This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add PL_foldable - a list of chars that participate in folds
[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))
b5ccf5f2 69/* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
765f542d 70 on-write. */
765f542d 71#endif
645c22ef
DM
72
73/* ============================================================================
74
75=head1 Allocation and deallocation of SVs.
76
d2a0f284
JC
77An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
78sv, av, hv...) contains type and reference count information, and for
79many types, a pointer to the body (struct xrv, xpv, xpviv...), which
80contains fields specific to each type. Some types store all they need
81in the head, so don't have a body.
82
486ec47a 83In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
84and bodies are allocated out of arenas, which by default are
85approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
86Sv-bodies are allocated by their sv-type, guaranteeing size
87consistency needed to allocate safely from arrays.
88
d2a0f284
JC
89For SV-heads, the first slot in each arena is reserved, and holds a
90link to the next arena, some flags, and a note of the number of slots.
91Snaked through each arena chain is a linked list of free items; when
92this becomes empty, an extra arena is allocated and divided up into N
93items which are threaded into the free list.
94
95SV-bodies are similar, but they use arena-sets by default, which
96separate the link and info from the arena itself, and reclaim the 1st
97slot in the arena. SV-bodies are further described later.
645c22ef
DM
98
99The following global variables are associated with arenas:
100
101 PL_sv_arenaroot pointer to list of SV arenas
102 PL_sv_root pointer to list of free SV structures
103
d2a0f284
JC
104 PL_body_arenas head of linked-list of body arenas
105 PL_body_roots[] array of pointers to list of free bodies of svtype
106 arrays are indexed by the svtype needed
93e68bfb 107
d2a0f284
JC
108A few special SV heads are not allocated from an arena, but are
109instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
110The size of arenas can be changed from the default by setting
111PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
112
113The SV arena serves the secondary purpose of allowing still-live SVs
114to be located and destroyed during final cleanup.
115
116At the lowest level, the macros new_SV() and del_SV() grab and free
117an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
118to return the SV to the free list with error checking.) new_SV() calls
119more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
120SVs in the free list have their SvTYPE field set to all ones.
121
ff276b08 122At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 123perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 124start of the interpreter.
645c22ef 125
645c22ef
DM
126The function visit() scans the SV arenas list, and calls a specified
127function for each SV it finds which is still live - ie which has an SvTYPE
128other than all 1's, and a non-zero SvREFCNT. visit() is used by the
129following functions (specified as [function that calls visit()] / [function
130called by visit() for each SV]):
131
132 sv_report_used() / do_report_used()
f2524eef 133 dump all remaining SVs (debugging aid)
645c22ef 134
e4487e9b
DM
135 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
136 do_clean_named_io_objs()
645c22ef 137 Attempt to free all objects pointed to by RVs,
c3b19b5c 138 and try to do the same for all objects indirectly
645c22ef
DM
139 referenced by typeglobs too. Called once from
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
22162ca8 185# define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
d7a2c63c
MHM
186# define DEBUG_SV_SERIAL(sv) \
187 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
188 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
189#else
190# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 191# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
192#endif
193
48614a46
NC
194#ifdef PERL_POISON
195# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 196# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
197/* Whilst I'd love to do this, it seems that things like to check on
198 unreferenced scalars
7e337ee0 199# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 200*/
7e337ee0
JH
201# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
202 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
203#else
204# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 205# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
206# define POSION_SV_HEAD(sv)
207#endif
208
990198f0
DM
209/* Mark an SV head as unused, and add to free list.
210 *
211 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
212 * its refcount artificially decremented during global destruction, so
213 * there may be dangling pointers to it. The last thing we want in that
214 * case is for it to be reused. */
215
053fc874
GS
216#define plant_SV(p) \
217 STMT_START { \
990198f0 218 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
219 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
220 DEBUG_SV_SERIAL(p); \
fd0854ff 221 FREE_SV_DEBUG_FILE(p); \
48614a46 222 POSION_SV_HEAD(p); \
053fc874 223 SvFLAGS(p) = SVTYPEMASK; \
990198f0 224 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 225 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
226 PL_sv_root = (p); \
227 } \
053fc874
GS
228 --PL_sv_count; \
229 } STMT_END
a0d0e21e 230
053fc874
GS
231#define uproot_SV(p) \
232 STMT_START { \
233 (p) = PL_sv_root; \
daba3364 234 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
235 ++PL_sv_count; \
236 } STMT_END
237
645c22ef 238
cac9b346
NC
239/* make some more SVs by adding another arena */
240
cac9b346
NC
241STATIC SV*
242S_more_sv(pTHX)
243{
97aff369 244 dVAR;
cac9b346 245 SV* sv;
9a87bd09
NC
246 char *chunk; /* must use New here to match call to */
247 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
248 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
249 uproot_SV(sv);
250 return sv;
251}
252
645c22ef
DM
253/* new_SV(): return a new, empty SV head */
254
eba0f806
DM
255#ifdef DEBUG_LEAKING_SCALARS
256/* provide a real function for a debugger to play with */
257STATIC SV*
d7a2c63c 258S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
259{
260 SV* sv;
261
eba0f806
DM
262 if (PL_sv_root)
263 uproot_SV(sv);
264 else
cac9b346 265 sv = S_more_sv(aTHX);
eba0f806
DM
266 SvANY(sv) = 0;
267 SvREFCNT(sv) = 1;
268 SvFLAGS(sv) = 0;
fd0854ff 269 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
270 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
271 ? PL_parser->copline
272 : PL_curcop
f24aceb1
DM
273 ? CopLINE(PL_curcop)
274 : 0
e385c3bf 275 );
fd0854ff 276 sv->sv_debug_inpad = 0;
cd676548 277 sv->sv_debug_parent = NULL;
fd0854ff 278 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
279
280 sv->sv_debug_serial = PL_sv_serial++;
281
282 MEM_LOG_NEW_SV(sv, file, line, func);
283 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
284 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
285
eba0f806
DM
286 return sv;
287}
d7a2c63c 288# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
289
290#else
291# define new_SV(p) \
053fc874 292 STMT_START { \
053fc874
GS
293 if (PL_sv_root) \
294 uproot_SV(p); \
295 else \
cac9b346 296 (p) = S_more_sv(aTHX); \
053fc874
GS
297 SvANY(p) = 0; \
298 SvREFCNT(p) = 1; \
299 SvFLAGS(p) = 0; \
d7a2c63c 300 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 301 } STMT_END
eba0f806 302#endif
463ee0b2 303
645c22ef
DM
304
305/* del_SV(): return an empty SV head to the free list */
306
a0d0e21e 307#ifdef DEBUGGING
4561caa4 308
053fc874
GS
309#define del_SV(p) \
310 STMT_START { \
aea4f609 311 if (DEBUG_D_TEST) \
053fc874
GS
312 del_sv(p); \
313 else \
314 plant_SV(p); \
053fc874 315 } STMT_END
a0d0e21e 316
76e3520e 317STATIC void
cea2e8a9 318S_del_sv(pTHX_ SV *p)
463ee0b2 319{
97aff369 320 dVAR;
7918f24d
NC
321
322 PERL_ARGS_ASSERT_DEL_SV;
323
aea4f609 324 if (DEBUG_D_TEST) {
4633a7c4 325 SV* sva;
a3b680e6 326 bool ok = 0;
daba3364 327 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
328 const SV * const sv = sva + 1;
329 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 330 if (p >= sv && p < svend) {
a0d0e21e 331 ok = 1;
c0ff570e
NC
332 break;
333 }
a0d0e21e
LW
334 }
335 if (!ok) {
9b387841
NC
336 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
337 "Attempt to free non-arena SV: 0x%"UVxf
338 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
339 return;
340 }
341 }
4561caa4 342 plant_SV(p);
463ee0b2 343}
a0d0e21e 344
4561caa4
CS
345#else /* ! DEBUGGING */
346
347#define del_SV(p) plant_SV(p)
348
349#endif /* DEBUGGING */
463ee0b2 350
645c22ef
DM
351
352/*
ccfc67b7
JH
353=head1 SV Manipulation Functions
354
645c22ef
DM
355=for apidoc sv_add_arena
356
357Given a chunk of memory, link it to the head of the list of arenas,
358and split it into a list of free SVs.
359
360=cut
361*/
362
d2bd4e7f
NC
363static void
364S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 365{
97aff369 366 dVAR;
daba3364 367 SV *const sva = MUTABLE_SV(ptr);
463ee0b2
LW
368 register SV* sv;
369 register SV* svend;
4633a7c4 370
7918f24d
NC
371 PERL_ARGS_ASSERT_SV_ADD_ARENA;
372
4633a7c4 373 /* The first SV in an arena isn't an SV. */
3280af22 374 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
375 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
376 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
377
3280af22
NIS
378 PL_sv_arenaroot = sva;
379 PL_sv_root = sva + 1;
4633a7c4
LW
380
381 svend = &sva[SvREFCNT(sva) - 1];
382 sv = sva + 1;
463ee0b2 383 while (sv < svend) {
3eef1deb 384 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 385#ifdef DEBUGGING
978b032e 386 SvREFCNT(sv) = 0;
03e36789 387#endif
4b69cbe3 388 /* Must always set typemask because it's always checked in on cleanup
03e36789 389 when the arenas are walked looking for objects. */
8990e307 390 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
391 sv++;
392 }
3eef1deb 393 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
394#ifdef DEBUGGING
395 SvREFCNT(sv) = 0;
396#endif
4633a7c4
LW
397 SvFLAGS(sv) = SVTYPEMASK;
398}
399
055972dc
DM
400/* visit(): call the named function for each non-free SV in the arenas
401 * whose flags field matches the flags/mask args. */
645c22ef 402
5226ed68 403STATIC I32
de37a194 404S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 405{
97aff369 406 dVAR;
4633a7c4 407 SV* sva;
5226ed68 408 I32 visited = 0;
8990e307 409
7918f24d
NC
410 PERL_ARGS_ASSERT_VISIT;
411
daba3364 412 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0 413 register const SV * const svend = &sva[SvREFCNT(sva)];
a3b680e6 414 register SV* sv;
4561caa4 415 for (sv = sva + 1; sv < svend; ++sv) {
055972dc
DM
416 if (SvTYPE(sv) != SVTYPEMASK
417 && (sv->sv_flags & mask) == flags
418 && SvREFCNT(sv))
419 {
acfe0abc 420 (FCALL)(aTHX_ sv);
5226ed68
JH
421 ++visited;
422 }
8990e307
LW
423 }
424 }
5226ed68 425 return visited;
8990e307
LW
426}
427
758a08c3
JH
428#ifdef DEBUGGING
429
645c22ef
DM
430/* called by sv_report_used() for each live SV */
431
432static void
5fa45a31 433do_report_used(pTHX_ SV *const sv)
645c22ef
DM
434{
435 if (SvTYPE(sv) != SVTYPEMASK) {
436 PerlIO_printf(Perl_debug_log, "****\n");
437 sv_dump(sv);
438 }
439}
758a08c3 440#endif
645c22ef
DM
441
442/*
443=for apidoc sv_report_used
444
445Dump the contents of all SVs not yet freed. (Debugging aid).
446
447=cut
448*/
449
8990e307 450void
864dbfa3 451Perl_sv_report_used(pTHX)
4561caa4 452{
ff270d3a 453#ifdef DEBUGGING
055972dc 454 visit(do_report_used, 0, 0);
96a5add6
AL
455#else
456 PERL_UNUSED_CONTEXT;
ff270d3a 457#endif
4561caa4
CS
458}
459
645c22ef
DM
460/* called by sv_clean_objs() for each live SV */
461
462static void
de37a194 463do_clean_objs(pTHX_ SV *const ref)
645c22ef 464{
97aff369 465 dVAR;
ea724faa
NC
466 assert (SvROK(ref));
467 {
823a54a3
AL
468 SV * const target = SvRV(ref);
469 if (SvOBJECT(target)) {
470 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
471 if (SvWEAKREF(ref)) {
472 sv_del_backref(target, ref);
473 SvWEAKREF_off(ref);
474 SvRV_set(ref, NULL);
475 } else {
476 SvROK_off(ref);
477 SvRV_set(ref, NULL);
478 SvREFCNT_dec(target);
479 }
645c22ef
DM
480 }
481 }
482
483 /* XXX Might want to check arrays, etc. */
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;
508 SvREFCNT_dec(obj);
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;
514 SvREFCNT_dec(obj);
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;
520 SvREFCNT_dec(obj);
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);
57ef47cc
DM
526 SvREFCNT_dec(obj);
527 }
e4487e9b
DM
528 SvREFCNT_dec(sv); /* undo the inc above */
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;
549 SvREFCNT_dec(obj);
645c22ef 550 }
57ef47cc 551 SvREFCNT_dec(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
566Attempt to destroy all objects not yet freed
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.... */
7f586e41 585 /* XXX This line breaks Tk and Gtk2. See [perl #82542].
4155e4fe 586 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
7f586e41 587 */
68b590d9
DM
588 olddef = PL_defoutgv;
589 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
590 if (olddef && isGV_with_GP(olddef))
591 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
592 olderr = PL_stderrgv;
593 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
594 if (olderr && isGV_with_GP(olderr))
595 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
596 SvREFCNT_dec(olddef);
3280af22 597 PL_in_clean_objs = FALSE;
4561caa4
CS
598}
599
645c22ef
DM
600/* called by sv_clean_all() for each live SV */
601
602static void
de37a194 603do_clean_all(pTHX_ SV *const sv)
645c22ef 604{
97aff369 605 dVAR;
daba3364 606 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 607 /* don't clean pid table and strtab */
d17ea597 608 return;
cddfcddc 609 }
645c22ef
DM
610 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
611 SvFLAGS(sv) |= SVf_BREAK;
612 SvREFCNT_dec(sv);
613}
614
615/*
616=for apidoc sv_clean_all
617
618Decrement the refcnt of each remaining SV, possibly triggering a
619cleanup. This function may have to be called multiple times to free
ff276b08 620SVs which are in complex self-referential hierarchies.
645c22ef
DM
621
622=cut
623*/
624
5226ed68 625I32
864dbfa3 626Perl_sv_clean_all(pTHX)
8990e307 627{
97aff369 628 dVAR;
5226ed68 629 I32 cleaned;
3280af22 630 PL_in_clean_all = TRUE;
055972dc 631 cleaned = visit(do_clean_all, 0,0);
5226ed68 632 return cleaned;
8990e307 633}
463ee0b2 634
5e258f8c
JC
635/*
636 ARENASETS: a meta-arena implementation which separates arena-info
637 into struct arena_set, which contains an array of struct
638 arena_descs, each holding info for a single arena. By separating
639 the meta-info from the arena, we recover the 1st slot, formerly
640 borrowed for list management. The arena_set is about the size of an
39244528 641 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
642
643 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
644 memory in the last arena-set (1/2 on average). In trade, we get
645 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 646 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
647 small arenas for large, rare body types, by changing array* fields
648 in body_details_by_type[] below.
