined(__riscos__) /* set up %ENV for iteration */ |
ad64d0ec NC |
2568 | if (!entry && SvRMAGICAL((const SV *)hv) |
2569 | && mg_find((const SV *)hv, PERL_MAGIC_env)) { | |
f675dbe5 | 2570 | prime_env_iter(); |
03026e68 JM |
2571 | #ifdef VMS |
2572 | /* The prime_env_iter() on VMS just loaded up new hash values | |
2573 | * so the iteration count needs to be reset back to the beginning | |
2574 | */ | |
2575 | hv_iterinit(hv); | |
2576 | iter = HvAUX(hv); | |
2577 | oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ | |
2578 | #endif | |
2579 | } | |
f675dbe5 | 2580 | #endif |
463ee0b2 | 2581 | |
bfaf5b52 | 2582 | /* hv_iterinit now ensures this. */ |
b79f7545 NC |
2583 | assert (HvARRAY(hv)); |
2584 | ||
015a5f36 | 2585 | /* At start of hash, entry is NULL. */ |
fde52b5c | 2586 | if (entry) |
8aacddc1 | 2587 | { |
fde52b5c | 2588 | entry = HeNEXT(entry); |
e16e2ff8 NC |
2589 | if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) { |
2590 | /* | |
2591 | * Skip past any placeholders -- don't want to include them in | |
2592 | * any iteration. | |
2593 | */ | |
7996736c | 2594 | while (entry && HeVAL(entry) == &PL_sv_placeholder) { |
e16e2ff8 NC |
2595 | entry = HeNEXT(entry); |
2596 | } | |
8aacddc1 NIS |
2597 | } |
2598 | } | |
6a5b4183 YO |
2599 | |
2600 | #ifdef PERL_HASH_RANDOMIZE_KEYS | |
a7b39f85 YO |
2601 | if (iter->xhv_last_rand != iter->xhv_rand) { |
2602 | if (iter->xhv_riter != -1) { | |
2603 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), | |
2604 | "Use of each() on hash after insertion without resetting hash iterator results in undefined behavior" | |
2605 | pTHX__FORMAT | |
2606 | pTHX__VALUE); | |
2607 | } | |
339441ef | 2608 | iter = HvAUX(hv); /* may been realloced */ |
a7b39f85 YO |
2609 | iter->xhv_last_rand = iter->xhv_rand; |
2610 | } | |
6a5b4183 | 2611 | #endif |
015a5f36 | 2612 | |
9eb4ebd1 NC |
2613 | /* Skip the entire loop if the hash is empty. */ |
2614 | if ((flags & HV_ITERNEXT_WANTPLACEHOLDERS) | |
2615 | ? HvTOTALKEYS(hv) : HvUSEDKEYS(hv)) { | |
900ac051 MM |
2616 | while (!entry) { |
2617 | /* OK. Come to the end of the current list. Grab the next one. */ | |
2618 | ||
2619 | iter->xhv_riter++; /* HvRITER(hv)++ */ | |
2620 | if (iter->xhv_riter > (I32)xhv->xhv_max /* HvRITER(hv) > HvMAX(hv) */) { | |
2621 | /* There is no next one. End of the hash. */ | |
2622 | iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ | |
6a5b4183 YO |
2623 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
2624 | iter->xhv_last_rand = iter->xhv_rand; /* reset xhv_last_rand so we can detect inserts during traversal */ | |
2625 | #endif | |
900ac051 MM |
2626 | break; |
2627 | } | |
6a5b4183 | 2628 | entry = (HvARRAY(hv))[ PERL_HASH_ITER_BUCKET(iter) & xhv->xhv_max ]; |
8aacddc1 | 2629 | |
900ac051 MM |
2630 | if (!(flags & HV_ITERNEXT_WANTPLACEHOLDERS)) { |
2631 | /* If we have an entry, but it's a placeholder, don't count it. | |
2632 | Try the next. */ | |
2633 | while (entry && HeVAL(entry) == &PL_sv_placeholder) | |
2634 | entry = HeNEXT(entry); | |
2635 | } | |
2636 | /* Will loop again if this linked list starts NULL | |
2637 | (for HV_ITERNEXT_WANTPLACEHOLDERS) | |
2638 | or if we run through it and find only placeholders. */ | |
015a5f36 | 2639 | } |
fde52b5c | 2640 | } |
a7b39f85 YO |
2641 | else { |
2642 | iter->xhv_riter = -1; | |
6a5b4183 | 2643 | #ifdef PERL_HASH_RANDOMIZE_KEYS |
a7b39f85 | 2644 | iter->xhv_last_rand = iter->xhv_rand; |
6a5b4183 | 2645 | #endif |
a7b39f85 | 2646 | } |
79072805 | 2647 | |
72940dca | 2648 | if (oldentry && HvLAZYDEL(hv)) { /* was deleted earlier? */ |
2649 | HvLAZYDEL_off(hv); | |
68dc0745 | 2650 | hv_free_ent(hv, oldentry); |
72940dca | 2651 | } |
a0d0e21e | 2652 | |
339441ef | 2653 | iter = HvAUX(hv); /* may been realloced */ |
bfcb3514 | 2654 | iter->xhv_eiter = entry; /* HvEITER(hv) = entry */ |
79072805 LW |
2655 | return entry; |
2656 | } | |
2657 | ||
954c1994 GS |
2658 | /* |
2659 | =for apidoc hv_iterkey | |
2660 | ||
2661 | Returns the key from the current position of the hash iterator. See | |
2662 | C<hv_iterinit>. | |
2663 | ||
2664 | =cut | |
2665 | */ | |
2666 | ||
79072805 | 2667 | char * |
5aaab254 | 2668 | Perl_hv_iterkey(pTHX_ HE *entry, I32 *retlen) |
79072805 | 2669 | { |
7918f24d NC |
2670 | PERL_ARGS_ASSERT_HV_ITERKEY; |
2671 | ||
fde52b5c | 2672 | if (HeKLEN(entry) == HEf_SVKEY) { |
fb73857a | 2673 | STRLEN len; |
0bd48802 | 2674 | char * const p = SvPV(HeKEY_sv(entry), len); |
fb73857a | 2675 | *retlen = len; |
2676 | return p; | |
fde52b5c | 2677 | } |
2678 | else { | |
2679 | *retlen = HeKLEN(entry); | |
2680 | return HeKEY(entry); | |
2681 | } | |
2682 | } | |
2683 | ||
2684 | /* unlike hv_iterval(), this always returns a mortal copy of the key */ | |
954c1994 GS |
2685 | /* |
2686 | =for apidoc hv_iterkeysv | |
2687 | ||
2688 | Returns the key as an C<SV*> from the current position of the hash | |
2689 | iterator. The return value will always be a mortal copy of the key. Also | |
2690 | see C<hv_iterinit>. | |
2691 | ||
2692 | =cut | |
2693 | */ | |
2694 | ||
fde52b5c | 2695 | SV * |
5aaab254 | 2696 | Perl_hv_iterkeysv(pTHX_ HE *entry) |
fde52b5c | 2697 | { |
7918f24d NC |
2698 | PERL_ARGS_ASSERT_HV_ITERKEYSV; |
2699 | ||
c1b02ed8 | 2700 | return sv_2mortal(newSVhek(HeKEY_hek(entry))); |
79072805 LW |
2701 | } |
2702 | ||
954c1994 GS |
2703 | /* |
2704 | =for apidoc hv_iterval | |
2705 | ||
2706 | Returns the value from the current position of the hash iterator. See | |
2707 | C<hv_iterkey>. | |
2708 | ||
2709 | =cut | |
2710 | */ | |
2711 | ||
79072805 | 2712 | SV * |
5aaab254 | 2713 | Perl_hv_iterval(pTHX_ HV *hv, HE *entry) |
79072805 | 2714 | { |
7918f24d NC |
2715 | PERL_ARGS_ASSERT_HV_ITERVAL; |
2716 | ||
8990e307 | 2717 | if (SvRMAGICAL(hv)) { |
