This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
reduce cost of SvVALID()
[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.
d2a0f284
JC
134An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135sv, av, hv...) contains type and reference count information, and for
136many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137contains fields specific to each type. Some types store all they need
138in the head, so don't have a body.
139
486ec47a 140In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
141and bodies are allocated out of arenas, which by default are
142approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
143Sv-bodies are allocated by their sv-type, guaranteeing size
144consistency needed to allocate safely from arrays.
145
d2a0f284
JC
146For SV-heads, the first slot in each arena is reserved, and holds a
147link to the next arena, some flags, and a note of the number of slots.
148Snaked through each arena chain is a linked list of free items; when
149this becomes empty, an extra arena is allocated and divided up into N
150items which are threaded into the free list.
151
152SV-bodies are similar, but they use arena-sets by default, which
153separate the link and info from the arena itself, and reclaim the 1st
154slot in the arena. SV-bodies are further described later.
645c22ef
DM
155
156The following global variables are associated with arenas:
157
7fefc6c1
KW
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
645c22ef 160
7fefc6c1
KW
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
93e68bfb 164
d2a0f284
JC
165A few special SV heads are not allocated from an arena, but are
166instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
167The size of arenas can be changed from the default by setting
168PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
169
170The SV arena serves the secondary purpose of allowing still-live SVs
171to be located and destroyed during final cleanup.
172
173At the lowest level, the macros new_SV() and del_SV() grab and free
174an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175to return the SV to the free list with error checking.) new_SV() calls
176more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177SVs in the free list have their SvTYPE field set to all ones.
178
ff276b08 179At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 180perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 181start of the interpreter.
645c22ef 182
645c22ef
DM
183The function visit() scans the SV arenas list, and calls a specified
184function for each SV it finds which is still live - ie which has an SvTYPE
185other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186following functions (specified as [function that calls visit()] / [function
187called by visit() for each SV]):
188
189 sv_report_used() / do_report_used()
f2524eef 190 dump all remaining SVs (debugging aid)
645c22ef 191
e4487e9b 192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 193 do_clean_named_io_objs(),do_curse()
645c22ef 194 Attempt to free all objects pointed to by RVs,
e76981f9
FC
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
645c22ef
DM
199 perl_destruct(), prior to calling sv_clean_all()
200 below.
201
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
211
93e68bfb 212=head2 Arena allocator API Summary
645c22ef
DM
213
214Private API to rest of sv.c
215
216 new_SV(), del_SV(),
217
df0f0429 218 new_XPVNV(), del_XPVGV(),
645c22ef
DM
219 etc
220
221Public API:
222
8cf8f3d1 223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 224
645c22ef
DM
225=cut
226
3e8320cc 227 * ========================================================================= */
645c22ef 228
4561caa4
CS
229/*
230 * "A time to plant, and a time to uproot what was planted..."
231 */
232
d7a2c63c
MHM
233#ifdef PERL_MEM_LOG
234# define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236# define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
238#else
239# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
241#endif
242
fd0854ff 243#ifdef DEBUG_LEAKING_SCALARS
484e6108
FC
244# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
246 } STMT_END
d7a2c63c
MHM
247# define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
250#else
251# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 252# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
253#endif
254
48614a46
NC
255#ifdef PERL_POISON
256# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 257# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
258/* Whilst I'd love to do this, it seems that things like to check on
259 unreferenced scalars
ce5dbf61 260# define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 261*/
ce5dbf61 262# define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
7e337ee0 263 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
264#else
265# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 266# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
ce5dbf61 267# define POISON_SV_HEAD(sv)
48614a46
NC
268#endif
269
990198f0
DM
270/* Mark an SV head as unused, and add to free list.
271 *
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
276
053fc874
GS
277#define plant_SV(p) \
278 STMT_START { \
990198f0 279 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
fd0854ff 282 FREE_SV_DEBUG_FILE(p); \
ce5dbf61 283 POISON_SV_HEAD(p); \
053fc874 284 SvFLAGS(p) = SVTYPEMASK; \
990198f0 285 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 286 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
287 PL_sv_root = (p); \
288 } \
053fc874
GS
289 --PL_sv_count; \
290 } STMT_END
a0d0e21e 291
053fc874
GS
292#define uproot_SV(p) \
293 STMT_START { \
294 (p) = PL_sv_root; \
daba3364 295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
296 ++PL_sv_count; \
297 } STMT_END
298
645c22ef 299
cac9b346
NC
300/* make some more SVs by adding another arena */
301
cac9b346
NC
302STATIC SV*
303S_more_sv(pTHX)
304{
305 SV* sv;
9a87bd09
NC
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
309 uproot_SV(sv);
310 return sv;
311}
312
645c22ef
DM
313/* new_SV(): return a new, empty SV head */
314
eba0f806
DM
315#ifdef DEBUG_LEAKING_SCALARS
316/* provide a real function for a debugger to play with */
317STATIC SV*
d7a2c63c 318S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
319{
320 SV* sv;
321
eba0f806
DM
322 if (PL_sv_root)
323 uproot_SV(sv);
324 else
cac9b346 325 sv = S_more_sv(aTHX);
eba0f806
DM
326 SvANY(sv) = 0;
327 SvREFCNT(sv) = 1;
328 SvFLAGS(sv) = 0;
fd0854ff 329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
331 ? PL_parser->copline
332 : PL_curcop
f24aceb1
DM
333 ? CopLINE(PL_curcop)
334 : 0
e385c3bf 335 );
fd0854ff 336 sv->sv_debug_inpad = 0;
cd676548 337 sv->sv_debug_parent = NULL;
484e6108 338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
339
340 sv->sv_debug_serial = PL_sv_serial++;
341
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
345
eba0f806
DM
346 return sv;
347}
d7a2c63c 348# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
349
350#else
351# define new_SV(p) \
053fc874 352 STMT_START { \
053fc874
GS
353 if (PL_sv_root) \
354 uproot_SV(p); \
355 else \
cac9b346 356 (p) = S_more_sv(aTHX); \
053fc874
GS
357 SvANY(p) = 0; \
358 SvREFCNT(p) = 1; \
359 SvFLAGS(p) = 0; \
d7a2c63c 360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 361 } STMT_END
eba0f806 362#endif
463ee0b2 363
645c22ef
DM
364
365/* del_SV(): return an empty SV head to the free list */
366
a0d0e21e 367#ifdef DEBUGGING
4561caa4 368
053fc874
GS
369#define del_SV(p) \
370 STMT_START { \
aea4f609 371 if (DEBUG_D_TEST) \
053fc874
GS
372 del_sv(p); \
373 else \
374 plant_SV(p); \
053fc874 375 } STMT_END
a0d0e21e 376
76e3520e 377STATIC void
cea2e8a9 378S_del_sv(pTHX_ SV *p)
463ee0b2 379{
7918f24d
NC
380 PERL_ARGS_ASSERT_DEL_SV;
381
aea4f609 382 if (DEBUG_D_TEST) {
4633a7c4 383 SV* sva;
a3b680e6 384 bool ok = 0;
daba3364 385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 388 if (p >= sv && p < svend) {
a0d0e21e 389 ok = 1;
c0ff570e
NC
390 break;
391 }
a0d0e21e
LW
392 }
393 if (!ok) {
9b387841
NC
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%"UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
397 return;
398 }
399 }
4561caa4 400 plant_SV(p);
463ee0b2 401}
a0d0e21e 402
4561caa4
CS
403#else /* ! DEBUGGING */
404
405#define del_SV(p) plant_SV(p)
406
407#endif /* DEBUGGING */
463ee0b2 408
645c22ef
DM
409
410/*
ccfc67b7
JH
411=head1 SV Manipulation Functions
412
645c22ef
DM
413=for apidoc sv_add_arena
414
415Given a chunk of memory, link it to the head of the list of arenas,
416and split it into a list of free SVs.
417
418=cut
419*/
420
d2bd4e7f
NC
421static void
422S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 423{
daba3364 424 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
425 SV* sv;
426 SV* svend;
4633a7c4 427
7918f24d
NC
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
429
4633a7c4 430 /* The first SV in an arena isn't an SV. */
3280af22 431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
434
3280af22
NIS
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
4633a7c4
LW
437
438 svend = &sva[SvREFCNT(sva) - 1];
439 sv = sva + 1;
463ee0b2 440 while (sv < svend) {
3eef1deb 441 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 442#ifdef DEBUGGING
978b032e 443 SvREFCNT(sv) = 0;
03e36789 444#endif
4b69cbe3 445 /* Must always set typemask because it's always checked in on cleanup
03e36789 446 when the arenas are walked looking for objects. */
8990e307 447 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
448 sv++;
449 }
3eef1deb 450 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
451#ifdef DEBUGGING
452 SvREFCNT(sv) = 0;
453#endif
4633a7c4
LW
454 SvFLAGS(sv) = SVTYPEMASK;
455}
456
055972dc
DM
457/* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
645c22ef 459
5226ed68 460STATIC I32
de37a194 461S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 462{
4633a7c4 463 SV* sva;
5226ed68 464 I32 visited = 0;
8990e307 465
7918f24d
NC
466 PERL_ARGS_ASSERT_VISIT;
467
daba3364 468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
469 const SV * const svend = &sva[SvREFCNT(sva)];
470 SV* sv;
4561caa4 471 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
473 && (sv->sv_flags & mask) == flags
474 && SvREFCNT(sv))
475 {
942481a7 476 (*f)(aTHX_ sv);
5226ed68
JH
477 ++visited;
478 }
8990e307
LW
479 }
480 }
5226ed68 481 return visited;
8990e307
LW
482}
483
758a08c3
JH
484#ifdef DEBUGGING
485
645c22ef
DM
486/* called by sv_report_used() for each live SV */
487
488static void
5fa45a31 489do_report_used(pTHX_ SV *const sv)
645c22ef 490{
e4787c0c 491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
492 PerlIO_printf(Perl_debug_log, "****\n");
493 sv_dump(sv);
494 }
495}
758a08c3 496#endif
645c22ef
DM
497
498/*
499=for apidoc sv_report_used
500
fde67290 501Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
502
503=cut
504*/
505
8990e307 506void
864dbfa3 507Perl_sv_report_used(pTHX)
4561caa4 508{
ff270d3a 509#ifdef DEBUGGING
055972dc 510 visit(do_report_used, 0, 0);
96a5add6
AL
511#else
512 PERL_UNUSED_CONTEXT;
ff270d3a 513#endif
4561caa4
CS
514}
515
645c22ef
DM
516/* called by sv_clean_objs() for each live SV */
517
518static void
de37a194 519do_clean_objs(pTHX_ SV *const ref)
645c22ef 520{
ea724faa
NC
521 assert (SvROK(ref));
522 {
823a54a3
AL
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
528 SvWEAKREF_off(ref);
529 SvRV_set(ref, NULL);
530 } else {
531 SvROK_off(ref);
532 SvRV_set(ref, NULL);
fc2b2dca 533 SvREFCNT_dec_NN(target);
823a54a3 534 }
645c22ef
DM
535 }
536 }
645c22ef
DM
537}
538
645c22ef 539
e4487e9b
DM
540/* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
542
645c22ef 543static void
f30de749 544do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 545{
57ef47cc 546 SV *obj;
ea724faa 547 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 548 assert(isGV_with_GP(sv));
57ef47cc
DM
549 if (!GvGP(sv))
550 return;
551
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
554 SvREFCNT_inc(sv);
555
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
559 GvSV(sv) = NULL;
fc2b2dca 560 SvREFCNT_dec_NN(obj);
57ef47cc
DM
561 }
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
565 GvAV(sv) = NULL;
fc2b2dca 566 SvREFCNT_dec_NN(obj);
57ef47cc
DM
567 }
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
571 GvHV(sv) = NULL;
fc2b2dca 572 SvREFCNT_dec_NN(obj);
57ef47cc
DM
573 }
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 577 GvCV_set(sv, NULL);
fc2b2dca 578 SvREFCNT_dec_NN(obj);
57ef47cc 579 }
fc2b2dca 580 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
581}
582
68b590d9 583/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
584 * called by sv_clean_objs() for each live GV */
585
586static void
587do_clean_named_io_objs(pTHX_ SV *const sv)
588{
e4487e9b
DM
589 SV *obj;
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
68b590d9 592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
593 return;
594
595 SvREFCNT_inc(sv);
57ef47cc
DM
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
599 GvIOp(sv) = NULL;
fc2b2dca 600 SvREFCNT_dec_NN(obj);
645c22ef 601 }
fc2b2dca 602 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 603}
645c22ef 604
4155e4fe
FC
605/* Void wrapper to pass to visit() */
606static void
607do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
610 return;
611 (void)curse(sv, 0);
612}
613
645c22ef
DM
614/*
615=for apidoc sv_clean_objs
616
fde67290 617Attempt to destroy all objects not yet freed.
