This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Revert "Use new Svt_INVLIST for inversion lists."
[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
2e0b8fbe
KW
901 /* The invlist placeholder pretends to be an RV for now.
902 Also it's marked as "can't upgrade" to stop anyone using it before it's
903 implemented. */
904 { 0, 0, 0, SVt_INVLIST, TRUE, NONV, NOARENA, 0 },
e045dbed 905
bc337e5c 906 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
907 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
908 + STRUCT_OFFSET(XPV, xpv_cur),
909 SVt_PVIV, FALSE, NONV, HASARENA,
910 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 911
bc337e5c 912 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
913 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
914 + STRUCT_OFFSET(XPV, xpv_cur),
915 SVt_PVNV, FALSE, HADNV, HASARENA,
916 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 917
6e128786 918 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 919 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 920
601dfd0a
NC
921 { sizeof(regexp),
922 sizeof(regexp),
923 0,
ecff11eb 924 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 925 FIT_ARENA(0, sizeof(regexp))
5c35adbb 926 },
4df7f6af 927
10666ae3 928 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
929 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
930
10666ae3 931 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
932 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
933
601dfd0a 934 { sizeof(XPVAV),
4f7003f5 935 copy_length(XPVAV, xav_alloc),
601dfd0a 936 0,
69ba284b 937 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 938 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 939
601dfd0a 940 { sizeof(XPVHV),
359164a0 941 copy_length(XPVHV, xhv_max),
601dfd0a 942 0,
69ba284b 943 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 944 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 945
601dfd0a
NC
946 { sizeof(XPVCV),
947 sizeof(XPVCV),
948 0,
69ba284b 949 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 950 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 951
601dfd0a
NC
952 { sizeof(XPVFM),
953 sizeof(XPVFM),
954 0,
69ba284b 955 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 956 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 957
601dfd0a
NC
958 { sizeof(XPVIO),
959 sizeof(XPVIO),
960 0,
b6f60916 961 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 962 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 963};
29489e7c 964
bd81e77b 965#define new_body_allocated(sv_type) \
d2a0f284 966 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 967 - bodies_by_type[sv_type].offset)
29489e7c 968
26359cfa
NC
969/* return a thing to the free list */
970
971#define del_body(thing, root) \
972 STMT_START { \
973 void ** const thing_copy = (void **)thing; \
974 *thing_copy = *root; \
975 *root = (void*)thing_copy; \
976 } STMT_END
29489e7c 977
bd81e77b 978#ifdef PURIFY
29489e7c 979
beeec492
NC
980#define new_XNV() safemalloc(sizeof(XPVNV))
981#define new_XPVNV() safemalloc(sizeof(XPVNV))
982#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 983
beeec492 984#define del_XPVGV(p) safefree(p)
29489e7c 985
bd81e77b 986#else /* !PURIFY */
29489e7c 987
65ac1738 988#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 989#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 990#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 991
26359cfa
NC
992#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
993 &PL_body_roots[SVt_PVGV])
1d7c1841 994
bd81e77b 995#endif /* PURIFY */
93e68bfb 996
bd81e77b 997/* no arena for you! */
93e68bfb 998
bd81e77b 999#define new_NOARENA(details) \
beeec492 1000 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1001#define new_NOARENAZ(details) \
beeec492 1002 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1003
1e30fcd5
NC
1004void *
1005Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1006 const size_t arena_size)
d2a0f284
JC
1007{
1008 dVAR;
1009 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1010 struct arena_desc *adesc;
1011 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1012 unsigned int curr;
d2a0f284
JC
1013 char *start;
1014 const char *end;
02982131 1015 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1016#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1017 static bool done_sanity_check;
1018
0b2d3faa
JH
1019 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1020 * variables like done_sanity_check. */
10666ae3 1021 if (!done_sanity_check) {
ea471437 1022 unsigned int i = SVt_LAST;
10666ae3
NC
1023
1024 done_sanity_check = TRUE;
1025
1026 while (i--)
1027 assert (bodies_by_type[i].type == i);
1028 }
1029#endif
1030
02982131 1031 assert(arena_size);
23e9d66c 1032
99816f8d
NC
1033 /* may need new arena-set to hold new arena */
1034 if (!aroot || aroot->curr >= aroot->set_size) {
1035 struct arena_set *newroot;
1036 Newxz(newroot, 1, struct arena_set);
1037 newroot->set_size = ARENAS_PER_SET;
1038 newroot->next = aroot;
1039 aroot = newroot;
1040 PL_body_arenas = (void *) newroot;
1041 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1042 }
1043
1044 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1045 curr = aroot->curr++;
1046 adesc = &(aroot->set[curr]);
1047 assert(!adesc->arena);
1048
1049 Newx(adesc->arena, good_arena_size, char);
1050 adesc->size = good_arena_size;
1051 adesc->utype = sv_type;
1052 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1053 curr, (void*)adesc->arena, (UV)good_arena_size));
1054
1055 start = (char *) adesc->arena;
d2a0f284 1056
29657bb6
NC
1057 /* Get the address of the byte after the end of the last body we can fit.
1058 Remember, this is integer division: */
02982131 1059 end = start + good_arena_size / body_size * body_size;
d2a0f284 1060
486ec47a 1061 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1062#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1063 DEBUG_m(PerlIO_printf(Perl_debug_log,
1064 "arena %p end %p arena-size %d (from %d) type %d "
1065 "size %d ct %d\n",
02982131
NC
1066 (void*)start, (void*)end, (int)good_arena_size,
1067 (int)arena_size, sv_type, (int)body_size,
1068 (int)good_arena_size / (int)body_size));
d8fca402 1069#else
d2a0f284
JC
1070 DEBUG_m(PerlIO_printf(Perl_debug_log,
1071 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1072 (void*)start, (void*)end,
02982131
NC
1073 (int)arena_size, sv_type, (int)body_size,
1074 (int)good_arena_size / (int)body_size));
d8fca402 1075#endif
d2a0f284
JC
1076 *root = (void *)start;
1077
29657bb6
NC
1078 while (1) {
1079 /* Where the next body would start: */
d2a0f284 1080 char * const next = start + body_size;
29657bb6
NC
1081
1082 if (next >= end) {
1083 /* This is the last body: */
1084 assert(next == end);
1085
1086 *(void **)start = 0;
1087 return *root;
1088 }
1089
d2a0f284
JC
1090 *(void**) start = (void *)next;
1091 start = next;
1092 }
d2a0f284
JC
1093}
1094
1095/* grab a new thing from the free list, allocating more if necessary.
1096 The inline version is used for speed in hot routines, and the
1097 function using it serves the rest (unless PURIFY).
1098*/
1099#define new_body_inline(xpv, sv_type) \
1100 STMT_START { \
1101 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1102 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1103 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1104 bodies_by_type[sv_type].body_size,\
1105 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1106 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1107 } STMT_END
1108
1109#ifndef PURIFY
1110
1111STATIC void *
de37a194 1112S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1113{
1114 dVAR;
1115 void *xpv;
1116 new_body_inline(xpv, sv_type);
1117 return xpv;
1118}
1119
1120#endif
93e68bfb 1121
238b27b3
NC
1122static const struct body_details fake_rv =
1123 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1124
bd81e77b
NC
1125/*
1126=for apidoc sv_upgrade
93e68bfb 1127
bd81e77b
NC
1128Upgrade an SV to a more complex form. Generally adds a new body type to the
1129SV, then copies across as much information as possible from the old body.
9521ca61
FC
1130It croaks if the SV is already in a more complex form than requested. You
1131generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1132before calling C<sv_upgrade>, and hence does not croak. See also
1133C<svtype>.
93e68bfb 1134
bd81e77b 1135=cut
93e68bfb 1136*/
93e68bfb 1137
bd81e77b 1138void
5aaab254 1139Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1140{
97aff369 1141 dVAR;
bd81e77b
NC
1142 void* old_body;
1143 void* new_body;
42d0e0b7 1144 const svtype old_type = SvTYPE(sv);
d2a0f284 1145 const struct body_details *new_type_details;
238b27b3 1146 const struct body_details *old_type_details
bd81e77b 1147 = bodies_by_type + old_type;
4df7f6af 1148 SV *referant = NULL;
cac9b346 1149
7918f24d
NC
1150 PERL_ARGS_ASSERT_SV_UPGRADE;
1151
1776cbe8
NC
1152 if (old_type == new_type)
1153 return;
1154
1155 /* This clause was purposefully added ahead of the early return above to
1156 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1157 inference by Nick I-S that it would fix other troublesome cases. See
1158 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1159
1160 Given that shared hash key scalars are no longer PVIV, but PV, there is
1161 no longer need to unshare so as to free up the IVX slot for its proper
1162 purpose. So it's safe to move the early return earlier. */
1163
093085a8 1164 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1165 sv_force_normal_flags(sv, 0);
1166 }
cac9b346 1167
bd81e77b 1168 old_body = SvANY(sv);
de042e1d 1169
bd81e77b
NC
1170 /* Copying structures onto other structures that have been neatly zeroed
1171 has a subtle gotcha. Consider XPVMG
cac9b346 1172
bd81e77b
NC
1173 +------+------+------+------+------+-------+-------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH |
1175 +------+------+------+------+------+-------+-------+
1176 0 4 8 12 16 20 24 28
645c22ef 1177
bd81e77b
NC
1178 where NVs are aligned to 8 bytes, so that sizeof that structure is
1179 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1180
bd81e77b
NC
1181 +------+------+------+------+------+-------+-------+------+
1182 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1183 +------+------+------+------+------+-------+-------+------+
1184 0 4 8 12 16 20 24 28 32
08742458 1185
bd81e77b 1186 so what happens if you allocate memory for this structure:
30f9da9e 1187
bd81e77b
NC
1188 +------+------+------+------+------+-------+-------+------+------+...
1189 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 0 4 8 12 16 20 24 28 32 36
bfc44f79 1192
bd81e77b
NC
1193 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1194 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1195 started out as zero once, but it's quite possible that it isn't. So now,
1196 rather than a nicely zeroed GP, you have it pointing somewhere random.
1197 Bugs ensue.
bfc44f79 1198
bd81e77b
NC
1199 (In fact, GP ends up pointing at a previous GP structure, because the
1200 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1201 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1202 this happens to be moot because XPVGV has been re-ordered, with GP
1203 no longer after STASH)
30f9da9e 1204
bd81e77b
NC
1205 So we are careful and work out the size of used parts of all the
1206 structures. */
bfc44f79 1207
bd81e77b
NC
1208 switch (old_type) {
1209 case SVt_NULL:
1210 break;
1211 case SVt_IV:
4df7f6af
NC
1212 if (SvROK(sv)) {
1213 referant = SvRV(sv);
238b27b3
NC
1214 old_type_details = &fake_rv;
1215 if (new_type == SVt_NV)
1216 new_type = SVt_PVNV;
4df7f6af
NC
1217 } else {
1218 if (new_type < SVt_PVIV) {
1219 new_type = (new_type == SVt_NV)
1220 ? SVt_PVNV : SVt_PVIV;
1221 }
bd81e77b
NC
1222 }
1223 break;
1224 case SVt_NV:
1225 if (new_type < SVt_PVNV) {
1226 new_type = SVt_PVNV;
bd81e77b
NC
1227 }
1228 break;
bd81e77b
NC
1229 case SVt_PV:
1230 assert(new_type > SVt_PV);
1231 assert(SVt_IV < SVt_PV);
1232 assert(SVt_NV < SVt_PV);
1233 break;
1234 case SVt_PVIV:
1235 break;
1236 case SVt_PVNV:
1237 break;
1238 case SVt_PVMG:
1239 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1240 there's no way that it can be safely upgraded, because perl.c
1241 expects to Safefree(SvANY(PL_mess_sv)) */
1242 assert(sv != PL_mess_sv);
1243 /* This flag bit is used to mean other things in other scalar types.
