This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Use static asserts when comparing sizeof() to a constant
[perl5.git] / sv.c
CommitLineData
5356d32e 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"
9d9a81f0
CB
34#ifdef __VMS
35# include <rms.h>
36#endif
79072805 37
2f8ed50e
OS
38#ifdef __Lynx__
39/* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
41#endif
42
a4eca1d4
JH
43#ifdef USE_QUADMATH
44# define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
46#else
47# define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
49#endif
50
9f53080a 51#ifndef SV_COW_THRESHOLD
e8c6a474 52# define SV_COW_THRESHOLD 0 /* COW iff len > K */
9f53080a
FC
53#endif
54#ifndef SV_COWBUF_THRESHOLD
e8c6a474 55# define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
9f53080a
FC
56#endif
57#ifndef SV_COW_MAX_WASTE_THRESHOLD
e8c6a474 58# define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
9f53080a
FC
59#endif
60#ifndef SV_COWBUF_WASTE_THRESHOLD
e8c6a474 61# define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
9f53080a
FC
62#endif
63#ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
e8c6a474 64# define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
9f53080a
FC
65#endif
66#ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
e8c6a474 67# define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
e8c6a474
YO
68#endif
69/* Work around compiler warnings about unsigned >= THRESHOLD when thres-
70 hold is 0. */
71#if SV_COW_THRESHOLD
72# define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
73#else
74# define GE_COW_THRESHOLD(cur) 1
75#endif
76#if SV_COWBUF_THRESHOLD
77# define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
78#else
79# define GE_COWBUF_THRESHOLD(cur) 1
80#endif
81#if SV_COW_MAX_WASTE_THRESHOLD
82# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
83#else
84# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
85#endif
86#if SV_COWBUF_WASTE_THRESHOLD
87# define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
88#else
89# define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
90#endif
91#if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
93#else
94# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
95#endif
96#if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
98#else
99# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
100#endif
101
102#define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
106)
107#define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
111)
cca0492e 112
e23c8137 113#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 114/* if adding more checks watch out for the following tests:
e23c8137
JH
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
117 * --jhi
118 */
6f207bd3 119# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
123 } STMT_END
e23c8137 124#else
6f207bd3 125# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
126#endif
127
958cdeac
TC
128static const char S_destroy[] = "DESTROY";
129#define S_destroy_len (sizeof(S_destroy)-1)
130
645c22ef
DM
131/* ============================================================================
132
133=head1 Allocation and deallocation of SVs.
17c7a51d 134
d2a0f284
JC
135An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
136sv, av, hv...) contains type and reference count information, and for
137many types, a pointer to the body (struct xrv, xpv, xpviv...), which
138contains fields specific to each type. Some types store all they need
139in the head, so don't have a body.
140
486ec47a 141In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
142and bodies are allocated out of arenas, which by default are
143approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
144Sv-bodies are allocated by their sv-type, guaranteeing size
145consistency needed to allocate safely from arrays.
146
d2a0f284
JC
147For SV-heads, the first slot in each arena is reserved, and holds a
148link to the next arena, some flags, and a note of the number of slots.
149Snaked through each arena chain is a linked list of free items; when
150this becomes empty, an extra arena is allocated and divided up into N
151items which are threaded into the free list.
152
153SV-bodies are similar, but they use arena-sets by default, which
154separate the link and info from the arena itself, and reclaim the 1st
155slot in the arena. SV-bodies are further described later.
645c22ef
DM
156
157The following global variables are associated with arenas:
158
7fefc6c1
KW
159 PL_sv_arenaroot pointer to list of SV arenas
160 PL_sv_root pointer to list of free SV structures
645c22ef 161
7fefc6c1
KW
162 PL_body_arenas head of linked-list of body arenas
163 PL_body_roots[] array of pointers to list of free bodies of svtype
164 arrays are indexed by the svtype needed
93e68bfb 165
d2a0f284
JC
166A few special SV heads are not allocated from an arena, but are
167instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
168The size of arenas can be changed from the default by setting
169PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
170
171The SV arena serves the secondary purpose of allowing still-live SVs
172to be located and destroyed during final cleanup.
173
174At the lowest level, the macros new_SV() and del_SV() grab and free
175an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
176to return the SV to the free list with error checking.) new_SV() calls
177more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
178SVs in the free list have their SvTYPE field set to all ones.
179
ff276b08 180At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 181perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 182start of the interpreter.
645c22ef 183
645c22ef
DM
184The function visit() scans the SV arenas list, and calls a specified
185function for each SV it finds which is still live - ie which has an SvTYPE
186other than all 1's, and a non-zero SvREFCNT. visit() is used by the
187following functions (specified as [function that calls visit()] / [function
188called by visit() for each SV]):
189
190 sv_report_used() / do_report_used()
f2524eef 191 dump all remaining SVs (debugging aid)
645c22ef 192
e4487e9b 193 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 194 do_clean_named_io_objs(),do_curse()
645c22ef 195 Attempt to free all objects pointed to by RVs,
e76981f9
FC
196 try to do the same for all objects indir-
197 ectly referenced by typeglobs too, and
198 then do a final sweep, cursing any
199 objects that remain. Called once from
645c22ef
DM
200 perl_destruct(), prior to calling sv_clean_all()
201 below.
202
203 sv_clean_all() / do_clean_all()
204 SvREFCNT_dec(sv) each remaining SV, possibly
205 triggering an sv_free(). It also sets the
206 SVf_BREAK flag on the SV to indicate that the
207 refcnt has been artificially lowered, and thus
208 stopping sv_free() from giving spurious warnings
209 about SVs which unexpectedly have a refcnt
210 of zero. called repeatedly from perl_destruct()
211 until there are no SVs left.
212
93e68bfb 213=head2 Arena allocator API Summary
645c22ef
DM
214
215Private API to rest of sv.c
216
217 new_SV(), del_SV(),
218
df0f0429 219 new_XPVNV(), del_XPVGV(),
645c22ef
DM
220 etc
221
222Public API:
223
8cf8f3d1 224 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 225
645c22ef
DM
226=cut
227
3e8320cc 228 * ========================================================================= */
645c22ef 229
4561caa4
CS
230/*
231 * "A time to plant, and a time to uproot what was planted..."
232 */
233
d7a2c63c
MHM
234#ifdef PERL_MEM_LOG
235# define MEM_LOG_NEW_SV(sv, file, line, func) \
236 Perl_mem_log_new_sv(sv, file, line, func)
237# define MEM_LOG_DEL_SV(sv, file, line, func) \
238 Perl_mem_log_del_sv(sv, file, line, func)
239#else
240# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
241# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
242#endif
243
fd0854ff 244#ifdef DEBUG_LEAKING_SCALARS
484e6108
FC
245# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
246 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 } STMT_END
d7a2c63c 248# define DEBUG_SV_SERIAL(sv) \
147e3846 249 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
d7a2c63c 250 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
251#else
252# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 253# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
254#endif
255
48614a46
NC
256#ifdef PERL_POISON
257# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 258# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
259/* Whilst I'd love to do this, it seems that things like to check on
260 unreferenced scalars
ce5dbf61 261# define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 262*/
ce5dbf61 263# define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
7e337ee0 264 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
265#else
266# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 267# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
ce5dbf61 268# define POISON_SV_HEAD(sv)
48614a46
NC
269#endif
270
990198f0
DM
271/* Mark an SV head as unused, and add to free list.
272 *
273 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
274 * its refcount artificially decremented during global destruction, so
275 * there may be dangling pointers to it. The last thing we want in that
276 * case is for it to be reused. */
277
053fc874
GS
278#define plant_SV(p) \
279 STMT_START { \
990198f0 280 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
281 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
282 DEBUG_SV_SERIAL(p); \
fd0854ff 283 FREE_SV_DEBUG_FILE(p); \
ce5dbf61 284 POISON_SV_HEAD(p); \
053fc874 285 SvFLAGS(p) = SVTYPEMASK; \
990198f0 286 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 287 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
288 PL_sv_root = (p); \
289 } \
053fc874
GS
290 --PL_sv_count; \
291 } STMT_END
a0d0e21e 292
053fc874
GS
293#define uproot_SV(p) \
294 STMT_START { \
295 (p) = PL_sv_root; \
daba3364 296 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
297 ++PL_sv_count; \
298 } STMT_END
299
645c22ef 300
cac9b346
NC
301/* make some more SVs by adding another arena */
302
cac9b346
NC
303STATIC SV*
304S_more_sv(pTHX)
305{
306 SV* sv;
9a87bd09
NC
307 char *chunk; /* must use New here to match call to */
308 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
309 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
310 uproot_SV(sv);
311 return sv;
312}
313
645c22ef
DM
314/* new_SV(): return a new, empty SV head */
315
eba0f806
DM
316#ifdef DEBUG_LEAKING_SCALARS
317/* provide a real function for a debugger to play with */
318STATIC SV*
d7a2c63c 319S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
320{
321 SV* sv;
322
eba0f806
DM
323 if (PL_sv_root)
324 uproot_SV(sv);
325 else
cac9b346 326 sv = S_more_sv(aTHX);
eba0f806
DM
327 SvANY(sv) = 0;
328 SvREFCNT(sv) = 1;
329 SvFLAGS(sv) = 0;
fd0854ff 330 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
331 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
332 ? PL_parser->copline
333 : PL_curcop
f24aceb1
DM
334 ? CopLINE(PL_curcop)
335 : 0
e385c3bf 336 );
fd0854ff 337 sv->sv_debug_inpad = 0;
cd676548 338 sv->sv_debug_parent = NULL;
484e6108 339 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
340
341 sv->sv_debug_serial = PL_sv_serial++;
342
343 MEM_LOG_NEW_SV(sv, file, line, func);
147e3846 344 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
d7a2c63c
MHM
345 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
346
eba0f806
DM
347 return sv;
348}
d7a2c63c 349# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
350
351#else
352# define new_SV(p) \
053fc874 353 STMT_START { \
053fc874
GS
354 if (PL_sv_root) \
355 uproot_SV(p); \
356 else \
cac9b346 357 (p) = S_more_sv(aTHX); \
053fc874
GS
358 SvANY(p) = 0; \
359 SvREFCNT(p) = 1; \
360 SvFLAGS(p) = 0; \
d7a2c63c 361 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 362 } STMT_END
eba0f806 363#endif
463ee0b2 364
645c22ef
DM
365
366/* del_SV(): return an empty SV head to the free list */
367
a0d0e21e 368#ifdef DEBUGGING
4561caa4 369
053fc874
GS
370#define del_SV(p) \
371 STMT_START { \
aea4f609 372 if (DEBUG_D_TEST) \
053fc874
GS
373 del_sv(p); \
374 else \
375 plant_SV(p); \
053fc874 376 } STMT_END
a0d0e21e 377
76e3520e 378STATIC void
cea2e8a9 379S_del_sv(pTHX_ SV *p)
463ee0b2 380{
7918f24d
NC
381 PERL_ARGS_ASSERT_DEL_SV;
382
aea4f609 383 if (DEBUG_D_TEST) {
4633a7c4 384 SV* sva;
a3b680e6 385 bool ok = 0;
daba3364 386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
387 const SV * const sv = sva + 1;
388 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 389 if (p >= sv && p < svend) {
a0d0e21e 390 ok = 1;
c0ff570e
NC
391 break;
392 }
a0d0e21e
LW
393 }
394 if (!ok) {
9b387841 395 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
147e3846 396 "Attempt to free non-arena SV: 0x%" UVxf
9b387841 397 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
398 return;
399 }
400 }
4561caa4 401 plant_SV(p);
463ee0b2 402}
a0d0e21e 403
4561caa4
CS
404#else /* ! DEBUGGING */
405
406#define del_SV(p) plant_SV(p)
407
408#endif /* DEBUGGING */
463ee0b2 409
645c22ef
DM
410
411/*
ccfc67b7
JH
412=head1 SV Manipulation Functions
413
645c22ef
DM
414=for apidoc sv_add_arena
415
416Given a chunk of memory, link it to the head of the list of arenas,
417and split it into a list of free SVs.
418
419=cut
420*/
421
d2bd4e7f
NC
422static void
423S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 424{
daba3364 425 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
426 SV* sv;
427 SV* svend;
4633a7c4 428
7918f24d
NC
429 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430
4633a7c4 431 /* The first SV in an arena isn't an SV. */
3280af22 432 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
433 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
434 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435
3280af22
NIS
436 PL_sv_arenaroot = sva;
437 PL_sv_root = sva + 1;
4633a7c4
LW
438
439 svend = &sva[SvREFCNT(sva) - 1];
440 sv = sva + 1;
463ee0b2 441 while (sv < svend) {
3eef1deb 442 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 443#ifdef DEBUGGING
978b032e 444 SvREFCNT(sv) = 0;
03e36789 445#endif
4b69cbe3 446 /* Must always set typemask because it's always checked in on cleanup
03e36789 447 when the arenas are walked looking for objects. */
8990e307 448 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
449 sv++;
450 }
3eef1deb 451 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
452#ifdef DEBUGGING
453 SvREFCNT(sv) = 0;
454#endif
4633a7c4
LW
455 SvFLAGS(sv) = SVTYPEMASK;
456}
457
055972dc
DM
458/* visit(): call the named function for each non-free SV in the arenas
459 * whose flags field matches the flags/mask args. */
645c22ef 460
5226ed68 461STATIC I32
de37a194 462S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 463{
4633a7c4 464 SV* sva;
5226ed68 465 I32 visited = 0;
8990e307 466
7918f24d
NC
467 PERL_ARGS_ASSERT_VISIT;
468
daba3364 469 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
470 const SV * const svend = &sva[SvREFCNT(sva)];
471 SV* sv;
4561caa4 472 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 473 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
474 && (sv->sv_flags & mask) == flags
475 && SvREFCNT(sv))
476 {
942481a7 477 (*f)(aTHX_ sv);
5226ed68
JH
478 ++visited;
479 }
8990e307
LW
480 }
481 }
5226ed68 482 return visited;
8990e307
LW
483}
484
758a08c3
JH
485#ifdef DEBUGGING
486
645c22ef
DM
487/* called by sv_report_used() for each live SV */
488
489static void
5fa45a31 490do_report_used(pTHX_ SV *const sv)
645c22ef 491{
e4787c0c 492 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
493 PerlIO_printf(Perl_debug_log, "****\n");
494 sv_dump(sv);
495 }
496}
758a08c3 497#endif
645c22ef
DM
498
499/*
500=for apidoc sv_report_used
501
fde67290 502Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
503
504=cut
505*/
506
8990e307 507void
864dbfa3 508Perl_sv_report_used(pTHX)
4561caa4 509{
ff270d3a 510#ifdef DEBUGGING
055972dc 511 visit(do_report_used, 0, 0);
96a5add6
AL
512#else
513 PERL_UNUSED_CONTEXT;
ff270d3a 514#endif
4561caa4
CS
515}
516
645c22ef
DM
517/* called by sv_clean_objs() for each live SV */
518
519static void
de37a194 520do_clean_objs(pTHX_ SV *const ref)
645c22ef 521{
ea724faa
NC
522 assert (SvROK(ref));
523 {
823a54a3
AL
524 SV * const target = SvRV(ref);
525 if (SvOBJECT(target)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
527 if (SvWEAKREF(ref)) {
528 sv_del_backref(target, ref);
529 SvWEAKREF_off(ref);
530 SvRV_set(ref, NULL);
531 } else {
532 SvROK_off(ref);
533 SvRV_set(ref, NULL);
fc2b2dca 534 SvREFCNT_dec_NN(target);
823a54a3 535 }
645c22ef
DM
536 }
537 }
645c22ef
DM
538}
539
645c22ef 540
e4487e9b
DM
541/* clear any slots in a GV which hold objects - except IO;
542 * called by sv_clean_objs() for each live GV */
543
645c22ef 544static void
f30de749 545do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 546{
57ef47cc 547 SV *obj;
ea724faa 548 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 549 assert(isGV_with_GP(sv));
57ef47cc
DM
550 if (!GvGP(sv))
551 return;
552
553 /* freeing GP entries may indirectly free the current GV;
554 * hold onto it while we mess with the GP slots */
555 SvREFCNT_inc(sv);
556
557 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
558 DEBUG_D((PerlIO_printf(Perl_debug_log,
559 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 GvSV(sv) = NULL;
fc2b2dca 561 SvREFCNT_dec_NN(obj);
57ef47cc
DM
562 }
563 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
564 DEBUG_D((PerlIO_printf(Perl_debug_log,
565 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 GvAV(sv) = NULL;
fc2b2dca 567 SvREFCNT_dec_NN(obj);
57ef47cc
DM
568 }
569 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
570 DEBUG_D((PerlIO_printf(Perl_debug_log,
571 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 GvHV(sv) = NULL;
fc2b2dca 573 SvREFCNT_dec_NN(obj);
57ef47cc
DM
574 }
575 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
576 DEBUG_D((PerlIO_printf(Perl_debug_log,
577 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 578 GvCV_set(sv, NULL);
fc2b2dca 579 SvREFCNT_dec_NN(obj);
57ef47cc 580 }
fc2b2dca 581 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
582}
583
68b590d9 584/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
585 * called by sv_clean_objs() for each live GV */
586
587static void
588do_clean_named_io_objs(pTHX_ SV *const sv)
589{
e4487e9b
DM
590 SV *obj;
591 assert(SvTYPE(sv) == SVt_PVGV);
592 assert(isGV_with_GP(sv));
68b590d9 593 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
594 return;
595
596 SvREFCNT_inc(sv);
57ef47cc
DM
597 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
598 DEBUG_D((PerlIO_printf(Perl_debug_log,
599 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 GvIOp(sv) = NULL;
fc2b2dca 601 SvREFCNT_dec_NN(obj);
645c22ef 602 }
fc2b2dca 603 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 604}
645c22ef 605
4155e4fe
FC
606/* Void wrapper to pass to visit() */
607static void
608do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
609 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
610 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
611 return;
612 (void)curse(sv, 0);
613}
614
645c22ef
DM
615/*
616=for apidoc sv_clean_objs
617
fde67290 618Attempt to destroy all objects not yet freed.