5e258f8c 649*/
5e258f8c 650struct arena_desc {
398c677b
NC
651 char *arena; /* the raw storage, allocated aligned */
652 size_t size; /* its size ~4k typ */
e5973ed5 653 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
654};
655
e6148039
NC
656struct arena_set;
657
658/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 659 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
660 therefore likely to be 1 aligned memory page. */
661
662#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
663 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
664
665struct arena_set {
666 struct arena_set* next;
0a848332
NC
667 unsigned int set_size; /* ie ARENAS_PER_SET */
668 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
669 struct arena_desc set[ARENAS_PER_SET];
670};
671
645c22ef
DM
672/*
673=for apidoc sv_free_arenas
674
675Deallocate the memory used by all arenas. Note that all the individual SV
676heads and bodies within the arenas must already have been freed.
677
678=cut
679*/
4633a7c4 680void
864dbfa3 681Perl_sv_free_arenas(pTHX)
4633a7c4 682{
97aff369 683 dVAR;
4633a7c4
LW
684 SV* sva;
685 SV* svanext;
0a848332 686 unsigned int i;
4633a7c4
LW
687
688 /* Free arenas here, but be careful about fake ones. (We assume
689 contiguity of the fake ones with the corresponding real ones.) */
690
3280af22 691 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 692 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 693 while (svanext && SvFAKE(svanext))
daba3364 694 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
695
696 if (!SvFAKE(sva))
1df70142 697 Safefree(sva);
4633a7c4 698 }
93e68bfb 699
5e258f8c 700 {
0a848332
NC
701 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
702
703 while (aroot) {
704 struct arena_set *current = aroot;
705 i = aroot->curr;
706 while (i--) {
5e258f8c
JC
707 assert(aroot->set[i].arena);
708 Safefree(aroot->set[i].arena);
709 }
0a848332
NC
710 aroot = aroot->next;
711 Safefree(current);
5e258f8c
JC
712 }
713 }
dc8220bf 714 PL_body_arenas = 0;
fdda85ca 715
0a848332
NC
716 i = PERL_ARENA_ROOTS_SIZE;
717 while (i--)
93e68bfb 718 PL_body_roots[i] = 0;
93e68bfb 719
3280af22
NIS
720 PL_sv_arenaroot = 0;
721 PL_sv_root = 0;
4633a7c4
LW
722}
723
bd81e77b
NC
724/*
725 Here are mid-level routines that manage the allocation of bodies out
726 of the various arenas. There are 5 kinds of arenas:
29489e7c 727
bd81e77b
NC
728 1. SV-head arenas, which are discussed and handled above
729 2. regular body arenas
730 3. arenas for reduced-size bodies
731 4. Hash-Entry arenas
29489e7c 732
bd81e77b
NC
733 Arena types 2 & 3 are chained by body-type off an array of
734 arena-root pointers, which is indexed by svtype. Some of the
735 larger/less used body types are malloced singly, since a large
736 unused block of them is wasteful. Also, several svtypes dont have
737 bodies; the data fits into the sv-head itself. The arena-root
738 pointer thus has a few unused root-pointers (which may be hijacked
739 later for arena types 4,5)
29489e7c 740
bd81e77b
NC
741 3 differs from 2 as an optimization; some body types have several
742 unused fields in the front of the structure (which are kept in-place
743 for consistency). These bodies can be allocated in smaller chunks,
744 because the leading fields arent accessed. Pointers to such bodies
745 are decremented to point at the unused 'ghost' memory, knowing that
746 the pointers are used with offsets to the real memory.
29489e7c 747
d2a0f284
JC
748
749=head1 SV-Body Allocation
750
751Allocation of SV-bodies is similar to SV-heads, differing as follows;
752the allocation mechanism is used for many body types, so is somewhat
753more complicated, it uses arena-sets, and has no need for still-live
754SV detection.
755
756At the outermost level, (new|del)_X*V macros return bodies of the
757appropriate type. These macros call either (new|del)_body_type or
758(new|del)_body_allocated macro pairs, depending on specifics of the
759type. Most body types use the former pair, the latter pair is used to
760allocate body types with "ghost fields".
761
762"ghost fields" are fields that are unused in certain types, and
69ba284b 763consequently don't need to actually exist. They are declared because
d2a0f284
JC
764they're part of a "base type", which allows use of functions as
765methods. The simplest examples are AVs and HVs, 2 aggregate types
766which don't use the fields which support SCALAR semantics.
767
69ba284b 768For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
769chunks, we thus avoid wasted memory for those unaccessed members.
770When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 771size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
772structure. (But things will all go boom if you write to the part that
773is "not there", because you'll be overwriting the last members of the
774preceding structure in memory.)
775
69ba284b
NC
776We calculate the correction using the STRUCT_OFFSET macro on the first
777member present. If the allocated structure is smaller (no initial NV
778actually allocated) then the net effect is to subtract the size of the NV
779from the pointer, to return a new pointer as if an initial NV were actually
780allocated. (We were using structures named *_allocated for this, but
781this turned out to be a subtle bug, because a structure without an NV
782could have a lower alignment constraint, but the compiler is allowed to
783optimised accesses based on the alignment constraint of the actual pointer
784to the full structure, for example, using a single 64 bit load instruction
785because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284
JC
786
787This is the same trick as was used for NV and IV bodies. Ironically it
788doesn't need to be used for NV bodies any more, because NV is now at
789the start of the structure. IV bodies don't need it either, because
790they are no longer allocated.
791
792In turn, the new_body_* allocators call S_new_body(), which invokes
793new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 794linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
795necessary to refresh an empty list. Then the lock is released, and
796the body is returned.
797
99816f8d 798Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
799bodies, which it strings into a linked list. It looks up arena-size
800and body-size from the body_details table described below, thus
801supporting the multiple body-types.
802
803If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
804the (new|del)_X*V macros are mapped directly to malloc/free.
805
d2a0f284
JC
806For each sv-type, struct body_details bodies_by_type[] carries
807parameters which control these aspects of SV handling:
808
809Arena_size determines whether arenas are used for this body type, and if
810so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
811zero, forcing individual mallocs and frees.
812
813Body_size determines how big a body is, and therefore how many fit into
814each arena. Offset carries the body-pointer adjustment needed for
69ba284b 815"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
816
817But its main purpose is to parameterize info needed in
818Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 819vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
820are used for this, except for arena_size.
821
822For the sv-types that have no bodies, arenas are not used, so those
823PL_body_roots[sv_type] are unused, and can be overloaded. In
824something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 825PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 826bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 827available in hv.c.
d2a0f284 828
29489e7c
DM
829*/
830
bd81e77b 831struct body_details {
0fb58b32 832 U8 body_size; /* Size to allocate */
10666ae3 833 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 834 U8 offset;
10666ae3
NC
835 unsigned int type : 4; /* We have space for a sanity check. */
836 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
837 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
838 unsigned int arena : 1; /* Allocated from an arena */
839 size_t arena_size; /* Size of arena to allocate */
bd81e77b 840};
29489e7c 841
bd81e77b
NC
842#define HADNV FALSE
843#define NONV TRUE
29489e7c 844
d2a0f284 845
bd81e77b
NC
846#ifdef PURIFY
847/* With -DPURFIY we allocate everything directly, and don't use arenas.
848 This seems a rather elegant way to simplify some of the code below. */
849#define HASARENA FALSE
850#else
851#define HASARENA TRUE
852#endif
853#define NOARENA FALSE
29489e7c 854
d2a0f284
JC
855/* Size the arenas to exactly fit a given number of bodies. A count
856 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
857 simplifying the default. If count > 0, the arena is sized to fit
858 only that many bodies, allowing arenas to be used for large, rare
859 bodies (XPVFM, XPVIO) without undue waste. The arena size is
860 limited by PERL_ARENA_SIZE, so we can safely oversize the
861 declarations.
862 */
95db5f15
MB
863#define FIT_ARENA0(body_size) \
864 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
865#define FIT_ARENAn(count,body_size) \
866 ( count * body_size <= PERL_ARENA_SIZE) \
867 ? count * body_size \
868 : FIT_ARENA0 (body_size)
869#define FIT_ARENA(count,body_size) \
870 count \
871 ? FIT_ARENAn (count, body_size) \
872 : FIT_ARENA0 (body_size)
d2a0f284 873
bd81e77b
NC
874/* Calculate the length to copy. Specifically work out the length less any
875 final padding the compiler needed to add. See the comment in sv_upgrade
876 for why copying the padding proved to be a bug. */
29489e7c 877
bd81e77b
NC
878#define copy_length(type, last_member) \
879 STRUCT_OFFSET(type, last_member) \
daba3364 880 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 881
bd81e77b 882static const struct body_details bodies_by_type[] = {
829cd18a
NC
883 /* HEs use this offset for their arena. */
884 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 885
1cb9cd50 886 /* The bind placeholder pretends to be an RV for now.
c6f8b1d0 887 Also it's marked as "can't upgrade" to stop anyone using it before it's
1cb9cd50
NC
888 implemented. */
889 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
890
db93c0c4
NC
891 /* IVs are in the head, so the allocation size is 0. */
892 { 0,
d2a0f284 893 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 894 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 895 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
896 },
897
bd81e77b 898 /* 8 bytes on most ILP32 with IEEE doubles */
6e128786
NC
899 { sizeof(NV), sizeof(NV),
900 STRUCT_OFFSET(XPVNV, xnv_u),
901 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 902
bd81e77b 903 /* 8 bytes on most ILP32 with IEEE doubles */
bc337e5c 904 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
905 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
906 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 907 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 908 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 909
bd81e77b 910 /* 12 */
bc337e5c 911 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
912 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
913 + STRUCT_OFFSET(XPV, xpv_cur),
914 SVt_PVIV, FALSE, NONV, HASARENA,
915 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 916
889d28b2 917 /* 20 */
bc337e5c 918 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
919 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
920 + STRUCT_OFFSET(XPV, xpv_cur),
921 SVt_PVNV, FALSE, HADNV, HASARENA,
922 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 923
bd81e77b 924 /* 28 */
6e128786 925 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 926 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 927
288b8c02 928 /* something big */
601dfd0a
NC
929 { sizeof(regexp),
930 sizeof(regexp),
931 0,
08e44740 932 SVt_REGEXP, FALSE, NONV, HASARENA,
eaeb1e7f 933 FIT_ARENA(0, sizeof(regexp))
5c35adbb 934 },
4df7f6af 935
bd81e77b 936 /* 48 */
10666ae3 937 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
938 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
939
bd81e77b 940 /* 64 */
10666ae3 941 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
942 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
943
601dfd0a 944 { sizeof(XPVAV),
4f7003f5 945 copy_length(XPVAV, xav_alloc),
601dfd0a 946 0,
69ba284b 947 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 948 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 949
601dfd0a 950 { sizeof(XPVHV),
359164a0 951 copy_length(XPVHV, xhv_max),
601dfd0a 952 0,
69ba284b 953 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 954 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 955
c84c4652 956 /* 56 */
601dfd0a
NC
957 { sizeof(XPVCV),
958 sizeof(XPVCV),
959 0,
69ba284b 960 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 961 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 962
601dfd0a
NC
963 { sizeof(XPVFM),
964 sizeof(XPVFM),
965 0,
69ba284b 966 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 967 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284
JC
968
969 /* XPVIO is 84 bytes, fits 48x */
601dfd0a
NC
970 { sizeof(XPVIO),
971 sizeof(XPVIO),
972 0,
b6f60916 973 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 974 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 975};
29489e7c 976
bd81e77b 977#define new_body_allocated(sv_type) \
d2a0f284 978 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 979 - bodies_by_type[sv_type].offset)
29489e7c 980
26359cfa
NC
981/* return a thing to the free list */
982
983#define del_body(thing, root) \
984 STMT_START { \
985 void ** const thing_copy = (void **)thing; \
986 *thing_copy = *root; \
987 *root = (void*)thing_copy; \
988 } STMT_END
29489e7c 989
bd81e77b 990#ifdef PURIFY
29489e7c 991
beeec492
NC
992#define new_XNV() safemalloc(sizeof(XPVNV))
993#define new_XPVNV() safemalloc(sizeof(XPVNV))
994#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 995
beeec492 996#define del_XPVGV(p) safefree(p)
29489e7c 997
bd81e77b 998#else /* !PURIFY */
29489e7c 999
65ac1738 1000#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 1001#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1002#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1003
26359cfa
NC
1004#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1005 &PL_body_roots[SVt_PVGV])
1d7c1841 1006
bd81e77b 1007#endif /* PURIFY */
93e68bfb 1008
bd81e77b 1009/* no arena for you! */
93e68bfb 1010
bd81e77b 1011#define new_NOARENA(details) \
beeec492 1012 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1013#define new_NOARENAZ(details) \
beeec492 1014 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1015
1e30fcd5
NC
1016void *
1017Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1018 const size_t arena_size)
d2a0f284
JC
1019{
1020 dVAR;
1021 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1022 struct arena_desc *adesc;
1023 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1024 unsigned int curr;
d2a0f284
JC
1025 char *start;
1026 const char *end;
02982131 1027 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1028#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1029 static bool done_sanity_check;
1030
0b2d3faa
JH
1031 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1032 * variables like done_sanity_check. */
10666ae3 1033 if (!done_sanity_check) {
ea471437 1034 unsigned int i = SVt_LAST;
10666ae3
NC
1035
1036 done_sanity_check = TRUE;
1037
1038 while (i--)
1039 assert (bodies_by_type[i].type == i);
1040 }
1041#endif
1042
02982131 1043 assert(arena_size);
23e9d66c 1044
99816f8d
NC
1045 /* may need new arena-set to hold new arena */
1046 if (!aroot || aroot->curr >= aroot->set_size) {
1047 struct arena_set *newroot;
1048 Newxz(newroot, 1, struct arena_set);
1049 newroot->set_size = ARENAS_PER_SET;
1050 newroot->next = aroot;
1051 aroot = newroot;
1052 PL_body_arenas = (void *) newroot;
1053 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1054 }
1055
1056 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1057 curr = aroot->curr++;
1058 adesc = &(aroot->set[curr]);
1059 assert(!adesc->arena);
1060
1061 Newx(adesc->arena, good_arena_size, char);
1062 adesc->size = good_arena_size;
1063 adesc->utype = sv_type;
1064 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1065 curr, (void*)adesc->arena, (UV)good_arena_size));
1066
1067 start = (char *) adesc->arena;
d2a0f284 1068
29657bb6
NC
1069 /* Get the address of the byte after the end of the last body we can fit.
1070 Remember, this is integer division: */
02982131 1071 end = start + good_arena_size / body_size * body_size;
d2a0f284 1072
486ec47a 1073 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1074#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1075 DEBUG_m(PerlIO_printf(Perl_debug_log,
1076 "arena %p end %p arena-size %d (from %d) type %d "
1077 "size %d ct %d\n",
02982131
NC
1078 (void*)start, (void*)end, (int)good_arena_size,
1079 (int)arena_size, sv_type, (int)body_size,
1080 (int)good_arena_size / (int)body_size));
d8fca402 1081#else
d2a0f284
JC
1082 DEBUG_m(PerlIO_printf(Perl_debug_log,
1083 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1084 (void*)start, (void*)end,
02982131
NC
1085 (int)arena_size, sv_type, (int)body_size,
1086 (int)good_arena_size / (int)body_size));
d8fca402 1087#endif
d2a0f284
JC
1088 *root = (void *)start;
1089
29657bb6
NC
1090 while (1) {
1091 /* Where the next body would start: */
d2a0f284 1092 char * const next = start + body_size;
29657bb6
NC
1093
1094 if (next >= end) {
1095 /* This is the last body: */
1096 assert(next == end);
1097
1098 *(void **)start = 0;
1099 return *root;
1100 }
1101
d2a0f284
JC
1102 *(void**) start = (void *)next;
1103 start = next;
1104 }
d2a0f284
JC
1105}
1106
1107/* grab a new thing from the free list, allocating more if necessary.
