This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
sv.c:sv_reset: Remove %ENV handling; fix for VMS
[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
bd81e77b 1473char *
5aaab254 1474Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1475{
eb578fdb 1476 char *s;
93e68bfb 1477
7918f24d
NC
1478 PERL_ARGS_ASSERT_SV_GROW;
1479
bd81e77b
NC
1480#ifdef HAS_64K_LIMIT
1481 if (newlen >= 0x10000) {
1482 PerlIO_printf(Perl_debug_log,
1483 "Allocation too large: %"UVxf"\n", (UV)newlen);
1484 my_exit(1);
1485 }
1486#endif /* HAS_64K_LIMIT */
1487 if (SvROK(sv))
1488 sv_unref(sv);
1489 if (SvTYPE(sv) < SVt_PV) {
1490 sv_upgrade(sv, SVt_PV);
1491 s = SvPVX_mutable(sv);
1492 }
1493 else if (SvOOK(sv)) { /* pv is offset? */
1494 sv_backoff(sv);
1495 s = SvPVX_mutable(sv);
1496 if (newlen > SvLEN(sv))
1497 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1498#ifdef HAS_64K_LIMIT
1499 if (newlen >= 0x10000)
1500 newlen = 0xFFFF;
1501#endif
1502 }
1503 else
db2c6cb3
FC
1504 {
1505 if (SvIsCOW(sv)) sv_force_normal(sv);
bd81e77b 1506 s = SvPVX_mutable(sv);
db2c6cb3 1507 }
aeb18a1e 1508
cbcb2a16
DM
1509#ifdef PERL_NEW_COPY_ON_WRITE
1510 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1511 * to store the COW count. So in general, allocate one more byte than
1512 * asked for, to make it likely this byte is always spare: and thus
1513 * make more strings COW-able.
1514 * If the new size is a big power of two, don't bother: we assume the
1515 * caller wanted a nice 2^N sized block and will be annoyed at getting
1516 * 2^N+1 */
1517 if (newlen & 0xff)
1518 newlen++;
1519#endif
1520
bd81e77b 1521 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1522 STRLEN minlen = SvCUR(sv);
1523 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1524 if (newlen < minlen)
1525 newlen = minlen;
aedff202 1526#ifndef Perl_safesysmalloc_size
bd81e77b 1527 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1528#endif
98653f18 1529 if (SvLEN(sv) && s) {
10edeb5d 1530 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1531 }
1532 else {
10edeb5d 1533 s = (char*)safemalloc(newlen);
bd81e77b
NC
1534 if (SvPVX_const(sv) && SvCUR(sv)) {
1535 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1536 }
1537 }
1538 SvPV_set(sv, s);
ca7c1a29 1539#ifdef Perl_safesysmalloc_size
98653f18
NC
1540 /* Do this here, do it once, do it right, and then we will never get
1541 called back into sv_grow() unless there really is some growing
1542 needed. */
ca7c1a29 1543 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1544#else
bd81e77b 1545 SvLEN_set(sv, newlen);
98653f18 1546#endif
bd81e77b
NC
1547 }
1548 return s;
1549}
aeb18a1e 1550
bd81e77b
NC
1551/*
1552=for apidoc sv_setiv
932e9ff9 1553
bd81e77b
NC
1554Copies an integer into the given SV, upgrading first if necessary.
1555Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1556
bd81e77b
NC
1557=cut
1558*/
463ee0b2 1559
bd81e77b 1560void
5aaab254 1561Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1562{
97aff369 1563 dVAR;
7918f24d
NC
1564
1565 PERL_ARGS_ASSERT_SV_SETIV;
1566
bd81e77b
NC
1567 SV_CHECK_THINKFIRST_COW_DROP(sv);
1568 switch (SvTYPE(sv)) {
1569 case SVt_NULL:
bd81e77b 1570 case SVt_NV:
3376de98 1571 sv_upgrade(sv, SVt_IV);
bd81e77b 1572 break;
bd81e77b
NC
1573 case SVt_PV:
1574 sv_upgrade(sv, SVt_PVIV);
1575 break;
463ee0b2 1576
bd81e77b 1577 case SVt_PVGV:
6e592b3a
BM
1578 if (!isGV_with_GP(sv))
1579 break;
bd81e77b
NC
1580 case SVt_PVAV:
1581 case SVt_PVHV:
1582 case SVt_PVCV:
1583 case SVt_PVFM:
1584 case SVt_PVIO:
22e74366 1585 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1586 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1587 OP_DESC(PL_op));
42d0e0b7 1588 default: NOOP;
bd81e77b
NC
1589 }
1590 (void)SvIOK_only(sv); /* validate number */
1591 SvIV_set(sv, i);
1592 SvTAINT(sv);
1593}
932e9ff9 1594
bd81e77b
NC
1595/*
1596=for apidoc sv_setiv_mg
d33b2eba 1597
bd81e77b 1598Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1599
bd81e77b
NC
1600=cut
1601*/
d33b2eba 1602
bd81e77b 1603void
5aaab254 1604Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1605{
7918f24d
NC
1606 PERL_ARGS_ASSERT_SV_SETIV_MG;
1607
bd81e77b
NC
1608 sv_setiv(sv,i);
1609 SvSETMAGIC(sv);
1610}
727879eb 1611
bd81e77b
NC
1612/*
1613=for apidoc sv_setuv
d33b2eba 1614
bd81e77b
NC
1615Copies an unsigned integer into the given SV, upgrading first if necessary.
1616Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1617
bd81e77b
NC
1618=cut
1619*/
d33b2eba 1620
bd81e77b 1621void
5aaab254 1622Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1623{
7918f24d
NC
1624 PERL_ARGS_ASSERT_SV_SETUV;
1625
013abb9b
NC
1626 /* With the if statement to ensure that integers are stored as IVs whenever
1627 possible:
bd81e77b 1628 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1629
bd81e77b
NC
1630 without
1631 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1632
013abb9b
NC
1633 If you wish to remove the following if statement, so that this routine
1634 (and its callers) always return UVs, please benchmark to see what the
1635 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1636 */
1637 if (u <= (UV)IV_MAX) {
1638 sv_setiv(sv, (IV)u);
1639 return;
1640 }
1641 sv_setiv(sv, 0);
1642 SvIsUV_on(sv);
1643 SvUV_set(sv, u);
1644}
d33b2eba 1645
bd81e77b
NC
1646/*
1647=for apidoc sv_setuv_mg
727879eb 1648
bd81e77b 1649Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1650
bd81e77b
NC
1651=cut
1652*/
5e2fc214 1653
bd81e77b 1654void
5aaab254 1655Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1656{
7918f24d
NC
1657 PERL_ARGS_ASSERT_SV_SETUV_MG;
1658
bd81e77b
NC
1659 sv_setuv(sv,u);
1660 SvSETMAGIC(sv);
1661}
5e2fc214 1662
954c1994 1663/*
bd81e77b 1664=for apidoc sv_setnv
954c1994 1665
bd81e77b
NC
1666Copies a double into the given SV, upgrading first if necessary.
1667Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1668
1669=cut
1670*/
1671
63f97190 1672void
5aaab254 1673Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1674{
97aff369 1675 dVAR;
7918f24d
NC
1676
1677 PERL_ARGS_ASSERT_SV_SETNV;
1678
bd81e77b
NC
1679 SV_CHECK_THINKFIRST_COW_DROP(sv);
1680 switch (SvTYPE(sv)) {
79072805 1681 case SVt_NULL:
79072805 1682 case SVt_IV:
bd81e77b 1683 sv_upgrade(sv, SVt_NV);
79072805
LW
1684 break;
1685 case SVt_PV:
79072805 1686 case SVt_PVIV:
bd81e77b 1687 sv_upgrade(sv, SVt_PVNV);
79072805 1688 break;
bd4b1eb5 1689
bd4b1eb5 1690 case SVt_PVGV:
6e592b3a
BM
1691 if (!isGV_with_GP(sv))
1692 break;
bd81e77b
NC
1693 case SVt_PVAV:
1694 case SVt_PVHV:
79072805 1695 case SVt_PVCV:
bd81e77b
NC
1696 case SVt_PVFM:
1697 case SVt_PVIO:
22e74366 1698 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1699 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1700 OP_DESC(PL_op));
42d0e0b7 1701 default: NOOP;
2068cd4d 1702 }
bd81e77b
NC
1703 SvNV_set(sv, num);
1704 (void)SvNOK_only(sv); /* validate number */
1705 SvTAINT(sv);
79072805
LW
1706}
1707
645c22ef 1708/*
bd81e77b 1709=for apidoc sv_setnv_mg
645c22ef 1710
bd81e77b 1711Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1712
1713=cut
1714*/
1715
bd81e77b 1716void
5aaab254 1717Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1718{
7918f24d
NC
1719 PERL_ARGS_ASSERT_SV_SETNV_MG;
1720
bd81e77b
NC
1721 sv_setnv(sv,num);
1722 SvSETMAGIC(sv);
79072805
LW
1723}
1724
bd81e77b
NC
1725/* Print an "isn't numeric" warning, using a cleaned-up,
1726 * printable version of the offending string
1727 */
954c1994 1728
bd81e77b 1729STATIC void
aad570aa 1730S_not_a_number(pTHX_ SV *const sv)
79072805 1731{
97aff369 1732 dVAR;
bd81e77b
NC
1733 SV *dsv;
1734 char tmpbuf[64];
1735 const char *pv;
94463019 1736
7918f24d
NC
1737 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1738
94463019 1739 if (DO_UTF8(sv)) {
84bafc02 1740 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;
551405c4 1744 const char * const limit = tmpbuf + sizeof(tmpbuf) - 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
533c011a 1793 if (PL_op)
9014280d 1794 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1795 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1796 "Argument \"%s\" isn't numeric in %s", pv,
1797 OP_DESC(PL_op));
a0d0e21e 1798 else
9014280d 1799 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1800 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1801 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1802}
1803
c2988b20
NC
1804/*
1805=for apidoc looks_like_number
1806
645c22ef
DM
1807Test if the content of an SV looks like a number (or is a number).