645c22ef
DM
619
620=cut
621*/
622
4561caa4 623void
864dbfa3 624Perl_sv_clean_objs(pTHX)
4561caa4 625{
68b590d9 626 GV *olddef, *olderr;
3280af22 627 PL_in_clean_objs = TRUE;
055972dc 628 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
629 /* Some barnacles may yet remain, clinging to typeglobs.
630 * Run the non-IO destructors first: they may want to output
631 * error messages, close files etc */
d011219a 632 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 633 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
634 /* And if there are some very tenacious barnacles clinging to arrays,
635 closures, or what have you.... */
636 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
637 olddef = PL_defoutgv;
638 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
639 if (olddef && isGV_with_GP(olddef))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
641 olderr = PL_stderrgv;
642 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
643 if (olderr && isGV_with_GP(olderr))
644 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
645 SvREFCNT_dec(olddef);
3280af22 646 PL_in_clean_objs = FALSE;
4561caa4
CS
647}
648
645c22ef
DM
649/* called by sv_clean_all() for each live SV */
650
651static void
de37a194 652do_clean_all(pTHX_ SV *const sv)
645c22ef 653{
daba3364 654 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 655 /* don't clean pid table and strtab */
d17ea597 656 return;
cddfcddc 657 }
147e3846 658 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
645c22ef 659 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 660 SvREFCNT_dec_NN(sv);
645c22ef
DM
661}
662
663/*
664=for apidoc sv_clean_all
665
666Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 667cleanup. This function may have to be called multiple times to free
ff276b08 668SVs which are in complex self-referential hierarchies.
645c22ef
DM
669
670=cut
671*/
672
5226ed68 673I32
864dbfa3 674Perl_sv_clean_all(pTHX)
8990e307 675{
5226ed68 676 I32 cleaned;
3280af22 677 PL_in_clean_all = TRUE;
055972dc 678 cleaned = visit(do_clean_all, 0,0);
5226ed68 679 return cleaned;
8990e307 680}
463ee0b2 681
5e258f8c
JC
682/*
683 ARENASETS: a meta-arena implementation which separates arena-info
684 into struct arena_set, which contains an array of struct
685 arena_descs, each holding info for a single arena. By separating
686 the meta-info from the arena, we recover the 1st slot, formerly
687 borrowed for list management. The arena_set is about the size of an
39244528 688 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
689
690 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
691 memory in the last arena-set (1/2 on average). In trade, we get
692 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 693 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
694 small arenas for large, rare body types, by changing array* fields
695 in body_details_by_type[] below.
5e258f8c 696*/
5e258f8c 697struct arena_desc {
398c677b
NC
698 char *arena; /* the raw storage, allocated aligned */
699 size_t size; /* its size ~4k typ */
e5973ed5 700 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
701};
702
e6148039
NC
703struct arena_set;
704
705/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 706 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
707 therefore likely to be 1 aligned memory page. */
708
709#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
710 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
711
712struct arena_set {
713 struct arena_set* next;
0a848332
NC
714 unsigned int set_size; /* ie ARENAS_PER_SET */
715 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
716 struct arena_desc set[ARENAS_PER_SET];
717};
718
645c22ef
DM
719/*
720=for apidoc sv_free_arenas
721
fde67290 722Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
723heads and bodies within the arenas must already have been freed.
724
725=cut
7fefc6c1 726
645c22ef 727*/
4633a7c4 728void
864dbfa3 729Perl_sv_free_arenas(pTHX)
4633a7c4
LW
730{
731 SV* sva;
732 SV* svanext;
0a848332 733 unsigned int i;
4633a7c4
LW
734
735 /* Free arenas here, but be careful about fake ones. (We assume
736 contiguity of the fake ones with the corresponding real ones.) */
737
3280af22 738 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 739 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 740 while (svanext && SvFAKE(svanext))
daba3364 741 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
742
743 if (!SvFAKE(sva))
1df70142 744 Safefree(sva);
4633a7c4 745 }
93e68bfb 746
5e258f8c 747 {
0a848332
NC
748 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
749
750 while (aroot) {
751 struct arena_set *current = aroot;
752 i = aroot->curr;
753 while (i--) {
5e258f8c
JC
754 assert(aroot->set[i].arena);
755 Safefree(aroot->set[i].arena);
756 }
0a848332
NC
757 aroot = aroot->next;
758 Safefree(current);
5e258f8c
JC
759 }
760 }
dc8220bf 761 PL_body_arenas = 0;
fdda85ca 762
0a848332
NC
763 i = PERL_ARENA_ROOTS_SIZE;
764 while (i--)
93e68bfb 765 PL_body_roots[i] = 0;
93e68bfb 766
3280af22
NIS
767 PL_sv_arenaroot = 0;
768 PL_sv_root = 0;
4633a7c4
LW
769}
770
bd81e77b
NC
771/*
772 Here are mid-level routines that manage the allocation of bodies out
cc463ce5 773 of the various arenas. There are 4 kinds of arenas:
29489e7c 774
bd81e77b
NC
775 1. SV-head arenas, which are discussed and handled above
776 2. regular body arenas
777 3. arenas for reduced-size bodies
778 4. Hash-Entry arenas
29489e7c 779
bd81e77b
NC
780 Arena types 2 & 3 are chained by body-type off an array of
781 arena-root pointers, which is indexed by svtype. Some of the
782 larger/less used body types are malloced singly, since a large
783 unused block of them is wasteful. Also, several svtypes dont have
784 bodies; the data fits into the sv-head itself. The arena-root
785 pointer thus has a few unused root-pointers (which may be hijacked
cc463ce5 786 later for arena type 4)
29489e7c 787
bd81e77b
NC
788 3 differs from 2 as an optimization; some body types have several
789 unused fields in the front of the structure (which are kept in-place
790 for consistency). These bodies can be allocated in smaller chunks,
791 because the leading fields arent accessed. Pointers to such bodies
792 are decremented to point at the unused 'ghost' memory, knowing that
793 the pointers are used with offsets to the real memory.
29489e7c 794
d2a0f284
JC
795Allocation of SV-bodies is similar to SV-heads, differing as follows;
796the allocation mechanism is used for many body types, so is somewhat
797more complicated, it uses arena-sets, and has no need for still-live
798SV detection.
799
800At the outermost level, (new|del)_X*V macros return bodies of the
801appropriate type. These macros call either (new|del)_body_type or
802(new|del)_body_allocated macro pairs, depending on specifics of the
803type. Most body types use the former pair, the latter pair is used to
804allocate body types with "ghost fields".
805
806"ghost fields" are fields that are unused in certain types, and
69ba284b 807consequently don't need to actually exist. They are declared because
d2a0f284
JC
808they're part of a "base type", which allows use of functions as
809methods. The simplest examples are AVs and HVs, 2 aggregate types
810which don't use the fields which support SCALAR semantics.
811
69ba284b 812For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
813chunks, we thus avoid wasted memory for those unaccessed members.
814When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 815size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
816structure. (But things will all go boom if you write to the part that
817is "not there", because you'll be overwriting the last members of the
818preceding structure in memory.)
819
69ba284b 820We calculate the correction using the STRUCT_OFFSET macro on the first
a05ea1cf 821member present. If the allocated structure is smaller (no initial NV
69ba284b
NC
822actually allocated) then the net effect is to subtract the size of the NV
823from the pointer, to return a new pointer as if an initial NV were actually
a05ea1cf 824allocated. (We were using structures named *_allocated for this, but
69ba284b
NC
825this turned out to be a subtle bug, because a structure without an NV
826could have a lower alignment constraint, but the compiler is allowed to
827optimised accesses based on the alignment constraint of the actual pointer
828to the full structure, for example, using a single 64 bit load instruction
829because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284 830
a05ea1cf 831This is the same trick as was used for NV and IV bodies. Ironically it
d2a0f284 832doesn't need to be used for NV bodies any more, because NV is now at
5b306eef
DD
833the start of the structure. IV bodies, and also in some builds NV bodies,
834don't need it either, because they are no longer allocated.
d2a0f284
JC
835
836In turn, the new_body_* allocators call S_new_body(), which invokes
837new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 838linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
839necessary to refresh an empty list. Then the lock is released, and
840the body is returned.
841
99816f8d 842Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
843bodies, which it strings into a linked list. It looks up arena-size
844and body-size from the body_details table described below, thus
845supporting the multiple body-types.
846
847If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848the (new|del)_X*V macros are mapped directly to malloc/free.
849
d2a0f284
JC
850For each sv-type, struct body_details bodies_by_type[] carries
851parameters which control these aspects of SV handling:
852
853Arena_size determines whether arenas are used for this body type, and if
854so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855zero, forcing individual mallocs and frees.
856
857Body_size determines how big a body is, and therefore how many fit into
858each arena. Offset carries the body-pointer adjustment needed for
69ba284b 859"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
860
861But its main purpose is to parameterize info needed in
862Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 863vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
864are used for this, except for arena_size.
865
866For the sv-types that have no bodies, arenas are not used, so those
867PL_body_roots[sv_type] are unused, and can be overloaded. In
868something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 869PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 870bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 871available in hv.c.
d2a0f284 872
29489e7c
DM
873*/
874
bd81e77b 875struct body_details {
0fb58b32 876 U8 body_size; /* Size to allocate */
10666ae3 877 U8 copy; /* Size of structure to copy (may be shorter) */
5b306eef
DD
878 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
879 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
880 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
881 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
882 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
883 U32 arena_size; /* Size of arena to allocate */
bd81e77b 884};
29489e7c 885
b11187bf
TC
886#define ALIGNED_TYPE_NAME(name) name##_aligned
887#define ALIGNED_TYPE(name) \
888 typedef union { \
889 name align_me; \
890 NV nv; \
891 IV iv; \
892 } ALIGNED_TYPE_NAME(name);
893
894ALIGNED_TYPE(regexp);
895ALIGNED_TYPE(XPVGV);
896ALIGNED_TYPE(XPVLV);
897ALIGNED_TYPE(XPVAV);
898ALIGNED_TYPE(XPVHV);
899ALIGNED_TYPE(XPVCV);
900ALIGNED_TYPE(XPVFM);
901ALIGNED_TYPE(XPVIO);
902
bd81e77b
NC
903#define HADNV FALSE
904#define NONV TRUE
29489e7c 905
d2a0f284 906
bd81e77b
NC
907#ifdef PURIFY
908/* With -DPURFIY we allocate everything directly, and don't use arenas.
909 This seems a rather elegant way to simplify some of the code below. */
910#define HASARENA FALSE
911#else
912#define HASARENA TRUE
913#endif
914#define NOARENA FALSE
29489e7c 915
d2a0f284
JC
916/* Size the arenas to exactly fit a given number of bodies. A count
917 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
918 simplifying the default. If count > 0, the arena is sized to fit
919 only that many bodies, allowing arenas to be used for large, rare
920 bodies (XPVFM, XPVIO) without undue waste. The arena size is
921 limited by PERL_ARENA_SIZE, so we can safely oversize the
922 declarations.