1108 The inline version is used for speed in hot routines, and the
1109 function using it serves the rest (unless PURIFY).
1110*/
1111#define new_body_inline(xpv, sv_type) \
1112 STMT_START { \
1113 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1114 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1115 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1116 bodies_by_type[sv_type].body_size,\
1117 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1118 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1119 } STMT_END
1120
1121#ifndef PURIFY
1122
1123STATIC void *
de37a194 1124S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1125{
1126 dVAR;
1127 void *xpv;
1128 new_body_inline(xpv, sv_type);
1129 return xpv;
1130}
1131
1132#endif
93e68bfb 1133
238b27b3
NC
1134static const struct body_details fake_rv =
1135 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1136
bd81e77b
NC
1137/*
1138=for apidoc sv_upgrade
93e68bfb 1139
bd81e77b
NC
1140Upgrade an SV to a more complex form. Generally adds a new body type to the
1141SV, then copies across as much information as possible from the old body.
1142You generally want to use the C<SvUPGRADE> macro wrapper. See also C<svtype>.
93e68bfb 1143
bd81e77b 1144=cut
93e68bfb 1145*/
93e68bfb 1146
bd81e77b 1147void
aad570aa 1148Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
cac9b346 1149{
97aff369 1150 dVAR;
bd81e77b
NC
1151 void* old_body;
1152 void* new_body;
42d0e0b7 1153 const svtype old_type = SvTYPE(sv);
d2a0f284 1154 const struct body_details *new_type_details;
238b27b3 1155 const struct body_details *old_type_details
bd81e77b 1156 = bodies_by_type + old_type;
4df7f6af 1157 SV *referant = NULL;
cac9b346 1158
7918f24d
NC
1159 PERL_ARGS_ASSERT_SV_UPGRADE;
1160
1776cbe8
NC
1161 if (old_type == new_type)
1162 return;
1163
1164 /* This clause was purposefully added ahead of the early return above to
1165 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1166 inference by Nick I-S that it would fix other troublesome cases. See
1167 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1168
1169 Given that shared hash key scalars are no longer PVIV, but PV, there is
1170 no longer need to unshare so as to free up the IVX slot for its proper
1171 purpose. So it's safe to move the early return earlier. */
1172
bd81e77b
NC
1173 if (new_type != SVt_PV && SvIsCOW(sv)) {
1174 sv_force_normal_flags(sv, 0);
1175 }
cac9b346 1176
bd81e77b 1177 old_body = SvANY(sv);
de042e1d 1178
bd81e77b
NC
1179 /* Copying structures onto other structures that have been neatly zeroed
1180 has a subtle gotcha. Consider XPVMG
cac9b346 1181
bd81e77b
NC
1182 +------+------+------+------+------+-------+-------+
1183 | NV | CUR | LEN | IV | MAGIC | STASH |
1184 +------+------+------+------+------+-------+-------+
1185 0 4 8 12 16 20 24 28
645c22ef 1186
bd81e77b
NC
1187 where NVs are aligned to 8 bytes, so that sizeof that structure is
1188 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1189
bd81e77b
NC
1190 +------+------+------+------+------+-------+-------+------+
1191 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1192 +------+------+------+------+------+-------+-------+------+
1193 0 4 8 12 16 20 24 28 32
08742458 1194
bd81e77b 1195 so what happens if you allocate memory for this structure:
30f9da9e 1196
bd81e77b
NC
1197 +------+------+------+------+------+-------+-------+------+------+...
1198 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1199 +------+------+------+------+------+-------+-------+------+------+...
1200 0 4 8 12 16 20 24 28 32 36
bfc44f79 1201
bd81e77b
NC
1202 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1203 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1204 started out as zero once, but it's quite possible that it isn't. So now,
1205 rather than a nicely zeroed GP, you have it pointing somewhere random.
1206 Bugs ensue.
bfc44f79 1207
bd81e77b
NC
1208 (In fact, GP ends up pointing at a previous GP structure, because the
1209 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1210 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1211 this happens to be moot because XPVGV has been re-ordered, with GP
1212 no longer after STASH)
30f9da9e 1213
bd81e77b
NC
1214 So we are careful and work out the size of used parts of all the
1215 structures. */
bfc44f79 1216
bd81e77b
NC
1217 switch (old_type) {
1218 case SVt_NULL:
1219 break;
1220 case SVt_IV:
4df7f6af
NC
1221 if (SvROK(sv)) {
1222 referant = SvRV(sv);
238b27b3
NC
1223 old_type_details = &fake_rv;
1224 if (new_type == SVt_NV)
1225 new_type = SVt_PVNV;
4df7f6af
NC
1226 } else {
1227 if (new_type < SVt_PVIV) {
1228 new_type = (new_type == SVt_NV)
1229 ? SVt_PVNV : SVt_PVIV;
1230 }
bd81e77b
NC
1231 }
1232 break;
1233 case SVt_NV:
1234 if (new_type < SVt_PVNV) {
1235 new_type = SVt_PVNV;
bd81e77b
NC
1236 }
1237 break;
bd81e77b
NC
1238 case SVt_PV:
1239 assert(new_type > SVt_PV);
1240 assert(SVt_IV < SVt_PV);
1241 assert(SVt_NV < SVt_PV);
1242 break;
1243 case SVt_PVIV:
1244 break;
1245 case SVt_PVNV:
1246 break;
1247 case SVt_PVMG:
1248 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1249 there's no way that it can be safely upgraded, because perl.c
1250 expects to Safefree(SvANY(PL_mess_sv)) */
1251 assert(sv != PL_mess_sv);
1252 /* This flag bit is used to mean other things in other scalar types.
1253 Given that it only has meaning inside the pad, it shouldn't be set
1254 on anything that can get upgraded. */
00b1698f 1255 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1256 break;
1257 default:
1258 if (old_type_details->cant_upgrade)
c81225bc
NC
1259 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1260 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1261 }
3376de98
NC
1262
1263 if (old_type > new_type)
1264 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1265 (int)old_type, (int)new_type);
1266
2fa1109b 1267 new_type_details = bodies_by_type + new_type;
645c22ef 1268
bd81e77b
NC
1269 SvFLAGS(sv) &= ~SVTYPEMASK;
1270 SvFLAGS(sv) |= new_type;
932e9ff9 1271
ab4416c0
NC
1272 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1273 the return statements above will have triggered. */
1274 assert (new_type != SVt_NULL);
bd81e77b 1275 switch (new_type) {
bd81e77b
NC
1276 case SVt_IV:
1277 assert(old_type == SVt_NULL);
1278 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1279 SvIV_set(sv, 0);
1280 return;
1281 case SVt_NV:
1282 assert(old_type == SVt_NULL);
1283 SvANY(sv) = new_XNV();
1284 SvNV_set(sv, 0);
1285 return;
bd81e77b 1286 case SVt_PVHV:
bd81e77b 1287 case SVt_PVAV:
d2a0f284 1288 assert(new_type_details->body_size);
c1ae03ae
NC
1289
1290#ifndef PURIFY
1291 assert(new_type_details->arena);
d2a0f284 1292 assert(new_type_details->arena_size);
c1ae03ae 1293 /* This points to the start of the allocated area. */
d2a0f284
JC
1294 new_body_inline(new_body, new_type);
1295 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1296 new_body = ((char *)new_body) - new_type_details->offset;
1297#else
1298 /* We always allocated the full length item with PURIFY. To do this
1299 we fake things so that arena is false for all 16 types.. */
1300 new_body = new_NOARENAZ(new_type_details);
1301#endif
1302 SvANY(sv) = new_body;
1303 if (new_type == SVt_PVAV) {
1304 AvMAX(sv) = -1;
1305 AvFILLp(sv) = -1;
1306 AvREAL_only(sv);
64484faa 1307 if (old_type_details->body_size) {
ac572bf4
NC
1308 AvALLOC(sv) = 0;
1309 } else {
1310 /* It will have been zeroed when the new body was allocated.
1311 Lets not write to it, in case it confuses a write-back
1312 cache. */
1313 }
78ac7dd9
NC
1314 } else {
1315 assert(!SvOK(sv));
1316 SvOK_off(sv);
1317#ifndef NODEFAULT_SHAREKEYS
1318 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1319#endif
1320 HvMAX(sv) = 7; /* (start with 8 buckets) */
c1ae03ae 1321 }
aeb18a1e 1322
bd81e77b
NC
1323 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1324 The target created by newSVrv also is, and it can have magic.
1325 However, it never has SvPVX set.
1326 */
4df7f6af
NC
1327 if (old_type == SVt_IV) {
1328 assert(!SvROK(sv));
1329 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1330 assert(SvPVX_const(sv) == 0);
1331 }
aeb18a1e 1332
bd81e77b 1333 if (old_type >= SVt_PVMG) {
e736a858 1334 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1335 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1336 } else {
1337 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1338 }
1339 break;
93e68bfb 1340
93e68bfb 1341
b9ad13ac
NC
1342 case SVt_REGEXP:
1343 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
1344 sv_force_normal_flags(sv) is called. */
1345 SvFAKE_on(sv);
bd81e77b
NC
1346 case SVt_PVIV:
1347 /* XXX Is this still needed? Was it ever needed? Surely as there is
1348 no route from NV to PVIV, NOK can never be true */
1349 assert(!SvNOKp(sv));
1350 assert(!SvNOK(sv));
1351 case SVt_PVIO:
1352 case SVt_PVFM:
bd81e77b
NC
1353 case SVt_PVGV:
1354 case SVt_PVCV:
1355 case SVt_PVLV:
1356 case SVt_PVMG:
1357 case SVt_PVNV:
1358 case SVt_PV:
93e68bfb 1359
d2a0f284 1360 assert(new_type_details->body_size);
bd81e77b
NC
1361 /* We always allocated the full length item with PURIFY. To do this
1362 we fake things so that arena is false for all 16 types.. */
1363 if(new_type_details->arena) {
1364 /* This points to the start of the allocated area. */
d2a0f284
JC
1365 new_body_inline(new_body, new_type);
1366 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1367 new_body = ((char *)new_body) - new_type_details->offset;
1368 } else {
1369 new_body = new_NOARENAZ(new_type_details);
1370 }
1371 SvANY(sv) = new_body;
5e2fc214 1372
bd81e77b 1373 if (old_type_details->copy) {
f9ba3d20
NC
1374 /* There is now the potential for an upgrade from something without
1375 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1376 int offset = old_type_details->offset;
1377 int length = old_type_details->copy;
1378
1379 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1380 const int difference
f9ba3d20
NC
1381 = new_type_details->offset - old_type_details->offset;
1382 offset += difference;
1383 length -= difference;
1384 }
1385 assert (length >= 0);
1386
1387 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1388 char);
bd81e77b
NC
1389 }
1390
1391#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1392 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1393 * correct 0.0 for us. Otherwise, if the old body didn't have an
1394 * NV slot, but the new one does, then we need to initialise the
1395 * freshly created NV slot with whatever the correct bit pattern is
1396 * for 0.0 */
e22a937e
NC
1397 if (old_type_details->zero_nv && !new_type_details->zero_nv
1398 && !isGV_with_GP(sv))
bd81e77b 1399 SvNV_set(sv, 0);
82048762 1400#endif
5e2fc214 1401
85dca89a
NC
1402 if (new_type == SVt_PVIO) {
1403 IO * const io = MUTABLE_IO(sv);
d963bf01 1404 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1405
1406 SvOBJECT_on(io);
1407 /* Clear the stashcache because a new IO could overrule a package
1408 name */
1409 hv_clear(PL_stashcache);
1410
85dca89a 1411 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1412 IoPAGE_LEN(sv) = 60;
85dca89a 1413 }
4df7f6af
NC
1414 if (old_type < SVt_PV) {
1415 /* referant will be NULL unless the old type was SVt_IV emulating
1416 SVt_RV */
1417 sv->sv_u.svu_rv = referant;
1418 }
bd81e77b
NC
1419 break;
1420 default:
afd78fd5
JH
1421 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1422 (unsigned long)new_type);
bd81e77b 1423 }
73171d91 1424
db93c0c4 1425 if (old_type > SVt_IV) {
bd81e77b 1426#ifdef PURIFY
beeec492 1427 safefree(old_body);
bd81e77b 1428#else
bc786448
GG
1429 /* Note that there is an assumption that all bodies of types that
1430 can be upgraded came from arenas. Only the more complex non-
1431 upgradable types are allowed to be directly malloc()ed. */
1432 assert(old_type_details->arena);
bd81e77b
NC
1433 del_body((void*)((char*)old_body + old_type_details->offset),
1434 &PL_body_roots[old_type]);
1435#endif
1436 }
1437}
73171d91 1438
bd81e77b
NC
1439/*
1440=for apidoc sv_backoff
73171d91 1441
bd81e77b
NC
1442Remove any string offset. You should normally use the C<SvOOK_off> macro
1443wrapper instead.
73171d91 1444
bd81e77b 1445=cut
73171d91
NC
1446*/
1447
bd81e77b 1448int
aad570aa 1449Perl_sv_backoff(pTHX_ register SV *const sv)
bd81e77b 1450{
69240efd 1451 STRLEN delta;
7a4bba22 1452 const char * const s = SvPVX_const(sv);
7918f24d
NC
1453
1454 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1455 PERL_UNUSED_CONTEXT;
7918f24d 1456
bd81e77b
NC
1457 assert(SvOOK(sv));
1458 assert(SvTYPE(sv) != SVt_PVHV);
1459 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1460
69240efd
NC
1461 SvOOK_offset(sv, delta);
1462
7a4bba22
NC
1463 SvLEN_set(sv, SvLEN(sv) + delta);
1464 SvPV_set(sv, SvPVX(sv) - delta);
1465 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1466 SvFLAGS(sv) &= ~SVf_OOK;
1467 return 0;
1468}
73171d91 1469
bd81e77b
NC
1470/*
1471=for apidoc sv_grow
73171d91 1472
bd81e77b
NC
1473Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1474upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1475Use the C<SvGROW> wrapper instead.