ad64d0ec | 2718 | if (mg_find((const SV *)hv, PERL_MAGIC_tied)) { |
c4420975 | 2719 | SV* const sv = sv_newmortal(); |
bbce6d69 | 2720 | if (HeKLEN(entry) == HEf_SVKEY) |
ad64d0ec | 2721 | mg_copy(MUTABLE_SV(hv), sv, (char*)HeKEY_sv(entry), HEf_SVKEY); |
a3b680e6 | 2722 | else |
ad64d0ec | 2723 | mg_copy(MUTABLE_SV(hv), sv, HeKEY(entry), HeKLEN(entry)); |
463ee0b2 LW |
2724 | return sv; |
2725 | } | |
79072805 | 2726 | } |
fde52b5c | 2727 | return HeVAL(entry); |
79072805 LW |
2728 | } |
2729 | ||
954c1994 GS |
2730 | /* |
2731 | =for apidoc hv_iternextsv | |
2732 | ||
2733 | Performs an C<hv_iternext>, C<hv_iterkey>, and C<hv_iterval> in one | |
2734 | operation. | |
2735 | ||
2736 | =cut | |
2737 | */ | |
2738 | ||
a0d0e21e | 2739 | SV * |
864dbfa3 | 2740 | Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen) |
a0d0e21e | 2741 | { |
0bd48802 AL |
2742 | HE * const he = hv_iternext_flags(hv, 0); |
2743 | ||
7918f24d NC |
2744 | PERL_ARGS_ASSERT_HV_ITERNEXTSV; |
2745 | ||
0bd48802 | 2746 | if (!he) |
a0d0e21e LW |
2747 | return NULL; |
2748 | *key = hv_iterkey(he, retlen); | |
2749 | return hv_iterval(hv, he); | |
2750 | } | |
2751 | ||
954c1994 | 2752 | /* |
bc5cdc23 NC |
2753 | |
2754 | Now a macro in hv.h | |
2755 | ||
954c1994 GS |
2756 | =for apidoc hv_magic |
2757 | ||
2758 | Adds magic to a hash. See C<sv_magic>. | |
2759 | ||
2760 | =cut | |
2761 | */ | |
2762 | ||
bbce6d69 | 2763 | /* possibly free a shared string if no one has access to it |
fde52b5c | 2764 | * len and hash must both be valid for str. |
2765 | */ | |
bbce6d69 | 2766 | void |
864dbfa3 | 2767 | Perl_unsharepvn(pTHX_ const char *str, I32 len, U32 hash) |
fde52b5c | 2768 | { |
19692e8d NC |
2769 | unshare_hek_or_pvn (NULL, str, len, hash); |
2770 | } | |
2771 | ||
2772 | ||
2773 | void | |
2774 | Perl_unshare_hek(pTHX_ HEK *hek) | |
2775 | { | |
bf11fd37 | 2776 | assert(hek); |
19692e8d NC |
2777 | unshare_hek_or_pvn(hek, NULL, 0, 0); |
2778 | } | |
2779 | ||
2780 | /* possibly free a shared string if no one has access to it | |
2781 | hek if non-NULL takes priority over the other 3, else str, len and hash | |
2782 | are used. If so, len and hash must both be valid for str. | |
2783 | */ | |
df132699 | 2784 | STATIC void |
97ddebaf | 2785 | S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash) |
19692e8d | 2786 | { |
eb578fdb | 2787 | XPVHV* xhv; |
20454177 | 2788 | HE *entry; |
eb578fdb | 2789 | HE **oentry; |
c3654f1a | 2790 | bool is_utf8 = FALSE; |
19692e8d | 2791 | int k_flags = 0; |
aec46f14 | 2792 | const char * const save = str; |
cbbf8932 | 2793 | struct shared_he *he = NULL; |
c3654f1a | 2794 | |
19692e8d | 2795 | if (hek) { |
cbae3960 NC |
2796 | /* Find the shared he which is just before us in memory. */ |
2797 | he = (struct shared_he *)(((char *)hek) | |
2798 | - STRUCT_OFFSET(struct shared_he, | |
2799 | shared_he_hek)); | |
2800 | ||
2801 | /* Assert that the caller passed us a genuine (or at least consistent) | |
2802 | shared hek */ | |
2803 | assert (he->shared_he_he.hent_hek == hek); | |
29404ae0 | 2804 | |
de616631 NC |
2805 | if (he->shared_he_he.he_valu.hent_refcount - 1) { |
2806 | --he->shared_he_he.he_valu.hent_refcount; | |
29404ae0 NC |
2807 | return; |
2808 | } | |
29404ae0 | 2809 | |
19692e8d NC |
2810 | hash = HEK_HASH(hek); |
2811 | } else if (len < 0) { | |
2812 | STRLEN tmplen = -len; | |
2813 | is_utf8 = TRUE; | |
2814 | /* See the note in hv_fetch(). --jhi */ | |
2815 | str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8); | |
2816 | len = tmplen; | |
2817 | if (is_utf8) | |
2818 | k_flags = HVhek_UTF8; | |
2819 | if (str != save) | |
2820 | k_flags |= HVhek_WASUTF8 | HVhek_FREEKEY; | |
c3654f1a | 2821 | } |
1c846c1f | 2822 | |
de616631 | 2823 | /* what follows was the moral equivalent of: |
6b88bc9c | 2824 | if ((Svp = hv_fetch(PL_strtab, tmpsv, FALSE, hash))) { |
a0714e2c | 2825 | if (--*Svp == NULL) |
6b88bc9c | 2826 | hv_delete(PL_strtab, str, len, G_DISCARD, hash); |
bbce6d69 | 2827 | } */ |
cbec9347 | 2828 | xhv = (XPVHV*)SvANY(PL_strtab); |
fde52b5c | 2829 | /* assert(xhv_array != 0) */ |
9de10d5c | 2830 | oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)]; |
6c1b96a1 NC |
2831 | if (he) { |
2832 | const HE *const he_he = &(he->shared_he_he); | |
45d1cc86 | 2833 | for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
35ab5632 NC |
2834 | if (entry == he_he) |
2835 | break; | |
19692e8d NC |
2836 | } |
2837 | } else { | |
35a4481c | 2838 | const int flags_masked = k_flags & HVhek_MASK; |
45d1cc86 | 2839 | for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { |
19692e8d NC |
2840 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
2841 | continue; | |
2842 | if (HeKLEN(entry) != len) | |
2843 | continue; | |
2844 | if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */ | |
2845 | continue; | |
2846 | if (HeKFLAGS(entry) != flags_masked) | |
2847 | continue; | |
19692e8d NC |
2848 | break; |
2849 | } | |
2850 | } | |
2851 | ||
35ab5632 NC |
2852 | if (entry) { |
2853 | if (--entry->he_valu.hent_refcount == 0) { | |
19692e8d | 2854 | *oentry = HeNEXT(entry); |
cbae3960 | 2855 | Safefree(entry); |
4c7185a0 | 2856 | xhv->xhv_keys--; /* HvTOTALKEYS(hv)-- */ |
19692e8d | 2857 | } |
fde52b5c | 2858 | } |
19692e8d | 2859 | |
9b387841 NC |
2860 | if (!entry) |
2861 | Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), | |
12578ffb | 2862 | "Attempt to free nonexistent shared string '%s'%s" |
9b387841 NC |
2863 | pTHX__FORMAT, |
2864 | hek ? HEK_KEY(hek) : str, | |
2865 | ((k_flags & HVhek_UTF8) ? " (utf8)" : "") pTHX__VALUE); | |
19692e8d NC |
2866 | if (k_flags & HVhek_FREEKEY) |
2867 | Safefree(str); | |
fde52b5c | 2868 | } |
2869 | ||
bbce6d69 | 2870 | /* get a (constant) string ptr from the global string table |