645c22ef
DM
618
619=cut
620*/
621
4561caa4 622void
864dbfa3 623Perl_sv_clean_objs(pTHX)
4561caa4 624{
68b590d9 625 GV *olddef, *olderr;
3280af22 626 PL_in_clean_objs = TRUE;
055972dc 627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
d011219a 631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
3280af22 645 PL_in_clean_objs = FALSE;
4561caa4
CS
646}
647
645c22ef
DM
648/* called by sv_clean_all() for each live SV */
649
650static void
de37a194 651do_clean_all(pTHX_ SV *const sv)
645c22ef 652{
daba3364 653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 654 /* don't clean pid table and strtab */
d17ea597 655 return;
cddfcddc 656 }
645c22ef
DM
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 659 SvREFCNT_dec_NN(sv);
645c22ef
DM
660}
661
662/*
663=for apidoc sv_clean_all
664
665Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 666cleanup. This function may have to be called multiple times to free
ff276b08 667SVs which are in complex self-referential hierarchies.
645c22ef
DM
668
669=cut
670*/
671
5226ed68 672I32
864dbfa3 673Perl_sv_clean_all(pTHX)
8990e307 674{
5226ed68 675 I32 cleaned;
3280af22 676 PL_in_clean_all = TRUE;
055972dc 677 cleaned = visit(do_clean_all, 0,0);
5226ed68 678 return cleaned;
8990e307 679}
463ee0b2 680
5e258f8c
JC
681/*
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
39244528 687 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
688
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 692 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
5e258f8c 695*/
5e258f8c 696struct arena_desc {
398c677b
NC
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
e5973ed5 699 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
700};
701
e6148039
NC
702struct arena_set;
703
704/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
706 therefore likely to be 1 aligned memory page. */
707
708#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
710
711struct arena_set {
712 struct arena_set* next;
0a848332
NC
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
715 struct arena_desc set[ARENAS_PER_SET];
716};
717
645c22ef
DM
718/*
719=for apidoc sv_free_arenas
720
fde67290 721Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
722heads and bodies within the arenas must already have been freed.
723
724=cut
7fefc6c1 725
645c22ef 726*/
4633a7c4 727void
864dbfa3 728Perl_sv_free_arenas(pTHX)
4633a7c4
LW
729{
730 SV* sva;
731 SV* svanext;
0a848332 732 unsigned int i;
4633a7c4
LW
733
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
736
3280af22 737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 738 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 739 while (svanext && SvFAKE(svanext))
daba3364 740 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
741
742 if (!SvFAKE(sva))
1df70142 743 Safefree(sva);
4633a7c4 744 }
93e68bfb 745
5e258f8c 746 {
0a848332
NC
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
748
749 while (aroot) {
750 struct arena_set *current = aroot;
751 i = aroot->curr;
752 while (i--) {
5e258f8c
JC
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
755 }
0a848332
NC
756 aroot = aroot->next;
757 Safefree(current);
5e258f8c
JC
758 }
759 }
dc8220bf 760 PL_body_arenas = 0;
fdda85ca 761
0a848332
NC
762 i = PERL_ARENA_ROOTS_SIZE;
763 while (i--)
93e68bfb 764 PL_body_roots[i] = 0;
93e68bfb 765
3280af22
NIS
766 PL_sv_arenaroot = 0;
767 PL_sv_root = 0;
4633a7c4
LW
768}
769
bd81e77b
NC
770/*
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
29489e7c 773
bd81e77b
NC
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
777 4. Hash-Entry arenas
29489e7c 778
bd81e77b
NC
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
29489e7c 786
bd81e77b
NC
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
29489e7c 793
d2a0f284
JC
794
795=head1 SV-Body Allocation
796
f554dfc5
MH
797=cut
798
d2a0f284
JC
799Allocation of SV-bodies is similar to SV-heads, differing as follows;
800the allocation mechanism is used for many body types, so is somewhat
801more complicated, it uses arena-sets, and has no need for still-live
802SV detection.
803
804At the outermost level, (new|del)_X*V macros return bodies of the
805appropriate type. These macros call either (new|del)_body_type or
806(new|del)_body_allocated macro pairs, depending on specifics of the
807type. Most body types use the former pair, the latter pair is used to
808allocate body types with "ghost fields".
809
810"ghost fields" are fields that are unused in certain types, and
69ba284b 811consequently don't need to actually exist. They are declared because
d2a0f284
JC
812they're part of a "base type", which allows use of functions as
813methods. The simplest examples are AVs and HVs, 2 aggregate types
814which don't use the fields which support SCALAR semantics.
815
69ba284b 816For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
817chunks, we thus avoid wasted memory for those unaccessed members.
818When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 819size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
820structure. (But things will all go boom if you write to the part that
821is "not there", because you'll be overwriting the last members of the
822preceding structure in memory.)
823
69ba284b 824We calculate the correction using the STRUCT_OFFSET macro on the first
a05ea1cf 825member present. If the allocated structure is smaller (no initial NV
69ba284b
NC
826actually allocated) then the net effect is to subtract the size of the NV
827from the pointer, to return a new pointer as if an initial NV were actually
a05ea1cf 828allocated. (We were using structures named *_allocated for this, but
69ba284b
NC
829this turned out to be a subtle bug, because a structure without an NV
830could have a lower alignment constraint, but the compiler is allowed to
831optimised accesses based on the alignment constraint of the actual pointer
832to the full structure, for example, using a single 64 bit load instruction
833because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284 834
a05ea1cf 835This is the same trick as was used for NV and IV bodies. Ironically it
d2a0f284 836doesn't need to be used for NV bodies any more, because NV is now at
5b306eef
DD
837the start of the structure. IV bodies, and also in some builds NV bodies,
838don't need it either, because they are no longer allocated.
d2a0f284
JC
839
840In turn, the new_body_* allocators call S_new_body(), which invokes
841new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 842linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
843necessary to refresh an empty list. Then the lock is released, and
844the body is returned.
845
99816f8d 846Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
847bodies, which it strings into a linked list. It looks up arena-size
848and body-size from the body_details table described below, thus
849supporting the multiple body-types.
850
851If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852the (new|del)_X*V macros are mapped directly to malloc/free.
853
d2a0f284
JC
854For each sv-type, struct body_details bodies_by_type[] carries
855parameters which control these aspects of SV handling:
856
857Arena_size determines whether arenas are used for this body type, and if
858so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859zero, forcing individual mallocs and frees.
860
861Body_size determines how big a body is, and therefore how many fit into
862each arena. Offset carries the body-pointer adjustment needed for
69ba284b 863"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
864
865But its main purpose is to parameterize info needed in
866Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 867vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
868are used for this, except for arena_size.
869
870For the sv-types that have no bodies, arenas are not used, so those
871PL_body_roots[sv_type] are unused, and can be overloaded. In
872something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 873PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 874bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 875available in hv.c.
d2a0f284 876
29489e7c
DM
877*/
878
bd81e77b 879struct body_details {
0fb58b32 880 U8 body_size; /* Size to allocate */
10666ae3 881 U8 copy; /* Size of structure to copy (may be shorter) */
5b306eef
DD
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
bd81e77b 888};
29489e7c 889
bd81e77b
NC
890#define HADNV FALSE
891#define NONV TRUE
29489e7c 892
d2a0f284 893
bd81e77b
NC
894#ifdef PURIFY
895/* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897#define HASARENA FALSE
898#else
899#define HASARENA TRUE
900#endif
901#define NOARENA FALSE
29489e7c 902
d2a0f284
JC
903/* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
909 declarations.
910 */
95db5f15
MB
911#define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913#define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917#define FIT_ARENA(count,body_size) \
cd1dc8e2 918 (U32)(count \
95db5f15 919 ? FIT_ARENAn (count, body_size) \
cd1dc8e2 920 : FIT_ARENA0 (body_size))
d2a0f284 921
bd81e77b
NC
922/* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
29489e7c 925
bd81e77b
NC
926#define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
daba3364 928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 929
bd81e77b 930static const struct body_details bodies_by_type[] = {
829cd18a
NC
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 933
db93c0c4
NC
934 /* IVs are in the head, so the allocation size is 0. */
935 { 0,
d2a0f284 936 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 938 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
939 },
940
5b306eef
DD
941#if NVSIZE <= IVSIZE
942 { 0, sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
945#else
6e128786
NC
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
5b306eef 949#endif
d2a0f284 950
bc337e5c 951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 954 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 956
d361b004
KW
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 962
bc337e5c 963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 968
bc337e5c 969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 974
6e128786 975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 977
601dfd0a
NC
978 { sizeof(regexp),
979 sizeof(regexp),
980 0,
ecff11eb 981 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 982 FIT_ARENA(0, sizeof(regexp))
5c35adbb 983 },
4df7f6af 984
10666ae3 985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
987
10666ae3 988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
990
601dfd0a 991 { sizeof(XPVAV),
4f7003f5 992 copy_length(XPVAV, xav_alloc),
601dfd0a 993 0,
69ba284b 994 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 995 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 996
601dfd0a 997 { sizeof(XPVHV),
359164a0 998 copy_length(XPVHV, xhv_max),
601dfd0a 999 0,
69ba284b 1000 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 1001 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 1002
601dfd0a
NC
1003 { sizeof(XPVCV),
1004 sizeof(XPVCV),
1005 0,
69ba284b 1006 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 1007 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 1008
601dfd0a
NC
1009 { sizeof(XPVFM),
1010 sizeof(XPVFM),
1011 0,
69ba284b 1012 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 1013 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 1014
601dfd0a
NC
1015 { sizeof(XPVIO),
1016 sizeof(XPVIO),
1017 0,
b6f60916 1018 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 1019 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 1020};
29489e7c 1021
bd81e77b 1022#define new_body_allocated(sv_type) \
d2a0f284 1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1024 - bodies_by_type[sv_type].offset)
29489e7c 1025
26359cfa
NC
1026/* return a thing to the free list */
1027
1028#define del_body(thing, root) \
1029 STMT_START { \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1033 } STMT_END
29489e7c 1034
bd81e77b 1035#ifdef PURIFY
5b306eef
DD
1036#if !(NVSIZE <= IVSIZE)
1037# define new_XNV() safemalloc(sizeof(XPVNV))
1038#endif
beeec492
NC
1039#define new_XPVNV() safemalloc(sizeof(XPVNV))
1040#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 1041
beeec492 1042#define del_XPVGV(p) safefree(p)
29489e7c 1043
bd81e77b 1044#else /* !PURIFY */
29489e7c 1045
5b306eef
DD
1046#if !(NVSIZE <= IVSIZE)
1047# define new_XNV() new_body_allocated(SVt_NV)
1048#endif
65ac1738 1049#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1050#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1051
26359cfa
NC
1052#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1d7c1841 1054
bd81e77b 1055#endif /* PURIFY */
93e68bfb 1056
bd81e77b 1057/* no arena for you! */
93e68bfb 1058
bd81e77b 1059#define new_NOARENA(details) \
beeec492 1060 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1061#define new_NOARENAZ(details) \
beeec492 1062 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1063
1e30fcd5
NC
1064void *
1065Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
d2a0f284 1067{
d2a0f284 1068 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1071 unsigned int curr;
d2a0f284
JC
1072 char *start;
1073 const char *end;
02982131 1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
20b7effb
JH
1075#if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1076 dVAR;
1077#endif
0b2d3faa 1078#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1079 static bool done_sanity_check;
1080
0b2d3faa
JH
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
10666ae3 1083 if (!done_sanity_check) {
ea471437 1084 unsigned int i = SVt_LAST;
10666ae3
NC
1085
1086 done_sanity_check = TRUE;
1087
1088 while (i--)
1089 assert (bodies_by_type[i].type == i);
1090 }
1091#endif
1092
02982131 1093 assert(arena_size);
23e9d66c 1094
99816f8d
NC
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1101 aroot = newroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1104 }
1105
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1110
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1116
1117 start = (char *) adesc->arena;
d2a0f284 1118
29657bb6
NC
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
02982131 1121 end = start + good_arena_size / body_size * body_size;
d2a0f284 1122
486ec47a 1123 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1124#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1127 "size %d ct %d\n",
02982131
NC
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
d8fca402 1131#else
d2a0f284
JC
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1134 (void*)start, (void*)end,
02982131
NC
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
d8fca402 1137#endif
d2a0f284
JC
1138 *root = (void *)start;
1139
29657bb6
NC
1140 while (1) {
1141 /* Where the next body would start: */
d2a0f284 1142 char * const next = start + body_size;
29657bb6
NC
1143
1144 if (next >= end) {
1145 /* This is the last body: */
1146 assert(next == end);
1147
1148 *(void **)start = 0;
1149 return *root;
1150 }
1151
d2a0f284
JC
1152 *(void**) start = (void *)next;
1153 start = next;
1154 }
d2a0f284
JC
1155}
1156
1157/* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1160*/
1161#define new_body_inline(xpv, sv_type) \
1162 STMT_START { \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1168 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1169 } STMT_END
1170
1171#ifndef PURIFY
1172
1173STATIC void *
de37a194 1174S_new_body(pTHX_ const svtype sv_type)
d2a0f284 1175{
d2a0f284
JC
1176 void *xpv;
1177 new_body_inline(xpv, sv_type);
1178 return xpv;
1179}
1180
1181#endif
93e68bfb 1182
238b27b3
NC
1183static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1185
bd81e77b
NC
1186/*
1187=for apidoc sv_upgrade
93e68bfb 1188
bd81e77b
NC
1189Upgrade an SV to a more complex form. Generally adds a new body type to the
1190SV, then copies across as much information as possible from the old body.