93e68bfb 1476
bd81e77b
NC
1477=cut
1478*/
93e68bfb 1479
bd81e77b 1480char *
aad570aa 1481Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
bd81e77b
NC
1482{
1483 register char *s;
93e68bfb 1484
7918f24d
NC
1485 PERL_ARGS_ASSERT_SV_GROW;
1486
5db06880
NC
1487 if (PL_madskills && newlen >= 0x100000) {
1488 PerlIO_printf(Perl_debug_log,
1489 "Allocation too large: %"UVxf"\n", (UV)newlen);
1490 }
bd81e77b
NC
1491#ifdef HAS_64K_LIMIT
1492 if (newlen >= 0x10000) {
1493 PerlIO_printf(Perl_debug_log,
1494 "Allocation too large: %"UVxf"\n", (UV)newlen);
1495 my_exit(1);
1496 }
1497#endif /* HAS_64K_LIMIT */
1498 if (SvROK(sv))
1499 sv_unref(sv);
1500 if (SvTYPE(sv) < SVt_PV) {
1501 sv_upgrade(sv, SVt_PV);
1502 s = SvPVX_mutable(sv);
1503 }
1504 else if (SvOOK(sv)) { /* pv is offset? */
1505 sv_backoff(sv);
1506 s = SvPVX_mutable(sv);
1507 if (newlen > SvLEN(sv))
1508 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1509#ifdef HAS_64K_LIMIT
1510 if (newlen >= 0x10000)
1511 newlen = 0xFFFF;
1512#endif
1513 }
1514 else
1515 s = SvPVX_mutable(sv);
aeb18a1e 1516
bd81e77b 1517 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1518 STRLEN minlen = SvCUR(sv);
1519 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1520 if (newlen < minlen)
1521 newlen = minlen;
aedff202 1522#ifndef Perl_safesysmalloc_size
bd81e77b 1523 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1524#endif
98653f18 1525 if (SvLEN(sv) && s) {
10edeb5d 1526 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1527 }
1528 else {
10edeb5d 1529 s = (char*)safemalloc(newlen);
bd81e77b
NC
1530 if (SvPVX_const(sv) && SvCUR(sv)) {
1531 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1532 }
1533 }
1534 SvPV_set(sv, s);
ca7c1a29 1535#ifdef Perl_safesysmalloc_size
98653f18
NC
1536 /* Do this here, do it once, do it right, and then we will never get
1537 called back into sv_grow() unless there really is some growing
1538 needed. */
ca7c1a29 1539 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1540#else
bd81e77b 1541 SvLEN_set(sv, newlen);
98653f18 1542#endif
bd81e77b
NC
1543 }
1544 return s;
1545}
aeb18a1e 1546
bd81e77b
NC
1547/*
1548=for apidoc sv_setiv
932e9ff9 1549
bd81e77b
NC
1550Copies an integer into the given SV, upgrading first if necessary.
1551Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1552
bd81e77b
NC
1553=cut
1554*/
463ee0b2 1555
bd81e77b 1556void
aad570aa 1557Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
bd81e77b 1558{
97aff369 1559 dVAR;
7918f24d
NC
1560
1561 PERL_ARGS_ASSERT_SV_SETIV;
1562
bd81e77b
NC
1563 SV_CHECK_THINKFIRST_COW_DROP(sv);
1564 switch (SvTYPE(sv)) {
1565 case SVt_NULL:
bd81e77b 1566 case SVt_NV:
3376de98 1567 sv_upgrade(sv, SVt_IV);
bd81e77b 1568 break;
bd81e77b
NC
1569 case SVt_PV:
1570 sv_upgrade(sv, SVt_PVIV);
1571 break;
463ee0b2 1572
bd81e77b 1573 case SVt_PVGV:
6e592b3a
BM
1574 if (!isGV_with_GP(sv))
1575 break;
bd81e77b
NC
1576 case SVt_PVAV:
1577 case SVt_PVHV:
1578 case SVt_PVCV:
1579 case SVt_PVFM:
1580 case SVt_PVIO:
1581 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1582 OP_DESC(PL_op));
42d0e0b7 1583 default: NOOP;
bd81e77b
NC
1584 }
1585 (void)SvIOK_only(sv); /* validate number */
1586 SvIV_set(sv, i);
1587 SvTAINT(sv);
1588}
932e9ff9 1589
bd81e77b
NC
1590/*
1591=for apidoc sv_setiv_mg
d33b2eba 1592
bd81e77b 1593Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1594
bd81e77b
NC
1595=cut
1596*/
d33b2eba 1597
bd81e77b 1598void
aad570aa 1599Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
bd81e77b 1600{
7918f24d
NC
1601 PERL_ARGS_ASSERT_SV_SETIV_MG;
1602
bd81e77b
NC
1603 sv_setiv(sv,i);
1604 SvSETMAGIC(sv);
1605}
727879eb 1606
bd81e77b
NC
1607/*
1608=for apidoc sv_setuv
d33b2eba 1609
bd81e77b
NC
1610Copies an unsigned integer into the given SV, upgrading first if necessary.
1611Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1612
bd81e77b
NC
1613=cut
1614*/
d33b2eba 1615
bd81e77b 1616void
aad570aa 1617Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
bd81e77b 1618{
7918f24d
NC
1619 PERL_ARGS_ASSERT_SV_SETUV;
1620
bd81e77b
NC
1621 /* With these two if statements:
1622 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1623
bd81e77b
NC
1624 without
1625 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1626
bd81e77b
NC
1627 If you wish to remove them, please benchmark to see what the effect is
1628 */
1629 if (u <= (UV)IV_MAX) {
1630 sv_setiv(sv, (IV)u);
1631 return;
1632 }
1633 sv_setiv(sv, 0);
1634 SvIsUV_on(sv);
1635 SvUV_set(sv, u);
1636}
d33b2eba 1637
bd81e77b
NC
1638/*
1639=for apidoc sv_setuv_mg
727879eb 1640
bd81e77b 1641Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1642
bd81e77b
NC
1643=cut
1644*/
5e2fc214 1645
bd81e77b 1646void
aad570aa 1647Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
bd81e77b 1648{
7918f24d
NC
1649 PERL_ARGS_ASSERT_SV_SETUV_MG;
1650
bd81e77b
NC
1651 sv_setuv(sv,u);
1652 SvSETMAGIC(sv);
1653}
5e2fc214 1654
954c1994 1655/*
bd81e77b 1656=for apidoc sv_setnv
954c1994 1657
bd81e77b
NC
1658Copies a double into the given SV, upgrading first if necessary.
1659Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1660
1661=cut
1662*/
1663
63f97190 1664void
aad570aa 1665Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
79072805 1666{
97aff369 1667 dVAR;
7918f24d
NC
1668
1669 PERL_ARGS_ASSERT_SV_SETNV;
1670
bd81e77b
NC
1671 SV_CHECK_THINKFIRST_COW_DROP(sv);
1672 switch (SvTYPE(sv)) {
79072805 1673 case SVt_NULL:
79072805 1674 case SVt_IV:
bd81e77b 1675 sv_upgrade(sv, SVt_NV);
79072805
LW
1676 break;
1677 case SVt_PV:
79072805 1678 case SVt_PVIV:
bd81e77b 1679 sv_upgrade(sv, SVt_PVNV);
79072805 1680 break;
bd4b1eb5 1681
bd4b1eb5 1682 case SVt_PVGV:
6e592b3a
BM
1683 if (!isGV_with_GP(sv))
1684 break;
bd81e77b
NC
1685 case SVt_PVAV:
1686 case SVt_PVHV:
79072805 1687 case SVt_PVCV:
bd81e77b
NC
1688 case SVt_PVFM:
1689 case SVt_PVIO:
1690 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1691 OP_DESC(PL_op));
42d0e0b7 1692 default: NOOP;
2068cd4d 1693 }
bd81e77b
NC
1694 SvNV_set(sv, num);
1695 (void)SvNOK_only(sv); /* validate number */
1696 SvTAINT(sv);
79072805
LW
1697}
1698
645c22ef 1699/*
bd81e77b 1700=for apidoc sv_setnv_mg
645c22ef 1701
bd81e77b 1702Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1703
1704=cut
1705*/
1706
bd81e77b 1707void
aad570aa 1708Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
79072805 1709{
7918f24d
NC
1710 PERL_ARGS_ASSERT_SV_SETNV_MG;
1711
bd81e77b
NC
1712 sv_setnv(sv,num);
1713 SvSETMAGIC(sv);
79072805
LW
1714}
1715
bd81e77b
NC
1716/* Print an "isn't numeric" warning, using a cleaned-up,
1717 * printable version of the offending string
1718 */
954c1994 1719
bd81e77b 1720STATIC void
aad570aa 1721S_not_a_number(pTHX_ SV *const sv)
79072805 1722{
97aff369 1723 dVAR;
bd81e77b
NC
1724 SV *dsv;
1725 char tmpbuf[64];
1726 const char *pv;
94463019 1727
7918f24d
NC
1728 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1729
94463019 1730 if (DO_UTF8(sv)) {
84bafc02 1731 dsv = newSVpvs_flags("", SVs_TEMP);
94463019
JH
1732 pv = sv_uni_display(dsv, sv, 10, 0);
1733 } else {
1734 char *d = tmpbuf;
551405c4 1735 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1736 /* each *s can expand to 4 chars + "...\0",
1737 i.e. need room for 8 chars */
ecdeb87c 1738
00b6aa41
AL
1739 const char *s = SvPVX_const(sv);
1740 const char * const end = s + SvCUR(sv);
1741 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1742 int ch = *s & 0xFF;
1743 if (ch & 128 && !isPRINT_LC(ch)) {
1744 *d++ = 'M';
1745 *d++ = '-';
1746 ch &= 127;
1747 }
1748 if (ch == '\n') {
1749 *d++ = '\\';
1750 *d++ = 'n';
1751 }
1752 else if (ch == '\r') {
1753 *d++ = '\\';
1754 *d++ = 'r';
1755 }
1756 else if (ch == '\f') {
1757 *d++ = '\\';
1758 *d++ = 'f';
1759 }
1760 else if (ch == '\\') {
1761 *d++ = '\\';
1762 *d++ = '\\';
1763 }
1764 else if (ch == '\0') {
1765 *d++ = '\\';
1766 *d++ = '0';
1767 }
1768 else if (isPRINT_LC(ch))
1769 *d++ = ch;
1770 else {
1771 *d++ = '^';
1772 *d++ = toCTRL(ch);
1773 }
1774 }
1775 if (s < end) {
1776 *d++ = '.';
1777 *d++ = '.';
1778 *d++ = '.';
1779 }
1780 *d = '\0';
1781 pv = tmpbuf;
a0d0e21e 1782 }
a0d0e21e 1783
533c011a 1784 if (PL_op)
9014280d 1785 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019
JH
1786 "Argument \"%s\" isn't numeric in %s", pv,
1787 OP_DESC(PL_op));
a0d0e21e 1788 else
9014280d 1789 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
94463019 1790 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1791}
1792
c2988b20
NC
1793/*
1794=for apidoc looks_like_number
1795
645c22ef
DM
1796Test if the content of an SV looks like a number (or is a number).
1797C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1798non-numeric warning), even if your atof() doesn't grok them.
c2988b20
NC
1799
1800=cut
1801*/
1802
1803I32
aad570aa 1804Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1805{
a3b680e6 1806 register const char *sbegin;
c2988b20
NC
1807 STRLEN len;
1808
7918f24d
NC
1809 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1810
c2988b20 1811 if (SvPOK(sv)) {
3f7c398e 1812 sbegin = SvPVX_const(sv);
c2988b20
NC
1813 len = SvCUR(sv);
1814 }
1815 else if (SvPOKp(sv))
83003860 1816 sbegin = SvPV_const(sv, len);
c2988b20 1817 else
e0ab1c0e 1818 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1819 return grok_number(sbegin, len, NULL);
1820}
25da4f38 1821
19f6321d
NC
1822STATIC bool
1823S_glob_2number(pTHX_ GV * const gv)
180488f8
NC
1824{
1825 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1826 SV *const buffer = sv_newmortal();
1827
7918f24d
NC
1828 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1829
180488f8
NC
1830 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1831 is on. */
1832 SvFAKE_off(gv);
1833 gv_efullname3(buffer, gv, "*");
1834 SvFLAGS(gv) |= wasfake;
1835
675c862f
AL
1836 /* We know that all GVs stringify to something that is not-a-number,
1837 so no need to test that. */
1838 if (ckWARN(WARN_NUMERIC))
1839 not_a_number(buffer);
1840 /* We just want something true to return, so that S_sv_2iuv_common
1841 can tail call us and return true. */
19f6321d 1842 return TRUE;
675c862f
AL
1843}
1844
25da4f38
IZ
1845/* Actually, ISO C leaves conversion of UV to IV undefined, but
1846 until proven guilty, assume that things are not that bad... */
1847
645c22ef
DM
1848/*
1849 NV_PRESERVES_UV:
1850
1851 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1852 an IV (an assumption perl has been based on to date) it becomes necessary
1853 to remove the assumption that the NV always carries enough precision to
1854 recreate the IV whenever needed, and that the NV is the canonical form.
1855 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1856 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1857 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1858 1) to distinguish between IV/UV/NV slots that have cached a valid
1859 conversion where precision was lost and IV/UV/NV slots that have a
1860 valid conversion which has lost no precision
645c22ef 1861 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1862 would lose precision, the precise conversion (or differently
1863 imprecise conversion) is also performed and cached, to prevent
1864 requests for different numeric formats on the same SV causing
1865 lossy conversion chains. (lossless conversion chains are perfectly
1866 acceptable (still))
1867
1868
1869 flags are used:
1870 SvIOKp is true if the IV slot contains a valid value
1871 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1872 SvNOKp is true if the NV slot contains a valid value
1873 SvNOK is true only if the NV value is accurate
1874
1875 so
645c22ef 1876 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1877 IV(or UV) would lose accuracy over a direct conversion from PV to
1878 IV(or UV). If it would, cache both conversions, return NV, but mark
1879 SV as IOK NOKp (ie not NOK).
1880
645c22ef 1881 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1882 NV would lose accuracy over a direct conversion from PV to NV. If it
1883 would, cache both conversions, flag similarly.
1884
1885 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1886 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1887 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1888 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1889 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1890
645c22ef
DM
1891 The benefit of this is that operations such as pp_add know that if
1892 SvIOK is true for both left and right operands, then integer addition
1893 can be used instead of floating point (for cases where the result won't
1894 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1895 loss of precision compared with integer addition.
1896
1897 * making IV and NV equal status should make maths accurate on 64 bit
1898 platforms
1899 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1900 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1901 looking for SvIOK and checking for overflow will not outweigh the
1902 fp to integer speedup)
1903 * will slow down integer operations (callers of SvIV) on "inaccurate"
1904 values, as the change from SvIOK to SvIOKp will cause a call into
1905 sv_2iv each time rather than a macro access direct to the IV slot
1906 * should speed up number->string conversion on integers as IV is
645c22ef 1907 favoured when IV and NV are equally accurate
28e5dec8
JH
1908
1909 ####################################################################
645c22ef
DM
1910 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1911 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1912 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1913 ####################################################################
1914
645c22ef 1915 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1916 performance ratio.
