This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlapi: Fix pod for perl_clone()
[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
bd81e77b
NC
886#define HADNV FALSE
887#define NONV TRUE
29489e7c 888
d2a0f284 889
bd81e77b
NC
890#ifdef PURIFY
891/* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893#define HASARENA FALSE
894#else
895#define HASARENA TRUE
896#endif
897#define NOARENA FALSE
29489e7c 898
d2a0f284
JC
899/* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
905 declarations.
906 */
95db5f15
MB
907#define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909#define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913#define FIT_ARENA(count,body_size) \
cd1dc8e2 914 (U32)(count \
95db5f15 915 ? FIT_ARENAn (count, body_size) \
cd1dc8e2 916 : FIT_ARENA0 (body_size))
d2a0f284 917
bd81e77b
NC
918/* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
29489e7c 921
bd81e77b
NC
922#define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
daba3364 924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 925
bd81e77b 926static const struct body_details bodies_by_type[] = {
829cd18a
NC
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 929
db93c0c4
NC
930 /* IVs are in the head, so the allocation size is 0. */
931 { 0,
d2a0f284 932 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 934 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
935 },
936
5b306eef
DD
937#if NVSIZE <= IVSIZE
938 { 0, sizeof(NV),
939 STRUCT_OFFSET(XPVNV, xnv_u),
940 SVt_NV, FALSE, HADNV, NOARENA, 0 },
941#else
6e128786
NC
942 { sizeof(NV), sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
5b306eef 945#endif
d2a0f284 946
bc337e5c 947 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
948 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 950 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 951 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 952
d361b004
KW
953 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_INVLIST, TRUE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 958
bc337e5c 959 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
960 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVIV, FALSE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 964
bc337e5c 965 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
966 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVNV, FALSE, HADNV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 970
6e128786 971 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 972 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 973
601dfd0a
NC
974 { sizeof(regexp),
975 sizeof(regexp),
976 0,
ecff11eb 977 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 978 FIT_ARENA(0, sizeof(regexp))
5c35adbb 979 },
4df7f6af 980
10666ae3 981 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
982 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
983
10666ae3 984 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
985 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
986
601dfd0a 987 { sizeof(XPVAV),
4f7003f5 988 copy_length(XPVAV, xav_alloc),
601dfd0a 989 0,
69ba284b 990 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 991 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 992
601dfd0a 993 { sizeof(XPVHV),
359164a0 994 copy_length(XPVHV, xhv_max),
601dfd0a 995 0,
69ba284b 996 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 997 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 998
601dfd0a
NC
999 { sizeof(XPVCV),
1000 sizeof(XPVCV),
1001 0,
69ba284b 1002 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 1003 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 1004
601dfd0a
NC
1005 { sizeof(XPVFM),
1006 sizeof(XPVFM),
1007 0,
69ba284b 1008 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 1009 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 1010
601dfd0a
NC
1011 { sizeof(XPVIO),
1012 sizeof(XPVIO),
1013 0,
b6f60916 1014 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 1015 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 1016};
29489e7c 1017
bd81e77b 1018#define new_body_allocated(sv_type) \
d2a0f284 1019 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1020 - bodies_by_type[sv_type].offset)
29489e7c 1021
26359cfa
NC
1022/* return a thing to the free list */
1023
1024#define del_body(thing, root) \
1025 STMT_START { \
1026 void ** const thing_copy = (void **)thing; \
1027 *thing_copy = *root; \
1028 *root = (void*)thing_copy; \
1029 } STMT_END
29489e7c 1030
bd81e77b 1031#ifdef PURIFY
5b306eef
DD
1032#if !(NVSIZE <= IVSIZE)
1033# define new_XNV() safemalloc(sizeof(XPVNV))
1034#endif
beeec492
NC
1035#define new_XPVNV() safemalloc(sizeof(XPVNV))
1036#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 1037
beeec492 1038#define del_XPVGV(p) safefree(p)
29489e7c 1039
bd81e77b 1040#else /* !PURIFY */
29489e7c 1041
5b306eef
DD
1042#if !(NVSIZE <= IVSIZE)
1043# define new_XNV() new_body_allocated(SVt_NV)
1044#endif
65ac1738 1045#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1046#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1047
26359cfa
NC
1048#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1049 &PL_body_roots[SVt_PVGV])
1d7c1841 1050
bd81e77b 1051#endif /* PURIFY */
93e68bfb 1052
bd81e77b 1053/* no arena for you! */
93e68bfb 1054
bd81e77b 1055#define new_NOARENA(details) \
beeec492 1056 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1057#define new_NOARENAZ(details) \
beeec492 1058 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1059
1e30fcd5
NC
1060void *
1061Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1062 const size_t arena_size)
d2a0f284 1063{
d2a0f284 1064 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1065 struct arena_desc *adesc;
1066 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1067 unsigned int curr;
d2a0f284
JC
1068 char *start;
1069 const char *end;
02982131 1070 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
20b7effb
JH
1071#if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1072 dVAR;
1073#endif
a6098fe8 1074#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
23e9d66c
NC
1075 static bool done_sanity_check;
1076
a6098fe8 1077 /* PERL_GLOBAL_STRUCT cannot coexist with global
0b2d3faa 1078 * variables like done_sanity_check. */
10666ae3 1079 if (!done_sanity_check) {
ea471437 1080 unsigned int i = SVt_LAST;
10666ae3
NC
1081
1082 done_sanity_check = TRUE;
1083
1084 while (i--)
1085 assert (bodies_by_type[i].type == i);
1086 }
1087#endif
1088
02982131 1089 assert(arena_size);
23e9d66c 1090
99816f8d
NC
1091 /* may need new arena-set to hold new arena */
1092 if (!aroot || aroot->curr >= aroot->set_size) {
1093 struct arena_set *newroot;
1094 Newxz(newroot, 1, struct arena_set);
1095 newroot->set_size = ARENAS_PER_SET;
1096 newroot->next = aroot;
1097 aroot = newroot;
1098 PL_body_arenas = (void *) newroot;
1099 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1100 }
1101
1102 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1103 curr = aroot->curr++;
1104 adesc = &(aroot->set[curr]);
1105 assert(!adesc->arena);
1106
1107 Newx(adesc->arena, good_arena_size, char);
1108 adesc->size = good_arena_size;
1109 adesc->utype = sv_type;
147e3846 1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
99816f8d
NC
1111 curr, (void*)adesc->arena, (UV)good_arena_size));
1112
1113 start = (char *) adesc->arena;
d2a0f284 1114
29657bb6
NC
1115 /* Get the address of the byte after the end of the last body we can fit.
1116 Remember, this is integer division: */
02982131 1117 end = start + good_arena_size / body_size * body_size;
d2a0f284 1118
486ec47a 1119 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1120#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1121 DEBUG_m(PerlIO_printf(Perl_debug_log,
1122 "arena %p end %p arena-size %d (from %d) type %d "
1123 "size %d ct %d\n",
02982131
NC
1124 (void*)start, (void*)end, (int)good_arena_size,
1125 (int)arena_size, sv_type, (int)body_size,
1126 (int)good_arena_size / (int)body_size));
d8fca402 1127#else
d2a0f284
JC
1128 DEBUG_m(PerlIO_printf(Perl_debug_log,
1129 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1130 (void*)start, (void*)end,
02982131
NC
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
d8fca402 1133#endif
d2a0f284
JC
1134 *root = (void *)start;
1135
29657bb6
NC
1136 while (1) {
1137 /* Where the next body would start: */
d2a0f284 1138 char * const next = start + body_size;
29657bb6
NC
1139
1140 if (next >= end) {
1141 /* This is the last body: */
1142 assert(next == end);
1143
1144 *(void **)start = 0;
1145 return *root;
1146 }
1147
d2a0f284
JC
1148 *(void**) start = (void *)next;
1149 start = next;
1150 }
d2a0f284
JC
1151}
1152
1153/* grab a new thing from the free list, allocating more if necessary.
1154 The inline version is used for speed in hot routines, and the
1155 function using it serves the rest (unless PURIFY).
1156*/
1157#define new_body_inline(xpv, sv_type) \
1158 STMT_START { \
1159 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1160 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1161 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1162 bodies_by_type[sv_type].body_size,\
1163 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1164 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1165 } STMT_END
1166
1167#ifndef PURIFY
1168
1169STATIC void *
de37a194 1170S_new_body(pTHX_ const svtype sv_type)
d2a0f284 1171{
d2a0f284
JC
1172 void *xpv;
1173 new_body_inline(xpv, sv_type);
1174 return xpv;
1175}
1176
1177#endif
93e68bfb 1178
238b27b3
NC
1179static const struct body_details fake_rv =
1180 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1181
bd81e77b
NC
1182/*
1183=for apidoc sv_upgrade
93e68bfb 1184
bd81e77b
NC
1185Upgrade an SV to a more complex form. Generally adds a new body type to the
1186SV, then copies across as much information as possible from the old body.
9521ca61
FC
1187It croaks if the SV is already in a more complex form than requested. You
1188generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1189before calling C<sv_upgrade>, and hence does not croak. See also
fbe13c60 1190C<L</svtype>>.
93e68bfb 1191
bd81e77b 1192=cut
93e68bfb 1193*/
93e68bfb 1194
bd81e77b 1195void
5aaab254 1196Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1197{
bd81e77b
NC
1198 void* old_body;
1199 void* new_body;
42d0e0b7 1200 const svtype old_type = SvTYPE(sv);
d2a0f284 1201 const struct body_details *new_type_details;
238b27b3 1202 const struct body_details *old_type_details
bd81e77b 1203 = bodies_by_type + old_type;
ed7df46e 1204 SV *referent = NULL;
cac9b346 1205
7918f24d
NC
1206 PERL_ARGS_ASSERT_SV_UPGRADE;
1207
1776cbe8
NC
1208 if (old_type == new_type)
1209 return;
1210
1211 /* This clause was purposefully added ahead of the early return above to
1212 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1213 inference by Nick I-S that it would fix other troublesome cases. See
1214 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1215
1216 Given that shared hash key scalars are no longer PVIV, but PV, there is
1217 no longer need to unshare so as to free up the IVX slot for its proper
1218 purpose. So it's safe to move the early return earlier. */
1219
093085a8 1220 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1221 sv_force_normal_flags(sv, 0);
1222 }
cac9b346 1223
bd81e77b 1224 old_body = SvANY(sv);
de042e1d 1225
bd81e77b
NC
1226 /* Copying structures onto other structures that have been neatly zeroed
1227 has a subtle gotcha. Consider XPVMG
cac9b346 1228
bd81e77b
NC
1229 +------+------+------+------+------+-------+-------+
1230 | NV | CUR | LEN | IV | MAGIC | STASH |
1231 +------+------+------+------+------+-------+-------+
1232 0 4 8 12 16 20 24 28
645c22ef 1233
bd81e77b
NC
1234 where NVs are aligned to 8 bytes, so that sizeof that structure is
1235 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1236
bd81e77b
NC
1237 +------+------+------+------+------+-------+-------+------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1239 +------+------+------+------+------+-------+-------+------+
1240 0 4 8 12 16 20 24 28 32
08742458 1241
bd81e77b 1242 so what happens if you allocate memory for this structure:
30f9da9e 1243
bd81e77b
NC
1244 +------+------+------+------+------+-------+-------+------+------+...
