This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
[perl #114864] Don’t use amt for DESTROY
[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
484e6108
FC
185# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
186 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
187 } STMT_END
d7a2c63c
MHM
188# define DEBUG_SV_SERIAL(sv) \
189 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
190 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
191#else
192# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 193# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
194#endif
195
48614a46
NC
196#ifdef PERL_POISON
197# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 198# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
199/* Whilst I'd love to do this, it seems that things like to check on
200 unreferenced scalars
7e337ee0 201# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 202*/
7e337ee0
JH
203# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
204 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
205#else
206# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 207# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
208# define POSION_SV_HEAD(sv)
209#endif
210
990198f0
DM
211/* Mark an SV head as unused, and add to free list.
212 *
213 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
214 * its refcount artificially decremented during global destruction, so
215 * there may be dangling pointers to it. The last thing we want in that
216 * case is for it to be reused. */
217
053fc874
GS
218#define plant_SV(p) \
219 STMT_START { \
990198f0 220 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
221 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
222 DEBUG_SV_SERIAL(p); \
fd0854ff 223 FREE_SV_DEBUG_FILE(p); \
48614a46 224 POSION_SV_HEAD(p); \
053fc874 225 SvFLAGS(p) = SVTYPEMASK; \
990198f0 226 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 227 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
228 PL_sv_root = (p); \
229 } \
053fc874
GS
230 --PL_sv_count; \
231 } STMT_END
a0d0e21e 232
053fc874
GS
233#define uproot_SV(p) \
234 STMT_START { \
235 (p) = PL_sv_root; \
daba3364 236 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
237 ++PL_sv_count; \
238 } STMT_END
239
645c22ef 240
cac9b346
NC
241/* make some more SVs by adding another arena */
242
cac9b346
NC
243STATIC SV*
244S_more_sv(pTHX)
245{
97aff369 246 dVAR;
cac9b346 247 SV* sv;
9a87bd09
NC
248 char *chunk; /* must use New here to match call to */
249 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
250 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
251 uproot_SV(sv);
252 return sv;
253}
254
645c22ef
DM
255/* new_SV(): return a new, empty SV head */
256
eba0f806
DM
257#ifdef DEBUG_LEAKING_SCALARS
258/* provide a real function for a debugger to play with */
259STATIC SV*
d7a2c63c 260S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
261{
262 SV* sv;
263
eba0f806
DM
264 if (PL_sv_root)
265 uproot_SV(sv);
266 else
cac9b346 267 sv = S_more_sv(aTHX);
eba0f806
DM
268 SvANY(sv) = 0;
269 SvREFCNT(sv) = 1;
270 SvFLAGS(sv) = 0;
fd0854ff 271 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
272 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
273 ? PL_parser->copline
274 : PL_curcop
f24aceb1
DM
275 ? CopLINE(PL_curcop)
276 : 0
e385c3bf 277 );
fd0854ff 278 sv->sv_debug_inpad = 0;
cd676548 279 sv->sv_debug_parent = NULL;
484e6108 280 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
281
282 sv->sv_debug_serial = PL_sv_serial++;
283
284 MEM_LOG_NEW_SV(sv, file, line, func);
285 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
286 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
287
eba0f806
DM
288 return sv;
289}
d7a2c63c 290# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
291
292#else
293# define new_SV(p) \
053fc874 294 STMT_START { \
053fc874
GS
295 if (PL_sv_root) \
296 uproot_SV(p); \
297 else \
cac9b346 298 (p) = S_more_sv(aTHX); \
053fc874
GS
299 SvANY(p) = 0; \
300 SvREFCNT(p) = 1; \
301 SvFLAGS(p) = 0; \
d7a2c63c 302 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 303 } STMT_END
eba0f806 304#endif
463ee0b2 305
645c22ef
DM
306
307/* del_SV(): return an empty SV head to the free list */
308
a0d0e21e 309#ifdef DEBUGGING
4561caa4 310
053fc874
GS
311#define del_SV(p) \
312 STMT_START { \
aea4f609 313 if (DEBUG_D_TEST) \
053fc874
GS
314 del_sv(p); \
315 else \
316 plant_SV(p); \
053fc874 317 } STMT_END
a0d0e21e 318
76e3520e 319STATIC void
cea2e8a9 320S_del_sv(pTHX_ SV *p)
463ee0b2 321{
97aff369 322 dVAR;
7918f24d
NC
323
324 PERL_ARGS_ASSERT_DEL_SV;
325
aea4f609 326 if (DEBUG_D_TEST) {
4633a7c4 327 SV* sva;
a3b680e6 328 bool ok = 0;
daba3364 329 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
330 const SV * const sv = sva + 1;
331 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 332 if (p >= sv && p < svend) {
a0d0e21e 333 ok = 1;
c0ff570e
NC
334 break;
335 }
a0d0e21e
LW
336 }
337 if (!ok) {
9b387841
NC
338 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
339 "Attempt to free non-arena SV: 0x%"UVxf
340 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
341 return;
342 }
343 }
4561caa4 344 plant_SV(p);
463ee0b2 345}
a0d0e21e 346
4561caa4
CS
347#else /* ! DEBUGGING */
348
349#define del_SV(p) plant_SV(p)
350
351#endif /* DEBUGGING */
463ee0b2 352
645c22ef
DM
353
354/*
ccfc67b7
JH
355=head1 SV Manipulation Functions
356
645c22ef
DM
357=for apidoc sv_add_arena
358
359Given a chunk of memory, link it to the head of the list of arenas,
360and split it into a list of free SVs.
361
362=cut
363*/
364
d2bd4e7f
NC
365static void
366S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 367{
97aff369 368 dVAR;
daba3364 369 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
370 SV* sv;
371 SV* svend;
4633a7c4 372
7918f24d
NC
373 PERL_ARGS_ASSERT_SV_ADD_ARENA;
374
4633a7c4 375 /* The first SV in an arena isn't an SV. */
3280af22 376 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
377 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
378 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
379
3280af22
NIS
380 PL_sv_arenaroot = sva;
381 PL_sv_root = sva + 1;
4633a7c4
LW
382
383 svend = &sva[SvREFCNT(sva) - 1];
384 sv = sva + 1;
463ee0b2 385 while (sv < svend) {
3eef1deb 386 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 387#ifdef DEBUGGING
978b032e 388 SvREFCNT(sv) = 0;
03e36789 389#endif
4b69cbe3 390 /* Must always set typemask because it's always checked in on cleanup
03e36789 391 when the arenas are walked looking for objects. */
8990e307 392 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
393 sv++;
394 }
3eef1deb 395 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
396#ifdef DEBUGGING
397 SvREFCNT(sv) = 0;
398#endif
4633a7c4
LW
399 SvFLAGS(sv) = SVTYPEMASK;
400}
401
055972dc
DM
402/* visit(): call the named function for each non-free SV in the arenas
403 * whose flags field matches the flags/mask args. */
645c22ef 404
5226ed68 405STATIC I32
de37a194 406S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 407{
97aff369 408 dVAR;
4633a7c4 409 SV* sva;
5226ed68 410 I32 visited = 0;
8990e307 411
7918f24d
NC
412 PERL_ARGS_ASSERT_VISIT;
413
daba3364 414 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
415 const SV * const svend = &sva[SvREFCNT(sva)];
416 SV* sv;
4561caa4 417 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 418 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
419 && (sv->sv_flags & mask) == flags
420 && SvREFCNT(sv))
421 {
acfe0abc 422 (FCALL)(aTHX_ sv);
5226ed68
JH
423 ++visited;
424 }
8990e307
LW
425 }
426 }
5226ed68 427 return visited;
8990e307
LW
428}
429
758a08c3
JH
430#ifdef DEBUGGING
431
645c22ef
DM
432/* called by sv_report_used() for each live SV */
433
434static void
5fa45a31 435do_report_used(pTHX_ SV *const sv)
645c22ef 436{
e4787c0c 437 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
438 PerlIO_printf(Perl_debug_log, "****\n");
439 sv_dump(sv);
440 }
441}
758a08c3 442#endif
645c22ef
DM
443
444/*
445=for apidoc sv_report_used
446
fde67290 447Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
448
449=cut
450*/
451
8990e307 452void
864dbfa3 453Perl_sv_report_used(pTHX)
4561caa4 454{
ff270d3a 455#ifdef DEBUGGING
055972dc 456 visit(do_report_used, 0, 0);
96a5add6
AL
457#else
458 PERL_UNUSED_CONTEXT;
ff270d3a 459#endif
4561caa4
CS
460}
461
645c22ef
DM
462/* called by sv_clean_objs() for each live SV */
463
464static void
de37a194 465do_clean_objs(pTHX_ SV *const ref)
645c22ef 466{
97aff369 467 dVAR;
ea724faa
NC
468 assert (SvROK(ref));
469 {
823a54a3
AL
470 SV * const target = SvRV(ref);
471 if (SvOBJECT(target)) {
472 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
473 if (SvWEAKREF(ref)) {
474 sv_del_backref(target, ref);
475 SvWEAKREF_off(ref);
476 SvRV_set(ref, NULL);
477 } else {
478 SvROK_off(ref);
479 SvRV_set(ref, NULL);
480 SvREFCNT_dec(target);
481 }
645c22ef
DM
482 }
483 }
484
485 /* XXX Might want to check arrays, etc. */
486}
487
645c22ef 488
e4487e9b
DM
489/* clear any slots in a GV which hold objects - except IO;
490 * called by sv_clean_objs() for each live GV */
491
645c22ef 492static void
f30de749 493do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 494{
97aff369 495 dVAR;
57ef47cc 496 SV *obj;
ea724faa 497 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 498 assert(isGV_with_GP(sv));
57ef47cc
DM
499 if (!GvGP(sv))
500 return;
501
502 /* freeing GP entries may indirectly free the current GV;
503 * hold onto it while we mess with the GP slots */
504 SvREFCNT_inc(sv);
505
506 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
507 DEBUG_D((PerlIO_printf(Perl_debug_log,
508 "Cleaning named glob SV object:\n "), sv_dump(obj)));
509 GvSV(sv) = NULL;
510 SvREFCNT_dec(obj);
511 }
512 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
513 DEBUG_D((PerlIO_printf(Perl_debug_log,
514 "Cleaning named glob AV object:\n "), sv_dump(obj)));
515 GvAV(sv) = NULL;
516 SvREFCNT_dec(obj);
517 }
518 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
519 DEBUG_D((PerlIO_printf(Perl_debug_log,
520 "Cleaning named glob HV object:\n "), sv_dump(obj)));
521 GvHV(sv) = NULL;
522 SvREFCNT_dec(obj);
523 }
524 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log,
526 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 527 GvCV_set(sv, NULL);
57ef47cc
DM
528 SvREFCNT_dec(obj);
529 }
e4487e9b
DM
530 SvREFCNT_dec(sv); /* undo the inc above */
531}
532
68b590d9 533/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
534 * called by sv_clean_objs() for each live GV */
535
536static void
537do_clean_named_io_objs(pTHX_ SV *const sv)
538{
539 dVAR;
540 SV *obj;
541 assert(SvTYPE(sv) == SVt_PVGV);
542 assert(isGV_with_GP(sv));
68b590d9 543 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
544 return;
545
546 SvREFCNT_inc(sv);
57ef47cc
DM
547 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
548 DEBUG_D((PerlIO_printf(Perl_debug_log,
549 "Cleaning named glob IO object:\n "), sv_dump(obj)));
550 GvIOp(sv) = NULL;
551 SvREFCNT_dec(obj);
645c22ef 552 }
57ef47cc 553 SvREFCNT_dec(sv); /* undo the inc above */
645c22ef 554}
645c22ef 555
4155e4fe
FC
556/* Void wrapper to pass to visit() */
557static void
558do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
559 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
560 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
561 return;
562 (void)curse(sv, 0);
563}
564
645c22ef
DM
565/*
566=for apidoc sv_clean_objs
567
fde67290 568Attempt to destroy all objects not yet freed.
645c22ef
DM
569
570=cut
571*/
572
4561caa4 573void
864dbfa3 574Perl_sv_clean_objs(pTHX)
4561caa4 575{
97aff369 576 dVAR;
68b590d9 577 GV *olddef, *olderr;
3280af22 578 PL_in_clean_objs = TRUE;
055972dc 579 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
580 /* Some barnacles may yet remain, clinging to typeglobs.
