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
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.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
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
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
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))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
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)) \
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)) \
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
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]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 #ifdef PERL_OLD_COPY_ON_WRITE
129 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
130 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
133 /* ============================================================================
135 =head1 Allocation and deallocation of SVs.
136 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
137 sv, av, hv...) contains type and reference count information, and for
138 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
139 contains fields specific to each type. Some types store all they need
140 in the head, so don't have a body.
142 In all but the most memory-paranoid configurations (ex: PURIFY), heads
143 and bodies are allocated out of arenas, which by default are
144 approximately 4K chunks of memory parcelled up into N heads or bodies.
145 Sv-bodies are allocated by their sv-type, guaranteeing size
146 consistency needed to allocate safely from arrays.
148 For SV-heads, the first slot in each arena is reserved, and holds a
149 link to the next arena, some flags, and a note of the number of slots.
150 Snaked through each arena chain is a linked list of free items; when
151 this becomes empty, an extra arena is allocated and divided up into N
152 items which are threaded into the free list.
154 SV-bodies are similar, but they use arena-sets by default, which
155 separate the link and info from the arena itself, and reclaim the 1st
156 slot in the arena. SV-bodies are further described later.
158 The following global variables are associated with arenas:
160 PL_sv_arenaroot pointer to list of SV arenas
161 PL_sv_root pointer to list of free SV structures
163 PL_body_arenas head of linked-list of body arenas
164 PL_body_roots[] array of pointers to list of free bodies of svtype
165 arrays are indexed by the svtype needed
167 A few special SV heads are not allocated from an arena, but are
168 instead directly created in the interpreter structure, eg PL_sv_undef.
169 The size of arenas can be changed from the default by setting
170 PERL_ARENA_SIZE appropriately at compile time.
172 The SV arena serves the secondary purpose of allowing still-live SVs
173 to be located and destroyed during final cleanup.
175 At the lowest level, the macros new_SV() and del_SV() grab and free
176 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
177 to return the SV to the free list with error checking.) new_SV() calls
178 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
179 SVs in the free list have their SvTYPE field set to all ones.
181 At the time of very final cleanup, sv_free_arenas() is called from
182 perl_destruct() to physically free all the arenas allocated since the
183 start of the interpreter.
185 The function visit() scans the SV arenas list, and calls a specified
186 function for each SV it finds which is still live - ie which has an SvTYPE
187 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
188 following functions (specified as [function that calls visit()] / [function
189 called by visit() for each SV]):
191 sv_report_used() / do_report_used()
192 dump all remaining SVs (debugging aid)
194 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
195 do_clean_named_io_objs(),do_curse()
196 Attempt to free all objects pointed to by RVs,
197 try to do the same for all objects indir-
198 ectly referenced by typeglobs too, and
199 then do a final sweep, cursing any
200 objects that remain. Called once from
201 perl_destruct(), prior to calling sv_clean_all()
204 sv_clean_all() / do_clean_all()
205 SvREFCNT_dec(sv) each remaining SV, possibly
206 triggering an sv_free(). It also sets the
207 SVf_BREAK flag on the SV to indicate that the
208 refcnt has been artificially lowered, and thus
209 stopping sv_free() from giving spurious warnings
210 about SVs which unexpectedly have a refcnt
211 of zero. called repeatedly from perl_destruct()
212 until there are no SVs left.
214 =head2 Arena allocator API Summary
216 Private API to rest of sv.c
220 new_XPVNV(), del_XPVGV(),
225 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
229 * ========================================================================= */
232 * "A time to plant, and a time to uproot what was planted..."
236 # define MEM_LOG_NEW_SV(sv, file, line, func) \
237 Perl_mem_log_new_sv(sv, file, line, func)
238 # define MEM_LOG_DEL_SV(sv, file, line, func) \
239 Perl_mem_log_del_sv(sv, file, line, func)
241 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
242 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
245 #ifdef DEBUG_LEAKING_SCALARS
246 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
247 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
249 # define DEBUG_SV_SERIAL(sv) \
250 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
251 PTR2UV(sv), (long)(sv)->sv_debug_serial))
253 # define FREE_SV_DEBUG_FILE(sv)
254 # define DEBUG_SV_SERIAL(sv) NOOP
258 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
259 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
260 /* Whilst I'd love to do this, it seems that things like to check on
262 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
264 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
265 PoisonNew(&SvREFCNT(sv), 1, U32)
267 # define SvARENA_CHAIN(sv) SvANY(sv)
268 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
269 # define POISON_SV_HEAD(sv)
272 /* Mark an SV head as unused, and add to free list.
274 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
275 * its refcount artificially decremented during global destruction, so
276 * there may be dangling pointers to it. The last thing we want in that
277 * case is for it to be reused. */
279 #define plant_SV(p) \
281 const U32 old_flags = SvFLAGS(p); \
282 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
283 DEBUG_SV_SERIAL(p); \
284 FREE_SV_DEBUG_FILE(p); \
286 SvFLAGS(p) = SVTYPEMASK; \
287 if (!(old_flags & SVf_BREAK)) { \
288 SvARENA_CHAIN_SET(p, PL_sv_root); \
294 #define uproot_SV(p) \
297 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
302 /* make some more SVs by adding another arena */
308 char *chunk; /* must use New here to match call to */
309 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
310 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
315 /* new_SV(): return a new, empty SV head */
317 #ifdef DEBUG_LEAKING_SCALARS
318 /* provide a real function for a debugger to play with */
320 S_new_SV(pTHX_ const char *file, int line, const char *func)
327 sv = S_more_sv(aTHX);
331 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
332 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
338 sv->sv_debug_inpad = 0;
339 sv->sv_debug_parent = NULL;
340 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
342 sv->sv_debug_serial = PL_sv_serial++;
344 MEM_LOG_NEW_SV(sv, file, line, func);
345 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
346 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
350 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
358 (p) = S_more_sv(aTHX); \
362 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
367 /* del_SV(): return an empty SV head to the free list */
380 S_del_sv(pTHX_ SV *p)
382 PERL_ARGS_ASSERT_DEL_SV;
387 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
388 const SV * const sv = sva + 1;
389 const SV * const svend = &sva[SvREFCNT(sva)];
390 if (p >= sv && p < svend) {
396 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
397 "Attempt to free non-arena SV: 0x%"UVxf
398 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
405 #else /* ! DEBUGGING */
407 #define del_SV(p) plant_SV(p)
409 #endif /* DEBUGGING */
412 * Bodyless IVs and NVs!
414 * Since 5.9.2, we can avoid allocating a body for SVt_IV-type SVs.
415 * Since the larger IV-holding variants of SVs store their integer
416 * values in their respective bodies, the family of SvIV() accessor
417 * macros would naively have to branch on the SV type to find the
418 * integer value either in the HEAD or BODY. In order to avoid this
419 * expensive branch, a clever soul has deployed a great hack:
420 * We set up the SvANY pointer such that instead of pointing to a
421 * real body, it points into the memory before the location of the
422 * head. We compute this pointer such that the location of
423 * the integer member of the hypothetical body struct happens to
424 * be the same as the location of the integer member of the bodyless
425 * SV head. This now means that the SvIV() family of accessors can
426 * always read from the (hypothetical or real) body via SvANY.
428 * Since the 5.21 dev series, we employ the same trick for NVs
429 * if the architecture can support it (NVSIZE <= IVSIZE).
432 /* The following two macros compute the necessary offsets for the above
433 * trick and store them in SvANY for SvIV() (and friends) to use. */
434 #define SET_SVANY_FOR_BODYLESS_IV(sv) \
435 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv))
437 #define SET_SVANY_FOR_BODYLESS_NV(sv) \
438 SvANY(sv) = (XPVNV*)((char*)&(sv->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv))
441 =head1 SV Manipulation Functions
443 =for apidoc sv_add_arena
445 Given a chunk of memory, link it to the head of the list of arenas,
446 and split it into a list of free SVs.
452 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
454 SV *const sva = MUTABLE_SV(ptr);
458 PERL_ARGS_ASSERT_SV_ADD_ARENA;
460 /* The first SV in an arena isn't an SV. */
461 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
462 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
463 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
465 PL_sv_arenaroot = sva;
466 PL_sv_root = sva + 1;
468 svend = &sva[SvREFCNT(sva) - 1];
471 SvARENA_CHAIN_SET(sv, (sv + 1));
475 /* Must always set typemask because it's always checked in on cleanup
476 when the arenas are walked looking for objects. */
477 SvFLAGS(sv) = SVTYPEMASK;
480 SvARENA_CHAIN_SET(sv, 0);
484 SvFLAGS(sv) = SVTYPEMASK;
487 /* visit(): call the named function for each non-free SV in the arenas
488 * whose flags field matches the flags/mask args. */
491 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
496 PERL_ARGS_ASSERT_VISIT;
498 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
499 const SV * const svend = &sva[SvREFCNT(sva)];
501 for (sv = sva + 1; sv < svend; ++sv) {
502 if (SvTYPE(sv) != (svtype)SVTYPEMASK
503 && (sv->sv_flags & mask) == flags
516 /* called by sv_report_used() for each live SV */
519 do_report_used(pTHX_ SV *const sv)
521 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
522 PerlIO_printf(Perl_debug_log, "****\n");
529 =for apidoc sv_report_used
531 Dump the contents of all SVs not yet freed (debugging aid).
537 Perl_sv_report_used(pTHX)
540 visit(do_report_used, 0, 0);
546 /* called by sv_clean_objs() for each live SV */
549 do_clean_objs(pTHX_ SV *const ref)
553 SV * const target = SvRV(ref);
554 if (SvOBJECT(target)) {
555 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
556 if (SvWEAKREF(ref)) {
557 sv_del_backref(target, ref);
563 SvREFCNT_dec_NN(target);
570 /* clear any slots in a GV which hold objects - except IO;
571 * called by sv_clean_objs() for each live GV */
574 do_clean_named_objs(pTHX_ SV *const sv)
577 assert(SvTYPE(sv) == SVt_PVGV);
578 assert(isGV_with_GP(sv));
582 /* freeing GP entries may indirectly free the current GV;
583 * hold onto it while we mess with the GP slots */
586 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
587 DEBUG_D((PerlIO_printf(Perl_debug_log,
588 "Cleaning named glob SV object:\n "), sv_dump(obj)));
590 SvREFCNT_dec_NN(obj);
592 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
593 DEBUG_D((PerlIO_printf(Perl_debug_log,
594 "Cleaning named glob AV object:\n "), sv_dump(obj)));
596 SvREFCNT_dec_NN(obj);
598 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
599 DEBUG_D((PerlIO_printf(Perl_debug_log,
600 "Cleaning named glob HV object:\n "), sv_dump(obj)));
602 SvREFCNT_dec_NN(obj);
604 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
605 DEBUG_D((PerlIO_printf(Perl_debug_log,
606 "Cleaning named glob CV object:\n "), sv_dump(obj)));
608 SvREFCNT_dec_NN(obj);
610 SvREFCNT_dec_NN(sv); /* undo the inc above */
613 /* clear any IO slots in a GV which hold objects (except stderr, defout);
614 * called by sv_clean_objs() for each live GV */
617 do_clean_named_io_objs(pTHX_ SV *const sv)
620 assert(SvTYPE(sv) == SVt_PVGV);
621 assert(isGV_with_GP(sv));
622 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
626 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
627 DEBUG_D((PerlIO_printf(Perl_debug_log,
628 "Cleaning named glob IO object:\n "), sv_dump(obj)));
630 SvREFCNT_dec_NN(obj);
632 SvREFCNT_dec_NN(sv); /* undo the inc above */
635 /* Void wrapper to pass to visit() */
637 do_curse(pTHX_ SV * const sv) {
638 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
639 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
645 =for apidoc sv_clean_objs
647 Attempt to destroy all objects not yet freed.
653 Perl_sv_clean_objs(pTHX)
656 PL_in_clean_objs = TRUE;
657 visit(do_clean_objs, SVf_ROK, SVf_ROK);
658 /* Some barnacles may yet remain, clinging to typeglobs.
659 * Run the non-IO destructors first: they may want to output
660 * error messages, close files etc */
661 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
662 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
663 /* And if there are some very tenacious barnacles clinging to arrays,
664 closures, or what have you.... */
665 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
666 olddef = PL_defoutgv;
667 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
668 if (olddef && isGV_with_GP(olddef))
669 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
670 olderr = PL_stderrgv;
671 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
672 if (olderr && isGV_with_GP(olderr))
673 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
674 SvREFCNT_dec(olddef);
675 PL_in_clean_objs = FALSE;
678 /* called by sv_clean_all() for each live SV */
681 do_clean_all(pTHX_ SV *const sv)
683 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
684 /* don't clean pid table and strtab */
687 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
688 SvFLAGS(sv) |= SVf_BREAK;
693 =for apidoc sv_clean_all
695 Decrement the refcnt of each remaining SV, possibly triggering a
696 cleanup. This function may have to be called multiple times to free
697 SVs which are in complex self-referential hierarchies.
703 Perl_sv_clean_all(pTHX)
706 PL_in_clean_all = TRUE;
707 cleaned = visit(do_clean_all, 0,0);
712 ARENASETS: a meta-arena implementation which separates arena-info
713 into struct arena_set, which contains an array of struct
714 arena_descs, each holding info for a single arena. By separating
715 the meta-info from the arena, we recover the 1st slot, formerly
716 borrowed for list management. The arena_set is about the size of an
717 arena, avoiding the needless malloc overhead of a naive linked-list.
719 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
720 memory in the last arena-set (1/2 on average). In trade, we get
721 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
722 smaller types). The recovery of the wasted space allows use of
723 small arenas for large, rare body types, by changing array* fields
724 in body_details_by_type[] below.
727 char *arena; /* the raw storage, allocated aligned */
728 size_t size; /* its size ~4k typ */
729 svtype utype; /* bodytype stored in arena */
734 /* Get the maximum number of elements in set[] such that struct arena_set
735 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
736 therefore likely to be 1 aligned memory page. */
738 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
739 - 2 * sizeof(int)) / sizeof (struct arena_desc))
742 struct arena_set* next;
743 unsigned int set_size; /* ie ARENAS_PER_SET */
744 unsigned int curr; /* index of next available arena-desc */
745 struct arena_desc set[ARENAS_PER_SET];
749 =for apidoc sv_free_arenas
751 Deallocate the memory used by all arenas. Note that all the individual SV
752 heads and bodies within the arenas must already have been freed.
758 Perl_sv_free_arenas(pTHX)
764 /* Free arenas here, but be careful about fake ones. (We assume
765 contiguity of the fake ones with the corresponding real ones.) */
767 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
768 svanext = MUTABLE_SV(SvANY(sva));
769 while (svanext && SvFAKE(svanext))
770 svanext = MUTABLE_SV(SvANY(svanext));
777 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
780 struct arena_set *current = aroot;
783 assert(aroot->set[i].arena);
784 Safefree(aroot->set[i].arena);
792 i = PERL_ARENA_ROOTS_SIZE;
794 PL_body_roots[i] = 0;
801 Here are mid-level routines that manage the allocation of bodies out
802 of the various arenas. There are 5 kinds of arenas:
804 1. SV-head arenas, which are discussed and handled above
805 2. regular body arenas
806 3. arenas for reduced-size bodies
809 Arena types 2 & 3 are chained by body-type off an array of
810 arena-root pointers, which is indexed by svtype. Some of the
811 larger/less used body types are malloced singly, since a large
812 unused block of them is wasteful. Also, several svtypes dont have
813 bodies; the data fits into the sv-head itself. The arena-root
814 pointer thus has a few unused root-pointers (which may be hijacked
815 later for arena types 4,5)
817 3 differs from 2 as an optimization; some body types have several
818 unused fields in the front of the structure (which are kept in-place
819 for consistency). These bodies can be allocated in smaller chunks,
820 because the leading fields arent accessed. Pointers to such bodies
821 are decremented to point at the unused 'ghost' memory, knowing that
822 the pointers are used with offsets to the real memory.
825 =head1 SV-Body Allocation
829 Allocation of SV-bodies is similar to SV-heads, differing as follows;
830 the allocation mechanism is used for many body types, so is somewhat
831 more complicated, it uses arena-sets, and has no need for still-live
834 At the outermost level, (new|del)_X*V macros return bodies of the
835 appropriate type. These macros call either (new|del)_body_type or
836 (new|del)_body_allocated macro pairs, depending on specifics of the
837 type. Most body types use the former pair, the latter pair is used to
838 allocate body types with "ghost fields".
840 "ghost fields" are fields that are unused in certain types, and
841 consequently don't need to actually exist. They are declared because
842 they're part of a "base type", which allows use of functions as
843 methods. The simplest examples are AVs and HVs, 2 aggregate types
844 which don't use the fields which support SCALAR semantics.
846 For these types, the arenas are carved up into appropriately sized
847 chunks, we thus avoid wasted memory for those unaccessed members.
848 When bodies are allocated, we adjust the pointer back in memory by the
849 size of the part not allocated, so it's as if we allocated the full
850 structure. (But things will all go boom if you write to the part that
851 is "not there", because you'll be overwriting the last members of the
852 preceding structure in memory.)
854 We calculate the correction using the STRUCT_OFFSET macro on the first
855 member present. If the allocated structure is smaller (no initial NV
856 actually allocated) then the net effect is to subtract the size of the NV
857 from the pointer, to return a new pointer as if an initial NV were actually
858 allocated. (We were using structures named *_allocated for this, but
859 this turned out to be a subtle bug, because a structure without an NV
860 could have a lower alignment constraint, but the compiler is allowed to
861 optimised accesses based on the alignment constraint of the actual pointer
862 to the full structure, for example, using a single 64 bit load instruction
863 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
865 This is the same trick as was used for NV and IV bodies. Ironically it
866 doesn't need to be used for NV bodies any more, because NV is now at
867 the start of the structure. IV bodies, and also in some builds NV bodies,
868 don't need it either, because they are no longer allocated.
870 In turn, the new_body_* allocators call S_new_body(), which invokes
871 new_body_inline macro, which takes a lock, and takes a body off the
872 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
873 necessary to refresh an empty list. Then the lock is released, and
874 the body is returned.
876 Perl_more_bodies allocates a new arena, and carves it up into an array of N
877 bodies, which it strings into a linked list. It looks up arena-size
878 and body-size from the body_details table described below, thus
879 supporting the multiple body-types.
881 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
882 the (new|del)_X*V macros are mapped directly to malloc/free.
884 For each sv-type, struct body_details bodies_by_type[] carries
885 parameters which control these aspects of SV handling:
887 Arena_size determines whether arenas are used for this body type, and if
888 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
889 zero, forcing individual mallocs and frees.
891 Body_size determines how big a body is, and therefore how many fit into
892 each arena. Offset carries the body-pointer adjustment needed for
893 "ghost fields", and is used in *_allocated macros.
895 But its main purpose is to parameterize info needed in
896 Perl_sv_upgrade(). The info here dramatically simplifies the function
897 vs the implementation in 5.8.8, making it table-driven. All fields
898 are used for this, except for arena_size.
900 For the sv-types that have no bodies, arenas are not used, so those
901 PL_body_roots[sv_type] are unused, and can be overloaded. In
902 something of a special case, SVt_NULL is borrowed for HE arenas;
903 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
904 bodies_by_type[SVt_NULL] slot is not used, as the table is not
909 struct body_details {
910 U8 body_size; /* Size to allocate */
911 U8 copy; /* Size of structure to copy (may be shorter) */
912 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
913 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
914 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
915 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
916 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
917 U32 arena_size; /* Size of arena to allocate */
925 /* With -DPURFIY we allocate everything directly, and don't use arenas.
926 This seems a rather elegant way to simplify some of the code below. */
927 #define HASARENA FALSE
929 #define HASARENA TRUE
931 #define NOARENA FALSE
933 /* Size the arenas to exactly fit a given number of bodies. A count
934 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
935 simplifying the default. If count > 0, the arena is sized to fit
936 only that many bodies, allowing arenas to be used for large, rare
937 bodies (XPVFM, XPVIO) without undue waste. The arena size is
938 limited by PERL_ARENA_SIZE, so we can safely oversize the
941 #define FIT_ARENA0(body_size) \
942 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
943 #define FIT_ARENAn(count,body_size) \
944 ( count * body_size <= PERL_ARENA_SIZE) \
945 ? count * body_size \
946 : FIT_ARENA0 (body_size)
947 #define FIT_ARENA(count,body_size) \
949 ? FIT_ARENAn (count, body_size) \
950 : FIT_ARENA0 (body_size))
952 /* Calculate the length to copy. Specifically work out the length less any
953 final padding the compiler needed to add. See the comment in sv_upgrade
954 for why copying the padding proved to be a bug. */
956 #define copy_length(type, last_member) \
957 STRUCT_OFFSET(type, last_member) \
958 + sizeof (((type*)SvANY((const SV *)0))->last_member)
960 static const struct body_details bodies_by_type[] = {
961 /* HEs use this offset for their arena. */
962 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
964 /* IVs are in the head, so the allocation size is 0. */
966 sizeof(IV), /* This is used to copy out the IV body. */
967 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
968 NOARENA /* IVS don't need an arena */, 0
973 STRUCT_OFFSET(XPVNV, xnv_u),
974 SVt_NV, FALSE, HADNV, NOARENA, 0 },
976 { sizeof(NV), sizeof(NV),
977 STRUCT_OFFSET(XPVNV, xnv_u),
978 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
981 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
982 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
983 + STRUCT_OFFSET(XPV, xpv_cur),
984 SVt_PV, FALSE, NONV, HASARENA,
985 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
987 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
988 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
989 + STRUCT_OFFSET(XPV, xpv_cur),
990 SVt_INVLIST, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
993 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
994 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
995 + STRUCT_OFFSET(XPV, xpv_cur),
996 SVt_PVIV, FALSE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
999 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
1000 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
1001 + STRUCT_OFFSET(XPV, xpv_cur),
1002 SVt_PVNV, FALSE, HADNV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
1005 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
1006 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
1011 SVt_REGEXP, TRUE, NONV, HASARENA,
1012 FIT_ARENA(0, sizeof(regexp))
1015 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
1016 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
1018 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1019 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
1022 copy_length(XPVAV, xav_alloc),
1024 SVt_PVAV, TRUE, NONV, HASARENA,
1025 FIT_ARENA(0, sizeof(XPVAV)) },
1028 copy_length(XPVHV, xhv_max),
1030 SVt_PVHV, TRUE, NONV, HASARENA,
1031 FIT_ARENA(0, sizeof(XPVHV)) },
1036 SVt_PVCV, TRUE, NONV, HASARENA,
1037 FIT_ARENA(0, sizeof(XPVCV)) },
1042 SVt_PVFM, TRUE, NONV, NOARENA,
1043 FIT_ARENA(20, sizeof(XPVFM)) },
1048 SVt_PVIO, TRUE, NONV, HASARENA,
1049 FIT_ARENA(24, sizeof(XPVIO)) },
1052 #define new_body_allocated(sv_type) \
1053 (void *)((char *)S_new_body(aTHX_ sv_type) \
1054 - bodies_by_type[sv_type].offset)
1056 /* return a thing to the free list */
1058 #define del_body(thing, root) \
1060 void ** const thing_copy = (void **)thing; \
1061 *thing_copy = *root; \
1062 *root = (void*)thing_copy; \
1066 #if !(NVSIZE <= IVSIZE)
1067 # define new_XNV() safemalloc(sizeof(XPVNV))
1069 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1070 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1072 #define del_XPVGV(p) safefree(p)
1076 #if !(NVSIZE <= IVSIZE)
1077 # define new_XNV() new_body_allocated(SVt_NV)
1079 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1080 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1082 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1083 &PL_body_roots[SVt_PVGV])
1087 /* no arena for you! */
1089 #define new_NOARENA(details) \
1090 safemalloc((details)->body_size + (details)->offset)
1091 #define new_NOARENAZ(details) \
1092 safecalloc((details)->body_size + (details)->offset, 1)
1095 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1096 const size_t arena_size)
1098 void ** const root = &PL_body_roots[sv_type];
1099 struct arena_desc *adesc;
1100 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1104 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1105 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1108 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1109 static bool done_sanity_check;
1111 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1112 * variables like done_sanity_check. */
1113 if (!done_sanity_check) {
1114 unsigned int i = SVt_LAST;
1116 done_sanity_check = TRUE;
1119 assert (bodies_by_type[i].type == i);
1125 /* may need new arena-set to hold new arena */
1126 if (!aroot || aroot->curr >= aroot->set_size) {
1127 struct arena_set *newroot;
1128 Newxz(newroot, 1, struct arena_set);
1129 newroot->set_size = ARENAS_PER_SET;
1130 newroot->next = aroot;
1132 PL_body_arenas = (void *) newroot;
1133 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1136 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1137 curr = aroot->curr++;
1138 adesc = &(aroot->set[curr]);
1139 assert(!adesc->arena);
1141 Newx(adesc->arena, good_arena_size, char);
1142 adesc->size = good_arena_size;
1143 adesc->utype = sv_type;
1144 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1145 curr, (void*)adesc->arena, (UV)good_arena_size));
1147 start = (char *) adesc->arena;
1149 /* Get the address of the byte after the end of the last body we can fit.
1150 Remember, this is integer division: */
1151 end = start + good_arena_size / body_size * body_size;
1153 /* computed count doesn't reflect the 1st slot reservation */
1154 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1155 DEBUG_m(PerlIO_printf(Perl_debug_log,
1156 "arena %p end %p arena-size %d (from %d) type %d "
1158 (void*)start, (void*)end, (int)good_arena_size,
1159 (int)arena_size, sv_type, (int)body_size,
1160 (int)good_arena_size / (int)body_size));
1162 DEBUG_m(PerlIO_printf(Perl_debug_log,
1163 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1164 (void*)start, (void*)end,
1165 (int)arena_size, sv_type, (int)body_size,
1166 (int)good_arena_size / (int)body_size));
1168 *root = (void *)start;
1171 /* Where the next body would start: */
1172 char * const next = start + body_size;
1175 /* This is the last body: */
1176 assert(next == end);
1178 *(void **)start = 0;
1182 *(void**) start = (void *)next;
1187 /* grab a new thing from the free list, allocating more if necessary.
1188 The inline version is used for speed in hot routines, and the
1189 function using it serves the rest (unless PURIFY).
1191 #define new_body_inline(xpv, sv_type) \
1193 void ** const r3wt = &PL_body_roots[sv_type]; \
1194 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1195 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1196 bodies_by_type[sv_type].body_size,\
1197 bodies_by_type[sv_type].arena_size)); \
1198 *(r3wt) = *(void**)(xpv); \
1204 S_new_body(pTHX_ const svtype sv_type)
1207 new_body_inline(xpv, sv_type);
1213 static const struct body_details fake_rv =
1214 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1217 =for apidoc sv_upgrade
1219 Upgrade an SV to a more complex form. Generally adds a new body type to the
1220 SV, then copies across as much information as possible from the old body.
1221 It croaks if the SV is already in a more complex form than requested. You
1222 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1223 before calling C<sv_upgrade>, and hence does not croak. See also
1230 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1234 const svtype old_type = SvTYPE(sv);
1235 const struct body_details *new_type_details;
1236 const struct body_details *old_type_details
1237 = bodies_by_type + old_type;
1238 SV *referant = NULL;
1240 PERL_ARGS_ASSERT_SV_UPGRADE;
1242 if (old_type == new_type)
1245 /* This clause was purposefully added ahead of the early return above to
1246 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1247 inference by Nick I-S that it would fix other troublesome cases. See
1248 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1250 Given that shared hash key scalars are no longer PVIV, but PV, there is
1251 no longer need to unshare so as to free up the IVX slot for its proper
1252 purpose. So it's safe to move the early return earlier. */
1254 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1255 sv_force_normal_flags(sv, 0);
1258 old_body = SvANY(sv);
1260 /* Copying structures onto other structures that have been neatly zeroed
1261 has a subtle gotcha. Consider XPVMG
1263 +------+------+------+------+------+-------+-------+
1264 | NV | CUR | LEN | IV | MAGIC | STASH |
1265 +------+------+------+------+------+-------+-------+
1266 0 4 8 12 16 20 24 28
1268 where NVs are aligned to 8 bytes, so that sizeof that structure is
1269 actually 32 bytes long, with 4 bytes of padding at the end:
1271 +------+------+------+------+------+-------+-------+------+
1272 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1273 +------+------+------+------+------+-------+-------+------+
1274 0 4 8 12 16 20 24 28 32
1276 so what happens if you allocate memory for this structure:
1278 +------+------+------+------+------+-------+-------+------+------+...
1279 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1280 +------+------+------+------+------+-------+-------+------+------+...
1281 0 4 8 12 16 20 24 28 32 36
1283 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1284 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1285 started out as zero once, but it's quite possible that it isn't. So now,
1286 rather than a nicely zeroed GP, you have it pointing somewhere random.
1289 (In fact, GP ends up pointing at a previous GP structure, because the
1290 principle cause of the padding in XPVMG getting garbage is a copy of
1291 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1292 this happens to be moot because XPVGV has been re-ordered, with GP
1293 no longer after STASH)
1295 So we are careful and work out the size of used parts of all the
1303 referant = SvRV(sv);
1304 old_type_details = &fake_rv;
1305 if (new_type == SVt_NV)
1306 new_type = SVt_PVNV;
1308 if (new_type < SVt_PVIV) {
1309 new_type = (new_type == SVt_NV)
1310 ? SVt_PVNV : SVt_PVIV;
1315 if (new_type < SVt_PVNV) {
1316 new_type = SVt_PVNV;
1320 assert(new_type > SVt_PV);
1321 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1322 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1329 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1330 there's no way that it can be safely upgraded, because perl.c
1331 expects to Safefree(SvANY(PL_mess_sv)) */
1332 assert(sv != PL_mess_sv);
1335 if (UNLIKELY(old_type_details->cant_upgrade))
1336 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1337 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1340 if (UNLIKELY(old_type > new_type))
1341 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1342 (int)old_type, (int)new_type);
1344 new_type_details = bodies_by_type + new_type;
1346 SvFLAGS(sv) &= ~SVTYPEMASK;
1347 SvFLAGS(sv) |= new_type;
1349 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1350 the return statements above will have triggered. */
1351 assert (new_type != SVt_NULL);
1354 assert(old_type == SVt_NULL);
1355 SET_SVANY_FOR_BODYLESS_IV(sv);
1359 assert(old_type == SVt_NULL);
1360 #if NVSIZE <= IVSIZE
1361 SET_SVANY_FOR_BODYLESS_NV(sv);
1363 SvANY(sv) = new_XNV();
1369 assert(new_type_details->body_size);
1372 assert(new_type_details->arena);
1373 assert(new_type_details->arena_size);
1374 /* This points to the start of the allocated area. */
1375 new_body_inline(new_body, new_type);
1376 Zero(new_body, new_type_details->body_size, char);
1377 new_body = ((char *)new_body) - new_type_details->offset;
1379 /* We always allocated the full length item with PURIFY. To do this
1380 we fake things so that arena is false for all 16 types.. */
1381 new_body = new_NOARENAZ(new_type_details);
1383 SvANY(sv) = new_body;
1384 if (new_type == SVt_PVAV) {
1388 if (old_type_details->body_size) {
1391 /* It will have been zeroed when the new body was allocated.
1392 Lets not write to it, in case it confuses a write-back
1398 #ifndef NODEFAULT_SHAREKEYS
1399 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1401 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1402 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1405 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1406 The target created by newSVrv also is, and it can have magic.
1407 However, it never has SvPVX set.
1409 if (old_type == SVt_IV) {
1411 } else if (old_type >= SVt_PV) {
1412 assert(SvPVX_const(sv) == 0);
1415 if (old_type >= SVt_PVMG) {
1416 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1417 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1419 sv->sv_u.svu_array = NULL; /* or svu_hash */
1424 /* XXX Is this still needed? Was it ever needed? Surely as there is
1425 no route from NV to PVIV, NOK can never be true */
1426 assert(!SvNOKp(sv));
1440 assert(new_type_details->body_size);
1441 /* We always allocated the full length item with PURIFY. To do this
1442 we fake things so that arena is false for all 16 types.. */
1443 if(new_type_details->arena) {
1444 /* This points to the start of the allocated area. */
1445 new_body_inline(new_body, new_type);
1446 Zero(new_body, new_type_details->body_size, char);
1447 new_body = ((char *)new_body) - new_type_details->offset;
1449 new_body = new_NOARENAZ(new_type_details);
1451 SvANY(sv) = new_body;
1453 if (old_type_details->copy) {
1454 /* There is now the potential for an upgrade from something without
1455 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1456 int offset = old_type_details->offset;
1457 int length = old_type_details->copy;
1459 if (new_type_details->offset > old_type_details->offset) {
1460 const int difference
1461 = new_type_details->offset - old_type_details->offset;
1462 offset += difference;
1463 length -= difference;
1465 assert (length >= 0);
1467 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1471 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1472 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1473 * correct 0.0 for us. Otherwise, if the old body didn't have an
1474 * NV slot, but the new one does, then we need to initialise the
1475 * freshly created NV slot with whatever the correct bit pattern is
1477 if (old_type_details->zero_nv && !new_type_details->zero_nv
1478 && !isGV_with_GP(sv))
1482 if (UNLIKELY(new_type == SVt_PVIO)) {
1483 IO * const io = MUTABLE_IO(sv);
1484 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1487 /* Clear the stashcache because a new IO could overrule a package
1489 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1490 hv_clear(PL_stashcache);
1492 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1493 IoPAGE_LEN(sv) = 60;
1495 if (UNLIKELY(new_type == SVt_REGEXP))
1496 sv->sv_u.svu_rx = (regexp *)new_body;
1497 else if (old_type < SVt_PV) {
1498 /* referant will be NULL unless the old type was SVt_IV emulating
1500 sv->sv_u.svu_rv = referant;
1504 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1505 (unsigned long)new_type);
1508 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1509 and sometimes SVt_NV */
1510 if (old_type_details->body_size) {
1514 /* Note that there is an assumption that all bodies of types that
1515 can be upgraded came from arenas. Only the more complex non-
1516 upgradable types are allowed to be directly malloc()ed. */
1517 assert(old_type_details->arena);
1518 del_body((void*)((char*)old_body + old_type_details->offset),
1519 &PL_body_roots[old_type]);
1525 =for apidoc sv_backoff
1527 Remove any string offset. You should normally use the C<SvOOK_off> macro
1534 Perl_sv_backoff(SV *const sv)
1537 const char * const s = SvPVX_const(sv);
1539 PERL_ARGS_ASSERT_SV_BACKOFF;
1542 assert(SvTYPE(sv) != SVt_PVHV);
1543 assert(SvTYPE(sv) != SVt_PVAV);
1545 SvOOK_offset(sv, delta);
1547 SvLEN_set(sv, SvLEN(sv) + delta);
1548 SvPV_set(sv, SvPVX(sv) - delta);
1549 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1550 SvFLAGS(sv) &= ~SVf_OOK;
1557 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1558 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1559 Use the C<SvGROW> wrapper instead.
1564 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1567 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1571 PERL_ARGS_ASSERT_SV_GROW;
1575 if (SvTYPE(sv) < SVt_PV) {
1576 sv_upgrade(sv, SVt_PV);
1577 s = SvPVX_mutable(sv);
1579 else if (SvOOK(sv)) { /* pv is offset? */
1581 s = SvPVX_mutable(sv);
1582 if (newlen > SvLEN(sv))
1583 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1587 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1588 s = SvPVX_mutable(sv);
1591 #ifdef PERL_NEW_COPY_ON_WRITE
1592 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1593 * to store the COW count. So in general, allocate one more byte than
1594 * asked for, to make it likely this byte is always spare: and thus
1595 * make more strings COW-able.
1596 * If the new size is a big power of two, don't bother: we assume the
1597 * caller wanted a nice 2^N sized block and will be annoyed at getting
1599 * Only increment if the allocation isn't MEM_SIZE_MAX,
1600 * otherwise it will wrap to 0.
1602 if (newlen & 0xff && newlen != MEM_SIZE_MAX)
1606 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1607 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1610 if (newlen > SvLEN(sv)) { /* need more room? */
1611 STRLEN minlen = SvCUR(sv);
1612 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1613 if (newlen < minlen)
1615 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1617 /* Don't round up on the first allocation, as odds are pretty good that
1618 * the initial request is accurate as to what is really needed */
1620 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1621 if (rounded > newlen)
1625 if (SvLEN(sv) && s) {
1626 s = (char*)saferealloc(s, newlen);
1629 s = (char*)safemalloc(newlen);
1630 if (SvPVX_const(sv) && SvCUR(sv)) {
1631 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1635 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1636 /* Do this here, do it once, do it right, and then we will never get
1637 called back into sv_grow() unless there really is some growing
1639 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1641 SvLEN_set(sv, newlen);
1648 =for apidoc sv_setiv
1650 Copies an integer into the given SV, upgrading first if necessary.
1651 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1657 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1659 PERL_ARGS_ASSERT_SV_SETIV;
1661 SV_CHECK_THINKFIRST_COW_DROP(sv);
1662 switch (SvTYPE(sv)) {
1665 sv_upgrade(sv, SVt_IV);
1668 sv_upgrade(sv, SVt_PVIV);
1672 if (!isGV_with_GP(sv))
1679 /* diag_listed_as: Can't coerce %s to %s in %s */
1680 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1684 (void)SvIOK_only(sv); /* validate number */
1690 =for apidoc sv_setiv_mg
1692 Like C<sv_setiv>, but also handles 'set' magic.
1698 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1700 PERL_ARGS_ASSERT_SV_SETIV_MG;
1707 =for apidoc sv_setuv
1709 Copies an unsigned integer into the given SV, upgrading first if necessary.
1710 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1716 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1718 PERL_ARGS_ASSERT_SV_SETUV;
1720 /* With the if statement to ensure that integers are stored as IVs whenever
1722 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1725 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1727 If you wish to remove the following if statement, so that this routine
1728 (and its callers) always return UVs, please benchmark to see what the
1729 effect is. Modern CPUs may be different. Or may not :-)
1731 if (u <= (UV)IV_MAX) {
1732 sv_setiv(sv, (IV)u);
1741 =for apidoc sv_setuv_mg
1743 Like C<sv_setuv>, but also handles 'set' magic.
1749 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1751 PERL_ARGS_ASSERT_SV_SETUV_MG;
1758 =for apidoc sv_setnv
1760 Copies a double into the given SV, upgrading first if necessary.
1761 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1767 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1769 PERL_ARGS_ASSERT_SV_SETNV;
1771 SV_CHECK_THINKFIRST_COW_DROP(sv);
1772 switch (SvTYPE(sv)) {
1775 sv_upgrade(sv, SVt_NV);
1779 sv_upgrade(sv, SVt_PVNV);
1783 if (!isGV_with_GP(sv))
1790 /* diag_listed_as: Can't coerce %s to %s in %s */
1791 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1796 (void)SvNOK_only(sv); /* validate number */
1801 =for apidoc sv_setnv_mg
1803 Like C<sv_setnv>, but also handles 'set' magic.
1809 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1811 PERL_ARGS_ASSERT_SV_SETNV_MG;
1817 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1818 * not incrementable warning display.
1819 * Originally part of S_not_a_number().
1820 * The return value may be != tmpbuf.
1824 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1827 PERL_ARGS_ASSERT_SV_DISPLAY;
1830 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1831 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1834 const char * const limit = tmpbuf + tmpbuf_size - 8;
1835 /* each *s can expand to 4 chars + "...\0",
1836 i.e. need room for 8 chars */
1838 const char *s = SvPVX_const(sv);
1839 const char * const end = s + SvCUR(sv);
1840 for ( ; s < end && d < limit; s++ ) {
1842 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1846 /* Map to ASCII "equivalent" of Latin1 */
1847 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1853 else if (ch == '\r') {
1857 else if (ch == '\f') {
1861 else if (ch == '\\') {
1865 else if (ch == '\0') {
1869 else if (isPRINT_LC(ch))
1888 /* Print an "isn't numeric" warning, using a cleaned-up,
1889 * printable version of the offending string
1893 S_not_a_number(pTHX_ SV *const sv)
1898 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1900 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1903 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1904 /* diag_listed_as: Argument "%s" isn't numeric%s */
1905 "Argument \"%s\" isn't numeric in %s", pv,
1908 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1909 /* diag_listed_as: Argument "%s" isn't numeric%s */
1910 "Argument \"%s\" isn't numeric", pv);
1914 S_not_incrementable(pTHX_ SV *const sv) {
1918 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1920 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1922 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1923 "Argument \"%s\" treated as 0 in increment (++)", pv);
1927 =for apidoc looks_like_number
1929 Test if the content of an SV looks like a number (or is a number).
1930 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1931 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1938 Perl_looks_like_number(pTHX_ SV *const sv)
1944 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1946 if (SvPOK(sv) || SvPOKp(sv)) {
1947 sbegin = SvPV_nomg_const(sv, len);
1950 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1951 numtype = grok_number(sbegin, len, NULL);
1952 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1956 S_glob_2number(pTHX_ GV * const gv)
1958 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1960 /* We know that all GVs stringify to something that is not-a-number,
1961 so no need to test that. */
1962 if (ckWARN(WARN_NUMERIC))
1964 SV *const buffer = sv_newmortal();
1965 gv_efullname3(buffer, gv, "*");
1966 not_a_number(buffer);
1968 /* We just want something true to return, so that S_sv_2iuv_common
1969 can tail call us and return true. */
1973 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1974 until proven guilty, assume that things are not that bad... */
1979 As 64 bit platforms often have an NV that doesn't preserve all bits of
1980 an IV (an assumption perl has been based on to date) it becomes necessary
1981 to remove the assumption that the NV always carries enough precision to
1982 recreate the IV whenever needed, and that the NV is the canonical form.
