This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Remove union _xivu from struct xpvhv - replace it with a non-union xav_keys.
[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
51371543 35#define FCALL *f
2c5424a7 36
2f8ed50e
OS
37#ifdef __Lynx__
38/* Missing proto on LynxOS */
39 char *gconvert(double, int, int, char *);
40#endif
41
e23c8137 42#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 43/* if adding more checks watch out for the following tests:
e23c8137
JH
44 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
45 * lib/utf8.t lib/Unicode/Collate/t/index.t
46 * --jhi
47 */
6f207bd3 48# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
49 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
50 assert((cache)[2] <= (cache)[3]); \
51 assert((cache)[3] <= (cache)[1]);} \
52 } STMT_END
e23c8137 53#else
6f207bd3 54# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
55#endif
56
f8c7b90f 57#ifdef PERL_OLD_COPY_ON_WRITE
765f542d 58#define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
607fa7f2 59#define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
b5ccf5f2 60/* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
765f542d 61 on-write. */
765f542d 62#endif
645c22ef
DM
63
64/* ============================================================================
65
66=head1 Allocation and deallocation of SVs.
67
d2a0f284
JC
68An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
69sv, av, hv...) contains type and reference count information, and for
70many types, a pointer to the body (struct xrv, xpv, xpviv...), which
71contains fields specific to each type. Some types store all they need
72in the head, so don't have a body.
73
74In all but the most memory-paranoid configuations (ex: PURIFY), heads
75and bodies are allocated out of arenas, which by default are
76approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
77Sv-bodies are allocated by their sv-type, guaranteeing size
78consistency needed to allocate safely from arrays.
79
d2a0f284
JC
80For SV-heads, the first slot in each arena is reserved, and holds a
81link to the next arena, some flags, and a note of the number of slots.
82Snaked through each arena chain is a linked list of free items; when
83this becomes empty, an extra arena is allocated and divided up into N
84items which are threaded into the free list.
85
86SV-bodies are similar, but they use arena-sets by default, which
87separate the link and info from the arena itself, and reclaim the 1st
88slot in the arena. SV-bodies are further described later.
645c22ef
DM
89
90The following global variables are associated with arenas:
91
92 PL_sv_arenaroot pointer to list of SV arenas
93 PL_sv_root pointer to list of free SV structures
94
d2a0f284
JC
95 PL_body_arenas head of linked-list of body arenas
96 PL_body_roots[] array of pointers to list of free bodies of svtype
97 arrays are indexed by the svtype needed
93e68bfb 98
d2a0f284
JC
99A few special SV heads are not allocated from an arena, but are
100instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
101The size of arenas can be changed from the default by setting
102PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
103
104The SV arena serves the secondary purpose of allowing still-live SVs
105to be located and destroyed during final cleanup.
106
107At the lowest level, the macros new_SV() and del_SV() grab and free
108an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
109to return the SV to the free list with error checking.) new_SV() calls
110more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
111SVs in the free list have their SvTYPE field set to all ones.
112
ff276b08 113At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 114perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 115start of the interpreter.
645c22ef 116
645c22ef
DM
117The function visit() scans the SV arenas list, and calls a specified
118function for each SV it finds which is still live - ie which has an SvTYPE
119other than all 1's, and a non-zero SvREFCNT. visit() is used by the
120following functions (specified as [function that calls visit()] / [function
121called by visit() for each SV]):
122
123 sv_report_used() / do_report_used()
f2524eef 124 dump all remaining SVs (debugging aid)
645c22ef
DM
125
126 sv_clean_objs() / do_clean_objs(),do_clean_named_objs()
127 Attempt to free all objects pointed to by RVs,
128 and, unless DISABLE_DESTRUCTOR_KLUDGE is defined,
129 try to do the same for all objects indirectly
130 referenced by typeglobs too. Called once from
131 perl_destruct(), prior to calling sv_clean_all()
132 below.
133
134 sv_clean_all() / do_clean_all()
135 SvREFCNT_dec(sv) each remaining SV, possibly
136 triggering an sv_free(). It also sets the
137 SVf_BREAK flag on the SV to indicate that the
138 refcnt has been artificially lowered, and thus
139 stopping sv_free() from giving spurious warnings
140 about SVs which unexpectedly have a refcnt
141 of zero. called repeatedly from perl_destruct()
142 until there are no SVs left.
143
93e68bfb 144=head2 Arena allocator API Summary
645c22ef
DM
145
146Private API to rest of sv.c
147
148 new_SV(), del_SV(),
149
150 new_XIV(), del_XIV(),
151 new_XNV(), del_XNV(),
152 etc
153
154Public API:
155
8cf8f3d1 156 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 157
645c22ef
DM
158=cut
159
3e8320cc 160 * ========================================================================= */
645c22ef 161
4561caa4
CS
162/*
163 * "A time to plant, and a time to uproot what was planted..."
164 */
165
77354fb4 166void
de37a194 167Perl_offer_nice_chunk(pTHX_ void *const chunk, const U32 chunk_size)
77354fb4 168{
97aff369 169 dVAR;
77354fb4
NC
170 void *new_chunk;
171 U32 new_chunk_size;
7918f24d
NC
172
173 PERL_ARGS_ASSERT_OFFER_NICE_CHUNK;
174
77354fb4
NC
175 new_chunk = (void *)(chunk);
176 new_chunk_size = (chunk_size);
177 if (new_chunk_size > PL_nice_chunk_size) {
178 Safefree(PL_nice_chunk);
179 PL_nice_chunk = (char *) new_chunk;
180 PL_nice_chunk_size = new_chunk_size;
181 } else {
182 Safefree(chunk);
183 }
77354fb4 184}
cac9b346 185
d7a2c63c
MHM
186#ifdef PERL_MEM_LOG
187# define MEM_LOG_NEW_SV(sv, file, line, func) \
188 Perl_mem_log_new_sv(sv, file, line, func)
189# define MEM_LOG_DEL_SV(sv, file, line, func) \
190 Perl_mem_log_del_sv(sv, file, line, func)
191#else
192# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
193# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
194#endif
195
fd0854ff 196#ifdef DEBUG_LEAKING_SCALARS
22162ca8 197# define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
d7a2c63c
MHM
198# define DEBUG_SV_SERIAL(sv) \
199 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
200 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
201#else
202# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 203# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
204#endif
205
48614a46
NC
206#ifdef PERL_POISON
207# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 208# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
209/* Whilst I'd love to do this, it seems that things like to check on
210 unreferenced scalars
7e337ee0 211# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 212*/
7e337ee0
JH
213# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
214 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
215#else
216# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 217# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
218# define POSION_SV_HEAD(sv)
219#endif
220
990198f0
DM
221/* Mark an SV head as unused, and add to free list.
222 *
223 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
224 * its refcount artificially decremented during global destruction, so
225 * there may be dangling pointers to it. The last thing we want in that
226 * case is for it to be reused. */
227
053fc874
GS
228#define plant_SV(p) \
229 STMT_START { \
990198f0 230 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
231 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
232 DEBUG_SV_SERIAL(p); \
fd0854ff 233 FREE_SV_DEBUG_FILE(p); \
48614a46 234 POSION_SV_HEAD(p); \
053fc874 235 SvFLAGS(p) = SVTYPEMASK; \
990198f0 236 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 237 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
238 PL_sv_root = (p); \
239 } \
053fc874
GS
240 --PL_sv_count; \
241 } STMT_END
a0d0e21e 242
053fc874
GS
243#define uproot_SV(p) \
244 STMT_START { \
245 (p) = PL_sv_root; \
daba3364 246 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
247 ++PL_sv_count; \
248 } STMT_END
249
645c22ef 250
cac9b346
NC
251/* make some more SVs by adding another arena */
252
cac9b346
NC
253STATIC SV*
254S_more_sv(pTHX)
255{
97aff369 256 dVAR;
cac9b346
NC
257 SV* sv;
258
259 if (PL_nice_chunk) {
260 sv_add_arena(PL_nice_chunk, PL_nice_chunk_size, 0);
bd61b366 261 PL_nice_chunk = NULL;
cac9b346
NC
262 PL_nice_chunk_size = 0;
263 }
264 else {
265 char *chunk; /* must use New here to match call to */
d2a0f284 266 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
2e7ed132 267 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
268 }
269 uproot_SV(sv);
270 return sv;
271}
272
645c22ef
DM
273/* new_SV(): return a new, empty SV head */
274
eba0f806
DM
275#ifdef DEBUG_LEAKING_SCALARS
276/* provide a real function for a debugger to play with */
277STATIC SV*
d7a2c63c 278S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
279{
280 SV* sv;
281
eba0f806
DM
282 if (PL_sv_root)
283 uproot_SV(sv);
284 else
cac9b346 285 sv = S_more_sv(aTHX);
eba0f806
DM
286 SvANY(sv) = 0;
287 SvREFCNT(sv) = 1;
288 SvFLAGS(sv) = 0;
fd0854ff 289 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
290 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
291 ? PL_parser->copline
292 : PL_curcop
f24aceb1
DM
293 ? CopLINE(PL_curcop)
294 : 0
e385c3bf 295 );
fd0854ff
DM
296 sv->sv_debug_inpad = 0;
297 sv->sv_debug_cloned = 0;
fd0854ff 298 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
299
300 sv->sv_debug_serial = PL_sv_serial++;
301
302 MEM_LOG_NEW_SV(sv, file, line, func);
303 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
304 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
305
eba0f806
DM
306 return sv;
307}
d7a2c63c 308# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
309
310#else
311# define new_SV(p) \
053fc874 312 STMT_START { \
053fc874
GS
313 if (PL_sv_root) \
314 uproot_SV(p); \
315 else \
cac9b346 316 (p) = S_more_sv(aTHX); \
053fc874
GS
317 SvANY(p) = 0; \
318 SvREFCNT(p) = 1; \
319 SvFLAGS(p) = 0; \
d7a2c63c 320 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 321 } STMT_END
eba0f806 322#endif
463ee0b2 323
645c22ef
DM
324
325/* del_SV(): return an empty SV head to the free list */
326
a0d0e21e 327#ifdef DEBUGGING
4561caa4 328
053fc874
GS
329#define del_SV(p) \
330 STMT_START { \
aea4f609 331 if (DEBUG_D_TEST) \
053fc874
GS
332 del_sv(p); \
333 else \
334 plant_SV(p); \
053fc874 335 } STMT_END
a0d0e21e 336
76e3520e 337STATIC void
cea2e8a9 338S_del_sv(pTHX_ SV *p)
463ee0b2 339{
97aff369 340 dVAR;
7918f24d
NC
341
342 PERL_ARGS_ASSERT_DEL_SV;
343
aea4f609 344 if (DEBUG_D_TEST) {
4633a7c4 345 SV* sva;
a3b680e6 346 bool ok = 0;
daba3364 347 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
348 const SV * const sv = sva + 1;
349 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 350 if (p >= sv && p < svend) {
a0d0e21e 351 ok = 1;
c0ff570e
NC
352 break;
353 }
a0d0e21e
LW
354 }
355 if (!ok) {
9b387841
NC
356 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
357 "Attempt to free non-arena SV: 0x%"UVxf
358 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
359 return;
360 }
361 }
4561caa4 362 plant_SV(p);
463ee0b2 363}
a0d0e21e 364
4561caa4
CS
365#else /* ! DEBUGGING */
366
367#define del_SV(p) plant_SV(p)
368
369#endif /* DEBUGGING */
463ee0b2 370
645c22ef
DM
371
372/*
ccfc67b7
JH
373=head1 SV Manipulation Functions
374
645c22ef
DM
375=for apidoc sv_add_arena
376
377Given a chunk of memory, link it to the head of the list of arenas,
378and split it into a list of free SVs.
379
380=cut
381*/
382
d2bd4e7f
NC
383static void
384S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 385{
97aff369 386 dVAR;
daba3364 387 SV *const sva = MUTABLE_SV(ptr);
463ee0b2
LW
388 register SV* sv;
389 register SV* svend;
4633a7c4 390
7918f24d
NC
391 PERL_ARGS_ASSERT_SV_ADD_ARENA;
392
4633a7c4 393 /* The first SV in an arena isn't an SV. */
3280af22 394 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
395 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
396 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
397
3280af22
NIS
398 PL_sv_arenaroot = sva;
399 PL_sv_root = sva + 1;
4633a7c4
LW
400
401 svend = &sva[SvREFCNT(sva) - 1];
402 sv = sva + 1;
463ee0b2 403 while (sv < svend) {
3eef1deb 404 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 405#ifdef DEBUGGING
978b032e 406 SvREFCNT(sv) = 0;
03e36789 407#endif
4b69cbe3 408 /* Must always set typemask because it's always checked in on cleanup
03e36789 409 when the arenas are walked looking for objects. */
8990e307 410 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
411 sv++;
412 }
3eef1deb 413 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
414#ifdef DEBUGGING
415 SvREFCNT(sv) = 0;
416#endif
4633a7c4
LW
417 SvFLAGS(sv) = SVTYPEMASK;
418}
419
055972dc
DM
420/* visit(): call the named function for each non-free SV in the arenas
421 * whose flags field matches the flags/mask args. */
645c22ef 422
5226ed68 423STATIC I32
de37a194 424S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 425{
97aff369 426 dVAR;
4633a7c4 427 SV* sva;
5226ed68 428 I32 visited = 0;
8990e307 429
7918f24d
NC
430 PERL_ARGS_ASSERT_VISIT;
431
daba3364 432 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0 433 register const SV * const svend = &sva[SvREFCNT(sva)];
a3b680e6 434 register SV* sv;
4561caa4 435 for (sv = sva + 1; sv < svend; ++sv) {
055972dc
DM
436 if (SvTYPE(sv) != SVTYPEMASK
437 && (sv->sv_flags & mask) == flags
438 && SvREFCNT(sv))
439 {
acfe0abc 440 (FCALL)(aTHX_ sv);
5226ed68
JH
441 ++visited;
442 }
8990e307
LW
443 }
444 }
5226ed68 445 return visited;
8990e307
LW
446}
447
758a08c3
JH
448#ifdef DEBUGGING
449
645c22ef
DM
450/* called by sv_report_used() for each live SV */
451
452static void
5fa45a31 453do_report_used(pTHX_ SV *const sv)
645c22ef
DM
454{
455 if (SvTYPE(sv) != SVTYPEMASK) {
456 PerlIO_printf(Perl_debug_log, "****\n");
457 sv_dump(sv);
458 }
459}
758a08c3 460#endif
645c22ef
DM
461
462/*
463=for apidoc sv_report_used
464
465Dump the contents of all SVs not yet freed. (Debugging aid).
