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, 32, 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)
2903 * returns zero, assert-fails on maxlen being too short.
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) {
2911 assert(maxlen >= 4);
2912 if (Perl_isinf(nv)) {
2914 if (maxlen < 5) /* "-Inf\0" */
2924 else if (Perl_isnan(nv)) {
2928 /* XXX optionally output the payload mantissa bits as
2929 * "(unsigned)" (to match the nan("...") C99 function,
2930 * or maybe as "(0xhhh...)" would make more sense...
2931 * provide a format string so that the user can decide?
2932 * NOTE: would affect the maxlen and assert() logic.*/
2937 assert((s == buffer + 3) || (s == buffer + 4));
2939 return s - buffer - 1; /* -1: excluding the zero byte */
2943 =for apidoc sv_2pv_flags
2945 Returns a pointer to the string value of an SV, and sets *lp to its length.
2946 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2947 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2948 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2954 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2958 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2960 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2961 && SvTYPE(sv) != SVt_PVFM);
2962 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2967 if (flags & SV_SKIP_OVERLOAD)
2969 tmpstr = AMG_CALLunary(sv, string_amg);
2970 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2971 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2973 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2977 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2978 if (flags & SV_CONST_RETURN) {
2979 pv = (char *) SvPVX_const(tmpstr);
2981 pv = (flags & SV_MUTABLE_RETURN)
2982 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2985 *lp = SvCUR(tmpstr);
2987 pv = sv_2pv_flags(tmpstr, lp, flags);
3000 SV *const referent = SvRV(sv);
3004 retval = buffer = savepvn("NULLREF", len);
3005 } else if (SvTYPE(referent) == SVt_REGEXP &&
3006 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3007 amagic_is_enabled(string_amg))) {
3008 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3012 /* If the regex is UTF-8 we want the containing scalar to
3013 have an UTF-8 flag too */
3020 *lp = RX_WRAPLEN(re);
3022 return RX_WRAPPED(re);
3024 const char *const typestr = sv_reftype(referent, 0);
3025 const STRLEN typelen = strlen(typestr);
3026 UV addr = PTR2UV(referent);
3027 const char *stashname = NULL;
3028 STRLEN stashnamelen = 0; /* hush, gcc */
3029 const char *buffer_end;
3031 if (SvOBJECT(referent)) {
3032 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3035 stashname = HEK_KEY(name);
3036 stashnamelen = HEK_LEN(name);
3038 if (HEK_UTF8(name)) {
3044 stashname = "__ANON__";
3047 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3048 + 2 * sizeof(UV) + 2 /* )\0 */;
3050 len = typelen + 3 /* (0x */
3051 + 2 * sizeof(UV) + 2 /* )\0 */;
3054 Newx(buffer, len, char);
3055 buffer_end = retval = buffer + len;
3057 /* Working backwards */
3061 *--retval = PL_hexdigit[addr & 15];
3062 } while (addr >>= 4);
3068 memcpy(retval, typestr, typelen);
3072 retval -= stashnamelen;
3073 memcpy(retval, stashname, stashnamelen);
3075 /* retval may not necessarily have reached the start of the
3077 assert (retval >= buffer);
3079 len = buffer_end - retval - 1; /* -1 for that \0 */
3091 if (flags & SV_MUTABLE_RETURN)
3092 return SvPVX_mutable(sv);
3093 if (flags & SV_CONST_RETURN)
3094 return (char *)SvPVX_const(sv);
3099 /* I'm assuming that if both IV and NV are equally valid then
3100 converting the IV is going to be more efficient */
3101 const U32 isUIOK = SvIsUV(sv);
3102 char buf[TYPE_CHARS(UV)];
3106 if (SvTYPE(sv) < SVt_PVIV)
3107 sv_upgrade(sv, SVt_PVIV);
3108 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3110 /* inlined from sv_setpvn */
3111 s = SvGROW_mutable(sv, len + 1);
3112 Move(ptr, s, len, char);
3117 else if (SvNOK(sv)) {
3118 if (SvTYPE(sv) < SVt_PVNV)
3119 sv_upgrade(sv, SVt_PVNV);
3120 if (SvNVX(sv) == 0.0
3121 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3122 && !Perl_isnan(SvNVX(sv))
3125 s = SvGROW_mutable(sv, 2);
3130 STRLEN size = 5; /* "-Inf\0" */
3132 s = SvGROW_mutable(sv, size);
3133 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3139 /* some Xenix systems wipe out errno here */
3148 5 + /* exponent digits */
3152 s = SvGROW_mutable(sv, size);
3153 #ifndef USE_LOCALE_NUMERIC
3154 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3160 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3161 STORE_LC_NUMERIC_SET_TO_NEEDED();
3165 PL_numeric_radix_sv &&
3166 SvUTF8(PL_numeric_radix_sv);
3167 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3168 size += SvLEN(PL_numeric_radix_sv) - 1;
3169 s = SvGROW_mutable(sv, size);
3172 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3174 /* If the radix character is UTF-8, and actually is in the
3175 * output, turn on the UTF-8 flag for the scalar */
3177 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3181 RESTORE_LC_NUMERIC();
3184 /* We don't call SvPOK_on(), because it may come to
3185 * pass that the locale changes so that the
3186 * stringification we just did is no longer correct. We
3187 * will have to re-stringify every time it is needed */
3194 else if (isGV_with_GP(sv)) {
3195 GV *const gv = MUTABLE_GV(sv);
3196 SV *const buffer = sv_newmortal();
3198 gv_efullname3(buffer, gv, "*");
3200 assert(SvPOK(buffer));
3204 *lp = SvCUR(buffer);
3205 return SvPVX(buffer);
3207 else if (isREGEXP(sv)) {
3208 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3209 return RX_WRAPPED((REGEXP *)sv);
3214 if (flags & SV_UNDEF_RETURNS_NULL)
3216 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3218 /* Typically the caller expects that sv_any is not NULL now. */
3219 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3220 sv_upgrade(sv, SVt_PV);
3225 const STRLEN len = s - SvPVX_const(sv);
3230 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3231 PTR2UV(sv),SvPVX_const(sv)));
3232 if (flags & SV_CONST_RETURN)
3233 return (char *)SvPVX_const(sv);
3234 if (flags & SV_MUTABLE_RETURN)
3235 return SvPVX_mutable(sv);
3240 =for apidoc sv_copypv
3242 Copies a stringified representation of the source SV into the
3243 destination SV. Automatically performs any necessary mg_get and
3244 coercion of numeric values into strings. Guaranteed to preserve
3245 UTF8 flag even from overloaded objects. Similar in nature to
3246 sv_2pv[_flags] but operates directly on an SV instead of just the
3247 string. Mostly uses sv_2pv_flags to do its work, except when that
3248 would lose the UTF-8'ness of the PV.
3250 =for apidoc sv_copypv_nomg
3252 Like sv_copypv, but doesn't invoke get magic first.
3254 =for apidoc sv_copypv_flags
3256 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3263 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3268 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3270 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3271 sv_setpvn(dsv,s,len);
3279 =for apidoc sv_2pvbyte
3281 Return a pointer to the byte-encoded representation of the SV, and set *lp
3282 to its length. May cause the SV to be downgraded from UTF-8 as a
3285 Usually accessed via the C<SvPVbyte> macro.
3291 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3293 PERL_ARGS_ASSERT_SV_2PVBYTE;
3296 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3297 || isGV_with_GP(sv) || SvROK(sv)) {
3298 SV *sv2 = sv_newmortal();
3299 sv_copypv_nomg(sv2,sv);
3302 sv_utf8_downgrade(sv,0);
3303 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3307 =for apidoc sv_2pvutf8
3309 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3310 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3312 Usually accessed via the C<SvPVutf8> macro.
3318 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3320 PERL_ARGS_ASSERT_SV_2PVUTF8;
3322 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3323 || isGV_with_GP(sv) || SvROK(sv))
3324 sv = sv_mortalcopy(sv);
3327 sv_utf8_upgrade_nomg(sv);
3328 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3333 =for apidoc sv_2bool
3335 This macro is only used by sv_true() or its macro equivalent, and only if
3336 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3337 It calls sv_2bool_flags with the SV_GMAGIC flag.
3339 =for apidoc sv_2bool_flags
3341 This function is only used by sv_true() and friends, and only if
3342 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3343 contain SV_GMAGIC, then it does an mg_get() first.
3350 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3352 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3355 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3361 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3362 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3365 if(SvGMAGICAL(sv)) {
3367 goto restart; /* call sv_2bool */
3369 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3370 else if(!SvOK(sv)) {
3373 else if(SvPOK(sv)) {
3374 svb = SvPVXtrue(sv);
3376 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3377 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3378 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3382 goto restart; /* call sv_2bool_nomg */
3387 return SvRV(sv) != 0;
3391 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3392 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3396 =for apidoc sv_utf8_upgrade
3398 Converts the PV of an SV to its UTF-8-encoded form.
3399 Forces the SV to string form if it is not already.
3400 Will C<mg_get> on C<sv> if appropriate.
3401 Always sets the SvUTF8 flag to avoid future validity checks even
3402 if the whole string is the same in UTF-8 as not.
3403 Returns the number of bytes in the converted string
3405 This is not a general purpose byte encoding to Unicode interface:
3406 use the Encode extension for that.
3408 =for apidoc sv_utf8_upgrade_nomg
3410 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3412 =for apidoc sv_utf8_upgrade_flags
3414 Converts the PV of an SV to its UTF-8-encoded form.
3415 Forces the SV to string form if it is not already.
3416 Always sets the SvUTF8 flag to avoid future validity checks even
3417 if all the bytes are invariant in UTF-8.
3418 If C<flags> has C<SV_GMAGIC> bit set,
3419 will C<mg_get> on C<sv> if appropriate, else not.
3421 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3422 will expand when converted to UTF-8, and skips the extra work of checking for
3423 that. Typically this flag is used by a routine that has already parsed the
3424 string and found such characters, and passes this information on so that the
3425 work doesn't have to be repeated.
3427 Returns the number of bytes in the converted string.
3429 This is not a general purpose byte encoding to Unicode interface:
3430 use the Encode extension for that.
3432 =for apidoc sv_utf8_upgrade_flags_grow
3434 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3435 the number of unused bytes the string of 'sv' is guaranteed to have free after
3436 it upon return. This allows the caller to reserve extra space that it intends
3437 to fill, to avoid extra grows.
3439 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3440 are implemented in terms of this function.
3442 Returns the number of bytes in the converted string (not including the spares).
3446 (One might think that the calling routine could pass in the position of the
3447 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3448 have to be found again. But that is not the case, because typically when the
3449 caller is likely to use this flag, it won't be calling this routine unless it
3450 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3451 and just use bytes. But some things that do fit into a byte are variants in
3452 utf8, and the caller may not have been keeping track of these.)
3454 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3455 C<NUL> isn't guaranteed due to having other routines do the work in some input
3456 cases, or if the input is already flagged as being in utf8.
3458 The speed of this could perhaps be improved for many cases if someone wanted to
3459 write a fast function that counts the number of variant characters in a string,
3460 especially if it could return the position of the first one.
3465 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3467 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3469 if (sv == &PL_sv_undef)
3471 if (!SvPOK_nog(sv)) {
3473 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3474 (void) sv_2pv_flags(sv,&len, flags);
3476 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3480 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3485 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3490 S_sv_uncow(aTHX_ sv, 0);
3493 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3494 sv_recode_to_utf8(sv, _get_encoding());
3495 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3499 if (SvCUR(sv) == 0) {
3500 if (extra) SvGROW(sv, extra);
3501 } else { /* Assume Latin-1/EBCDIC */
3502 /* This function could be much more efficient if we
3503 * had a FLAG in SVs to signal if there are any variant
3504 * chars in the PV. Given that there isn't such a flag
3505 * make the loop as fast as possible (although there are certainly ways
3506 * to speed this up, eg. through vectorization) */
3507 U8 * s = (U8 *) SvPVX_const(sv);
3508 U8 * e = (U8 *) SvEND(sv);
3510 STRLEN two_byte_count = 0;
3512 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3514 /* See if really will need to convert to utf8. We mustn't rely on our
3515 * incoming SV being well formed and having a trailing '\0', as certain
3516 * code in pp_formline can send us partially built SVs. */
3520 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3522 t--; /* t already incremented; re-point to first variant */
3527 /* utf8 conversion not needed because all are invariants. Mark as
3528 * UTF-8 even if no variant - saves scanning loop */
3530 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3535 /* Here, the string should be converted to utf8, either because of an
3536 * input flag (two_byte_count = 0), or because a character that
3537 * requires 2 bytes was found (two_byte_count = 1). t points either to
3538 * the beginning of the string (if we didn't examine anything), or to
3539 * the first variant. In either case, everything from s to t - 1 will
3540 * occupy only 1 byte each on output.
3542 * There are two main ways to convert. One is to create a new string
3543 * and go through the input starting from the beginning, appending each
3544 * converted value onto the new string as we go along. It's probably
3545 * best to allocate enough space in the string for the worst possible
3546 * case rather than possibly running out of space and having to
3547 * reallocate and then copy what we've done so far. Since everything
3548 * from s to t - 1 is invariant, the destination can be initialized
3549 * with these using a fast memory copy
3551 * The other way is to figure out exactly how big the string should be
3552 * by parsing the entire input. Then you don't have to make it big
3553 * enough to handle the worst possible case, and more importantly, if
3554 * the string you already have is large enough, you don't have to
3555 * allocate a new string, you can copy the last character in the input
3556 * string to the final position(s) that will be occupied by the
3557 * converted string and go backwards, stopping at t, since everything
3558 * before that is invariant.