1983 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1984 precision as a side effect of conversion (which would lead to insanity
1985 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1986 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1987 where precision was lost, and IV/UV/NV slots that have a valid conversion
1988 which has lost no precision
1989 2) to ensure that if a numeric conversion to one form is requested that
1990 would lose precision, the precise conversion (or differently
1991 imprecise conversion) is also performed and cached, to prevent
1992 requests for different numeric formats on the same SV causing
1993 lossy conversion chains. (lossless conversion chains are perfectly
1998 SvIOKp is true if the IV slot contains a valid value
1999 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
2000 SvNOKp is true if the NV slot contains a valid value
2001 SvNOK is true only if the NV value is accurate
2004 while converting from PV to NV, check to see if converting that NV to an
2005 IV(or UV) would lose accuracy over a direct conversion from PV to
2006 IV(or UV). If it would, cache both conversions, return NV, but mark
2007 SV as IOK NOKp (ie not NOK).
2009 While converting from PV to IV, check to see if converting that IV to an
2010 NV would lose accuracy over a direct conversion from PV to NV. If it
2011 would, cache both conversions, flag similarly.
2013 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2014 correctly because if IV & NV were set NV *always* overruled.
2015 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2016 changes - now IV and NV together means that the two are interchangeable:
2017 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
2019 The benefit of this is that operations such as pp_add know that if
2020 SvIOK is true for both left and right operands, then integer addition
2021 can be used instead of floating point (for cases where the result won't
2022 overflow). Before, floating point was always used, which could lead to
2023 loss of precision compared with integer addition.
2025 * making IV and NV equal status should make maths accurate on 64 bit
2027 * may speed up maths somewhat if pp_add and friends start to use
2028 integers when possible instead of fp. (Hopefully the overhead in
2029 looking for SvIOK and checking for overflow will not outweigh the
2030 fp to integer speedup)
2031 * will slow down integer operations (callers of SvIV) on "inaccurate"
2032 values, as the change from SvIOK to SvIOKp will cause a call into
2033 sv_2iv each time rather than a macro access direct to the IV slot
2034 * should speed up number->string conversion on integers as IV is
2035 favoured when IV and NV are equally accurate
2037 ####################################################################
2038 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2039 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2040 On the other hand, SvUOK is true iff UV.
2041 ####################################################################
2043 Your mileage will vary depending your CPU's relative fp to integer
2047 #ifndef NV_PRESERVES_UV
2048 # define IS_NUMBER_UNDERFLOW_IV 1
2049 # define IS_NUMBER_UNDERFLOW_UV 2
2050 # define IS_NUMBER_IV_AND_UV 2
2051 # define IS_NUMBER_OVERFLOW_IV 4
2052 # define IS_NUMBER_OVERFLOW_UV 5
2054 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2056 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2058 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2064 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2065 PERL_UNUSED_CONTEXT;
2067 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));
2068 if (SvNVX(sv) < (NV)IV_MIN) {
2069 (void)SvIOKp_on(sv);
2071 SvIV_set(sv, IV_MIN);
2072 return IS_NUMBER_UNDERFLOW_IV;
2074 if (SvNVX(sv) > (NV)UV_MAX) {
2075 (void)SvIOKp_on(sv);
2078 SvUV_set(sv, UV_MAX);
2079 return IS_NUMBER_OVERFLOW_UV;
2081 (void)SvIOKp_on(sv);
2083 /* Can't use strtol etc to convert this string. (See truth table in
2085 if (SvNVX(sv) <= (UV)IV_MAX) {
2086 SvIV_set(sv, I_V(SvNVX(sv)));
2087 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2088 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2090 /* Integer is imprecise. NOK, IOKp */
2092 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2095 SvUV_set(sv, U_V(SvNVX(sv)));
2096 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2097 if (SvUVX(sv) == UV_MAX) {
2098 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2099 possibly be preserved by NV. Hence, it must be overflow.
2101 return IS_NUMBER_OVERFLOW_UV;
2103 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2105 /* Integer is imprecise. NOK, IOKp */
2107 return IS_NUMBER_OVERFLOW_IV;
2109 #endif /* !NV_PRESERVES_UV*/
2111 /* If numtype is infnan, set the NV of the sv accordingly.
2112 * If numtype is anything else, try setting the NV using Atof(PV). */
2114 # pragma warning(push)
2115 # pragma warning(disable:4756;disable:4056)
2118 S_sv_setnv(pTHX_ SV* sv, int numtype)
2120 bool pok = cBOOL(SvPOK(sv));
2122 if ((numtype & IS_NUMBER_INFINITY)) {
2123 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2126 else if ((numtype & IS_NUMBER_NAN)) {
2127 SvNV_set(sv, NV_NAN);
2131 SvNV_set(sv, Atof(SvPVX_const(sv)));
2132 /* Purposefully no true nok here, since we don't want to blow
2133 * away the possible IOK/UV of an existing sv. */
2136 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2138 SvPOK_on(sv); /* PV is okay, though. */
2142 # pragma warning(pop)
2146 S_sv_2iuv_common(pTHX_ SV *const sv)
2148 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2151 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2152 * without also getting a cached IV/UV from it at the same time
2153 * (ie PV->NV conversion should detect loss of accuracy and cache
2154 * IV or UV at same time to avoid this. */
2155 /* IV-over-UV optimisation - choose to cache IV if possible */
2157 if (SvTYPE(sv) == SVt_NV)
2158 sv_upgrade(sv, SVt_PVNV);
2160 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2161 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2162 certainly cast into the IV range at IV_MAX, whereas the correct
2163 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2165 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2166 if (Perl_isnan(SvNVX(sv))) {
2172 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2173 SvIV_set(sv, I_V(SvNVX(sv)));
2174 if (SvNVX(sv) == (NV) SvIVX(sv)
2175 #ifndef NV_PRESERVES_UV
2176 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2177 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2178 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2179 /* Don't flag it as "accurately an integer" if the number
2180 came from a (by definition imprecise) NV operation, and
2181 we're outside the range of NV integer precision */
2185 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2187 /* scalar has trailing garbage, eg "42a" */
2189 DEBUG_c(PerlIO_printf(Perl_debug_log,
2190 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2196 /* IV not precise. No need to convert from PV, as NV
2197 conversion would already have cached IV if it detected
2198 that PV->IV would be better than PV->NV->IV
2199 flags already correct - don't set public IOK. */
2200 DEBUG_c(PerlIO_printf(Perl_debug_log,
2201 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2206 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2207 but the cast (NV)IV_MIN rounds to a the value less (more
2208 negative) than IV_MIN which happens to be equal to SvNVX ??
2209 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2210 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2211 (NV)UVX == NVX are both true, but the values differ. :-(
2212 Hopefully for 2s complement IV_MIN is something like
2213 0x8000000000000000 which will be exact. NWC */
2216 SvUV_set(sv, U_V(SvNVX(sv)));
2218 (SvNVX(sv) == (NV) SvUVX(sv))
2219 #ifndef NV_PRESERVES_UV
2220 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2221 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2222 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2223 /* Don't flag it as "accurately an integer" if the number
2224 came from a (by definition imprecise) NV operation, and
2225 we're outside the range of NV integer precision */
2231 DEBUG_c(PerlIO_printf(Perl_debug_log,
2232 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2238 else if (SvPOKp(sv)) {
2240 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2241 /* We want to avoid a possible problem when we cache an IV/ a UV which
2242 may be later translated to an NV, and the resulting NV is not
2243 the same as the direct translation of the initial string
2244 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2245 be careful to ensure that the value with the .456 is around if the
2246 NV value is requested in the future).
2248 This means that if we cache such an IV/a UV, we need to cache the
2249 NV as well. Moreover, we trade speed for space, and do not
2250 cache the NV if we are sure it's not needed.
2253 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2254 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2255 == IS_NUMBER_IN_UV) {
2256 /* It's definitely an integer, only upgrade to PVIV */
2257 if (SvTYPE(sv) < SVt_PVIV)
2258 sv_upgrade(sv, SVt_PVIV);
2260 } else if (SvTYPE(sv) < SVt_PVNV)
2261 sv_upgrade(sv, SVt_PVNV);
2263 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2264 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2266 S_sv_setnv(aTHX_ sv, numtype);
2270 /* If NVs preserve UVs then we only use the UV value if we know that
2271 we aren't going to call atof() below. If NVs don't preserve UVs
2272 then the value returned may have more precision than atof() will
2273 return, even though value isn't perfectly accurate. */
2274 if ((numtype & (IS_NUMBER_IN_UV
2275 #ifdef NV_PRESERVES_UV
2278 )) == IS_NUMBER_IN_UV) {
2279 /* This won't turn off the public IOK flag if it was set above */
2280 (void)SvIOKp_on(sv);
2282 if (!(numtype & IS_NUMBER_NEG)) {
2284 if (value <= (UV)IV_MAX) {
2285 SvIV_set(sv, (IV)value);
2287 /* it didn't overflow, and it was positive. */
2288 SvUV_set(sv, value);
2292 /* 2s complement assumption */
2293 if (value <= (UV)IV_MIN) {
2294 SvIV_set(sv, value == (UV)IV_MIN
2295 ? IV_MIN : -(IV)value);
2297 /* Too negative for an IV. This is a double upgrade, but
2298 I'm assuming it will be rare. */
2299 if (SvTYPE(sv) < SVt_PVNV)
2300 sv_upgrade(sv, SVt_PVNV);
2304 SvNV_set(sv, -(NV)value);
2305 SvIV_set(sv, IV_MIN);
2309 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2310 will be in the previous block to set the IV slot, and the next
2311 block to set the NV slot. So no else here. */
2313 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2314 != IS_NUMBER_IN_UV) {
2315 /* It wasn't an (integer that doesn't overflow the UV). */
2316 S_sv_setnv(aTHX_ sv, numtype);
2318 if (! numtype && ckWARN(WARN_NUMERIC))
2321 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2322 PTR2UV(sv), SvNVX(sv)));
2324 #ifdef NV_PRESERVES_UV
2325 (void)SvIOKp_on(sv);
2327 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2328 if (Perl_isnan(SvNVX(sv))) {
2334 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2335 SvIV_set(sv, I_V(SvNVX(sv)));
2336 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2339 NOOP; /* Integer is imprecise. NOK, IOKp */
2341 /* UV will not work better than IV */
2343 if (SvNVX(sv) > (NV)UV_MAX) {
2345 /* Integer is inaccurate. NOK, IOKp, is UV */
2346 SvUV_set(sv, UV_MAX);
2348 SvUV_set(sv, U_V(SvNVX(sv)));
2349 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2350 NV preservse UV so can do correct comparison. */
2351 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2354 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2359 #else /* NV_PRESERVES_UV */
2360 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2361 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2362 /* The IV/UV slot will have been set from value returned by
2363 grok_number above. The NV slot has just been set using
2366 assert (SvIOKp(sv));
2368 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2369 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2370 /* Small enough to preserve all bits. */
2371 (void)SvIOKp_on(sv);
2373 SvIV_set(sv, I_V(SvNVX(sv)));
2374 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2376 /* Assumption: first non-preserved integer is < IV_MAX,
2377 this NV is in the preserved range, therefore: */
2378 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2380 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);
2384 0 0 already failed to read UV.
2385 0 1 already failed to read UV.
2386 1 0 you won't get here in this case. IV/UV
2387 slot set, public IOK, Atof() unneeded.
2388 1 1 already read UV.
2389 so there's no point in sv_2iuv_non_preserve() attempting
2390 to use atol, strtol, strtoul etc. */
2392 sv_2iuv_non_preserve (sv, numtype);
2394 sv_2iuv_non_preserve (sv);
2398 #endif /* NV_PRESERVES_UV */
2399 /* It might be more code efficient to go through the entire logic above
2400 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2401 gets complex and potentially buggy, so more programmer efficient
2402 to do it this way, by turning off the public flags: */
2404 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2408 if (isGV_with_GP(sv))
2409 return glob_2number(MUTABLE_GV(sv));
2411 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2413 if (SvTYPE(sv) < SVt_IV)
2414 /* Typically the caller expects that sv_any is not NULL now. */
2415 sv_upgrade(sv, SVt_IV);
2416 /* Return 0 from the caller. */
2423 =for apidoc sv_2iv_flags
2425 Return the integer value of an SV, doing any necessary string
2426 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2427 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2433 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2435 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2437 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2438 && SvTYPE(sv) != SVt_PVFM);
2440 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2446 if (flags & SV_SKIP_OVERLOAD)
2448 tmpstr = AMG_CALLunary(sv, numer_amg);
2449 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2450 return SvIV(tmpstr);
2453 return PTR2IV(SvRV(sv));
2456 if (SvVALID(sv) || isREGEXP(sv)) {
2457 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2458 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2459 In practice they are extremely unlikely to actually get anywhere
2460 accessible by user Perl code - the only way that I'm aware of is when
2461 a constant subroutine which is used as the second argument to index.
2463 Regexps have no SvIVX and SvNVX fields.
2465 assert(isREGEXP(sv) || SvPOKp(sv));
2468 const char * const ptr =
2469 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2471 = grok_number(ptr, SvCUR(sv), &value);
2473 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2474 == IS_NUMBER_IN_UV) {
2475 /* It's definitely an integer */
2476 if (numtype & IS_NUMBER_NEG) {
2477 if (value < (UV)IV_MIN)
2480 if (value < (UV)IV_MAX)
2485 /* Quite wrong but no good choices. */
2486 if ((numtype & IS_NUMBER_INFINITY)) {
2487 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2488 } else if ((numtype & IS_NUMBER_NAN)) {
2489 return 0; /* So wrong. */
2493 if (ckWARN(WARN_NUMERIC))
2496 return I_V(Atof(ptr));
2500 if (SvTHINKFIRST(sv)) {
2501 #ifdef PERL_OLD_COPY_ON_WRITE
2503 sv_force_normal_flags(sv, 0);
2506 if (SvREADONLY(sv) && !SvOK(sv)) {
2507 if (ckWARN(WARN_UNINITIALIZED))
2514 if (S_sv_2iuv_common(aTHX_ sv))
2518 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2519 PTR2UV(sv),SvIVX(sv)));
2520 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2524 =for apidoc sv_2uv_flags
2526 Return the unsigned integer value of an SV, doing any necessary string
2527 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2528 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2534 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2536 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2538 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2544 if (flags & SV_SKIP_OVERLOAD)
2546 tmpstr = AMG_CALLunary(sv, numer_amg);
2547 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2548 return SvUV(tmpstr);
2551 return PTR2UV(SvRV(sv));
2554 if (SvVALID(sv) || isREGEXP(sv)) {
2555 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2556 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2557 Regexps have no SvIVX and SvNVX fields. */
2558 assert(isREGEXP(sv) || SvPOKp(sv));
2561 const char * const ptr =
2562 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2564 = grok_number(ptr, SvCUR(sv), &value);
2566 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2567 == IS_NUMBER_IN_UV) {
2568 /* It's definitely an integer */
2569 if (!(numtype & IS_NUMBER_NEG))
2573 /* Quite wrong but no good choices. */
2574 if ((numtype & IS_NUMBER_INFINITY)) {
2575 return UV_MAX; /* So wrong. */
2576 } else if ((numtype & IS_NUMBER_NAN)) {
2577 return 0; /* So wrong. */
2581 if (ckWARN(WARN_NUMERIC))
2584 return U_V(Atof(ptr));
2588 if (SvTHINKFIRST(sv)) {
2589 #ifdef PERL_OLD_COPY_ON_WRITE
2591 sv_force_normal_flags(sv, 0);
2594 if (SvREADONLY(sv) && !SvOK(sv)) {
2595 if (ckWARN(WARN_UNINITIALIZED))
2602 if (S_sv_2iuv_common(aTHX_ sv))
2606 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2607 PTR2UV(sv),SvUVX(sv)));
2608 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2612 =for apidoc sv_2nv_flags
2614 Return the num value of an SV, doing any necessary string or integer
2615 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2616 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2622 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2624 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2626 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2627 && SvTYPE(sv) != SVt_PVFM);
2628 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2629 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2630 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2631 Regexps have no SvIVX and SvNVX fields. */
2633 if (flags & SV_GMAGIC)
2637 if (SvPOKp(sv) && !SvIOKp(sv)) {
2638 ptr = SvPVX_const(sv);
2640 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2641 !grok_number(ptr, SvCUR(sv), NULL))
2647 return (NV)SvUVX(sv);
2649 return (NV)SvIVX(sv);
2655 ptr = RX_WRAPPED((REGEXP *)sv);
2658 assert(SvTYPE(sv) >= SVt_PVMG);
2659 /* This falls through to the report_uninit near the end of the
2661 } else if (SvTHINKFIRST(sv)) {
2666 if (flags & SV_SKIP_OVERLOAD)
2668 tmpstr = AMG_CALLunary(sv, numer_amg);
2669 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2670 return SvNV(tmpstr);
2673 return PTR2NV(SvRV(sv));
2675 #ifdef PERL_OLD_COPY_ON_WRITE
2677 sv_force_normal_flags(sv, 0);
2680 if (SvREADONLY(sv) && !SvOK(sv)) {
2681 if (ckWARN(WARN_UNINITIALIZED))
2686 if (SvTYPE(sv) < SVt_NV) {
2687 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2688 sv_upgrade(sv, SVt_NV);
2690 STORE_NUMERIC_LOCAL_SET_STANDARD();
2691 PerlIO_printf(Perl_debug_log,
2692 "0x%"UVxf" num(%" NVgf ")\n",
2693 PTR2UV(sv), SvNVX(sv));
2694 RESTORE_NUMERIC_LOCAL();
2697 else if (SvTYPE(sv) < SVt_PVNV)
2698 sv_upgrade(sv, SVt_PVNV);
2703 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2704 #ifdef NV_PRESERVES_UV
2710 /* Only set the public NV OK flag if this NV preserves the IV */
2711 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2713 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2714 : (SvIVX(sv) == I_V(SvNVX(sv))))
2720 else if (SvPOKp(sv)) {
2722 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2723 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2725 #ifdef NV_PRESERVES_UV
2726 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2727 == IS_NUMBER_IN_UV) {
2728 /* It's definitely an integer */
2729 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2731 S_sv_setnv(aTHX_ sv, numtype);
2738 SvNV_set(sv, Atof(SvPVX_const(sv)));
2739 /* Only set the public NV OK flag if this NV preserves the value in
2740 the PV at least as well as an IV/UV would.
2741 Not sure how to do this 100% reliably. */
2742 /* if that shift count is out of range then Configure's test is
2743 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2745 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2746 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2747 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2748 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2749 /* Can't use strtol etc to convert this string, so don't try.
2750 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2753 /* value has been set. It may not be precise. */
2754 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2755 /* 2s complement assumption for (UV)IV_MIN */
2756 SvNOK_on(sv); /* Integer is too negative. */
2761 if (numtype & IS_NUMBER_NEG) {
2762 /* -IV_MIN is undefined, but we should never reach
2763 * this point with both IS_NUMBER_NEG and value ==
2765 assert(value != (UV)IV_MIN);
2766 SvIV_set(sv, -(IV)value);
2767 } else if (value <= (UV)IV_MAX) {
2768 SvIV_set(sv, (IV)value);
2770 SvUV_set(sv, value);
2774 if (numtype & IS_NUMBER_NOT_INT) {
2775 /* I believe that even if the original PV had decimals,
2776 they are lost beyond the limit of the FP precision.
2777 However, neither is canonical, so both only get p
2778 flags. NWC, 2000/11/25 */
2779 /* Both already have p flags, so do nothing */
2781 const NV nv = SvNVX(sv);
2782 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2783 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2784 if (SvIVX(sv) == I_V(nv)) {
2787 /* It had no "." so it must be integer. */
2791 /* between IV_MAX and NV(UV_MAX).
2792 Could be slightly > UV_MAX */
2794 if (numtype & IS_NUMBER_NOT_INT) {
2795 /* UV and NV both imprecise. */
2797 const UV nv_as_uv = U_V(nv);
2799 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2808 /* It might be more code efficient to go through the entire logic above
2809 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2810 gets complex and potentially buggy, so more programmer efficient
2811 to do it this way, by turning off the public flags: */
2813 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2814 #endif /* NV_PRESERVES_UV */
2817 if (isGV_with_GP(sv)) {
2818 glob_2number(MUTABLE_GV(sv));
2822 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2824 assert (SvTYPE(sv) >= SVt_NV);
2825 /* Typically the caller expects that sv_any is not NULL now. */
2826 /* XXX Ilya implies that this is a bug in callers that assume this
2827 and ideally should be fixed. */
2831 STORE_NUMERIC_LOCAL_SET_STANDARD();
2832 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2833 PTR2UV(sv), SvNVX(sv));
2834 RESTORE_NUMERIC_LOCAL();
2842 Return an SV with the numeric value of the source SV, doing any necessary
2843 reference or overload conversion. The caller is expected to have handled
2850 Perl_sv_2num(pTHX_ SV *const sv)
2852 PERL_ARGS_ASSERT_SV_2NUM;
2857 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2858 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2859 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2860 return sv_2num(tmpsv);
2862 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2865 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2866 * UV as a string towards the end of buf, and return pointers to start and
2869 * We assume that buf is at least TYPE_CHARS(UV) long.
2873 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2875 char *ptr = buf + TYPE_CHARS(UV);
2876 char * const ebuf = ptr;
2879 PERL_ARGS_ASSERT_UIV_2BUF;
2887 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2891 *--ptr = '0' + (char)(uv % 10);
2899 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2900 * infinity or a not-a-number, writes the appropriate strings to the
2901 * buffer, including a zero byte. On success returns the written length,
2902 * excluding the zero byte, on failure (not an infinity, not a nan, or the
2903 * maxlen too small) returns zero.
2905 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2906 * shared string constants we point to, instead of generating a new
2907 * string for each instance. */
2909 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2910 assert(maxlen >= 4);
2911 if (maxlen < 4) /* "Inf\0", "NaN\0" */
2915 if (Perl_isinf(nv)) {
2917 if (maxlen < 5) /* "-Inf\0" */
2926 } else if (Perl_isnan(nv)) {
2930 /* XXX optionally output the payload mantissa bits as
2931 * "(unsigned)" (to match the nan("...") C99 function,
2932 * or maybe as "(0xhhh...)" would make more sense...
2933 * provide a format string so that the user can decide?
2934 * NOTE: would affect the maxlen and assert() logic.*/
2939 assert((s == buffer + 3) || (s == buffer + 4));
2941 return s - buffer - 1; /* -1: excluding the zero byte */
2946 =for apidoc sv_2pv_flags
2948 Returns a pointer to the string value of an SV, and sets *lp to its length.
2949 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2950 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2951 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2957 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2961 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2963 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2964 && SvTYPE(sv) != SVt_PVFM);
2965 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2970 if (flags & SV_SKIP_OVERLOAD)
2972 tmpstr = AMG_CALLunary(sv, string_amg);
2973 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2974 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2976 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2980 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2981 if (flags & SV_CONST_RETURN) {
2982 pv = (char *) SvPVX_const(tmpstr);
2984 pv = (flags & SV_MUTABLE_RETURN)
2985 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2988 *lp = SvCUR(tmpstr);
2990 pv = sv_2pv_flags(tmpstr, lp, flags);
3003 SV *const referent = SvRV(sv);
3007 retval = buffer = savepvn("NULLREF", len);
3008 } else if (SvTYPE(referent) == SVt_REGEXP &&
3009 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3010 amagic_is_enabled(string_amg))) {
3011 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3015 /* If the regex is UTF-8 we want the containing scalar to
3016 have an UTF-8 flag too */
3023 *lp = RX_WRAPLEN(re);
3025 return RX_WRAPPED(re);
3027 const char *const typestr = sv_reftype(referent, 0);
3028 const STRLEN typelen = strlen(typestr);
3029 UV addr = PTR2UV(referent);
3030 const char *stashname = NULL;
3031 STRLEN stashnamelen = 0; /* hush, gcc */
3032 const char *buffer_end;
3034 if (SvOBJECT(referent)) {
3035 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3038 stashname = HEK_KEY(name);
3039 stashnamelen = HEK_LEN(name);
3041 if (HEK_UTF8(name)) {
3047 stashname = "__ANON__";
3050 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3051 + 2 * sizeof(UV) + 2 /* )\0 */;
3053 len = typelen + 3 /* (0x */
3054 + 2 * sizeof(UV) + 2 /* )\0 */;
3057 Newx(buffer, len, char);
3058 buffer_end = retval = buffer + len;
3060 /* Working backwards */
3064 *--retval = PL_hexdigit[addr & 15];
3065 } while (addr >>= 4);
3071 memcpy(retval, typestr, typelen);
3075 retval -= stashnamelen;
3076 memcpy(retval, stashname, stashnamelen);
3078 /* retval may not necessarily have reached the start of the
3080 assert (retval >= buffer);
3082 len = buffer_end - retval - 1; /* -1 for that \0 */
3094 if (flags & SV_MUTABLE_RETURN)
3095 return SvPVX_mutable(sv);
3096 if (flags & SV_CONST_RETURN)
3097 return (char *)SvPVX_const(sv);
3102 /* I'm assuming that if both IV and NV are equally valid then
3103 converting the IV is going to be more efficient */
3104 const U32 isUIOK = SvIsUV(sv);
3105 char buf[TYPE_CHARS(UV)];
3109 if (SvTYPE(sv) < SVt_PVIV)
3110 sv_upgrade(sv, SVt_PVIV);
3111 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3113 /* inlined from sv_setpvn */
3114 s = SvGROW_mutable(sv, len + 1);
3115 Move(ptr, s, len, char);
3120 else if (SvNOK(sv)) {
3121 if (SvTYPE(sv) < SVt_PVNV)
3122 sv_upgrade(sv, SVt_PVNV);
3123 if (SvNVX(sv) == 0.0
3124 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3125 && !Perl_isnan(SvNVX(sv))
3128 s = SvGROW_mutable(sv, 2);
3133 STRLEN size = 5; /* "-Inf\0" */
3135 s = SvGROW_mutable(sv, size);
3136 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3142 /* some Xenix systems wipe out errno here */
3151 5 + /* exponent digits */
3155 s = SvGROW_mutable(sv, size);
3156 #ifndef USE_LOCALE_NUMERIC
3157 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3163 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3167 PL_numeric_radix_sv &&
3168 SvUTF8(PL_numeric_radix_sv);
3169 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3170 size += SvLEN(PL_numeric_radix_sv) - 1;
3171 s = SvGROW_mutable(sv, size);
3174 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3176 /* If the radix character is UTF-8, and actually is in the
3177 * output, turn on the UTF-8 flag for the scalar */
3179 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3183 RESTORE_LC_NUMERIC();
3186 /* We don't call SvPOK_on(), because it may come to
3187 * pass that the locale changes so that the
3188 * stringification we just did is no longer correct. We
3189 * will have to re-stringify every time it is needed */
3196 else if (isGV_with_GP(sv)) {
3197 GV *const gv = MUTABLE_GV(sv);
3198 SV *const buffer = sv_newmortal();
3200 gv_efullname3(buffer, gv, "*");
3202 assert(SvPOK(buffer));
3206 *lp = SvCUR(buffer);
3207 return SvPVX(buffer);
3209 else if (isREGEXP(sv)) {
3210 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3211 return RX_WRAPPED((REGEXP *)sv);
3216 if (flags & SV_UNDEF_RETURNS_NULL)
3218 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3220 /* Typically the caller expects that sv_any is not NULL now. */
3221 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3222 sv_upgrade(sv, SVt_PV);
3227 const STRLEN len = s - SvPVX_const(sv);
3232 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3233 PTR2UV(sv),SvPVX_const(sv)));
3234 if (flags & SV_CONST_RETURN)
3235 return (char *)SvPVX_const(sv);
3236 if (flags & SV_MUTABLE_RETURN)
3237 return SvPVX_mutable(sv);
3242 =for apidoc sv_copypv
3244 Copies a stringified representation of the source SV into the
3245 destination SV. Automatically performs any necessary mg_get and
3246 coercion of numeric values into strings. Guaranteed to preserve
3247 UTF8 flag even from overloaded objects. Similar in nature to
3248 sv_2pv[_flags] but operates directly on an SV instead of just the
3249 string. Mostly uses sv_2pv_flags to do its work, except when that
3250 would lose the UTF-8'ness of the PV.
3252 =for apidoc sv_copypv_nomg
3254 Like sv_copypv, but doesn't invoke get magic first.
3256 =for apidoc sv_copypv_flags
3258 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3265 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3270 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3272 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3273 sv_setpvn(dsv,s,len);
3281 =for apidoc sv_2pvbyte
3283 Return a pointer to the byte-encoded representation of the SV, and set *lp
3284 to its length. May cause the SV to be downgraded from UTF-8 as a
3287 Usually accessed via the C<SvPVbyte> macro.
3293 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3295 PERL_ARGS_ASSERT_SV_2PVBYTE;
3298 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3299 || isGV_with_GP(sv) || SvROK(sv)) {
3300 SV *sv2 = sv_newmortal();
3301 sv_copypv_nomg(sv2,sv);
3304 sv_utf8_downgrade(sv,0);
3305 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3309 =for apidoc sv_2pvutf8
3311 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3312 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3314 Usually accessed via the C<SvPVutf8> macro.
3320 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3322 PERL_ARGS_ASSERT_SV_2PVUTF8;
3324 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3325 || isGV_with_GP(sv) || SvROK(sv))
3326 sv = sv_mortalcopy(sv);
3329 sv_utf8_upgrade_nomg(sv);
3330 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3335 =for apidoc sv_2bool
3337 This macro is only used by sv_true() or its macro equivalent, and only if
3338 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3339 It calls sv_2bool_flags with the SV_GMAGIC flag.
3341 =for apidoc sv_2bool_flags
3343 This function is only used by sv_true() and friends, and only if
3344 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3345 contain SV_GMAGIC, then it does an mg_get() first.
3352 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3354 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3357 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3363 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3364 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3367 if(SvGMAGICAL(sv)) {
3369 goto restart; /* call sv_2bool */
3371 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3372 else if(!SvOK(sv)) {
3375 else if(SvPOK(sv)) {
3376 svb = SvPVXtrue(sv);
3378 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3379 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3380 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3384 goto restart; /* call sv_2bool_nomg */
3389 return SvRV(sv) != 0;
3393 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3394 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3398 =for apidoc sv_utf8_upgrade
3400 Converts the PV of an SV to its UTF-8-encoded form.
3401 Forces the SV to string form if it is not already.
3402 Will C<mg_get> on C<sv> if appropriate.
3403 Always sets the SvUTF8 flag to avoid future validity checks even
3404 if the whole string is the same in UTF-8 as not.
3405 Returns the number of bytes in the converted string
3407 This is not a general purpose byte encoding to Unicode interface:
3408 use the Encode extension for that.
3410 =for apidoc sv_utf8_upgrade_nomg
3412 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3414 =for apidoc sv_utf8_upgrade_flags
3416 Converts the PV of an SV to its UTF-8-encoded form.
3417 Forces the SV to string form if it is not already.
3418 Always sets the SvUTF8 flag to avoid future validity checks even
3419 if all the bytes are invariant in UTF-8.
3420 If C<flags> has C<SV_GMAGIC> bit set,
3421 will C<mg_get> on C<sv> if appropriate, else not.
3423 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3424 will expand when converted to UTF-8, and skips the extra work of checking for
3425 that. Typically this flag is used by a routine that has already parsed the
3426 string and found such characters, and passes this information on so that the
3427 work doesn't have to be repeated.
3429 Returns the number of bytes in the converted string.
3431 This is not a general purpose byte encoding to Unicode interface:
3432 use the Encode extension for that.
3434 =for apidoc sv_utf8_upgrade_flags_grow
3436 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3437 the number of unused bytes the string of 'sv' is guaranteed to have free after
3438 it upon return. This allows the caller to reserve extra space that it intends
3439 to fill, to avoid extra grows.
3441 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3442 are implemented in terms of this function.
3444 Returns the number of bytes in the converted string (not including the spares).
3448 (One might think that the calling routine could pass in the position of the
3449 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3450 have to be found again. But that is not the case, because typically when the
3451 caller is likely to use this flag, it won't be calling this routine unless it
3452 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3453 and just use bytes. But some things that do fit into a byte are variants in
3454 utf8, and the caller may not have been keeping track of these.)
3456 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3457 C<NUL> isn't guaranteed due to having other routines do the work in some input
3458 cases, or if the input is already flagged as being in utf8.
3460 The speed of this could perhaps be improved for many cases if someone wanted to
3461 write a fast function that counts the number of variant characters in a string,
3462 especially if it could return the position of the first one.
3467 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3469 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3471 if (sv == &PL_sv_undef)
3473 if (!SvPOK_nog(sv)) {
3475 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3476 (void) sv_2pv_flags(sv,&len, flags);
3478 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3482 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3487 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3492 S_sv_uncow(aTHX_ sv, 0);
3495 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3496 sv_recode_to_utf8(sv, _get_encoding());
3497 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3501 if (SvCUR(sv) == 0) {
3502 if (extra) SvGROW(sv, extra);
3503 } else { /* Assume Latin-1/EBCDIC */
3504 /* This function could be much more efficient if we
3505 * had a FLAG in SVs to signal if there are any variant
3506 * chars in the PV. Given that there isn't such a flag
3507 * make the loop as fast as possible (although there are certainly ways
3508 * to speed this up, eg. through vectorization) */
3509 U8 * s = (U8 *) SvPVX_const(sv);
3510 U8 * e = (U8 *) SvEND(sv);
3512 STRLEN two_byte_count = 0;
3514 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3516 /* See if really will need to convert to utf8. We mustn't rely on our
3517 * incoming SV being well formed and having a trailing '\0', as certain
3518 * code in pp_formline can send us partially built SVs. */
3522 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3524 t--; /* t already incremented; re-point to first variant */
3529 /* utf8 conversion not needed because all are invariants. Mark as
3530 * UTF-8 even if no variant - saves scanning loop */
3532 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3537 /* Here, the string should be converted to utf8, either because of an
3538 * input flag (two_byte_count = 0), or because a character that
3539 * requires 2 bytes was found (two_byte_count = 1). t points either to
3540 * the beginning of the string (if we didn't examine anything), or to
3541 * the first variant. In either case, everything from s to t - 1 will
3542 * occupy only 1 byte each on output.
3544 * There are two main ways to convert. One is to create a new string
3545 * and go through the input starting from the beginning, appending each
3546 * converted value onto the new string as we go along. It's probably
3547 * best to allocate enough space in the string for the worst possible
3548 * case rather than possibly running out of space and having to
3549 * reallocate and then copy what we've done so far. Since everything
3550 * from s to t - 1 is invariant, the destination can be initialized
3551 * with these using a fast memory copy
3553 * The other way is to figure out exactly how big the string should be
3554 * by parsing the entire input. Then you don't have to make it big
3555 * enough to handle the worst possible case, and more importantly, if
3556 * the string you already have is large enough, you don't have to
3557 * allocate a new string, you can copy the last character in the input
3558 * string to the final position(s) that will be occupied by the
3559 * converted string and go backwards, stopping at t, since everything
3560 * before that is invariant.
3562 * There are advantages and disadvantages to each method.
3564 * In the first method, we can allocate a new string, do the memory
3565 * copy from the s to t - 1, and then proceed through the rest of the
3566 * string byte-by-byte.
3568 * In the second method, we proceed through the rest of the input
3569 * string just calculating how big the converted string will be. Then
3570 * there are two cases:
3571 * 1) if the string has enough extra space to handle the converted
3572 * value. We go backwards through the string, converting until we
3573 * get to the position we are at now, and then stop. If this
3574 * position is far enough along in the string, this method is
3575 * faster than the other method. If the memory copy were the same
3576 * speed as the byte-by-byte loop, that position would be about
3577 * half-way, as at the half-way mark, parsing to the end and back
3578 * is one complete string's parse, the same amount as starting
3579 * over and going all the way through. Actually, it would be
3580 * somewhat less than half-way, as it's faster to just count bytes
3581 * than to also copy, and we don't have the overhead of allocating
3582 * a new string, changing the scalar to use it, and freeing the
3583 * existing one. But if the memory copy is fast, the break-even
3584 * point is somewhere after half way. The counting loop could be
3585 * sped up by vectorization, etc, to move the break-even point
3586 * further towards the beginning.
3587 * 2) if the string doesn't have enough space to handle the converted
3588 * value. A new string will have to be allocated, and one might
3589 * as well, given that, start from the beginning doing the first
3590 * method. We've spent extra time parsing the string and in
3591 * exchange all we've gotten is that we know precisely how big to
3592 * make the new one. Perl is more optimized for time than space,
3593 * so this case is a loser.
3594 * So what I've decided to do is not use the 2nd method unless it is
3595 * guaranteed that a new string won't have to be allocated, assuming
3596 * the worst case. I also decided not to put any more conditions on it
3597 * than this, for now. It seems likely that, since the worst case is
3598 * twice as big as the unknown portion of the string (plus 1), we won't
3599 * be guaranteed enough space, causing us to go to the first method,
3600 * unless the string is short, or the first variant character is near
3601 * the end of it. In either of these cases, it seems best to use the
3602 * 2nd method. The only circumstance I can think of where this would
3603 * be really slower is if the string had once had much more data in it
3604 * than it does now, but there is still a substantial amount in it */
3607 STRLEN invariant_head = t - s;
3608 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3609 if (SvLEN(sv) < size) {
3611 /* Here, have decided to allocate a new string */
3616 Newx(dst, size, U8);
3618 /* If no known invariants at the beginning of the input string,
3619 * set so starts from there. Otherwise, can use memory copy to
3620 * get up to where we are now, and then start from here */
3622 if (invariant_head == 0) {
3625 Copy(s, dst, invariant_head, char);
3626 d = dst + invariant_head;
3630 append_utf8_from_native_byte(*t, &d);
3634 SvPV_free(sv); /* No longer using pre-existing string */
3635 SvPV_set(sv, (char*)dst);
3636 SvCUR_set(sv, d - dst);
3637 SvLEN_set(sv, size);
3640 /* Here, have decided to get the exact size of the string.
3641 * Currently this happens only when we know that there is
3642 * guaranteed enough space to fit the converted string, so
3643 * don't have to worry about growing. If two_byte_count is 0,
3644 * then t points to the first byte of the string which hasn't
3645 * been examined yet. Otherwise two_byte_count is 1, and t
3646 * points to the first byte in the string that will expand to
3647 * two. Depending on this, start examining at t or 1 after t.
3650 U8 *d = t + two_byte_count;
3653 /* Count up the remaining bytes that expand to two */
3656 const U8 chr = *d++;
3657 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3660 /* The string will expand by just the number of bytes that
3661 * occupy two positions. But we are one afterwards because of
3662 * the increment just above. This is the place to put the
3663 * trailing NUL, and to set the length before we decrement */
3665 d += two_byte_count;
3666 SvCUR_set(sv, d - s);
3670 /* Having decremented d, it points to the position to put the
3671 * very last byte of the expanded string. Go backwards through
3672 * the string, copying and expanding as we go, stopping when we
3673 * get to the part that is invariant the rest of the way down */
3677 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3680 *d-- = UTF8_EIGHT_BIT_LO(*e);
3681 *d-- = UTF8_EIGHT_BIT_HI(*e);
3687 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3688 /* Update pos. We do it at the end rather than during
3689 * the upgrade, to avoid slowing down the common case
3690 * (upgrade without pos).
3691 * pos can be stored as either bytes or characters. Since
3692 * this was previously a byte string we can just turn off
3693 * the bytes flag. */
3694 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3696 mg->mg_flags &= ~MGf_BYTES;
3698 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3699 magic_setutf8(sv,mg); /* clear UTF8 cache */
3704 /* Mark as UTF-8 even if no variant - saves scanning loop */
3710 =for apidoc sv_utf8_downgrade
3712 Attempts to convert the PV of an SV from characters to bytes.
3713 If the PV contains a character that cannot fit
3714 in a byte, this conversion will fail;
3715 in this case, either returns false or, if C<fail_ok> is not
3718 This is not a general purpose Unicode to byte encoding interface:
3719 use the Encode extension for that.
3725 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3727 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3729 if (SvPOKp(sv) && SvUTF8(sv)) {
3733 int mg_flags = SV_GMAGIC;
3736 S_sv_uncow(aTHX_ sv, 0);
3738 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3740 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3741 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3742 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3743 SV_GMAGIC|SV_CONST_RETURN);
3744 mg_flags = 0; /* sv_pos_b2u does get magic */
3746 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3747 magic_setutf8(sv,mg); /* clear UTF8 cache */
3750 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3752 if (!utf8_to_bytes(s, &len)) {
3757 Perl_croak(aTHX_ "Wide character in %s",
3760 Perl_croak(aTHX_ "Wide character");
3771 =for apidoc sv_utf8_encode
3773 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3774 flag off so that it looks like octets again.
3780 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3782 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3784 if (SvREADONLY(sv)) {
3785 sv_force_normal_flags(sv, 0);
3787 (void) sv_utf8_upgrade(sv);
3792 =for apidoc sv_utf8_decode
3794 If the PV of the SV is an octet sequence in UTF-8
3795 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3796 so that it looks like a character. If the PV contains only single-byte
3797 characters, the C<SvUTF8> flag stays off.
3798 Scans PV for validity and returns false if the PV is invalid UTF-8.
3804 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3806 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3809 const U8 *start, *c;
3812 /* The octets may have got themselves encoded - get them back as
3815 if (!sv_utf8_downgrade(sv, TRUE))
3818 /* it is actually just a matter of turning the utf8 flag on, but
3819 * we want to make sure everything inside is valid utf8 first.