1245 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1246 +------+------+------+------+------+-------+-------+------+------+...
1247 0 4 8 12 16 20 24 28 32 36
bfc44f79 1248
bd81e77b
NC
1249 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1250 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1251 started out as zero once, but it's quite possible that it isn't. So now,
1252 rather than a nicely zeroed GP, you have it pointing somewhere random.
1253 Bugs ensue.
bfc44f79 1254
bd81e77b
NC
1255 (In fact, GP ends up pointing at a previous GP structure, because the
1256 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1257 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1258 this happens to be moot because XPVGV has been re-ordered, with GP
1259 no longer after STASH)
30f9da9e 1260
bd81e77b
NC
1261 So we are careful and work out the size of used parts of all the
1262 structures. */
bfc44f79 1263
bd81e77b
NC
1264 switch (old_type) {
1265 case SVt_NULL:
1266 break;
1267 case SVt_IV:
4df7f6af 1268 if (SvROK(sv)) {
ed7df46e 1269 referent = SvRV(sv);
238b27b3
NC
1270 old_type_details = &fake_rv;
1271 if (new_type == SVt_NV)
1272 new_type = SVt_PVNV;
4df7f6af
NC
1273 } else {
1274 if (new_type < SVt_PVIV) {
1275 new_type = (new_type == SVt_NV)
1276 ? SVt_PVNV : SVt_PVIV;
1277 }
bd81e77b
NC
1278 }
1279 break;
1280 case SVt_NV:
1281 if (new_type < SVt_PVNV) {
1282 new_type = SVt_PVNV;
bd81e77b
NC
1283 }
1284 break;
bd81e77b
NC
1285 case SVt_PV:
1286 assert(new_type > SVt_PV);
6d59e610
LM
1287 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1288 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
bd81e77b
NC
1289 break;
1290 case SVt_PVIV:
1291 break;
1292 case SVt_PVNV:
1293 break;
1294 case SVt_PVMG:
1295 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1296 there's no way that it can be safely upgraded, because perl.c
1297 expects to Safefree(SvANY(PL_mess_sv)) */
1298 assert(sv != PL_mess_sv);
bd81e77b
NC
1299 break;
1300 default:
2439e033 1301 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1302 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1303 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1304 }
3376de98 1305
2439e033 1306 if (UNLIKELY(old_type > new_type))
3376de98
NC
1307 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1308 (int)old_type, (int)new_type);
1309
2fa1109b 1310 new_type_details = bodies_by_type + new_type;
645c22ef 1311
bd81e77b
NC
1312 SvFLAGS(sv) &= ~SVTYPEMASK;
1313 SvFLAGS(sv) |= new_type;
932e9ff9 1314
ab4416c0
NC
1315 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1316 the return statements above will have triggered. */
1317 assert (new_type != SVt_NULL);
bd81e77b 1318 switch (new_type) {
bd81e77b
NC
1319 case SVt_IV:
1320 assert(old_type == SVt_NULL);
dc6369ef 1321 SET_SVANY_FOR_BODYLESS_IV(sv);
bd81e77b
NC
1322 SvIV_set(sv, 0);
1323 return;
1324 case SVt_NV:
1325 assert(old_type == SVt_NULL);
5b306eef 1326#if NVSIZE <= IVSIZE
dc6369ef 1327 SET_SVANY_FOR_BODYLESS_NV(sv);
5b306eef 1328#else
bd81e77b 1329 SvANY(sv) = new_XNV();
5b306eef 1330#endif
bd81e77b
NC
1331 SvNV_set(sv, 0);
1332 return;
bd81e77b 1333 case SVt_PVHV:
bd81e77b 1334 case SVt_PVAV:
d2a0f284 1335 assert(new_type_details->body_size);
c1ae03ae
NC
1336
1337#ifndef PURIFY
1338 assert(new_type_details->arena);
d2a0f284 1339 assert(new_type_details->arena_size);
c1ae03ae 1340 /* This points to the start of the allocated area. */
d2a0f284
JC
1341 new_body_inline(new_body, new_type);
1342 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1343 new_body = ((char *)new_body) - new_type_details->offset;
1344#else
1345 /* We always allocated the full length item with PURIFY. To do this
1346 we fake things so that arena is false for all 16 types.. */
1347 new_body = new_NOARENAZ(new_type_details);
1348#endif
1349 SvANY(sv) = new_body;
1350 if (new_type == SVt_PVAV) {
1351 AvMAX(sv) = -1;
1352 AvFILLp(sv) = -1;
1353 AvREAL_only(sv);
64484faa 1354 if (old_type_details->body_size) {
ac572bf4
NC
1355 AvALLOC(sv) = 0;
1356 } else {
1357 /* It will have been zeroed when the new body was allocated.
1358 Lets not write to it, in case it confuses a write-back
1359 cache. */
1360 }
78ac7dd9
NC
1361 } else {
1362 assert(!SvOK(sv));
1363 SvOK_off(sv);
1364#ifndef NODEFAULT_SHAREKEYS
1365 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1366#endif
586fc6a3
SM
1367 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1368 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1369 }
aeb18a1e 1370
bd81e77b
NC
1371 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1372 The target created by newSVrv also is, and it can have magic.
1373 However, it never has SvPVX set.
1374 */
4df7f6af
NC
1375 if (old_type == SVt_IV) {
1376 assert(!SvROK(sv));
1377 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1378 assert(SvPVX_const(sv) == 0);
1379 }
aeb18a1e 1380
bd81e77b 1381 if (old_type >= SVt_PVMG) {
e736a858 1382 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1383 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1384 } else {
1385 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1386 }
1387 break;
93e68bfb 1388
bd81e77b
NC
1389 case SVt_PVIV:
1390 /* XXX Is this still needed? Was it ever needed? Surely as there is
1391 no route from NV to PVIV, NOK can never be true */
1392 assert(!SvNOKp(sv));
1393 assert(!SvNOK(sv));
2b5060ae 1394 /* FALLTHROUGH */
bd81e77b
NC
1395 case SVt_PVIO:
1396 case SVt_PVFM:
bd81e77b
NC
1397 case SVt_PVGV:
1398 case SVt_PVCV:
1399 case SVt_PVLV:
d361b004 1400 case SVt_INVLIST:
12c45b25 1401 case SVt_REGEXP:
bd81e77b
NC
1402 case SVt_PVMG:
1403 case SVt_PVNV:
1404 case SVt_PV:
93e68bfb 1405
d2a0f284 1406 assert(new_type_details->body_size);
bd81e77b
NC
1407 /* We always allocated the full length item with PURIFY. To do this
1408 we fake things so that arena is false for all 16 types.. */
1409 if(new_type_details->arena) {
1410 /* This points to the start of the allocated area. */
d2a0f284
JC
1411 new_body_inline(new_body, new_type);
1412 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1413 new_body = ((char *)new_body) - new_type_details->offset;
1414 } else {
1415 new_body = new_NOARENAZ(new_type_details);
1416 }
1417 SvANY(sv) = new_body;
5e2fc214 1418
bd81e77b 1419 if (old_type_details->copy) {
f9ba3d20
NC
1420 /* There is now the potential for an upgrade from something without
1421 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1422 int offset = old_type_details->offset;
1423 int length = old_type_details->copy;
1424
1425 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1426 const int difference
f9ba3d20
NC
1427 = new_type_details->offset - old_type_details->offset;
1428 offset += difference;
1429 length -= difference;
1430 }
1431 assert (length >= 0);
1432
1433 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1434 char);
bd81e77b
NC
1435 }
1436
1437#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1438 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1439 * correct 0.0 for us. Otherwise, if the old body didn't have an
1440 * NV slot, but the new one does, then we need to initialise the
1441 * freshly created NV slot with whatever the correct bit pattern is
1442 * for 0.0 */
e22a937e
NC
1443 if (old_type_details->zero_nv && !new_type_details->zero_nv
1444 && !isGV_with_GP(sv))
bd81e77b 1445 SvNV_set(sv, 0);
82048762 1446#endif
5e2fc214 1447
2439e033 1448 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1449 IO * const io = MUTABLE_IO(sv);
d963bf01 1450 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1451
1452 SvOBJECT_on(io);
1453 /* Clear the stashcache because a new IO could overrule a package
1454 name */
103f5a36 1455 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1456 hv_clear(PL_stashcache);
1457
85dca89a 1458 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1459 IoPAGE_LEN(sv) = 60;
85dca89a 1460 }
df6b4bd5 1461 if (old_type < SVt_PV) {
ed7df46e 1462 /* referent will be NULL unless the old type was SVt_IV emulating
4df7f6af 1463 SVt_RV */
ed7df46e 1464 sv->sv_u.svu_rv = referent;
4df7f6af 1465 }
bd81e77b
NC
1466 break;
1467 default:
afd78fd5
JH
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
bd81e77b 1470 }
73171d91 1471
5b306eef
DD
1472 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1473 and sometimes SVt_NV */
1474 if (old_type_details->body_size) {
bd81e77b 1475#ifdef PURIFY
beeec492 1476 safefree(old_body);
bd81e77b 1477#else
bc786448
GG
1478 /* Note that there is an assumption that all bodies of types that
1479 can be upgraded came from arenas. Only the more complex non-
1480 upgradable types are allowed to be directly malloc()ed. */
1481 assert(old_type_details->arena);
bd81e77b
NC
1482 del_body((void*)((char*)old_body + old_type_details->offset),
1483 &PL_body_roots[old_type]);
1484#endif
1485 }
1486}
73171d91 1487
bd81e77b
NC
1488/*
1489=for apidoc sv_backoff
73171d91 1490
fde67290 1491Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1492wrapper instead.
73171d91 1493
bd81e77b 1494=cut
73171d91
NC
1495*/
1496
fa7a1e49
DD
1497/* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1498 prior to 5.23.4 this function always returned 0
1499*/
1500
1501void
ddeaf645 1502Perl_sv_backoff(SV *const sv)
bd81e77b 1503{
69240efd 1504 STRLEN delta;
7a4bba22 1505 const char * const s = SvPVX_const(sv);
7918f24d
NC
1506
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
7918f24d 1508
bd81e77b
NC
1509 assert(SvOOK(sv));
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1512
69240efd
NC
1513 SvOOK_offset(sv, delta);
1514
7a4bba22
NC
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
bd81e77b 1517 SvFLAGS(sv) &= ~SVf_OOK;
fa7a1e49
DD
1518 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1519 return;
bd81e77b 1520}
73171d91 1521
03885497
DM
1522
1523/* forward declaration */
1524static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1525
1526
bd81e77b
NC
1527/*
1528=for apidoc sv_grow
73171d91 1529
bd81e77b
NC
1530Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1531upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1532Use the C<SvGROW> wrapper instead.
