This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
sv.c/find_uninit_var: Explain kid-scanning loop better
[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
07208e09 35#ifndef HAS_C99
4b2d6671 36# if __STDC_VERSION__ >= 199901L && !defined(VMS)
07208e09
CS
37# define HAS_C99 1
38# endif
39#endif
40#if HAS_C99
41# include <stdint.h>
42#endif
43
51371543 44#define FCALL *f
2c5424a7 45
2f8ed50e
OS
46#ifdef __Lynx__
47/* Missing proto on LynxOS */
48 char *gconvert(double, int, int, char *);
49#endif
50
e23c8137 51#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 52/* if adding more checks watch out for the following tests:
e23c8137
JH
53 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
54 * lib/utf8.t lib/Unicode/Collate/t/index.t
55 * --jhi
56 */
6f207bd3 57# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
58 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
59 assert((cache)[2] <= (cache)[3]); \
60 assert((cache)[3] <= (cache)[1]);} \
61 } STMT_END
e23c8137 62#else
6f207bd3 63# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
64#endif
65
f8c7b90f 66#ifdef PERL_OLD_COPY_ON_WRITE
765f542d 67#define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
607fa7f2 68#define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
b5ccf5f2 69/* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
765f542d 70 on-write. */
765f542d 71#endif
645c22ef
DM
72
73/* ============================================================================
74
75=head1 Allocation and deallocation of SVs.
76
d2a0f284
JC
77An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
78sv, av, hv...) contains type and reference count information, and for
79many types, a pointer to the body (struct xrv, xpv, xpviv...), which
80contains fields specific to each type. Some types store all they need
81in the head, so don't have a body.
82
486ec47a 83In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
84and bodies are allocated out of arenas, which by default are
85approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
86Sv-bodies are allocated by their sv-type, guaranteeing size
87consistency needed to allocate safely from arrays.
88
d2a0f284
JC
89For SV-heads, the first slot in each arena is reserved, and holds a
90link to the next arena, some flags, and a note of the number of slots.
91Snaked through each arena chain is a linked list of free items; when
92this becomes empty, an extra arena is allocated and divided up into N
93items which are threaded into the free list.
94
95SV-bodies are similar, but they use arena-sets by default, which
96separate the link and info from the arena itself, and reclaim the 1st
97slot in the arena. SV-bodies are further described later.
645c22ef
DM
98
99The following global variables are associated with arenas:
100
101 PL_sv_arenaroot pointer to list of SV arenas
102 PL_sv_root pointer to list of free SV structures
103
d2a0f284
JC
104 PL_body_arenas head of linked-list of body arenas
105 PL_body_roots[] array of pointers to list of free bodies of svtype
106 arrays are indexed by the svtype needed
93e68bfb 107
d2a0f284
JC
108A few special SV heads are not allocated from an arena, but are
109instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
110The size of arenas can be changed from the default by setting
111PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
112
113The SV arena serves the secondary purpose of allowing still-live SVs
114to be located and destroyed during final cleanup.
115
116At the lowest level, the macros new_SV() and del_SV() grab and free
117an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
118to return the SV to the free list with error checking.) new_SV() calls
119more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
120SVs in the free list have their SvTYPE field set to all ones.
121
ff276b08 122At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 123perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 124start of the interpreter.
645c22ef 125
645c22ef
DM
126The function visit() scans the SV arenas list, and calls a specified
127function for each SV it finds which is still live - ie which has an SvTYPE
128other than all 1's, and a non-zero SvREFCNT. visit() is used by the
129following functions (specified as [function that calls visit()] / [function
130called by visit() for each SV]):
131
132 sv_report_used() / do_report_used()
f2524eef 133 dump all remaining SVs (debugging aid)
645c22ef 134
e4487e9b
DM
135 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
136 do_clean_named_io_objs()
645c22ef 137 Attempt to free all objects pointed to by RVs,
c3b19b5c 138 and try to do the same for all objects indirectly
645c22ef
DM
139 referenced by typeglobs too. Called once from
140 perl_destruct(), prior to calling sv_clean_all()
141 below.
142
143 sv_clean_all() / do_clean_all()
144 SvREFCNT_dec(sv) each remaining SV, possibly
145 triggering an sv_free(). It also sets the
146 SVf_BREAK flag on the SV to indicate that the
147 refcnt has been artificially lowered, and thus
148 stopping sv_free() from giving spurious warnings
149 about SVs which unexpectedly have a refcnt
150 of zero. called repeatedly from perl_destruct()
151 until there are no SVs left.
152
93e68bfb 153=head2 Arena allocator API Summary
645c22ef
DM
154
155Private API to rest of sv.c
156
157 new_SV(), del_SV(),
158
df0f0429 159 new_XPVNV(), del_XPVGV(),
645c22ef
DM
160 etc
161
162Public API:
163
8cf8f3d1 164 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 165
645c22ef
DM
166=cut
167
3e8320cc 168 * ========================================================================= */
645c22ef 169
4561caa4
CS
170/*
171 * "A time to plant, and a time to uproot what was planted..."
172 */
173
d7a2c63c
MHM
174#ifdef PERL_MEM_LOG
175# define MEM_LOG_NEW_SV(sv, file, line, func) \
176 Perl_mem_log_new_sv(sv, file, line, func)
177# define MEM_LOG_DEL_SV(sv, file, line, func) \
178 Perl_mem_log_del_sv(sv, file, line, func)
179#else
180# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
181# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
182#endif
183
fd0854ff 184#ifdef DEBUG_LEAKING_SCALARS
22162ca8 185# define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
d7a2c63c
MHM
186# define DEBUG_SV_SERIAL(sv) \
187 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
188 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
189#else
190# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 191# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
192#endif
193
48614a46
NC
194#ifdef PERL_POISON
195# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 196# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
197/* Whilst I'd love to do this, it seems that things like to check on
198 unreferenced scalars
7e337ee0 199# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 200*/
7e337ee0
JH
201# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
202 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
203#else
204# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 205# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
206# define POSION_SV_HEAD(sv)
207#endif
208
990198f0
DM
209/* Mark an SV head as unused, and add to free list.
210 *
211 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
212 * its refcount artificially decremented during global destruction, so
213 * there may be dangling pointers to it. The last thing we want in that
214 * case is for it to be reused. */
215
053fc874
GS
216#define plant_SV(p) \
217 STMT_START { \
990198f0 218 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
219 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
220 DEBUG_SV_SERIAL(p); \
fd0854ff 221 FREE_SV_DEBUG_FILE(p); \
48614a46 222 POSION_SV_HEAD(p); \
053fc874 223 SvFLAGS(p) = SVTYPEMASK; \
990198f0 224 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 225 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
226 PL_sv_root = (p); \
227 } \
053fc874
GS
228 --PL_sv_count; \
229 } STMT_END
a0d0e21e 230
053fc874
GS
231#define uproot_SV(p) \
232 STMT_START { \
233 (p) = PL_sv_root; \
daba3364 234 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
235 ++PL_sv_count; \
236 } STMT_END
237
645c22ef 238
cac9b346
NC
239/* make some more SVs by adding another arena */
240
cac9b346
NC
241STATIC SV*
242S_more_sv(pTHX)
243{
97aff369 244 dVAR;
cac9b346 245 SV* sv;
9a87bd09
NC
246 char *chunk; /* must use New here to match call to */
247 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
248 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
249 uproot_SV(sv);
250 return sv;
251}
252
645c22ef
DM
253/* new_SV(): return a new, empty SV head */
254
eba0f806
DM
255#ifdef DEBUG_LEAKING_SCALARS
256/* provide a real function for a debugger to play with */
257STATIC SV*
d7a2c63c 258S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
259{
260 SV* sv;
261
eba0f806
DM
262 if (PL_sv_root)
263 uproot_SV(sv);
264 else
cac9b346 265 sv = S_more_sv(aTHX);
eba0f806
DM
266 SvANY(sv) = 0;
267 SvREFCNT(sv) = 1;
268 SvFLAGS(sv) = 0;
fd0854ff 269 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
270 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
271 ? PL_parser->copline
272 : PL_curcop
f24aceb1
DM
273 ? CopLINE(PL_curcop)
274 : 0
e385c3bf 275 );
fd0854ff 276 sv->sv_debug_inpad = 0;
cd676548 277 sv->sv_debug_parent = NULL;
fd0854ff 278 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
279
280 sv->sv_debug_serial = PL_sv_serial++;
281
282 MEM_LOG_NEW_SV(sv, file, line, func);
283 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
284 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
285
eba0f806
DM
286 return sv;
287}
d7a2c63c 288# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
289
290#else
291# define new_SV(p) \
053fc874 292 STMT_START { \
053fc874
GS
293 if (PL_sv_root) \
294 uproot_SV(p); \
295 else \
cac9b346 296 (p) = S_more_sv(aTHX); \
053fc874
GS
297 SvANY(p) = 0; \
298 SvREFCNT(p) = 1; \
299 SvFLAGS(p) = 0; \
d7a2c63c 300 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 301 } STMT_END
eba0f806 302#endif
463ee0b2 303
645c22ef
DM
304
305/* del_SV(): return an empty SV head to the free list */
306
a0d0e21e 307#ifdef DEBUGGING
4561caa4 308
053fc874
GS
309#define del_SV(p) \
310 STMT_START { \
aea4f609 311 if (DEBUG_D_TEST) \
053fc874
GS
312 del_sv(p); \
313 else \
314 plant_SV(p); \
053fc874 315 } STMT_END
a0d0e21e 316
76e3520e 317STATIC void
cea2e8a9 318S_del_sv(pTHX_ SV *p)
463ee0b2 319{
97aff369 320 dVAR;
7918f24d
NC
321
322 PERL_ARGS_ASSERT_DEL_SV;
323
aea4f609 324 if (DEBUG_D_TEST) {
4633a7c4 325 SV* sva;
a3b680e6 326 bool ok = 0;
daba3364 327 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
328 const SV * const sv = sva + 1;
329 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 330 if (p >= sv && p < svend) {
a0d0e21e 331 ok = 1;
c0ff570e
NC
332 break;
333 }
a0d0e21e
LW
334 }
335 if (!ok) {
9b387841
NC
336 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
337 "Attempt to free non-arena SV: 0x%"UVxf
338 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
339 return;
340 }
341 }
4561caa4 342 plant_SV(p);
463ee0b2 343}
a0d0e21e 344
4561caa4
CS
345#else /* ! DEBUGGING */
346
347#define del_SV(p) plant_SV(p)
348
349#endif /* DEBUGGING */
463ee0b2 350
645c22ef
DM
351
352/*
ccfc67b7
JH
353=head1 SV Manipulation Functions
354
645c22ef
DM
355=for apidoc sv_add_arena
356
357Given a chunk of memory, link it to the head of the list of arenas,
358and split it into a list of free SVs.
359
360=cut
361*/
362
d2bd4e7f
NC
363static void
364S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 365{
97aff369 366 dVAR;
daba3364 367 SV *const sva = MUTABLE_SV(ptr);
463ee0b2
LW
368 register SV* sv;
369 register SV* svend;
4633a7c4 370
7918f24d
NC
371 PERL_ARGS_ASSERT_SV_ADD_ARENA;
372
4633a7c4 373 /* The first SV in an arena isn't an SV. */
3280af22 374 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
375 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
376 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
377
3280af22
NIS
378 PL_sv_arenaroot = sva;
379 PL_sv_root = sva + 1;
4633a7c4
LW
380
381 svend = &sva[SvREFCNT(sva) - 1];
382 sv = sva + 1;
463ee0b2 383 while (sv < svend) {
3eef1deb 384 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 385#ifdef DEBUGGING
978b032e 386 SvREFCNT(sv) = 0;
03e36789 387#endif
4b69cbe3 388 /* Must always set typemask because it's always checked in on cleanup
03e36789 389 when the arenas are walked looking for objects. */
8990e307 390 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
391 sv++;
392 }
3eef1deb 393 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
394#ifdef DEBUGGING
395 SvREFCNT(sv) = 0;
396#endif
4633a7c4
LW
397 SvFLAGS(sv) = SVTYPEMASK;
398}
399
055972dc
DM
400/* visit(): call the named function for each non-free SV in the arenas
401 * whose flags field matches the flags/mask args. */
645c22ef 402
5226ed68 403STATIC I32
de37a194 404S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 405{
97aff369 406 dVAR;
4633a7c4 407 SV* sva;
5226ed68 408 I32 visited = 0;
8990e307 409
7918f24d
NC
410 PERL_ARGS_ASSERT_VISIT;
411
daba3364 412 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0 413 register const SV * const svend = &sva[SvREFCNT(sva)];
a3b680e6 414 register SV* sv;
4561caa4 415 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 416 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
417 && (sv->sv_flags & mask) == flags
418 && SvREFCNT(sv))
419 {
acfe0abc 420 (FCALL)(aTHX_ sv);
5226ed68
JH
421 ++visited;
422 }
8990e307
LW
423 }
424 }
5226ed68 425 return visited;
8990e307
LW
426}
427
758a08c3
JH
428#ifdef DEBUGGING
429
645c22ef
DM
430/* called by sv_report_used() for each live SV */
431
432static void
5fa45a31 433do_report_used(pTHX_ SV *const sv)
645c22ef 434{
e4787c0c 435 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
436 PerlIO_printf(Perl_debug_log, "****\n");
437 sv_dump(sv);
438 }
439}
758a08c3 440#endif
645c22ef
DM
441
442/*
443=for apidoc sv_report_used
444
fde67290 445Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
446
447=cut
448*/
449
8990e307 450void
864dbfa3 451Perl_sv_report_used(pTHX)
4561caa4 452{
ff270d3a 453#ifdef DEBUGGING
055972dc 454 visit(do_report_used, 0, 0);
96a5add6
AL
455#else
456 PERL_UNUSED_CONTEXT;
ff270d3a 457#endif
4561caa4
CS
458}
459
645c22ef
DM
460/* called by sv_clean_objs() for each live SV */
461
462static void
de37a194 463do_clean_objs(pTHX_ SV *const ref)
645c22ef 464{
97aff369 465 dVAR;
ea724faa
NC
466 assert (SvROK(ref));
467 {
823a54a3
AL
468 SV * const target = SvRV(ref);
469 if (SvOBJECT(target)) {
470 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
471 if (SvWEAKREF(ref)) {
472 sv_del_backref(target, ref);
473 SvWEAKREF_off(ref);
474 SvRV_set(ref, NULL);
475 } else {
476 SvROK_off(ref);
477 SvRV_set(ref, NULL);
478 SvREFCNT_dec(target);
479 }
645c22ef
DM
480 }
481 }
482
483 /* XXX Might want to check arrays, etc. */
484}
485
645c22ef 486
e4487e9b
DM
487/* clear any slots in a GV which hold objects - except IO;
488 * called by sv_clean_objs() for each live GV */
489
645c22ef 490static void
f30de749 491do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 492{
97aff369 493 dVAR;
57ef47cc 494 SV *obj;
ea724faa 495 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 496 assert(isGV_with_GP(sv));
57ef47cc
DM
497 if (!GvGP(sv))
498 return;
499
500 /* freeing GP entries may indirectly free the current GV;
501 * hold onto it while we mess with the GP slots */
502 SvREFCNT_inc(sv);
503
504 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
505 DEBUG_D((PerlIO_printf(Perl_debug_log,
506 "Cleaning named glob SV object:\n "), sv_dump(obj)));
507 GvSV(sv) = NULL;
508 SvREFCNT_dec(obj);
509 }
510 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
511 DEBUG_D((PerlIO_printf(Perl_debug_log,
512 "Cleaning named glob AV object:\n "), sv_dump(obj)));
513 GvAV(sv) = NULL;
514 SvREFCNT_dec(obj);
515 }
516 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
517 DEBUG_D((PerlIO_printf(Perl_debug_log,
518 "Cleaning named glob HV object:\n "), sv_dump(obj)));
519 GvHV(sv) = NULL;
520 SvREFCNT_dec(obj);
521 }
522 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
523 DEBUG_D((PerlIO_printf(Perl_debug_log,
524 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 525 GvCV_set(sv, NULL);
57ef47cc
DM
526 SvREFCNT_dec(obj);
527 }
e4487e9b
DM
528 SvREFCNT_dec(sv); /* undo the inc above */
529}
530
68b590d9 531/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
532 * called by sv_clean_objs() for each live GV */
533
534static void
535do_clean_named_io_objs(pTHX_ SV *const sv)
536{
537 dVAR;
538 SV *obj;
539 assert(SvTYPE(sv) == SVt_PVGV);
540 assert(isGV_with_GP(sv));
68b590d9 541 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
542 return;
543
544 SvREFCNT_inc(sv);
57ef47cc
DM
545 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
546 DEBUG_D((PerlIO_printf(Perl_debug_log,
547 "Cleaning named glob IO object:\n "), sv_dump(obj)));
548 GvIOp(sv) = NULL;
549 SvREFCNT_dec(obj);
645c22ef 550 }
57ef47cc 551 SvREFCNT_dec(sv); /* undo the inc above */
645c22ef 552}
645c22ef 553
4155e4fe
FC
554/* Void wrapper to pass to visit() */
555static void
556do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
557 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
558 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
559 return;
560 (void)curse(sv, 0);
561}
562
645c22ef
DM
563/*
564=for apidoc sv_clean_objs
565
fde67290 566Attempt to destroy all objects not yet freed.
645c22ef
DM
567
568=cut
569*/
570
4561caa4 571void
864dbfa3 572Perl_sv_clean_objs(pTHX)
4561caa4 573{
97aff369 574 dVAR;
68b590d9 575 GV *olddef, *olderr;
3280af22 576 PL_in_clean_objs = TRUE;
055972dc 577 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
578 /* Some barnacles may yet remain, clinging to typeglobs.
579 * Run the non-IO destructors first: they may want to output
580 * error messages, close files etc */
d011219a 581 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 582 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
583 /* And if there are some very tenacious barnacles clinging to arrays,
584 closures, or what have you.... */
585 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
586 olddef = PL_defoutgv;
587 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
588 if (olddef && isGV_with_GP(olddef))
589 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
590 olderr = PL_stderrgv;
591 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
592 if (olderr && isGV_with_GP(olderr))
593 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
594 SvREFCNT_dec(olddef);
3280af22 595 PL_in_clean_objs = FALSE;
4561caa4
CS
596}
597
645c22ef
DM
598/* called by sv_clean_all() for each live SV */
599
600static void
de37a194 601do_clean_all(pTHX_ SV *const sv)
645c22ef 602{
97aff369 603 dVAR;
daba3364 604 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 605 /* don't clean pid table and strtab */
d17ea597 606 return;
cddfcddc 607 }
645c22ef
DM
608 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
609 SvFLAGS(sv) |= SVf_BREAK;
610 SvREFCNT_dec(sv);
611}
612
613/*
614=for apidoc sv_clean_all
615
616Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 617cleanup. This function may have to be called multiple times to free
ff276b08 618SVs which are in complex self-referential hierarchies.
645c22ef
DM
619
620=cut
621*/
622
5226ed68 623I32
864dbfa3 624Perl_sv_clean_all(pTHX)
8990e307 625{
97aff369 626 dVAR;
5226ed68 627 I32 cleaned;
3280af22 628 PL_in_clean_all = TRUE;
055972dc 629 cleaned = visit(do_clean_all, 0,0);
5226ed68 630 return cleaned;
8990e307 631}
463ee0b2 632
5e258f8c
JC
633/*
634 ARENASETS: a meta-arena implementation which separates arena-info
635 into struct arena_set, which contains an array of struct
636 arena_descs, each holding info for a single arena. By separating
637 the meta-info from the arena, we recover the 1st slot, formerly
638 borrowed for list management. The arena_set is about the size of an
39244528 639 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
640
641 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
642 memory in the last arena-set (1/2 on average). In trade, we get
643 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 644 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
645 small arenas for large, rare body types, by changing array* fields
646 in body_details_by_type[] below.
5e258f8c 647*/
5e258f8c 648struct arena_desc {
398c677b
NC
649 char *arena; /* the raw storage, allocated aligned */
650 size_t size; /* its size ~4k typ */
e5973ed5 651 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
652};
653
e6148039
NC
654struct arena_set;
655
656/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 657 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
658 therefore likely to be 1 aligned memory page. */
659
660#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
661 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
662
663struct arena_set {
664 struct arena_set* next;
0a848332
NC
665 unsigned int set_size; /* ie ARENAS_PER_SET */
666 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
667 struct arena_desc set[ARENAS_PER_SET];
668};
669
645c22ef
DM
670/*
671=for apidoc sv_free_arenas
672
fde67290 673Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
674heads and bodies within the arenas must already have been freed.
675
676=cut
677*/
4633a7c4 678void
864dbfa3 679Perl_sv_free_arenas(pTHX)
4633a7c4 680{
97aff369 681 dVAR;
4633a7c4
LW
682 SV* sva;
683 SV* svanext;
0a848332 684 unsigned int i;
4633a7c4
LW
685
686 /* Free arenas here, but be careful about fake ones. (We assume
687 contiguity of the fake ones with the corresponding real ones.) */
688
3280af22 689 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 690 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 691 while (svanext && SvFAKE(svanext))
daba3364 692 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
693
694 if (!SvFAKE(sva))
1df70142 695 Safefree(sva);
4633a7c4 696 }
93e68bfb 697
5e258f8c 698 {
0a848332
NC
699 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
700
701 while (aroot) {
702 struct arena_set *current = aroot;
703 i = aroot->curr;
704 while (i--) {
5e258f8c
JC
705 assert(aroot->set[i].arena);
706 Safefree(aroot->set[i].arena);
707 }
0a848332
NC
708 aroot = aroot->next;
709 Safefree(current);
5e258f8c
JC
710 }
711 }
dc8220bf 712 PL_body_arenas = 0;
fdda85ca 713
0a848332
NC
714 i = PERL_ARENA_ROOTS_SIZE;
715 while (i--)
93e68bfb 716 PL_body_roots[i] = 0;
93e68bfb 717
3280af22
NIS
718 PL_sv_arenaroot = 0;
719 PL_sv_root = 0;
4633a7c4
LW
720}
721
bd81e77b
NC
722/*
723 Here are mid-level routines that manage the allocation of bodies out
724 of the various arenas. There are 5 kinds of arenas:
29489e7c 725
bd81e77b
NC
726 1. SV-head arenas, which are discussed and handled above
727 2. regular body arenas
728 3. arenas for reduced-size bodies
729 4. Hash-Entry arenas
29489e7c 730
bd81e77b
NC
731 Arena types 2 & 3 are chained by body-type off an array of
732 arena-root pointers, which is indexed by svtype. Some of the
733 larger/less used body types are malloced singly, since a large
734 unused block of them is wasteful. Also, several svtypes dont have
735 bodies; the data fits into the sv-head itself. The arena-root
736 pointer thus has a few unused root-pointers (which may be hijacked
737 later for arena types 4,5)
29489e7c 738
bd81e77b
NC
739 3 differs from 2 as an optimization; some body types have several
740 unused fields in the front of the structure (which are kept in-place
741 for consistency). These bodies can be allocated in smaller chunks,
742 because the leading fields arent accessed. Pointers to such bodies
743 are decremented to point at the unused 'ghost' memory, knowing that
744 the pointers are used with offsets to the real memory.
29489e7c 745
d2a0f284
JC
746
747=head1 SV-Body Allocation
748
749Allocation of SV-bodies is similar to SV-heads, differing as follows;
750the allocation mechanism is used for many body types, so is somewhat
751more complicated, it uses arena-sets, and has no need for still-live
752SV detection.
753
754At the outermost level, (new|del)_X*V macros return bodies of the
755appropriate type. These macros call either (new|del)_body_type or
756(new|del)_body_allocated macro pairs, depending on specifics of the
757type. Most body types use the former pair, the latter pair is used to
758allocate body types with "ghost fields".
759
760"ghost fields" are fields that are unused in certain types, and
69ba284b 761consequently don't need to actually exist. They are declared because
d2a0f284
JC
762they're part of a "base type", which allows use of functions as
763methods. The simplest examples are AVs and HVs, 2 aggregate types
764which don't use the fields which support SCALAR semantics.
765
69ba284b 766For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
767chunks, we thus avoid wasted memory for those unaccessed members.
768When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 769size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
770structure. (But things will all go boom if you write to the part that
771is "not there", because you'll be overwriting the last members of the
772preceding structure in memory.)
773
69ba284b
NC
774We calculate the correction using the STRUCT_OFFSET macro on the first
775member present. If the allocated structure is smaller (no initial NV
776actually allocated) then the net effect is to subtract the size of the NV
777from the pointer, to return a new pointer as if an initial NV were actually
778allocated. (We were using structures named *_allocated for this, but
779this turned out to be a subtle bug, because a structure without an NV
780could have a lower alignment constraint, but the compiler is allowed to
781optimised accesses based on the alignment constraint of the actual pointer
782to the full structure, for example, using a single 64 bit load instruction
783because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284
JC
784
785This is the same trick as was used for NV and IV bodies. Ironically it
786doesn't need to be used for NV bodies any more, because NV is now at
787the start of the structure. IV bodies don't need it either, because
788they are no longer allocated.
789
790In turn, the new_body_* allocators call S_new_body(), which invokes
791new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 792linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
793necessary to refresh an empty list. Then the lock is released, and
794the body is returned.
795
99816f8d 796Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
797bodies, which it strings into a linked list. It looks up arena-size
798and body-size from the body_details table described below, thus
799supporting the multiple body-types.
800
801If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
802the (new|del)_X*V macros are mapped directly to malloc/free.
803
d2a0f284
JC
804For each sv-type, struct body_details bodies_by_type[] carries
805parameters which control these aspects of SV handling:
806
807Arena_size determines whether arenas are used for this body type, and if
808so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
809zero, forcing individual mallocs and frees.
810
811Body_size determines how big a body is, and therefore how many fit into
812each arena. Offset carries the body-pointer adjustment needed for
69ba284b 813"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
814
815But its main purpose is to parameterize info needed in
816Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 817vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
818are used for this, except for arena_size.
819
820For the sv-types that have no bodies, arenas are not used, so those
821PL_body_roots[sv_type] are unused, and can be overloaded. In
822something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 823PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 824bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 825available in hv.c.
d2a0f284 826
29489e7c
DM
827*/
828
bd81e77b 829struct body_details {
0fb58b32 830 U8 body_size; /* Size to allocate */
10666ae3 831 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 832 U8 offset;
10666ae3
NC
833 unsigned int type : 4; /* We have space for a sanity check. */
834 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
835 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
836 unsigned int arena : 1; /* Allocated from an arena */
837 size_t arena_size; /* Size of arena to allocate */
bd81e77b 838};
29489e7c 839
bd81e77b
NC
840#define HADNV FALSE
841#define NONV TRUE
29489e7c 842
d2a0f284 843
bd81e77b
NC
844#ifdef PURIFY
845/* With -DPURFIY we allocate everything directly, and don't use arenas.
846 This seems a rather elegant way to simplify some of the code below. */
847#define HASARENA FALSE
848#else
849#define HASARENA TRUE
850#endif
851#define NOARENA FALSE
29489e7c 852
d2a0f284
JC
853/* Size the arenas to exactly fit a given number of bodies. A count
854 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
855 simplifying the default. If count > 0, the arena is sized to fit
856 only that many bodies, allowing arenas to be used for large, rare
857 bodies (XPVFM, XPVIO) without undue waste. The arena size is
858 limited by PERL_ARENA_SIZE, so we can safely oversize the
859 declarations.
860 */
95db5f15
MB
861#define FIT_ARENA0(body_size) \
862 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
863#define FIT_ARENAn(count,body_size) \
864 ( count * body_size <= PERL_ARENA_SIZE) \
865 ? count * body_size \
866 : FIT_ARENA0 (body_size)
867#define FIT_ARENA(count,body_size) \
868 count \
869 ? FIT_ARENAn (count, body_size) \
870 : FIT_ARENA0 (body_size)
d2a0f284 871
bd81e77b
NC
872/* Calculate the length to copy. Specifically work out the length less any
873 final padding the compiler needed to add. See the comment in sv_upgrade
874 for why copying the padding proved to be a bug. */
29489e7c 875
bd81e77b
NC
876#define copy_length(type, last_member) \
877 STRUCT_OFFSET(type, last_member) \
daba3364 878 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 879
bd81e77b 880static const struct body_details bodies_by_type[] = {
829cd18a
NC
881 /* HEs use this offset for their arena. */
882 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 883
1cb9cd50 884 /* The bind placeholder pretends to be an RV for now.
c6f8b1d0 885 Also it's marked as "can't upgrade" to stop anyone using it before it's
1cb9cd50
NC
886 implemented. */
887 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
888
db93c0c4
NC
889 /* IVs are in the head, so the allocation size is 0. */
890 { 0,
d2a0f284 891 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 892 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 893 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
894 },
895
6e128786
NC
896 { sizeof(NV), sizeof(NV),
897 STRUCT_OFFSET(XPVNV, xnv_u),
898 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 899
bc337e5c 900 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
901 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
902 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 903 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 904 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 905
bc337e5c 906 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
907 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
908 + STRUCT_OFFSET(XPV, xpv_cur),
909 SVt_PVIV, FALSE, NONV, HASARENA,
910 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 911
bc337e5c 912 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
913 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
914 + STRUCT_OFFSET(XPV, xpv_cur),
915 SVt_PVNV, FALSE, HADNV, HASARENA,
916 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 917
6e128786 918 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 919 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 920
601dfd0a
NC
921 { sizeof(regexp),
922 sizeof(regexp),
923 0,
08e44740 924 SVt_REGEXP, FALSE, NONV, HASARENA,
eaeb1e7f 925 FIT_ARENA(0, sizeof(regexp))
5c35adbb 926 },
4df7f6af 927
10666ae3 928 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
929 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
930
10666ae3 931 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
932 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
933
601dfd0a 934 { sizeof(XPVAV),
4f7003f5 935 copy_length(XPVAV, xav_alloc),
601dfd0a 936 0,
69ba284b 937 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 938 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 939
601dfd0a 940 { sizeof(XPVHV),
359164a0 941 copy_length(XPVHV, xhv_max),
601dfd0a 942 0,
69ba284b 943 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 944 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 945
601dfd0a
NC
946 { sizeof(XPVCV),
947 sizeof(XPVCV),
948 0,
69ba284b 949 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 950 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 951
601dfd0a
NC
952 { sizeof(XPVFM),
953 sizeof(XPVFM),
954 0,
69ba284b 955 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 956 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 957
601dfd0a
NC
958 { sizeof(XPVIO),
959 sizeof(XPVIO),
960 0,
b6f60916 961 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 962 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 963};
29489e7c 964
bd81e77b 965#define new_body_allocated(sv_type) \
d2a0f284 966 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 967 - bodies_by_type[sv_type].offset)
29489e7c 968
26359cfa
NC
969/* return a thing to the free list */
970
971#define del_body(thing, root) \
972 STMT_START { \
973 void ** const thing_copy = (void **)thing; \
974 *thing_copy = *root; \
975 *root = (void*)thing_copy; \
976 } STMT_END
29489e7c 977
bd81e77b 978#ifdef PURIFY
29489e7c 979
beeec492
NC
980#define new_XNV() safemalloc(sizeof(XPVNV))
981#define new_XPVNV() safemalloc(sizeof(XPVNV))
982#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 983
beeec492 984#define del_XPVGV(p) safefree(p)
29489e7c 985
bd81e77b 986#else /* !PURIFY */
29489e7c 987
65ac1738 988#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 989#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 990#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 991
26359cfa
NC
992#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
993 &PL_body_roots[SVt_PVGV])
1d7c1841 994
bd81e77b 995#endif /* PURIFY */
93e68bfb 996
bd81e77b 997/* no arena for you! */
93e68bfb 998
bd81e77b 999#define new_NOARENA(details) \
beeec492 1000 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1001#define new_NOARENAZ(details) \
beeec492 1002 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1003
1e30fcd5
NC
1004void *
1005Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1006 const size_t arena_size)
d2a0f284
JC
1007{
1008 dVAR;
1009 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1010 struct arena_desc *adesc;
1011 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1012 unsigned int curr;
d2a0f284
JC
1013 char *start;
1014 const char *end;
02982131 1015 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1016#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1017 static bool done_sanity_check;
1018
0b2d3faa
JH
1019 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1020 * variables like done_sanity_check. */
10666ae3 1021 if (!done_sanity_check) {
ea471437 1022 unsigned int i = SVt_LAST;
10666ae3
NC
1023
1024 done_sanity_check = TRUE;
1025
1026 while (i--)
1027 assert (bodies_by_type[i].type == i);
1028 }
1029#endif
1030
02982131 1031 assert(arena_size);
23e9d66c 1032
99816f8d
NC
1033 /* may need new arena-set to hold new arena */
1034 if (!aroot || aroot->curr >= aroot->set_size) {
1035 struct arena_set *newroot;
1036 Newxz(newroot, 1, struct arena_set);
1037 newroot->set_size = ARENAS_PER_SET;
1038 newroot->next = aroot;
1039 aroot = newroot;
1040 PL_body_arenas = (void *) newroot;
1041 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1042 }
1043
1044 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1045 curr = aroot->curr++;
1046 adesc = &(aroot->set[curr]);
1047 assert(!adesc->arena);
1048
1049 Newx(adesc->arena, good_arena_size, char);
1050 adesc->size = good_arena_size;
1051 adesc->utype = sv_type;
1052 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1053 curr, (void*)adesc->arena, (UV)good_arena_size));
1054
1055 start = (char *) adesc->arena;
d2a0f284 1056
29657bb6
NC
1057 /* Get the address of the byte after the end of the last body we can fit.
1058 Remember, this is integer division: */
02982131 1059 end = start + good_arena_size / body_size * body_size;
d2a0f284 1060
486ec47a 1061 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1062#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1063 DEBUG_m(PerlIO_printf(Perl_debug_log,
1064 "arena %p end %p arena-size %d (from %d) type %d "
1065 "size %d ct %d\n",
02982131
NC
1066 (void*)start, (void*)end, (int)good_arena_size,
1067 (int)arena_size, sv_type, (int)body_size,
1068 (int)good_arena_size / (int)body_size));
d8fca402 1069#else
d2a0f284
JC
1070 DEBUG_m(PerlIO_printf(Perl_debug_log,
1071 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1072 (void*)start, (void*)end,
02982131
NC
1073 (int)arena_size, sv_type, (int)body_size,
1074 (int)good_arena_size / (int)body_size));
d8fca402 1075#endif
d2a0f284
JC
1076 *root = (void *)start;
1077
29657bb6
NC
1078 while (1) {
1079 /* Where the next body would start: */
d2a0f284 1080 char * const next = start + body_size;
29657bb6
NC
1081
1082 if (next >= end) {
1083 /* This is the last body: */
1084 assert(next == end);
1085
1086 *(void **)start = 0;
1087 return *root;
1088 }
1089
d2a0f284
JC
1090 *(void**) start = (void *)next;
1091 start = next;
1092 }
d2a0f284
JC
1093}
1094
1095/* grab a new thing from the free list, allocating more if necessary.
1096 The inline version is used for speed in hot routines, and the
1097 function using it serves the rest (unless PURIFY).
1098*/
1099#define new_body_inline(xpv, sv_type) \
1100 STMT_START { \
1101 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1102 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1103 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1104 bodies_by_type[sv_type].body_size,\
1105 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1106 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1107 } STMT_END
1108
1109#ifndef PURIFY
1110
1111STATIC void *
de37a194 1112S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1113{
1114 dVAR;
1115 void *xpv;
1116 new_body_inline(xpv, sv_type);
1117 return xpv;
1118}
1119
1120#endif
93e68bfb 1121
238b27b3
NC
1122static const struct body_details fake_rv =
1123 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1124
bd81e77b
NC
1125/*
1126=for apidoc sv_upgrade
93e68bfb 1127
bd81e77b
NC
1128Upgrade an SV to a more complex form. Generally adds a new body type to the
1129SV, then copies across as much information as possible from the old body.
9521ca61
FC
1130It croaks if the SV is already in a more complex form than requested. You
1131generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1132before calling C<sv_upgrade>, and hence does not croak. See also
1133C<svtype>.
93e68bfb 1134
bd81e77b 1135=cut
93e68bfb 1136*/
93e68bfb 1137
bd81e77b 1138void
aad570aa 1139Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
cac9b346 1140{
97aff369 1141 dVAR;
bd81e77b
NC
1142 void* old_body;
1143 void* new_body;
42d0e0b7 1144 const svtype old_type = SvTYPE(sv);
d2a0f284 1145 const struct body_details *new_type_details;
238b27b3 1146 const struct body_details *old_type_details
bd81e77b 1147 = bodies_by_type + old_type;
4df7f6af 1148 SV *referant = NULL;
cac9b346 1149
7918f24d
NC
1150 PERL_ARGS_ASSERT_SV_UPGRADE;
1151
1776cbe8
NC
1152 if (old_type == new_type)
1153 return;
1154
1155 /* This clause was purposefully added ahead of the early return above to
1156 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1157 inference by Nick I-S that it would fix other troublesome cases. See
1158 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1159
1160 Given that shared hash key scalars are no longer PVIV, but PV, there is
1161 no longer need to unshare so as to free up the IVX slot for its proper
1162 purpose. So it's safe to move the early return earlier. */
1163
bd81e77b
NC
1164 if (new_type != SVt_PV && SvIsCOW(sv)) {
1165 sv_force_normal_flags(sv, 0);
1166 }
cac9b346 1167
bd81e77b 1168 old_body = SvANY(sv);
de042e1d 1169
bd81e77b
NC
1170 /* Copying structures onto other structures that have been neatly zeroed
1171 has a subtle gotcha. Consider XPVMG
cac9b346 1172
bd81e77b
NC
1173 +------+------+------+------+------+-------+-------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH |
1175 +------+------+------+------+------+-------+-------+
1176 0 4 8 12 16 20 24 28
645c22ef 1177
bd81e77b
NC
1178 where NVs are aligned to 8 bytes, so that sizeof that structure is
1179 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1180
bd81e77b
NC
1181 +------+------+------+------+------+-------+-------+------+
1182 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1183 +------+------+------+------+------+-------+-------+------+
1184 0 4 8 12 16 20 24 28 32
08742458 1185
bd81e77b 1186 so what happens if you allocate memory for this structure:
30f9da9e 1187
bd81e77b
NC
1188 +------+------+------+------+------+-------+-------+------+------+...
1189 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 0 4 8 12 16 20 24 28 32 36
bfc44f79 1192
bd81e77b
NC
1193 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1194 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1195 started out as zero once, but it's quite possible that it isn't. So now,
1196 rather than a nicely zeroed GP, you have it pointing somewhere random.
1197 Bugs ensue.
bfc44f79 1198
bd81e77b
NC
1199 (In fact, GP ends up pointing at a previous GP structure, because the
1200 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1201 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1202 this happens to be moot because XPVGV has been re-ordered, with GP
1203 no longer after STASH)
30f9da9e 1204
bd81e77b
NC
1205 So we are careful and work out the size of used parts of all the
1206 structures. */
bfc44f79 1207
bd81e77b
NC
1208 switch (old_type) {
1209 case SVt_NULL:
1210 break;
1211 case SVt_IV:
4df7f6af
NC
1212 if (SvROK(sv)) {
1213 referant = SvRV(sv);
238b27b3
NC
1214 old_type_details = &fake_rv;
1215 if (new_type == SVt_NV)
1216 new_type = SVt_PVNV;
4df7f6af
NC
1217 } else {
1218 if (new_type < SVt_PVIV) {
1219 new_type = (new_type == SVt_NV)
1220 ? SVt_PVNV : SVt_PVIV;
1221 }
bd81e77b
NC
1222 }
1223 break;
1224 case SVt_NV:
1225 if (new_type < SVt_PVNV) {
1226 new_type = SVt_PVNV;
bd81e77b
NC
1227 }
1228 break;
bd81e77b
NC
1229 case SVt_PV:
1230 assert(new_type > SVt_PV);
1231 assert(SVt_IV < SVt_PV);
1232 assert(SVt_NV < SVt_PV);
1233 break;
1234 case SVt_PVIV:
1235 break;
1236 case SVt_PVNV:
1237 break;
1238 case SVt_PVMG:
1239 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1240 there's no way that it can be safely upgraded, because perl.c
1241 expects to Safefree(SvANY(PL_mess_sv)) */
1242 assert(sv != PL_mess_sv);
1243 /* This flag bit is used to mean other things in other scalar types.
1244 Given that it only has meaning inside the pad, it shouldn't be set
1245 on anything that can get upgraded. */
00b1698f 1246 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1247 break;
1248 default:
1249 if (old_type_details->cant_upgrade)
c81225bc
NC
1250 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1251 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1252 }
3376de98
NC
1253
1254 if (old_type > new_type)
1255 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1256 (int)old_type, (int)new_type);
1257
2fa1109b 1258 new_type_details = bodies_by_type + new_type;
645c22ef 1259
bd81e77b
NC
1260 SvFLAGS(sv) &= ~SVTYPEMASK;
1261 SvFLAGS(sv) |= new_type;
932e9ff9 1262
ab4416c0
NC
1263 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1264 the return statements above will have triggered. */
1265 assert (new_type != SVt_NULL);
bd81e77b 1266 switch (new_type) {
bd81e77b
NC
1267 case SVt_IV:
1268 assert(old_type == SVt_NULL);
1269 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1270 SvIV_set(sv, 0);
1271 return;
1272 case SVt_NV:
1273 assert(old_type == SVt_NULL);
1274 SvANY(sv) = new_XNV();
1275 SvNV_set(sv, 0);
1276 return;
bd81e77b 1277 case SVt_PVHV:
bd81e77b 1278 case SVt_PVAV:
d2a0f284 1279 assert(new_type_details->body_size);
c1ae03ae
NC
1280
1281#ifndef PURIFY
1282 assert(new_type_details->arena);
d2a0f284 1283 assert(new_type_details->arena_size);
c1ae03ae 1284 /* This points to the start of the allocated area. */
d2a0f284
JC
1285 new_body_inline(new_body, new_type);
1286 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1287 new_body = ((char *)new_body) - new_type_details->offset;
1288#else
1289 /* We always allocated the full length item with PURIFY. To do this
1290 we fake things so that arena is false for all 16 types.. */
1291 new_body = new_NOARENAZ(new_type_details);
1292#endif
1293 SvANY(sv) = new_body;
1294 if (new_type == SVt_PVAV) {
1295 AvMAX(sv) = -1;
1296 AvFILLp(sv) = -1;
1297 AvREAL_only(sv);
64484faa 1298 if (old_type_details->body_size) {
ac572bf4
NC
1299 AvALLOC(sv) = 0;
1300 } else {
1301 /* It will have been zeroed when the new body was allocated.
1302 Lets not write to it, in case it confuses a write-back
1303 cache. */
1304 }
78ac7dd9
NC
1305 } else {
1306 assert(!SvOK(sv));
1307 SvOK_off(sv);
1308#ifndef NODEFAULT_SHAREKEYS
1309 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1310#endif
1311 HvMAX(sv) = 7; /* (start with 8 buckets) */
c1ae03ae 1312 }
aeb18a1e 1313
bd81e77b
NC
1314 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1315 The target created by newSVrv also is, and it can have magic.
1316 However, it never has SvPVX set.
1317 */
4df7f6af
NC
1318 if (old_type == SVt_IV) {
1319 assert(!SvROK(sv));
1320 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1321 assert(SvPVX_const(sv) == 0);
1322 }
aeb18a1e 1323
bd81e77b 1324 if (old_type >= SVt_PVMG) {
e736a858 1325 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1326 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1327 } else {
1328 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1329 }
1330 break;
93e68bfb 1331
93e68bfb 1332
b9ad13ac
NC
1333 case SVt_REGEXP:
1334 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
1335 sv_force_normal_flags(sv) is called. */
1336 SvFAKE_on(sv);
bd81e77b
NC
1337 case SVt_PVIV:
1338 /* XXX Is this still needed? Was it ever needed? Surely as there is
1339 no route from NV to PVIV, NOK can never be true */
1340 assert(!SvNOKp(sv));
1341 assert(!SvNOK(sv));
1342 case SVt_PVIO:
1343 case SVt_PVFM:
bd81e77b
NC
1344 case SVt_PVGV:
1345 case SVt_PVCV:
1346 case SVt_PVLV:
1347 case SVt_PVMG:
1348 case SVt_PVNV:
1349 case SVt_PV:
93e68bfb 1350
d2a0f284 1351 assert(new_type_details->body_size);
bd81e77b
NC
1352 /* We always allocated the full length item with PURIFY. To do this
1353 we fake things so that arena is false for all 16 types.. */
1354 if(new_type_details->arena) {
1355 /* This points to the start of the allocated area. */
d2a0f284
JC
1356 new_body_inline(new_body, new_type);
1357 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1358 new_body = ((char *)new_body) - new_type_details->offset;
1359 } else {
1360 new_body = new_NOARENAZ(new_type_details);
1361 }
1362 SvANY(sv) = new_body;
5e2fc214 1363
bd81e77b 1364 if (old_type_details->copy) {
f9ba3d20
NC
1365 /* There is now the potential for an upgrade from something without
1366 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1367 int offset = old_type_details->offset;
1368 int length = old_type_details->copy;
1369
1370 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1371 const int difference
f9ba3d20
NC
1372 = new_type_details->offset - old_type_details->offset;
1373 offset += difference;
1374 length -= difference;
1375 }
1376 assert (length >= 0);
1377
1378 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1379 char);
bd81e77b
NC
1380 }
1381
1382#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1383 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1384 * correct 0.0 for us. Otherwise, if the old body didn't have an
1385 * NV slot, but the new one does, then we need to initialise the
1386 * freshly created NV slot with whatever the correct bit pattern is
1387 * for 0.0 */
e22a937e
NC
1388 if (old_type_details->zero_nv && !new_type_details->zero_nv
1389 && !isGV_with_GP(sv))
bd81e77b 1390 SvNV_set(sv, 0);
82048762 1391#endif
5e2fc214 1392
85dca89a
NC
1393 if (new_type == SVt_PVIO) {
1394 IO * const io = MUTABLE_IO(sv);
d963bf01 1395 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1396
1397 SvOBJECT_on(io);
1398 /* Clear the stashcache because a new IO could overrule a package
1399 name */
1400 hv_clear(PL_stashcache);
1401
85dca89a 1402 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1403 IoPAGE_LEN(sv) = 60;
85dca89a 1404 }
4df7f6af
NC
1405 if (old_type < SVt_PV) {
1406 /* referant will be NULL unless the old type was SVt_IV emulating
1407 SVt_RV */
1408 sv->sv_u.svu_rv = referant;
1409 }
bd81e77b
NC
1410 break;
1411 default:
afd78fd5
JH
1412 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1413 (unsigned long)new_type);
bd81e77b 1414 }
73171d91 1415
db93c0c4 1416 if (old_type > SVt_IV) {
bd81e77b 1417#ifdef PURIFY
beeec492 1418 safefree(old_body);
bd81e77b 1419#else
bc786448
GG
1420 /* Note that there is an assumption that all bodies of types that
1421 can be upgraded came from arenas. Only the more complex non-
1422 upgradable types are allowed to be directly malloc()ed. */
1423 assert(old_type_details->arena);
bd81e77b
NC
1424 del_body((void*)((char*)old_body + old_type_details->offset),
1425 &PL_body_roots[old_type]);
1426#endif
1427 }
1428}
73171d91 1429
bd81e77b
NC
1430/*
1431=for apidoc sv_backoff
73171d91 1432
fde67290 1433Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1434wrapper instead.
73171d91 1435
bd81e77b 1436=cut
73171d91
NC
1437*/
1438
bd81e77b 1439int
aad570aa 1440Perl_sv_backoff(pTHX_ register SV *const sv)
bd81e77b 1441{
69240efd 1442 STRLEN delta;
7a4bba22 1443 const char * const s = SvPVX_const(sv);
7918f24d
NC
1444
1445 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1446 PERL_UNUSED_CONTEXT;
7918f24d 1447
bd81e77b
NC
1448 assert(SvOOK(sv));
1449 assert(SvTYPE(sv) != SVt_PVHV);
1450 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1451
69240efd
NC
1452 SvOOK_offset(sv, delta);
1453
7a4bba22
NC
1454 SvLEN_set(sv, SvLEN(sv) + delta);
1455 SvPV_set(sv, SvPVX(sv) - delta);
1456 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1457 SvFLAGS(sv) &= ~SVf_OOK;
1458 return 0;
1459}
73171d91 1460
bd81e77b
NC
1461/*
1462=for apidoc sv_grow
73171d91 1463
bd81e77b
NC
1464Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1465upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1466Use the C<SvGROW> wrapper instead.
93e68bfb 1467
bd81e77b
NC
1468=cut
1469*/
93e68bfb 1470
bd81e77b 1471char *
aad570aa 1472Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
bd81e77b
NC
1473{
1474 register char *s;
93e68bfb 1475
7918f24d
NC
1476 PERL_ARGS_ASSERT_SV_GROW;
1477
5db06880
NC
1478 if (PL_madskills && newlen >= 0x100000) {
1479 PerlIO_printf(Perl_debug_log,
1480 "Allocation too large: %"UVxf"\n", (UV)newlen);
1481 }
bd81e77b
NC
1482#ifdef HAS_64K_LIMIT
1483 if (newlen >= 0x10000) {
1484 PerlIO_printf(Perl_debug_log,
1485 "Allocation too large: %"UVxf"\n", (UV)newlen);
1486 my_exit(1);
1487 }
1488#endif /* HAS_64K_LIMIT */
1489 if (SvROK(sv))
1490 sv_unref(sv);
1491 if (SvTYPE(sv) < SVt_PV) {
1492 sv_upgrade(sv, SVt_PV);
1493 s = SvPVX_mutable(sv);
1494 }
1495 else if (SvOOK(sv)) { /* pv is offset? */
1496 sv_backoff(sv);
1497 s = SvPVX_mutable(sv);
1498 if (newlen > SvLEN(sv))
1499 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1500#ifdef HAS_64K_LIMIT
1501 if (newlen >= 0x10000)
1502 newlen = 0xFFFF;
1503#endif
1504 }
1505 else
1506 s = SvPVX_mutable(sv);
aeb18a1e 1507
bd81e77b 1508 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1509 STRLEN minlen = SvCUR(sv);
1510 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1511 if (newlen < minlen)
1512 newlen = minlen;
aedff202 1513#ifndef Perl_safesysmalloc_size
bd81e77b 1514 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1515#endif
98653f18 1516 if (SvLEN(sv) && s) {
10edeb5d 1517 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1518 }
1519 else {
10edeb5d 1520 s = (char*)safemalloc(newlen);
bd81e77b
NC
1521 if (SvPVX_const(sv) && SvCUR(sv)) {
1522 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1523 }
1524 }
1525 SvPV_set(sv, s);
ca7c1a29 1526#ifdef Perl_safesysmalloc_size
98653f18
NC
1527 /* Do this here, do it once, do it right, and then we will never get
1528 called back into sv_grow() unless there really is some growing
1529 needed. */
ca7c1a29 1530 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1531#else
bd81e77b 1532 SvLEN_set(sv, newlen);
98653f18 1533#endif
bd81e77b
NC
1534 }
1535 return s;
1536}
aeb18a1e 1537
bd81e77b
NC
1538/*
1539=for apidoc sv_setiv
932e9ff9 1540
bd81e77b
NC
1541Copies an integer into the given SV, upgrading first if necessary.
1542Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1543
bd81e77b
NC
1544=cut
1545*/
463ee0b2 1546
bd81e77b 1547void
aad570aa 1548Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
bd81e77b 1549{
97aff369 1550 dVAR;
7918f24d
NC
1551
1552 PERL_ARGS_ASSERT_SV_SETIV;
1553
bd81e77b
NC
1554 SV_CHECK_THINKFIRST_COW_DROP(sv);
1555 switch (SvTYPE(sv)) {
1556 case SVt_NULL:
bd81e77b 1557 case SVt_NV:
3376de98 1558 sv_upgrade(sv, SVt_IV);
bd81e77b 1559 break;
bd81e77b
NC
1560 case SVt_PV:
1561 sv_upgrade(sv, SVt_PVIV);
1562 break;
463ee0b2 1563
bd81e77b 1564 case SVt_PVGV:
6e592b3a
BM
1565 if (!isGV_with_GP(sv))
1566 break;
bd81e77b
NC
1567 case SVt_PVAV:
1568 case SVt_PVHV:
1569 case SVt_PVCV:
1570 case SVt_PVFM:
1571 case SVt_PVIO:
22e74366 1572 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1573 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1574 OP_DESC(PL_op));
42d0e0b7 1575 default: NOOP;
bd81e77b
NC
1576 }
1577 (void)SvIOK_only(sv); /* validate number */
1578 SvIV_set(sv, i);
1579 SvTAINT(sv);
1580}
932e9ff9 1581
bd81e77b
NC
1582/*
1583=for apidoc sv_setiv_mg
d33b2eba 1584
bd81e77b 1585Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1586
bd81e77b
NC
1587=cut
1588*/
d33b2eba 1589
bd81e77b 1590void
aad570aa 1591Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
bd81e77b 1592{
7918f24d
NC
1593 PERL_ARGS_ASSERT_SV_SETIV_MG;
1594
bd81e77b
NC
1595 sv_setiv(sv,i);
1596 SvSETMAGIC(sv);
1597}
727879eb 1598
bd81e77b
NC
1599/*
1600=for apidoc sv_setuv
d33b2eba 1601
bd81e77b
NC
1602Copies an unsigned integer into the given SV, upgrading first if necessary.
1603Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1604
bd81e77b
NC
1605=cut
1606*/
d33b2eba 1607
bd81e77b 1608void
aad570aa 1609Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
bd81e77b 1610{
7918f24d
NC
1611 PERL_ARGS_ASSERT_SV_SETUV;
1612
bd81e77b
NC
1613 /* With these two if statements:
1614 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1615
bd81e77b
NC
1616 without
1617 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1618
bd81e77b
NC
1619 If you wish to remove them, please benchmark to see what the effect is
1620 */
1621 if (u <= (UV)IV_MAX) {
1622 sv_setiv(sv, (IV)u);
1623 return;
1624 }
1625 sv_setiv(sv, 0);
1626 SvIsUV_on(sv);
1627 SvUV_set(sv, u);
1628}
d33b2eba 1629
bd81e77b
NC
1630/*
1631=for apidoc sv_setuv_mg
727879eb 1632
bd81e77b 1633Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1634
bd81e77b
NC
1635=cut
1636*/
5e2fc214 1637
bd81e77b 1638void
aad570aa 1639Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
bd81e77b 1640{
7918f24d
NC
1641 PERL_ARGS_ASSERT_SV_SETUV_MG;
1642
bd81e77b
NC
1643 sv_setuv(sv,u);
1644 SvSETMAGIC(sv);
1645}
5e2fc214 1646
954c1994 1647/*
bd81e77b 1648=for apidoc sv_setnv
954c1994 1649
bd81e77b
NC
1650Copies a double into the given SV, upgrading first if necessary.
1651Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1652
1653=cut
1654*/
1655
63f97190 1656void
aad570aa 1657Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
79072805 1658{
97aff369 1659 dVAR;
7918f24d
NC
1660
1661 PERL_ARGS_ASSERT_SV_SETNV;
1662
bd81e77b
NC
1663 SV_CHECK_THINKFIRST_COW_DROP(sv);
1664 switch (SvTYPE(sv)) {
79072805 1665 case SVt_NULL:
79072805 1666 case SVt_IV:
bd81e77b 1667 sv_upgrade(sv, SVt_NV);
79072805
LW
1668 break;
1669 case SVt_PV:
79072805 1670 case SVt_PVIV:
bd81e77b 1671 sv_upgrade(sv, SVt_PVNV);
79072805 1672 break;
bd4b1eb5 1673
bd4b1eb5 1674 case SVt_PVGV:
6e592b3a
BM
1675 if (!isGV_with_GP(sv))
1676 break;
bd81e77b
NC
1677 case SVt_PVAV:
1678 case SVt_PVHV:
79072805 1679 case SVt_PVCV:
bd81e77b
NC
1680 case SVt_PVFM:
1681 case SVt_PVIO:
22e74366 1682 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1683 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1684 OP_DESC(PL_op));
42d0e0b7 1685 default: NOOP;
2068cd4d 1686 }
bd81e77b
NC
1687 SvNV_set(sv, num);
1688 (void)SvNOK_only(sv); /* validate number */
1689 SvTAINT(sv);
79072805
LW
1690}
1691
645c22ef 1692/*
bd81e77b 1693=for apidoc sv_setnv_mg
645c22ef 1694
bd81e77b 1695Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1696
1697=cut
1698*/
1699
bd81e77b 1700void
aad570aa 1701Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
79072805 1702{
7918f24d
NC
1703 PERL_ARGS_ASSERT_SV_SETNV_MG;
1704
bd81e77b
NC
1705 sv_setnv(sv,num);
1706 SvSETMAGIC(sv);
79072805
LW
1707}
1708
bd81e77b
NC
1709/* Print an "isn't numeric" warning, using a cleaned-up,
1710 * printable version of the offending string
1711 */
954c1994 1712
bd81e77b 1713STATIC void
aad570aa 1714S_not_a_number(pTHX_ SV *const sv)
79072805 1715{
97aff369 1716 dVAR;
bd81e77b
NC
1717 SV *dsv;
1718 char tmpbuf[64];
1719 const char *pv;
94463019 1720
7918f24d
NC
1721 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1722
94463019 1723 if (DO_UTF8(sv)) {
84bafc02 1724 dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1725 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1726 } else {
1727 char *d = tmpbuf;
551405c4 1728 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1729 /* each *s can expand to 4 chars + "...\0",
1730 i.e. need room for 8 chars */
ecdeb87c 1731
00b6aa41
AL
1732 const char *s = SvPVX_const(sv);
1733 const char * const end = s + SvCUR(sv);
1734 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1735 int ch = *s & 0xFF;
1736 if (ch & 128 && !isPRINT_LC(ch)) {
1737 *d++ = 'M';
1738 *d++ = '-';
1739 ch &= 127;
1740 }
1741 if (ch == '\n') {
1742 *d++ = '\\';
1743 *d++ = 'n';
1744 }
1745 else if (ch == '\r') {
1746 *d++ = '\\';
1747 *d++ = 'r';
1748 }
1749 else if (ch == '\f') {
1750 *d++ = '\\';
1751 *d++ = 'f';
1752 }
1753 else if (ch == '\\') {
1754 *d++ = '\\';
1755 *d++ = '\\';
1756 }
1757 else if (ch == '\0') {
1758 *d++ = '\\';
1759 *d++ = '0';
1760 }
1761 else if (isPRINT_LC(ch))
1762 *d++ = ch;
1763 else {
1764 *d++ = '^';
1765 *d++ = toCTRL(ch);
1766 }
1767 }
1768 if (s < end) {
1769 *d++ = '.';
1770 *d++ = '.';
1771 *d++ = '.';
1772 }
1773 *d = '\0';
1774 pv = tmpbuf;
a0d0e21e 1775 }
a0d0e21e 1776
533c011a 1777 if (PL_op)
9014280d 1778 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019
JH
1779 "Argument \"%s\" isn't numeric in %s", pv,
1780 OP_DESC(PL_op));
a0d0e21e 1781 else
9014280d 1782 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019 1783 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1784}
1785
c2988b20
NC
1786/*
1787=for apidoc looks_like_number
1788
645c22ef
DM
1789Test if the content of an SV looks like a number (or is a number).
1790C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1791non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1792ignored.
c2988b20
NC
1793
1794=cut
1795*/
1796
1797I32
aad570aa 1798Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1799{
a3b680e6 1800 register const char *sbegin;
c2988b20
NC
1801 STRLEN len;
1802
7918f24d
NC
1803 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1804
f52e41ad
FC
1805 if (SvPOK(sv) || SvPOKp(sv)) {
1806 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1807 }
c2988b20 1808 else
e0ab1c0e 1809 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1810 return grok_number(sbegin, len, NULL);
1811}
25da4f38 1812
19f6321d
NC
1813STATIC bool
1814S_glob_2number(pTHX_ GV * const gv)
180488f8 1815{
180488f8
NC
1816 SV *const buffer = sv_newmortal();
1817
7918f24d
NC
1818 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1819
180488f8 1820 gv_efullname3(buffer, gv, "*");
180488f8 1821
675c862f
AL
1822 /* We know that all GVs stringify to something that is not-a-number,
1823 so no need to test that. */
1824 if (ckWARN(WARN_NUMERIC))
1825 not_a_number(buffer);
1826 /* We just want something true to return, so that S_sv_2iuv_common
1827 can tail call us and return true. */
19f6321d 1828 return TRUE;
675c862f
AL
1829}
1830
25da4f38
IZ
1831/* Actually, ISO C leaves conversion of UV to IV undefined, but
1832 until proven guilty, assume that things are not that bad... */
1833
645c22ef
DM
1834/*
1835 NV_PRESERVES_UV:
1836
1837 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1838 an IV (an assumption perl has been based on to date) it becomes necessary
1839 to remove the assumption that the NV always carries enough precision to
1840 recreate the IV whenever needed, and that the NV is the canonical form.
1841 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1842 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1843 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1844 1) to distinguish between IV/UV/NV slots that have cached a valid
1845 conversion where precision was lost and IV/UV/NV slots that have a
1846 valid conversion which has lost no precision
645c22ef 1847 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1848 would lose precision, the precise conversion (or differently
1849 imprecise conversion) is also performed and cached, to prevent
1850 requests for different numeric formats on the same SV causing
1851 lossy conversion chains. (lossless conversion chains are perfectly
1852 acceptable (still))
1853
1854
1855 flags are used:
1856 SvIOKp is true if the IV slot contains a valid value
1857 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1858 SvNOKp is true if the NV slot contains a valid value
1859 SvNOK is true only if the NV value is accurate
1860
1861 so
645c22ef 1862 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1863 IV(or UV) would lose accuracy over a direct conversion from PV to
1864 IV(or UV). If it would, cache both conversions, return NV, but mark
1865 SV as IOK NOKp (ie not NOK).
1866
645c22ef 1867 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1868 NV would lose accuracy over a direct conversion from PV to NV. If it
1869 would, cache both conversions, flag similarly.
1870
1871 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1872 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1873 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1874 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1875 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1876
645c22ef
DM
1877 The benefit of this is that operations such as pp_add know that if
1878 SvIOK is true for both left and right operands, then integer addition
1879 can be used instead of floating point (for cases where the result won't
1880 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1881 loss of precision compared with integer addition.
1882
1883 * making IV and NV equal status should make maths accurate on 64 bit
1884 platforms
1885 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1886 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1887 looking for SvIOK and checking for overflow will not outweigh the
1888 fp to integer speedup)
1889 * will slow down integer operations (callers of SvIV) on "inaccurate"
1890 values, as the change from SvIOK to SvIOKp will cause a call into
1891 sv_2iv each time rather than a macro access direct to the IV slot
1892 * should speed up number->string conversion on integers as IV is
645c22ef 1893 favoured when IV and NV are equally accurate
28e5dec8
JH
1894
1895 ####################################################################
645c22ef
DM
1896 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1897 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1898 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1899 ####################################################################
1900
645c22ef 1901 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1902 performance ratio.
1903*/
1904
1905#ifndef NV_PRESERVES_UV
645c22ef
DM
1906# define IS_NUMBER_UNDERFLOW_IV 1
1907# define IS_NUMBER_UNDERFLOW_UV 2
1908# define IS_NUMBER_IV_AND_UV 2
1909# define IS_NUMBER_OVERFLOW_IV 4
1910# define IS_NUMBER_OVERFLOW_UV 5
1911
1912/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1913
1914/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1915STATIC int
5de3775c 1916S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1917# ifdef DEBUGGING
1918 , I32 numtype
1919# endif
1920 )
28e5dec8 1921{
97aff369 1922 dVAR;
7918f24d
NC
1923
1924 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1925
3f7c398e 1926 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
1927 if (SvNVX(sv) < (NV)IV_MIN) {
1928 (void)SvIOKp_on(sv);
1929 (void)SvNOK_on(sv);
45977657 1930 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1931 return IS_NUMBER_UNDERFLOW_IV;
1932 }
1933 if (SvNVX(sv) > (NV)UV_MAX) {
1934 (void)SvIOKp_on(sv);
1935 (void)SvNOK_on(sv);
1936 SvIsUV_on(sv);
607fa7f2 1937 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1938 return IS_NUMBER_OVERFLOW_UV;
1939 }
c2988b20
NC
1940 (void)SvIOKp_on(sv);
1941 (void)SvNOK_on(sv);
1942 /* Can't use strtol etc to convert this string. (See truth table in
1943 sv_2iv */
1944 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1945 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1946 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1947 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1948 } else {
1949 /* Integer is imprecise. NOK, IOKp */
1950 }
1951 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1952 }
1953 SvIsUV_on(sv);
607fa7f2 1954 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1955 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1956 if (SvUVX(sv) == UV_MAX) {
1957 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1958 possibly be preserved by NV. Hence, it must be overflow.
1959 NOK, IOKp */
1960 return IS_NUMBER_OVERFLOW_UV;
1961 }
1962 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1963 } else {
1964 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1965 }
c2988b20 1966 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1967}
645c22ef
DM
1968#endif /* !NV_PRESERVES_UV*/
1969
af359546 1970STATIC bool
7918f24d
NC
1971S_sv_2iuv_common(pTHX_ SV *const sv)
1972{
97aff369 1973 dVAR;
7918f24d
NC
1974
1975 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1976
af359546 1977 if (SvNOKp(sv)) {
28e5dec8
JH
1978 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1979 * without also getting a cached IV/UV from it at the same time
1980 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1981 * IV or UV at same time to avoid this. */
1982 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1983
1984 if (SvTYPE(sv) == SVt_NV)
1985 sv_upgrade(sv, SVt_PVNV);
1986
28e5dec8
JH
1987 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1988 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1989 certainly cast into the IV range at IV_MAX, whereas the correct
1990 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1991 cases go to UV */
cab190d4
JD
1992#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1993 if (Perl_isnan(SvNVX(sv))) {
1994 SvUV_set(sv, 0);
1995 SvIsUV_on(sv);
fdbe6d7c 1996 return FALSE;
cab190d4 1997 }
cab190d4 1998#endif
28e5dec8 1999 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2000 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2001 if (SvNVX(sv) == (NV) SvIVX(sv)
2002#ifndef NV_PRESERVES_UV
2003 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2004 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2005 /* Don't flag it as "accurately an integer" if the number
2006 came from a (by definition imprecise) NV operation, and
2007 we're outside the range of NV integer precision */
2008#endif
2009 ) {
a43d94f2
NC
2010 if (SvNOK(sv))
2011 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2012 else {
2013 /* scalar has trailing garbage, eg "42a" */
2014 }
28e5dec8 2015 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2016 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2017 PTR2UV(sv),
2018 SvNVX(sv),
2019 SvIVX(sv)));
2020
2021 } else {
2022 /* IV not precise. No need to convert from PV, as NV
2023 conversion would already have cached IV if it detected
2024 that PV->IV would be better than PV->NV->IV
2025 flags already correct - don't set public IOK. */
2026 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2027 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2028 PTR2UV(sv),
2029 SvNVX(sv),
2030 SvIVX(sv)));
2031 }
2032 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2033 but the cast (NV)IV_MIN rounds to a the value less (more
2034 negative) than IV_MIN which happens to be equal to SvNVX ??
2035 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2036 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2037 (NV)UVX == NVX are both true, but the values differ. :-(
2038 Hopefully for 2s complement IV_MIN is something like
2039 0x8000000000000000 which will be exact. NWC */
d460ef45 2040 }
25da4f38 2041 else {
607fa7f2 2042 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2043 if (
2044 (SvNVX(sv) == (NV) SvUVX(sv))
2045#ifndef NV_PRESERVES_UV
2046 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2047 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2048 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2049 /* Don't flag it as "accurately an integer" if the number
2050 came from a (by definition imprecise) NV operation, and
2051 we're outside the range of NV integer precision */
2052#endif
a43d94f2 2053 && SvNOK(sv)
28e5dec8
JH
2054 )
2055 SvIOK_on(sv);
25da4f38 2056 SvIsUV_on(sv);
1c846c1f 2057 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2058 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2059 PTR2UV(sv),
57def98f
JH
2060 SvUVX(sv),
2061 SvUVX(sv)));
25da4f38 2062 }
748a9306
LW
2063 }
2064 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2065 UV value;
504618e9 2066 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2067 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2068 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2069 the same as the direct translation of the initial string
2070 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2071 be careful to ensure that the value with the .456 is around if the
2072 NV value is requested in the future).
1c846c1f 2073
af359546 2074 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2075 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2076 cache the NV if we are sure it's not needed.
25da4f38 2077 */
16b7a9a4 2078
c2988b20
NC
2079 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2080 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2081 == IS_NUMBER_IN_UV) {
5e045b90 2082 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2083 if (SvTYPE(sv) < SVt_PVIV)
2084 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2085 (void)SvIOK_on(sv);
c2988b20
NC
2086 } else if (SvTYPE(sv) < SVt_PVNV)
2087 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2088
f2524eef 2089 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2090 we aren't going to call atof() below. If NVs don't preserve UVs
2091 then the value returned may have more precision than atof() will
2092 return, even though value isn't perfectly accurate. */
2093 if ((numtype & (IS_NUMBER_IN_UV
2094#ifdef NV_PRESERVES_UV
2095 | IS_NUMBER_NOT_INT
2096#endif
2097 )) == IS_NUMBER_IN_UV) {
2098 /* This won't turn off the public IOK flag if it was set above */
2099 (void)SvIOKp_on(sv);
2100
2101 if (!(numtype & IS_NUMBER_NEG)) {
2102 /* positive */;
2103 if (value <= (UV)IV_MAX) {
45977657 2104 SvIV_set(sv, (IV)value);
c2988b20 2105 } else {
af359546 2106 /* it didn't overflow, and it was positive. */
607fa7f2 2107 SvUV_set(sv, value);
c2988b20
NC
2108 SvIsUV_on(sv);
2109 }
2110 } else {
2111 /* 2s complement assumption */
2112 if (value <= (UV)IV_MIN) {
45977657 2113 SvIV_set(sv, -(IV)value);
c2988b20
NC
2114 } else {
2115 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2116 I'm assuming it will be rare. */
c2988b20
NC
2117 if (SvTYPE(sv) < SVt_PVNV)
2118 sv_upgrade(sv, SVt_PVNV);
2119 SvNOK_on(sv);
2120 SvIOK_off(sv);
2121 SvIOKp_on(sv);
9d6ce603 2122 SvNV_set(sv, -(NV)value);
45977657 2123 SvIV_set(sv, IV_MIN);
c2988b20
NC
2124 }
2125 }
2126 }
2127 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2128 will be in the previous block to set the IV slot, and the next
2129 block to set the NV slot. So no else here. */
2130
2131 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2132 != IS_NUMBER_IN_UV) {
2133 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2134 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2135
c2988b20
NC
2136 if (! numtype && ckWARN(WARN_NUMERIC))
2137 not_a_number(sv);
28e5dec8 2138
65202027 2139#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2140 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2141 PTR2UV(sv), SvNVX(sv)));
65202027 2142#else
1779d84d 2143 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2144 PTR2UV(sv), SvNVX(sv)));
65202027 2145#endif
28e5dec8 2146
28e5dec8 2147#ifdef NV_PRESERVES_UV
af359546
NC
2148 (void)SvIOKp_on(sv);
2149 (void)SvNOK_on(sv);
2150 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2151 SvIV_set(sv, I_V(SvNVX(sv)));
2152 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2153 SvIOK_on(sv);
2154 } else {
6f207bd3 2155 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2156 }
2157 /* UV will not work better than IV */
2158 } else {
2159 if (SvNVX(sv) > (NV)UV_MAX) {
2160 SvIsUV_on(sv);
2161 /* Integer is inaccurate. NOK, IOKp, is UV */
2162 SvUV_set(sv, UV_MAX);
af359546
NC
2163 } else {
2164 SvUV_set(sv, U_V(SvNVX(sv)));
2165 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2166 NV preservse UV so can do correct comparison. */
2167 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2168 SvIOK_on(sv);
af359546 2169 } else {
6f207bd3 2170 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2171 }
2172 }
4b0c9573 2173 SvIsUV_on(sv);
af359546 2174 }
28e5dec8 2175#else /* NV_PRESERVES_UV */
c2988b20
NC
2176 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2177 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2178 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2179 grok_number above. The NV slot has just been set using
2180 Atof. */
560b0c46 2181 SvNOK_on(sv);
c2988b20
NC
2182 assert (SvIOKp(sv));
2183 } else {
2184 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2185 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2186 /* Small enough to preserve all bits. */
2187 (void)SvIOKp_on(sv);
2188 SvNOK_on(sv);
45977657 2189 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2190 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2191 SvIOK_on(sv);
2192 /* Assumption: first non-preserved integer is < IV_MAX,
2193 this NV is in the preserved range, therefore: */
2194 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2195 < (UV)IV_MAX)) {
32fdb065 2196 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
2197 }
2198 } else {
2199 /* IN_UV NOT_INT
2200 0 0 already failed to read UV.
2201 0 1 already failed to read UV.
2202 1 0 you won't get here in this case. IV/UV
2203 slot set, public IOK, Atof() unneeded.
2204 1 1 already read UV.
2205 so there's no point in sv_2iuv_non_preserve() attempting
2206 to use atol, strtol, strtoul etc. */
47031da6 2207# ifdef DEBUGGING
40a17c4c 2208 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2209# else
2210 sv_2iuv_non_preserve (sv);
2211# endif
c2988b20
NC
2212 }
2213 }
28e5dec8 2214#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2215 /* It might be more code efficient to go through the entire logic above
2216 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2217 gets complex and potentially buggy, so more programmer efficient
2218 to do it this way, by turning off the public flags: */
2219 if (!numtype)
2220 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2221 }
af359546
NC
2222 }
2223 else {
675c862f 2224 if (isGV_with_GP(sv))
159b6efe 2225 return glob_2number(MUTABLE_GV(sv));
180488f8 2226
9a214eec 2227 if (!SvPADTMP(sv)) {
af359546
NC
2228 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2229 report_uninit(sv);
2230 }
25da4f38
IZ
2231 if (SvTYPE(sv) < SVt_IV)
2232 /* Typically the caller expects that sv_any is not NULL now. */
2233 sv_upgrade(sv, SVt_IV);
af359546
NC
2234 /* Return 0 from the caller. */
2235 return TRUE;
2236 }
2237 return FALSE;
2238}
2239
2240/*
2241=for apidoc sv_2iv_flags
2242
2243Return the integer value of an SV, doing any necessary string
2244conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2245Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2246
2247=cut
2248*/
2249
2250IV
5de3775c 2251Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2252{
97aff369 2253 dVAR;
af359546 2254 if (!sv)
a0d0e21e 2255 return 0;
a672f009 2256 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2257 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2258 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2259 In practice they are extremely unlikely to actually get anywhere
2260 accessible by user Perl code - the only way that I'm aware of is when
2261 a constant subroutine which is used as the second argument to index.
2262 */
af359546
NC
2263 if (flags & SV_GMAGIC)
2264 mg_get(sv);
2265 if (SvIOKp(sv))
2266 return SvIVX(sv);
2267 if (SvNOKp(sv)) {
2268 return I_V(SvNVX(sv));
2269 }
71c558c3
NC
2270 if (SvPOKp(sv) && SvLEN(sv)) {
2271 UV value;
2272 const int numtype
2273 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2274
2275 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2276 == IS_NUMBER_IN_UV) {
2277 /* It's definitely an integer */
2278 if (numtype & IS_NUMBER_NEG) {
2279 if (value < (UV)IV_MIN)
2280 return -(IV)value;
2281 } else {
2282 if (value < (UV)IV_MAX)
2283 return (IV)value;
2284 }
2285 }
2286 if (!numtype) {
2287 if (ckWARN(WARN_NUMERIC))
2288 not_a_number(sv);
2289 }
2290 return I_V(Atof(SvPVX_const(sv)));
2291 }
1c7ff15e
NC
2292 if (SvROK(sv)) {
2293 goto return_rok;
af359546 2294 }
1c7ff15e
NC
2295 assert(SvTYPE(sv) >= SVt_PVMG);
2296 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2297 } else if (SvTHINKFIRST(sv)) {
af359546 2298 if (SvROK(sv)) {
1c7ff15e 2299 return_rok:
af359546 2300 if (SvAMAGIC(sv)) {
aee036bb
DM
2301 SV * tmpstr;
2302 if (flags & SV_SKIP_OVERLOAD)
2303 return 0;
31d632c3 2304 tmpstr = AMG_CALLunary(sv, numer_amg);
af359546
NC
2305 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2306 return SvIV(tmpstr);
2307 }
2308 }
2309 return PTR2IV(SvRV(sv));
2310 }
2311 if (SvIsCOW(sv)) {
2312 sv_force_normal_flags(sv, 0);
2313 }
2314 if (SvREADONLY(sv) && !SvOK(sv)) {
2315 if (ckWARN(WARN_UNINITIALIZED))
2316 report_uninit(sv);
2317 return 0;
2318 }
2319 }
2320 if (!SvIOKp(sv)) {
2321 if (S_sv_2iuv_common(aTHX_ sv))
2322 return 0;
79072805 2323 }
1d7c1841
GS
2324 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2325 PTR2UV(sv),SvIVX(sv)));
25da4f38 2326 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2327}
2328
645c22ef 2329/*
891f9566 2330=for apidoc sv_2uv_flags
645c22ef
DM
2331
2332Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2333conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2334Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2335
2336=cut
2337*/
2338
ff68c719 2339UV
5de3775c 2340Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2341{
97aff369 2342 dVAR;
ff68c719
PP
2343 if (!sv)
2344 return 0;
a672f009 2345 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2346 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2347 the same flag bit as SVf_IVisUV, so must not let them cache IVs. */
891f9566
YST
2348 if (flags & SV_GMAGIC)
2349 mg_get(sv);
ff68c719
PP
2350 if (SvIOKp(sv))
2351 return SvUVX(sv);
2352 if (SvNOKp(sv))
2353 return U_V(SvNVX(sv));
71c558c3
NC
2354 if (SvPOKp(sv) && SvLEN(sv)) {
2355 UV value;
2356 const int numtype
2357 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2358
2359 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2360 == IS_NUMBER_IN_UV) {
2361 /* It's definitely an integer */
2362 if (!(numtype & IS_NUMBER_NEG))
2363 return value;
2364 }
2365 if (!numtype) {
2366 if (ckWARN(WARN_NUMERIC))
2367 not_a_number(sv);
2368 }
2369 return U_V(Atof(SvPVX_const(sv)));
2370 }
1c7ff15e
NC
2371 if (SvROK(sv)) {
2372 goto return_rok;
3fe9a6f1 2373 }
1c7ff15e
NC
2374 assert(SvTYPE(sv) >= SVt_PVMG);
2375 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2376 } else if (SvTHINKFIRST(sv)) {
ff68c719 2377 if (SvROK(sv)) {
1c7ff15e 2378 return_rok:
deb46114 2379 if (SvAMAGIC(sv)) {
aee036bb
DM
2380 SV *tmpstr;
2381 if (flags & SV_SKIP_OVERLOAD)
2382 return 0;
31d632c3 2383 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2384 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2385 return SvUV(tmpstr);
2386 }
2387 }
2388 return PTR2UV(SvRV(sv));
ff68c719 2389 }
765f542d
NC
2390 if (SvIsCOW(sv)) {
2391 sv_force_normal_flags(sv, 0);
8a818333 2392 }
0336b60e 2393 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2394 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2395 report_uninit(sv);
ff68c719
PP
2396 return 0;
2397 }
2398 }
af359546
NC
2399 if (!SvIOKp(sv)) {
2400 if (S_sv_2iuv_common(aTHX_ sv))
2401 return 0;
ff68c719 2402 }
25da4f38 2403
1d7c1841
GS
2404 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2405 PTR2UV(sv),SvUVX(sv)));
25da4f38 2406 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2407}
2408
645c22ef 2409/*
196007d1 2410=for apidoc sv_2nv_flags
645c22ef
DM
2411
2412Return the num value of an SV, doing any necessary string or integer
fde67290 2413conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2414Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2415
2416=cut
2417*/
2418
65202027 2419NV
39d5de13 2420Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
79072805 2421{
97aff369 2422 dVAR;
79072805
LW
2423 if (!sv)
2424 return 0.0;
a672f009 2425 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2426 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2427 the same flag bit as SVf_IVisUV, so must not let them cache NVs. */
39d5de13
DM
2428 if (flags & SV_GMAGIC)
2429 mg_get(sv);
463ee0b2
LW
2430 if (SvNOKp(sv))
2431 return SvNVX(sv);
0aa395f8 2432 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
041457d9 2433 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
504618e9 2434 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
a0d0e21e 2435 not_a_number(sv);
3f7c398e 2436 return Atof(SvPVX_const(sv));
a0d0e21e 2437 }
25da4f38 2438 if (SvIOKp(sv)) {
1c846c1f 2439 if (SvIsUV(sv))
65202027 2440 return (NV)SvUVX(sv);
25da4f38 2441 else
65202027 2442 return (NV)SvIVX(sv);
47a72cb8
NC
2443 }
2444 if (SvROK(sv)) {
2445 goto return_rok;
2446 }
2447 assert(SvTYPE(sv) >= SVt_PVMG);
2448 /* This falls through to the report_uninit near the end of the
2449 function. */
2450 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2451 if (SvROK(sv)) {
47a72cb8 2452 return_rok:
deb46114 2453 if (SvAMAGIC(sv)) {
aee036bb
DM
2454 SV *tmpstr;
2455 if (flags & SV_SKIP_OVERLOAD)
2456 return 0;
31d632c3 2457 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2458 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2459 return SvNV(tmpstr);
2460 }
2461 }
2462 return PTR2NV(SvRV(sv));
a0d0e21e 2463 }
765f542d
NC
2464 if (SvIsCOW(sv)) {
2465 sv_force_normal_flags(sv, 0);
8a818333 2466 }
0336b60e 2467 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2468 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2469 report_uninit(sv);
ed6116ce
LW
2470 return 0.0;
2471 }
79072805
LW
2472 }
2473 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2474 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2475 sv_upgrade(sv, SVt_NV);
906f284f 2476#ifdef USE_LONG_DOUBLE
097ee67d 2477 DEBUG_c({
f93f4e46 2478 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2479 PerlIO_printf(Perl_debug_log,
2480 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2481 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2482 RESTORE_NUMERIC_LOCAL();
2483 });
65202027 2484#else
572bbb43 2485 DEBUG_c({
f93f4e46 2486 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2487 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2488 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2489 RESTORE_NUMERIC_LOCAL();
2490 });
572bbb43 2491#endif
79072805
LW
2492 }
2493 else if (SvTYPE(sv) < SVt_PVNV)
2494 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2495 if (SvNOKp(sv)) {
2496 return SvNVX(sv);
61604483 2497 }
59d8ce62 2498 if (SvIOKp(sv)) {
9d6ce603 2499 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2500#ifdef NV_PRESERVES_UV
a43d94f2
NC
2501 if (SvIOK(sv))
2502 SvNOK_on(sv);
2503 else
2504 SvNOKp_on(sv);
28e5dec8
JH
2505#else
2506 /* Only set the public NV OK flag if this NV preserves the IV */
2507 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2508 if (SvIOK(sv) &&
2509 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2510 : (SvIVX(sv) == I_V(SvNVX(sv))))
2511 SvNOK_on(sv);
2512 else
2513 SvNOKp_on(sv);
2514#endif
93a17b20 2515 }
748a9306 2516 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2517 UV value;
3f7c398e 2518 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2519 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2520 not_a_number(sv);
28e5dec8 2521#ifdef NV_PRESERVES_UV
c2988b20
NC
2522 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2523 == IS_NUMBER_IN_UV) {
5e045b90 2524 /* It's definitely an integer */
9d6ce603 2525 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2526 } else
3f7c398e 2527 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2528 if (numtype)
2529 SvNOK_on(sv);
2530 else
2531 SvNOKp_on(sv);
28e5dec8 2532#else
3f7c398e 2533 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2534 /* Only set the public NV OK flag if this NV preserves the value in
2535 the PV at least as well as an IV/UV would.
2536 Not sure how to do this 100% reliably. */
2537 /* if that shift count is out of range then Configure's test is
2538 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2539 UV_BITS */
2540 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2541 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2542 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2543 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2544 /* Can't use strtol etc to convert this string, so don't try.
2545 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2546 SvNOK_on(sv);
2547 } else {
2548 /* value has been set. It may not be precise. */
2549 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2550 /* 2s complement assumption for (UV)IV_MIN */
2551 SvNOK_on(sv); /* Integer is too negative. */
2552 } else {
2553 SvNOKp_on(sv);
2554 SvIOKp_on(sv);
6fa402ec 2555
c2988b20 2556 if (numtype & IS_NUMBER_NEG) {
45977657 2557 SvIV_set(sv, -(IV)value);
c2988b20 2558 } else if (value <= (UV)IV_MAX) {
45977657 2559 SvIV_set(sv, (IV)value);
c2988b20 2560 } else {
607fa7f2 2561 SvUV_set(sv, value);
c2988b20
NC
2562 SvIsUV_on(sv);
2563 }
2564
2565 if (numtype & IS_NUMBER_NOT_INT) {
2566 /* I believe that even if the original PV had decimals,
2567 they are lost beyond the limit of the FP precision.
2568 However, neither is canonical, so both only get p
2569 flags. NWC, 2000/11/25 */
2570 /* Both already have p flags, so do nothing */
2571 } else {
66a1b24b 2572 const NV nv = SvNVX(sv);
c2988b20
NC
2573 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2574 if (SvIVX(sv) == I_V(nv)) {
2575 SvNOK_on(sv);
c2988b20 2576 } else {
c2988b20
NC
2577 /* It had no "." so it must be integer. */
2578 }
00b6aa41 2579 SvIOK_on(sv);
c2988b20
NC
2580 } else {
2581 /* between IV_MAX and NV(UV_MAX).
2582 Could be slightly > UV_MAX */
6fa402ec 2583
c2988b20
NC
2584 if (numtype & IS_NUMBER_NOT_INT) {
2585 /* UV and NV both imprecise. */
2586 } else {
66a1b24b 2587 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2588
2589 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2590 SvNOK_on(sv);
c2988b20 2591 }
00b6aa41 2592 SvIOK_on(sv);
c2988b20
NC
2593 }
2594 }
2595 }
2596 }
2597 }
a43d94f2
NC
2598 /* It might be more code efficient to go through the entire logic above
2599 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2600 gets complex and potentially buggy, so more programmer efficient
2601 to do it this way, by turning off the public flags: */
2602 if (!numtype)
2603 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2604#endif /* NV_PRESERVES_UV */
93a17b20 2605 }
79072805 2606 else {
f7877b28 2607 if (isGV_with_GP(sv)) {
159b6efe 2608 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2609 return 0.0;
2610 }
2611
9a214eec 2612 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2613 report_uninit(sv);
7e25a7e9
NC
2614 assert (SvTYPE(sv) >= SVt_NV);
2615 /* Typically the caller expects that sv_any is not NULL now. */
2616 /* XXX Ilya implies that this is a bug in callers that assume this
2617 and ideally should be fixed. */
a0d0e21e 2618 return 0.0;
79072805 2619 }
572bbb43 2620#if defined(USE_LONG_DOUBLE)
097ee67d 2621 DEBUG_c({
f93f4e46 2622 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2623 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2624 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2625 RESTORE_NUMERIC_LOCAL();
2626 });
65202027 2627#else
572bbb43 2628 DEBUG_c({
f93f4e46 2629 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2630 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2631 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2632 RESTORE_NUMERIC_LOCAL();
2633 });
572bbb43 2634#endif
463ee0b2 2635 return SvNVX(sv);
79072805
LW
2636}
2637
800401ee
JH
2638/*
2639=for apidoc sv_2num
2640
2641Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2642reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2643access this function.
800401ee
JH
2644
2645=cut
2646*/
2647
2648SV *
5de3775c 2649Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2650{
7918f24d
NC
2651 PERL_ARGS_ASSERT_SV_2NUM;
2652
b9ee0594
RGS
2653 if (!SvROK(sv))
2654 return sv;
800401ee 2655 if (SvAMAGIC(sv)) {
31d632c3 2656 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2657 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2658 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2659 return sv_2num(tmpsv);
2660 }
2661 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2662}
2663
645c22ef
DM
2664/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2665 * UV as a string towards the end of buf, and return pointers to start and
2666 * end of it.
2667 *
2668 * We assume that buf is at least TYPE_CHARS(UV) long.
2669 */
2670
864dbfa3 2671static char *
5de3775c 2672S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2673{
25da4f38 2674 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2675 char * const ebuf = ptr;
25da4f38 2676 int sign;
25da4f38 2677
7918f24d
NC
2678 PERL_ARGS_ASSERT_UIV_2BUF;
2679
25da4f38
IZ
2680 if (is_uv)
2681 sign = 0;
2682 else if (iv >= 0) {
2683 uv = iv;
2684 sign = 0;
2685 } else {
2686 uv = -iv;
2687 sign = 1;
2688 }
2689 do {
eb160463 2690 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2691 } while (uv /= 10);
2692 if (sign)
2693 *--ptr = '-';
2694 *peob = ebuf;
2695 return ptr;
2696}
2697
645c22ef
DM
2698/*
2699=for apidoc sv_2pv_flags
2700
ff276b08 2701Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2702If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2703string if necessary. Normally invoked via the C<SvPV_flags> macro.
2704C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2705
2706=cut
2707*/
2708
8d6d96c1 2709char *
5de3775c 2710Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2711{
97aff369 2712 dVAR;
79072805 2713 register char *s;
79072805 2714
463ee0b2 2715 if (!sv) {
cdb061a3
NC
2716 if (lp)
2717 *lp = 0;
73d840c0 2718 return (char *)"";
463ee0b2 2719 }
8990e307 2720 if (SvGMAGICAL(sv)) {
8d6d96c1
HS
2721 if (flags & SV_GMAGIC)
2722 mg_get(sv);
463ee0b2 2723 if (SvPOKp(sv)) {
cdb061a3
NC
2724 if (lp)
2725 *lp = SvCUR(sv);
10516c54
NC
2726 if (flags & SV_MUTABLE_RETURN)
2727 return SvPVX_mutable(sv);
4d84ee25
NC
2728 if (flags & SV_CONST_RETURN)
2729 return (char *)SvPVX_const(sv);
463ee0b2
LW
2730 return SvPVX(sv);
2731 }
75dfc8ec
NC
2732 if (SvIOKp(sv) || SvNOKp(sv)) {
2733 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
75dfc8ec
NC
2734 STRLEN len;
2735
2736 if (SvIOKp(sv)) {
e80fed9d 2737 len = SvIsUV(sv)
d9fad198
JH
2738 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2739 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
29912d93
NC
2740 } else if(SvNVX(sv) == 0.0) {
2741 tbuf[0] = '0';
2742 tbuf[1] = 0;
2743 len = 1;
75dfc8ec 2744 } else {
e8ada2d0
NC
2745 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2746 len = strlen(tbuf);
75dfc8ec 2747 }
b5b886f0
NC
2748 assert(!SvROK(sv));
2749 {
75dfc8ec
NC
2750 dVAR;
2751
75dfc8ec
NC
2752 SvUPGRADE(sv, SVt_PV);
2753 if (lp)
2754 *lp = len;
2755 s = SvGROW_mutable(sv, len + 1);
2756 SvCUR_set(sv, len);
2757 SvPOKp_on(sv);
10edeb5d 2758 return (char*)memcpy(s, tbuf, len + 1);
75dfc8ec 2759 }
463ee0b2 2760 }
1c7ff15e
NC
2761 if (SvROK(sv)) {
2762 goto return_rok;
2763 }
2764 assert(SvTYPE(sv) >= SVt_PVMG);
2765 /* This falls through to the report_uninit near the end of the
2766 function. */
2767 } else if (SvTHINKFIRST(sv)) {
ed6116ce 2768 if (SvROK(sv)) {
1c7ff15e 2769 return_rok:
deb46114 2770 if (SvAMAGIC(sv)) {
aee036bb
DM
2771 SV *tmpstr;
2772 if (flags & SV_SKIP_OVERLOAD)
2773 return NULL;
31d632c3 2774 tmpstr = AMG_CALLunary(sv, string_amg);
a02ec77a 2775 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
deb46114
NC
2776 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2777 /* Unwrap this: */
2778 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2779 */
2780
2781 char *pv;
2782 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2783 if (flags & SV_CONST_RETURN) {
2784 pv = (char *) SvPVX_const(tmpstr);
2785 } else {
2786 pv = (flags & SV_MUTABLE_RETURN)
2787 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2788 }
2789 if (lp)
2790 *lp = SvCUR(tmpstr);
50adf7d2 2791 } else {
deb46114 2792 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2793 }
deb46114
NC
2794 if (SvUTF8(tmpstr))
2795 SvUTF8_on(sv);
2796 else
2797 SvUTF8_off(sv);
2798 return pv;
50adf7d2 2799 }
deb46114
NC
2800 }
2801 {
fafee734
NC
2802 STRLEN len;
2803 char *retval;
2804 char *buffer;
d2c6dc5e 2805 SV *const referent = SvRV(sv);
d8eae41e
NC
2806
2807 if (!referent) {
fafee734
NC
2808 len = 7;
2809 retval = buffer = savepvn("NULLREF", len);
5c35adbb 2810 } else if (SvTYPE(referent) == SVt_REGEXP) {
d2c6dc5e 2811 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
67d2d14d
AB
2812 I32 seen_evals = 0;
2813
2814 assert(re);
2815
2816 /* If the regex is UTF-8 we want the containing scalar to
2817 have an UTF-8 flag too */
2818 if (RX_UTF8(re))
2819 SvUTF8_on(sv);
2820 else
2821 SvUTF8_off(sv);
2822
2823 if ((seen_evals = RX_SEEN_EVALS(re)))
2824 PL_reginterp_cnt += seen_evals;
2825
2826 if (lp)
2827 *lp = RX_WRAPLEN(re);
2828
2829 return RX_WRAPPED(re);
d8eae41e
NC
2830 } else {
2831 const char *const typestr = sv_reftype(referent, 0);
fafee734
NC
2832 const STRLEN typelen = strlen(typestr);
2833 UV addr = PTR2UV(referent);
2834 const char *stashname = NULL;
2835 STRLEN stashnamelen = 0; /* hush, gcc */
2836 const char *buffer_end;
d8eae41e 2837
d8eae41e 2838 if (SvOBJECT(referent)) {
fafee734
NC
2839 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2840
2841 if (name) {
2842 stashname = HEK_KEY(name);
2843 stashnamelen = HEK_LEN(name);
2844
2845 if (HEK_UTF8(name)) {
2846 SvUTF8_on(sv);
2847 } else {
2848 SvUTF8_off(sv);
2849 }
2850 } else {
2851 stashname = "__ANON__";
2852 stashnamelen = 8;
2853 }
2854 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2855 + 2 * sizeof(UV) + 2 /* )\0 */;
2856 } else {
2857 len = typelen + 3 /* (0x */
2858 + 2 * sizeof(UV) + 2 /* )\0 */;
d8eae41e 2859 }
fafee734
NC
2860
2861 Newx(buffer, len, char);
2862 buffer_end = retval = buffer + len;
2863
2864 /* Working backwards */
2865 *--retval = '\0';
2866 *--retval = ')';
2867 do {
2868 *--retval = PL_hexdigit[addr & 15];
2869 } while (addr >>= 4);
2870 *--retval = 'x';
2871 *--retval = '0';
2872 *--retval = '(';
2873
2874 retval -= typelen;
2875 memcpy(retval, typestr, typelen);
2876
2877 if (stashname) {
2878 *--retval = '=';
2879 retval -= stashnamelen;
2880 memcpy(retval, stashname, stashnamelen);
2881 }
486ec47a 2882 /* retval may not necessarily have reached the start of the
fafee734
NC
2883 buffer here. */
2884 assert (retval >= buffer);
2885
2886 len = buffer_end - retval - 1; /* -1 for that \0 */
c080367d 2887 }
042dae7a 2888 if (lp)
fafee734
NC
2889 *lp = len;
2890 SAVEFREEPV(buffer);
2891 return retval;
463ee0b2 2892 }
79072805 2893 }
0336b60e 2894 if (SvREADONLY(sv) && !SvOK(sv)) {
cdb061a3
NC
2895 if (lp)
2896 *lp = 0;
9f621bb0
NC
2897 if (flags & SV_UNDEF_RETURNS_NULL)
2898 return NULL;
2899 if (ckWARN(WARN_UNINITIALIZED))
2900 report_uninit(sv);
73d840c0 2901 return (char *)"";
79072805 2902 }
79072805 2903 }
28e5dec8
JH
2904 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2905 /* I'm assuming that if both IV and NV are equally valid then
2906 converting the IV is going to be more efficient */
e1ec3a88 2907 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2908 char buf[TYPE_CHARS(UV)];
2909 char *ebuf, *ptr;
97a130b8 2910 STRLEN len;
28e5dec8
JH
2911
2912 if (SvTYPE(sv) < SVt_PVIV)
2913 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2914 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2915 len = ebuf - ptr;
5902b6a9 2916 /* inlined from sv_setpvn */
97a130b8
NC
2917 s = SvGROW_mutable(sv, len + 1);
2918 Move(ptr, s, len, char);
2919 s += len;
28e5dec8 2920 *s = '\0';
28e5dec8
JH
2921 }
2922 else if (SvNOKp(sv)) {
79072805
LW
2923 if (SvTYPE(sv) < SVt_PVNV)
2924 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2925 if (SvNVX(sv) == 0.0) {
2926 s = SvGROW_mutable(sv, 2);
2927 *s++ = '0';
2928 *s = '\0';
2929 } else {
2930 dSAVE_ERRNO;
2931 /* The +20 is pure guesswork. Configure test needed. --jhi */
2932 s = SvGROW_mutable(sv, NV_DIG + 20);
2933 /* some Xenix systems wipe out errno here */
2d4389e4 2934 Gconvert(SvNVX(sv), NV_DIG, 0, s);
29912d93
NC
2935 RESTORE_ERRNO;
2936 while (*s) s++;
bbce6d69 2937 }
79072805
LW
2938#ifdef hcx
2939 if (s[-1] == '.')
46fc3d4c 2940 *--s = '\0';
79072805
LW
2941#endif
2942 }
79072805 2943 else {
8d1c3e26
NC
2944 if (isGV_with_GP(sv)) {
2945 GV *const gv = MUTABLE_GV(sv);
8d1c3e26
NC
2946 SV *const buffer = sv_newmortal();
2947
8d1c3e26 2948 gv_efullname3(buffer, gv, "*");
8d1c3e26 2949
52a6327b
FC
2950 assert(SvPOK(buffer));
2951 if (lp) {
1809c940 2952 *lp = SvCUR(buffer);
8d1c3e26 2953 }
52a6327b
FC
2954 if ( SvUTF8(buffer) ) SvUTF8_on(sv);
2955 return SvPVX(buffer);
8d1c3e26 2956 }
180488f8 2957
cdb061a3 2958 if (lp)
00b6aa41 2959 *lp = 0;
9f621bb0
NC
2960 if (flags & SV_UNDEF_RETURNS_NULL)
2961 return NULL;
9a214eec 2962 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2963 report_uninit(sv);
25da4f38
IZ
2964 if (SvTYPE(sv) < SVt_PV)
2965 /* Typically the caller expects that sv_any is not NULL now. */
2966 sv_upgrade(sv, SVt_PV);
73d840c0 2967 return (char *)"";
79072805 2968 }
cdb061a3 2969 {
823a54a3 2970 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2971 if (lp)
2972 *lp = len;
2973 SvCUR_set(sv, len);
2974 }
79072805 2975 SvPOK_on(sv);
1d7c1841 2976 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 2977 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
2978 if (flags & SV_CONST_RETURN)
2979 return (char *)SvPVX_const(sv);
10516c54
NC
2980 if (flags & SV_MUTABLE_RETURN)
2981 return SvPVX_mutable(sv);
463ee0b2
LW
2982 return SvPVX(sv);
2983}
2984
645c22ef 2985/*
6050d10e
JP
2986=for apidoc sv_copypv
2987
2988Copies a stringified representation of the source SV into the
2989destination SV. Automatically performs any necessary mg_get and
54f0641b 2990coercion of numeric values into strings. Guaranteed to preserve
2575c402 2991UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
2992sv_2pv[_flags] but operates directly on an SV instead of just the
2993string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
2994would lose the UTF-8'ness of the PV.
2995
2996=cut
2997*/
2998
2999void
5de3775c 3000Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 3001{
446eaa42 3002 STRLEN len;
53c1dcc0 3003 const char * const s = SvPV_const(ssv,len);
7918f24d
NC
3004
3005 PERL_ARGS_ASSERT_SV_COPYPV;
3006
cb50f42d 3007 sv_setpvn(dsv,s,len);
446eaa42 3008 if (SvUTF8(ssv))
cb50f42d 3009 SvUTF8_on(dsv);
446eaa42 3010 else
cb50f42d 3011 SvUTF8_off(dsv);
6050d10e
JP
3012}
3013
3014/*
645c22ef
DM
3015=for apidoc sv_2pvbyte
3016
3017Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3018to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3019side-effect.
3020
3021Usually accessed via the C<SvPVbyte> macro.
3022
3023=cut
3024*/
3025
7340a771 3026char *
5de3775c 3027Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3028{
7918f24d
NC
3029 PERL_ARGS_ASSERT_SV_2PVBYTE;
3030
71eb6d8c 3031 SvGETMAGIC(sv);
0875d2fe 3032 sv_utf8_downgrade(sv,0);
71eb6d8c 3033 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3034}
3035
645c22ef 3036/*
035cbb0e
RGS
3037=for apidoc sv_2pvutf8
3038
3039Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3040to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3041
3042Usually accessed via the C<SvPVutf8> macro.
3043
3044=cut
3045*/
645c22ef 3046
7340a771 3047char *
7bc54cea 3048Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3049{
7918f24d
NC
3050 PERL_ARGS_ASSERT_SV_2PVUTF8;
3051
035cbb0e
RGS
3052 sv_utf8_upgrade(sv);
3053 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771 3054}
1c846c1f 3055
7ee2227d 3056
645c22ef
DM
3057/*
3058=for apidoc sv_2bool
3059
06c841cf
FC
3060This macro is only used by sv_true() or its macro equivalent, and only if
3061the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3062It calls sv_2bool_flags with the SV_GMAGIC flag.
3063
3064=for apidoc sv_2bool_flags
3065
3066This function is only used by sv_true() and friends, and only if
fde67290 3067the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3068contain SV_GMAGIC, then it does an mg_get() first.
3069
645c22ef
DM
3070
3071=cut
3072*/
3073
463ee0b2 3074bool
06c841cf 3075Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
463ee0b2 3076{
97aff369 3077 dVAR;
7918f24d 3078
06c841cf 3079 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3080
06c841cf 3081 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3082
a0d0e21e
LW
3083 if (!SvOK(sv))
3084 return 0;
3085 if (SvROK(sv)) {
fabdb6c0 3086 if (SvAMAGIC(sv)) {
31d632c3 3087 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3088 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3089 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3090 }
3091 return SvRV(sv) != 0;
a0d0e21e 3092 }
463ee0b2 3093 if (SvPOKp(sv)) {
53c1dcc0
AL
3094 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3095 if (Xpvtmp &&
339049b0 3096 (*sv->sv_u.svu_pv > '0' ||
11343788 3097 Xpvtmp->xpv_cur > 1 ||
339049b0 3098 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
463ee0b2
LW
3099 return 1;
3100 else
3101 return 0;
3102 }
3103 else {
3104 if (SvIOKp(sv))
3105 return SvIVX(sv) != 0;
3106 else {
3107 if (SvNOKp(sv))
3108 return SvNVX(sv) != 0.0;
180488f8 3109 else {
f7877b28 3110 if (isGV_with_GP(sv))
180488f8
NC
3111 return TRUE;
3112 else
3113 return FALSE;
3114 }
463ee0b2
LW
3115 }
3116 }
79072805
LW
3117}
3118
c461cf8f
JH
3119/*
3120=for apidoc sv_utf8_upgrade
3121
78ea37eb 3122Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3123Forces the SV to string form if it is not already.
2bbc8d55 3124Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3125Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3126if the whole string is the same in UTF-8 as not.
3127Returns the number of bytes in the converted string
c461cf8f 3128
13a6c0e0
JH
3129This is not as a general purpose byte encoding to Unicode interface:
3130use the Encode extension for that.
3131
fe749c9a
KW
3132=for apidoc sv_utf8_upgrade_nomg
3133
fde67290 3134Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3135
8d6d96c1
HS
3136=for apidoc sv_utf8_upgrade_flags
3137
78ea37eb 3138Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3139Forces the SV to string form if it is not already.
8d6d96c1 3140Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3141if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3142will C<mg_get> on C<sv> if appropriate, else not.
3143Returns the number of bytes in the converted string
3144C<sv_utf8_upgrade> and
8d6d96c1
HS
3145C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3146
13a6c0e0
JH
3147This is not as a general purpose byte encoding to Unicode interface:
3148use the Encode extension for that.
3149
8d6d96c1 3150=cut
b3ab6785
KW
3151
3152The grow version is currently not externally documented. It adds a parameter,
3153extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3154have free after it upon return. This allows the caller to reserve extra space
3155that it intends to fill, to avoid extra grows.
3156
3157Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3158which can be used to tell this function to not first check to see if there are
3159any characters that are different in UTF-8 (variant characters) which would
3160force it to allocate a new string to sv, but to assume there are. Typically
3161this flag is used by a routine that has already parsed the string to find that
3162there are such characters, and passes this information on so that the work
3163doesn't have to be repeated.
3164
3165(One might think that the calling routine could pass in the position of the
3166first such variant, so it wouldn't have to be found again. But that is not the
3167case, because typically when the caller is likely to use this flag, it won't be
3168calling this routine unless it finds something that won't fit into a byte.
3169Otherwise it tries to not upgrade and just use bytes. But some things that
3170do fit into a byte are variants in utf8, and the caller may not have been
3171keeping track of these.)
3172
3173If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3174isn't guaranteed due to having other routines do the work in some input cases,
3175or if the input is already flagged as being in utf8.
3176
3177The speed of this could perhaps be improved for many cases if someone wanted to
3178write a fast function that counts the number of variant characters in a string,
3179especially if it could return the position of the first one.
3180
8d6d96c1
HS
3181*/
3182
3183STRLEN
b3ab6785 3184Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3185{
97aff369 3186 dVAR;
7918f24d 3187
b3ab6785 3188 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3189
808c356f
RGS
3190 if (sv == &PL_sv_undef)
3191 return 0;
e0e62c2a
NIS
3192 if (!SvPOK(sv)) {
3193 STRLEN len = 0;
d52b7888
NC
3194 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3195 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3196 if (SvUTF8(sv)) {
3197 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3198 return len;
b3ab6785 3199 }
d52b7888 3200 } else {
33fb6f35 3201 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3202 }
e0e62c2a 3203 }
4411f3b6 3204
f5cee72b 3205 if (SvUTF8(sv)) {
b3ab6785 3206 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3207 return SvCUR(sv);
f5cee72b 3208 }
5fec3b1d 3209
765f542d
NC
3210 if (SvIsCOW(sv)) {
3211 sv_force_normal_flags(sv, 0);
db42d148
NIS
3212 }
3213
b3ab6785 3214 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3215 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3216 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3217 return SvCUR(sv);
3218 }
3219
4e93345f
KW
3220 if (SvCUR(sv) == 0) {
3221 if (extra) SvGROW(sv, extra);
3222 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3223 /* This function could be much more efficient if we
2bbc8d55 3224 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3225 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3226 * make the loop as fast as possible (although there are certainly ways
3227 * to speed this up, eg. through vectorization) */
3228 U8 * s = (U8 *) SvPVX_const(sv);
3229 U8 * e = (U8 *) SvEND(sv);
3230 U8 *t = s;
3231 STRLEN two_byte_count = 0;
c4e7c712 3232
b3ab6785
KW
3233 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3234
3235 /* See if really will need to convert to utf8. We mustn't rely on our
3236 * incoming SV being well formed and having a trailing '\0', as certain
3237 * code in pp_formline can send us partially built SVs. */
3238
c4e7c712 3239 while (t < e) {
53c1dcc0 3240 const U8 ch = *t++;
b3ab6785
KW
3241 if (NATIVE_IS_INVARIANT(ch)) continue;
3242
3243 t--; /* t already incremented; re-point to first variant */
3244 two_byte_count = 1;
3245 goto must_be_utf8;
c4e7c712 3246 }
b3ab6785
KW
3247
3248 /* utf8 conversion not needed because all are invariants. Mark as
3249 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3250 SvUTF8_on(sv);
7f0bfbea 3251 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3252 return SvCUR(sv);
3253
3254must_be_utf8:
3255
3256 /* Here, the string should be converted to utf8, either because of an
3257 * input flag (two_byte_count = 0), or because a character that
3258 * requires 2 bytes was found (two_byte_count = 1). t points either to
3259 * the beginning of the string (if we didn't examine anything), or to
3260 * the first variant. In either case, everything from s to t - 1 will
3261 * occupy only 1 byte each on output.
3262 *
3263 * There are two main ways to convert. One is to create a new string
3264 * and go through the input starting from the beginning, appending each
3265 * converted value onto the new string as we go along. It's probably
3266 * best to allocate enough space in the string for the worst possible
3267 * case rather than possibly running out of space and having to
3268 * reallocate and then copy what we've done so far. Since everything
3269 * from s to t - 1 is invariant, the destination can be initialized
3270 * with these using a fast memory copy
3271 *
3272 * The other way is to figure out exactly how big the string should be
3273 * by parsing the entire input. Then you don't have to make it big
3274 * enough to handle the worst possible case, and more importantly, if
3275 * the string you already have is large enough, you don't have to
3276 * allocate a new string, you can copy the last character in the input
3277 * string to the final position(s) that will be occupied by the
3278 * converted string and go backwards, stopping at t, since everything
3279 * before that is invariant.
3280 *
3281 * There are advantages and disadvantages to each method.
3282 *
3283 * In the first method, we can allocate a new string, do the memory
3284 * copy from the s to t - 1, and then proceed through the rest of the
3285 * string byte-by-byte.
3286 *
3287 * In the second method, we proceed through the rest of the input
3288 * string just calculating how big the converted string will be. Then
3289 * there are two cases:
3290 * 1) if the string has enough extra space to handle the converted
3291 * value. We go backwards through the string, converting until we
3292 * get to the position we are at now, and then stop. If this
3293 * position is far enough along in the string, this method is
3294 * faster than the other method. If the memory copy were the same
3295 * speed as the byte-by-byte loop, that position would be about
3296 * half-way, as at the half-way mark, parsing to the end and back
3297 * is one complete string's parse, the same amount as starting
3298 * over and going all the way through. Actually, it would be
3299 * somewhat less than half-way, as it's faster to just count bytes
3300 * than to also copy, and we don't have the overhead of allocating
3301 * a new string, changing the scalar to use it, and freeing the
3302 * existing one. But if the memory copy is fast, the break-even
3303 * point is somewhere after half way. The counting loop could be
3304 * sped up by vectorization, etc, to move the break-even point
3305 * further towards the beginning.
3306 * 2) if the string doesn't have enough space to handle the converted
3307 * value. A new string will have to be allocated, and one might
3308 * as well, given that, start from the beginning doing the first
3309 * method. We've spent extra time parsing the string and in
3310 * exchange all we've gotten is that we know precisely how big to
3311 * make the new one. Perl is more optimized for time than space,
3312 * so this case is a loser.
3313 * So what I've decided to do is not use the 2nd method unless it is
3314 * guaranteed that a new string won't have to be allocated, assuming
3315 * the worst case. I also decided not to put any more conditions on it
3316 * than this, for now. It seems likely that, since the worst case is
3317 * twice as big as the unknown portion of the string (plus 1), we won't
3318 * be guaranteed enough space, causing us to go to the first method,
3319 * unless the string is short, or the first variant character is near
3320 * the end of it. In either of these cases, it seems best to use the
3321 * 2nd method. The only circumstance I can think of where this would
3322 * be really slower is if the string had once had much more data in it
3323 * than it does now, but there is still a substantial amount in it */
3324
3325 {
3326 STRLEN invariant_head = t - s;
3327 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3328 if (SvLEN(sv) < size) {
3329
3330 /* Here, have decided to allocate a new string */
3331
3332 U8 *dst;
3333 U8 *d;
3334
3335 Newx(dst, size, U8);
3336
3337 /* If no known invariants at the beginning of the input string,
3338 * set so starts from there. Otherwise, can use memory copy to
3339 * get up to where we are now, and then start from here */
3340
3341 if (invariant_head <= 0) {
3342 d = dst;
3343 } else {
3344 Copy(s, dst, invariant_head, char);
3345 d = dst + invariant_head;
3346 }
3347
3348 while (t < e) {
3349 const UV uv = NATIVE8_TO_UNI(*t++);
3350 if (UNI_IS_INVARIANT(uv))
3351 *d++ = (U8)UNI_TO_NATIVE(uv);
3352 else {
3353 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3354 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3355 }
3356 }
3357 *d = '\0';
3358 SvPV_free(sv); /* No longer using pre-existing string */
3359 SvPV_set(sv, (char*)dst);
3360 SvCUR_set(sv, d - dst);
3361 SvLEN_set(sv, size);
3362 } else {
3363
3364 /* Here, have decided to get the exact size of the string.
3365 * Currently this happens only when we know that there is
3366 * guaranteed enough space to fit the converted string, so
3367 * don't have to worry about growing. If two_byte_count is 0,
3368 * then t points to the first byte of the string which hasn't
3369 * been examined yet. Otherwise two_byte_count is 1, and t
3370 * points to the first byte in the string that will expand to
3371 * two. Depending on this, start examining at t or 1 after t.
3372 * */
3373
3374 U8 *d = t + two_byte_count;
3375
3376
3377 /* Count up the remaining bytes that expand to two */
3378
3379 while (d < e) {
3380 const U8 chr = *d++;
3381 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3382 }
3383
3384 /* The string will expand by just the number of bytes that
3385 * occupy two positions. But we are one afterwards because of
3386 * the increment just above. This is the place to put the
3387 * trailing NUL, and to set the length before we decrement */
3388
3389 d += two_byte_count;
3390 SvCUR_set(sv, d - s);
3391 *d-- = '\0';
3392
3393
3394 /* Having decremented d, it points to the position to put the
3395 * very last byte of the expanded string. Go backwards through
3396 * the string, copying and expanding as we go, stopping when we
3397 * get to the part that is invariant the rest of the way down */
3398
3399 e--;
3400 while (e >= t) {
3401 const U8 ch = NATIVE8_TO_UNI(*e--);
3402 if (UNI_IS_INVARIANT(ch)) {
3403 *d-- = UNI_TO_NATIVE(ch);
3404 } else {
3405 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3406 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3407 }
3408 }
3409 }
75da9d4c
DM
3410
3411 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3412 /* Update pos. We do it at the end rather than during
3413 * the upgrade, to avoid slowing down the common case
3414 * (upgrade without pos) */
3415 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3416 if (mg) {
3417 I32 pos = mg->mg_len;
3418 if (pos > 0 && (U32)pos > invariant_head) {
3419 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3420 STRLEN n = (U32)pos - invariant_head;
3421 while (n > 0) {
3422 if (UTF8_IS_START(*d))
3423 d++;
3424 d++;
3425 n--;
3426 }
3427 mg->mg_len = d - (U8*)SvPVX(sv);
3428 }
3429 }
3430 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3431 magic_setutf8(sv,mg); /* clear UTF8 cache */
3432 }
b3ab6785 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;
75da9d4c 3467 int mg_flags = SV_GMAGIC;
fa301091 3468
765f542d
NC
3469 if (SvIsCOW(sv)) {
3470 sv_force_normal_flags(sv, 0);
3471 }
75da9d4c
DM
3472 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3473 /* update pos */
3474 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3475 if (mg) {
3476 I32 pos = mg->mg_len;
3477 if (pos > 0) {
3478 sv_pos_b2u(sv, &pos);
3479 mg_flags = 0; /* sv_pos_b2u does get magic */
3480 mg->mg_len = pos;
3481 }
3482 }
3483 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3484 magic_setutf8(sv,mg); /* clear UTF8 cache */
3485
3486 }
3487 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3488
03cfe0ae 3489 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3490 if (fail_ok)
3491 return FALSE;
3492 else {
3493 if (PL_op)
3494 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3495 OP_DESC(PL_op));
fa301091
JH
3496 else
3497 Perl_croak(aTHX_ "Wide character");
3498 }
4b3603a4 3499 }
b162af07 3500 SvCUR_set(sv, len);
67e989fb 3501 }
560a288e 3502 }
ffebcc3e 3503 SvUTF8_off(sv);
560a288e
GS
3504 return TRUE;
3505}
3506
c461cf8f
JH
3507/*
3508=for apidoc sv_utf8_encode
3509
78ea37eb
ST
3510Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3511flag off so that it looks like octets again.
c461cf8f
JH
3512
3513=cut
3514*/
3515
560a288e 3516void
7bc54cea 3517Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3518{
7918f24d
NC
3519 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3520
4c94c214
NC
3521 if (SvIsCOW(sv)) {
3522 sv_force_normal_flags(sv, 0);
3523 }
3524 if (SvREADONLY(sv)) {
6ad8f254 3525 Perl_croak_no_modify(aTHX);
4c94c214 3526 }
a5f5288a 3527 (void) sv_utf8_upgrade(sv);
560a288e
GS
3528 SvUTF8_off(sv);
3529}
3530
4411f3b6
NIS
3531/*
3532=for apidoc sv_utf8_decode
3533
78ea37eb
ST
3534If the PV of the SV is an octet sequence in UTF-8
3535and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3536so that it looks like a character. If the PV contains only single-byte
694cf0d2 3537characters, the C<SvUTF8> flag stays off.
78ea37eb 3538Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3539
3540=cut
3541*/
3542
560a288e 3543bool
7bc54cea 3544Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3545{
7918f24d
NC
3546 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3547
78ea37eb 3548 if (SvPOKp(sv)) {
75da9d4c 3549 const U8 *start, *c;
93524f2b 3550 const U8 *e;
9cbac4c7 3551
645c22ef
DM
3552 /* The octets may have got themselves encoded - get them back as
3553 * bytes
3554 */
3555 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3556 return FALSE;
3557
3558 /* it is actually just a matter of turning the utf8 flag on, but
3559 * we want to make sure everything inside is valid utf8 first.
3560 */
75da9d4c 3561 c = start = (const U8 *) SvPVX_const(sv);
63cd0674 3562 if (!is_utf8_string(c, SvCUR(sv)+1))
67e989fb 3563 return FALSE;
93524f2b 3564 e = (const U8 *) SvEND(sv);
511c2ff0 3565 while (c < e) {
b64e5050 3566 const U8 ch = *c++;
c4d5f83a 3567 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3568 SvUTF8_on(sv);
3569 break;
3570 }
560a288e 3571 }
75da9d4c
DM
3572 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3573 /* adjust pos to the start of a UTF8 char sequence */
3574 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3575 if (mg) {
3576 I32 pos = mg->mg_len;
3577 if (pos > 0) {
3578 for (c = start + pos; c > start; c--) {
3579 if (UTF8_IS_START(*c))
3580 break;
3581 }
3582 mg->mg_len = c - start;
3583 }
3584 }
3585 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3586 magic_setutf8(sv,mg); /* clear UTF8 cache */
3587 }
560a288e
GS
3588 }
3589 return TRUE;
3590}
3591
954c1994
GS
3592/*
3593=for apidoc sv_setsv
3594
645c22ef
DM
3595Copies the contents of the source SV C<ssv> into the destination SV
3596C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3597function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3598Loosely speaking, it performs a copy-by-value, obliterating any previous
3599content of the destination.
3600
3601You probably want to use one of the assortment of wrappers, such as
3602C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3603C<SvSetMagicSV_nosteal>.
3604
8d6d96c1
HS
3605=for apidoc sv_setsv_flags
3606
645c22ef
DM
3607Copies the contents of the source SV C<ssv> into the destination SV
3608C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3609function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3610Loosely speaking, it performs a copy-by-value, obliterating any previous
3611content of the destination.
3612If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
fde67290
FC
3613C<ssv> if appropriate, else not. If the C<flags>
3614parameter has the C<NOSTEAL> bit set then the
3615buffers of temps will not be stolen. <sv_setsv>
5fcdf167 3616and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3617
3618You probably want to use one of the assortment of wrappers, such as
3619C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3620C<SvSetMagicSV_nosteal>.
3621
3622This is the primary function for copying scalars, and most other
3623copy-ish functions and macros use this underneath.
8d6d96c1
HS
3624
3625=cut
3626*/
3627
5d0301b7 3628static void
7bc54cea 3629S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3630{
c8bbf675 3631 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3e6edce2 3632 HV *old_stash = NULL;
dd69841b 3633
7918f24d
NC
3634 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3635
bec4f4b4 3636 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
5d0301b7
NC
3637 const char * const name = GvNAME(sstr);
3638 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3639 {
f7877b28
NC
3640 if (dtype >= SVt_PV) {
3641 SvPV_free(dstr);
3642 SvPV_set(dstr, 0);