1917*/
1918
1919#ifndef NV_PRESERVES_UV
645c22ef
DM
1920# define IS_NUMBER_UNDERFLOW_IV 1
1921# define IS_NUMBER_UNDERFLOW_UV 2
1922# define IS_NUMBER_IV_AND_UV 2
1923# define IS_NUMBER_OVERFLOW_IV 4
1924# define IS_NUMBER_OVERFLOW_UV 5
1925
1926/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1927
1928/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1929STATIC int
5de3775c 1930S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1931# ifdef DEBUGGING
1932 , I32 numtype
1933# endif
1934 )
28e5dec8 1935{
97aff369 1936 dVAR;
7918f24d
NC
1937
1938 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1939
3f7c398e 1940 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
1941 if (SvNVX(sv) < (NV)IV_MIN) {
1942 (void)SvIOKp_on(sv);
1943 (void)SvNOK_on(sv);
45977657 1944 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1945 return IS_NUMBER_UNDERFLOW_IV;
1946 }
1947 if (SvNVX(sv) > (NV)UV_MAX) {
1948 (void)SvIOKp_on(sv);
1949 (void)SvNOK_on(sv);
1950 SvIsUV_on(sv);
607fa7f2 1951 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1952 return IS_NUMBER_OVERFLOW_UV;
1953 }
c2988b20
NC
1954 (void)SvIOKp_on(sv);
1955 (void)SvNOK_on(sv);
1956 /* Can't use strtol etc to convert this string. (See truth table in
1957 sv_2iv */
1958 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1959 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1960 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1961 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1962 } else {
1963 /* Integer is imprecise. NOK, IOKp */
1964 }
1965 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1966 }
1967 SvIsUV_on(sv);
607fa7f2 1968 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1969 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1970 if (SvUVX(sv) == UV_MAX) {
1971 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1972 possibly be preserved by NV. Hence, it must be overflow.
1973 NOK, IOKp */
1974 return IS_NUMBER_OVERFLOW_UV;
1975 }
1976 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1977 } else {
1978 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1979 }
c2988b20 1980 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1981}
645c22ef
DM
1982#endif /* !NV_PRESERVES_UV*/
1983
af359546 1984STATIC bool
7918f24d
NC
1985S_sv_2iuv_common(pTHX_ SV *const sv)
1986{
97aff369 1987 dVAR;
7918f24d
NC
1988
1989 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1990
af359546 1991 if (SvNOKp(sv)) {
28e5dec8
JH
1992 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1993 * without also getting a cached IV/UV from it at the same time
1994 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1995 * IV or UV at same time to avoid this. */
1996 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1997
1998 if (SvTYPE(sv) == SVt_NV)
1999 sv_upgrade(sv, SVt_PVNV);
2000
28e5dec8
JH
2001 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2002 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2003 certainly cast into the IV range at IV_MAX, whereas the correct
2004 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2005 cases go to UV */
cab190d4
JD
2006#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2007 if (Perl_isnan(SvNVX(sv))) {
2008 SvUV_set(sv, 0);
2009 SvIsUV_on(sv);
fdbe6d7c 2010 return FALSE;
cab190d4 2011 }
cab190d4 2012#endif
28e5dec8 2013 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2014 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2015 if (SvNVX(sv) == (NV) SvIVX(sv)
2016#ifndef NV_PRESERVES_UV
2017 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2018 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2019 /* Don't flag it as "accurately an integer" if the number
2020 came from a (by definition imprecise) NV operation, and
2021 we're outside the range of NV integer precision */
2022#endif
2023 ) {
a43d94f2
NC
2024 if (SvNOK(sv))
2025 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2026 else {
2027 /* scalar has trailing garbage, eg "42a" */
2028 }
28e5dec8 2029 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2030 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2031 PTR2UV(sv),
2032 SvNVX(sv),
2033 SvIVX(sv)));
2034
2035 } else {
2036 /* IV not precise. No need to convert from PV, as NV
2037 conversion would already have cached IV if it detected
2038 that PV->IV would be better than PV->NV->IV
2039 flags already correct - don't set public IOK. */
2040 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2041 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2042 PTR2UV(sv),
2043 SvNVX(sv),
2044 SvIVX(sv)));
2045 }
2046 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2047 but the cast (NV)IV_MIN rounds to a the value less (more
2048 negative) than IV_MIN which happens to be equal to SvNVX ??
2049 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2050 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2051 (NV)UVX == NVX are both true, but the values differ. :-(
2052 Hopefully for 2s complement IV_MIN is something like
2053 0x8000000000000000 which will be exact. NWC */
d460ef45 2054 }
25da4f38 2055 else {
607fa7f2 2056 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2057 if (
2058 (SvNVX(sv) == (NV) SvUVX(sv))
2059#ifndef NV_PRESERVES_UV
2060 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2061 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2062 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2063 /* Don't flag it as "accurately an integer" if the number
2064 came from a (by definition imprecise) NV operation, and
2065 we're outside the range of NV integer precision */
2066#endif
a43d94f2 2067 && SvNOK(sv)
28e5dec8
JH
2068 )
2069 SvIOK_on(sv);
25da4f38 2070 SvIsUV_on(sv);
1c846c1f 2071 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2072 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2073 PTR2UV(sv),
57def98f
JH
2074 SvUVX(sv),
2075 SvUVX(sv)));
25da4f38 2076 }
748a9306
LW
2077 }
2078 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2079 UV value;
504618e9 2080 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2081 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2082 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2083 the same as the direct translation of the initial string
2084 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2085 be careful to ensure that the value with the .456 is around if the
2086 NV value is requested in the future).
1c846c1f 2087
af359546 2088 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2089 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2090 cache the NV if we are sure it's not needed.
25da4f38 2091 */
16b7a9a4 2092
c2988b20
NC
2093 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2094 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2095 == IS_NUMBER_IN_UV) {
5e045b90 2096 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2097 if (SvTYPE(sv) < SVt_PVIV)
2098 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2099 (void)SvIOK_on(sv);
c2988b20
NC
2100 } else if (SvTYPE(sv) < SVt_PVNV)
2101 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2102
f2524eef 2103 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2104 we aren't going to call atof() below. If NVs don't preserve UVs
2105 then the value returned may have more precision than atof() will
2106 return, even though value isn't perfectly accurate. */
2107 if ((numtype & (IS_NUMBER_IN_UV
2108#ifdef NV_PRESERVES_UV
2109 | IS_NUMBER_NOT_INT
2110#endif
2111 )) == IS_NUMBER_IN_UV) {
2112 /* This won't turn off the public IOK flag if it was set above */
2113 (void)SvIOKp_on(sv);
2114
2115 if (!(numtype & IS_NUMBER_NEG)) {
2116 /* positive */;
2117 if (value <= (UV)IV_MAX) {
45977657 2118 SvIV_set(sv, (IV)value);
c2988b20 2119 } else {
af359546 2120 /* it didn't overflow, and it was positive. */
607fa7f2 2121 SvUV_set(sv, value);
c2988b20
NC
2122 SvIsUV_on(sv);
2123 }
2124 } else {
2125 /* 2s complement assumption */
2126 if (value <= (UV)IV_MIN) {
45977657 2127 SvIV_set(sv, -(IV)value);
c2988b20
NC
2128 } else {
2129 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2130 I'm assuming it will be rare. */
c2988b20
NC
2131 if (SvTYPE(sv) < SVt_PVNV)
2132 sv_upgrade(sv, SVt_PVNV);
2133 SvNOK_on(sv);
2134 SvIOK_off(sv);
2135 SvIOKp_on(sv);
9d6ce603 2136 SvNV_set(sv, -(NV)value);
45977657 2137 SvIV_set(sv, IV_MIN);
c2988b20
NC
2138 }
2139 }
2140 }
2141 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2142 will be in the previous block to set the IV slot, and the next
2143 block to set the NV slot. So no else here. */
2144
2145 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2146 != IS_NUMBER_IN_UV) {
2147 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2148 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2149
c2988b20
NC
2150 if (! numtype && ckWARN(WARN_NUMERIC))
2151 not_a_number(sv);
28e5dec8 2152
65202027 2153#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2154 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2155 PTR2UV(sv), SvNVX(sv)));
65202027 2156#else
1779d84d 2157 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2158 PTR2UV(sv), SvNVX(sv)));
65202027 2159#endif
28e5dec8 2160
28e5dec8 2161#ifdef NV_PRESERVES_UV
af359546
NC
2162 (void)SvIOKp_on(sv);
2163 (void)SvNOK_on(sv);
2164 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2165 SvIV_set(sv, I_V(SvNVX(sv)));
2166 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2167 SvIOK_on(sv);
2168 } else {
6f207bd3 2169 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2170 }
2171 /* UV will not work better than IV */
2172 } else {
2173 if (SvNVX(sv) > (NV)UV_MAX) {
2174 SvIsUV_on(sv);
2175 /* Integer is inaccurate. NOK, IOKp, is UV */
2176 SvUV_set(sv, UV_MAX);
af359546
NC
2177 } else {
2178 SvUV_set(sv, U_V(SvNVX(sv)));
2179 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2180 NV preservse UV so can do correct comparison. */
2181 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2182 SvIOK_on(sv);
af359546 2183 } else {
6f207bd3 2184 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2185 }
2186 }
4b0c9573 2187 SvIsUV_on(sv);
af359546 2188 }
28e5dec8 2189#else /* NV_PRESERVES_UV */
c2988b20
NC
2190 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2191 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2192 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2193 grok_number above. The NV slot has just been set using
2194 Atof. */
560b0c46 2195 SvNOK_on(sv);
c2988b20
NC
2196 assert (SvIOKp(sv));
2197 } else {
2198 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2199 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2200 /* Small enough to preserve all bits. */
2201 (void)SvIOKp_on(sv);
2202 SvNOK_on(sv);
45977657 2203 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2204 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2205 SvIOK_on(sv);
2206 /* Assumption: first non-preserved integer is < IV_MAX,
2207 this NV is in the preserved range, therefore: */
2208 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2209 < (UV)IV_MAX)) {
32fdb065 2210 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
2211 }
2212 } else {
2213 /* IN_UV NOT_INT
2214 0 0 already failed to read UV.
2215 0 1 already failed to read UV.
2216 1 0 you won't get here in this case. IV/UV
2217 slot set, public IOK, Atof() unneeded.
2218 1 1 already read UV.
2219 so there's no point in sv_2iuv_non_preserve() attempting
2220 to use atol, strtol, strtoul etc. */
47031da6 2221# ifdef DEBUGGING
40a17c4c 2222 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2223# else
2224 sv_2iuv_non_preserve (sv);
2225# endif
c2988b20
NC
2226 }
2227 }
28e5dec8 2228#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2229 /* It might be more code efficient to go through the entire logic above
2230 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2231 gets complex and potentially buggy, so more programmer efficient
2232 to do it this way, by turning off the public flags: */
2233 if (!numtype)
2234 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2235 }
af359546
NC
2236 }
2237 else {
675c862f 2238 if (isGV_with_GP(sv))
159b6efe 2239 return glob_2number(MUTABLE_GV(sv));
180488f8 2240
af359546
NC
2241 if (!(SvFLAGS(sv) & SVs_PADTMP)) {
2242 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2243 report_uninit(sv);
2244 }
25da4f38
IZ
2245 if (SvTYPE(sv) < SVt_IV)
2246 /* Typically the caller expects that sv_any is not NULL now. */
2247 sv_upgrade(sv, SVt_IV);
af359546
NC
2248 /* Return 0 from the caller. */
2249 return TRUE;
2250 }
2251 return FALSE;
2252}
2253
2254/*
2255=for apidoc sv_2iv_flags
2256
2257Return the integer value of an SV, doing any necessary string
2258conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2259Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2260
2261=cut
2262*/
2263
2264IV
5de3775c 2265Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2266{
97aff369 2267 dVAR;
af359546 2268 if (!sv)
a0d0e21e 2269 return 0;
cecf5685
NC
2270 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2271 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e
NC
2272 cache IVs just in case. In practice it seems that they never
2273 actually anywhere accessible by user Perl code, let alone get used
2274 in anything other than a string context. */
af359546
NC
2275 if (flags & SV_GMAGIC)
2276 mg_get(sv);
2277 if (SvIOKp(sv))
2278 return SvIVX(sv);
2279 if (SvNOKp(sv)) {
2280 return I_V(SvNVX(sv));
2281 }
71c558c3
NC
2282 if (SvPOKp(sv) && SvLEN(sv)) {
2283 UV value;
2284 const int numtype
2285 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2286
2287 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2288 == IS_NUMBER_IN_UV) {
2289 /* It's definitely an integer */
2290 if (numtype & IS_NUMBER_NEG) {
2291 if (value < (UV)IV_MIN)