581 * Run the non-IO destructors first: they may want to output
582 * error messages, close files etc */
d011219a 583 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 584 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
585 /* And if there are some very tenacious barnacles clinging to arrays,
586 closures, or what have you.... */
587 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
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
fde67290 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
fde67290 675Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
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
6e128786
NC
898 { sizeof(NV), sizeof(NV),
899 STRUCT_OFFSET(XPVNV, xnv_u),
900 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 901
bc337e5c 902 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
903 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
904 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 905 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 906 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 907
bc337e5c 908 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
909 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
910 + STRUCT_OFFSET(XPV, xpv_cur),
911 SVt_PVIV, FALSE, NONV, HASARENA,
912 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 913
bc337e5c 914 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
915 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
916 + STRUCT_OFFSET(XPV, xpv_cur),
917 SVt_PVNV, FALSE, HADNV, HASARENA,
918 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 919
6e128786 920 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 921 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 922
601dfd0a
NC
923 { sizeof(regexp),
924 sizeof(regexp),
925 0,
08e44740 926 SVt_REGEXP, FALSE, NONV, HASARENA,
eaeb1e7f 927 FIT_ARENA(0, sizeof(regexp))
5c35adbb 928 },
4df7f6af 929
10666ae3 930 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
931 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
932
10666ae3 933 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
934 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
935
601dfd0a 936 { sizeof(XPVAV),
4f7003f5 937 copy_length(XPVAV, xav_alloc),
601dfd0a 938 0,
69ba284b 939 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 940 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 941
601dfd0a 942 { sizeof(XPVHV),
359164a0 943 copy_length(XPVHV, xhv_max),
601dfd0a 944 0,
69ba284b 945 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 946 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 947
601dfd0a
NC
948 { sizeof(XPVCV),
949 sizeof(XPVCV),
950 0,
69ba284b 951 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 952 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 953
601dfd0a
NC
954 { sizeof(XPVFM),
955 sizeof(XPVFM),
956 0,
69ba284b 957 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 958 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 959
601dfd0a
NC
960 { sizeof(XPVIO),
961 sizeof(XPVIO),
962 0,
b6f60916 963 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 964 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 965};
29489e7c 966
bd81e77b 967#define new_body_allocated(sv_type) \
d2a0f284 968 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 969 - bodies_by_type[sv_type].offset)
29489e7c 970
26359cfa
NC
971/* return a thing to the free list */
972
973#define del_body(thing, root) \
974 STMT_START { \
975 void ** const thing_copy = (void **)thing; \
976 *thing_copy = *root; \
977 *root = (void*)thing_copy; \
978 } STMT_END
29489e7c 979
bd81e77b 980#ifdef PURIFY
29489e7c 981
beeec492
NC
982#define new_XNV() safemalloc(sizeof(XPVNV))
983#define new_XPVNV() safemalloc(sizeof(XPVNV))
984#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 985
beeec492 986#define del_XPVGV(p) safefree(p)
29489e7c 987
bd81e77b 988#else /* !PURIFY */
29489e7c 989
65ac1738 990#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 991#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 992#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 993
26359cfa
NC
994#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
995 &PL_body_roots[SVt_PVGV])
1d7c1841 996
bd81e77b 997#endif /* PURIFY */
93e68bfb 998
bd81e77b 999/* no arena for you! */
93e68bfb 1000
bd81e77b 1001#define new_NOARENA(details) \
beeec492 1002 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1003#define new_NOARENAZ(details) \
beeec492 1004 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1005
1e30fcd5
NC
1006void *
1007Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1008 const size_t arena_size)
d2a0f284
JC
1009{
1010 dVAR;
1011 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1012 struct arena_desc *adesc;
1013 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1014 unsigned int curr;
d2a0f284
JC
1015 char *start;
1016 const char *end;
02982131 1017 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
0b2d3faa 1018#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1019 static bool done_sanity_check;
1020
0b2d3faa
JH
1021 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1022 * variables like done_sanity_check. */
10666ae3 1023 if (!done_sanity_check) {
ea471437 1024 unsigned int i = SVt_LAST;
10666ae3
NC
1025
1026 done_sanity_check = TRUE;
1027
1028 while (i--)
1029 assert (bodies_by_type[i].type == i);
1030 }
1031#endif
1032
02982131 1033 assert(arena_size);
23e9d66c 1034
99816f8d
NC
1035 /* may need new arena-set to hold new arena */
1036 if (!aroot || aroot->curr >= aroot->set_size) {
1037 struct arena_set *newroot;
1038 Newxz(newroot, 1, struct arena_set);
1039 newroot->set_size = ARENAS_PER_SET;
1040 newroot->next = aroot;
1041 aroot = newroot;
1042 PL_body_arenas = (void *) newroot;
1043 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1044 }
1045
1046 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1047 curr = aroot->curr++;
1048 adesc = &(aroot->set[curr]);
1049 assert(!adesc->arena);
1050
1051 Newx(adesc->arena, good_arena_size, char);
1052 adesc->size = good_arena_size;
1053 adesc->utype = sv_type;
1054 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1055 curr, (void*)adesc->arena, (UV)good_arena_size));
1056
1057 start = (char *) adesc->arena;
d2a0f284 1058
29657bb6
NC
1059 /* Get the address of the byte after the end of the last body we can fit.
1060 Remember, this is integer division: */
02982131 1061 end = start + good_arena_size / body_size * body_size;
d2a0f284 1062
486ec47a 1063 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1064#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1065 DEBUG_m(PerlIO_printf(Perl_debug_log,
1066 "arena %p end %p arena-size %d (from %d) type %d "
1067 "size %d ct %d\n",
02982131
NC
1068 (void*)start, (void*)end, (int)good_arena_size,
1069 (int)arena_size, sv_type, (int)body_size,
1070 (int)good_arena_size / (int)body_size));
d8fca402 1071#else
d2a0f284
JC
1072 DEBUG_m(PerlIO_printf(Perl_debug_log,
1073 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1074 (void*)start, (void*)end,
02982131
NC
1075 (int)arena_size, sv_type, (int)body_size,
1076 (int)good_arena_size / (int)body_size));
d8fca402 1077#endif
d2a0f284
JC
1078 *root = (void *)start;
1079
29657bb6
NC
1080 while (1) {
1081 /* Where the next body would start: */
d2a0f284 1082 char * const next = start + body_size;
29657bb6
NC
1083
1084 if (next >= end) {
1085 /* This is the last body: */
1086 assert(next == end);
1087
1088 *(void **)start = 0;
1089 return *root;
1090 }
1091
d2a0f284
JC
1092 *(void**) start = (void *)next;
1093 start = next;
1094 }
d2a0f284
JC
1095}
1096
1097/* grab a new thing from the free list, allocating more if necessary.
1098 The inline version is used for speed in hot routines, and the
1099 function using it serves the rest (unless PURIFY).
1100*/
1101#define new_body_inline(xpv, sv_type) \
1102 STMT_START { \
1103 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1104 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1105 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1106 bodies_by_type[sv_type].body_size,\
1107 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1108 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1109 } STMT_END
1110
1111#ifndef PURIFY
1112
1113STATIC void *
de37a194 1114S_new_body(pTHX_ const svtype sv_type)
d2a0f284
JC
1115{
1116 dVAR;
1117 void *xpv;
1118 new_body_inline(xpv, sv_type);
1119 return xpv;
1120}
1121
1122#endif
93e68bfb 1123
238b27b3
NC
1124static const struct body_details fake_rv =
1125 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1126
bd81e77b
NC
1127/*
1128=for apidoc sv_upgrade
93e68bfb 1129
bd81e77b
NC
1130Upgrade an SV to a more complex form. Generally adds a new body type to the
1131SV, then copies across as much information as possible from the old body.
9521ca61
FC
1132It croaks if the SV is already in a more complex form than requested. You
1133generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1134before calling C<sv_upgrade>, and hence does not croak. See also
1135C<svtype>.
93e68bfb 1136
bd81e77b 1137=cut
93e68bfb 1138*/
93e68bfb 1139
bd81e77b 1140void
aad570aa 1141Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
cac9b346 1142{
97aff369 1143 dVAR;
bd81e77b
NC
1144 void* old_body;
1145 void* new_body;
42d0e0b7 1146 const svtype old_type = SvTYPE(sv);
d2a0f284 1147 const struct body_details *new_type_details;
238b27b3 1148 const struct body_details *old_type_details
bd81e77b 1149 = bodies_by_type + old_type;
4df7f6af 1150 SV *referant = NULL;
cac9b346 1151
7918f24d
NC
1152 PERL_ARGS_ASSERT_SV_UPGRADE;
1153
1776cbe8
NC
1154 if (old_type == new_type)
1155 return;
1156
1157 /* This clause was purposefully added ahead of the early return above to
1158 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1159 inference by Nick I-S that it would fix other troublesome cases. See
1160 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1161
1162 Given that shared hash key scalars are no longer PVIV, but PV, there is
1163 no longer need to unshare so as to free up the IVX slot for its proper
1164 purpose. So it's safe to move the early return earlier. */
1165
093085a8 1166 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1167 sv_force_normal_flags(sv, 0);
1168 }
cac9b346 1169
bd81e77b 1170 old_body = SvANY(sv);
de042e1d 1171
bd81e77b
NC
1172 /* Copying structures onto other structures that have been neatly zeroed
1173 has a subtle gotcha. Consider XPVMG
cac9b346 1174
bd81e77b
NC
1175 +------+------+------+------+------+-------+-------+
1176 | NV | CUR | LEN | IV | MAGIC | STASH |
1177 +------+------+------+------+------+-------+-------+
1178 0 4 8 12 16 20 24 28
645c22ef 1179
bd81e77b
NC
1180 where NVs are aligned to 8 bytes, so that sizeof that structure is
1181 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1182
bd81e77b
NC
1183 +------+------+------+------+------+-------+-------+------+
1184 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1185 +------+------+------+------+------+-------+-------+------+
1186 0 4 8 12 16 20 24 28 32
08742458 1187
bd81e77b 1188 so what happens if you allocate memory for this structure:
30f9da9e 1189
bd81e77b
NC
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1192 +------+------+------+------+------+-------+-------+------+------+...
1193 0 4 8 12 16 20 24 28 32 36
bfc44f79 1194
bd81e77b
NC
1195 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1196 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1197 started out as zero once, but it's quite possible that it isn't. So now,
1198 rather than a nicely zeroed GP, you have it pointing somewhere random.
1199 Bugs ensue.
bfc44f79 1200
bd81e77b
NC
1201 (In fact, GP ends up pointing at a previous GP structure, because the
1202 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1203 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1204 this happens to be moot because XPVGV has been re-ordered, with GP
1205 no longer after STASH)
30f9da9e 1206
bd81e77b
NC
1207 So we are careful and work out the size of used parts of all the
1208 structures. */
bfc44f79 1209
bd81e77b
NC
1210 switch (old_type) {
1211 case SVt_NULL:
1212 break;
1213 case SVt_IV:
4df7f6af
NC
1214 if (SvROK(sv)) {
1215 referant = SvRV(sv);
238b27b3
NC
1216 old_type_details = &fake_rv;
1217 if (new_type == SVt_NV)
1218 new_type = SVt_PVNV;
4df7f6af
NC
1219 } else {
1220 if (new_type < SVt_PVIV) {
1221 new_type = (new_type == SVt_NV)
1222 ? SVt_PVNV : SVt_PVIV;
1223 }
bd81e77b
NC
1224 }
1225 break;
1226 case SVt_NV:
1227 if (new_type < SVt_PVNV) {
1228 new_type = SVt_PVNV;
bd81e77b
NC
1229 }
1230 break;
bd81e77b
NC
1231 case SVt_PV:
1232 assert(new_type > SVt_PV);
1233 assert(SVt_IV < SVt_PV);
1234 assert(SVt_NV < SVt_PV);
1235 break;
1236 case SVt_PVIV:
1237 break;
1238 case SVt_PVNV:
1239 break;
1240 case SVt_PVMG:
1241 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1242 there's no way that it can be safely upgraded, because perl.c
1243 expects to Safefree(SvANY(PL_mess_sv)) */
1244 assert(sv != PL_mess_sv);
1245 /* This flag bit is used to mean other things in other scalar types.
1246 Given that it only has meaning inside the pad, it shouldn't be set
1247 on anything that can get upgraded. */
00b1698f 1248 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1249 break;
1250 default:
1251 if (old_type_details->cant_upgrade)
c81225bc
NC
1252 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1253 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1254 }
3376de98
NC
1255
1256 if (old_type > new_type)
1257 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1258 (int)old_type, (int)new_type);
1259
2fa1109b 1260 new_type_details = bodies_by_type + new_type;
645c22ef 1261
bd81e77b
NC
1262 SvFLAGS(sv) &= ~SVTYPEMASK;
1263 SvFLAGS(sv) |= new_type;
932e9ff9 1264
ab4416c0
NC
1265 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1266 the return statements above will have triggered. */
1267 assert (new_type != SVt_NULL);
bd81e77b 1268 switch (new_type) {
bd81e77b
NC
1269 case SVt_IV:
1270 assert(old_type == SVt_NULL);
1271 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1272 SvIV_set(sv, 0);
1273 return;
1274 case SVt_NV:
1275 assert(old_type == SVt_NULL);
1276 SvANY(sv) = new_XNV();
1277 SvNV_set(sv, 0);
1278 return;
bd81e77b 1279 case SVt_PVHV:
bd81e77b 1280 case SVt_PVAV:
d2a0f284 1281 assert(new_type_details->body_size);
c1ae03ae
NC
1282
1283#ifndef PURIFY
1284 assert(new_type_details->arena);
d2a0f284 1285 assert(new_type_details->arena_size);
c1ae03ae 1286 /* This points to the start of the allocated area. */
d2a0f284
JC
1287 new_body_inline(new_body, new_type);
1288 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1289 new_body = ((char *)new_body) - new_type_details->offset;
1290#else
1291 /* We always allocated the full length item with PURIFY. To do this
1292 we fake things so that arena is false for all 16 types.. */
1293 new_body = new_NOARENAZ(new_type_details);
1294#endif
1295 SvANY(sv) = new_body;
1296 if (new_type == SVt_PVAV) {
1297 AvMAX(sv) = -1;
1298 AvFILLp(sv) = -1;
1299 AvREAL_only(sv);
64484faa 1300 if (old_type_details->body_size) {
ac572bf4
NC
1301 AvALLOC(sv) = 0;
1302 } else {
1303 /* It will have been zeroed when the new body was allocated.
1304 Lets not write to it, in case it confuses a write-back
1305 cache. */
1306 }
78ac7dd9
NC
1307 } else {
1308 assert(!SvOK(sv));
1309 SvOK_off(sv);
1310#ifndef NODEFAULT_SHAREKEYS
1311 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1312#endif
1313 HvMAX(sv) = 7; /* (start with 8 buckets) */
c1ae03ae 1314 }
aeb18a1e 1315
bd81e77b
NC
1316 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1317 The target created by newSVrv also is, and it can have magic.