2871 | * string will get added if it is not already there. | |
fde52b5c | 2872 | * len and hash must both be valid for str. |
2873 | */ | |
bbce6d69 | 2874 | HEK * |
5aaab254 | 2875 | Perl_share_hek(pTHX_ const char *str, I32 len, U32 hash) |
fde52b5c | 2876 | { |
da58a35d | 2877 | bool is_utf8 = FALSE; |
19692e8d | 2878 | int flags = 0; |
aec46f14 | 2879 | const char * const save = str; |
da58a35d | 2880 | |
7918f24d NC |
2881 | PERL_ARGS_ASSERT_SHARE_HEK; |
2882 | ||
da58a35d | 2883 | if (len < 0) { |
77caf834 | 2884 | STRLEN tmplen = -len; |
da58a35d | 2885 | is_utf8 = TRUE; |
77caf834 JH |
2886 | /* See the note in hv_fetch(). --jhi */ |
2887 | str = (char*)bytes_from_utf8((U8*)str, &tmplen, &is_utf8); | |
2888 | len = tmplen; | |
19692e8d NC |
2889 | /* If we were able to downgrade here, then than means that we were passed |
2890 | in a key which only had chars 0-255, but was utf8 encoded. */ | |
2891 | if (is_utf8) | |
2892 | flags = HVhek_UTF8; | |
2893 | /* If we found we were able to downgrade the string to bytes, then | |
2894 | we should flag that it needs upgrading on keys or each. Also flag | |
2895 | that we need share_hek_flags to free the string. */ | |
4643eb69 | 2896 | if (str != save) { |
c2587955 | 2897 | dVAR; |
4643eb69 | 2898 | PERL_HASH(hash, str, len); |
19692e8d | 2899 | flags |= HVhek_WASUTF8 | HVhek_FREEKEY; |
4643eb69 | 2900 | } |
19692e8d NC |
2901 | } |
2902 | ||
6e838c70 | 2903 | return share_hek_flags (str, len, hash, flags); |
19692e8d NC |
2904 | } |
2905 | ||
6e838c70 | 2906 | STATIC HEK * |
5aaab254 | 2907 | S_share_hek_flags(pTHX_ const char *str, I32 len, U32 hash, int flags) |
19692e8d | 2908 | { |
eb578fdb | 2909 | HE *entry; |
35a4481c | 2910 | const int flags_masked = flags & HVhek_MASK; |
263cb4a6 | 2911 | const U32 hindex = hash & (I32) HvMAX(PL_strtab); |
eb578fdb | 2912 | XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab); |
7918f24d NC |
2913 | |
2914 | PERL_ARGS_ASSERT_SHARE_HEK_FLAGS; | |
bbce6d69 | 2915 | |
fde52b5c | 2916 | /* what follows is the moral equivalent of: |
1c846c1f | 2917 | |
6b88bc9c | 2918 | if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE))) |
a0714e2c | 2919 | hv_store(PL_strtab, str, len, NULL, hash); |
fdcd69b6 NC |
2920 | |
2921 | Can't rehash the shared string table, so not sure if it's worth | |
2922 | counting the number of entries in the linked list | |
bbce6d69 | 2923 | */ |
7918f24d | 2924 | |
fde52b5c | 2925 | /* assert(xhv_array != 0) */ |
263cb4a6 NC |
2926 | entry = (HvARRAY(PL_strtab))[hindex]; |
2927 | for (;entry; entry = HeNEXT(entry)) { | |
fde52b5c | 2928 | if (HeHASH(entry) != hash) /* strings can't be equal */ |
2929 | continue; | |
2930 | if (HeKLEN(entry) != len) | |
2931 | continue; | |
1c846c1f | 2932 | if (HeKEY(entry) != str && memNE(HeKEY(entry),str,len)) /* is this it? */ |
fde52b5c | 2933 | continue; |
19692e8d | 2934 | if (HeKFLAGS(entry) != flags_masked) |
c3654f1a | 2935 | continue; |
fde52b5c | 2936 | break; |
2937 | } | |
263cb4a6 NC |
2938 | |
2939 | if (!entry) { | |
45d1cc86 NC |
2940 | /* What used to be head of the list. |
2941 | If this is NULL, then we're the first entry for this slot, which | |
2942 | means we need to increate fill. */ | |
cbae3960 NC |
2943 | struct shared_he *new_entry; |
2944 | HEK *hek; | |
2945 | char *k; | |
263cb4a6 NC |
2946 | HE **const head = &HvARRAY(PL_strtab)[hindex]; |
2947 | HE *const next = *head; | |
cbae3960 NC |
2948 | |
2949 | /* We don't actually store a HE from the arena and a regular HEK. | |
2950 | Instead we allocate one chunk of memory big enough for both, | |
2951 | and put the HEK straight after the HE. This way we can find the | |
f52337cf | 2952 | HE directly from the HEK. |
cbae3960 NC |
2953 | */ |
2954 | ||
a02a5408 | 2955 | Newx(k, STRUCT_OFFSET(struct shared_he, |
cbae3960 NC |
2956 | shared_he_hek.hek_key[0]) + len + 2, char); |
2957 | new_entry = (struct shared_he *)k; | |
2958 | entry = &(new_entry->shared_he_he); | |
2959 | hek = &(new_entry->shared_he_hek); | |
2960 | ||
2961 | Copy(str, HEK_KEY(hek), len, char); | |
2962 | HEK_KEY(hek)[len] = 0; | |
2963 | HEK_LEN(hek) = len; | |
2964 | HEK_HASH(hek) = hash; | |
2965 | HEK_FLAGS(hek) = (unsigned char)flags_masked; | |
2966 | ||
2967 | /* Still "point" to the HEK, so that other code need not know what | |
2968 | we're up to. */ | |
2969 | HeKEY_hek(entry) = hek; | |
de616631 | 2970 | entry->he_valu.hent_refcount = 0; |
263cb4a6 NC |
2971 | HeNEXT(entry) = next; |
2972 | *head = entry; | |
cbae3960 | 2973 | |
4c7185a0 | 2974 | xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */ |
263cb4a6 | 2975 | if (!next) { /* initial entry? */ |
8e317198 | 2976 | } else if ( DO_HSPLIT(xhv) ) { |
adf6906b NC |
2977 | const STRLEN oldsize = xhv->xhv_max + 1; |
2978 | hsplit(PL_strtab, oldsize, oldsize * 2); | |
bbce6d69 | 2979 | } |
2980 | } | |
2981 | ||
de616631 | 2982 | ++entry->he_valu.hent_refcount; |
19692e8d NC |
2983 | |
2984 | if (flags & HVhek_FREEKEY) | |
f9a63242 | 2985 | Safefree(str); |
19692e8d | 2986 | |
6e838c70 | 2987 | return HeKEY_hek(entry); |
fde52b5c | 2988 | } |
ecae49c0 | 2989 | |
6174b39a | 2990 | SSize_t * |
ca732855 NC |
2991 | Perl_hv_placeholders_p(pTHX_ HV *hv) |
2992 | { | |
ad64d0ec | 2993 | MAGIC *mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
ca732855 | 2994 | |
7918f24d NC |
2995 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_P; |
2996 | ||
ca732855 | 2997 | if (!mg) { |
ad64d0ec | 2998 | mg = sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, 0); |
ca732855 NC |
2999 | |
3000 | if (!mg) { | |
3001 | Perl_die(aTHX_ "panic: hv_placeholders_p"); | |
3002 | } | |
3003 | } | |
3004 | return &(mg->mg_len); | |
3005 | } | |
3006 | ||
3007 | ||
3008 | I32 | |
0c289d13 | 3009 | Perl_hv_placeholders_get(pTHX_ const HV *hv) |
ca732855 | 3010 | { |
0c289d13 | 3011 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
ca732855 | 3012 | |
7918f24d | 3013 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_GET; |
23491f1d | 3014 | PERL_UNUSED_CONTEXT; |
7918f24d | 3015 | |
ca732855 NC |
3016 | return mg ? mg->mg_len : 0; |
3017 | } | |
3018 | ||
3019 | void | |
ac1e784a | 3020 | Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph) |
ca732855 | 3021 | { |
ad64d0ec | 3022 | MAGIC * const mg = mg_find((const SV *)hv, PERL_MAGIC_rhash); |
ca732855 | 3023 | |
7918f24d NC |
3024 | PERL_ARGS_ASSERT_HV_PLACEHOLDERS_SET; |
3025 | ||
ca732855 NC |
3026 | if (mg) { |
3027 | mg->mg_len = ph; | |
3028 | } else if (ph) { | |
ad64d0ec | 3029 | if (!sv_magicext(MUTABLE_SV(hv), 0, PERL_MAGIC_rhash, 0, 0, ph)) |
ca732855 NC |
3030 | Perl_die(aTHX_ "panic: hv_placeholders_set"); |
3031 | } | |
3032 | /* else we don't need to add magic to record 0 placeholders. */ | |
3033 | } | |
ecae49c0 | 3034 | |
2a49f0f5 | 3035 | STATIC SV * |
7b0bddfa NC |
3036 | S_refcounted_he_value(pTHX_ const struct refcounted_he *he) |
3037 | { | |
0b2d3faa | 3038 | dVAR; |
7b0bddfa | 3039 | SV *value; |
7918f24d NC |
3040 | |
3041 | PERL_ARGS_ASSERT_REFCOUNTED_HE_VALUE; | |
3042 | ||
7b0bddfa NC |
3043 | switch(he->refcounted_he_data[0] & HVrhek_typemask) { |
3044 | case HVrhek_undef: | |
3045 | value = newSV(0); | |
3046 | break; | |
3047 | case HVrhek_delete: | |
3048 | value = &PL_sv_placeholder; | |
3049 | break; | |
3050 | case HVrhek_IV: | |
44ebaf21 NC |
3051 | value = newSViv(he->refcounted_he_val.refcounted_he_u_iv); |
3052 | break; | |
3053 | case HVrhek_UV: | |
3054 | value = newSVuv(he->refcounted_he_val.refcounted_he_u_uv); | |
7b0bddfa NC |
3055 | break; |
3056 | case HVrhek_PV: | |
44ebaf21 | 3057 | case HVrhek_PV_UTF8: |
7b0bddfa NC |
3058 | /* Create a string SV that directly points to the bytes in our |
3059 | structure. */ | |
b9f83d2f | 3060 | value = newSV_type(SVt_PV); |
7b0bddfa NC |
3061 | SvPV_set(value, (char *) he->refcounted_he_data + 1); |
3062 | SvCUR_set(value, he->refcounted_he_val.refcounted_he_u_len); | |
3063 | /* This stops anything trying to free it */ | |
3064 | SvLEN_set(value, 0); | |
3065 | SvPOK_on(value); | |
3066 | SvREADONLY_on(value); | |
44ebaf21 | 3067 | if ((he->refcounted_he_data[0] & HVrhek_typemask) == HVrhek_PV_UTF8) |
7b0bddfa NC |
3068 | SvUTF8_on(value); |
3069 | break; | |
3070 | default: | |
20439bc7 Z |
3071 | Perl_croak(aTHX_ "panic: refcounted_he_value bad flags %"UVxf, |
3072 | (UV)he->refcounted_he_data[0]); | |
7b0bddfa NC |
3073 | } |
3074 | return value; | |
3075 | } | |
3076 | ||
ecae49c0 | 3077 | /* |
20439bc7 | 3078 | =for apidoc m|HV *|refcounted_he_chain_2hv|const struct refcounted_he *c|U32 flags |
8dff4fc5 | 3079 | |
20439bc7 Z |
3080 | Generates and returns a C<HV *> representing the content of a |
3081 | C<refcounted_he> chain. | |
3082 | I<flags> is currently unused and must be zero. | |
8dff4fc5 BM |
3083 | |
3084 | =cut | |
3085 | */ | |
3086 | HV * | |
20439bc7 | 3087 | Perl_refcounted_he_chain_2hv(pTHX_ const struct refcounted_he *chain, U32 flags) |
8dff4fc5 | 3088 | { |
20439bc7 Z |
3089 | dVAR; |
3090 | HV *hv; | |
3091 | U32 placeholders, max; | |
b3ca2e83 | 3092 | |
20439bc7 Z |
3093 | if (flags) |
3094 | Perl_croak(aTHX_ "panic: refcounted_he_chain_2hv bad flags %"UVxf, | |
3095 | (UV)flags); | |
b3ca2e83 | 3096 | |
b3ca2e83 NC |
3097 | /* We could chase the chain once to get an idea of the number of keys, |
3098 | and call ksplit. But for now we'll make a potentially inefficient | |
3099 | hash with only 8 entries in its array. */ | |
20439bc7 Z |
3100 | hv = newHV(); |
3101 | max = HvMAX(hv); | |
b3ca2e83 NC |
3102 | if (!HvARRAY(hv)) { |
3103 | char *array; | |
3104 | Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(max + 1), char); | |
3105 | HvARRAY(hv) = (HE**)array; | |
3106 | } | |
3107 | ||
20439bc7 | 3108 | placeholders = 0; |
b3ca2e83 | 3109 | while (chain) { |
cbb1fbea | 3110 | #ifdef USE_ITHREADS |
b6bbf3fa | 3111 | U32 hash = chain->refcounted_he_hash; |
cbb1fbea NC |
3112 | #else |
3113 | U32 hash = HEK_HASH(chain->refcounted_he_hek); | |
3114 | #endif | |
b3ca2e83 NC |
3115 | HE **oentry = &((HvARRAY(hv))[hash & max]); |
3116 | HE *entry = *oentry; | |
b6bbf3fa | 3117 | SV *value; |
cbb1fbea | 3118 | |
b3ca2e83 NC |
3119 | for (; entry; entry = HeNEXT(entry)) { |
3120 | if (HeHASH(entry) == hash) { | |
9f769845 NC |
3121 | /* We might have a duplicate key here. If so, entry is older |
3122 | than the key we've already put in the hash, so if they are | |
3123 | the same, skip adding entry. */ | |
3124 | #ifdef USE_ITHREADS | |
3125 | const STRLEN klen = HeKLEN(entry); | |
3126 | const char *const key = HeKEY(entry); | |
3127 | if (klen == chain->refcounted_he_keylen | |
3128 | && (!!HeKUTF8(entry) | |
3129 | == !!(chain->refcounted_he_data[0] & HVhek_UTF8)) | |
3130 | && memEQ(key, REF_HE_KEY(chain), klen)) | |
3131 | goto next_please; | |
3132 | #else | |
3133 | if (HeKEY_hek(entry) == chain->refcounted_he_hek) | |
3134 | goto next_please; | |
3135 | if (HeKLEN(entry) == HEK_LEN(chain->refcounted_he_hek) | |
3136 | && HeKUTF8(entry) == HEK_UTF8(chain->refcounted_he_hek) | |
3137 | && memEQ(HeKEY(entry), HEK_KEY(chain->refcounted_he_hek), | |
3138 | HeKLEN(entry))) | |
3139 | goto next_please; | |
3140 | #endif | |
b3ca2e83 NC |
3141 | } |
3142 | } | |
3143 | assert (!entry); | |
3144 | entry = new_HE(); | |
3145 | ||
cbb1fbea NC |
3146 | #ifdef USE_ITHREADS |
3147 | HeKEY_hek(entry) | |
7b0bddfa | 3148 | = share_hek_flags(REF_HE_KEY(chain), |
b6bbf3fa NC |
3149 | chain->refcounted_he_keylen, |
3150 | chain->refcounted_he_hash, | |
3151 | (chain->refcounted_he_data[0] | |
3152 | & (HVhek_UTF8|HVhek_WASUTF8))); | |
cbb1fbea | 3153 | #else |
71ad1b0c | 3154 | HeKEY_hek(entry) = share_hek_hek(chain->refcounted_he_hek); |
cbb1fbea | 3155 | #endif |
7b0bddfa NC |
3156 | value = refcounted_he_value(chain); |
3157 | if (value == &PL_sv_placeholder) | |
b3ca2e83 | 3158 | placeholders++; |
b6bbf3fa | 3159 | HeVAL(entry) = value; |
b3ca2e83 NC |
3160 | |
3161 | /* Link it into the chain. */ | |
3162 | HeNEXT(entry) = *oentry; | |
b3ca2e83 NC |
3163 | *oentry = entry; |
3164 | ||
3165 | HvTOTALKEYS(hv)++; | |
3166 | ||
3167 | next_please: | |
71ad1b0c | 3168 | chain = chain->refcounted_he_next; |
b3ca2e83 NC |
3169 | } |
3170 | ||
3171 | if (placeholders) { | |
3172 | clear_placeholders(hv, placeholders); | |
3173 | HvTOTALKEYS(hv) -= placeholders; | |
3174 | } | |
3175 | ||
3176 | /* We could check in the loop to see if we encounter any keys with key | |
3177 | flags, but it's probably not worth it, as this per-hash flag is only | |
3178 | really meant as an optimisation for things like Storable. */ | |
3179 | HvHASKFLAGS_on(hv); | |
def9038f | 3180 | DEBUG_A(Perl_hv_assert(aTHX_ hv)); |
b3ca2e83 NC |
3181 | |
3182 | return hv; | |
3183 | } | |
3184 | ||
20439bc7 Z |
3185 | /* |
3186 | =for apidoc m|SV *|refcounted_he_fetch_pvn|const struct refcounted_he *chain|const char *keypv|STRLEN keylen|U32 hash|U32 flags | |
3187 | ||
3188 | Search along a C<refcounted_he> chain for an entry with the key specified | |
3189 | by I<keypv> and I<keylen>. If I<flags> has the C<REFCOUNTED_HE_KEY_UTF8> | |
3190 | bit set, the key octets are interpreted as UTF-8, otherwise they | |
3191 | are interpreted as Latin-1. I<hash> is a precomputed hash of the key | |
3192 | string, or zero if it has not been precomputed. Returns a mortal scalar | |
3193 | representing the value associated with the key, or C<&PL_sv_placeholder> | |
3194 | if there is no value associated with the key. | |
3195 | ||
3196 | =cut | |
3197 | */ | |
3198 | ||
7b0bddfa | 3199 | SV * |
20439bc7 Z |
3200 | Perl_refcounted_he_fetch_pvn(pTHX_ const struct refcounted_he *chain, |
3201 | const char *keypv, STRLEN keylen, U32 hash, U32 flags) | |
7b0bddfa | 3202 | { |
0b2d3faa | 3203 | dVAR; |
20439bc7 Z |
3204 | U8 utf8_flag; |
3205 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PVN; | |
7b0bddfa | 3206 | |
94250aee | 3207 | if (flags & ~(REFCOUNTED_HE_KEY_UTF8|REFCOUNTED_HE_EXISTS)) |
20439bc7 Z |
3208 | Perl_croak(aTHX_ "panic: refcounted_he_fetch_pvn bad flags %"UVxf, |
3209 | (UV)flags); | |
3210 | if (!chain) | |
3211 | return &PL_sv_placeholder; | |
3212 | if (flags & REFCOUNTED_HE_KEY_UTF8) { | |
3213 | /* For searching purposes, canonicalise to Latin-1 where possible. */ | |
3214 | const char *keyend = keypv + keylen, *p; | |
3215 | STRLEN nonascii_count = 0; | |
3216 | for (p = keypv; p != keyend; p++) { | |
e8e5e5b3 KW |
3217 | if (! UTF8_IS_INVARIANT(*p)) { |
3218 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) { | |
20439bc7 | 3219 | goto canonicalised_key; |
e8e5e5b3 | 3220 | } |
20439bc7 | 3221 | nonascii_count++; |
e8e5e5b3 | 3222 | p++; |
20439bc7 | 3223 | } |
cd1d2f8a | 3224 | } |
20439bc7 Z |
3225 | if (nonascii_count) { |
3226 | char *q; | |
3227 | const char *p = keypv, *keyend = keypv + keylen; | |
3228 | keylen -= nonascii_count; | |
3229 | Newx(q, keylen, char); | |
3230 | SAVEFREEPV(q); | |
3231 | keypv = q; | |
3232 | for (; p != keyend; p++, q++) { | |
3233 | U8 c = (U8)*p; | |
e8e5e5b3 KW |
3234 | if (UTF8_IS_INVARIANT(c)) { |
3235 | *q = (char) c; | |
3236 | } | |
3237 | else { | |
3238 | p++; | |
3239 | *q = (char) TWO_BYTE_UTF8_TO_NATIVE(c, *p); | |
3240 | } | |
cd1d2f8a NC |
3241 | } |
3242 | } | |
20439bc7 Z |
3243 | flags &= ~REFCOUNTED_HE_KEY_UTF8; |
3244 | canonicalised_key: ; | |
3245 | } | |
3246 | utf8_flag = (flags & REFCOUNTED_HE_KEY_UTF8) ? HVhek_UTF8 : 0; | |
3247 | if (!hash) | |
3248 | PERL_HASH(hash, keypv, keylen); | |
7b0bddfa | 3249 | |
20439bc7 Z |
3250 | for (; chain; chain = chain->refcounted_he_next) { |
3251 | if ( | |
7b0bddfa | 3252 | #ifdef USE_ITHREADS |
20439bc7 Z |
3253 | hash == chain->refcounted_he_hash && |
3254 | keylen == chain->refcounted_he_keylen && | |
3255 | memEQ(REF_HE_KEY(chain), keypv, keylen) && | |
3256 | utf8_flag == (chain->refcounted_he_data[0] & HVhek_UTF8) | |
7b0bddfa | 3257 | #else |
20439bc7 Z |
3258 | hash == HEK_HASH(chain->refcounted_he_hek) && |
3259 | keylen == (STRLEN)HEK_LEN(chain->refcounted_he_hek) && | |
3260 | memEQ(HEK_KEY(chain->refcounted_he_hek), keypv, keylen) && | |
3261 | utf8_flag == (HEK_FLAGS(chain->refcounted_he_hek) & HVhek_UTF8) | |
7b0bddfa | 3262 | #endif |
ef8156f5 NC |
3263 | ) { |
3264 | if (flags & REFCOUNTED_HE_EXISTS) | |
3265 | return (chain->refcounted_he_data[0] & HVrhek_typemask) | |
3266 | == HVrhek_delete | |
3267 | ? NULL : &PL_sv_yes; | |
3268 | return sv_2mortal(refcounted_he_value(chain)); | |
3269 | } | |
94250aee FC |
3270 | } |
3271 | return flags & REFCOUNTED_HE_EXISTS ? NULL : &PL_sv_placeholder; | |
20439bc7 | 3272 | } |
7b0bddfa | 3273 | |
20439bc7 Z |
3274 | /* |
3275 | =for apidoc m|SV *|refcounted_he_fetch_pv|const struct refcounted_he *chain|const char *key|U32 hash|U32 flags | |
7b0bddfa | 3276 | |
20439bc7 Z |
3277 | Like L</refcounted_he_fetch_pvn>, but takes a nul-terminated string |
3278 | instead of a string/length pair. | |
3279 | ||
3280 | =cut | |
3281 | */ | |
3282 | ||
3283 | SV * | |
3284 | Perl_refcounted_he_fetch_pv(pTHX_ const struct refcounted_he *chain, | |
3285 | const char *key, U32 hash, U32 flags) | |
3286 | { | |
3287 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_PV; | |
3288 | return refcounted_he_fetch_pvn(chain, key, strlen(key), hash, flags); | |
7b0bddfa NC |
3289 | } |
3290 | ||
b3ca2e83 | 3291 | /* |
20439bc7 Z |
3292 | =for apidoc m|SV *|refcounted_he_fetch_sv|const struct refcounted_he *chain|SV *key|U32 hash|U32 flags |
3293 | ||
3294 | Like L</refcounted_he_fetch_pvn>, but takes a Perl scalar instead of a | |
3295 | string/length pair. | |
3296 | ||
3297 | =cut | |
3298 | */ | |
b3ca2e83 | 3299 | |
20439bc7 Z |
3300 | SV * |
3301 | Perl_refcounted_he_fetch_sv(pTHX_ const struct refcounted_he *chain, | |
3302 | SV *key, U32 hash, U32 flags) | |
3303 | { | |
3304 | const char *keypv; | |
3305 | STRLEN keylen; | |
3306 | PERL_ARGS_ASSERT_REFCOUNTED_HE_FETCH_SV; | |
3307 | if (flags & REFCOUNTED_HE_KEY_UTF8) | |
3308 | Perl_croak(aTHX_ "panic: refcounted_he_fetch_sv bad flags %"UVxf, | |
3309 | (UV)flags); | |
3310 | keypv = SvPV_const(key, keylen); | |
3311 | if (SvUTF8(key)) | |
3312 | flags |= REFCOUNTED_HE_KEY_UTF8; | |
3313 | if (!hash && SvIsCOW_shared_hash(key)) | |
3314 | hash = SvSHARED_HASH(key); | |
3315 | return refcounted_he_fetch_pvn(chain, keypv, keylen, hash, flags); | |
3316 | } | |
3317 | ||
3318 | /* | |
3319 | =for apidoc m|struct refcounted_he *|refcounted_he_new_pvn|struct refcounted_he *parent|const char *keypv|STRLEN keylen|U32 hash|SV *value|U32 flags | |
3320 | ||
3321 | Creates a new C<refcounted_he>. This consists of a single key/value | |
3322 | pair and a reference to an existing C<refcounted_he> chain (which may | |
3323 | be empty), and thus forms a longer chain. When using the longer chain, | |
3324 | the new key/value pair takes precedence over any entry for the same key | |
3325 | further along the chain. | |
3326 | ||
3327 | The new key is specified by I<keypv> and I<keylen>. If I<flags> has | |
3328 | the C<REFCOUNTED_HE_KEY_UTF8> bit set, the key octets are interpreted | |
3329 | as UTF-8, otherwise they are interpreted as Latin-1. I<hash> is | |
3330 | a precomputed hash of the key string, or zero if it has not been | |
3331 | precomputed. | |
3332 | ||
3333 | I<value> is the scalar value to store for this key. I<value> is copied | |
3334 | by this function, which thus does not take ownership of any reference | |
3335 | to it, and later changes to the scalar will not be reflected in the | |
3336 | value visible in the C<refcounted_he>. Complex types of scalar will not | |
3337 | be stored with referential integrity, but will be coerced to strings. | |
3338 | I<value> may be either null or C<&PL_sv_placeholder> to indicate that no | |
3339 | value is to be associated with the key; this, as with any non-null value, | |
3340 | takes precedence over the existence of a value for the key further along | |
3341 | the chain. | |
3342 | ||
3343 | I<parent> points to the rest of the C<refcounted_he> chain to be | |
3344 | attached to the new C<refcounted_he>. This function takes ownership | |
3345 | of one reference to I<parent>, and returns one reference to the new | |
3346 | C<refcounted_he>. | |
b3ca2e83 NC |
3347 | |
3348 | =cut | |
3349 | */ | |
3350 | ||
3351 | struct refcounted_he * | |
20439bc7 Z |
3352 | Perl_refcounted_he_new_pvn(pTHX_ struct refcounted_he *parent, |
3353 | const char *keypv, STRLEN keylen, U32 hash, SV *value, U32 flags) | |
3354 | { | |
7a89be66 | 3355 | dVAR; |
b6bbf3fa | 3356 | STRLEN value_len = 0; |
95b63a38 | 3357 | const char *value_p = NULL; |
20439bc7 | 3358 | bool is_pv; |
b6bbf3fa | 3359 | char value_type; |
20439bc7 Z |
3360 | char hekflags; |
3361 | STRLEN key_offset = 1; | |
3362 | struct refcounted_he *he; | |
3363 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PVN; | |
b6bbf3fa | 3364 | |
20439bc7 Z |
3365 | if (!value || value == &PL_sv_placeholder) { |
3366 | value_type = HVrhek_delete; | |
3367 | } else if (SvPOK(value)) { | |
b6bbf3fa NC |
3368 | value_type = HVrhek_PV; |
3369 | } else if (SvIOK(value)) { | |
ad64d0ec | 3370 | value_type = SvUOK((const SV *)value) ? HVrhek_UV : HVrhek_IV; |
b6bbf3fa NC |
3371 | } else if (!SvOK(value)) { |
3372 | value_type = HVrhek_undef; | |
3373 | } else { | |
3374 | value_type = HVrhek_PV; | |
3375 | } | |
20439bc7 Z |
3376 | is_pv = value_type == HVrhek_PV; |
3377 | if (is_pv) { | |
012da8e5 NC |
3378 | /* Do it this way so that the SvUTF8() test is after the SvPV, in case |
3379 | the value is overloaded, and doesn't yet have the UTF-8flag set. */ | |
b6bbf3fa | 3380 | value_p = SvPV_const(value, value_len); |
012da8e5 NC |
3381 | if (SvUTF8(value)) |
3382 | value_type = HVrhek_PV_UTF8; | |
20439bc7 Z |
3383 | key_offset = value_len + 2; |
3384 | } | |
3385 | hekflags = value_type; | |
3386 | ||
3387 | if (flags & REFCOUNTED_HE_KEY_UTF8) { | |
3388 | /* Canonicalise to Latin-1 where possible. */ | |
3389 | const char *keyend = keypv + keylen, *p; | |
3390 | STRLEN nonascii_count = 0; | |
3391 | for (p = keypv; p != keyend; p++) { | |
e8e5e5b3 KW |
3392 | if (! UTF8_IS_INVARIANT(*p)) { |
3393 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, keyend)) { | |
20439bc7 | 3394 | goto canonicalised_key; |
e8e5e5b3 | 3395 | } |
20439bc7 | 3396 | nonascii_count++; |
e8e5e5b3 | 3397 | p++; |
20439bc7 Z |
3398 | } |
3399 | } | |
3400 | if (nonascii_count) { | |
3401 | char *q; | |
3402 | const char *p = keypv, *keyend = keypv + keylen; | |
3403 | keylen -= nonascii_count; | |
3404 | Newx(q, keylen, char); | |
3405 | SAVEFREEPV(q); | |
3406 | keypv = q; | |
3407 | for (; p != keyend; p++, q++) { | |
3408 | U8 c = (U8)*p; | |
e8e5e5b3 KW |
3409 | if (UTF8_IS_INVARIANT(c)) { |
3410 | *q = (char) c; | |
3411 | } | |
3412 | else { | |
3413 | p++; | |
3414 | *q = (char) TWO_BYTE_UTF8_TO_NATIVE(c, *p); | |
3415 | } | |
20439bc7 Z |
3416 | } |
3417 | } | |
3418 | flags &= ~REFCOUNTED_HE_KEY_UTF8; | |
3419 | canonicalised_key: ; | |
b6bbf3fa | 3420 | } |
20439bc7 Z |
3421 | if (flags & REFCOUNTED_HE_KEY_UTF8) |
3422 | hekflags |= HVhek_UTF8; | |
3423 | if (!hash) | |
3424 | PERL_HASH(hash, keypv, keylen); | |
012da8e5 | 3425 | |
0de694c5 | 3426 | #ifdef USE_ITHREADS |
10edeb5d JH |
3427 | he = (struct refcounted_he*) |
3428 | PerlMemShared_malloc(sizeof(struct refcounted_he) - 1 | |
20439bc7 | 3429 | + keylen |
20439bc7 | 3430 | + key_offset); |
0de694c5 NC |
3431 | #else |
3432 | he = (struct refcounted_he*) | |
3433 | PerlMemShared_malloc(sizeof(struct refcounted_he) - 1 | |
3434 | + key_offset); | |
3435 | #endif | |
b3ca2e83 | 3436 | |
71ad1b0c | 3437 | he->refcounted_he_next = parent; |
b6bbf3fa | 3438 | |
012da8e5 | 3439 | if (is_pv) { |
20439bc7 | 3440 | Copy(value_p, he->refcounted_he_data + 1, value_len + 1, char); |
b6bbf3fa | 3441 | he->refcounted_he_val.refcounted_he_u_len = value_len; |
b6bbf3fa | 3442 | } else if (value_type == HVrhek_IV) { |
20439bc7 | 3443 | he->refcounted_he_val.refcounted_he_u_iv = SvIVX(value); |
012da8e5 | 3444 | } else if (value_type == HVrhek_UV) { |
20439bc7 | 3445 | he->refcounted_he_val.refcounted_he_u_uv = SvUVX(value); |
b6bbf3fa NC |
3446 | } |
3447 | ||
cbb1fbea | 3448 | #ifdef USE_ITHREADS |
b6bbf3fa | 3449 | he->refcounted_he_hash = hash; |
20439bc7 Z |
3450 | he->refcounted_he_keylen = keylen; |
3451 | Copy(keypv, he->refcounted_he_data + key_offset, keylen, char); | |
cbb1fbea | 3452 | #else |
20439bc7 | 3453 | he->refcounted_he_hek = share_hek_flags(keypv, keylen, hash, hekflags); |
cbb1fbea | 3454 | #endif |
b6bbf3fa | 3455 | |
20439bc7 | 3456 | he->refcounted_he_data[0] = hekflags; |
b3ca2e83 NC |
3457 | he->refcounted_he_refcnt = 1; |
3458 | ||
3459 | return he; | |
3460 | } | |
3461 | ||
3462 | /* | |
20439bc7 | 3463 | =for apidoc m|struct refcounted_he *|refcounted_he_new_pv|struct refcounted_he *parent|const char *key|U32 hash|SV *value|U32 flags |
b3ca2e83 | 3464 | |
20439bc7 Z |
3465 | Like L</refcounted_he_new_pvn>, but takes a nul-terminated string instead |
3466 | of a string/length pair. | |
3467 | ||
3468 | =cut | |
3469 | */ | |
3470 | ||
3471 | struct refcounted_he * | |
3472 | Perl_refcounted_he_new_pv(pTHX_ struct refcounted_he *parent, | |
3473 | const char *key, U32 hash, SV *value, U32 flags) | |
3474 | { | |
3475 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_PV; | |
3476 | return refcounted_he_new_pvn(parent, key, strlen(key), hash, value, flags); | |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | =for apidoc m|struct refcounted_he *|refcounted_he_new_sv|struct refcounted_he *parent|SV *key|U32 hash|SV *value|U32 flags | |
3481 | ||
3482 | Like L</refcounted_he_new_pvn>, but takes a Perl scalar instead of a | |
3483 | string/length pair. | |
3484 | ||
3485 | =cut | |
3486 | */ | |
3487 | ||
3488 | struct refcounted_he * | |
3489 | Perl_refcounted_he_new_sv(pTHX_ struct refcounted_he *parent, | |
3490 | SV *key, U32 hash, SV *value, U32 flags) | |
3491 | { | |
3492 | const char *keypv; | |
3493 | STRLEN keylen; | |
3494 | PERL_ARGS_ASSERT_REFCOUNTED_HE_NEW_SV; | |
3495 | if (flags & REFCOUNTED_HE_KEY_UTF8) | |
3496 | Perl_croak(aTHX_ "panic: refcounted_he_new_sv bad flags %"UVxf, | |
3497 | (UV)flags); | |
3498 | keypv = SvPV_const(key, keylen); | |
3499 | if (SvUTF8(key)) | |
3500 | flags |= REFCOUNTED_HE_KEY_UTF8; | |
3501 | if (!hash && SvIsCOW_shared_hash(key)) | |
3502 | hash = SvSHARED_HASH(key); | |
3503 | return refcounted_he_new_pvn(parent, keypv, keylen, hash, value, flags); | |
3504 | } | |
3505 | ||
3506 | /* | |
3507 | =for apidoc m|void|refcounted_he_free|struct refcounted_he *he | |
3508 | ||
3509 | Decrements the reference count of a C<refcounted_he> by one. If the | |
3510 | reference count reaches zero the structure's memory is freed, which | |
3511 | (recursively) causes a reduction of its parent C<refcounted_he>'s | |
3512 | reference count. It is safe to pass a null pointer to this function: | |
3513 | no action occurs in this case. | |
b3ca2e83 NC |
3514 | |
3515 | =cut | |
3516 | */ | |
3517 | ||
3518 | void | |
3519 | Perl_refcounted_he_free(pTHX_ struct refcounted_he *he) { | |
20b7effb | 3520 | #ifdef USE_ITHREADS |
53d44271 | 3521 | dVAR; |
20b7effb | 3522 | #endif |
57ca3b03 AL |
3523 | PERL_UNUSED_CONTEXT; |
3524 | ||
b3ca2e83 NC |
3525 | while (he) { |
3526 | struct refcounted_he *copy; | |
cbb1fbea | 3527 | U32 new_count; |
b3ca2e83 | 3528 | |
cbb1fbea NC |
3529 | HINTS_REFCNT_LOCK; |
3530 | new_count = --he->refcounted_he_refcnt; | |
3531 | HINTS_REFCNT_UNLOCK; | |
3532 | ||
3533 | if (new_count) { | |
b3ca2e83 | 3534 | return; |
cbb1fbea | 3535 | } |
b3ca2e83 | 3536 | |
b6bbf3fa | 3537 | #ifndef USE_ITHREADS |
71ad1b0c | 3538 | unshare_hek_or_pvn (he->refcounted_he_hek, 0, 0, 0); |
cbb1fbea | 3539 | #endif |
b3ca2e83 | 3540 | copy = he; |
71ad1b0c | 3541 | he = he->refcounted_he_next; |
b6bbf3fa | 3542 | PerlMemShared_free(copy); |
b3ca2e83 NC |
3543 | } |
3544 | } | |
3545 | ||
20439bc7 Z |
3546 | /* |
3547 | =for apidoc m|struct refcounted_he *|refcounted_he_inc|struct refcounted_he *he | |
3548 | ||
3549 | Increment the reference count of a C<refcounted_he>. The pointer to the | |
3550 | C<refcounted_he> is also returned. It is safe to pass a null pointer | |
3551 | to this function: no action occurs and a null pointer is returned. | |
3552 | ||
3553 | =cut | |
3554 | */ | |
3555 | ||
3556 | struct refcounted_he * | |
3557 | Perl_refcounted_he_inc(pTHX_ struct refcounted_he *he) | |
3558 | { | |
20b7effb | 3559 | #ifdef USE_ITHREADS |
09ddd873 | 3560 | dVAR; |
20b7effb | 3561 | #endif |
dc3bf405 | 3562 | PERL_UNUSED_CONTEXT; |
20439bc7 Z |
3563 | if (he) { |
3564 | HINTS_REFCNT_LOCK; | |
3565 | he->refcounted_he_refcnt++; | |
3566 | HINTS_REFCNT_UNLOCK; | |
3567 | } | |
3568 | return he; | |
3569 | } | |
3570 | ||
8375c93e | 3571 | /* |
aebc0cbe | 3572 | =for apidoc cop_fetch_label |
8375c93e RU |
3573 | |
3574 | Returns the label attached to a cop. | |
3575 | The flags pointer may be set to C<SVf_UTF8> or 0. | |
3576 | ||
3577 | =cut | |
3578 | */ | |
3579 | ||
47550813 NC |
3580 | /* pp_entereval is aware that labels are stored with a key ':' at the top of |
3581 | the linked list. */ | |
dca6062a | 3582 | const char * |
aebc0cbe | 3583 | Perl_cop_fetch_label(pTHX_ COP *const cop, STRLEN *len, U32 *flags) { |
d6747b7a NC |
3584 | struct refcounted_he *const chain = cop->cop_hints_hash; |
3585 | ||
aebc0cbe | 3586 | PERL_ARGS_ASSERT_COP_FETCH_LABEL; |
dc3bf405 | 3587 | PERL_UNUSED_CONTEXT; |
d6747b7a | 3588 | |
dca6062a NC |
3589 | if (!chain) |
3590 | return NULL; | |
3591 | #ifdef USE_ITHREADS | |
3592 | if (chain->refcounted_he_keylen != 1) | |
3593 | return NULL; | |
3594 | if (*REF_HE_KEY(chain) != ':') | |
3595 | return NULL; | |
3596 | #else | |
3597 | if ((STRLEN)HEK_LEN(chain->refcounted_he_hek) != 1) | |
3598 | return NULL; | |
3599 | if (*HEK_KEY(chain->refcounted_he_hek) != ':') | |
3600 | return NULL; | |
3601 | #endif | |
012da8e5 NC |
3602 | /* Stop anyone trying to really mess us up by adding their own value for |
3603 | ':' into %^H */ | |
3604 | if ((chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV | |
3605 | && (chain->refcounted_he_data[0] & HVrhek_typemask) != HVrhek_PV_UTF8) | |
3606 | return NULL; | |
3607 | ||
dca6062a NC |
3608 | if (len) |
3609 | *len = chain->refcounted_he_val.refcounted_he_u_len; | |
3610 | if (flags) { | |
3611 | *flags = ((chain->refcounted_he_data[0] & HVrhek_typemask) | |
3612 | == HVrhek_PV_UTF8) ? SVf_UTF8 : 0; | |
3613 | } | |
3614 | return chain->refcounted_he_data + 1; | |
3615 | } | |
3616 | ||
8375c93e | 3617 | /* |
aebc0cbe | 3618 | =for apidoc cop_store_label |
8375c93e | 3619 | |
72d33970 FC |
3620 | Save a label into a C<cop_hints_hash>. |
3621 | You need to set flags to C<SVf_UTF8> | |
8375c93e RU |
3622 | for a utf-8 label. |
3623 | ||
3624 | =cut | |
3625 | */ | |
3626 | ||
a77ac40c | 3627 | void |
aebc0cbe | 3628 | Perl_cop_store_label(pTHX_ COP *const cop, const char *label, STRLEN len, |
a77ac40c | 3629 | U32 flags) |
012da8e5 | 3630 | { |
20439bc7 | 3631 | SV *labelsv; |
aebc0cbe | 3632 | PERL_ARGS_ASSERT_COP_STORE_LABEL; |
547bb267 | 3633 | |
a77ac40c | 3634 | if (flags & ~(SVf_UTF8)) |
aebc0cbe | 3635 | Perl_croak(aTHX_ "panic: cop_store_label illegal flag bits 0x%" UVxf, |
a77ac40c | 3636 | (UV)flags); |
a3179684 | 3637 | labelsv = newSVpvn_flags(label, len, SVs_TEMP); |
20439bc7 Z |
3638 | if (flags & SVf_UTF8) |
3639 | SvUTF8_on(labelsv); | |
a77ac40c | 3640 | cop->cop_hints_hash |
20439bc7 | 3641 | = refcounted_he_new_pvs(cop->cop_hints_hash, ":", labelsv, 0); |
012da8e5 NC |
3642 | } |
3643 | ||
b3ca2e83 | 3644 | /* |
ecae49c0 NC |
3645 | =for apidoc hv_assert |
3646 | ||
3647 | Check that a hash is in an internally consistent state. | |
3648 | ||
3649 | =cut | |
3650 | */ | |
3651 | ||
943795c2 NC |
3652 | #ifdef DEBUGGING |
3653 | ||
ecae49c0 NC |
3654 | void |
3655 | Perl_hv_assert(pTHX_ HV *hv) | |
3656 | { | |
57ca3b03 AL |
3657 | dVAR; |
3658 | HE* entry; | |
3659 | int withflags = 0; | |
3660 | int placeholders = 0; | |
3661 | int real = 0; | |
3662 | int bad = 0; | |
3663 | const I32 riter = HvRITER_get(hv); | |
3664 | HE *eiter = HvEITER_get(hv); | |
3665 | ||
7918f24d NC |
3666 | PERL_ARGS_ASSERT_HV_ASSERT; |
3667 | ||
57ca3b03 AL |
3668 | (void)hv_iterinit(hv); |
3669 | ||
3670 | while ((entry = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS))) { | |
3671 | /* sanity check the values */ | |
3672 | if (HeVAL(entry) == &PL_sv_placeholder) | |
3673 | placeholders++; | |
3674 | else | |
3675 | real++; | |
3676 | /* sanity check the keys */ | |
3677 | if (HeSVKEY(entry)) { | |
6f207bd3 | 3678 | NOOP; /* Don't know what to check on SV keys. */ |
57ca3b03 AL |
3679 | } else if (HeKUTF8(entry)) { |
3680 | withflags++; | |
3681 | if (HeKWASUTF8(entry)) { | |
3682 | PerlIO_printf(Perl_debug_log, | |
d2a455e7 | 3683 | "hash key has both WASUTF8 and UTF8: '%.*s'\n", |
57ca3b03 AL |
3684 | (int) HeKLEN(entry), HeKEY(entry)); |
3685 | bad = 1; | |
3686 | } | |
3687 | } else if (HeKWASUTF8(entry)) | |
3688 | withflags++; | |
3689 | } | |
ad64d0ec | 3690 | if (!SvTIED_mg((const SV *)hv, PERL_MAGIC_tied)) { |
57ca3b03 AL |
3691 | static const char bad_count[] = "Count %d %s(s), but hash reports %d\n"; |
3692 | const int nhashkeys = HvUSEDKEYS(hv); | |
3693 | const int nhashplaceholders = HvPLACEHOLDERS_get(hv); | |
3694 | ||
3695 | if (nhashkeys != real) { | |
3696 | PerlIO_printf(Perl_debug_log, bad_count, real, "keys", nhashkeys ); | |
3697 | bad = 1; | |
3698 | } | |
3699 | if (nhashplaceholders != placeholders) { | |
3700 | PerlIO_printf(Perl_debug_log, bad_count, placeholders, "placeholder", nhashplaceholders ); | |
3701 | bad = 1; | |
3702 | } | |
3703 | } | |
3704 | if (withflags && ! HvHASKFLAGS(hv)) { | |
3705 | PerlIO_printf(Perl_debug_log, | |
3706 | "Hash has HASKFLAGS off but I count %d key(s) with flags\n", | |
3707 | withflags); | |
3708 | bad = 1; | |
3709 | } | |
3710 | if (bad) { | |
ad64d0ec | 3711 | sv_dump(MUTABLE_SV(hv)); |
57ca3b03 AL |
3712 | } |
3713 | HvRITER_set(hv, riter); /* Restore hash iterator state */ | |
3714 | HvEITER_set(hv, eiter); | |
ecae49c0 | 3715 | } |
af3babe4 | 3716 | |
943795c2 NC |
3717 | #endif |
3718 | ||
af3babe4 NC |
3719 | /* |
3720 | * Local variables: | |
3721 | * c-indentation-style: bsd | |
3722 | * c-basic-offset: 4 | |
14d04a33 | 3723 | * indent-tabs-mode: nil |
af3babe4 NC |
3724 | * End: |
3725 | * | |
14d04a33 | 3726 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 3727 | */ |