1808C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1809non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1810ignored.
c2988b20
NC
1811
1812=cut
1813*/
1814
1815I32
aad570aa 1816Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1817{
eb578fdb 1818 const char *sbegin;
c2988b20
NC
1819 STRLEN len;
1820
7918f24d
NC
1821 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1822
f52e41ad
FC
1823 if (SvPOK(sv) || SvPOKp(sv)) {
1824 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1825 }
c2988b20 1826 else
e0ab1c0e 1827 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1828 return grok_number(sbegin, len, NULL);
1829}
25da4f38 1830
19f6321d
NC
1831STATIC bool
1832S_glob_2number(pTHX_ GV * const gv)
180488f8 1833{
7918f24d
NC
1834 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1835
675c862f
AL
1836 /* We know that all GVs stringify to something that is not-a-number,
1837 so no need to test that. */
1838 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1839 {
1840 SV *const buffer = sv_newmortal();
1841 gv_efullname3(buffer, gv, "*");
675c862f 1842 not_a_number(buffer);
8e629ff4 1843 }
675c862f
AL
1844 /* We just want something true to return, so that S_sv_2iuv_common
1845 can tail call us and return true. */
19f6321d 1846 return TRUE;
675c862f
AL
1847}
1848
25da4f38
IZ
1849/* Actually, ISO C leaves conversion of UV to IV undefined, but
1850 until proven guilty, assume that things are not that bad... */
1851
645c22ef
DM
1852/*
1853 NV_PRESERVES_UV:
1854
1855 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1856 an IV (an assumption perl has been based on to date) it becomes necessary
1857 to remove the assumption that the NV always carries enough precision to
1858 recreate the IV whenever needed, and that the NV is the canonical form.
1859 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1860 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1861 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1862 1) to distinguish between IV/UV/NV slots that have cached a valid
1863 conversion where precision was lost and IV/UV/NV slots that have a
1864 valid conversion which has lost no precision
645c22ef 1865 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1866 would lose precision, the precise conversion (or differently
1867 imprecise conversion) is also performed and cached, to prevent
1868 requests for different numeric formats on the same SV causing
1869 lossy conversion chains. (lossless conversion chains are perfectly
1870 acceptable (still))
1871
1872
1873 flags are used:
1874 SvIOKp is true if the IV slot contains a valid value
1875 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1876 SvNOKp is true if the NV slot contains a valid value
1877 SvNOK is true only if the NV value is accurate
1878
1879 so
645c22ef 1880 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1881 IV(or UV) would lose accuracy over a direct conversion from PV to
1882 IV(or UV). If it would, cache both conversions, return NV, but mark
1883 SV as IOK NOKp (ie not NOK).
1884
645c22ef 1885 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1886 NV would lose accuracy over a direct conversion from PV to NV. If it
1887 would, cache both conversions, flag similarly.
1888
1889 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1890 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1891 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1892 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1893 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1894
645c22ef
DM
1895 The benefit of this is that operations such as pp_add know that if
1896 SvIOK is true for both left and right operands, then integer addition
1897 can be used instead of floating point (for cases where the result won't
1898 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1899 loss of precision compared with integer addition.
1900
1901 * making IV and NV equal status should make maths accurate on 64 bit
1902 platforms
1903 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1904 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1905 looking for SvIOK and checking for overflow will not outweigh the
1906 fp to integer speedup)
1907 * will slow down integer operations (callers of SvIV) on "inaccurate"
1908 values, as the change from SvIOK to SvIOKp will cause a call into
1909 sv_2iv each time rather than a macro access direct to the IV slot
1910 * should speed up number->string conversion on integers as IV is
645c22ef 1911 favoured when IV and NV are equally accurate
28e5dec8
JH
1912
1913 ####################################################################
645c22ef
DM
1914 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1915 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1916 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1917 ####################################################################
1918
645c22ef 1919 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1920 performance ratio.
1921*/
1922
1923#ifndef NV_PRESERVES_UV
645c22ef
DM
1924# define IS_NUMBER_UNDERFLOW_IV 1
1925# define IS_NUMBER_UNDERFLOW_UV 2
1926# define IS_NUMBER_IV_AND_UV 2
1927# define IS_NUMBER_OVERFLOW_IV 4
1928# define IS_NUMBER_OVERFLOW_UV 5
1929
1930/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1931
1932/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1933STATIC int
5aaab254 1934S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
1935# ifdef DEBUGGING
1936 , I32 numtype
1937# endif
1938 )
28e5dec8 1939{
97aff369 1940 dVAR;
7918f24d
NC
1941
1942 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1943
3f7c398e 1944 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
1945 if (SvNVX(sv) < (NV)IV_MIN) {
1946 (void)SvIOKp_on(sv);
1947 (void)SvNOK_on(sv);
45977657 1948 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1949 return IS_NUMBER_UNDERFLOW_IV;
1950 }
1951 if (SvNVX(sv) > (NV)UV_MAX) {
1952 (void)SvIOKp_on(sv);
1953 (void)SvNOK_on(sv);
1954 SvIsUV_on(sv);
607fa7f2 1955 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1956 return IS_NUMBER_OVERFLOW_UV;
1957 }
c2988b20
NC
1958 (void)SvIOKp_on(sv);
1959 (void)SvNOK_on(sv);
1960 /* Can't use strtol etc to convert this string. (See truth table in
1961 sv_2iv */
1962 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1963 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1964 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1965 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1966 } else {
1967 /* Integer is imprecise. NOK, IOKp */
1968 }
1969 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1970 }
1971 SvIsUV_on(sv);
607fa7f2 1972 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1973 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1974 if (SvUVX(sv) == UV_MAX) {
1975 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1976 possibly be preserved by NV. Hence, it must be overflow.
1977 NOK, IOKp */
1978 return IS_NUMBER_OVERFLOW_UV;
1979 }
1980 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1981 } else {
1982 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1983 }
c2988b20 1984 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1985}
645c22ef
DM
1986#endif /* !NV_PRESERVES_UV*/
1987
af359546 1988STATIC bool
7918f24d
NC
1989S_sv_2iuv_common(pTHX_ SV *const sv)
1990{
97aff369 1991 dVAR;
7918f24d
NC
1992
1993 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1994
af359546 1995 if (SvNOKp(sv)) {
28e5dec8
JH
1996 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1997 * without also getting a cached IV/UV from it at the same time
1998 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1999 * IV or UV at same time to avoid this. */
2000 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2001
2002 if (SvTYPE(sv) == SVt_NV)
2003 sv_upgrade(sv, SVt_PVNV);
2004
28e5dec8
JH
2005 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2006 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2007 certainly cast into the IV range at IV_MAX, whereas the correct
2008 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2009 cases go to UV */
cab190d4
JD
2010#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2011 if (Perl_isnan(SvNVX(sv))) {
2012 SvUV_set(sv, 0);
2013 SvIsUV_on(sv);
fdbe6d7c 2014 return FALSE;
cab190d4 2015 }
cab190d4 2016#endif
28e5dec8 2017 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2018 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2019 if (SvNVX(sv) == (NV) SvIVX(sv)
2020#ifndef NV_PRESERVES_UV
2021 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2022 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2023 /* Don't flag it as "accurately an integer" if the number
2024 came from a (by definition imprecise) NV operation, and
2025 we're outside the range of NV integer precision */
2026#endif
2027 ) {
a43d94f2
NC
2028 if (SvNOK(sv))
2029 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2030 else {
2031 /* scalar has trailing garbage, eg "42a" */
2032 }
28e5dec8 2033 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2034 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2035 PTR2UV(sv),
2036 SvNVX(sv),
2037 SvIVX(sv)));
2038
2039 } else {
2040 /* IV not precise. No need to convert from PV, as NV
2041 conversion would already have cached IV if it detected
2042 that PV->IV would be better than PV->NV->IV
2043 flags already correct - don't set public IOK. */
2044 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2045 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2046 PTR2UV(sv),
2047 SvNVX(sv),
2048 SvIVX(sv)));
2049 }
2050 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2051 but the cast (NV)IV_MIN rounds to a the value less (more
2052 negative) than IV_MIN which happens to be equal to SvNVX ??
