This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
move SET_SVANY_FOR_BODYLESS_IV() from sv.c to sv.h
[perl5.git] / sv.c
CommitLineData
5356d32e 1/* sv.c
79072805 2 *
1129b882 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
83706693
RGS
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
5 * and others
79072805
LW
6 *
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
9 *
4ac71550
TC
10 */
11
12/*
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
14 * --Pippin
15 *
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
17 */
18
19/*
645c22ef
DM
20 *
21 *
5e045b90
AMS
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
27 * in the pp*.c files.
79072805
LW
28 */
29
30#include "EXTERN.h"
864dbfa3 31#define PERL_IN_SV_C
79072805 32#include "perl.h"
d2f185dc 33#include "regcomp.h"
9d9a81f0
CB
34#ifdef __VMS
35# include <rms.h>
36#endif
79072805 37
2f8ed50e
OS
38#ifdef __Lynx__
39/* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
41#endif
42
a4eca1d4
JH
43#ifdef USE_QUADMATH
44# define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
46#else
47# define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
49#endif
50
9f53080a 51#ifndef SV_COW_THRESHOLD
e8c6a474 52# define SV_COW_THRESHOLD 0 /* COW iff len > K */
9f53080a
FC
53#endif
54#ifndef SV_COWBUF_THRESHOLD
e8c6a474 55# define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
9f53080a
FC
56#endif
57#ifndef SV_COW_MAX_WASTE_THRESHOLD
e8c6a474 58# define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
9f53080a
FC
59#endif
60#ifndef SV_COWBUF_WASTE_THRESHOLD
e8c6a474 61# define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
9f53080a
FC
62#endif
63#ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
e8c6a474 64# define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
9f53080a
FC
65#endif
66#ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
e8c6a474 67# define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
e8c6a474
YO
68#endif
69/* Work around compiler warnings about unsigned >= THRESHOLD when thres-
70 hold is 0. */
71#if SV_COW_THRESHOLD
72# define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
73#else
74# define GE_COW_THRESHOLD(cur) 1
75#endif
76#if SV_COWBUF_THRESHOLD
77# define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
78#else
79# define GE_COWBUF_THRESHOLD(cur) 1
80#endif
81#if SV_COW_MAX_WASTE_THRESHOLD
82# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
83#else
84# define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
85#endif
86#if SV_COWBUF_WASTE_THRESHOLD
87# define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
88#else
89# define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
90#endif
91#if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
93#else
94# define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
95#endif
96#if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
98#else
99# define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
100#endif
101
102#define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
106)
107#define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
111)
cca0492e 112
e23c8137 113#ifdef PERL_UTF8_CACHE_ASSERT
ab455f60 114/* if adding more checks watch out for the following tests:
e23c8137
JH
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
117 * --jhi
118 */
6f207bd3 119# define ASSERT_UTF8_CACHE(cache) \
ab455f60
NC
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
123 } STMT_END
e23c8137 124#else
6f207bd3 125# define ASSERT_UTF8_CACHE(cache) NOOP
e23c8137
JH
126#endif
127
645c22ef
DM
128/* ============================================================================
129
130=head1 Allocation and deallocation of SVs.
d2a0f284
JC
131An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
132sv, av, hv...) contains type and reference count information, and for
133many types, a pointer to the body (struct xrv, xpv, xpviv...), which
134contains fields specific to each type. Some types store all they need
135in the head, so don't have a body.
136
486ec47a 137In all but the most memory-paranoid configurations (ex: PURIFY), heads
d2a0f284
JC
138and bodies are allocated out of arenas, which by default are
139approximately 4K chunks of memory parcelled up into N heads or bodies.
93e68bfb
JC
140Sv-bodies are allocated by their sv-type, guaranteeing size
141consistency needed to allocate safely from arrays.
142
d2a0f284
JC
143For SV-heads, the first slot in each arena is reserved, and holds a
144link to the next arena, some flags, and a note of the number of slots.
145Snaked through each arena chain is a linked list of free items; when
146this becomes empty, an extra arena is allocated and divided up into N
147items which are threaded into the free list.
148
149SV-bodies are similar, but they use arena-sets by default, which
150separate the link and info from the arena itself, and reclaim the 1st
151slot in the arena. SV-bodies are further described later.
645c22ef
DM
152
153The following global variables are associated with arenas:
154
7fefc6c1
KW
155 PL_sv_arenaroot pointer to list of SV arenas
156 PL_sv_root pointer to list of free SV structures
645c22ef 157
7fefc6c1
KW
158 PL_body_arenas head of linked-list of body arenas
159 PL_body_roots[] array of pointers to list of free bodies of svtype
160 arrays are indexed by the svtype needed
93e68bfb 161
d2a0f284
JC
162A few special SV heads are not allocated from an arena, but are
163instead directly created in the interpreter structure, eg PL_sv_undef.
93e68bfb
JC
164The size of arenas can be changed from the default by setting
165PERL_ARENA_SIZE appropriately at compile time.
645c22ef
DM
166
167The SV arena serves the secondary purpose of allowing still-live SVs
168to be located and destroyed during final cleanup.
169
170At the lowest level, the macros new_SV() and del_SV() grab and free
171an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
172to return the SV to the free list with error checking.) new_SV() calls
173more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
174SVs in the free list have their SvTYPE field set to all ones.
175
ff276b08 176At the time of very final cleanup, sv_free_arenas() is called from
645c22ef 177perl_destruct() to physically free all the arenas allocated since the
6a93a7e5 178start of the interpreter.
645c22ef 179
645c22ef
DM
180The function visit() scans the SV arenas list, and calls a specified
181function for each SV it finds which is still live - ie which has an SvTYPE
182other than all 1's, and a non-zero SvREFCNT. visit() is used by the
183following functions (specified as [function that calls visit()] / [function
184called by visit() for each SV]):
185
186 sv_report_used() / do_report_used()
f2524eef 187 dump all remaining SVs (debugging aid)
645c22ef 188
e4487e9b 189 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
e76981f9 190 do_clean_named_io_objs(),do_curse()
645c22ef 191 Attempt to free all objects pointed to by RVs,
e76981f9
FC
192 try to do the same for all objects indir-
193 ectly referenced by typeglobs too, and
194 then do a final sweep, cursing any
195 objects that remain. Called once from
645c22ef
DM
196 perl_destruct(), prior to calling sv_clean_all()
197 below.
198
199 sv_clean_all() / do_clean_all()
200 SvREFCNT_dec(sv) each remaining SV, possibly
201 triggering an sv_free(). It also sets the
202 SVf_BREAK flag on the SV to indicate that the
203 refcnt has been artificially lowered, and thus
204 stopping sv_free() from giving spurious warnings
205 about SVs which unexpectedly have a refcnt
206 of zero. called repeatedly from perl_destruct()
207 until there are no SVs left.
208
93e68bfb 209=head2 Arena allocator API Summary
645c22ef
DM
210
211Private API to rest of sv.c
212
213 new_SV(), del_SV(),
214
df0f0429 215 new_XPVNV(), del_XPVGV(),
645c22ef
DM
216 etc
217
218Public API:
219
8cf8f3d1 220 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
645c22ef 221
645c22ef
DM
222=cut
223
3e8320cc 224 * ========================================================================= */
645c22ef 225
4561caa4
CS
226/*
227 * "A time to plant, and a time to uproot what was planted..."
228 */
229
d7a2c63c
MHM
230#ifdef PERL_MEM_LOG
231# define MEM_LOG_NEW_SV(sv, file, line, func) \
232 Perl_mem_log_new_sv(sv, file, line, func)
233# define MEM_LOG_DEL_SV(sv, file, line, func) \
234 Perl_mem_log_del_sv(sv, file, line, func)
235#else
236# define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
237# define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
238#endif
239
fd0854ff 240#ifdef DEBUG_LEAKING_SCALARS
484e6108
FC
241# define FREE_SV_DEBUG_FILE(sv) STMT_START { \
242 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
243 } STMT_END
d7a2c63c
MHM
244# define DEBUG_SV_SERIAL(sv) \
245 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
246 PTR2UV(sv), (long)(sv)->sv_debug_serial))
fd0854ff
DM
247#else
248# define FREE_SV_DEBUG_FILE(sv)
d7a2c63c 249# define DEBUG_SV_SERIAL(sv) NOOP
fd0854ff
DM
250#endif
251
48614a46
NC
252#ifdef PERL_POISON
253# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
daba3364 254# define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
48614a46
NC
255/* Whilst I'd love to do this, it seems that things like to check on
256 unreferenced scalars
ce5dbf61 257# define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
48614a46 258*/
ce5dbf61 259# define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
7e337ee0 260 PoisonNew(&SvREFCNT(sv), 1, U32)
48614a46
NC
261#else
262# define SvARENA_CHAIN(sv) SvANY(sv)
3eef1deb 263# define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
ce5dbf61 264# define POISON_SV_HEAD(sv)
48614a46
NC
265#endif
266
990198f0
DM
267/* Mark an SV head as unused, and add to free list.
268 *
269 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
270 * its refcount artificially decremented during global destruction, so
271 * there may be dangling pointers to it. The last thing we want in that
272 * case is for it to be reused. */
273
053fc874
GS
274#define plant_SV(p) \
275 STMT_START { \
990198f0 276 const U32 old_flags = SvFLAGS(p); \
d7a2c63c
MHM
277 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
278 DEBUG_SV_SERIAL(p); \
fd0854ff 279 FREE_SV_DEBUG_FILE(p); \
ce5dbf61 280 POISON_SV_HEAD(p); \
053fc874 281 SvFLAGS(p) = SVTYPEMASK; \
990198f0 282 if (!(old_flags & SVf_BREAK)) { \
3eef1deb 283 SvARENA_CHAIN_SET(p, PL_sv_root); \
990198f0
DM
284 PL_sv_root = (p); \
285 } \
053fc874
GS
286 --PL_sv_count; \
287 } STMT_END
a0d0e21e 288
053fc874
GS
289#define uproot_SV(p) \
290 STMT_START { \
291 (p) = PL_sv_root; \
daba3364 292 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
053fc874
GS
293 ++PL_sv_count; \
294 } STMT_END
295
645c22ef 296
cac9b346
NC
297/* make some more SVs by adding another arena */
298
cac9b346
NC
299STATIC SV*
300S_more_sv(pTHX)
301{
302 SV* sv;
9a87bd09
NC
303 char *chunk; /* must use New here to match call to */
304 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
305 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
cac9b346
NC
306 uproot_SV(sv);
307 return sv;
308}
309
645c22ef
DM
310/* new_SV(): return a new, empty SV head */
311
eba0f806
DM
312#ifdef DEBUG_LEAKING_SCALARS
313/* provide a real function for a debugger to play with */
314STATIC SV*
d7a2c63c 315S_new_SV(pTHX_ const char *file, int line, const char *func)
eba0f806
DM
316{
317 SV* sv;
318
eba0f806
DM
319 if (PL_sv_root)
320 uproot_SV(sv);
321 else
cac9b346 322 sv = S_more_sv(aTHX);
eba0f806
DM
323 SvANY(sv) = 0;
324 SvREFCNT(sv) = 1;
325 SvFLAGS(sv) = 0;
fd0854ff 326 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
e385c3bf
DM
327 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
328 ? PL_parser->copline
329 : PL_curcop
f24aceb1
DM
330 ? CopLINE(PL_curcop)
331 : 0
e385c3bf 332 );
fd0854ff 333 sv->sv_debug_inpad = 0;
cd676548 334 sv->sv_debug_parent = NULL;
484e6108 335 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
d7a2c63c
MHM
336
337 sv->sv_debug_serial = PL_sv_serial++;
338
339 MEM_LOG_NEW_SV(sv, file, line, func);
340 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
341 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
342
eba0f806
DM
343 return sv;
344}
d7a2c63c 345# define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
eba0f806
DM
346
347#else
348# define new_SV(p) \
053fc874 349 STMT_START { \
053fc874
GS
350 if (PL_sv_root) \
351 uproot_SV(p); \
352 else \
cac9b346 353 (p) = S_more_sv(aTHX); \
053fc874
GS
354 SvANY(p) = 0; \
355 SvREFCNT(p) = 1; \
356 SvFLAGS(p) = 0; \
d7a2c63c 357 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
053fc874 358 } STMT_END
eba0f806 359#endif
463ee0b2 360
645c22ef
DM
361
362/* del_SV(): return an empty SV head to the free list */
363
a0d0e21e 364#ifdef DEBUGGING
4561caa4 365
053fc874
GS
366#define del_SV(p) \
367 STMT_START { \
aea4f609 368 if (DEBUG_D_TEST) \
053fc874
GS
369 del_sv(p); \
370 else \
371 plant_SV(p); \
053fc874 372 } STMT_END
a0d0e21e 373
76e3520e 374STATIC void
cea2e8a9 375S_del_sv(pTHX_ SV *p)
463ee0b2 376{
7918f24d
NC
377 PERL_ARGS_ASSERT_DEL_SV;
378
aea4f609 379 if (DEBUG_D_TEST) {
4633a7c4 380 SV* sva;
a3b680e6 381 bool ok = 0;
daba3364 382 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
53c1dcc0
AL
383 const SV * const sv = sva + 1;
384 const SV * const svend = &sva[SvREFCNT(sva)];
c0ff570e 385 if (p >= sv && p < svend) {
a0d0e21e 386 ok = 1;
c0ff570e
NC
387 break;
388 }
a0d0e21e
LW
389 }
390 if (!ok) {
9b387841
NC
391 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
392 "Attempt to free non-arena SV: 0x%"UVxf
393 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
a0d0e21e
LW
394 return;
395 }
396 }
4561caa4 397 plant_SV(p);
463ee0b2 398}
a0d0e21e 399
4561caa4
CS
400#else /* ! DEBUGGING */
401
402#define del_SV(p) plant_SV(p)
403
404#endif /* DEBUGGING */
463ee0b2 405
645c22ef
DM
406
407/*
ccfc67b7
JH
408=head1 SV Manipulation Functions
409
645c22ef
DM
410=for apidoc sv_add_arena
411
412Given a chunk of memory, link it to the head of the list of arenas,
413and split it into a list of free SVs.
414
415=cut
416*/
417
d2bd4e7f
NC
418static void
419S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
463ee0b2 420{
daba3364 421 SV *const sva = MUTABLE_SV(ptr);
eb578fdb
KW
422 SV* sv;
423 SV* svend;
4633a7c4 424
7918f24d
NC
425 PERL_ARGS_ASSERT_SV_ADD_ARENA;
426
4633a7c4 427 /* The first SV in an arena isn't an SV. */
3280af22 428 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
4633a7c4
LW
429 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
430 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
431
3280af22
NIS
432 PL_sv_arenaroot = sva;
433 PL_sv_root = sva + 1;
4633a7c4
LW
434
435 svend = &sva[SvREFCNT(sva) - 1];
436 sv = sva + 1;
463ee0b2 437 while (sv < svend) {
3eef1deb 438 SvARENA_CHAIN_SET(sv, (sv + 1));
03e36789 439#ifdef DEBUGGING
978b032e 440 SvREFCNT(sv) = 0;
03e36789 441#endif
4b69cbe3 442 /* Must always set typemask because it's always checked in on cleanup
03e36789 443 when the arenas are walked looking for objects. */
8990e307 444 SvFLAGS(sv) = SVTYPEMASK;
463ee0b2
LW
445 sv++;
446 }
3eef1deb 447 SvARENA_CHAIN_SET(sv, 0);
03e36789
NC
448#ifdef DEBUGGING
449 SvREFCNT(sv) = 0;
450#endif
4633a7c4
LW
451 SvFLAGS(sv) = SVTYPEMASK;
452}
453
055972dc
DM
454/* visit(): call the named function for each non-free SV in the arenas
455 * whose flags field matches the flags/mask args. */
645c22ef 456
5226ed68 457STATIC I32
de37a194 458S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
8990e307 459{
4633a7c4 460 SV* sva;
5226ed68 461 I32 visited = 0;
8990e307 462
7918f24d
NC
463 PERL_ARGS_ASSERT_VISIT;
464
daba3364 465 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
eb578fdb
KW
466 const SV * const svend = &sva[SvREFCNT(sva)];
467 SV* sv;
4561caa4 468 for (sv = sva + 1; sv < svend; ++sv) {
e4787c0c 469 if (SvTYPE(sv) != (svtype)SVTYPEMASK
055972dc
DM
470 && (sv->sv_flags & mask) == flags
471 && SvREFCNT(sv))
472 {
942481a7 473 (*f)(aTHX_ sv);
5226ed68
JH
474 ++visited;
475 }
8990e307
LW
476 }
477 }
5226ed68 478 return visited;
8990e307
LW
479}
480
758a08c3
JH
481#ifdef DEBUGGING
482
645c22ef
DM
483/* called by sv_report_used() for each live SV */
484
485static void
5fa45a31 486do_report_used(pTHX_ SV *const sv)
645c22ef 487{
e4787c0c 488 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
645c22ef
DM
489 PerlIO_printf(Perl_debug_log, "****\n");
490 sv_dump(sv);
491 }
492}
758a08c3 493#endif
645c22ef
DM
494
495/*
496=for apidoc sv_report_used
497
fde67290 498Dump the contents of all SVs not yet freed (debugging aid).