2292 return -(IV)value;
2293 } else {
2294 if (value < (UV)IV_MAX)
2295 return (IV)value;
2296 }
2297 }
2298 if (!numtype) {
2299 if (ckWARN(WARN_NUMERIC))
2300 not_a_number(sv);
2301 }
2302 return I_V(Atof(SvPVX_const(sv)));
2303 }
1c7ff15e
NC
2304 if (SvROK(sv)) {
2305 goto return_rok;
af359546 2306 }
1c7ff15e
NC
2307 assert(SvTYPE(sv) >= SVt_PVMG);
2308 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2309 } else if (SvTHINKFIRST(sv)) {
af359546 2310 if (SvROK(sv)) {
1c7ff15e 2311 return_rok:
af359546 2312 if (SvAMAGIC(sv)) {
aee036bb
DM
2313 SV * tmpstr;
2314 if (flags & SV_SKIP_OVERLOAD)
2315 return 0;
31d632c3 2316 tmpstr = AMG_CALLunary(sv, numer_amg);
af359546
NC
2317 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2318 return SvIV(tmpstr);
2319 }
2320 }
2321 return PTR2IV(SvRV(sv));
2322 }
2323 if (SvIsCOW(sv)) {
2324 sv_force_normal_flags(sv, 0);
2325 }
2326 if (SvREADONLY(sv) && !SvOK(sv)) {
2327 if (ckWARN(WARN_UNINITIALIZED))
2328 report_uninit(sv);
2329 return 0;
2330 }
2331 }
2332 if (!SvIOKp(sv)) {
2333 if (S_sv_2iuv_common(aTHX_ sv))
2334 return 0;
79072805 2335 }
1d7c1841
GS
2336 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2337 PTR2UV(sv),SvIVX(sv)));
25da4f38 2338 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2339}
2340
645c22ef 2341/*
891f9566 2342=for apidoc sv_2uv_flags
645c22ef
DM
2343
2344Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2345conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2346Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2347
2348=cut
2349*/
2350
ff68c719 2351UV
5de3775c 2352Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2353{
97aff369 2354 dVAR;
ff68c719 2355 if (!sv)
2356 return 0;
cecf5685
NC
2357 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2358 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2359 cache IVs just in case. */
891f9566
YST
2360 if (flags & SV_GMAGIC)
2361 mg_get(sv);
ff68c719 2362 if (SvIOKp(sv))
2363 return SvUVX(sv);
2364 if (SvNOKp(sv))
2365 return U_V(SvNVX(sv));
71c558c3
NC
2366 if (SvPOKp(sv) && SvLEN(sv)) {
2367 UV value;
2368 const int numtype
2369 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2370
2371 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2372 == IS_NUMBER_IN_UV) {
2373 /* It's definitely an integer */
2374 if (!(numtype & IS_NUMBER_NEG))
2375 return value;
2376 }
2377 if (!numtype) {
2378 if (ckWARN(WARN_NUMERIC))
2379 not_a_number(sv);
2380 }
2381 return U_V(Atof(SvPVX_const(sv)));
2382 }
1c7ff15e
NC
2383 if (SvROK(sv)) {
2384 goto return_rok;
3fe9a6f1 2385 }
1c7ff15e
NC
2386 assert(SvTYPE(sv) >= SVt_PVMG);
2387 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
4cb1ec55 2388 } else if (SvTHINKFIRST(sv)) {
ff68c719 2389 if (SvROK(sv)) {
1c7ff15e 2390 return_rok:
deb46114 2391 if (SvAMAGIC(sv)) {
aee036bb
DM
2392 SV *tmpstr;
2393 if (flags & SV_SKIP_OVERLOAD)
2394 return 0;
31d632c3 2395 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2396 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2397 return SvUV(tmpstr);
2398 }
2399 }
2400 return PTR2UV(SvRV(sv));
ff68c719 2401 }
765f542d
NC
2402 if (SvIsCOW(sv)) {
2403 sv_force_normal_flags(sv, 0);
8a818333 2404 }
0336b60e 2405 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2406 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2407 report_uninit(sv);
ff68c719 2408 return 0;
2409 }
2410 }
af359546
NC
2411 if (!SvIOKp(sv)) {
2412 if (S_sv_2iuv_common(aTHX_ sv))
2413 return 0;
ff68c719 2414 }
25da4f38 2415
1d7c1841
GS
2416 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2417 PTR2UV(sv),SvUVX(sv)));
25da4f38 2418 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719 2419}
2420
645c22ef 2421/*
196007d1 2422=for apidoc sv_2nv_flags
645c22ef
DM
2423
2424Return the num value of an SV, doing any necessary string or integer
39d5de13
DM
2425conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2426Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2427
2428=cut
2429*/
2430
65202027 2431NV
39d5de13 2432Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
79072805 2433{
97aff369 2434 dVAR;
79072805
LW
2435 if (!sv)
2436 return 0.0;
cecf5685
NC
2437 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2438 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
50caf62e 2439 cache IVs just in case. */
39d5de13
DM
2440 if (flags & SV_GMAGIC)
2441 mg_get(sv);
463ee0b2
LW
2442 if (SvNOKp(sv))
2443 return SvNVX(sv);
0aa395f8 2444 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
041457d9 2445 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
504618e9 2446 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
a0d0e21e 2447 not_a_number(sv);
3f7c398e 2448 return Atof(SvPVX_const(sv));
a0d0e21e 2449 }
25da4f38 2450 if (SvIOKp(sv)) {
1c846c1f 2451 if (SvIsUV(sv))
65202027 2452 return (NV)SvUVX(sv);
25da4f38 2453 else
65202027 2454 return (NV)SvIVX(sv);
47a72cb8
NC
2455 }
2456 if (SvROK(sv)) {
2457 goto return_rok;
2458 }
2459 assert(SvTYPE(sv) >= SVt_PVMG);
2460 /* This falls through to the report_uninit near the end of the
2461 function. */
2462 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2463 if (SvROK(sv)) {
47a72cb8 2464 return_rok:
deb46114 2465 if (SvAMAGIC(sv)) {
aee036bb
DM
2466 SV *tmpstr;
2467 if (flags & SV_SKIP_OVERLOAD)
2468 return 0;
31d632c3 2469 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2470 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2471 return SvNV(tmpstr);
2472 }
2473 }
2474 return PTR2NV(SvRV(sv));
a0d0e21e 2475 }
765f542d
NC
2476 if (SvIsCOW(sv)) {
2477 sv_force_normal_flags(sv, 0);
8a818333 2478 }
0336b60e 2479 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2480 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2481 report_uninit(sv);
ed6116ce
LW
2482 return 0.0;
2483 }
79072805
LW
2484 }
2485 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2486 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2487 sv_upgrade(sv, SVt_NV);
906f284f 2488#ifdef USE_LONG_DOUBLE
097ee67d 2489 DEBUG_c({
f93f4e46 2490 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2491 PerlIO_printf(Perl_debug_log,
2492 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2493 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2494 RESTORE_NUMERIC_LOCAL();
2495 });
65202027 2496#else
572bbb43 2497 DEBUG_c({
f93f4e46 2498 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2499 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2500 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2501 RESTORE_NUMERIC_LOCAL();
2502 });
572bbb43 2503#endif
79072805
LW
2504 }
2505 else if (SvTYPE(sv) < SVt_PVNV)
2506 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2507 if (SvNOKp(sv)) {
2508 return SvNVX(sv);
61604483 2509 }
59d8ce62 2510 if (SvIOKp(sv)) {
9d6ce603 2511 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2512#ifdef NV_PRESERVES_UV
a43d94f2
NC
2513 if (SvIOK(sv))
2514 SvNOK_on(sv);
2515 else
2516 SvNOKp_on(sv);
28e5dec8
JH
2517#else
2518 /* Only set the public NV OK flag if this NV preserves the IV */
2519 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2520 if (SvIOK(sv) &&
2521 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2522 : (SvIVX(sv) == I_V(SvNVX(sv))))
2523 SvNOK_on(sv);
2524 else
2525 SvNOKp_on(sv);
2526#endif
93a17b20 2527 }
748a9306 2528 else if (SvPOKp(sv) && SvLEN(sv)) {
c2988b20 2529 UV value;
3f7c398e 2530 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2531 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2532 not_a_number(sv);
28e5dec8 2533#ifdef NV_PRESERVES_UV
c2988b20
NC
2534 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2535 == IS_NUMBER_IN_UV) {
5e045b90 2536 /* It's definitely an integer */
9d6ce603 2537 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2538 } else
3f7c398e 2539 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2540 if (numtype)
2541 SvNOK_on(sv);
2542 else
2543 SvNOKp_on(sv);
28e5dec8 2544#else
3f7c398e 2545 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2546 /* Only set the public NV OK flag if this NV preserves the value in
2547 the PV at least as well as an IV/UV would.
2548 Not sure how to do this 100% reliably. */
2549 /* if that shift count is out of range then Configure's test is
2550 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2551 UV_BITS */
2552 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2553 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2554 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2555 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2556 /* Can't use strtol etc to convert this string, so don't try.
2557 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2558 SvNOK_on(sv);
2559 } else {
2560 /* value has been set. It may not be precise. */
2561 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2562 /* 2s complement assumption for (UV)IV_MIN */
2563 SvNOK_on(sv); /* Integer is too negative. */
2564 } else {
2565 SvNOKp_on(sv);
2566 SvIOKp_on(sv);
6fa402ec 2567
c2988b20 2568 if (numtype & IS_NUMBER_NEG) {
45977657 2569 SvIV_set(sv, -(IV)value);
c2988b20 2570 } else if (value <= (UV)IV_MAX) {
45977657 2571 SvIV_set(sv, (IV)value);
c2988b20 2572 } else {
607fa7f2 2573 SvUV_set(sv, value);
c2988b20
NC
2574 SvIsUV_on(sv);
2575 }
2576
2577 if (numtype & IS_NUMBER_NOT_INT) {
2578 /* I believe that even if the original PV had decimals,
2579 they are lost beyond the limit of the FP precision.
2580 However, neither is canonical, so both only get p
2581 flags. NWC, 2000/11/25 */
2582 /* Both already have p flags, so do nothing */
2583 } else {
66a1b24b 2584 const NV nv = SvNVX(sv);
c2988b20
NC
2585 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2586 if (SvIVX(sv) == I_V(nv)) {
2587 SvNOK_on(sv);
c2988b20 2588 } else {
c2988b20
NC
2589 /* It had no "." so it must be integer. */
2590 }
00b6aa41 2591 SvIOK_on(sv);
c2988b20
NC
2592 } else {
2593 /* between IV_MAX and NV(UV_MAX).
2594 Could be slightly > UV_MAX */
6fa402ec 2595
c2988b20
NC
2596 if (numtype & IS_NUMBER_NOT_INT) {
2597 /* UV and NV both imprecise. */
2598 } else {
66a1b24b 2599 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2600
2601 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2602 SvNOK_on(sv);
c2988b20 2603 }
00b6aa41 2604 SvIOK_on(sv);
c2988b20
NC
2605 }
2606 }
2607 }
2608 }
2609 }
a43d94f2
NC
2610 /* It might be more code efficient to go through the entire logic above
2611 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2612 gets complex and potentially buggy, so more programmer efficient
2613 to do it this way, by turning off the public flags: */
2614 if (!numtype)
2615 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2616#endif /* NV_PRESERVES_UV */
93a17b20 2617 }
79072805 2618 else {
f7877b28 2619 if (isGV_with_GP(sv)) {
159b6efe 2620 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2621 return 0.0;
2622 }
2623
041457d9 2624 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2625 report_uninit(sv);
7e25a7e9
NC
2626 assert (SvTYPE(sv) >= SVt_NV);
2627 /* Typically the caller expects that sv_any is not NULL now. */
2628 /* XXX Ilya implies that this is a bug in callers that assume this
2629 and ideally should be fixed. */
a0d0e21e 2630 return 0.0;
79072805 2631 }
572bbb43 2632#if defined(USE_LONG_DOUBLE)
097ee67d 2633 DEBUG_c({
f93f4e46 2634 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2635 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2636 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2637 RESTORE_NUMERIC_LOCAL();
2638 });
65202027 2639#else
572bbb43 2640 DEBUG_c({
f93f4e46 2641 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2642 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2643 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2644 RESTORE_NUMERIC_LOCAL();
2645 });
572bbb43 2646#endif
463ee0b2 2647 return SvNVX(sv);
79072805
LW
2648}
2649
800401ee
JH
2650/*
2651=for apidoc sv_2num
2652
2653Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2654reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2655access this function.
800401ee
JH
2656
2657=cut
2658*/
2659
2660SV *
5de3775c 2661Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2662{
7918f24d
NC
2663 PERL_ARGS_ASSERT_SV_2NUM;
2664
b9ee0594
RGS
2665 if (!SvROK(sv))
2666 return sv;
800401ee 2667 if (SvAMAGIC(sv)) {
31d632c3 2668 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2669 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2670 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2671 return sv_2num(tmpsv);
2672 }
2673 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2674}
2675
645c22ef
DM
2676/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2677 * UV as a string towards the end of buf, and return pointers to start and
2678 * end of it.
2679 *
2680 * We assume that buf is at least TYPE_CHARS(UV) long.
2681 */
2682
864dbfa3 2683static char *
5de3775c 2684S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2685{
25da4f38 2686 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2687 char * const ebuf = ptr;
25da4f38 2688 int sign;
25da4f38 2689
7918f24d
NC
2690 PERL_ARGS_ASSERT_UIV_2BUF;
2691
25da4f38
IZ
2692 if (is_uv)
2693 sign = 0;
2694 else if (iv >= 0) {
2695 uv = iv;
2696 sign = 0;
2697 } else {
2698 uv = -iv;
2699 sign = 1;
2700 }
2701 do {
eb160463 2702 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2703 } while (uv /= 10);
2704 if (sign)
2705 *--ptr = '-';
2706 *peob = ebuf;
2707 return ptr;
2708}
2709
645c22ef
DM
2710/*
2711=for apidoc sv_2pv_flags
2712
ff276b08 2713Returns a pointer to the string value of an SV, and sets *lp to its length.
645c22ef
DM
2714If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2715if necessary.
2716Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2717usually end up here too.