9521ca61
FC
1191It croaks if the SV is already in a more complex form than requested. You
1192generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193before calling C<sv_upgrade>, and hence does not croak. See also
fbe13c60 1194C<L</svtype>>.
93e68bfb 1195
bd81e77b 1196=cut
93e68bfb 1197*/
93e68bfb 1198
bd81e77b 1199void
5aaab254 1200Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1201{
bd81e77b
NC
1202 void* old_body;
1203 void* new_body;
42d0e0b7 1204 const svtype old_type = SvTYPE(sv);
d2a0f284 1205 const struct body_details *new_type_details;
238b27b3 1206 const struct body_details *old_type_details
bd81e77b 1207 = bodies_by_type + old_type;
ed7df46e 1208 SV *referent = NULL;
cac9b346 1209
7918f24d
NC
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1211
1776cbe8
NC
1212 if (old_type == new_type)
1213 return;
1214
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1219
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1223
093085a8 1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1225 sv_force_normal_flags(sv, 0);
1226 }
cac9b346 1227
bd81e77b 1228 old_body = SvANY(sv);
de042e1d 1229
bd81e77b
NC
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
cac9b346 1232
bd81e77b
NC
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
645c22ef 1237
bd81e77b
NC
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1240
bd81e77b
NC
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
08742458 1245
bd81e77b 1246 so what happens if you allocate memory for this structure:
30f9da9e 1247
bd81e77b
NC
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
bfc44f79 1252
bd81e77b
NC
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1257 Bugs ensue.
bfc44f79 1258
bd81e77b
NC
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
30f9da9e 1264
bd81e77b
NC
1265 So we are careful and work out the size of used parts of all the
1266 structures. */
bfc44f79 1267
bd81e77b
NC
1268 switch (old_type) {
1269 case SVt_NULL:
1270 break;
1271 case SVt_IV:
4df7f6af 1272 if (SvROK(sv)) {
ed7df46e 1273 referent = SvRV(sv);
238b27b3
NC
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
4df7f6af
NC
1277 } else {
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1281 }
bd81e77b
NC
1282 }
1283 break;
1284 case SVt_NV:
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
bd81e77b
NC
1287 }
1288 break;
bd81e77b
NC
1289 case SVt_PV:
1290 assert(new_type > SVt_PV);
6d59e610
LM
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
bd81e77b
NC
1293 break;
1294 case SVt_PVIV:
1295 break;
1296 case SVt_PVNV:
1297 break;
1298 case SVt_PVMG:
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
bd81e77b
NC
1303 break;
1304 default:
2439e033 1305 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1308 }
3376de98 1309
2439e033 1310 if (UNLIKELY(old_type > new_type))
3376de98
NC
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1313
2fa1109b 1314 new_type_details = bodies_by_type + new_type;
645c22ef 1315
bd81e77b
NC
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
932e9ff9 1318
ab4416c0
NC
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
bd81e77b 1322 switch (new_type) {
bd81e77b
NC
1323 case SVt_IV:
1324 assert(old_type == SVt_NULL);
dc6369ef 1325 SET_SVANY_FOR_BODYLESS_IV(sv);
bd81e77b
NC
1326 SvIV_set(sv, 0);
1327 return;
1328 case SVt_NV:
1329 assert(old_type == SVt_NULL);
5b306eef 1330#if NVSIZE <= IVSIZE
dc6369ef 1331 SET_SVANY_FOR_BODYLESS_NV(sv);
5b306eef 1332#else
bd81e77b 1333 SvANY(sv) = new_XNV();
5b306eef 1334#endif
bd81e77b
NC
1335 SvNV_set(sv, 0);
1336 return;
bd81e77b 1337 case SVt_PVHV:
bd81e77b 1338 case SVt_PVAV:
d2a0f284 1339 assert(new_type_details->body_size);
c1ae03ae
NC
1340
1341#ifndef PURIFY
1342 assert(new_type_details->arena);
d2a0f284 1343 assert(new_type_details->arena_size);
c1ae03ae 1344 /* This points to the start of the allocated area. */
d2a0f284
JC
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1347 new_body = ((char *)new_body) - new_type_details->offset;
1348#else
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1352#endif
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1355 AvMAX(sv) = -1;
1356 AvFILLp(sv) = -1;
1357 AvREAL_only(sv);
64484faa 1358 if (old_type_details->body_size) {
ac572bf4
NC
1359 AvALLOC(sv) = 0;
1360 } else {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1363 cache. */
1364 }
78ac7dd9
NC
1365 } else {
1366 assert(!SvOK(sv));
1367 SvOK_off(sv);
1368#ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1370#endif
586fc6a3
SM
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1373 }
aeb18a1e 1374
bd81e77b
NC
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1378 */
4df7f6af
NC
1379 if (old_type == SVt_IV) {
1380 assert(!SvROK(sv));
1381 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1382 assert(SvPVX_const(sv) == 0);
1383 }
aeb18a1e 1384
bd81e77b 1385 if (old_type >= SVt_PVMG) {
e736a858 1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1388 } else {
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1390 }
1391 break;
93e68bfb 1392
bd81e77b
NC
1393 case SVt_PVIV:
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1397 assert(!SvNOK(sv));
2b5060ae 1398 /* FALLTHROUGH */
bd81e77b
NC
1399 case SVt_PVIO:
1400 case SVt_PVFM:
bd81e77b
NC
1401 case SVt_PVGV:
1402 case SVt_PVCV:
1403 case SVt_PVLV:
d361b004 1404 case SVt_INVLIST:
12c45b25 1405 case SVt_REGEXP:
bd81e77b
NC
1406 case SVt_PVMG:
1407 case SVt_PVNV:
1408 case SVt_PV:
93e68bfb 1409
d2a0f284 1410 assert(new_type_details->body_size);
bd81e77b
NC
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
d2a0f284
JC
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1417 new_body = ((char *)new_body) - new_type_details->offset;
1418 } else {
1419 new_body = new_NOARENAZ(new_type_details);
1420 }
1421 SvANY(sv) = new_body;
5e2fc214 1422
bd81e77b 1423 if (old_type_details->copy) {
f9ba3d20
NC
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1428
1429 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1430 const int difference
f9ba3d20
NC
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1434 }
1435 assert (length >= 0);
1436
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1438 char);
bd81e77b
NC
1439 }
1440
1441#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1446 * for 0.0 */
e22a937e
NC
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
bd81e77b 1449 SvNV_set(sv, 0);
82048762 1450#endif
5e2fc214 1451
2439e033 1452 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1453 IO * const io = MUTABLE_IO(sv);
d963bf01 1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1455
1456 SvOBJECT_on(io);
1457 /* Clear the stashcache because a new IO could overrule a package
1458 name */
103f5a36 1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1460 hv_clear(PL_stashcache);
1461
85dca89a 1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1463 IoPAGE_LEN(sv) = 60;
85dca89a 1464 }
2439e033 1465 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
ed7df46e 1468 /* referent will be NULL unless the old type was SVt_IV emulating
4df7f6af 1469 SVt_RV */
ed7df46e 1470 sv->sv_u.svu_rv = referent;
4df7f6af 1471 }
bd81e77b
NC
1472 break;
1473 default:
afd78fd5
JH
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
bd81e77b 1476 }
73171d91 1477
5b306eef
DD
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
bd81e77b 1481#ifdef PURIFY
beeec492 1482 safefree(old_body);
bd81e77b 1483#else
bc786448
GG
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
bd81e77b
NC
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1490#endif
1491 }
1492}
73171d91 1493
bd81e77b
NC
1494/*
1495=for apidoc sv_backoff
73171d91 1496
fde67290 1497Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1498wrapper instead.
73171d91 1499
bd81e77b 1500=cut
73171d91
NC
1501*/
1502
fa7a1e49
DD
1503/* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1505*/
1506
1507void
ddeaf645 1508Perl_sv_backoff(SV *const sv)
bd81e77b 1509{
69240efd 1510 STRLEN delta;
7a4bba22 1511 const char * const s = SvPVX_const(sv);
7918f24d
NC
1512
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
7918f24d 1514
bd81e77b
NC
1515 assert(SvOOK(sv));
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1518
69240efd
NC
1519 SvOOK_offset(sv, delta);
1520
7a4bba22
NC
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
bd81e77b 1523 SvFLAGS(sv) &= ~SVf_OOK;
fa7a1e49
DD
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1525 return;
bd81e77b 1526}
73171d91 1527
bd81e77b
NC
1528/*
1529=for apidoc sv_grow
73171d91 1530
bd81e77b
NC
1531Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533Use the C<SvGROW> wrapper instead.
93e68bfb 1534
bd81e77b
NC
1535=cut
1536*/
93e68bfb 1537
e0060e30
FC
1538static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1539
bd81e77b 1540char *
5aaab254 1541Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1542{
eb578fdb 1543 char *s;
93e68bfb 1544
7918f24d
NC
1545 PERL_ARGS_ASSERT_SV_GROW;
1546
bd81e77b
NC
1547 if (SvROK(sv))
1548 sv_unref(sv);
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1552 }
1553 else if (SvOOK(sv)) { /* pv is offset? */
1554 sv_backoff(sv);
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
bd81e77b
NC
1558 }
1559 else
db2c6cb3 1560 {
e0060e30 1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1562 s = SvPVX_mutable(sv);
db2c6cb3 1563 }
aeb18a1e 1564
93c10d60 1565#ifdef PERL_COPY_ON_WRITE
cbcb2a16 1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
3c239bea 1567 * to store the COW count. So in general, allocate one more byte than
cbcb2a16
DM
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
fe546b38 1570 *
fa8f4f85
TC
1571 * Only increment if the allocation isn't MEM_SIZE_MAX,
1572 * otherwise it will wrap to 0.
1573 */
fe546b38 1574 if ( newlen != MEM_SIZE_MAX )
cbcb2a16
DM
1575 newlen++;
1576#endif
1577
ce861ea7
YO
1578#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1579#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1580#endif
1581
bd81e77b 1582 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559 1583 STRLEN minlen = SvCUR(sv);
3c239bea 1584 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
f1200559
WH
1585 if (newlen < minlen)
1586 newlen = minlen;
ce861ea7 1587#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
7c641603
KW
1588
1589 /* Don't round up on the first allocation, as odds are pretty good that
1590 * the initial request is accurate as to what is really needed */
ce861ea7 1591 if (SvLEN(sv)) {
9efda33a
TC
1592 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1593 if (rounded > newlen)
1594 newlen = rounded;
ce861ea7 1595 }
bd81e77b 1596#endif
98653f18 1597 if (SvLEN(sv) && s) {
10edeb5d 1598 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1599 }
1600 else {
10edeb5d 1601 s = (char*)safemalloc(newlen);
bd81e77b 1602 if (SvPVX_const(sv) && SvCUR(sv)) {
0a5fcc38 1603 Move(SvPVX_const(sv), s, SvCUR(sv), char);
bd81e77b
NC
1604 }
1605 }
1606 SvPV_set(sv, s);
ce861ea7 1607#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
98653f18
NC
1608 /* Do this here, do it once, do it right, and then we will never get
1609 called back into sv_grow() unless there really is some growing
1610 needed. */
ca7c1a29 1611 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1612#else
bd81e77b 1613 SvLEN_set(sv, newlen);
98653f18 1614#endif
bd81e77b
NC
1615 }
1616 return s;
1617}
aeb18a1e 1618
bd81e77b
NC
1619/*
1620=for apidoc sv_setiv
932e9ff9 1621
bd81e77b 1622Copies an integer into the given SV, upgrading first if necessary.
fbe13c60 1623Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
463ee0b2 1624
bd81e77b
NC
1625=cut
1626*/
463ee0b2 1627
bd81e77b 1628void
5aaab254 1629Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1630{
7918f24d
NC
1631 PERL_ARGS_ASSERT_SV_SETIV;
1632
bd81e77b
NC
1633 SV_CHECK_THINKFIRST_COW_DROP(sv);
1634 switch (SvTYPE(sv)) {
1635 case SVt_NULL:
bd81e77b 1636 case SVt_NV:
3376de98 1637 sv_upgrade(sv, SVt_IV);
bd81e77b 1638 break;
bd81e77b
NC
1639 case SVt_PV:
1640 sv_upgrade(sv, SVt_PVIV);
1641 break;
463ee0b2 1642
bd81e77b 1643 case SVt_PVGV:
6e592b3a
BM
1644 if (!isGV_with_GP(sv))
1645 break;
bd81e77b
NC
1646 case SVt_PVAV:
1647 case SVt_PVHV:
1648 case SVt_PVCV:
1649 case SVt_PVFM:
1650 case SVt_PVIO:
22e74366 1651 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1652 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1653 OP_DESC(PL_op));
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;
bd81e77b
NC
1758 case SVt_PVAV:
1759 case SVt_PVHV:
79072805 1760 case SVt_PVCV:
bd81e77b
NC
1761 case SVt_PVFM:
1762 case SVt_PVIO:
22e74366 1763 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1764 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1765 OP_DESC(PL_op));
0103ca14 1766 break;
42d0e0b7 1767 default: NOOP;
2068cd4d 1768 }
bd81e77b
NC
1769 SvNV_set(sv, num);
1770 (void)SvNOK_only(sv); /* validate number */
1771 SvTAINT(sv);
79072805
LW
1772}
1773
645c22ef 1774/*
bd81e77b 1775=for apidoc sv_setnv_mg
645c22ef 1776
bd81e77b 1777Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1778
1779=cut
1780*/
1781
bd81e77b 1782void
5aaab254 1783Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1784{
7918f24d
NC
1785 PERL_ARGS_ASSERT_SV_SETNV_MG;
1786
bd81e77b
NC
1787 sv_setnv(sv,num);
1788 SvSETMAGIC(sv);
79072805
LW
1789}
1790
3f7602fa
TC
1791/* Return a cleaned-up, printable version of sv, for non-numeric, or
1792 * not incrementable warning display.