645c22ef
DM
499
500=cut
501*/
502
8990e307 503void
864dbfa3 504Perl_sv_report_used(pTHX)
4561caa4 505{
ff270d3a 506#ifdef DEBUGGING
055972dc 507 visit(do_report_used, 0, 0);
96a5add6
AL
508#else
509 PERL_UNUSED_CONTEXT;
ff270d3a 510#endif
4561caa4
CS
511}
512
645c22ef
DM
513/* called by sv_clean_objs() for each live SV */
514
515static void
de37a194 516do_clean_objs(pTHX_ SV *const ref)
645c22ef 517{
ea724faa
NC
518 assert (SvROK(ref));
519 {
823a54a3
AL
520 SV * const target = SvRV(ref);
521 if (SvOBJECT(target)) {
522 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
523 if (SvWEAKREF(ref)) {
524 sv_del_backref(target, ref);
525 SvWEAKREF_off(ref);
526 SvRV_set(ref, NULL);
527 } else {
528 SvROK_off(ref);
529 SvRV_set(ref, NULL);
fc2b2dca 530 SvREFCNT_dec_NN(target);
823a54a3 531 }
645c22ef
DM
532 }
533 }
645c22ef
DM
534}
535
645c22ef 536
e4487e9b
DM
537/* clear any slots in a GV which hold objects - except IO;
538 * called by sv_clean_objs() for each live GV */
539
645c22ef 540static void
f30de749 541do_clean_named_objs(pTHX_ SV *const sv)
645c22ef 542{
57ef47cc 543 SV *obj;
ea724faa 544 assert(SvTYPE(sv) == SVt_PVGV);
d011219a 545 assert(isGV_with_GP(sv));
57ef47cc
DM
546 if (!GvGP(sv))
547 return;
548
549 /* freeing GP entries may indirectly free the current GV;
550 * hold onto it while we mess with the GP slots */
551 SvREFCNT_inc(sv);
552
553 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
554 DEBUG_D((PerlIO_printf(Perl_debug_log,
555 "Cleaning named glob SV object:\n "), sv_dump(obj)));
556 GvSV(sv) = NULL;
fc2b2dca 557 SvREFCNT_dec_NN(obj);
57ef47cc
DM
558 }
559 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
560 DEBUG_D((PerlIO_printf(Perl_debug_log,
561 "Cleaning named glob AV object:\n "), sv_dump(obj)));
562 GvAV(sv) = NULL;
fc2b2dca 563 SvREFCNT_dec_NN(obj);
57ef47cc
DM
564 }
565 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
566 DEBUG_D((PerlIO_printf(Perl_debug_log,
567 "Cleaning named glob HV object:\n "), sv_dump(obj)));
568 GvHV(sv) = NULL;
fc2b2dca 569 SvREFCNT_dec_NN(obj);
57ef47cc
DM
570 }
571 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
572 DEBUG_D((PerlIO_printf(Perl_debug_log,
573 "Cleaning named glob CV object:\n "), sv_dump(obj)));
c43ae56f 574 GvCV_set(sv, NULL);
fc2b2dca 575 SvREFCNT_dec_NN(obj);
57ef47cc 576 }
fc2b2dca 577 SvREFCNT_dec_NN(sv); /* undo the inc above */
e4487e9b
DM
578}
579
68b590d9 580/* clear any IO slots in a GV which hold objects (except stderr, defout);
e4487e9b
DM
581 * called by sv_clean_objs() for each live GV */
582
583static void
584do_clean_named_io_objs(pTHX_ SV *const sv)
585{
e4487e9b
DM
586 SV *obj;
587 assert(SvTYPE(sv) == SVt_PVGV);
588 assert(isGV_with_GP(sv));
68b590d9 589 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
e4487e9b
DM
590 return;
591
592 SvREFCNT_inc(sv);
57ef47cc
DM
593 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
594 DEBUG_D((PerlIO_printf(Perl_debug_log,
595 "Cleaning named glob IO object:\n "), sv_dump(obj)));
596 GvIOp(sv) = NULL;
fc2b2dca 597 SvREFCNT_dec_NN(obj);
645c22ef 598 }
fc2b2dca 599 SvREFCNT_dec_NN(sv); /* undo the inc above */
645c22ef 600}
645c22ef 601
4155e4fe
FC
602/* Void wrapper to pass to visit() */
603static void
604do_curse(pTHX_ SV * const sv) {
c2910e6c
FC
605 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
606 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
4155e4fe
FC
607 return;
608 (void)curse(sv, 0);
609}
610
645c22ef
DM
611/*
612=for apidoc sv_clean_objs
613
fde67290 614Attempt to destroy all objects not yet freed.
645c22ef
DM
615
616=cut
617*/
618
4561caa4 619void
864dbfa3 620Perl_sv_clean_objs(pTHX)
4561caa4 621{
68b590d9 622 GV *olddef, *olderr;
3280af22 623 PL_in_clean_objs = TRUE;
055972dc 624 visit(do_clean_objs, SVf_ROK, SVf_ROK);
e4487e9b
DM
625 /* Some barnacles may yet remain, clinging to typeglobs.
626 * Run the non-IO destructors first: they may want to output
627 * error messages, close files etc */
d011219a 628 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
e4487e9b 629 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
4155e4fe
FC
630 /* And if there are some very tenacious barnacles clinging to arrays,
631 closures, or what have you.... */
632 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
68b590d9
DM
633 olddef = PL_defoutgv;
634 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
635 if (olddef && isGV_with_GP(olddef))
636 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
637 olderr = PL_stderrgv;
638 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
639 if (olderr && isGV_with_GP(olderr))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
641 SvREFCNT_dec(olddef);
3280af22 642 PL_in_clean_objs = FALSE;
4561caa4
CS
643}
644
645c22ef
DM
645/* called by sv_clean_all() for each live SV */
646
647static void
de37a194 648do_clean_all(pTHX_ SV *const sv)
645c22ef 649{
daba3364 650 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
cddfcddc 651 /* don't clean pid table and strtab */
d17ea597 652 return;
cddfcddc 653 }
645c22ef
DM
654 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
655 SvFLAGS(sv) |= SVf_BREAK;
fc2b2dca 656 SvREFCNT_dec_NN(sv);
645c22ef
DM
657}
658
659/*
660=for apidoc sv_clean_all
661
662Decrement the refcnt of each remaining SV, possibly triggering a
fde67290 663cleanup. This function may have to be called multiple times to free
ff276b08 664SVs which are in complex self-referential hierarchies.
645c22ef
DM
665
666=cut
667*/
668
5226ed68 669I32
864dbfa3 670Perl_sv_clean_all(pTHX)
8990e307 671{
5226ed68 672 I32 cleaned;
3280af22 673 PL_in_clean_all = TRUE;
055972dc 674 cleaned = visit(do_clean_all, 0,0);
5226ed68 675 return cleaned;
8990e307 676}
463ee0b2 677
5e258f8c
JC
678/*
679 ARENASETS: a meta-arena implementation which separates arena-info
680 into struct arena_set, which contains an array of struct
681 arena_descs, each holding info for a single arena. By separating
682 the meta-info from the arena, we recover the 1st slot, formerly
683 borrowed for list management. The arena_set is about the size of an
39244528 684 arena, avoiding the needless malloc overhead of a naive linked-list.
5e258f8c
JC
685
686 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
687 memory in the last arena-set (1/2 on average). In trade, we get
688 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
d2a0f284 689 smaller types). The recovery of the wasted space allows use of
e15dad31
JC
690 small arenas for large, rare body types, by changing array* fields
691 in body_details_by_type[] below.
5e258f8c 692*/
5e258f8c 693struct arena_desc {
398c677b
NC
694 char *arena; /* the raw storage, allocated aligned */
695 size_t size; /* its size ~4k typ */
e5973ed5 696 svtype utype; /* bodytype stored in arena */
5e258f8c
JC
697};
698
e6148039
NC
699struct arena_set;
700
701/* Get the maximum number of elements in set[] such that struct arena_set
e15dad31 702 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
e6148039
NC
703 therefore likely to be 1 aligned memory page. */
704
705#define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
706 - 2 * sizeof(int)) / sizeof (struct arena_desc))
5e258f8c
JC
707
708struct arena_set {
709 struct arena_set* next;
0a848332
NC
710 unsigned int set_size; /* ie ARENAS_PER_SET */
711 unsigned int curr; /* index of next available arena-desc */
5e258f8c
JC
712 struct arena_desc set[ARENAS_PER_SET];
713};
714
645c22ef
DM
715/*
716=for apidoc sv_free_arenas
717
fde67290 718Deallocate the memory used by all arenas. Note that all the individual SV
645c22ef
DM
719heads and bodies within the arenas must already have been freed.
720
721=cut
7fefc6c1 722
645c22ef 723*/
4633a7c4 724void
864dbfa3 725Perl_sv_free_arenas(pTHX)
4633a7c4
LW
726{
727 SV* sva;
728 SV* svanext;
0a848332 729 unsigned int i;
4633a7c4
LW
730
731 /* Free arenas here, but be careful about fake ones. (We assume
732 contiguity of the fake ones with the corresponding real ones.) */
733
3280af22 734 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
daba3364 735 svanext = MUTABLE_SV(SvANY(sva));
4633a7c4 736 while (svanext && SvFAKE(svanext))
daba3364 737 svanext = MUTABLE_SV(SvANY(svanext));
4633a7c4
LW
738
739 if (!SvFAKE(sva))
1df70142 740 Safefree(sva);
4633a7c4 741 }
93e68bfb 742
5e258f8c 743 {
0a848332
NC
744 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
745
746 while (aroot) {
747 struct arena_set *current = aroot;
748 i = aroot->curr;
749 while (i--) {
5e258f8c
JC
750 assert(aroot->set[i].arena);
751 Safefree(aroot->set[i].arena);
752 }
0a848332
NC
753 aroot = aroot->next;
754 Safefree(current);
5e258f8c
JC
755 }
756 }
dc8220bf 757 PL_body_arenas = 0;
fdda85ca 758
0a848332
NC
759 i = PERL_ARENA_ROOTS_SIZE;
760 while (i--)
93e68bfb 761 PL_body_roots[i] = 0;
93e68bfb 762
3280af22
NIS
763 PL_sv_arenaroot = 0;
764 PL_sv_root = 0;
4633a7c4
LW
765}
766
bd81e77b
NC
767/*
768 Here are mid-level routines that manage the allocation of bodies out
769 of the various arenas. There are 5 kinds of arenas:
29489e7c 770
bd81e77b
NC
771 1. SV-head arenas, which are discussed and handled above
772 2. regular body arenas
773 3. arenas for reduced-size bodies
774 4. Hash-Entry arenas
29489e7c 775
bd81e77b
NC
776 Arena types 2 & 3 are chained by body-type off an array of
777 arena-root pointers, which is indexed by svtype. Some of the
778 larger/less used body types are malloced singly, since a large
779 unused block of them is wasteful. Also, several svtypes dont have
780 bodies; the data fits into the sv-head itself. The arena-root
781 pointer thus has a few unused root-pointers (which may be hijacked
782 later for arena types 4,5)
29489e7c 783
bd81e77b
NC
784 3 differs from 2 as an optimization; some body types have several
785 unused fields in the front of the structure (which are kept in-place
786 for consistency). These bodies can be allocated in smaller chunks,
787 because the leading fields arent accessed. Pointers to such bodies
788 are decremented to point at the unused 'ghost' memory, knowing that
789 the pointers are used with offsets to the real memory.
29489e7c 790
d2a0f284
JC
791
792=head1 SV-Body Allocation
793
f554dfc5
MH
794=cut
795
d2a0f284
JC
796Allocation of SV-bodies is similar to SV-heads, differing as follows;
797the allocation mechanism is used for many body types, so is somewhat
798more complicated, it uses arena-sets, and has no need for still-live
799SV detection.
800
801At the outermost level, (new|del)_X*V macros return bodies of the
802appropriate type. These macros call either (new|del)_body_type or
803(new|del)_body_allocated macro pairs, depending on specifics of the
804type. Most body types use the former pair, the latter pair is used to
805allocate body types with "ghost fields".
806
807"ghost fields" are fields that are unused in certain types, and
69ba284b 808consequently don't need to actually exist. They are declared because
d2a0f284
JC
809they're part of a "base type", which allows use of functions as
810methods. The simplest examples are AVs and HVs, 2 aggregate types
811which don't use the fields which support SCALAR semantics.
812
69ba284b 813For these types, the arenas are carved up into appropriately sized
d2a0f284
JC
814chunks, we thus avoid wasted memory for those unaccessed members.
815When bodies are allocated, we adjust the pointer back in memory by the
69ba284b 816size of the part not allocated, so it's as if we allocated the full
d2a0f284
JC
817structure. (But things will all go boom if you write to the part that
818is "not there", because you'll be overwriting the last members of the
819preceding structure in memory.)
820
69ba284b 821We calculate the correction using the STRUCT_OFFSET macro on the first
a05ea1cf 822member present. If the allocated structure is smaller (no initial NV
69ba284b
NC
823actually allocated) then the net effect is to subtract the size of the NV
824from the pointer, to return a new pointer as if an initial NV were actually
a05ea1cf 825allocated. (We were using structures named *_allocated for this, but
69ba284b
NC
826this turned out to be a subtle bug, because a structure without an NV
827could have a lower alignment constraint, but the compiler is allowed to
828optimised accesses based on the alignment constraint of the actual pointer
829to the full structure, for example, using a single 64 bit load instruction
830because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
d2a0f284 831
a05ea1cf 832This is the same trick as was used for NV and IV bodies. Ironically it
d2a0f284 833doesn't need to be used for NV bodies any more, because NV is now at
5b306eef
DD
834the start of the structure. IV bodies, and also in some builds NV bodies,
835don't need it either, because they are no longer allocated.
d2a0f284
JC
836
837In turn, the new_body_* allocators call S_new_body(), which invokes
838new_body_inline macro, which takes a lock, and takes a body off the
1e30fcd5 839linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
d2a0f284
JC
840necessary to refresh an empty list. Then the lock is released, and
841the body is returned.
842
99816f8d 843Perl_more_bodies allocates a new arena, and carves it up into an array of N
d2a0f284
JC
844bodies, which it strings into a linked list. It looks up arena-size
845and body-size from the body_details table described below, thus
846supporting the multiple body-types.
847
848If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
849the (new|del)_X*V macros are mapped directly to malloc/free.
850
d2a0f284
JC
851For each sv-type, struct body_details bodies_by_type[] carries
852parameters which control these aspects of SV handling:
853
854Arena_size determines whether arenas are used for this body type, and if
855so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
856zero, forcing individual mallocs and frees.
857
858Body_size determines how big a body is, and therefore how many fit into
859each arena. Offset carries the body-pointer adjustment needed for
69ba284b 860"ghost fields", and is used in *_allocated macros.
d2a0f284
JC
861
862But its main purpose is to parameterize info needed in
863Perl_sv_upgrade(). The info here dramatically simplifies the function
69ba284b 864vs the implementation in 5.8.8, making it table-driven. All fields
d2a0f284
JC
865are used for this, except for arena_size.
866
867For the sv-types that have no bodies, arenas are not used, so those
868PL_body_roots[sv_type] are unused, and can be overloaded. In
869something of a special case, SVt_NULL is borrowed for HE arenas;
c6f8b1d0 870PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
d2a0f284 871bodies_by_type[SVt_NULL] slot is not used, as the table is not
c6f8b1d0 872available in hv.c.
d2a0f284 873
29489e7c
DM
874*/
875
bd81e77b 876struct body_details {
0fb58b32 877 U8 body_size; /* Size to allocate */
10666ae3 878 U8 copy; /* Size of structure to copy (may be shorter) */
5b306eef
DD
879 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
880 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
881 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
882 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
883 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
884 U32 arena_size; /* Size of arena to allocate */
bd81e77b 885};
29489e7c 886
bd81e77b
NC
887#define HADNV FALSE
888#define NONV TRUE
29489e7c 889
d2a0f284 890
bd81e77b
NC
891#ifdef PURIFY
892/* With -DPURFIY we allocate everything directly, and don't use arenas.
893 This seems a rather elegant way to simplify some of the code below. */
894#define HASARENA FALSE
895#else
896#define HASARENA TRUE
897#endif
898#define NOARENA FALSE
29489e7c 899
d2a0f284
JC
900/* Size the arenas to exactly fit a given number of bodies. A count
901 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
902 simplifying the default. If count > 0, the arena is sized to fit
903 only that many bodies, allowing arenas to be used for large, rare
904 bodies (XPVFM, XPVIO) without undue waste. The arena size is
905 limited by PERL_ARENA_SIZE, so we can safely oversize the
906 declarations.