1318 However, it never has SvPVX set.
1319 */
4df7f6af
NC
1320 if (old_type == SVt_IV) {
1321 assert(!SvROK(sv));
1322 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1323 assert(SvPVX_const(sv) == 0);
1324 }
aeb18a1e 1325
bd81e77b 1326 if (old_type >= SVt_PVMG) {
e736a858 1327 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1328 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1329 } else {
1330 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1331 }
1332 break;
93e68bfb 1333
bd81e77b
NC
1334 case SVt_PVIV:
1335 /* XXX Is this still needed? Was it ever needed? Surely as there is
1336 no route from NV to PVIV, NOK can never be true */
1337 assert(!SvNOKp(sv));
1338 assert(!SvNOK(sv));
1339 case SVt_PVIO:
1340 case SVt_PVFM:
bd81e77b
NC
1341 case SVt_PVGV:
1342 case SVt_PVCV:
1343 case SVt_PVLV:
12c45b25 1344 case SVt_REGEXP:
bd81e77b
NC
1345 case SVt_PVMG:
1346 case SVt_PVNV:
1347 case SVt_PV:
93e68bfb 1348
d2a0f284 1349 assert(new_type_details->body_size);
bd81e77b
NC
1350 /* We always allocated the full length item with PURIFY. To do this
1351 we fake things so that arena is false for all 16 types.. */
1352 if(new_type_details->arena) {
1353 /* This points to the start of the allocated area. */
d2a0f284
JC
1354 new_body_inline(new_body, new_type);
1355 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1356 new_body = ((char *)new_body) - new_type_details->offset;
1357 } else {
1358 new_body = new_NOARENAZ(new_type_details);
1359 }
1360 SvANY(sv) = new_body;
5e2fc214 1361
bd81e77b 1362 if (old_type_details->copy) {
f9ba3d20
NC
1363 /* There is now the potential for an upgrade from something without
1364 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1365 int offset = old_type_details->offset;
1366 int length = old_type_details->copy;
1367
1368 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1369 const int difference
f9ba3d20
NC
1370 = new_type_details->offset - old_type_details->offset;
1371 offset += difference;
1372 length -= difference;
1373 }
1374 assert (length >= 0);
1375
1376 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1377 char);
bd81e77b
NC
1378 }
1379
1380#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1381 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1382 * correct 0.0 for us. Otherwise, if the old body didn't have an
1383 * NV slot, but the new one does, then we need to initialise the
1384 * freshly created NV slot with whatever the correct bit pattern is
1385 * for 0.0 */
e22a937e
NC
1386 if (old_type_details->zero_nv && !new_type_details->zero_nv
1387 && !isGV_with_GP(sv))
bd81e77b 1388 SvNV_set(sv, 0);
82048762 1389#endif
5e2fc214 1390
85dca89a
NC
1391 if (new_type == SVt_PVIO) {
1392 IO * const io = MUTABLE_IO(sv);
d963bf01 1393 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1394
1395 SvOBJECT_on(io);
1396 /* Clear the stashcache because a new IO could overrule a package
1397 name */
103f5a36 1398 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1399 hv_clear(PL_stashcache);
1400
85dca89a 1401 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1402 IoPAGE_LEN(sv) = 60;
85dca89a 1403 }
8d919b0a
FC
1404 if (new_type == SVt_REGEXP)
1405 sv->sv_u.svu_rx = (regexp *)new_body;
1406 else if (old_type < SVt_PV) {
4df7f6af
NC
1407 /* referant will be NULL unless the old type was SVt_IV emulating
1408 SVt_RV */
1409 sv->sv_u.svu_rv = referant;
1410 }
bd81e77b
NC
1411 break;
1412 default:
afd78fd5
JH
1413 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1414 (unsigned long)new_type);
bd81e77b 1415 }
73171d91 1416
db93c0c4 1417 if (old_type > SVt_IV) {
bd81e77b 1418#ifdef PURIFY
beeec492 1419 safefree(old_body);
bd81e77b 1420#else
bc786448
GG
1421 /* Note that there is an assumption that all bodies of types that
1422 can be upgraded came from arenas. Only the more complex non-
1423 upgradable types are allowed to be directly malloc()ed. */
1424 assert(old_type_details->arena);
bd81e77b
NC
1425 del_body((void*)((char*)old_body + old_type_details->offset),
1426 &PL_body_roots[old_type]);
1427#endif
1428 }
1429}
73171d91 1430
bd81e77b
NC
1431/*
1432=for apidoc sv_backoff
73171d91 1433
fde67290 1434Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1435wrapper instead.
73171d91 1436
bd81e77b 1437=cut
73171d91
NC
1438*/
1439
bd81e77b 1440int
aad570aa 1441Perl_sv_backoff(pTHX_ register SV *const sv)
bd81e77b 1442{
69240efd 1443 STRLEN delta;
7a4bba22 1444 const char * const s = SvPVX_const(sv);
7918f24d
NC
1445
1446 PERL_ARGS_ASSERT_SV_BACKOFF;
96a5add6 1447 PERL_UNUSED_CONTEXT;
7918f24d 1448
bd81e77b
NC
1449 assert(SvOOK(sv));
1450 assert(SvTYPE(sv) != SVt_PVHV);
1451 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1452
69240efd
NC
1453 SvOOK_offset(sv, delta);
1454
7a4bba22
NC
1455 SvLEN_set(sv, SvLEN(sv) + delta);
1456 SvPV_set(sv, SvPVX(sv) - delta);
1457 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1458 SvFLAGS(sv) &= ~SVf_OOK;
1459 return 0;
1460}
73171d91 1461
bd81e77b
NC
1462/*
1463=for apidoc sv_grow
73171d91 1464
bd81e77b
NC
1465Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1466upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1467Use the C<SvGROW> wrapper instead.
93e68bfb 1468
bd81e77b
NC
1469=cut
1470*/
93e68bfb 1471
bd81e77b 1472char *
aad570aa 1473Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
bd81e77b 1474{
eb578fdb 1475 char *s;
93e68bfb 1476
7918f24d
NC
1477 PERL_ARGS_ASSERT_SV_GROW;
1478
5db06880
NC
1479 if (PL_madskills && newlen >= 0x100000) {
1480 PerlIO_printf(Perl_debug_log,
1481 "Allocation too large: %"UVxf"\n", (UV)newlen);
1482 }
bd81e77b
NC
1483#ifdef HAS_64K_LIMIT
1484 if (newlen >= 0x10000) {
1485 PerlIO_printf(Perl_debug_log,
1486 "Allocation too large: %"UVxf"\n", (UV)newlen);
1487 my_exit(1);
1488 }
1489#endif /* HAS_64K_LIMIT */
1490 if (SvROK(sv))
1491 sv_unref(sv);
1492 if (SvTYPE(sv) < SVt_PV) {
1493 sv_upgrade(sv, SVt_PV);
1494 s = SvPVX_mutable(sv);
1495 }
1496 else if (SvOOK(sv)) { /* pv is offset? */
1497 sv_backoff(sv);
1498 s = SvPVX_mutable(sv);
1499 if (newlen > SvLEN(sv))
1500 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1501#ifdef HAS_64K_LIMIT
1502 if (newlen >= 0x10000)
1503 newlen = 0xFFFF;
1504#endif
1505 }
1506 else
1507 s = SvPVX_mutable(sv);
aeb18a1e 1508
bd81e77b 1509 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559
WH
1510 STRLEN minlen = SvCUR(sv);
1511 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1512 if (newlen < minlen)
1513 newlen = minlen;
aedff202 1514#ifndef Perl_safesysmalloc_size
bd81e77b 1515 newlen = PERL_STRLEN_ROUNDUP(newlen);
bd81e77b 1516#endif
98653f18 1517 if (SvLEN(sv) && s) {
10edeb5d 1518 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1519 }
1520 else {
10edeb5d 1521 s = (char*)safemalloc(newlen);
bd81e77b
NC
1522 if (SvPVX_const(sv) && SvCUR(sv)) {
1523 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1524 }
1525 }
1526 SvPV_set(sv, s);
ca7c1a29 1527#ifdef Perl_safesysmalloc_size
98653f18
NC
1528 /* Do this here, do it once, do it right, and then we will never get
1529 called back into sv_grow() unless there really is some growing
1530 needed. */
ca7c1a29 1531 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1532#else
bd81e77b 1533 SvLEN_set(sv, newlen);
98653f18 1534#endif
bd81e77b
NC
1535 }
1536 return s;
1537}
aeb18a1e 1538
bd81e77b
NC
1539/*
1540=for apidoc sv_setiv
932e9ff9 1541
bd81e77b
NC
1542Copies an integer into the given SV, upgrading first if necessary.
1543Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1544
bd81e77b
NC
1545=cut
1546*/
463ee0b2 1547
bd81e77b 1548void
aad570aa 1549Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
bd81e77b 1550{
97aff369 1551 dVAR;
7918f24d
NC
1552
1553 PERL_ARGS_ASSERT_SV_SETIV;
1554
bd81e77b
NC
1555 SV_CHECK_THINKFIRST_COW_DROP(sv);
1556 switch (SvTYPE(sv)) {
1557 case SVt_NULL:
bd81e77b 1558 case SVt_NV:
3376de98 1559 sv_upgrade(sv, SVt_IV);
bd81e77b 1560 break;
bd81e77b
NC
1561 case SVt_PV:
1562 sv_upgrade(sv, SVt_PVIV);
1563 break;
463ee0b2 1564
bd81e77b 1565 case SVt_PVGV:
6e592b3a
BM
1566 if (!isGV_with_GP(sv))
1567 break;
bd81e77b
NC
1568 case SVt_PVAV:
1569 case SVt_PVHV:
1570 case SVt_PVCV:
1571 case SVt_PVFM:
1572 case SVt_PVIO:
22e74366 1573 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1574 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1575 OP_DESC(PL_op));
42d0e0b7 1576 default: NOOP;
bd81e77b
NC
1577 }
1578 (void)SvIOK_only(sv); /* validate number */
1579 SvIV_set(sv, i);
1580 SvTAINT(sv);
1581}
932e9ff9 1582
bd81e77b
NC
1583/*
1584=for apidoc sv_setiv_mg
d33b2eba 1585
bd81e77b 1586Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1587
bd81e77b
NC
1588=cut
1589*/
d33b2eba 1590
bd81e77b 1591void
aad570aa 1592Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
bd81e77b 1593{
7918f24d
NC
1594 PERL_ARGS_ASSERT_SV_SETIV_MG;
1595
bd81e77b
NC
1596 sv_setiv(sv,i);
1597 SvSETMAGIC(sv);
1598}
727879eb 1599
bd81e77b
NC
1600/*
1601=for apidoc sv_setuv
d33b2eba 1602
bd81e77b
NC
1603Copies an unsigned integer into the given SV, upgrading first if necessary.
1604Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1605
bd81e77b
NC
1606=cut
1607*/
d33b2eba 1608
bd81e77b 1609void
aad570aa 1610Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
bd81e77b 1611{
7918f24d
NC
1612 PERL_ARGS_ASSERT_SV_SETUV;
1613
013abb9b
NC
1614 /* With the if statement to ensure that integers are stored as IVs whenever
1615 possible:
bd81e77b 1616 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1617
bd81e77b
NC
1618 without
1619 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1620
013abb9b
NC
1621 If you wish to remove the following if statement, so that this routine
1622 (and its callers) always return UVs, please benchmark to see what the
1623 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1624 */
1625 if (u <= (UV)IV_MAX) {
1626 sv_setiv(sv, (IV)u);
1627 return;
1628 }
1629 sv_setiv(sv, 0);
1630 SvIsUV_on(sv);
1631 SvUV_set(sv, u);
1632}
d33b2eba 1633
bd81e77b
NC
1634/*
1635=for apidoc sv_setuv_mg
727879eb 1636
bd81e77b 1637Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1638
bd81e77b
NC
1639=cut
1640*/
5e2fc214 1641
bd81e77b 1642void
aad570aa 1643Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
bd81e77b 1644{
7918f24d
NC
1645 PERL_ARGS_ASSERT_SV_SETUV_MG;
1646
bd81e77b
NC
1647 sv_setuv(sv,u);
1648 SvSETMAGIC(sv);
1649}
5e2fc214 1650
954c1994 1651/*
bd81e77b 1652=for apidoc sv_setnv
954c1994 1653
bd81e77b
NC
1654Copies a double into the given SV, upgrading first if necessary.