466
467=cut
468*/
469
8990e307 470void
864dbfa3 471Perl_sv_report_used(pTHX)
4561caa4 472{
ff270d3a 473#ifdef DEBUGGING
055972dc 474 visit(do_report_used, 0, 0);
96a5add6
AL
475#else
476 PERL_UNUSED_CONTEXT;
ff270d3a 477#endif
4561caa4
CS
478}
479
645c22ef
DM
480/* called by sv_clean_objs() for each live SV */
481
482static void
de37a194 483do_clean_objs(pTHX_ SV *const ref)
645c22ef 484{
97aff369 485 dVAR;
ea724faa
NC
486 assert (SvROK(ref));
487 {
823a54a3
AL
488 SV * const target = SvRV(ref);
489 if (SvOBJECT(target)) {
490 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
491 if (SvWEAKREF(ref)) {
492 sv_del_backref(target, ref);
493 SvWEAKREF_off(ref);
494 SvRV_set(ref, NULL);
495 } else {
496 SvROK_off(ref);
497 SvRV_set(ref, NULL);
498 SvREFCNT_dec(target);
499 }
645c22ef
DM
500 }
501 }
502
503 /* XXX Might want to check arrays, etc. */
504}
505
506/* called by sv_clean_objs() for each live SV */
507
508#ifndef DISABLE_DESTRUCTOR_KLUDGE
509static void
f30de749 510do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 511{
97aff369 512 dVAR;
ea724faa 513 assert(SvTYPE(sv) == SVt_PVGV);
d011219a
NC
514 assert(isGV_with_GP(sv));
515 if (GvGP(sv)) {
c69033f2
NC
516 if ((
517#ifdef PERL_DONT_CREATE_GVSV
518 GvSV(sv) &&
519#endif
520 SvOBJECT(GvSV(sv))) ||
645c22ef
DM
521 (GvAV(sv) && SvOBJECT(GvAV(sv))) ||
522 (GvHV(sv) && SvOBJECT(GvHV(sv))) ||
9c12f1e5
RGS
523 /* In certain rare cases GvIOp(sv) can be NULL, which would make SvOBJECT(GvIO(sv)) dereference NULL. */
524 (GvIO(sv) ? (SvFLAGS(GvIOp(sv)) & SVs_OBJECT) : 0) ||
645c22ef
DM
525 (GvCV(sv) && SvOBJECT(GvCV(sv))) )
526 {
527 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning named glob object:\n "), sv_dump(sv)));
ec5f3c78 528 SvFLAGS(sv) |= SVf_BREAK;
645c22ef
DM
529 SvREFCNT_dec(sv);
530 }
531 }
532}
533#endif
534
535/*
536=for apidoc sv_clean_objs
537
538Attempt to destroy all objects not yet freed
539
540=cut
541*/
542
4561caa4 543void
864dbfa3 544Perl_sv_clean_objs(pTHX)
4561caa4 545{
97aff369 546 dVAR;
3280af22 547 PL_in_clean_objs = TRUE;
055972dc 548 visit(do_clean_objs, SVf_ROK, SVf_ROK);
4561caa4 549#ifndef DISABLE_DESTRUCTOR_KLUDGE
2d0f3c12 550 /* some barnacles may yet remain, clinging to typeglobs */
d011219a 551 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4561caa4 552#endif
3280af22 553 PL_in_clean_objs = FALSE;
4561caa4
CS
554}
555
645c22ef
DM
556/* called by sv_clean_all() for each live SV */
557
558static void
de37a194 559do_clean_all(pTHX_ SV *const sv)
645c22ef 560{
97aff369 561 dVAR;
daba3364 562 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 563 /* don't clean pid table and strtab */
d17ea597 564 return;
cddfcddc 565 }
645c22ef
DM
566 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
567 SvFLAGS(sv) |= SVf_BREAK;
568 SvREFCNT_dec(sv);
569}
570
571/*
572=for apidoc sv_clean_all
573
574Decrement the refcnt of each remaining SV, possibly triggering a
575cleanup. This function may have to be called multiple times to free
ff276b08 576SVs which are in complex self-referential hierarchies.
645c22ef
DM
577
578=cut
579*/
580
5226ed68 581I32
864dbfa3 582Perl_sv_clean_all(pTHX)
8990e307 583{
97aff369 584 dVAR;
5226ed68 585 I32 cleaned;
3280af22 586 PL_in_clean_all = TRUE;
055972dc 587 cleaned = visit(do_clean_all, 0,0);
3280af22 588 PL_in_clean_all = FALSE;
5226ed68 589 return cleaned;
8990e307 590}
463ee0b2 591
5e258f8c
JC
592/*
593 ARENASETS: a meta-arena implementation which separates arena-info
594 into struct arena_set, which contains an array of struct
595 arena_descs, each holding info for a single arena. By separating
596 the meta-info from the arena, we recover the 1st slot, formerly
597 borrowed for list management. The arena_set is about the size of an
39244528 598 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
599
600 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
601 memory in the last arena-set (1/2 on average). In trade, we get
602 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 603 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
604 small arenas for large, rare body types, by changing array* fields
605 in body_details_by_type[] below.
5e258f8c 606*/
5e258f8c 607struct arena_desc {
398c677b
NC
608 char *arena; /* the raw storage, allocated aligned */
609 size_t size; /* its size ~4k typ */
e5973ed5 610 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
611};
612
e6148039
NC
613struct arena_set;
614
615/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 616 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
617 therefore likely to be 1 aligned memory page. */
618
619#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
620 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
621
622struct arena_set {
623 struct arena_set* next;
0a848332
NC
624 unsigned int set_size; /* ie ARENAS_PER_SET */
625 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
626 struct arena_desc set[ARENAS_PER_SET];
627};
628
645c22ef
DM
629/*
630=for apidoc sv_free_arenas
631
632Deallocate the memory used by all arenas. Note that all the individual SV
633heads and bodies within the arenas must already have been freed.
634
635=cut
636*/
4633a7c4 637void
864dbfa3 638Perl_sv_free_arenas(pTHX)
4633a7c4 639{
97aff369 640 dVAR;
4633a7c4
LW
641 SV* sva;
642 SV* svanext;
0a848332 643 unsigned int i;
4633a7c4
LW
644
645 /* Free arenas here, but be careful about fake ones. (We assume
646 contiguity of the fake ones with the corresponding real ones.) */
647
3280af22 648 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 649 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 650 while (svanext && SvFAKE(svanext))
daba3364 651 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
652
653 if (!SvFAKE(sva))
1df70142 654 Safefree(sva);
4633a7c4 655 }
93e68bfb 656
5e258f8c 657 {
0a848332
NC
658 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
659
660 while (aroot) {
661 struct arena_set *current = aroot;
662 i = aroot->curr;
663 while (i--) {
5e258f8c
JC
664 assert(aroot->set[i].arena);
665 Safefree(aroot->set[i].arena);
666 }
0a848332
NC
667 aroot = aroot->next;
668 Safefree(current);
5e258f8c
JC
669 }
670 }
dc8220bf 671 PL_body_arenas = 0;
fdda85ca 672
0a848332
NC
673 i = PERL_ARENA_ROOTS_SIZE;
674 while (i--)
93e68bfb 675 PL_body_roots[i] = 0;
93e68bfb 676
43c5f42d 677 Safefree(PL_nice_chunk);
bd61b366 678 PL_nice_chunk = NULL;
3280af22
NIS
679 PL_nice_chunk_size = 0;
680 PL_sv_arenaroot = 0;
681 PL_sv_root = 0;
4633a7c4
LW
682}
683
bd81e77b
NC
684/*
685 Here are mid-level routines that manage the allocation of bodies out
686 of the various arenas. There are 5 kinds of arenas:
29489e7c 687
bd81e77b
NC
688 1. SV-head arenas, which are discussed and handled above
689 2. regular body arenas
690 3. arenas for reduced-size bodies
691 4. Hash-Entry arenas
29489e7c 692
bd81e77b
NC
693 Arena types 2 & 3 are chained by body-type off an array of
694 arena-root pointers, which is indexed by svtype. Some of the
695 larger/less used body types are malloced singly, since a large
696 unused block of them is wasteful. Also, several svtypes dont have
697 bodies; the data fits into the sv-head itself. The arena-root
698 pointer thus has a few unused root-pointers (which may be hijacked
699 later for arena types 4,5)
29489e7c 700
bd81e77b
NC
701 3 differs from 2 as an optimization; some body types have several
702 unused fields in the front of the structure (which are kept in-place
703 for consistency). These bodies can be allocated in smaller chunks,
704 because the leading fields arent accessed. Pointers to such bodies
705 are decremented to point at the unused 'ghost' memory, knowing that
706 the pointers are used with offsets to the real memory.
29489e7c 707
bd81e77b
NC
708 HE, HEK arenas are managed separately, with separate code, but may
709 be merge-able later..
bd81e77b
NC
710*/
711
bd26d9a3 712/* get_arena(size): this creates custom-sized arenas
5e258f8c
JC
713 TBD: export properly for hv.c: S_more_he().
714*/
715void*
e5973ed5 716Perl_get_arena(pTHX_ const size_t arena_size, const svtype bodytype)
5e258f8c 717{
7a89be66 718 dVAR;
5e258f8c 719 struct arena_desc* adesc;
39244528 720 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
0a848332 721 unsigned int curr;
5e258f8c 722
476a1e16
JC
723 /* shouldnt need this
724 if (!arena_size) arena_size = PERL_ARENA_SIZE;
725 */
5e258f8c
JC
726
727 /* may need new arena-set to hold new arena */
39244528
NC
728 if (!aroot || aroot->curr >= aroot->set_size) {
729 struct arena_set *newroot;
5e258f8c
JC
730 Newxz(newroot, 1, struct arena_set);
731 newroot->set_size = ARENAS_PER_SET;
39244528
NC
732 newroot->next = aroot;
733 aroot = newroot;
734 PL_body_arenas = (void *) newroot;
52944de8 735 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
5e258f8c
JC
736 }
737
738 /* ok, now have arena-set with at least 1 empty/available arena-desc */
39244528
NC
739 curr = aroot->curr++;
740 adesc = &(aroot->set[curr]);
5e258f8c
JC
741 assert(!adesc->arena);
742
89086707 743 Newx(adesc->arena, arena_size, char);
5e258f8c 744 adesc->size = arena_size;
e5973ed5 745 adesc->utype = bodytype;
d67b3c53
JH
746 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
747 curr, (void*)adesc->arena, (UV)arena_size));
5e258f8c
JC
748
749 return adesc->arena;
5e258f8c
JC
750}
751
53c1dcc0 752
bd81e77b 753/* return a thing to the free list */
29489e7c 754
bd81e77b
NC
755#define del_body(thing, root) \
756 STMT_START { \
00b6aa41 757 void ** const thing_copy = (void **)thing;\
bd81e77b
NC
758 *thing_copy = *root; \
759 *root = (void*)thing_copy; \
bd81e77b 760 } STMT_END
29489e7c 761
bd81e77b 762/*
d2a0f284
JC
763
764=head1 SV-Body Allocation
765
766Allocation of SV-bodies is similar to SV-heads, differing as follows;
767the allocation mechanism is used for many body types, so is somewhat
768more complicated, it uses arena-sets, and has no need for still-live
769SV detection.
770
771At the outermost level, (new|del)_X*V macros return bodies of the
772appropriate type. These macros call either (new|del)_body_type or
773(new|del)_body_allocated macro pairs, depending on specifics of the
774type. Most body types use the former pair, the latter pair is used to
775allocate body types with "ghost fields".
776
777"ghost fields" are fields that are unused in certain types, and
69ba284b 778consequently don't need to actually exist. They are declared because
d2a0f284
JC
779they're part of a "base type", which allows use of functions as
780methods. The simplest examples are AVs and HVs, 2 aggregate types
781which don't use the fields which support SCALAR semantics.
782
69ba284b 783For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
784chunks, we thus avoid wasted memory for those unaccessed members.
785When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 786size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
787structure. (But things will all go boom if you write to the part that
788is "not there", because you'll be overwriting the last members of the
789preceding structure in memory.)
790
69ba284b
NC
791We calculate the correction using the STRUCT_OFFSET macro on the first
792member present. If the allocated structure is smaller (no initial NV
793actually allocated) then the net effect is to subtract the size of the NV
794from the pointer, to return a new pointer as if an initial NV were actually
795allocated. (We were using structures named *_allocated for this, but
796this turned out to be a subtle bug, because a structure without an NV
797could have a lower alignment constraint, but the compiler is allowed to
798optimised accesses based on the alignment constraint of the actual pointer
799to the full structure, for example, using a single 64 bit load instruction
800because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284
JC
801
802This is the same trick as was used for NV and IV bodies. Ironically it
803doesn't need to be used for NV bodies any more, because NV is now at
804the start of the structure. IV bodies don't need it either, because
805they are no longer allocated.
806
807In turn, the new_body_* allocators call S_new_body(), which invokes
808new_body_inline macro, which takes a lock, and takes a body off the
809linked list at PL_body_roots[sv_type], calling S_more_bodies() if
810necessary to refresh an empty list. Then the lock is released, and
811the body is returned.
812
813S_more_bodies calls get_arena(), and carves it up into an array of N
814bodies, which it strings into a linked list. It looks up arena-size
815and body-size from the body_details table described below, thus
816supporting the multiple body-types.
817
818If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
819the (new|del)_X*V macros are mapped directly to malloc/free.
820
821*/
822
823/*
824
825For each sv-type, struct body_details bodies_by_type[] carries
826parameters which control these aspects of SV handling:
827
828Arena_size determines whether arenas are used for this body type, and if
829so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
830zero, forcing individual mallocs and frees.
831
832Body_size determines how big a body is, and therefore how many fit into
833each arena. Offset carries the body-pointer adjustment needed for
69ba284b 834"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
835
836But its main purpose is to parameterize info needed in
837Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 838vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
839are used for this, except for arena_size.
840
841For the sv-types that have no bodies, arenas are not used, so those
842PL_body_roots[sv_type] are unused, and can be overloaded. In
843something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 844PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 845bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 846available in hv.c.
d2a0f284 847
29489e7c
DM
848*/
849
bd81e77b 850struct body_details {
0fb58b32 851 U8 body_size; /* Size to allocate */
10666ae3 852 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 853 U8 offset;
10666ae3
NC
854 unsigned int type : 4; /* We have space for a sanity check. */
855 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
856 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
857 unsigned int arena : 1; /* Allocated from an arena */
858 size_t arena_size; /* Size of arena to allocate */
bd81e77b 859};
29489e7c 860
bd81e77b
NC
861#define HADNV FALSE
862#define NONV TRUE
29489e7c 863
d2a0f284 864
bd81e77b
NC
865#ifdef PURIFY
866/* With -DPURFIY we allocate everything directly, and don't use arenas.
867 This seems a rather elegant way to simplify some of the code below. */
868#define HASARENA FALSE
869#else
870#define HASARENA TRUE
871#endif
872#define NOARENA FALSE
29489e7c 873
d2a0f284
JC
874/* Size the arenas to exactly fit a given number of bodies. A count
875 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
876 simplifying the default. If count > 0, the arena is sized to fit
877 only that many bodies, allowing arenas to be used for large, rare
878 bodies (XPVFM, XPVIO) without undue waste. The arena size is
879 limited by PERL_ARENA_SIZE, so we can safely oversize the
880 declarations.
881 */
95db5f15
MB
882#define FIT_ARENA0(body_size) \
883 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
884#define FIT_ARENAn(count,body_size) \
885 ( count * body_size <= PERL_ARENA_SIZE) \
886 ? count * body_size \
887 : FIT_ARENA0 (body_size)
888#define FIT_ARENA(count,body_size) \
889 count \
890 ? FIT_ARENAn (count, body_size) \
891 : FIT_ARENA0 (body_size)
d2a0f284 892
bd81e77b
NC
893/* Calculate the length to copy. Specifically work out the length less any
894 final padding the compiler needed to add. See the comment in sv_upgrade
895 for why copying the padding proved to be a bug. */
29489e7c 896
bd81e77b
NC
897#define copy_length(type, last_member) \
898 STRUCT_OFFSET(type, last_member) \
daba3364 899 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 900
bd81e77b 901static const struct body_details bodies_by_type[] = {
10666ae3
NC
902 { sizeof(HE), 0, 0, SVt_NULL,
903 FALSE, NONV, NOARENA, FIT_ARENA(0, sizeof(HE)) },
d2a0f284 904
1cb9cd50 905 /* The bind placeholder pretends to be an RV for now.