3821 c = start = (const U8 *) SvPVX_const(sv);
3822 if (!is_utf8_string(c, SvCUR(sv)))
3824 e = (const U8 *) SvEND(sv);
3827 if (!UTF8_IS_INVARIANT(ch)) {
3832 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3833 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3834 after this, clearing pos. Does anything on CPAN
3836 /* adjust pos to the start of a UTF8 char sequence */
3837 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3839 I32 pos = mg->mg_len;
3841 for (c = start + pos; c > start; c--) {
3842 if (UTF8_IS_START(*c))
3845 mg->mg_len = c - start;
3848 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3849 magic_setutf8(sv,mg); /* clear UTF8 cache */
3856 =for apidoc sv_setsv
3858 Copies the contents of the source SV C<ssv> into the destination SV
3859 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3860 function if the source SV needs to be reused. Does not handle 'set' magic on
3861 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3862 performs a copy-by-value, obliterating any previous content of the
3865 You probably want to use one of the assortment of wrappers, such as
3866 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3867 C<SvSetMagicSV_nosteal>.
3869 =for apidoc sv_setsv_flags
3871 Copies the contents of the source SV C<ssv> into the destination SV
3872 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3873 function if the source SV needs to be reused. Does not handle 'set' magic.
3874 Loosely speaking, it performs a copy-by-value, obliterating any previous
3875 content of the destination.
3876 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3877 C<ssv> if appropriate, else not. If the C<flags>
3878 parameter has the C<SV_NOSTEAL> bit set then the
3879 buffers of temps will not be stolen. <sv_setsv>
3880 and C<sv_setsv_nomg> are implemented in terms of this function.
3882 You probably want to use one of the assortment of wrappers, such as
3883 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3884 C<SvSetMagicSV_nosteal>.
3886 This is the primary function for copying scalars, and most other
3887 copy-ish functions and macros use this underneath.
3893 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3895 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3896 HV *old_stash = NULL;
3898 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3900 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3901 const char * const name = GvNAME(sstr);
3902 const STRLEN len = GvNAMELEN(sstr);
3904 if (dtype >= SVt_PV) {
3910 SvUPGRADE(dstr, SVt_PVGV);
3911 (void)SvOK_off(dstr);
3912 isGV_with_GP_on(dstr);
3914 GvSTASH(dstr) = GvSTASH(sstr);
3916 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3917 gv_name_set(MUTABLE_GV(dstr), name, len,
3918 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3919 SvFAKE_on(dstr); /* can coerce to non-glob */
3922 if(GvGP(MUTABLE_GV(sstr))) {
3923 /* If source has method cache entry, clear it */
3925 SvREFCNT_dec(GvCV(sstr));
3926 GvCV_set(sstr, NULL);
3929 /* If source has a real method, then a method is
3932 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3938 /* If dest already had a real method, that's a change as well */
3940 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3941 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3946 /* We don't need to check the name of the destination if it was not a
3947 glob to begin with. */
3948 if(dtype == SVt_PVGV) {
3949 const char * const name = GvNAME((const GV *)dstr);
3952 /* The stash may have been detached from the symbol table, so
3954 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3958 const STRLEN len = GvNAMELEN(dstr);
3959 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3960 || (len == 1 && name[0] == ':')) {
3963 /* Set aside the old stash, so we can reset isa caches on
3965 if((old_stash = GvHV(dstr)))
3966 /* Make sure we do not lose it early. */
3967 SvREFCNT_inc_simple_void_NN(
3968 sv_2mortal((SV *)old_stash)
3973 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3976 gp_free(MUTABLE_GV(dstr));
3977 GvINTRO_off(dstr); /* one-shot flag */
3978 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3979 if (SvTAINTED(sstr))
3981 if (GvIMPORTED(dstr) != GVf_IMPORTED
3982 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3984 GvIMPORTED_on(dstr);
3987 if(mro_changes == 2) {
3988 if (GvAV((const GV *)sstr)) {
3990 SV * const sref = (SV *)GvAV((const GV *)dstr);
3991 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3992 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3993 AV * const ary = newAV();
3994 av_push(ary, mg->mg_obj); /* takes the refcount */
3995 mg->mg_obj = (SV *)ary;
3997 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3999 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
4001 mro_isa_changed_in(GvSTASH(dstr));
4003 else if(mro_changes == 3) {
4004 HV * const stash = GvHV(dstr);
4005 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
4011 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
4012 if (GvIO(dstr) && dtype == SVt_PVGV) {
4013 DEBUG_o(Perl_deb(aTHX_
4014 "glob_assign_glob clearing PL_stashcache\n"));
4015 /* It's a cache. It will rebuild itself quite happily.
4016 It's a lot of effort to work out exactly which key (or keys)
4017 might be invalidated by the creation of the this file handle.
4019 hv_clear(PL_stashcache);
4025 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4027 SV * const sref = SvRV(sstr);
4029 const int intro = GvINTRO(dstr);
4032 const U32 stype = SvTYPE(sref);
4034 PERL_ARGS_ASSERT_GV_SETREF;
4037 GvINTRO_off(dstr); /* one-shot flag */
4038 GvLINE(dstr) = CopLINE(PL_curcop);
4039 GvEGV(dstr) = MUTABLE_GV(dstr);
4044 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4045 import_flag = GVf_IMPORTED_CV;
4048 location = (SV **) &GvHV(dstr);
4049 import_flag = GVf_IMPORTED_HV;
4052 location = (SV **) &GvAV(dstr);
4053 import_flag = GVf_IMPORTED_AV;
4056 location = (SV **) &GvIOp(dstr);
4059 location = (SV **) &GvFORM(dstr);
4062 location = &GvSV(dstr);
4063 import_flag = GVf_IMPORTED_SV;
4066 if (stype == SVt_PVCV) {
4067 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4068 if (GvCVGEN(dstr)) {
4069 SvREFCNT_dec(GvCV(dstr));
4070 GvCV_set(dstr, NULL);
4071 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4074 /* SAVEt_GVSLOT takes more room on the savestack and has more
4075 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4076 leave_scope needs access to the GV so it can reset method
4077 caches. We must use SAVEt_GVSLOT whenever the type is
4078 SVt_PVCV, even if the stash is anonymous, as the stash may
4079 gain a name somehow before leave_scope. */
4080 if (stype == SVt_PVCV) {
4081 /* There is no save_pushptrptrptr. Creating it for this
4082 one call site would be overkill. So inline the ss add
4086 SS_ADD_PTR(location);
4087 SS_ADD_PTR(SvREFCNT_inc(*location));
4088 SS_ADD_UV(SAVEt_GVSLOT);
4091 else SAVEGENERICSV(*location);
4094 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4095 CV* const cv = MUTABLE_CV(*location);
4097 if (!GvCVGEN((const GV *)dstr) &&
4098 (CvROOT(cv) || CvXSUB(cv)) &&
4099 /* redundant check that avoids creating the extra SV
4100 most of the time: */
4101 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4103 SV * const new_const_sv =
4104 CvCONST((const CV *)sref)
4105 ? cv_const_sv((const CV *)sref)
4107 report_redefined_cv(
4108 sv_2mortal(Perl_newSVpvf(aTHX_
4111 HvNAME_HEK(GvSTASH((const GV *)dstr))
4113 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4116 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4120 cv_ckproto_len_flags(cv, (const GV *)dstr,
4121 SvPOK(sref) ? CvPROTO(sref) : NULL,
4122 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4123 SvPOK(sref) ? SvUTF8(sref) : 0);
4125 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4126 GvASSUMECV_on(dstr);
4127 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4128 if (intro && GvREFCNT(dstr) > 1) {
4129 /* temporary remove extra savestack's ref */
4131 gv_method_changed(dstr);
4134 else gv_method_changed(dstr);
4137 *location = SvREFCNT_inc_simple_NN(sref);
4138 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4139 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4140 GvFLAGS(dstr) |= import_flag;
4142 if (import_flag == GVf_IMPORTED_SV) {
4144 save_aliased_sv((GV *)dstr);
4146 /* Turn off the flag if sref is not referenced elsewhere,
4147 even by weak refs. (SvRMAGICAL is a pessimistic check for
4149 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4150 GvALIASED_SV_off(dstr);
4152 GvALIASED_SV_on(dstr);
4154 if (stype == SVt_PVHV) {
4155 const char * const name = GvNAME((GV*)dstr);
4156 const STRLEN len = GvNAMELEN(dstr);
4159 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4160 || (len == 1 && name[0] == ':')
4162 && (!dref || HvENAME_get(dref))
4165 (HV *)sref, (HV *)dref,
4171 stype == SVt_PVAV && sref != dref
4172 && strEQ(GvNAME((GV*)dstr), "ISA")
4173 /* The stash may have been detached from the symbol table, so
4174 check its name before doing anything. */
4175 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4178 MAGIC * const omg = dref && SvSMAGICAL(dref)
4179 ? mg_find(dref, PERL_MAGIC_isa)
4181 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4182 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4183 AV * const ary = newAV();
4184 av_push(ary, mg->mg_obj); /* takes the refcount */
4185 mg->mg_obj = (SV *)ary;
4188 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4189 SV **svp = AvARRAY((AV *)omg->mg_obj);
4190 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4194 SvREFCNT_inc_simple_NN(*svp++)
4200 SvREFCNT_inc_simple_NN(omg->mg_obj)
4204 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4209 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4211 mg = mg_find(sref, PERL_MAGIC_isa);
4213 /* Since the *ISA assignment could have affected more than
4214 one stash, don't call mro_isa_changed_in directly, but let
4215 magic_clearisa do it for us, as it already has the logic for
4216 dealing with globs vs arrays of globs. */
4218 Perl_magic_clearisa(aTHX_ NULL, mg);
4220 else if (stype == SVt_PVIO) {
4221 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4222 /* It's a cache. It will rebuild itself quite happily.
4223 It's a lot of effort to work out exactly which key (or keys)
4224 might be invalidated by the creation of the this file handle.
4226 hv_clear(PL_stashcache);
4230 if (!intro) SvREFCNT_dec(dref);
4231 if (SvTAINTED(sstr))
4239 #ifdef PERL_DEBUG_READONLY_COW
4240 # include <sys/mman.h>
4242 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4243 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4247 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4249 struct perl_memory_debug_header * const header =
4250 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4251 const MEM_SIZE len = header->size;
4252 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4253 # ifdef PERL_TRACK_MEMPOOL
4254 if (!header->readonly) header->readonly = 1;
4256 if (mprotect(header, len, PROT_READ))
4257 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4258 header, len, errno);
4262 S_sv_buf_to_rw(pTHX_ SV *sv)
4264 struct perl_memory_debug_header * const header =
4265 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4266 const MEM_SIZE len = header->size;
4267 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4268 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4269 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4270 header, len, errno);
4271 # ifdef PERL_TRACK_MEMPOOL
4272 header->readonly = 0;
4277 # define sv_buf_to_ro(sv) NOOP
4278 # define sv_buf_to_rw(sv) NOOP
4282 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4288 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4290 if (UNLIKELY( sstr == dstr ))
4293 if (SvIS_FREED(dstr)) {
4294 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4295 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4297 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4298 if (UNLIKELY( !sstr ))
4299 sstr = &PL_sv_undef;
4300 if (SvIS_FREED(sstr)) {
4301 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4302 (void*)sstr, (void*)dstr);
4304 stype = SvTYPE(sstr);
4305 dtype = SvTYPE(dstr);
4307 /* There's a lot of redundancy below but we're going for speed here */
4312 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4313 (void)SvOK_off(dstr);
4321 /* For performance, we inline promoting to type SVt_IV. */
4322 /* We're starting from SVt_NULL, so provided that define is
4323 * actual 0, we don't have to unset any SV type flags
4324 * to promote to SVt_IV. */
4325 STATIC_ASSERT_STMT(SVt_NULL == 0);
4326 SET_SVANY_FOR_BODYLESS_IV(dstr);
4327 SvFLAGS(dstr) |= SVt_IV;
4331 sv_upgrade(dstr, SVt_PVIV);
4335 goto end_of_first_switch;
4337 (void)SvIOK_only(dstr);
4338 SvIV_set(dstr, SvIVX(sstr));
4341 /* SvTAINTED can only be true if the SV has taint magic, which in
4342 turn means that the SV type is PVMG (or greater). This is the
4343 case statement for SVt_IV, so this cannot be true (whatever gcov
4345 assert(!SvTAINTED(sstr));
4350 if (dtype < SVt_PV && dtype != SVt_IV)
4351 sv_upgrade(dstr, SVt_IV);
4355 if (LIKELY( SvNOK(sstr) )) {
4359 sv_upgrade(dstr, SVt_NV);
4363 sv_upgrade(dstr, SVt_PVNV);
4367 goto end_of_first_switch;
4369 SvNV_set(dstr, SvNVX(sstr));
4370 (void)SvNOK_only(dstr);
4371 /* SvTAINTED can only be true if the SV has taint magic, which in
4372 turn means that the SV type is PVMG (or greater). This is the
4373 case statement for SVt_NV, so this cannot be true (whatever gcov
4375 assert(!SvTAINTED(sstr));
4382 sv_upgrade(dstr, SVt_PV);
4385 if (dtype < SVt_PVIV)
4386 sv_upgrade(dstr, SVt_PVIV);
4389 if (dtype < SVt_PVNV)
4390 sv_upgrade(dstr, SVt_PVNV);
4394 const char * const type = sv_reftype(sstr,0);
4396 /* diag_listed_as: Bizarre copy of %s */
4397 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4399 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4401 NOT_REACHED; /* NOTREACHED */
4405 if (dtype < SVt_REGEXP)
4407 if (dtype >= SVt_PV) {
4413 sv_upgrade(dstr, SVt_REGEXP);
4421 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4423 if (SvTYPE(sstr) != stype)
4424 stype = SvTYPE(sstr);
4426 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4427 glob_assign_glob(dstr, sstr, dtype);
4430 if (stype == SVt_PVLV)
4432 if (isREGEXP(sstr)) goto upgregexp;
4433 SvUPGRADE(dstr, SVt_PVNV);
4436 SvUPGRADE(dstr, (svtype)stype);
4438 end_of_first_switch:
4440 /* dstr may have been upgraded. */
4441 dtype = SvTYPE(dstr);
4442 sflags = SvFLAGS(sstr);
4444 if (UNLIKELY( dtype == SVt_PVCV )) {
4445 /* Assigning to a subroutine sets the prototype. */
4448 const char *const ptr = SvPV_const(sstr, len);
4450 SvGROW(dstr, len + 1);
4451 Copy(ptr, SvPVX(dstr), len + 1, char);
4452 SvCUR_set(dstr, len);
4454 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4455 CvAUTOLOAD_off(dstr);
4460 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4461 || dtype == SVt_PVFM))
4463 const char * const type = sv_reftype(dstr,0);
4465 /* diag_listed_as: Cannot copy to %s */
4466 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4468 Perl_croak(aTHX_ "Cannot copy to %s", type);
4469 } else if (sflags & SVf_ROK) {
4470 if (isGV_with_GP(dstr)
4471 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4474 if (GvIMPORTED(dstr) != GVf_IMPORTED
4475 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4477 GvIMPORTED_on(dstr);
4482 glob_assign_glob(dstr, sstr, dtype);
4486 if (dtype >= SVt_PV) {
4487 if (isGV_with_GP(dstr)) {
4488 gv_setref(dstr, sstr);
4491 if (SvPVX_const(dstr)) {
4497 (void)SvOK_off(dstr);
4498 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4499 SvFLAGS(dstr) |= sflags & SVf_ROK;
4500 assert(!(sflags & SVp_NOK));
4501 assert(!(sflags & SVp_IOK));
4502 assert(!(sflags & SVf_NOK));
4503 assert(!(sflags & SVf_IOK));
4505 else if (isGV_with_GP(dstr)) {
4506 if (!(sflags & SVf_OK)) {
4507 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4508 "Undefined value assigned to typeglob");
4511 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4512 if (dstr != (const SV *)gv) {
4513 const char * const name = GvNAME((const GV *)dstr);
4514 const STRLEN len = GvNAMELEN(dstr);
4515 HV *old_stash = NULL;
4516 bool reset_isa = FALSE;
4517 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4518 || (len == 1 && name[0] == ':')) {
4519 /* Set aside the old stash, so we can reset isa caches
4520 on its subclasses. */
4521 if((old_stash = GvHV(dstr))) {
4522 /* Make sure we do not lose it early. */
4523 SvREFCNT_inc_simple_void_NN(
4524 sv_2mortal((SV *)old_stash)
4531 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4532 gp_free(MUTABLE_GV(dstr));
4534 GvGP_set(dstr, gp_ref(GvGP(gv)));
4537 HV * const stash = GvHV(dstr);
4539 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4549 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4550 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4551 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4553 else if (sflags & SVp_POK) {
4554 const STRLEN cur = SvCUR(sstr);
4555 const STRLEN len = SvLEN(sstr);
4558 * We have three basic ways to copy the string:
4564 * Which we choose is based on various factors. The following
4565 * things are listed in order of speed, fastest to slowest:
4567 * - Copying a short string
4568 * - Copy-on-write bookkeeping
4570 * - Copying a long string
4572 * We swipe the string (steal the string buffer) if the SV on the
4573 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4574 * big win on long strings. It should be a win on short strings if
4575 * SvPVX_const(dstr) has to be allocated. If not, it should not
4576 * slow things down, as SvPVX_const(sstr) would have been freed
4579 * We also steal the buffer from a PADTMP (operator target) if it
4580 * is ‘long enough’. For short strings, a swipe does not help
4581 * here, as it causes more malloc calls the next time the target
4582 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4583 * be allocated it is still not worth swiping PADTMPs for short
4584 * strings, as the savings here are small.
4586 * If swiping is not an option, then we see whether it is
4587 * worth using copy-on-write. If the lhs already has a buf-
4588 * fer big enough and the string is short, we skip it and fall back
4589 * to method 3, since memcpy is faster for short strings than the
4590 * later bookkeeping overhead that copy-on-write entails.
4592 * If the rhs is not a copy-on-write string yet, then we also
4593 * consider whether the buffer is too large relative to the string
4594 * it holds. Some operations such as readline allocate a large
4595 * buffer in the expectation of reusing it. But turning such into
4596 * a COW buffer is counter-productive because it increases memory
4597 * usage by making readline allocate a new large buffer the sec-
4598 * ond time round. So, if the buffer is too large, again, we use
4601 * Finally, if there is no buffer on the left, or the buffer is too
4602 * small, then we use copy-on-write and make both SVs share the
4607 /* Whichever path we take through the next code, we want this true,
4608 and doing it now facilitates the COW check. */
4609 (void)SvPOK_only(dstr);
4613 /* slated for free anyway (and not COW)? */
4614 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4615 /* or a swipable TARG */
4617 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4619 /* whose buffer is worth stealing */
4620 && CHECK_COWBUF_THRESHOLD(cur,len)
4623 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4624 (!(flags & SV_NOSTEAL)) &&
4625 /* and we're allowed to steal temps */
4626 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4627 len) /* and really is a string */
4628 { /* Passes the swipe test. */
4629 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4631 SvPV_set(dstr, SvPVX_mutable(sstr));
4632 SvLEN_set(dstr, SvLEN(sstr));
4633 SvCUR_set(dstr, SvCUR(sstr));
4636 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4637 SvPV_set(sstr, NULL);
4642 else if (flags & SV_COW_SHARED_HASH_KEYS
4644 #ifdef PERL_OLD_COPY_ON_WRITE
4645 ( sflags & SVf_IsCOW
4646 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4647 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4648 && SvTYPE(sstr) >= SVt_PVIV && len
4651 #elif defined(PERL_NEW_COPY_ON_WRITE)
4654 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4655 /* If this is a regular (non-hek) COW, only so
4656 many COW "copies" are possible. */
4657 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4658 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4659 && !(SvFLAGS(dstr) & SVf_BREAK)
4660 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4661 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4665 && !(SvFLAGS(dstr) & SVf_BREAK)
4668 /* Either it's a shared hash key, or it's suitable for
4671 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4676 if (!(sflags & SVf_IsCOW)) {
4678 # ifdef PERL_OLD_COPY_ON_WRITE
4679 /* Make the source SV into a loop of 1.
4680 (about to become 2) */
4681 SV_COW_NEXT_SV_SET(sstr, sstr);
4683 CowREFCNT(sstr) = 0;
4687 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4693 # ifdef PERL_OLD_COPY_ON_WRITE
4694 assert (SvTYPE(dstr) >= SVt_PVIV);
4695 /* SvIsCOW_normal */
4696 /* splice us in between source and next-after-source. */
4697 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4698 SV_COW_NEXT_SV_SET(sstr, dstr);
4700 if (sflags & SVf_IsCOW) {
4705 SvPV_set(dstr, SvPVX_mutable(sstr));
4710 /* SvIsCOW_shared_hash */
4711 DEBUG_C(PerlIO_printf(Perl_debug_log,
4712 "Copy on write: Sharing hash\n"));
4714 assert (SvTYPE(dstr) >= SVt_PV);
4716 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4718 SvLEN_set(dstr, len);
4719 SvCUR_set(dstr, cur);
4722 /* Failed the swipe test, and we cannot do copy-on-write either.
4723 Have to copy the string. */
4724 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4725 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4726 SvCUR_set(dstr, cur);
4727 *SvEND(dstr) = '\0';
4729 if (sflags & SVp_NOK) {
4730 SvNV_set(dstr, SvNVX(sstr));
4732 if (sflags & SVp_IOK) {
4733 SvIV_set(dstr, SvIVX(sstr));
4734 /* Must do this otherwise some other overloaded use of 0x80000000
4735 gets confused. I guess SVpbm_VALID */
4736 if (sflags & SVf_IVisUV)
4739 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4741 const MAGIC * const smg = SvVSTRING_mg(sstr);
4743 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4744 smg->mg_ptr, smg->mg_len);
4745 SvRMAGICAL_on(dstr);
4749 else if (sflags & (SVp_IOK|SVp_NOK)) {
4750 (void)SvOK_off(dstr);
4751 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4752 if (sflags & SVp_IOK) {
4753 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4754 SvIV_set(dstr, SvIVX(sstr));
4756 if (sflags & SVp_NOK) {
4757 SvNV_set(dstr, SvNVX(sstr));
4761 if (isGV_with_GP(sstr)) {
4762 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4765 (void)SvOK_off(dstr);
4767 if (SvTAINTED(sstr))
4772 =for apidoc sv_setsv_mg
4774 Like C<sv_setsv>, but also handles 'set' magic.
4780 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4782 PERL_ARGS_ASSERT_SV_SETSV_MG;
4784 sv_setsv(dstr,sstr);
4789 # ifdef PERL_OLD_COPY_ON_WRITE
4790 # define SVt_COW SVt_PVIV
4792 # define SVt_COW SVt_PV
4795 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4797 STRLEN cur = SvCUR(sstr);
4798 STRLEN len = SvLEN(sstr);
4800 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4801 const bool already = cBOOL(SvIsCOW(sstr));
4804 PERL_ARGS_ASSERT_SV_SETSV_COW;
4807 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4808 (void*)sstr, (void*)dstr);
4815 if (SvTHINKFIRST(dstr))
4816 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4817 else if (SvPVX_const(dstr))
4818 Safefree(SvPVX_mutable(dstr));
4822 SvUPGRADE(dstr, SVt_COW);
4824 assert (SvPOK(sstr));
4825 assert (SvPOKp(sstr));
4826 # ifdef PERL_OLD_COPY_ON_WRITE
4827 assert (!SvIOK(sstr));
4828 assert (!SvIOKp(sstr));
4829 assert (!SvNOK(sstr));
4830 assert (!SvNOKp(sstr));
4833 if (SvIsCOW(sstr)) {
4835 if (SvLEN(sstr) == 0) {
4836 /* source is a COW shared hash key. */
4837 DEBUG_C(PerlIO_printf(Perl_debug_log,
4838 "Fast copy on write: Sharing hash\n"));
4839 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4842 # ifdef PERL_OLD_COPY_ON_WRITE
4843 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4845 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4846 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4849 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4850 SvUPGRADE(sstr, SVt_COW);
4852 DEBUG_C(PerlIO_printf(Perl_debug_log,
4853 "Fast copy on write: Converting sstr to COW\n"));
4854 # ifdef PERL_OLD_COPY_ON_WRITE
4855 SV_COW_NEXT_SV_SET(dstr, sstr);
4857 CowREFCNT(sstr) = 0;
4860 # ifdef PERL_OLD_COPY_ON_WRITE
4861 SV_COW_NEXT_SV_SET(sstr, dstr);
4863 # ifdef PERL_DEBUG_READONLY_COW
4864 if (already) sv_buf_to_rw(sstr);
4868 new_pv = SvPVX_mutable(sstr);
4872 SvPV_set(dstr, new_pv);
4873 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4876 SvLEN_set(dstr, len);
4877 SvCUR_set(dstr, cur);
4886 =for apidoc sv_setpvn
4888 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4889 The C<len> parameter indicates the number of
4890 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4891 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4897 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4901 PERL_ARGS_ASSERT_SV_SETPVN;
4903 SV_CHECK_THINKFIRST_COW_DROP(sv);
4909 /* len is STRLEN which is unsigned, need to copy to signed */
4912 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4915 SvUPGRADE(sv, SVt_PV);
4917 dptr = SvGROW(sv, len + 1);
4918 Move(ptr,dptr,len,char);
4921 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4923 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4927 =for apidoc sv_setpvn_mg
4929 Like C<sv_setpvn>, but also handles 'set' magic.
4935 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4937 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4939 sv_setpvn(sv,ptr,len);
4944 =for apidoc sv_setpv
4946 Copies a string into an SV. The string must be terminated with a C<NUL>
4948 Does not handle 'set' magic. See C<sv_setpv_mg>.
4954 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4958 PERL_ARGS_ASSERT_SV_SETPV;
4960 SV_CHECK_THINKFIRST_COW_DROP(sv);
4966 SvUPGRADE(sv, SVt_PV);
4968 SvGROW(sv, len + 1);
4969 Move(ptr,SvPVX(sv),len+1,char);
4971 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4973 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4977 =for apidoc sv_setpv_mg
4979 Like C<sv_setpv>, but also handles 'set' magic.
4985 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4987 PERL_ARGS_ASSERT_SV_SETPV_MG;
4994 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4996 PERL_ARGS_ASSERT_SV_SETHEK;
5002 if (HEK_LEN(hek) == HEf_SVKEY) {
5003 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5006 const int flags = HEK_FLAGS(hek);
5007 if (flags & HVhek_WASUTF8) {
5008 STRLEN utf8_len = HEK_LEN(hek);
5009 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5010 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5013 } else if (flags & HVhek_UNSHARED) {
5014 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5017 else SvUTF8_off(sv);
5021 SV_CHECK_THINKFIRST_COW_DROP(sv);
5022 SvUPGRADE(sv, SVt_PV);
5024 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5025 SvCUR_set(sv, HEK_LEN(hek));
5031 else SvUTF8_off(sv);
5039 =for apidoc sv_usepvn_flags
5041 Tells an SV to use C<ptr> to find its string value. Normally the
5042 string is stored inside the SV, but sv_usepvn allows the SV to use an
5043 outside string. The C<ptr> should point to memory that was allocated
5044 by L<Newx|perlclib/Memory Management and String Handling>. It must be
5045 the start of a Newx-ed block of memory, and not a pointer to the
5046 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
5047 and not be from a non-Newx memory allocator like C<malloc>. The
5048 string length, C<len>, must be supplied. By default this function
5049 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5050 so that pointer should not be freed or used by the programmer after
5051 giving it to sv_usepvn, and neither should any pointers from "behind"
5052 that pointer (e.g. ptr + 1) be used.
5054 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
5055 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
5056 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5057 C<len>, and already meets the requirements for storing in C<SvPVX>).
5063 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5067 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5069 SV_CHECK_THINKFIRST_COW_DROP(sv);
5070 SvUPGRADE(sv, SVt_PV);
5073 if (flags & SV_SMAGIC)
5077 if (SvPVX_const(sv))
5081 if (flags & SV_HAS_TRAILING_NUL)
5082 assert(ptr[len] == '\0');
5085 allocate = (flags & SV_HAS_TRAILING_NUL)
5087 #ifdef Perl_safesysmalloc_size
5090 PERL_STRLEN_ROUNDUP(len + 1);
5092 if (flags & SV_HAS_TRAILING_NUL) {
5093 /* It's long enough - do nothing.
5094 Specifically Perl_newCONSTSUB is relying on this. */
5097 /* Force a move to shake out bugs in callers. */
5098 char *new_ptr = (char*)safemalloc(allocate);
5099 Copy(ptr, new_ptr, len, char);
5100 PoisonFree(ptr,len,char);
5104 ptr = (char*) saferealloc (ptr, allocate);
5107 #ifdef Perl_safesysmalloc_size
5108 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5110 SvLEN_set(sv, allocate);
5114 if (!(flags & SV_HAS_TRAILING_NUL)) {
5117 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5119 if (flags & SV_SMAGIC)
5123 #ifdef PERL_OLD_COPY_ON_WRITE
5124 /* Need to do this *after* making the SV normal, as we need the buffer
5125 pointer to remain valid until after we've copied it. If we let go too early,
5126 another thread could invalidate it by unsharing last of the same hash key
5127 (which it can do by means other than releasing copy-on-write Svs)
5128 or by changing the other copy-on-write SVs in the loop. */
5130 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5132 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5134 { /* this SV was SvIsCOW_normal(sv) */
5135 /* we need to find the SV pointing to us. */
5136 SV *current = SV_COW_NEXT_SV(after);
5138 if (current == sv) {
5139 /* The SV we point to points back to us (there were only two of us
5141 Hence other SV is no longer copy on write either. */
5143 sv_buf_to_rw(after);
5145 /* We need to follow the pointers around the loop. */
5147 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5150 /* don't loop forever if the structure is bust, and we have
5151 a pointer into a closed loop. */
5152 assert (current != after);
5153 assert (SvPVX_const(current) == pvx);
5155 /* Make the SV before us point to the SV after us. */
5156 SV_COW_NEXT_SV_SET(current, after);
5162 =for apidoc sv_force_normal_flags
5164 Undo various types of fakery on an SV, where fakery means
5165 "more than" a string: if the PV is a shared string, make
5166 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5167 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5168 we do the copy, and is also used locally; if this is a
5169 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5170 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5171 SvPOK_off rather than making a copy. (Used where this
5172 scalar is about to be set to some other value.) In addition,
5173 the C<flags> parameter gets passed to C<sv_unref_flags()>
5174 when unreffing. C<sv_force_normal> calls this function
5175 with flags set to 0.
5177 This function is expected to be used to signal to perl that this SV is
5178 about to be written to, and any extra book-keeping needs to be taken care
5179 of. Hence, it croaks on read-only values.
5185 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5187 assert(SvIsCOW(sv));
5190 const char * const pvx = SvPVX_const(sv);
5191 const STRLEN len = SvLEN(sv);
5192 const STRLEN cur = SvCUR(sv);
5193 # ifdef PERL_OLD_COPY_ON_WRITE
5194 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5195 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5196 we'll fail an assertion. */
5197 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5201 PerlIO_printf(Perl_debug_log,
5202 "Copy on write: Force normal %ld\n",
5207 # ifdef PERL_NEW_COPY_ON_WRITE
5209 /* Must do this first, since the CowREFCNT uses SvPVX and
5210 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5211 the only owner left of the buffer. */
5212 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5214 U8 cowrefcnt = CowREFCNT(sv);
5215 if(cowrefcnt != 0) {
5217 CowREFCNT(sv) = cowrefcnt;
5222 /* Else we are the only owner of the buffer. */
5227 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5232 if (flags & SV_COW_DROP_PV) {
5233 /* OK, so we don't need to copy our buffer. */
5236 SvGROW(sv, cur + 1);
5237 Move(pvx,SvPVX(sv),cur,char);
5242 # ifdef PERL_OLD_COPY_ON_WRITE
5243 sv_release_COW(sv, pvx, next);
5246 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5253 const char * const pvx = SvPVX_const(sv);
5254 const STRLEN len = SvCUR(sv);
5258 if (flags & SV_COW_DROP_PV) {
5259 /* OK, so we don't need to copy our buffer. */
5262 SvGROW(sv, len + 1);
5263 Move(pvx,SvPVX(sv),len,char);
5266 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5272 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5274 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5277 Perl_croak_no_modify();
5278 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5279 S_sv_uncow(aTHX_ sv, flags);
5281 sv_unref_flags(sv, flags);
5282 else if (SvFAKE(sv) && isGV_with_GP(sv))
5283 sv_unglob(sv, flags);
5284 else if (SvFAKE(sv) && isREGEXP(sv)) {
5285 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5286 to sv_unglob. We only need it here, so inline it. */
5287 const bool islv = SvTYPE(sv) == SVt_PVLV;
5288 const svtype new_type =
5289 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5290 SV *const temp = newSV_type(new_type);
5291 regexp *const temp_p = ReANY((REGEXP *)sv);
5293 if (new_type == SVt_PVMG) {
5294 SvMAGIC_set(temp, SvMAGIC(sv));
5295 SvMAGIC_set(sv, NULL);
5296 SvSTASH_set(temp, SvSTASH(sv));
5297 SvSTASH_set(sv, NULL);
5299 if (!islv) SvCUR_set(temp, SvCUR(sv));
5300 /* Remember that SvPVX is in the head, not the body. But
5301 RX_WRAPPED is in the body. */
5302 assert(ReANY((REGEXP *)sv)->mother_re);
5303 /* Their buffer is already owned by someone else. */
5304 if (flags & SV_COW_DROP_PV) {
5305 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5306 zeroed body. For SVt_PVLV, it should have been set to 0
5307 before turning into a regexp. */
5308 assert(!SvLEN(islv ? sv : temp));
5309 sv->sv_u.svu_pv = 0;
5312 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5313 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5317 /* Now swap the rest of the bodies. */
5321 SvFLAGS(sv) &= ~SVTYPEMASK;
5322 SvFLAGS(sv) |= new_type;
5323 SvANY(sv) = SvANY(temp);
5326 SvFLAGS(temp) &= ~(SVTYPEMASK);
5327 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5328 SvANY(temp) = temp_p;
5329 temp->sv_u.svu_rx = (regexp *)temp_p;
5331 SvREFCNT_dec_NN(temp);
5333 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5339 Efficient removal of characters from the beginning of the string buffer.
5340 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5341 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5342 character of the adjusted string. Uses the "OOK hack". On return, only
5343 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5345 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5346 refer to the same chunk of data.
5348 The unfortunate similarity of this function's name to that of Perl's C<chop>
5349 operator is strictly coincidental. This function works from the left;
5350 C<chop> works from the right.
5356 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5367 PERL_ARGS_ASSERT_SV_CHOP;
5369 if (!ptr || !SvPOKp(sv))
5371 delta = ptr - SvPVX_const(sv);
5373 /* Nothing to do. */
5376 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5377 if (delta > max_delta)
5378 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5379 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5380 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5381 SV_CHECK_THINKFIRST(sv);
5382 SvPOK_only_UTF8(sv);
5385 if (!SvLEN(sv)) { /* make copy of shared string */
5386 const char *pvx = SvPVX_const(sv);
5387 const STRLEN len = SvCUR(sv);
5388 SvGROW(sv, len + 1);
5389 Move(pvx,SvPVX(sv),len,char);
5395 SvOOK_offset(sv, old_delta);
5397 SvLEN_set(sv, SvLEN(sv) - delta);
5398 SvCUR_set(sv, SvCUR(sv) - delta);
5399 SvPV_set(sv, SvPVX(sv) + delta);
5401 p = (U8 *)SvPVX_const(sv);
5404 /* how many bytes were evacuated? we will fill them with sentinel
5405 bytes, except for the part holding the new offset of course. */
5408 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5410 assert(evacn <= delta + old_delta);
5414 /* This sets 'delta' to the accumulated value of all deltas so far */
5418 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5419 * the string; otherwise store a 0 byte there and store 'delta' just prior
5420 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5421 * portion of the chopped part of the string */
5422 if (delta < 0x100) {
5426 p -= sizeof(STRLEN);
5427 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5431 /* Fill the preceding buffer with sentinals to verify that no-one is
5441 =for apidoc sv_catpvn
5443 Concatenates the string onto the end of the string which is in the SV. The
5444 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5445 status set, then the bytes appended should be valid UTF-8.
5446 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5448 =for apidoc sv_catpvn_flags
5450 Concatenates the string onto the end of the string which is in the SV. The
5451 C<len> indicates number of bytes to copy.
5453 By default, the string appended is assumed to be valid UTF-8 if the SV has
5454 the UTF-8 status set, and a string of bytes otherwise. One can force the
5455 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5456 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5457 string appended will be upgraded to UTF-8 if necessary.
5459 If C<flags> has the C<SV_SMAGIC> bit set, will
5460 C<mg_set> on C<dsv> afterwards if appropriate.
5461 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5462 in terms of this function.
5468 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5471 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5473 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5474 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5476 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5477 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5478 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5481 else SvGROW(dsv, dlen + slen + 1);
5483 sstr = SvPVX_const(dsv);
5484 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5485 SvCUR_set(dsv, SvCUR(dsv) + slen);
5488 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5489 const char * const send = sstr + slen;
5492 /* Something this code does not account for, which I think is
5493 impossible; it would require the same pv to be treated as
5494 bytes *and* utf8, which would indicate a bug elsewhere. */
5495 assert(sstr != dstr);
5497 SvGROW(dsv, dlen + slen * 2 + 1);
5498 d = (U8 *)SvPVX(dsv) + dlen;
5500 while (sstr < send) {
5501 append_utf8_from_native_byte(*sstr, &d);
5504 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5507 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5509 if (flags & SV_SMAGIC)
5514 =for apidoc sv_catsv
5516 Concatenates the string from SV C<ssv> onto the end of the string in SV
5517 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5518 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5521 =for apidoc sv_catsv_flags
5523 Concatenates the string from SV C<ssv> onto the end of the string in SV
5524 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5525 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5526 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5527 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5528 and C<sv_catsv_mg> are implemented in terms of this function.
5533 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5535 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5539 const char *spv = SvPV_flags_const(ssv, slen, flags);
5540 if (flags & SV_GMAGIC)
5542 sv_catpvn_flags(dsv, spv, slen,
5543 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5544 if (flags & SV_SMAGIC)
5550 =for apidoc sv_catpv
5552 Concatenates the C<NUL>-terminated string onto the end of the string which is
5554 If the SV has the UTF-8 status set, then the bytes appended should be
5555 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5560 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5566 PERL_ARGS_ASSERT_SV_CATPV;
5570 junk = SvPV_force(sv, tlen);
5572 SvGROW(sv, tlen + len + 1);
5574 ptr = SvPVX_const(sv);
5575 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5576 SvCUR_set(sv, SvCUR(sv) + len);
5577 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5582 =for apidoc sv_catpv_flags
5584 Concatenates the C<NUL>-terminated string onto the end of the string which is
5586 If the SV has the UTF-8 status set, then the bytes appended should
5587 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5588 on the modified SV if appropriate.
5594 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5596 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5597 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5601 =for apidoc sv_catpv_mg
5603 Like C<sv_catpv>, but also handles 'set' magic.
5609 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5611 PERL_ARGS_ASSERT_SV_CATPV_MG;
5620 Creates a new SV. A non-zero C<len> parameter indicates the number of
5621 bytes of preallocated string space the SV should have. An extra byte for a
5622 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5623 space is allocated.) The reference count for the new SV is set to 1.
5625 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5626 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5627 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5628 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5629 modules supporting older perls.
5635 Perl_newSV(pTHX_ const STRLEN len)
5641 sv_grow(sv, len + 1);
5646 =for apidoc sv_magicext
5648 Adds magic to an SV, upgrading it if necessary. Applies the
5649 supplied vtable and returns a pointer to the magic added.
5651 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5652 In particular, you can add magic to SvREADONLY SVs, and add more than
5653 one instance of the same 'how'.
5655 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5656 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5657 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5658 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5660 (This is now used as a subroutine by C<sv_magic>.)
5665 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5666 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5670 PERL_ARGS_ASSERT_SV_MAGICEXT;
5672 SvUPGRADE(sv, SVt_PVMG);
5673 Newxz(mg, 1, MAGIC);
5674 mg->mg_moremagic = SvMAGIC(sv);
5675 SvMAGIC_set(sv, mg);
5677 /* Sometimes a magic contains a reference loop, where the sv and
5678 object refer to each other. To prevent a reference loop that
5679 would prevent such objects being freed, we look for such loops
5680 and if we find one we avoid incrementing the object refcount.
5682 Note we cannot do this to avoid self-tie loops as intervening RV must
5683 have its REFCNT incremented to keep it in existence.
