This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
sv.c:sv_grow: accept read-only COWs
[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))
765f542d 69#endif
645c22ef
DM
70
71/* ============================================================================
72
73=head1 Allocation and deallocation of SVs.
74
d2a0f284
JC
75An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
76sv, av, hv...) contains type and reference count information, and for
77many types, a pointer to the body (struct xrv, xpv, xpviv...), which
78contains fields specific to each type. Some types store all they need
79in the head, so don't have a body.
80
486ec47a 81In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
82and bodies are allocated out of arenas, which by default are
83approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
84Sv-bodies are allocated by their sv-type, guaranteeing size
85consistency needed to allocate safely from arrays.
86
d2a0f284
JC
87For SV-heads, the first slot in each arena is reserved, and holds a
88link to the next arena, some flags, and a note of the number of slots.
89Snaked through each arena chain is a linked list of free items; when
90this becomes empty, an extra arena is allocated and divided up into N
91items which are threaded into the free list.
92
93SV-bodies are similar, but they use arena-sets by default, which
94separate the link and info from the arena itself, and reclaim the 1st
95slot in the arena. SV-bodies are further described later.
645c22ef
DM
96
97The following global variables are associated with arenas:
98
99 PL_sv_arenaroot pointer to list of SV arenas
100 PL_sv_root pointer to list of free SV structures
101
d2a0f284
JC
102 PL_body_arenas head of linked-list of body arenas
103 PL_body_roots[] array of pointers to list of free bodies of svtype
104 arrays are indexed by the svtype needed
93e68bfb 105
d2a0f284
JC
106A few special SV heads are not allocated from an arena, but are
107instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
108The size of arenas can be changed from the default by setting
109PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
110
111The SV arena serves the secondary purpose of allowing still-live SVs
112to be located and destroyed during final cleanup.
113
114At the lowest level, the macros new_SV() and del_SV() grab and free
115an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
116to return the SV to the free list with error checking.) new_SV() calls
117more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
118SVs in the free list have their SvTYPE field set to all ones.
119
ff276b08 120At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 121perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 122start of the interpreter.
645c22ef 123
645c22ef
DM
124The function visit() scans the SV arenas list, and calls a specified
125function for each SV it finds which is still live - ie which has an SvTYPE
126other than all 1's, and a non-zero SvREFCNT. visit() is used by the
127following functions (specified as [function that calls visit()] / [function
128called by visit() for each SV]):
129
130 sv_report_used() / do_report_used()
f2524eef 131 dump all remaining SVs (debugging aid)
645c22ef 132
e4487e9b 133 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 134 do_clean_named_io_objs(),do_curse()
645c22ef 135 Attempt to free all objects pointed to by RVs,
e76981f9
FC
136 try to do the same for all objects indir-
137 ectly referenced by typeglobs too, and
138 then do a final sweep, cursing any
139 objects that remain. Called once from
645c22ef
DM
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
484e6108
FC
185# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
186 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
187 } STMT_END
d7a2c63c
MHM
188# define DEBUG_SV_SERIAL(sv) \
189 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
190 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
191#else
192# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 193# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
194#endif
195
48614a46
NC
196#ifdef PERL_POISON
197# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 198# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
199/* Whilst I'd love to do this, it seems that things like to check on
200 unreferenced scalars
7e337ee0 201# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 202*/
7e337ee0
JH
203# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
204 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
205#else
206# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 207# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
208# define POSION_SV_HEAD(sv)
209#endif
210
990198f0
DM
211/* Mark an SV head as unused, and add to free list.
212 *
213 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
214 * its refcount artificially decremented during global destruction, so
215 * there may be dangling pointers to it. The last thing we want in that
216 * case is for it to be reused. */
217
053fc874
GS
218#define plant_SV(p) \
219 STMT_START { \
990198f0 220 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
221 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
222 DEBUG_SV_SERIAL(p); \
fd0854ff 223 FREE_SV_DEBUG_FILE(p); \
48614a46 224 POSION_SV_HEAD(p); \
053fc874 225 SvFLAGS(p) = SVTYPEMASK; \
990198f0 226 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 227 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
228 PL_sv_root = (p); \
229 } \
053fc874
GS
230 --PL_sv_count; \
231 } STMT_END
a0d0e21e 232
053fc874
GS
233#define uproot_SV(p) \
234 STMT_START { \
235 (p) = PL_sv_root; \
daba3364 236 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
237 ++PL_sv_count; \
238 } STMT_END
239
645c22ef 240
cac9b346
NC
241/* make some more SVs by adding another arena */
242
cac9b346
NC
243STATIC SV*
244S_more_sv(pTHX)
245{
97aff369 246 dVAR;
cac9b346 247 SV* sv;
9a87bd09
NC
248 char *chunk; /* must use New here to match call to */
249 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
250 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
251 uproot_SV(sv);
252 return sv;
253}
254
645c22ef
DM
255/* new_SV(): return a new, empty SV head */
256
eba0f806
DM
257#ifdef DEBUG_LEAKING_SCALARS
258/* provide a real function for a debugger to play with */
259STATIC SV*
d7a2c63c 260S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
261{
262 SV* sv;
263
eba0f806
DM
264 if (PL_sv_root)
265 uproot_SV(sv);
266 else
cac9b346 267 sv = S_more_sv(aTHX);
eba0f806
DM
268 SvANY(sv) = 0;
269 SvREFCNT(sv) = 1;
270 SvFLAGS(sv) = 0;
fd0854ff 271 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
272 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
273 ? PL_parser->copline
274 : PL_curcop
f24aceb1
DM
275 ? CopLINE(PL_curcop)
276 : 0
e385c3bf 277 );
fd0854ff 278 sv->sv_debug_inpad = 0;
cd676548 279 sv->sv_debug_parent = NULL;
484e6108 280 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
281
282 sv->sv_debug_serial = PL_sv_serial++;
283
284 MEM_LOG_NEW_SV(sv, file, line, func);
285 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
286 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
287
eba0f806
DM
288 return sv;
289}
d7a2c63c 290# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
291
292#else
293# define new_SV(p) \
053fc874 294 STMT_START { \
053fc874
GS
295 if (PL_sv_root) \
296 uproot_SV(p); \
297 else \
cac9b346 298 (p) = S_more_sv(aTHX); \
053fc874
GS
299 SvANY(p) = 0; \
300 SvREFCNT(p) = 1; \
301 SvFLAGS(p) = 0; \
d7a2c63c 302 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 303 } STMT_END
eba0f806 304#endif
463ee0b2 305
645c22ef
DM
306
307/* del_SV(): return an empty SV head to the free list */
308
a0d0e21e 309#ifdef DEBUGGING
4561caa4 310
053fc874
GS
311#define del_SV(p) \
312 STMT_START { \
aea4f609 313 if (DEBUG_D_TEST) \
053fc874
GS
314 del_sv(p); \
315 else \
316 plant_SV(p); \
053fc874 317 } STMT_END
a0d0e21e 318
76e3520e 319STATIC void
cea2e8a9 320S_del_sv(pTHX_ SV *p)
463ee0b2 321{
97aff369 322 dVAR;
7918f24d
NC
323
324 PERL_ARGS_ASSERT_DEL_SV;
325
aea4f609 326 if (DEBUG_D_TEST) {
4633a7c4 327 SV* sva;
a3b680e6 328 bool ok = 0;
daba3364 329 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
330 const SV * const sv = sva + 1;
331 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 332 if (p >= sv && p < svend) {
a0d0e21e 333 ok = 1;
c0ff570e
NC
334 break;
335 }
a0d0e21e
LW
336 }
337 if (!ok) {
9b387841
NC
338 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
339 "Attempt to free non-arena SV: 0x%"UVxf
340 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
341 return;
342 }
343 }
4561caa4 344 plant_SV(p);
463ee0b2 345}
a0d0e21e 346
4561caa4
CS
347#else /* ! DEBUGGING */
348
349#define del_SV(p) plant_SV(p)
350
351#endif /* DEBUGGING */
463ee0b2 352
645c22ef
DM
353
354/*
ccfc67b7
JH
355=head1 SV Manipulation Functions
356
645c22ef
DM
357=for apidoc sv_add_arena
358
359Given a chunk of memory, link it to the head of the list of arenas,
360and split it into a list of free SVs.
361
362=cut
363*/
364
d2bd4e7f
NC
365static void
366S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 367{
97aff369 368 dVAR;
daba3364 369 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
370 SV* sv;
371 SV* svend;
4633a7c4 372
7918f24d
NC
373 PERL_ARGS_ASSERT_SV_ADD_ARENA;
374
4633a7c4 375 /* The first SV in an arena isn't an SV. */
3280af22 376 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
377 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
378 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
379
3280af22
NIS
380 PL_sv_arenaroot = sva;
381 PL_sv_root = sva + 1;
4633a7c4
LW
382
383 svend = &sva[SvREFCNT(sva) - 1];
384 sv = sva + 1;
463ee0b2 385 while (sv < svend) {
3eef1deb 386 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 387#ifdef DEBUGGING
978b032e 388 SvREFCNT(sv) = 0;
03e36789 389#endif
4b69cbe3 390 /* Must always set typemask because it's always checked in on cleanup
03e36789 391 when the arenas are walked looking for objects. */
8990e307 392 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
393 sv++;
394 }
3eef1deb 395 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
396#ifdef DEBUGGING
397 SvREFCNT(sv) = 0;
398#endif
4633a7c4
LW
399 SvFLAGS(sv) = SVTYPEMASK;
400}
401
055972dc
DM
402/* visit(): call the named function for each non-free SV in the arenas
403 * whose flags field matches the flags/mask args. */
645c22ef 404
5226ed68 405STATIC I32
de37a194 406S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 407{
97aff369 408 dVAR;
4633a7c4 409 SV* sva;
5226ed68 410 I32 visited = 0;
8990e307 411
7918f24d
NC
412 PERL_ARGS_ASSERT_VISIT;
413
daba3364 414 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
415 const SV * const svend = &sva[SvREFCNT(sva)];
416 SV* sv;
4561caa4 417 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 418 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
419 && (sv->sv_flags & mask) == flags
420 && SvREFCNT(sv))
421 {
acfe0abc 422 (FCALL)(aTHX_ sv);
5226ed68
JH
423 ++visited;
424 }
8990e307
LW
425 }
426 }
5226ed68 427 return visited;
8990e307
LW
428}
429
758a08c3
JH
430#ifdef DEBUGGING
431
645c22ef
DM
432/* called by sv_report_used() for each live SV */
433
434static void
5fa45a31 435do_report_used(pTHX_ SV *const sv)
645c22ef 436{
e4787c0c 437 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
438 PerlIO_printf(Perl_debug_log, "****\n");
439 sv_dump(sv);
440 }
441}
758a08c3 442#endif
645c22ef
DM
443
444/*
445=for apidoc sv_report_used
446
fde67290 447Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
448
449=cut
450*/
451
8990e307 452void
864dbfa3 453Perl_sv_report_used(pTHX)
4561caa4 454{
ff270d3a 455#ifdef DEBUGGING
055972dc 456 visit(do_report_used, 0, 0);
96a5add6
AL
457#else
458 PERL_UNUSED_CONTEXT;
ff270d3a 459#endif
4561caa4
CS
460}
461
645c22ef
DM
462/* called by sv_clean_objs() for each live SV */
463
464static void
de37a194 465do_clean_objs(pTHX_ SV *const ref)
645c22ef 466{
97aff369 467 dVAR;
ea724faa
NC
468 assert (SvROK(ref));
469 {
823a54a3
AL
470 SV * const target = SvRV(ref);
471 if (SvOBJECT(target)) {
472 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
473 if (SvWEAKREF(ref)) {
474 sv_del_backref(target, ref);
475 SvWEAKREF_off(ref);
476 SvRV_set(ref, NULL);
477 } else {
478 SvROK_off(ref);
479 SvRV_set(ref, NULL);
fc2b2dca 480 SvREFCNT_dec_NN(target);
823a54a3 481 }
645c22ef
DM
482 }
483 }
645c22ef
DM
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;
fc2b2dca 508 SvREFCNT_dec_NN(obj);
57ef47cc
DM
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;
fc2b2dca 514 SvREFCNT_dec_NN(obj);
57ef47cc
DM
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;
fc2b2dca 520 SvREFCNT_dec_NN(obj);
57ef47cc
DM
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);
fc2b2dca 526 SvREFCNT_dec_NN(obj);
57ef47cc 527 }
fc2b2dca 528 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
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;
fc2b2dca 549 SvREFCNT_dec_NN(obj);
645c22ef 550 }
fc2b2dca 551 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 552}
645c22ef 553
4155e4fe
FC
554/* Void wrapper to pass to visit() */
555static void
556do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
557 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
558 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
559 return;
560 (void)curse(sv, 0);
561}
562
645c22ef
DM
563/*
564=for apidoc sv_clean_objs
565
fde67290 566Attempt to destroy all objects not yet freed.
645c22ef
DM
567
568=cut
569*/
570
4561caa4 571void
864dbfa3 572Perl_sv_clean_objs(pTHX)
4561caa4 573{
97aff369 574 dVAR;
68b590d9 575 GV *olddef, *olderr;
3280af22 576 PL_in_clean_objs = TRUE;
055972dc 577 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
578 /* Some barnacles may yet remain, clinging to typeglobs.
579 * Run the non-IO destructors first: they may want to output
580 * error messages, close files etc */
d011219a 581 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 582 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
583 /* And if there are some very tenacious barnacles clinging to arrays,
584 closures, or what have you.... */
585 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
586 olddef = PL_defoutgv;
587 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
588 if (olddef && isGV_with_GP(olddef))
589 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
590 olderr = PL_stderrgv;
591 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
592 if (olderr && isGV_with_GP(olderr))
593 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
594 SvREFCNT_dec(olddef);
3280af22 595 PL_in_clean_objs = FALSE;
4561caa4
CS
596}
597
645c22ef
DM
598/* called by sv_clean_all() for each live SV */
599
600static void
de37a194 601do_clean_all(pTHX_ SV *const sv)
645c22ef 602{
97aff369 603 dVAR;
daba3364 604 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 605 /* don't clean pid table and strtab */
d17ea597 606 return;
cddfcddc 607 }
645c22ef
DM
608 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
609 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 610 SvREFCNT_dec_NN(sv);
645c22ef
DM
611}
612
613/*
614=for apidoc sv_clean_all
615
616Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 617cleanup. This function may have to be called multiple times to free
ff276b08 618SVs which are in complex self-referential hierarchies.
