This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
remove -Dmad string length restriction
[perl5.git] / sv.c
CommitLineData
a0d0e21e 1/* sv.c
79072805 2 *
1129b882 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
83706693
RGS
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
5 * and others
79072805
LW
6 *
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
9 *
4ac71550
TC
10 */
11
12/*
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
14 * --Pippin
15 *
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
17 */
18
19/*
645c22ef
DM
20 *
21 *
5e045b90
AMS
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
27 * in the pp*.c files.
79072805
LW
28 */
29
30#include "EXTERN.h"
864dbfa3 31#define PERL_IN_SV_C
79072805 32#include "perl.h"
d2f185dc 33#include "regcomp.h"
79072805 34
07208e09 35#ifndef HAS_C99
4b2d6671 36# if __STDC_VERSION__ >= 199901L && !defined(VMS)
07208e09
CS
37# define HAS_C99 1
38# endif
39#endif
40#if HAS_C99
41# include <stdint.h>
42#endif
43
51371543 44#define FCALL *f
2c5424a7 45
2f8ed50e
OS
46#ifdef __Lynx__
47/* Missing proto on LynxOS */
48 char *gconvert(double, int, int, char *);
49#endif
50
e23c8137 51#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 52/* if adding more checks watch out for the following tests:
e23c8137
JH
53 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
54 * lib/utf8.t lib/Unicode/Collate/t/index.t
55 * --jhi
56 */
6f207bd3 57# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
58 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
59 assert((cache)[2] <= (cache)[3]); \
60 assert((cache)[3] <= (cache)[1]);} \
61 } STMT_END
e23c8137 62#else
6f207bd3 63# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
64#endif
65
f8c7b90f 66#ifdef PERL_OLD_COPY_ON_WRITE
765f542d 67#define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
607fa7f2 68#define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
765f542d 69#endif
645c22ef
DM
70
71/* ============================================================================
72
73=head1 Allocation and deallocation of SVs.
74
d2a0f284
JC
75An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
76sv, av, hv...) contains type and reference count information, and for
77many types, a pointer to the body (struct xrv, xpv, xpviv...), which
78contains fields specific to each type. Some types store all they need
79in the head, so don't have a body.
80
486ec47a 81In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
82and bodies are allocated out of arenas, which by default are
83approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
84Sv-bodies are allocated by their sv-type, guaranteeing size
85consistency needed to allocate safely from arrays.
86
d2a0f284
JC
87For SV-heads, the first slot in each arena is reserved, and holds a
88link to the next arena, some flags, and a note of the number of slots.
89Snaked through each arena chain is a linked list of free items; when
90this becomes empty, an extra arena is allocated and divided up into N
91items which are threaded into the free list.
92
93SV-bodies are similar, but they use arena-sets by default, which
94separate the link and info from the arena itself, and reclaim the 1st
95slot in the arena. SV-bodies are further described later.
645c22ef
DM
96
97The following global variables are associated with arenas:
98
99 PL_sv_arenaroot pointer to list of SV arenas
100 PL_sv_root pointer to list of free SV structures
101
d2a0f284
JC
102 PL_body_arenas head of linked-list of body arenas
103 PL_body_roots[] array of pointers to list of free bodies of svtype
104 arrays are indexed by the svtype needed
93e68bfb 105
d2a0f284
JC
106A few special SV heads are not allocated from an arena, but are
107instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
108The size of arenas can be changed from the default by setting
109PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
110
111The SV arena serves the secondary purpose of allowing still-live SVs
112to be located and destroyed during final cleanup.
113
114At the lowest level, the macros new_SV() and del_SV() grab and free
115an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
116to return the SV to the free list with error checking.) new_SV() calls
117more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
118SVs in the free list have their SvTYPE field set to all ones.
119
ff276b08 120At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 121perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 122start of the interpreter.
645c22ef 123
645c22ef
DM
124The function visit() scans the SV arenas list, and calls a specified
125function for each SV it finds which is still live - ie which has an SvTYPE
126other than all 1's, and a non-zero SvREFCNT. visit() is used by the
127following functions (specified as [function that calls visit()] / [function
128called by visit() for each SV]):
129
130 sv_report_used() / do_report_used()
f2524eef 131 dump all remaining SVs (debugging aid)
645c22ef 132
e4487e9b 133 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 134 do_clean_named_io_objs(),do_curse()
645c22ef 135 Attempt to free all objects pointed to by RVs,
e76981f9
FC
136 try to do the same for all objects indir-
137 ectly referenced by typeglobs too, and
138 then do a final sweep, cursing any
139 objects that remain. Called once from
645c22ef
DM
140 perl_destruct(), prior to calling sv_clean_all()
141 below.
142
143 sv_clean_all() / do_clean_all()
144 SvREFCNT_dec(sv) each remaining SV, possibly
145 triggering an sv_free(). It also sets the
146 SVf_BREAK flag on the SV to indicate that the
147 refcnt has been artificially lowered, and thus
148 stopping sv_free() from giving spurious warnings
149 about SVs which unexpectedly have a refcnt
150 of zero. called repeatedly from perl_destruct()
151 until there are no SVs left.
152
93e68bfb 153=head2 Arena allocator API Summary
645c22ef
DM
154
155Private API to rest of sv.c
156
157 new_SV(), del_SV(),
158
df0f0429 159 new_XPVNV(), del_XPVGV(),
645c22ef
DM
160 etc
161
162Public API:
163
8cf8f3d1 164 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 165
645c22ef
DM
166=cut
167
3e8320cc 168 * ========================================================================= */
645c22ef 169
4561caa4
CS
170/*
171 * "A time to plant, and a time to uproot what was planted..."
172 */
173
d7a2c63c
MHM
174#ifdef PERL_MEM_LOG
175# define MEM_LOG_NEW_SV(sv, file, line, func) \
176 Perl_mem_log_new_sv(sv, file, line, func)
177# define MEM_LOG_DEL_SV(sv, file, line, func) \
178 Perl_mem_log_del_sv(sv, file, line, func)
179#else
180# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
181# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
182#endif
183
fd0854ff 184#ifdef DEBUG_LEAKING_SCALARS
484e6108
FC
185# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
186 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
187 } STMT_END
d7a2c63c
MHM
188# define DEBUG_SV_SERIAL(sv) \
189 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
190 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
191#else
192# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 193# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
194#endif
195
48614a46
NC
196#ifdef PERL_POISON
197# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 198# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
199/* Whilst I'd love to do this, it seems that things like to check on
200 unreferenced scalars
7e337ee0 201# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 202*/
7e337ee0
JH
203# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
204 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
205#else
206# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 207# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
208# define POSION_SV_HEAD(sv)
209#endif
210
990198f0
DM
211/* Mark an SV head as unused, and add to free list.
212 *
213 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
214 * its refcount artificially decremented during global destruction, so
215 * there may be dangling pointers to it. The last thing we want in that
216 * case is for it to be reused. */
217
053fc874
GS
218#define plant_SV(p) \
219 STMT_START { \
990198f0 220 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
221 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
222 DEBUG_SV_SERIAL(p); \
fd0854ff 223 FREE_SV_DEBUG_FILE(p); \
48614a46 224 POSION_SV_HEAD(p); \
053fc874 225 SvFLAGS(p) = SVTYPEMASK; \
990198f0 226 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 227 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
228 PL_sv_root = (p); \
229 } \
053fc874
GS
230 --PL_sv_count; \
231 } STMT_END
a0d0e21e 232
053fc874
GS
233#define uproot_SV(p) \
234 STMT_START { \
235 (p) = PL_sv_root; \
daba3364 236 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
237 ++PL_sv_count; \
238 } STMT_END
239
645c22ef 240
cac9b346
NC
241/* make some more SVs by adding another arena */
242
cac9b346
NC
243STATIC SV*
244S_more_sv(pTHX)
245{
97aff369 246 dVAR;
cac9b346 247 SV* sv;
9a87bd09
NC
248 char *chunk; /* must use New here to match call to */
249 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
250 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
251 uproot_SV(sv);
252 return sv;
253}
254
645c22ef
DM
255/* new_SV(): return a new, empty SV head */
256
eba0f806
DM
257#ifdef DEBUG_LEAKING_SCALARS
258/* provide a real function for a debugger to play with */
259STATIC SV*
d7a2c63c 260S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
261{
262 SV* sv;
263
eba0f806
DM
264 if (PL_sv_root)
265 uproot_SV(sv);
266 else
cac9b346 267 sv = S_more_sv(aTHX);
eba0f806
DM
268 SvANY(sv) = 0;
269 SvREFCNT(sv) = 1;
270 SvFLAGS(sv) = 0;
fd0854ff 271 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
272 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
273 ? PL_parser->copline
274 : PL_curcop
f24aceb1
DM
275 ? CopLINE(PL_curcop)
276 : 0
e385c3bf 277 );
fd0854ff 278 sv->sv_debug_inpad = 0;
cd676548 279 sv->sv_debug_parent = NULL;
484e6108 280 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
281
282 sv->sv_debug_serial = PL_sv_serial++;
283
284 MEM_LOG_NEW_SV(sv, file, line, func);
285 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
286 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
287
eba0f806
DM
288 return sv;
289}
d7a2c63c 290# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
291
292#else
293# define new_SV(p) \
053fc874 294 STMT_START { \
053fc874
GS
295 if (PL_sv_root) \
296 uproot_SV(p); \
297 else \
cac9b346 298 (p) = S_more_sv(aTHX); \
053fc874
GS
299 SvANY(p) = 0; \
300 SvREFCNT(p) = 1; \
301 SvFLAGS(p) = 0; \
d7a2c63c 302 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 303 } STMT_END
eba0f806 304#endif
463ee0b2 305
645c22ef
DM
306
307/* del_SV(): return an empty SV head to the free list */
308
a0d0e21e 309#ifdef DEBUGGING
4561caa4 310
053fc874
GS
311#define del_SV(p) \
312 STMT_START { \
aea4f609 313 if (DEBUG_D_TEST) \
053fc874
GS
314 del_sv(p); \
315 else \
316 plant_SV(p); \
053fc874 317 } STMT_END
a0d0e21e 318
76e3520e 319STATIC void
cea2e8a9 320S_del_sv(pTHX_ SV *p)
463ee0b2 321{
97aff369 322 dVAR;
7918f24d
NC
323
324 PERL_ARGS_ASSERT_DEL_SV;
325
aea4f609 326 if (DEBUG_D_TEST) {
4633a7c4 327 SV* sva;
a3b680e6 328 bool ok = 0;
daba3364 329 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
330 const SV * const sv = sva + 1;
331 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 332 if (p >= sv && p < svend) {
a0d0e21e 333 ok = 1;
c0ff570e
NC
334 break;
335 }
a0d0e21e
LW
336 }
337 if (!ok) {
9b387841
NC
338 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
339 "Attempt to free non-arena SV: 0x%"UVxf
340 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
341 return;
342 }
343 }
4561caa4 344 plant_SV(p);
463ee0b2 345}
a0d0e21e 346
4561caa4
CS
347#else /* ! DEBUGGING */
348
349#define del_SV(p) plant_SV(p)
350
351#endif /* DEBUGGING */
463ee0b2 352
645c22ef
DM
353
354/*
ccfc67b7
JH
355=head1 SV Manipulation Functions
356
645c22ef
DM
357=for apidoc sv_add_arena
358
359Given a chunk of memory, link it to the head of the list of arenas,
360and split it into a list of free SVs.
361
362=cut
363*/
364
d2bd4e7f
NC
365static void
366S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 367{
97aff369 368 dVAR;
daba3364 369 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
370 SV* sv;
371 SV* svend;
4633a7c4 372
7918f24d
NC
373 PERL_ARGS_ASSERT_SV_ADD_ARENA;
374
4633a7c4 375 /* The first SV in an arena isn't an SV. */
3280af22 376 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
377 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
378 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
379
3280af22
NIS
380 PL_sv_arenaroot = sva;
381 PL_sv_root = sva + 1;
4633a7c4
LW
382
383 svend = &sva[SvREFCNT(sva) - 1];
384 sv = sva + 1;
463ee0b2 385 while (sv < svend) {
3eef1deb 386 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 387#ifdef DEBUGGING
978b032e 388 SvREFCNT(sv) = 0;
03e36789 389#endif
4b69cbe3 390 /* Must always set typemask because it's always checked in on cleanup
03e36789 391 when the arenas are walked looking for objects. */
8990e307 392 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
393 sv++;
394 }
3eef1deb 395 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
396#ifdef DEBUGGING
397 SvREFCNT(sv) = 0;
398#endif
4633a7c4
LW
399 SvFLAGS(sv) = SVTYPEMASK;
400}
401
055972dc
DM
402/* visit(): call the named function for each non-free SV in the arenas
403 * whose flags field matches the flags/mask args. */
645c22ef 404
5226ed68 405STATIC I32
de37a194 406S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 407{
97aff369 408 dVAR;
4633a7c4 409 SV* sva;
5226ed68 410 I32 visited = 0;
8990e307 411
7918f24d
NC
412 PERL_ARGS_ASSERT_VISIT;
413
daba3364 414 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
415 const SV * const svend = &sva[SvREFCNT(sva)];
416 SV* sv;
4561caa4 417 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 418 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
419 && (sv->sv_flags & mask) == flags
420 && SvREFCNT(sv))
421 {
acfe0abc 422 (FCALL)(aTHX_ sv);
5226ed68
JH
423 ++visited;
424 }
8990e307
LW
425 }
426 }
5226ed68 427 return visited;
8990e307
LW
428}
429
758a08c3
JH
430#ifdef DEBUGGING
431
645c22ef
DM
432/* called by sv_report_used() for each live SV */
433
434static void
5fa45a31 435do_report_used(pTHX_ SV *const sv)
645c22ef 436{
e4787c0c 437 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
438 PerlIO_printf(Perl_debug_log, "****\n");
439 sv_dump(sv);
440 }
441}
758a08c3 442#endif
645c22ef
DM
443
444/*
445=for apidoc sv_report_used
446
fde67290 447Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
448
449=cut
450*/
451
8990e307 452void
864dbfa3 453Perl_sv_report_used(pTHX)
4561caa4 454{
ff270d3a 455#ifdef DEBUGGING
055972dc 456 visit(do_report_used, 0, 0);
96a5add6
AL
457#else
458 PERL_UNUSED_CONTEXT;
ff270d3a 459#endif
4561caa4
CS
460}
461
645c22ef
DM
462/* called by sv_clean_objs() for each live SV */
463
464static void
de37a194 465do_clean_objs(pTHX_ SV *const ref)
645c22ef 466{
97aff369 467 dVAR;
ea724faa
NC
468 assert (SvROK(ref));
469 {
823a54a3
AL
470 SV * const target = SvRV(ref);
471 if (SvOBJECT(target)) {
472 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
473 if (SvWEAKREF(ref)) {
474 sv_del_backref(target, ref);
475 SvWEAKREF_off(ref);
476 SvRV_set(ref, NULL);
477 } else {
478 SvROK_off(ref);
479 SvRV_set(ref, NULL);
fc2b2dca 480 SvREFCNT_dec_NN(target);
823a54a3 481 }
645c22ef
DM
482 }
483 }
645c22ef
DM
484}
485
645c22ef 486
e4487e9b
DM
487/* clear any slots in a GV which hold objects - except IO;
488 * called by sv_clean_objs() for each live GV */
489
645c22ef 490static void
f30de749 491do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 492{
97aff369 493 dVAR;
57ef47cc 494 SV *obj;
ea724faa 495 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 496 assert(isGV_with_GP(sv));
57ef47cc
DM
497 if (!GvGP(sv))
498 return;
499
500 /* freeing GP entries may indirectly free the current GV;
501 * hold onto it while we mess with the GP slots */
502 SvREFCNT_inc(sv);
503
504 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
505 DEBUG_D((PerlIO_printf(Perl_debug_log,
506 "Cleaning named glob SV object:\n "), sv_dump(obj)));
507 GvSV(sv) = NULL;
fc2b2dca 508 SvREFCNT_dec_NN(obj);
57ef47cc
DM
509 }
510 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
511 DEBUG_D((PerlIO_printf(Perl_debug_log,
512 "Cleaning named glob AV object:\n "), sv_dump(obj)));
513 GvAV(sv) = NULL;
fc2b2dca 514 SvREFCNT_dec_NN(obj);
57ef47cc
DM
515 }
516 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
517 DEBUG_D((PerlIO_printf(Perl_debug_log,
518 "Cleaning named glob HV object:\n "), sv_dump(obj)));
519 GvHV(sv) = NULL;
fc2b2dca 520 SvREFCNT_dec_NN(obj);
57ef47cc
DM
521 }
522 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
523 DEBUG_D((PerlIO_printf(Perl_debug_log,
524 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 525 GvCV_set(sv, NULL);
fc2b2dca 526 SvREFCNT_dec_NN(obj);
57ef47cc 527 }
fc2b2dca 528 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
529}
530
68b590d9 531/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
532 * called by sv_clean_objs() for each live GV */
533
534static void
535do_clean_named_io_objs(pTHX_ SV *const sv)
536{
537 dVAR;
538 SV *obj;
539 assert(SvTYPE(sv) == SVt_PVGV);
540 assert(isGV_with_GP(sv));
68b590d9 541 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
542 return;
543
544 SvREFCNT_inc(sv);
57ef47cc
DM
545 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
546 DEBUG_D((PerlIO_printf(Perl_debug_log,
547 "Cleaning named glob IO object:\n "), sv_dump(obj)));
548 GvIOp(sv) = NULL;
fc2b2dca 549 SvREFCNT_dec_NN(obj);
645c22ef 550 }
fc2b2dca 551 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 552}
645c22ef 553
4155e4fe
FC
554/* Void wrapper to pass to visit() */
555static void
556do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
557 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
558 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
559 return;
560 (void)curse(sv, 0);
561}
562
645c22ef
DM
563/*
564=for apidoc sv_clean_objs
565
fde67290 566Attempt to destroy all objects not yet freed.