5686 if (!obj || obj == sv ||
5687 how == PERL_MAGIC_arylen ||
5688 how == PERL_MAGIC_symtab ||
5689 (SvTYPE(obj) == SVt_PVGV &&
5690 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5691 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5692 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5697 mg->mg_obj = SvREFCNT_inc_simple(obj);
5698 mg->mg_flags |= MGf_REFCOUNTED;
5701 /* Normal self-ties simply pass a null object, and instead of
5702 using mg_obj directly, use the SvTIED_obj macro to produce a
5703 new RV as needed. For glob "self-ties", we are tieing the PVIO
5704 with an RV obj pointing to the glob containing the PVIO. In
5705 this case, to avoid a reference loop, we need to weaken the
5709 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5710 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5716 mg->mg_len = namlen;
5719 mg->mg_ptr = savepvn(name, namlen);
5720 else if (namlen == HEf_SVKEY) {
5721 /* Yes, this is casting away const. This is only for the case of
5722 HEf_SVKEY. I think we need to document this aberation of the
5723 constness of the API, rather than making name non-const, as
5724 that change propagating outwards a long way. */
5725 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5727 mg->mg_ptr = (char *) name;
5729 mg->mg_virtual = (MGVTBL *) vtable;
5736 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5738 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5739 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5740 /* This sv is only a delegate. //g magic must be attached to
5745 #ifdef PERL_OLD_COPY_ON_WRITE
5747 sv_force_normal_flags(sv, 0);
5749 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5750 &PL_vtbl_mglob, 0, 0);
5754 =for apidoc sv_magic
5756 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5757 necessary, then adds a new magic item of type C<how> to the head of the
5760 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5761 handling of the C<name> and C<namlen> arguments.
5763 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5764 to add more than one instance of the same 'how'.
5770 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5771 const char *const name, const I32 namlen)
5773 const MGVTBL *vtable;
5776 unsigned int vtable_index;
5778 PERL_ARGS_ASSERT_SV_MAGIC;
5780 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5781 || ((flags = PL_magic_data[how]),
5782 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5783 > magic_vtable_max))
5784 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5786 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5787 Useful for attaching extension internal data to perl vars.
5788 Note that multiple extensions may clash if magical scalars
5789 etc holding private data from one are passed to another. */
5791 vtable = (vtable_index == magic_vtable_max)
5792 ? NULL : PL_magic_vtables + vtable_index;
5794 #ifdef PERL_OLD_COPY_ON_WRITE
5796 sv_force_normal_flags(sv, 0);
5798 if (SvREADONLY(sv)) {
5800 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5803 Perl_croak_no_modify();
5806 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5807 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5808 /* sv_magic() refuses to add a magic of the same 'how' as an
5811 if (how == PERL_MAGIC_taint)
5817 /* Force pos to be stored as characters, not bytes. */
5818 if (SvMAGICAL(sv) && DO_UTF8(sv)
5819 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5821 && mg->mg_flags & MGf_BYTES) {
5822 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5824 mg->mg_flags &= ~MGf_BYTES;
5827 /* Rest of work is done else where */
5828 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5831 case PERL_MAGIC_taint:
5834 case PERL_MAGIC_ext:
5835 case PERL_MAGIC_dbfile:
5842 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5849 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5851 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5852 for (mg = *mgp; mg; mg = *mgp) {
5853 const MGVTBL* const virt = mg->mg_virtual;
5854 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5855 *mgp = mg->mg_moremagic;
5856 if (virt && virt->svt_free)
5857 virt->svt_free(aTHX_ sv, mg);
5858 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5860 Safefree(mg->mg_ptr);
5861 else if (mg->mg_len == HEf_SVKEY)
5862 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5863 else if (mg->mg_type == PERL_MAGIC_utf8)
5864 Safefree(mg->mg_ptr);
5866 if (mg->mg_flags & MGf_REFCOUNTED)
5867 SvREFCNT_dec(mg->mg_obj);
5871 mgp = &mg->mg_moremagic;
5874 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5875 mg_magical(sv); /* else fix the flags now */
5879 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5885 =for apidoc sv_unmagic
5887 Removes all magic of type C<type> from an SV.
5893 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5895 PERL_ARGS_ASSERT_SV_UNMAGIC;
5896 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5900 =for apidoc sv_unmagicext
5902 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5908 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5910 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5911 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5915 =for apidoc sv_rvweaken
5917 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5918 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5919 push a back-reference to this RV onto the array of backreferences
5920 associated with that magic. If the RV is magical, set magic will be
5921 called after the RV is cleared.
5927 Perl_sv_rvweaken(pTHX_ SV *const sv)
5931 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5933 if (!SvOK(sv)) /* let undefs pass */
5936 Perl_croak(aTHX_ "Can't weaken a nonreference");
5937 else if (SvWEAKREF(sv)) {
5938 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5941 else if (SvREADONLY(sv)) croak_no_modify();
5943 Perl_sv_add_backref(aTHX_ tsv, sv);
5945 SvREFCNT_dec_NN(tsv);
5950 =for apidoc sv_get_backrefs
5952 If the sv is the target of a weakrefence then return
5953 the backrefs structure associated with the sv, otherwise
5956 When returning a non-null result the type of the return
5957 is relevant. If it is an AV then the contents of the AV
5958 are the weakrefs which point at this item. If it is any
5959 other type then the item itself is the weakref.
5961 See also Perl_sv_add_backref(), Perl_sv_del_backref(),
5962 Perl_sv_kill_backrefs()
5968 Perl_sv_get_backrefs(SV *const sv)
5972 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5974 /* find slot to store array or singleton backref */
5976 if (SvTYPE(sv) == SVt_PVHV) {
5978 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5979 backrefs = (SV *)iter->xhv_backreferences;
5981 } else if (SvMAGICAL(sv)) {
5982 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5984 backrefs = mg->mg_obj;
5989 /* Give tsv backref magic if it hasn't already got it, then push a
5990 * back-reference to sv onto the array associated with the backref magic.
5992 * As an optimisation, if there's only one backref and it's not an AV,
5993 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5994 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5998 /* A discussion about the backreferences array and its refcount:
6000 * The AV holding the backreferences is pointed to either as the mg_obj of
6001 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6002 * xhv_backreferences field. The array is created with a refcount
6003 * of 2. This means that if during global destruction the array gets
6004 * picked on before its parent to have its refcount decremented by the
6005 * random zapper, it won't actually be freed, meaning it's still there for
6006 * when its parent gets freed.
6008 * When the parent SV is freed, the extra ref is killed by
6009 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6010 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6012 * When a single backref SV is stored directly, it is not reference
6017 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6023 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6025 /* find slot to store array or singleton backref */
6027 if (SvTYPE(tsv) == SVt_PVHV) {
6028 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6031 mg = mg_find(tsv, PERL_MAGIC_backref);
6033 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6034 svp = &(mg->mg_obj);
6037 /* create or retrieve the array */
6039 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6040 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6044 mg->mg_flags |= MGf_REFCOUNTED;
6047 SvREFCNT_inc_simple_void_NN(av);
6048 /* av now has a refcnt of 2; see discussion above */
6049 av_extend(av, *svp ? 2 : 1);
6051 /* move single existing backref to the array */
6052 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6057 av = MUTABLE_AV(*svp);
6059 /* optimisation: store single backref directly in HvAUX or mg_obj */
6063 assert(SvTYPE(av) == SVt_PVAV);
6064 if (AvFILLp(av) >= AvMAX(av)) {
6065 av_extend(av, AvFILLp(av)+1);
6068 /* push new backref */
6069 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6072 /* delete a back-reference to ourselves from the backref magic associated
6073 * with the SV we point to.
6077 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6081 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6083 if (SvTYPE(tsv) == SVt_PVHV) {
6085 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6087 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6088 /* It's possible for the the last (strong) reference to tsv to have
6089 become freed *before* the last thing holding a weak reference.
6090 If both survive longer than the backreferences array, then when
6091 the referent's reference count drops to 0 and it is freed, it's
6092 not able to chase the backreferences, so they aren't NULLed.
6094 For example, a CV holds a weak reference to its stash. If both the
6095 CV and the stash survive longer than the backreferences array,
6096 and the CV gets picked for the SvBREAK() treatment first,
6097 *and* it turns out that the stash is only being kept alive because
6098 of an our variable in the pad of the CV, then midway during CV
6099 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6100 It ends up pointing to the freed HV. Hence it's chased in here, and
6101 if this block wasn't here, it would hit the !svp panic just below.
6103 I don't believe that "better" destruction ordering is going to help
6104 here - during global destruction there's always going to be the
6105 chance that something goes out of order. We've tried to make it
6106 foolproof before, and it only resulted in evolutionary pressure on
6107 fools. Which made us look foolish for our hubris. :-(
6113 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6114 svp = mg ? &(mg->mg_obj) : NULL;
6118 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6120 /* It's possible that sv is being freed recursively part way through the
6121 freeing of tsv. If this happens, the backreferences array of tsv has
6122 already been freed, and so svp will be NULL. If this is the case,
6123 we should not panic. Instead, nothing needs doing, so return. */
6124 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6126 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6127 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6130 if (SvTYPE(*svp) == SVt_PVAV) {
6134 AV * const av = (AV*)*svp;
6136 assert(!SvIS_FREED(av));
6140 /* for an SV with N weak references to it, if all those
6141 * weak refs are deleted, then sv_del_backref will be called
6142 * N times and O(N^2) compares will be done within the backref
6143 * array. To ameliorate this potential slowness, we:
6144 * 1) make sure this code is as tight as possible;
6145 * 2) when looking for SV, look for it at both the head and tail of the
6146 * array first before searching the rest, since some create/destroy
6147 * patterns will cause the backrefs to be freed in order.
6154 SV **p = &svp[fill];
6155 SV *const topsv = *p;
6162 /* We weren't the last entry.
6163 An unordered list has this property that you
6164 can take the last element off the end to fill
6165 the hole, and it's still an unordered list :-)
6171 break; /* should only be one */
6178 AvFILLp(av) = fill-1;
6180 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6181 /* freed AV; skip */
6184 /* optimisation: only a single backref, stored directly */
6186 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6187 (void*)*svp, (void*)sv);
6194 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6200 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6205 /* after multiple passes through Perl_sv_clean_all() for a thingy
6206 * that has badly leaked, the backref array may have gotten freed,
6207 * since we only protect it against 1 round of cleanup */
6208 if (SvIS_FREED(av)) {
6209 if (PL_in_clean_all) /* All is fair */
6212 "panic: magic_killbackrefs (freed backref AV/SV)");
6216 is_array = (SvTYPE(av) == SVt_PVAV);
6218 assert(!SvIS_FREED(av));
6221 last = svp + AvFILLp(av);
6224 /* optimisation: only a single backref, stored directly */
6230 while (svp <= last) {
6232 SV *const referrer = *svp;
6233 if (SvWEAKREF(referrer)) {
6234 /* XXX Should we check that it hasn't changed? */
6235 assert(SvROK(referrer));
6236 SvRV_set(referrer, 0);
6238 SvWEAKREF_off(referrer);
6239 SvSETMAGIC(referrer);
6240 } else if (SvTYPE(referrer) == SVt_PVGV ||
6241 SvTYPE(referrer) == SVt_PVLV) {
6242 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6243 /* You lookin' at me? */
6244 assert(GvSTASH(referrer));
6245 assert(GvSTASH(referrer) == (const HV *)sv);
6246 GvSTASH(referrer) = 0;
6247 } else if (SvTYPE(referrer) == SVt_PVCV ||
6248 SvTYPE(referrer) == SVt_PVFM) {
6249 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6250 /* You lookin' at me? */
6251 assert(CvSTASH(referrer));
6252 assert(CvSTASH(referrer) == (const HV *)sv);
6253 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6256 assert(SvTYPE(sv) == SVt_PVGV);
6257 /* You lookin' at me? */
6258 assert(CvGV(referrer));
6259 assert(CvGV(referrer) == (const GV *)sv);
6260 anonymise_cv_maybe(MUTABLE_GV(sv),
6261 MUTABLE_CV(referrer));
6266 "panic: magic_killbackrefs (flags=%"UVxf")",
6267 (UV)SvFLAGS(referrer));
6278 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6284 =for apidoc sv_insert
6286 Inserts a string at the specified offset/length within the SV. Similar to
6287 the Perl substr() function. Handles get magic.
6289 =for apidoc sv_insert_flags
6291 Same as C<sv_insert>, but the extra C<flags> are passed to the
6292 C<SvPV_force_flags> that applies to C<bigstr>.
6298 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6304 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6307 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6310 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6311 SvPV_force_flags(bigstr, curlen, flags);
6312 (void)SvPOK_only_UTF8(bigstr);
6313 if (offset + len > curlen) {
6314 SvGROW(bigstr, offset+len+1);
6315 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6316 SvCUR_set(bigstr, offset+len);
6320 i = littlelen - len;
6321 if (i > 0) { /* string might grow */
6322 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6323 mid = big + offset + len;
6324 midend = bigend = big + SvCUR(bigstr);
6327 while (midend > mid) /* shove everything down */
6328 *--bigend = *--midend;
6329 Move(little,big+offset,littlelen,char);
6330 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6335 Move(little,SvPVX(bigstr)+offset,len,char);
6340 big = SvPVX(bigstr);
6343 bigend = big + SvCUR(bigstr);
6345 if (midend > bigend)
6346 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6349 if (mid - big > bigend - midend) { /* faster to shorten from end */
6351 Move(little, mid, littlelen,char);
6354 i = bigend - midend;
6356 Move(midend, mid, i,char);
6360 SvCUR_set(bigstr, mid - big);
6362 else if ((i = mid - big)) { /* faster from front */
6363 midend -= littlelen;
6365 Move(big, midend - i, i, char);
6366 sv_chop(bigstr,midend-i);
6368 Move(little, mid, littlelen,char);
6370 else if (littlelen) {
6371 midend -= littlelen;
6372 sv_chop(bigstr,midend);
6373 Move(little,midend,littlelen,char);
6376 sv_chop(bigstr,midend);
6382 =for apidoc sv_replace
6384 Make the first argument a copy of the second, then delete the original.
6385 The target SV physically takes over ownership of the body of the source SV
6386 and inherits its flags; however, the target keeps any magic it owns,
6387 and any magic in the source is discarded.
6388 Note that this is a rather specialist SV copying operation; most of the
6389 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6395 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6397 const U32 refcnt = SvREFCNT(sv);
6399 PERL_ARGS_ASSERT_SV_REPLACE;
6401 SV_CHECK_THINKFIRST_COW_DROP(sv);
6402 if (SvREFCNT(nsv) != 1) {
6403 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6404 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6406 if (SvMAGICAL(sv)) {
6410 sv_upgrade(nsv, SVt_PVMG);
6411 SvMAGIC_set(nsv, SvMAGIC(sv));
6412 SvFLAGS(nsv) |= SvMAGICAL(sv);
6414 SvMAGIC_set(sv, NULL);
6418 assert(!SvREFCNT(sv));
6419 #ifdef DEBUG_LEAKING_SCALARS
6420 sv->sv_flags = nsv->sv_flags;
6421 sv->sv_any = nsv->sv_any;
6422 sv->sv_refcnt = nsv->sv_refcnt;
6423 sv->sv_u = nsv->sv_u;
6425 StructCopy(nsv,sv,SV);
6427 if(SvTYPE(sv) == SVt_IV) {
6428 SET_SVANY_FOR_BODYLESS_IV(sv);
6432 #ifdef PERL_OLD_COPY_ON_WRITE
6433 if (SvIsCOW_normal(nsv)) {
6434 /* We need to follow the pointers around the loop to make the
6435 previous SV point to sv, rather than nsv. */
6438 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6441 assert(SvPVX_const(current) == SvPVX_const(nsv));
6443 /* Make the SV before us point to the SV after us. */
6445 PerlIO_printf(Perl_debug_log, "previous is\n");
6447 PerlIO_printf(Perl_debug_log,
6448 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6449 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6451 SV_COW_NEXT_SV_SET(current, sv);
6454 SvREFCNT(sv) = refcnt;
6455 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6460 /* We're about to free a GV which has a CV that refers back to us.
6461 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6465 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6470 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6473 assert(SvREFCNT(gv) == 0);
6474 assert(isGV(gv) && isGV_with_GP(gv));
6476 assert(!CvANON(cv));
6477 assert(CvGV(cv) == gv);
6478 assert(!CvNAMED(cv));
6480 /* will the CV shortly be freed by gp_free() ? */
6481 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6482 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6486 /* if not, anonymise: */
6487 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6488 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6489 : newSVpvn_flags( "__ANON__", 8, 0 );
6490 sv_catpvs(gvname, "::__ANON__");
6491 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6492 SvREFCNT_dec_NN(gvname);
6496 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6501 =for apidoc sv_clear
6503 Clear an SV: call any destructors, free up any memory used by the body,
6504 and free the body itself. The SV's head is I<not> freed, although
6505 its type is set to all 1's so that it won't inadvertently be assumed
6506 to be live during global destruction etc.
6507 This function should only be called when REFCNT is zero. Most of the time
6508 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6515 Perl_sv_clear(pTHX_ SV *const orig_sv)
6520 const struct body_details *sv_type_details;
6524 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6525 Not strictly necessary */
6527 PERL_ARGS_ASSERT_SV_CLEAR;
6529 /* within this loop, sv is the SV currently being freed, and
6530 * iter_sv is the most recent AV or whatever that's being iterated
6531 * over to provide more SVs */
6537 assert(SvREFCNT(sv) == 0);
6538 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6540 if (type <= SVt_IV) {
6541 /* See the comment in sv.h about the collusion between this
6542 * early return and the overloading of the NULL slots in the
6546 SvFLAGS(sv) &= SVf_BREAK;
6547 SvFLAGS(sv) |= SVTYPEMASK;
6551 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6552 for another purpose */
6553 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6555 if (type >= SVt_PVMG) {
6557 if (!curse(sv, 1)) goto get_next_sv;
6558 type = SvTYPE(sv); /* destructor may have changed it */
6560 /* Free back-references before magic, in case the magic calls
6561 * Perl code that has weak references to sv. */
6562 if (type == SVt_PVHV) {
6563 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6567 else if (SvMAGIC(sv)) {
6568 /* Free back-references before other types of magic. */
6569 sv_unmagic(sv, PERL_MAGIC_backref);
6575 /* case SVt_INVLIST: */
6578 IoIFP(sv) != PerlIO_stdin() &&
6579 IoIFP(sv) != PerlIO_stdout() &&
6580 IoIFP(sv) != PerlIO_stderr() &&
6581 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6583 io_close(MUTABLE_IO(sv), NULL, FALSE,
6584 (IoTYPE(sv) == IoTYPE_WRONLY ||
6585 IoTYPE(sv) == IoTYPE_RDWR ||
6586 IoTYPE(sv) == IoTYPE_APPEND));
6588 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6589 PerlDir_close(IoDIRP(sv));
6590 IoDIRP(sv) = (DIR*)NULL;
6591 Safefree(IoTOP_NAME(sv));
6592 Safefree(IoFMT_NAME(sv));
6593 Safefree(IoBOTTOM_NAME(sv));
6594 if ((const GV *)sv == PL_statgv)
6598 /* FIXME for plugins */
6600 pregfree2((REGEXP*) sv);
6604 cv_undef(MUTABLE_CV(sv));
6605 /* If we're in a stash, we don't own a reference to it.
6606 * However it does have a back reference to us, which needs to
6608 if ((stash = CvSTASH(sv)))
6609 sv_del_backref(MUTABLE_SV(stash), sv);
6612 if (PL_last_swash_hv == (const HV *)sv) {
6613 PL_last_swash_hv = NULL;
6615 if (HvTOTALKEYS((HV*)sv) > 0) {
6617 /* this statement should match the one at the beginning of
6618 * hv_undef_flags() */
6619 if ( PL_phase != PERL_PHASE_DESTRUCT
6620 && (hek = HvNAME_HEK((HV*)sv)))
6622 if (PL_stashcache) {
6623 DEBUG_o(Perl_deb(aTHX_
6624 "sv_clear clearing PL_stashcache for '%"HEKf
6627 (void)hv_deletehek(PL_stashcache,
6630 hv_name_set((HV*)sv, NULL, 0, 0);
6633 /* save old iter_sv in unused SvSTASH field */
6634 assert(!SvOBJECT(sv));
6635 SvSTASH(sv) = (HV*)iter_sv;
6638 /* save old hash_index in unused SvMAGIC field */
6639 assert(!SvMAGICAL(sv));
6640 assert(!SvMAGIC(sv));
6641 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6644 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6645 goto get_next_sv; /* process this new sv */
6647 /* free empty hash */
6648 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6649 assert(!HvARRAY((HV*)sv));
6653 AV* av = MUTABLE_AV(sv);
6654 if (PL_comppad == av) {
6658 if (AvREAL(av) && AvFILLp(av) > -1) {
6659 next_sv = AvARRAY(av)[AvFILLp(av)--];
6660 /* save old iter_sv in top-most slot of AV,
6661 * and pray that it doesn't get wiped in the meantime */
6662 AvARRAY(av)[AvMAX(av)] = iter_sv;
6664 goto get_next_sv; /* process this new sv */
6666 Safefree(AvALLOC(av));
6671 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6672 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6673 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6674 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6676 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6677 SvREFCNT_dec(LvTARG(sv));
6678 if (isREGEXP(sv)) goto freeregexp;
6681 if (isGV_with_GP(sv)) {
6682 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6683 && HvENAME_get(stash))
6684 mro_method_changed_in(stash);
6685 gp_free(MUTABLE_GV(sv));
6687 unshare_hek(GvNAME_HEK(sv));
6688 /* If we're in a stash, we don't own a reference to it.
6689 * However it does have a back reference to us, which
6690 * needs to be cleared. */
6691 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6692 sv_del_backref(MUTABLE_SV(stash), sv);
6694 /* FIXME. There are probably more unreferenced pointers to SVs
6695 * in the interpreter struct that we should check and tidy in
6696 * a similar fashion to this: */
6697 /* See also S_sv_unglob, which does the same thing. */
6698 if ((const GV *)sv == PL_last_in_gv)
6699 PL_last_in_gv = NULL;
6700 else if ((const GV *)sv == PL_statgv)
6702 else if ((const GV *)sv == PL_stderrgv)
6711 /* Don't bother with SvOOK_off(sv); as we're only going to
6715 SvOOK_offset(sv, offset);
6716 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6717 /* Don't even bother with turning off the OOK flag. */
6722 SV * const target = SvRV(sv);
6724 sv_del_backref(target, sv);
6730 else if (SvPVX_const(sv)
6731 && !(SvTYPE(sv) == SVt_PVIO
6732 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6736 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6740 # ifdef PERL_OLD_COPY_ON_WRITE
6741 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6743 if (CowREFCNT(sv)) {
6751 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6755 # ifdef PERL_OLD_COPY_ON_WRITE
6759 Safefree(SvPVX_mutable(sv));
6763 else if (SvPVX_const(sv) && SvLEN(sv)
6764 && !(SvTYPE(sv) == SVt_PVIO
6765 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6766 Safefree(SvPVX_mutable(sv));
6767 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6768 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6778 SvFLAGS(sv) &= SVf_BREAK;
6779 SvFLAGS(sv) |= SVTYPEMASK;
6781 sv_type_details = bodies_by_type + type;
6782 if (sv_type_details->arena) {
6783 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6784 &PL_body_roots[type]);
6786 else if (sv_type_details->body_size) {
6787 safefree(SvANY(sv));
6791 /* caller is responsible for freeing the head of the original sv */
6792 if (sv != orig_sv && !SvREFCNT(sv))
6795 /* grab and free next sv, if any */
6803 else if (!iter_sv) {
6805 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6806 AV *const av = (AV*)iter_sv;
6807 if (AvFILLp(av) > -1) {
6808 sv = AvARRAY(av)[AvFILLp(av)--];
6810 else { /* no more elements of current AV to free */
6813 /* restore previous value, squirrelled away */
6814 iter_sv = AvARRAY(av)[AvMAX(av)];
6815 Safefree(AvALLOC(av));
6818 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6819 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6820 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6821 /* no more elements of current HV to free */
6824 /* Restore previous values of iter_sv and hash_index,
6825 * squirrelled away */
6826 assert(!SvOBJECT(sv));
6827 iter_sv = (SV*)SvSTASH(sv);
6828 assert(!SvMAGICAL(sv));
6829 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6831 /* perl -DA does not like rubbish in SvMAGIC. */
6835 /* free any remaining detritus from the hash struct */
6836 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6837 assert(!HvARRAY((HV*)sv));
6842 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6846 if (!SvREFCNT(sv)) {
6850 if (--(SvREFCNT(sv)))
6854 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6855 "Attempt to free temp prematurely: SV 0x%"UVxf
6856 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6860 if (SvIMMORTAL(sv)) {
6861 /* make sure SvREFCNT(sv)==0 happens very seldom */
6862 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6871 /* This routine curses the sv itself, not the object referenced by sv. So
6872 sv does not have to be ROK. */
6875 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6876 PERL_ARGS_ASSERT_CURSE;
6877 assert(SvOBJECT(sv));
6879 if (PL_defstash && /* Still have a symbol table? */
6885 stash = SvSTASH(sv);
6886 assert(SvTYPE(stash) == SVt_PVHV);
6887 if (HvNAME(stash)) {
6888 CV* destructor = NULL;
6889 assert (SvOOK(stash));
6890 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6891 if (!destructor || HvMROMETA(stash)->destroy_gen
6892 != PL_sub_generation)
6895 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6896 if (gv) destructor = GvCV(gv);
6897 if (!SvOBJECT(stash))
6900 destructor ? (HV *)destructor : ((HV *)0)+1;
6901 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6905 assert(!destructor || destructor == ((CV *)0)+1
6906 || SvTYPE(destructor) == SVt_PVCV);
6907 if (destructor && destructor != ((CV *)0)+1
6908 /* A constant subroutine can have no side effects, so
6909 don't bother calling it. */
6910 && !CvCONST(destructor)
6911 /* Don't bother calling an empty destructor or one that
6912 returns immediately. */
6913 && (CvISXSUB(destructor)
6914 || (CvSTART(destructor)
6915 && (CvSTART(destructor)->op_next->op_type
6917 && (CvSTART(destructor)->op_next->op_type
6919 || CvSTART(destructor)->op_next->op_next->op_type
6925 SV* const tmpref = newRV(sv);
6926 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6928 PUSHSTACKi(PERLSI_DESTROY);
6933 call_sv(MUTABLE_SV(destructor),
6934 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6938 if(SvREFCNT(tmpref) < 2) {
6939 /* tmpref is not kept alive! */
6941 SvRV_set(tmpref, NULL);
6944 SvREFCNT_dec_NN(tmpref);
6947 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6950 if (check_refcnt && SvREFCNT(sv)) {
6951 if (PL_in_clean_objs)
6953 "DESTROY created new reference to dead object '%"HEKf"'",
6954 HEKfARG(HvNAME_HEK(stash)));
6955 /* DESTROY gave object new lease on life */
6961 HV * const stash = SvSTASH(sv);
6962 /* Curse before freeing the stash, as freeing the stash could cause
6963 a recursive call into S_curse. */
6964 SvOBJECT_off(sv); /* Curse the object. */
6965 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6966 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6972 =for apidoc sv_newref
6974 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6981 Perl_sv_newref(pTHX_ SV *const sv)
6983 PERL_UNUSED_CONTEXT;
6992 Decrement an SV's reference count, and if it drops to zero, call
6993 C<sv_clear> to invoke destructors and free up any memory used by
6994 the body; finally, deallocate the SV's head itself.
6995 Normally called via a wrapper macro C<SvREFCNT_dec>.
7001 Perl_sv_free(pTHX_ SV *const sv)
7007 /* Private helper function for SvREFCNT_dec().
7008 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7011 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7015 PERL_ARGS_ASSERT_SV_FREE2;
7017 if (LIKELY( rc == 1 )) {
7023 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7024 "Attempt to free temp prematurely: SV 0x%"UVxf
7025 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7029 if (SvIMMORTAL(sv)) {
7030 /* make sure SvREFCNT(sv)==0 happens very seldom */
7031 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7035 if (! SvREFCNT(sv)) /* may have have been resurrected */
7040 /* handle exceptional cases */
7044 if (SvFLAGS(sv) & SVf_BREAK)
7045 /* this SV's refcnt has been artificially decremented to
7046 * trigger cleanup */
7048 if (PL_in_clean_all) /* All is fair */
7050 if (SvIMMORTAL(sv)) {
7051 /* make sure SvREFCNT(sv)==0 happens very seldom */
7052 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7055 if (ckWARN_d(WARN_INTERNAL)) {
7056 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7057 Perl_dump_sv_child(aTHX_ sv);
7059 #ifdef DEBUG_LEAKING_SCALARS
7062 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7063 if (PL_warnhook == PERL_WARNHOOK_FATAL
7064 || ckDEAD(packWARN(WARN_INTERNAL))) {
7065 /* Don't let Perl_warner cause us to escape our fate: */
7069 /* This may not return: */
7070 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7071 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7072 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7075 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7085 Returns the length of the string in the SV. Handles magic and type
7086 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
7087 gives raw access to the xpv_cur slot.
7093 Perl_sv_len(pTHX_ SV *const sv)
7100 (void)SvPV_const(sv, len);
7105 =for apidoc sv_len_utf8
7107 Returns the number of characters in the string in an SV, counting wide
7108 UTF-8 bytes as a single character. Handles magic and type coercion.
7114 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7115 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7116 * (Note that the mg_len is not the length of the mg_ptr field.
7117 * This allows the cache to store the character length of the string without
7118 * needing to malloc() extra storage to attach to the mg_ptr.)
7123 Perl_sv_len_utf8(pTHX_ SV *const sv)
7129 return sv_len_utf8_nomg(sv);
7133 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7136 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7138 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7140 if (PL_utf8cache && SvUTF8(sv)) {
7142 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7144 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7145 if (mg->mg_len != -1)
7148 /* We can use the offset cache for a headstart.
7149 The longer value is stored in the first pair. */
7150 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7152 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7156 if (PL_utf8cache < 0) {
7157 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7158 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7162 ulen = Perl_utf8_length(aTHX_ s, s + len);
7163 utf8_mg_len_cache_update(sv, &mg, ulen);
7167 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7170 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7173 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7174 STRLEN *const uoffset_p, bool *const at_end)
7176 const U8 *s = start;
7177 STRLEN uoffset = *uoffset_p;
7179 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7181 while (s < send && uoffset) {
7188 else if (s > send) {
7190 /* This is the existing behaviour. Possibly it should be a croak, as
7191 it's actually a bounds error */
7194 *uoffset_p -= uoffset;
7198 /* Given the length of the string in both bytes and UTF-8 characters, decide
7199 whether to walk forwards or backwards to find the byte corresponding to
7200 the passed in UTF-8 offset. */
7202 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7203 STRLEN uoffset, const STRLEN uend)
7205 STRLEN backw = uend - uoffset;
7207 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7209 if (uoffset < 2 * backw) {
7210 /* The assumption is that going forwards is twice the speed of going
7211 forward (that's where the 2 * backw comes from).
7212 (The real figure of course depends on the UTF-8 data.) */
7213 const U8 *s = start;
7215 while (s < send && uoffset--)
7225 while (UTF8_IS_CONTINUATION(*send))
7228 return send - start;
7231 /* For the string representation of the given scalar, find the byte
7232 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7233 give another position in the string, *before* the sought offset, which
7234 (which is always true, as 0, 0 is a valid pair of positions), which should
7235 help reduce the amount of linear searching.
7236 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7237 will be used to reduce the amount of linear searching. The cache will be
7238 created if necessary, and the found value offered to it for update. */
7240 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7241 const U8 *const send, STRLEN uoffset,
7242 STRLEN uoffset0, STRLEN boffset0)
7244 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7246 bool at_end = FALSE;
7248 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7250 assert (uoffset >= uoffset0);
7255 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7257 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7258 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7259 if ((*mgp)->mg_ptr) {
7260 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7261 if (cache[0] == uoffset) {
7262 /* An exact match. */
7265 if (cache[2] == uoffset) {
7266 /* An exact match. */
7270 if (cache[0] < uoffset) {
7271 /* The cache already knows part of the way. */
7272 if (cache[0] > uoffset0) {
7273 /* The cache knows more than the passed in pair */
7274 uoffset0 = cache[0];
7275 boffset0 = cache[1];
7277 if ((*mgp)->mg_len != -1) {
7278 /* And we know the end too. */
7280 + sv_pos_u2b_midway(start + boffset0, send,
7282 (*mgp)->mg_len - uoffset0);
7284 uoffset -= uoffset0;
7286 + sv_pos_u2b_forwards(start + boffset0,
7287 send, &uoffset, &at_end);
7288 uoffset += uoffset0;
7291 else if (cache[2] < uoffset) {
7292 /* We're between the two cache entries. */
7293 if (cache[2] > uoffset0) {
7294 /* and the cache knows more than the passed in pair */
7295 uoffset0 = cache[2];
7296 boffset0 = cache[3];
7300 + sv_pos_u2b_midway(start + boffset0,
7303 cache[0] - uoffset0);
7306 + sv_pos_u2b_midway(start + boffset0,
7309 cache[2] - uoffset0);
7313 else if ((*mgp)->mg_len != -1) {
7314 /* If we can take advantage of a passed in offset, do so. */
7315 /* In fact, offset0 is either 0, or less than offset, so don't
7316 need to worry about the other possibility. */
7318 + sv_pos_u2b_midway(start + boffset0, send,
7320 (*mgp)->mg_len - uoffset0);
7325 if (!found || PL_utf8cache < 0) {
7326 STRLEN real_boffset;
7327 uoffset -= uoffset0;
7328 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7329 send, &uoffset, &at_end);
7330 uoffset += uoffset0;
7332 if (found && PL_utf8cache < 0)
7333 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7335 boffset = real_boffset;
7338 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7340 utf8_mg_len_cache_update(sv, mgp, uoffset);
7342 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7349 =for apidoc sv_pos_u2b_flags
7351 Converts the offset from a count of UTF-8 chars from
7352 the start of the string, to a count of the equivalent number of bytes; if
7353 lenp is non-zero, it does the same to lenp, but this time starting from
7354 the offset, rather than from the start
7355 of the string. Handles type coercion.
7356 I<flags> is passed to C<SvPV_flags>, and usually should be
7357 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7363 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7364 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7365 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7370 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7377 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7379 start = (U8*)SvPV_flags(sv, len, flags);
7381 const U8 * const send = start + len;
7383 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7386 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7387 is 0, and *lenp is already set to that. */) {
7388 /* Convert the relative offset to absolute. */
7389 const STRLEN uoffset2 = uoffset + *lenp;
7390 const STRLEN boffset2
7391 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7392 uoffset, boffset) - boffset;
7406 =for apidoc sv_pos_u2b
7408 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7409 the start of the string, to a count of the equivalent number of bytes; if
7410 lenp is non-zero, it does the same to lenp, but this time starting from
7411 the offset, rather than from the start of the string. Handles magic and
7414 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7421 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7422 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7423 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7427 /* This function is subject to size and sign problems */
7430 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7432 PERL_ARGS_ASSERT_SV_POS_U2B;
7435 STRLEN ulen = (STRLEN)*lenp;
7436 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7437 SV_GMAGIC|SV_CONST_RETURN);
7440 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7441 SV_GMAGIC|SV_CONST_RETURN);
7446 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7449 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7450 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7453 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7454 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7455 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7459 (*mgp)->mg_len = ulen;
7462 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7463 byte length pairing. The (byte) length of the total SV is passed in too,
7464 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7465 may not have updated SvCUR, so we can't rely on reading it directly.
7467 The proffered utf8/byte length pairing isn't used if the cache already has
7468 two pairs, and swapping either for the proffered pair would increase the
7469 RMS of the intervals between known byte offsets.
7471 The cache itself consists of 4 STRLEN values
7472 0: larger UTF-8 offset
7473 1: corresponding byte offset
7474 2: smaller UTF-8 offset
7475 3: corresponding byte offset
7477 Unused cache pairs have the value 0, 0.
7478 Keeping the cache "backwards" means that the invariant of
7479 cache[0] >= cache[2] is maintained even with empty slots, which means that
7480 the code that uses it doesn't need to worry if only 1 entry has actually
7481 been set to non-zero. It also makes the "position beyond the end of the
7482 cache" logic much simpler, as the first slot is always the one to start
7486 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7487 const STRLEN utf8, const STRLEN blen)
7491 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7496 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7497 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7498 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7500 (*mgp)->mg_len = -1;
7504 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7505 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7506 (*mgp)->mg_ptr = (char *) cache;
7510 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7511 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7512 a pointer. Note that we no longer cache utf8 offsets on refer-
7513 ences, but this check is still a good idea, for robustness. */
7514 const U8 *start = (const U8 *) SvPVX_const(sv);
7515 const STRLEN realutf8 = utf8_length(start, start + byte);
7517 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7521 /* Cache is held with the later position first, to simplify the code
7522 that deals with unbounded ends. */
7524 ASSERT_UTF8_CACHE(cache);
7525 if (cache[1] == 0) {
7526 /* Cache is totally empty */
7529 } else if (cache[3] == 0) {
7530 if (byte > cache[1]) {
7531 /* New one is larger, so goes first. */
7532 cache[2] = cache[0];
7533 cache[3] = cache[1];
7541 /* float casts necessary? XXX */
7542 #define THREEWAY_SQUARE(a,b,c,d) \
7543 ((float)((d) - (c))) * ((float)((d) - (c))) \
7544 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7545 + ((float)((b) - (a))) * ((float)((b) - (a)))
7547 /* Cache has 2 slots in use, and we know three potential pairs.
7548 Keep the two that give the lowest RMS distance. Do the
7549 calculation in bytes simply because we always know the byte
7550 length. squareroot has the same ordering as the positive value,
7551 so don't bother with the actual square root. */
7552 if (byte > cache[1]) {
7553 /* New position is after the existing pair of pairs. */
7554 const float keep_earlier
7555 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7556 const float keep_later
7557 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7559 if (keep_later < keep_earlier) {
7560 cache[2] = cache[0];
7561 cache[3] = cache[1];
7567 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7568 float b, c, keep_earlier;
7569 if (byte > cache[3]) {
7570 /* New position is between the existing pair of pairs. */
7571 b = (float)cache[3];
7574 /* New position is before the existing pair of pairs. */
7576 c = (float)cache[3];
7578 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7579 if (byte > cache[3]) {
7580 if (keep_later < keep_earlier) {
7590 if (! (keep_later < keep_earlier)) {
7591 cache[0] = cache[2];
7592 cache[1] = cache[3];
7599 ASSERT_UTF8_CACHE(cache);
7602 /* We already know all of the way, now we may be able to walk back. The same
7603 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7604 backward is half the speed of walking forward. */
7606 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7607 const U8 *end, STRLEN endu)
7609 const STRLEN forw = target - s;
7610 STRLEN backw = end - target;
7612 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7614 if (forw < 2 * backw) {
7615 return utf8_length(s, target);
7618 while (end > target) {
7620 while (UTF8_IS_CONTINUATION(*end)) {
7629 =for apidoc sv_pos_b2u_flags
7631 Converts the offset from a count of bytes from the start of the string, to
7632 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7633 I<flags> is passed to C<SvPV_flags>, and usually should be
7634 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7640 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7641 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7646 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7649 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7655 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7657 s = (const U8*)SvPV_flags(sv, blen, flags);
7660 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7661 ", byte=%"UVuf, (UV)blen, (UV)offset);
7667 && SvTYPE(sv) >= SVt_PVMG
7668 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7671 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7672 if (cache[1] == offset) {
7673 /* An exact match. */
7676 if (cache[3] == offset) {
7677 /* An exact match. */
7681 if (cache[1] < offset) {
7682 /* We already know part of the way. */
7683 if (mg->mg_len != -1) {
7684 /* Actually, we know the end too. */
7686 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7687 s + blen, mg->mg_len - cache[0]);
7689 len = cache[0] + utf8_length(s + cache[1], send);
7692 else if (cache[3] < offset) {
7693 /* We're between the two cached pairs, so we do the calculation
7694 offset by the byte/utf-8 positions for the earlier pair,
7695 then add the utf-8 characters from the string start to
7697 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7698 s + cache[1], cache[0] - cache[2])
7702 else { /* cache[3] > offset */
7703 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7707 ASSERT_UTF8_CACHE(cache);
7709 } else if (mg->mg_len != -1) {
7710 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7714 if (!found || PL_utf8cache < 0) {
7715 const STRLEN real_len = utf8_length(s, send);
7717 if (found && PL_utf8cache < 0)
7718 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7724 utf8_mg_len_cache_update(sv, &mg, len);
7726 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7733 =for apidoc sv_pos_b2u
7735 Converts the value pointed to by offsetp from a count of bytes from the
7736 start of the string, to a count of the equivalent number of UTF-8 chars.
7737 Handles magic and type coercion.
7739 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7746 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7747 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7752 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7754 PERL_ARGS_ASSERT_SV_POS_B2U;
7759 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7760 SV_GMAGIC|SV_CONST_RETURN);
7764 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7765 STRLEN real, SV *const sv)
7767 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7769 /* As this is debugging only code, save space by keeping this test here,
7770 rather than inlining it in all the callers. */
7771 if (from_cache == real)
7774 /* Need to turn the assertions off otherwise we may recurse infinitely
7775 while printing error messages. */
7776 SAVEI8(PL_utf8cache);
7778 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7779 func, (UV) from_cache, (UV) real, SVfARG(sv));
7785 Returns a boolean indicating whether the strings in the two SVs are
7786 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7787 coerce its args to strings if necessary.
7789 =for apidoc sv_eq_flags
7791 Returns a boolean indicating whether the strings in the two SVs are
7792 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7793 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7799 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7806 SV* svrecode = NULL;
7813 /* if pv1 and pv2 are the same, second SvPV_const call may
7814 * invalidate pv1 (if we are handling magic), so we may need to
7816 if (sv1 == sv2 && flags & SV_GMAGIC
7817 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7818 pv1 = SvPV_const(sv1, cur1);
7819 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7821 pv1 = SvPV_flags_const(sv1, cur1, flags);
7829 pv2 = SvPV_flags_const(sv2, cur2, flags);
7831 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7832 /* Differing utf8ness.
