This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Fill out sv_upgrade docs
[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
445Dump the contents of all SVs not yet freed. (Debugging aid).
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
566Attempt to destroy all objects not yet freed
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
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
673Deallocate the memory used by all arenas. Note that all the individual SV
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
bd81e77b
NC
1433Remove any string offset. You should normally use the C<SvOOK_off> macro
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
NC
1815{
1816 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1817 SV *const buffer = sv_newmortal();
1818
7918f24d
NC
1819 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1820
180488f8
NC
1821 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1822 is on. */
1823 SvFAKE_off(gv);
1824 gv_efullname3(buffer, gv, "*");
1825 SvFLAGS(gv) |= wasfake;
1826
675c862f
AL
1827 /* We know that all GVs stringify to something that is not-a-number,
1828 so no need to test that. */
1829 if (ckWARN(WARN_NUMERIC))
1830 not_a_number(buffer);
1831 /* We just want something true to return, so that S_sv_2iuv_common
1832 can tail call us and return true. */
19f6321d 1833 return TRUE;
675c862f
AL
1834}
1835
25da4f38
IZ
1836/* Actually, ISO C leaves conversion of UV to IV undefined, but
1837 until proven guilty, assume that things are not that bad... */
1838
645c22ef
DM
1839/*
1840 NV_PRESERVES_UV:
1841
1842 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1843 an IV (an assumption perl has been based on to date) it becomes necessary
1844 to remove the assumption that the NV always carries enough precision to
1845 recreate the IV whenever needed, and that the NV is the canonical form.
1846 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1847 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1848 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1849 1) to distinguish between IV/UV/NV slots that have cached a valid
1850 conversion where precision was lost and IV/UV/NV slots that have a
1851 valid conversion which has lost no precision
645c22ef 1852 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1853 would lose precision, the precise conversion (or differently
1854 imprecise conversion) is also performed and cached, to prevent
1855 requests for different numeric formats on the same SV causing
1856 lossy conversion chains. (lossless conversion chains are perfectly
1857 acceptable (still))
1858
1859
1860 flags are used:
1861 SvIOKp is true if the IV slot contains a valid value
1862 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1863 SvNOKp is true if the NV slot contains a valid value
1864 SvNOK is true only if the NV value is accurate
1865
1866 so
645c22ef 1867 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1868 IV(or UV) would lose accuracy over a direct conversion from PV to
1869 IV(or UV). If it would, cache both conversions, return NV, but mark
1870 SV as IOK NOKp (ie not NOK).
1871
645c22ef 1872 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1873 NV would lose accuracy over a direct conversion from PV to NV. If it
1874 would, cache both conversions, flag similarly.
1875
1876 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1877 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1878 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1879 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1880 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1881
645c22ef
DM
1882 The benefit of this is that operations such as pp_add know that if
1883 SvIOK is true for both left and right operands, then integer addition
1884 can be used instead of floating point (for cases where the result won't
1885 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1886 loss of precision compared with integer addition.
1887
1888 * making IV and NV equal status should make maths accurate on 64 bit
1889 platforms
1890 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1891 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1892 looking for SvIOK and checking for overflow will not outweigh the
1893 fp to integer speedup)
1894 * will slow down integer operations (callers of SvIV) on "inaccurate"
1895 values, as the change from SvIOK to SvIOKp will cause a call into
1896 sv_2iv each time rather than a macro access direct to the IV slot
1897 * should speed up number->string conversion on integers as IV is
645c22ef 1898 favoured when IV and NV are equally accurate
28e5dec8
JH
1899
1900 ####################################################################
645c22ef
DM
1901 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1902 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1903 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1904 ####################################################################
1905
645c22ef 1906 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1907 performance ratio.
1908*/
1909
1910#ifndef NV_PRESERVES_UV
645c22ef
DM
1911# define IS_NUMBER_UNDERFLOW_IV 1
1912# define IS_NUMBER_UNDERFLOW_UV 2
1913# define IS_NUMBER_IV_AND_UV 2
1914# define IS_NUMBER_OVERFLOW_IV 4
1915# define IS_NUMBER_OVERFLOW_UV 5
1916
1917/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1918
1919/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1920STATIC int
5de3775c 1921S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1922# ifdef DEBUGGING
1923 , I32 numtype
1924# endif
1925 )
28e5dec8 1926{
97aff369 1927 dVAR;
7918f24d
NC
1928
1929 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1930
3f7c398e 1931 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
1932 if (SvNVX(sv) < (NV)IV_MIN) {
1933 (void)SvIOKp_on(sv);
1934 (void)SvNOK_on(sv);
45977657 1935 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1936 return IS_NUMBER_UNDERFLOW_IV;
1937 }
1938 if (SvNVX(sv) > (NV)UV_MAX) {
1939 (void)SvIOKp_on(sv);
1940 (void)SvNOK_on(sv);
1941 SvIsUV_on(sv);
607fa7f2 1942 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1943 return IS_NUMBER_OVERFLOW_UV;
1944 }
c2988b20
NC
1945 (void)SvIOKp_on(sv);
1946 (void)SvNOK_on(sv);
1947 /* Can't use strtol etc to convert this string. (See truth table in
1948 sv_2iv */
1949 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1950 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1951 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1952 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1953 } else {
1954 /* Integer is imprecise. NOK, IOKp */
1955 }
1956 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1957 }
1958 SvIsUV_on(sv);
607fa7f2 1959 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1960 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1961 if (SvUVX(sv) == UV_MAX) {
1962 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1963 possibly be preserved by NV. Hence, it must be overflow.
1964 NOK, IOKp */
1965 return IS_NUMBER_OVERFLOW_UV;
1966 }
1967 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1968 } else {
1969 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1970 }
c2988b20 1971 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1972}
645c22ef
DM
1973#endif /* !NV_PRESERVES_UV*/
1974
af359546 1975STATIC bool
7918f24d
NC
1976S_sv_2iuv_common(pTHX_ SV *const sv)
1977{
97aff369 1978 dVAR;
7918f24d
NC
1979
1980 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1981
af359546 1982 if (SvNOKp(sv)) {
28e5dec8
JH
1983 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1984 * without also getting a cached IV/UV from it at the same time
1985 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1986 * IV or UV at same time to avoid this. */
1987 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1988
1989 if (SvTYPE(sv) == SVt_NV)
1990 sv_upgrade(sv, SVt_PVNV);
1991
28e5dec8
JH
1992 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1993 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1994 certainly cast into the IV range at IV_MAX, whereas the correct
1995 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1996 cases go to UV */
cab190d4
JD
1997#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1998 if (Perl_isnan(SvNVX(sv))) {
1999 SvUV_set(sv, 0);
2000 SvIsUV_on(sv);
fdbe6d7c 2001 return FALSE;
cab190d4 2002 }
cab190d4 2003#endif
28e5dec8 2004 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2005 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2006 if (SvNVX(sv) == (NV) SvIVX(sv)
2007#ifndef NV_PRESERVES_UV
2008 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2009 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2010 /* Don't flag it as "accurately an integer" if the number
2011 came from a (by definition imprecise) NV operation, and
2012 we're outside the range of NV integer precision */
2013#endif
2014 ) {
a43d94f2
NC
2015 if (SvNOK(sv))
2016 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2017 else {
2018 /* scalar has trailing garbage, eg "42a" */
2019 }
28e5dec8 2020 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2021 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2022 PTR2UV(sv),
2023 SvNVX(sv),
2024 SvIVX(sv)));
2025
2026 } else {
2027 /* IV not precise. No need to convert from PV, as NV
2028 conversion would already have cached IV if it detected
2029 that PV->IV would be better than PV->NV->IV
2030 flags already correct - don't set public IOK. */
2031 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2032 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2033 PTR2UV(sv),
2034 SvNVX(sv),
2035 SvIVX(sv)));
2036 }
2037 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2038 but the cast (NV)IV_MIN rounds to a the value less (more
2039 negative) than IV_MIN which happens to be equal to SvNVX ??
2040 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2041 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2042 (NV)UVX == NVX are both true, but the values differ. :-(
2043 Hopefully for 2s complement IV_MIN is something like
2044 0x8000000000000000 which will be exact. NWC */
d460ef45 2045 }
25da4f38 2046 else {
607fa7f2 2047 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2048 if (
2049 (SvNVX(sv) == (NV) SvUVX(sv))
2050#ifndef NV_PRESERVES_UV
2051 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2052 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2053 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2054 /* Don't flag it as "accurately an integer" if the number
2055 came from a (by definition imprecise) NV operation, and
2056 we're outside the range of NV integer precision */
2057#endif
a43d94f2 2058 && SvNOK(sv)
28e5dec8
JH
2059 )
2060 SvIOK_on(sv);
25da4f38 2061 SvIsUV_on(sv);
1c846c1f 2062 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2063 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2064 PTR2UV(sv),
57def98f
JH
2065 SvUVX(sv),
2066 SvUVX(sv)));
25da4f38 2067 }
748a9306
LW
2068 }
2069 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2070 UV value;
504618e9 2071 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2072 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2073 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2074 the same as the direct translation of the initial string
2075 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2076 be careful to ensure that the value with the .456 is around if the
2077 NV value is requested in the future).
1c846c1f 2078
af359546 2079 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2080 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2081 cache the NV if we are sure it's not needed.
25da4f38 2082 */
16b7a9a4 2083
c2988b20
NC
2084 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2085 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2086 == IS_NUMBER_IN_UV) {
5e045b90 2087 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2088 if (SvTYPE(sv) < SVt_PVIV)
2089 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2090 (void)SvIOK_on(sv);
c2988b20
NC
2091 } else if (SvTYPE(sv) < SVt_PVNV)
2092 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2093
f2524eef 2094 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2095 we aren't going to call atof() below. If NVs don't preserve UVs
2096 then the value returned may have more precision than atof() will
2097 return, even though value isn't perfectly accurate. */
2098 if ((numtype & (IS_NUMBER_IN_UV
2099#ifdef NV_PRESERVES_UV
2100 | IS_NUMBER_NOT_INT
2101#endif
2102 )) == IS_NUMBER_IN_UV) {
2103 /* This won't turn off the public IOK flag if it was set above */
2104 (void)SvIOKp_on(sv);
2105
2106 if (!(numtype & IS_NUMBER_NEG)) {
2107 /* positive */;
2108 if (value <= (UV)IV_MAX) {
45977657 2109 SvIV_set(sv, (IV)value);
c2988b20 2110 } else {
af359546 2111 /* it didn't overflow, and it was positive. */
607fa7f2 2112 SvUV_set(sv, value);
c2988b20
NC
2113 SvIsUV_on(sv);
2114 }
2115 } else {
2116 /* 2s complement assumption */
2117 if (value <= (UV)IV_MIN) {
45977657 2118 SvIV_set(sv, -(IV)value);
c2988b20
NC
2119 } else {
2120 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2121 I'm assuming it will be rare. */
c2988b20
NC
2122 if (SvTYPE(sv) < SVt_PVNV)
2123 sv_upgrade(sv, SVt_PVNV);
2124 SvNOK_on(sv);
2125 SvIOK_off(sv);
2126 SvIOKp_on(sv);
9d6ce603 2127 SvNV_set(sv, -(NV)value);
45977657 2128 SvIV_set(sv, IV_MIN);
c2988b20
NC
2129 }
2130 }
2131 }
2132 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2133 will be in the previous block to set the IV slot, and the next
2134 block to set the NV slot. So no else here. */
2135
2136 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2137 != IS_NUMBER_IN_UV) {
2138 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2139 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2140
c2988b20
NC
2141 if (! numtype && ckWARN(WARN_NUMERIC))
2142 not_a_number(sv);
28e5dec8 2143
65202027 2144#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2145 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2146 PTR2UV(sv), SvNVX(sv)));
65202027 2147#else
1779d84d 2148 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2149 PTR2UV(sv), SvNVX(sv)));
65202027 2150#endif
28e5dec8 2151
28e5dec8 2152#ifdef NV_PRESERVES_UV
af359546
NC
2153 (void)SvIOKp_on(sv);
2154 (void)SvNOK_on(sv);
2155 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2156 SvIV_set(sv, I_V(SvNVX(sv)));
2157 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2158 SvIOK_on(sv);
2159 } else {
6f207bd3 2160 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2161 }
2162 /* UV will not work better than IV */
2163 } else {
2164 if (SvNVX(sv) > (NV)UV_MAX) {
2165 SvIsUV_on(sv);
2166 /* Integer is inaccurate. NOK, IOKp, is UV */
2167 SvUV_set(sv, UV_MAX);
af359546
NC
2168 } else {
2169 SvUV_set(sv, U_V(SvNVX(sv)));
2170 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2171 NV preservse UV so can do correct comparison. */
2172 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2173 SvIOK_on(sv);
af359546 2174 } else {
6f207bd3 2175 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2176 }
2177 }
4b0c9573 2178 SvIsUV_on(sv);
af359546 2179 }
28e5dec8 2180#else /* NV_PRESERVES_UV */
c2988b20
NC
2181 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2182 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2183 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2184 grok_number above. The NV slot has just been set using
2185 Atof. */
560b0c46 2186 SvNOK_on(sv);
c2988b20
NC
2187 assert (SvIOKp(sv));
2188 } else {
2189 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2190 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2191 /* Small enough to preserve all bits. */
2192 (void)SvIOKp_on(sv);
2193 SvNOK_on(sv);
45977657 2194 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2195 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2196 SvIOK_on(sv);
2197 /* Assumption: first non-preserved integer is < IV_MAX,
2198 this NV is in the preserved range, therefore: */
2199 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2200 < (UV)IV_MAX)) {
32fdb065 2201 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
2202 }
2203 } else {
2204 /* IN_UV NOT_INT
2205 0 0 already failed to read UV.
2206 0 1 already failed to read UV.
2207 1 0 you won't get here in this case. IV/UV
2208 slot set, public IOK, Atof() unneeded.
2209 1 1 already read UV.
2210 so there's no point in sv_2iuv_non_preserve() attempting
2211 to use atol, strtol, strtoul etc. */
47031da6 2212# ifdef DEBUGGING
40a17c4c 2213 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2214# else
2215 sv_2iuv_non_preserve (sv);
2216# endif
c2988b20
NC
2217 }
2218 }
28e5dec8 2219#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2220 /* It might be more code efficient to go through the entire logic above
2221 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2222 gets complex and potentially buggy, so more programmer efficient
2223 to do it this way, by turning off the public flags: */
2224 if (!numtype)
2225 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2226 }
af359546
NC
2227 }
2228 else {
675c862f 2229 if (isGV_with_GP(sv))
159b6efe 2230 return glob_2number(MUTABLE_GV(sv));
180488f8 2231
9a214eec 2232 if (!SvPADTMP(sv)) {
af359546
NC
2233 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2234 report_uninit(sv);
2235 }
25da4f38
IZ
2236 if (SvTYPE(sv) < SVt_IV)
2237 /* Typically the caller expects that sv_any is not NULL now. */
2238 sv_upgrade(sv, SVt_IV);
af359546
NC
2239 /* Return 0 from the caller. */
2240 return TRUE;
2241 }
2242 return FALSE;
2243}
2244
2245/*
2246=for apidoc sv_2iv_flags
2247
2248Return the integer value of an SV, doing any necessary string
2249conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2250Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2251
2252=cut
2253*/
2254
2255IV
5de3775c 2256Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2257{
97aff369 2258 dVAR;
af359546 2259 if (!sv)
a0d0e21e 2260 return 0;
a672f009 2261 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2262 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2263 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2264 In practice they are extremely unlikely to actually get anywhere
2265 accessible by user Perl code - the only way that I'm aware of is when
2266 a constant subroutine which is used as the second argument to index.