1793 * Originally part of S_not_a_number().
1794 * The return value may be != tmpbuf.
bd81e77b 1795 */
954c1994 1796
3f7602fa
TC
1797STATIC const char *
1798S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1799 const char *pv;
94463019 1800
3f7602fa 1801 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1802
94463019 1803 if (DO_UTF8(sv)) {
3f7602fa 1804 SV *dsv = newSVpvs_flags("", SVs_TEMP);
37b8cdd1 1805 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
94463019
JH
1806 } else {
1807 char *d = tmpbuf;
3f7602fa 1808 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1809 /* each *s can expand to 4 chars + "...\0",
1810 i.e. need room for 8 chars */
ecdeb87c 1811
00b6aa41
AL
1812 const char *s = SvPVX_const(sv);
1813 const char * const end = s + SvCUR(sv);
1814 for ( ; s < end && d < limit; s++ ) {
94463019 1815 int ch = *s & 0xFF;
bd27cf70 1816 if (! isASCII(ch) && !isPRINT_LC(ch)) {
94463019
JH
1817 *d++ = 'M';
1818 *d++ = '-';
bd27cf70
KW
1819
1820 /* Map to ASCII "equivalent" of Latin1 */
1821 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
94463019
JH
1822 }
1823 if (ch == '\n') {
1824 *d++ = '\\';
1825 *d++ = 'n';
1826 }
1827 else if (ch == '\r') {
1828 *d++ = '\\';
1829 *d++ = 'r';
1830 }
1831 else if (ch == '\f') {
1832 *d++ = '\\';
1833 *d++ = 'f';
1834 }
1835 else if (ch == '\\') {
1836 *d++ = '\\';
1837 *d++ = '\\';
1838 }
1839 else if (ch == '\0') {
1840 *d++ = '\\';
1841 *d++ = '0';
1842 }
1843 else if (isPRINT_LC(ch))
1844 *d++ = ch;
1845 else {
1846 *d++ = '^';
1847 *d++ = toCTRL(ch);
1848 }
1849 }
1850 if (s < end) {
1851 *d++ = '.';
1852 *d++ = '.';
1853 *d++ = '.';
1854 }
1855 *d = '\0';
1856 pv = tmpbuf;
a0d0e21e 1857 }
a0d0e21e 1858
3f7602fa
TC
1859 return pv;
1860}
1861
1862/* Print an "isn't numeric" warning, using a cleaned-up,
1863 * printable version of the offending string
1864 */
1865
1866STATIC void
1867S_not_a_number(pTHX_ SV *const sv)
1868{
3f7602fa
TC
1869 char tmpbuf[64];
1870 const char *pv;
1871
1872 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1873
1874 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1875
533c011a 1876 if (PL_op)
9014280d 1877 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1878 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1879 "Argument \"%s\" isn't numeric in %s", pv,
1880 OP_DESC(PL_op));
a0d0e21e 1881 else
9014280d 1882 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1883 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1884 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1885}
1886
3f7602fa
TC
1887STATIC void
1888S_not_incrementable(pTHX_ SV *const sv) {
3f7602fa
TC
1889 char tmpbuf[64];
1890 const char *pv;
1891
1892 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1893
1894 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1895
1896 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1897 "Argument \"%s\" treated as 0 in increment (++)", pv);
1898}
1899
c2988b20
NC
1900/*
1901=for apidoc looks_like_number
1902
645c22ef
DM
1903Test if the content of an SV looks like a number (or is a number).
1904C<Inf> and C<Infinity> are treated as numbers (so will not issue a
796b6530 1905non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
f52e41ad 1906ignored.
c2988b20
NC
1907
1908=cut
1909*/
1910
1911I32
aad570aa 1912Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1913{
eb578fdb 1914 const char *sbegin;
c2988b20 1915 STRLEN len;
ea2485eb 1916 int numtype;
c2988b20 1917
7918f24d
NC
1918 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1919
f52e41ad
FC
1920 if (SvPOK(sv) || SvPOKp(sv)) {
1921 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1922 }
c2988b20 1923 else
e0ab1c0e 1924 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
ea2485eb
JH
1925 numtype = grok_number(sbegin, len, NULL);
1926 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
c2988b20 1927}
25da4f38 1928
19f6321d
NC
1929STATIC bool
1930S_glob_2number(pTHX_ GV * const gv)
180488f8 1931{
7918f24d
NC
1932 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1933
675c862f
AL
1934 /* We know that all GVs stringify to something that is not-a-number,
1935 so no need to test that. */
1936 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1937 {
1938 SV *const buffer = sv_newmortal();
1939 gv_efullname3(buffer, gv, "*");
675c862f 1940 not_a_number(buffer);
8e629ff4 1941 }
675c862f
AL
1942 /* We just want something true to return, so that S_sv_2iuv_common
1943 can tail call us and return true. */
19f6321d 1944 return TRUE;
675c862f
AL
1945}
1946
25da4f38
IZ
1947/* Actually, ISO C leaves conversion of UV to IV undefined, but
1948 until proven guilty, assume that things are not that bad... */
1949
645c22ef
DM
1950/*
1951 NV_PRESERVES_UV:
1952
1953 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1954 an IV (an assumption perl has been based on to date) it becomes necessary
1955 to remove the assumption that the NV always carries enough precision to
1956 recreate the IV whenever needed, and that the NV is the canonical form.
1957 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1958 precision as a side effect of conversion (which would lead to insanity
28e5dec8 1959 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
8a4a3196
KW
1960 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1961 where precision was lost, and IV/UV/NV slots that have a valid conversion
1962 which has lost no precision
645c22ef 1963 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1964 would lose precision, the precise conversion (or differently
1965 imprecise conversion) is also performed and cached, to prevent
1966 requests for different numeric formats on the same SV causing
1967 lossy conversion chains. (lossless conversion chains are perfectly
1968 acceptable (still))
1969
1970
1971 flags are used:
1972 SvIOKp is true if the IV slot contains a valid value
1973 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1974 SvNOKp is true if the NV slot contains a valid value
1975 SvNOK is true only if the NV value is accurate
1976
1977 so
645c22ef 1978 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1979 IV(or UV) would lose accuracy over a direct conversion from PV to
1980 IV(or UV). If it would, cache both conversions, return NV, but mark
1981 SV as IOK NOKp (ie not NOK).
1982
645c22ef 1983 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1984 NV would lose accuracy over a direct conversion from PV to NV. If it
1985 would, cache both conversions, flag similarly.
1986
1987 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1988 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1989 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1990 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1991 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1992
645c22ef
DM
1993 The benefit of this is that operations such as pp_add know that if
1994 SvIOK is true for both left and right operands, then integer addition
1995 can be used instead of floating point (for cases where the result won't
1996 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1997 loss of precision compared with integer addition.
1998
1999 * making IV and NV equal status should make maths accurate on 64 bit
2000 platforms
2001 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 2002 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
2003 looking for SvIOK and checking for overflow will not outweigh the
2004 fp to integer speedup)
2005 * will slow down integer operations (callers of SvIV) on "inaccurate"
2006 values, as the change from SvIOK to SvIOKp will cause a call into
2007 sv_2iv each time rather than a macro access direct to the IV slot
2008 * should speed up number->string conversion on integers as IV is
645c22ef 2009 favoured when IV and NV are equally accurate
28e5dec8
JH
2010
2011 ####################################################################
645c22ef
DM
2012 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2013 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2014 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
2015 ####################################################################
2016
645c22ef 2017 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
2018 performance ratio.
2019*/
2020
2021#ifndef NV_PRESERVES_UV
645c22ef
DM
2022# define IS_NUMBER_UNDERFLOW_IV 1
2023# define IS_NUMBER_UNDERFLOW_UV 2
2024# define IS_NUMBER_IV_AND_UV 2
2025# define IS_NUMBER_OVERFLOW_IV 4
2026# define IS_NUMBER_OVERFLOW_UV 5
2027
2028/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
2029
2030/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2031STATIC int
5aaab254 2032S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
2033# ifdef DEBUGGING
2034 , I32 numtype
2035# endif
2036 )
28e5dec8 2037{
7918f24d 2038 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
23491f1d 2039 PERL_UNUSED_CONTEXT;
7918f24d 2040
3f7c398e 2041 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
2042 if (SvNVX(sv) < (NV)IV_MIN) {
2043 (void)SvIOKp_on(sv);
2044 (void)SvNOK_on(sv);
45977657 2045 SvIV_set(sv, IV_MIN);
28e5dec8
JH
2046 return IS_NUMBER_UNDERFLOW_IV;
2047 }
2048 if (SvNVX(sv) > (NV)UV_MAX) {
2049 (void)SvIOKp_on(sv);
2050 (void)SvNOK_on(sv);
2051 SvIsUV_on(sv);
607fa7f2 2052 SvUV_set(sv, UV_MAX);
28e5dec8
JH
2053 return IS_NUMBER_OVERFLOW_UV;
2054 }
c2988b20
NC
2055 (void)SvIOKp_on(sv);
2056 (void)SvNOK_on(sv);
2057 /* Can't use strtol etc to convert this string. (See truth table in
2058 sv_2iv */
2059 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 2060 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2061 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2062 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2063 } else {
2064 /* Integer is imprecise. NOK, IOKp */
2065 }
2066 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2067 }
2068 SvIsUV_on(sv);
607fa7f2 2069 SvUV_set(sv, U_V(SvNVX(sv)));
659c4b96 2070 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2071 if (SvUVX(sv) == UV_MAX) {
2072 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2073 possibly be preserved by NV. Hence, it must be overflow.
2074 NOK, IOKp */
2075 return IS_NUMBER_OVERFLOW_UV;
2076 }
2077 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2078 } else {
2079 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2080 }
c2988b20 2081 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2082}
645c22ef
DM
2083#endif /* !NV_PRESERVES_UV*/
2084
a13f4dff 2085/* If numtype is infnan, set the NV of the sv accordingly.
5564cd7f 2086 * If numtype is anything else, try setting the NV using Atof(PV). */
3c81f0b3
DD
2087#ifdef USING_MSVC6
2088# pragma warning(push)
2089# pragma warning(disable:4756;disable:4056)
2090#endif
a13f4dff 2091static void
3823048b 2092S_sv_setnv(pTHX_ SV* sv, int numtype)
a13f4dff 2093{
07925c5e 2094 bool pok = cBOOL(SvPOK(sv));
5564cd7f 2095 bool nok = FALSE;
a7157111 2096#ifdef NV_INF
a13f4dff
JH
2097 if ((numtype & IS_NUMBER_INFINITY)) {
2098 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
5564cd7f 2099 nok = TRUE;
a7157111
JH
2100 } else
2101#endif
2102#ifdef NV_NAN
2103 if ((numtype & IS_NUMBER_NAN)) {
3823048b 2104 SvNV_set(sv, NV_NAN);
d48bd569 2105 nok = TRUE;
a7157111
JH
2106 } else
2107#endif
2108 if (pok) {
a13f4dff 2109 SvNV_set(sv, Atof(SvPVX_const(sv)));
d48bd569
JH
2110 /* Purposefully no true nok here, since we don't want to blow
2111 * away the possible IOK/UV of an existing sv. */
2112 }
5564cd7f 2113 if (nok) {
d48bd569 2114 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
5564cd7f
JH
2115 if (pok)
2116 SvPOK_on(sv); /* PV is okay, though. */
2117 }
a13f4dff 2118}
3c81f0b3
DD
2119#ifdef USING_MSVC6
2120# pragma warning(pop)
2121#endif
a13f4dff 2122
af359546 2123STATIC bool
7918f24d
NC
2124S_sv_2iuv_common(pTHX_ SV *const sv)
2125{
7918f24d
NC
2126 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2127
af359546 2128 if (SvNOKp(sv)) {
28e5dec8
JH
2129 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2130 * without also getting a cached IV/UV from it at the same time
2131 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2132 * IV or UV at same time to avoid this. */
2133 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2134
2135 if (SvTYPE(sv) == SVt_NV)
2136 sv_upgrade(sv, SVt_PVNV);
2137
28e5dec8
JH
2138 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2139 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2140 certainly cast into the IV range at IV_MAX, whereas the correct
2141 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2142 cases go to UV */
e91de695
JH
2143#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2144 if (Perl_isnan(SvNVX(sv))) {
2145 SvUV_set(sv, 0);
2146 SvIsUV_on(sv);
2147 return FALSE;
2148 }
2149#endif
28e5dec8 2150 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2151 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2152 if (SvNVX(sv) == (NV) SvIVX(sv)
28e5dec8 2153#ifndef NV_PRESERVES_UV
53e2bfb7 2154 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
28e5dec8
JH
2155 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2156 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2157 /* Don't flag it as "accurately an integer" if the number
2158 came from a (by definition imprecise) NV operation, and
2159 we're outside the range of NV integer precision */
2160#endif
2161 ) {
a43d94f2
NC
2162 if (SvNOK(sv))
2163 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2164 else {
2165 /* scalar has trailing garbage, eg "42a" */
2166 }
28e5dec8 2167 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2168 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2169 PTR2UV(sv),
2170 SvNVX(sv),
2171 SvIVX(sv)));
2172
2173 } else {
2174 /* IV not precise. No need to convert from PV, as NV
2175 conversion would already have cached IV if it detected
2176 that PV->IV would be better than PV->NV->IV
2177 flags already correct - don't set public IOK. */
2178 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2179 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2180 PTR2UV(sv),
2181 SvNVX(sv),
2182 SvIVX(sv)));
2183 }
2184 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2185 but the cast (NV)IV_MIN rounds to a the value less (more
2186 negative) than IV_MIN which happens to be equal to SvNVX ??