2053 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2054 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2055 (NV)UVX == NVX are both true, but the values differ. :-(
2056 Hopefully for 2s complement IV_MIN is something like
2057 0x8000000000000000 which will be exact. NWC */
d460ef45 2058 }
25da4f38 2059 else {
607fa7f2 2060 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2061 if (
2062 (SvNVX(sv) == (NV) SvUVX(sv))
2063#ifndef NV_PRESERVES_UV
2064 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2065 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2066 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2067 /* Don't flag it as "accurately an integer" if the number
2068 came from a (by definition imprecise) NV operation, and
2069 we're outside the range of NV integer precision */
2070#endif
a43d94f2 2071 && SvNOK(sv)
28e5dec8
JH
2072 )
2073 SvIOK_on(sv);
25da4f38 2074 SvIsUV_on(sv);
1c846c1f 2075 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2076 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2077 PTR2UV(sv),
57def98f
JH
2078 SvUVX(sv),
2079 SvUVX(sv)));
25da4f38 2080 }
748a9306 2081 }
cd84013a 2082 else if (SvPOKp(sv)) {
c2988b20 2083 UV value;
504618e9 2084 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2085 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2086 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2087 the same as the direct translation of the initial string
2088 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2089 be careful to ensure that the value with the .456 is around if the
2090 NV value is requested in the future).
1c846c1f 2091
af359546 2092 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2093 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2094 cache the NV if we are sure it's not needed.
25da4f38 2095 */
16b7a9a4 2096
c2988b20
NC
2097 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2098 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2099 == IS_NUMBER_IN_UV) {
5e045b90 2100 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2101 if (SvTYPE(sv) < SVt_PVIV)
2102 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2103 (void)SvIOK_on(sv);
c2988b20
NC
2104 } else if (SvTYPE(sv) < SVt_PVNV)
2105 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2106
f2524eef 2107 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2108 we aren't going to call atof() below. If NVs don't preserve UVs
2109 then the value returned may have more precision than atof() will
2110 return, even though value isn't perfectly accurate. */
2111 if ((numtype & (IS_NUMBER_IN_UV
2112#ifdef NV_PRESERVES_UV
2113 | IS_NUMBER_NOT_INT
2114#endif
2115 )) == IS_NUMBER_IN_UV) {
2116 /* This won't turn off the public IOK flag if it was set above */
2117 (void)SvIOKp_on(sv);
2118
2119 if (!(numtype & IS_NUMBER_NEG)) {
2120 /* positive */;
2121 if (value <= (UV)IV_MAX) {
45977657 2122 SvIV_set(sv, (IV)value);
c2988b20 2123 } else {
af359546 2124 /* it didn't overflow, and it was positive. */
607fa7f2 2125 SvUV_set(sv, value);
c2988b20
NC
2126 SvIsUV_on(sv);
2127 }
2128 } else {
2129 /* 2s complement assumption */
2130 if (value <= (UV)IV_MIN) {
45977657 2131 SvIV_set(sv, -(IV)value);
c2988b20
NC
2132 } else {
2133 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2134 I'm assuming it will be rare. */
c2988b20
NC
2135 if (SvTYPE(sv) < SVt_PVNV)
2136 sv_upgrade(sv, SVt_PVNV);
2137 SvNOK_on(sv);
2138 SvIOK_off(sv);
2139 SvIOKp_on(sv);
9d6ce603 2140 SvNV_set(sv, -(NV)value);
45977657 2141 SvIV_set(sv, IV_MIN);
c2988b20
NC
2142 }
2143 }
2144 }
2145 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2146 will be in the previous block to set the IV slot, and the next
2147 block to set the NV slot. So no else here. */
2148
2149 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2150 != IS_NUMBER_IN_UV) {
2151 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2152 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2153
c2988b20
NC
2154 if (! numtype && ckWARN(WARN_NUMERIC))
2155 not_a_number(sv);
28e5dec8 2156
65202027 2157#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2158 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2159 PTR2UV(sv), SvNVX(sv)));
65202027 2160#else
1779d84d 2161 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2162 PTR2UV(sv), SvNVX(sv)));
65202027 2163#endif
28e5dec8 2164
28e5dec8 2165#ifdef NV_PRESERVES_UV
af359546
NC
2166 (void)SvIOKp_on(sv);
2167 (void)SvNOK_on(sv);
2168 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2169 SvIV_set(sv, I_V(SvNVX(sv)));
2170 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2171 SvIOK_on(sv);
2172 } else {
6f207bd3 2173 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2174 }
2175 /* UV will not work better than IV */
2176 } else {
2177 if (SvNVX(sv) > (NV)UV_MAX) {
2178 SvIsUV_on(sv);
2179 /* Integer is inaccurate. NOK, IOKp, is UV */
2180 SvUV_set(sv, UV_MAX);
af359546
NC
2181 } else {
2182 SvUV_set(sv, U_V(SvNVX(sv)));
2183 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2184 NV preservse UV so can do correct comparison. */
2185 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2186 SvIOK_on(sv);
af359546 2187 } else {
6f207bd3 2188 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2189 }
2190 }
4b0c9573 2191 SvIsUV_on(sv);
af359546 2192 }
28e5dec8 2193#else /* NV_PRESERVES_UV */
c2988b20
NC
2194 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2195 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2196 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2197 grok_number above. The NV slot has just been set using
2198 Atof. */
560b0c46 2199 SvNOK_on(sv);
c2988b20
NC
2200 assert (SvIOKp(sv));
2201 } else {
2202 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2203 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2204 /* Small enough to preserve all bits. */
2205 (void)SvIOKp_on(sv);
2206 SvNOK_on(sv);
45977657 2207 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2208 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2209 SvIOK_on(sv);
2210 /* Assumption: first non-preserved integer is < IV_MAX,
2211 this NV is in the preserved range, therefore: */
2212 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2213 < (UV)IV_MAX)) {
32fdb065 2214 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
2215 }
2216 } else {
2217 /* IN_UV NOT_INT
2218 0 0 already failed to read UV.
2219 0 1 already failed to read UV.
2220 1 0 you won't get here in this case. IV/UV
2221 slot set, public IOK, Atof() unneeded.
2222 1 1 already read UV.
2223 so there's no point in sv_2iuv_non_preserve() attempting
2224 to use atol, strtol, strtoul etc. */
47031da6 2225# ifdef DEBUGGING
40a17c4c 2226 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2227# else
2228 sv_2iuv_non_preserve (sv);
2229# endif
c2988b20
NC
2230 }
2231 }
28e5dec8 2232#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2233 /* It might be more code efficient to go through the entire logic above
2234 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2235 gets complex and potentially buggy, so more programmer efficient
2236 to do it this way, by turning off the public flags: */
2237 if (!numtype)
2238 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2239 }
af359546
NC
2240 }
2241 else {
675c862f 2242 if (isGV_with_GP(sv))
159b6efe 2243 return glob_2number(MUTABLE_GV(sv));
180488f8 2244
9a214eec 2245 if (!SvPADTMP(sv)) {
af359546
NC
2246 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2247 report_uninit(sv);
2248 }
25da4f38
IZ
2249 if (SvTYPE(sv) < SVt_IV)
2250 /* Typically the caller expects that sv_any is not NULL now. */
2251 sv_upgrade(sv, SVt_IV);
af359546
NC
2252 /* Return 0 from the caller. */
2253 return TRUE;
2254 }
2255 return FALSE;
2256}
2257
2258/*
2259=for apidoc sv_2iv_flags
2260
2261Return the integer value of an SV, doing any necessary string
2262conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2263Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2264
2265=cut
2266*/
2267
2268IV
5aaab254 2269Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2270{
97aff369 2271 dVAR;
4bac9ae4 2272
af359546 2273 if (!sv)
a0d0e21e 2274 return 0;
4bac9ae4 2275
217f6fa3
FC
2276 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2277 && SvTYPE(sv) != SVt_PVFM);
2278
4bac9ae4
CS
2279 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2280 mg_get(sv);
2281
2282 if (SvROK(sv)) {
2283 if (SvAMAGIC(sv)) {
2284 SV * tmpstr;
2285 if (flags & SV_SKIP_OVERLOAD)
2286 return 0;
2287 tmpstr = AMG_CALLunary(sv, numer_amg);
2288 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2289 return SvIV(tmpstr);
2290 }
2291 }
2292 return PTR2IV(SvRV(sv));
2293 }
2294
8d919b0a 2295 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2296 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2297 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2298 In practice they are extremely unlikely to actually get anywhere
2299 accessible by user Perl code - the only way that I'm aware of is when
2300 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2301
2302 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2303 */
8d919b0a 2304 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2305 {
71c558c3 2306 UV value;
8d919b0a
FC
2307 const char * const ptr =
2308 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2309 const int numtype
8d919b0a 2310 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2311
2312 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2313 == IS_NUMBER_IN_UV) {
2314 /* It's definitely an integer */
2315 if (numtype & IS_NUMBER_NEG) {
2316 if (value < (UV)IV_MIN)
2317 return -(IV)value;
2318 } else {
2319 if (value < (UV)IV_MAX)
2320 return (IV)value;
2321 }
2322 }
2323 if (!numtype) {
2324 if (ckWARN(WARN_NUMERIC))
2325 not_a_number(sv);
2326 }
8d919b0a 2327 return I_V(Atof(ptr));
e20b6c3b 2328 }
4bac9ae4
CS
2329 }
2330
2331 if (SvTHINKFIRST(sv)) {
73381561 2332#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2333 if (SvIsCOW(sv)) {
2334 sv_force_normal_flags(sv, 0);
2335 }
73381561 2336#endif
af359546
NC
2337 if (SvREADONLY(sv) && !SvOK(sv)) {
2338 if (ckWARN(WARN_UNINITIALIZED))
2339 report_uninit(sv);
2340 return 0;
2341 }
2342 }
4bac9ae4 2343
af359546
NC
2344 if (!SvIOKp(sv)) {
2345 if (S_sv_2iuv_common(aTHX_ sv))
2346 return 0;
79072805 2347 }
4bac9ae4 2348
1d7c1841
GS
2349 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2350 PTR2UV(sv),SvIVX(sv)));
25da4f38 2351 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2352}
2353
645c22ef 2354/*
891f9566 2355=for apidoc sv_2uv_flags
645c22ef
DM
2356
2357Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2358conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2359Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2360
2361=cut
2362*/
2363
ff68c719 2364UV
5aaab254 2365Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2366{
97aff369 2367 dVAR;
4bac9ae4 2368
ff68c719
PP
2369 if (!sv)
2370 return 0;
4bac9ae4
CS
2371
2372 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2373 mg_get(sv);
2374
2375 if (SvROK(sv)) {
2376 if (SvAMAGIC(sv)) {
2377 SV *tmpstr;
2378 if (flags & SV_SKIP_OVERLOAD)
2379 return 0;
2380 tmpstr = AMG_CALLunary(sv, numer_amg);
2381 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2382 return SvUV(tmpstr);
2383 }
2384 }
2385 return PTR2UV(SvRV(sv));
2386 }
2387
8d919b0a 2388 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2389 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2390 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2391 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2392 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2393 {
71c558c3 2394 UV value;
8d919b0a
FC
2395 const char * const ptr =
2396 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2397 const int numtype
8d919b0a 2398 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2399
2400 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2401 == IS_NUMBER_IN_UV) {
2402 /* It's definitely an integer */
2403 if (!(numtype & IS_NUMBER_NEG))
2404 return value;
2405 }
2406 if (!numtype) {
2407 if (ckWARN(WARN_NUMERIC))
2408 not_a_number(sv);
2409 }
8d919b0a 2410 return U_V(Atof(ptr));
e20b6c3b 2411 }
4bac9ae4
CS
2412 }
2413
2414 if (SvTHINKFIRST(sv)) {
73381561 2415#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2416 if (SvIsCOW(sv)) {
2417 sv_force_normal_flags(sv, 0);
8a818333 2418 }
73381561 2419#endif
0336b60e 2420 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2421 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2422 report_uninit(sv);
ff68c719
PP
2423 return 0;
2424 }
2425 }
4bac9ae4 2426
af359546
NC
2427 if (!SvIOKp(sv)) {
2428 if (S_sv_2iuv_common(aTHX_ sv))
2429 return 0;
ff68c719 2430 }
25da4f38 2431
1d7c1841
GS
2432 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2433 PTR2UV(sv),SvUVX(sv)));
25da4f38 2434 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2435}
2436
645c22ef 2437/*
196007d1 2438=for apidoc sv_2nv_flags
645c22ef
DM
2439
2440Return the num value of an SV, doing any necessary string or integer
fde67290 2441conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2442Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2443
2444=cut
2445*/
2446
65202027 2447NV
5aaab254 2448Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2449{
97aff369 2450 dVAR;
79072805
LW
2451 if (!sv)
2452 return 0.0;
217f6fa3
FC
2453 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2454 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2455 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2456 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2457 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2458 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2459 const char *ptr;
39d5de13
DM
2460 if (flags & SV_GMAGIC)
2461 mg_get(sv);
463ee0b2
LW
2462 if (SvNOKp(sv))
2463 return SvNVX(sv);
cd84013a 2464 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2465 ptr = SvPVX_const(sv);
2466 grokpv:
041457d9 2467 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2468 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2469 not_a_number(sv);
8d919b0a 2470 return Atof(ptr);
a0d0e21e 2471 }
25da4f38 2472 if (SvIOKp(sv)) {
1c846c1f 2473 if (SvIsUV(sv))
65202027 2474 return (NV)SvUVX(sv);
25da4f38 2475 else
65202027 2476 return (NV)SvIVX(sv);
47a72cb8
NC
2477 }
2478 if (SvROK(sv)) {
2479 goto return_rok;
2480 }
8d919b0a
FC
2481 if (isREGEXP(sv)) {
2482 ptr = RX_WRAPPED((REGEXP *)sv);
2483 goto grokpv;
2484 }
47a72cb8
NC
2485 assert(SvTYPE(sv) >= SVt_PVMG);
2486 /* This falls through to the report_uninit near the end of the
2487 function. */
2488 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2489 if (SvROK(sv)) {
47a72cb8 2490 return_rok:
deb46114 2491 if (SvAMAGIC(sv)) {
aee036bb
DM
2492 SV *tmpstr;
2493 if (flags & SV_SKIP_OVERLOAD)
2494 return 0;
31d632c3 2495 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2496 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2497 return SvNV(tmpstr);
2498 }
2499 }
2500 return PTR2NV(SvRV(sv));
a0d0e21e 2501 }
73381561 2502#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2503 if (SvIsCOW(sv)) {
2504 sv_force_normal_flags(sv, 0);
8a818333 2505 }
73381561 2506#endif
0336b60e 2507 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2508 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2509 report_uninit(sv);
ed6116ce
LW
2510 return 0.0;
2511 }
79072805
LW
2512 }
2513 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2514 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2515 sv_upgrade(sv, SVt_NV);
906f284f 2516#ifdef USE_LONG_DOUBLE
097ee67d 2517 DEBUG_c({
f93f4e46 2518 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2519 PerlIO_printf(Perl_debug_log,
2520 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2521 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2522 RESTORE_NUMERIC_LOCAL();
2523 });
65202027 2524#else
572bbb43 2525 DEBUG_c({
f93f4e46 2526 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2527 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2528 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2529 RESTORE_NUMERIC_LOCAL();
2530 });
572bbb43 2531#endif
79072805
LW
2532 }
2533 else if (SvTYPE(sv) < SVt_PVNV)
2534 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2535 if (SvNOKp(sv)) {
2536 return SvNVX(sv);
61604483 2537 }
59d8ce62 2538 if (SvIOKp(sv)) {
9d6ce603 2539 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2540#ifdef NV_PRESERVES_UV
a43d94f2
NC
2541 if (SvIOK(sv))
2542 SvNOK_on(sv);
2543 else
2544 SvNOKp_on(sv);
28e5dec8
JH
2545#else
2546 /* Only set the public NV OK flag if this NV preserves the IV */
2547 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2548 if (SvIOK(sv) &&
2549 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2550 : (SvIVX(sv) == I_V(SvNVX(sv))))
2551 SvNOK_on(sv);
2552 else
2553 SvNOKp_on(sv);
2554#endif
93a17b20 2555 }
cd84013a 2556 else if (SvPOKp(sv)) {
c2988b20 2557 UV value;
3f7c398e 2558 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2559 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2560 not_a_number(sv);
28e5dec8 2561#ifdef NV_PRESERVES_UV
c2988b20
NC
2562 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2563 == IS_NUMBER_IN_UV) {
5e045b90 2564 /* It's definitely an integer */
9d6ce603 2565 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2566 } else
3f7c398e 2567 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2568 if (numtype)
2569 SvNOK_on(sv);
2570 else
2571 SvNOKp_on(sv);
28e5dec8 2572#else
3f7c398e 2573 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2574 /* Only set the public NV OK flag if this NV preserves the value in
2575 the PV at least as well as an IV/UV would.
2576 Not sure how to do this 100% reliably. */
2577 /* if that shift count is out of range then Configure's test is
2578 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2579 UV_BITS */
2580 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2581 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2582 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2583 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2584 /* Can't use strtol etc to convert this string, so don't try.
2585 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2586 SvNOK_on(sv);
2587 } else {
2588 /* value has been set. It may not be precise. */
2589 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2590 /* 2s complement assumption for (UV)IV_MIN */
2591 SvNOK_on(sv); /* Integer is too negative. */
2592 } else {
2593 SvNOKp_on(sv);
2594 SvIOKp_on(sv);
6fa402ec 2595
c2988b20 2596 if (numtype & IS_NUMBER_NEG) {
45977657 2597 SvIV_set(sv, -(IV)value);
c2988b20 2598 } else if (value <= (UV)IV_MAX) {
45977657 2599 SvIV_set(sv, (IV)value);
c2988b20 2600 } else {
607fa7f2 2601 SvUV_set(sv, value);
c2988b20
NC
2602 SvIsUV_on(sv);
2603 }
2604
2605 if (numtype & IS_NUMBER_NOT_INT) {
2606 /* I believe that even if the original PV had decimals,
2607 they are lost beyond the limit of the FP precision.
2608 However, neither is canonical, so both only get p
2609 flags. NWC, 2000/11/25 */
2610 /* Both already have p flags, so do nothing */
2611 } else {
66a1b24b 2612 const NV nv = SvNVX(sv);
c2988b20
NC
2613 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2614 if (SvIVX(sv) == I_V(nv)) {
2615 SvNOK_on(sv);
c2988b20 2616 } else {
c2988b20
NC
2617 /* It had no "." so it must be integer. */
2618 }
00b6aa41 2619 SvIOK_on(sv);
c2988b20
NC
2620 } else {
2621 /* between IV_MAX and NV(UV_MAX).
2622 Could be slightly > UV_MAX */
6fa402ec 2623
c2988b20
NC
2624 if (numtype & IS_NUMBER_NOT_INT) {
2625 /* UV and NV both imprecise. */
2626 } else {
66a1b24b 2627 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2628
2629 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2630 SvNOK_on(sv);
c2988b20 2631 }
00b6aa41 2632 SvIOK_on(sv);
c2988b20
NC
2633 }
2634 }
2635 }
2636 }
2637 }
a43d94f2
NC
2638 /* It might be more code efficient to go through the entire logic above
2639 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2640 gets complex and potentially buggy, so more programmer efficient
2641 to do it this way, by turning off the public flags: */
2642 if (!numtype)
2643 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2644#endif /* NV_PRESERVES_UV */
93a17b20 2645 }
79072805 2646 else {
f7877b28 2647 if (isGV_with_GP(sv)) {
159b6efe 2648 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2649 return 0.0;
2650 }
2651
9a214eec 2652 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2653 report_uninit(sv);
7e25a7e9
NC
2654 assert (SvTYPE(sv) >= SVt_NV);
2655 /* Typically the caller expects that sv_any is not NULL now. */
2656 /* XXX Ilya implies that this is a bug in callers that assume this
2657 and ideally should be fixed. */
a0d0e21e 2658 return 0.0;
79072805 2659 }
572bbb43 2660#if defined(USE_LONG_DOUBLE)
097ee67d 2661 DEBUG_c({
f93f4e46 2662 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2663 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2664 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2665 RESTORE_NUMERIC_LOCAL();
2666 });
65202027 2667#else
572bbb43 2668 DEBUG_c({
f93f4e46 2669 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2670 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2671 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2672 RESTORE_NUMERIC_LOCAL();
2673 });
572bbb43 2674#endif
463ee0b2 2675 return SvNVX(sv);
79072805
LW
2676}
2677
800401ee
JH
2678/*
2679=for apidoc sv_2num
2680
2681Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2682reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2683access this function.
800401ee
JH
2684
2685=cut
2686*/
2687
2688SV *
5aaab254 2689Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2690{
7918f24d
NC
2691 PERL_ARGS_ASSERT_SV_2NUM;
2692
b9ee0594
RGS
2693 if (!SvROK(sv))
2694 return sv;
800401ee 2695 if (SvAMAGIC(sv)) {
31d632c3 2696 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2697 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2698 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2699 return sv_2num(tmpsv);
2700 }
2701 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2702}
2703
645c22ef
DM
2704/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2705 * UV as a string towards the end of buf, and return pointers to start and
2706 * end of it.
2707 *
2708 * We assume that buf is at least TYPE_CHARS(UV) long.
2709 */
2710
864dbfa3 2711static char *
5de3775c 2712S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2713{
25da4f38 2714 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2715 char * const ebuf = ptr;
25da4f38 2716 int sign;
25da4f38 2717
7918f24d
NC
2718 PERL_ARGS_ASSERT_UIV_2BUF;
2719
25da4f38
IZ
2720 if (is_uv)
2721 sign = 0;
2722 else if (iv >= 0) {
2723 uv = iv;
2724 sign = 0;
2725 } else {
2726 uv = -iv;
2727 sign = 1;
2728 }
2729 do {
eb160463 2730 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2731 } while (uv /= 10);
2732 if (sign)
2733 *--ptr = '-';
2734 *peob = ebuf;
2735 return ptr;
2736}
2737
645c22ef
DM
2738/*
2739=for apidoc sv_2pv_flags
2740
ff276b08 2741Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2742If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2743string if necessary. Normally invoked via the C<SvPV_flags> macro.
2744C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2745
2746=cut
2747*/
2748
8d6d96c1 2749char *
5aaab254 2750Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2751{
97aff369 2752 dVAR;
eb578fdb 2753 char *s;
79072805 2754
463ee0b2 2755 if (!sv) {
cdb061a3
NC
2756 if (lp)
2757 *lp = 0;
73d840c0 2758 return (char *)"";
463ee0b2 2759 }
217f6fa3
FC
2760 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2761 && SvTYPE(sv) != SVt_PVFM);
4bac9ae4
CS
2762 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2763 mg_get(sv);
2764 if (SvROK(sv)) {
2765 if (SvAMAGIC(sv)) {
2766 SV *tmpstr;
2767 if (flags & SV_SKIP_OVERLOAD)
2768 return NULL;
2769 tmpstr = AMG_CALLunary(sv, string_amg);
2770 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2771 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2772 /* Unwrap this: */
2773 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2774 */
75dfc8ec 2775
4bac9ae4
CS
2776 char *pv;
2777 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2778 if (flags & SV_CONST_RETURN) {
2779 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2780 } else {
4bac9ae4
CS
2781 pv = (flags & SV_MUTABLE_RETURN)
2782 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2783 }
4bac9ae4
CS
2784 if (lp)
2785 *lp = SvCUR(tmpstr);
2786 } else {
2787 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2788 }
4bac9ae4
CS
2789 if (SvUTF8(tmpstr))
2790 SvUTF8_on(sv);
2791 else
2792 SvUTF8_off(sv);
2793 return pv;
deb46114 2794 }
4bac9ae4
CS
2795 }
2796 {
2797 STRLEN len;
2798 char *retval;
2799 char *buffer;
2800 SV *const referent = SvRV(sv);
2801
2802 if (!referent) {
2803 len = 7;
2804 retval = buffer = savepvn("NULLREF", len);
2805 } else if (SvTYPE(referent) == SVt_REGEXP &&
2806 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2807 amagic_is_enabled(string_amg))) {
2808 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2809
2810 assert(re);
67d2d14d 2811
4bac9ae4
CS
2812 /* If the regex is UTF-8 we want the containing scalar to
2813 have an UTF-8 flag too */
2814 if (RX_UTF8(re))
2815 SvUTF8_on(sv);
2816 else
2817 SvUTF8_off(sv);
67d2d14d 2818
4bac9ae4
CS
2819 if (lp)
2820 *lp = RX_WRAPLEN(re);
67d2d14d 2821
4bac9ae4
CS
2822 return RX_WRAPPED(re);
2823 } else {
2824 const char *const typestr = sv_reftype(referent, 0);
2825 const STRLEN typelen = strlen(typestr);
2826 UV addr = PTR2UV(referent);
2827 const char *stashname = NULL;
2828 STRLEN stashnamelen = 0; /* hush, gcc */
2829 const char *buffer_end;
2830
2831 if (SvOBJECT(referent)) {
2832 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2833
2834 if (name) {
2835 stashname = HEK_KEY(name);
2836 stashnamelen = HEK_LEN(name);
2837
2838 if (HEK_UTF8(name)) {
2839 SvUTF8_on(sv);
fafee734 2840 } else {
4bac9ae4 2841 SvUTF8_off(sv);
fafee734 2842 }
fafee734 2843 } else {
4bac9ae4
CS
2844 stashname = "__ANON__";
2845 stashnamelen = 8;
fafee734 2846 }
4bac9ae4
CS
2847 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2848 + 2 * sizeof(UV) + 2 /* )\0 */;
2849 } else {
2850 len = typelen + 3 /* (0x */
2851 + 2 * sizeof(UV) + 2 /* )\0 */;
2852 }
fafee734 2853
4bac9ae4
CS
2854 Newx(buffer, len, char);
2855 buffer_end = retval = buffer + len;
2856
2857 /* Working backwards */
2858 *--retval = '\0';
2859 *--retval = ')';
2860 do {
2861 *--retval = PL_hexdigit[addr & 15];
2862 } while (addr >>= 4);
2863 *--retval = 'x';
2864 *--retval = '0';
2865 *--retval = '(';
2866
2867 retval -= typelen;
2868 memcpy(retval, typestr, typelen);
2869
2870 if (stashname) {
2871 *--retval = '=';
2872 retval -= stashnamelen;
2873 memcpy(retval, stashname, stashnamelen);
c080367d 2874 }
4bac9ae4
CS
2875 /* retval may not necessarily have reached the start of the
2876 buffer here. */
2877 assert (retval >= buffer);
2878
2879 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 2880 }
cdb061a3 2881 if (lp)
4bac9ae4
CS
2882 *lp = len;
2883 SAVEFREEPV(buffer);
2884 return retval;
79072805 2885 }
79072805 2886 }
4bac9ae4
CS
2887
2888 if (SvPOKp(sv)) {
2889 if (lp)
2890 *lp = SvCUR(sv);
2891 if (flags & SV_MUTABLE_RETURN)
2892 return SvPVX_mutable(sv);
2893 if (flags & SV_CONST_RETURN)
2894 return (char *)SvPVX_const(sv);
2895 return SvPVX(sv);
2896 }
2897
2898 if (SvIOK(sv)) {
28e5dec8
JH
2899 /* I'm assuming that if both IV and NV are equally valid then
2900 converting the IV is going to be more efficient */
e1ec3a88 2901 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2902 char buf[TYPE_CHARS(UV)];
2903 char *ebuf, *ptr;
97a130b8 2904 STRLEN len;
28e5dec8
JH
2905
2906 if (SvTYPE(sv) < SVt_PVIV)
2907 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2908 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2909 len = ebuf - ptr;
5902b6a9 2910 /* inlined from sv_setpvn */
97a130b8
NC
2911 s = SvGROW_mutable(sv, len + 1);
2912 Move(ptr, s, len, char);
2913 s += len;
28e5dec8 2914 *s = '\0';
b127e37e 2915 SvPOK_on(sv);
28e5dec8 2916 }
4bac9ae4 2917 else if (SvNOK(sv)) {
79072805
LW
2918 if (SvTYPE(sv) < SVt_PVNV)
2919 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2920 if (SvNVX(sv) == 0.0) {
2921 s = SvGROW_mutable(sv, 2);
2922 *s++ = '0';
2923 *s = '\0';
2924 } else {
2925 dSAVE_ERRNO;
2926 /* The +20 is pure guesswork. Configure test needed. --jhi */
2927 s = SvGROW_mutable(sv, NV_DIG + 20);
2928 /* some Xenix systems wipe out errno here */
b127e37e 2929
b127e37e 2930#ifndef USE_LOCALE_NUMERIC
68e8f474 2931 Gconvert(SvNVX(sv), NV_DIG, 0, s);
b127e37e 2932 SvPOK_on(sv);
68e8f474
KW
2933#else
2934 /* Gconvert always uses the current locale. That's the right thing
2935 * to do if we're supposed to be using locales. But otherwise, we
2936 * want the result to be based on the C locale, so we need to
2937 * change to the C locale during the Gconvert and then change back.
2938 * But if we're already in the C locale (PL_numeric_standard is
2939 * TRUE in that case), no need to do any changing */
7187d38e 2940 if (PL_numeric_standard || IN_SOME_LOCALE_FORM_RUNTIME) {
68e8f474 2941 Gconvert(SvNVX(sv), NV_DIG, 0, s);
28acfe03
KW
2942
2943 /* If the radix character is UTF-8, and actually is in the
2944 * output, turn on the UTF-8 flag for the scalar */
2945 if (! PL_numeric_standard
2946 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
2947 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
2948 {
2949 SvUTF8_on(sv);
2950 }
68e8f474
KW
2951 }
2952 else {
2953 char *loc = savepv(setlocale(LC_NUMERIC, NULL));
2954 setlocale(LC_NUMERIC, "C");
2955 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2956 setlocale(LC_NUMERIC, loc);
2957 Safefree(loc);
28acfe03 2958
68e8f474
KW
2959 }
2960
2961 /* We don't call SvPOK_on(), because it may come to pass that the
2962 * locale changes so that the stringification we just did is no
2963 * longer correct. We will have to re-stringify every time it is
2964 * needed */
b127e37e 2965#endif
29912d93
NC
2966 RESTORE_ERRNO;
2967 while (*s) s++;
bbce6d69 2968 }
79072805
LW
2969#ifdef hcx
2970 if (s[-1] == '.')
46fc3d4c 2971 *--s = '\0';
79072805
LW
2972#endif
2973 }
4bac9ae4
CS
2974 else if (isGV_with_GP(sv)) {
2975 GV *const gv = MUTABLE_GV(sv);
2976 SV *const buffer = sv_newmortal();
8d1c3e26 2977
4bac9ae4 2978 gv_efullname3(buffer, gv, "*");
180488f8 2979
4bac9ae4
CS
2980 assert(SvPOK(buffer));
2981 if (SvUTF8(buffer))
2982 SvUTF8_on(sv);
2983 if (lp)
2984 *lp = SvCUR(buffer);
2985 return SvPVX(buffer);
2986 }
8d919b0a
FC
2987 else if (isREGEXP(sv)) {
2988 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2989 return RX_WRAPPED((REGEXP *)sv);
2990 }
4bac9ae4 2991 else {
cdb061a3 2992 if (lp)
00b6aa41 2993 *lp = 0;
9f621bb0
NC
2994 if (flags & SV_UNDEF_RETURNS_NULL)
2995 return NULL;
9a214eec 2996 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2997 report_uninit(sv);
4bac9ae4
CS
2998 /* Typically the caller expects that sv_any is not NULL now. */
2999 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3000 sv_upgrade(sv, SVt_PV);
73d840c0 3001 return (char *)"";
79072805 3002 }
4bac9ae4 3003
cdb061a3 3004 {
823a54a3 3005 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3006 if (lp)
3007 *lp = len;
3008 SvCUR_set(sv, len);
3009 }
1d7c1841 3010 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3011 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3012 if (flags & SV_CONST_RETURN)
3013 return (char *)SvPVX_const(sv);
10516c54
NC
3014 if (flags & SV_MUTABLE_RETURN)
3015 return SvPVX_mutable(sv);
463ee0b2
LW
3016 return SvPVX(sv);
3017}
3018
645c22ef 3019/*
6050d10e
JP
3020=for apidoc sv_copypv
3021
3022Copies a stringified representation of the source SV into the
3023destination SV. Automatically performs any necessary mg_get and
54f0641b 3024coercion of numeric values into strings. Guaranteed to preserve
2575c402 3025UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3026sv_2pv[_flags] but operates directly on an SV instead of just the
3027string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3028would lose the UTF-8'ness of the PV.
3029
4bac9ae4
CS
3030=for apidoc sv_copypv_nomg
3031
3032Like sv_copypv, but doesn't invoke get magic first.
3033
3034=for apidoc sv_copypv_flags
3035
3036Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3037include SV_GMAGIC.
3038
6050d10e
JP
3039=cut
3040*/
3041
3042void
5aaab254 3043Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
6050d10e 3044{
4bac9ae4
CS
3045 PERL_ARGS_ASSERT_SV_COPYPV;
3046
3047 sv_copypv_flags(dsv, ssv, 0);
3048}
3049
3050void
5aaab254 3051Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3052{
446eaa42 3053 STRLEN len;
4bac9ae4 3054 const char *s;
7918f24d 3055
4bac9ae4 3056 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3057
4bac9ae4
CS
3058 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3059 mg_get(ssv);
3060 s = SvPV_nomg_const(ssv,len);
cb50f42d 3061 sv_setpvn(dsv,s,len);
446eaa42 3062 if (SvUTF8(ssv))
cb50f42d 3063 SvUTF8_on(dsv);
446eaa42 3064 else
cb50f42d 3065 SvUTF8_off(dsv);
6050d10e
JP
3066}
3067
3068/*
645c22ef
DM
3069=for apidoc sv_2pvbyte
3070
3071Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3072to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3073side-effect.
3074
3075Usually accessed via the C<SvPVbyte> macro.
3076
3077=cut
3078*/
3079
7340a771 3080char *
5aaab254 3081Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3082{
7918f24d
NC
3083 PERL_ARGS_ASSERT_SV_2PVBYTE;
3084
48120f8f 3085 SvGETMAGIC(sv);
4499db73
FC
3086 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3087 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3088 SV *sv2 = sv_newmortal();
48120f8f 3089 sv_copypv_nomg(sv2,sv);
a901b181
FC
3090 sv = sv2;
3091 }
0875d2fe 3092 sv_utf8_downgrade(sv,0);
71eb6d8c 3093 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3094}
3095
645c22ef 3096/*
035cbb0e
RGS
3097=for apidoc sv_2pvutf8
3098
3099Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3100to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3101
3102Usually accessed via the C<SvPVutf8> macro.
3103
3104=cut
3105*/
645c22ef 3106
7340a771 3107char *
5aaab254 3108Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3109{
7918f24d
NC
3110 PERL_ARGS_ASSERT_SV_2PVUTF8;
3111
4499db73
FC
3112 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3113 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3114 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3115 else
3116 SvGETMAGIC(sv);
3117 sv_utf8_upgrade_nomg(sv);
c3ec315f 3118 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3119}
1c846c1f 3120
7ee2227d 3121
645c22ef
DM
3122/*
3123=for apidoc sv_2bool
3124
06c841cf
FC
3125This macro is only used by sv_true() or its macro equivalent, and only if
3126the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3127It calls sv_2bool_flags with the SV_GMAGIC flag.
3128
3129=for apidoc sv_2bool_flags
3130
3131This function is only used by sv_true() and friends, and only if
fde67290 3132the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3133contain SV_GMAGIC, then it does an mg_get() first.
3134
645c22ef
DM
3135
3136=cut
3137*/
3138
463ee0b2 3139bool
5aaab254 3140Perl_sv_2bool_flags(pTHX_ SV *const sv, const I32 flags)
463ee0b2 3141{
97aff369 3142 dVAR;
7918f24d 3143
06c841cf 3144 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3145
06c841cf 3146 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3147
a0d0e21e
LW
3148 if (!SvOK(sv))
3149 return 0;
3150 if (SvROK(sv)) {
fabdb6c0 3151 if (SvAMAGIC(sv)) {
31d632c3 3152 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3153 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3154 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3155 }
3156 return SvRV(sv) != 0;
a0d0e21e 3157 }
85b7d9b3
FC
3158 if (isREGEXP(sv))
3159 return
3160 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
4bac9ae4 3161 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3162}
3163
c461cf8f
JH
3164/*
3165=for apidoc sv_utf8_upgrade
3166
78ea37eb 3167Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3168Forces the SV to string form if it is not already.
2bbc8d55 3169Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3170Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3171if the whole string is the same in UTF-8 as not.
3172Returns the number of bytes in the converted string
c461cf8f 3173
0efd0472 3174This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3175use the Encode extension for that.
3176
fe749c9a
KW
3177=for apidoc sv_utf8_upgrade_nomg
3178
fde67290 3179Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3180
8d6d96c1
HS
3181=for apidoc sv_utf8_upgrade_flags
3182
78ea37eb 3183Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3184Forces the SV to string form if it is not already.
8d6d96c1 3185Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3186if all the bytes are invariant in UTF-8.
3187If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55
SP
3188will C<mg_get> on C<sv> if appropriate, else not.
3189Returns the number of bytes in the converted string
3190C<sv_utf8_upgrade> and
8d6d96c1
HS
3191C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3192
0efd0472 3193This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3194use the Encode extension for that.
3195
8d6d96c1 3196=cut
b3ab6785
KW
3197
3198The grow version is currently not externally documented. It adds a parameter,
3199extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3200have free after it upon return. This allows the caller to reserve extra space
3201that it intends to fill, to avoid extra grows.
3202
3203Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3204which can be used to tell this function to not first check to see if there are
3205any characters that are different in UTF-8 (variant characters) which would
3206force it to allocate a new string to sv, but to assume there are. Typically
3207this flag is used by a routine that has already parsed the string to find that
3208there are such characters, and passes this information on so that the work
3209doesn't have to be repeated.
3210
3211(One might think that the calling routine could pass in the position of the
3212first such variant, so it wouldn't have to be found again. But that is not the
3213case, because typically when the caller is likely to use this flag, it won't be
3214calling this routine unless it finds something that won't fit into a byte.
3215Otherwise it tries to not upgrade and just use bytes. But some things that
3216do fit into a byte are variants in utf8, and the caller may not have been
3217keeping track of these.)
3218
3219If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3220isn't guaranteed due to having other routines do the work in some input cases,
3221or if the input is already flagged as being in utf8.
3222
3223The speed of this could perhaps be improved for many cases if someone wanted to
3224write a fast function that counts the number of variant characters in a string,
3225especially if it could return the position of the first one.
3226
8d6d96c1
HS
3227*/
3228
c56ed9f6
FC
3229static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
3230
8d6d96c1 3231STRLEN
5aaab254 3232Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3233{
97aff369 3234 dVAR;
7918f24d 3235
b3ab6785 3236 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3237
808c356f
RGS
3238 if (sv == &PL_sv_undef)
3239 return 0;
892f9127 3240 if (!SvPOK_nog(sv)) {
e0e62c2a 3241 STRLEN len = 0;
d52b7888
NC
3242 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3243 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3244 if (SvUTF8(sv)) {
3245 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3246 return len;
b3ab6785 3247 }
d52b7888 3248 } else {
33fb6f35 3249 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3250 }
e0e62c2a 3251 }
4411f3b6 3252
f5cee72b 3253 if (SvUTF8(sv)) {
b3ab6785 3254 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3255 return SvCUR(sv);
f5cee72b 3256 }
5fec3b1d 3257
765f542d 3258 if (SvIsCOW(sv)) {
c56ed9f6 3259 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3260 }
3261
b3ab6785 3262 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3263 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3264 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3265 return SvCUR(sv);
3266 }
3267
4e93345f
KW
3268 if (SvCUR(sv) == 0) {
3269 if (extra) SvGROW(sv, extra);
3270 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3271 /* This function could be much more efficient if we
2bbc8d55 3272 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3273 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3274 * make the loop as fast as possible (although there are certainly ways
3275 * to speed this up, eg. through vectorization) */
3276 U8 * s = (U8 *) SvPVX_const(sv);
3277 U8 * e = (U8 *) SvEND(sv);
3278 U8 *t = s;
3279 STRLEN two_byte_count = 0;
c4e7c712 3280
b3ab6785
KW
3281 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3282
3283 /* See if really will need to convert to utf8. We mustn't rely on our
3284 * incoming SV being well formed and having a trailing '\0', as certain
3285 * code in pp_formline can send us partially built SVs. */
3286
c4e7c712 3287 while (t < e) {
53c1dcc0 3288 const U8 ch = *t++;
b3ab6785
KW
3289 if (NATIVE_IS_INVARIANT(ch)) continue;
3290
3291 t--; /* t already incremented; re-point to first variant */
3292 two_byte_count = 1;
3293 goto must_be_utf8;
c4e7c712 3294 }
b3ab6785
KW
3295
3296 /* utf8 conversion not needed because all are invariants. Mark as
3297 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3298 SvUTF8_on(sv);
7f0bfbea 3299 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3300 return SvCUR(sv);
3301
3302must_be_utf8:
3303
3304 /* Here, the string should be converted to utf8, either because of an
3305 * input flag (two_byte_count = 0), or because a character that
3306 * requires 2 bytes was found (two_byte_count = 1). t points either to
3307 * the beginning of the string (if we didn't examine anything), or to
3308 * the first variant. In either case, everything from s to t - 1 will
3309 * occupy only 1 byte each on output.
3310 *
3311 * There are two main ways to convert. One is to create a new string
3312 * and go through the input starting from the beginning, appending each
3313 * converted value onto the new string as we go along. It's probably
3314 * best to allocate enough space in the string for the worst possible
3315 * case rather than possibly running out of space and having to
3316 * reallocate and then copy what we've done so far. Since everything
3317 * from s to t - 1 is invariant, the destination can be initialized
3318 * with these using a fast memory copy
3319 *
3320 * The other way is to figure out exactly how big the string should be
3321 * by parsing the entire input. Then you don't have to make it big
3322 * enough to handle the worst possible case, and more importantly, if
3323 * the string you already have is large enough, you don't have to
3324 * allocate a new string, you can copy the last character in the input
3325 * string to the final position(s) that will be occupied by the
3326 * converted string and go backwards, stopping at t, since everything
3327 * before that is invariant.
3328 *
3329 * There are advantages and disadvantages to each method.
3330 *
3331 * In the first method, we can allocate a new string, do the memory
3332 * copy from the s to t - 1, and then proceed through the rest of the
3333 * string byte-by-byte.
3334 *
3335 * In the second method, we proceed through the rest of the input
3336 * string just calculating how big the converted string will be. Then
3337 * there are two cases:
3338 * 1) if the string has enough extra space to handle the converted
3339 * value. We go backwards through the string, converting until we
3340 * get to the position we are at now, and then stop. If this
3341 * position is far enough along in the string, this method is
3342 * faster than the other method. If the memory copy were the same
3343 * speed as the byte-by-byte loop, that position would be about
3344 * half-way, as at the half-way mark, parsing to the end and back
3345 * is one complete string's parse, the same amount as starting
3346 * over and going all the way through. Actually, it would be
3347 * somewhat less than half-way, as it's faster to just count bytes
3348 * than to also copy, and we don't have the overhead of allocating
3349 * a new string, changing the scalar to use it, and freeing the
3350 * existing one. But if the memory copy is fast, the break-even
3351 * point is somewhere after half way. The counting loop could be
3352 * sped up by vectorization, etc, to move the break-even point
3353 * further towards the beginning.
3354 * 2) if the string doesn't have enough space to handle the converted
3355 * value. A new string will have to be allocated, and one might
3356 * as well, given that, start from the beginning doing the first
3357 * method. We've spent extra time parsing the string and in
3358 * exchange all we've gotten is that we know precisely how big to
3359 * make the new one. Perl is more optimized for time than space,
3360 * so this case is a loser.
3361 * So what I've decided to do is not use the 2nd method unless it is
3362 * guaranteed that a new string won't have to be allocated, assuming
3363 * the worst case. I also decided not to put any more conditions on it
3364 * than this, for now. It seems likely that, since the worst case is
3365 * twice as big as the unknown portion of the string (plus 1), we won't
3366 * be guaranteed enough space, causing us to go to the first method,
3367 * unless the string is short, or the first variant character is near
3368 * the end of it. In either of these cases, it seems best to use the
3369 * 2nd method. The only circumstance I can think of where this would
3370 * be really slower is if the string had once had much more data in it
3371 * than it does now, but there is still a substantial amount in it */
3372
3373 {
3374 STRLEN invariant_head = t - s;
3375 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3376 if (SvLEN(sv) < size) {
3377
3378 /* Here, have decided to allocate a new string */
3379
3380 U8 *dst;
3381 U8 *d;
3382
3383 Newx(dst, size, U8);
3384
3385 /* If no known invariants at the beginning of the input string,
3386 * set so starts from there. Otherwise, can use memory copy to
3387 * get up to where we are now, and then start from here */
3388
3389 if (invariant_head <= 0) {
3390 d = dst;
3391 } else {
3392 Copy(s, dst, invariant_head, char);
3393 d = dst + invariant_head;
3394 }
3395
3396 while (t < e) {
3397 const UV uv = NATIVE8_TO_UNI(*t++);
3398 if (UNI_IS_INVARIANT(uv))
3399 *d++ = (U8)UNI_TO_NATIVE(uv);
3400 else {
3401 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3402 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3403 }
3404 }
3405 *d = '\0';
3406 SvPV_free(sv); /* No longer using pre-existing string */
3407 SvPV_set(sv, (char*)dst);
3408 SvCUR_set(sv, d - dst);
3409 SvLEN_set(sv, size);
3410 } else {
3411
3412 /* Here, have decided to get the exact size of the string.
3413 * Currently this happens only when we know that there is
3414 * guaranteed enough space to fit the converted string, so
3415 * don't have to worry about growing. If two_byte_count is 0,
3416 * then t points to the first byte of the string which hasn't
3417 * been examined yet. Otherwise two_byte_count is 1, and t
3418 * points to the first byte in the string that will expand to
3419 * two. Depending on this, start examining at t or 1 after t.
3420 * */
3421
3422 U8 *d = t + two_byte_count;
3423
3424
3425 /* Count up the remaining bytes that expand to two */
3426
3427 while (d < e) {
3428 const U8 chr = *d++;
3429 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3430 }
3431
3432 /* The string will expand by just the number of bytes that
3433 * occupy two positions. But we are one afterwards because of
3434 * the increment just above. This is the place to put the
3435 * trailing NUL, and to set the length before we decrement */
3436
3437 d += two_byte_count;
3438 SvCUR_set(sv, d - s);
3439 *d-- = '\0';
3440
3441
3442 /* Having decremented d, it points to the position to put the
3443 * very last byte of the expanded string. Go backwards through
3444 * the string, copying and expanding as we go, stopping when we
3445 * get to the part that is invariant the rest of the way down */
3446
3447 e--;
3448 while (e >= t) {
3449 const U8 ch = NATIVE8_TO_UNI(*e--);
3450 if (UNI_IS_INVARIANT(ch)) {
3451 *d-- = UNI_TO_NATIVE(ch);
3452 } else {
3453 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3454 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3455 }
3456 }
3457 }
75da9d4c
DM
3458
3459 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3460 /* Update pos. We do it at the end rather than during
3461 * the upgrade, to avoid slowing down the common case
3462 * (upgrade without pos) */
3463 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3464 if (mg) {
3465 I32 pos = mg->mg_len;
3466 if (pos > 0 && (U32)pos > invariant_head) {
3467 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3468 STRLEN n = (U32)pos - invariant_head;
3469 while (n > 0) {
3470 if (UTF8_IS_START(*d))
3471 d++;
3472 d++;
3473 n--;
3474 }
3475 mg->mg_len = d - (U8*)SvPVX(sv);
3476 }
3477 }
3478 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3479 magic_setutf8(sv,mg); /* clear UTF8 cache */
3480 }
b3ab6785 3481 }
560a288e 3482 }
b3ab6785
KW
3483
3484 /* Mark as UTF-8 even if no variant - saves scanning loop */
3485 SvUTF8_on(sv);
4411f3b6 3486 return SvCUR(sv);
560a288e
GS
3487}
3488
c461cf8f
JH
3489/*
3490=for apidoc sv_utf8_downgrade
3491
78ea37eb 3492Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3493If the PV contains a character that cannot fit
3494in a byte, this conversion will fail;
78ea37eb 3495in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3496true, croaks.
3497
0efd0472 3498This is not a general purpose Unicode to byte encoding interface:
13a6c0e0
JH
3499use the Encode extension for that.
3500
c461cf8f
JH
3501=cut
3502*/
3503
560a288e 3504bool
5aaab254 3505Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
560a288e 3506{
97aff369 3507 dVAR;
7918f24d
NC
3508
3509 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3510
78ea37eb 3511 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3512 if (SvCUR(sv)) {
03cfe0ae 3513 U8 *s;
652088fc 3514 STRLEN len;
75da9d4c 3515 int mg_flags = SV_GMAGIC;
fa301091 3516
765f542d 3517 if (SvIsCOW(sv)) {
c56ed9f6 3518 S_sv_uncow(aTHX_ sv, 0);
765f542d 3519 }
75da9d4c
DM
3520 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3521 /* update pos */
3522 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3523 if (mg) {
3524 I32 pos = mg->mg_len;
3525 if (pos > 0) {
3526 sv_pos_b2u(sv, &pos);
3527 mg_flags = 0; /* sv_pos_b2u does get magic */
3528 mg->mg_len = pos;
3529 }
3530 }
3531 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3532 magic_setutf8(sv,mg); /* clear UTF8 cache */
3533
3534 }
3535 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3536
03cfe0ae 3537 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3538 if (fail_ok)
3539 return FALSE;
3540 else {
3541 if (PL_op)
3542 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3543 OP_DESC(PL_op));
fa301091
JH
3544 else
3545 Perl_croak(aTHX_ "Wide character");
3546 }
4b3603a4 3547 }
b162af07 3548 SvCUR_set(sv, len);
67e989fb 3549 }
560a288e 3550 }
ffebcc3e 3551 SvUTF8_off(sv);
560a288e
GS
3552 return TRUE;
3553}
3554
c461cf8f
JH
3555/*
3556=for apidoc sv_utf8_encode
3557
78ea37eb
ST
3558Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3559flag off so that it looks like octets again.
c461cf8f
JH
3560
3561=cut
3562*/
3563
560a288e 3564void
5aaab254 3565Perl_sv_utf8_encode(pTHX_ SV *const sv)
560a288e 3566{
7918f24d
NC
3567 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3568
4c94c214 3569 if (SvREADONLY(sv)) {
654c723f 3570 sv_force_normal_flags(sv, 0);
4c94c214 3571 }
a5f5288a 3572 (void) sv_utf8_upgrade(sv);
560a288e
GS
3573 SvUTF8_off(sv);
3574}
3575
4411f3b6
NIS
3576/*
3577=for apidoc sv_utf8_decode
3578
78ea37eb
ST
3579If the PV of the SV is an octet sequence in UTF-8
3580and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3581so that it looks like a character. If the PV contains only single-byte
694cf0d2 3582characters, the C<SvUTF8> flag stays off.
78ea37eb 3583Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3584
3585=cut
3586*/
3587
560a288e 3588bool
5aaab254 3589Perl_sv_utf8_decode(pTHX_ SV *const sv)
560a288e 3590{
7918f24d
NC
3591 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3592
78ea37eb 3593 if (SvPOKp(sv)) {
75da9d4c 3594 const U8 *start, *c;
93524f2b 3595 const U8 *e;
9cbac4c7 3596
645c22ef
DM
3597 /* The octets may have got themselves encoded - get them back as
3598 * bytes
3599 */
3600 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3601 return FALSE;
3602
3603 /* it is actually just a matter of turning the utf8 flag on, but
3604 * we want to make sure everything inside is valid utf8 first.
3605 */
75da9d4c 3606 c = start = (const U8 *) SvPVX_const(sv);
f9cf141a 3607 if (!is_utf8_string(c, SvCUR(sv)))
67e989fb 3608 return FALSE;
93524f2b 3609 e = (const U8 *) SvEND(sv);
511c2ff0 3610 while (c < e) {
b64e5050 3611 const U8 ch = *c++;
c4d5f83a 3612 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3613 SvUTF8_on(sv);
3614 break;
3615 }
560a288e 3616 }
75da9d4c
DM
3617 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3618 /* adjust pos to the start of a UTF8 char sequence */
3619 MAGIC * mg = mg_find(sv, P