2267 */
af359546
NC
2268 if (flags & SV_GMAGIC)
2269 mg_get(sv);
2270 if (SvIOKp(sv))
2271 return SvIVX(sv);
2272 if (SvNOKp(sv)) {
2273 return I_V(SvNVX(sv));
2274 }
71c558c3
NC
2275 if (SvPOKp(sv) && SvLEN(sv)) {
2276 UV value;
2277 const int numtype
2278 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2279
2280 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2281 == IS_NUMBER_IN_UV) {
2282 /* It's definitely an integer */
2283 if (numtype & IS_NUMBER_NEG) {
2284 if (value < (UV)IV_MIN)
2285 return -(IV)value;
2286 } else {
2287 if (value < (UV)IV_MAX)
2288 return (IV)value;
2289 }
2290 }
2291 if (!numtype) {
2292 if (ckWARN(WARN_NUMERIC))
2293 not_a_number(sv);
2294 }
2295 return I_V(Atof(SvPVX_const(sv)));
2296 }
1c7ff15e
NC
2297 if (SvROK(sv)) {
2298 goto return_rok;
af359546 2299 }
1c7ff15e
NC
2300 assert(SvTYPE(sv) >= SVt_PVMG);
2301 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2302 } else if (SvTHINKFIRST(sv)) {
af359546 2303 if (SvROK(sv)) {
1c7ff15e 2304 return_rok:
af359546 2305 if (SvAMAGIC(sv)) {
aee036bb
DM
2306 SV * tmpstr;
2307 if (flags & SV_SKIP_OVERLOAD)
2308 return 0;
31d632c3 2309 tmpstr = AMG_CALLunary(sv, numer_amg);
af359546
NC
2310 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2311 return SvIV(tmpstr);
2312 }
2313 }
2314 return PTR2IV(SvRV(sv));
2315 }
2316 if (SvIsCOW(sv)) {
2317 sv_force_normal_flags(sv, 0);
2318 }
2319 if (SvREADONLY(sv) && !SvOK(sv)) {
2320 if (ckWARN(WARN_UNINITIALIZED))
2321 report_uninit(sv);
2322 return 0;
2323 }
2324 }
2325 if (!SvIOKp(sv)) {
2326 if (S_sv_2iuv_common(aTHX_ sv))
2327 return 0;
79072805 2328 }
1d7c1841
GS
2329 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2330 PTR2UV(sv),SvIVX(sv)));
25da4f38 2331 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2332}
2333
645c22ef 2334/*
891f9566 2335=for apidoc sv_2uv_flags
645c22ef
DM
2336
2337Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2338conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2339Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2340
2341=cut
2342*/
2343
ff68c719 2344UV
5de3775c 2345Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2346{
97aff369 2347 dVAR;
ff68c719
PP
2348 if (!sv)
2349 return 0;
a672f009 2350 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2351 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2352 the same flag bit as SVf_IVisUV, so must not let them cache IVs. */
891f9566
YST
2353 if (flags & SV_GMAGIC)
2354 mg_get(sv);
ff68c719
PP
2355 if (SvIOKp(sv))
2356 return SvUVX(sv);
2357 if (SvNOKp(sv))
2358 return U_V(SvNVX(sv));
71c558c3
NC
2359 if (SvPOKp(sv) && SvLEN(sv)) {
2360 UV value;
2361 const int numtype
2362 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2363
2364 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2365 == IS_NUMBER_IN_UV) {
2366 /* It's definitely an integer */
2367 if (!(numtype & IS_NUMBER_NEG))
2368 return value;
2369 }
2370 if (!numtype) {
2371 if (ckWARN(WARN_NUMERIC))
2372 not_a_number(sv);
2373 }
2374 return U_V(Atof(SvPVX_const(sv)));
2375 }
1c7ff15e
NC
2376 if (SvROK(sv)) {
2377 goto return_rok;
3fe9a6f1 2378 }
1c7ff15e
NC
2379 assert(SvTYPE(sv) >= SVt_PVMG);
2380 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2381 } else if (SvTHINKFIRST(sv)) {
ff68c719 2382 if (SvROK(sv)) {
1c7ff15e 2383 return_rok:
deb46114 2384 if (SvAMAGIC(sv)) {
aee036bb
DM
2385 SV *tmpstr;
2386 if (flags & SV_SKIP_OVERLOAD)
2387 return 0;
31d632c3 2388 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2389 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2390 return SvUV(tmpstr);
2391 }
2392 }
2393 return PTR2UV(SvRV(sv));
ff68c719 2394 }
765f542d
NC
2395 if (SvIsCOW(sv)) {
2396 sv_force_normal_flags(sv, 0);
8a818333 2397 }
0336b60e 2398 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2399 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2400 report_uninit(sv);
ff68c719
PP
2401 return 0;
2402 }
2403 }
af359546
NC
2404 if (!SvIOKp(sv)) {
2405 if (S_sv_2iuv_common(aTHX_ sv))
2406 return 0;
ff68c719 2407 }
25da4f38 2408
1d7c1841
GS
2409 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2410 PTR2UV(sv),SvUVX(sv)));
25da4f38 2411 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2412}
2413
645c22ef 2414/*
196007d1 2415=for apidoc sv_2nv_flags
645c22ef
DM
2416
2417Return the num value of an SV, doing any necessary string or integer
39d5de13
DM
2418conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2419Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2420
2421=cut
2422*/
2423
65202027 2424NV
39d5de13 2425Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
79072805 2426{
97aff369 2427 dVAR;
79072805
LW
2428 if (!sv)
2429 return 0.0;
a672f009 2430 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2b2b6d6d
NC
2431 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2432 the same flag bit as SVf_IVisUV, so must not let them cache NVs. */
39d5de13
DM
2433 if (flags & SV_GMAGIC)
2434 mg_get(sv);
463ee0b2
LW
2435 if (SvNOKp(sv))
2436 return SvNVX(sv);
0aa395f8 2437 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
041457d9 2438 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
504618e9 2439 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
a0d0e21e 2440 not_a_number(sv);
3f7c398e 2441 return Atof(SvPVX_const(sv));
a0d0e21e 2442 }
25da4f38 2443 if (SvIOKp(sv)) {
1c846c1f 2444 if (SvIsUV(sv))
65202027 2445 return (NV)SvUVX(sv);
25da4f38 2446 else
65202027 2447 return (NV)SvIVX(sv);
47a72cb8
NC
2448 }
2449 if (SvROK(sv)) {
2450 goto return_rok;
2451 }
2452 assert(SvTYPE(sv) >= SVt_PVMG);
2453 /* This falls through to the report_uninit near the end of the
2454 function. */
2455 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2456 if (SvROK(sv)) {
47a72cb8 2457 return_rok:
deb46114 2458 if (SvAMAGIC(sv)) {
aee036bb
DM
2459 SV *tmpstr;
2460 if (flags & SV_SKIP_OVERLOAD)
2461 return 0;
31d632c3 2462 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2463 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2464 return SvNV(tmpstr);
2465 }
2466 }
2467 return PTR2NV(SvRV(sv));
a0d0e21e 2468 }
765f542d
NC
2469 if (SvIsCOW(sv)) {
2470 sv_force_normal_flags(sv, 0);
8a818333 2471 }
0336b60e 2472 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2473 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2474 report_uninit(sv);
ed6116ce
LW
2475 return 0.0;
2476 }
79072805
LW
2477 }
2478 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2479 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2480 sv_upgrade(sv, SVt_NV);
906f284f 2481#ifdef USE_LONG_DOUBLE
097ee67d 2482 DEBUG_c({
f93f4e46 2483 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2484 PerlIO_printf(Perl_debug_log,
2485 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2486 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2487 RESTORE_NUMERIC_LOCAL();
2488 });
65202027 2489#else
572bbb43 2490 DEBUG_c({
f93f4e46 2491 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2492 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2493 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2494 RESTORE_NUMERIC_LOCAL();
2495 });
572bbb43 2496#endif
79072805
LW
2497 }
2498 else if (SvTYPE(sv) < SVt_PVNV)
2499 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2500 if (SvNOKp(sv)) {
2501 return SvNVX(sv);
61604483 2502 }
59d8ce62 2503 if (SvIOKp(sv)) {
9d6ce603 2504 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2505#ifdef NV_PRESERVES_UV
a43d94f2
NC
2506 if (SvIOK(sv))
2507 SvNOK_on(sv);
2508 else
2509 SvNOKp_on(sv);
28e5dec8
JH
2510#else
2511 /* Only set the public NV OK flag if this NV preserves the IV */
2512 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2513 if (SvIOK(sv) &&
2514 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2515 : (SvIVX(sv) == I_V(SvNVX(sv))))
2516 SvNOK_on(sv);
2517 else
2518 SvNOKp_on(sv);
2519#endif
93a17b20 2520 }
748a9306 2521 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2522 UV value;
3f7c398e 2523 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2524 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2525 not_a_number(sv);
28e5dec8 2526#ifdef NV_PRESERVES_UV
c2988b20
NC
2527 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2528 == IS_NUMBER_IN_UV) {
5e045b90 2529 /* It's definitely an integer */
9d6ce603 2530 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2531 } else
3f7c398e 2532 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2533 if (numtype)
2534 SvNOK_on(sv);
2535 else
2536 SvNOKp_on(sv);
28e5dec8 2537#else
3f7c398e 2538 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2539 /* Only set the public NV OK flag if this NV preserves the value in
2540 the PV at least as well as an IV/UV would.