7833 * Do not UTF8size the comparands as a side-effect. */
7836 svrecode = newSVpvn(pv2, cur2);
7837 sv_recode_to_utf8(svrecode, _get_encoding());
7838 pv2 = SvPV_const(svrecode, cur2);
7841 svrecode = newSVpvn(pv1, cur1);
7842 sv_recode_to_utf8(svrecode, _get_encoding());
7843 pv1 = SvPV_const(svrecode, cur1);
7845 /* Now both are in UTF-8. */
7847 SvREFCNT_dec_NN(svrecode);
7853 /* sv1 is the UTF-8 one */
7854 return bytes_cmp_utf8((const U8*)pv2, cur2,
7855 (const U8*)pv1, cur1) == 0;
7858 /* sv2 is the UTF-8 one */
7859 return bytes_cmp_utf8((const U8*)pv1, cur1,
7860 (const U8*)pv2, cur2) == 0;
7866 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7868 SvREFCNT_dec(svrecode);
7876 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7877 string in C<sv1> is less than, equal to, or greater than the string in
7878 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7879 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7881 =for apidoc sv_cmp_flags
7883 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7884 string in C<sv1> is less than, equal to, or greater than the string in
7885 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7886 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7887 also C<sv_cmp_locale_flags>.
7893 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7895 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7899 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7903 const char *pv1, *pv2;
7905 SV *svrecode = NULL;
7912 pv1 = SvPV_flags_const(sv1, cur1, flags);
7919 pv2 = SvPV_flags_const(sv2, cur2, flags);
7921 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7922 /* Differing utf8ness.
7923 * Do not UTF8size the comparands as a side-effect. */
7926 svrecode = newSVpvn(pv2, cur2);
7927 sv_recode_to_utf8(svrecode, _get_encoding());
7928 pv2 = SvPV_const(svrecode, cur2);
7931 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7932 (const U8*)pv1, cur1);
7933 return retval ? retval < 0 ? -1 : +1 : 0;
7938 svrecode = newSVpvn(pv1, cur1);
7939 sv_recode_to_utf8(svrecode, _get_encoding());
7940 pv1 = SvPV_const(svrecode, cur1);
7943 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7944 (const U8*)pv2, cur2);
7945 return retval ? retval < 0 ? -1 : +1 : 0;
7951 cmp = cur2 ? -1 : 0;
7955 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7958 cmp = retval < 0 ? -1 : 1;
7959 } else if (cur1 == cur2) {
7962 cmp = cur1 < cur2 ? -1 : 1;
7966 SvREFCNT_dec(svrecode);
7972 =for apidoc sv_cmp_locale
7974 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7975 'use bytes' aware, handles get magic, and will coerce its args to strings
7976 if necessary. See also C<sv_cmp>.
7978 =for apidoc sv_cmp_locale_flags
7980 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7981 'use bytes' aware and will coerce its args to strings if necessary. If the
7982 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7988 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7990 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7994 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7997 #ifdef USE_LOCALE_COLLATE
8003 if (PL_collation_standard)
8007 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8009 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8011 if (!pv1 || !len1) {
8022 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8025 return retval < 0 ? -1 : 1;
8028 * When the result of collation is equality, that doesn't mean
8029 * that there are no differences -- some locales exclude some
8030 * characters from consideration. So to avoid false equalities,
8031 * we use the raw string as a tiebreaker.
8038 PERL_UNUSED_ARG(flags);
8039 #endif /* USE_LOCALE_COLLATE */
8041 return sv_cmp(sv1, sv2);
8045 #ifdef USE_LOCALE_COLLATE
8048 =for apidoc sv_collxfrm
8050 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8051 C<sv_collxfrm_flags>.
8053 =for apidoc sv_collxfrm_flags
8055 Add Collate Transform magic to an SV if it doesn't already have it. If the
8056 flags contain SV_GMAGIC, it handles get-magic.
8058 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
8059 scalar data of the variable, but transformed to such a format that a normal
8060 memory comparison can be used to compare the data according to the locale
8067 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8071 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8073 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8074 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8080 Safefree(mg->mg_ptr);
8081 s = SvPV_flags_const(sv, len, flags);
8082 if ((xf = mem_collxfrm(s, len, &xlen))) {
8084 #ifdef PERL_OLD_COPY_ON_WRITE
8086 sv_force_normal_flags(sv, 0);
8088 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8102 if (mg && mg->mg_ptr) {
8104 return mg->mg_ptr + sizeof(PL_collation_ix);
8112 #endif /* USE_LOCALE_COLLATE */
8115 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8117 SV * const tsv = newSV(0);
8120 sv_gets(tsv, fp, 0);
8121 sv_utf8_upgrade_nomg(tsv);
8122 SvCUR_set(sv,append);
8125 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8129 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8132 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8133 /* Grab the size of the record we're getting */
8134 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8141 /* With a true, record-oriented file on VMS, we need to use read directly
8142 * to ensure that we respect RMS record boundaries. The user is responsible
8143 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8144 * record size) field. N.B. This is likely to produce invalid results on
8145 * varying-width character data when a record ends mid-character.
8147 fd = PerlIO_fileno(fp);
8149 && PerlLIO_fstat(fd, &st) == 0
8150 && (st.st_fab_rfm == FAB$C_VAR
8151 || st.st_fab_rfm == FAB$C_VFC
8152 || st.st_fab_rfm == FAB$C_FIX)) {
8154 bytesread = PerlLIO_read(fd, buffer, recsize);
8156 else /* in-memory file from PerlIO::Scalar
8157 * or not a record-oriented file
8161 bytesread = PerlIO_read(fp, buffer, recsize);
8163 /* At this point, the logic in sv_get() means that sv will
8164 be treated as utf-8 if the handle is utf8.
8166 if (PerlIO_isutf8(fp) && bytesread > 0) {
8167 char *bend = buffer + bytesread;
8168 char *bufp = buffer;
8169 size_t charcount = 0;
8170 bool charstart = TRUE;
8173 while (charcount < recsize) {
8174 /* count accumulated characters */
8175 while (bufp < bend) {
8177 skip = UTF8SKIP(bufp);
8179 if (bufp + skip > bend) {
8180 /* partial at the end */
8191 if (charcount < recsize) {
8193 STRLEN bufp_offset = bufp - buffer;
8194 SSize_t morebytesread;
8196 /* originally I read enough to fill any incomplete
8197 character and the first byte of the next
8198 character if needed, but if there's many
8199 multi-byte encoded characters we're going to be
8200 making a read call for every character beyond
8201 the original read size.
8203 So instead, read the rest of the character if
8204 any, and enough bytes to match at least the
8205 start bytes for each character we're going to
8209 readsize = recsize - charcount;
8211 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8212 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8213 bend = buffer + bytesread;
8214 morebytesread = PerlIO_read(fp, bend, readsize);
8215 if (morebytesread <= 0) {
8216 /* we're done, if we still have incomplete
8217 characters the check code in sv_gets() will
8220 I'd originally considered doing
8221 PerlIO_ungetc() on all but the lead
8222 character of the incomplete character, but
8223 read() doesn't do that, so I don't.
8228 /* prepare to scan some more */
8229 bytesread += morebytesread;
8230 bend = buffer + bytesread;
8231 bufp = buffer + bufp_offset;
8239 SvCUR_set(sv, bytesread + append);
8240 buffer[bytesread] = '\0';
8241 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8247 Get a line from the filehandle and store it into the SV, optionally
8248 appending to the currently-stored string. If C<append> is not 0, the
8249 line is appended to the SV instead of overwriting it. C<append> should
8250 be set to the byte offset that the appended string should start at
8251 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8257 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8267 PERL_ARGS_ASSERT_SV_GETS;
8269 if (SvTHINKFIRST(sv))
8270 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8271 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8273 However, perlbench says it's slower, because the existing swipe code
8274 is faster than copy on write.
8275 Swings and roundabouts. */
8276 SvUPGRADE(sv, SVt_PV);
8279 /* line is going to be appended to the existing buffer in the sv */
8280 if (PerlIO_isutf8(fp)) {
8282 sv_utf8_upgrade_nomg(sv);
8283 sv_pos_u2b(sv,&append,0);
8285 } else if (SvUTF8(sv)) {
8286 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8292 /* not appending - "clear" the string by setting SvCUR to 0,
8293 * the pv is still avaiable. */
8296 if (PerlIO_isutf8(fp))
8299 if (IN_PERL_COMPILETIME) {
8300 /* we always read code in line mode */
8304 else if (RsSNARF(PL_rs)) {
8305 /* If it is a regular disk file use size from stat() as estimate
8306 of amount we are going to read -- may result in mallocing
8307 more memory than we really need if the layers below reduce
8308 the size we read (e.g. CRLF or a gzip layer).
8311 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8312 const Off_t offset = PerlIO_tell(fp);
8313 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8314 #ifdef PERL_NEW_COPY_ON_WRITE
8315 /* Add an extra byte for the sake of copy-on-write's
8316 * buffer reference count. */
8317 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8319 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8326 else if (RsRECORD(PL_rs)) {
8327 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8329 else if (RsPARA(PL_rs)) {
8335 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8336 if (PerlIO_isutf8(fp)) {
8337 rsptr = SvPVutf8(PL_rs, rslen);
8340 if (SvUTF8(PL_rs)) {
8341 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8342 Perl_croak(aTHX_ "Wide character in $/");
8345 /* extract the raw pointer to the record separator */
8346 rsptr = SvPV_const(PL_rs, rslen);
8350 /* rslast is the last character in the record separator
8351 * note we don't use rslast except when rslen is true, so the
8352 * null assign is a placeholder. */
8353 rslast = rslen ? rsptr[rslen - 1] : '\0';
8355 if (rspara) { /* have to do this both before and after */
8356 do { /* to make sure file boundaries work right */
8359 i = PerlIO_getc(fp);
8363 PerlIO_ungetc(fp,i);
8369 /* See if we know enough about I/O mechanism to cheat it ! */
8371 /* This used to be #ifdef test - it is made run-time test for ease
8372 of abstracting out stdio interface. One call should be cheap
8373 enough here - and may even be a macro allowing compile
8377 if (PerlIO_fast_gets(fp)) {
8379 * We can do buffer based IO operations on this filehandle.
8381 * This means we can bypass a lot of subcalls and process
8382 * the buffer directly, it also means we know the upper bound
8383 * on the amount of data we might read of the current buffer
8384 * into our sv. Knowing this allows us to preallocate the pv
8385 * to be able to hold that maximum, which allows us to simplify
8386 * a lot of logic. */
8389 * We're going to steal some values from the stdio struct
8390 * and put EVERYTHING in the innermost loop into registers.
8392 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8393 STRLEN bpx; /* length of the data in the target sv
8394 used to fix pointers after a SvGROW */
8395 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8396 of data left in the read-ahead buffer.
8397 If 0 then the pv buffer can hold the full
8398 amount left, otherwise this is the amount it
8401 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8402 /* An ungetc()d char is handled separately from the regular
8403 * buffer, so we getc() it back out and stuff it in the buffer.
8405 i = PerlIO_getc(fp);
8406 if (i == EOF) return 0;
8407 *(--((*fp)->_ptr)) = (unsigned char) i;
8411 /* Here is some breathtakingly efficient cheating */
8413 /* When you read the following logic resist the urge to think
8414 * of record separators that are 1 byte long. They are an
8415 * uninteresting special (simple) case.
8417 * Instead think of record separators which are at least 2 bytes
8418 * long, and keep in mind that we need to deal with such
8419 * separators when they cross a read-ahead buffer boundary.
8421 * Also consider that we need to gracefully deal with separators
8422 * that may be longer than a single read ahead buffer.
8424 * Lastly do not forget we want to copy the delimiter as well. We
8425 * are copying all data in the file _up_to_and_including_ the separator
8428 * Now that you have all that in mind here is what is happening below:
8430 * 1. When we first enter the loop we do some memory book keeping to see
8431 * how much free space there is in the target SV. (This sub assumes that
8432 * it is operating on the same SV most of the time via $_ and that it is
8433 * going to be able to reuse the same pv buffer each call.) If there is
8434 * "enough" room then we set "shortbuffered" to how much space there is
8435 * and start reading forward.
8437 * 2. When we scan forward we copy from the read-ahead buffer to the target
8438 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8439 * and the end of the of pv, as well as for the "rslast", which is the last
8440 * char of the separator.
8442 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8443 * (which has a "complete" record up to the point we saw rslast) and check
8444 * it to see if it matches the separator. If it does we are done. If it doesn't
8445 * we continue on with the scan/copy.
8447 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8448 * the IO system to read the next buffer. We do this by doing a getc(), which
8449 * returns a single char read (or EOF), and prefills the buffer, and also
8450 * allows us to find out how full the buffer is. We use this information to
8451 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8452 * the returned single char into the target sv, and then go back into scan
8455 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8456 * remaining space in the read-buffer.
8458 * Note that this code despite its twisty-turny nature is pretty darn slick.
8459 * It manages single byte separators, multi-byte cross boundary separators,
8460 * and cross-read-buffer separators cleanly and efficiently at the cost
8461 * of potentially greatly overallocating the target SV.
8467 /* get the number of bytes remaining in the read-ahead buffer
8468 * on first call on a given fp this will return 0.*/
8469 cnt = PerlIO_get_cnt(fp);
8471 /* make sure we have the room */
8472 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8473 /* Not room for all of it
8474 if we are looking for a separator and room for some
8476 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8477 /* just process what we have room for */
8478 shortbuffered = cnt - SvLEN(sv) + append + 1;
8479 cnt -= shortbuffered;
8482 /* ensure that the target sv has enough room to hold
8483 * the rest of the read-ahead buffer */
8485 /* remember that cnt can be negative */
8486 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8490 /* we have enough room to hold the full buffer, lets scream */
8494 /* extract the pointer to sv's string buffer, offset by append as necessary */
8495 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8496 /* extract the point to the read-ahead buffer */
8497 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8499 /* some trace debug output */
8500 DEBUG_P(PerlIO_printf(Perl_debug_log,
8501 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8502 DEBUG_P(PerlIO_printf(Perl_debug_log,
8503 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8505 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8506 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8510 /* if there is stuff left in the read-ahead buffer */
8512 /* if there is a separator */
8514 /* loop until we hit the end of the read-ahead buffer */
8515 while (cnt > 0) { /* this | eat */
8516 /* scan forward copying and searching for rslast as we go */
8518 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8519 goto thats_all_folks; /* screams | sed :-) */
8523 /* no separator, slurp the full buffer */
8524 Copy(ptr, bp, cnt, char); /* this | eat */
8525 bp += cnt; /* screams | dust */
8526 ptr += cnt; /* louder | sed :-) */
8528 assert (!shortbuffered);
8529 goto cannot_be_shortbuffered;
8533 if (shortbuffered) { /* oh well, must extend */
8534 /* we didnt have enough room to fit the line into the target buffer
8535 * so we must extend the target buffer and keep going */
8536 cnt = shortbuffered;
8538 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8540 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8541 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8542 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8546 cannot_be_shortbuffered:
8547 /* we need to refill the read-ahead buffer if possible */
8549 DEBUG_P(PerlIO_printf(Perl_debug_log,
8550 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8551 PTR2UV(ptr),(IV)cnt));
8552 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8554 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8555 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8556 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8557 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8560 call PerlIO_getc() to let it prefill the lookahead buffer
8562 This used to call 'filbuf' in stdio form, but as that behaves like
8563 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8564 another abstraction.
8566 Note we have to deal with the char in 'i' if we are not at EOF
8568 i = PerlIO_getc(fp); /* get more characters */
8570 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8571 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8572 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8573 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8575 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8576 cnt = PerlIO_get_cnt(fp);
8577 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8578 DEBUG_P(PerlIO_printf(Perl_debug_log,
8579 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8580 PTR2UV(ptr),(IV)cnt));
8582 if (i == EOF) /* all done for ever? */
8583 goto thats_really_all_folks;
8585 /* make sure we have enough space in the target sv */
8586 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8588 SvGROW(sv, bpx + cnt + 2);
8589 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8591 /* copy of the char we got from getc() */
8592 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8594 /* make sure we deal with the i being the last character of a separator */
8595 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8596 goto thats_all_folks;
8600 /* check if we have actually found the separator - only really applies
8602 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8603 memNE((char*)bp - rslen, rsptr, rslen))
8604 goto screamer; /* go back to the fray */
8605 thats_really_all_folks:
8607 cnt += shortbuffered;
8608 DEBUG_P(PerlIO_printf(Perl_debug_log,
8609 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8610 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8611 DEBUG_P(PerlIO_printf(Perl_debug_log,
8612 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8614 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8615 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8617 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8618 DEBUG_P(PerlIO_printf(Perl_debug_log,
8619 "Screamer: done, len=%ld, string=|%.*s|\n",
8620 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8624 /*The big, slow, and stupid way. */
8625 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8626 STDCHAR *buf = NULL;
8627 Newx(buf, 8192, STDCHAR);
8635 const STDCHAR * const bpe = buf + sizeof(buf);
8637 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8638 ; /* keep reading */
8642 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8643 /* Accommodate broken VAXC compiler, which applies U8 cast to
8644 * both args of ?: operator, causing EOF to change into 255
8647 i = (U8)buf[cnt - 1];
8653 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8655 sv_catpvn_nomg(sv, (char *) buf, cnt);
8657 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8659 if (i != EOF && /* joy */
8661 SvCUR(sv) < rslen ||
8662 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8666 * If we're reading from a TTY and we get a short read,
8667 * indicating that the user hit his EOF character, we need
8668 * to notice it now, because if we try to read from the TTY
8669 * again, the EOF condition will disappear.
8671 * The comparison of cnt to sizeof(buf) is an optimization
8672 * that prevents unnecessary calls to feof().
8676 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8680 #ifdef USE_HEAP_INSTEAD_OF_STACK
8685 if (rspara) { /* have to do this both before and after */
8686 while (i != EOF) { /* to make sure file boundaries work right */
8687 i = PerlIO_getc(fp);
8689 PerlIO_ungetc(fp,i);
8695 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8701 Auto-increment of the value in the SV, doing string to numeric conversion
8702 if necessary. Handles 'get' magic and operator overloading.
8708 Perl_sv_inc(pTHX_ SV *const sv)
8717 =for apidoc sv_inc_nomg
8719 Auto-increment of the value in the SV, doing string to numeric conversion
8720 if necessary. Handles operator overloading. Skips handling 'get' magic.
8726 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8733 if (SvTHINKFIRST(sv)) {
8734 if (SvREADONLY(sv)) {
8735 Perl_croak_no_modify();
8739 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8741 i = PTR2IV(SvRV(sv));
8745 else sv_force_normal_flags(sv, 0);
8747 flags = SvFLAGS(sv);
8748 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8749 /* It's (privately or publicly) a float, but not tested as an
8750 integer, so test it to see. */
8752 flags = SvFLAGS(sv);
8754 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8755 /* It's publicly an integer, or privately an integer-not-float */
8756 #ifdef PERL_PRESERVE_IVUV
8760 if (SvUVX(sv) == UV_MAX)
8761 sv_setnv(sv, UV_MAX_P1);
8763 (void)SvIOK_only_UV(sv);
8764 SvUV_set(sv, SvUVX(sv) + 1);
8766 if (SvIVX(sv) == IV_MAX)
8767 sv_setuv(sv, (UV)IV_MAX + 1);
8769 (void)SvIOK_only(sv);
8770 SvIV_set(sv, SvIVX(sv) + 1);
8775 if (flags & SVp_NOK) {
8776 const NV was = SvNVX(sv);
8777 if (LIKELY(!Perl_isinfnan(was)) &&
8778 NV_OVERFLOWS_INTEGERS_AT &&
8779 was >= NV_OVERFLOWS_INTEGERS_AT) {
8780 /* diag_listed_as: Lost precision when %s %f by 1 */
8781 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8782 "Lost precision when incrementing %" NVff " by 1",
8785 (void)SvNOK_only(sv);
8786 SvNV_set(sv, was + 1.0);
8790 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8791 if ((flags & SVTYPEMASK) < SVt_PVIV)
8792 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8793 (void)SvIOK_only(sv);
8798 while (isALPHA(*d)) d++;
8799 while (isDIGIT(*d)) d++;
8800 if (d < SvEND(sv)) {
8801 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8802 #ifdef PERL_PRESERVE_IVUV
8803 /* Got to punt this as an integer if needs be, but we don't issue
8804 warnings. Probably ought to make the sv_iv_please() that does
8805 the conversion if possible, and silently. */
8806 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8807 /* Need to try really hard to see if it's an integer.
8808 9.22337203685478e+18 is an integer.
8809 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8810 so $a="9.22337203685478e+18"; $a+0; $a++
8811 needs to be the same as $a="9.22337203685478e+18"; $a++
8818 /* sv_2iv *should* have made this an NV */
8819 if (flags & SVp_NOK) {
8820 (void)SvNOK_only(sv);
8821 SvNV_set(sv, SvNVX(sv) + 1.0);
8824 /* I don't think we can get here. Maybe I should assert this
8825 And if we do get here I suspect that sv_setnv will croak. NWC
8827 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8828 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8830 #endif /* PERL_PRESERVE_IVUV */
8831 if (!numtype && ckWARN(WARN_NUMERIC))
8832 not_incrementable(sv);
8833 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8837 while (d >= SvPVX_const(sv)) {
8845 /* MKS: The original code here died if letters weren't consecutive.
8846 * at least it didn't have to worry about non-C locales. The
8847 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8848 * arranged in order (although not consecutively) and that only
8849 * [A-Za-z] are accepted by isALPHA in the C locale.
8851 if (isALPHA_FOLD_NE(*d, 'z')) {
8852 do { ++*d; } while (!isALPHA(*d));
8855 *(d--) -= 'z' - 'a';
8860 *(d--) -= 'z' - 'a' + 1;
8864 /* oh,oh, the number grew */
8865 SvGROW(sv, SvCUR(sv) + 2);
8866 SvCUR_set(sv, SvCUR(sv) + 1);
8867 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8878 Auto-decrement of the value in the SV, doing string to numeric conversion
8879 if necessary. Handles 'get' magic and operator overloading.
8885 Perl_sv_dec(pTHX_ SV *const sv)
8894 =for apidoc sv_dec_nomg
8896 Auto-decrement of the value in the SV, doing string to numeric conversion
8897 if necessary. Handles operator overloading. Skips handling 'get' magic.
8903 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8909 if (SvTHINKFIRST(sv)) {
8910 if (SvREADONLY(sv)) {
8911 Perl_croak_no_modify();
8915 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8917 i = PTR2IV(SvRV(sv));
8921 else sv_force_normal_flags(sv, 0);
8923 /* Unlike sv_inc we don't have to worry about string-never-numbers
8924 and keeping them magic. But we mustn't warn on punting */
8925 flags = SvFLAGS(sv);
8926 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8927 /* It's publicly an integer, or privately an integer-not-float */
8928 #ifdef PERL_PRESERVE_IVUV
8932 if (SvUVX(sv) == 0) {
8933 (void)SvIOK_only(sv);
8937 (void)SvIOK_only_UV(sv);
8938 SvUV_set(sv, SvUVX(sv) - 1);
8941 if (SvIVX(sv) == IV_MIN) {
8942 sv_setnv(sv, (NV)IV_MIN);
8946 (void)SvIOK_only(sv);
8947 SvIV_set(sv, SvIVX(sv) - 1);
8952 if (flags & SVp_NOK) {
8955 const NV was = SvNVX(sv);
8956 if (LIKELY(!Perl_isinfnan(was)) &&
8957 NV_OVERFLOWS_INTEGERS_AT &&
8958 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8959 /* diag_listed_as: Lost precision when %s %f by 1 */
8960 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8961 "Lost precision when decrementing %" NVff " by 1",
8964 (void)SvNOK_only(sv);
8965 SvNV_set(sv, was - 1.0);
8969 if (!(flags & SVp_POK)) {
8970 if ((flags & SVTYPEMASK) < SVt_PVIV)
8971 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8973 (void)SvIOK_only(sv);
8976 #ifdef PERL_PRESERVE_IVUV
8978 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8979 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8980 /* Need to try really hard to see if it's an integer.
8981 9.22337203685478e+18 is an integer.
8982 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8983 so $a="9.22337203685478e+18"; $a+0; $a--
8984 needs to be the same as $a="9.22337203685478e+18"; $a--
8991 /* sv_2iv *should* have made this an NV */
8992 if (flags & SVp_NOK) {
8993 (void)SvNOK_only(sv);
8994 SvNV_set(sv, SvNVX(sv) - 1.0);
8997 /* I don't think we can get here. Maybe I should assert this
8998 And if we do get here I suspect that sv_setnv will croak. NWC
9000 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9001 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9004 #endif /* PERL_PRESERVE_IVUV */
9005 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9008 /* this define is used to eliminate a chunk of duplicated but shared logic
9009 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9010 * used anywhere but here - yves
9012 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9014 SSize_t ix = ++PL_tmps_ix; \
9015 if (UNLIKELY(ix >= PL_tmps_max)) \
9016 ix = tmps_grow_p(ix); \
9017 PL_tmps_stack[ix] = (AnSv); \
9021 =for apidoc sv_mortalcopy
9023 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9024 The new SV is marked as mortal. It will be destroyed "soon", either by an
9025 explicit call to FREETMPS, or by an implicit call at places such as
9026 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
9031 /* Make a string that will exist for the duration of the expression
9032 * evaluation. Actually, it may have to last longer than that, but
9033 * hopefully we won't free it until it has been assigned to a
9034 * permanent location. */
9037 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9041 if (flags & SV_GMAGIC)
9042 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9044 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9045 PUSH_EXTEND_MORTAL__SV_C(sv);
9051 =for apidoc sv_newmortal
9053 Creates a new null SV which is mortal. The reference count of the SV is
9054 set to 1. It will be destroyed "soon", either by an explicit call to
9055 FREETMPS, or by an implicit call at places such as statement boundaries.
9056 See also C<sv_mortalcopy> and C<sv_2mortal>.
9062 Perl_sv_newmortal(pTHX)
9067 SvFLAGS(sv) = SVs_TEMP;
9068 PUSH_EXTEND_MORTAL__SV_C(sv);
9074 =for apidoc newSVpvn_flags
9076 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9077 characters) into it. The reference count for the
9078 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9079 string. You are responsible for ensuring that the source string is at least
9080 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9081 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9082 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9083 returning. If C<SVf_UTF8> is set, C<s>
9084 is considered to be in UTF-8 and the
9085 C<SVf_UTF8> flag will be set on the new SV.
9086 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9088 #define newSVpvn_utf8(s, len, u) \
9089 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9095 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9099 /* All the flags we don't support must be zero.
9100 And we're new code so I'm going to assert this from the start. */
9101 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9103 sv_setpvn(sv,s,len);
9105 /* This code used to do a sv_2mortal(), however we now unroll the call to
9106 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9107 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9108 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9109 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9110 * means that we eliminate quite a few steps than it looks - Yves
9111 * (explaining patch by gfx) */
9113 SvFLAGS(sv) |= flags;
9115 if(flags & SVs_TEMP){
9116 PUSH_EXTEND_MORTAL__SV_C(sv);
9123 =for apidoc sv_2mortal
9125 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9126 by an explicit call to FREETMPS, or by an implicit call at places such as
9127 statement boundaries. SvTEMP() is turned on which means that the SV's
9128 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9129 and C<sv_mortalcopy>.
9135 Perl_sv_2mortal(pTHX_ SV *const sv)
9142 PUSH_EXTEND_MORTAL__SV_C(sv);
9150 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9151 characters) into it. The reference count for the
9152 SV is set to 1. If C<len> is zero, Perl will compute the length using
9153 strlen(), (which means if you use this option, that C<s> can't have embedded
9154 C<NUL> characters and has to have a terminating C<NUL> byte).
9156 For efficiency, consider using C<newSVpvn> instead.
9162 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9167 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9172 =for apidoc newSVpvn
9174 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9175 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9176 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9177 are responsible for ensuring that the source buffer is at least
9178 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9185 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9189 sv_setpvn(sv,buffer,len);
9194 =for apidoc newSVhek
9196 Creates a new SV from the hash key structure. It will generate scalars that
9197 point to the shared string table where possible. Returns a new (undefined)
9198 SV if the hek is NULL.
9204 Perl_newSVhek(pTHX_ const HEK *const hek)
9213 if (HEK_LEN(hek) == HEf_SVKEY) {
9214 return newSVsv(*(SV**)HEK_KEY(hek));
9216 const int flags = HEK_FLAGS(hek);
9217 if (flags & HVhek_WASUTF8) {
9219 Andreas would like keys he put in as utf8 to come back as utf8
9221 STRLEN utf8_len = HEK_LEN(hek);
9222 SV * const sv = newSV_type(SVt_PV);
9223 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9224 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9225 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9228 } else if (flags & HVhek_UNSHARED) {
9229 /* A hash that isn't using shared hash keys has to have
9230 the flag in every key so that we know not to try to call
9231 share_hek_hek on it. */
9233 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9238 /* This will be overwhelminly the most common case. */
9240 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9241 more efficient than sharepvn(). */
9245 sv_upgrade(sv, SVt_PV);
9246 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9247 SvCUR_set(sv, HEK_LEN(hek));
9259 =for apidoc newSVpvn_share
9261 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9262 table. If the string does not already exist in the table, it is
9263 created first. Turns on the SvIsCOW flag (or READONLY
9264 and FAKE in 5.16 and earlier). If the C<hash> parameter
9265 is non-zero, that value is used; otherwise the hash is computed.
9266 The string's hash can later be retrieved from the SV
9267 with the C<SvSHARED_HASH()> macro. The idea here is
9268 that as the string table is used for shared hash keys these strings will have
9269 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9275 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9279 bool is_utf8 = FALSE;
9280 const char *const orig_src = src;
9283 STRLEN tmplen = -len;
9285 /* See the note in hv.c:hv_fetch() --jhi */
9286 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9290 PERL_HASH(hash, src, len);
9292 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9293 changes here, update it there too. */
9294 sv_upgrade(sv, SVt_PV);
9295 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9302 if (src != orig_src)
9308 =for apidoc newSVpv_share
9310 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9317 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9319 return newSVpvn_share(src, strlen(src), hash);
9322 #if defined(PERL_IMPLICIT_CONTEXT)
9324 /* pTHX_ magic can't cope with varargs, so this is a no-context
9325 * version of the main function, (which may itself be aliased to us).
9326 * Don't access this version directly.
9330 Perl_newSVpvf_nocontext(const char *const pat, ...)
9336 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9338 va_start(args, pat);
9339 sv = vnewSVpvf(pat, &args);
9346 =for apidoc newSVpvf
9348 Creates a new SV and initializes it with the string formatted like
9355 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9360 PERL_ARGS_ASSERT_NEWSVPVF;
9362 va_start(args, pat);
9363 sv = vnewSVpvf(pat, &args);
9368 /* backend for newSVpvf() and newSVpvf_nocontext() */
9371 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9375 PERL_ARGS_ASSERT_VNEWSVPVF;
9378 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9385 Creates a new SV and copies a floating point value into it.
9386 The reference count for the SV is set to 1.
9392 Perl_newSVnv(pTHX_ const NV n)
9404 Creates a new SV and copies an integer into it. The reference count for the
9411 Perl_newSViv(pTHX_ const IV i)
9417 /* Inlining ONLY the small relevant subset of sv_setiv here
9418 * for performance. Makes a significant difference. */
9420 /* We're starting from SVt_FIRST, so provided that's
9421 * actual 0, we don't have to unset any SV type flags
9422 * to promote to SVt_IV. */
9423 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9425 SET_SVANY_FOR_BODYLESS_IV(sv);
9426 SvFLAGS(sv) |= SVt_IV;
9438 Creates a new SV and copies an unsigned integer into it.
9439 The reference count for the SV is set to 1.
9445 Perl_newSVuv(pTHX_ const UV u)
9449 /* Inlining ONLY the small relevant subset of sv_setuv here
9450 * for performance. Makes a significant difference. */
9452 /* Using ivs is more efficient than using uvs - see sv_setuv */
9453 if (u <= (UV)IV_MAX) {
9454 return newSViv((IV)u);
9459 /* We're starting from SVt_FIRST, so provided that's
9460 * actual 0, we don't have to unset any SV type flags
9461 * to promote to SVt_IV. */
9462 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9464 SET_SVANY_FOR_BODYLESS_IV(sv);
9465 SvFLAGS(sv) |= SVt_IV;
9467 (void)SvIsUV_on(sv);
9476 =for apidoc newSV_type
9478 Creates a new SV, of the type specified. The reference count for the new SV
9485 Perl_newSV_type(pTHX_ const svtype type)
9490 ASSUME(SvTYPE(sv) == SVt_FIRST);
9491 if(type != SVt_FIRST)
9492 sv_upgrade(sv, type);
9497 =for apidoc newRV_noinc
9499 Creates an RV wrapper for an SV. The reference count for the original
9500 SV is B<not> incremented.
9506 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9510 PERL_ARGS_ASSERT_NEWRV_NOINC;
9514 /* We're starting from SVt_FIRST, so provided that's
9515 * actual 0, we don't have to unset any SV type flags
9516 * to promote to SVt_IV. */
9517 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9519 SET_SVANY_FOR_BODYLESS_IV(sv);
9520 SvFLAGS(sv) |= SVt_IV;
9525 SvRV_set(sv, tmpRef);
9530 /* newRV_inc is the official function name to use now.
9531 * newRV_inc is in fact #defined to newRV in sv.h
9535 Perl_newRV(pTHX_ SV *const sv)
9537 PERL_ARGS_ASSERT_NEWRV;
9539 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9545 Creates a new SV which is an exact duplicate of the original SV.
9552 Perl_newSVsv(pTHX_ SV *const old)
9558 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9559 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9562 /* Do this here, otherwise we leak the new SV if this croaks. */
9565 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9566 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9567 sv_setsv_flags(sv, old, SV_NOSTEAL);
9572 =for apidoc sv_reset
9574 Underlying implementation for the C<reset> Perl function.
9575 Note that the perl-level function is vaguely deprecated.
9581 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9583 PERL_ARGS_ASSERT_SV_RESET;
9585 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9589 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9591 char todo[PERL_UCHAR_MAX+1];
9594 if (!stash || SvTYPE(stash) != SVt_PVHV)
9597 if (!s) { /* reset ?? searches */
9598 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9600 const U32 count = mg->mg_len / sizeof(PMOP**);
9601 PMOP **pmp = (PMOP**) mg->mg_ptr;
9602 PMOP *const *const end = pmp + count;
9606 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9608 (*pmp)->op_pmflags &= ~PMf_USED;
9616 /* reset variables */
9618 if (!HvARRAY(stash))
9621 Zero(todo, 256, char);
9625 I32 i = (unsigned char)*s;
9629 max = (unsigned char)*s++;
9630 for ( ; i <= max; i++) {
9633 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9635 for (entry = HvARRAY(stash)[i];
9637 entry = HeNEXT(entry))
9642 if (!todo[(U8)*HeKEY(entry)])
9644 gv = MUTABLE_GV(HeVAL(entry));
9646 if (sv && !SvREADONLY(sv)) {
9647 SV_CHECK_THINKFIRST_COW_DROP(sv);
9648 if (!isGV(sv)) SvOK_off(sv);
9653 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9664 Using various gambits, try to get an IO from an SV: the IO slot if its a
9665 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9666 named after the PV if we're a string.
9668 'Get' magic is ignored on the sv passed in, but will be called on
9669 C<SvRV(sv)> if sv is an RV.
9675 Perl_sv_2io(pTHX_ SV *const sv)
9680 PERL_ARGS_ASSERT_SV_2IO;
9682 switch (SvTYPE(sv)) {
9684 io = MUTABLE_IO(sv);
9688 if (isGV_with_GP(sv)) {
9689 gv = MUTABLE_GV(sv);
9692 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9693 HEKfARG(GvNAME_HEK(gv)));
9699 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9701 SvGETMAGIC(SvRV(sv));
9702 return sv_2io(SvRV(sv));
9704 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9711 if (SvGMAGICAL(sv)) {
9712 newsv = sv_newmortal();
9713 sv_setsv_nomg(newsv, sv);
9715 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9725 Using various gambits, try to get a CV from an SV; in addition, try if
9726 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9727 The flags in C<lref> are passed to gv_fetchsv.
9733 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9738 PERL_ARGS_ASSERT_SV_2CV;
9745 switch (SvTYPE(sv)) {
9749 return MUTABLE_CV(sv);
9759 sv = amagic_deref_call(sv, to_cv_amg);
9762 if (SvTYPE(sv) == SVt_PVCV) {
9763 cv = MUTABLE_CV(sv);
9768 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9769 gv = MUTABLE_GV(sv);
9771 Perl_croak(aTHX_ "Not a subroutine reference");
9773 else if (isGV_with_GP(sv)) {
9774 gv = MUTABLE_GV(sv);
9777 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9784 /* Some flags to gv_fetchsv mean don't really create the GV */
9785 if (!isGV_with_GP(gv)) {
9790 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9791 /* XXX this is probably not what they think they're getting.
9792 * It has the same effect as "sub name;", i.e. just a forward
9803 Returns true if the SV has a true value by Perl's rules.
9804 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9805 instead use an in-line version.
9811 Perl_sv_true(pTHX_ SV *const sv)
9816 const XPV* const tXpv = (XPV*)SvANY(sv);
9818 (tXpv->xpv_cur > 1 ||
9819 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9826 return SvIVX(sv) != 0;
9829 return SvNVX(sv) != 0.0;
9831 return sv_2bool(sv);
9837 =for apidoc sv_pvn_force
9839 Get a sensible string out of the SV somehow.
9840 A private implementation of the C<SvPV_force> macro for compilers which
9841 can't cope with complex macro expressions. Always use the macro instead.
9843 =for apidoc sv_pvn_force_flags
9845 Get a sensible string out of the SV somehow.
9846 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9847 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9848 implemented in terms of this function.
9849 You normally want to use the various wrapper macros instead: see
9850 C<SvPV_force> and C<SvPV_force_nomg>
9856 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9858 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9860 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9861 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9862 sv_force_normal_flags(sv, 0);
9872 if (SvTYPE(sv) > SVt_PVLV
9873 || isGV_with_GP(sv))
9874 /* diag_listed_as: Can't coerce %s to %s in %s */
9875 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9877 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9884 if (SvTYPE(sv) < SVt_PV ||
9885 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9888 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9889 SvGROW(sv, len + 1);
9890 Move(s,SvPVX(sv),len,char);
9892 SvPVX(sv)[len] = '\0';
9895 SvPOK_on(sv); /* validate pointer */
9897 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9898 PTR2UV(sv),SvPVX_const(sv)));
9901 (void)SvPOK_only_UTF8(sv);
9902 return SvPVX_mutable(sv);
9906 =for apidoc sv_pvbyten_force
9908 The backend for the C<SvPVbytex_force> macro. Always use the macro
9915 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9917 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9919 sv_pvn_force(sv,lp);
9920 sv_utf8_downgrade(sv,0);
9926 =for apidoc sv_pvutf8n_force
9928 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9935 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9937 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9940 sv_utf8_upgrade_nomg(sv);
9946 =for apidoc sv_reftype
9948 Returns a string describing what the SV is a reference to.
9954 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9956 PERL_ARGS_ASSERT_SV_REFTYPE;
9957 if (ob && SvOBJECT(sv)) {
9958 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9961 /* WARNING - There is code, for instance in mg.c, that assumes that
9962 * the only reason that sv_reftype(sv,0) would return a string starting
9963 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9964 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9965 * this routine inside other subs, and it saves time.
9966 * Do not change this assumption without searching for "dodgy type check" in
9969 switch (SvTYPE(sv)) {
9984 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9985 /* tied lvalues should appear to be
9986 * scalars for backwards compatibility */
9987 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9988 ? "SCALAR" : "LVALUE");
9989 case SVt_PVAV: return "ARRAY";
9990 case SVt_PVHV: return "HASH";
9991 case SVt_PVCV: return "CODE";
9992 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9993 ? "GLOB" : "SCALAR");
9994 case SVt_PVFM: return "FORMAT";
9995 case SVt_PVIO: return "IO";
9996 case SVt_INVLIST: return "INVLIST";
9997 case SVt_REGEXP: return "REGEXP";
9998 default: return "UNKNOWN";
10006 Returns a SV describing what the SV passed in is a reference to.
10012 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10014 PERL_ARGS_ASSERT_SV_REF;
10017 dst = sv_newmortal();
10019 if (ob && SvOBJECT(sv)) {
10020 HvNAME_get(SvSTASH(sv))
10021 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10022 : sv_setpvn(dst, "__ANON__", 8);
10025 const char * reftype = sv_reftype(sv, 0);
10026 sv_setpv(dst, reftype);
10032 =for apidoc sv_isobject
10034 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10035 object. If the SV is not an RV, or if the object is not blessed, then this
10042 Perl_sv_isobject(pTHX_ SV *sv)
10058 Returns a boolean indicating whether the SV is blessed into the specified
10059 class. This does not check for subtypes; use C<sv_derived_from> to verify
10060 an inheritance relationship.
10066 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10068 const char *hvname;
10070 PERL_ARGS_ASSERT_SV_ISA;
10080 hvname = HvNAME_get(SvSTASH(sv));
10084 return strEQ(hvname, name);
10088 =for apidoc newSVrv
10090 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10091 RV then it will be upgraded to one. If C<classname> is non-null then the new
10092 SV will be blessed in the specified package. The new SV is returned and its
10093 reference count is 1. The reference count 1 is owned by C<rv>.