93e68bfb 1533
bd81e77b
NC
1534=cut
1535*/
93e68bfb 1536
e0060e30 1537
bd81e77b 1538char *
5aaab254 1539Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1540{
eb578fdb 1541 char *s;
93e68bfb 1542
7918f24d
NC
1543 PERL_ARGS_ASSERT_SV_GROW;
1544
bd81e77b
NC
1545 if (SvROK(sv))
1546 sv_unref(sv);
1547 if (SvTYPE(sv) < SVt_PV) {
1548 sv_upgrade(sv, SVt_PV);
1549 s = SvPVX_mutable(sv);
1550 }
1551 else if (SvOOK(sv)) { /* pv is offset? */
1552 sv_backoff(sv);
1553 s = SvPVX_mutable(sv);
1554 if (newlen > SvLEN(sv))
1555 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
bd81e77b
NC
1556 }
1557 else
db2c6cb3 1558 {
e0060e30 1559 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1560 s = SvPVX_mutable(sv);
db2c6cb3 1561 }
aeb18a1e 1562
93c10d60 1563#ifdef PERL_COPY_ON_WRITE
cbcb2a16 1564 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
3c239bea 1565 * to store the COW count. So in general, allocate one more byte than
cbcb2a16
DM
1566 * asked for, to make it likely this byte is always spare: and thus
1567 * make more strings COW-able.
fe546b38 1568 *
fa8f4f85
TC
1569 * Only increment if the allocation isn't MEM_SIZE_MAX,
1570 * otherwise it will wrap to 0.
1571 */
fe546b38 1572 if ( newlen != MEM_SIZE_MAX )
cbcb2a16
DM
1573 newlen++;
1574#endif
1575
ce861ea7
YO
1576#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1577#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1578#endif
1579
bd81e77b 1580 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559 1581 STRLEN minlen = SvCUR(sv);
3c239bea 1582 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
f1200559
WH
1583 if (newlen < minlen)
1584 newlen = minlen;
ce861ea7 1585#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
7c641603
KW
1586
1587 /* Don't round up on the first allocation, as odds are pretty good that
1588 * the initial request is accurate as to what is really needed */
ce861ea7 1589 if (SvLEN(sv)) {
9efda33a
TC
1590 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1591 if (rounded > newlen)
1592 newlen = rounded;
ce861ea7 1593 }
bd81e77b 1594#endif
98653f18 1595 if (SvLEN(sv) && s) {
10edeb5d 1596 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1597 }
1598 else {
10edeb5d 1599 s = (char*)safemalloc(newlen);
bd81e77b 1600 if (SvPVX_const(sv) && SvCUR(sv)) {
0a5fcc38 1601 Move(SvPVX_const(sv), s, SvCUR(sv), char);
bd81e77b
NC
1602 }
1603 }
1604 SvPV_set(sv, s);
ce861ea7 1605#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
98653f18
NC
1606 /* Do this here, do it once, do it right, and then we will never get
1607 called back into sv_grow() unless there really is some growing
1608 needed. */
ca7c1a29 1609 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1610#else
bd81e77b 1611 SvLEN_set(sv, newlen);
98653f18 1612#endif
bd81e77b
NC
1613 }
1614 return s;
1615}
aeb18a1e 1616
bd81e77b
NC
1617/*
1618=for apidoc sv_setiv
932e9ff9 1619
bd81e77b 1620Copies an integer into the given SV, upgrading first if necessary.
fbe13c60 1621Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
463ee0b2 1622
bd81e77b
NC
1623=cut
1624*/
463ee0b2 1625
bd81e77b 1626void
5aaab254 1627Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1628{
7918f24d
NC
1629 PERL_ARGS_ASSERT_SV_SETIV;
1630
bd81e77b
NC
1631 SV_CHECK_THINKFIRST_COW_DROP(sv);
1632 switch (SvTYPE(sv)) {
1633 case SVt_NULL:
bd81e77b 1634 case SVt_NV:
3376de98 1635 sv_upgrade(sv, SVt_IV);
bd81e77b 1636 break;
bd81e77b
NC
1637 case SVt_PV:
1638 sv_upgrade(sv, SVt_PVIV);
1639 break;
463ee0b2 1640
bd81e77b 1641 case SVt_PVGV:
6e592b3a
BM
1642 if (!isGV_with_GP(sv))
1643 break;
2165bd23 1644 /* FALLTHROUGH */
bd81e77b
NC
1645 case SVt_PVAV:
1646 case SVt_PVHV:
1647 case SVt_PVCV:
1648 case SVt_PVFM:
1649 case SVt_PVIO:
22e74366 1650 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1652 OP_DESC(PL_op));
c9a0dcdc 1653 NOT_REACHED; /* NOTREACHED */
0103ca14 1654 break;
42d0e0b7 1655 default: NOOP;
bd81e77b
NC
1656 }
1657 (void)SvIOK_only(sv); /* validate number */
1658 SvIV_set(sv, i);
1659 SvTAINT(sv);
1660}
932e9ff9 1661
bd81e77b
NC
1662/*
1663=for apidoc sv_setiv_mg
d33b2eba 1664
bd81e77b 1665Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1666
bd81e77b
NC
1667=cut
1668*/
d33b2eba 1669
bd81e77b 1670void
5aaab254 1671Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1672{
7918f24d
NC
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1674
bd81e77b
NC
1675 sv_setiv(sv,i);
1676 SvSETMAGIC(sv);
1677}
727879eb 1678
bd81e77b
NC
1679/*
1680=for apidoc sv_setuv
d33b2eba 1681
bd81e77b 1682Copies an unsigned integer into the given SV, upgrading first if necessary.
fbe13c60 1683Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
9b94d1dd 1684
bd81e77b
NC
1685=cut
1686*/
d33b2eba 1687
bd81e77b 1688void
5aaab254 1689Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1690{
7918f24d
NC
1691 PERL_ARGS_ASSERT_SV_SETUV;
1692
013abb9b
NC
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1694 possible:
bd81e77b 1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1696
bd81e77b
NC
1697 without
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1699
013abb9b
NC
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1703 */
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1706 return;
1707 }
1708 sv_setiv(sv, 0);
1709 SvIsUV_on(sv);
1710 SvUV_set(sv, u);
1711}
d33b2eba 1712
bd81e77b
NC
1713/*
1714=for apidoc sv_setuv_mg
727879eb 1715
bd81e77b 1716Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1717
bd81e77b
NC
1718=cut
1719*/
5e2fc214 1720
bd81e77b 1721void
5aaab254 1722Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1723{
7918f24d
NC
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1725
bd81e77b
NC
1726 sv_setuv(sv,u);
1727 SvSETMAGIC(sv);
1728}
5e2fc214 1729
954c1994 1730/*
bd81e77b 1731=for apidoc sv_setnv
954c1994 1732
bd81e77b 1733Copies a double into the given SV, upgrading first if necessary.
fbe13c60 1734Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
954c1994
GS
1735
1736=cut
1737*/
1738
63f97190 1739void
5aaab254 1740Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1741{
7918f24d
NC
1742 PERL_ARGS_ASSERT_SV_SETNV;
1743
bd81e77b
NC
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
79072805 1746 case SVt_NULL:
79072805 1747 case SVt_IV:
bd81e77b 1748 sv_upgrade(sv, SVt_NV);
79072805
LW
1749 break;
1750 case SVt_PV:
79072805 1751 case SVt_PVIV:
bd81e77b 1752 sv_upgrade(sv, SVt_PVNV);
79072805 1753 break;
bd4b1eb5 1754
bd4b1eb5 1755 case SVt_PVGV:
6e592b3a
BM
1756 if (!isGV_with_GP(sv))
1757 break;
2165bd23 1758 /* FALLTHROUGH */
bd81e77b
NC
1759 case SVt_PVAV:
1760 case SVt_PVHV:
79072805 1761 case SVt_PVCV:
bd81e77b
NC
1762 case SVt_PVFM:
1763 case SVt_PVIO:
22e74366 1764 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1765 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1766 OP_DESC(PL_op));
c9a0dcdc 1767 NOT_REACHED; /* NOTREACHED */
0103ca14 1768 break;
42d0e0b7 1769 default: NOOP;
2068cd4d 1770 }
bd81e77b
NC
1771 SvNV_set(sv, num);
1772 (void)SvNOK_only(sv); /* validate number */
1773 SvTAINT(sv);
79072805
LW
1774}
1775
645c22ef 1776/*
bd81e77b 1777=for apidoc sv_setnv_mg
645c22ef 1778
bd81e77b 1779Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1780
1781=cut
1782*/
1783
bd81e77b 1784void
5aaab254 1785Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1786{
7918f24d
NC
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1788
bd81e77b
NC
1789 sv_setnv(sv,num);
1790 SvSETMAGIC(sv);
79072805
LW
1791}
1792
3f7602fa
TC
1793/* Return a cleaned-up, printable version of sv, for non-numeric, or
1794 * not incrementable warning display.
1795 * Originally part of S_not_a_number().
1796 * The return value may be != tmpbuf.
bd81e77b 1797 */
954c1994 1798
3f7602fa
TC
1799STATIC const char *
1800S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1801 const char *pv;
94463019 1802
3f7602fa 1803 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1804
94463019 1805 if (DO_UTF8(sv)) {
3f7602fa 1806 SV *dsv = newSVpvs_flags("", SVs_TEMP);
37b8cdd1 1807 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
94463019
JH
1808 } else {
1809 char *d = tmpbuf;
3f7602fa 1810 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1811 /* each *s can expand to 4 chars + "...\0",
1812 i.e. need room for 8 chars */
ecdeb87c 1813
00b6aa41
AL
1814 const char *s = SvPVX_const(sv);
1815 const char * const end = s + SvCUR(sv);
1816 for ( ; s < end && d < limit; s++ ) {
94463019 1817 int ch = *s & 0xFF;
bd27cf70 1818 if (! isASCII(ch) && !isPRINT_LC(ch)) {
94463019
JH
1819 *d++ = 'M';
1820 *d++ = '-';
bd27cf70
KW
1821
1822 /* Map to ASCII "equivalent" of Latin1 */
1823 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
94463019
JH
1824 }
1825 if (ch == '\n') {
1826 *d++ = '\\';
1827 *d++ = 'n';
1828 }
1829 else if (ch == '\r') {
1830 *d++ = '\\';
1831 *d++ = 'r';
1832 }
1833 else if (ch == '\f') {
1834 *d++ = '\\';
1835 *d++ = 'f';
1836 }
1837 else if (ch == '\\') {
1838 *d++ = '\\';
1839 *d++ = '\\';
1840 }
1841 else if (ch == '\0') {
1842 *d++ = '\\';
1843 *d++ = '0';
1844 }
1845 else if (isPRINT_LC(ch))
1846 *d++ = ch;
1847 else {
1848 *d++ = '^';
1849 *d++ = toCTRL(ch);
1850 }
1851 }
1852 if (s < end) {
1853 *d++ = '.';
1854 *d++ = '.';
1855 *d++ = '.';
1856 }
1857 *d = '\0';
1858 pv = tmpbuf;
a0d0e21e 1859 }
a0d0e21e 1860
3f7602fa
TC
1861 return pv;
1862}
1863
1864/* Print an "isn't numeric" warning, using a cleaned-up,
1865 * printable version of the offending string
1866 */
1867
1868STATIC void
1869S_not_a_number(pTHX_ SV *const sv)
1870{
3f7602fa
TC
1871 char tmpbuf[64];
1872 const char *pv;
1873
1874 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1875
1876 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1877
533c011a 1878 if (PL_op)
9014280d 1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1880 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1881 "Argument \"%s\" isn't numeric in %s", pv,
1882 OP_DESC(PL_op));
a0d0e21e 1883 else
9014280d 1884 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1885 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1886 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1887}
1888
3f7602fa
TC
1889STATIC void
1890S_not_incrementable(pTHX_ SV *const sv) {
3f7602fa
TC
1891 char tmpbuf[64];
1892 const char *pv;
1893
1894 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1895
1896 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1897
1898 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1899 "Argument \"%s\" treated as 0 in increment (++)", pv);
1900}
1901
c2988b20
NC
1902/*
1903=for apidoc looks_like_number
1904
645c22ef
DM
1905Test if the content of an SV looks like a number (or is a number).
1906C<Inf> and C<Infinity> are treated as numbers (so will not issue a
796b6530 1907non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
f52e41ad 1908ignored.
c2988b20
NC
1909
1910=cut
1911*/
1912
1913I32
aad570aa 1914Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1915{
eb578fdb 1916 const char *sbegin;
c2988b20 1917 STRLEN len;
ea2485eb 1918 int numtype;
c2988b20 1919
7918f24d
NC
1920 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1921
f52e41ad
FC
1922 if (SvPOK(sv) || SvPOKp(sv)) {
1923 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1924 }
c2988b20 1925 else
e0ab1c0e 1926 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
ea2485eb
JH
1927 numtype = grok_number(sbegin, len, NULL);
1928 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
c2988b20 1929}
25da4f38 1930
19f6321d
NC
1931STATIC bool
1932S_glob_2number(pTHX_ GV * const gv)
180488f8 1933{
7918f24d
NC
1934 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1935
675c862f
AL
1936 /* We know that all GVs stringify to something that is not-a-number,
1937 so no need to test that. */
1938 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1939 {
1940 SV *const buffer = sv_newmortal();
1941 gv_efullname3(buffer, gv, "*");
675c862f 1942 not_a_number(buffer);
8e629ff4 1943 }
675c862f
AL
1944 /* We just want something true to return, so that S_sv_2iuv_common
1945 can tail call us and return true. */
19f6321d 1946 return TRUE;
675c862f
AL
1947}
1948
25da4f38
IZ
1949/* Actually, ISO C leaves conversion of UV to IV undefined, but
1950 until proven guilty, assume that things are not that bad... */
1951
645c22ef
DM
1952/*
1953 NV_PRESERVES_UV:
1954
1955 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1956 an IV (an assumption perl has been based on to date) it becomes necessary
1957 to remove the assumption that the NV always carries enough precision to
1958 recreate the IV whenever needed, and that the NV is the canonical form.
1959 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1960 precision as a side effect of conversion (which would lead to insanity
28e5dec8 1961 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
8a4a3196
KW
1962 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1963 where precision was lost, and IV/UV/NV slots that have a valid conversion
1964 which has lost no precision
645c22ef 1965 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1966 would lose precision, the precise conversion (or differently
1967 imprecise conversion) is also performed and cached, to prevent
1968 requests for different numeric formats on the same SV causing
1969 lossy conversion chains. (lossless conversion chains are perfectly
1970 acceptable (still))
1971
1972
1973 flags are used:
1974 SvIOKp is true if the IV slot contains a valid value
1975 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1976 SvNOKp is true if the NV slot contains a valid value
1977 SvNOK is true only if the NV value is accurate
1978
1979 so
645c22ef 1980 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1981 IV(or UV) would lose accuracy over a direct conversion from PV to
1982 IV(or UV). If it would, cache both conversions, return NV, but mark
1983 SV as IOK NOKp (ie not NOK).
1984
645c22ef 1985 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1986 NV would lose accuracy over a direct conversion from PV to NV. If it
1987 would, cache both conversions, flag similarly.
1988
1989 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1990 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1991 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1992 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1993 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1994
645c22ef
DM
1995 The benefit of this is that operations such as pp_add know that if
1996 SvIOK is true for both left and right operands, then integer addition
1997 can be used instead of floating point (for cases where the result won't
1998 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1999 loss of precision compared with integer addition.
2000
2001 * making IV and NV equal status should make maths accurate on 64 bit
2002 platforms
2003 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 2004 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
2005 looking for SvIOK and checking for overflow will not outweigh the
2006 fp to integer speedup)
2007 * will slow down integer operations (callers of SvIV) on "inaccurate"
2008 values, as the change from SvIOK to SvIOKp will cause a call into
2009 sv_2iv each time rather than a macro access direct to the IV slot
2010 * should speed up number->string conversion on integers as IV is
645c22ef 2011 favoured when IV and NV are equally accurate
28e5dec8
JH
2012
2013 ####################################################################
645c22ef
DM
2014 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2015 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2016 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
2017 ####################################################################
2018
645c22ef 2019 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
2020 performance ratio.
2021*/
2022
2023#ifndef NV_PRESERVES_UV
645c22ef
DM
2024# define IS_NUMBER_UNDERFLOW_IV 1
2025# define IS_NUMBER_UNDERFLOW_UV 2
2026# define IS_NUMBER_IV_AND_UV 2
2027# define IS_NUMBER_OVERFLOW_IV 4
2028# define IS_NUMBER_OVERFLOW_UV 5
2029
2030/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
2031
2032/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2033STATIC int
5aaab254 2034S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
2035# ifdef DEBUGGING
2036 , I32 numtype
2037# endif
2038 )
28e5dec8 2039{
7918f24d 2040 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
23491f1d 2041 PERL_UNUSED_CONTEXT;
7918f24d 2042
147e3846 2043 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
2044 if (SvNVX(sv) < (NV)IV_MIN) {
2045 (void)SvIOKp_on(sv);
2046 (void)SvNOK_on(sv);
45977657 2047 SvIV_set(sv, IV_MIN);
28e5dec8
JH
2048 return IS_NUMBER_UNDERFLOW_IV;
2049 }
2050 if (SvNVX(sv) > (NV)UV_MAX) {
2051 (void)SvIOKp_on(sv);
2052 (void)SvNOK_on(sv);
2053 SvIsUV_on(sv);
607fa7f2 2054 SvUV_set(sv, UV_MAX);
28e5dec8
JH
2055 return IS_NUMBER_OVERFLOW_UV;
2056 }
c2988b20
NC
2057 (void)SvIOKp_on(sv);
2058 (void)SvNOK_on(sv);
2059 /* Can't use strtol etc to convert this string. (See truth table in
2060 sv_2iv */
2061 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 2062 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2063 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2064 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2065 } else {
2066 /* Integer is imprecise. NOK, IOKp */
2067 }
2068 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2069 }
2070 SvIsUV_on(sv);
607fa7f2 2071 SvUV_set(sv, U_V(SvNVX(sv)));
659c4b96 2072 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2073 if (SvUVX(sv) == UV_MAX) {
2074 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2075 possibly be preserved by NV. Hence, it must be overflow.
2076 NOK, IOKp */
2077 return IS_NUMBER_OVERFLOW_UV;
2078 }
2079 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2080 } else {
2081 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2082 }
c2988b20 2083 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2084}
645c22ef
DM
2085#endif /* !NV_PRESERVES_UV*/
2086
a13f4dff 2087/* If numtype is infnan, set the NV of the sv accordingly.
5564cd7f 2088 * If numtype is anything else, try setting the NV using Atof(PV). */
3c81f0b3
DD
2089#ifdef USING_MSVC6
2090# pragma warning(push)
2091# pragma warning(disable:4756;disable:4056)
2092#endif
a13f4dff 2093static void
3823048b 2094S_sv_setnv(pTHX_ SV* sv, int numtype)
a13f4dff 2095{
07925c5e 2096 bool pok = cBOOL(SvPOK(sv));
5564cd7f 2097 bool nok = FALSE;
a7157111 2098#ifdef NV_INF
a13f4dff
JH
2099 if ((numtype & IS_NUMBER_INFINITY)) {
2100 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
5564cd7f 2101 nok = TRUE;
a7157111
JH
2102 } else
2103#endif
2104#ifdef NV_NAN
2105 if ((numtype & IS_NUMBER_NAN)) {
3823048b 2106 SvNV_set(sv, NV_NAN);
d48bd569 2107 nok = TRUE;
a7157111
JH
2108 } else
2109#endif
2110 if (pok) {
a13f4dff 2111 SvNV_set(sv, Atof(SvPVX_const(sv)));
d48bd569
JH
2112 /* Purposefully no true nok here, since we don't want to blow
2113 * away the possible IOK/UV of an existing sv. */
2114 }
5564cd7f 2115 if (nok) {
d48bd569 2116 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
5564cd7f
JH
2117 if (pok)
2118 SvPOK_on(sv); /* PV is okay, though. */
2119 }
a13f4dff 2120}
3c81f0b3
DD
2121#ifdef USING_MSVC6
2122# pragma warning(pop)
2123#endif
a13f4dff 2124
af359546 2125STATIC bool
7918f24d
NC
2126S_sv_2iuv_common(pTHX_ SV *const sv)
2127{
7918f24d
NC
2128 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2129
af359546 2130 if (SvNOKp(sv)) {
28e5dec8
JH
2131 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2132 * without also getting a cached IV/UV from it at the same time
2133 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2134 * IV or UV at same time to avoid this. */
2135 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2136
2137 if (SvTYPE(sv) == SVt_NV)
2138 sv_upgrade(sv, SVt_PVNV);
2139
28e5dec8
JH
2140 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2141 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2142 certainly cast into the IV range at IV_MAX, whereas the correct
2143 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2144 cases go to UV */
e91de695
JH
2145#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2146 if (Perl_isnan(SvNVX(sv))) {
2147 SvUV_set(sv, 0);
2148 SvIsUV_on(sv);
2149 return FALSE;
2150 }
2151#endif
28e5dec8 2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2153 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2154 if (SvNVX(sv) == (NV) SvIVX(sv)
28e5dec8 2155#ifndef NV_PRESERVES_UV
53e2bfb7 2156 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
28e5dec8
JH
2157 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2158 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2159 /* Don't flag it as "accurately an integer" if the number
2160 came from a (by definition imprecise) NV operation, and
2161 we're outside the range of NV integer precision */
2162#endif
2163 ) {
a43d94f2
NC
2164 if (SvNOK(sv))
2165 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2166 else {
2167 /* scalar has trailing garbage, eg "42a" */
2168 }
28e5dec8 2169 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2170 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
28e5dec8
JH
2171 PTR2UV(sv),
2172 SvNVX(sv),
2173 SvIVX(sv)));
2174
2175 } else {
2176 /* IV not precise. No need to convert from PV, as NV
2177 conversion would already have cached IV if it detected
2178 that PV->IV would be better than PV->NV->IV
2179 flags already correct - don't set public IOK. */
2180 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2181 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
28e5dec8
JH
2182 PTR2UV(sv),
2183 SvNVX(sv),
2184 SvIVX(sv)));
2185 }
2186 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2187 but the cast (NV)IV_MIN rounds to a the value less (more
2188 negative) than IV_MIN which happens to be equal to SvNVX ??
2189 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2190 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2191 (NV)UVX == NVX are both true, but the values differ. :-(
2192 Hopefully for 2s complement IV_MIN is something like
2193 0x8000000000000000 which will be exact. NWC */
d460ef45 2194 }
25da4f38 2195 else {
607fa7f2 2196 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8 2197 if (
659c4b96 2198 (SvNVX(sv) == (NV) SvUVX(sv))
28e5dec8
JH
2199#ifndef NV_PRESERVES_UV
2200 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2201 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2202 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2203 /* Don't flag it as "accurately an integer" if the number
2204 came from a (by definition imprecise) NV operation, and
2205 we're outside the range of NV integer precision */
2206#endif
a43d94f2 2207 && SvNOK(sv)
28e5dec8
JH
2208 )
2209 SvIOK_on(sv);
25da4f38 2210 SvIsUV_on(sv);
1c846c1f 2211 DEBUG_c(PerlIO_printf(Perl_debug_log,
147e3846 2212 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
56431972 2213 PTR2UV(sv),
57def98f
JH
2214 SvUVX(sv),
2215 SvUVX(sv)));
25da4f38 2216 }
748a9306 2217 }
cd84013a 2218 else if (SvPOKp(sv)) {
c2988b20 2219 UV value;
80e5abf2
DM
2220 int numtype;
2221 const char *s = SvPVX_const(sv);
2222 const STRLEN cur = SvCUR(sv);
2223
2224 /* short-cut for a single digit string like "1" */
2225
2226 if (cur == 1) {
2227 char c = *s;
2228 if (isDIGIT(c)) {
2229 if (SvTYPE(sv) < SVt_PVIV)
2230 sv_upgrade(sv, SVt_PVIV);
2231 (void)SvIOK_on(sv);
2232 SvIV_set(sv, (IV)(c - '0'));
2233 return FALSE;
2234 }
2235 }
2236
2237 numtype = grok_number(s, cur, &value);
af359546 2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2239 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
1c846c1f 2244
af359546 2245 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2246 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2247 cache the NV if we are sure it's not needed.
25da4f38 2248 */
16b7a9a4 2249
c2988b20
NC
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
5e045b90 2253 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2256 (void)SvIOK_on(sv);
c2988b20
NC
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2259
a13f4dff 2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
75a57a38 2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
6b322424 2262 not_a_number(sv);
3823048b 2263 S_sv_setnv(aTHX_ sv, numtype);
a13f4dff
JH
2264 return FALSE;
2265 }
2266
f2524eef 2267 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272#ifdef NV_PRESERVES_UV
2273 | IS_NUMBER_NOT_INT
2274#endif
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2278
2279 if (!(numtype & IS_NUMBER_NEG)) {
2280 /* positive */;
2281 if (value <= (UV)IV_MAX) {
45977657 2282 SvIV_set(sv, (IV)value);
c2988b20 2283 } else {
af359546 2284 /* it didn't overflow, and it was positive. */
607fa7f2 2285 SvUV_set(sv, value);
c2988b20
NC
2286 SvIsUV_on(sv);
2287 }
2288 } else {
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
53e2bfb7
DM
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
c2988b20
NC
2293 } else {
2294 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2295 I'm assuming it will be rare. */
c2988b20
NC
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2298 SvNOK_on(sv);
2299 SvIOK_off(sv);
2300 SvIOKp_on(sv);
9d6ce603 2301 SvNV_set(sv, -(NV)value);
45977657 2302 SvIV_set(sv, IV_MIN);
c2988b20
NC
2303 }
2304 }
2305 }
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2309
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
3823048b 2313 S_sv_setnv(aTHX_ sv, numtype);
28e5dec8 2314
c2988b20
NC
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2316 not_a_number(sv);
28e5dec8 2317
147e3846 2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
c2988b20 2319 PTR2UV(sv), SvNVX(sv)));
28e5dec8 2320
28e5dec8 2321#ifdef NV_PRESERVES_UV
af359546
NC
2322 (void)SvIOKp_on(sv);
2323 (void)SvNOK_on(sv);
e91de695
JH
2324#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2326 SvUV_set(sv, 0);
2327 SvIsUV_on(sv);
2328 return FALSE;
2329 }
2330#endif
af359546
NC
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2334 SvIOK_on(sv);
2335 } else {
6f207bd3 2336 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2337 }
2338 /* UV will not work better than IV */
2339 } else {
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2341 SvIsUV_on(sv);
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
af359546
NC
2344 } else {
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2349 SvIOK_on(sv);
af359546 2350 } else {
6f207bd3 2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2352 }
2353 }
4b0c9573 2354 SvIsUV_on(sv);
af359546 2355 }
28e5dec8 2356#else /* NV_PRESERVES_UV */
c2988b20
NC
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2359 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2360 grok_number above. The NV slot has just been set using
2361 Atof. */
560b0c46 2362 SvNOK_on(sv);
c2988b20
NC
2363 assert (SvIOKp(sv));
2364 } else {
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2369 SvNOK_on(sv);
45977657 2370 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
c2988b20
NC
2372 SvIOK_on(sv);
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2376 < (UV)IV_MAX)) {
147e3846 2377 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
2378 }
2379 } else {
2380 /* IN_UV NOT_INT
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
47031da6 2388# ifdef DEBUGGING
40a17c4c 2389 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2390# else
2391 sv_2iuv_non_preserve (sv);
2392# endif
c2988b20
NC
2393 }
2394 }
28e5dec8 2395#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2400 if (!numtype)
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2402 }
af359546
NC
2403 }
2404 else {
675c862f 2405 if (isGV_with_GP(sv))
159b6efe 2406 return glob_2number(MUTABLE_GV(sv));
180488f8 2407
4f62cd62 2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
af359546 2409 report_uninit(sv);
25da4f38
IZ
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
af359546
NC
2413 /* Return 0 from the caller. */
2414 return TRUE;
2415 }
2416 return FALSE;
2417}
2418
2419/*
2420=for apidoc sv_2iv_flags
2421
2422Return the integer value of an SV, doing any necessary string
c5608a1f 2423conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
af359546
NC
2424Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2425
2426=cut
2427*/
2428
2429IV
5aaab254 2430Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2431{
1061065f 2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
4bac9ae4 2433
217f6fa3
FC
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2436
4bac9ae4
CS
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2438 mg_get(sv);
2439
2440 if (SvROK(sv)) {
2441 if (SvAMAGIC(sv)) {
2442 SV * tmpstr;
2443 if (flags & SV_SKIP_OVERLOAD)
2444 return 0;
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2448 }
2449 }
2450 return PTR2IV(SvRV(sv));
2451 }
2452
8d919b0a 2453 if (SvVALID(sv) || isREGEXP(sv)) {
4e8879f3
DM
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2455 must not let them cache IVs.
2b2b6d6d
NC
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2459
2460 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2461 */
df6b4bd5 2462 assert(SvPOKp(sv));
e20b6c3b 2463 {
71c558c3 2464 UV value;
8d919b0a
FC
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2467 const int numtype
2468 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2469
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2475 return -(IV)value;
2476 } else {
2477 if (value < (UV)IV_MAX)
2478 return (IV)value;
2479 }
2480 }
058b8ae2 2481
e91de695
JH
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2487 }
2488
71c558c3
NC
2489 if (!numtype) {
2490 if (ckWARN(WARN_NUMERIC))
2491 not_a_number(sv);
2492 }
8d919b0a 2493 return I_V(Atof(ptr));
e20b6c3b 2494 }
4bac9ae4
CS
2495 }
2496
2497 if (SvTHINKFIRST(sv)) {
af359546
NC
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2500 report_uninit(sv);
2501 return 0;
2502 }
2503 }
4bac9ae4 2504
af359546
NC
2505 if (!SvIOKp(sv)) {
2506 if (S_sv_2iuv_common(aTHX_ sv))
2507 return 0;
79072805 2508 }
4bac9ae4 2509
147e3846 2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
1d7c1841 2511 PTR2UV(sv),SvIVX(sv)));
25da4f38 2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2513}
2514
645c22ef 2515/*
891f9566 2516=for apidoc sv_2uv_flags
645c22ef
DM
2517
2518Return the unsigned integer value of an SV, doing any necessary string
c5608a1f 2519conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
891f9566 2520Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2521
2522=cut
2523*/
2524
ff68c719 2525UV
5aaab254 2526Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2527{
1061065f 2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
4bac9ae4
CS
2529
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2531 mg_get(sv);
2532
2533 if (SvROK(sv)) {
2534 if (SvAMAGIC(sv)) {
2535 SV *tmpstr;
2536 if (flags & SV_SKIP_OVERLOAD)
2537 return 0;
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2541 }
2542 }
2543 return PTR2UV(SvRV(sv));
2544 }
2545
8d919b0a 2546 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
df6b4bd5 2550 assert(SvPOKp(sv));
e20b6c3b 2551 {
71c558c3 2552 UV value;
8d919b0a
FC
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2555 const int numtype
2556 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2557
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2562 return value;
2563 }
058b8ae2 2564
e91de695
JH
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2570 }
2571
71c558c3
NC
2572 if (!numtype) {
2573 if (ckWARN(WARN_NUMERIC))
2574 not_a_number(sv);
2575 }
8d919b0a 2576 return U_V(Atof(ptr));
e20b6c3b 2577 }
4bac9ae4
CS
2578 }
2579
2580 if (SvTHINKFIRST(sv)) {
0336b60e 2581 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2582 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2583 report_uninit(sv);
ff68c719
PP
2584 return 0;
2585 }
2586 }
4bac9ae4 2587
af359546
NC
2588 if (!SvIOKp(sv)) {
2589 if (S_sv_2iuv_common(aTHX_ sv))
2590 return 0;
ff68c719 2591 }
25da4f38 2592
147e3846 2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
1d7c1841 2594 PTR2UV(sv),SvUVX(sv)));
25da4f38 2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2596}
2597
645c22ef 2598/*
196007d1 2599=for apidoc sv_2nv_flags
645c22ef
DM
2600
2601Return the num value of an SV, doing any necessary string or integer
c5608a1f 2602conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
39d5de13 2603Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2604
2605=cut
2606*/
2607
65202027 2608NV
5aaab254 2609Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2610{
1061065f
DD
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2612
217f6fa3
FC
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2619 const char *ptr;
39d5de13
DM
2620 if (flags & SV_GMAGIC)
2621 mg_get(sv);
463ee0b2
LW
2622 if (SvNOKp(sv))
2623 return SvNVX(sv);
cd84013a 2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a 2625 ptr = SvPVX_const(sv);
041457d9 2626 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2627 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2628 not_a_number(sv);
8d919b0a 2629 return Atof(ptr);
a0d0e21e 2630 }
25da4f38 2631 if (SvIOKp(sv)) {
1c846c1f 2632 if (SvIsUV(sv))
65202027 2633 return (NV)SvUVX(sv);
25da4f38 2634 else
65202027 2635 return (NV)SvIVX(sv);
47a72cb8
NC
2636 }
2637 if (SvROK(sv)) {
2638 goto return_rok;
2639 }
2640 assert(SvTYPE(sv) >= SVt_PVMG);
2641 /* This falls through to the report_uninit near the end of the
2642 function. */
2643 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2644 if (SvROK(sv)) {
47a72cb8 2645 return_rok:
deb46114 2646 if (SvAMAGIC(sv)) {
aee036bb
DM
2647 SV *tmpstr;
2648 if (flags & SV_SKIP_OVERLOAD)
2649 return 0;
31d632c3 2650 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2651 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2652 return SvNV(tmpstr);
2653 }
2654 }
2655 return PTR2NV(SvRV(sv));
a0d0e21e 2656 }
0336b60e 2657 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2658 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2659 report_uninit(sv);
ed6116ce
LW
2660 return 0.0;
2661 }
79072805
LW
2662 }
2663 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2664 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2665 sv_upgrade(sv, SVt_NV);
b2f82b52 2666 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
097ee67d 2667 DEBUG_c({
688523a0
KW
2668 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2669 STORE_LC_NUMERIC_SET_STANDARD();
1d7c1841 2670 PerlIO_printf(Perl_debug_log,
147e3846 2671 "0x%" UVxf " num(%" NVgf ")\n",
1d7c1841 2672 PTR2UV(sv), SvNVX(sv));
688523a0 2673 RESTORE_LC_NUMERIC();
097ee67d 2674 });
b2f82b52
KW
2675 CLANG_DIAG_RESTORE_STMT;
2676
79072805
LW
2677 }
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2680 if (SvNOKp(sv)) {
2681 return SvNVX(sv);
61604483 2682 }
59d8ce62 2683 if (SvIOKp(sv)) {
9d6ce603 2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2685#ifdef NV_PRESERVES_UV
a43d94f2
NC
2686 if (SvIOK(sv))
2687 SvNOK_on(sv);
2688 else
2689 SvNOKp_on(sv);
28e5dec8
JH
2690#else
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2693 if (SvIOK(sv) &&
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2696 SvNOK_on(sv);
2697 else
2698 SvNOKp_on(sv);
2699#endif
93a17b20 2700 }
cd84013a 2701 else if (SvPOKp(sv)) {
c2988b20 2702 UV value;
3823048b 2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2705 not_a_number(sv);
28e5dec8 2706#ifdef NV_PRESERVES_UV
c2988b20
NC
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
5e045b90 2709 /* It's definitely an integer */
9d6ce603 2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
66d83377 2711 } else {
3823048b 2712 S_sv_setnv(aTHX_ sv, numtype);
66d83377 2713 }
a43d94f2
NC
2714 if (numtype)
2715 SvNOK_on(sv);
2716 else
2717 SvNOKp_on(sv);
28e5dec8 2718#else
e91de695
JH
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2725 UV_BITS */
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
c2988b20
NC
2732 SvNOK_on(sv);
2733 } else {
e91de695 2734 /* value has been set. It may not be precise. */
53e2bfb7 2735 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
e91de695
JH
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
c2988b20 2738 } else {
e91de695
JH
2739 SvNOKp_on(sv);
2740 SvIOKp_on(sv);
6fa402ec 2741
e91de695 2742 if (numtype & IS_NUMBER_NEG) {
02b08bbc
DM
2743 /* -IV_MIN is undefined, but we should never reach
2744 * this point with both IS_NUMBER_NEG and value ==
2745 * (UV)IV_MIN */
2746 assert(value != (UV)IV_MIN);
e91de695
JH
2747 SvIV_set(sv, -(IV)value);
2748 } else if (value <= (UV)IV_MAX) {
2749 SvIV_set(sv, (IV)value);
2750 } else {
2751 SvUV_set(sv, value);
2752 SvIsUV_on(sv);
2753 }
c2988b20 2754
e91de695
JH
2755 if (numtype & IS_NUMBER_NOT_INT) {
2756 /* I believe that even if the original PV had decimals,
2757 they are lost beyond the limit of the FP precision.
2758 However, neither is canonical, so both only get p
2759 flags. NWC, 2000/11/25 */
2760 /* Both already have p flags, so do nothing */
2761 } else {
2762 const NV nv = SvNVX(sv);
2763 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2764 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2765 if (SvIVX(sv) == I_V(nv)) {
2766 SvNOK_on(sv);
2767 } else {
2768 /* It had no "." so it must be integer. */
2769 }
2770 SvIOK_on(sv);
0f83c5a4 2771 } else {
e91de695
JH
2772 /* between IV_MAX and NV(UV_MAX).
2773 Could be slightly > UV_MAX */
2774
2775 if (numtype & IS_NUMBER_NOT_INT) {
2776 /* UV and NV both imprecise. */
0f83c5a4 2777 } else {
e91de695
JH
2778 const UV nv_as_uv = U_V(nv);
2779
2780 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2781 SvNOK_on(sv);
c2988b20 2782 }
e91de695 2783 SvIOK_on(sv);
c2988b20
NC
2784 }
2785 }
2786 }
2787 }
0f83c5a4 2788 }
e91de695
JH
2789 /* It might be more code efficient to go through the entire logic above
2790 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2791 gets complex and potentially buggy, so more programmer efficient
2792 to do it this way, by turning off the public flags: */
2793 if (!numtype)
2794 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2795#endif /* NV_PRESERVES_UV */
93a17b20 2796 }
79072805 2797 else {
e91de695
JH
2798 if (isGV_with_GP(sv)) {
2799 glob_2number(MUTABLE_GV(sv));
2800 return 0.0;
2801 }
180488f8 2802
e91de695
JH
2803 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2804 report_uninit(sv);
2805 assert (SvTYPE(sv) >= SVt_NV);
2806 /* Typically the caller expects that sv_any is not NULL now. */
2807 /* XXX Ilya implies that this is a bug in callers that assume this
2808 and ideally should be fixed. */
2809 return 0.0;
79072805 2810 }
b2f82b52 2811 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
097ee67d 2812 DEBUG_c({
688523a0
KW
2813 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2814 STORE_LC_NUMERIC_SET_STANDARD();
147e3846 2815 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
e91de695 2816 PTR2UV(sv), SvNVX(sv));
688523a0 2817 RESTORE_LC_NUMERIC();
e91de695 2818 });
b2f82b52 2819 CLANG_DIAG_RESTORE_STMT;
463ee0b2 2820 return SvNVX(sv);
79072805
LW
2821}
2822
800401ee
JH
2823/*
2824=for apidoc sv_2num
2825
2826Return an SV with the numeric value of the source SV, doing any necessary
d024d1a7
FC
2827reference or overload conversion. The caller is expected to have handled
2828get-magic already.
800401ee
JH
2829
2830=cut
2831*/
2832
2833SV *
5aaab254 2834Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2835{
7918f24d
NC
2836 PERL_ARGS_ASSERT_SV_2NUM;
2837
b9ee0594
RGS
2838 if (!SvROK(sv))
2839 return sv;
800401ee 2840 if (SvAMAGIC(sv)) {
31d632c3 2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2845 }
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2847}
2848
dd0a5f5f
TK
2849/* int2str_table: lookup table containing string representations of all
2850 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2851 * int2str_table.arr[12*2] is "12".
2852 *
2853 * We are going to read two bytes at a time, so we have to ensure that
2854 * the array is aligned to a 2 byte boundary. That's why it was made a
2855 * union with a dummy U16 member. */
2856static const union {
2857 char arr[200];
2858 U16 dummy;
2859} int2str_table = {{
2860 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2861 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2862 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2863 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2864 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2865 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2866 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2867 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2868 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2869 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2870 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2871 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2872 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2873 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2874 '9', '8', '9', '9'
2875}};
2876
645c22ef
DM
2877/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2878 * UV as a string towards the end of buf, and return pointers to start and
2879 * end of it.
2880 *
2881 * We assume that buf is at least TYPE_CHARS(UV) long.
2882 */
2883
dd0a5f5f 2884PERL_STATIC_INLINE char *
5de3775c 2885S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2886{
25da4f38 2887 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2888 char * const ebuf = ptr;
25da4f38 2889 int sign;
dd0a5f5f 2890 U16 *word_ptr, *word_table;
25da4f38 2891
7918f24d
NC
2892 PERL_ARGS_ASSERT_UIV_2BUF;
2893
dd0a5f5f
TK
2894 /* ptr has to be properly aligned, because we will cast it to U16* */
2895 assert(PTR2nat(ptr) % 2 == 0);
2896 /* we are going to read/write two bytes at a time */
2897 word_ptr = (U16*)ptr;
2898 word_table = (U16*)int2str_table.arr;
2899
2900 if (UNLIKELY(is_uv))
25da4f38
IZ
2901 sign = 0;
2902 else if (iv >= 0) {
2903 uv = iv;
2904 sign = 0;
2905 } else {
7895b980
KW
2906 /* Using 0- here to silence bogus warning from MS VC */
2907 uv = (UV) (0 - (UV) iv);
25da4f38
IZ
2908 sign = 1;
2909 }
dd0a5f5f
TK
2910
2911 while (uv > 99) {
2912 *--word_ptr = word_table[uv % 100];
2913 uv /= 100;
2914 }
2915 ptr = (char*)word_ptr;
2916
2917 if (uv < 10)
2918 *--ptr = (char)uv + '0';
2919 else {
2920 *--word_ptr = word_table[uv];
2921 ptr = (char*)word_ptr;
2922 }
2923
25da4f38 2924 if (sign)
dd0a5f5f
TK
2925 *--ptr = '-';
2926
25da4f38
IZ
2927 *peob = ebuf;
2928 return ptr;
2929}
2930
bfaa02d5
JH
2931/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2932 * infinity or a not-a-number, writes the appropriate strings to the
2933 * buffer, including a zero byte. On success returns the written length,
3bde2d43
JH
2934 * excluding the zero byte, on failure (not an infinity, not a nan)
2935 * returns zero, assert-fails on maxlen being too short.
3823048b
JH
2936 *
2937 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2938 * shared string constants we point to, instead of generating a new
2939 * string for each instance. */
bfaa02d5 2940STATIC size_t
3823048b 2941S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
3bde2d43 2942 char* s = buffer;
bfaa02d5 2943 assert(maxlen >= 4);
3bde2d43
JH
2944 if (Perl_isinf(nv)) {
2945 if (nv < 0) {
2946 if (maxlen < 5) /* "-Inf\0" */
2947 return 0;
2948 *s++ = '-';
2949 } else if (plus) {
2950 *s++ = '+';
6e915616 2951 }
3bde2d43
JH
2952 *s++ = 'I';
2953 *s++ = 'n';
2954 *s++ = 'f';
2955 }
2956 else if (Perl_isnan(nv)) {
2957 *s++ = 'N';
2958 *s++ = 'a';
2959 *s++ = 'N';
2960 /* XXX optionally output the payload mantissa bits as
2961 * "(unsigned)" (to match the nan("...") C99 function,
2962 * or maybe as "(0xhhh...)" would make more sense...
2963 * provide a format string so that the user can decide?
2964 * NOTE: would affect the maxlen and assert() logic.*/
2965 }
2966 else {
2967 return 0;
bfaa02d5 2968 }
3bde2d43 2969 assert((s == buffer + 3) || (s == buffer + 4));
defe49c8
AL
2970 *s = 0;
2971 return s - buffer;
bfaa02d5
JH
2972}
2973
2974/*
2975=for apidoc sv_2pv_flags
2976
796b6530 2977Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
c5608a1f 2978If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
bfaa02d5
JH
2979string if necessary. Normally invoked via the C<SvPV_flags> macro.
2980C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2981
2982=cut
2983*/
2984
2985char *
2986Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2987{
2988 char *s;
2989
2990 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2991
2992 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2993 && SvTYPE(sv) != SVt_PVFM);
2994 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2995 mg_get(sv);
2996 if (SvROK(sv)) {
2997 if (SvAMAGIC(sv)) {
2998 SV *tmpstr;
2999 if (flags & SV_SKIP_OVERLOAD)
3000 return NULL;
3001 tmpstr = AMG_CALLunary(sv, string_amg);
3002 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3003 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3004 /* Unwrap this: */
3005 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3006 */
3007
3008 char *pv;
3009 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3010 if (flags & SV_CONST_RETURN) {
3011 pv = (char *) SvPVX_const(tmpstr);
3012 } else {
3013 pv = (flags & SV_MUTABLE_RETURN)
3014 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3015 }
3016 if (lp)
3017 *lp = SvCUR(tmpstr);
3018 } else {
3019 pv = sv_2pv_flags(tmpstr, lp, flags);
3020 }
3021 if (SvUTF8(tmpstr))
3022 SvUTF8_on(sv);
3023 else
3024 SvUTF8_off(sv);
3025 return pv;
3026 }
3027 }
3028 {
3029 STRLEN len;
3030 char *retval;
3031 char *buffer;
3032 SV *const referent = SvRV(sv);
3033
3034 if (!referent) {
3035 len = 7;
3036 retval = buffer = savepvn("NULLREF", len);
3037 } else if (SvTYPE(referent) == SVt_REGEXP &&
3038 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3039 amagic_is_enabled(string_amg))) {
3040 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3041
3042 assert(re);
3043
3044 /* If the regex is UTF-8 we want the containing scalar to
3045 have an UTF-8 flag too */
3046 if (RX_UTF8(re))
3047 SvUTF8_on(sv);
3048 else
3049 SvUTF8_off(sv);
3050
3051 if (lp)
3052 *lp = RX_WRAPLEN(re);
3053
3054 return RX_WRAPPED(re);
3055 } else {
3056 const char *const typestr = sv_reftype(referent, 0);
3057 const STRLEN typelen = strlen(typestr);
3058 UV addr = PTR2UV(referent);
3059 const char *stashname = NULL;
3060 STRLEN stashnamelen = 0; /* hush, gcc */
3061 const char *buffer_end;
3062
3063 if (SvOBJECT(referent)) {
3064 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3065
3066 if (name) {
3067 stashname = HEK_KEY(name);
3068 stashnamelen = HEK_LEN(name);
3069
3070 if (HEK_UTF8(name)) {
3071 SvUTF8_on(sv);
3072 } else {
3073 SvUTF8_off(sv);
3074 }
3075 } else {
3076 stashname = "__ANON__";
3077 stashnamelen = 8;
3078 }
3079 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3080 + 2 * sizeof(UV) + 2 /* )\0 */;
3081 } else {
3082 len = typelen + 3 /* (0x */
3083 + 2 * sizeof(UV) + 2 /* )\0 */;
3084 }
fafee734 3085
4bac9ae4
CS
3086 Newx(buffer, len, char);
3087 buffer_end = retval = buffer + len;
3088
3089 /* Working backwards */
3090 *--retval = '\0';
3091 *--retval = ')';
3092 do {
3093 *--retval = PL_hexdigit[addr & 15];
3094 } while (addr >>= 4);
3095 *--retval = 'x';
3096 *--retval = '0';
3097 *--retval = '(';
3098
3099 retval -= typelen;
3100 memcpy(retval, typestr, typelen);
3101
3102 if (stashname) {
3103 *--retval = '=';
3104 retval -= stashnamelen;
3105 memcpy(retval, stashname, stashnamelen);
c080367d 3106 }
4bac9ae4
CS
3107 /* retval may not necessarily have reached the start of the
3108 buffer here. */
3109 assert (retval >= buffer);
3110
3111 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 3112 }
cdb061a3 3113 if (lp)
4bac9ae4
CS
3114 *lp = len;
3115 SAVEFREEPV(buffer);
3116 return retval;
79072805 3117 }
79072805 3118 }
4bac9ae4
CS
3119
3120 if (SvPOKp(sv)) {
3121 if (lp)
3122 *lp = SvCUR(sv);
3123 if (flags & SV_MUTABLE_RETURN)
3124 return SvPVX_mutable(sv);
3125 if (flags & SV_CONST_RETURN)
3126 return (char *)SvPVX_const(sv);
3127 return SvPVX(sv);
3128 }
3129
3130 if (SvIOK(sv)) {
28e5dec8
JH
3131 /* I'm assuming that if both IV and NV are equally valid then
3132 converting the IV is going to be more efficient */
e1ec3a88 3133 const U32 isUIOK = SvIsUV(sv);
dd0a5f5f
TK
3134 /* The purpose of this union is to ensure that arr is aligned on
3135 a 2 byte boundary, because that is what uiv_2buf() requires */
3136 union {
3137 char arr[TYPE_CHARS(UV)];
3138 U16 dummy;
3139 } buf;
28e5dec8 3140 char *ebuf, *ptr;
97a130b8 3141 STRLEN len;
28e5dec8
JH
3142
3143 if (SvTYPE(sv) < SVt_PVIV)
3144 sv_upgrade(sv, SVt_PVIV);
dd0a5f5f 3145 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 3146 len = ebuf - ptr;
5902b6a9 3147 /* inlined from sv_setpvn */
97a130b8
NC
3148 s = SvGROW_mutable(sv, len + 1);
3149 Move(ptr, s, len, char);
3150 s += len;
28e5dec8 3151 *s = '\0';
b127e37e 3152 SvPOK_on(sv);
28e5dec8 3153 }
4bac9ae4 3154 else if (SvNOK(sv)) {
79072805
LW
3155 if (SvTYPE(sv) < SVt_PVNV)
3156 sv_upgrade(sv, SVt_PVNV);
128eeacb
DD
3157 if (SvNVX(sv) == 0.0
3158#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3159 && !Perl_isnan(SvNVX(sv))
3160#endif
3161 ) {
29912d93
NC
3162 s = SvGROW_mutable(sv, 2);
3163 *s++ = '0';
3164 *s = '\0';
3165 } else {
5e85836e 3166 STRLEN len;
fb8cdbc5 3167 STRLEN size = 5; /* "-Inf\0" */
0c7e610f 3168
fb8cdbc5 3169 s = SvGROW_mutable(sv, size);
3823048b 3170 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
fb8cdbc5 3171 if (len > 0) {
0c7e610f 3172 s += len;
fb8cdbc5
JH
3173 SvPOK_on(sv);
3174 }
0c7e610f 3175 else {
0c7e610f 3176 /* some Xenix systems wipe out errno here */
fb8cdbc5
JH
3177 dSAVE_ERRNO;
3178
3840bff0
JH
3179 size =
3180 1 + /* sign */
3181 1 + /* "." */
3182 NV_DIG +
3183 1 + /* "e" */
3184 1 + /* sign */
3185 5 + /* exponent digits */
3186 1 + /* \0 */
3187 2; /* paranoia */
b127e37e 3188
fb8cdbc5 3189 s = SvGROW_mutable(sv, size);
b127e37e 3190#ifndef USE_LOCALE_NUMERIC
a4eca1d4
JH
3191 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3192
0c7e610f
JH
3193 SvPOK_on(sv);
3194#else
28acfe03 3195 {
3840bff0 3196 bool local_radix;
67d796ae
KW
3197 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3198 STORE_LC_NUMERIC_SET_TO_NEEDED();
3840bff0 3199
dd2dbc5f 3200 local_radix = _NOT_IN_NUMERIC_STANDARD;
4c039fd8
DM
3201 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3202 size += SvCUR(PL_numeric_radix_sv) - 1;
3840bff0
JH
3203 s = SvGROW_mutable(sv, size);
3204 }
3205
a4eca1d4 3206 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
0c7e610f
JH
3207
3208 /* If the radix character is UTF-8, and actually is in the
3209 * output, turn on the UTF-8 flag for the scalar */
3dbc6af5
KW
3210 if ( local_radix
3211 && SvUTF8(PL_numeric_radix_sv)
3212 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3213 {
3840bff0
JH
3214 SvUTF8_on(sv);
3215 }
3216
0c7e610f 3217 RESTORE_LC_NUMERIC();
28acfe03 3218 }
68e8f474 3219
0c7e610f
JH
3220 /* We don't call SvPOK_on(), because it may come to
3221 * pass that the locale changes so that the
3222 * stringification we just did is no longer correct. We
3223 * will have to re-stringify every time it is needed */
b127e37e 3224#endif
0c7e610f
JH
3225 RESTORE_ERRNO;
3226 }
3227 while (*s) s++;
bbce6d69 3228 }
79072805 3229 }
4bac9ae4
CS
3230 else if (isGV_with_GP(sv)) {
3231 GV *const gv = MUTABLE_GV(sv);
3232 SV *const buffer = sv_newmortal();
8d1c3e26 3233
4bac9ae4 3234 gv_efullname3(buffer, gv, "*");
180488f8 3235
4bac9ae4
CS
3236 assert(SvPOK(buffer));
3237 if (SvUTF8(buffer))
3238 SvUTF8_on(sv);
1097da16
TC
3239 else
3240 SvUTF8_off(sv);
4bac9ae4
CS
3241 if (lp)
3242 *lp = SvCUR(buffer);
3243 return SvPVX(buffer);
3244 }
3245 else {
cdb061a3 3246 if (lp)
00b6aa41 3247 *lp = 0;
9f621bb0
NC
3248 if (flags & SV_UNDEF_RETURNS_NULL)
3249 return NULL;
4f62cd62 3250 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3251 report_uninit(sv);
4bac9ae4
CS
3252 /* Typically the caller expects that sv_any is not NULL now. */
3253 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3254 sv_upgrade(sv, SVt_PV);
73d840c0 3255 return (char *)"";
79072805 3256 }
4bac9ae4 3257
cdb061a3 3258 {
823a54a3 3259 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3260 if (lp)
3261 *lp = len;
3262 SvCUR_set(sv, len);
3263 }
147e3846 3264 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3f7c398e 3265 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3266 if (flags & SV_CONST_RETURN)
3267 return (char *)SvPVX_const(sv);
10516c54
NC
3268 if (flags & SV_MUTABLE_RETURN)
3269 return SvPVX_mutable(sv);
463ee0b2
LW
3270 return SvPVX(sv);
3271}
3272
645c22ef 3273/*
6050d10e
JP
3274=for apidoc sv_copypv
3275
3276Copies a stringified representation of the source SV into the
796b6530 3277destination SV. Automatically performs any necessary C<mg_get> and
54f0641b 3278coercion of numeric values into strings. Guaranteed to preserve
796b6530
KW
3279C<UTF8> flag even from overloaded objects. Similar in nature to
3280C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3281string. Mostly uses C<sv_2pv_flags> to do its work, except when that
6050d10e
JP
3282would lose the UTF-8'ness of the PV.