907 */
95db5f15
MB
908#define FIT_ARENA0(body_size) \
909 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
910#define FIT_ARENAn(count,body_size) \
911 ( count * body_size <= PERL_ARENA_SIZE) \
912 ? count * body_size \
913 : FIT_ARENA0 (body_size)
914#define FIT_ARENA(count,body_size) \
cd1dc8e2 915 (U32)(count \
95db5f15 916 ? FIT_ARENAn (count, body_size) \
cd1dc8e2 917 : FIT_ARENA0 (body_size))
d2a0f284 918
bd81e77b
NC
919/* Calculate the length to copy. Specifically work out the length less any
920 final padding the compiler needed to add. See the comment in sv_upgrade
921 for why copying the padding proved to be a bug. */
29489e7c 922
bd81e77b
NC
923#define copy_length(type, last_member) \
924 STRUCT_OFFSET(type, last_member) \
daba3364 925 + sizeof (((type*)SvANY((const SV *)0))->last_member)
29489e7c 926
bd81e77b 927static const struct body_details bodies_by_type[] = {
829cd18a
NC
928 /* HEs use this offset for their arena. */
929 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
d2a0f284 930
db93c0c4
NC
931 /* IVs are in the head, so the allocation size is 0. */
932 { 0,
d2a0f284 933 sizeof(IV), /* This is used to copy out the IV body. */
10666ae3 934 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
db93c0c4 935 NOARENA /* IVS don't need an arena */, 0
d2a0f284
JC
936 },
937
5b306eef
DD
938#if NVSIZE <= IVSIZE
939 { 0, sizeof(NV),
940 STRUCT_OFFSET(XPVNV, xnv_u),
941 SVt_NV, FALSE, HADNV, NOARENA, 0 },
942#else
6e128786
NC
943 { sizeof(NV), sizeof(NV),
944 STRUCT_OFFSET(XPVNV, xnv_u),
945 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
5b306eef 946#endif
d2a0f284 947
bc337e5c 948 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
949 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
950 + STRUCT_OFFSET(XPV, xpv_cur),
69ba284b 951 SVt_PV, FALSE, NONV, HASARENA,
889d28b2 952 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 953
d361b004
KW
954 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
955 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
956 + STRUCT_OFFSET(XPV, xpv_cur),
957 SVt_INVLIST, TRUE, NONV, HASARENA,
958 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
e94d9b54 959
bc337e5c 960 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
961 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
962 + STRUCT_OFFSET(XPV, xpv_cur),
963 SVt_PVIV, FALSE, NONV, HASARENA,
964 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 965
bc337e5c 966 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
889d28b2
NC
967 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
968 + STRUCT_OFFSET(XPV, xpv_cur),
969 SVt_PVNV, FALSE, HADNV, HASARENA,
970 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
d2a0f284 971
6e128786 972 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
d2a0f284 973 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
4df7f6af 974
601dfd0a
NC
975 { sizeof(regexp),
976 sizeof(regexp),
977 0,
ecff11eb 978 SVt_REGEXP, TRUE, NONV, HASARENA,
eaeb1e7f 979 FIT_ARENA(0, sizeof(regexp))
5c35adbb 980 },
4df7f6af 981
10666ae3 982 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
d2a0f284
JC
983 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
984
10666ae3 985 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
d2a0f284
JC
986 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
987
601dfd0a 988 { sizeof(XPVAV),
4f7003f5 989 copy_length(XPVAV, xav_alloc),
601dfd0a 990 0,
69ba284b 991 SVt_PVAV, TRUE, NONV, HASARENA,
601dfd0a 992 FIT_ARENA(0, sizeof(XPVAV)) },
d2a0f284 993
601dfd0a 994 { sizeof(XPVHV),
359164a0 995 copy_length(XPVHV, xhv_max),
601dfd0a 996 0,
69ba284b 997 SVt_PVHV, TRUE, NONV, HASARENA,
601dfd0a 998 FIT_ARENA(0, sizeof(XPVHV)) },
d2a0f284 999
601dfd0a
NC
1000 { sizeof(XPVCV),
1001 sizeof(XPVCV),
1002 0,
69ba284b 1003 SVt_PVCV, TRUE, NONV, HASARENA,
601dfd0a 1004 FIT_ARENA(0, sizeof(XPVCV)) },
69ba284b 1005
601dfd0a
NC
1006 { sizeof(XPVFM),
1007 sizeof(XPVFM),
1008 0,
69ba284b 1009 SVt_PVFM, TRUE, NONV, NOARENA,
601dfd0a 1010 FIT_ARENA(20, sizeof(XPVFM)) },
d2a0f284 1011
601dfd0a
NC
1012 { sizeof(XPVIO),
1013 sizeof(XPVIO),
1014 0,
b6f60916 1015 SVt_PVIO, TRUE, NONV, HASARENA,
601dfd0a 1016 FIT_ARENA(24, sizeof(XPVIO)) },
bd81e77b 1017};
29489e7c 1018
bd81e77b 1019#define new_body_allocated(sv_type) \
d2a0f284 1020 (void *)((char *)S_new_body(aTHX_ sv_type) \
bd81e77b 1021 - bodies_by_type[sv_type].offset)
29489e7c 1022
26359cfa
NC
1023/* return a thing to the free list */
1024
1025#define del_body(thing, root) \
1026 STMT_START { \
1027 void ** const thing_copy = (void **)thing; \
1028 *thing_copy = *root; \
1029 *root = (void*)thing_copy; \
1030 } STMT_END
29489e7c 1031
bd81e77b 1032#ifdef PURIFY
5b306eef
DD
1033#if !(NVSIZE <= IVSIZE)
1034# define new_XNV() safemalloc(sizeof(XPVNV))
1035#endif
beeec492
NC
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
5b306eef
DD
1043#if !(NVSIZE <= IVSIZE)
1044# define new_XNV() new_body_allocated(SVt_NV)
1045#endif
65ac1738 1046#define new_XPVNV() new_body_allocated(SVt_PVNV)
65ac1738 1047#define new_XPVMG() new_body_allocated(SVt_PVMG)
645c22ef 1048
26359cfa
NC
1049#define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1050 &PL_body_roots[SVt_PVGV])
1d7c1841 1051
bd81e77b 1052#endif /* PURIFY */
93e68bfb 1053
bd81e77b 1054/* no arena for you! */
93e68bfb 1055
bd81e77b 1056#define new_NOARENA(details) \
beeec492 1057 safemalloc((details)->body_size + (details)->offset)
bd81e77b 1058#define new_NOARENAZ(details) \
beeec492 1059 safecalloc((details)->body_size + (details)->offset, 1)
d2a0f284 1060
1e30fcd5
NC
1061void *
1062Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1063 const size_t arena_size)
d2a0f284 1064{
d2a0f284 1065 void ** const root = &PL_body_roots[sv_type];
99816f8d
NC
1066 struct arena_desc *adesc;
1067 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1068 unsigned int curr;
d2a0f284
JC
1069 char *start;
1070 const char *end;
02982131 1071 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
20b7effb
JH
1072#if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1073 dVAR;
1074#endif
0b2d3faa 1075#if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
23e9d66c
NC
1076 static bool done_sanity_check;
1077
0b2d3faa
JH
1078 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1079 * variables like done_sanity_check. */
10666ae3 1080 if (!done_sanity_check) {
ea471437 1081 unsigned int i = SVt_LAST;
10666ae3
NC
1082
1083 done_sanity_check = TRUE;
1084
1085 while (i--)
1086 assert (bodies_by_type[i].type == i);
1087 }
1088#endif
1089
02982131 1090 assert(arena_size);
23e9d66c 1091
99816f8d
NC
1092 /* may need new arena-set to hold new arena */
1093 if (!aroot || aroot->curr >= aroot->set_size) {
1094 struct arena_set *newroot;
1095 Newxz(newroot, 1, struct arena_set);
1096 newroot->set_size = ARENAS_PER_SET;
1097 newroot->next = aroot;
1098 aroot = newroot;
1099 PL_body_arenas = (void *) newroot;
1100 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1101 }
1102
1103 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1104 curr = aroot->curr++;
1105 adesc = &(aroot->set[curr]);
1106 assert(!adesc->arena);
1107
1108 Newx(adesc->arena, good_arena_size, char);
1109 adesc->size = good_arena_size;
1110 adesc->utype = sv_type;
1111 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1112 curr, (void*)adesc->arena, (UV)good_arena_size));
1113
1114 start = (char *) adesc->arena;
d2a0f284 1115
29657bb6
NC
1116 /* Get the address of the byte after the end of the last body we can fit.
1117 Remember, this is integer division: */
02982131 1118 end = start + good_arena_size / body_size * body_size;
d2a0f284 1119
486ec47a 1120 /* computed count doesn't reflect the 1st slot reservation */
d8fca402
NC
1121#if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1122 DEBUG_m(PerlIO_printf(Perl_debug_log,
1123 "arena %p end %p arena-size %d (from %d) type %d "
1124 "size %d ct %d\n",
02982131
NC
1125 (void*)start, (void*)end, (int)good_arena_size,
1126 (int)arena_size, sv_type, (int)body_size,
1127 (int)good_arena_size / (int)body_size));
d8fca402 1128#else
d2a0f284
JC
1129 DEBUG_m(PerlIO_printf(Perl_debug_log,
1130 "arena %p end %p arena-size %d type %d size %d ct %d\n",
6c9570dc 1131 (void*)start, (void*)end,
02982131
NC
1132 (int)arena_size, sv_type, (int)body_size,
1133 (int)good_arena_size / (int)body_size));
d8fca402 1134#endif
d2a0f284
JC
1135 *root = (void *)start;
1136
29657bb6
NC
1137 while (1) {
1138 /* Where the next body would start: */
d2a0f284 1139 char * const next = start + body_size;
29657bb6
NC
1140
1141 if (next >= end) {
1142 /* This is the last body: */
1143 assert(next == end);
1144
1145 *(void **)start = 0;
1146 return *root;
1147 }
1148
d2a0f284
JC
1149 *(void**) start = (void *)next;
1150 start = next;
1151 }
d2a0f284
JC
1152}
1153
1154/* grab a new thing from the free list, allocating more if necessary.
1155 The inline version is used for speed in hot routines, and the
1156 function using it serves the rest (unless PURIFY).
1157*/
1158#define new_body_inline(xpv, sv_type) \
1159 STMT_START { \
1160 void ** const r3wt = &PL_body_roots[sv_type]; \
11b79775 1161 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1e30fcd5 1162 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
02982131
NC
1163 bodies_by_type[sv_type].body_size,\
1164 bodies_by_type[sv_type].arena_size)); \
d2a0f284 1165 *(r3wt) = *(void**)(xpv); \
d2a0f284
JC
1166 } STMT_END
1167
1168#ifndef PURIFY
1169
1170STATIC void *
de37a194 1171S_new_body(pTHX_ const svtype sv_type)
d2a0f284 1172{
d2a0f284
JC
1173 void *xpv;
1174 new_body_inline(xpv, sv_type);
1175 return xpv;
1176}
1177
1178#endif
93e68bfb 1179
238b27b3
NC
1180static const struct body_details fake_rv =
1181 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1182
bd81e77b
NC
1183/*
1184=for apidoc sv_upgrade
93e68bfb 1185
bd81e77b
NC
1186Upgrade an SV to a more complex form. Generally adds a new body type to the
1187SV, then copies across as much information as possible from the old body.
9521ca61
FC
1188It croaks if the SV is already in a more complex form than requested. You
1189generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1190before calling C<sv_upgrade>, and hence does not croak. See also
fbe13c60 1191C<L</svtype>>.
93e68bfb 1192
bd81e77b 1193=cut
93e68bfb 1194*/
93e68bfb 1195
bd81e77b 1196void
5aaab254 1197Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
cac9b346 1198{
bd81e77b
NC
1199 void* old_body;
1200 void* new_body;
42d0e0b7 1201 const svtype old_type = SvTYPE(sv);
d2a0f284 1202 const struct body_details *new_type_details;
238b27b3 1203 const struct body_details *old_type_details
bd81e77b 1204 = bodies_by_type + old_type;
4df7f6af 1205 SV *referant = NULL;
cac9b346 1206
7918f24d
NC
1207 PERL_ARGS_ASSERT_SV_UPGRADE;
1208
1776cbe8
NC
1209 if (old_type == new_type)
1210 return;
1211
1212 /* This clause was purposefully added ahead of the early return above to
1213 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1214 inference by Nick I-S that it would fix other troublesome cases. See
1215 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1216
1217 Given that shared hash key scalars are no longer PVIV, but PV, there is
1218 no longer need to unshare so as to free up the IVX slot for its proper
1219 purpose. So it's safe to move the early return earlier. */
1220
093085a8 1221 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
bd81e77b
NC
1222 sv_force_normal_flags(sv, 0);
1223 }
cac9b346 1224
bd81e77b 1225 old_body = SvANY(sv);
de042e1d 1226
bd81e77b
NC
1227 /* Copying structures onto other structures that have been neatly zeroed
1228 has a subtle gotcha. Consider XPVMG
cac9b346 1229
bd81e77b
NC
1230 +------+------+------+------+------+-------+-------+
1231 | NV | CUR | LEN | IV | MAGIC | STASH |
1232 +------+------+------+------+------+-------+-------+
1233 0 4 8 12 16 20 24 28
645c22ef 1234
bd81e77b
NC
1235 where NVs are aligned to 8 bytes, so that sizeof that structure is
1236 actually 32 bytes long, with 4 bytes of padding at the end:
08742458 1237
bd81e77b
NC
1238 +------+------+------+------+------+-------+-------+------+
1239 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1240 +------+------+------+------+------+-------+-------+------+
1241 0 4 8 12 16 20 24 28 32
08742458 1242
bd81e77b 1243 so what happens if you allocate memory for this structure:
30f9da9e 1244
bd81e77b
NC
1245 +------+------+------+------+------+-------+-------+------+------+...
1246 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1247 +------+------+------+------+------+-------+-------+------+------+...
1248 0 4 8 12 16 20 24 28 32 36
bfc44f79 1249
bd81e77b
NC
1250 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1251 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1252 started out as zero once, but it's quite possible that it isn't. So now,
1253 rather than a nicely zeroed GP, you have it pointing somewhere random.
1254 Bugs ensue.
bfc44f79 1255
bd81e77b
NC
1256 (In fact, GP ends up pointing at a previous GP structure, because the
1257 principle cause of the padding in XPVMG getting garbage is a copy of
6c9e42f7
NC
1258 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1259 this happens to be moot because XPVGV has been re-ordered, with GP
1260 no longer after STASH)
30f9da9e 1261
bd81e77b
NC
1262 So we are careful and work out the size of used parts of all the
1263 structures. */
bfc44f79 1264
bd81e77b
NC
1265 switch (old_type) {
1266 case SVt_NULL:
1267 break;
1268 case SVt_IV:
4df7f6af
NC
1269 if (SvROK(sv)) {
1270 referant = SvRV(sv);
238b27b3
NC
1271 old_type_details = &fake_rv;
1272 if (new_type == SVt_NV)
1273 new_type = SVt_PVNV;
4df7f6af
NC
1274 } else {
1275 if (new_type < SVt_PVIV) {
1276 new_type = (new_type == SVt_NV)
1277 ? SVt_PVNV : SVt_PVIV;
1278 }
bd81e77b
NC
1279 }
1280 break;
1281 case SVt_NV:
1282 if (new_type < SVt_PVNV) {
1283 new_type = SVt_PVNV;
bd81e77b
NC
1284 }
1285 break;
bd81e77b
NC
1286 case SVt_PV:
1287 assert(new_type > SVt_PV);
6d59e610
LM
1288 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1289 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
bd81e77b
NC
1290 break;
1291 case SVt_PVIV:
1292 break;
1293 case SVt_PVNV:
1294 break;
1295 case SVt_PVMG:
1296 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1297 there's no way that it can be safely upgraded, because perl.c
1298 expects to Safefree(SvANY(PL_mess_sv)) */
1299 assert(sv != PL_mess_sv);
bd81e77b
NC
1300 break;
1301 default:
2439e033 1302 if (UNLIKELY(old_type_details->cant_upgrade))
c81225bc
NC
1303 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1304 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
bd81e77b 1305 }
3376de98 1306
2439e033 1307 if (UNLIKELY(old_type > new_type))
3376de98
NC
1308 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1309 (int)old_type, (int)new_type);
1310
2fa1109b 1311 new_type_details = bodies_by_type + new_type;
645c22ef 1312
bd81e77b
NC
1313 SvFLAGS(sv) &= ~SVTYPEMASK;
1314 SvFLAGS(sv) |= new_type;
932e9ff9 1315
ab4416c0
NC
1316 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1317 the return statements above will have triggered. */
1318 assert (new_type != SVt_NULL);
bd81e77b 1319 switch (new_type) {
bd81e77b
NC
1320 case SVt_IV:
1321 assert(old_type == SVt_NULL);
dc6369ef 1322 SET_SVANY_FOR_BODYLESS_IV(sv);
bd81e77b
NC
1323 SvIV_set(sv, 0);
1324 return;
1325 case SVt_NV:
1326 assert(old_type == SVt_NULL);
5b306eef 1327#if NVSIZE <= IVSIZE
dc6369ef 1328 SET_SVANY_FOR_BODYLESS_NV(sv);
5b306eef 1329#else
bd81e77b 1330 SvANY(sv) = new_XNV();
5b306eef 1331#endif
bd81e77b
NC
1332 SvNV_set(sv, 0);
1333 return;
bd81e77b 1334 case SVt_PVHV:
bd81e77b 1335 case SVt_PVAV:
d2a0f284 1336 assert(new_type_details->body_size);
c1ae03ae
NC
1337
1338#ifndef PURIFY
1339 assert(new_type_details->arena);
d2a0f284 1340 assert(new_type_details->arena_size);
c1ae03ae 1341 /* This points to the start of the allocated area. */
d2a0f284
JC
1342 new_body_inline(new_body, new_type);
1343 Zero(new_body, new_type_details->body_size, char);
c1ae03ae
NC
1344 new_body = ((char *)new_body) - new_type_details->offset;
1345#else
1346 /* We always allocated the full length item with PURIFY. To do this
1347 we fake things so that arena is false for all 16 types.. */
1348 new_body = new_NOARENAZ(new_type_details);
1349#endif
1350 SvANY(sv) = new_body;
1351 if (new_type == SVt_PVAV) {
1352 AvMAX(sv) = -1;
1353 AvFILLp(sv) = -1;
1354 AvREAL_only(sv);
64484faa 1355 if (old_type_details->body_size) {
ac572bf4
NC
1356 AvALLOC(sv) = 0;
1357 } else {
1358 /* It will have been zeroed when the new body was allocated.