10099 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10103 PERL_ARGS_ASSERT_NEWSVRV;
10107 SV_CHECK_THINKFIRST_COW_DROP(rv);
10109 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10110 const U32 refcnt = SvREFCNT(rv);
10114 SvREFCNT(rv) = refcnt;
10116 sv_upgrade(rv, SVt_IV);
10117 } else if (SvROK(rv)) {
10118 SvREFCNT_dec(SvRV(rv));
10120 prepare_SV_for_RV(rv);
10128 HV* const stash = gv_stashpv(classname, GV_ADD);
10129 (void)sv_bless(rv, stash);
10135 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10137 SV * const lv = newSV_type(SVt_PVLV);
10138 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10140 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10141 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10142 LvSTARGOFF(lv) = ix;
10143 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10148 =for apidoc sv_setref_pv
10150 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10151 argument will be upgraded to an RV. That RV will be modified to point to
10152 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10153 into the SV. The C<classname> argument indicates the package for the
10154 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10155 will have a reference count of 1, and the RV will be returned.
10157 Do not use with other Perl types such as HV, AV, SV, CV, because those
10158 objects will become corrupted by the pointer copy process.
10160 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10166 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10168 PERL_ARGS_ASSERT_SV_SETREF_PV;
10171 sv_setsv(rv, &PL_sv_undef);
10175 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10180 =for apidoc sv_setref_iv
10182 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10183 argument will be upgraded to an RV. That RV will be modified to point to
10184 the new SV. The C<classname> argument indicates the package for the
10185 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10186 will have a reference count of 1, and the RV will be returned.
10192 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10194 PERL_ARGS_ASSERT_SV_SETREF_IV;
10196 sv_setiv(newSVrv(rv,classname), iv);
10201 =for apidoc sv_setref_uv
10203 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10204 argument will be upgraded to an RV. That RV will be modified to point to
10205 the new SV. The C<classname> argument indicates the package for the
10206 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10207 will have a reference count of 1, and the RV will be returned.
10213 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10215 PERL_ARGS_ASSERT_SV_SETREF_UV;
10217 sv_setuv(newSVrv(rv,classname), uv);
10222 =for apidoc sv_setref_nv
10224 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10225 argument will be upgraded to an RV. That RV will be modified to point to
10226 the new SV. The C<classname> argument indicates the package for the
10227 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10228 will have a reference count of 1, and the RV will be returned.
10234 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10236 PERL_ARGS_ASSERT_SV_SETREF_NV;
10238 sv_setnv(newSVrv(rv,classname), nv);
10243 =for apidoc sv_setref_pvn
10245 Copies a string into a new SV, optionally blessing the SV. The length of the
10246 string must be specified with C<n>. The C<rv> argument will be upgraded to
10247 an RV. That RV will be modified to point to the new SV. The C<classname>
10248 argument indicates the package for the blessing. Set C<classname> to
10249 C<NULL> to avoid the blessing. The new SV will have a reference count
10250 of 1, and the RV will be returned.
10252 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10258 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10259 const char *const pv, const STRLEN n)
10261 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10263 sv_setpvn(newSVrv(rv,classname), pv, n);
10268 =for apidoc sv_bless
10270 Blesses an SV into a specified package. The SV must be an RV. The package
10271 must be designated by its stash (see C<gv_stashpv()>). The reference count
10272 of the SV is unaffected.
10278 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10281 HV *oldstash = NULL;
10283 PERL_ARGS_ASSERT_SV_BLESS;
10287 Perl_croak(aTHX_ "Can't bless non-reference value");
10289 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10290 if (SvREADONLY(tmpRef))
10291 Perl_croak_no_modify();
10292 if (SvOBJECT(tmpRef)) {
10293 oldstash = SvSTASH(tmpRef);
10296 SvOBJECT_on(tmpRef);
10297 SvUPGRADE(tmpRef, SVt_PVMG);
10298 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10299 SvREFCNT_dec(oldstash);
10301 if(SvSMAGICAL(tmpRef))
10302 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10310 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10311 * as it is after unglobbing it.
10314 PERL_STATIC_INLINE void
10315 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10319 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10321 PERL_ARGS_ASSERT_SV_UNGLOB;
10323 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10325 if (!(flags & SV_COW_DROP_PV))
10326 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10328 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10330 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10331 && HvNAME_get(stash))
10332 mro_method_changed_in(stash);
10333 gp_free(MUTABLE_GV(sv));
10336 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10337 GvSTASH(sv) = NULL;
10340 if (GvNAME_HEK(sv)) {
10341 unshare_hek(GvNAME_HEK(sv));
10343 isGV_with_GP_off(sv);
10345 if(SvTYPE(sv) == SVt_PVGV) {
10346 /* need to keep SvANY(sv) in the right arena */
10347 xpvmg = new_XPVMG();
10348 StructCopy(SvANY(sv), xpvmg, XPVMG);
10349 del_XPVGV(SvANY(sv));
10352 SvFLAGS(sv) &= ~SVTYPEMASK;
10353 SvFLAGS(sv) |= SVt_PVMG;
10356 /* Intentionally not calling any local SET magic, as this isn't so much a
10357 set operation as merely an internal storage change. */
10358 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10359 else sv_setsv_flags(sv, temp, 0);
10361 if ((const GV *)sv == PL_last_in_gv)
10362 PL_last_in_gv = NULL;
10363 else if ((const GV *)sv == PL_statgv)
10368 =for apidoc sv_unref_flags
10370 Unsets the RV status of the SV, and decrements the reference count of
10371 whatever was being referenced by the RV. This can almost be thought of
10372 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10373 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10374 (otherwise the decrementing is conditional on the reference count being
10375 different from one or the reference being a readonly SV).
10382 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10384 SV* const target = SvRV(ref);
10386 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10388 if (SvWEAKREF(ref)) {
10389 sv_del_backref(target, ref);
10390 SvWEAKREF_off(ref);
10391 SvRV_set(ref, NULL);
10394 SvRV_set(ref, NULL);
10396 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10397 assigned to as BEGIN {$a = \"Foo"} will fail. */
10398 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10399 SvREFCNT_dec_NN(target);
10400 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10401 sv_2mortal(target); /* Schedule for freeing later */
10405 =for apidoc sv_untaint
10407 Untaint an SV. Use C<SvTAINTED_off> instead.
10413 Perl_sv_untaint(pTHX_ SV *const sv)
10415 PERL_ARGS_ASSERT_SV_UNTAINT;
10416 PERL_UNUSED_CONTEXT;
10418 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10419 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10426 =for apidoc sv_tainted
10428 Test an SV for taintedness. Use C<SvTAINTED> instead.
10434 Perl_sv_tainted(pTHX_ SV *const sv)
10436 PERL_ARGS_ASSERT_SV_TAINTED;
10437 PERL_UNUSED_CONTEXT;
10439 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10440 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10441 if (mg && (mg->mg_len & 1) )
10448 =for apidoc sv_setpviv
10450 Copies an integer into the given SV, also updating its string value.
10451 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10457 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10459 char buf[TYPE_CHARS(UV)];
10461 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10463 PERL_ARGS_ASSERT_SV_SETPVIV;
10465 sv_setpvn(sv, ptr, ebuf - ptr);
10469 =for apidoc sv_setpviv_mg
10471 Like C<sv_setpviv>, but also handles 'set' magic.
10477 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10479 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10481 sv_setpviv(sv, iv);
10485 #if defined(PERL_IMPLICIT_CONTEXT)
10487 /* pTHX_ magic can't cope with varargs, so this is a no-context
10488 * version of the main function, (which may itself be aliased to us).
10489 * Don't access this version directly.
10493 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10498 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10500 va_start(args, pat);
10501 sv_vsetpvf(sv, pat, &args);
10505 /* pTHX_ magic can't cope with varargs, so this is a no-context
10506 * version of the main function, (which may itself be aliased to us).
10507 * Don't access this version directly.
10511 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10516 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10518 va_start(args, pat);
10519 sv_vsetpvf_mg(sv, pat, &args);
10525 =for apidoc sv_setpvf
10527 Works like C<sv_catpvf> but copies the text into the SV instead of
10528 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10534 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10538 PERL_ARGS_ASSERT_SV_SETPVF;
10540 va_start(args, pat);
10541 sv_vsetpvf(sv, pat, &args);
10546 =for apidoc sv_vsetpvf
10548 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10549 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10551 Usually used via its frontend C<sv_setpvf>.
10557 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10559 PERL_ARGS_ASSERT_SV_VSETPVF;
10561 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10565 =for apidoc sv_setpvf_mg
10567 Like C<sv_setpvf>, but also handles 'set' magic.
10573 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10577 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10579 va_start(args, pat);
10580 sv_vsetpvf_mg(sv, pat, &args);
10585 =for apidoc sv_vsetpvf_mg
10587 Like C<sv_vsetpvf>, but also handles 'set' magic.
10589 Usually used via its frontend C<sv_setpvf_mg>.
10595 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10597 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10599 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10603 #if defined(PERL_IMPLICIT_CONTEXT)
10605 /* pTHX_ magic can't cope with varargs, so this is a no-context
10606 * version of the main function, (which may itself be aliased to us).
10607 * Don't access this version directly.
10611 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10616 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10618 va_start(args, pat);
10619 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10623 /* pTHX_ magic can't cope with varargs, so this is a no-context
10624 * version of the main function, (which may itself be aliased to us).
10625 * Don't access this version directly.
10629 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10634 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10636 va_start(args, pat);
10637 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10644 =for apidoc sv_catpvf
10646 Processes its arguments like C<sprintf> and appends the formatted
10647 output to an SV. If the appended data contains "wide" characters
10648 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10649 and characters >255 formatted with %c), the original SV might get
10650 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10651 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10652 valid UTF-8; if the original SV was bytes, the pattern should be too.
10657 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10661 PERL_ARGS_ASSERT_SV_CATPVF;
10663 va_start(args, pat);
10664 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10669 =for apidoc sv_vcatpvf
10671 Processes its arguments like C<vsprintf> and appends the formatted output
10672 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10674 Usually used via its frontend C<sv_catpvf>.
10680 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10682 PERL_ARGS_ASSERT_SV_VCATPVF;
10684 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10688 =for apidoc sv_catpvf_mg
10690 Like C<sv_catpvf>, but also handles 'set' magic.
10696 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10700 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10702 va_start(args, pat);
10703 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10709 =for apidoc sv_vcatpvf_mg
10711 Like C<sv_vcatpvf>, but also handles 'set' magic.
10713 Usually used via its frontend C<sv_catpvf_mg>.
10719 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10721 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10723 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10728 =for apidoc sv_vsetpvfn
10730 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10733 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10739 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10740 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10742 PERL_ARGS_ASSERT_SV_VSETPVFN;
10745 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10750 * Warn of missing argument to sprintf, and then return a defined value
10751 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10754 S_vcatpvfn_missing_argument(pTHX) {
10755 if (ckWARN(WARN_MISSING)) {
10756 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10757 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10764 S_expect_number(pTHX_ char **const pattern)
10768 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10770 switch (**pattern) {
10771 case '1': case '2': case '3':
10772 case '4': case '5': case '6':
10773 case '7': case '8': case '9':
10774 var = *(*pattern)++ - '0';
10775 while (isDIGIT(**pattern)) {
10776 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10778 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10786 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10788 const int neg = nv < 0;
10791 PERL_ARGS_ASSERT_F0CONVERT;
10793 if (UNLIKELY(Perl_isinfnan(nv))) {
10794 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10804 if (uv & 1 && uv == nv)
10805 uv--; /* Round to even */
10807 const unsigned dig = uv % 10;
10809 } while (uv /= 10);
10820 =for apidoc sv_vcatpvfn
10822 =for apidoc sv_vcatpvfn_flags
10824 Processes its arguments like C<vsprintf> and appends the formatted output
10825 to an SV. Uses an array of SVs if the C style variable argument list is
10826 missing (NULL). When running with taint checks enabled, indicates via
10827 C<maybe_tainted> if results are untrustworthy (often due to the use of
10830 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10832 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10837 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10838 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10839 vec_utf8 = DO_UTF8(vecsv);
10841 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10844 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10845 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10847 PERL_ARGS_ASSERT_SV_VCATPVFN;
10849 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10852 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10853 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10854 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10855 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10856 * after the first 1023 zero bits.
10858 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10859 * of dynamically growing buffer might be better, start at just 16 bytes
10860 * (for example) and grow only when necessary. Or maybe just by looking
10861 * at the exponents of the two doubles? */
10862 # define DOUBLEDOUBLE_MAXBITS 2098
10865 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10866 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10867 * per xdigit. For the double-double case, this can be rather many.
10868 * The non-double-double-long-double overshoots since all bits of NV
10869 * are not mantissa bits, there are also exponent bits. */
10870 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10871 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10873 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10876 /* If we do not have a known long double format, (including not using
10877 * long doubles, or long doubles being equal to doubles) then we will
10878 * fall back to the ldexp/frexp route, with which we can retrieve at
10879 * most as many bits as our widest unsigned integer type is. We try
10880 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10882 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10883 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10885 #if defined(HAS_QUAD) && defined(Uquad_t)
10886 # define MANTISSATYPE Uquad_t
10887 # define MANTISSASIZE 8
10889 # define MANTISSATYPE UV
10890 # define MANTISSASIZE UVSIZE
10893 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10894 # define HEXTRACT_LITTLE_ENDIAN
10895 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10896 # define HEXTRACT_BIG_ENDIAN
10898 # define HEXTRACT_MIX_ENDIAN
10901 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10902 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10903 * are being extracted from (either directly from the long double in-memory
10904 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10905 * is used to update the exponent. vhex is the pointer to the beginning
10906 * of the output buffer (of VHEX_SIZE).
10908 * The tricky part is that S_hextract() needs to be called twice:
10909 * the first time with vend as NULL, and the second time with vend as
10910 * the pointer returned by the first call. What happens is that on
10911 * the first round the output size is computed, and the intended
10912 * extraction sanity checked. On the second round the actual output
10913 * (the extraction of the hexadecimal values) takes place.
10914 * Sanity failures cause fatal failures during both rounds. */
10916 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10920 int ixmin = 0, ixmax = 0;
10922 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10923 * and elsewhere. */
10925 /* These macros are just to reduce typos, they have multiple
10926 * repetitions below, but usually only one (or sometimes two)
10927 * of them is really being used. */
10928 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10929 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10930 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10931 #define HEXTRACT_OUTPUT(ix) \
10933 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10935 #define HEXTRACT_COUNT(ix, c) \
10937 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10939 #define HEXTRACT_BYTE(ix) \
10941 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10943 #define HEXTRACT_LO_NYBBLE(ix) \
10945 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10947 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10948 * to make it look less odd when the top bits of a NV
10949 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10950 * order bits can be in the "low nybble" of a byte. */
10951 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10952 #define HEXTRACT_BYTES_LE(a, b) \
10953 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10954 #define HEXTRACT_BYTES_BE(a, b) \
10955 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10956 #define HEXTRACT_IMPLICIT_BIT(nv) \
10958 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10961 /* Most formats do. Those which don't should undef this. */
10962 #define HEXTRACT_HAS_IMPLICIT_BIT
10963 /* Many formats do. Those which don't should undef this. */
10964 #define HEXTRACT_HAS_TOP_NYBBLE
10966 /* HEXTRACTSIZE is the maximum number of xdigits. */
10967 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10968 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/4)
10970 # define HEXTRACTSIZE 2 * NVSIZE
10973 const U8* vmaxend = vhex + HEXTRACTSIZE;
10974 PERL_UNUSED_VAR(ix); /* might happen */
10975 (void)Perl_frexp(PERL_ABS(nv), exponent);
10976 if (vend && (vend <= vhex || vend > vmaxend))
10977 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10979 /* First check if using long doubles. */
10980 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10981 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10982 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10983 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10984 /* The bytes 13..0 are the mantissa/fraction,
10985 * the 15,14 are the sign+exponent. */
10986 const U8* nvp = (const U8*)(&nv);
10987 HEXTRACT_IMPLICIT_BIT(nv);
10988 # undef HEXTRACT_HAS_TOP_NYBBLE
10989 HEXTRACT_BYTES_LE(13, 0);
10990 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10991 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10992 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10993 /* The bytes 2..15 are the mantissa/fraction,
10994 * the 0,1 are the sign+exponent. */
10995 const U8* nvp = (const U8*)(&nv);
10996 HEXTRACT_IMPLICIT_BIT(nv);
10997 # undef HEXTRACT_HAS_TOP_NYBBLE
10998 HEXTRACT_BYTES_BE(2, 15);
10999 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11000 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11001 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
11002 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
11003 * meaning that 2 or 6 bytes are empty padding. */
11004 /* The bytes 7..0 are the mantissa/fraction */
11005 const U8* nvp = (const U8*)(&nv);
11006 # undef HEXTRACT_HAS_IMPLICIT_BIT
11007 # undef HEXTRACT_HAS_TOP_NYBBLE
11008 HEXTRACT_BYTES_LE(7, 0);
11009 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11010 /* Does this format ever happen? (Wikipedia says the Motorola
11011 * 6888x math coprocessors used format _like_ this but padded
11012 * to 96 bits with 16 unused bits between the exponent and the
11014 const U8* nvp = (const U8*)(&nv);
11015 # undef HEXTRACT_HAS_IMPLICIT_BIT
11016 # undef HEXTRACT_HAS_TOP_NYBBLE
11017 HEXTRACT_BYTES_BE(0, 7);
11019 # define HEXTRACT_FALLBACK
11020 /* Double-double format: two doubles next to each other.
11021 * The first double is the high-order one, exactly like
11022 * it would be for a "lone" double. The second double
11023 * is shifted down using the exponent so that that there
11024 * are no common bits. The tricky part is that the value
11025 * of the double-double is the SUM of the two doubles and
11026 * the second one can be also NEGATIVE.
11028 * Because of this tricky construction the bytewise extraction we
11029 * use for the other long double formats doesn't work, we must
11030 * extract the values bit by bit.
11032 * The little-endian double-double is used .. somewhere?
11034 * The big endian double-double is used in e.g. PPC/Power (AIX)
11037 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11038 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11039 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11042 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11043 /* Using normal doubles, not long doubles.
11045 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11046 * bytes, since we might need to handle printf precision, and
11047 * also need to insert the radix. */
11049 # ifdef HEXTRACT_LITTLE_ENDIAN
11050 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11051 const U8* nvp = (const U8*)(&nv);
11052 HEXTRACT_IMPLICIT_BIT(nv);
11053 HEXTRACT_TOP_NYBBLE(6);
11054 HEXTRACT_BYTES_LE(5, 0);
11055 # elif defined(HEXTRACT_BIG_ENDIAN)
11056 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11057 const U8* nvp = (const U8*)(&nv);
11058 HEXTRACT_IMPLICIT_BIT(nv);
11059 HEXTRACT_TOP_NYBBLE(1);
11060 HEXTRACT_BYTES_BE(2, 7);
11061 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11062 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11063 const U8* nvp = (const U8*)(&nv);
11064 HEXTRACT_IMPLICIT_BIT(nv);
11065 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11066 HEXTRACT_BYTE(1); /* 5 */
11067 HEXTRACT_BYTE(0); /* 4 */
11068 HEXTRACT_BYTE(7); /* 3 */
11069 HEXTRACT_BYTE(6); /* 2 */
11070 HEXTRACT_BYTE(5); /* 1 */
11071 HEXTRACT_BYTE(4); /* 0 */
11072 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11073 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11074 const U8* nvp = (const U8*)(&nv);
11075 HEXTRACT_IMPLICIT_BIT(nv);
11076 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11077 HEXTRACT_BYTE(6); /* 5 */
11078 HEXTRACT_BYTE(7); /* 4 */
11079 HEXTRACT_BYTE(0); /* 3 */
11080 HEXTRACT_BYTE(1); /* 2 */
11081 HEXTRACT_BYTE(2); /* 1 */
11082 HEXTRACT_BYTE(3); /* 0 */
11084 # define HEXTRACT_FALLBACK
11087 # define HEXTRACT_FALLBACK
11089 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11090 # ifdef HEXTRACT_FALLBACK
11091 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11092 /* The fallback is used for the double-double format, and
11093 * for unknown long double formats, and for unknown double
11094 * formats, or in general unknown NV formats. */
11095 if (nv == (NV)0.0) {
11103 NV d = nv < 0 ? -nv : nv;
11105 U8 ha = 0x0; /* hexvalue accumulator */
11106 U8 hd = 0x8; /* hexvalue digit */
11108 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11109 * this is essentially manual frexp(). Multiplying by 0.5 and
11110 * doubling should be lossless in binary floating point. */
11120 while (d >= e + e) {
11124 /* Now e <= d < 2*e */
11126 /* First extract the leading hexdigit (the implicit bit). */
11142 /* Then extract the remaining hexdigits. */
11143 while (d > (NV)0.0) {
11149 /* Output or count in groups of four bits,
11150 * that is, when the hexdigit is down to one. */
11155 /* Reset the hexvalue. */
11164 /* Flush possible pending hexvalue. */
11174 /* Croak for various reasons: if the output pointer escaped the
11175 * output buffer, if the extraction index escaped the extraction
11176 * buffer, or if the ending output pointer didn't match the
11177 * previously computed value. */
11178 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11179 /* For double-double the ixmin and ixmax stay at zero,
11180 * which is convenient since the HEXTRACTSIZE is tricky
11181 * for double-double. */
11182 ixmin < 0 || ixmax >= NVSIZE ||
11183 (vend && v != vend))
11184 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11189 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11190 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11195 const char *patend;
11198 static const char nullstr[] = "(null)";
11200 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11201 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11203 /* Times 4: a decimal digit takes more than 3 binary digits.
11204 * NV_DIG: mantissa takes than many decimal digits.
11205 * Plus 32: Playing safe. */
11206 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11207 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11208 bool hexfp = FALSE; /* hexadecimal floating point? */
11210 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
11212 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11213 PERL_UNUSED_ARG(maybe_tainted);
11215 if (flags & SV_GMAGIC)
11218 /* no matter what, this is a string now */
11219 (void)SvPV_force_nomg(sv, origlen);
11221 /* special-case "", "%s", and "%-p" (SVf - see below) */
11223 if (svmax && ckWARN(WARN_REDUNDANT))
11224 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11225 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11228 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11229 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11230 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11231 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11234 const char * const s = va_arg(*args, char*);
11235 sv_catpv_nomg(sv, s ? s : nullstr);
11237 else if (svix < svmax) {
11238 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11239 SvGETMAGIC(*svargs);
11240 sv_catsv_nomg(sv, *svargs);
11243 S_vcatpvfn_missing_argument(aTHX);
11246 if (args && patlen == 3 && pat[0] == '%' &&
11247 pat[1] == '-' && pat[2] == 'p') {
11248 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11249 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11250 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11251 argsv = MUTABLE_SV(va_arg(*args, void*));
11252 sv_catsv_nomg(sv, argsv);
11256 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11257 /* special-case "%.<number>[gf]" */
11258 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11259 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11260 unsigned digits = 0;
11264 while (*pp >= '0' && *pp <= '9')
11265 digits = 10 * digits + (*pp++ - '0');
11267 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11268 format the first argument and WARN_REDUNDANT if svmax > 1?
11269 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11270 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11271 const NV nv = SvNV(*svargs);
11272 if (LIKELY(!Perl_isinfnan(nv))) {
11274 /* Add check for digits != 0 because it seems that some
11275 gconverts are buggy in this case, and we don't yet have
11276 a Configure test for this. */
11277 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11278 /* 0, point, slack */
11279 STORE_LC_NUMERIC_SET_TO_NEEDED();
11280 SNPRINTF_G(nv, ebuf, size, digits);
11281 sv_catpv_nomg(sv, ebuf);
11282 if (*ebuf) /* May return an empty string for digits==0 */
11285 } else if (!digits) {
11288 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11289 sv_catpvn_nomg(sv, p, l);
11296 #endif /* !USE_LONG_DOUBLE */
11298 if (!args && svix < svmax && DO_UTF8(*svargs))
11301 patend = (char*)pat + patlen;
11302 for (p = (char*)pat; p < patend; p = q) {
11305 bool vectorize = FALSE;
11306 bool vectorarg = FALSE;
11307 bool vec_utf8 = FALSE;
11313 bool has_precis = FALSE;
11315 const I32 osvix = svix;
11316 bool is_utf8 = FALSE; /* is this item utf8? */
11317 #ifdef HAS_LDBL_SPRINTF_BUG
11318 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11319 with sfio - Allen <allens@cpan.org> */
11320 bool fix_ldbl_sprintf_bug = FALSE;
11324 U8 utf8buf[UTF8_MAXBYTES+1];
11325 STRLEN esignlen = 0;
11327 const char *eptr = NULL;
11328 const char *fmtstart;
11331 const U8 *vecstr = NULL;
11338 /* We need a long double target in case HAS_LONG_DOUBLE,
11339 * even without USE_LONG_DOUBLE, so that we can printf with
11340 * long double formats, even without NV being long double.
11341 * But we call the target 'fv' instead of 'nv', since most of
11342 * the time it is not (most compilers these days recognize
11343 * "long double", even if only as a synonym for "double").
11345 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11346 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11348 # ifdef Perl_isfinitel
11349 # define FV_ISFINITE(x) Perl_isfinitel(x)
11351 # define FV_GF PERL_PRIgldbl
11352 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11353 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11354 # define NV_TO_FV(nv,fv) STMT_START { \
11356 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11359 # define NV_TO_FV(nv,fv) (fv)=(nv)
11363 # define FV_GF NVgf
11364 # define NV_TO_FV(nv,fv) (fv)=(nv)
11366 #ifndef FV_ISFINITE
11367 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11373 const char *dotstr = ".";
11374 STRLEN dotstrlen = 1;
11375 I32 efix = 0; /* explicit format parameter index */
11376 I32 ewix = 0; /* explicit width index */
11377 I32 epix = 0; /* explicit precision index */
11378 I32 evix = 0; /* explicit vector index */
11379 bool asterisk = FALSE;
11380 bool infnan = FALSE;
11382 /* echo everything up to the next format specification */
11383 for (q = p; q < patend && *q != '%'; ++q) ;
11385 if (has_utf8 && !pat_utf8)
11386 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11388 sv_catpvn_nomg(sv, p, q - p);
11397 We allow format specification elements in this order:
11398 \d+\$ explicit format parameter index
11400 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11401 0 flag (as above): repeated to allow "v02"
11402 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11403 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11405 [%bcdefginopsuxDFOUX] format (mandatory)
11410 As of perl5.9.3, printf format checking is on by default.
11411 Internally, perl uses %p formats to provide an escape to
11412 some extended formatting. This block deals with those
11413 extensions: if it does not match, (char*)q is reset and
11414 the normal format processing code is used.
11416 Currently defined extensions are:
11417 %p include pointer address (standard)
11418 %-p (SVf) include an SV (previously %_)
11419 %-<num>p include an SV with precision <num>
11421 %3p include a HEK with precision of 256
11422 %4p char* preceded by utf8 flag and length
11423 %<num>p (where num is 1 or > 4) reserved for future
11426 Robin Barker 2005-07-14 (but modified since)
11428 %1p (VDf) removed. RMB 2007-10-19
11435 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11436 /* The argument has already gone through cBOOL, so the cast
11438 is_utf8 = (bool)va_arg(*args, int);
11439 elen = va_arg(*args, UV);
11440 if ((IV)elen < 0) {
11441 /* check if utf8 length is larger than 0 when cast to IV */
11442 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11443 elen= 0; /* otherwise we want to treat this as an empty string */
11445 eptr = va_arg(*args, char *);
11446 q += sizeof(UTF8f)-1;
11449 n = expect_number(&q);
11451 if (sv) { /* SVf */
11456 argsv = MUTABLE_SV(va_arg(*args, void*));
11457 eptr = SvPV_const(argsv, elen);
11458 if (DO_UTF8(argsv))
11462 else if (n==2 || n==3) { /* HEKf */
11463 HEK * const hek = va_arg(*args, HEK *);
11464 eptr = HEK_KEY(hek);
11465 elen = HEK_LEN(hek);
11466 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11467 if (n==3) precis = 256, has_precis = TRUE;
11471 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11472 "internal %%<num>p might conflict with future printf extensions");
11478 if ( (width = expect_number(&q)) ) {
11482 if (!no_redundant_warning)
11483 /* I've forgotten if it's a better
11484 micro-optimization to always set this or to
11485 only set it if it's unset */
11486 no_redundant_warning = TRUE;
11498 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11527 if ( (ewix = expect_number(&q)) )
11536 if ((vectorarg = asterisk)) {
11549 width = expect_number(&q);
11552 if (vectorize && vectorarg) {
11553 /* vectorizing, but not with the default "." */
11555 vecsv = va_arg(*args, SV*);
11557 vecsv = (evix > 0 && evix <= svmax)
11558 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11560 vecsv = svix < svmax
11561 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11563 dotstr = SvPV_const(vecsv, dotstrlen);
11564 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11565 bad with tied or overloaded values that return UTF8. */
11566 if (DO_UTF8(vecsv))
11568 else if (has_utf8) {
11569 vecsv = sv_mortalcopy(vecsv);
11570 sv_utf8_upgrade(vecsv);
11571 dotstr = SvPV_const(vecsv, dotstrlen);
11578 i = va_arg(*args, int);
11580 i = (ewix ? ewix <= svmax : svix < svmax) ?
11581 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11583 width = (i < 0) ? -i : i;
11593 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11595 /* XXX: todo, support specified precision parameter */
11599 i = va_arg(*args, int);
11601 i = (ewix ? ewix <= svmax : svix < svmax)
11602 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11604 has_precis = !(i < 0);
11608 while (isDIGIT(*q))
11609 precis = precis * 10 + (*q++ - '0');
11618 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11619 vecsv = svargs[efix ? efix-1 : svix++];
11620 vecstr = (U8*)SvPV_const(vecsv,veclen);
11621 vec_utf8 = DO_UTF8(vecsv);
11623 /* if this is a version object, we need to convert
11624 * back into v-string notation and then let the
11625 * vectorize happen normally
11627 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11628 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11629 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11630 "vector argument not supported with alpha versions");
11633 vecsv = sv_newmortal();
11634 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11636 vecstr = (U8*)SvPV_const(vecsv, veclen);
11637 vec_utf8 = DO_UTF8(vecsv);
11651 case 'I': /* Ix, I32x, and I64x */
11652 # ifdef USE_64_BIT_INT
11653 if (q[1] == '6' && q[2] == '4') {
11659 if (q[1] == '3' && q[2] == '2') {
11663 # ifdef USE_64_BIT_INT
11669 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11670 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11673 # ifdef USE_QUADMATH
11686 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11687 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11688 if (*q == 'l') { /* lld, llf */
11697 if (*++q == 'h') { /* hhd, hhu */
11726 if (!vectorize && !args) {
11728 const I32 i = efix-1;
11729 argsv = (i >= 0 && i < svmax)
11730 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11732 argsv = (svix >= 0 && svix < svmax)
11733 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11737 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11738 /* XXX va_arg(*args) case? need peek, use va_copy? */
11740 if (UNLIKELY(SvAMAGIC(argsv)))
11741 argsv = sv_2num(argsv);
11742 infnan = UNLIKELY(isinfnansv(argsv));
11745 switch (c = *q++) {
11753 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11754 /* no va_arg() case */
11755 SvNV_nomg(argsv), (int)c);
11756 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11758 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11760 eptr = (char*)utf8buf;
11761 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11775 eptr = va_arg(*args, char*);
11777 elen = strlen(eptr);
11779 eptr = (char *)nullstr;
11780 elen = sizeof nullstr - 1;
11784 eptr = SvPV_const(argsv, elen);
11785 if (DO_UTF8(argsv)) {
11786 STRLEN old_precis = precis;
11787 if (has_precis && precis < elen) {
11788 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11789 STRLEN p = precis > ulen ? ulen : precis;
11790 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11791 /* sticks at end */
11793 if (width) { /* fudge width (can't fudge elen) */
11794 if (has_precis && precis < elen)
11795 width += precis - old_precis;
11798 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11805 if (has_precis && precis < elen)
11813 goto floating_point;
11815 if (alt || vectorize)
11817 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11831 goto floating_point;
11838 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11847 esignbuf[esignlen++] = plus;
11851 case 'c': iv = (char)va_arg(*args, int); break;
11852 case 'h': iv = (short)va_arg(*args, int); break;
11853 case 'l': iv = va_arg(*args, long); break;
11854 case 'V': iv = va_arg(*args, IV); break;
11855 case 'z': iv = va_arg(*args, SSize_t); break;
11856 #ifdef HAS_PTRDIFF_T
11857 case 't': iv = va_arg(*args, ptrdiff_t); break;
11859 default: iv = va_arg(*args, int); break;
11861 case 'j': iv = va_arg(*args, intmax_t); break;
11865 iv = va_arg(*args, Quad_t); break;
11872 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11874 case 'c': iv = (char)tiv; break;
11875 case 'h': iv = (short)tiv; break;
11876 case 'l': iv = (long)tiv; break;
11878 default: iv = tiv; break;
11881 iv = (Quad_t)tiv; break;
11887 if ( !vectorize ) /* we already set uv above */
11892 esignbuf[esignlen++] = plus;
11895 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
11896 esignbuf[esignlen++] = '-';
11935 goto floating_point;
11943 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11954 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11955 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11956 case 'l': uv = va_arg(*args, unsigned long); break;
11957 case 'V': uv = va_arg(*args, UV); break;
11958 case 'z': uv = va_arg(*args, Size_t); break;
11959 #ifdef HAS_PTRDIFF_T
11960 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11963 case 'j': uv = va_arg(*args, uintmax_t); break;
11965 default: uv = va_arg(*args, unsigned); break;
11968 uv = va_arg(*args, Uquad_t); break;
11975 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11977 case 'c': uv = (unsigned char)tuv; break;
11978 case 'h': uv = (unsigned short)tuv; break;
11979 case 'l': uv = (unsigned long)tuv; break;
11981 default: uv = tuv; break;
11984 uv = (Uquad_t)tuv; break;
11993 char *ptr = ebuf + sizeof ebuf;
11994 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12000 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12004 } while (uv >>= 4);
12006 esignbuf[esignlen++] = '0';
12007 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12013 *--ptr = '0' + dig;
12014 } while (uv >>= 3);
12015 if (alt && *ptr != '0')
12021 *--ptr = '0' + dig;
12022 } while (uv >>= 1);
12024 esignbuf[esignlen++] = '0';
12025 esignbuf[esignlen++] = c;
12028 default: /* it had better be ten or less */
12031 *--ptr = '0' + dig;
12032 } while (uv /= base);
12035 elen = (ebuf + sizeof ebuf) - ptr;
12039 zeros = precis - elen;
12040 else if (precis == 0 && elen == 1 && *eptr == '0'
12041 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12044 /* a precision nullifies the 0 flag. */
12051 /* FLOATING POINT */
12056 c = 'f'; /* maybe %F isn't supported here */
12058 case 'e': case 'E':
12060 case 'g': case 'G':
12061 case 'a': case 'A':
12065 /* This is evil, but floating point is even more evil */
12067 /* for SV-style calling, we can only get NV
12068 for C-style calling, we assume %f is double;
12069 for simplicity we allow any of %Lf, %llf, %qf for long double
12073 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12077 /* [perl #20339] - we should accept and ignore %lf rather than die */
12081 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12082 intsize = args ? 0 : 'q';
12086 #if defined(HAS_LONG_DOUBLE)
12099 /* Now we need (long double) if intsize == 'q', else (double). */
12101 /* Note: do not pull NVs off the va_list with va_arg()
12102 * (pull doubles instead) because if you have a build
12103 * with long doubles, you would always be pulling long
12104 * doubles, which would badly break anyone using only
12105 * doubles (i.e. the majority of builds). In other
12106 * words, you cannot mix doubles and long doubles.
12107 * The only case where you can pull off long doubles
12108 * is when the format specifier explicitly asks so with
12110 #ifdef USE_QUADMATH
12111 fv = intsize == 'q' ?
12112 va_arg(*args, NV) : va_arg(*args, double);
12114 #elif LONG_DOUBLESIZE > DOUBLESIZE
12115 if (intsize == 'q') {
12116 fv = va_arg(*args, long double);
12119 nv = va_arg(*args, double);
12123 nv = va_arg(*args, double);
12129 if (!infnan) SvGETMAGIC(argsv);
12130 nv = SvNV_nomg(argsv);
12135 /* frexp() (or frexpl) has some unspecified behaviour for
12136 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12137 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12139 (void)Perl_frexp((NV)fv, &i);
12140 if (i == PERL_INT_MIN)
12141 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12142 /* Do not set hexfp earlier since we want to printf
12143 * Inf/NaN for Inf/NaN, not their hexfp. */
12144 hexfp = isALPHA_FOLD_EQ(c, 'a');
12145 if (UNLIKELY(hexfp)) {
12146 /* This seriously overshoots in most cases, but
12147 * better the undershooting. Firstly, all bytes
12148 * of the NV are not mantissa, some of them are
12149 * exponent. Secondly, for the reasonably common
12150 * long doubles case, the "80-bit extended", two
12151 * or six bytes of the NV are unused. */
12153 (fv < 0) ? 1 : 0 + /* possible unary minus */
12155 1 + /* the very unlikely carry */
12158 2 * NVSIZE + /* 2 hexdigits for each byte */
12160 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12162 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12163 /* However, for the "double double", we need more.
12164 * Since each double has their own exponent, the
12165 * doubles may float (haha) rather far from each
12166 * other, and the number of required bits is much
12167 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12168 * See the definition of DOUBLEDOUBLE_MAXBITS.
12170 * Need 2 hexdigits for each byte. */
12171 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12172 /* the size for the exponent already added */
12174 #ifdef USE_LOCALE_NUMERIC
12175 STORE_LC_NUMERIC_SET_TO_NEEDED();
12176 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12177 need += SvLEN(PL_numeric_radix_sv);
12178 RESTORE_LC_NUMERIC();
12182 need = BIT_DIGITS(i);
12183 } /* if i < 0, the number of digits is hard to predict. */
12185 need += has_precis ? precis : 6; /* known default */
12190 #ifdef HAS_LDBL_SPRINTF_BUG
12191 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12192 with sfio - Allen <allens@cpan.org> */
12195 # define MY_DBL_MAX DBL_MAX
12196 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12197 # if DOUBLESIZE >= 8
12198 # define MY_DBL_MAX 1.7976931348623157E+308L
12200 # define MY_DBL_MAX 3.40282347E+38L
12204 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12205 # define MY_DBL_MAX_BUG 1L
12207 # define MY_DBL_MAX_BUG MY_DBL_MAX
12211 # define MY_DBL_MIN DBL_MIN
12212 # else /* XXX guessing! -Allen */
12213 # if DOUBLESIZE >= 8
12214 # define MY_DBL_MIN 2.2250738585072014E-308L
12216 # define MY_DBL_MIN 1.17549435E-38L
12220 if ((intsize == 'q') && (c == 'f') &&
12221 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12222 (need < DBL_DIG)) {
12223 /* it's going to be short enough that
12224 * long double precision is not needed */
12226 if ((fv <= 0L) && (fv >= -0L))
12227 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12229 /* would use Perl_fp_class as a double-check but not
12230 * functional on IRIX - see perl.h comments */
12232 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12233 /* It's within the range that a double can represent */
12234 #if defined(DBL_MAX) && !defined(DBL_MIN)
12235 if ((fv >= ((long double)1/DBL_MAX)) ||
12236 (fv <= (-(long double)1/DBL_MAX)))
12238 fix_ldbl_sprintf_bug = TRUE;
12241 if (fix_ldbl_sprintf_bug == TRUE) {
12251 # undef MY_DBL_MAX_BUG
12254 #endif /* HAS_LDBL_SPRINTF_BUG */
12256 need += 20; /* fudge factor */
12257 if (PL_efloatsize < need) {
12258 Safefree(PL_efloatbuf);
12259 PL_efloatsize = need + 20; /* more fudge */
12260 Newx(PL_efloatbuf, PL_efloatsize, char);
12261 PL_efloatbuf[0] = '\0';
12264 if ( !(width || left || plus || alt) && fill != '0'
12265 && has_precis && intsize != 'q' /* Shortcuts */
12266 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12267 /* See earlier comment about buggy Gconvert when digits,
12269 if ( c == 'g' && precis ) {
12270 STORE_LC_NUMERIC_SET_TO_NEEDED();
12271 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12272 /* May return an empty string for digits==0 */
12273 if (*PL_efloatbuf) {
12274 elen = strlen(PL_efloatbuf);
12275 goto float_converted;
12277 } else if ( c == 'f' && !precis ) {
12278 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12283 if (UNLIKELY(hexfp)) {
12284 /* Hexadecimal floating point. */
12285 char* p = PL_efloatbuf;
12286 U8 vhex[VHEX_SIZE];
12287 U8* v = vhex; /* working pointer to vhex */
12288 U8* vend; /* pointer to one beyond last digit of vhex */
12289 U8* vfnz = NULL; /* first non-zero */
12290 const bool lower = (c == 'a');
12291 /* At output the values of vhex (up to vend) will
12292 * be mapped through the xdig to get the actual
12293 * human-readable xdigits. */
12294 const char* xdig = PL_hexdigit;
12295 int zerotail = 0; /* how many extra zeros to append */
12296 int exponent = 0; /* exponent of the floating point input */
12298 /* XXX: denormals, NaN, Inf.