1244 Given that it only has meaning inside the pad, it shouldn't be set
1245 on anything that can get upgraded. */
00b1698f 1246 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1247 break;
1248 default:
2439e033 1249 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1250 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1251 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1252 }
3376de98 1253
2439e033 1254 if (UNLIKELY(old_type > new_type))
3376de98
NC
1255 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1256 (int)old_type, (int)new_type);
1257
2fa1109b 1258 new_type_details = bodies_by_type + new_type;
645c22ef 1259
bd81e77b
NC
1260 SvFLAGS(sv) &= ~SVTYPEMASK;
1261 SvFLAGS(sv) |= new_type;
932e9ff9 1262
ab4416c0
NC
1263 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1264 the return statements above will have triggered. */
1265 assert (new_type != SVt_NULL);
bd81e77b 1266 switch (new_type) {
bd81e77b
NC
1267 case SVt_IV:
1268 assert(old_type == SVt_NULL);
1269 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1270 SvIV_set(sv, 0);
1271 return;
1272 case SVt_NV:
1273 assert(old_type == SVt_NULL);
1274 SvANY(sv) = new_XNV();
1275 SvNV_set(sv, 0);
1276 return;
bd81e77b 1277 case SVt_PVHV:
bd81e77b 1278 case SVt_PVAV:
d2a0f284 1279 assert(new_type_details->body_size);
c1ae03ae
NC
1280
1281#ifndef PURIFY
1282 assert(new_type_details->arena);
d2a0f284 1283 assert(new_type_details->arena_size);
c1ae03ae 1284 /* This points to the start of the allocated area. */
d2a0f284
JC
1285 new_body_inline(new_body, new_type);
1286 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1287 new_body = ((char *)new_body) - new_type_details->offset;
1288#else
1289 /* We always allocated the full length item with PURIFY. To do this
1290 we fake things so that arena is false for all 16 types.. */
1291 new_body = new_NOARENAZ(new_type_details);
1292#endif
1293 SvANY(sv) = new_body;
1294 if (new_type == SVt_PVAV) {
1295 AvMAX(sv) = -1;
1296 AvFILLp(sv) = -1;
1297 AvREAL_only(sv);
64484faa 1298 if (old_type_details->body_size) {
ac572bf4
NC
1299 AvALLOC(sv) = 0;
1300 } else {
1301 /* It will have been zeroed when the new body was allocated.
1302 Lets not write to it, in case it confuses a write-back
1303 cache. */
1304 }
78ac7dd9
NC
1305 } else {
1306 assert(!SvOK(sv));
1307 SvOK_off(sv);
1308#ifndef NODEFAULT_SHAREKEYS
1309 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1310#endif
586fc6a3
SM
1311 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1312 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1313 }
aeb18a1e 1314
bd81e77b
NC
1315 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1316 The target created by newSVrv also is, and it can have magic.
1317 However, it never has SvPVX set.
1318 */
4df7f6af
NC
1319 if (old_type == SVt_IV) {
1320 assert(!SvROK(sv));
1321 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1322 assert(SvPVX_const(sv) == 0);
1323 }
aeb18a1e 1324
bd81e77b 1325 if (old_type >= SVt_PVMG) {
e736a858 1326 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1327 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1328 } else {
1329 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1330 }
1331 break;
93e68bfb 1332
bd81e77b
NC
1333 case SVt_PVIV:
1334 /* XXX Is this still needed? Was it ever needed? Surely as there is
1335 no route from NV to PVIV, NOK can never be true */
1336 assert(!SvNOKp(sv));
1337 assert(!SvNOK(sv));
1338 case SVt_PVIO:
1339 case SVt_PVFM:
bd81e77b
NC
1340 case SVt_PVGV:
1341 case SVt_PVCV:
1342 case SVt_PVLV:
12c45b25 1343 case SVt_REGEXP:
bd81e77b
NC
1344 case SVt_PVMG:
1345 case SVt_PVNV:
1346 case SVt_PV:
93e68bfb 1347
d2a0f284 1348 assert(new_type_details->body_size);
bd81e77b
NC
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 if(new_type_details->arena) {
1352 /* This points to the start of the allocated area. */
d2a0f284
JC
1353 new_body_inline(new_body, new_type);
1354 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1355 new_body = ((char *)new_body) - new_type_details->offset;
1356 } else {
1357 new_body = new_NOARENAZ(new_type_details);
1358 }
1359 SvANY(sv) = new_body;
5e2fc214 1360
bd81e77b 1361 if (old_type_details->copy) {
f9ba3d20
NC
1362 /* There is now the potential for an upgrade from something without
1363 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1364 int offset = old_type_details->offset;
1365 int length = old_type_details->copy;
1366
1367 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1368 const int difference
f9ba3d20
NC
1369 = new_type_details->offset - old_type_details->offset;
1370 offset += difference;
1371 length -= difference;
1372 }
1373 assert (length >= 0);
1374
1375 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1376 char);
bd81e77b
NC
1377 }
1378
1379#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1380 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1381 * correct 0.0 for us. Otherwise, if the old body didn't have an
1382 * NV slot, but the new one does, then we need to initialise the
1383 * freshly created NV slot with whatever the correct bit pattern is
1384 * for 0.0 */
e22a937e
NC
1385 if (old_type_details->zero_nv && !new_type_details->zero_nv
1386 && !isGV_with_GP(sv))
bd81e77b 1387 SvNV_set(sv, 0);
82048762 1388#endif
5e2fc214 1389
2439e033 1390 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1391 IO * const io = MUTABLE_IO(sv);
d963bf01 1392 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1393
1394 SvOBJECT_on(io);
1395 /* Clear the stashcache because a new IO could overrule a package
1396 name */
103f5a36 1397 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1398 hv_clear(PL_stashcache);
1399
85dca89a 1400 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1401 IoPAGE_LEN(sv) = 60;
85dca89a 1402 }
2439e033 1403 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1404 sv->sv_u.svu_rx = (regexp *)new_body;
1405 else if (old_type < SVt_PV) {
4df7f6af
NC
1406 /* referant will be NULL unless the old type was SVt_IV emulating
1407 SVt_RV */
1408 sv->sv_u.svu_rv = referant;
1409 }
bd81e77b
NC
1410 break;
1411 default:
afd78fd5
JH
1412 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1413 (unsigned long)new_type);
bd81e77b 1414 }
73171d91 1415
db93c0c4 1416 if (old_type > SVt_IV) {
bd81e77b 1417#ifdef PURIFY
beeec492 1418 safefree(old_body);
bd81e77b 1419#else
bc786448
GG
1420 /* Note that there is an assumption that all bodies of types that
1421 can be upgraded came from arenas. Only the more complex non-
1422 upgradable types are allowed to be directly malloc()ed. */
1423 assert(old_type_details->arena);
bd81e77b
NC
1424 del_body((void*)((char*)old_body + old_type_details->offset),
1425 &PL_body_roots[old_type]);
1426#endif
1427 }
1428}
73171d91 1429
bd81e77b
NC
1430/*
1431=for apidoc sv_backoff
73171d91 1432
fde67290 1433Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1434wrapper instead.
73171d91 1435
bd81e77b 1436=cut
73171d91
NC
1437*/
1438
bd81e77b 1439int
5aaab254 1440Perl_sv_backoff(pTHX_ SV *const sv)
bd81e77b 1441{
69240efd 1442 STRLEN delta;
7a4bba22 1443 const char * const s = SvPVX_const(sv);
7918f24d
NC
1444
1445 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1446 PERL_UNUSED_CONTEXT;
7918f24d 1447
bd81e77b
NC
1448 assert(SvOOK(sv));
1449 assert(SvTYPE(sv) != SVt_PVHV);
1450 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1451
69240efd
NC
1452 SvOOK_offset(sv, delta);
1453
7a4bba22
NC
1454 SvLEN_set(sv, SvLEN(sv) + delta);
1455 SvPV_set(sv, SvPVX(sv) - delta);
1456 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1457 SvFLAGS(sv) &= ~SVf_OOK;
1458 return 0;
1459}
73171d91 1460
bd81e77b
NC
1461/*
1462=for apidoc sv_grow
73171d91 1463
bd81e77b
NC
1464Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1465upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1466Use the C<SvGROW> wrapper instead.
93e68bfb 1467
bd81e77b
NC
1468=cut
1469*/
93e68bfb 1470
bd81e77b 1471char *
5aaab254 1472Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1473{
eb578fdb 1474 char *s;
93e68bfb 1475
7918f24d
NC
1476 PERL_ARGS_ASSERT_SV_GROW;
1477
bd81e77b
NC
1478#ifdef HAS_64K_LIMIT
1479 if (newlen >= 0x10000) {
1480 PerlIO_printf(Perl_debug_log,
1481 "Allocation too large: %"UVxf"\n", (UV)newlen);
1482 my_exit(1);
1483 }
1484#endif /* HAS_64K_LIMIT */
1485 if (SvROK(sv))
1486 sv_unref(sv);
1487 if (SvTYPE(sv) < SVt_PV) {
1488 sv_upgrade(sv, SVt_PV);
1489 s = SvPVX_mutable(sv);
1490 }
1491 else if (SvOOK(sv)) { /* pv is offset? */
1492 sv_backoff(sv);
1493 s = SvPVX_mutable(sv);
1494 if (newlen > SvLEN(sv))
1495 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1496#ifdef HAS_64K_LIMIT
1497 if (newlen >= 0x10000)
1498 newlen = 0xFFFF;
1499#endif
1500 }
1501 else
db2c6cb3
FC
1502 {
1503 if (SvIsCOW(sv)) sv_force_normal(sv);
bd81e77b 1504 s = SvPVX_mutable(sv);
db2c6cb3 1505 }
aeb18a1e 1506
cbcb2a16
DM
1507#ifdef PERL_NEW_COPY_ON_WRITE
1508 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1509 * to store the COW count. So in general, allocate one more byte than
1510 * asked for, to make it likely this byte is always spare: and thus
1511 * make more strings COW-able.
1512 * If the new size is a big power of two, don't bother: we assume the
1513 * caller wanted a nice 2^N sized block and will be annoyed at getting
1514 * 2^N+1 */
1515 if (newlen & 0xff)
1516 newlen++;
1517#endif
1518
bd81e77b 1519 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1520 STRLEN minlen = SvCUR(sv);
1521 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1522 if (newlen < minlen)
1523 newlen = minlen;
aedff202 1524#ifndef Perl_safesysmalloc_size
bd81e77b 1525 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1526#endif
98653f18 1527 if (SvLEN(sv) && s) {
10edeb5d 1528 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1529 }
1530 else {
10edeb5d 1531 s = (char*)safemalloc(newlen);
bd81e77b
NC
1532 if (SvPVX_const(sv) && SvCUR(sv)) {
1533 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1534 }
1535 }
1536 SvPV_set(sv, s);
ca7c1a29 1537#ifdef Perl_safesysmalloc_size
98653f18
NC
1538 /* Do this here, do it once, do it right, and then we will never get
1539 called back into sv_grow() unless there really is some growing
1540 needed. */
ca7c1a29 1541 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1542#else
bd81e77b 1543 SvLEN_set(sv, newlen);
98653f18 1544#endif
bd81e77b
NC
1545 }
1546 return s;
1547}
aeb18a1e 1548
bd81e77b
NC
1549/*
1550=for apidoc sv_setiv
932e9ff9 1551
bd81e77b
NC
1552Copies an integer into the given SV, upgrading first if necessary.
1553Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1554
bd81e77b
NC
1555=cut
1556*/
463ee0b2 1557
bd81e77b 1558void
5aaab254 1559Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1560{
97aff369 1561 dVAR;
7918f24d
NC
1562
1563 PERL_ARGS_ASSERT_SV_SETIV;
1564
bd81e77b
NC
1565 SV_CHECK_THINKFIRST_COW_DROP(sv);
1566 switch (SvTYPE(sv)) {
1567 case SVt_NULL:
bd81e77b 1568 case SVt_NV:
3376de98 1569 sv_upgrade(sv, SVt_IV);
bd81e77b 1570 break;
bd81e77b
NC
1571 case SVt_PV:
1572 sv_upgrade(sv, SVt_PVIV);
1573 break;
463ee0b2 1574
bd81e77b 1575 case SVt_PVGV:
6e592b3a
BM
1576 if (!isGV_with_GP(sv))
1577 break;
bd81e77b
NC
1578 case SVt_PVAV:
1579 case SVt_PVHV:
1580 case SVt_PVCV:
1581 case SVt_PVFM:
1582 case SVt_PVIO:
22e74366 1583 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1584 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1585 OP_DESC(PL_op));
42d0e0b7 1586 default: NOOP;
bd81e77b
NC
1587 }
1588 (void)SvIOK_only(sv); /* validate number */
1589 SvIV_set(sv, i);
1590 SvTAINT(sv);
1591}
932e9ff9 1592
bd81e77b
NC
1593/*
1594=for apidoc sv_setiv_mg
d33b2eba 1595
bd81e77b 1596Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1597
bd81e77b
NC
1598=cut
1599*/
d33b2eba 1600
bd81e77b 1601void
5aaab254 1602Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1603{
7918f24d
NC
1604 PERL_ARGS_ASSERT_SV_SETIV_MG;
1605
bd81e77b
NC
1606 sv_setiv(sv,i);
1607 SvSETMAGIC(sv);
1608}
727879eb 1609
bd81e77b
NC
1610/*
1611=for apidoc sv_setuv
d33b2eba 1612
bd81e77b
NC
1613Copies an unsigned integer into the given SV, upgrading first if necessary.
1614Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1615
bd81e77b
NC
1616=cut
1617*/
d33b2eba 1618
bd81e77b 1619void
5aaab254 1620Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1621{
7918f24d
NC
1622 PERL_ARGS_ASSERT_SV_SETUV;
1623
013abb9b
NC
1624 /* With the if statement to ensure that integers are stored as IVs whenever
1625 possible:
bd81e77b 1626 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1627
bd81e77b
NC
1628 without
1629 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1630
013abb9b
NC
1631 If you wish to remove the following if statement, so that this routine
1632 (and its callers) always return UVs, please benchmark to see what the
1633 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1634 */
1635 if (u <= (UV)IV_MAX) {
1636 sv_setiv(sv, (IV)u);
1637 return;
1638 }
1639 sv_setiv(sv, 0);
1640 SvIsUV_on(sv);
1641 SvUV_set(sv, u);
1642}
d33b2eba 1643
bd81e77b
NC
1644/*
1645=for apidoc sv_setuv_mg
727879eb 1646
bd81e77b 1647Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1648
bd81e77b
NC
1649=cut
1650*/
5e2fc214 1651
bd81e77b 1652void
5aaab254 1653Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1654{
7918f24d
NC
1655 PERL_ARGS_ASSERT_SV_SETUV_MG;
1656
bd81e77b
NC
1657 sv_setuv(sv,u);
1658 SvSETMAGIC(sv);
1659}
5e2fc214 1660
954c1994 1661/*
bd81e77b 1662=for apidoc sv_setnv
954c1994 1663
bd81e77b
NC
1664Copies a double into the given SV, upgrading first if necessary.
1665Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1666
1667=cut
1668*/
1669
63f97190 1670void
5aaab254 1671Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1672{
97aff369 1673 dVAR;
7918f24d
NC
1674
1675 PERL_ARGS_ASSERT_SV_SETNV;
1676
bd81e77b
NC
1677 SV_CHECK_THINKFIRST_COW_DROP(sv);
1678 switch (SvTYPE(sv)) {
79072805 1679 case SVt_NULL:
79072805 1680 case SVt_IV:
bd81e77b 1681 sv_upgrade(sv, SVt_NV);
79072805
LW
1682 break;
1683 case SVt_PV:
79072805 1684 case SVt_PVIV:
bd81e77b 1685 sv_upgrade(sv, SVt_PVNV);
79072805 1686 break;
bd4b1eb5 1687
bd4b1eb5 1688 case SVt_PVGV:
6e592b3a
BM
1689 if (!isGV_with_GP(sv))
1690 break;
bd81e77b
NC
1691 case SVt_PVAV:
1692 case SVt_PVHV:
79072805 1693 case SVt_PVCV:
bd81e77b
NC
1694 case SVt_PVFM:
1695 case SVt_PVIO:
22e74366 1696 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1697 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1698 OP_DESC(PL_op));
42d0e0b7 1699 default: NOOP;
2068cd4d 1700 }
bd81e77b
NC
1701 SvNV_set(sv, num);
1702 (void)SvNOK_only(sv); /* validate number */
1703 SvTAINT(sv);
79072805
LW
1704}
1705
645c22ef 1706/*
bd81e77b 1707=for apidoc sv_setnv_mg
645c22ef 1708
bd81e77b 1709Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1710
1711=cut
1712*/
1713
bd81e77b 1714void
5aaab254 1715Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1716{
7918f24d
NC
1717 PERL_ARGS_ASSERT_SV_SETNV_MG;
1718
bd81e77b
NC
1719 sv_setnv(sv,num);
1720 SvSETMAGIC(sv);
79072805
LW
1721}
1722
bd81e77b
NC
1723/* Print an "isn't numeric" warning, using a cleaned-up,
1724 * printable version of the offending string
1725 */
954c1994 1726
bd81e77b 1727STATIC void
aad570aa 1728S_not_a_number(pTHX_ SV *const sv)
79072805 1729{
97aff369 1730 dVAR;
bd81e77b
NC
1731 SV *dsv;
1732 char tmpbuf[64];
1733 const char *pv;
94463019 1734
7918f24d
NC
1735 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1736
94463019 1737 if (DO_UTF8(sv)) {
84bafc02 1738 dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1739 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1740 } else {
1741 char *d = tmpbuf;
551405c4 1742 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1743 /* each *s can expand to 4 chars + "...\0",
1744 i.e. need room for 8 chars */
ecdeb87c 1745
00b6aa41
AL
1746 const char *s = SvPVX_const(sv);
1747 const char * const end = s + SvCUR(sv);
1748 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1749 int ch = *s & 0xFF;
1750 if (ch & 128 && !isPRINT_LC(ch)) {
1751 *d++ = 'M';
1752 *d++ = '-';
1753 ch &= 127;
1754 }
1755 if (ch == '\n') {
1756 *d++ = '\\';
1757 *d++ = 'n';
1758 }
1759 else if (ch == '\r') {
1760 *d++ = '\\';
1761 *d++ = 'r';
1762 }
1763 else if (ch == '\f') {
1764 *d++ = '\\';
1765 *d++ = 'f';
1766 }
1767 else if (ch == '\\') {
1768 *d++ = '\\';
1769 *d++ = '\\';
1770 }
1771 else if (ch == '\0') {
1772 *d++ = '\\';
1773 *d++ = '0';
1774 }
1775 else if (isPRINT_LC(ch))
1776 *d++ = ch;
1777 else {
1778 *d++ = '^';
1779 *d++ = toCTRL(ch);
1780 }
1781 }
1782 if (s < end) {
1783 *d++ = '.';
1784 *d++ = '.';
1785 *d++ = '.';
1786 }
1787 *d = '\0';
1788 pv = tmpbuf;
a0d0e21e 1789 }
a0d0e21e 1790
533c011a 1791 if (PL_op)
9014280d 1792 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1793 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1794 "Argument \"%s\" isn't numeric in %s", pv,
1795 OP_DESC(PL_op));
a0d0e21e 1796 else
9014280d 1797 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1798 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1799 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1800}
1801
c2988b20
NC
1802/*
1803=for apidoc looks_like_number
1804
645c22ef
DM
1805Test if the content of an SV looks like a number (or is a number).
1806C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1807non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1808ignored.
c2988b20
NC
1809
1810=cut
1811*/
1812
1813I32
aad570aa 1814Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1815{
eb578fdb 1816 const char *sbegin;
c2988b20
NC
1817 STRLEN len;
1818
7918f24d
NC
1819 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1820
f52e41ad
FC
1821 if (SvPOK(sv) || SvPOKp(sv)) {
1822 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1823 }
c2988b20 1824 else
e0ab1c0e 1825 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1826 return grok_number(sbegin, len, NULL);
1827}
25da4f38 1828
19f6321d
NC
1829STATIC bool
1830S_glob_2number(pTHX_ GV * const gv)
180488f8 1831{
7918f24d
NC
1832 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1833
675c862f
AL
1834 /* We know that all GVs stringify to something that is not-a-number,
1835 so no need to test that. */
1836 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1837 {
1838 SV *const buffer = sv_newmortal();
1839 gv_efullname3(buffer, gv, "*");
675c862f 1840 not_a_number(buffer);
8e629ff4 1841 }
675c862f
AL
1842 /* We just want something true to return, so that S_sv_2iuv_common
1843 can tail call us and return true. */
19f6321d 1844 return TRUE;
675c862f
AL
1845}
1846
25da4f38
IZ
1847/* Actually, ISO C leaves conversion of UV to IV undefined, but
1848 until proven guilty, assume that things are not that bad... */
1849
645c22ef
DM
1850/*
1851 NV_PRESERVES_UV:
1852
1853 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1854 an IV (an assumption perl has been based on to date) it becomes necessary
1855 to remove the assumption that the NV always carries enough precision to
1856 recreate the IV whenever needed, and that the NV is the canonical form.
1857 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1858 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1859 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1860 1) to distinguish between IV/UV/NV slots that have cached a valid
1861 conversion where precision was lost and IV/UV/NV slots that have a
1862 valid conversion which has lost no precision
645c22ef 1863 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1864 would lose precision, the precise conversion (or differently
1865 imprecise conversion) is also performed and cached, to prevent
1866 requests for different numeric formats on the same SV causing
1867 lossy conversion chains. (lossless conversion chains are perfectly
1868 acceptable (still))
1869
1870
1871 flags are used:
1872 SvIOKp is true if the IV slot contains a valid value
1873 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1874 SvNOKp is true if the NV slot contains a valid value
1875 SvNOK is true only if the NV value is accurate
1876
1877 so
645c22ef 1878 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1879 IV(or UV) would lose accuracy over a direct conversion from PV to
1880 IV(or UV). If it would, cache both conversions, return NV, but mark
1881 SV as IOK NOKp (ie not NOK).
1882
645c22ef 1883 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1884 NV would lose accuracy over a direct conversion from PV to NV. If it
1885 would, cache both conversions, flag similarly.
1886
1887 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1888 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1889 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1890 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1891 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1892
645c22ef
DM
1893 The benefit of this is that operations such as pp_add know that if
1894 SvIOK is true for both left and right operands, then integer addition
1895 can be used instead of floating point (for cases where the result won't
1896 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1897 loss of precision compared with integer addition.
1898
1899 * making IV and NV equal status should make maths accurate on 64 bit
1900 platforms
1901 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1902 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1903 looking for SvIOK and checking for overflow will not outweigh the
1904 fp to integer speedup)
1905 * will slow down integer operations (callers of SvIV) on "inaccurate"
1906 values, as the change from SvIOK to SvIOKp will cause a call into
1907 sv_2iv each time rather than a macro access direct to the IV slot
1908 * should speed up number->string conversion on integers as IV is
645c22ef 1909 favoured when IV and NV are equally accurate
28e5dec8
JH
1910
1911 ####################################################################
645c22ef
DM
1912 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1913 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1914 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1915 ####################################################################
1916
645c22ef 1917 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1918 performance ratio.
1919*/
1920
1921#ifndef NV_PRESERVES_UV
645c22ef
DM
1922# define IS_NUMBER_UNDERFLOW_IV 1
1923# define IS_NUMBER_UNDERFLOW_UV 2
1924# define IS_NUMBER_IV_AND_UV 2
1925# define IS_NUMBER_OVERFLOW_IV 4
1926# define IS_NUMBER_OVERFLOW_UV 5
1927
1928/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1929
1930/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1931STATIC int
5aaab254 1932S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
1933# ifdef DEBUGGING
1934 , I32 numtype
1935# endif
1936 )
28e5dec8 1937{
97aff369 1938 dVAR;
7918f24d
NC
1939
1940 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1941
3f7c398e 1942 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
1943 if (SvNVX(sv) < (NV)IV_MIN) {
1944 (void)SvIOKp_on(sv);
1945 (void)SvNOK_on(sv);
45977657 1946 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1947 return IS_NUMBER_UNDERFLOW_IV;
1948 }
1949 if (SvNVX(sv) > (NV)UV_MAX) {
1950 (void)SvIOKp_on(sv);
1951 (void)SvNOK_on(sv);
1952 SvIsUV_on(sv);
607fa7f2 1953 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1954 return IS_NUMBER_OVERFLOW_UV;
1955 }
c2988b20
NC
1956 (void)SvIOKp_on(sv);
1957 (void)SvNOK_on(sv);
1958 /* Can't use strtol etc to convert this string. (See truth table in
1959 sv_2iv */
1960 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1961 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1962 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1963 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1964 } else {
1965 /* Integer is imprecise. NOK, IOKp */
1966 }
1967 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1968 }
1969 SvIsUV_on(sv);
607fa7f2 1970 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1971 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1972 if (SvUVX(sv) == UV_MAX) {
1973 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1974 possibly be preserved by NV. Hence, it must be overflow.
1975 NOK, IOKp */
1976 return IS_NUMBER_OVERFLOW_UV;
1977 }
1978 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1979 } else {
1980 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1981 }
c2988b20 1982 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1983}
645c22ef
DM
1984#endif /* !NV_PRESERVES_UV*/
1985
af359546 1986STATIC bool
7918f24d
NC
1987S_sv_2iuv_common(pTHX_ SV *const sv)
1988{
97aff369 1989 dVAR;
7918f24d
NC
1990
1991 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1992
af359546 1993 if (SvNOKp(sv)) {
28e5dec8
JH
1994 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1995 * without also getting a cached IV/UV from it at the same time
1996 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1997 * IV or UV at same time to avoid this. */
1998 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1999
2000 if (SvTYPE(sv) == SVt_NV)
2001 sv_upgrade(sv, SVt_PVNV);
2002
28e5dec8
JH
2003 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2004 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2005 certainly cast into the IV range at IV_MAX, whereas the correct
2006 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2007 cases go to UV */
cab190d4
JD
2008#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2009 if (Perl_isnan(SvNVX(sv))) {
2010 SvUV_set(sv, 0);
2011 SvIsUV_on(sv);
fdbe6d7c 2012 return FALSE;
cab190d4 2013 }
cab190d4 2014#endif
28e5dec8 2015 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2016 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2017 if (SvNVX(sv) == (NV) SvIVX(sv)
2018#ifndef NV_PRESERVES_UV
2019 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2020 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2021 /* Don't flag it as "accurately an integer" if the number
2022 came from a (by definition imprecise) NV operation, and
2023 we're outside the range of NV integer precision */
2024#endif
2025 ) {
a43d94f2
NC
2026 if (SvNOK(sv))
2027 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2028 else {
2029 /* scalar has trailing garbage, eg "42a" */
2030 }
28e5dec8 2031 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2032 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2033 PTR2UV(sv),
2034 SvNVX(sv),
2035 SvIVX(sv)));
2036
2037 } else {
2038 /* IV not precise. No need to convert from PV, as NV
2039 conversion would already have cached IV if it detected
2040 that PV->IV would be better than PV->NV->IV
2041 flags already correct - don't set public IOK. */
2042 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2043 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2044 PTR2UV(sv),
2045 SvNVX(sv),
2046 SvIVX(sv)));
2047 }
2048 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2049 but the cast (NV)IV_MIN rounds to a the value less (more
2050 negative) than IV_MIN which happens to be equal to SvNVX ??
2051 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2052 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2053 (NV)UVX == NVX are both true, but the values differ. :-(
2054 Hopefully for 2s complement IV_MIN is something like
2055 0x8000000000000000 which will be exact. NWC */
d460ef45 2056 }
25da4f38 2057 else {
607fa7f2 2058 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2059 if (
2060 (SvNVX(sv) == (NV) SvUVX(sv))
2061#ifndef NV_PRESERVES_UV
2062 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2063 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2064 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2065 /* Don't flag it as "accurately an integer" if the number
2066 came from a (by definition imprecise) NV operation, and
2067 we're outside the range of NV integer precision */
2068#endif
a43d94f2 2069 && SvNOK(sv)
28e5dec8
JH
2070 )
2071 SvIOK_on(sv);
25da4f38 2072 SvIsUV_on(sv);
1c846c1f 2073 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2074 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2075 PTR2UV(sv),
57def98f
JH
2076 SvUVX(sv),
2077 SvUVX(sv)));
25da4f38 2078 }
748a9306 2079 }
cd84013a 2080 else if (SvPOKp(sv)) {
c2988b20 2081 UV value;
504618e9 2082 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2083 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2084 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2085 the same as the direct translation of the initial string
2086 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2087 be careful to ensure that the value with the .456 is around if the
2088 NV value is requested in the future).
1c846c1f 2089
af359546 2090 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2091 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2092 cache the NV if we are sure it's not needed.
25da4f38 2093 */
16b7a9a4 2094
c2988b20
NC
2095 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2096 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2097 == IS_NUMBER_IN_UV) {
5e045b90 2098 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2099 if (SvTYPE(sv) < SVt_PVIV)
2100 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2101 (void)SvIOK_on(sv);
c2988b20
NC
2102 } else if (SvTYPE(sv) < SVt_PVNV)
2103 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2104
f2524eef 2105 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2106 we aren't going to call atof() below. If NVs don't preserve UVs
2107 then the value returned may have more precision than atof() will
2108 return, even though value isn't perfectly accurate. */
2109 if ((numtype & (IS_NUMBER_IN_UV
2110#ifdef NV_PRESERVES_UV
2111 | IS_NUMBER_NOT_INT
2112#endif
2113 )) == IS_NUMBER_IN_UV) {
2114 /* This won't turn off the public IOK flag if it was set above */
2115 (void)SvIOKp_on(sv);
2116
2117 if (!(numtype & IS_NUMBER_NEG)) {
2118 /* positive */;
2119 if (value <= (UV)IV_MAX) {
45977657 2120 SvIV_set(sv, (IV)value);
c2988b20 2121 } else {
af359546 2122 /* it didn't overflow, and it was positive. */
607fa7f2 2123 SvUV_set(sv, value);
c2988b20
NC
2124 SvIsUV_on(sv);
2125 }
2126 } else {
2127 /* 2s complement assumption */
2128 if (value <= (UV)IV_MIN) {
45977657 2129 SvIV_set(sv, -(IV)value);
c2988b20
NC
2130 } else {
2131 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2132 I'm assuming it will be rare. */
c2988b20
NC
2133 if (SvTYPE(sv) < SVt_PVNV)
2134 sv_upgrade(sv, SVt_PVNV);
2135 SvNOK_on(sv);
2136 SvIOK_off(sv);
2137 SvIOKp_on(sv);
9d6ce603 2138 SvNV_set(sv, -(NV)value);
45977657 2139 SvIV_set(sv, IV_MIN);
c2988b20
NC
2140 }
2141 }
2142 }
2143 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2144 will be in the previous block to set the IV slot, and the next
2145 block to set the NV slot. So no else here. */
2146
2147 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2148 != IS_NUMBER_IN_UV) {
2149 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2150 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2151
c2988b20
NC
2152 if (! numtype && ckWARN(WARN_NUMERIC))
2153 not_a_number(sv);
28e5dec8 2154
65202027 2155#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2156 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2157 PTR2UV(sv), SvNVX(sv)));
65202027 2158#else
1779d84d 2159 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2160 PTR2UV(sv), SvNVX(sv)));
65202027 2161#endif
28e5dec8 2162
28e5dec8 2163#ifdef NV_PRESERVES_UV
af359546
NC
2164 (void)SvIOKp_on(sv);
2165 (void)SvNOK_on(sv);
2166 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2167 SvIV_set(sv, I_V(SvNVX(sv)));
2168 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2169 SvIOK_on(sv);
2170 } else {
6f207bd3 2171 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2172 }
2173 /* UV will not work better than IV */
2174 } else {
2175 if (SvNVX(sv) > (NV)UV_MAX) {
2176 SvIsUV_on(sv);
2177 /* Integer is inaccurate. NOK, IOKp, is UV */
2178 SvUV_set(sv, UV_MAX);
af359546
NC
2179 } else {
2180 SvUV_set(sv, U_V(SvNVX(sv)));
2181 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2182 NV preservse UV so can do correct comparison. */
2183 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2184 SvIOK_on(sv);
af359546 2185 } else {
6f207bd3 2186 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2187 }
2188 }
4b0c9573 2189 SvIsUV_on(sv);
af359546 2190 }
28e5dec8 2191#else /* NV_PRESERVES_UV */
c2988b20
NC
2192 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2193 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2194 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2195 grok_number above. The NV slot has just been set using
2196 Atof. */
560b0c46 2197 SvNOK_on(sv);
c2988b20
NC
2198 assert (SvIOKp(sv));
2199 } else {
2200 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2201 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2202 /* Small enough to preserve all bits. */
2203 (void)SvIOKp_on(sv);
2204 SvNOK_on(sv);
45977657 2205 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2206 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2207 SvIOK_on(sv);
2208 /* Assumption: first non-preserved integer is < IV_MAX,
2209 this NV is in the preserved range, therefore: */
2210 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2211 < (UV)IV_MAX)) {
32fdb065 2212 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
2213 }
2214 } else {
2215 /* IN_UV NOT_INT
2216 0 0 already failed to read UV.
2217 0 1 already failed to read UV.
2218 1 0 you won't get here in this case. IV/UV
2219 slot set, public IOK, Atof() unneeded.
2220 1 1 already read UV.
2221 so there's no point in sv_2iuv_non_preserve() attempting
2222 to use atol, strtol, strtoul etc. */
47031da6 2223# ifdef DEBUGGING
40a17c4c 2224 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2225# else
2226 sv_2iuv_non_preserve (sv);
2227# endif
c2988b20
NC
2228 }
2229 }
28e5dec8 2230#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2231 /* It might be more code efficient to go through the entire logic above
2232 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2233 gets complex and potentially buggy, so more programmer efficient
2234 to do it this way, by turning off the public flags: */
2235 if (!numtype)
2236 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2237 }
af359546
NC
2238 }
2239 else {
675c862f 2240 if (isGV_with_GP(sv))
159b6efe 2241 return glob_2number(MUTABLE_GV(sv));
180488f8 2242
9a214eec 2243 if (!SvPADTMP(sv)) {
af359546
NC
2244 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2245 report_uninit(sv);
2246 }
25da4f38
IZ
2247 if (SvTYPE(sv) < SVt_IV)
2248 /* Typically the caller expects that sv_any is not NULL now. */
2249 sv_upgrade(sv, SVt_IV);
af359546
NC
2250 /* Return 0 from the caller. */
2251 return TRUE;
2252 }
2253 return FALSE;
2254}
2255
2256/*
2257=for apidoc sv_2iv_flags
2258
2259Return the integer value of an SV, doing any necessary string
2260conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2261Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2262
2263=cut
2264*/
2265
2266IV
5aaab254 2267Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2268{
97aff369 2269 dVAR;
4bac9ae4 2270
af359546 2271 if (!sv)
a0d0e21e 2272 return 0;
4bac9ae4
CS
2273
2274 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2275 mg_get(sv);
2276
2277 if (SvROK(sv)) {
2278 if (SvAMAGIC(sv)) {
2279 SV * tmpstr;
2280 if (flags & SV_SKIP_OVERLOAD)
2281 return 0;
2282 tmpstr = AMG_CALLunary(sv, numer_amg);
2283 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2284 return SvIV(tmpstr);
2285 }
2286 }
2287 return PTR2IV(SvRV(sv));
2288 }
2289
8d919b0a 2290 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2291 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2292 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2293 In practice they are extremely unlikely to actually get anywhere
2294 accessible by user Perl code - the only way that I'm aware of is when
2295 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2296
2297 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2298 */
8d919b0a 2299 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2300 {
71c558c3 2301 UV value;
8d919b0a
FC
2302 const char * const ptr =
2303 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2304 const int numtype
8d919b0a 2305 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2306
2307 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2308 == IS_NUMBER_IN_UV) {
2309 /* It's definitely an integer */
2310 if (numtype & IS_NUMBER_NEG) {
2311 if (value < (UV)IV_MIN)
2312 return -(IV)value;
2313 } else {
2314 if (value < (UV)IV_MAX)
2315 return (IV)value;
2316 }
2317 }
2318 if (!numtype) {
2319 if (ckWARN(WARN_NUMERIC))
2320 not_a_number(sv);
2321 }
8d919b0a 2322 return I_V(Atof(ptr));
e20b6c3b 2323 }
4bac9ae4
CS
2324 }
2325
2326 if (SvTHINKFIRST(sv)) {
73381561 2327#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2328 if (SvIsCOW(sv)) {
2329 sv_force_normal_flags(sv, 0);
2330 }
73381561 2331#endif
af359546
NC
2332 if (SvREADONLY(sv) && !SvOK(sv)) {
2333 if (ckWARN(WARN_UNINITIALIZED))
2334 report_uninit(sv);
2335 return 0;
2336 }
2337 }
4bac9ae4 2338
af359546
NC
2339 if (!SvIOKp(sv)) {
2340 if (S_sv_2iuv_common(aTHX_ sv))
2341 return 0;
79072805 2342 }
4bac9ae4 2343
1d7c1841
GS
2344 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2345 PTR2UV(sv),SvIVX(sv)));
25da4f38 2346 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2347}
2348
645c22ef 2349/*
891f9566 2350=for apidoc sv_2uv_flags
645c22ef
DM
2351
2352Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2353conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2354Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2355
2356=cut
2357*/
2358
ff68c719 2359UV
5aaab254 2360Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2361{
97aff369 2362 dVAR;
4bac9ae4 2363
ff68c719
PP
2364 if (!sv)
2365 return 0;
4bac9ae4
CS
2366
2367 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2368 mg_get(sv);
2369
2370 if (SvROK(sv)) {
2371 if (SvAMAGIC(sv)) {
2372 SV *tmpstr;
2373 if (flags & SV_SKIP_OVERLOAD)
2374 return 0;
2375 tmpstr = AMG_CALLunary(sv, numer_amg);
2376 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2377 return SvUV(tmpstr);
2378 }
2379 }
2380 return PTR2UV(SvRV(sv));
2381 }
2382
8d919b0a 2383 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2384 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2385 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2386 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2387 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2388 {
71c558c3 2389 UV value;
8d919b0a
FC
2390 const char * const ptr =
2391 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2392 const int numtype
8d919b0a 2393 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2394
2395 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2396 == IS_NUMBER_IN_UV) {
2397 /* It's definitely an integer */
2398 if (!(numtype & IS_NUMBER_NEG))
2399 return value;
2400 }
2401 if (!numtype) {
2402 if (ckWARN(WARN_NUMERIC))
2403 not_a_number(sv);
2404 }
8d919b0a 2405 return U_V(Atof(ptr));
e20b6c3b 2406 }
4bac9ae4
CS
2407 }
2408
2409 if (SvTHINKFIRST(sv)) {
73381561 2410#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2411 if (SvIsCOW(sv)) {
2412 sv_force_normal_flags(sv, 0);
8a818333 2413 }
73381561 2414#endif
0336b60e 2415 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2416 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2417 report_uninit(sv);
ff68c719
PP
2418 return 0;
2419 }
2420 }
4bac9ae4 2421
af359546
NC
2422 if (!SvIOKp(sv)) {
2423 if (S_sv_2iuv_common(aTHX_ sv))
2424 return 0;
ff68c719 2425 }
25da4f38 2426
1d7c1841
GS
2427 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2428 PTR2UV(sv),SvUVX(sv)));
25da4f38 2429 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2430}
2431
645c22ef 2432/*
196007d1 2433=for apidoc sv_2nv_flags
645c22ef
DM
2434
2435Return the num value of an SV, doing any necessary string or integer
fde67290 2436conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2437Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2438
2439=cut
2440*/
2441
65202027 2442NV
5aaab254 2443Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2444{
97aff369 2445 dVAR;
79072805
LW
2446 if (!sv)
2447 return 0.0;
8d919b0a 2448 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2449 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2450 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2451 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2452 const char *ptr;
39d5de13
DM
2453 if (flags & SV_GMAGIC)
2454 mg_get(sv);
463ee0b2
LW
2455 if (SvNOKp(sv))
2456 return SvNVX(sv);
cd84013a 2457 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2458 ptr = SvPVX_const(sv);
2459 grokpv:
041457d9 2460 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2461 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2462 not_a_number(sv);
8d919b0a 2463 return Atof(ptr);
a0d0e21e 2464 }
25da4f38 2465 if (SvIOKp(sv)) {
1c846c1f 2466 if (SvIsUV(sv))
65202027 2467 return (NV)SvUVX(sv);
25da4f38 2468 else
65202027 2469 return (NV)SvIVX(sv);
47a72cb8
NC
2470 }
2471 if (SvROK(sv)) {
2472 goto return_rok;
2473 }
8d919b0a
FC
2474 if (isREGEXP(sv)) {
2475 ptr = RX_WRAPPED((REGEXP *)sv);
2476 goto grokpv;
2477 }
47a72cb8
NC
2478 assert(SvTYPE(sv) >= SVt_PVMG);
2479 /* This falls through to the report_uninit near the end of the
2480 function. */
2481 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2482 if (SvROK(sv)) {
47a72cb8 2483 return_rok:
deb46114 2484 if (SvAMAGIC(sv)) {
aee036bb
DM
2485 SV *tmpstr;
2486 if (flags & SV_SKIP_OVERLOAD)
2487 return 0;
31d632c3 2488 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2489 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2490 return SvNV(tmpstr);
2491 }
2492 }
2493 return PTR2NV(SvRV(sv));
a0d0e21e 2494 }
73381561 2495#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2496 if (SvIsCOW(sv)) {
2497 sv_force_normal_flags(sv, 0);
8a818333 2498 }
73381561 2499#endif
0336b60e 2500 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2501 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2502 report_uninit(sv);
ed6116ce
LW
2503 return 0.0;
2504 }
79072805
LW
2505 }
2506 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2507 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2508 sv_upgrade(sv, SVt_NV);
906f284f 2509#ifdef USE_LONG_DOUBLE
097ee67d 2510 DEBUG_c({
f93f4e46 2511 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2512 PerlIO_printf(Perl_debug_log,
2513 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2514 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2515 RESTORE_NUMERIC_LOCAL();
2516 });
65202027 2517#else
572bbb43 2518 DEBUG_c({
f93f4e46 2519 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2520 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2521 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2522 RESTORE_NUMERIC_LOCAL();
2523 });
572bbb43 2524#endif
79072805
LW
2525 }
2526 else if (SvTYPE(sv) < SVt_PVNV)
2527 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2528 if (SvNOKp(sv)) {
2529 return SvNVX(sv);
61604483 2530 }
59d8ce62 2531 if (SvIOKp(sv)) {
9d6ce603 2532 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2533#ifdef NV_PRESERVES_UV
a43d94f2
NC
2534 if (SvIOK(sv))
2535 SvNOK_on(sv);
2536 else
2537 SvNOKp_on(sv);
28e5dec8
JH
2538#else
2539 /* Only set the public NV OK flag if this NV preserves the IV */
2540 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2541 if (SvIOK(sv) &&
2542 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2543 : (SvIVX(sv) == I_V(SvNVX(sv))))
2544 SvNOK_on(sv);
2545 else
2546 SvNOKp_on(sv);
2547#endif
93a17b20 2548 }
cd84013a 2549 else if (SvPOKp(sv)) {
c2988b20 2550 UV value;
3f7c398e 2551 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2552 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2553 not_a_number(sv);
28e5dec8 2554#ifdef NV_PRESERVES_UV
c2988b20
NC
2555 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2556 == IS_NUMBER_IN_UV) {
5e045b90 2557 /* It's definitely an integer */
9d6ce603 2558 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2559 } else
3f7c398e 2560 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2561 if (numtype)
2562 SvNOK_on(sv);
2563 else
2564 SvNOKp_on(sv);
28e5dec8 2565#else
3f7c398e 2566 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2567 /* Only set the public NV OK flag if this NV preserves the value in
2568 the PV at least as well as an IV/UV would.
2569 Not sure how to do this 100% reliably. */
2570 /* if that shift count is out of range then Configure's test is
2571 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2572 UV_BITS */
2573 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2574 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2575 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2576 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2577 /* Can't use strtol etc to convert this string, so don't try.
2578 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2579 SvNOK_on(sv);
2580 } else {
2581 /* value has been set. It may not be precise. */
2582 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2583 /* 2s complement assumption for (UV)IV_MIN */
2584 SvNOK_on(sv); /* Integer is too negative. */
2585 } else {
2586 SvNOKp_on(sv);
2587 SvIOKp_on(sv);
6fa402ec 2588
c2988b20 2589 if (numtype & IS_NUMBER_NEG) {
45977657 2590 SvIV_set(sv, -(IV)value);
c2988b20 2591 } else if (value <= (UV)IV_MAX) {
45977657 2592 SvIV_set(sv, (IV)value);
c2988b20 2593 } else {
607fa7f2 2594 SvUV_set(sv, value);
c2988b20
NC
2595 SvIsUV_on(sv);
2596 }
2597
2598 if (numtype & IS_NUMBER_NOT_INT) {
2599 /* I believe that even if the original PV had decimals,
2600 they are lost beyond the limit of the FP precision.
2601 However, neither is canonical, so both only get p
2602 flags. NWC, 2000/11/25 */
2603 /* Both already have p flags, so do nothing */
2604 } else {
66a1b24b 2605 const NV nv = SvNVX(sv);
c2988b20
NC
2606 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2607 if (SvIVX(sv) == I_V(nv)) {
2608 SvNOK_on(sv);
c2988b20 2609 } else {
c2988b20
NC
2610 /* It had no "." so it must be integer. */
2611 }
00b6aa41 2612 SvIOK_on(sv);
c2988b20
NC
2613 } else {
2614 /* between IV_MAX and NV(UV_MAX).
2615 Could be slightly > UV_MAX */
6fa402ec 2616
c2988b20
NC
2617 if (numtype & IS_NUMBER_NOT_INT) {
2618 /* UV and NV both imprecise. */
2619 } else {
66a1b24b 2620 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2621
2622 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2623 SvNOK_on(sv);
c2988b20 2624 }
00b6aa41 2625 SvIOK_on(sv);
c2988b20
NC
2626 }
2627 }
2628 }
2629 }
2630 }
a43d94f2
NC
2631 /* It might be more code efficient to go through the entire logic above
2632 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2633 gets complex and potentially buggy, so more programmer efficient
2634 to do it this way, by turning off the public flags: */
2635 if (!numtype)
2636 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2637#endif /* NV_PRESERVES_UV */
93a17b20 2638 }
79072805 2639 else {
f7877b28 2640 if (isGV_with_GP(sv)) {
159b6efe 2641 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2642 return 0.0;
2643 }
2644
9a214eec 2645 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2646 report_uninit(sv);
7e25a7e9
NC
2647 assert (SvTYPE(sv) >= SVt_NV);
2648 /* Typically the caller expects that sv_any is not NULL now. */
2649 /* XXX Ilya implies that this is a bug in callers that assume this
2650 and ideally should be fixed. */
a0d0e21e 2651 return 0.0;
79072805 2652 }
572bbb43 2653#if defined(USE_LONG_DOUBLE)
097ee67d 2654 DEBUG_c({
f93f4e46 2655 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2656 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2657 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2658 RESTORE_NUMERIC_LOCAL();
2659 });
65202027 2660#else
572bbb43 2661 DEBUG_c({
f93f4e46 2662 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2663 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2664 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2665 RESTORE_NUMERIC_LOCAL();
2666 });
572bbb43 2667#endif
463ee0b2 2668 return SvNVX(sv);
79072805
LW
2669}
2670
800401ee
JH
2671/*
2672=for apidoc sv_2num
2673
2674Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2675reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2676access this function.
800401ee
JH
2677
2678=cut
2679*/
2680
2681SV *
5aaab254 2682Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2683{
7918f24d
NC
2684 PERL_ARGS_ASSERT_SV_2NUM;
2685
b9ee0594
RGS
2686 if (!SvROK(sv))
2687 return sv;
800401ee 2688 if (SvAMAGIC(sv)) {
31d632c3 2689 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2690 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2691 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2692 return sv_2num(tmpsv);
2693 }
2694 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2695}
2696
645c22ef
DM
2697/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2698 * UV as a string towards the end of buf, and return pointers to start and
2699 * end of it.
2700 *
2701 * We assume that buf is at least TYPE_CHARS(UV) long.
2702 */
2703
864dbfa3 2704static char *
5de3775c 2705S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2706{
25da4f38 2707 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2708 char * const ebuf = ptr;
25da4f38 2709 int sign;
25da4f38 2710
7918f24d
NC
2711 PERL_ARGS_ASSERT_UIV_2BUF;
2712
25da4f38
IZ
2713 if (is_uv)
2714 sign = 0;
2715 else if (iv >= 0) {
2716 uv = iv;
2717 sign = 0;
2718 } else {
2719 uv = -iv;
2720 sign = 1;
2721 }
2722 do {
eb160463 2723 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2724 } while (uv /= 10);
2725 if (sign)
2726 *--ptr = '-';
2727 *peob = ebuf;
2728 return ptr;
2729}
2730
645c22ef
DM
2731/*
2732=for apidoc sv_2pv_flags
2733
ff276b08 2734Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2735If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2736string if necessary. Normally invoked via the C<SvPV_flags> macro.
2737C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2738
2739=cut
2740*/
2741
8d6d96c1 2742char *
5aaab254 2743Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2744{
97aff369 2745 dVAR;
eb578fdb 2746 char *s;
79072805 2747
463ee0b2 2748 if (!sv) {
cdb061a3
NC
2749 if (lp)
2750 *lp = 0;
73d840c0 2751 return (char *)"";
463ee0b2 2752 }
4bac9ae4
CS
2753 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2754 mg_get(sv);
2755 if (SvROK(sv)) {
2756 if (SvAMAGIC(sv)) {
2757 SV *tmpstr;
2758 if (flags & SV_SKIP_OVERLOAD)
2759 return NULL;
2760 tmpstr = AMG_CALLunary(sv, string_amg);
2761 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2762 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2763 /* Unwrap this: */
2764 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2765 */
75dfc8ec 2766
4bac9ae4
CS
2767 char *pv;
2768 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2769 if (flags & SV_CONST_RETURN) {
2770 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2771 } else {
4bac9ae4
CS
2772 pv = (flags & SV_MUTABLE_RETURN)
2773 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2774 }
4bac9ae4
CS
2775 if (lp)
2776 *lp = SvCUR(tmpstr);
2777 } else {
2778 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2779 }
4bac9ae4
CS
2780 if (SvUTF8(tmpstr))
2781 SvUTF8_on(sv);
2782 else
2783 SvUTF8_off(sv);
2784 return pv;
deb46114 2785 }
4bac9ae4
CS
2786 }
2787 {
2788 STRLEN len;
2789 char *retval;
2790 char *buffer;
2791 SV *const referent = SvRV(sv);
2792
2793 if (!referent) {
2794 len = 7;
2795 retval = buffer = savepvn("NULLREF", len);
2796 } else if (SvTYPE(referent) == SVt_REGEXP &&
2797 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2798 amagic_is_enabled(string_amg))) {
2799 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2800
2801 assert(re);
67d2d14d 2802
4bac9ae4
CS
2803 /* If the regex is UTF-8 we want the containing scalar to
2804 have an UTF-8 flag too */
2805 if (RX_UTF8(re))
2806 SvUTF8_on(sv);
2807 else
2808 SvUTF8_off(sv);
67d2d14d 2809
4bac9ae4
CS
2810 if (lp)
2811 *lp = RX_WRAPLEN(re);
67d2d14d 2812
4bac9ae4
CS
2813 return RX_WRAPPED(re);
2814 } else {
2815 const char *const typestr = sv_reftype(referent, 0);
2816 const STRLEN typelen = strlen(typestr);
2817 UV addr = PTR2UV(referent);
2818 const char *stashname = NULL;
2819 STRLEN stashnamelen = 0; /* hush, gcc */
2820 const char *buffer_end;
2821
2822 if (SvOBJECT(referent)) {
2823 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2824
2825 if (name) {
2826 stashname = HEK_KEY(name);
2827 stashnamelen = HEK_LEN(name);
2828
2829 if (HEK_UTF8(name)) {
2830 SvUTF8_on(sv);
fafee734 2831 } else {
4bac9ae4 2832 SvUTF8_off(sv);
fafee734 2833 }
fafee734 2834 } else {
4bac9ae4
CS
2835 stashname = "__ANON__";
2836 stashnamelen = 8;
fafee734 2837 }
4bac9ae4
CS
2838 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2839 + 2 * sizeof(UV) + 2 /* )\0 */;
2840 } else {
2841 len = typelen + 3 /* (0x */
2842 + 2 * sizeof(UV) + 2 /* )\0 */;
2843 }
fafee734 2844
4bac9ae4
CS
2845 Newx(buffer, len, char);
2846 buffer_end = retval = buffer + len;
2847
2848 /* Working backwards */
2849 *--retval = '\0';
2850 *--retval = ')';
2851 do {
2852 *--retval = PL_hexdigit[addr & 15];
2853 } while (addr >>= 4);
2854 *--retval = 'x';
2855 *--retval = '0';
2856 *--retval = '(';
2857
2858 retval -= typelen;
2859 memcpy(retval, typestr, typelen);
2860
2861 if (stashname) {
2862 *--retval = '=';
2863 retval -= stashnamelen;
2864 memcpy(retval, stashname, stashnamelen);
c080367d 2865 }
4bac9ae4
CS
2866 /* retval may not necessarily have reached the start of the
2867 buffer here. */
2868 assert (retval >= buffer);
2869
2870 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 2871 }
cdb061a3 2872 if (lp)
4bac9ae4
CS
2873 *lp = len;
2874 SAVEFREEPV(buffer);
2875 return retval;
79072805 2876 }
79072805 2877 }
4bac9ae4
CS
2878
2879 if (SvPOKp(sv)) {
2880 if (lp)
2881 *lp = SvCUR(sv);
2882 if (flags & SV_MUTABLE_RETURN)
2883 return SvPVX_mutable(sv);
2884 if (flags & SV_CONST_RETURN)
2885 return (char *)SvPVX_const(sv);
2886 return SvPVX(sv);
2887 }
2888
2889 if (SvIOK(sv)) {
28e5dec8
JH
2890 /* I'm assuming that if both IV and NV are equally valid then
2891 converting the IV is going to be more efficient */
e1ec3a88 2892 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2893 char buf[TYPE_CHARS(UV)];
2894 char *ebuf, *ptr;
97a130b8 2895 STRLEN len;
28e5dec8
JH
2896
2897 if (SvTYPE(sv) < SVt_PVIV)
2898 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2899 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2900 len = ebuf - ptr;
5902b6a9 2901 /* inlined from sv_setpvn */
97a130b8
NC
2902 s = SvGROW_mutable(sv, len + 1);
2903 Move(ptr, s, len, char);
2904 s += len;
28e5dec8 2905 *s = '\0';
b127e37e 2906 SvPOK_on(sv);
28e5dec8 2907 }
4bac9ae4 2908 else if (SvNOK(sv)) {
79072805
LW
2909 if (SvTYPE(sv) < SVt_PVNV)
2910 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2911 if (SvNVX(sv) == 0.0) {
2912 s = SvGROW_mutable(sv, 2);
2913 *s++ = '0';
2914 *s = '\0';
2915 } else {
2916 dSAVE_ERRNO;
2917 /* The +20 is pure guesswork. Configure test needed. --jhi */
2918 s = SvGROW_mutable(sv, NV_DIG + 20);
2919 /* some Xenix systems wipe out errno here */
b127e37e 2920
b127e37e 2921#ifndef USE_LOCALE_NUMERIC
68e8f474 2922 Gconvert(SvNVX(sv), NV_DIG, 0, s);
b127e37e 2923 SvPOK_on(sv);
68e8f474
KW
2924#else
2925 /* Gconvert always uses the current locale. That's the right thing
2926 * to do if we're supposed to be using locales. But otherwise, we
2927 * want the result to be based on the C locale, so we need to
2928 * change to the C locale during the Gconvert and then change back.
2929 * But if we're already in the C locale (PL_numeric_standard is
2930 * TRUE in that case), no need to do any changing */
2931 if (PL_numeric_standard || IN_LOCALE_RUNTIME) {
2932 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2933 }
2934 else {
2935 char *loc = savepv(setlocale(LC_NUMERIC, NULL));
2936 setlocale(LC_NUMERIC, "C");
2937 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2938 setlocale(LC_NUMERIC, loc);
2939 Safefree(loc);
2940 }
2941
2942 /* We don't call SvPOK_on(), because it may come to pass that the
2943 * locale changes so that the stringification we just did is no
2944 * longer correct. We will have to re-stringify every time it is
2945 * needed */
b127e37e 2946#endif
29912d93
NC
2947 RESTORE_ERRNO;
2948 while (*s) s++;
bbce6d69 2949 }
79072805
LW
2950#ifdef hcx
2951 if (s[-1] == '.')
46fc3d4c 2952 *--s = '\0';
79072805
LW
2953#endif
2954 }
4bac9ae4
CS
2955 else if (isGV_with_GP(sv)) {
2956 GV *const gv = MUTABLE_GV(sv);
2957 SV *const buffer = sv_newmortal();
8d1c3e26 2958
4bac9ae4 2959 gv_efullname3(buffer, gv, "*");
180488f8 2960
4bac9ae4
CS
2961 assert(SvPOK(buffer));
2962 if (SvUTF8(buffer))
2963 SvUTF8_on(sv);
2964 if (lp)
2965 *lp = SvCUR(buffer);
2966 return SvPVX(buffer);
2967 }
8d919b0a
FC
2968 else if (isREGEXP(sv)) {
2969 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2970 return RX_WRAPPED((REGEXP *)sv);
2971 }
4bac9ae4 2972 else {
cdb061a3 2973 if (lp)
00b6aa41 2974 *lp = 0;
9f621bb0
NC
2975 if (flags & SV_UNDEF_RETURNS_NULL)
2976 return NULL;
9a214eec 2977 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2978 report_uninit(sv);
4bac9ae4
CS
2979 /* Typically the caller expects that sv_any is not NULL now. */
2980 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 2981 sv_upgrade(sv, SVt_PV);
73d840c0 2982 return (char *)"";
79072805 2983 }
4bac9ae4 2984
cdb061a3 2985 {
823a54a3 2986 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2987 if (lp)
2988 *lp = len;
2989 SvCUR_set(sv, len);
2990 }
1d7c1841 2991 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 2992 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
2993 if (flags & SV_CONST_RETURN)
2994 return (char *)SvPVX_const(sv);
10516c54
NC
2995 if (flags & SV_MUTABLE_RETURN)
2996 return SvPVX_mutable(sv);
463ee0b2
LW
2997 return SvPVX(sv);
2998}
2999
645c22ef 3000/*
6050d10e
JP
3001=for apidoc sv_copypv
3002
3003Copies a stringified representation of the source SV into the
3004destination SV. Automatically performs any necessary mg_get and
54f0641b 3005coercion of numeric values into strings. Guaranteed to preserve
2575c402 3006UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3007sv_2pv[_flags] but operates directly on an SV instead of just the
3008string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3009would lose the UTF-8'ness of the PV.
3010
4bac9ae4
CS
3011=for apidoc sv_copypv_nomg
3012
3013Like sv_copypv, but doesn't invoke get magic first.
3014
3015=for apidoc sv_copypv_flags
3016
3017Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3018include SV_GMAGIC.
3019
6050d10e
JP
3020=cut
3021*/
3022
3023void
5aaab254 3024Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
6050d10e 3025{
4bac9ae4
CS
3026 PERL_ARGS_ASSERT_SV_COPYPV;
3027
3028 sv_copypv_flags(dsv, ssv, 0);
3029}
3030
3031void
5aaab254 3032Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3033{
446eaa42 3034 STRLEN len;
4bac9ae4 3035 const char *s;
7918f24d 3036
4bac9ae4 3037 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3038
4bac9ae4
CS
3039 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3040 mg_get(ssv);
3041 s = SvPV_nomg_const(ssv,len);
cb50f42d 3042 sv_setpvn(dsv,s,len);
446eaa42 3043 if (SvUTF8(ssv))
cb50f42d 3044 SvUTF8_on(dsv);
446eaa42 3045 else
cb50f42d 3046 SvUTF8_off(dsv);
6050d10e
JP
3047}
3048
3049/*
645c22ef
DM
3050=for apidoc sv_2pvbyte
3051
3052Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3053to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3054side-effect.
3055
3056Usually accessed via the C<SvPVbyte> macro.
3057
3058=cut
3059*/
3060
7340a771 3061char *
5aaab254 3062Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3063{
7918f24d
NC
3064 PERL_ARGS_ASSERT_SV_2PVBYTE;
3065
4499db73
FC
3066 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3067 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181
FC
3068 SV *sv2 = sv_newmortal();
3069 sv_copypv(sv2,sv);
3070 sv = sv2;
3071 }
3072 else SvGETMAGIC(sv);
0875d2fe 3073 sv_utf8_downgrade(sv,0);
71eb6d8c 3074 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3075}
3076
645c22ef 3077/*
035cbb0e
RGS
3078=for apidoc sv_2pvutf8
3079
3080Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3081to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3082
3083Usually accessed via the C<SvPVutf8> macro.
3084
3085=cut
3086*/
645c22ef 3087
7340a771 3088char *
5aaab254 3089Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3090{
7918f24d
NC
3091 PERL_ARGS_ASSERT_SV_2PVUTF8;
3092
4499db73
FC
3093 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3094 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3095 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3096 else
3097 SvGETMAGIC(sv);
3098 sv_utf8_upgrade_nomg(sv);
c3ec315f 3099 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3100}
1c846c1f 3101
7ee2227d 3102
645c22ef
DM
3103/*
3104=for apidoc sv_2bool
3105
06c841cf
FC
3106This macro is only used by sv_true() or its macro equivalent, and only if
3107the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3108It calls sv_2bool_flags with the SV_GMAGIC flag.
3109
3110=for apidoc sv_2bool_flags
3111
3112This function is only used by sv_true() and friends, and only if
fde67290 3113the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3114contain SV_GMAGIC, then it does an mg_get() first.
3115
645c22ef
DM
3116
3117=cut
3118*/
3119
463ee0b2 3120bool
5aaab254 3121Perl_sv_2bool_flags(pTHX_ SV *const sv, const I32 flags)
463ee0b2 3122{
97aff369 3123 dVAR;
7918f24d 3124
06c841cf 3125 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3126
06c841cf 3127 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3128
a0d0e21e
LW
3129 if (!SvOK(sv))
3130 return 0;
3131 if (SvROK(sv)) {
fabdb6c0 3132 if (SvAMAGIC(sv)) {
31d632c3 3133 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3134 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3135 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3136 }
3137 return SvRV(sv) != 0;
a0d0e21e 3138 }
4bac9ae4 3139 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3140}
3141
c461cf8f
JH
3142/*
3143=for apidoc sv_utf8_upgrade
3144
78ea37eb 3145Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3146Forces the SV to string form if it is not already.
2bbc8d55 3147Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3148Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3149if the whole string is the same in UTF-8 as not.
3150Returns the number of bytes in the converted string
c461cf8f 3151
0efd0472 3152This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3153use the Encode extension for that.
3154
fe749c9a
KW
3155=for apidoc sv_utf8_upgrade_nomg
3156
fde67290 3157Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3158
8d6d96c1
HS
3159=for apidoc sv_utf8_upgrade_flags
3160
78ea37eb 3161Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3162Forces the SV to string form if it is not already.
8d6d96c1 3163Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3164if all the bytes are invariant in UTF-8.
3165If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55
SP
3166will C<mg_get> on C<sv> if appropriate, else not.
3167Returns the number of bytes in the converted string
3168C<sv_utf8_upgrade> and
8d6d96c1
HS
3169C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3170
0efd0472 3171This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3172use the Encode extension for that.
3173
8d6d96c1 3174=cut
b3ab6785
KW
3175
3176The grow version is currently not externally documented. It adds a parameter,
3177extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3178have free after it upon return. This allows the caller to reserve extra space
3179that it intends to fill, to avoid extra grows.
3180
3181Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3182which can be used to tell this function to not first check to see if there are
3183any characters that are different in UTF-8 (variant characters) which would
3184force it to allocate a new string to sv, but to assume there are. Typically
3185this flag is used by a routine that has already parsed the string to find that
3186there are such characters, and passes this information on so that the work
3187doesn't have to be repeated.
3188
3189(One might think that the calling routine could pass in the position of the
3190first such variant, so it wouldn't have to be found again. But that is not the
3191case, because typically when the caller is likely to use this flag, it won't be
3192calling this routine unless it finds something that won't fit into a byte.
3193Otherwise it tries to not upgrade and just use bytes. But some things that
3194do fit into a byte are variants in utf8, and the caller may not have been
3195keeping track of these.)
3196
3197If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3198isn't guaranteed due to having other routines do the work in some input cases,
3199or if the input is already flagged as being in utf8.
3200
3201The speed of this could perhaps be improved for many cases if someone wanted to
3202write a fast function that counts the number of variant characters in a string,
3203especially if it could return the position of the first one.
3204
8d6d96c1
HS
3205*/
3206
3207STRLEN
5aaab254 3208Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3209{
97aff369 3210 dVAR;
7918f24d 3211
b3ab6785 3212 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3213
808c356f
RGS
3214 if (sv == &PL_sv_undef)
3215 return 0;
892f9127 3216 if (!SvPOK_nog(sv)) {
e0e62c2a 3217 STRLEN len = 0;
d52b7888
NC
3218 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3219 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3220 if (SvUTF8(sv)) {
3221 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3222 return len;
b3ab6785 3223 }
d52b7888 3224 } else {
33fb6f35 3225 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3226 }
e0e62c2a 3227 }
4411f3b6 3228
f5cee72b 3229 if (SvUTF8(sv)) {
b3ab6785 3230 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3231 return SvCUR(sv);
f5cee72b 3232 }
5fec3b1d 3233
765f542d
NC
3234 if (SvIsCOW(sv)) {
3235 sv_force_normal_flags(sv, 0);
db42d148
NIS
3236 }
3237
b3ab6785 3238 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3239 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3240 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3241 return SvCUR(sv);
3242 }
3243
4e93345f
KW
3244 if (SvCUR(sv) == 0) {
3245 if (extra) SvGROW(sv, extra);
3246 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3247 /* This function could be much more efficient if we
2bbc8d55 3248 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3249 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3250 * make the loop as fast as possible (although there are certainly ways
3251 * to speed this up, eg. through vectorization) */
3252 U8 * s = (U8 *) SvPVX_const(sv);
3253 U8 * e = (U8 *) SvEND(sv);
3254 U8 *t = s;
3255 STRLEN two_byte_count = 0;
c4e7c712 3256
b3ab6785
KW
3257 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3258
3259 /* See if really will need to convert to utf8. We mustn't rely on our
3260 * incoming SV being well formed and having a trailing '\0', as certain
3261 * code in pp_formline can send us partially built SVs. */
3262
c4e7c712 3263 while (t < e) {
53c1dcc0 3264 const U8 ch = *t++;
b3ab6785
KW
3265 if (NATIVE_IS_INVARIANT(ch)) continue;
3266
3267 t--; /* t already incremented; re-point to first variant */
3268 two_byte_count = 1;
3269 goto must_be_utf8;
c4e7c712 3270 }
b3ab6785
KW
3271
3272 /* utf8 conversion not needed because all are invariants. Mark as
3273 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3274 SvUTF8_on(sv);
7f0bfbea 3275 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3276 return SvCUR(sv);
3277
3278must_be_utf8:
3279
3280 /* Here, the string should be converted to utf8, either because of an
3281 * input flag (two_byte_count = 0), or because a character that
3282 * requires 2 bytes was found (two_byte_count = 1). t points either to
3283 * the beginning of the string (if we didn't examine anything), or to
3284 * the first variant. In either case, everything from s to t - 1 will
3285 * occupy only 1 byte each on output.
3286 *
3287 * There are two main ways to convert. One is to create a new string
3288 * and go through the input starting from the beginning, appending each
3289 * converted value onto the new string as we go along. It's probably
3290 * best to allocate enough space in the string for the worst possible
3291 * case rather than possibly running out of space and having to
3292 * reallocate and then copy what we've done so far. Since everything
3293 * from s to t - 1 is invariant, the destination can be initialized
3294 * with these using a fast memory copy
3295 *
3296 * The other way is to figure out exactly how big the string should be
3297 * by parsing the entire input. Then you don't have to make it big
3298 * enough to handle the worst possible case, and more importantly, if
3299 * the string you already have is large enough, you don't have to
3300 * allocate a new string, you can copy the last character in the input
3301 * string to the final position(s) that will be occupied by the
3302 * converted string and go backwards, stopping at t, since everything
3303 * before that is invariant.