2718
2719=cut
2720*/
2721
8d6d96c1 2722char *
5de3775c 2723Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2724{
97aff369 2725 dVAR;
79072805 2726 register char *s;
79072805 2727
463ee0b2 2728 if (!sv) {
cdb061a3
NC
2729 if (lp)
2730 *lp = 0;
73d840c0 2731 return (char *)"";
463ee0b2 2732 }
8990e307 2733 if (SvGMAGICAL(sv)) {
8d6d96c1
HS
2734 if (flags & SV_GMAGIC)
2735 mg_get(sv);
463ee0b2 2736 if (SvPOKp(sv)) {
cdb061a3
NC
2737 if (lp)
2738 *lp = SvCUR(sv);
10516c54
NC
2739 if (flags & SV_MUTABLE_RETURN)
2740 return SvPVX_mutable(sv);
4d84ee25
NC
2741 if (flags & SV_CONST_RETURN)
2742 return (char *)SvPVX_const(sv);
463ee0b2
LW
2743 return SvPVX(sv);
2744 }
75dfc8ec
NC
2745 if (SvIOKp(sv) || SvNOKp(sv)) {
2746 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
75dfc8ec
NC
2747 STRLEN len;
2748
2749 if (SvIOKp(sv)) {
e80fed9d 2750 len = SvIsUV(sv)
d9fad198
JH
2751 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2752 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
29912d93
NC
2753 } else if(SvNVX(sv) == 0.0) {
2754 tbuf[0] = '0';
2755 tbuf[1] = 0;
2756 len = 1;
75dfc8ec 2757 } else {
e8ada2d0
NC
2758 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2759 len = strlen(tbuf);
75dfc8ec 2760 }
b5b886f0
NC
2761 assert(!SvROK(sv));
2762 {
75dfc8ec
NC
2763 dVAR;
2764
75dfc8ec
NC
2765 SvUPGRADE(sv, SVt_PV);
2766 if (lp)
2767 *lp = len;
2768 s = SvGROW_mutable(sv, len + 1);
2769 SvCUR_set(sv, len);
2770 SvPOKp_on(sv);
10edeb5d 2771 return (char*)memcpy(s, tbuf, len + 1);
75dfc8ec 2772 }
463ee0b2 2773 }
1c7ff15e
NC
2774 if (SvROK(sv)) {
2775 goto return_rok;
2776 }
2777 assert(SvTYPE(sv) >= SVt_PVMG);
2778 /* This falls through to the report_uninit near the end of the
2779 function. */
2780 } else if (SvTHINKFIRST(sv)) {
ed6116ce 2781 if (SvROK(sv)) {
1c7ff15e 2782 return_rok:
deb46114 2783 if (SvAMAGIC(sv)) {
aee036bb
DM
2784 SV *tmpstr;
2785 if (flags & SV_SKIP_OVERLOAD)
2786 return NULL;
31d632c3 2787 tmpstr = AMG_CALLunary(sv, string_amg);
a02ec77a 2788 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
deb46114
NC
2789 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2790 /* Unwrap this: */
2791 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2792 */
2793
2794 char *pv;
2795 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2796 if (flags & SV_CONST_RETURN) {
2797 pv = (char *) SvPVX_const(tmpstr);
2798 } else {
2799 pv = (flags & SV_MUTABLE_RETURN)
2800 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2801 }
2802 if (lp)
2803 *lp = SvCUR(tmpstr);
50adf7d2 2804 } else {
deb46114 2805 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2806 }
deb46114
NC
2807 if (SvUTF8(tmpstr))
2808 SvUTF8_on(sv);
2809 else
2810 SvUTF8_off(sv);
2811 return pv;
50adf7d2 2812 }
deb46114
NC
2813 }
2814 {
fafee734
NC
2815 STRLEN len;
2816 char *retval;
2817 char *buffer;
d2c6dc5e 2818 SV *const referent = SvRV(sv);
d8eae41e
NC
2819
2820 if (!referent) {
fafee734
NC
2821 len = 7;
2822 retval = buffer = savepvn("NULLREF", len);
5c35adbb 2823 } else if (SvTYPE(referent) == SVt_REGEXP) {
d2c6dc5e 2824 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
67d2d14d
AB
2825 I32 seen_evals = 0;
2826
2827 assert(re);
2828
2829 /* If the regex is UTF-8 we want the containing scalar to
2830 have an UTF-8 flag too */
2831 if (RX_UTF8(re))
2832 SvUTF8_on(sv);
2833 else
2834 SvUTF8_off(sv);
2835
2836 if ((seen_evals = RX_SEEN_EVALS(re)))
2837 PL_reginterp_cnt += seen_evals;
2838
2839 if (lp)
2840 *lp = RX_WRAPLEN(re);
2841
2842 return RX_WRAPPED(re);
d8eae41e
NC
2843 } else {
2844 const char *const typestr = sv_reftype(referent, 0);
fafee734
NC
2845 const STRLEN typelen = strlen(typestr);
2846 UV addr = PTR2UV(referent);
2847 const char *stashname = NULL;
2848 STRLEN stashnamelen = 0; /* hush, gcc */
2849 const char *buffer_end;
d8eae41e 2850
d8eae41e 2851 if (SvOBJECT(referent)) {
fafee734
NC
2852 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2853
2854 if (name) {
2855 stashname = HEK_KEY(name);
2856 stashnamelen = HEK_LEN(name);
2857
2858 if (HEK_UTF8(name)) {
2859 SvUTF8_on(sv);
2860 } else {
2861 SvUTF8_off(sv);
2862 }
2863 } else {
2864 stashname = "__ANON__";
2865 stashnamelen = 8;
2866 }
2867 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2868 + 2 * sizeof(UV) + 2 /* )\0 */;
2869 } else {
2870 len = typelen + 3 /* (0x */
2871 + 2 * sizeof(UV) + 2 /* )\0 */;
d8eae41e 2872 }
fafee734
NC
2873
2874 Newx(buffer, len, char);
2875 buffer_end = retval = buffer + len;
2876
2877 /* Working backwards */
2878 *--retval = '\0';
2879 *--retval = ')';
2880 do {
2881 *--retval = PL_hexdigit[addr & 15];
2882 } while (addr >>= 4);
2883 *--retval = 'x';
2884 *--retval = '0';
2885 *--retval = '(';
2886
2887 retval -= typelen;
2888 memcpy(retval, typestr, typelen);
2889
2890 if (stashname) {
2891 *--retval = '=';
2892 retval -= stashnamelen;
2893 memcpy(retval, stashname, stashnamelen);
2894 }
486ec47a 2895 /* retval may not necessarily have reached the start of the
fafee734
NC
2896 buffer here. */
2897 assert (retval >= buffer);
2898
2899 len = buffer_end - retval - 1; /* -1 for that \0 */
c080367d 2900 }
042dae7a 2901 if (lp)
fafee734
NC
2902 *lp = len;
2903 SAVEFREEPV(buffer);
2904 return retval;
463ee0b2 2905 }
79072805 2906 }
0336b60e 2907 if (SvREADONLY(sv) && !SvOK(sv)) {
cdb061a3
NC
2908 if (lp)
2909 *lp = 0;
9f621bb0
NC
2910 if (flags & SV_UNDEF_RETURNS_NULL)
2911 return NULL;
2912 if (ckWARN(WARN_UNINITIALIZED))
2913 report_uninit(sv);
73d840c0 2914 return (char *)"";
79072805 2915 }
79072805 2916 }
28e5dec8
JH
2917 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2918 /* I'm assuming that if both IV and NV are equally valid then
2919 converting the IV is going to be more efficient */
e1ec3a88 2920 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2921 char buf[TYPE_CHARS(UV)];
2922 char *ebuf, *ptr;
97a130b8 2923 STRLEN len;
28e5dec8
JH
2924
2925 if (SvTYPE(sv) < SVt_PVIV)
2926 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2927 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2928 len = ebuf - ptr;
5902b6a9 2929 /* inlined from sv_setpvn */
97a130b8
NC
2930 s = SvGROW_mutable(sv, len + 1);
2931 Move(ptr, s, len, char);
2932 s += len;
28e5dec8 2933 *s = '\0';
28e5dec8
JH
2934 }
2935 else if (SvNOKp(sv)) {
79072805
LW
2936 if (SvTYPE(sv) < SVt_PVNV)
2937 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2938 if (SvNVX(sv) == 0.0) {
2939 s = SvGROW_mutable(sv, 2);
2940 *s++ = '0';
2941 *s = '\0';
2942 } else {
2943 dSAVE_ERRNO;
2944 /* The +20 is pure guesswork. Configure test needed. --jhi */
2945 s = SvGROW_mutable(sv, NV_DIG + 20);
2946 /* some Xenix systems wipe out errno here */
2d4389e4 2947 Gconvert(SvNVX(sv), NV_DIG, 0, s);
29912d93
NC
2948 RESTORE_ERRNO;
2949 while (*s) s++;
bbce6d69 2950 }
79072805
LW
2951#ifdef hcx
2952 if (s[-1] == '.')
46fc3d4c 2953 *--s = '\0';
79072805
LW
2954#endif
2955 }
79072805 2956 else {
8d1c3e26
NC
2957 if (isGV_with_GP(sv)) {
2958 GV *const gv = MUTABLE_GV(sv);
2959 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
2960 SV *const buffer = sv_newmortal();
2961
2962 /* FAKE globs can get coerced, so need to turn this off temporarily
2963 if it is on. */
2964 SvFAKE_off(gv);
2965 gv_efullname3(buffer, gv, "*");
2966 SvFLAGS(gv) |= wasfake;
2967
1809c940
DM
2968 if (SvPOK(buffer)) {
2969 if (lp) {
2970 *lp = SvCUR(buffer);
2971 }
2972 return SvPVX(buffer);
2973 }
2974 else {
2975 if (lp)
2976 *lp = 0;
2977 return (char *)"";
8d1c3e26 2978 }
8d1c3e26 2979 }
180488f8 2980
cdb061a3 2981 if (lp)
00b6aa41 2982 *lp = 0;
9f621bb0
NC
2983 if (flags & SV_UNDEF_RETURNS_NULL)
2984 return NULL;
2985 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
2986 report_uninit(sv);
25da4f38
IZ
2987 if (SvTYPE(sv) < SVt_PV)
2988 /* Typically the caller expects that sv_any is not NULL now. */
2989 sv_upgrade(sv, SVt_PV);
73d840c0 2990 return (char *)"";
79072805 2991 }
cdb061a3 2992 {
823a54a3 2993 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2994 if (lp)
2995 *lp = len;
2996 SvCUR_set(sv, len);
2997 }
79072805 2998 SvPOK_on(sv);
1d7c1841 2999 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3000 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3001 if (flags & SV_CONST_RETURN)
3002 return (char *)SvPVX_const(sv);
10516c54
NC
3003 if (flags & SV_MUTABLE_RETURN)
3004 return SvPVX_mutable(sv);
463ee0b2
LW
3005 return SvPVX(sv);
3006}
3007
645c22ef 3008/*
6050d10e
JP
3009=for apidoc sv_copypv
3010
3011Copies a stringified representation of the source SV into the
3012destination SV. Automatically performs any necessary mg_get and
54f0641b 3013coercion of numeric values into strings. Guaranteed to preserve
2575c402 3014UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3015sv_2pv[_flags] but operates directly on an SV instead of just the
3016string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3017would lose the UTF-8'ness of the PV.
3018
3019=cut
3020*/
3021
3022void
5de3775c 3023Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 3024{
446eaa42 3025 STRLEN len;
53c1dcc0 3026 const char * const s = SvPV_const(ssv,len);
7918f24d
NC
3027
3028 PERL_ARGS_ASSERT_SV_COPYPV;
3029
cb50f42d 3030 sv_setpvn(dsv,s,len);
446eaa42 3031 if (SvUTF8(ssv))
cb50f42d 3032 SvUTF8_on(dsv);
446eaa42 3033 else
cb50f42d 3034 SvUTF8_off(dsv);
6050d10e
JP
3035}
3036
3037/*
645c22ef
DM
3038=for apidoc sv_2pvbyte
3039
3040Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3041to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3042side-effect.
3043
3044Usually accessed via the C<SvPVbyte> macro.
3045
3046=cut
3047*/
3048
7340a771 3049char *
5de3775c 3050Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3051{
7918f24d
NC
3052 PERL_ARGS_ASSERT_SV_2PVBYTE;
3053
71eb6d8c 3054 SvGETMAGIC(sv);
0875d2fe 3055 sv_utf8_downgrade(sv,0);
71eb6d8c 3056 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3057}
3058
645c22ef 3059/*
035cbb0e
RGS
3060=for apidoc sv_2pvutf8
3061
3062Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3063to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3064
3065Usually accessed via the C<SvPVutf8> macro.
3066
3067=cut
3068*/
645c22ef 3069
7340a771 3070char *
7bc54cea 3071Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
7340a771 3072{
7918f24d
NC
3073 PERL_ARGS_ASSERT_SV_2PVUTF8;
3074
035cbb0e
RGS
3075 sv_utf8_upgrade(sv);
3076 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
7340a771 3077}
1c846c1f 3078
7ee2227d 3079
645c22ef
DM
3080/*
3081=for apidoc sv_2bool
3082
06c841cf
FC
3083This macro is only used by sv_true() or its macro equivalent, and only if
3084the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3085It calls sv_2bool_flags with the SV_GMAGIC flag.
3086
3087=for apidoc sv_2bool_flags
3088
3089This function is only used by sv_true() and friends, and only if
3090the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3091contain SV_GMAGIC, then it does an mg_get() first.
3092
645c22ef
DM
3093
3094=cut
3095*/
3096
463ee0b2 3097bool
06c841cf 3098Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
463ee0b2 3099{
97aff369 3100 dVAR;
7918f24d 3101
06c841cf 3102 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3103
06c841cf 3104 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3105
a0d0e21e
LW
3106 if (!SvOK(sv))
3107 return 0;
3108 if (SvROK(sv)) {
fabdb6c0 3109 if (SvAMAGIC(sv)) {
31d632c3 3110 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3111 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3112 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3113 }
3114 return SvRV(sv) != 0;
a0d0e21e 3115 }
463ee0b2 3116 if (SvPOKp(sv)) {
53c1dcc0
AL
3117 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3118 if (Xpvtmp &&
339049b0 3119 (*sv->sv_u.svu_pv > '0' ||
11343788 3120 Xpvtmp->xpv_cur > 1 ||
339049b0 3121 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
463ee0b2
LW
3122 return 1;
3123 else
3124 return 0;
3125 }
3126 else {
3127 if (SvIOKp(sv))
3128 return SvIVX(sv) != 0;
3129 else {
3130 if (SvNOKp(sv))
3131 return SvNVX(sv) != 0.0;
180488f8 3132 else {
f7877b28 3133 if (isGV_with_GP(sv))
180488f8
NC
3134 return TRUE;
3135 else
3136 return FALSE;
3137 }
463ee0b2
LW
3138 }
3139 }
79072805
LW
3140}
3141
c461cf8f
JH
3142/*
3143=for apidoc sv_utf8_upgrade
3144
78ea37eb 3145Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3146Forces the SV to string form if it is not already.
2bbc8d55 3147Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3148Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3149if the whole string is the same in UTF-8 as not.
3150Returns the number of bytes in the converted string
c461cf8f 3151
13a6c0e0
JH
3152This is not as a general purpose byte encoding to Unicode interface:
3153use the Encode extension for that.
3154
fe749c9a
KW
3155=for apidoc sv_utf8_upgrade_nomg
3156
3157Like sv_utf8_upgrade, but doesn't do magic on C<sv>
3158
8d6d96c1
HS
3159=for apidoc sv_utf8_upgrade_flags
3160
78ea37eb 3161Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3162Forces the SV to string form if it is not already.
8d6d96c1 3163Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3164if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3165will C<mg_get> on C<sv> if appropriate, else not.
3166Returns the number of bytes in the converted string
3167C<sv_utf8_upgrade> and
8d6d96c1
HS
3168C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3169
13a6c0e0
JH
3170This is not as a general purpose byte encoding to Unicode interface:
3171use the Encode extension for that.
3172
8d6d96c1 3173=cut
b3ab6785
KW
3174
3175The grow version is currently not externally documented. It adds a parameter,
3176extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3177have free after it upon return. This allows the caller to reserve extra space
3178that it intends to fill, to avoid extra grows.
3179
3180Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3181which can be used to tell this function to not first check to see if there are
3182any characters that are different in UTF-8 (variant characters) which would
3183force it to allocate a new string to sv, but to assume there are. Typically
3184this flag is used by a routine that has already parsed the string to find that
3185there are such characters, and passes this information on so that the work
3186doesn't have to be repeated.
3187
3188(One might think that the calling routine could pass in the position of the
3189first such variant, so it wouldn't have to be found again. But that is not the
3190case, because typically when the caller is likely to use this flag, it won't be
3191calling this routine unless it finds something that won't fit into a byte.
3192Otherwise it tries to not upgrade and just use bytes. But some things that
3193do fit into a byte are variants in utf8, and the caller may not have been
3194keeping track of these.)
3195
3196If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3197isn't guaranteed due to having other routines do the work in some input cases,
3198or if the input is already flagged as being in utf8.
3199
3200The speed of this could perhaps be improved for many cases if someone wanted to
3201write a fast function that counts the number of variant characters in a string,
3202especially if it could return the position of the first one.
3203
8d6d96c1
HS
3204*/
3205
3206STRLEN
b3ab6785 3207Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3208{
97aff369 3209 dVAR;
7918f24d 3210
b3ab6785 3211 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3212
808c356f
RGS
3213 if (sv == &PL_sv_undef)
3214 return 0;
e0e62c2a
NIS
3215 if (!SvPOK(sv)) {
3216 STRLEN len = 0;
d52b7888
NC
3217 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3218 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3219 if (SvUTF8(sv)) {
3220 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3221 return len;
b3ab6785 3222 }
d52b7888
NC
3223 } else {
3224 (void) SvPV_force(sv,len);
3225 }
e0e62c2a 3226 }
4411f3b6 3227
f5cee72b 3228 if (SvUTF8(sv)) {
b3ab6785 3229 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3230 return SvCUR(sv);
f5cee72b 3231 }
5fec3b1d 3232
765f542d
NC
3233 if (SvIsCOW(sv)) {
3234 sv_force_normal_flags(sv, 0);
db42d148
NIS
3235 }
3236
b3ab6785 3237 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3238 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3239 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3240 return SvCUR(sv);
3241 }
3242
4e93345f
KW
3243 if (SvCUR(sv) == 0) {
3244 if (extra) SvGROW(sv, extra);
3245 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3246 /* This function could be much more efficient if we
2bbc8d55 3247 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3248 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3249 * make the loop as fast as possible (although there are certainly ways
3250 * to speed this up, eg. through vectorization) */
3251 U8 * s = (U8 *) SvPVX_const(sv);
3252 U8 * e = (U8 *) SvEND(sv);
3253 U8 *t = s;
3254 STRLEN two_byte_count = 0;
c4e7c712 3255
b3ab6785
KW
3256 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3257
3258 /* See if really will need to convert to utf8. We mustn't rely on our
3259 * incoming SV being well formed and having a trailing '\0', as certain
3260 * code in pp_formline can send us partially built SVs. */
3261
c4e7c712 3262 while (t < e) {
53c1dcc0 3263 const U8 ch = *t++;
b3ab6785
KW
3264 if (NATIVE_IS_INVARIANT(ch)) continue;
3265
3266 t--; /* t already incremented; re-point to first variant */
3267 two_byte_count = 1;
3268 goto must_be_utf8;
c4e7c712 3269 }
b3ab6785
KW
3270
3271 /* utf8 conversion not needed because all are invariants. Mark as
3272 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3273 SvUTF8_on(sv);
b3ab6785
KW
3274 return SvCUR(sv);
3275
3276must_be_utf8:
3277
3278 /* Here, the string should be converted to utf8, either because of an
3279 * input flag (two_byte_count = 0), or because a character that
3280 * requires 2 bytes was found (two_byte_count = 1). t points either to
3281 * the beginning of the string (if we didn't examine anything), or to
3282 * the first variant. In either case, everything from s to t - 1 will
3283 * occupy only 1 byte each on output.
3284 *
3285 * There are two main ways to convert. One is to create a new string
3286 * and go through the input starting from the beginning, appending each
3287 * converted value onto the new string as we go along. It's probably
3288 * best to allocate enough space in the string for the worst possible
3289 * case rather than possibly running out of space and having to
3290 * reallocate and then copy what we've done so far. Since everything
3291 * from s to t - 1 is invariant, the destination can be initialized
3292 * with these using a fast memory copy
3293 *
3294 * The other way is to figure out exactly how big the string should be
3295 * by parsing the entire input. Then you don't have to make it big
3296 * enough to handle the worst possible case, and more importantly, if
3297 * the string you already have is large enough, you don't have to
3298 * allocate a new string, you can copy the last character in the input
3299 * string to the final position(s) that will be occupied by the
3300 * converted string and go backwards, stopping at t, since everything
3301 * before that is invariant.
3302 *
3303 * There are advantages and disadvantages to each method.
3304 *
3305 * In the first method, we can allocate a new string, do the memory
3306 * copy from the s to t - 1, and then proceed through the rest of the
3307 * string byte-by-byte.
3308 *
3309 * In the second method, we proceed through the rest of the input
3310 * string just calculating how big the converted string will be. Then
3311 * there are two cases:
3312 * 1) if the string has enough extra space to handle the converted
3313 * value. We go backwards through the string, converting until we
3314 * get to the position we are at now, and then stop. If this
3315 * position is far enough along in the string, this method is
3316 * faster than the other method. If the memory copy were the same
3317 * speed as the byte-by-byte loop, that position would be about
3318 * half-way, as at the half-way mark, parsing to the end and back
3319 * is one complete string's parse, the same amount as starting
3320 * over and going all the way through. Actually, it would be
3321 * somewhat less than half-way, as it's faster to just count bytes
3322 * than to also copy, and we don't have the overhead of allocating
3323 * a new string, changing the scalar to use it, and freeing the
3324 * existing one. But if the memory copy is fast, the break-even
3325 * point is somewhere after half way. The counting loop could be
3326 * sped up by vectorization, etc, to move the break-even point
3327 * further towards the beginning.
3328 * 2) if the string doesn't have enough space to handle the converted
3329 * value. A new string will have to be allocated, and one might
3330 * as well, given that, start from the beginning doing the first
3331 * method. We've spent extra time parsing the string and in
3332 * exchange all we've gotten is that we know precisely how big to
3333 * make the new one. Perl is more optimized for time than space,
3334 * so this case is a loser.
3335 * So what I've decided to do is not use the 2nd method unless it is
3336 * guaranteed that a new string won't have to be allocated, assuming
3337 * the worst case. I also decided not to put any more conditions on it
3338 * than this, for now. It seems likely that, since the worst case is
3339 * twice as big as the unknown portion of the string (plus 1), we won't
3340 * be guaranteed enough space, causing us to go to the first method,
3341 * unless the string is short, or the first variant character is near
3342 * the end of it. In either of these cases, it seems best to use the
3343 * 2nd method. The only circumstance I can think of where this would
3344 * be really slower is if the string had once had much more data in it
3345 * than it does now, but there is still a substantial amount in it */
3346
3347 {
3348 STRLEN invariant_head = t - s;
3349 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3350 if (SvLEN(sv) < size) {
3351
3352 /* Here, have decided to allocate a new string */
3353
3354 U8 *dst;
3355 U8 *d;
3356
3357 Newx(dst, size, U8);
3358
3359 /* If no known invariants at the beginning of the input string,
3360 * set so starts from there. Otherwise, can use memory copy to
3361 * get up to where we are now, and then start from here */
3362
3363 if (invariant_head <= 0) {
3364 d = dst;
3365 } else {
3366 Copy(s, dst, invariant_head, char);
3367 d = dst + invariant_head;
3368 }
3369
3370 while (t < e) {
3371 const UV uv = NATIVE8_TO_UNI(*t++);
3372 if (UNI_IS_INVARIANT(uv))
3373 *d++ = (U8)UNI_TO_NATIVE(uv);
3374 else {
3375 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3376 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3377 }
3378 }
3379 *d = '\0';
3380 SvPV_free(sv); /* No longer using pre-existing string */
3381 SvPV_set(sv, (char*)dst);
3382 SvCUR_set(sv, d - dst);
3383 SvLEN_set(sv, size);
3384 } else {
3385
3386 /* Here, have decided to get the exact size of the string.
3387 * Currently this happens only when we know that there is
3388 * guaranteed enough space to fit the converted string, so
3389 * don't have to worry about growing. If two_byte_count is 0,
3390 * then t points to the first byte of the string which hasn't
3391 * been examined yet. Otherwise two_byte_count is 1, and t
3392 * points to the first byte in the string that will expand to
3393 * two. Depending on this, start examining at t or 1 after t.
3394 * */
3395
3396 U8 *d = t + two_byte_count;
3397
3398
3399 /* Count up the remaining bytes that expand to two */
3400
3401 while (d < e) {
3402 const U8 chr = *d++;
3403 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3404 }
3405
3406 /* The string will expand by just the number of bytes that
3407 * occupy two positions. But we are one afterwards because of
3408 * the increment just above. This is the place to put the
3409 * trailing NUL, and to set the length before we decrement */
3410
3411 d += two_byte_count;
3412 SvCUR_set(sv, d - s);
3413 *d-- = '\0';
3414
3415
3416 /* Having decremented d, it points to the position to put the
3417 * very last byte of the expanded string. Go backwards through
3418 * the string, copying and expanding as we go, stopping when we
3419 * get to the part that is invariant the rest of the way down */
3420
3421 e--;
3422 while (e >= t) {
3423 const U8 ch = NATIVE8_TO_UNI(*e--);
3424 if (UNI_IS_INVARIANT(ch)) {
3425 *d-- = UNI_TO_NATIVE(ch);
3426 } else {
3427 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3428 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3429 }
3430 }
3431 }
3432 }
560a288e 3433 }
b3ab6785
KW
3434
3435 /* Mark as UTF-8 even if no variant - saves scanning loop */
3436 SvUTF8_on(sv);
4411f3b6 3437 return SvCUR(sv);
560a288e
GS
3438}
3439
c461cf8f
JH
3440/*
3441=for apidoc sv_utf8_downgrade
3442
78ea37eb 3443Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3444If the PV contains a character that cannot fit
3445in a byte, this conversion will fail;
78ea37eb 3446in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3447true, croaks.
3448
13a6c0e0
JH
3449This is not as a general purpose Unicode to byte encoding interface:
3450use the Encode extension for that.
3451
c461cf8f
JH
3452=cut
3453*/
3454
560a288e 3455bool
7bc54cea 3456Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
560a288e 3457{
97aff369 3458 dVAR;
7918f24d
NC
3459
3460 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3461
78ea37eb 3462 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3463 if (SvCUR(sv)) {
03cfe0ae 3464 U8 *s;
652088fc 3465 STRLEN len;
fa301091 3466
765f542d
NC
3467 if (SvIsCOW(sv)) {
3468 sv_force_normal_flags(sv, 0);
3469 }
03cfe0ae
NIS
3470 s = (U8 *) SvPV(sv, len);
3471 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3472 if (fail_ok)
3473 return FALSE;
3474 else {
3475 if (PL_op)
3476 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3477 OP_DESC(PL_op));
fa301091
JH
3478 else
3479 Perl_croak(aTHX_ "Wide character");
3480 }
4b3603a4 3481 }
b162af07 3482 SvCUR_set(sv, len);
67e989fb 3483 }
560a288e 3484 }
ffebcc3e 3485 SvUTF8_off(sv);
560a288e
GS
3486 return TRUE;
3487}
3488
c461cf8f
JH
3489/*
3490=for apidoc sv_utf8_encode
3491
78ea37eb
TS
3492Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3493flag off so that it looks like octets again.
c461cf8f
JH
3494
3495=cut
3496*/
3497
560a288e 3498void
7bc54cea 3499Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3500{
7918f24d
NC
3501 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3502
4c94c214
NC
3503 if (SvIsCOW(sv)) {
3504 sv_force_normal_flags(sv, 0);
3505 }
3506 if (SvREADONLY(sv)) {
6ad8f254 3507 Perl_croak_no_modify(aTHX);
4c94c214 3508 }
a5f5288a 3509 (void) sv_utf8_upgrade(sv);
560a288e
GS
3510 SvUTF8_off(sv);
3511}
3512
4411f3b6
NIS
3513/*
3514=for apidoc sv_utf8_decode
3515
78ea37eb
TS
3516If the PV of the SV is an octet sequence in UTF-8
3517and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3518so that it looks like a character. If the PV contains only single-byte
3519characters, the C<SvUTF8> flag stays being off.
3520Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3521
3522=cut
3523*/
3524
560a288e 3525bool
7bc54cea 3526Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3527{
7918f24d
NC
3528 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3529
78ea37eb 3530 if (SvPOKp(sv)) {
93524f2b
NC
3531 const U8 *c;
3532 const U8 *e;
9cbac4c7 3533
645c22ef
DM
3534 /* The octets may have got themselves encoded - get them back as
3535 * bytes
3536 */
3537 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3538 return FALSE;
3539
3540 /* it is actually just a matter of turning the utf8 flag on, but
3541 * we want to make sure everything inside is valid utf8 first.
3542 */
93524f2b 3543 c = (const U8 *) SvPVX_const(sv);
63cd0674 3544 if (!is_utf8_string(c, SvCUR(sv)+1))
67e989fb 3545 return FALSE;
93524f2b 3546 e = (const U8 *) SvEND(sv);
511c2ff0 3547 while (c < e) {
b64e5050 3548 const U8 ch = *c++;
c4d5f83a 3549 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3550 SvUTF8_on(sv);
3551 break;
3552 }
560a288e 3553 }
560a288e
GS
3554 }
3555 return TRUE;
3556}
3557
954c1994
GS
3558/*
3559=for apidoc sv_setsv
3560
645c22ef
DM
3561Copies the contents of the source SV C<ssv> into the destination SV
3562C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3563function if the source SV needs to be reused. Does not handle 'set' magic.
3564Loosely speaking, it performs a copy-by-value, obliterating any previous
3565content of the destination.
3566
3567You probably want to use one of the assortment of wrappers, such as
3568C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3569C<SvSetMagicSV_nosteal>.
3570
8d6d96c1
HS
3571=for apidoc sv_setsv_flags
3572
645c22ef
DM
3573Copies the contents of the source SV C<ssv> into the destination SV
3574C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3575function if the source SV needs to be reused. Does not handle 'set' magic.
3576Loosely speaking, it performs a copy-by-value, obliterating any previous
3577content of the destination.
3578If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
5fcdf167
NC
3579C<ssv> if appropriate, else not. If the C<flags> parameter has the
3580C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3581and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3582
3583You probably want to use one of the assortment of wrappers, such as
3584C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3585C<SvSetMagicSV_nosteal>.
3586
3587This is the primary function for copying scalars, and most other
3588copy-ish functions and macros use this underneath.
8d6d96c1
HS
3589
3590=cut
3591*/
3592
5d0301b7 3593static void
7bc54cea 3594S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3595{
c8bbf675 3596 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3e6edce2 3597 HV *old_stash = NULL;
dd69841b 3598
7918f24d
NC
3599 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3600
bec4f4b4 3601 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
5d0301b7
NC
3602 const char * const name = GvNAME(sstr);
3603 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3604 {
f7877b28
NC
3605 if (dtype >= SVt_PV) {
3606 SvPV_free(dstr);
3607 SvPV_set(dstr, 0);
3608 SvLEN_set(dstr, 0);
3609 SvCUR_set(dstr, 0);
3610 }
0d092c36 3611 SvUPGRADE(dstr, SVt_PVGV);
dedf8e73 3612 (void)SvOK_off(dstr);
2e5b91de
NC
3613 /* FIXME - why are we doing this, then turning it off and on again
3614 below? */
3615 isGV_with_GP_on(dstr);
f7877b28 3616 }
5d0301b7
NC
3617 GvSTASH(dstr) = GvSTASH(sstr);
3618 if (GvSTASH(dstr))
daba3364 3619 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
159b6efe 3620 gv_name_set(MUTABLE_GV(dstr), name, len, GV_ADD);
5d0301b7
NC
3621 SvFAKE_on(dstr); /* can coerce to non-glob */
3622 }
3623
159b6efe 3624 if(GvGP(MUTABLE_GV(sstr))) {
dd69841b
BB
3625 /* If source has method cache entry, clear it */
3626 if(GvCVGEN(sstr)) {
3627 SvREFCNT_dec(GvCV(sstr));
c43ae56f 3628 GvCV_set(sstr, NULL);
dd69841b
BB
3629 GvCVGEN(sstr) = 0;
3630 }
3631 /* If source has a real method, then a method is
3632 going to change */
00169e2c
FC
3633 else if(
3634 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3635 ) {
70cd14a1 3636 mro_changes = 1;
dd69841b
BB
3637 }
3638 }
3639
3640 /* If dest already had a real method, that's a change as well */
00169e2c
FC
3641 if(
3642 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3643 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3644 ) {
70cd14a1 3645 mro_changes = 1;
dd69841b
BB
3646 }
3647
c8bbf675
FC
3648 /* We don’t need to check the name of the destination if it was not a
3649 glob to begin with. */
3650 if(dtype == SVt_PVGV) {
3651 const char * const name = GvNAME((const GV *)dstr);
00169e2c
FC
3652 if(
3653 strEQ(name,"ISA")
3654 /* The stash may have been detached from the symbol table, so
3655 check its name. */