3283
4bac9ae4
CS
3284=for apidoc sv_copypv_nomg
3285
796b6530 3286Like C<sv_copypv>, but doesn't invoke get magic first.
4bac9ae4
CS
3287
3288=for apidoc sv_copypv_flags
3289
796b6530 3290Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
c5608a1f 3291has the C<SV_GMAGIC> bit set.
4bac9ae4 3292
6050d10e
JP
3293=cut
3294*/
3295
3296void
5aaab254 3297Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3298{
446eaa42 3299 STRLEN len;
4bac9ae4 3300 const char *s;
7918f24d 3301
4bac9ae4 3302 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3303
c77ed9ca 3304 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
cb50f42d 3305 sv_setpvn(dsv,s,len);
446eaa42 3306 if (SvUTF8(ssv))
cb50f42d 3307 SvUTF8_on(dsv);
446eaa42 3308 else
cb50f42d 3309 SvUTF8_off(dsv);
6050d10e
JP
3310}
3311
3312/*
645c22ef
DM
3313=for apidoc sv_2pvbyte
3314
796b6530 3315Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
1e54db1a 3316to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3317side-effect.
3318
3319Usually accessed via the C<SvPVbyte> macro.
3320
3321=cut
3322*/
3323
7340a771 3324char *
757fc329 3325Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
7340a771 3326{
757fc329 3327 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS;
7918f24d 3328
757fc329
P
3329 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3330 mg_get(sv);
4499db73
FC
3331 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3332 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3333 SV *sv2 = sv_newmortal();
48120f8f 3334 sv_copypv_nomg(sv2,sv);
a901b181
FC
3335 sv = sv2;
3336 }
757fc329 3337 sv_utf8_downgrade_nomg(sv,0);
71eb6d8c 3338 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3339}
3340
645c22ef 3341/*
035cbb0e
RGS
3342=for apidoc sv_2pvutf8
3343
796b6530 3344Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
035cbb0e
RGS
3345to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3346
3347Usually accessed via the C<SvPVutf8> macro.
3348
3349=cut
3350*/
645c22ef 3351
7340a771 3352char *
757fc329 3353Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
7340a771 3354{
757fc329 3355 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS;
7918f24d 3356
757fc329
P
3357 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3358 mg_get(sv);
4499db73 3359 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
757fc329
P
3360 || isGV_with_GP(sv) || SvROK(sv)) {
3361 SV *sv2 = sv_newmortal();
3362 sv_copypv_nomg(sv2,sv);
3363 sv = sv2;
3364 }
4bac9ae4 3365 sv_utf8_upgrade_nomg(sv);
c3ec315f 3366 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3367}
1c846c1f 3368
7ee2227d 3369
645c22ef
DM
3370/*
3371=for apidoc sv_2bool
3372
796b6530
KW
3373This macro is only used by C<sv_true()> or its macro equivalent, and only if
3374the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3375It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
06c841cf
FC
3376
3377=for apidoc sv_2bool_flags
3378
796b6530
KW
3379This function is only used by C<sv_true()> and friends, and only if
3380the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3381contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
06c841cf 3382
645c22ef
DM
3383
3384=cut
3385*/
3386
463ee0b2 3387bool
9d176cd8 3388Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
463ee0b2 3389{
06c841cf 3390 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3391
9d176cd8 3392 restart:
06c841cf 3393 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3394
a0d0e21e
LW
3395 if (!SvOK(sv))
3396 return 0;
3397 if (SvROK(sv)) {
fabdb6c0 3398 if (SvAMAGIC(sv)) {
31d632c3 3399 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
9d176cd8
DD
3400 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3401 bool svb;
3402 sv = tmpsv;
3403 if(SvGMAGICAL(sv)) {
3404 flags = SV_GMAGIC;
3405 goto restart; /* call sv_2bool */
3406 }
3407 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3408 else if(!SvOK(sv)) {
3409 svb = 0;
3410 }
3411 else if(SvPOK(sv)) {
3412 svb = SvPVXtrue(sv);
3413 }
3414 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3415 svb = (SvIOK(sv) && SvIVX(sv) != 0)
659c4b96 3416 || (SvNOK(sv) && SvNVX(sv) != 0.0);
9d176cd8
DD
3417 }
3418 else {
3419 flags = 0;
3420 goto restart; /* call sv_2bool_nomg */
3421 }
3422 return cBOOL(svb);
3423 }
fabdb6c0 3424 }
e8f01ee5
DM
3425 assert(SvRV(sv));
3426 return TRUE;
a0d0e21e 3427 }
85b7d9b3
FC
3428 if (isREGEXP(sv))
3429 return
3430 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
1a436fbe
DM
3431
3432 if (SvNOK(sv) && !SvPOK(sv))
3433 return SvNVX(sv) != 0.0;
3434
4bac9ae4 3435 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3436}
3437
c461cf8f
JH
3438/*
3439=for apidoc sv_utf8_upgrade
3440
78ea37eb 3441Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3442Forces the SV to string form if it is not already.
2bbc8d55 3443Will C<mg_get> on C<sv> if appropriate.
796b6530 3444Always sets the C<SvUTF8> flag to avoid future validity checks even
2bbc8d55
SP
3445if the whole string is the same in UTF-8 as not.
3446Returns the number of bytes in the converted string
c461cf8f 3447
0efd0472 3448This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3449use the Encode extension for that.
3450
fe749c9a
KW
3451=for apidoc sv_utf8_upgrade_nomg
3452
796b6530 3453Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
fe749c9a 3454
8d6d96c1
HS
3455=for apidoc sv_utf8_upgrade_flags
3456
78ea37eb 3457Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3458Forces the SV to string form if it is not already.
8d6d96c1 3459Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3460if all the bytes are invariant in UTF-8.
3461If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55 3462will C<mg_get> on C<sv> if appropriate, else not.
2a590426 3463
c58971e9 3464The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
2a590426
KW
3465
3466Returns the number of bytes in the converted string.
8d6d96c1 3467
0efd0472 3468This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3469use the Encode extension for that.
3470
2a590426 3471=for apidoc sv_utf8_upgrade_flags_grow
b3ab6785 3472
796b6530
KW
3473Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3474the number of unused bytes the string of C<sv> is guaranteed to have free after
2a590426
KW
3475it upon return. This allows the caller to reserve extra space that it intends
3476to fill, to avoid extra grows.
b3ab6785 3477
2a590426
KW
3478C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3479are implemented in terms of this function.
3480
3481Returns the number of bytes in the converted string (not including the spares).
3482
3483=cut
b3ab6785 3484
6602b933
KW
3485If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3486C<NUL> isn't guaranteed due to having other routines do the work in some input
3487cases, or if the input is already flagged as being in utf8.
b3ab6785 3488
8d6d96c1
HS
3489*/
3490
3491STRLEN
5aaab254 3492Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3493{
b3ab6785 3494 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3495
808c356f
RGS
3496 if (sv == &PL_sv_undef)
3497 return 0;
892f9127 3498 if (!SvPOK_nog(sv)) {
e0e62c2a 3499 STRLEN len = 0;
d52b7888
NC
3500 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3501 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3502 if (SvUTF8(sv)) {
3503 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3504 return len;
b3ab6785 3505 }
d52b7888 3506 } else {
33fb6f35 3507 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3508 }
e0e62c2a 3509 }
4411f3b6 3510
fde84d2e
DM
3511 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3512 * compiled and individual nodes will remain non-utf8 even if the
3513 * stringified version of the pattern gets upgraded. Whether the
3514 * PVX of a REGEXP should be grown or we should just croak, I don't
3515 * know - DAPM */
3516 if (SvUTF8(sv) || isREGEXP(sv)) {
b3ab6785 3517 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3518 return SvCUR(sv);
f5cee72b 3519 }
5fec3b1d 3520
765f542d 3521 if (SvIsCOW(sv)) {
c56ed9f6 3522 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3523 }
3524
4e93345f 3525 if (SvCUR(sv) == 0) {
e2e3bb6a
KW
3526 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3527 byte */
4e93345f 3528 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3529 /* This function could be much more efficient if we
2bbc8d55 3530 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3531 * chars in the PV. Given that there isn't such a flag
c58971e9 3532 * make the loop as fast as possible. */
b3ab6785 3533 U8 * s = (U8 *) SvPVX_const(sv);
b3ab6785 3534 U8 *t = s;
c4e7c712 3535
c58971e9 3536 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
b3ab6785 3537
c58971e9
KW
3538 /* utf8 conversion not needed because all are invariants. Mark
3539 * as UTF-8 even if no variant - saves scanning loop */
3540 SvUTF8_on(sv);
3541 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3542 return SvCUR(sv);
dc772057 3543 }
b3ab6785 3544
c58971e9
KW
3545 /* Here, there is at least one variant (t points to the first one), so
3546 * the string should be converted to utf8. Everything from 's' to
3547 * 't - 1' will occupy only 1 byte each on output.
dc772057 3548 *
c58971e9
KW
3549 * Note that the incoming SV may not have a trailing '\0', as certain
3550 * code in pp_formline can send us partially built SVs.
b3ab6785
KW
3551 *
3552 * There are two main ways to convert. One is to create a new string
3553 * and go through the input starting from the beginning, appending each
c58971e9
KW
3554 * converted value onto the new string as we go along. Going this
3555 * route, it's probably best to initially allocate enough space in the
3556 * string rather than possibly running out of space and having to
3557 * reallocate and then copy what we've done so far. Since everything
3558 * from 's' to 't - 1' is invariant, the destination can be initialized
3559 * with these using a fast memory copy. To be sure to allocate enough
3560 * space, one could use the worst case scenario, where every remaining
3561 * byte expands to two under UTF-8, or one could parse it and count
3562 * exactly how many do expand.
b3ab6785 3563 *
c58971e9
KW
3564 * The other way is to unconditionally parse the remainder of the
3565 * string to figure out exactly how big the expanded string will be,
3566 * growing if needed. Then start at the end of the string and place
3567 * the character there at the end of the unfilled space in the expanded
3568 * one, working backwards until reaching 't'.
b3ab6785 3569 *
c58971e9
KW
3570 * The problem with assuming the worst case scenario is that for very
3571 * long strings, we could allocate much more memory than actually
3572 * needed, which can create performance problems. If we have to parse
3573 * anyway, the second method is the winner as it may avoid an extra
3574 * copy.