3304 *
3305 * There are advantages and disadvantages to each method.
3306 *
3307 * In the first method, we can allocate a new string, do the memory
3308 * copy from the s to t - 1, and then proceed through the rest of the
3309 * string byte-by-byte.
3310 *
3311 * In the second method, we proceed through the rest of the input
3312 * string just calculating how big the converted string will be. Then
3313 * there are two cases:
3314 * 1) if the string has enough extra space to handle the converted
3315 * value. We go backwards through the string, converting until we
3316 * get to the position we are at now, and then stop. If this
3317 * position is far enough along in the string, this method is
3318 * faster than the other method. If the memory copy were the same
3319 * speed as the byte-by-byte loop, that position would be about
3320 * half-way, as at the half-way mark, parsing to the end and back
3321 * is one complete string's parse, the same amount as starting
3322 * over and going all the way through. Actually, it would be
3323 * somewhat less than half-way, as it's faster to just count bytes
3324 * than to also copy, and we don't have the overhead of allocating
3325 * a new string, changing the scalar to use it, and freeing the
3326 * existing one. But if the memory copy is fast, the break-even
3327 * point is somewhere after half way. The counting loop could be
3328 * sped up by vectorization, etc, to move the break-even point
3329 * further towards the beginning.
3330 * 2) if the string doesn't have enough space to handle the converted
3331 * value. A new string will have to be allocated, and one might
3332 * as well, given that, start from the beginning doing the first
3333 * method. We've spent extra time parsing the string and in
3334 * exchange all we've gotten is that we know precisely how big to
3335 * make the new one. Perl is more optimized for time than space,
3336 * so this case is a loser.
3337 * So what I've decided to do is not use the 2nd method unless it is
3338 * guaranteed that a new string won't have to be allocated, assuming
3339 * the worst case. I also decided not to put any more conditions on it
3340 * than this, for now. It seems likely that, since the worst case is
3341 * twice as big as the unknown portion of the string (plus 1), we won't
3342 * be guaranteed enough space, causing us to go to the first method,
3343 * unless the string is short, or the first variant character is near
3344 * the end of it. In either of these cases, it seems best to use the
3345 * 2nd method. The only circumstance I can think of where this would
3346 * be really slower is if the string had once had much more data in it
3347 * than it does now, but there is still a substantial amount in it */
3348
3349 {
3350 STRLEN invariant_head = t - s;
3351 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3352 if (SvLEN(sv) < size) {
3353
3354 /* Here, have decided to allocate a new string */
3355
3356 U8 *dst;
3357 U8 *d;
3358
3359 Newx(dst, size, U8);
3360
3361 /* If no known invariants at the beginning of the input string,
3362 * set so starts from there. Otherwise, can use memory copy to
3363 * get up to where we are now, and then start from here */
3364
3365 if (invariant_head <= 0) {
3366 d = dst;
3367 } else {
3368 Copy(s, dst, invariant_head, char);
3369 d = dst + invariant_head;
3370 }
3371
3372 while (t < e) {
3373 const UV uv = NATIVE8_TO_UNI(*t++);
3374 if (UNI_IS_INVARIANT(uv))
3375 *d++ = (U8)UNI_TO_NATIVE(uv);
3376 else {
3377 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3378 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3379 }
3380 }
3381 *d = '\0';
3382 SvPV_free(sv); /* No longer using pre-existing string */
3383 SvPV_set(sv, (char*)dst);
3384 SvCUR_set(sv, d - dst);
3385 SvLEN_set(sv, size);
3386 } else {
3387
3388 /* Here, have decided to get the exact size of the string.
3389 * Currently this happens only when we know that there is
3390 * guaranteed enough space to fit the converted string, so
3391 * don't have to worry about growing. If two_byte_count is 0,
3392 * then t points to the first byte of the string which hasn't
3393 * been examined yet. Otherwise two_byte_count is 1, and t
3394 * points to the first byte in the string that will expand to
3395 * two. Depending on this, start examining at t or 1 after t.
3396 * */
3397
3398 U8 *d = t + two_byte_count;
3399
3400
3401 /* Count up the remaining bytes that expand to two */
3402
3403 while (d < e) {
3404 const U8 chr = *d++;
3405 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3406 }
3407
3408 /* The string will expand by just the number of bytes that
3409 * occupy two positions. But we are one afterwards because of
3410 * the increment just above. This is the place to put the
3411 * trailing NUL, and to set the length before we decrement */
3412
3413 d += two_byte_count;
3414 SvCUR_set(sv, d - s);
3415 *d-- = '\0';
3416
3417
3418 /* Having decremented d, it points to the position to put the
3419 * very last byte of the expanded string. Go backwards through
3420 * the string, copying and expanding as we go, stopping when we
3421 * get to the part that is invariant the rest of the way down */
3422
3423 e--;
3424 while (e >= t) {
3425 const U8 ch = NATIVE8_TO_UNI(*e--);
3426 if (UNI_IS_INVARIANT(ch)) {
3427 *d-- = UNI_TO_NATIVE(ch);
3428 } else {
3429 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3430 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3431 }
3432 }
3433 }
75da9d4c
DM
3434
3435 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3436 /* Update pos. We do it at the end rather than during
3437 * the upgrade, to avoid slowing down the common case
3438 * (upgrade without pos) */
3439 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3440 if (mg) {
3441 I32 pos = mg->mg_len;
3442 if (pos > 0 && (U32)pos > invariant_head) {
3443 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3444 STRLEN n = (U32)pos - invariant_head;
3445 while (n > 0) {
3446 if (UTF8_IS_START(*d))
3447 d++;
3448 d++;
3449 n--;
3450 }
3451 mg->mg_len = d - (U8*)SvPVX(sv);
3452 }
3453 }
3454 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3455 magic_setutf8(sv,mg); /* clear UTF8 cache */
3456 }
b3ab6785 3457 }
560a288e 3458 }
b3ab6785
KW
3459
3460 /* Mark as UTF-8 even if no variant - saves scanning loop */
3461 SvUTF8_on(sv);
4411f3b6 3462 return SvCUR(sv);
560a288e
GS
3463}
3464
c461cf8f
JH
3465/*
3466=for apidoc sv_utf8_downgrade
3467
78ea37eb 3468Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3469If the PV contains a character that cannot fit
3470in a byte, this conversion will fail;
78ea37eb 3471in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3472true, croaks.
3473
0efd0472 3474This is not a general purpose Unicode to byte encoding interface:
13a6c0e0
JH
3475use the Encode extension for that.
3476
c461cf8f
JH
3477=cut
3478*/
3479
560a288e 3480bool
5aaab254 3481Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
560a288e 3482{
97aff369 3483 dVAR;
7918f24d
NC
3484
3485 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3486
78ea37eb 3487 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3488 if (SvCUR(sv)) {
03cfe0ae 3489 U8 *s;
652088fc 3490 STRLEN len;
75da9d4c 3491 int mg_flags = SV_GMAGIC;
fa301091 3492
765f542d
NC
3493 if (SvIsCOW(sv)) {
3494 sv_force_normal_flags(sv, 0);
3495 }
75da9d4c
DM
3496 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3497 /* update pos */
3498 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3499 if (mg) {
3500 I32 pos = mg->mg_len;
3501 if (pos > 0) {
3502 sv_pos_b2u(sv, &pos);
3503 mg_flags = 0; /* sv_pos_b2u does get magic */
3504 mg->mg_len = pos;
3505 }
3506 }
3507 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3508 magic_setutf8(sv,mg); /* clear UTF8 cache */
3509
3510 }
3511 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3512
03cfe0ae 3513 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3514 if (fail_ok)
3515 return FALSE;
3516 else {
3517 if (PL_op)
3518 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3519 OP_DESC(PL_op));
fa301091
JH
3520 else
3521 Perl_croak(aTHX_ "Wide character");
3522 }
4b3603a4 3523 }
b162af07 3524 SvCUR_set(sv, len);
67e989fb 3525 }
560a288e 3526 }
ffebcc3e 3527 SvUTF8_off(sv);
560a288e
GS
3528 return TRUE;
3529}
3530
c461cf8f
JH
3531/*
3532=for apidoc sv_utf8_encode
3533
78ea37eb
ST
3534Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3535flag off so that it looks like octets again.
c461cf8f
JH
3536
3537=cut
3538*/
3539
560a288e 3540void
5aaab254 3541Perl_sv_utf8_encode(pTHX_ SV *const sv)
560a288e 3542{
7918f24d
NC
3543 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3544
4c94c214 3545 if (SvREADONLY(sv)) {
654c723f 3546 sv_force_normal_flags(sv, 0);
4c94c214 3547 }
a5f5288a 3548 (void) sv_utf8_upgrade(sv);
560a288e
GS
3549 SvUTF8_off(sv);
3550}
3551
4411f3b6
NIS
3552/*
3553=for apidoc sv_utf8_decode
3554
78ea37eb
ST
3555If the PV of the SV is an octet sequence in UTF-8
3556and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3557so that it looks like a character. If the PV contains only single-byte
694cf0d2 3558characters, the C<SvUTF8> flag stays off.
78ea37eb 3559Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3560
3561=cut
3562*/
3563
560a288e 3564bool
5aaab254 3565Perl_sv_utf8_decode(pTHX_ SV *const sv)
560a288e 3566{
7918f24d
NC
3567 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3568
78ea37eb 3569 if (SvPOKp(sv)) {
75da9d4c 3570 const U8 *start, *c;
93524f2b 3571 const U8 *e;
9cbac4c7 3572
645c22ef
DM
3573 /* The octets may have got themselves encoded - get them back as
3574 * bytes
3575 */
3576 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3577 return FALSE;
3578
3579 /* it is actually just a matter of turning the utf8 flag on, but
3580 * we want to make sure everything inside is valid utf8 first.
3581 */
75da9d4c 3582 c = start = (const U8 *) SvPVX_const(sv);
f9cf141a 3583 if (!is_utf8_string(c, SvCUR(sv)))
67e989fb 3584 return FALSE;
93524f2b 3585 e = (const U8 *) SvEND(sv);
511c2ff0 3586 while (c < e) {
b64e5050 3587 const U8 ch = *c++;
c4d5f83a 3588 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3589 SvUTF8_on(sv);
3590 break;
3591 }
560a288e 3592 }
75da9d4c
DM
3593 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3594 /* adjust pos to the start of a UTF8 char sequence */
3595 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3596 if (mg) {
3597 I32 pos = mg->mg_len;
3598 if (pos > 0) {
3599 for (c = start + pos; c > start; c--) {
3600 if (UTF8_IS_START(*c))
3601 break;
3602 }
3603 mg->mg_len = c - start;
3604 }
3605 }
3606 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3607 magic_setutf8(sv,mg); /* clear UTF8 cache */
3608 }
560a288e
GS
3609 }
3610 return TRUE;
3611}
3612
954c1994
GS
3613/*
3614=for apidoc sv_setsv
3615
645c22ef
DM
3616Copies the contents of the source SV C<ssv> into the destination SV
3617C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3618function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3619Loosely speaking, it performs a copy-by-value, obliterating any previous
3620content of the destination.
3621
3622You probably want to use one of the assortment of wrappers, such as
3623C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3624C<SvSetMagicSV_nosteal>.
3625
8d6d96c1
HS
3626=for apidoc sv_setsv_flags
3627
645c22ef
DM
3628Copies the contents of the source SV C<ssv> into the destination SV
3629C<dsv>. The source SV may be destroyed if it is mortal, so don't use this