1655Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1656
1657=cut
1658*/
1659
63f97190 1660void
aad570aa 1661Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
79072805 1662{
97aff369 1663 dVAR;
7918f24d
NC
1664
1665 PERL_ARGS_ASSERT_SV_SETNV;
1666
bd81e77b
NC
1667 SV_CHECK_THINKFIRST_COW_DROP(sv);
1668 switch (SvTYPE(sv)) {
79072805 1669 case SVt_NULL:
79072805 1670 case SVt_IV:
bd81e77b 1671 sv_upgrade(sv, SVt_NV);
79072805
LW
1672 break;
1673 case SVt_PV:
79072805 1674 case SVt_PVIV:
bd81e77b 1675 sv_upgrade(sv, SVt_PVNV);
79072805 1676 break;
bd4b1eb5 1677
bd4b1eb5 1678 case SVt_PVGV:
6e592b3a
BM
1679 if (!isGV_with_GP(sv))
1680 break;
bd81e77b
NC
1681 case SVt_PVAV:
1682 case SVt_PVHV:
79072805 1683 case SVt_PVCV:
bd81e77b
NC
1684 case SVt_PVFM:
1685 case SVt_PVIO:
22e74366 1686 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1687 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1688 OP_DESC(PL_op));
42d0e0b7 1689 default: NOOP;
2068cd4d 1690 }
bd81e77b
NC
1691 SvNV_set(sv, num);
1692 (void)SvNOK_only(sv); /* validate number */
1693 SvTAINT(sv);
79072805
LW
1694}
1695
645c22ef 1696/*
bd81e77b 1697=for apidoc sv_setnv_mg
645c22ef 1698
bd81e77b 1699Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1700
1701=cut
1702*/
1703
bd81e77b 1704void
aad570aa 1705Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
79072805 1706{
7918f24d
NC
1707 PERL_ARGS_ASSERT_SV_SETNV_MG;
1708
bd81e77b
NC
1709 sv_setnv(sv,num);
1710 SvSETMAGIC(sv);
79072805
LW
1711}
1712
bd81e77b
NC
1713/* Print an "isn't numeric" warning, using a cleaned-up,
1714 * printable version of the offending string
1715 */
954c1994 1716
bd81e77b 1717STATIC void
aad570aa 1718S_not_a_number(pTHX_ SV *const sv)
79072805 1719{
97aff369 1720 dVAR;
bd81e77b
NC
1721 SV *dsv;
1722 char tmpbuf[64];
1723 const char *pv;
94463019 1724
7918f24d
NC
1725 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1726
94463019 1727 if (DO_UTF8(sv)) {
84bafc02 1728 dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1729 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1730 } else {
1731 char *d = tmpbuf;
551405c4 1732 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
94463019
JH
1733 /* each *s can expand to 4 chars + "...\0",
1734 i.e. need room for 8 chars */
ecdeb87c 1735
00b6aa41
AL
1736 const char *s = SvPVX_const(sv);
1737 const char * const end = s + SvCUR(sv);
1738 for ( ; s < end && d < limit; s++ ) {
94463019
JH
1739 int ch = *s & 0xFF;
1740 if (ch & 128 && !isPRINT_LC(ch)) {
1741 *d++ = 'M';
1742 *d++ = '-';
1743 ch &= 127;
1744 }
1745 if (ch == '\n') {
1746 *d++ = '\\';
1747 *d++ = 'n';
1748 }
1749 else if (ch == '\r') {
1750 *d++ = '\\';
1751 *d++ = 'r';
1752 }
1753 else if (ch == '\f') {
1754 *d++ = '\\';
1755 *d++ = 'f';
1756 }
1757 else if (ch == '\\') {
1758 *d++ = '\\';
1759 *d++ = '\\';
1760 }
1761 else if (ch == '\0') {
1762 *d++ = '\\';
1763 *d++ = '0';
1764 }
1765 else if (isPRINT_LC(ch))
1766 *d++ = ch;
1767 else {
1768 *d++ = '^';
1769 *d++ = toCTRL(ch);
1770 }
1771 }
1772 if (s < end) {
1773 *d++ = '.';
1774 *d++ = '.';
1775 *d++ = '.';
1776 }
1777 *d = '\0';
1778 pv = tmpbuf;
a0d0e21e 1779 }
a0d0e21e 1780
533c011a 1781 if (PL_op)
9014280d 1782 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1783 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1784 "Argument \"%s\" isn't numeric in %s", pv,
1785 OP_DESC(PL_op));
a0d0e21e 1786 else
9014280d 1787 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1788 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1789 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1790}
1791
c2988b20
NC
1792/*
1793=for apidoc looks_like_number
1794
645c22ef
DM
1795Test if the content of an SV looks like a number (or is a number).
1796C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1797non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1798ignored.
c2988b20
NC
1799
1800=cut
1801*/
1802
1803I32
aad570aa 1804Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1805{
eb578fdb 1806 const char *sbegin;
c2988b20
NC
1807 STRLEN len;
1808
7918f24d
NC
1809 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1810
f52e41ad
FC
1811 if (SvPOK(sv) || SvPOKp(sv)) {
1812 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1813 }
c2988b20 1814 else
e0ab1c0e 1815 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1816 return grok_number(sbegin, len, NULL);
1817}
25da4f38 1818
19f6321d
NC
1819STATIC bool
1820S_glob_2number(pTHX_ GV * const gv)
180488f8 1821{
7918f24d
NC
1822 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1823
675c862f
AL
1824 /* We know that all GVs stringify to something that is not-a-number,
1825 so no need to test that. */
1826 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1827 {
1828 SV *const buffer = sv_newmortal();
1829 gv_efullname3(buffer, gv, "*");
675c862f 1830 not_a_number(buffer);
8e629ff4 1831 }
675c862f
AL
1832 /* We just want something true to return, so that S_sv_2iuv_common
1833 can tail call us and return true. */
19f6321d 1834 return TRUE;
675c862f
AL
1835}
1836
25da4f38
IZ
1837/* Actually, ISO C leaves conversion of UV to IV undefined, but
1838 until proven guilty, assume that things are not that bad... */
1839
645c22ef
DM
1840/*
1841 NV_PRESERVES_UV:
1842
1843 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1844 an IV (an assumption perl has been based on to date) it becomes necessary
1845 to remove the assumption that the NV always carries enough precision to
1846 recreate the IV whenever needed, and that the NV is the canonical form.
1847 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1848 precision as a side effect of conversion (which would lead to insanity
28e5dec8
JH
1849 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1850 1) to distinguish between IV/UV/NV slots that have cached a valid
1851 conversion where precision was lost and IV/UV/NV slots that have a
1852 valid conversion which has lost no precision
645c22ef 1853 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1854 would lose precision, the precise conversion (or differently
1855 imprecise conversion) is also performed and cached, to prevent
1856 requests for different numeric formats on the same SV causing
1857 lossy conversion chains. (lossless conversion chains are perfectly
1858 acceptable (still))
1859
1860
1861 flags are used:
1862 SvIOKp is true if the IV slot contains a valid value
1863 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1864 SvNOKp is true if the NV slot contains a valid value
1865 SvNOK is true only if the NV value is accurate
1866
1867 so
645c22ef 1868 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1869 IV(or UV) would lose accuracy over a direct conversion from PV to
1870 IV(or UV). If it would, cache both conversions, return NV, but mark
1871 SV as IOK NOKp (ie not NOK).
1872
645c22ef 1873 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1874 NV would lose accuracy over a direct conversion from PV to NV. If it
1875 would, cache both conversions, flag similarly.
1876
1877 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1878 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1879 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1880 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1881 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1882
645c22ef
DM
1883 The benefit of this is that operations such as pp_add know that if
1884 SvIOK is true for both left and right operands, then integer addition
1885 can be used instead of floating point (for cases where the result won't
1886 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1887 loss of precision compared with integer addition.
1888
1889 * making IV and NV equal status should make maths accurate on 64 bit
1890 platforms
1891 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1892 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1893 looking for SvIOK and checking for overflow will not outweigh the
1894 fp to integer speedup)
1895 * will slow down integer operations (callers of SvIV) on "inaccurate"
1896 values, as the change from SvIOK to SvIOKp will cause a call into
1897 sv_2iv each time rather than a macro access direct to the IV slot
1898 * should speed up number->string conversion on integers as IV is
645c22ef 1899 favoured when IV and NV are equally accurate
28e5dec8
JH
1900
1901 ####################################################################
645c22ef
DM
1902 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1903 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1904 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1905 ####################################################################
1906
645c22ef 1907 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1908 performance ratio.
1909*/
1910
1911#ifndef NV_PRESERVES_UV
645c22ef
DM
1912# define IS_NUMBER_UNDERFLOW_IV 1
1913# define IS_NUMBER_UNDERFLOW_UV 2
1914# define IS_NUMBER_IV_AND_UV 2
1915# define IS_NUMBER_OVERFLOW_IV 4
1916# define IS_NUMBER_OVERFLOW_UV 5
1917
1918/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
1919
1920/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1921STATIC int
5de3775c 1922S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
47031da6
NC
1923# ifdef DEBUGGING
1924 , I32 numtype
1925# endif
1926 )
28e5dec8 1927{
97aff369 1928 dVAR;
7918f24d
NC
1929
1930 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1931
3f7c398e 1932 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
1933 if (SvNVX(sv) < (NV)IV_MIN) {
1934 (void)SvIOKp_on(sv);
1935 (void)SvNOK_on(sv);
45977657 1936 SvIV_set(sv, IV_MIN);
28e5dec8
JH
1937 return IS_NUMBER_UNDERFLOW_IV;
1938 }
1939 if (SvNVX(sv) > (NV)UV_MAX) {
1940 (void)SvIOKp_on(sv);
1941 (void)SvNOK_on(sv);
1942 SvIsUV_on(sv);
607fa7f2 1943 SvUV_set(sv, UV_MAX);
28e5dec8
JH
1944 return IS_NUMBER_OVERFLOW_UV;
1945 }
c2988b20
NC
1946 (void)SvIOKp_on(sv);
1947 (void)SvNOK_on(sv);
1948 /* Can't use strtol etc to convert this string. (See truth table in
1949 sv_2iv */
1950 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 1951 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
1952 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1953 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1954 } else {
1955 /* Integer is imprecise. NOK, IOKp */
1956 }
1957 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1958 }
1959 SvIsUV_on(sv);
607fa7f2 1960 SvUV_set(sv, U_V(SvNVX(sv)));
c2988b20
NC
1961 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1962 if (SvUVX(sv) == UV_MAX) {
1963 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1964 possibly be preserved by NV. Hence, it must be overflow.
1965 NOK, IOKp */
1966 return IS_NUMBER_OVERFLOW_UV;
1967 }
1968 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1969 } else {
1970 /* Integer is imprecise. NOK, IOKp */
28e5dec8 1971 }
c2988b20 1972 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 1973}
645c22ef
DM
1974#endif /* !NV_PRESERVES_UV*/
1975
af359546 1976STATIC bool
7918f24d
NC
1977S_sv_2iuv_common(pTHX_ SV *const sv)
1978{
97aff369 1979 dVAR;
7918f24d
NC
1980
1981 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1982
af359546 1983 if (SvNOKp(sv)) {
28e5dec8
JH
1984 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1985 * without also getting a cached IV/UV from it at the same time
1986 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
1987 * IV or UV at same time to avoid this. */
1988 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
1989
1990 if (SvTYPE(sv) == SVt_NV)
1991 sv_upgrade(sv, SVt_PVNV);
1992
28e5dec8
JH
1993 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1994 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1995 certainly cast into the IV range at IV_MAX, whereas the correct
1996 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1997 cases go to UV */
cab190d4
JD
1998#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1999 if (Perl_isnan(SvNVX(sv))) {
2000 SvUV_set(sv, 0);
2001 SvIsUV_on(sv);
fdbe6d7c 2002 return FALSE;
cab190d4 2003 }
cab190d4 2004#endif
28e5dec8 2005 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2006 SvIV_set(sv, I_V(SvNVX(sv)));
28e5dec8
JH
2007 if (SvNVX(sv) == (NV) SvIVX(sv)
2008#ifndef NV_PRESERVES_UV
2009 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2010 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2011 /* Don't flag it as "accurately an integer" if the number
2012 came from a (by definition imprecise) NV operation, and
2013 we're outside the range of NV integer precision */
2014#endif
2015 ) {
a43d94f2
NC
2016 if (SvNOK(sv))
2017 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2018 else {
2019 /* scalar has trailing garbage, eg "42a" */
2020 }
28e5dec8 2021 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2022 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2023 PTR2UV(sv),
2024 SvNVX(sv),
2025 SvIVX(sv)));
2026
2027 } else {
2028 /* IV not precise. No need to convert from PV, as NV
2029 conversion would already have cached IV if it detected
2030 that PV->IV would be better than PV->NV->IV
2031 flags already correct - don't set public IOK. */
2032 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2033 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2034 PTR2UV(sv),
2035 SvNVX(sv),
2036 SvIVX(sv)));
2037 }
2038 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2039 but the cast (NV)IV_MIN rounds to a the value less (more
2040 negative) than IV_MIN which happens to be equal to SvNVX ??
2041 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2042 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2043 (NV)UVX == NVX are both true, but the values differ. :-(
2044 Hopefully for 2s complement IV_MIN is something like
2045 0x8000000000000000 which will be exact. NWC */
d460ef45 2046 }
25da4f38 2047 else {
607fa7f2 2048 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8
JH
2049 if (
2050 (SvNVX(sv) == (NV) SvUVX(sv))
2051#ifndef NV_PRESERVES_UV
2052 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2053 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2054 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2055 /* Don't flag it as "accurately an integer" if the number
2056 came from a (by definition imprecise) NV operation, and
2057 we're outside the range of NV integer precision */
2058#endif
a43d94f2 2059 && SvNOK(sv)
28e5dec8
JH
2060 )
2061 SvIOK_on(sv);
25da4f38 2062 SvIsUV_on(sv);
1c846c1f 2063 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2064 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2065 PTR2UV(sv),
57def98f
JH
2066 SvUVX(sv),
2067 SvUVX(sv)));
25da4f38 2068 }
748a9306 2069 }
cd84013a 2070 else if (SvPOKp(sv)) {
c2988b20 2071 UV value;
504618e9 2072 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2073 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2074 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2075 the same as the direct translation of the initial string
2076 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2077 be careful to ensure that the value with the .456 is around if the
2078 NV value is requested in the future).
1c846c1f 2079
af359546 2080 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2081 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2082 cache the NV if we are sure it's not needed.
