This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
sv.c: Remove redundant assignments
[perl5.git] / sv.c
CommitLineData
a0d0e21e 1/* sv.c
79072805 2 *
1129b882 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
83706693
RGS
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
5 * and others
79072805
LW
6 *
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
9 *
4ac71550
TC
10 */
11
12/*
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
14 * --Pippin
15 *
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
17 */
18
19/*
645c22ef
DM
20 *
21 *
5e045b90
AMS
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
27 * in the pp*.c files.
79072805
LW
28 */
29
30#include "EXTERN.h"
864dbfa3 31#define PERL_IN_SV_C
79072805 32#include "perl.h"
d2f185dc 33#include "regcomp.h"
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
e8c6a474
YO
51#ifdef PERL_NEW_COPY_ON_WRITE
52# ifndef SV_COW_THRESHOLD
53# define SV_COW_THRESHOLD 0 /* COW iff len > K */
54# endif
55# ifndef SV_COWBUF_THRESHOLD
56# define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57# endif
58# ifndef SV_COW_MAX_WASTE_THRESHOLD
59# define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60# endif
61# ifndef SV_COWBUF_WASTE_THRESHOLD
62# define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63# endif
64# ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
65# define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66# endif
67# ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
68# define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69# endif
70#endif
71/* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 hold is 0. */
73#if SV_COW_THRESHOLD
74# define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
75#else
76# define GE_COW_THRESHOLD(cur) 1
77#endif
78#if SV_COWBUF_THRESHOLD
79# define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
80#else
81# define GE_COWBUF_THRESHOLD(cur) 1
82#endif
83#if SV_COW_MAX_WASTE_THRESHOLD
84# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
85#else
86# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
87#endif
88#if SV_COWBUF_WASTE_THRESHOLD
89# define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
90#else
91# define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
92#endif
93#if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
94# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
95#else
96# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
97#endif
98#if SV_COWBUF_WASTE_FACTOR_THRESHOLD
99# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
100#else
101# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102#endif
103
104#define CHECK_COW_THRESHOLD(cur,len) (\
105 GE_COW_THRESHOLD((cur)) && \
106 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
107 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
108)
109#define CHECK_COWBUF_THRESHOLD(cur,len) (\
110 GE_COWBUF_THRESHOLD((cur)) && \
111 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
112 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113)
cca0492e 114
e23c8137 115#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 116/* if adding more checks watch out for the following tests:
e23c8137
JH
117 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
118 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 * --jhi
120 */
6f207bd3 121# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
122 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
123 assert((cache)[2] <= (cache)[3]); \
124 assert((cache)[3] <= (cache)[1]);} \
125 } STMT_END
e23c8137 126#else
6f207bd3 127# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
128#endif
129
f8c7b90f 130#ifdef PERL_OLD_COPY_ON_WRITE
765f542d 131#define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
607fa7f2 132#define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
765f542d 133#endif
645c22ef
DM
134
135/* ============================================================================
136
137=head1 Allocation and deallocation of SVs.
d2a0f284
JC
138An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
139sv, av, hv...) contains type and reference count information, and for
140many types, a pointer to the body (struct xrv, xpv, xpviv...), which
141contains fields specific to each type. Some types store all they need
142in the head, so don't have a body.
143
486ec47a 144In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
145and bodies are allocated out of arenas, which by default are
146approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
147Sv-bodies are allocated by their sv-type, guaranteeing size
148consistency needed to allocate safely from arrays.
149
d2a0f284
JC
150For SV-heads, the first slot in each arena is reserved, and holds a
151link to the next arena, some flags, and a note of the number of slots.
152Snaked through each arena chain is a linked list of free items; when
153this becomes empty, an extra arena is allocated and divided up into N
154items which are threaded into the free list.
155
156SV-bodies are similar, but they use arena-sets by default, which
157separate the link and info from the arena itself, and reclaim the 1st
158slot in the arena. SV-bodies are further described later.
645c22ef
DM
159
160The following global variables are associated with arenas:
161
7fefc6c1
KW
162 PL_sv_arenaroot pointer to list of SV arenas
163 PL_sv_root pointer to list of free SV structures
645c22ef 164
7fefc6c1
KW
165 PL_body_arenas head of linked-list of body arenas
166 PL_body_roots[] array of pointers to list of free bodies of svtype
167 arrays are indexed by the svtype needed
93e68bfb 168
d2a0f284
JC
169A few special SV heads are not allocated from an arena, but are
170instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
171The size of arenas can be changed from the default by setting
172PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
173
174The SV arena serves the secondary purpose of allowing still-live SVs
175to be located and destroyed during final cleanup.
176
177At the lowest level, the macros new_SV() and del_SV() grab and free
178an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
179to return the SV to the free list with error checking.) new_SV() calls
180more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
181SVs in the free list have their SvTYPE field set to all ones.
182
ff276b08 183At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 184perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 185start of the interpreter.
645c22ef 186
645c22ef
DM
187The function visit() scans the SV arenas list, and calls a specified
188function for each SV it finds which is still live - ie which has an SvTYPE
189other than all 1's, and a non-zero SvREFCNT. visit() is used by the
190following functions (specified as [function that calls visit()] / [function
191called by visit() for each SV]):
192
193 sv_report_used() / do_report_used()
f2524eef 194 dump all remaining SVs (debugging aid)
645c22ef 195
e4487e9b 196 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 197 do_clean_named_io_objs(),do_curse()
645c22ef 198 Attempt to free all objects pointed to by RVs,
e76981f9
FC
199 try to do the same for all objects indir-
200 ectly referenced by typeglobs too, and
201 then do a final sweep, cursing any
202 objects that remain. Called once from
645c22ef
DM
203 perl_destruct(), prior to calling sv_clean_all()
204 below.
205
206 sv_clean_all() / do_clean_all()
207 SvREFCNT_dec(sv) each remaining SV, possibly
208 triggering an sv_free(). It also sets the
209 SVf_BREAK flag on the SV to indicate that the
210 refcnt has been artificially lowered, and thus
211 stopping sv_free() from giving spurious warnings
212 about SVs which unexpectedly have a refcnt
213 of zero. called repeatedly from perl_destruct()
214 until there are no SVs left.
215
93e68bfb 216=head2 Arena allocator API Summary
645c22ef
DM
217
218Private API to rest of sv.c
219
220 new_SV(), del_SV(),
221
df0f0429 222 new_XPVNV(), del_XPVGV(),
645c22ef
DM
223 etc
224
225Public API:
226
8cf8f3d1 227 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 228
645c22ef
DM
229=cut
230
3e8320cc 231 * ========================================================================= */
645c22ef 232
4561caa4
CS
233/*
234 * "A time to plant, and a time to uproot what was planted..."
235 */
236
d7a2c63c
MHM
237#ifdef PERL_MEM_LOG
238# define MEM_LOG_NEW_SV(sv, file, line, func) \
239 Perl_mem_log_new_sv(sv, file, line, func)
240# define MEM_LOG_DEL_SV(sv, file, line, func) \
241 Perl_mem_log_del_sv(sv, file, line, func)
242#else
243# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
244# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
245#endif
246
fd0854ff 247#ifdef DEBUG_LEAKING_SCALARS
484e6108
FC
248# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
249 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
250 } STMT_END
d7a2c63c
MHM
251# define DEBUG_SV_SERIAL(sv) \
252 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
253 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
254#else
255# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 256# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
257#endif
258
48614a46
NC
259#ifdef PERL_POISON
260# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 261# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
262/* Whilst I'd love to do this, it seems that things like to check on
263 unreferenced scalars
7e337ee0 264# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 265*/
7e337ee0
JH
266# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
267 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
268#else
269# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 270# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
48614a46
NC
271# define POSION_SV_HEAD(sv)
272#endif
273
990198f0
DM
274/* Mark an SV head as unused, and add to free list.
275 *
276 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
277 * its refcount artificially decremented during global destruction, so
278 * there may be dangling pointers to it. The last thing we want in that
279 * case is for it to be reused. */
280
053fc874
GS
281#define plant_SV(p) \
282 STMT_START { \
990198f0 283 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
284 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
285 DEBUG_SV_SERIAL(p); \
fd0854ff 286 FREE_SV_DEBUG_FILE(p); \
48614a46 287 POSION_SV_HEAD(p); \
053fc874 288 SvFLAGS(p) = SVTYPEMASK; \
990198f0 289 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 290 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
291 PL_sv_root = (p); \
292 } \
053fc874
GS
293 --PL_sv_count; \
294 } STMT_END
a0d0e21e 295
053fc874
GS
296#define uproot_SV(p) \
297 STMT_START { \
298 (p) = PL_sv_root; \
daba3364 299 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
300 ++PL_sv_count; \
301 } STMT_END
302
645c22ef 303
cac9b346
NC
304/* make some more SVs by adding another arena */
305
cac9b346
NC
306STATIC SV*
307S_more_sv(pTHX)
308{
309 SV* sv;
9a87bd09
NC
310 char *chunk; /* must use New here to match call to */
311 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
312 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
313 uproot_SV(sv);
314 return sv;
315}
316
645c22ef
DM
317/* new_SV(): return a new, empty SV head */
318
eba0f806
DM
319#ifdef DEBUG_LEAKING_SCALARS
320/* provide a real function for a debugger to play with */
321STATIC SV*
d7a2c63c 322S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
323{
324 SV* sv;
325
eba0f806
DM
326 if (PL_sv_root)
327 uproot_SV(sv);
328 else
cac9b346 329 sv = S_more_sv(aTHX);
eba0f806
DM
330 SvANY(sv) = 0;
331 SvREFCNT(sv) = 1;
332 SvFLAGS(sv) = 0;
fd0854ff 333 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
334 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
335 ? PL_parser->copline
336 : PL_curcop
f24aceb1
DM
337 ? CopLINE(PL_curcop)
338 : 0
e385c3bf 339 );
fd0854ff 340 sv->sv_debug_inpad = 0;
cd676548 341 sv->sv_debug_parent = NULL;
484e6108 342 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
343
344 sv->sv_debug_serial = PL_sv_serial++;
345
346 MEM_LOG_NEW_SV(sv, file, line, func);
347 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
348 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
349
eba0f806
DM
350 return sv;
351}
d7a2c63c 352# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
353
354#else
355# define new_SV(p) \
053fc874 356 STMT_START { \
053fc874
GS
357 if (PL_sv_root) \
358 uproot_SV(p); \
359 else \
cac9b346 360 (p) = S_more_sv(aTHX); \
053fc874
GS
361 SvANY(p) = 0; \
362 SvREFCNT(p) = 1; \
363 SvFLAGS(p) = 0; \
d7a2c63c 364 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 365 } STMT_END
eba0f806 366#endif
463ee0b2 367
645c22ef
DM
368
369/* del_SV(): return an empty SV head to the free list */
370
a0d0e21e 371#ifdef DEBUGGING
4561caa4 372
053fc874
GS
373#define del_SV(p) \
374 STMT_START { \
aea4f609 375 if (DEBUG_D_TEST) \
053fc874
GS
376 del_sv(p); \
377 else \
378 plant_SV(p); \
053fc874 379 } STMT_END
a0d0e21e 380
76e3520e 381STATIC void
cea2e8a9 382S_del_sv(pTHX_ SV *p)
463ee0b2 383{
7918f24d
NC
384 PERL_ARGS_ASSERT_DEL_SV;
385
aea4f609 386 if (DEBUG_D_TEST) {
4633a7c4 387 SV* sva;
a3b680e6 388 bool ok = 0;
daba3364 389 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
390 const SV * const sv = sva + 1;
391 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 392 if (p >= sv && p < svend) {
a0d0e21e 393 ok = 1;
c0ff570e
NC
394 break;
395 }
a0d0e21e
LW
396 }
397 if (!ok) {
9b387841
NC
398 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
399 "Attempt to free non-arena SV: 0x%"UVxf
400 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
401 return;
402 }
403 }
4561caa4 404 plant_SV(p);
463ee0b2 405}
a0d0e21e 406
4561caa4
CS
407#else /* ! DEBUGGING */
408
409#define del_SV(p) plant_SV(p)
410
411#endif /* DEBUGGING */
463ee0b2 412
645c22ef
DM
413
414/*
ccfc67b7
JH
415=head1 SV Manipulation Functions
416
645c22ef
DM
417=for apidoc sv_add_arena
418
419Given a chunk of memory, link it to the head of the list of arenas,
420and split it into a list of free SVs.
421
422=cut
423*/
424
d2bd4e7f
NC
425static void
426S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 427{
daba3364 428 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
429 SV* sv;
430 SV* svend;
4633a7c4 431
7918f24d
NC
432 PERL_ARGS_ASSERT_SV_ADD_ARENA;
433
4633a7c4 434 /* The first SV in an arena isn't an SV. */
3280af22 435 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
436 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
437 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
438
3280af22
NIS
439 PL_sv_arenaroot = sva;
440 PL_sv_root = sva + 1;
4633a7c4
LW
441
442 svend = &sva[SvREFCNT(sva) - 1];
443 sv = sva + 1;
463ee0b2 444 while (sv < svend) {
3eef1deb 445 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 446#ifdef DEBUGGING
978b032e 447 SvREFCNT(sv) = 0;
03e36789 448#endif
4b69cbe3 449 /* Must always set typemask because it's always checked in on cleanup
03e36789 450 when the arenas are walked looking for objects. */
8990e307 451 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
452 sv++;
453 }
3eef1deb 454 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
455#ifdef DEBUGGING
456 SvREFCNT(sv) = 0;
457#endif
4633a7c4
LW
458 SvFLAGS(sv) = SVTYPEMASK;
459}
460
055972dc
DM
461/* visit(): call the named function for each non-free SV in the arenas
462 * whose flags field matches the flags/mask args. */
645c22ef 463
5226ed68 464STATIC I32
de37a194 465S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 466{
4633a7c4 467 SV* sva;
5226ed68 468 I32 visited = 0;
8990e307 469
7918f24d
NC
470 PERL_ARGS_ASSERT_VISIT;
471
daba3364 472 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
473 const SV * const svend = &sva[SvREFCNT(sva)];
474 SV* sv;
4561caa4 475 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 476 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
477 && (sv->sv_flags & mask) == flags
478 && SvREFCNT(sv))
479 {
942481a7 480 (*f)(aTHX_ sv);
5226ed68
JH
481 ++visited;
482 }
8990e307
LW
483 }
484 }
5226ed68 485 return visited;
8990e307
LW
486}
487
758a08c3
JH
488#ifdef DEBUGGING
489
645c22ef
DM
490/* called by sv_report_used() for each live SV */
491
492static void
5fa45a31 493do_report_used(pTHX_ SV *const sv)
645c22ef 494{
e4787c0c 495 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
496 PerlIO_printf(Perl_debug_log, "****\n");
497 sv_dump(sv);
498 }
499}
758a08c3 500#endif
645c22ef
DM
501
502/*
503=for apidoc sv_report_used
504
fde67290 505Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
506
507=cut
508*/
509
8990e307 510void
864dbfa3 511Perl_sv_report_used(pTHX)
4561caa4 512{
ff270d3a 513#ifdef DEBUGGING
055972dc 514 visit(do_report_used, 0, 0);
96a5add6
AL
515#else
516 PERL_UNUSED_CONTEXT;
ff270d3a 517#endif
4561caa4
CS
518}
519
645c22ef
DM
520/* called by sv_clean_objs() for each live SV */
521
522static void
de37a194 523do_clean_objs(pTHX_ SV *const ref)
645c22ef 524{
ea724faa
NC
525 assert (SvROK(ref));
526 {
823a54a3
AL
527 SV * const target = SvRV(ref);
528 if (SvOBJECT(target)) {
529 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
530 if (SvWEAKREF(ref)) {
531 sv_del_backref(target, ref);
532 SvWEAKREF_off(ref);
533 SvRV_set(ref, NULL);
534 } else {
535 SvROK_off(ref);
536 SvRV_set(ref, NULL);
fc2b2dca 537 SvREFCNT_dec_NN(target);
823a54a3 538 }
645c22ef
DM
539 }
540 }
645c22ef
DM
541}
542
645c22ef 543
e4487e9b
DM
544/* clear any slots in a GV which hold objects - except IO;
545 * called by sv_clean_objs() for each live GV */
546
645c22ef 547static void
f30de749 548do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 549{
57ef47cc 550 SV *obj;
ea724faa 551 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 552 assert(isGV_with_GP(sv));
57ef47cc
DM
553 if (!GvGP(sv))
554 return;
555
556 /* freeing GP entries may indirectly free the current GV;
557 * hold onto it while we mess with the GP slots */
558 SvREFCNT_inc(sv);
559
560 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
561 DEBUG_D((PerlIO_printf(Perl_debug_log,
562 "Cleaning named glob SV object:\n "), sv_dump(obj)));
563 GvSV(sv) = NULL;
fc2b2dca 564 SvREFCNT_dec_NN(obj);
57ef47cc
DM
565 }
566 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
567 DEBUG_D((PerlIO_printf(Perl_debug_log,
568 "Cleaning named glob AV object:\n "), sv_dump(obj)));
569 GvAV(sv) = NULL;
fc2b2dca 570 SvREFCNT_dec_NN(obj);
57ef47cc
DM
571 }
572 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
573 DEBUG_D((PerlIO_printf(Perl_debug_log,
574 "Cleaning named glob HV object:\n "), sv_dump(obj)));
575 GvHV(sv) = NULL;
fc2b2dca 576 SvREFCNT_dec_NN(obj);
57ef47cc
DM
577 }
578 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
579 DEBUG_D((PerlIO_printf(Perl_debug_log,
580 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 581 GvCV_set(sv, NULL);
fc2b2dca 582 SvREFCNT_dec_NN(obj);
57ef47cc 583 }
fc2b2dca 584 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
585}
586
68b590d9 587/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
588 * called by sv_clean_objs() for each live GV */
589
590static void
591do_clean_named_io_objs(pTHX_ SV *const sv)
592{
e4487e9b
DM
593 SV *obj;
594 assert(SvTYPE(sv) == SVt_PVGV);
595 assert(isGV_with_GP(sv));
68b590d9 596 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
597 return;
598
599 SvREFCNT_inc(sv);
57ef47cc
DM
600 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
601 DEBUG_D((PerlIO_printf(Perl_debug_log,
602 "Cleaning named glob IO object:\n "), sv_dump(obj)));
603 GvIOp(sv) = NULL;
fc2b2dca 604 SvREFCNT_dec_NN(obj);
645c22ef 605 }
fc2b2dca 606 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 607}
645c22ef 608
4155e4fe
FC
609/* Void wrapper to pass to visit() */
610static void
611do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
612 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
613 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
614 return;
615 (void)curse(sv, 0);
616}
617
645c22ef
DM
618/*
619=for apidoc sv_clean_objs
620
fde67290 621Attempt to destroy all objects not yet freed.