1359 Lets not write to it, in case it confuses a write-back
1360 cache. */
1361 }
78ac7dd9
NC
1362 } else {
1363 assert(!SvOK(sv));
1364 SvOK_off(sv);
1365#ifndef NODEFAULT_SHAREKEYS
1366 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1367#endif
586fc6a3
SM
1368 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1369 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
c1ae03ae 1370 }
aeb18a1e 1371
bd81e77b
NC
1372 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1373 The target created by newSVrv also is, and it can have magic.
1374 However, it never has SvPVX set.
1375 */
4df7f6af
NC
1376 if (old_type == SVt_IV) {
1377 assert(!SvROK(sv));
1378 } else if (old_type >= SVt_PV) {
bd81e77b
NC
1379 assert(SvPVX_const(sv) == 0);
1380 }
aeb18a1e 1381
bd81e77b 1382 if (old_type >= SVt_PVMG) {
e736a858 1383 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
bd81e77b 1384 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
797c7171
NC
1385 } else {
1386 sv->sv_u.svu_array = NULL; /* or svu_hash */
bd81e77b
NC
1387 }
1388 break;
93e68bfb 1389
bd81e77b
NC
1390 case SVt_PVIV:
1391 /* XXX Is this still needed? Was it ever needed? Surely as there is
1392 no route from NV to PVIV, NOK can never be true */
1393 assert(!SvNOKp(sv));
1394 assert(!SvNOK(sv));
2b5060ae 1395 /* FALLTHROUGH */
bd81e77b
NC
1396 case SVt_PVIO:
1397 case SVt_PVFM:
bd81e77b
NC
1398 case SVt_PVGV:
1399 case SVt_PVCV:
1400 case SVt_PVLV:
d361b004 1401 case SVt_INVLIST:
12c45b25 1402 case SVt_REGEXP:
bd81e77b
NC
1403 case SVt_PVMG:
1404 case SVt_PVNV:
1405 case SVt_PV:
93e68bfb 1406
d2a0f284 1407 assert(new_type_details->body_size);
bd81e77b
NC
1408 /* We always allocated the full length item with PURIFY. To do this
1409 we fake things so that arena is false for all 16 types.. */
1410 if(new_type_details->arena) {
1411 /* This points to the start of the allocated area. */
d2a0f284
JC
1412 new_body_inline(new_body, new_type);
1413 Zero(new_body, new_type_details->body_size, char);
bd81e77b
NC
1414 new_body = ((char *)new_body) - new_type_details->offset;
1415 } else {
1416 new_body = new_NOARENAZ(new_type_details);
1417 }
1418 SvANY(sv) = new_body;
5e2fc214 1419
bd81e77b 1420 if (old_type_details->copy) {
f9ba3d20
NC
1421 /* There is now the potential for an upgrade from something without
1422 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1423 int offset = old_type_details->offset;
1424 int length = old_type_details->copy;
1425
1426 if (new_type_details->offset > old_type_details->offset) {
d4c19fe8 1427 const int difference
f9ba3d20
NC
1428 = new_type_details->offset - old_type_details->offset;
1429 offset += difference;
1430 length -= difference;
1431 }
1432 assert (length >= 0);
1433
1434 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1435 char);
bd81e77b
NC
1436 }
1437
1438#ifndef NV_ZERO_IS_ALLBITS_ZERO
f2524eef 1439 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
e5ce394c
NC
1440 * correct 0.0 for us. Otherwise, if the old body didn't have an
1441 * NV slot, but the new one does, then we need to initialise the
1442 * freshly created NV slot with whatever the correct bit pattern is
1443 * for 0.0 */
e22a937e
NC
1444 if (old_type_details->zero_nv && !new_type_details->zero_nv
1445 && !isGV_with_GP(sv))
bd81e77b 1446 SvNV_set(sv, 0);
82048762 1447#endif
5e2fc214 1448
2439e033 1449 if (UNLIKELY(new_type == SVt_PVIO)) {
85dca89a 1450 IO * const io = MUTABLE_IO(sv);
d963bf01 1451 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
85dca89a
NC
1452
1453 SvOBJECT_on(io);
1454 /* Clear the stashcache because a new IO could overrule a package
1455 name */
103f5a36 1456 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
85dca89a
NC
1457 hv_clear(PL_stashcache);
1458
85dca89a 1459 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
f2524eef 1460 IoPAGE_LEN(sv) = 60;
85dca89a 1461 }
2439e033 1462 if (UNLIKELY(new_type == SVt_REGEXP))
8d919b0a
FC
1463 sv->sv_u.svu_rx = (regexp *)new_body;
1464 else if (old_type < SVt_PV) {
4df7f6af
NC
1465 /* referant will be NULL unless the old type was SVt_IV emulating
1466 SVt_RV */
1467 sv->sv_u.svu_rv = referant;
1468 }
bd81e77b
NC
1469 break;
1470 default:
afd78fd5
JH
1471 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1472 (unsigned long)new_type);
bd81e77b 1473 }
73171d91 1474
5b306eef
DD
1475 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1476 and sometimes SVt_NV */
1477 if (old_type_details->body_size) {
bd81e77b 1478#ifdef PURIFY
beeec492 1479 safefree(old_body);
bd81e77b 1480#else
bc786448
GG
1481 /* Note that there is an assumption that all bodies of types that
1482 can be upgraded came from arenas. Only the more complex non-
1483 upgradable types are allowed to be directly malloc()ed. */
1484 assert(old_type_details->arena);
bd81e77b
NC
1485 del_body((void*)((char*)old_body + old_type_details->offset),
1486 &PL_body_roots[old_type]);
1487#endif
1488 }
1489}
73171d91 1490
bd81e77b
NC
1491/*
1492=for apidoc sv_backoff
73171d91 1493
fde67290 1494Remove any string offset. You should normally use the C<SvOOK_off> macro
bd81e77b 1495wrapper instead.
73171d91 1496
bd81e77b 1497=cut
73171d91
NC
1498*/
1499
fa7a1e49
DD
1500/* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1501 prior to 5.23.4 this function always returned 0
1502*/
1503
1504void
ddeaf645 1505Perl_sv_backoff(SV *const sv)
bd81e77b 1506{
69240efd 1507 STRLEN delta;
7a4bba22 1508 const char * const s = SvPVX_const(sv);
7918f24d
NC
1509
1510 PERL_ARGS_ASSERT_SV_BACKOFF;
7918f24d 1511
bd81e77b
NC
1512 assert(SvOOK(sv));
1513 assert(SvTYPE(sv) != SVt_PVHV);
1514 assert(SvTYPE(sv) != SVt_PVAV);
7a4bba22 1515
69240efd
NC
1516 SvOOK_offset(sv, delta);
1517
7a4bba22
NC
1518 SvLEN_set(sv, SvLEN(sv) + delta);
1519 SvPV_set(sv, SvPVX(sv) - delta);
bd81e77b 1520 SvFLAGS(sv) &= ~SVf_OOK;
fa7a1e49
DD
1521 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1522 return;
bd81e77b 1523}
73171d91 1524
bd81e77b
NC
1525/*
1526=for apidoc sv_grow
73171d91 1527
bd81e77b
NC
1528Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1529upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1530Use the C<SvGROW> wrapper instead.
93e68bfb 1531
bd81e77b
NC
1532=cut
1533*/
93e68bfb 1534
e0060e30
FC
1535static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1536
bd81e77b 1537char *
5aaab254 1538Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
bd81e77b 1539{
eb578fdb 1540 char *s;
93e68bfb 1541
7918f24d
NC
1542 PERL_ARGS_ASSERT_SV_GROW;
1543
bd81e77b
NC
1544 if (SvROK(sv))
1545 sv_unref(sv);
1546 if (SvTYPE(sv) < SVt_PV) {
1547 sv_upgrade(sv, SVt_PV);
1548 s = SvPVX_mutable(sv);
1549 }
1550 else if (SvOOK(sv)) { /* pv is offset? */
1551 sv_backoff(sv);
1552 s = SvPVX_mutable(sv);
1553 if (newlen > SvLEN(sv))
1554 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
bd81e77b
NC
1555 }
1556 else
db2c6cb3 1557 {
e0060e30 1558 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
bd81e77b 1559 s = SvPVX_mutable(sv);
db2c6cb3 1560 }
aeb18a1e 1561
93c10d60 1562#ifdef PERL_COPY_ON_WRITE
cbcb2a16 1563 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
3c239bea 1564 * to store the COW count. So in general, allocate one more byte than
cbcb2a16
DM
1565 * asked for, to make it likely this byte is always spare: and thus
1566 * make more strings COW-able.
1567 * If the new size is a big power of two, don't bother: we assume the
1568 * caller wanted a nice 2^N sized block and will be annoyed at getting
fa8f4f85
TC
1569 * 2^N+1.
1570 * Only increment if the allocation isn't MEM_SIZE_MAX,
1571 * otherwise it will wrap to 0.
1572 */
1573 if (newlen & 0xff && newlen != MEM_SIZE_MAX)
cbcb2a16
DM
1574 newlen++;
1575#endif
1576
ce861ea7
YO
1577#if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1578#define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1579#endif
1580
bd81e77b 1581 if (newlen > SvLEN(sv)) { /* need more room? */
f1200559 1582 STRLEN minlen = SvCUR(sv);
3c239bea 1583 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
f1200559
WH
1584 if (newlen < minlen)
1585 newlen = minlen;
ce861ea7 1586#ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
7c641603
KW
1587
1588 /* Don't round up on the first allocation, as odds are pretty good that
1589 * the initial request is accurate as to what is really needed */
ce861ea7 1590 if (SvLEN(sv)) {
9efda33a
TC
1591 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1592 if (rounded > newlen)
1593 newlen = rounded;
ce861ea7 1594 }
bd81e77b 1595#endif
98653f18 1596 if (SvLEN(sv) && s) {
10edeb5d 1597 s = (char*)saferealloc(s, newlen);
bd81e77b
NC
1598 }
1599 else {
10edeb5d 1600 s = (char*)safemalloc(newlen);
bd81e77b
NC
1601 if (SvPVX_const(sv) && SvCUR(sv)) {
1602 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1603 }
1604 }
1605 SvPV_set(sv, s);
ce861ea7 1606#ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
98653f18
NC
1607 /* Do this here, do it once, do it right, and then we will never get
1608 called back into sv_grow() unless there really is some growing
1609 needed. */
ca7c1a29 1610 SvLEN_set(sv, Perl_safesysmalloc_size(s));
98653f18 1611#else
bd81e77b 1612 SvLEN_set(sv, newlen);
98653f18 1613#endif
bd81e77b
NC
1614 }
1615 return s;
1616}
aeb18a1e 1617
bd81e77b
NC
1618/*
1619=for apidoc sv_setiv
932e9ff9 1620
bd81e77b 1621Copies an integer into the given SV, upgrading first if necessary.
fbe13c60 1622Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
463ee0b2 1623
bd81e77b
NC
1624=cut
1625*/
463ee0b2 1626
bd81e77b 1627void
5aaab254 1628Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
bd81e77b 1629{
7918f24d
NC
1630 PERL_ARGS_ASSERT_SV_SETIV;
1631
bd81e77b
NC
1632 SV_CHECK_THINKFIRST_COW_DROP(sv);
1633 switch (SvTYPE(sv)) {
1634 case SVt_NULL:
bd81e77b 1635 case SVt_NV:
3376de98 1636 sv_upgrade(sv, SVt_IV);
bd81e77b 1637 break;
bd81e77b
NC
1638 case SVt_PV:
1639 sv_upgrade(sv, SVt_PVIV);
1640 break;
463ee0b2 1641
bd81e77b 1642 case SVt_PVGV:
6e592b3a
BM
1643 if (!isGV_with_GP(sv))
1644 break;
bd81e77b
NC
1645 case SVt_PVAV:
1646 case SVt_PVHV:
1647 case SVt_PVCV:
1648 case SVt_PVFM:
1649 case SVt_PVIO:
22e74366 1650 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b
NC
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1652 OP_DESC(PL_op));
0103ca14 1653 break;
42d0e0b7 1654 default: NOOP;
bd81e77b
NC
1655 }
1656 (void)SvIOK_only(sv); /* validate number */
1657 SvIV_set(sv, i);
1658 SvTAINT(sv);
1659}
932e9ff9 1660
bd81e77b
NC
1661/*
1662=for apidoc sv_setiv_mg
d33b2eba 1663
bd81e77b 1664Like C<sv_setiv>, but also handles 'set' magic.
1c846c1f 1665
bd81e77b
NC
1666=cut
1667*/
d33b2eba 1668
bd81e77b 1669void
5aaab254 1670Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
bd81e77b 1671{
7918f24d
NC
1672 PERL_ARGS_ASSERT_SV_SETIV_MG;
1673
bd81e77b
NC
1674 sv_setiv(sv,i);
1675 SvSETMAGIC(sv);
1676}
727879eb 1677
bd81e77b
NC
1678/*
1679=for apidoc sv_setuv
d33b2eba 1680
bd81e77b 1681Copies an unsigned integer into the given SV, upgrading first if necessary.
fbe13c60 1682Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
9b94d1dd 1683
bd81e77b
NC
1684=cut
1685*/
d33b2eba 1686
bd81e77b 1687void
5aaab254 1688Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
bd81e77b 1689{
7918f24d
NC
1690 PERL_ARGS_ASSERT_SV_SETUV;
1691
013abb9b
NC
1692 /* With the if statement to ensure that integers are stored as IVs whenever
1693 possible:
bd81e77b 1694 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
d33b2eba 1695
bd81e77b
NC
1696 without
1697 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1c846c1f 1698
013abb9b
NC
1699 If you wish to remove the following if statement, so that this routine
1700 (and its callers) always return UVs, please benchmark to see what the
1701 effect is. Modern CPUs may be different. Or may not :-)
bd81e77b
NC
1702 */
1703 if (u <= (UV)IV_MAX) {
1704 sv_setiv(sv, (IV)u);
1705 return;
1706 }
1707 sv_setiv(sv, 0);
1708 SvIsUV_on(sv);
1709 SvUV_set(sv, u);
1710}
d33b2eba 1711
bd81e77b
NC
1712/*
1713=for apidoc sv_setuv_mg
727879eb 1714
bd81e77b 1715Like C<sv_setuv>, but also handles 'set' magic.
9b94d1dd 1716
bd81e77b
NC
1717=cut
1718*/
5e2fc214 1719
bd81e77b 1720void
5aaab254 1721Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
bd81e77b 1722{
7918f24d
NC
1723 PERL_ARGS_ASSERT_SV_SETUV_MG;
1724
bd81e77b
NC
1725 sv_setuv(sv,u);
1726 SvSETMAGIC(sv);
1727}
5e2fc214 1728
954c1994 1729/*
bd81e77b 1730=for apidoc sv_setnv
954c1994 1731
bd81e77b 1732Copies a double into the given SV, upgrading first if necessary.
fbe13c60 1733Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
954c1994
GS
1734
1735=cut
1736*/
1737
63f97190 1738void
5aaab254 1739Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
79072805 1740{
7918f24d
NC
1741 PERL_ARGS_ASSERT_SV_SETNV;
1742
bd81e77b
NC
1743 SV_CHECK_THINKFIRST_COW_DROP(sv);
1744 switch (SvTYPE(sv)) {
79072805 1745 case SVt_NULL:
79072805 1746 case SVt_IV:
bd81e77b 1747 sv_upgrade(sv, SVt_NV);
79072805
LW
1748 break;
1749 case SVt_PV:
79072805 1750 case SVt_PVIV:
bd81e77b 1751 sv_upgrade(sv, SVt_PVNV);
79072805 1752 break;
bd4b1eb5 1753
bd4b1eb5 1754 case SVt_PVGV:
6e592b3a
BM
1755 if (!isGV_with_GP(sv))
1756 break;
bd81e77b
NC
1757 case SVt_PVAV:
1758 case SVt_PVHV:
79072805 1759 case SVt_PVCV:
bd81e77b
NC
1760 case SVt_PVFM:
1761 case SVt_PVIO:
22e74366 1762 /* diag_listed_as: Can't coerce %s to %s in %s */
bd81e77b 1763 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
94bbb3f4 1764 OP_DESC(PL_op));
0103ca14 1765 break;
42d0e0b7 1766 default: NOOP;
2068cd4d 1767 }
bd81e77b
NC
1768 SvNV_set(sv, num);
1769 (void)SvNOK_only(sv); /* validate number */
1770 SvTAINT(sv);
79072805
LW
1771}
1772
645c22ef 1773/*
bd81e77b 1774=for apidoc sv_setnv_mg
645c22ef 1775
bd81e77b 1776Like C<sv_setnv>, but also handles 'set' magic.
645c22ef
DM
1777
1778=cut
1779*/
1780
bd81e77b 1781void
5aaab254 1782Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
79072805 1783{
7918f24d
NC
1784 PERL_ARGS_ASSERT_SV_SETNV_MG;
1785
bd81e77b
NC
1786 sv_setnv(sv,num);
1787 SvSETMAGIC(sv);
79072805
LW
1788}
1789
3f7602fa
TC
1790/* Return a cleaned-up, printable version of sv, for non-numeric, or
1791 * not incrementable warning display.
1792 * Originally part of S_not_a_number().
1793 * The return value may be != tmpbuf.
bd81e77b 1794 */
954c1994 1795
3f7602fa
TC
1796STATIC const char *
1797S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1798 const char *pv;
94463019 1799
3f7602fa 1800 PERL_ARGS_ASSERT_SV_DISPLAY;
7918f24d 1801
94463019 1802 if (DO_UTF8(sv)) {
3f7602fa 1803 SV *dsv = newSVpvs_flags("", SVs_TEMP);
37b8cdd1 1804 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
94463019
JH
1805 } else {
1806 char *d = tmpbuf;
3f7602fa 1807 const char * const limit = tmpbuf + tmpbuf_size - 8;
94463019
JH
1808 /* each *s can expand to 4 chars + "...\0",
1809 i.e. need room for 8 chars */
ecdeb87c 1810
00b6aa41
AL
1811 const char *s = SvPVX_const(sv);
1812 const char * const end = s + SvCUR(sv);
1813 for ( ; s < end && d < limit; s++ ) {
94463019 1814 int ch = *s & 0xFF;
bd27cf70 1815 if (! isASCII(ch) && !isPRINT_LC(ch)) {
94463019
JH
1816 *d++ = 'M';
1817 *d++ = '-';
bd27cf70
KW
1818
1819 /* Map to ASCII "equivalent" of Latin1 */
1820 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
94463019
JH
1821 }
1822 if (ch == '\n') {
1823 *d++ = '\\';
1824 *d++ = 'n';
1825 }
1826 else if (ch == '\r') {
1827 *d++ = '\\';
1828 *d++ = 'r';
1829 }
1830 else if (ch == '\f') {
1831 *d++ = '\\';
1832 *d++ = 'f';
1833 }
1834 else if (ch == '\\') {
1835 *d++ = '\\';
1836 *d++ = '\\';
1837 }
1838 else if (ch == '\0') {
1839 *d++ = '\\';
1840 *d++ = '0';
1841 }
1842 else if (isPRINT_LC(ch))
1843 *d++ = ch;
1844 else {
1845 *d++ = '^';
1846 *d++ = toCTRL(ch);
1847 }
1848 }
1849 if (s < end) {
1850 *d++ = '.';
1851 *d++ = '.';
1852 *d++ = '.';
1853 }
1854 *d = '\0';
1855 pv = tmpbuf;
a0d0e21e 1856 }
a0d0e21e 1857
3f7602fa
TC
1858 return pv;
1859}
1860
1861/* Print an "isn't numeric" warning, using a cleaned-up,
1862 * printable version of the offending string
1863 */
1864
1865STATIC void
1866S_not_a_number(pTHX_ SV *const sv)
1867{
3f7602fa
TC
1868 char tmpbuf[64];
1869 const char *pv;
1870
1871 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1872
1873 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1874
533c011a 1875 if (PL_op)
9014280d 1876 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1877 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019
JH
1878 "Argument \"%s\" isn't numeric in %s", pv,
1879 OP_DESC(PL_op));
a0d0e21e 1880 else
9014280d 1881 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
734856a2 1882 /* diag_listed_as: Argument "%s" isn't numeric%s */
94463019 1883 "Argument \"%s\" isn't numeric", pv);
a0d0e21e
LW
1884}
1885
3f7602fa
TC
1886STATIC void
1887S_not_incrementable(pTHX_ SV *const sv) {
3f7602fa
TC
1888 char tmpbuf[64];
1889 const char *pv;
1890
1891 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1892
1893 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1894
1895 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1896 "Argument \"%s\" treated as 0 in increment (++)", pv);
1897}
1898
c2988b20
NC
1899/*
1900=for apidoc looks_like_number
1901
645c22ef
DM
1902Test if the content of an SV looks like a number (or is a number).
1903C<Inf> and C<Infinity> are treated as numbers (so will not issue a
796b6530 1904non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
f52e41ad 1905ignored.
c2988b20
NC
1906
1907=cut
1908*/
1909
1910I32
aad570aa 1911Perl_looks_like_number(pTHX_ SV *const sv)
c2988b20 1912{
eb578fdb 1913 const char *sbegin;
c2988b20 1914 STRLEN len;
ea2485eb 1915 int numtype;
c2988b20 1916
7918f24d
NC
1917 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1918
f52e41ad
FC
1919 if (SvPOK(sv) || SvPOKp(sv)) {
1920 sbegin = SvPV_nomg_const(sv, len);
c2988b20 1921 }
c2988b20 1922 else
e0ab1c0e 1923 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
ea2485eb
JH
1924 numtype = grok_number(sbegin, len, NULL);
1925 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
c2988b20 1926}
25da4f38 1927
19f6321d
NC
1928STATIC bool
1929S_glob_2number(pTHX_ GV * const gv)
180488f8 1930{
7918f24d
NC
1931 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1932
675c862f
AL
1933 /* We know that all GVs stringify to something that is not-a-number,
1934 so no need to test that. */
1935 if (ckWARN(WARN_NUMERIC))
8e629ff4
FC
1936 {
1937 SV *const buffer = sv_newmortal();
1938 gv_efullname3(buffer, gv, "*");
675c862f 1939 not_a_number(buffer);
8e629ff4 1940 }
675c862f
AL
1941 /* We just want something true to return, so that S_sv_2iuv_common
1942 can tail call us and return true. */
19f6321d 1943 return TRUE;
675c862f
AL
1944}
1945
25da4f38
IZ
1946/* Actually, ISO C leaves conversion of UV to IV undefined, but
1947 until proven guilty, assume that things are not that bad... */
1948
645c22ef
DM
1949/*
1950 NV_PRESERVES_UV:
1951
1952 As 64 bit platforms often have an NV that doesn't preserve all bits of
28e5dec8
JH
1953 an IV (an assumption perl has been based on to date) it becomes necessary
1954 to remove the assumption that the NV always carries enough precision to
1955 recreate the IV whenever needed, and that the NV is the canonical form.
1956 Instead, IV/UV and NV need to be given equal rights. So as to not lose
645c22ef 1957 precision as a side effect of conversion (which would lead to insanity
28e5dec8 1958 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
8a4a3196
KW
1959 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1960 where precision was lost, and IV/UV/NV slots that have a valid conversion
1961 which has lost no precision
645c22ef 1962 2) to ensure that if a numeric conversion to one form is requested that
28e5dec8
JH
1963 would lose precision, the precise conversion (or differently
1964 imprecise conversion) is also performed and cached, to prevent
1965 requests for different numeric formats on the same SV causing
1966 lossy conversion chains. (lossless conversion chains are perfectly
1967 acceptable (still))
1968
1969
1970 flags are used:
1971 SvIOKp is true if the IV slot contains a valid value
1972 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1973 SvNOKp is true if the NV slot contains a valid value
1974 SvNOK is true only if the NV value is accurate
1975
1976 so
645c22ef 1977 while converting from PV to NV, check to see if converting that NV to an
28e5dec8
JH
1978 IV(or UV) would lose accuracy over a direct conversion from PV to
1979 IV(or UV). If it would, cache both conversions, return NV, but mark
1980 SV as IOK NOKp (ie not NOK).
1981
645c22ef 1982 While converting from PV to IV, check to see if converting that IV to an
28e5dec8
JH
1983 NV would lose accuracy over a direct conversion from PV to NV. If it
1984 would, cache both conversions, flag similarly.
1985
1986 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1987 correctly because if IV & NV were set NV *always* overruled.
645c22ef
DM
1988 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1989 changes - now IV and NV together means that the two are interchangeable:
28e5dec8 1990 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
d460ef45 1991
645c22ef
DM
1992 The benefit of this is that operations such as pp_add know that if
1993 SvIOK is true for both left and right operands, then integer addition
1994 can be used instead of floating point (for cases where the result won't
1995 overflow). Before, floating point was always used, which could lead to
28e5dec8
JH
1996 loss of precision compared with integer addition.
1997
1998 * making IV and NV equal status should make maths accurate on 64 bit
1999 platforms
2000 * may speed up maths somewhat if pp_add and friends start to use
645c22ef 2001 integers when possible instead of fp. (Hopefully the overhead in
28e5dec8
JH
2002 looking for SvIOK and checking for overflow will not outweigh the
2003 fp to integer speedup)
2004 * will slow down integer operations (callers of SvIV) on "inaccurate"
2005 values, as the change from SvIOK to SvIOKp will cause a call into
2006 sv_2iv each time rather than a macro access direct to the IV slot
2007 * should speed up number->string conversion on integers as IV is
645c22ef 2008 favoured when IV and NV are equally accurate
28e5dec8
JH
2009
2010 ####################################################################
645c22ef
DM
2011 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2012 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2013 On the other hand, SvUOK is true iff UV.
28e5dec8
JH
2014 ####################################################################
2015
645c22ef 2016 Your mileage will vary depending your CPU's relative fp to integer
28e5dec8
JH
2017 performance ratio.
2018*/
2019
2020#ifndef NV_PRESERVES_UV
645c22ef
DM
2021# define IS_NUMBER_UNDERFLOW_IV 1
2022# define IS_NUMBER_UNDERFLOW_UV 2
2023# define IS_NUMBER_IV_AND_UV 2
2024# define IS_NUMBER_OVERFLOW_IV 4
2025# define IS_NUMBER_OVERFLOW_UV 5
2026
2027/* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
28e5dec8
JH
2028
2029/* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2030STATIC int
5aaab254 2031S_sv_2iuv_non_preserve(pTHX_ SV *const sv
47031da6
NC
2032# ifdef DEBUGGING
2033 , I32 numtype
2034# endif
2035 )
28e5dec8 2036{
7918f24d 2037 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
23491f1d 2038 PERL_UNUSED_CONTEXT;
7918f24d 2039
3f7c398e 2040 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
2041 if (SvNVX(sv) < (NV)IV_MIN) {
2042 (void)SvIOKp_on(sv);
2043 (void)SvNOK_on(sv);
45977657 2044 SvIV_set(sv, IV_MIN);
28e5dec8
JH
2045 return IS_NUMBER_UNDERFLOW_IV;
2046 }
2047 if (SvNVX(sv) > (NV)UV_MAX) {
2048 (void)SvIOKp_on(sv);
2049 (void)SvNOK_on(sv);
2050 SvIsUV_on(sv);
607fa7f2 2051 SvUV_set(sv, UV_MAX);
28e5dec8
JH
2052 return IS_NUMBER_OVERFLOW_UV;
2053 }
c2988b20
NC
2054 (void)SvIOKp_on(sv);
2055 (void)SvNOK_on(sv);
2056 /* Can't use strtol etc to convert this string. (See truth table in
2057 sv_2iv */
2058 if (SvNVX(sv) <= (UV)IV_MAX) {
45977657 2059 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2060 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2061 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2062 } else {
2063 /* Integer is imprecise. NOK, IOKp */
2064 }
2065 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2066 }
2067 SvIsUV_on(sv);
607fa7f2 2068 SvUV_set(sv, U_V(SvNVX(sv)));
659c4b96 2069 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
c2988b20
NC
2070 if (SvUVX(sv) == UV_MAX) {
2071 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2072 possibly be preserved by NV. Hence, it must be overflow.
2073 NOK, IOKp */
2074 return IS_NUMBER_OVERFLOW_UV;
2075 }
2076 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2077 } else {
2078 /* Integer is imprecise. NOK, IOKp */
28e5dec8 2079 }
c2988b20 2080 return IS_NUMBER_OVERFLOW_IV;
28e5dec8 2081}
645c22ef
DM
2082#endif /* !NV_PRESERVES_UV*/
2083
a13f4dff 2084/* If numtype is infnan, set the NV of the sv accordingly.
5564cd7f 2085 * If numtype is anything else, try setting the NV using Atof(PV). */
3c81f0b3
DD
2086#ifdef USING_MSVC6
2087# pragma warning(push)
2088# pragma warning(disable:4756;disable:4056)
2089#endif
a13f4dff 2090static void
3823048b 2091S_sv_setnv(pTHX_ SV* sv, int numtype)
a13f4dff 2092{
07925c5e 2093 bool pok = cBOOL(SvPOK(sv));
5564cd7f 2094 bool nok = FALSE;
a13f4dff
JH
2095 if ((numtype & IS_NUMBER_INFINITY)) {
2096 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
5564cd7f 2097 nok = TRUE;
a13f4dff
JH
2098 }
2099 else if ((numtype & IS_NUMBER_NAN)) {
3823048b 2100 SvNV_set(sv, NV_NAN);
d48bd569 2101 nok = TRUE;
a13f4dff 2102 }
d48bd569 2103 else if (pok) {
a13f4dff 2104 SvNV_set(sv, Atof(SvPVX_const(sv)));
d48bd569
JH
2105 /* Purposefully no true nok here, since we don't want to blow
2106 * away the possible IOK/UV of an existing sv. */
2107 }
5564cd7f 2108 if (nok) {
d48bd569 2109 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
5564cd7f
JH
2110 if (pok)
2111 SvPOK_on(sv); /* PV is okay, though. */
2112 }
a13f4dff 2113}
3c81f0b3
DD
2114#ifdef USING_MSVC6
2115# pragma warning(pop)
2116#endif
a13f4dff 2117
af359546 2118STATIC bool
7918f24d
NC
2119S_sv_2iuv_common(pTHX_ SV *const sv)
2120{
7918f24d
NC
2121 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2122
af359546 2123 if (SvNOKp(sv)) {
28e5dec8
JH
2124 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2125 * without also getting a cached IV/UV from it at the same time
2126 * (ie PV->NV conversion should detect loss of accuracy and cache
af359546
NC
2127 * IV or UV at same time to avoid this. */
2128 /* IV-over-UV optimisation - choose to cache IV if possible */
25da4f38
IZ
2129
2130 if (SvTYPE(sv) == SVt_NV)
2131 sv_upgrade(sv, SVt_PVNV);
2132
28e5dec8
JH
2133 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2134 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2135 certainly cast into the IV range at IV_MAX, whereas the correct
2136 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2137 cases go to UV */
e91de695
JH
2138#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2139 if (Perl_isnan(SvNVX(sv))) {
2140 SvUV_set(sv, 0);
2141 SvIsUV_on(sv);
2142 return FALSE;
2143 }
2144#endif
28e5dec8 2145 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
45977657 2146 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2147 if (SvNVX(sv) == (NV) SvIVX(sv)
28e5dec8 2148#ifndef NV_PRESERVES_UV
53e2bfb7 2149 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
28e5dec8
JH
2150 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2151 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2152 /* Don't flag it as "accurately an integer" if the number
2153 came from a (by definition imprecise) NV operation, and
2154 we're outside the range of NV integer precision */
2155#endif
2156 ) {
a43d94f2
NC
2157 if (SvNOK(sv))
2158 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2159 else {
2160 /* scalar has trailing garbage, eg "42a" */
2161 }
28e5dec8 2162 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2163 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
28e5dec8
JH
2164 PTR2UV(sv),
2165 SvNVX(sv),
2166 SvIVX(sv)));
2167
2168 } else {
2169 /* IV not precise. No need to convert from PV, as NV
2170 conversion would already have cached IV if it detected
2171 that PV->IV would be better than PV->NV->IV
2172 flags already correct - don't set public IOK. */
2173 DEBUG_c(PerlIO_printf(Perl_debug_log,
7234c960 2174 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
28e5dec8
JH
2175 PTR2UV(sv),
2176 SvNVX(sv),
2177 SvIVX(sv)));
2178 }
2179 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2180 but the cast (NV)IV_MIN rounds to a the value less (more
2181 negative) than IV_MIN which happens to be equal to SvNVX ??
2182 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2183 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2184 (NV)UVX == NVX are both true, but the values differ. :-(
2185 Hopefully for 2s complement IV_MIN is something like
2186 0x8000000000000000 which will be exact. NWC */
d460ef45 2187 }
25da4f38 2188 else {
607fa7f2 2189 SvUV_set(sv, U_V(SvNVX(sv)));
28e5dec8 2190 if (
659c4b96 2191 (SvNVX(sv) == (NV) SvUVX(sv))
28e5dec8
JH
2192#ifndef NV_PRESERVES_UV
2193 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2194 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2195 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2196 /* Don't flag it as "accurately an integer" if the number
2197 came from a (by definition imprecise) NV operation, and
2198 we're outside the range of NV integer precision */
2199#endif
a43d94f2 2200 && SvNOK(sv)
28e5dec8
JH
2201 )
2202 SvIOK_on(sv);
25da4f38 2203 SvIsUV_on(sv);
1c846c1f 2204 DEBUG_c(PerlIO_printf(Perl_debug_log,
57def98f 2205 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
56431972 2206 PTR2UV(sv),
57def98f
JH
2207 SvUVX(sv),
2208 SvUVX(sv)));
25da4f38 2209 }
748a9306 2210 }
cd84013a 2211 else if (SvPOKp(sv)) {
c2988b20 2212 UV value;
3823048b 2213 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
af359546 2214 /* We want to avoid a possible problem when we cache an IV/ a UV which
25da4f38 2215 may be later translated to an NV, and the resulting NV is not
c2988b20
NC
2216 the same as the direct translation of the initial string
2217 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2218 be careful to ensure that the value with the .456 is around if the
2219 NV value is requested in the future).
1c846c1f 2220
af359546 2221 This means that if we cache such an IV/a UV, we need to cache the
25da4f38 2222 NV as well. Moreover, we trade speed for space, and do not
28e5dec8 2223 cache the NV if we are sure it's not needed.
25da4f38 2224 */
16b7a9a4 2225
c2988b20
NC
2226 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2227 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2228 == IS_NUMBER_IN_UV) {
5e045b90 2229 /* It's definitely an integer, only upgrade to PVIV */
28e5dec8
JH
2230 if (SvTYPE(sv) < SVt_PVIV)
2231 sv_upgrade(sv, SVt_PVIV);
f7bbb42a 2232 (void)SvIOK_on(sv);
c2988b20
NC
2233 } else if (SvTYPE(sv) < SVt_PVNV)
2234 sv_upgrade(sv, SVt_PVNV);
28e5dec8 2235
a13f4dff 2236 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
75a57a38 2237 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
6b322424 2238 not_a_number(sv);
3823048b 2239 S_sv_setnv(aTHX_ sv, numtype);
a13f4dff
JH
2240 return FALSE;
2241 }
2242
f2524eef 2243 /* If NVs preserve UVs then we only use the UV value if we know that
c2988b20
NC
2244 we aren't going to call atof() below. If NVs don't preserve UVs
2245 then the value returned may have more precision than atof() will
2246 return, even though value isn't perfectly accurate. */
2247 if ((numtype & (IS_NUMBER_IN_UV
2248#ifdef NV_PRESERVES_UV
2249 | IS_NUMBER_NOT_INT
2250#endif
2251 )) == IS_NUMBER_IN_UV) {
2252 /* This won't turn off the public IOK flag if it was set above */
2253 (void)SvIOKp_on(sv);
2254
2255 if (!(numtype & IS_NUMBER_NEG)) {
2256 /* positive */;
2257 if (value <= (UV)IV_MAX) {
45977657 2258 SvIV_set(sv, (IV)value);
c2988b20 2259 } else {
af359546 2260 /* it didn't overflow, and it was positive. */
607fa7f2 2261 SvUV_set(sv, value);
c2988b20
NC
2262 SvIsUV_on(sv);
2263 }
2264 } else {
2265 /* 2s complement assumption */
2266 if (value <= (UV)IV_MIN) {
53e2bfb7
DM
2267 SvIV_set(sv, value == (UV)IV_MIN
2268 ? IV_MIN : -(IV)value);
c2988b20
NC
2269 } else {
2270 /* Too negative for an IV. This is a double upgrade, but
d1be9408 2271 I'm assuming it will be rare. */
c2988b20
NC
2272 if (SvTYPE(sv) < SVt_PVNV)
2273 sv_upgrade(sv, SVt_PVNV);
2274 SvNOK_on(sv);
2275 SvIOK_off(sv);
2276 SvIOKp_on(sv);
9d6ce603 2277 SvNV_set(sv, -(NV)value);
45977657 2278 SvIV_set(sv, IV_MIN);
c2988b20
NC
2279 }
2280 }
2281 }
2282 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2283 will be in the previous block to set the IV slot, and the next
2284 block to set the NV slot. So no else here. */
2285
2286 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2287 != IS_NUMBER_IN_UV) {
2288 /* It wasn't an (integer that doesn't overflow the UV). */
3823048b 2289 S_sv_setnv(aTHX_ sv, numtype);
28e5dec8 2290
c2988b20
NC
2291 if (! numtype && ckWARN(WARN_NUMERIC))
2292 not_a_number(sv);
28e5dec8 2293
88cb8500 2294 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
c2988b20 2295 PTR2UV(sv), SvNVX(sv)));
28e5dec8 2296
28e5dec8 2297#ifdef NV_PRESERVES_UV
af359546
NC
2298 (void)SvIOKp_on(sv);
2299 (void)SvNOK_on(sv);
e91de695
JH
2300#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2301 if (Perl_isnan(SvNVX(sv))) {
2302 SvUV_set(sv, 0);
2303 SvIsUV_on(sv);
2304 return FALSE;
2305 }
2306#endif
af359546
NC
2307 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2308 SvIV_set(sv, I_V(SvNVX(sv)));
2309 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2310 SvIOK_on(sv);
2311 } else {
6f207bd3 2312 NOOP; /* Integer is imprecise. NOK, IOKp */
af359546
NC
2313 }
2314 /* UV will not work better than IV */
2315 } else {
2316 if (SvNVX(sv) > (NV)UV_MAX) {
2317 SvIsUV_on(sv);
2318 /* Integer is inaccurate. NOK, IOKp, is UV */
2319 SvUV_set(sv, UV_MAX);
af359546
NC
2320 } else {
2321 SvUV_set(sv, U_V(SvNVX(sv)));
2322 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2323 NV preservse UV so can do correct comparison. */
2324 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2325 SvIOK_on(sv);
af359546 2326 } else {
6f207bd3 2327 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
af359546
NC
2328 }
2329 }
4b0c9573 2330 SvIsUV_on(sv);
af359546 2331 }
28e5dec8 2332#else /* NV_PRESERVES_UV */
c2988b20
NC
2333 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2334 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
af359546 2335 /* The IV/UV slot will have been set from value returned by
c2988b20
NC
2336 grok_number above. The NV slot has just been set using
2337 Atof. */
560b0c46 2338 SvNOK_on(sv);
c2988b20
NC
2339 assert (SvIOKp(sv));
2340 } else {
2341 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2342 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2343 /* Small enough to preserve all bits. */
2344 (void)SvIOKp_on(sv);
2345 SvNOK_on(sv);
45977657 2346 SvIV_set(sv, I_V(SvNVX(sv)));
659c4b96 2347 if ((NV)(SvIVX(sv)) == SvNVX(sv))
c2988b20
NC
2348 SvIOK_on(sv);
2349 /* Assumption: first non-preserved integer is < IV_MAX,
2350 this NV is in the preserved range, therefore: */
2351 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2352 < (UV)IV_MAX)) {
32fdb065 2353 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
2354 }
2355 } else {
2356 /* IN_UV NOT_INT
2357 0 0 already failed to read UV.
2358 0 1 already failed to read UV.
2359 1 0 you won't get here in this case. IV/UV
2360 slot set, public IOK, Atof() unneeded.
2361 1 1 already read UV.
2362 so there's no point in sv_2iuv_non_preserve() attempting
2363 to use atol, strtol, strtoul etc. */
47031da6 2364# ifdef DEBUGGING
40a17c4c 2365 sv_2iuv_non_preserve (sv, numtype);
47031da6
NC
2366# else
2367 sv_2iuv_non_preserve (sv);
2368# endif
c2988b20
NC
2369 }
2370 }
28e5dec8 2371#endif /* NV_PRESERVES_UV */
a43d94f2
NC
2372 /* It might be more code efficient to go through the entire logic above
2373 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2374 gets complex and potentially buggy, so more programmer efficient
2375 to do it this way, by turning off the public flags: */
2376 if (!numtype)
2377 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
25da4f38 2378 }
af359546
NC
2379 }
2380 else {
675c862f 2381 if (isGV_with_GP(sv))
159b6efe 2382 return glob_2number(MUTABLE_GV(sv));
180488f8 2383
4f62cd62 2384 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
af359546 2385 report_uninit(sv);
25da4f38
IZ
2386 if (SvTYPE(sv) < SVt_IV)
2387 /* Typically the caller expects that sv_any is not NULL now. */
2388 sv_upgrade(sv, SVt_IV);
af359546
NC
2389 /* Return 0 from the caller. */
2390 return TRUE;
2391 }
2392 return FALSE;
2393}
2394
2395/*
2396=for apidoc sv_2iv_flags
2397
2398Return the integer value of an SV, doing any necessary string
c5608a1f 2399conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
af359546
NC
2400Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2401
2402=cut
2403*/
2404
2405IV
5aaab254 2406Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
af359546 2407{
1061065f 2408 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
4bac9ae4 2409
217f6fa3
FC
2410 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2411 && SvTYPE(sv) != SVt_PVFM);
2412
4bac9ae4
CS
2413 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2414 mg_get(sv);
2415
2416 if (SvROK(sv)) {
2417 if (SvAMAGIC(sv)) {
2418 SV * tmpstr;
2419 if (flags & SV_SKIP_OVERLOAD)
2420 return 0;
2421 tmpstr = AMG_CALLunary(sv, numer_amg);
2422 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2423 return SvIV(tmpstr);
2424 }
2425 }
2426 return PTR2IV(SvRV(sv));
2427 }
2428
8d919b0a 2429 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d
NC
2430 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2431 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2432 In practice they are extremely unlikely to actually get anywhere
2433 accessible by user Perl code - the only way that I'm aware of is when
2434 a constant subroutine which is used as the second argument to index.
cd84013a
FC
2435
2436 Regexps have no SvIVX and SvNVX fields.
2b2b6d6d 2437 */
8d919b0a 2438 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2439 {
71c558c3 2440 UV value;
8d919b0a
FC
2441 const char * const ptr =
2442 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2443 const int numtype
2444 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2445
2446 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2447 == IS_NUMBER_IN_UV) {
2448 /* It's definitely an integer */
2449 if (numtype & IS_NUMBER_NEG) {
2450 if (value < (UV)IV_MIN)
2451 return -(IV)value;
2452 } else {
2453 if (value < (UV)IV_MAX)
2454 return (IV)value;
2455 }
2456 }
058b8ae2 2457
e91de695
JH
2458 /* Quite wrong but no good choices. */
2459 if ((numtype & IS_NUMBER_INFINITY)) {
2460 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2461 } else if ((numtype & IS_NUMBER_NAN)) {
2462 return 0; /* So wrong. */
2463 }
2464
71c558c3
NC
2465 if (!numtype) {
2466 if (ckWARN(WARN_NUMERIC))
2467 not_a_number(sv);
2468 }
8d919b0a 2469 return I_V(Atof(ptr));
e20b6c3b 2470 }
4bac9ae4
CS
2471 }
2472
2473 if (SvTHINKFIRST(sv)) {
af359546
NC
2474 if (SvREADONLY(sv) && !SvOK(sv)) {
2475 if (ckWARN(WARN_UNINITIALIZED))
2476 report_uninit(sv);
2477 return 0;
2478 }
2479 }
4bac9ae4 2480
af359546
NC
2481 if (!SvIOKp(sv)) {
2482 if (S_sv_2iuv_common(aTHX_ sv))
2483 return 0;
79072805 2484 }
4bac9ae4 2485
1d7c1841
GS
2486 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2487 PTR2UV(sv),SvIVX(sv)));
25da4f38 2488 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
79072805
LW
2489}
2490
645c22ef 2491/*
891f9566 2492=for apidoc sv_2uv_flags
645c22ef
DM
2493
2494Return the unsigned integer value of an SV, doing any necessary string
c5608a1f 2495conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
891f9566 2496Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
645c22ef
DM
2497
2498=cut
2499*/
2500
ff68c719 2501UV
5aaab254 2502Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
ff68c719 2503{
1061065f 2504 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
4bac9ae4
CS
2505
2506 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2507 mg_get(sv);
2508
2509 if (SvROK(sv)) {
2510 if (SvAMAGIC(sv)) {
2511 SV *tmpstr;
2512 if (flags & SV_SKIP_OVERLOAD)
2513 return 0;
2514 tmpstr = AMG_CALLunary(sv, numer_amg);
2515 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2516 return SvUV(tmpstr);
2517 }
2518 }
2519 return PTR2UV(SvRV(sv));
2520 }
2521
8d919b0a 2522 if (SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2523 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2524 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2525 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2526 assert(isREGEXP(sv) || SvPOKp(sv));
e20b6c3b 2527 {
71c558c3 2528 UV value;
8d919b0a
FC
2529 const char * const ptr =
2530 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
e91de695
JH
2531 const int numtype
2532 = grok_number(ptr, SvCUR(sv), &value);
71c558c3
NC
2533
2534 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2535 == IS_NUMBER_IN_UV) {
2536 /* It's definitely an integer */
2537 if (!(numtype & IS_NUMBER_NEG))
2538 return value;
2539 }
058b8ae2 2540
e91de695
JH
2541 /* Quite wrong but no good choices. */
2542 if ((numtype & IS_NUMBER_INFINITY)) {
2543 return UV_MAX; /* So wrong. */
2544 } else if ((numtype & IS_NUMBER_NAN)) {
2545 return 0; /* So wrong. */
2546 }
2547
71c558c3
NC
2548 if (!numtype) {
2549 if (ckWARN(WARN_NUMERIC))
2550 not_a_number(sv);
2551 }
8d919b0a 2552 return U_V(Atof(ptr));
e20b6c3b 2553 }
4bac9ae4
CS
2554 }
2555
2556 if (SvTHINKFIRST(sv)) {
0336b60e 2557 if (SvREADONLY(sv) && !SvOK(sv)) {
0336b60e 2558 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2559 report_uninit(sv);
ff68c719
PP
2560 return 0;
2561 }
2562 }
4bac9ae4 2563
af359546
NC
2564 if (!SvIOKp(sv)) {
2565 if (S_sv_2iuv_common(aTHX_ sv))
2566 return 0;
ff68c719 2567 }
25da4f38 2568
1d7c1841
GS
2569 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2570 PTR2UV(sv),SvUVX(sv)));
25da4f38 2571 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
ff68c719
PP
2572}
2573
645c22ef 2574/*
196007d1 2575=for apidoc sv_2nv_flags
645c22ef
DM
2576
2577Return the num value of an SV, doing any necessary string or integer
c5608a1f 2578conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
39d5de13 2579Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
645c22ef
DM
2580
2581=cut
2582*/
2583
65202027 2584NV
5aaab254 2585Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
79072805 2586{
1061065f
DD
2587 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2588
217f6fa3
FC
2589 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2590 && SvTYPE(sv) != SVt_PVFM);
8d919b0a 2591 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2b2b6d6d 2592 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
cd84013a
FC
2593 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2594 Regexps have no SvIVX and SvNVX fields. */
8d919b0a 2595 const char *ptr;
39d5de13
DM
2596 if (flags & SV_GMAGIC)
2597 mg_get(sv);
463ee0b2
LW
2598 if (SvNOKp(sv))
2599 return SvNVX(sv);
cd84013a 2600 if (SvPOKp(sv) && !SvIOKp(sv)) {
8d919b0a
FC
2601 ptr = SvPVX_const(sv);
2602 grokpv:
041457d9 2603 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
8d919b0a 2604 !grok_number(ptr, SvCUR(sv), NULL))
a0d0e21e 2605 not_a_number(sv);
8d919b0a 2606 return Atof(ptr);
a0d0e21e 2607 }
25da4f38 2608 if (SvIOKp(sv)) {
1c846c1f 2609 if (SvIsUV(sv))
65202027 2610 return (NV)SvUVX(sv);
25da4f38 2611 else
65202027 2612 return (NV)SvIVX(sv);
47a72cb8
NC
2613 }
2614 if (SvROK(sv)) {
2615 goto return_rok;
2616 }
8d919b0a
FC
2617 if (isREGEXP(sv)) {
2618 ptr = RX_WRAPPED((REGEXP *)sv);
2619 goto grokpv;
2620 }
47a72cb8
NC
2621 assert(SvTYPE(sv) >= SVt_PVMG);
2622 /* This falls through to the report_uninit near the end of the
2623 function. */
2624 } else if (SvTHINKFIRST(sv)) {
a0d0e21e 2625 if (SvROK(sv)) {
47a72cb8 2626 return_rok:
deb46114 2627 if (SvAMAGIC(sv)) {
aee036bb
DM
2628 SV *tmpstr;
2629 if (flags & SV_SKIP_OVERLOAD)
2630 return 0;
31d632c3 2631 tmpstr = AMG_CALLunary(sv, numer_amg);
deb46114
NC
2632 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2633 return SvNV(tmpstr);
2634 }
2635 }
2636 return PTR2NV(SvRV(sv));
a0d0e21e 2637 }
0336b60e 2638 if (SvREADONLY(sv) && !SvOK(sv)) {
599cee73 2639 if (ckWARN(WARN_UNINITIALIZED))
29489e7c 2640 report_uninit(sv);
ed6116ce
LW
2641 return 0.0;
2642 }
79072805
LW
2643 }
2644 if (SvTYPE(sv) < SVt_NV) {
7e25a7e9
NC
2645 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2646 sv_upgrade(sv, SVt_NV);
097ee67d 2647 DEBUG_c({
f93f4e46 2648 STORE_NUMERIC_LOCAL_SET_STANDARD();
1d7c1841 2649 PerlIO_printf(Perl_debug_log,
88cb8500 2650 "0x%"UVxf" num(%" NVgf ")\n",
1d7c1841 2651 PTR2UV(sv), SvNVX(sv));
097ee67d
JH
2652 RESTORE_NUMERIC_LOCAL();
2653 });
79072805
LW
2654 }
2655 else if (SvTYPE(sv) < SVt_PVNV)
2656 sv_upgrade(sv, SVt_PVNV);
59d8ce62
NC
2657 if (SvNOKp(sv)) {
2658 return SvNVX(sv);
61604483 2659 }
59d8ce62 2660 if (SvIOKp(sv)) {
9d6ce603 2661 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
28e5dec8 2662#ifdef NV_PRESERVES_UV
a43d94f2
NC
2663 if (SvIOK(sv))
2664 SvNOK_on(sv);
2665 else
2666 SvNOKp_on(sv);
28e5dec8
JH
2667#else
2668 /* Only set the public NV OK flag if this NV preserves the IV */
2669 /* Check it's not 0xFFFFFFFFFFFFFFFF */
a43d94f2
NC
2670 if (SvIOK(sv) &&
2671 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
28e5dec8
JH
2672 : (SvIVX(sv) == I_V(SvNVX(sv))))
2673 SvNOK_on(sv);
2674 else
2675 SvNOKp_on(sv);
2676#endif
93a17b20 2677 }
cd84013a 2678 else if (SvPOKp(sv)) {
c2988b20 2679 UV value;
3823048b 2680 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
041457d9 2681 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
a0d0e21e 2682 not_a_number(sv);
28e5dec8 2683#ifdef NV_PRESERVES_UV
c2988b20
NC
2684 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2685 == IS_NUMBER_IN_UV) {
5e045b90 2686 /* It's definitely an integer */
9d6ce603 2687 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
66d83377 2688 } else {
3823048b 2689 S_sv_setnv(aTHX_ sv, numtype);
66d83377 2690 }
a43d94f2
NC
2691 if (numtype)
2692 SvNOK_on(sv);
2693 else
2694 SvNOKp_on(sv);
28e5dec8 2695#else
e91de695
JH
2696 SvNV_set(sv, Atof(SvPVX_const(sv)));
2697 /* Only set the public NV OK flag if this NV preserves the value in
2698 the PV at least as well as an IV/UV would.
2699 Not sure how to do this 100% reliably. */
2700 /* if that shift count is out of range then Configure's test is
2701 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2702 UV_BITS */
2703 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2704 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2705 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2706 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2707 /* Can't use strtol etc to convert this string, so don't try.
2708 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
c2988b20
NC
2709 SvNOK_on(sv);
2710 } else {
e91de695 2711 /* value has been set. It may not be precise. */
53e2bfb7 2712 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
e91de695
JH
2713 /* 2s complement assumption for (UV)IV_MIN */
2714 SvNOK_on(sv); /* Integer is too negative. */
c2988b20 2715 } else {
e91de695
JH
2716 SvNOKp_on(sv);
2717 SvIOKp_on(sv);
6fa402ec 2718
e91de695 2719 if (numtype & IS_NUMBER_NEG) {
02b08bbc
DM
2720 /* -IV_MIN is undefined, but we should never reach
2721 * this point with both IS_NUMBER_NEG and value ==
2722 * (UV)IV_MIN */
2723 assert(value != (UV)IV_MIN);
e91de695
JH
2724 SvIV_set(sv, -(IV)value);
2725 } else if (value <= (UV)IV_MAX) {
2726 SvIV_set(sv, (IV)value);
2727 } else {
2728 SvUV_set(sv, value);
2729 SvIsUV_on(sv);
2730 }
c2988b20 2731
e91de695
JH
2732 if (numtype & IS_NUMBER_NOT_INT) {
2733 /* I believe that even if the original PV had decimals,
2734 they are lost beyond the limit of the FP precision.
2735 However, neither is canonical, so both only get p
2736 flags. NWC, 2000/11/25 */
2737 /* Both already have p flags, so do nothing */
2738 } else {
2739 const NV nv = SvNVX(sv);
2740 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2741 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2742 if (SvIVX(sv) == I_V(nv)) {
2743 SvNOK_on(sv);
2744 } else {
2745 /* It had no "." so it must be integer. */
2746 }
2747 SvIOK_on(sv);
0f83c5a4 2748 } else {
e91de695
JH
2749 /* between IV_MAX and NV(UV_MAX).
2750 Could be slightly > UV_MAX */
2751
2752 if (numtype & IS_NUMBER_NOT_INT) {
2753 /* UV and NV both imprecise. */
0f83c5a4 2754 } else {
e91de695
JH
2755 const UV nv_as_uv = U_V(nv);
2756
2757 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2758 SvNOK_on(sv);
c2988b20 2759 }
e91de695 2760 SvIOK_on(sv);
c2988b20
NC
2761 }
2762 }
2763 }
2764 }
0f83c5a4 2765 }
e91de695
JH
2766 /* It might be more code efficient to go through the entire logic above
2767 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2768 gets complex and potentially buggy, so more programmer efficient
2769 to do it this way, by turning off the public flags: */
2770 if (!numtype)
2771 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
28e5dec8 2772#endif /* NV_PRESERVES_UV */
93a17b20 2773 }
79072805 2774 else {
e91de695
JH
2775 if (isGV_with_GP(sv)) {
2776 glob_2number(MUTABLE_GV(sv));
2777 return 0.0;
2778 }
180488f8 2779
e91de695
JH
2780 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2781 report_uninit(sv);
2782 assert (SvTYPE(sv) >= SVt_NV);
2783 /* Typically the caller expects that sv_any is not NULL now. */
2784 /* XXX Ilya implies that this is a bug in callers that assume this
2785 and ideally should be fixed. */
2786 return 0.0;
79072805 2787 }
097ee67d 2788 DEBUG_c({
e91de695
JH
2789 STORE_NUMERIC_LOCAL_SET_STANDARD();
2790 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2791 PTR2UV(sv), SvNVX(sv));
2792 RESTORE_NUMERIC_LOCAL();
2793 });
463ee0b2 2794 return SvNVX(sv);
79072805
LW
2795}
2796
800401ee
JH
2797/*
2798=for apidoc sv_2num
2799
2800Return an SV with the numeric value of the source SV, doing any necessary
d024d1a7
FC
2801reference or overload conversion. The caller is expected to have handled
2802get-magic already.
800401ee
JH
2803
2804=cut
2805*/
2806
2807SV *
5aaab254 2808Perl_sv_2num(pTHX_ SV *const sv)
800401ee 2809{
7918f24d
NC
2810 PERL_ARGS_ASSERT_SV_2NUM;
2811
b9ee0594
RGS
2812 if (!SvROK(sv))
2813 return sv;
800401ee 2814 if (SvAMAGIC(sv)) {
31d632c3 2815 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
a02ec77a 2816 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
800401ee
JH
2817 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2818 return sv_2num(tmpsv);
2819 }
2820 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2821}
2822
645c22ef
DM
2823/* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2824 * UV as a string towards the end of buf, and return pointers to start and
2825 * end of it.
2826 *
2827 * We assume that buf is at least TYPE_CHARS(UV) long.
2828 */
2829
864dbfa3 2830static char *
5de3775c 2831S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
25da4f38 2832{
25da4f38 2833 char *ptr = buf + TYPE_CHARS(UV);
823a54a3 2834 char * const ebuf = ptr;
25da4f38 2835 int sign;
25da4f38 2836
7918f24d
NC
2837 PERL_ARGS_ASSERT_UIV_2BUF;
2838
25da4f38
IZ
2839 if (is_uv)
2840 sign = 0;
2841 else if (iv >= 0) {
2842 uv = iv;
2843 sign = 0;
2844 } else {
53e2bfb7 2845 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
25da4f38
IZ
2846 sign = 1;
2847 }
2848 do {
eb160463 2849 *--ptr = '0' + (char)(uv % 10);
25da4f38
IZ
2850 } while (uv /= 10);
2851 if (sign)
2852 *--ptr = '-';
2853 *peob = ebuf;
2854 return ptr;
2855}
2856
bfaa02d5
JH
2857/* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2858 * infinity or a not-a-number, writes the appropriate strings to the
2859 * buffer, including a zero byte. On success returns the written length,
3bde2d43
JH
2860 * excluding the zero byte, on failure (not an infinity, not a nan)
2861 * returns zero, assert-fails on maxlen being too short.
3823048b
JH
2862 *
2863 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2864 * shared string constants we point to, instead of generating a new
2865 * string for each instance. */
bfaa02d5 2866STATIC size_t
3823048b 2867S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
3bde2d43 2868 char* s = buffer;
bfaa02d5 2869 assert(maxlen >= 4);
3bde2d43
JH
2870 if (Perl_isinf(nv)) {
2871 if (nv < 0) {
2872 if (maxlen < 5) /* "-Inf\0" */
2873 return 0;
2874 *s++ = '-';
2875 } else if (plus) {
2876 *s++ = '+';
6e915616 2877 }
3bde2d43
JH
2878 *s++ = 'I';
2879 *s++ = 'n';
2880 *s++ = 'f';
2881 }
2882 else if (Perl_isnan(nv)) {
2883 *s++ = 'N';
2884 *s++ = 'a';
2885 *s++ = 'N';
2886 /* XXX optionally output the payload mantissa bits as
2887 * "(unsigned)" (to match the nan("...") C99 function,
2888 * or maybe as "(0xhhh...)" would make more sense...
2889 * provide a format string so that the user can decide?
2890 * NOTE: would affect the maxlen and assert() logic.*/
2891 }
2892 else {
2893 return 0;
bfaa02d5 2894 }
3bde2d43
JH
2895 assert((s == buffer + 3) || (s == buffer + 4));
2896 *s++ = 0;
2897 return s - buffer - 1; /* -1: excluding the zero byte */
bfaa02d5
JH
2898}
2899
2900/*
2901=for apidoc sv_2pv_flags
2902
796b6530 2903Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
c5608a1f 2904If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
bfaa02d5
JH
2905string if necessary. Normally invoked via the C<SvPV_flags> macro.
2906C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2907
2908=cut
2909*/
2910
2911char *
2912Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2913{
2914 char *s;
2915
2916 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2917
2918 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2919 && SvTYPE(sv) != SVt_PVFM);
2920 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2921 mg_get(sv);
2922 if (SvROK(sv)) {
2923 if (SvAMAGIC(sv)) {
2924 SV *tmpstr;
2925 if (flags & SV_SKIP_OVERLOAD)
2926 return NULL;
2927 tmpstr = AMG_CALLunary(sv, string_amg);
2928 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2929 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2930 /* Unwrap this: */
2931 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2932 */
2933
2934 char *pv;
2935 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2936 if (flags & SV_CONST_RETURN) {
2937 pv = (char *) SvPVX_const(tmpstr);
2938 } else {
2939 pv = (flags & SV_MUTABLE_RETURN)
2940 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2941 }
2942 if (lp)
2943 *lp = SvCUR(tmpstr);
2944 } else {
2945 pv = sv_2pv_flags(tmpstr, lp, flags);
2946 }
2947 if (SvUTF8(tmpstr))
2948 SvUTF8_on(sv);
2949 else
2950 SvUTF8_off(sv);
2951 return pv;
2952 }
2953 }
2954 {
2955 STRLEN len;
2956 char *retval;
2957 char *buffer;
2958 SV *const referent = SvRV(sv);
2959
2960 if (!referent) {
2961 len = 7;
2962 retval = buffer = savepvn("NULLREF", len);
2963 } else if (SvTYPE(referent) == SVt_REGEXP &&
2964 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2965 amagic_is_enabled(string_amg))) {
2966 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2967
2968 assert(re);
2969
2970 /* If the regex is UTF-8 we want the containing scalar to
2971 have an UTF-8 flag too */
2972 if (RX_UTF8(re))
2973 SvUTF8_on(sv);
2974 else
2975 SvUTF8_off(sv);
2976
2977 if (lp)
2978 *lp = RX_WRAPLEN(re);
2979
2980 return RX_WRAPPED(re);
2981 } else {
2982 const char *const typestr = sv_reftype(referent, 0);
2983 const STRLEN typelen = strlen(typestr);
2984 UV addr = PTR2UV(referent);
2985 const char *stashname = NULL;
2986 STRLEN stashnamelen = 0; /* hush, gcc */
2987 const char *buffer_end;
2988
2989 if (SvOBJECT(referent)) {
2990 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2991
2992 if (name) {
2993 stashname = HEK_KEY(name);
2994 stashnamelen = HEK_LEN(name);
2995
2996 if (HEK_UTF8(name)) {
2997 SvUTF8_on(sv);
2998 } else {
2999 SvUTF8_off(sv);
3000 }
3001 } else {
3002 stashname = "__ANON__";
3003 stashnamelen = 8;
3004 }
3005 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3006 + 2 * sizeof(UV) + 2 /* )\0 */;
3007 } else {
3008 len = typelen + 3 /* (0x */
3009 + 2 * sizeof(UV) + 2 /* )\0 */;
3010 }
fafee734 3011
4bac9ae4
CS
3012 Newx(buffer, len, char);
3013 buffer_end = retval = buffer + len;
3014
3015 /* Working backwards */
3016 *--retval = '\0';
3017 *--retval = ')';
3018 do {
3019 *--retval = PL_hexdigit[addr & 15];
3020 } while (addr >>= 4);
3021 *--retval = 'x';
3022 *--retval = '0';
3023 *--retval = '(';
3024
3025 retval -= typelen;
3026 memcpy(retval, typestr, typelen);
3027
3028 if (stashname) {
3029 *--retval = '=';
3030 retval -= stashnamelen;
3031 memcpy(retval, stashname, stashnamelen);
c080367d 3032 }
4bac9ae4
CS
3033 /* retval may not necessarily have reached the start of the
3034 buffer here. */
3035 assert (retval >= buffer);
3036
3037 len = buffer_end - retval - 1; /* -1 for that \0 */
463ee0b2 3038 }
cdb061a3 3039 if (lp)
4bac9ae4
CS
3040 *lp = len;
3041 SAVEFREEPV(buffer);
3042 return retval;
79072805 3043 }
79072805 3044 }
4bac9ae4
CS
3045
3046 if (SvPOKp(sv)) {
3047 if (lp)
3048 *lp = SvCUR(sv);
3049 if (flags & SV_MUTABLE_RETURN)
3050 return SvPVX_mutable(sv);
3051 if (flags & SV_CONST_RETURN)
3052 return (char *)SvPVX_const(sv);
3053 return SvPVX(sv);
3054 }
3055
3056 if (SvIOK(sv)) {
28e5dec8
JH
3057 /* I'm assuming that if both IV and NV are equally valid then
3058 converting the IV is going to be more efficient */
e1ec3a88 3059 const U32 isUIOK = SvIsUV(sv);
28e5dec8
JH
3060 char buf[TYPE_CHARS(UV)];
3061 char *ebuf, *ptr;
97a130b8 3062 STRLEN len;
28e5dec8
JH
3063
3064 if (SvTYPE(sv) < SVt_PVIV)
3065 sv_upgrade(sv, SVt_PVIV);
4ea1d550 3066 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
97a130b8 3067 len = ebuf - ptr;
5902b6a9 3068 /* inlined from sv_setpvn */
97a130b8
NC
3069 s = SvGROW_mutable(sv, len + 1);
3070 Move(ptr, s, len, char);
3071 s += len;
28e5dec8 3072 *s = '\0';
b127e37e 3073 SvPOK_on(sv);
28e5dec8 3074 }
4bac9ae4 3075 else if (SvNOK(sv)) {
79072805
LW
3076 if (SvTYPE(sv) < SVt_PVNV)
3077 sv_upgrade(sv, SVt_PVNV);
128eeacb
DD
3078 if (SvNVX(sv) == 0.0
3079#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3080 && !Perl_isnan(SvNVX(sv))
3081#endif
3082 ) {
29912d93
NC
3083 s = SvGROW_mutable(sv, 2);
3084 *s++ = '0';
3085 *s = '\0';
3086 } else {
5e85836e 3087 STRLEN len;
fb8cdbc5 3088 STRLEN size = 5; /* "-Inf\0" */
0c7e610f 3089
fb8cdbc5 3090 s = SvGROW_mutable(sv, size);
3823048b 3091 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
fb8cdbc5 3092 if (len > 0) {
0c7e610f 3093 s += len;
fb8cdbc5
JH
3094 SvPOK_on(sv);
3095 }
0c7e610f 3096 else {
0c7e610f 3097 /* some Xenix systems wipe out errno here */
fb8cdbc5
JH
3098 dSAVE_ERRNO;
3099
3840bff0
JH
3100 size =
3101 1 + /* sign */
3102 1 + /* "." */
3103 NV_DIG +
3104 1 + /* "e" */
3105 1 + /* sign */
3106 5 + /* exponent digits */
3107 1 + /* \0 */
3108 2; /* paranoia */
b127e37e 3109
fb8cdbc5 3110 s = SvGROW_mutable(sv, size);
b127e37e 3111#ifndef USE_LOCALE_NUMERIC
a4eca1d4
JH
3112 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3113
0c7e610f
JH
3114 SvPOK_on(sv);
3115#else
28acfe03 3116 {
3840bff0 3117 bool local_radix;
67d796ae
KW
3118 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3119 STORE_LC_NUMERIC_SET_TO_NEEDED();
3840bff0
JH
3120
3121 local_radix =
3122 PL_numeric_local &&
3123 PL_numeric_radix_sv &&
3124 SvUTF8(PL_numeric_radix_sv);
3125 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3126 size += SvLEN(PL_numeric_radix_sv) - 1;
3127 s = SvGROW_mutable(sv, size);
3128 }
3129
a4eca1d4 3130 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
0c7e610f
JH
3131
3132 /* If the radix character is UTF-8, and actually is in the
3133 * output, turn on the UTF-8 flag for the scalar */
3840bff0
JH
3134 if (local_radix &&
3135 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3136 SvUTF8_on(sv);
3137 }
3138
0c7e610f 3139 RESTORE_LC_NUMERIC();
28acfe03 3140 }
68e8f474 3141
0c7e610f
JH
3142 /* We don't call SvPOK_on(), because it may come to
3143 * pass that the locale changes so that the
3144 * stringification we just did is no longer correct. We
3145 * will have to re-stringify every time it is needed */
b127e37e 3146#endif
0c7e610f
JH
3147 RESTORE_ERRNO;
3148 }
3149 while (*s) s++;
bbce6d69 3150 }
79072805 3151 }
4bac9ae4
CS
3152 else if (isGV_with_GP(sv)) {
3153 GV *const gv = MUTABLE_GV(sv);
3154 SV *const buffer = sv_newmortal();
8d1c3e26 3155
4bac9ae4 3156 gv_efullname3(buffer, gv, "*");
180488f8 3157
4bac9ae4
CS
3158 assert(SvPOK(buffer));
3159 if (SvUTF8(buffer))
3160 SvUTF8_on(sv);
3161 if (lp)
3162 *lp = SvCUR(buffer);
3163 return SvPVX(buffer);
3164 }
8d919b0a
FC
3165 else if (isREGEXP(sv)) {
3166 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3167 return RX_WRAPPED((REGEXP *)sv);
3168 }
4bac9ae4 3169 else {
cdb061a3 3170 if (lp)
00b6aa41 3171 *lp = 0;
9f621bb0
NC
3172 if (flags & SV_UNDEF_RETURNS_NULL)
3173 return NULL;
4f62cd62 3174 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
9f621bb0 3175 report_uninit(sv);
4bac9ae4
CS
3176 /* Typically the caller expects that sv_any is not NULL now. */
3177 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
25da4f38 3178 sv_upgrade(sv, SVt_PV);
73d840c0 3179 return (char *)"";
79072805 3180 }
4bac9ae4 3181
cdb061a3 3182 {
823a54a3 3183 const STRLEN len = s - SvPVX_const(sv);
cdb061a3
NC
3184 if (lp)
3185 *lp = len;
3186 SvCUR_set(sv, len);
3187 }
1d7c1841 3188 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3f7c398e 3189 PTR2UV(sv),SvPVX_const(sv)));
4d84ee25
NC
3190 if (flags & SV_CONST_RETURN)
3191 return (char *)SvPVX_const(sv);
10516c54
NC
3192 if (flags & SV_MUTABLE_RETURN)
3193 return SvPVX_mutable(sv);
463ee0b2
LW
3194 return SvPVX(sv);
3195}
3196
645c22ef 3197/*
6050d10e
JP
3198=for apidoc sv_copypv
3199
3200Copies a stringified representation of the source SV into the
796b6530 3201destination SV. Automatically performs any necessary C<mg_get> and
54f0641b 3202coercion of numeric values into strings. Guaranteed to preserve
796b6530
KW
3203C<UTF8> flag even from overloaded objects. Similar in nature to
3204C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3205string. Mostly uses C<sv_2pv_flags> to do its work, except when that
6050d10e
JP
3206would lose the UTF-8'ness of the PV.