25da4f38 2083 */
16b7a9a4 2084
c2988b20
NC
2085 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2086 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2087 == IS_NUMBER_IN_UV) {
5e045b90 2088 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2089 if (SvTYPE(sv) < SVt_PVIV)
2090 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2091 (void)SvIOK_on(sv);
c2988b20
NC
2092 } else if (SvTYPE(sv) < SVt_PVNV)
2093 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2094
f2524eef 2095 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2096 we aren't going to call atof() below. If NVs don't preserve UVs
2097 then the value returned may have more precision than atof() will
2098 return, even though value isn't perfectly accurate. */
2099 if ((numtype & (IS_NUMBER_IN_UV
2100#ifdef NV_PRESERVES_UV
2101 | IS_NUMBER_NOT_INT
2102#endif
2103 )) == IS_NUMBER_IN_UV) {
2104 /* This won't turn off the public IOK flag if it was set above */
2105 (void)SvIOKp_on(sv);
2106
2107 if (!(numtype & IS_NUMBER_NEG)) {
2108 /* positive */;
2109 if (value <= (UV)IV_MAX) {
45977657 2110 SvIV_set(sv, (IV)value);
c2988b20 2111 } else {
af359546 2112 /* it didn't overflow, and it was positive. */
607fa7f2 2113 SvUV_set(sv, value);
c2988b20
NC
2114 SvIsUV_on(sv);
2115 }
2116 } else {
2117 /* 2s complement assumption */
2118 if (value <= (UV)IV_MIN) {
45977657 2119 SvIV_set(sv, -(IV)value);
c2988b20
NC
2120 } else {
2121 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2122 I'm assuming it will be rare. */
c2988b20
NC
2123 if (SvTYPE(sv) < SVt_PVNV)
2124 sv_upgrade(sv, SVt_PVNV);
2125 SvNOK_on(sv);
2126 SvIOK_off(sv);
2127 SvIOKp_on(sv);
9d6ce603 2128 SvNV_set(sv, -(NV)value);
45977657 2129 SvIV_set(sv, IV_MIN);
c2988b20
NC
2130 }
2131 }
2132 }
2133 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2134 will be in the previous block to set the IV slot, and the next
2135 block to set the NV slot. So no else here. */
2136
2137 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2138 != IS_NUMBER_IN_UV) {
2139 /* It wasn't an (integer that doesn't overflow the UV). */
3f7c398e 2140 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8 2141
c2988b20
NC
2142 if (! numtype && ckWARN(WARN_NUMERIC))
2143 not_a_number(sv);
28e5dec8 2144
65202027 2145#if defined(USE_LONG_DOUBLE)
c2988b20
NC
2146 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2147 PTR2UV(sv), SvNVX(sv)));
65202027 2148#else
1779d84d 2149 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
c2988b20 2150 PTR2UV(sv), SvNVX(sv)));
65202027 2151#endif
28e5dec8 2152
28e5dec8 2153#ifdef NV_PRESERVES_UV
af359546
NC
2154 (void)SvIOKp_on(sv);
2155 (void)SvNOK_on(sv);
2156 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2157 SvIV_set(sv, I_V(SvNVX(sv)));
2158 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2159 SvIOK_on(sv);
2160 } else {
6f207bd3 2161 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2162 }
2163 /* UV will not work better than IV */
2164 } else {
2165 if (SvNVX(sv) > (NV)UV_MAX) {
2166 SvIsUV_on(sv);
2167 /* Integer is inaccurate. NOK, IOKp, is UV */
2168 SvUV_set(sv, UV_MAX);
af359546
NC
2169 } else {
2170 SvUV_set(sv, U_V(SvNVX(sv)));
2171 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2172 NV preservse UV so can do correct comparison. */
2173 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2174 SvIOK_on(sv);
af359546 2175 } else {
6f207bd3 2176 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2177 }
2178 }
4b0c9573 2179 SvIsUV_on(sv);
af359546 2180 }
28e5dec8 2181#else /* NV_PRESERVES_UV */
c2988b20
NC
2182 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2183 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2184 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2185 grok_number above. The NV slot has just been set using
2186 Atof. */
560b0c46 2187 SvNOK_on(sv);
c2988b20
NC
2188 assert (SvIOKp(sv));
2189 } else {
2190 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2191 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2192 /* Small enough to preserve all bits. */
2193 (void)SvIOKp_on(sv);
2194 SvNOK_on(sv);
45977657 2195 SvIV_set(sv, I_V(SvNVX(sv)));
c2988b20
NC
2196 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2197 SvIOK_on(sv);
2198 /* Assumption: first non-preserved integer is < IV_MAX,
2199 this NV is in the preserved range, therefore: */
2200 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2201 < (UV)IV_MAX)) {
32fdb065 2202 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
2203 }
2204 } else {
2205 /* IN_UV NOT_INT
2206 0 0 already failed to read UV.
2207 0 1 already failed to read UV.
2208 1 0 you won't get here in this case. IV/UV
2209 slot set, public IOK, Atof() unneeded.
2210 1 1 already read UV.
2211 so there's no point in sv_2iuv_non_preserve() attempting
2212 to use atol, strtol, strtoul etc. */
47031da6 2213# ifdef DEBUGGING
40a17c4c 2214 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2215# else
2216 sv_2iuv_non_preserve (sv);
2217# endif
c2988b20
NC
2218 }
2219 }
28e5dec8 2220#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2221 /* It might be more code efficient to go through the entire logic above
2222 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2223 gets complex and potentially buggy, so more programmer efficient
2224 to do it this way, by turning off the public flags: */
2225 if (!numtype)
2226 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2227 }
af359546
NC
2228 }
2229 else {
675c862f 2230 if (isGV_with_GP(sv))
159b6efe 2231 return glob_2number(MUTABLE_GV(sv));
180488f8 2232
9a214eec 2233 if (!SvPADTMP(sv)) {
af359546
NC
2234 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2235 report_uninit(sv);
2236 }
25da4f38
IZ
2237 if (SvTYPE(sv) < SVt_IV)
2238 /* Typically the caller expects that sv_any is not NULL now. */
2239 sv_upgrade(sv, SVt_IV);
af359546
NC
2240 /* Return 0 from the caller. */
2241 return TRUE;
2242 }
2243 return FALSE;
2244}
2245
2246/*
2247=for apidoc sv_2iv_flags
2248
2249Return the integer value of an SV, doing any necessary string
2250conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2251Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2252
2253=cut
2254*/
2255
2256IV
5de3775c 2257Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
af359546 2258{
97aff369 2259 dVAR;
4bac9ae4 2260
af359546 2261 if (!sv)
a0d0e21e 2262 return 0;
4bac9ae4
CS
2263
2264 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2265 mg_get(sv);
2266
2267 if (SvROK(sv)) {
2268 if (SvAMAGIC(sv)) {
2269 SV * tmpstr;
2270 if (flags & SV_SKIP_OVERLOAD)
2271 return 0;
2272 tmpstr = AMG_CALLunary(sv, numer_amg);
2273 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2274 return SvIV(tmpstr);
2275 }
2276 }
2277 return PTR2IV(SvRV(sv));
2278 }
2279
8d919b0a 2280 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2281 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2282 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2283 In practice they are extremely unlikely to actually get anywhere
2284 accessible by user Perl code - the only way that I'm aware of is when
2285 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2286
2287 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2288 */
8d919b0a 2289 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2290 {
71c558c3 2291 UV value;
8d919b0a
FC
2292 const char * const ptr =
2293 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2294 const int numtype
8d919b0a 2295 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2296
2297 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2298 == IS_NUMBER_IN_UV) {
2299 /* It's definitely an integer */
2300 if (numtype & IS_NUMBER_NEG) {
2301 if (value < (UV)IV_MIN)
2302 return -(IV)value;
2303 } else {
2304 if (value < (UV)IV_MAX)
2305 return (IV)value;
2306 }
2307 }
2308 if (!numtype) {
2309 if (ckWARN(WARN_NUMERIC))
2310 not_a_number(sv);
2311 }
8d919b0a 2312 return I_V(Atof(ptr));
e20b6c3b 2313 }
4bac9ae4
CS
2314 }
2315
2316 if (SvTHINKFIRST(sv)) {
73381561 2317#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2318 if (SvIsCOW(sv)) {
2319 sv_force_normal_flags(sv, 0);
2320 }
73381561 2321#endif
af359546
NC
2322 if (SvREADONLY(sv) && !SvOK(sv)) {
2323 if (ckWARN(WARN_UNINITIALIZED))
2324 report_uninit(sv);
2325 return 0;
2326 }
2327 }
4bac9ae4 2328
af359546
NC
2329 if (!SvIOKp(sv)) {
2330 if (S_sv_2iuv_common(aTHX_ sv))
2331 return 0;
79072805 2332 }
4bac9ae4 2333
1d7c1841
GS
2334 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2335 PTR2UV(sv),SvIVX(sv)));
25da4f38 2336 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2337}
2338
645c22ef 2339/*
891f9566 2340=for apidoc sv_2uv_flags
645c22ef
DM
2341
2342Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2343conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2344Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2345
2346=cut
2347*/
2348
ff68c719 2349UV
5de3775c 2350Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
ff68c719 2351{
97aff369 2352 dVAR;
4bac9ae4 2353
ff68c719
PP
2354 if (!sv)
2355 return 0;
4bac9ae4
CS
2356
2357 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2358 mg_get(sv);
2359
2360 if (SvROK(sv)) {
2361 if (SvAMAGIC(sv)) {
2362 SV *tmpstr;
2363 if (flags & SV_SKIP_OVERLOAD)
2364 return 0;
2365 tmpstr = AMG_CALLunary(sv, numer_amg);
2366 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2367 return SvUV(tmpstr);
2368 }
2369 }
2370 return PTR2UV(SvRV(sv));
2371 }
2372
8d919b0a 2373 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2374 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2375 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2376 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2377 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2378 {
71c558c3 2379 UV value;
8d919b0a
FC
2380 const char * const ptr =
2381 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
71c558c3 2382 const int numtype
8d919b0a 2383 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2384
2385 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2386 == IS_NUMBER_IN_UV) {
2387 /* It's definitely an integer */
2388 if (!(numtype & IS_NUMBER_NEG))
2389 return value;
2390 }
2391 if (!numtype) {
2392 if (ckWARN(WARN_NUMERIC))
2393 not_a_number(sv);
2394 }
8d919b0a 2395 return U_V(Atof(ptr));
e20b6c3b 2396 }
4bac9ae4
CS
2397 }
2398
2399 if (SvTHINKFIRST(sv)) {
73381561 2400#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2401 if (SvIsCOW(sv)) {
2402 sv_force_normal_flags(sv, 0);
8a818333 2403 }
73381561 2404#endif
0336b60e 2405 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2406 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2407 report_uninit(sv);
ff68c719
PP
2408 return 0;
2409 }
2410 }
4bac9ae4 2411
af359546
NC
2412 if (!SvIOKp(sv)) {
2413 if (S_sv_2iuv_common(aTHX_ sv))
2414 return 0;
ff68c719 2415 }
25da4f38 2416
1d7c1841
GS
2417 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2418 PTR2UV(sv),SvUVX(sv)));
25da4f38 2419 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2420}
2421
645c22ef 2422/*
196007d1 2423=for apidoc sv_2nv_flags
645c22ef
DM
2424
2425Return the num value of an SV, doing any necessary string or integer
fde67290 2426conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2427Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2428
2429=cut
2430*/
2431
65202027 2432NV
39d5de13 2433Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
79072805 2434{
97aff369 2435 dVAR;
79072805
LW
2436 if (!sv)
2437 return 0.0;
8d919b0a 2438 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2439 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2440 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2441 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2442 const char *ptr;
39d5de13
DM
2443 if (flags & SV_GMAGIC)
2444 mg_get(sv);
463ee0b2
LW
2445 if (SvNOKp(sv))
2446 return SvNVX(sv);
cd84013a 2447 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2448 ptr = SvPVX_const(sv);
2449 grokpv:
041457d9 2450 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2451 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2452 not_a_number(sv);
8d919b0a 2453 return Atof(ptr);
a0d0e21e 2454 }
25da4f38 2455 if (SvIOKp(sv)) {
1c846c1f 2456 if (SvIsUV(sv))
65202027 2457 return (NV)SvUVX(sv);
25da4f38 2458 else
65202027 2459 return (NV)SvIVX(sv);
47a72cb8
NC
2460 }
2461 if (SvROK(sv)) {
2462 goto return_rok;
2463 }
8d919b0a
FC
2464 if (isREGEXP(sv)) {
2465 ptr = RX_WRAPPED((REGEXP *)sv);
2466 goto grokpv;
2467 }
47a72cb8
NC
2468 assert(SvTYPE(sv) >= SVt_PVMG);
2469 /* This falls through to the report_uninit near the end of the
2470 function. */
2471 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2472 if (SvROK(sv)) {
47a72cb8 2473 return_rok:
deb46114 2474 if (SvAMAGIC(sv)) {
aee036bb
DM
2475 SV *tmpstr;
2476 if (flags & SV_SKIP_OVERLOAD)
2477 return 0;
31d632c3 2478 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2479 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2480 return SvNV(tmpstr);
2481 }
2482 }
2483 return PTR2NV(SvRV(sv));
a0d0e21e 2484 }
73381561 2485#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2486 if (SvIsCOW(sv)) {
2487 sv_force_normal_flags(sv, 0);
8a818333 2488 }
73381561 2489#endif
0336b60e 2490 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2491 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2492 report_uninit(sv);
ed6116ce
LW
2493 return 0.0;
2494 }
79072805
LW
2495 }
2496 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2497 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2498 sv_upgrade(sv, SVt_NV);
906f284f 2499#ifdef USE_LONG_DOUBLE
097ee67d 2500 DEBUG_c({
f93f4e46 2501 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2502 PerlIO_printf(Perl_debug_log,
2503 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2504 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2505 RESTORE_NUMERIC_LOCAL();
2506 });
65202027 2507#else
572bbb43 2508 DEBUG_c({
f93f4e46 2509 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2510 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
1d7c1841 2511 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2512 RESTORE_NUMERIC_LOCAL();
2513 });
572bbb43 2514#endif
79072805
LW
2515 }
2516 else if (SvTYPE(sv) < SVt_PVNV)
2517 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2518 if (SvNOKp(sv)) {
2519 return SvNVX(sv);
61604483 2520 }
59d8ce62 2521 if (SvIOKp(sv)) {
9d6ce603 2522 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2523#ifdef NV_PRESERVES_UV
a43d94f2
NC
2524 if (SvIOK(sv))
2525 SvNOK_on(sv);
2526 else
2527 SvNOKp_on(sv);
28e5dec8
JH
2528#else
2529 /* Only set the public NV OK flag if this NV preserves the IV */
2530 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2531 if (SvIOK(sv) &&
2532 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2533 : (SvIVX(sv) == I_V(SvNVX(sv))))
2534 SvNOK_on(sv);
2535 else
2536 SvNOKp_on(sv);
2537#endif
93a17b20 2538 }
cd84013a 2539 else if (SvPOKp(sv)) {
c2988b20 2540 UV value;
3f7c398e 2541 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2542 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2543 not_a_number(sv);
28e5dec8 2544#ifdef NV_PRESERVES_UV
c2988b20
NC
2545 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2546 == IS_NUMBER_IN_UV) {
5e045b90 2547 /* It's definitely an integer */
9d6ce603 2548 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
c2988b20 2549 } else
3f7c398e 2550 SvNV_set(sv, Atof(SvPVX_const(sv)));
a43d94f2
NC
2551 if (numtype)
2552 SvNOK_on(sv);
2553 else
2554 SvNOKp_on(sv);
28e5dec8 2555#else
3f7c398e 2556 SvNV_set(sv, Atof(SvPVX_const(sv)));
28e5dec8
JH
2557 /* Only set the public NV OK flag if this NV preserves the value in
2558 the PV at least as well as an IV/UV would.