645c22ef
DM
567
568=cut
569*/
570
4561caa4 571void
864dbfa3 572Perl_sv_clean_objs(pTHX)
4561caa4 573{
97aff369 574 dVAR;
68b590d9 575 GV *olddef, *olderr;
3280af22 576 PL_in_clean_objs = TRUE;
055972dc 577 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
578 /* Some barnacles may yet remain, clinging to typeglobs.
579 * Run the non-IO destructors first: they may want to output
580 * error messages, close files etc */
d011219a 581 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 582 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
583 /* And if there are some very tenacious barnacles clinging to arrays,
584 closures, or what have you.... */
585 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
586 olddef = PL_defoutgv;
587 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
588 if (olddef && isGV_with_GP(olddef))
589 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
590 olderr = PL_stderrgv;
591 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
592 if (olderr && isGV_with_GP(olderr))
593 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
594 SvREFCNT_dec(olddef);
3280af22 595 PL_in_clean_objs = FALSE;
4561caa4
CS
596}
597
645c22ef
DM
598/* called by sv_clean_all() for each live SV */
599
600static void
de37a194 601do_clean_all(pTHX_ SV *const sv)
645c22ef 602{
97aff369 603 dVAR;
daba3364 604 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 605 /* don't clean pid table and strtab */
d17ea597 606 return;
cddfcddc 607 }
645c22ef
DM
608 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
609 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 610 SvREFCNT_dec_NN(sv);
645c22ef
DM
611}
612
613/*
614=for apidoc sv_clean_all
615
616Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 617cleanup. This function may have to be called multiple times to free
ff276b08 618SVs which are in complex self-referential hierarchies.
645c22ef
DM
619
620=cut
621*/
622
5226ed68 623I32
864dbfa3 624Perl_sv_clean_all(pTHX)
8990e307 625{
97aff369 626 dVAR;
5226ed68 627 I32 cleaned;
3280af22 628 PL_in_clean_all = TRUE;
055972dc 629 cleaned = visit(do_clean_all, 0,0);
5226ed68 630 return cleaned;
8990e307 631}
463ee0b2 632
5e258f8c
JC
633/*
634 ARENASETS: a meta-arena implementation which separates arena-info
635 into struct arena_set, which contains an array of struct
636 arena_descs, each holding info for a single arena. By separating
637 the meta-info from the arena, we recover the 1st slot, formerly
638 borrowed for list management. The arena_set is about the size of an
39244528 639 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
640
641 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
642 memory in the last arena-set (1/2 on average). In trade, we get
643 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 644 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
645 small arenas for large, rare body types, by changing array* fields
646 in body_details_by_type[] below.
5e258f8c 647*/
5e258f8c 648struct arena_desc {
398c677b
NC
649 char *arena; /* the raw storage, allocated aligned */
650 size_t size; /* its size ~4k typ */
e5973ed5 651 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
652};
653
e6148039
NC
654struct arena_set;
655
656/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 657 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
658 therefore likely to be 1 aligned memory page. */
659
660#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
661 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
662
663struct arena_set {
664 struct arena_set* next;
0a848332
NC
665 unsigned int set_size; /* ie ARENAS_PER_SET */
666 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
667 struct arena_desc set[ARENAS_PER_SET];
668};
669
645c22ef
DM
670/*
671=for apidoc sv_free_arenas
672
fde67290 673Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
674heads and bodies within the arenas must already have been freed.
675
676=cut
677*/
4633a7c4 678void
864dbfa3 679Perl_sv_free_arenas(pTHX)
4633a7c4 680{
97aff369 681 dVAR;
4633a7c4
LW
682 SV* sva;
683 SV* svanext;
0a848332 684 unsigned int i;
4633a7c4
LW
685
686 /* Free arenas here, but be careful about fake ones. (We assume
687 contiguity of the fake ones with the corresponding real ones.) */
688
3280af22 689 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 690 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 691 while (svanext && SvFAKE(svanext))
daba3364 692 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
693
694 if (!SvFAKE(sva))
1df70142 695 Safefree(sva);
4633a7c4 696 }
93e68bfb 697
5e258f8c 698 {
0a848332
NC
699 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
700
701 while (aroot) {
702 struct arena_set *current = aroot;
703 i = aroot->curr;
704 while (i--) {
5e258f8c
JC
705 assert(aroot->set[i].arena);
706 Safefree(aroot->set[i].arena);
707 }
0a848332
NC
708 aroot = aroot->next;
709 Safefree(current);
5e258f8c
JC
710 }
711 }
dc8220bf 712 PL_body_arenas = 0;
fdda85ca 713
0a848332
NC
714 i = PERL_ARENA_ROOTS_SIZE;
715 while (i--)
93e68bfb 716 PL_body_roots[i] = 0;
93e68bfb 717
3280af22
NIS
718 PL_sv_arenaroot = 0;
719 PL_sv_root = 0;
4633a7c4
LW
720}
721
bd81e77b
NC
722/*
723 Here are mid-level routines that manage the allocation of bodies out
724 of the various arenas. There are 5 kinds of arenas:
29489e7c 725
bd81e77b
NC
726 1. SV-head arenas, which are discussed and handled above
727 2. regular body arenas
728 3. arenas for reduced-size bodies
729 4. Hash-Entry arenas
29489e7c 730
bd81e77b
NC
731 Arena types 2 & 3 are chained by body-type off an array of
732 arena-root pointers, which is indexed by svtype. Some of the
733 larger/less used body types are malloced singly, since a large
734 unused block of them is wasteful. Also, several svtypes dont have
735 bodies; the data fits into the sv-head itself. The arena-root
736 pointer thus has a few unused root-pointers (which may be hijacked
737 later for arena types 4,5)
29489e7c 738
bd81e77b
NC
739 3 differs from 2 as an optimization; some body types have several
740 unused fields in the front of the structure (which are kept in-place
741 for consistency). These bodies can be allocated in smaller chunks,
742 because the leading fields arent accessed. Pointers to such bodies
743 are decremented to point at the unused 'ghost' memory, knowing that
744 the pointers are used with offsets to the real memory.
29489e7c 745
d2a0f284
JC
746
747=head1 SV-Body Allocation
748
749Allocation of SV-bodies is similar to SV-heads, differing as follows;
750the allocation mechanism is used for many body types, so is somewhat
751more complicated, it uses arena-sets, and has no need for still-live
752SV detection.
753
754At the outermost level, (new|del)_X*V macros return bodies of the
755appropriate type. These macros call either (new|del)_body_type or
756(new|del)_body_allocated macro pairs, depending on specifics of the
757type. Most body types use the former pair, the latter pair is used to
758allocate body types with "ghost fields".
759
760"ghost fields" are fields that are unused in certain types, and
69ba284b 761consequently don't need to actually exist. They are declared because
d2a0f284
JC
762they're part of a "base type", which allows use of functions as
763methods. The simplest examples are AVs and HVs, 2 aggregate types
764which don't use the fields which support SCALAR semantics.
765
69ba284b 766For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
767chunks, we thus avoid wasted memory for those unaccessed members.
768When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 769size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
770structure. (But things will all go boom if you write to the part that
771is "not there", because you'll be overwriting the last members of the
772preceding structure in memory.)
773
69ba284b
NC
774We calculate the correction using the STRUCT_OFFSET macro on the first
775member present. If the allocated structure is smaller (no initial NV
776actually allocated) then the net effect is to subtract the size of the NV
777from the pointer, to return a new pointer as if an initial NV were actually
778allocated. (We were using structures named *_allocated for this, but
779this turned out to be a subtle bug, because a structure without an NV
780could have a lower alignment constraint, but the compiler is allowed to
781optimised accesses based on the alignment constraint of the actual pointer
782to the full structure, for example, using a single 64 bit load instruction
783because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284
JC
784
785This is the same trick as was used for NV and IV bodies. Ironically it
786doesn't need to be used for NV bodies any more, because NV is now at
787the start of the structure. IV bodies don't need it either, because
788they are no longer allocated.
789
790In turn, the new_body_* allocators call S_new_body(), which invokes
791new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 792linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
793necessary to refresh an empty list. Then the lock is released, and
794the body is returned.
795
99816f8d 796Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
797bodies, which it strings into a linked list. It looks up arena-size
798and body-size from the body_details table described below, thus
799supporting the multiple body-types.
800
801If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
802the (new|del)_X*V macros are mapped directly to malloc/free.
803
d2a0f284
JC
804For each sv-type, struct body_details bodies_by_type[] carries
805parameters which control these aspects of SV handling:
806
807Arena_size determines whether arenas are used for this body type, and if
808so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
809zero, forcing individual mallocs and frees.
810
811Body_size determines how big a body is, and therefore how many fit into
812each arena. Offset carries the body-pointer adjustment needed for
69ba284b 813"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
814
815But its main purpose is to parameterize info needed in
816Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 817vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
818are used for this, except for arena_size.
819
820For the sv-types that have no bodies, arenas are not used, so those
821PL_body_roots[sv_type] are unused, and can be overloaded. In
822something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 823PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 824bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 825available in hv.c.
d2a0f284 826
29489e7c
DM
827*/
828
bd81e77b 829struct body_details {
0fb58b32 830 U8 body_size; /* Size to allocate */
10666ae3 831 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 832 U8 offset;
10666ae3
NC
833 unsigned int type : 4; /* We have space for a sanity check. */
834 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
835 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
836 unsigned int arena : 1; /* Allocated from an arena */
837 size_t arena_size; /* Size of arena to allocate */
bd81e77b 838};
29489e7c 839
bd81e77b
NC
840#define HADNV FALSE
841#define NONV TRUE
29489e7c 842
d2a0f284 843
bd81e77b
NC
844#ifdef PURIFY
845/* With -DPURFIY we allocate everything directly, and don't use arenas.
846 This seems a rather elegant way to simplify some of the code below. */
847#define HASARENA FALSE
848#else
849#define HASARENA TRUE
850#endif
851#define NOARENA FALSE
29489e7c 852
d2a0f284
JC
853/* Size the arenas to exactly fit a given number of bodies. A count
854 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
855 simplifying the default. If count > 0, the arena is sized to fit
856 only that many bodies, allowing arenas to be used for large, rare
857 bodies (XPVFM, XPVIO) without undue waste. The arena size is
858 limited by PERL_ARENA_SIZE, so we can safely oversize the
859 declarations.