3207
4bac9ae4
CS
3208=for apidoc sv_copypv_nomg
3209
796b6530 3210Like C<sv_copypv>, but doesn't invoke get magic first.
4bac9ae4
CS
3211
3212=for apidoc sv_copypv_flags
3213
796b6530 3214Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
c5608a1f 3215has the C<SV_GMAGIC> bit set.
4bac9ae4 3216
6050d10e
JP
3217=cut
3218*/
3219
3220void
5aaab254 3221Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
4bac9ae4 3222{
446eaa42 3223 STRLEN len;
4bac9ae4 3224 const char *s;
7918f24d 3225
4bac9ae4 3226 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
7918f24d 3227
c77ed9ca 3228 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
cb50f42d 3229 sv_setpvn(dsv,s,len);
446eaa42 3230 if (SvUTF8(ssv))
cb50f42d 3231 SvUTF8_on(dsv);
446eaa42 3232 else
cb50f42d 3233 SvUTF8_off(dsv);
6050d10e
JP
3234}
3235
3236/*
645c22ef
DM
3237=for apidoc sv_2pvbyte
3238
796b6530 3239Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
1e54db1a 3240to its length. May cause the SV to be downgraded from UTF-8 as a
645c22ef
DM
3241side-effect.
3242
3243Usually accessed via the C<SvPVbyte> macro.
3244
3245=cut
3246*/
3247
7340a771 3248char *
5aaab254 3249Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3250{
7918f24d
NC
3251 PERL_ARGS_ASSERT_SV_2PVBYTE;
3252
48120f8f 3253 SvGETMAGIC(sv);
4499db73
FC
3254 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3255 || isGV_with_GP(sv) || SvROK(sv)) {
a901b181 3256 SV *sv2 = sv_newmortal();
48120f8f 3257 sv_copypv_nomg(sv2,sv);
a901b181
FC
3258 sv = sv2;
3259 }
0875d2fe 3260 sv_utf8_downgrade(sv,0);
71eb6d8c 3261 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771
GS
3262}
3263
645c22ef 3264/*
035cbb0e
RGS
3265=for apidoc sv_2pvutf8
3266
796b6530 3267Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
035cbb0e
RGS
3268to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3269
3270Usually accessed via the C<SvPVutf8> macro.
3271
3272=cut
3273*/
645c22ef 3274
7340a771 3275char *
5aaab254 3276Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
7340a771 3277{
7918f24d
NC
3278 PERL_ARGS_ASSERT_SV_2PVUTF8;
3279
4499db73
FC
3280 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3281 || isGV_with_GP(sv) || SvROK(sv))
fe46cbda 3282 sv = sv_mortalcopy(sv);
4bac9ae4
CS
3283 else
3284 SvGETMAGIC(sv);
3285 sv_utf8_upgrade_nomg(sv);
c3ec315f 3286 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
7340a771 3287}
1c846c1f 3288
7ee2227d 3289
645c22ef
DM
3290/*
3291=for apidoc sv_2bool
3292
796b6530
KW
3293This macro is only used by C<sv_true()> or its macro equivalent, and only if
3294the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3295It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
06c841cf
FC
3296
3297=for apidoc sv_2bool_flags
3298
796b6530
KW
3299This function is only used by C<sv_true()> and friends, and only if
3300the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3301contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
06c841cf 3302
645c22ef
DM
3303
3304=cut
3305*/
3306
463ee0b2 3307bool
9d176cd8 3308Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
463ee0b2 3309{
06c841cf 3310 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
7918f24d 3311
9d176cd8 3312 restart:
06c841cf 3313 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
463ee0b2 3314
a0d0e21e
LW
3315 if (!SvOK(sv))
3316 return 0;
3317 if (SvROK(sv)) {
fabdb6c0 3318 if (SvAMAGIC(sv)) {
31d632c3 3319 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
9d176cd8
DD
3320 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3321 bool svb;
3322 sv = tmpsv;
3323 if(SvGMAGICAL(sv)) {
3324 flags = SV_GMAGIC;
3325 goto restart; /* call sv_2bool */
3326 }
3327 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3328 else if(!SvOK(sv)) {
3329 svb = 0;
3330 }
3331 else if(SvPOK(sv)) {
3332 svb = SvPVXtrue(sv);
3333 }
3334 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3335 svb = (SvIOK(sv) && SvIVX(sv) != 0)
659c4b96 3336 || (SvNOK(sv) && SvNVX(sv) != 0.0);
9d176cd8
DD
3337 }
3338 else {
3339 flags = 0;
3340 goto restart; /* call sv_2bool_nomg */
3341 }
3342 return cBOOL(svb);
3343 }
fabdb6c0
AL
3344 }
3345 return SvRV(sv) != 0;
a0d0e21e 3346 }
85b7d9b3
FC
3347 if (isREGEXP(sv))
3348 return
3349 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
4bac9ae4 3350 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
79072805
LW
3351}
3352
c461cf8f
JH
3353/*
3354=for apidoc sv_utf8_upgrade
3355
78ea37eb 3356Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3357Forces the SV to string form if it is not already.
2bbc8d55 3358Will C<mg_get> on C<sv> if appropriate.
796b6530 3359Always sets the C<SvUTF8> flag to avoid future validity checks even
2bbc8d55
SP
3360if the whole string is the same in UTF-8 as not.
3361Returns the number of bytes in the converted string
c461cf8f 3362
0efd0472 3363This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3364use the Encode extension for that.
3365
fe749c9a
KW
3366=for apidoc sv_utf8_upgrade_nomg
3367
796b6530 3368Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
fe749c9a 3369
8d6d96c1
HS
3370=for apidoc sv_utf8_upgrade_flags
3371
78ea37eb 3372Converts the PV of an SV to its UTF-8-encoded form.
645c22ef 3373Forces the SV to string form if it is not already.
8d6d96c1 3374Always sets the SvUTF8 flag to avoid future validity checks even
960b0271
FC
3375if all the bytes are invariant in UTF-8.
3376If C<flags> has C<SV_GMAGIC> bit set,
2bbc8d55 3377will C<mg_get> on C<sv> if appropriate, else not.
2a590426 3378
796b6530 3379If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
2a590426
KW
3380will expand when converted to UTF-8, and skips the extra work of checking for
3381that. Typically this flag is used by a routine that has already parsed the
3382string and found such characters, and passes this information on so that the
3383work doesn't have to be repeated.
3384
3385Returns the number of bytes in the converted string.
8d6d96c1 3386
0efd0472 3387This is not a general purpose byte encoding to Unicode interface:
13a6c0e0
JH
3388use the Encode extension for that.
3389
2a590426 3390=for apidoc sv_utf8_upgrade_flags_grow
b3ab6785 3391
796b6530
KW
3392Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3393the number of unused bytes the string of C<sv> is guaranteed to have free after
2a590426
KW
3394it upon return. This allows the caller to reserve extra space that it intends
3395to fill, to avoid extra grows.
b3ab6785 3396
2a590426
KW
3397C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3398are implemented in terms of this function.
3399
3400Returns the number of bytes in the converted string (not including the spares).
3401
3402=cut
b3ab6785
KW
3403
3404(One might think that the calling routine could pass in the position of the
2a590426
KW
3405first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3406have to be found again. But that is not the case, because typically when the
3407caller is likely to use this flag, it won't be calling this routine unless it
3408finds something that won't fit into a byte. Otherwise it tries to not upgrade
3409and just use bytes. But some things that do fit into a byte are variants in
3410utf8, and the caller may not have been keeping track of these.)
b3ab6785 3411
6602b933
KW
3412If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3413C<NUL> isn't guaranteed due to having other routines do the work in some input
3414cases, or if the input is already flagged as being in utf8.
b3ab6785
KW
3415
3416The speed of this could perhaps be improved for many cases if someone wanted to
3417write a fast function that counts the number of variant characters in a string,
3418especially if it could return the position of the first one.
3419
8d6d96c1
HS
3420*/
3421
3422STRLEN
5aaab254 3423Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
8d6d96c1 3424{
b3ab6785 3425 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
7918f24d 3426
808c356f
RGS
3427 if (sv == &PL_sv_undef)
3428 return 0;
892f9127 3429 if (!SvPOK_nog(sv)) {
e0e62c2a 3430 STRLEN len = 0;
d52b7888
NC
3431 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3432 (void) sv_2pv_flags(sv,&len, flags);
b3ab6785
KW
3433 if (SvUTF8(sv)) {
3434 if (extra) SvGROW(sv, SvCUR(sv) + extra);
d52b7888 3435 return len;
b3ab6785 3436 }
d52b7888 3437 } else {
33fb6f35 3438 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
d52b7888 3439 }
e0e62c2a 3440 }
4411f3b6 3441
f5cee72b 3442 if (SvUTF8(sv)) {
b3ab6785 3443 if (extra) SvGROW(sv, SvCUR(sv) + extra);
5fec3b1d 3444 return SvCUR(sv);
f5cee72b 3445 }
5fec3b1d 3446
765f542d 3447 if (SvIsCOW(sv)) {
c56ed9f6 3448 S_sv_uncow(aTHX_ sv, 0);
db42d148
NIS
3449 }
3450
47e13f24 3451 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
ad2de1b2 3452 sv_recode_to_utf8(sv, _get_encoding());
b3ab6785
KW
3453 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3454 return SvCUR(sv);
3455 }
3456
4e93345f
KW
3457 if (SvCUR(sv) == 0) {
3458 if (extra) SvGROW(sv, extra);
3459 } else { /* Assume Latin-1/EBCDIC */
c4e7c712 3460 /* This function could be much more efficient if we
2bbc8d55 3461 * had a FLAG in SVs to signal if there are any variant
c4e7c712 3462 * chars in the PV. Given that there isn't such a flag
b3ab6785
KW
3463 * make the loop as fast as possible (although there are certainly ways
3464 * to speed this up, eg. through vectorization) */
3465 U8 * s = (U8 *) SvPVX_const(sv);
3466 U8 * e = (U8 *) SvEND(sv);
3467 U8 *t = s;
3468 STRLEN two_byte_count = 0;
c4e7c712 3469
b3ab6785
KW
3470 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3471
3472 /* See if really will need to convert to utf8. We mustn't rely on our
3473 * incoming SV being well formed and having a trailing '\0', as certain
3474 * code in pp_formline can send us partially built SVs. */
3475
c4e7c712 3476 while (t < e) {
53c1dcc0 3477 const U8 ch = *t++;
6f2d5cbc 3478 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
b3ab6785
KW
3479
3480 t--; /* t already incremented; re-point to first variant */
3481 two_byte_count = 1;
3482 goto must_be_utf8;
c4e7c712 3483 }
b3ab6785
KW
3484
3485 /* utf8 conversion not needed because all are invariants. Mark as
3486 * UTF-8 even if no variant - saves scanning loop */
c4e7c712 3487 SvUTF8_on(sv);
7f0bfbea 3488 if (extra) SvGROW(sv, SvCUR(sv) + extra);
b3ab6785
KW
3489 return SvCUR(sv);
3490
7b52d656 3491 must_be_utf8:
b3ab6785
KW
3492
3493 /* Here, the string should be converted to utf8, either because of an
3494 * input flag (two_byte_count = 0), or because a character that
3495 * requires 2 bytes was found (two_byte_count = 1). t points either to
3496 * the beginning of the string (if we didn't examine anything), or to
3497 * the first variant. In either case, everything from s to t - 1 will
3498 * occupy only 1 byte each on output.
3499 *
3500 * There are two main ways to convert. One is to create a new string
3501 * and go through the input starting from the beginning, appending each
3502 * converted value onto the new string as we go along. It's probably
3503 * best to allocate enough space in the string for the worst possible
3504 * case rather than possibly running out of space and having to
3505 * reallocate and then copy what we've done so far. Since everything
3506 * from s to t - 1 is invariant, the destination can be initialized
3507 * with these using a fast memory copy
3508 *
3509 * The other way is to figure out exactly how big the string should be
3510 * by parsing the entire input. Then you don't have to make it big
3511 * enough to handle the worst possible case, and more importantly, if
3512 * the string you already have is large enough, you don't have to
3513 * allocate a new string, you can copy the last character in the input
3514 * string to the final position(s) that will be occupied by the
3515 * converted string and go backwards, stopping at t, since everything
3516 * before that is invariant.
3517 *
3518 * There are advantages and disadvantages to each method.
3519 *
3520 * In the first method, we can allocate a new string, do the memory
3521 * copy from the s to t - 1, and then proceed through the rest of the
3522 * string byte-by-byte.
3523 *
3524 * In the second method, we proceed through the rest of the input
3525 * string just calculating how big the converted string will be. Then
3526 * there are two cases:
3527 * 1) if the string has enough extra space to handle the converted
3528 * value. We go backwards through the string, converting until we
3529 * get to the position we are at now, and then stop. If this
3530 * position is far enough along in the string, this method is
3531 * faster than the other method. If the memory copy were the same
3532 * speed as the byte-by-byte loop, that position would be about
3533 * half-way, as at the half-way mark, parsing to the end and back
3534 * is one complete string's parse, the same amount as starting
3535 * over and going all the way through. Actually, it would be
3536 * somewhat less than half-way, as it's faster to just count bytes
3537 * than to also copy, and we don't have the overhead of allocating
3538 * a new string, changing the scalar to use it, and freeing the
3539 * existing one. But if the memory copy is fast, the break-even
3540 * point is somewhere after half way. The counting loop could be
3541 * sped up by vectorization, etc, to move the break-even point
3542 * further towards the beginning.
3543 * 2) if the string doesn't have enough space to handle the converted
3544 * value. A new string will have to be allocated, and one might
3545 * as well, given that, start from the beginning doing the first
3546 * method. We've spent extra time parsing the string and in
3547 * exchange all we've gotten is that we know precisely how big to
3548 * make the new one. Perl is more optimized for time than space,
3549 * so this case is a loser.
3550 * So what I've decided to do is not use the 2nd method unless it is
3551 * guaranteed that a new string won't have to be allocated, assuming
3552 * the worst case. I also decided not to put any more conditions on it
3553 * than this, for now. It seems likely that, since the worst case is
3554 * twice as big as the unknown portion of the string (plus 1), we won't
3555 * be guaranteed enough space, causing us to go to the first method,
3556 * unless the string is short, or the first variant character is near
3557 * the end of it. In either of these cases, it seems best to use the
3558 * 2nd method. The only circumstance I can think of where this would
3559 * be really slower is if the string had once had much more data in it
3560 * than it does now, but there is still a substantial amount in it */
3561
3562 {
3563 STRLEN invariant_head = t - s;
3564 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3565 if (SvLEN(sv) < size) {
3566
3567 /* Here, have decided to allocate a new string */
3568
3569 U8 *dst;
3570 U8 *d;
3571
3572 Newx(dst, size, U8);
3573
3574 /* If no known invariants at the beginning of the input string,
3575 * set so starts from there. Otherwise, can use memory copy to
3576 * get up to where we are now, and then start from here */
3577
5b26a7b3 3578 if (invariant_head == 0) {
b3ab6785
KW
3579 d = dst;
3580 } else {
3581 Copy(s, dst, invariant_head, char);
3582 d = dst + invariant_head;
3583 }
3584
3585 while (t < e) {
55d09dc8
KW
3586 append_utf8_from_native_byte(*t, &d);
3587 t++;
b3ab6785
KW
3588 }
3589 *d = '\0';
3590 SvPV_free(sv); /* No longer using pre-existing string */
3591 SvPV_set(sv, (char*)dst);
3592 SvCUR_set(sv, d - dst);
3593 SvLEN_set(sv, size);
3594 } else {
3595
3596 /* Here, have decided to get the exact size of the string.
3597 * Currently this happens only when we know that there is
3598 * guaranteed enough space to fit the converted string, so
3599 * don't have to worry about growing. If two_byte_count is 0,
3600 * then t points to the first byte of the string which hasn't
3601 * been examined yet. Otherwise two_byte_count is 1, and t
3602 * points to the first byte in the string that will expand to
3603 * two. Depending on this, start examining at t or 1 after t.
3604 * */