12300 * For example with denormals, (assuming the vanilla
12301 * 64-bit double): the exponent is zero. 1xp-1074 is
12302 * the smallest denormal and the smallest double, it
12303 * should be output as 0x0.0000000000001p-1022 to
12304 * match its internal structure. */
12306 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
12307 S_hextract(aTHX_ nv, &exponent, vhex, vend);
12309 #if NVSIZE > DOUBLESIZE
12310 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12311 /* In this case there is an implicit bit,
12312 * and therefore the exponent is shifted shift by one. */
12315 /* In this case there is no implicit bit,
12316 * and the exponent is shifted by the first xdigit. */
12331 xdig += 16; /* Use uppercase hex. */
12334 /* Find the first non-zero xdigit. */
12335 for (v = vhex; v < vend; v++) {
12343 U8* vlnz = NULL; /* The last non-zero. */
12345 /* Find the last non-zero xdigit. */
12346 for (v = vend - 1; v >= vhex; v--) {
12353 #if NVSIZE == DOUBLESIZE
12359 if ((SSize_t)(precis + 1) < vend - vhex) {
12362 v = vhex + precis + 1;
12363 /* Round away from zero: if the tail
12364 * beyond the precis xdigits is equal to
12365 * or greater than 0x8000... */
12367 if (!round && *v == 0x8) {
12368 for (v++; v < vend; v++) {
12376 for (v = vhex + precis; v >= vhex; v--) {
12383 /* If the carry goes all the way to
12384 * the front, we need to output
12385 * a single '1'. This goes against
12386 * the "xdigit and then radix"
12387 * but since this is "cannot happen"
12388 * category, that is probably good. */
12393 /* The new effective "last non zero". */
12394 vlnz = vhex + precis;
12397 zerotail = precis - (vlnz - vhex);
12404 /* The radix is always output after the first
12405 * non-zero xdigit, or if alt. */
12406 if (vfnz < vlnz || alt) {
12407 #ifndef USE_LOCALE_NUMERIC
12410 STORE_LC_NUMERIC_SET_TO_NEEDED();
12411 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12413 const char* r = SvPV(PL_numeric_radix_sv, n);
12414 Copy(r, p, n, char);
12420 RESTORE_LC_NUMERIC();
12435 elen = p - PL_efloatbuf;
12436 elen += my_snprintf(p, PL_efloatsize - elen,
12437 "%c%+d", lower ? 'p' : 'P',
12440 if (elen < width) {
12442 /* Pad the back with spaces. */
12443 memset(PL_efloatbuf + elen, ' ', width - elen);
12445 else if (fill == '0') {
12446 /* Insert the zeros between the "0x" and
12447 * the digits, otherwise we end up with
12449 STRLEN nzero = width - elen;
12450 char* zerox = PL_efloatbuf + 2;
12451 Move(zerox, zerox + nzero, elen - 2, char);
12452 memset(zerox, fill, nzero);
12455 /* Move it to the right. */
12456 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12458 /* Pad the front with spaces. */
12459 memset(PL_efloatbuf, ' ', width - elen);
12465 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12467 /* Not affecting infnan output: precision, alt, fill. */
12468 if (elen < width) {
12470 /* Pack the back with spaces. */
12471 memset(PL_efloatbuf + elen, ' ', width - elen);
12473 /* Move it to the right. */
12474 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12476 /* Pad the front with spaces. */
12477 memset(PL_efloatbuf, ' ', width - elen);
12485 char *ptr = ebuf + sizeof ebuf;
12488 #if defined(USE_QUADMATH)
12489 if (intsize == 'q') {
12493 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12494 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12495 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12496 * not USE_LONG_DOUBLE and NVff. In other words,
12497 * this needs to work without USE_LONG_DOUBLE. */
12498 if (intsize == 'q') {
12499 /* Copy the one or more characters in a long double
12500 * format before the 'base' ([efgEFG]) character to
12501 * the format string. */
12502 static char const ldblf[] = PERL_PRIfldbl;
12503 char const *p = ldblf + sizeof(ldblf) - 3;
12504 while (p >= ldblf) { *--ptr = *p--; }
12509 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12514 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12526 /* No taint. Otherwise we are in the strange situation
12527 * where printf() taints but print($float) doesn't.
12530 STORE_LC_NUMERIC_SET_TO_NEEDED();
12532 /* hopefully the above makes ptr a very constrained format
12533 * that is safe to use, even though it's not literal */
12534 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12535 #ifdef USE_QUADMATH
12537 const char* qfmt = quadmath_format_single(ptr);
12539 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12540 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12542 if ((IV)elen == -1)
12543 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12547 #elif defined(HAS_LONG_DOUBLE)
12548 elen = ((intsize == 'q')
12549 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12550 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12552 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12558 eptr = PL_efloatbuf;
12559 assert((IV)elen > 0); /* here zero elen is bad */
12561 #ifdef USE_LOCALE_NUMERIC
12562 /* If the decimal point character in the string is UTF-8, make the
12564 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12565 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12578 i = SvCUR(sv) - origlen;
12581 case 'c': *(va_arg(*args, char*)) = i; break;
12582 case 'h': *(va_arg(*args, short*)) = i; break;
12583 default: *(va_arg(*args, int*)) = i; break;
12584 case 'l': *(va_arg(*args, long*)) = i; break;
12585 case 'V': *(va_arg(*args, IV*)) = i; break;
12586 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12587 #ifdef HAS_PTRDIFF_T
12588 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12591 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12595 *(va_arg(*args, Quad_t*)) = i; break;
12602 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12603 continue; /* not "break" */
12610 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12611 && ckWARN(WARN_PRINTF))
12613 SV * const msg = sv_newmortal();
12614 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12615 (PL_op->op_type == OP_PRTF) ? "" : "s");
12616 if (fmtstart < patend) {
12617 const char * const fmtend = q < patend ? q : patend;
12619 sv_catpvs(msg, "\"%");
12620 for (f = fmtstart; f < fmtend; f++) {
12622 sv_catpvn_nomg(msg, f, 1);
12624 Perl_sv_catpvf(aTHX_ msg,
12625 "\\%03"UVof, (UV)*f & 0xFF);
12628 sv_catpvs(msg, "\"");
12630 sv_catpvs(msg, "end of string");
12632 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12635 /* output mangled stuff ... */
12641 /* ... right here, because formatting flags should not apply */
12642 SvGROW(sv, SvCUR(sv) + elen + 1);
12644 Copy(eptr, p, elen, char);
12647 SvCUR_set(sv, p - SvPVX_const(sv));
12649 continue; /* not "break" */
12652 if (is_utf8 != has_utf8) {
12655 sv_utf8_upgrade(sv);
12658 const STRLEN old_elen = elen;
12659 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12660 sv_utf8_upgrade(nsv);
12661 eptr = SvPVX_const(nsv);
12664 if (width) { /* fudge width (can't fudge elen) */
12665 width += elen - old_elen;
12671 assert((IV)elen >= 0); /* here zero elen is fine */
12672 have = esignlen + zeros + elen;
12674 croak_memory_wrap();
12676 need = (have > width ? have : width);
12679 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12680 croak_memory_wrap();
12681 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12683 if (esignlen && fill == '0') {
12685 for (i = 0; i < (int)esignlen; i++)
12686 *p++ = esignbuf[i];
12688 if (gap && !left) {
12689 memset(p, fill, gap);
12692 if (esignlen && fill != '0') {
12694 for (i = 0; i < (int)esignlen; i++)
12695 *p++ = esignbuf[i];
12699 for (i = zeros; i; i--)
12703 Copy(eptr, p, elen, char);
12707 memset(p, ' ', gap);
12712 Copy(dotstr, p, dotstrlen, char);
12716 vectorize = FALSE; /* done iterating over vecstr */
12723 SvCUR_set(sv, p - SvPVX_const(sv));
12730 /* Now that we've consumed all our printf format arguments (svix)
12731 * do we have things left on the stack that we didn't use?
12733 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12734 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12735 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12740 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12744 /* =========================================================================
12746 =head1 Cloning an interpreter
12750 All the macros and functions in this section are for the private use of
12751 the main function, perl_clone().
12753 The foo_dup() functions make an exact copy of an existing foo thingy.
12754 During the course of a cloning, a hash table is used to map old addresses
12755 to new addresses. The table is created and manipulated with the
12756 ptr_table_* functions.
12758 * =========================================================================*/
12761 #if defined(USE_ITHREADS)
12763 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12764 #ifndef GpREFCNT_inc
12765 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12769 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12770 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12771 If this changes, please unmerge ss_dup.
12772 Likewise, sv_dup_inc_multiple() relies on this fact. */
12773 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12774 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12775 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12776 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12777 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12778 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12779 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12780 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12781 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12782 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12783 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12784 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12785 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12787 /* clone a parser */
12790 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12794 PERL_ARGS_ASSERT_PARSER_DUP;
12799 /* look for it in the table first */
12800 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12804 /* create anew and remember what it is */
12805 Newxz(parser, 1, yy_parser);
12806 ptr_table_store(PL_ptr_table, proto, parser);
12808 /* XXX these not yet duped */
12809 parser->old_parser = NULL;
12810 parser->stack = NULL;
12812 parser->stack_size = 0;
12813 /* XXX parser->stack->state = 0; */
12815 /* XXX eventually, just Copy() most of the parser struct ? */
12817 parser->lex_brackets = proto->lex_brackets;
12818 parser->lex_casemods = proto->lex_casemods;
12819 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12820 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12821 parser->lex_casestack = savepvn(proto->lex_casestack,
12822 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12823 parser->lex_defer = proto->lex_defer;
12824 parser->lex_dojoin = proto->lex_dojoin;
12825 parser->lex_formbrack = proto->lex_formbrack;
12826 parser->lex_inpat = proto->lex_inpat;
12827 parser->lex_inwhat = proto->lex_inwhat;
12828 parser->lex_op = proto->lex_op;
12829 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12830 parser->lex_starts = proto->lex_starts;
12831 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12832 parser->multi_close = proto->multi_close;
12833 parser->multi_open = proto->multi_open;
12834 parser->multi_start = proto->multi_start;
12835 parser->multi_end = proto->multi_end;
12836 parser->preambled = proto->preambled;
12837 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12838 parser->linestr = sv_dup_inc(proto->linestr, param);
12839 parser->expect = proto->expect;
12840 parser->copline = proto->copline;
12841 parser->last_lop_op = proto->last_lop_op;
12842 parser->lex_state = proto->lex_state;
12843 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12844 /* rsfp_filters entries have fake IoDIRP() */
12845 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12846 parser->in_my = proto->in_my;
12847 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12848 parser->error_count = proto->error_count;
12851 parser->linestr = sv_dup_inc(proto->linestr, param);
12854 char * const ols = SvPVX(proto->linestr);
12855 char * const ls = SvPVX(parser->linestr);
12857 parser->bufptr = ls + (proto->bufptr >= ols ?
12858 proto->bufptr - ols : 0);
12859 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12860 proto->oldbufptr - ols : 0);
12861 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12862 proto->oldoldbufptr - ols : 0);
12863 parser->linestart = ls + (proto->linestart >= ols ?
12864 proto->linestart - ols : 0);
12865 parser->last_uni = ls + (proto->last_uni >= ols ?
12866 proto->last_uni - ols : 0);
12867 parser->last_lop = ls + (proto->last_lop >= ols ?
12868 proto->last_lop - ols : 0);
12870 parser->bufend = ls + SvCUR(parser->linestr);
12873 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12876 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12877 Copy(proto->nexttype, parser->nexttype, 5, I32);
12878 parser->nexttoke = proto->nexttoke;
12880 /* XXX should clone saved_curcop here, but we aren't passed
12881 * proto_perl; so do it in perl_clone_using instead */
12887 /* duplicate a file handle */
12890 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12894 PERL_ARGS_ASSERT_FP_DUP;
12895 PERL_UNUSED_ARG(type);
12898 return (PerlIO*)NULL;
12900 /* look for it in the table first */
12901 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12905 /* create anew and remember what it is */
12906 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12907 ptr_table_store(PL_ptr_table, fp, ret);
12911 /* duplicate a directory handle */
12914 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12918 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12920 const Direntry_t *dirent;
12921 char smallbuf[256];
12927 PERL_UNUSED_CONTEXT;
12928 PERL_ARGS_ASSERT_DIRP_DUP;
12933 /* look for it in the table first */
12934 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12938 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12940 PERL_UNUSED_ARG(param);
12944 /* open the current directory (so we can switch back) */
12945 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12947 /* chdir to our dir handle and open the present working directory */
12948 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12949 PerlDir_close(pwd);
12950 return (DIR *)NULL;
12952 /* Now we should have two dir handles pointing to the same dir. */
12954 /* Be nice to the calling code and chdir back to where we were. */
12955 /* XXX If this fails, then what? */
12956 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12958 /* We have no need of the pwd handle any more. */
12959 PerlDir_close(pwd);
12962 # define d_namlen(d) (d)->d_namlen
12964 # define d_namlen(d) strlen((d)->d_name)
12966 /* Iterate once through dp, to get the file name at the current posi-
12967 tion. Then step back. */
12968 pos = PerlDir_tell(dp);
12969 if ((dirent = PerlDir_read(dp))) {
12970 len = d_namlen(dirent);
12971 if (len <= sizeof smallbuf) name = smallbuf;
12972 else Newx(name, len, char);
12973 Move(dirent->d_name, name, len, char);
12975 PerlDir_seek(dp, pos);
12977 /* Iterate through the new dir handle, till we find a file with the
12979 if (!dirent) /* just before the end */
12981 pos = PerlDir_tell(ret);
12982 if (PerlDir_read(ret)) continue; /* not there yet */
12983 PerlDir_seek(ret, pos); /* step back */
12987 const long pos0 = PerlDir_tell(ret);
12989 pos = PerlDir_tell(ret);
12990 if ((dirent = PerlDir_read(ret))) {
12991 if (len == (STRLEN)d_namlen(dirent)
12992 && memEQ(name, dirent->d_name, len)) {
12994 PerlDir_seek(ret, pos); /* step back */
12997 /* else we are not there yet; keep iterating */
12999 else { /* This is not meant to happen. The best we can do is
13000 reset the iterator to the beginning. */
13001 PerlDir_seek(ret, pos0);
13008 if (name && name != smallbuf)
13013 ret = win32_dirp_dup(dp, param);
13016 /* pop it in the pointer table */
13018 ptr_table_store(PL_ptr_table, dp, ret);
13023 /* duplicate a typeglob */
13026 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13030 PERL_ARGS_ASSERT_GP_DUP;
13034 /* look for it in the table first */
13035 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13039 /* create anew and remember what it is */
13041 ptr_table_store(PL_ptr_table, gp, ret);
13044 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13045 on Newxz() to do this for us. */
13046 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13047 ret->gp_io = io_dup_inc(gp->gp_io, param);
13048 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13049 ret->gp_av = av_dup_inc(gp->gp_av, param);
13050 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13051 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13052 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13053 ret->gp_cvgen = gp->gp_cvgen;
13054 ret->gp_line = gp->gp_line;
13055 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13059 /* duplicate a chain of magic */
13062 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13064 MAGIC *mgret = NULL;
13065 MAGIC **mgprev_p = &mgret;
13067 PERL_ARGS_ASSERT_MG_DUP;
13069 for (; mg; mg = mg->mg_moremagic) {
13072 if ((param->flags & CLONEf_JOIN_IN)
13073 && mg->mg_type == PERL_MAGIC_backref)
13074 /* when joining, we let the individual SVs add themselves to
13075 * backref as needed. */
13078 Newx(nmg, 1, MAGIC);
13080 mgprev_p = &(nmg->mg_moremagic);
13082 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13083 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13084 from the original commit adding Perl_mg_dup() - revision 4538.
13085 Similarly there is the annotation "XXX random ptr?" next to the
13086 assignment to nmg->mg_ptr. */
13089 /* FIXME for plugins
13090 if (nmg->mg_type == PERL_MAGIC_qr) {
13091 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13095 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13096 ? nmg->mg_type == PERL_MAGIC_backref
13097 /* The backref AV has its reference
13098 * count deliberately bumped by 1 */
13099 ? SvREFCNT_inc(av_dup_inc((const AV *)
13100 nmg->mg_obj, param))
13101 : sv_dup_inc(nmg->mg_obj, param)
13102 : sv_dup(nmg->mg_obj, param);
13104 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13105 if (nmg->mg_len > 0) {
13106 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13107 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13108 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13110 AMT * const namtp = (AMT*)nmg->mg_ptr;
13111 sv_dup_inc_multiple((SV**)(namtp->table),
13112 (SV**)(namtp->table), NofAMmeth, param);
13115 else if (nmg->mg_len == HEf_SVKEY)
13116 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13118 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13119 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13125 #endif /* USE_ITHREADS */
13127 struct ptr_tbl_arena {
13128 struct ptr_tbl_arena *next;
13129 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13132 /* create a new pointer-mapping table */
13135 Perl_ptr_table_new(pTHX)
13138 PERL_UNUSED_CONTEXT;
13140 Newx(tbl, 1, PTR_TBL_t);
13141 tbl->tbl_max = 511;
13142 tbl->tbl_items = 0;
13143 tbl->tbl_arena = NULL;
13144 tbl->tbl_arena_next = NULL;
13145 tbl->tbl_arena_end = NULL;
13146 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13150 #define PTR_TABLE_HASH(ptr) \
13151 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13153 /* map an existing pointer using a table */
13155 STATIC PTR_TBL_ENT_t *
13156 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13158 PTR_TBL_ENT_t *tblent;
13159 const UV hash = PTR_TABLE_HASH(sv);
13161 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13163 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13164 for (; tblent; tblent = tblent->next) {
13165 if (tblent->oldval == sv)
13172 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13174 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13176 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13177 PERL_UNUSED_CONTEXT;
13179 return tblent ? tblent->newval : NULL;
13182 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13183 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13184 * the core's typical use of ptr_tables in thread cloning. */
13187 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13189 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13191 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13192 PERL_UNUSED_CONTEXT;
13195 tblent->newval = newsv;
13197 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13199 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13200 struct ptr_tbl_arena *new_arena;
13202 Newx(new_arena, 1, struct ptr_tbl_arena);
13203 new_arena->next = tbl->tbl_arena;
13204 tbl->tbl_arena = new_arena;
13205 tbl->tbl_arena_next = new_arena->array;
13206 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13209 tblent = tbl->tbl_arena_next++;
13211 tblent->oldval = oldsv;
13212 tblent->newval = newsv;
13213 tblent->next = tbl->tbl_ary[entry];
13214 tbl->tbl_ary[entry] = tblent;
13216 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13217 ptr_table_split(tbl);
13221 /* double the hash bucket size of an existing ptr table */
13224 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13226 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13227 const UV oldsize = tbl->tbl_max + 1;
13228 UV newsize = oldsize * 2;
13231 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13232 PERL_UNUSED_CONTEXT;
13234 Renew(ary, newsize, PTR_TBL_ENT_t*);
13235 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13236 tbl->tbl_max = --newsize;
13237 tbl->tbl_ary = ary;
13238 for (i=0; i < oldsize; i++, ary++) {
13239 PTR_TBL_ENT_t **entp = ary;
13240 PTR_TBL_ENT_t *ent = *ary;
13241 PTR_TBL_ENT_t **curentp;
13244 curentp = ary + oldsize;
13246 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13248 ent->next = *curentp;
13258 /* remove all the entries from a ptr table */
13259 /* Deprecated - will be removed post 5.14 */
13262 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13264 PERL_UNUSED_CONTEXT;
13265 if (tbl && tbl->tbl_items) {
13266 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13268 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
13271 struct ptr_tbl_arena *next = arena->next;
13277 tbl->tbl_items = 0;
13278 tbl->tbl_arena = NULL;
13279 tbl->tbl_arena_next = NULL;
13280 tbl->tbl_arena_end = NULL;
13284 /* clear and free a ptr table */
13287 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13289 struct ptr_tbl_arena *arena;
13291 PERL_UNUSED_CONTEXT;
13297 arena = tbl->tbl_arena;
13300 struct ptr_tbl_arena *next = arena->next;
13306 Safefree(tbl->tbl_ary);
13310 #if defined(USE_ITHREADS)
13313 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13315 PERL_ARGS_ASSERT_RVPV_DUP;
13317 assert(!isREGEXP(sstr));
13319 if (SvWEAKREF(sstr)) {
13320 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13321 if (param->flags & CLONEf_JOIN_IN) {
13322 /* if joining, we add any back references individually rather
13323 * than copying the whole backref array */
13324 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13328 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13330 else if (SvPVX_const(sstr)) {
13331 /* Has something there */
13333 /* Normal PV - clone whole allocated space */
13334 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13335 /* sstr may not be that normal, but actually copy on write.
13336 But we are a true, independent SV, so: */
13340 /* Special case - not normally malloced for some reason */
13341 if (isGV_with_GP(sstr)) {
13342 /* Don't need to do anything here. */
13344 else if ((SvIsCOW(sstr))) {
13345 /* A "shared" PV - clone it as "shared" PV */
13347 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13351 /* Some other special case - random pointer */
13352 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13357 /* Copy the NULL */
13358 SvPV_set(dstr, NULL);
13362 /* duplicate a list of SVs. source and dest may point to the same memory. */
13364 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13365 SSize_t items, CLONE_PARAMS *const param)
13367 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13369 while (items-- > 0) {
13370 *dest++ = sv_dup_inc(*source++, param);
13376 /* duplicate an SV of any type (including AV, HV etc) */
13379 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13384 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13386 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13387 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13392 /* look for it in the table first */
13393 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13397 if(param->flags & CLONEf_JOIN_IN) {
13398 /** We are joining here so we don't want do clone
13399 something that is bad **/
13400 if (SvTYPE(sstr) == SVt_PVHV) {
13401 const HEK * const hvname = HvNAME_HEK(sstr);
13403 /** don't clone stashes if they already exist **/
13404 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13405 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13406 ptr_table_store(PL_ptr_table, sstr, dstr);
13410 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13411 HV *stash = GvSTASH(sstr);
13412 const HEK * hvname;
13413 if (stash && (hvname = HvNAME_HEK(stash))) {
13414 /** don't clone GVs if they already exist **/
13416 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13417 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13419 stash, GvNAME(sstr),
13425 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13426 ptr_table_store(PL_ptr_table, sstr, *svp);
13433 /* create anew and remember what it is */
13436 #ifdef DEBUG_LEAKING_SCALARS
13437 dstr->sv_debug_optype = sstr->sv_debug_optype;
13438 dstr->sv_debug_line = sstr->sv_debug_line;
13439 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13440 dstr->sv_debug_parent = (SV*)sstr;
13441 FREE_SV_DEBUG_FILE(dstr);
13442 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13445 ptr_table_store(PL_ptr_table, sstr, dstr);
13448 SvFLAGS(dstr) = SvFLAGS(sstr);
13449 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13450 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13453 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13454 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13455 (void*)PL_watch_pvx, SvPVX_const(sstr));
13458 /* don't clone objects whose class has asked us not to */
13460 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13466 switch (SvTYPE(sstr)) {
13468 SvANY(dstr) = NULL;
13471 SET_SVANY_FOR_BODYLESS_IV(dstr);
13473 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13475 SvIV_set(dstr, SvIVX(sstr));
13479 #if NVSIZE <= IVSIZE
13480 SET_SVANY_FOR_BODYLESS_NV(dstr);
13482 SvANY(dstr) = new_XNV();
13484 SvNV_set(dstr, SvNVX(sstr));
13488 /* These are all the types that need complex bodies allocating. */
13490 const svtype sv_type = SvTYPE(sstr);
13491 const struct body_details *const sv_type_details
13492 = bodies_by_type + sv_type;
13496 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13512 assert(sv_type_details->body_size);
13513 if (sv_type_details->arena) {
13514 new_body_inline(new_body, sv_type);
13516 = (void*)((char*)new_body - sv_type_details->offset);
13518 new_body = new_NOARENA(sv_type_details);
13522 SvANY(dstr) = new_body;
13525 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13526 ((char*)SvANY(dstr)) + sv_type_details->offset,
13527 sv_type_details->copy, char);
13529 Copy(((char*)SvANY(sstr)),
13530 ((char*)SvANY(dstr)),
13531 sv_type_details->body_size + sv_type_details->offset, char);
13534 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13535 && !isGV_with_GP(dstr)
13537 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13538 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13540 /* The Copy above means that all the source (unduplicated) pointers
13541 are now in the destination. We can check the flags and the
13542 pointers in either, but it's possible that there's less cache
13543 missing by always going for the destination.
13544 FIXME - instrument and check that assumption */
13545 if (sv_type >= SVt_PVMG) {
13547 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13548 if (SvOBJECT(dstr) && SvSTASH(dstr))
13549 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13550 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13553 /* The cast silences a GCC warning about unhandled types. */
13554 switch ((int)sv_type) {
13565 /* FIXME for plugins */
13566 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13567 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13570 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13571 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13572 LvTARG(dstr) = dstr;
13573 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13574 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13576 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13577 if (isREGEXP(sstr)) goto duprex;
13579 /* non-GP case already handled above */
13580 if(isGV_with_GP(sstr)) {
13581 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13582 /* Don't call sv_add_backref here as it's going to be
13583 created as part of the magic cloning of the symbol
13584 table--unless this is during a join and the stash
13585 is not actually being cloned. */
13586 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13587 at the point of this comment. */
13588 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13589 if (param->flags & CLONEf_JOIN_IN)
13590 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13591 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13592 (void)GpREFCNT_inc(GvGP(dstr));
13596 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13597 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13598 /* I have no idea why fake dirp (rsfps)
13599 should be treated differently but otherwise
13600 we end up with leaks -- sky*/
13601 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13602 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13603 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13605 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13606 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13607 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13608 if (IoDIRP(dstr)) {
13609 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13612 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13614 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13616 if (IoOFP(dstr) == IoIFP(sstr))
13617 IoOFP(dstr) = IoIFP(dstr);
13619 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13620 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13621 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13622 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13625 /* avoid cloning an empty array */
13626 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13627 SV **dst_ary, **src_ary;
13628 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13630 src_ary = AvARRAY((const AV *)sstr);
13631 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13632 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13633 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13634 AvALLOC((const AV *)dstr) = dst_ary;
13635 if (AvREAL((const AV *)sstr)) {
13636 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13640 while (items-- > 0)
13641 *dst_ary++ = sv_dup(*src_ary++, param);
13643 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13644 while (items-- > 0) {
13645 *dst_ary++ = &PL_sv_undef;
13649 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13650 AvALLOC((const AV *)dstr) = (SV**)NULL;
13651 AvMAX( (const AV *)dstr) = -1;
13652 AvFILLp((const AV *)dstr) = -1;
13656 if (HvARRAY((const HV *)sstr)) {
13658 const bool sharekeys = !!HvSHAREKEYS(sstr);
13659 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13660 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13662 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13663 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13665 HvARRAY(dstr) = (HE**)darray;
13666 while (i <= sxhv->xhv_max) {
13667 const HE * const source = HvARRAY(sstr)[i];
13668 HvARRAY(dstr)[i] = source
13669 ? he_dup(source, sharekeys, param) : 0;
13673 const struct xpvhv_aux * const saux = HvAUX(sstr);
13674 struct xpvhv_aux * const daux = HvAUX(dstr);
13675 /* This flag isn't copied. */
13678 if (saux->xhv_name_count) {
13679 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13681 = saux->xhv_name_count < 0
13682 ? -saux->xhv_name_count
13683 : saux->xhv_name_count;
13684 HEK **shekp = sname + count;
13686 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13687 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13688 while (shekp-- > sname) {
13690 *dhekp = hek_dup(*shekp, param);
13694 daux->xhv_name_u.xhvnameu_name
13695 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13698 daux->xhv_name_count = saux->xhv_name_count;
13700 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13701 daux->xhv_aux_flags = saux->xhv_aux_flags;
13702 #ifdef PERL_HASH_RANDOMIZE_KEYS
13703 daux->xhv_rand = saux->xhv_rand;
13704 daux->xhv_last_rand = saux->xhv_last_rand;
13706 daux->xhv_riter = saux->xhv_riter;
13707 daux->xhv_eiter = saux->xhv_eiter
13708 ? he_dup(saux->xhv_eiter,
13709 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13710 /* backref array needs refcnt=2; see sv_add_backref */
13711 daux->xhv_backreferences =
13712 (param->flags & CLONEf_JOIN_IN)
13713 /* when joining, we let the individual GVs and
13714 * CVs add themselves to backref as
13715 * needed. This avoids pulling in stuff
13716 * that isn't required, and simplifies the
13717 * case where stashes aren't cloned back
13718 * if they already exist in the parent
13721 : saux->xhv_backreferences
13722 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13723 ? MUTABLE_AV(SvREFCNT_inc(
13724 sv_dup_inc((const SV *)
13725 saux->xhv_backreferences, param)))
13726 : MUTABLE_AV(sv_dup((const SV *)
13727 saux->xhv_backreferences, param))
13730 daux->xhv_mro_meta = saux->xhv_mro_meta
13731 ? mro_meta_dup(saux->xhv_mro_meta, param)
13734 /* Record stashes for possible cloning in Perl_clone(). */
13736 av_push(param->stashes, dstr);
13740 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13743 if (!(param->flags & CLONEf_COPY_STACKS)) {
13748 /* NOTE: not refcounted */
13749 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13750 hv_dup(CvSTASH(dstr), param);
13751 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13752 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13753 if (!CvISXSUB(dstr)) {
13755 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13757 CvSLABBED_off(dstr);
13758 } else if (CvCONST(dstr)) {
13759 CvXSUBANY(dstr).any_ptr =
13760 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13762 assert(!CvSLABBED(dstr));
13763 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13765 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13766 hek_dup(CvNAME_HEK((CV *)sstr), param);
13767 /* don't dup if copying back - CvGV isn't refcounted, so the
13768 * duped GV may never be freed. A bit of a hack! DAPM */
13770 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13772 ? gv_dup_inc(CvGV(sstr), param)
13773 : (param->flags & CLONEf_JOIN_IN)
13775 : gv_dup(CvGV(sstr), param);
13777 if (!CvISXSUB(sstr)) {
13778 PADLIST * padlist = CvPADLIST(sstr);
13780 padlist = padlist_dup(padlist, param);
13781 CvPADLIST_set(dstr, padlist);
13783 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13784 PoisonPADLIST(dstr);
13787 CvWEAKOUTSIDE(sstr)
13788 ? cv_dup( CvOUTSIDE(dstr), param)
13789 : cv_dup_inc(CvOUTSIDE(dstr), param);
13799 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13801 PERL_ARGS_ASSERT_SV_DUP_INC;
13802 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13806 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13808 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13809 PERL_ARGS_ASSERT_SV_DUP;
13811 /* Track every SV that (at least initially) had a reference count of 0.
13812 We need to do this by holding an actual reference to it in this array.
13813 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13814 (akin to the stashes hash, and the perl stack), we come unstuck if
13815 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13816 thread) is manipulated in a CLONE method, because CLONE runs before the
13817 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13818 (and fix things up by giving each a reference via the temps stack).
13819 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13820 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13821 before the walk of unreferenced happens and a reference to that is SV
13822 added to the temps stack. At which point we have the same SV considered
13823 to be in use, and free to be re-used. Not good.
13825 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13826 assert(param->unreferenced);
13827 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13833 /* duplicate a context */
13836 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13838 PERL_CONTEXT *ncxs;
13840 PERL_ARGS_ASSERT_CX_DUP;
13843 return (PERL_CONTEXT*)NULL;
13845 /* look for it in the table first */
13846 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13850 /* create anew and remember what it is */
13851 Newx(ncxs, max + 1, PERL_CONTEXT);
13852 ptr_table_store(PL_ptr_table, cxs, ncxs);
13853 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13856 PERL_CONTEXT * const ncx = &ncxs[ix];
13857 if (CxTYPE(ncx) == CXt_SUBST) {
13858 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13861 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13862 switch (CxTYPE(ncx)) {
13864 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13865 ? cv_dup_inc(ncx->blk_sub.cv, param)
13866 : cv_dup(ncx->blk_sub.cv,param));
13867 if(CxHASARGS(ncx)){
13868 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13869 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13871 ncx->blk_sub.argarray = NULL;
13872 ncx->blk_sub.savearray = NULL;
13874 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13875 ncx->blk_sub.oldcomppad);
13878 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13880 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13881 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13883 case CXt_LOOP_LAZYSV:
13884 ncx->blk_loop.state_u.lazysv.end
13885 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13886 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
13887 duplication code instead.
13888 We are taking advantage of (1) av_dup_inc and sv_dup_inc
13889 actually being the same function, and (2) order
13890 equivalence of the two unions.
13891 We can assert the later [but only at run time :-(] */
13892 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13893 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13896 ncx->blk_loop.state_u.ary.ary
13897 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13899 case CXt_LOOP_LAZYIV:
13900 case CXt_LOOP_PLAIN:
13901 /* code common to all CXt_LOOP_* types */
13902 if (CxPADLOOP(ncx)) {
13903 ncx->blk_loop.itervar_u.oldcomppad
13904 = (PAD*)ptr_table_fetch(PL_ptr_table,
13905 ncx->blk_loop.itervar_u.oldcomppad);
13907 ncx->blk_loop.itervar_u.gv
13908 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13913 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13914 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13915 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13930 /* duplicate a stack info structure */
13933 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13937 PERL_ARGS_ASSERT_SI_DUP;
13940 return (PERL_SI*)NULL;
13942 /* look for it in the table first */
13943 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13947 /* create anew and remember what it is */
13948 Newxz(nsi, 1, PERL_SI);
13949 ptr_table_store(PL_ptr_table, si, nsi);
13951 nsi->si_stack = av_dup_inc(si->si_stack, param);
13952 nsi->si_cxix = si->si_cxix;
13953 nsi->si_cxmax = si->si_cxmax;
13954 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13955 nsi->si_type = si->si_type;
13956 nsi->si_prev = si_dup(si->si_prev, param);
13957 nsi->si_next = si_dup(si->si_next, param);
13958 nsi->si_markoff = si->si_markoff;
13963 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13964 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13965 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13966 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13967 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13968 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13969 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13970 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13971 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13972 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13973 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13974 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13975 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13976 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13977 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13978 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13981 #define pv_dup_inc(p) SAVEPV(p)
13982 #define pv_dup(p) SAVEPV(p)
13983 #define svp_dup_inc(p,pp) any_dup(p,pp)
13985 /* map any object to the new equivent - either something in the
13986 * ptr table, or something in the interpreter structure
13990 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13994 PERL_ARGS_ASSERT_ANY_DUP;
13997 return (void*)NULL;
13999 /* look for it in the table first */
14000 ret = ptr_table_fetch(PL_ptr_table, v);
14004 /* see if it is part of the interpreter structure */
14005 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14006 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14014 /* duplicate the save stack */
14017 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14020 ANY * const ss = proto_perl->Isavestack;
14021 const I32 max = proto_perl->Isavestack_max;
14022 I32 ix = proto_perl->Isavestack_ix;
14035 void (*dptr) (void*);
14036 void (*dxptr) (pTHX_ void*);
14038 PERL_ARGS_ASSERT_SS_DUP;
14040 Newxz(nss, max, ANY);
14043 const UV uv = POPUV(ss,ix);
14044 const U8 type = (U8)uv & SAVE_MASK;
14046 TOPUV(nss,ix) = uv;
14048 case SAVEt_CLEARSV:
14049 case SAVEt_CLEARPADRANGE:
14051 case SAVEt_HELEM: /* hash element */
14052 case SAVEt_SV: /* scalar reference */
14053 sv = (const SV *)POPPTR(ss,ix);
14054 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14056 case SAVEt_ITEM: /* normal string */
14057 case SAVEt_GVSV: /* scalar slot in GV */
14058 sv = (const SV *)POPPTR(ss,ix);
14059 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14060 if (type == SAVEt_SV)
14064 case SAVEt_MORTALIZESV:
14065 case SAVEt_READONLY_OFF:
14066 sv = (const SV *)POPPTR(ss,ix);
14067 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14069 case SAVEt_FREEPADNAME:
14070 ptr = POPPTR(ss,ix);
14071 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14072 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14074 case SAVEt_SHARED_PVREF: /* char* in shared space */
14075 c = (char*)POPPTR(ss,ix);
14076 TOPPTR(nss,ix) = savesharedpv(c);
14077 ptr = POPPTR(ss,ix);
14078 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14080 case SAVEt_GENERIC_SVREF: /* generic sv */
14081 case SAVEt_SVREF: /* scalar reference */
14082 sv = (const SV *)POPPTR(ss,ix);
14083 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14084 if (type == SAVEt_SVREF)
14085 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14086 ptr = POPPTR(ss,ix);
14087 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14089 case SAVEt_GVSLOT: /* any slot in GV */
14090 sv = (const SV *)POPPTR(ss,ix);
14091 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14092 ptr = POPPTR(ss,ix);
14093 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14094 sv = (const SV *)POPPTR(ss,ix);
14095 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14097 case SAVEt_HV: /* hash reference */
14098 case SAVEt_AV: /* array reference */
14099 sv = (const SV *) POPPTR(ss,ix);
14100 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14102 case SAVEt_COMPPAD:
14104 sv = (const SV *) POPPTR(ss,ix);
14105 TOPPTR(nss,ix) = sv_dup(sv, param);
14107 case SAVEt_INT: /* int reference */
14108 ptr = POPPTR(ss,ix);
14109 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14110 intval = (int)POPINT(ss,ix);
14111 TOPINT(nss,ix) = intval;
14113 case SAVEt_LONG: /* long reference */
14114 ptr = POPPTR(ss,ix);
14115 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14116 longval = (long)POPLONG(ss,ix);
14117 TOPLONG(nss,ix) = longval;
14119 case SAVEt_I32: /* I32 reference */
14120 ptr = POPPTR(ss,ix);
14121 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14123 TOPINT(nss,ix) = i;
14125 case SAVEt_IV: /* IV reference */
14126 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14127 ptr = POPPTR(ss,ix);
14128 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14130 TOPIV(nss,ix) = iv;
14132 case SAVEt_HPTR: /* HV* reference */
14133 case SAVEt_APTR: /* AV* reference */
14134 case SAVEt_SPTR: /* SV* reference */
14135 ptr = POPPTR(ss,ix);
14136 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14137 sv = (const SV *)POPPTR(ss,ix);
14138 TOPPTR(nss,ix) = sv_dup(sv, param);
14140 case SAVEt_VPTR: /* random* reference */
14141 ptr = POPPTR(ss,ix);
14142 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14144 case SAVEt_INT_SMALL:
14145 case SAVEt_I32_SMALL:
14146 case SAVEt_I16: /* I16 reference */
14147 case SAVEt_I8: /* I8 reference */
14149 ptr = POPPTR(ss,ix);
14150 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14152 case SAVEt_GENERIC_PVREF: /* generic char* */
14153 case SAVEt_PPTR: /* char* reference */
14154 ptr = POPPTR(ss,ix);
14155 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14156 c = (char*)POPPTR(ss,ix);
14157 TOPPTR(nss,ix) = pv_dup(c);
14159 case SAVEt_GP: /* scalar reference */
14160 gp = (GP*)POPPTR(ss,ix);
14161 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14162 (void)GpREFCNT_inc(gp);
14163 gv = (const GV *)POPPTR(ss,ix);
14164 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14167 ptr = POPPTR(ss,ix);
14168 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14169 /* these are assumed to be refcounted properly */
14171 switch (((OP*)ptr)->op_type) {
14173 case OP_LEAVESUBLV:
14177 case OP_LEAVEWRITE:
14178 TOPPTR(nss,ix) = ptr;
14181 (void) OpREFCNT_inc(o);
14185 TOPPTR(nss,ix) = NULL;
14190 TOPPTR(nss,ix) = NULL;
14192 case SAVEt_FREECOPHH:
14193 ptr = POPPTR(ss,ix);
14194 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14196 case SAVEt_ADELETE:
14197 av = (const AV *)POPPTR(ss,ix);
14198 TOPPTR(nss,ix) = av_dup_inc(av, param);
14200 TOPINT(nss,ix) = i;
14203 hv = (const HV *)POPPTR(ss,ix);
14204 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14206 TOPINT(nss,ix) = i;
14209 c = (char*)POPPTR(ss,ix);
14210 TOPPTR(nss,ix) = pv_dup_inc(c);
14212 case SAVEt_STACK_POS: /* Position on Perl stack */
14214 TOPINT(nss,ix) = i;
14216 case SAVEt_DESTRUCTOR:
14217 ptr = POPPTR(ss,ix);
14218 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14219 dptr = POPDPTR(ss,ix);
14220 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14221 any_dup(FPTR2DPTR(void *, dptr),
14224 case SAVEt_DESTRUCTOR_X:
14225 ptr = POPPTR(ss,ix);
14226 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14227 dxptr = POPDXPTR(ss,ix);
14228 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14229 any_dup(FPTR2DPTR(void *, dxptr),
14232 case SAVEt_REGCONTEXT:
14234 ix -= uv >> SAVE_TIGHT_SHIFT;
14236 case SAVEt_AELEM: /* array element */
14237 sv = (const SV *)POPPTR(ss,ix);
14238 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14240 TOPINT(nss,ix) = i;
14241 av = (const AV *)POPPTR(ss,ix);
14242 TOPPTR(nss,ix) = av_dup_inc(av, param);
14245 ptr = POPPTR(ss,ix);
14246 TOPPTR(nss,ix) = ptr;
14249 ptr = POPPTR(ss,ix);
14250 ptr = cophh_copy((COPHH*)ptr);
14251 TOPPTR(nss,ix) = ptr;
14253 TOPINT(nss,ix) = i;
14254 if (i & HINT_LOCALIZE_HH) {
14255 hv = (const HV *)POPPTR(ss,ix);
14256 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14259 case SAVEt_PADSV_AND_MORTALIZE:
14260 longval = (long)POPLONG(ss,ix);
14261 TOPLONG(nss,ix) = longval;
14262 ptr = POPPTR(ss,ix);
14263 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14264 sv = (const SV *)POPPTR(ss,ix);
14265 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14267 case SAVEt_SET_SVFLAGS:
14269 TOPINT(nss,ix) = i;
14271 TOPINT(nss,ix) = i;
14272 sv = (const SV *)POPPTR(ss,ix);
14273 TOPPTR(nss,ix) = sv_dup(sv, param);
14275 case SAVEt_COMPILE_WARNINGS:
14276 ptr = POPPTR(ss,ix);
14277 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14280 ptr = POPPTR(ss,ix);
14281 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14283 case SAVEt_GP_ALIASED_SV: {
14284 GP * gp_ptr = (GP *)POPPTR(ss,ix);
14285 GP * new_gp_ptr = gp_dup(gp_ptr, param);
14286 TOPPTR(nss,ix) = new_gp_ptr;
14287 new_gp_ptr->gp_refcnt++;
14292 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14300 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14301 * flag to the result. This is done for each stash before cloning starts,
14302 * so we know which stashes want their objects cloned */
14305 do_mark_cloneable_stash(pTHX_ SV *const sv)
14307 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14309 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14310 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14311 if (cloner && GvCV(cloner)) {
14318 mXPUSHs(newSVhek(hvname));
14320 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14327 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14335 =for apidoc perl_clone
14337 Create and return a new interpreter by cloning the current one.
14339 perl_clone takes these flags as parameters:
14341 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14342 without it we only clone the data and zero the stacks,
14343 with it we copy the stacks and the new perl interpreter is
14344 ready to run at the exact same point as the previous one.