2541 Not sure how to do this 100% reliably. */
2542 /* if that shift count is out of range then Configure's test is
2543 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2544 UV_BITS */
2545 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2546 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2547 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2548 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2549 /* Can't use strtol etc to convert this string, so don't try.
2550 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2551 SvNOK_on(sv);
2552 } else {
2553 /* value has been set. It may not be precise. */
2554 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2555 /* 2s complement assumption for (UV)IV_MIN */
2556 SvNOK_on(sv); /* Integer is too negative. */
2557 } else {
2558 SvNOKp_on(sv);
2559 SvIOKp_on(sv);
6fa402ec 2560
c2988b20 2561 if (numtype & IS_NUMBER_NEG) {
45977657 2562 SvIV_set(sv, -(IV)value);
c2988b20 2563 } else if (value <= (UV)IV_MAX) {
45977657 2564 SvIV_set(sv, (IV)value);
c2988b20 2565 } else {
607fa7f2 2566 SvUV_set(sv, value);
c2988b20
NC
2567 SvIsUV_on(sv);
2568 }
2569
2570 if (numtype & IS_NUMBER_NOT_INT) {
2571 /* I believe that even if the original PV had decimals,
2572 they are lost beyond the limit of the FP precision.
2573 However, neither is canonical, so both only get p
2574 flags. NWC, 2000/11/25 */
2575 /* Both already have p flags, so do nothing */
2576 } else {
66a1b24b 2577 const NV nv = SvNVX(sv);
c2988b20
NC
2578 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2579 if (SvIVX(sv) == I_V(nv)) {
2580 SvNOK_on(sv);
c2988b20 2581 } else {
c2988b20
NC
2582 /* It had no "." so it must be integer. */
2583 }
00b6aa41 2584 SvIOK_on(sv);
c2988b20
NC
2585 } else {
2586 /* between IV_MAX and NV(UV_MAX).
2587 Could be slightly > UV_MAX */
6fa402ec 2588
c2988b20
NC
2589 if (numtype & IS_NUMBER_NOT_INT) {
2590 /* UV and NV both imprecise. */
2591 } else {
66a1b24b 2592 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2593
2594 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2595 SvNOK_on(sv);
c2988b20 2596 }
00b6aa41 2597 SvIOK_on(sv);
c2988b20
NC
2598 }
2599 }
2600 }
2601 }
2602 }
a43d94f2
NC
2603 /* It might be more code efficient to go through the entire logic above
2604 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2605 gets complex and potentially buggy, so more programmer efficient
2606 to do it this way, by turning off the public flags: */
2607 if (!numtype)
2608 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2609#endif /* NV_PRESERVES_UV */
93a17b20 2610 }
79072805 2611 else {
f7877b28 2612 if (isGV_with_GP(sv)) {
159b6efe 2613 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2614 return 0.0;
2615 }
2616
9a214eec 2617 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2618 report_uninit(sv);
7e25a7e9
NC
2619 assert (SvTYPE(sv) >= SVt_NV);
2620 /* Typically the caller expects that sv_any is not NULL now. */
2621 /* XXX Ilya implies that this is a bug in callers that assume this
2622 and ideally should be fixed. */
a0d0e21e 2623 return 0.0;
79072805 2624 }
572bbb43 2625#if defined(USE_LONG_DOUBLE)
097ee67d 2626 DEBUG_c({
f93f4e46 2627 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2628 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2629 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2630 RESTORE_NUMERIC_LOCAL();
2631 });
65202027 2632#else
572bbb43 2633 DEBUG_c({
f93f4e46 2634 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2635 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2636 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2637 RESTORE_NUMERIC_LOCAL();
2638 });
572bbb43 2639#endif
463ee0b2 2640 return SvNVX(sv);
79072805
LW
2641}
2642
800401ee
JH
2643/*
2644=for apidoc sv_2num
2645
2646Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2647reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2648access this function.
800401ee
JH
2649
2650=cut
2651*/
2652
2653SV *
5de3775c 2654Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2655{
7918f24d
NC
2656 PERL_ARGS_ASSERT_SV_2NUM;
2657
b9ee0594
RGS
2658 if (!SvROK(sv))
2659 return sv;
800401ee 2660 if (SvAMAGIC(sv)) {
31d632c3 2661 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2662 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2663 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2664 return sv_2num(tmpsv);
2665 }
2666 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2667}
2668
645c22ef
DM
2669/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2670 * UV as a string towards the end of buf, and return pointers to start and
2671 * end of it.
2672 *
2673 * We assume that buf is at least TYPE_CHARS(UV) long.
2674 */
2675
864dbfa3 2676static char *
5de3775c 2677S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2678{
25da4f38 2679 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2680 char * const ebuf = ptr;
25da4f38 2681 int sign;
25da4f38 2682
7918f24d
NC
2683 PERL_ARGS_ASSERT_UIV_2BUF;
2684
25da4f38
IZ
2685 if (is_uv)
2686 sign = 0;
2687 else if (iv >= 0) {
2688 uv = iv;
2689 sign = 0;
2690 } else {
2691 uv = -iv;
2692 sign = 1;
2693 }
2694 do {
eb160463 2695 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2696 } while (uv /= 10);
2697 if (sign)
2698 *--ptr = '-';
2699 *peob = ebuf;
2700 return ptr;
2701}
2702
645c22ef
DM
2703/*
2704=for apidoc sv_2pv_flags
2705
ff276b08 2706Returns a pointer to the string value of an SV, and sets *lp to its length.
645c22ef
DM
2707If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2708if necessary.
2709Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2710usually end up here too.
2711
2712=cut
2713*/
2714
8d6d96c1 2715char *
5de3775c 2716Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2717{
97aff369 2718 dVAR;
79072805 2719 register char *s;
79072805 2720
463ee0b2 2721 if (!sv) {
cdb061a3
NC
2722 if (lp)
2723 *lp = 0;
73d840c0 2724 return (char *)"";
463ee0b2 2725 }
8990e307 2726 if (SvGMAGICAL(sv)) {
8d6d96c1
HS
2727 if (flags & SV_GMAGIC)
2728 mg_get(sv);
463ee0b2 2729 if (SvPOKp(sv)) {
cdb061a3
NC
2730 if (lp)
2731 *lp = SvCUR(sv);
10516c54
NC
2732 if (flags & SV_MUTABLE_RETURN)
2733 return SvPVX_mutable(sv);
4d84ee25
NC
2734 if (flags & SV_CONST_RETURN)
2735 return (char *)SvPVX_const(sv);
463ee0b2
LW
2736 return SvPVX(sv);
2737 }
75dfc8ec
NC
2738 if (SvIOKp(sv) || SvNOKp(sv)) {
2739 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
75dfc8ec
NC
2740 STRLEN len;
2741
2742 if (SvIOKp(sv)) {
e80fed9d 2743 len = SvIsUV(sv)
d9fad198
JH
2744 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2745 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
29912d93
NC
2746 } else if(SvNVX(sv) == 0.0) {
2747 tbuf[0] = '0';
2748 tbuf[1] = 0;
2749 len = 1;
75dfc8ec 2750 } else {
e8ada2d0
NC
2751 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2752 len = strlen(tbuf);
75dfc8ec 2753 }
b5b886f0
NC
2754 assert(!SvROK(sv));
2755 {
75dfc8ec
NC
2756 dVAR;
2757
75dfc8ec
NC
2758 SvUPGRADE(sv, SVt_PV);
2759 if (lp)
2760 *lp = len;
2761 s = SvGROW_mutable(sv, len + 1);
2762 SvCUR_set(sv, len);
2763 SvPOKp_on(sv);
10edeb5d 2764 return (char*)memcpy(s, tbuf, len + 1);
75dfc8ec 2765 }
463ee0b2 2766 }
1c7ff15e
NC
2767 if (SvROK(sv)) {
2768 goto return_rok;
2769 }
2770 assert(SvTYPE(sv) >= SVt_PVMG);
2771 /* This falls through to the report_uninit near the end of the
2772 function. */
2773 } else if (SvTHINKFIRST(sv)) {
ed6116ce 2774 if (SvROK(sv)) {
1c7ff15e 2775 return_rok:
deb46114 2776 if (SvAMAGIC(sv)) {
aee036bb
DM
2777 SV *tmpstr;
2778 if (flags & SV_SKIP_OVERLOAD)
2779 return NULL;
31d632c3 2780 tmpstr = AMG_CALLunary(sv, string_amg);
a02ec77a 2781 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
deb46114
NC
2782 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2783 /* Unwrap this: */
2784 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2785 */
2786
2787 char *pv;
2788 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2789 if (flags & SV_CONST_RETURN) {
2790 pv = (char *) SvPVX_const(tmpstr);
2791 } else {
2792 pv = (flags & SV_MUTABLE_RETURN)
2793 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2794 }
2795 if (lp)
2796 *lp = SvCUR(tmpstr);
50adf7d2 2797 } else {
deb46114 2798 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2799 }
deb46114
NC
2800 if (SvUTF8(tmpstr))
2801 SvUTF8_on(sv);
2802 else
2803 SvUTF8_off(sv);
2804 return pv;
50adf7d2 2805 }
deb46114
NC
2806 }
2807 {
fafee734
NC
2808 STRLEN len;
2809 char *retval;
2810 char *buffer;
d2c6dc5e 2811 SV *const referent = SvRV(sv);
d8eae41e
NC
2812
2813 if (!referent) {
fafee734
NC
2814 len = 7;
2815 retval = buffer = savepvn("NULLREF", len);
5c35adbb 2816 } else if (SvTYPE(referent) == SVt_REGEXP) {
d2c6dc5e 2817 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
67d2d14d
AB
2818 I32 seen_evals = 0;
2819
2820 assert(re);
2821
2822 /* If the regex is UTF-8 we want the containing scalar to
2823 have an UTF-8 flag too */
2824 if (RX_UTF8(re))
2825 SvUTF8_on(sv);
2826 else
2827 SvUTF8_off(sv);
2828
2829 if ((seen_evals = RX_SEEN_EVALS(re)))
2830 PL_reginterp_cnt += seen_evals;
2831
2832 if (lp)
2833 *lp = RX_WRAPLEN(re);
2834
2835 return RX_WRAPPED(re);
d8eae41e
NC
2836 } else {
2837 const char *const typestr = sv_reftype(referent, 0);
fafee734
NC
2838 const STRLEN typelen = strlen(typestr);
2839 UV addr = PTR2UV(referent);
2840 const char *stashname = NULL;
2841 STRLEN stashnamelen = 0; /* hush, gcc */
2842 const char *buffer_end;
d8eae41e 2843
d8eae41e 2844 if (SvOBJECT(referent)) {
fafee734
NC
2845 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2846
2847 if (name) {
2848 stashname = HEK_KEY(name);
2849 stashnamelen = HEK_LEN(name);
2850
2851 if (HEK_UTF8(name)) {
2852 SvUTF8_on(sv);
2853 } else {
2854 SvUTF8_off(sv);
2855 }
2856 } else {
2857 stashname = "__ANON__";
2858 stashnamelen = 8;
2859 }
2860 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2861 + 2 * sizeof(UV) + 2 /* )\0 */;
2862 } else {
2863 len = typelen + 3 /* (0x */
2864 + 2 * sizeof(UV) + 2 /* )\0 */;
d8eae41e 2865 }
fafee734
NC
2866
2867 Newx(buffer, len, char);
2868 buffer_end = retval = buffer + len;
2869
2870 /* Working backwards */
2871 *--retval = '\0';
2872 *--retval = ')';
2873 do {
2874 *--retval = PL_hexdigit[addr & 15];
2875 } while (addr >>= 4);
2876 *--retval = 'x';
2877 *--retval = '0';
2878 *--retval = '(';
2879
2880 retval -= typelen;
2881 memcpy(retval, typestr, typelen);
2882
2883 if (stashname) {
2884 *--retval = '=';
2885 retval -= stashnamelen;
2886 memcpy(retval, stashname, stashnamelen);
2887 }
486ec47a 2888 /* retval may not necessarily have reached the start of the
fafee734
NC
2889 buffer here. */
2890 assert (retval >= buffer);
2891
2892 len = buffer_end - retval - 1; /* -1 for that \0 */
c080367d 2893 }
042dae7a 2894 if (lp)
fafee734
NC
2895 *lp = len;
2896 SAVEFREEPV(buffer);
2897 return retval;
463ee0b2 2898 }
79072805 2899 }
0336b60e 2900 if (SvREADONLY(sv) && !SvOK(sv)) {
cdb061a3
NC
2901 if (lp)
2902 *lp = 0;
9f621bb0
NC
2903 if (flags & SV_UNDEF_RETURNS_NULL)
2904 return NULL;
2905 if (ckWARN(WARN_UNINITIALIZED))
2906 report_uninit(sv);
73d840c0 2907 return (char *)"";
79072805 2908 }
79072805 2909 }
28e5dec8
JH
2910 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2911 /* I'm assuming that if both IV and NV are equally valid then
2912 converting the IV is going to be more efficient */
e1ec3a88 2913 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2914 char buf[TYPE_CHARS(UV)];
2915 char *ebuf, *ptr;
97a130b8 2916 STRLEN len;
28e5dec8
JH
2917
2918 if (SvTYPE(sv) < SVt_PVIV)
2919 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2920 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2921 len = ebuf - ptr;
5902b6a9 2922 /* inlined from sv_setpvn */
97a130b8
NC
2923 s = SvGROW_mutable(sv, len + 1);
2924 Move(ptr, s, len, char);
2925 s += len;
28e5dec8 2926 *s = '\0';
28e5dec8
JH
2927 }
2928 else if (SvNOKp(sv)) {
79072805
LW
2929 if (SvTYPE(sv) < SVt_PVNV)
2930 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2931 if (SvNVX(sv) == 0.0) {
2932 s = SvGROW_mutable(sv, 2);
2933 *s++ = '0';
2934 *s = '\0';
2935 } else {
2936 dSAVE_ERRNO;
2937 /* The +20 is pure guesswork. Configure test needed. --jhi */
2938 s = SvGROW_mutable(sv, NV_DIG + 20);
2939 /* some Xenix systems wipe out errno here */
2d4389e4 2940 Gconvert(SvNVX(sv), NV_DIG, 0, s);
29912d93
NC
2941 RESTORE_ERRNO;
2942 while (*s) s++;
bbce6d69 2943 }
79072805
LW
2944#ifdef hcx
2945 if (s[-1] == '.')
46fc3d4c 2946 *--s = '\0';
79072805
LW
2947#endif
2948 }
79072805 2949 else {
8d1c3e26
NC
2950 if (isGV_with_GP(sv)) {
2951 GV *const gv = MUTABLE_GV(sv);
2952 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
2953 SV *const buffer = sv_newmortal();
2954
2955 /* FAKE globs can get coerced, so need to turn this off temporarily
2956 if it is on. */
2957 SvFAKE_off(gv);
2958 gv_efullname3(buffer, gv, "*");
2959 SvFLAGS(gv) |= wasfake;
2960
1809c940
DM
2961 if (SvPOK(buffer)) {
2962 if (lp) {
2963 *lp = SvCUR(buffer);
2964 }
bc1c56e9 2965 if ( SvUTF8(buffer) ) SvUTF8_on(sv);
1809c940
DM
2966 return SvPVX(buffer);
2967 }
2968 else {
2969 if (lp)
2970 *lp = 0;
2971 return (char *)"";
8d1c3e26 2972 }
8d1c3e26 2973 }
180488f8 2974
cdb061a3 2975 if (lp)
00b6aa41 2976 *lp = 0;
9f621bb0
NC
2977 if (flags & SV_UNDEF_RETURNS_NULL)
2978 return NULL;
9a214eec 2979 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2980 report_uninit(sv);
25da4f38
IZ
2981 if (SvTYPE(sv) < SVt_PV)
2982 /* Typically the caller expects that sv_any is not NULL now. */
2983 sv_upgrade(sv, SVt_PV);
73d840c0 2984 return (char *)"";
79072805 2985 }
cdb061a3 2986 {
823a54a3 2987 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2988 if (lp)
2989 *lp = len;
2990 SvCUR_set(sv, len);
2991 }
79072805 2992 SvPOK_on(sv);
1d7c1841 2993 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 2994 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
2995 if (flags & SV_CONST_RETURN)
2996 return (char *)SvPVX_const(sv);
10516c54
NC
2997 if (flags & SV_MUTABLE_RETURN)
2998 return SvPVX_mutable(sv);
463ee0b2
LW
2999 return SvPVX(sv);
3000}
3001
645c22ef 3002/*
6050d10e
JP
3003=for apidoc sv_copypv
3004
3005Copies a stringified representation of the source SV into the
3006destination SV. Automatically performs any necessary mg_get and
54f0641b 3007coercion of numeric values into strings. Guaranteed to preserve
2575c402 3008UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3009sv_2pv[_flags] but operates directly on an SV instead of just the
3010string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3011would lose the UTF-8'ness of the PV.
3012
3013=cut
3014*/
3015
3016void
5de3775c 3017Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 3018{
446eaa42 3019 STRLEN len;
53c1dcc0 3020 const char * const s = SvPV_const(ssv,len);
7918f24d
NC
3021
3022 PERL_ARGS_ASSERT_SV_COPYPV;
3023
cb50f42d 3024 sv_setpvn(dsv,s,len);
446eaa42 3025 if (SvUTF8(ssv))
cb50f42d 3026 SvUTF8_on(dsv);
446eaa42 3027 else
cb50f42d 3028 SvUTF8_off(dsv);
6050d10e
JP
3029}
3030
3031/*
645c22ef
DM
3032=for apidoc sv_2pvbyte
3033
3034Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3035to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3036side-effect.
3037
3038Usually accessed via the C<SvPVbyte> macro.
3039
3040=cut
3041*/
3042
7340a771 3043char *
5de3775c 3044Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3045{
7918f24d
NC
3046 PERL_ARGS_ASSERT_SV_2PVBYTE;
3047
71eb6d8c 3048 SvGETMAGIC(sv);
0875d2fe 3049 sv_utf8_downgrade(sv,0);
71eb6d8c 3050 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3051}
3052
645c22ef 3053/*
035cbb0e
RGS
3054=for apidoc sv_2pvutf8
3055
3056Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3057to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3058
3059Usually accessed via the C<SvPVutf8> macro.
3060
3061=cut
3062*/
645c22ef 3063
7340a771 3064char *
7bc54cea 3065Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3066{
7918f24d
NC
3067 PERL_ARGS_ASSERT_SV_2PVUTF8;
3068
035cbb0e
RGS
3069 sv_utf8_upgrade(sv);
3070 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771 3071}
1c846c1f 3072
7ee2227d 3073
645c22ef
DM
3074/*
3075=for apidoc sv_2bool
3076
06c841cf
FC
3077This macro is only used by sv_true() or its macro equivalent, and only if
3078the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3079It calls sv_2bool_flags with the SV_GMAGIC flag.
3080
3081=for apidoc sv_2bool_flags
3082
3083This function is only used by sv_true() and friends, and only if
3084the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3085contain SV_GMAGIC, then it does an mg_get() first.
3086
645c22ef
DM
3087
3088=cut
3089*/
3090
463ee0b2 3091bool
06c841cf 3092Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
463ee0b2 3093{
97aff369 3094 dVAR;
7918f24d 3095
06c841cf 3096 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3097
06c841cf 3098 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3099
a0d0e21e
LW
3100 if (!SvOK(sv))
3101 return 0;
3102 if (SvROK(sv)) {
fabdb6c0 3103 if (SvAMAGIC(sv)) {
31d632c3 3104 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3105 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3106 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3107 }
3108 return SvRV(sv) != 0;
a0d0e21e 3109 }
463ee0b2 3110 if (SvPOKp(sv)) {
53c1dcc0
AL
3111 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3112 if (Xpvtmp &&
339049b0 3113 (*sv->sv_u.svu_pv > '0' ||
11343788 3114 Xpvtmp->xpv_cur > 1 ||
339049b0 3115 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
463ee0b2
LW
3116 return 1;
3117 else
3118 return 0;
3119 }
3120 else {
3121 if (SvIOKp(sv))
3122 return SvIVX(sv) != 0;
3123 else {
3124 if (SvNOKp(sv))
3125 return SvNVX(sv) != 0.0;
180488f8 3126 else {
f7877b28 3127 if (isGV_with_GP(sv))
180488f8
NC
3128 return TRUE;
3129 else
3130 return FALSE;
3131 }
463ee0b2
LW
3132 }
3133 }
79072805
LW
3134}
3135
c461cf8f
JH
3136/*
3137=for apidoc sv_utf8_upgrade
3138
78ea37eb 3139Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3140Forces the SV to string form if it is not already.
2bbc8d55 3141Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3142Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3143if the whole string is the same in UTF-8 as not.
3144Returns the number of bytes in the converted string
c461cf8f 3145
13a6c0e0
JH
3146This is not as a general purpose byte encoding to Unicode interface:
3147use the Encode extension for that.
3148
fe749c9a
KW
3149=for apidoc sv_utf8_upgrade_nomg
3150
3151Like sv_utf8_upgrade, but doesn't do magic on C<sv>
3152
8d6d96c1
HS
3153=for apidoc sv_utf8_upgrade_flags
3154
78ea37eb 3155Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3156Forces the SV to string form if it is not already.
8d6d96c1 3157Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3158if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3159will C<mg_get> on C<sv> if appropriate, else not.
3160Returns the number of bytes in the converted string
3161C<sv_utf8_upgrade> and
8d6d96c1
HS
3162C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3163
13a6c0e0
JH
3164This is not as a general purpose byte encoding to Unicode interface:
3165use the Encode extension for that.
3166
8d6d96c1 3167=cut
b3ab6785
KW
3168
3169The grow version is currently not externally documented. It adds a parameter,
3170extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3171have free after it upon return. This allows the caller to reserve extra space
3172that it intends to fill, to avoid extra grows.
3173
3174Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3175which can be used to tell this function to not first check to see if there are
3176any characters that are different in UTF-8 (variant characters) which would
3177force it to allocate a new string to sv, but to assume there are. Typically
3178this flag is used by a routine that has already parsed the string to find that
3179there are such characters, and passes this information on so that the work
3180doesn't have to be repeated.
3181
3182(One might think that the calling routine could pass in the position of the
3183first such variant, so it wouldn't have to be found again. But that is not the
3184case, because typically when the caller is likely to use this flag, it won't be
3185calling this routine unless it finds something that won't fit into a byte.
3186Otherwise it tries to not upgrade and just use bytes. But some things that
3187do fit into a byte are variants in utf8, and the caller may not have been
3188keeping track of these.)
3189
3190If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3191isn't guaranteed due to having other routines do the work in some input cases,
3192or if the input is already flagged as being in utf8.
3193
3194The speed of this could perhaps be improved for many cases if someone wanted to
3195write a fast function that counts the number of variant characters in a string,
3196especially if it could return the position of the first one.
3197
8d6d96c1
HS
3198*/
3199
3200STRLEN
b3ab6785 3201Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3202{
97aff369 3203 dVAR;
7918f24d 3204
b3ab6785 3205 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3206
808c356f
RGS
3207 if (sv == &PL_sv_undef)
3208 return 0;
e0e62c2a
NIS
3209 if (!SvPOK(sv)) {
3210 STRLEN len = 0;
d52b7888
NC
3211 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3212 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3213 if (SvUTF8(sv)) {
3214 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3215 return len;
b3ab6785 3216 }
d52b7888 3217 } else {
33fb6f35 3218 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3219 }
e0e62c2a 3220 }
4411f3b6 3221
f5cee72b 3222 if (SvUTF8(sv)) {
b3ab6785 3223 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3224 return SvCUR(sv);
f5cee72b 3225 }
5fec3b1d 3226
765f542d
NC
3227 if (SvIsCOW(sv)) {
3228 sv_force_normal_flags(sv, 0);
db42d148
NIS
3229 }
3230
b3ab6785 3231 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3232 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3233 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3234 return SvCUR(sv);
3235 }
3236
4e93345f
KW
3237 if (SvCUR(sv) == 0) {
3238 if (extra) SvGROW(sv, extra);
3239 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3240 /* This function could be much more efficient if we
2bbc8d55 3241 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3242 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3243 * make the loop as fast as possible (although there are certainly ways
3244 * to speed this up, eg. through vectorization) */
3245 U8 * s = (U8 *) SvPVX_const(sv);
3246 U8 * e = (U8 *) SvEND(sv);
3247 U8 *t = s;
3248 STRLEN two_byte_count = 0;
c4e7c712 3249
b3ab6785
KW
3250 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3251
3252 /* See if really will need to convert to utf8. We mustn't rely on our
3253 * incoming SV being well formed and having a trailing '\0', as certain
3254 * code in pp_formline can send us partially built SVs. */
3255
c4e7c712 3256 while (t < e) {
53c1dcc0 3257 const U8 ch = *t++;
b3ab6785
KW
3258 if (NATIVE_IS_INVARIANT(ch)) continue;
3259
3260 t--; /* t already incremented; re-point to first variant */
3261 two_byte_count = 1;
3262 goto must_be_utf8;
c4e7c712 3263 }
b3ab6785
KW
3264
3265 /* utf8 conversion not needed because all are invariants. Mark as
3266 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3267 SvUTF8_on(sv);
7f0bfbea 3268 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3269 return SvCUR(sv);
3270
3271must_be_utf8:
3272
3273 /* Here, the string should be converted to utf8, either because of an
3274 * input flag (two_byte_count = 0), or because a character that
3275 * requires 2 bytes was found (two_byte_count = 1). t points either to
3276 * the beginning of the string (if we didn't examine anything), or to
3277 * the first variant. In either case, everything from s to t - 1 will
3278 * occupy only 1 byte each on output.
3279 *
3280 * There are two main ways to convert. One is to create a new string
3281 * and go through the input starting from the beginning, appending each
3282 * converted value onto the new string as we go along. It's probably
3283 * best to allocate enough space in the string for the worst possible
3284 * case rather than possibly running out of space and having to
3285 * reallocate and then copy what we've done so far. Since everything
3286 * from s to t - 1 is invariant, the destination can be initialized
3287 * with these using a fast memory copy
3288 *
3289 * The other way is to figure out exactly how big the string should be
3290 * by parsing the entire input. Then you don't have to make it big
3291 * enough to handle the worst possible case, and more importantly, if
3292 * the string you already have is large enough, you don't have to
3293 * allocate a new string, you can copy the last character in the input
3294 * string to the final position(s) that will be occupied by the
3295 * converted string and go backwards, stopping at t, since everything
3296 * before that is invariant.
3297 *
3298 * There are advantages and disadvantages to each method.
3299 *
3300 * In the first method, we can allocate a new string, do the memory
3301 * copy from the s to t - 1, and then proceed through the rest of the
3302 * string byte-by-byte.
3303 *
3304 * In the second method, we proceed through the rest of the input
3305 * string just calculating how big the converted string will be. Then
3306 * there are two cases:
3307 * 1) if the string has enough extra space to handle the converted
3308 * value. We go backwards through the string, converting until we
3309 * get to the position we are at now, and then stop. If this
3310 * position is far enough along in the string, this method is
3311 * faster than the other method. If the memory copy were the same
3312 * speed as the byte-by-byte loop, that position would be about
3313 * half-way, as at the half-way mark, parsing to the end and back
3314 * is one complete string's parse, the same amount as starting
3315 * over and going all the way through. Actually, it would be
3316 * somewhat less than half-way, as it's faster to just count bytes
3317 * than to also copy, and we don't have the overhead of allocating
3318 * a new string, changing the scalar to use it, and freeing the
3319 * existing one. But if the memory copy is fast, the break-even
3320 * point is somewhere after half way. The counting loop could be
3321 * sped up by vectorization, etc, to move the break-even point
3322 * further towards the beginning.
3323 * 2) if the string doesn't have enough space to handle the converted
3324 * value. A new string will have to be allocated, and one might
3325 * as well, given that, start from the beginning doing the first
3326 * method. We've spent extra time parsing the string and in
3327 * exchange all we've gotten is that we know precisely how big to
3328 * make the new one. Perl is more optimized for time than space,
3329 * so this case is a loser.
3330 * So what I've decided to do is not use the 2nd method unless it is
3331 * guaranteed that a new string won't have to be allocated, assuming
3332 * the worst case. I also decided not to put any more conditions on it
3333 * than this, for now. It seems likely that, since the worst case is
3334 * twice as big as the unknown portion of the string (plus 1), we won't
3335 * be guaranteed enough space, causing us to go to the first method,
3336 * unless the string is short, or the first variant character is near
3337 * the end of it. In either of these cases, it seems best to use the
3338 * 2nd method. The only circumstance I can think of where this would
3339 * be really slower is if the string had once had much more data in it
3340 * than it does now, but there is still a substantial amount in it */
3341
3342 {
3343 STRLEN invariant_head = t - s;
3344 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3345 if (SvLEN(sv) < size) {
3346
3347 /* Here, have decided to allocate a new string */
3348
3349 U8 *dst;
3350 U8 *d;
3351
3352 Newx(dst, size, U8);
3353
3354 /* If no known invariants at the beginning of the input string,
3355 * set so starts from there. Otherwise, can use memory copy to
3356 * get up to where we are now, and then start from here */
3357
3358 if (invariant_head <= 0) {
3359 d = dst;
3360 } else {
3361 Copy(s, dst, invariant_head, char);
3362 d = dst + invariant_head;
3363 }
3364
3365 while (t < e) {
3366 const UV uv = NATIVE8_TO_UNI(*t++);
3367 if (UNI_IS_INVARIANT(uv))
3368 *d++ = (U8)UNI_TO_NATIVE(uv);
3369 else {
3370 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3371 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3372 }
3373 }
3374 *d = '\0';
3375 SvPV_free(sv); /* No longer using pre-existing string */
3376 SvPV_set(sv, (char*)dst);
3377 SvCUR_set(sv, d - dst);
3378 SvLEN_set(sv, size);
3379 } else {
3380
3381 /* Here, have decided to get the exact size of the string.
3382 * Currently this happens only when we know that there is
3383 * guaranteed enough space to fit the converted string, so
3384 * don't have to worry about growing. If two_byte_count is 0,
3385 * then t points to the first byte of the string which hasn't
3386 * been examined yet. Otherwise two_byte_count is 1, and t
3387 * points to the first byte in the string that will expand to
3388 * two. Depending on this, start examining at t or 1 after t.