2187 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2188 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2189 (NV)UVX == NVX are both true, but the values differ. :-(
2190 Hopefully for 2s complement IV_MIN is something like
2191 0x8000000000000000 which will be exact. NWC */
d460ef45 2192 }
25da4f38 2193 else {
607fa7f2 2194 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8 2195 if (
659c4b96 2196 (SvNVX(sv) == (NV) SvUVX(sv))
28e5dec8
JH
2197#ifndef NV_PRESERVES_UV
2198 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2199 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2200 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2201 /* Don't flag it as "accurately an integer" if the number
2202 came from a (by definition imprecise) NV operation, and
2203 we're outside the range of NV integer precision */
2204#endif
a43d94f2 2205 && SvNOK(sv)
28e5dec8
JH
2206 )
2207 SvIOK_on(sv);
25da4f38 2208 SvIsUV_on(sv);
1c846c1f 2209 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2210 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2211 PTR2UV(sv),
57def98f
JH
2212 SvUVX(sv),
2213 SvUVX(sv)));
25da4f38 2214 }
748a9306 2215 }
cd84013a 2216 else if (SvPOKp(sv)) {
c2988b20 2217 UV value;
80e5abf2
DM
2218 int numtype;
2219 const char *s = SvPVX_const(sv);
2220 const STRLEN cur = SvCUR(sv);
2221
2222 /* short-cut for a single digit string like "1" */
2223
2224 if (cur == 1) {
2225 char c = *s;
2226 if (isDIGIT(c)) {
2227 if (SvTYPE(sv) < SVt_PVIV)
2228 sv_upgrade(sv, SVt_PVIV);
2229 (void)SvIOK_on(sv);
2230 SvIV_set(sv, (IV)(c - '0'));
2231 return FALSE;
2232 }
2233 }
2234
2235 numtype = grok_number(s, cur, &value);
af359546 2236 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2237 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2238 the same as the direct translation of the initial string
2239 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2240 be careful to ensure that the value with the .456 is around if the
2241 NV value is requested in the future).
1c846c1f 2242
af359546 2243 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2244 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2245 cache the NV if we are sure it's not needed.
25da4f38 2246 */
16b7a9a4 2247
c2988b20
NC
2248 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2249 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2250 == IS_NUMBER_IN_UV) {
5e045b90 2251 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2252 if (SvTYPE(sv) < SVt_PVIV)
2253 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2254 (void)SvIOK_on(sv);
c2988b20
NC
2255 } else if (SvTYPE(sv) < SVt_PVNV)
2256 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2257
a13f4dff 2258 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
75a57a38 2259 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
6b322424 2260 not_a_number(sv);
3823048b 2261 S_sv_setnv(aTHX_ sv, numtype);
a13f4dff
JH
2262 return FALSE;
2263 }
2264
f2524eef 2265 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2266 we aren't going to call atof() below. If NVs don't preserve UVs
2267 then the value returned may have more precision than atof() will
2268 return, even though value isn't perfectly accurate. */
2269 if ((numtype & (IS_NUMBER_IN_UV
2270#ifdef NV_PRESERVES_UV
2271 | IS_NUMBER_NOT_INT
2272#endif
2273 )) == IS_NUMBER_IN_UV) {
2274 /* This won't turn off the public IOK flag if it was set above */
2275 (void)SvIOKp_on(sv);
2276
2277 if (!(numtype & IS_NUMBER_NEG)) {
2278 /* positive */;
2279 if (value <= (UV)IV_MAX) {
45977657 2280 SvIV_set(sv, (IV)value);
c2988b20 2281 } else {
af359546 2282 /* it didn't overflow, and it was positive. */
607fa7f2 2283 SvUV_set(sv, value);
c2988b20
NC
2284 SvIsUV_on(sv);
2285 }
2286 } else {
2287 /* 2s complement assumption */
2288 if (value <= (UV)IV_MIN) {
53e2bfb7
DM
2289 SvIV_set(sv, value == (UV)IV_MIN
2290 ? IV_MIN : -(IV)value);
c2988b20
NC
2291 } else {
2292 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2293 I'm assuming it will be rare. */
c2988b20
NC
2294 if (SvTYPE(sv) < SVt_PVNV)
2295 sv_upgrade(sv, SVt_PVNV);
2296 SvNOK_on(sv);
2297 SvIOK_off(sv);
2298 SvIOKp_on(sv);
9d6ce603 2299 SvNV_set(sv, -(NV)value);
45977657 2300 SvIV_set(sv, IV_MIN);
c2988b20
NC
2301 }
2302 }
2303 }
2304 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2305 will be in the previous block to set the IV slot, and the next
2306 block to set the NV slot. So no else here. */
2307
2308 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2309 != IS_NUMBER_IN_UV) {
2310 /* It wasn't an (integer that doesn't overflow the UV). */
3823048b 2311 S_sv_setnv(aTHX_ sv, numtype);
28e5dec8 2312
c2988b20
NC
2313 if (! numtype && ckWARN(WARN_NUMERIC))
2314 not_a_number(sv);
28e5dec8 2315
88cb8500 2316 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
c2988b20 2317 PTR2UV(sv), SvNVX(sv)));
28e5dec8 2318
28e5dec8 2319#ifdef NV_PRESERVES_UV
af359546
NC
2320 (void)SvIOKp_on(sv);
2321 (void)SvNOK_on(sv);
e91de695
JH
2322#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2323 if (Perl_isnan(SvNVX(sv))) {
2324 SvUV_set(sv, 0);
2325 SvIsUV_on(sv);
2326 return FALSE;
2327 }
2328#endif
af359546
NC
2329 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2330 SvIV_set(sv, I_V(SvNVX(sv)));
2331 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2332 SvIOK_on(sv);
2333 } else {
6f207bd3 2334 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2335 }
2336 /* UV will not work better than IV */
2337 } else {
2338 if (SvNVX(sv) > (NV)UV_MAX) {
2339 SvIsUV_on(sv);
2340 /* Integer is inaccurate. NOK, IOKp, is UV */
2341 SvUV_set(sv, UV_MAX);
af359546
NC
2342 } else {
2343 SvUV_set(sv, U_V(SvNVX(sv)));
2344 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2345 NV preservse UV so can do correct comparison. */
2346 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2347 SvIOK_on(sv);
af359546 2348 } else {
6f207bd3 2349 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2350 }
2351 }
4b0c9573 2352 SvIsUV_on(sv);
af359546 2353 }
28e5dec8 2354#else /* NV_PRESERVES_UV */
c2988b20
NC
2355 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2356 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2357 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2358 grok_number above. The NV slot has just been set using
2359 Atof. */
560b0c46 2360 SvNOK_on(sv);
c2988b20
NC
2361 assert (SvIOKp(sv));
2362 } else {
2363 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2364 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2365 /* Small enough to preserve all bits. */
2366 (void)SvIOKp_on(sv);
2367 SvNOK_on(sv);
45977657 2368 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2369 if ((NV)(SvIVX(sv)) == SvNVX(sv))
c2988b20
NC
2370 SvIOK_on(sv);
2371 /* Assumption: first non-preserved integer is < IV_MAX,
2372 this NV is in the preserved range, therefore: */
2373 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2374 < (UV)IV_MAX)) {
32fdb065 2375 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
2376 }
2377 } else {
2378 /* IN_UV NOT_INT
2379 0 0 already failed to read UV.
2380 0 1 already failed to read UV.
2381 1 0 you won't get here in this case. IV/UV
2382 slot set, public IOK, Atof() unneeded.
2383 1 1 already read UV.
2384 so there's no point in sv_2iuv_non_preserve() attempting
2385 to use atol, strtol, strtoul etc. */
47031da6 2386# ifdef DEBUGGING
40a17c4c 2387 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2388# else
2389 sv_2iuv_non_preserve (sv);
2390# endif
c2988b20
NC
2391 }
2392 }
28e5dec8 2393#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2394 /* It might be more code efficient to go through the entire logic above
2395 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2396 gets complex and potentially buggy, so more programmer efficient
2397 to do it this way, by turning off the public flags: */
2398 if (!numtype)
2399 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2400 }
af359546
NC
2401 }
2402 else {
675c862f 2403 if (isGV_with_GP(sv))
159b6efe 2404 return glob_2number(MUTABLE_GV(sv));
180488f8 2405
4f62cd62 2406 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
af359546 2407 report_uninit(sv);
25da4f38
IZ
2408 if (SvTYPE(sv) < SVt_IV)
2409 /* Typically the caller expects that sv_any is not NULL now. */
2410 sv_upgrade(sv, SVt_IV);
af359546
NC
2411 /* Return 0 from the caller. */
2412 return TRUE;
2413 }
2414 return FALSE;
2415}
2416
2417/*
2418=for apidoc sv_2iv_flags
2419
2420Return the integer value of an SV, doing any necessary string
c5608a1f 2421conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
af359546
NC
2422Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2423
2424=cut
2425*/
2426
2427IV
5aaab254 2428Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2429{
1061065f 2430 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
4bac9ae4 2431
217f6fa3
FC
2432 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2433 && SvTYPE(sv) != SVt_PVFM);
2434
4bac9ae4
CS
2435 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2436 mg_get(sv);
2437
2438 if (SvROK(sv)) {
2439 if (SvAMAGIC(sv)) {
2440 SV * tmpstr;
2441 if (flags & SV_SKIP_OVERLOAD)
2442 return 0;
2443 tmpstr = AMG_CALLunary(sv, numer_amg);
2444 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2445 return SvIV(tmpstr);
2446 }
2447 }
2448 return PTR2IV(SvRV(sv));
2449 }
2450
8d919b0a 2451 if (SvVALID(sv) || isREGEXP(sv)) {
4e8879f3
DM
2452 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2453 must not let them cache IVs.
2b2b6d6d
NC
2454 In practice they are extremely unlikely to actually get anywhere
2455 accessible by user Perl code - the only way that I'm aware of is when
2456 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2457
2458 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2459 */
8d919b0a 2460 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2461 {
71c558c3 2462 UV value;
8d919b0a
FC
2463 const char * const ptr =
2464 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2465 const int numtype
2466 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2467
2468 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2469 == IS_NUMBER_IN_UV) {
2470 /* It's definitely an integer */
2471 if (numtype & IS_NUMBER_NEG) {
2472 if (value < (UV)IV_MIN)
2473 return -(IV)value;
2474 } else {
2475 if (value < (UV)IV_MAX)
2476 return (IV)value;
2477 }
2478 }
058b8ae2 2479
e91de695
JH
2480 /* Quite wrong but no good choices. */
2481 if ((numtype & IS_NUMBER_INFINITY)) {
2482 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2483 } else if ((numtype & IS_NUMBER_NAN)) {
2484 return 0; /* So wrong. */
2485 }
2486
71c558c3
NC
2487 if (!numtype) {
2488 if (ckWARN(WARN_NUMERIC))
2489 not_a_number(sv);
2490 }
8d919b0a 2491 return I_V(Atof(ptr));
e20b6c3b 2492 }
4bac9ae4
CS
2493 }
2494
2495 if (SvTHINKFIRST(sv)) {
af359546
NC
2496 if (SvREADONLY(sv) && !SvOK(sv)) {
2497 if (ckWARN(WARN_UNINITIALIZED))
2498 report_uninit(sv);
2499 return 0;
2500 }
2501 }
4bac9ae4 2502
af359546
NC
2503 if (!SvIOKp(sv)) {
2504 if (S_sv_2iuv_common(aTHX_ sv))
2505 return 0;
79072805 2506 }
4bac9ae4 2507
1d7c1841
GS
2508 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2509 PTR2UV(sv),SvIVX(sv)));
25da4f38 2510 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2511}
2512
645c22ef 2513/*
891f9566 2514=for apidoc sv_2uv_flags
645c22ef
DM
2515
2516Return the unsigned integer value of an SV, doing any necessary string
c5608a1f 2517conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
891f9566 2518Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2519
2520=cut
2521*/
2522
ff68c719 2523UV
5aaab254 2524Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2525{
1061065f 2526 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
4bac9ae4
CS
2527
2528 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2529 mg_get(sv);
2530
2531 if (SvROK(sv)) {
2532 if (SvAMAGIC(sv)) {
2533 SV *tmpstr;
2534 if (flags & SV_SKIP_OVERLOAD)
2535 return 0;
2536 tmpstr = AMG_CALLunary(sv, numer_amg);
2537 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2538 return SvUV(tmpstr);
2539 }
2540 }
2541 return PTR2UV(SvRV(sv));
2542 }
2543
8d919b0a 2544 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2545 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2546 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2547 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2548 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2549 {
71c558c3 2550 UV value;
8d919b0a
FC
2551 const char * const ptr =
2552 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2553 const int numtype
2554 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2555
2556 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2557 == IS_NUMBER_IN_UV) {
2558 /* It's definitely an integer */
2559 if (!(numtype & IS_NUMBER_NEG))
2560 return value;
2561 }
058b8ae2 2562
e91de695
JH
2563 /* Quite wrong but no good choices. */
2564 if ((numtype & IS_NUMBER_INFINITY)) {
2565 return UV_MAX; /* So wrong. */
2566 } else if ((numtype & IS_NUMBER_NAN)) {
2567 return 0; /* So wrong. */
2568 }
2569
71c558c3
NC
2570 if (!numtype) {
2571 if (ckWARN(WARN_NUMERIC))
2572 not_a_number(sv);
2573 }
8d919b0a 2574 return U_V(Atof(ptr));
e20b6c3b 2575 }
4bac9ae4
CS
2576 }
2577
2578 if (SvTHINKFIRST(sv)) {
0336b60e 2579 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2580 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2581 report_uninit(sv);
ff68c719
PP
2582 return 0;
2583 }
2584 }
4bac9ae4 2585
af359546
NC
2586 if (!SvIOKp(sv)) {
2587 if (S_sv_2iuv_common(aTHX_ sv))
2588 return 0;
ff68c719 2589 }
25da4f38 2590
1d7c1841
GS
2591 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2592 PTR2UV(sv),SvUVX(sv)));
25da4f38 2593 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2594}
2595
645c22ef 2596/*
196007d1 2597=for apidoc sv_2nv_flags
645c22ef
DM
2598
2599Return the num value of an SV, doing any necessary string or integer
c5608a1f 2600conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
39d5de13 2601Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2602
2603=cut
2604*/
2605
65202027 2606NV
5aaab254 2607Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2608{
1061065f
DD
2609 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2610
217f6fa3
FC
2611 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2612 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2613 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2614 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2615 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2616 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2617 const char *ptr;
39d5de13
DM
2618 if (flags & SV_GMAGIC)
2619 mg_get(sv);
463ee0b2
LW
2620 if (SvNOKp(sv))
2621 return SvNVX(sv);
cd84013a 2622 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2623 ptr = SvPVX_const(sv);
2624 grokpv:
041457d9 2625 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2626 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2627 not_a_number(sv);
8d919b0a 2628 return Atof(ptr);
a0d0e21e 2629 }
25da4f38 2630 if (SvIOKp(sv)) {
1c846c1f 2631 if (SvIsUV(sv))
65202027 2632 return (NV)SvUVX(sv);
25da4f38 2633 else
65202027 2634 return (NV)SvIVX(sv);
47a72cb8
NC
2635 }
2636 if (SvROK(sv)) {
2637 goto return_rok;
2638 }
8d919b0a
FC
2639 if (isREGEXP(sv)) {
2640 ptr = RX_WRAPPED((REGEXP *)sv);
2641 goto grokpv;
2642 }
47a72cb8
NC
2643 assert(SvTYPE(sv) >= SVt_PVMG);
2644 /* This falls through to the report_uninit near the end of the
2645 function. */
2646 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2647 if (SvROK(sv)) {
47a72cb8 2648 return_rok:
deb46114 2649 if (SvAMAGIC(sv)) {
aee036bb
DM
2650 SV *tmpstr;
2651 if (flags & SV_SKIP_OVERLOAD)
2652 return 0;
31d632c3 2653 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2654 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2655 return SvNV(tmpstr);
2656 }
2657 }
2658 return PTR2NV(SvRV(sv));
a0d0e21e 2659 }
0336b60e 2660 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2661 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2662 report_uninit(sv);
ed6116ce
LW
2663 return 0.0;
2664 }
79072805
LW
2665 }
2666 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2667 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2668 sv_upgrade(sv, SVt_NV);
097ee67d 2669 DEBUG_c({
f93f4e46 2670 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841 2671 PerlIO_printf(Perl_debug_log,
88cb8500 2672 "0x%"UVxf" num(%" NVgf ")\n",
1d7c1841 2673 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2674 RESTORE_NUMERIC_LOCAL();
2675 });
79072805
LW
2676 }
2677 else if (SvTYPE(sv) < SVt_PVNV)
2678 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2679 if (SvNOKp(sv)) {
2680 return SvNVX(sv);
61604483 2681 }
59d8ce62 2682 if (SvIOKp(sv)) {
9d6ce603 2683 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2684#ifdef NV_PRESERVES_UV
a43d94f2
NC
2685 if (SvIOK(sv))
2686 SvNOK_on(sv);
2687 else
2688 SvNOKp_on(sv);
28e5dec8
JH
2689#else
2690 /* Only set the public NV OK flag if this NV preserves the IV */
2691 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2692 if (SvIOK(sv) &&
2693 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2694 : (SvIVX(sv) == I_V(SvNVX(sv))))
2695 SvNOK_on(sv);
2696 else
2697 SvNOKp_on(sv);
2698#endif
93a17b20 2699 }
cd84013a 2700 else if (SvPOKp(sv)) {
c2988b20 2701 UV value;
3823048b 2702 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2703 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2704 not_a_number(sv);
28e5dec8 2705#ifdef NV_PRESERVES_UV
c2988b20
NC
2706 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2707 == IS_NUMBER_IN_UV) {
5e045b90 2708 /* It's definitely an integer */
9d6ce603 2709 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
66d83377 2710 } else {
3823048b 2711 S_sv_setnv(aTHX_ sv, numtype);
66d83377 2712 }
a43d94f2
NC
2713 if (numtype)
2714 SvNOK_on(sv);
2715 else
2716 SvNOKp_on(sv);
28e5dec8 2717#else
e91de695
JH
2718 SvNV_set(sv, Atof(SvPVX_const(sv)));
2719 /* Only set the public NV OK flag if this NV preserves the value in
2720 the PV at least as well as an IV/UV would.
2721 Not sure how to do this 100% reliably. */
2722 /* if that shift count is out of range then Configure's test is
2723 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2724 UV_BITS */
2725 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2726 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2727 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2728 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2729 /* Can't use strtol etc to convert this string, so don't try.
2730 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
c2988b20
NC
2731 SvNOK_on(sv);
2732 } else {
e91de695 2733 /* value has been set. It may not be precise. */
53e2bfb7 2734 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
e91de695
JH
2735 /* 2s complement assumption for (UV)IV_MIN */
2736 SvNOK_on(sv); /* Integer is too negative. */
c2988b20 2737 } else {
e91de695
JH
2738 SvNOKp_on(sv);
2739 SvIOKp_on(sv);
6fa402ec 2740
e91de695 2741 if (numtype & IS_NUMBER_NEG) {
02b08bbc
DM
2742 /* -IV_MIN is undefined, but we should never reach
2743 * this point with both IS_NUMBER_NEG and value ==
2744 * (UV)IV_MIN */
2745 assert(value != (UV)IV_MIN);
e91de695
JH
2746 SvIV_set(sv, -(IV)value);
2747 } else if (value <= (UV)IV_MAX) {
2748 SvIV_set(sv, (IV)value);
2749 } else {
2750 SvUV_set(sv, value);
2751 SvIsUV_on(sv);
2752 }
c2988b20 2753
e91de695
JH
2754 if (numtype & IS_NUMBER_NOT_INT) {
2755 /* I believe that even if the original PV had decimals,
2756 they are lost beyond the limit of the FP precision.
2757 However, neither is canonical, so both only get p
2758 flags. NWC, 2000/11/25 */
2759 /* Both already have p flags, so do nothing */
2760 } else {
2761 const NV nv = SvNVX(sv);
2762 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2763 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2764 if (SvIVX(sv) == I_V(nv)) {
2765 SvNOK_on(sv);
2766 } else {
2767 /* It had no "." so it must be integer. */
2768 }
2769 SvIOK_on(sv);
0f83c5a4 2770 } else {
e91de695
JH
2771 /* between IV_MAX and NV(UV_MAX).
2772 Could be slightly > UV_MAX */
2773
2774 if (numtype & IS_NUMBER_NOT_INT) {
2775 /* UV and NV both imprecise. */
0f83c5a4 2776 } else {
e91de695
JH
2777 const UV nv_as_uv = U_V(nv);
2778
2779 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2780 SvNOK_on(sv);
c2988b20 2781 }
e91de695 2782 SvIOK_on(sv);
c2988b20
NC
2783 }
2784 }
2785 }
2786 }
0f83c5a4 2787 }
e91de695
JH
2788 /* It might be more code efficient to go through the entire logic above
2789 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2790 gets complex and potentially buggy, so more programmer efficient
2791 to do it this way, by turning off the public flags: */
2792 if (!numtype)
2793 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2794#endif /* NV_PRESERVES_UV */
93a17b20 2795 }
79072805 2796 else {
e91de695
JH
2797 if (isGV_with_GP(sv)) {
2798 glob_2number(MUTABLE_GV(sv));
2799 return 0.0;
2800 }
180488f8 2801
e91de695
JH
2802 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2803 report_uninit(sv);
2804 assert (SvTYPE(sv) >= SVt_NV);
2805 /* Typically the caller expects that sv_any is not NULL now. */
2806 /* XXX Ilya implies that this is a bug in callers that assume this
2807 and ideally should be fixed. */
2808 return 0.0;
79072805 2809 }
097ee67d 2810 DEBUG_c({
e91de695
JH
2811 STORE_NUMERIC_LOCAL_SET_STANDARD();
2812 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2813 PTR2UV(sv), SvNVX(sv));
2814 RESTORE_NUMERIC_LOCAL();
2815 });
463ee0b2 2816 return SvNVX(sv);
79072805
LW
2817}
2818
800401ee
JH
2819/*
2820=for apidoc sv_2num
2821
2822Return an SV with the numeric value of the source SV, doing any necessary
d024d1a7
FC
2823reference or overload conversion. The caller is expected to have handled
2824get-magic already.
800401ee
JH
2825
2826=cut
2827*/
2828
2829SV *
5aaab254 2830Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2831{
7918f24d
NC
2832 PERL_ARGS_ASSERT_SV_2NUM;
2833
b9ee0594
RGS
2834 if (!SvROK(sv))
2835 return sv;
800401ee 2836 if (SvAMAGIC(sv)) {
31d632c3 2837 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2838 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2839 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2840 return sv_2num(tmpsv);
2841 }
2842 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2843}
2844
645c22ef
DM
2845/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2846 * UV as a string towards the end of buf, and return pointers to start and
2847 * end of it.
2848 *
2849 * We assume that buf is at least TYPE_CHARS(UV) long.
2850 */
2851
864dbfa3 2852static char *
5de3775c 2853S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2854{
25da4f38 2855 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2856 char * const ebuf = ptr;
25da4f38 2857 int sign;
25da4f38 2858
7918f24d
NC
2859 PERL_ARGS_ASSERT_UIV_2BUF;
2860
25da4f38
IZ
2861 if (is_uv)
2862 sign = 0;
2863 else if (iv >= 0) {
2864 uv = iv;
2865 sign = 0;
2866 } else {
53e2bfb7 2867 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
25da4f38
IZ
2868 sign = 1;
2869 }
2870 do {
eb160463 2871 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2872 } while (uv /= 10);
2873 if (sign)
2874 *--ptr = '-';
2875 *peob = ebuf;
2876 return ptr;
2877}
2878
bfaa02d5
JH
2879/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2880 * infinity or a not-a-number, writes the appropriate strings to the
2881 * buffer, including a zero byte. On success returns the written length,
3bde2d43
JH
2882 * excluding the zero byte, on failure (not an infinity, not a nan)
2883 * returns zero, assert-fails on maxlen being too short.