2559 Not sure how to do this 100% reliably. */
2560 /* if that shift count is out of range then Configure's test is
2561 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2562 UV_BITS */
2563 if (((UV)1 << NV_PRESERVES_UV_BITS) >
c2988b20 2564 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
28e5dec8 2565 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
c2988b20
NC
2566 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2567 /* Can't use strtol etc to convert this string, so don't try.
2568 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2569 SvNOK_on(sv);
2570 } else {
2571 /* value has been set. It may not be precise. */
2572 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2573 /* 2s complement assumption for (UV)IV_MIN */
2574 SvNOK_on(sv); /* Integer is too negative. */
2575 } else {
2576 SvNOKp_on(sv);
2577 SvIOKp_on(sv);
6fa402ec 2578
c2988b20 2579 if (numtype & IS_NUMBER_NEG) {
45977657 2580 SvIV_set(sv, -(IV)value);
c2988b20 2581 } else if (value <= (UV)IV_MAX) {
45977657 2582 SvIV_set(sv, (IV)value);
c2988b20 2583 } else {
607fa7f2 2584 SvUV_set(sv, value);
c2988b20
NC
2585 SvIsUV_on(sv);
2586 }
2587
2588 if (numtype & IS_NUMBER_NOT_INT) {
2589 /* I believe that even if the original PV had decimals,
2590 they are lost beyond the limit of the FP precision.
2591 However, neither is canonical, so both only get p
2592 flags. NWC, 2000/11/25 */
2593 /* Both already have p flags, so do nothing */
2594 } else {
66a1b24b 2595 const NV nv = SvNVX(sv);
c2988b20
NC
2596 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2597 if (SvIVX(sv) == I_V(nv)) {
2598 SvNOK_on(sv);
c2988b20 2599 } else {
c2988b20
NC
2600 /* It had no "." so it must be integer. */
2601 }
00b6aa41 2602 SvIOK_on(sv);
c2988b20
NC
2603 } else {
2604 /* between IV_MAX and NV(UV_MAX).
2605 Could be slightly > UV_MAX */
6fa402ec 2606
c2988b20
NC
2607 if (numtype & IS_NUMBER_NOT_INT) {
2608 /* UV and NV both imprecise. */
2609 } else {
66a1b24b 2610 const UV nv_as_uv = U_V(nv);
c2988b20
NC
2611
2612 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2613 SvNOK_on(sv);
c2988b20 2614 }
00b6aa41 2615 SvIOK_on(sv);
c2988b20
NC
2616 }
2617 }
2618 }
2619 }
2620 }
a43d94f2
NC
2621 /* It might be more code efficient to go through the entire logic above
2622 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2623 gets complex and potentially buggy, so more programmer efficient
2624 to do it this way, by turning off the public flags: */
2625 if (!numtype)
2626 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2627#endif /* NV_PRESERVES_UV */
93a17b20 2628 }
79072805 2629 else {
f7877b28 2630 if (isGV_with_GP(sv)) {
159b6efe 2631 glob_2number(MUTABLE_GV(sv));
180488f8
NC
2632 return 0.0;
2633 }
2634
9a214eec 2635 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
29489e7c 2636 report_uninit(sv);
7e25a7e9
NC
2637 assert (SvTYPE(sv) >= SVt_NV);
2638 /* Typically the caller expects that sv_any is not NULL now. */
2639 /* XXX Ilya implies that this is a bug in callers that assume this
2640 and ideally should be fixed. */
a0d0e21e 2641 return 0.0;
79072805 2642 }
572bbb43 2643#if defined(USE_LONG_DOUBLE)
097ee67d 2644 DEBUG_c({
f93f4e46 2645 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841
GS
2646 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2647 PTR2UV(sv), SvNVX(sv));
572bbb43
GS
2648 RESTORE_NUMERIC_LOCAL();
2649 });
65202027 2650#else
572bbb43 2651 DEBUG_c({
f93f4e46 2652 STORE_NUMERIC_LOCAL_SET_STANDARD();
1779d84d 2653 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
1d7c1841 2654 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2655 RESTORE_NUMERIC_LOCAL();
2656 });
572bbb43 2657#endif
463ee0b2 2658 return SvNVX(sv);
79072805
LW
2659}
2660
800401ee
JH
2661/*
2662=for apidoc sv_2num
2663
2664Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2665reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2666access this function.
800401ee
JH
2667
2668=cut
2669*/
2670
2671SV *
5de3775c 2672Perl_sv_2num(pTHX_ register SV *const sv)
800401ee 2673{
7918f24d
NC
2674 PERL_ARGS_ASSERT_SV_2NUM;
2675
b9ee0594
RGS
2676 if (!SvROK(sv))
2677 return sv;
800401ee 2678 if (SvAMAGIC(sv)) {
31d632c3 2679 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2680 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2681 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2682 return sv_2num(tmpsv);
2683 }
2684 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2685}
2686
645c22ef
DM
2687/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2688 * UV as a string towards the end of buf, and return pointers to start and
2689 * end of it.
2690 *
2691 * We assume that buf is at least TYPE_CHARS(UV) long.
2692 */
2693
864dbfa3 2694static char *
5de3775c 2695S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2696{
25da4f38 2697 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2698 char * const ebuf = ptr;
25da4f38 2699 int sign;
25da4f38 2700
7918f24d
NC
2701 PERL_ARGS_ASSERT_UIV_2BUF;
2702
25da4f38
IZ
2703 if (is_uv)
2704 sign = 0;
2705 else if (iv >= 0) {
2706 uv = iv;
2707 sign = 0;
2708 } else {
2709 uv = -iv;
2710 sign = 1;
2711 }
2712 do {
eb160463 2713 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2714 } while (uv /= 10);
2715 if (sign)
2716 *--ptr = '-';
2717 *peob = ebuf;
2718 return ptr;
2719}
2720
645c22ef
DM
2721/*
2722=for apidoc sv_2pv_flags
2723
ff276b08 2724Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2725If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2726string if necessary. Normally invoked via the C<SvPV_flags> macro.
2727C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2728
2729=cut
2730*/
2731
8d6d96c1 2732char *
5de3775c 2733Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2734{
97aff369 2735 dVAR;
eb578fdb 2736 char *s;
79072805 2737
463ee0b2 2738 if (!sv) {
cdb061a3
NC
2739 if (lp)
2740 *lp = 0;
73d840c0 2741 return (char *)"";
463ee0b2 2742 }
4bac9ae4
CS
2743 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2744 mg_get(sv);
2745 if (SvROK(sv)) {
2746 if (SvAMAGIC(sv)) {
2747 SV *tmpstr;
2748 if (flags & SV_SKIP_OVERLOAD)
2749 return NULL;
2750 tmpstr = AMG_CALLunary(sv, string_amg);
2751 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2752 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2753 /* Unwrap this: */
2754 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2755 */
75dfc8ec 2756
4bac9ae4
CS
2757 char *pv;
2758 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2759 if (flags & SV_CONST_RETURN) {
2760 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2761 } else {
4bac9ae4
CS
2762 pv = (flags & SV_MUTABLE_RETURN)
2763 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2764 }
4bac9ae4
CS
2765 if (lp)
2766 *lp = SvCUR(tmpstr);
2767 } else {
2768 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2769 }
4bac9ae4
CS
2770 if (SvUTF8(tmpstr))
2771 SvUTF8_on(sv);
2772 else
2773 SvUTF8_off(sv);
2774 return pv;
deb46114 2775 }
4bac9ae4
CS
2776 }
2777 {
2778 STRLEN len;
2779 char *retval;
2780 char *buffer;
2781 SV *const referent = SvRV(sv);
2782
2783 if (!referent) {
2784 len = 7;
2785 retval = buffer = savepvn("NULLREF", len);
2786 } else if (SvTYPE(referent) == SVt_REGEXP &&
2787 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2788 amagic_is_enabled(string_amg))) {
2789 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2790
2791 assert(re);
67d2d14d 2792
4bac9ae4
CS
2793 /* If the regex is UTF-8 we want the containing scalar to
2794 have an UTF-8 flag too */
2795 if (RX_UTF8(re))
2796 SvUTF8_on(sv);
2797 else
2798 SvUTF8_off(sv);
67d2d14d 2799
4bac9ae4
CS
2800 if (lp)
2801 *lp = RX_WRAPLEN(re);
67d2d14d 2802
4bac9ae4
CS
2803 return RX_WRAPPED(re);
2804 } else {
2805 const char *const typestr = sv_reftype(referent, 0);
2806 const STRLEN typelen = strlen(typestr);
2807 UV addr = PTR2UV(referent);
2808 const char *stashname = NULL;
2809 STRLEN stashnamelen = 0; /* hush, gcc */
2810 const char *buffer_end;
2811
2812 if (SvOBJECT(referent)) {
2813 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2814
2815 if (name) {
2816 stashname = HEK_KEY(name);
2817 stashnamelen = HEK_LEN(name);
2818
2819 if (HEK_UTF8(name)) {
2820 SvUTF8_on(sv);
fafee734 2821 } else {
4bac9ae4 2822 SvUTF8_off(sv);
fafee734 2823 }
fafee734 2824 } else {
4bac9ae4
CS
2825 stashname = "__ANON__";
2826 stashnamelen = 8;
fafee734 2827 }
4bac9ae4
CS
2828 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2829 + 2 * sizeof(UV) + 2 /* )\0 */;
2830 } else {
2831 len = typelen + 3 /* (0x */
2832 + 2 * sizeof(UV) + 2 /* )\0 */;
2833 }
fafee734 2834
4bac9ae4
CS
2835 Newx(buffer, len, char);
2836 buffer_end = retval = buffer + len;
2837
2838 /* Working backwards */
2839 *--retval = '\0';
2840 *--retval = ')';
2841 do {
2842 *--retval = PL_hexdigit[addr & 15];
2843 } while (addr >>= 4);
2844 *--retval = 'x';
2845 *--retval = '0';
2846 *--retval = '(';
2847
2848 retval -= typelen;
2849 memcpy(retval, typestr, typelen);
2850
2851 if (stashname) {
2852 *--retval = '=';
2853 retval -= stashnamelen;
2854 memcpy(retval, stashname, stashnamelen);
c080367d 2855 }
4bac9ae4
CS
2856 /* retval may not necessarily have reached the start of the
2857 buffer here. */
2858 assert (retval >= buffer);
2859
2860 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 2861 }
cdb061a3 2862 if (lp)
4bac9ae4
CS
2863 *lp = len;
2864 SAVEFREEPV(buffer);
2865 return retval;
79072805 2866 }
79072805 2867 }
4bac9ae4
CS
2868
2869 if (SvPOKp(sv)) {
2870 if (lp)
2871 *lp = SvCUR(sv);
2872 if (flags & SV_MUTABLE_RETURN)
2873 return SvPVX_mutable(sv);
2874 if (flags & SV_CONST_RETURN)
2875 return (char *)SvPVX_const(sv);
2876 return SvPVX(sv);
2877 }
2878
2879 if (SvIOK(sv)) {
28e5dec8
JH
2880 /* I'm assuming that if both IV and NV are equally valid then
2881 converting the IV is going to be more efficient */
e1ec3a88 2882 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
2883 char buf[TYPE_CHARS(UV)];
2884 char *ebuf, *ptr;
97a130b8 2885 STRLEN len;
28e5dec8
JH
2886
2887 if (SvTYPE(sv) < SVt_PVIV)
2888 sv_upgrade(sv, SVt_PVIV);
4ea1d550 2889 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 2890 len = ebuf - ptr;
5902b6a9 2891 /* inlined from sv_setpvn */
97a130b8
NC
2892 s = SvGROW_mutable(sv, len + 1);
2893 Move(ptr, s, len, char);
2894 s += len;
28e5dec8 2895 *s = '\0';
28e5dec8 2896 }
4bac9ae4 2897 else if (SvNOK(sv)) {
79072805
LW
2898 if (SvTYPE(sv) < SVt_PVNV)
2899 sv_upgrade(sv, SVt_PVNV);
29912d93
NC
2900 if (SvNVX(sv) == 0.0) {
2901 s = SvGROW_mutable(sv, 2);
2902 *s++ = '0';
2903 *s = '\0';
2904 } else {
2905 dSAVE_ERRNO;
2906 /* The +20 is pure guesswork. Configure test needed. --jhi */
2907 s = SvGROW_mutable(sv, NV_DIG + 20);
2908 /* some Xenix systems wipe out errno here */
2d4389e4 2909 Gconvert(SvNVX(sv), NV_DIG, 0, s);
29912d93
NC
2910 RESTORE_ERRNO;
2911 while (*s) s++;
bbce6d69 2912 }
79072805
LW
2913#ifdef hcx
2914 if (s[-1] == '.')