923 */
95db5f15
MB
924#define FIT_ARENA0(body_size) \
925 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
926#define FIT_ARENAn(count,body_size) \
927 ( count * body_size <= PERL_ARENA_SIZE) \
928 ? count * body_size \
929 : FIT_ARENA0 (body_size)
930#define FIT_ARENA(count,body_size) \
cd1dc8e2 931 (U32)(count \
95db5f15 932 ? FIT_ARENAn (count, body_size) \
cd1dc8e2 933 : FIT_ARENA0 (body_size))
d2a0f284 934
bd81e77b
NC
935/* Calculate the length to copy. Specifically work out the length less any
936 final padding the compiler needed to add. See the comment in sv_upgrade
937 for why copying the padding proved to be a bug. */
29489e7c 938
bd81e77b
NC
939#define copy_length(type, last_member) \
940 STRUCT_OFFSET(type, last_member) \
daba3364 941 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 942
bd81e77b 943static const struct body_details bodies_by_type[] = {
829cd18a
NC
944 /* HEs use this offset for their arena. */
945 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 946
db93c0c4
NC
947 /* IVs are in the head, so the allocation size is 0. */
948 { 0,
d2a0f284 949 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 950 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 951 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
952 },
953
5b306eef
DD
954#if NVSIZE <= IVSIZE
955 { 0, sizeof(NV),
956 STRUCT_OFFSET(XPVNV, xnv_u),
957 SVt_NV, FALSE, HADNV, NOARENA, 0 },
958#else
6e128786
NC
959 { sizeof(NV), sizeof(NV),
960 STRUCT_OFFSET(XPVNV, xnv_u),
961 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
5b306eef 962#endif
d2a0f284 963
bc337e5c 964 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
965 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
966 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 967 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 968 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 969
d361b004
KW
970 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
971 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
972 + STRUCT_OFFSET(XPV, xpv_cur),
973 SVt_INVLIST, TRUE, NONV, HASARENA,
974 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 975
bc337e5c 976 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
977 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
978 + STRUCT_OFFSET(XPV, xpv_cur),
979 SVt_PVIV, FALSE, NONV, HASARENA,
980 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 981
bc337e5c 982 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
983 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
984 + STRUCT_OFFSET(XPV, xpv_cur),
985 SVt_PVNV, FALSE, HADNV, HASARENA,
986 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 987
6e128786 988 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 989 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 990
b11187bf 991 { sizeof(ALIGNED_TYPE_NAME(regexp)),
601dfd0a
NC
992 sizeof(regexp),
993 0,
ecff11eb 994 SVt_REGEXP, TRUE, NONV, HASARENA,
b11187bf 995 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(regexp)))
5c35adbb 996 },
4df7f6af 997
b11187bf
TC
998 { sizeof(ALIGNED_TYPE_NAME(XPVGV)), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
999 HASARENA, FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVGV))) },
d2a0f284 1000
b11187bf
TC
1001 { sizeof(ALIGNED_TYPE_NAME(XPVLV)), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1002 HASARENA, FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVLV))) },
d2a0f284 1003
b11187bf 1004 { sizeof(ALIGNED_TYPE_NAME(XPVAV)),
4f7003f5 1005 copy_length(XPVAV, xav_alloc),
601dfd0a 1006 0,
69ba284b 1007 SVt_PVAV, TRUE, NONV, HASARENA,
b11187bf 1008 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVAV))) },
d2a0f284 1009
b11187bf 1010 { sizeof(ALIGNED_TYPE_NAME(XPVHV)),
359164a0 1011 copy_length(XPVHV, xhv_max),
601dfd0a 1012 0,
69ba284b 1013 SVt_PVHV, TRUE, NONV, HASARENA,
b11187bf 1014 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVHV))) },
d2a0f284 1015
b11187bf 1016 { sizeof(ALIGNED_TYPE_NAME(XPVCV)),
601dfd0a
NC
1017 sizeof(XPVCV),
1018 0,
69ba284b 1019 SVt_PVCV, TRUE, NONV, HASARENA,
b11187bf 1020 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVCV))) },
69ba284b 1021
b11187bf 1022 { sizeof(ALIGNED_TYPE_NAME(XPVFM)),
601dfd0a
NC
1023 sizeof(XPVFM),
1024 0,
69ba284b 1025 SVt_PVFM, TRUE, NONV, NOARENA,
b11187bf 1026 FIT_ARENA(20, sizeof(ALIGNED_TYPE_NAME(XPVFM))) },
d2a0f284 1027
b11187bf 1028 { sizeof(ALIGNED_TYPE_NAME(XPVIO)),
601dfd0a
NC
1029 sizeof(XPVIO),
1030 0,
b6f60916 1031 SVt_PVIO, TRUE, NONV, HASARENA,
b11187bf 1032 FIT_ARENA(24, sizeof(ALIGNED_TYPE_NAME(XPVIO))) },
bd81e77b 1033};
29489e7c 1034
bd81e77b 1035#define new_body_allocated(sv_type) \
d2a0f284 1036 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1037 - bodies_by_type[sv_type].offset)
29489e7c 1038
26359cfa
NC
1039/* return a thing to the free list */
1040
1041#define del_body(thing, root) \
1042 STMT_START { \
1043 void ** const thing_copy = (void **)thing; \
1044 *thing_copy = *root; \
1045 *root = (void*)thing_copy; \
1046 } STMT_END
29489e7c 1047
bd81e77b 1048#ifdef PURIFY
5b306eef
DD
1049#if !(NVSIZE <= IVSIZE)
1050# define new_XNV() safemalloc(sizeof(XPVNV))
1051#endif
beeec492
NC
1052#define new_XPVNV() safemalloc(sizeof(XPVNV))
1053#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 1054
beeec492 1055#define del_XPVGV(p) safefree(p)
29489e7c 1056
bd81e77b 1057#else /* !PURIFY */
29489e7c 1058
5b306eef
DD
1059#if !(NVSIZE <= IVSIZE)
1060# define new_XNV() new_body_allocated(SVt_NV)
1061#endif
65ac1738 1062#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1063#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1064
26359cfa
NC
1065#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1066 &PL_body_roots[SVt_PVGV])
1d7c1841 1067
bd81e77b 1068#endif /* PURIFY */
93e68bfb 1069
bd81e77b 1070/* no arena for you! */
93e68bfb 1071
bd81e77b 1072#define new_NOARENA(details) \
beeec492 1073 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1074#define new_NOARENAZ(details) \
beeec492 1075 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1076
1e30fcd5
NC
1077void *
1078Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1079 const size_t arena_size)
d2a0f284 1080{
d2a0f284 1081 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1082 struct arena_desc *adesc;
1083 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1084 unsigned int curr;
d2a0f284
JC
1085 char *start;
1086 const char *end;
02982131 1087 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
20b7effb
JH
1088#if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1089 dVAR;
1090#endif
a6098fe8 1091#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
23e9d66c
NC
1092 static bool done_sanity_check;
1093
a6098fe8 1094 /* PERL_GLOBAL_STRUCT cannot coexist with global
0b2d3faa 1095 * variables like done_sanity_check. */
10666ae3 1096 if (!done_sanity_check) {
ea471437 1097 unsigned int i = SVt_LAST;
10666ae3
NC
1098
1099 done_sanity_check = TRUE;
1100
1101 while (i--)
1102 assert (bodies_by_type[i].type == i);
1103 }
1104#endif
1105
02982131 1106 assert(arena_size);
23e9d66c 1107
99816f8d
NC
1108 /* may need new arena-set to hold new arena */
1109 if (!aroot || aroot->curr >= aroot->set_size) {
1110 struct arena_set *newroot;
1111 Newxz(newroot, 1, struct arena_set);
1112 newroot->set_size = ARENAS_PER_SET;
1113 newroot->next = aroot;
1114 aroot = newroot;
1115 PL_body_arenas = (void *) newroot;
1116 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1117 }
1118
1119 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1120 curr = aroot->curr++;
1121 adesc = &(aroot->set[curr]);
1122 assert(!adesc->arena);
1123
1124 Newx(adesc->arena, good_arena_size, char);
1125 adesc->size = good_arena_size;
1126 adesc->utype = sv_type;
147e3846 1127 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
99816f8d
NC
1128 curr, (void*)adesc->arena, (UV)good_arena_size));
1129
1130 start = (char *) adesc->arena;
d2a0f284 1131
29657bb6
NC
1132 /* Get the address of the byte after the end of the last body we can fit.
1133 Remember, this is integer division: */
02982131 1134 end = start + good_arena_size / body_size * body_size;
d2a0f284 1135
486ec47a 1136 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1137#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1138 DEBUG_m(PerlIO_printf(Perl_debug_log,
1139 "arena %p end %p arena-size %d (from %d) type %d "
1140 "size %d ct %d\n",
02982131
NC
1141 (void*)start, (void*)end, (int)good_arena_size,
1142 (int)arena_size, sv_type, (int)body_size,
1143 (int)good_arena_size / (int)body_size));
d8fca402 1144#else
d2a0f284
JC
1145 DEBUG_m(PerlIO_printf(Perl_debug_log,
1146 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1147 (void*)start, (void*)end,
02982131
NC
1148 (int)arena_size, sv_type, (int)body_size,
1149 (int)good_arena_size / (int)body_size));
d8fca402 1150#endif
d2a0f284
JC
1151 *root = (void *)start;
1152
29657bb6
NC
1153 while (1) {
1154 /* Where the next body would start: */
d2a0f284 1155 char * const next = start + body_size;
29657bb6
NC
1156
1157 if (next >= end) {
1158 /* This is the last body: */
1159 assert(next == end);
1160
1161 *(void **)start = 0;
1162 return *root;
1163 }
1164
d2a0f284
JC
1165 *(void**) start = (void *)next;
1166 start = next;
1167 }
d2a0f284
JC
1168}
1169
1170/* grab a new thing from the free list, allocating more if necessary.
1171 The inline version is used for speed in hot routines, and the
1172 function using it serves the rest (unless PURIFY).
1173*/
1174#define new_body_inline(xpv, sv_type) \
1175 STMT_START { \
1176 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1177 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1178 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1179 bodies_by_type[sv_type].body_size,\
1180 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1181 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1182 } STMT_END
1183
1184#ifndef PURIFY
1185
1186STATIC void *
de37a194 1187S_new_body(pTHX_ const svtype sv_type)
d2a0f284 1188{
d2a0f284
JC
1189 void *xpv;
1190 new_body_inline(xpv, sv_type);
1191 return xpv;
1192}
1193
1194#endif
93e68bfb 1195
238b27b3
NC
1196static const struct body_details fake_rv =
1197 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1198
bd81e77b
NC
1199/*
1200=for apidoc sv_upgrade
93e68bfb 1201
bd81e77b
NC
1202Upgrade an SV to a more complex form. Generally adds a new body type to the
1203SV, then copies across as much information as possible from the old body.
9521ca61
FC
1204It croaks if the SV is already in a more complex form than requested. You
1205generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1206before calling C<sv_upgrade>, and hence does not croak. See also
fbe13c60 1207C<L</svtype>>.
93e68bfb 1208
bd81e77b 1209=cut
93e68bfb 1210*/
93e68bfb 1211
bd81e77b 1212void
5aaab254 1213Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1214{
bd81e77b
NC
1215 void* old_body;
1216 void* new_body;
42d0e0b7 1217 const svtype old_type = SvTYPE(sv);
d2a0f284 1218 const struct body_details *new_type_details;
238b27b3 1219 const struct body_details *old_type_details
bd81e77b 1220 = bodies_by_type + old_type;
ed7df46e 1221 SV *referent = NULL;
cac9b346 1222
7918f24d
NC
1223 PERL_ARGS_ASSERT_SV_UPGRADE;
1224
1776cbe8
NC
1225 if (old_type == new_type)
1226 return;
1227
1228 /* This clause was purposefully added ahead of the early return above to
1229 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1230 inference by Nick I-S that it would fix other troublesome cases. See
1231 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1232
1233 Given that shared hash key scalars are no longer PVIV, but PV, there is
1234 no longer need to unshare so as to free up the IVX slot for its proper
1235 purpose. So it's safe to move the early return earlier. */
1236
093085a8 1237 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1238 sv_force_normal_flags(sv, 0);
1239 }
cac9b346 1240
bd81e77b 1241 old_body = SvANY(sv);
de042e1d 1242
bd81e77b
NC
1243 /* Copying structures onto other structures that have been neatly zeroed
1244 has a subtle gotcha. Consider XPVMG
cac9b346 1245
bd81e77b
NC
1246 +------+------+------+------+------+-------+-------+
1247 | NV | CUR | LEN | IV | MAGIC | STASH |
1248 +------+------+------+------+------+-------+-------+
1249 0 4 8 12 16 20 24 28
645c22ef 1250
bd81e77b
NC
1251 where NVs are aligned to 8 bytes, so that sizeof that structure is
1252 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1253
bd81e77b
NC
1254 +------+------+------+------+------+-------+-------+------+
1255 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1256 +------+------+------+------+------+-------+-------+------+
1257 0 4 8 12 16 20 24 28 32
08742458 1258
bd81e77b 1259 so what happens if you allocate memory for this structure:
30f9da9e 1260
bd81e77b
NC
1261 +------+------+------+------+------+-------+-------+------+------+...
1262 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1263 +------+------+------+------+------+-------+-------+------+------+...
1264 0 4 8 12 16 20 24 28 32 36
bfc44f79 1265
bd81e77b
NC
1266 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1267 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1268 started out as zero once, but it's quite possible that it isn't. So now,
1269 rather than a nicely zeroed GP, you have it pointing somewhere random.
1270 Bugs ensue.
bfc44f79 1271
bd81e77b
NC
1272 (In fact, GP ends up pointing at a previous GP structure, because the
1273 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1274 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1275 this happens to be moot because XPVGV has been re-ordered, with GP
1276 no longer after STASH)
30f9da9e 1277
bd81e77b
NC
1278 So we are careful and work out the size of used parts of all the
1279 structures. */
bfc44f79 1280
bd81e77b
NC
1281 switch (old_type) {
1282 case SVt_NULL:
1283 break;
1284 case SVt_IV:
4df7f6af 1285 if (SvROK(sv)) {
ed7df46e 1286 referent = SvRV(sv);
238b27b3
NC
1287 old_type_details = &fake_rv;
1288 if (new_type == SVt_NV)
1289 new_type = SVt_PVNV;
4df7f6af
NC
1290 } else {
1291 if (new_type < SVt_PVIV) {
1292 new_type = (new_type == SVt_NV)
1293 ? SVt_PVNV : SVt_PVIV;
1294 }
bd81e77b
NC
1295 }
1296 break;
1297 case SVt_NV:
1298 if (new_type < SVt_PVNV) {
1299 new_type = SVt_PVNV;
bd81e77b
NC
1300 }
1301 break;
bd81e77b
NC
1302 case SVt_PV:
1303 assert(new_type > SVt_PV);
6d59e610
LM
1304 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1305 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
bd81e77b
NC
1306 break;
1307 case SVt_PVIV:
1308 break;
1309 case SVt_PVNV:
1310 break;
1311 case SVt_PVMG:
1312 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1313 there's no way that it can be safely upgraded, because perl.c
1314 expects to Safefree(SvANY(PL_mess_sv)) */
1315 assert(sv != PL_mess_sv);
bd81e77b
NC
1316 break;
1317 default:
2439e033 1318 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1319 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1320 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1321 }
3376de98 1322
2439e033 1323 if (UNLIKELY(old_type > new_type))
3376de98
NC
1324 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1325 (int)old_type, (int)new_type);
1326
2fa1109b 1327 new_type_details = bodies_by_type + new_type;
645c22ef 1328
bd81e77b
NC
1329 SvFLAGS(sv) &= ~SVTYPEMASK;
1330 SvFLAGS(sv) |= new_type;
932e9ff9 1331
ab4416c0
NC
1332 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1333 the return statements above will have triggered. */
1334 assert (new_type != SVt_NULL);
bd81e77b 1335 switch (new_type) {
bd81e77b
NC
1336 case SVt_IV:
1337 assert(old_type == SVt_NULL);
dc6369ef 1338 SET_SVANY_FOR_BODYLESS_IV(sv);
bd81e77b
NC
1339 SvIV_set(sv, 0);
1340 return;
1341 case SVt_NV:
1342 assert(old_type == SVt_NULL);
5b306eef 1343#if NVSIZE <= IVSIZE
dc6369ef 1344 SET_SVANY_FOR_BODYLESS_NV(sv);
5b306eef 1345#else
bd81e77b 1346 SvANY(sv) = new_XNV();
5b306eef 1347#endif
bd81e77b
NC
1348 SvNV_set(sv, 0);
1349 return;
bd81e77b 1350 case SVt_PVHV:
bd81e77b 1351 case SVt_PVAV:
d2a0f284 1352 assert(new_type_details->body_size);
c1ae03ae
NC
1353
1354#ifndef PURIFY
1355 assert(new_type_details->arena);
d2a0f284 1356 assert(new_type_details->arena_size);
c1ae03ae 1357 /* This points to the start of the allocated area. */
d2a0f284
JC
1358 new_body_inline(new_body, new_type);
1359 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1360 new_body = ((char *)new_body) - new_type_details->offset;
1361#else
1362 /* We always allocated the full length item with PURIFY. To do this
1363 we fake things so that arena is false for all 16 types.. */
1364 new_body = new_NOARENAZ(new_type_details);
1365#endif
1366 SvANY(sv) = new_body;
1367 if (new_type == SVt_PVAV) {
1368 AvMAX(sv) = -1;
1369 AvFILLp(sv) = -1;
1370 AvREAL_only(sv);
64484faa 1371 if (old_type_details->body_size) {
ac572bf4
NC
1372 AvALLOC(sv) = 0;
1373 } else {
1374 /* It will have been zeroed when the new body was allocated.