3389 * */
3390
3391 U8 *d = t + two_byte_count;
3392
3393
3394 /* Count up the remaining bytes that expand to two */
3395
3396 while (d < e) {
3397 const U8 chr = *d++;
3398 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3399 }
3400
3401 /* The string will expand by just the number of bytes that
3402 * occupy two positions. But we are one afterwards because of
3403 * the increment just above. This is the place to put the
3404 * trailing NUL, and to set the length before we decrement */
3405
3406 d += two_byte_count;
3407 SvCUR_set(sv, d - s);
3408 *d-- = '\0';
3409
3410
3411 /* Having decremented d, it points to the position to put the
3412 * very last byte of the expanded string. Go backwards through
3413 * the string, copying and expanding as we go, stopping when we
3414 * get to the part that is invariant the rest of the way down */
3415
3416 e--;
3417 while (e >= t) {
3418 const U8 ch = NATIVE8_TO_UNI(*e--);
3419 if (UNI_IS_INVARIANT(ch)) {
3420 *d-- = UNI_TO_NATIVE(ch);
3421 } else {
3422 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3423 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3424 }
3425 }
3426 }
75da9d4c
DM
3427
3428 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3429 /* Update pos. We do it at the end rather than during
3430 * the upgrade, to avoid slowing down the common case
3431 * (upgrade without pos) */
3432 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3433 if (mg) {
3434 I32 pos = mg->mg_len;
3435 if (pos > 0 && (U32)pos > invariant_head) {
3436 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3437 STRLEN n = (U32)pos - invariant_head;
3438 while (n > 0) {
3439 if (UTF8_IS_START(*d))
3440 d++;
3441 d++;
3442 n--;
3443 }
3444 mg->mg_len = d - (U8*)SvPVX(sv);
3445 }
3446 }
3447 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3448 magic_setutf8(sv,mg); /* clear UTF8 cache */
3449 }
b3ab6785 3450 }
560a288e 3451 }
b3ab6785
KW
3452
3453 /* Mark as UTF-8 even if no variant - saves scanning loop */
3454 SvUTF8_on(sv);
4411f3b6 3455 return SvCUR(sv);
560a288e
GS
3456}
3457
c461cf8f
JH
3458/*
3459=for apidoc sv_utf8_downgrade
3460
78ea37eb 3461Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3462If the PV contains a character that cannot fit
3463in a byte, this conversion will fail;
78ea37eb 3464in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3465true, croaks.
3466
13a6c0e0
JH
3467This is not as a general purpose Unicode to byte encoding interface:
3468use the Encode extension for that.
3469
c461cf8f
JH
3470=cut
3471*/
3472
560a288e 3473bool
7bc54cea 3474Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
560a288e 3475{
97aff369 3476 dVAR;
7918f24d
NC
3477
3478 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3479
78ea37eb 3480 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3481 if (SvCUR(sv)) {
03cfe0ae 3482 U8 *s;
652088fc 3483 STRLEN len;
75da9d4c 3484 int mg_flags = SV_GMAGIC;
fa301091 3485
765f542d
NC
3486 if (SvIsCOW(sv)) {
3487 sv_force_normal_flags(sv, 0);
3488 }
75da9d4c
DM
3489 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3490 /* update pos */
3491 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3492 if (mg) {
3493 I32 pos = mg->mg_len;
3494 if (pos > 0) {
3495 sv_pos_b2u(sv, &pos);
3496 mg_flags = 0; /* sv_pos_b2u does get magic */
3497 mg->mg_len = pos;
3498 }
3499 }
3500 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3501 magic_setutf8(sv,mg); /* clear UTF8 cache */
3502
3503 }
3504 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3505
03cfe0ae 3506 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3507 if (fail_ok)
3508 return FALSE;
3509 else {
3510 if (PL_op)
3511 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3512 OP_DESC(PL_op));
fa301091
JH
3513 else
3514 Perl_croak(aTHX_ "Wide character");
3515 }
4b3603a4 3516 }
b162af07 3517 SvCUR_set(sv, len);
67e989fb 3518 }
560a288e 3519 }
ffebcc3e 3520 SvUTF8_off(sv);
560a288e
GS
3521 return TRUE;
3522}
3523
c461cf8f
JH
3524/*
3525=for apidoc sv_utf8_encode
3526
78ea37eb
ST
3527Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3528flag off so that it looks like octets again.
c461cf8f
JH
3529
3530=cut
3531*/
3532
560a288e 3533void
7bc54cea 3534Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3535{
7918f24d
NC
3536 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3537
4c94c214
NC
3538 if (SvIsCOW(sv)) {
3539 sv_force_normal_flags(sv, 0);
3540 }
3541 if (SvREADONLY(sv)) {
6ad8f254 3542 Perl_croak_no_modify(aTHX);
4c94c214 3543 }
a5f5288a 3544 (void) sv_utf8_upgrade(sv);
560a288e
GS
3545 SvUTF8_off(sv);
3546}
3547
4411f3b6
NIS
3548/*
3549=for apidoc sv_utf8_decode
3550
78ea37eb
ST
3551If the PV of the SV is an octet sequence in UTF-8
3552and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3553so that it looks like a character. If the PV contains only single-byte
694cf0d2 3554characters, the C<SvUTF8> flag stays off.
78ea37eb 3555Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3556
3557=cut
3558*/
3559
560a288e 3560bool
7bc54cea 3561Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3562{
7918f24d
NC
3563 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3564
78ea37eb 3565 if (SvPOKp(sv)) {
75da9d4c 3566 const U8 *start, *c;
93524f2b 3567 const U8 *e;
9cbac4c7 3568
645c22ef
DM
3569 /* The octets may have got themselves encoded - get them back as
3570 * bytes
3571 */
3572 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3573 return FALSE;
3574
3575 /* it is actually just a matter of turning the utf8 flag on, but
3576 * we want to make sure everything inside is valid utf8 first.
3577 */
75da9d4c 3578 c = start = (const U8 *) SvPVX_const(sv);
63cd0674 3579 if (!is_utf8_string(c, SvCUR(sv)+1))
67e989fb 3580 return FALSE;
93524f2b 3581 e = (const U8 *) SvEND(sv);
511c2ff0 3582 while (c < e) {
b64e5050 3583 const U8 ch = *c++;
c4d5f83a 3584 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3585 SvUTF8_on(sv);
3586 break;
3587 }
560a288e 3588 }
75da9d4c
DM
3589 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3590 /* adjust pos to the start of a UTF8 char sequence */
3591 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3592 if (mg) {
3593 I32 pos = mg->mg_len;
3594 if (pos > 0) {
3595 for (c = start + pos; c > start; c--) {
3596 if (UTF8_IS_START(*c))
3597 break;
3598 }
3599 mg->mg_len = c - start;
3600 }
3601 }
3602 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3603 magic_setutf8(sv,mg); /* clear UTF8 cache */
3604 }
560a288e
GS
3605 }
3606 return TRUE;
3607}
3608
954c1994
GS
3609/*
3610=for apidoc sv_setsv
3611
645c22ef
DM
3612Copies the contents of the source SV C<ssv> into the destination SV
3613C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3614function if the source SV needs to be reused. Does not handle 'set' magic.
3615Loosely speaking, it performs a copy-by-value, obliterating any previous
3616content of the destination.
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
8d6d96c1
HS
3622=for apidoc sv_setsv_flags
3623
645c22ef
DM
3624Copies the contents of the source SV C<ssv> into the destination SV
3625C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3626function if the source SV needs to be reused. Does not handle 'set' magic.
3627Loosely speaking, it performs a copy-by-value, obliterating any previous
3628content of the destination.
3629If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
5fcdf167
NC
3630C<ssv> if appropriate, else not. If the C<flags> parameter has the
3631C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3632and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3633
3634You probably want to use one of the assortment of wrappers, such as
3635C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3636C<SvSetMagicSV_nosteal>.
3637
3638This is the primary function for copying scalars, and most other
3639copy-ish functions and macros use this underneath.
8d6d96c1
HS
3640
3641=cut
3642*/
3643
5d0301b7 3644static void
7bc54cea 3645S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3646{
c8bbf675 3647 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3e6edce2 3648 HV *old_stash = NULL;
dd69841b 3649
7918f24d
NC
3650 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3651
bec4f4b4 3652 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
5d0301b7
NC
3653 const char * const name = GvNAME(sstr);
3654 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3655 {
f7877b28
NC
3656 if (dtype >= SVt_PV) {
3657 SvPV_free(dstr);
3658 SvPV_set(dstr, 0);
3659 SvLEN_set(dstr, 0);
3660 SvCUR_set(dstr, 0);
3661 }