3560 * There are advantages and disadvantages to each method.
3562 * In the first method, we can allocate a new string, do the memory
3563 * copy from the s to t - 1, and then proceed through the rest of the
3564 * string byte-by-byte.
3566 * In the second method, we proceed through the rest of the input
3567 * string just calculating how big the converted string will be. Then
3568 * there are two cases:
3569 * 1) if the string has enough extra space to handle the converted
3570 * value. We go backwards through the string, converting until we
3571 * get to the position we are at now, and then stop. If this
3572 * position is far enough along in the string, this method is
3573 * faster than the other method. If the memory copy were the same
3574 * speed as the byte-by-byte loop, that position would be about
3575 * half-way, as at the half-way mark, parsing to the end and back
3576 * is one complete string's parse, the same amount as starting
3577 * over and going all the way through. Actually, it would be
3578 * somewhat less than half-way, as it's faster to just count bytes
3579 * than to also copy, and we don't have the overhead of allocating
3580 * a new string, changing the scalar to use it, and freeing the
3581 * existing one. But if the memory copy is fast, the break-even
3582 * point is somewhere after half way. The counting loop could be
3583 * sped up by vectorization, etc, to move the break-even point
3584 * further towards the beginning.
3585 * 2) if the string doesn't have enough space to handle the converted
3586 * value. A new string will have to be allocated, and one might
3587 * as well, given that, start from the beginning doing the first
3588 * method. We've spent extra time parsing the string and in
3589 * exchange all we've gotten is that we know precisely how big to
3590 * make the new one. Perl is more optimized for time than space,
3591 * so this case is a loser.
3592 * So what I've decided to do is not use the 2nd method unless it is
3593 * guaranteed that a new string won't have to be allocated, assuming
3594 * the worst case. I also decided not to put any more conditions on it
3595 * than this, for now. It seems likely that, since the worst case is
3596 * twice as big as the unknown portion of the string (plus 1), we won't
3597 * be guaranteed enough space, causing us to go to the first method,
3598 * unless the string is short, or the first variant character is near
3599 * the end of it. In either of these cases, it seems best to use the
3600 * 2nd method. The only circumstance I can think of where this would
3601 * be really slower is if the string had once had much more data in it
3602 * than it does now, but there is still a substantial amount in it */
3605 STRLEN invariant_head = t - s;
3606 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3607 if (SvLEN(sv) < size) {
3609 /* Here, have decided to allocate a new string */
3614 Newx(dst, size, U8);
3616 /* If no known invariants at the beginning of the input string,
3617 * set so starts from there. Otherwise, can use memory copy to
3618 * get up to where we are now, and then start from here */
3620 if (invariant_head == 0) {
3623 Copy(s, dst, invariant_head, char);
3624 d = dst + invariant_head;
3628 append_utf8_from_native_byte(*t, &d);
3632 SvPV_free(sv); /* No longer using pre-existing string */
3633 SvPV_set(sv, (char*)dst);
3634 SvCUR_set(sv, d - dst);
3635 SvLEN_set(sv, size);
3638 /* Here, have decided to get the exact size of the string.
3639 * Currently this happens only when we know that there is
3640 * guaranteed enough space to fit the converted string, so
3641 * don't have to worry about growing. If two_byte_count is 0,
3642 * then t points to the first byte of the string which hasn't
3643 * been examined yet. Otherwise two_byte_count is 1, and t
3644 * points to the first byte in the string that will expand to
3645 * two. Depending on this, start examining at t or 1 after t.
3648 U8 *d = t + two_byte_count;
3651 /* Count up the remaining bytes that expand to two */
3654 const U8 chr = *d++;
3655 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3658 /* The string will expand by just the number of bytes that
3659 * occupy two positions. But we are one afterwards because of
3660 * the increment just above. This is the place to put the
3661 * trailing NUL, and to set the length before we decrement */
3663 d += two_byte_count;
3664 SvCUR_set(sv, d - s);
3668 /* Having decremented d, it points to the position to put the
3669 * very last byte of the expanded string. Go backwards through
3670 * the string, copying and expanding as we go, stopping when we
3671 * get to the part that is invariant the rest of the way down */
3675 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3678 *d-- = UTF8_EIGHT_BIT_LO(*e);
3679 *d-- = UTF8_EIGHT_BIT_HI(*e);
3685 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3686 /* Update pos. We do it at the end rather than during
3687 * the upgrade, to avoid slowing down the common case
3688 * (upgrade without pos).
3689 * pos can be stored as either bytes or characters. Since
3690 * this was previously a byte string we can just turn off
3691 * the bytes flag. */
3692 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3694 mg->mg_flags &= ~MGf_BYTES;
3696 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3697 magic_setutf8(sv,mg); /* clear UTF8 cache */
3702 /* Mark as UTF-8 even if no variant - saves scanning loop */
3708 =for apidoc sv_utf8_downgrade
3710 Attempts to convert the PV of an SV from characters to bytes.
3711 If the PV contains a character that cannot fit
3712 in a byte, this conversion will fail;
3713 in this case, either returns false or, if C<fail_ok> is not
3716 This is not a general purpose Unicode to byte encoding interface:
3717 use the Encode extension for that.
3723 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3725 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3727 if (SvPOKp(sv) && SvUTF8(sv)) {
3731 int mg_flags = SV_GMAGIC;
3734 S_sv_uncow(aTHX_ sv, 0);
3736 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3738 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3739 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3740 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3741 SV_GMAGIC|SV_CONST_RETURN);
3742 mg_flags = 0; /* sv_pos_b2u does get magic */
3744 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3745 magic_setutf8(sv,mg); /* clear UTF8 cache */
3748 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3750 if (!utf8_to_bytes(s, &len)) {
3755 Perl_croak(aTHX_ "Wide character in %s",
3758 Perl_croak(aTHX_ "Wide character");
3769 =for apidoc sv_utf8_encode
3771 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3772 flag off so that it looks like octets again.
3778 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3780 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3782 if (SvREADONLY(sv)) {
3783 sv_force_normal_flags(sv, 0);
3785 (void) sv_utf8_upgrade(sv);
3790 =for apidoc sv_utf8_decode
3792 If the PV of the SV is an octet sequence in UTF-8
3793 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3794 so that it looks like a character. If the PV contains only single-byte
3795 characters, the C<SvUTF8> flag stays off.
3796 Scans PV for validity and returns false if the PV is invalid UTF-8.
3802 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3804 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3807 const U8 *start, *c;
3810 /* The octets may have got themselves encoded - get them back as
3813 if (!sv_utf8_downgrade(sv, TRUE))
3816 /* it is actually just a matter of turning the utf8 flag on, but
3817 * we want to make sure everything inside is valid utf8 first.
3819 c = start = (const U8 *) SvPVX_const(sv);
3820 if (!is_utf8_string(c, SvCUR(sv)))
3822 e = (const U8 *) SvEND(sv);
3825 if (!UTF8_IS_INVARIANT(ch)) {
3830 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3831 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3832 after this, clearing pos. Does anything on CPAN
3834 /* adjust pos to the start of a UTF8 char sequence */
3835 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3837 I32 pos = mg->mg_len;
3839 for (c = start + pos; c > start; c--) {
3840 if (UTF8_IS_START(*c))
3843 mg->mg_len = c - start;
3846 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3847 magic_setutf8(sv,mg); /* clear UTF8 cache */
3854 =for apidoc sv_setsv
3856 Copies the contents of the source SV C<ssv> into the destination SV
3857 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3858 function if the source SV needs to be reused. Does not handle 'set' magic on
3859 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3860 performs a copy-by-value, obliterating any previous content of the
3863 You probably want to use one of the assortment of wrappers, such as
3864 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3865 C<SvSetMagicSV_nosteal>.
3867 =for apidoc sv_setsv_flags
3869 Copies the contents of the source SV C<ssv> into the destination SV
3870 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3871 function if the source SV needs to be reused. Does not handle 'set' magic.
3872 Loosely speaking, it performs a copy-by-value, obliterating any previous
3873 content of the destination.
3874 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3875 C<ssv> if appropriate, else not. If the C<flags>
3876 parameter has the C<SV_NOSTEAL> bit set then the
3877 buffers of temps will not be stolen. <sv_setsv>
3878 and C<sv_setsv_nomg> are implemented in terms of this function.
3880 You probably want to use one of the assortment of wrappers, such as
3881 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3882 C<SvSetMagicSV_nosteal>.
3884 This is the primary function for copying scalars, and most other
3885 copy-ish functions and macros use this underneath.
3891 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3893 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3894 HV *old_stash = NULL;
3896 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3898 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3899 const char * const name = GvNAME(sstr);
3900 const STRLEN len = GvNAMELEN(sstr);
3902 if (dtype >= SVt_PV) {
3908 SvUPGRADE(dstr, SVt_PVGV);
3909 (void)SvOK_off(dstr);
3910 isGV_with_GP_on(dstr);
3912 GvSTASH(dstr) = GvSTASH(sstr);
3914 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3915 gv_name_set(MUTABLE_GV(dstr), name, len,
3916 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3917 SvFAKE_on(dstr); /* can coerce to non-glob */
3920 if(GvGP(MUTABLE_GV(sstr))) {
3921 /* If source has method cache entry, clear it */
3923 SvREFCNT_dec(GvCV(sstr));
3924 GvCV_set(sstr, NULL);
3927 /* If source has a real method, then a method is
3930 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3936 /* If dest already had a real method, that's a change as well */
3938 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3939 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3944 /* We don't need to check the name of the destination if it was not a
3945 glob to begin with. */
3946 if(dtype == SVt_PVGV) {
3947 const char * const name = GvNAME((const GV *)dstr);
3950 /* The stash may have been detached from the symbol table, so
3952 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3956 const STRLEN len = GvNAMELEN(dstr);
3957 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3958 || (len == 1 && name[0] == ':')) {
3961 /* Set aside the old stash, so we can reset isa caches on
3963 if((old_stash = GvHV(dstr)))
3964 /* Make sure we do not lose it early. */
3965 SvREFCNT_inc_simple_void_NN(
3966 sv_2mortal((SV *)old_stash)
3971 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3974 gp_free(MUTABLE_GV(dstr));
3975 GvINTRO_off(dstr); /* one-shot flag */
3976 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3977 if (SvTAINTED(sstr))
3979 if (GvIMPORTED(dstr) != GVf_IMPORTED
3980 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3982 GvIMPORTED_on(dstr);
3985 if(mro_changes == 2) {
3986 if (GvAV((const GV *)sstr)) {
3988 SV * const sref = (SV *)GvAV((const GV *)dstr);
3989 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3990 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3991 AV * const ary = newAV();
3992 av_push(ary, mg->mg_obj); /* takes the refcount */
3993 mg->mg_obj = (SV *)ary;
3995 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3997 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3999 mro_isa_changed_in(GvSTASH(dstr));
4001 else if(mro_changes == 3) {
4002 HV * const stash = GvHV(dstr);
4003 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
4009 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
4010 if (GvIO(dstr) && dtype == SVt_PVGV) {
4011 DEBUG_o(Perl_deb(aTHX_
4012 "glob_assign_glob clearing PL_stashcache\n"));
4013 /* It's a cache. It will rebuild itself quite happily.
4014 It's a lot of effort to work out exactly which key (or keys)
4015 might be invalidated by the creation of the this file handle.