c6f8b1d0 906 Also it's marked as "can't upgrade" to stop anyone using it before it's
1cb9cd50
NC
907 implemented. */
908 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
909
db93c0c4
NC
910 /* IVs are in the head, so the allocation size is 0. */
911 { 0,
d2a0f284 912 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 913 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 914 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
915 },
916
bd81e77b 917 /* 8 bytes on most ILP32 with IEEE doubles */
6e128786
NC
918 { sizeof(NV), sizeof(NV),
919 STRUCT_OFFSET(XPVNV, xnv_u),
920 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 921
bd81e77b 922 /* 8 bytes on most ILP32 with IEEE doubles */
601dfd0a 923 { sizeof(XPV),
889d28b2
NC
924 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
925 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 926 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 927 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 928
889d28b2 929#if 2 *PTRSIZE <= IVSIZE
bd81e77b 930 /* 12 */
601dfd0a 931 { sizeof(XPVIV),
889d28b2
NC
932 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
933 + STRUCT_OFFSET(XPV, xpv_cur),
934 SVt_PVIV, FALSE, NONV, HASARENA,
935 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
936 /* 12 */
937#else
938 { sizeof(XPVIV),
601dfd0a
NC
939 copy_length(XPVIV, xiv_u),
940 0,
69ba284b 941 SVt_PVIV, FALSE, NONV, HASARENA,
889d28b2
NC
942 FIT_ARENA(0, sizeof(XPVIV)) },
943#endif
d2a0f284 944
889d28b2
NC
945#if (2 *PTRSIZE <= IVSIZE) && (2 *PTRSIZE <= NVSIZE)
946 /* 20 */
947 { sizeof(XPVNV),
948 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_PVNV, FALSE, HADNV, HASARENA,
951 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
952#else
bd81e77b 953 /* 20 */
6e128786 954 { sizeof(XPVNV), copy_length(XPVNV, xnv_u), 0, SVt_PVNV, FALSE, HADNV,
d2a0f284 955 HASARENA, FIT_ARENA(0, sizeof(XPVNV)) },
889d28b2 956#endif
d2a0f284 957
bd81e77b 958 /* 28 */
6e128786 959 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 960 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 961
288b8c02 962 /* something big */
601dfd0a
NC
963 { sizeof(regexp),
964 sizeof(regexp),
965 0,
08e44740 966 SVt_REGEXP, FALSE, NONV, HASARENA,
b6f60916 967 FIT_ARENA(0, sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur))
5c35adbb 968 },
4df7f6af 969
bd81e77b 970 /* 48 */
10666ae3 971 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
972 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
973
bd81e77b 974 /* 64 */
10666ae3 975 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
976 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
977
601dfd0a 978 { sizeof(XPVAV),
4f7003f5 979 copy_length(XPVAV, xav_alloc),
601dfd0a 980 0,
69ba284b 981 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 982 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 983
601dfd0a 984 { sizeof(XPVHV),
de0a224a 985 copy_length(XPVHV, xhv_keys),
601dfd0a 986 0,
69ba284b 987 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 988 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 989
c84c4652 990 /* 56 */
601dfd0a
NC
991 { sizeof(XPVCV),
992 sizeof(XPVCV),
993 0,
69ba284b 994 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 995 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 996
601dfd0a
NC
997 { sizeof(XPVFM),
998 sizeof(XPVFM),
999 0,
69ba284b 1000 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 1001 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284
JC
1002
1003 /* XPVIO is 84 bytes, fits 48x */
601dfd0a
NC
1004 { sizeof(XPVIO),
1005 sizeof(XPVIO),
1006 0,
b6f60916 1007 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 1008 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 1009};
29489e7c 1010
bd81e77b 1011#define new_body_allocated(sv_type) \
d2a0f284 1012 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1013 - bodies_by_type[sv_type].offset)
29489e7c 1014
bd81e77b
NC
1015#define del_body_allocated(p, sv_type) \
1016 del_body(p + bodies_by_type[sv_type].offset, &PL_body_roots[sv_type])
29489e7c 1017
29489e7c 1018
bd81e77b
NC
1019#define my_safemalloc(s) (void*)safemalloc(s)
1020#define my_safecalloc(s) (void*)safecalloc(s, 1)
1021#define my_safefree(p) safefree((char*)p)
29489e7c 1022
bd81e77b 1023#ifdef PURIFY
29489e7c 1024
bd81e77b
NC
1025#define new_XNV() my_safemalloc(sizeof(XPVNV))
1026#define del_XNV(p) my_safefree(p)
29489e7c 1027
bd81e77b
NC
1028#define new_XPVNV() my_safemalloc(sizeof(XPVNV))
1029#define del_XPVNV(p) my_safefree(p)
29489e7c 1030
bd81e77b
NC
1031#define new_XPVAV() my_safemalloc(sizeof(XPVAV))
1032#define del_XPVAV(p) my_safefree(p)
29489e7c 1033
bd81e77b
NC
1034#define new_XPVHV() my_safemalloc(sizeof(XPVHV))
1035#define del_XPVHV(p) my_safefree(p)
29489e7c 1036
bd81e77b
NC
1037#define new_XPVMG() my_safemalloc(sizeof(XPVMG))
1038#define del_XPVMG(p) my_safefree(p)
29489e7c 1039
bd81e77b
NC
1040#define new_XPVGV() my_safemalloc(sizeof(XPVGV))
1041#define del_XPVGV(p) my_safefree(p)
29489e7c 1042
bd81e77b 1043#else /* !PURIFY */
29489e7c 1044
65ac1738
NC
1045#define new_XNV() new_body_allocated(SVt_NV)
1046#define del_XNV(p) del_body_allocated(p, SVt_NV)
29489e7c 1047
65ac1738
NC
1048#define new_XPVNV() new_body_allocated(SVt_PVNV)
1049#define del_XPVNV(p) del_body_allocated(p, SVt_PVNV)
29489e7c 1050
bd81e77b
NC
1051#define new_XPVAV() new_body_allocated(SVt_PVAV)
1052#define del_XPVAV(p) del_body_allocated(p, SVt_PVAV)
645c22ef 1053
bd81e77b
NC
1054#define new_XPVHV() new_body_allocated(SVt_PVHV)
1055#define del_XPVHV(p) del_body_allocated(p, SVt_PVHV)
645c22ef 1056
65ac1738
NC
1057#define new_XPVMG() new_body_allocated(SVt_PVMG)
1058#define del_XPVMG(p) del_body_allocated(p, SVt_PVMG)
645c22ef 1059
65ac1738
NC
1060#define new_XPVGV() new_body_allocated(SVt_PVGV)
1061#define del_XPVGV(p) del_body_allocated(p, SVt_PVGV)
1d7c1841 1062
bd81e77b 1063#endif /* PURIFY */
93e68bfb 1064
bd81e77b 1065/* no arena for you! */
93e68bfb 1066
bd81e77b 1067#define new_NOARENA(details) \
d2a0f284 1068 my_safemalloc((details)->body_size + (details)->offset)
bd81e77b 1069#define new_NOARENAZ(details) \
d2a0f284
JC
1070 my_safecalloc((details)->body_size + (details)->offset)
1071
1072STATIC void *
de37a194 1073S_more_bodies (pTHX_ const svtype sv_type)
d2a0f284
JC
1074{
1075 dVAR;
1076 void ** const root = &PL_body_roots[sv_type];
96a5add6 1077 const struct body_details * const bdp = &bodies_by_type[sv_type];
d2a0f284
JC
1078 const size_t body_size = bdp->body_size;
1079 char *start;
1080 const char *end;
d8fca402 1081 const size_t arena_size = Perl_malloc_good_size(bdp->arena_size);
0b2d3faa 1082#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1083 static bool done_sanity_check;
1084
0b2d3faa
JH
1085 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1086 * variables like done_sanity_check. */
10666ae3 1087 if (!done_sanity_check) {
ea471437 1088 unsigned int i = SVt_LAST;
10666ae3
NC
1089
1090 done_sanity_check = TRUE;
1091
1092 while (i--)
1093 assert (bodies_by_type[i].type == i);
1094 }
1095#endif
1096
23e9d66c
NC
1097 assert(bdp->arena_size);
1098
d8fca402 1099 start = (char*) Perl_get_arena(aTHX_ arena_size, sv_type);
d2a0f284 1100
d8fca402 1101 end = start + arena_size - 2 * body_size;
d2a0f284 1102
d2a0f284 1103 /* computed count doesnt reflect the 1st slot reservation */
d8fca402
NC
1104#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1105 DEBUG_m(PerlIO_printf(Perl_debug_log,
1106 "arena %p end %p arena-size %d (from %d) type %d "
1107 "size %d ct %d\n",
1108 (void*)start, (void*)end, (int)arena_size,
1109 (int)bdp->arena_size, sv_type, (int)body_size,
1110 (int)arena_size / (int)body_size));
1111#else
d2a0f284
JC
1112 DEBUG_m(PerlIO_printf(Perl_debug_log,
1113 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1114 (void*)start, (void*)end,
0e84aef4
JH
1115 (int)bdp->arena_size, sv_type, (int)body_size,
1116 (int)bdp->arena_size / (int)body_size));
d8fca402 1117#endif
d2a0f284
JC
1118 *root = (void *)start;
1119
d8fca402 1120 while (start <= end) {
d2a0f284
JC
1121 char * const next = start + body_size;
1122 *(void**) start = (void *)next;
1123 start = next;
1124 }
1125 *(void **)start = 0;
1126
1127 return *root;
1128}
1129
1130/* grab a new thing from the free list, allocating more if necessary.
1131 The inline version is used for speed in hot routines, and the
1132 function using it serves the rest (unless PURIFY).
1133*/
1134#define new_body_inline(xpv, sv_type) \
1135 STMT_START { \
1136 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775
DD
1137 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1138 ? *((void **)(r3wt)) : more_bodies(sv_type)); \
d2a0f284 1139 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1140 } STMT_END
1141
1142#ifndef PURIFY
1143
1144STATIC void *
de37a194 1145S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1146{
1147 dVAR;
1148 void *xpv;
1149 new_body_inline(xpv, sv_type);
1150 return xpv;
1151}
1152
1153#endif
93e68bfb 1154
238b27b3
NC
1155static const struct body_details fake_rv =
1156 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1157
bd81e77b
NC
1158/*
1159=for apidoc sv_upgrade
93e68bfb 1160
bd81e77b
NC
1161Upgrade an SV to a more complex form. Generally adds a new body type to the
1162SV, then copies across as much information as possible from the old body.
1163You generally want to use the C<SvUPGRADE> macro wrapper. See also C<svtype>.
93e68bfb 1164
bd81e77b 1165=cut
93e68bfb 1166*/
93e68bfb 1167
bd81e77b 1168void
aad570aa 1169Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
cac9b346 1170{
97aff369 1171 dVAR;
bd81e77b
NC
1172 void* old_body;
1173 void* new_body;
42d0e0b7 1174 const svtype old_type = SvTYPE(sv);
d2a0f284 1175 const struct body_details *new_type_details;
238b27b3 1176 const struct body_details *old_type_details
bd81e77b 1177 = bodies_by_type + old_type;
4df7f6af 1178 SV *referant = NULL;
cac9b346 1179
7918f24d
NC
1180 PERL_ARGS_ASSERT_SV_UPGRADE;
1181
1776cbe8
NC
1182 if (old_type == new_type)
1183 return;
1184
1185 /* This clause was purposefully added ahead of the early return above to
1186 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1187 inference by Nick I-S that it would fix other troublesome cases. See
1188 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1189
1190 Given that shared hash key scalars are no longer PVIV, but PV, there is
1191 no longer need to unshare so as to free up the IVX slot for its proper
1192 purpose. So it's safe to move the early return earlier. */
1193
bd81e77b
NC
1194 if (new_type != SVt_PV && SvIsCOW(sv)) {
1195 sv_force_normal_flags(sv, 0);
1196 }
cac9b346 1197
bd81e77b 1198 old_body = SvANY(sv);
de042e1d 1199
bd81e77b
NC
1200 /* Copying structures onto other structures that have been neatly zeroed
1201 has a subtle gotcha. Consider XPVMG
cac9b346 1202
bd81e77b
NC
1203 +------+------+------+------+------+-------+-------+
1204 | NV | CUR | LEN | IV | MAGIC | STASH |
1205 +------+------+------+------+------+-------+-------+
1206 0 4 8 12 16 20 24 28
645c22ef 1207
bd81e77b
NC
1208 where NVs are aligned to 8 bytes, so that sizeof that structure is
1209 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1210
bd81e77b
NC
1211 +------+------+------+------+------+-------+-------+------+
1212 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1213 +------+------+------+------+------+-------+-------+------+
1214 0 4 8 12 16 20 24 28 32
08742458 1215
bd81e77b 1216 so what happens if you allocate memory for this structure:
30f9da9e 1217
bd81e77b
NC
1218 +------+------+------+------+------+-------+-------+------+------+...
1219 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1220 +------+------+------+------+------+-------+-------+------+------+...
1221 0 4 8 12 16 20 24 28 32 36
bfc44f79 1222
bd81e77b
NC
1223 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1224 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1225 started out as zero once, but it's quite possible that it isn't. So now,
1226 rather than a nicely zeroed GP, you have it pointing somewhere random.
1227 Bugs ensue.
bfc44f79 1228
bd81e77b
NC
1229 (In fact, GP ends up pointing at a previous GP structure, because the
1230 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1231 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1232 this happens to be moot because XPVGV has been re-ordered, with GP
1233 no longer after STASH)
30f9da9e 1234
bd81e77b
NC
1235 So we are careful and work out the size of used parts of all the
1236 structures. */
bfc44f79 1237
bd81e77b
NC
1238 switch (old_type) {
1239 case SVt_NULL:
1240 break;
1241 case SVt_IV:
4df7f6af
NC
1242 if (SvROK(sv)) {
1243 referant = SvRV(sv);
238b27b3
NC
1244 old_type_details = &fake_rv;
1245 if (new_type == SVt_NV)
1246 new_type = SVt_PVNV;
4df7f6af
NC
1247 } else {
1248 if (new_type < SVt_PVIV) {
1249 new_type = (new_type == SVt_NV)
1250 ? SVt_PVNV : SVt_PVIV;
1251 }
bd81e77b
NC
1252 }
1253 break;
1254 case SVt_NV:
1255 if (new_type < SVt_PVNV) {
1256 new_type = SVt_PVNV;
bd81e77b
NC
1257 }
1258 break;
bd81e77b
NC
1259 case SVt_PV:
1260 assert(new_type > SVt_PV);
1261 assert(SVt_IV < SVt_PV);
1262 assert(SVt_NV < SVt_PV);
1263 break;
1264 case SVt_PVIV:
1265 break;
1266 case SVt_PVNV:
1267 break;
1268 case SVt_PVMG:
1269 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1270 there's no way that it can be safely upgraded, because perl.c
1271 expects to Safefree(SvANY(PL_mess_sv)) */
1272 assert(sv != PL_mess_sv);
1273 /* This flag bit is used to mean other things in other scalar types.
1274 Given that it only has meaning inside the pad, it shouldn't be set
1275 on anything that can get upgraded. */
00b1698f 1276 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1277 break;
1278 default:
1279 if (old_type_details->cant_upgrade)
c81225bc
NC
1280 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1281 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1282 }
3376de98
NC
1283
1284 if (old_type > new_type)
1285 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1286 (int)old_type, (int)new_type);
1287
2fa1109b 1288 new_type_details = bodies_by_type + new_type;
645c22ef 1289
bd81e77b
NC
1290 SvFLAGS(sv) &= ~SVTYPEMASK;
1291 SvFLAGS(sv) |= new_type;
932e9ff9 1292
ab4416c0
NC
1293 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1294 the return statements above will have triggered. */
1295 assert (new_type != SVt_NULL);
bd81e77b 1296 switch (new_type) {
bd81e77b
NC
1297 case SVt_IV:
1298 assert(old_type == SVt_NULL);
1299 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1300 SvIV_set(sv, 0);
1301 return;
1302 case SVt_NV:
1303 assert(old_type == SVt_NULL);
1304 SvANY(sv) = new_XNV();
1305 SvNV_set(sv, 0);
1306 return;
bd81e77b 1307 case SVt_PVHV:
bd81e77b 1308 case SVt_PVAV:
d2a0f284 1309 assert(new_type_details->body_size);
c1ae03ae
NC
1310
1311#ifndef PURIFY
1312 assert(new_type_details->arena);
d2a0f284 1313 assert(new_type_details->arena_size);
c1ae03ae 1314 /* This points to the start of the allocated area. */
d2a0f284
JC
1315 new_body_inline(new_body, new_type);
1316 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1317 new_body = ((char *)new_body) - new_type_details->offset;
1318#else
1319 /* We always allocated the full length item with PURIFY. To do this
1320 we fake things so that arena is false for all 16 types.. */
1321 new_body = new_NOARENAZ(new_type_details);
1322#endif
1323 SvANY(sv) = new_body;
1324 if (new_type == SVt_PVAV) {
1325 AvMAX(sv) = -1;
1326 AvFILLp(sv) = -1;
1327 AvREAL_only(sv);
64484faa 1328 if (old_type_details->body_size) {
ac572bf4
NC
1329 AvALLOC(sv) = 0;
1330 } else {
1331 /* It will have been zeroed when the new body was allocated.
1332 Lets not write to it, in case it confuses a write-back
1333 cache. */
1334 }
78ac7dd9
NC
1335 } else {
1336 assert(!SvOK(sv));
1337 SvOK_off(sv);
1338#ifndef NODEFAULT_SHAREKEYS
1339 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1340#endif
1341 HvMAX(sv) = 7; /* (start with 8 buckets) */
64484faa 1342 if (old_type_details->body_size) {
78ac7dd9
NC
1343 HvFILL(sv) = 0;
1344 } else {
1345 /* It will have been zeroed when the new body was allocated.
1346 Lets not write to it, in case it confuses a write-back
1347 cache. */
1348 }
c1ae03ae 1349 }
aeb18a1e 1350
bd81e77b
NC
1351 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1352 The target created by newSVrv also is, and it can have magic.
1353 However, it never has SvPVX set.
1354 */
4df7f6af
NC
1355 if (old_type == SVt_IV) {
1356 assert(!SvROK(sv));
1357 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1358 assert(SvPVX_const(sv) == 0);
1359 }
aeb18a1e 1360
bd81e77b 1361 if (old_type >= SVt_PVMG) {
e736a858 1362 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1363 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1364 } else {
1365 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1366 }
1367 break;
93e68bfb 1368
93e68bfb 1369
b9ad13ac
NC
1370 case SVt_REGEXP:
1371 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
1372 sv_force_normal_flags(sv) is called. */
1373 SvFAKE_on(sv);
bd81e77b
NC
1374 case SVt_PVIV:
1375 /* XXX Is this still needed? Was it ever needed? Surely as there is
1376 no route from NV to PVIV, NOK can never be true */
1377 assert(!SvNOKp(sv));
1378 assert(!SvNOK(sv));
1379 case SVt_PVIO:
1380 case SVt_PVFM:
bd81e77b
NC
1381 case SVt_PVGV:
1382 case SVt_PVCV:
1383 case SVt_PVLV:
1384 case SVt_PVMG:
1385 case SVt_PVNV:
1386 case SVt_PV:
93e68bfb 1387
d2a0f284 1388 assert(new_type_details->body_size);
bd81e77b
NC
1389 /* We always allocated the full length item with PURIFY. To do this
1390 we fake things so that arena is false for all 16 types.. */
1391 if(new_type_details->arena) {
1392 /* This points to the start of the allocated area. */
d2a0f284
JC
1393 new_body_inline(new_body, new_type);
1394 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1395 new_body = ((char *)new_body) - new_type_details->offset;
1396 } else {
1397 new_body = new_NOARENAZ(new_type_details);
1398 }
1399 SvANY(sv) = new_body;
5e2fc214 1400
bd81e77b 1401 if (old_type_details->copy) {
f9ba3d20
NC
1402 /* There is now the potential for an upgrade from something without
1403 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1404 int offset = old_type_details->offset;
1405 int length = old_type_details->copy;
1406
1407 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1408 const int difference
f9ba3d20
NC
1409 = new_type_details->offset - old_type_details->offset;
1410 offset += difference;
1411 length -= difference;
1412 }
1413 assert (length >= 0);
1414
1415 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1416 char);
bd81e77b
NC
1417 }
1418
1419#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1420 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1421 * correct 0.0 for us. Otherwise, if the old body didn't have an
1422 * NV slot, but the new one does, then we need to initialise the
1423 * freshly created NV slot with whatever the correct bit pattern is
1424 * for 0.0 */
e22a937e
NC
1425 if (old_type_details->zero_nv && !new_type_details->zero_nv
1426 && !isGV_with_GP(sv))
bd81e77b 1427 SvNV_set(sv, 0);
82048762 1428#endif
5e2fc214 1429
85dca89a
NC
1430 if (new_type == SVt_PVIO) {
1431 IO * const io = MUTABLE_IO(sv);
d963bf01 1432 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1433
1434 SvOBJECT_on(io);
1435 /* Clear the stashcache because a new IO could overrule a package
1436 name */
1437 hv_clear(PL_stashcache);
1438
85dca89a 1439 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1440 IoPAGE_LEN(sv) = 60;
85dca89a 1441 }
4df7f6af
NC
1442 if (old_type < SVt_PV) {
1443 /* referant will be NULL unless the old type was SVt_IV emulating
1444 SVt_RV */
1445 sv->sv_u.svu_rv = referant;
1446 }
bd81e77b
NC
1447 break;
1448 default:
afd78fd5
JH
1449 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1450 (unsigned long)new_type);
bd81e77b 1451 }
73171d91 1452
db93c0c4 1453 if (old_type > SVt_IV) {
bd81e77b
NC
1454#ifdef PURIFY
1455 my_safefree(old_body);
1456#else
bc786448
GG
1457 /* Note that there is an assumption that all bodies of types that
1458 can be upgraded came from arenas. Only the more complex non-
1459 upgradable types are allowed to be directly malloc()ed. */
1460 assert(old_type_details->arena);
bd81e77b
NC
1461 del_body((void*)((char*)old_body + old_type_details->offset),
1462 &PL_body_roots[old_type]);
1463#endif
1464 }
1465}
73171d91 1466
bd81e77b
NC
1467/*
1468=for apidoc sv_backoff
73171d91 1469
bd81e77b
NC
1470Remove any string offset. You should normally use the C<SvOOK_off> macro
1471wrapper instead.
73171d91 1472
bd81e77b 1473=cut
73171d91
NC
1474*/
1475
bd81e77b 1476int
aad570aa 1477Perl_sv_backoff(pTHX_ register SV *const sv)
bd81e77b 1478{
69240efd 1479 STRLEN delta;
7a4bba22 1480 const char * const s = SvPVX_const(sv);
7918f24d
NC
1481
1482 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1483 PERL_UNUSED_CONTEXT;
7918f24d 1484
bd81e77b
NC
1485 assert(SvOOK(sv));
1486 assert(SvTYPE(sv) != SVt_PVHV);
1487 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1488
69240efd
NC
1489 SvOOK_offset(sv, delta);
1490
7a4bba22
NC
1491 SvLEN_set(sv, SvLEN(sv) + delta);
1492 SvPV_set(sv, SvPVX(sv) - delta);
1493 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1494 SvFLAGS(sv) &= ~SVf_OOK;
1495 return 0;
1496}
73171d91 1497
bd81e77b
NC
1498/*
1499=for apidoc sv_grow
73171d91 1500
bd81e77b
NC
1501Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1502upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1503Use the C<SvGROW> wrapper instead.
93e68bfb 1504
bd81e77b
NC
1505=cut
1506*/
93e68bfb 1507
bd81e77b 1508char *
aad570aa 1509Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
bd81e77b
NC
1510{
1511 register char *s;
93e68bfb 1512
7918f24d
NC
1513 PERL_ARGS_ASSERT_SV_GROW;
1514
5db06880
NC
1515 if (PL_madskills && newlen >= 0x100000) {
1516 PerlIO_printf(Perl_debug_log,
1517 "Allocation too large: %"UVxf"\n", (UV)newlen);
1518 }
bd81e77b
NC
1519#ifdef HAS_64K_LIMIT
1520 if (newlen >= 0x10000) {
1521 PerlIO_printf(Perl_debug_log,
1522 "Allocation too large: %"UVxf"\n", (UV)newlen);
1523 my_exit(1);
1524 }
1525#endif /* HAS_64K_LIMIT */
1526 if (SvROK(sv))
1527 sv_unref(sv);
1528 if (SvTYPE(sv) < SVt_PV) {
1529 sv_upgrade(sv, SVt_PV);
1530 s = SvPVX_mutable(sv);
1531 }
1532 else if (SvOOK(sv)) { /* pv is offset? */
1533 sv_backoff(sv);
1534 s = SvPVX_mutable(sv);
1535 if (newlen > SvLEN(sv))
1536 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1537#ifdef HAS_64K_LIMIT
1538 if (newlen >= 0x10000)
1539 newlen = 0xFFFF;
1540#endif
1541 }
1542 else
1543 s = SvPVX_mutable(sv);
aeb18a1e 1544
bd81e77b 1545 if (newlen > SvLEN(sv)) { /* need more room? */
aedff202 1546#ifndef Perl_safesysmalloc_size
bd81e77b 1547 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1548#endif
98653f18 1549 if (SvLEN(sv) && s) {
10edeb5d 1550 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1551 }
1552 else {
10edeb5d 1553 s = (char*)safemalloc(newlen);
bd81e77b
NC
1554 if (SvPVX_const(sv) && SvCUR(sv)) {
1555 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1556 }
1557 }
1558 SvPV_set(sv, s);
ca7c1a29 1559#ifdef Perl_safesysmalloc_size
98653f18
NC
1560 /* Do this here, do it once, do it right, and then we will never get
1561 called back into sv_grow() unless there really is some growing
1562 needed. */
ca7c1a29 1563 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1564#else
bd81e77b 1565 SvLEN_set(sv, newlen);
98653f18 1566#endif
bd81e77b
NC
1567 }
1568 return s;
1569}
aeb18a1e 1570
bd81e77b
NC
1571/*
1572=for apidoc sv_setiv
932e9ff9 1573
bd81e77b
NC
1574Copies an integer into the given SV, upgrading first if necessary.
1575Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1576
bd81e77b
NC
1577=cut
1578*/
463ee0b2 1579
bd81e77b 1580void
aad570aa 1581Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
bd81e77b 1582{
97aff369 1583 dVAR;
7918f24d
NC
1584
1585 PERL_ARGS_ASSERT_SV_SETIV;
1586
bd81e77b
NC
1587 SV_CHECK_THINKFIRST_COW_DROP(sv);
1588 switch (SvTYPE(sv)) {
1589 case SVt_NULL:
bd81e77b 1590 case SVt_NV:
3376de98 1591 sv_upgrade(sv, SVt_IV);
bd81e77b 1592 break;
bd81e77b
NC
1593 case SVt_PV:
1594 sv_upgrade(sv, SVt_PVIV);
1595 break;
463ee0b2 1596
bd81e77b 1597 case SVt_PVGV:
6e592b3a
BM
1598 if (!isGV_with_GP(sv))
1599 break;
bd81e77b
NC
1600 case SVt_PVAV:
1601 case SVt_PVHV:
1602 case SVt_PVCV:
1603 case SVt_PVFM:
1604 case SVt_PVIO:
1605 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1606 OP_DESC(PL_op));
42d0e0b7 1607 default: NOOP;
bd81e77b
NC
1608 }
1609 (void)SvIOK_only(sv); /* validate number */
1610 SvIV_set(sv, i);
1611 SvTAINT(sv);
1612}
932e9ff9 1613
bd81e77b
NC
1614/*
1615=for apidoc sv_setiv_mg
d33b2eba 1616
bd81e77b 1617Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1618
bd81e77b
NC
1619=cut
1620*/
d33b2eba 1621
bd81e77b 1622void
aad570aa 1623Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
bd81e77b 1624{
7918f24d
NC
1625 PERL_ARGS_ASSERT_SV_SETIV_MG;
1626
bd81e77b
NC
1627 sv_setiv(sv,i);
1628 SvSETMAGIC(sv);
1629}
727879eb 1630
bd81e77b
NC
1631/*
1632=for apidoc sv_setuv
d33b2eba 1633
bd81e77b
NC
1634Copies an unsigned integer into the given SV, upgrading first if necessary.
1635Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1636
bd81e77b
NC
1637=cut
1638*/
d33b2eba 1639
bd81e77b 1640void
aad570aa 1641Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
bd81e77b 1642{
7918f24d
NC
1643 PERL_ARGS_ASSERT_SV_SETUV;
1644
bd81e77b
NC
1645 /* With these two if statements:
1646 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1647
bd81e77b
NC
1648 without
1649 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1650
bd81e77b
NC
1651 If you wish to remove them, please benchmark to see what the effect is
1652 */
1653 if (u <= (UV)IV_MAX) {
1654 sv_setiv(sv, (IV)u);
1655 return;
1656 }
1657 sv_setiv(sv, 0);
1658 SvIsUV_on(sv);
1659 SvUV_set(sv, u);
1660}
d33b2eba 1661
bd81e77b
NC
1662/*
1663=for apidoc sv_setuv_mg
727879eb 1664
bd81e77b 1665Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1666
bd81e77b
NC
1667=cut
1668*/
5e2fc214 1669
bd81e77b 1670void
aad570aa 1671Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
bd81e77b 1672{
7918f24d
NC
1673 PERL_ARGS_ASSERT_SV_SETUV_MG;
1674
bd81e77b
NC
1675 sv_setuv(sv,u);
1676 SvSETMAGIC(sv);
1677}
5e2fc214 1678
954c1994 1679/*
bd81e77b 1680=for apidoc sv_setnv
954c1994 1681
bd81e77b
NC
1682Copies a double into the given SV, upgrading first if necessary.
1683Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1684
1685=cut
1686*/
1687
63f97190 1688void
aad570aa 1689Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
79072805 1690{
97aff369 1691 dVAR;
7918f24d
NC
1692
1693 PERL_ARGS_ASSERT_SV_SETNV;
1694
bd81e77b
NC
1695 SV_CHECK_THINKFIRST_COW_DROP(sv);
1696 switch (SvTYPE(sv)) {
79072805 1697 case SVt_NULL:
79072805 1698 case SVt_IV:
bd81e77b 1699 sv_upgrade(sv, SVt_NV);
79072805
LW
1700 break;
1701 case SVt_PV:
79072805 1702 case SVt_PVIV:
bd81e77b 1703 sv_upgrade(sv, SVt_PVNV);
79072805 1704 break;
bd4b1eb5 1705
bd4b1eb5 1706 case SVt_PVGV:
6e592b3a
BM
1707 if (!isGV_with_GP(sv))
1708 break;
bd81e77b
NC
1709 case SVt_PVAV:
1710 case SVt_PVHV:
79072805 1711 case SVt_PVCV:
bd81e77b
NC
1712 case SVt_PVFM:
1713 case SVt_PVIO:
1714 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1715 OP_DESC(PL_op));
42d0e0b7 1716 default: NOOP;
2068cd4d 1717 }
bd81e77b
NC
1718 SvNV_set(sv, num);
1719 (void)SvNOK_only(sv); /* validate number */
1720 SvTAINT(sv);
79072805
LW
1721}
1722
645c22ef 1723/*
bd81e77b 1724=for apidoc sv_setnv_mg
645c22ef 1725
bd81e77b 1726Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1727
1728=cut
1729*/
1730
bd81e77b 1731void
aad570aa 1732Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
79072805 1733{
7918f24d
NC
1734 PERL_ARGS_ASSERT_SV_SETNV_MG;
1735
bd81e77b
NC
1736 sv_setnv(sv,num);
1737 SvSETMAGIC(sv);
79072805
LW
1738}
1739
bd81e77b
NC
1740/* Print an "isn't numeric" warning, using a cleaned-up,
1741 * printable version of the offending string
1742 */
954c1994 1743
bd81e77b 1744STATIC void
aad570aa 1745S_not_a_number(pTHX_ SV *const sv)
79072805 1746{
97aff369 1747 dVAR;
bd81e77b
NC
1748 SV *dsv;
1749 char tmpbuf[64];
1750 const char *pv;
94463019 1751
7918f24d
NC
1752 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1753
94463019 1754 if (DO_UTF8(sv)) {
84bafc02 1755 dsv = newSVpvs_flags("", SVs_TEMP);
94463019
JH
1756 pv = sv_uni_display(dsv, sv, 10, 0);
1757 } else {
1758 char *d = tmpbuf;
551405c4 1759 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1760 /* each *s can expand to 4 chars + "...\0",
1761 i.e. need room for 8 chars */
ecdeb87c 1762
00b6aa41
AL
1763 const char *s = SvPVX_const(sv);
1764 const char * const end = s + SvCUR(sv);
1765 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1766 int ch = *s & 0xFF;
1767 if (ch & 128 && !isPRINT_LC(ch)) {
1768 *d++ = 'M';
1769 *d++ = '-';
1770 ch &= 127;
1771 }
1772 if (ch == '\n') {
1773 *d++ = '\\';
1774 *d++ = 'n';
1775 }
1776 else if (ch == '\r') {
1777 *d++ = '\\';
1778 *d++ = 'r';
1779 }
1780 else if (ch == '\f') {
1781 *d++ = '\\';
1782 *d++ = 'f';
1783 }
1784 else if (ch == '\\') {
1785 *d++ = '\\';
1786 *d++ = '\\';
1787 }
1788 else if (ch == '\0') {
1789 *d++ = '\\';
1790 *d++ = '0';
1791 }
1792 else if (isPRINT_LC(ch))
1793 *d++ = ch;
1794 else {
1795 *d++ = '^';
1796 *d++ = toCTRL(ch);
1797 }
1798 }
1799 if (s < end) {
1800 *d++ = '.';
1801 *d++ = '.';
1802 *d++ = '.';
1803 }
1804 *d = '\0';
1805 pv = tmpbuf;
a0d0e21e 1806 }
a0d0e21e 1807
533c011a 1808 if (PL_op)
9014280d 1809 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019
JH
1810 "Argument \"%s\" isn't numeric in %s", pv,
1811 OP_DESC(PL_op));
a0d0e21e 1812 else
9014280d 1813 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019 1814 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1815}
1816
c2988b20
NC
1817/*
1818=for apidoc looks_like_number
1819
645c22ef
DM
1820Test if the content of an SV looks like a number (or is a number).
1821C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1822non-numeric warning), even if your atof() doesn't grok them.
c2988b20
NC
1823
1824=cut
1825*/
1826
1827I32
aad570aa 1828Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1829{
a3b680e6 1830 register const char *sbegin;
c2988b20
NC
1831 STRLEN len;
1832
7918f24d
NC
1833 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1834
c2988b20 1835 if (SvPOK(sv)) {
3f7c398e 1836 sbegin = SvPVX_const(sv);
c2988b20
NC
1837 len = SvCUR(sv);
1838 }
1839 else if (SvPOKp(sv))
83003860 1840 sbegin = SvPV_const(sv, len);
c2988b20 1841 else
e0ab1c0e 1842 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1843 return grok_number(sbegin, len, NULL);
1844}
25da4f38 1845
19f6321d
NC
1846STATIC bool
1847S_glob_2number(pTHX_ GV * const gv)
180488f8
NC
1848{
1849 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1850 SV *const buffer = sv_newmortal();
1851
7918f24d
NC
1852 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1853
180488f8
NC
1854 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1855 is on. */
1856 SvFAKE_off(gv);
1857 gv_efullname3(buffer, gv, "*");
1858 SvFLAGS(gv) |= wasfake;
1859
675c862f
AL
1860 /* We know that all GVs stringify to something that is not-a-number,
1861 so no need to test that. */
1862 if (ckWARN(WARN_NUMERIC))
1863 not_a_number(buffer);
1864 /* We just want something true to return, so that S_sv_2iuv_common
1865 can tail call us and return true. */
19f6321d 1866 return TRUE;
675c862f
AL
1867}
1868
25da4f38
IZ
1869/* Actually, ISO C leaves conversion of UV to IV undefined, but
1870 until proven guilty, assume that things are not that bad... */
1871
645c22ef
DM
1872/*
1873 NV_PRESERVES_UV:
1874
1875 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1876 an IV (an assumption perl has been based on to date) it becomes necessary
1877 to remove the assumption that the NV always carries enough precision to
1878 recreate the IV whenever needed, and that the NV is the canonical form.
1879 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1880 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1881 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1882 1) to distinguish between IV/UV/NV slots that have cached a valid
1883 conversion where precision was lost and IV/UV/NV slots that have a
1884 valid conversion which has lost no precision
645c22ef 1885 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1886 would lose precision, the precise conversion (or differently
1887 imprecise conversion) is also performed and cached, to prevent
1888 requests for different numeric formats on the same SV causing
1889 lossy conversion chains. (lossless conversion chains are perfectly
1890 acceptable (still))
1891
1892
1893 flags are used:
1894 SvIOKp is true if the IV slot contains a valid value
1895 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1896 SvNOKp is true if the NV slot contains a valid value
1897 SvNOK is true only if the NV value is accurate
1898
1899 so
645c22ef 1900 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1901 IV(or UV) would lose accuracy over a direct conversion from PV to
1902 IV(or UV). If it would, cache both conversions, return NV, but mark
1903 SV as IOK NOKp (ie not NOK).
1904
645c22ef 1905 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1906 NV would lose accuracy over a direct conversion from PV to NV. If it
1907 would, cache both conversions, flag similarly.
1908
1909 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1910 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1911 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1912 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1913 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1914
645c22ef
DM
1915 The benefit of this is that operations such as pp_add know that if
1916 SvIOK is true for both left and right operands, then integer addition
1917 can be used instead of floating point (for cases where the result won't
1918 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1919 loss of precision compared with integer addition.
1920
1921 * making IV and NV equal status should make maths accurate on 64 bit
1922 platforms
1923 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1924 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1925 looking for SvIOK and checking for overflow will not outweigh the
1926 fp to integer speedup)
1927 * will slow down integer operations (callers of SvIV) on "inaccurate"
1928 values, as the change from SvIOK to SvIOKp will cause a call into
1929 sv_2iv each time rather than a macro access direct to the IV slot
1930 * should speed up number->string conversion on integers as IV is
645c22ef 1931 favoured when IV and NV are equally accurate
28e5dec8
JH
1932
1933 ####################################################################
645c22ef
DM
1934 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1935 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1936 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1937 ####################################################################
1938
645c22ef 1939 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1940 performance ratio.
1941*/
1942
1943#ifndef NV_PRESERVES_UV
645c22ef
DM
1944# define IS_NUMBER_UNDERFLOW_IV 1
1945# define IS_NUMBER_UNDERFLOW_UV 2
1946# define IS_NUMBER_IV_AND_UV 2
1947# define IS_NUMBER_OVERFLOW_IV 4
1948# define IS_NUMBER_OVERFLOW_UV 5
1949
1950/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1951
1952/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1953STATIC int
5de3775c 1954S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1955# ifdef DEBUGGING
1956 , I32 numtype
1957# endif
1958 )
28e5dec8 1959{
97aff369 1960 dVAR;
7918f24d
NC
1961
1962 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1963
3f7c398e 1964 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
1965 if (SvNVX(sv) < (NV)IV_MIN) {
1966 (void)SvIOKp_on(sv);
1967 (void)SvNOK_on(sv);
45977657 1968 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1969 return IS_NUMBER_UNDERFLOW_IV;
1970 }
1971 if (SvNVX(sv) > (NV)UV_MAX) {
1972 (void)SvIOKp_on(sv);
1973 (void)SvNOK_on(sv);
1974 SvIsUV_on(sv);
607fa7f2 1975 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1976 return IS_NUMBER_OVERFLOW_UV;
1977 }
c2988b20
NC
1978 (void)SvIOKp_on(sv);
1979 (void)SvNOK_on(sv);
1980 /* Can't use strtol etc to convert this string. (See truth table in
1981 sv_2iv */
1982 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1983 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1984 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1985 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1986 } else {
1987 /* Integer is imprecise. NOK, IOKp */
1988 }
1989 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1990 }
1991 SvIsUV_on(sv);
607fa7f2 1992 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1993 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1994 if (SvUVX(sv) == UV_MAX) {
1995 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1996 possibly be preserved by NV. Hence, it must be overflow.
1997 NOK, IOKp */
1998 return IS_NUMBER_OVERFLOW_UV;
1999 }
2000 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2001 } else {
2002 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2003 }
c2988b20 2004 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2005}
645c22ef
DM
2006#endif /* !NV_PRESERVES_UV*/
2007
af359546 2008STATIC bool
7918f24d
NC
2009S_sv_2iuv_common(pTHX_ SV *const sv)
2010{
97aff369 2011 dVAR;
7918f24d
NC
2012
2013 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2014
af359546 2015 if (SvNOKp(sv)) {
28e5dec8
JH
2016 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2017 * without also getting a cached IV/UV from it at the same time
2018 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2019 * IV or UV at same time to avoid this. */
2020 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2021
2022 if (SvTYPE(sv) == SVt_NV)
2023 sv_upgrade(sv, SVt_PVNV);
2024
28e5dec8
JH
2025 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2026 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2027 certainly cast into the IV range at IV_MAX, whereas the correct
2028 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2029 cases go to UV */
cab190d4
JD
2030#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2031 if (Perl_isnan(SvNVX(sv))) {
2032 SvUV_set(sv, 0);
2033 SvIsUV_on(sv);
fdbe6d7c 2034 return FALSE;
cab190d4 2035 }
cab190d4 2036#endif
28e5dec8 2037 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2038 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2039 if (SvNVX(sv) == (NV) SvIVX(sv)
2040#ifndef NV_PRESERVES_UV
2041 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2042 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2043 /* Don't flag it as "accurately an integer" if the number
2044 came from a (by definition imprecise) NV operation, and
2045 we're outside the range of NV integer precision */
2046#endif
2047 ) {
a43d94f2
NC
2048 if (SvNOK(sv))
2049 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2050 else {
2051 /* scalar has trailing garbage, eg "42a" */
2052 }
28e5dec8 2053 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2054 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2055 PTR2UV(sv),
2056 SvNVX(sv),
2057 SvIVX(sv)));
2058
2059 } else {
2060 /* IV not precise. No need to convert from PV, as NV
2061 conversion would already have cached IV if it detected
2062 that PV->IV would be better than PV->NV->IV
2063 flags already correct - don't set public IOK. */
2064 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2065 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2066 PTR2UV(sv),
2067 SvNVX(sv),
2068 SvIVX(sv)));
2069 }
2070 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2071 but the cast (NV)IV_MIN rounds to a the value less (more
2072 negative) than IV_MIN which happens to be equal to SvNVX ??
2073 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2074 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2075 (NV)UVX == NVX are both true, but the values differ. :-(
2076 Hopefully for 2s complement IV_MIN is something like
2077 0x8000000000000000 which will be exact. NWC */
d460ef45 2078 }
25da4f38 2079 else {
607fa7f2 2080 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2081 if (
2082 (SvNVX(sv) == (NV) SvUVX(sv))
2083#ifndef NV_PRESERVES_UV
2084 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2085 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2086 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2087 /* Don't flag it as "accurately an integer" if the number
2088 came from a (by definition imprecise) NV operation, and
2089 we're outside the range of NV integer precision */
2090#endif
a43d94f2 2091 && SvNOK(sv)
28e5dec8
JH
2092 )
2093 SvIOK_on(sv);
25da4f38 2094 SvIsUV_on(sv);
1c846c1f 2095 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2096 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2097 PTR2UV(sv),
57def98f
JH
2098 SvUVX(sv),
2099 SvUVX(sv)));
25da4f38 2100 }
748a9306
LW
2101 }
2102 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2103 UV value;
504618e9 2104 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2105 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2106 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2107 the same as the direct translation of the initial string
2108 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2109 be careful to ensure that the value with the .456 is around if the
2110 NV value is requested in the future).
1c846c1f 2111
af359546 2112 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2113 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2114 cache the NV if we are sure it's not needed.
25da4f38 2115 */
16b7a9a4 2116
c2988b20
NC
2117 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2118 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2119 == IS_NUMBER_IN_UV) {
5e045b90 2120 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2121 if (SvTYPE(sv) < SVt_PVIV)
2122 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2123 (void)SvIOK_on(sv);
c2988b20
NC
2124 } else if (SvTYPE(sv) < SVt_PVNV)
2125 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2126
f2524eef 2127 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2128 we aren't going to call atof() below. If NVs don't preserve UVs
2129 then the value returned may have more precision than atof() will
2130 return, even though value isn't perfectly accurate. */
2131 if ((numtype & (IS_NUMBER_IN_UV
2132#ifdef NV_PRESERVES_UV
2133 | IS_NUMBER_NOT_INT
2134#endif
2135 )) == IS_NUMBER_IN_UV) {
2136 /* This won't turn off the public IOK flag if it was set above */
2137 (void)SvIOKp_on(sv);
2138
2139 if (!(numtype & IS_NUMBER_NEG)) {
2140 /* positive */;
2141 if (value <= (UV)IV_MAX) {
45977657 2142 SvIV_set(sv, (IV)value);
c2988b20 2143 } else {
af359546 2144 /* it didn't overflow, and it was positive. */
607fa7f2 2145 SvUV_set(sv, value);
c2988b20
NC
2146 SvIsUV_on(sv);
2147 }
2148 } else {
2149 /* 2s complement assumption */
2150 if (value <= (UV)IV_MIN) {
45977657 2151 SvIV_set(sv, -(IV)value);
c2988b20
NC
2152 } else {
2153 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2154 I'm assuming it will be rare. */
c2988b20
NC
2155 if (SvTYPE(sv) < SVt_PVNV)
2156 sv_upgrade(sv, SVt_PVNV);
2157 SvNOK_on(sv);
2158 SvIOK_off(sv);
2159 SvIOKp_on(sv);
9d6ce603 2160 SvNV_set(sv, -(NV)value);
45977657 2161 SvIV_set(sv, IV_MIN);
c2988b20
NC
2162 }
2163 }
2164 }
2165 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2166 will be in the previous block to set the IV slot, and the next
2167 block to set the NV slot. So no else here. */
2168
2169 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2170 != IS_NUMBER_IN_UV) {
2171 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2172 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2173
c2988b20
NC
2174 if (! numtype && ckWARN(WARN_NUMERIC))
2175 not_a_number(sv);
28e5dec8 2176
65202027 2177#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2178 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2179 PTR2UV(sv), SvNVX(sv)));
65202027 2180#else
1779d84d 2181 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2182 PTR2UV(sv), SvNVX(sv)));
65202027 2183#endif
28e5dec8 2184
28e5dec8 2185#ifdef NV_PRESERVES_UV
af359546
NC
2186 (void)SvIOKp_on(sv);
2187 (void)SvNOK_on(sv);
2188 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2189 SvIV_set(sv, I_V(SvNVX(sv)));
2190 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2191 SvIOK_on(sv);
2192 } else {
6f207bd3 2193 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2194 }
2195 /* UV will not work better than IV */
2196 } else {
2197 if (SvNVX(sv) > (NV)UV_MAX) {
2198 SvIsUV_on(sv);
2199 /* Integer is inaccurate. NOK, IOKp, is UV */
2200 SvUV_set(sv, UV_MAX);
af359546
NC
2201 } else {
2202 SvUV_set(sv, U_V(SvNVX(sv)));
2203 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2204 NV preservse UV so can do correct comparison. */
2205 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2206 SvIOK_on(sv);
af359546 2207 } else {
6f207bd3 2208 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2209 }
2210 }
4b0c9573 2211 SvIsUV_on(sv);
af359546 2212 }
28e5dec8 2213#else /* NV_PRESERVES_UV */
c2988b20
NC
2214 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2215 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2216 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2217 grok_number above. The NV slot has just been set using
2218 Atof. */
560b0c46 2219 SvNOK_on(sv);
c2988b20
NC
2220 assert (SvIOKp(sv));
2221 } else {
2222 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2223 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2224 /* Small enough to preserve all bits. */
2225 (void)SvIOKp_on(sv);
2226 SvNOK_on(sv);
45977657 2227 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2228 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2229 SvIOK_on(sv);
2230 /* Assumption: first non-preserved integer is < IV_MAX,
2231 this NV is in the preserved range, therefore: */
2232 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2233 < (UV)IV_MAX)) {
32fdb065 2234 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
2235 }
2236 } else {
2237 /* IN_UV NOT_INT
2238 0 0 already failed to read UV.
2239 0 1 already failed to read UV.
2240 1 0 you won't get here in this case. IV/UV
2241 slot set, public IOK, Atof() unneeded.
2242 1 1 already read UV.
2243 so there's no point in sv_2iuv_non_preserve() attempting
2244 to use atol, strtol, strtoul etc. */
47031da6 2245# ifdef DEBUGGING
40a17c4c 2246 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2247# else
2248 sv_2iuv_non_preserve (sv);
2249# endif
c2988b20
NC
2250 }
2251 }
28e5dec8 2252#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2253 /* It might be more code efficient to go through the entire logic above
2254 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2255 gets complex and potentially buggy, so more programmer efficient
2256 to do it this way, by turning off the public flags: */
2257 if (!numtype)
2258 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2259 }
af359546
NC
2260 }
2261 else {
675c862f 2262 if (isGV_with_GP(sv))
159b6efe 2263 return glob_2number(MUTABLE_GV(sv));
180488f8 2264
af359546
NC
2265 if (!(SvFLAGS(sv) & SVs_PADTMP)) {
2266 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2267 report_uninit(sv);
2268 }
25da4f38
IZ
2269 if (SvTYPE(sv) < SVt_IV)
2270 /* Typically the caller expects that sv_any is not NULL now. */
2271 sv_upgrade(sv, SVt_IV);
af359546
NC
2272 /* Return 0 from the caller. */
2273 return TRUE;
2274 }
2275 return FALSE;
2276}
2277
2278/*
2279=for apidoc sv_2iv_flags
2280
2281Return the integer value of an SV, doing any necessary string
2282conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2283Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2284
2285=cut
2286*/
2287
2288IV
5de3775c 2289Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2290{
97aff369 2291 dVAR;
af359546 2292 if (!sv)
a0d0e21e 2293 return 0;
cecf5685
NC
2294 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2295 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e
NC
2296 cache IVs just in case. In practice it seems that they never
2297 actually anywhere accessible by user Perl code, let alone get used
2298 in anything other than a string context. */
af359546
NC
2299 if (flags & SV_GMAGIC)
2300 mg_get(sv);
2301 if (SvIOKp(sv))
2302 return SvIVX(sv);
2303 if (SvNOKp(sv)) {
2304 return I_V(SvNVX(sv));
2305 }
71c558c3
NC
2306 if (SvPOKp(sv) && SvLEN(sv)) {
2307 UV value;
2308 const int numtype
2309 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2310
2311 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2312 == IS_NUMBER_IN_UV) {
2313 /* It's definitely an integer */
2314 if (numtype & IS_NUMBER_NEG) {
2315 if (value < (UV)IV_MIN)
2316 return -(IV)value;
2317 } else {
2318 if (value < (UV)IV_MAX)
2319 return (IV)value;
2320 }
2321 }
2322 if (!numtype) {
2323 if (ckWARN(WARN_NUMERIC))
2324 not_a_number(sv);
2325 }
2326 return I_V(Atof(SvPVX_const(sv)));
2327 }
1c7ff15e
NC
2328 if (SvROK(sv)) {
2329 goto return_rok;
af359546 2330 }
1c7ff15e
NC
2331 assert(SvTYPE(sv) >= SVt_PVMG);
2332 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2333 } else if (SvTHINKFIRST(sv)) {
af359546 2334 if (SvROK(sv)) {
1c7ff15e 2335 return_rok:
af359546 2336 if (SvAMAGIC(sv)) {
aee036bb
DM
2337 SV * tmpstr;
2338 if (flags & SV_SKIP_OVERLOAD)
2339 return 0;
2340 tmpstr=AMG_CALLun(sv,numer);
af359546
NC
2341 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2342 return SvIV(tmpstr);
2343 }
2344 }
2345 return PTR2IV(SvRV(sv));
2346 }
2347 if (SvIsCOW(sv)) {
2348 sv_force_normal_flags(sv, 0);
2349 }
2350 if (SvREADONLY(sv) && !SvOK(sv)) {
2351 if (ckWARN(WARN_UNINITIALIZED))
2352 report_uninit(sv);
2353 return 0;
2354 }
2355 }
2356 if (!SvIOKp(sv)) {
2357 if (S_sv_2iuv_common(aTHX_ sv))
2358 return 0;
79072805 2359 }
1d7c1841
GS
2360 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2361 PTR2UV(sv),SvIVX(sv)));
25da4f38 2362 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2363}
2364
645c22ef 2365/*
891f9566 2366=for apidoc sv_2uv_flags
645c22ef
DM
2367
2368Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2369conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2370Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2371
2372=cut
2373*/
2374
ff68c719 2375UV
5de3775c 2376Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2377{
97aff369 2378 dVAR;
ff68c719
PP
2379 if (!sv)
2380 return 0;
cecf5685
NC
2381 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2382 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2383 cache IVs just in case. */
891f9566
YST
2384 if (flags & SV_GMAGIC)
2385 mg_get(sv);
ff68c719
PP
2386 if (SvIOKp(sv))
2387 return SvUVX(sv);
2388 if (SvNOKp(sv))
2389 return U_V(SvNVX(sv));
71c558c3
NC
2390 if (SvPOKp(sv) && SvLEN(sv)) {
2391 UV value;
2392 const int numtype
2393 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
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 }
2405 return U_V(Atof(SvPVX_const(sv)));
2406 }
1c7ff15e
NC
2407 if (SvROK(sv)) {
2408 goto return_rok;
3fe9a6f1 2409 }
1c7ff15e
NC
2410 assert(SvTYPE(sv) >= SVt_PVMG);
2411 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2412 } else if (SvTHINKFIRST(sv)) {
ff68c719 2413 if (SvROK(sv)) {
1c7ff15e 2414 return_rok:
deb46114 2415 if (SvAMAGIC(sv)) {
aee036bb
DM
2416 SV *tmpstr;
2417 if (flags & SV_SKIP_OVERLOAD)
2418 return 0;
2419 tmpstr = AMG_CALLun(sv,numer);
deb46114
NC
2420 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2421 return SvUV(tmpstr);
2422 }
2423 }
2424 return PTR2UV(SvRV(sv));
ff68c719 2425 }
765f542d
NC
2426 if (SvIsCOW(sv)) {
2427 sv_force_normal_flags(sv, 0);
8a818333 2428 }
0336b60e 2429 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2430 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2431 report_uninit(sv);
ff68c719
PP
2432 return 0;
2433 }
2434 }
af359546
NC
2435 if (!SvIOKp(sv)) {
2436 if (S_sv_2iuv_common(aTHX_ sv))
2437 return 0;
ff68c719 2438 }
25da4f38 2439
1d7c1841
GS
2440 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2441 PTR2UV(sv),SvUVX(sv)));
25da4f38 2442 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2443}
2444
645c22ef
DM
2445/*
2446=for apidoc sv_2nv
2447
2448Return the num value of an SV, doing any necessary string or integer
39d5de13
DM
2449conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2450Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2451
2452=cut
2453*/
2454
65202027 2455NV
39d5de13 2456Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
79072805 2457{
97aff369 2458 dVAR;
79072805
LW
2459 if (!sv)
2460 return 0.0;
cecf5685
NC
2461 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2462 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2463 cache IVs just in case. */
39d5de13
DM
2464 if (flags & SV_GMAGIC)
2465 mg_get(sv);
463ee0b2
LW
2466 if (SvNOKp(sv))
2467 return SvNVX(sv);
0aa395f8 2468 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
041457d9 2469 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
504618e9 2470 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
a0d0e21e 2471 not_a_number(sv);
3f7c398e 2472 return Atof(SvPVX_const(sv));
a0d0e21e 2473 }
25da4f38 2474 if (SvIOKp(sv)) {
1c846c1f 2475 if (SvIsUV(sv))
65202027 2476 return (NV)SvUVX(sv);
25da4f38 2477 else
65202027 2478 return (NV)SvIVX(sv);
47a72cb8
NC
2479 }
2480 if (SvROK(sv)) {
2481 goto return_rok;
2482 }
2483 assert(SvTYPE(sv) >= SVt_PVMG);
2484 /* This falls through to the report_uninit near the end of the
2485 function. */
2486 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2487 if (SvROK(sv)) {
47a72cb8 2488 return_rok:
deb46114 2489 if (SvAMAGIC(sv)) {
aee036bb
DM
2490 SV *tmpstr;
2491 if (flags & SV_SKIP_OVERLOAD)
2492 return 0;
2493 tmpstr = AMG_CALLun(sv,numer);
deb46114
NC
2494 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2495 return SvNV(tmpstr);
2496 }
2497 }
2498 return PTR2NV(SvRV(sv));
a0d0e21e 2499 }
765f542d
NC
2500 if (SvIsCOW(sv)) {
2501 sv_force_normal_flags(sv, 0);
8a818333 2502 }
0336b60e 2503 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2504 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2505 report_uninit(sv);
ed6116ce
LW
2506 return 0.0;
2507 }
79072805
LW
2508 }
2509 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2510 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2511 sv_upgrade(sv, SVt_NV);
906f284f 2512#ifdef USE_LONG_DOUBLE
097ee67d 2513 DEBUG_c({
f93f4e46 2514 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2515 PerlIO_printf(Perl_debug_log,
2516 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2517 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2518 RESTORE_NUMERIC_LOCAL();
2519 });
65202027 2520#else
572bbb43 2521 DEBUG_c({
f93f4e46 2522 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2523 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2524 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2525 RESTORE_NUMERIC_LOCAL();
2526 });
572bbb43 2527#endif
79072805
LW
2528 }
2529 else if (SvTYPE(sv) < SVt_PVNV)
2530 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2531 if (SvNOKp(sv)) {
2532 return SvNVX(sv);
61604483 2533 }
59d8ce62 2534 if (SvIOKp(sv)) {
9d6ce603 2535 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2536#ifdef NV_PRESERVES_UV
a43d94f2
NC
2537 if (SvIOK(sv))
2538 SvNOK_on(sv);
2539 else
2540 SvNOKp_on(sv);
28e5dec8
JH
2541#else
2542 /* Only set the public NV OK flag if this NV preserves the IV */
2543 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2544 if (SvIOK(sv) &&
2545 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2546 : (SvIVX(sv) == I_V(SvNVX(sv))))
2547 SvNOK_on(sv);
2548 else
2549 SvNOKp_on(sv);
2550#endif
93a17b20 2551 }
748a9306 2552 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2553 UV value;
3f7c398e 2554 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2555 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2556 not_a_number(sv);
28e5dec8 2557#ifdef NV_PRESERVES_UV
c2988b20
NC
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
5e045b90 2560 /* It's definitely an integer */
9d6ce603 2561 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2562 } else
3f7c398e 2563 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2564 if (numtype)
2565 SvNOK_on(sv);
2566 else
2567 SvNOKp_on(sv);
28e5dec8 2568#else
3f7c398e 2569 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2570 /* Only set the public NV OK flag if this NV preserves the value in
2571 the PV at least as well as an IV/UV would.
2572 Not sure how to do this 100% reliably. */
2573 /* if that shift count is out of range then Configure's test is
2574 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2575 UV_BITS */
2576 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2577 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2578 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2579 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2580 /* Can't use strtol etc to convert this string, so don't try.
2581 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2582 SvNOK_on(sv);
2583 } else {
2584 /* value has been set. It may not be precise. */
2585 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2586 /* 2s complement assumption for (UV)IV_MIN */
2587 SvNOK_on(sv); /* Integer is too negative. */
2588 } else {
2589 SvNOKp_on(sv);
2590 SvIOKp_on(sv);
6fa402ec 2591
c2988b20 2592 if (numtype & IS_NUMBER_NEG) {
45977657 2593 SvIV_set(sv, -(IV)value);
c2988b20 2594 } else if (value <= (UV)IV_MAX) {
45977657 2595 SvIV_set(sv, (IV)value);
c2988b20 2596 } else {
607fa7f2 2597 SvUV_set(sv, value);
c2988b20
NC
2598 SvIsUV_on(sv);
2599 }
2600
2601 if (numtype & IS_NUMBER_NOT_INT) {
2602 /* I believe that even if the original PV had decimals,
2603 they are lost beyond the limit of the FP precision.
2604 However, neither is canonical, so both only get p
2605 flags. NWC, 2000/11/25 */
2606 /* Both already have p flags, so do nothing */
2607 } else {
66a1b24b 2608 const NV nv = SvNVX(sv);
c2988b20
NC
2609 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2610 if (SvIVX(sv) == I_V(nv)) {
2611 SvNOK_on(sv);
c2988b20 2612 } else {
c2988b20
NC
2613 /* It had no "." so it must be integer. */
2614 }
00b6aa41 2615 SvIOK_on(sv);
c2988b20
NC
2616 } else {
2617 /* between IV_MAX and NV(UV_MAX).
2618 Could be slightly > UV_MAX */
6fa402ec 2619
c2988b20
NC
2620 if (numtype & IS_NUMBER_NOT_INT) {
2621 /* UV and NV both imprecise. */
2622 } else {
66a1b24b 2623 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2624
2625 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2626 SvNOK_on(sv);
c2988b20 2627 }
00b6aa41 2628 SvIOK_on(sv);
c2988b20
NC
2629 }
2630 }
2631 }
2632 }
2633 }
a43d94f2
NC
2634 /* It might be more code efficient to go through the entire logic above
2635 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2636 gets complex and potentially buggy, so more programmer efficient
2637 to do it this way, by turning off the public flags: */
2638 if (!numtype)
2639 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2640#endif /* NV_PRESERVES_UV */
93a17b20 2641 }
79072805 2642 else {
f7877b28 2643 if (isGV_with_GP(sv)) {
159b6efe 2644 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2645 return 0.0;
2646 }
2647
041457d9 2648 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2649 report_uninit(sv);
7e25a7e9
NC
2650 assert (SvTYPE(sv) >= SVt_NV);
2651 /* Typically the caller expects that sv_any is not NULL now. */
2652 /* XXX Ilya implies that this is a bug in callers that assume this
2653 and ideally should be fixed. */
a0d0e21e 2654 return 0.0;
79072805 2655 }
572bbb43 2656#if defined(USE_LONG_DOUBLE)
097ee67d 2657 DEBUG_c({
f93f4e46 2658 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2659 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2660 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2661 RESTORE_NUMERIC_LOCAL();
2662 });
65202027 2663#else
572bbb43 2664 DEBUG_c({
f93f4e46 2665 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2666 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2667 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2668 RESTORE_NUMERIC_LOCAL();
2669 });
572bbb43 2670#endif
463ee0b2 2671 return SvNVX(sv);
79072805
LW
2672}
2673
800401ee
JH
2674/*
2675=for apidoc sv_2num
2676
2677Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2678reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2679access this function.
800401ee
JH
2680
2681=cut
2682*/
2683
2684SV *
5de3775c 2685Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2686{
7918f24d
NC
2687 PERL_ARGS_ASSERT_SV_2NUM;
2688
b9ee0594
RGS
2689 if (!SvROK(sv))
2690 return sv;
800401ee
JH
2691 if (SvAMAGIC(sv)) {
2692 SV * const tmpsv = AMG_CALLun(sv,numer);
2693 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2694 return sv_2num(tmpsv);
2695 }
2696 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2697}
2698
645c22ef
DM
2699/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2700 * UV as a string towards the end of buf, and return pointers to start and
2701 * end of it.
2702 *
2703 * We assume that buf is at least TYPE_CHARS(UV) long.
2704 */
2705
864dbfa3 2706static char *
5de3775c 2707S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2708{
25da4f38 2709 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2710 char * const ebuf = ptr;
25da4f38 2711 int sign;
25da4f38 2712
7918f24d
NC
2713 PERL_ARGS_ASSERT_UIV_2BUF;
2714
25da4f38
IZ
2715 if (is_uv)
2716 sign = 0;
2717 else if (iv >= 0) {
2718 uv = iv;
2719 sign = 0;
2720 } else {
2721 uv = -iv;
2722 sign = 1;
2723 }
2724 do {
eb160463 2725 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2726 } while (uv /= 10);
2727 if (sign)
2728 *--ptr = '-';
2729 *peob = ebuf;
2730 return ptr;
2731}
2732
645c22ef
DM
2733/*
2734=for apidoc sv_2pv_flags
2735
ff276b08 2736Returns a pointer to the string value of an SV, and sets *lp to its length.
645c22ef
DM
2737If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2738if necessary.
2739Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2740usually end up here too.
2741
2742=cut
2743*/
2744
8d6d96c1 2745char *
5de3775c 2746Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2747{
97aff369 2748 dVAR;
79072805 2749 register char *s;
79072805 2750
463ee0b2 2751 if (!sv) {
cdb061a3
NC
2752 if (lp)
2753 *lp = 0;
73d840c0 2754 return (char *)"";
463ee0b2 2755 }
8990e307 2756 if (SvGMAGICAL(sv)) {
8d6d96c1
HS
2757 if (flags & SV_GMAGIC)
2758 mg_get(sv);
463ee0b2 2759 if (SvPOKp(sv)) {
cdb061a3
NC
2760 if (lp)
2761 *lp = SvCUR(sv);
10516c54
NC
2762 if (flags & SV_MUTABLE_RETURN)
2763 return SvPVX_mutable(sv);
4d84ee25
NC
2764 if (flags & SV_CONST_RETURN)
2765 return (char *)SvPVX_const(sv);
463ee0b2
LW
2766 return SvPVX(sv);
2767 }
75dfc8ec
NC
2768 if (SvIOKp(sv) || SvNOKp(sv)) {
2769 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
75dfc8ec
NC
2770 STRLEN len;
2771
2772 if (SvIOKp(sv)) {
e80fed9d 2773 len = SvIsUV(sv)
d9fad198
JH
2774 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2775 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
75dfc8ec 2776 } else {
e8ada2d0
NC
2777 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2778 len = strlen(tbuf);
75dfc8ec 2779 }
b5b886f0
NC
2780 assert(!SvROK(sv));
2781 {
75dfc8ec
NC
2782 dVAR;
2783
2784#ifdef FIXNEGATIVEZERO
e8ada2d0
NC
2785 if (len == 2 && tbuf[0] == '-' && tbuf[1] == '0') {
2786 tbuf[0] = '0';
2787 tbuf[1] = 0;
75dfc8ec
NC
2788 len = 1;
2789 }
2790#endif
2791 SvUPGRADE(sv, SVt_PV);
2792 if (lp)
2793 *lp = len;
2794 s = SvGROW_mutable(sv, len + 1);
2795 SvCUR_set(sv, len);
2796 SvPOKp_on(sv);
10edeb5d 2797 return (char*)memcpy(s, tbuf, len + 1);
75dfc8ec 2798 }
463ee0b2 2799 }
1c7ff15e
NC
2800 if (SvROK(sv)) {
2801 goto return_rok;
2802 }
2803 assert(SvTYPE(sv) >= SVt_PVMG);
2804 /* This falls through to the report_uninit near the end of the
2805 function. */
2806 } else if (SvTHINKFIRST(sv)) {
ed6116ce 2807 if (SvROK(sv)) {
1c7ff15e 2808 return_rok:
deb46114 2809 if (SvAMAGIC(sv)) {
aee036bb
DM
2810 SV *tmpstr;
2811 if (flags & SV_SKIP_OVERLOAD)
2812 return NULL;
2813 tmpstr = AMG_CALLun(sv,string);
deb46114
NC
2814 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2815 /* Unwrap this: */
2816 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2817 */
2818
2819 char *pv;
2820 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2821 if (flags & SV_CONST_RETURN) {
2822 pv = (char *) SvPVX_const(tmpstr);
2823 } else {
2824 pv = (flags & SV_MUTABLE_RETURN)
2825 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2826 }
2827 if (lp)
2828 *lp = SvCUR(tmpstr);
50adf7d2 2829 } else {
deb46114 2830 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2831 }
deb46114
NC
2832 if (SvUTF8(tmpstr))
2833 SvUTF8_on(sv);
2834 else
2835 SvUTF8_off(sv);
2836 return pv;
50adf7d2 2837 }
deb46114
NC
2838 }
2839 {
fafee734
NC
2840 STRLEN len;
2841 char *retval;
2842 char *buffer;
d2c6dc5e 2843 SV *const referent = SvRV(sv);
d8eae41e
NC
2844
2845 if (!referent) {
fafee734
NC
2846 len = 7;
2847 retval = buffer = savepvn("NULLREF", len);
5c35adbb 2848 } else if (SvTYPE(referent) == SVt_REGEXP) {
d2c6dc5e 2849 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
67d2d14d
AB
2850 I32 seen_evals = 0;
2851
2852 assert(re);
2853
2854 /* If the regex is UTF-8 we want the containing scalar to
2855 have an UTF-8 flag too */
2856 if (RX_UTF8(re))
2857 SvUTF8_on(sv);
2858 else
2859 SvUTF8_off(sv);
2860
2861 if ((seen_evals = RX_SEEN_EVALS(re)))
2862 PL_reginterp_cnt += seen_evals;
2863
2864 if (lp)
2865 *lp = RX_WRAPLEN(re);
2866
2867 return RX_WRAPPED(re);
d8eae41e
NC
2868 } else {
2869 const char *const typestr = sv_reftype(referent, 0);
fafee734
NC
2870 const STRLEN typelen = strlen(typestr);
2871 UV addr = PTR2UV(referent);
2872 const char *stashname = NULL;
2873 STRLEN stashnamelen = 0; /* hush, gcc */
2874 const char *buffer_end;
d8eae41e 2875
d8eae41e 2876 if (SvOBJECT(referent)) {
fafee734
NC
2877 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2878
2879 if (name) {
2880 stashname = HEK_KEY(name);
2881 stashnamelen = HEK_LEN(name);
2882
2883 if (HEK_UTF8(name)) {
2884 SvUTF8_on(sv);
2885 } else {
2886 SvUTF8_off(sv);
2887 }
2888 } else {
2889 stashname = "__ANON__";
2890 stashnamelen = 8;
2891 }
2892 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2893 + 2 * sizeof(UV) + 2 /* )\0 */;
2894 } else {
2895 len = typelen + 3 /* (0x */
2896 + 2 * sizeof(UV) + 2 /* )\0 */;
d8eae41e 2897 }
fafee734
NC
2898
2899 Newx(buffer, len, char);
2900 buffer_end = retval = buffer + len;
2901
2902 /* Working backwards */
2903 *--retval = '\0';
2904 *--retval = ')';
2905 do {
2906 *--retval = PL_hexdigit[addr & 15];
2907 } while (addr >>= 4);
2908 *--retval = 'x';
2909 *--retval = '0';
2910 *--retval = '(';
2911
2912 retval -= typelen;
2913 memcpy(retval, typestr, typelen);
2914
2915 if (stashname) {
2916 *--retval = '=';
2917 retval -= stashnamelen;
2918 memcpy(retval, stashname, stashnamelen);
2919 }
2920 /* retval may not neccesarily have reached the start of the
2921 buffer here. */
2922 assert (retval >= buffer);
2923
2924 len = buffer_end - retval - 1; /* -1 for that \0 */
c080367d 2925 }
042dae7a 2926 if (lp)
fafee734
NC
2927 *lp = len;
2928 SAVEFREEPV(buffer);
2929 return retval;
463ee0b2 2930 }
79072805 2931 }
0336b60e 2932 if (SvREADONLY(sv) && !SvOK(sv)) {
cdb061a3
NC
2933 if (lp)
2934 *lp = 0;
9f621bb0
NC
2935 if (flags & SV_UNDEF_RETURNS_NULL)
2936 return NULL;
2937 if (ckWARN(WARN_UNINITIALIZED))
2938 report_uninit(sv);
73d840c0 2939 return (char *)"";
79072805 2940 }
79072805 2941 }
28e5dec8
JH
2942 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2943 /* I'm assuming that if both IV and NV are equally valid then
2944 converting the IV is going to be more efficient */
e1ec3a88 2945 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2946 char buf[TYPE_CHARS(UV)];
2947 char *ebuf, *ptr;
97a130b8 2948 STRLEN len;
28e5dec8
JH
2949
2950 if (SvTYPE(sv) < SVt_PVIV)
2951 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2952 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2953 len = ebuf - ptr;
5902b6a9 2954 /* inlined from sv_setpvn */
97a130b8
NC
2955 s = SvGROW_mutable(sv, len + 1);
2956 Move(ptr, s, len, char);
2957 s += len;
28e5dec8 2958 *s = '\0';
28e5dec8
JH
2959 }
2960 else if (SvNOKp(sv)) {
4ee39169 2961 dSAVE_ERRNO;
79072805
LW
2962 if (SvTYPE(sv) < SVt_PVNV)
2963 sv_upgrade(sv, SVt_PVNV);
1c846c1f 2964 /* The +20 is pure guesswork. Configure test needed. --jhi */
5902b6a9 2965 s = SvGROW_mutable(sv, NV_DIG + 20);
c81271c3 2966 /* some Xenix systems wipe out errno here */
79072805 2967#ifdef apollo
463ee0b2 2968 if (SvNVX(sv) == 0.0)
d1307786 2969 my_strlcpy(s, "0", SvLEN(sv));
79072805
LW
2970 else
2971#endif /*apollo*/
bbce6d69 2972 {
2d4389e4 2973 Gconvert(SvNVX(sv), NV_DIG, 0, s);
bbce6d69 2974 }
4ee39169 2975 RESTORE_ERRNO;
a0d0e21e 2976#ifdef FIXNEGATIVEZERO
20773dcd
NC
2977 if (*s == '-' && s[1] == '0' && !s[2]) {
2978 s[0] = '0';
2979 s[1] = 0;
2980 }
a0d0e21e 2981#endif
79072805
LW
2982 while (*s) s++;
2983#ifdef hcx
2984 if (s[-1] == '.')
46fc3d4c 2985 *--s = '\0';
79072805
LW
2986#endif
2987 }
79072805 2988 else {
8d1c3e26
NC
2989 if (isGV_with_GP(sv)) {
2990 GV *const gv = MUTABLE_GV(sv);
2991 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
2992 SV *const buffer = sv_newmortal();
2993
2994 /* FAKE globs can get coerced, so need to turn this off temporarily
2995 if it is on. */
2996 SvFAKE_off(gv);
2997 gv_efullname3(buffer, gv, "*");
2998 SvFLAGS(gv) |= wasfake;
2999
1809c940
DM
3000 if (SvPOK(buffer)) {
3001 if (lp) {
3002 *lp = SvCUR(buffer);
3003 }
3004 return SvPVX(buffer);
3005 }
3006 else {
3007 if (lp)
3008 *lp = 0;
3009 return (char *)"";
8d1c3e26 3010 }
8d1c3e26 3011 }
180488f8 3012
cdb061a3 3013 if (lp)
00b6aa41 3014 *lp = 0;
9f621bb0
NC
3015 if (flags & SV_UNDEF_RETURNS_NULL)
3016 return NULL;
3017 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
3018 report_uninit(sv);
25da4f38
IZ
3019 if (SvTYPE(sv) < SVt_PV)
3020 /* Typically the caller expects that sv_any is not NULL now. */
3021 sv_upgrade(sv, SVt_PV);
73d840c0 3022 return (char *)"";
79072805 3023 }
cdb061a3 3024 {
823a54a3 3025 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3026 if (lp)
3027 *lp = len;
3028 SvCUR_set(sv, len);
3029 }
79072805 3030 SvPOK_on(sv);
1d7c1841 3031 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3032 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3033 if (flags & SV_CONST_RETURN)
3034 return (char *)SvPVX_const(sv);
10516c54
NC
3035 if (flags & SV_MUTABLE_RETURN)
3036 return SvPVX_mutable(sv);
463ee0b2
LW
3037 return SvPVX(sv);
3038}
3039
645c22ef 3040/*
6050d10e
JP
3041=for apidoc sv_copypv
3042
3043Copies a stringified representation of the source SV into the
3044destination SV. Automatically performs any necessary mg_get and
54f0641b 3045coercion of numeric values into strings. Guaranteed to preserve
2575c402 3046UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3047sv_2pv[_flags] but operates directly on an SV instead of just the
3048string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3049would lose the UTF-8'ness of the PV.
3050
3051=cut
3052*/
3053
3054void
5de3775c 3055Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 3056{
446eaa42 3057 STRLEN len;
53c1dcc0 3058 const char * const s = SvPV_const(ssv,len);
7918f24d
NC
3059
3060 PERL_ARGS_ASSERT_SV_COPYPV;
3061
cb50f42d 3062 sv_setpvn(dsv,s,len);
446eaa42 3063 if (SvUTF8(ssv))
cb50f42d 3064 SvUTF8_on(dsv);
446eaa42 3065 else
cb50f42d 3066 SvUTF8_off(dsv);
6050d10e
JP
3067}
3068
3069/*
645c22ef
DM
3070=for apidoc sv_2pvbyte
3071
3072Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3073to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3074side-effect.
3075
3076Usually accessed via the C<SvPVbyte> macro.
3077
3078=cut
3079*/
3080
7340a771 3081char *
5de3775c 3082Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3083{
7918f24d
NC
3084 PERL_ARGS_ASSERT_SV_2PVBYTE;
3085
0875d2fe 3086 sv_utf8_downgrade(sv,0);
97972285 3087 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771
GS
3088}
3089
645c22ef 3090/*
035cbb0e
RGS
3091=for apidoc sv_2pvutf8
3092
3093Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3094to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3095
3096Usually accessed via the C<SvPVutf8> macro.
3097
3098=cut
3099*/
645c22ef 3100
7340a771 3101char *
7bc54cea 3102Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3103{
7918f24d
NC
3104 PERL_ARGS_ASSERT_SV_2PVUTF8;
3105
035cbb0e
RGS
3106 sv_utf8_upgrade(sv);
3107 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771 3108}
1c846c1f 3109
7ee2227d 3110
645c22ef
DM
3111/*
3112=for apidoc sv_2bool
3113
3114This function is only called on magical items, and is only used by
8cf8f3d1 3115sv_true() or its macro equivalent.
645c22ef
DM
3116
3117=cut
3118*/
3119
463ee0b2 3120bool
7bc54cea 3121Perl_sv_2bool(pTHX_ register SV *const sv)
463ee0b2 3122{
97aff369 3123 dVAR;
7918f24d
NC
3124
3125 PERL_ARGS_ASSERT_SV_2BOOL;
3126
5b295bef 3127 SvGETMAGIC(sv);
463ee0b2 3128
a0d0e21e
LW
3129 if (!SvOK(sv))
3130 return 0;
3131 if (SvROK(sv)) {
fabdb6c0
AL
3132 if (SvAMAGIC(sv)) {
3133 SV * const tmpsv = AMG_CALLun(sv,bool_);
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 }
463ee0b2 3139 if (SvPOKp(sv)) {
53c1dcc0
AL
3140 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3141 if (Xpvtmp &&
339049b0 3142 (*sv->sv_u.svu_pv > '0' ||
11343788 3143 Xpvtmp->xpv_cur > 1 ||
339049b0 3144 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
463ee0b2
LW
3145 return 1;
3146 else
3147 return 0;
3148 }
3149 else {
3150 if (SvIOKp(sv))
3151 return SvIVX(sv) != 0;
3152 else {
3153 if (SvNOKp(sv))
3154 return SvNVX(sv) != 0.0;
180488f8 3155 else {
f7877b28 3156 if (isGV_with_GP(sv))
180488f8
NC
3157 return TRUE;
3158 else
3159 return FALSE;
3160 }
463ee0b2
LW
3161 }
3162 }
79072805
LW
3163}
3164
c461cf8f
JH
3165/*
3166=for apidoc sv_utf8_upgrade
3167
78ea37eb 3168Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3169Forces the SV to string form if it is not already.
2bbc8d55 3170Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3171Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3172if the whole string is the same in UTF-8 as not.
3173Returns the number of bytes in the converted string
c461cf8f 3174
13a6c0e0
JH
3175This is not as a general purpose byte encoding to Unicode interface:
3176use the Encode extension for that.
3177
fe749c9a
KW
3178=for apidoc sv_utf8_upgrade_nomg
3179
3180Like sv_utf8_upgrade, but doesn't do magic on C<sv>
3181
8d6d96c1
HS
3182=for apidoc sv_utf8_upgrade_flags
3183
78ea37eb 3184Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3185Forces the SV to string form if it is not already.
8d6d96c1 3186Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3187if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3188will C<mg_get> on C<sv> if appropriate, else not.
3189Returns the number of bytes in the converted string
3190C<sv_utf8_upgrade> and
8d6d96c1
HS
3191C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3192
13a6c0e0
JH
3193This is not as a general purpose byte encoding to Unicode interface:
3194use the Encode extension for that.
3195
8d6d96c1 3196=cut
b3ab6785
KW
3197
3198The grow version is currently not externally documented. It adds a parameter,
3199extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3200have free after it upon return. This allows the caller to reserve extra space
3201that it intends to fill, to avoid extra grows.
3202
3203Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3204which can be used to tell this function to not first check to see if there are
3205any characters that are different in UTF-8 (variant characters) which would
3206force it to allocate a new string to sv, but to assume there are. Typically
3207this flag is used by a routine that has already parsed the string to find that
3208there are such characters, and passes this information on so that the work
3209doesn't have to be repeated.
3210
3211(One might think that the calling routine could pass in the position of the
3212first such variant, so it wouldn't have to be found again. But that is not the
3213case, because typically when the caller is likely to use this flag, it won't be
3214calling this routine unless it finds something that won't fit into a byte.
3215Otherwise it tries to not upgrade and just use bytes. But some things that
3216do fit into a byte are variants in utf8, and the caller may not have been
3217keeping track of these.)
3218
3219If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3220isn't guaranteed due to having other routines do the work in some input cases,
3221or if the input is already flagged as being in utf8.
3222
3223The speed of this could perhaps be improved for many cases if someone wanted to
3224write a fast function that counts the number of variant characters in a string,
3225especially if it could return the position of the first one.
3226
8d6d96c1
HS
3227*/
3228
3229STRLEN
b3ab6785 3230Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3231{
97aff369 3232 dVAR;
7918f24d 3233
b3ab6785 3234 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3235
808c356f
RGS
3236 if (sv == &PL_sv_undef)
3237 return 0;
e0e62c2a
NIS
3238 if (!SvPOK(sv)) {
3239 STRLEN len = 0;
d52b7888
NC
3240 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3241 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3242 if (SvUTF8(sv)) {
3243 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3244 return len;
b3ab6785 3245 }
d52b7888
NC
3246 } else {
3247 (void) SvPV_force(sv,len);
3248 }
e0e62c2a 3249 }
4411f3b6 3250
f5cee72b 3251 if (SvUTF8(sv)) {
b3ab6785 3252 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3253 return SvCUR(sv);
f5cee72b 3254 }
5fec3b1d 3255
765f542d
NC
3256 if (SvIsCOW(sv)) {
3257 sv_force_normal_flags(sv, 0);
db42d148
NIS
3258 }
3259
b3ab6785 3260 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3261 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3262 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3263 return SvCUR(sv);
3264 }
3265
4e93345f
KW
3266 if (SvCUR(sv) == 0) {
3267 if (extra) SvGROW(sv, extra);
3268 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3269 /* This function could be much more efficient if we
2bbc8d55 3270 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3271 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3272 * make the loop as fast as possible (although there are certainly ways
3273 * to speed this up, eg. through vectorization) */
3274 U8 * s = (U8 *) SvPVX_const(sv);
3275 U8 * e = (U8 *) SvEND(sv);
3276 U8 *t = s;
3277 STRLEN two_byte_count = 0;
c4e7c712 3278
b3ab6785
KW
3279 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3280
3281 /* See if really will need to convert to utf8. We mustn't rely on our
3282 * incoming SV being well formed and having a trailing '\0', as certain
3283 * code in pp_formline can send us partially built SVs. */
3284
c4e7c712 3285 while (t < e) {
53c1dcc0 3286 const U8 ch = *t++;
b3ab6785
KW
3287 if (NATIVE_IS_INVARIANT(ch)) continue;
3288
3289 t--; /* t already incremented; re-point to first variant */
3290 two_byte_count = 1;
3291 goto must_be_utf8;
c4e7c712 3292 }
b3ab6785
KW
3293
3294 /* utf8 conversion not needed because all are invariants. Mark as
3295 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3296 SvUTF8_on(sv);
b3ab6785
KW
3297 return SvCUR(sv);
3298
3299must_be_utf8:
3300
3301 /* Here, the string should be converted to utf8, either because of an
3302 * input flag (two_byte_count = 0), or because a character that
3303 * requires 2 bytes was found (two_byte_count = 1). t points either to
3304 * the beginning of the string (if we didn't examine anything), or to
3305 * the first variant. In either case, everything from s to t - 1 will
3306 * occupy only 1 byte each on output.
3307 *
3308 * There are two main ways to convert. One is to create a new string
3309 * and go through the input starting from the beginning, appending each
3310 * converted value onto the new string as we go along. It's probably
3311 * best to allocate enough space in the string for the worst possible
3312 * case rather than possibly running out of space and having to
3313 * reallocate and then copy what we've done so far. Since everything
3314 * from s to t - 1 is invariant, the destination can be initialized
3315 * with these using a fast memory copy
3316 *
3317 * The other way is to figure out exactly how big the string should be
3318 * by parsing the entire input. Then you don't have to make it big
3319 * enough to handle the worst possible case, and more importantly, if
3320 * the string you already have is large enough, you don't have to
3321 * allocate a new string, you can copy the last character in the input
3322 * string to the final position(s) that will be occupied by the
3323 * converted string and go backwards, stopping at t, since everything
3324 * before that is invariant.
3325 *
3326 * There are advantages and disadvantages to each method.
3327 *
3328 * In the first method, we can allocate a new string, do the memory
3329 * copy from the s to t - 1, and then proceed through the rest of the
3330 * string byte-by-byte.
3331 *
3332 * In the second method, we proceed through the rest of the input
3333 * string just calculating how big the converted string will be. Then
3334 * there are two cases:
3335 * 1) if the string has enough extra space to handle the converted
3336 * value. We go backwards through the string, converting until we
3337 * get to the position we are at now, and then stop. If this
3338 * position is far enough along in the string, this method is
3339 * faster than the other method. If the memory copy were the same
3340 * speed as the byte-by-byte loop, that position would be about
3341 * half-way, as at the half-way mark, parsing to the end and back
3342 * is one complete string's parse, the same amount as starting
3343 * over and going all the way through. Actually, it would be
3344 * somewhat less than half-way, as it's faster to just count bytes
3345 * than to also copy, and we don't have the overhead of allocating
3346 * a new string, changing the scalar to use it, and freeing the
3347 * existing one. But if the memory copy is fast, the break-even
3348 * point is somewhere after half way. The counting loop could be
3349 * sped up by vectorization, etc, to move the break-even point
3350 * further towards the beginning.
3351 * 2) if the string doesn't have enough space to handle the converted
3352 * value. A new string will have to be allocated, and one might
3353 * as well, given that, start from the beginning doing the first
3354 * method. We've spent extra time parsing the string and in
3355 * exchange all we've gotten is that we know precisely how big to
3356 * make the new one. Perl is more optimized for time than space,
3357 * so this case is a loser.
3358 * So what I've decided to do is not use the 2nd method unless it is
3359 * guaranteed that a new string won't have to be allocated, assuming
3360 * the worst case. I also decided not to put any more conditions on it
3361 * than this, for now. It seems likely that, since the worst case is
3362 * twice as big as the unknown portion of the string (plus 1), we won't
3363 * be guaranteed enough space, causing us to go to the first method,
3364 * unless the string is short, or the first variant character is near
3365 * the end of it. In either of these cases, it seems best to use the
3366 * 2nd method. The only circumstance I can think of where this would
3367 * be really slower is if the string had once had much more data in it
3368 * than it does now, but there is still a substantial amount in it */
3369
3370 {
3371 STRLEN invariant_head = t - s;
3372 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3373 if (SvLEN(sv) < size) {
3374
3375 /* Here, have decided to allocate a new string */
3376
3377 U8 *dst;
3378 U8 *d;
3379
3380 Newx(dst, size, U8);
3381
3382 /* If no known invariants at the beginning of the input string,
3383 * set so starts from there. Otherwise, can use memory copy to
3384 * get up to where we are now, and then start from here */
3385
3386 if (invariant_head <= 0) {
3387 d = dst;
3388 } else {
3389 Copy(s, dst, invariant_head, char);
3390 d = dst + invariant_head;
3391 }
3392
3393 while (t < e) {
3394 const UV uv = NATIVE8_TO_UNI(*t++);
3395 if (UNI_IS_INVARIANT(uv))
3396 *d++ = (U8)UNI_TO_NATIVE(uv);
3397 else {
3398 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3399 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3400 }
3401 }
3402 *d = '\0';
3403 SvPV_free(sv); /* No longer using pre-existing string */
3404 SvPV_set(sv, (char*)dst);
3405 SvCUR_set(sv, d - dst);
3406 SvLEN_set(sv, size);
3407 } else {
3408
3409 /* Here, have decided to get the exact size of the string.
3410 * Currently this happens only when we know that there is
3411 * guaranteed enough space to fit the converted string, so
3412 * don't have to worry about growing. If two_byte_count is 0,
3413 * then t points to the first byte of the string which hasn't
3414 * been examined yet. Otherwise two_byte_count is 1, and t
3415 * points to the first byte in the string that will expand to
3416 * two. Depending on this, start examining at t or 1 after t.
3417 * */
3418
3419 U8 *d = t + two_byte_count;
3420
3421
3422 /* Count up the remaining bytes that expand to two */
3423
3424 while (d < e) {
3425 const U8 chr = *d++;
3426 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3427 }
3428
3429 /* The string will expand by just the number of bytes that
3430 * occupy two positions. But we are one afterwards because of
3431 * the increment just above. This is the place to put the
3432 * trailing NUL, and to set the length before we decrement */
3433
3434 d += two_byte_count;
3435 SvCUR_set(sv, d - s);
3436 *d-- = '\0';
3437
3438
3439 /* Having decremented d, it points to the position to put the
3440 * very last byte of the expanded string. Go backwards through
3441 * the string, copying and expanding as we go, stopping when we
3442 * get to the part that is invariant the rest of the way down */
3443
3444 e--;
3445 while (e >= t) {
3446 const U8 ch = NATIVE8_TO_UNI(*e--);
3447 if (UNI_IS_INVARIANT(ch)) {
3448 *d-- = UNI_TO_NATIVE(ch);
3449 } else {
3450 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3451 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3452 }
3453 }
3454 }
3455 }
560a288e 3456 }
b3ab6785
KW
3457
3458 /* Mark as UTF-8 even if no variant - saves scanning loop */
3459 SvUTF8_on(sv);
4411f3b6 3460 return SvCUR(sv);
560a288e
GS
3461}
3462
c461cf8f
JH
3463/*
3464=for apidoc sv_utf8_downgrade
3465
78ea37eb 3466Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3467If the PV contains a character that cannot fit
3468in a byte, this conversion will fail;
78ea37eb 3469in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3470true, croaks.
3471
13a6c0e0
JH
3472This is not as a general purpose Unicode to byte encoding interface:
3473use the Encode extension for that.
3474
c461cf8f
JH
3475=cut
3476*/
3477
560a288e 3478bool
7bc54cea 3479Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
560a288e 3480{
97aff369 3481 dVAR;
7918f24d
NC
3482
3483 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3484
78ea37eb 3485 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3486 if (SvCUR(sv)) {
03cfe0ae 3487 U8 *s;
652088fc 3488 STRLEN len;
fa301091 3489
765f542d
NC
3490 if (SvIsCOW(sv)) {
3491 sv_force_normal_flags(sv, 0);
3492 }
03cfe0ae
NIS
3493 s = (U8 *) SvPV(sv, len);
3494 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3495 if (fail_ok)
3496 return FALSE;
3497 else {
3498 if (PL_op)
3499 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3500 OP_DESC(PL_op));
fa301091
JH
3501 else
3502 Perl_croak(aTHX_ "Wide character");
3503 }
4b3603a4 3504 }
b162af07 3505 SvCUR_set(sv, len);
67e989fb 3506 }
560a288e 3507 }
ffebcc3e 3508 SvUTF8_off(sv);
560a288e
GS
3509 return TRUE;
3510}
3511
c461cf8f
JH
3512/*
3513=for apidoc sv_utf8_encode
3514
78ea37eb
ST
3515Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3516flag off so that it looks like octets again.
c461cf8f
JH
3517
3518=cut
3519*/
3520
560a288e 3521void
7bc54cea 3522Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3523{
7918f24d
NC
3524 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3525
4c94c214
NC
3526 if (SvIsCOW(sv)) {
3527 sv_force_normal_flags(sv, 0);
3528 }
3529 if (SvREADONLY(sv)) {
f1f66076 3530 Perl_croak(aTHX_ "%s", PL_no_modify);
4c94c214 3531 }
a5f5288a 3532 (void) sv_utf8_upgrade(sv);
560a288e
GS
3533 SvUTF8_off(sv);
3534}
3535
4411f3b6
NIS
3536/*
3537=for apidoc sv_utf8_decode
3538
78ea37eb
ST
3539If the PV of the SV is an octet sequence in UTF-8
3540and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3541so that it looks like a character. If the PV contains only single-byte
3542characters, the C<SvUTF8> flag stays being off.
3543Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3544
3545=cut
3546*/
3547
560a288e 3548bool
7bc54cea 3549Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3550{
7918f24d
NC
3551 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3552
78ea37eb 3553 if (SvPOKp(sv)) {
93524f2b
NC
3554 const U8 *c;
3555 const U8 *e;
9cbac4c7 3556
645c22ef
DM
3557 /* The octets may have got themselves encoded - get them back as
3558 * bytes
3559 */
3560 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3561 return FALSE;
3562
3563 /* it is actually just a matter of turning the utf8 flag on, but
3564 * we want to make sure everything inside is valid utf8 first.
3565 */
93524f2b 3566 c = (const U8 *) SvPVX_const(sv);
63cd0674 3567 if (!is_utf8_string(c, SvCUR(sv)+1))
67e989fb 3568 return FALSE;
93524f2b 3569 e = (const U8 *) SvEND(sv);
511c2ff0 3570 while (c < e) {
b64e5050 3571 const U8 ch = *c++;
c4d5f83a 3572 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3573 SvUTF8_on(sv);
3574 break;
3575 }
560a288e 3576 }
560a288e
GS
3577 }
3578 return TRUE;
3579}
3580
954c1994
GS
3581/*
3582=for apidoc sv_setsv
3583
645c22ef
DM
3584Copies the contents of the source SV C<ssv> into the destination SV
3585C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3586function if the source SV needs to be reused. Does not handle 'set' magic.
3587Loosely speaking, it performs a copy-by-value, obliterating any previous
3588content of the destination.
3589
3590You probably want to use one of the assortment of wrappers, such as
3591C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3592C<SvSetMagicSV_nosteal>.
3593
8d6d96c1
HS
3594=for apidoc sv_setsv_flags
3595
645c22ef
DM
3596Copies the contents of the source SV C<ssv> into the destination SV
3597C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3598function if the source SV needs to be reused. Does not handle 'set' magic.
3599Loosely speaking, it performs a copy-by-value, obliterating any previous
3600content of the destination.
3601If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
5fcdf167
NC
3602C<ssv> if appropriate, else not. If the C<flags> parameter has the
3603C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3604and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3605
3606You probably want to use one of the assortment of wrappers, such as
3607C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3608C<SvSetMagicSV_nosteal>.
3609
3610This is the primary function for copying scalars, and most other
3611copy-ish functions and macros use this underneath.
8d6d96c1
HS
3612
3613=cut
3614*/
3615
5d0301b7 3616static void
7bc54cea 3617S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3618{
70cd14a1 3619 I32 mro_changes = 0; /* 1 = method, 2 = isa */
dd69841b 3620
7918f24d
NC
3621 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3622
5d0301b7
NC
3623 if (dtype != SVt_PVGV) {
3624 const char * const name = GvNAME(sstr);
3625 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3626 {
f7877b28
NC
3627 if (dtype >= SVt_PV) {
3628 SvPV_free(dstr);
3629 SvPV_set(dstr, 0);
3630 SvLEN_set(dstr, 0);
3631 SvCUR_set(dstr, 0);
3632 }
0d092c36 3633 SvUPGRADE(dstr, SVt_PVGV);
dedf8e73 3634 (void)SvOK_off(dstr);
2e5b91de
NC
3635 /* FIXME - why are we doing this, then turning it off and on again
3636 below? */
3637 isGV_with_GP_on(dstr);
f7877b28 3638 }
5d0301b7
NC
3639 GvSTASH(dstr) = GvSTASH(sstr);
3640 if (GvSTASH(dstr))
daba3364 3641 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
159b6efe 3642 gv_name_set(MUTABLE_GV(dstr), name, len, GV_ADD);
5d0301b7
NC
3643 SvFAKE_on(dstr); /* can coerce to non-glob */
3644 }
3645
159b6efe 3646 if(GvGP(MUTABLE_GV(sstr))) {
dd69841b
BB
3647 /* If source has method cache entry, clear it */
3648 if(GvCVGEN(sstr)) {
3649 SvREFCNT_dec(GvCV(sstr));
3650 GvCV(sstr) = NULL;
3651 GvCVGEN(sstr) = 0;
3652 }
3653 /* If source has a real method, then a method is
3654 going to change */
159b6efe 3655 else if(GvCV((const GV *)sstr)) {
70cd14a1