3823048b
JH
2884 *
2885 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2886 * shared string constants we point to, instead of generating a new
2887 * string for each instance. */
bfaa02d5 2888STATIC size_t
3823048b 2889S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
3bde2d43 2890 char* s = buffer;
bfaa02d5 2891 assert(maxlen >= 4);
3bde2d43
JH
2892 if (Perl_isinf(nv)) {
2893 if (nv < 0) {
2894 if (maxlen < 5) /* "-Inf\0" */
2895 return 0;
2896 *s++ = '-';
2897 } else if (plus) {
2898 *s++ = '+';
6e915616 2899 }
3bde2d43
JH
2900 *s++ = 'I';
2901 *s++ = 'n';
2902 *s++ = 'f';
2903 }
2904 else if (Perl_isnan(nv)) {
2905 *s++ = 'N';
2906 *s++ = 'a';
2907 *s++ = 'N';
2908 /* XXX optionally output the payload mantissa bits as
2909 * "(unsigned)" (to match the nan("...") C99 function,
2910 * or maybe as "(0xhhh...)" would make more sense...
2911 * provide a format string so that the user can decide?
2912 * NOTE: would affect the maxlen and assert() logic.*/
2913 }
2914 else {
2915 return 0;
bfaa02d5 2916 }
3bde2d43
JH
2917 assert((s == buffer + 3) || (s == buffer + 4));
2918 *s++ = 0;
2919 return s - buffer - 1; /* -1: excluding the zero byte */
bfaa02d5
JH
2920}
2921
2922/*
2923=for apidoc sv_2pv_flags
2924
796b6530 2925Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
c5608a1f 2926If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
bfaa02d5
JH
2927string if necessary. Normally invoked via the C<SvPV_flags> macro.
2928C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2929
2930=cut
2931*/
2932
2933char *
2934Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2935{
2936 char *s;
2937
2938 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2939
2940 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2941 && SvTYPE(sv) != SVt_PVFM);
2942 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2943 mg_get(sv);
2944 if (SvROK(sv)) {
2945 if (SvAMAGIC(sv)) {
2946 SV *tmpstr;
2947 if (flags & SV_SKIP_OVERLOAD)
2948 return NULL;
2949 tmpstr = AMG_CALLunary(sv, string_amg);
2950 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2951 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2952 /* Unwrap this: */
2953 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2954 */
2955
2956 char *pv;
2957 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2958 if (flags & SV_CONST_RETURN) {
2959 pv = (char *) SvPVX_const(tmpstr);
2960 } else {
2961 pv = (flags & SV_MUTABLE_RETURN)
2962 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2963 }
2964 if (lp)
2965 *lp = SvCUR(tmpstr);
2966 } else {
2967 pv = sv_2pv_flags(tmpstr, lp, flags);
2968 }
2969 if (SvUTF8(tmpstr))
2970 SvUTF8_on(sv);
2971 else
2972 SvUTF8_off(sv);
2973 return pv;
2974 }
2975 }
2976 {
2977 STRLEN len;
2978 char *retval;
2979 char *buffer;
2980 SV *const referent = SvRV(sv);
2981
2982 if (!referent) {
2983 len = 7;
2984 retval = buffer = savepvn("NULLREF", len);
2985 } else if (SvTYPE(referent) == SVt_REGEXP &&
2986 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2987 amagic_is_enabled(string_amg))) {
2988 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2989
2990 assert(re);
2991
2992 /* If the regex is UTF-8 we want the containing scalar to
2993 have an UTF-8 flag too */
2994 if (RX_UTF8(re))
2995 SvUTF8_on(sv);
2996 else
2997 SvUTF8_off(sv);
2998
2999 if (lp)
3000 *lp = RX_WRAPLEN(re);
3001
3002 return RX_WRAPPED(re);
3003 } else {
3004 const char *const typestr = sv_reftype(referent, 0);
3005 const STRLEN typelen = strlen(typestr);
3006 UV addr = PTR2UV(referent);
3007 const char *stashname = NULL;
3008 STRLEN stashnamelen = 0; /* hush, gcc */
3009 const char *buffer_end;
3010
3011 if (SvOBJECT(referent)) {
3012 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3013
3014 if (name) {
3015 stashname = HEK_KEY(name);
3016 stashnamelen = HEK_LEN(name);
3017
3018 if (HEK_UTF8(name)) {
3019 SvUTF8_on(sv);
3020 } else {
3021 SvUTF8_off(sv);
3022 }
3023 } else {
3024 stashname = "__ANON__";
3025 stashnamelen = 8;
3026 }
3027 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3028 + 2 * sizeof(UV) + 2 /* )\0 */;
3029 } else {
3030 len = typelen + 3 /* (0x */
3031 + 2 * sizeof(UV) + 2 /* )\0 */;
3032 }
fafee734 3033
4bac9ae4
CS
3034 Newx(buffer, len, char);
3035 buffer_end = retval = buffer + len;
3036
3037 /* Working backwards */
3038 *--retval = '\0';
3039 *--retval = ')';
3040 do {
3041 *--retval = PL_hexdigit[addr & 15];
3042 } while (addr >>= 4);
3043 *--retval = 'x';
3044 *--retval = '0';
3045 *--retval = '(';
3046
3047 retval -= typelen;
3048 memcpy(retval, typestr, typelen);
3049
3050 if (stashname) {
3051 *--retval = '=';
3052 retval -= stashnamelen;
3053 memcpy(retval, stashname, stashnamelen);
c080367d 3054 }
4bac9ae4
CS
3055 /* retval may not necessarily have reached the start of the
3056 buffer here. */
3057 assert (retval >= buffer);
3058
3059 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 3060 }
cdb061a3 3061 if (lp)
4bac9ae4
CS
3062 *lp = len;
3063 SAVEFREEPV(buffer);
3064 return retval;
79072805 3065 }
79072805 3066 }
4bac9ae4
CS
3067
3068 if (SvPOKp(sv)) {
3069 if (lp)
3070 *lp = SvCUR(sv);
3071 if (flags & SV_MUTABLE_RETURN)
3072 return SvPVX_mutable(sv);
3073 if (flags & SV_CONST_RETURN)
3074 return (char *)SvPVX_const(sv);
3075 return SvPVX(sv);
3076 }
3077
3078 if (SvIOK(sv)) {
28e5dec8
JH
3079 /* I'm assuming that if both IV and NV are equally valid then
3080 converting the IV is going to be more efficient */
e1ec3a88 3081 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
3082 char buf[TYPE_CHARS(UV)];
3083 char *ebuf, *ptr;
97a130b8 3084 STRLEN len;
28e5dec8
JH
3085
3086 if (SvTYPE(sv) < SVt_PVIV)
3087 sv_upgrade(sv, SVt_PVIV);
4ea1d550 3088 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 3089 len = ebuf - ptr;
5902b6a9 3090 /* inlined from sv_setpvn */
97a130b8
NC
3091 s = SvGROW_mutable(sv, len + 1);
3092 Move(ptr, s, len, char);
3093 s += len;
28e5dec8 3094 *s = '\0';
b127e37e 3095 SvPOK_on(sv);
28e5dec8 3096 }
4bac9ae4 3097 else if (SvNOK(sv)) {
79072805
LW
3098 if (SvTYPE(sv) < SVt_PVNV)
3099 sv_upgrade(sv, SVt_PVNV);
128eeacb
DD
3100 if (SvNVX(sv) == 0.0
3101#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3102 && !Perl_isnan(SvNVX(sv))
3103#endif
3104 ) {
29912d93
NC
3105 s = SvGROW_mutable(sv, 2);
3106 *s++ = '0';
3107 *s = '\0';
3108 } else {
5e85836e 3109 STRLEN len;
fb8cdbc5 3110 STRLEN size = 5; /* "-Inf\0" */
0c7e610f 3111
fb8cdbc5 3112 s = SvGROW_mutable(sv, size);
3823048b 3113 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
fb8cdbc5 3114 if (len > 0) {
0c7e610f 3115 s += len;
fb8cdbc5
JH
3116 SvPOK_on(sv);
3117 }
0c7e610f 3118 else {
0c7e610f 3119 /* some Xenix systems wipe out errno here */
fb8cdbc5
JH
3120 dSAVE_ERRNO;
3121
3840bff0
JH
3122 size =
3123 1 + /* sign */
3124 1 + /* "." */
3125 NV_DIG +
3126 1 + /* "e" */
3127 1 + /* sign */
3128 5 + /* exponent digits */
3129 1 + /* \0 */
3130 2; /* paranoia */
b127e37e 3131
fb8cdbc5 3132 s = SvGROW_mutable(sv, size);
b127e37e 3133#ifndef USE_LOCALE_NUMERIC
a4eca1d4
JH
3134 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3135
0c7e610f
JH
3136 SvPOK_on(sv);
3137#else
28acfe03 3138 {
3840bff0 3139 bool local_radix;
67d796ae
KW
3140 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3141 STORE_LC_NUMERIC_SET_TO_NEEDED();
3840bff0 3142
3dbc6af5 3143 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3840bff0
JH
3144 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3145 size += SvLEN(PL_numeric_radix_sv) - 1;
3146 s = SvGROW_mutable(sv, size);
3147 }
3148
a4eca1d4 3149 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
0c7e610f
JH
3150
3151 /* If the radix character is UTF-8, and actually is in the
3152 * output, turn on the UTF-8 flag for the scalar */
3dbc6af5
KW
3153 if ( local_radix
3154 && SvUTF8(PL_numeric_radix_sv)
3155 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3156 {
3840bff0
JH
3157 SvUTF8_on(sv);
3158 }
3159
0c7e610f 3160 RESTORE_LC_NUMERIC();
28acfe03 3161 }
68e8f474 3162
0c7e610f
JH
3163 /* We don't call SvPOK_on(), because it may come to
3164 * pass that the locale changes so that the
3165 * stringification we just did is no longer correct. We
3166 * will have to re-stringify every time it is needed */
b127e37e 3167#endif
0c7e610f
JH
3168 RESTORE_ERRNO;
3169 }
3170 while (*s) s++;
bbce6d69 3171 }
79072805 3172 }
4bac9ae4
CS
3173 else if (isGV_with_GP(sv)) {
3174 GV *const gv = MUTABLE_GV(sv);
3175 SV *const buffer = sv_newmortal();
8d1c3e26 3176
4bac9ae4 3177 gv_efullname3(buffer, gv, "*");
180488f8 3178
4bac9ae4
CS
3179 assert(SvPOK(buffer));
3180 if (SvUTF8(buffer))
3181 SvUTF8_on(sv);
3182 if (lp)
3183 *lp = SvCUR(buffer);
3184 return SvPVX(buffer);
3185 }
8d919b0a
FC
3186 else if (isREGEXP(sv)) {
3187 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3188 return RX_WRAPPED((REGEXP *)sv);
3189 }
4bac9ae4 3190 else {
cdb061a3 3191 if (lp)
00b6aa41 3192 *lp = 0;
9f621bb0
NC
3193 if (flags & SV_UNDEF_RETURNS_NULL)
3194 return NULL;
4f62cd62 3195 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3196 report_uninit(sv);
4bac9ae4
CS
3197 /* Typically the caller expects that sv_any is not NULL now. */
3198 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3199 sv_upgrade(sv, SVt_PV);
73d840c0 3200 return (char *)"";
79072805 3201 }
4bac9ae4 3202
cdb061a3 3203 {
823a54a3 3204 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3205 if (lp)
3206 *lp = len;
3207 SvCUR_set(sv, len);
3208 }
1d7c1841 3209 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3210 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3211 if (flags & SV_CONST_RETURN)
3212 return (char *)SvPVX_const(sv);
10516c54
NC
3213 if (flags & SV_MUTABLE_RETURN)
3214 return SvPVX_mutable(sv);
463ee0b2
LW
3215 return SvPVX(sv);
3216}
3217
645c22ef 3218/*
6050d10e
JP
3219=for apidoc sv_copypv
3220
3221Copies a stringified representation of the source SV into the
796b6530 3222destination SV. Automatically performs any necessary C<mg_get> and
54f0641b 3223coercion of numeric values into strings. Guaranteed to preserve
796b6530
KW
3224C<UTF8> flag even from overloaded objects. Similar in nature to
3225C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3226string. Mostly uses C<sv_2pv_flags> to do its work, except when that
6050d10e
JP
3227would lose the UTF-8'ness of the PV.
3228
4bac9ae4
CS
3229=for apidoc sv_copypv_nomg
3230
796b6530 3231Like C<sv_copypv>, but doesn't invoke get magic first.
4bac9ae4
CS
3232
3233=for apidoc sv_copypv_flags
3234
796b6530 3235Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
c5608a1f 3236has the C<SV_GMAGIC> bit set.