4017 hv_clear(PL_stashcache);
4023 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4025 SV * const sref = SvRV(sstr);
4027 const int intro = GvINTRO(dstr);
4030 const U32 stype = SvTYPE(sref);
4032 PERL_ARGS_ASSERT_GV_SETREF;
4035 GvINTRO_off(dstr); /* one-shot flag */
4036 GvLINE(dstr) = CopLINE(PL_curcop);
4037 GvEGV(dstr) = MUTABLE_GV(dstr);
4042 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4043 import_flag = GVf_IMPORTED_CV;
4046 location = (SV **) &GvHV(dstr);
4047 import_flag = GVf_IMPORTED_HV;
4050 location = (SV **) &GvAV(dstr);
4051 import_flag = GVf_IMPORTED_AV;
4054 location = (SV **) &GvIOp(dstr);
4057 location = (SV **) &GvFORM(dstr);
4060 location = &GvSV(dstr);
4061 import_flag = GVf_IMPORTED_SV;
4064 if (stype == SVt_PVCV) {
4065 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4066 if (GvCVGEN(dstr)) {
4067 SvREFCNT_dec(GvCV(dstr));
4068 GvCV_set(dstr, NULL);
4069 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4072 /* SAVEt_GVSLOT takes more room on the savestack and has more
4073 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4074 leave_scope needs access to the GV so it can reset method
4075 caches. We must use SAVEt_GVSLOT whenever the type is
4076 SVt_PVCV, even if the stash is anonymous, as the stash may
4077 gain a name somehow before leave_scope. */
4078 if (stype == SVt_PVCV) {
4079 /* There is no save_pushptrptrptr. Creating it for this
4080 one call site would be overkill. So inline the ss add
4084 SS_ADD_PTR(location);
4085 SS_ADD_PTR(SvREFCNT_inc(*location));
4086 SS_ADD_UV(SAVEt_GVSLOT);
4089 else SAVEGENERICSV(*location);
4092 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4093 CV* const cv = MUTABLE_CV(*location);
4095 if (!GvCVGEN((const GV *)dstr) &&
4096 (CvROOT(cv) || CvXSUB(cv)) &&
4097 /* redundant check that avoids creating the extra SV
4098 most of the time: */
4099 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4101 SV * const new_const_sv =
4102 CvCONST((const CV *)sref)
4103 ? cv_const_sv((const CV *)sref)
4105 report_redefined_cv(
4106 sv_2mortal(Perl_newSVpvf(aTHX_
4109 HvNAME_HEK(GvSTASH((const GV *)dstr))
4111 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4114 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4118 cv_ckproto_len_flags(cv, (const GV *)dstr,
4119 SvPOK(sref) ? CvPROTO(sref) : NULL,
4120 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4121 SvPOK(sref) ? SvUTF8(sref) : 0);
4123 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4124 GvASSUMECV_on(dstr);
4125 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4126 if (intro && GvREFCNT(dstr) > 1) {
4127 /* temporary remove extra savestack's ref */
4129 gv_method_changed(dstr);
4132 else gv_method_changed(dstr);
4135 *location = SvREFCNT_inc_simple_NN(sref);
4136 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4137 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4138 GvFLAGS(dstr) |= import_flag;
4140 if (import_flag == GVf_IMPORTED_SV) {
4142 save_aliased_sv((GV *)dstr);
4144 /* Turn off the flag if sref is not referenced elsewhere,
4145 even by weak refs. (SvRMAGICAL is a pessimistic check for
4147 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4148 GvALIASED_SV_off(dstr);
4150 GvALIASED_SV_on(dstr);
4152 if (stype == SVt_PVHV) {
4153 const char * const name = GvNAME((GV*)dstr);
4154 const STRLEN len = GvNAMELEN(dstr);
4157 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4158 || (len == 1 && name[0] == ':')
4160 && (!dref || HvENAME_get(dref))
4163 (HV *)sref, (HV *)dref,
4169 stype == SVt_PVAV && sref != dref
4170 && strEQ(GvNAME((GV*)dstr), "ISA")
4171 /* The stash may have been detached from the symbol table, so
4172 check its name before doing anything. */
4173 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4176 MAGIC * const omg = dref && SvSMAGICAL(dref)
4177 ? mg_find(dref, PERL_MAGIC_isa)
4179 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4180 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4181 AV * const ary = newAV();
4182 av_push(ary, mg->mg_obj); /* takes the refcount */
4183 mg->mg_obj = (SV *)ary;
4186 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4187 SV **svp = AvARRAY((AV *)omg->mg_obj);
4188 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4192 SvREFCNT_inc_simple_NN(*svp++)
4198 SvREFCNT_inc_simple_NN(omg->mg_obj)
4202 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4207 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4209 mg = mg_find(sref, PERL_MAGIC_isa);
4211 /* Since the *ISA assignment could have affected more than
4212 one stash, don't call mro_isa_changed_in directly, but let
4213 magic_clearisa do it for us, as it already has the logic for
4214 dealing with globs vs arrays of globs. */
4216 Perl_magic_clearisa(aTHX_ NULL, mg);
4218 else if (stype == SVt_PVIO) {
4219 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4220 /* It's a cache. It will rebuild itself quite happily.
4221 It's a lot of effort to work out exactly which key (or keys)
4222 might be invalidated by the creation of the this file handle.
4224 hv_clear(PL_stashcache);
4228 if (!intro) SvREFCNT_dec(dref);
4229 if (SvTAINTED(sstr))
4237 #ifdef PERL_DEBUG_READONLY_COW
4238 # include <sys/mman.h>
4240 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4241 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4245 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4247 struct perl_memory_debug_header * const header =
4248 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4249 const MEM_SIZE len = header->size;
4250 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4251 # ifdef PERL_TRACK_MEMPOOL
4252 if (!header->readonly) header->readonly = 1;
4254 if (mprotect(header, len, PROT_READ))
4255 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4256 header, len, errno);
4260 S_sv_buf_to_rw(pTHX_ SV *sv)
4262 struct perl_memory_debug_header * const header =
4263 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4264 const MEM_SIZE len = header->size;
4265 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4266 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4267 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4268 header, len, errno);
4269 # ifdef PERL_TRACK_MEMPOOL
4270 header->readonly = 0;
4275 # define sv_buf_to_ro(sv) NOOP
4276 # define sv_buf_to_rw(sv) NOOP
4280 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4286 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4288 if (UNLIKELY( sstr == dstr ))
4291 if (SvIS_FREED(dstr)) {
4292 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4293 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4295 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4296 if (UNLIKELY( !sstr ))
4297 sstr = &PL_sv_undef;
4298 if (SvIS_FREED(sstr)) {
4299 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4300 (void*)sstr, (void*)dstr);
4302 stype = SvTYPE(sstr);
4303 dtype = SvTYPE(dstr);
4305 /* There's a lot of redundancy below but we're going for speed here */
4310 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4311 (void)SvOK_off(dstr);
4319 /* For performance, we inline promoting to type SVt_IV. */
4320 /* We're starting from SVt_NULL, so provided that define is
4321 * actual 0, we don't have to unset any SV type flags
4322 * to promote to SVt_IV. */
4323 STATIC_ASSERT_STMT(SVt_NULL == 0);
4324 SET_SVANY_FOR_BODYLESS_IV(dstr);
4325 SvFLAGS(dstr) |= SVt_IV;
4329 sv_upgrade(dstr, SVt_PVIV);
4333 goto end_of_first_switch;
4335 (void)SvIOK_only(dstr);
4336 SvIV_set(dstr, SvIVX(sstr));
4339 /* SvTAINTED can only be true if the SV has taint magic, which in
4340 turn means that the SV type is PVMG (or greater). This is the
4341 case statement for SVt_IV, so this cannot be true (whatever gcov
4343 assert(!SvTAINTED(sstr));
4348 if (dtype < SVt_PV && dtype != SVt_IV)
4349 sv_upgrade(dstr, SVt_IV);
4353 if (LIKELY( SvNOK(sstr) )) {
4357 sv_upgrade(dstr, SVt_NV);
4361 sv_upgrade(dstr, SVt_PVNV);
4365 goto end_of_first_switch;
4367 SvNV_set(dstr, SvNVX(sstr));
4368 (void)SvNOK_only(dstr);
4369 /* SvTAINTED can only be true if the SV has taint magic, which in
4370 turn means that the SV type is PVMG (or greater). This is the
4371 case statement for SVt_NV, so this cannot be true (whatever gcov
4373 assert(!SvTAINTED(sstr));
4380 sv_upgrade(dstr, SVt_PV);
4383 if (dtype < SVt_PVIV)
4384 sv_upgrade(dstr, SVt_PVIV);
4387 if (dtype < SVt_PVNV)
4388 sv_upgrade(dstr, SVt_PVNV);
4392 const char * const type = sv_reftype(sstr,0);
4394 /* diag_listed_as: Bizarre copy of %s */
4395 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4397 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4399 NOT_REACHED; /* NOTREACHED */
4403 if (dtype < SVt_REGEXP)
4405 if (dtype >= SVt_PV) {
4411 sv_upgrade(dstr, SVt_REGEXP);
4419 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4421 if (SvTYPE(sstr) != stype)
4422 stype = SvTYPE(sstr);
4424 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4425 glob_assign_glob(dstr, sstr, dtype);
4428 if (stype == SVt_PVLV)
4430 if (isREGEXP(sstr)) goto upgregexp;
4431 SvUPGRADE(dstr, SVt_PVNV);
4434 SvUPGRADE(dstr, (svtype)stype);
4436 end_of_first_switch:
4438 /* dstr may have been upgraded. */
4439 dtype = SvTYPE(dstr);
4440 sflags = SvFLAGS(sstr);
4442 if (UNLIKELY( dtype == SVt_PVCV )) {
4443 /* Assigning to a subroutine sets the prototype. */
4446 const char *const ptr = SvPV_const(sstr, len);
4448 SvGROW(dstr, len + 1);
4449 Copy(ptr, SvPVX(dstr), len + 1, char);
4450 SvCUR_set(dstr, len);
4452 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4453 CvAUTOLOAD_off(dstr);
4458 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4459 || dtype == SVt_PVFM))
4461 const char * const type = sv_reftype(dstr,0);
4463 /* diag_listed_as: Cannot copy to %s */
4464 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4466 Perl_croak(aTHX_ "Cannot copy to %s", type);
4467 } else if (sflags & SVf_ROK) {
4468 if (isGV_with_GP(dstr)
4469 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4472 if (GvIMPORTED(dstr) != GVf_IMPORTED
4473 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4475 GvIMPORTED_on(dstr);
4480 glob_assign_glob(dstr, sstr, dtype);
4484 if (dtype >= SVt_PV) {
4485 if (isGV_with_GP(dstr)) {
4486 gv_setref(dstr, sstr);
4489 if (SvPVX_const(dstr)) {
4495 (void)SvOK_off(dstr);
4496 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4497 SvFLAGS(dstr) |= sflags & SVf_ROK;
4498 assert(!(sflags & SVp_NOK));
4499 assert(!(sflags & SVp_IOK));
4500 assert(!(sflags & SVf_NOK));
4501 assert(!(sflags & SVf_IOK));
4503 else if (isGV_with_GP(dstr)) {
4504 if (!(sflags & SVf_OK)) {
4505 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4506 "Undefined value assigned to typeglob");
4509 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4510 if (dstr != (const SV *)gv) {
4511 const char * const name = GvNAME((const GV *)dstr);
4512 const STRLEN len = GvNAMELEN(dstr);
4513 HV *old_stash = NULL;
4514 bool reset_isa = FALSE;
4515 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4516 || (len == 1 && name[0] == ':')) {
4517 /* Set aside the old stash, so we can reset isa caches
4518 on its subclasses. */
4519 if((old_stash = GvHV(dstr))) {
4520 /* Make sure we do not lose it early. */
4521 SvREFCNT_inc_simple_void_NN(
4522 sv_2mortal((SV *)old_stash)
4529 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4530 gp_free(MUTABLE_GV(dstr));
4532 GvGP_set(dstr, gp_ref(GvGP(gv)));
4535 HV * const stash = GvHV(dstr);
4537 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4547 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4548 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4549 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4551 else if (sflags & SVp_POK) {
4552 const STRLEN cur = SvCUR(sstr);
4553 const STRLEN len = SvLEN(sstr);
4556 * We have three basic ways to copy the string:
4562 * Which we choose is based on various factors. The following
4563 * things are listed in order of speed, fastest to slowest:
4565 * - Copying a short string
4566 * - Copy-on-write bookkeeping
4568 * - Copying a long string
4570 * We swipe the string (steal the string buffer) if the SV on the
4571 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4572 * big win on long strings. It should be a win on short strings if
4573 * SvPVX_const(dstr) has to be allocated. If not, it should not
4574 * slow things down, as SvPVX_const(sstr) would have been freed
4577 * We also steal the buffer from a PADTMP (operator target) if it
4578 * is ‘long enough’. For short strings, a swipe does not help
4579 * here, as it causes more malloc calls the next time the target
4580 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4581 * be allocated it is still not worth swiping PADTMPs for short
4582 * strings, as the savings here are small.
4584 * If swiping is not an option, then we see whether it is
4585 * worth using copy-on-write. If the lhs already has a buf-
4586 * fer big enough and the string is short, we skip it and fall back
4587 * to method 3, since memcpy is faster for short strings than the
4588 * later bookkeeping overhead that copy-on-write entails.
4590 * If the rhs is not a copy-on-write string yet, then we also
4591 * consider whether the buffer is too large relative to the string
4592 * it holds. Some operations such as readline allocate a large
4593 * buffer in the expectation of reusing it. But turning such into
4594 * a COW buffer is counter-productive because it increases memory
4595 * usage by making readline allocate a new large buffer the sec-
4596 * ond time round. So, if the buffer is too large, again, we use
4599 * Finally, if there is no buffer on the left, or the buffer is too
4600 * small, then we use copy-on-write and make both SVs share the
4605 /* Whichever path we take through the next code, we want this true,
4606 and doing it now facilitates the COW check. */
4607 (void)SvPOK_only(dstr);
4611 /* slated for free anyway (and not COW)? */
4612 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4613 /* or a swipable TARG */
4615 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4617 /* whose buffer is worth stealing */
4618 && CHECK_COWBUF_THRESHOLD(cur,len)
4621 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4622 (!(flags & SV_NOSTEAL)) &&
4623 /* and we're allowed to steal temps */
4624 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4625 len) /* and really is a string */
4626 { /* Passes the swipe test. */
4627 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4629 SvPV_set(dstr, SvPVX_mutable(sstr));
4630 SvLEN_set(dstr, SvLEN(sstr));
4631 SvCUR_set(dstr, SvCUR(sstr));
4634 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4635 SvPV_set(sstr, NULL);
4640 else if (flags & SV_COW_SHARED_HASH_KEYS
4642 #ifdef PERL_OLD_COPY_ON_WRITE
4643 ( sflags & SVf_IsCOW
4644 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4645 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4646 && SvTYPE(sstr) >= SVt_PVIV && len
4649 #elif defined(PERL_NEW_COPY_ON_WRITE)
4652 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4653 /* If this is a regular (non-hek) COW, only so
4654 many COW "copies" are possible. */
4655 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4656 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4657 && !(SvFLAGS(dstr) & SVf_BREAK)
4658 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4659 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4663 && !(SvFLAGS(dstr) & SVf_BREAK)
4666 /* Either it's a shared hash key, or it's suitable for
4669 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4674 if (!(sflags & SVf_IsCOW)) {
4676 # ifdef PERL_OLD_COPY_ON_WRITE
4677 /* Make the source SV into a loop of 1.
4678 (about to become 2) */
4679 SV_COW_NEXT_SV_SET(sstr, sstr);
4681 CowREFCNT(sstr) = 0;
4685 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4691 # ifdef PERL_OLD_COPY_ON_WRITE
4692 assert (SvTYPE(dstr) >= SVt_PVIV);
4693 /* SvIsCOW_normal */
4694 /* splice us in between source and next-after-source. */
4695 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4696 SV_COW_NEXT_SV_SET(sstr, dstr);
4698 if (sflags & SVf_IsCOW) {
4703 SvPV_set(dstr, SvPVX_mutable(sstr));
4708 /* SvIsCOW_shared_hash */
4709 DEBUG_C(PerlIO_printf(Perl_debug_log,
4710 "Copy on write: Sharing hash\n"));
4712 assert (SvTYPE(dstr) >= SVt_PV);
4714 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4716 SvLEN_set(dstr, len);
4717 SvCUR_set(dstr, cur);
4720 /* Failed the swipe test, and we cannot do copy-on-write either.
4721 Have to copy the string. */
4722 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4723 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4724 SvCUR_set(dstr, cur);
4725 *SvEND(dstr) = '\0';
4727 if (sflags & SVp_NOK) {
4728 SvNV_set(dstr, SvNVX(sstr));
4730 if (sflags & SVp_IOK) {
4731 SvIV_set(dstr, SvIVX(sstr));
4732 /* Must do this otherwise some other overloaded use of 0x80000000
4733 gets confused. I guess SVpbm_VALID */
4734 if (sflags & SVf_IVisUV)
4737 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4739 const MAGIC * const smg = SvVSTRING_mg(sstr);
4741 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4742 smg->mg_ptr, smg->mg_len);
4743 SvRMAGICAL_on(dstr);
4747 else if (sflags & (SVp_IOK|SVp_NOK)) {
4748 (void)SvOK_off(dstr);
4749 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4750 if (sflags & SVp_IOK) {
4751 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4752 SvIV_set(dstr, SvIVX(sstr));
4754 if (sflags & SVp_NOK) {
4755 SvNV_set(dstr, SvNVX(sstr));
4759 if (isGV_with_GP(sstr)) {
4760 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4763 (void)SvOK_off(dstr);
4765 if (SvTAINTED(sstr))
4770 =for apidoc sv_setsv_mg
4772 Like C<sv_setsv>, but also handles 'set' magic.
4778 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4780 PERL_ARGS_ASSERT_SV_SETSV_MG;
4782 sv_setsv(dstr,sstr);
4787 # ifdef PERL_OLD_COPY_ON_WRITE
4788 # define SVt_COW SVt_PVIV
4790 # define SVt_COW SVt_PV
4793 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4795 STRLEN cur = SvCUR(sstr);
4796 STRLEN len = SvLEN(sstr);
4798 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4799 const bool already = cBOOL(SvIsCOW(sstr));
4802 PERL_ARGS_ASSERT_SV_SETSV_COW;
4805 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4806 (void*)sstr, (void*)dstr);
4813 if (SvTHINKFIRST(dstr))
4814 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4815 else if (SvPVX_const(dstr))
4816 Safefree(SvPVX_mutable(dstr));
4820 SvUPGRADE(dstr, SVt_COW);
4822 assert (SvPOK(sstr));
4823 assert (SvPOKp(sstr));
4824 # ifdef PERL_OLD_COPY_ON_WRITE
4825 assert (!SvIOK(sstr));
4826 assert (!SvIOKp(sstr));
4827 assert (!SvNOK(sstr));
4828 assert (!SvNOKp(sstr));
4831 if (SvIsCOW(sstr)) {
4833 if (SvLEN(sstr) == 0) {
4834 /* source is a COW shared hash key. */
4835 DEBUG_C(PerlIO_printf(Perl_debug_log,
4836 "Fast copy on write: Sharing hash\n"));
4837 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4840 # ifdef PERL_OLD_COPY_ON_WRITE
4841 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4843 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4844 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4847 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4848 SvUPGRADE(sstr, SVt_COW);
4850 DEBUG_C(PerlIO_printf(Perl_debug_log,
4851 "Fast copy on write: Converting sstr to COW\n"));
4852 # ifdef PERL_OLD_COPY_ON_WRITE
4853 SV_COW_NEXT_SV_SET(dstr, sstr);
4855 CowREFCNT(sstr) = 0;
4858 # ifdef PERL_OLD_COPY_ON_WRITE
4859 SV_COW_NEXT_SV_SET(sstr, dstr);
4861 # ifdef PERL_DEBUG_READONLY_COW
4862 if (already) sv_buf_to_rw(sstr);
4866 new_pv = SvPVX_mutable(sstr);
4870 SvPV_set(dstr, new_pv);
4871 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4874 SvLEN_set(dstr, len);
4875 SvCUR_set(dstr, cur);
4884 =for apidoc sv_setpvn
4886 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4887 The C<len> parameter indicates the number of
4888 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4889 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4895 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4899 PERL_ARGS_ASSERT_SV_SETPVN;
4901 SV_CHECK_THINKFIRST_COW_DROP(sv);
4907 /* len is STRLEN which is unsigned, need to copy to signed */
4910 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4913 SvUPGRADE(sv, SVt_PV);
4915 dptr = SvGROW(sv, len + 1);
4916 Move(ptr,dptr,len,char);
4919 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4921 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4925 =for apidoc sv_setpvn_mg
4927 Like C<sv_setpvn>, but also handles 'set' magic.
4933 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4935 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4937 sv_setpvn(sv,ptr,len);
4942 =for apidoc sv_setpv
4944 Copies a string into an SV. The string must be terminated with a C<NUL>
4946 Does not handle 'set' magic. See C<sv_setpv_mg>.
4952 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4956 PERL_ARGS_ASSERT_SV_SETPV;
4958 SV_CHECK_THINKFIRST_COW_DROP(sv);
4964 SvUPGRADE(sv, SVt_PV);
4966 SvGROW(sv, len + 1);
4967 Move(ptr,SvPVX(sv),len+1,char);
4969 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4971 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4975 =for apidoc sv_setpv_mg
4977 Like C<sv_setpv>, but also handles 'set' magic.
4983 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4985 PERL_ARGS_ASSERT_SV_SETPV_MG;
4992 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4994 PERL_ARGS_ASSERT_SV_SETHEK;
5000 if (HEK_LEN(hek) == HEf_SVKEY) {
5001 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5004 const int flags = HEK_FLAGS(hek);
5005 if (flags & HVhek_WASUTF8) {
5006 STRLEN utf8_len = HEK_LEN(hek);
5007 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5008 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5011 } else if (flags & HVhek_UNSHARED) {
5012 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5015 else SvUTF8_off(sv);
5019 SV_CHECK_THINKFIRST_COW_DROP(sv);
5020 SvUPGRADE(sv, SVt_PV);
5022 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5023 SvCUR_set(sv, HEK_LEN(hek));
5029 else SvUTF8_off(sv);
5037 =for apidoc sv_usepvn_flags
5039 Tells an SV to use C<ptr> to find its string value. Normally the
5040 string is stored inside the SV, but sv_usepvn allows the SV to use an
5041 outside string. The C<ptr> should point to memory that was allocated
5042 by L<Newx|perlclib/Memory Management and String Handling>. It must be
5043 the start of a Newx-ed block of memory, and not a pointer to the
5044 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
5045 and not be from a non-Newx memory allocator like C<malloc>. The
5046 string length, C<len>, must be supplied. By default this function
5047 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5048 so that pointer should not be freed or used by the programmer after
5049 giving it to sv_usepvn, and neither should any pointers from "behind"
5050 that pointer (e.g. ptr + 1) be used.
5052 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
5053 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
5054 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5055 C<len>, and already meets the requirements for storing in C<SvPVX>).
5061 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5065 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5067 SV_CHECK_THINKFIRST_COW_DROP(sv);
5068 SvUPGRADE(sv, SVt_PV);
5071 if (flags & SV_SMAGIC)
5075 if (SvPVX_const(sv))
5079 if (flags & SV_HAS_TRAILING_NUL)
5080 assert(ptr[len] == '\0');
5083 allocate = (flags & SV_HAS_TRAILING_NUL)
5085 #ifdef Perl_safesysmalloc_size
5088 PERL_STRLEN_ROUNDUP(len + 1);
5090 if (flags & SV_HAS_TRAILING_NUL) {
5091 /* It's long enough - do nothing.
5092 Specifically Perl_newCONSTSUB is relying on this. */
5095 /* Force a move to shake out bugs in callers. */
5096 char *new_ptr = (char*)safemalloc(allocate);
5097 Copy(ptr, new_ptr, len, char);
5098 PoisonFree(ptr,len,char);
5102 ptr = (char*) saferealloc (ptr, allocate);
5105 #ifdef Perl_safesysmalloc_size
5106 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5108 SvLEN_set(sv, allocate);
5112 if (!(flags & SV_HAS_TRAILING_NUL)) {
5115 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5117 if (flags & SV_SMAGIC)
5121 #ifdef PERL_OLD_COPY_ON_WRITE
5122 /* Need to do this *after* making the SV normal, as we need the buffer
5123 pointer to remain valid until after we've copied it. If we let go too early,
5124 another thread could invalidate it by unsharing last of the same hash key
5125 (which it can do by means other than releasing copy-on-write Svs)
5126 or by changing the other copy-on-write SVs in the loop. */
5128 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5130 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5132 { /* this SV was SvIsCOW_normal(sv) */
5133 /* we need to find the SV pointing to us. */
5134 SV *current = SV_COW_NEXT_SV(after);
5136 if (current == sv) {
5137 /* The SV we point to points back to us (there were only two of us
5139 Hence other SV is no longer copy on write either. */
5141 sv_buf_to_rw(after);
5143 /* We need to follow the pointers around the loop. */
5145 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5148 /* don't loop forever if the structure is bust, and we have
5149 a pointer into a closed loop. */
5150 assert (current != after);
5151 assert (SvPVX_const(current) == pvx);
5153 /* Make the SV before us point to the SV after us. */
5154 SV_COW_NEXT_SV_SET(current, after);
5160 =for apidoc sv_force_normal_flags
5162 Undo various types of fakery on an SV, where fakery means
5163 "more than" a string: if the PV is a shared string, make
5164 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5165 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5166 we do the copy, and is also used locally; if this is a
5167 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5168 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5169 SvPOK_off rather than making a copy. (Used where this
5170 scalar is about to be set to some other value.) In addition,
5171 the C<flags> parameter gets passed to C<sv_unref_flags()>
5172 when unreffing. C<sv_force_normal> calls this function
5173 with flags set to 0.
5175 This function is expected to be used to signal to perl that this SV is
5176 about to be written to, and any extra book-keeping needs to be taken care
5177 of. Hence, it croaks on read-only values.
5183 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5185 assert(SvIsCOW(sv));
5188 const char * const pvx = SvPVX_const(sv);
5189 const STRLEN len = SvLEN(sv);
5190 const STRLEN cur = SvCUR(sv);
5191 # ifdef PERL_OLD_COPY_ON_WRITE
5192 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5193 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5194 we'll fail an assertion. */
5195 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5199 PerlIO_printf(Perl_debug_log,
5200 "Copy on write: Force normal %ld\n",
5205 # ifdef PERL_NEW_COPY_ON_WRITE
5207 /* Must do this first, since the CowREFCNT uses SvPVX and
5208 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5209 the only owner left of the buffer. */
5210 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5212 U8 cowrefcnt = CowREFCNT(sv);
5213 if(cowrefcnt != 0) {
5215 CowREFCNT(sv) = cowrefcnt;
5220 /* Else we are the only owner of the buffer. */
5225 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5230 if (flags & SV_COW_DROP_PV) {
5231 /* OK, so we don't need to copy our buffer. */
5234 SvGROW(sv, cur + 1);
5235 Move(pvx,SvPVX(sv),cur,char);
5240 # ifdef PERL_OLD_COPY_ON_WRITE
5241 sv_release_COW(sv, pvx, next);
5244 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5251 const char * const pvx = SvPVX_const(sv);
5252 const STRLEN len = SvCUR(sv);
5256 if (flags & SV_COW_DROP_PV) {
5257 /* OK, so we don't need to copy our buffer. */
5260 SvGROW(sv, len + 1);
5261 Move(pvx,SvPVX(sv),len,char);
5264 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5270 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5272 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5275 Perl_croak_no_modify();
5276 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5277 S_sv_uncow(aTHX_ sv, flags);
5279 sv_unref_flags(sv, flags);
5280 else if (SvFAKE(sv) && isGV_with_GP(sv))
5281 sv_unglob(sv, flags);
5282 else if (SvFAKE(sv) && isREGEXP(sv)) {
5283 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5284 to sv_unglob. We only need it here, so inline it. */
5285 const bool islv = SvTYPE(sv) == SVt_PVLV;
5286 const svtype new_type =
5287 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5288 SV *const temp = newSV_type(new_type);
5289 regexp *const temp_p = ReANY((REGEXP *)sv);
5291 if (new_type == SVt_PVMG) {
5292 SvMAGIC_set(temp, SvMAGIC(sv));
5293 SvMAGIC_set(sv, NULL);
5294 SvSTASH_set(temp, SvSTASH(sv));
5295 SvSTASH_set(sv, NULL);
5297 if (!islv) SvCUR_set(temp, SvCUR(sv));
5298 /* Remember that SvPVX is in the head, not the body. But
5299 RX_WRAPPED is in the body. */
5300 assert(ReANY((REGEXP *)sv)->mother_re);
5301 /* Their buffer is already owned by someone else. */
5302 if (flags & SV_COW_DROP_PV) {
5303 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5304 zeroed body. For SVt_PVLV, it should have been set to 0
5305 before turning into a regexp. */
5306 assert(!SvLEN(islv ? sv : temp));
5307 sv->sv_u.svu_pv = 0;
5310 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5311 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5315 /* Now swap the rest of the bodies. */
5319 SvFLAGS(sv) &= ~SVTYPEMASK;
5320 SvFLAGS(sv) |= new_type;
5321 SvANY(sv) = SvANY(temp);
5324 SvFLAGS(temp) &= ~(SVTYPEMASK);
5325 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5326 SvANY(temp) = temp_p;
5327 temp->sv_u.svu_rx = (regexp *)temp_p;
5329 SvREFCNT_dec_NN(temp);
5331 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5337 Efficient removal of characters from the beginning of the string buffer.
5338 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5339 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5340 character of the adjusted string. Uses the "OOK hack". On return, only
5341 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5343 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5344 refer to the same chunk of data.
5346 The unfortunate similarity of this function's name to that of Perl's C<chop>
5347 operator is strictly coincidental. This function works from the left;
5348 C<chop> works from the right.
5354 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5365 PERL_ARGS_ASSERT_SV_CHOP;
5367 if (!ptr || !SvPOKp(sv))
5369 delta = ptr - SvPVX_const(sv);
5371 /* Nothing to do. */
5374 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5375 if (delta > max_delta)
5376 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5377 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5378 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5379 SV_CHECK_THINKFIRST(sv);
5380 SvPOK_only_UTF8(sv);
5383 if (!SvLEN(sv)) { /* make copy of shared string */
5384 const char *pvx = SvPVX_const(sv);
5385 const STRLEN len = SvCUR(sv);
5386 SvGROW(sv, len + 1);
5387 Move(pvx,SvPVX(sv),len,char);
5393 SvOOK_offset(sv, old_delta);
5395 SvLEN_set(sv, SvLEN(sv) - delta);
5396 SvCUR_set(sv, SvCUR(sv) - delta);
5397 SvPV_set(sv, SvPVX(sv) + delta);
5399 p = (U8 *)SvPVX_const(sv);
5402 /* how many bytes were evacuated? we will fill them with sentinel
5403 bytes, except for the part holding the new offset of course. */
5406 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5408 assert(evacn <= delta + old_delta);
5412 /* This sets 'delta' to the accumulated value of all deltas so far */
5416 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5417 * the string; otherwise store a 0 byte there and store 'delta' just prior
5418 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5419 * portion of the chopped part of the string */
5420 if (delta < 0x100) {
5424 p -= sizeof(STRLEN);
5425 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5429 /* Fill the preceding buffer with sentinals to verify that no-one is
5439 =for apidoc sv_catpvn
5441 Concatenates the string onto the end of the string which is in the SV. The
5442 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5443 status set, then the bytes appended should be valid UTF-8.
5444 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5446 =for apidoc sv_catpvn_flags
5448 Concatenates the string onto the end of the string which is in the SV. The
5449 C<len> indicates number of bytes to copy.
5451 By default, the string appended is assumed to be valid UTF-8 if the SV has
5452 the UTF-8 status set, and a string of bytes otherwise. One can force the
5453 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5454 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5455 string appended will be upgraded to UTF-8 if necessary.
5457 If C<flags> has the C<SV_SMAGIC> bit set, will
5458 C<mg_set> on C<dsv> afterwards if appropriate.
5459 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5460 in terms of this function.
5466 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5469 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5471 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5472 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5474 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5475 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5476 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5479 else SvGROW(dsv, dlen + slen + 1);
5481 sstr = SvPVX_const(dsv);
5482 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5483 SvCUR_set(dsv, SvCUR(dsv) + slen);
5486 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5487 const char * const send = sstr + slen;
5490 /* Something this code does not account for, which I think is
5491 impossible; it would require the same pv to be treated as
5492 bytes *and* utf8, which would indicate a bug elsewhere. */
5493 assert(sstr != dstr);
5495 SvGROW(dsv, dlen + slen * 2 + 1);
5496 d = (U8 *)SvPVX(dsv) + dlen;
5498 while (sstr < send) {
5499 append_utf8_from_native_byte(*sstr, &d);
5502 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5505 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5507 if (flags & SV_SMAGIC)
5512 =for apidoc sv_catsv
5514 Concatenates the string from SV C<ssv> onto the end of the string in SV
5515 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5516 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5519 =for apidoc sv_catsv_flags
5521 Concatenates the string from SV C<ssv> onto the end of the string in SV
5522 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5523 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5524 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5525 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5526 and C<sv_catsv_mg> are implemented in terms of this function.
5531 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5533 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5537 const char *spv = SvPV_flags_const(ssv, slen, flags);
5538 if (flags & SV_GMAGIC)
5540 sv_catpvn_flags(dsv, spv, slen,
5541 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5542 if (flags & SV_SMAGIC)
5548 =for apidoc sv_catpv
5550 Concatenates the C<NUL>-terminated string onto the end of the string which is
5552 If the SV has the UTF-8 status set, then the bytes appended should be
5553 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5558 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5564 PERL_ARGS_ASSERT_SV_CATPV;
5568 junk = SvPV_force(sv, tlen);
5570 SvGROW(sv, tlen + len + 1);
5572 ptr = SvPVX_const(sv);
5573 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5574 SvCUR_set(sv, SvCUR(sv) + len);
5575 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5580 =for apidoc sv_catpv_flags
5582 Concatenates the C<NUL>-terminated string onto the end of the string which is
5584 If the SV has the UTF-8 status set, then the bytes appended should
5585 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5586 on the modified SV if appropriate.
5592 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5594 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5595 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5599 =for apidoc sv_catpv_mg
5601 Like C<sv_catpv>, but also handles 'set' magic.
5607 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5609 PERL_ARGS_ASSERT_SV_CATPV_MG;
5618 Creates a new SV. A non-zero C<len> parameter indicates the number of
5619 bytes of preallocated string space the SV should have. An extra byte for a
5620 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5621 space is allocated.) The reference count for the new SV is set to 1.
5623 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5624 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5625 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5626 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5627 modules supporting older perls.
5633 Perl_newSV(pTHX_ const STRLEN len)
5639 sv_grow(sv, len + 1);
5644 =for apidoc sv_magicext
5646 Adds magic to an SV, upgrading it if necessary. Applies the
5647 supplied vtable and returns a pointer to the magic added.
5649 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5650 In particular, you can add magic to SvREADONLY SVs, and add more than
5651 one instance of the same 'how'.
5653 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5654 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5655 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5656 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5658 (This is now used as a subroutine by C<sv_magic>.)
5663 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5664 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5668 PERL_ARGS_ASSERT_SV_MAGICEXT;
5670 SvUPGRADE(sv, SVt_PVMG);
5671 Newxz(mg, 1, MAGIC);
5672 mg->mg_moremagic = SvMAGIC(sv);
5673 SvMAGIC_set(sv, mg);
5675 /* Sometimes a magic contains a reference loop, where the sv and
5676 object refer to each other. To prevent a reference loop that
5677 would prevent such objects being freed, we look for such loops
5678 and if we find one we avoid incrementing the object refcount.
5680 Note we cannot do this to avoid self-tie loops as intervening RV must
5681 have its REFCNT incremented to keep it in existence.
5684 if (!obj || obj == sv ||
5685 how == PERL_MAGIC_arylen ||
5686 how == PERL_MAGIC_symtab ||
5687 (SvTYPE(obj) == SVt_PVGV &&
5688 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5689 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5690 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5695 mg->mg_obj = SvREFCNT_inc_simple(obj);
5696 mg->mg_flags |= MGf_REFCOUNTED;
5699 /* Normal self-ties simply pass a null object, and instead of
5700 using mg_obj directly, use the SvTIED_obj macro to produce a
5701 new RV as needed. For glob "self-ties", we are tieing the PVIO
5702 with an RV obj pointing to the glob containing the PVIO. In
5703 this case, to avoid a reference loop, we need to weaken the
5707 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5708 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5714 mg->mg_len = namlen;
5717 mg->mg_ptr = savepvn(name, namlen);
5718 else if (namlen == HEf_SVKEY) {
5719 /* Yes, this is casting away const. This is only for the case of
5720 HEf_SVKEY. I think we need to document this aberation of the
5721 constness of the API, rather than making name non-const, as
5722 that change propagating outwards a long way. */
5723 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5725 mg->mg_ptr = (char *) name;
5727 mg->mg_virtual = (MGVTBL *) vtable;
5734 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5736 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5737 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5738 /* This sv is only a delegate. //g magic must be attached to
5743 #ifdef PERL_OLD_COPY_ON_WRITE
5745 sv_force_normal_flags(sv, 0);
5747 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5748 &PL_vtbl_mglob, 0, 0);
5752 =for apidoc sv_magic
5754 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5755 necessary, then adds a new magic item of type C<how> to the head of the
5758 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5759 handling of the C<name> and C<namlen> arguments.
5761 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5762 to add more than one instance of the same 'how'.
5768 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5769 const char *const name, const I32 namlen)
5771 const MGVTBL *vtable;
5774 unsigned int vtable_index;
5776 PERL_ARGS_ASSERT_SV_MAGIC;
5778 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5779 || ((flags = PL_magic_data[how]),
5780 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5781 > magic_vtable_max))
5782 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5784 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5785 Useful for attaching extension internal data to perl vars.
5786 Note that multiple extensions may clash if magical scalars
5787 etc holding private data from one are passed to another. */
5789 vtable = (vtable_index == magic_vtable_max)
5790 ? NULL : PL_magic_vtables + vtable_index;
5792 #ifdef PERL_OLD_COPY_ON_WRITE
5794 sv_force_normal_flags(sv, 0);
5796 if (SvREADONLY(sv)) {
5798 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5801 Perl_croak_no_modify();
5804 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5805 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5806 /* sv_magic() refuses to add a magic of the same 'how' as an
5809 if (how == PERL_MAGIC_taint)
5815 /* Force pos to be stored as characters, not bytes. */
5816 if (SvMAGICAL(sv) && DO_UTF8(sv)
5817 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5819 && mg->mg_flags & MGf_BYTES) {
5820 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5822 mg->mg_flags &= ~MGf_BYTES;
5825 /* Rest of work is done else where */
5826 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5829 case PERL_MAGIC_taint:
5832 case PERL_MAGIC_ext:
5833 case PERL_MAGIC_dbfile:
5840 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5847 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5849 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5850 for (mg = *mgp; mg; mg = *mgp) {
5851 const MGVTBL* const virt = mg->mg_virtual;
5852 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5853 *mgp = mg->mg_moremagic;
5854 if (virt && virt->svt_free)
5855 virt->svt_free(aTHX_ sv, mg);
5856 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5858 Safefree(mg->mg_ptr);
5859 else if (mg->mg_len == HEf_SVKEY)
5860 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5861 else if (mg->mg_type == PERL_MAGIC_utf8)
5862 Safefree(mg->mg_ptr);
5864 if (mg->mg_flags & MGf_REFCOUNTED)
5865 SvREFCNT_dec(mg->mg_obj);
5869 mgp = &mg->mg_moremagic;
5872 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5873 mg_magical(sv); /* else fix the flags now */
5877 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5883 =for apidoc sv_unmagic
5885 Removes all magic of type C<type> from an SV.
5891 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5893 PERL_ARGS_ASSERT_SV_UNMAGIC;
5894 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5898 =for apidoc sv_unmagicext
5900 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5906 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5908 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5909 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5913 =for apidoc sv_rvweaken
5915 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5916 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5917 push a back-reference to this RV onto the array of backreferences
5918 associated with that magic. If the RV is magical, set magic will be
5919 called after the RV is cleared.
5925 Perl_sv_rvweaken(pTHX_ SV *const sv)
5929 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5931 if (!SvOK(sv)) /* let undefs pass */
5934 Perl_croak(aTHX_ "Can't weaken a nonreference");
5935 else if (SvWEAKREF(sv)) {
5936 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5939 else if (SvREADONLY(sv)) croak_no_modify();
5941 Perl_sv_add_backref(aTHX_ tsv, sv);
5943 SvREFCNT_dec_NN(tsv);
5948 =for apidoc sv_get_backrefs
5950 If the sv is the target of a weak reference then it returns the back
5951 references structure associated with the sv; otherwise return NULL.
5953 When returning a non-null result the type of the return is relevant. If it
5954 is an AV then the elements of the AV are the weak reference RVs which
5955 point at this item. If it is any other type then the item itself is the
5958 See also Perl_sv_add_backref(), Perl_sv_del_backref(),
5959 Perl_sv_kill_backrefs()
5965 Perl_sv_get_backrefs(SV *const sv)
5969 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5971 /* find slot to store array or singleton backref */
5973 if (SvTYPE(sv) == SVt_PVHV) {
5975 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5976 backrefs = (SV *)iter->xhv_backreferences;
5978 } else if (SvMAGICAL(sv)) {
5979 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5981 backrefs = mg->mg_obj;
5986 /* Give tsv backref magic if it hasn't already got it, then push a
5987 * back-reference to sv onto the array associated with the backref magic.
5989 * As an optimisation, if there's only one backref and it's not an AV,
5990 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5991 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5995 /* A discussion about the backreferences array and its refcount:
5997 * The AV holding the backreferences is pointed to either as the mg_obj of
5998 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5999 * xhv_backreferences field. The array is created with a refcount
6000 * of 2. This means that if during global destruction the array gets
6001 * picked on before its parent to have its refcount decremented by the
6002 * random zapper, it won't actually be freed, meaning it's still there for
6003 * when its parent gets freed.
6005 * When the parent SV is freed, the extra ref is killed by
6006 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6007 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6009 * When a single backref SV is stored directly, it is not reference
6014 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6020 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6022 /* find slot to store array or singleton backref */
6024 if (SvTYPE(tsv) == SVt_PVHV) {
6025 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6028 mg = mg_find(tsv, PERL_MAGIC_backref);
6030 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6031 svp = &(mg->mg_obj);
6034 /* create or retrieve the array */
6036 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6037 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6041 mg->mg_flags |= MGf_REFCOUNTED;
6044 SvREFCNT_inc_simple_void_NN(av);
6045 /* av now has a refcnt of 2; see discussion above */
6046 av_extend(av, *svp ? 2 : 1);
6048 /* move single existing backref to the array */
6049 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6054 av = MUTABLE_AV(*svp);
6056 /* optimisation: store single backref directly in HvAUX or mg_obj */
6060 assert(SvTYPE(av) == SVt_PVAV);
6061 if (AvFILLp(av) >= AvMAX(av)) {
6062 av_extend(av, AvFILLp(av)+1);
6065 /* push new backref */
6066 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6069 /* delete a back-reference to ourselves from the backref magic associated
6070 * with the SV we point to.
6074 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6078 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6080 if (SvTYPE(tsv) == SVt_PVHV) {
6082 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6084 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6085 /* It's possible for the the last (strong) reference to tsv to have
6086 become freed *before* the last thing holding a weak reference.
6087 If both survive longer than the backreferences array, then when
6088 the referent's reference count drops to 0 and it is freed, it's
6089 not able to chase the backreferences, so they aren't NULLed.
6091 For example, a CV holds a weak reference to its stash. If both the
6092 CV and the stash survive longer than the backreferences array,
6093 and the CV gets picked for the SvBREAK() treatment first,
6094 *and* it turns out that the stash is only being kept alive because
6095 of an our variable in the pad of the CV, then midway during CV
6096 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6097 It ends up pointing to the freed HV. Hence it's chased in here, and
6098 if this block wasn't here, it would hit the !svp panic just below.
6100 I don't believe that "better" destruction ordering is going to help
6101 here - during global destruction there's always going to be the
6102 chance that something goes out of order. We've tried to make it
6103 foolproof before, and it only resulted in evolutionary pressure on
6104 fools. Which made us look foolish for our hubris. :-(
6110 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6111 svp = mg ? &(mg->mg_obj) : NULL;
6115 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6117 /* It's possible that sv is being freed recursively part way through the
6118 freeing of tsv. If this happens, the backreferences array of tsv has
6119 already been freed, and so svp will be NULL. If this is the case,
6120 we should not panic. Instead, nothing needs doing, so return. */
6121 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6123 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6124 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6127 if (SvTYPE(*svp) == SVt_PVAV) {
6131 AV * const av = (AV*)*svp;
6133 assert(!SvIS_FREED(av));
6137 /* for an SV with N weak references to it, if all those
6138 * weak refs are deleted, then sv_del_backref will be called
6139 * N times and O(N^2) compares will be done within the backref
6140 * array. To ameliorate this potential slowness, we:
6141 * 1) make sure this code is as tight as possible;
6142 * 2) when looking for SV, look for it at both the head and tail of the
6143 * array first before searching the rest, since some create/destroy
6144 * patterns will cause the backrefs to be freed in order.
6151 SV **p = &svp[fill];
6152 SV *const topsv = *p;
6159 /* We weren't the last entry.
6160 An unordered list has this property that you
6161 can take the last element off the end to fill
6162 the hole, and it's still an unordered list :-)
6168 break; /* should only be one */
6175 AvFILLp(av) = fill-1;
6177 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6178 /* freed AV; skip */
6181 /* optimisation: only a single backref, stored directly */
6183 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6184 (void*)*svp, (void*)sv);
6191 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6197 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6202 /* after multiple passes through Perl_sv_clean_all() for a thingy
6203 * that has badly leaked, the backref array may have gotten freed,
6204 * since we only protect it against 1 round of cleanup */
6205 if (SvIS_FREED(av)) {
6206 if (PL_in_clean_all) /* All is fair */
6209 "panic: magic_killbackrefs (freed backref AV/SV)");
6213 is_array = (SvTYPE(av) == SVt_PVAV);
6215 assert(!SvIS_FREED(av));
6218 last = svp + AvFILLp(av);
6221 /* optimisation: only a single backref, stored directly */
6227 while (svp <= last) {
6229 SV *const referrer = *svp;
6230 if (SvWEAKREF(referrer)) {
6231 /* XXX Should we check that it hasn't changed? */
6232 assert(SvROK(referrer));
6233 SvRV_set(referrer, 0);
6235 SvWEAKREF_off(referrer);
6236 SvSETMAGIC(referrer);
6237 } else if (SvTYPE(referrer) == SVt_PVGV ||
6238 SvTYPE(referrer) == SVt_PVLV) {
6239 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6240 /* You lookin' at me? */
6241 assert(GvSTASH(referrer));
6242 assert(GvSTASH(referrer) == (const HV *)sv);
6243 GvSTASH(referrer) = 0;
6244 } else if (SvTYPE(referrer) == SVt_PVCV ||
6245 SvTYPE(referrer) == SVt_PVFM) {
6246 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6247 /* You lookin' at me? */
6248 assert(CvSTASH(referrer));
6249 assert(CvSTASH(referrer) == (const HV *)sv);
6250 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6253 assert(SvTYPE(sv) == SVt_PVGV);
6254 /* You lookin' at me? */
6255 assert(CvGV(referrer));
6256 assert(CvGV(referrer) == (const GV *)sv);
6257 anonymise_cv_maybe(MUTABLE_GV(sv),
6258 MUTABLE_CV(referrer));
6263 "panic: magic_killbackrefs (flags=%"UVxf")",
6264 (UV)SvFLAGS(referrer));
6275 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6281 =for apidoc sv_insert
6283 Inserts a string at the specified offset/length within the SV. Similar to
6284 the Perl substr() function. Handles get magic.
6286 =for apidoc sv_insert_flags
6288 Same as C<sv_insert>, but the extra C<flags> are passed to the
6289 C<SvPV_force_flags> that applies to C<bigstr>.
6295 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6301 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6304 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6306 SvPV_force_flags(bigstr, curlen, flags);
6307 (void)SvPOK_only_UTF8(bigstr);
6308 if (offset + len > curlen) {
6309 SvGROW(bigstr, offset+len+1);
6310 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6311 SvCUR_set(bigstr, offset+len);
6315 i = littlelen - len;
6316 if (i > 0) { /* string might grow */
6317 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6318 mid = big + offset + len;
6319 midend = bigend = big + SvCUR(bigstr);
6322 while (midend > mid) /* shove everything down */
6323 *--bigend = *--midend;
6324 Move(little,big+offset,littlelen,char);
6325 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6330 Move(little,SvPVX(bigstr)+offset,len,char);
6335 big = SvPVX(bigstr);
6338 bigend = big + SvCUR(bigstr);
6340 if (midend > bigend)
6341 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6344 if (mid - big > bigend - midend) { /* faster to shorten from end */
6346 Move(little, mid, littlelen,char);
6349 i = bigend - midend;
6351 Move(midend, mid, i,char);
6355 SvCUR_set(bigstr, mid - big);
6357 else if ((i = mid - big)) { /* faster from front */
6358 midend -= littlelen;
6360 Move(big, midend - i, i, char);
6361 sv_chop(bigstr,midend-i);
6363 Move(little, mid, littlelen,char);
6365 else if (littlelen) {
6366 midend -= littlelen;
6367 sv_chop(bigstr,midend);
6368 Move(little,midend,littlelen,char);
6371 sv_chop(bigstr,midend);
6377 =for apidoc sv_replace
6379 Make the first argument a copy of the second, then delete the original.
6380 The target SV physically takes over ownership of the body of the source SV
6381 and inherits its flags; however, the target keeps any magic it owns,
6382 and any magic in the source is discarded.
6383 Note that this is a rather specialist SV copying operation; most of the
6384 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6390 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6392 const U32 refcnt = SvREFCNT(sv);
6394 PERL_ARGS_ASSERT_SV_REPLACE;
6396 SV_CHECK_THINKFIRST_COW_DROP(sv);
6397 if (SvREFCNT(nsv) != 1) {
6398 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6399 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6401 if (SvMAGICAL(sv)) {
6405 sv_upgrade(nsv, SVt_PVMG);
6406 SvMAGIC_set(nsv, SvMAGIC(sv));
6407 SvFLAGS(nsv) |= SvMAGICAL(sv);
6409 SvMAGIC_set(sv, NULL);
6413 assert(!SvREFCNT(sv));
6414 #ifdef DEBUG_LEAKING_SCALARS
6415 sv->sv_flags = nsv->sv_flags;
6416 sv->sv_any = nsv->sv_any;
6417 sv->sv_refcnt = nsv->sv_refcnt;
6418 sv->sv_u = nsv->sv_u;
6420 StructCopy(nsv,sv,SV);
6422 if(SvTYPE(sv) == SVt_IV) {
6423 SET_SVANY_FOR_BODYLESS_IV(sv);
6427 #ifdef PERL_OLD_COPY_ON_WRITE
6428 if (SvIsCOW_normal(nsv)) {
6429 /* We need to follow the pointers around the loop to make the
6430 previous SV point to sv, rather than nsv. */
6433 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6436 assert(SvPVX_const(current) == SvPVX_const(nsv));
6438 /* Make the SV before us point to the SV after us. */
6440 PerlIO_printf(Perl_debug_log, "previous is\n");
6442 PerlIO_printf(Perl_debug_log,
6443 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6444 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6446 SV_COW_NEXT_SV_SET(current, sv);
6449 SvREFCNT(sv) = refcnt;
6450 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6455 /* We're about to free a GV which has a CV that refers back to us.
6456 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6460 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6465 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6468 assert(SvREFCNT(gv) == 0);
6469 assert(isGV(gv) && isGV_with_GP(gv));
6471 assert(!CvANON(cv));
6472 assert(CvGV(cv) == gv);
6473 assert(!CvNAMED(cv));
6475 /* will the CV shortly be freed by gp_free() ? */
6476 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6477 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6481 /* if not, anonymise: */
6482 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6483 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6484 : newSVpvn_flags( "__ANON__", 8, 0 );
6485 sv_catpvs(gvname, "::__ANON__");
6486 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6487 SvREFCNT_dec_NN(gvname);
6491 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6496 =for apidoc sv_clear
6498 Clear an SV: call any destructors, free up any memory used by the body,
6499 and free the body itself. The SV's head is I<not> freed, although
6500 its type is set to all 1's so that it won't inadvertently be assumed
6501 to be live during global destruction etc.
6502 This function should only be called when REFCNT is zero. Most of the time
6503 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6510 Perl_sv_clear(pTHX_ SV *const orig_sv)
6515 const struct body_details *sv_type_details;
6519 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6520 Not strictly necessary */
6522 PERL_ARGS_ASSERT_SV_CLEAR;
6524 /* within this loop, sv is the SV currently being freed, and
6525 * iter_sv is the most recent AV or whatever that's being iterated
6526 * over to provide more SVs */
6532 assert(SvREFCNT(sv) == 0);
6533 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6535 if (type <= SVt_IV) {
6536 /* See the comment in sv.h about the collusion between this
6537 * early return and the overloading of the NULL slots in the
6541 SvFLAGS(sv) &= SVf_BREAK;
6542 SvFLAGS(sv) |= SVTYPEMASK;
6546 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6547 for another purpose */
6548 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6550 if (type >= SVt_PVMG) {
6552 if (!curse(sv, 1)) goto get_next_sv;
6553 type = SvTYPE(sv); /* destructor may have changed it */
6555 /* Free back-references before magic, in case the magic calls
6556 * Perl code that has weak references to sv. */
6557 if (type == SVt_PVHV) {
6558 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6562 else if (SvMAGIC(sv)) {
6563 /* Free back-references before other types of magic. */
6564 sv_unmagic(sv, PERL_MAGIC_backref);
6570 /* case SVt_INVLIST: */
6573 IoIFP(sv) != PerlIO_stdin() &&
6574 IoIFP(sv) != PerlIO_stdout() &&
6575 IoIFP(sv) != PerlIO_stderr() &&
6576 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6578 io_close(MUTABLE_IO(sv), NULL, FALSE,
6579 (IoTYPE(sv) == IoTYPE_WRONLY ||
6580 IoTYPE(sv) == IoTYPE_RDWR ||
6581 IoTYPE(sv) == IoTYPE_APPEND));
6583 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6584 PerlDir_close(IoDIRP(sv));
6585 IoDIRP(sv) = (DIR*)NULL;
6586 Safefree(IoTOP_NAME(sv));
6587 Safefree(IoFMT_NAME(sv));
6588 Safefree(IoBOTTOM_NAME(sv));
6589 if ((const GV *)sv == PL_statgv)
6593 /* FIXME for plugins */
6595 pregfree2((REGEXP*) sv);
6599 cv_undef(MUTABLE_CV(sv));
6600 /* If we're in a stash, we don't own a reference to it.
6601 * However it does have a back reference to us, which needs to
6603 if ((stash = CvSTASH(sv)))
6604 sv_del_backref(MUTABLE_SV(stash), sv);
6607 if (PL_last_swash_hv == (const HV *)sv) {
6608 PL_last_swash_hv = NULL;
6610 if (HvTOTALKEYS((HV*)sv) > 0) {
6612 /* this statement should match the one at the beginning of
6613 * hv_undef_flags() */
6614 if ( PL_phase != PERL_PHASE_DESTRUCT
6615 && (hek = HvNAME_HEK((HV*)sv)))
6617 if (PL_stashcache) {
6618 DEBUG_o(Perl_deb(aTHX_
6619 "sv_clear clearing PL_stashcache for '%"HEKf
6622 (void)hv_deletehek(PL_stashcache,
6625 hv_name_set((HV*)sv, NULL, 0, 0);
6628 /* save old iter_sv in unused SvSTASH field */
6629 assert(!SvOBJECT(sv));
6630 SvSTASH(sv) = (HV*)iter_sv;
6633 /* save old hash_index in unused SvMAGIC field */
6634 assert(!SvMAGICAL(sv));
6635 assert(!SvMAGIC(sv));
6636 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6639 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6640 goto get_next_sv; /* process this new sv */
6642 /* free empty hash */
6643 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6644 assert(!HvARRAY((HV*)sv));
6648 AV* av = MUTABLE_AV(sv);
6649 if (PL_comppad == av) {
6653 if (AvREAL(av) && AvFILLp(av) > -1) {
6654 next_sv = AvARRAY(av)[AvFILLp(av)--];
6655 /* save old iter_sv in top-most slot of AV,
6656 * and pray that it doesn't get wiped in the meantime */
6657 AvARRAY(av)[AvMAX(av)] = iter_sv;
6659 goto get_next_sv; /* process this new sv */
6661 Safefree(AvALLOC(av));
6666 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6667 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6668 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6669 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6671 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6672 SvREFCNT_dec(LvTARG(sv));
6673 if (isREGEXP(sv)) goto freeregexp;
6676 if (isGV_with_GP(sv)) {
6677 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6678 && HvENAME_get(stash))
6679 mro_method_changed_in(stash);
6680 gp_free(MUTABLE_GV(sv));
6682 unshare_hek(GvNAME_HEK(sv));
6683 /* If we're in a stash, we don't own a reference to it.
6684 * However it does have a back reference to us, which
6685 * needs to be cleared. */
6686 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6687 sv_del_backref(MUTABLE_SV(stash), sv);
6689 /* FIXME. There are probably more unreferenced pointers to SVs
6690 * in the interpreter struct that we should check and tidy in
6691 * a similar fashion to this: */
6692 /* See also S_sv_unglob, which does the same thing. */
6693 if ((const GV *)sv == PL_last_in_gv)
6694 PL_last_in_gv = NULL;
6695 else if ((const GV *)sv == PL_statgv)
6697 else if ((const GV *)sv == PL_stderrgv)
6706 /* Don't bother with SvOOK_off(sv); as we're only going to
6710 SvOOK_offset(sv, offset);
6711 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6712 /* Don't even bother with turning off the OOK flag. */
6717 SV * const target = SvRV(sv);
6719 sv_del_backref(target, sv);
6725 else if (SvPVX_const(sv)
6726 && !(SvTYPE(sv) == SVt_PVIO
6727 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6731 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6735 # ifdef PERL_OLD_COPY_ON_WRITE
6736 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6738 if (CowREFCNT(sv)) {
6746 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6750 # ifdef PERL_OLD_COPY_ON_WRITE
6754 Safefree(SvPVX_mutable(sv));
6758 else if (SvPVX_const(sv) && SvLEN(sv)
6759 && !(SvTYPE(sv) == SVt_PVIO
6760 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6761 Safefree(SvPVX_mutable(sv));
6762 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6763 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6773 SvFLAGS(sv) &= SVf_BREAK;
6774 SvFLAGS(sv) |= SVTYPEMASK;
6776 sv_type_details = bodies_by_type + type;
6777 if (sv_type_details->arena) {
6778 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6779 &PL_body_roots[type]);
6781 else if (sv_type_details->body_size) {
6782 safefree(SvANY(sv));
6786 /* caller is responsible for freeing the head of the original sv */
6787 if (sv != orig_sv && !SvREFCNT(sv))
6790 /* grab and free next sv, if any */
6798 else if (!iter_sv) {
6800 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6801 AV *const av = (AV*)iter_sv;
6802 if (AvFILLp(av) > -1) {
6803 sv = AvARRAY(av)[AvFILLp(av)--];
6805 else { /* no more elements of current AV to free */
6808 /* restore previous value, squirrelled away */
6809 iter_sv = AvARRAY(av)[AvMAX(av)];
6810 Safefree(AvALLOC(av));
6813 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6814 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6815 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6816 /* no more elements of current HV to free */
6819 /* Restore previous values of iter_sv and hash_index,
6820 * squirrelled away */
6821 assert(!SvOBJECT(sv));
6822 iter_sv = (SV*)SvSTASH(sv);
6823 assert(!SvMAGICAL(sv));
6824 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6826 /* perl -DA does not like rubbish in SvMAGIC. */
6830 /* free any remaining detritus from the hash struct */
6831 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6832 assert(!HvARRAY((HV*)sv));
6837 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6841 if (!SvREFCNT(sv)) {
6845 if (--(SvREFCNT(sv)))
6849 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6850 "Attempt to free temp prematurely: SV 0x%"UVxf
6851 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6855 if (SvIMMORTAL(sv)) {
6856 /* make sure SvREFCNT(sv)==0 happens very seldom */
6857 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6866 /* This routine curses the sv itself, not the object referenced by sv. So
6867 sv does not have to be ROK. */
6870 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6871 PERL_ARGS_ASSERT_CURSE;
6872 assert(SvOBJECT(sv));
6874 if (PL_defstash && /* Still have a symbol table? */
6880 stash = SvSTASH(sv);
6881 assert(SvTYPE(stash) == SVt_PVHV);
6882 if (HvNAME(stash)) {
6883 CV* destructor = NULL;
6884 assert (SvOOK(stash));
6885 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6886 if (!destructor || HvMROMETA(stash)->destroy_gen
6887 != PL_sub_generation)
6890 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6891 if (gv) destructor = GvCV(gv);
6892 if (!SvOBJECT(stash))
6895 destructor ? (HV *)destructor : ((HV *)0)+1;
6896 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6900 assert(!destructor || destructor == ((CV *)0)+1
6901 || SvTYPE(destructor) == SVt_PVCV);
6902 if (destructor && destructor != ((CV *)0)+1
6903 /* A constant subroutine can have no side effects, so
6904 don't bother calling it. */
6905 && !CvCONST(destructor)
6906 /* Don't bother calling an empty destructor or one that
6907 returns immediately. */
6908 && (CvISXSUB(destructor)
6909 || (CvSTART(destructor)
6910 && (CvSTART(destructor)->op_next->op_type
6912 && (CvSTART(destructor)->op_next->op_type
6914 || CvSTART(destructor)->op_next->op_next->op_type
6920 SV* const tmpref = newRV(sv);
6921 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6923 PUSHSTACKi(PERLSI_DESTROY);
6928 call_sv(MUTABLE_SV(destructor),
6929 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6933 if(SvREFCNT(tmpref) < 2) {
6934 /* tmpref is not kept alive! */
6936 SvRV_set(tmpref, NULL);
6939 SvREFCNT_dec_NN(tmpref);
6942 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6945 if (check_refcnt && SvREFCNT(sv)) {
6946 if (PL_in_clean_objs)
6948 "DESTROY created new reference to dead object '%"HEKf"'",
6949 HEKfARG(HvNAME_HEK(stash)));
6950 /* DESTROY gave object new lease on life */
6956 HV * const stash = SvSTASH(sv);
6957 /* Curse before freeing the stash, as freeing the stash could cause
6958 a recursive call into S_curse. */
6959 SvOBJECT_off(sv); /* Curse the object. */
6960 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6961 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6967 =for apidoc sv_newref
6969 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6976 Perl_sv_newref(pTHX_ SV *const sv)
6978 PERL_UNUSED_CONTEXT;
6987 Decrement an SV's reference count, and if it drops to zero, call
6988 C<sv_clear> to invoke destructors and free up any memory used by
6989 the body; finally, deallocate the SV's head itself.
6990 Normally called via a wrapper macro C<SvREFCNT_dec>.
6996 Perl_sv_free(pTHX_ SV *const sv)
7002 /* Private helper function for SvREFCNT_dec().
7003 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7006 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7010 PERL_ARGS_ASSERT_SV_FREE2;
7012 if (LIKELY( rc == 1 )) {
7018 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7019 "Attempt to free temp prematurely: SV 0x%"UVxf
7020 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7024 if (SvIMMORTAL(sv)) {
7025 /* make sure SvREFCNT(sv)==0 happens very seldom */
7026 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7030 if (! SvREFCNT(sv)) /* may have have been resurrected */
7035 /* handle exceptional cases */
7039 if (SvFLAGS(sv) & SVf_BREAK)
7040 /* this SV's refcnt has been artificially decremented to
7041 * trigger cleanup */
7043 if (PL_in_clean_all) /* All is fair */
7045 if (SvIMMORTAL(sv)) {
7046 /* make sure SvREFCNT(sv)==0 happens very seldom */
7047 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7050 if (ckWARN_d(WARN_INTERNAL)) {
7051 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7052 Perl_dump_sv_child(aTHX_ sv);
7054 #ifdef DEBUG_LEAKING_SCALARS
7057 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7058 if (PL_warnhook == PERL_WARNHOOK_FATAL
7059 || ckDEAD(packWARN(WARN_INTERNAL))) {
7060 /* Don't let Perl_warner cause us to escape our fate: */
7064 /* This may not return: */
7065 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7066 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7067 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7070 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7080 Returns the length of the string in the SV. Handles magic and type
7081 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
7082 gives raw access to the xpv_cur slot.
7088 Perl_sv_len(pTHX_ SV *const sv)
7095 (void)SvPV_const(sv, len);
7100 =for apidoc sv_len_utf8
7102 Returns the number of characters in the string in an SV, counting wide
7103 UTF-8 bytes as a single character. Handles magic and type coercion.
7109 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7110 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7111 * (Note that the mg_len is not the length of the mg_ptr field.
7112 * This allows the cache to store the character length of the string without
7113 * needing to malloc() extra storage to attach to the mg_ptr.)
7118 Perl_sv_len_utf8(pTHX_ SV *const sv)
7124 return sv_len_utf8_nomg(sv);
7128 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7131 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7133 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7135 if (PL_utf8cache && SvUTF8(sv)) {
7137 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7139 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7140 if (mg->mg_len != -1)
7143 /* We can use the offset cache for a headstart.
7144 The longer value is stored in the first pair. */
7145 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7147 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7151 if (PL_utf8cache < 0) {
7152 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7153 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7157 ulen = Perl_utf8_length(aTHX_ s, s + len);
7158 utf8_mg_len_cache_update(sv, &mg, ulen);
7162 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7165 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7168 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7169 STRLEN *const uoffset_p, bool *const at_end)
7171 const U8 *s = start;
7172 STRLEN uoffset = *uoffset_p;
7174 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7176 while (s < send && uoffset) {
7183 else if (s > send) {
7185 /* This is the existing behaviour. Possibly it should be a croak, as
7186 it's actually a bounds error */
7189 *uoffset_p -= uoffset;
7193 /* Given the length of the string in both bytes and UTF-8 characters, decide
7194 whether to walk forwards or backwards to find the byte corresponding to
7195 the passed in UTF-8 offset. */
7197 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7198 STRLEN uoffset, const STRLEN uend)
7200 STRLEN backw = uend - uoffset;
7202 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7204 if (uoffset < 2 * backw) {
7205 /* The assumption is that going forwards is twice the speed of going
7206 forward (that's where the 2 * backw comes from).
7207 (The real figure of course depends on the UTF-8 data.) */
7208 const U8 *s = start;
7210 while (s < send && uoffset--)
7220 while (UTF8_IS_CONTINUATION(*send))
7223 return send - start;
7226 /* For the string representation of the given scalar, find the byte
7227 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7228 give another position in the string, *before* the sought offset, which
7229 (which is always true, as 0, 0 is a valid pair of positions), which should
7230 help reduce the amount of linear searching.
7231 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7232 will be used to reduce the amount of linear searching. The cache will be
7233 created if necessary, and the found value offered to it for update. */
7235 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7236 const U8 *const send, STRLEN uoffset,
7237 STRLEN uoffset0, STRLEN boffset0)
7239 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7241 bool at_end = FALSE;
7243 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7245 assert (uoffset >= uoffset0);
7250 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7252 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7253 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7254 if ((*mgp)->mg_ptr) {
7255 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7256 if (cache[0] == uoffset) {
7257 /* An exact match. */
7260 if (cache[2] == uoffset) {
7261 /* An exact match. */
7265 if (cache[0] < uoffset) {
7266 /* The cache already knows part of the way. */
7267 if (cache[0] > uoffset0) {
7268 /* The cache knows more than the passed in pair */
7269 uoffset0 = cache[0];
7270 boffset0 = cache[1];
7272 if ((*mgp)->mg_len != -1) {
7273 /* And we know the end too. */
7275 + sv_pos_u2b_midway(start + boffset0, send,
7277 (*mgp)->mg_len - uoffset0);
7279 uoffset -= uoffset0;
7281 + sv_pos_u2b_forwards(start + boffset0,
7282 send, &uoffset, &at_end);
7283 uoffset += uoffset0;
7286 else if (cache[2] < uoffset) {
7287 /* We're between the two cache entries. */
7288 if (cache[2] > uoffset0) {
7289 /* and the cache knows more than the passed in pair */
7290 uoffset0 = cache[2];
7291 boffset0 = cache[3];
7295 + sv_pos_u2b_midway(start + boffset0,
7298 cache[0] - uoffset0);
7301 + sv_pos_u2b_midway(start + boffset0,
7304 cache[2] - uoffset0);
7308 else if ((*mgp)->mg_len != -1) {
7309 /* If we can take advantage of a passed in offset, do so. */
7310 /* In fact, offset0 is either 0, or less than offset, so don't
7311 need to worry about the other possibility. */
7313 + sv_pos_u2b_midway(start + boffset0, send,
7315 (*mgp)->mg_len - uoffset0);
7320 if (!found || PL_utf8cache < 0) {
7321 STRLEN real_boffset;
7322 uoffset -= uoffset0;
7323 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7324 send, &uoffset, &at_end);
7325 uoffset += uoffset0;
7327 if (found && PL_utf8cache < 0)
7328 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7330 boffset = real_boffset;
7333 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7335 utf8_mg_len_cache_update(sv, mgp, uoffset);
7337 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7344 =for apidoc sv_pos_u2b_flags
7346 Converts the offset from a count of UTF-8 chars from
7347 the start of the string, to a count of the equivalent number of bytes; if
7348 lenp is non-zero, it does the same to lenp, but this time starting from
7349 the offset, rather than from the start
7350 of the string. Handles type coercion.