1375 Lets not write to it, in case it confuses a write-back
1376 cache. */
1377 }
78ac7dd9
NC
1378 } else {
1379 assert(!SvOK(sv));
1380 SvOK_off(sv);
1381#ifndef NODEFAULT_SHAREKEYS
1382 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1383#endif
586fc6a3
SM
1384 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1385 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1386 }
aeb18a1e 1387
bd81e77b
NC
1388 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1389 The target created by newSVrv also is, and it can have magic.
1390 However, it never has SvPVX set.
1391 */
4df7f6af
NC
1392 if (old_type == SVt_IV) {
1393 assert(!SvROK(sv));
1394 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1395 assert(SvPVX_const(sv) == 0);
1396 }
aeb18a1e 1397
bd81e77b 1398 if (old_type >= SVt_PVMG) {
e736a858 1399 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1400 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1401 } else {
1402 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1403 }
1404 break;
93e68bfb 1405
bd81e77b
NC
1406 case SVt_PVIV:
1407 /* XXX Is this still needed? Was it ever needed? Surely as there is
1408 no route from NV to PVIV, NOK can never be true */
1409 assert(!SvNOKp(sv));
1410 assert(!SvNOK(sv));
2b5060ae 1411 /* FALLTHROUGH */
bd81e77b
NC
1412 case SVt_PVIO:
1413 case SVt_PVFM:
bd81e77b
NC
1414 case SVt_PVGV:
1415 case SVt_PVCV:
1416 case SVt_PVLV:
d361b004 1417 case SVt_INVLIST:
12c45b25 1418 case SVt_REGEXP:
bd81e77b
NC
1419 case SVt_PVMG:
1420 case SVt_PVNV:
1421 case SVt_PV:
93e68bfb 1422
d2a0f284 1423 assert(new_type_details->body_size);
bd81e77b
NC
1424 /* We always allocated the full length item with PURIFY. To do this
1425 we fake things so that arena is false for all 16 types.. */
1426 if(new_type_details->arena) {
1427 /* This points to the start of the allocated area. */
d2a0f284
JC
1428 new_body_inline(new_body, new_type);
1429 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1430 new_body = ((char *)new_body) - new_type_details->offset;
1431 } else {
1432 new_body = new_NOARENAZ(new_type_details);
1433 }
1434 SvANY(sv) = new_body;
5e2fc214 1435
bd81e77b 1436 if (old_type_details->copy) {
f9ba3d20
NC
1437 /* There is now the potential for an upgrade from something without
1438 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1439 int offset = old_type_details->offset;
1440 int length = old_type_details->copy;
1441
1442 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1443 const int difference
f9ba3d20
NC
1444 = new_type_details->offset - old_type_details->offset;
1445 offset += difference;
1446 length -= difference;
1447 }
1448 assert (length >= 0);
1449
1450 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1451 char);
bd81e77b
NC
1452 }
1453
1454#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1455 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1456 * correct 0.0 for us. Otherwise, if the old body didn't have an
1457 * NV slot, but the new one does, then we need to initialise the
1458 * freshly created NV slot with whatever the correct bit pattern is
1459 * for 0.0 */
e22a937e
NC
1460 if (old_type_details->zero_nv && !new_type_details->zero_nv
1461 && !isGV_with_GP(sv))
bd81e77b 1462 SvNV_set(sv, 0);
82048762 1463#endif
5e2fc214 1464
2439e033 1465 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1466 IO * const io = MUTABLE_IO(sv);
d963bf01 1467 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1468
1469 SvOBJECT_on(io);
1470 /* Clear the stashcache because a new IO could overrule a package
1471 name */
103f5a36 1472 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1473 hv_clear(PL_stashcache);
1474
85dca89a 1475 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1476 IoPAGE_LEN(sv) = 60;
85dca89a 1477 }
df6b4bd5 1478 if (old_type < SVt_PV) {
ed7df46e 1479 /* referent will be NULL unless the old type was SVt_IV emulating
4df7f6af 1480 SVt_RV */
ed7df46e 1481 sv->sv_u.svu_rv = referent;
4df7f6af 1482 }
bd81e77b
NC
1483 break;
1484 default:
afd78fd5
JH
1485 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1486 (unsigned long)new_type);
bd81e77b 1487 }
73171d91 1488
5b306eef
DD
1489 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1490 and sometimes SVt_NV */
1491 if (old_type_details->body_size) {
bd81e77b 1492#ifdef PURIFY
beeec492 1493 safefree(old_body);
bd81e77b 1494#else
bc786448
GG
1495 /* Note that there is an assumption that all bodies of types that
1496 can be upgraded came from arenas. Only the more complex non-
1497 upgradable types are allowed to be directly malloc()ed. */
1498 assert(old_type_details->arena);
bd81e77b
NC
1499 del_body((void*)((char*)old_body + old_type_details->offset),
1500 &PL_body_roots[old_type]);
1501#endif
1502 }
1503}
73171d91 1504
bd81e77b
NC
1505/*
1506=for apidoc sv_backoff
73171d91 1507
fde67290 1508Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1509wrapper instead.
73171d91 1510
bd81e77b 1511=cut
73171d91
NC
1512*/
1513
fa7a1e49
DD
1514/* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1515 prior to 5.23.4 this function always returned 0
1516*/
1517
1518void
ddeaf645 1519Perl_sv_backoff(SV *const sv)
bd81e77b 1520{
69240efd 1521 STRLEN delta;
7a4bba22 1522 const char * const s = SvPVX_const(sv);
7918f24d
NC
1523
1524 PERL_ARGS_ASSERT_SV_BACKOFF;
7918f24d 1525
bd81e77b
NC
1526 assert(SvOOK(sv));
1527 assert(SvTYPE(sv) != SVt_PVHV);
1528 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1529
69240efd
NC
1530 SvOOK_offset(sv, delta);
1531
7a4bba22
NC
1532 SvLEN_set(sv, SvLEN(sv) + delta);
1533 SvPV_set(sv, SvPVX(sv) - delta);
bd81e77b 1534 SvFLAGS(sv) &= ~SVf_OOK;
fa7a1e49
DD
1535 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1536 return;
bd81e77b 1537}
73171d91 1538
03885497
DM
1539
1540/* forward declaration */
1541static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1542
1543
bd81e77b
NC
1544/*
1545=for apidoc sv_grow
73171d91 1546
bd81e77b
NC
1547Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1548upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1549Use the C<SvGROW> wrapper instead.
93e68bfb 1550
bd81e77b
NC
1551=cut
1552*/
93e68bfb 1553
e0060e30 1554
bd81e77b 1555char *
5aaab254 1556Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1557{
eb578fdb 1558 char *s;
93e68bfb 1559
7918f24d
NC
1560 PERL_ARGS_ASSERT_SV_GROW;
1561
bd81e77b
NC
1562 if (SvROK(sv))
1563 sv_unref(sv);
1564 if (SvTYPE(sv) < SVt_PV) {
1565 sv_upgrade(sv, SVt_PV);
1566 s = SvPVX_mutable(sv);
1567 }
1568 else if (SvOOK(sv)) { /* pv is offset? */
1569 sv_backoff(sv);
1570 s = SvPVX_mutable(sv);
1571 if (newlen > SvLEN(sv))
1572 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
bd81e77b
NC
1573 }
1574 else
db2c6cb3 1575 {
e0060e30 1576 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1577 s = SvPVX_mutable(sv);
db2c6cb3 1578 }
aeb18a1e 1579
93c10d60 1580#ifdef PERL_COPY_ON_WRITE
cbcb2a16 1581 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
3c239bea 1582 * to store the COW count. So in general, allocate one more byte than
cbcb2a16
DM
1583 * asked for, to make it likely this byte is always spare: and thus
1584 * make more strings COW-able.
fe546b38 1585 *
fa8f4f85
TC
1586 * Only increment if the allocation isn't MEM_SIZE_MAX,
1587 * otherwise it will wrap to 0.
1588 */
fe546b38 1589 if ( newlen != MEM_SIZE_MAX )
cbcb2a16
DM
1590 newlen++;
1591#endif
1592
ce861ea7
YO
1593#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1594#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1595#endif
1596
bd81e77b 1597 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559 1598 STRLEN minlen = SvCUR(sv);
3c239bea 1599 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
f1200559
WH
1600 if (newlen < minlen)
1601 newlen = minlen;
ce861ea7 1602#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
7c641603
KW
1603
1604 /* Don't round up on the first allocation, as odds are pretty good that
1605 * the initial request is accurate as to what is really needed */
ce861ea7 1606 if (SvLEN(sv)) {
9efda33a
TC
1607 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1608 if (rounded > newlen)
1609 newlen = rounded;
ce861ea7 1610 }
bd81e77b 1611#endif
98653f18 1612 if (SvLEN(sv) && s) {
10edeb5d 1613 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1614 }
1615 else {
10edeb5d 1616 s = (char*)safemalloc(newlen);
bd81e77b 1617 if (SvPVX_const(sv) && SvCUR(sv)) {
0a5fcc38 1618 Move(SvPVX_const(sv), s, SvCUR(sv), char);
bd81e77b
NC
1619 }
1620 }
1621 SvPV_set(sv, s);
ce861ea7 1622#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
98653f18
NC
1623 /* Do this here, do it once, do it right, and then we will never get
1624 called back into sv_grow() unless there really is some growing
1625 needed. */
ca7c1a29 1626 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1627#else
bd81e77b 1628 SvLEN_set(sv, newlen);
98653f18 1629#endif
bd81e77b
NC
1630 }
1631 return s;
1632}
aeb18a1e 1633
bd81e77b
NC
1634/*
1635=for apidoc sv_setiv
932e9ff9 1636
bd81e77b 1637Copies an integer into the given SV, upgrading first if necessary.
fbe13c60 1638Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
463ee0b2 1639
bd81e77b
NC
1640=cut
1641*/
463ee0b2 1642
bd81e77b 1643void
5aaab254 1644Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1645{
7918f24d
NC
1646 PERL_ARGS_ASSERT_SV_SETIV;
1647
bd81e77b
NC
1648 SV_CHECK_THINKFIRST_COW_DROP(sv);
1649 switch (SvTYPE(sv)) {
1650 case SVt_NULL:
bd81e77b 1651 case SVt_NV:
3376de98 1652 sv_upgrade(sv, SVt_IV);
bd81e77b 1653 break;
bd81e77b
NC
1654 case SVt_PV:
1655 sv_upgrade(sv, SVt_PVIV);
1656 break;
463ee0b2 1657
bd81e77b 1658 case SVt_PVGV:
6e592b3a
BM
1659 if (!isGV_with_GP(sv))
1660 break;
2165bd23 1661 /* FALLTHROUGH */
bd81e77b
NC
1662 case SVt_PVAV:
1663 case SVt_PVHV:
1664 case SVt_PVCV:
1665 case SVt_PVFM:
1666 case SVt_PVIO:
22e74366 1667 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1668 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1669 OP_DESC(PL_op));
c9a0dcdc 1670 NOT_REACHED; /* NOTREACHED */
0103ca14 1671 break;
42d0e0b7 1672 default: NOOP;
bd81e77b
NC
1673 }
1674 (void)SvIOK_only(sv); /* validate number */
1675 SvIV_set(sv, i);
1676 SvTAINT(sv);
1677}
932e9ff9 1678
bd81e77b
NC
1679/*
1680=for apidoc sv_setiv_mg
d33b2eba 1681
bd81e77b 1682Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1683
bd81e77b
NC
1684=cut
1685*/
d33b2eba 1686
bd81e77b 1687void
5aaab254 1688Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1689{
7918f24d
NC
1690 PERL_ARGS_ASSERT_SV_SETIV_MG;
1691
bd81e77b
NC
1692 sv_setiv(sv,i);
1693 SvSETMAGIC(sv);
1694}
727879eb 1695
bd81e77b
NC
1696/*
1697=for apidoc sv_setuv
d33b2eba 1698
bd81e77b 1699Copies an unsigned integer into the given SV, upgrading first if necessary.
fbe13c60 1700Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
9b94d1dd 1701
bd81e77b
NC
1702=cut
1703*/
d33b2eba 1704
bd81e77b 1705void
5aaab254 1706Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1707{
7918f24d
NC
1708 PERL_ARGS_ASSERT_SV_SETUV;
1709
013abb9b
NC
1710 /* With the if statement to ensure that integers are stored as IVs whenever
1711 possible:
bd81e77b 1712 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1713
bd81e77b
NC
1714 without
1715 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1716
013abb9b
NC
1717 If you wish to remove the following if statement, so that this routine
1718 (and its callers) always return UVs, please benchmark to see what the
1719 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1720 */
1721 if (u <= (UV)IV_MAX) {
1722 sv_setiv(sv, (IV)u);
1723 return;
1724 }
1725 sv_setiv(sv, 0);
1726 SvIsUV_on(sv);
1727 SvUV_set(sv, u);
1728}
d33b2eba 1729
bd81e77b
NC
1730/*
1731=for apidoc sv_setuv_mg
727879eb 1732
bd81e77b 1733Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1734
bd81e77b
NC
1735=cut
1736*/
5e2fc214 1737
bd81e77b 1738void
5aaab254 1739Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1740{
7918f24d
NC
1741 PERL_ARGS_ASSERT_SV_SETUV_MG;
1742
bd81e77b
NC
1743 sv_setuv(sv,u);
1744 SvSETMAGIC(sv);
1745}
5e2fc214 1746
954c1994 1747/*
bd81e77b 1748=for apidoc sv_setnv
954c1994 1749
bd81e77b 1750Copies a double into the given SV, upgrading first if necessary.
fbe13c60 1751Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
954c1994
GS
1752
1753=cut
1754*/
1755
63f97190 1756void
5aaab254 1757Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1758{
7918f24d
NC
1759 PERL_ARGS_ASSERT_SV_SETNV;
1760
bd81e77b
NC
1761 SV_CHECK_THINKFIRST_COW_DROP(sv);
1762 switch (SvTYPE(sv)) {
79072805 1763 case SVt_NULL:
79072805 1764 case SVt_IV:
bd81e77b 1765 sv_upgrade(sv, SVt_NV);
79072805
LW
1766 break;
1767 case SVt_PV:
79072805 1768 case SVt_PVIV:
bd81e77b 1769 sv_upgrade(sv, SVt_PVNV);
79072805 1770 break;
bd4b1eb5 1771
bd4b1eb5 1772 case SVt_PVGV:
6e592b3a
BM
1773 if (!isGV_with_GP(sv))
1774 break;
2165bd23 1775 /* FALLTHROUGH */
bd81e77b
NC
1776 case SVt_PVAV:
1777 case SVt_PVHV:
79072805 1778 case SVt_PVCV:
bd81e77b
NC
1779 case SVt_PVFM:
1780 case SVt_PVIO:
22e74366 1781 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1782 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1783 OP_DESC(PL_op));
c9a0dcdc 1784 NOT_REACHED; /* NOTREACHED */
0103ca14 1785 break;
42d0e0b7 1786 default: NOOP;
2068cd4d 1787 }
bd81e77b
NC
1788 SvNV_set(sv, num);
1789 (void)SvNOK_only(sv); /* validate number */
1790 SvTAINT(sv);
79072805
LW
1791}
1792
645c22ef 1793/*
bd81e77b 1794=for apidoc sv_setnv_mg
645c22ef 1795
bd81e77b 1796Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1797
1798=cut
1799*/
1800
bd81e77b 1801void
5aaab254 1802Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1803{
7918f24d
NC
1804 PERL_ARGS_ASSERT_SV_SETNV_MG;
1805
bd81e77b
NC
1806 sv_setnv(sv,num);
1807 SvSETMAGIC(sv);
79072805
LW
1808}
1809
3f7602fa
TC
1810/* Return a cleaned-up, printable version of sv, for non-numeric, or
1811 * not incrementable warning display.
1812 * Originally part of S_not_a_number().
1813 * The return value may be != tmpbuf.
bd81e77b 1814 */
954c1994 1815
3f7602fa
TC
1816STATIC const char *
1817S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1818 const char *pv;
94463019 1819
3f7602fa 1820 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1821
94463019 1822 if (DO_UTF8(sv)) {
3f7602fa 1823 SV *dsv = newSVpvs_flags("", SVs_TEMP);
37b8cdd1 1824 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
94463019
JH
1825 } else {
1826 char *d = tmpbuf;
3f7602fa 1827 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1828 /* each *s can expand to 4 chars + "...\0",
1829 i.e. need room for 8 chars */
ecdeb87c 1830
00b6aa41
AL
1831 const char *s = SvPVX_const(sv);
1832 const char * const end = s + SvCUR(sv);
1833 for ( ; s < end && d < limit; s++ ) {
94463019 1834 int ch = *s & 0xFF;
bd27cf70 1835 if (! isASCII(ch) && !isPRINT_LC(ch)) {
94463019
JH
1836 *d++ = 'M';
1837 *d++ = '-';
bd27cf70
KW
1838
1839 /* Map to ASCII "equivalent" of Latin1 */
1840 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
94463019
JH
1841 }
1842 if (ch == '\n') {
1843 *d++ = '\\';
1844 *d++ = 'n';
1845 }
1846 else if (ch == '\r') {
1847 *d++ = '\\';
1848 *d++ = 'r';
1849 }
1850 else if (ch == '\f') {
1851 *d++ = '\\';
1852 *d++ = 'f';
1853 }
1854 else if (ch == '\\') {
1855 *d++ = '\\';
1856 *d++ = '\\';
1857 }
1858 else if (ch == '\0') {
1859 *d++ = '\\';
1860 *d++ = '0';
1861 }
1862 else if (isPRINT_LC(ch))
1863 *d++ = ch;
1864 else {
1865 *d++ = '^';
1866 *d++ = toCTRL(ch);
1867 }
1868 }
1869 if (s < end) {
1870 *d++ = '.';
1871 *d++ = '.';
1872 *d++ = '.';
1873 }
1874 *d = '\0';
1875 pv = tmpbuf;
a0d0e21e 1876 }
a0d0e21e 1877
3f7602fa
TC
1878 return pv;
1879}
1880
1881/* Print an "isn't numeric" warning, using a cleaned-up,
1882 * printable version of the offending string
1883 */
1884
1885STATIC void
1886S_not_a_number(pTHX_ SV *const sv)
1887{
3f7602fa
TC
1888 char tmpbuf[64];
1889 const char *pv;
1890
1891 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1892
1893 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1894
533c011a 1895 if (PL_op)
9014280d 1896 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1897 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1898 "Argument \"%s\" isn't numeric in %s", pv,
1899 OP_DESC(PL_op));
a0d0e21e 1900 else
9014280d 1901 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1902 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1903 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1904}
1905
3f7602fa
TC
1906STATIC void
1907S_not_incrementable(pTHX_ SV *const sv) {
3f7602fa
TC
1908 char tmpbuf[64];
1909 const char *pv;
1910
1911 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1912
1913 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1914
1915 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1916 "Argument \"%s\" treated as 0 in increment (++)", pv);
1917}
1918
c2988b20
NC
1919/*
1920=for apidoc looks_like_number
1921
645c22ef
DM
1922Test if the content of an SV looks like a number (or is a number).
1923C<Inf> and C<Infinity> are treated as numbers (so will not issue a
796b6530 1924non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
f52e41ad 1925ignored.
c2988b20
NC
1926
1927=cut
1928*/
1929
1930I32
aad570aa 1931Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1932{
eb578fdb 1933 const char *sbegin;
c2988b20 1934 STRLEN len;
ea2485eb 1935 int numtype;
c2988b20 1936
7918f24d
NC
1937 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1938
f52e41ad
FC
1939 if (SvPOK(sv) || SvPOKp(sv)) {
1940 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1941 }
c2988b20 1942 else
e0ab1c0e 1943 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
ea2485eb
JH
1944 numtype = grok_number(sbegin, len, NULL);
1945 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
c2988b20 1946}
25da4f38 1947
19f6321d
NC
1948STATIC bool
1949S_glob_2number(pTHX_ GV * const gv)
180488f8 1950{
7918f24d
NC
1951 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1952
675c862f
AL
1953 /* We know that all GVs stringify to something that is not-a-number,
1954 so no need to test that. */
1955 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1956 {
1957 SV *const buffer = sv_newmortal();
1958 gv_efullname3(buffer, gv, "*");
675c862f 1959 not_a_number(buffer);
8e629ff4 1960 }
675c862f
AL
1961 /* We just want something true to return, so that S_sv_2iuv_common
1962 can tail call us and return true. */
19f6321d 1963 return TRUE;
675c862f
AL
1964}
1965
25da4f38
IZ
1966/* Actually, ISO C leaves conversion of UV to IV undefined, but
1967 until proven guilty, assume that things are not that bad... */
1968
645c22ef
DM
1969/*
1970 NV_PRESERVES_UV:
1971
1972 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1973 an IV (an assumption perl has been based on to date) it becomes necessary
1974 to remove the assumption that the NV always carries enough precision to
1975 recreate the IV whenever needed, and that the NV is the canonical form.
1976 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1977 precision as a side effect of conversion (which would lead to insanity
28e5dec8 1978 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
8a4a3196
KW
1979 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1980 where precision was lost, and IV/UV/NV slots that have a valid conversion
1981 which has lost no precision
645c22ef 1982 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1983 would lose precision, the precise conversion (or differently
1984 imprecise conversion) is also performed and cached, to prevent
1985 requests for different numeric formats on the same SV causing
1986 lossy conversion chains. (lossless conversion chains are perfectly
1987 acceptable (still))
1988
1989
1990 flags are used:
1991 SvIOKp is true if the IV slot contains a valid value
1992 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1993 SvNOKp is true if the NV slot contains a valid value
1994 SvNOK is true only if the NV value is accurate
1995
1996 so
645c22ef 1997 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1998 IV(or UV) would lose accuracy over a direct conversion from PV to
1999 IV(or UV). If it would, cache both conversions, return NV, but mark
2000 SV as IOK NOKp (ie not NOK).
2001
645c22ef 2002 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
2003 NV would lose accuracy over a direct conversion from PV to NV. If it
2004 would, cache both conversions, flag similarly.
2005
2006 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2007 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
2008 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2009 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 2010 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 2011
645c22ef
DM
2012 The benefit of this is that operations such as pp_add know that if
2013 SvIOK is true for both left and right operands, then integer addition
2014 can be used instead of floating point (for cases where the result won't
2015 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
2016 loss of precision compared with integer addition.
2017
2018 * making IV and NV equal status should make maths accurate on 64 bit
2019 platforms
2020 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 2021 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
2022 looking for SvIOK and checking for overflow will not outweigh the
2023 fp to integer speedup)
2024 * will slow down integer operations (callers of SvIV) on "inaccurate"
2025 values, as the change from SvIOK to SvIOKp will cause a call into
2026 sv_2iv each time rather than a macro access direct to the IV slot
2027 * should speed up number->string conversion on integers as IV is
645c22ef 2028 favoured when IV and NV are equally accurate
28e5dec8
JH
2029
2030 ####################################################################
645c22ef
DM
2031 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2032 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2033 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
2034 ####################################################################
2035
645c22ef 2036 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
2037 performance ratio.
2038*/
2039
2040#ifndef NV_PRESERVES_UV
645c22ef
DM
2041# define IS_NUMBER_UNDERFLOW_IV 1
2042# define IS_NUMBER_UNDERFLOW_UV 2
2043# define IS_NUMBER_IV_AND_UV 2
2044# define IS_NUMBER_OVERFLOW_IV 4
2045# define IS_NUMBER_OVERFLOW_UV 5
2046
2047/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
2048
2049/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2050STATIC int
5aaab254 2051S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
2052# ifdef DEBUGGING
2053 , I32 numtype
2054# endif
2055 )
28e5dec8 2056{
7918f24d 2057 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
23491f1d 2058 PERL_UNUSED_CONTEXT;
7918f24d 2059
147e3846 2060 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
2061 if (SvNVX(sv) < (NV)IV_MIN) {
2062 (void)SvIOKp_on(sv);
2063 (void)SvNOK_on(sv);
45977657 2064 SvIV_set(sv, IV_MIN);
28e5dec8
JH
2065 return IS_NUMBER_UNDERFLOW_IV;
2066 }
2067 if (SvNVX(sv) > (NV)UV_MAX) {
2068 (void)SvIOKp_on(sv);
2069 (void)SvNOK_on(sv);
2070 SvIsUV_on(sv);
607fa7f2 2071 SvUV_set(sv, UV_MAX);
28e5dec8
JH
2072 return IS_NUMBER_OVERFLOW_UV;
2073 }
c2988b20
NC
2074 (void)SvIOKp_on(sv);
2075 (void)SvNOK_on(sv);
2076 /* Can't use strtol etc to convert this string. (See truth table in
2077 sv_2iv */
2078 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 2079 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2080 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2081 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2082 } else {
2083 /* Integer is imprecise. NOK, IOKp */
2084 }
2085 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2086 }
2087 SvIsUV_on(sv);
607fa7f2 2088 SvUV_set(sv, U_V(SvNVX(sv)));
659c4b96 2089 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2090 if (SvUVX(sv) == UV_MAX) {
2091 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2092 possibly be preserved by NV. Hence, it must be overflow.
2093 NOK, IOKp */
2094 return IS_NUMBER_OVERFLOW_UV;
2095 }
2096 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2097 } else {
2098 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2099 }
c2988b20 2100 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2101}
645c22ef
DM
2102#endif /* !NV_PRESERVES_UV*/
2103
a13f4dff 2104/* If numtype is infnan, set the NV of the sv accordingly.
5564cd7f 2105 * If numtype is anything else, try setting the NV using Atof(PV). */
a13f4dff 2106static void
3823048b 2107S_sv_setnv(pTHX_ SV* sv, int numtype)
a13f4dff 2108{
07925c5e 2109 bool pok = cBOOL(SvPOK(sv));
5564cd7f 2110 bool nok = FALSE;
a7157111 2111#ifdef NV_INF
a13f4dff
JH
2112 if ((numtype & IS_NUMBER_INFINITY)) {
2113 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
5564cd7f 2114 nok = TRUE;
a7157111
JH
2115 } else
2116#endif
2117#ifdef NV_NAN
2118 if ((numtype & IS_NUMBER_NAN)) {
3823048b 2119 SvNV_set(sv, NV_NAN);
d48bd569 2120 nok = TRUE;
a7157111
JH
2121 } else
2122#endif
2123 if (pok) {
a13f4dff 2124 SvNV_set(sv, Atof(SvPVX_const(sv)));
d48bd569
JH
2125 /* Purposefully no true nok here, since we don't want to blow
2126 * away the possible IOK/UV of an existing sv. */
2127 }
5564cd7f 2128 if (nok) {
d48bd569 2129 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
5564cd7f
JH
2130 if (pok)
2131 SvPOK_on(sv); /* PV is okay, though. */
2132 }
a13f4dff
JH
2133}
2134
af359546 2135STATIC bool
7918f24d
NC
2136S_sv_2iuv_common(pTHX_ SV *const sv)
2137{
7918f24d
NC
2138 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2139
af359546 2140 if (SvNOKp(sv)) {
28e5dec8
JH
2141 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2142 * without also getting a cached IV/UV from it at the same time
2143 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2144 * IV or UV at same time to avoid this. */
2145 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2146
2147 if (SvTYPE(sv) == SVt_NV)
2148 sv_upgrade(sv, SVt_PVNV);
2149
28e5dec8
JH
2150 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2151 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2152 certainly cast into the IV range at IV_MAX, whereas the correct
2153 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2154 cases go to UV */
e91de695
JH
2155#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2156 if (Perl_isnan(SvNVX(sv))) {
2157 SvUV_set(sv, 0);
2158 SvIsUV_on(sv);
2159 return FALSE;
2160 }
2161#endif
28e5dec8 2162 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2163 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2164 if (SvNVX(sv) == (NV) SvIVX(sv)
28e5dec8 2165#ifndef NV_PRESERVES_UV
53e2bfb7 2166 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
28e5dec8
JH
2167 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2168 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2169 /* Don't flag it as "accurately an integer" if the number
2170 came from a (by definition imprecise) NV operation, and
2171 we're outside the range of NV integer precision */
2172#endif
2173 ) {
a43d94f2
NC
2174 if (SvNOK(sv))
2175 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2176 else {
2177 /* scalar has trailing garbage, eg "42a" */
2178 }
28e5dec8 2179 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2180 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
28e5dec8
JH
2181 PTR2UV(sv),
2182 SvNVX(sv),
2183 SvIVX(sv)));
2184
2185 } else {
2186 /* IV not precise. No need to convert from PV, as NV
2187 conversion would already have cached IV if it detected
2188 that PV->IV would be better than PV->NV->IV
2189 flags already correct - don't set public IOK. */
2190 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2191 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
28e5dec8
JH
2192 PTR2UV(sv),
2193 SvNVX(sv),
2194 SvIVX(sv)));
2195 }
2196 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2197 but the cast (NV)IV_MIN rounds to a the value less (more
2198 negative) than IV_MIN which happens to be equal to SvNVX ??
2199 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2200 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2201 (NV)UVX == NVX are both true, but the values differ. :-(
2202 Hopefully for 2s complement IV_MIN is something like
2203 0x8000000000000000 which will be exact. NWC */
d460ef45 2204 }
25da4f38 2205 else {
607fa7f2 2206 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8 2207 if (
659c4b96 2208 (SvNVX(sv) == (NV) SvUVX(sv))
28e5dec8
JH
2209#ifndef NV_PRESERVES_UV
2210 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2211 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2212 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2213 /* Don't flag it as "accurately an integer" if the number
2214 came from a (by definition imprecise) NV operation, and
2215 we're outside the range of NV integer precision */
2216#endif
a43d94f2 2217 && SvNOK(sv)
28e5dec8
JH
2218 )
2219 SvIOK_on(sv);
25da4f38 2220 SvIsUV_on(sv);
1c846c1f 2221 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2222 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
56431972 2223 PTR2UV(sv),
57def98f
JH
2224 SvUVX(sv),
2225 SvUVX(sv)));
25da4f38 2226 }
748a9306 2227 }
cd84013a 2228 else if (SvPOKp(sv)) {
c2988b20 2229 UV value;
80e5abf2
DM
2230 int numtype;
2231 const char *s = SvPVX_const(sv);
2232 const STRLEN cur = SvCUR(sv);
2233
2234 /* short-cut for a single digit string like "1" */
2235
2236 if (cur == 1) {
2237 char c = *s;
2238 if (isDIGIT(c)) {
2239 if (SvTYPE(sv) < SVt_PVIV)
2240 sv_upgrade(sv, SVt_PVIV);
2241 (void)SvIOK_on(sv);
2242 SvIV_set(sv, (IV)(c - '0'));
2243 return FALSE;
2244 }
2245 }
2246
2247 numtype = grok_number(s, cur, &value);
af359546 2248 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2249 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2250 the same as the direct translation of the initial string
2251 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2252 be careful to ensure that the value with the .456 is around if the
2253 NV value is requested in the future).
1c846c1f 2254
af359546 2255 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2256 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2257 cache the NV if we are sure it's not needed.
25da4f38 2258 */
16b7a9a4 2259
c2988b20
NC
2260 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2261 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2262 == IS_NUMBER_IN_UV) {
5e045b90 2263 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2264 if (SvTYPE(sv) < SVt_PVIV)
2265 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2266 (void)SvIOK_on(sv);
c2988b20
NC
2267 } else if (SvTYPE(sv) < SVt_PVNV)
2268 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2269
a13f4dff 2270 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
75a57a38 2271 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
6b322424 2272 not_a_number(sv);
3823048b 2273 S_sv_setnv(aTHX_ sv, numtype);
a13f4dff
JH
2274 return FALSE;
2275 }
2276
f2524eef 2277 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2278 we aren't going to call atof() below. If NVs don't preserve UVs
2279 then the value returned may have more precision than atof() will
2280 return, even though value isn't perfectly accurate. */
2281 if ((numtype & (IS_NUMBER_IN_UV
2282#ifdef NV_PRESERVES_UV
2283 | IS_NUMBER_NOT_INT
2284#endif
2285 )) == IS_NUMBER_IN_UV) {
2286 /* This won't turn off the public IOK flag if it was set above */
2287 (void)SvIOKp_on(sv);
2288
2289 if (!(numtype & IS_NUMBER_NEG)) {
2290 /* positive */;
2291 if (value <= (UV)IV_MAX) {
45977657 2292 SvIV_set(sv, (IV)value);
c2988b20 2293 } else {
af359546 2294 /* it didn't overflow, and it was positive. */
607fa7f2 2295 SvUV_set(sv, value);
c2988b20
NC
2296 SvIsUV_on(sv);
2297 }
2298 } else {
2299 /* 2s complement assumption */
2300 if (value <= (UV)IV_MIN) {
53e2bfb7
DM
2301 SvIV_set(sv, value == (UV)IV_MIN
2302 ? IV_MIN : -(IV)value);
c2988b20
NC
2303 } else {
2304 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2305 I'm assuming it will be rare. */
c2988b20
NC
2306 if (SvTYPE(sv) < SVt_PVNV)
2307 sv_upgrade(sv, SVt_PVNV);
2308 SvNOK_on(sv);
2309 SvIOK_off(sv);
2310 SvIOKp_on(sv);
9d6ce603 2311 SvNV_set(sv, -(NV)value);
45977657 2312 SvIV_set(sv, IV_MIN);
c2988b20
NC
2313 }
2314 }
2315 }
2316 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2317 will be in the previous block to set the IV slot, and the next
2318 block to set the NV slot. So no else here. */
2319
2320 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2321 != IS_NUMBER_IN_UV) {
2322 /* It wasn't an (integer that doesn't overflow the UV). */
3823048b 2323 S_sv_setnv(aTHX_ sv, numtype);
28e5dec8 2324
c2988b20
NC
2325 if (! numtype && ckWARN(WARN_NUMERIC))
2326 not_a_number(sv);
28e5dec8 2327
147e3846 2328 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
c2988b20 2329 PTR2UV(sv), SvNVX(sv)));
28e5dec8 2330
28e5dec8 2331#ifdef NV_PRESERVES_UV
af359546
NC
2332 (void)SvIOKp_on(sv);
2333 (void)SvNOK_on(sv);
e91de695
JH
2334#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2335 if (Perl_isnan(SvNVX(sv))) {
2336 SvUV_set(sv, 0);
2337 SvIsUV_on(sv);
2338 return FALSE;
2339 }
2340#endif
af359546
NC
2341 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2342 SvIV_set(sv, I_V(SvNVX(sv)));
2343 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2344 SvIOK_on(sv);
2345 } else {
6f207bd3 2346 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2347 }
2348 /* UV will not work better than IV */
2349 } else {
2350 if (SvNVX(sv) > (NV)UV_MAX) {
2351 SvIsUV_on(sv);
2352 /* Integer is inaccurate. NOK, IOKp, is UV */
2353 SvUV_set(sv, UV_MAX);
af359546
NC
2354 } else {
2355 SvUV_set(sv, U_V(SvNVX(sv)));
2356 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2357 NV preservse UV so can do correct comparison. */
2358 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2359 SvIOK_on(sv);
af359546 2360 } else {
6f207bd3 2361 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2362 }
2363 }
4b0c9573 2364 SvIsUV_on(sv);
af359546 2365 }
28e5dec8 2366#else /* NV_PRESERVES_UV */
c2988b20
NC
2367 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2368 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2369 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2370 grok_number above. The NV slot has just been set using
2371 Atof. */
560b0c46 2372 SvNOK_on(sv);
c2988b20
NC
2373 assert (SvIOKp(sv));
2374 } else {
2375 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2376 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2377 /* Small enough to preserve all bits. */
2378 (void)SvIOKp_on(sv);
2379 SvNOK_on(sv);
45977657 2380 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2381 if ((NV)(SvIVX(sv)) == SvNVX(sv))
c2988b20
NC
2382 SvIOK_on(sv);
2383 /* Assumption: first non-preserved integer is < IV_MAX,
2384 this NV is in the preserved range, therefore: */
2385 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2386 < (UV)IV_MAX)) {
147e3846 2387 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
2388 }
2389 } else {
2390 /* IN_UV NOT_INT
2391 0 0 already failed to read UV.
2392 0 1 already failed to read UV.
2393 1 0 you won't get here in this case. IV/UV
2394 slot set, public IOK, Atof() unneeded.
2395 1 1 already read UV.
2396 so there's no point in sv_2iuv_non_preserve() attempting
2397 to use atol, strtol, strtoul etc. */
47031da6 2398# ifdef DEBUGGING
40a17c4c 2399 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2400# else
2401 sv_2iuv_non_preserve (sv);
2402# endif
c2988b20
NC
2403 }
2404 }
28e5dec8 2405#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2406 /* It might be more code efficient to go through the entire logic above
2407 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2408 gets complex and potentially buggy, so more programmer efficient
2409 to do it this way, by turning off the public flags: */
2410 if (!numtype)
2411 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2412 }
af359546 2413 }
c0443cc0 2414 else {
675c862f 2415 if (isGV_with_GP(sv))
159b6efe 2416 return glob_2number(MUTABLE_GV(sv));
180488f8 2417
4f62cd62 2418 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
af359546 2419 report_uninit(sv);
25da4f38
IZ
2420 if (SvTYPE(sv) < SVt_IV)
2421 /* Typically the caller expects that sv_any is not NULL now. */
2422 sv_upgrade(sv, SVt_IV);
af359546
NC
2423 /* Return 0 from the caller. */
2424 return TRUE;
2425 }
2426 return FALSE;
2427}
2428
2429/*
2430=for apidoc sv_2iv_flags
2431
2432Return the integer value of an SV, doing any necessary string
c5608a1f 2433conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
af359546
NC
2434Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2435
2436=cut
2437*/
2438
2439IV
5aaab254 2440Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2441{
1061065f 2442 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
4bac9ae4 2443
217f6fa3
FC
2444 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2445 && SvTYPE(sv) != SVt_PVFM);
2446
4bac9ae4
CS
2447 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2448 mg_get(sv);
2449
2450 if (SvROK(sv)) {
2451 if (SvAMAGIC(sv)) {
2452 SV * tmpstr;
2453 if (flags & SV_SKIP_OVERLOAD)
2454 return 0;
2455 tmpstr = AMG_CALLunary(sv, numer_amg);
2456 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2457 return SvIV(tmpstr);
2458 }
2459 }
2460 return PTR2IV(SvRV(sv));
2461 }
2462
8d919b0a 2463 if (SvVALID(sv) || isREGEXP(sv)) {
4e8879f3
DM
2464 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2465 must not let them cache IVs.
2b2b6d6d
NC
2466 In practice they are extremely unlikely to actually get anywhere
2467 accessible by user Perl code - the only way that I'm aware of is when
2468 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2469
2470 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2471 */
df6b4bd5 2472 assert(SvPOKp(sv));
e20b6c3b 2473 {
71c558c3 2474 UV value;
8d919b0a
FC
2475 const char * const ptr =
2476 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2477 const int numtype
2478 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2479
2480 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2481 == IS_NUMBER_IN_UV) {
2482 /* It's definitely an integer */
2483 if (numtype & IS_NUMBER_NEG) {
2484 if (value < (UV)IV_MIN)
2485 return -(IV)value;
2486 } else {
2487 if (value < (UV)IV_MAX)
2488 return (IV)value;
2489 }
2490 }
058b8ae2 2491
e91de695
JH
2492 /* Quite wrong but no good choices. */
2493 if ((numtype & IS_NUMBER_INFINITY)) {
2494 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2495 } else if ((numtype & IS_NUMBER_NAN)) {
2496 return 0; /* So wrong. */
2497 }
2498
71c558c3
NC
2499 if (!numtype) {
2500 if (ckWARN(WARN_NUMERIC))
2501 not_a_number(sv);
2502 }
8d919b0a 2503 return I_V(Atof(ptr));
e20b6c3b 2504 }
4bac9ae4
CS
2505 }
2506
2507 if (SvTHINKFIRST(sv)) {
af359546
NC
2508 if (SvREADONLY(sv) && !SvOK(sv)) {
2509 if (ckWARN(WARN_UNINITIALIZED))
2510 report_uninit(sv);
2511 return 0;
2512 }
2513 }
4bac9ae4 2514
af359546
NC
2515 if (!SvIOKp(sv)) {
2516 if (S_sv_2iuv_common(aTHX_ sv))
2517 return 0;
79072805 2518 }
4bac9ae4 2519
147e3846 2520 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
1d7c1841 2521 PTR2UV(sv),SvIVX(sv)));
25da4f38 2522 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2523}
2524
645c22ef 2525/*
891f9566 2526=for apidoc sv_2uv_flags
645c22ef
DM
2527
2528Return the unsigned integer value of an SV, doing any necessary string
c5608a1f 2529conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
891f9566 2530Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef 2531
58ae9391
KW
2532=for apidoc Amnh||SV_GMAGIC
2533
645c22ef
DM
2534=cut
2535*/
2536
ff68c719 2537UV
5aaab254 2538Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2539{
1061065f 2540 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
4bac9ae4
CS
2541
2542 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2543 mg_get(sv);
2544
2545 if (SvROK(sv)) {
2546 if (SvAMAGIC(sv)) {
2547 SV *tmpstr;
2548 if (flags & SV_SKIP_OVERLOAD)
2549 return 0;
2550 tmpstr = AMG_CALLunary(sv, numer_amg);
2551 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2552 return SvUV(tmpstr);
2553 }
2554 }
2555 return PTR2UV(SvRV(sv));
2556 }
2557
8d919b0a 2558 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2559 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2560 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2561 Regexps have no SvIVX and SvNVX fields. */
df6b4bd5 2562 assert(SvPOKp(sv));
e20b6c3b 2563 {
71c558c3 2564 UV value;
8d919b0a
FC
2565 const char * const ptr =
2566 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2567 const int numtype
2568 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2569
2570 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2571 == IS_NUMBER_IN_UV) {
2572 /* It's definitely an integer */
2573 if (!(numtype & IS_NUMBER_NEG))
2574 return value;
2575 }
058b8ae2 2576
e91de695
JH
2577 /* Quite wrong but no good choices. */
2578 if ((numtype & IS_NUMBER_INFINITY)) {
2579 return UV_MAX; /* So wrong. */
2580 } else if ((numtype & IS_NUMBER_NAN)) {
2581 return 0; /* So wrong. */
2582 }
2583
71c558c3
NC
2584 if (!numtype) {
2585 if (ckWARN(WARN_NUMERIC))
2586 not_a_number(sv);
2587 }
8d919b0a 2588 return U_V(Atof(ptr));
e20b6c3b 2589 }
4bac9ae4
CS
2590 }
2591
2592 if (SvTHINKFIRST(sv)) {
0336b60e 2593 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2594 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2595 report_uninit(sv);
ff68c719
PP
2596 return 0;
2597 }
2598 }
4bac9ae4 2599
af359546
NC
2600 if (!SvIOKp(sv)) {
2601 if (S_sv_2iuv_common(aTHX_ sv))
2602 return 0;
ff68c719 2603 }
25da4f38 2604
147e3846 2605 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
1d7c1841 2606 PTR2UV(sv),SvUVX(sv)));
25da4f38 2607 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2608}
2609
645c22ef 2610/*
196007d1 2611=for apidoc sv_2nv_flags
645c22ef
DM
2612
2613Return the num value of an SV, doing any necessary string or integer
c5608a1f 2614conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
39d5de13 2615Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2616
2617=cut
2618*/
2619
65202027 2620NV
5aaab254 2621Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2622{
1061065f
DD
2623 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2624
217f6fa3
FC
2625 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2626 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2627 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2628 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2629 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2630 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2631 const char *ptr;
39d5de13
DM
2632 if (flags & SV_GMAGIC)
2633 mg_get(sv);
463ee0b2
LW
2634 if (SvNOKp(sv))
2635 return SvNVX(sv);
cd84013a 2636 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a 2637 ptr = SvPVX_const(sv);
041457d9 2638 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2639 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2640 not_a_number(sv);
8d919b0a 2641 return Atof(ptr);
a0d0e21e 2642 }
25da4f38 2643 if (SvIOKp(sv)) {
1c846c1f 2644 if (SvIsUV(sv))
65202027 2645 return (NV)SvUVX(sv);
25da4f38 2646 else
65202027 2647 return (NV)SvIVX(sv);
47a72cb8
NC
2648 }
2649 if (SvROK(sv)) {
2650 goto return_rok;
2651 }
2652 assert(SvTYPE(sv) >= SVt_PVMG);
2653 /* This falls through to the report_uninit near the end of the
2654 function. */
2655 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2656 if (SvROK(sv)) {
47a72cb8 2657 return_rok:
deb46114 2658 if (SvAMAGIC(sv)) {
aee036bb
DM
2659 SV *tmpstr;
2660 if (flags & SV_SKIP_OVERLOAD)
2661 return 0;
31d632c3 2662 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2663 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2664 return SvNV(tmpstr);
2665 }
2666 }
2667 return PTR2NV(SvRV(sv));
a0d0e21e 2668 }
0336b60e 2669 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2670 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2671 report_uninit(sv);
ed6116ce
LW
2672 return 0.0;
2673 }
79072805
LW
2674 }
2675 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2676 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2677 sv_upgrade(sv, SVt_NV);
b2f82b52 2678 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
097ee67d 2679 DEBUG_c({
688523a0
KW
2680 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2681 STORE_LC_NUMERIC_SET_STANDARD();
1d7c1841 2682 PerlIO_printf(Perl_debug_log,
147e3846 2683 "0x%" UVxf " num(%" NVgf ")\n",
1d7c1841 2684 PTR2UV(sv), SvNVX(sv));
688523a0 2685 RESTORE_LC_NUMERIC();
097ee67d 2686 });
b2f82b52
KW
2687 CLANG_DIAG_RESTORE_STMT;
2688
79072805
LW
2689 }
2690 else if (SvTYPE(sv) < SVt_PVNV)
2691 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2692 if (SvNOKp(sv)) {
2693 return SvNVX(sv);
61604483 2694 }
59d8ce62 2695 if (SvIOKp(sv)) {
9d6ce603 2696 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2697#ifdef NV_PRESERVES_UV
a43d94f2
NC
2698 if (SvIOK(sv))
2699 SvNOK_on(sv);
2700 else
2701 SvNOKp_on(sv);
28e5dec8
JH
2702#else
2703 /* Only set the public NV OK flag if this NV preserves the IV */
2704 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2705 if (SvIOK(sv) &&
2706 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2707 : (SvIVX(sv) == I_V(SvNVX(sv))))
2708 SvNOK_on(sv);
2709 else
2710 SvNOKp_on(sv);
2711#endif
93a17b20 2712 }
cd84013a 2713 else if (SvPOKp(sv)) {
c2988b20 2714 UV value;
3823048b 2715 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2716 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2717 not_a_number(sv);
28e5dec8 2718#ifdef NV_PRESERVES_UV
c2988b20
NC
2719 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2720 == IS_NUMBER_IN_UV) {
5e045b90 2721 /* It's definitely an integer */
9d6ce603 2722 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
66d83377 2723 } else {
3823048b 2724 S_sv_setnv(aTHX_ sv, numtype);
66d83377 2725 }
a43d94f2
NC
2726 if (numtype)
2727 SvNOK_on(sv);
2728 else
2729 SvNOKp_on(sv);
28e5dec8 2730#else
e91de695
JH
2731 SvNV_set(sv, Atof(SvPVX_const(sv)));
2732 /* Only set the public NV OK flag if this NV preserves the value in
2733 the PV at least as well as an IV/UV would.
2734 Not sure how to do this 100% reliably. */
2735 /* if that shift count is out of range then Configure's test is
2736 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2737 UV_BITS */
2738 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2739 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2740 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2741 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2742 /* Can't use strtol etc to convert this string, so don't try.
2743 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
c2988b20
NC
2744 SvNOK_on(sv);
2745 } else {
e91de695 2746 /* value has been set. It may not be precise. */
53e2bfb7 2747 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
e91de695
JH
2748 /* 2s complement assumption for (UV)IV_MIN */
2749 SvNOK_on(sv); /* Integer is too negative. */
c2988b20 2750 } else {
e91de695
JH
2751 SvNOKp_on(sv);
2752 SvIOKp_on(sv);
6fa402ec 2753
e91de695 2754 if (numtype & IS_NUMBER_NEG) {
02b08bbc
DM
2755 /* -IV_MIN is undefined, but we should never reach
2756 * this point with both IS_NUMBER_NEG and value ==
2757 * (UV)IV_MIN */
2758 assert(value != (UV)IV_MIN);
e91de695
JH
2759 SvIV_set(sv, -(IV)value);
2760 } else if (value <= (UV)IV_MAX) {
2761 SvIV_set(sv, (IV)value);
2762 } else {
2763 SvUV_set(sv, value);
2764 SvIsUV_on(sv);
2765 }
c2988b20 2766
e91de695
JH
2767 if (numtype & IS_NUMBER_NOT_INT) {
2768 /* I believe that even if the original PV had decimals,
2769 they are lost beyond the limit of the FP precision.
2770 However, neither is canonical, so both only get p
2771 flags. NWC, 2000/11/25 */
2772 /* Both already have p flags, so do nothing */
2773 } else {
2774 const NV nv = SvNVX(sv);
2775 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2776 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2777 if (SvIVX(sv) == I_V(nv)) {
2778 SvNOK_on(sv);
2779 } else {
2780 /* It had no "." so it must be integer. */
2781 }
2782 SvIOK_on(sv);
0f83c5a4 2783 } else {
e91de695
JH
2784 /* between IV_MAX and NV(UV_MAX).
2785 Could be slightly > UV_MAX */
2786
2787 if (numtype & IS_NUMBER_NOT_INT) {
2788 /* UV and NV both imprecise. */
0f83c5a4 2789 } else {
e91de695
JH
2790 const UV nv_as_uv = U_V(nv);
2791
2792 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2793 SvNOK_on(sv);
c2988b20 2794 }
e91de695 2795 SvIOK_on(sv);
c2988b20
NC
2796 }
2797 }
2798 }
2799 }
0f83c5a4 2800 }
e91de695
JH
2801 /* It might be more code efficient to go through the entire logic above
2802 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2803 gets complex and potentially buggy, so more programmer efficient
2804 to do it this way, by turning off the public flags: */
2805 if (!numtype)
2806 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2807#endif /* NV_PRESERVES_UV */
93a17b20 2808 }
c0443cc0 2809 else {
e91de695
JH
2810 if (isGV_with_GP(sv)) {
2811 glob_2number(MUTABLE_GV(sv));
2812 return 0.0;
2813 }
180488f8 2814
e91de695
JH
2815 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2816 report_uninit(sv);
2817 assert (SvTYPE(sv) >= SVt_NV);
2818 /* Typically the caller expects that sv_any is not NULL now. */
2819 /* XXX Ilya implies that this is a bug in callers that assume this
2820 and ideally should be fixed. */
2821 return 0.0;
79072805 2822 }
b2f82b52 2823 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
097ee67d 2824 DEBUG_c({
688523a0
KW
2825 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2826 STORE_LC_NUMERIC_SET_STANDARD();
147e3846 2827 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
e91de695 2828 PTR2UV(sv), SvNVX(sv));
688523a0 2829 RESTORE_LC_NUMERIC();
e91de695 2830 });
b2f82b52 2831 CLANG_DIAG_RESTORE_STMT;
463ee0b2 2832 return SvNVX(sv);
79072805
LW
2833}
2834
800401ee
JH
2835/*
2836=for apidoc sv_2num
2837
2838Return an SV with the numeric value of the source SV, doing any necessary
d024d1a7
FC
2839reference or overload conversion. The caller is expected to have handled
2840get-magic already.
800401ee
JH
2841
2842=cut
2843*/
2844
2845SV *
5aaab254 2846Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2847{
7918f24d
NC
2848 PERL_ARGS_ASSERT_SV_2NUM;
2849
b9ee0594
RGS
2850 if (!SvROK(sv))
2851 return sv;
800401ee 2852 if (SvAMAGIC(sv)) {
31d632c3 2853 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2854 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2855 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2856 return sv_2num(tmpsv);
2857 }
2858 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2859}
2860
dd0a5f5f
TK
2861/* int2str_table: lookup table containing string representations of all
2862 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2863 * int2str_table.arr[12*2] is "12".
2864 *
2865 * We are going to read two bytes at a time, so we have to ensure that
2866 * the array is aligned to a 2 byte boundary. That's why it was made a
2867 * union with a dummy U16 member. */
2868static const union {
2869 char arr[200];
2870 U16 dummy;
2871} int2str_table = {{
2872 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2873 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2874 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2875 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2876 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2877 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2878 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2879 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2880 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2881 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2882 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2883 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2884 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2885 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2886 '9', '8', '9', '9'
2887}};
2888
645c22ef
DM
2889/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2890 * UV as a string towards the end of buf, and return pointers to start and
2891 * end of it.
2892 *
2893 * We assume that buf is at least TYPE_CHARS(UV) long.
2894 */
2895
dd0a5f5f 2896PERL_STATIC_INLINE char *
5de3775c 2897S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2898{
25da4f38 2899 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2900 char * const ebuf = ptr;
25da4f38 2901 int sign;
dd0a5f5f 2902 U16 *word_ptr, *word_table;
25da4f38 2903
7918f24d
NC
2904 PERL_ARGS_ASSERT_UIV_2BUF;
2905
dd0a5f5f
TK
2906 /* ptr has to be properly aligned, because we will cast it to U16* */
2907 assert(PTR2nat(ptr) % 2 == 0);
2908 /* we are going to read/write two bytes at a time */
2909 word_ptr = (U16*)ptr;
2910 word_table = (U16*)int2str_table.arr;
2911
2912 if (UNLIKELY(is_uv))
25da4f38
IZ
2913 sign = 0;
2914 else if (iv >= 0) {
2915 uv = iv;
2916 sign = 0;
2917 } else {
7895b980
KW
2918 /* Using 0- here to silence bogus warning from MS VC */
2919 uv = (UV) (0 - (UV) iv);
25da4f38
IZ
2920 sign = 1;
2921 }
dd0a5f5f
TK
2922
2923 while (uv > 99) {
2924 *--word_ptr = word_table[uv % 100];
2925 uv /= 100;
2926 }
2927 ptr = (char*)word_ptr;
2928
2929 if (uv < 10)
2930 *--ptr = (char)uv + '0';
2931 else {
2932 *--word_ptr = word_table[uv];
2933 ptr = (char*)word_ptr;
2934 }
2935
25da4f38 2936 if (sign)
dd0a5f5f
TK
2937 *--ptr = '-';
2938
25da4f38
IZ
2939 *peob = ebuf;
2940 return ptr;
2941}
2942
bfaa02d5
JH
2943/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2944 * infinity or a not-a-number, writes the appropriate strings to the
2945 * buffer, including a zero byte. On success returns the written length,
3bde2d43
JH
2946 * excluding the zero byte, on failure (not an infinity, not a nan)
2947 * returns zero, assert-fails on maxlen being too short.
3823048b
JH
2948 *
2949 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2950 * shared string constants we point to, instead of generating a new
2951 * string for each instance. */
bfaa02d5 2952STATIC size_t
3823048b 2953S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
3bde2d43 2954 char* s = buffer;
bfaa02d5 2955 assert(maxlen >= 4);
3bde2d43
JH
2956 if (Perl_isinf(nv)) {
2957 if (nv < 0) {
2958 if (maxlen < 5) /* "-Inf\0" */
2959 return 0;
2960 *s++ = '-';
2961 } else if (plus) {
2962 *s++ = '+';
6e915616 2963 }
3bde2d43
JH
2964 *s++ = 'I';
2965 *s++ = 'n';
2966 *s++ = 'f';
2967 }
2968 else if (Perl_isnan(nv)) {
2969 *s++ = 'N';
2970 *s++ = 'a';
2971 *s++ = 'N';
2972 /* XXX optionally output the payload mantissa bits as
2973 * "(unsigned)" (to match the nan("...") C99 function,
2974 * or maybe as "(0xhhh...)" would make more sense...
2975 * provide a format string so that the user can decide?
2976 * NOTE: would affect the maxlen and assert() logic.*/
2977 }
2978 else {
2979 return 0;
bfaa02d5 2980 }
3bde2d43 2981 assert((s == buffer + 3) || (s == buffer + 4));
defe49c8
AL
2982 *s = 0;
2983 return s - buffer;
bfaa02d5
JH
2984}
2985
2986/*
2987=for apidoc sv_2pv_flags
2988
796b6530 2989Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
c5608a1f 2990If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
bfaa02d5
JH
2991string if necessary. Normally invoked via the C<SvPV_flags> macro.
2992C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2993
2994=cut
2995*/
2996
2997char *
2998Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2999{
3000 char *s;
3001
3002 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
3003
3004 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
3005 && SvTYPE(sv) != SVt_PVFM);
3006 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3007 mg_get(sv);
3008 if (SvROK(sv)) {
3009 if (SvAMAGIC(sv)) {
3010 SV *tmpstr;
3011 if (flags & SV_SKIP_OVERLOAD)
3012 return NULL;
3013 tmpstr = AMG_CALLunary(sv, string_amg);
3014 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3015 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3016 /* Unwrap this: */
3017 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3018 */
3019
3020 char *pv;
3021 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3022 if (flags & SV_CONST_RETURN) {
3023 pv = (char *) SvPVX_const(tmpstr);
3024 } else {
3025 pv = (flags & SV_MUTABLE_RETURN)
3026 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3027 }
3028 if (lp)
3029 *lp = SvCUR(tmpstr);
3030 } else {
3031 pv = sv_2pv_flags(tmpstr, lp, flags);
3032 }
3033 if (SvUTF8(tmpstr))
3034 SvUTF8_on(sv);
3035 else
3036 SvUTF8_off(sv);
3037 return pv;
3038 }
3039 }
3040 {
3041 STRLEN len;
3042 char *retval;
3043 char *buffer;
3044 SV *const referent = SvRV(sv);
3045
3046 if (!referent) {
3047 len = 7;
3048 retval = buffer = savepvn("NULLREF", len);
3049 } else if (SvTYPE(referent) == SVt_REGEXP &&
3050 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3051 amagic_is_enabled(string_amg))) {
3052 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3053
3054 assert(re);
3055
3056 /* If the regex is UTF-8 we want the containing scalar to
3057 have an UTF-8 flag too */
3058 if (RX_UTF8(re))
3059 SvUTF8_on(sv);
3060 else
3061 SvUTF8_off(sv);
3062
3063 if (lp)
3064 *lp = RX_WRAPLEN(re);
3065
3066 return RX_WRAPPED(re);
3067 } else {
3068 const char *const typestr = sv_reftype(referent, 0);
3069 const STRLEN typelen = strlen(typestr);
3070 UV addr = PTR2UV(referent);
3071 const char *stashname = NULL;
3072 STRLEN stashnamelen = 0; /* hush, gcc */
3073 const char *buffer_end;
3074
3075 if (SvOBJECT(referent)) {
3076 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3077
3078 if (name) {
3079 stashname = HEK_KEY(name);
3080 stashnamelen = HEK_LEN(name);
3081
3082 if (HEK_UTF8(name)) {
3083 SvUTF8_on(sv);
3084 } else {
3085 SvUTF8_off(sv);
3086 }
3087 } else {
3088 stashname = "__ANON__";
3089 stashnamelen = 8;
3090 }
3091 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3092 + 2 * sizeof(UV) + 2 /* )\0 */;
3093 } else {
3094 len = typelen + 3 /* (0x */
3095 + 2 * sizeof(UV) + 2 /* )\0 */;
3096 }
fafee734 3097
4bac9ae4
CS
3098 Newx(buffer, len, char);
3099 buffer_end = retval = buffer + len;
3100
3101 /* Working backwards */
3102 *--retval = '\0';
3103 *--retval = ')';
3104 do {
3105 *--retval = PL_hexdigit[addr & 15];
3106 } while (addr >>= 4);
3107 *--retval = 'x';
3108 *--retval = '0';
3109 *--retval = '(';
3110
3111 retval -= typelen;
3112 memcpy(retval, typestr, typelen);
3113
3114 if (stashname) {
3115 *--retval = '=';
3116 retval -= stashnamelen;
3117 memcpy(retval, stashname, stashnamelen);
c080367d 3118 }
4bac9ae4
CS
3119 /* retval may not necessarily have reached the start of the
3120 buffer here. */
3121 assert (retval >= buffer);
3122
3123 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 3124 }
cdb061a3 3125 if (lp)
4bac9ae4
CS
3126 *lp = len;
3127 SAVEFREEPV(buffer);
3128 return retval;
79072805 3129 }
79072805 3130 }
4bac9ae4
CS
3131
3132 if (SvPOKp(sv)) {
3133 if (lp)
3134 *lp = SvCUR(sv);
3135 if (flags & SV_MUTABLE_RETURN)
3136 return SvPVX_mutable(sv);
3137 if (flags & SV_CONST_RETURN)
3138 return (char *)SvPVX_const(sv);
3139 return SvPVX(sv);
3140 }
3141
3142 if (SvIOK(sv)) {
28e5dec8
JH
3143 /* I'm assuming that if both IV and NV are equally valid then
3144 converting the IV is going to be more efficient */
e1ec3a88 3145 const U32 isUIOK = SvIsUV(sv);
dd0a5f5f
TK
3146 /* The purpose of this union is to ensure that arr is aligned on
3147 a 2 byte boundary, because that is what uiv_2buf() requires */
3148 union {
3149 char arr[TYPE_CHARS(UV)];
3150 U16 dummy;
3151 } buf;
28e5dec8 3152 char *ebuf, *ptr;
97a130b8 3153 STRLEN len;
28e5dec8
JH
3154
3155 if (SvTYPE(sv) < SVt_PVIV)
3156 sv_upgrade(sv, SVt_PVIV);
dd0a5f5f 3157 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 3158 len = ebuf - ptr;
5902b6a9 3159 /* inlined from sv_setpvn */
97a130b8
NC
3160 s = SvGROW_mutable(sv, len + 1);
3161 Move(ptr, s, len, char);
3162 s += len;
28e5dec8 3163 *s = '\0';
b127e37e 3164 SvPOK_on(sv);
28e5dec8 3165 }
4bac9ae4 3166 else if (SvNOK(sv)) {
79072805
LW
3167 if (SvTYPE(sv) < SVt_PVNV)
3168 sv_upgrade(sv, SVt_PVNV);
128eeacb
DD
3169 if (SvNVX(sv) == 0.0
3170#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3171 && !Perl_isnan(SvNVX(sv))
3172#endif
3173 ) {
29912d93
NC
3174 s = SvGROW_mutable(sv, 2);
3175 *s++ = '0';
3176 *s = '\0';
3177 } else {
5e85836e 3178 STRLEN len;
fb8cdbc5 3179 STRLEN size = 5; /* "-Inf\0" */
0c7e610f 3180
fb8cdbc5 3181 s = SvGROW_mutable(sv, size);
3823048b 3182 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
fb8cdbc5 3183 if (len > 0) {
0c7e610f 3184 s += len;
fb8cdbc5
JH
3185 SvPOK_on(sv);
3186 }
0c7e610f 3187 else {
0c7e610f 3188 /* some Xenix systems wipe out errno here */
fb8cdbc5
JH
3189 dSAVE_ERRNO;
3190
3840bff0
JH
3191 size =
3192 1 + /* sign */
3193 1 + /* "." */
3194 NV_DIG +
3195 1 + /* "e" */
3196 1 + /* sign */
3197 5 + /* exponent digits */
3198 1 + /* \0 */
3199 2; /* paranoia */
b127e37e 3200
fb8cdbc5 3201 s = SvGROW_mutable(sv, size);
b127e37e 3202#ifndef USE_LOCALE_NUMERIC
a4eca1d4
JH
3203 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3204
0c7e610f
JH
3205 SvPOK_on(sv);
3206#else
28acfe03 3207 {
3840bff0 3208 bool local_radix;
67d796ae
KW
3209 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3210 STORE_LC_NUMERIC_SET_TO_NEEDED();
3840bff0 3211
dd2dbc5f 3212 local_radix = _NOT_IN_NUMERIC_STANDARD;
4c039fd8
DM
3213 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3214 size += SvCUR(PL_numeric_radix_sv) - 1;
3840bff0
JH
3215 s = SvGROW_mutable(sv, size);
3216 }
3217
a4eca1d4 3218 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
0c7e610f
JH
3219
3220 /* If the radix character is UTF-8, and actually is in the
3221 * output, turn on the UTF-8 flag for the scalar */
3dbc6af5
KW
3222 if ( local_radix
3223 && SvUTF8(PL_numeric_radix_sv)
3224 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3225 {
3840bff0
JH
3226 SvUTF8_on(sv);
3227 }
3228
0c7e610f 3229 RESTORE_LC_NUMERIC();
28acfe03 3230 }
68e8f474 3231
0c7e610f
JH
3232 /* We don't call SvPOK_on(), because it may come to
3233 * pass that the locale changes so that the
3234 * stringification we just did is no longer correct. We
3235 * will have to re-stringify every time it is needed */
b127e37e 3236#endif
0c7e610f
JH
3237 RESTORE_ERRNO;
3238 }
3239 while (*s) s++;
bbce6d69 3240 }
79072805 3241 }
4bac9ae4
CS
3242 else if (isGV_with_GP(sv)) {
3243 GV *const gv = MUTABLE_GV(sv);
3244 SV *const buffer = sv_newmortal();
8d1c3e26 3245
4bac9ae4 3246 gv_efullname3(buffer, gv, "*");
180488f8 3247
4bac9ae4
CS
3248 assert(SvPOK(buffer));
3249 if (SvUTF8(buffer))
3250 SvUTF8_on(sv);
1097da16
TC
3251 else
3252 SvUTF8_off(sv);
4bac9ae4
CS
3253 if (lp)
3254 *lp = SvCUR(buffer);
3255 return SvPVX(buffer);
3256 }
3257 else {
cdb061a3 3258 if (lp)
00b6aa41 3259 *lp = 0;
9f621bb0
NC
3260 if (flags & SV_UNDEF_RETURNS_NULL)
3261 return NULL;
4f62cd62 3262 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3263 report_uninit(sv);
4bac9ae4
CS
3264 /* Typically the caller expects that sv_any is not NULL now. */
3265 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3266 sv_upgrade(sv, SVt_PV);
73d840c0 3267 return (char *)"";
79072805 3268 }
4bac9ae4 3269
cdb061a3 3270 {
823a54a3 3271 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3272 if (lp)
3273 *lp = len;
3274 SvCUR_set(sv, len);
3275 }
147e3846 3276 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3f7c398e 3277 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3278 if (flags & SV_CONST_RETURN)
3279 return (char *)SvPVX_const(sv);
10516c54
NC
3280 if (flags & SV_MUTABLE_RETURN)
3281 return SvPVX_mutable(sv);
463ee0b2
LW
3282 return SvPVX(sv);
3283}
3284
645c22ef 3285/*
6050d10e
JP
3286=for apidoc sv_copypv
3287
3288Copies a stringified representation of the source SV into the
796b6530 3289destination SV. Automatically performs any necessary C<mg_get> and
54f0641b 3290coercion of numeric values into strings. Guaranteed to preserve
796b6530
KW
3291C<UTF8> flag even from overloaded objects. Similar in nature to
3292C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3293string. Mostly uses C<sv_2pv_flags> to do its work, except when that
6050d10e
JP
3294would lose the UTF-8'ness of the PV.
3295
4bac9ae4
CS
3296=for apidoc sv_copypv_nomg
3297
796b6530 3298Like C<sv_copypv>, but doesn't invoke get magic first.
4bac9ae4
CS
3299
3300=for apidoc sv_copypv_flags
3301
796b6530 3302Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
c5608a1f 3303has the C<SV_GMAGIC> bit set.
4bac9ae4 3304
6050d10e
JP
3305=cut
3306*/
3307
3308void
5aaab254 3309Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3310{
446eaa42 3311 STRLEN len;
4bac9ae4 3312 const char *s;
7918f24d 3313
4bac9ae4 3314 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3315
c77ed9ca 3316 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
cb50f42d 3317 sv_setpvn(dsv,s,len);
446eaa42 3318 if (SvUTF8(ssv))
cb50f42d 3319 SvUTF8_on(dsv);
446eaa42 3320 else
cb50f42d 3321 SvUTF8_off(dsv);
6050d10e
JP
3322}
3323
3324/*
645c22ef
DM
3325=for apidoc sv_2pvbyte
3326
796b6530 3327Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
b661dc90
KW
3328to its length. If the SV is marked as being encoded as UTF-8, it will
3329downgrade it to a byte string as a side-effect, if possible. If the SV cannot
3330be downgraded, this croaks.
645c22ef
DM
3331
3332Usually accessed via the C<SvPVbyte> macro.
3333
3334=cut
3335*/
3336
7340a771 3337char *
757fc329 3338Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
7340a771 3339{
757fc329 3340 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS;
7918f24d 3341
757fc329
P
3342 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3343 mg_get(sv);
4499db73
FC
3344 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3345 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3346 SV *sv2 = sv_newmortal();
48120f8f 3347 sv_copypv_nomg(sv2,sv);
a901b181
FC
3348 sv = sv2;
3349 }
757fc329 3350 sv_utf8_downgrade_nomg(sv,0);
71eb6d8c 3351 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3352}
3353
645c22ef 3354/*
035cbb0e
RGS
3355=for apidoc sv_2pvutf8
3356
796b6530 3357Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
035cbb0e
RGS
3358to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3359
3360Usually accessed via the C<SvPVutf8> macro.
3361
3362=cut
3363*/
645c22ef 3364
7340a771 3365char *
757fc329 3366Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
7340a771 3367{
757fc329 3368 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS;
7918f24d 3369
757fc329
P
3370 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3371 mg_get(sv);
4499db73 3372 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
757fc329
P
3373 || isGV_with_GP(sv) || SvROK(sv)) {
3374 SV *sv2 = sv_newmortal();
3375 sv_copypv_nomg(sv2,sv);
3376 sv = sv2;
3377 }
4bac9ae4 3378 sv_utf8_upgrade_nomg(sv);
c3ec315f 3379 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3380}
1c846c1f 3381
7ee2227d 3382
645c22ef
DM
3383/*
3384=for apidoc sv_2bool
3385
796b6530
KW
3386This macro is only used by C<sv_true()> or its macro equivalent, and only if
3387the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3388It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
06c841cf
FC
3389
3390=for apidoc sv_2bool_flags
3391
796b6530
KW
3392This function is only used by C<sv_true()> and friends, and only if
3393the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3394contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
06c841cf 3395
645c22ef
DM
3396
3397=cut
3398*/
3399
463ee0b2 3400bool
9d176cd8 3401Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
463ee0b2 3402{
06c841cf 3403 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3404
9d176cd8 3405 restart:
06c841cf 3406 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3407
a0d0e21e
LW
3408 if (!SvOK(sv))
3409 return 0;
3410 if (SvROK(sv)) {
fabdb6c0 3411 if (SvAMAGIC(sv)) {
31d632c3 3412 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
9d176cd8
DD
3413 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3414 bool svb;
3415 sv = tmpsv;
3416 if(SvGMAGICAL(sv)) {
3417 flags = SV_GMAGIC;
3418 goto restart; /* call sv_2bool */
3419 }
3420 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3421 else if(!SvOK(sv)) {
3422 svb = 0;
3423 }
3424 else if(SvPOK(sv)) {
3425 svb = SvPVXtrue(sv);
3426 }
3427 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3428 svb = (SvIOK(sv) && SvIVX(sv) != 0)
659c4b96 3429 || (SvNOK(sv) && SvNVX(sv) != 0.0);
9d176cd8
DD
3430 }
3431 else {
3432 flags = 0;
3433 goto restart; /* call sv_2bool_nomg */
3434 }
3435 return cBOOL(svb);
3436 }
fabdb6c0 3437 }
e8f01ee5
DM
3438 assert(SvRV(sv));
3439 return TRUE;
a0d0e21e 3440 }
85b7d9b3
FC
3441 if (isREGEXP(sv))
3442 return
3443 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
1a436fbe
DM
3444
3445 if (SvNOK(sv) && !SvPOK(sv))
3446 return SvNVX(sv) != 0.0;
3447
4bac9ae4 3448 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3449}
3450
c461cf8f
JH
3451/*
3452=for apidoc sv_utf8_upgrade
3453
78ea37eb 3454Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3455Forces the SV to string form if it is not already.
2bbc8d55 3456Will C<mg_get> on C<sv> if appropriate.
796b6530 3457Always sets the C<SvUTF8> flag to avoid future validity checks even
2bbc8d55
SP
3458if the whole string is the same in UTF-8 as not.
3459Returns the number of bytes in the converted string
c461cf8f 3460
0efd0472 3461This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3462use the Encode extension for that.
3463
fe749c9a
KW
3464=for apidoc sv_utf8_upgrade_nomg
3465
796b6530 3466Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
fe749c9a 3467
8d6d96c1
HS
3468=for apidoc sv_utf8_upgrade_flags
3469
78ea37eb 3470Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3471Forces the SV to string form if it is not already.
8d6d96c1 3472Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3473if all the bytes are invariant in UTF-8.
3474If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55 3475will C<mg_get> on C<sv> if appropriate, else not.
2a590426 3476
c58971e9 3477The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
2a590426
KW
3478
3479Returns the number of bytes in the converted string.
8d6d96c1 3480
0efd0472 3481This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3482use the Encode extension for that.
3483
2a590426 3484=for apidoc sv_utf8_upgrade_flags_grow
b3ab6785 3485
796b6530
KW
3486Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3487the number of unused bytes the string of C<sv> is guaranteed to have free after
2a590426
KW
3488it upon return. This allows the caller to reserve extra space that it intends
3489to fill, to avoid extra grows.
b3ab6785 3490
2a590426
KW
3491C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3492are implemented in terms of this function.
3493
3494Returns the number of bytes in the converted string (not including the spares).
3495
3496=cut
b3ab6785 3497
6602b933
KW
3498If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3499C<NUL> isn't guaranteed due to having other routines do the work in some input
3500cases, or if the input is already flagged as being in utf8.
b3ab6785 3501
8d6d96c1
HS
3502*/
3503
3504STRLEN
5aaab254 3505Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3506{
b3ab6785 3507 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3508
808c356f
RGS
3509 if (sv == &PL_sv_undef)
3510 return 0;
892f9127 3511 if (!SvPOK_nog(sv)) {
e0e62c2a 3512 STRLEN len = 0;
d52b7888
NC
3513 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3514 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3515 if (SvUTF8(sv)) {
3516 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3517 return len;
b3ab6785 3518 }
d52b7888 3519 } else {
33fb6f35 3520 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3521 }
e0e62c2a 3522 }
4411f3b6 3523
fde84d2e
DM
3524 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3525 * compiled and individual nodes will remain non-utf8 even if the
3526 * stringified version of the pattern gets upgraded. Whether the
3527 * PVX of a REGEXP should be grown or we should just croak, I don't
3528 * know - DAPM */
3529 if (SvUTF8(sv) || isREGEXP(sv)) {
b3ab6785 3530 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3531 return SvCUR(sv);
f5cee72b 3532 }
5fec3b1d 3533
765f542d 3534 if (SvIsCOW(sv)) {
c56ed9f6 3535 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3536 }
3537
4e93345f 3538 if (SvCUR(sv) == 0) {
e2e3bb6a
KW
3539 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3540 byte */
4e93345f 3541 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3542 /* This function could be much more efficient if we
2bbc8d55 3543 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3544 * chars in the PV. Given that there isn't such a flag
c58971e9 3545 * make the loop as fast as possible. */
b3ab6785 3546 U8 * s = (U8 *) SvPVX_const(sv);
b3ab6785 3547 U8 *t = s;
c4e7c712 3548
c58971e9 3549 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
b3ab6785 3550
c58971e9
KW
3551 /* utf8 conversion not needed because all are invariants. Mark
3552 * as UTF-8 even if no variant - saves scanning loop */
3553 SvUTF8_on(sv);
3554 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3555 return SvCUR(sv);
dc772057 3556 }
b3ab6785 3557
c58971e9
KW
3558 /* Here, there is at least one variant (t points to the first one), so
3559 * the string should be converted to utf8. Everything from 's' to
3560 * 't - 1' will occupy only 1 byte each on output.
dc772057 3561 *
c58971e9
KW
3562 * Note that the incoming SV may not have a trailing '\0', as certain
3563 * code in pp_formline can send us partially built SVs.
b3ab6785
KW
3564 *
3565 * There are two main ways to convert. One is to create a new string
3566 * and go through the input starting from the beginning, appending each
c58971e9
KW
3567 * converted value onto the new string as we go along. Going this
3568 * route, it's probably best to initially allocate enough space in the
3569 * string rather than possibly running out of space and having to
3570 * reallocate and then copy what we've done so far. Since everything
3571 * from 's' to 't - 1' is invariant, the destination can be initialized
3572 * with these using a fast memory copy. To be sure to allocate enough
3573 * space, one could use the worst case scenario, where every remaining
3574<