4bac9ae4 3237
6050d10e
JP
3238=cut
3239*/
3240
3241void
5aaab254 3242Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3243{
446eaa42 3244 STRLEN len;
4bac9ae4 3245 const char *s;
7918f24d 3246
4bac9ae4 3247 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3248
c77ed9ca 3249 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
cb50f42d 3250 sv_setpvn(dsv,s,len);
446eaa42 3251 if (SvUTF8(ssv))
cb50f42d 3252 SvUTF8_on(dsv);
446eaa42 3253 else
cb50f42d 3254 SvUTF8_off(dsv);
6050d10e
JP
3255}
3256
3257/*
645c22ef
DM
3258=for apidoc sv_2pvbyte
3259
796b6530 3260Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
1e54db1a 3261to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3262side-effect.
3263
3264Usually accessed via the C<SvPVbyte> macro.
3265
3266=cut
3267*/
3268
7340a771 3269char *
5aaab254 3270Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3271{
7918f24d
NC
3272 PERL_ARGS_ASSERT_SV_2PVBYTE;
3273
48120f8f 3274 SvGETMAGIC(sv);
4499db73
FC
3275 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3276 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3277 SV *sv2 = sv_newmortal();
48120f8f 3278 sv_copypv_nomg(sv2,sv);
a901b181
FC
3279 sv = sv2;
3280 }
0875d2fe 3281 sv_utf8_downgrade(sv,0);
71eb6d8c 3282 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3283}
3284
645c22ef 3285/*
035cbb0e
RGS
3286=for apidoc sv_2pvutf8
3287
796b6530 3288Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
035cbb0e
RGS
3289to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3290
3291Usually accessed via the C<SvPVutf8> macro.
3292
3293=cut
3294*/
645c22ef 3295
7340a771 3296char *
5aaab254 3297Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3298{
7918f24d
NC
3299 PERL_ARGS_ASSERT_SV_2PVUTF8;
3300
4499db73
FC
3301 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3302 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3303 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3304 else
3305 SvGETMAGIC(sv);
3306 sv_utf8_upgrade_nomg(sv);
c3ec315f 3307 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3308}
1c846c1f 3309
7ee2227d 3310
645c22ef
DM
3311/*
3312=for apidoc sv_2bool
3313
796b6530
KW
3314This macro is only used by C<sv_true()> or its macro equivalent, and only if
3315the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3316It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
06c841cf
FC
3317
3318=for apidoc sv_2bool_flags
3319
796b6530
KW
3320This function is only used by C<sv_true()> and friends, and only if
3321the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3322contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
06c841cf 3323
645c22ef
DM
3324
3325=cut
3326*/
3327
463ee0b2 3328bool
9d176cd8 3329Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
463ee0b2 3330{
06c841cf 3331 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3332
9d176cd8 3333 restart:
06c841cf 3334 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3335
a0d0e21e
LW
3336 if (!SvOK(sv))
3337 return 0;
3338 if (SvROK(sv)) {
fabdb6c0 3339 if (SvAMAGIC(sv)) {
31d632c3 3340 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
9d176cd8
DD
3341 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3342 bool svb;
3343 sv = tmpsv;
3344 if(SvGMAGICAL(sv)) {
3345 flags = SV_GMAGIC;
3346 goto restart; /* call sv_2bool */
3347 }
3348 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3349 else if(!SvOK(sv)) {
3350 svb = 0;
3351 }
3352 else if(SvPOK(sv)) {
3353 svb = SvPVXtrue(sv);
3354 }
3355 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3356 svb = (SvIOK(sv) && SvIVX(sv) != 0)
659c4b96 3357 || (SvNOK(sv) && SvNVX(sv) != 0.0);
9d176cd8
DD
3358 }
3359 else {
3360 flags = 0;
3361 goto restart; /* call sv_2bool_nomg */
3362 }
3363 return cBOOL(svb);
3364 }
fabdb6c0
AL
3365 }
3366 return SvRV(sv) != 0;
a0d0e21e 3367 }
85b7d9b3
FC
3368 if (isREGEXP(sv))
3369 return
3370 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
4bac9ae4 3371 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3372}
3373
c461cf8f
JH
3374/*
3375=for apidoc sv_utf8_upgrade
3376
78ea37eb 3377Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3378Forces the SV to string form if it is not already.
2bbc8d55 3379Will C<mg_get> on C<sv> if appropriate.
796b6530 3380Always sets the C<SvUTF8> flag to avoid future validity checks even
2bbc8d55
SP
3381if the whole string is the same in UTF-8 as not.
3382Returns the number of bytes in the converted string
c461cf8f 3383
0efd0472 3384This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3385use the Encode extension for that.
3386
fe749c9a
KW
3387=for apidoc sv_utf8_upgrade_nomg
3388
796b6530 3389Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
fe749c9a 3390
8d6d96c1
HS
3391=for apidoc sv_utf8_upgrade_flags
3392
78ea37eb 3393Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3394Forces the SV to string form if it is not already.
8d6d96c1 3395Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3396if all the bytes are invariant in UTF-8.
3397If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55 3398will C<mg_get> on C<sv> if appropriate, else not.
2a590426 3399
796b6530 3400If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
2a590426
KW
3401will expand when converted to UTF-8, and skips the extra work of checking for
3402that. Typically this flag is used by a routine that has already parsed the
3403string and found such characters, and passes this information on so that the
3404work doesn't have to be repeated.
3405
3406Returns the number of bytes in the converted string.
8d6d96c1 3407
0efd0472 3408This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3409use the Encode extension for that.
3410
2a590426 3411=for apidoc sv_utf8_upgrade_flags_grow
b3ab6785 3412
796b6530
KW
3413Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3414the number of unused bytes the string of C<sv> is guaranteed to have free after
2a590426
KW
3415it upon return. This allows the caller to reserve extra space that it intends
3416to fill, to avoid extra grows.
b3ab6785 3417
2a590426
KW
3418C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3419are implemented in terms of this function.
3420
3421Returns the number of bytes in the converted string (not including the spares).
3422
3423=cut
b3ab6785
KW
3424
3425(One might think that the calling routine could pass in the position of the
2a590426
KW
3426first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3427have to be found again. But that is not the case, because typically when the
3428caller is likely to use this flag, it won't be calling this routine unless it
3429finds something that won't fit into a byte. Otherwise it tries to not upgrade
3430and just use bytes. But some things that do fit into a byte are variants in
3431utf8, and the caller may not have been keeping track of these.)
b3ab6785 3432
6602b933
KW
3433If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3434C<NUL> isn't guaranteed due to having other routines do the work in some input
3435cases, or if the input is already flagged as being in utf8.
b3ab6785
KW
3436
3437The speed of this could perhaps be improved for many cases if someone wanted to
3438write a fast function that counts the number of variant characters in a string,
3439especially if it could return the position of the first one.
3440
8d6d96c1
HS
3441*/
3442
3443STRLEN
5aaab254 3444Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3445{
b3ab6785 3446 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3447
808c356f
RGS
3448 if (sv == &PL_sv_undef)
3449 return 0;
892f9127 3450 if (!SvPOK_nog(sv)) {
e0e62c2a 3451 STRLEN len = 0;
d52b7888
NC
3452 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3453 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3454 if (SvUTF8(sv)) {
3455 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3456 return len;
b3ab6785 3457 }
d52b7888 3458 } else {
33fb6f35 3459 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3460 }
e0e62c2a 3461 }
4411f3b6 3462
f5cee72b 3463 if (SvUTF8(sv)) {
b3ab6785 3464 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3465 return SvCUR(sv);
f5cee72b 3466 }
5fec3b1d 3467
765f542d 3468 if (SvIsCOW(sv)) {
c56ed9f6 3469 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3470 }
3471
4e93345f
KW
3472 if (SvCUR(sv) == 0) {
3473 if (extra) SvGROW(sv, extra);
3474 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3475 /* This function could be much more efficient if we
2bbc8d55 3476 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3477 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3478 * make the loop as fast as possible (although there are certainly ways
3479 * to speed this up, eg. through vectorization) */
3480 U8 * s = (U8 *) SvPVX_const(sv);
3481 U8 * e = (U8 *) SvEND(sv);
3482 U8 *t = s;
3483 STRLEN two_byte_count = 0;
c4e7c712 3484
b3ab6785
KW
3485 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3486
3487 /* See if really will need to convert to utf8. We mustn't rely on our
3488 * incoming SV being well formed and having a trailing '\0', as certain
3489 * code in pp_formline can send us partially built SVs. */
3490
c4e7c712 3491 while (t < e) {
53c1dcc0 3492 const U8 ch = *t++;
6f2d5cbc 3493 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
b3ab6785
KW
3494
3495 t--; /* t already incremented; re-point to first variant */
3496 two_byte_count = 1;
3497 goto must_be_utf8;
c4e7c712 3498 }
b3ab6785
KW
3499
3500 /* utf8 conversion not needed because all are invariants. Mark as
3501 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3502 SvUTF8_on(sv);
7f0bfbea 3503 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3504 return SvCUR(sv);
3505
7b52d656 3506 must_be_utf8:
b3ab6785
KW
3507
3508 /* Here, the string should be converted to utf8, either because of an
3509 * input flag (two_byte_count = 0), or because a character that
3510 * requires 2 bytes was found (two_byte_count = 1). t points either to
3511 * the beginning of the string (if we didn't examine anything), or to
3512 * the first variant. In either case, everything from s to t - 1 will
3513 * occupy only 1 byte each on output.
3514 *
3515 * There are two main ways to convert. One is to create a new string
3516 * and go through the input starting from the beginning, appending each
3517 * converted value onto the new string as we go along. It's probably
3518 * best to allocate enough space in the string for the worst possible
3519 * case rather than possibly running out of space and having to
3520 * reallocate and then copy what we've done so far. Since everything
3521 * from s to t - 1 is invariant, the destination can be initialized
3522 * with these using a fast memory copy
3523 *
3524 * The other way is to figure out exactly how big the string should be
3525 * by parsing the entire input. Then you don't have to make it big
3526 * enough to handle the worst possible case, and more importantly, if
3527 * the string you already have is large enough, you don't have to
3528 * allocate a new string, you can copy the last character in the input
3529 * string to the final position(s) that will be occupied by the
3530 * converted string and go backwards, stopping at t, since everything
3531 * before that is invariant.
3532 *
3533 * There are advantages and disadvantages to each method.
3534 *
3535 * In the first method, we can allocate a new string, do the memory
3536 * copy from the s to t - 1, and then proceed through the rest of the
3537 * string byte-by-byte.
3538 *
3539 * In the second method, we proceed through the rest of the input
3540 * string just calculating how big the converted string will be. Then
3541 * there are two cases:
3542 * 1) if the string has enough extra space to handle the converted
3543 * value. We go backwards through the string, converting until we
3544 * get to the position we are at now, and then stop. If this
3545 * position is far enough along in the string, this method is
3546 * faster than the other method. If the memory copy were the same
3547 * speed as the byte-by-byte loop, that position would be about
3548 * half-way, as at the half-way mark, parsing to the end and back
3549 * is one complete string's parse, the same amount as starting
3550 * over and going all the way through. Actually, it would be
3551 * somewhat less than half-way, as it's faster to just count bytes
3552 * than to also copy, and we don't have the overhead of allocating
3553 * a new string, changing the scalar to use it, and freeing the
3554 * existing one. But if the memory copy is fast, the break-even
3555 * point is somewhere after half way. The counting loop could be
3556 * sped up by vectorization, etc, to move the break-even point
3557 * further towards the beginning.
3558 * 2) if the string doesn't have enough space to handle the converted
3559 * value. A new string will have to be allocated, and one might
3560 * as well, given that, start from the beginning doing the first
3561 * method. We've spent extra time parsing the string and in
3562 * exchange all we've gotten is that we know precisely how big to
3563 * make the new one. Perl is more optimized for time than space,
3564 * so this case is a loser.
3565 * So what I've decided to do is not use the 2nd method unless it is
3566 * guaranteed that a new string won't have to be allocated, assuming
3567 * the worst case. I also decided not to put any more conditions on it
3568 * than this, for now. It seems likely that, since the worst case is
3569 * twice as big as the unknown portion of the string (plus 1), we won't
3570 * be guaranteed enough space, causing us to go to the first method,
3571 * unless the string is short, or the first variant character is near
3572 * the end of it. In either of these cases, it seems best to use the
3573 * 2nd method. The only circumstance I can think of where this would
3574 * be really slower is if the string had once had much more data in it
3575 * than it does now, but there is still a substantial amount in it */
3576
3577 {
3578 STRLEN invariant_head = t - s;
3579 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3580 if (SvLEN(sv) < size) {
3581
3582 /* Here, have decided to allocate a new string */
3583
3584 U8 *dst;
3585 U8 *d;
3586
3587 Newx(dst, size, U8);
3588
3589 /* If no known invariants at the beginning of the input string,
3590 * set so starts from there. Otherwise, can use memory copy to
3591 * get up to where we are now, and then start from here */
3592
5b26a7b3 3593 if (invariant_head == 0) {
b3ab6785
KW
3594 d = dst;
3595 } else {
3596 Copy(s, dst, invariant_head, char);
3597 d = dst + invariant_head;
3598 }
3599
3600 while (t < e) {