14345 The pseudo-fork code uses COPY_STACKS while the
14346 threads->create doesn't.
14348 CLONEf_KEEP_PTR_TABLE -
14349 perl_clone keeps a ptr_table with the pointer of the old
14350 variable as a key and the new variable as a value,
14351 this allows it to check if something has been cloned and not
14352 clone it again but rather just use the value and increase the
14353 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14354 the ptr_table using the function
14355 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14356 reason to keep it around is if you want to dup some of your own
14357 variable who are outside the graph perl scans, example of this
14358 code is in threads.xs create.
14360 CLONEf_CLONE_HOST -
14361 This is a win32 thing, it is ignored on unix, it tells perls
14362 win32host code (which is c++) to clone itself, this is needed on
14363 win32 if you want to run two threads at the same time,
14364 if you just want to do some stuff in a separate perl interpreter
14365 and then throw it away and return to the original one,
14366 you don't need to do anything.
14371 /* XXX the above needs expanding by someone who actually understands it ! */
14372 EXTERN_C PerlInterpreter *
14373 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14376 perl_clone(PerlInterpreter *proto_perl, UV flags)
14379 #ifdef PERL_IMPLICIT_SYS
14381 PERL_ARGS_ASSERT_PERL_CLONE;
14383 /* perlhost.h so we need to call into it
14384 to clone the host, CPerlHost should have a c interface, sky */
14386 if (flags & CLONEf_CLONE_HOST) {
14387 return perl_clone_host(proto_perl,flags);
14389 return perl_clone_using(proto_perl, flags,
14391 proto_perl->IMemShared,
14392 proto_perl->IMemParse,
14394 proto_perl->IStdIO,
14398 proto_perl->IProc);
14402 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14403 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14404 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14405 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14406 struct IPerlDir* ipD, struct IPerlSock* ipS,
14407 struct IPerlProc* ipP)
14409 /* XXX many of the string copies here can be optimized if they're
14410 * constants; they need to be allocated as common memory and just
14411 * their pointers copied. */
14414 CLONE_PARAMS clone_params;
14415 CLONE_PARAMS* const param = &clone_params;
14417 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14419 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14420 #else /* !PERL_IMPLICIT_SYS */
14422 CLONE_PARAMS clone_params;
14423 CLONE_PARAMS* param = &clone_params;
14424 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14426 PERL_ARGS_ASSERT_PERL_CLONE;
14427 #endif /* PERL_IMPLICIT_SYS */
14429 /* for each stash, determine whether its objects should be cloned */
14430 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14431 PERL_SET_THX(my_perl);
14434 PoisonNew(my_perl, 1, PerlInterpreter);
14437 PL_defstash = NULL; /* may be used by perl malloc() */
14440 PL_scopestack_name = 0;
14442 PL_savestack_ix = 0;
14443 PL_savestack_max = -1;
14444 PL_sig_pending = 0;
14446 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14447 Zero(&PL_padname_undef, 1, PADNAME);
14448 Zero(&PL_padname_const, 1, PADNAME);
14449 # ifdef DEBUG_LEAKING_SCALARS
14450 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14452 #else /* !DEBUGGING */
14453 Zero(my_perl, 1, PerlInterpreter);
14454 #endif /* DEBUGGING */
14456 #ifdef PERL_IMPLICIT_SYS
14457 /* host pointers */
14459 PL_MemShared = ipMS;
14460 PL_MemParse = ipMP;
14467 #endif /* PERL_IMPLICIT_SYS */
14470 param->flags = flags;
14471 /* Nothing in the core code uses this, but we make it available to
14472 extensions (using mg_dup). */
14473 param->proto_perl = proto_perl;
14474 /* Likely nothing will use this, but it is initialised to be consistent
14475 with Perl_clone_params_new(). */
14476 param->new_perl = my_perl;
14477 param->unreferenced = NULL;
14480 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14482 PL_body_arenas = NULL;
14483 Zero(&PL_body_roots, 1, PL_body_roots);
14487 PL_sv_arenaroot = NULL;
14489 PL_debug = proto_perl->Idebug;
14491 /* dbargs array probably holds garbage */
14494 PL_compiling = proto_perl->Icompiling;
14496 /* pseudo environmental stuff */
14497 PL_origargc = proto_perl->Iorigargc;
14498 PL_origargv = proto_perl->Iorigargv;
14500 #ifndef NO_TAINT_SUPPORT
14501 /* Set tainting stuff before PerlIO_debug can possibly get called */
14502 PL_tainting = proto_perl->Itainting;
14503 PL_taint_warn = proto_perl->Itaint_warn;
14505 PL_tainting = FALSE;
14506 PL_taint_warn = FALSE;
14509 PL_minus_c = proto_perl->Iminus_c;
14511 PL_localpatches = proto_perl->Ilocalpatches;
14512 PL_splitstr = proto_perl->Isplitstr;
14513 PL_minus_n = proto_perl->Iminus_n;
14514 PL_minus_p = proto_perl->Iminus_p;
14515 PL_minus_l = proto_perl->Iminus_l;
14516 PL_minus_a = proto_perl->Iminus_a;
14517 PL_minus_E = proto_perl->Iminus_E;
14518 PL_minus_F = proto_perl->Iminus_F;
14519 PL_doswitches = proto_perl->Idoswitches;
14520 PL_dowarn = proto_perl->Idowarn;
14521 PL_sawalias = proto_perl->Isawalias;
14522 #ifdef PERL_SAWAMPERSAND
14523 PL_sawampersand = proto_perl->Isawampersand;
14525 PL_unsafe = proto_perl->Iunsafe;
14526 PL_perldb = proto_perl->Iperldb;
14527 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14528 PL_exit_flags = proto_perl->Iexit_flags;
14530 /* XXX time(&PL_basetime) when asked for? */
14531 PL_basetime = proto_perl->Ibasetime;
14533 PL_maxsysfd = proto_perl->Imaxsysfd;
14534 PL_statusvalue = proto_perl->Istatusvalue;
14536 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14538 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14541 /* RE engine related */
14542 PL_regmatch_slab = NULL;
14543 PL_reg_curpm = NULL;
14545 PL_sub_generation = proto_perl->Isub_generation;
14547 /* funky return mechanisms */
14548 PL_forkprocess = proto_perl->Iforkprocess;
14550 /* internal state */
14551 PL_maxo = proto_perl->Imaxo;
14553 PL_main_start = proto_perl->Imain_start;
14554 PL_eval_root = proto_perl->Ieval_root;
14555 PL_eval_start = proto_perl->Ieval_start;
14557 PL_filemode = proto_perl->Ifilemode;
14558 PL_lastfd = proto_perl->Ilastfd;
14559 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14562 PL_gensym = proto_perl->Igensym;
14564 PL_laststatval = proto_perl->Ilaststatval;
14565 PL_laststype = proto_perl->Ilaststype;
14568 PL_profiledata = NULL;
14570 PL_generation = proto_perl->Igeneration;
14572 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14573 PL_in_clean_all = proto_perl->Iin_clean_all;
14575 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14576 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14577 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14578 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14579 PL_nomemok = proto_perl->Inomemok;
14580 PL_an = proto_perl->Ian;
14581 PL_evalseq = proto_perl->Ievalseq;
14582 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14583 PL_origalen = proto_perl->Iorigalen;
14585 PL_sighandlerp = proto_perl->Isighandlerp;
14587 PL_runops = proto_perl->Irunops;
14589 PL_subline = proto_perl->Isubline;
14591 PL_cv_has_eval = proto_perl->Icv_has_eval;
14594 PL_cryptseen = proto_perl->Icryptseen;
14597 #ifdef USE_LOCALE_COLLATE
14598 PL_collation_ix = proto_perl->Icollation_ix;
14599 PL_collation_standard = proto_perl->Icollation_standard;
14600 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14601 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14602 #endif /* USE_LOCALE_COLLATE */
14604 #ifdef USE_LOCALE_NUMERIC
14605 PL_numeric_standard = proto_perl->Inumeric_standard;
14606 PL_numeric_local = proto_perl->Inumeric_local;
14607 #endif /* !USE_LOCALE_NUMERIC */
14609 /* Did the locale setup indicate UTF-8? */
14610 PL_utf8locale = proto_perl->Iutf8locale;
14611 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14612 /* Unicode features (see perlrun/-C) */
14613 PL_unicode = proto_perl->Iunicode;
14615 /* Pre-5.8 signals control */
14616 PL_signals = proto_perl->Isignals;
14618 /* times() ticks per second */
14619 PL_clocktick = proto_perl->Iclocktick;
14621 /* Recursion stopper for PerlIO_find_layer */
14622 PL_in_load_module = proto_perl->Iin_load_module;
14624 /* sort() routine */
14625 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14627 /* Not really needed/useful since the reenrant_retint is "volatile",
14628 * but do it for consistency's sake. */
14629 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14631 /* Hooks to shared SVs and locks. */
14632 PL_sharehook = proto_perl->Isharehook;
14633 PL_lockhook = proto_perl->Ilockhook;
14634 PL_unlockhook = proto_perl->Iunlockhook;
14635 PL_threadhook = proto_perl->Ithreadhook;
14636 PL_destroyhook = proto_perl->Idestroyhook;
14637 PL_signalhook = proto_perl->Isignalhook;
14639 PL_globhook = proto_perl->Iglobhook;
14642 PL_last_swash_hv = NULL; /* reinits on demand */
14643 PL_last_swash_klen = 0;
14644 PL_last_swash_key[0]= '\0';
14645 PL_last_swash_tmps = (U8*)NULL;
14646 PL_last_swash_slen = 0;
14648 PL_srand_called = proto_perl->Isrand_called;
14649 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14651 if (flags & CLONEf_COPY_STACKS) {
14652 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14653 PL_tmps_ix = proto_perl->Itmps_ix;
14654 PL_tmps_max = proto_perl->Itmps_max;
14655 PL_tmps_floor = proto_perl->Itmps_floor;
14657 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14658 * NOTE: unlike the others! */
14659 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14660 PL_scopestack_max = proto_perl->Iscopestack_max;
14662 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14663 * NOTE: unlike the others! */
14664 PL_savestack_ix = proto_perl->Isavestack_ix;
14665 PL_savestack_max = proto_perl->Isavestack_max;
14668 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14669 PL_top_env = &PL_start_env;
14671 PL_op = proto_perl->Iop;
14674 PL_Xpv = (XPV*)NULL;
14675 my_perl->Ina = proto_perl->Ina;
14677 PL_statbuf = proto_perl->Istatbuf;
14678 PL_statcache = proto_perl->Istatcache;
14680 #ifndef NO_TAINT_SUPPORT
14681 PL_tainted = proto_perl->Itainted;
14683 PL_tainted = FALSE;
14685 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14687 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14689 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14690 PL_restartop = proto_perl->Irestartop;
14691 PL_in_eval = proto_perl->Iin_eval;
14692 PL_delaymagic = proto_perl->Idelaymagic;
14693 PL_phase = proto_perl->Iphase;
14694 PL_localizing = proto_perl->Ilocalizing;
14696 PL_hv_fetch_ent_mh = NULL;
14697 PL_modcount = proto_perl->Imodcount;
14698 PL_lastgotoprobe = NULL;
14699 PL_dumpindent = proto_perl->Idumpindent;
14701 PL_efloatbuf = NULL; /* reinits on demand */
14702 PL_efloatsize = 0; /* reinits on demand */
14706 PL_colorset = 0; /* reinits PL_colors[] */
14707 /*PL_colors[6] = {0,0,0,0,0,0};*/
14709 /* Pluggable optimizer */
14710 PL_peepp = proto_perl->Ipeepp;
14711 PL_rpeepp = proto_perl->Irpeepp;
14712 /* op_free() hook */
14713 PL_opfreehook = proto_perl->Iopfreehook;
14715 #ifdef USE_REENTRANT_API
14716 /* XXX: things like -Dm will segfault here in perlio, but doing
14717 * PERL_SET_CONTEXT(proto_perl);
14718 * breaks too many other things
14720 Perl_reentrant_init(aTHX);
14723 /* create SV map for pointer relocation */
14724 PL_ptr_table = ptr_table_new();
14726 /* initialize these special pointers as early as possible */
14728 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14729 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14730 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14731 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14732 &PL_padname_const);
14734 /* create (a non-shared!) shared string table */
14735 PL_strtab = newHV();
14736 HvSHAREKEYS_off(PL_strtab);
14737 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14738 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14740 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14742 /* This PV will be free'd special way so must set it same way op.c does */
14743 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14744 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14746 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14747 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14748 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14749 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14751 param->stashes = newAV(); /* Setup array of objects to call clone on */
14752 /* This makes no difference to the implementation, as it always pushes
14753 and shifts pointers to other SVs without changing their reference
14754 count, with the array becoming empty before it is freed. However, it
14755 makes it conceptually clear what is going on, and will avoid some
14756 work inside av.c, filling slots between AvFILL() and AvMAX() with
14757 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14758 AvREAL_off(param->stashes);
14760 if (!(flags & CLONEf_COPY_STACKS)) {
14761 param->unreferenced = newAV();
14764 #ifdef PERLIO_LAYERS
14765 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14766 PerlIO_clone(aTHX_ proto_perl, param);
14769 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14770 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14771 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14772 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14773 PL_xsubfilename = proto_perl->Ixsubfilename;
14774 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14775 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14778 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14779 PL_inplace = SAVEPV(proto_perl->Iinplace);
14780 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14782 /* magical thingies */
14784 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14785 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14787 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14788 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14789 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14792 /* Clone the regex array */
14793 /* ORANGE FIXME for plugins, probably in the SV dup code.
14794 newSViv(PTR2IV(CALLREGDUPE(
14795 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14797 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14798 PL_regex_pad = AvARRAY(PL_regex_padav);
14800 PL_stashpadmax = proto_perl->Istashpadmax;
14801 PL_stashpadix = proto_perl->Istashpadix ;
14802 Newx(PL_stashpad, PL_stashpadmax, HV *);
14805 for (; o < PL_stashpadmax; ++o)
14806 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14809 /* shortcuts to various I/O objects */
14810 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14811 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14812 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14813 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14814 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14815 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14816 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14818 /* shortcuts to regexp stuff */
14819 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14821 /* shortcuts to misc objects */
14822 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14824 /* shortcuts to debugging objects */
14825 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14826 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14827 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14828 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14829 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14830 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14831 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14833 /* symbol tables */
14834 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14835 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14836 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14837 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14838 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14840 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14841 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14842 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14843 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14844 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14845 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14846 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14847 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14848 PL_savebegin = proto_perl->Isavebegin;
14850 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14852 /* subprocess state */
14853 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14855 if (proto_perl->Iop_mask)
14856 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14859 /* PL_asserting = proto_perl->Iasserting; */
14861 /* current interpreter roots */
14862 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14864 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14867 /* runtime control stuff */
14868 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14870 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14872 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14874 /* interpreter atexit processing */
14875 PL_exitlistlen = proto_perl->Iexitlistlen;
14876 if (PL_exitlistlen) {
14877 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14878 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14881 PL_exitlist = (PerlExitListEntry*)NULL;
14883 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14884 if (PL_my_cxt_size) {
14885 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14886 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14887 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14888 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14889 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14893 PL_my_cxt_list = (void**)NULL;
14894 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14895 PL_my_cxt_keys = (const char**)NULL;
14898 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14899 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14900 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14901 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14903 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14905 PAD_CLONE_VARS(proto_perl, param);
14907 #ifdef HAVE_INTERP_INTERN
14908 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14911 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14913 #ifdef PERL_USES_PL_PIDSTATUS
14914 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14916 PL_osname = SAVEPV(proto_perl->Iosname);
14917 PL_parser = parser_dup(proto_perl->Iparser, param);
14919 /* XXX this only works if the saved cop has already been cloned */
14920 if (proto_perl->Iparser) {
14921 PL_parser->saved_curcop = (COP*)any_dup(
14922 proto_perl->Iparser->saved_curcop,
14926 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14928 #ifdef USE_LOCALE_CTYPE
14929 /* Should we warn if uses locale? */
14930 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
14933 #ifdef USE_LOCALE_COLLATE
14934 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14935 #endif /* USE_LOCALE_COLLATE */
14937 #ifdef USE_LOCALE_NUMERIC
14938 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14939 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14940 #endif /* !USE_LOCALE_NUMERIC */
14942 /* Unicode inversion lists */
14943 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14944 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14945 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14946 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14948 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14949 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14951 /* utf8 character class swashes */
14952 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14953 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14955 for (i = 0; i < POSIX_CC_COUNT; i++) {
14956 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14958 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
14959 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
14960 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
14961 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14962 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14963 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14964 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14965 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14966 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14967 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14968 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14969 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14970 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14971 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14972 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14973 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14974 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14976 if (proto_perl->Ipsig_pend) {
14977 Newxz(PL_psig_pend, SIG_SIZE, int);
14980 PL_psig_pend = (int*)NULL;
14983 if (proto_perl->Ipsig_name) {
14984 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14985 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14987 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14990 PL_psig_ptr = (SV**)NULL;
14991 PL_psig_name = (SV**)NULL;
14994 if (flags & CLONEf_COPY_STACKS) {
14995 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14996 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14997 PL_tmps_ix+1, param);
14999 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15000 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15001 Newxz(PL_markstack, i, I32);
15002 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15003 - proto_perl->Imarkstack);
15004 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15005 - proto_perl->Imarkstack);
15006 Copy(proto_perl->Imarkstack, PL_markstack,
15007 PL_markstack_ptr - PL_markstack + 1, I32);
15009 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15010 * NOTE: unlike the others! */
15011 Newxz(PL_scopestack, PL_scopestack_max, I32);
15012 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15015 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15016 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15018 /* reset stack AV to correct length before its duped via
15019 * PL_curstackinfo */
15020 AvFILLp(proto_perl->Icurstack) =
15021 proto_perl->Istack_sp - proto_perl->Istack_base;
15023 /* NOTE: si_dup() looks at PL_markstack */
15024 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15026 /* PL_curstack = PL_curstackinfo->si_stack; */
15027 PL_curstack = av_dup(proto_perl->Icurstack, param);
15028 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15030 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15031 PL_stack_base = AvARRAY(PL_curstack);
15032 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15033 - proto_perl->Istack_base);
15034 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15036 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15037 PL_savestack = ss_dup(proto_perl, param);
15041 ENTER; /* perl_destruct() wants to LEAVE; */
15044 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15045 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15047 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15048 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15049 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15050 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15051 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15052 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15054 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15056 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15057 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15058 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15060 PL_stashcache = newHV();
15062 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15063 proto_perl->Iwatchaddr);
15064 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15065 if (PL_debug && PL_watchaddr) {
15066 PerlIO_printf(Perl_debug_log,
15067 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15068 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15069 PTR2UV(PL_watchok));
15072 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15073 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15074 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15076 /* Call the ->CLONE method, if it exists, for each of the stashes
15077 identified by sv_dup() above.
15079 while(av_tindex(param->stashes) != -1) {
15080 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15081 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15082 if (cloner && GvCV(cloner)) {
15087 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15089 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15095 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15096 ptr_table_free(PL_ptr_table);
15097 PL_ptr_table = NULL;
15100 if (!(flags & CLONEf_COPY_STACKS)) {
15101 unreferenced_to_tmp_stack(param->unreferenced);
15104 SvREFCNT_dec(param->stashes);
15106 /* orphaned? eg threads->new inside BEGIN or use */
15107 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15108 SvREFCNT_inc_simple_void(PL_compcv);
15109 SAVEFREESV(PL_compcv);
15116 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15118 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15120 if (AvFILLp(unreferenced) > -1) {
15121 SV **svp = AvARRAY(unreferenced);
15122 SV **const last = svp + AvFILLp(unreferenced);
15126 if (SvREFCNT(*svp) == 1)
15128 } while (++svp <= last);
15130 EXTEND_MORTAL(count);
15131 svp = AvARRAY(unreferenced);
15134 if (SvREFCNT(*svp) == 1) {
15135 /* Our reference is the only one to this SV. This means that
15136 in this thread, the scalar effectively has a 0 reference.
15137 That doesn't work (cleanup never happens), so donate our
15138 reference to it onto the save stack. */
15139 PL_tmps_stack[++PL_tmps_ix] = *svp;
15141 /* As an optimisation, because we are already walking the
15142 entire array, instead of above doing either
15143 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15144 release our reference to the scalar, so that at the end of
15145 the array owns zero references to the scalars it happens to
15146 point to. We are effectively converting the array from
15147 AvREAL() on to AvREAL() off. This saves the av_clear()
15148 (triggered by the SvREFCNT_dec(unreferenced) below) from
15149 walking the array a second time. */
15150 SvREFCNT_dec(*svp);
15153 } while (++svp <= last);
15154 AvREAL_off(unreferenced);
15156 SvREFCNT_dec_NN(unreferenced);
15160 Perl_clone_params_del(CLONE_PARAMS *param)
15162 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15164 PerlInterpreter *const to = param->new_perl;
15166 PerlInterpreter *const was = PERL_GET_THX;
15168 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15174 SvREFCNT_dec(param->stashes);
15175 if (param->unreferenced)
15176 unreferenced_to_tmp_stack(param->unreferenced);
15186 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15189 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15190 does a dTHX; to get the context from thread local storage.
15191 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15192 a version that passes in my_perl. */
15193 PerlInterpreter *const was = PERL_GET_THX;
15194 CLONE_PARAMS *param;
15196 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15202 /* Given that we've set the context, we can do this unshared. */
15203 Newx(param, 1, CLONE_PARAMS);
15206 param->proto_perl = from;
15207 param->new_perl = to;
15208 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15209 AvREAL_off(param->stashes);
15210 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15218 #endif /* USE_ITHREADS */
15221 Perl_init_constants(pTHX)
15223 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15224 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15225 SvANY(&PL_sv_undef) = NULL;
15227 SvANY(&PL_sv_no) = new_XPVNV();
15228 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15229 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15230 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15233 SvANY(&PL_sv_yes) = new_XPVNV();
15234 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15235 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15236 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15239 SvPV_set(&PL_sv_no, (char*)PL_No);
15240 SvCUR_set(&PL_sv_no, 0);
15241 SvLEN_set(&PL_sv_no, 0);
15242 SvIV_set(&PL_sv_no, 0);
15243 SvNV_set(&PL_sv_no, 0);
15245 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15246 SvCUR_set(&PL_sv_yes, 1);
15247 SvLEN_set(&PL_sv_yes, 0);
15248 SvIV_set(&PL_sv_yes, 1);
15249 SvNV_set(&PL_sv_yes, 1);
15251 PadnamePV(&PL_padname_const) = (char *)PL_No;
15255 =head1 Unicode Support
15257 =for apidoc sv_recode_to_utf8
15259 The encoding is assumed to be an Encode object, on entry the PV
15260 of the sv is assumed to be octets in that encoding, and the sv
15261 will be converted into Unicode (and UTF-8).
15263 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15264 is not a reference, nothing is done to the sv. If the encoding is not
15265 an C<Encode::XS> Encoding object, bad things will happen.
15266 (See F<lib/encoding.pm> and L<Encode>.)
15268 The PV of the sv is returned.
15273 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15275 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15277 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15286 if (SvPADTMP(nsv)) {
15287 nsv = sv_newmortal();
15288 SvSetSV_nosteal(nsv, sv);
15296 Passing sv_yes is wrong - it needs to be or'ed set of constants
15297 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15298 remove converted chars from source.
15300 Both will default the value - let them.
15302 XPUSHs(&PL_sv_yes);
15305 call_method("decode", G_SCALAR);
15309 s = SvPV_const(uni, len);
15310 if (s != SvPVX_const(sv)) {
15311 SvGROW(sv, len + 1);
15312 Move(s, SvPVX(sv), len + 1, char);
15313 SvCUR_set(sv, len);
15318 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15319 /* clear pos and any utf8 cache */
15320 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15323 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15324 magic_setutf8(sv,mg); /* clear UTF8 cache */
15329 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15333 =for apidoc sv_cat_decode
15335 The encoding is assumed to be an Encode object, the PV of the ssv is
15336 assumed to be octets in that encoding and decoding the input starts
15337 from the position which (PV + *offset) pointed to. The dsv will be
15338 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
15339 when the string tstr appears in decoding output or the input ends on
15340 the PV of the ssv. The value which the offset points will be modified
15341 to the last input position on the ssv.
15343 Returns TRUE if the terminator was found, else returns FALSE.
15348 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15349 SV *ssv, int *offset, char *tstr, int tlen)
15353 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15355 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
15365 offsv = newSViv(*offset);
15367 mPUSHp(tstr, tlen);
15369 call_method("cat_decode", G_SCALAR);
15371 ret = SvTRUE(TOPs);
15372 *offset = SvIV(offsv);
15378 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15383 /* ---------------------------------------------------------------------
15385 * support functions for report_uninit()
15388 /* the maxiumum size of array or hash where we will scan looking
15389 * for the undefined element that triggered the warning */
15391 #define FUV_MAX_SEARCH_SIZE 1000
15393 /* Look for an entry in the hash whose value has the same SV as val;
15394 * If so, return a mortal copy of the key. */
15397 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15403 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15405 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15406 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15409 array = HvARRAY(hv);
15411 for (i=HvMAX(hv); i>=0; i--) {
15413 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15414 if (HeVAL(entry) != val)
15416 if ( HeVAL(entry) == &PL_sv_undef ||
15417 HeVAL(entry) == &PL_sv_placeholder)
15421 if (HeKLEN(entry) == HEf_SVKEY)
15422 return sv_mortalcopy(HeKEY_sv(entry));
15423 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15429 /* Look for an entry in the array whose value has the same SV as val;
15430 * If so, return the index, otherwise return -1. */
15433 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15435 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15437 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15438 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15441 if (val != &PL_sv_undef) {
15442 SV ** const svp = AvARRAY(av);
15445 for (i=AvFILLp(av); i>=0; i--)
15452 /* varname(): return the name of a variable, optionally with a subscript.
15453 * If gv is non-zero, use the name of that global, along with gvtype (one
15454 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15455 * targ. Depending on the value of the subscript_type flag, return:
15458 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15459 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15460 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15461 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15464 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15465 const SV *const keyname, I32 aindex, int subscript_type)
15468 SV * const name = sv_newmortal();
15469 if (gv && isGV(gv)) {
15471 buffer[0] = gvtype;
15474 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15476 gv_fullname4(name, gv, buffer, 0);
15478 if ((unsigned int)SvPVX(name)[1] <= 26) {
15480 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15482 /* Swap the 1 unprintable control character for the 2 byte pretty
15483 version - ie substr($name, 1, 1) = $buffer; */
15484 sv_insert(name, 1, 1, buffer, 2);
15488 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15491 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15493 if (!cv || !CvPADLIST(cv))
15495 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15496 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15500 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15501 SV * const sv = newSV(0);
15502 *SvPVX(name) = '$';
15503 Perl_sv_catpvf(aTHX_ name, "{%s}",
15504 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15505 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15506 SvREFCNT_dec_NN(sv);
15508 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15509 *SvPVX(name) = '$';
15510 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15512 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15513 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15514 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15522 =for apidoc find_uninit_var
15524 Find the name of the undefined variable (if any) that caused the operator
15525 to issue a "Use of uninitialized value" warning.
15526 If match is true, only return a name if its value matches uninit_sv.
15527 So roughly speaking, if a unary operator (such as OP_COS) generates a
15528 warning, then following the direct child of the op may yield an
15529 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15530 other hand, with OP_ADD there are two branches to follow, so we only print
15531 the variable name if we get an exact match.
15532 desc_p points to a string pointer holding the description of the op.
15533 This may be updated if needed.
15535 The name is returned as a mortal SV.
15537 Assumes that PL_op is the op that originally triggered the error, and that
15538 PL_comppad/PL_curpad points to the currently executing pad.
15544 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15545 bool match, const char **desc_p)
15550 const OP *o, *o2, *kid;
15552 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15554 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15555 uninit_sv == &PL_sv_placeholder)))
15558 switch (obase->op_type) {
15565 const bool pad = ( obase->op_type == OP_PADAV
15566 || obase->op_type == OP_PADHV
15567 || obase->op_type == OP_PADRANGE
15570 const bool hash = ( obase->op_type == OP_PADHV
15571 || obase->op_type == OP_RV2HV
15572 || (obase->op_type == OP_PADRANGE
15573 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15577 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15579 if (pad) { /* @lex, %lex */
15580 sv = PAD_SVl(obase->op_targ);
15584 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15585 /* @global, %global */
15586 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15589 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15591 else if (obase == PL_op) /* @{expr}, %{expr} */
15592 return find_uninit_var(cUNOPx(obase)->op_first,
15593 uninit_sv, match, desc_p);
15594 else /* @{expr}, %{expr} as a sub-expression */
15598 /* attempt to find a match within the aggregate */
15600 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15602 subscript_type = FUV_SUBSCRIPT_HASH;
15605 index = find_array_subscript((const AV *)sv, uninit_sv);
15607 subscript_type = FUV_SUBSCRIPT_ARRAY;
15610 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15613 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15614 keysv, index, subscript_type);
15618 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15620 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15621 if (!gv || !GvSTASH(gv))
15623 if (match && (GvSV(gv) != uninit_sv))
15625 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15628 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15631 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15633 return varname(NULL, '$', obase->op_targ,
15634 NULL, 0, FUV_SUBSCRIPT_NONE);
15637 gv = cGVOPx_gv(obase);
15638 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15640 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15642 case OP_AELEMFAST_LEX:
15645 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15646 if (!av || SvRMAGICAL(av))
15648 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15649 if (!svp || *svp != uninit_sv)
15652 return varname(NULL, '$', obase->op_targ,
15653 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15656 gv = cGVOPx_gv(obase);
15661 AV *const av = GvAV(gv);
15662 if (!av || SvRMAGICAL(av))
15664 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15665 if (!svp || *svp != uninit_sv)
15668 return varname(gv, '$', 0,
15669 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15671 NOT_REACHED; /* NOTREACHED */
15674 o = cUNOPx(obase)->op_first;
15675 if (!o || o->op_type != OP_NULL ||
15676 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15678 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15683 bool negate = FALSE;
15685 if (PL_op == obase)
15686 /* $a[uninit_expr] or $h{uninit_expr} */
15687 return find_uninit_var(cBINOPx(obase)->op_last,
15688 uninit_sv, match, desc_p);
15691 o = cBINOPx(obase)->op_first;
15692 kid = cBINOPx(obase)->op_last;
15694 /* get the av or hv, and optionally the gv */
15696 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15697 sv = PAD_SV(o->op_targ);
15699 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15700 && cUNOPo->op_first->op_type == OP_GV)
15702 gv = cGVOPx_gv(cUNOPo->op_first);
15706 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15711 if (kid && kid->op_type == OP_NEGATE) {
15713 kid = cUNOPx(kid)->op_first;
15716 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15717 /* index is constant */
15720 kidsv = newSVpvs_flags("-", SVs_TEMP);
15721 sv_catsv(kidsv, cSVOPx_sv(kid));
15724 kidsv = cSVOPx_sv(kid);
15728 if (obase->op_type == OP_HELEM) {
15729 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15730 if (!he || HeVAL(he) != uninit_sv)
15734 SV * const opsv = cSVOPx_sv(kid);
15735 const IV opsviv = SvIV(opsv);
15736 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15737 negate ? - opsviv : opsviv,
15739 if (!svp || *svp != uninit_sv)
15743 if (obase->op_type == OP_HELEM)
15744 return varname(gv, '%', o->op_targ,
15745 kidsv, 0, FUV_SUBSCRIPT_HASH);
15747 return varname(gv, '@', o->op_targ, NULL,
15748 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15749 FUV_SUBSCRIPT_ARRAY);
15752 /* index is an expression;
15753 * attempt to find a match within the aggregate */
15754 if (obase->op_type == OP_HELEM) {
15755 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15757 return varname(gv, '%', o->op_targ,
15758 keysv, 0, FUV_SUBSCRIPT_HASH);
15762 = find_array_subscript((const AV *)sv, uninit_sv);
15764 return varname(gv, '@', o->op_targ,
15765 NULL, index, FUV_SUBSCRIPT_ARRAY);
15770 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15772 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15774 NOT_REACHED; /* NOTREACHED */
15777 case OP_MULTIDEREF: {
15778 /* If we were executing OP_MULTIDEREF when the undef warning
15779 * triggered, then it must be one of the index values within
15780 * that triggered it. If not, then the only possibility is that
15781 * the value retrieved by the last aggregate lookup might be the
15782 * culprit. For the former, we set PL_multideref_pc each time before
15783 * using an index, so work though the item list until we reach
15784 * that point. For the latter, just work through the entire item
15785 * list; the last aggregate retrieved will be the candidate.
15788 /* the named aggregate, if any */
15789 PADOFFSET agg_targ = 0;
15791 /* the last-seen index */
15793 PADOFFSET index_targ;
15795 IV index_const_iv = 0; /* init for spurious compiler warn */
15796 SV *index_const_sv;
15797 int depth = 0; /* how many array/hash lookups we've done */
15799 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15800 UNOP_AUX_item *last = NULL;
15801 UV actions = items->uv;
15804 if (PL_op == obase) {
15805 last = PL_multideref_pc;
15806 assert(last >= items && last <= items + items[-1].uv);
15813 switch (actions & MDEREF_ACTION_MASK) {
15815 case MDEREF_reload:
15816 actions = (++items)->uv;
15819 case MDEREF_HV_padhv_helem: /* $lex{...} */
15822 case MDEREF_AV_padav_aelem: /* $lex[...] */
15823 agg_targ = (++items)->pad_offset;
15827 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15830 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15832 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15833 assert(isGV_with_GP(agg_gv));
15836 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15837 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15840 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15841 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15847 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15848 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15851 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15852 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15859 index_const_sv = NULL;
15861 index_type = (actions & MDEREF_INDEX_MASK);
15862 switch (index_type) {
15863 case MDEREF_INDEX_none:
15865 case MDEREF_INDEX_const:
15867 index_const_sv = UNOP_AUX_item_sv(++items)
15869 index_const_iv = (++items)->iv;
15871 case MDEREF_INDEX_padsv:
15872 index_targ = (++items)->pad_offset;
15874 case MDEREF_INDEX_gvsv:
15875 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15876 assert(isGV_with_GP(index_gv));
15880 if (index_type != MDEREF_INDEX_none)
15883 if ( index_type == MDEREF_INDEX_none
15884 || (actions & MDEREF_FLAG_last)
15885 || (last && items == last)
15889 actions >>= MDEREF_SHIFT;
15892 if (PL_op == obase) {
15893 /* index was undef */
15895 *desc_p = ( (actions & MDEREF_FLAG_last)
15896 && (obase->op_private
15897 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15899 (obase->op_private & OPpMULTIDEREF_EXISTS)
15902 : is_hv ? "hash element" : "array element";
15903 assert(index_type != MDEREF_INDEX_none);
15905 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15907 return varname(NULL, '$', index_targ,
15908 NULL, 0, FUV_SUBSCRIPT_NONE);
15909 assert(is_hv); /* AV index is an IV and can't be undef */
15910 /* can a const HV index ever be undef? */
15914 /* the SV returned by pp_multideref() was undef, if anything was */
15920 sv = PAD_SV(agg_targ);
15922 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15926 if (index_type == MDEREF_INDEX_const) {
15931 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15932 if (!he || HeVAL(he) != uninit_sv)
15936 SV * const * const svp =
15937 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15938 if (!svp || *svp != uninit_sv)
15943 ? varname(agg_gv, '%', agg_targ,
15944 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15945 : varname(agg_gv, '@', agg_targ,
15946 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15949 /* index is an var */
15951 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15953 return varname(agg_gv, '%', agg_targ,
15954 keysv, 0, FUV_SUBSCRIPT_HASH);
15958 = find_array_subscript((const AV *)sv, uninit_sv);
15960 return varname(agg_gv, '@', agg_targ,
15961 NULL, index, FUV_SUBSCRIPT_ARRAY);
15965 return varname(agg_gv,
15967 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15969 NOT_REACHED; /* NOTREACHED */
15973 /* only examine RHS */
15974 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
15978 o = cUNOPx(obase)->op_first;
15979 if ( o->op_type == OP_PUSHMARK
15980 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15984 if (!OpHAS_SIBLING(o)) {
15985 /* one-arg version of open is highly magical */
15987 if (o->op_type == OP_GV) { /* open FOO; */
15989 if (match && GvSV(gv) != uninit_sv)
15991 return varname(gv, '$', 0,
15992 NULL, 0, FUV_SUBSCRIPT_NONE);
15994 /* other possibilities not handled are:
15995 * open $x; or open my $x; should return '${*$x}'
15996 * open expr; should return '$'.expr ideally
16002 /* ops where $_ may be an implicit arg */
16007 if ( !(obase->op_flags & OPf_STACKED)) {
16008 if (uninit_sv == DEFSV)
16009 return newSVpvs_flags("$_", SVs_TEMP);
16010 else if (obase->op_targ
16011 && uninit_sv == PAD_SVl(obase->op_targ))
16012 return varname(NULL, '$', obase->op_targ, NULL, 0,
16013 FUV_SUBSCRIPT_NONE);
16020 match = 1; /* print etc can return undef on defined args */
16021 /* skip filehandle as it can't produce 'undef' warning */
16022 o = cUNOPx(obase)->op_first;
16023 if ((obase->op_flags & OPf_STACKED)
16025 ( o->op_type == OP_PUSHMARK
16026 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16027 o = OpSIBLING(OpSIBLING(o));
16031 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16032 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16034 /* the following ops are capable of returning PL_sv_undef even for
16035 * defined arg(s) */
16054 case OP_GETPEERNAME:
16102 case OP_SMARTMATCH:
16111 /* XXX tmp hack: these two may call an XS sub, and currently
16112 XS subs don't have a SUB entry on the context stack, so CV and
16113 pad determination goes wrong, and BAD things happen. So, just
16114 don't try to determine the value under those circumstances.
16115 Need a better fix at dome point. DAPM 11/2007 */
16121 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16122 if (gv && GvSV(gv) == uninit_sv)
16123 return newSVpvs_flags("$.", SVs_TEMP);
16128 /* def-ness of rval pos() is independent of the def-ness of its arg */
16129 if ( !(obase->op_flags & OPf_MOD))
16134 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16135 return newSVpvs_flags("${$/}", SVs_TEMP);
16140 if (!(obase->op_flags & OPf_KIDS))
16142 o = cUNOPx(obase)->op_first;
16148 /* This loop checks all the kid ops, skipping any that cannot pos-
16149 * sibly be responsible for the uninitialized value; i.e., defined
16150 * constants and ops that return nothing. If there is only one op
16151 * left that is not skipped, then we *know* it is responsible for
16152 * the uninitialized value. If there is more than one op left, we
16153 * have to look for an exact match in the while() loop below.
16154 * Note that we skip padrange, because the individual pad ops that
16155 * it replaced are still in the tree, so we work on them instead.
16158 for (kid=o; kid; kid = OpSIBLING(kid)) {
16159 const OPCODE type = kid->op_type;
16160 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16161 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16162 || (type == OP_PUSHMARK)
16163 || (type == OP_PADRANGE)
16167 if (o2) { /* more than one found */
16174 return find_uninit_var(o2, uninit_sv, match, desc_p);
16176 /* scan all args */
16178 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16190 =for apidoc report_uninit
16192 Print appropriate "Use of uninitialized variable" warning.
16198 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16201 SV* varname = NULL;
16204 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16207 if (uninit_sv && PL_curpad) {
16208 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16210 sv_insert(varname, 0, 0, " ", 1);
16212 /* PL_warn_uninit_sv is constant */
16213 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16214 /* diag_listed_as: Use of uninitialized value%s */
16215 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16216 SVfARG(varname ? varname : &PL_sv_no),
16221 /* PL_warn_uninit is constant */
16222 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16223 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16231 * c-indentation-style: bsd
16232 * c-basic-offset: 4
16233 * indent-tabs-mode: nil
16236 * ex: set ts=8 sts=4 sw=4 et: