This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add perl589delta.pod to pod.lst, and run pod/buildtoc --build-all.
[perl5.git] / sv.c
CommitLineData
a0d0e21e 1/* sv.c
79072805 2 *
1129b882 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
83706693
RGS
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
5 * and others
79072805
LW
6 *
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
9 *
4ac71550
TC
10 */
11
12/*
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
14 * --Pippin
15 *
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
17 */
18
19/*
645c22ef
DM
20 *
21 *
5e045b90
AMS
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
27 * in the pp*.c files.
79072805
LW
28 */
29
30#include "EXTERN.h"
864dbfa3 31#define PERL_IN_SV_C
79072805 32#include "perl.h"
d2f185dc 33#include "regcomp.h"
79072805 34
51371543 35#define FCALL *f
2c5424a7 36
2f8ed50e
OS
37#ifdef __Lynx__
38/* Missing proto on LynxOS */
39 char *gconvert(double, int, int, char *);
40#endif
41
e23c8137 42#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 43/* if adding more checks watch out for the following tests:
e23c8137
JH
44 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
45 * lib/utf8.t lib/Unicode/Collate/t/index.t
46 * --jhi
47 */
6f207bd3 48# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
49 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
50 assert((cache)[2] <= (cache)[3]); \
51 assert((cache)[3] <= (cache)[1]);} \
52 } STMT_END
e23c8137 53#else
6f207bd3 54# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
55#endif
56
f8c7b90f 57#ifdef PERL_OLD_COPY_ON_WRITE
765f542d 58#define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
607fa7f2 59#define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
b5ccf5f2 60/* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
765f542d 61 on-write. */
765f542d 62#endif
645c22ef
DM
63
64/* ============================================================================
65
66=head1 Allocation and deallocation of SVs.
67
d2a0f284
JC
68An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
69sv, av, hv...) contains type and reference count information, and for
70many types, a pointer to the body (struct xrv, xpv, xpviv...), which
71contains fields specific to each type. Some types store all they need
72in the head, so don't have a body.
73
74In all but the most memory-paranoid configuations (ex: PURIFY), heads
75and bodies are allocated out of arenas, which by default are
76approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
77Sv-bodies are allocated by their sv-type, guaranteeing size
78consistency needed to allocate safely from arrays.
79
d2a0f284
JC
80For SV-heads, the first slot in each arena is reserved, and holds a
81link to the next arena, some flags, and a note of the number of slots.
82Snaked through each arena chain is a linked list of free items; when
83this becomes empty, an extra arena is allocated and divided up into N
84items which are threaded into the free list.
85
86SV-bodies are similar, but they use arena-sets by default, which
87separate the link and info from the arena itself, and reclaim the 1st
88slot in the arena. SV-bodies are further described later.
645c22ef
DM
89
90The following global variables are associated with arenas:
91
92 PL_sv_arenaroot pointer to list of SV arenas
93 PL_sv_root pointer to list of free SV structures
94
d2a0f284
JC
95 PL_body_arenas head of linked-list of body arenas
96 PL_body_roots[] array of pointers to list of free bodies of svtype
97 arrays are indexed by the svtype needed
93e68bfb 98
d2a0f284
JC
99A few special SV heads are not allocated from an arena, but are
100instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
101The size of arenas can be changed from the default by setting
102PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
103
104The SV arena serves the secondary purpose of allowing still-live SVs
105to be located and destroyed during final cleanup.
106
107At the lowest level, the macros new_SV() and del_SV() grab and free
108an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
109to return the SV to the free list with error checking.) new_SV() calls
110more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
111SVs in the free list have their SvTYPE field set to all ones.
112
ff276b08 113At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 114perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 115start of the interpreter.
645c22ef 116
645c22ef
DM
117The function visit() scans the SV arenas list, and calls a specified
118function for each SV it finds which is still live - ie which has an SvTYPE
119other than all 1's, and a non-zero SvREFCNT. visit() is used by the
120following functions (specified as [function that calls visit()] / [function
121called by visit() for each SV]):
122
123 sv_report_used() / do_report_used()
f2524eef 124 dump all remaining SVs (debugging aid)
645c22ef
DM
125
126 sv_clean_objs() / do_clean_objs(),do_clean_named_objs()
127 Attempt to free all objects pointed to by RVs,
128 and, unless DISABLE_DESTRUCTOR_KLUDGE is defined,
129 try to do the same for all objects indirectly
130 referenced by typeglobs too. Called once from
131 perl_destruct(), prior to calling sv_clean_all()
132 below.
133
134 sv_clean_all() / do_clean_all()
135 SvREFCNT_dec(sv) each remaining SV, possibly
136 triggering an sv_free(). It also sets the
137 SVf_BREAK flag on the SV to indicate that the
138 refcnt has been artificially lowered, and thus
139 stopping sv_free() from giving spurious warnings
140 about SVs which unexpectedly have a refcnt
141 of zero. called repeatedly from perl_destruct()
142 until there are no SVs left.
143
93e68bfb 144=head2 Arena allocator API Summary
645c22ef
DM
145
146Private API to rest of sv.c
147
148 new_SV(), del_SV(),
149
150 new_XIV(), del_XIV(),
151 new_XNV(), del_XNV(),
152 etc
153
154Public API:
155
8cf8f3d1 156 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 157
645c22ef
DM
158=cut
159
3e8320cc 160 * ========================================================================= */
645c22ef 161
4561caa4
CS
162/*
163 * "A time to plant, and a time to uproot what was planted..."
164 */
165
77354fb4 166void
de37a194 167Perl_offer_nice_chunk(pTHX_ void *const chunk, const U32 chunk_size)
77354fb4 168{
97aff369 169 dVAR;
77354fb4
NC
170 void *new_chunk;
171 U32 new_chunk_size;
7918f24d
NC
172
173 PERL_ARGS_ASSERT_OFFER_NICE_CHUNK;
174
77354fb4
NC
175 new_chunk = (void *)(chunk);
176 new_chunk_size = (chunk_size);
177 if (new_chunk_size > PL_nice_chunk_size) {
178 Safefree(PL_nice_chunk);
179 PL_nice_chunk = (char *) new_chunk;
180 PL_nice_chunk_size = new_chunk_size;
181 } else {
182 Safefree(chunk);
183 }
77354fb4 184}
cac9b346 185
d7a2c63c
MHM
186#ifdef PERL_MEM_LOG
187# define MEM_LOG_NEW_SV(sv, file, line, func) \
188 Perl_mem_log_new_sv(sv, file, line, func)
189# define MEM_LOG_DEL_SV(sv, file, line, func) \
190 Perl_mem_log_del_sv(sv, file, line, func)
191#else
192# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
193# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
194#endif
195
fd0854ff 196#ifdef DEBUG_LEAKING_SCALARS
22162ca8 197# define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
d7a2c63c
MHM
198# define DEBUG_SV_SERIAL(sv) \
199 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
200 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
201#else
202# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 203# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
204#endif
205
48614a46
NC
206#ifdef PERL_POISON
207# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 208# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
209/* Whilst I'd love to do this, it seems that things like to check on
210 unreferenced scalars
7e337ee0 211# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 212*/
7e337ee0
JH
213# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
214 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
215#else
216# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 217# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
218# define POSION_SV_HEAD(sv)
219#endif
220
990198f0
DM
221/* Mark an SV head as unused, and add to free list.
222 *
223 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
224 * its refcount artificially decremented during global destruction, so
225 * there may be dangling pointers to it. The last thing we want in that
226 * case is for it to be reused. */
227
053fc874
GS
228#define plant_SV(p) \
229 STMT_START { \
990198f0 230 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
231 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
232 DEBUG_SV_SERIAL(p); \
fd0854ff 233 FREE_SV_DEBUG_FILE(p); \
48614a46 234 POSION_SV_HEAD(p); \
053fc874 235 SvFLAGS(p) = SVTYPEMASK; \
990198f0 236 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 237 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
238 PL_sv_root = (p); \
239 } \
053fc874
GS
240 --PL_sv_count; \
241 } STMT_END
a0d0e21e 242
053fc874
GS
243#define uproot_SV(p) \
244 STMT_START { \
245 (p) = PL_sv_root; \
daba3364 246 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
247 ++PL_sv_count; \
248 } STMT_END
249
645c22ef 250
cac9b346
NC
251/* make some more SVs by adding another arena */
252
cac9b346
NC
253STATIC SV*
254S_more_sv(pTHX)
255{
97aff369 256 dVAR;
cac9b346
NC
257 SV* sv;
258
259 if (PL_nice_chunk) {
260 sv_add_arena(PL_nice_chunk, PL_nice_chunk_size, 0);
bd61b366 261 PL_nice_chunk = NULL;
cac9b346
NC
262 PL_nice_chunk_size = 0;
263 }
264 else {
265 char *chunk; /* must use New here to match call to */
d2a0f284 266 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
2e7ed132 267 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
268 }
269 uproot_SV(sv);
270 return sv;
271}
272
645c22ef
DM
273/* new_SV(): return a new, empty SV head */
274
eba0f806
DM
275#ifdef DEBUG_LEAKING_SCALARS
276/* provide a real function for a debugger to play with */
277STATIC SV*
d7a2c63c 278S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
279{
280 SV* sv;
281
eba0f806
DM
282 if (PL_sv_root)
283 uproot_SV(sv);
284 else
cac9b346 285 sv = S_more_sv(aTHX);
eba0f806
DM
286 SvANY(sv) = 0;
287 SvREFCNT(sv) = 1;
288 SvFLAGS(sv) = 0;
fd0854ff 289 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
290 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
291 ? PL_parser->copline
292 : PL_curcop
f24aceb1
DM
293 ? CopLINE(PL_curcop)
294 : 0
e385c3bf 295 );
fd0854ff
DM
296 sv->sv_debug_inpad = 0;
297 sv->sv_debug_cloned = 0;
fd0854ff 298 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
299
300 sv->sv_debug_serial = PL_sv_serial++;
301
302 MEM_LOG_NEW_SV(sv, file, line, func);
303 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
304 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
305
eba0f806
DM
306 return sv;
307}
d7a2c63c 308# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
309
310#else
311# define new_SV(p) \
053fc874 312 STMT_START { \
053fc874
GS
313 if (PL_sv_root) \
314 uproot_SV(p); \
315 else \
cac9b346 316 (p) = S_more_sv(aTHX); \
053fc874
GS
317 SvANY(p) = 0; \
318 SvREFCNT(p) = 1; \
319 SvFLAGS(p) = 0; \
d7a2c63c 320 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 321 } STMT_END
eba0f806 322#endif
463ee0b2 323
645c22ef
DM
324
325/* del_SV(): return an empty SV head to the free list */
326
a0d0e21e 327#ifdef DEBUGGING
4561caa4 328
053fc874
GS
329#define del_SV(p) \
330 STMT_START { \
aea4f609 331 if (DEBUG_D_TEST) \
053fc874
GS
332 del_sv(p); \
333 else \
334 plant_SV(p); \
053fc874 335 } STMT_END
a0d0e21e 336
76e3520e 337STATIC void
cea2e8a9 338S_del_sv(pTHX_ SV *p)
463ee0b2 339{
97aff369 340 dVAR;
7918f24d
NC
341
342 PERL_ARGS_ASSERT_DEL_SV;
343
aea4f609 344 if (DEBUG_D_TEST) {
4633a7c4 345 SV* sva;
a3b680e6 346 bool ok = 0;
daba3364 347 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
348 const SV * const sv = sva + 1;
349 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 350 if (p >= sv && p < svend) {
a0d0e21e 351 ok = 1;
c0ff570e
NC
352 break;
353 }
a0d0e21e
LW
354 }
355 if (!ok) {
0453d815 356 if (ckWARN_d(WARN_INTERNAL))
9014280d 357 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
472d47bc
SB
358 "Attempt to free non-arena SV: 0x%"UVxf
359 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
360 return;
361 }
362 }
4561caa4 363 plant_SV(p);
463ee0b2 364}
a0d0e21e 365
4561caa4
CS
366#else /* ! DEBUGGING */
367
368#define del_SV(p) plant_SV(p)
369
370#endif /* DEBUGGING */
463ee0b2 371
645c22ef
DM
372
373/*
ccfc67b7
JH
374=head1 SV Manipulation Functions
375
645c22ef
DM
376=for apidoc sv_add_arena
377
378Given a chunk of memory, link it to the head of the list of arenas,
379and split it into a list of free SVs.
380
381=cut
382*/
383
d2bd4e7f
NC
384static void
385S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 386{
97aff369 387 dVAR;
daba3364 388 SV *const sva = MUTABLE_SV(ptr);
463ee0b2
LW
389 register SV* sv;
390 register SV* svend;
4633a7c4 391
7918f24d
NC
392 PERL_ARGS_ASSERT_SV_ADD_ARENA;
393
4633a7c4 394 /* The first SV in an arena isn't an SV. */
3280af22 395 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
396 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
397 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
398
3280af22
NIS
399 PL_sv_arenaroot = sva;
400 PL_sv_root = sva + 1;
4633a7c4
LW
401
402 svend = &sva[SvREFCNT(sva) - 1];
403 sv = sva + 1;
463ee0b2 404 while (sv < svend) {
3eef1deb 405 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 406#ifdef DEBUGGING
978b032e 407 SvREFCNT(sv) = 0;
03e36789 408#endif
4b69cbe3 409 /* Must always set typemask because it's always checked in on cleanup
03e36789 410 when the arenas are walked looking for objects. */
8990e307 411 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
412 sv++;
413 }
3eef1deb 414 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
415#ifdef DEBUGGING
416 SvREFCNT(sv) = 0;
417#endif
4633a7c4
LW
418 SvFLAGS(sv) = SVTYPEMASK;
419}
420
055972dc
DM
421/* visit(): call the named function for each non-free SV in the arenas
422 * whose flags field matches the flags/mask args. */
645c22ef 423
5226ed68 424STATIC I32
de37a194 425S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 426{
97aff369 427 dVAR;
4633a7c4 428 SV* sva;
5226ed68 429 I32 visited = 0;
8990e307 430
7918f24d
NC
431 PERL_ARGS_ASSERT_VISIT;
432
daba3364 433 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0 434 register const SV * const svend = &sva[SvREFCNT(sva)];
a3b680e6 435 register SV* sv;
4561caa4 436 for (sv = sva + 1; sv < svend; ++sv) {
055972dc
DM
437 if (SvTYPE(sv) != SVTYPEMASK
438 && (sv->sv_flags & mask) == flags
439 && SvREFCNT(sv))
440 {
acfe0abc 441 (FCALL)(aTHX_ sv);
5226ed68
JH
442 ++visited;
443 }
8990e307
LW
444 }
445 }
5226ed68 446 return visited;
8990e307
LW
447}
448
758a08c3
JH
449#ifdef DEBUGGING
450
645c22ef
DM
451/* called by sv_report_used() for each live SV */
452
453static void
5fa45a31 454do_report_used(pTHX_ SV *const sv)
645c22ef
DM
455{
456 if (SvTYPE(sv) != SVTYPEMASK) {
457 PerlIO_printf(Perl_debug_log, "****\n");
458 sv_dump(sv);
459 }
460}
758a08c3 461#endif
645c22ef
DM
462
463/*
464=for apidoc sv_report_used
465
466Dump the contents of all SVs not yet freed. (Debugging aid).
467
468=cut
469*/
470
8990e307 471void
864dbfa3 472Perl_sv_report_used(pTHX)
4561caa4 473{
ff270d3a 474#ifdef DEBUGGING
055972dc 475 visit(do_report_used, 0, 0);
96a5add6
AL
476#else
477 PERL_UNUSED_CONTEXT;
ff270d3a 478#endif
4561caa4
CS
479}
480
645c22ef
DM
481/* called by sv_clean_objs() for each live SV */
482
483static void
de37a194 484do_clean_objs(pTHX_ SV *const ref)
645c22ef 485{
97aff369 486 dVAR;
ea724faa
NC
487 assert (SvROK(ref));
488 {
823a54a3
AL
489 SV * const target = SvRV(ref);
490 if (SvOBJECT(target)) {
491 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
492 if (SvWEAKREF(ref)) {
493 sv_del_backref(target, ref);
494 SvWEAKREF_off(ref);
495 SvRV_set(ref, NULL);
496 } else {
497 SvROK_off(ref);
498 SvRV_set(ref, NULL);
499 SvREFCNT_dec(target);
500 }
645c22ef
DM
501 }
502 }
503
504 /* XXX Might want to check arrays, etc. */
505}
506
507/* called by sv_clean_objs() for each live SV */
508
509#ifndef DISABLE_DESTRUCTOR_KLUDGE
510static void
f30de749 511do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 512{
97aff369 513 dVAR;
ea724faa 514 assert(SvTYPE(sv) == SVt_PVGV);
d011219a
NC
515 assert(isGV_with_GP(sv));
516 if (GvGP(sv)) {
c69033f2
NC
517 if ((
518#ifdef PERL_DONT_CREATE_GVSV
519 GvSV(sv) &&
520#endif
521 SvOBJECT(GvSV(sv))) ||
645c22ef
DM
522 (GvAV(sv) && SvOBJECT(GvAV(sv))) ||
523 (GvHV(sv) && SvOBJECT(GvHV(sv))) ||
9c12f1e5
RGS
524 /* In certain rare cases GvIOp(sv) can be NULL, which would make SvOBJECT(GvIO(sv)) dereference NULL. */
525 (GvIO(sv) ? (SvFLAGS(GvIOp(sv)) & SVs_OBJECT) : 0) ||
645c22ef
DM
526 (GvCV(sv) && SvOBJECT(GvCV(sv))) )
527 {
528 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning named glob object:\n "), sv_dump(sv)));
ec5f3c78 529 SvFLAGS(sv) |= SVf_BREAK;
645c22ef
DM
530 SvREFCNT_dec(sv);
531 }
532 }
533}
534#endif
535
536/*
537=for apidoc sv_clean_objs
538
539Attempt to destroy all objects not yet freed
540
541=cut
542*/
543
4561caa4 544void
864dbfa3 545Perl_sv_clean_objs(pTHX)
4561caa4 546{
97aff369 547 dVAR;
3280af22 548 PL_in_clean_objs = TRUE;
055972dc 549 visit(do_clean_objs, SVf_ROK, SVf_ROK);
4561caa4 550#ifndef DISABLE_DESTRUCTOR_KLUDGE
2d0f3c12 551 /* some barnacles may yet remain, clinging to typeglobs */
d011219a 552 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4561caa4 553#endif
3280af22 554 PL_in_clean_objs = FALSE;
4561caa4
CS
555}
556
645c22ef
DM
557/* called by sv_clean_all() for each live SV */
558
559static void
de37a194 560do_clean_all(pTHX_ SV *const sv)
645c22ef 561{
97aff369 562 dVAR;
daba3364 563 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 564 /* don't clean pid table and strtab */
d17ea597 565 return;
cddfcddc 566 }
645c22ef
DM
567 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
568 SvFLAGS(sv) |= SVf_BREAK;
569 SvREFCNT_dec(sv);
570}
571
572/*
573=for apidoc sv_clean_all
574
575Decrement the refcnt of each remaining SV, possibly triggering a
576cleanup. This function may have to be called multiple times to free
ff276b08 577SVs which are in complex self-referential hierarchies.
645c22ef
DM
578
579=cut
580*/
581
5226ed68 582I32
864dbfa3 583Perl_sv_clean_all(pTHX)
8990e307 584{
97aff369 585 dVAR;
5226ed68 586 I32 cleaned;
3280af22 587 PL_in_clean_all = TRUE;
055972dc 588 cleaned = visit(do_clean_all, 0,0);
3280af22 589 PL_in_clean_all = FALSE;
5226ed68 590 return cleaned;
8990e307 591}
463ee0b2 592
5e258f8c
JC
593/*
594 ARENASETS: a meta-arena implementation which separates arena-info
595 into struct arena_set, which contains an array of struct
596 arena_descs, each holding info for a single arena. By separating
597 the meta-info from the arena, we recover the 1st slot, formerly
598 borrowed for list management. The arena_set is about the size of an
39244528 599 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
600
601 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
602 memory in the last arena-set (1/2 on average). In trade, we get
603 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 604 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
605 small arenas for large, rare body types, by changing array* fields
606 in body_details_by_type[] below.
5e258f8c 607*/
5e258f8c 608struct arena_desc {
398c677b
NC
609 char *arena; /* the raw storage, allocated aligned */
610 size_t size; /* its size ~4k typ */
0a848332 611 U32 misc; /* type, and in future other things. */
5e258f8c
JC
612};
613
e6148039
NC
614struct arena_set;
615
616/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 617 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
618 therefore likely to be 1 aligned memory page. */
619
620#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
621 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
622
623struct arena_set {
624 struct arena_set* next;
0a848332
NC
625 unsigned int set_size; /* ie ARENAS_PER_SET */
626 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
627 struct arena_desc set[ARENAS_PER_SET];
628};
629
645c22ef
DM
630/*
631=for apidoc sv_free_arenas
632
633Deallocate the memory used by all arenas. Note that all the individual SV
634heads and bodies within the arenas must already have been freed.
635
636=cut
637*/
4633a7c4 638void
864dbfa3 639Perl_sv_free_arenas(pTHX)
4633a7c4 640{
97aff369 641 dVAR;
4633a7c4
LW
642 SV* sva;
643 SV* svanext;
0a848332 644 unsigned int i;
4633a7c4
LW
645
646 /* Free arenas here, but be careful about fake ones. (We assume
647 contiguity of the fake ones with the corresponding real ones.) */
648
3280af22 649 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 650 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 651 while (svanext && SvFAKE(svanext))
daba3364 652 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
653
654 if (!SvFAKE(sva))
1df70142 655 Safefree(sva);
4633a7c4 656 }
93e68bfb 657
5e258f8c 658 {
0a848332
NC
659 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
660
661 while (aroot) {
662 struct arena_set *current = aroot;
663 i = aroot->curr;
664 while (i--) {
5e258f8c
JC
665 assert(aroot->set[i].arena);
666 Safefree(aroot->set[i].arena);
667 }
0a848332
NC
668 aroot = aroot->next;
669 Safefree(current);
5e258f8c
JC
670 }
671 }
dc8220bf 672 PL_body_arenas = 0;
fdda85ca 673
0a848332
NC
674 i = PERL_ARENA_ROOTS_SIZE;
675 while (i--)
93e68bfb 676 PL_body_roots[i] = 0;
93e68bfb 677
43c5f42d 678 Safefree(PL_nice_chunk);
bd61b366 679 PL_nice_chunk = NULL;
3280af22
NIS
680 PL_nice_chunk_size = 0;
681 PL_sv_arenaroot = 0;
682 PL_sv_root = 0;
4633a7c4
LW
683}
684
bd81e77b
NC
685/*
686 Here are mid-level routines that manage the allocation of bodies out
687 of the various arenas. There are 5 kinds of arenas:
29489e7c 688
bd81e77b
NC
689 1. SV-head arenas, which are discussed and handled above
690 2. regular body arenas
691 3. arenas for reduced-size bodies
692 4. Hash-Entry arenas
693 5. pte arenas (thread related)
29489e7c 694
bd81e77b
NC
695 Arena types 2 & 3 are chained by body-type off an array of
696 arena-root pointers, which is indexed by svtype. Some of the
697 larger/less used body types are malloced singly, since a large
698 unused block of them is wasteful. Also, several svtypes dont have
699 bodies; the data fits into the sv-head itself. The arena-root
700 pointer thus has a few unused root-pointers (which may be hijacked
701 later for arena types 4,5)
29489e7c 702
bd81e77b
NC
703 3 differs from 2 as an optimization; some body types have several
704 unused fields in the front of the structure (which are kept in-place
705 for consistency). These bodies can be allocated in smaller chunks,
706 because the leading fields arent accessed. Pointers to such bodies
707 are decremented to point at the unused 'ghost' memory, knowing that
708 the pointers are used with offsets to the real memory.
29489e7c 709
bd81e77b
NC
710 HE, HEK arenas are managed separately, with separate code, but may
711 be merge-able later..
712
713 PTE arenas are not sv-bodies, but they share these mid-level
714 mechanics, so are considered here. The new mid-level mechanics rely
715 on the sv_type of the body being allocated, so we just reserve one
716 of the unused body-slots for PTEs, then use it in those (2) PTE
717 contexts below (line ~10k)
718*/
719
bd26d9a3 720/* get_arena(size): this creates custom-sized arenas
5e258f8c
JC
721 TBD: export properly for hv.c: S_more_he().
722*/
723void*
de37a194 724Perl_get_arena(pTHX_ const size_t arena_size, const U32 misc)
5e258f8c 725{
7a89be66 726 dVAR;
5e258f8c 727 struct arena_desc* adesc;
39244528 728 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
0a848332 729 unsigned int curr;
5e258f8c 730
476a1e16
JC
731 /* shouldnt need this
732 if (!arena_size) arena_size = PERL_ARENA_SIZE;
733 */
5e258f8c
JC
734
735 /* may need new arena-set to hold new arena */
39244528
NC
736 if (!aroot || aroot->curr >= aroot->set_size) {
737 struct arena_set *newroot;
5e258f8c
JC
738 Newxz(newroot, 1, struct arena_set);
739 newroot->set_size = ARENAS_PER_SET;
39244528
NC
740 newroot->next = aroot;
741 aroot = newroot;
742 PL_body_arenas = (void *) newroot;
52944de8 743 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
5e258f8c
JC
744 }
745
746 /* ok, now have arena-set with at least 1 empty/available arena-desc */
39244528
NC
747 curr = aroot->curr++;
748 adesc = &(aroot->set[curr]);
5e258f8c
JC
749 assert(!adesc->arena);
750
89086707 751 Newx(adesc->arena, arena_size, char);
5e258f8c 752 adesc->size = arena_size;
0a848332 753 adesc->misc = misc;
d67b3c53
JH
754 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
755 curr, (void*)adesc->arena, (UV)arena_size));
5e258f8c
JC
756
757 return adesc->arena;
5e258f8c
JC
758}
759
53c1dcc0 760
bd81e77b 761/* return a thing to the free list */
29489e7c 762
bd81e77b
NC
763#define del_body(thing, root) \
764 STMT_START { \
00b6aa41 765 void ** const thing_copy = (void **)thing;\
bd81e77b
NC
766 *thing_copy = *root; \
767 *root = (void*)thing_copy; \
bd81e77b 768 } STMT_END
29489e7c 769
bd81e77b 770/*
d2a0f284
JC
771
772=head1 SV-Body Allocation
773
774Allocation of SV-bodies is similar to SV-heads, differing as follows;
775the allocation mechanism is used for many body types, so is somewhat
776more complicated, it uses arena-sets, and has no need for still-live
777SV detection.
778
779At the outermost level, (new|del)_X*V macros return bodies of the
780appropriate type. These macros call either (new|del)_body_type or
781(new|del)_body_allocated macro pairs, depending on specifics of the
782type. Most body types use the former pair, the latter pair is used to
783allocate body types with "ghost fields".
784
785"ghost fields" are fields that are unused in certain types, and
786consequently dont need to actually exist. They are declared because
787they're part of a "base type", which allows use of functions as
788methods. The simplest examples are AVs and HVs, 2 aggregate types
789which don't use the fields which support SCALAR semantics.
790
791For these types, the arenas are carved up into *_allocated size
792chunks, we thus avoid wasted memory for those unaccessed members.
793When bodies are allocated, we adjust the pointer back in memory by the
794size of the bit not allocated, so it's as if we allocated the full
795structure. (But things will all go boom if you write to the part that
796is "not there", because you'll be overwriting the last members of the
797preceding structure in memory.)
798
799We calculate the correction using the STRUCT_OFFSET macro. For
800example, if xpv_allocated is the same structure as XPV then the two
801OFFSETs sum to zero, and the pointer is unchanged. If the allocated
802structure is smaller (no initial NV actually allocated) then the net
803effect is to subtract the size of the NV from the pointer, to return a
804new pointer as if an initial NV were actually allocated.
805
806This is the same trick as was used for NV and IV bodies. Ironically it
807doesn't need to be used for NV bodies any more, because NV is now at
808the start of the structure. IV bodies don't need it either, because
809they are no longer allocated.
810
811In turn, the new_body_* allocators call S_new_body(), which invokes
812new_body_inline macro, which takes a lock, and takes a body off the
813linked list at PL_body_roots[sv_type], calling S_more_bodies() if
814necessary to refresh an empty list. Then the lock is released, and
815the body is returned.
816
817S_more_bodies calls get_arena(), and carves it up into an array of N
818bodies, which it strings into a linked list. It looks up arena-size
819and body-size from the body_details table described below, thus
820supporting the multiple body-types.
821
822If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
823the (new|del)_X*V macros are mapped directly to malloc/free.
824
825*/
826
827/*
828
829For each sv-type, struct body_details bodies_by_type[] carries
830parameters which control these aspects of SV handling:
831
832Arena_size determines whether arenas are used for this body type, and if
833so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
834zero, forcing individual mallocs and frees.
835
836Body_size determines how big a body is, and therefore how many fit into
837each arena. Offset carries the body-pointer adjustment needed for
838*_allocated body types, and is used in *_allocated macros.
839
840But its main purpose is to parameterize info needed in
841Perl_sv_upgrade(). The info here dramatically simplifies the function
842vs the implementation in 5.8.7, making it table-driven. All fields
843are used for this, except for arena_size.
844
845For the sv-types that have no bodies, arenas are not used, so those
846PL_body_roots[sv_type] are unused, and can be overloaded. In
847something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 848PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 849bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 850available in hv.c.
d2a0f284 851
c6f8b1d0
JC
852PTEs also use arenas, but are never seen in Perl_sv_upgrade. Nonetheless,
853they get their own slot in bodies_by_type[PTE_SVSLOT =SVt_IV], so they can
854just use the same allocation semantics. At first, PTEs were also
855overloaded to a non-body sv-type, but this yielded hard-to-find malloc
856bugs, so was simplified by claiming a new slot. This choice has no
857consequence at this time.
d2a0f284 858
29489e7c
DM
859*/
860
bd81e77b 861struct body_details {
0fb58b32 862 U8 body_size; /* Size to allocate */
10666ae3 863 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 864 U8 offset;
10666ae3
NC
865 unsigned int type : 4; /* We have space for a sanity check. */
866 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
867 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
868 unsigned int arena : 1; /* Allocated from an arena */
869 size_t arena_size; /* Size of arena to allocate */
bd81e77b 870};
29489e7c 871
bd81e77b
NC
872#define HADNV FALSE
873#define NONV TRUE
29489e7c 874
d2a0f284 875
bd81e77b
NC
876#ifdef PURIFY
877/* With -DPURFIY we allocate everything directly, and don't use arenas.
878 This seems a rather elegant way to simplify some of the code below. */
879#define HASARENA FALSE
880#else
881#define HASARENA TRUE
882#endif
883#define NOARENA FALSE
29489e7c 884
d2a0f284
JC
885/* Size the arenas to exactly fit a given number of bodies. A count
886 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
887 simplifying the default. If count > 0, the arena is sized to fit
888 only that many bodies, allowing arenas to be used for large, rare
889 bodies (XPVFM, XPVIO) without undue waste. The arena size is
890 limited by PERL_ARENA_SIZE, so we can safely oversize the
891 declarations.
892 */
95db5f15
MB
893#define FIT_ARENA0(body_size) \
894 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
895#define FIT_ARENAn(count,body_size) \
896 ( count * body_size <= PERL_ARENA_SIZE) \
897 ? count * body_size \
898 : FIT_ARENA0 (body_size)
899#define FIT_ARENA(count,body_size) \
900 count \
901 ? FIT_ARENAn (count, body_size) \
902 : FIT_ARENA0 (body_size)
d2a0f284 903
bd81e77b 904/* A macro to work out the offset needed to subtract from a pointer to (say)
29489e7c 905
bd81e77b
NC
906typedef struct {
907 STRLEN xpv_cur;
908 STRLEN xpv_len;
909} xpv_allocated;
29489e7c 910
bd81e77b 911to make its members accessible via a pointer to (say)
29489e7c 912
bd81e77b
NC
913struct xpv {
914 NV xnv_nv;
915 STRLEN xpv_cur;
916 STRLEN xpv_len;
917};
29489e7c 918
bd81e77b 919*/
29489e7c 920
bd81e77b
NC
921#define relative_STRUCT_OFFSET(longer, shorter, member) \
922 (STRUCT_OFFSET(shorter, member) - STRUCT_OFFSET(longer, member))
29489e7c 923
bd81e77b
NC
924/* Calculate the length to copy. Specifically work out the length less any
925 final padding the compiler needed to add. See the comment in sv_upgrade
926 for why copying the padding proved to be a bug. */
29489e7c 927
bd81e77b
NC
928#define copy_length(type, last_member) \
929 STRUCT_OFFSET(type, last_member) \
daba3364 930 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 931
bd81e77b 932static const struct body_details bodies_by_type[] = {
10666ae3
NC
933 { sizeof(HE), 0, 0, SVt_NULL,
934 FALSE, NONV, NOARENA, FIT_ARENA(0, sizeof(HE)) },
d2a0f284 935
1cb9cd50 936 /* The bind placeholder pretends to be an RV for now.
c6f8b1d0 937 Also it's marked as "can't upgrade" to stop anyone using it before it's
1cb9cd50
NC
938 implemented. */
939 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
940
d2a0f284
JC
941 /* IVs are in the head, so the allocation size is 0.
942 However, the slot is overloaded for PTEs. */
943 { sizeof(struct ptr_tbl_ent), /* This is used for PTEs. */
944 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 945 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
d2a0f284
JC
946 NOARENA /* IVS don't need an arena */,
947 /* But PTEs need to know the size of their arena */
948 FIT_ARENA(0, sizeof(struct ptr_tbl_ent))
949 },
950
bd81e77b 951 /* 8 bytes on most ILP32 with IEEE doubles */
10666ae3 952 { sizeof(NV), sizeof(NV), 0, SVt_NV, FALSE, HADNV, HASARENA,
d2a0f284
JC
953 FIT_ARENA(0, sizeof(NV)) },
954
bd81e77b 955 /* 8 bytes on most ILP32 with IEEE doubles */
d2a0f284
JC
956 { sizeof(xpv_allocated),
957 copy_length(XPV, xpv_len)
958 - relative_STRUCT_OFFSET(xpv_allocated, XPV, xpv_cur),
959 + relative_STRUCT_OFFSET(xpv_allocated, XPV, xpv_cur),
10666ae3 960 SVt_PV, FALSE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpv_allocated)) },
d2a0f284 961
bd81e77b 962 /* 12 */
d2a0f284
JC
963 { sizeof(xpviv_allocated),
964 copy_length(XPVIV, xiv_u)
965 - relative_STRUCT_OFFSET(xpviv_allocated, XPVIV, xpv_cur),
966 + relative_STRUCT_OFFSET(xpviv_allocated, XPVIV, xpv_cur),
10666ae3 967 SVt_PVIV, FALSE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpviv_allocated)) },
d2a0f284 968
bd81e77b 969 /* 20 */
10666ae3 970 { sizeof(XPVNV), copy_length(XPVNV, xiv_u), 0, SVt_PVNV, FALSE, HADNV,
d2a0f284
JC
971 HASARENA, FIT_ARENA(0, sizeof(XPVNV)) },
972
bd81e77b 973 /* 28 */
10666ae3 974 { sizeof(XPVMG), copy_length(XPVMG, xmg_stash), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 975 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 976
288b8c02 977 /* something big */
08e44740
NC
978 { sizeof(struct regexp_allocated), sizeof(struct regexp_allocated),
979 + relative_STRUCT_OFFSET(struct regexp_allocated, regexp, xpv_cur),
980 SVt_REGEXP, FALSE, NONV, HASARENA,
981 FIT_ARENA(0, sizeof(struct regexp_allocated))
5c35adbb 982 },
4df7f6af 983
bd81e77b 984 /* 48 */
10666ae3 985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
987
bd81e77b 988 /* 64 */
10666ae3 989 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
990 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
991
992 { sizeof(xpvav_allocated),
993 copy_length(XPVAV, xmg_stash)
994 - relative_STRUCT_OFFSET(xpvav_allocated, XPVAV, xav_fill),
995 + relative_STRUCT_OFFSET(xpvav_allocated, XPVAV, xav_fill),
9c59bb28 996 SVt_PVAV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvav_allocated)) },
d2a0f284
JC
997
998 { sizeof(xpvhv_allocated),
999 copy_length(XPVHV, xmg_stash)
1000 - relative_STRUCT_OFFSET(xpvhv_allocated, XPVHV, xhv_fill),
1001 + relative_STRUCT_OFFSET(xpvhv_allocated, XPVHV, xhv_fill),
9c59bb28 1002 SVt_PVHV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvhv_allocated)) },
d2a0f284 1003
c84c4652 1004 /* 56 */
4115f141 1005 { sizeof(xpvcv_allocated), sizeof(xpvcv_allocated),
c84c4652 1006 + relative_STRUCT_OFFSET(xpvcv_allocated, XPVCV, xpv_cur),
10666ae3 1007 SVt_PVCV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvcv_allocated)) },
d2a0f284 1008
4115f141 1009 { sizeof(xpvfm_allocated), sizeof(xpvfm_allocated),
3038937b 1010 + relative_STRUCT_OFFSET(xpvfm_allocated, XPVFM, xpv_cur),
10666ae3 1011 SVt_PVFM, TRUE, NONV, NOARENA, FIT_ARENA(20, sizeof(xpvfm_allocated)) },
d2a0f284
JC
1012
1013 /* XPVIO is 84 bytes, fits 48x */
167f2c4d
NC
1014 { sizeof(xpvio_allocated), sizeof(xpvio_allocated),
1015 + relative_STRUCT_OFFSET(xpvio_allocated, XPVIO, xpv_cur),
1016 SVt_PVIO, TRUE, NONV, HASARENA, FIT_ARENA(24, sizeof(xpvio_allocated)) },
bd81e77b 1017};
29489e7c 1018
d2a0f284
JC
1019#define new_body_type(sv_type) \
1020 (void *)((char *)S_new_body(aTHX_ sv_type))
29489e7c 1021
bd81e77b
NC
1022#define del_body_type(p, sv_type) \
1023 del_body(p, &PL_body_roots[sv_type])
29489e7c 1024
29489e7c 1025
bd81e77b 1026#define new_body_allocated(sv_type) \
d2a0f284 1027 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1028 - bodies_by_type[sv_type].offset)
29489e7c 1029
bd81e77b
NC
1030#define del_body_allocated(p, sv_type) \
1031 del_body(p + bodies_by_type[sv_type].offset, &PL_body_roots[sv_type])
29489e7c 1032
29489e7c 1033
bd81e77b
NC
1034#define my_safemalloc(s) (void*)safemalloc(s)
1035#define my_safecalloc(s) (void*)safecalloc(s, 1)
1036#define my_safefree(p) safefree((char*)p)
29489e7c 1037
bd81e77b 1038#ifdef PURIFY
29489e7c 1039
bd81e77b
NC
1040#define new_XNV() my_safemalloc(sizeof(XPVNV))
1041#define del_XNV(p) my_safefree(p)
29489e7c 1042
bd81e77b
NC
1043#define new_XPVNV() my_safemalloc(sizeof(XPVNV))
1044#define del_XPVNV(p) my_safefree(p)
29489e7c 1045
bd81e77b
NC
1046#define new_XPVAV() my_safemalloc(sizeof(XPVAV))
1047#define del_XPVAV(p) my_safefree(p)
29489e7c 1048
bd81e77b
NC
1049#define new_XPVHV() my_safemalloc(sizeof(XPVHV))
1050#define del_XPVHV(p) my_safefree(p)
29489e7c 1051
bd81e77b
NC
1052#define new_XPVMG() my_safemalloc(sizeof(XPVMG))
1053#define del_XPVMG(p) my_safefree(p)
29489e7c 1054
bd81e77b
NC
1055#define new_XPVGV() my_safemalloc(sizeof(XPVGV))
1056#define del_XPVGV(p) my_safefree(p)
29489e7c 1057
bd81e77b 1058#else /* !PURIFY */
29489e7c 1059
bd81e77b
NC
1060#define new_XNV() new_body_type(SVt_NV)
1061#define del_XNV(p) del_body_type(p, SVt_NV)
29489e7c 1062
bd81e77b
NC
1063#define new_XPVNV() new_body_type(SVt_PVNV)
1064#define del_XPVNV(p) del_body_type(p, SVt_PVNV)
29489e7c 1065
bd81e77b
NC
1066#define new_XPVAV() new_body_allocated(SVt_PVAV)
1067#define del_XPVAV(p) del_body_allocated(p, SVt_PVAV)
645c22ef 1068
bd81e77b
NC
1069#define new_XPVHV() new_body_allocated(SVt_PVHV)
1070#define del_XPVHV(p) del_body_allocated(p, SVt_PVHV)
645c22ef 1071
bd81e77b
NC
1072#define new_XPVMG() new_body_type(SVt_PVMG)
1073#define del_XPVMG(p) del_body_type(p, SVt_PVMG)
645c22ef 1074
bd81e77b
NC
1075#define new_XPVGV() new_body_type(SVt_PVGV)
1076#define del_XPVGV(p) del_body_type(p, SVt_PVGV)
1d7c1841 1077
bd81e77b 1078#endif /* PURIFY */
93e68bfb 1079
bd81e77b 1080/* no arena for you! */
93e68bfb 1081
bd81e77b 1082#define new_NOARENA(details) \
d2a0f284 1083 my_safemalloc((details)->body_size + (details)->offset)
bd81e77b 1084#define new_NOARENAZ(details) \
d2a0f284
JC
1085 my_safecalloc((details)->body_size + (details)->offset)
1086
1087STATIC void *
de37a194 1088S_more_bodies (pTHX_ const svtype sv_type)
d2a0f284
JC
1089{
1090 dVAR;
1091 void ** const root = &PL_body_roots[sv_type];
96a5add6 1092 const struct body_details * const bdp = &bodies_by_type[sv_type];
d2a0f284
JC
1093 const size_t body_size = bdp->body_size;
1094 char *start;
1095 const char *end;
d8fca402 1096 const size_t arena_size = Perl_malloc_good_size(bdp->arena_size);
0b2d3faa 1097#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1098 static bool done_sanity_check;
1099
0b2d3faa
JH
1100 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1101 * variables like done_sanity_check. */
10666ae3 1102 if (!done_sanity_check) {
ea471437 1103 unsigned int i = SVt_LAST;
10666ae3
NC
1104
1105 done_sanity_check = TRUE;
1106
1107 while (i--)
1108 assert (bodies_by_type[i].type == i);
1109 }
1110#endif
1111
23e9d66c
NC
1112 assert(bdp->arena_size);
1113
d8fca402 1114 start = (char*) Perl_get_arena(aTHX_ arena_size, sv_type);
d2a0f284 1115
d8fca402 1116 end = start + arena_size - 2 * body_size;
d2a0f284 1117
d2a0f284 1118 /* computed count doesnt reflect the 1st slot reservation */
d8fca402
NC
1119#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1120 DEBUG_m(PerlIO_printf(Perl_debug_log,
1121 "arena %p end %p arena-size %d (from %d) type %d "
1122 "size %d ct %d\n",
1123 (void*)start, (void*)end, (int)arena_size,
1124 (int)bdp->arena_size, sv_type, (int)body_size,
1125 (int)arena_size / (int)body_size));
1126#else
d2a0f284
JC
1127 DEBUG_m(PerlIO_printf(Perl_debug_log,
1128 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1129 (void*)start, (void*)end,
0e84aef4
JH
1130 (int)bdp->arena_size, sv_type, (int)body_size,
1131 (int)bdp->arena_size / (int)body_size));
d8fca402 1132#endif
d2a0f284
JC
1133 *root = (void *)start;
1134
d8fca402 1135 while (start <= end) {
d2a0f284
JC
1136 char * const next = start + body_size;
1137 *(void**) start = (void *)next;
1138 start = next;
1139 }
1140 *(void **)start = 0;
1141
1142 return *root;
1143}
1144
1145/* grab a new thing from the free list, allocating more if necessary.
1146 The inline version is used for speed in hot routines, and the
1147 function using it serves the rest (unless PURIFY).
1148*/
1149#define new_body_inline(xpv, sv_type) \
1150 STMT_START { \
1151 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775
DD
1152 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1153 ? *((void **)(r3wt)) : more_bodies(sv_type)); \
d2a0f284 1154 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1155 } STMT_END
1156
1157#ifndef PURIFY
1158
1159STATIC void *
de37a194 1160S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1161{
1162 dVAR;
1163 void *xpv;
1164 new_body_inline(xpv, sv_type);
1165 return xpv;
1166}
1167
1168#endif
93e68bfb 1169
238b27b3
NC
1170static const struct body_details fake_rv =
1171 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1172
bd81e77b
NC
1173/*
1174=for apidoc sv_upgrade
93e68bfb 1175
bd81e77b
NC
1176Upgrade an SV to a more complex form. Generally adds a new body type to the
1177SV, then copies across as much information as possible from the old body.
1178You generally want to use the C<SvUPGRADE> macro wrapper. See also C<svtype>.
93e68bfb 1179
bd81e77b 1180=cut
93e68bfb 1181*/
93e68bfb 1182
bd81e77b 1183void
aad570aa 1184Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
cac9b346 1185{
97aff369 1186 dVAR;
bd81e77b
NC
1187 void* old_body;
1188 void* new_body;
42d0e0b7 1189 const svtype old_type = SvTYPE(sv);
d2a0f284 1190 const struct body_details *new_type_details;
238b27b3 1191 const struct body_details *old_type_details
bd81e77b 1192 = bodies_by_type + old_type;
4df7f6af 1193 SV *referant = NULL;
cac9b346 1194
7918f24d
NC
1195 PERL_ARGS_ASSERT_SV_UPGRADE;
1196
bd81e77b
NC
1197 if (new_type != SVt_PV && SvIsCOW(sv)) {
1198 sv_force_normal_flags(sv, 0);
1199 }
cac9b346 1200
bd81e77b
NC
1201 if (old_type == new_type)
1202 return;
cac9b346 1203
bd81e77b 1204 old_body = SvANY(sv);
de042e1d 1205
bd81e77b
NC
1206 /* Copying structures onto other structures that have been neatly zeroed
1207 has a subtle gotcha. Consider XPVMG
cac9b346 1208
bd81e77b
NC
1209 +------+------+------+------+------+-------+-------+
1210 | NV | CUR | LEN | IV | MAGIC | STASH |
1211 +------+------+------+------+------+-------+-------+
1212 0 4 8 12 16 20 24 28
645c22ef 1213
bd81e77b
NC
1214 where NVs are aligned to 8 bytes, so that sizeof that structure is
1215 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1216
bd81e77b
NC
1217 +------+------+------+------+------+-------+-------+------+
1218 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1219 +------+------+------+------+------+-------+-------+------+
1220 0 4 8 12 16 20 24 28 32
08742458 1221
bd81e77b 1222 so what happens if you allocate memory for this structure:
30f9da9e 1223
bd81e77b
NC
1224 +------+------+------+------+------+-------+-------+------+------+...
1225 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1226 +------+------+------+------+------+-------+-------+------+------+...
1227 0 4 8 12 16 20 24 28 32 36
bfc44f79 1228
bd81e77b
NC
1229 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1230 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1231 started out as zero once, but it's quite possible that it isn't. So now,
1232 rather than a nicely zeroed GP, you have it pointing somewhere random.
1233 Bugs ensue.
bfc44f79 1234
bd81e77b
NC
1235 (In fact, GP ends up pointing at a previous GP structure, because the
1236 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1237 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1238 this happens to be moot because XPVGV has been re-ordered, with GP
1239 no longer after STASH)
30f9da9e 1240
bd81e77b
NC
1241 So we are careful and work out the size of used parts of all the
1242 structures. */
bfc44f79 1243
bd81e77b
NC
1244 switch (old_type) {
1245 case SVt_NULL:
1246 break;
1247 case SVt_IV:
4df7f6af
NC
1248 if (SvROK(sv)) {
1249 referant = SvRV(sv);
238b27b3
NC
1250 old_type_details = &fake_rv;
1251 if (new_type == SVt_NV)
1252 new_type = SVt_PVNV;
4df7f6af
NC
1253 } else {
1254 if (new_type < SVt_PVIV) {
1255 new_type = (new_type == SVt_NV)
1256 ? SVt_PVNV : SVt_PVIV;
1257 }
bd81e77b
NC
1258 }
1259 break;
1260 case SVt_NV:
1261 if (new_type < SVt_PVNV) {
1262 new_type = SVt_PVNV;
bd81e77b
NC
1263 }
1264 break;
bd81e77b
NC
1265 case SVt_PV:
1266 assert(new_type > SVt_PV);
1267 assert(SVt_IV < SVt_PV);
1268 assert(SVt_NV < SVt_PV);
1269 break;
1270 case SVt_PVIV:
1271 break;
1272 case SVt_PVNV:
1273 break;
1274 case SVt_PVMG:
1275 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1276 there's no way that it can be safely upgraded, because perl.c
1277 expects to Safefree(SvANY(PL_mess_sv)) */
1278 assert(sv != PL_mess_sv);
1279 /* This flag bit is used to mean other things in other scalar types.
1280 Given that it only has meaning inside the pad, it shouldn't be set
1281 on anything that can get upgraded. */
00b1698f 1282 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1283 break;
1284 default:
1285 if (old_type_details->cant_upgrade)
c81225bc
NC
1286 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1287 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1288 }
3376de98
NC
1289
1290 if (old_type > new_type)
1291 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1292 (int)old_type, (int)new_type);
1293
2fa1109b 1294 new_type_details = bodies_by_type + new_type;
645c22ef 1295
bd81e77b
NC
1296 SvFLAGS(sv) &= ~SVTYPEMASK;
1297 SvFLAGS(sv) |= new_type;
932e9ff9 1298
ab4416c0
NC
1299 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1300 the return statements above will have triggered. */
1301 assert (new_type != SVt_NULL);
bd81e77b 1302 switch (new_type) {
bd81e77b
NC
1303 case SVt_IV:
1304 assert(old_type == SVt_NULL);
1305 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1306 SvIV_set(sv, 0);
1307 return;
1308 case SVt_NV:
1309 assert(old_type == SVt_NULL);
1310 SvANY(sv) = new_XNV();
1311 SvNV_set(sv, 0);
1312 return;
bd81e77b 1313 case SVt_PVHV:
bd81e77b 1314 case SVt_PVAV:
d2a0f284 1315 assert(new_type_details->body_size);
c1ae03ae
NC
1316
1317#ifndef PURIFY
1318 assert(new_type_details->arena);
d2a0f284 1319 assert(new_type_details->arena_size);
c1ae03ae 1320 /* This points to the start of the allocated area. */
d2a0f284
JC
1321 new_body_inline(new_body, new_type);
1322 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1323 new_body = ((char *)new_body) - new_type_details->offset;
1324#else
1325 /* We always allocated the full length item with PURIFY. To do this
1326 we fake things so that arena is false for all 16 types.. */
1327 new_body = new_NOARENAZ(new_type_details);
1328#endif
1329 SvANY(sv) = new_body;
1330 if (new_type == SVt_PVAV) {
1331 AvMAX(sv) = -1;
1332 AvFILLp(sv) = -1;
1333 AvREAL_only(sv);
64484faa 1334 if (old_type_details->body_size) {
ac572bf4
NC
1335 AvALLOC(sv) = 0;
1336 } else {
1337 /* It will have been zeroed when the new body was allocated.
1338 Lets not write to it, in case it confuses a write-back
1339 cache. */
1340 }
78ac7dd9
NC
1341 } else {
1342 assert(!SvOK(sv));
1343 SvOK_off(sv);
1344#ifndef NODEFAULT_SHAREKEYS
1345 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1346#endif
1347 HvMAX(sv) = 7; /* (start with 8 buckets) */
64484faa 1348 if (old_type_details->body_size) {
78ac7dd9
NC
1349 HvFILL(sv) = 0;
1350 } else {
1351 /* It will have been zeroed when the new body was allocated.
1352 Lets not write to it, in case it confuses a write-back
1353 cache. */
1354 }
c1ae03ae 1355 }
aeb18a1e 1356
bd81e77b
NC
1357 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1358 The target created by newSVrv also is, and it can have magic.
1359 However, it never has SvPVX set.
1360 */
4df7f6af
NC
1361 if (old_type == SVt_IV) {
1362 assert(!SvROK(sv));
1363 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1364 assert(SvPVX_const(sv) == 0);
1365 }
aeb18a1e 1366
bd81e77b 1367 if (old_type >= SVt_PVMG) {
e736a858 1368 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1369 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1370 } else {
1371 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1372 }
1373 break;
93e68bfb 1374
93e68bfb 1375
bd81e77b
NC
1376 case SVt_PVIV:
1377 /* XXX Is this still needed? Was it ever needed? Surely as there is
1378 no route from NV to PVIV, NOK can never be true */
1379 assert(!SvNOKp(sv));
1380 assert(!SvNOK(sv));
1381 case SVt_PVIO:
1382 case SVt_PVFM:
bd81e77b
NC
1383 case SVt_PVGV:
1384 case SVt_PVCV:
1385 case SVt_PVLV:
5c35adbb 1386 case SVt_REGEXP:
bd81e77b
NC
1387 case SVt_PVMG:
1388 case SVt_PVNV:
1389 case SVt_PV:
93e68bfb 1390
d2a0f284 1391 assert(new_type_details->body_size);
bd81e77b
NC
1392 /* We always allocated the full length item with PURIFY. To do this
1393 we fake things so that arena is false for all 16 types.. */
1394 if(new_type_details->arena) {
1395 /* This points to the start of the allocated area. */
d2a0f284
JC
1396 new_body_inline(new_body, new_type);
1397 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1398 new_body = ((char *)new_body) - new_type_details->offset;
1399 } else {
1400 new_body = new_NOARENAZ(new_type_details);
1401 }
1402 SvANY(sv) = new_body;
5e2fc214 1403
bd81e77b 1404 if (old_type_details->copy) {
f9ba3d20
NC
1405 /* There is now the potential for an upgrade from something without
1406 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1407 int offset = old_type_details->offset;
1408 int length = old_type_details->copy;
1409
1410 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1411 const int difference
f9ba3d20
NC
1412 = new_type_details->offset - old_type_details->offset;
1413 offset += difference;
1414 length -= difference;
1415 }
1416 assert (length >= 0);
1417
1418 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1419 char);
bd81e77b
NC
1420 }
1421
1422#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1423 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1424 * correct 0.0 for us. Otherwise, if the old body didn't have an
1425 * NV slot, but the new one does, then we need to initialise the
1426 * freshly created NV slot with whatever the correct bit pattern is
1427 * for 0.0 */
e22a937e
NC
1428 if (old_type_details->zero_nv && !new_type_details->zero_nv
1429 && !isGV_with_GP(sv))
bd81e77b 1430 SvNV_set(sv, 0);
82048762 1431#endif
5e2fc214 1432
bd81e77b 1433 if (new_type == SVt_PVIO)
f2524eef 1434 IoPAGE_LEN(sv) = 60;
4df7f6af
NC
1435 if (old_type < SVt_PV) {
1436 /* referant will be NULL unless the old type was SVt_IV emulating
1437 SVt_RV */
1438 sv->sv_u.svu_rv = referant;
1439 }
bd81e77b
NC
1440 break;
1441 default:
afd78fd5
JH
1442 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1443 (unsigned long)new_type);
bd81e77b 1444 }
73171d91 1445
d2a0f284
JC
1446 if (old_type_details->arena) {
1447 /* If there was an old body, then we need to free it.
1448 Note that there is an assumption that all bodies of types that
1449 can be upgraded came from arenas. Only the more complex non-
1450 upgradable types are allowed to be directly malloc()ed. */
bd81e77b
NC
1451#ifdef PURIFY
1452 my_safefree(old_body);
1453#else
1454 del_body((void*)((char*)old_body + old_type_details->offset),
1455 &PL_body_roots[old_type]);
1456#endif
1457 }
1458}
73171d91 1459
bd81e77b
NC
1460/*
1461=for apidoc sv_backoff
73171d91 1462
bd81e77b
NC
1463Remove any string offset. You should normally use the C<SvOOK_off> macro
1464wrapper instead.
73171d91 1465
bd81e77b 1466=cut
73171d91
NC
1467*/
1468
bd81e77b 1469int
aad570aa 1470Perl_sv_backoff(pTHX_ register SV *const sv)
bd81e77b 1471{
69240efd 1472 STRLEN delta;
7a4bba22 1473 const char * const s = SvPVX_const(sv);
7918f24d
NC
1474
1475 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1476 PERL_UNUSED_CONTEXT;
7918f24d 1477
bd81e77b
NC
1478 assert(SvOOK(sv));
1479 assert(SvTYPE(sv) != SVt_PVHV);
1480 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1481
69240efd
NC
1482 SvOOK_offset(sv, delta);
1483
7a4bba22
NC
1484 SvLEN_set(sv, SvLEN(sv) + delta);
1485 SvPV_set(sv, SvPVX(sv) - delta);
1486 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1487 SvFLAGS(sv) &= ~SVf_OOK;
1488 return 0;
1489}
73171d91 1490
bd81e77b
NC
1491/*
1492=for apidoc sv_grow
73171d91 1493
bd81e77b
NC
1494Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1495upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1496Use the C<SvGROW> wrapper instead.
93e68bfb 1497
bd81e77b
NC
1498=cut
1499*/
93e68bfb 1500
bd81e77b 1501char *
aad570aa 1502Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
bd81e77b
NC
1503{
1504 register char *s;
93e68bfb 1505
7918f24d
NC
1506 PERL_ARGS_ASSERT_SV_GROW;
1507
5db06880
NC
1508 if (PL_madskills && newlen >= 0x100000) {
1509 PerlIO_printf(Perl_debug_log,
1510 "Allocation too large: %"UVxf"\n", (UV)newlen);
1511 }
bd81e77b
NC
1512#ifdef HAS_64K_LIMIT
1513 if (newlen >= 0x10000) {
1514 PerlIO_printf(Perl_debug_log,
1515 "Allocation too large: %"UVxf"\n", (UV)newlen);
1516 my_exit(1);
1517 }
1518#endif /* HAS_64K_LIMIT */
1519 if (SvROK(sv))
1520 sv_unref(sv);
1521 if (SvTYPE(sv) < SVt_PV) {
1522 sv_upgrade(sv, SVt_PV);
1523 s = SvPVX_mutable(sv);
1524 }
1525 else if (SvOOK(sv)) { /* pv is offset? */
1526 sv_backoff(sv);
1527 s = SvPVX_mutable(sv);
1528 if (newlen > SvLEN(sv))
1529 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1530#ifdef HAS_64K_LIMIT
1531 if (newlen >= 0x10000)
1532 newlen = 0xFFFF;
1533#endif
1534 }
1535 else
1536 s = SvPVX_mutable(sv);
aeb18a1e 1537
bd81e77b 1538 if (newlen > SvLEN(sv)) { /* need more room? */
aedff202 1539#ifndef Perl_safesysmalloc_size
bd81e77b 1540 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1541#endif
98653f18 1542 if (SvLEN(sv) && s) {
10edeb5d 1543 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1544 }
1545 else {
10edeb5d 1546 s = (char*)safemalloc(newlen);
bd81e77b
NC
1547 if (SvPVX_const(sv) && SvCUR(sv)) {
1548 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1549 }
1550 }
1551 SvPV_set(sv, s);
ca7c1a29 1552#ifdef Perl_safesysmalloc_size
98653f18
NC
1553 /* Do this here, do it once, do it right, and then we will never get
1554 called back into sv_grow() unless there really is some growing
1555 needed. */
ca7c1a29 1556 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1557#else
bd81e77b 1558 SvLEN_set(sv, newlen);
98653f18 1559#endif
bd81e77b
NC
1560 }
1561 return s;
1562}
aeb18a1e 1563
bd81e77b
NC
1564/*
1565=for apidoc sv_setiv
932e9ff9 1566
bd81e77b
NC
1567Copies an integer into the given SV, upgrading first if necessary.
1568Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1569
bd81e77b
NC
1570=cut
1571*/
463ee0b2 1572
bd81e77b 1573void
aad570aa 1574Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
bd81e77b 1575{
97aff369 1576 dVAR;
7918f24d
NC
1577
1578 PERL_ARGS_ASSERT_SV_SETIV;
1579
bd81e77b
NC
1580 SV_CHECK_THINKFIRST_COW_DROP(sv);
1581 switch (SvTYPE(sv)) {
1582 case SVt_NULL:
bd81e77b 1583 case SVt_NV:
3376de98 1584 sv_upgrade(sv, SVt_IV);
bd81e77b 1585 break;
bd81e77b
NC
1586 case SVt_PV:
1587 sv_upgrade(sv, SVt_PVIV);
1588 break;
463ee0b2 1589
bd81e77b 1590 case SVt_PVGV:
6e592b3a
BM
1591 if (!isGV_with_GP(sv))
1592 break;
bd81e77b
NC
1593 case SVt_PVAV:
1594 case SVt_PVHV:
1595 case SVt_PVCV:
1596 case SVt_PVFM:
1597 case SVt_PVIO:
1598 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1599 OP_DESC(PL_op));
42d0e0b7 1600 default: NOOP;
bd81e77b
NC
1601 }
1602 (void)SvIOK_only(sv); /* validate number */
1603 SvIV_set(sv, i);
1604 SvTAINT(sv);
1605}
932e9ff9 1606
bd81e77b
NC
1607/*
1608=for apidoc sv_setiv_mg
d33b2eba 1609
bd81e77b 1610Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1611
bd81e77b
NC
1612=cut
1613*/
d33b2eba 1614
bd81e77b 1615void
aad570aa 1616Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
bd81e77b 1617{
7918f24d
NC
1618 PERL_ARGS_ASSERT_SV_SETIV_MG;
1619
bd81e77b
NC
1620 sv_setiv(sv,i);
1621 SvSETMAGIC(sv);
1622}
727879eb 1623
bd81e77b
NC
1624/*
1625=for apidoc sv_setuv
d33b2eba 1626
bd81e77b
NC
1627Copies an unsigned integer into the given SV, upgrading first if necessary.
1628Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1629
bd81e77b
NC
1630=cut
1631*/
d33b2eba 1632
bd81e77b 1633void
aad570aa 1634Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
bd81e77b 1635{
7918f24d
NC
1636 PERL_ARGS_ASSERT_SV_SETUV;
1637
bd81e77b
NC
1638 /* With these two if statements:
1639 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1640
bd81e77b
NC
1641 without
1642 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1643
bd81e77b
NC
1644 If you wish to remove them, please benchmark to see what the effect is
1645 */
1646 if (u <= (UV)IV_MAX) {
1647 sv_setiv(sv, (IV)u);
1648 return;
1649 }
1650 sv_setiv(sv, 0);
1651 SvIsUV_on(sv);
1652 SvUV_set(sv, u);
1653}
d33b2eba 1654
bd81e77b
NC
1655/*
1656=for apidoc sv_setuv_mg
727879eb 1657
bd81e77b 1658Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1659
bd81e77b
NC
1660=cut
1661*/
5e2fc214 1662
bd81e77b 1663void
aad570aa 1664Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
bd81e77b 1665{
7918f24d
NC
1666 PERL_ARGS_ASSERT_SV_SETUV_MG;
1667
bd81e77b
NC
1668 sv_setuv(sv,u);
1669 SvSETMAGIC(sv);
1670}
5e2fc214 1671
954c1994 1672/*
bd81e77b 1673=for apidoc sv_setnv
954c1994 1674
bd81e77b
NC
1675Copies a double into the given SV, upgrading first if necessary.
1676Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1677
1678=cut
1679*/
1680
63f97190 1681void
aad570aa 1682Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
79072805 1683{
97aff369 1684 dVAR;
7918f24d
NC
1685
1686 PERL_ARGS_ASSERT_SV_SETNV;
1687
bd81e77b
NC
1688 SV_CHECK_THINKFIRST_COW_DROP(sv);
1689 switch (SvTYPE(sv)) {
79072805 1690 case SVt_NULL:
79072805 1691 case SVt_IV:
bd81e77b 1692 sv_upgrade(sv, SVt_NV);
79072805
LW
1693 break;
1694 case SVt_PV:
79072805 1695 case SVt_PVIV:
bd81e77b 1696 sv_upgrade(sv, SVt_PVNV);
79072805 1697 break;
bd4b1eb5 1698
bd4b1eb5 1699 case SVt_PVGV:
6e592b3a
BM
1700 if (!isGV_with_GP(sv))
1701 break;
bd81e77b
NC
1702 case SVt_PVAV:
1703 case SVt_PVHV:
79072805 1704 case SVt_PVCV:
bd81e77b
NC
1705 case SVt_PVFM:
1706 case SVt_PVIO:
1707 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1708 OP_NAME(PL_op));
42d0e0b7 1709 default: NOOP;
2068cd4d 1710 }
bd81e77b
NC
1711 SvNV_set(sv, num);
1712 (void)SvNOK_only(sv); /* validate number */
1713 SvTAINT(sv);
79072805
LW
1714}
1715
645c22ef 1716/*
bd81e77b 1717=for apidoc sv_setnv_mg
645c22ef 1718
bd81e77b 1719Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1720
1721=cut
1722*/
1723
bd81e77b 1724void
aad570aa 1725Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
79072805 1726{
7918f24d
NC
1727 PERL_ARGS_ASSERT_SV_SETNV_MG;
1728
bd81e77b
NC
1729 sv_setnv(sv,num);
1730 SvSETMAGIC(sv);
79072805
LW
1731}
1732
bd81e77b
NC
1733/* Print an "isn't numeric" warning, using a cleaned-up,
1734 * printable version of the offending string
1735 */
954c1994 1736
bd81e77b 1737STATIC void
aad570aa 1738S_not_a_number(pTHX_ SV *const sv)
79072805 1739{
97aff369 1740 dVAR;
bd81e77b
NC
1741 SV *dsv;
1742 char tmpbuf[64];
1743 const char *pv;
94463019 1744
7918f24d
NC
1745 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1746
94463019 1747 if (DO_UTF8(sv)) {
84bafc02 1748 dsv = newSVpvs_flags("", SVs_TEMP);
94463019
JH
1749 pv = sv_uni_display(dsv, sv, 10, 0);
1750 } else {
1751 char *d = tmpbuf;
551405c4 1752 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1753 /* each *s can expand to 4 chars + "...\0",
1754 i.e. need room for 8 chars */
ecdeb87c 1755
00b6aa41
AL
1756 const char *s = SvPVX_const(sv);
1757 const char * const end = s + SvCUR(sv);
1758 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1759 int ch = *s & 0xFF;
1760 if (ch & 128 && !isPRINT_LC(ch)) {
1761 *d++ = 'M';
1762 *d++ = '-';
1763 ch &= 127;
1764 }
1765 if (ch == '\n') {
1766 *d++ = '\\';
1767 *d++ = 'n';
1768 }
1769 else if (ch == '\r') {
1770 *d++ = '\\';
1771 *d++ = 'r';
1772 }
1773 else if (ch == '\f') {
1774 *d++ = '\\';
1775 *d++ = 'f';
1776 }
1777 else if (ch == '\\') {
1778 *d++ = '\\';
1779 *d++ = '\\';
1780 }
1781 else if (ch == '\0') {
1782 *d++ = '\\';
1783 *d++ = '0';
1784 }
1785 else if (isPRINT_LC(ch))
1786 *d++ = ch;
1787 else {
1788 *d++ = '^';
1789 *d++ = toCTRL(ch);
1790 }
1791 }
1792 if (s < end) {
1793 *d++ = '.';
1794 *d++ = '.';
1795 *d++ = '.';
1796 }
1797 *d = '\0';
1798 pv = tmpbuf;
a0d0e21e 1799 }
a0d0e21e 1800
533c011a 1801 if (PL_op)
9014280d 1802 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019
JH
1803 "Argument \"%s\" isn't numeric in %s", pv,
1804 OP_DESC(PL_op));
a0d0e21e 1805 else
9014280d 1806 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019 1807 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1808}
1809
c2988b20
NC
1810/*
1811=for apidoc looks_like_number
1812
645c22ef
DM
1813Test if the content of an SV looks like a number (or is a number).
