1 /* sv.c
2 *
3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
5 * and others
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 *
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 /*
20 *
21 *
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.
28 */
29
30 #include "EXTERN.h"
31 #define PERL_IN_SV_C
32 #include "perl.h"
33 #include "regcomp.h"
34 #ifdef __VMS
35 # include <rms.h>
36 #endif
37
38 #ifdef __Lynx__
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
41 #endif
42
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
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
53 #endif
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
56 #endif
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
59 #endif
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
62 #endif
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
65 #endif
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
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 )
112
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
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 */
119 # define ASSERT_UTF8_CACHE(cache) \
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
124 #else
125 # define ASSERT_UTF8_CACHE(cache) NOOP
126 #endif
127
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
130
131 /* ============================================================================
132
133 =head1 Allocation and deallocation of SVs.
134
135 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
136 sv, av, hv...) contains type and reference count information, and for
137 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
138 contains fields specific to each type. Some types store all they need
139 in the head, so don't have a body.
140
141 In all but the most memory-paranoid configurations (ex: PURIFY), heads
142 and bodies are allocated out of arenas, which by default are
143 approximately 4K chunks of memory parcelled up into N heads or bodies.
144 Sv-bodies are allocated by their sv-type, guaranteeing size
145 consistency needed to allocate safely from arrays.
146
147 For SV-heads, the first slot in each arena is reserved, and holds a
148 link to the next arena, some flags, and a note of the number of slots.
149 Snaked through each arena chain is a linked list of free items; when
150 this becomes empty, an extra arena is allocated and divided up into N
151 items which are threaded into the free list.
152
153 SV-bodies are similar, but they use arena-sets by default, which
154 separate the link and info from the arena itself, and reclaim the 1st
155 slot in the arena. SV-bodies are further described later.
156
157 The following global variables are associated with arenas:
158
159 PL_sv_arenaroot pointer to list of SV arenas
160 PL_sv_root pointer to list of free SV structures
161
162 PL_body_arenas head of linked-list of body arenas
163 PL_body_roots[] array of pointers to list of free bodies of svtype
164 arrays are indexed by the svtype needed
165
166 A few special SV heads are not allocated from an arena, but are
167 instead directly created in the interpreter structure, eg PL_sv_undef.
168 The size of arenas can be changed from the default by setting
169 PERL_ARENA_SIZE appropriately at compile time.
170
171 The SV arena serves the secondary purpose of allowing still-live SVs
172 to be located and destroyed during final cleanup.
173
174 At the lowest level, the macros new_SV() and del_SV() grab and free
175 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
176 to return the SV to the free list with error checking.) new_SV() calls
177 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
178 SVs in the free list have their SvTYPE field set to all ones.
179
180 At the time of very final cleanup, sv_free_arenas() is called from
181 perl_destruct() to physically free all the arenas allocated since the
182 start of the interpreter.
183
184 The function visit() scans the SV arenas list, and calls a specified
185 function for each SV it finds which is still live - ie which has an SvTYPE
186 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
187 following functions (specified as [function that calls visit()] / [function
188 called by visit() for each SV]):
189
190 sv_report_used() / do_report_used()
191 dump all remaining SVs (debugging aid)
192
193 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
194 do_clean_named_io_objs(),do_curse()
195 Attempt to free all objects pointed to by RVs,
196 try to do the same for all objects indir-
197 ectly referenced by typeglobs too, and
198 then do a final sweep, cursing any
199 objects that remain. Called once from
200 perl_destruct(), prior to calling sv_clean_all()
201 below.
202
203 sv_clean_all() / do_clean_all()
204 SvREFCNT_dec(sv) each remaining SV, possibly
205 triggering an sv_free(). It also sets the
206 SVf_BREAK flag on the SV to indicate that the
207 refcnt has been artificially lowered, and thus
208 stopping sv_free() from giving spurious warnings
209 about SVs which unexpectedly have a refcnt
210 of zero. called repeatedly from perl_destruct()
211 until there are no SVs left.
212
213 =head2 Arena allocator API Summary
214
215 Private API to rest of sv.c
216
217 new_SV(), del_SV(),
218
219 new_XPVNV(), del_XPVGV(),
220 etc
221
222 Public API:
223
224 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
225
226 =cut
227
228 * ========================================================================= */
229
230 /*
231 * "A time to plant, and a time to uproot what was planted..."
232 */
233
234 #ifdef PERL_MEM_LOG
235 # define MEM_LOG_NEW_SV(sv, file, line, func) \
236 Perl_mem_log_new_sv(sv, file, line, func)
237 # define MEM_LOG_DEL_SV(sv, file, line, func) \
238 Perl_mem_log_del_sv(sv, file, line, func)
239 #else
240 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
241 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
242 #endif
243
244 #ifdef DEBUG_LEAKING_SCALARS
245 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
246 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 } STMT_END
248 # define DEBUG_SV_SERIAL(sv) \
249 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
250 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 #else
252 # define FREE_SV_DEBUG_FILE(sv)
253 # define DEBUG_SV_SERIAL(sv) NOOP
254 #endif
255
256 #ifdef PERL_POISON
257 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
258 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
259 /* Whilst I'd love to do this, it seems that things like to check on
260 unreferenced scalars
261 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 */
263 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
264 PoisonNew(&SvREFCNT(sv), 1, U32)
265 #else
266 # define SvARENA_CHAIN(sv) SvANY(sv)
267 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
268 # define POISON_SV_HEAD(sv)
269 #endif
270
271 /* Mark an SV head as unused, and add to free list.
272 *
273 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
274 * its refcount artificially decremented during global destruction, so
275 * there may be dangling pointers to it. The last thing we want in that
276 * case is for it to be reused. */
277
278 #define plant_SV(p) \
279 STMT_START { \
280 const U32 old_flags = SvFLAGS(p); \
281 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
282 DEBUG_SV_SERIAL(p); \
283 FREE_SV_DEBUG_FILE(p); \
284 POISON_SV_HEAD(p); \
285 SvFLAGS(p) = SVTYPEMASK; \
286 if (!(old_flags & SVf_BREAK)) { \
287 SvARENA_CHAIN_SET(p, PL_sv_root); \
288 PL_sv_root = (p); \
289 } \
290 --PL_sv_count; \
291 } STMT_END
292
293 #define uproot_SV(p) \
294 STMT_START { \
295 (p) = PL_sv_root; \
296 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
297 ++PL_sv_count; \
298 } STMT_END
299
300
301 /* make some more SVs by adding another arena */
302
303 STATIC SV*
S_more_sv(pTHX)304 S_more_sv(pTHX)
305 {
306 SV* sv;
307 char *chunk; /* must use New here to match call to */
308 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
309 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
310 uproot_SV(sv);
311 return sv;
312 }
313
314 /* new_SV(): return a new, empty SV head */
315
316 #ifdef DEBUG_LEAKING_SCALARS
317 /* provide a real function for a debugger to play with */
318 STATIC SV*
S_new_SV(pTHX_ const char * file,int line,const char * func)319 S_new_SV(pTHX_ const char *file, int line, const char *func)
320 {
321 SV* sv;
322
323 if (PL_sv_root)
324 uproot_SV(sv);
325 else
326 sv = S_more_sv(aTHX);
327 SvANY(sv) = 0;
328 SvREFCNT(sv) = 1;
329 SvFLAGS(sv) = 0;
330 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
331 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
332 ? PL_parser->copline
333 : PL_curcop
334 ? CopLINE(PL_curcop)
335 : 0
336 );
337 sv->sv_debug_inpad = 0;
338 sv->sv_debug_parent = NULL;
339 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340
341 sv->sv_debug_serial = PL_sv_serial++;
342
343 MEM_LOG_NEW_SV(sv, file, line, func);
344 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
345 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
346
347 return sv;
348 }
349 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
350
351 #else
352 # define new_SV(p) \
353 STMT_START { \
354 if (PL_sv_root) \
355 uproot_SV(p); \
356 else \
357 (p) = S_more_sv(aTHX); \
358 SvANY(p) = 0; \
359 SvREFCNT(p) = 1; \
360 SvFLAGS(p) = 0; \
361 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
362 } STMT_END
363 #endif
364
365
366 /* del_SV(): return an empty SV head to the free list */
367
368 #ifdef DEBUGGING
369
370 #define del_SV(p) \
371 STMT_START { \
372 if (DEBUG_D_TEST) \
373 del_sv(p); \
374 else \
375 plant_SV(p); \
376 } STMT_END
377
378 STATIC void
S_del_sv(pTHX_ SV * p)379 S_del_sv(pTHX_ SV *p)
380 {
381 PERL_ARGS_ASSERT_DEL_SV;
382
383 if (DEBUG_D_TEST) {
384 SV* sva;
385 bool ok = 0;
386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
387 const SV * const sv = sva + 1;
388 const SV * const svend = &sva[SvREFCNT(sva)];
389 if (p >= sv && p < svend) {
390 ok = 1;
391 break;
392 }
393 }
394 if (!ok) {
395 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
396 "Attempt to free non-arena SV: 0x%" UVxf
397 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
398 return;
399 }
400 }
401 plant_SV(p);
402 }
403
404 #else /* ! DEBUGGING */
405
406 #define del_SV(p) plant_SV(p)
407
408 #endif /* DEBUGGING */
409
410
411 /*
412 =head1 SV Manipulation Functions
413
414 =for apidoc sv_add_arena
415
416 Given a chunk of memory, link it to the head of the list of arenas,
417 and split it into a list of free SVs.
418
419 =cut
420 */
421
422 static void
S_sv_add_arena(pTHX_ char * const ptr,const U32 size,const U32 flags)423 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 {
425 SV *const sva = MUTABLE_SV(ptr);
426 SV* sv;
427 SV* svend;
428
429 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430
431 /* The first SV in an arena isn't an SV. */
432 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
433 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
434 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435
436 PL_sv_arenaroot = sva;
437 PL_sv_root = sva + 1;
438
439 svend = &sva[SvREFCNT(sva) - 1];
440 sv = sva + 1;
441 while (sv < svend) {
442 SvARENA_CHAIN_SET(sv, (sv + 1));
443 #ifdef DEBUGGING
444 SvREFCNT(sv) = 0;
445 #endif
446 /* Must always set typemask because it's always checked in on cleanup
447 when the arenas are walked looking for objects. */
448 SvFLAGS(sv) = SVTYPEMASK;
449 sv++;
450 }
451 SvARENA_CHAIN_SET(sv, 0);
452 #ifdef DEBUGGING
453 SvREFCNT(sv) = 0;
454 #endif
455 SvFLAGS(sv) = SVTYPEMASK;
456 }
457
458 /* visit(): call the named function for each non-free SV in the arenas
459 * whose flags field matches the flags/mask args. */
460
461 STATIC I32
S_visit(pTHX_ SVFUNC_t f,const U32 flags,const U32 mask)462 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
463 {
464 SV* sva;
465 I32 visited = 0;
466
467 PERL_ARGS_ASSERT_VISIT;
468
469 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
470 const SV * const svend = &sva[SvREFCNT(sva)];
471 SV* sv;
472 for (sv = sva + 1; sv < svend; ++sv) {
473 if (SvTYPE(sv) != (svtype)SVTYPEMASK
474 && (sv->sv_flags & mask) == flags
475 && SvREFCNT(sv))
476 {
477 (*f)(aTHX_ sv);
478 ++visited;
479 }
480 }
481 }
482 return visited;
483 }
484
485 #ifdef DEBUGGING
486
487 /* called by sv_report_used() for each live SV */
488
489 static void
do_report_used(pTHX_ SV * const sv)490 do_report_used(pTHX_ SV *const sv)
491 {
492 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
493 PerlIO_printf(Perl_debug_log, "****\n");
494 sv_dump(sv);
495 }
496 }
497 #endif
498
499 /*
500 =for apidoc sv_report_used
501
502 Dump the contents of all SVs not yet freed (debugging aid).
503
504 =cut
505 */
506
507 void
Perl_sv_report_used(pTHX)508 Perl_sv_report_used(pTHX)
509 {
510 #ifdef DEBUGGING
511 visit(do_report_used, 0, 0);
512 #else
513 PERL_UNUSED_CONTEXT;
514 #endif
515 }
516
517 /* called by sv_clean_objs() for each live SV */
518
519 static void
do_clean_objs(pTHX_ SV * const ref)520 do_clean_objs(pTHX_ SV *const ref)
521 {
522 assert (SvROK(ref));
523 {
524 SV * const target = SvRV(ref);
525 if (SvOBJECT(target)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
527 if (SvWEAKREF(ref)) {
528 sv_del_backref(target, ref);
529 SvWEAKREF_off(ref);
530 SvRV_set(ref, NULL);
531 } else {
532 SvROK_off(ref);
533 SvRV_set(ref, NULL);
534 SvREFCNT_dec_NN(target);
535 }
536 }
537 }
538 }
539
540
541 /* clear any slots in a GV which hold objects - except IO;
542 * called by sv_clean_objs() for each live GV */
543
544 static void
do_clean_named_objs(pTHX_ SV * const sv)545 do_clean_named_objs(pTHX_ SV *const sv)
546 {
547 SV *obj;
548 assert(SvTYPE(sv) == SVt_PVGV);
549 assert(isGV_with_GP(sv));
550 if (!GvGP(sv))
551 return;
552
553 /* freeing GP entries may indirectly free the current GV;
554 * hold onto it while we mess with the GP slots */
555 SvREFCNT_inc(sv);
556
557 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
558 DEBUG_D((PerlIO_printf(Perl_debug_log,
559 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 GvSV(sv) = NULL;
561 SvREFCNT_dec_NN(obj);
562 }
563 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
564 DEBUG_D((PerlIO_printf(Perl_debug_log,
565 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 GvAV(sv) = NULL;
567 SvREFCNT_dec_NN(obj);
568 }
569 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
570 DEBUG_D((PerlIO_printf(Perl_debug_log,
571 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 GvHV(sv) = NULL;
573 SvREFCNT_dec_NN(obj);
574 }
575 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
576 DEBUG_D((PerlIO_printf(Perl_debug_log,
577 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 GvCV_set(sv, NULL);
579 SvREFCNT_dec_NN(obj);
580 }
581 SvREFCNT_dec_NN(sv); /* undo the inc above */
582 }
583
584 /* clear any IO slots in a GV which hold objects (except stderr, defout);
585 * called by sv_clean_objs() for each live GV */
586
587 static void
do_clean_named_io_objs(pTHX_ SV * const sv)588 do_clean_named_io_objs(pTHX_ SV *const sv)
589 {
590 SV *obj;
591 assert(SvTYPE(sv) == SVt_PVGV);
592 assert(isGV_with_GP(sv));
593 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
594 return;
595
596 SvREFCNT_inc(sv);
597 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
598 DEBUG_D((PerlIO_printf(Perl_debug_log,
599 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 GvIOp(sv) = NULL;
601 SvREFCNT_dec_NN(obj);
602 }
603 SvREFCNT_dec_NN(sv); /* undo the inc above */
604 }
605
606 /* Void wrapper to pass to visit() */
607 static void
do_curse(pTHX_ SV * const sv)608 do_curse(pTHX_ SV * const sv) {
609 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
610 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
611 return;
612 (void)curse(sv, 0);
613 }
614
615 /*
616 =for apidoc sv_clean_objs
617
618 Attempt to destroy all objects not yet freed.
619
620 =cut
621 */
622
623 void
Perl_sv_clean_objs(pTHX)624 Perl_sv_clean_objs(pTHX)
625 {
626 GV *olddef, *olderr;
627 PL_in_clean_objs = TRUE;
628 visit(do_clean_objs, SVf_ROK, SVf_ROK);
629 /* Some barnacles may yet remain, clinging to typeglobs.
630 * Run the non-IO destructors first: they may want to output
631 * error messages, close files etc */
632 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
634 /* And if there are some very tenacious barnacles clinging to arrays,
635 closures, or what have you.... */
636 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
637 olddef = PL_defoutgv;
638 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
639 if (olddef && isGV_with_GP(olddef))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
641 olderr = PL_stderrgv;
642 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
643 if (olderr && isGV_with_GP(olderr))
644 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
645 SvREFCNT_dec(olddef);
646 PL_in_clean_objs = FALSE;
647 }
648
649 /* called by sv_clean_all() for each live SV */
650
651 static void
do_clean_all(pTHX_ SV * const sv)652 do_clean_all(pTHX_ SV *const sv)
653 {
654 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
655 /* don't clean pid table and strtab */
656 return;
657 }
658 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
659 SvFLAGS(sv) |= SVf_BREAK;
660 SvREFCNT_dec_NN(sv);
661 }
662
663 /*
664 =for apidoc sv_clean_all
665
666 Decrement the refcnt of each remaining SV, possibly triggering a
667 cleanup. This function may have to be called multiple times to free
668 SVs which are in complex self-referential hierarchies.
669
670 =cut
671 */
672
673 I32
Perl_sv_clean_all(pTHX)674 Perl_sv_clean_all(pTHX)
675 {
676 I32 cleaned;
677 PL_in_clean_all = TRUE;
678 cleaned = visit(do_clean_all, 0,0);
679 return cleaned;
680 }
681
682 /*
683 ARENASETS: a meta-arena implementation which separates arena-info
684 into struct arena_set, which contains an array of struct
685 arena_descs, each holding info for a single arena. By separating
686 the meta-info from the arena, we recover the 1st slot, formerly
687 borrowed for list management. The arena_set is about the size of an
688 arena, avoiding the needless malloc overhead of a naive linked-list.
689
690 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
691 memory in the last arena-set (1/2 on average). In trade, we get
692 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
693 smaller types). The recovery of the wasted space allows use of
694 small arenas for large, rare body types, by changing array* fields
695 in body_details_by_type[] below.
696 */
697 struct arena_desc {
698 char *arena; /* the raw storage, allocated aligned */
699 size_t size; /* its size ~4k typ */
700 svtype utype; /* bodytype stored in arena */
701 };
702
703 struct arena_set;
704
705 /* Get the maximum number of elements in set[] such that struct arena_set
706 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
707 therefore likely to be 1 aligned memory page. */
708
709 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
710 - 2 * sizeof(int)) / sizeof (struct arena_desc))
711
712 struct arena_set {
713 struct arena_set* next;
714 unsigned int set_size; /* ie ARENAS_PER_SET */
715 unsigned int curr; /* index of next available arena-desc */
716 struct arena_desc set[ARENAS_PER_SET];
717 };
718
719 /*
720 =for apidoc sv_free_arenas
721
722 Deallocate the memory used by all arenas. Note that all the individual SV
723 heads and bodies within the arenas must already have been freed.
724
725 =cut
726
727 */
728 void
Perl_sv_free_arenas(pTHX)729 Perl_sv_free_arenas(pTHX)
730 {
731 SV* sva;
732 SV* svanext;
733 unsigned int i;
734
735 /* Free arenas here, but be careful about fake ones. (We assume
736 contiguity of the fake ones with the corresponding real ones.) */
737
738 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
739 svanext = MUTABLE_SV(SvANY(sva));
740 while (svanext && SvFAKE(svanext))
741 svanext = MUTABLE_SV(SvANY(svanext));
742
743 if (!SvFAKE(sva))
744 Safefree(sva);
745 }
746
747 {
748 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
749
750 while (aroot) {
751 struct arena_set *current = aroot;
752 i = aroot->curr;
753 while (i--) {
754 assert(aroot->set[i].arena);
755 Safefree(aroot->set[i].arena);
756 }
757 aroot = aroot->next;
758 Safefree(current);
759 }
760 }
761 PL_body_arenas = 0;
762
763 i = PERL_ARENA_ROOTS_SIZE;
764 while (i--)
765 PL_body_roots[i] = 0;
766
767 PL_sv_arenaroot = 0;
768 PL_sv_root = 0;
769 }
770
771 /*
772 Here are mid-level routines that manage the allocation of bodies out
773 of the various arenas. There are 4 kinds of arenas:
774
775 1. SV-head arenas, which are discussed and handled above
776 2. regular body arenas
777 3. arenas for reduced-size bodies
778 4. Hash-Entry arenas
779
780 Arena types 2 & 3 are chained by body-type off an array of
781 arena-root pointers, which is indexed by svtype. Some of the
782 larger/less used body types are malloced singly, since a large
783 unused block of them is wasteful. Also, several svtypes dont have
784 bodies; the data fits into the sv-head itself. The arena-root
785 pointer thus has a few unused root-pointers (which may be hijacked
786 later for arena type 4)
787
788 3 differs from 2 as an optimization; some body types have several
789 unused fields in the front of the structure (which are kept in-place
790 for consistency). These bodies can be allocated in smaller chunks,
791 because the leading fields arent accessed. Pointers to such bodies
792 are decremented to point at the unused 'ghost' memory, knowing that
793 the pointers are used with offsets to the real memory.
794
795 Allocation of SV-bodies is similar to SV-heads, differing as follows;
796 the allocation mechanism is used for many body types, so is somewhat
797 more complicated, it uses arena-sets, and has no need for still-live
798 SV detection.
799
800 At the outermost level, (new|del)_X*V macros return bodies of the
801 appropriate type. These macros call either (new|del)_body_type or
802 (new|del)_body_allocated macro pairs, depending on specifics of the
803 type. Most body types use the former pair, the latter pair is used to
804 allocate body types with "ghost fields".
805
806 "ghost fields" are fields that are unused in certain types, and
807 consequently don't need to actually exist. They are declared because
808 they're part of a "base type", which allows use of functions as
809 methods. The simplest examples are AVs and HVs, 2 aggregate types
810 which don't use the fields which support SCALAR semantics.
811
812 For these types, the arenas are carved up into appropriately sized
813 chunks, we thus avoid wasted memory for those unaccessed members.
814 When bodies are allocated, we adjust the pointer back in memory by the
815 size of the part not allocated, so it's as if we allocated the full
816 structure. (But things will all go boom if you write to the part that
817 is "not there", because you'll be overwriting the last members of the
818 preceding structure in memory.)
819
820 We calculate the correction using the STRUCT_OFFSET macro on the first
821 member present. If the allocated structure is smaller (no initial NV
822 actually allocated) then the net effect is to subtract the size of the NV
823 from the pointer, to return a new pointer as if an initial NV were actually
824 allocated. (We were using structures named *_allocated for this, but
825 this turned out to be a subtle bug, because a structure without an NV
826 could have a lower alignment constraint, but the compiler is allowed to
827 optimised accesses based on the alignment constraint of the actual pointer
828 to the full structure, for example, using a single 64 bit load instruction
829 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
830
831 This is the same trick as was used for NV and IV bodies. Ironically it
832 doesn't need to be used for NV bodies any more, because NV is now at
833 the start of the structure. IV bodies, and also in some builds NV bodies,
834 don't need it either, because they are no longer allocated.
835
836 In turn, the new_body_* allocators call S_new_body(), which invokes
837 new_body_inline macro, which takes a lock, and takes a body off the
838 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
839 necessary to refresh an empty list. Then the lock is released, and
840 the body is returned.
841
842 Perl_more_bodies allocates a new arena, and carves it up into an array of N
843 bodies, which it strings into a linked list. It looks up arena-size
844 and body-size from the body_details table described below, thus
845 supporting the multiple body-types.
846
847 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848 the (new|del)_X*V macros are mapped directly to malloc/free.
849
850 For each sv-type, struct body_details bodies_by_type[] carries
851 parameters which control these aspects of SV handling:
852
853 Arena_size determines whether arenas are used for this body type, and if
854 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855 zero, forcing individual mallocs and frees.
856
857 Body_size determines how big a body is, and therefore how many fit into
858 each arena. Offset carries the body-pointer adjustment needed for
859 "ghost fields", and is used in *_allocated macros.
860
861 But its main purpose is to parameterize info needed in
862 Perl_sv_upgrade(). The info here dramatically simplifies the function
863 vs the implementation in 5.8.8, making it table-driven. All fields
864 are used for this, except for arena_size.
865
866 For the sv-types that have no bodies, arenas are not used, so those
867 PL_body_roots[sv_type] are unused, and can be overloaded. In
868 something of a special case, SVt_NULL is borrowed for HE arenas;
869 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
870 bodies_by_type[SVt_NULL] slot is not used, as the table is not
871 available in hv.c.
872
873 */
874
875 struct body_details {
876 U8 body_size; /* Size to allocate */
877 U8 copy; /* Size of structure to copy (may be shorter) */
878 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
879 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
880 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
881 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
882 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
883 U32 arena_size; /* Size of arena to allocate */
884 };
885
886 #define HADNV FALSE
887 #define NONV TRUE
888
889
890 #ifdef PURIFY
891 /* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893 #define HASARENA FALSE
894 #else
895 #define HASARENA TRUE
896 #endif
897 #define NOARENA FALSE
898
899 /* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
905 declarations.
906 */
907 #define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909 #define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913 #define FIT_ARENA(count,body_size) \
914 (U32)(count \
915 ? FIT_ARENAn (count, body_size) \
916 : FIT_ARENA0 (body_size))
917
918 /* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
921
922 #define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
925
926 static const struct body_details bodies_by_type[] = {
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
929
930 /* IVs are in the head, so the allocation size is 0. */
931 { 0,
932 sizeof(IV), /* This is used to copy out the IV body. */
933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
934 NOARENA /* IVS don't need an arena */, 0
935 },
936
937 #if NVSIZE <= IVSIZE
938 { 0, sizeof(NV),
939 STRUCT_OFFSET(XPVNV, xnv_u),
940 SVt_NV, FALSE, HADNV, NOARENA, 0 },
941 #else
942 { sizeof(NV), sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
945 #endif
946
947 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_PV, FALSE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
952
953 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_INVLIST, TRUE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
958
959 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVIV, FALSE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
964
965 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
966 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVNV, FALSE, HADNV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
970
971 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
972 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
973
974 { sizeof(regexp),
975 sizeof(regexp),
976 0,
977 SVt_REGEXP, TRUE, NONV, HASARENA,
978 FIT_ARENA(0, sizeof(regexp))
979 },
980
981 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
982 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
983
984 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
985 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
986
987 { sizeof(XPVAV),
988 copy_length(XPVAV, xav_alloc),
989 0,
990 SVt_PVAV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVAV)) },
992
993 { sizeof(XPVHV),
994 copy_length(XPVHV, xhv_max),
995 0,
996 SVt_PVHV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVHV)) },
998
999 { sizeof(XPVCV),
1000 sizeof(XPVCV),
1001 0,
1002 SVt_PVCV, TRUE, NONV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVCV)) },
1004
1005 { sizeof(XPVFM),
1006 sizeof(XPVFM),
1007 0,
1008 SVt_PVFM, TRUE, NONV, NOARENA,
1009 FIT_ARENA(20, sizeof(XPVFM)) },
1010
1011 { sizeof(XPVIO),
1012 sizeof(XPVIO),
1013 0,
1014 SVt_PVIO, TRUE, NONV, HASARENA,
1015 FIT_ARENA(24, sizeof(XPVIO)) },
1016 };
1017
1018 #define new_body_allocated(sv_type) \
1019 (void *)((char *)S_new_body(aTHX_ sv_type) \
1020 - bodies_by_type[sv_type].offset)
1021
1022 /* return a thing to the free list */
1023
1024 #define del_body(thing, root) \
1025 STMT_START { \
1026 void ** const thing_copy = (void **)thing; \
1027 *thing_copy = *root; \
1028 *root = (void*)thing_copy; \
1029 } STMT_END
1030
1031 #ifdef PURIFY
1032 #if !(NVSIZE <= IVSIZE)
1033 # define new_XNV() safemalloc(sizeof(XPVNV))
1034 #endif
1035 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1036 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1037
1038 #define del_XPVGV(p) safefree(p)
1039
1040 #else /* !PURIFY */
1041
1042 #if !(NVSIZE <= IVSIZE)
1043 # define new_XNV() new_body_allocated(SVt_NV)
1044 #endif
1045 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1046 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1047
1048 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1049 &PL_body_roots[SVt_PVGV])
1050
1051 #endif /* PURIFY */
1052
1053 /* no arena for you! */
1054
1055 #define new_NOARENA(details) \
1056 safemalloc((details)->body_size + (details)->offset)
1057 #define new_NOARENAZ(details) \
1058 safecalloc((details)->body_size + (details)->offset, 1)
1059
1060 void *
Perl_more_bodies(pTHX_ const svtype sv_type,const size_t body_size,const size_t arena_size)1061 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1062 const size_t arena_size)
1063 {
1064 void ** const root = &PL_body_roots[sv_type];
1065 struct arena_desc *adesc;
1066 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1067 unsigned int curr;
1068 char *start;
1069 const char *end;
1070 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1071 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1072 dVAR;
1073 #endif
1074 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
1075 static bool done_sanity_check;
1076
1077 /* PERL_GLOBAL_STRUCT cannot coexist with global
1078 * variables like done_sanity_check. */
1079 if (!done_sanity_check) {
1080 unsigned int i = SVt_LAST;
1081
1082 done_sanity_check = TRUE;
1083
1084 while (i--)
1085 assert (bodies_by_type[i].type == i);
1086 }
1087 #endif
1088
1089 assert(arena_size);
1090
1091 /* may need new arena-set to hold new arena */
1092 if (!aroot || aroot->curr >= aroot->set_size) {
1093 struct arena_set *newroot;
1094 Newxz(newroot, 1, struct arena_set);
1095 newroot->set_size = ARENAS_PER_SET;
1096 newroot->next = aroot;
1097 aroot = newroot;
1098 PL_body_arenas = (void *) newroot;
1099 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1100 }
1101
1102 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1103 curr = aroot->curr++;
1104 adesc = &(aroot->set[curr]);
1105 assert(!adesc->arena);
1106
1107 Newx(adesc->arena, good_arena_size, char);
1108 adesc->size = good_arena_size;
1109 adesc->utype = sv_type;
1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1111 curr, (void*)adesc->arena, (UV)good_arena_size));
1112
1113 start = (char *) adesc->arena;
1114
1115 /* Get the address of the byte after the end of the last body we can fit.
1116 Remember, this is integer division: */
1117 end = start + good_arena_size / body_size * body_size;
1118
1119 /* computed count doesn't reflect the 1st slot reservation */
1120 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1121 DEBUG_m(PerlIO_printf(Perl_debug_log,
1122 "arena %p end %p arena-size %d (from %d) type %d "
1123 "size %d ct %d\n",
1124 (void*)start, (void*)end, (int)good_arena_size,
1125 (int)arena_size, sv_type, (int)body_size,
1126 (int)good_arena_size / (int)body_size));
1127 #else
1128 DEBUG_m(PerlIO_printf(Perl_debug_log,
1129 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1130 (void*)start, (void*)end,
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
1133 #endif
1134 *root = (void *)start;
1135
1136 while (1) {
1137 /* Where the next body would start: */
1138 char * const next = start + body_size;
1139
1140 if (next >= end) {
1141 /* This is the last body: */
1142 assert(next == end);
1143
1144 *(void **)start = 0;
1145 return *root;
1146 }
1147
1148 *(void**) start = (void *)next;
1149 start = next;
1150 }
1151 }
1152
1153 /* grab a new thing from the free list, allocating more if necessary.
1154 The inline version is used for speed in hot routines, and the
1155 function using it serves the rest (unless PURIFY).
1156 */
1157 #define new_body_inline(xpv, sv_type) \
1158 STMT_START { \
1159 void ** const r3wt = &PL_body_roots[sv_type]; \
1160 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1161 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1162 bodies_by_type[sv_type].body_size,\
1163 bodies_by_type[sv_type].arena_size)); \
1164 *(r3wt) = *(void**)(xpv); \
1165 } STMT_END
1166
1167 #ifndef PURIFY
1168
1169 STATIC void *
S_new_body(pTHX_ const svtype sv_type)1170 S_new_body(pTHX_ const svtype sv_type)
1171 {
1172 void *xpv;
1173 new_body_inline(xpv, sv_type);
1174 return xpv;
1175 }
1176
1177 #endif
1178
1179 static const struct body_details fake_rv =
1180 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1181
1182 /*
1183 =for apidoc sv_upgrade
1184
1185 Upgrade an SV to a more complex form. Generally adds a new body type to the
1186 SV, then copies across as much information as possible from the old body.
1187 It croaks if the SV is already in a more complex form than requested. You
1188 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1189 before calling C<sv_upgrade>, and hence does not croak. See also
1190 C<L</svtype>>.
1191
1192 =cut
1193 */
1194
1195 void
Perl_sv_upgrade(pTHX_ SV * const sv,svtype new_type)1196 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1197 {
1198 void* old_body;
1199 void* new_body;
1200 const svtype old_type = SvTYPE(sv);
1201 const struct body_details *new_type_details;
1202 const struct body_details *old_type_details
1203 = bodies_by_type + old_type;
1204 SV *referent = NULL;
1205
1206 PERL_ARGS_ASSERT_SV_UPGRADE;
1207
1208 if (old_type == new_type)
1209 return;
1210
1211 /* This clause was purposefully added ahead of the early return above to
1212 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1213 inference by Nick I-S that it would fix other troublesome cases. See
1214 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1215
1216 Given that shared hash key scalars are no longer PVIV, but PV, there is
1217 no longer need to unshare so as to free up the IVX slot for its proper
1218 purpose. So it's safe to move the early return earlier. */
1219
1220 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1221 sv_force_normal_flags(sv, 0);
1222 }
1223
1224 old_body = SvANY(sv);
1225
1226 /* Copying structures onto other structures that have been neatly zeroed
1227 has a subtle gotcha. Consider XPVMG
1228
1229 +------+------+------+------+------+-------+-------+
1230 | NV | CUR | LEN | IV | MAGIC | STASH |
1231 +------+------+------+------+------+-------+-------+
1232 0 4 8 12 16 20 24 28
1233
1234 where NVs are aligned to 8 bytes, so that sizeof that structure is
1235 actually 32 bytes long, with 4 bytes of padding at the end:
1236
1237 +------+------+------+------+------+-------+-------+------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1239 +------+------+------+------+------+-------+-------+------+
1240 0 4 8 12 16 20 24 28 32
1241
1242 so what happens if you allocate memory for this structure:
1243
1244 +------+------+------+------+------+-------+-------+------+------+...
1245 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1246 +------+------+------+------+------+-------+-------+------+------+...
1247 0 4 8 12 16 20 24 28 32 36
1248
1249 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1250 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1251 started out as zero once, but it's quite possible that it isn't. So now,
1252 rather than a nicely zeroed GP, you have it pointing somewhere random.
1253 Bugs ensue.
1254
1255 (In fact, GP ends up pointing at a previous GP structure, because the
1256 principle cause of the padding in XPVMG getting garbage is a copy of
1257 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1258 this happens to be moot because XPVGV has been re-ordered, with GP
1259 no longer after STASH)
1260
1261 So we are careful and work out the size of used parts of all the
1262 structures. */
1263
1264 switch (old_type) {
1265 case SVt_NULL:
1266 break;
1267 case SVt_IV:
1268 if (SvROK(sv)) {
1269 referent = SvRV(sv);
1270 old_type_details = &fake_rv;
1271 if (new_type == SVt_NV)
1272 new_type = SVt_PVNV;
1273 } else {
1274 if (new_type < SVt_PVIV) {
1275 new_type = (new_type == SVt_NV)
1276 ? SVt_PVNV : SVt_PVIV;
1277 }
1278 }
1279 break;
1280 case SVt_NV:
1281 if (new_type < SVt_PVNV) {
1282 new_type = SVt_PVNV;
1283 }
1284 break;
1285 case SVt_PV:
1286 assert(new_type > SVt_PV);
1287 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1288 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1289 break;
1290 case SVt_PVIV:
1291 break;
1292 case SVt_PVNV:
1293 break;
1294 case SVt_PVMG:
1295 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1296 there's no way that it can be safely upgraded, because perl.c
1297 expects to Safefree(SvANY(PL_mess_sv)) */
1298 assert(sv != PL_mess_sv);
1299 break;
1300 default:
1301 if (UNLIKELY(old_type_details->cant_upgrade))
1302 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1303 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1304 }
1305
1306 if (UNLIKELY(old_type > new_type))
1307 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1308 (int)old_type, (int)new_type);
1309
1310 new_type_details = bodies_by_type + new_type;
1311
1312 SvFLAGS(sv) &= ~SVTYPEMASK;
1313 SvFLAGS(sv) |= new_type;
1314
1315 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1316 the return statements above will have triggered. */
1317 assert (new_type != SVt_NULL);
1318 switch (new_type) {
1319 case SVt_IV:
1320 assert(old_type == SVt_NULL);
1321 SET_SVANY_FOR_BODYLESS_IV(sv);
1322 SvIV_set(sv, 0);
1323 return;
1324 case SVt_NV:
1325 assert(old_type == SVt_NULL);
1326 #if NVSIZE <= IVSIZE
1327 SET_SVANY_FOR_BODYLESS_NV(sv);
1328 #else
1329 SvANY(sv) = new_XNV();
1330 #endif
1331 SvNV_set(sv, 0);
1332 return;
1333 case SVt_PVHV:
1334 case SVt_PVAV:
1335 assert(new_type_details->body_size);
1336
1337 #ifndef PURIFY
1338 assert(new_type_details->arena);
1339 assert(new_type_details->arena_size);
1340 /* This points to the start of the allocated area. */
1341 new_body_inline(new_body, new_type);
1342 Zero(new_body, new_type_details->body_size, char);
1343 new_body = ((char *)new_body) - new_type_details->offset;
1344 #else
1345 /* We always allocated the full length item with PURIFY. To do this
1346 we fake things so that arena is false for all 16 types.. */
1347 new_body = new_NOARENAZ(new_type_details);
1348 #endif
1349 SvANY(sv) = new_body;
1350 if (new_type == SVt_PVAV) {
1351 AvMAX(sv) = -1;
1352 AvFILLp(sv) = -1;
1353 AvREAL_only(sv);
1354 if (old_type_details->body_size) {
1355 AvALLOC(sv) = 0;
1356 } else {
1357 /* It will have been zeroed when the new body was allocated.
1358 Lets not write to it, in case it confuses a write-back
1359 cache. */
1360 }
1361 } else {
1362 assert(!SvOK(sv));
1363 SvOK_off(sv);
1364 #ifndef NODEFAULT_SHAREKEYS
1365 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1366 #endif
1367 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1368 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1369 }
1370
1371 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1372 The target created by newSVrv also is, and it can have magic.
1373 However, it never has SvPVX set.
1374 */
1375 if (old_type == SVt_IV) {
1376 assert(!SvROK(sv));
1377 } else if (old_type >= SVt_PV) {
1378 assert(SvPVX_const(sv) == 0);
1379 }
1380
1381 if (old_type >= SVt_PVMG) {
1382 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1383 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1384 } else {
1385 sv->sv_u.svu_array = NULL; /* or svu_hash */
1386 }
1387 break;
1388
1389 case SVt_PVIV:
1390 /* XXX Is this still needed? Was it ever needed? Surely as there is
1391 no route from NV to PVIV, NOK can never be true */
1392 assert(!SvNOKp(sv));
1393 assert(!SvNOK(sv));
1394 /* FALLTHROUGH */
1395 case SVt_PVIO:
1396 case SVt_PVFM:
1397 case SVt_PVGV:
1398 case SVt_PVCV:
1399 case SVt_PVLV:
1400 case SVt_INVLIST:
1401 case SVt_REGEXP:
1402 case SVt_PVMG:
1403 case SVt_PVNV:
1404 case SVt_PV:
1405
1406 assert(new_type_details->body_size);
1407 /* We always allocated the full length item with PURIFY. To do this
1408 we fake things so that arena is false for all 16 types.. */
1409 if(new_type_details->arena) {
1410 /* This points to the start of the allocated area. */
1411 new_body_inline(new_body, new_type);
1412 Zero(new_body, new_type_details->body_size, char);
1413 new_body = ((char *)new_body) - new_type_details->offset;
1414 } else {
1415 new_body = new_NOARENAZ(new_type_details);
1416 }
1417 SvANY(sv) = new_body;
1418
1419 if (old_type_details->copy) {
1420 /* There is now the potential for an upgrade from something without
1421 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1422 int offset = old_type_details->offset;
1423 int length = old_type_details->copy;
1424
1425 if (new_type_details->offset > old_type_details->offset) {
1426 const int difference
1427 = new_type_details->offset - old_type_details->offset;
1428 offset += difference;
1429 length -= difference;
1430 }
1431 assert (length >= 0);
1432
1433 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1434 char);
1435 }
1436
1437 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1438 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1439 * correct 0.0 for us. Otherwise, if the old body didn't have an
1440 * NV slot, but the new one does, then we need to initialise the
1441 * freshly created NV slot with whatever the correct bit pattern is
1442 * for 0.0 */
1443 if (old_type_details->zero_nv && !new_type_details->zero_nv
1444 && !isGV_with_GP(sv))
1445 SvNV_set(sv, 0);
1446 #endif
1447
1448 if (UNLIKELY(new_type == SVt_PVIO)) {
1449 IO * const io = MUTABLE_IO(sv);
1450 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1451
1452 SvOBJECT_on(io);
1453 /* Clear the stashcache because a new IO could overrule a package
1454 name */
1455 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1456 hv_clear(PL_stashcache);
1457
1458 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1459 IoPAGE_LEN(sv) = 60;
1460 }
1461 if (old_type < SVt_PV) {
1462 /* referent will be NULL unless the old type was SVt_IV emulating
1463 SVt_RV */
1464 sv->sv_u.svu_rv = referent;
1465 }
1466 break;
1467 default:
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
1470 }
1471
1472 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1473 and sometimes SVt_NV */
1474 if (old_type_details->body_size) {
1475 #ifdef PURIFY
1476 safefree(old_body);
1477 #else
1478 /* Note that there is an assumption that all bodies of types that
1479 can be upgraded came from arenas. Only the more complex non-
1480 upgradable types are allowed to be directly malloc()ed. */
1481 assert(old_type_details->arena);
1482 del_body((void*)((char*)old_body + old_type_details->offset),
1483 &PL_body_roots[old_type]);
1484 #endif
1485 }
1486 }
1487
1488 /*
1489 =for apidoc sv_backoff
1490
1491 Remove any string offset. You should normally use the C<SvOOK_off> macro
1492 wrapper instead.
1493
1494 =cut
1495 */
1496
1497 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1498 prior to 5.23.4 this function always returned 0
1499 */
1500
1501 void
Perl_sv_backoff(SV * const sv)1502 Perl_sv_backoff(SV *const sv)
1503 {
1504 STRLEN delta;
1505 const char * const s = SvPVX_const(sv);
1506
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
1508
1509 assert(SvOOK(sv));
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
1512
1513 SvOOK_offset(sv, delta);
1514
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
1517 SvFLAGS(sv) &= ~SVf_OOK;
1518 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1519 return;
1520 }
1521
1522
1523 /* forward declaration */
1524 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1525
1526
1527 /*
1528 =for apidoc sv_grow
1529
1530 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1531 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1532 Use the C<SvGROW> wrapper instead.
1533
1534 =cut
1535 */
1536
1537
1538 char *
Perl_sv_grow(pTHX_ SV * const sv,STRLEN newlen)1539 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1540 {
1541 char *s;
1542
1543 PERL_ARGS_ASSERT_SV_GROW;
1544
1545 if (SvROK(sv))
1546 sv_unref(sv);
1547 if (SvTYPE(sv) < SVt_PV) {
1548 sv_upgrade(sv, SVt_PV);
1549 s = SvPVX_mutable(sv);
1550 }
1551 else if (SvOOK(sv)) { /* pv is offset? */
1552 sv_backoff(sv);
1553 s = SvPVX_mutable(sv);
1554 if (newlen > SvLEN(sv))
1555 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1556 }
1557 else
1558 {
1559 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1560 s = SvPVX_mutable(sv);
1561 }
1562
1563 #ifdef PERL_COPY_ON_WRITE
1564 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1565 * to store the COW count. So in general, allocate one more byte than
1566 * asked for, to make it likely this byte is always spare: and thus
1567 * make more strings COW-able.
1568 *
1569 * Only increment if the allocation isn't MEM_SIZE_MAX,
1570 * otherwise it will wrap to 0.
1571 */
1572 if ( newlen != MEM_SIZE_MAX )
1573 newlen++;
1574 #endif
1575
1576 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1577 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1578 #endif
1579
1580 if (newlen > SvLEN(sv)) { /* need more room? */
1581 STRLEN minlen = SvCUR(sv);
1582 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1583 if (newlen < minlen)
1584 newlen = minlen;
1585 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586
1587 /* Don't round up on the first allocation, as odds are pretty good that
1588 * the initial request is accurate as to what is really needed */
1589 if (SvLEN(sv)) {
1590 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1591 if (rounded > newlen)
1592 newlen = rounded;
1593 }
1594 #endif
1595 if (SvLEN(sv) && s) {
1596 s = (char*)saferealloc(s, newlen);
1597 }
1598 else {
1599 s = (char*)safemalloc(newlen);
1600 if (SvPVX_const(sv) && SvCUR(sv)) {
1601 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1602 }
1603 }
1604 SvPV_set(sv, s);
1605 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1606 /* Do this here, do it once, do it right, and then we will never get
1607 called back into sv_grow() unless there really is some growing
1608 needed. */
1609 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1610 #else
1611 SvLEN_set(sv, newlen);
1612 #endif
1613 }
1614 return s;
1615 }
1616
1617 /*
1618 =for apidoc sv_setiv
1619
1620 Copies an integer into the given SV, upgrading first if necessary.
1621 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1622
1623 =cut
1624 */
1625
1626 void
Perl_sv_setiv(pTHX_ SV * const sv,const IV i)1627 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1628 {
1629 PERL_ARGS_ASSERT_SV_SETIV;
1630
1631 SV_CHECK_THINKFIRST_COW_DROP(sv);
1632 switch (SvTYPE(sv)) {
1633 case SVt_NULL:
1634 case SVt_NV:
1635 sv_upgrade(sv, SVt_IV);
1636 break;
1637 case SVt_PV:
1638 sv_upgrade(sv, SVt_PVIV);
1639 break;
1640
1641 case SVt_PVGV:
1642 if (!isGV_with_GP(sv))
1643 break;
1644 /* FALLTHROUGH */
1645 case SVt_PVAV:
1646 case SVt_PVHV:
1647 case SVt_PVCV:
1648 case SVt_PVFM:
1649 case SVt_PVIO:
1650 /* diag_listed_as: Can't coerce %s to %s in %s */
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1652 OP_DESC(PL_op));
1653 NOT_REACHED; /* NOTREACHED */
1654 break;
1655 default: NOOP;
1656 }
1657 (void)SvIOK_only(sv); /* validate number */
1658 SvIV_set(sv, i);
1659 SvTAINT(sv);
1660 }
1661
1662 /*
1663 =for apidoc sv_setiv_mg
1664
1665 Like C<sv_setiv>, but also handles 'set' magic.
1666
1667 =cut
1668 */
1669
1670 void
Perl_sv_setiv_mg(pTHX_ SV * const sv,const IV i)1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1672 {
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1674
1675 sv_setiv(sv,i);
1676 SvSETMAGIC(sv);
1677 }
1678
1679 /*
1680 =for apidoc sv_setuv
1681
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1684
1685 =cut
1686 */
1687
1688 void
Perl_sv_setuv(pTHX_ SV * const sv,const UV u)1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1690 {
1691 PERL_ARGS_ASSERT_SV_SETUV;
1692
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1694 possible:
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1696
1697 without
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1699
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1703 */
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1706 return;
1707 }
1708 sv_setiv(sv, 0);
1709 SvIsUV_on(sv);
1710 SvUV_set(sv, u);
1711 }
1712
1713 /*
1714 =for apidoc sv_setuv_mg
1715
1716 Like C<sv_setuv>, but also handles 'set' magic.
1717
1718 =cut
1719 */
1720
1721 void
Perl_sv_setuv_mg(pTHX_ SV * const sv,const UV u)1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1723 {
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1725
1726 sv_setuv(sv,u);
1727 SvSETMAGIC(sv);
1728 }
1729
1730 /*
1731 =for apidoc sv_setnv
1732
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1735
1736 =cut
1737 */
1738
1739 void
Perl_sv_setnv(pTHX_ SV * const sv,const NV num)1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1741 {
1742 PERL_ARGS_ASSERT_SV_SETNV;
1743
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1746 case SVt_NULL:
1747 case SVt_IV:
1748 sv_upgrade(sv, SVt_NV);
1749 break;
1750 case SVt_PV:
1751 case SVt_PVIV:
1752 sv_upgrade(sv, SVt_PVNV);
1753 break;
1754
1755 case SVt_PVGV:
1756 if (!isGV_with_GP(sv))
1757 break;
1758 /* FALLTHROUGH */
1759 case SVt_PVAV:
1760 case SVt_PVHV:
1761 case SVt_PVCV:
1762 case SVt_PVFM:
1763 case SVt_PVIO:
1764 /* diag_listed_as: Can't coerce %s to %s in %s */
1765 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1766 OP_DESC(PL_op));
1767 NOT_REACHED; /* NOTREACHED */
1768 break;
1769 default: NOOP;
1770 }
1771 SvNV_set(sv, num);
1772 (void)SvNOK_only(sv); /* validate number */
1773 SvTAINT(sv);
1774 }
1775
1776 /*
1777 =for apidoc sv_setnv_mg
1778
1779 Like C<sv_setnv>, but also handles 'set' magic.
1780
1781 =cut
1782 */
1783
1784 void
Perl_sv_setnv_mg(pTHX_ SV * const sv,const NV num)1785 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1786 {
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1788
1789 sv_setnv(sv,num);
1790 SvSETMAGIC(sv);
1791 }
1792
1793 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1794 * not incrementable warning display.
1795 * Originally part of S_not_a_number().
1796 * The return value may be != tmpbuf.
1797 */
1798
1799 STATIC const char *
S_sv_display(pTHX_ SV * const sv,char * tmpbuf,STRLEN tmpbuf_size)1800 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1801 const char *pv;
1802
1803 PERL_ARGS_ASSERT_SV_DISPLAY;
1804
1805 if (DO_UTF8(sv)) {
1806 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1807 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1808 } else {
1809 char *d = tmpbuf;
1810 const char * const limit = tmpbuf + tmpbuf_size - 8;
1811 /* each *s can expand to 4 chars + "...\0",
1812 i.e. need room for 8 chars */
1813
1814 const char *s = SvPVX_const(sv);
1815 const char * const end = s + SvCUR(sv);
1816 for ( ; s < end && d < limit; s++ ) {
1817 int ch = *s & 0xFF;
1818 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1819 *d++ = 'M';
1820 *d++ = '-';
1821
1822 /* Map to ASCII "equivalent" of Latin1 */
1823 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1824 }
1825 if (ch == '\n') {
1826 *d++ = '\\';
1827 *d++ = 'n';
1828 }
1829 else if (ch == '\r') {
1830 *d++ = '\\';
1831 *d++ = 'r';
1832 }
1833 else if (ch == '\f') {
1834 *d++ = '\\';
1835 *d++ = 'f';
1836 }
1837 else if (ch == '\\') {
1838 *d++ = '\\';
1839 *d++ = '\\';
1840 }
1841 else if (ch == '\0') {
1842 *d++ = '\\';
1843 *d++ = '0';
1844 }
1845 else if (isPRINT_LC(ch))
1846 *d++ = ch;
1847 else {
1848 *d++ = '^';
1849 *d++ = toCTRL(ch);
1850 }
1851 }
1852 if (s < end) {
1853 *d++ = '.';
1854 *d++ = '.';
1855 *d++ = '.';
1856 }
1857 *d = '\0';
1858 pv = tmpbuf;
1859 }
1860
1861 return pv;
1862 }
1863
1864 /* Print an "isn't numeric" warning, using a cleaned-up,
1865 * printable version of the offending string
1866 */
1867
1868 STATIC void
S_not_a_number(pTHX_ SV * const sv)1869 S_not_a_number(pTHX_ SV *const sv)
1870 {
1871 char tmpbuf[64];
1872 const char *pv;
1873
1874 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1875
1876 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1877
1878 if (PL_op)
1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1880 /* diag_listed_as: Argument "%s" isn't numeric%s */
1881 "Argument \"%s\" isn't numeric in %s", pv,
1882 OP_DESC(PL_op));
1883 else
1884 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1885 /* diag_listed_as: Argument "%s" isn't numeric%s */
1886 "Argument \"%s\" isn't numeric", pv);
1887 }
1888
1889 STATIC void
S_not_incrementable(pTHX_ SV * const sv)1890 S_not_incrementable(pTHX_ SV *const sv) {
1891 char tmpbuf[64];
1892 const char *pv;
1893
1894 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1895
1896 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1897
1898 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1899 "Argument \"%s\" treated as 0 in increment (++)", pv);
1900 }
1901
1902 /*
1903 =for apidoc looks_like_number
1904
1905 Test if the content of an SV looks like a number (or is a number).
1906 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1907 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1908 ignored.
1909
1910 =cut
1911 */
1912
1913 I32
Perl_looks_like_number(pTHX_ SV * const sv)1914 Perl_looks_like_number(pTHX_ SV *const sv)
1915 {
1916 const char *sbegin;
1917 STRLEN len;
1918 int numtype;
1919
1920 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1921
1922 if (SvPOK(sv) || SvPOKp(sv)) {
1923 sbegin = SvPV_nomg_const(sv, len);
1924 }
1925 else
1926 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1927 numtype = grok_number(sbegin, len, NULL);
1928 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1929 }
1930
1931 STATIC bool
S_glob_2number(pTHX_ GV * const gv)1932 S_glob_2number(pTHX_ GV * const gv)
1933 {
1934 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1935
1936 /* We know that all GVs stringify to something that is not-a-number,
1937 so no need to test that. */
1938 if (ckWARN(WARN_NUMERIC))
1939 {
1940 SV *const buffer = sv_newmortal();
1941 gv_efullname3(buffer, gv, "*");
1942 not_a_number(buffer);
1943 }
1944 /* We just want something true to return, so that S_sv_2iuv_common
1945 can tail call us and return true. */
1946 return TRUE;
1947 }
1948
1949 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1950 until proven guilty, assume that things are not that bad... */
1951
1952 /*
1953 NV_PRESERVES_UV:
1954
1955 As 64 bit platforms often have an NV that doesn't preserve all bits of
1956 an IV (an assumption perl has been based on to date) it becomes necessary
1957 to remove the assumption that the NV always carries enough precision to
1958 recreate the IV whenever needed, and that the NV is the canonical form.
1959 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1960 precision as a side effect of conversion (which would lead to insanity
1961 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1962 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1963 where precision was lost, and IV/UV/NV slots that have a valid conversion
1964 which has lost no precision
1965 2) to ensure that if a numeric conversion to one form is requested that
1966 would lose precision, the precise conversion (or differently
1967 imprecise conversion) is also performed and cached, to prevent
1968 requests for different numeric formats on the same SV causing
1969 lossy conversion chains. (lossless conversion chains are perfectly
1970 acceptable (still))
1971
1972
1973 flags are used:
1974 SvIOKp is true if the IV slot contains a valid value
1975 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1976 SvNOKp is true if the NV slot contains a valid value
1977 SvNOK is true only if the NV value is accurate
1978
1979 so
1980 while converting from PV to NV, check to see if converting that NV to an
1981 IV(or UV) would lose accuracy over a direct conversion from PV to
1982 IV(or UV). If it would, cache both conversions, return NV, but mark
1983 SV as IOK NOKp (ie not NOK).
1984
1985 While converting from PV to IV, check to see if converting that IV to an
1986 NV would lose accuracy over a direct conversion from PV to NV. If it
1987 would, cache both conversions, flag similarly.
1988
1989 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1990 correctly because if IV & NV were set NV *always* overruled.
1991 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1992 changes - now IV and NV together means that the two are interchangeable:
1993 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1994
1995 The benefit of this is that operations such as pp_add know that if
1996 SvIOK is true for both left and right operands, then integer addition
1997 can be used instead of floating point (for cases where the result won't
1998 overflow). Before, floating point was always used, which could lead to
1999 loss of precision compared with integer addition.
2000
2001 * making IV and NV equal status should make maths accurate on 64 bit
2002 platforms
2003 * may speed up maths somewhat if pp_add and friends start to use
2004 integers when possible instead of fp. (Hopefully the overhead in
2005 looking for SvIOK and checking for overflow will not outweigh the
2006 fp to integer speedup)
2007 * will slow down integer operations (callers of SvIV) on "inaccurate"
2008 values, as the change from SvIOK to SvIOKp will cause a call into
2009 sv_2iv each time rather than a macro access direct to the IV slot
2010 * should speed up number->string conversion on integers as IV is
2011 favoured when IV and NV are equally accurate
2012
2013 ####################################################################
2014 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2015 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2016 On the other hand, SvUOK is true iff UV.
2017 ####################################################################
2018
2019 Your mileage will vary depending your CPU's relative fp to integer
2020 performance ratio.
2021 */
2022
2023 #ifndef NV_PRESERVES_UV
2024 # define IS_NUMBER_UNDERFLOW_IV 1
2025 # define IS_NUMBER_UNDERFLOW_UV 2
2026 # define IS_NUMBER_IV_AND_UV 2
2027 # define IS_NUMBER_OVERFLOW_IV 4
2028 # define IS_NUMBER_OVERFLOW_UV 5
2029
2030 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2031
2032 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2033 STATIC int
S_sv_2iuv_non_preserve(pTHX_ SV * const sv,I32 numtype)2034 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2035 # ifdef DEBUGGING
2036 , I32 numtype
2037 # endif
2038 )
2039 {
2040 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2041 PERL_UNUSED_CONTEXT;
2042
2043 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%" UVxf " NV=%" NVgf " inttype=%" UVXf "\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2044 if (SvNVX(sv) < (NV)IV_MIN) {
2045 (void)SvIOKp_on(sv);
2046 (void)SvNOK_on(sv);
2047 SvIV_set(sv, IV_MIN);
2048 return IS_NUMBER_UNDERFLOW_IV;
2049 }
2050 if (SvNVX(sv) > (NV)UV_MAX) {
2051 (void)SvIOKp_on(sv);
2052 (void)SvNOK_on(sv);
2053 SvIsUV_on(sv);
2054 SvUV_set(sv, UV_MAX);
2055 return IS_NUMBER_OVERFLOW_UV;
2056 }
2057 (void)SvIOKp_on(sv);
2058 (void)SvNOK_on(sv);
2059 /* Can't use strtol etc to convert this string. (See truth table in
2060 sv_2iv */
2061 if (SvNVX(sv) <= (UV)IV_MAX) {
2062 SvIV_set(sv, I_V(SvNVX(sv)));
2063 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2064 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2065 } else {
2066 /* Integer is imprecise. NOK, IOKp */
2067 }
2068 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2069 }
2070 SvIsUV_on(sv);
2071 SvUV_set(sv, U_V(SvNVX(sv)));
2072 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2073 if (SvUVX(sv) == UV_MAX) {
2074 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2075 possibly be preserved by NV. Hence, it must be overflow.
2076 NOK, IOKp */
2077 return IS_NUMBER_OVERFLOW_UV;
2078 }
2079 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2080 } else {
2081 /* Integer is imprecise. NOK, IOKp */
2082 }
2083 return IS_NUMBER_OVERFLOW_IV;
2084 }
2085 #endif /* !NV_PRESERVES_UV*/
2086
2087 /* If numtype is infnan, set the NV of the sv accordingly.
2088 * If numtype is anything else, try setting the NV using Atof(PV). */
2089 #ifdef USING_MSVC6
2090 # pragma warning(push)
2091 # pragma warning(disable:4756;disable:4056)
2092 #endif
2093 static void
S_sv_setnv(pTHX_ SV * sv,int numtype)2094 S_sv_setnv(pTHX_ SV* sv, int numtype)
2095 {
2096 bool pok = cBOOL(SvPOK(sv));
2097 bool nok = FALSE;
2098 #ifdef NV_INF
2099 if ((numtype & IS_NUMBER_INFINITY)) {
2100 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2101 nok = TRUE;
2102 } else
2103 #endif
2104 #ifdef NV_NAN
2105 if ((numtype & IS_NUMBER_NAN)) {
2106 SvNV_set(sv, NV_NAN);
2107 nok = TRUE;
2108 } else
2109 #endif
2110 if (pok) {
2111 SvNV_set(sv, Atof(SvPVX_const(sv)));
2112 /* Purposefully no true nok here, since we don't want to blow
2113 * away the possible IOK/UV of an existing sv. */
2114 }
2115 if (nok) {
2116 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2117 if (pok)
2118 SvPOK_on(sv); /* PV is okay, though. */
2119 }
2120 }
2121 #ifdef USING_MSVC6
2122 # pragma warning(pop)
2123 #endif
2124
2125 STATIC bool
S_sv_2iuv_common(pTHX_ SV * const sv)2126 S_sv_2iuv_common(pTHX_ SV *const sv)
2127 {
2128 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2129
2130 if (SvNOKp(sv)) {
2131 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2132 * without also getting a cached IV/UV from it at the same time
2133 * (ie PV->NV conversion should detect loss of accuracy and cache
2134 * IV or UV at same time to avoid this. */
2135 /* IV-over-UV optimisation - choose to cache IV if possible */
2136
2137 if (SvTYPE(sv) == SVt_NV)
2138 sv_upgrade(sv, SVt_PVNV);
2139
2140 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2141 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2142 certainly cast into the IV range at IV_MAX, whereas the correct
2143 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2144 cases go to UV */
2145 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2146 if (Perl_isnan(SvNVX(sv))) {
2147 SvUV_set(sv, 0);
2148 SvIsUV_on(sv);
2149 return FALSE;
2150 }
2151 #endif
2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2153 SvIV_set(sv, I_V(SvNVX(sv)));
2154 if (SvNVX(sv) == (NV) SvIVX(sv)
2155 #ifndef NV_PRESERVES_UV
2156 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2157 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2158 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2159 /* Don't flag it as "accurately an integer" if the number
2160 came from a (by definition imprecise) NV operation, and
2161 we're outside the range of NV integer precision */
2162 #endif
2163 ) {
2164 if (SvNOK(sv))
2165 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2166 else {
2167 /* scalar has trailing garbage, eg "42a" */
2168 }
2169 DEBUG_c(PerlIO_printf(Perl_debug_log,
2170 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2171 PTR2UV(sv),
2172 SvNVX(sv),
2173 SvIVX(sv)));
2174
2175 } else {
2176 /* IV not precise. No need to convert from PV, as NV
2177 conversion would already have cached IV if it detected
2178 that PV->IV would be better than PV->NV->IV
2179 flags already correct - don't set public IOK. */
2180 DEBUG_c(PerlIO_printf(Perl_debug_log,
2181 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2182 PTR2UV(sv),
2183 SvNVX(sv),
2184 SvIVX(sv)));
2185 }
2186 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2187 but the cast (NV)IV_MIN rounds to a the value less (more
2188 negative) than IV_MIN which happens to be equal to SvNVX ??
2189 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2190 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2191 (NV)UVX == NVX are both true, but the values differ. :-(
2192 Hopefully for 2s complement IV_MIN is something like
2193 0x8000000000000000 which will be exact. NWC */
2194 }
2195 else {
2196 SvUV_set(sv, U_V(SvNVX(sv)));
2197 if (
2198 (SvNVX(sv) == (NV) SvUVX(sv))
2199 #ifndef NV_PRESERVES_UV
2200 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2201 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2202 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2203 /* Don't flag it as "accurately an integer" if the number
2204 came from a (by definition imprecise) NV operation, and
2205 we're outside the range of NV integer precision */
2206 #endif
2207 && SvNOK(sv)
2208 )
2209 SvIOK_on(sv);
2210 SvIsUV_on(sv);
2211 DEBUG_c(PerlIO_printf(Perl_debug_log,
2212 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2213 PTR2UV(sv),
2214 SvUVX(sv),
2215 SvUVX(sv)));
2216 }
2217 }
2218 else if (SvPOKp(sv)) {
2219 UV value;
2220 int numtype;
2221 const char *s = SvPVX_const(sv);
2222 const STRLEN cur = SvCUR(sv);
2223
2224 /* short-cut for a single digit string like "1" */
2225
2226 if (cur == 1) {
2227 char c = *s;
2228 if (isDIGIT(c)) {
2229 if (SvTYPE(sv) < SVt_PVIV)
2230 sv_upgrade(sv, SVt_PVIV);
2231 (void)SvIOK_on(sv);
2232 SvIV_set(sv, (IV)(c - '0'));
2233 return FALSE;
2234 }
2235 }
2236
2237 numtype = grok_number(s, cur, &value);
2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
2239 may be later translated to an NV, and the resulting NV is not
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
2244
2245 This means that if we cache such an IV/a UV, we need to cache the
2246 NV as well. Moreover, we trade speed for space, and do not
2247 cache the NV if we are sure it's not needed.
2248 */
2249
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
2253 /* It's definitely an integer, only upgrade to PVIV */
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
2256 (void)SvIOK_on(sv);
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
2259
2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2262 not_a_number(sv);
2263 S_sv_setnv(aTHX_ sv, numtype);
2264 return FALSE;
2265 }
2266
2267 /* If NVs preserve UVs then we only use the UV value if we know that
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272 #ifdef NV_PRESERVES_UV
2273 | IS_NUMBER_NOT_INT
2274 #endif
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2278
2279 if (!(numtype & IS_NUMBER_NEG)) {
2280 /* positive */;
2281 if (value <= (UV)IV_MAX) {
2282 SvIV_set(sv, (IV)value);
2283 } else {
2284 /* it didn't overflow, and it was positive. */
2285 SvUV_set(sv, value);
2286 SvIsUV_on(sv);
2287 }
2288 } else {
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
2293 } else {
2294 /* Too negative for an IV. This is a double upgrade, but
2295 I'm assuming it will be rare. */
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2298 SvNOK_on(sv);
2299 SvIOK_off(sv);
2300 SvIOKp_on(sv);
2301 SvNV_set(sv, -(NV)value);
2302 SvIV_set(sv, IV_MIN);
2303 }
2304 }
2305 }
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2309
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
2313 S_sv_setnv(aTHX_ sv, numtype);
2314
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2316 not_a_number(sv);
2317
2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2319 PTR2UV(sv), SvNVX(sv)));
2320
2321 #ifdef NV_PRESERVES_UV
2322 (void)SvIOKp_on(sv);
2323 (void)SvNOK_on(sv);
2324 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2326 SvUV_set(sv, 0);
2327 SvIsUV_on(sv);
2328 return FALSE;
2329 }
2330 #endif
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2334 SvIOK_on(sv);
2335 } else {
2336 NOOP; /* Integer is imprecise. NOK, IOKp */
2337 }
2338 /* UV will not work better than IV */
2339 } else {
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2341 SvIsUV_on(sv);
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
2344 } else {
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2349 SvIOK_on(sv);
2350 } else {
2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2352 }
2353 }
2354 SvIsUV_on(sv);
2355 }
2356 #else /* NV_PRESERVES_UV */
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2359 /* The IV/UV slot will have been set from value returned by
2360 grok_number above. The NV slot has just been set using
2361 Atof. */
2362 SvNOK_on(sv);
2363 assert (SvIOKp(sv));
2364 } else {
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2369 SvNOK_on(sv);
2370 SvIV_set(sv, I_V(SvNVX(sv)));
2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2372 SvIOK_on(sv);
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2376 < (UV)IV_MAX)) {
2377 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%" NVgf " U_V is 0x%" UVxf ", IV_MAX is 0x%" UVxf "\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2378 }
2379 } else {
2380 /* IN_UV NOT_INT
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
2388 # ifdef DEBUGGING
2389 sv_2iuv_non_preserve (sv, numtype);
2390 # else
2391 sv_2iuv_non_preserve (sv);
2392 # endif
2393 }
2394 }
2395 #endif /* NV_PRESERVES_UV */
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2400 if (!numtype)
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2402 }
2403 }
2404 else {
2405 if (isGV_with_GP(sv))
2406 return glob_2number(MUTABLE_GV(sv));
2407
2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2409 report_uninit(sv);
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
2413 /* Return 0 from the caller. */
2414 return TRUE;
2415 }
2416 return FALSE;
2417 }
2418
2419 /*
2420 =for apidoc sv_2iv_flags
2421
2422 Return the integer value of an SV, doing any necessary string
2423 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2424 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2425
2426 =cut
2427 */
2428
2429 IV
Perl_sv_2iv_flags(pTHX_ SV * const sv,const I32 flags)2430 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2431 {
2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2433
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2436
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2438 mg_get(sv);
2439
2440 if (SvROK(sv)) {
2441 if (SvAMAGIC(sv)) {
2442 SV * tmpstr;
2443 if (flags & SV_SKIP_OVERLOAD)
2444 return 0;
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2448 }
2449 }
2450 return PTR2IV(SvRV(sv));
2451 }
2452
2453 if (SvVALID(sv) || isREGEXP(sv)) {
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2455 must not let them cache IVs.
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
2459
2460 Regexps have no SvIVX and SvNVX fields.
2461 */
2462 assert(SvPOKp(sv));
2463 {
2464 UV value;
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2467 const int numtype
2468 = grok_number(ptr, SvCUR(sv), &value);
2469
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2475 return -(IV)value;
2476 } else {
2477 if (value < (UV)IV_MAX)
2478 return (IV)value;
2479 }
2480 }
2481
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2487 }
2488
2489 if (!numtype) {
2490 if (ckWARN(WARN_NUMERIC))
2491 not_a_number(sv);
2492 }
2493 return I_V(Atof(ptr));
2494 }
2495 }
2496
2497 if (SvTHINKFIRST(sv)) {
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2500 report_uninit(sv);
2501 return 0;
2502 }
2503 }
2504
2505 if (!SvIOKp(sv)) {
2506 if (S_sv_2iuv_common(aTHX_ sv))
2507 return 0;
2508 }
2509
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2511 PTR2UV(sv),SvIVX(sv)));
2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2513 }
2514
2515 /*
2516 =for apidoc sv_2uv_flags
2517
2518 Return the unsigned integer value of an SV, doing any necessary string
2519 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2520 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2521
2522 =cut
2523 */
2524
2525 UV
Perl_sv_2uv_flags(pTHX_ SV * const sv,const I32 flags)2526 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2527 {
2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2529
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2531 mg_get(sv);
2532
2533 if (SvROK(sv)) {
2534 if (SvAMAGIC(sv)) {
2535 SV *tmpstr;
2536 if (flags & SV_SKIP_OVERLOAD)
2537 return 0;
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2541 }
2542 }
2543 return PTR2UV(SvRV(sv));
2544 }
2545
2546 if (SvVALID(sv) || isREGEXP(sv)) {
2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
2550 assert(SvPOKp(sv));
2551 {
2552 UV value;
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2555 const int numtype
2556 = grok_number(ptr, SvCUR(sv), &value);
2557
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2562 return value;
2563 }
2564
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2570 }
2571
2572 if (!numtype) {
2573 if (ckWARN(WARN_NUMERIC))
2574 not_a_number(sv);
2575 }
2576 return U_V(Atof(ptr));
2577 }
2578 }
2579
2580 if (SvTHINKFIRST(sv)) {
2581 if (SvREADONLY(sv) && !SvOK(sv)) {
2582 if (ckWARN(WARN_UNINITIALIZED))
2583 report_uninit(sv);
2584 return 0;
2585 }
2586 }
2587
2588 if (!SvIOKp(sv)) {
2589 if (S_sv_2iuv_common(aTHX_ sv))
2590 return 0;
2591 }
2592
2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2594 PTR2UV(sv),SvUVX(sv)));
2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2596 }
2597
2598 /*
2599 =for apidoc sv_2nv_flags
2600
2601 Return the num value of an SV, doing any necessary string or integer
2602 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2603 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2604
2605 =cut
2606 */
2607
2608 NV
Perl_sv_2nv_flags(pTHX_ SV * const sv,const I32 flags)2609 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2610 {
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2612
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
2619 const char *ptr;
2620 if (flags & SV_GMAGIC)
2621 mg_get(sv);
2622 if (SvNOKp(sv))
2623 return SvNVX(sv);
2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
2625 ptr = SvPVX_const(sv);
2626 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2627 !grok_number(ptr, SvCUR(sv), NULL))
2628 not_a_number(sv);
2629 return Atof(ptr);
2630 }
2631 if (SvIOKp(sv)) {
2632 if (SvIsUV(sv))
2633 return (NV)SvUVX(sv);
2634 else
2635 return (NV)SvIVX(sv);
2636 }
2637 if (SvROK(sv)) {
2638 goto return_rok;
2639 }
2640 assert(SvTYPE(sv) >= SVt_PVMG);
2641 /* This falls through to the report_uninit near the end of the
2642 function. */
2643 } else if (SvTHINKFIRST(sv)) {
2644 if (SvROK(sv)) {
2645 return_rok:
2646 if (SvAMAGIC(sv)) {
2647 SV *tmpstr;
2648 if (flags & SV_SKIP_OVERLOAD)
2649 return 0;
2650 tmpstr = AMG_CALLunary(sv, numer_amg);
2651 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2652 return SvNV(tmpstr);
2653 }
2654 }
2655 return PTR2NV(SvRV(sv));
2656 }
2657 if (SvREADONLY(sv) && !SvOK(sv)) {
2658 if (ckWARN(WARN_UNINITIALIZED))
2659 report_uninit(sv);
2660 return 0.0;
2661 }
2662 }
2663 if (SvTYPE(sv) < SVt_NV) {
2664 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2665 sv_upgrade(sv, SVt_NV);
2666 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2667 DEBUG_c({
2668 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2669 STORE_LC_NUMERIC_SET_STANDARD();
2670 PerlIO_printf(Perl_debug_log,
2671 "0x%" UVxf " num(%" NVgf ")\n",
2672 PTR2UV(sv), SvNVX(sv));
2673 RESTORE_LC_NUMERIC();
2674 });
2675 CLANG_DIAG_RESTORE_STMT;
2676
2677 }
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
2680 if (SvNOKp(sv)) {
2681 return SvNVX(sv);
2682 }
2683 if (SvIOKp(sv)) {
2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2685 #ifdef NV_PRESERVES_UV
2686 if (SvIOK(sv))
2687 SvNOK_on(sv);
2688 else
2689 SvNOKp_on(sv);
2690 #else
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2693 if (SvIOK(sv) &&
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2696 SvNOK_on(sv);
2697 else
2698 SvNOKp_on(sv);
2699 #endif
2700 }
2701 else if (SvPOKp(sv)) {
2702 UV value;
2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2705 not_a_number(sv);
2706 #ifdef NV_PRESERVES_UV
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
2709 /* It's definitely an integer */
2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2711 } else {
2712 S_sv_setnv(aTHX_ sv, numtype);
2713 }
2714 if (numtype)
2715 SvNOK_on(sv);
2716 else
2717 SvNOKp_on(sv);
2718 #else
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2725 UV_BITS */
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2732 SvNOK_on(sv);
2733 } else {
2734 /* value has been set. It may not be precise. */
2735 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
2738 } else {
2739 SvNOKp_on(sv);
2740 SvIOKp_on(sv);
2741
2742 if (numtype & IS_NUMBER_NEG) {
2743 /* -IV_MIN is undefined, but we should never reach
2744 * this point with both IS_NUMBER_NEG and value ==
2745 * (UV)IV_MIN */
2746 assert(value != (UV)IV_MIN);
2747 SvIV_set(sv, -(IV)value);
2748 } else if (value <= (UV)IV_MAX) {
2749 SvIV_set(sv, (IV)value);
2750 } else {
2751 SvUV_set(sv, value);
2752 SvIsUV_on(sv);
2753 }
2754
2755 if (numtype & IS_NUMBER_NOT_INT) {
2756 /* I believe that even if the original PV had decimals,
2757 they are lost beyond the limit of the FP precision.
2758 However, neither is canonical, so both only get p
2759 flags. NWC, 2000/11/25 */
2760 /* Both already have p flags, so do nothing */
2761 } else {
2762 const NV nv = SvNVX(sv);
2763 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2764 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2765 if (SvIVX(sv) == I_V(nv)) {
2766 SvNOK_on(sv);
2767 } else {
2768 /* It had no "." so it must be integer. */
2769 }
2770 SvIOK_on(sv);
2771 } else {
2772 /* between IV_MAX and NV(UV_MAX).
2773 Could be slightly > UV_MAX */
2774
2775 if (numtype & IS_NUMBER_NOT_INT) {
2776 /* UV and NV both imprecise. */
2777 } else {
2778 const UV nv_as_uv = U_V(nv);
2779
2780 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2781 SvNOK_on(sv);
2782 }
2783 SvIOK_on(sv);
2784 }
2785 }
2786 }
2787 }
2788 }
2789 /* It might be more code efficient to go through the entire logic above
2790 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2791 gets complex and potentially buggy, so more programmer efficient
2792 to do it this way, by turning off the public flags: */
2793 if (!numtype)
2794 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2795 #endif /* NV_PRESERVES_UV */
2796 }
2797 else {
2798 if (isGV_with_GP(sv)) {
2799 glob_2number(MUTABLE_GV(sv));
2800 return 0.0;
2801 }
2802
2803 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2804 report_uninit(sv);
2805 assert (SvTYPE(sv) >= SVt_NV);
2806 /* Typically the caller expects that sv_any is not NULL now. */
2807 /* XXX Ilya implies that this is a bug in callers that assume this
2808 and ideally should be fixed. */
2809 return 0.0;
2810 }
2811 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2812 DEBUG_c({
2813 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2814 STORE_LC_NUMERIC_SET_STANDARD();
2815 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2816 PTR2UV(sv), SvNVX(sv));
2817 RESTORE_LC_NUMERIC();
2818 });
2819 CLANG_DIAG_RESTORE_STMT;
2820 return SvNVX(sv);
2821 }
2822
2823 /*
2824 =for apidoc sv_2num
2825
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2828 get-magic already.
2829
2830 =cut
2831 */
2832
2833 SV *
Perl_sv_2num(pTHX_ SV * const sv)2834 Perl_sv_2num(pTHX_ SV *const sv)
2835 {
2836 PERL_ARGS_ASSERT_SV_2NUM;
2837
2838 if (!SvROK(sv))
2839 return sv;
2840 if (SvAMAGIC(sv)) {
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2845 }
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2847 }
2848
2849 /* int2str_table: lookup table containing string representations of all
2850 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2851 * int2str_table.arr[12*2] is "12".
2852 *
2853 * We are going to read two bytes at a time, so we have to ensure that
2854 * the array is aligned to a 2 byte boundary. That's why it was made a
2855 * union with a dummy U16 member. */
2856 static const union {
2857 char arr[200];
2858 U16 dummy;
2859 } int2str_table = {{
2860 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2861 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2862 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2863 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2864 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2865 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2866 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2867 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2868 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2869 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2870 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2871 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2872 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2873 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2874 '9', '8', '9', '9'
2875 }};
2876
2877 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2878 * UV as a string towards the end of buf, and return pointers to start and
2879 * end of it.
2880 *
2881 * We assume that buf is at least TYPE_CHARS(UV) long.
2882 */
2883
2884 PERL_STATIC_INLINE char *
S_uiv_2buf(char * const buf,const IV iv,UV uv,const int is_uv,char ** const peob)2885 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2886 {
2887 char *ptr = buf + TYPE_CHARS(UV);
2888 char * const ebuf = ptr;
2889 int sign;
2890 U16 *word_ptr, *word_table;
2891
2892 PERL_ARGS_ASSERT_UIV_2BUF;
2893
2894 /* ptr has to be properly aligned, because we will cast it to U16* */
2895 assert(PTR2nat(ptr) % 2 == 0);
2896 /* we are going to read/write two bytes at a time */
2897 word_ptr = (U16*)ptr;
2898 word_table = (U16*)int2str_table.arr;
2899
2900 if (UNLIKELY(is_uv))
2901 sign = 0;
2902 else if (iv >= 0) {
2903 uv = iv;
2904 sign = 0;
2905 } else {
2906 /* Using 0- here to silence bogus warning from MS VC */
2907 uv = (UV) (0 - (UV) iv);
2908 sign = 1;
2909 }
2910
2911 while (uv > 99) {
2912 *--word_ptr = word_table[uv % 100];
2913 uv /= 100;
2914 }
2915 ptr = (char*)word_ptr;
2916
2917 if (uv < 10)
2918 *--ptr = (char)uv + '0';
2919 else {
2920 *--word_ptr = word_table[uv];
2921 ptr = (char*)word_ptr;
2922 }
2923
2924 if (sign)
2925 *--ptr = '-';
2926
2927 *peob = ebuf;
2928 return ptr;
2929 }
2930
2931 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2932 * infinity or a not-a-number, writes the appropriate strings to the
2933 * buffer, including a zero byte. On success returns the written length,
2934 * excluding the zero byte, on failure (not an infinity, not a nan)
2935 * returns zero, assert-fails on maxlen being too short.
2936 *
2937 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2938 * shared string constants we point to, instead of generating a new
2939 * string for each instance. */
2940 STATIC size_t
S_infnan_2pv(NV nv,char * buffer,size_t maxlen,char plus)2941 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2942 char* s = buffer;
2943 assert(maxlen >= 4);
2944 if (Perl_isinf(nv)) {
2945 if (nv < 0) {
2946 if (maxlen < 5) /* "-Inf\0" */
2947 return 0;
2948 *s++ = '-';
2949 } else if (plus) {
2950 *s++ = '+';
2951 }
2952 *s++ = 'I';
2953 *s++ = 'n';
2954 *s++ = 'f';
2955 }
2956 else if (Perl_isnan(nv)) {
2957 *s++ = 'N';
2958 *s++ = 'a';
2959 *s++ = 'N';
2960 /* XXX optionally output the payload mantissa bits as
2961 * "(unsigned)" (to match the nan("...") C99 function,
2962 * or maybe as "(0xhhh...)" would make more sense...
2963 * provide a format string so that the user can decide?
2964 * NOTE: would affect the maxlen and assert() logic.*/
2965 }
2966 else {
2967 return 0;
2968 }
2969 assert((s == buffer + 3) || (s == buffer + 4));
2970 *s = 0;
2971 return s - buffer;
2972 }
2973
2974 /*
2975 =for apidoc sv_2pv_flags
2976
2977 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2978 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2979 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2980 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2981
2982 =cut
2983 */
2984
2985 char *
Perl_sv_2pv_flags(pTHX_ SV * const sv,STRLEN * const lp,const I32 flags)2986 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2987 {
2988 char *s;
2989
2990 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2991
2992 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2993 && SvTYPE(sv) != SVt_PVFM);
2994 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2995 mg_get(sv);
2996 if (SvROK(sv)) {
2997 if (SvAMAGIC(sv)) {
2998 SV *tmpstr;
2999 if (flags & SV_SKIP_OVERLOAD)
3000 return NULL;
3001 tmpstr = AMG_CALLunary(sv, string_amg);
3002 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3003 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3004 /* Unwrap this: */
3005 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3006 */
3007
3008 char *pv;
3009 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3010 if (flags & SV_CONST_RETURN) {
3011 pv = (char *) SvPVX_const(tmpstr);
3012 } else {
3013 pv = (flags & SV_MUTABLE_RETURN)
3014 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3015 }
3016 if (lp)
3017 *lp = SvCUR(tmpstr);
3018 } else {
3019 pv = sv_2pv_flags(tmpstr, lp, flags);
3020 }
3021 if (SvUTF8(tmpstr))
3022 SvUTF8_on(sv);
3023 else
3024 SvUTF8_off(sv);
3025 return pv;
3026 }
3027 }
3028 {
3029 STRLEN len;
3030 char *retval;
3031 char *buffer;
3032 SV *const referent = SvRV(sv);
3033
3034 if (!referent) {
3035 len = 7;
3036 retval = buffer = savepvn("NULLREF", len);
3037 } else if (SvTYPE(referent) == SVt_REGEXP &&
3038 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3039 amagic_is_enabled(string_amg))) {
3040 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3041
3042 assert(re);
3043
3044 /* If the regex is UTF-8 we want the containing scalar to
3045 have an UTF-8 flag too */
3046 if (RX_UTF8(re))
3047 SvUTF8_on(sv);
3048 else
3049 SvUTF8_off(sv);
3050
3051 if (lp)
3052 *lp = RX_WRAPLEN(re);
3053
3054 return RX_WRAPPED(re);
3055 } else {
3056 const char *const typestr = sv_reftype(referent, 0);
3057 const STRLEN typelen = strlen(typestr);
3058 UV addr = PTR2UV(referent);
3059 const char *stashname = NULL;
3060 STRLEN stashnamelen = 0; /* hush, gcc */
3061 const char *buffer_end;
3062
3063 if (SvOBJECT(referent)) {
3064 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3065
3066 if (name) {
3067 stashname = HEK_KEY(name);
3068 stashnamelen = HEK_LEN(name);
3069
3070 if (HEK_UTF8(name)) {
3071 SvUTF8_on(sv);
3072 } else {
3073 SvUTF8_off(sv);
3074 }
3075 } else {
3076 stashname = "__ANON__";
3077 stashnamelen = 8;
3078 }
3079 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3080 + 2 * sizeof(UV) + 2 /* )\0 */;
3081 } else {
3082 len = typelen + 3 /* (0x */
3083 + 2 * sizeof(UV) + 2 /* )\0 */;
3084 }
3085
3086 Newx(buffer, len, char);
3087 buffer_end = retval = buffer + len;
3088
3089 /* Working backwards */
3090 *--retval = '\0';
3091 *--retval = ')';
3092 do {
3093 *--retval = PL_hexdigit[addr & 15];
3094 } while (addr >>= 4);
3095 *--retval = 'x';
3096 *--retval = '0';
3097 *--retval = '(';
3098
3099 retval -= typelen;
3100 memcpy(retval, typestr, typelen);
3101
3102 if (stashname) {
3103 *--retval = '=';
3104 retval -= stashnamelen;
3105 memcpy(retval, stashname, stashnamelen);
3106 }
3107 /* retval may not necessarily have reached the start of the
3108 buffer here. */
3109 assert (retval >= buffer);
3110
3111 len = buffer_end - retval - 1; /* -1 for that \0 */
3112 }
3113 if (lp)
3114 *lp = len;
3115 SAVEFREEPV(buffer);
3116 return retval;
3117 }
3118 }
3119
3120 if (SvPOKp(sv)) {
3121 if (lp)
3122 *lp = SvCUR(sv);
3123 if (flags & SV_MUTABLE_RETURN)
3124 return SvPVX_mutable(sv);
3125 if (flags & SV_CONST_RETURN)
3126 return (char *)SvPVX_const(sv);
3127 return SvPVX(sv);
3128 }
3129
3130 if (SvIOK(sv)) {
3131 /* I'm assuming that if both IV and NV are equally valid then
3132 converting the IV is going to be more efficient */
3133 const U32 isUIOK = SvIsUV(sv);
3134 /* The purpose of this union is to ensure that arr is aligned on
3135 a 2 byte boundary, because that is what uiv_2buf() requires */
3136 union {
3137 char arr[TYPE_CHARS(UV)];
3138 U16 dummy;
3139 } buf;
3140 char *ebuf, *ptr;
3141 STRLEN len;
3142
3143 if (SvTYPE(sv) < SVt_PVIV)
3144 sv_upgrade(sv, SVt_PVIV);
3145 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3146 len = ebuf - ptr;
3147 /* inlined from sv_setpvn */
3148 s = SvGROW_mutable(sv, len + 1);
3149 Move(ptr, s, len, char);
3150 s += len;
3151 *s = '\0';
3152 SvPOK_on(sv);
3153 }
3154 else if (SvNOK(sv)) {
3155 if (SvTYPE(sv) < SVt_PVNV)
3156 sv_upgrade(sv, SVt_PVNV);
3157 if (SvNVX(sv) == 0.0
3158 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3159 && !Perl_isnan(SvNVX(sv))
3160 #endif
3161 ) {
3162 s = SvGROW_mutable(sv, 2);
3163 *s++ = '0';
3164 *s = '\0';
3165 } else {
3166 STRLEN len;
3167 STRLEN size = 5; /* "-Inf\0" */
3168
3169 s = SvGROW_mutable(sv, size);
3170 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3171 if (len > 0) {
3172 s += len;
3173 SvPOK_on(sv);
3174 }
3175 else {
3176 /* some Xenix systems wipe out errno here */
3177 dSAVE_ERRNO;
3178
3179 size =
3180 1 + /* sign */
3181 1 + /* "." */
3182 NV_DIG +
3183 1 + /* "e" */
3184 1 + /* sign */
3185 5 + /* exponent digits */
3186 1 + /* \0 */
3187 2; /* paranoia */
3188
3189 s = SvGROW_mutable(sv, size);
3190 #ifndef USE_LOCALE_NUMERIC
3191 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3192
3193 SvPOK_on(sv);
3194 #else
3195 {
3196 bool local_radix;
3197 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3198 STORE_LC_NUMERIC_SET_TO_NEEDED();
3199
3200 local_radix = _NOT_IN_NUMERIC_STANDARD;
3201 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3202 size += SvCUR(PL_numeric_radix_sv) - 1;
3203 s = SvGROW_mutable(sv, size);
3204 }
3205
3206 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3207
3208 /* If the radix character is UTF-8, and actually is in the
3209 * output, turn on the UTF-8 flag for the scalar */
3210 if ( local_radix
3211 && SvUTF8(PL_numeric_radix_sv)
3212 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3213 {
3214 SvUTF8_on(sv);
3215 }
3216
3217 RESTORE_LC_NUMERIC();
3218 }
3219
3220 /* We don't call SvPOK_on(), because it may come to
3221 * pass that the locale changes so that the
3222 * stringification we just did is no longer correct. We
3223 * will have to re-stringify every time it is needed */
3224 #endif
3225 RESTORE_ERRNO;
3226 }
3227 while (*s) s++;
3228 }
3229 }
3230 else if (isGV_with_GP(sv)) {
3231 GV *const gv = MUTABLE_GV(sv);
3232 SV *const buffer = sv_newmortal();
3233
3234 gv_efullname3(buffer, gv, "*");
3235
3236 assert(SvPOK(buffer));
3237 if (SvUTF8(buffer))
3238 SvUTF8_on(sv);
3239 else
3240 SvUTF8_off(sv);
3241 if (lp)
3242 *lp = SvCUR(buffer);
3243 return SvPVX(buffer);
3244 }
3245 else {
3246 if (lp)
3247 *lp = 0;
3248 if (flags & SV_UNDEF_RETURNS_NULL)
3249 return NULL;
3250 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3251 report_uninit(sv);
3252 /* Typically the caller expects that sv_any is not NULL now. */
3253 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3254 sv_upgrade(sv, SVt_PV);
3255 return (char *)"";
3256 }
3257
3258 {
3259 const STRLEN len = s - SvPVX_const(sv);
3260 if (lp)
3261 *lp = len;
3262 SvCUR_set(sv, len);
3263 }
3264 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3265 PTR2UV(sv),SvPVX_const(sv)));
3266 if (flags & SV_CONST_RETURN)
3267 return (char *)SvPVX_const(sv);
3268 if (flags & SV_MUTABLE_RETURN)
3269 return SvPVX_mutable(sv);
3270 return SvPVX(sv);
3271 }
3272
3273 /*
3274 =for apidoc sv_copypv
3275
3276 Copies a stringified representation of the source SV into the
3277 destination SV. Automatically performs any necessary C<mg_get> and
3278 coercion of numeric values into strings. Guaranteed to preserve
3279 C<UTF8> flag even from overloaded objects. Similar in nature to
3280 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3281 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3282 would lose the UTF-8'ness of the PV.
3283
3284 =for apidoc sv_copypv_nomg
3285
3286 Like C<sv_copypv>, but doesn't invoke get magic first.
3287
3288 =for apidoc sv_copypv_flags
3289
3290 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3291 has the C<SV_GMAGIC> bit set.
3292
3293 =cut
3294 */
3295
3296 void
Perl_sv_copypv_flags(pTHX_ SV * const dsv,SV * const ssv,const I32 flags)3297 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3298 {
3299 STRLEN len;
3300 const char *s;
3301
3302 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3303
3304 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3305 sv_setpvn(dsv,s,len);
3306 if (SvUTF8(ssv))
3307 SvUTF8_on(dsv);
3308 else
3309 SvUTF8_off(dsv);
3310 }
3311
3312 /*
3313 =for apidoc sv_2pvbyte
3314
3315 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3316 to its length. May cause the SV to be downgraded from UTF-8 as a
3317 side-effect.
3318
3319 Usually accessed via the C<SvPVbyte> macro.
3320
3321 =cut
3322 */
3323
3324 char *
Perl_sv_2pvbyte(pTHX_ SV * sv,STRLEN * const lp)3325 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3326 {
3327 PERL_ARGS_ASSERT_SV_2PVBYTE;
3328
3329 SvGETMAGIC(sv);
3330 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3331 || isGV_with_GP(sv) || SvROK(sv)) {
3332 SV *sv2 = sv_newmortal();
3333 sv_copypv_nomg(sv2,sv);
3334 sv = sv2;
3335 }
3336 sv_utf8_downgrade(sv,0);
3337 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3338 }
3339
3340 /*
3341 =for apidoc sv_2pvutf8
3342
3343 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3344 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3345
3346 Usually accessed via the C<SvPVutf8> macro.
3347
3348 =cut
3349 */
3350
3351 char *
Perl_sv_2pvutf8(pTHX_ SV * sv,STRLEN * const lp)3352 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3353 {
3354 PERL_ARGS_ASSERT_SV_2PVUTF8;
3355
3356 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3357 || isGV_with_GP(sv) || SvROK(sv))
3358 sv = sv_mortalcopy(sv);
3359 else
3360 SvGETMAGIC(sv);
3361 sv_utf8_upgrade_nomg(sv);
3362 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3363 }
3364
3365
3366 /*
3367 =for apidoc sv_2bool
3368
3369 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3370 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3371 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3372
3373 =for apidoc sv_2bool_flags
3374
3375 This function is only used by C<sv_true()> and friends, and only if
3376 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3377 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3378
3379
3380 =cut
3381 */
3382
3383 bool
Perl_sv_2bool_flags(pTHX_ SV * sv,I32 flags)3384 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3385 {
3386 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3387
3388 restart:
3389 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3390
3391 if (!SvOK(sv))
3392 return 0;
3393 if (SvROK(sv)) {
3394 if (SvAMAGIC(sv)) {
3395 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3396 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3397 bool svb;
3398 sv = tmpsv;
3399 if(SvGMAGICAL(sv)) {
3400 flags = SV_GMAGIC;
3401 goto restart; /* call sv_2bool */
3402 }
3403 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3404 else if(!SvOK(sv)) {
3405 svb = 0;
3406 }
3407 else if(SvPOK(sv)) {
3408 svb = SvPVXtrue(sv);
3409 }
3410 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3411 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3412 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3413 }
3414 else {
3415 flags = 0;
3416 goto restart; /* call sv_2bool_nomg */
3417 }
3418 return cBOOL(svb);
3419 }
3420 }
3421 assert(SvRV(sv));
3422 return TRUE;
3423 }
3424 if (isREGEXP(sv))
3425 return
3426 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3427
3428 if (SvNOK(sv) && !SvPOK(sv))
3429 return SvNVX(sv) != 0.0;
3430
3431 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3432 }
3433
3434 /*
3435 =for apidoc sv_utf8_upgrade
3436
3437 Converts the PV of an SV to its UTF-8-encoded form.
3438 Forces the SV to string form if it is not already.
3439 Will C<mg_get> on C<sv> if appropriate.
3440 Always sets the C<SvUTF8> flag to avoid future validity checks even
3441 if the whole string is the same in UTF-8 as not.
3442 Returns the number of bytes in the converted string
3443
3444 This is not a general purpose byte encoding to Unicode interface:
3445 use the Encode extension for that.
3446
3447 =for apidoc sv_utf8_upgrade_nomg
3448
3449 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3450
3451 =for apidoc sv_utf8_upgrade_flags
3452
3453 Converts the PV of an SV to its UTF-8-encoded form.
3454 Forces the SV to string form if it is not already.
3455 Always sets the SvUTF8 flag to avoid future validity checks even
3456 if all the bytes are invariant in UTF-8.
3457 If C<flags> has C<SV_GMAGIC> bit set,
3458 will C<mg_get> on C<sv> if appropriate, else not.
3459
3460 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3461
3462 Returns the number of bytes in the converted string.
3463
3464 This is not a general purpose byte encoding to Unicode interface:
3465 use the Encode extension for that.
3466
3467 =for apidoc sv_utf8_upgrade_flags_grow
3468
3469 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3470 the number of unused bytes the string of C<sv> is guaranteed to have free after
3471 it upon return. This allows the caller to reserve extra space that it intends
3472 to fill, to avoid extra grows.
3473
3474 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3475 are implemented in terms of this function.
3476
3477 Returns the number of bytes in the converted string (not including the spares).
3478
3479 =cut
3480
3481 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3482 C<NUL> isn't guaranteed due to having other routines do the work in some input
3483 cases, or if the input is already flagged as being in utf8.
3484
3485 */
3486
3487 STRLEN
Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV * const sv,const I32 flags,STRLEN extra)3488 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3489 {
3490 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3491
3492 if (sv == &PL_sv_undef)
3493 return 0;
3494 if (!SvPOK_nog(sv)) {
3495 STRLEN len = 0;
3496 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3497 (void) sv_2pv_flags(sv,&len, flags);
3498 if (SvUTF8(sv)) {
3499 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3500 return len;
3501 }
3502 } else {
3503 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3504 }
3505 }
3506
3507 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3508 * compiled and individual nodes will remain non-utf8 even if the
3509 * stringified version of the pattern gets upgraded. Whether the
3510 * PVX of a REGEXP should be grown or we should just croak, I don't
3511 * know - DAPM */
3512 if (SvUTF8(sv) || isREGEXP(sv)) {
3513 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3514 return SvCUR(sv);
3515 }
3516
3517 if (SvIsCOW(sv)) {
3518 S_sv_uncow(aTHX_ sv, 0);
3519 }
3520
3521 if (SvCUR(sv) == 0) {
3522 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3523 byte */
3524 } else { /* Assume Latin-1/EBCDIC */
3525 /* This function could be much more efficient if we
3526 * had a FLAG in SVs to signal if there are any variant
3527 * chars in the PV. Given that there isn't such a flag
3528 * make the loop as fast as possible. */
3529 U8 * s = (U8 *) SvPVX_const(sv);
3530 U8 *t = s;
3531
3532 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3533
3534 /* utf8 conversion not needed because all are invariants. Mark
3535 * as UTF-8 even if no variant - saves scanning loop */
3536 SvUTF8_on(sv);
3537 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3538 return SvCUR(sv);
3539 }
3540
3541 /* Here, there is at least one variant (t points to the first one), so
3542 * the string should be converted to utf8. Everything from 's' to
3543 * 't - 1' will occupy only 1 byte each on output.
3544 *
3545 * Note that the incoming SV may not have a trailing '\0', as certain
3546 * code in pp_formline can send us partially built SVs.
3547 *
3548 * There are two main ways to convert. One is to create a new string
3549 * and go through the input starting from the beginning, appending each
3550 * converted value onto the new string as we go along. Going this
3551 * route, it's probably best to initially allocate enough space in the
3552 * string rather than possibly running out of space and having to
3553 * reallocate and then copy what we've done so far. Since everything
3554 * from 's' to 't - 1' is invariant, the destination can be initialized
3555 * with these using a fast memory copy. To be sure to allocate enough
3556 * space, one could use the worst case scenario, where every remaining
3557 * byte expands to two under UTF-8, or one could parse it and count
3558 * exactly how many do expand.
3559 *
3560 * The other way is to unconditionally parse the remainder of the
3561 * string to figure out exactly how big the expanded string will be,
3562 * growing if needed. Then start at the end of the string and place
3563 * the character there at the end of the unfilled space in the expanded
3564 * one, working backwards until reaching 't'.
3565 *
3566 * The problem with assuming the worst case scenario is that for very
3567 * long strings, we could allocate much more memory than actually
3568 * needed, which can create performance problems. If we have to parse
3569 * anyway, the second method is the winner as it may avoid an extra
3570 * copy. The code used to use the first method under some
3571 * circumstances, but now that there is faster variant counting on
3572 * ASCII platforms, the second method is used exclusively, eliminating
3573 * some code that no longer has to be maintained. */
3574
3575 {
3576 /* Count the total number of variants there are. We can start
3577 * just beyond the first one, which is known to be at 't' */
3578 const Size_t invariant_length = t - s;
3579 U8 * e = (U8 *) SvEND(sv);
3580
3581 /* The length of the left overs, plus 1. */
3582 const Size_t remaining_length_p1 = e - t;
3583
3584 /* We expand by 1 for the variant at 't' and one for each remaining
3585 * variant (we start looking at 't+1') */
3586 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3587
3588 /* +1 = trailing NUL */
3589 Size_t need = SvCUR(sv) + expansion + extra + 1;
3590 U8 * d;
3591
3592 /* Grow if needed */
3593 if (SvLEN(sv) < need) {
3594 t = invariant_length + (U8*) SvGROW(sv, need);
3595 e = t + remaining_length_p1;
3596 }
3597 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3598
3599 /* Set the NUL at the end */
3600 d = (U8 *) SvEND(sv);
3601 *d-- = '\0';
3602
3603 /* Having decremented d, it points to the position to put the
3604 * very last byte of the expanded string. Go backwards through
3605 * the string, copying and expanding as we go, stopping when we
3606 * get to the part that is invariant the rest of the way down */
3607
3608 e--;
3609 while (e >= t) {
3610 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3611 *d-- = *e;
3612 } else {
3613 *d-- = UTF8_EIGHT_BIT_LO(*e);
3614 *d-- = UTF8_EIGHT_BIT_HI(*e);
3615 }
3616 e--;
3617 }
3618
3619 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3620 /* Update pos. We do it at the end rather than during
3621 * the upgrade, to avoid slowing down the common case
3622 * (upgrade without pos).
3623 * pos can be stored as either bytes or characters. Since
3624 * this was previously a byte string we can just turn off
3625 * the bytes flag. */
3626 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3627 if (mg) {
3628 mg->mg_flags &= ~MGf_BYTES;
3629 }
3630 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3631 magic_setutf8(sv,mg); /* clear UTF8 cache */
3632 }
3633 }
3634 }
3635
3636 SvUTF8_on(sv);
3637 return SvCUR(sv);
3638 }
3639
3640 /*
3641 =for apidoc sv_utf8_downgrade
3642
3643 Attempts to convert the PV of an SV from characters to bytes.
3644 If the PV contains a character that cannot fit
3645 in a byte, this conversion will fail;
3646 in this case, either returns false or, if C<fail_ok> is not
3647 true, croaks.
3648
3649 This is not a general purpose Unicode to byte encoding interface:
3650 use the C<Encode> extension for that.
3651
3652 =cut
3653 */
3654
3655 bool
Perl_sv_utf8_downgrade(pTHX_ SV * const sv,const bool fail_ok)3656 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3657 {
3658 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3659
3660 if (SvPOKp(sv) && SvUTF8(sv)) {
3661 if (SvCUR(sv)) {
3662 U8 *s;
3663 STRLEN len;
3664 int mg_flags = SV_GMAGIC;
3665
3666 if (SvIsCOW(sv)) {
3667 S_sv_uncow(aTHX_ sv, 0);
3668 }
3669 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3670 /* update pos */
3671 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3672 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3673 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3674 SV_GMAGIC|SV_CONST_RETURN);
3675 mg_flags = 0; /* sv_pos_b2u does get magic */
3676 }
3677 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3678 magic_setutf8(sv,mg); /* clear UTF8 cache */
3679
3680 }
3681 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3682
3683 if (!utf8_to_bytes(s, &len)) {
3684 if (fail_ok)
3685 return FALSE;
3686 else {
3687 if (PL_op)
3688 Perl_croak(aTHX_ "Wide character in %s",
3689 OP_DESC(PL_op));
3690 else
3691 Perl_croak(aTHX_ "Wide character");
3692 }
3693 }
3694 SvCUR_set(sv, len);
3695 }
3696 }
3697 SvUTF8_off(sv);
3698 return TRUE;
3699 }
3700
3701 /*
3702 =for apidoc sv_utf8_encode
3703
3704 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3705 flag off so that it looks like octets again.
3706
3707 =cut
3708 */
3709
3710 void
Perl_sv_utf8_encode(pTHX_ SV * const sv)3711 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3712 {
3713 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3714
3715 if (SvREADONLY(sv)) {
3716 sv_force_normal_flags(sv, 0);
3717 }
3718 (void) sv_utf8_upgrade(sv);
3719 SvUTF8_off(sv);
3720 }
3721
3722 /*
3723 =for apidoc sv_utf8_decode
3724
3725 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3726 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3727 so that it looks like a character. If the PV contains only single-byte
3728 characters, the C<SvUTF8> flag stays off.
3729 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3730
3731 =cut
3732 */
3733
3734 bool
Perl_sv_utf8_decode(pTHX_ SV * const sv)3735 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3736 {
3737 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3738
3739 if (SvPOKp(sv)) {
3740 const U8 *start, *c, *first_variant;
3741
3742 /* The octets may have got themselves encoded - get them back as
3743 * bytes
3744 */
3745 if (!sv_utf8_downgrade(sv, TRUE))
3746 return FALSE;
3747
3748 /* it is actually just a matter of turning the utf8 flag on, but
3749 * we want to make sure everything inside is valid utf8 first.
3750 */
3751 c = start = (const U8 *) SvPVX_const(sv);
3752 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3753 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3754 return FALSE;
3755 SvUTF8_on(sv);
3756 }
3757 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3758 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3759 after this, clearing pos. Does anything on CPAN
3760 need this? */
3761 /* adjust pos to the start of a UTF8 char sequence */
3762 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3763 if (mg) {
3764 I32 pos = mg->mg_len;
3765 if (pos > 0) {
3766 for (c = start + pos; c > start; c--) {
3767 if (UTF8_IS_START(*c))
3768 break;
3769 }
3770 mg->mg_len = c - start;
3771 }
3772 }
3773 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3774 magic_setutf8(sv,mg); /* clear UTF8 cache */
3775 }
3776 }
3777 return TRUE;
3778 }
3779
3780 /*
3781 =for apidoc sv_setsv
3782
3783 Copies the contents of the source SV C<ssv> into the destination SV
3784 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3785 function if the source SV needs to be reused. Does not handle 'set' magic on
3786 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3787 performs a copy-by-value, obliterating any previous content of the
3788 destination.
3789
3790 You probably want to use one of the assortment of wrappers, such as
3791 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3792 C<SvSetMagicSV_nosteal>.
3793
3794 =for apidoc sv_setsv_flags
3795
3796 Copies the contents of the source SV C<ssv> into the destination SV
3797 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3798 function if the source SV needs to be reused. Does not handle 'set' magic.
3799 Loosely speaking, it performs a copy-by-value, obliterating any previous
3800 content of the destination.
3801 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3802 C<ssv> if appropriate, else not. If the C<flags>
3803 parameter has the C<SV_NOSTEAL> bit set then the
3804 buffers of temps will not be stolen. C<sv_setsv>
3805 and C<sv_setsv_nomg> are implemented in terms of this function.
3806
3807 You probably want to use one of the assortment of wrappers, such as
3808 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3809 C<SvSetMagicSV_nosteal>.
3810
3811 This is the primary function for copying scalars, and most other
3812 copy-ish functions and macros use this underneath.
3813
3814 =cut
3815 */
3816
3817 static void
S_glob_assign_glob(pTHX_ SV * const dstr,SV * const sstr,const int dtype)3818 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3819 {
3820 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3821 HV *old_stash = NULL;
3822
3823 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3824
3825 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3826 const char * const name = GvNAME(sstr);
3827 const STRLEN len = GvNAMELEN(sstr);
3828 {
3829 if (dtype >= SVt_PV) {
3830 SvPV_free(dstr);
3831 SvPV_set(dstr, 0);
3832 SvLEN_set(dstr, 0);
3833 SvCUR_set(dstr, 0);
3834 }
3835 SvUPGRADE(dstr, SVt_PVGV);
3836 (void)SvOK_off(dstr);
3837 isGV_with_GP_on(dstr);
3838 }
3839 GvSTASH(dstr) = GvSTASH(sstr);
3840 if (GvSTASH(dstr))
3841 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3842 gv_name_set(MUTABLE_GV(dstr), name, len,
3843 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3844 SvFAKE_on(dstr); /* can coerce to non-glob */
3845 }
3846
3847 if(GvGP(MUTABLE_GV(sstr))) {
3848 /* If source has method cache entry, clear it */
3849 if(GvCVGEN(sstr)) {
3850 SvREFCNT_dec(GvCV(sstr));
3851 GvCV_set(sstr, NULL);
3852 GvCVGEN(sstr) = 0;
3853 }
3854 /* If source has a real method, then a method is
3855 going to change */
3856 else if(
3857 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3858 ) {
3859 mro_changes = 1;
3860 }
3861 }
3862
3863 /* If dest already had a real method, that's a change as well */
3864 if(
3865 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3866 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3867 ) {
3868 mro_changes = 1;
3869 }
3870
3871 /* We don't need to check the name of the destination if it was not a
3872 glob to begin with. */
3873 if(dtype == SVt_PVGV) {
3874 const char * const name = GvNAME((const GV *)dstr);
3875 const STRLEN len = GvNAMELEN(dstr);
3876 if(memEQs(name, len, "ISA")
3877 /* The stash may have been detached from the symbol table, so
3878 check its name. */
3879 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3880 )
3881 mro_changes = 2;
3882 else {
3883 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3884 || (len == 1 && name[0] == ':')) {
3885 mro_changes = 3;
3886
3887 /* Set aside the old stash, so we can reset isa caches on
3888 its subclasses. */
3889 if((old_stash = GvHV(dstr)))
3890 /* Make sure we do not lose it early. */
3891 SvREFCNT_inc_simple_void_NN(
3892 sv_2mortal((SV *)old_stash)
3893 );
3894 }
3895 }
3896
3897 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3898 }
3899
3900 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3901 * so temporarily protect it */
3902 ENTER;
3903 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3904 gp_free(MUTABLE_GV(dstr));
3905 GvINTRO_off(dstr); /* one-shot flag */
3906 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3907 LEAVE;
3908
3909 if (SvTAINTED(sstr))
3910 SvTAINT(dstr);
3911 if (GvIMPORTED(dstr) != GVf_IMPORTED
3912 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3913 {
3914 GvIMPORTED_on(dstr);
3915 }
3916 GvMULTI_on(dstr);
3917 if(mro_changes == 2) {
3918 if (GvAV((const GV *)sstr)) {
3919 MAGIC *mg;
3920 SV * const sref = (SV *)GvAV((const GV *)dstr);
3921 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3922 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3923 AV * const ary = newAV();
3924 av_push(ary, mg->mg_obj); /* takes the refcount */
3925 mg->mg_obj = (SV *)ary;
3926 }
3927 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3928 }
3929 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3930 }
3931 mro_isa_changed_in(GvSTASH(dstr));
3932 }
3933 else if(mro_changes == 3) {
3934 HV * const stash = GvHV(dstr);
3935 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3936 mro_package_moved(
3937 stash, old_stash,
3938 (GV *)dstr, 0
3939 );
3940 }
3941 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3942 if (GvIO(dstr) && dtype == SVt_PVGV) {
3943 DEBUG_o(Perl_deb(aTHX_
3944 "glob_assign_glob clearing PL_stashcache\n"));
3945 /* It's a cache. It will rebuild itself quite happily.
3946 It's a lot of effort to work out exactly which key (or keys)
3947 might be invalidated by the creation of the this file handle.
3948 */
3949 hv_clear(PL_stashcache);
3950 }
3951 return;
3952 }
3953
3954 void
Perl_gv_setref(pTHX_ SV * const dstr,SV * const sstr)3955 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3956 {
3957 SV * const sref = SvRV(sstr);
3958 SV *dref;
3959 const int intro = GvINTRO(dstr);
3960 SV **location;
3961 U8 import_flag = 0;
3962 const U32 stype = SvTYPE(sref);
3963
3964 PERL_ARGS_ASSERT_GV_SETREF;
3965
3966 if (intro) {
3967 GvINTRO_off(dstr); /* one-shot flag */
3968 GvLINE(dstr) = CopLINE(PL_curcop);
3969 GvEGV(dstr) = MUTABLE_GV(dstr);
3970 }
3971 GvMULTI_on(dstr);
3972 switch (stype) {
3973 case SVt_PVCV:
3974 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3975 import_flag = GVf_IMPORTED_CV;
3976 goto common;
3977 case SVt_PVHV:
3978 location = (SV **) &GvHV(dstr);
3979 import_flag = GVf_IMPORTED_HV;
3980 goto common;
3981 case SVt_PVAV:
3982 location = (SV **) &GvAV(dstr);
3983 import_flag = GVf_IMPORTED_AV;
3984 goto common;
3985 case SVt_PVIO:
3986 location = (SV **) &GvIOp(dstr);
3987 goto common;
3988 case SVt_PVFM:
3989 location = (SV **) &GvFORM(dstr);
3990 goto common;
3991 default:
3992 location = &GvSV(dstr);
3993 import_flag = GVf_IMPORTED_SV;
3994 common:
3995 if (intro) {
3996 if (stype == SVt_PVCV) {
3997 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3998 if (GvCVGEN(dstr)) {
3999 SvREFCNT_dec(GvCV(dstr));
4000 GvCV_set(dstr, NULL);
4001 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4002 }
4003 }
4004 /* SAVEt_GVSLOT takes more room on the savestack and has more
4005 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4006 leave_scope needs access to the GV so it can reset method
4007 caches. We must use SAVEt_GVSLOT whenever the type is
4008 SVt_PVCV, even if the stash is anonymous, as the stash may
4009 gain a name somehow before leave_scope. */
4010 if (stype == SVt_PVCV) {
4011 /* There is no save_pushptrptrptr. Creating it for this
4012 one call site would be overkill. So inline the ss add
4013 routines here. */
4014 dSS_ADD;
4015 SS_ADD_PTR(dstr);
4016 SS_ADD_PTR(location);
4017 SS_ADD_PTR(SvREFCNT_inc(*location));
4018 SS_ADD_UV(SAVEt_GVSLOT);
4019 SS_ADD_END(4);
4020 }
4021 else SAVEGENERICSV(*location);
4022 }
4023 dref = *location;
4024 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4025 CV* const cv = MUTABLE_CV(*location);
4026 if (cv) {
4027 if (!GvCVGEN((const GV *)dstr) &&
4028 (CvROOT(cv) || CvXSUB(cv)) &&
4029 /* redundant check that avoids creating the extra SV
4030 most of the time: */
4031 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4032 {
4033 SV * const new_const_sv =
4034 CvCONST((const CV *)sref)
4035 ? cv_const_sv((const CV *)sref)
4036 : NULL;
4037 HV * const stash = GvSTASH((const GV *)dstr);
4038 report_redefined_cv(
4039 sv_2mortal(
4040 stash
4041 ? Perl_newSVpvf(aTHX_
4042 "%" HEKf "::%" HEKf,
4043 HEKfARG(HvNAME_HEK(stash)),
4044 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4045 : Perl_newSVpvf(aTHX_
4046 "%" HEKf,
4047 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4048 ),
4049 cv,
4050 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4051 );
4052 }
4053 if (!intro)
4054 cv_ckproto_len_flags(cv, (const GV *)dstr,
4055 SvPOK(sref) ? CvPROTO(sref) : NULL,
4056 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4057 SvPOK(sref) ? SvUTF8(sref) : 0);
4058 }
4059 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4060 GvASSUMECV_on(dstr);
4061 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4062 if (intro && GvREFCNT(dstr) > 1) {
4063 /* temporary remove extra savestack's ref */
4064 --GvREFCNT(dstr);
4065 gv_method_changed(dstr);
4066 ++GvREFCNT(dstr);
4067 }
4068 else gv_method_changed(dstr);
4069 }
4070 }
4071 *location = SvREFCNT_inc_simple_NN(sref);
4072 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4073 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4074 GvFLAGS(dstr) |= import_flag;
4075 }
4076
4077 if (stype == SVt_PVHV) {
4078 const char * const name = GvNAME((GV*)dstr);
4079 const STRLEN len = GvNAMELEN(dstr);
4080 if (
4081 (
4082 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4083 || (len == 1 && name[0] == ':')
4084 )
4085 && (!dref || HvENAME_get(dref))
4086 ) {
4087 mro_package_moved(
4088 (HV *)sref, (HV *)dref,
4089 (GV *)dstr, 0
4090 );
4091 }
4092 }
4093 else if (
4094 stype == SVt_PVAV && sref != dref
4095 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4096 /* The stash may have been detached from the symbol table, so
4097 check its name before doing anything. */
4098 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4099 ) {
4100 MAGIC *mg;
4101 MAGIC * const omg = dref && SvSMAGICAL(dref)
4102 ? mg_find(dref, PERL_MAGIC_isa)
4103 : NULL;
4104 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4105 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4106 AV * const ary = newAV();
4107 av_push(ary, mg->mg_obj); /* takes the refcount */
4108 mg->mg_obj = (SV *)ary;
4109 }
4110 if (omg) {
4111 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4112 SV **svp = AvARRAY((AV *)omg->mg_obj);
4113 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4114 while (items--)
4115 av_push(
4116 (AV *)mg->mg_obj,
4117 SvREFCNT_inc_simple_NN(*svp++)
4118 );
4119 }
4120 else
4121 av_push(
4122 (AV *)mg->mg_obj,
4123 SvREFCNT_inc_simple_NN(omg->mg_obj)
4124 );
4125 }
4126 else
4127 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4128 }
4129 else
4130 {
4131 SSize_t i;
4132 sv_magic(
4133 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4134 );
4135 for (i = 0; i <= AvFILL(sref); ++i) {
4136 SV **elem = av_fetch ((AV*)sref, i, 0);
4137 if (elem) {
4138 sv_magic(
4139 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4140 );
4141 }
4142 }
4143 mg = mg_find(sref, PERL_MAGIC_isa);
4144 }
4145 /* Since the *ISA assignment could have affected more than
4146 one stash, don't call mro_isa_changed_in directly, but let
4147 magic_clearisa do it for us, as it already has the logic for
4148 dealing with globs vs arrays of globs. */
4149 assert(mg);
4150 Perl_magic_clearisa(aTHX_ NULL, mg);
4151 }
4152 else if (stype == SVt_PVIO) {
4153 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4154 /* It's a cache. It will rebuild itself quite happily.
4155 It's a lot of effort to work out exactly which key (or keys)
4156 might be invalidated by the creation of the this file handle.
4157 */
4158 hv_clear(PL_stashcache);
4159 }
4160 break;
4161 }
4162 if (!intro) SvREFCNT_dec(dref);
4163 if (SvTAINTED(sstr))
4164 SvTAINT(dstr);
4165 return;
4166 }
4167
4168
4169
4170
4171 #ifdef PERL_DEBUG_READONLY_COW
4172 # include <sys/mman.h>
4173
4174 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4175 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4176 # endif
4177
4178 void
Perl_sv_buf_to_ro(pTHX_ SV * sv)4179 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4180 {
4181 struct perl_memory_debug_header * const header =
4182 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4183 const MEM_SIZE len = header->size;
4184 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4185 # ifdef PERL_TRACK_MEMPOOL
4186 if (!header->readonly) header->readonly = 1;
4187 # endif
4188 if (mprotect(header, len, PROT_READ))
4189 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4190 header, len, errno);
4191 }
4192
4193 static void
S_sv_buf_to_rw(pTHX_ SV * sv)4194 S_sv_buf_to_rw(pTHX_ SV *sv)
4195 {
4196 struct perl_memory_debug_header * const header =
4197 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4198 const MEM_SIZE len = header->size;
4199 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4200 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4201 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4202 header, len, errno);
4203 # ifdef PERL_TRACK_MEMPOOL
4204 header->readonly = 0;
4205 # endif
4206 }
4207
4208 #else
4209 # define sv_buf_to_ro(sv) NOOP
4210 # define sv_buf_to_rw(sv) NOOP
4211 #endif
4212
4213 void
Perl_sv_setsv_flags(pTHX_ SV * dstr,SV * sstr,const I32 flags)4214 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4215 {
4216 U32 sflags;
4217 int dtype;
4218 svtype stype;
4219 unsigned int both_type;
4220
4221 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4222
4223 if (UNLIKELY( sstr == dstr ))
4224 return;
4225
4226 if (UNLIKELY( !sstr ))
4227 sstr = &PL_sv_undef;
4228
4229 stype = SvTYPE(sstr);
4230 dtype = SvTYPE(dstr);
4231 both_type = (stype | dtype);
4232
4233 /* with these values, we can check that both SVs are NULL/IV (and not
4234 * freed) just by testing the or'ed types */
4235 STATIC_ASSERT_STMT(SVt_NULL == 0);
4236 STATIC_ASSERT_STMT(SVt_IV == 1);
4237 if (both_type <= 1) {
4238 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4239 * special-casing */
4240 U32 sflags;
4241 U32 new_dflags;
4242 SV *old_rv = NULL;
4243
4244 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4245 if (SvREADONLY(dstr))
4246 Perl_croak_no_modify();
4247 if (SvROK(dstr)) {
4248 if (SvWEAKREF(dstr))
4249 sv_unref_flags(dstr, 0);
4250 else
4251 old_rv = SvRV(dstr);
4252 }
4253
4254 assert(!SvGMAGICAL(sstr));
4255 assert(!SvGMAGICAL(dstr));
4256
4257 sflags = SvFLAGS(sstr);
4258 if (sflags & (SVf_IOK|SVf_ROK)) {
4259 SET_SVANY_FOR_BODYLESS_IV(dstr);
4260 new_dflags = SVt_IV;
4261
4262 if (sflags & SVf_ROK) {
4263 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4264 new_dflags |= SVf_ROK;
4265 }
4266 else {
4267 /* both src and dst are <= SVt_IV, so sv_any points to the
4268 * head; so access the head directly
4269 */
4270 assert( &(sstr->sv_u.svu_iv)
4271 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4272 assert( &(dstr->sv_u.svu_iv)
4273 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4274 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4275 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4276 }
4277 }
4278 else {
4279 new_dflags = dtype; /* turn off everything except the type */
4280 }
4281 SvFLAGS(dstr) = new_dflags;
4282 SvREFCNT_dec(old_rv);
4283
4284 return;
4285 }
4286
4287 if (UNLIKELY(both_type == SVTYPEMASK)) {
4288 if (SvIS_FREED(dstr)) {
4289 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4290 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4291 }
4292 if (SvIS_FREED(sstr)) {
4293 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4294 (void*)sstr, (void*)dstr);
4295 }
4296 }
4297
4298
4299
4300 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4301 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4302
4303 /* There's a lot of redundancy below but we're going for speed here */
4304
4305 switch (stype) {
4306 case SVt_NULL:
4307 undef_sstr:
4308 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4309 (void)SvOK_off(dstr);
4310 return;
4311 }
4312 break;
4313 case SVt_IV:
4314 if (SvIOK(sstr)) {
4315 switch (dtype) {
4316 case SVt_NULL:
4317 /* For performance, we inline promoting to type SVt_IV. */
4318 /* We're starting from SVt_NULL, so provided that define is
4319 * actual 0, we don't have to unset any SV type flags
4320 * to promote to SVt_IV. */
4321 STATIC_ASSERT_STMT(SVt_NULL == 0);
4322 SET_SVANY_FOR_BODYLESS_IV(dstr);
4323 SvFLAGS(dstr) |= SVt_IV;
4324 break;
4325 case SVt_NV:
4326 case SVt_PV:
4327 sv_upgrade(dstr, SVt_PVIV);
4328 break;
4329 case SVt_PVGV:
4330 case SVt_PVLV:
4331 goto end_of_first_switch;
4332 }
4333 (void)SvIOK_only(dstr);
4334 SvIV_set(dstr, SvIVX(sstr));
4335 if (SvIsUV(sstr))
4336 SvIsUV_on(dstr);
4337 /* SvTAINTED can only be true if the SV has taint magic, which in
4338 turn means that the SV type is PVMG (or greater). This is the
4339 case statement for SVt_IV, so this cannot be true (whatever gcov
4340 may say). */
4341 assert(!SvTAINTED(sstr));
4342 return;
4343 }
4344 if (!SvROK(sstr))
4345 goto undef_sstr;
4346 if (dtype < SVt_PV && dtype != SVt_IV)
4347 sv_upgrade(dstr, SVt_IV);
4348 break;
4349
4350 case SVt_NV:
4351 if (LIKELY( SvNOK(sstr) )) {
4352 switch (dtype) {
4353 case SVt_NULL:
4354 case SVt_IV:
4355 sv_upgrade(dstr, SVt_NV);
4356 break;
4357 case SVt_PV:
4358 case SVt_PVIV:
4359 sv_upgrade(dstr, SVt_PVNV);
4360 break;
4361 case SVt_PVGV:
4362 case SVt_PVLV:
4363 goto end_of_first_switch;
4364 }
4365 SvNV_set(dstr, SvNVX(sstr));
4366 (void)SvNOK_only(dstr);
4367 /* SvTAINTED can only be true if the SV has taint magic, which in
4368 turn means that the SV type is PVMG (or greater). This is the
4369 case statement for SVt_NV, so this cannot be true (whatever gcov
4370 may say). */
4371 assert(!SvTAINTED(sstr));
4372 return;
4373 }
4374 goto undef_sstr;
4375
4376 case SVt_PV:
4377 if (dtype < SVt_PV)
4378 sv_upgrade(dstr, SVt_PV);
4379 break;
4380 case SVt_PVIV:
4381 if (dtype < SVt_PVIV)
4382 sv_upgrade(dstr, SVt_PVIV);
4383 break;
4384 case SVt_PVNV:
4385 if (dtype < SVt_PVNV)
4386 sv_upgrade(dstr, SVt_PVNV);
4387 break;
4388
4389 case SVt_INVLIST:
4390 invlist_clone(sstr, dstr);
4391 break;
4392 default:
4393 {
4394 const char * const type = sv_reftype(sstr,0);
4395 if (PL_op)
4396 /* diag_listed_as: Bizarre copy of %s */
4397 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4398 else
4399 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4400 }
4401 NOT_REACHED; /* NOTREACHED */
4402
4403 case SVt_REGEXP:
4404 upgregexp:
4405 if (dtype < SVt_REGEXP)
4406 sv_upgrade(dstr, SVt_REGEXP);
4407 break;
4408
4409 case SVt_PVLV:
4410 case SVt_PVGV:
4411 case SVt_PVMG:
4412 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4413 mg_get(sstr);
4414 if (SvTYPE(sstr) != stype)
4415 stype = SvTYPE(sstr);
4416 }
4417 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4418 glob_assign_glob(dstr, sstr, dtype);
4419 return;
4420 }
4421 if (stype == SVt_PVLV)
4422 {
4423 if (isREGEXP(sstr)) goto upgregexp;
4424 SvUPGRADE(dstr, SVt_PVNV);
4425 }
4426 else
4427 SvUPGRADE(dstr, (svtype)stype);
4428 }
4429 end_of_first_switch:
4430
4431 /* dstr may have been upgraded. */
4432 dtype = SvTYPE(dstr);
4433 sflags = SvFLAGS(sstr);
4434
4435 if (UNLIKELY( dtype == SVt_PVCV )) {
4436 /* Assigning to a subroutine sets the prototype. */
4437 if (SvOK(sstr)) {
4438 STRLEN len;
4439 const char *const ptr = SvPV_const(sstr, len);
4440
4441 SvGROW(dstr, len + 1);
4442 Copy(ptr, SvPVX(dstr), len + 1, char);
4443 SvCUR_set(dstr, len);
4444 SvPOK_only(dstr);
4445 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4446 CvAUTOLOAD_off(dstr);
4447 } else {
4448 SvOK_off(dstr);
4449 }
4450 }
4451 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4452 || dtype == SVt_PVFM))
4453 {
4454 const char * const type = sv_reftype(dstr,0);
4455 if (PL_op)
4456 /* diag_listed_as: Cannot copy to %s */
4457 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4458 else
4459 Perl_croak(aTHX_ "Cannot copy to %s", type);
4460 } else if (sflags & SVf_ROK) {
4461 if (isGV_with_GP(dstr)
4462 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4463 sstr = SvRV(sstr);
4464 if (sstr == dstr) {
4465 if (GvIMPORTED(dstr) != GVf_IMPORTED
4466 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4467 {
4468 GvIMPORTED_on(dstr);
4469 }
4470 GvMULTI_on(dstr);
4471 return;
4472 }
4473 glob_assign_glob(dstr, sstr, dtype);
4474 return;
4475 }
4476
4477 if (dtype >= SVt_PV) {
4478 if (isGV_with_GP(dstr)) {
4479 gv_setref(dstr, sstr);
4480 return;
4481 }
4482 if (SvPVX_const(dstr)) {
4483 SvPV_free(dstr);
4484 SvLEN_set(dstr, 0);
4485 SvCUR_set(dstr, 0);
4486 }
4487 }
4488 (void)SvOK_off(dstr);
4489 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4490 SvFLAGS(dstr) |= sflags & SVf_ROK;
4491 assert(!(sflags & SVp_NOK));
4492 assert(!(sflags & SVp_IOK));
4493 assert(!(sflags & SVf_NOK));
4494 assert(!(sflags & SVf_IOK));
4495 }
4496 else if (isGV_with_GP(dstr)) {
4497 if (!(sflags & SVf_OK)) {
4498 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4499 "Undefined value assigned to typeglob");
4500 }
4501 else {
4502 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4503 if (dstr != (const SV *)gv) {
4504 const char * const name = GvNAME((const GV *)dstr);
4505 const STRLEN len = GvNAMELEN(dstr);
4506 HV *old_stash = NULL;
4507 bool reset_isa = FALSE;
4508 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4509 || (len == 1 && name[0] == ':')) {
4510 /* Set aside the old stash, so we can reset isa caches
4511 on its subclasses. */
4512 if((old_stash = GvHV(dstr))) {
4513 /* Make sure we do not lose it early. */
4514 SvREFCNT_inc_simple_void_NN(
4515 sv_2mortal((SV *)old_stash)
4516 );
4517 }
4518 reset_isa = TRUE;
4519 }
4520
4521 if (GvGP(dstr)) {
4522 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4523 gp_free(MUTABLE_GV(dstr));
4524 }
4525 GvGP_set(dstr, gp_ref(GvGP(gv)));
4526
4527 if (reset_isa) {
4528 HV * const stash = GvHV(dstr);
4529 if(
4530 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4531 )
4532 mro_package_moved(
4533 stash, old_stash,
4534 (GV *)dstr, 0
4535 );
4536 }
4537 }
4538 }
4539 }
4540 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4541 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4542 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4543 }
4544 else if (sflags & SVp_POK) {
4545 const STRLEN cur = SvCUR(sstr);
4546 const STRLEN len = SvLEN(sstr);
4547
4548 /*
4549 * We have three basic ways to copy the string:
4550 *
4551 * 1. Swipe
4552 * 2. Copy-on-write
4553 * 3. Actual copy
4554 *
4555 * Which we choose is based on various factors. The following
4556 * things are listed in order of speed, fastest to slowest:
4557 * - Swipe
4558 * - Copying a short string
4559 * - Copy-on-write bookkeeping
4560 * - malloc
4561 * - Copying a long string
4562 *
4563 * We swipe the string (steal the string buffer) if the SV on the
4564 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4565 * big win on long strings. It should be a win on short strings if
4566 * SvPVX_const(dstr) has to be allocated. If not, it should not
4567 * slow things down, as SvPVX_const(sstr) would have been freed
4568 * soon anyway.
4569 *
4570 * We also steal the buffer from a PADTMP (operator target) if it
4571 * is ‘long enough’. For short strings, a swipe does not help
4572 * here, as it causes more malloc calls the next time the target
4573 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4574 * be allocated it is still not worth swiping PADTMPs for short
4575 * strings, as the savings here are small.
4576 *
4577 * If swiping is not an option, then we see whether it is
4578 * worth using copy-on-write. If the lhs already has a buf-
4579 * fer big enough and the string is short, we skip it and fall back
4580 * to method 3, since memcpy is faster for short strings than the
4581 * later bookkeeping overhead that copy-on-write entails.
4582
4583 * If the rhs is not a copy-on-write string yet, then we also
4584 * consider whether the buffer is too large relative to the string
4585 * it holds. Some operations such as readline allocate a large
4586 * buffer in the expectation of reusing it. But turning such into
4587 * a COW buffer is counter-productive because it increases memory
4588 * usage by making readline allocate a new large buffer the sec-
4589 * ond time round. So, if the buffer is too large, again, we use
4590 * method 3 (copy).
4591 *
4592 * Finally, if there is no buffer on the left, or the buffer is too
4593 * small, then we use copy-on-write and make both SVs share the
4594 * string buffer.
4595 *
4596 */
4597
4598 /* Whichever path we take through the next code, we want this true,
4599 and doing it now facilitates the COW check. */
4600 (void)SvPOK_only(dstr);
4601
4602 if (
4603 ( /* Either ... */
4604 /* slated for free anyway (and not COW)? */
4605 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4606 /* or a swipable TARG */
4607 || ((sflags &
4608 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4609 == SVs_PADTMP
4610 /* whose buffer is worth stealing */
4611 && CHECK_COWBUF_THRESHOLD(cur,len)
4612 )
4613 ) &&
4614 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4615 (!(flags & SV_NOSTEAL)) &&
4616 /* and we're allowed to steal temps */
4617 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4618 len) /* and really is a string */
4619 { /* Passes the swipe test. */
4620 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4621 SvPV_free(dstr);
4622 SvPV_set(dstr, SvPVX_mutable(sstr));
4623 SvLEN_set(dstr, SvLEN(sstr));
4624 SvCUR_set(dstr, SvCUR(sstr));
4625
4626 SvTEMP_off(dstr);
4627 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4628 SvPV_set(sstr, NULL);
4629 SvLEN_set(sstr, 0);
4630 SvCUR_set(sstr, 0);
4631 SvTEMP_off(sstr);
4632 }
4633 else if (flags & SV_COW_SHARED_HASH_KEYS
4634 &&
4635 #ifdef PERL_COPY_ON_WRITE
4636 (sflags & SVf_IsCOW
4637 ? (!len ||
4638 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4639 /* If this is a regular (non-hek) COW, only so
4640 many COW "copies" are possible. */
4641 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4642 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4643 && !(SvFLAGS(dstr) & SVf_BREAK)
4644 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4645 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4646 ))
4647 #else
4648 sflags & SVf_IsCOW
4649 && !(SvFLAGS(dstr) & SVf_BREAK)
4650 #endif
4651 ) {
4652 /* Either it's a shared hash key, or it's suitable for
4653 copy-on-write. */
4654 #ifdef DEBUGGING
4655 if (DEBUG_C_TEST) {
4656 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4657 sv_dump(sstr);
4658 sv_dump(dstr);
4659 }
4660 #endif
4661 #ifdef PERL_ANY_COW
4662 if (!(sflags & SVf_IsCOW)) {
4663 SvIsCOW_on(sstr);
4664 CowREFCNT(sstr) = 0;
4665 }
4666 #endif
4667 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4668 SvPV_free(dstr);
4669 }
4670
4671 #ifdef PERL_ANY_COW
4672 if (len) {
4673 if (sflags & SVf_IsCOW) {
4674 sv_buf_to_rw(sstr);
4675 }
4676 CowREFCNT(sstr)++;
4677 SvPV_set(dstr, SvPVX_mutable(sstr));
4678 sv_buf_to_ro(sstr);
4679 } else
4680 #endif
4681 {
4682 /* SvIsCOW_shared_hash */
4683 DEBUG_C(PerlIO_printf(Perl_debug_log,
4684 "Copy on write: Sharing hash\n"));
4685
4686 assert (SvTYPE(dstr) >= SVt_PV);
4687 SvPV_set(dstr,
4688 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4689 }
4690 SvLEN_set(dstr, len);
4691 SvCUR_set(dstr, cur);
4692 SvIsCOW_on(dstr);
4693 } else {
4694 /* Failed the swipe test, and we cannot do copy-on-write either.
4695 Have to copy the string. */
4696 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4697 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4698 SvCUR_set(dstr, cur);
4699 *SvEND(dstr) = '\0';
4700 }
4701 if (sflags & SVp_NOK) {
4702 SvNV_set(dstr, SvNVX(sstr));
4703 }
4704 if (sflags & SVp_IOK) {
4705 SvIV_set(dstr, SvIVX(sstr));
4706 if (sflags & SVf_IVisUV)
4707 SvIsUV_on(dstr);
4708 }
4709 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4710 {
4711 const MAGIC * const smg = SvVSTRING_mg(sstr);
4712 if (smg) {
4713 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4714 smg->mg_ptr, smg->mg_len);
4715 SvRMAGICAL_on(dstr);
4716 }
4717 }
4718 }
4719 else if (sflags & (SVp_IOK|SVp_NOK)) {
4720 (void)SvOK_off(dstr);
4721 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4722 if (sflags & SVp_IOK) {
4723 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4724 SvIV_set(dstr, SvIVX(sstr));
4725 }
4726 if (sflags & SVp_NOK) {
4727 SvNV_set(dstr, SvNVX(sstr));
4728 }
4729 }
4730 else {
4731 if (isGV_with_GP(sstr)) {
4732 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4733 }
4734 else
4735 (void)SvOK_off(dstr);
4736 }
4737 if (SvTAINTED(sstr))
4738 SvTAINT(dstr);
4739 }
4740
4741
4742 /*
4743 =for apidoc sv_set_undef
4744
4745 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4746 Doesn't handle set magic.
4747
4748 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4749 buffer, unlike C<undef $sv>.
4750
4751 Introduced in perl 5.25.12.
4752
4753 =cut
4754 */
4755
4756 void
Perl_sv_set_undef(pTHX_ SV * sv)4757 Perl_sv_set_undef(pTHX_ SV *sv)
4758 {
4759 U32 type = SvTYPE(sv);
4760
4761 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4762
4763 /* shortcut, NULL, IV, RV */
4764
4765 if (type <= SVt_IV) {
4766 assert(!SvGMAGICAL(sv));
4767 if (SvREADONLY(sv)) {
4768 /* does undeffing PL_sv_undef count as modifying a read-only
4769 * variable? Some XS code does this */
4770 if (sv == &PL_sv_undef)
4771 return;
4772 Perl_croak_no_modify();
4773 }
4774
4775 if (SvROK(sv)) {
4776 if (SvWEAKREF(sv))
4777 sv_unref_flags(sv, 0);
4778 else {
4779 SV *rv = SvRV(sv);
4780 SvFLAGS(sv) = type; /* quickly turn off all flags */
4781 SvREFCNT_dec_NN(rv);
4782 return;
4783 }
4784 }
4785 SvFLAGS(sv) = type; /* quickly turn off all flags */
4786 return;
4787 }
4788
4789 if (SvIS_FREED(sv))
4790 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4791 (void *)sv);
4792
4793 SV_CHECK_THINKFIRST_COW_DROP(sv);
4794
4795 if (isGV_with_GP(sv))
4796 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4797 "Undefined value assigned to typeglob");
4798 else
4799 SvOK_off(sv);
4800 }
4801
4802
4803
4804 /*
4805 =for apidoc sv_setsv_mg
4806
4807 Like C<sv_setsv>, but also handles 'set' magic.
4808
4809 =cut
4810 */
4811
4812 void
Perl_sv_setsv_mg(pTHX_ SV * const dstr,SV * const sstr)4813 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4814 {
4815 PERL_ARGS_ASSERT_SV_SETSV_MG;
4816
4817 sv_setsv(dstr,sstr);
4818 SvSETMAGIC(dstr);
4819 }
4820
4821 #ifdef PERL_ANY_COW
4822 # define SVt_COW SVt_PV
4823 SV *
Perl_sv_setsv_cow(pTHX_ SV * dstr,SV * sstr)4824 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4825 {
4826 STRLEN cur = SvCUR(sstr);
4827 STRLEN len = SvLEN(sstr);
4828 char *new_pv;
4829 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4830 const bool already = cBOOL(SvIsCOW(sstr));
4831 #endif
4832
4833 PERL_ARGS_ASSERT_SV_SETSV_COW;
4834 #ifdef DEBUGGING
4835 if (DEBUG_C_TEST) {
4836 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4837 (void*)sstr, (void*)dstr);
4838 sv_dump(sstr);
4839 if (dstr)
4840 sv_dump(dstr);
4841 }
4842 #endif
4843 if (dstr) {
4844 if (SvTHINKFIRST(dstr))
4845 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4846 else if (SvPVX_const(dstr))
4847 Safefree(SvPVX_mutable(dstr));
4848 }
4849 else
4850 new_SV(dstr);
4851 SvUPGRADE(dstr, SVt_COW);
4852
4853 assert (SvPOK(sstr));
4854 assert (SvPOKp(sstr));
4855
4856 if (SvIsCOW(sstr)) {
4857
4858 if (SvLEN(sstr) == 0) {
4859 /* source is a COW shared hash key. */
4860 DEBUG_C(PerlIO_printf(Perl_debug_log,
4861 "Fast copy on write: Sharing hash\n"));
4862 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4863 goto common_exit;
4864 }
4865 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4866 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4867 } else {
4868 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4869 SvUPGRADE(sstr, SVt_COW);
4870 SvIsCOW_on(sstr);
4871 DEBUG_C(PerlIO_printf(Perl_debug_log,
4872 "Fast copy on write: Converting sstr to COW\n"));
4873 CowREFCNT(sstr) = 0;
4874 }
4875 # ifdef PERL_DEBUG_READONLY_COW
4876 if (already) sv_buf_to_rw(sstr);
4877 # endif
4878 CowREFCNT(sstr)++;
4879 new_pv = SvPVX_mutable(sstr);
4880 sv_buf_to_ro(sstr);
4881
4882 common_exit:
4883 SvPV_set(dstr, new_pv);
4884 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4885 if (SvUTF8(sstr))
4886 SvUTF8_on(dstr);
4887 SvLEN_set(dstr, len);
4888 SvCUR_set(dstr, cur);
4889 #ifdef DEBUGGING
4890 if (DEBUG_C_TEST)
4891 sv_dump(dstr);
4892 #endif
4893 return dstr;
4894 }
4895 #endif
4896
4897 /*
4898 =for apidoc sv_setpv_bufsize
4899
4900 Sets the SV to be a string of cur bytes length, with at least
4901 len bytes available. Ensures that there is a null byte at SvEND.
4902 Returns a char * pointer to the SvPV buffer.
4903
4904 =cut
4905 */
4906
4907 char *
Perl_sv_setpv_bufsize(pTHX_ SV * const sv,const STRLEN cur,const STRLEN len)4908 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4909 {
4910 char *pv;
4911
4912 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4913
4914 SV_CHECK_THINKFIRST_COW_DROP(sv);
4915 SvUPGRADE(sv, SVt_PV);
4916 pv = SvGROW(sv, len + 1);
4917 SvCUR_set(sv, cur);
4918 *(SvEND(sv))= '\0';
4919 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4920
4921 SvTAINT(sv);
4922 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4923 return pv;
4924 }
4925
4926 /*
4927 =for apidoc sv_setpvn
4928
4929 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4930 The C<len> parameter indicates the number of
4931 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4932 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4933
4934 =cut
4935 */
4936
4937 void
Perl_sv_setpvn(pTHX_ SV * const sv,const char * const ptr,const STRLEN len)4938 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4939 {
4940 char *dptr;
4941
4942 PERL_ARGS_ASSERT_SV_SETPVN;
4943
4944 SV_CHECK_THINKFIRST_COW_DROP(sv);
4945 if (isGV_with_GP(sv))
4946 Perl_croak_no_modify();
4947 if (!ptr) {
4948 (void)SvOK_off(sv);
4949 return;
4950 }
4951 else {
4952 /* len is STRLEN which is unsigned, need to copy to signed */
4953 const IV iv = len;
4954 if (iv < 0)
4955 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4956 IVdf, iv);
4957 }
4958 SvUPGRADE(sv, SVt_PV);
4959
4960 dptr = SvGROW(sv, len + 1);
4961 Move(ptr,dptr,len,char);
4962 dptr[len] = '\0';
4963 SvCUR_set(sv, len);
4964 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4965 SvTAINT(sv);
4966 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4967 }
4968
4969 /*
4970 =for apidoc sv_setpvn_mg
4971
4972 Like C<sv_setpvn>, but also handles 'set' magic.
4973
4974 =cut
4975 */
4976
4977 void
Perl_sv_setpvn_mg(pTHX_ SV * const sv,const char * const ptr,const STRLEN len)4978 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4979 {
4980 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4981
4982 sv_setpvn(sv,ptr,len);
4983 SvSETMAGIC(sv);
4984 }
4985
4986 /*
4987 =for apidoc sv_setpv
4988
4989 Copies a string into an SV. The string must be terminated with a C<NUL>
4990 character, and not contain embeded C<NUL>'s.
4991 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4992
4993 =cut
4994 */
4995
4996 void
Perl_sv_setpv(pTHX_ SV * const sv,const char * const ptr)4997 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4998 {
4999 STRLEN len;
5000
5001 PERL_ARGS_ASSERT_SV_SETPV;
5002
5003 SV_CHECK_THINKFIRST_COW_DROP(sv);
5004 if (!ptr) {
5005 (void)SvOK_off(sv);
5006 return;
5007 }
5008 len = strlen(ptr);
5009 SvUPGRADE(sv, SVt_PV);
5010
5011 SvGROW(sv, len + 1);
5012 Move(ptr,SvPVX(sv),len+1,char);
5013 SvCUR_set(sv, len);
5014 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5015 SvTAINT(sv);
5016 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5017 }
5018
5019 /*
5020 =for apidoc sv_setpv_mg
5021
5022 Like C<sv_setpv>, but also handles 'set' magic.
5023
5024 =cut
5025 */
5026
5027 void
Perl_sv_setpv_mg(pTHX_ SV * const sv,const char * const ptr)5028 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5029 {
5030 PERL_ARGS_ASSERT_SV_SETPV_MG;
5031
5032 sv_setpv(sv,ptr);
5033 SvSETMAGIC(sv);
5034 }
5035
5036 void
Perl_sv_sethek(pTHX_ SV * const sv,const HEK * const hek)5037 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5038 {
5039 PERL_ARGS_ASSERT_SV_SETHEK;
5040
5041 if (!hek) {
5042 return;
5043 }
5044
5045 if (HEK_LEN(hek) == HEf_SVKEY) {
5046 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5047 return;
5048 } else {
5049 const int flags = HEK_FLAGS(hek);
5050 if (flags & HVhek_WASUTF8) {
5051 STRLEN utf8_len = HEK_LEN(hek);
5052 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5053 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5054 SvUTF8_on(sv);
5055 return;
5056 } else if (flags & HVhek_UNSHARED) {
5057 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5058 if (HEK_UTF8(hek))
5059 SvUTF8_on(sv);
5060 else SvUTF8_off(sv);
5061 return;
5062 }
5063 {
5064 SV_CHECK_THINKFIRST_COW_DROP(sv);
5065 SvUPGRADE(sv, SVt_PV);
5066 SvPV_free(sv);
5067 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5068 SvCUR_set(sv, HEK_LEN(hek));
5069 SvLEN_set(sv, 0);
5070 SvIsCOW_on(sv);
5071 SvPOK_on(sv);
5072 if (HEK_UTF8(hek))
5073 SvUTF8_on(sv);
5074 else SvUTF8_off(sv);
5075 return;
5076 }
5077 }
5078 }
5079
5080
5081 /*
5082 =for apidoc sv_usepvn_flags
5083
5084 Tells an SV to use C<ptr> to find its string value. Normally the
5085 string is stored inside the SV, but sv_usepvn allows the SV to use an
5086 outside string. C<ptr> should point to memory that was allocated
5087 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5088 the start of a C<Newx>-ed block of memory, and not a pointer to the
5089 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5090 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5091 string length, C<len>, must be supplied. By default this function
5092 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5093 so that pointer should not be freed or used by the programmer after
5094 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5095 that pointer (e.g. ptr + 1) be used.
5096
5097 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5098 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5099 and the realloc
5100 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5101 C<len>, and already meets the requirements for storing in C<SvPVX>).
5102
5103 =cut
5104 */
5105
5106 void
Perl_sv_usepvn_flags(pTHX_ SV * const sv,char * ptr,const STRLEN len,const U32 flags)5107 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5108 {
5109 STRLEN allocate;
5110
5111 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5112
5113 SV_CHECK_THINKFIRST_COW_DROP(sv);
5114 SvUPGRADE(sv, SVt_PV);
5115 if (!ptr) {
5116 (void)SvOK_off(sv);
5117 if (flags & SV_SMAGIC)
5118 SvSETMAGIC(sv);
5119 return;
5120 }
5121 if (SvPVX_const(sv))
5122 SvPV_free(sv);
5123
5124 #ifdef DEBUGGING
5125 if (flags & SV_HAS_TRAILING_NUL)
5126 assert(ptr[len] == '\0');
5127 #endif
5128
5129 allocate = (flags & SV_HAS_TRAILING_NUL)
5130 ? len + 1 :
5131 #ifdef Perl_safesysmalloc_size
5132 len + 1;
5133 #else
5134 PERL_STRLEN_ROUNDUP(len + 1);
5135 #endif
5136 if (flags & SV_HAS_TRAILING_NUL) {
5137 /* It's long enough - do nothing.
5138 Specifically Perl_newCONSTSUB is relying on this. */
5139 } else {
5140 #ifdef DEBUGGING
5141 /* Force a move to shake out bugs in callers. */
5142 char *new_ptr = (char*)safemalloc(allocate);
5143 Copy(ptr, new_ptr, len, char);
5144 PoisonFree(ptr,len,char);
5145 Safefree(ptr);
5146 ptr = new_ptr;
5147 #else
5148 ptr = (char*) saferealloc (ptr, allocate);
5149 #endif
5150 }
5151 #ifdef Perl_safesysmalloc_size
5152 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5153 #else
5154 SvLEN_set(sv, allocate);
5155 #endif
5156 SvCUR_set(sv, len);
5157 SvPV_set(sv, ptr);
5158 if (!(flags & SV_HAS_TRAILING_NUL)) {
5159 ptr[len] = '\0';
5160 }
5161 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5162 SvTAINT(sv);
5163 if (flags & SV_SMAGIC)
5164 SvSETMAGIC(sv);
5165 }
5166
5167
5168 static void
S_sv_uncow(pTHX_ SV * const sv,const U32 flags)5169 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5170 {
5171 assert(SvIsCOW(sv));
5172 {
5173 #ifdef PERL_ANY_COW
5174 const char * const pvx = SvPVX_const(sv);
5175 const STRLEN len = SvLEN(sv);
5176 const STRLEN cur = SvCUR(sv);
5177
5178 #ifdef DEBUGGING
5179 if (DEBUG_C_TEST) {
5180 PerlIO_printf(Perl_debug_log,
5181 "Copy on write: Force normal %ld\n",
5182 (long) flags);
5183 sv_dump(sv);
5184 }
5185 #endif
5186 SvIsCOW_off(sv);
5187 # ifdef PERL_COPY_ON_WRITE
5188 if (len) {
5189 /* Must do this first, since the CowREFCNT uses SvPVX and
5190 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5191 the only owner left of the buffer. */
5192 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5193 {
5194 U8 cowrefcnt = CowREFCNT(sv);
5195 if(cowrefcnt != 0) {
5196 cowrefcnt--;
5197 CowREFCNT(sv) = cowrefcnt;
5198 sv_buf_to_ro(sv);
5199 goto copy_over;
5200 }
5201 }
5202 /* Else we are the only owner of the buffer. */
5203 }
5204 else
5205 # endif
5206 {
5207 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5208 copy_over:
5209 SvPV_set(sv, NULL);
5210 SvCUR_set(sv, 0);
5211 SvLEN_set(sv, 0);
5212 if (flags & SV_COW_DROP_PV) {
5213 /* OK, so we don't need to copy our buffer. */
5214 SvPOK_off(sv);
5215 } else {
5216 SvGROW(sv, cur + 1);
5217 Move(pvx,SvPVX(sv),cur,char);
5218 SvCUR_set(sv, cur);
5219 *SvEND(sv) = '\0';
5220 }
5221 if (! len) {
5222 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5223 }
5224 #ifdef DEBUGGING
5225 if (DEBUG_C_TEST)
5226 sv_dump(sv);
5227 #endif
5228 }
5229 #else
5230 const char * const pvx = SvPVX_const(sv);
5231 const STRLEN len = SvCUR(sv);
5232 SvIsCOW_off(sv);
5233 SvPV_set(sv, NULL);
5234 SvLEN_set(sv, 0);
5235 if (flags & SV_COW_DROP_PV) {
5236 /* OK, so we don't need to copy our buffer. */
5237 SvPOK_off(sv);
5238 } else {
5239 SvGROW(sv, len + 1);
5240 Move(pvx,SvPVX(sv),len,char);
5241 *SvEND(sv) = '\0';
5242 }
5243 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5244 #endif
5245 }
5246 }
5247
5248
5249 /*
5250 =for apidoc sv_force_normal_flags
5251
5252 Undo various types of fakery on an SV, where fakery means
5253 "more than" a string: if the PV is a shared string, make
5254 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5255 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5256 we do the copy, and is also used locally; if this is a
5257 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5258 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5259 C<SvPOK_off> rather than making a copy. (Used where this
5260 scalar is about to be set to some other value.) In addition,
5261 the C<flags> parameter gets passed to C<sv_unref_flags()>
5262 when unreffing. C<sv_force_normal> calls this function
5263 with flags set to 0.
5264
5265 This function is expected to be used to signal to perl that this SV is
5266 about to be written to, and any extra book-keeping needs to be taken care
5267 of. Hence, it croaks on read-only values.
5268
5269 =cut
5270 */
5271
5272 void
Perl_sv_force_normal_flags(pTHX_ SV * const sv,const U32 flags)5273 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5274 {
5275 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5276
5277 if (SvREADONLY(sv))
5278 Perl_croak_no_modify();
5279 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5280 S_sv_uncow(aTHX_ sv, flags);
5281 if (SvROK(sv))
5282 sv_unref_flags(sv, flags);
5283 else if (SvFAKE(sv) && isGV_with_GP(sv))
5284 sv_unglob(sv, flags);
5285 else if (SvFAKE(sv) && isREGEXP(sv)) {
5286 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5287 to sv_unglob. We only need it here, so inline it. */
5288 const bool islv = SvTYPE(sv) == SVt_PVLV;
5289 const svtype new_type =
5290 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5291 SV *const temp = newSV_type(new_type);
5292 regexp *old_rx_body;
5293
5294 if (new_type == SVt_PVMG) {
5295 SvMAGIC_set(temp, SvMAGIC(sv));
5296 SvMAGIC_set(sv, NULL);
5297 SvSTASH_set(temp, SvSTASH(sv));
5298 SvSTASH_set(sv, NULL);
5299 }
5300 if (!islv)
5301 SvCUR_set(temp, SvCUR(sv));
5302 /* Remember that SvPVX is in the head, not the body. */
5303 assert(ReANY((REGEXP *)sv)->mother_re);
5304
5305 if (islv) {
5306 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5307 * whose xpvlenu_rx field points to the regex body */
5308 XPV *xpv = (XPV*)(SvANY(sv));
5309 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5310 xpv->xpv_len_u.xpvlenu_rx = NULL;
5311 }
5312 else
5313 old_rx_body = ReANY((REGEXP *)sv);
5314
5315 /* Their buffer is already owned by someone else. */
5316 if (flags & SV_COW_DROP_PV) {
5317 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5318 zeroed body. For SVt_PVLV, we zeroed it above (len field
5319 a union with xpvlenu_rx) */
5320 assert(!SvLEN(islv ? sv : temp));
5321 sv->sv_u.svu_pv = 0;
5322 }
5323 else {
5324 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5325 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5326 SvPOK_on(sv);
5327 }
5328
5329 /* Now swap the rest of the bodies. */
5330
5331 SvFAKE_off(sv);
5332 if (!islv) {
5333 SvFLAGS(sv) &= ~SVTYPEMASK;
5334 SvFLAGS(sv) |= new_type;
5335 SvANY(sv) = SvANY(temp);
5336 }
5337
5338 SvFLAGS(temp) &= ~(SVTYPEMASK);
5339 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5340 SvANY(temp) = old_rx_body;
5341
5342 SvREFCNT_dec_NN(temp);
5343 }
5344 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5345 }
5346
5347 /*
5348 =for apidoc sv_chop
5349
5350 Efficient removal of characters from the beginning of the string buffer.
5351 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5352 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5353 character of the adjusted string. Uses the C<OOK> hack. On return, only
5354 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5355
5356 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5357 refer to the same chunk of data.
5358
5359 The unfortunate similarity of this function's name to that of Perl's C<chop>
5360 operator is strictly coincidental. This function works from the left;
5361 C<chop> works from the right.
5362
5363 =cut
5364 */
5365
5366 void
Perl_sv_chop(pTHX_ SV * const sv,const char * const ptr)5367 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5368 {
5369 STRLEN delta;
5370 STRLEN old_delta;
5371 U8 *p;
5372 #ifdef DEBUGGING
5373 const U8 *evacp;
5374 STRLEN evacn;
5375 #endif
5376 STRLEN max_delta;
5377
5378 PERL_ARGS_ASSERT_SV_CHOP;
5379
5380 if (!ptr || !SvPOKp(sv))
5381 return;
5382 delta = ptr - SvPVX_const(sv);
5383 if (!delta) {
5384 /* Nothing to do. */
5385 return;
5386 }
5387 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5388 if (delta > max_delta)
5389 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5390 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5391 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5392 SV_CHECK_THINKFIRST(sv);
5393 SvPOK_only_UTF8(sv);
5394
5395 if (!SvOOK(sv)) {
5396 if (!SvLEN(sv)) { /* make copy of shared string */
5397 const char *pvx = SvPVX_const(sv);
5398 const STRLEN len = SvCUR(sv);
5399 SvGROW(sv, len + 1);
5400 Move(pvx,SvPVX(sv),len,char);
5401 *SvEND(sv) = '\0';
5402 }
5403 SvOOK_on(sv);
5404 old_delta = 0;
5405 } else {
5406 SvOOK_offset(sv, old_delta);
5407 }
5408 SvLEN_set(sv, SvLEN(sv) - delta);
5409 SvCUR_set(sv, SvCUR(sv) - delta);
5410 SvPV_set(sv, SvPVX(sv) + delta);
5411
5412 p = (U8 *)SvPVX_const(sv);
5413
5414 #ifdef DEBUGGING
5415 /* how many bytes were evacuated? we will fill them with sentinel
5416 bytes, except for the part holding the new offset of course. */
5417 evacn = delta;
5418 if (old_delta)
5419 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5420 assert(evacn);
5421 assert(evacn <= delta + old_delta);
5422 evacp = p - evacn;
5423 #endif
5424
5425 /* This sets 'delta' to the accumulated value of all deltas so far */
5426 delta += old_delta;
5427 assert(delta);
5428
5429 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5430 * the string; otherwise store a 0 byte there and store 'delta' just prior
5431 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5432 * portion of the chopped part of the string */
5433 if (delta < 0x100) {
5434 *--p = (U8) delta;
5435 } else {
5436 *--p = 0;
5437 p -= sizeof(STRLEN);
5438 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5439 }
5440
5441 #ifdef DEBUGGING
5442 /* Fill the preceding buffer with sentinals to verify that no-one is
5443 using it. */
5444 while (p > evacp) {
5445 --p;
5446 *p = (U8)PTR2UV(p);
5447 }
5448 #endif
5449 }
5450
5451 /*
5452 =for apidoc sv_catpvn
5453
5454 Concatenates the string onto the end of the string which is in the SV.
5455 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5456 status set, then the bytes appended should be valid UTF-8.
5457 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5458
5459 =for apidoc sv_catpvn_flags
5460
5461 Concatenates the string onto the end of the string which is in the SV. The
5462 C<len> indicates number of bytes to copy.
5463
5464 By default, the string appended is assumed to be valid UTF-8 if the SV has
5465 the UTF-8 status set, and a string of bytes otherwise. One can force the
5466 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5467 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5468 string appended will be upgraded to UTF-8 if necessary.
5469
5470 If C<flags> has the C<SV_SMAGIC> bit set, will
5471 C<mg_set> on C<dsv> afterwards if appropriate.
5472 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5473 in terms of this function.
5474
5475 =cut
5476 */
5477
5478 void
Perl_sv_catpvn_flags(pTHX_ SV * const dsv,const char * sstr,const STRLEN slen,const I32 flags)5479 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5480 {
5481 STRLEN dlen;
5482 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5483
5484 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5485 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5486
5487 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5488 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5489 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5490 dlen = SvCUR(dsv);
5491 }
5492 else SvGROW(dsv, dlen + slen + 3);
5493 if (sstr == dstr)
5494 sstr = SvPVX_const(dsv);
5495 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5496 SvCUR_set(dsv, SvCUR(dsv) + slen);
5497 }
5498 else {
5499 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5500 const char * const send = sstr + slen;
5501 U8 *d;
5502
5503 /* Something this code does not account for, which I think is
5504 impossible; it would require the same pv to be treated as
5505 bytes *and* utf8, which would indicate a bug elsewhere. */
5506 assert(sstr != dstr);
5507
5508 SvGROW(dsv, dlen + slen * 2 + 3);
5509 d = (U8 *)SvPVX(dsv) + dlen;
5510
5511 while (sstr < send) {
5512 append_utf8_from_native_byte(*sstr, &d);
5513 sstr++;
5514 }
5515 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5516 }
5517 *SvEND(dsv) = '\0';
5518 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5519 SvTAINT(dsv);
5520 if (flags & SV_SMAGIC)
5521 SvSETMAGIC(dsv);
5522 }
5523
5524 /*
5525 =for apidoc sv_catsv
5526
5527 Concatenates the string from SV C<ssv> onto the end of the string in SV
5528 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5529 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5530 and C<L</sv_catsv_nomg>>.
5531
5532 =for apidoc sv_catsv_flags
5533
5534 Concatenates the string from SV C<ssv> onto the end of the string in SV
5535 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5536 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5537 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5538 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5539 and C<sv_catsv_mg> are implemented in terms of this function.
5540
5541 =cut */
5542
5543 void
Perl_sv_catsv_flags(pTHX_ SV * const dsv,SV * const ssv,const I32 flags)5544 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5545 {
5546 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5547
5548 if (ssv) {
5549 STRLEN slen;
5550 const char *spv = SvPV_flags_const(ssv, slen, flags);
5551 if (flags & SV_GMAGIC)
5552 SvGETMAGIC(dsv);
5553 sv_catpvn_flags(dsv, spv, slen,
5554 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5555 if (flags & SV_SMAGIC)
5556 SvSETMAGIC(dsv);
5557 }
5558 }
5559
5560 /*
5561 =for apidoc sv_catpv
5562
5563 Concatenates the C<NUL>-terminated string onto the end of the string which is
5564 in the SV.
5565 If the SV has the UTF-8 status set, then the bytes appended should be
5566 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5567 C<L</sv_catpv_mg>>.
5568
5569 =cut */
5570
5571 void
Perl_sv_catpv(pTHX_ SV * const sv,const char * ptr)5572 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5573 {
5574 STRLEN len;
5575 STRLEN tlen;
5576 char *junk;
5577
5578 PERL_ARGS_ASSERT_SV_CATPV;
5579
5580 if (!ptr)
5581 return;
5582 junk = SvPV_force(sv, tlen);
5583 len = strlen(ptr);
5584 SvGROW(sv, tlen + len + 1);
5585 if (ptr == junk)
5586 ptr = SvPVX_const(sv);
5587 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5588 SvCUR_set(sv, SvCUR(sv) + len);
5589 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5590 SvTAINT(sv);
5591 }
5592
5593 /*
5594 =for apidoc sv_catpv_flags
5595
5596 Concatenates the C<NUL>-terminated string onto the end of the string which is
5597 in the SV.
5598 If the SV has the UTF-8 status set, then the bytes appended should
5599 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5600 on the modified SV if appropriate.
5601
5602 =cut
5603 */
5604
5605 void
Perl_sv_catpv_flags(pTHX_ SV * dstr,const char * sstr,const I32 flags)5606 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5607 {
5608 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5609 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5610 }
5611
5612 /*
5613 =for apidoc sv_catpv_mg
5614
5615 Like C<sv_catpv>, but also handles 'set' magic.
5616
5617 =cut
5618 */
5619
5620 void
Perl_sv_catpv_mg(pTHX_ SV * const sv,const char * const ptr)5621 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5622 {
5623 PERL_ARGS_ASSERT_SV_CATPV_MG;
5624
5625 sv_catpv(sv,ptr);
5626 SvSETMAGIC(sv);
5627 }
5628
5629 /*
5630 =for apidoc newSV
5631
5632 Creates a new SV. A non-zero C<len> parameter indicates the number of
5633 bytes of preallocated string space the SV should have. An extra byte for a
5634 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5635 space is allocated.) The reference count for the new SV is set to 1.
5636
5637 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5638 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5639 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5640 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5641 modules supporting older perls.
5642
5643 =cut
5644 */
5645
5646 SV *
Perl_newSV(pTHX_ const STRLEN len)5647 Perl_newSV(pTHX_ const STRLEN len)
5648 {
5649 SV *sv;
5650
5651 new_SV(sv);
5652 if (len) {
5653 sv_grow(sv, len + 1);
5654 }
5655 return sv;
5656 }
5657 /*
5658 =for apidoc sv_magicext
5659
5660 Adds magic to an SV, upgrading it if necessary. Applies the
5661 supplied C<vtable> and returns a pointer to the magic added.
5662
5663 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5664 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5665 one instance of the same C<how>.
5666
5667 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5668 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5669 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5670 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5671
5672 (This is now used as a subroutine by C<sv_magic>.)
5673
5674 =cut
5675 */
5676 MAGIC *
Perl_sv_magicext(pTHX_ SV * const sv,SV * const obj,const int how,const MGVTBL * const vtable,const char * const name,const I32 namlen)5677 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5678 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5679 {
5680 MAGIC* mg;
5681
5682 PERL_ARGS_ASSERT_SV_MAGICEXT;
5683
5684 SvUPGRADE(sv, SVt_PVMG);
5685 Newxz(mg, 1, MAGIC);
5686 mg->mg_moremagic = SvMAGIC(sv);
5687 SvMAGIC_set(sv, mg);
5688
5689 /* Sometimes a magic contains a reference loop, where the sv and
5690 object refer to each other. To prevent a reference loop that
5691 would prevent such objects being freed, we look for such loops
5692 and if we find one we avoid incrementing the object refcount.
5693
5694 Note we cannot do this to avoid self-tie loops as intervening RV must
5695 have its REFCNT incremented to keep it in existence.
5696
5697 */
5698 if (!obj || obj == sv ||
5699 how == PERL_MAGIC_arylen ||
5700 how == PERL_MAGIC_regdata ||
5701 how == PERL_MAGIC_regdatum ||
5702 how == PERL_MAGIC_symtab ||
5703 (SvTYPE(obj) == SVt_PVGV &&
5704 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5705 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5706 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5707 {
5708 mg->mg_obj = obj;
5709 }
5710 else {
5711 mg->mg_obj = SvREFCNT_inc_simple(obj);
5712 mg->mg_flags |= MGf_REFCOUNTED;
5713 }
5714
5715 /* Normal self-ties simply pass a null object, and instead of
5716 using mg_obj directly, use the SvTIED_obj macro to produce a
5717 new RV as needed. For glob "self-ties", we are tieing the PVIO
5718 with an RV obj pointing to the glob containing the PVIO. In
5719 this case, to avoid a reference loop, we need to weaken the
5720 reference.
5721 */
5722
5723 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5724 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5725 {
5726 sv_rvweaken(obj);
5727 }
5728
5729 mg->mg_type = how;
5730 mg->mg_len = namlen;
5731 if (name) {
5732 if (namlen > 0)
5733 mg->mg_ptr = savepvn(name, namlen);
5734 else if (namlen == HEf_SVKEY) {
5735 /* Yes, this is casting away const. This is only for the case of
5736 HEf_SVKEY. I think we need to document this aberation of the
5737 constness of the API, rather than making name non-const, as
5738 that change propagating outwards a long way. */
5739 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5740 } else
5741 mg->mg_ptr = (char *) name;
5742 }
5743 mg->mg_virtual = (MGVTBL *) vtable;
5744
5745 mg_magical(sv);
5746 return mg;
5747 }
5748
5749 MAGIC *
Perl_sv_magicext_mglob(pTHX_ SV * sv)5750 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5751 {
5752 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5753 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5754 /* This sv is only a delegate. //g magic must be attached to
5755 its target. */
5756 vivify_defelem(sv);
5757 sv = LvTARG(sv);
5758 }
5759 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5760 &PL_vtbl_mglob, 0, 0);
5761 }
5762
5763 /*
5764 =for apidoc sv_magic
5765
5766 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5767 necessary, then adds a new magic item of type C<how> to the head of the
5768 magic list.
5769
5770 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5771 handling of the C<name> and C<namlen> arguments.
5772
5773 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5774 to add more than one instance of the same C<how>.
5775
5776 =cut
5777 */
5778
5779 void
Perl_sv_magic(pTHX_ SV * const sv,SV * const obj,const int how,const char * const name,const I32 namlen)5780 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5781 const char *const name, const I32 namlen)
5782 {
5783 const MGVTBL *vtable;
5784 MAGIC* mg;
5785 unsigned int flags;
5786 unsigned int vtable_index;
5787
5788 PERL_ARGS_ASSERT_SV_MAGIC;
5789
5790 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5791 || ((flags = PL_magic_data[how]),
5792 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5793 > magic_vtable_max))
5794 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5795
5796 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5797 Useful for attaching extension internal data to perl vars.
5798 Note that multiple extensions may clash if magical scalars
5799 etc holding private data from one are passed to another. */
5800
5801 vtable = (vtable_index == magic_vtable_max)
5802 ? NULL : PL_magic_vtables + vtable_index;
5803
5804 if (SvREADONLY(sv)) {
5805 if (
5806 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5807 )
5808 {
5809 Perl_croak_no_modify();
5810 }
5811 }
5812 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5813 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5814 /* sv_magic() refuses to add a magic of the same 'how' as an
5815 existing one
5816 */
5817 if (how == PERL_MAGIC_taint)
5818 mg->mg_len |= 1;
5819 return;
5820 }
5821 }
5822
5823 /* Force pos to be stored as characters, not bytes. */
5824 if (SvMAGICAL(sv) && DO_UTF8(sv)
5825 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5826 && mg->mg_len != -1
5827 && mg->mg_flags & MGf_BYTES) {
5828 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5829 SV_CONST_RETURN);
5830 mg->mg_flags &= ~MGf_BYTES;
5831 }
5832
5833 /* Rest of work is done else where */
5834 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5835
5836 switch (how) {
5837 case PERL_MAGIC_taint:
5838 mg->mg_len = 1;
5839 break;
5840 case PERL_MAGIC_ext:
5841 case PERL_MAGIC_dbfile:
5842 SvRMAGICAL_on(sv);
5843 break;
5844 }
5845 }
5846
5847 static int
S_sv_unmagicext_flags(pTHX_ SV * const sv,const int type,MGVTBL * vtbl,const U32 flags)5848 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5849 {
5850 MAGIC* mg;
5851 MAGIC** mgp;
5852
5853 assert(flags <= 1);
5854
5855 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5856 return 0;
5857 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5858 for (mg = *mgp; mg; mg = *mgp) {
5859 const MGVTBL* const virt = mg->mg_virtual;
5860 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5861 *mgp = mg->mg_moremagic;
5862 if (virt && virt->svt_free)
5863 virt->svt_free(aTHX_ sv, mg);
5864 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5865 if (mg->mg_len > 0)
5866 Safefree(mg->mg_ptr);
5867 else if (mg->mg_len == HEf_SVKEY)
5868 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5869 else if (mg->mg_type == PERL_MAGIC_utf8)
5870 Safefree(mg->mg_ptr);
5871 }
5872 if (mg->mg_flags & MGf_REFCOUNTED)
5873 SvREFCNT_dec(mg->mg_obj);
5874 Safefree(mg);
5875 }
5876 else
5877 mgp = &mg->mg_moremagic;
5878 }
5879 if (SvMAGIC(sv)) {
5880 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5881 mg_magical(sv); /* else fix the flags now */
5882 }
5883 else
5884 SvMAGICAL_off(sv);
5885
5886 return 0;
5887 }
5888
5889 /*
5890 =for apidoc sv_unmagic
5891
5892 Removes all magic of type C<type> from an SV.
5893
5894 =cut
5895 */
5896
5897 int
Perl_sv_unmagic(pTHX_ SV * const sv,const int type)5898 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5899 {
5900 PERL_ARGS_ASSERT_SV_UNMAGIC;
5901 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5902 }
5903
5904 /*
5905 =for apidoc sv_unmagicext
5906
5907 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5908
5909 =cut
5910 */
5911
5912 int
Perl_sv_unmagicext(pTHX_ SV * const sv,const int type,MGVTBL * vtbl)5913 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5914 {
5915 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5916 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5917 }
5918
5919 /*
5920 =for apidoc sv_rvweaken
5921
5922 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5923 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5924 push a back-reference to this RV onto the array of backreferences
5925 associated with that magic. If the RV is magical, set magic will be
5926 called after the RV is cleared. Silently ignores C<undef> and warns
5927 on already-weak references.
5928
5929 =cut
5930 */
5931
5932 SV *
Perl_sv_rvweaken(pTHX_ SV * const sv)5933 Perl_sv_rvweaken(pTHX_ SV *const sv)
5934 {
5935 SV *tsv;
5936
5937 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5938
5939 if (!SvOK(sv)) /* let undefs pass */
5940 return sv;
5941 if (!SvROK(sv))
5942 Perl_croak(aTHX_ "Can't weaken a nonreference");
5943 else if (SvWEAKREF(sv)) {
5944 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5945 return sv;
5946 }
5947 else if (SvREADONLY(sv)) croak_no_modify();
5948 tsv = SvRV(sv);
5949 Perl_sv_add_backref(aTHX_ tsv, sv);
5950 SvWEAKREF_on(sv);
5951 SvREFCNT_dec_NN(tsv);
5952 return sv;
5953 }
5954
5955 /*
5956 =for apidoc sv_rvunweaken
5957
5958 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5959 the backreference to this RV from the array of backreferences
5960 associated with the target SV, increment the refcount of the target.
5961 Silently ignores C<undef> and warns on non-weak references.
5962
5963 =cut
5964 */
5965
5966 SV *
Perl_sv_rvunweaken(pTHX_ SV * const sv)5967 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5968 {
5969 SV *tsv;
5970
5971 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5972
5973 if (!SvOK(sv)) /* let undefs pass */
5974 return sv;
5975 if (!SvROK(sv))
5976 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5977 else if (!SvWEAKREF(sv)) {
5978 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5979 return sv;
5980 }
5981 else if (SvREADONLY(sv)) croak_no_modify();
5982
5983 tsv = SvRV(sv);
5984 SvWEAKREF_off(sv);
5985 SvROK_on(sv);
5986 SvREFCNT_inc_NN(tsv);
5987 Perl_sv_del_backref(aTHX_ tsv, sv);
5988 return sv;
5989 }
5990
5991 /*
5992 =for apidoc sv_get_backrefs
5993
5994 If C<sv> is the target of a weak reference then it returns the back
5995 references structure associated with the sv; otherwise return C<NULL>.
5996
5997 When returning a non-null result the type of the return is relevant. If it
5998 is an AV then the elements of the AV are the weak reference RVs which
5999 point at this item. If it is any other type then the item itself is the
6000 weak reference.
6001
6002 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6003 C<Perl_sv_kill_backrefs()>
6004
6005 =cut
6006 */
6007
6008 SV *
Perl_sv_get_backrefs(SV * const sv)6009 Perl_sv_get_backrefs(SV *const sv)
6010 {
6011 SV *backrefs= NULL;
6012
6013 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6014
6015 /* find slot to store array or singleton backref */
6016
6017 if (SvTYPE(sv) == SVt_PVHV) {
6018 if (SvOOK(sv)) {
6019 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6020 backrefs = (SV *)iter->xhv_backreferences;
6021 }
6022 } else if (SvMAGICAL(sv)) {
6023 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6024 if (mg)
6025 backrefs = mg->mg_obj;
6026 }
6027 return backrefs;
6028 }
6029
6030 /* Give tsv backref magic if it hasn't already got it, then push a
6031 * back-reference to sv onto the array associated with the backref magic.
6032 *
6033 * As an optimisation, if there's only one backref and it's not an AV,
6034 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6035 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6036 * active.)
6037 */
6038
6039 /* A discussion about the backreferences array and its refcount:
6040 *
6041 * The AV holding the backreferences is pointed to either as the mg_obj of
6042 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6043 * xhv_backreferences field. The array is created with a refcount
6044 * of 2. This means that if during global destruction the array gets
6045 * picked on before its parent to have its refcount decremented by the
6046 * random zapper, it won't actually be freed, meaning it's still there for
6047 * when its parent gets freed.
6048 *
6049 * When the parent SV is freed, the extra ref is killed by
6050 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6051 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6052 *
6053 * When a single backref SV is stored directly, it is not reference
6054 * counted.
6055 */
6056
6057 void
Perl_sv_add_backref(pTHX_ SV * const tsv,SV * const sv)6058 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6059 {
6060 SV **svp;
6061 AV *av = NULL;
6062 MAGIC *mg = NULL;
6063
6064 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6065
6066 /* find slot to store array or singleton backref */
6067
6068 if (SvTYPE(tsv) == SVt_PVHV) {
6069 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6070 } else {
6071 if (SvMAGICAL(tsv))
6072 mg = mg_find(tsv, PERL_MAGIC_backref);
6073 if (!mg)
6074 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6075 svp = &(mg->mg_obj);
6076 }
6077
6078 /* create or retrieve the array */
6079
6080 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6081 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6082 ) {
6083 /* create array */
6084 if (mg)
6085 mg->mg_flags |= MGf_REFCOUNTED;
6086 av = newAV();
6087 AvREAL_off(av);
6088 SvREFCNT_inc_simple_void_NN(av);
6089 /* av now has a refcnt of 2; see discussion above */
6090 av_extend(av, *svp ? 2 : 1);
6091 if (*svp) {
6092 /* move single existing backref to the array */
6093 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6094 }
6095 *svp = (SV*)av;
6096 }
6097 else {
6098 av = MUTABLE_AV(*svp);
6099 if (!av) {
6100 /* optimisation: store single backref directly in HvAUX or mg_obj */
6101 *svp = sv;
6102 return;
6103 }
6104 assert(SvTYPE(av) == SVt_PVAV);
6105 if (AvFILLp(av) >= AvMAX(av)) {
6106 av_extend(av, AvFILLp(av)+1);
6107 }
6108 }
6109 /* push new backref */
6110 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6111 }
6112
6113 /* delete a back-reference to ourselves from the backref magic associated
6114 * with the SV we point to.
6115 */
6116
6117 void
Perl_sv_del_backref(pTHX_ SV * const tsv,SV * const sv)6118 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6119 {
6120 SV **svp = NULL;
6121
6122 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6123
6124 if (SvTYPE(tsv) == SVt_PVHV) {
6125 if (SvOOK(tsv))
6126 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6127 }
6128 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6129 /* It's possible for the the last (strong) reference to tsv to have
6130 become freed *before* the last thing holding a weak reference.
6131 If both survive longer than the backreferences array, then when
6132 the referent's reference count drops to 0 and it is freed, it's
6133 not able to chase the backreferences, so they aren't NULLed.
6134
6135 For example, a CV holds a weak reference to its stash. If both the
6136 CV and the stash survive longer than the backreferences array,
6137 and the CV gets picked for the SvBREAK() treatment first,
6138 *and* it turns out that the stash is only being kept alive because
6139 of an our variable in the pad of the CV, then midway during CV
6140 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6141 It ends up pointing to the freed HV. Hence it's chased in here, and
6142 if this block wasn't here, it would hit the !svp panic just below.
6143
6144 I don't believe that "better" destruction ordering is going to help
6145 here - during global destruction there's always going to be the
6146 chance that something goes out of order. We've tried to make it
6147 foolproof before, and it only resulted in evolutionary pressure on
6148 fools. Which made us look foolish for our hubris. :-(
6149 */
6150 return;
6151 }
6152 else {
6153 MAGIC *const mg
6154 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6155 svp = mg ? &(mg->mg_obj) : NULL;
6156 }
6157
6158 if (!svp)
6159 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6160 if (!*svp) {
6161 /* It's possible that sv is being freed recursively part way through the
6162 freeing of tsv. If this happens, the backreferences array of tsv has
6163 already been freed, and so svp will be NULL. If this is the case,
6164 we should not panic. Instead, nothing needs doing, so return. */
6165 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6166 return;
6167 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6168 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6169 }
6170
6171 if (SvTYPE(*svp) == SVt_PVAV) {
6172 #ifdef DEBUGGING
6173 int count = 1;
6174 #endif
6175 AV * const av = (AV*)*svp;
6176 SSize_t fill;
6177 assert(!SvIS_FREED(av));
6178 fill = AvFILLp(av);
6179 assert(fill > -1);
6180 svp = AvARRAY(av);
6181 /* for an SV with N weak references to it, if all those
6182 * weak refs are deleted, then sv_del_backref will be called
6183 * N times and O(N^2) compares will be done within the backref
6184 * array. To ameliorate this potential slowness, we:
6185 * 1) make sure this code is as tight as possible;
6186 * 2) when looking for SV, look for it at both the head and tail of the
6187 * array first before searching the rest, since some create/destroy
6188 * patterns will cause the backrefs to be freed in order.
6189 */
6190 if (*svp == sv) {
6191 AvARRAY(av)++;
6192 AvMAX(av)--;
6193 }
6194 else {
6195 SV **p = &svp[fill];
6196 SV *const topsv = *p;
6197 if (topsv != sv) {
6198 #ifdef DEBUGGING
6199 count = 0;
6200 #endif
6201 while (--p > svp) {
6202 if (*p == sv) {
6203 /* We weren't the last entry.
6204 An unordered list has this property that you
6205 can take the last element off the end to fill
6206 the hole, and it's still an unordered list :-)
6207 */
6208 *p = topsv;
6209 #ifdef DEBUGGING
6210 count++;
6211 #else
6212 break; /* should only be one */
6213 #endif
6214 }
6215 }
6216 }
6217 }
6218 assert(count ==1);
6219 AvFILLp(av) = fill-1;
6220 }
6221 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6222 /* freed AV; skip */
6223 }
6224 else {
6225 /* optimisation: only a single backref, stored directly */
6226 if (*svp != sv)
6227 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6228 (void*)*svp, (void*)sv);
6229 *svp = NULL;
6230 }
6231
6232 }
6233
6234 void
Perl_sv_kill_backrefs(pTHX_ SV * const sv,AV * const av)6235 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6236 {
6237 SV **svp;
6238 SV **last;
6239 bool is_array;
6240
6241 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6242
6243 if (!av)
6244 return;
6245
6246 /* after multiple passes through Perl_sv_clean_all() for a thingy
6247 * that has badly leaked, the backref array may have gotten freed,
6248 * since we only protect it against 1 round of cleanup */
6249 if (SvIS_FREED(av)) {
6250 if (PL_in_clean_all) /* All is fair */
6251 return;
6252 Perl_croak(aTHX_
6253 "panic: magic_killbackrefs (freed backref AV/SV)");
6254 }
6255
6256
6257 is_array = (SvTYPE(av) == SVt_PVAV);
6258 if (is_array) {
6259 assert(!SvIS_FREED(av));
6260 svp = AvARRAY(av);
6261 if (svp)
6262 last = svp + AvFILLp(av);
6263 }
6264 else {
6265 /* optimisation: only a single backref, stored directly */
6266 svp = (SV**)&av;
6267 last = svp;
6268 }
6269
6270 if (svp) {
6271 while (svp <= last) {
6272 if (*svp) {
6273 SV *const referrer = *svp;
6274 if (SvWEAKREF(referrer)) {
6275 /* XXX Should we check that it hasn't changed? */
6276 assert(SvROK(referrer));
6277 SvRV_set(referrer, 0);
6278 SvOK_off(referrer);
6279 SvWEAKREF_off(referrer);
6280 SvSETMAGIC(referrer);
6281 } else if (SvTYPE(referrer) == SVt_PVGV ||
6282 SvTYPE(referrer) == SVt_PVLV) {
6283 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6284 /* You lookin' at me? */
6285 assert(GvSTASH(referrer));
6286 assert(GvSTASH(referrer) == (const HV *)sv);
6287 GvSTASH(referrer) = 0;
6288 } else if (SvTYPE(referrer) == SVt_PVCV ||
6289 SvTYPE(referrer) == SVt_PVFM) {
6290 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6291 /* You lookin' at me? */
6292 assert(CvSTASH(referrer));
6293 assert(CvSTASH(referrer) == (const HV *)sv);
6294 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6295 }
6296 else {
6297 assert(SvTYPE(sv) == SVt_PVGV);
6298 /* You lookin' at me? */
6299 assert(CvGV(referrer));
6300 assert(CvGV(referrer) == (const GV *)sv);
6301 anonymise_cv_maybe(MUTABLE_GV(sv),
6302 MUTABLE_CV(referrer));
6303 }
6304
6305 } else {
6306 Perl_croak(aTHX_
6307 "panic: magic_killbackrefs (flags=%" UVxf ")",
6308 (UV)SvFLAGS(referrer));
6309 }
6310
6311 if (is_array)
6312 *svp = NULL;
6313 }
6314 svp++;
6315 }
6316 }
6317 if (is_array) {
6318 AvFILLp(av) = -1;
6319 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6320 }
6321 return;
6322 }
6323
6324 /*
6325 =for apidoc sv_insert
6326
6327 Inserts and/or replaces a string at the specified offset/length within the SV.
6328 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6329 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6330 C<offset>. Handles get magic.
6331
6332 =for apidoc sv_insert_flags
6333
6334 Same as C<sv_insert>, but the extra C<flags> are passed to the
6335 C<SvPV_force_flags> that applies to C<bigstr>.
6336
6337 =cut
6338 */
6339
6340 void
Perl_sv_insert_flags(pTHX_ SV * const bigstr,const STRLEN offset,const STRLEN len,const char * little,const STRLEN littlelen,const U32 flags)6341 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6342 {
6343 char *big;
6344 char *mid;
6345 char *midend;
6346 char *bigend;
6347 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6348 STRLEN curlen;
6349
6350 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6351
6352 SvPV_force_flags(bigstr, curlen, flags);
6353 (void)SvPOK_only_UTF8(bigstr);
6354
6355 if (little >= SvPVX(bigstr) &&
6356 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6357 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6358 or little...little+littlelen might overlap offset...offset+len we make a copy
6359 */
6360 little = savepvn(little, littlelen);
6361 SAVEFREEPV(little);
6362 }
6363
6364 if (offset + len > curlen) {
6365 SvGROW(bigstr, offset+len+1);
6366 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6367 SvCUR_set(bigstr, offset+len);
6368 }
6369
6370 SvTAINT(bigstr);
6371 i = littlelen - len;
6372 if (i > 0) { /* string might grow */
6373 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6374 mid = big + offset + len;
6375 midend = bigend = big + SvCUR(bigstr);
6376 bigend += i;
6377 *bigend = '\0';
6378 while (midend > mid) /* shove everything down */
6379 *--bigend = *--midend;
6380 Move(little,big+offset,littlelen,char);
6381 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6382 SvSETMAGIC(bigstr);
6383 return;
6384 }
6385 else if (i == 0) {
6386 Move(little,SvPVX(bigstr)+offset,len,char);
6387 SvSETMAGIC(bigstr);
6388 return;
6389 }
6390
6391 big = SvPVX(bigstr);
6392 mid = big + offset;
6393 midend = mid + len;
6394 bigend = big + SvCUR(bigstr);
6395
6396 if (midend > bigend)
6397 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6398 midend, bigend);
6399
6400 if (mid - big > bigend - midend) { /* faster to shorten from end */
6401 if (littlelen) {
6402 Move(little, mid, littlelen,char);
6403 mid += littlelen;
6404 }
6405 i = bigend - midend;
6406 if (i > 0) {
6407 Move(midend, mid, i,char);
6408 mid += i;
6409 }
6410 *mid = '\0';
6411 SvCUR_set(bigstr, mid - big);
6412 }
6413 else if ((i = mid - big)) { /* faster from front */
6414 midend -= littlelen;
6415 mid = midend;
6416 Move(big, midend - i, i, char);
6417 sv_chop(bigstr,midend-i);
6418 if (littlelen)
6419 Move(little, mid, littlelen,char);
6420 }
6421 else if (littlelen) {
6422 midend -= littlelen;
6423 sv_chop(bigstr,midend);
6424 Move(little,midend,littlelen,char);
6425 }
6426 else {
6427 sv_chop(bigstr,midend);
6428 }
6429 SvSETMAGIC(bigstr);
6430 }
6431
6432 /*
6433 =for apidoc sv_replace
6434
6435 Make the first argument a copy of the second, then delete the original.
6436 The target SV physically takes over ownership of the body of the source SV
6437 and inherits its flags; however, the target keeps any magic it owns,
6438 and any magic in the source is discarded.
6439 Note that this is a rather specialist SV copying operation; most of the
6440 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6441
6442 =cut
6443 */
6444
6445 void
Perl_sv_replace(pTHX_ SV * const sv,SV * const nsv)6446 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6447 {
6448 const U32 refcnt = SvREFCNT(sv);
6449
6450 PERL_ARGS_ASSERT_SV_REPLACE;
6451
6452 SV_CHECK_THINKFIRST_COW_DROP(sv);
6453 if (SvREFCNT(nsv) != 1) {
6454 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6455 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6456 }
6457 if (SvMAGICAL(sv)) {
6458 if (SvMAGICAL(nsv))
6459 mg_free(nsv);
6460 else
6461 sv_upgrade(nsv, SVt_PVMG);
6462 SvMAGIC_set(nsv, SvMAGIC(sv));
6463 SvFLAGS(nsv) |= SvMAGICAL(sv);
6464 SvMAGICAL_off(sv);
6465 SvMAGIC_set(sv, NULL);
6466 }
6467 SvREFCNT(sv) = 0;
6468 sv_clear(sv);
6469 assert(!SvREFCNT(sv));
6470 #ifdef DEBUG_LEAKING_SCALARS
6471 sv->sv_flags = nsv->sv_flags;
6472 sv->sv_any = nsv->sv_any;
6473 sv->sv_refcnt = nsv->sv_refcnt;
6474 sv->sv_u = nsv->sv_u;
6475 #else
6476 StructCopy(nsv,sv,SV);
6477 #endif
6478 if(SvTYPE(sv) == SVt_IV) {
6479 SET_SVANY_FOR_BODYLESS_IV(sv);
6480 }
6481
6482
6483 SvREFCNT(sv) = refcnt;
6484 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6485 SvREFCNT(nsv) = 0;
6486 del_SV(nsv);
6487 }
6488
6489 /* We're about to free a GV which has a CV that refers back to us.
6490 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6491 * field) */
6492
6493 STATIC void
S_anonymise_cv_maybe(pTHX_ GV * gv,CV * cv)6494 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6495 {
6496 SV *gvname;
6497 GV *anongv;
6498
6499 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6500
6501 /* be assertive! */
6502 assert(SvREFCNT(gv) == 0);
6503 assert(isGV(gv) && isGV_with_GP(gv));
6504 assert(GvGP(gv));
6505 assert(!CvANON(cv));
6506 assert(CvGV(cv) == gv);
6507 assert(!CvNAMED(cv));
6508
6509 /* will the CV shortly be freed by gp_free() ? */
6510 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6511 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6512 return;
6513 }
6514
6515 /* if not, anonymise: */
6516 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6517 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6518 : newSVpvn_flags( "__ANON__", 8, 0 );
6519 sv_catpvs(gvname, "::__ANON__");
6520 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6521 SvREFCNT_dec_NN(gvname);
6522
6523 CvANON_on(cv);
6524 CvCVGV_RC_on(cv);
6525 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6526 }
6527
6528
6529 /*
6530 =for apidoc sv_clear
6531
6532 Clear an SV: call any destructors, free up any memory used by the body,
6533 and free the body itself. The SV's head is I<not> freed, although
6534 its type is set to all 1's so that it won't inadvertently be assumed
6535 to be live during global destruction etc.
6536 This function should only be called when C<REFCNT> is zero. Most of the time
6537 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6538 instead.
6539
6540 =cut
6541 */
6542
6543 void
Perl_sv_clear(pTHX_ SV * const orig_sv)6544 Perl_sv_clear(pTHX_ SV *const orig_sv)
6545 {
6546 dVAR;
6547 HV *stash;
6548 U32 type;
6549 const struct body_details *sv_type_details;
6550 SV* iter_sv = NULL;
6551 SV* next_sv = NULL;
6552 SV *sv = orig_sv;
6553 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6554 Not strictly necessary */
6555
6556 PERL_ARGS_ASSERT_SV_CLEAR;
6557
6558 /* within this loop, sv is the SV currently being freed, and
6559 * iter_sv is the most recent AV or whatever that's being iterated
6560 * over to provide more SVs */
6561
6562 while (sv) {
6563
6564 type = SvTYPE(sv);
6565
6566 assert(SvREFCNT(sv) == 0);
6567 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6568
6569 if (type <= SVt_IV) {
6570 /* See the comment in sv.h about the collusion between this
6571 * early return and the overloading of the NULL slots in the
6572 * size table. */
6573 if (SvROK(sv))
6574 goto free_rv;
6575 SvFLAGS(sv) &= SVf_BREAK;
6576 SvFLAGS(sv) |= SVTYPEMASK;
6577 goto free_head;
6578 }
6579
6580 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6581 for another purpose */
6582 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6583
6584 if (type >= SVt_PVMG) {
6585 if (SvOBJECT(sv)) {
6586 if (!curse(sv, 1)) goto get_next_sv;
6587 type = SvTYPE(sv); /* destructor may have changed it */
6588 }
6589 /* Free back-references before magic, in case the magic calls
6590 * Perl code that has weak references to sv. */
6591 if (type == SVt_PVHV) {
6592 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6593 if (SvMAGIC(sv))
6594 mg_free(sv);
6595 }
6596 else if (SvMAGIC(sv)) {
6597 /* Free back-references before other types of magic. */
6598 sv_unmagic(sv, PERL_MAGIC_backref);
6599 mg_free(sv);
6600 }
6601 SvMAGICAL_off(sv);
6602 }
6603 switch (type) {
6604 /* case SVt_INVLIST: */
6605 case SVt_PVIO:
6606 if (IoIFP(sv) &&
6607 IoIFP(sv) != PerlIO_stdin() &&
6608 IoIFP(sv) != PerlIO_stdout() &&
6609 IoIFP(sv) != PerlIO_stderr() &&
6610 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6611 {
6612 io_close(MUTABLE_IO(sv), NULL, FALSE,
6613 (IoTYPE(sv) == IoTYPE_WRONLY ||
6614 IoTYPE(sv) == IoTYPE_RDWR ||
6615 IoTYPE(sv) == IoTYPE_APPEND));
6616 }
6617 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6618 PerlDir_close(IoDIRP(sv));
6619 IoDIRP(sv) = (DIR*)NULL;
6620 Safefree(IoTOP_NAME(sv));
6621 Safefree(IoFMT_NAME(sv));
6622 Safefree(IoBOTTOM_NAME(sv));
6623 if ((const GV *)sv == PL_statgv)
6624 PL_statgv = NULL;
6625 goto freescalar;
6626 case SVt_REGEXP:
6627 /* FIXME for plugins */
6628 pregfree2((REGEXP*) sv);
6629 goto freescalar;
6630 case SVt_PVCV:
6631 case SVt_PVFM:
6632 cv_undef(MUTABLE_CV(sv));
6633 /* If we're in a stash, we don't own a reference to it.
6634 * However it does have a back reference to us, which needs to
6635 * be cleared. */
6636 if ((stash = CvSTASH(sv)))
6637 sv_del_backref(MUTABLE_SV(stash), sv);
6638 goto freescalar;
6639 case SVt_PVHV:
6640 if (PL_last_swash_hv == (const HV *)sv) {
6641 PL_last_swash_hv = NULL;
6642 }
6643 if (HvTOTALKEYS((HV*)sv) > 0) {
6644 const HEK *hek;
6645 /* this statement should match the one at the beginning of
6646 * hv_undef_flags() */
6647 if ( PL_phase != PERL_PHASE_DESTRUCT
6648 && (hek = HvNAME_HEK((HV*)sv)))
6649 {
6650 if (PL_stashcache) {
6651 DEBUG_o(Perl_deb(aTHX_
6652 "sv_clear clearing PL_stashcache for '%" HEKf
6653 "'\n",
6654 HEKfARG(hek)));
6655 (void)hv_deletehek(PL_stashcache,
6656 hek, G_DISCARD);
6657 }
6658 hv_name_set((HV*)sv, NULL, 0, 0);
6659 }
6660
6661 /* save old iter_sv in unused SvSTASH field */
6662 assert(!SvOBJECT(sv));
6663 SvSTASH(sv) = (HV*)iter_sv;
6664 iter_sv = sv;
6665
6666 /* save old hash_index in unused SvMAGIC field */
6667 assert(!SvMAGICAL(sv));
6668 assert(!SvMAGIC(sv));
6669 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6670 hash_index = 0;
6671
6672 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6673 goto get_next_sv; /* process this new sv */
6674 }
6675 /* free empty hash */
6676 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6677 assert(!HvARRAY((HV*)sv));
6678 break;
6679 case SVt_PVAV:
6680 {
6681 AV* av = MUTABLE_AV(sv);
6682 if (PL_comppad == av) {
6683 PL_comppad = NULL;
6684 PL_curpad = NULL;
6685 }
6686 if (AvREAL(av) && AvFILLp(av) > -1) {
6687 next_sv = AvARRAY(av)[AvFILLp(av)--];
6688 /* save old iter_sv in top-most slot of AV,
6689 * and pray that it doesn't get wiped in the meantime */
6690 AvARRAY(av)[AvMAX(av)] = iter_sv;
6691 iter_sv = sv;
6692 goto get_next_sv; /* process this new sv */
6693 }
6694 Safefree(AvALLOC(av));
6695 }
6696
6697 break;
6698 case SVt_PVLV:
6699 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6700 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6701 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6702 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6703 }
6704 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6705 SvREFCNT_dec(LvTARG(sv));
6706 if (isREGEXP(sv)) {
6707 /* SvLEN points to a regex body. Free the body, then
6708 * set SvLEN to whatever value was in the now-freed
6709 * regex body. The PVX buffer is shared by multiple re's
6710 * and only freed once, by the re whose len in non-null */
6711 STRLEN len = ReANY(sv)->xpv_len;
6712 pregfree2((REGEXP*) sv);
6713 SvLEN_set((sv), len);
6714 goto freescalar;
6715 }
6716 /* FALLTHROUGH */
6717 case SVt_PVGV:
6718 if (isGV_with_GP(sv)) {
6719 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6720 && HvENAME_get(stash))
6721 mro_method_changed_in(stash);
6722 gp_free(MUTABLE_GV(sv));
6723 if (GvNAME_HEK(sv))
6724 unshare_hek(GvNAME_HEK(sv));
6725 /* If we're in a stash, we don't own a reference to it.
6726 * However it does have a back reference to us, which
6727 * needs to be cleared. */
6728 if ((stash = GvSTASH(sv)))
6729 sv_del_backref(MUTABLE_SV(stash), sv);
6730 }
6731 /* FIXME. There are probably more unreferenced pointers to SVs
6732 * in the interpreter struct that we should check and tidy in
6733 * a similar fashion to this: */
6734 /* See also S_sv_unglob, which does the same thing. */
6735 if ((const GV *)sv == PL_last_in_gv)
6736 PL_last_in_gv = NULL;
6737 else if ((const GV *)sv == PL_statgv)
6738 PL_statgv = NULL;
6739 else if ((const GV *)sv == PL_stderrgv)
6740 PL_stderrgv = NULL;
6741 /* FALLTHROUGH */
6742 case SVt_PVMG:
6743 case SVt_PVNV:
6744 case SVt_PVIV:
6745 case SVt_INVLIST:
6746 case SVt_PV:
6747 freescalar:
6748 /* Don't bother with SvOOK_off(sv); as we're only going to
6749 * free it. */
6750 if (SvOOK(sv)) {
6751 STRLEN offset;
6752 SvOOK_offset(sv, offset);
6753 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6754 /* Don't even bother with turning off the OOK flag. */
6755 }
6756 if (SvROK(sv)) {
6757 free_rv:
6758 {
6759 SV * const target = SvRV(sv);
6760 if (SvWEAKREF(sv))
6761 sv_del_backref(target, sv);
6762 else
6763 next_sv = target;
6764 }
6765 }
6766 #ifdef PERL_ANY_COW
6767 else if (SvPVX_const(sv)
6768 && !(SvTYPE(sv) == SVt_PVIO
6769 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6770 {
6771 if (SvIsCOW(sv)) {
6772 #ifdef DEBUGGING
6773 if (DEBUG_C_TEST) {
6774 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6775 sv_dump(sv);
6776 }
6777 #endif
6778 if (SvLEN(sv)) {
6779 if (CowREFCNT(sv)) {
6780 sv_buf_to_rw(sv);
6781 CowREFCNT(sv)--;
6782 sv_buf_to_ro(sv);
6783 SvLEN_set(sv, 0);
6784 }
6785 } else {
6786 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6787 }
6788
6789 }
6790 if (SvLEN(sv)) {
6791 Safefree(SvPVX_mutable(sv));
6792 }
6793 }
6794 #else
6795 else if (SvPVX_const(sv) && SvLEN(sv)
6796 && !(SvTYPE(sv) == SVt_PVIO
6797 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6798 Safefree(SvPVX_mutable(sv));
6799 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6800 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6801 }
6802 #endif
6803 break;
6804 case SVt_NV:
6805 break;
6806 }
6807
6808 free_body:
6809
6810 SvFLAGS(sv) &= SVf_BREAK;
6811 SvFLAGS(sv) |= SVTYPEMASK;
6812
6813 sv_type_details = bodies_by_type + type;
6814 if (sv_type_details->arena) {
6815 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6816 &PL_body_roots[type]);
6817 }
6818 else if (sv_type_details->body_size) {
6819 safefree(SvANY(sv));
6820 }
6821
6822 free_head:
6823 /* caller is responsible for freeing the head of the original sv */
6824 if (sv != orig_sv && !SvREFCNT(sv))
6825 del_SV(sv);
6826
6827 /* grab and free next sv, if any */
6828 get_next_sv:
6829 while (1) {
6830 sv = NULL;
6831 if (next_sv) {
6832 sv = next_sv;
6833 next_sv = NULL;
6834 }
6835 else if (!iter_sv) {
6836 break;
6837 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6838 AV *const av = (AV*)iter_sv;
6839 if (AvFILLp(av) > -1) {
6840 sv = AvARRAY(av)[AvFILLp(av)--];
6841 }
6842 else { /* no more elements of current AV to free */
6843 sv = iter_sv;
6844 type = SvTYPE(sv);
6845 /* restore previous value, squirrelled away */
6846 iter_sv = AvARRAY(av)[AvMAX(av)];
6847 Safefree(AvALLOC(av));
6848 goto free_body;
6849 }
6850 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6851 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6852 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6853 /* no more elements of current HV to free */
6854 sv = iter_sv;
6855 type = SvTYPE(sv);
6856 /* Restore previous values of iter_sv and hash_index,
6857 * squirrelled away */
6858 assert(!SvOBJECT(sv));
6859 iter_sv = (SV*)SvSTASH(sv);
6860 assert(!SvMAGICAL(sv));
6861 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6862 #ifdef DEBUGGING
6863 /* perl -DA does not like rubbish in SvMAGIC. */
6864 SvMAGIC_set(sv, 0);
6865 #endif
6866
6867 /* free any remaining detritus from the hash struct */
6868 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6869 assert(!HvARRAY((HV*)sv));
6870 goto free_body;
6871 }
6872 }
6873
6874 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6875
6876 if (!sv)
6877 continue;
6878 if (!SvREFCNT(sv)) {
6879 sv_free(sv);
6880 continue;
6881 }
6882 if (--(SvREFCNT(sv)))
6883 continue;
6884 #ifdef DEBUGGING
6885 if (SvTEMP(sv)) {
6886 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6887 "Attempt to free temp prematurely: SV 0x%" UVxf
6888 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6889 continue;
6890 }
6891 #endif
6892 if (SvIMMORTAL(sv)) {
6893 /* make sure SvREFCNT(sv)==0 happens very seldom */
6894 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6895 continue;
6896 }
6897 break;
6898 } /* while 1 */
6899
6900 } /* while sv */
6901 }
6902
6903 /* This routine curses the sv itself, not the object referenced by sv. So
6904 sv does not have to be ROK. */
6905
6906 static bool
S_curse(pTHX_ SV * const sv,const bool check_refcnt)6907 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6908 PERL_ARGS_ASSERT_CURSE;
6909 assert(SvOBJECT(sv));
6910
6911 if (PL_defstash && /* Still have a symbol table? */
6912 SvDESTROYABLE(sv))
6913 {
6914 dSP;
6915 HV* stash;
6916 do {
6917 stash = SvSTASH(sv);
6918 assert(SvTYPE(stash) == SVt_PVHV);
6919 if (HvNAME(stash)) {
6920 CV* destructor = NULL;
6921 struct mro_meta *meta;
6922
6923 assert (SvOOK(stash));
6924
6925 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6926 HvNAME(stash)) );
6927
6928 /* don't make this an initialization above the assert, since it needs
6929 an AUX structure */
6930 meta = HvMROMETA(stash);
6931 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6932 destructor = meta->destroy;
6933 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6934 (void *)destructor, HvNAME(stash)) );
6935 }
6936 else {
6937 bool autoload = FALSE;
6938 GV *gv =
6939 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6940 if (gv)
6941 destructor = GvCV(gv);
6942 if (!destructor) {
6943 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6944 GV_AUTOLOAD_ISMETHOD);
6945 if (gv)
6946 destructor = GvCV(gv);
6947 if (destructor)
6948 autoload = TRUE;
6949 }
6950 /* we don't cache AUTOLOAD for DESTROY, since this code
6951 would then need to set $__PACKAGE__::AUTOLOAD, or the
6952 equivalent for XS AUTOLOADs */
6953 if (!autoload) {
6954 meta->destroy_gen = PL_sub_generation;
6955 meta->destroy = destructor;
6956
6957 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6958 (void *)destructor, HvNAME(stash)) );
6959 }
6960 else {
6961 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6962 HvNAME(stash)) );
6963 }
6964 }
6965 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6966 if (destructor
6967 /* A constant subroutine can have no side effects, so
6968 don't bother calling it. */
6969 && !CvCONST(destructor)
6970 /* Don't bother calling an empty destructor or one that
6971 returns immediately. */
6972 && (CvISXSUB(destructor)
6973 || (CvSTART(destructor)
6974 && (CvSTART(destructor)->op_next->op_type
6975 != OP_LEAVESUB)
6976 && (CvSTART(destructor)->op_next->op_type
6977 != OP_PUSHMARK
6978 || CvSTART(destructor)->op_next->op_next->op_type
6979 != OP_RETURN
6980 )
6981 ))
6982 )
6983 {
6984 SV* const tmpref = newRV(sv);
6985 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6986 ENTER;
6987 PUSHSTACKi(PERLSI_DESTROY);
6988 EXTEND(SP, 2);
6989 PUSHMARK(SP);
6990 PUSHs(tmpref);
6991 PUTBACK;
6992 call_sv(MUTABLE_SV(destructor),
6993 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6994 POPSTACK;
6995 SPAGAIN;
6996 LEAVE;
6997 if(SvREFCNT(tmpref) < 2) {
6998 /* tmpref is not kept alive! */
6999 SvREFCNT(sv)--;
7000 SvRV_set(tmpref, NULL);
7001 SvROK_off(tmpref);
7002 }
7003 SvREFCNT_dec_NN(tmpref);
7004 }
7005 }
7006 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7007
7008
7009 if (check_refcnt && SvREFCNT(sv)) {
7010 if (PL_in_clean_objs)
7011 Perl_croak(aTHX_
7012 "DESTROY created new reference to dead object '%" HEKf "'",
7013 HEKfARG(HvNAME_HEK(stash)));
7014 /* DESTROY gave object new lease on life */
7015 return FALSE;
7016 }
7017 }
7018
7019 if (SvOBJECT(sv)) {
7020 HV * const stash = SvSTASH(sv);
7021 /* Curse before freeing the stash, as freeing the stash could cause
7022 a recursive call into S_curse. */
7023 SvOBJECT_off(sv); /* Curse the object. */
7024 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7025 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7026 }
7027 return TRUE;
7028 }
7029
7030 /*
7031 =for apidoc sv_newref
7032
7033 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7034 instead.
7035
7036 =cut
7037 */
7038
7039 SV *
Perl_sv_newref(pTHX_ SV * const sv)7040 Perl_sv_newref(pTHX_ SV *const sv)
7041 {
7042 PERL_UNUSED_CONTEXT;
7043 if (sv)
7044 (SvREFCNT(sv))++;
7045 return sv;
7046 }
7047
7048 /*
7049 =for apidoc sv_free
7050
7051 Decrement an SV's reference count, and if it drops to zero, call
7052 C<sv_clear> to invoke destructors and free up any memory used by
7053 the body; finally, deallocating the SV's head itself.
7054 Normally called via a wrapper macro C<SvREFCNT_dec>.
7055
7056 =cut
7057 */
7058
7059 void
Perl_sv_free(pTHX_ SV * const sv)7060 Perl_sv_free(pTHX_ SV *const sv)
7061 {
7062 SvREFCNT_dec(sv);
7063 }
7064
7065
7066 /* Private helper function for SvREFCNT_dec().
7067 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7068
7069 void
Perl_sv_free2(pTHX_ SV * const sv,const U32 rc)7070 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7071 {
7072 dVAR;
7073
7074 PERL_ARGS_ASSERT_SV_FREE2;
7075
7076 if (LIKELY( rc == 1 )) {
7077 /* normal case */
7078 SvREFCNT(sv) = 0;
7079
7080 #ifdef DEBUGGING
7081 if (SvTEMP(sv)) {
7082 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7083 "Attempt to free temp prematurely: SV 0x%" UVxf
7084 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7085 return;
7086 }
7087 #endif
7088 if (SvIMMORTAL(sv)) {
7089 /* make sure SvREFCNT(sv)==0 happens very seldom */
7090 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7091 return;
7092 }
7093 sv_clear(sv);
7094 if (! SvREFCNT(sv)) /* may have have been resurrected */
7095 del_SV(sv);
7096 return;
7097 }
7098
7099 /* handle exceptional cases */
7100
7101 assert(rc == 0);
7102
7103 if (SvFLAGS(sv) & SVf_BREAK)
7104 /* this SV's refcnt has been artificially decremented to
7105 * trigger cleanup */
7106 return;
7107 if (PL_in_clean_all) /* All is fair */
7108 return;
7109 if (SvIMMORTAL(sv)) {
7110 /* make sure SvREFCNT(sv)==0 happens very seldom */
7111 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7112 return;
7113 }
7114 if (ckWARN_d(WARN_INTERNAL)) {
7115 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7116 Perl_dump_sv_child(aTHX_ sv);
7117 #else
7118 #ifdef DEBUG_LEAKING_SCALARS
7119 sv_dump(sv);
7120 #endif
7121 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7122 if (PL_warnhook == PERL_WARNHOOK_FATAL
7123 || ckDEAD(packWARN(WARN_INTERNAL))) {
7124 /* Don't let Perl_warner cause us to escape our fate: */
7125 abort();
7126 }
7127 #endif
7128 /* This may not return: */
7129 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7130 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7131 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7132 #endif
7133 }
7134 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7135 abort();
7136 #endif
7137
7138 }
7139
7140
7141 /*
7142 =for apidoc sv_len
7143
7144 Returns the length of the string in the SV. Handles magic and type
7145 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7146 gives raw access to the C<xpv_cur> slot.
7147
7148 =cut
7149 */
7150
7151 STRLEN
Perl_sv_len(pTHX_ SV * const sv)7152 Perl_sv_len(pTHX_ SV *const sv)
7153 {
7154 STRLEN len;
7155
7156 if (!sv)
7157 return 0;
7158
7159 (void)SvPV_const(sv, len);
7160 return len;
7161 }
7162
7163 /*
7164 =for apidoc sv_len_utf8
7165
7166 Returns the number of characters in the string in an SV, counting wide
7167 UTF-8 bytes as a single character. Handles magic and type coercion.
7168
7169 =cut
7170 */
7171
7172 /*
7173 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7174 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7175 * (Note that the mg_len is not the length of the mg_ptr field.
7176 * This allows the cache to store the character length of the string without
7177 * needing to malloc() extra storage to attach to the mg_ptr.)
7178 *
7179 */
7180
7181 STRLEN
Perl_sv_len_utf8(pTHX_ SV * const sv)7182 Perl_sv_len_utf8(pTHX_ SV *const sv)
7183 {
7184 if (!sv)
7185 return 0;
7186
7187 SvGETMAGIC(sv);
7188 return sv_len_utf8_nomg(sv);
7189 }
7190
7191 STRLEN
Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)7192 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7193 {
7194 STRLEN len;
7195 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7196
7197 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7198
7199 if (PL_utf8cache && SvUTF8(sv)) {
7200 STRLEN ulen;
7201 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7202
7203 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7204 if (mg->mg_len != -1)
7205 ulen = mg->mg_len;
7206 else {
7207 /* We can use the offset cache for a headstart.
7208 The longer value is stored in the first pair. */
7209 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7210
7211 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7212 s + len);
7213 }
7214
7215 if (PL_utf8cache < 0) {
7216 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7217 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7218 }
7219 }
7220 else {
7221 ulen = Perl_utf8_length(aTHX_ s, s + len);
7222 utf8_mg_len_cache_update(sv, &mg, ulen);
7223 }
7224 return ulen;
7225 }
7226 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7227 }
7228
7229 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7230 offset. */
7231 static STRLEN
S_sv_pos_u2b_forwards(const U8 * const start,const U8 * const send,STRLEN * const uoffset_p,bool * const at_end)7232 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7233 STRLEN *const uoffset_p, bool *const at_end)
7234 {
7235 const U8 *s = start;
7236 STRLEN uoffset = *uoffset_p;
7237
7238 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7239
7240 while (s < send && uoffset) {
7241 --uoffset;
7242 s += UTF8SKIP(s);
7243 }
7244 if (s == send) {
7245 *at_end = TRUE;
7246 }
7247 else if (s > send) {
7248 *at_end = TRUE;
7249 /* This is the existing behaviour. Possibly it should be a croak, as
7250 it's actually a bounds error */
7251 s = send;
7252 }
7253 *uoffset_p -= uoffset;
7254 return s - start;
7255 }
7256
7257 /* Given the length of the string in both bytes and UTF-8 characters, decide
7258 whether to walk forwards or backwards to find the byte corresponding to
7259 the passed in UTF-8 offset. */
7260 static STRLEN
S_sv_pos_u2b_midway(const U8 * const start,const U8 * send,STRLEN uoffset,const STRLEN uend)7261 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7262 STRLEN uoffset, const STRLEN uend)
7263 {
7264 STRLEN backw = uend - uoffset;
7265
7266 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7267
7268 if (uoffset < 2 * backw) {
7269 /* The assumption is that going forwards is twice the speed of going
7270 forward (that's where the 2 * backw comes from).
7271 (The real figure of course depends on the UTF-8 data.) */
7272 const U8 *s = start;
7273
7274 while (s < send && uoffset--)
7275 s += UTF8SKIP(s);
7276 assert (s <= send);
7277 if (s > send)
7278 s = send;
7279 return s - start;
7280 }
7281
7282 while (backw--) {
7283 send--;
7284 while (UTF8_IS_CONTINUATION(*send))
7285 send--;
7286 }
7287 return send - start;
7288 }
7289
7290 /* For the string representation of the given scalar, find the byte
7291 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7292 give another position in the string, *before* the sought offset, which
7293 (which is always true, as 0, 0 is a valid pair of positions), which should
7294 help reduce the amount of linear searching.
7295 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7296 will be used to reduce the amount of linear searching. The cache will be
7297 created if necessary, and the found value offered to it for update. */
7298 static STRLEN
S_sv_pos_u2b_cached(pTHX_ SV * const sv,MAGIC ** const mgp,const U8 * const start,const U8 * const send,STRLEN uoffset,STRLEN uoffset0,STRLEN boffset0)7299 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7300 const U8 *const send, STRLEN uoffset,
7301 STRLEN uoffset0, STRLEN boffset0)
7302 {
7303 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7304 bool found = FALSE;
7305 bool at_end = FALSE;
7306
7307 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7308
7309 assert (uoffset >= uoffset0);
7310
7311 if (!uoffset)
7312 return 0;
7313
7314 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7315 && PL_utf8cache
7316 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7317 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7318 if ((*mgp)->mg_ptr) {
7319 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7320 if (cache[0] == uoffset) {
7321 /* An exact match. */
7322 return cache[1];
7323 }
7324 if (cache[2] == uoffset) {
7325 /* An exact match. */
7326 return cache[3];
7327 }
7328
7329 if (cache[0] < uoffset) {
7330 /* The cache already knows part of the way. */
7331 if (cache[0] > uoffset0) {
7332 /* The cache knows more than the passed in pair */
7333 uoffset0 = cache[0];
7334 boffset0 = cache[1];
7335 }
7336 if ((*mgp)->mg_len != -1) {
7337 /* And we know the end too. */
7338 boffset = boffset0
7339 + sv_pos_u2b_midway(start + boffset0, send,
7340 uoffset - uoffset0,
7341 (*mgp)->mg_len - uoffset0);
7342 } else {
7343 uoffset -= uoffset0;
7344 boffset = boffset0
7345 + sv_pos_u2b_forwards(start + boffset0,
7346 send, &uoffset, &at_end);
7347 uoffset += uoffset0;
7348 }
7349 }
7350 else if (cache[2] < uoffset) {
7351 /* We're between the two cache entries. */
7352 if (cache[2] > uoffset0) {
7353 /* and the cache knows more than the passed in pair */
7354 uoffset0 = cache[2];
7355 boffset0 = cache[3];
7356 }
7357
7358 boffset = boffset0
7359 + sv_pos_u2b_midway(start + boffset0,
7360 start + cache[1],
7361 uoffset - uoffset0,
7362 cache[0] - uoffset0);
7363 } else {
7364 boffset = boffset0
7365 + sv_pos_u2b_midway(start + boffset0,
7366 start + cache[3],
7367 uoffset - uoffset0,
7368 cache[2] - uoffset0);
7369 }
7370 found = TRUE;
7371 }
7372 else if ((*mgp)->mg_len != -1) {
7373 /* If we can take advantage of a passed in offset, do so. */
7374 /* In fact, offset0 is either 0, or less than offset, so don't
7375 need to worry about the other possibility. */
7376 boffset = boffset0
7377 + sv_pos_u2b_midway(start + boffset0, send,
7378 uoffset - uoffset0,
7379 (*mgp)->mg_len - uoffset0);
7380 found = TRUE;
7381 }
7382 }
7383
7384 if (!found || PL_utf8cache < 0) {
7385 STRLEN real_boffset;
7386 uoffset -= uoffset0;
7387 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7388 send, &uoffset, &at_end);
7389 uoffset += uoffset0;
7390
7391 if (found && PL_utf8cache < 0)
7392 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7393 real_boffset, sv);
7394 boffset = real_boffset;
7395 }
7396
7397 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7398 if (at_end)
7399 utf8_mg_len_cache_update(sv, mgp, uoffset);
7400 else
7401 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7402 }
7403 return boffset;
7404 }
7405
7406
7407 /*
7408 =for apidoc sv_pos_u2b_flags
7409
7410 Converts the offset from a count of UTF-8 chars from
7411 the start of the string, to a count of the equivalent number of bytes; if
7412 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7413 C<offset>, rather than from the start
7414 of the string. Handles type coercion.
7415 C<flags> is passed to C<SvPV_flags>, and usually should be
7416 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7417
7418 =cut
7419 */
7420
7421 /*
7422 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7423 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7424 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7425 *
7426 */
7427
7428 STRLEN
Perl_sv_pos_u2b_flags(pTHX_ SV * const sv,STRLEN uoffset,STRLEN * const lenp,U32 flags)7429 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7430 U32 flags)
7431 {
7432 const U8 *start;
7433 STRLEN len;
7434 STRLEN boffset;
7435
7436 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7437
7438 start = (U8*)SvPV_flags(sv, len, flags);
7439 if (len) {
7440 const U8 * const send = start + len;
7441 MAGIC *mg = NULL;
7442 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7443
7444 if (lenp
7445 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7446 is 0, and *lenp is already set to that. */) {
7447 /* Convert the relative offset to absolute. */
7448 const STRLEN uoffset2 = uoffset + *lenp;
7449 const STRLEN boffset2
7450 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7451 uoffset, boffset) - boffset;
7452
7453 *lenp = boffset2;
7454 }
7455 } else {
7456 if (lenp)
7457 *lenp = 0;
7458 boffset = 0;
7459 }
7460
7461 return boffset;
7462 }
7463
7464 /*
7465 =for apidoc sv_pos_u2b
7466
7467 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7468 the start of the string, to a count of the equivalent number of bytes; if
7469 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7470 the offset, rather than from the start of the string. Handles magic and
7471 type coercion.
7472
7473 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7474 than 2Gb.
7475
7476 =cut
7477 */
7478
7479 /*
7480 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7481 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7482 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7483 *
7484 */
7485
7486 /* This function is subject to size and sign problems */
7487
7488 void
Perl_sv_pos_u2b(pTHX_ SV * const sv,I32 * const offsetp,I32 * const lenp)7489 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7490 {
7491 PERL_ARGS_ASSERT_SV_POS_U2B;
7492
7493 if (lenp) {
7494 STRLEN ulen = (STRLEN)*lenp;
7495 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7496 SV_GMAGIC|SV_CONST_RETURN);
7497 *lenp = (I32)ulen;
7498 } else {
7499 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7500 SV_GMAGIC|SV_CONST_RETURN);
7501 }
7502 }
7503
7504 static void
S_utf8_mg_len_cache_update(pTHX_ SV * const sv,MAGIC ** const mgp,const STRLEN ulen)7505 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7506 const STRLEN ulen)
7507 {
7508 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7509 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7510 return;
7511
7512 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7513 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7514 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7515 }
7516 assert(*mgp);
7517
7518 (*mgp)->mg_len = ulen;
7519 }
7520
7521 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7522 byte length pairing. The (byte) length of the total SV is passed in too,
7523 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7524 may not have updated SvCUR, so we can't rely on reading it directly.
7525
7526 The proffered utf8/byte length pairing isn't used if the cache already has
7527 two pairs, and swapping either for the proffered pair would increase the
7528 RMS of the intervals between known byte offsets.
7529
7530 The cache itself consists of 4 STRLEN values
7531 0: larger UTF-8 offset
7532 1: corresponding byte offset
7533 2: smaller UTF-8 offset
7534 3: corresponding byte offset
7535
7536 Unused cache pairs have the value 0, 0.
7537 Keeping the cache "backwards" means that the invariant of
7538 cache[0] >= cache[2] is maintained even with empty slots, which means that
7539 the code that uses it doesn't need to worry if only 1 entry has actually
7540 been set to non-zero. It also makes the "position beyond the end of the
7541 cache" logic much simpler, as the first slot is always the one to start
7542 from.
7543 */
7544 static void
S_utf8_mg_pos_cache_update(pTHX_ SV * const sv,MAGIC ** const mgp,const STRLEN byte,const STRLEN utf8,const STRLEN blen)7545 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7546 const STRLEN utf8, const STRLEN blen)
7547 {
7548 STRLEN *cache;
7549
7550 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7551
7552 if (SvREADONLY(sv))
7553 return;
7554
7555 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7556 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7557 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7558 0);
7559 (*mgp)->mg_len = -1;
7560 }
7561 assert(*mgp);
7562
7563 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7564 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7565 (*mgp)->mg_ptr = (char *) cache;
7566 }
7567 assert(cache);
7568
7569 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7570 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7571 a pointer. Note that we no longer cache utf8 offsets on refer-
7572 ences, but this check is still a good idea, for robustness. */
7573 const U8 *start = (const U8 *) SvPVX_const(sv);
7574 const STRLEN realutf8 = utf8_length(start, start + byte);
7575
7576 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7577 sv);
7578 }
7579
7580 /* Cache is held with the later position first, to simplify the code
7581 that deals with unbounded ends. */
7582
7583 ASSERT_UTF8_CACHE(cache);
7584 if (cache[1] == 0) {
7585 /* Cache is totally empty */
7586 cache[0] = utf8;
7587 cache[1] = byte;
7588 } else if (cache[3] == 0) {
7589 if (byte > cache[1]) {
7590 /* New one is larger, so goes first. */
7591 cache[2] = cache[0];
7592 cache[3] = cache[1];
7593 cache[0] = utf8;
7594 cache[1] = byte;
7595 } else {
7596 cache[2] = utf8;
7597 cache[3] = byte;
7598 }
7599 } else {
7600 /* float casts necessary? XXX */
7601 #define THREEWAY_SQUARE(a,b,c,d) \
7602 ((float)((d) - (c))) * ((float)((d) - (c))) \
7603 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7604 + ((float)((b) - (a))) * ((float)((b) - (a)))
7605
7606 /* Cache has 2 slots in use, and we know three potential pairs.
7607 Keep the two that give the lowest RMS distance. Do the
7608 calculation in bytes simply because we always know the byte
7609 length. squareroot has the same ordering as the positive value,
7610 so don't bother with the actual square root. */
7611 if (byte > cache[1]) {
7612 /* New position is after the existing pair of pairs. */
7613 const float keep_earlier
7614 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7615 const float keep_later
7616 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7617
7618 if (keep_later < keep_earlier) {
7619 cache[2] = cache[0];
7620 cache[3] = cache[1];
7621 }
7622 cache[0] = utf8;
7623 cache[1] = byte;
7624 }
7625 else {
7626 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7627 float b, c, keep_earlier;
7628 if (byte > cache[3]) {
7629 /* New position is between the existing pair of pairs. */
7630 b = (float)cache[3];
7631 c = (float)byte;
7632 } else {
7633 /* New position is before the existing pair of pairs. */
7634 b = (float)byte;
7635 c = (float)cache[3];
7636 }
7637 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7638 if (byte > cache[3]) {
7639 if (keep_later < keep_earlier) {
7640 cache[2] = utf8;
7641 cache[3] = byte;
7642 }
7643 else {
7644 cache[0] = utf8;
7645 cache[1] = byte;
7646 }
7647 }
7648 else {
7649 if (! (keep_later < keep_earlier)) {
7650 cache[0] = cache[2];
7651 cache[1] = cache[3];
7652 }
7653 cache[2] = utf8;
7654 cache[3] = byte;
7655 }
7656 }
7657 }
7658 ASSERT_UTF8_CACHE(cache);
7659 }
7660
7661 /* We already know all of the way, now we may be able to walk back. The same
7662 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7663 backward is half the speed of walking forward. */
7664 static STRLEN
S_sv_pos_b2u_midway(pTHX_ const U8 * const s,const U8 * const target,const U8 * end,STRLEN endu)7665 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7666 const U8 *end, STRLEN endu)
7667 {
7668 const STRLEN forw = target - s;
7669 STRLEN backw = end - target;
7670
7671 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7672
7673 if (forw < 2 * backw) {
7674 return utf8_length(s, target);
7675 }
7676
7677 while (end > target) {
7678 end--;
7679 while (UTF8_IS_CONTINUATION(*end)) {
7680 end--;
7681 }
7682 endu--;
7683 }
7684 return endu;
7685 }
7686
7687 /*
7688 =for apidoc sv_pos_b2u_flags
7689
7690 Converts C<offset> from a count of bytes from the start of the string, to
7691 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7692 C<flags> is passed to C<SvPV_flags>, and usually should be
7693 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7694
7695 =cut
7696 */
7697
7698 /*
7699 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7700 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7701 * and byte offsets.
7702 *
7703 */
7704 STRLEN
Perl_sv_pos_b2u_flags(pTHX_ SV * const sv,STRLEN const offset,U32 flags)7705 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7706 {
7707 const U8* s;
7708 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7709 STRLEN blen;
7710 MAGIC* mg = NULL;
7711 const U8* send;
7712 bool found = FALSE;
7713
7714 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7715
7716 s = (const U8*)SvPV_flags(sv, blen, flags);
7717
7718 if (blen < offset)
7719 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7720 ", byte=%" UVuf, (UV)blen, (UV)offset);
7721
7722 send = s + offset;
7723
7724 if (!SvREADONLY(sv)
7725 && PL_utf8cache
7726 && SvTYPE(sv) >= SVt_PVMG
7727 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7728 {
7729 if (mg->mg_ptr) {
7730 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7731 if (cache[1] == offset) {
7732 /* An exact match. */
7733 return cache[0];
7734 }
7735 if (cache[3] == offset) {
7736 /* An exact match. */
7737 return cache[2];
7738 }
7739
7740 if (cache[1] < offset) {
7741 /* We already know part of the way. */
7742 if (mg->mg_len != -1) {
7743 /* Actually, we know the end too. */
7744 len = cache[0]
7745 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7746 s + blen, mg->mg_len - cache[0]);
7747 } else {
7748 len = cache[0] + utf8_length(s + cache[1], send);
7749 }
7750 }
7751 else if (cache[3] < offset) {
7752 /* We're between the two cached pairs, so we do the calculation
7753 offset by the byte/utf-8 positions for the earlier pair,
7754 then add the utf-8 characters from the string start to
7755 there. */
7756 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7757 s + cache[1], cache[0] - cache[2])
7758 + cache[2];
7759
7760 }
7761 else { /* cache[3] > offset */
7762 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7763 cache[2]);
7764
7765 }
7766 ASSERT_UTF8_CACHE(cache);
7767 found = TRUE;
7768 } else if (mg->mg_len != -1) {
7769 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7770 found = TRUE;
7771 }
7772 }
7773 if (!found || PL_utf8cache < 0) {
7774 const STRLEN real_len = utf8_length(s, send);
7775
7776 if (found && PL_utf8cache < 0)
7777 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7778 len = real_len;
7779 }
7780
7781 if (PL_utf8cache) {
7782 if (blen == offset)
7783 utf8_mg_len_cache_update(sv, &mg, len);
7784 else
7785 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7786 }
7787
7788 return len;
7789 }
7790
7791 /*
7792 =for apidoc sv_pos_b2u
7793
7794 Converts the value pointed to by C<offsetp> from a count of bytes from the
7795 start of the string, to a count of the equivalent number of UTF-8 chars.
7796 Handles magic and type coercion.
7797
7798 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7799 longer than 2Gb.
7800
7801 =cut
7802 */
7803
7804 /*
7805 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7806 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7807 * byte offsets.
7808 *
7809 */
7810 void
Perl_sv_pos_b2u(pTHX_ SV * const sv,I32 * const offsetp)7811 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7812 {
7813 PERL_ARGS_ASSERT_SV_POS_B2U;
7814
7815 if (!sv)
7816 return;
7817
7818 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7819 SV_GMAGIC|SV_CONST_RETURN);
7820 }
7821
7822 static void
S_assert_uft8_cache_coherent(pTHX_ const char * const func,STRLEN from_cache,STRLEN real,SV * const sv)7823 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7824 STRLEN real, SV *const sv)
7825 {
7826 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7827
7828 /* As this is debugging only code, save space by keeping this test here,
7829 rather than inlining it in all the callers. */
7830 if (from_cache == real)
7831 return;
7832
7833 /* Need to turn the assertions off otherwise we may recurse infinitely
7834 while printing error messages. */
7835 SAVEI8(PL_utf8cache);
7836 PL_utf8cache = 0;
7837 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7838 func, (UV) from_cache, (UV) real, SVfARG(sv));
7839 }
7840
7841 /*
7842 =for apidoc sv_eq
7843
7844 Returns a boolean indicating whether the strings in the two SVs are
7845 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7846 coerce its args to strings if necessary.
7847
7848 =for apidoc sv_eq_flags
7849
7850 Returns a boolean indicating whether the strings in the two SVs are
7851 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7852 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7853
7854 =cut
7855 */
7856
7857 I32
Perl_sv_eq_flags(pTHX_ SV * sv1,SV * sv2,const U32 flags)7858 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7859 {
7860 const char *pv1;
7861 STRLEN cur1;
7862 const char *pv2;
7863 STRLEN cur2;
7864
7865 if (!sv1) {
7866 pv1 = "";
7867 cur1 = 0;
7868 }
7869 else {
7870 /* if pv1 and pv2 are the same, second SvPV_const call may
7871 * invalidate pv1 (if we are handling magic), so we may need to
7872 * make a copy */
7873 if (sv1 == sv2 && flags & SV_GMAGIC
7874 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7875 pv1 = SvPV_const(sv1, cur1);
7876 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7877 }
7878 pv1 = SvPV_flags_const(sv1, cur1, flags);
7879 }
7880
7881 if (!sv2){
7882 pv2 = "";
7883 cur2 = 0;
7884 }
7885 else
7886 pv2 = SvPV_flags_const(sv2, cur2, flags);
7887
7888 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7889 /* Differing utf8ness. */
7890 if (SvUTF8(sv1)) {
7891 /* sv1 is the UTF-8 one */
7892 return bytes_cmp_utf8((const U8*)pv2, cur2,
7893 (const U8*)pv1, cur1) == 0;
7894 }
7895 else {
7896 /* sv2 is the UTF-8 one */
7897 return bytes_cmp_utf8((const U8*)pv1, cur1,
7898 (const U8*)pv2, cur2) == 0;
7899 }
7900 }
7901
7902 if (cur1 == cur2)
7903 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7904 else
7905 return 0;
7906 }
7907
7908 /*
7909 =for apidoc sv_cmp
7910
7911 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7912 string in C<sv1> is less than, equal to, or greater than the string in
7913 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7914 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7915
7916 =for apidoc sv_cmp_flags
7917
7918 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7919 string in C<sv1> is less than, equal to, or greater than the string in
7920 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7921 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7922 also C<L</sv_cmp_locale_flags>>.
7923
7924 =cut
7925 */
7926
7927 I32
Perl_sv_cmp(pTHX_ SV * const sv1,SV * const sv2)7928 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7929 {
7930 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7931 }
7932
7933 I32
Perl_sv_cmp_flags(pTHX_ SV * const sv1,SV * const sv2,const U32 flags)7934 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7935 const U32 flags)
7936 {
7937 STRLEN cur1, cur2;
7938 const char *pv1, *pv2;
7939 I32 cmp;
7940 SV *svrecode = NULL;
7941
7942 if (!sv1) {
7943 pv1 = "";
7944 cur1 = 0;
7945 }
7946 else
7947 pv1 = SvPV_flags_const(sv1, cur1, flags);
7948
7949 if (!sv2) {
7950 pv2 = "";
7951 cur2 = 0;
7952 }
7953 else
7954 pv2 = SvPV_flags_const(sv2, cur2, flags);
7955
7956 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7957 /* Differing utf8ness. */
7958 if (SvUTF8(sv1)) {
7959 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7960 (const U8*)pv1, cur1);
7961 return retval ? retval < 0 ? -1 : +1 : 0;
7962 }
7963 else {
7964 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7965 (const U8*)pv2, cur2);
7966 return retval ? retval < 0 ? -1 : +1 : 0;
7967 }
7968 }
7969
7970 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7971
7972 if (!cur1) {
7973 cmp = cur2 ? -1 : 0;
7974 } else if (!cur2) {
7975 cmp = 1;
7976 } else {
7977 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7978
7979 #ifdef EBCDIC
7980 if (! DO_UTF8(sv1)) {
7981 #endif
7982 const I32 retval = memcmp((const void*)pv1,
7983 (const void*)pv2,
7984 shortest_len);
7985 if (retval) {
7986 cmp = retval < 0 ? -1 : 1;
7987 } else if (cur1 == cur2) {
7988 cmp = 0;
7989 } else {
7990 cmp = cur1 < cur2 ? -1 : 1;
7991 }
7992 #ifdef EBCDIC
7993 }
7994 else { /* Both are to be treated as UTF-EBCDIC */
7995
7996 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7997 * which remaps code points 0-255. We therefore generally have to
7998 * unmap back to the original values to get an accurate comparison.
7999 * But we don't have to do that for UTF-8 invariants, as by
8000 * definition, they aren't remapped, nor do we have to do it for
8001 * above-latin1 code points, as they also aren't remapped. (This
8002 * code also works on ASCII platforms, but the memcmp() above is
8003 * much faster). */
8004
8005 const char *e = pv1 + shortest_len;
8006
8007 /* Find the first bytes that differ between the two strings */
8008 while (pv1 < e && *pv1 == *pv2) {
8009 pv1++;
8010 pv2++;
8011 }
8012
8013
8014 if (pv1 == e) { /* Are the same all the way to the end */
8015 if (cur1 == cur2) {
8016 cmp = 0;
8017 } else {
8018 cmp = cur1 < cur2 ? -1 : 1;
8019 }
8020 }
8021 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8022 * in the strings were. The current bytes may or may not be
8023 * at the beginning of a character. But neither or both are
8024 * (or else earlier bytes would have been different). And
8025 * if we are in the middle of a character, the two
8026 * characters are comprised of the same number of bytes
8027 * (because in this case the start bytes are the same, and
8028 * the start bytes encode the character's length). */
8029 if (UTF8_IS_INVARIANT(*pv1))
8030 {
8031 /* If both are invariants; can just compare directly */
8032 if (UTF8_IS_INVARIANT(*pv2)) {
8033 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8034 }
8035 else /* Since *pv1 is invariant, it is the whole character,
8036 which means it is at the beginning of a character.
8037 That means pv2 is also at the beginning of a
8038 character (see earlier comment). Since it isn't
8039 invariant, it must be a start byte. If it starts a
8040 character whose code point is above 255, that
8041 character is greater than any single-byte char, which
8042 *pv1 is */
8043 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8044 {
8045 cmp = -1;
8046 }
8047 else {
8048 /* Here, pv2 points to a character composed of 2 bytes
8049 * whose code point is < 256. Get its code point and
8050 * compare with *pv1 */
8051 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8052 ? -1
8053 : 1;
8054 }
8055 }
8056 else /* The code point starting at pv1 isn't a single byte */
8057 if (UTF8_IS_INVARIANT(*pv2))
8058 {
8059 /* But here, the code point starting at *pv2 is a single byte,
8060 * and so *pv1 must begin a character, hence is a start byte.
8061 * If that character is above 255, it is larger than any
8062 * single-byte char, which *pv2 is */
8063 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8064 cmp = 1;
8065 }
8066 else {
8067 /* Here, pv1 points to a character composed of 2 bytes
8068 * whose code point is < 256. Get its code point and
8069 * compare with the single byte character *pv2 */
8070 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8071 ? -1
8072 : 1;
8073 }
8074 }
8075 else /* Here, we've ruled out either *pv1 and *pv2 being
8076 invariant. That means both are part of variants, but not
8077 necessarily at the start of a character */
8078 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8079 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8080 {
8081 /* Here, at least one is the start of a character, which means
8082 * the other is also a start byte. And the code point of at
8083 * least one of the characters is above 255. It is a
8084 * characteristic of UTF-EBCDIC that all start bytes for
8085 * above-latin1 code points are well behaved as far as code
8086 * point comparisons go, and all are larger than all other
8087 * start bytes, so the comparison with those is also well
8088 * behaved */
8089 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8090 }
8091 else {
8092 /* Here both *pv1 and *pv2 are part of variant characters.
8093 * They could be both continuations, or both start characters.
8094 * (One or both could even be an illegal start character (for
8095 * an overlong) which for the purposes of sorting we treat as
8096 * legal. */
8097 if (UTF8_IS_CONTINUATION(*pv1)) {
8098
8099 /* If they are continuations for code points above 255,
8100 * then comparing the current byte is sufficient, as there
8101 * is no remapping of these and so the comparison is
8102 * well-behaved. We determine if they are such
8103 * continuations by looking at the preceding byte. It
8104 * could be a start byte, from which we can tell if it is
8105 * for an above 255 code point. Or it could be a
8106 * continuation, which means the character occupies at
8107 * least 3 bytes, so must be above 255. */
8108 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8109 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8110 {
8111 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8112 goto cmp_done;
8113 }
8114
8115 /* Here, the continuations are for code points below 256;
8116 * back up one to get to the start byte */
8117 pv1--;
8118 pv2--;
8119 }
8120
8121 /* We need to get the actual native code point of each of these
8122 * variants in order to compare them */
8123 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8124 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8125 ? -1
8126 : 1;
8127 }
8128 }
8129 cmp_done: ;
8130 #endif
8131 }
8132
8133 SvREFCNT_dec(svrecode);
8134
8135 return cmp;
8136 }
8137
8138 /*
8139 =for apidoc sv_cmp_locale
8140
8141 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8142 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8143 if necessary. See also C<L</sv_cmp>>.
8144
8145 =for apidoc sv_cmp_locale_flags
8146
8147 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8148 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8149 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8150 C<L</sv_cmp_flags>>.
8151
8152 =cut
8153 */
8154
8155 I32
Perl_sv_cmp_locale(pTHX_ SV * const sv1,SV * const sv2)8156 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8157 {
8158 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8159 }
8160
8161 I32
Perl_sv_cmp_locale_flags(pTHX_ SV * const sv1,SV * const sv2,const U32 flags)8162 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8163 const U32 flags)
8164 {
8165 #ifdef USE_LOCALE_COLLATE
8166
8167 char *pv1, *pv2;
8168 STRLEN len1, len2;
8169 I32 retval;
8170
8171 if (PL_collation_standard)
8172 goto raw_compare;
8173
8174 len1 = len2 = 0;
8175
8176 /* Revert to using raw compare if both operands exist, but either one
8177 * doesn't transform properly for collation */
8178 if (sv1 && sv2) {
8179 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8180 if (! pv1) {
8181 goto raw_compare;
8182 }
8183 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8184 if (! pv2) {
8185 goto raw_compare;
8186 }
8187 }
8188 else {
8189 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8190 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8191 }
8192
8193 if (!pv1 || !len1) {
8194 if (pv2 && len2)
8195 return -1;
8196 else
8197 goto raw_compare;
8198 }
8199 else {
8200 if (!pv2 || !len2)
8201 return 1;
8202 }
8203
8204 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8205
8206 if (retval)
8207 return retval < 0 ? -1 : 1;
8208
8209 /*
8210 * When the result of collation is equality, that doesn't mean
8211 * that there are no differences -- some locales exclude some
8212 * characters from consideration. So to avoid false equalities,
8213 * we use the raw string as a tiebreaker.
8214 */
8215
8216 raw_compare:
8217 /* FALLTHROUGH */
8218
8219 #else
8220 PERL_UNUSED_ARG(flags);
8221 #endif /* USE_LOCALE_COLLATE */
8222
8223 return sv_cmp(sv1, sv2);
8224 }
8225
8226
8227 #ifdef USE_LOCALE_COLLATE
8228
8229 /*
8230 =for apidoc sv_collxfrm
8231
8232 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8233 C<L</sv_collxfrm_flags>>.
8234
8235 =for apidoc sv_collxfrm_flags
8236
8237 Add Collate Transform magic to an SV if it doesn't already have it. If the
8238 flags contain C<SV_GMAGIC>, it handles get-magic.
8239
8240 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8241 scalar data of the variable, but transformed to such a format that a normal
8242 memory comparison can be used to compare the data according to the locale
8243 settings.
8244
8245 =cut
8246 */
8247
8248 char *
Perl_sv_collxfrm_flags(pTHX_ SV * const sv,STRLEN * const nxp,const I32 flags)8249 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8250 {
8251 MAGIC *mg;
8252
8253 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8254
8255 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8256
8257 /* If we don't have collation magic on 'sv', or the locale has changed
8258 * since the last time we calculated it, get it and save it now */
8259 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8260 const char *s;
8261 char *xf;
8262 STRLEN len, xlen;
8263
8264 /* Free the old space */
8265 if (mg)
8266 Safefree(mg->mg_ptr);
8267
8268 s = SvPV_flags_const(sv, len, flags);
8269 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8270 if (! mg) {
8271 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8272 0, 0);
8273 assert(mg);
8274 }
8275 mg->mg_ptr = xf;
8276 mg->mg_len = xlen;
8277 }
8278 else {
8279 if (mg) {
8280 mg->mg_ptr = NULL;
8281 mg->mg_len = -1;
8282 }
8283 }
8284 }
8285
8286 if (mg && mg->mg_ptr) {
8287 *nxp = mg->mg_len;
8288 return mg->mg_ptr + sizeof(PL_collation_ix);
8289 }
8290 else {
8291 *nxp = 0;
8292 return NULL;
8293 }
8294 }
8295
8296 #endif /* USE_LOCALE_COLLATE */
8297
8298 static char *
S_sv_gets_append_to_utf8(pTHX_ SV * const sv,PerlIO * const fp,I32 append)8299 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8300 {
8301 SV * const tsv = newSV(0);
8302 ENTER;
8303 SAVEFREESV(tsv);
8304 sv_gets(tsv, fp, 0);
8305 sv_utf8_upgrade_nomg(tsv);
8306 SvCUR_set(sv,append);
8307 sv_catsv(sv,tsv);
8308 LEAVE;
8309 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8310 }
8311
8312 static char *
S_sv_gets_read_record(pTHX_ SV * const sv,PerlIO * const fp,I32 append)8313 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8314 {
8315 SSize_t bytesread;
8316 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8317 /* Grab the size of the record we're getting */
8318 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8319
8320 /* Go yank in */
8321 #ifdef __VMS
8322 int fd;
8323 Stat_t st;
8324
8325 /* With a true, record-oriented file on VMS, we need to use read directly
8326 * to ensure that we respect RMS record boundaries. The user is responsible
8327 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8328 * record size) field. N.B. This is likely to produce invalid results on
8329 * varying-width character data when a record ends mid-character.
8330 */
8331 fd = PerlIO_fileno(fp);
8332 if (fd != -1
8333 && PerlLIO_fstat(fd, &st) == 0
8334 && (st.st_fab_rfm == FAB$C_VAR
8335 || st.st_fab_rfm == FAB$C_VFC
8336 || st.st_fab_rfm == FAB$C_FIX)) {
8337
8338 bytesread = PerlLIO_read(fd, buffer, recsize);
8339 }
8340 else /* in-memory file from PerlIO::Scalar
8341 * or not a record-oriented file
8342 */
8343 #endif
8344 {
8345 bytesread = PerlIO_read(fp, buffer, recsize);
8346
8347 /* At this point, the logic in sv_get() means that sv will
8348 be treated as utf-8 if the handle is utf8.
8349 */
8350 if (PerlIO_isutf8(fp) && bytesread > 0) {
8351 char *bend = buffer + bytesread;
8352 char *bufp = buffer;
8353 size_t charcount = 0;
8354 bool charstart = TRUE;
8355 STRLEN skip = 0;
8356
8357 while (charcount < recsize) {
8358 /* count accumulated characters */
8359 while (bufp < bend) {
8360 if (charstart) {
8361 skip = UTF8SKIP(bufp);
8362 }
8363 if (bufp + skip > bend) {
8364 /* partial at the end */
8365 charstart = FALSE;
8366 break;
8367 }
8368 else {
8369 ++charcount;
8370 bufp += skip;
8371 charstart = TRUE;
8372 }
8373 }
8374
8375 if (charcount < recsize) {
8376 STRLEN readsize;
8377 STRLEN bufp_offset = bufp - buffer;
8378 SSize_t morebytesread;
8379
8380 /* originally I read enough to fill any incomplete
8381 character and the first byte of the next
8382 character if needed, but if there's many
8383 multi-byte encoded characters we're going to be
8384 making a read call for every character beyond
8385 the original read size.
8386
8387 So instead, read the rest of the character if
8388 any, and enough bytes to match at least the
8389 start bytes for each character we're going to
8390 read.
8391 */
8392 if (charstart)
8393 readsize = recsize - charcount;
8394 else
8395 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8396 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8397 bend = buffer + bytesread;
8398 morebytesread = PerlIO_read(fp, bend, readsize);
8399 if (morebytesread <= 0) {
8400 /* we're done, if we still have incomplete
8401 characters the check code in sv_gets() will
8402 warn about them.
8403
8404 I'd originally considered doing
8405 PerlIO_ungetc() on all but the lead
8406 character of the incomplete character, but
8407 read() doesn't do that, so I don't.
8408 */
8409 break;
8410 }
8411
8412 /* prepare to scan some more */
8413 bytesread += morebytesread;
8414 bend = buffer + bytesread;
8415 bufp = buffer + bufp_offset;
8416 }
8417 }
8418 }
8419 }
8420
8421 if (bytesread < 0)
8422 bytesread = 0;
8423 SvCUR_set(sv, bytesread + append);
8424 buffer[bytesread] = '\0';
8425 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8426 }
8427
8428 /*
8429 =for apidoc sv_gets
8430
8431 Get a line from the filehandle and store it into the SV, optionally
8432 appending to the currently-stored string. If C<append> is not 0, the
8433 line is appended to the SV instead of overwriting it. C<append> should
8434 be set to the byte offset that the appended string should start at
8435 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8436
8437 =cut
8438 */
8439
8440 char *
Perl_sv_gets(pTHX_ SV * const sv,PerlIO * const fp,I32 append)8441 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8442 {
8443 const char *rsptr;
8444 STRLEN rslen;
8445 STDCHAR rslast;
8446 STDCHAR *bp;
8447 SSize_t cnt;
8448 int i = 0;
8449 int rspara = 0;
8450
8451 PERL_ARGS_ASSERT_SV_GETS;
8452
8453 if (SvTHINKFIRST(sv))
8454 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8455 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8456 from <>.
8457 However, perlbench says it's slower, because the existing swipe code
8458 is faster than copy on write.
8459 Swings and roundabouts. */
8460 SvUPGRADE(sv, SVt_PV);
8461
8462 if (append) {
8463 /* line is going to be appended to the existing buffer in the sv */
8464 if (PerlIO_isutf8(fp)) {
8465 if (!SvUTF8(sv)) {
8466 sv_utf8_upgrade_nomg(sv);
8467 sv_pos_u2b(sv,&append,0);
8468 }
8469 } else if (SvUTF8(sv)) {
8470 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8471 }
8472 }
8473
8474 SvPOK_only(sv);
8475 if (!append) {
8476 /* not appending - "clear" the string by setting SvCUR to 0,
8477 * the pv is still avaiable. */
8478 SvCUR_set(sv,0);
8479 }
8480 if (PerlIO_isutf8(fp))
8481 SvUTF8_on(sv);
8482
8483 if (IN_PERL_COMPILETIME) {
8484 /* we always read code in line mode */
8485 rsptr = "\n";
8486 rslen = 1;
8487 }
8488 else if (RsSNARF(PL_rs)) {
8489 /* If it is a regular disk file use size from stat() as estimate
8490 of amount we are going to read -- may result in mallocing
8491 more memory than we really need if the layers below reduce
8492 the size we read (e.g. CRLF or a gzip layer).
8493 */
8494 Stat_t st;
8495 int fd = PerlIO_fileno(fp);
8496 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8497 const Off_t offset = PerlIO_tell(fp);
8498 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8499 #ifdef PERL_COPY_ON_WRITE
8500 /* Add an extra byte for the sake of copy-on-write's
8501 * buffer reference count. */
8502 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8503 #else
8504 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8505 #endif
8506 }
8507 }
8508 rsptr = NULL;
8509 rslen = 0;
8510 }
8511 else if (RsRECORD(PL_rs)) {
8512 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8513 }
8514 else if (RsPARA(PL_rs)) {
8515 rsptr = "\n\n";
8516 rslen = 2;
8517 rspara = 1;
8518 }
8519 else {
8520 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8521 if (PerlIO_isutf8(fp)) {
8522 rsptr = SvPVutf8(PL_rs, rslen);
8523 }
8524 else {
8525 if (SvUTF8(PL_rs)) {
8526 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8527 Perl_croak(aTHX_ "Wide character in $/");
8528 }
8529 }
8530 /* extract the raw pointer to the record separator */
8531 rsptr = SvPV_const(PL_rs, rslen);
8532 }
8533 }
8534
8535 /* rslast is the last character in the record separator
8536 * note we don't use rslast except when rslen is true, so the
8537 * null assign is a placeholder. */
8538 rslast = rslen ? rsptr[rslen - 1] : '\0';
8539
8540 if (rspara) { /* have to do this both before and after */
8541 /* to make sure file boundaries work right */
8542 while (1) {
8543 if (PerlIO_eof(fp))
8544 return 0;
8545 i = PerlIO_getc(fp);
8546 if (i != '\n') {
8547 if (i == -1)
8548 return 0;
8549 PerlIO_ungetc(fp,i);
8550 break;
8551 }
8552 }
8553 }
8554
8555 /* See if we know enough about I/O mechanism to cheat it ! */
8556
8557 /* This used to be #ifdef test - it is made run-time test for ease
8558 of abstracting out stdio interface. One call should be cheap
8559 enough here - and may even be a macro allowing compile
8560 time optimization.
8561 */
8562
8563 if (PerlIO_fast_gets(fp)) {
8564 /*
8565 * We can do buffer based IO operations on this filehandle.
8566 *
8567 * This means we can bypass a lot of subcalls and process
8568 * the buffer directly, it also means we know the upper bound
8569 * on the amount of data we might read of the current buffer
8570 * into our sv. Knowing this allows us to preallocate the pv
8571 * to be able to hold that maximum, which allows us to simplify
8572 * a lot of logic. */
8573
8574 /*
8575 * We're going to steal some values from the stdio struct
8576 * and put EVERYTHING in the innermost loop into registers.
8577 */
8578 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8579 STRLEN bpx; /* length of the data in the target sv
8580 used to fix pointers after a SvGROW */
8581 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8582 of data left in the read-ahead buffer.
8583 If 0 then the pv buffer can hold the full
8584 amount left, otherwise this is the amount it
8585 can hold. */
8586
8587 /* Here is some breathtakingly efficient cheating */
8588
8589 /* When you read the following logic resist the urge to think
8590 * of record separators that are 1 byte long. They are an
8591 * uninteresting special (simple) case.
8592 *
8593 * Instead think of record separators which are at least 2 bytes
8594 * long, and keep in mind that we need to deal with such
8595 * separators when they cross a read-ahead buffer boundary.
8596 *
8597 * Also consider that we need to gracefully deal with separators
8598 * that may be longer than a single read ahead buffer.
8599 *
8600 * Lastly do not forget we want to copy the delimiter as well. We
8601 * are copying all data in the file _up_to_and_including_ the separator
8602 * itself.
8603 *
8604 * Now that you have all that in mind here is what is happening below:
8605 *
8606 * 1. When we first enter the loop we do some memory book keeping to see
8607 * how much free space there is in the target SV. (This sub assumes that
8608 * it is operating on the same SV most of the time via $_ and that it is
8609 * going to be able to reuse the same pv buffer each call.) If there is
8610 * "enough" room then we set "shortbuffered" to how much space there is
8611 * and start reading forward.
8612 *
8613 * 2. When we scan forward we copy from the read-ahead buffer to the target
8614 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8615 * and the end of the of pv, as well as for the "rslast", which is the last
8616 * char of the separator.
8617 *
8618 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8619 * (which has a "complete" record up to the point we saw rslast) and check
8620 * it to see if it matches the separator. If it does we are done. If it doesn't
8621 * we continue on with the scan/copy.
8622 *
8623 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8624 * the IO system to read the next buffer. We do this by doing a getc(), which
8625 * returns a single char read (or EOF), and prefills the buffer, and also
8626 * allows us to find out how full the buffer is. We use this information to
8627 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8628 * the returned single char into the target sv, and then go back into scan
8629 * forward mode.
8630 *
8631 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8632 * remaining space in the read-buffer.
8633 *
8634 * Note that this code despite its twisty-turny nature is pretty darn slick.
8635 * It manages single byte separators, multi-byte cross boundary separators,
8636 * and cross-read-buffer separators cleanly and efficiently at the cost
8637 * of potentially greatly overallocating the target SV.
8638 *
8639 * Yves
8640 */
8641
8642
8643 /* get the number of bytes remaining in the read-ahead buffer
8644 * on first call on a given fp this will return 0.*/
8645 cnt = PerlIO_get_cnt(fp);
8646
8647 /* make sure we have the room */
8648 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8649 /* Not room for all of it
8650 if we are looking for a separator and room for some
8651 */
8652 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8653 /* just process what we have room for */
8654 shortbuffered = cnt - SvLEN(sv) + append + 1;
8655 cnt -= shortbuffered;
8656 }
8657 else {
8658 /* ensure that the target sv has enough room to hold
8659 * the rest of the read-ahead buffer */
8660 shortbuffered = 0;
8661 /* remember that cnt can be negative */
8662 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8663 }
8664 }
8665 else {
8666 /* we have enough room to hold the full buffer, lets scream */
8667 shortbuffered = 0;
8668 }
8669
8670 /* extract the pointer to sv's string buffer, offset by append as necessary */
8671 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8672 /* extract the point to the read-ahead buffer */
8673 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8674
8675 /* some trace debug output */
8676 DEBUG_P(PerlIO_printf(Perl_debug_log,
8677 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8678 DEBUG_P(PerlIO_printf(Perl_debug_log,
8679 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8680 UVuf "\n",
8681 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8682 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8683
8684 for (;;) {
8685 screamer:
8686 /* if there is stuff left in the read-ahead buffer */
8687 if (cnt > 0) {
8688 /* if there is a separator */
8689 if (rslen) {
8690 /* find next rslast */
8691 STDCHAR *p;
8692
8693 /* shortcut common case of blank line */
8694 cnt--;
8695 if ((*bp++ = *ptr++) == rslast)
8696 goto thats_all_folks;
8697
8698 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8699 if (p) {
8700 SSize_t got = p - ptr + 1;
8701 Copy(ptr, bp, got, STDCHAR);
8702 ptr += got;
8703 bp += got;
8704 cnt -= got;
8705 goto thats_all_folks;
8706 }
8707 Copy(ptr, bp, cnt, STDCHAR);
8708 ptr += cnt;
8709 bp += cnt;
8710 cnt = 0;
8711 }
8712 else {
8713 /* no separator, slurp the full buffer */
8714 Copy(ptr, bp, cnt, char); /* this | eat */
8715 bp += cnt; /* screams | dust */
8716 ptr += cnt; /* louder | sed :-) */
8717 cnt = 0;
8718 assert (!shortbuffered);
8719 goto cannot_be_shortbuffered;
8720 }
8721 }
8722
8723 if (shortbuffered) { /* oh well, must extend */
8724 /* we didnt have enough room to fit the line into the target buffer
8725 * so we must extend the target buffer and keep going */
8726 cnt = shortbuffered;
8727 shortbuffered = 0;
8728 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8729 SvCUR_set(sv, bpx);
8730 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8731 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8732 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8733 continue;
8734 }
8735
8736 cannot_be_shortbuffered:
8737 /* we need to refill the read-ahead buffer if possible */
8738
8739 DEBUG_P(PerlIO_printf(Perl_debug_log,
8740 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8741 PTR2UV(ptr),(IV)cnt));
8742 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8743
8744 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8745 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8746 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8747 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8748
8749 /*
8750 call PerlIO_getc() to let it prefill the lookahead buffer
8751
8752 This used to call 'filbuf' in stdio form, but as that behaves like
8753 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8754 another abstraction.
8755
8756 Note we have to deal with the char in 'i' if we are not at EOF
8757 */
8758 bpx = bp - (STDCHAR*)SvPVX_const(sv);
8759 /* signals might be called here, possibly modifying sv */
8760 i = PerlIO_getc(fp); /* get more characters */
8761 bp = (STDCHAR*)SvPVX_const(sv) + bpx;
8762
8763 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8764 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8765 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8766 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8767
8768 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8769 cnt = PerlIO_get_cnt(fp);
8770 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8771 DEBUG_P(PerlIO_printf(Perl_debug_log,
8772 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8773 PTR2UV(ptr),(IV)cnt));
8774
8775 if (i == EOF) /* all done for ever? */
8776 goto thats_really_all_folks;
8777
8778 /* make sure we have enough space in the target sv */
8779 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8780 SvCUR_set(sv, bpx);
8781 SvGROW(sv, bpx + cnt + 2);
8782 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8783
8784 /* copy of the char we got from getc() */
8785 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8786
8787 /* make sure we deal with the i being the last character of a separator */
8788 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8789 goto thats_all_folks;
8790 }
8791
8792 thats_all_folks:
8793 /* check if we have actually found the separator - only really applies
8794 * when rslen > 1 */
8795 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8796 memNE((char*)bp - rslen, rsptr, rslen))
8797 goto screamer; /* go back to the fray */
8798 thats_really_all_folks:
8799 if (shortbuffered)
8800 cnt += shortbuffered;
8801 DEBUG_P(PerlIO_printf(Perl_debug_log,
8802 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8803 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8804 DEBUG_P(PerlIO_printf(Perl_debug_log,
8805 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8806 "\n",
8807 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8808 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8809 *bp = '\0';
8810 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8811 DEBUG_P(PerlIO_printf(Perl_debug_log,
8812 "Screamer: done, len=%ld, string=|%.*s|\n",
8813 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8814 }
8815 else
8816 {
8817 /*The big, slow, and stupid way. */
8818 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8819 STDCHAR *buf = NULL;
8820 Newx(buf, 8192, STDCHAR);
8821 assert(buf);
8822 #else
8823 STDCHAR buf[8192];
8824 #endif
8825
8826 screamer2:
8827 if (rslen) {
8828 const STDCHAR * const bpe = buf + sizeof(buf);
8829 bp = buf;
8830 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8831 ; /* keep reading */
8832 cnt = bp - buf;
8833 }
8834 else {
8835 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8836 /* Accommodate broken VAXC compiler, which applies U8 cast to
8837 * both args of ?: operator, causing EOF to change into 255
8838 */
8839 if (cnt > 0)
8840 i = (U8)buf[cnt - 1];
8841 else
8842 i = EOF;
8843 }
8844
8845 if (cnt < 0)
8846 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8847 if (append)
8848 sv_catpvn_nomg(sv, (char *) buf, cnt);
8849 else
8850 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8851
8852 if (i != EOF && /* joy */
8853 (!rslen ||
8854 SvCUR(sv) < rslen ||
8855 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8856 {
8857 append = -1;
8858 /*
8859 * If we're reading from a TTY and we get a short read,
8860 * indicating that the user hit his EOF character, we need
8861 * to notice it now, because if we try to read from the TTY
8862 * again, the EOF condition will disappear.
8863 *
8864 * The comparison of cnt to sizeof(buf) is an optimization
8865 * that prevents unnecessary calls to feof().
8866 *
8867 * - jik 9/25/96
8868 */
8869 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8870 goto screamer2;
8871 }
8872
8873 #ifdef USE_HEAP_INSTEAD_OF_STACK
8874 Safefree(buf);
8875 #endif
8876 }
8877
8878 if (rspara) { /* have to do this both before and after */
8879 while (i != EOF) { /* to make sure file boundaries work right */
8880 i = PerlIO_getc(fp);
8881 if (i != '\n') {
8882 PerlIO_ungetc(fp,i);
8883 break;
8884 }
8885 }
8886 }
8887
8888 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8889 }
8890
8891 /*
8892 =for apidoc sv_inc
8893
8894 Auto-increment of the value in the SV, doing string to numeric conversion
8895 if necessary. Handles 'get' magic and operator overloading.
8896
8897 =cut
8898 */
8899
8900 void
Perl_sv_inc(pTHX_ SV * const sv)8901 Perl_sv_inc(pTHX_ SV *const sv)
8902 {
8903 if (!sv)
8904 return;
8905 SvGETMAGIC(sv);
8906 sv_inc_nomg(sv);
8907 }
8908
8909 /*
8910 =for apidoc sv_inc_nomg
8911
8912 Auto-increment of the value in the SV, doing string to numeric conversion
8913 if necessary. Handles operator overloading. Skips handling 'get' magic.
8914
8915 =cut
8916 */
8917
8918 void
Perl_sv_inc_nomg(pTHX_ SV * const sv)8919 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8920 {
8921 char *d;
8922 int flags;
8923
8924 if (!sv)
8925 return;
8926 if (SvTHINKFIRST(sv)) {
8927 if (SvREADONLY(sv)) {
8928 Perl_croak_no_modify();
8929 }
8930 if (SvROK(sv)) {
8931 IV i;
8932 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8933 return;
8934 i = PTR2IV(SvRV(sv));
8935 sv_unref(sv);
8936 sv_setiv(sv, i);
8937 }
8938 else sv_force_normal_flags(sv, 0);
8939 }
8940 flags = SvFLAGS(sv);
8941 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8942 /* It's (privately or publicly) a float, but not tested as an
8943 integer, so test it to see. */
8944 (void) SvIV(sv);
8945 flags = SvFLAGS(sv);
8946 }
8947 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8948 /* It's publicly an integer, or privately an integer-not-float */
8949 #ifdef PERL_PRESERVE_IVUV
8950 oops_its_int:
8951 #endif
8952 if (SvIsUV(sv)) {
8953 if (SvUVX(sv) == UV_MAX)
8954 sv_setnv(sv, UV_MAX_P1);
8955 else
8956 (void)SvIOK_only_UV(sv);
8957 SvUV_set(sv, SvUVX(sv) + 1);
8958 } else {
8959 if (SvIVX(sv) == IV_MAX)
8960 sv_setuv(sv, (UV)IV_MAX + 1);
8961 else {
8962 (void)SvIOK_only(sv);
8963 SvIV_set(sv, SvIVX(sv) + 1);
8964 }
8965 }
8966 return;
8967 }
8968 if (flags & SVp_NOK) {
8969 const NV was = SvNVX(sv);
8970 if (LIKELY(!Perl_isinfnan(was)) &&
8971 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8972 was >= NV_OVERFLOWS_INTEGERS_AT) {
8973 /* diag_listed_as: Lost precision when %s %f by 1 */
8974 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8975 "Lost precision when incrementing %" NVff " by 1",
8976 was);
8977 }
8978 (void)SvNOK_only(sv);
8979 SvNV_set(sv, was + 1.0);
8980 return;
8981 }
8982
8983 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8984 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8985 Perl_croak_no_modify();
8986
8987 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8988 if ((flags & SVTYPEMASK) < SVt_PVIV)
8989 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8990 (void)SvIOK_only(sv);
8991 SvIV_set(sv, 1);
8992 return;
8993 }
8994 d = SvPVX(sv);
8995 while (isALPHA(*d)) d++;
8996 while (isDIGIT(*d)) d++;
8997 if (d < SvEND(sv)) {
8998 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8999 #ifdef PERL_PRESERVE_IVUV
9000 /* Got to punt this as an integer if needs be, but we don't issue
9001 warnings. Probably ought to make the sv_iv_please() that does
9002 the conversion if possible, and silently. */
9003 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9004 /* Need to try really hard to see if it's an integer.
9005 9.22337203685478e+18 is an integer.
9006 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9007 so $a="9.22337203685478e+18"; $a+0; $a++
9008 needs to be the same as $a="9.22337203685478e+18"; $a++
9009 or we go insane. */
9010
9011 (void) sv_2iv(sv);
9012 if (SvIOK(sv))
9013 goto oops_its_int;
9014
9015 /* sv_2iv *should* have made this an NV */
9016 if (flags & SVp_NOK) {
9017 (void)SvNOK_only(sv);
9018 SvNV_set(sv, SvNVX(sv) + 1.0);
9019 return;
9020 }
9021 /* I don't think we can get here. Maybe I should assert this
9022 And if we do get here I suspect that sv_setnv will croak. NWC
9023 Fall through. */
9024 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9025 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9026 }
9027 #endif /* PERL_PRESERVE_IVUV */
9028 if (!numtype && ckWARN(WARN_NUMERIC))
9029 not_incrementable(sv);
9030 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9031 return;
9032 }
9033 d--;
9034 while (d >= SvPVX_const(sv)) {
9035 if (isDIGIT(*d)) {
9036 if (++*d <= '9')
9037 return;
9038 *(d--) = '0';
9039 }
9040 else {
9041 #ifdef EBCDIC
9042 /* MKS: The original code here died if letters weren't consecutive.
9043 * at least it didn't have to worry about non-C locales. The
9044 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9045 * arranged in order (although not consecutively) and that only
9046 * [A-Za-z] are accepted by isALPHA in the C locale.
9047 */
9048 if (isALPHA_FOLD_NE(*d, 'z')) {
9049 do { ++*d; } while (!isALPHA(*d));
9050 return;
9051 }
9052 *(d--) -= 'z' - 'a';
9053 #else
9054 ++*d;
9055 if (isALPHA(*d))
9056 return;
9057 *(d--) -= 'z' - 'a' + 1;
9058 #endif
9059 }
9060 }
9061 /* oh,oh, the number grew */
9062 SvGROW(sv, SvCUR(sv) + 2);
9063 SvCUR_set(sv, SvCUR(sv) + 1);
9064 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9065 *d = d[-1];
9066 if (isDIGIT(d[1]))
9067 *d = '1';
9068 else
9069 *d = d[1];
9070 }
9071
9072 /*
9073 =for apidoc sv_dec
9074
9075 Auto-decrement of the value in the SV, doing string to numeric conversion
9076 if necessary. Handles 'get' magic and operator overloading.
9077
9078 =cut
9079 */
9080
9081 void
Perl_sv_dec(pTHX_ SV * const sv)9082 Perl_sv_dec(pTHX_ SV *const sv)
9083 {
9084 if (!sv)
9085 return;
9086 SvGETMAGIC(sv);
9087 sv_dec_nomg(sv);
9088 }
9089
9090 /*
9091 =for apidoc sv_dec_nomg
9092
9093 Auto-decrement of the value in the SV, doing string to numeric conversion
9094 if necessary. Handles operator overloading. Skips handling 'get' magic.
9095
9096 =cut
9097 */
9098
9099 void
Perl_sv_dec_nomg(pTHX_ SV * const sv)9100 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9101 {
9102 int flags;
9103
9104 if (!sv)
9105 return;
9106 if (SvTHINKFIRST(sv)) {
9107 if (SvREADONLY(sv)) {
9108 Perl_croak_no_modify();
9109 }
9110 if (SvROK(sv)) {
9111 IV i;
9112 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9113 return;
9114 i = PTR2IV(SvRV(sv));
9115 sv_unref(sv);
9116 sv_setiv(sv, i);
9117 }
9118 else sv_force_normal_flags(sv, 0);
9119 }
9120 /* Unlike sv_inc we don't have to worry about string-never-numbers
9121 and keeping them magic. But we mustn't warn on punting */
9122 flags = SvFLAGS(sv);
9123 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9124 /* It's publicly an integer, or privately an integer-not-float */
9125 #ifdef PERL_PRESERVE_IVUV
9126 oops_its_int:
9127 #endif
9128 if (SvIsUV(sv)) {
9129 if (SvUVX(sv) == 0) {
9130 (void)SvIOK_only(sv);
9131 SvIV_set(sv, -1);
9132 }
9133 else {
9134 (void)SvIOK_only_UV(sv);
9135 SvUV_set(sv, SvUVX(sv) - 1);
9136 }
9137 } else {
9138 if (SvIVX(sv) == IV_MIN) {
9139 sv_setnv(sv, (NV)IV_MIN);
9140 goto oops_its_num;
9141 }
9142 else {
9143 (void)SvIOK_only(sv);
9144 SvIV_set(sv, SvIVX(sv) - 1);
9145 }
9146 }
9147 return;
9148 }
9149 if (flags & SVp_NOK) {
9150 oops_its_num:
9151 {
9152 const NV was = SvNVX(sv);
9153 if (LIKELY(!Perl_isinfnan(was)) &&
9154 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9155 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9156 /* diag_listed_as: Lost precision when %s %f by 1 */
9157 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9158 "Lost precision when decrementing %" NVff " by 1",
9159 was);
9160 }
9161 (void)SvNOK_only(sv);
9162 SvNV_set(sv, was - 1.0);
9163 return;
9164 }
9165 }
9166
9167 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9168 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9169 Perl_croak_no_modify();
9170
9171 if (!(flags & SVp_POK)) {
9172 if ((flags & SVTYPEMASK) < SVt_PVIV)
9173 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9174 SvIV_set(sv, -1);
9175 (void)SvIOK_only(sv);
9176 return;
9177 }
9178 #ifdef PERL_PRESERVE_IVUV
9179 {
9180 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9181 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9182 /* Need to try really hard to see if it's an integer.
9183 9.22337203685478e+18 is an integer.
9184 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9185 so $a="9.22337203685478e+18"; $a+0; $a--
9186 needs to be the same as $a="9.22337203685478e+18"; $a--
9187 or we go insane. */
9188
9189 (void) sv_2iv(sv);
9190 if (SvIOK(sv))
9191 goto oops_its_int;
9192
9193 /* sv_2iv *should* have made this an NV */
9194 if (flags & SVp_NOK) {
9195 (void)SvNOK_only(sv);
9196 SvNV_set(sv, SvNVX(sv) - 1.0);
9197 return;
9198 }
9199 /* I don't think we can get here. Maybe I should assert this
9200 And if we do get here I suspect that sv_setnv will croak. NWC
9201 Fall through. */
9202 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9203 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9204 }
9205 }
9206 #endif /* PERL_PRESERVE_IVUV */
9207 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9208 }
9209
9210 /* this define is used to eliminate a chunk of duplicated but shared logic
9211 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9212 * used anywhere but here - yves
9213 */
9214 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9215 STMT_START { \
9216 SSize_t ix = ++PL_tmps_ix; \
9217 if (UNLIKELY(ix >= PL_tmps_max)) \
9218 ix = tmps_grow_p(ix); \
9219 PL_tmps_stack[ix] = (AnSv); \
9220 } STMT_END
9221
9222 /*
9223 =for apidoc sv_mortalcopy
9224
9225 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9226 The new SV is marked as mortal. It will be destroyed "soon", either by an
9227 explicit call to C<FREETMPS>, or by an implicit call at places such as
9228 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9229
9230 =cut
9231 */
9232
9233 /* Make a string that will exist for the duration of the expression
9234 * evaluation. Actually, it may have to last longer than that, but
9235 * hopefully we won't free it until it has been assigned to a
9236 * permanent location. */
9237
9238 SV *
Perl_sv_mortalcopy_flags(pTHX_ SV * const oldstr,U32 flags)9239 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9240 {
9241 SV *sv;
9242
9243 if (flags & SV_GMAGIC)
9244 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9245 new_SV(sv);
9246 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9247 PUSH_EXTEND_MORTAL__SV_C(sv);
9248 SvTEMP_on(sv);
9249 return sv;
9250 }
9251
9252 /*
9253 =for apidoc sv_newmortal
9254
9255 Creates a new null SV which is mortal. The reference count of the SV is
9256 set to 1. It will be destroyed "soon", either by an explicit call to
9257 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9258 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9259
9260 =cut
9261 */
9262
9263 SV *
Perl_sv_newmortal(pTHX)9264 Perl_sv_newmortal(pTHX)
9265 {
9266 SV *sv;
9267
9268 new_SV(sv);
9269 SvFLAGS(sv) = SVs_TEMP;
9270 PUSH_EXTEND_MORTAL__SV_C(sv);
9271 return sv;
9272 }
9273
9274
9275 /*
9276 =for apidoc newSVpvn_flags
9277
9278 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9279 characters) into it. The reference count for the
9280 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9281 string. You are responsible for ensuring that the source string is at least
9282 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9283 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9284 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9285 returning. If C<SVf_UTF8> is set, C<s>
9286 is considered to be in UTF-8 and the
9287 C<SVf_UTF8> flag will be set on the new SV.
9288 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9289
9290 #define newSVpvn_utf8(s, len, u) \
9291 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9292
9293 =cut
9294 */
9295
9296 SV *
Perl_newSVpvn_flags(pTHX_ const char * const s,const STRLEN len,const U32 flags)9297 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9298 {
9299 SV *sv;
9300
9301 /* All the flags we don't support must be zero.
9302 And we're new code so I'm going to assert this from the start. */
9303 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9304 new_SV(sv);
9305 sv_setpvn(sv,s,len);
9306
9307 /* This code used to do a sv_2mortal(), however we now unroll the call to
9308 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9309 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9310 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9311 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9312 * means that we eliminate quite a few steps than it looks - Yves
9313 * (explaining patch by gfx) */
9314
9315 SvFLAGS(sv) |= flags;
9316
9317 if(flags & SVs_TEMP){
9318 PUSH_EXTEND_MORTAL__SV_C(sv);
9319 }
9320
9321 return sv;
9322 }
9323
9324 /*
9325 =for apidoc sv_2mortal
9326
9327 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9328 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9329 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9330 string buffer can be "stolen" if this SV is copied. See also
9331 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9332
9333 =cut
9334 */
9335
9336 SV *
Perl_sv_2mortal(pTHX_ SV * const sv)9337 Perl_sv_2mortal(pTHX_ SV *const sv)
9338 {
9339 dVAR;
9340 if (!sv)
9341 return sv;
9342 if (SvIMMORTAL(sv))
9343 return sv;
9344 PUSH_EXTEND_MORTAL__SV_C(sv);
9345 SvTEMP_on(sv);
9346 return sv;
9347 }
9348
9349 /*
9350 =for apidoc newSVpv
9351
9352 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9353 characters) into it. The reference count for the
9354 SV is set to 1. If C<len> is zero, Perl will compute the length using
9355 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9356 C<NUL> characters and has to have a terminating C<NUL> byte).
9357
9358 This function can cause reliability issues if you are likely to pass in
9359 empty strings that are not null terminated, because it will run
9360 strlen on the string and potentially run past valid memory.
9361
9362 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9363 For string literals use L</newSVpvs> instead. This function will work fine for
9364 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9365 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9366
9367 =cut
9368 */
9369
9370 SV *
Perl_newSVpv(pTHX_ const char * const s,const STRLEN len)9371 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9372 {
9373 SV *sv;
9374
9375 new_SV(sv);
9376 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9377 return sv;
9378 }
9379
9380 /*
9381 =for apidoc newSVpvn
9382
9383 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9384 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9385 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9386 are responsible for ensuring that the source buffer is at least
9387 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9388 undefined.
9389
9390 =cut
9391 */
9392
9393 SV *
Perl_newSVpvn(pTHX_ const char * const buffer,const STRLEN len)9394 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9395 {
9396 SV *sv;
9397 new_SV(sv);
9398 sv_setpvn(sv,buffer,len);
9399 return sv;
9400 }
9401
9402 /*
9403 =for apidoc newSVhek
9404
9405 Creates a new SV from the hash key structure. It will generate scalars that
9406 point to the shared string table where possible. Returns a new (undefined)
9407 SV if C<hek> is NULL.
9408
9409 =cut
9410 */
9411
9412 SV *
Perl_newSVhek(pTHX_ const HEK * const hek)9413 Perl_newSVhek(pTHX_ const HEK *const hek)
9414 {
9415 if (!hek) {
9416 SV *sv;
9417
9418 new_SV(sv);
9419 return sv;
9420 }
9421
9422 if (HEK_LEN(hek) == HEf_SVKEY) {
9423 return newSVsv(*(SV**)HEK_KEY(hek));
9424 } else {
9425 const int flags = HEK_FLAGS(hek);
9426 if (flags & HVhek_WASUTF8) {
9427 /* Trouble :-)
9428 Andreas would like keys he put in as utf8 to come back as utf8
9429 */
9430 STRLEN utf8_len = HEK_LEN(hek);
9431 SV * const sv = newSV_type(SVt_PV);
9432 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9433 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9434 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9435 SvUTF8_on (sv);
9436 return sv;
9437 } else if (flags & HVhek_UNSHARED) {
9438 /* A hash that isn't using shared hash keys has to have
9439 the flag in every key so that we know not to try to call
9440 share_hek_hek on it. */
9441
9442 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9443 if (HEK_UTF8(hek))
9444 SvUTF8_on (sv);
9445 return sv;
9446 }
9447 /* This will be overwhelminly the most common case. */
9448 {
9449 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9450 more efficient than sharepvn(). */
9451 SV *sv;
9452
9453 new_SV(sv);
9454 sv_upgrade(sv, SVt_PV);
9455 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9456 SvCUR_set(sv, HEK_LEN(hek));
9457 SvLEN_set(sv, 0);
9458 SvIsCOW_on(sv);
9459 SvPOK_on(sv);
9460 if (HEK_UTF8(hek))
9461 SvUTF8_on(sv);
9462 return sv;
9463 }
9464 }
9465 }
9466
9467 /*
9468 =for apidoc newSVpvn_share
9469
9470 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9471 table. If the string does not already exist in the table, it is
9472 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9473 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9474 is non-zero, that value is used; otherwise the hash is computed.
9475 The string's hash can later be retrieved from the SV
9476 with the C<SvSHARED_HASH()> macro. The idea here is
9477 that as the string table is used for shared hash keys these strings will have
9478 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9479
9480 =cut
9481 */
9482
9483 SV *
Perl_newSVpvn_share(pTHX_ const char * src,I32 len,U32 hash)9484 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9485 {
9486 dVAR;
9487 SV *sv;
9488 bool is_utf8 = FALSE;
9489 const char *const orig_src = src;
9490
9491 if (len < 0) {
9492 STRLEN tmplen = -len;
9493 is_utf8 = TRUE;
9494 /* See the note in hv.c:hv_fetch() --jhi */
9495 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9496 len = tmplen;
9497 }
9498 if (!hash)
9499 PERL_HASH(hash, src, len);
9500 new_SV(sv);
9501 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9502 changes here, update it there too. */
9503 sv_upgrade(sv, SVt_PV);
9504 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9505 SvCUR_set(sv, len);
9506 SvLEN_set(sv, 0);
9507 SvIsCOW_on(sv);
9508 SvPOK_on(sv);
9509 if (is_utf8)
9510 SvUTF8_on(sv);
9511 if (src != orig_src)
9512 Safefree(src);
9513 return sv;
9514 }
9515
9516 /*
9517 =for apidoc newSVpv_share
9518
9519 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9520 string/length pair.
9521
9522 =cut
9523 */
9524
9525 SV *
Perl_newSVpv_share(pTHX_ const char * src,U32 hash)9526 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9527 {
9528 return newSVpvn_share(src, strlen(src), hash);
9529 }
9530
9531 #if defined(PERL_IMPLICIT_CONTEXT)
9532
9533 /* pTHX_ magic can't cope with varargs, so this is a no-context
9534 * version of the main function, (which may itself be aliased to us).
9535 * Don't access this version directly.
9536 */
9537
9538 SV *
Perl_newSVpvf_nocontext(const char * const pat,...)9539 Perl_newSVpvf_nocontext(const char *const pat, ...)
9540 {
9541 dTHX;
9542 SV *sv;
9543 va_list args;
9544
9545 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9546
9547 va_start(args, pat);
9548 sv = vnewSVpvf(pat, &args);
9549 va_end(args);
9550 return sv;
9551 }
9552 #endif
9553
9554 /*
9555 =for apidoc newSVpvf
9556
9557 Creates a new SV and initializes it with the string formatted like
9558 C<sv_catpvf>.
9559
9560 =cut
9561 */
9562
9563 SV *
Perl_newSVpvf(pTHX_ const char * const pat,...)9564 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9565 {
9566 SV *sv;
9567 va_list args;
9568
9569 PERL_ARGS_ASSERT_NEWSVPVF;
9570
9571 va_start(args, pat);
9572 sv = vnewSVpvf(pat, &args);
9573 va_end(args);
9574 return sv;
9575 }
9576
9577 /* backend for newSVpvf() and newSVpvf_nocontext() */
9578
9579 SV *
Perl_vnewSVpvf(pTHX_ const char * const pat,va_list * const args)9580 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9581 {
9582 SV *sv;
9583
9584 PERL_ARGS_ASSERT_VNEWSVPVF;
9585
9586 new_SV(sv);
9587 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9588 return sv;
9589 }
9590
9591 /*
9592 =for apidoc newSVnv
9593
9594 Creates a new SV and copies a floating point value into it.
9595 The reference count for the SV is set to 1.
9596
9597 =cut
9598 */
9599
9600 SV *
Perl_newSVnv(pTHX_ const NV n)9601 Perl_newSVnv(pTHX_ const NV n)
9602 {
9603 SV *sv;
9604
9605 new_SV(sv);
9606 sv_setnv(sv,n);
9607 return sv;
9608 }
9609
9610 /*
9611 =for apidoc newSViv
9612
9613 Creates a new SV and copies an integer into it. The reference count for the
9614 SV is set to 1.
9615
9616 =cut
9617 */
9618
9619 SV *
Perl_newSViv(pTHX_ const IV i)9620 Perl_newSViv(pTHX_ const IV i)
9621 {
9622 SV *sv;
9623
9624 new_SV(sv);
9625
9626 /* Inlining ONLY the small relevant subset of sv_setiv here
9627 * for performance. Makes a significant difference. */
9628
9629 /* We're starting from SVt_FIRST, so provided that's
9630 * actual 0, we don't have to unset any SV type flags
9631 * to promote to SVt_IV. */
9632 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9633
9634 SET_SVANY_FOR_BODYLESS_IV(sv);
9635 SvFLAGS(sv) |= SVt_IV;
9636 (void)SvIOK_on(sv);
9637
9638 SvIV_set(sv, i);
9639 SvTAINT(sv);
9640
9641 return sv;
9642 }
9643
9644 /*
9645 =for apidoc newSVuv
9646
9647 Creates a new SV and copies an unsigned integer into it.
9648 The reference count for the SV is set to 1.
9649
9650 =cut
9651 */
9652
9653 SV *
Perl_newSVuv(pTHX_ const UV u)9654 Perl_newSVuv(pTHX_ const UV u)
9655 {
9656 SV *sv;
9657
9658 /* Inlining ONLY the small relevant subset of sv_setuv here
9659 * for performance. Makes a significant difference. */
9660
9661 /* Using ivs is more efficient than using uvs - see sv_setuv */
9662 if (u <= (UV)IV_MAX) {
9663 return newSViv((IV)u);
9664 }
9665
9666 new_SV(sv);
9667
9668 /* We're starting from SVt_FIRST, so provided that's
9669 * actual 0, we don't have to unset any SV type flags
9670 * to promote to SVt_IV. */
9671 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9672
9673 SET_SVANY_FOR_BODYLESS_IV(sv);
9674 SvFLAGS(sv) |= SVt_IV;
9675 (void)SvIOK_on(sv);
9676 (void)SvIsUV_on(sv);
9677
9678 SvUV_set(sv, u);
9679 SvTAINT(sv);
9680
9681 return sv;
9682 }
9683
9684 /*
9685 =for apidoc newSV_type
9686
9687 Creates a new SV, of the type specified. The reference count for the new SV
9688 is set to 1.
9689
9690 =cut
9691 */
9692
9693 SV *
Perl_newSV_type(pTHX_ const svtype type)9694 Perl_newSV_type(pTHX_ const svtype type)
9695 {
9696 SV *sv;
9697
9698 new_SV(sv);
9699 ASSUME(SvTYPE(sv) == SVt_FIRST);
9700 if(type != SVt_FIRST)
9701 sv_upgrade(sv, type);
9702 return sv;
9703 }
9704
9705 /*
9706 =for apidoc newRV_noinc
9707
9708 Creates an RV wrapper for an SV. The reference count for the original
9709 SV is B<not> incremented.
9710
9711 =cut
9712 */
9713
9714 SV *
Perl_newRV_noinc(pTHX_ SV * const tmpRef)9715 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9716 {
9717 SV *sv;
9718
9719 PERL_ARGS_ASSERT_NEWRV_NOINC;
9720
9721 new_SV(sv);
9722
9723 /* We're starting from SVt_FIRST, so provided that's
9724 * actual 0, we don't have to unset any SV type flags
9725 * to promote to SVt_IV. */
9726 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9727
9728 SET_SVANY_FOR_BODYLESS_IV(sv);
9729 SvFLAGS(sv) |= SVt_IV;
9730 SvROK_on(sv);
9731 SvIV_set(sv, 0);
9732
9733 SvTEMP_off(tmpRef);
9734 SvRV_set(sv, tmpRef);
9735
9736 return sv;
9737 }
9738
9739 /* newRV_inc is the official function name to use now.
9740 * newRV_inc is in fact #defined to newRV in sv.h
9741 */
9742
9743 SV *
Perl_newRV(pTHX_ SV * const sv)9744 Perl_newRV(pTHX_ SV *const sv)
9745 {
9746 PERL_ARGS_ASSERT_NEWRV;
9747
9748 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9749 }
9750
9751 /*
9752 =for apidoc newSVsv
9753
9754 Creates a new SV which is an exact duplicate of the original SV.
9755 (Uses C<sv_setsv>.)
9756
9757 =for apidoc newSVsv_nomg
9758
9759 Like C<newSVsv> but does not process get magic.
9760
9761 =cut
9762 */
9763
9764 SV *
Perl_newSVsv_flags(pTHX_ SV * const old,I32 flags)9765 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9766 {
9767 SV *sv;
9768
9769 if (!old)
9770 return NULL;
9771 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9772 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9773 return NULL;
9774 }
9775 /* Do this here, otherwise we leak the new SV if this croaks. */
9776 if (flags & SV_GMAGIC)
9777 SvGETMAGIC(old);
9778 new_SV(sv);
9779 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9780 return sv;
9781 }
9782
9783 /*
9784 =for apidoc sv_reset
9785
9786 Underlying implementation for the C<reset> Perl function.
9787 Note that the perl-level function is vaguely deprecated.
9788
9789 =cut
9790 */
9791
9792 void
Perl_sv_reset(pTHX_ const char * s,HV * const stash)9793 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9794 {
9795 PERL_ARGS_ASSERT_SV_RESET;
9796
9797 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9798 }
9799
9800 void
Perl_sv_resetpvn(pTHX_ const char * s,STRLEN len,HV * const stash)9801 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9802 {
9803 char todo[PERL_UCHAR_MAX+1];
9804 const char *send;
9805
9806 if (!stash || SvTYPE(stash) != SVt_PVHV)
9807 return;
9808
9809 if (!s) { /* reset ?? searches */
9810 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9811 if (mg) {
9812 const U32 count = mg->mg_len / sizeof(PMOP**);
9813 PMOP **pmp = (PMOP**) mg->mg_ptr;
9814 PMOP *const *const end = pmp + count;
9815
9816 while (pmp < end) {
9817 #ifdef USE_ITHREADS
9818 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9819 #else
9820 (*pmp)->op_pmflags &= ~PMf_USED;
9821 #endif
9822 ++pmp;
9823 }
9824 }
9825 return;
9826 }
9827
9828 /* reset variables */
9829
9830 if (!HvARRAY(stash))
9831 return;
9832
9833 Zero(todo, 256, char);
9834 send = s + len;
9835 while (s < send) {
9836 I32 max;
9837 I32 i = (unsigned char)*s;
9838 if (s[1] == '-') {
9839 s += 2;
9840 }
9841 max = (unsigned char)*s++;
9842 for ( ; i <= max; i++) {
9843 todo[i] = 1;
9844 }
9845 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9846 HE *entry;
9847 for (entry = HvARRAY(stash)[i];
9848 entry;
9849 entry = HeNEXT(entry))
9850 {
9851 GV *gv;
9852 SV *sv;
9853
9854 if (!todo[(U8)*HeKEY(entry)])
9855 continue;
9856 gv = MUTABLE_GV(HeVAL(entry));
9857 if (!isGV(gv))
9858 continue;
9859 sv = GvSV(gv);
9860 if (sv && !SvREADONLY(sv)) {
9861 SV_CHECK_THINKFIRST_COW_DROP(sv);
9862 if (!isGV(sv)) SvOK_off(sv);
9863 }
9864 if (GvAV(gv)) {
9865 av_clear(GvAV(gv));
9866 }
9867 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9868 hv_clear(GvHV(gv));
9869 }
9870 }
9871 }
9872 }
9873 }
9874
9875 /*
9876 =for apidoc sv_2io
9877
9878 Using various gambits, try to get an IO from an SV: the IO slot if its a
9879 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9880 named after the PV if we're a string.
9881
9882 'Get' magic is ignored on the C<sv> passed in, but will be called on
9883 C<SvRV(sv)> if C<sv> is an RV.
9884
9885 =cut
9886 */
9887
9888 IO*
Perl_sv_2io(pTHX_ SV * const sv)9889 Perl_sv_2io(pTHX_ SV *const sv)
9890 {
9891 IO* io;
9892 GV* gv;
9893
9894 PERL_ARGS_ASSERT_SV_2IO;
9895
9896 switch (SvTYPE(sv)) {
9897 case SVt_PVIO:
9898 io = MUTABLE_IO(sv);
9899 break;
9900 case SVt_PVGV:
9901 case SVt_PVLV:
9902 if (isGV_with_GP(sv)) {
9903 gv = MUTABLE_GV(sv);
9904 io = GvIO(gv);
9905 if (!io)
9906 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9907 HEKfARG(GvNAME_HEK(gv)));
9908 break;
9909 }
9910 /* FALLTHROUGH */
9911 default:
9912 if (!SvOK(sv))
9913 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9914 if (SvROK(sv)) {
9915 SvGETMAGIC(SvRV(sv));
9916 return sv_2io(SvRV(sv));
9917 }
9918 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9919 if (gv)
9920 io = GvIO(gv);
9921 else
9922 io = 0;
9923 if (!io) {
9924 SV *newsv = sv;
9925 if (SvGMAGICAL(sv)) {
9926 newsv = sv_newmortal();
9927 sv_setsv_nomg(newsv, sv);
9928 }
9929 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9930 }
9931 break;
9932 }
9933 return io;
9934 }
9935
9936 /*
9937 =for apidoc sv_2cv
9938
9939 Using various gambits, try to get a CV from an SV; in addition, try if
9940 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9941 The flags in C<lref> are passed to C<gv_fetchsv>.
9942
9943 =cut
9944 */
9945
9946 CV *
Perl_sv_2cv(pTHX_ SV * sv,HV ** const st,GV ** const gvp,const I32 lref)9947 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9948 {
9949 GV *gv = NULL;
9950 CV *cv = NULL;
9951
9952 PERL_ARGS_ASSERT_SV_2CV;
9953
9954 if (!sv) {
9955 *st = NULL;
9956 *gvp = NULL;
9957 return NULL;
9958 }
9959 switch (SvTYPE(sv)) {
9960 case SVt_PVCV:
9961 *st = CvSTASH(sv);
9962 *gvp = NULL;
9963 return MUTABLE_CV(sv);
9964 case SVt_PVHV:
9965 case SVt_PVAV:
9966 *st = NULL;
9967 *gvp = NULL;
9968 return NULL;
9969 default:
9970 SvGETMAGIC(sv);
9971 if (SvROK(sv)) {
9972 if (SvAMAGIC(sv))
9973 sv = amagic_deref_call(sv, to_cv_amg);
9974
9975 sv = SvRV(sv);
9976 if (SvTYPE(sv) == SVt_PVCV) {
9977 cv = MUTABLE_CV(sv);
9978 *gvp = NULL;
9979 *st = CvSTASH(cv);
9980 return cv;
9981 }
9982 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9983 gv = MUTABLE_GV(sv);
9984 else
9985 Perl_croak(aTHX_ "Not a subroutine reference");
9986 }
9987 else if (isGV_with_GP(sv)) {
9988 gv = MUTABLE_GV(sv);
9989 }
9990 else {
9991 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9992 }
9993 *gvp = gv;
9994 if (!gv) {
9995 *st = NULL;
9996 return NULL;
9997 }
9998 /* Some flags to gv_fetchsv mean don't really create the GV */
9999 if (!isGV_with_GP(gv)) {
10000 *st = NULL;
10001 return NULL;
10002 }
10003 *st = GvESTASH(gv);
10004 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10005 /* XXX this is probably not what they think they're getting.
10006 * It has the same effect as "sub name;", i.e. just a forward
10007 * declaration! */
10008 newSTUB(gv,0);
10009 }
10010 return GvCVu(gv);
10011 }
10012 }
10013
10014 /*
10015 =for apidoc sv_true
10016
10017 Returns true if the SV has a true value by Perl's rules.
10018 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10019 instead use an in-line version.
10020
10021 =cut
10022 */
10023
10024 I32
Perl_sv_true(pTHX_ SV * const sv)10025 Perl_sv_true(pTHX_ SV *const sv)
10026 {
10027 if (!sv)
10028 return 0;
10029 if (SvPOK(sv)) {
10030 const XPV* const tXpv = (XPV*)SvANY(sv);
10031 if (tXpv &&
10032 (tXpv->xpv_cur > 1 ||
10033 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10034 return 1;
10035 else
10036 return 0;
10037 }
10038 else {
10039 if (SvIOK(sv))
10040 return SvIVX(sv) != 0;
10041 else {
10042 if (SvNOK(sv))
10043 return SvNVX(sv) != 0.0;
10044 else
10045 return sv_2bool(sv);
10046 }
10047 }
10048 }
10049
10050 /*
10051 =for apidoc sv_pvn_force
10052
10053 Get a sensible string out of the SV somehow.
10054 A private implementation of the C<SvPV_force> macro for compilers which
10055 can't cope with complex macro expressions. Always use the macro instead.
10056
10057 =for apidoc sv_pvn_force_flags
10058
10059 Get a sensible string out of the SV somehow.
10060 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10061 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10062 implemented in terms of this function.
10063 You normally want to use the various wrapper macros instead: see
10064 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10065
10066 =cut
10067 */
10068
10069 char *
Perl_sv_pvn_force_flags(pTHX_ SV * const sv,STRLEN * const lp,const I32 flags)10070 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10071 {
10072 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10073
10074 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10075 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10076 sv_force_normal_flags(sv, 0);
10077
10078 if (SvPOK(sv)) {
10079 if (lp)
10080 *lp = SvCUR(sv);
10081 }
10082 else {
10083 char *s;
10084 STRLEN len;
10085
10086 if (SvTYPE(sv) > SVt_PVLV
10087 || isGV_with_GP(sv))
10088 /* diag_listed_as: Can't coerce %s to %s in %s */
10089 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10090 OP_DESC(PL_op));
10091 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10092 if (!s) {
10093 s = (char *)"";
10094 }
10095 if (lp)
10096 *lp = len;
10097
10098 if (SvTYPE(sv) < SVt_PV ||
10099 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10100 if (SvROK(sv))
10101 sv_unref(sv);
10102 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10103 SvGROW(sv, len + 1);
10104 Move(s,SvPVX(sv),len,char);
10105 SvCUR_set(sv, len);
10106 SvPVX(sv)[len] = '\0';
10107 }
10108 if (!SvPOK(sv)) {
10109 SvPOK_on(sv); /* validate pointer */
10110 SvTAINT(sv);
10111 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10112 PTR2UV(sv),SvPVX_const(sv)));
10113 }
10114 }
10115 (void)SvPOK_only_UTF8(sv);
10116 return SvPVX_mutable(sv);
10117 }
10118
10119 /*
10120 =for apidoc sv_pvbyten_force
10121
10122 The backend for the C<SvPVbytex_force> macro. Always use the macro
10123 instead.
10124
10125 =cut
10126 */
10127
10128 char *
Perl_sv_pvbyten_force(pTHX_ SV * const sv,STRLEN * const lp)10129 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10130 {
10131 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10132
10133 sv_pvn_force(sv,lp);
10134 sv_utf8_downgrade(sv,0);
10135 *lp = SvCUR(sv);
10136 return SvPVX(sv);
10137 }
10138
10139 /*
10140 =for apidoc sv_pvutf8n_force
10141
10142 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10143 instead.
10144
10145 =cut
10146 */
10147
10148 char *
Perl_sv_pvutf8n_force(pTHX_ SV * const sv,STRLEN * const lp)10149 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10150 {
10151 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10152
10153 sv_pvn_force(sv,0);
10154 sv_utf8_upgrade_nomg(sv);
10155 *lp = SvCUR(sv);
10156 return SvPVX(sv);
10157 }
10158
10159 /*
10160 =for apidoc sv_reftype
10161
10162 Returns a string describing what the SV is a reference to.
10163
10164 If ob is true and the SV is blessed, the string is the class name,
10165 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10166
10167 =cut
10168 */
10169
10170 const char *
Perl_sv_reftype(pTHX_ const SV * const sv,const int ob)10171 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10172 {
10173 PERL_ARGS_ASSERT_SV_REFTYPE;
10174 if (ob && SvOBJECT(sv)) {
10175 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10176 }
10177 else {
10178 /* WARNING - There is code, for instance in mg.c, that assumes that
10179 * the only reason that sv_reftype(sv,0) would return a string starting
10180 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10181 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10182 * this routine inside other subs, and it saves time.
10183 * Do not change this assumption without searching for "dodgy type check" in
10184 * the code.
10185 * - Yves */
10186 switch (SvTYPE(sv)) {
10187 case SVt_NULL:
10188 case SVt_IV:
10189 case SVt_NV:
10190 case SVt_PV:
10191 case SVt_PVIV:
10192 case SVt_PVNV:
10193 case SVt_PVMG:
10194 if (SvVOK(sv))
10195 return "VSTRING";
10196 if (SvROK(sv))
10197 return "REF";
10198 else
10199 return "SCALAR";
10200
10201 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10202 /* tied lvalues should appear to be
10203 * scalars for backwards compatibility */
10204 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10205 ? "SCALAR" : "LVALUE");
10206 case SVt_PVAV: return "ARRAY";
10207 case SVt_PVHV: return "HASH";
10208 case SVt_PVCV: return "CODE";
10209 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10210 ? "GLOB" : "SCALAR");
10211 case SVt_PVFM: return "FORMAT";
10212 case SVt_PVIO: return "IO";
10213 case SVt_INVLIST: return "INVLIST";
10214 case SVt_REGEXP: return "REGEXP";
10215 default: return "UNKNOWN";
10216 }
10217 }
10218 }
10219
10220 /*
10221 =for apidoc sv_ref
10222
10223 Returns a SV describing what the SV passed in is a reference to.
10224
10225 dst can be a SV to be set to the description or NULL, in which case a
10226 mortal SV is returned.
10227
10228 If ob is true and the SV is blessed, the description is the class
10229 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10230
10231 =cut
10232 */
10233
10234 SV *
Perl_sv_ref(pTHX_ SV * dst,const SV * const sv,const int ob)10235 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10236 {
10237 PERL_ARGS_ASSERT_SV_REF;
10238
10239 if (!dst)
10240 dst = sv_newmortal();
10241
10242 if (ob && SvOBJECT(sv)) {
10243 HvNAME_get(SvSTASH(sv))
10244 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10245 : sv_setpvs(dst, "__ANON__");
10246 }
10247 else {
10248 const char * reftype = sv_reftype(sv, 0);
10249 sv_setpv(dst, reftype);
10250 }
10251 return dst;
10252 }
10253
10254 /*
10255 =for apidoc sv_isobject
10256
10257 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10258 object. If the SV is not an RV, or if the object is not blessed, then this
10259 will return false.
10260
10261 =cut
10262 */
10263
10264 int
Perl_sv_isobject(pTHX_ SV * sv)10265 Perl_sv_isobject(pTHX_ SV *sv)
10266 {
10267 if (!sv)
10268 return 0;
10269 SvGETMAGIC(sv);
10270 if (!SvROK(sv))
10271 return 0;
10272 sv = SvRV(sv);
10273 if (!SvOBJECT(sv))
10274 return 0;
10275 return 1;
10276 }
10277
10278 /*
10279 =for apidoc sv_isa
10280
10281 Returns a boolean indicating whether the SV is blessed into the specified
10282 class. This does not check for subtypes; use C<sv_derived_from> to verify
10283 an inheritance relationship.
10284
10285 =cut
10286 */
10287
10288 int
Perl_sv_isa(pTHX_ SV * sv,const char * const name)10289 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10290 {
10291 const char *hvname;
10292
10293 PERL_ARGS_ASSERT_SV_ISA;
10294
10295 if (!sv)
10296 return 0;
10297 SvGETMAGIC(sv);
10298 if (!SvROK(sv))
10299 return 0;
10300 sv = SvRV(sv);
10301 if (!SvOBJECT(sv))
10302 return 0;
10303 hvname = HvNAME_get(SvSTASH(sv));
10304 if (!hvname)
10305 return 0;
10306
10307 return strEQ(hvname, name);
10308 }
10309
10310 /*
10311 =for apidoc newSVrv
10312
10313 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10314 RV then it will be upgraded to one. If C<classname> is non-null then the new
10315 SV will be blessed in the specified package. The new SV is returned and its
10316 reference count is 1. The reference count 1 is owned by C<rv>. See also
10317 newRV_inc() and newRV_noinc() for creating a new RV properly.
10318
10319 =cut
10320 */
10321
10322 SV*
Perl_newSVrv(pTHX_ SV * const rv,const char * const classname)10323 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10324 {
10325 SV *sv;
10326
10327 PERL_ARGS_ASSERT_NEWSVRV;
10328
10329 new_SV(sv);
10330
10331 SV_CHECK_THINKFIRST_COW_DROP(rv);
10332
10333 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10334 const U32 refcnt = SvREFCNT(rv);
10335 SvREFCNT(rv) = 0;
10336 sv_clear(rv);
10337 SvFLAGS(rv) = 0;
10338 SvREFCNT(rv) = refcnt;
10339
10340 sv_upgrade(rv, SVt_IV);
10341 } else if (SvROK(rv)) {
10342 SvREFCNT_dec(SvRV(rv));
10343 } else {
10344 prepare_SV_for_RV(rv);
10345 }
10346
10347 SvOK_off(rv);
10348 SvRV_set(rv, sv);
10349 SvROK_on(rv);
10350
10351 if (classname) {
10352 HV* const stash = gv_stashpv(classname, GV_ADD);
10353 (void)sv_bless(rv, stash);
10354 }
10355 return sv;
10356 }
10357
10358 SV *
Perl_newSVavdefelem(pTHX_ AV * av,SSize_t ix,bool extendible)10359 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10360 {
10361 SV * const lv = newSV_type(SVt_PVLV);
10362 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10363 LvTYPE(lv) = 'y';
10364 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10365 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10366 LvSTARGOFF(lv) = ix;
10367 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10368 return lv;
10369 }
10370
10371 /*
10372 =for apidoc sv_setref_pv
10373
10374 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10375 argument will be upgraded to an RV. That RV will be modified to point to
10376 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10377 into the SV. The C<classname> argument indicates the package for the
10378 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10379 will have a reference count of 1, and the RV will be returned.
10380
10381 Do not use with other Perl types such as HV, AV, SV, CV, because those
10382 objects will become corrupted by the pointer copy process.
10383
10384 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10385
10386 =cut
10387 */
10388
10389 SV*
Perl_sv_setref_pv(pTHX_ SV * const rv,const char * const classname,void * const pv)10390 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10391 {
10392 PERL_ARGS_ASSERT_SV_SETREF_PV;
10393
10394 if (!pv) {
10395 sv_set_undef(rv);
10396 SvSETMAGIC(rv);
10397 }
10398 else
10399 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10400 return rv;
10401 }
10402
10403 /*
10404 =for apidoc sv_setref_iv
10405
10406 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10407 argument will be upgraded to an RV. That RV will be modified to point to
10408 the new SV. The C<classname> argument indicates the package for the
10409 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10410 will have a reference count of 1, and the RV will be returned.
10411
10412 =cut
10413 */
10414
10415 SV*
Perl_sv_setref_iv(pTHX_ SV * const rv,const char * const classname,const IV iv)10416 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10417 {
10418 PERL_ARGS_ASSERT_SV_SETREF_IV;
10419
10420 sv_setiv(newSVrv(rv,classname), iv);
10421 return rv;
10422 }
10423
10424 /*
10425 =for apidoc sv_setref_uv
10426
10427 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10428 argument will be upgraded to an RV. That RV will be modified to point to
10429 the new SV. The C<classname> argument indicates the package for the
10430 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10431 will have a reference count of 1, and the RV will be returned.
10432
10433 =cut
10434 */
10435
10436 SV*
Perl_sv_setref_uv(pTHX_ SV * const rv,const char * const classname,const UV uv)10437 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10438 {
10439 PERL_ARGS_ASSERT_SV_SETREF_UV;
10440
10441 sv_setuv(newSVrv(rv,classname), uv);
10442 return rv;
10443 }
10444
10445 /*
10446 =for apidoc sv_setref_nv
10447
10448 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10449 argument will be upgraded to an RV. That RV will be modified to point to
10450 the new SV. The C<classname> argument indicates the package for the
10451 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10452 will have a reference count of 1, and the RV will be returned.
10453
10454 =cut
10455 */
10456
10457 SV*
Perl_sv_setref_nv(pTHX_ SV * const rv,const char * const classname,const NV nv)10458 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10459 {
10460 PERL_ARGS_ASSERT_SV_SETREF_NV;
10461
10462 sv_setnv(newSVrv(rv,classname), nv);
10463 return rv;
10464 }
10465
10466 /*
10467 =for apidoc sv_setref_pvn
10468
10469 Copies a string into a new SV, optionally blessing the SV. The length of the
10470 string must be specified with C<n>. The C<rv> argument will be upgraded to
10471 an RV. That RV will be modified to point to the new SV. The C<classname>
10472 argument indicates the package for the blessing. Set C<classname> to
10473 C<NULL> to avoid the blessing. The new SV will have a reference count
10474 of 1, and the RV will be returned.
10475
10476 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10477
10478 =cut
10479 */
10480
10481 SV*
Perl_sv_setref_pvn(pTHX_ SV * const rv,const char * const classname,const char * const pv,const STRLEN n)10482 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10483 const char *const pv, const STRLEN n)
10484 {
10485 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10486
10487 sv_setpvn(newSVrv(rv,classname), pv, n);
10488 return rv;
10489 }
10490
10491 /*
10492 =for apidoc sv_bless
10493
10494 Blesses an SV into a specified package. The SV must be an RV. The package
10495 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10496 of the SV is unaffected.
10497
10498 =cut
10499 */
10500
10501 SV*
Perl_sv_bless(pTHX_ SV * const sv,HV * const stash)10502 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10503 {
10504 SV *tmpRef;
10505 HV *oldstash = NULL;
10506
10507 PERL_ARGS_ASSERT_SV_BLESS;
10508
10509 SvGETMAGIC(sv);
10510 if (!SvROK(sv))
10511 Perl_croak(aTHX_ "Can't bless non-reference value");
10512 tmpRef = SvRV(sv);
10513 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10514 if (SvREADONLY(tmpRef))
10515 Perl_croak_no_modify();
10516 if (SvOBJECT(tmpRef)) {
10517 oldstash = SvSTASH(tmpRef);
10518 }
10519 }
10520 SvOBJECT_on(tmpRef);
10521 SvUPGRADE(tmpRef, SVt_PVMG);
10522 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10523 SvREFCNT_dec(oldstash);
10524
10525 if(SvSMAGICAL(tmpRef))
10526 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10527 mg_set(tmpRef);
10528
10529
10530
10531 return sv;
10532 }
10533
10534 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10535 * as it is after unglobbing it.
10536 */
10537
10538 PERL_STATIC_INLINE void
S_sv_unglob(pTHX_ SV * const sv,U32 flags)10539 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10540 {
10541 void *xpvmg;
10542 HV *stash;
10543 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10544
10545 PERL_ARGS_ASSERT_SV_UNGLOB;
10546
10547 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10548 SvFAKE_off(sv);
10549 if (!(flags & SV_COW_DROP_PV))
10550 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10551
10552 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10553 if (GvGP(sv)) {
10554 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10555 && HvNAME_get(stash))
10556 mro_method_changed_in(stash);
10557 gp_free(MUTABLE_GV(sv));
10558 }
10559 if (GvSTASH(sv)) {
10560 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10561 GvSTASH(sv) = NULL;
10562 }
10563 GvMULTI_off(sv);
10564 if (GvNAME_HEK(sv)) {
10565 unshare_hek(GvNAME_HEK(sv));
10566 }
10567 isGV_with_GP_off(sv);
10568
10569 if(SvTYPE(sv) == SVt_PVGV) {
10570 /* need to keep SvANY(sv) in the right arena */
10571 xpvmg = new_XPVMG();
10572 StructCopy(SvANY(sv), xpvmg, XPVMG);
10573 del_XPVGV(SvANY(sv));
10574 SvANY(sv) = xpvmg;
10575
10576 SvFLAGS(sv) &= ~SVTYPEMASK;
10577 SvFLAGS(sv) |= SVt_PVMG;
10578 }
10579
10580 /* Intentionally not calling any local SET magic, as this isn't so much a
10581 set operation as merely an internal storage change. */
10582 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10583 else sv_setsv_flags(sv, temp, 0);
10584
10585 if ((const GV *)sv == PL_last_in_gv)
10586 PL_last_in_gv = NULL;
10587 else if ((const GV *)sv == PL_statgv)
10588 PL_statgv = NULL;
10589 }
10590
10591 /*
10592 =for apidoc sv_unref_flags
10593
10594 Unsets the RV status of the SV, and decrements the reference count of
10595 whatever was being referenced by the RV. This can almost be thought of
10596 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10597 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10598 (otherwise the decrementing is conditional on the reference count being
10599 different from one or the reference being a readonly SV).
10600 See C<L</SvROK_off>>.
10601
10602 =cut
10603 */
10604
10605 void
Perl_sv_unref_flags(pTHX_ SV * const ref,const U32 flags)10606 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10607 {
10608 SV* const target = SvRV(ref);
10609
10610 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10611
10612 if (SvWEAKREF(ref)) {
10613 sv_del_backref(target, ref);
10614 SvWEAKREF_off(ref);
10615 SvRV_set(ref, NULL);
10616 return;
10617 }
10618 SvRV_set(ref, NULL);
10619 SvROK_off(ref);
10620 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10621 assigned to as BEGIN {$a = \"Foo"} will fail. */
10622 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10623 SvREFCNT_dec_NN(target);
10624 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10625 sv_2mortal(target); /* Schedule for freeing later */
10626 }
10627
10628 /*
10629 =for apidoc sv_untaint
10630
10631 Untaint an SV. Use C<SvTAINTED_off> instead.
10632
10633 =cut
10634 */
10635
10636 void
Perl_sv_untaint(pTHX_ SV * const sv)10637 Perl_sv_untaint(pTHX_ SV *const sv)
10638 {
10639 PERL_ARGS_ASSERT_SV_UNTAINT;
10640 PERL_UNUSED_CONTEXT;
10641
10642 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10643 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10644 if (mg)
10645 mg->mg_len &= ~1;
10646 }
10647 }
10648
10649 /*
10650 =for apidoc sv_tainted
10651
10652 Test an SV for taintedness. Use C<SvTAINTED> instead.
10653
10654 =cut
10655 */
10656
10657 bool
Perl_sv_tainted(pTHX_ SV * const sv)10658 Perl_sv_tainted(pTHX_ SV *const sv)
10659 {
10660 PERL_ARGS_ASSERT_SV_TAINTED;
10661 PERL_UNUSED_CONTEXT;
10662
10663 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10664 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10665 if (mg && (mg->mg_len & 1) )
10666 return TRUE;
10667 }
10668 return FALSE;
10669 }
10670
10671 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10672 private to this file */
10673
10674 /*
10675 =for apidoc sv_setpviv
10676
10677 Copies an integer into the given SV, also updating its string value.
10678 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10679
10680 =cut
10681 */
10682
10683 void
Perl_sv_setpviv(pTHX_ SV * const sv,const IV iv)10684 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10685 {
10686 /* The purpose of this union is to ensure that arr is aligned on
10687 a 2 byte boundary, because that is what uiv_2buf() requires */
10688 union {
10689 char arr[TYPE_CHARS(UV)];
10690 U16 dummy;
10691 } buf;
10692 char *ebuf;
10693 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10694
10695 PERL_ARGS_ASSERT_SV_SETPVIV;
10696
10697 sv_setpvn(sv, ptr, ebuf - ptr);
10698 }
10699
10700 /*
10701 =for apidoc sv_setpviv_mg
10702
10703 Like C<sv_setpviv>, but also handles 'set' magic.
10704
10705 =cut
10706 */
10707
10708 void
Perl_sv_setpviv_mg(pTHX_ SV * const sv,const IV iv)10709 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10710 {
10711 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10712
10713 sv_setpviv(sv, iv);
10714 SvSETMAGIC(sv);
10715 }
10716
10717 #endif /* NO_MATHOMS */
10718
10719 #if defined(PERL_IMPLICIT_CONTEXT)
10720
10721 /* pTHX_ magic can't cope with varargs, so this is a no-context
10722 * version of the main function, (which may itself be aliased to us).
10723 * Don't access this version directly.
10724 */
10725
10726 void
Perl_sv_setpvf_nocontext(SV * const sv,const char * const pat,...)10727 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10728 {
10729 dTHX;
10730 va_list args;
10731
10732 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10733
10734 va_start(args, pat);
10735 sv_vsetpvf(sv, pat, &args);
10736 va_end(args);
10737 }
10738
10739 /* pTHX_ magic can't cope with varargs, so this is a no-context
10740 * version of the main function, (which may itself be aliased to us).
10741 * Don't access this version directly.
10742 */
10743
10744 void
Perl_sv_setpvf_mg_nocontext(SV * const sv,const char * const pat,...)10745 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10746 {
10747 dTHX;
10748 va_list args;
10749
10750 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10751
10752 va_start(args, pat);
10753 sv_vsetpvf_mg(sv, pat, &args);
10754 va_end(args);
10755 }
10756 #endif
10757
10758 /*
10759 =for apidoc sv_setpvf
10760
10761 Works like C<sv_catpvf> but copies the text into the SV instead of
10762 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10763
10764 =cut
10765 */
10766
10767 void
Perl_sv_setpvf(pTHX_ SV * const sv,const char * const pat,...)10768 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10769 {
10770 va_list args;
10771
10772 PERL_ARGS_ASSERT_SV_SETPVF;
10773
10774 va_start(args, pat);
10775 sv_vsetpvf(sv, pat, &args);
10776 va_end(args);
10777 }
10778
10779 /*
10780 =for apidoc sv_vsetpvf
10781
10782 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10783 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10784
10785 Usually used via its frontend C<sv_setpvf>.
10786
10787 =cut
10788 */
10789
10790 void
Perl_sv_vsetpvf(pTHX_ SV * const sv,const char * const pat,va_list * const args)10791 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10792 {
10793 PERL_ARGS_ASSERT_SV_VSETPVF;
10794
10795 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10796 }
10797
10798 /*
10799 =for apidoc sv_setpvf_mg
10800
10801 Like C<sv_setpvf>, but also handles 'set' magic.
10802
10803 =cut
10804 */
10805
10806 void
Perl_sv_setpvf_mg(pTHX_ SV * const sv,const char * const pat,...)10807 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10808 {
10809 va_list args;
10810
10811 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10812
10813 va_start(args, pat);
10814 sv_vsetpvf_mg(sv, pat, &args);
10815 va_end(args);
10816 }
10817
10818 /*
10819 =for apidoc sv_vsetpvf_mg
10820
10821 Like C<sv_vsetpvf>, but also handles 'set' magic.
10822
10823 Usually used via its frontend C<sv_setpvf_mg>.
10824
10825 =cut
10826 */
10827
10828 void
Perl_sv_vsetpvf_mg(pTHX_ SV * const sv,const char * const pat,va_list * const args)10829 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10830 {
10831 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10832
10833 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10834 SvSETMAGIC(sv);
10835 }
10836
10837 #if defined(PERL_IMPLICIT_CONTEXT)
10838
10839 /* pTHX_ magic can't cope with varargs, so this is a no-context
10840 * version of the main function, (which may itself be aliased to us).
10841 * Don't access this version directly.
10842 */
10843
10844 void
Perl_sv_catpvf_nocontext(SV * const sv,const char * const pat,...)10845 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10846 {
10847 dTHX;
10848 va_list args;
10849
10850 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10851
10852 va_start(args, pat);
10853 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10854 va_end(args);
10855 }
10856
10857 /* pTHX_ magic can't cope with varargs, so this is a no-context
10858 * version of the main function, (which may itself be aliased to us).
10859 * Don't access this version directly.
10860 */
10861
10862 void
Perl_sv_catpvf_mg_nocontext(SV * const sv,const char * const pat,...)10863 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10864 {
10865 dTHX;
10866 va_list args;
10867
10868 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10869
10870 va_start(args, pat);
10871 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10872 SvSETMAGIC(sv);
10873 va_end(args);
10874 }
10875 #endif
10876
10877 /*
10878 =for apidoc sv_catpvf
10879
10880 Processes its arguments like C<sprintf>, and appends the formatted
10881 output to an SV. As with C<sv_vcatpvfn> called with a non-null C-style
10882 variable argument list, argument reordering is not supported.
10883 If the appended data contains "wide" characters
10884 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10885 and characters >255 formatted with C<%c>), the original SV might get
10886 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10887 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10888 valid UTF-8; if the original SV was bytes, the pattern should be too.
10889
10890 =cut */
10891
10892 void
Perl_sv_catpvf(pTHX_ SV * const sv,const char * const pat,...)10893 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10894 {
10895 va_list args;
10896
10897 PERL_ARGS_ASSERT_SV_CATPVF;
10898
10899 va_start(args, pat);
10900 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10901 va_end(args);
10902 }
10903
10904 /*
10905 =for apidoc sv_vcatpvf
10906
10907 Processes its arguments like C<sv_vcatpvfn> called with a non-null C-style
10908 variable argument list, and appends the formatted output
10909 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10910
10911 Usually used via its frontend C<sv_catpvf>.
10912
10913 =cut
10914 */
10915
10916 void
Perl_sv_vcatpvf(pTHX_ SV * const sv,const char * const pat,va_list * const args)10917 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10918 {
10919 PERL_ARGS_ASSERT_SV_VCATPVF;
10920
10921 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10922 }
10923
10924 /*
10925 =for apidoc sv_catpvf_mg
10926
10927 Like C<sv_catpvf>, but also handles 'set' magic.
10928
10929 =cut
10930 */
10931
10932 void
Perl_sv_catpvf_mg(pTHX_ SV * const sv,const char * const pat,...)10933 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10934 {
10935 va_list args;
10936
10937 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10938
10939 va_start(args, pat);
10940 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10941 SvSETMAGIC(sv);
10942 va_end(args);
10943 }
10944
10945 /*
10946 =for apidoc sv_vcatpvf_mg
10947
10948 Like C<sv_vcatpvf>, but also handles 'set' magic.
10949
10950 Usually used via its frontend C<sv_catpvf_mg>.
10951
10952 =cut
10953 */
10954
10955 void
Perl_sv_vcatpvf_mg(pTHX_ SV * const sv,const char * const pat,va_list * const args)10956 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10957 {
10958 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10959
10960 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10961 SvSETMAGIC(sv);
10962 }
10963
10964 /*
10965 =for apidoc sv_vsetpvfn
10966
10967 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10968 appending it.
10969
10970 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10971
10972 =cut
10973 */
10974
10975 void
Perl_sv_vsetpvfn(pTHX_ SV * const sv,const char * const pat,const STRLEN patlen,va_list * const args,SV ** const svargs,const Size_t sv_count,bool * const maybe_tainted)10976 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10977 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10978 {
10979 PERL_ARGS_ASSERT_SV_VSETPVFN;
10980
10981 SvPVCLEAR(sv);
10982 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10983 }
10984
10985
10986 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10987
10988 PERL_STATIC_INLINE void
S_sv_catpvn_simple(pTHX_ SV * const sv,const char * const buf,const STRLEN len)10989 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10990 {
10991 STRLEN const need = len + SvCUR(sv) + 1;
10992 char *end;
10993
10994 /* can't wrap as both len and SvCUR() are allocated in
10995 * memory and together can't consume all the address space
10996 */
10997 assert(need > len);
10998
10999 assert(SvPOK(sv));
11000 SvGROW(sv, need);
11001 end = SvEND(sv);
11002 Copy(buf, end, len, char);
11003 end += len;
11004 *end = '\0';
11005 SvCUR_set(sv, need - 1);
11006 }
11007
11008
11009 /*
11010 * Warn of missing argument to sprintf. The value used in place of such
11011 * arguments should be &PL_sv_no; an undefined value would yield
11012 * inappropriate "use of uninit" warnings [perl #71000].
11013 */
11014 STATIC void
S_warn_vcatpvfn_missing_argument(pTHX)11015 S_warn_vcatpvfn_missing_argument(pTHX) {
11016 if (ckWARN(WARN_MISSING)) {
11017 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11018 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11019 }
11020 }
11021
11022
11023 static void
S_croak_overflow()11024 S_croak_overflow()
11025 {
11026 dTHX;
11027 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11028 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11029 }
11030
11031
11032 /* Given an int i from the next arg (if args is true) or an sv from an arg
11033 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11034 * with overflow checking.
11035 * Sets *neg to true if the value was negative (untouched otherwise.
11036 * Returns the absolute value.
11037 * As an extra margin of safety, it croaks if the returned value would
11038 * exceed the maximum value of a STRLEN / 4.
11039 */
11040
11041 static STRLEN
S_sprintf_arg_num_val(pTHX_ va_list * const args,int i,SV * sv,bool * neg)11042 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11043 {
11044 IV iv;
11045
11046 if (args) {
11047 iv = i;
11048 goto do_iv;
11049 }
11050
11051 if (!sv)
11052 return 0;
11053
11054 SvGETMAGIC(sv);
11055
11056 if (UNLIKELY(SvIsUV(sv))) {
11057 UV uv = SvUV_nomg(sv);
11058 if (uv > IV_MAX)
11059 S_croak_overflow();
11060 iv = uv;
11061 }
11062 else {
11063 iv = SvIV_nomg(sv);
11064 do_iv:
11065 if (iv < 0) {
11066 if (iv < -IV_MAX)
11067 S_croak_overflow();
11068 iv = -iv;
11069 *neg = TRUE;
11070 }
11071 }
11072
11073 if (iv > (IV)(((STRLEN)~0) / 4))
11074 S_croak_overflow();
11075
11076 return (STRLEN)iv;
11077 }
11078
11079 /* Read in and return a number. Updates *pattern to point to the char
11080 * following the number. Expects the first char to 1..9.
11081 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11082 * This is a belt-and-braces safety measure to complement any
11083 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11084 * It means that e.g. on a 32-bit system the width/precision can't be more
11085 * than 1G, which seems reasonable.
11086 */
11087
11088 STATIC STRLEN
S_expect_number(pTHX_ const char ** const pattern)11089 S_expect_number(pTHX_ const char **const pattern)
11090 {
11091 STRLEN var;
11092
11093 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11094
11095 assert(inRANGE(**pattern, '1', '9'));
11096
11097 var = *(*pattern)++ - '0';
11098 while (isDIGIT(**pattern)) {
11099 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11100 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11101 S_croak_overflow();
11102 var = var * 10 + (*(*pattern)++ - '0');
11103 }
11104 return var;
11105 }
11106
11107 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11108 * ensures it's big enough), back fill it with the rounded integer part of
11109 * nv. Returns ptr to start of string, and sets *len to its length.
11110 * Returns NULL if not convertible.
11111 */
11112
11113 STATIC char *
S_F0convert(NV nv,char * const endbuf,STRLEN * const len)11114 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11115 {
11116 const int neg = nv < 0;
11117 UV uv;
11118
11119 PERL_ARGS_ASSERT_F0CONVERT;
11120
11121 assert(!Perl_isinfnan(nv));
11122 if (neg)
11123 nv = -nv;
11124 if (nv != 0.0 && nv < UV_MAX) {
11125 char *p = endbuf;
11126 uv = (UV)nv;
11127 if (uv != nv) {
11128 nv += 0.5;
11129 uv = (UV)nv;
11130 if (uv & 1 && uv == nv)
11131 uv--; /* Round to even */
11132 }
11133 do {
11134 const unsigned dig = uv % 10;
11135 *--p = '0' + dig;
11136 } while (uv /= 10);
11137 if (neg)
11138 *--p = '-';
11139 *len = endbuf - p;
11140 return p;
11141 }
11142 return NULL;
11143 }
11144
11145
11146 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11147
11148 void
Perl_sv_vcatpvfn(pTHX_ SV * const sv,const char * const pat,const STRLEN patlen,va_list * const args,SV ** const svargs,const Size_t sv_count,bool * const maybe_tainted)11149 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11150 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11151 {
11152 PERL_ARGS_ASSERT_SV_VCATPVFN;
11153
11154 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11155 }
11156
11157
11158 /* For the vcatpvfn code, we need a long double target in case
11159 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11160 * with long double formats, even without NV being long double. But we
11161 * call the target 'fv' instead of 'nv', since most of the time it is not
11162 * (most compilers these days recognize "long double", even if only as a
11163 * synonym for "double").
11164 */
11165 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11166 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11167 # define VCATPVFN_FV_GF PERL_PRIgldbl
11168 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11169 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11170 # define VCATPVFN_NV_TO_FV(nv,fv) \
11171 STMT_START { \
11172 double _dv = nv; \
11173 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11174 } STMT_END
11175 # else
11176 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11177 # endif
11178 typedef long double vcatpvfn_long_double_t;
11179 #else
11180 # define VCATPVFN_FV_GF NVgf
11181 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11182 typedef NV vcatpvfn_long_double_t;
11183 #endif
11184
11185 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11186 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11187 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11188 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11189 * after the first 1023 zero bits.
11190 *
11191 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11192 * of dynamically growing buffer might be better, start at just 16 bytes
11193 * (for example) and grow only when necessary. Or maybe just by looking
11194 * at the exponents of the two doubles? */
11195 # define DOUBLEDOUBLE_MAXBITS 2098
11196 #endif
11197
11198 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11199 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11200 * per xdigit. For the double-double case, this can be rather many.
11201 * The non-double-double-long-double overshoots since all bits of NV
11202 * are not mantissa bits, there are also exponent bits. */
11203 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11204 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11205 #else
11206 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11207 #endif
11208
11209 /* If we do not have a known long double format, (including not using
11210 * long doubles, or long doubles being equal to doubles) then we will
11211 * fall back to the ldexp/frexp route, with which we can retrieve at
11212 * most as many bits as our widest unsigned integer type is. We try
11213 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11214 *
11215 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11216 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11217 */
11218 #if defined(HAS_QUAD) && defined(Uquad_t)
11219 # define MANTISSATYPE Uquad_t
11220 # define MANTISSASIZE 8
11221 #else
11222 # define MANTISSATYPE UV
11223 # define MANTISSASIZE UVSIZE
11224 #endif
11225
11226 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11227 # define HEXTRACT_LITTLE_ENDIAN
11228 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11229 # define HEXTRACT_BIG_ENDIAN
11230 #else
11231 # define HEXTRACT_MIX_ENDIAN
11232 #endif
11233
11234 /* S_hextract() is a helper for S_format_hexfp, for extracting
11235 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11236 * are being extracted from (either directly from the long double in-memory
11237 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11238 * is used to update the exponent. The subnormal is set to true
11239 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11240 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11241 *
11242 * The tricky part is that S_hextract() needs to be called twice:
11243 * the first time with vend as NULL, and the second time with vend as
11244 * the pointer returned by the first call. What happens is that on
11245 * the first round the output size is computed, and the intended
11246 * extraction sanity checked. On the second round the actual output
11247 * (the extraction of the hexadecimal values) takes place.
11248 * Sanity failures cause fatal failures during both rounds. */
11249 STATIC U8*
S_hextract(pTHX_ const NV nv,int * exponent,bool * subnormal,U8 * vhex,U8 * vend)11250 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11251 U8* vhex, U8* vend)
11252 {
11253 U8* v = vhex;
11254 int ix;
11255 int ixmin = 0, ixmax = 0;
11256
11257 /* XXX Inf/NaN are not handled here, since it is
11258 * assumed they are to be output as "Inf" and "NaN". */
11259
11260 /* These macros are just to reduce typos, they have multiple
11261 * repetitions below, but usually only one (or sometimes two)
11262 * of them is really being used. */
11263 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11264 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11265 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11266 #define HEXTRACT_OUTPUT(ix) \
11267 STMT_START { \
11268 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11269 } STMT_END
11270 #define HEXTRACT_COUNT(ix, c) \
11271 STMT_START { \
11272 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11273 } STMT_END
11274 #define HEXTRACT_BYTE(ix) \
11275 STMT_START { \
11276 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11277 } STMT_END
11278 #define HEXTRACT_LO_NYBBLE(ix) \
11279 STMT_START { \
11280 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11281 } STMT_END
11282 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11283 * to make it look less odd when the top bits of a NV
11284 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11285 * order bits can be in the "low nybble" of a byte. */
11286 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11287 #define HEXTRACT_BYTES_LE(a, b) \
11288 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11289 #define HEXTRACT_BYTES_BE(a, b) \
11290 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11291 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11292 #define HEXTRACT_IMPLICIT_BIT(nv) \
11293 STMT_START { \
11294 if (!*subnormal) { \
11295 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11296 } \
11297 } STMT_END
11298
11299 /* Most formats do. Those which don't should undef this.
11300 *
11301 * But also note that IEEE 754 subnormals do not have it, or,
11302 * expressed alternatively, their implicit bit is zero. */
11303 #define HEXTRACT_HAS_IMPLICIT_BIT
11304
11305 /* Many formats do. Those which don't should undef this. */
11306 #define HEXTRACT_HAS_TOP_NYBBLE
11307
11308 /* HEXTRACTSIZE is the maximum number of xdigits. */
11309 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11310 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11311 #else
11312 # define HEXTRACTSIZE 2 * NVSIZE
11313 #endif
11314
11315 const U8* vmaxend = vhex + HEXTRACTSIZE;
11316
11317 assert(HEXTRACTSIZE <= VHEX_SIZE);
11318
11319 PERL_UNUSED_VAR(ix); /* might happen */
11320 (void)Perl_frexp(PERL_ABS(nv), exponent);
11321 *subnormal = FALSE;
11322 if (vend && (vend <= vhex || vend > vmaxend)) {
11323 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11324 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11325 }
11326 {
11327 /* First check if using long doubles. */
11328 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11329 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11330 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11331 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11332 /* The bytes 13..0 are the mantissa/fraction,
11333 * the 15,14 are the sign+exponent. */
11334 const U8* nvp = (const U8*)(&nv);
11335 HEXTRACT_GET_SUBNORMAL(nv);
11336 HEXTRACT_IMPLICIT_BIT(nv);
11337 # undef HEXTRACT_HAS_TOP_NYBBLE
11338 HEXTRACT_BYTES_LE(13, 0);
11339 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11340 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11341 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11342 /* The bytes 2..15 are the mantissa/fraction,
11343 * the 0,1 are the sign+exponent. */
11344 const U8* nvp = (const U8*)(&nv);
11345 HEXTRACT_GET_SUBNORMAL(nv);
11346 HEXTRACT_IMPLICIT_BIT(nv);
11347 # undef HEXTRACT_HAS_TOP_NYBBLE
11348 HEXTRACT_BYTES_BE(2, 15);
11349 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11350 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11351 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11352 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11353 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11354 /* The bytes 0..1 are the sign+exponent,
11355 * the bytes 2..9 are the mantissa/fraction. */
11356 const U8* nvp = (const U8*)(&nv);
11357 # undef HEXTRACT_HAS_IMPLICIT_BIT
11358 # undef HEXTRACT_HAS_TOP_NYBBLE
11359 HEXTRACT_GET_SUBNORMAL(nv);
11360 HEXTRACT_BYTES_LE(7, 0);
11361 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11362 /* Does this format ever happen? (Wikipedia says the Motorola
11363 * 6888x math coprocessors used format _like_ this but padded
11364 * to 96 bits with 16 unused bits between the exponent and the
11365 * mantissa.) */
11366 const U8* nvp = (const U8*)(&nv);
11367 # undef HEXTRACT_HAS_IMPLICIT_BIT
11368 # undef HEXTRACT_HAS_TOP_NYBBLE
11369 HEXTRACT_GET_SUBNORMAL(nv);
11370 HEXTRACT_BYTES_BE(0, 7);
11371 # else
11372 # define HEXTRACT_FALLBACK
11373 /* Double-double format: two doubles next to each other.
11374 * The first double is the high-order one, exactly like
11375 * it would be for a "lone" double. The second double
11376 * is shifted down using the exponent so that that there
11377 * are no common bits. The tricky part is that the value
11378 * of the double-double is the SUM of the two doubles and
11379 * the second one can be also NEGATIVE.
11380 *
11381 * Because of this tricky construction the bytewise extraction we
11382 * use for the other long double formats doesn't work, we must
11383 * extract the values bit by bit.
11384 *
11385 * The little-endian double-double is used .. somewhere?
11386 *
11387 * The big endian double-double is used in e.g. PPC/Power (AIX)
11388 * and MIPS (SGI).
11389 *
11390 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11391 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11392 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11393 */
11394 # endif
11395 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11396 /* Using normal doubles, not long doubles.
11397 *
11398 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11399 * bytes, since we might need to handle printf precision, and
11400 * also need to insert the radix. */
11401 # if NVSIZE == 8
11402 # ifdef HEXTRACT_LITTLE_ENDIAN
11403 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11404 const U8* nvp = (const U8*)(&nv);
11405 HEXTRACT_GET_SUBNORMAL(nv);
11406 HEXTRACT_IMPLICIT_BIT(nv);
11407 HEXTRACT_TOP_NYBBLE(6);
11408 HEXTRACT_BYTES_LE(5, 0);
11409 # elif defined(HEXTRACT_BIG_ENDIAN)
11410 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11411 const U8* nvp = (const U8*)(&nv);
11412 HEXTRACT_GET_SUBNORMAL(nv);
11413 HEXTRACT_IMPLICIT_BIT(nv);
11414 HEXTRACT_TOP_NYBBLE(1);
11415 HEXTRACT_BYTES_BE(2, 7);
11416 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11417 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11418 const U8* nvp = (const U8*)(&nv);
11419 HEXTRACT_GET_SUBNORMAL(nv);
11420 HEXTRACT_IMPLICIT_BIT(nv);
11421 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11422 HEXTRACT_BYTE(1); /* 5 */
11423 HEXTRACT_BYTE(0); /* 4 */
11424 HEXTRACT_BYTE(7); /* 3 */
11425 HEXTRACT_BYTE(6); /* 2 */
11426 HEXTRACT_BYTE(5); /* 1 */
11427 HEXTRACT_BYTE(4); /* 0 */
11428 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11429 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11430 const U8* nvp = (const U8*)(&nv);
11431 HEXTRACT_GET_SUBNORMAL(nv);
11432 HEXTRACT_IMPLICIT_BIT(nv);
11433 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11434 HEXTRACT_BYTE(6); /* 5 */
11435 HEXTRACT_BYTE(7); /* 4 */
11436 HEXTRACT_BYTE(0); /* 3 */
11437 HEXTRACT_BYTE(1); /* 2 */
11438 HEXTRACT_BYTE(2); /* 1 */
11439 HEXTRACT_BYTE(3); /* 0 */
11440 # else
11441 # define HEXTRACT_FALLBACK
11442 # endif
11443 # else
11444 # define HEXTRACT_FALLBACK
11445 # endif
11446 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11447
11448 #ifdef HEXTRACT_FALLBACK
11449 HEXTRACT_GET_SUBNORMAL(nv);
11450 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11451 /* The fallback is used for the double-double format, and
11452 * for unknown long double formats, and for unknown double
11453 * formats, or in general unknown NV formats. */
11454 if (nv == (NV)0.0) {
11455 if (vend)
11456 *v++ = 0;
11457 else
11458 v++;
11459 *exponent = 0;
11460 }
11461 else {
11462 NV d = nv < 0 ? -nv : nv;
11463 NV e = (NV)1.0;
11464 U8 ha = 0x0; /* hexvalue accumulator */
11465 U8 hd = 0x8; /* hexvalue digit */
11466
11467 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11468 * this is essentially manual frexp(). Multiplying by 0.5 and
11469 * doubling should be lossless in binary floating point. */
11470
11471 *exponent = 1;
11472
11473 while (e > d) {
11474 e *= (NV)0.5;
11475 (*exponent)--;
11476 }
11477 /* Now d >= e */
11478
11479 while (d >= e + e) {
11480 e += e;
11481 (*exponent)++;
11482 }
11483 /* Now e <= d < 2*e */
11484
11485 /* First extract the leading hexdigit (the implicit bit). */
11486 if (d >= e) {
11487 d -= e;
11488 if (vend)
11489 *v++ = 1;
11490 else
11491 v++;
11492 }
11493 else {
11494 if (vend)
11495 *v++ = 0;
11496 else
11497 v++;
11498 }
11499 e *= (NV)0.5;
11500
11501 /* Then extract the remaining hexdigits. */
11502 while (d > (NV)0.0) {
11503 if (d >= e) {
11504 ha |= hd;
11505 d -= e;
11506 }
11507 if (hd == 1) {
11508 /* Output or count in groups of four bits,
11509 * that is, when the hexdigit is down to one. */
11510 if (vend)
11511 *v++ = ha;
11512 else
11513 v++;
11514 /* Reset the hexvalue. */
11515 ha = 0x0;
11516 hd = 0x8;
11517 }
11518 else
11519 hd >>= 1;
11520 e *= (NV)0.5;
11521 }
11522
11523 /* Flush possible pending hexvalue. */
11524 if (ha) {
11525 if (vend)
11526 *v++ = ha;
11527 else
11528 v++;
11529 }
11530 }
11531 #endif
11532 }
11533 /* Croak for various reasons: if the output pointer escaped the
11534 * output buffer, if the extraction index escaped the extraction
11535 * buffer, or if the ending output pointer didn't match the
11536 * previously computed value. */
11537 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11538 /* For double-double the ixmin and ixmax stay at zero,
11539 * which is convenient since the HEXTRACTSIZE is tricky
11540 * for double-double. */
11541 ixmin < 0 || ixmax >= NVSIZE ||
11542 (vend && v != vend)) {
11543 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11544 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11545 }
11546 return v;
11547 }
11548
11549
11550 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11551 *
11552 * Processes the %a/%A hexadecimal floating-point format, since the
11553 * built-in snprintf()s which are used for most of the f/p formats, don't
11554 * universally handle %a/%A.
11555 * Populates buf of length bufsize, and returns the length of the created
11556 * string.
11557 * The rest of the args have the same meaning as the local vars of the
11558 * same name within Perl_sv_vcatpvfn_flags().
11559 *
11560 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11561 *
11562 * It requires the caller to make buf large enough.
11563 */
11564
11565 static STRLEN
S_format_hexfp(pTHX_ char * const buf,const STRLEN bufsize,const char c,const NV nv,const vcatpvfn_long_double_t fv,bool has_precis,STRLEN precis,STRLEN width,bool alt,char plus,bool left,bool fill)11566 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11567 const NV nv, const vcatpvfn_long_double_t fv,
11568 bool has_precis, STRLEN precis, STRLEN width,
11569 bool alt, char plus, bool left, bool fill)
11570 {
11571 /* Hexadecimal floating point. */
11572 char* p = buf;
11573 U8 vhex[VHEX_SIZE];
11574 U8* v = vhex; /* working pointer to vhex */
11575 U8* vend; /* pointer to one beyond last digit of vhex */
11576 U8* vfnz = NULL; /* first non-zero */
11577 U8* vlnz = NULL; /* last non-zero */
11578 U8* v0 = NULL; /* first output */
11579 const bool lower = (c == 'a');
11580 /* At output the values of vhex (up to vend) will
11581 * be mapped through the xdig to get the actual
11582 * human-readable xdigits. */
11583 const char* xdig = PL_hexdigit;
11584 STRLEN zerotail = 0; /* how many extra zeros to append */
11585 int exponent = 0; /* exponent of the floating point input */
11586 bool hexradix = FALSE; /* should we output the radix */
11587 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11588 bool negative = FALSE;
11589 STRLEN elen;
11590
11591 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11592 *
11593 * For example with denormals, (assuming the vanilla
11594 * 64-bit double): the exponent is zero. 1xp-1074 is
11595 * the smallest denormal and the smallest double, it
11596 * could be output also as 0x0.0000000000001p-1022 to
11597 * match its internal structure. */
11598
11599 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11600 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11601
11602 #if NVSIZE > DOUBLESIZE
11603 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11604 /* In this case there is an implicit bit,
11605 * and therefore the exponent is shifted by one. */
11606 exponent--;
11607 # elif defined(NV_X86_80_BIT)
11608 if (subnormal) {
11609 /* The subnormals of the x86-80 have a base exponent of -16382,
11610 * (while the physical exponent bits are zero) but the frexp()
11611 * returned the scientific-style floating exponent. We want
11612 * to map the last one as:
11613 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11614 * -16835..-16388 -> -16384
11615 * since we want to keep the first hexdigit
11616 * as one of the [8421]. */
11617 exponent = -4 * ( (exponent + 1) / -4) - 2;
11618 } else {
11619 exponent -= 4;
11620 }
11621 /* TBD: other non-implicit-bit platforms than the x86-80. */
11622 # endif
11623 #endif
11624
11625 negative = fv < 0 || Perl_signbit(nv);
11626 if (negative)
11627 *p++ = '-';
11628 else if (plus)
11629 *p++ = plus;
11630 *p++ = '0';
11631 if (lower) {
11632 *p++ = 'x';
11633 }
11634 else {
11635 *p++ = 'X';
11636 xdig += 16; /* Use uppercase hex. */
11637 }
11638
11639 /* Find the first non-zero xdigit. */
11640 for (v = vhex; v < vend; v++) {
11641 if (*v) {
11642 vfnz = v;
11643 break;
11644 }
11645 }
11646
11647 if (vfnz) {
11648 /* Find the last non-zero xdigit. */
11649 for (v = vend - 1; v >= vhex; v--) {
11650 if (*v) {
11651 vlnz = v;
11652 break;
11653 }
11654 }
11655
11656 #if NVSIZE == DOUBLESIZE
11657 if (fv != 0.0)
11658 exponent--;
11659 #endif
11660
11661 if (subnormal) {
11662 #ifndef NV_X86_80_BIT
11663 if (vfnz[0] > 1) {
11664 /* IEEE 754 subnormals (but not the x86 80-bit):
11665 * we want "normalize" the subnormal,
11666 * so we need to right shift the hex nybbles
11667 * so that the output of the subnormal starts
11668 * from the first true bit. (Another, equally
11669 * valid, policy would be to dump the subnormal
11670 * nybbles as-is, to display the "physical" layout.) */
11671 int i, n;
11672 U8 *vshr;
11673 /* Find the ceil(log2(v[0])) of
11674 * the top non-zero nybble. */
11675 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11676 assert(n < 4);
11677 assert(vlnz);
11678 vlnz[1] = 0;
11679 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11680 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11681 vshr[0] >>= n;
11682 }
11683 if (vlnz[1]) {
11684 vlnz++;
11685 }
11686 }
11687 #endif
11688 v0 = vfnz;
11689 } else {
11690 v0 = vhex;
11691 }
11692
11693 if (has_precis) {
11694 U8* ve = (subnormal ? vlnz + 1 : vend);
11695 SSize_t vn = ve - v0;
11696 assert(vn >= 1);
11697 if (precis < (Size_t)(vn - 1)) {
11698 bool overflow = FALSE;
11699 if (v0[precis + 1] < 0x8) {
11700 /* Round down, nothing to do. */
11701 } else if (v0[precis + 1] > 0x8) {
11702 /* Round up. */
11703 v0[precis]++;
11704 overflow = v0[precis] > 0xF;
11705 v0[precis] &= 0xF;
11706 } else { /* v0[precis] == 0x8 */
11707 /* Half-point: round towards the one
11708 * with the even least-significant digit:
11709 * 08 -> 0 88 -> 8
11710 * 18 -> 2 98 -> a
11711 * 28 -> 2 a8 -> a
11712 * 38 -> 4 b8 -> c
11713 * 48 -> 4 c8 -> c
11714 * 58 -> 6 d8 -> e
11715 * 68 -> 6 e8 -> e
11716 * 78 -> 8 f8 -> 10 */
11717 if ((v0[precis] & 0x1)) {
11718 v0[precis]++;
11719 }
11720 overflow = v0[precis] > 0xF;
11721 v0[precis] &= 0xF;
11722 }
11723
11724 if (overflow) {
11725 for (v = v0 + precis - 1; v >= v0; v--) {
11726 (*v)++;
11727 overflow = *v > 0xF;
11728 (*v) &= 0xF;
11729 if (!overflow) {
11730 break;
11731 }
11732 }
11733 if (v == v0 - 1 && overflow) {
11734 /* If the overflow goes all the
11735 * way to the front, we need to
11736 * insert 0x1 in front, and adjust
11737 * the exponent. */
11738 Move(v0, v0 + 1, vn - 1, char);
11739 *v0 = 0x1;
11740 exponent += 4;
11741 }
11742 }
11743
11744 /* The new effective "last non zero". */
11745 vlnz = v0 + precis;
11746 }
11747 else {
11748 zerotail =
11749 subnormal ? precis - vn + 1 :
11750 precis - (vlnz - vhex);
11751 }
11752 }
11753
11754 v = v0;
11755 *p++ = xdig[*v++];
11756
11757 /* If there are non-zero xdigits, the radix
11758 * is output after the first one. */
11759 if (vfnz < vlnz) {
11760 hexradix = TRUE;
11761 }
11762 }
11763 else {
11764 *p++ = '0';
11765 exponent = 0;
11766 zerotail = has_precis ? precis : 0;
11767 }
11768
11769 /* The radix is always output if precis, or if alt. */
11770 if ((has_precis && precis > 0) || alt) {
11771 hexradix = TRUE;
11772 }
11773
11774 if (hexradix) {
11775 #ifndef USE_LOCALE_NUMERIC
11776 *p++ = '.';
11777 #else
11778 if (IN_LC(LC_NUMERIC)) {
11779 STRLEN n;
11780 const char* r = SvPV(PL_numeric_radix_sv, n);
11781 Copy(r, p, n, char);
11782 p += n;
11783 }
11784 else {
11785 *p++ = '.';
11786 }
11787 #endif
11788 }
11789
11790 if (vlnz) {
11791 while (v <= vlnz)
11792 *p++ = xdig[*v++];
11793 }
11794
11795 if (zerotail > 0) {
11796 while (zerotail--) {
11797 *p++ = '0';
11798 }
11799 }
11800
11801 elen = p - buf;
11802
11803 /* sanity checks */
11804 if (elen >= bufsize || width >= bufsize)
11805 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11806 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11807
11808 elen += my_snprintf(p, bufsize - elen,
11809 "%c%+d", lower ? 'p' : 'P',
11810 exponent);
11811
11812 if (elen < width) {
11813 STRLEN gap = (STRLEN)(width - elen);
11814 if (left) {
11815 /* Pad the back with spaces. */
11816 memset(buf + elen, ' ', gap);
11817 }
11818 else if (fill) {
11819 /* Insert the zeros after the "0x" and the
11820 * the potential sign, but before the digits,
11821 * otherwise we end up with "0000xH.HHH...",
11822 * when we want "0x000H.HHH..." */
11823 STRLEN nzero = gap;
11824 char* zerox = buf + 2;
11825 STRLEN nmove = elen - 2;
11826 if (negative || plus) {
11827 zerox++;
11828 nmove--;
11829 }
11830 Move(zerox, zerox + nzero, nmove, char);
11831 memset(zerox, fill ? '0' : ' ', nzero);
11832 }
11833 else {
11834 /* Move it to the right. */
11835 Move(buf, buf + gap,
11836 elen, char);
11837 /* Pad the front with spaces. */
11838 memset(buf, ' ', gap);
11839 }
11840 elen = width;
11841 }
11842 return elen;
11843 }
11844
11845
11846 /*
11847 =for apidoc sv_vcatpvfn
11848
11849 =for apidoc sv_vcatpvfn_flags
11850
11851 Processes its arguments like C<vsprintf> and appends the formatted output
11852 to an SV. Uses an array of SVs if the C-style variable argument list is
11853 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11854 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11855 C<va_list> argument list with a format string that uses argument reordering
11856 will yield an exception.
11857
11858 When running with taint checks enabled, indicates via
11859 C<maybe_tainted> if results are untrustworthy (often due to the use of
11860 locales).
11861
11862 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11863
11864 It assumes that pat has the same utf8-ness as sv. It's the caller's
11865 responsibility to ensure that this is so.
11866
11867 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11868
11869 =cut
11870 */
11871
11872
11873 void
Perl_sv_vcatpvfn_flags(pTHX_ SV * const sv,const char * const pat,const STRLEN patlen,va_list * const args,SV ** const svargs,const Size_t sv_count,bool * const maybe_tainted,const U32 flags)11874 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11875 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11876 const U32 flags)
11877 {
11878 const char *fmtstart; /* character following the current '%' */
11879 const char *q; /* current position within format */
11880 const char *patend;
11881 STRLEN origlen;
11882 Size_t svix = 0;
11883 static const char nullstr[] = "(null)";
11884 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11885 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11886 /* Times 4: a decimal digit takes more than 3 binary digits.
11887 * NV_DIG: mantissa takes that many decimal digits.
11888 * Plus 32: Playing safe. */
11889 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11890 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11891 #ifdef USE_LOCALE_NUMERIC
11892 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11893 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11894 #endif
11895
11896 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11897 PERL_UNUSED_ARG(maybe_tainted);
11898
11899 if (flags & SV_GMAGIC)
11900 SvGETMAGIC(sv);
11901
11902 /* no matter what, this is a string now */
11903 (void)SvPV_force_nomg(sv, origlen);
11904
11905 /* the code that scans for flags etc following a % relies on
11906 * a '\0' being present to avoid falling off the end. Ideally that
11907 * should be fixed */
11908 assert(pat[patlen] == '\0');
11909
11910
11911 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11912 * In each case, if there isn't the correct number of args, instead
11913 * fall through to the main code to handle the issuing of any
11914 * warnings etc.
11915 */
11916
11917 if (patlen == 0 && (args || sv_count == 0))
11918 return;
11919
11920 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11921
11922 /* "%s" */
11923 if (patlen == 2 && pat[1] == 's') {
11924 if (args) {
11925 const char * const s = va_arg(*args, char*);
11926 sv_catpv_nomg(sv, s ? s : nullstr);
11927 }
11928 else {
11929 /* we want get magic on the source but not the target.
11930 * sv_catsv can't do that, though */
11931 SvGETMAGIC(*svargs);
11932 sv_catsv_nomg(sv, *svargs);
11933 }
11934 return;
11935 }
11936
11937 /* "%-p" */
11938 if (args) {
11939 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11940 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11941 sv_catsv_nomg(sv, asv);
11942 return;
11943 }
11944 }
11945 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11946 /* special-case "%.0f" */
11947 else if ( patlen == 4
11948 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11949 {
11950 const NV nv = SvNV(*svargs);
11951 if (LIKELY(!Perl_isinfnan(nv))) {
11952 STRLEN l;
11953 char *p;
11954
11955 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11956 sv_catpvn_nomg(sv, p, l);
11957 return;
11958 }
11959 }
11960 }
11961 #endif /* !USE_LONG_DOUBLE */
11962 }
11963
11964
11965 patend = (char*)pat + patlen;
11966 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11967 char intsize = 0; /* size qualifier in "%hi..." etc */
11968 bool alt = FALSE; /* has "%#..." */
11969 bool left = FALSE; /* has "%-..." */
11970 bool fill = FALSE; /* has "%0..." */
11971 char plus = 0; /* has "%+..." */
11972 STRLEN width = 0; /* value of "%NNN..." */
11973 bool has_precis = FALSE; /* has "%.NNN..." */
11974 STRLEN precis = 0; /* value of "%.NNN..." */
11975 int base = 0; /* base to print in, e.g. 8 for %o */
11976 UV uv = 0; /* the value to print of int-ish args */
11977
11978 bool vectorize = FALSE; /* has "%v..." */
11979 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11980 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11981 STRLEN veclen = 0; /* SvCUR(vec arg) */
11982 const char *dotstr = NULL; /* separator string for %v */
11983 STRLEN dotstrlen; /* length of separator string for %v */
11984
11985 Size_t efix = 0; /* explicit format parameter index */
11986 const Size_t osvix = svix; /* original index in case of bad fmt */
11987
11988 SV *argsv = NULL;
11989 bool is_utf8 = FALSE; /* is this item utf8? */
11990 bool arg_missing = FALSE; /* give "Missing argument" warning */
11991 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11992 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11993 STRLEN zeros = 0; /* how many '0' to prepend */
11994
11995 const char *eptr = NULL; /* the address of the element string */
11996 STRLEN elen = 0; /* the length of the element string */
11997
11998 char c; /* the actual format ('d', s' etc) */
11999
12000
12001 /* echo everything up to the next format specification */
12002 for (q = fmtstart; q < patend && *q != '%'; ++q)
12003 {};
12004
12005 if (q > fmtstart) {
12006 if (has_utf8 && !pat_utf8) {
12007 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12008 * the fly */
12009 const char *p;
12010 char *dst;
12011 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12012
12013 for (p = fmtstart; p < q; p++)
12014 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12015 need++;
12016 SvGROW(sv, need);
12017
12018 dst = SvEND(sv);
12019 for (p = fmtstart; p < q; p++)
12020 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12021 *dst = '\0';
12022 SvCUR_set(sv, need - 1);
12023 }
12024 else
12025 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12026 }
12027 if (q++ >= patend)
12028 break;
12029
12030 fmtstart = q; /* fmtstart is char following the '%' */
12031
12032 /*
12033 We allow format specification elements in this order:
12034 \d+\$ explicit format parameter index
12035 [-+ 0#]+ flags
12036 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12037 0 flag (as above): repeated to allow "v02"
12038 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12039 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12040 [hlqLV] size
12041 [%bcdefginopsuxDFOUX] format (mandatory)
12042 */
12043
12044 if (inRANGE(*q, '1', '9')) {
12045 width = expect_number(&q);
12046 if (*q == '$') {
12047 if (args)
12048 Perl_croak_nocontext(
12049 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12050 ++q;
12051 efix = (Size_t)width;
12052 width = 0;
12053 no_redundant_warning = TRUE;
12054 } else {
12055 goto gotwidth;
12056 }
12057 }
12058
12059 /* FLAGS */
12060
12061 while (*q) {
12062 switch (*q) {
12063 case ' ':
12064 case '+':
12065 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12066 q++;
12067 else
12068 plus = *q++;
12069 continue;
12070
12071 case '-':
12072 left = TRUE;
12073 q++;
12074 continue;
12075
12076 case '0':
12077 fill = TRUE;
12078 q++;
12079 continue;
12080
12081 case '#':
12082 alt = TRUE;
12083 q++;
12084 continue;
12085
12086 default:
12087 break;
12088 }
12089 break;
12090 }
12091
12092 /* at this point we can expect one of:
12093 *
12094 * 123 an explicit width
12095 * * width taken from next arg
12096 * *12$ width taken from 12th arg
12097 * or no width
12098 *
12099 * But any width specification may be preceded by a v, in one of its
12100 * forms:
12101 * v
12102 * *v
12103 * *12$v
12104 * So an asterisk may be either a width specifier or a vector
12105 * separator arg specifier, and we don't know which initially
12106 */
12107
12108 tryasterisk:
12109 if (*q == '*') {
12110 STRLEN ix; /* explicit width/vector separator index */
12111 q++;
12112 if (inRANGE(*q, '1', '9')) {
12113 ix = expect_number(&q);
12114 if (*q++ == '$') {
12115 if (args)
12116 Perl_croak_nocontext(
12117 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12118 no_redundant_warning = TRUE;
12119 } else
12120 goto unknown;
12121 }
12122 else
12123 ix = 0;
12124
12125 if (*q == 'v') {
12126 SV *vecsv;
12127 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12128 * with the default "." */
12129 q++;
12130 if (vectorize)
12131 goto unknown;
12132 if (args)
12133 vecsv = va_arg(*args, SV*);
12134 else {
12135 ix = ix ? ix - 1 : svix++;
12136 vecsv = ix < sv_count ? svargs[ix]
12137 : (arg_missing = TRUE, &PL_sv_no);
12138 }
12139 dotstr = SvPV_const(vecsv, dotstrlen);
12140 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12141 bad with tied or overloaded values that return UTF8. */
12142 if (DO_UTF8(vecsv))
12143 is_utf8 = TRUE;
12144 else if (has_utf8) {
12145 vecsv = sv_mortalcopy(vecsv);
12146 sv_utf8_upgrade(vecsv);
12147 dotstr = SvPV_const(vecsv, dotstrlen);
12148 is_utf8 = TRUE;
12149 }
12150 vectorize = TRUE;
12151 goto tryasterisk;
12152 }
12153
12154 /* the asterisk specified a width */
12155 {
12156 int i = 0;
12157 SV *sv = NULL;
12158 if (args)
12159 i = va_arg(*args, int);
12160 else {
12161 ix = ix ? ix - 1 : svix++;
12162 sv = (ix < sv_count) ? svargs[ix]
12163 : (arg_missing = TRUE, (SV*)NULL);
12164 }
12165 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12166 }
12167 }
12168 else if (*q == 'v') {
12169 q++;
12170 if (vectorize)
12171 goto unknown;
12172 vectorize = TRUE;
12173 dotstr = ".";
12174 dotstrlen = 1;
12175 goto tryasterisk;
12176
12177 }
12178 else {
12179 /* explicit width? */
12180 if(*q == '0') {
12181 fill = TRUE;
12182 q++;
12183 }
12184 if (inRANGE(*q, '1', '9'))
12185 width = expect_number(&q);
12186 }
12187
12188 gotwidth:
12189
12190 /* PRECISION */
12191
12192 if (*q == '.') {
12193 q++;
12194 if (*q == '*') {
12195 STRLEN ix; /* explicit precision index */
12196 q++;
12197 if (inRANGE(*q, '1', '9')) {
12198 ix = expect_number(&q);
12199 if (*q++ == '$') {
12200 if (args)
12201 Perl_croak_nocontext(
12202 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12203 no_redundant_warning = TRUE;
12204 } else
12205 goto unknown;
12206 }
12207 else
12208 ix = 0;
12209
12210 {
12211 int i = 0;
12212 SV *sv = NULL;
12213 bool neg = FALSE;
12214
12215 if (args)
12216 i = va_arg(*args, int);
12217 else {
12218 ix = ix ? ix - 1 : svix++;
12219 sv = (ix < sv_count) ? svargs[ix]
12220 : (arg_missing = TRUE, (SV*)NULL);
12221 }
12222 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12223 has_precis = !neg;
12224 /* ignore negative precision */
12225 if (!has_precis)
12226 precis = 0;
12227 }
12228 }
12229 else {
12230 /* although it doesn't seem documented, this code has long
12231 * behaved so that:
12232 * no digits following the '.' is treated like '.0'
12233 * the number may be preceded by any number of zeroes,
12234 * e.g. "%.0001f", which is the same as "%.1f"
12235 * so I've kept that behaviour. DAPM May 2017
12236 */
12237 while (*q == '0')
12238 q++;
12239 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0;
12240 has_precis = TRUE;
12241 }
12242 }
12243
12244 /* SIZE */
12245
12246 switch (*q) {
12247 #ifdef WIN32
12248 case 'I': /* Ix, I32x, and I64x */
12249 # ifdef USE_64_BIT_INT
12250 if (q[1] == '6' && q[2] == '4') {
12251 q += 3;
12252 intsize = 'q';
12253 break;
12254 }
12255 # endif
12256 if (q[1] == '3' && q[2] == '2') {
12257 q += 3;
12258 break;
12259 }
12260 # ifdef USE_64_BIT_INT
12261 intsize = 'q';
12262 # endif
12263 q++;
12264 break;
12265 #endif
12266 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12267 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12268 case 'L': /* Ld */
12269 /* FALLTHROUGH */
12270 # ifdef USE_QUADMATH
12271 case 'Q':
12272 /* FALLTHROUGH */
12273 # endif
12274 # if IVSIZE >= 8
12275 case 'q': /* qd */
12276 # endif
12277 intsize = 'q';
12278 q++;
12279 break;
12280 #endif
12281 case 'l':
12282 ++q;
12283 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12284 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12285 if (*q == 'l') { /* lld, llf */
12286 intsize = 'q';
12287 ++q;
12288 }
12289 else
12290 #endif
12291 intsize = 'l';
12292 break;
12293 case 'h':
12294 if (*++q == 'h') { /* hhd, hhu */
12295 intsize = 'c';
12296 ++q;
12297 }
12298 else
12299 intsize = 'h';
12300 break;
12301 case 'V':
12302 case 'z':
12303 case 't':
12304 case 'j':
12305 intsize = *q++;
12306 break;
12307 }
12308
12309 /* CONVERSION */
12310
12311 c = *q++; /* c now holds the conversion type */
12312
12313 /* '%' doesn't have an arg, so skip arg processing */
12314 if (c == '%') {
12315 eptr = q - 1;
12316 elen = 1;
12317 if (vectorize)
12318 goto unknown;
12319 goto string;
12320 }
12321
12322 if (vectorize && !strchr("BbDdiOouUXx", c))
12323 goto unknown;
12324
12325 /* get next arg (individual branches do their own va_arg()
12326 * handling for the args case) */
12327
12328 if (!args) {
12329 efix = efix ? efix - 1 : svix++;
12330 argsv = efix < sv_count ? svargs[efix]
12331 : (arg_missing = TRUE, &PL_sv_no);
12332 }
12333
12334
12335 switch (c) {
12336
12337 /* STRINGS */
12338
12339 case 's':
12340 if (args) {
12341 eptr = va_arg(*args, char*);
12342 if (eptr)
12343 if (has_precis)
12344 elen = my_strnlen(eptr, precis);
12345 else
12346 elen = strlen(eptr);
12347 else {
12348 eptr = (char *)nullstr;
12349 elen = sizeof nullstr - 1;
12350 }
12351 }
12352 else {
12353 eptr = SvPV_const(argsv, elen);
12354 if (DO_UTF8(argsv)) {
12355 STRLEN old_precis = precis;
12356 if (has_precis && precis < elen) {
12357 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12358 STRLEN p = precis > ulen ? ulen : precis;
12359 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12360 /* sticks at end */
12361 }
12362 if (width) { /* fudge width (can't fudge elen) */
12363 if (has_precis && precis < elen)
12364 width += precis - old_precis;
12365 else
12366 width +=
12367 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12368 }
12369 is_utf8 = TRUE;
12370 }
12371 }
12372
12373 string:
12374 if (has_precis && precis < elen)
12375 elen = precis;
12376 break;
12377
12378 /* INTEGERS */
12379
12380 case 'p':
12381 if (alt)
12382 goto unknown;
12383
12384 /* %p extensions:
12385 *
12386 * "%...p" is normally treated like "%...x", except that the
12387 * number to print is the SV's address (or a pointer address
12388 * for C-ish sprintf).
12389 *
12390 * However, the C-ish sprintf variant allows a few special
12391 * extensions. These are currently:
12392 *
12393 * %-p (SVf) Like %s, but gets the string from an SV*
12394 * arg rather than a char* arg.
12395 * (This was previously %_).
12396 *
12397 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12398 *
12399 * %2p (HEKf) Like %s, but using the key string in a HEK
12400 *
12401 * %3p (HEKf256) Ditto but like %.256s
12402 *
12403 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12404 * (cBOOL(utf8), len, string_buf).
12405 * It's handled by the "case 'd'" branch
12406 * rather than here.
12407 *
12408 * %<num>p where num is 1 or > 4: reserved for future
12409 * extensions. Warns, but then is treated as a
12410 * general %p (print hex address) format.
12411 */
12412
12413 if ( args
12414 && !intsize
12415 && !fill
12416 && !plus
12417 && !has_precis
12418 /* not %*p or %*1$p - any width was explicit */
12419 && q[-2] != '*'
12420 && q[-2] != '$'
12421 ) {
12422 if (left) { /* %-p (SVf), %-NNNp */
12423 if (width) {
12424 precis = width;
12425 has_precis = TRUE;
12426 }
12427 argsv = MUTABLE_SV(va_arg(*args, void*));
12428 eptr = SvPV_const(argsv, elen);
12429 if (DO_UTF8(argsv))
12430 is_utf8 = TRUE;
12431 width = 0;
12432 goto string;
12433 }
12434 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12435 HEK * const hek = va_arg(*args, HEK *);
12436 eptr = HEK_KEY(hek);
12437 elen = HEK_LEN(hek);
12438 if (HEK_UTF8(hek))
12439 is_utf8 = TRUE;
12440 if (width == 3) {
12441 precis = 256;
12442 has_precis = TRUE;
12443 }
12444 width = 0;
12445 goto string;
12446 }
12447 else if (width) {
12448 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12449 "internal %%<num>p might conflict with future printf extensions");
12450 }
12451 }
12452
12453 /* treat as normal %...p */
12454
12455 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12456 base = 16;
12457 goto do_integer;
12458
12459 case 'c':
12460 /* Ignore any size specifiers, since they're not documented as
12461 * being allowed for %c (ideally we should warn on e.g. '%hc').
12462 * Setting a default intsize, along with a positive
12463 * (which signals unsigned) base, causes, for C-ish use, the
12464 * va_arg to be interpreted as as unsigned int, when it's
12465 * actually signed, which will convert -ve values to high +ve
12466 * values. Note that unlike the libc %c, values > 255 will
12467 * convert to high unicode points rather than being truncated
12468 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12469 * will again convert -ve args to high -ve values.
12470 */
12471 intsize = 0;
12472 base = 1; /* special value that indicates we're doing a 'c' */
12473 goto get_int_arg_val;
12474
12475 case 'D':
12476 #ifdef IV_IS_QUAD
12477 intsize = 'q';
12478 #else
12479 intsize = 'l';
12480 #endif
12481 base = -10;
12482 goto get_int_arg_val;
12483
12484 case 'd':
12485 /* probably just a plain %d, but it might be the start of the
12486 * special UTF8f format, which usually looks something like
12487 * "%d%lu%4p" (the lu may vary by platform)
12488 */
12489 assert((UTF8f)[0] == 'd');
12490 assert((UTF8f)[1] == '%');
12491
12492 if ( args /* UTF8f only valid for C-ish sprintf */
12493 && q == fmtstart + 1 /* plain %d, not %....d */
12494 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12495 && *q == '%'
12496 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12497 {
12498 /* The argument has already gone through cBOOL, so the cast
12499 is safe. */
12500 is_utf8 = (bool)va_arg(*args, int);
12501 elen = va_arg(*args, UV);
12502 /* if utf8 length is larger than 0x7ffff..., then it might
12503 * have been a signed value that wrapped */
12504 if (elen > ((~(STRLEN)0) >> 1)) {
12505 assert(0); /* in DEBUGGING build we want to crash */
12506 elen = 0; /* otherwise we want to treat this as an empty string */
12507 }
12508 eptr = va_arg(*args, char *);
12509 q += sizeof(UTF8f) - 2;
12510 goto string;
12511 }
12512
12513 /* FALLTHROUGH */
12514 case 'i':
12515 base = -10;
12516 goto get_int_arg_val;
12517
12518 case 'U':
12519 #ifdef IV_IS_QUAD
12520 intsize = 'q';
12521 #else
12522 intsize = 'l';
12523 #endif
12524 /* FALLTHROUGH */
12525 case 'u':
12526 base = 10;
12527 goto get_int_arg_val;
12528
12529 case 'B':
12530 case 'b':
12531 base = 2;
12532 goto get_int_arg_val;
12533
12534 case 'O':
12535 #ifdef IV_IS_QUAD
12536 intsize = 'q';
12537 #else
12538 intsize = 'l';
12539 #endif
12540 /* FALLTHROUGH */
12541 case 'o':
12542 base = 8;
12543 goto get_int_arg_val;
12544
12545 case 'X':
12546 case 'x':
12547 base = 16;
12548
12549 get_int_arg_val:
12550
12551 if (vectorize) {
12552 STRLEN ulen;
12553 SV *vecsv;
12554
12555 if (base < 0) {
12556 base = -base;
12557 if (plus)
12558 esignbuf[esignlen++] = plus;
12559 }
12560
12561 /* initialise the vector string to iterate over */
12562
12563 vecsv = args ? va_arg(*args, SV*) : argsv;
12564
12565 /* if this is a version object, we need to convert
12566 * back into v-string notation and then let the
12567 * vectorize happen normally
12568 */
12569 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12570 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12571 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12572 "vector argument not supported with alpha versions");
12573 vecsv = &PL_sv_no;
12574 }
12575 else {
12576 vecstr = (U8*)SvPV_const(vecsv,veclen);
12577 vecsv = sv_newmortal();
12578 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12579 vecsv);
12580 }
12581 }
12582 vecstr = (U8*)SvPV_const(vecsv, veclen);
12583 vec_utf8 = DO_UTF8(vecsv);
12584
12585 /* This is the re-entry point for when we're iterating
12586 * over the individual characters of a vector arg */
12587 vector:
12588 if (!veclen)
12589 goto done_valid_conversion;
12590 if (vec_utf8)
12591 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12592 UTF8_ALLOW_ANYUV);
12593 else {
12594 uv = *vecstr;
12595 ulen = 1;
12596 }
12597 vecstr += ulen;
12598 veclen -= ulen;
12599 }
12600 else {
12601 /* test arg for inf/nan. This can trigger an unwanted
12602 * 'str' overload, so manually force 'num' overload first
12603 * if necessary */
12604 if (argsv) {
12605 SvGETMAGIC(argsv);
12606 if (UNLIKELY(SvAMAGIC(argsv)))
12607 argsv = sv_2num(argsv);
12608 if (UNLIKELY(isinfnansv(argsv)))
12609 goto handle_infnan_argsv;
12610 }
12611
12612 if (base < 0) {
12613 /* signed int type */
12614 IV iv;
12615 base = -base;
12616 if (args) {
12617 switch (intsize) {
12618 case 'c': iv = (char)va_arg(*args, int); break;
12619 case 'h': iv = (short)va_arg(*args, int); break;
12620 case 'l': iv = va_arg(*args, long); break;
12621 case 'V': iv = va_arg(*args, IV); break;
12622 case 'z': iv = va_arg(*args, SSize_t); break;
12623 #ifdef HAS_PTRDIFF_T
12624 case 't': iv = va_arg(*args, ptrdiff_t); break;
12625 #endif
12626 default: iv = va_arg(*args, int); break;
12627 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break;
12628 case 'q':
12629 #if IVSIZE >= 8
12630 iv = va_arg(*args, Quad_t); break;
12631 #else
12632 goto unknown;
12633 #endif
12634 }
12635 }
12636 else {
12637 /* assign to tiv then cast to iv to work around
12638 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12639 IV tiv = SvIV_nomg(argsv);
12640 switch (intsize) {
12641 case 'c': iv = (char)tiv; break;
12642 case 'h': iv = (short)tiv; break;
12643 case 'l': iv = (long)tiv; break;
12644 case 'V':
12645 default: iv = tiv; break;
12646 case 'q':
12647 #if IVSIZE >= 8
12648 iv = (Quad_t)tiv; break;
12649 #else
12650 goto unknown;
12651 #endif
12652 }
12653 }
12654
12655 /* now convert iv to uv */
12656 if (iv >= 0) {
12657 uv = iv;
12658 if (plus)
12659 esignbuf[esignlen++] = plus;
12660 }
12661 else {
12662 /* Using 0- here to silence bogus warning from MS VC */
12663 uv = (UV) (0 - (UV) iv);
12664 esignbuf[esignlen++] = '-';
12665 }
12666 }
12667 else {
12668 /* unsigned int type */
12669 if (args) {
12670 switch (intsize) {
12671 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12672 break;
12673 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12674 break;
12675 case 'l': uv = va_arg(*args, unsigned long); break;
12676 case 'V': uv = va_arg(*args, UV); break;
12677 case 'z': uv = va_arg(*args, Size_t); break;
12678 #ifdef HAS_PTRDIFF_T
12679 /* will sign extend, but there is no
12680 * uptrdiff_t, so oh well */
12681 case 't': uv = va_arg(*args, ptrdiff_t); break;
12682 #endif
12683 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break;
12684 default: uv = va_arg(*args, unsigned); break;
12685 case 'q':
12686 #if IVSIZE >= 8
12687 uv = va_arg(*args, Uquad_t); break;
12688 #else
12689 goto unknown;
12690 #endif
12691 }
12692 }
12693 else {
12694 /* assign to tiv then cast to iv to work around
12695 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12696 UV tuv = SvUV_nomg(argsv);
12697 switch (intsize) {
12698 case 'c': uv = (unsigned char)tuv; break;
12699 case 'h': uv = (unsigned short)tuv; break;
12700 case 'l': uv = (unsigned long)tuv; break;
12701 case 'V':
12702 default: uv = tuv; break;
12703 case 'q':
12704 #if IVSIZE >= 8
12705 uv = (Uquad_t)tuv; break;
12706 #else
12707 goto unknown;
12708 #endif
12709 }
12710 }
12711 }
12712 }
12713
12714 do_integer:
12715 {
12716 char *ptr = ebuf + sizeof ebuf;
12717 unsigned dig;
12718 zeros = 0;
12719
12720 switch (base) {
12721 case 16:
12722 {
12723 const char * const p =
12724 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12725
12726 do {
12727 dig = uv & 15;
12728 *--ptr = p[dig];
12729 } while (uv >>= 4);
12730 if (alt && *ptr != '0') {
12731 esignbuf[esignlen++] = '0';
12732 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12733 }
12734 break;
12735 }
12736 case 8:
12737 do {
12738 dig = uv & 7;
12739 *--ptr = '0' + dig;
12740 } while (uv >>= 3);
12741 if (alt && *ptr != '0')
12742 *--ptr = '0';
12743 break;
12744 case 2:
12745 do {
12746 dig = uv & 1;
12747 *--ptr = '0' + dig;
12748 } while (uv >>= 1);
12749 if (alt && *ptr != '0') {
12750 esignbuf[esignlen++] = '0';
12751 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12752 }
12753 break;
12754
12755 case 1:
12756 /* special-case: base 1 indicates a 'c' format:
12757 * we use the common code for extracting a uv,
12758 * but handle that value differently here than
12759 * all the other int types */
12760 if ((uv > 255 ||
12761 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12762 && !IN_BYTES)
12763 {
12764 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12765 eptr = ebuf;
12766 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12767 is_utf8 = TRUE;
12768 }
12769 else {
12770 eptr = ebuf;
12771 ebuf[0] = (char)uv;
12772 elen = 1;
12773 }
12774 goto string;
12775
12776 default: /* it had better be ten or less */
12777 do {
12778 dig = uv % base;
12779 *--ptr = '0' + dig;
12780 } while (uv /= base);
12781 break;
12782 }
12783 elen = (ebuf + sizeof ebuf) - ptr;
12784 eptr = ptr;
12785 if (has_precis) {
12786 if (precis > elen)
12787 zeros = precis - elen;
12788 else if (precis == 0 && elen == 1 && *eptr == '0'
12789 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12790 elen = 0;
12791
12792 /* a precision nullifies the 0 flag. */
12793 fill = FALSE;
12794 }
12795 }
12796 break;
12797
12798 /* FLOATING POINT */
12799
12800 case 'F':
12801 c = 'f'; /* maybe %F isn't supported here */
12802 /* FALLTHROUGH */
12803 case 'e': case 'E':
12804 case 'f':
12805 case 'g': case 'G':
12806 case 'a': case 'A':
12807
12808 {
12809 STRLEN float_need; /* what PL_efloatsize needs to become */
12810 bool hexfp; /* hexadecimal floating point? */
12811
12812 vcatpvfn_long_double_t fv;
12813 NV nv;
12814
12815 /* This is evil, but floating point is even more evil */
12816
12817 /* for SV-style calling, we can only get NV
12818 for C-style calling, we assume %f is double;
12819 for simplicity we allow any of %Lf, %llf, %qf for long double
12820 */
12821 switch (intsize) {
12822 case 'V':
12823 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12824 intsize = 'q';
12825 #endif
12826 break;
12827 /* [perl #20339] - we should accept and ignore %lf rather than die */
12828 case 'l':
12829 /* FALLTHROUGH */
12830 default:
12831 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12832 intsize = args ? 0 : 'q';
12833 #endif
12834 break;
12835 case 'q':
12836 #if defined(HAS_LONG_DOUBLE)
12837 break;
12838 #else
12839 /* FALLTHROUGH */
12840 #endif
12841 case 'c':
12842 case 'h':
12843 case 'z':
12844 case 't':
12845 case 'j':
12846 goto unknown;
12847 }
12848
12849 /* Now we need (long double) if intsize == 'q', else (double). */
12850 if (args) {
12851 /* Note: do not pull NVs off the va_list with va_arg()
12852 * (pull doubles instead) because if you have a build
12853 * with long doubles, you would always be pulling long
12854 * doubles, which would badly break anyone using only
12855 * doubles (i.e. the majority of builds). In other
12856 * words, you cannot mix doubles and long doubles.
12857 * The only case where you can pull off long doubles
12858 * is when the format specifier explicitly asks so with
12859 * e.g. "%Lg". */
12860 #ifdef USE_QUADMATH
12861 fv = intsize == 'q' ?
12862 va_arg(*args, NV) : va_arg(*args, double);
12863 nv = fv;
12864 #elif LONG_DOUBLESIZE > DOUBLESIZE
12865 if (intsize == 'q') {
12866 fv = va_arg(*args, long double);
12867 nv = fv;
12868 } else {
12869 nv = va_arg(*args, double);
12870 VCATPVFN_NV_TO_FV(nv, fv);
12871 }
12872 #else
12873 nv = va_arg(*args, double);
12874 fv = nv;
12875 #endif
12876 }
12877 else
12878 {
12879 SvGETMAGIC(argsv);
12880 /* we jump here if an int-ish format encountered an
12881 * infinite/Nan argsv. After setting nv/fv, it falls
12882 * into the isinfnan block which follows */
12883 handle_infnan_argsv:
12884 nv = SvNV_nomg(argsv);
12885 VCATPVFN_NV_TO_FV(nv, fv);
12886 }
12887
12888 if (Perl_isinfnan(nv)) {
12889 if (c == 'c')
12890 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12891 SvNV_nomg(argsv), (int)c);
12892
12893 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12894 assert(elen);
12895 eptr = ebuf;
12896 zeros = 0;
12897 esignlen = 0;
12898 dotstrlen = 0;
12899 break;
12900 }
12901
12902 /* special-case "%.0f" */
12903 if ( c == 'f'
12904 && !precis
12905 && has_precis
12906 && !(width || left || plus || alt)
12907 && !fill
12908 && intsize != 'q'
12909 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12910 )
12911 goto float_concat;
12912
12913 /* Determine the buffer size needed for the various
12914 * floating-point formats.
12915 *
12916 * The basic possibilities are:
12917 *
12918 * <---P--->
12919 * %f 1111111.123456789
12920 * %e 1.111111123e+06
12921 * %a 0x1.0f4471f9bp+20
12922 * %g 1111111.12
12923 * %g 1.11111112e+15
12924 *
12925 * where P is the value of the precision in the format, or 6
12926 * if not specified. Note the two possible output formats of
12927 * %g; in both cases the number of significant digits is <=
12928 * precision.
12929 *
12930 * For most of the format types the maximum buffer size needed
12931 * is precision, plus: any leading 1 or 0x1, the radix
12932 * point, and an exponent. The difficult one is %f: for a
12933 * large positive exponent it can have many leading digits,
12934 * which needs to be calculated specially. Also %a is slightly
12935 * different in that in the absence of a specified precision,
12936 * it uses as many digits as necessary to distinguish
12937 * different values.
12938 *
12939 * First, here are the constant bits. For ease of calculation
12940 * we over-estimate the needed buffer size, for example by
12941 * assuming all formats have an exponent and a leading 0x1.
12942 *
12943 * Also for production use, add a little extra overhead for
12944 * safety's sake. Under debugging don't, as it means we're
12945 * more likely to quickly spot issues during development.
12946 */
12947
12948 float_need = 1 /* possible unary minus */
12949 + 4 /* "0x1" plus very unlikely carry */
12950 + 1 /* default radix point '.' */
12951 + 2 /* "e-", "p+" etc */
12952 + 6 /* exponent: up to 16383 (quad fp) */
12953 #ifndef DEBUGGING
12954 + 20 /* safety net */
12955 #endif
12956 + 1; /* \0 */
12957
12958
12959 /* determine the radix point len, e.g. length(".") in "1.2" */
12960 #ifdef USE_LOCALE_NUMERIC
12961 /* note that we may either explicitly use PL_numeric_radix_sv
12962 * below, or implicitly, via an snprintf() variant.
12963 * Note also things like ps_AF.utf8 which has
12964 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12965 if (!lc_numeric_set) {
12966 /* only set once and reuse in-locale value on subsequent
12967 * iterations.
12968 * XXX what happens if we die in an eval?
12969 */
12970 STORE_LC_NUMERIC_SET_TO_NEEDED();
12971 lc_numeric_set = TRUE;
12972 }
12973
12974 if (IN_LC(LC_NUMERIC)) {
12975 /* this can't wrap unless PL_numeric_radix_sv is a string
12976 * consuming virtually all the 32-bit or 64-bit address
12977 * space
12978 */
12979 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12980
12981 /* floating-point formats only get utf8 if the radix point
12982 * is utf8. All other characters in the string are < 128
12983 * and so can be safely appended to both a non-utf8 and utf8
12984 * string as-is.
12985 * Note that this will convert the output to utf8 even if
12986 * the radix point didn't get output.
12987 */
12988 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12989 sv_utf8_upgrade(sv);
12990 has_utf8 = TRUE;
12991 }
12992 }
12993 #endif
12994
12995 hexfp = FALSE;
12996
12997 if (isALPHA_FOLD_EQ(c, 'f')) {
12998 /* Determine how many digits before the radix point
12999 * might be emitted. frexp() (or frexpl) has some
13000 * unspecified behaviour for nan/inf/-inf, so lucky we've
13001 * already handled them above */
13002 STRLEN digits;
13003 int i = PERL_INT_MIN;
13004 (void)Perl_frexp((NV)fv, &i);
13005 if (i == PERL_INT_MIN)
13006 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13007
13008 if (i > 0) {
13009 digits = BIT_DIGITS(i);
13010 /* this can't overflow. 'digits' will only be a few
13011 * thousand even for the largest floating-point types.
13012 * And up until now float_need is just some small
13013 * constants plus radix len, which can't be in
13014 * overflow territory unless the radix SV is consuming
13015 * over 1/2 the address space */
13016 assert(float_need < ((STRLEN)~0) - digits);
13017 float_need += digits;
13018 }
13019 }
13020 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13021 hexfp = TRUE;
13022 if (!has_precis) {
13023 /* %a in the absence of precision may print as many
13024 * digits as needed to represent the entire mantissa
13025 * bit pattern.
13026 * This estimate seriously overshoots in most cases,
13027 * but better the undershooting. Firstly, all bytes
13028 * of the NV are not mantissa, some of them are
13029 * exponent. Secondly, for the reasonably common
13030 * long doubles case, the "80-bit extended", two
13031 * or six bytes of the NV are unused. Also, we'll
13032 * still pick up an extra +6 from the default
13033 * precision calculation below. */
13034 STRLEN digits =
13035 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13036 /* For the "double double", we need more.
13037 * Since each double has their own exponent, the
13038 * doubles may float (haha) rather far from each
13039 * other, and the number of required bits is much
13040 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13041 * See the definition of DOUBLEDOUBLE_MAXBITS.
13042 *
13043 * Need 2 hexdigits for each byte. */
13044 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13045 #else
13046 NVSIZE * 2; /* 2 hexdigits for each byte */
13047 #endif
13048 /* see "this can't overflow" comment above */
13049 assert(float_need < ((STRLEN)~0) - digits);
13050 float_need += digits;
13051 }
13052 }
13053 /* special-case "%.<number>g" if it will fit in ebuf */
13054 else if (c == 'g'
13055 && precis /* See earlier comment about buggy Gconvert
13056 when digits, aka precis, is 0 */
13057 && has_precis
13058 /* check, in manner not involving wrapping, that it will
13059 * fit in ebuf */
13060 && float_need < sizeof(ebuf)
13061 && sizeof(ebuf) - float_need > precis
13062 && !(width || left || plus || alt)
13063 && !fill
13064 && intsize != 'q'
13065 ) {
13066 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13067 elen = strlen(ebuf);
13068 eptr = ebuf;
13069 goto float_concat;
13070 }
13071
13072
13073 {
13074 STRLEN pr = has_precis ? precis : 6; /* known default */
13075 /* this probably can't wrap, since precis is limited
13076 * to 1/4 address space size, but better safe than sorry
13077 */
13078 if (float_need >= ((STRLEN)~0) - pr)
13079 croak_memory_wrap();
13080 float_need += pr;
13081 }
13082
13083 if (float_need < width)
13084 float_need = width;
13085
13086 if (PL_efloatsize <= float_need) {
13087 /* PL_efloatbuf should be at least 1 greater than
13088 * float_need to allow a trailing \0 to be returned by
13089 * snprintf(). If we need to grow, overgrow for the
13090 * benefit of future generations */
13091 const STRLEN extra = 0x20;
13092 if (float_need >= ((STRLEN)~0) - extra)
13093 croak_memory_wrap();
13094 float_need += extra;
13095 Safefree(PL_efloatbuf);
13096 PL_efloatsize = float_need;
13097 Newx(PL_efloatbuf, PL_efloatsize, char);
13098 PL_efloatbuf[0] = '\0';
13099 }
13100
13101 if (UNLIKELY(hexfp)) {
13102 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13103 nv, fv, has_precis, precis, width,
13104 alt, plus, left, fill);
13105 }
13106 else {
13107 char *ptr = ebuf + sizeof ebuf;
13108 *--ptr = '\0';
13109 *--ptr = c;
13110 #if defined(USE_QUADMATH)
13111 if (intsize == 'q') {
13112 /* "g" -> "Qg" */
13113 *--ptr = 'Q';
13114 }
13115 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13116 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13117 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13118 * not USE_LONG_DOUBLE and NVff. In other words,
13119 * this needs to work without USE_LONG_DOUBLE. */
13120 if (intsize == 'q') {
13121 /* Copy the one or more characters in a long double
13122 * format before the 'base' ([efgEFG]) character to
13123 * the format string. */
13124 static char const ldblf[] = PERL_PRIfldbl;
13125 char const *p = ldblf + sizeof(ldblf) - 3;
13126 while (p >= ldblf) { *--ptr = *p--; }
13127 }
13128 #endif
13129 if (has_precis) {
13130 base = precis;
13131 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13132 *--ptr = '.';
13133 }
13134 if (width) {
13135 base = width;
13136 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13137 }
13138 if (fill)
13139 *--ptr = '0';
13140 if (left)
13141 *--ptr = '-';
13142 if (plus)
13143 *--ptr = plus;
13144 if (alt)
13145 *--ptr = '#';
13146 *--ptr = '%';
13147
13148 /* No taint. Otherwise we are in the strange situation
13149 * where printf() taints but print($float) doesn't.
13150 * --jhi */
13151
13152 /* hopefully the above makes ptr a very constrained format
13153 * that is safe to use, even though it's not literal */
13154 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13155 #ifdef USE_QUADMATH
13156 {
13157 const char* qfmt = quadmath_format_single(ptr);
13158 if (!qfmt)
13159 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13160 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13161 qfmt, nv);
13162 if ((IV)elen == -1) {
13163 if (qfmt != ptr)
13164 SAVEFREEPV(qfmt);
13165 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13166 }
13167 if (qfmt != ptr)
13168 Safefree(qfmt);
13169 }
13170 #elif defined(HAS_LONG_DOUBLE)
13171 elen = ((intsize == 'q')
13172 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13173 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13174 #else
13175 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13176 #endif
13177 GCC_DIAG_RESTORE_STMT;
13178 }
13179
13180 eptr = PL_efloatbuf;
13181
13182 float_concat:
13183
13184 /* Since floating-point formats do their own formatting and
13185 * padding, we skip the main block of code at the end of this
13186 * loop which handles appending eptr to sv, and do our own
13187 * stripped-down version */
13188
13189 assert(!zeros);
13190 assert(!esignlen);
13191 assert(elen);
13192 assert(elen >= width);
13193
13194 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13195
13196 goto done_valid_conversion;
13197 }
13198
13199 /* SPECIAL */
13200
13201 case 'n':
13202 {
13203 STRLEN len;
13204 /* XXX ideally we should warn if any flags etc have been
13205 * set, e.g. "%-4.5n" */
13206 /* XXX if sv was originally non-utf8 with a char in the
13207 * range 0x80-0xff, then if it got upgraded, we should
13208 * calculate char len rather than byte len here */
13209 len = SvCUR(sv) - origlen;
13210 if (args) {
13211 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13212
13213 switch (intsize) {
13214 case 'c': *(va_arg(*args, char*)) = i; break;
13215 case 'h': *(va_arg(*args, short*)) = i; break;
13216 default: *(va_arg(*args, int*)) = i; break;
13217 case 'l': *(va_arg(*args, long*)) = i; break;
13218 case 'V': *(va_arg(*args, IV*)) = i; break;
13219 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13220 #ifdef HAS_PTRDIFF_T
13221 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13222 #endif
13223 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13224 case 'q':
13225 #if IVSIZE >= 8
13226 *(va_arg(*args, Quad_t*)) = i; break;
13227 #else
13228 goto unknown;
13229 #endif
13230 }
13231 }
13232 else {
13233 if (arg_missing)
13234 Perl_croak_nocontext(
13235 "Missing argument for %%n in %s",
13236 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13237 sv_setuv_mg(argsv, has_utf8
13238 ? (UV)utf8_length((U8*)SvPVX(sv), (U8*)SvEND(sv))
13239 : (UV)len);
13240 }
13241 goto done_valid_conversion;
13242 }
13243
13244 /* UNKNOWN */
13245
13246 default:
13247 unknown:
13248 if (!args
13249 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13250 && ckWARN(WARN_PRINTF))
13251 {
13252 SV * const msg = sv_newmortal();
13253 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13254 (PL_op->op_type == OP_PRTF) ? "" : "s");
13255 if (fmtstart < patend) {
13256 const char * const fmtend = q < patend ? q : patend;
13257 const char * f;
13258 sv_catpvs(msg, "\"%");
13259 for (f = fmtstart; f < fmtend; f++) {
13260 if (isPRINT(*f)) {
13261 sv_catpvn_nomg(msg, f, 1);
13262 } else {
13263 Perl_sv_catpvf(aTHX_ msg,
13264 "\\%03" UVof, (UV)*f & 0xFF);
13265 }
13266 }
13267 sv_catpvs(msg, "\"");
13268 } else {
13269 sv_catpvs(msg, "end of string");
13270 }
13271 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13272 }
13273
13274 /* mangled format: output the '%', then continue from the
13275 * character following that */
13276 sv_catpvn_nomg(sv, fmtstart-1, 1);
13277 q = fmtstart;
13278 svix = osvix;
13279 /* Any "redundant arg" warning from now onwards will probably
13280 * just be misleading, so don't bother. */
13281 no_redundant_warning = TRUE;
13282 continue; /* not "break" */
13283 }
13284
13285 if (is_utf8 != has_utf8) {
13286 if (is_utf8) {
13287 if (SvCUR(sv))
13288 sv_utf8_upgrade(sv);
13289 }
13290 else {
13291 const STRLEN old_elen = elen;
13292 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13293 sv_utf8_upgrade(nsv);
13294 eptr = SvPVX_const(nsv);
13295 elen = SvCUR(nsv);
13296
13297 if (width) { /* fudge width (can't fudge elen) */
13298 width += elen - old_elen;
13299 }
13300 is_utf8 = TRUE;
13301 }
13302 }
13303
13304
13305 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13306
13307 {
13308 STRLEN need, have, gap;
13309 STRLEN i;
13310 char *s;
13311
13312 /* signed value that's wrapped? */
13313 assert(elen <= ((~(STRLEN)0) >> 1));
13314
13315 /* if zeros is non-zero, then it represents filler between
13316 * elen and precis. So adding elen and zeros together will
13317 * always be <= precis, and the addition can never wrap */
13318 assert(!zeros || (precis > elen && precis - elen == zeros));
13319 have = elen + zeros;
13320
13321 if (have >= (((STRLEN)~0) - esignlen))
13322 croak_memory_wrap();
13323 have += esignlen;
13324
13325 need = (have > width ? have : width);
13326 gap = need - have;
13327
13328 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13329 croak_memory_wrap();
13330 need += (SvCUR(sv) + 1);
13331
13332 SvGROW(sv, need);
13333
13334 s = SvEND(sv);
13335
13336 if (left) {
13337 for (i = 0; i < esignlen; i++)
13338 *s++ = esignbuf[i];
13339 for (i = zeros; i; i--)
13340 *s++ = '0';
13341 Copy(eptr, s, elen, char);
13342 s += elen;
13343 for (i = gap; i; i--)
13344 *s++ = ' ';
13345 }
13346 else {
13347 if (fill) {
13348 for (i = 0; i < esignlen; i++)
13349 *s++ = esignbuf[i];
13350 assert(!zeros);
13351 zeros = gap;
13352 }
13353 else {
13354 for (i = gap; i; i--)
13355 *s++ = ' ';
13356 for (i = 0; i < esignlen; i++)
13357 *s++ = esignbuf[i];
13358 }
13359
13360 for (i = zeros; i; i--)
13361 *s++ = '0';
13362 Copy(eptr, s, elen, char);
13363 s += elen;
13364 }
13365
13366 *s = '\0';
13367 SvCUR_set(sv, s - SvPVX_const(sv));
13368
13369 if (is_utf8)
13370 has_utf8 = TRUE;
13371 if (has_utf8)
13372 SvUTF8_on(sv);
13373 }
13374
13375 if (vectorize && veclen) {
13376 /* we append the vector separator separately since %v isn't
13377 * very common: don't slow down the general case by adding
13378 * dotstrlen to need etc */
13379 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13380 esignlen = 0;
13381 goto vector; /* do next iteration */
13382 }
13383
13384 done_valid_conversion:
13385
13386 if (arg_missing)
13387 S_warn_vcatpvfn_missing_argument(aTHX);
13388 }
13389
13390 /* Now that we've consumed all our printf format arguments (svix)
13391 * do we have things left on the stack that we didn't use?
13392 */
13393 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13394 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13395 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13396 }
13397
13398 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
13399 /* while we shouldn't set the cache, it may have been previously
13400 set in the caller, so clear it */
13401 MAGIC *mg = mg_find(sv, PERL_MAGIC_utf8);
13402 if (mg)
13403 magic_setutf8(sv,mg); /* clear UTF8 cache */
13404 }
13405 SvTAINT(sv);
13406
13407 #ifdef USE_LOCALE_NUMERIC
13408
13409 if (lc_numeric_set) {
13410 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to
13411 save/restore each iteration. */
13412 }
13413
13414 #endif
13415
13416 }
13417
13418 /* =========================================================================
13419
13420 =head1 Cloning an interpreter
13421
13422 =cut
13423
13424 All the macros and functions in this section are for the private use of
13425 the main function, perl_clone().
13426
13427 The foo_dup() functions make an exact copy of an existing foo thingy.
13428 During the course of a cloning, a hash table is used to map old addresses
13429 to new addresses. The table is created and manipulated with the
13430 ptr_table_* functions.
13431
13432 * =========================================================================*/
13433
13434
13435 #if defined(USE_ITHREADS)
13436
13437 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13438 #ifndef GpREFCNT_inc
13439 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13440 #endif
13441
13442
13443 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13444 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13445 If this changes, please unmerge ss_dup.
13446 Likewise, sv_dup_inc_multiple() relies on this fact. */
13447 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13448 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13449 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13450 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13451 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13452 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13453 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13454 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13455 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13456 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13457 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13458 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13459 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13460
13461 /* clone a parser */
13462
13463 yy_parser *
Perl_parser_dup(pTHX_ const yy_parser * const proto,CLONE_PARAMS * const param)13464 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13465 {
13466 yy_parser *parser;
13467
13468 PERL_ARGS_ASSERT_PARSER_DUP;
13469
13470 if (!proto)
13471 return NULL;
13472
13473 /* look for it in the table first */
13474 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13475 if (parser)
13476 return parser;
13477
13478 /* create anew and remember what it is */
13479 Newxz(parser, 1, yy_parser);
13480 ptr_table_store(PL_ptr_table, proto, parser);
13481
13482 /* XXX eventually, just Copy() most of the parser struct ? */
13483
13484 parser->lex_brackets = proto->lex_brackets;
13485 parser->lex_casemods = proto->lex_casemods;
13486 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13487 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13488 parser->lex_casestack = savepvn(proto->lex_casestack,
13489 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13490 parser->lex_defer = proto->lex_defer;
13491 parser->lex_dojoin = proto->lex_dojoin;
13492 parser->lex_formbrack = proto->lex_formbrack;
13493 parser->lex_inpat = proto->lex_inpat;
13494 parser->lex_inwhat = proto->lex_inwhat;
13495 parser->lex_op = proto->lex_op;
13496 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13497 parser->lex_starts = proto->lex_starts;
13498 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13499 parser->multi_close = proto->multi_close;
13500 parser->multi_open = proto->multi_open;
13501 parser->multi_start = proto->multi_start;
13502 parser->multi_end = proto->multi_end;
13503 parser->preambled = proto->preambled;
13504 parser->lex_super_state = proto->lex_super_state;
13505 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13506 parser->lex_sub_op = proto->lex_sub_op;
13507 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13508 parser->linestr = sv_dup_inc(proto->linestr, param);
13509 parser->expect = proto->expect;
13510 parser->copline = proto->copline;
13511 parser->last_lop_op = proto->last_lop_op;
13512 parser->lex_state = proto->lex_state;
13513 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13514 /* rsfp_filters entries have fake IoDIRP() */
13515 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13516 parser->in_my = proto->in_my;
13517 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13518 parser->error_count = proto->error_count;
13519 parser->sig_elems = proto->sig_elems;
13520 parser->sig_optelems= proto->sig_optelems;
13521 parser->sig_slurpy = proto->sig_slurpy;
13522 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13523
13524 {
13525 char * const ols = SvPVX(proto->linestr);
13526 char * const ls = SvPVX(parser->linestr);
13527
13528 parser->bufptr = ls + (proto->bufptr >= ols ?
13529 proto->bufptr - ols : 0);
13530 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13531 proto->oldbufptr - ols : 0);
13532 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13533 proto->oldoldbufptr - ols : 0);
13534 parser->linestart = ls + (proto->linestart >= ols ?
13535 proto->linestart - ols : 0);
13536 parser->last_uni = ls + (proto->last_uni >= ols ?
13537 proto->last_uni - ols : 0);
13538 parser->last_lop = ls + (proto->last_lop >= ols ?
13539 proto->last_lop - ols : 0);
13540
13541 parser->bufend = ls + SvCUR(parser->linestr);
13542 }
13543
13544 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13545
13546
13547 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13548 Copy(proto->nexttype, parser->nexttype, 5, I32);
13549 parser->nexttoke = proto->nexttoke;
13550
13551 /* XXX should clone saved_curcop here, but we aren't passed
13552 * proto_perl; so do it in perl_clone_using instead */
13553
13554 return parser;
13555 }
13556
13557
13558 /* duplicate a file handle */
13559
13560 PerlIO *
Perl_fp_dup(pTHX_ PerlIO * const fp,const char type,CLONE_PARAMS * const param)13561 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13562 {
13563 PerlIO *ret;
13564
13565 PERL_ARGS_ASSERT_FP_DUP;
13566 PERL_UNUSED_ARG(type);
13567
13568 if (!fp)
13569 return (PerlIO*)NULL;
13570
13571 /* look for it in the table first */
13572 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13573 if (ret)
13574 return ret;
13575
13576 /* create anew and remember what it is */
13577 #ifdef __amigaos4__
13578 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13579 #else
13580 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13581 #endif
13582 ptr_table_store(PL_ptr_table, fp, ret);
13583 return ret;
13584 }
13585
13586 /* duplicate a directory handle */
13587
13588 DIR *
Perl_dirp_dup(pTHX_ DIR * const dp,CLONE_PARAMS * const param)13589 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13590 {
13591 DIR *ret;
13592
13593 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13594 DIR *pwd;
13595 const Direntry_t *dirent;
13596 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13597 char *name = NULL;
13598 STRLEN len = 0;
13599 long pos;
13600 #endif
13601
13602 PERL_UNUSED_CONTEXT;
13603 PERL_ARGS_ASSERT_DIRP_DUP;
13604
13605 if (!dp)
13606 return (DIR*)NULL;
13607
13608 /* look for it in the table first */
13609 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13610 if (ret)
13611 return ret;
13612
13613 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13614
13615 PERL_UNUSED_ARG(param);
13616
13617 /* create anew */
13618
13619 /* open the current directory (so we can switch back) */
13620 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13621
13622 /* chdir to our dir handle and open the present working directory */
13623 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13624 PerlDir_close(pwd);
13625 return (DIR *)NULL;
13626 }
13627 /* Now we should have two dir handles pointing to the same dir. */
13628
13629 /* Be nice to the calling code and chdir back to where we were. */
13630 /* XXX If this fails, then what? */
13631 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13632
13633 /* We have no need of the pwd handle any more. */
13634 PerlDir_close(pwd);
13635
13636 #ifdef DIRNAMLEN
13637 # define d_namlen(d) (d)->d_namlen
13638 #else
13639 # define d_namlen(d) strlen((d)->d_name)
13640 #endif
13641 /* Iterate once through dp, to get the file name at the current posi-
13642 tion. Then step back. */
13643 pos = PerlDir_tell(dp);
13644 if ((dirent = PerlDir_read(dp))) {
13645 len = d_namlen(dirent);
13646 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13647 /* If the len is somehow magically longer than the
13648 * maximum length of the directory entry, even though
13649 * we could fit it in a buffer, we could not copy it
13650 * from the dirent. Bail out. */
13651 PerlDir_close(ret);
13652 return (DIR*)NULL;
13653 }
13654 if (len <= sizeof smallbuf) name = smallbuf;
13655 else Newx(name, len, char);
13656 Move(dirent->d_name, name, len, char);
13657 }
13658 PerlDir_seek(dp, pos);
13659
13660 /* Iterate through the new dir handle, till we find a file with the
13661 right name. */
13662 if (!dirent) /* just before the end */
13663 for(;;) {
13664 pos = PerlDir_tell(ret);
13665 if (PerlDir_read(ret)) continue; /* not there yet */
13666 PerlDir_seek(ret, pos); /* step back */
13667 break;
13668 }
13669 else {
13670 const long pos0 = PerlDir_tell(ret);
13671 for(;;) {
13672 pos = PerlDir_tell(ret);
13673 if ((dirent = PerlDir_read(ret))) {
13674 if (len == (STRLEN)d_namlen(dirent)
13675 && memEQ(name, dirent->d_name, len)) {
13676 /* found it */
13677 PerlDir_seek(ret, pos); /* step back */
13678 break;
13679 }
13680 /* else we are not there yet; keep iterating */
13681 }
13682 else { /* This is not meant to happen. The best we can do is
13683 reset the iterator to the beginning. */
13684 PerlDir_seek(ret, pos0);
13685 break;
13686 }
13687 }
13688 }
13689 #undef d_namlen
13690
13691 if (name && name != smallbuf)
13692 Safefree(name);
13693 #endif
13694
13695 #ifdef WIN32
13696 ret = win32_dirp_dup(dp, param);
13697 #endif
13698
13699 /* pop it in the pointer table */
13700 if (ret)
13701 ptr_table_store(PL_ptr_table, dp, ret);
13702
13703 return ret;
13704 }
13705
13706 /* duplicate a typeglob */
13707
13708 GP *
Perl_gp_dup(pTHX_ GP * const gp,CLONE_PARAMS * const param)13709 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13710 {
13711 GP *ret;
13712
13713 PERL_ARGS_ASSERT_GP_DUP;
13714
13715 if (!gp)
13716 return (GP*)NULL;
13717 /* look for it in the table first */
13718 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13719 if (ret)
13720 return ret;
13721
13722 /* create anew and remember what it is */
13723 Newxz(ret, 1, GP);
13724 ptr_table_store(PL_ptr_table, gp, ret);
13725
13726 /* clone */
13727 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13728 on Newxz() to do this for us. */
13729 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13730 ret->gp_io = io_dup_inc(gp->gp_io, param);
13731 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13732 ret->gp_av = av_dup_inc(gp->gp_av, param);
13733 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13734 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13735 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13736 ret->gp_cvgen = gp->gp_cvgen;
13737 ret->gp_line = gp->gp_line;
13738 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13739 return ret;
13740 }
13741
13742 /* duplicate a chain of magic */
13743
13744 MAGIC *
Perl_mg_dup(pTHX_ MAGIC * mg,CLONE_PARAMS * const param)13745 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13746 {
13747 MAGIC *mgret = NULL;
13748 MAGIC **mgprev_p = &mgret;
13749
13750 PERL_ARGS_ASSERT_MG_DUP;
13751
13752 for (; mg; mg = mg->mg_moremagic) {
13753 MAGIC *nmg;
13754
13755 if ((param->flags & CLONEf_JOIN_IN)
13756 && mg->mg_type == PERL_MAGIC_backref)
13757 /* when joining, we let the individual SVs add themselves to
13758 * backref as needed. */
13759 continue;
13760
13761 Newx(nmg, 1, MAGIC);
13762 *mgprev_p = nmg;
13763 mgprev_p = &(nmg->mg_moremagic);
13764
13765 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13766 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13767 from the original commit adding Perl_mg_dup() - revision 4538.
13768 Similarly there is the annotation "XXX random ptr?" next to the
13769 assignment to nmg->mg_ptr. */
13770 *nmg = *mg;
13771
13772 /* FIXME for plugins
13773 if (nmg->mg_type == PERL_MAGIC_qr) {
13774 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13775 }
13776 else
13777 */
13778 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13779 ? nmg->mg_type == PERL_MAGIC_backref
13780 /* The backref AV has its reference
13781 * count deliberately bumped by 1 */
13782 ? SvREFCNT_inc(av_dup_inc((const AV *)
13783 nmg->mg_obj, param))
13784 : sv_dup_inc(nmg->mg_obj, param)
13785 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13786 nmg->mg_type == PERL_MAGIC_regdata)
13787 ? nmg->mg_obj
13788 : sv_dup(nmg->mg_obj, param);
13789
13790 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13791 if (nmg->mg_len > 0) {
13792 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13793 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13794 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13795 {
13796 AMT * const namtp = (AMT*)nmg->mg_ptr;
13797 sv_dup_inc_multiple((SV**)(namtp->table),
13798 (SV**)(namtp->table), NofAMmeth, param);
13799 }
13800 }
13801 else if (nmg->mg_len == HEf_SVKEY)
13802 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13803 }
13804 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13805 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13806 }
13807 }
13808 return mgret;
13809 }
13810
13811 #endif /* USE_ITHREADS */
13812
13813 struct ptr_tbl_arena {
13814 struct ptr_tbl_arena *next;
13815 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13816 };
13817
13818 /* create a new pointer-mapping table */
13819
13820 PTR_TBL_t *
Perl_ptr_table_new(pTHX)13821 Perl_ptr_table_new(pTHX)
13822 {
13823 PTR_TBL_t *tbl;
13824 PERL_UNUSED_CONTEXT;
13825
13826 Newx(tbl, 1, PTR_TBL_t);
13827 tbl->tbl_max = 511;
13828 tbl->tbl_items = 0;
13829 tbl->tbl_arena = NULL;
13830 tbl->tbl_arena_next = NULL;
13831 tbl->tbl_arena_end = NULL;
13832 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13833 return tbl;
13834 }
13835
13836 #define PTR_TABLE_HASH(ptr) \
13837 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13838
13839 /* map an existing pointer using a table */
13840
13841 STATIC PTR_TBL_ENT_t *
S_ptr_table_find(PTR_TBL_t * const tbl,const void * const sv)13842 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13843 {
13844 PTR_TBL_ENT_t *tblent;
13845 const UV hash = PTR_TABLE_HASH(sv);
13846
13847 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13848
13849 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13850 for (; tblent; tblent = tblent->next) {
13851 if (tblent->oldval == sv)
13852 return tblent;
13853 }
13854 return NULL;
13855 }
13856
13857 void *
Perl_ptr_table_fetch(pTHX_ PTR_TBL_t * const tbl,const void * const sv)13858 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13859 {
13860 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13861
13862 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13863 PERL_UNUSED_CONTEXT;
13864
13865 return tblent ? tblent->newval : NULL;
13866 }
13867
13868 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13869 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13870 * the core's typical use of ptr_tables in thread cloning. */
13871
13872 void
Perl_ptr_table_store(pTHX_ PTR_TBL_t * const tbl,const void * const oldsv,void * const newsv)13873 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13874 {
13875 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13876
13877 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13878 PERL_UNUSED_CONTEXT;
13879
13880 if (tblent) {
13881 tblent->newval = newsv;
13882 } else {
13883 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13884
13885 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13886 struct ptr_tbl_arena *new_arena;
13887
13888 Newx(new_arena, 1, struct ptr_tbl_arena);
13889 new_arena->next = tbl->tbl_arena;
13890 tbl->tbl_arena = new_arena;
13891 tbl->tbl_arena_next = new_arena->array;
13892 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13893 }
13894
13895 tblent = tbl->tbl_arena_next++;
13896
13897 tblent->oldval = oldsv;
13898 tblent->newval = newsv;
13899 tblent->next = tbl->tbl_ary[entry];
13900 tbl->tbl_ary[entry] = tblent;
13901 tbl->tbl_items++;
13902 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13903 ptr_table_split(tbl);
13904 }
13905 }
13906
13907 /* double the hash bucket size of an existing ptr table */
13908
13909 void
Perl_ptr_table_split(pTHX_ PTR_TBL_t * const tbl)13910 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13911 {
13912 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13913 const UV oldsize = tbl->tbl_max + 1;
13914 UV newsize = oldsize * 2;
13915 UV i;
13916
13917 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13918 PERL_UNUSED_CONTEXT;
13919
13920 Renew(ary, newsize, PTR_TBL_ENT_t*);
13921 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13922 tbl->tbl_max = --newsize;
13923 tbl->tbl_ary = ary;
13924 for (i=0; i < oldsize; i++, ary++) {
13925 PTR_TBL_ENT_t **entp = ary;
13926 PTR_TBL_ENT_t *ent = *ary;
13927 PTR_TBL_ENT_t **curentp;
13928 if (!ent)
13929 continue;
13930 curentp = ary + oldsize;
13931 do {
13932 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13933 *entp = ent->next;
13934 ent->next = *curentp;
13935 *curentp = ent;
13936 }
13937 else
13938 entp = &ent->next;
13939 ent = *entp;
13940 } while (ent);
13941 }
13942 }
13943
13944 /* remove all the entries from a ptr table */
13945 /* Deprecated - will be removed post 5.14 */
13946
13947 void
Perl_ptr_table_clear(pTHX_ PTR_TBL_t * const tbl)13948 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13949 {
13950 PERL_UNUSED_CONTEXT;
13951 if (tbl && tbl->tbl_items) {
13952 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13953
13954 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13955
13956 while (arena) {
13957 struct ptr_tbl_arena *next = arena->next;
13958
13959 Safefree(arena);
13960 arena = next;
13961 };
13962
13963 tbl->tbl_items = 0;
13964 tbl->tbl_arena = NULL;
13965 tbl->tbl_arena_next = NULL;
13966 tbl->tbl_arena_end = NULL;
13967 }
13968 }
13969
13970 /* clear and free a ptr table */
13971
13972 void
Perl_ptr_table_free(pTHX_ PTR_TBL_t * const tbl)13973 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13974 {
13975 struct ptr_tbl_arena *arena;
13976
13977 PERL_UNUSED_CONTEXT;
13978
13979 if (!tbl) {
13980 return;
13981 }
13982
13983 arena = tbl->tbl_arena;
13984
13985 while (arena) {
13986 struct ptr_tbl_arena *next = arena->next;
13987
13988 Safefree(arena);
13989 arena = next;
13990 }
13991
13992 Safefree(tbl->tbl_ary);
13993 Safefree(tbl);
13994 }
13995
13996 #if defined(USE_ITHREADS)
13997
13998 void
Perl_rvpv_dup(pTHX_ SV * const dstr,const SV * const sstr,CLONE_PARAMS * const param)13999 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14000 {
14001 PERL_ARGS_ASSERT_RVPV_DUP;
14002
14003 assert(!isREGEXP(sstr));
14004 if (SvROK(sstr)) {
14005 if (SvWEAKREF(sstr)) {
14006 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14007 if (param->flags & CLONEf_JOIN_IN) {
14008 /* if joining, we add any back references individually rather
14009 * than copying the whole backref array */
14010 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14011 }
14012 }
14013 else
14014 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14015 }
14016 else if (SvPVX_const(sstr)) {
14017 /* Has something there */
14018 if (SvLEN(sstr)) {
14019 /* Normal PV - clone whole allocated space */
14020 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14021 /* sstr may not be that normal, but actually copy on write.
14022 But we are a true, independent SV, so: */
14023 SvIsCOW_off(dstr);
14024 }
14025 else {
14026 /* Special case - not normally malloced for some reason */
14027 if (isGV_with_GP(sstr)) {
14028 /* Don't need to do anything here. */
14029 }
14030 else if ((SvIsCOW(sstr))) {
14031 /* A "shared" PV - clone it as "shared" PV */
14032 SvPV_set(dstr,
14033 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14034 param)));
14035 }
14036 else {
14037 /* Some other special case - random pointer */
14038 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14039 }
14040 }
14041 }
14042 else {
14043 /* Copy the NULL */
14044 SvPV_set(dstr, NULL);
14045 }
14046 }
14047
14048 /* duplicate a list of SVs. source and dest may point to the same memory. */
14049 static SV **
S_sv_dup_inc_multiple(pTHX_ SV * const * source,SV ** dest,SSize_t items,CLONE_PARAMS * const param)14050 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14051 SSize_t items, CLONE_PARAMS *const param)
14052 {
14053 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14054
14055 while (items-- > 0) {
14056 *dest++ = sv_dup_inc(*source++, param);
14057 }
14058
14059 return dest;
14060 }
14061
14062 /* duplicate an SV of any type (including AV, HV etc) */
14063
14064 static SV *
S_sv_dup_common(pTHX_ const SV * const sstr,CLONE_PARAMS * const param)14065 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14066 {
14067 dVAR;
14068 SV *dstr;
14069
14070 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14071
14072 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14073 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14074 abort();
14075 #endif
14076 return NULL;
14077 }
14078 /* look for it in the table first */
14079 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14080 if (dstr)
14081 return dstr;
14082
14083 if(param->flags & CLONEf_JOIN_IN) {
14084 /** We are joining here so we don't want do clone
14085 something that is bad **/
14086 if (SvTYPE(sstr) == SVt_PVHV) {
14087 const HEK * const hvname = HvNAME_HEK(sstr);
14088 if (hvname) {
14089 /** don't clone stashes if they already exist **/
14090 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14091 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14092 ptr_table_store(PL_ptr_table, sstr, dstr);
14093 return dstr;
14094 }
14095 }
14096 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14097 HV *stash = GvSTASH(sstr);
14098 const HEK * hvname;
14099 if (stash && (hvname = HvNAME_HEK(stash))) {
14100 /** don't clone GVs if they already exist **/
14101 SV **svp;
14102 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14103 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14104 svp = hv_fetch(
14105 stash, GvNAME(sstr),
14106 GvNAMEUTF8(sstr)
14107 ? -GvNAMELEN(sstr)
14108 : GvNAMELEN(sstr),
14109 0
14110 );
14111 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14112 ptr_table_store(PL_ptr_table, sstr, *svp);
14113 return *svp;
14114 }
14115 }
14116 }
14117 }
14118
14119 /* create anew and remember what it is */
14120 new_SV(dstr);
14121
14122 #ifdef DEBUG_LEAKING_SCALARS
14123 dstr->sv_debug_optype = sstr->sv_debug_optype;
14124 dstr->sv_debug_line = sstr->sv_debug_line;
14125 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14126 dstr->sv_debug_parent = (SV*)sstr;
14127 FREE_SV_DEBUG_FILE(dstr);
14128 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14129 #endif
14130
14131 ptr_table_store(PL_ptr_table, sstr, dstr);
14132
14133 /* clone */
14134 SvFLAGS(dstr) = SvFLAGS(sstr);
14135 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14136 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14137
14138 #ifdef DEBUGGING
14139 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14140 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14141 (void*)PL_watch_pvx, SvPVX_const(sstr));
14142 #endif
14143
14144 /* don't clone objects whose class has asked us not to */
14145 if (SvOBJECT(sstr)
14146 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14147 {
14148 SvFLAGS(dstr) = 0;
14149 return dstr;
14150 }
14151
14152 switch (SvTYPE(sstr)) {
14153 case SVt_NULL:
14154 SvANY(dstr) = NULL;
14155 break;
14156 case SVt_IV:
14157 SET_SVANY_FOR_BODYLESS_IV(dstr);
14158 if(SvROK(sstr)) {
14159 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14160 } else {
14161 SvIV_set(dstr, SvIVX(sstr));
14162 }
14163 break;
14164 case SVt_NV:
14165 #if NVSIZE <= IVSIZE
14166 SET_SVANY_FOR_BODYLESS_NV(dstr);
14167 #else
14168 SvANY(dstr) = new_XNV();
14169 #endif
14170 SvNV_set(dstr, SvNVX(sstr));
14171 break;
14172 default:
14173 {
14174 /* These are all the types that need complex bodies allocating. */
14175 void *new_body;
14176 const svtype sv_type = SvTYPE(sstr);
14177 const struct body_details *const sv_type_details
14178 = bodies_by_type + sv_type;
14179
14180 switch (sv_type) {
14181 default:
14182 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14183 NOT_REACHED; /* NOTREACHED */
14184 break;
14185
14186 case SVt_PVGV:
14187 case SVt_PVIO:
14188 case SVt_PVFM:
14189 case SVt_PVHV:
14190 case SVt_PVAV:
14191 case SVt_PVCV:
14192 case SVt_PVLV:
14193 case SVt_REGEXP:
14194 case SVt_PVMG:
14195 case SVt_PVNV:
14196 case SVt_PVIV:
14197 case SVt_INVLIST:
14198 case SVt_PV:
14199 assert(sv_type_details->body_size);
14200 if (sv_type_details->arena) {
14201 new_body_inline(new_body, sv_type);
14202 new_body
14203 = (void*)((char*)new_body - sv_type_details->offset);
14204 } else {
14205 new_body = new_NOARENA(sv_type_details);
14206 }
14207 }
14208 assert(new_body);
14209 SvANY(dstr) = new_body;
14210
14211 #ifndef PURIFY
14212 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14213 ((char*)SvANY(dstr)) + sv_type_details->offset,
14214 sv_type_details->copy, char);
14215 #else
14216 Copy(((char*)SvANY(sstr)),
14217 ((char*)SvANY(dstr)),
14218 sv_type_details->body_size + sv_type_details->offset, char);
14219 #endif
14220
14221 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14222 && !isGV_with_GP(dstr)
14223 && !isREGEXP(dstr)
14224 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14225 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14226
14227 /* The Copy above means that all the source (unduplicated) pointers
14228 are now in the destination. We can check the flags and the
14229 pointers in either, but it's possible that there's less cache
14230 missing by always going for the destination.
14231 FIXME - instrument and check that assumption */
14232 if (sv_type >= SVt_PVMG) {
14233 if (SvMAGIC(dstr))
14234 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14235 if (SvOBJECT(dstr) && SvSTASH(dstr))
14236 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14237 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14238 }
14239
14240 /* The cast silences a GCC warning about unhandled types. */
14241 switch ((int)sv_type) {
14242 case SVt_PV:
14243 break;
14244 case SVt_PVIV:
14245 break;
14246 case SVt_PVNV:
14247 break;
14248 case SVt_PVMG:
14249 break;
14250 case SVt_REGEXP:
14251 duprex:
14252 /* FIXME for plugins */
14253 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14254 break;
14255 case SVt_PVLV:
14256 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14257 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14258 LvTARG(dstr) = dstr;
14259 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14260 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14261 else
14262 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14263 if (isREGEXP(sstr)) goto duprex;
14264 /* FALLTHROUGH */
14265 case SVt_PVGV:
14266 /* non-GP case already handled above */
14267 if(isGV_with_GP(sstr)) {
14268 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14269 /* Don't call sv_add_backref here as it's going to be
14270 created as part of the magic cloning of the symbol
14271 table--unless this is during a join and the stash
14272 is not actually being cloned. */
14273 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14274 at the point of this comment. */
14275 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14276 if (param->flags & CLONEf_JOIN_IN)
14277 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14278 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14279 (void)GpREFCNT_inc(GvGP(dstr));
14280 }
14281 break;
14282 case SVt_PVIO:
14283 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14284 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14285 /* I have no idea why fake dirp (rsfps)
14286 should be treated differently but otherwise
14287 we end up with leaks -- sky*/
14288 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14289 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14290 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14291 } else {
14292 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14293 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14294 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14295 if (IoDIRP(dstr)) {
14296 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14297 } else {
14298 NOOP;
14299 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14300 }
14301 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14302 }
14303 if (IoOFP(dstr) == IoIFP(sstr))
14304 IoOFP(dstr) = IoIFP(dstr);
14305 else
14306 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14307 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14308 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14309 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14310 break;
14311 case SVt_PVAV:
14312 /* avoid cloning an empty array */
14313 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14314 SV **dst_ary, **src_ary;
14315 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14316
14317 src_ary = AvARRAY((const AV *)sstr);
14318 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14319 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14320 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14321 AvALLOC((const AV *)dstr) = dst_ary;
14322 if (AvREAL((const AV *)sstr)) {
14323 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14324 param);
14325 }
14326 else {
14327 while (items-- > 0)
14328 *dst_ary++ = sv_dup(*src_ary++, param);
14329 }
14330 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14331 while (items-- > 0) {
14332 *dst_ary++ = NULL;
14333 }
14334 }
14335 else {
14336 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14337 AvALLOC((const AV *)dstr) = (SV**)NULL;
14338 AvMAX( (const AV *)dstr) = -1;
14339 AvFILLp((const AV *)dstr) = -1;
14340 }
14341 break;
14342 case SVt_PVHV:
14343 if (HvARRAY((const HV *)sstr)) {
14344 STRLEN i = 0;
14345 const bool sharekeys = !!HvSHAREKEYS(sstr);
14346 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14347 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14348 char *darray;
14349 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14350 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14351 char);
14352 HvARRAY(dstr) = (HE**)darray;
14353 while (i <= sxhv->xhv_max) {
14354 const HE * const source = HvARRAY(sstr)[i];
14355 HvARRAY(dstr)[i] = source
14356 ? he_dup(source, sharekeys, param) : 0;
14357 ++i;
14358 }
14359 if (SvOOK(sstr)) {
14360 const struct xpvhv_aux * const saux = HvAUX(sstr);
14361 struct xpvhv_aux * const daux = HvAUX(dstr);
14362 /* This flag isn't copied. */
14363 SvOOK_on(dstr);
14364
14365 if (saux->xhv_name_count) {
14366 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14367 const I32 count
14368 = saux->xhv_name_count < 0
14369 ? -saux->xhv_name_count
14370 : saux->xhv_name_count;
14371 HEK **shekp = sname + count;
14372 HEK **dhekp;
14373 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14374 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14375 while (shekp-- > sname) {
14376 dhekp--;
14377 *dhekp = hek_dup(*shekp, param);
14378 }
14379 }
14380 else {
14381 daux->xhv_name_u.xhvnameu_name
14382 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14383 param);
14384 }
14385 daux->xhv_name_count = saux->xhv_name_count;
14386
14387 daux->xhv_aux_flags = saux->xhv_aux_flags;
14388 #ifdef PERL_HASH_RANDOMIZE_KEYS
14389 daux->xhv_rand = saux->xhv_rand;
14390 daux->xhv_last_rand = saux->xhv_last_rand;
14391 #endif
14392 daux->xhv_riter = saux->xhv_riter;
14393 daux->xhv_eiter = saux->xhv_eiter
14394 ? he_dup(saux->xhv_eiter,
14395 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14396 /* backref array needs refcnt=2; see sv_add_backref */
14397 daux->xhv_backreferences =
14398 (param->flags & CLONEf_JOIN_IN)
14399 /* when joining, we let the individual GVs and
14400 * CVs add themselves to backref as
14401 * needed. This avoids pulling in stuff
14402 * that isn't required, and simplifies the
14403 * case where stashes aren't cloned back
14404 * if they already exist in the parent
14405 * thread */
14406 ? NULL
14407 : saux->xhv_backreferences
14408 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14409 ? MUTABLE_AV(SvREFCNT_inc(
14410 sv_dup_inc((const SV *)
14411 saux->xhv_backreferences, param)))
14412 : MUTABLE_AV(sv_dup((const SV *)
14413 saux->xhv_backreferences, param))
14414 : 0;
14415
14416 daux->xhv_mro_meta = saux->xhv_mro_meta
14417 ? mro_meta_dup(saux->xhv_mro_meta, param)
14418 : 0;
14419
14420 /* Record stashes for possible cloning in Perl_clone(). */
14421 if (HvNAME(sstr))
14422 av_push(param->stashes, dstr);
14423 }
14424 }
14425 else
14426 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14427 break;
14428 case SVt_PVCV:
14429 if (!(param->flags & CLONEf_COPY_STACKS)) {
14430 CvDEPTH(dstr) = 0;
14431 }
14432 /* FALLTHROUGH */
14433 case SVt_PVFM:
14434 /* NOTE: not refcounted */
14435 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14436 hv_dup(CvSTASH(dstr), param);
14437 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14438 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14439 if (!CvISXSUB(dstr)) {
14440 OP_REFCNT_LOCK;
14441 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14442 OP_REFCNT_UNLOCK;
14443 CvSLABBED_off(dstr);
14444 } else if (CvCONST(dstr)) {
14445 CvXSUBANY(dstr).any_ptr =
14446 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14447 }
14448 assert(!CvSLABBED(dstr));
14449 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14450 if (CvNAMED(dstr))
14451 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14452 hek_dup(CvNAME_HEK((CV *)sstr), param);
14453 /* don't dup if copying back - CvGV isn't refcounted, so the
14454 * duped GV may never be freed. A bit of a hack! DAPM */
14455 else
14456 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14457 CvCVGV_RC(dstr)
14458 ? gv_dup_inc(CvGV(sstr), param)
14459 : (param->flags & CLONEf_JOIN_IN)
14460 ? NULL
14461 : gv_dup(CvGV(sstr), param);
14462
14463 if (!CvISXSUB(sstr)) {
14464 PADLIST * padlist = CvPADLIST(sstr);
14465 if(padlist)
14466 padlist = padlist_dup(padlist, param);
14467 CvPADLIST_set(dstr, padlist);
14468 } else
14469 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14470 PoisonPADLIST(dstr);
14471
14472 CvOUTSIDE(dstr) =
14473 CvWEAKOUTSIDE(sstr)
14474 ? cv_dup( CvOUTSIDE(dstr), param)
14475 : cv_dup_inc(CvOUTSIDE(dstr), param);
14476 break;
14477 }
14478 }
14479 }
14480
14481 return dstr;
14482 }
14483
14484 SV *
Perl_sv_dup_inc(pTHX_ const SV * const sstr,CLONE_PARAMS * const param)14485 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14486 {
14487 PERL_ARGS_ASSERT_SV_DUP_INC;
14488 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14489 }
14490
14491 SV *
Perl_sv_dup(pTHX_ const SV * const sstr,CLONE_PARAMS * const param)14492 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14493 {
14494 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14495 PERL_ARGS_ASSERT_SV_DUP;
14496
14497 /* Track every SV that (at least initially) had a reference count of 0.
14498 We need to do this by holding an actual reference to it in this array.
14499 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14500 (akin to the stashes hash, and the perl stack), we come unstuck if
14501 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14502 thread) is manipulated in a CLONE method, because CLONE runs before the
14503 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14504 (and fix things up by giving each a reference via the temps stack).
14505 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14506 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14507 before the walk of unreferenced happens and a reference to that is SV
14508 added to the temps stack. At which point we have the same SV considered
14509 to be in use, and free to be re-used. Not good.
14510 */
14511 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14512 assert(param->unreferenced);
14513 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14514 }
14515
14516 return dstr;
14517 }
14518
14519 /* duplicate a context */
14520
14521 PERL_CONTEXT *
Perl_cx_dup(pTHX_ PERL_CONTEXT * cxs,I32 ix,I32 max,CLONE_PARAMS * param)14522 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14523 {
14524 PERL_CONTEXT *ncxs;
14525
14526 PERL_ARGS_ASSERT_CX_DUP;
14527
14528 if (!cxs)
14529 return (PERL_CONTEXT*)NULL;
14530
14531 /* look for it in the table first */
14532 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14533 if (ncxs)
14534 return ncxs;
14535
14536 /* create anew and remember what it is */
14537 Newx(ncxs, max + 1, PERL_CONTEXT);
14538 ptr_table_store(PL_ptr_table, cxs, ncxs);
14539 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14540
14541 while (ix >= 0) {
14542 PERL_CONTEXT * const ncx = &ncxs[ix];
14543 if (CxTYPE(ncx) == CXt_SUBST) {
14544 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14545 }
14546 else {
14547 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14548 switch (CxTYPE(ncx)) {
14549 case CXt_SUB:
14550 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14551 if(CxHASARGS(ncx)){
14552 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14553 } else {
14554 ncx->blk_sub.savearray = NULL;
14555 }
14556 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14557 ncx->blk_sub.prevcomppad);
14558 break;
14559 case CXt_EVAL:
14560 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14561 param);
14562 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14563 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14564 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14565 /* XXX what do do with cur_top_env ???? */
14566 break;
14567 case CXt_LOOP_LAZYSV:
14568 ncx->blk_loop.state_u.lazysv.end
14569 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14570 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14571 duplication code instead.
14572 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14573 actually being the same function, and (2) order
14574 equivalence of the two unions.
14575 We can assert the later [but only at run time :-(] */
14576 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14577 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14578 /* FALLTHROUGH */
14579 case CXt_LOOP_ARY:
14580 ncx->blk_loop.state_u.ary.ary
14581 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14582 /* FALLTHROUGH */
14583 case CXt_LOOP_LIST:
14584 case CXt_LOOP_LAZYIV:
14585 /* code common to all 'for' CXt_LOOP_* types */
14586 ncx->blk_loop.itersave =
14587 sv_dup_inc(ncx->blk_loop.itersave, param);
14588 if (CxPADLOOP(ncx)) {
14589 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14590 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14591 ncx->blk_loop.oldcomppad =
14592 (PAD*)ptr_table_fetch(PL_ptr_table,
14593 ncx->blk_loop.oldcomppad);
14594 ncx->blk_loop.itervar_u.svp =
14595 &CX_CURPAD_SV(ncx->blk_loop, off);
14596 }
14597 else {
14598 /* this copies the GV if CXp_FOR_GV, or the SV for an
14599 * alias (for \$x (...)) - relies on gv_dup being the
14600 * same as sv_dup */
14601 ncx->blk_loop.itervar_u.gv
14602 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14603 param);
14604 }
14605 break;
14606 case CXt_LOOP_PLAIN:
14607 break;
14608 case CXt_FORMAT:
14609 ncx->blk_format.prevcomppad =
14610 (PAD*)ptr_table_fetch(PL_ptr_table,
14611 ncx->blk_format.prevcomppad);
14612 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14613 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14614 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14615 param);
14616 break;
14617 case CXt_GIVEN:
14618 ncx->blk_givwhen.defsv_save =
14619 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14620 break;
14621 case CXt_BLOCK:
14622 case CXt_NULL:
14623 case CXt_WHEN:
14624 break;
14625 }
14626 }
14627 --ix;
14628 }
14629 return ncxs;
14630 }
14631
14632 /* duplicate a stack info structure */
14633
14634 PERL_SI *
Perl_si_dup(pTHX_ PERL_SI * si,CLONE_PARAMS * param)14635 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14636 {
14637 PERL_SI *nsi;
14638
14639 PERL_ARGS_ASSERT_SI_DUP;
14640
14641 if (!si)
14642 return (PERL_SI*)NULL;
14643
14644 /* look for it in the table first */
14645 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14646 if (nsi)
14647 return nsi;
14648
14649 /* create anew and remember what it is */
14650 Newx(nsi, 1, PERL_SI);
14651 ptr_table_store(PL_ptr_table, si, nsi);
14652
14653 nsi->si_stack = av_dup_inc(si->si_stack, param);
14654 nsi->si_cxix = si->si_cxix;
14655 nsi->si_cxmax = si->si_cxmax;
14656 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14657 nsi->si_type = si->si_type;
14658 nsi->si_prev = si_dup(si->si_prev, param);
14659 nsi->si_next = si_dup(si->si_next, param);
14660 nsi->si_markoff = si->si_markoff;
14661 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14662 nsi->si_stack_hwm = 0;
14663 #endif
14664
14665 return nsi;
14666 }
14667
14668 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14669 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14670 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14671 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14672 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14673 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14674 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14675 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14676 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14677 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14678 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14679 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14680 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14681 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14682 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14683 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14684
14685 /* XXXXX todo */
14686 #define pv_dup_inc(p) SAVEPV(p)
14687 #define pv_dup(p) SAVEPV(p)
14688 #define svp_dup_inc(p,pp) any_dup(p,pp)
14689
14690 /* map any object to the new equivent - either something in the
14691 * ptr table, or something in the interpreter structure
14692 */
14693
14694 void *
Perl_any_dup(pTHX_ void * v,const PerlInterpreter * proto_perl)14695 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14696 {
14697 void *ret;
14698
14699 PERL_ARGS_ASSERT_ANY_DUP;
14700
14701 if (!v)
14702 return (void*)NULL;
14703
14704 /* look for it in the table first */
14705 ret = ptr_table_fetch(PL_ptr_table, v);
14706 if (ret)
14707 return ret;
14708
14709 /* see if it is part of the interpreter structure */
14710 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14711 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14712 else {
14713 ret = v;
14714 }
14715
14716 return ret;
14717 }
14718
14719 /* duplicate the save stack */
14720
14721 ANY *
Perl_ss_dup(pTHX_ PerlInterpreter * proto_perl,CLONE_PARAMS * param)14722 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14723 {
14724 dVAR;
14725 ANY * const ss = proto_perl->Isavestack;
14726 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14727 I32 ix = proto_perl->Isavestack_ix;
14728 ANY *nss;
14729 const SV *sv;
14730 const GV *gv;
14731 const AV *av;
14732 const HV *hv;
14733 void* ptr;
14734 int intval;
14735 long longval;
14736 GP *gp;
14737 IV iv;
14738 I32 i;
14739 char *c = NULL;
14740 void (*dptr) (void*);
14741 void (*dxptr) (pTHX_ void*);
14742
14743 PERL_ARGS_ASSERT_SS_DUP;
14744
14745 Newx(nss, max, ANY);
14746
14747 while (ix > 0) {
14748 const UV uv = POPUV(ss,ix);
14749 const U8 type = (U8)uv & SAVE_MASK;
14750
14751 TOPUV(nss,ix) = uv;
14752 switch (type) {
14753 case SAVEt_CLEARSV:
14754 case SAVEt_CLEARPADRANGE:
14755 break;
14756 case SAVEt_HELEM: /* hash element */
14757 case SAVEt_SV: /* scalar reference */
14758 sv = (const SV *)POPPTR(ss,ix);
14759 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14760 /* FALLTHROUGH */
14761 case SAVEt_ITEM: /* normal string */
14762 case SAVEt_GVSV: /* scalar slot in GV */
14763 sv = (const SV *)POPPTR(ss,ix);
14764 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14765 if (type == SAVEt_SV)
14766 break;
14767 /* FALLTHROUGH */
14768 case SAVEt_FREESV:
14769 case SAVEt_MORTALIZESV:
14770 case SAVEt_READONLY_OFF:
14771 sv = (const SV *)POPPTR(ss,ix);
14772 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14773 break;
14774 case SAVEt_FREEPADNAME:
14775 ptr = POPPTR(ss,ix);
14776 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14777 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14778 break;
14779 case SAVEt_SHARED_PVREF: /* char* in shared space */
14780 c = (char*)POPPTR(ss,ix);
14781 TOPPTR(nss,ix) = savesharedpv(c);
14782 ptr = POPPTR(ss,ix);
14783 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14784 break;
14785 case SAVEt_GENERIC_SVREF: /* generic sv */
14786 case SAVEt_SVREF: /* scalar reference */
14787 sv = (const SV *)POPPTR(ss,ix);
14788 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14789 if (type == SAVEt_SVREF)
14790 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14791 ptr = POPPTR(ss,ix);
14792 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14793 break;
14794 case SAVEt_GVSLOT: /* any slot in GV */
14795 sv = (const SV *)POPPTR(ss,ix);
14796 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14797 ptr = POPPTR(ss,ix);
14798 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14799 sv = (const SV *)POPPTR(ss,ix);
14800 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14801 break;
14802 case SAVEt_HV: /* hash reference */
14803 case SAVEt_AV: /* array reference */
14804 sv = (const SV *) POPPTR(ss,ix);
14805 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14806 /* FALLTHROUGH */
14807 case SAVEt_COMPPAD:
14808 case SAVEt_NSTAB:
14809 sv = (const SV *) POPPTR(ss,ix);
14810 TOPPTR(nss,ix) = sv_dup(sv, param);
14811 break;
14812 case SAVEt_INT: /* int reference */
14813 ptr = POPPTR(ss,ix);
14814 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14815 intval = (int)POPINT(ss,ix);
14816 TOPINT(nss,ix) = intval;
14817 break;
14818 case SAVEt_LONG: /* long reference */
14819 ptr = POPPTR(ss,ix);
14820 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14821 longval = (long)POPLONG(ss,ix);
14822 TOPLONG(nss,ix) = longval;
14823 break;
14824 case SAVEt_I32: /* I32 reference */
14825 ptr = POPPTR(ss,ix);
14826 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14827 i = POPINT(ss,ix);
14828 TOPINT(nss,ix) = i;
14829 break;
14830 case SAVEt_IV: /* IV reference */
14831 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14832 ptr = POPPTR(ss,ix);
14833 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14834 iv = POPIV(ss,ix);
14835 TOPIV(nss,ix) = iv;
14836 break;
14837 case SAVEt_TMPSFLOOR:
14838 iv = POPIV(ss,ix);
14839 TOPIV(nss,ix) = iv;
14840 break;
14841 case SAVEt_HPTR: /* HV* reference */
14842 case SAVEt_APTR: /* AV* reference */
14843 case SAVEt_SPTR: /* SV* reference */
14844 ptr = POPPTR(ss,ix);
14845 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14846 sv = (const SV *)POPPTR(ss,ix);
14847 TOPPTR(nss,ix) = sv_dup(sv, param);
14848 break;
14849 case SAVEt_VPTR: /* random* reference */
14850 ptr = POPPTR(ss,ix);
14851 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14852 /* FALLTHROUGH */
14853 case SAVEt_INT_SMALL:
14854 case SAVEt_I32_SMALL:
14855 case SAVEt_I16: /* I16 reference */
14856 case SAVEt_I8: /* I8 reference */
14857 case SAVEt_BOOL:
14858 ptr = POPPTR(ss,ix);
14859 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14860 break;
14861 case SAVEt_GENERIC_PVREF: /* generic char* */
14862 case SAVEt_PPTR: /* char* reference */
14863 ptr = POPPTR(ss,ix);
14864 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14865 c = (char*)POPPTR(ss,ix);
14866 TOPPTR(nss,ix) = pv_dup(c);
14867 break;
14868 case SAVEt_GP: /* scalar reference */
14869 gp = (GP*)POPPTR(ss,ix);
14870 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14871 (void)GpREFCNT_inc(gp);
14872 gv = (const GV *)POPPTR(ss,ix);
14873 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14874 break;
14875 case SAVEt_FREEOP:
14876 ptr = POPPTR(ss,ix);
14877 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14878 /* these are assumed to be refcounted properly */
14879 OP *o;
14880 switch (((OP*)ptr)->op_type) {
14881 case OP_LEAVESUB:
14882 case OP_LEAVESUBLV:
14883 case OP_LEAVEEVAL:
14884 case OP_LEAVE:
14885 case OP_SCOPE:
14886 case OP_LEAVEWRITE:
14887 TOPPTR(nss,ix) = ptr;
14888 o = (OP*)ptr;
14889 OP_REFCNT_LOCK;
14890 (void) OpREFCNT_inc(o);
14891 OP_REFCNT_UNLOCK;
14892 break;
14893 default:
14894 TOPPTR(nss,ix) = NULL;
14895 break;
14896 }
14897 }
14898 else
14899 TOPPTR(nss,ix) = NULL;
14900 break;
14901 case SAVEt_FREECOPHH:
14902 ptr = POPPTR(ss,ix);
14903 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14904 break;
14905 case SAVEt_ADELETE:
14906 av = (const AV *)POPPTR(ss,ix);
14907 TOPPTR(nss,ix) = av_dup_inc(av, param);
14908 i = POPINT(ss,ix);
14909 TOPINT(nss,ix) = i;
14910 break;
14911 case SAVEt_DELETE:
14912 hv = (const HV *)POPPTR(ss,ix);
14913 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14914 i = POPINT(ss,ix);
14915 TOPINT(nss,ix) = i;
14916 /* FALLTHROUGH */
14917 case SAVEt_FREEPV:
14918 c = (char*)POPPTR(ss,ix);
14919 TOPPTR(nss,ix) = pv_dup_inc(c);
14920 break;
14921 case SAVEt_STACK_POS: /* Position on Perl stack */
14922 i = POPINT(ss,ix);
14923 TOPINT(nss,ix) = i;
14924 break;
14925 case SAVEt_DESTRUCTOR:
14926 ptr = POPPTR(ss,ix);
14927 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14928 dptr = POPDPTR(ss,ix);
14929 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14930 any_dup(FPTR2DPTR(void *, dptr),
14931 proto_perl));
14932 break;
14933 case SAVEt_DESTRUCTOR_X:
14934 ptr = POPPTR(ss,ix);
14935 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14936 dxptr = POPDXPTR(ss,ix);
14937 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14938 any_dup(FPTR2DPTR(void *, dxptr),
14939 proto_perl));
14940 break;
14941 case SAVEt_REGCONTEXT:
14942 case SAVEt_ALLOC:
14943 ix -= uv >> SAVE_TIGHT_SHIFT;
14944 break;
14945 case SAVEt_AELEM: /* array element */
14946 sv = (const SV *)POPPTR(ss,ix);
14947 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14948 iv = POPIV(ss,ix);
14949 TOPIV(nss,ix) = iv;
14950 av = (const AV *)POPPTR(ss,ix);
14951 TOPPTR(nss,ix) = av_dup_inc(av, param);
14952 break;
14953 case SAVEt_OP:
14954 ptr = POPPTR(ss,ix);
14955 TOPPTR(nss,ix) = ptr;
14956 break;
14957 case SAVEt_HINTS:
14958 ptr = POPPTR(ss,ix);
14959 ptr = cophh_copy((COPHH*)ptr);
14960 TOPPTR(nss,ix) = ptr;
14961 i = POPINT(ss,ix);
14962 TOPINT(nss,ix) = i;
14963 if (i & HINT_LOCALIZE_HH) {
14964 hv = (const HV *)POPPTR(ss,ix);
14965 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14966 }
14967 break;
14968 case SAVEt_PADSV_AND_MORTALIZE:
14969 longval = (long)POPLONG(ss,ix);
14970 TOPLONG(nss,ix) = longval;
14971 ptr = POPPTR(ss,ix);
14972 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14973 sv = (const SV *)POPPTR(ss,ix);
14974 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14975 break;
14976 case SAVEt_SET_SVFLAGS:
14977 i = POPINT(ss,ix);
14978 TOPINT(nss,ix) = i;
14979 i = POPINT(ss,ix);
14980 TOPINT(nss,ix) = i;
14981 sv = (const SV *)POPPTR(ss,ix);
14982 TOPPTR(nss,ix) = sv_dup(sv, param);
14983 break;
14984 case SAVEt_COMPILE_WARNINGS:
14985 ptr = POPPTR(ss,ix);
14986 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14987 break;
14988 case SAVEt_PARSER:
14989 ptr = POPPTR(ss,ix);
14990 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14991 break;
14992 default:
14993 Perl_croak(aTHX_
14994 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14995 }
14996 }
14997
14998 return nss;
14999 }
15000
15001
15002 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15003 * flag to the result. This is done for each stash before cloning starts,
15004 * so we know which stashes want their objects cloned */
15005
15006 static void
do_mark_cloneable_stash(pTHX_ SV * const sv)15007 do_mark_cloneable_stash(pTHX_ SV *const sv)
15008 {
15009 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15010 if (hvname) {
15011 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15012 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15013 if (cloner && GvCV(cloner)) {
15014 dSP;
15015 UV status;
15016
15017 ENTER;
15018 SAVETMPS;
15019 PUSHMARK(SP);
15020 mXPUSHs(newSVhek(hvname));
15021 PUTBACK;
15022 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15023 SPAGAIN;
15024 status = POPu;
15025 PUTBACK;
15026 FREETMPS;
15027 LEAVE;
15028 if (status)
15029 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15030 }
15031 }
15032 }
15033
15034
15035
15036 /*
15037 =for apidoc perl_clone
15038
15039 Create and return a new interpreter by cloning the current one.
15040
15041 C<perl_clone> takes these flags as parameters:
15042
15043 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15044 without it we only clone the data and zero the stacks,
15045 with it we copy the stacks and the new perl interpreter is
15046 ready to run at the exact same point as the previous one.
15047 The pseudo-fork code uses C<COPY_STACKS> while the
15048 threads->create doesn't.
15049
15050 C<CLONEf_KEEP_PTR_TABLE> -
15051 C<perl_clone> keeps a ptr_table with the pointer of the old
15052 variable as a key and the new variable as a value,
15053 this allows it to check if something has been cloned and not
15054 clone it again but rather just use the value and increase the
15055 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15056 the ptr_table using the function
15057 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15058 reason to keep it around is if you want to dup some of your own
15059 variable who are outside the graph perl scans, an example of this
15060 code is in F<threads.xs> create.
15061
15062 C<CLONEf_CLONE_HOST> -
15063 This is a win32 thing, it is ignored on unix, it tells perls
15064 win32host code (which is c++) to clone itself, this is needed on
15065 win32 if you want to run two threads at the same time,
15066 if you just want to do some stuff in a separate perl interpreter
15067 and then throw it away and return to the original one,
15068 you don't need to do anything.
15069
15070 =cut
15071 */
15072
15073 /* XXX the above needs expanding by someone who actually understands it ! */
15074 EXTERN_C PerlInterpreter *
15075 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15076
15077 PerlInterpreter *
perl_clone(PerlInterpreter * proto_perl,UV flags)15078 perl_clone(PerlInterpreter *proto_perl, UV flags)
15079 {
15080 dVAR;
15081 #ifdef PERL_IMPLICIT_SYS
15082
15083 PERL_ARGS_ASSERT_PERL_CLONE;
15084
15085 /* perlhost.h so we need to call into it
15086 to clone the host, CPerlHost should have a c interface, sky */
15087
15088 #ifndef __amigaos4__
15089 if (flags & CLONEf_CLONE_HOST) {
15090 return perl_clone_host(proto_perl,flags);
15091 }
15092 #endif
15093 return perl_clone_using(proto_perl, flags,
15094 proto_perl->IMem,
15095 proto_perl->IMemShared,
15096 proto_perl->IMemParse,
15097 proto_perl->IEnv,
15098 proto_perl->IStdIO,
15099 proto_perl->ILIO,
15100 proto_perl->IDir,
15101 proto_perl->ISock,
15102 proto_perl->IProc);
15103 }
15104
15105 PerlInterpreter *
perl_clone_using(PerlInterpreter * proto_perl,UV flags,struct IPerlMem * ipM,struct IPerlMem * ipMS,struct IPerlMem * ipMP,struct IPerlEnv * ipE,struct IPerlStdIO * ipStd,struct IPerlLIO * ipLIO,struct IPerlDir * ipD,struct IPerlSock * ipS,struct IPerlProc * ipP)15106 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15107 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15108 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15109 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15110 struct IPerlDir* ipD, struct IPerlSock* ipS,
15111 struct IPerlProc* ipP)
15112 {
15113 /* XXX many of the string copies here can be optimized if they're
15114 * constants; they need to be allocated as common memory and just
15115 * their pointers copied. */
15116
15117 IV i;
15118 CLONE_PARAMS clone_params;
15119 CLONE_PARAMS* const param = &clone_params;
15120
15121 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15122
15123 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15124 #else /* !PERL_IMPLICIT_SYS */
15125 IV i;
15126 CLONE_PARAMS clone_params;
15127 CLONE_PARAMS* param = &clone_params;
15128 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15129
15130 PERL_ARGS_ASSERT_PERL_CLONE;
15131 #endif /* PERL_IMPLICIT_SYS */
15132
15133 /* for each stash, determine whether its objects should be cloned */
15134 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15135 PERL_SET_THX(my_perl);
15136
15137 #ifdef DEBUGGING
15138 PoisonNew(my_perl, 1, PerlInterpreter);
15139 PL_op = NULL;
15140 PL_curcop = NULL;
15141 PL_defstash = NULL; /* may be used by perl malloc() */
15142 PL_markstack = 0;
15143 PL_scopestack = 0;
15144 PL_scopestack_name = 0;
15145 PL_savestack = 0;
15146 PL_savestack_ix = 0;
15147 PL_savestack_max = -1;
15148 PL_sig_pending = 0;
15149 PL_parser = NULL;
15150 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15151 Zero(&PL_padname_undef, 1, PADNAME);
15152 Zero(&PL_padname_const, 1, PADNAME);
15153 # ifdef DEBUG_LEAKING_SCALARS
15154 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15155 # endif
15156 # ifdef PERL_TRACE_OPS
15157 Zero(PL_op_exec_cnt, OP_max+2, UV);
15158 # endif
15159 #else /* !DEBUGGING */
15160 Zero(my_perl, 1, PerlInterpreter);
15161 #endif /* DEBUGGING */
15162
15163 #ifdef PERL_IMPLICIT_SYS
15164 /* host pointers */
15165 PL_Mem = ipM;
15166 PL_MemShared = ipMS;
15167 PL_MemParse = ipMP;
15168 PL_Env = ipE;
15169 PL_StdIO = ipStd;
15170 PL_LIO = ipLIO;
15171 PL_Dir = ipD;
15172 PL_Sock = ipS;
15173 PL_Proc = ipP;
15174 #endif /* PERL_IMPLICIT_SYS */
15175
15176
15177 param->flags = flags;
15178 /* Nothing in the core code uses this, but we make it available to
15179 extensions (using mg_dup). */
15180 param->proto_perl = proto_perl;
15181 /* Likely nothing will use this, but it is initialised to be consistent
15182 with Perl_clone_params_new(). */
15183 param->new_perl = my_perl;
15184 param->unreferenced = NULL;
15185
15186
15187 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15188
15189 PL_body_arenas = NULL;
15190 Zero(&PL_body_roots, 1, PL_body_roots);
15191
15192 PL_sv_count = 0;
15193 PL_sv_root = NULL;
15194 PL_sv_arenaroot = NULL;
15195
15196 PL_debug = proto_perl->Idebug;
15197
15198 /* dbargs array probably holds garbage */
15199 PL_dbargs = NULL;
15200
15201 PL_compiling = proto_perl->Icompiling;
15202
15203 /* pseudo environmental stuff */
15204 PL_origargc = proto_perl->Iorigargc;
15205 PL_origargv = proto_perl->Iorigargv;
15206
15207 #ifndef NO_TAINT_SUPPORT
15208 /* Set tainting stuff before PerlIO_debug can possibly get called */
15209 PL_tainting = proto_perl->Itainting;
15210 PL_taint_warn = proto_perl->Itaint_warn;
15211 #else
15212 PL_tainting = FALSE;
15213 PL_taint_warn = FALSE;
15214 #endif
15215
15216 PL_minus_c = proto_perl->Iminus_c;
15217
15218 PL_localpatches = proto_perl->Ilocalpatches;
15219 PL_splitstr = proto_perl->Isplitstr;
15220 PL_minus_n = proto_perl->Iminus_n;
15221 PL_minus_p = proto_perl->Iminus_p;
15222 PL_minus_l = proto_perl->Iminus_l;
15223 PL_minus_a = proto_perl->Iminus_a;
15224 PL_minus_E = proto_perl->Iminus_E;
15225 PL_minus_F = proto_perl->Iminus_F;
15226 PL_doswitches = proto_perl->Idoswitches;
15227 PL_dowarn = proto_perl->Idowarn;
15228 #ifdef PERL_SAWAMPERSAND
15229 PL_sawampersand = proto_perl->Isawampersand;
15230 #endif
15231 PL_unsafe = proto_perl->Iunsafe;
15232 PL_perldb = proto_perl->Iperldb;
15233 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15234 PL_exit_flags = proto_perl->Iexit_flags;
15235
15236 /* XXX time(&PL_basetime) when asked for? */
15237 PL_basetime = proto_perl->Ibasetime;
15238
15239 PL_maxsysfd = proto_perl->Imaxsysfd;
15240 PL_statusvalue = proto_perl->Istatusvalue;
15241 #ifdef __VMS
15242 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15243 #else
15244 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15245 #endif
15246
15247 /* RE engine related */
15248 PL_regmatch_slab = NULL;
15249 PL_reg_curpm = NULL;
15250
15251 PL_sub_generation = proto_perl->Isub_generation;
15252
15253 /* funky return mechanisms */
15254 PL_forkprocess = proto_perl->Iforkprocess;
15255
15256 /* internal state */
15257 PL_main_start = proto_perl->Imain_start;
15258 PL_eval_root = proto_perl->Ieval_root;
15259 PL_eval_start = proto_perl->Ieval_start;
15260
15261 PL_filemode = proto_perl->Ifilemode;
15262 PL_lastfd = proto_perl->Ilastfd;
15263 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15264 PL_gensym = proto_perl->Igensym;
15265
15266 PL_laststatval = proto_perl->Ilaststatval;
15267 PL_laststype = proto_perl->Ilaststype;
15268 PL_mess_sv = NULL;
15269
15270 PL_profiledata = NULL;
15271
15272 PL_generation = proto_perl->Igeneration;
15273
15274 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15275 PL_in_clean_all = proto_perl->Iin_clean_all;
15276
15277 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15278 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15279 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15280 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15281 PL_nomemok = proto_perl->Inomemok;
15282 PL_an = proto_perl->Ian;
15283 PL_evalseq = proto_perl->Ievalseq;
15284 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15285 PL_origalen = proto_perl->Iorigalen;
15286
15287 PL_sighandlerp = proto_perl->Isighandlerp;
15288
15289 PL_runops = proto_perl->Irunops;
15290
15291 PL_subline = proto_perl->Isubline;
15292
15293 PL_cv_has_eval = proto_perl->Icv_has_eval;
15294
15295 #ifdef FCRYPT
15296 PL_cryptseen = proto_perl->Icryptseen;
15297 #endif
15298
15299 #ifdef USE_LOCALE_COLLATE
15300 PL_collation_ix = proto_perl->Icollation_ix;
15301 PL_collation_standard = proto_perl->Icollation_standard;
15302 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15303 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15304 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15305 #endif /* USE_LOCALE_COLLATE */
15306
15307 #ifdef USE_LOCALE_NUMERIC
15308 PL_numeric_standard = proto_perl->Inumeric_standard;
15309 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15310 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15311 #endif /* !USE_LOCALE_NUMERIC */
15312
15313 /* Did the locale setup indicate UTF-8? */
15314 PL_utf8locale = proto_perl->Iutf8locale;
15315 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15316 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15317 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15318 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15319 PL_lc_numeric_mutex_depth = 0;
15320 #endif
15321 /* Unicode features (see perlrun/-C) */
15322 PL_unicode = proto_perl->Iunicode;
15323
15324 /* Pre-5.8 signals control */
15325 PL_signals = proto_perl->Isignals;
15326
15327 /* times() ticks per second */
15328 PL_clocktick = proto_perl->Iclocktick;
15329
15330 /* Recursion stopper for PerlIO_find_layer */
15331 PL_in_load_module = proto_perl->Iin_load_module;
15332
15333 /* sort() routine */
15334 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15335
15336 /* Not really needed/useful since the reenrant_retint is "volatile",
15337 * but do it for consistency's sake. */
15338 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15339
15340 /* Hooks to shared SVs and locks. */
15341 PL_sharehook = proto_perl->Isharehook;
15342 PL_lockhook = proto_perl->Ilockhook;
15343 PL_unlockhook = proto_perl->Iunlockhook;
15344 PL_threadhook = proto_perl->Ithreadhook;
15345 PL_destroyhook = proto_perl->Idestroyhook;
15346 PL_signalhook = proto_perl->Isignalhook;
15347
15348 PL_globhook = proto_perl->Iglobhook;
15349
15350 /* swatch cache */
15351 PL_last_swash_hv = NULL; /* reinits on demand */
15352 PL_last_swash_klen = 0;
15353 PL_last_swash_key[0]= '\0';
15354 PL_last_swash_tmps = (U8*)NULL;
15355 PL_last_swash_slen = 0;
15356
15357 PL_srand_called = proto_perl->Isrand_called;
15358 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15359
15360 if (flags & CLONEf_COPY_STACKS) {
15361 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15362 PL_tmps_ix = proto_perl->Itmps_ix;
15363 PL_tmps_max = proto_perl->Itmps_max;
15364 PL_tmps_floor = proto_perl->Itmps_floor;
15365
15366 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15367 * NOTE: unlike the others! */
15368 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15369 PL_scopestack_max = proto_perl->Iscopestack_max;
15370
15371 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15372 * NOTE: unlike the others! */
15373 PL_savestack_ix = proto_perl->Isavestack_ix;
15374 PL_savestack_max = proto_perl->Isavestack_max;
15375 }
15376
15377 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15378 PL_top_env = &PL_start_env;
15379
15380 PL_op = proto_perl->Iop;
15381
15382 PL_Sv = NULL;
15383 PL_Xpv = (XPV*)NULL;
15384 my_perl->Ina = proto_perl->Ina;
15385
15386 PL_statcache = proto_perl->Istatcache;
15387
15388 #ifndef NO_TAINT_SUPPORT
15389 PL_tainted = proto_perl->Itainted;
15390 #else
15391 PL_tainted = FALSE;
15392 #endif
15393 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15394
15395 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15396
15397 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15398 PL_restartop = proto_perl->Irestartop;
15399 PL_in_eval = proto_perl->Iin_eval;
15400 PL_delaymagic = proto_perl->Idelaymagic;
15401 PL_phase = proto_perl->Iphase;
15402 PL_localizing = proto_perl->Ilocalizing;
15403
15404 PL_hv_fetch_ent_mh = NULL;
15405 PL_modcount = proto_perl->Imodcount;
15406 PL_lastgotoprobe = NULL;
15407 PL_dumpindent = proto_perl->Idumpindent;
15408
15409 PL_efloatbuf = NULL; /* reinits on demand */
15410 PL_efloatsize = 0; /* reinits on demand */
15411
15412 /* regex stuff */
15413
15414 PL_colorset = 0; /* reinits PL_colors[] */
15415 /*PL_colors[6] = {0,0,0,0,0,0};*/
15416
15417 /* Pluggable optimizer */
15418 PL_peepp = proto_perl->Ipeepp;
15419 PL_rpeepp = proto_perl->Irpeepp;
15420 /* op_free() hook */
15421 PL_opfreehook = proto_perl->Iopfreehook;
15422
15423 #ifdef USE_REENTRANT_API
15424 /* XXX: things like -Dm will segfault here in perlio, but doing
15425 * PERL_SET_CONTEXT(proto_perl);
15426 * breaks too many other things
15427 */
15428 Perl_reentrant_init(aTHX);
15429 #endif
15430
15431 /* create SV map for pointer relocation */
15432 PL_ptr_table = ptr_table_new();
15433
15434 /* initialize these special pointers as early as possible */
15435 init_constants();
15436 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15437 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15438 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15439 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15440 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15441 &PL_padname_const);
15442
15443 /* create (a non-shared!) shared string table */
15444 PL_strtab = newHV();
15445 HvSHAREKEYS_off(PL_strtab);
15446 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15447 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15448
15449 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15450
15451 /* This PV will be free'd special way so must set it same way op.c does */
15452 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15453 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15454
15455 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15456 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15457 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15458 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15459
15460 param->stashes = newAV(); /* Setup array of objects to call clone on */
15461 /* This makes no difference to the implementation, as it always pushes
15462 and shifts pointers to other SVs without changing their reference
15463 count, with the array becoming empty before it is freed. However, it
15464 makes it conceptually clear what is going on, and will avoid some
15465 work inside av.c, filling slots between AvFILL() and AvMAX() with
15466 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15467 AvREAL_off(param->stashes);
15468
15469 if (!(flags & CLONEf_COPY_STACKS)) {
15470 param->unreferenced = newAV();
15471 }
15472
15473 #ifdef PERLIO_LAYERS
15474 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15475 PerlIO_clone(aTHX_ proto_perl, param);
15476 #endif
15477
15478 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15479 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15480 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15481 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15482 PL_xsubfilename = proto_perl->Ixsubfilename;
15483 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15484 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15485
15486 /* switches */
15487 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15488 PL_inplace = SAVEPV(proto_perl->Iinplace);
15489 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15490
15491 /* magical thingies */
15492
15493 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15494 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15495 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15496
15497
15498 /* Clone the regex array */
15499 /* ORANGE FIXME for plugins, probably in the SV dup code.
15500 newSViv(PTR2IV(CALLREGDUPE(
15501 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15502 */
15503 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15504 PL_regex_pad = AvARRAY(PL_regex_padav);
15505
15506 PL_stashpadmax = proto_perl->Istashpadmax;
15507 PL_stashpadix = proto_perl->Istashpadix ;
15508 Newx(PL_stashpad, PL_stashpadmax, HV *);
15509 {
15510 PADOFFSET o = 0;
15511 for (; o < PL_stashpadmax; ++o)
15512 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15513 }
15514
15515 /* shortcuts to various I/O objects */
15516 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15517 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15518 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15519 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15520 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15521 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15522 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15523
15524 /* shortcuts to regexp stuff */
15525 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15526
15527 /* shortcuts to misc objects */
15528 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15529
15530 /* shortcuts to debugging objects */
15531 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15532 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15533 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15534 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15535 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15536 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15537 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15538
15539 /* symbol tables */
15540 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15541 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15542 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15543 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15544 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15545
15546 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15547 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15548 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15549 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15550 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15551 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15552 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15553 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15554 PL_savebegin = proto_perl->Isavebegin;
15555
15556 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15557
15558 /* subprocess state */
15559 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15560
15561 if (proto_perl->Iop_mask)
15562 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15563 else
15564 PL_op_mask = NULL;
15565 /* PL_asserting = proto_perl->Iasserting; */
15566
15567 /* current interpreter roots */
15568 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15569 OP_REFCNT_LOCK;
15570 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15571 OP_REFCNT_UNLOCK;
15572
15573 /* runtime control stuff */
15574 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15575
15576 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15577
15578 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15579
15580 /* interpreter atexit processing */
15581 PL_exitlistlen = proto_perl->Iexitlistlen;
15582 if (PL_exitlistlen) {
15583 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15584 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15585 }
15586 else
15587 PL_exitlist = (PerlExitListEntry*)NULL;
15588
15589 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15590 if (PL_my_cxt_size) {
15591 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15592 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15593 }
15594 else {
15595 PL_my_cxt_list = (void**)NULL;
15596 }
15597 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15598 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15599 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15600 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15601
15602 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15603
15604 PAD_CLONE_VARS(proto_perl, param);
15605
15606 #ifdef HAVE_INTERP_INTERN
15607 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15608 #endif
15609
15610 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15611
15612 #ifdef PERL_USES_PL_PIDSTATUS
15613 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15614 #endif
15615 PL_osname = SAVEPV(proto_perl->Iosname);
15616 PL_parser = parser_dup(proto_perl->Iparser, param);
15617
15618 /* XXX this only works if the saved cop has already been cloned */
15619 if (proto_perl->Iparser) {
15620 PL_parser->saved_curcop = (COP*)any_dup(
15621 proto_perl->Iparser->saved_curcop,
15622 proto_perl);
15623 }
15624
15625 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15626
15627 #if defined(USE_POSIX_2008_LOCALE) \
15628 && defined(USE_THREAD_SAFE_LOCALE) \
15629 && ! defined(HAS_QUERYLOCALE)
15630 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15631 PL_curlocales[i] = savepv("."); /* An illegal value */
15632 }
15633 #endif
15634 #ifdef USE_LOCALE_CTYPE
15635 /* Should we warn if uses locale? */
15636 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15637 #endif
15638
15639 #ifdef USE_LOCALE_COLLATE
15640 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15641 #endif /* USE_LOCALE_COLLATE */
15642
15643 #ifdef USE_LOCALE_NUMERIC
15644 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15645 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15646
15647 # if defined(HAS_POSIX_2008_LOCALE)
15648 PL_underlying_numeric_obj = NULL;
15649 # endif
15650 #endif /* !USE_LOCALE_NUMERIC */
15651
15652 PL_langinfo_buf = NULL;
15653 PL_langinfo_bufsize = 0;
15654
15655 PL_setlocale_buf = NULL;
15656 PL_setlocale_bufsize = 0;
15657
15658 /* utf8 character class swashes */
15659 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15660
15661 if (proto_perl->Ipsig_pend) {
15662 Newxz(PL_psig_pend, SIG_SIZE, int);
15663 }
15664 else {
15665 PL_psig_pend = (int*)NULL;
15666 }
15667
15668 if (proto_perl->Ipsig_name) {
15669 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15670 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15671 param);
15672 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15673 }
15674 else {
15675 PL_psig_ptr = (SV**)NULL;
15676 PL_psig_name = (SV**)NULL;
15677 }
15678
15679 if (flags & CLONEf_COPY_STACKS) {
15680 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15681 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15682 PL_tmps_ix+1, param);
15683
15684 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15685 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15686 Newx(PL_markstack, i, I32);
15687 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15688 - proto_perl->Imarkstack);
15689 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15690 - proto_perl->Imarkstack);
15691 Copy(proto_perl->Imarkstack, PL_markstack,
15692 PL_markstack_ptr - PL_markstack + 1, I32);
15693
15694 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15695 * NOTE: unlike the others! */
15696 Newx(PL_scopestack, PL_scopestack_max, I32);
15697 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15698
15699 #ifdef DEBUGGING
15700 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15701 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15702 #endif
15703 /* reset stack AV to correct length before its duped via
15704 * PL_curstackinfo */
15705 AvFILLp(proto_perl->Icurstack) =
15706 proto_perl->Istack_sp - proto_perl->Istack_base;
15707
15708 /* NOTE: si_dup() looks at PL_markstack */
15709 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15710
15711 /* PL_curstack = PL_curstackinfo->si_stack; */
15712 PL_curstack = av_dup(proto_perl->Icurstack, param);
15713 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15714
15715 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15716 PL_stack_base = AvARRAY(PL_curstack);
15717 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15718 - proto_perl->Istack_base);
15719 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15720
15721 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15722 PL_savestack = ss_dup(proto_perl, param);
15723 }
15724 else {
15725 init_stacks();
15726 ENTER; /* perl_destruct() wants to LEAVE; */
15727 }
15728
15729 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15730 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15731
15732 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15733 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15734 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15735 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15736 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15737 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15738
15739 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15740
15741 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15742 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15743 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15744
15745 PL_stashcache = newHV();
15746
15747 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15748 proto_perl->Iwatchaddr);
15749 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15750 if (PL_debug && PL_watchaddr) {
15751 PerlIO_printf(Perl_debug_log,
15752 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15753 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15754 PTR2UV(PL_watchok));
15755 }
15756
15757 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15758 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15759
15760 /* Call the ->CLONE method, if it exists, for each of the stashes
15761 identified by sv_dup() above.
15762 */
15763 while(av_tindex(param->stashes) != -1) {
15764 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15765 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15766 if (cloner && GvCV(cloner)) {
15767 dSP;
15768 ENTER;
15769 SAVETMPS;
15770 PUSHMARK(SP);
15771 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15772 PUTBACK;
15773 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15774 FREETMPS;
15775 LEAVE;
15776 }
15777 }
15778
15779 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15780 ptr_table_free(PL_ptr_table);
15781 PL_ptr_table = NULL;
15782 }
15783
15784 if (!(flags & CLONEf_COPY_STACKS)) {
15785 unreferenced_to_tmp_stack(param->unreferenced);
15786 }
15787
15788 SvREFCNT_dec(param->stashes);
15789
15790 /* orphaned? eg threads->new inside BEGIN or use */
15791 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15792 SvREFCNT_inc_simple_void(PL_compcv);
15793 SAVEFREESV(PL_compcv);
15794 }
15795
15796 return my_perl;
15797 }
15798
15799 static void
S_unreferenced_to_tmp_stack(pTHX_ AV * const unreferenced)15800 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15801 {
15802 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15803
15804 if (AvFILLp(unreferenced) > -1) {
15805 SV **svp = AvARRAY(unreferenced);
15806 SV **const last = svp + AvFILLp(unreferenced);
15807 SSize_t count = 0;
15808
15809 do {
15810 if (SvREFCNT(*svp) == 1)
15811 ++count;
15812 } while (++svp <= last);
15813
15814 EXTEND_MORTAL(count);
15815 svp = AvARRAY(unreferenced);
15816
15817 do {
15818 if (SvREFCNT(*svp) == 1) {
15819 /* Our reference is the only one to this SV. This means that
15820 in this thread, the scalar effectively has a 0 reference.
15821 That doesn't work (cleanup never happens), so donate our
15822 reference to it onto the save stack. */
15823 PL_tmps_stack[++PL_tmps_ix] = *svp;
15824 } else {
15825 /* As an optimisation, because we are already walking the
15826 entire array, instead of above doing either
15827 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15828 release our reference to the scalar, so that at the end of
15829 the array owns zero references to the scalars it happens to
15830 point to. We are effectively converting the array from
15831 AvREAL() on to AvREAL() off. This saves the av_clear()
15832 (triggered by the SvREFCNT_dec(unreferenced) below) from
15833 walking the array a second time. */
15834 SvREFCNT_dec(*svp);
15835 }
15836
15837 } while (++svp <= last);
15838 AvREAL_off(unreferenced);
15839 }
15840 SvREFCNT_dec_NN(unreferenced);
15841 }
15842
15843 void
Perl_clone_params_del(CLONE_PARAMS * param)15844 Perl_clone_params_del(CLONE_PARAMS *param)
15845 {
15846 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15847 happy: */
15848 PerlInterpreter *const to = param->new_perl;
15849 dTHXa(to);
15850 PerlInterpreter *const was = PERL_GET_THX;
15851
15852 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15853
15854 if (was != to) {
15855 PERL_SET_THX(to);
15856 }
15857
15858 SvREFCNT_dec(param->stashes);
15859 if (param->unreferenced)
15860 unreferenced_to_tmp_stack(param->unreferenced);
15861
15862 Safefree(param);
15863
15864 if (was != to) {
15865 PERL_SET_THX(was);
15866 }
15867 }
15868
15869 CLONE_PARAMS *
Perl_clone_params_new(PerlInterpreter * const from,PerlInterpreter * const to)15870 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15871 {
15872 dVAR;
15873 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15874 does a dTHX; to get the context from thread local storage.
15875 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15876 a version that passes in my_perl. */
15877 PerlInterpreter *const was = PERL_GET_THX;
15878 CLONE_PARAMS *param;
15879
15880 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15881
15882 if (was != to) {
15883 PERL_SET_THX(to);
15884 }
15885
15886 /* Given that we've set the context, we can do this unshared. */
15887 Newx(param, 1, CLONE_PARAMS);
15888
15889 param->flags = 0;
15890 param->proto_perl = from;
15891 param->new_perl = to;
15892 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15893 AvREAL_off(param->stashes);
15894 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15895
15896 if (was != to) {
15897 PERL_SET_THX(was);
15898 }
15899 return param;
15900 }
15901
15902 #endif /* USE_ITHREADS */
15903
15904 void
Perl_init_constants(pTHX)15905 Perl_init_constants(pTHX)
15906 {
15907 dVAR;
15908
15909 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15910 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15911 SvANY(&PL_sv_undef) = NULL;
15912
15913 SvANY(&PL_sv_no) = new_XPVNV();
15914 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15915 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15916 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15917 |SVp_POK|SVf_POK;
15918
15919 SvANY(&PL_sv_yes) = new_XPVNV();
15920 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15921 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15922 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15923 |SVp_POK|SVf_POK;
15924
15925 SvANY(&PL_sv_zero) = new_XPVNV();
15926 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15927 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15928 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15929 |SVp_POK|SVf_POK
15930 |SVs_PADTMP;
15931
15932 SvPV_set(&PL_sv_no, (char*)PL_No);
15933 SvCUR_set(&PL_sv_no, 0);
15934 SvLEN_set(&PL_sv_no, 0);
15935 SvIV_set(&PL_sv_no, 0);
15936 SvNV_set(&PL_sv_no, 0);
15937
15938 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15939 SvCUR_set(&PL_sv_yes, 1);
15940 SvLEN_set(&PL_sv_yes, 0);
15941 SvIV_set(&PL_sv_yes, 1);
15942 SvNV_set(&PL_sv_yes, 1);
15943
15944 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15945 SvCUR_set(&PL_sv_zero, 1);
15946 SvLEN_set(&PL_sv_zero, 0);
15947 SvIV_set(&PL_sv_zero, 0);
15948 SvNV_set(&PL_sv_zero, 0);
15949
15950 PadnamePV(&PL_padname_const) = (char *)PL_No;
15951
15952 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15953 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15954 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15955 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15956
15957 assert(SvIMMORTAL(&PL_sv_yes));
15958 assert(SvIMMORTAL(&PL_sv_undef));
15959 assert(SvIMMORTAL(&PL_sv_no));
15960 assert(SvIMMORTAL(&PL_sv_zero));
15961
15962 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15963 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15964 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15965 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15966
15967 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15968 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
15969 assert(!SvTRUE_nomg_NN(&PL_sv_no));
15970 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
15971 }
15972
15973 /*
15974 =head1 Unicode Support
15975
15976 =for apidoc sv_recode_to_utf8
15977
15978 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15979 of C<sv> is assumed to be octets in that encoding, and C<sv>
15980 will be converted into Unicode (and UTF-8).
15981
15982 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15983 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15984 an C<Encode::XS> Encoding object, bad things will happen.
15985 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15986
15987 The PV of C<sv> is returned.
15988
15989 =cut */
15990
15991 char *
Perl_sv_recode_to_utf8(pTHX_ SV * sv,SV * encoding)15992 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15993 {
15994 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15995
15996 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15997 SV *uni;
15998 STRLEN len;
15999 const char *s;
16000 dSP;
16001 SV *nsv = sv;
16002 ENTER;
16003 PUSHSTACK;
16004 SAVETMPS;
16005 if (SvPADTMP(nsv)) {
16006 nsv = sv_newmortal();
16007 SvSetSV_nosteal(nsv, sv);
16008 }
16009 save_re_context();
16010 PUSHMARK(sp);
16011 EXTEND(SP, 3);
16012 PUSHs(encoding);
16013 PUSHs(nsv);
16014 /*
16015 NI-S 2002/07/09
16016 Passing sv_yes is wrong - it needs to be or'ed set of constants
16017 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16018 remove converted chars from source.
16019
16020 Both will default the value - let them.
16021
16022 XPUSHs(&PL_sv_yes);
16023 */
16024 PUTBACK;
16025 call_method("decode", G_SCALAR);
16026 SPAGAIN;
16027 uni = POPs;
16028 PUTBACK;
16029 s = SvPV_const(uni, len);
16030 if (s != SvPVX_const(sv)) {
16031 SvGROW(sv, len + 1);
16032 Move(s, SvPVX(sv), len + 1, char);
16033 SvCUR_set(sv, len);
16034 }
16035 FREETMPS;
16036 POPSTACK;
16037 LEAVE;
16038 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16039 /* clear pos and any utf8 cache */
16040 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16041 if (mg)
16042 mg->mg_len = -1;
16043 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16044 magic_setutf8(sv,mg); /* clear UTF8 cache */
16045 }
16046 SvUTF8_on(sv);
16047 return SvPVX(sv);
16048 }
16049 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16050 }
16051
16052 /*
16053 =for apidoc sv_cat_decode
16054
16055 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16056 assumed to be octets in that encoding and decoding the input starts
16057 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16058 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16059 when the string C<tstr> appears in decoding output or the input ends on
16060 the PV of C<ssv>. The value which C<offset> points will be modified
16061 to the last input position on C<ssv>.
16062
16063 Returns TRUE if the terminator was found, else returns FALSE.
16064
16065 =cut */
16066
16067 bool
Perl_sv_cat_decode(pTHX_ SV * dsv,SV * encoding,SV * ssv,int * offset,char * tstr,int tlen)16068 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16069 SV *ssv, int *offset, char *tstr, int tlen)
16070 {
16071 bool ret = FALSE;
16072
16073 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16074
16075 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16076 SV *offsv;
16077 dSP;
16078 ENTER;
16079 SAVETMPS;
16080 save_re_context();
16081 PUSHMARK(sp);
16082 EXTEND(SP, 6);
16083 PUSHs(encoding);
16084 PUSHs(dsv);
16085 PUSHs(ssv);
16086 offsv = newSViv(*offset);
16087 mPUSHs(offsv);
16088 mPUSHp(tstr, tlen);
16089 PUTBACK;
16090 call_method("cat_decode", G_SCALAR);
16091 SPAGAIN;
16092 ret = SvTRUE(TOPs);
16093 *offset = SvIV(offsv);
16094 PUTBACK;
16095 FREETMPS;
16096 LEAVE;
16097 }
16098 else
16099 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16100 return ret;
16101
16102 }
16103
16104 /* ---------------------------------------------------------------------
16105 *
16106 * support functions for report_uninit()
16107 */
16108
16109 /* the maxiumum size of array or hash where we will scan looking
16110 * for the undefined element that triggered the warning */
16111
16112 #define FUV_MAX_SEARCH_SIZE 1000
16113
16114 /* Look for an entry in the hash whose value has the same SV as val;
16115 * If so, return a mortal copy of the key. */
16116
16117 STATIC SV*
S_find_hash_subscript(pTHX_ const HV * const hv,const SV * const val)16118 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16119 {
16120 dVAR;
16121 HE **array;
16122 I32 i;
16123
16124 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16125
16126 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16127 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16128 return NULL;
16129
16130 array = HvARRAY(hv);
16131
16132 for (i=HvMAX(hv); i>=0; i--) {
16133 HE *entry;
16134 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16135 if (HeVAL(entry) != val)
16136 continue;
16137 if ( HeVAL(entry) == &PL_sv_undef ||
16138 HeVAL(entry) == &PL_sv_placeholder)
16139 continue;
16140 if (!HeKEY(entry))
16141 return NULL;
16142 if (HeKLEN(entry) == HEf_SVKEY)
16143 return sv_mortalcopy(HeKEY_sv(entry));
16144 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16145 }
16146 }
16147 return NULL;
16148 }
16149
16150 /* Look for an entry in the array whose value has the same SV as val;
16151 * If so, return the index, otherwise return -1. */
16152
16153 STATIC SSize_t
S_find_array_subscript(pTHX_ const AV * const av,const SV * const val)16154 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16155 {
16156 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16157
16158 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16159 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16160 return -1;
16161
16162 if (val != &PL_sv_undef) {
16163 SV ** const svp = AvARRAY(av);
16164 SSize_t i;
16165
16166 for (i=AvFILLp(av); i>=0; i--)
16167 if (svp[i] == val)
16168 return i;
16169 }
16170 return -1;
16171 }
16172
16173 /* varname(): return the name of a variable, optionally with a subscript.
16174 * If gv is non-zero, use the name of that global, along with gvtype (one
16175 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16176 * targ. Depending on the value of the subscript_type flag, return:
16177 */
16178
16179 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16180 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16181 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16182 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16183
16184 SV*
Perl_varname(pTHX_ const GV * const gv,const char gvtype,PADOFFSET targ,const SV * const keyname,SSize_t aindex,int subscript_type)16185 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16186 const SV *const keyname, SSize_t aindex, int subscript_type)
16187 {
16188
16189 SV * const name = sv_newmortal();
16190 if (gv && isGV(gv)) {
16191 char buffer[2];
16192 buffer[0] = gvtype;
16193 buffer[1] = 0;
16194
16195 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16196
16197 gv_fullname4(name, gv, buffer, 0);
16198
16199 if ((unsigned int)SvPVX(name)[1] <= 26) {
16200 buffer[0] = '^';
16201 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16202
16203 /* Swap the 1 unprintable control character for the 2 byte pretty
16204 version - ie substr($name, 1, 1) = $buffer; */
16205 sv_insert(name, 1, 1, buffer, 2);
16206 }
16207 }
16208 else {
16209 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16210 PADNAME *sv;
16211
16212 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16213
16214 if (!cv || !CvPADLIST(cv))
16215 return NULL;
16216 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16217 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16218 SvUTF8_on(name);
16219 }
16220
16221 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16222 SV * const sv = newSV(0);
16223 STRLEN len;
16224 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16225
16226 *SvPVX(name) = '$';
16227 Perl_sv_catpvf(aTHX_ name, "{%s}",
16228 pv_pretty(sv, pv, len, 32, NULL, NULL,
16229 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16230 SvREFCNT_dec_NN(sv);
16231 }
16232 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16233 *SvPVX(name) = '$';
16234 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16235 }
16236 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16237 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16238 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16239 }
16240
16241 return name;
16242 }
16243
16244
16245 /*
16246 =for apidoc find_uninit_var
16247
16248 Find the name of the undefined variable (if any) that caused the operator
16249 to issue a "Use of uninitialized value" warning.
16250 If match is true, only return a name if its value matches C<uninit_sv>.
16251 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16252 warning, then following the direct child of the op may yield an
16253 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16254 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16255 the variable name if we get an exact match.
16256 C<desc_p> points to a string pointer holding the description of the op.
16257 This may be updated if needed.
16258
16259 The name is returned as a mortal SV.
16260
16261 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16262 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16263
16264 =cut
16265 */
16266
16267 STATIC SV *
S_find_uninit_var(pTHX_ const OP * const obase,const SV * const uninit_sv,bool match,const char ** desc_p)16268 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16269 bool match, const char **desc_p)
16270 {
16271 dVAR;
16272 SV *sv;
16273 const GV *gv;
16274 const OP *o, *o2, *kid;
16275
16276 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16277
16278 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16279 uninit_sv == &PL_sv_placeholder)))
16280 return NULL;
16281
16282 switch (obase->op_type) {
16283
16284 case OP_UNDEF:
16285 /* undef should care if its args are undef - any warnings
16286 * will be from tied/magic vars */
16287 break;
16288
16289 case OP_RV2AV:
16290 case OP_RV2HV:
16291 case OP_PADAV:
16292 case OP_PADHV:
16293 {
16294 const bool pad = ( obase->op_type == OP_PADAV
16295 || obase->op_type == OP_PADHV
16296 || obase->op_type == OP_PADRANGE
16297 );
16298
16299 const bool hash = ( obase->op_type == OP_PADHV
16300 || obase->op_type == OP_RV2HV
16301 || (obase->op_type == OP_PADRANGE
16302 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16303 );
16304 SSize_t index = 0;
16305 SV *keysv = NULL;
16306 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16307
16308 if (pad) { /* @lex, %lex */
16309 sv = PAD_SVl(obase->op_targ);
16310 gv = NULL;
16311 }
16312 else {
16313 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16314 /* @global, %global */
16315 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16316 if (!gv)
16317 break;
16318 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16319 }
16320 else if (obase == PL_op) /* @{expr}, %{expr} */
16321 return find_uninit_var(cUNOPx(obase)->op_first,
16322 uninit_sv, match, desc_p);
16323 else /* @{expr}, %{expr} as a sub-expression */
16324 return NULL;
16325 }
16326
16327 /* attempt to find a match within the aggregate */
16328 if (hash) {
16329 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16330 if (keysv)
16331 subscript_type = FUV_SUBSCRIPT_HASH;
16332 }
16333 else {
16334 index = find_array_subscript((const AV *)sv, uninit_sv);
16335 if (index >= 0)
16336 subscript_type = FUV_SUBSCRIPT_ARRAY;
16337 }
16338
16339 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16340 break;
16341
16342 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16343 keysv, index, subscript_type);
16344 }
16345
16346 case OP_RV2SV:
16347 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16348 /* $global */
16349 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16350 if (!gv || !GvSTASH(gv))
16351 break;
16352 if (match && (GvSV(gv) != uninit_sv))
16353 break;
16354 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16355 }
16356 /* ${expr} */
16357 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16358
16359 case OP_PADSV:
16360 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16361 break;
16362 return varname(NULL, '$', obase->op_targ,
16363 NULL, 0, FUV_SUBSCRIPT_NONE);
16364
16365 case OP_GVSV:
16366 gv = cGVOPx_gv(obase);
16367 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16368 break;
16369 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16370
16371 case OP_AELEMFAST_LEX:
16372 if (match) {
16373 SV **svp;
16374 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16375 if (!av || SvRMAGICAL(av))
16376 break;
16377 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16378 if (!svp || *svp != uninit_sv)
16379 break;
16380 }
16381 return varname(NULL, '$', obase->op_targ,
16382 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16383 case OP_AELEMFAST:
16384 {
16385 gv = cGVOPx_gv(obase);
16386 if (!gv)
16387 break;
16388 if (match) {
16389 SV **svp;
16390 AV *const av = GvAV(gv);
16391 if (!av || SvRMAGICAL(av))
16392 break;
16393 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16394 if (!svp || *svp != uninit_sv)
16395 break;
16396 }
16397 return varname(gv, '$', 0,
16398 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16399 }
16400 NOT_REACHED; /* NOTREACHED */
16401
16402 case OP_EXISTS:
16403 o = cUNOPx(obase)->op_first;
16404 if (!o || o->op_type != OP_NULL ||
16405 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16406 break;
16407 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16408
16409 case OP_AELEM:
16410 case OP_HELEM:
16411 {
16412 bool negate = FALSE;
16413
16414 if (PL_op == obase)
16415 /* $a[uninit_expr] or $h{uninit_expr} */
16416 return find_uninit_var(cBINOPx(obase)->op_last,
16417 uninit_sv, match, desc_p);
16418
16419 gv = NULL;
16420 o = cBINOPx(obase)->op_first;
16421 kid = cBINOPx(obase)->op_last;
16422
16423 /* get the av or hv, and optionally the gv */
16424 sv = NULL;
16425 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16426 sv = PAD_SV(o->op_targ);
16427 }
16428 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16429 && cUNOPo->op_first->op_type == OP_GV)
16430 {
16431 gv = cGVOPx_gv(cUNOPo->op_first);
16432 if (!gv)
16433 break;
16434 sv = o->op_type
16435 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16436 }
16437 if (!sv)
16438 break;
16439
16440 if (kid && kid->op_type == OP_NEGATE) {
16441 negate = TRUE;
16442 kid = cUNOPx(kid)->op_first;
16443 }
16444
16445 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16446 /* index is constant */
16447 SV* kidsv;
16448 if (negate) {
16449 kidsv = newSVpvs_flags("-", SVs_TEMP);
16450 sv_catsv(kidsv, cSVOPx_sv(kid));
16451 }
16452 else
16453 kidsv = cSVOPx_sv(kid);
16454 if (match) {
16455 if (SvMAGICAL(sv))
16456 break;
16457 if (obase->op_type == OP_HELEM) {
16458 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16459 if (!he || HeVAL(he) != uninit_sv)
16460 break;
16461 }
16462 else {
16463 SV * const opsv = cSVOPx_sv(kid);
16464 const IV opsviv = SvIV(opsv);
16465 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16466 negate ? - opsviv : opsviv,
16467 FALSE);
16468 if (!svp || *svp != uninit_sv)
16469 break;
16470 }
16471 }
16472 if (obase->op_type == OP_HELEM)
16473 return varname(gv, '%', o->op_targ,
16474 kidsv, 0, FUV_SUBSCRIPT_HASH);
16475 else
16476 return varname(gv, '@', o->op_targ, NULL,
16477 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16478 FUV_SUBSCRIPT_ARRAY);
16479 }
16480 else {
16481 /* index is an expression;
16482 * attempt to find a match within the aggregate */
16483 if (obase->op_type == OP_HELEM) {
16484 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16485 if (keysv)
16486 return varname(gv, '%', o->op_targ,
16487 keysv, 0, FUV_SUBSCRIPT_HASH);
16488 }
16489 else {
16490 const SSize_t index
16491 = find_array_subscript((const AV *)sv, uninit_sv);
16492 if (index >= 0)
16493 return varname(gv, '@', o->op_targ,
16494 NULL, index, FUV_SUBSCRIPT_ARRAY);
16495 }
16496 if (match)
16497 break;
16498 return varname(gv,
16499 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16500 ? '@' : '%'),
16501 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16502 }
16503 NOT_REACHED; /* NOTREACHED */
16504 }
16505
16506 case OP_MULTIDEREF: {
16507 /* If we were executing OP_MULTIDEREF when the undef warning
16508 * triggered, then it must be one of the index values within
16509 * that triggered it. If not, then the only possibility is that
16510 * the value retrieved by the last aggregate index might be the
16511 * culprit. For the former, we set PL_multideref_pc each time before
16512 * using an index, so work though the item list until we reach
16513 * that point. For the latter, just work through the entire item
16514 * list; the last aggregate retrieved will be the candidate.
16515 * There is a third rare possibility: something triggered
16516 * magic while fetching an array/hash element. Just display
16517 * nothing in this case.
16518 */
16519
16520 /* the named aggregate, if any */
16521 PADOFFSET agg_targ = 0;
16522 GV *agg_gv = NULL;
16523 /* the last-seen index */
16524 UV index_type;
16525 PADOFFSET index_targ;
16526 GV *index_gv;
16527 IV index_const_iv = 0; /* init for spurious compiler warn */
16528 SV *index_const_sv;
16529 int depth = 0; /* how many array/hash lookups we've done */
16530
16531 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16532 UNOP_AUX_item *last = NULL;
16533 UV actions = items->uv;
16534 bool is_hv;
16535
16536 if (PL_op == obase) {
16537 last = PL_multideref_pc;
16538 assert(last >= items && last <= items + items[-1].uv);
16539 }
16540
16541 assert(actions);
16542
16543 while (1) {
16544 is_hv = FALSE;
16545 switch (actions & MDEREF_ACTION_MASK) {
16546
16547 case MDEREF_reload:
16548 actions = (++items)->uv;
16549 continue;
16550
16551 case MDEREF_HV_padhv_helem: /* $lex{...} */
16552 is_hv = TRUE;
16553 /* FALLTHROUGH */
16554 case MDEREF_AV_padav_aelem: /* $lex[...] */
16555 agg_targ = (++items)->pad_offset;
16556 agg_gv = NULL;
16557 break;
16558
16559 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16560 is_hv = TRUE;
16561 /* FALLTHROUGH */
16562 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16563 agg_targ = 0;
16564 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16565 assert(isGV_with_GP(agg_gv));
16566 break;
16567
16568 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16569 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16570 ++items;
16571 /* FALLTHROUGH */
16572 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16573 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16574 agg_targ = 0;
16575 agg_gv = NULL;
16576 is_hv = TRUE;
16577 break;
16578
16579 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16580 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16581 ++items;
16582 /* FALLTHROUGH */
16583 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16584 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16585 agg_targ = 0;
16586 agg_gv = NULL;
16587 } /* switch */
16588
16589 index_targ = 0;
16590 index_gv = NULL;
16591 index_const_sv = NULL;
16592
16593 index_type = (actions & MDEREF_INDEX_MASK);
16594 switch (index_type) {
16595 case MDEREF_INDEX_none:
16596 break;
16597 case MDEREF_INDEX_const:
16598 if (is_hv)
16599 index_const_sv = UNOP_AUX_item_sv(++items)
16600 else
16601 index_const_iv = (++items)->iv;
16602 break;
16603 case MDEREF_INDEX_padsv:
16604 index_targ = (++items)->pad_offset;
16605 break;
16606 case MDEREF_INDEX_gvsv:
16607 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16608 assert(isGV_with_GP(index_gv));
16609 break;
16610 }
16611
16612 if (index_type != MDEREF_INDEX_none)
16613 depth++;
16614
16615 if ( index_type == MDEREF_INDEX_none
16616 || (actions & MDEREF_FLAG_last)
16617 || (last && items >= last)
16618 )
16619 break;
16620
16621 actions >>= MDEREF_SHIFT;
16622 } /* while */
16623
16624 if (PL_op == obase) {
16625 /* most likely index was undef */
16626
16627 *desc_p = ( (actions & MDEREF_FLAG_last)
16628 && (obase->op_private
16629 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16630 ?
16631 (obase->op_private & OPpMULTIDEREF_EXISTS)
16632 ? "exists"
16633 : "delete"
16634 : is_hv ? "hash element" : "array element";
16635 assert(index_type != MDEREF_INDEX_none);
16636 if (index_gv) {
16637 if (GvSV(index_gv) == uninit_sv)
16638 return varname(index_gv, '$', 0, NULL, 0,
16639 FUV_SUBSCRIPT_NONE);
16640 else
16641 return NULL;
16642 }
16643 if (index_targ) {
16644 if (PL_curpad[index_targ] == uninit_sv)
16645 return varname(NULL, '$', index_targ,
16646 NULL, 0, FUV_SUBSCRIPT_NONE);
16647 else
16648 return NULL;
16649 }
16650 /* If we got to this point it was undef on a const subscript,
16651 * so magic probably involved, e.g. $ISA[0]. Give up. */
16652 return NULL;
16653 }
16654
16655 /* the SV returned by pp_multideref() was undef, if anything was */
16656
16657 if (depth != 1)
16658 break;
16659
16660 if (agg_targ)
16661 sv = PAD_SV(agg_targ);
16662 else if (agg_gv)
16663 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16664 else
16665 break;
16666
16667 if (index_type == MDEREF_INDEX_const) {
16668 if (match) {
16669 if (SvMAGICAL(sv))
16670 break;
16671 if (is_hv) {
16672 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16673 if (!he || HeVAL(he) != uninit_sv)
16674 break;
16675 }
16676 else {
16677 SV * const * const svp =
16678 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16679 if (!svp || *svp != uninit_sv)
16680 break;
16681 }
16682 }
16683 return is_hv
16684 ? varname(agg_gv, '%', agg_targ,
16685 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16686 : varname(agg_gv, '@', agg_targ,
16687 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16688 }
16689 else {
16690 /* index is an var */
16691 if (is_hv) {
16692 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16693 if (keysv)
16694 return varname(agg_gv, '%', agg_targ,
16695 keysv, 0, FUV_SUBSCRIPT_HASH);
16696 }
16697 else {
16698 const SSize_t index
16699 = find_array_subscript((const AV *)sv, uninit_sv);
16700 if (index >= 0)
16701 return varname(agg_gv, '@', agg_targ,
16702 NULL, index, FUV_SUBSCRIPT_ARRAY);
16703 }
16704 if (match)
16705 break;
16706 return varname(agg_gv,
16707 is_hv ? '%' : '@',
16708 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16709 }
16710 NOT_REACHED; /* NOTREACHED */
16711 }
16712
16713 case OP_AASSIGN:
16714 /* only examine RHS */
16715 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16716 match, desc_p);
16717
16718 case OP_OPEN:
16719 o = cUNOPx(obase)->op_first;
16720 if ( o->op_type == OP_PUSHMARK
16721 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16722 )
16723 o = OpSIBLING(o);
16724
16725 if (!OpHAS_SIBLING(o)) {
16726 /* one-arg version of open is highly magical */
16727
16728 if (o->op_type == OP_GV) { /* open FOO; */
16729 gv = cGVOPx_gv(o);
16730 if (match && GvSV(gv) != uninit_sv)
16731 break;
16732 return varname(gv, '$', 0,
16733 NULL, 0, FUV_SUBSCRIPT_NONE);
16734 }
16735 /* other possibilities not handled are:
16736 * open $x; or open my $x; should return '${*$x}'
16737 * open expr; should return '$'.expr ideally
16738 */
16739 break;
16740 }
16741 match = 1;
16742 goto do_op;
16743
16744 /* ops where $_ may be an implicit arg */
16745 case OP_TRANS:
16746 case OP_TRANSR:
16747 case OP_SUBST:
16748 case OP_MATCH:
16749 if ( !(obase->op_flags & OPf_STACKED)) {
16750 if (uninit_sv == DEFSV)
16751 return newSVpvs_flags("$_", SVs_TEMP);
16752 else if (obase->op_targ
16753 && uninit_sv == PAD_SVl(obase->op_targ))
16754 return varname(NULL, '$', obase->op_targ, NULL, 0,
16755 FUV_SUBSCRIPT_NONE);
16756 }
16757 goto do_op;
16758
16759 case OP_PRTF:
16760 case OP_PRINT:
16761 case OP_SAY:
16762 match = 1; /* print etc can return undef on defined args */
16763 /* skip filehandle as it can't produce 'undef' warning */
16764 o = cUNOPx(obase)->op_first;
16765 if ((obase->op_flags & OPf_STACKED)
16766 &&
16767 ( o->op_type == OP_PUSHMARK
16768 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16769 o = OpSIBLING(OpSIBLING(o));
16770 goto do_op2;
16771
16772
16773 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16774 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16775
16776 /* the following ops are capable of returning PL_sv_undef even for
16777 * defined arg(s) */
16778
16779 case OP_BACKTICK:
16780 case OP_PIPE_OP:
16781 case OP_FILENO:
16782 case OP_BINMODE:
16783 case OP_TIED:
16784 case OP_GETC:
16785 case OP_SYSREAD:
16786 case OP_SEND:
16787 case OP_IOCTL:
16788 case OP_SOCKET:
16789 case OP_SOCKPAIR:
16790 case OP_BIND:
16791 case OP_CONNECT:
16792 case OP_LISTEN:
16793 case OP_ACCEPT:
16794 case OP_SHUTDOWN:
16795 case OP_SSOCKOPT:
16796 case OP_GETPEERNAME:
16797 case OP_FTRREAD:
16798 case OP_FTRWRITE:
16799 case OP_FTREXEC:
16800 case OP_FTROWNED:
16801 case OP_FTEREAD:
16802 case OP_FTEWRITE:
16803 case OP_FTEEXEC:
16804 case OP_FTEOWNED:
16805 case OP_FTIS:
16806 case OP_FTZERO:
16807 case OP_FTSIZE:
16808 case OP_FTFILE:
16809 case OP_FTDIR:
16810 case OP_FTLINK:
16811 case OP_FTPIPE:
16812 case OP_FTSOCK:
16813 case OP_FTBLK:
16814 case OP_FTCHR:
16815 case OP_FTTTY:
16816 case OP_FTSUID:
16817 case OP_FTSGID:
16818 case OP_FTSVTX:
16819 case OP_FTTEXT:
16820 case OP_FTBINARY:
16821 case OP_FTMTIME:
16822 case OP_FTATIME:
16823 case OP_FTCTIME:
16824 case OP_READLINK:
16825 case OP_OPEN_DIR:
16826 case OP_READDIR:
16827 case OP_TELLDIR:
16828 case OP_SEEKDIR:
16829 case OP_REWINDDIR:
16830 case OP_CLOSEDIR:
16831 case OP_GMTIME:
16832 case OP_ALARM:
16833 case OP_SEMGET:
16834 case OP_GETLOGIN:
16835 case OP_SUBSTR:
16836 case OP_AEACH:
16837 case OP_EACH:
16838 case OP_SORT:
16839 case OP_CALLER:
16840 case OP_DOFILE:
16841 case OP_PROTOTYPE:
16842 case OP_NCMP:
16843 case OP_SMARTMATCH:
16844 case OP_UNPACK:
16845 case OP_SYSOPEN:
16846 case OP_SYSSEEK:
16847 match = 1;
16848 goto do_op;
16849
16850 case OP_ENTERSUB:
16851 case OP_GOTO:
16852 /* XXX tmp hack: these two may call an XS sub, and currently
16853 XS subs don't have a SUB entry on the context stack, so CV and
16854 pad determination goes wrong, and BAD things happen. So, just
16855 don't try to determine the value under those circumstances.
16856 Need a better fix at dome point. DAPM 11/2007 */
16857 break;
16858
16859 case OP_FLIP:
16860 case OP_FLOP:
16861 {
16862 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16863 if (gv && GvSV(gv) == uninit_sv)
16864 return newSVpvs_flags("$.", SVs_TEMP);
16865 goto do_op;
16866 }
16867
16868 case OP_POS:
16869 /* def-ness of rval pos() is independent of the def-ness of its arg */
16870 if ( !(obase->op_flags & OPf_MOD))
16871 break;
16872 /* FALLTHROUGH */
16873
16874 case OP_SCHOMP:
16875 case OP_CHOMP:
16876 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16877 return newSVpvs_flags("${$/}", SVs_TEMP);
16878 /* FALLTHROUGH */
16879
16880 default:
16881 do_op:
16882 if (!(obase->op_flags & OPf_KIDS))
16883 break;
16884 o = cUNOPx(obase)->op_first;
16885
16886 do_op2:
16887 if (!o)
16888 break;
16889
16890 /* This loop checks all the kid ops, skipping any that cannot pos-
16891 * sibly be responsible for the uninitialized value; i.e., defined
16892 * constants and ops that return nothing. If there is only one op
16893 * left that is not skipped, then we *know* it is responsible for
16894 * the uninitialized value. If there is more than one op left, we
16895 * have to look for an exact match in the while() loop below.
16896 * Note that we skip padrange, because the individual pad ops that
16897 * it replaced are still in the tree, so we work on them instead.
16898 */
16899 o2 = NULL;
16900 for (kid=o; kid; kid = OpSIBLING(kid)) {
16901 const OPCODE type = kid->op_type;
16902 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16903 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16904 || (type == OP_PUSHMARK)
16905 || (type == OP_PADRANGE)
16906 )
16907 continue;
16908
16909 if (o2) { /* more than one found */
16910 o2 = NULL;
16911 break;
16912 }
16913 o2 = kid;
16914 }
16915 if (o2)
16916 return find_uninit_var(o2, uninit_sv, match, desc_p);
16917
16918 /* scan all args */
16919 while (o) {
16920 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16921 if (sv)
16922 return sv;
16923 o = OpSIBLING(o);
16924 }
16925 break;
16926 }
16927 return NULL;
16928 }
16929
16930
16931 /*
16932 =for apidoc report_uninit
16933
16934 Print appropriate "Use of uninitialized variable" warning.
16935
16936 =cut
16937 */
16938
16939 void
Perl_report_uninit(pTHX_ const SV * uninit_sv)16940 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16941 {
16942 const char *desc = NULL;
16943 SV* varname = NULL;
16944
16945 if (PL_op) {
16946 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16947 ? "join or string"
16948 : PL_op->op_type == OP_MULTICONCAT
16949 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16950 ? "sprintf"
16951 : OP_DESC(PL_op);
16952 if (uninit_sv && PL_curpad) {
16953 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16954 if (varname)
16955 sv_insert(varname, 0, 0, " ", 1);
16956 }
16957 }
16958 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16959 /* we've reached the end of a sort block or sub,
16960 * and the uninit value is probably what that code returned */
16961 desc = "sort";
16962
16963 /* PL_warn_uninit_sv is constant */
16964 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
16965 if (desc)
16966 /* diag_listed_as: Use of uninitialized value%s */
16967 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16968 SVfARG(varname ? varname : &PL_sv_no),
16969 " in ", desc);
16970 else
16971 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16972 "", "", "");
16973 GCC_DIAG_RESTORE_STMT;
16974 }
16975
16976 /*
16977 * ex: set ts=8 sts=4 sw=4 et:
16978 */
16979