860 */
95db5f15
MB
861#define FIT_ARENA0(body_size) \
862 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
863#define FIT_ARENAn(count,body_size) \
864 ( count * body_size <= PERL_ARENA_SIZE) \
865 ? count * body_size \
866 : FIT_ARENA0 (body_size)
867#define FIT_ARENA(count,body_size) \
868 count \
869 ? FIT_ARENAn (count, body_size) \
870 : FIT_ARENA0 (body_size)
d2a0f284 871
bd81e77b
NC
872/* Calculate the length to copy. Specifically work out the length less any
873 final padding the compiler needed to add. See the comment in sv_upgrade
874 for why copying the padding proved to be a bug. */
29489e7c 875
bd81e77b
NC
876#define copy_length(type, last_member) \
877 STRUCT_OFFSET(type, last_member) \
daba3364 878 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 879
bd81e77b 880static const struct body_details bodies_by_type[] = {
829cd18a
NC
881 /* HEs use this offset for their arena. */
882 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 883
1cb9cd50 884 /* The bind placeholder pretends to be an RV for now.
c6f8b1d0 885 Also it's marked as "can't upgrade" to stop anyone using it before it's
1cb9cd50 886 implemented. */
a9032aa0 887 { 0, 0, 0, SVt_DUMMY, TRUE, NONV, NOARENA, 0 },
1cb9cd50 888
db93c0c4
NC
889 /* IVs are in the head, so the allocation size is 0. */
890 { 0,
d2a0f284 891 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 892 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 893 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
894 },
895
6e128786
NC
896 { sizeof(NV), sizeof(NV),
897 STRUCT_OFFSET(XPVNV, xnv_u),
898 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 899
bc337e5c 900 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
901 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
902 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 903 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 904 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 905
bc337e5c 906 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
907 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
908 + STRUCT_OFFSET(XPV, xpv_cur),
909 SVt_PVIV, FALSE, NONV, HASARENA,
910 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 911
bc337e5c 912 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
913 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
914 + STRUCT_OFFSET(XPV, xpv_cur),
915 SVt_PVNV, FALSE, HADNV, HASARENA,
916 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 917
6e128786 918 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 919 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 920
601dfd0a
NC
921 { sizeof(regexp),
922 sizeof(regexp),
923 0,
08e44740 924 SVt_REGEXP, FALSE, NONV, HASARENA,
eaeb1e7f 925 FIT_ARENA(0, sizeof(regexp))
5c35adbb 926 },
4df7f6af 927
10666ae3 928 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
929 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
930
10666ae3 931 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
932 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
933
601dfd0a 934 { sizeof(XPVAV),
4f7003f5 935 copy_length(XPVAV, xav_alloc),
601dfd0a 936 0,
69ba284b 937 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 938 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 939
601dfd0a 940 { sizeof(XPVHV),
359164a0 941 copy_length(XPVHV, xhv_max),
601dfd0a 942 0,
69ba284b 943 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 944 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 945
601dfd0a
NC
946 { sizeof(XPVCV),
947 sizeof(XPVCV),
948 0,
69ba284b 949 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 950 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 951
601dfd0a
NC
952 { sizeof(XPVFM),
953 sizeof(XPVFM),
954 0,
69ba284b 955 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 956 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 957
601dfd0a
NC
958 { sizeof(XPVIO),
959 sizeof(XPVIO),
960 0,
b6f60916 961 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 962 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 963};
29489e7c 964
bd81e77b 965#define new_body_allocated(sv_type) \
d2a0f284 966 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 967 - bodies_by_type[sv_type].offset)
29489e7c 968
26359cfa
NC
969/* return a thing to the free list */
970
971#define del_body(thing, root) \
972 STMT_START { \
973 void ** const thing_copy = (void **)thing; \
974 *thing_copy = *root; \
975 *root = (void*)thing_copy; \
976 } STMT_END
29489e7c 977
bd81e77b 978#ifdef PURIFY
29489e7c 979
beeec492
NC
980#define new_XNV() safemalloc(sizeof(XPVNV))
981#define new_XPVNV() safemalloc(sizeof(XPVNV))
982#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 983
beeec492 984#define del_XPVGV(p) safefree(p)
29489e7c 985
bd81e77b 986#else /* !PURIFY */
29489e7c 987
65ac1738 988#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 989#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 990#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 991
26359cfa
NC
992#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
993 &PL_body_roots[SVt_PVGV])
1d7c1841 994
bd81e77b 995#endif /* PURIFY */
93e68bfb 996
bd81e77b 997/* no arena for you! */
93e68bfb 998
bd81e77b 999#define new_NOARENA(details) \
beeec492 1000 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1001#define new_NOARENAZ(details) \
beeec492 1002 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1003
1e30fcd5
NC
1004void *
1005Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1006 const size_t arena_size)
d2a0f284
JC
1007{
1008 dVAR;
1009 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1010 struct arena_desc *adesc;
1011 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1012 unsigned int curr;
d2a0f284
JC
1013 char *start;
1014 const char *end;
02982131 1015 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1016#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1017 static bool done_sanity_check;
1018
0b2d3faa
JH
1019 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1020 * variables like done_sanity_check. */
10666ae3 1021 if (!done_sanity_check) {
ea471437 1022 unsigned int i = SVt_LAST;
10666ae3
NC
1023
1024 done_sanity_check = TRUE;
1025
1026 while (i--)
1027 assert (bodies_by_type[i].type == i);
1028 }
1029#endif
1030
02982131 1031 assert(arena_size);
23e9d66c 1032
99816f8d
NC
1033 /* may need new arena-set to hold new arena */
1034 if (!aroot || aroot->curr >= aroot->set_size) {
1035 struct arena_set *newroot;
1036 Newxz(newroot, 1, struct arena_set);
1037 newroot->set_size = ARENAS_PER_SET;
1038 newroot->next = aroot;
1039 aroot = newroot;
1040 PL_body_arenas = (void *) newroot;
1041 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1042 }
1043
1044 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1045 curr = aroot->curr++;
1046 adesc = &(aroot->set[curr]);
1047 assert(!adesc->arena);
1048
1049 Newx(adesc->arena, good_arena_size, char);
1050 adesc->size = good_arena_size;
1051 adesc->utype = sv_type;
1052 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1053 curr, (void*)adesc->arena, (UV)good_arena_size));
1054
1055 start = (char *) adesc->arena;
d2a0f284 1056
29657bb6
NC
1057 /* Get the address of the byte after the end of the last body we can fit.
1058 Remember, this is integer division: */
02982131 1059 end = start + good_arena_size / body_size * body_size;
d2a0f284 1060
486ec47a 1061 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1062#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1063 DEBUG_m(PerlIO_printf(Perl_debug_log,
1064 "arena %p end %p arena-size %d (from %d) type %d "
1065 "size %d ct %d\n",
02982131
NC
1066 (void*)start, (void*)end, (int)good_arena_size,
1067 (int)arena_size, sv_type, (int)body_size,
1068 (int)good_arena_size / (int)body_size));
d8fca402 1069#else
d2a0f284
JC
1070 DEBUG_m(PerlIO_printf(Perl_debug_log,
1071 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1072 (void*)start, (void*)end,
02982131
NC
1073 (int)arena_size, sv_type, (int)body_size,
1074 (int)good_arena_size / (int)body_size));
d8fca402 1075#endif
d2a0f284
JC
1076 *root = (void *)start;
1077
29657bb6
NC
1078 while (1) {
1079 /* Where the next body would start: */
d2a0f284 1080 char * const next = start + body_size;
29657bb6
NC
1081
1082 if (next >= end) {
1083 /* This is the last body: */
1084 assert(next == end);
1085
1086 *(void **)start = 0;
1087 return *root;
1088 }
1089
d2a0f284
JC
1090 *(void**) start = (void *)next;
1091 start = next;
1092 }
d2a0f284
JC
1093}
1094
1095/* grab a new thing from the free list, allocating more if necessary.
1096 The inline version is used for speed in hot routines, and the
1097 function using it serves the rest (unless PURIFY).
1098*/
1099#define new_body_inline(xpv, sv_type) \
1100 STMT_START { \
1101 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1102 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1103 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1104 bodies_by_type[sv_type].body_size,\
1105 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1106 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1107 } STMT_END
1108
1109#ifndef PURIFY
1110
1111STATIC void *
de37a194 1112S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1113{
1114 dVAR;
1115 void *xpv;
1116 new_body_inline(xpv, sv_type);
1117 return xpv;
1118}
1119
1120#endif
93e68bfb 1121
238b27b3
NC
1122static const struct body_details fake_rv =
1123 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1124
bd81e77b
NC
1125/*
1126=for apidoc sv_upgrade
93e68bfb 1127
bd81e77b
NC
1128Upgrade an SV to a more complex form. Generally adds a new body type to the
1129SV, then copies across as much information as possible from the old body.
9521ca61
FC
1130It croaks if the SV is already in a more complex form than requested. You
1131generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1132before calling C<sv_upgrade>, and hence does not croak. See also
1133C<svtype>.
93e68bfb 1134
bd81e77b 1135=cut
93e68bfb 1136*/
93e68bfb 1137
bd81e77b 1138void
5aaab254 1139Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1140{
97aff369 1141 dVAR;
bd81e77b
NC
1142 void* old_body;
1143 void* new_body;
42d0e0b7 1144 const svtype old_type = SvTYPE(sv);
d2a0f284 1145 const struct body_details *new_type_details;
238b27b3 1146 const struct body_details *old_type_details
bd81e77b 1147 = bodies_by_type + old_type;
4df7f6af 1148 SV *referant = NULL;
cac9b346 1149
7918f24d
NC
1150 PERL_ARGS_ASSERT_SV_UPGRADE;
1151
1776cbe8
NC
1152 if (old_type == new_type)
1153 return;
1154
1155 /* This clause was purposefully added ahead of the early return above to
1156 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1157 inference by Nick I-S that it would fix other troublesome cases. See
1158 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1159
1160 Given that shared hash key scalars are no longer PVIV, but PV, there is
1161 no longer need to unshare so as to free up the IVX slot for its proper
1162 purpose. So it's safe to move the early return earlier. */
1163
093085a8 1164 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1165 sv_force_normal_flags(sv, 0);
1166 }
cac9b346 1167
bd81e77b 1168 old_body = SvANY(sv);
de042e1d 1169
bd81e77b
NC
1170 /* Copying structures onto other structures that have been neatly zeroed
1171 has a subtle gotcha. Consider XPVMG
cac9b346 1172
bd81e77b
NC
1173 +------+------+------+------+------+-------+-------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH |
1175 +------+------+------+------+------+-------+-------+
1176 0 4 8 12 16 20 24 28
645c22ef 1177
bd81e77b
NC
1178 where NVs are aligned to 8 bytes, so that sizeof that structure is
1179 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1180
bd81e77b
NC
1181 +------+------+------+------+------+-------+-------+------+
1182 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1183 +------+------+------+------+------+-------+-------+------+
1184 0 4 8 12 16 20 24 28 32
08742458 1185
bd81e77b 1186 so what happens if you allocate memory for this structure:
30f9da9e 1187
bd81e77b
NC
1188 +------+------+------+------+------+-------+-------+------+------+...
1189 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 0 4 8 12 16 20 24 28 32 36
bfc44f79 1192
bd81e77b
NC
1193 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1194 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1195 started out as zero once, but it's quite possible that it isn't. So now,
1196 rather than a nicely zeroed GP, you have it pointing somewhere random.
1197 Bugs ensue.
bfc44f79 1198
bd81e77b
NC
1199 (In fact, GP ends up pointing at a previous GP structure, because the
1200 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1201 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1202 this happens to be moot because XPVGV has been re-ordered, with GP
1203 no longer after STASH)
30f9da9e 1204
bd81e77b
NC
1205 So we are careful and work out the size of used parts of all the
1206 structures. */
bfc44f79 1207
bd81e77b
NC
1208 switch (old_type) {
1209 case SVt_NULL:
1210 break;
1211 case SVt_IV:
4df7f6af
NC
1212 if (SvROK(sv)) {
1213 referant = SvRV(sv);
238b27b3
NC
1214 old_type_details = &fake_rv;
1215 if (new_type == SVt_NV)
1216 new_type = SVt_PVNV;
4df7f6af
NC
1217 } else {
1218 if (new_type < SVt_PVIV) {
1219 new_type = (new_type == SVt_NV)
1220 ? SVt_PVNV : SVt_PVIV;
1221 }
bd81e77b
NC
1222 }
1223 break;
1224 case SVt_NV:
1225 if (new_type < SVt_PVNV) {
1226 new_type = SVt_PVNV;
bd81e77b
NC
1227 }
1228 break;
bd81e77b
NC
1229 case SVt_PV:
1230 assert(new_type > SVt_PV);
1231 assert(SVt_IV < SVt_PV);
1232 assert(SVt_NV < SVt_PV);
1233 break;
1234 case SVt_PVIV:
1235 break;
1236 case SVt_PVNV:
1237 break;
1238 case SVt_PVMG:
1239 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1240 there's no way that it can be safely upgraded, because perl.c
1241 expects to Safefree(SvANY(PL_mess_sv)) */
1242 assert(sv != PL_mess_sv);
1243 /* This flag bit is used to mean other things in other scalar types.
1244 Given that it only has meaning inside the pad, it shouldn't be set
1245 on anything that can get upgraded. */
00b1698f 1246 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1247 break;
1248 default:
2439e033 1249 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1250 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1251 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1252 }
3376de98 1253
2439e033 1254 if (UNLIKELY(old_type > new_type))
3376de98
NC
1255 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1256 (int)old_type, (int)new_type);
1257
2fa1109b 1258 new_type_details = bodies_by_type + new_type;
645c22ef 1259
bd81e77b
NC
1260 SvFLAGS(sv) &= ~SVTYPEMASK;
1261 SvFLAGS(sv) |= new_type;
932e9ff9 1262
ab4416c0
NC
1263 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1264 the return statements above will have triggered. */
1265 assert (new_type != SVt_NULL);
bd81e77b 1266 switch (new_type) {
bd81e77b
NC
1267 case SVt_IV:
1268 assert(old_type == SVt_NULL);
1269 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1270 SvIV_set(sv, 0);
1271 return;
1272 case SVt_NV:
1273 assert(old_type == SVt_NULL);
1274 SvANY(sv) = new_XNV();
1275 SvNV_set(sv, 0);
1276 return;
bd81e77b 1277 case SVt_PVHV:
bd81e77b 1278 case SVt_PVAV:
d2a0f284 1279 assert(new_type_details->body_size);
c1ae03ae
NC
1280
1281#ifndef PURIFY
1282 assert(new_type_details->arena);
d2a0f284 1283 assert(new_type_details->arena_size);
c1ae03ae 1284 /* This points to the start of the allocated area. */
d2a0f284
JC
1285 new_body_inline(new_body, new_type);
1286 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1287 new_body = ((char *)new_body) - new_type_details->offset;
1288#else
1289 /* We always allocated the full length item with PURIFY. To do this
1290 we fake things so that arena is false for all 16 types.. */
1291 new_body = new_NOARENAZ(new_type_details);
1292#endif
1293 SvANY(sv) = new_body;
1294 if (new_type == SVt_PVAV) {
1295 AvMAX(sv) = -1;
1296 AvFILLp(sv) = -1;
1297 AvREAL_only(sv);
64484faa 1298 if (old_type_details->body_size) {
ac572bf4
NC
1299 AvALLOC(sv) = 0;
1300 } else {
1301 /* It will have been zeroed when the new body was allocated.
1302 Lets not write to it, in case it confuses a write-back
1303 cache. */
1304 }
78ac7dd9
NC
1305 } else {
1306 assert(!SvOK(sv));
1307 SvOK_off(sv);
1308#ifndef NODEFAULT_SHAREKEYS
1309 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1310#endif
1311 HvMAX(sv) = 7; /* (start with 8 buckets) */
c1ae03ae 1312 }
aeb18a1e 1313
bd81e77b
NC
1314 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1315 The target created by newSVrv also is, and it can have magic.
1316 However, it never has SvPVX set.
1317 */
4df7f6af
NC
1318 if (old_type == SVt_IV) {
1319 assert(!SvROK(sv));
1320 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1321 assert(SvPVX_const(sv) == 0);
1322 }
aeb18a1e 1323
bd81e77b 1324 if (old_type >= SVt_PVMG) {
e736a858 1325 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1326 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1327 } else {
1328 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1329 }
1330 break;
93e68bfb 1331
bd81e77b
NC
1332 case SVt_PVIV:
1333 /* XXX Is this still needed? Was it ever needed? Surely as there is
1334 no route from NV to PVIV, NOK can never be true */
1335 assert(!SvNOKp(sv));
1336 assert(!SvNOK(sv));
1337 case SVt_PVIO:
1338 case SVt_PVFM:
bd81e77b
NC
1339 case SVt_PVGV:
1340 case SVt_PVCV:
1341 case SVt_PVLV:
12c45b25 1342 case SVt_REGEXP:
bd81e77b
NC
1343 case SVt_PVMG:
1344 case SVt_PVNV:
1345 case SVt_PV:
93e68bfb 1346
d2a0f284 1347 assert(new_type_details->body_size);
bd81e77b
NC
1348 /* We always allocated the full length item with PURIFY. To do this
1349 we fake things so that arena is false for all 16 types.. */
1350 if(new_type_details->arena) {
1351 /* This points to the start of the allocated area. */
d2a0f284
JC
1352 new_body_inline(new_body, new_type);
1353 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1354 new_body = ((char *)new_body) - new_type_details->offset;
1355 } else {
1356 new_body = new_NOARENAZ(new_type_details);
1357 }
1358 SvANY(sv) = new_body;
5e2fc214 1359
bd81e77b 1360 if (old_type_details->copy) {
f9ba3d20
NC
1361 /* There is now the potential for an upgrade from something without
1362 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1363 int offset = old_type_details->offset;
1364 int length = old_type_details->copy;
1365
1366 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1367 const int difference
f9ba3d20
NC
1368 = new_type_details->offset - old_type_details->offset;
1369 offset += difference;
1370 length -= difference;
1371 }
1372 assert (length >= 0);
1373
1374 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1375 char);
bd81e77b
NC
1376 }
1377
1378#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1379 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1380 * correct 0.0 for us. Otherwise, if the old body didn't have an
1381 * NV slot, but the new one does, then we need to initialise the
1382 * freshly created NV slot with whatever the correct bit pattern is
1383 * for 0.0 */
e22a937e
NC
1384 if (old_type_details->zero_nv && !new_type_details->zero_nv
1385 && !isGV_with_GP(sv))
bd81e77b 1386 SvNV_set(sv, 0);
82048762 1387#endif
5e2fc214 1388
2439e033 1389 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1390 IO * const io = MUTABLE_IO(sv);
d963bf01 1391 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1392
1393 SvOBJECT_on(io);
1394 /* Clear the stashcache because a new IO could overrule a package
1395 name */
103f5a36 1396 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1397 hv_clear(PL_stashcache);
1398
85dca89a 1399 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1400 IoPAGE_LEN(sv) = 60;
85dca89a 1401 }
2439e033 1402 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1403 sv->sv_u.svu_rx = (regexp *)new_body;
1404 else if (old_type < SVt_PV) {
4df7f6af
NC
1405 /* referant will be NULL unless the old type was SVt_IV emulating
1406 SVt_RV */
1407 sv->sv_u.svu_rv = referant;
1408 }
bd81e77b
NC
1409 break;
1410 default:
afd78fd5
JH
1411 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1412 (unsigned long)new_type);
bd81e77b 1413 }
73171d91 1414
db93c0c4 1415 if (old_type > SVt_IV) {
bd81e77b 1416#ifdef PURIFY
beeec492 1417 safefree(old_body);
bd81e77b 1418#else
bc786448
GG
1419 /* Note that there is an assumption that all bodies of types that
1420 can be upgraded came from arenas. Only the more complex non-
1421 upgradable types are allowed to be directly malloc()ed. */
1422 assert(old_type_details->arena);
bd81e77b
NC
1423 del_body((void*)((char*)old_body + old_type_details->offset),
1424 &PL_body_roots[old_type]);
1425#endif
1426 }
1427}
73171d91 1428
bd81e77b
NC
1429/*
1430=for apidoc sv_backoff
73171d91 1431
fde67290 1432Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1433wrapper instead.
73171d91 1434
bd81e77b 1435=cut
73171d91
NC
1436*/
1437
bd81e77b 1438int
5aaab254 1439Perl_sv_backoff(pTHX_ SV *const sv)
bd81e77b 1440{
69240efd 1441 STRLEN delta;
7a4bba22 1442 const char * const s = SvPVX_const(sv);
7918f24d
NC
1443
1444 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1445 PERL_UNUSED_CONTEXT;
7918f24d 1446
bd81e77b
NC
1447 assert(SvOOK(sv));
1448 assert(SvTYPE(sv) != SVt_PVHV);
1449 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1450
69240efd
NC
1451 SvOOK_offset(sv, delta);
1452
7a4bba22
NC
1453 SvLEN_set(sv, SvLEN(sv) + delta);
1454 SvPV_set(sv, SvPVX(sv) - delta);
1455 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1456 SvFLAGS(sv) &= ~SVf_OOK;
1457 return 0;
1458}
73171d91 1459
bd81e77b
NC
1460/*
1461=for apidoc sv_grow
73171d91 1462
bd81e77b
NC
1463Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1464upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1465Use the C<SvGROW> wrapper instead.
93e68bfb 1466
bd81e77b
NC
1467=cut
1468*/
93e68bfb 1469
bd81e77b 1470char *
5aaab254 1471Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1472{
eb578fdb 1473 char *s;
93e68bfb 1474
7918f24d
NC
1475 PERL_ARGS_ASSERT_SV_GROW;
1476
bd81e77b
NC
1477#ifdef HAS_64K_LIMIT
1478 if (newlen >= 0x10000) {
1479 PerlIO_printf(Perl_debug_log,
1480 "Allocation too large: %"UVxf"\n", (UV)newlen);
1481 my_exit(1);
1482 }
1483#endif /* HAS_64K_LIMIT */
1484 if (SvROK(sv))
1485 sv_unref(sv);
1486 if (SvTYPE(sv) < SVt_PV) {
1487 sv_upgrade(sv, SVt_PV);
1488 s = SvPVX_mutable(sv);
1489 }
1490 else if (SvOOK(sv)) { /* pv is offset? */
1491 sv_backoff(sv);
1492 s = SvPVX_mutable(sv);
1493 if (newlen > SvLEN(sv))
1494 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1495#ifdef HAS_64K_LIMIT
1496 if (newlen >= 0x10000)
1497 newlen = 0xFFFF;
1498#endif
1499 }
1500 else
db2c6cb3
FC
1501 {
1502 if (SvIsCOW(sv)) sv_force_normal(sv);
bd81e77b 1503 s = SvPVX_mutable(sv);
db2c6cb3 1504 }
aeb18a1e 1505
cbcb2a16
DM
1506#ifdef PERL_NEW_COPY_ON_WRITE
1507 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1508 * to store the COW count. So in general, allocate one more byte than
1509 * asked for, to make it likely this byte is always spare: and thus
1510 * make more strings COW-able.
1511 * If the new size is a big power of two, don't bother: we assume the
1512 * caller wanted a nice 2^N sized block and will be annoyed at getting
1513 * 2^N+1 */
1514 if (newlen & 0xff)
1515 newlen++;
1516#endif
1517
bd81e77b 1518 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1519 STRLEN minlen = SvCUR(sv);
1520 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1521 if (newlen < minlen)
1522 newlen = minlen;
aedff202 1523#ifndef Perl_safesysmalloc_size
bd81e77b 1524 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1525#endif
98653f18 1526 if (SvLEN(sv) && s) {
10edeb5d 1527 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1528 }
1529 else {
10edeb5d 1530 s = (char*)safemalloc(newlen);
bd81e77b
NC
1531 if (SvPVX_const(sv) && SvCUR(sv)) {
1532 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1533 }
1534 }
1535 SvPV_set(sv, s);
ca7c1a29 1536#ifdef Perl_safesysmalloc_size
98653f18
NC
1537 /* Do this here, do it once, do it right, and then we will never get
1538 called back into sv_grow() unless there really is some growing
1539 needed. */
ca7c1a29 1540 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1541#else
bd81e77b 1542 SvLEN_set(sv, newlen);
98653f18 1543#endif
bd81e77b
NC
1544 }
1545 return s;
1546}
aeb18a1e 1547
bd81e77b
NC
1548/*
1549=for apidoc sv_setiv
932e9ff9 1550
bd81e77b
NC
1551Copies an integer into the given SV, upgrading first if necessary.
1552Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1553
bd81e77b
NC
1554=cut
1555*/
463ee0b2 1556
bd81e77b 1557void
5aaab254 1558Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1559{
97aff369 1560 dVAR;
7918f24d
NC
1561
1562 PERL_ARGS_ASSERT_SV_SETIV;
1563
bd81e77b
NC
1564 SV_CHECK_THINKFIRST_COW_DROP(sv);
1565 switch (SvTYPE(sv)) {
1566 case SVt_NULL:
bd81e77b 1567 case SVt_NV:
3376de98 1568 sv_upgrade(sv, SVt_IV);
bd81e77b 1569 break;
bd81e77b
NC
1570 case SVt_PV:
1571 sv_upgrade(sv, SVt_PVIV);
1572 break;
463ee0b2 1573
bd81e77b 1574 case SVt_PVGV:
6e592b3a
BM
1575 if (!isGV_with_GP(sv))
1576 break;
bd81e77b
NC
1577 case SVt_PVAV:
1578 case SVt_PVHV:
1579 case SVt_PVCV:
1580 case SVt_PVFM:
1581 case SVt_PVIO:
22e74366 1582 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1583 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1584 OP_DESC(PL_op));
42d0e0b7 1585 default: NOOP;
bd81e77b
NC
1586 }
1587 (void)SvIOK_only(sv); /* validate number */
1588 SvIV_set(sv, i);
1589 SvTAINT(sv);
1590}
932e9ff9 1591
bd81e77b
NC
1592/*
1593=for apidoc sv_setiv_mg
d33b2eba 1594
bd81e77b 1595Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1596
bd81e77b
NC
1597=cut
1598*/
d33b2eba 1599
bd81e77b 1600void
5aaab254 1601Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1602{
7918f24d
NC
1603 PERL_ARGS_ASSERT_SV_SETIV_MG;
1604
bd81e77b
NC
1605 sv_setiv(sv,i);
1606 SvSETMAGIC(sv);
1607}
727879eb 1608
bd81e77b
NC
1609/*
1610=for apidoc sv_setuv
d33b2eba 1611
bd81e77b
NC
1612Copies an unsigned integer into the given SV, upgrading first if necessary.
1613Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1614
bd81e77b
NC
1615=cut
1616*/
d33b2eba 1617
bd81e77b 1618void
5aaab254 1619Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1620{
7918f24d
NC
1621 PERL_ARGS_ASSERT_SV_SETUV;
1622
013abb9b
NC
1623 /* With the if statement to ensure that integers are stored as IVs whenever
1624 possible:
bd81e77b 1625 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1626
bd81e77b
NC
1627 without
1628 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1629
013abb9b
NC
1630 If you wish to remove the following if statement, so that this routine
1631 (and its callers) always return UVs, please benchmark to see what the
1632 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1633 */
1634 if (u <= (UV)IV_MAX) {
1635 sv_setiv(sv, (IV)u);
1636 return;
1637 }
1638 sv_setiv(sv, 0);
1639 SvIsUV_on(sv);
1640 SvUV_set(sv, u);
1641}
d33b2eba 1642
bd81e77b
NC
1643/*
1644=for apidoc sv_setuv_mg
727879eb 1645
bd81e77b 1646Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1647
bd81e77b
NC
1648=cut
1649*/
5e2fc214 1650
bd81e77b 1651void
5aaab254 1652Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1653{
7918f24d
NC
1654 PERL_ARGS_ASSERT_SV_SETUV_MG;
1655
bd81e77b
NC
1656 sv_setuv(sv,u);
1657 SvSETMAGIC(sv);
1658}
5e2fc214 1659
954c1994 1660/*
bd81e77b 1661=for apidoc sv_setnv
954c1994 1662
bd81e77b
NC
1663Copies a double into the given SV, upgrading first if necessary.
1664Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1665
1666=cut
1667*/
1668
63f97190 1669void
5aaab254 1670Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1671{
97aff369 1672 dVAR;
7918f24d
NC
1673
1674 PERL_ARGS_ASSERT_SV_SETNV;
1675
bd81e77b
NC
1676 SV_CHECK_THINKFIRST_COW_DROP(sv);
1677 switch (SvTYPE(sv)) {
79072805 1678 case SVt_NULL:
79072805 1679 case SVt_IV:
bd81e77b 1680 sv_upgrade(sv, SVt_NV);
79072805
LW
1681 break;
1682 case SVt_PV:
79072805 1683 case SVt_PVIV:
bd81e77b 1684 sv_upgrade(sv, SVt_PVNV);
79072805 1685 break;
bd4b1eb5 1686
bd4b1eb5 1687 case SVt_PVGV:
6e592b3a
BM
1688 if (!isGV_with_GP(sv))
1689 break;
bd81e77b
NC
1690 case SVt_PVAV:
1691 case SVt_PVHV:
79072805 1692 case SVt_PVCV:
bd81e77b
NC
1693 case SVt_PVFM:
1694 case SVt_PVIO:
22e74366 1695 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1696 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1697 OP_DESC(PL_op));
42d0e0b7 1698 default: NOOP;
2068cd4d 1699 }
bd81e77b
NC
1700 SvNV_set(sv, num);
1701 (void)SvNOK_only(sv); /* validate number */
1702 SvTAINT(sv);
79072805
LW
1703}
1704
645c22ef 1705/*
bd81e77b 1706=for apidoc sv_setnv_mg
645c22ef 1707
bd81e77b 1708Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1709
1710=cut
1711*/
1712
bd81e77b 1713void
5aaab254 1714Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1715{
7918f24d
NC
1716 PERL_ARGS_ASSERT_SV_SETNV_MG;
1717
bd81e77b
NC
1718 sv_setnv(sv,num);
1719 SvSETMAGIC(sv);
79072805
LW
1720}
1721
bd81e77b
NC
1722/* Print an "isn't numeric" warning, using a cleaned-up,
1723 * printable version of the offending string
1724 */
954c1994 1725
bd81e77b 1726STATIC void
aad570aa 1727S_not_a_number(pTHX_ SV *const sv)
79072805 1728{
97aff369 1729 dVAR;
bd81e77b
NC
1730 SV *dsv;
1731 char tmpbuf[64];
1732 const char *pv;
94463019 1733
7918f24d
NC
1734 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1735
94463019 1736 if (DO_UTF8(sv)) {
84bafc02 1737 dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1738 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1739 } else {
1740 char *d = tmpbuf;
551405c4 1741 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1742 /* each *s can expand to 4 chars + "...\0",
1743 i.e. need room for 8 chars */
ecdeb87c 1744
00b6aa41
AL
1745 const char *s = SvPVX_const(sv);
1746 const char * const end = s + SvCUR(sv);
1747 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1748 int ch = *s & 0xFF;
1749 if (ch & 128 && !isPRINT_LC(ch)) {
1750 *d++ = 'M';
1751 *d++ = '-';
1752 ch &= 127;
1753 }
1754 if (ch == '\n') {
1755 *d++ = '\\';
1756 *d++ = 'n';
1757 }
1758 else if (ch == '\r') {
1759 *d++ = '\\';
1760 *d++ = 'r';
1761 }
1762 else if (ch == '\f') {
1763 *d++ = '\\';
1764 *d++ = 'f';
1765 }
1766 else if (ch == '\\') {
1767 *d++ = '\\';
1768 *d++ = '\\';
1769 }
1770 else if (ch == '\0') {
1771 *d++ = '\\';
1772 *d++ = '0';
1773 }
1774 else if (isPRINT_LC(ch))
1775 *d++ = ch;
1776 else {
1777 *d++ = '^';
1778 *d++ = toCTRL(ch);
1779 }
1780 }
1781 if (s < end) {
1782 *d++ = '.';
1783 *d++ = '.';
1784 *d++ = '.';
1785 }
1786 *d = '\0';
1787 pv = tmpbuf;
a0d0e21e 1788 }
a0d0e21e 1789
533c011a 1790 if (PL_op)
9014280d 1791 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1792 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1793 "Argument \"%s\" isn't numeric in %s", pv,
1794 OP_DESC(PL_op));
a0d0e21e 1795 else
9014280d 1796 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1797 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1798 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1799}
1800
c2988b20
NC
1801/*
1802=for apidoc looks_like_number
1803
645c22ef
DM
1804Test if the content of an SV looks like a number (or is a number).
1805C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1806non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1807ignored.
c2988b20
NC
1808
1809=cut
1810*/
1811
1812I32
aad570aa 1813Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1814{
eb578fdb 1815 const char *sbegin;
c2988b20
NC
1816 STRLEN len;
1817
7918f24d
NC
1818 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1819
f52e41ad
FC
1820 if (SvPOK(sv) || SvPOKp(sv)) {
1821 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1822 }
c2988b20 1823 else
e0ab1c0e 1824 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1825 return grok_number(sbegin, len, NULL);
1826}
25da4f38 1827
19f6321d
NC
1828STATIC bool
1829S_glob_2number(pTHX_ GV * const gv)
180488f8 1830{
7918f24d
NC
1831 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1832
675c862f
AL
1833 /* We know that all GVs stringify to something that is not-a-number,
1834 so no need to test that. */
1835 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1836 {
1837 SV *const buffer = sv_newmortal();
1838 gv_efullname3(buffer, gv, "*");
675c862f 1839 not_a_number(buffer);
8e629ff4 1840 }
675c862f
AL
1841 /* We just want something true to return, so that S_sv_2iuv_common
1842 can tail call us and return true. */
19f6321d 1843 return TRUE;
675c862f
AL
1844}
1845
25da4f38
IZ
1846/* Actually, ISO C leaves conversion of UV to IV undefined, but
1847 until proven guilty, assume that things are not that bad... */
1848
645c22ef
DM
1849/*
1850 NV_PRESERVES_UV:
1851
1852 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1853 an IV (an assumption perl has been based on to date) it becomes necessary
1854 to remove the assumption that the NV always carries enough precision to
1855 recreate the IV whenever needed, and that the NV is the canonical form.
1856 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1857 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1858 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1859 1) to distinguish between IV/UV/NV slots that have cached a valid
1860 conversion where precision was lost and IV/UV/NV slots that have a
1861 valid conversion which has lost no precision
645c22ef 1862 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1863 would lose precision, the precise conversion (or differently
1864 imprecise conversion) is also performed and cached, to prevent
1865 requests for different numeric formats on the same SV causing
1866 lossy conversion chains. (lossless conversion chains are perfectly
1867 acceptable (still))
1868
1869
1870 flags are used:
1871 SvIOKp is true if the IV slot contains a valid value
1872 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1873 SvNOKp is true if the NV slot contains a valid value
1874 SvNOK is true only if the NV value is accurate
1875
1876 so
645c22ef 1877 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1878 IV(or UV) would lose accuracy over a direct conversion from PV to
1879 IV(or UV). If it would, cache both conversions, return NV, but mark
1880 SV as IOK NOKp (ie not NOK).
1881
645c22ef 1882 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1883 NV would lose accuracy over a direct conversion from PV to NV. If it
1884 would, cache both conversions, flag similarly.
1885
1886 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1887 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1888 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1889 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1890 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1891
645c22ef
DM
1892 The benefit of this is that operations such as pp_add know that if
1893 SvIOK is true for both left and right operands, then integer addition
1894 can be used instead of floating point (for cases where the result won't
1895 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1896 loss of precision compared with integer addition.
1897
1898 * making IV and NV equal status should make maths accurate on 64 bit
1899 platforms
1900 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1901 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1902 looking for SvIOK and checking for overflow will not outweigh the
1903 fp to integer speedup)
1904 * will slow down integer operations (callers of SvIV) on "inaccurate"
1905 values, as the change from SvIOK to SvIOKp will cause a call into
1906 sv_2iv each time rather than a macro access direct to the IV slot
1907 * should speed up number->string conversion on integers as IV is
645c22ef 1908 favoured when IV and NV are equally accurate
28e5dec8
JH
1909
1910 ####################################################################
645c22ef
DM
1911 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1912 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1913 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1914 ####################################################################
1915
645c22ef 1916 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1917 performance ratio.
1918*/
1919
1920#ifndef NV_PRESERVES_UV
645c22ef
DM
1921# define IS_NUMBER_UNDERFLOW_IV 1
1922# define IS_NUMBER_UNDERFLOW_UV 2
1923# define IS_NUMBER_IV_AND_UV 2
1924# define IS_NUMBER_OVERFLOW_IV 4
1925# define IS_NUMBER_OVERFLOW_UV 5
1926
1927/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1928
1929/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1930STATIC int
5aaab254 1931S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
1932# ifdef DEBUGGING
1933 , I32 numtype
1934# endif
1935 )
28e5dec8 1936{
97aff369 1937 dVAR;
7918f24d
NC
1938
1939 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1940
3f7c398e 1941 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
1942 if (SvNVX(sv) < (NV)IV_MIN) {
1943 (void)SvIOKp_on(sv);
1944 (void)SvNOK_on(sv);
45977657 1945 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1946 return IS_NUMBER_UNDERFLOW_IV;
1947 }
1948 if (SvNVX(sv) > (NV)UV_MAX) {
1949 (void)SvIOKp_on(sv);
1950 (void)SvNOK_on(sv);
1951 SvIsUV_on(sv);
607fa7f2 1952 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1953 return IS_NUMBER_OVERFLOW_UV;
1954 }
c2988b20
NC
1955 (void)SvIOKp_on(sv);
1956 (void)SvNOK_on(sv);
1957 /* Can't use strtol etc to convert this string. (See truth table in
1958 sv_2iv */
1959 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1960 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1961 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1962 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1963 } else {
1964 /* Integer is imprecise. NOK, IOKp */
1965 }
1966 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1967 }
1968 SvIsUV_on(sv);
607fa7f2 1969 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1970 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1971 if (SvUVX(sv) == UV_MAX) {
1972 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1973 possibly be preserved by NV. Hence, it must be overflow.
1974 NOK, IOKp */
1975 return IS_NUMBER_OVERFLOW_UV;
1976 }
1977 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1978 } else {
1979 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1980 }
c2988b20 1981 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1982}
645c22ef
DM
1983#endif /* !NV_PRESERVES_UV*/
1984
af359546 1985STATIC bool
7918f24d
NC
1986S_sv_2iuv_common(pTHX_ SV *const sv)
1987{
97aff369 1988 dVAR;
7918f24d
NC
1989
1990 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1991
af359546 1992 if (SvNOKp(sv)) {
28e5dec8
JH
1993 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1994 * without also getting a cached IV/UV from it at the same time
1995 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1996 * IV or UV at same time to avoid this. */
1997 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1998
1999 if (SvTYPE(sv) == SVt_NV)
2000 sv_upgrade(sv, SVt_PVNV);
2001
28e5dec8
JH
2002 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2003 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2004 certainly cast into the IV range at IV_MAX, whereas the correct
2005 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2006 cases go to UV */
cab190d4
JD
2007#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2008 if (Perl_isnan(SvNVX(sv))) {
2009 SvUV_set(sv, 0);
2010 SvIsUV_on(sv);
fdbe6d7c 2011 return FALSE;
cab190d4 2012 }
cab190d4 2013#endif
28e5dec8 2014 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2015 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2016 if (SvNVX(sv) == (NV) SvIVX(sv)
2017#ifndef NV_PRESERVES_UV
2018 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2019 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2020 /* Don't flag it as "accurately an integer" if the number
2021 came from a (by definition imprecise) NV operation, and
2022 we're outside the range of NV integer precision */
2023#endif
2024 ) {
a43d94f2
NC
2025 if (SvNOK(sv))
2026 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2027 else {
2028 /* scalar has trailing garbage, eg "42a" */
2029 }
28e5dec8 2030 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2031 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2032 PTR2UV(sv),
2033 SvNVX(sv),
2034 SvIVX(sv)));
2035
2036 } else {
2037 /* IV not precise. No need to convert from PV, as NV
2038 conversion would already have cached IV if it detected
2039 that PV->IV would be better than PV->NV->IV
2040 flags already correct - don't set public IOK. */
2041 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2042 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2043 PTR2UV(sv),
2044 SvNVX(sv),
2045 SvIVX(sv)));
2046 }
2047 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2048 but the cast (NV)IV_MIN rounds to a the value less (more
2049 negative) than IV_MIN which happens to be equal to SvNVX ??
2050 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2051 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2052 (NV)UVX == NVX are both true, but the values differ. :-(
2053 Hopefully for 2s complement IV_MIN is something like
2054 0x8000000000000000 which will be exact. NWC */
d460ef45 2055 }
25da4f38 2056 else {
607fa7f2 2057 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2058 if (
2059 (SvNVX(sv) == (NV) SvUVX(sv))
2060#ifndef NV_PRESERVES_UV
2061 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2062 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2063 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2064 /* Don't flag it as "accurately an integer" if the number
2065 came from a (by definition imprecise) NV operation, and
2066 we're outside the range of NV integer precision */
2067#endif
a43d94f2 2068 && SvNOK(sv)
28e5dec8
JH
2069 )
2070 SvIOK_on(sv);
25da4f38 2071 SvIsUV_on(sv);
1c846c1f 2072 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2073 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2074 PTR2UV(sv),
57def98f
JH
2075 SvUVX(sv),
2076 SvUVX(sv)));
25da4f38 2077 }
748a9306 2078 }
cd84013a 2079 else if (SvPOKp(sv)) {
c2988b20 2080 UV value;
504618e9 2081 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2082 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2083 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2084 the same as the direct translation of the initial string
2085 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2086 be careful to ensure that the value with the .456 is around if the
2087 NV value is requested in the future).
1c846c1f 2088
af359546 2089 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2090 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2091 cache the NV if we are sure it's not needed.
25da4f38 2092 */
16b7a9a4 2093
c2988b20
NC
2094 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2095 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2096 == IS_NUMBER_IN_UV) {
5e045b90 2097 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2098 if (SvTYPE(sv) < SVt_PVIV)
2099 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2100 (void)SvIOK_on(sv);
c2988b20
NC
2101 } else if (SvTYPE(sv) < SVt_PVNV)
2102 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2103
f2524eef 2104 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2105 we aren't going to call atof() below. If NVs don't preserve UVs
2106 then the value returned may have more precision than atof() will
2107 return, even though value isn't perfectly accurate. */
2108 if ((numtype & (IS_NUMBER_IN_UV
2109#ifdef NV_PRESERVES_UV
2110 | IS_NUMBER_NOT_INT
2111#endif
2112 )) == IS_NUMBER_IN_UV) {
2113 /* This won't turn off the public IOK flag if it was set above */
2114 (void)SvIOKp_on(sv);
2115
2116 if (!(numtype & IS_NUMBER_NEG)) {
2117 /* positive */;
2118 if (value <= (UV)IV_MAX) {
45977657 2119 SvIV_set(sv, (IV)value);
c2988b20 2120 } else {
af359546 2121 /* it didn't overflow, and it was positive. */
607fa7f2 2122 SvUV_set(sv, value);
c2988b20
NC
2123 SvIsUV_on(sv);
2124 }
2125 } else {
2126 /* 2s complement assumption */
2127 if (value <= (UV)IV_MIN) {
45977657 2128 SvIV_set(sv, -(IV)value);
c2988b20
NC
2129 } else {
2130 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2131 I'm assuming it will be rare. */
c2988b20
NC
2132 if (SvTYPE(sv) < SVt_PVNV)
2133 sv_upgrade(sv, SVt_PVNV);
2134 SvNOK_on(sv);
2135 SvIOK_off(sv);
2136 SvIOKp_on(sv);
9d6ce603 2137 SvNV_set(sv, -(NV)value);
45977657 2138 SvIV_set(sv, IV_MIN);
c2988b20
NC
2139 }
2140 }
2141 }
2142 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2143 will be in the previous block to set the IV slot, and the next
2144 block to set the NV slot. So no else here. */
2145
2146 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2147 != IS_NUMBER_IN_UV) {
2148 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2149 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2150
c2988b20
NC
2151 if (! numtype && ckWARN(WARN_NUMERIC))
2152 not_a_number(sv);
28e5dec8 2153
65202027 2154#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2155 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2156 PTR2UV(sv), SvNVX(sv)));
65202027 2157#else
1779d84d 2158 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2159 PTR2UV(sv), SvNVX(sv)));
65202027 2160#endif
28e5dec8 2161
28e5dec8 2162#ifdef NV_PRESERVES_UV
af359546
NC
2163 (void)SvIOKp_on(sv);
2164 (void)SvNOK_on(sv);
2165 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2166 SvIV_set(sv, I_V(SvNVX(sv)));
2167 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2168 SvIOK_on(sv);
2169 } else {
6f207bd3 2170 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2171 }
2172 /* UV will not work better than IV */
2173 } else {
2174 if (SvNVX(sv) > (NV)UV_MAX) {
2175 SvIsUV_on(sv);
2176 /* Integer is inaccurate. NOK, IOKp, is UV */
2177 SvUV_set(sv, UV_MAX);
af359546
NC
2178 } else {
2179 SvUV_set(sv, U_V(SvNVX(sv)));
2180 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2181 NV preservse UV so can do correct comparison. */
2182 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2183 SvIOK_on(sv);
af359546 2184 } else {
6f207bd3 2185 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2186 }
2187 }
4b0c9573 2188 SvIsUV_on(sv);
af359546 2189 }
28e5dec8 2190#else /* NV_PRESERVES_UV */
c2988b20
NC
2191 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2192 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2193 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2194 grok_number above. The NV slot has just been set using
2195 Atof. */
560b0c46 2196 SvNOK_on(sv);
c2988b20
NC
2197 assert (SvIOKp(sv));
2198 } else {
2199 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2200 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2201 /* Small enough to preserve all bits. */
2202 (void)SvIOKp_on(sv);
2203 SvNOK_on(sv);
45977657 2204 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2205 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2206 SvIOK_on(sv);
2207 /* Assumption: first non-preserved integer is < IV_MAX,
2208 this NV is in the preserved range, therefore: */
2209 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2210 < (UV)IV_MAX)) {
32fdb065 2211 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
2212 }
2213 } else {
2214 /* IN_UV NOT_INT
2215 0 0 already failed to read UV.
2216 0 1 already failed to read UV.
2217 1 0 you won't get here in this case. IV/UV
2218 slot set, public IOK, Atof() unneeded.
2219 1 1 already read UV.
2220 so there's no point in sv_2iuv_non_preserve() attempting
2221 to use atol, strtol, strtoul etc. */
47031da6 2222# ifdef DEBUGGING
40a17c4c 2223 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2224# else
2225 sv_2iuv_non_preserve (sv);
2226# endif
c2988b20
NC
2227 }
2228 }
28e5dec8 2229#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2230 /* It might be more code efficient to go through the entire logic above
2231 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2232 gets complex and potentially buggy, so more programmer efficient
2233 to do it this way, by turning off the public flags: */
2234 if (!numtype)
2235 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2236 }
af359546
NC
2237 }
2238 else {
675c862f 2239 if (isGV_with_GP(sv))
159b6efe 2240 return glob_2number(MUTABLE_GV(sv));
180488f8 2241
9a214eec 2242 if (!SvPADTMP(sv)) {
af359546
NC
2243 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2244 report_uninit(sv);
2245 }
25da4f38
IZ
2246 if (SvTYPE(sv) < SVt_IV)
2247 /* Typically the caller expects that sv_any is not NULL now. */
2248 sv_upgrade(sv, SVt_IV);
af359546
NC
2249 /* Return 0 from the caller. */
2250 return TRUE;
2251 }
2252 return FALSE;
2253}
2254
2255/*
2256=for apidoc sv_2iv_flags
2257
2258Return the integer value of an SV, doing any necessary string
2259conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2260Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2261
2262=cut
2263*/
2264
2265IV
5aaab254 2266Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2267{
97aff369 2268 dVAR;
4bac9ae4 2269
af359546 2270 if (!sv)
a0d0e21e 2271 return 0;
4bac9ae4
CS
2272
2273 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2274 mg_get(sv);
2275
2276 if (SvROK(sv)) {
2277 if (SvAMAGIC(sv)) {
2278 SV * tmpstr;
2279 if (flags & SV_SKIP_OVERLOAD)
2280 return 0;
2281 tmpstr = AMG_CALLunary(sv, numer_amg);
2282 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2283 return SvIV(tmpstr);
2284 }
2285 }
2286 return PTR2IV(SvRV(sv));
2287 }
2288
8d919b0a 2289 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2290 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2291 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2292 In practice they are extremely unlikely to actually get anywhere
2293 accessible by user Perl code - the only way that I'm aware of is when
2294 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2295
2296 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2297 */
8d919b0a 2298 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2299 {
71c558c3 2300 UV value;
8d919b0a
FC
2301 const char * const ptr =
2302 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2303 const int numtype
8d919b0a 2304 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2305
2306 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2307 == IS_NUMBER_IN_UV) {
2308 /* It's definitely an integer */
2309 if (numtype & IS_NUMBER_NEG) {
2310 if (value < (UV)IV_MIN)
2311 return -(IV)value;
2312 } else {
2313 if (value < (UV)IV_MAX)
2314 return (IV)value;
2315 }
2316 }
2317 if (!numtype) {
2318 if (ckWARN(WARN_NUMERIC))
2319 not_a_number(sv);
2320 }
8d919b0a 2321 return I_V(Atof(ptr));
e20b6c3b 2322 }
4bac9ae4
CS
2323 }
2324
2325 if (SvTHINKFIRST(sv)) {
73381561 2326#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2327 if (SvIsCOW(sv)) {
2328 sv_force_normal_flags(sv, 0);
2329 }
73381561 2330#endif
af359546
NC
2331 if (SvREADONLY(sv) && !SvOK(sv)) {
2332 if (ckWARN(WARN_UNINITIALIZED))
2333 report_uninit(sv);
2334 return 0;
2335 }
2336 }
4bac9ae4 2337
af359546
NC
2338 if (!SvIOKp(sv)) {
2339 if (S_sv_2iuv_common(aTHX_ sv))
2340 return 0;
79072805 2341 }
4bac9ae4 2342
1d7c1841
GS
2343 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2344 PTR2UV(sv),SvIVX(sv)));
25da4f38 2345 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2346}
2347
645c22ef 2348/*
891f9566 2349=for apidoc sv_2uv_flags
645c22ef
DM
2350
2351Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2352conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2353Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2354
2355=cut
2356*/
2357
ff68c719 2358UV
5aaab254 2359Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2360{
97aff369 2361 dVAR;
4bac9ae4 2362
ff68c719
PP
2363 if (!sv)
2364 return 0;
4bac9ae4
CS
2365
2366 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2367 mg_get(sv);
2368
2369 if (SvROK(sv)) {
2370 if (SvAMAGIC(sv)) {
2371 SV *tmpstr;
2372 if (flags & SV_SKIP_OVERLOAD)
2373 return 0;
2374 tmpstr = AMG_CALLunary(sv, numer_amg);
2375 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2376 return SvUV(tmpstr);
2377 }
2378 }
2379 return PTR2UV(SvRV(sv));
2380 }
2381
8d919b0a 2382 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2383 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2384 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2385 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2386 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2387 {
71c558c3 2388 UV value;
8d919b0a
FC
2389 const char * const ptr =
2390 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2391 const int numtype
8d919b0a 2392 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2393
2394 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2395 == IS_NUMBER_IN_UV) {
2396 /* It's definitely an integer */
2397 if (!(numtype & IS_NUMBER_NEG))
2398 return value;
2399 }
2400 if (!numtype) {
2401 if (ckWARN(WARN_NUMERIC))
2402 not_a_number(sv);
2403 }
8d919b0a 2404 return U_V(Atof(ptr));
e20b6c3b 2405 }
4bac9ae4
CS
2406 }
2407
2408 if (SvTHINKFIRST(sv)) {
73381561 2409#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2410 if (SvIsCOW(sv)) {
2411 sv_force_normal_flags(sv, 0);
8a818333 2412 }
73381561 2413#endif
0336b60e 2414 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2415 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2416 report_uninit(sv);
ff68c719
PP
2417 return 0;
2418 }
2419 }
4bac9ae4 2420
af359546
NC
2421 if (!SvIOKp(sv)) {
2422 if (S_sv_2iuv_common(aTHX_ sv))
2423 return 0;
ff68c719 2424 }
25da4f38 2425
1d7c1841
GS
2426 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2427 PTR2UV(sv),SvUVX(sv)));
25da4f38 2428 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2429}
2430
645c22ef 2431/*
196007d1 2432=for apidoc sv_2nv_flags
645c22ef
DM
2433
2434Return the num value of an SV, doing any necessary string or integer
fde67290 2435conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2436Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2437
2438=cut
2439*/
2440
65202027 2441NV
5aaab254 2442Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2443{
97aff369 2444 dVAR;
79072805
LW
2445 if (!sv)
2446 return 0.0;
8d919b0a 2447 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2448 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2449 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2450 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2451 const char *ptr;
39d5de13
DM
2452 if (flags & SV_GMAGIC)
2453 mg_get(sv);
463ee0b2
LW
2454 if (SvNOKp(sv))
2455 return SvNVX(sv);
cd84013a 2456 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2457 ptr = SvPVX_const(sv);
2458 grokpv:
041457d9 2459 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2460 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2461 not_a_number(sv);
8d919b0a 2462 return Atof(ptr);
a0d0e21e 2463 }
25da4f38 2464 if (SvIOKp(sv)) {
1c846c1f 2465 if (SvIsUV(sv))
65202027 2466 return (NV)SvUVX(sv);
25da4f38 2467 else
65202027 2468 return (NV)SvIVX(sv);
47a72cb8
NC
2469 }
2470 if (SvROK(sv)) {
2471 goto return_rok;
2472 }
8d919b0a
FC
2473 if (isREGEXP(sv)) {
2474 ptr = RX_WRAPPED((REGEXP *)sv);
2475 goto grokpv;
2476 }
47a72cb8
NC
2477 assert(SvTYPE(sv) >= SVt_PVMG);
2478 /* This falls through to the report_uninit near the end of the
2479 function. */
2480 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2481 if (SvROK(sv)) {
47a72cb8 2482 return_rok:
deb46114 2483 if (SvAMAGIC(sv)) {
aee036bb
DM
2484 SV *tmpstr;
2485 if (flags & SV_SKIP_OVERLOAD)
2486 return 0;
31d632c3 2487 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2488 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2489 return SvNV(tmpstr);
2490 }
2491 }
2492 return PTR2NV(SvRV(sv));
a0d0e21e 2493 }
73381561 2494#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2495 if (SvIsCOW(sv)) {
2496 sv_force_normal_flags(sv, 0);
8a818333 2497 }
73381561 2498#endif
0336b60e 2499 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2500 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2501 report_uninit(sv);
ed6116ce
LW
2502 return 0.0;
2503 }
79072805
LW
2504 }
2505 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2506 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2507 sv_upgrade(sv, SVt_NV);
906f284f 2508#ifdef USE_LONG_DOUBLE
097ee67d 2509 DEBUG_c({
f93f4e46 2510 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2511 PerlIO_printf(Perl_debug_log,
2512 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2513 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2514 RESTORE_NUMERIC_LOCAL();
2515 });
65202027 2516#else
572bbb43 2517 DEBUG_c({
f93f4e46 2518 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2519 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2520 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2521 RESTORE_NUMERIC_LOCAL();
2522 });
572bbb43 2523#endif
79072805
LW
2524 }
2525 else if (SvTYPE(sv) < SVt_PVNV)
2526 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2527 if (SvNOKp(sv)) {
2528 return SvNVX(sv);
61604483 2529 }
59d8ce62 2530 if (SvIOKp(sv)) {
9d6ce603 2531 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2532#ifdef NV_PRESERVES_UV
a43d94f2
NC
2533 if (SvIOK(sv))
2534 SvNOK_on(sv);
2535 else
2536 SvNOKp_on(sv);
28e5dec8
JH
2537#else
2538 /* Only set the public NV OK flag if this NV preserves the IV */
2539 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2540 if (SvIOK(sv) &&
2541 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2542 : (SvIVX(sv) == I_V(SvNVX(sv))))
2543 SvNOK_on(sv);
2544 else
2545 SvNOKp_on(sv);
2546#endif
93a17b20 2547 }
cd84013a 2548 else if (SvPOKp(sv)) {
c2988b20 2549 UV value;
3f7c398e 2550 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2551 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2552 not_a_number(sv);
28e5dec8 2553#ifdef NV_PRESERVES_UV
c2988b20
NC
2554 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2555 == IS_NUMBER_IN_UV) {
5e045b90 2556 /* It's definitely an integer */
9d6ce603 2557 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2558 } else
3f7c398e 2559 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2560 if (numtype)
2561 SvNOK_on(sv);
2562 else
2563 SvNOKp_on(sv);
28e5dec8 2564#else
3f7c398e 2565 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2566 /* Only set the public NV OK flag if this NV preserves the value in
2567 the PV at least as well as an IV/UV would.
2568 Not sure how to do this 100% reliably. */
2569 /* if that shift count is out of range then Configure's test is
2570 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2571 UV_BITS */
2572 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2573 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2574 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2575 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2576 /* Can't use strtol etc to convert this string, so don't try.
2577 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2578 SvNOK_on(sv);
2579 } else {
2580 /* value has been set. It may not be precise. */
2581 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2582 /* 2s complement assumption for (UV)IV_MIN */
2583 SvNOK_on(sv); /* Integer is too negative. */
2584 } else {
2585 SvNOKp_on(sv);
2586 SvIOKp_on(sv);
6fa402ec 2587
c2988b20 2588 if (numtype & IS_NUMBER_NEG) {
45977657 2589 SvIV_set(sv, -(IV)value);
c2988b20 2590 } else if (value <= (UV)IV_MAX) {
45977657 2591 SvIV_set(sv, (IV)value);
c2988b20 2592 } else {
607fa7f2 2593 SvUV_set(sv, value);
c2988b20
NC
2594 SvIsUV_on(sv);
2595 }
2596
2597 if (numtype & IS_NUMBER_NOT_INT) {
2598 /* I believe that even if the original PV had decimals,
2599 they are lost beyond the limit of the FP precision.
2600 However, neither is canonical, so both only get p
2601 flags. NWC, 2000/11/25 */
2602 /* Both already have p flags, so do nothing */
2603 } else {
66a1b24b 2604 const NV nv = SvNVX(sv);
c2988b20
NC
2605 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2606 if (SvIVX(sv) == I_V(nv)) {
2607 SvNOK_on(sv);
c2988b20 2608 } else {
c2988b20
NC
2609 /* It had no "." so it must be integer. */
2610 }
00b6aa41 2611 SvIOK_on(sv);
c2988b20
NC
2612 } else {
2613 /* between IV_MAX and NV(UV_MAX).
2614 Could be slightly > UV_MAX */
6fa402ec 2615
c2988b20
NC
2616 if (numtype & IS_NUMBER_NOT_INT) {
2617 /* UV and NV both imprecise. */
2618 } else {
66a1b24b 2619 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2620
2621 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2622 SvNOK_on(sv);
c2988b20 2623 }
00b6aa41 2624 SvIOK_on(sv);
c2988b20
NC
2625 }
2626 }
2627 }
2628 }
2629 }
a43d94f2
NC
2630 /* It might be more code efficient to go through the entire logic above
2631 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2632 gets complex and potentially buggy, so more programmer efficient
2633 to do it this way, by turning off the public flags: */
2634 if (!numtype)
2635 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2636#endif /* NV_PRESERVES_UV */
93a17b20 2637 }
79072805 2638 else {
f7877b28 2639 if (isGV_with_GP(sv)) {
159b6efe 2640 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2641 return 0.0;
2642 }
2643
9a214eec 2644 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2645 report_uninit(sv);
7e25a7e9
NC
2646 assert (SvTYPE(sv) >= SVt_NV);
2647 /* Typically the caller expects that sv_any is not NULL now. */
2648 /* XXX Ilya implies that this is a bug in callers that assume this
2649 and ideally should be fixed. */
a0d0e21e 2650 return 0.0;
79072805 2651 }
572bbb43 2652#if defined(USE_LONG_DOUBLE)
097ee67d 2653 DEBUG_c({
f93f4e46 2654 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2655 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2656 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2657 RESTORE_NUMERIC_LOCAL();
2658 });
65202027 2659#else
572bbb43 2660 DEBUG_c({
f93f4e46 2661 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2662 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2663 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2664 RESTORE_NUMERIC_LOCAL();
2665 });
572bbb43 2666#endif
463ee0b2 2667 return SvNVX(sv);
79072805
LW
2668}
2669
800401ee
JH
2670/*
2671=for apidoc sv_2num
2672
2673Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2674reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2675access this function.
800401ee
JH
2676
2677=cut
2678*/
2679
2680SV *
5aaab254 2681Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2682{
7918f24d
NC
2683 PERL_ARGS_ASSERT_SV_2NUM;
2684
b9ee0594
RGS
2685 if (!SvROK(sv))
2686 return sv;
800401ee 2687 if (SvAMAGIC(sv)) {
31d632c3 2688 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2689 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2690 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2691 return sv_2num(tmpsv);
2692 }
2693 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2694}
2695
645c22ef
DM
2696/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2697 * UV as a string towards the end of buf, and return pointers to start and
2698 * end of it.
2699 *
2700 * We assume that buf is at least TYPE_CHARS(UV) long.
2701 */
2702
864dbfa3 2703static char *
5de3775c 2704S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2705{
25da4f38 2706 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2707 char * const ebuf = ptr;
25da4f38 2708 int sign;
25da4f38 2709
7918f24d
NC
2710 PERL_ARGS_ASSERT_UIV_2BUF;
2711
25da4f38
IZ
2712 if (is_uv)
2713 sign = 0;
2714 else if (iv >= 0) {
2715 uv = iv;
2716 sign = 0;
2717 } else {
2718 uv = -iv;
2719 sign = 1;
2720 }
2721 do {
eb160463 2722 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2723 } while (uv /= 10);
2724 if (sign)
2725 *--ptr = '-';
2726 *peob = ebuf;
2727 return ptr;
2728}
2729
645c22ef
DM
2730/*
2731=for apidoc sv_2pv_flags
2732
ff276b08 2733Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2734If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2735string if necessary. Normally invoked via the C<SvPV_flags> macro.
2736C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2737
2738=cut
2739*/
2740
8d6d96c1 2741char *
5aaab254 2742Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2743{
97aff369 2744 dVAR;
eb578fdb 2745 char *s;
79072805 2746
463ee0b2 2747 if (!sv) {
cdb061a3
NC
2748 if (lp)
2749 *lp = 0;
73d840c0 2750 return (char *)"";
463ee0b2 2751 }
4bac9ae4
CS
2752 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2753 mg_get(sv);
2754 if (SvROK(sv)) {
2755 if (SvAMAGIC(sv)) {
2756 SV *tmpstr;
2757 if (flags & SV_SKIP_OVERLOAD)
2758 return NULL;
2759 tmpstr = AMG_CALLunary(sv, string_amg);
2760 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2761 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2762 /* Unwrap this: */
2763 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2764 */
75dfc8ec 2765
4bac9ae4
CS
2766 char *pv;
2767 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2768 if (flags & SV_CONST_RETURN) {
2769 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2770 } else {
4bac9ae4
CS
2771 pv = (flags & SV_MUTABLE_RETURN)
2772 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2773 }
4bac9ae4
CS
2774 if (lp)
2775 *lp = SvCUR(tmpstr);
2776 } else {
2777 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2778 }
4bac9ae4
CS
2779 if (SvUTF8(tmpstr))
2780 SvUTF8_on(sv);
2781 else
2782 SvUTF8_off(sv);
2783 return pv;
deb46114 2784 }
4bac9ae4
CS
2785 }
2786 {
2787 STRLEN len;
2788 char *retval;
2789 char *buffer;
2790 SV *const referent = SvRV(sv);
2791
2792 if (!referent) {
2793 len = 7;
2794 retval = buffer = savepvn("NULLREF", len);
2795 } else if (SvTYPE(referent) == SVt_REGEXP &&
2796 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2797 amagic_is_enabled(string_amg))) {
2798 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2799
2800 assert(re);
67d2d14d 2801
4bac9ae4
CS
2802 /* If the regex is UTF-8 we want the containing scalar to
2803 have an UTF-8 flag too */
2804 if (RX_UTF8(re))
2805 SvUTF8_on(sv);
2806 else
2807 SvUTF8_off(sv);
67d2d14d 2808
4bac9ae4
CS
2809 if (lp)
2810 *lp = RX_WRAPLEN(re);
67d2d14d 2811
4bac9ae4
CS
2812 return RX_WRAPPED(re);
2813 } else {
2814 const char *const typestr = sv_reftype(referent, 0);
2815 const STRLEN typelen = strlen(typestr);
2816 UV addr = PTR2UV(referent);
2817 const char *stashname = NULL;
2818 STRLEN stashnamelen = 0; /* hush, gcc */
2819 const char *buffer_end;
2820
2821 if (SvOBJECT(referent)) {
2822 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2823
2824 if (name) {
2825 stashname = HEK_KEY(name);
2826 stashnamelen = HEK_LEN(name);
2827
2828 if (HEK_UTF8(name)) {
2829 SvUTF8_on(sv);
fafee734 2830 } else {
4bac9ae4 2831 SvUTF8_off(sv);
fafee734 2832 }
fafee734 2833 } else {
4bac9ae4
CS
2834 stashname = "__ANON__";
2835 stashnamelen = 8;
fafee734 2836 }
4bac9ae4
CS
2837 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2838 + 2 * sizeof(UV) + 2 /* )\0 */;
2839 } else {
2840 len = typelen + 3 /* (0x */
2841 + 2 * sizeof(UV) + 2 /* )\0 */;
2842 }
fafee734 2843
4bac9ae4
CS
2844 Newx(buffer, len, char);
2845 buffer_end = retval = buffer + len;
2846
2847 /* Working backwards */
2848 *--retval = '\0';
2849 *--retval = ')';
2850 do {
2851 *--retval = PL_hexdigit[addr & 15];
2852 } while (addr >>= 4);
2853 *--retval = 'x';
2854 *--retval = '0';
2855 *--retval = '(';
2856
2857 retval -= typelen;
2858 memcpy(retval, typestr, typelen);
2859
2860 if (stashname) {
2861 *--retval = '=';
2862 retval -= stashnamelen;
2863 memcpy(retval, stashname, stashnamelen);
c080367d 2864 }
4bac9ae4
CS
2865 /* retval may not necessarily have reached the start of the
2866 buffer here. */
2867 assert (retval >= buffer);
2868
2869 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 2870 }
cdb061a3 2871 if (lp)
4bac9ae4
CS
2872 *lp = len;
2873 SAVEFREEPV(buffer);
2874 return retval;
79072805 2875 }
79072805 2876 }
4bac9ae4
CS
2877
2878 if (SvPOKp(sv)) {
2879 if (lp)
2880 *lp = SvCUR(sv);
2881 if (flags & SV_MUTABLE_RETURN)
2882 return SvPVX_mutable(sv);
2883 if (flags & SV_CONST_RETURN)
2884 return (char *)SvPVX_const(sv);
2885 return SvPVX(sv);
2886 }
2887
2888 if (SvIOK(sv)) {
28e5dec8
JH
2889 /* I'm assuming that if both IV and NV are equally valid then
2890 converting the IV is going to be more efficient */
e1ec3a88 2891 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2892 char buf[TYPE_CHARS(UV)];
2893 char *ebuf, *ptr;
97a130b8 2894 STRLEN len;
28e5dec8
JH
2895
2896 if (SvTYPE(sv) < SVt_PVIV)
2897 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2898 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2899 len = ebuf - ptr;
5902b6a9 2900 /* inlined from sv_setpvn */
97a130b8
NC
2901 s = SvGROW_mutable(sv, len + 1);
2902 Move(ptr, s, len, char);
2903 s += len;
28e5dec8 2904 *s = '\0';
28e5dec8 2905 }
4bac9ae4 2906 else if (SvNOK(sv)) {
79072805
LW
2907 if (SvTYPE(sv) < SVt_PVNV)
2908 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2909 if (SvNVX(sv) == 0.0) {
2910 s = SvGROW_mutable(sv, 2);
2911 *s++ = '0';
2912 *s = '\0';
2913 } else {
2914 dSAVE_ERRNO;
2915 /* The +20 is pure guesswork. Configure test needed. --jhi */
2916 s = SvGROW_mutable(sv, NV_DIG + 20);
2917 /* some Xenix systems wipe out errno here */
2d4389e4 2918 Gconvert(SvNVX(sv), NV_DIG, 0, s);
29912d93
NC
2919 RESTORE_ERRNO;
2920 while (*s) s++;
bbce6d69 2921 }
79072805
LW
2922#ifdef hcx
2923 if (s[-1] == '.')
46fc3d4c 2924 *--s = '\0';
79072805
LW
2925#endif
2926 }
4bac9ae4
CS
2927 else if (isGV_with_GP(sv)) {
2928 GV *const gv = MUTABLE_GV(sv);
2929 SV *const buffer = sv_newmortal();
8d1c3e26 2930
4bac9ae4 2931 gv_efullname3(buffer, gv, "*");
180488f8 2932
4bac9ae4
CS
2933 assert(SvPOK(buffer));
2934 if (SvUTF8(buffer))
2935 SvUTF8_on(sv);
2936 if (lp)
2937 *lp = SvCUR(buffer);
2938 return SvPVX(buffer);
2939 }
8d919b0a
FC
2940 else if (isREGEXP(sv)) {
2941 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2942 return RX_WRAPPED((REGEXP *)sv);
2943 }
4bac9ae4 2944 else {
cdb061a3 2945 if (lp)
00b6aa41 2946 *lp = 0;
9f621bb0
NC
2947 if (flags & SV_UNDEF_RETURNS_NULL)
2948 return NULL;
9a214eec 2949 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2950 report_uninit(sv);
4bac9ae4
CS
2951 /* Typically the caller expects that sv_any is not NULL now. */
2952 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 2953 sv_upgrade(sv, SVt_PV);
73d840c0 2954 return (char *)"";
79072805 2955 }
4bac9ae4 2956
cdb061a3 2957 {
823a54a3 2958 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2959 if (lp)
2960 *lp = len;
2961 SvCUR_set(sv, len);
2962 }
79072805 2963 SvPOK_on(sv);
1d7c1841 2964 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 2965 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
2966 if (flags & SV_CONST_RETURN)
2967 return (char *)SvPVX_const(sv);
10516c54
NC
2968 if (flags & SV_MUTABLE_RETURN)
2969 return SvPVX_mutable(sv);
463ee0b2
LW
2970 return SvPVX(sv);
2971}
2972
645c22ef 2973/*
6050d10e
JP
2974=for apidoc sv_copypv
2975
2976Copies a stringified representation of the source SV into the
2977destination SV. Automatically performs any necessary mg_get and
54f0641b 2978coercion of numeric values into strings. Guaranteed to preserve
2575c402 2979UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
2980sv_2pv[_flags] but operates directly on an SV instead of just the
2981string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
2982would lose the UTF-8'ness of the PV.
2983
4bac9ae4
CS
2984=for apidoc sv_copypv_nomg
2985
2986Like sv_copypv, but doesn't invoke get magic first.
2987
2988=for apidoc sv_copypv_flags
2989
2990Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
2991include SV_GMAGIC.
2992
6050d10e
JP
2993=cut
2994*/
2995
2996void
5aaab254 2997Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
6050d10e 2998{
4bac9ae4
CS
2999 PERL_ARGS_ASSERT_SV_COPYPV;
3000
3001 sv_copypv_flags(dsv, ssv, 0);
3002}
3003
3004void
5aaab254 3005Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3006{
446eaa42 3007 STRLEN len;
4bac9ae4 3008 const char *s;
7918f24d 3009
4bac9ae4 3010 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3011
4bac9ae4
CS
3012 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3013 mg_get(ssv);
3014 s = SvPV_nomg_const(ssv,len);
cb50f42d 3015 sv_setpvn(dsv,s,len);
446eaa42 3016 if (SvUTF8(ssv))
cb50f42d 3017 SvUTF8_on(dsv);
446eaa42 3018 else
cb50f42d 3019 SvUTF8_off(dsv);
6050d10e
JP
3020}
3021
3022/*
645c22ef
DM
3023=for apidoc sv_2pvbyte
3024
3025Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3026to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3027side-effect.
3028
3029Usually accessed via the C<SvPVbyte> macro.
3030
3031=cut
3032*/
3033
7340a771 3034char *
5aaab254 3035Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3036{
7918f24d
NC
3037 PERL_ARGS_ASSERT_SV_2PVBYTE;
3038
4499db73
FC
3039 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3040 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181
FC
3041 SV *sv2 = sv_newmortal();
3042 sv_copypv(sv2,sv);
3043 sv = sv2;
3044 }
3045 else SvGETMAGIC(sv);
0875d2fe 3046 sv_utf8_downgrade(sv,0);
71eb6d8c 3047 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3048}
3049
645c22ef 3050/*
035cbb0e
RGS
3051=for apidoc sv_2pvutf8
3052
3053Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3054to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3055
3056Usually accessed via the C<SvPVutf8> macro.
3057
3058=cut
3059*/
645c22ef 3060
7340a771 3061char *
5aaab254 3062Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3063{
7918f24d
NC
3064 PERL_ARGS_ASSERT_SV_2PVUTF8;
3065
4499db73
FC
3066 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3067 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3068 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3069 else
3070 SvGETMAGIC(sv);
3071 sv_utf8_upgrade_nomg(sv);
c3ec315f 3072 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3073}
1c846c1f 3074
7ee2227d 3075
645c22ef
DM
3076/*
3077=for apidoc sv_2bool
3078
06c841cf
FC
3079This macro is only used by sv_true() or its macro equivalent, and only if
3080the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3081It calls sv_2bool_flags with the SV_GMAGIC flag.
3082
3083=for apidoc sv_2bool_flags
3084
3085This function is only used by sv_true() and friends, and only if
fde67290 3086the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3087contain SV_GMAGIC, then it does an mg_get() first.
3088
645c22ef
DM
3089
3090=cut
3091*/
3092
463ee0b2 3093bool
5aaab254 3094Perl_sv_2bool_flags(pTHX_ SV *const sv, const I32 flags)
463ee0b2 3095{
97aff369 3096 dVAR;
7918f24d 3097
06c841cf 3098 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3099
06c841cf 3100 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3101
a0d0e21e
LW
3102 if (!SvOK(sv))
3103 return 0;
3104 if (SvROK(sv)) {
fabdb6c0 3105 if (SvAMAGIC(sv)) {
31d632c3 3106 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3107 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3108 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3109 }
3110 return SvRV(sv) != 0;
a0d0e21e 3111 }
4bac9ae4 3112 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3113}
3114
c461cf8f
JH
3115/*
3116=for apidoc sv_utf8_upgrade
3117
78ea37eb 3118Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3119Forces the SV to string form if it is not already.
2bbc8d55 3120Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3121Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3122if the whole string is the same in UTF-8 as not.
3123Returns the number of bytes in the converted string
c461cf8f 3124
0efd0472 3125This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3126use the Encode extension for that.
3127
fe749c9a
KW
3128=for apidoc sv_utf8_upgrade_nomg
3129
fde67290 3130Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3131
8d6d96c1
HS
3132=for apidoc sv_utf8_upgrade_flags
3133
78ea37eb 3134Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3135Forces the SV to string form if it is not already.
8d6d96c1 3136Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3137if all the bytes are invariant in UTF-8.
3138If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55
SP
3139will C<mg_get> on C<sv> if appropriate, else not.
3140Returns the number of bytes in the converted string
3141C<sv_utf8_upgrade> and
8d6d96c1
HS
3142C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3143
0efd0472 3144This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3145use the Encode extension for that.
3146
8d6d96c1 3147=cut
b3ab6785
KW
3148
3149The grow version is currently not externally documented. It adds a parameter,
3150extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3151have free after it upon return. This allows the caller to reserve extra space
3152that it intends to fill, to avoid extra grows.
3153
3154Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3155which can be used to tell this function to not first check to see if there are
3156any characters that are different in UTF-8 (variant characters) which would
3157force it to allocate a new string to sv, but to assume there are. Typically
3158this flag is used by a routine that has already parsed the string to find that
3159there are such characters, and passes this information on so that the work
3160doesn't have to be repeated.
3161
3162(One might think that the calling routine could pass in the position of the
3163first such variant, so it wouldn't have to be found again. But that is not the
3164case, because typically when the caller is likely to use this flag, it won't be
3165calling this routine unless it finds something that won't fit into a byte.
3166Otherwise it tries to not upgrade and just use bytes. But some things that
3167do fit into a byte are variants in utf8, and the caller may not have been
3168keeping track of these.)
3169
3170If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3171isn't guaranteed due to having other routines do the work in some input cases,
3172or if the input is already flagged as being in utf8.
3173
3174The speed of this could perhaps be improved for many cases if someone wanted to
3175write a fast function that counts the number of variant characters in a string,
3176especially if it could return the position of the first one.
3177
8d6d96c1
HS
3178*/
3179
3180STRLEN
5aaab254 3181Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3182{
97aff369 3183 dVAR;
7918f24d 3184
b3ab6785 3185 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3186
808c356f
RGS
3187 if (sv == &PL_sv_undef)
3188 return 0;
892f9127 3189 if (!SvPOK_nog(sv)) {
e0e62c2a 3190 STRLEN len = 0;
d52b7888
NC
3191 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3192 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3193 if (SvUTF8(sv)) {
3194 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3195 return len;
b3ab6785 3196 }
d52b7888 3197 } else {
33fb6f35 3198 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3199 }
e0e62c2a 3200 }
4411f3b6 3201
f5cee72b 3202 if (SvUTF8(sv)) {
b3ab6785 3203 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3204 return SvCUR(sv);
f5cee72b 3205 }
5fec3b1d 3206
765f542d
NC
3207 if (SvIsCOW(sv)) {
3208 sv_force_normal_flags(sv, 0);
db42d148
NIS
3209 }
3210
b3ab6785 3211 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3212 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3213 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3214 return SvCUR(sv);
3215 }
3216
4e93345f
KW
3217 if (SvCUR(sv) == 0) {
3218 if (extra) SvGROW(sv, extra);
3219 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3220 /* This function could be much more efficient if we
2bbc8d55 3221 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3222 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3223 * make the loop as fast as possible (although there are certainly ways
3224 * to speed this up, eg. through vectorization) */
3225 U8 * s = (U8 *) SvPVX_const(sv);
3226 U8 * e = (U8 *) SvEND(sv);
3227 U8 *t = s;
3228 STRLEN two_byte_count = 0;
c4e7c712 3229
b3ab6785
KW
3230 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3231
3232 /* See if really will need to convert to utf8. We mustn't rely on our
3233 * incoming SV being well formed and having a trailing '\0', as certain
3234 * code in pp_formline can send us partially built SVs. */
3235
c4e7c712 3236 while (t < e) {
53c1dcc0 3237 const U8 ch = *t++;
b3ab6785
KW
3238 if (NATIVE_IS_INVARIANT(ch)) continue;
3239
3240 t--; /* t already incremented; re-point to first variant */
3241 two_byte_count = 1;
3242 goto must_be_utf8;
c4e7c712 3243 }
b3ab6785
KW
3244
3245 /* utf8 conversion not needed because all are invariants. Mark as
3246 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3247 SvUTF8_on(sv);
7f0bfbea 3248 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3249 return SvCUR(sv);
3250
3251must_be_utf8:
3252
3253 /* Here, the string should be converted to utf8, either because of an
3254 * input flag (two_byte_count = 0), or because a character that
3255 * requires 2 bytes was found (two_byte_count = 1). t points either to
3256 * the beginning of the string (if we didn't examine anything), or to
3257 * the first variant. In either case, everything from s to t - 1 will
3258 * occupy only 1 byte each on output.
3259 *
3260 * There are two main ways to convert. One is to create a new string
3261 * and go through the input starting from the beginning, appending each
3262 * converted value onto the new string as we go along. It's probably
3263 * best to allocate enough space in the string for the worst possible
3264 * case rather than possibly running out of space and having to
3265 * reallocate and then copy what we've done so far. Since everything
3266 * from s to t - 1 is invariant, the destination can be initialized
3267 * with these using a fast memory copy
3268 *
3269 * The other way is to figure out exactly how big the string should be
3270 * by parsing the entire input. Then you don't have to make it big
3271 * enough to handle the worst possible case, and more importantly, if
3272 * the string you already have is large enough, you don't have to
3273 * allocate a new string, you can copy the last character in the input
3274 * string to the final position(s) that will be occupied by the
3275 * converted string and go backwards, stopping at t, since everything
3276 * before that is invariant.
3277 *
3278 * There are advantages and disadvantages to each method.
3279 *
3280 * In the first method, we can allocate a new string, do the memory
3281 * copy from the s to t - 1, and then proceed through the rest of the
3282 * string byte-by-byte.
3283 *
3284 * In the second method, we proceed through the rest of the input
3285 * string just calculating how big the converted string will be. Then
3286 * there are two cases:
3287 * 1) if the string has enough extra space to handle the converted
3288 * value. We go backwards through the string, converting until we
3289 * get to the position we are at now, and then stop. If this
3290 * position is far enough along in the string, this method is
3291 * faster than the other method. If the memory copy were the same
3292 * speed as the byte-by-byte loop, that position would be about
3293 * half-way, as at the half-way mark, parsing to the end and back
3294 * is one complete string's parse, the same amount as starting
3295 * over and going all the way through. Actually, it would be
3296 * somewhat less than half-way, as it's faster to just count bytes
3297 * than to also copy, and we don't have the overhead of allocating
3298 * a new string, changing the scalar to use it, and freeing the
3299 * existing one. But if the memory copy is fast, the break-even
3300 * point is somewhere after half way. The counting loop could be
3301 * sped up by vectorization, etc, to move the break-even point
3302 * further towards the beginning.
3303 * 2) if the string doesn't have enough space to handle the converted
3304 * value. A new string will have to be allocated, and one might
3305 * as well, given that, start from the beginning doing the first
3306 * method. We've spent extra time parsing the string and in
3307 * exchange all we've gotten is that we know precisely how big to
3308 * make the new one. Perl is more optimized for time than space,
3309 * so this case is a loser.
3310 * So what I've decided to do is not use the 2nd method unless it is
3311 * guaranteed that a new string won't have to be allocated, assuming
3312 * the worst case. I also decided not to put any more conditions on it
3313 * than this, for now. It seems likely that, since the worst case is
3314 * twice as big as the unknown portion of the string (plus 1), we won't
3315 * be guaranteed enough space, causing us to go to the first method,
3316 * unless the string is short, or the first variant character is near
3317 * the end of it. In either of these cases, it seems best to use the
3318 * 2nd method. The only circumstance I can think of where this would
3319 * be really slower is if the string had once had much more data in it
3320 * than it does now, but there is still a substantial amount in it */
3321
3322 {
3323 STRLEN invariant_head = t - s;
3324 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3325 if (SvLEN(sv) < size) {
3326
3327 /* Here, have decided to allocate a new string */
3328
3329 U8 *dst;
3330 U8 *d;
3331
3332 Newx(dst, size, U8);
3333
3334 /* If no known invariants at the beginning of the input string,
3335 * set so starts from there. Otherwise, can use memory copy to
3336 * get up to where we are now, and then start from here */
3337
3338 if (invariant_head <= 0) {
3339 d = dst;
3340 } else {
3341 Copy(s, dst, invariant_head, char);
3342 d = dst + invariant_head;
3343 }
3344
3345 while (t < e) {
3346 const UV uv = NATIVE8_TO_UNI(*t++);
3347 if (UNI_IS_INVARIANT(uv))
3348 *d++ = (U8)UNI_TO_NATIVE(uv);
3349 else {
3350 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3351 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3352 }
3353 }
3354 *d = '\0';
3355 SvPV_free(sv); /* No longer using pre-existing string */
3356 SvPV_set(sv, (char*)dst);
3357 SvCUR_set(sv, d - dst);
3358 SvLEN_set(sv, size);
3359 } else {
3360
3361 /* Here, have decided to get the exact size of the string.
3362 * Currently this happens only when we know that there is
3363 * guaranteed enough space to fit the converted string, so
3364 * don't have to worry about growing. If two_byte_count is 0,
3365 * then t points to the first byte of the string which hasn't
3366 * been examined yet. Otherwise two_byte_count is 1, and t
3367 * points to the first byte in the string that will expand to
3368 * two. Depending on this, start examining at t or 1 after t.
3369 * */
3370
3371 U8 *d = t + two_byte_count;
3372
3373
3374 /* Count up the remaining bytes that expand to two */
3375
3376 while (d < e) {
3377 const U8 chr = *d++;
3378 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3379 }
3380
3381 /* The string will expand by just the number of bytes that
3382 * occupy two positions. But we are one afterwards because of
3383 * the increment just above. This is the place to put the
3384 * trailing NUL, and to set the length before we decrement */
3385
3386 d += two_byte_count;
3387 SvCUR_set(sv, d - s);
3388 *d-- = '\0';
3389
3390
3391 /* Having decremented d, it points to the position to put the
3392 * very last byte of the expanded string. Go backwards through
3393 * the string, copying and expanding as we go, stopping when we
3394 * get to the part that is invariant the rest of the way down */
3395
3396 e--;
3397 while (e >= t) {
3398 const U8 ch = NATIVE8_TO_UNI(*e--);
3399 if (UNI_IS_INVARIANT(ch)) {
3400 *d-- = UNI_TO_NATIVE(ch);
3401 } else {
3402 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3403 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3404 }
3405 }
3406 }
75da9d4c
DM
3407
3408 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3409 /* Update pos. We do it at the end rather than during
3410 * the upgrade, to avoid slowing down the common case
3411 * (upgrade without pos) */
3412 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3413 if (mg) {
3414 I32 pos = mg->mg_len;
3415 if (pos > 0 && (U32)pos > invariant_head) {
3416 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3417 STRLEN n = (U32)pos - invariant_head;
3418 while (n > 0) {
3419 if (UTF8_IS_START(*d))
3420 d++;
3421 d++;
3422 n--;
3423 }
3424 mg->mg_len = d - (U8*)SvPVX(sv);
3425 }
3426 }
3427 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3428 magic_setutf8(sv,mg); /* clear UTF8 cache */
3429 }
b3ab6785 3430 }
560a288e 3431 }
b3ab6785
KW
3432
3433 /* Mark as UTF-8 even if no variant - saves scanning loop */
3434 SvUTF8_on(sv);
4411f3b6 3435 return SvCUR(sv);
560a288e
GS
3436}
3437
c461cf8f
JH
3438/*
3439=for apidoc sv_utf8_downgrade
3440
78ea37eb 3441Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3442If the PV contains a character that cannot fit
3443in a byte, this conversion will fail;
78ea37eb 3444in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3445true, croaks.
3446
0efd0472 3447This is not a general purpose Unicode to byte encoding interface:
13a6c0e0
JH
3448use the Encode extension for that.
3449
c461cf8f
JH
3450=cut
3451*/
3452
560a288e 3453bool
5aaab254 3454Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
560a288e 3455{
97aff369 3456 dVAR;
7918f24d
NC
3457
3458 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3459
78ea37eb 3460 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3461 if (SvCUR(sv)) {
03cfe0ae 3462 U8 *s;
652088fc 3463 STRLEN len;
75da9d4c 3464 int mg_flags = SV_GMAGIC;
fa301091 3465
765f542d
NC
3466 if (SvIsCOW(sv)) {
3467 sv_force_normal_flags(sv, 0);
3468 }
75da9d4c
DM
3469 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3470 /* update pos */
3471 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3472 if (mg) {
3473 I32 pos = mg->mg_len;
3474 if (pos > 0) {
3475 sv_pos_b2u(sv, &pos);
3476 mg_flags = 0; /* sv_pos_b2u does get magic */
3477 mg->mg_len = pos;
3478 }
3479 }
3480 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3481 magic_setutf8(sv,mg); /* clear UTF8 cache */
3482
3483 }
3484 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3485
03cfe0ae 3486 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3487 if (fail_ok)
3488 return FALSE;
3489 else {
3490 if (PL_op)
3491 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3492 OP_DESC(PL_op));
fa301091
JH
3493 else
3494 Perl_croak(aTHX_ "Wide character");
3495 }
4b3603a4 3496 }
b162af07 3497 SvCUR_set(sv, len);
67e989fb 3498 }
560a288e 3499 }
ffebcc3e 3500 SvUTF8_off(sv);
560a288e
GS
3501 return TRUE;
3502}
3503
c461cf8f
JH
3504/*
3505=for apidoc sv_utf8_encode
3506
78ea37eb
ST
3507Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3508flag off so that it looks like octets again.
c461cf8f
JH
3509
3510=cut
3511*/
3512
560a288e 3513void
5aaab254 3514Perl_sv_utf8_encode(pTHX_ SV *const sv)
560a288e 3515{
7918f24d
NC
3516 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3517
4c94c214 3518 if (SvREADONLY(sv)) {
654c723f 3519 sv_force_normal_flags(sv, 0);
4c94c214 3520 }
a5f5288a 3521 (void) sv_utf8_upgrade(sv);
560a288e
GS
3522 SvUTF8_off(sv);
3523}
3524
4411f3b6
NIS
3525/*
3526=for apidoc sv_utf8_decode
3527
78ea37eb
ST
3528If the PV of the SV is an octet sequence in UTF-8
3529and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3530so that it looks like a character. If the PV contains only single-byte
694cf0d2 3531characters, the C<SvUTF8> flag stays off.
78ea37eb 3532Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3533
3534=cut
3535*/
3536
560a288e 3537bool
5aaab254 3538Perl_sv_utf8_decode(pTHX_ SV *const sv)
560a288e 3539{
7918f24d
NC
3540 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3541
78ea37eb 3542 if (SvPOKp(sv)) {
75da9d4c 3543 const U8 *start, *c;
93524f2b 3544 const U8 *e;
9cbac4c7 3545
645c22ef
DM
3546 /* The octets may have got themselves encoded - get them back as
3547 * bytes
3548 */
3549 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3550 return FALSE;
3551
3552 /* it is actually just a matter of turning the utf8 flag on, but
3553 * we want to make sure everything inside is valid utf8 first.
3554 */
75da9d4c 3555 c = start = (const U8 *) SvPVX_const(sv);
f9cf141a 3556 if (!is_utf8_string(c, SvCUR(sv)))
67e989fb 3557 return FALSE;
93524f2b 3558 e = (const U8 *) SvEND(sv);
511c2ff0 3559 while (c < e) {
b64e5050 3560 const U8 ch = *c++;
c4d5f83a 3561 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3562 SvUTF8_on(sv);
3563 break;
3564 }
560a288e 3565 }
75da9d4c
DM
3566 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3567 /* adjust pos to the start of a UTF8 char sequence */
3568 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3569 if (mg) {
3570 I32 pos = mg->mg_len;
3571 if (pos > 0) {
3572 for (c = start + pos; c > start; c--) {
3573 if (UTF8_IS_START(*c))
3574 break;
3575 }
3576 mg->mg_len = c - start;
3577 }
3578 }
3579 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3580 magic_setutf8(sv,mg); /* clear UTF8 cache */
3581 }
560a288e
GS
3582 }
3583 return TRUE;
3584}
3585
954c1994
GS
3586/*
3587=for apidoc sv_setsv
3588
645c22ef
DM
3589Copies the contents of the source SV C<ssv> into the destination SV
3590C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3591function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3592Loosely speaking, it performs a copy-by-value, obliterating any previous
3593content of the destination.
3594
3595You probably want to use one of the assortment of wrappers, such as
3596C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3597C<SvSetMagicSV_nosteal>.
3598
8d6d96c1
HS
3599=for apidoc sv_setsv_flags
3600
645c22ef
DM
3601Copies the contents of the source SV C<ssv> into the destination SV
3602C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3603function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3604Loosely speaking, it performs a copy-by-value, obliterating any previous
3605content of the destination.
3606If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
fde67290
FC
3607C<ssv> if appropriate, else not. If the C<flags>
3608parameter has the C<NOSTEAL> bit set then the
3609buffers of temps will not be stolen. <sv_setsv>
5fcdf167 3610and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3611
3612You probably want to use one of the assortment of wrappers, such as
3613C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3614C<SvSetMagicSV_nosteal>.
3615
3616This is the primary function for copying scalars, and most other
3617copy-ish functions and macros use this underneath.
8d6d96c1
HS
3618
3619=cut
3620*/
3621
5d0301b7 3622static void
7bc54cea 3623S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3624{
c8bbf675 3625 I32 mro_changes = 0; /*