645c22ef
DM
622
623=cut
624*/
625
4561caa4 626void
864dbfa3 627Perl_sv_clean_objs(pTHX)
4561caa4 628{
68b590d9 629 GV *olddef, *olderr;
3280af22 630 PL_in_clean_objs = TRUE;
055972dc 631 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
632 /* Some barnacles may yet remain, clinging to typeglobs.
633 * Run the non-IO destructors first: they may want to output
634 * error messages, close files etc */
d011219a 635 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 636 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
637 /* And if there are some very tenacious barnacles clinging to arrays,
638 closures, or what have you.... */
639 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
640 olddef = PL_defoutgv;
641 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
642 if (olddef && isGV_with_GP(olddef))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
644 olderr = PL_stderrgv;
645 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
646 if (olderr && isGV_with_GP(olderr))
647 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
648 SvREFCNT_dec(olddef);
3280af22 649 PL_in_clean_objs = FALSE;
4561caa4
CS
650}
651
645c22ef
DM
652/* called by sv_clean_all() for each live SV */
653
654static void
de37a194 655do_clean_all(pTHX_ SV *const sv)
645c22ef 656{
daba3364 657 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 658 /* don't clean pid table and strtab */
d17ea597 659 return;
cddfcddc 660 }
645c22ef
DM
661 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
662 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 663 SvREFCNT_dec_NN(sv);
645c22ef
DM
664}
665
666/*
667=for apidoc sv_clean_all
668
669Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 670cleanup. This function may have to be called multiple times to free
ff276b08 671SVs which are in complex self-referential hierarchies.
645c22ef
DM
672
673=cut
674*/
675
5226ed68 676I32
864dbfa3 677Perl_sv_clean_all(pTHX)
8990e307 678{
5226ed68 679 I32 cleaned;
3280af22 680 PL_in_clean_all = TRUE;
055972dc 681 cleaned = visit(do_clean_all, 0,0);
5226ed68 682 return cleaned;
8990e307 683}
463ee0b2 684
5e258f8c
JC
685/*
686 ARENASETS: a meta-arena implementation which separates arena-info
687 into struct arena_set, which contains an array of struct
688 arena_descs, each holding info for a single arena. By separating
689 the meta-info from the arena, we recover the 1st slot, formerly
690 borrowed for list management. The arena_set is about the size of an
39244528 691 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
692
693 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
694 memory in the last arena-set (1/2 on average). In trade, we get
695 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 696 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
697 small arenas for large, rare body types, by changing array* fields
698 in body_details_by_type[] below.
5e258f8c 699*/
5e258f8c 700struct arena_desc {
398c677b
NC
701 char *arena; /* the raw storage, allocated aligned */
702 size_t size; /* its size ~4k typ */
e5973ed5 703 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
704};
705
e6148039
NC
706struct arena_set;
707
708/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 709 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
710 therefore likely to be 1 aligned memory page. */
711
712#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
713 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
714
715struct arena_set {
716 struct arena_set* next;
0a848332
NC
717 unsigned int set_size; /* ie ARENAS_PER_SET */
718 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
719 struct arena_desc set[ARENAS_PER_SET];
720};
721
645c22ef
DM
722/*
723=for apidoc sv_free_arenas
724
fde67290 725Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
726heads and bodies within the arenas must already have been freed.
727
728=cut
7fefc6c1 729
645c22ef 730*/
4633a7c4 731void
864dbfa3 732Perl_sv_free_arenas(pTHX)
4633a7c4
LW
733{
734 SV* sva;
735 SV* svanext;
0a848332 736 unsigned int i;
4633a7c4
LW
737
738 /* Free arenas here, but be careful about fake ones. (We assume
739 contiguity of the fake ones with the corresponding real ones.) */
740
3280af22 741 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 742 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 743 while (svanext && SvFAKE(svanext))
daba3364 744 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
745
746 if (!SvFAKE(sva))
1df70142 747 Safefree(sva);
4633a7c4 748 }
93e68bfb 749
5e258f8c 750 {
0a848332
NC
751 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
752
753 while (aroot) {
754 struct arena_set *current = aroot;
755 i = aroot->curr;
756 while (i--) {
5e258f8c
JC
757 assert(aroot->set[i].arena);
758 Safefree(aroot->set[i].arena);
759 }
0a848332
NC
760 aroot = aroot->next;
761 Safefree(current);
5e258f8c
JC
762 }
763 }
dc8220bf 764 PL_body_arenas = 0;
fdda85ca 765
0a848332
NC
766 i = PERL_ARENA_ROOTS_SIZE;
767 while (i--)
93e68bfb 768 PL_body_roots[i] = 0;
93e68bfb 769
3280af22
NIS
770 PL_sv_arenaroot = 0;
771 PL_sv_root = 0;
4633a7c4
LW
772}
773
bd81e77b
NC
774/*
775 Here are mid-level routines that manage the allocation of bodies out
776 of the various arenas. There are 5 kinds of arenas:
29489e7c 777
bd81e77b
NC
778 1. SV-head arenas, which are discussed and handled above
779 2. regular body arenas
780 3. arenas for reduced-size bodies
781 4. Hash-Entry arenas
29489e7c 782
bd81e77b
NC
783 Arena types 2 & 3 are chained by body-type off an array of
784 arena-root pointers, which is indexed by svtype. Some of the
785 larger/less used body types are malloced singly, since a large
786 unused block of them is wasteful. Also, several svtypes dont have
787 bodies; the data fits into the sv-head itself. The arena-root
788 pointer thus has a few unused root-pointers (which may be hijacked
789 later for arena types 4,5)
29489e7c 790
bd81e77b
NC
791 3 differs from 2 as an optimization; some body types have several
792 unused fields in the front of the structure (which are kept in-place
793 for consistency). These bodies can be allocated in smaller chunks,
794 because the leading fields arent accessed. Pointers to such bodies
795 are decremented to point at the unused 'ghost' memory, knowing that
796 the pointers are used with offsets to the real memory.
29489e7c 797
d2a0f284
JC
798
799=head1 SV-Body Allocation
800
f554dfc5
MH
801=cut
802
d2a0f284
JC
803Allocation of SV-bodies is similar to SV-heads, differing as follows;
804the allocation mechanism is used for many body types, so is somewhat
805more complicated, it uses arena-sets, and has no need for still-live
806SV detection.
807
808At the outermost level, (new|del)_X*V macros return bodies of the
809appropriate type. These macros call either (new|del)_body_type or
810(new|del)_body_allocated macro pairs, depending on specifics of the
811type. Most body types use the former pair, the latter pair is used to
812allocate body types with "ghost fields".
813
814"ghost fields" are fields that are unused in certain types, and
69ba284b 815consequently don't need to actually exist. They are declared because
d2a0f284
JC
816they're part of a "base type", which allows use of functions as
817methods. The simplest examples are AVs and HVs, 2 aggregate types
818which don't use the fields which support SCALAR semantics.
819
69ba284b 820For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
821chunks, we thus avoid wasted memory for those unaccessed members.
822When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 823size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
824structure. (But things will all go boom if you write to the part that
825is "not there", because you'll be overwriting the last members of the
826preceding structure in memory.)
827
69ba284b 828We calculate the correction using the STRUCT_OFFSET macro on the first
a05ea1cf 829member present. If the allocated structure is smaller (no initial NV
69ba284b
NC
830actually allocated) then the net effect is to subtract the size of the NV
831from the pointer, to return a new pointer as if an initial NV were actually
a05ea1cf 832allocated. (We were using structures named *_allocated for this, but
69ba284b
NC
833this turned out to be a subtle bug, because a structure without an NV
834could have a lower alignment constraint, but the compiler is allowed to
835optimised accesses based on the alignment constraint of the actual pointer
836to the full structure, for example, using a single 64 bit load instruction
837because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284 838
a05ea1cf 839This is the same trick as was used for NV and IV bodies. Ironically it
d2a0f284 840doesn't need to be used for NV bodies any more, because NV is now at
a05ea1cf 841the start of the structure. IV bodies don't need it either, because
d2a0f284
JC
842they are no longer allocated.
843
844In turn, the new_body_* allocators call S_new_body(), which invokes
845new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 846linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
847necessary to refresh an empty list. Then the lock is released, and
848the body is returned.
849
99816f8d 850Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
851bodies, which it strings into a linked list. It looks up arena-size
852and body-size from the body_details table described below, thus
853supporting the multiple body-types.
854
855If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
856the (new|del)_X*V macros are mapped directly to malloc/free.
857
d2a0f284
JC
858For each sv-type, struct body_details bodies_by_type[] carries
859parameters which control these aspects of SV handling:
860
861Arena_size determines whether arenas are used for this body type, and if
862so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
863zero, forcing individual mallocs and frees.
864
865Body_size determines how big a body is, and therefore how many fit into
866each arena. Offset carries the body-pointer adjustment needed for
69ba284b 867"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
868
869But its main purpose is to parameterize info needed in
870Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 871vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
872are used for this, except for arena_size.
873
874For the sv-types that have no bodies, arenas are not used, so those
875PL_body_roots[sv_type] are unused, and can be overloaded. In
876something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 877PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 878bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 879available in hv.c.
d2a0f284 880
29489e7c
DM
881*/
882
bd81e77b 883struct body_details {
0fb58b32 884 U8 body_size; /* Size to allocate */
10666ae3 885 U8 copy; /* Size of structure to copy (may be shorter) */
0fb58b32 886 U8 offset;
10666ae3
NC
887 unsigned int type : 4; /* We have space for a sanity check. */
888 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
889 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
890 unsigned int arena : 1; /* Allocated from an arena */
891 size_t arena_size; /* Size of arena to allocate */
bd81e77b 892};
29489e7c 893
bd81e77b
NC
894#define HADNV FALSE
895#define NONV TRUE
29489e7c 896
d2a0f284 897
bd81e77b
NC
898#ifdef PURIFY
899/* With -DPURFIY we allocate everything directly, and don't use arenas.
900 This seems a rather elegant way to simplify some of the code below. */
901#define HASARENA FALSE
902#else
903#define HASARENA TRUE
904#endif
905#define NOARENA FALSE
29489e7c 906
d2a0f284
JC
907/* Size the arenas to exactly fit a given number of bodies. A count
908 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
909 simplifying the default. If count > 0, the arena is sized to fit
910 only that many bodies, allowing arenas to be used for large, rare
911 bodies (XPVFM, XPVIO) without undue waste. The arena size is
912 limited by PERL_ARENA_SIZE, so we can safely oversize the
913 declarations.
914 */
95db5f15
MB
915#define FIT_ARENA0(body_size) \
916 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
917#define FIT_ARENAn(count,body_size) \
918 ( count * body_size <= PERL_ARENA_SIZE) \
919 ? count * body_size \
920 : FIT_ARENA0 (body_size)
921#define FIT_ARENA(count,body_size) \
922 count \
923 ? FIT_ARENAn (count, body_size) \
924 : FIT_ARENA0 (body_size)
d2a0f284 925
bd81e77b
NC
926/* Calculate the length to copy. Specifically work out the length less any
927 final padding the compiler needed to add. See the comment in sv_upgrade
928 for why copying the padding proved to be a bug. */
29489e7c 929
bd81e77b
NC
930#define copy_length(type, last_member) \
931 STRUCT_OFFSET(type, last_member) \
daba3364 932 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 933
bd81e77b 934static const struct body_details bodies_by_type[] = {
829cd18a
NC
935 /* HEs use this offset for their arena. */
936 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 937
db93c0c4
NC
938 /* IVs are in the head, so the allocation size is 0. */
939 { 0,
d2a0f284 940 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 941 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 942 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
943 },
944
6e128786
NC
945 { sizeof(NV), sizeof(NV),
946 STRUCT_OFFSET(XPVNV, xnv_u),
947 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
d2a0f284 948
bc337e5c 949 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
950 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
951 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 952 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 953 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 954
d361b004
KW
955 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
956 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
957 + STRUCT_OFFSET(XPV, xpv_cur),
958 SVt_INVLIST, TRUE, NONV, HASARENA,
959 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 960
bc337e5c 961 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
962 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
963 + STRUCT_OFFSET(XPV, xpv_cur),
964 SVt_PVIV, FALSE, NONV, HASARENA,
965 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 966
bc337e5c 967 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
968 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
969 + STRUCT_OFFSET(XPV, xpv_cur),
970 SVt_PVNV, FALSE, HADNV, HASARENA,
971 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 972
6e128786 973 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 974 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 975
601dfd0a
NC
976 { sizeof(regexp),
977 sizeof(regexp),
978 0,
ecff11eb 979 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 980 FIT_ARENA(0, sizeof(regexp))
5c35adbb 981 },
4df7f6af 982
10666ae3 983 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
984 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
985
10666ae3 986 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
987 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
988
601dfd0a 989 { sizeof(XPVAV),
4f7003f5 990 copy_length(XPVAV, xav_alloc),
601dfd0a 991 0,
69ba284b 992 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 993 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 994
601dfd0a 995 { sizeof(XPVHV),
359164a0 996 copy_length(XPVHV, xhv_max),
601dfd0a 997 0,
69ba284b 998 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 999 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 1000
601dfd0a
NC
1001 { sizeof(XPVCV),
1002 sizeof(XPVCV),
1003 0,
69ba284b 1004 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 1005 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 1006
601dfd0a
NC
1007 { sizeof(XPVFM),
1008 sizeof(XPVFM),
1009 0,
69ba284b 1010 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 1011 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 1012
601dfd0a
NC
1013 { sizeof(XPVIO),
1014 sizeof(XPVIO),
1015 0,
b6f60916 1016 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 1017 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 1018};
29489e7c 1019
bd81e77b 1020#define new_body_allocated(sv_type) \
d2a0f284 1021 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1022 - bodies_by_type[sv_type].offset)
29489e7c 1023
26359cfa
NC
1024/* return a thing to the free list */
1025
1026#define del_body(thing, root) \
1027 STMT_START { \
1028 void ** const thing_copy = (void **)thing; \
1029 *thing_copy = *root; \
1030 *root = (void*)thing_copy; \
1031 } STMT_END
29489e7c 1032
bd81e77b 1033#ifdef PURIFY
29489e7c 1034
beeec492
NC
1035#define new_XNV() safemalloc(sizeof(XPVNV))
1036#define new_XPVNV() safemalloc(sizeof(XPVNV))
1037#define new_XPVMG() safemalloc(sizeof(XPVMG))
29489e7c 1038
beeec492 1039#define del_XPVGV(p) safefree(p)
29489e7c 1040
bd81e77b 1041#else /* !PURIFY */
29489e7c 1042
65ac1738 1043#define new_XNV() new_body_allocated(SVt_NV)
65ac1738 1044#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1045#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1046
26359cfa
NC
1047#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1048 &PL_body_roots[SVt_PVGV])
1d7c1841 1049
bd81e77b 1050#endif /* PURIFY */
93e68bfb 1051
bd81e77b 1052/* no arena for you! */
93e68bfb 1053
bd81e77b 1054#define new_NOARENA(details) \
beeec492 1055 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1056#define new_NOARENAZ(details) \
beeec492 1057 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1058
1e30fcd5
NC
1059void *
1060Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1061 const size_t arena_size)
d2a0f284 1062{
d2a0f284 1063 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1064 struct arena_desc *adesc;
1065 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1066 unsigned int curr;
d2a0f284
JC
1067 char *start;
1068 const char *end;
02982131 1069 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
20b7effb
JH
1070#if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1071 dVAR;
1072#endif
0b2d3faa 1073#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1074 static bool done_sanity_check;
1075
0b2d3faa
JH
1076 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1077 * variables like done_sanity_check. */
10666ae3 1078 if (!done_sanity_check) {
ea471437 1079 unsigned int i = SVt_LAST;
10666ae3
NC
1080
1081 done_sanity_check = TRUE;
1082
1083 while (i--)
1084 assert (bodies_by_type[i].type == i);
1085 }
1086#endif
1087
02982131 1088 assert(arena_size);
23e9d66c 1089
99816f8d
NC
1090 /* may need new arena-set to hold new arena */
1091 if (!aroot || aroot->curr >= aroot->set_size) {
1092 struct arena_set *newroot;
1093 Newxz(newroot, 1, struct arena_set);
1094 newroot->set_size = ARENAS_PER_SET;
1095 newroot->next = aroot;
1096 aroot = newroot;
1097 PL_body_arenas = (void *) newroot;
1098 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1099 }
1100
1101 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1102 curr = aroot->curr++;
1103 adesc = &(aroot->set[curr]);
1104 assert(!adesc->arena);
1105
1106 Newx(adesc->arena, good_arena_size, char);
1107 adesc->size = good_arena_size;
1108 adesc->utype = sv_type;
1109 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1110 curr, (void*)adesc->arena, (UV)good_arena_size));
1111
1112 start = (char *) adesc->arena;
d2a0f284 1113
29657bb6
NC
1114 /* Get the address of the byte after the end of the last body we can fit.
1115 Remember, this is integer division: */
02982131 1116 end = start + good_arena_size / body_size * body_size;
d2a0f284 1117
486ec47a 1118 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1119#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1120 DEBUG_m(PerlIO_printf(Perl_debug_log,
1121 "arena %p end %p arena-size %d (from %d) type %d "
1122 "size %d ct %d\n",
02982131
NC
1123 (void*)start, (void*)end, (int)good_arena_size,
1124 (int)arena_size, sv_type, (int)body_size,
1125 (int)good_arena_size / (int)body_size));
d8fca402 1126#else
d2a0f284
JC
1127 DEBUG_m(PerlIO_printf(Perl_debug_log,
1128 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1129 (void*)start, (void*)end,
02982131
NC
1130 (int)arena_size, sv_type, (int)body_size,
1131 (int)good_arena_size / (int)body_size));
d8fca402 1132#endif
d2a0f284
JC
1133 *root = (void *)start;
1134
29657bb6
NC
1135 while (1) {
1136 /* Where the next body would start: */
d2a0f284 1137 char * const next = start + body_size;
29657bb6
NC
1138
1139 if (next >= end) {
1140 /* This is the last body: */
1141 assert(next == end);
1142
1143 *(void **)start = 0;
1144 return *root;
1145 }
1146
d2a0f284
JC
1147 *(void**) start = (void *)next;
1148 start = next;
1149 }
d2a0f284
JC
1150}
1151
1152/* grab a new thing from the free list, allocating more if necessary.
1153 The inline version is used for speed in hot routines, and the
1154 function using it serves the rest (unless PURIFY).
1155*/
1156#define new_body_inline(xpv, sv_type) \
1157 STMT_START { \
1158 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1159 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1160 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1161 bodies_by_type[sv_type].body_size,\
1162 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1163 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1164 } STMT_END
1165
1166#ifndef PURIFY
1167
1168STATIC void *
de37a194 1169S_new_body(pTHX_ const svtype sv_type)
d2a0f284 1170{
d2a0f284
JC
1171 void *xpv;
1172 new_body_inline(xpv, sv_type);
1173 return xpv;
1174}
1175
1176#endif
93e68bfb 1177
238b27b3
NC
1178static const struct body_details fake_rv =
1179 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1180
bd81e77b
NC
1181/*
1182=for apidoc sv_upgrade
93e68bfb 1183
bd81e77b
NC
1184Upgrade an SV to a more complex form. Generally adds a new body type to the
1185SV, then copies across as much information as possible from the old body.
9521ca61
FC
1186It croaks if the SV is already in a more complex form than requested. You
1187generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1188before calling C<sv_upgrade>, and hence does not croak. See also
1189C<svtype>.
93e68bfb 1190
bd81e77b 1191=cut
93e68bfb 1192*/
93e68bfb 1193
bd81e77b 1194void
5aaab254 1195Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1196{
bd81e77b
NC
1197 void* old_body;
1198 void* new_body;
42d0e0b7 1199 const svtype old_type = SvTYPE(sv);
d2a0f284 1200 const struct body_details *new_type_details;
238b27b3 1201 const struct body_details *old_type_details
bd81e77b 1202 = bodies_by_type + old_type;
4df7f6af 1203 SV *referant = NULL;
cac9b346 1204
7918f24d
NC
1205 PERL_ARGS_ASSERT_SV_UPGRADE;
1206
1776cbe8
NC
1207 if (old_type == new_type)
1208 return;
1209
1210 /* This clause was purposefully added ahead of the early return above to
1211 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1212 inference by Nick I-S that it would fix other troublesome cases. See
1213 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1214
1215 Given that shared hash key scalars are no longer PVIV, but PV, there is
1216 no longer need to unshare so as to free up the IVX slot for its proper
1217 purpose. So it's safe to move the early return earlier. */
1218
093085a8 1219 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1220 sv_force_normal_flags(sv, 0);
1221 }
cac9b346 1222
bd81e77b 1223 old_body = SvANY(sv);
de042e1d 1224
bd81e77b
NC
1225 /* Copying structures onto other structures that have been neatly zeroed
1226 has a subtle gotcha. Consider XPVMG
cac9b346 1227
bd81e77b
NC
1228 +------+------+------+------+------+-------+-------+
1229 | NV | CUR | LEN | IV | MAGIC | STASH |
1230 +------+------+------+------+------+-------+-------+
1231 0 4 8 12 16 20 24 28
645c22ef 1232
bd81e77b
NC
1233 where NVs are aligned to 8 bytes, so that sizeof that structure is
1234 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1235
bd81e77b
NC
1236 +------+------+------+------+------+-------+-------+------+
1237 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1238 +------+------+------+------+------+-------+-------+------+
1239 0 4 8 12 16 20 24 28 32
08742458 1240
bd81e77b 1241 so what happens if you allocate memory for this structure:
30f9da9e 1242
bd81e77b
NC
1243 +------+------+------+------+------+-------+-------+------+------+...
1244 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1245 +------+------+------+------+------+-------+-------+------+------+...
1246 0 4 8 12 16 20 24 28 32 36
bfc44f79 1247
bd81e77b
NC
1248 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1249 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1250 started out as zero once, but it's quite possible that it isn't. So now,
1251 rather than a nicely zeroed GP, you have it pointing somewhere random.
1252 Bugs ensue.
bfc44f79 1253
bd81e77b
NC
1254 (In fact, GP ends up pointing at a previous GP structure, because the
1255 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1256 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1257 this happens to be moot because XPVGV has been re-ordered, with GP
1258 no longer after STASH)
30f9da9e 1259
bd81e77b
NC
1260 So we are careful and work out the size of used parts of all the
1261 structures. */
bfc44f79 1262
bd81e77b
NC
1263 switch (old_type) {
1264 case SVt_NULL:
1265 break;
1266 case SVt_IV:
4df7f6af
NC
1267 if (SvROK(sv)) {
1268 referant = SvRV(sv);
238b27b3
NC
1269 old_type_details = &fake_rv;
1270 if (new_type == SVt_NV)
1271 new_type = SVt_PVNV;
4df7f6af
NC
1272 } else {
1273 if (new_type < SVt_PVIV) {
1274 new_type = (new_type == SVt_NV)
1275 ? SVt_PVNV : SVt_PVIV;
1276 }
bd81e77b
NC
1277 }
1278 break;
1279 case SVt_NV:
1280 if (new_type < SVt_PVNV) {
1281 new_type = SVt_PVNV;
bd81e77b
NC
1282 }
1283 break;
bd81e77b
NC
1284 case SVt_PV:
1285 assert(new_type > SVt_PV);
1286 assert(SVt_IV < SVt_PV);
1287 assert(SVt_NV < SVt_PV);
1288 break;
1289 case SVt_PVIV:
1290 break;
1291 case SVt_PVNV:
1292 break;
1293 case SVt_PVMG:
1294 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1295 there's no way that it can be safely upgraded, because perl.c
1296 expects to Safefree(SvANY(PL_mess_sv)) */
1297 assert(sv != PL_mess_sv);
1298 /* This flag bit is used to mean other things in other scalar types.
1299 Given that it only has meaning inside the pad, it shouldn't be set
1300 on anything that can get upgraded. */
00b1698f 1301 assert(!SvPAD_TYPED(sv));
bd81e77b
NC
1302 break;
1303 default:
2439e033 1304 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1305 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1306 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1307 }
3376de98 1308
2439e033 1309 if (UNLIKELY(old_type > new_type))
3376de98
NC
1310 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1311 (int)old_type, (int)new_type);
1312
2fa1109b 1313 new_type_details = bodies_by_type + new_type;
645c22ef 1314
bd81e77b
NC
1315 SvFLAGS(sv) &= ~SVTYPEMASK;
1316 SvFLAGS(sv) |= new_type;
932e9ff9 1317
ab4416c0
NC
1318 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1319 the return statements above will have triggered. */
1320 assert (new_type != SVt_NULL);
bd81e77b 1321 switch (new_type) {
bd81e77b
NC
1322 case SVt_IV:
1323 assert(old_type == SVt_NULL);
1324 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1325 SvIV_set(sv, 0);
1326 return;
1327 case SVt_NV:
1328 assert(old_type == SVt_NULL);
1329 SvANY(sv) = new_XNV();
1330 SvNV_set(sv, 0);
1331 return;
bd81e77b 1332 case SVt_PVHV:
bd81e77b 1333 case SVt_PVAV:
d2a0f284 1334 assert(new_type_details->body_size);
c1ae03ae
NC
1335
1336#ifndef PURIFY
1337 assert(new_type_details->arena);
d2a0f284 1338 assert(new_type_details->arena_size);
c1ae03ae 1339 /* This points to the start of the allocated area. */
d2a0f284
JC
1340 new_body_inline(new_body, new_type);
1341 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1342 new_body = ((char *)new_body) - new_type_details->offset;
1343#else
1344 /* We always allocated the full length item with PURIFY. To do this
1345 we fake things so that arena is false for all 16 types.. */
1346 new_body = new_NOARENAZ(new_type_details);
1347#endif
1348 SvANY(sv) = new_body;
1349 if (new_type == SVt_PVAV) {
1350 AvMAX(sv) = -1;
1351 AvFILLp(sv) = -1;
1352 AvREAL_only(sv);
64484faa 1353 if (old_type_details->body_size) {
ac572bf4
NC
1354 AvALLOC(sv) = 0;
1355 } else {
1356 /* It will have been zeroed when the new body was allocated.
1357 Lets not write to it, in case it confuses a write-back
1358 cache. */
1359 }
78ac7dd9
NC
1360 } else {
1361 assert(!SvOK(sv));
1362 SvOK_off(sv);
1363#ifndef NODEFAULT_SHAREKEYS
1364 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1365#endif
586fc6a3
SM
1366 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1367 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1368 }
aeb18a1e 1369
bd81e77b
NC
1370 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1371 The target created by newSVrv also is, and it can have magic.
1372 However, it never has SvPVX set.
1373 */
4df7f6af
NC
1374 if (old_type == SVt_IV) {
1375 assert(!SvROK(sv));
1376 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1377 assert(SvPVX_const(sv) == 0);
1378 }
aeb18a1e 1379
bd81e77b 1380 if (old_type >= SVt_PVMG) {
e736a858 1381 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1382 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1383 } else {
1384 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1385 }
1386 break;
93e68bfb 1387
bd81e77b
NC
1388 case SVt_PVIV:
1389 /* XXX Is this still needed? Was it ever needed? Surely as there is
1390 no route from NV to PVIV, NOK can never be true */
1391 assert(!SvNOKp(sv));
1392 assert(!SvNOK(sv));
1393 case SVt_PVIO:
1394 case SVt_PVFM:
bd81e77b
NC
1395 case SVt_PVGV:
1396 case SVt_PVCV:
1397 case SVt_PVLV:
d361b004 1398 case SVt_INVLIST:
12c45b25 1399 case SVt_REGEXP:
bd81e77b
NC
1400 case SVt_PVMG:
1401 case SVt_PVNV:
1402 case SVt_PV:
93e68bfb 1403
d2a0f284 1404 assert(new_type_details->body_size);
bd81e77b
NC
1405 /* We always allocated the full length item with PURIFY. To do this
1406 we fake things so that arena is false for all 16 types.. */
1407 if(new_type_details->arena) {
1408 /* This points to the start of the allocated area. */
d2a0f284
JC
1409 new_body_inline(new_body, new_type);
1410 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1411 new_body = ((char *)new_body) - new_type_details->offset;
1412 } else {
1413 new_body = new_NOARENAZ(new_type_details);
1414 }
1415 SvANY(sv) = new_body;
5e2fc214 1416
bd81e77b 1417 if (old_type_details->copy) {
f9ba3d20
NC
1418 /* There is now the potential for an upgrade from something without
1419 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1420 int offset = old_type_details->offset;
1421 int length = old_type_details->copy;
1422
1423 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1424 const int difference
f9ba3d20
NC
1425 = new_type_details->offset - old_type_details->offset;
1426 offset += difference;
1427 length -= difference;
1428 }
1429 assert (length >= 0);
1430
1431 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1432 char);
bd81e77b
NC
1433 }
1434
1435#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1436 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1437 * correct 0.0 for us. Otherwise, if the old body didn't have an
1438 * NV slot, but the new one does, then we need to initialise the
1439 * freshly created NV slot with whatever the correct bit pattern is
1440 * for 0.0 */
e22a937e
NC
1441 if (old_type_details->zero_nv && !new_type_details->zero_nv
1442 && !isGV_with_GP(sv))
bd81e77b 1443 SvNV_set(sv, 0);
82048762 1444#endif
5e2fc214 1445
2439e033 1446 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1447 IO * const io = MUTABLE_IO(sv);
d963bf01 1448 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1449
1450 SvOBJECT_on(io);
1451 /* Clear the stashcache because a new IO could overrule a package
1452 name */
103f5a36 1453 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1454 hv_clear(PL_stashcache);
1455
85dca89a 1456 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1457 IoPAGE_LEN(sv) = 60;
85dca89a 1458 }
2439e033 1459 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1460 sv->sv_u.svu_rx = (regexp *)new_body;
1461 else if (old_type < SVt_PV) {
4df7f6af
NC
1462 /* referant will be NULL unless the old type was SVt_IV emulating
1463 SVt_RV */
1464 sv->sv_u.svu_rv = referant;
1465 }
bd81e77b
NC
1466 break;
1467 default:
afd78fd5
JH
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
bd81e77b 1470 }
73171d91 1471
db93c0c4 1472 if (old_type > SVt_IV) {
bd81e77b 1473#ifdef PURIFY
beeec492 1474 safefree(old_body);
bd81e77b 1475#else
bc786448
GG
1476 /* Note that there is an assumption that all bodies of types that
1477 can be upgraded came from arenas. Only the more complex non-
1478 upgradable types are allowed to be directly malloc()ed. */
1479 assert(old_type_details->arena);
bd81e77b
NC
1480 del_body((void*)((char*)old_body + old_type_details->offset),
1481 &PL_body_roots[old_type]);
1482#endif
1483 }
1484}
73171d91 1485
bd81e77b
NC
1486/*
1487=for apidoc sv_backoff
73171d91 1488
fde67290 1489Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1490wrapper instead.
73171d91 1491
bd81e77b 1492=cut
73171d91
NC
1493*/
1494
bd81e77b 1495int
ddeaf645 1496Perl_sv_backoff(SV *const sv)
bd81e77b 1497{
69240efd 1498 STRLEN delta;
7a4bba22 1499 const char * const s = SvPVX_const(sv);
7918f24d
NC
1500
1501 PERL_ARGS_ASSERT_SV_BACKOFF;
7918f24d 1502
bd81e77b
NC
1503 assert(SvOOK(sv));
1504 assert(SvTYPE(sv) != SVt_PVHV);
1505 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1506
69240efd
NC
1507 SvOOK_offset(sv, delta);
1508
7a4bba22
NC
1509 SvLEN_set(sv, SvLEN(sv) + delta);
1510 SvPV_set(sv, SvPVX(sv) - delta);
1511 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
bd81e77b
NC
1512 SvFLAGS(sv) &= ~SVf_OOK;
1513 return 0;
1514}
73171d91 1515
bd81e77b
NC
1516/*
1517=for apidoc sv_grow
73171d91 1518
bd81e77b
NC
1519Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1520upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1521Use the C<SvGROW> wrapper instead.
93e68bfb 1522
bd81e77b
NC
1523=cut
1524*/
93e68bfb 1525
e0060e30
FC
1526static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1527
bd81e77b 1528char *
5aaab254 1529Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1530{
eb578fdb 1531 char *s;
93e68bfb 1532
7918f24d
NC
1533 PERL_ARGS_ASSERT_SV_GROW;
1534
bd81e77b
NC
1535 if (SvROK(sv))
1536 sv_unref(sv);
1537 if (SvTYPE(sv) < SVt_PV) {
1538 sv_upgrade(sv, SVt_PV);
1539 s = SvPVX_mutable(sv);
1540 }
1541 else if (SvOOK(sv)) { /* pv is offset? */
1542 sv_backoff(sv);
1543 s = SvPVX_mutable(sv);
1544 if (newlen > SvLEN(sv))
1545 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
bd81e77b
NC
1546 }
1547 else
db2c6cb3 1548 {
e0060e30 1549 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1550 s = SvPVX_mutable(sv);
db2c6cb3 1551 }
aeb18a1e 1552
cbcb2a16
DM
1553#ifdef PERL_NEW_COPY_ON_WRITE
1554 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
3c239bea 1555 * to store the COW count. So in general, allocate one more byte than
cbcb2a16
DM
1556 * asked for, to make it likely this byte is always spare: and thus
1557 * make more strings COW-able.
1558 * If the new size is a big power of two, don't bother: we assume the
1559 * caller wanted a nice 2^N sized block and will be annoyed at getting
1560 * 2^N+1 */
1561 if (newlen & 0xff)
1562 newlen++;
1563#endif
1564
ce861ea7
YO
1565#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1566#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1567#endif
1568
bd81e77b 1569 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559 1570 STRLEN minlen = SvCUR(sv);
3c239bea 1571 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
f1200559
WH
1572 if (newlen < minlen)
1573 newlen = minlen;
ce861ea7 1574#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
7c641603
KW
1575
1576 /* Don't round up on the first allocation, as odds are pretty good that
1577 * the initial request is accurate as to what is really needed */
ce861ea7 1578 if (SvLEN(sv)) {
e8c6a474 1579 newlen = PERL_STRLEN_ROUNDUP(newlen);
ce861ea7 1580 }
bd81e77b 1581#endif
98653f18 1582 if (SvLEN(sv) && s) {
10edeb5d 1583 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1584 }
1585 else {
10edeb5d 1586 s = (char*)safemalloc(newlen);
bd81e77b
NC
1587 if (SvPVX_const(sv) && SvCUR(sv)) {
1588 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1589 }
1590 }
1591 SvPV_set(sv, s);
ce861ea7 1592#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
98653f18
NC
1593 /* Do this here, do it once, do it right, and then we will never get
1594 called back into sv_grow() unless there really is some growing
1595 needed. */
ca7c1a29 1596 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1597#else
bd81e77b 1598 SvLEN_set(sv, newlen);
98653f18 1599#endif
bd81e77b
NC
1600 }
1601 return s;
1602}
aeb18a1e 1603
bd81e77b
NC
1604/*
1605=for apidoc sv_setiv
932e9ff9 1606
bd81e77b
NC
1607Copies an integer into the given SV, upgrading first if necessary.
1608Does not handle 'set' magic. See also C<sv_setiv_mg>.
463ee0b2 1609
bd81e77b
NC
1610=cut
1611*/
463ee0b2 1612
bd81e77b 1613void
5aaab254 1614Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1615{
7918f24d
NC
1616 PERL_ARGS_ASSERT_SV_SETIV;
1617
bd81e77b
NC
1618 SV_CHECK_THINKFIRST_COW_DROP(sv);
1619 switch (SvTYPE(sv)) {
1620 case SVt_NULL:
bd81e77b 1621 case SVt_NV:
3376de98 1622 sv_upgrade(sv, SVt_IV);
bd81e77b 1623 break;
bd81e77b
NC
1624 case SVt_PV:
1625 sv_upgrade(sv, SVt_PVIV);
1626 break;
463ee0b2 1627
bd81e77b 1628 case SVt_PVGV:
6e592b3a
BM
1629 if (!isGV_with_GP(sv))
1630 break;
bd81e77b
NC
1631 case SVt_PVAV:
1632 case SVt_PVHV:
1633 case SVt_PVCV:
1634 case SVt_PVFM:
1635 case SVt_PVIO:
22e74366 1636 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1637 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1638 OP_DESC(PL_op));
42d0e0b7 1639 default: NOOP;
bd81e77b
NC
1640 }
1641 (void)SvIOK_only(sv); /* validate number */
1642 SvIV_set(sv, i);
1643 SvTAINT(sv);
1644}
932e9ff9 1645
bd81e77b
NC
1646/*
1647=for apidoc sv_setiv_mg
d33b2eba 1648
bd81e77b 1649Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1650
bd81e77b
NC
1651=cut
1652*/
d33b2eba 1653
bd81e77b 1654void
5aaab254 1655Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1656{
7918f24d
NC
1657 PERL_ARGS_ASSERT_SV_SETIV_MG;
1658
bd81e77b
NC
1659 sv_setiv(sv,i);
1660 SvSETMAGIC(sv);
1661}
727879eb 1662
bd81e77b
NC
1663/*
1664=for apidoc sv_setuv
d33b2eba 1665
bd81e77b
NC
1666Copies an unsigned integer into the given SV, upgrading first if necessary.
1667Does not handle 'set' magic. See also C<sv_setuv_mg>.
9b94d1dd 1668
bd81e77b
NC
1669=cut
1670*/
d33b2eba 1671
bd81e77b 1672void
5aaab254 1673Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1674{
7918f24d
NC
1675 PERL_ARGS_ASSERT_SV_SETUV;
1676
013abb9b
NC
1677 /* With the if statement to ensure that integers are stored as IVs whenever
1678 possible:
bd81e77b 1679 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1680
bd81e77b
NC
1681 without
1682 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1683
013abb9b
NC
1684 If you wish to remove the following if statement, so that this routine
1685 (and its callers) always return UVs, please benchmark to see what the
1686 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1687 */
1688 if (u <= (UV)IV_MAX) {
1689 sv_setiv(sv, (IV)u);
1690 return;
1691 }
1692 sv_setiv(sv, 0);
1693 SvIsUV_on(sv);
1694 SvUV_set(sv, u);
1695}
d33b2eba 1696
bd81e77b
NC
1697/*
1698=for apidoc sv_setuv_mg
727879eb 1699
bd81e77b 1700Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1701
bd81e77b
NC
1702=cut
1703*/
5e2fc214 1704
bd81e77b 1705void
5aaab254 1706Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1707{
7918f24d
NC
1708 PERL_ARGS_ASSERT_SV_SETUV_MG;
1709
bd81e77b
NC
1710 sv_setuv(sv,u);
1711 SvSETMAGIC(sv);
1712}
5e2fc214 1713
954c1994 1714/*
bd81e77b 1715=for apidoc sv_setnv
954c1994 1716
bd81e77b
NC
1717Copies a double into the given SV, upgrading first if necessary.
1718Does not handle 'set' magic. See also C<sv_setnv_mg>.
954c1994
GS
1719
1720=cut
1721*/
1722
63f97190 1723void
5aaab254 1724Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1725{
7918f24d
NC
1726 PERL_ARGS_ASSERT_SV_SETNV;
1727
bd81e77b
NC
1728 SV_CHECK_THINKFIRST_COW_DROP(sv);
1729 switch (SvTYPE(sv)) {
79072805 1730 case SVt_NULL:
79072805 1731 case SVt_IV:
bd81e77b 1732 sv_upgrade(sv, SVt_NV);
79072805
LW
1733 break;
1734 case SVt_PV:
79072805 1735 case SVt_PVIV:
bd81e77b 1736 sv_upgrade(sv, SVt_PVNV);
79072805 1737 break;
bd4b1eb5 1738
bd4b1eb5 1739 case SVt_PVGV:
6e592b3a
BM
1740 if (!isGV_with_GP(sv))
1741 break;
bd81e77b
NC
1742 case SVt_PVAV:
1743 case SVt_PVHV:
79072805 1744 case SVt_PVCV:
bd81e77b
NC
1745 case SVt_PVFM:
1746 case SVt_PVIO:
22e74366 1747 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1748 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1749 OP_DESC(PL_op));
42d0e0b7 1750 default: NOOP;
2068cd4d 1751 }
bd81e77b
NC
1752 SvNV_set(sv, num);
1753 (void)SvNOK_only(sv); /* validate number */
1754 SvTAINT(sv);
79072805
LW
1755}
1756
645c22ef 1757/*
bd81e77b 1758=for apidoc sv_setnv_mg
645c22ef 1759
bd81e77b 1760Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1761
1762=cut
1763*/
1764
bd81e77b 1765void
5aaab254 1766Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1767{
7918f24d
NC
1768 PERL_ARGS_ASSERT_SV_SETNV_MG;
1769
bd81e77b
NC
1770 sv_setnv(sv,num);
1771 SvSETMAGIC(sv);
79072805
LW
1772}
1773
3f7602fa
TC
1774/* Return a cleaned-up, printable version of sv, for non-numeric, or
1775 * not incrementable warning display.
1776 * Originally part of S_not_a_number().
1777 * The return value may be != tmpbuf.
bd81e77b 1778 */
954c1994 1779
3f7602fa
TC
1780STATIC const char *
1781S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1782 const char *pv;
94463019 1783
3f7602fa 1784 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1785
94463019 1786 if (DO_UTF8(sv)) {
3f7602fa 1787 SV *dsv = newSVpvs_flags("", SVs_TEMP);
472394e4 1788 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
94463019
JH
1789 } else {
1790 char *d = tmpbuf;
3f7602fa 1791 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1792 /* each *s can expand to 4 chars + "...\0",
1793 i.e. need room for 8 chars */
ecdeb87c 1794
00b6aa41
AL
1795 const char *s = SvPVX_const(sv);
1796 const char * const end = s + SvCUR(sv);
1797 for ( ; s < end && d < limit; s++ ) {
94463019 1798 int ch = *s & 0xFF;
bd27cf70 1799 if (! isASCII(ch) && !isPRINT_LC(ch)) {
94463019
JH
1800 *d++ = 'M';
1801 *d++ = '-';
bd27cf70
KW
1802
1803 /* Map to ASCII "equivalent" of Latin1 */
1804 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
94463019
JH
1805 }
1806 if (ch == '\n') {
1807 *d++ = '\\';
1808 *d++ = 'n';
1809 }
1810 else if (ch == '\r') {
1811 *d++ = '\\';
1812 *d++ = 'r';
1813 }
1814 else if (ch == '\f') {
1815 *d++ = '\\';
1816 *d++ = 'f';
1817 }
1818 else if (ch == '\\') {
1819 *d++ = '\\';
1820 *d++ = '\\';
1821 }
1822 else if (ch == '\0') {
1823 *d++ = '\\';
1824 *d++ = '0';
1825 }
1826 else if (isPRINT_LC(ch))
1827 *d++ = ch;
1828 else {
1829 *d++ = '^';
1830 *d++ = toCTRL(ch);
1831 }
1832 }
1833 if (s < end) {
1834 *d++ = '.';
1835 *d++ = '.';
1836 *d++ = '.';
1837 }
1838 *d = '\0';
1839 pv = tmpbuf;
a0d0e21e 1840 }
a0d0e21e 1841
3f7602fa
TC
1842 return pv;
1843}
1844
1845/* Print an "isn't numeric" warning, using a cleaned-up,
1846 * printable version of the offending string
1847 */
1848
1849STATIC void
1850S_not_a_number(pTHX_ SV *const sv)
1851{
3f7602fa
TC
1852 char tmpbuf[64];
1853 const char *pv;
1854
1855 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1856
1857 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1858
533c011a 1859 if (PL_op)
9014280d 1860 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1861 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1862 "Argument \"%s\" isn't numeric in %s", pv,
1863 OP_DESC(PL_op));
a0d0e21e 1864 else
9014280d 1865 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1866 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1867 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1868}
1869
3f7602fa
TC
1870STATIC void
1871S_not_incrementable(pTHX_ SV *const sv) {
3f7602fa
TC
1872 char tmpbuf[64];
1873 const char *pv;
1874
1875 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1876
1877 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1878
1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1880 "Argument \"%s\" treated as 0 in increment (++)", pv);
1881}
1882
c2988b20
NC
1883/*
1884=for apidoc looks_like_number
1885
645c22ef
DM
1886Test if the content of an SV looks like a number (or is a number).
1887C<Inf> and C<Infinity> are treated as numbers (so will not issue a
f52e41ad
FC
1888non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1889ignored.
c2988b20
NC
1890
1891=cut
1892*/
1893
1894I32
aad570aa 1895Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1896{
eb578fdb 1897 const char *sbegin;
c2988b20
NC
1898 STRLEN len;
1899
7918f24d
NC
1900 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1901
f52e41ad
FC
1902 if (SvPOK(sv) || SvPOKp(sv)) {
1903 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1904 }
c2988b20 1905 else
e0ab1c0e 1906 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
c2988b20
NC
1907 return grok_number(sbegin, len, NULL);
1908}
25da4f38 1909
19f6321d
NC
1910STATIC bool
1911S_glob_2number(pTHX_ GV * const gv)
180488f8 1912{
7918f24d
NC
1913 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1914
675c862f
AL
1915 /* We know that all GVs stringify to something that is not-a-number,
1916 so no need to test that. */
1917 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1918 {
1919 SV *const buffer = sv_newmortal();
1920 gv_efullname3(buffer, gv, "*");
675c862f 1921 not_a_number(buffer);
8e629ff4 1922 }
675c862f
AL
1923 /* We just want something true to return, so that S_sv_2iuv_common
1924 can tail call us and return true. */
19f6321d 1925 return TRUE;
675c862f
AL
1926}
1927
25da4f38
IZ
1928/* Actually, ISO C leaves conversion of UV to IV undefined, but
1929 until proven guilty, assume that things are not that bad... */
1930
645c22ef
DM
1931/*
1932 NV_PRESERVES_UV:
1933
1934 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1935 an IV (an assumption perl has been based on to date) it becomes necessary
1936 to remove the assumption that the NV always carries enough precision to
1937 recreate the IV whenever needed, and that the NV is the canonical form.
1938 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1939 precision as a side effect of conversion (which would lead to insanity
28e5dec8 1940 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
8a4a3196
KW
1941 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1942 where precision was lost, and IV/UV/NV slots that have a valid conversion
1943 which has lost no precision
645c22ef 1944 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1945 would lose precision, the precise conversion (or differently
1946 imprecise conversion) is also performed and cached, to prevent
1947 requests for different numeric formats on the same SV causing
1948 lossy conversion chains. (lossless conversion chains are perfectly
1949 acceptable (still))
1950
1951
1952 flags are used:
1953 SvIOKp is true if the IV slot contains a valid value
1954 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1955 SvNOKp is true if the NV slot contains a valid value
1956 SvNOK is true only if the NV value is accurate
1957
1958 so
645c22ef 1959 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1960 IV(or UV) would lose accuracy over a direct conversion from PV to
1961 IV(or UV). If it would, cache both conversions, return NV, but mark
1962 SV as IOK NOKp (ie not NOK).
1963
645c22ef 1964 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1965 NV would lose accuracy over a direct conversion from PV to NV. If it
1966 would, cache both conversions, flag similarly.
1967
1968 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1969 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1970 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1971 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1972 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1973
645c22ef
DM
1974 The benefit of this is that operations such as pp_add know that if
1975 SvIOK is true for both left and right operands, then integer addition
1976 can be used instead of floating point (for cases where the result won't
1977 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1978 loss of precision compared with integer addition.
1979
1980 * making IV and NV equal status should make maths accurate on 64 bit
1981 platforms
1982 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 1983 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
1984 looking for SvIOK and checking for overflow will not outweigh the
1985 fp to integer speedup)
1986 * will slow down integer operations (callers of SvIV) on "inaccurate"
1987 values, as the change from SvIOK to SvIOKp will cause a call into
1988 sv_2iv each time rather than a macro access direct to the IV slot
1989 * should speed up number->string conversion on integers as IV is
645c22ef 1990 favoured when IV and NV are equally accurate
28e5dec8
JH
1991
1992 ####################################################################
645c22ef
DM
1993 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1994 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1995 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
1996 ####################################################################
1997
645c22ef 1998 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
1999 performance ratio.
2000*/
2001
2002#ifndef NV_PRESERVES_UV
645c22ef
DM
2003# define IS_NUMBER_UNDERFLOW_IV 1
2004# define IS_NUMBER_UNDERFLOW_UV 2
2005# define IS_NUMBER_IV_AND_UV 2
2006# define IS_NUMBER_OVERFLOW_IV 4
2007# define IS_NUMBER_OVERFLOW_UV 5
2008
2009/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
2010
2011/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2012STATIC int
5aaab254 2013S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
2014# ifdef DEBUGGING
2015 , I32 numtype
2016# endif
2017 )
28e5dec8 2018{
7918f24d 2019 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
23491f1d 2020 PERL_UNUSED_CONTEXT;
7918f24d 2021
3f7c398e 2022 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
2023 if (SvNVX(sv) < (NV)IV_MIN) {
2024 (void)SvIOKp_on(sv);
2025 (void)SvNOK_on(sv);
45977657 2026 SvIV_set(sv, IV_MIN);
28e5dec8
JH
2027 return IS_NUMBER_UNDERFLOW_IV;
2028 }
2029 if (SvNVX(sv) > (NV)UV_MAX) {
2030 (void)SvIOKp_on(sv);
2031 (void)SvNOK_on(sv);
2032 SvIsUV_on(sv);
607fa7f2 2033 SvUV_set(sv, UV_MAX);
28e5dec8
JH
2034 return IS_NUMBER_OVERFLOW_UV;
2035 }
c2988b20
NC
2036 (void)SvIOKp_on(sv);
2037 (void)SvNOK_on(sv);
2038 /* Can't use strtol etc to convert this string. (See truth table in
2039 sv_2iv */
2040 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 2041 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2042 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2043 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2044 } else {
2045 /* Integer is imprecise. NOK, IOKp */
2046 }
2047 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2048 }
2049 SvIsUV_on(sv);
607fa7f2 2050 SvUV_set(sv, U_V(SvNVX(sv)));
659c4b96 2051 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2052 if (SvUVX(sv) == UV_MAX) {
2053 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2054 possibly be preserved by NV. Hence, it must be overflow.
2055 NOK, IOKp */
2056 return IS_NUMBER_OVERFLOW_UV;
2057 }
2058 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2059 } else {
2060 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2061 }
c2988b20 2062 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2063}
645c22ef
DM
2064#endif /* !NV_PRESERVES_UV*/
2065
a13f4dff 2066/* If numtype is infnan, set the NV of the sv accordingly.
5564cd7f 2067 * If numtype is anything else, try setting the NV using Atof(PV). */
a13f4dff 2068static void
5564cd7f 2069S_sv_setnv(pTHX_ SV* sv, int numtype)
a13f4dff 2070{
07925c5e 2071 bool pok = cBOOL(SvPOK(sv));
5564cd7f 2072 bool nok = FALSE;
a13f4dff
JH
2073 if ((numtype & IS_NUMBER_INFINITY)) {
2074 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
5564cd7f 2075 nok = TRUE;
a13f4dff
JH
2076 }
2077 else if ((numtype & IS_NUMBER_NAN)) {
2078 SvNV_set(sv, NV_NAN);
d48bd569 2079 nok = TRUE;
a13f4dff 2080 }
d48bd569 2081 else if (pok) {
a13f4dff 2082 SvNV_set(sv, Atof(SvPVX_const(sv)));
d48bd569
JH
2083 /* Purposefully no true nok here, since we don't want to blow
2084 * away the possible IOK/UV of an existing sv. */
2085 }
5564cd7f 2086 if (nok) {
d48bd569 2087 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
5564cd7f
JH
2088 if (pok)
2089 SvPOK_on(sv); /* PV is okay, though. */
2090 }
a13f4dff
JH
2091}
2092
af359546 2093STATIC bool
7918f24d
NC
2094S_sv_2iuv_common(pTHX_ SV *const sv)
2095{
7918f24d
NC
2096 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2097
af359546 2098 if (SvNOKp(sv)) {
28e5dec8
JH
2099 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2100 * without also getting a cached IV/UV from it at the same time
2101 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2102 * IV or UV at same time to avoid this. */
2103 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38 2104
a83034fc 2105 if (UNLIKELY(Perl_isinfnan(SvNVX(sv))))
0f83c5a4
JH
2106 return FALSE;
2107
25da4f38
IZ
2108 if (SvTYPE(sv) == SVt_NV)
2109 sv_upgrade(sv, SVt_PVNV);
2110
28e5dec8
JH
2111 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2112 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2113 certainly cast into the IV range at IV_MAX, whereas the correct
2114 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2115 cases go to UV */
2116 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2117 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2118 if (SvNVX(sv) == (NV) SvIVX(sv)
28e5dec8
JH
2119#ifndef NV_PRESERVES_UV
2120 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2121 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2122 /* Don't flag it as "accurately an integer" if the number
2123 came from a (by definition imprecise) NV operation, and
2124 we're outside the range of NV integer precision */
2125#endif
2126 ) {
a43d94f2
NC
2127 if (SvNOK(sv))
2128 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2129 else {
2130 /* scalar has trailing garbage, eg "42a" */
2131 }
28e5dec8 2132 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2133 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2134 PTR2UV(sv),
2135 SvNVX(sv),
2136 SvIVX(sv)));
2137
2138 } else {
2139 /* IV not precise. No need to convert from PV, as NV
2140 conversion would already have cached IV if it detected
2141 that PV->IV would be better than PV->NV->IV
2142 flags already correct - don't set public IOK. */
2143 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2144 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2145 PTR2UV(sv),
2146 SvNVX(sv),
2147 SvIVX(sv)));
2148 }
2149 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2150 but the cast (NV)IV_MIN rounds to a the value less (more
2151 negative) than IV_MIN which happens to be equal to SvNVX ??
2152 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2153 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2154 (NV)UVX == NVX are both true, but the values differ. :-(
2155 Hopefully for 2s complement IV_MIN is something like
2156 0x8000000000000000 which will be exact. NWC */
d460ef45 2157 }
25da4f38 2158 else {
607fa7f2 2159 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8 2160 if (
659c4b96 2161 (SvNVX(sv) == (NV) SvUVX(sv))
28e5dec8
JH
2162#ifndef NV_PRESERVES_UV
2163 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2164 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2165 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2166 /* Don't flag it as "accurately an integer" if the number
2167 came from a (by definition imprecise) NV operation, and
2168 we're outside the range of NV integer precision */
2169#endif
a43d94f2 2170 && SvNOK(sv)
28e5dec8
JH
2171 )
2172 SvIOK_on(sv);
25da4f38 2173 SvIsUV_on(sv);
1c846c1f 2174 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2175 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2176 PTR2UV(sv),
57def98f
JH
2177 SvUVX(sv),
2178 SvUVX(sv)));
25da4f38 2179 }
748a9306 2180 }
cd84013a 2181 else if (SvPOKp(sv)) {
c2988b20 2182 UV value;
504618e9 2183 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2184 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2185 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2186 the same as the direct translation of the initial string
2187 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2188 be careful to ensure that the value with the .456 is around if the
2189 NV value is requested in the future).
1c846c1f 2190
af359546 2191 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2192 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2193 cache the NV if we are sure it's not needed.
25da4f38 2194 */
16b7a9a4 2195
c2988b20
NC
2196 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2197 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2198 == IS_NUMBER_IN_UV) {
5e045b90 2199 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2200 if (SvTYPE(sv) < SVt_PVIV)
2201 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2202 (void)SvIOK_on(sv);
c2988b20
NC
2203 } else if (SvTYPE(sv) < SVt_PVNV)
2204 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2205
a13f4dff 2206 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
5564cd7f 2207 S_sv_setnv(aTHX_ sv, numtype);
a13f4dff
JH
2208 return FALSE;
2209 }
2210
f2524eef 2211 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2212 we aren't going to call atof() below. If NVs don't preserve UVs
2213 then the value returned may have more precision than atof() will
2214 return, even though value isn't perfectly accurate. */
2215 if ((numtype & (IS_NUMBER_IN_UV
2216#ifdef NV_PRESERVES_UV
2217 | IS_NUMBER_NOT_INT
2218#endif
2219 )) == IS_NUMBER_IN_UV) {
2220 /* This won't turn off the public IOK flag if it was set above */
2221 (void)SvIOKp_on(sv);
2222
2223 if (!(numtype & IS_NUMBER_NEG)) {
2224 /* positive */;
2225 if (value <= (UV)IV_MAX) {
45977657 2226 SvIV_set(sv, (IV)value);
c2988b20 2227 } else {
af359546 2228 /* it didn't overflow, and it was positive. */
607fa7f2 2229 SvUV_set(sv, value);
c2988b20
NC
2230 SvIsUV_on(sv);
2231 }
2232 } else {
2233 /* 2s complement assumption */
2234 if (value <= (UV)IV_MIN) {
45977657 2235 SvIV_set(sv, -(IV)value);
c2988b20
NC
2236 } else {
2237 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2238 I'm assuming it will be rare. */
c2988b20
NC
2239 if (SvTYPE(sv) < SVt_PVNV)
2240 sv_upgrade(sv, SVt_PVNV);
2241 SvNOK_on(sv);
2242 SvIOK_off(sv);
2243 SvIOKp_on(sv);
9d6ce603 2244 SvNV_set(sv, -(NV)value);
45977657 2245 SvIV_set(sv, IV_MIN);
c2988b20
NC
2246 }
2247 }
2248 }
2249 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2250 will be in the previous block to set the IV slot, and the next
2251 block to set the NV slot. So no else here. */
2252
2253 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2254 != IS_NUMBER_IN_UV) {
2255 /* It wasn't an (integer that doesn't overflow the UV). */
5564cd7f 2256 S_sv_setnv(aTHX_ sv, numtype);
28e5dec8 2257
c2988b20
NC
2258 if (! numtype && ckWARN(WARN_NUMERIC))
2259 not_a_number(sv);
28e5dec8 2260
88cb8500 2261 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
c2988b20 2262 PTR2UV(sv), SvNVX(sv)));
28e5dec8 2263
28e5dec8 2264#ifdef NV_PRESERVES_UV
af359546
NC
2265 (void)SvIOKp_on(sv);
2266 (void)SvNOK_on(sv);
2267 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2268 SvIV_set(sv, I_V(SvNVX(sv)));
2269 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2270 SvIOK_on(sv);
2271 } else {
6f207bd3 2272 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2273 }
2274 /* UV will not work better than IV */
2275 } else {
2276 if (SvNVX(sv) > (NV)UV_MAX) {
2277 SvIsUV_on(sv);
2278 /* Integer is inaccurate. NOK, IOKp, is UV */
2279 SvUV_set(sv, UV_MAX);
af359546
NC
2280 } else {
2281 SvUV_set(sv, U_V(SvNVX(sv)));
2282 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2283 NV preservse UV so can do correct comparison. */
2284 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2285 SvIOK_on(sv);
af359546 2286 } else {
6f207bd3 2287 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2288 }
2289 }
4b0c9573 2290 SvIsUV_on(sv);
af359546 2291 }
28e5dec8 2292#else /* NV_PRESERVES_UV */
c2988b20
NC
2293 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2294 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2295 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2296 grok_number above. The NV slot has just been set using
2297 Atof. */
560b0c46 2298 SvNOK_on(sv);
c2988b20
NC
2299 assert (SvIOKp(sv));
2300 } else {
2301 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2302 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2303 /* Small enough to preserve all bits. */
2304 (void)SvIOKp_on(sv);
2305 SvNOK_on(sv);
45977657 2306 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2307 if ((NV)(SvIVX(sv)) == SvNVX(sv))
c2988b20
NC
2308 SvIOK_on(sv);
2309 /* Assumption: first non-preserved integer is < IV_MAX,
2310 this NV is in the preserved range, therefore: */
2311 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2312 < (UV)IV_MAX)) {
32fdb065 2313 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
2314 }
2315 } else {
2316 /* IN_UV NOT_INT
2317 0 0 already failed to read UV.
2318 0 1 already failed to read UV.
2319 1 0 you won't get here in this case. IV/UV
2320 slot set, public IOK, Atof() unneeded.
2321 1 1 already read UV.
2322 so there's no point in sv_2iuv_non_preserve() attempting
2323 to use atol, strtol, strtoul etc. */
47031da6 2324# ifdef DEBUGGING
40a17c4c 2325 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2326# else
2327 sv_2iuv_non_preserve (sv);
2328# endif
c2988b20
NC
2329 }
2330 }
28e5dec8 2331#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2332 /* It might be more code efficient to go through the entire logic above
2333 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2334 gets complex and potentially buggy, so more programmer efficient
2335 to do it this way, by turning off the public flags: */
2336 if (!numtype)
2337 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2338 }
af359546
NC
2339 }
2340 else {
675c862f 2341 if (isGV_with_GP(sv))
159b6efe 2342 return glob_2number(MUTABLE_GV(sv));
180488f8 2343
4f62cd62 2344 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
af359546 2345 report_uninit(sv);
25da4f38
IZ
2346 if (SvTYPE(sv) < SVt_IV)
2347 /* Typically the caller expects that sv_any is not NULL now. */
2348 sv_upgrade(sv, SVt_IV);
af359546
NC
2349 /* Return 0 from the caller. */
2350 return TRUE;
2351 }
2352 return FALSE;
2353}
2354
2355/*
2356=for apidoc sv_2iv_flags
2357
2358Return the integer value of an SV, doing any necessary string
2359conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2360Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2361
2362=cut
2363*/
2364
2365IV
5aaab254 2366Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2367{
1061065f 2368 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
4bac9ae4 2369
217f6fa3
FC
2370 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2371 && SvTYPE(sv) != SVt_PVFM);
2372
4bac9ae4
CS
2373 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2374 mg_get(sv);
2375
a83034fc 2376 if (SvNOK(sv) && UNLIKELY(Perl_isinfnan(SvNVX(sv))))
0f83c5a4
JH
2377 return 0; /* So wrong but what can we do. */
2378
4bac9ae4
CS
2379 if (SvROK(sv)) {
2380 if (SvAMAGIC(sv)) {
2381 SV * tmpstr;
2382 if (flags & SV_SKIP_OVERLOAD)
2383 return 0;
2384 tmpstr = AMG_CALLunary(sv, numer_amg);
2385 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2386 return SvIV(tmpstr);
2387 }
2388 }
2389 return PTR2IV(SvRV(sv));
2390 }
2391
8d919b0a 2392 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2393 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2394 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2395 In practice they are extremely unlikely to actually get anywhere
2396 accessible by user Perl code - the only way that I'm aware of is when
2397 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2398
2399 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2400 */
8d919b0a 2401 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2402 {
71c558c3 2403 UV value;
8d919b0a
FC
2404 const char * const ptr =
2405 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
0f83c5a4
JH
2406 const int numtype = grok_number(ptr, SvCUR(sv), &value);
2407
2408 assert((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN)) == 0);
71c558c3
NC
2409
2410 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2411 == IS_NUMBER_IN_UV) {
2412 /* It's definitely an integer */
2413 if (numtype & IS_NUMBER_NEG) {
2414 if (value < (UV)IV_MIN)
2415 return -(IV)value;
2416 } else {
2417 if (value < (UV)IV_MAX)
2418 return (IV)value;
2419 }
2420 }
058b8ae2 2421
71c558c3
NC
2422 if (!numtype) {
2423 if (ckWARN(WARN_NUMERIC))
2424 not_a_number(sv);
2425 }
8d919b0a 2426 return I_V(Atof(ptr));
e20b6c3b 2427 }
4bac9ae4
CS
2428 }
2429
2430 if (SvTHINKFIRST(sv)) {
73381561 2431#ifdef PERL_OLD_COPY_ON_WRITE
af359546
NC
2432 if (SvIsCOW(sv)) {
2433 sv_force_normal_flags(sv, 0);
2434 }
73381561 2435#endif
af359546
NC
2436 if (SvREADONLY(sv) && !SvOK(sv)) {
2437 if (ckWARN(WARN_UNINITIALIZED))
2438 report_uninit(sv);
2439 return 0;
2440 }
2441 }
4bac9ae4 2442
af359546
NC
2443 if (!SvIOKp(sv)) {
2444 if (S_sv_2iuv_common(aTHX_ sv))
2445 return 0;
79072805 2446 }
4bac9ae4 2447
1d7c1841
GS
2448 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2449 PTR2UV(sv),SvIVX(sv)));
25da4f38 2450 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2451}
2452
645c22ef 2453/*
891f9566 2454=for apidoc sv_2uv_flags
645c22ef
DM
2455
2456Return the unsigned integer value of an SV, doing any necessary string
891f9566
YST
2457conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2458Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2459
2460=cut
2461*/
2462
ff68c719 2463UV
5aaab254 2464Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2465{
1061065f 2466 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
4bac9ae4
CS
2467
2468 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2469 mg_get(sv);
2470
a83034fc 2471 if (SvNOK(sv) && UNLIKELY(Perl_isinfnan(SvNVX(sv))))
0f83c5a4
JH
2472 return 0; /* So wrong but what can we do. */
2473
4bac9ae4
CS
2474 if (SvROK(sv)) {
2475 if (SvAMAGIC(sv)) {
2476 SV *tmpstr;
2477 if (flags & SV_SKIP_OVERLOAD)
2478 return 0;
2479 tmpstr = AMG_CALLunary(sv, numer_amg);
2480 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2481 return SvUV(tmpstr);
2482 }
2483 }
2484 return PTR2UV(SvRV(sv));
2485 }
2486
8d919b0a 2487 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2488 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2489 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2490 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2491 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2492 {
71c558c3 2493 UV value;
8d919b0a
FC
2494 const char * const ptr =
2495 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
0f83c5a4
JH
2496 const int numtype = grok_number(ptr, SvCUR(sv), &value);
2497
2498 assert((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN)) == 0);
71c558c3
NC
2499
2500 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2501 == IS_NUMBER_IN_UV) {
2502 /* It's definitely an integer */
2503 if (!(numtype & IS_NUMBER_NEG))
2504 return value;
2505 }
058b8ae2 2506
71c558c3
NC
2507 if (!numtype) {
2508 if (ckWARN(WARN_NUMERIC))
2509 not_a_number(sv);
2510 }
8d919b0a 2511 return U_V(Atof(ptr));
e20b6c3b 2512 }
4bac9ae4
CS
2513 }
2514
2515 if (SvTHINKFIRST(sv)) {
73381561 2516#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2517 if (SvIsCOW(sv)) {
2518 sv_force_normal_flags(sv, 0);
8a818333 2519 }
73381561 2520#endif
0336b60e 2521 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2522 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2523 report_uninit(sv);
ff68c719
PP
2524 return 0;
2525 }
2526 }
4bac9ae4 2527
af359546
NC
2528 if (!SvIOKp(sv)) {
2529 if (S_sv_2iuv_common(aTHX_ sv))
2530 return 0;
ff68c719 2531 }
25da4f38 2532
1d7c1841
GS
2533 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2534 PTR2UV(sv),SvUVX(sv)));
25da4f38 2535 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2536}
2537
645c22ef 2538/*
196007d1 2539=for apidoc sv_2nv_flags
645c22ef
DM
2540
2541Return the num value of an SV, doing any necessary string or integer
fde67290 2542conversion. If flags includes SV_GMAGIC, does an mg_get() first.
39d5de13 2543Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2544
2545=cut
2546*/
2547
65202027 2548NV
5aaab254 2549Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2550{
1061065f
DD
2551 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2552
217f6fa3
FC
2553 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2554 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2555 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2556 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2557 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2558 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2559 const char *ptr;
39d5de13
DM
2560 if (flags & SV_GMAGIC)
2561 mg_get(sv);
463ee0b2
LW
2562 if (SvNOKp(sv))
2563 return SvNVX(sv);
cd84013a 2564 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2565 ptr = SvPVX_const(sv);
2566 grokpv:
041457d9 2567 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2568 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2569 not_a_number(sv);
8d919b0a 2570 return Atof(ptr);
a0d0e21e 2571 }
25da4f38 2572 if (SvIOKp(sv)) {
1c846c1f 2573 if (SvIsUV(sv))
65202027 2574 return (NV)SvUVX(sv);
25da4f38 2575 else
65202027 2576 return (NV)SvIVX(sv);
47a72cb8
NC
2577 }
2578 if (SvROK(sv)) {
2579 goto return_rok;
2580 }
8d919b0a
FC
2581 if (isREGEXP(sv)) {
2582 ptr = RX_WRAPPED((REGEXP *)sv);
2583 goto grokpv;
2584 }
47a72cb8
NC
2585 assert(SvTYPE(sv) >= SVt_PVMG);
2586 /* This falls through to the report_uninit near the end of the
2587 function. */
2588 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2589 if (SvROK(sv)) {
47a72cb8 2590 return_rok:
deb46114 2591 if (SvAMAGIC(sv)) {
aee036bb
DM
2592 SV *tmpstr;
2593 if (flags & SV_SKIP_OVERLOAD)
2594 return 0;
31d632c3 2595 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2596 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2597 return SvNV(tmpstr);
2598 }
2599 }
2600 return PTR2NV(SvRV(sv));
a0d0e21e 2601 }
73381561 2602#ifdef PERL_OLD_COPY_ON_WRITE
765f542d
NC
2603 if (SvIsCOW(sv)) {
2604 sv_force_normal_flags(sv, 0);
8a818333 2605 }
73381561 2606#endif
0336b60e 2607 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2608 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2609 report_uninit(sv);
ed6116ce
LW
2610 return 0.0;
2611 }
79072805
LW
2612 }
2613 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2614 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2615 sv_upgrade(sv, SVt_NV);
097ee67d 2616 DEBUG_c({
f93f4e46 2617 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841 2618 PerlIO_printf(Perl_debug_log,
88cb8500 2619 "0x%"UVxf" num(%" NVgf ")\n",
1d7c1841 2620 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2621 RESTORE_NUMERIC_LOCAL();
2622 });
79072805
LW
2623 }
2624 else if (SvTYPE(sv) < SVt_PVNV)
2625 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2626 if (SvNOKp(sv)) {
2627 return SvNVX(sv);
61604483 2628 }
59d8ce62 2629 if (SvIOKp(sv)) {
9d6ce603 2630 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2631#ifdef NV_PRESERVES_UV
a43d94f2
NC
2632 if (SvIOK(sv))
2633 SvNOK_on(sv);
2634 else
2635 SvNOKp_on(sv);
28e5dec8
JH
2636#else
2637 /* Only set the public NV OK flag if this NV preserves the IV */
2638 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2639 if (SvIOK(sv) &&
2640 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2641 : (SvIVX(sv) == I_V(SvNVX(sv))))
2642 SvNOK_on(sv);
2643 else
2644 SvNOKp_on(sv);
2645#endif
93a17b20 2646 }
cd84013a 2647 else if (SvPOKp(sv)) {
c2988b20 2648 UV value;
3f7c398e 2649 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2650 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2651 not_a_number(sv);
28e5dec8 2652#ifdef NV_PRESERVES_UV
c2988b20
NC
2653 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2654 == IS_NUMBER_IN_UV) {
5e045b90 2655 /* It's definitely an integer */
9d6ce603 2656 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
66d83377 2657 } else {
5564cd7f 2658 S_sv_setnv(aTHX_ sv, numtype);
66d83377 2659 }
a43d94f2
NC
2660 if (numtype)
2661 SvNOK_on(sv);
2662 else
2663 SvNOKp_on(sv);
28e5dec8 2664#else
0f83c5a4
JH
2665 if ((numtype & IS_NUMBER_INFINITY)) {
2666 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2667 SvNOK_on(sv);
2668 } else if ((numtype & IS_NUMBER_NAN)) {
2669 SvNV_set(sv, NV_NAN);
c2988b20
NC
2670 SvNOK_on(sv);
2671 } else {
0f83c5a4
JH
2672 SvNV_set(sv, Atof(SvPVX_const(sv)));
2673 /* Only set the public NV OK flag if this NV preserves the value in
2674 the PV at least as well as an IV/UV would.
2675 Not sure how to do this 100% reliably. */
2676 /* if that shift count is out of range then Configure's test is
2677 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2678 UV_BITS */
2679 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2680 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2681 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2682 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2683 /* Can't use strtol etc to convert this string, so don't try.
2684 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2685 SvNOK_on(sv);
c2988b20 2686 } else {
0f83c5a4
JH
2687 /* value has been set. It may not be precise. */
2688 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2689 /* 2s complement assumption for (UV)IV_MIN */
2690 SvNOK_on(sv); /* Integer is too negative. */
c2988b20 2691 } else {
0f83c5a4
JH
2692 SvNOKp_on(sv);
2693 SvIOKp_on(sv);
6fa402ec 2694
0f83c5a4
JH
2695 if (numtype & IS_NUMBER_NEG) {
2696 SvIV_set(sv, -(IV)value);
2697 } else if (value <= (UV)IV_MAX) {
2698 SvIV_set(sv, (IV)value);
2699 } else {
2700 SvUV_set(sv, value);
2701 SvIsUV_on(sv);
2702 }
c2988b20 2703
0f83c5a4
JH
2704 if (numtype & IS_NUMBER_NOT_INT) {
2705 /* I believe that even if the original PV had decimals,
2706 they are lost beyond the limit of the FP precision.
2707 However, neither is canonical, so both only get p
2708 flags. NWC, 2000/11/25 */
2709 /* Both already have p flags, so do nothing */
2710 } else {
2711 const NV nv = SvNVX(sv);
2712 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2713 if (SvIVX(sv) == I_V(nv)) {
c2988b20 2714 SvNOK_on(sv);
0f83c5a4
JH
2715 } else {
2716 /* It had no "." so it must be integer. */
2717 }
2718 SvIOK_on(sv);
2719 } else {
2720 /* between IV_MAX and NV(UV_MAX).
2721 Could be slightly > UV_MAX */
2722
2723 if (numtype & IS_NUMBER_NOT_INT) {
2724 /* UV and NV both imprecise. */
2725 } else {
2726 const UV nv_as_uv = U_V(nv);
2727
2728 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2729 SvNOK_on(sv);
2730 }
2731 SvIOK_on(sv);
c2988b20
NC
2732 }
2733 }
2734 }
2735 }
2736 }
0f83c5a4
JH
2737 /* It might be more code efficient to go through the entire logic above
2738 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2739 gets complex and potentially buggy, so more programmer efficient
a43d94f2 2740 to do it this way, by turning off the public flags: */
0f83c5a4
JH
2741 if (!numtype)
2742 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2743 }
28e5dec8 2744#endif /* NV_PRESERVES_UV */
93a17b20 2745 }
79072805 2746 else {
0f83c5a4
JH
2747 if (isGV_with_GP(sv)) {
2748 glob_2number(MUTABLE_GV(sv));
2749 return 0.0;
2750 }
180488f8 2751
0f83c5a4
JH
2752 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2753 report_uninit(sv);
2754 assert (SvTYPE(sv) >= SVt_NV);
2755 /* Typically the caller expects that sv_any is not NULL now. */
2756 /* XXX Ilya implies that this is a bug in callers that assume this
2757 and ideally should be fixed. */
2758 return 0.0;
79072805 2759 }
097ee67d 2760 DEBUG_c({
0f83c5a4
JH
2761 STORE_NUMERIC_LOCAL_SET_STANDARD();
2762 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2763 PTR2UV(sv), SvNVX(sv));
2764 RESTORE_NUMERIC_LOCAL();
2765 });
463ee0b2 2766 return SvNVX(sv);
79072805
LW
2767}
2768
800401ee
JH
2769/*
2770=for apidoc sv_2num
2771
2772Return an SV with the numeric value of the source SV, doing any necessary
a196a5fa
JH
2773reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2774access this function.
800401ee
JH
2775
2776=cut
2777*/
2778
2779SV *
5aaab254 2780Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2781{
7918f24d
NC
2782 PERL_ARGS_ASSERT_SV_2NUM;
2783
b9ee0594
RGS
2784 if (!SvROK(sv))
2785 return sv;
800401ee 2786 if (SvAMAGIC(sv)) {
31d632c3 2787 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2788 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2789 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2790 return sv_2num(tmpsv);
2791 }
2792 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2793}
2794
645c22ef
DM
2795/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2796 * UV as a string towards the end of buf, and return pointers to start and
2797 * end of it.
2798 *
2799 * We assume that buf is at least TYPE_CHARS(UV) long.
2800 */
2801
864dbfa3 2802static char *
5de3775c 2803S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2804{
25da4f38 2805 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2806 char * const ebuf = ptr;
25da4f38 2807 int sign;
25da4f38 2808
7918f24d
NC
2809 PERL_ARGS_ASSERT_UIV_2BUF;
2810
25da4f38
IZ
2811 if (is_uv)
2812 sign = 0;
2813 else if (iv >= 0) {
2814 uv = iv;
2815 sign = 0;
2816 } else {
2817 uv = -iv;
2818 sign = 1;
2819 }
2820 do {
eb160463 2821 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2822 } while (uv /= 10);
2823 if (sign)
2824 *--ptr = '-';
2825 *peob = ebuf;
2826 return ptr;
2827}
2828
0c7e610f 2829/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
fde71869
JH
2830 * infinity or a not-a-number, writes the appropriate strings to the
2831 * buffer, including a zero byte. On success returns the written length,
5e85836e 2832 * excluding the zero byte, on failure (not an infinity, not a nan, or the
fb8cdbc5
JH
2833 * maxlen too small) returns zero.
2834 *
2835 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2836 * shared string constants we point to, instead of generating a new
2837 * string for each instance. */
0c7e610f 2838STATIC size_t
5e85836e 2839S_infnan_2pv(NV nv, char* buffer, size_t maxlen) {
fb8cdbc5 2840 assert(maxlen >= 4);
5e85836e 2841 if (maxlen < 4) /* "Inf\0", "NaN\0" */
0c7e610f
JH
2842 return 0;
2843 else {
2844 char* s = buffer;
3fd158ea 2845 if (Perl_isinf(nv)) {
128eeacb
DD
2846 if (nv < 0) {
2847 if (maxlen < 5) /* "-Inf\0" */
2848 return 0;
2849 *s++ = '-';
2850 }
2851 *s++ = 'I';
2852 *s++ = 'n';
2853 *s++ = 'f';
3fd158ea
JH
2854 } else if (Perl_isnan(nv)) {
2855 *s++ = 'N';
2856 *s++ = 'a';
2857 *s++ = 'N';
2858 /* XXX optionally output the payload mantissa bits as
2859 * "(unsigned)" (to match the nan("...") C99 function,
2860 * or maybe as "(0xhhh...)" would make more sense...
2861 * provide a format string so that the user can decide?
2862 * NOTE: would affect the maxlen and assert() logic.*/
128eeacb
DD
2863 }
2864
0c7e610f
JH
2865 else
2866 return 0;
5e85836e 2867 assert((s == buffer + 3) || (s == buffer + 4));
0c7e610f 2868 *s++ = 0;
5e85836e 2869 return s - buffer - 1; /* -1: excluding the zero byte */
0c7e610f
JH
2870 }
2871}
2872
645c22ef
DM
2873/*
2874=for apidoc sv_2pv_flags
2875
ff276b08 2876Returns a pointer to the string value of an SV, and sets *lp to its length.
fde67290
FC
2877If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2878string if necessary. Normally invoked via the C<SvPV_flags> macro.
2879C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
645c22ef
DM
2880
2881=cut
2882*/
2883
8d6d96c1 2884char *
5aaab254 2885Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
8d6d96c1 2886{
eb578fdb 2887 char *s;
79072805 2888
1061065f
DD
2889 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2890
217f6fa3
FC
2891 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2892 && SvTYPE(sv) != SVt_PVFM);
4bac9ae4
CS
2893 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2894 mg_get(sv);
2895 if (SvROK(sv)) {
2896 if (SvAMAGIC(sv)) {
2897 SV *tmpstr;
2898 if (flags & SV_SKIP_OVERLOAD)
2899 return NULL;
2900 tmpstr = AMG_CALLunary(sv, string_amg);
2901 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2902 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2903 /* Unwrap this: */
2904 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2905 */
75dfc8ec 2906
4bac9ae4
CS
2907 char *pv;
2908 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2909 if (flags & SV_CONST_RETURN) {
2910 pv = (char *) SvPVX_const(tmpstr);
50adf7d2 2911 } else {
4bac9ae4
CS
2912 pv = (flags & SV_MUTABLE_RETURN)
2913 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
50adf7d2 2914 }
4bac9ae4
CS
2915 if (lp)
2916 *lp = SvCUR(tmpstr);
2917 } else {
2918 pv = sv_2pv_flags(tmpstr, lp, flags);
50adf7d2 2919 }
4bac9ae4
CS
2920 if (SvUTF8(tmpstr))
2921 SvUTF8_on(sv);
2922 else
2923 SvUTF8_off(sv);
2924 return pv;
deb46114 2925 }
4bac9ae4
CS
2926 }
2927 {
2928 STRLEN len;
2929 char *retval;
2930 char *buffer;
2931 SV *const referent = SvRV(sv);
2932
2933 if (!referent) {
2934 len = 7;
2935 retval = buffer = savepvn("NULLREF", len);
2936 } else if (SvTYPE(referent) == SVt_REGEXP &&
2937 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2938 amagic_is_enabled(string_amg))) {
2939 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2940
2941 assert(re);
67d2d14d 2942
4bac9ae4
CS
2943 /* If the regex is UTF-8 we want the containing scalar to
2944 have an UTF-8 flag too */
2945 if (RX_UTF8(re))
2946 SvUTF8_on(sv);
2947 else
2948 SvUTF8_off(sv);
67d2d14d 2949
4bac9ae4
CS
2950 if (lp)
2951 *lp = RX_WRAPLEN(re);
67d2d14d 2952
4bac9ae4
CS
2953 return RX_WRAPPED(re);
2954 } else {
2955 const char *const typestr = sv_reftype(referent, 0);
2956 const STRLEN typelen = strlen(typestr);
2957 UV addr = PTR2UV(referent);
2958 const char *stashname = NULL;
2959 STRLEN stashnamelen = 0; /* hush, gcc */
2960 const char *buffer_end;
2961
2962 if (SvOBJECT(referent)) {
2963 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2964
2965 if (name) {
2966 stashname = HEK_KEY(name);
2967 stashnamelen = HEK_LEN(name);
2968
2969 if (HEK_UTF8(name)) {
2970 SvUTF8_on(sv);
fafee734 2971 } else {
4bac9ae4 2972 SvUTF8_off(sv);
fafee734 2973 }
fafee734 2974 } else {
4bac9ae4
CS
2975 stashname = "__ANON__";
2976 stashnamelen = 8;
fafee734 2977 }
4bac9ae4
CS
2978 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2979 + 2 * sizeof(UV) + 2 /* )\0 */;
2980 } else {
2981 len = typelen + 3 /* (0x */
2982 + 2 * sizeof(UV) + 2 /* )\0 */;
2983 }
fafee734 2984
4bac9ae4
CS
2985 Newx(buffer, len, char);
2986 buffer_end = retval = buffer + len;
2987
2988 /* Working backwards */
2989 *--retval = '\0';
2990 *--retval = ')';
2991 do {
2992 *--retval = PL_hexdigit[addr & 15];
2993 } while (addr >>= 4);
2994 *--retval = 'x';
2995 *--retval = '0';
2996 *--retval = '(';
2997
2998 retval -= typelen;
2999 memcpy(retval, typestr, typelen);
3000
3001 if (stashname) {
3002 *--retval = '=';
3003 retval -= stashnamelen;
3004 memcpy(retval, stashname, stashnamelen);
c080367d 3005 }
4bac9ae4
CS
3006 /* retval may not necessarily have reached the start of the
3007 buffer here. */
3008 assert (retval >= buffer);
3009
3010 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 3011 }
cdb061a3 3012 if (lp)
4bac9ae4
CS
3013 *lp = len;
3014 SAVEFREEPV(buffer);
3015 return retval;
79072805 3016 }
79072805 3017 }
4bac9ae4
CS
3018
3019 if (SvPOKp(sv)) {
3020 if (lp)
3021 *lp = SvCUR(sv);
3022 if (flags & SV_MUTABLE_RETURN)
3023 return SvPVX_mutable(sv);
3024 if (flags & SV_CONST_RETURN)
3025 return (char *)SvPVX_const(sv);
3026 return SvPVX(sv);
3027 }
3028
3029 if (SvIOK(sv)) {
28e5dec8
JH
3030 /* I'm assuming that if both IV and NV are equally valid then
3031 converting the IV is going to be more efficient */
e1ec3a88 3032 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
3033 char buf[TYPE_CHARS(UV)];
3034 char *ebuf, *ptr;
97a130b8 3035 STRLEN len;
28e5dec8
JH
3036
3037 if (SvTYPE(sv) < SVt_PVIV)
3038 sv_upgrade(sv, SVt_PVIV);
4ea1d550 3039 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 3040 len = ebuf - ptr;
5902b6a9 3041 /* inlined from sv_setpvn */
97a130b8
NC
3042 s = SvGROW_mutable(sv, len + 1);
3043 Move(ptr, s, len, char);
3044 s += len;
28e5dec8 3045 *s = '\0';
b127e37e 3046 SvPOK_on(sv);
28e5dec8 3047 }
4bac9ae4 3048 else if (SvNOK(sv)) {
79072805
LW
3049 if (SvTYPE(sv) < SVt_PVNV)
3050 sv_upgrade(sv, SVt_PVNV);
128eeacb
DD
3051 if (SvNVX(sv) == 0.0
3052#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
0f83c5a4 3053 /* XXX Create SvNVXeq(sv, x)? Or just SvNVXzero(sv)? */
128eeacb
DD
3054 && !Perl_isnan(SvNVX(sv))
3055#endif
3056 ) {
29912d93
NC
3057 s = SvGROW_mutable(sv, 2);
3058 *s++ = '0';
3059 *s = '\0';
3060 } else {
5e85836e 3061 STRLEN len;
fb8cdbc5 3062 STRLEN size = 5; /* "-Inf\0" */
0c7e610f 3063
fb8cdbc5 3064 s = SvGROW_mutable(sv, size);
5e85836e 3065 len = S_infnan_2pv(SvNVX(sv), s, size);
fb8cdbc5 3066 if (len > 0) {
0c7e610f 3067 s += len;
fb8cdbc5
JH
3068 SvPOK_on(sv);
3069 }
0c7e610f 3070 else {
0c7e610f 3071 /* some Xenix systems wipe out errno here */
fb8cdbc5
JH
3072 dSAVE_ERRNO;
3073
3840bff0
JH
3074 size =
3075 1 + /* sign */
3076 1 + /* "." */
3077 NV_DIG +
3078 1 + /* "e" */
3079 1 + /* sign */
3080 5 + /* exponent digits */
3081 1 + /* \0 */
3082 2; /* paranoia */
b127e37e 3083
fb8cdbc5 3084 s = SvGROW_mutable(sv, size);
b127e37e 3085#ifndef USE_LOCALE_NUMERIC
a4eca1d4
JH
3086 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3087
0c7e610f
JH
3088 SvPOK_on(sv);
3089#else
28acfe03 3090 {
3840bff0 3091 bool local_radix;
0c7e610f 3092 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3840bff0
JH
3093
3094 local_radix =
3095 PL_numeric_local &&
3096 PL_numeric_radix_sv &&
3097 SvUTF8(PL_numeric_radix_sv);
3098 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3099 size += SvLEN(PL_numeric_radix_sv) - 1;
3100 s = SvGROW_mutable(sv, size);
3101 }
3102
a4eca1d4 3103 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
0c7e610f
JH
3104
3105 /* If the radix character is UTF-8, and actually is in the
3106 * output, turn on the UTF-8 flag for the scalar */
3840bff0
JH
3107 if (local_radix &&
3108 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3109 SvUTF8_on(sv);
3110 }
3111
0c7e610f 3112 RESTORE_LC_NUMERIC();
28acfe03 3113 }
68e8f474 3114
0c7e610f
JH
3115 /* We don't call SvPOK_on(), because it may come to
3116 * pass that the locale changes so that the
3117 * stringification we just did is no longer correct. We
3118 * will have to re-stringify every time it is needed */
b127e37e 3119#endif
0c7e610f
JH
3120 RESTORE_ERRNO;
3121 }
3122 while (*s) s++;
bbce6d69 3123 }
79072805 3124 }
4bac9ae4
CS
3125 else if (isGV_with_GP(sv)) {
3126 GV *const gv = MUTABLE_GV(sv);
3127 SV *const buffer = sv_newmortal();
8d1c3e26 3128
4bac9ae4 3129 gv_efullname3(buffer, gv, "*");
180488f8 3130
4bac9ae4
CS
3131 assert(SvPOK(buffer));
3132 if (SvUTF8(buffer))
3133 SvUTF8_on(sv);
3134 if (lp)
3135 *lp = SvCUR(buffer);
3136 return SvPVX(buffer);
3137 }
8d919b0a
FC
3138 else if (isREGEXP(sv)) {
3139 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3140 return RX_WRAPPED((REGEXP *)sv);
3141 }
4bac9ae4 3142 else {
cdb061a3 3143 if (lp)
00b6aa41 3144 *lp = 0;
9f621bb0
NC
3145 if (flags & SV_UNDEF_RETURNS_NULL)
3146 return NULL;
4f62cd62 3147 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3148 report_uninit(sv);
4bac9ae4
CS
3149 /* Typically the caller expects that sv_any is not NULL now. */
3150 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3151 sv_upgrade(sv, SVt_PV);
73d840c0 3152 return (char *)"";
79072805 3153 }
4bac9ae4 3154
cdb061a3 3155 {
823a54a3 3156 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3157 if (lp)
3158 *lp = len;
3159 SvCUR_set(sv, len);
3160 }
1d7c1841 3161 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3162 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3163 if (flags & SV_CONST_RETURN)
3164 return (char *)SvPVX_const(sv);
10516c54
NC
3165 if (flags & SV_MUTABLE_RETURN)
3166 return SvPVX_mutable(sv);
463ee0b2
LW
3167 return SvPVX(sv);
3168}
3169
645c22ef 3170/*
6050d10e
JP
3171=for apidoc sv_copypv
3172
3173Copies a stringified representation of the source SV into the
3174destination SV. Automatically performs any necessary mg_get and
54f0641b 3175coercion of numeric values into strings. Guaranteed to preserve
2575c402 3176UTF8 flag even from overloaded objects. Similar in nature to
54f0641b
NIS
3177sv_2pv[_flags] but operates directly on an SV instead of just the
3178string. Mostly uses sv_2pv_flags to do its work, except when that
6050d10e
JP
3179would lose the UTF-8'ness of the PV.
3180
4bac9ae4
CS
3181=for apidoc sv_copypv_nomg
3182
3183Like sv_copypv, but doesn't invoke get magic first.
3184
3185=for apidoc sv_copypv_flags
3186
3187Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3188include SV_GMAGIC.
3189
6050d10e
JP
3190=cut
3191*/
3192
3193void
5aaab254 3194Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
6050d10e 3195{
4bac9ae4
CS
3196 PERL_ARGS_ASSERT_SV_COPYPV;
3197
3198 sv_copypv_flags(dsv, ssv, 0);
3199}
3200
3201void
5aaab254 3202Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3203{
446eaa42 3204 STRLEN len;
4bac9ae4 3205 const char *s;
7918f24d 3206
4bac9ae4 3207 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3208
c77ed9ca 3209 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
cb50f42d 3210 sv_setpvn(dsv,s,len);
446eaa42 3211 if (SvUTF8(ssv))
cb50f42d 3212 SvUTF8_on(dsv);
446eaa42 3213 else
cb50f42d 3214 SvUTF8_off(dsv);
6050d10e
JP
3215}
3216
3217/*
645c22ef
DM
3218=for apidoc sv_2pvbyte
3219
3220Return a pointer to the byte-encoded representation of the SV, and set *lp
1e54db1a 3221to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3222side-effect.
3223
3224Usually accessed via the C<SvPVbyte> macro.
3225
3226=cut
3227*/
3228
7340a771 3229char *
5aaab254 3230Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3231{
7918f24d
NC
3232 PERL_ARGS_ASSERT_SV_2PVBYTE;
3233
48120f8f 3234 SvGETMAGIC(sv);
4499db73
FC
3235 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3236 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3237 SV *sv2 = sv_newmortal();
48120f8f 3238 sv_copypv_nomg(sv2,sv);
a901b181
FC
3239 sv = sv2;
3240 }
0875d2fe 3241 sv_utf8_downgrade(sv,0);
71eb6d8c 3242 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3243}
3244
645c22ef 3245/*
035cbb0e
RGS
3246=for apidoc sv_2pvutf8
3247
3248Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3249to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3250
3251Usually accessed via the C<SvPVutf8> macro.
3252
3253=cut
3254*/
645c22ef 3255
7340a771 3256char *
5aaab254 3257Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3258{
7918f24d
NC
3259 PERL_ARGS_ASSERT_SV_2PVUTF8;
3260
4499db73
FC
3261 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3262 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3263 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3264 else
3265 SvGETMAGIC(sv);
3266 sv_utf8_upgrade_nomg(sv);
c3ec315f 3267 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3268}
1c846c1f 3269
7ee2227d 3270
645c22ef
DM
3271/*
3272=for apidoc sv_2bool
3273
06c841cf
FC
3274This macro is only used by sv_true() or its macro equivalent, and only if
3275the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3276It calls sv_2bool_flags with the SV_GMAGIC flag.
3277
3278=for apidoc sv_2bool_flags
3279
3280This function is only used by sv_true() and friends, and only if
fde67290 3281the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
06c841cf
FC
3282contain SV_GMAGIC, then it does an mg_get() first.
3283
645c22ef
DM
3284
3285=cut
3286*/
3287
463ee0b2 3288bool
9d176cd8 3289Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
463ee0b2 3290{
06c841cf 3291 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3292
9d176cd8 3293 restart:
06c841cf 3294 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3295
a0d0e21e
LW
3296 if (!SvOK(sv))
3297 return 0;
3298 if (SvROK(sv)) {
fabdb6c0 3299 if (SvAMAGIC(sv)) {
31d632c3 3300 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
9d176cd8
DD
3301 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3302 bool svb;
3303 sv = tmpsv;
3304 if(SvGMAGICAL(sv)) {
3305 flags = SV_GMAGIC;
3306 goto restart; /* call sv_2bool */
3307 }
3308 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3309 else if(!SvOK(sv)) {
3310 svb = 0;
3311 }
3312 else if(SvPOK(sv)) {
3313 svb = SvPVXtrue(sv);
3314 }
3315 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3316 svb = (SvIOK(sv) && SvIVX(sv) != 0)
659c4b96 3317 || (SvNOK(sv) && SvNVX(sv) != 0.0);
9d176cd8
DD
3318 }
3319 else {
3320 flags = 0;
3321 goto restart; /* call sv_2bool_nomg */
3322 }
3323 return cBOOL(svb);
3324 }
fabdb6c0
AL
3325 }
3326 return SvRV(sv) != 0;
a0d0e21e 3327 }
85b7d9b3
FC
3328 if (isREGEXP(sv))
3329 return
3330 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
4bac9ae4 3331 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3332}
3333
c461cf8f
JH
3334/*
3335=for apidoc sv_utf8_upgrade
3336
78ea37eb 3337Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3338Forces the SV to string form if it is not already.
2bbc8d55 3339Will C<mg_get> on C<sv> if appropriate.
4411f3b6 3340Always sets the SvUTF8 flag to avoid future validity checks even
2bbc8d55
SP
3341if the whole string is the same in UTF-8 as not.
3342Returns the number of bytes in the converted string
c461cf8f 3343
0efd0472 3344This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3345use the Encode extension for that.
3346
fe749c9a
KW
3347=for apidoc sv_utf8_upgrade_nomg
3348
fde67290 3349Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
fe749c9a 3350
8d6d96c1
HS
3351=for apidoc sv_utf8_upgrade_flags
3352
78ea37eb 3353Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3354Forces the SV to string form if it is not already.
8d6d96c1 3355Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3356if all the bytes are invariant in UTF-8.
3357If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55 3358will C<mg_get> on C<sv> if appropriate, else not.
2a590426
KW
3359
3360If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3361will expand when converted to UTF-8, and skips the extra work of checking for
3362that. Typically this flag is used by a routine that has already parsed the
3363string and found such characters, and passes this information on so that the
3364work doesn't have to be repeated.
3365
3366Returns the number of bytes in the converted string.
8d6d96c1 3367
0efd0472 3368This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3369use the Encode extension for that.
3370
2a590426 3371=for apidoc sv_utf8_upgrade_flags_grow
b3ab6785 3372
2a590426
KW
3373Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3374the number of unused bytes the string of 'sv' is guaranteed to have free after
3375it upon return. This allows the caller to reserve extra space that it intends
3376to fill, to avoid extra grows.
b3ab6785 3377
2a590426
KW
3378C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3379are implemented in terms of this function.
3380
3381Returns the number of bytes in the converted string (not including the spares).
3382
3383=cut
b3ab6785
KW
3384
3385(One might think that the calling routine could pass in the position of the
2a590426
KW
3386first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3387have to be found again. But that is not the case, because typically when the
3388caller is likely to use this flag, it won't be calling this routine unless it
3389finds something that won't fit into a byte. Otherwise it tries to not upgrade
3390and just use bytes. But some things that do fit into a byte are variants in
3391utf8, and the caller may not have been keeping track of these.)
b3ab6785 3392
6602b933
KW
3393If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3394C<NUL> isn't guaranteed due to having other routines do the work in some input
3395cases, or if the input is already flagged as being in utf8.
b3ab6785
KW
3396
3397The speed of this could perhaps be improved for many cases if someone wanted to
3398write a fast function that counts the number of variant characters in a string,
3399especially if it could return the position of the first one.
3400
8d6d96c1
HS
3401*/
3402
3403STRLEN
5aaab254 3404Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3405{
b3ab6785 3406 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3407
808c356f
RGS
3408 if (sv == &PL_sv_undef)
3409 return 0;
892f9127 3410 if (!SvPOK_nog(sv)) {
e0e62c2a 3411 STRLEN len = 0;
d52b7888
NC
3412 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3413 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3414 if (SvUTF8(sv)) {
3415 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3416 return len;
b3ab6785 3417 }
d52b7888 3418 } else {
33fb6f35 3419 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3420 }
e0e62c2a 3421 }
4411f3b6 3422
f5cee72b 3423 if (SvUTF8(sv)) {
b3ab6785 3424 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3425 return SvCUR(sv);
f5cee72b 3426 }
5fec3b1d 3427
765f542d 3428 if (SvIsCOW(sv)) {
c56ed9f6 3429 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3430 }
3431
b3ab6785 3432 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
799ef3cb 3433 sv_recode_to_utf8(sv, PL_encoding);
b3ab6785
KW
3434 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3435 return SvCUR(sv);
3436 }
3437
4e93345f
KW
3438 if (SvCUR(sv) == 0) {
3439 if (extra) SvGROW(sv, extra);
3440 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3441 /* This function could be much more efficient if we
2bbc8d55 3442 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3443 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3444 * make the loop as fast as possible (although there are certainly ways
3445 * to speed this up, eg. through vectorization) */
3446 U8 * s = (U8 *) SvPVX_const(sv);
3447 U8 * e = (U8 *) SvEND(sv);
3448 U8 *t = s;
3449 STRLEN two_byte_count = 0;
c4e7c712 3450
b3ab6785
KW
3451 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3452
3453 /* See if really will need to convert to utf8. We mustn't rely on our
3454 * incoming SV being well formed and having a trailing '\0', as certain
3455 * code in pp_formline can send us partially built SVs. */
3456
c4e7c712 3457 while (t < e) {
53c1dcc0 3458 const U8 ch = *t++;
6f2d5cbc 3459 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
b3ab6785
KW
3460
3461 t--; /* t already incremented; re-point to first variant */
3462 two_byte_count = 1;
3463 goto must_be_utf8;
c4e7c712 3464 }
b3ab6785
KW
3465
3466 /* utf8 conversion not needed because all are invariants. Mark as
3467 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3468 SvUTF8_on(sv);
7f0bfbea 3469 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3470 return SvCUR(sv);
3471
3472must_be_utf8:
3473
3474 /* Here, the string should be converted to utf8, either because of an
3475 * input flag (two_byte_count = 0), or because a character that
3476 * requires 2 bytes was found (two_byte_count = 1). t points either to
3477 * the beginning of the string (if we didn't examine anything), or to
3478 * the first variant. In either case, everything from s to t - 1 will
3479 * occupy only 1 byte each on output.
3480 *
3481 * There are two main ways to convert. One is to create a new string
3482 * and go through the input starting from the beginning, appending each
3483 * converted value onto the new string as we go along. It's probably
3484 * best to allocate enough space in the string for the worst possible
3485 * case rather than possibly running out of space and having to
3486 * reallocate and then copy what we've done so far. Since everything
3487 * from s to t - 1 is invariant, the destination can be initialized
3488 * with these using a fast memory copy
3489 *
3490 * The other way is to figure out exactly how big the string should be
3491 * by parsing the entire input. Then you don't have to make it big
3492 * enough to handle the worst possible case, and more importantly, if
3493 * the string you already have is large enough, you don't have to
3494 * allocate a new string, you can copy the last character in the input
3495 * string to the final position(s) that will be occupied by the
3496 * converted string and go backwards, stopping at t, since everything
3497 * before that is invariant.
3498 *
3499 * There are advantages and disadvantages to each method.
3500 *
3501 * In the first method, we can allocate a new string, do the memory
3502 * copy from the s to t - 1, and then proceed through the rest of the
3503 * string byte-by-byte.
3504 *
3505 * In the second method, we proceed through the rest of the input
3506 * string just calculating how big the converted string will be. Then
3507 * there are two cases:
3508 * 1) if the string has enough extra space to handle the converted
3509 * value. We go backwards through the string, converting until we
3510 * get to the position we are at now, and then stop. If this
3511 * position is far enough along in the string, this method is
3512 * faster than the other method. If the memory copy were the same
3513 * speed as the byte-by-byte loop, that position would be about
3514 * half-way, as at the half-way mark, parsing to the end and back
3515 * is one complete string's parse, the same amount as starting
3516 * over and going all the way through. Actually, it would be
3517 * somewhat less than half-way, as it's faster to just count bytes
3518 * than to also copy, and we don't have the overhead of allocating
3519 * a new string, changing the scalar to use it, and freeing the
3520 * existing one. But if the memory copy is fast, the break-even
3521 * point is somewhere after half way. The counting loop could be
3522 * sped up by vectorization, etc, to move the break-even point
3523 * further towards the beginning.
3524 * 2) if the string doesn't have enough space to handle the converted
3525 * value. A new string will have to be allocated, and one might
3526 * as well, given that, start from the beginning doing the first
3527 * method. We've spent extra time parsing the string and in
3528 * exchange all we've gotten is that we know precisely how big to
3529 * make the new one. Perl is more optimized for time than space,
3530 * so this case is a loser.
3531 * So what I've decided to do is not use the 2nd method unless it is
3532 * guaranteed that a new string won't have to be allocated, assuming
3533 * the worst case. I also decided not to put any more conditions on it
3534 * than this, for now. It seems likely that, since the worst case is
3535 * twice as big as the unknown portion of the string (plus 1), we won't
3536 * be guaranteed enough space, causing us to go to the first method,
3537 * unless the string is short, or the first variant character is near
3538 * the end of it. In either of these cases, it seems best to use the
3539 * 2nd method. The only circumstance I can think of where this would
3540 * be really slower is if the string had once had much more data in it
3541 * than it does now, but there is still a substantial amount in it */
3542
3543 {
3544 STRLEN invariant_head = t - s;
3545 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3546 if (SvLEN(sv) < size) {
3547
3548 /* Here, have decided to allocate a new string */
3549
3550 U8 *dst;
3551 U8 *d;
3552
3553 Newx(dst, size, U8);
3554
3555 /* If no known invariants at the beginning of the input string,
3556 * set so starts from there. Otherwise, can use memory copy to
3557 * get up to where we are now, and then start from here */
3558
5b26a7b3 3559 if (invariant_head == 0) {
b3ab6785
KW
3560 d = dst;
3561 } else {
3562 Copy(s, dst, invariant_head, char);
3563 d = dst + invariant_head;
3564 }
3565
3566 while (t < e) {
55d09dc8
KW
3567 append_utf8_from_native_byte(*t, &d);
3568 t++;
b3ab6785
KW
3569 }
3570 *d = '\0';
3571 SvPV_free(sv); /* No longer using pre-existing string */
3572 SvPV_set(sv, (char*)dst);
3573 SvCUR_set(sv, d - dst);
3574 SvLEN_set(sv, size);
3575 } else {
3576
3577 /* Here, have decided to get the exact size of the string.
3578 * Currently this happens only when we know that there is
3579 * guaranteed enough space to fit the converted string, so
3580 * don't have to worry about growing. If two_byte_count is 0,
3581 * then t points to the first byte of the string which hasn't
3582 * been examined yet. Otherwise two_byte_count is 1, and t
3583 * points to the first byte in the string that will expand to
3584 * two. Depending on this, start examining at t or 1 after t.
3585 * */
3586
3587 U8 *d = t + two_byte_count;
3588
3589
3590 /* Count up the remaining bytes that expand to two */
3591
3592 while (d < e) {
3593 const U8 chr = *d++;
6f2d5cbc 3594 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
b3ab6785
KW
3595 }
3596
3597 /* The string will expand by just the number of bytes that
3598 * occupy two positions. But we are one afterwards because of
3599 * the increment just above. This is the place to put the
3600 * trailing NUL, and to set the length before we decrement */
3601
3602 d += two_byte_count;
3603 SvCUR_set(sv, d - s);
3604 *d-- = '\0';
3605
3606
3607 /* Having decremented d, it points to the position to put the
3608 * very last byte of the expanded string. Go backwards through
3609 * the string, copying and expanding as we go, stopping when we
3610 * get to the part that is invariant the rest of the way down */
3611
3612 e--;
3613 while (e >= t) {
6f2d5cbc 3614 if (NATIVE_BYTE_IS_INVARIANT(*e)) {