1814C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1815non-numeric warning), even if your atof() doesn't grok them.
c2988b20
NC
1816
1817=cut
1818*/
1819
1820I32
aad570aa 1821Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1822{
a3b680e6 1823 register const char *sbegin;
c2988b20
NC
1824 STRLEN len;
1825
7918f24d
NC
1826 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1827
c2988b20 1828 if (SvPOK(sv)) {
3f7c398e 1829 sbegin = SvPVX_const(sv);
c2988b20
NC
1830 len = SvCUR(sv);
1831 }
1832 else if (SvPOKp(sv))
83003860 1833 sbegin = SvPV_const(sv, len);
c2988b20 1834 else
e0ab1c0e 1835 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1836 return grok_number(sbegin, len, NULL);
1837}
25da4f38 1838
19f6321d
NC
1839STATIC bool
1840S_glob_2number(pTHX_ GV * const gv)
180488f8
NC
1841{
1842 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1843 SV *const buffer = sv_newmortal();
1844
7918f24d
NC
1845 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1846
180488f8
NC
1847 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1848 is on. */
1849 SvFAKE_off(gv);
1850 gv_efullname3(buffer, gv, "*");
1851 SvFLAGS(gv) |= wasfake;
1852
675c862f
AL
1853 /* We know that all GVs stringify to something that is not-a-number,
1854 so no need to test that. */
1855 if (ckWARN(WARN_NUMERIC))
1856 not_a_number(buffer);
1857 /* We just want something true to return, so that S_sv_2iuv_common
1858 can tail call us and return true. */
19f6321d 1859 return TRUE;
675c862f
AL
1860}
1861
1862STATIC char *
19f6321d 1863S_glob_2pv(pTHX_ GV * const gv, STRLEN * const len)
675c862f
AL
1864{
1865 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1866 SV *const buffer = sv_newmortal();
1867
7918f24d
NC
1868 PERL_ARGS_ASSERT_GLOB_2PV;
1869
675c862f
AL
1870 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1871 is on. */
1872 SvFAKE_off(gv);
1873 gv_efullname3(buffer, gv, "*");
1874 SvFLAGS(gv) |= wasfake;
1875
1876 assert(SvPOK(buffer));
a6d61a6c
NC
1877 if (len) {
1878 *len = SvCUR(buffer);
1879 }
675c862f 1880 return SvPVX(buffer);
180488f8
NC
1881}
1882
25da4f38
IZ
1883/* Actually, ISO C leaves conversion of UV to IV undefined, but
1884 until proven guilty, assume that things are not that bad... */
1885
645c22ef
DM
1886/*
1887 NV_PRESERVES_UV:
1888
1889 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1890 an IV (an assumption perl has been based on to date) it becomes necessary
1891 to remove the assumption that the NV always carries enough precision to
1892 recreate the IV whenever needed, and that the NV is the canonical form.
1893 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1894 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1895 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1896 1) to distinguish between IV/UV/NV slots that have cached a valid
1897 conversion where precision was lost and IV/UV/NV slots that have a
1898 valid conversion which has lost no precision
645c22ef 1899 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1900 would lose precision, the precise conversion (or differently
1901 imprecise conversion) is also performed and cached, to prevent
1902 requests for different numeric formats on the same SV causing
1903 lossy conversion chains. (lossless conversion chains are perfectly
1904 acceptable (still))
1905
1906
1907 flags are used:
1908 SvIOKp is true if the IV slot contains a valid value
1909 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1910 SvNOKp is true if the NV slot contains a valid value
1911 SvNOK is true only if the NV value is accurate
1912
1913 so
645c22ef 1914 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1915 IV(or UV) would lose accuracy over a direct conversion from PV to
1916 IV(or UV). If it would, cache both conversions, return NV, but mark
1917 SV as IOK NOKp (ie not NOK).
1918
645c22ef 1919 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1920 NV would lose accuracy over a direct conversion from PV to NV. If it
1921 would, cache both conversions, flag similarly.
1922
1923 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1924 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1925 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1926 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1927 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1928
645c22ef
DM
1929 The benefit of this is that operations such as pp_add know that if
1930 SvIOK is true for both left and right operands, then integer addition
1931 can be used instead of floating point (for cases where the result won't
1932 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1933 loss of precision compared with integer addition.
1934
1935 * making IV and NV equal status should make maths accurate on 64 bit
1936 platforms
1937 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1938 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1939 looking for SvIOK and checking for overflow will not outweigh the
1940 fp to integer speedup)
1941 * will slow down integer operations (callers of SvIV) on "inaccurate"
1942 values, as the change from SvIOK to SvIOKp will cause a call into
1943 sv_2iv each time rather than a macro access direct to the IV slot
1944 * should speed up number->string conversion on integers as IV is
645c22ef 1945 favoured when IV and NV are equally accurate
28e5dec8
JH
1946
1947 ####################################################################
645c22ef
DM
1948 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1949 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1950 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1951 ####################################################################
1952
645c22ef 1953 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1954 performance ratio.
1955*/
1956
1957#ifndef NV_PRESERVES_UV
645c22ef
DM
1958# define IS_NUMBER_UNDERFLOW_IV 1
1959# define IS_NUMBER_UNDERFLOW_UV 2
1960# define IS_NUMBER_IV_AND_UV 2
1961# define IS_NUMBER_OVERFLOW_IV 4
1962# define IS_NUMBER_OVERFLOW_UV 5
1963
1964/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1965
1966/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1967STATIC int
5de3775c 1968S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1969# ifdef DEBUGGING
1970 , I32 numtype
1971# endif
1972 )
28e5dec8 1973{
97aff369 1974 dVAR;
7918f24d
NC
1975
1976 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1977
3f7c398e 1978 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%"UVxf" NV=%"NVgf" inttype=%"UVXf"\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
28e5dec8
JH
1979 if (SvNVX(sv) < (NV)IV_MIN) {
1980 (void)SvIOKp_on(sv);
1981 (void)SvNOK_on(sv);
45977657 1982 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1983 return IS_NUMBER_UNDERFLOW_IV;
1984 }
1985 if (SvNVX(sv) > (NV)UV_MAX) {
1986 (void)SvIOKp_on(sv);
1987 (void)SvNOK_on(sv);
1988 SvIsUV_on(sv);
607fa7f2 1989 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1990 return IS_NUMBER_OVERFLOW_UV;
1991 }
c2988b20
NC
1992 (void)SvIOKp_on(sv);
1993 (void)SvNOK_on(sv);
1994 /* Can't use strtol etc to convert this string. (See truth table in
1995 sv_2iv */
1996 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1997 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1998 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1999 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2000 } else {
2001 /* Integer is imprecise. NOK, IOKp */
2002 }
2003 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2004 }
2005 SvIsUV_on(sv);
607fa7f2 2006 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
2007 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2008 if (SvUVX(sv) == UV_MAX) {
2009 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2010 possibly be preserved by NV. Hence, it must be overflow.
2011 NOK, IOKp */
2012 return IS_NUMBER_OVERFLOW_UV;
2013 }
2014 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2015 } else {
2016 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2017 }
c2988b20 2018 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2019}
645c22ef
DM
2020#endif /* !NV_PRESERVES_UV*/
2021
af359546 2022STATIC bool
7918f24d
NC
2023S_sv_2iuv_common(pTHX_ SV *const sv)
2024{
97aff369 2025 dVAR;
7918f24d
NC
2026
2027 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2028
af359546 2029 if (SvNOKp(sv)) {
28e5dec8
JH
2030 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2031 * without also getting a cached IV/UV from it at the same time
2032 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2033 * IV or UV at same time to avoid this. */
2034 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2035
2036 if (SvTYPE(sv) == SVt_NV)
2037 sv_upgrade(sv, SVt_PVNV);
2038
28e5dec8
JH
2039 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2040 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2041 certainly cast into the IV range at IV_MAX, whereas the correct
2042 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2043 cases go to UV */
cab190d4
JD
2044#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2045 if (Perl_isnan(SvNVX(sv))) {
2046 SvUV_set(sv, 0);
2047 SvIsUV_on(sv);
fdbe6d7c 2048 return FALSE;
cab190d4 2049 }
cab190d4 2050#endif
28e5dec8 2051 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2052 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2053 if (SvNVX(sv) == (NV) SvIVX(sv)
2054#ifndef NV_PRESERVES_UV
2055 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2056 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2057 /* Don't flag it as "accurately an integer" if the number
2058 came from a (by definition imprecise) NV operation, and
2059 we're outside the range of NV integer precision */
2060#endif
2061 ) {
a43d94f2
NC
2062 if (SvNOK(sv))
2063 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2064 else {
2065 /* scalar has trailing garbage, eg "42a" */
2066 }
28e5dec8 2067 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2068 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2069 PTR2UV(sv),
2070 SvNVX(sv),
2071 SvIVX(sv)));
2072
2073 } else {
2074 /* IV not precise. No need to convert from PV, as NV
2075 conversion would already have cached IV if it detected
2076 that PV->IV would be better than PV->NV->IV
2077 flags already correct - don't set public IOK. */
2078 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2079 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2080 PTR2UV(sv),
2081 SvNVX(sv),
2082 SvIVX(sv)));
2083 }
2084 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2085 but the cast (NV)IV_MIN rounds to a the value less (more
2086 negative) than IV_MIN which happens to be equal to SvNVX ??
2087 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2088 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2089 (NV)UVX == NVX are both true, but the values differ. :-(
2090 Hopefully for 2s complement IV_MIN is something like
2091 0x8000000000000000 which will be exact. NWC */
d460ef45 2092 }
25da4f38 2093 else {
607fa7f2 2094 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2095 if (
2096 (SvNVX(sv) == (NV) SvUVX(sv))
2097#ifndef NV_PRESERVES_UV
2098 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2099 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2100 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2101 /* Don't flag it as "accurately an integer" if the number
2102 came from a (by definition imprecise) NV operation, and
2103 we're outside the range of NV integer precision */
2104#endif
a43d94f2 2105 && SvNOK(sv)
28e5dec8
JH
2106 )
2107 SvIOK_on(sv);
25da4f38 2108 SvIsUV_on(sv);
1c846c1f 2109 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2110 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2111 PTR2UV(sv),
57def98f
JH
2112 SvUVX(sv),
2113 SvUVX(sv)));
25da4f38 2114 }
748a9306
LW
2115 }
2116 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2117 UV value;
504618e9 2118 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2119 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2120 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2121 the same as the direct translation of the initial string
2122 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2123 be careful to ensure that the value with the .456 is around if the
2124 NV value is requested in the future).
1c846c1f 2125
af359546 2126 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2127 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2128 cache the NV if we are sure it's not needed.
25da4f38 2129 */
16b7a9a4 2130
c2988b20
NC
2131 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2132 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2133 == IS_NUMBER_IN_UV) {
5e045b90 2134 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2135 if (SvTYPE(sv) < SVt_PVIV)
2136 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2137 (void)SvIOK_on(sv);
c2988b20
NC
2138 } else if (SvTYPE(sv) < SVt_PVNV)
2139 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2140
f2524eef 2141 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2142 we aren't going to call atof() below. If NVs don't preserve UVs
2143 then the value returned may have more precision than atof() will
2144 return, even though value isn't perfectly accurate. */
2145 if ((numtype & (IS_NUMBER_IN_UV
2146#ifdef NV_PRESERVES_UV
2147 | IS_NUMBER_NOT_INT
2148#endif
2149 )) == IS_NUMBER_IN_UV) {
2150 /* This won't turn off the public IOK flag if it was set above */
2151 (void)SvIOKp_on(sv);
2152
2153 if (!(numtype & IS_NUMBER_NEG)) {
2154 /* positive */;
2155 if (value <= (UV)IV_MAX) {
45977657 2156 SvIV_set(sv, (IV)value);
c2988b20 2157 } else {
af359546 2158 /* it didn't overflow, and it was positive. */
607fa7f2 2159 SvUV_set(sv, value);
c2988b20
NC
2160 SvIsUV_on(sv);
2161 }
2162 } else {
2163 /* 2s complement assumption */
2164 if (value <= (UV)IV_MIN) {
45977657 2165 SvIV_set(sv, -(IV)value);
c2988b20
NC
2166 } else {
2167 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2168 I'm assuming it will be rare. */
c2988b20
NC
2169 if (SvTYPE(sv) < SVt_PVNV)
2170 sv_upgrade(sv, SVt_PVNV);
2171 SvNOK_on(sv);
2172 SvIOK_off(sv);
2173 SvIOKp_on(sv);
9d6ce603 2174 SvNV_set(sv, -(NV)value);
45977657 2175 SvIV_set(sv, IV_MIN);
c2988b20
NC
2176 }
2177 }
2178 }
2179 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2180 will be in the previous block to set the IV slot, and the next
2181 block to set the NV slot. So no else here. */
2182
2183 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2184 != IS_NUMBER_IN_UV) {
2185 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2186 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2187
c2988b20
NC
2188 if (! numtype && ckWARN(WARN_NUMERIC))
2189 not_a_number(sv);
28e5dec8 2190
65202027 2191#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2192 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2193 PTR2UV(sv), SvNVX(sv)));
65202027 2194#else
1779d84d 2195 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2196 PTR2UV(sv), SvNVX(sv)));
65202027 2197#endif
28e5dec8 2198
28e5dec8 2199#ifdef NV_PRESERVES_UV
af359546
NC
2200 (void)SvIOKp_on(sv);
2201 (void)SvNOK_on(sv);
2202 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2203 SvIV_set(sv, I_V(SvNVX(sv)));
2204 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2205 SvIOK_on(sv);
2206 } else {
6f207bd3 2207 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2208 }
2209 /* UV will not work better than IV */
2210 } else {
2211 if (SvNVX(sv) > (NV)UV_MAX) {
2212 SvIsUV_on(sv);
2213 /* Integer is inaccurate. NOK, IOKp, is UV */
2214 SvUV_set(sv, UV_MAX);
af359546
NC
2215 } else {
2216 SvUV_set(sv, U_V(SvNVX(sv)));
2217 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2218 NV preservse UV so can do correct comparison. */
2219 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2220 SvIOK_on(sv);
af359546 2221 } else {
6f207bd3 2222 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2223 }
2224 }
4b0c9573 2225 SvIsUV_on(sv);
af359546 2226 }
28e5dec8 2227#else /* NV_PRESERVES_UV */
c2988b20
NC
2228 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2229 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2230 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2231 grok_number above. The NV slot has just been set using
2232 Atof. */
560b0c46 2233 SvNOK_on(sv);
c2988b20
NC
2234 assert (SvIOKp(sv));
2235 } else {
2236 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2237 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2238 /* Small enough to preserve all bits. */
2239 (void)SvIOKp_on(sv);
2240 SvNOK_on(sv);
45977657 2241 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2242 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2243 SvIOK_on(sv);
2244 /* Assumption: first non-preserved integer is < IV_MAX,
2245 this NV is in the preserved range, therefore: */
2246 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2247 < (UV)IV_MAX)) {
32fdb065 2248 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
c2988b20
NC
2249 }
2250 } else {
2251 /* IN_UV NOT_INT
2252 0 0 already failed to read UV.
2253 0 1 already failed to read UV.
2254 1 0 you won't get here in this case. IV/UV
2255 slot set, public IOK, Atof() unneeded.
2256 1 1 already read UV.
2257 so there's no point in sv_2iuv_non_preserve() attempting
2258 to use atol, strtol, strtoul etc. */
47031da6 2259# ifdef DEBUGGING
40a17c4c 2260 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2261# else
2262 sv_2iuv_non_preserve (sv);
2263# endif
c2988b20
NC
2264 }
2265 }
28e5dec8 2266#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2267 /* It might be more code efficient to go through the entire logic above
2268 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2269 gets complex and potentially buggy, so more programmer efficient
2270 to do it this way, by turning off the public flags: */
2271 if (!numtype)
2272 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2273 }
af359546
NC
2274 }
2275 else {
675c862f 2276 if (isGV_with_GP(sv))
159b6efe 2277 return glob_2number(MUTABLE_GV(sv));
180488f8 2278
af359546
NC
2279 if (!(SvFLAGS(sv) & SVs_PADTMP)) {
2280 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2281 report_uninit(sv);
2282 }
25da4f38
IZ
2283 if (SvTYPE(sv) < SVt_IV)
2284 /* Typically the caller expects that sv_any is not NULL now. */
2285 sv_upgrade(sv, SVt_IV);
af359546
NC
2286 /* Return 0 from the caller. */
2287 return TRUE;
2288 }
2289 return FALSE;
2290}
2291
2292/*
2293=for apidoc sv_2iv_flags
2294
2295Return the integer value of an SV, doing any necessary string
2296conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2297Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2298
2299=cut
2300*/
2301
2302IV
5de3775c 2303Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2304{
97aff369 2305 dVAR;
af359546 2306 if (!sv)
a0d0e21e 2307 return 0;
cecf5685
NC
2308 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2309 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e
NC
2310 cache IVs just in case. In practice it seems that they never
2311 actually anywhere accessible by user Perl code, let alone get used
2312 in anything other than a string context. */
af359546
NC
2313 if (flags & SV_GMAGIC)
2314 mg_get(sv);
2315 if (SvIOKp(sv))
2316 return SvIVX(sv);
2317 if (SvNOKp(sv)) {
2318 return I_V(SvNVX(sv));
2319 }
71c558c3
NC
2320 if (SvPOKp(sv) && SvLEN(sv)) {
2321 UV value;
2322 const int numtype
2323 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2324
2325 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2326 == IS_NUMBER_IN_UV) {
2327 /* It's definitely an integer */
2328 if (numtype & IS_NUMBER_NEG) {
2329 if (value < (UV)IV_MIN)
2330 return -(IV)value;
2331 } else {
2332 if (value < (UV)IV_MAX)
2333 return (IV)value;
2334 }
2335 }
2336 if (!numtype) {
2337 if (ckWARN(WARN_NUMERIC))
2338 not_a_number(sv);
2339 }
2340 return I_V(Atof(SvPVX_const(sv)));
2341 }
1c7ff15e
NC
2342 if (SvROK(sv)) {
2343 goto return_rok;
af359546 2344 }
1c7ff15e
NC
2345 assert(SvTYPE(sv) >= SVt_PVMG);
2346 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2347 } else if (SvTHINKFIRST(sv)) {
af359546 2348 if (SvROK(sv)) {
1c7ff15e 2349 return_rok:
af359546
NC
2350 if (SvAMAGIC(sv)) {
2351 SV * const tmpstr=AMG_CALLun(sv,numer);
2352 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2353 return SvIV(tmpstr);
2354 }
2355 }
2356 return PTR2IV(SvRV(sv));
2357 }
2358 if (SvIsCOW(sv)) {
2359 sv_force_normal_flags(sv, 0);
2360 }
2361 if (SvREADONLY(sv) && !SvOK(sv)) {
2362 if (ckWARN(WARN_UNINITIALIZED))
2363 report_uninit(sv);
2364 return 0;
2365 }
2366 }
2367 if (!SvIOKp(sv)) {
2368 if (S_sv_2iuv_common(aTHX_ sv))
2369 return 0;
79072805 2370 }
1d7c1841
GS
2371 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2372 PTR2UV(sv),SvIVX(sv)));
25da4f38 2373 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2374}
2375
645c22ef 2376/*
891f9566 2377=for apidoc sv_2uv_flags
645c22ef
DM
2378
2379Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2380conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2381Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2382
2383=cut
2384*/
2385
ff68c719 2386UV
5de3775c 2387Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2388{
97aff369 2389 dVAR;
ff68c719
PP
2390 if (!sv)
2391 return 0;
cecf5685
NC
2392 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2393 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2394 cache IVs just in case. */
891f9566
YST
2395 if (flags & SV_GMAGIC)
2396 mg_get(sv);
ff68c719
PP
2397 if (SvIOKp(sv))
2398 return SvUVX(sv);
2399 if (SvNOKp(sv))
2400 return U_V(SvNVX(sv));
71c558c3
NC
2401 if (SvPOKp(sv) && SvLEN(sv)) {
2402 UV value;
2403 const int numtype
2404 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2405
2406 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2407 == IS_NUMBER_IN_UV) {
2408 /* It's definitely an integer */
2409 if (!(numtype & IS_NUMBER_NEG))
2410 return value;
2411 }
2412 if (!numtype) {
2413 if (ckWARN(WARN_NUMERIC))
2414 not_a_number(sv);
2415 }
2416 return U_V(Atof(SvPVX_const(sv)));
2417 }
1c7ff15e
NC
2418 if (SvROK(sv)) {
2419 goto return_rok;
3fe9a6f1 2420 }
1c7ff15e
NC
2421 assert(SvTYPE(sv) >= SVt_PVMG);
2422 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2423 } else if (SvTHINKFIRST(sv)) {
ff68c719 2424 if (SvROK(sv)) {
1c7ff15e 2425 return_rok:
deb46114
NC
2426 if (SvAMAGIC(sv)) {
2427 SV *const tmpstr = AMG_CALLun(sv,numer);
2428 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2429 return SvUV(tmpstr);
2430 }
2431 }
2432 return PTR2UV(SvRV(sv));
ff68c719 2433 }
765f542d
NC
2434 if (SvIsCOW(sv)) {
2435 sv_force_normal_flags(sv, 0);
8a818333 2436 }
0336b60e 2437 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2438 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2439 report_uninit(sv);
ff68c719
PP
2440 return 0;
2441 }
2442 }
af359546
NC
2443 if (!SvIOKp(sv)) {
2444 if (S_sv_2iuv_common(aTHX_ sv))
2445 return 0;
ff68c719 2446 }
25da4f38 2447
1d7c1841
GS
2448 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2449 PTR2UV(sv),SvUVX(sv)));
25da4f38 2450 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2451}
2452
645c22ef
DM
2453/*
2454=for apidoc sv_2nv
2455
2456Return the num value of an SV, doing any necessary string or integer
2457conversion, magic etc. Normally used via the C<SvNV(sv)> and C<SvNVx(sv)>
2458macros.
2459
2460=cut
2461*/
2462
65202027 2463NV
5de3775c 2464Perl_sv_2nv(pTHX_ register SV *const sv)
79072805 2465{
97aff369 2466 dVAR;
79072805
LW
2467 if (!sv)
2468 return 0.0;
cecf5685
NC
2469 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2470 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2471 cache IVs just in case. */
463ee0b2
LW
2472 mg_get(sv);
2473 if (SvNOKp(sv))
2474 return SvNVX(sv);
0aa395f8 2475 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
041457d9 2476 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
504618e9 2477 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
a0d0e21e 2478 not_a_number(sv);
3f7c398e 2479 return Atof(SvPVX_const(sv));
a0d0e21e 2480 }
25da4f38 2481 if (SvIOKp(sv)) {
1c846c1f 2482 if (SvIsUV(sv))
65202027 2483 return (NV)SvUVX(sv);
25da4f38 2484 else
65202027 2485 return (NV)SvIVX(sv);
47a72cb8
NC
2486 }
2487 if (SvROK(sv)) {
2488 goto return_rok;
2489 }
2490 assert(SvTYPE(sv) >= SVt_PVMG);
2491 /* This falls through to the report_uninit near the end of the
2492 function. */
2493 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2494 if (SvROK(sv)) {
47a72cb8 2495 return_rok:
deb46114
NC
2496 if (SvAMAGIC(sv)) {
2497 SV *const tmpstr = AMG_CALLun(sv,numer);
2498 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2499 return SvNV(tmpstr);
2500 }
2501 }
2502 return PTR2NV(SvRV(sv));
a0d0e21e 2503 }
765f542d
NC
2504 if (SvIsCOW(sv)) {
2505 sv_force_normal_flags(sv, 0);
8a818333 2506 }
0336b60e 2507 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2508 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2509 report_uninit(sv);
ed6116ce
LW
2510 return 0.0;
2511 }
79072805
LW
2512 }
2513 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2514 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2515 sv_upgrade(sv, SVt_NV);
906f284f 2516#ifdef USE_LONG_DOUBLE
097ee67d 2517 DEBUG_c({
f93f4e46 2518 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2519 PerlIO_printf(Perl_debug_log,
2520 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2521 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2522 RESTORE_NUMERIC_LOCAL();
2523 });
65202027 2524#else
572bbb43 2525 DEBUG_c({
f93f4e46 2526 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2527 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2528 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2529 RESTORE_NUMERIC_LOCAL();
2530 });
572bbb43 2531#endif
79072805
LW
2532 }
2533 else if (SvTYPE(sv) < SVt_PVNV)
2534 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2535 if (SvNOKp(sv)) {
2536 return SvNVX(sv);
61604483 2537 }
59d8ce62 2538 if (SvIOKp(sv)) {
9d6ce603 2539 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2540#ifdef NV_PRESERVES_UV
a43d94f2
NC
2541 if (SvIOK(sv))
2542 SvNOK_on(sv);
2543 else
2544 SvNOKp_on(sv);
28e5dec8
JH
2545#else
2546 /* Only set the public NV OK flag if this NV preserves the IV */
2547 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2548 if (SvIOK(sv) &&
2549 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2550 : (SvIVX(sv) == I_V(SvNVX(sv))))
2551 SvNOK_on(sv);
2552 else
2553 SvNOKp_on(sv);
2554#endif
93a17b20 2555 }
748a9306 2556 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2557 UV value;
3f7c398e 2558 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2559 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2560 not_a_number(sv);
28e5dec8 2561#ifdef NV_PRESERVES_UV
c2988b20
NC
2562 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2563 == IS_NUMBER_IN_UV) {
5e045b90 2564 /* It's definitely an integer */
9d6ce603 2565 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2566 } else
3f7c398e 2567 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2568 if (numtype)
2569 SvNOK_on(sv);
2570 else
2571 SvNOKp_on(sv);
28e5dec8 2572#else
3f7c398e 2573 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2574 /* Only set the public NV OK flag if this NV preserves the value in
2575 the PV at least as well as an IV/UV would.
2576 Not sure how to do this 100% reliably. */
2577 /* if that shift count is out of range then Configure's test is
2578 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2579 UV_BITS */
2580 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2581 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2582 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2583 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2584 /* Can't use strtol etc to convert this string, so don't try.
2585 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2586 SvNOK_on(sv);
2587 } else {
2588 /* value has been set. It may not be precise. */
2589 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2590 /* 2s complement assumption for (UV)IV_MIN */
2591 SvNOK_on(sv); /* Integer is too negative. */
2592 } else {
2593 SvNOKp_on(sv);
2594 SvIOKp_on(sv);
6fa402ec 2595
c2988b20 2596 if (numtype & IS_NUMBER_NEG) {
45977657 2597 SvIV_set(sv, -(IV)value);
c2988b20 2598 } else if (value <= (UV)IV_MAX) {
45977657 2599 SvIV_set(sv, (IV)value);
c2988b20 2600 } else {
607fa7f2 2601 SvUV_set(sv, value);
c2988b20
NC
2602 SvIsUV_on(sv);
2603 }
2604
2605 if (numtype & IS_NUMBER_NOT_INT) {
2606 /* I believe that even if the original PV had decimals,
2607 they are lost beyond the limit of the FP precision.
2608 However, neither is canonical, so both only get p
2609 flags. NWC, 2000/11/25 */
2610 /* Both already have p flags, so do nothing */
2611 } else {
66a1b24b 2612 const NV nv = SvNVX(sv);
c2988b20
NC
2613 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2614 if (SvIVX(sv) == I_V(nv)) {
2615 SvNOK_on(sv);
c2988b20 2616 } else {
c2988b20
NC
2617 /* It had no "." so it must be integer. */
2618 }
00b6aa41 2619 SvIOK_on(sv);
c2988b20
NC
2620 } else {
2621 /* between IV_MAX and NV(UV_MAX).
2622 Could be slightly > UV_MAX */
6fa402ec 2623
c2988b20
NC
2624 if (numtype & IS_NUMBER_NOT_INT) {
2625 /* UV and NV both imprecise. */
2626 } else {
66a1b24b 2627 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2628
2629 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2630 SvNOK_on(sv);
c2988b20 2631 }
00b6aa41 2632 SvIOK_on(sv);
c2988b20
NC
2633 }
2634 }
2635 }
2636 }
2637 }
a43d94f2
NC
2638 /* It might be more code efficient to go through the entire logic above
2639 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2640 gets complex and potentially buggy, so more programmer efficient
2641 to do it this way, by turning off the public flags: */
2642 if (!numtype)
2643 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2644#endif /* NV_PRESERVES_UV */
93a17b20 2645 }
79072805 2646 else {
f7877b28 2647 if (isGV_with_GP(sv)) {
159b6efe 2648 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2649 return 0.0;
2650 }
2651
041457d9 2652 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2653 report_uninit(sv);
7e25a7e9
NC
2654 assert (SvTYPE(sv) >= SVt_NV);
2655 /* Typically the caller expects that sv_any is not NULL now. */
2656 /* XXX Ilya implies that this is a bug in callers that assume this
2657 and ideally should be fixed. */
a0d0e21e 2658 return 0.0;
79072805 2659 }
572bbb43 2660#if defined(USE_LONG_DOUBLE)
097ee67d 2661 DEBUG_c({
f93f4e46 2662 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2663 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2664 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2665 RESTORE_NUMERIC_LOCAL();
2666 });
65202027 2667#else
572bbb43 2668 DEBUG_c({
f93f4e46 2669 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2670 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2671 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2672 RESTORE_NUMERIC_LOCAL();
2673 });
572bbb43 2674#endif
463ee0b2 2675 return SvNVX(sv);
79072805
LW
2676}
2677
800401ee
JH
2678/*
2679=for apidoc sv_2num
2680
2681Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2682reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2683access this function.
800401ee
JH
2684
2685=cut
2686*/
2687
2688SV *
5de3775c 2689Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2690{
7918f24d
NC
2691 PERL_ARGS_ASSERT_SV_2NUM;
2692
b9ee0594
RGS
2693 if (!SvROK(sv))
2694 return sv;
800401ee
JH
2695 if (SvAMAGIC(sv)) {
2696 SV * const tmpsv = AMG_CALLun(sv,numer);
2697 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2698 return sv_2num(tmpsv);
2699 }
2700 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2701}
2702
645c22ef
DM
2703/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2704 * UV as a string towards the end of buf, and return pointers to start and
2705 * end of it.
2706 *
2707 * We assume that buf is at least TYPE_CHARS(UV) long.
2708 */
2709
864dbfa3 2710static char *
5de3775c 2711S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2712{
25da4f38 2713 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2714 char * const ebuf = ptr;
25da4f38 2715 int sign;
25da4f38 2716
7918f24d
NC
2717 PERL_ARGS_ASSERT_UIV_2BUF;
2718
25da4f38
IZ
2719 if (is_uv)
2720 sign = 0;
2721 else if (iv >= 0) {
2722 uv = iv;
2723 sign = 0;
2724 } else {
2725 uv = -iv;
2726 sign = 1;
2727 }
2728 do {
eb160463 2729 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2730 } while (uv /= 10);
2731 if (sign)
2732 *--ptr = '-';
2733 *peob = ebuf;
2734 return ptr;
2735}
2736
645c22ef
DM
2737/*
2738=for apidoc sv_2pv_flags
2739
ff276b08 2740Returns a pointer to the string value of an SV, and sets *lp to its length.
645c22ef
DM
2741If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2742if necessary.
2743Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2744usually end up here too.
2745
2746=cut
2747*/
2748
8d6d96c1 2749char *
5de3775c 2750Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2751{
97aff369 2752 dVAR;
79072805 2753 register char *s;
79072805 2754
463ee0b2 2755 if (!sv) {
cdb061a3
NC
2756 if (lp)
2757 *lp = 0;
73d840c0 2758 return (char *)"";
463ee0b2 2759 }
8990e307 2760 if (SvGMAGICAL(sv)) {
8d6d96c1
HS
2761 if (flags & SV_GMAGIC)
2762 mg_get(sv);
463ee0b2 2763 if (SvPOKp(sv)) {
cdb061a3
NC
2764 if (lp)
2765 *lp = SvCUR(sv);
10516c54
NC
2766 if (flags & SV_MUTABLE_RETURN)
2767 return SvPVX_mutable(sv);
4d84ee25
NC
2768 if (flags & SV_CONST_RETURN)
2769 return (char *)SvPVX_const(sv);
463ee0b2
LW
2770 return SvPVX(sv);
2771 }
75dfc8ec
NC
2772 if (SvIOKp(sv) || SvNOKp(sv)) {
2773 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
75dfc8ec
NC
2774 STRLEN len;
2775
2776 if (SvIOKp(sv)) {
e80fed9d 2777 len = SvIsUV(sv)
d9fad198
JH
2778 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2779 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
75dfc8ec 2780 } else {
e8ada2d0
NC
2781 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2782 len = strlen(tbuf);
75dfc8ec 2783 }
b5b886f0
NC
2784 assert(!SvROK(sv));
2785 {
75dfc8ec
NC
2786 dVAR;
2787
2788#ifdef FIXNEGATIVEZERO
e8ada2d0
NC
2789 if (len == 2 && tbuf[0] == '-' && tbuf[1] == '0') {
2790 tbuf[0] = '0';
2791 tbuf[1] = 0;
75dfc8ec
NC
2792 len = 1;
2793 }
2794#endif
2795 SvUPGRADE(sv, SVt_PV);
2796 if (lp)
2797 *lp = len;
2798 s = SvGROW_mutable(sv, len + 1);
2799 SvCUR_set(sv, len);
2800 SvPOKp_on(sv);
10edeb5d 2801 return (char*)memcpy(s, tbuf, len + 1);
75dfc8ec 2802 }
463ee0b2 2803 }
1c7ff15e
NC
2804 if (SvROK(sv)) {
2805 goto return_rok;
2806 }
2807 assert(SvTYPE(sv) >= SVt_PVMG);
2808 /* This falls through to the report_uninit near the end of the
2809 function. */
2810 } else if (SvTHINKFIRST(sv)) {
ed6116ce 2811 if (SvROK(sv)) {
1c7ff15e 2812 return_rok:
deb46114
NC
2813 if (SvAMAGIC(sv)) {
2814 SV *const tmpstr = AMG_CALLun(sv,string);
2815 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2816 /* Unwrap this: */
2817 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2818 */
2819
2820 char *pv;
2821 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2822 if (flags & SV_CONST_RETURN) {
2823 pv = (char *) SvPVX_const(tmpstr);
2824 } else {
2825 pv = (flags & SV_MUTABLE_RETURN)
2826 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2827 }
2828 if (lp)
2829 *lp = SvCUR(tmpstr);
50adf7d2 2830 } else {
deb46114 2831 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2832 }
deb46114
NC
2833 if (SvUTF8(tmpstr))
2834 SvUTF8_on(sv);
2835 else
2836 SvUTF8_off(sv);
2837 return pv;
50adf7d2 2838 }
deb46114
NC
2839 }
2840 {
fafee734
NC
2841 STRLEN len;
2842 char *retval;
2843 char *buffer;
d2c6dc5e 2844 SV *const referent = SvRV(sv);
d8eae41e
NC
2845
2846 if (!referent) {
fafee734
NC
2847 len = 7;
2848 retval = buffer = savepvn("NULLREF", len);
5c35adbb 2849 } else if (SvTYPE(referent) == SVt_REGEXP) {
d2c6dc5e 2850 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
67d2d14d
AB
2851 I32 seen_evals = 0;
2852
2853 assert(re);
2854
2855 /* If the regex is UTF-8 we want the containing scalar to
2856 have an UTF-8 flag too */
2857 if (RX_UTF8(re))
2858 SvUTF8_on(sv);
2859 else
2860 SvUTF8_off(sv);
2861
2862 if ((seen_evals = RX_SEEN_EVALS(re)))
2863 PL_reginterp_cnt += seen_evals;
2864
2865 if (lp)
2866 *lp = RX_WRAPLEN(re);
2867
2868 return RX_WRAPPED(re);
d8eae41e
NC
2869 } else {
2870 const char *const typestr = sv_reftype(referent, 0);
fafee734
NC
2871 const STRLEN typelen = strlen(typestr);
2872 UV addr = PTR2UV(referent);
2873 const char *stashname = NULL;
2874 STRLEN stashnamelen = 0; /* hush, gcc */
2875 const char *buffer_end;
d8eae41e 2876
d8eae41e 2877 if (SvOBJECT(referent)) {
fafee734
NC
2878 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2879
2880 if (name) {
2881 stashname = HEK_KEY(name);
2882 stashnamelen = HEK_LEN(name);
2883
2884 if (HEK_UTF8(name)) {
2885 SvUTF8_on(sv);
2886 } else {
2887 SvUTF8_off(sv);
2888 }
2889 } else {
2890 stashname = "__ANON__";
2891 stashnamelen = 8;
2892 }
2893 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2894 + 2 * sizeof(UV) + 2 /* )\0 */;
2895 } else {
2896 len = typelen + 3 /* (0x */
2897 + 2 * sizeof(UV) + 2 /* )\0 */;
d8eae41e 2898 }
fafee734
NC
2899
2900 Newx(buffer, len, char);
2901 buffer_end = retval = buffer + len;
2902
2903 /* Working backwards */
2904 *--retval = '\0';
2905 *--retval = ')';
2906 do {
2907 *--retval = PL_hexdigit[addr & 15];
2908 } while (addr >>= 4);
2909 *--retval = 'x';
2910 *--retval = '0';
2911 *--retval = '(';
2912
2913 retval -= typelen;
2914 memcpy(retval, typestr, typelen);
2915
2916 if (stashname) {
2917 *--retval = '=';
2918 retval -= stashnamelen;
2919 memcpy(retval, stashname, stashnamelen);
2920 }
2921 /* retval may not neccesarily have reached the start of the
2922 buffer here. */
2923 assert (retval >= buffer);
2924
2925 len = buffer_end - retval - 1; /* -1 for that \0 */
c080367d 2926 }
042dae7a 2927 if (lp)
fafee734
NC
2928 *lp = len;
2929 SAVEFREEPV(buffer);
2930 return retval;
463ee0b2 2931 }
79072805 2932 }
0336b60e 2933 if (SvREADONLY(sv) && !SvOK(sv)) {
cdb061a3
NC
2934 if (lp)
2935 *lp = 0;
9f621bb0
NC
2936 if (flags & SV_UNDEF_RETURNS_NULL)
2937 return NULL;
2938 if (ckWARN(WARN_UNINITIALIZED))
2939 report_uninit(sv);
73d840c0 2940 return (char *)"";
79072805 2941 }
79072805 2942 }
28e5dec8
JH
2943 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2944 /* I'm assuming that if both IV and NV are equally valid then
2945 converting the IV is going to be more efficient */
e1ec3a88 2946 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2947 char buf[TYPE_CHARS(UV)];
2948 char *ebuf, *ptr;
97a130b8 2949 STRLEN len;
28e5dec8
JH
2950
2951 if (SvTYPE(sv) < SVt_PVIV)
2952 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2953 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2954 len = ebuf - ptr;
5902b6a9 2955 /* inlined from sv_setpvn */
97a130b8
NC
2956 s = SvGROW_mutable(sv, len + 1);
2957 Move(ptr, s, len, char);
2958 s += len;
28e5dec8 2959 *s = '\0';
28e5dec8
JH
2960 }
2961 else if (SvNOKp(sv)) {
4ee39169 2962 dSAVE_ERRNO;
79072805
LW
2963 if (SvTYPE(sv) < SVt_PVNV)
2964 sv_upgrade(sv, SVt_PVNV);
1c846c1f 2965 /* The +20 is pure guesswork. Configure test needed. --jhi */
5902b6a9 2966 s = SvGROW_mutable(sv, NV_DIG + 20);
c81271c3 2967 /* some Xenix systems wipe out errno here */
79072805 2968#ifdef apollo
463ee0b2 2969 if (SvNVX(sv) == 0.0)
d1307786 2970 my_strlcpy(s, "0", SvLEN(sv));
79072805
LW
2971 else
2972#endif /*apollo*/
bbce6d69 2973 {
2d4389e4 2974 Gconvert(SvNVX(sv), NV_DIG, 0, s);
bbce6d69 2975 }
4ee39169 2976 RESTORE_ERRNO;
a0d0e21e 2977#ifdef FIXNEGATIVEZERO
20773dcd
NC
2978 if (*s == '-' && s[1] == '0' && !s[2]) {
2979 s[0] = '0';
2980 s[1] = 0;
2981 }
a0d0e21e 2982#endif
79072805
LW
2983 while (*s) s++;
2984#ifdef hcx
2985 if (s[-1] == '.')
46fc3d4c 2986 *--s = '\0';
79072805
LW
2987#endif
2988 }
79072805 2989 else {
675c862f 2990 if (isGV_with_GP(sv))
159b6efe 2991 return glob_2pv(MUTABLE_GV(sv), lp);
180488f8 2992
cdb061a3 2993 if (lp)
00b6aa41 2994 *lp = 0;
9f621bb0
NC
2995 if (flags & SV_UNDEF_RETURNS_NULL)
2996 return NULL;
2997 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
2998 report_uninit(sv);
25da4f38
IZ
2999 if (SvTYPE(sv) < SVt_PV)
3000 /* Typically the caller expects that sv_any is not NULL now. */
3001 sv_upgrade(sv, SVt_PV);
73d840c0 3002 return (char *)"";
79072805 3003 }
cdb061a3 3004 {
823a54a3 3005 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3006 if (lp)
3007 *lp = len;
3008 SvCUR_set(sv, len);
3009 }
79072805 3010 SvPOK_on(sv);
1d7c1841 3011 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3012 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3013 if (flags & SV_CONST_RETURN)
3014 return (char *)SvPVX_const(sv);
10516c54
NC
3015 if (flags & SV_MUTABLE_RETURN)
3016 return SvPVX_mutable(sv);
463ee0b2
LW
3017 return SvPVX(sv);
3018}
3019
645c22ef 3020/*
6050d10e
JP
3021=for apidoc sv_copypv
3022
3023Copies a stringified representation of the source SV into the
3024destination SV. Automatically performs any necessary mg_get and
54f0641b 3025coercion of numeric values into strings. Guaranteed to preserve
2575c402 3026UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3027sv_2pv[_flags] but operates directly on an SV instead of just the
3028string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3029would lose the UTF-8'ness of the PV.
3030
3031=cut
3032*/
3033
3034void
5de3775c 3035Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 3036{
446eaa42 3037 STRLEN len;
53c1dcc0 3038 const char * const s = SvPV_const(ssv,len);
7918f24d
NC
3039
3040 PERL_ARGS_ASSERT_SV_COPYPV;
3041
cb50f42d 3042 sv_setpvn(dsv,s,len);
446eaa42 3043 if (SvUTF8(ssv))
cb50f42d 3044 SvUTF8_on(dsv);
446eaa42 3045 else
cb50f42d 3046 SvUTF8_off(dsv);
6050d10e
JP
3047}
3048
3049/*
645c22ef
DM
3050=for apidoc sv_2pvbyte
3051
3052Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3053to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3054side-effect.
3055
3056Usually accessed via the C<SvPVbyte> macro.
3057
3058=cut
3059*/
3060
7340a771 3061char *
5de3775c 3062Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3063{
7918f24d
NC
3064 PERL_ARGS_ASSERT_SV_2PVBYTE;
3065
0875d2fe 3066 sv_utf8_downgrade(sv,0);
97972285 3067 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771
GS
3068}
3069
645c22ef 3070/*
035cbb0e
RGS
3071=for apidoc sv_2pvutf8
3072
3073Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3074to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3075
3076Usually accessed via the C<SvPVutf8> macro.
3077
3078=cut
3079*/
645c22ef 3080
7340a771 3081char *
7bc54cea 3082Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3083{
7918f24d
NC
3084 PERL_ARGS_ASSERT_SV_2PVUTF8;
3085
035cbb0e
RGS
3086 sv_utf8_upgrade(sv);
3087 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771 3088}
1c846c1f 3089
7ee2227d 3090
645c22ef
DM
3091/*
3092=for apidoc sv_2bool
3093
3094This function is only called on magical items, and is only used by
8cf8f3d1 3095sv_true() or its macro equivalent.
645c22ef
DM
3096
3097=cut
3098*/
3099
463ee0b2 3100bool
7bc54cea 3101Perl_sv_2bool(pTHX_ register SV *const sv)
463ee0b2 3102{
97aff369 3103 dVAR;
7918f24d
NC
3104
3105 PERL_ARGS_ASSERT_SV_2BOOL;
3106
5b295bef 3107 SvGETMAGIC(sv);
463ee0b2 3108
a0d0e21e
LW
3109 if (!SvOK(sv))
3110 return 0;
3111 if (SvROK(sv)) {
fabdb6c0
AL
3112 if (SvAMAGIC(sv)) {
3113 SV * const tmpsv = AMG_CALLun(sv,bool_);
3114 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3115 return (bool)SvTRUE(tmpsv);
3116 }
3117 return SvRV(sv) != 0;
a0d0e21e 3118 }
463ee0b2 3119 if (SvPOKp(sv)) {
53c1dcc0
AL
3120 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3121 if (Xpvtmp &&
339049b0 3122 (*sv->sv_u.svu_pv > '0' ||
11343788 3123 Xpvtmp->xpv_cur > 1 ||
339049b0 3124 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
463ee0b2
LW
3125 return 1;
3126 else
3127 return 0;
3128 }
3129 else {
3130 if (SvIOKp(sv))
3131 return SvIVX(sv) != 0;
3132 else {
3133 if (SvNOKp(sv))
3134 return SvNVX(sv) != 0.0;
180488f8 3135 else {
f7877b28 3136 if (isGV_with_GP(sv))
180488f8
NC
3137 return TRUE;
3138 else
3139 return FALSE;
3140 }
463ee0b2
LW
3141 }
3142 }
79072805
LW
3143}
3144
c461cf8f
JH
3145/*
3146=for apidoc sv_utf8_upgrade
3147
78ea37eb 3148Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3149Forces the SV to string form if it is not already.
2bbc8d55 3150Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3151Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3152if the whole string is the same in UTF-8 as not.
3153Returns the number of bytes in the converted string
c461cf8f 3154
13a6c0e0
JH
3155This is not as a general purpose byte encoding to Unicode interface:
3156use the Encode extension for that.
3157
fe749c9a
KW
3158=for apidoc sv_utf8_upgrade_nomg
3159
3160Like sv_utf8_upgrade, but doesn't do magic on C<sv>
3161
8d6d96c1
HS
3162=for apidoc sv_utf8_upgrade_flags
3163
78ea37eb 3164Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3165Forces the SV to string form if it is not already.
8d6d96c1 3166Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3167if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3168will C<mg_get> on C<sv> if appropriate, else not.
3169Returns the number of bytes in the converted string
3170C<sv_utf8_upgrade> and
8d6d96c1
HS
3171C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3172
13a6c0e0
JH
3173This is not as a general purpose byte encoding to Unicode interface:
3174use the Encode extension for that.
3175
8d6d96c1 3176=cut
b3ab6785
KW
3177
3178The grow version is currently not externally documented. It adds a parameter,
3179extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3180have free after it upon return. This allows the caller to reserve extra space
3181that it intends to fill, to avoid extra grows.
3182
3183Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3184which can be used to tell this function to not first check to see if there are
3185any characters that are different in UTF-8 (variant characters) which would
3186force it to allocate a new string to sv, but to assume there are. Typically
3187this flag is used by a routine that has already parsed the string to find that
3188there are such characters, and passes this information on so that the work
3189doesn't have to be repeated.
3190
3191(One might think that the calling routine could pass in the position of the
3192first such variant, so it wouldn't have to be found again. But that is not the
3193case, because typically when the caller is likely to use this flag, it won't be
3194calling this routine unless it finds something that won't fit into a byte.
3195Otherwise it tries to not upgrade and just use bytes. But some things that
3196do fit into a byte are variants in utf8, and the caller may not have been
3197keeping track of these.)
3198
3199If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3200isn't guaranteed due to having other routines do the work in some input cases,
3201or if the input is already flagged as being in utf8.
3202
3203The speed of this could perhaps be improved for many cases if someone wanted to
3204write a fast function that counts the number of variant characters in a string,
3205especially if it could return the position of the first one.