46fc3d4c 2915 *--s = '\0';
79072805
LW
2916#endif
2917 }
4bac9ae4
CS
2918 else if (isGV_with_GP(sv)) {
2919 GV *const gv = MUTABLE_GV(sv);
2920 SV *const buffer = sv_newmortal();
8d1c3e26 2921
4bac9ae4 2922 gv_efullname3(buffer, gv, "*");
180488f8 2923
4bac9ae4
CS
2924 assert(SvPOK(buffer));
2925 if (SvUTF8(buffer))
2926 SvUTF8_on(sv);
2927 if (lp)
2928 *lp = SvCUR(buffer);
2929 return SvPVX(buffer);
2930 }
8d919b0a
FC
2931 else if (isREGEXP(sv)) {
2932 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2933 return RX_WRAPPED((REGEXP *)sv);
2934 }
4bac9ae4 2935 else {
cdb061a3 2936 if (lp)
00b6aa41 2937 *lp = 0;
9f621bb0
NC
2938 if (flags & SV_UNDEF_RETURNS_NULL)
2939 return NULL;
9a214eec 2940 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
9f621bb0 2941 report_uninit(sv);
4bac9ae4
CS
2942 /* Typically the caller expects that sv_any is not NULL now. */
2943 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 2944 sv_upgrade(sv, SVt_PV);
73d840c0 2945 return (char *)"";
79072805 2946 }
4bac9ae4 2947
cdb061a3 2948 {
823a54a3 2949 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
2950 if (lp)
2951 *lp = len;
2952 SvCUR_set(sv, len);
2953 }
79072805 2954 SvPOK_on(sv);
1d7c1841 2955 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 2956 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
2957 if (flags & SV_CONST_RETURN)
2958 return (char *)SvPVX_const(sv);
10516c54
NC
2959 if (flags & SV_MUTABLE_RETURN)
2960 return SvPVX_mutable(sv);
463ee0b2
LW
2961 return SvPVX(sv);
2962}
2963
645c22ef 2964/*
6050d10e
JP
2965=for apidoc sv_copypv
2966
2967Copies a stringified representation of the source SV into the
2968destination SV. Automatically performs any necessary mg_get and
54f0641b 2969coercion of numeric values into strings. Guaranteed to preserve
2575c402 2970UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
2971sv_2pv[_flags] but operates directly on an SV instead of just the
2972string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
2973would lose the UTF-8'ness of the PV.
2974
4bac9ae4
CS
2975=for apidoc sv_copypv_nomg
2976
2977Like sv_copypv, but doesn't invoke get magic first.
2978
2979=for apidoc sv_copypv_flags
2980
2981Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
2982include SV_GMAGIC.
2983
6050d10e
JP
2984=cut
2985*/
2986
2987void
5de3775c 2988Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
6050d10e 2989{
4bac9ae4
CS
2990 PERL_ARGS_ASSERT_SV_COPYPV;
2991
2992 sv_copypv_flags(dsv, ssv, 0);
2993}
2994
2995void
2996Perl_sv_copypv_flags(pTHX_ SV *const dsv, register SV *const ssv, const I32 flags)
2997{
446eaa42 2998 STRLEN len;
4bac9ae4 2999 const char *s;
7918f24d 3000
4bac9ae4 3001 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3002
4bac9ae4
CS
3003 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3004 mg_get(ssv);
3005 s = SvPV_nomg_const(ssv,len);
cb50f42d 3006 sv_setpvn(dsv,s,len);
446eaa42 3007 if (SvUTF8(ssv))
cb50f42d 3008 SvUTF8_on(dsv);
446eaa42 3009 else
cb50f42d 3010 SvUTF8_off(dsv);
6050d10e
JP
3011}
3012
3013/*
645c22ef
DM
3014=for apidoc sv_2pvbyte
3015
3016Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3017to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3018side-effect.
3019
3020Usually accessed via the C<SvPVbyte> macro.
3021
3022=cut
3023*/
3024
7340a771 3025char *
a901b181 3026Perl_sv_2pvbyte(pTHX_ register SV *sv, STRLEN *const lp)
7340a771 3027{
7918f24d
NC
3028 PERL_ARGS_ASSERT_SV_2PVBYTE;
3029
4499db73
FC
3030 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3031 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181
FC
3032 SV *sv2 = sv_newmortal();
3033 sv_copypv(sv2,sv);
3034 sv = sv2;
3035 }
3036 else SvGETMAGIC(sv);
0875d2fe 3037 sv_utf8_downgrade(sv,0);
71eb6d8c 3038 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3039}
3040
645c22ef 3041/*
035cbb0e
RGS
3042=for apidoc sv_2pvutf8
3043
3044Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3045to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3046
3047Usually accessed via the C<SvPVutf8> macro.
3048
3049=cut
3050*/
645c22ef 3051
7340a771 3052char *
fe46cbda 3053Perl_sv_2pvutf8(pTHX_ register SV *sv, STRLEN *const lp)
7340a771 3054{
7918f24d
NC
3055 PERL_ARGS_ASSERT_SV_2PVUTF8;
3056
4499db73
FC
3057 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3058 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3059 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3060 else
3061 SvGETMAGIC(sv);
3062 sv_utf8_upgrade_nomg(sv);
c3ec315f 3063 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3064}
1c846c1f 3065
7ee2227d 3066
645c22ef
DM
3067/*
3068=for apidoc sv_2bool
3069
06c841cf
FC
3070This macro is only used by sv_true() or its macro equivalent, and only if
3071the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3072It calls sv_2bool_flags with the SV_GMAGIC flag.
3073
3074=for apidoc sv_2bool_flags
3075
3076This function is only used by sv_true() and friends, and only if
fde67290 3077the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3078contain SV_GMAGIC, then it does an mg_get() first.
3079
645c22ef
DM
3080
3081=cut
3082*/
3083
463ee0b2 3084bool
06c841cf 3085Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
463ee0b2 3086{
97aff369 3087 dVAR;
7918f24d 3088
06c841cf 3089 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3090
06c841cf 3091 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3092
a0d0e21e
LW
3093 if (!SvOK(sv))
3094 return 0;
3095 if (SvROK(sv)) {
fabdb6c0 3096 if (SvAMAGIC(sv)) {
31d632c3 3097 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
fabdb6c0 3098 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
f2338a2e 3099 return cBOOL(SvTRUE(tmpsv));
fabdb6c0
AL
3100 }
3101 return SvRV(sv) != 0;
a0d0e21e 3102 }
4bac9ae4 3103 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3104}
3105
c461cf8f
JH
3106/*
3107=for apidoc sv_utf8_upgrade
3108
78ea37eb 3109Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3110Forces the SV to string form if it is not already.
2bbc8d55 3111Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3112Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3113if the whole string is the same in UTF-8 as not.
3114Returns the number of bytes in the converted string
c461cf8f 3115
0efd0472 3116This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3117use the Encode extension for that.
3118
fe749c9a
KW
3119=for apidoc sv_utf8_upgrade_nomg
3120
fde67290 3121Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3122
8d6d96c1
HS
3123=for apidoc sv_utf8_upgrade_flags
3124
78ea37eb 3125Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3126Forces the SV to string form if it is not already.
8d6d96c1 3127Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3128if all the bytes are invariant in UTF-8.
3129If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55
SP
3130will C<mg_get> on C<sv> if appropriate, else not.
3131Returns the number of bytes in the converted string
3132C<sv_utf8_upgrade> and
8d6d96c1
HS
3133C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3134
0efd0472 3135This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3136use the Encode extension for that.
3137
8d6d96c1 3138=cut
b3ab6785
KW
3139
3140The grow version is currently not externally documented. It adds a parameter,
3141extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3142have free after it upon return. This allows the caller to reserve extra space
3143that it intends to fill, to avoid extra grows.
3144
3145Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3146which can be used to tell this function to not first check to see if there are
3147any characters that are different in UTF-8 (variant characters) which would
3148force it to allocate a new string to sv, but to assume there are. Typically
3149this flag is used by a routine that has already parsed the string to find that
3150there are such characters, and passes this information on so that the work
3151doesn't have to be repeated.
3152
3153(One might think that the calling routine could pass in the position of the
3154first such variant, so it wouldn't have to be found again. But that is not the
3155case, because typically when the caller is likely to use this flag, it won't be
3156calling this routine unless it finds something that won't fit into a byte.
3157Otherwise it tries to not upgrade and just use bytes. But some things that
3158do fit into a byte are variants in utf8, and the caller may not have been
3159keeping track of these.)
3160
3161If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3162isn't guaranteed due to having other routines do the work in some input cases,
3163or if the input is already flagged as being in utf8.
3164
3165The speed of this could perhaps be improved for many cases if someone wanted to
3166write a fast function that counts the number of variant characters in a string,
3167especially if it could return the position of the first one.
3168
8d6d96c1
HS
3169*/
3170
3171STRLEN
b3ab6785 3172Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3173{
97aff369 3174 dVAR;
7918f24d 3175
b3ab6785 3176 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3177
808c356f
RGS
3178 if (sv == &PL_sv_undef)
3179 return 0;
892f9127 3180 if (!SvPOK_nog(sv)) {
e0e62c2a 3181 STRLEN len = 0;
d52b7888
NC
3182 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3183 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3184 if (SvUTF8(sv)) {
3185 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3186 return len;
b3ab6785 3187 }
d52b7888 3188 } else {
33fb6f35 3189 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3190 }
e0e62c2a 3191 }
4411f3b6 3192
f5cee72b 3193 if (SvUTF8(sv)) {
b3ab6785 3194 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3195 return SvCUR(sv);
f5cee72b 3196 }
5fec3b1d 3197
765f542d
NC
3198 if (SvIsCOW(sv)) {
3199 sv_force_normal_flags(sv, 0);
db42d148
NIS
3200 }
3201
b3ab6785 3202 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3203 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3204 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3205 return SvCUR(sv);
3206 }
3207
4e93345f
KW
3208 if (SvCUR(sv) == 0) {
3209 if (extra) SvGROW(sv, extra);
3210 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3211 /* This function could be much more efficient if we
2bbc8d55 3212 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3213 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3214 * make the loop as fast as possible (although there are certainly ways
3215 * to speed this up, eg. through vectorization) */
3216 U8 * s = (U8 *) SvPVX_const(sv);
3217 U8 * e = (U8 *) SvEND(sv);
3218 U8 *t = s;
3219 STRLEN two_byte_count = 0;
c4e7c712 3220
b3ab6785
KW
3221 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3222
3223 /* See if really will need to convert to utf8. We mustn't rely on our
3224 * incoming SV being well formed and having a trailing '\0', as certain
3225 * code in pp_formline can send us partially built SVs. */
3226
c4e7c712 3227 while (t < e) {
53c1dcc0 3228 const U8 ch = *t++;
b3ab6785
KW
3229 if (NATIVE_IS_INVARIANT(ch)) continue;
3230
3231 t--; /* t already incremented; re-point to first variant */
3232 two_byte_count = 1;
3233 goto must_be_utf8;
c4e7c712 3234 }
b3ab6785
KW
3235
3236 /* utf8 conversion not needed because all are invariants. Mark as
3237 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3238 SvUTF8_on(sv);
7f0bfbea 3239 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3240 return SvCUR(sv);
3241
3242must_be_utf8:
3243
3244 /* Here, the string should be converted to utf8, either because of an
3245 * input flag (two_byte_count = 0), or because a character that
3246 * requires 2 bytes was found (two_byte_count = 1). t points either to
3247 * the beginning of the string (if we didn't examine anything), or to
3248 * the first variant. In either case, everything from s to t - 1 will
3249 * occupy only 1 byte each on output.
3250 *
3251 * There are two main ways to convert. One is to create a new string
3252 * and go through the input starting from the beginning, appending each
3253 * converted value onto the new string as we go along. It's probably
3254 * best to allocate enough space in the string for the worst possible
3255 * case rather than possibly running out of space and having to
3256 * reallocate and then copy what we've done so far. Since everything
3257 * from s to t - 1 is invariant, the destination can be initialized
3258 * with these using a fast memory copy
3259 *
3260 * The other way is to figure out exactly how big the string should be
3261 * by parsing the entire input. Then you don't have to make it big
3262 * enough to handle the worst possible case, and more importantly, if
3263 * the string you already have is large enough, you don't have to
3264 * allocate a new string, you can copy the last character in the input
3265 * string to the final position(s) that will be occupied by the
3266 * converted string and go backwards, stopping at t, since everything
3267 * before that is invariant.
3268 *
3269 * There are advantages and disadvantages to each method.
3270 *
3271 * In the first method, we can allocate a new string, do the memory
3272 * copy from the s to t - 1, and then proceed through the rest of the
3273 * string byte-by-byte.