645c22ef
DM
619
620=cut
621*/
622
5226ed68 623I32
864dbfa3 624Perl_sv_clean_all(pTHX)
8990e307 625{
97aff369 626 dVAR;
5226ed68 627 I32 cleaned;
3280af22 628 PL_in_clean_all = TRUE;
055972dc 629 cleaned = visit(do_clean_all, 0,0);
5226ed68 630 return cleaned;
8990e307 631}
463ee0b2 632
5e258f8c
JC
633/*
634 ARENASETS: a meta-arena implementation which separates arena-info
635 into struct arena_set, which contains an array of struct
636 arena_descs, each holding info for a single arena. By separating
637 the meta-info from the arena, we recover the 1st slot, formerly
638 borrowed for list management. The arena_set is about the size of an
39244528 639 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
640
641 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
642 memory in the last arena-set (1/2 on average). In trade, we get
643 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 644 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
645 small arenas for large, rare body types, by changing array* fields
646 in body_details_by_type[] below.
5e258f8c 647*/
5e258f8c 648struct arena_desc {
398c677b
NC
649 char *arena; /* the raw storage, allocated aligned */
650 size_t size; /* its size ~4k typ */
e5973ed5 651 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
652};
653
e6148039
NC
654struct arena_set;
655
656/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 657 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
658 therefore likely to be 1 aligned memory page. */
659
660#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
661 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
662
663struct arena_set {
664 struct arena_set* next;
0a848332
NC
665 unsigned int set_size; /* ie ARENAS_PER_SET */
666 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
667 struct arena_desc set[ARENAS_PER_SET];
668};
669
645c22ef
DM
670/*
671=for apidoc sv_free_arenas
672
fde67290 673Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
674heads and bodies within the arenas must already have been freed.
675
676=cut
677*/
4633a7c4 678void
864dbfa3 679Perl_sv_free_arenas(pTHX)
4633a7c4 680{
97aff369 681 dVAR;
4633a7c4
LW
682 SV* sva;
683 SV* svanext;
0a848332 684 unsigned int i;
4633a7c4
LW
685
686 /* Free arenas here, but be careful about fake ones. (We assume
687 contiguity of the fake ones with the corresponding real ones.) */
688
3280af22 689 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 690 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 691 while (svanext && SvFAKE(svanext))
daba3364 692 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
693
694 if (!SvFAKE(sva))
1df70142 695 Safefree(sva);
4633a7c4 696 }
93e68bfb 697
5e258f8c 698 {
0a848332
NC
699 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
700
701 while (aroot) {
702 struct arena_set *current = aroot;
703 i = aroot->curr;
704 while (i--) {
5e258f8c
JC
705 assert(aroot->set[i].arena);
706 Safefree(aroot->set[i].arena);
707 }
0a848332
NC
708 aroot = aroot->next;
709 Safefree(current);
5e258f8c
JC
710 }
711 }
dc8220bf 712 PL_body_arenas = 0;
fdda85ca 713
0a848332
NC
714 i = PERL_ARENA_ROOTS_SIZE;
715 while (i--)
93e68bfb 716 PL_body_roots[i] = 0;
93e68bfb 717
3280af22
NIS
718 PL_sv_arenaroot = 0;
719 PL_sv_root = 0;
4633a7c4
LW
720}
721
bd81e77b
NC
722/*
723 Here are mid-level routines that manage the allocation of bodies out
724 of the various arenas. There are 5 kinds of arenas:
29489e7c 725
bd81e77b
NC
726 1. SV-head arenas, which are discussed and handled above
727 2. regular body arenas
728 3. arenas for reduced-size bodies
729 4. Hash-Entry arenas
29489e7c 730
bd81e77b
NC
731 Arena types 2 & 3 are chained by body-type off an array of
732 arena-root pointers, which is indexed by svtype. Some of the
733 larger/less used body types are malloced singly, since a large
734 unused block of them is wasteful. Also, several svtypes dont have
735 bodies; the data fits into the sv-head itself. The arena-root
736 pointer thus has a few unused root-pointers (which may be hijacked
737 later for arena types 4,5)
29489e7c 738
bd81e77b
NC
739 3 differs from 2 as an optimization; some body types have several
740 unused fields in the front of the structure (which are kept in-place
741 for consistency). These bodies can be allocated in smaller chunks,
742 because the leading fields arent accessed. Pointers to such bodies
743 are decremented to point at the unused 'ghost' memory, knowing that
744 the pointers are used with offsets to the real memory.
29489e7c 745
d2a0f284
JC
746
747=head1 SV-Body Allocation
748
749Allocation of SV-bodies is similar to SV-heads, differing as follows;
750the allocation mechanism is used for many body types, so is somewhat
751more complicated, it uses arena-sets, and has no need for still-live
752SV detection.
753
754At the outermost level, (new|del)_X*V macros return bodies of the
755appropriate type. These macros call either (new|del)_body_type or
756(new|del)_body_allocated macro pairs, depending on specifics of the
757type. Most body types use the former pair, the latter pair is used to
758allocate body types with "ghost fields".
759
760"ghost fields" are fields that are unused in certain types, and
69ba284b 761consequently don't need to actually exist. They are declared because
d2a0f284
JC
762they're part of a "base type", which allows use of functions as
763methods. The simplest examples are AVs and HVs, 2 aggregate types
764which don't use the fields which support SCALAR semantics.
765
69ba284b 766For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
767chunks, we thus avoid wasted memory for those unaccessed members.
768When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 769size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
770structure. (But things will all go boom if you write to the part that
771is "not there", because you'll be overwriting the last members of the
772preceding structure in memory.)
773
69ba284b
NC
774We calculate the correction using the STRUCT_OFFSET macro on the first
775member present. If the allocated structure is smaller (no initial NV
776actually allocated) then the net effect is to subtract the size of the NV
777from the pointer, to return a new pointer as if an initial NV were actually
778allocated. (We were using structures named *_allocated for this, but
779this turned out to be a subtle bug, because a structure without an NV
780could have a lower alignment constraint, but the compiler is allowed to
781optimised accesses based on the alignment constraint of the actual pointer
782to the full structure, for example, using a single 64 bit load instruction
783because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284
JC
784
785This is the same trick as was used for NV and IV bodies. Ironically it
786doesn't need to be used for NV bodies any more, because NV is now at
787the start of the structure. IV bodies don't need it either, because
788they are no longer allocated.
789
790In turn, the new_body_* allocators call S_new_body(), which invokes
791new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 792linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
793necessary to refresh an empty list. Then the lock is released, and
794the body is returned.
795
99816f8d 796Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
797bodies, which it strings into a linked list. It looks up arena-size
798and body-size from the body_details table described below, thus
799supporting the multiple body-types.
800
801If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
802the (new|del)_X*V macros are mapped directly to malloc/free.
803
d2a0f284
JC
804For each sv-type, struct body_details bodies_by_type[] carries
805parameters which control these aspects of SV handling:
806
807Arena_size determines whether arenas are used for this body type, and if
808so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
809zero, forcing individual mallocs and frees.
810
811Body_size determines how big a body is, and therefore how many fit into
812each arena. Offset carries the body-pointer adjustment needed for
69ba284b 813"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
814
815But its main purpose is to parameterize info needed in
816Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 817vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
818are used for this, except for arena_size.
819
820For the sv-types that have no bodies, arenas are not used, so those
821PL_body_roots[sv_type] are unused, and can be overloaded. In
822something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 823PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 824bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 825available in hv.c.
d2a0f284 826
29489e7c
DM
827*/
828
bd81e77b 829struct body_details {
0fb58b32 830 U8 body_size; /* Size to allocate */
10666ae3 831 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 832 U8 offset;
10666ae3
NC
833 unsigned int type : 4; /* We have space for a sanity check. */
834 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
835 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
836 unsigned int arena : 1; /* Allocated from an arena */
837 size_t arena_size; /* Size of arena to allocate */
bd81e77b 838};
29489e7c 839
bd81e77b
NC
840#define HADNV FALSE
841#define NONV TRUE
29489e7c 842
d2a0f284 843
bd81e77b
NC
844#ifdef PURIFY
845/* With -DPURFIY we allocate everything directly, and don't use arenas.
846 This seems a rather elegant way to simplify some of the code below. */
847#define HASARENA FALSE
848#else
849#define HASARENA TRUE
850#endif
851#define NOARENA FALSE
29489e7c 852
d2a0f284
JC
853/* Size the arenas to exactly fit a given number of bodies. A count
854 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
855 simplifying the default. If count > 0, the arena is sized to fit
856 only that many bodies, allowing arenas to be used for large, rare
857 bodies (XPVFM, XPVIO) without undue waste. The arena size is
858 limited by PERL_ARENA_SIZE, so we can safely oversize the
859 declarations.
860 */
95db5f15
MB
861#define FIT_ARENA0(body_size) \
862 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
863#define FIT_ARENAn(count,body_size) \
864 ( count * body_size <= PERL_ARENA_SIZE) \
865 ? count * body_size \
866 : FIT_ARENA0 (body_size)
867#define FIT_ARENA(count,body_size) \
868 count \
869 ? FIT_ARENAn (count, body_size) \
870 : FIT_ARENA0 (body_size)
d2a0f284 871
bd81e77b
NC
872/* Calculate the length to copy. Specifically work out the length less any
873 final padding the compiler needed to add. See the comment in sv_upgrade
874 for why copying the padding proved to be a bug. */
29489e7c 875
bd81e77b
NC
876#define copy_length(type, last_member) \
877 STRUCT_OFFSET(type, last_member) \
daba3364 878 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 879
bd81e77b 880static const struct body_details bodies_by_type[] = {
829cd18a
NC
881 /* HEs use this offset for their arena. */
882 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 883
db93c0c4
NC
884 /* IVs are in the head, so the allocation size is 0. */
885 { 0,
d2a0f284 886 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 887 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 888 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
889 },
890
6e128786
NC
891 { sizeof(NV), sizeof(NV),
892 STRUCT_OFFSET(XPVNV, xnv_u),
893 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 894
bc337e5c 895 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
896 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
897 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 898 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 899 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 900
d361b004
KW
901 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
902 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
903 + STRUCT_OFFSET(XPV, xpv_cur),
904 SVt_INVLIST, TRUE, NONV, HASARENA,
905 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 906
bc337e5c 907 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
908 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
909 + STRUCT_OFFSET(XPV, xpv_cur),
910 SVt_PVIV, FALSE, NONV, HASARENA,
911 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 912
bc337e5c 913 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
914 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
915 + STRUCT_OFFSET(XPV, xpv_cur),
916 SVt_PVNV, FALSE, HADNV, HASARENA,
917 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 918
6e128786 919 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 920 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 921
601dfd0a
NC
922 { sizeof(regexp),
923 sizeof(regexp),
924 0,
ecff11eb 925 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 926 FIT_ARENA(0, sizeof(regexp))
5c35adbb 927 },
4df7f6af 928
10666ae3 929 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
930 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
931
10666ae3 932 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
933 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
934
601dfd0a 935 { sizeof(XPVAV),
4f7003f5 936 copy_length(XPVAV, xav_alloc),
601dfd0a 937 0,
69ba284b 938 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 939 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 940
601dfd0a 941 { sizeof(XPVHV),
359164a0 942 copy_length(XPVHV, xhv_max),
601dfd0a 943 0,
69ba284b 944 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 945 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 946
601dfd0a
NC
947 { sizeof(XPVCV),
948 sizeof(XPVCV),
949 0,
69ba284b 950 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 951 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 952
601dfd0a
NC
953 { sizeof(XPVFM),
954 sizeof(XPVFM),
955 0,
69ba284b 956 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 957 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 958
601dfd0a
NC
959 { sizeof(XPVIO),
960 sizeof(XPVIO),
961 0,
b6f60916 962 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 963 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 964};
29489e7c 965
bd81e77b 966#define new_body_allocated(sv_type) \
d2a0f284 967 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 968 - bodies_by_type[sv_type].offset)
29489e7c 969
26359cfa
NC
970/* return a thing to the free list */
971
972#define del_body(thing, root) \
973 STMT_START { \
974 void ** const thing_copy = (void **)thing; \
975 *thing_copy = *root; \
976 *root = (void*)thing_copy; \
977 } STMT_END
29489e7c 978
bd81e77b 979#ifdef PURIFY
29489e7c 980
beeec492
NC
981#define new_XNV() safemalloc(sizeof(XPVNV))
982#define new_XPVNV() safemalloc(sizeof(XPVNV))
983#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 984
beeec492 985#define del_XPVGV(p) safefree(p)
29489e7c 986
bd81e77b 987#else /* !PURIFY */
29489e7c 988
65ac1738 989#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 990#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 991#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 992
26359cfa
NC
993#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
994 &PL_body_roots[SVt_PVGV])
1d7c1841 995
bd81e77b 996#endif /* PURIFY */
93e68bfb 997
bd81e77b 998/* no arena for you! */
93e68bfb 999
bd81e77b 1000#define new_NOARENA(details) \
beeec492 1001 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1002#define new_NOARENAZ(details) \
beeec492 1003 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1004
1e30fcd5
NC
1005void *
1006Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1007 const size_t arena_size)
d2a0f284
JC
1008{
1009 dVAR;
1010 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1011 struct arena_desc *adesc;
1012 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1013 unsigned int curr;
d2a0f284
JC
1014 char *start;
1015 const char *end;
02982131 1016 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1017#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1018 static bool done_sanity_check;
1019
0b2d3faa
JH
1020 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1021 * variables like done_sanity_check. */
10666ae3 1022 if (!done_sanity_check) {
ea471437 1023 unsigned int i = SVt_LAST;
10666ae3
NC
1024
1025 done_sanity_check = TRUE;
1026
1027 while (i--)
1028 assert (bodies_by_type[i].type == i);
1029 }
1030#endif
1031
02982131 1032 assert(arena_size);
23e9d66c 1033
99816f8d
NC
1034 /* may need new arena-set to hold new arena */
1035 if (!aroot || aroot->curr >= aroot->set_size) {
1036 struct arena_set *newroot;
1037 Newxz(newroot, 1, struct arena_set);
1038 newroot->set_size = ARENAS_PER_SET;
1039 newroot->next = aroot;
1040 aroot = newroot;
1041 PL_body_arenas = (void *) newroot;
1042 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1043 }
1044
1045 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1046 curr = aroot->curr++;
1047 adesc = &(aroot->set[curr]);
1048 assert(!adesc->arena);
1049
1050 Newx(adesc->arena, good_arena_size, char);
1051 adesc->size = good_arena_size;
1052 adesc->utype = sv_type;
1053 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1054 curr, (void*)adesc->arena, (UV)good_arena_size));
1055
1056 start = (char *) adesc->arena;
d2a0f284 1057
29657bb6
NC
1058 /* Get the address of the byte after the end of the last body we can fit.
1059 Remember, this is integer division: */
02982131 1060 end = start + good_arena_size / body_size * body_size;
d2a0f284 1061
486ec47a 1062 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1063#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1064 DEBUG_m(PerlIO_printf(Perl_debug_log,
1065 "arena %p end %p arena-size %d (from %d) type %d "
1066 "size %d ct %d\n",
02982131
NC
1067 (void*)start, (void*)end, (int)good_arena_size,
1068 (int)arena_size, sv_type, (int)body_size,
1069 (int)good_arena_size / (int)body_size));
d8fca402 1070#else
d2a0f284
JC
1071 DEBUG_m(PerlIO_printf(Perl_debug_log,
1072 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1073 (void*)start, (void*)end,
02982131
NC
1074 (int)arena_size, sv_type, (int)body_size,
1075 (int)good_arena_size / (int)body_size));
d8fca402 1076#endif
d2a0f284
JC
1077 *root = (void *)start;
1078
29657bb6
NC
1079 while (1) {
1080 /* Where the next body would start: */
d2a0f284 1081 char * const next = start + body_size;
29657bb6
NC
1082
1083 if (next >= end) {
1084 /* This is the last body: */
1085 assert(next == end);
1086
1087 *(void **)start = 0;
1088 return *root;
1089 }
1090
d2a0f284
JC
1091 *(void**) start = (void *)next;
1092 start = next;
1093 }
d2a0f284
JC
1094}
1095
1096/* grab a new thing from the free list, allocating more if necessary.
1097 The inline version is used for speed in hot routines, and the
1098 function using it serves the rest (unless PURIFY).
1099*/
1100#define new_body_inline(xpv, sv_type) \
1101 STMT_START { \
1102 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1103 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1104 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1105 bodies_by_type[sv_type].body_size,\
1106 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1107 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1108 } STMT_END
1109
1110#ifndef PURIFY
1111
1112STATIC void *
de37a194 1113S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1114{
1115 dVAR;
1116 void *xpv;
1117 new_body_inline(xpv, sv_type);
1118 return xpv;
1119}
1120
1121#endif
93e68bfb 1122
238b27b3
NC
1123static const struct body_details fake_rv =
1124 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1125
bd81e77b
NC
1126/*
1127=for apidoc sv_upgrade
93e68bfb 1128
bd81e77b
NC
1129Upgrade an SV to a more complex form. Generally adds a new body type to the
1130SV, then copies across as much information as possible from the old body.
9521ca61
FC
1131It croaks if the SV is already in a more complex form than requested. You
1132generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1133before calling C<sv_upgrade>, and hence does not croak. See also
1134C<svtype>.
93e68bfb 1135
bd81e77b 1136=cut
93e68bfb 1137*/
93e68bfb 1138
bd81e77b 1139void
5aaab254 1140Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1141{
97aff369 1142 dVAR;
bd81e77b
NC
1143 void* old_body;
1144 void* new_body;
42d0e0b7 1145 const svtype old_type = SvTYPE(sv);
d2a0f284 1146 const struct body_details *new_type_details;
238b27b3 1147 const struct body_details *old_type_details
bd81e77b 1148 = bodies_by_type + old_type;
4df7f6af 1149 SV *referant = NULL;
cac9b346 1150
7918f24d
NC
1151 PERL_ARGS_ASSERT_SV_UPGRADE;
1152
1776cbe8
NC
1153 if (old_type == new_type)
1154 return;
1155
1156 /* This clause was purposefully added ahead of the early return above to
1157 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1158 inference by Nick I-S that it would fix other troublesome cases. See
1159 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1160
1161 Given that shared hash key scalars are no longer PVIV, but PV, there is
1162 no longer need to unshare so as to free up the IVX slot for its proper
1163 purpose. So it's safe to move the early return earlier. */
1164
093085a8 1165 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1166 sv_force_normal_flags(sv, 0);
1167 }
cac9b346 1168
bd81e77b 1169 old_body = SvANY(sv);
de042e1d 1170
bd81e77b
NC
1171 /* Copying structures onto other structures that have been neatly zeroed
1172 has a subtle gotcha. Consider XPVMG
cac9b346 1173
bd81e77b
NC
1174 +------+------+------+------+------+-------+-------+
1175 | NV | CUR | LEN | IV | MAGIC | STASH |
1176 +------+------+------+------+------+-------+-------+
1177 0 4 8 12 16 20 24 28
645c22ef 1178
bd81e77b
NC
1179 where NVs are aligned to 8 bytes, so that sizeof that structure is
1180 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1181
bd81e77b
NC
1182 +------+------+------+------+------+-------+-------+------+
1183 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1184 +------+------+------+------+------+-------+-------+------+
1185 0 4 8 12 16 20 24 28 32
08742458 1186
bd81e77b 1187 so what happens if you allocate memory for this structure:
30f9da9e 1188
bd81e77b
NC
1189 +------+------+------+------+------+-------+-------+------+------+...
1190 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1191 +------+------+------+------+------+-------+-------+------+------+...
1192 0 4 8 12 16 20 24 28 32 36
bfc44f79 1193
bd81e77b
NC
1194 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1195 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1196 started out as zero once, but it's quite possible that it isn't. So now,
1197 rather than a nicely zeroed GP, you have it pointing somewhere random.
1198 Bugs ensue.
bfc44f79 1199
bd81e77b
NC
1200 (In fact, GP ends up pointing at a previous GP structure, because the
1201 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1202 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1203 this happens to be moot because XPVGV has been re-ordered, with GP
1204 no longer after STASH)
30f9da9e 1205
bd81e77b
NC
1206 So we are careful and work out the size of used parts of all the
1207 structures. */
bfc44f79 1208
bd81e77b
NC
1209 switch (old_type) {
1210 case SVt_NULL:
1211 break;
1212 case SVt_IV:
4df7f6af
NC
1213 if (SvROK(sv)) {
1214 referant = SvRV(sv);
238b27b3
NC
1215 old_type_details = &fake_rv;
1216 if (new_type == SVt_NV)
1217 new_type = SVt_PVNV;
4df7f6af
NC
1218 } else {
1219 if (new_type < SVt_PVIV) {
1220 new_type = (new_type == SVt_NV)
1221 ? SVt_PVNV : SVt_PVIV;
1222 }
bd81e77b
NC
1223 }
1224 break;
1225 case SVt_NV:
1226 if (new_type < SVt_PVNV) {
1227 new_type = SVt_PVNV;
bd81e77b
NC
1228 }
1229 break;
bd81e77b
NC
1230 case SVt_PV:
1231 assert(new_type > SVt_PV);
1232 assert(SVt_IV < SVt_PV);
1233 assert(SVt_NV < SVt_PV);
1234 break;
1235 case SVt_PVIV:
1236 break;
1237 case SVt_PVNV:
1238 break;
1239 case SVt_PVMG:
1240 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1241 there's no way that it can be safely upgraded, because perl.c
1242 expects to Safefree(SvANY(PL_mess_sv)) */
1243 assert(sv != PL_mess_sv);
1244 /* This flag bit is used to mean other things in other scalar types.
1245 Given that it only has meaning inside the pad, it shouldn't be set
1246 on anything that can get upgraded. */
00b1698f 1247 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1248 break;
1249 default:
2439e033 1250 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1251 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1252 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1253 }
3376de98 1254
2439e033 1255 if (UNLIKELY(old_type > new_type))
3376de98
NC
1256 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1257 (int)old_type, (int)new_type);
1258
2fa1109b 1259 new_type_details = bodies_by_type + new_type;
645c22ef 1260
bd81e77b
NC
1261 SvFLAGS(sv) &= ~SVTYPEMASK;
1262 SvFLAGS(sv) |= new_type;
932e9ff9 1263
ab4416c0
NC
1264 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1265 the return statements above will have triggered. */
1266 assert (new_type != SVt_NULL);
bd81e77b 1267 switch (new_type) {
bd81e77b
NC
1268 case SVt_IV:
1269 assert(old_type == SVt_NULL);
1270 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1271 SvIV_set(sv, 0);
1272 return;
1273 case SVt_NV:
1274 assert(old_type == SVt_NULL);
1275 SvANY(sv) = new_XNV();
1276 SvNV_set(sv, 0);
1277 return;
bd81e77b 1278 case SVt_PVHV:
bd81e77b 1279 case SVt_PVAV:
d2a0f284 1280 assert(new_type_details->body_size);
c1ae03ae
NC
1281
1282#ifndef PURIFY
1283 assert(new_type_details->arena);
d2a0f284 1284 assert(new_type_details->arena_size);
c1ae03ae 1285 /* This points to the start of the allocated area. */
d2a0f284
JC
1286 new_body_inline(new_body, new_type);
1287 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1288 new_body = ((char *)new_body) - new_type_details->offset;
1289#else
1290 /* We always allocated the full length item with PURIFY. To do this
1291 we fake things so that arena is false for all 16 types.. */
1292 new_body = new_NOARENAZ(new_type_details);
1293#endif
1294 SvANY(sv) = new_body;
1295 if (new_type == SVt_PVAV) {
1296 AvMAX(sv) = -1;
1297 AvFILLp(sv) = -1;
1298 AvREAL_only(sv);
64484faa 1299 if (old_type_details->body_size) {
ac572bf4
NC
1300 AvALLOC(sv) = 0;
1301 } else {
1302 /* It will have been zeroed when the new body was allocated.
1303 Lets not write to it, in case it confuses a write-back
1304 cache. */
1305 }
78ac7dd9
NC
1306 } else {
1307 assert(!SvOK(sv));
1308 SvOK_off(sv);
1309#ifndef NODEFAULT_SHAREKEYS
1310 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1311#endif
586fc6a3
SM
1312 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1313 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1314 }
aeb18a1e 1315
bd81e77b
NC
1316 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1317 The target created by newSVrv also is, and it can have magic.
1318 However, it never has SvPVX set.
1319 */
4df7f6af
NC
1320 if (old_type == SVt_IV) {
1321 assert(!SvROK(sv));
1322 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1323 assert(SvPVX_const(sv) == 0);
1324 }
aeb18a1e 1325
bd81e77b 1326 if (old_type >= SVt_PVMG) {
e736a858 1327 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1328 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1329 } else {
1330 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1331 }
1332 break;
93e68bfb 1333
bd81e77b
NC
1334 case SVt_PVIV:
1335 /* XXX Is this still needed? Was it ever needed? Surely as there is
1336 no route from NV to PVIV, NOK can never be true */
1337 assert(!SvNOKp(sv));
1338 assert(!SvNOK(sv));
1339 case SVt_PVIO:
1340 case SVt_PVFM:
bd81e77b
NC
1341 case SVt_PVGV:
1342 case SVt_PVCV:
1343 case SVt_PVLV:
d361b004 1344 case SVt_INVLIST:
12c45b25 1345 case SVt_REGEXP:
bd81e77b
NC
1346 case SVt_PVMG:
1347 case SVt_PVNV:
1348 case SVt_PV:
93e68bfb 1349
d2a0f284 1350 assert(new_type_details->body_size);
bd81e77b
NC
1351 /* We always allocated the full length item with PURIFY. To do this
1352 we fake things so that arena is false for all 16 types.. */
1353 if(new_type_details->arena) {
1354 /* This points to the start of the allocated area. */
d2a0f284
JC
1355 new_body_inline(new_body, new_type);
1356 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1357 new_body = ((char *)new_body) - new_type_details->offset;
1358 } else {
1359 new_body = new_NOARENAZ(new_type_details);
1360 }
1361 SvANY(sv) = new_body;
5e2fc214 1362
bd81e77b 1363 if (old_type_details->copy) {
f9ba3d20
NC
1364 /* There is now the potential for an upgrade from something without
1365 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1366 int offset = old_type_details->offset;
1367 int length = old_type_details->copy;
1368
1369 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1370 const int difference
f9ba3d20
NC
1371 = new_type_details->offset - old_type_details->offset;
1372 offset += difference;
1373 length -= difference;
1374 }
1375 assert (length >= 0);
1376
1377 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1378 char);
bd81e77b
NC
1379 }
1380
1381#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1382 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1383 * correct 0.0 for us. Otherwise, if the old body didn't have an
1384 * NV slot, but the new one does, then we need to initialise the
1385 * freshly created NV slot with whatever the correct bit pattern is
1386 * for 0.0 */
e22a937e
NC
1387 if (old_type_details->zero_nv && !new_type_details->zero_nv
1388 && !isGV_with_GP(sv))
bd81e77b 1389 SvNV_set(sv, 0);
82048762 1390#endif
5e2fc214 1391
2439e033 1392 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1393 IO * const io = MUTABLE_IO(sv);
d963bf01 1394 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1395
1396 SvOBJECT_on(io);
1397 /* Clear the stashcache because a new IO could overrule a package
1398 name */
103f5a36 1399 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
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 }
2439e033 1405 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1406 sv->sv_u.svu_rx = (regexp *)new_body;
1407 else if (old_type < SVt_PV) {
4df7f6af
NC
1408 /* referant will be NULL unless the old type was SVt_IV emulating
1409 SVt_RV */
1410 sv->sv_u.svu_rv = referant;
1411 }
bd81e77b
NC
1412 break;
1413 default:
afd78fd5
JH
1414 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1415 (unsigned long)new_type);
bd81e77b 1416 }
73171d91 1417
db93c0c4 1418 if (old_type > SVt_IV) {
bd81e77b 1419#ifdef PURIFY
beeec492 1420 safefree(old_body);
bd81e77b 1421#else
bc786448
GG
1422 /* Note that there is an assumption that all bodies of types that
1423 can be upgraded came from arenas. Only the more complex non-
1424 upgradable types are allowed to be directly malloc()ed. */
1425 assert(old_type_details->arena);
bd81e77b
NC
1426 del_body((void*)((char*)old_body + old_type_details->offset),
1427 &PL_body_roots[old_type]);
1428#endif
1429 }
1430}
73171d91 1431
bd81e77b
NC
1432/*
1433=for apidoc sv_backoff
73171d91 1434
fde67290 1435Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1436wrapper instead.
73171d91 1437
bd81e77b 1438=cut
73171d91
NC
1439*/
1440
bd81e77b 1441int
5aaab254 1442Perl_sv_backoff(pTHX_ SV *const sv)
bd81e77b 1443{
69240efd 1444 STRLEN delta;
7a4bba22 1445 const char * const s = SvPVX_const(sv);
7918f24d
NC
1446
1447 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1448 PERL_UNUSED_CONTEXT;
7918f24d 1449
bd81e77b
NC
1450 assert(SvOOK(sv));
1451 assert(SvTYPE(sv) != SVt_PVHV);
1452 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1453
69240efd
NC
1454 SvOOK_offset(sv, delta);
1455
7a4bba22
NC
1456 SvLEN_set(sv, SvLEN(sv) + delta);
1457 SvPV_set(sv, SvPVX(sv) - delta);
1458 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1459 SvFLAGS(sv) &= ~SVf_OOK;
1460 return 0;
1461}
73171d91 1462
bd81e77b
NC
1463/*
1464=for apidoc sv_grow
73171d91 1465
bd81e77b
NC
1466Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1467upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1468Use the C<SvGROW> wrapper instead.
93e68bfb 1469
bd81e77b
NC
1470=cut
1471*/
93e68bfb 1472
e0060e30
FC
1473static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1474
bd81e77b 1475char *
5aaab254 1476Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1477{
eb578fdb 1478 char *s;
93e68bfb 1479
7918f24d
NC
1480 PERL_ARGS_ASSERT_SV_GROW;
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
db2c6cb3 1506 {
e0060e30 1507 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1508 s = SvPVX_mutable(sv);
db2c6cb3 1509 }
aeb18a1e 1510
cbcb2a16
DM
1511#ifdef PERL_NEW_COPY_ON_WRITE
1512 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1513 * to store the COW count. So in general, allocate one more byte than
1514 * asked for, to make it likely this byte is always spare: and thus
1515 * make more strings COW-able.
1516 * If the new size is a big power of two, don't bother: we assume the
1517 * caller wanted a nice 2^N sized block and will be annoyed at getting
1518 * 2^N+1 */
1519 if (newlen & 0xff)
1520 newlen++;
1521#endif
1522
bd81e77b 1523 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1524 STRLEN minlen = SvCUR(sv);
1525 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1526 if (newlen < minlen)
1527 newlen = minlen;
aedff202 1528#ifndef Perl_safesysmalloc_size
bd81e77b 1529 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1530#endif
98653f18 1531 if (SvLEN(sv) && s) {
10edeb5d 1532 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1533 }
1534 else {
10edeb5d 1535 s = (char*)safemalloc(newlen);
bd81e77b
NC
1536 if (SvPVX_const(sv) && SvCUR(sv)) {
1537 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1538 }
1539 }
1540 SvPV_set(sv, s);
ca7c1a29 1541#ifdef Perl_safesysmalloc_size
98653f18
NC
1542 /* Do this here, do it once, do it right, and then we will never get
1543 called back into sv_grow() unless there really is some growing
1544 needed. */
ca7c1a29 1545 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1546#else
bd81e77b 1547 SvLEN_set(sv, newlen);
98653f18 1548#endif
bd81e77b
NC
1549 }
1550 return s;
1551}
aeb18a1e 1552
bd81e77b
NC
1553/*
1554=for apidoc sv_setiv
932e9ff9 1555
bd81e77b
NC
1556Copies an integer into the given SV, upgrading first if necessary.
1557Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1558
bd81e77b
NC
1559=cut
1560*/
463ee0b2 1561
bd81e77b 1562void
5aaab254 1563Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1564{
97aff369 1565 dVAR;
7918f24d
NC
1566
1567 PERL_ARGS_ASSERT_SV_SETIV;
1568
bd81e77b
NC
1569 SV_CHECK_THINKFIRST_COW_DROP(sv);
1570 switch (SvTYPE(sv)) {
1571 case SVt_NULL:
bd81e77b 1572 case SVt_NV:
3376de98 1573 sv_upgrade(sv, SVt_IV);
bd81e77b 1574 break;
bd81e77b
NC
1575 case SVt_PV:
1576 sv_upgrade(sv, SVt_PVIV);
1577 break;
463ee0b2 1578
bd81e77b 1579 case SVt_PVGV:
6e592b3a
BM
1580 if (!isGV_with_GP(sv))
1581 break;
bd81e77b
NC
1582 case SVt_PVAV:
1583 case SVt_PVHV:
1584 case SVt_PVCV:
1585 case SVt_PVFM:
1586 case SVt_PVIO:
22e74366 1587 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1588 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1589 OP_DESC(PL_op));
42d0e0b7 1590 default: NOOP;
bd81e77b
NC
1591 }
1592 (void)SvIOK_only(sv); /* validate number */
1593 SvIV_set(sv, i);
1594 SvTAINT(sv);
1595}
932e9ff9 1596
bd81e77b
NC
1597/*
1598=for apidoc sv_setiv_mg
d33b2eba 1599
bd81e77b 1600Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1601
bd81e77b
NC
1602=cut
1603*/
d33b2eba 1604
bd81e77b 1605void
5aaab254 1606Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1607{
7918f24d
NC
1608 PERL_ARGS_ASSERT_SV_SETIV_MG;
1609
bd81e77b
NC
1610 sv_setiv(sv,i);
1611 SvSETMAGIC(sv);
1612}
727879eb 1613
bd81e77b
NC
1614/*
1615=for apidoc sv_setuv
d33b2eba 1616
bd81e77b
NC
1617Copies an unsigned integer into the given SV, upgrading first if necessary.
1618Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1619
bd81e77b
NC
1620=cut
1621*/
d33b2eba 1622
bd81e77b 1623void
5aaab254 1624Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1625{
7918f24d
NC
1626 PERL_ARGS_ASSERT_SV_SETUV;
1627
013abb9b
NC
1628 /* With the if statement to ensure that integers are stored as IVs whenever
1629 possible:
bd81e77b 1630 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1631
bd81e77b
NC
1632 without
1633 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1634
013abb9b
NC
1635 If you wish to remove the following if statement, so that this routine
1636 (and its callers) always return UVs, please benchmark to see what the
1637 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1638 */
1639 if (u <= (UV)IV_MAX) {
1640 sv_setiv(sv, (IV)u);
1641 return;
1642 }
1643 sv_setiv(sv, 0);
1644 SvIsUV_on(sv);
1645 SvUV_set(sv, u);
1646}
d33b2eba 1647
bd81e77b
NC
1648/*
1649=for apidoc sv_setuv_mg
727879eb 1650
bd81e77b 1651Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1652
bd81e77b
NC
1653=cut
1654*/
5e2fc214 1655
bd81e77b 1656void
5aaab254 1657Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1658{
7918f24d
NC
1659 PERL_ARGS_ASSERT_SV_SETUV_MG;
1660
bd81e77b
NC
1661 sv_setuv(sv,u);
1662 SvSETMAGIC(sv);
1663}
5e2fc214 1664
954c1994 1665/*
bd81e77b 1666=for apidoc sv_setnv
954c1994 1667
bd81e77b
NC
1668Copies a double into the given SV, upgrading first if necessary.
1669Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1670
1671=cut
1672*/
1673
63f97190 1674void
5aaab254 1675Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1676{
97aff369 1677 dVAR;
7918f24d
NC
1678
1679 PERL_ARGS_ASSERT_SV_SETNV;
1680
bd81e77b
NC
1681 SV_CHECK_THINKFIRST_COW_DROP(sv);
1682 switch (SvTYPE(sv)) {
79072805 1683 case SVt_NULL:
79072805 1684 case SVt_IV:
bd81e77b 1685 sv_upgrade(sv, SVt_NV);
79072805
LW
1686 break;
1687 case SVt_PV:
79072805 1688 case SVt_PVIV:
bd81e77b 1689 sv_upgrade(sv, SVt_PVNV);
79072805 1690 break;
bd4b1eb5 1691
bd4b1eb5 1692 case SVt_PVGV:
6e592b3a
BM
1693 if (!isGV_with_GP(sv))
1694 break;
bd81e77b
NC
1695 case SVt_PVAV:
1696 case SVt_PVHV:
79072805 1697 case SVt_PVCV:
bd81e77b
NC
1698 case SVt_PVFM:
1699 case SVt_PVIO:
22e74366 1700 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1701 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1702 OP_DESC(PL_op));
42d0e0b7 1703 default: NOOP;
2068cd4d 1704 }
bd81e77b
NC
1705 SvNV_set(sv, num);
1706 (void)SvNOK_only(sv); /* validate number */
1707 SvTAINT(sv);
79072805
LW
1708}
1709
645c22ef 1710/*
bd81e77b 1711=for apidoc sv_setnv_mg
645c22ef 1712
bd81e77b 1713Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1714
1715=cut
1716*/
1717
bd81e77b 1718void
5aaab254 1719Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1720{
7918f24d
NC
1721 PERL_ARGS_ASSERT_SV_SETNV_MG;
1722
bd81e77b
NC
1723 sv_setnv(sv,num);
1724 SvSETMAGIC(sv);
79072805
LW
1725}
1726
8140a7a8
TC
1727/* Return a cleaned-up, printable version of sv, for non-numeric, or
1728 * not incrementable warning display.
1729 * Originally part of S_not_a_number().
1730 * The return value may be != tmpbuf.
bd81e77b 1731 */
954c1994 1732
8140a7a8
TC
1733STATIC const char *
1734S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1735 const char *pv;
94463019 1736
8140a7a8 1737 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1738
94463019 1739 if (DO_UTF8(sv)) {
8140a7a8 1740 SV *dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1741 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1742 } else {
1743 char *d = tmpbuf;
8140a7a8 1744 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1745 /* each *s can expand to 4 chars + "...\0",
1746 i.e. need room for 8 chars */
ecdeb87c 1747
00b6aa41
AL
1748 const char *s = SvPVX_const(sv);
1749 const char * const end = s + SvCUR(sv);
1750 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1751 int ch = *s & 0xFF;
1752 if (ch & 128 && !isPRINT_LC(ch)) {
1753 *d++ = 'M';
1754 *d++ = '-';
1755 ch &= 127;
1756 }
1757 if (ch == '\n') {
1758 *d++ = '\\';
1759 *d++ = 'n';
1760 }
1761 else if (ch == '\r') {
1762 *d++ = '\\';
1763 *d++ = 'r';
1764 }
1765 else if (ch == '\f') {
1766 *d++ = '\\';
1767 *d++ = 'f';
1768 }
1769 else if (ch == '\\') {
1770 *d++ = '\\';
1771 *d++ = '\\';
1772 }
1773 else if (ch == '\0') {
1774 *d++ = '\\';
1775 *d++ = '0';
1776 }
1777 else if (isPRINT_LC(ch))
1778 *d++ = ch;
1779 else {
1780 *d++ = '^';
1781 *d++ = toCTRL(ch);
1782 }
1783 }
1784 if (s < end) {
1785 *d++ = '.';
1786 *d++ = '.';
1787 *d++ = '.';
1788 }
1789 *d = '\0';
1790 pv = tmpbuf;
a0d0e21e 1791 }
a0d0e21e 1792
8140a7a8
TC
1793 return pv;
1794}
1795
1796/* Print an "isn't numeric" warning, using a cleaned-up,
1797 * printable version of the offending string
1798 */
1799
1800STATIC void
1801S_not_a_number(pTHX_ SV *const sv)
1802{
1803 dVAR;
1804 char tmpbuf[64];
1805 const char *pv;
1806
1807 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1808
1809 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1810
533c011a 1811 if (PL_op)
9014280d 1812 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1813 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1814 "Argument \"%s\" isn't numeric in %s", pv,
1815 OP_DESC(PL_op));
a0d0e21e 1816 else
9014280d 1817 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1818 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1819 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1820}
1821
8140a7a8
TC
1822STATIC void
1823S_not_incrementable(pTHX_ SV *const sv) {
1824 dVAR;
1825 char tmpbuf[64];
1826 const char *pv;
1827
1828 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1829
1830 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1831
1832 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1833 "Argument \"%s\" treated as 0 in increment (++)", pv);
1834}
1835
c2988b20
NC
1836/*
1837=for apidoc looks_like_number
1838
645c22ef
DM
1839Test if the content of an SV looks like a number (or is a number).
1840C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1841non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1842ignored.
c2988b20
NC
1843
1844=cut
1845*/
1846
1847I32
aad570aa 1848Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1849{
eb578fdb 1850 const char *sbegin;
c2988b20
NC
1851 STRLEN len;
1852
7918f24d
NC
1853 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1854
f52e41ad
FC
1855 if (SvPOK(sv) || SvPOKp(sv)) {
1856 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1857 }
c2988b20 1858 else
e0ab1c0e 1859 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1860 return grok_number(sbegin, len, NULL);
1861}
25da4f38 1862
19f6321d
NC
1863STATIC bool
1864S_glob_2number(pTHX_ GV * const gv)
180488f8 1865{
7918f24d
NC
1866 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1867
675c862f
AL
1868 /* We know that all GVs stringify to something that is not-a-number,
1869 so no need to test that. */
1870 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1871 {
1872 SV *const buffer = sv_newmortal();
1873 gv_efullname3(buffer, gv, "*");
675c862f 1874 not_a_number(buffer);
8e629ff4 1875 }
675c862f
AL
1876 /* We just want something true to return, so that S_sv_2iuv_common
1877 can tail call us and return true. */
19f6321d 1878 return TRUE;
675c862f
AL
1879}
1880
25da4f38
IZ
1881/* Actually, ISO C leaves conversion of UV to IV undefined, but
1882 until proven guilty, assume that things are not that bad... */
1883
645c22ef
DM
1884/*
1885 NV_PRESERVES_UV:
1886
1887 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1888 an IV (an assumption perl has been based on to date) it becomes necessary
1889 to remove the assumption that the NV always carries enough precision to
1890 recreate the IV whenever needed, and that the NV is the canonical form.
1891 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1892 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1893 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1894 1) to distinguish between IV/UV/NV slots that have cached a valid
1895 conversion where precision was lost and IV/UV/NV slots that have a
1896 valid conversion which has lost no precision
645c22ef 1897 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1898 would lose precision, the precise conversion (or differently
1899 imprecise conversion) is also performed and cached, to prevent
1900 requests for different numeric formats on the same SV causing
1901 lossy conversion chains. (lossless conversion chains are perfectly
1902 acceptable (still))
1903
1904
1905 flags are used:
1906 SvIOKp is true if the IV slot contains a valid value
1907 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1908 SvNOKp is true if the NV slot contains a valid value
1909 SvNOK is true only if the NV value is accurate
1910
1911 so
645c22ef 1912 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1913 IV(or UV) would lose accuracy over a direct conversion from PV to
1914 IV(or UV). If it would, cache both conversions, return NV, but mark
1915 SV as IOK NOKp (ie not NOK).
1916
645c22ef 1917 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1918 NV would lose accuracy over a direct conversion from PV to NV. If it
1919 would, cache both conversions, flag similarly.
1920
1921 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1922 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1923 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1924 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1925 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1926
645c22ef
DM
1927 The benefit of this is that operations such as pp_add know that if
1928 SvIOK is true for both left and right operands, then integer addition
1929 can be used instead of floating point (for cases where the result won't
1930 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1931 loss of precision compared with integer addition.
1932
1933 * making IV and NV equal status should make maths accurate on 64 bit
1934 platforms
1935 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1936 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1937 looking for SvIOK and checking for overflow will not outweigh the
1938 fp to integer speedup)
1939 * will slow down integer operations (callers of SvIV) on "inaccurate"
1940 values, as the change from SvIOK to SvIOKp will cause a call into
1941 sv_2iv each time rather than a macro access direct to the IV slot
1942 * should speed up number->string conversion on integers as IV is
645c22ef 1943 favoured when IV and NV are equally accurate
28e5dec8
JH
1944
1945 ####################################################################
645c22ef
DM
1946 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1947 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1948 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1949 ####################################################################
1950
645c22ef 1951 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1952 performance ratio.
1953*/
1954
1955#ifndef NV_PRESERVES_UV
645c22ef
DM
1956# define IS_NUMBER_UNDERFLOW_IV 1
1957# define IS_NUMBER_UNDERFLOW_UV 2
1958# define IS_NUMBER_IV_AND_UV 2
1959# define IS_NUMBER_OVERFLOW_IV 4
1960# define IS_NUMBER_OVERFLOW_UV 5
1961
1962/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1963
1964/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1965STATIC int
5aaab254 1966S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
1967# ifdef DEBUGGING
1968 , I32 numtype
1969# endif
1970 )
28e5dec8 1971{
97aff369 1972 dVAR;
7918f24d
NC
1973
1974 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1975
3f7c398e 1976 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
1977 if (SvNVX(sv) < (NV)IV_MIN) {
1978 (void)SvIOKp_on(sv);
1979 (void)SvNOK_on(sv);
45977657 1980 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1981 return IS_NUMBER_UNDERFLOW_IV;
1982 }
1983 if (SvNVX(sv) > (NV)UV_MAX) {
1984 (void)SvIOKp_on(sv);
1985 (void)SvNOK_on(sv);
1986 SvIsUV_on(sv);
607fa7f2 1987 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1988 return IS_NUMBER_OVERFLOW_UV;
1989 }
c2988b20
NC
1990 (void)SvIOKp_on(sv);
1991 (void)SvNOK_on(sv);
1992 /* Can't use strtol etc to convert this string. (See truth table in
1993 sv_2iv */
1994 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1995 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1996 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1997 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1998 } else {
1999 /* Integer is imprecise. NOK, IOKp */
2000 }
2001 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2002 }
2003 SvIsUV_on(sv);
607fa7f2 2004 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
2005 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2006 if (SvUVX(sv) == UV_MAX) {
2007 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2008 possibly be preserved by NV. Hence, it must be overflow.
2009 NOK, IOKp */
2010 return IS_NUMBER_OVERFLOW_UV;
2011 }
2012 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2013 } else {
2014 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2015 }
c2988b20 2016 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2017}
645c22ef
DM
2018#endif /* !NV_PRESERVES_UV*/
2019
af359546 2020STATIC bool
7918f24d
NC
2021S_sv_2iuv_common(pTHX_ SV *const sv)
2022{
97aff369 2023 dVAR;
7918f24d
NC
2024
2025 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2026
af359546 2027 if (SvNOKp(sv)) {
28e5dec8
JH
2028 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2029 * without also getting a cached IV/UV from it at the same time
2030 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2031 * IV or UV at same time to avoid this. */
2032 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2033
2034 if (SvTYPE(sv) == SVt_NV)
2035 sv_upgrade(sv, SVt_PVNV);
2036
28e5dec8
JH
2037 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2038 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2039 certainly cast into the IV range at IV_MAX, whereas the correct
2040 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2041 cases go to UV */
cab190d4
JD
2042#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2043 if (Perl_isnan(SvNVX(sv))) {
2044 SvUV_set(sv, 0);
2045 SvIsUV_on(sv);
fdbe6d7c 2046 return FALSE;
cab190d4 2047 }
cab190d4 2048#endif
28e5dec8 2049 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2050 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2051 if (SvNVX(sv) == (NV) SvIVX(sv)
2052#ifndef NV_PRESERVES_UV
2053 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2054 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2055 /* Don't flag it as "accurately an integer" if the number
2056 came from a (by definition imprecise) NV operation, and
2057 we're outside the range of NV integer precision */
2058#endif
2059 ) {
a43d94f2
NC
2060 if (SvNOK(sv))
2061 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2062 else {
2063 /* scalar has trailing garbage, eg "42a" */
2064 }
28e5dec8 2065 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2066 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2067 PTR2UV(sv),
2068 SvNVX(sv),
2069 SvIVX(sv)));
2070
2071 } else {
2072 /* IV not precise. No need to convert from PV, as NV
2073 conversion would already have cached IV if it detected
2074 that PV->IV would be better than PV->NV->IV
2075 flags already correct - don't set public IOK. */
2076 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2077 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2078 PTR2UV(sv),
2079 SvNVX(sv),
2080 SvIVX(sv)));
2081 }
2082 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2083 but the cast (NV)IV_MIN rounds to a the value less (more
2084 negative) than IV_MIN which happens to be equal to SvNVX ??
2085 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2086 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2087 (NV)UVX == NVX are both true, but the values differ. :-(
2088 Hopefully for 2s complement IV_MIN is something like
2089 0x8000000000000000 which will be exact. NWC */
d460ef45 2090 }
25da4f38 2091 else {
607fa7f2 2092 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2093 if (
2094 (SvNVX(sv) == (NV) SvUVX(sv))
2095#ifndef NV_PRESERVES_UV
2096 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2097 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2098 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2099 /* Don't flag it as "accurately an integer" if the number
2100 came from a (by definition imprecise) NV operation, and
2101 we're outside the range of NV integer precision */
2102#endif
a43d94f2 2103 && SvNOK(sv)
28e5dec8
JH
2104 )
2105 SvIOK_on(sv);
25da4f38 2106 SvIsUV_on(sv);
1c846c1f 2107 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2108 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2109 PTR2UV(sv),
57def98f
JH
2110 SvUVX(sv),
2111 SvUVX(sv)));
25da4f38 2112 }
748a9306 2113 }
cd84013a 2114 else if (SvPOKp(sv)) {
c2988b20 2115 UV value;
504618e9 2116 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2117 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2118 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2119 the same as the direct translation of the initial string
2120 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2121 be careful to ensure that the value with the .456 is around if the
2122 NV value is requested in the future).
1c846c1f 2123
af359546 2124 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2125 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2126 cache the NV if we are sure it's not needed.
25da4f38 2127 */
16b7a9a4 2128
c2988b20
NC
2129 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2130 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2131 == IS_NUMBER_IN_UV) {
5e045b90 2132 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2133 if (SvTYPE(sv) < SVt_PVIV)
2134 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2135 (void)SvIOK_on(sv);
c2988b20
NC
2136 } else if (SvTYPE(sv) < SVt_PVNV)
2137 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2138
f2524eef 2139 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2140 we aren't going to call atof() below. If NVs don't preserve UVs
2141 then the value returned may have more precision than atof() will
2142 return, even though value isn't perfectly accurate. */
2143 if ((numtype & (IS_NUMBER_IN_UV
2144#ifdef NV_PRESERVES_UV
2145 | IS_NUMBER_NOT_INT
2146#endif
2147 )) == IS_NUMBER_IN_UV) {
2148 /* This won't turn off the public IOK flag if it was set above */
2149 (void)SvIOKp_on(sv);
2150
2151 if (!(numtype & IS_NUMBER_NEG)) {
2152 /* positive */;
2153 if (value <= (UV)IV_MAX) {
45977657 2154 SvIV_set(sv, (IV)value);
c2988b20 2155 } else {
af359546 2156 /* it didn't overflow, and it was positive. */
607fa7f2 2157 SvUV_set(sv, value);
c2988b20
NC
2158 SvIsUV_on(sv);
2159 }
2160 } else {
2161 /* 2s complement assumption */
2162 if (value <= (UV)IV_MIN) {
45977657 2163 SvIV_set(sv, -(IV)value);
c2988b20
NC
2164 } else {
2165 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2166 I'm assuming it will be rare. */
c2988b20
NC
2167 if (SvTYPE(sv) < SVt_PVNV)
2168 sv_upgrade(sv, SVt_PVNV);
2169 SvNOK_on(sv);
2170 SvIOK_off(sv);
2171 SvIOKp_on(sv);
9d6ce603 2172 SvNV_set(sv, -(NV)value);
45977657 2173 SvIV_set(sv, IV_MIN);
c2988b20
NC
2174 }
2175 }
2176 }
2177 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2178 will be in the previous block to set the IV slot, and the next
2179 block to set the NV slot. So no else here. */
2180
2181 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2182 != IS_NUMBER_IN_UV) {
2183 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2184 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2185
c2988b20
NC
2186 if (! numtype && ckWARN(WARN_NUMERIC))
2187 not_a_number(sv);
28e5dec8 2188
65202027 2189#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2190 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2191 PTR2UV(sv), SvNVX(sv)));
65202027 2192#else
1779d84d 2193 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2194 PTR2UV(sv), SvNVX(sv)));
65202027 2195#endif
28e5dec8 2196
28e5dec8 2197#ifdef NV_PRESERVES_UV
af359546
NC
2198 (void)SvIOKp_on(sv);
2199 (void)SvNOK_on(sv);
2200 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2201 SvIV_set(sv, I_V(SvNVX(sv)));
2202 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2203 SvIOK_on(sv);
2204 } else {
6f207bd3 2205 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2206 }
2207 /* UV will not work better than IV */
2208 } else {
2209 if (SvNVX(sv) > (NV)UV_MAX) {
2210 SvIsUV_on(sv);
2211 /* Integer is inaccurate. NOK, IOKp, is UV */
2212 SvUV_set(sv, UV_MAX);
af359546
NC
2213 } else {
2214 SvUV_set(sv, U_V(SvNVX(sv)));
2215 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2216 NV preservse UV so can do correct comparison. */
2217 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2218 SvIOK_on(sv);
af359546 2219 } else {
6f207bd3 2220 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2221 }
2222 }
4b0c9573 2223 SvIsUV_on(sv);
af359546 2224 }
28e5dec8 2225#else /* NV_PRESERVES_UV */
c2988b20
NC
2226 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2227 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2228 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2229 grok_number above. The NV slot has just been set using
2230 Atof. */
560b0c46 2231 SvNOK_on(sv);
c2988b20
NC
2232 assert (SvIOKp(sv));
2233 } else {
2234 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2235 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2236 /* Small enough to preserve all bits. */
2237 (void)SvIOKp_on(sv);
2238 SvNOK_on(sv);
45977657 2239 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2240 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2241 SvIOK_on(sv);
2242 /* Assumption: first non-preserved integer is < IV_MAX,
2243 this NV is in the preserved range, therefore: */
2244 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2245 < (UV)IV_MAX)) {
32fdb065 2246 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
2247 }
2248 } else {
2249 /* IN_UV NOT_INT
2250 0 0 already failed to read UV.
2251 0 1 already failed to read UV.
2252 1 0 you won't get here in this case. IV/UV
2253 slot set, public IOK, Atof() unneeded.
2254 1 1 already read UV.
2255 so there's no point in sv_2iuv_non_preserve() attempting
2256 to use atol, strtol, strtoul etc. */
47031da6 2257# ifdef DEBUGGING
40a17c4c 2258 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2259# else
2260 sv_2iuv_non_preserve (sv);
2261# endif
c2988b20
NC
2262 }
2263 }
28e5dec8 2264#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2265 /* It might be more code efficient to go through the entire logic above
2266 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2267 gets complex and potentially buggy, so more programmer efficient
2268 to do it this way, by turning off the public flags: */
2269 if (!numtype)
2270 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2271 }
af359546
NC
2272 }
2273 else {
675c862f 2274 if (isGV_with_GP(sv))
159b6efe 2275 return glob_2number(MUTABLE_GV(sv));
180488f8 2276
9a214eec 2277 if (!SvPADTMP(sv)) {
af359546
NC
2278 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2279 report_uninit(sv);
2280 }
25da4f38
IZ
2281 if (SvTYPE(sv) < SVt_IV)
2282 /* Typically the caller expects that sv_any is not NULL now. */
2283 sv_upgrade(sv, SVt_IV);
af359546
NC
2284 /* Return 0 from the caller. */
2285 return TRUE;
2286 }
2287 return FALSE;
2288}
2289
2290/*
2291=for apidoc sv_2iv_flags
2292
2293Return the integer value of an SV, doing any necessary string
2294conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2295Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2296
2297=cut
2298*/
2299
2300IV
5aaab254 2301Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2302{
97aff369 2303 dVAR;
4bac9ae4 2304
af359546 2305 if (!sv)
a0d0e21e 2306 return 0;
4bac9ae4 2307
217f6fa3
FC
2308 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2309 && SvTYPE(sv) != SVt_PVFM);
2310
4bac9ae4
CS
2311 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2312 mg_get(sv);
2313
2314 if (SvROK(sv)) {
2315 if (SvAMAGIC(sv)) {
2316 SV * tmpstr;
2317 if (flags & SV_SKIP_OVERLOAD)
2318 return 0;
2319 tmpstr = AMG_CALLunary(sv, numer_amg);
2320 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2321 return SvIV(tmpstr);
2322 }
2323 }
2324 return PTR2IV(SvRV(sv));
2325 }
2326
8d919b0a 2327 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2328 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2329 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2330 In practice they are extremely unlikely to actually get anywhere
2331 accessible by user Perl code - the only way that I'm aware of is when
2332 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2333
2334 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2335 */
8d919b0a 2336 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2337 {
71c558c3 2338 UV value;
8d919b0a
FC
2339 const char * const ptr =
2340 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2341 const int numtype
8d919b0a 2342 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2343
2344 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2345 == IS_NUMBER_IN_UV) {
2346 /* It's definitely an integer */
2347 if (numtype & IS_NUMBER_NEG) {
2348 if (value < (UV)IV_MIN)
2349 return -(IV)value;
2350 } else {
2351 if (value < (UV)IV_MAX)
2352 return (IV)value;
2353 }
2354 }
2355 if (!numtype) {
2356 if (ckWARN(WARN_NUMERIC))
2357 not_a_number(sv);
2358 }
8d919b0a 2359 return I_V(Atof(ptr));
e20b6c3b 2360 }
4bac9ae4
CS
2361 }
2362
2363 if (SvTHINKFIRST(sv)) {
73381561 2364#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2365 if (SvIsCOW(sv)) {
2366 sv_force_normal_flags(sv, 0);
2367 }
73381561 2368#endif
af359546
NC
2369 if (SvREADONLY(sv) && !SvOK(sv)) {
2370 if (ckWARN(WARN_UNINITIALIZED))
2371 report_uninit(sv);
2372 return 0;
2373 }
2374 }
4bac9ae4 2375
af359546
NC
2376 if (!SvIOKp(sv)) {
2377 if (S_sv_2iuv_common(aTHX_ sv))
2378 return 0;
79072805 2379 }
4bac9ae4 2380
1d7c1841
GS
2381 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2382 PTR2UV(sv),SvIVX(sv)));
25da4f38 2383 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2384}
2385
645c22ef 2386/*
891f9566 2387=for apidoc sv_2uv_flags
645c22ef
DM
2388
2389Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2390conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2391Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2392
2393=cut
2394*/
2395
ff68c719 2396UV
5aaab254 2397Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2398{
97aff369 2399 dVAR;
4bac9ae4 2400
ff68c719
PP
2401 if (!sv)
2402 return 0;
4bac9ae4
CS
2403
2404 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2405 mg_get(sv);
2406
2407 if (SvROK(sv)) {
2408 if (SvAMAGIC(sv)) {
2409 SV *tmpstr;
2410 if (flags & SV_SKIP_OVERLOAD)
2411 return 0;
2412 tmpstr = AMG_CALLunary(sv, numer_amg);
2413 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2414 return SvUV(tmpstr);
2415 }
2416 }
2417 return PTR2UV(SvRV(sv));
2418 }
2419
8d919b0a 2420 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2421 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2422 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2423 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2424 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2425 {
71c558c3 2426 UV value;
8d919b0a
FC
2427 const char * const ptr =
2428 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2429 const int numtype
8d919b0a 2430 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2431
2432 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2433 == IS_NUMBER_IN_UV) {
2434 /* It's definitely an integer */
2435 if (!(numtype & IS_NUMBER_NEG))
2436 return value;
2437 }
2438 if (!numtype) {
2439 if (ckWARN(WARN_NUMERIC))
2440 not_a_number(sv);
2441 }
8d919b0a 2442 return U_V(Atof(ptr));
e20b6c3b 2443 }
4bac9ae4
CS
2444 }
2445
2446 if (SvTHINKFIRST(sv)) {
73381561 2447#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2448 if (SvIsCOW(sv)) {
2449 sv_force_normal_flags(sv, 0);
8a818333 2450 }
73381561 2451#endif
0336b60e 2452 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2453 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2454 report_uninit(sv);
ff68c719
PP
2455 return 0;
2456 }
2457 }
4bac9ae4 2458
af359546
NC
2459 if (!SvIOKp(sv)) {
2460 if (S_sv_2iuv_common(aTHX_ sv))
2461 return 0;
ff68c719 2462 }
25da4f38 2463
1d7c1841
GS
2464 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2465 PTR2UV(sv),SvUVX(sv)));
25da4f38 2466 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2467}
2468
645c22ef 2469/*
196007d1 2470=for apidoc sv_2nv_flags
645c22ef
DM
2471
2472Return the num value of an SV, doing any necessary string or integer
fde67290 2473conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2474Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2475
2476=cut
2477*/
2478
65202027 2479NV
5aaab254 2480Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2481{
97aff369 2482 dVAR;
79072805
LW
2483 if (!sv)
2484 return 0.0;
217f6fa3
FC
2485 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2486 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2487 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2488 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2489 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2490 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2491 const char *ptr;
39d5de13
DM
2492 if (flags & SV_GMAGIC)
2493 mg_get(sv);
463ee0b2
LW
2494 if (SvNOKp(sv))
2495 return SvNVX(sv);
cd84013a 2496 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2497 ptr = SvPVX_const(sv);
2498 grokpv:
041457d9 2499 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2500 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2501 not_a_number(sv);
8d919b0a 2502 return Atof(ptr);
a0d0e21e 2503 }
25da4f38 2504 if (SvIOKp(sv)) {
1c846c1f 2505 if (SvIsUV(sv))
65202027 2506 return (NV)SvUVX(sv);
25da4f38 2507 else
65202027 2508 return (NV)SvIVX(sv);
47a72cb8
NC
2509 }
2510 if (SvROK(sv)) {
2511 goto return_rok;
2512 }
8d919b0a
FC
2513 if (isREGEXP(sv)) {
2514 ptr = RX_WRAPPED((REGEXP *)sv);
2515 goto grokpv;
2516 }
47a72cb8
NC
2517 assert(SvTYPE(sv) >= SVt_PVMG);
2518 /* This falls through to the report_uninit near the end of the
2519 function. */
2520 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2521 if (SvROK(sv)) {
47a72cb8 2522 return_rok:
deb46114 2523 if (SvAMAGIC(sv)) {
aee036bb
DM
2524 SV *tmpstr;
2525 if (flags & SV_SKIP_OVERLOAD)
2526 return 0;
31d632c3 2527 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2528 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2529 return SvNV(tmpstr);
2530 }
2531 }
2532 return PTR2NV(SvRV(sv));
a0d0e21e 2533 }
73381561 2534#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2535 if (SvIsCOW(sv)) {
2536 sv_force_normal_flags(sv, 0);
8a818333 2537 }
73381561 2538#endif
0336b60e 2539 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2540 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2541 report_uninit(sv);
ed6116ce
LW
2542 return 0.0;
2543 }
79072805
LW
2544 }
2545 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2546 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2547 sv_upgrade(sv, SVt_NV);
906f284f 2548#ifdef USE_LONG_DOUBLE
097ee67d 2549 DEBUG_c({
f93f4e46 2550 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2551 PerlIO_printf(Perl_debug_log,
2552 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2553 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2554 RESTORE_NUMERIC_LOCAL();
2555 });
65202027 2556#else
572bbb43 2557 DEBUG_c({
f93f4e46 2558 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2559 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2560 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2561 RESTORE_NUMERIC_LOCAL();
2562 });
572bbb43 2563#endif
79072805
LW
2564 }
2565 else if (SvTYPE(sv) < SVt_PVNV)
2566 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2567 if (SvNOKp(sv)) {
2568 return SvNVX(sv);
61604483 2569 }
59d8ce62 2570 if (SvIOKp(sv)) {
9d6ce603 2571 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2572#ifdef NV_PRESERVES_UV
a43d94f2
NC
2573 if (SvIOK(sv))
2574 SvNOK_on(sv);
2575 else
2576 SvNOKp_on(sv);
28e5dec8
JH
2577#else
2578 /* Only set the public NV OK flag if this NV preserves the IV */
2579 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2580 if (SvIOK(sv) &&
2581 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2582 : (SvIVX(sv) == I_V(SvNVX(sv))))
2583 SvNOK_on(sv);
2584 else
2585 SvNOKp_on(sv);
2586#endif
93a17b20 2587 }
cd84013a 2588 else if (SvPOKp(sv)) {
c2988b20 2589 UV value;
3f7c398e 2590 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2591 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2592 not_a_number(sv);
28e5dec8 2593#ifdef NV_PRESERVES_UV
c2988b20
NC
2594 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2595 == IS_NUMBER_IN_UV) {
5e045b90 2596 /* It's definitely an integer */
9d6ce603 2597 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2598 } else
3f7c398e 2599 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2600 if (numtype)
2601 SvNOK_on(sv);
2602 else
2603 SvNOKp_on(sv);
28e5dec8 2604#else
3f7c398e 2605 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2606 /* Only set the public NV OK flag if this NV preserves the value in
2607 the PV at least as well as an IV/UV would.
2608 Not sure how to do this 100% reliably. */
2609 /* if that shift count is out of range then Configure's test is
2610 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2611 UV_BITS */
2612 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2613 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2614 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2615 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2616 /* Can't use strtol etc to convert this string, so don't try.
2617 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2618 SvNOK_on(sv);
2619 } else {
2620 /* value has been set. It may not be precise. */
2621 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2622 /* 2s complement assumption for (UV)IV_MIN */
2623 SvNOK_on(sv); /* Integer is too negative. */
2624 } else {
2625 SvNOKp_on(sv);
2626 SvIOKp_on(sv);
6fa402ec 2627
c2988b20 2628 if (numtype & IS_NUMBER_NEG) {
45977657 2629 SvIV_set(sv, -(IV)value);
c2988b20 2630 } else if (value <= (UV)IV_MAX) {
45977657 2631 SvIV_set(sv, (IV)value);
c2988b20 2632 } else {
607fa7f2 2633 SvUV_set(sv, value);
c2988b20
NC
2634 SvIsUV_on(sv);
2635 }
2636
2637 if (numtype & IS_NUMBER_NOT_INT) {
2638 /* I believe that even if the original PV had decimals,
2639 they are lost beyond the limit of the FP precision.
2640 However, neither is canonical, so both only get p
2641 flags. NWC, 2000/11/25 */
2642 /* Both already have p flags, so do nothing */
2643 } else {
66a1b24b 2644 const NV nv = SvNVX(sv);
c2988b20
NC
2645 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2646 if (SvIVX(sv) == I_V(nv)) {
2647 SvNOK_on(sv);
c2988b20 2648 } else {
c2988b20
NC
2649 /* It had no "." so it must be integer. */
2650 }
00b6aa41 2651 SvIOK_on(sv);
c2988b20
NC
2652 } else {
2653 /* between IV_MAX and NV(UV_MAX).
2654 Could be slightly > UV_MAX */
6fa402ec 2655
c2988b20
NC
2656 if (numtype & IS_NUMBER_NOT_INT) {
2657 /* UV and NV both imprecise. */
2658 } else {
66a1b24b 2659 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2660
2661 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2662 SvNOK_on(sv);
c2988b20 2663 }
00b6aa41 2664 SvIOK_on(sv);
c2988b20
NC
2665 }
2666 }
2667 }
2668 }
2669 }
a43d94f2
NC
2670 /* It might be more code efficient to go through the entire logic above
2671 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2672 gets complex and potentially buggy, so more programmer efficient
2673 to do it this way, by turning off the public flags: */
2674 if (!numtype)
2675 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2676#endif /* NV_PRESERVES_UV */
93a17b20 2677 }
79072805 2678 else {
f7877b28 2679 if (isGV_with_GP(sv)) {
159b6efe 2680 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2681 return 0.0;
2682 }
2683
9a214eec 2684 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2685 report_uninit(sv);
7e25a7e9
NC
2686 assert (SvTYPE(sv) >= SVt_NV);
2687 /* Typically the caller expects that sv_any is not NULL now. */
2688 /* XXX Ilya implies that this is a bug in callers that assume this
2689 and ideally should be fixed. */
a0d0e21e 2690 return 0.0;
79072805 2691 }
572bbb43 2692#if defined(USE_LONG_DOUBLE)
097ee67d 2693 DEBUG_c({
f93f4e46 2694 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2695 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2696 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2697 RESTORE_NUMERIC_LOCAL();
2698 });
65202027 2699#else
572bbb43 2700 DEBUG_c({
f93f4e46 2701 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2702 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2703 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2704 RESTORE_NUMERIC_LOCAL();
2705 });
572bbb43 2706#endif
463ee0b2 2707 return SvNVX(sv);
79072805
LW
2708}
2709
800401ee
JH
2710/*
2711=for apidoc sv_2num
2712
2713Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2714reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2715access this function.
800401ee
JH
2716
2717=cut
2718*/
2719
2720SV *
5aaab254 2721Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2722{
7918f24d
NC
2723 PERL_ARGS_ASSERT_SV_2NUM;
2724
b9ee0594
RGS
2725 if (!SvROK(sv))
2726 return sv;
800401ee 2727 if (SvAMAGIC(sv)) {
31d632c3 2728 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2729 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2730 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2731 return sv_2num(tmpsv);
2732 }
2733 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2734}
2735
645c22ef
DM
2736/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2737 * UV as a string towards the end of buf, and return pointers to start and
2738 * end of it.
2739 *
2740 * We assume that buf is at least TYPE_CHARS(UV) long.
2741 */
2742
864dbfa3 2743static char *
5de3775c 2744S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2745{
25da4f38 2746 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2747 char * const ebuf = ptr;
25da4f38 2748 int sign;
25da4f38 2749
7918f24d
NC
2750 PERL_ARGS_ASSERT_UIV_2BUF;
2751
25da4f38
IZ
2752 if (is_uv)
2753 sign = 0;
2754 else if (iv >= 0) {
2755 uv = iv;
2756 sign = 0;
2757 } else {
2758 uv = -iv;
2759 sign = 1;
2760 }
2761 do {
eb160463 2762 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2763 } while (uv /= 10);
2764 if (sign)
2765 *--ptr = '-';
2766 *peob = ebuf;
2767 return ptr;
2768}
2769
645c22ef
DM
2770/*
2771=for apidoc sv_2pv_flags
2772
ff276b08 2773Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2774If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2775string if necessary. Normally invoked via the C<SvPV_flags> macro.
2776C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2777
2778=cut
2779*/
2780
8d6d96c1 2781char *
5aaab254 2782Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2783{
97aff369 2784 dVAR;
eb578fdb 2785 char *s;
79072805 2786
463ee0b2 2787 if (!sv) {
cdb061a3
NC
2788 if (lp)
2789 *lp = 0;
73d840c0 2790 return (char *)"";
463ee0b2 2791 }
217f6fa3
FC
2792 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2793 && SvTYPE(sv) != SVt_PVFM);
4bac9ae4
CS
2794 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2795 mg_get(sv);
2796 if (SvROK(sv)) {
2797 if (SvAMAGIC(sv)) {
2798 SV *tmpstr;
2799 if (flags & SV_SKIP_OVERLOAD)
2800 return NULL;
2801 tmpstr = AMG_CALLunary(sv, string_amg);
2802 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2803 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2804 /* Unwrap this: */
2805 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2806 */
75dfc8ec 2807
4bac9ae4
CS
2808 char *pv;
2809 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2810 if (flags & SV_CONST_RETURN) {
2811 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2812 } else {
4bac9ae4
CS
2813 pv = (flags & SV_MUTABLE_RETURN)
2814 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2815 }
4bac9ae4
CS
2816 if (lp)
2817 *lp = SvCUR(tmpstr);
2818 } else {
2819 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2820 }
4bac9ae4
CS
2821 if (SvUTF8(tmpstr))
2822 SvUTF8_on(sv);
2823 else
2824 SvUTF8_off(sv);
2825 return pv;
deb46114 2826 }
4bac9ae4
CS
2827 }
2828 {
2829 STRLEN len;
2830 char *retval;
2831 char *buffer;
2832 SV *const referent = SvRV(sv);
2833
2834 if (!referent) {
2835 len = 7;
2836 retval = buffer = savepvn("NULLREF", len);
2837 } else if (SvTYPE(referent) == SVt_REGEXP &&
2838 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2839 amagic_is_enabled(string_amg))) {
2840 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2841
2842 assert(re);
67d2d14d 2843
4bac9ae4
CS
2844 /* If the regex is UTF-8 we want the containing scalar to
2845 have an UTF-8 flag too */
2846 if (RX_UTF8(re))
2847 SvUTF8_on(sv);
2848 else
2849 SvUTF8_off(sv);
67d2d14d 2850
4bac9ae4
CS
2851 if (lp)
2852 *lp = RX_WRAPLEN(re);
67d2d14d 2853
4bac9ae4
CS
2854 return RX_WRAPPED(re);
2855 } else {
2856 const char *const typestr = sv_reftype(referent, 0);
2857 const STRLEN typelen = strlen(typestr);
2858 UV addr = PTR2UV(referent);
2859 const char *stashname = NULL;
2860 STRLEN stashnamelen = 0; /* hush, gcc */
2861 const char *buffer_end;
2862
2863 if (SvOBJECT(referent)) {
2864 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2865
2866 if (name) {
2867 stashname = HEK_KEY(name);
2868 stashnamelen = HEK_LEN(name);
2869
2870 if (HEK_UTF8(name)) {
2871 SvUTF8_on(sv);
fafee734 2872 } else {
4bac9ae4 2873 SvUTF8_off(sv);
fafee734 2874 }
fafee734 2875 } else {
4bac9ae4
CS
2876 stashname = "__ANON__";
2877 stashnamelen = 8;
fafee734 2878 }
4bac9ae4
CS
2879 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2880 + 2 * sizeof(UV) + 2 /* )\0 */;
2881 } else {
2882 len = typelen + 3 /* (0x */
2883 + 2 * sizeof(UV) + 2 /* )\0 */;
2884 }
fafee734 2885
4bac9ae4
CS
2886 Newx(buffer, len, char);
2887 buffer_end = retval = buffer + len;
2888
2889 /* Working backwards */
2890 *--retval = '\0';
2891 *--retval = ')';
2892 do {
2893 *--retval = PL_hexdigit[addr & 15];
2894 } while (addr >>= 4);
2895 *--retval = 'x';
2896 *--retval = '0';
2897 *--retval = '(';
2898
2899 retval -= typelen;
2900 memcpy(retval, typestr, typelen);
2901
2902 if (stashname) {
2903 *--retval = '=';
2904 retval -= stashnamelen;
2905 memcpy(retval, stashname, stashnamelen);
c080367d 2906 }
4bac9ae4
CS
2907 /* retval may not necessarily have reached the start of the
2908 buffer here. */
2909 assert (retval >= buffer);
2910
2911 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 2912 }
cdb061a3 2913 if (lp)
4bac9ae4
CS
2914 *lp = len;
2915 SAVEFREEPV(buffer);
2916 return retval;
79072805 2917 }
79072805 2918 }
4bac9ae4
CS
2919
2920 if (SvPOKp(sv)) {
2921 if (lp)
2922 *lp = SvCUR(sv);
2923 if (flags & SV_MUTABLE_RETURN)
2924 return SvPVX_mutable(sv);
2925 if (flags & SV_CONST_RETURN)
2926 return (char *)SvPVX_const(sv);
2927 return SvPVX(sv);
2928 }
2929
2930 if (SvIOK(sv)) {
28e5dec8
JH
2931 /* I'm assuming that if both IV and NV are equally valid then
2932 converting the IV is going to be more efficient */
e1ec3a88 2933 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2934 char buf[TYPE_CHARS(UV)];
2935 char *ebuf, *ptr;
97a130b8 2936 STRLEN len;
28e5dec8
JH
2937
2938 if (SvTYPE(sv) < SVt_PVIV)
2939 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2940 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2941 len = ebuf - ptr;
5902b6a9 2942 /* inlined from sv_setpvn */
97a130b8
NC
2943 s = SvGROW_mutable(sv, len + 1);
2944 Move(ptr, s, len, char);
2945 s += len;
28e5dec8 2946 *s = '\0';
b127e37e 2947 SvPOK_on(sv);
28e5dec8 2948 }
4bac9ae4 2949 else if (SvNOK(sv)) {
79072805
LW
2950 if (SvTYPE(sv) < SVt_PVNV)
2951 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2952 if (SvNVX(sv) == 0.0) {
2953 s = SvGROW_mutable(sv, 2);
2954 *s++ = '0';
2955 *s = '\0';
2956 } else {
2957 dSAVE_ERRNO;
2958 /* The +20 is pure guesswork. Configure test needed. --jhi */
2959 s = SvGROW_mutable(sv, NV_DIG + 20);
2960 /* some Xenix systems wipe out errno here */
b127e37e 2961
b127e37e 2962#ifndef USE_LOCALE_NUMERIC
68e8f474 2963 Gconvert(SvNVX(sv), NV_DIG, 0, s);
b127e37e 2964 SvPOK_on(sv);
68e8f474
KW
2965#else
2966 /* Gconvert always uses the current locale. That's the right thing
2967 * to do if we're supposed to be using locales. But otherwise, we
2968 * want the result to be based on the C locale, so we need to
2969 * change to the C locale during the Gconvert and then change back.
2970 * But if we're already in the C locale (PL_numeric_standard is
2971 * TRUE in that case), no need to do any changing */
7187d38e 2972 if (PL_numeric_standard || IN_SOME_LOCALE_FORM_RUNTIME) {
68e8f474 2973 Gconvert(SvNVX(sv), NV_DIG, 0, s);
28acfe03
KW
2974
2975 /* If the radix character is UTF-8, and actually is in the
2976 * output, turn on the UTF-8 flag for the scalar */
2977 if (! PL_numeric_standard
2978 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
2979 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
2980 {
2981 SvUTF8_on(sv);
2982 }
68e8f474
KW
2983 }
2984 else {
2985 char *loc = savepv(setlocale(LC_NUMERIC, NULL));
2986 setlocale(LC_NUMERIC, "C");
2987 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2988 setlocale(LC_NUMERIC, loc);
2989 Safefree(loc);
28acfe03 2990
68e8f474
KW
2991 }
2992
2993 /* We don't call SvPOK_on(), because it may come to pass that the
2994 * locale changes so that the stringification we just did is no
2995 * longer correct. We will have to re-stringify every time it is
2996 * needed */
b127e37e 2997#endif
29912d93
NC
2998 RESTORE_ERRNO;
2999 while (*s) s++;
bbce6d69 3000 }
79072805
LW
3001#ifdef hcx
3002 if (s[-1] == '.')
46fc3d4c 3003 *--s = '\0';
79072805
LW
3004#endif
3005 }
4bac9ae4
CS
3006 else if (isGV_with_GP(sv)) {
3007 GV *const gv = MUTABLE_GV(sv);
3008 SV *const buffer = sv_newmortal();
8d1c3e26 3009
4bac9ae4 3010 gv_efullname3(buffer, gv, "*");
180488f8 3011
4bac9ae4
CS
3012 assert(SvPOK(buffer));
3013 if (SvUTF8(buffer))
3014 SvUTF8_on(sv);
3015 if (lp)
3016 *lp = SvCUR(buffer);
3017 return SvPVX(buffer);
3018 }
8d919b0a
FC
3019 else if (isREGEXP(sv)) {
3020 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3021 return RX_WRAPPED((REGEXP *)sv);
3022 }
4bac9ae4 3023 else {
cdb061a3 3024 if (lp)
00b6aa41 3025 *lp = 0;
9f621bb0
NC
3026 if (flags & SV_UNDEF_RETURNS_NULL)
3027 return NULL;
9a214eec 3028 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3029 report_uninit(sv);
4bac9ae4
CS
3030 /* Typically the caller expects that sv_any is not NULL now. */
3031 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3032 sv_upgrade(sv, SVt_PV);
73d840c0 3033 return (char *)"";
79072805 3034 }
4bac9ae4 3035
cdb061a3 3036 {
823a54a3 3037 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3038 if (lp)
3039 *lp = len;
3040 SvCUR_set(sv, len);
3041 }
1d7c1841 3042 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3043 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3044 if (flags & SV_CONST_RETURN)
3045 return (char *)SvPVX_const(sv);
10516c54
NC
3046 if (flags & SV_MUTABLE_RETURN)
3047 return SvPVX_mutable(sv);
463ee0b2
LW
3048 return SvPVX(sv);
3049}
3050
645c22ef 3051/*
6050d10e
JP
3052=for apidoc sv_copypv
3053
3054Copies a stringified representation of the source SV into the
3055destination SV. Automatically performs any necessary mg_get and
54f0641b 3056coercion of numeric values into strings. Guaranteed to preserve
2575c402 3057UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3058sv_2pv[_flags] but operates directly on an SV instead of just the
3059string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3060would lose the UTF-8'ness of the PV.
3061
4bac9ae4
CS
3062=for apidoc sv_copypv_nomg
3063
3064Like sv_copypv, but doesn't invoke get magic first.
3065
3066=for apidoc sv_copypv_flags
3067
3068Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3069include SV_GMAGIC.
3070
6050d10e
JP
3071=cut
3072*/
3073
3074void
5aaab254 3075Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
6050d10e 3076{
4bac9ae4
CS
3077 PERL_ARGS_ASSERT_SV_COPYPV;
3078
3079 sv_copypv_flags(dsv, ssv, 0);
3080}
3081
3082void
5aaab254 3083Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3084{
446eaa42 3085 STRLEN len;
4bac9ae4 3086 const char *s;
7918f24d 3087
4bac9ae4 3088 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3089
4bac9ae4
CS
3090 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3091 mg_get(ssv);
3092 s = SvPV_nomg_const(ssv,len);
cb50f42d 3093 sv_setpvn(dsv,s,len);
446eaa42 3094 if (SvUTF8(ssv))
cb50f42d 3095 SvUTF8_on(dsv);
446eaa42 3096 else
cb50f42d 3097 SvUTF8_off(dsv);
6050d10e
JP
3098}
3099
3100/*
645c22ef
DM
3101=for apidoc sv_2pvbyte
3102
3103Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3104to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3105side-effect.
3106
3107Usually accessed via the C<SvPVbyte> macro.
3108
3109=cut
3110*/
3111
7340a771 3112char *
5aaab254 3113Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3114{
7918f24d
NC
3115 PERL_ARGS_ASSERT_SV_2PVBYTE;
3116
48120f8f 3117 SvGETMAGIC(sv);
4499db73
FC
3118 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3119 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3120 SV *sv2 = sv_newmortal();
48120f8f 3121 sv_copypv_nomg(sv2,sv);
a901b181
FC
3122 sv = sv2;
3123 }
0875d2fe 3124 sv_utf8_downgrade(sv,0);
71eb6d8c 3125 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3126}
3127
645c22ef 3128/*
035cbb0e
RGS
3129=for apidoc sv_2pvutf8
3130
3131Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3132to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3133
3134Usually accessed via the C<SvPVutf8> macro.
3135
3136=cut
3137*/
645c22ef 3138
7340a771 3139char *
5aaab254 3140Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3141{
7918f24d
NC
3142 PERL_ARGS_ASSERT_SV_2PVUTF8;
3143
4499db73
FC
3144 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3145 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3146 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3147 else
3148 SvGETMAGIC(sv);
3149 sv_utf8_upgrade_nomg(sv);
c3ec315f 3150 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3151}
1c846c1f 3152
7ee2227d 3153
645c22ef
DM
3154/*
3155=for apidoc sv_2bool
3156
06c841cf
FC
3157This macro is only used by sv_true() or its macro equivalent, and only if
3158the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3159It calls sv_2bool_flags with the SV_GMAGIC flag.
3160
3161=for apidoc sv_2bool_flags
3162
3163This function is only used by sv_true() and friends, and only if
fde67290 3164the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3165contain SV_GMAGIC, then it does an mg_get() first.
3166
645c22ef
DM
3167
3168=cut
3169*/
3170
463ee0b2 3171bool
5aaab254 3172Perl_sv_2bool_flags(pTHX_ SV *const sv, const I32 flags)
463ee0b2 3173{
97aff369 3174 dVAR;
7918f24d 3175
06c841cf 3176 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3177
06c841cf 3178 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3179
a0d0e21e
LW
3180 if (!SvOK(sv))
3181 return 0;
3182 if (SvROK(sv)) {
fabdb6c0 3183 if (SvAMAGIC(sv)) {
31d632c3 3184 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3185 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3186 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3187 }
3188 return SvRV(sv) != 0;
a0d0e21e 3189 }
85b7d9b3
FC
3190 if (isREGEXP(sv))
3191 return
3192 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
4bac9ae4 3193 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3194}
3195
c461cf8f
JH
3196/*
3197=for apidoc sv_utf8_upgrade
3198
78ea37eb 3199Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3200Forces the SV to string form if it is not already.
2bbc8d55 3201Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3202Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3203if the whole string is the same in UTF-8 as not.
3204Returns the number of bytes in the converted string
c461cf8f 3205
0efd0472 3206This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3207use the Encode extension for that.
3208
fe749c9a
KW
3209=for apidoc sv_utf8_upgrade_nomg
3210
fde67290 3211Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3212
8d6d96c1
HS
3213=for apidoc sv_utf8_upgrade_flags
3214
78ea37eb 3215Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3216Forces the SV to string form if it is not already.
8d6d96c1 3217Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3218if all the bytes are invariant in UTF-8.
3219If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55
SP
3220will C<mg_get> on C<sv> if appropriate, else not.
3221Returns the number of bytes in the converted string
3222C<sv_utf8_upgrade> and
8d6d96c1
HS
3223C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3224
0efd0472 3225This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3226use the Encode extension for that.
3227
8d6d96c1 3228=cut
b3ab6785
KW
3229
3230The grow version is currently not externally documented. It adds a parameter,
3231extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3232have free after it upon return. This allows the caller to reserve extra space
3233that it intends to fill, to avoid extra grows.
3234
3235Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3236which can be used to tell this function to not first check to see if there are
3237any characters that are different in UTF-8 (variant characters) which would
3238force it to allocate a new string to sv, but to assume there are. Typically
3239this flag is used by a routine that has already parsed the string to find that
3240there are such characters, and passes this information on so that the work
3241doesn't have to be repeated.
3242
3243(One might think that the calling routine could pass in the position of the
3244first such variant, so it wouldn't have to be found again. But that is not the
3245case, because typically when the caller is likely to use this flag, it won't be
3246calling this routine unless it finds something that won't fit into a byte.
3247Otherwise it tries to not upgrade and just use bytes. But some things that
3248do fit into a byte are variants in utf8, and the caller may not have been
3249keeping track of these.)
3250
3251If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3252isn't guaranteed due to having other routines do the work in some input cases,
3253or if the input is already flagged as being in utf8.
3254
3255The speed of this could perhaps be improved for many cases if someone wanted to
3256write a fast function that counts the number of variant characters in a string,
3257especially if it could return the position of the first one.
3258
8d6d96c1
HS
3259*/
3260
3261STRLEN
5aaab254 3262Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3263{
97aff369 3264 dVAR;
7918f24d 3265
b3ab6785 3266 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3267
808c356f
RGS
3268 if (sv == &PL_sv_undef)
3269 return 0;
892f9127 3270 if (!SvPOK_nog(sv)) {
e0e62c2a 3271 STRLEN len = 0;
d52b7888
NC
3272 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3273 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3274 if (SvUTF8(sv)) {
3275 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3276 return len;
b3ab6785 3277 }
d52b7888 3278 } else {
33fb6f35 3279 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3280 }
e0e62c2a 3281 }
4411f3b6 3282
f5cee72b 3283 if (SvUTF8(sv)) {
b3ab6785 3284 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3285 return SvCUR(sv);
f5cee72b 3286 }
5fec3b1d 3287
765f542d 3288 if (SvIsCOW(sv)) {
c56ed9f6 3289 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3290 }
3291
b3ab6785 3292 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3293 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3294 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3295 return SvCUR(sv);
3296 }
3297
4e93345f
KW
3298 if (SvCUR(sv) == 0) {
3299 if (extra) SvGROW(sv, extra);
3300 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3301 /* This function could be much more efficient if we
2bbc8d55 3302 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3303 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3304 * make the loop as fast as possible (although there are certainly ways
3305 * to speed this up, eg. through vectorization) */
3306 U8 * s = (U8 *) SvPVX_const(sv);
3307 U8 * e = (U8 *) SvEND(sv);
3308 U8 *t = s;
3309 STRLEN two_byte_count = 0;
c4e7c712 3310
b3ab6785
KW
3311 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3312
3313 /* See if really will need to convert to utf8. We mustn't rely on our
3314 * incoming SV being well formed and having a trailing '\0', as certain
3315 * code in pp_formline can send us partially built SVs. */
3316
c4e7c712 3317 while (t < e) {
53c1dcc0 3318 const U8 ch = *t++;
b3ab6785
KW
3319 if (NATIVE_IS_INVARIANT(ch)) continue;
3320
3321 t--; /* t already incremented; re-point to first variant */
3322 two_byte_count = 1;
3323 goto must_be_utf8;
c4e7c712 3324 }
b3ab6785
KW
3325
3326 /* utf8 conversion not needed because all are invariants. Mark as
3327 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3328 SvUTF8_on(sv);
7f0bfbea 3329 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3330 return SvCUR(sv);
3331
3332must_be_utf8:
3333
3334 /* Here, the string should be converted to utf8, either because of an
3335 * input flag (two_byte_count = 0), or because a character that
3336 * requires 2 bytes was found (two_byte_count = 1). t points either to
3337 * the beginning of the string (if we didn't examine anything), or to
3338 * the first variant. In either case, everything from s to t - 1 will
3339 * occupy only 1 byte each on output.
3340 *
3341 * There are two main ways to convert. One is to create a new string
3342 * and go through the input starting from the beginning, appending each
3343 * converted value onto the new string as we go along. It's probably
3344 * best to allocate enough space in the string for the worst possible
3345 * case rather than possibly running out of space and having to
3346 * reallocate and then copy what we've done so far. Since everything
3347 * from s to t - 1 is invariant, the destination can be initialized
3348 * with these using a fast memory copy
3349 *
3350 * The other way is to figure out exactly how big the string should be
3351 * by parsing the entire input. Then you don't have to make it big
3352 * enough to handle the worst possible case, and more importantly, if
3353 * the string you already have is large enough, you don't have to
3354 * allocate a new string, you can copy the last character in the input
3355 * string to the final position(s) that will be occupied by the
3356 * converted string and go backwards, stopping at t, since everything
3357 * before that is invariant.
3358 *
3359 * There are advantages and disadvantages to each method.
3360 *
3361 * In the first method, we can allocate a new string, do the memory
3362 * copy from the s to t - 1, and then proceed through the rest of the
3363 * string byte-by-byte.
3364 *
3365 * In the second method, we proceed through the rest of the input
3366 * string just calculating how big the converted string will be. Then
3367 * there are two cases:
3368 * 1) if the string has enough extra space to handle the converted
3369 * value. We go backwards through the string, converting until we
3370 * get to the position we are at now, and then stop. If this
3371 * position is far enough along in the string, this method is
3372 * faster than the other method. If the memory copy were the same
3373 * speed as the byte-by-byte loop, that position would be about
3374 * half-way, as at the half-way mark, parsing to the end and back
3375 * is one complete string's parse, the same amount as starting
3376 * over and going all the way through. Actually, it would be
3377 * somewhat less than half-way, as it's faster to just count bytes
3378 * than to also copy, and we don't have the overhead of allocating
3379 * a new string, changing the scalar to use it, and freeing the
3380 * existing one. But if the memory copy is fast, the break-even
3381 * point is somewhere after half way. The counting loop could be
3382 * sped up by vectorization, etc, to move the break-even point
3383 * further towards the beginning.
3384 * 2) if the string doesn't have enough space to handle the converted
3385 * value. A new string will have to be allocated, and one might
3386 * as well, given that, start from the beginning doing the first
3387 * method. We've spent extra time parsing the string and in
3388 * exchange all we've gotten is that we know precisely how big to
3389 * make the new one. Perl is more optimized for time than space,
3390 * so this case is a loser.
3391 * So what I've decided to do is not use the 2nd method unless it is
3392 * guaranteed that a new string won't have to be allocated, assuming
3393 * the worst case. I also decided not to put any more conditions on it
3394 * than this, for now. It seems likely that, since the worst case is
3395 * twice as big as the unknown portion of the string (plus 1), we won't
3396 * be guaranteed enough space, causing us to go to the first method,
3397 * unless the string is short, or the first variant character is near
3398 * the end of it. In either of these cases, it seems best to use the
3399 * 2nd method. The only circumstance I can think of where this would
3400 * be really slower is if the string had once had much more data in it
3401 * than it does now, but there is still a substantial amount in it */
3402
3403 {
3404 STRLEN invariant_head = t - s;
3405 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3406 if (SvLEN(sv) < size) {
3407
3408 /* Here, have decided to allocate a new string */
3409
3410 U8 *dst;
3411 U8 *d;
3412
3413 Newx(dst, size, U8);
3414
3415 /* If no known invariants at the beginning of the input string,
3416 * set so starts from there. Otherwise, can use memory copy to
3417 * get up to where we are now, and then start from here */
3418
3419 if (invariant_head <= 0) {
3420 d = dst;
3421 } else {
3422 Copy(s, dst, invariant_head, char);
3423 d = dst + invariant_head;
3424 }
3425
3426 while (t < e) {
3427 const UV uv = NATIVE8_TO_UNI(*t++);
3428 if (UNI_IS_INVARIANT(uv))
3429 *d++ = (U8)UNI_TO_NATIVE(uv);
3430 else {
3431 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3432 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3433 }
3434 }
3435 *d = '\0';
3436 SvPV_free(sv); /* No longer using pre-existing string */
3437 SvPV_set(sv, (char*)dst);
3438 SvCUR_set(sv, d - dst);
3439 SvLEN_set(sv, size);
3440 } else {
3441
3442 /* Here, have decided to get the exact size of the string.
3443 * Currently this happens only when we know that there is
3444 * guaranteed enough space to fit the converted string, so
3445 * don't have to worry about growing. If two_byte_count is 0,
3446 * then t points to the first byte of the string which hasn't
3447 * been examined yet. Otherwise two_byte_count is 1, and t
3448 * points to the first byte in the string that will expand to
3449 * two. Depending on this, start examining at t or 1 after t.
3450 * */
3451
3452 U8 *d = t + two_byte_count;
3453
3454
3455 /* Count up the remaining bytes that expand to two */
3456
3457 while (d < e) {
3458 const U8 chr = *d++;
3459 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3460 }
3461
3462 /* The string will expand by just the number of bytes that
3463 * occupy two positions. But we are one afterwards because of
3464 * the increment just above. This is the place to put the
3465 * trailing NUL, and to set the length before we decrement */
3466
3467 d += two_byte_count;
3468 SvCUR_set(sv, d - s);
3469 *d-- = '\0';
3470
3471
3472 /* Having decremented d, it points to the position to put the
3473 * very last byte of the expanded string. Go backwards through
3474 * the string, copying and expanding as we go, stopping when we
3475 * get to the part that is invariant the rest of the way down */
3476
3477 e--;
3478 while (e >= t) {
3479 const U8 ch = NATIVE8_TO_UNI(*e--);
3480 if (UNI_IS_INVARIANT(ch)) {
3481 *d-- = UNI_TO_NATIVE(ch);
3482 } else {
3483 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3484 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3485 }
3486 }
3487 }
75da9d4c
DM
3488
3489 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3490 /* Update pos. We do it at the end rather than during
3491 * the upgrade, to avoid slowing down the common case
3492 * (upgrade without pos) */
3493 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3494 if (mg) {
3495 I32 pos = mg->mg_len;
3496 if (pos > 0 && (U32)pos > invariant_head) {
3497 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3498 STRLEN n = (U32)pos - invariant_head;
3499 while (n > 0) {
3500 if (UTF8_IS_START(*d))
3501 d++;
3502 d++;
3503 n--;
3504 }
3505 mg->mg_len = d - (U8*)SvPVX(sv);
3506 }
3507 }
3508 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3509 magic_setutf8(sv,mg); /* clear UTF8 cache */
3510 }
b3ab6785 3511 }
560a288e 3512 }
b3ab6785
KW
3513
3514 /* Mark as UTF-8 even if no variant - saves scanning loop */
3515 SvUTF8_on(sv);
4411f3b6 3516 return SvCUR(sv);
560a288e
GS
3517}
3518
c461cf8f
JH
3519/*
3520=for apidoc sv_utf8_downgrade
3521
78ea37eb 3522Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3523If the PV contains a character that cannot fit
3524in a byte, this conversion will fail;
78ea37eb 3525in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3526true, croaks.
3527
0efd0472 3528This is not a general purpose Unicode to byte encoding interface:
13a6c0e0
JH
3529use the Encode extension for that.
3530
c461cf8f
JH
3531=cut
3532*/
3533
560a288e 3534bool
5aaab254 3535Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
560a288e 3536{
97aff369 3537 dVAR;
7918f24d
NC
3538
3539 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3540
78ea37eb 3541 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3542 if (SvCUR(sv)) {
03cfe0ae 3543 U8 *s;
652088fc 3544 STRLEN len;
75da9d4c 3545 int mg_flags = SV_GMAGIC;
fa301091 3546
765f542d 3547 if (SvIsCOW(sv)) {
c56ed9f6 3548 S_sv_uncow(aTHX_ sv, 0);
765f542d 3549 }
75da9d4c
DM
3550 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3551 /* update pos */
3552 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3553 if (mg) {
3554 I32 pos = mg->mg_len;
3555 if (pos > 0) {
3556 sv_pos_b2u(sv, &pos);
3557 mg_flags = 0; /* sv_pos_b2u does get magic */
3558 mg->mg_len = pos;
3559 }
3560 }
3561 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3562 magic_setutf8(sv,mg); /* clear UTF8 cache */
3563
3564 }
3565 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3566
03cfe0ae 3567 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3568 if (fail_ok)
3569 return FALSE;
3570 else {
3571 if (PL_op)
3572 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3573 OP_DESC(PL_op));
fa301091
JH
3574 else
3575 Perl_croak(aTHX_ "Wide character");
3576 }
4b3603a4 3577 }
b162af07 3578 SvCUR_set(sv, len);
67e989fb 3579 }
560a288e 3580 }
ffebcc3e 3581 SvUTF8_off(sv);
560a288e
GS
3582 return TRUE;
3583}
3584
c461cf8f
JH
3585/*
3586=for apidoc sv_utf8_encode
3587
78ea37eb
ST
3588Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3589flag off so that it looks like octets again.
c461cf8f
JH
3590
3591=cut
3592*/
3593
560a288e 3594void
5aaab254 3595Perl_sv_utf8_encode(pTHX_ SV *const sv)
560a288e 3596{
7918f24d
NC
3597 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3598
4c94c214 3599 if (SvREADONLY(sv)) {
654c723f 3600 sv_force_normal_flags(sv, 0);
4c94c214 3601 }
a5f5288a 3602 (void) sv_utf8_upgrade(sv);
560a288e
GS
3603 SvUTF8_off(sv);
3604}
3605
4411f3b6
NIS
3606/*
3607=for apidoc sv_utf8_decode
3608
78ea37eb
ST
3609If the PV of the SV is an octet sequence in UTF-8
3610and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3611so that it looks like a character. If the PV contains only single-byte
694cf0d2 3612characters, the C<SvUTF8> flag stays off.
78ea37eb 3613Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3614
3615=cut
3616*/
3617
560a288e 3618bool
5aaab254 3619Perl_sv_utf8_decode(pTHX_ SV *const sv)
560a288e 3620{
7918f24d
NC
3621 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3622
78ea37eb 3623 if (SvPOKp(sv)) {
75da9d4c 3624 const U8 *start, *c;
93524f2b 3625 const U8 *e;
9cbac4c7 3626
645c22ef
DM
3627 /* The octets may have got themselves encoded - get them back as
3628 * bytes