3206
8d6d96c1
HS
3207*/
3208
3209STRLEN
b3ab6785 3210Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3211{
97aff369 3212 dVAR;
7918f24d 3213
b3ab6785 3214 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3215
808c356f
RGS
3216 if (sv == &PL_sv_undef)
3217 return 0;
e0e62c2a
NIS
3218 if (!SvPOK(sv)) {
3219 STRLEN len = 0;
d52b7888
NC
3220 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3221 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3222 if (SvUTF8(sv)) {
3223 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3224 return len;
b3ab6785 3225 }
d52b7888
NC
3226 } else {
3227 (void) SvPV_force(sv,len);
3228 }
e0e62c2a 3229 }
4411f3b6 3230
f5cee72b 3231 if (SvUTF8(sv)) {
b3ab6785 3232 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3233 return SvCUR(sv);
f5cee72b 3234 }
5fec3b1d 3235
765f542d
NC
3236 if (SvIsCOW(sv)) {
3237 sv_force_normal_flags(sv, 0);
db42d148
NIS
3238 }
3239
b3ab6785 3240 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3241 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3242 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3243 return SvCUR(sv);
3244 }
3245
3246 if (SvCUR(sv) > 0) { /* Assume Latin-1/EBCDIC */
c4e7c712 3247 /* This function could be much more efficient if we
2bbc8d55 3248 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3249 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3250 * make the loop as fast as possible (although there are certainly ways
3251 * to speed this up, eg. through vectorization) */
3252 U8 * s = (U8 *) SvPVX_const(sv);
3253 U8 * e = (U8 *) SvEND(sv);
3254 U8 *t = s;
3255 STRLEN two_byte_count = 0;
c4e7c712 3256
b3ab6785
KW
3257 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3258
3259 /* See if really will need to convert to utf8. We mustn't rely on our
3260 * incoming SV being well formed and having a trailing '\0', as certain
3261 * code in pp_formline can send us partially built SVs. */
3262
c4e7c712 3263 while (t < e) {
53c1dcc0 3264 const U8 ch = *t++;
b3ab6785
KW
3265 if (NATIVE_IS_INVARIANT(ch)) continue;
3266
3267 t--; /* t already incremented; re-point to first variant */
3268 two_byte_count = 1;
3269 goto must_be_utf8;
c4e7c712 3270 }
b3ab6785
KW
3271
3272 /* utf8 conversion not needed because all are invariants. Mark as
3273 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3274 SvUTF8_on(sv);
b3ab6785
KW
3275 return SvCUR(sv);
3276
3277must_be_utf8:
3278
3279 /* Here, the string should be converted to utf8, either because of an
3280 * input flag (two_byte_count = 0), or because a character that
3281 * requires 2 bytes was found (two_byte_count = 1). t points either to
3282 * the beginning of the string (if we didn't examine anything), or to
3283 * the first variant. In either case, everything from s to t - 1 will
3284 * occupy only 1 byte each on output.
3285 *
3286 * There are two main ways to convert. One is to create a new string
3287 * and go through the input starting from the beginning, appending each
3288 * converted value onto the new string as we go along. It's probably
3289 * best to allocate enough space in the string for the worst possible
3290 * case rather than possibly running out of space and having to
3291 * reallocate and then copy what we've done so far. Since everything
3292 * from s to t - 1 is invariant, the destination can be initialized
3293 * with these using a fast memory copy
3294 *
3295 * The other way is to figure out exactly how big the string should be
3296 * by parsing the entire input. Then you don't have to make it big
3297 * enough to handle the worst possible case, and more importantly, if
3298 * the string you already have is large enough, you don't have to
3299 * allocate a new string, you can copy the last character in the input
3300 * string to the final position(s) that will be occupied by the
3301 * converted string and go backwards, stopping at t, since everything
3302 * before that is invariant.
3303 *
3304 * There are advantages and disadvantages to each method.
3305 *
3306 * In the first method, we can allocate a new string, do the memory
3307 * copy from the s to t - 1, and then proceed through the rest of the
3308 * string byte-by-byte.
3309 *
3310 * In the second method, we proceed through the rest of the input
3311 * string just calculating how big the converted string will be. Then
3312 * there are two cases:
3313 * 1) if the string has enough extra space to handle the converted
3314 * value. We go backwards through the string, converting until we
3315 * get to the position we are at now, and then stop. If this
3316 * position is far enough along in the string, this method is
3317 * faster than the other method. If the memory copy were the same
3318 * speed as the byte-by-byte loop, that position would be about
3319 * half-way, as at the half-way mark, parsing to the end and back
3320 * is one complete string's parse, the same amount as starting
3321 * over and going all the way through. Actually, it would be
3322 * somewhat less than half-way, as it's faster to just count bytes
3323 * than to also copy, and we don't have the overhead of allocating
3324 * a new string, changing the scalar to use it, and freeing the
3325 * existing one. But if the memory copy is fast, the break-even
3326 * point is somewhere after half way. The counting loop could be
3327 * sped up by vectorization, etc, to move the break-even point
3328 * further towards the beginning.
3329 * 2) if the string doesn't have enough space to handle the converted
3330 * value. A new string will have to be allocated, and one might
3331 * as well, given that, start from the beginning doing the first
3332 * method. We've spent extra time parsing the string and in
3333 * exchange all we've gotten is that we know precisely how big to
3334 * make the new one. Perl is more optimized for time than space,
3335 * so this case is a loser.
3336 * So what I've decided to do is not use the 2nd method unless it is
3337 * guaranteed that a new string won't have to be allocated, assuming
3338 * the worst case. I also decided not to put any more conditions on it
3339 * than this, for now. It seems likely that, since the worst case is
3340 * twice as big as the unknown portion of the string (plus 1), we won't
3341 * be guaranteed enough space, causing us to go to the first method,
3342 * unless the string is short, or the first variant character is near
3343 * the end of it. In either of these cases, it seems best to use the
3344 * 2nd method. The only circumstance I can think of where this would
3345 * be really slower is if the string had once had much more data in it
3346 * than it does now, but there is still a substantial amount in it */
3347
3348 {
3349 STRLEN invariant_head = t - s;
3350 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3351 if (SvLEN(sv) < size) {
3352
3353 /* Here, have decided to allocate a new string */
3354
3355 U8 *dst;
3356 U8 *d;
3357
3358 Newx(dst, size, U8);
3359
3360 /* If no known invariants at the beginning of the input string,
3361 * set so starts from there. Otherwise, can use memory copy to
3362 * get up to where we are now, and then start from here */
3363
3364 if (invariant_head <= 0) {
3365 d = dst;
3366 } else {
3367 Copy(s, dst, invariant_head, char);
3368 d = dst + invariant_head;
3369 }
3370
3371 while (t < e) {
3372 const UV uv = NATIVE8_TO_UNI(*t++);
3373 if (UNI_IS_INVARIANT(uv))
3374 *d++ = (U8)UNI_TO_NATIVE(uv);
3375 else {
3376 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3377 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3378 }
3379 }
3380 *d = '\0';
3381 SvPV_free(sv); /* No longer using pre-existing string */
3382 SvPV_set(sv, (char*)dst);
3383 SvCUR_set(sv, d - dst);
3384 SvLEN_set(sv, size);
3385 } else {
3386
3387 /* Here, have decided to get the exact size of the string.
3388 * Currently this happens only when we know that there is
3389 * guaranteed enough space to fit the converted string, so
3390 * don't have to worry about growing. If two_byte_count is 0,
3391 * then t points to the first byte of the string which hasn't
3392 * been examined yet. Otherwise two_byte_count is 1, and t
3393 * points to the first byte in the string that will expand to
3394 * two. Depending on this, start examining at t or 1 after t.
3395 * */
3396
3397 U8 *d = t + two_byte_count;
3398
3399
3400 /* Count up the remaining bytes that expand to two */
3401
3402 while (d < e) {
3403 const U8 chr = *d++;
3404 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3405 }
3406
3407 /* The string will expand by just the number of bytes that
3408 * occupy two positions. But we are one afterwards because of
3409 * the increment just above. This is the place to put the
3410 * trailing NUL, and to set the length before we decrement */
3411
3412 d += two_byte_count;
3413 SvCUR_set(sv, d - s);
3414 *d-- = '\0';
3415
3416
3417 /* Having decremented d, it points to the position to put the
3418 * very last byte of the expanded string. Go backwards through
3419 * the string, copying and expanding as we go, stopping when we
3420 * get to the part that is invariant the rest of the way down */
3421
3422 e--;
3423 while (e >= t) {
3424 const U8 ch = NATIVE8_TO_UNI(*e--);
3425 if (UNI_IS_INVARIANT(ch)) {
3426 *d-- = UNI_TO_NATIVE(ch);
3427 } else {
3428 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3429 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3430 }
3431 }
3432 }
3433 }
560a288e 3434 }
b3ab6785
KW
3435
3436 /* Mark as UTF-8 even if no variant - saves scanning loop */
3437 SvUTF8_on(sv);
4411f3b6 3438 return SvCUR(sv);
560a288e
GS
3439}
3440
c461cf8f
JH
3441/*
3442=for apidoc sv_utf8_downgrade
3443
78ea37eb 3444Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3445If the PV contains a character that cannot fit
3446in a byte, this conversion will fail;
78ea37eb 3447in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3448true, croaks.
3449
13a6c0e0
JH
3450This is not as a general purpose Unicode to byte encoding interface:
3451use the Encode extension for that.
3452
c461cf8f
JH
3453=cut
3454*/
3455
560a288e 3456bool
7bc54cea 3457Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
560a288e 3458{
97aff369 3459 dVAR;
7918f24d
NC
3460
3461 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3462
78ea37eb 3463 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3464 if (SvCUR(sv)) {
03cfe0ae 3465 U8 *s;
652088fc 3466 STRLEN len;
fa301091 3467
765f542d
NC
3468 if (SvIsCOW(sv)) {
3469 sv_force_normal_flags(sv, 0);
3470 }
03cfe0ae
NIS
3471 s = (U8 *) SvPV(sv, len);
3472 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3473 if (fail_ok)
3474 return FALSE;
3475 else {
3476 if (PL_op)
3477 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3478 OP_DESC(PL_op));
fa301091
JH
3479 else
3480 Perl_croak(aTHX_ "Wide character");
3481 }
4b3603a4 3482 }
b162af07 3483 SvCUR_set(sv, len);
67e989fb 3484 }
560a288e 3485 }
ffebcc3e 3486 SvUTF8_off(sv);
560a288e
GS
3487 return TRUE;
3488}
3489
c461cf8f
JH
3490/*
3491=for apidoc sv_utf8_encode
3492
78ea37eb
ST
3493Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3494flag off so that it looks like octets again.
c461cf8f
JH
3495
3496=cut
3497*/
3498
560a288e 3499void
7bc54cea 3500Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3501{
7918f24d
NC
3502 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3503
4c94c214
NC
3504 if (SvIsCOW(sv)) {
3505 sv_force_normal_flags(sv, 0);
3506 }
3507 if (SvREADONLY(sv)) {
f1f66076 3508 Perl_croak(aTHX_ "%s", PL_no_modify);
4c94c214 3509 }
a5f5288a 3510 (void) sv_utf8_upgrade(sv);
560a288e
GS
3511 SvUTF8_off(sv);
3512}
3513
4411f3b6
NIS
3514/*
3515=for apidoc sv_utf8_decode
3516
78ea37eb
ST
3517If the PV of the SV is an octet sequence in UTF-8
3518and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3519so that it looks like a character. If the PV contains only single-byte
3520characters, the C<SvUTF8> flag stays being off.
3521Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3522
3523=cut
3524*/
3525
560a288e 3526bool
7bc54cea 3527Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3528{
7918f24d
NC
3529 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3530
78ea37eb 3531 if (SvPOKp(sv)) {
93524f2b
NC
3532 const U8 *c;
3533 const U8 *e;
9cbac4c7 3534
645c22ef
DM
3535 /* The octets may have got themselves encoded - get them back as
3536 * bytes
3537 */
3538 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3539 return FALSE;
3540
3541 /* it is actually just a matter of turning the utf8 flag on, but
3542 * we want to make sure everything inside is valid utf8 first.
3543 */
93524f2b 3544 c = (const U8 *) SvPVX_const(sv);
63cd0674 3545 if (!is_utf8_string(c, SvCUR(sv)+1))
67e989fb 3546 return FALSE;
93524f2b 3547 e = (const U8 *) SvEND(sv);
511c2ff0 3548 while (c < e) {
b64e5050 3549 const U8 ch = *c++;
c4d5f83a 3550 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3551 SvUTF8_on(sv);
3552 break;
3553 }
560a288e 3554 }
560a288e
GS
3555 }
3556 return TRUE;
3557}
3558
954c1994
GS
3559/*
3560=for apidoc sv_setsv
3561
645c22ef
DM
3562Copies the contents of the source SV C<ssv> into the destination SV
3563C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3564function if the source SV needs to be reused. Does not handle 'set' magic.
3565Loosely speaking, it performs a copy-by-value, obliterating any previous
3566content of the destination.
3567
3568You probably want to use one of the assortment of wrappers, such as
3569C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3570C<SvSetMagicSV_nosteal>.
3571
8d6d96c1
HS
3572=for apidoc sv_setsv_flags
3573
645c22ef
DM
3574Copies the contents of the source SV C<ssv> into the destination SV
3575C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3576function if the source SV needs to be reused. Does not handle 'set' magic.
3577Loosely speaking, it performs a copy-by-value, obliterating any previous
3578content of the destination.
3579If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
5fcdf167
NC
3580C<ssv> if appropriate, else not. If the C<flags> parameter has the
3581C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3582and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3583
3584You probably want to use one of the assortment of wrappers, such as
3585C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3586C<SvSetMagicSV_nosteal>.
3587
3588This is the primary function for copying scalars, and most other
3589copy-ish functions and macros use this underneath.
8d6d96c1
HS
3590
3591=cut
3592*/
3593
5d0301b7 3594static void
7bc54cea 3595S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3596{
70cd14a1 3597 I32 mro_changes = 0; /* 1 = method, 2 = isa */
dd69841b 3598
7918f24d
NC
3599 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3600
5d0301b7
NC
3601 if (dtype != SVt_PVGV) {
3602 const char * const name = GvNAME(sstr);
3603 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3604 {
f7877b28
NC
3605 if (dtype >= SVt_PV) {
3606 SvPV_free(dstr);
3607 SvPV_set(dstr, 0);
3608 SvLEN_set(dstr, 0);
3609 SvCUR_set(dstr, 0);
3610 }
0d092c36 3611 SvUPGRADE(dstr, SVt_PVGV);
dedf8e73 3612 (void)SvOK_off(dstr);
2e5b91de
NC
3613 /* FIXME - why are we doing this, then turning it off and on again
3614 below? */
3615 isGV_with_GP_on(dstr);
f7877b28 3616 }
5d0301b7
NC
3617 GvSTASH(dstr) = GvSTASH(sstr);
3618 if (GvSTASH(dstr))
daba3364 3619 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
159b6efe 3620 gv_name_set(MUTABLE_GV(dstr), name, len, GV_ADD);
5d0301b7
NC
3621 SvFAKE_on(dstr); /* can coerce to non-glob */
3622 }
3623
159b6efe 3624 if(GvGP(MUTABLE_GV(sstr))) {
dd69841b
BB
3625 /* If source has method cache entry, clear it */
3626 if(GvCVGEN(sstr)) {
3627 SvREFCNT_dec(GvCV(sstr));
3628 GvCV(sstr) = NULL;
3629 GvCVGEN(sstr) = 0;
3630 }
3631 /* If source has a real method, then a method is
3632 going to change */
159b6efe 3633 else if(GvCV((const GV *)sstr)) {
70cd14a1 3634 mro_changes = 1;
dd69841b
BB
3635 }
3636 }
3637
3638 /* If dest already had a real method, that's a change as well */
159b6efe 3639 if(!mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)) {
70cd14a1 3640 mro_changes = 1;
dd69841b
BB
3641 }
3642
159b6efe 3643 if(strEQ(GvNAME((const GV *)dstr),"ISA"))
70cd14a1
CB
3644 mro_changes = 2;
3645
159b6efe 3646 gp_free(MUTABLE_GV(dstr));
2e5b91de 3647 isGV_with_GP_off(dstr);
5d0301b7 3648 (void)SvOK_off(dstr);
2e5b91de 3649 isGV_with_GP_on(dstr);
dedf8e73 3650 GvINTRO_off(dstr); /* one-shot flag */
5d0301b7
NC
3651 GvGP(dstr) = gp_ref(GvGP(sstr));
3652 if (SvTAINTED(sstr))
3653 SvTAINT(dstr);
3654 if (GvIMPORTED(dstr) != GVf_IMPORTED
3655 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3656 {
3657 GvIMPORTED_on(dstr);
3658 }
3659 GvMULTI_on(dstr);
70cd14a1
CB
3660 if(mro_changes == 2) mro_isa_changed_in(GvSTASH(dstr));
3661 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
5d0301b7
NC
3662 return;
3663}
3664