3274 *
3275 * In the second method, we proceed through the rest of the input
3276 * string just calculating how big the converted string will be. Then
3277 * there are two cases:
3278 * 1) if the string has enough extra space to handle the converted
3279 * value. We go backwards through the string, converting until we
3280 * get to the position we are at now, and then stop. If this
3281 * position is far enough along in the string, this method is
3282 * faster than the other method. If the memory copy were the same
3283 * speed as the byte-by-byte loop, that position would be about
3284 * half-way, as at the half-way mark, parsing to the end and back
3285 * is one complete string's parse, the same amount as starting
3286 * over and going all the way through. Actually, it would be
3287 * somewhat less than half-way, as it's faster to just count bytes
3288 * than to also copy, and we don't have the overhead of allocating
3289 * a new string, changing the scalar to use it, and freeing the
3290 * existing one. But if the memory copy is fast, the break-even
3291 * point is somewhere after half way. The counting loop could be
3292 * sped up by vectorization, etc, to move the break-even point
3293 * further towards the beginning.
3294 * 2) if the string doesn't have enough space to handle the converted
3295 * value. A new string will have to be allocated, and one might
3296 * as well, given that, start from the beginning doing the first
3297 * method. We've spent extra time parsing the string and in
3298 * exchange all we've gotten is that we know precisely how big to
3299 * make the new one. Perl is more optimized for time than space,
3300 * so this case is a loser.
3301 * So what I've decided to do is not use the 2nd method unless it is
3302 * guaranteed that a new string won't have to be allocated, assuming
3303 * the worst case. I also decided not to put any more conditions on it
3304 * than this, for now. It seems likely that, since the worst case is
3305 * twice as big as the unknown portion of the string (plus 1), we won't
3306 * be guaranteed enough space, causing us to go to the first method,
3307 * unless the string is short, or the first variant character is near
3308 * the end of it. In either of these cases, it seems best to use the
3309 * 2nd method. The only circumstance I can think of where this would
3310 * be really slower is if the string had once had much more data in it
3311 * than it does now, but there is still a substantial amount in it */
3312
3313 {
3314 STRLEN invariant_head = t - s;
3315 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3316 if (SvLEN(sv) < size) {
3317
3318 /* Here, have decided to allocate a new string */
3319
3320 U8 *dst;
3321 U8 *d;
3322
3323 Newx(dst, size, U8);
3324
3325 /* If no known invariants at the beginning of the input string,
3326 * set so starts from there. Otherwise, can use memory copy to
3327 * get up to where we are now, and then start from here */
3328
3329 if (invariant_head <= 0) {
3330 d = dst;
3331 } else {
3332 Copy(s, dst, invariant_head, char);
3333 d = dst + invariant_head;
3334 }
3335
3336 while (t < e) {
3337 const UV uv = NATIVE8_TO_UNI(*t++);
3338 if (UNI_IS_INVARIANT(uv))
3339 *d++ = (U8)UNI_TO_NATIVE(uv);
3340 else {
3341 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3342 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3343 }
3344 }
3345 *d = '\0';
3346 SvPV_free(sv); /* No longer using pre-existing string */
3347 SvPV_set(sv, (char*)dst);
3348 SvCUR_set(sv, d - dst);
3349 SvLEN_set(sv, size);
3350 } else {
3351
3352 /* Here, have decided to get the exact size of the string.
3353 * Currently this happens only when we know that there is
3354 * guaranteed enough space to fit the converted string, so
3355 * don't have to worry about growing. If two_byte_count is 0,
3356 * then t points to the first byte of the string which hasn't
3357 * been examined yet. Otherwise two_byte_count is 1, and t
3358 * points to the first byte in the string that will expand to
3359 * two. Depending on this, start examining at t or 1 after t.
3360 * */
3361
3362 U8 *d = t + two_byte_count;
3363
3364
3365 /* Count up the remaining bytes that expand to two */
3366
3367 while (d < e) {
3368 const U8 chr = *d++;
3369 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3370 }
3371
3372 /* The string will expand by just the number of bytes that
3373 * occupy two positions. But we are one afterwards because of
3374 * the increment just above. This is the place to put the
3375 * trailing NUL, and to set the length before we decrement */
3376
3377 d += two_byte_count;
3378 SvCUR_set(sv, d - s);
3379 *d-- = '\0';
3380
3381
3382 /* Having decremented d, it points to the position to put the
3383 * very last byte of the expanded string. Go backwards through
3384 * the string, copying and expanding as we go, stopping when we
3385 * get to the part that is invariant the rest of the way down */
3386
3387 e--;
3388 while (e >= t) {
3389 const U8 ch = NATIVE8_TO_UNI(*e--);
3390 if (UNI_IS_INVARIANT(ch)) {
3391 *d-- = UNI_TO_NATIVE(ch);
3392 } else {
3393 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3394 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3395 }
3396 }
3397 }
75da9d4c
DM
3398
3399 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3400 /* Update pos. We do it at the end rather than during
3401 * the upgrade, to avoid slowing down the common case
3402 * (upgrade without pos) */
3403 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3404 if (mg) {
3405 I32 pos = mg->mg_len;
3406 if (pos > 0 && (U32)pos > invariant_head) {
3407 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3408 STRLEN n = (U32)pos - invariant_head;
3409 while (n > 0) {
3410 if (UTF8_IS_START(*d))
3411 d++;
3412 d++;
3413 n--;
3414 }
3415 mg->mg_len = d - (U8*)SvPVX(sv);
3416 }
3417 }
3418 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3419 magic_setutf8(sv,mg); /* clear UTF8 cache */
3420 }
b3ab6785 3421 }
560a288e 3422 }
b3ab6785
KW
3423
3424 /* Mark as UTF-8 even if no variant - saves scanning loop */
3425 SvUTF8_on(sv);
4411f3b6 3426 return SvCUR(sv);
560a288e
GS
3427}
3428
c461cf8f
JH
3429/*
3430=for apidoc sv_utf8_downgrade
3431
78ea37eb 3432Attempts to convert the PV of an SV from characters to bytes.
2bbc8d55
SP
3433If the PV contains a character that cannot fit
3434in a byte, this conversion will fail;
78ea37eb 3435in this case, either returns false or, if C<fail_ok> is not
c461cf8f
JH
3436true, croaks.
3437
0efd0472 3438This is not a general purpose Unicode to byte encoding interface:
13a6c0e0
JH
3439use the Encode extension for that.
3440
c461cf8f
JH
3441=cut
3442*/
3443
560a288e 3444bool
7bc54cea 3445Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
560a288e 3446{
97aff369 3447 dVAR;
7918f24d
NC
3448
3449 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3450
78ea37eb 3451 if (SvPOKp(sv) && SvUTF8(sv)) {
fa301091 3452 if (SvCUR(sv)) {
03cfe0ae 3453 U8 *s;
652088fc 3454 STRLEN len;
75da9d4c 3455 int mg_flags = SV_GMAGIC;
fa301091 3456
765f542d
NC
3457 if (SvIsCOW(sv)) {
3458 sv_force_normal_flags(sv, 0);
3459 }
75da9d4c
DM
3460 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3461 /* update pos */
3462 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3463 if (mg) {
3464 I32 pos = mg->mg_len;
3465 if (pos > 0) {
3466 sv_pos_b2u(sv, &pos);
3467 mg_flags = 0; /* sv_pos_b2u does get magic */
3468 mg->mg_len = pos;
3469 }
3470 }
3471 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3472 magic_setutf8(sv,mg); /* clear UTF8 cache */
3473
3474 }
3475 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3476
03cfe0ae 3477 if (!utf8_to_bytes(s, &len)) {
fa301091
JH
3478 if (fail_ok)
3479 return FALSE;
3480 else {
3481 if (PL_op)
3482 Perl_croak(aTHX_ "Wide character in %s",
53e06cf0 3483 OP_DESC(PL_op));
fa301091
JH
3484 else
3485 Perl_croak(aTHX_ "Wide character");
3486 }
4b3603a4 3487 }
b162af07 3488 SvCUR_set(sv, len);
67e989fb 3489 }
560a288e 3490 }
ffebcc3e 3491 SvUTF8_off(sv);
560a288e
GS
3492 return TRUE;
3493}
3494
c461cf8f
JH
3495/*
3496=for apidoc sv_utf8_encode
3497
78ea37eb
ST
3498Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3499flag off so that it looks like octets again.
c461cf8f
JH
3500
3501=cut
3502*/
3503
560a288e 3504void
7bc54cea 3505Perl_sv_utf8_encode(pTHX_ register SV *const sv)
560a288e 3506{
7918f24d
NC
3507 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3508
4c94c214 3509 if (SvREADONLY(sv)) {
654c723f 3510 sv_force_normal_flags(sv, 0);
4c94c214 3511 }
a5f5288a 3512 (void) sv_utf8_upgrade(sv);
560a288e
GS
3513 SvUTF8_off(sv);
3514}
3515
4411f3b6
NIS
3516/*
3517=for apidoc sv_utf8_decode
3518
78ea37eb
ST
3519If the PV of the SV is an octet sequence in UTF-8
3520and contains a multiple-byte character, the C<SvUTF8> flag is turned on
fde67290 3521so that it looks like a character. If the PV contains only single-byte
694cf0d2 3522characters, the C<SvUTF8> flag stays off.
78ea37eb 3523Scans PV for validity and returns false if the PV is invalid UTF-8.
4411f3b6
NIS
3524
3525=cut
3526*/
3527
560a288e 3528bool
7bc54cea 3529Perl_sv_utf8_decode(pTHX_ register SV *const sv)
560a288e 3530{
7918f24d
NC
3531 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3532
78ea37eb 3533 if (SvPOKp(sv)) {
75da9d4c 3534 const U8 *start, *c;
93524f2b 3535 const U8 *e;
9cbac4c7 3536
645c22ef
DM
3537 /* The octets may have got themselves encoded - get them back as
3538 * bytes
3539 */
3540 if (!sv_utf8_downgrade(sv, TRUE))
560a288e
GS
3541 return FALSE;
3542
3543 /* it is actually just a matter of turning the utf8 flag on, but
3544 * we want to make sure everything inside is valid utf8 first.
3545 */
75da9d4c 3546 c = start = (const U8 *) SvPVX_const(sv);
f9cf141a 3547 if (!is_utf8_string(c, SvCUR(sv)))
67e989fb 3548 return FALSE;
93524f2b 3549 e = (const U8 *) SvEND(sv);
511c2ff0 3550 while (c < e) {
b64e5050 3551 const U8 ch = *c++;
c4d5f83a 3552 if (!UTF8_IS_INVARIANT(ch)) {
67e989fb
JH
3553 SvUTF8_on(sv);
3554 break;
3555 }
560a288e 3556 }
75da9d4c
DM
3557 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3558 /* adjust pos to the start of a UTF8 char sequence */
3559 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3560 if (mg) {
3561 I32 pos = mg->mg_len;
3562 if (pos > 0) {
3563 for (c = start + pos; c > start; c--) {
3564 if (UTF8_IS_START(*c))
3565 break;
3566 }
3567 mg->mg_len = c - start;
3568 }
3569 }
3570 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3571 magic_setutf8(sv,mg); /* clear UTF8 cache */
3572 }
560a288e
GS
3573 }
3574 return TRUE;
3575}
3576
954c1994
GS
3577/*
3578=for apidoc sv_setsv
3579
645c22ef
DM
3580Copies the contents of the source SV C<ssv> into the destination SV
3581C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3582function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3583Loosely speaking, it performs a copy-by-value, obliterating any previous
3584content of the destination.
3585
3586You probably want to use one of the assortment of wrappers, such as
3587C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3588C<SvSetMagicSV_nosteal>.
3589
8d6d96c1
HS
3590=for apidoc sv_setsv_flags
3591
645c22ef
DM
3592Copies the contents of the source SV C<ssv> into the destination SV
3593C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
fde67290 3594function if the source SV needs to be reused. Does not handle 'set' magic.
645c22ef
DM
3595Loosely speaking, it performs a copy-by-value, obliterating any previous
3596content of the destination.
3597If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
fde67290
FC
3598C<ssv> if appropriate, else not. If the C<flags>
3599parameter has the C<NOSTEAL> bit set then the
3600buffers of temps will not be stolen. <sv_setsv>
5fcdf167 3601and C<sv_setsv_nomg> are implemented in terms of this function.
645c22ef
DM
3602
3603You probably want to use one of the assortment of wrappers, such as
3604C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3605C<SvSetMagicSV_nosteal>.
3606
3607This is the primary function for copying scalars, and most other
3608copy-ish functions and macros use this underneath.
8d6d96c1
HS
3609
3610=cut
3611*/
3612
5d0301b7 3613static void
7bc54cea 3614S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
5d0301b7 3615{
c8bbf675 3616 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3e6edce2 3617 HV *old_stash = NULL;
dd69841b 3618
7918f24d
NC
3619 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3620
bec4f4b4 3621 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
5d0301b7
NC
3622 const char * const name = GvNAME(sstr);
3623 const STRLEN len = GvNAMELEN(sstr);
0d092c36 3624 {
f7877b28
NC
3625 if (dtype >= SVt_PV) {
3626 SvPV_free(dstr);
3627 SvPV_set(dstr, 0);
3628 SvLEN_set(dstr, 0);
3629 SvCUR_set(dstr, 0);
3630 }
0d092c36 3631 SvUPGRADE(dstr, SVt_PVGV);
dedf8e73 3632 (void)SvOK_off(dstr);
83504ee4 3633 /* We have to turn this on here, even though we turn it off
ce2077b1 3634 below, as GvSTASH will fail an assertion otherwise. */
2e5b91de