7351 I<flags> is passed to C<SvPV_flags>, and usually should be
7352 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7358 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7359 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7360 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7365 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7372 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7374 start = (U8*)SvPV_flags(sv, len, flags);
7376 const U8 * const send = start + len;
7378 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7381 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7382 is 0, and *lenp is already set to that. */) {
7383 /* Convert the relative offset to absolute. */
7384 const STRLEN uoffset2 = uoffset + *lenp;
7385 const STRLEN boffset2
7386 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7387 uoffset, boffset) - boffset;
7401 =for apidoc sv_pos_u2b
7403 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7404 the start of the string, to a count of the equivalent number of bytes; if
7405 lenp is non-zero, it does the same to lenp, but this time starting from
7406 the offset, rather than from the start of the string. Handles magic and
7409 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7416 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7417 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7418 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7422 /* This function is subject to size and sign problems */
7425 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7427 PERL_ARGS_ASSERT_SV_POS_U2B;
7430 STRLEN ulen = (STRLEN)*lenp;
7431 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7432 SV_GMAGIC|SV_CONST_RETURN);
7435 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7436 SV_GMAGIC|SV_CONST_RETURN);
7441 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7444 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7445 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7448 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7449 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7450 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7454 (*mgp)->mg_len = ulen;
7457 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7458 byte length pairing. The (byte) length of the total SV is passed in too,
7459 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7460 may not have updated SvCUR, so we can't rely on reading it directly.
7462 The proffered utf8/byte length pairing isn't used if the cache already has
7463 two pairs, and swapping either for the proffered pair would increase the
7464 RMS of the intervals between known byte offsets.
7466 The cache itself consists of 4 STRLEN values
7467 0: larger UTF-8 offset
7468 1: corresponding byte offset
7469 2: smaller UTF-8 offset
7470 3: corresponding byte offset
7472 Unused cache pairs have the value 0, 0.
7473 Keeping the cache "backwards" means that the invariant of
7474 cache[0] >= cache[2] is maintained even with empty slots, which means that
7475 the code that uses it doesn't need to worry if only 1 entry has actually
7476 been set to non-zero. It also makes the "position beyond the end of the
7477 cache" logic much simpler, as the first slot is always the one to start
7481 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7482 const STRLEN utf8, const STRLEN blen)
7486 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7491 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7492 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7493 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7495 (*mgp)->mg_len = -1;
7499 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7500 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7501 (*mgp)->mg_ptr = (char *) cache;
7505 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7506 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7507 a pointer. Note that we no longer cache utf8 offsets on refer-
7508 ences, but this check is still a good idea, for robustness. */
7509 const U8 *start = (const U8 *) SvPVX_const(sv);
7510 const STRLEN realutf8 = utf8_length(start, start + byte);
7512 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7516 /* Cache is held with the later position first, to simplify the code
7517 that deals with unbounded ends. */
7519 ASSERT_UTF8_CACHE(cache);
7520 if (cache[1] == 0) {
7521 /* Cache is totally empty */
7524 } else if (cache[3] == 0) {
7525 if (byte > cache[1]) {
7526 /* New one is larger, so goes first. */
7527 cache[2] = cache[0];
7528 cache[3] = cache[1];
7536 /* float casts necessary? XXX */
7537 #define THREEWAY_SQUARE(a,b,c,d) \
7538 ((float)((d) - (c))) * ((float)((d) - (c))) \
7539 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7540 + ((float)((b) - (a))) * ((float)((b) - (a)))
7542 /* Cache has 2 slots in use, and we know three potential pairs.
7543 Keep the two that give the lowest RMS distance. Do the
7544 calculation in bytes simply because we always know the byte
7545 length. squareroot has the same ordering as the positive value,
7546 so don't bother with the actual square root. */
7547 if (byte > cache[1]) {
7548 /* New position is after the existing pair of pairs. */
7549 const float keep_earlier
7550 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7551 const float keep_later
7552 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7554 if (keep_later < keep_earlier) {
7555 cache[2] = cache[0];
7556 cache[3] = cache[1];
7562 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7563 float b, c, keep_earlier;
7564 if (byte > cache[3]) {
7565 /* New position is between the existing pair of pairs. */
7566 b = (float)cache[3];
7569 /* New position is before the existing pair of pairs. */
7571 c = (float)cache[3];
7573 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7574 if (byte > cache[3]) {
7575 if (keep_later < keep_earlier) {
7585 if (! (keep_later < keep_earlier)) {
7586 cache[0] = cache[2];
7587 cache[1] = cache[3];
7594 ASSERT_UTF8_CACHE(cache);
7597 /* We already know all of the way, now we may be able to walk back. The same
7598 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7599 backward is half the speed of walking forward. */
7601 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7602 const U8 *end, STRLEN endu)
7604 const STRLEN forw = target - s;
7605 STRLEN backw = end - target;
7607 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7609 if (forw < 2 * backw) {
7610 return utf8_length(s, target);
7613 while (end > target) {
7615 while (UTF8_IS_CONTINUATION(*end)) {
7624 =for apidoc sv_pos_b2u_flags
7626 Converts the offset from a count of bytes from the start of the string, to
7627 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7628 I<flags> is passed to C<SvPV_flags>, and usually should be
7629 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7635 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7636 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7641 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7644 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7650 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7652 s = (const U8*)SvPV_flags(sv, blen, flags);
7655 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7656 ", byte=%"UVuf, (UV)blen, (UV)offset);
7662 && SvTYPE(sv) >= SVt_PVMG
7663 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7666 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7667 if (cache[1] == offset) {
7668 /* An exact match. */
7671 if (cache[3] == offset) {
7672 /* An exact match. */
7676 if (cache[1] < offset) {
7677 /* We already know part of the way. */
7678 if (mg->mg_len != -1) {
7679 /* Actually, we know the end too. */
7681 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7682 s + blen, mg->mg_len - cache[0]);
7684 len = cache[0] + utf8_length(s + cache[1], send);
7687 else if (cache[3] < offset) {
7688 /* We're between the two cached pairs, so we do the calculation
7689 offset by the byte/utf-8 positions for the earlier pair,
7690 then add the utf-8 characters from the string start to
7692 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7693 s + cache[1], cache[0] - cache[2])
7697 else { /* cache[3] > offset */
7698 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7702 ASSERT_UTF8_CACHE(cache);
7704 } else if (mg->mg_len != -1) {
7705 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7709 if (!found || PL_utf8cache < 0) {
7710 const STRLEN real_len = utf8_length(s, send);
7712 if (found && PL_utf8cache < 0)
7713 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7719 utf8_mg_len_cache_update(sv, &mg, len);
7721 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7728 =for apidoc sv_pos_b2u
7730 Converts the value pointed to by offsetp from a count of bytes from the
7731 start of the string, to a count of the equivalent number of UTF-8 chars.
7732 Handles magic and type coercion.
7734 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7741 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7742 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7747 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7749 PERL_ARGS_ASSERT_SV_POS_B2U;
7754 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7755 SV_GMAGIC|SV_CONST_RETURN);
7759 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7760 STRLEN real, SV *const sv)
7762 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7764 /* As this is debugging only code, save space by keeping this test here,
7765 rather than inlining it in all the callers. */
7766 if (from_cache == real)
7769 /* Need to turn the assertions off otherwise we may recurse infinitely
7770 while printing error messages. */
7771 SAVEI8(PL_utf8cache);
7773 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7774 func, (UV) from_cache, (UV) real, SVfARG(sv));
7780 Returns a boolean indicating whether the strings in the two SVs are
7781 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7782 coerce its args to strings if necessary.
7784 =for apidoc sv_eq_flags
7786 Returns a boolean indicating whether the strings in the two SVs are
7787 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7788 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7794 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7801 SV* svrecode = NULL;
7808 /* if pv1 and pv2 are the same, second SvPV_const call may
7809 * invalidate pv1 (if we are handling magic), so we may need to
7811 if (sv1 == sv2 && flags & SV_GMAGIC
7812 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7813 pv1 = SvPV_const(sv1, cur1);
7814 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7816 pv1 = SvPV_flags_const(sv1, cur1, flags);
7824 pv2 = SvPV_flags_const(sv2, cur2, flags);
7826 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7827 /* Differing utf8ness.
7828 * Do not UTF8size the comparands as a side-effect. */
7831 svrecode = newSVpvn(pv2, cur2);
7832 sv_recode_to_utf8(svrecode, _get_encoding());
7833 pv2 = SvPV_const(svrecode, cur2);
7836 svrecode = newSVpvn(pv1, cur1);
7837 sv_recode_to_utf8(svrecode, _get_encoding());
7838 pv1 = SvPV_const(svrecode, cur1);
7840 /* Now both are in UTF-8. */
7842 SvREFCNT_dec_NN(svrecode);
7848 /* sv1 is the UTF-8 one */
7849 return bytes_cmp_utf8((const U8*)pv2, cur2,
7850 (const U8*)pv1, cur1) == 0;
7853 /* sv2 is the UTF-8 one */
7854 return bytes_cmp_utf8((const U8*)pv1, cur1,
7855 (const U8*)pv2, cur2) == 0;
7861 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7863 SvREFCNT_dec(svrecode);
7871 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7872 string in C<sv1> is less than, equal to, or greater than the string in
7873 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7874 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7876 =for apidoc sv_cmp_flags
7878 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7879 string in C<sv1> is less than, equal to, or greater than the string in
7880 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7881 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7882 also C<sv_cmp_locale_flags>.
7888 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7890 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7894 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7898 const char *pv1, *pv2;
7900 SV *svrecode = NULL;
7907 pv1 = SvPV_flags_const(sv1, cur1, flags);
7914 pv2 = SvPV_flags_const(sv2, cur2, flags);
7916 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7917 /* Differing utf8ness.
7918 * Do not UTF8size the comparands as a side-effect. */
7921 svrecode = newSVpvn(pv2, cur2);
7922 sv_recode_to_utf8(svrecode, _get_encoding());
7923 pv2 = SvPV_const(svrecode, cur2);
7926 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7927 (const U8*)pv1, cur1);
7928 return retval ? retval < 0 ? -1 : +1 : 0;
7933 svrecode = newSVpvn(pv1, cur1);
7934 sv_recode_to_utf8(svrecode, _get_encoding());
7935 pv1 = SvPV_const(svrecode, cur1);
7938 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7939 (const U8*)pv2, cur2);
7940 return retval ? retval < 0 ? -1 : +1 : 0;
7946 cmp = cur2 ? -1 : 0;
7950 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7953 cmp = retval < 0 ? -1 : 1;
7954 } else if (cur1 == cur2) {
7957 cmp = cur1 < cur2 ? -1 : 1;
7961 SvREFCNT_dec(svrecode);
7967 =for apidoc sv_cmp_locale
7969 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7970 'use bytes' aware, handles get magic, and will coerce its args to strings
7971 if necessary. See also C<sv_cmp>.
7973 =for apidoc sv_cmp_locale_flags
7975 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7976 'use bytes' aware and will coerce its args to strings if necessary. If the
7977 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7983 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7985 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7989 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7992 #ifdef USE_LOCALE_COLLATE
7998 if (PL_collation_standard)
8002 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8004 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8006 if (!pv1 || !len1) {
8017 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8020 return retval < 0 ? -1 : 1;
8023 * When the result of collation is equality, that doesn't mean
8024 * that there are no differences -- some locales exclude some
8025 * characters from consideration. So to avoid false equalities,
8026 * we use the raw string as a tiebreaker.
8033 PERL_UNUSED_ARG(flags);
8034 #endif /* USE_LOCALE_COLLATE */
8036 return sv_cmp(sv1, sv2);
8040 #ifdef USE_LOCALE_COLLATE
8043 =for apidoc sv_collxfrm
8045 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8046 C<sv_collxfrm_flags>.
8048 =for apidoc sv_collxfrm_flags
8050 Add Collate Transform magic to an SV if it doesn't already have it. If the
8051 flags contain SV_GMAGIC, it handles get-magic.
8053 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
8054 scalar data of the variable, but transformed to such a format that a normal
8055 memory comparison can be used to compare the data according to the locale
8062 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8066 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8068 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8069 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8075 Safefree(mg->mg_ptr);
8076 s = SvPV_flags_const(sv, len, flags);
8077 if ((xf = mem_collxfrm(s, len, &xlen))) {
8079 #ifdef PERL_OLD_COPY_ON_WRITE
8081 sv_force_normal_flags(sv, 0);
8083 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8097 if (mg && mg->mg_ptr) {
8099 return mg->mg_ptr + sizeof(PL_collation_ix);
8107 #endif /* USE_LOCALE_COLLATE */
8110 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8112 SV * const tsv = newSV(0);
8115 sv_gets(tsv, fp, 0);
8116 sv_utf8_upgrade_nomg(tsv);
8117 SvCUR_set(sv,append);
8120 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8124 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8127 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8128 /* Grab the size of the record we're getting */
8129 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8136 /* With a true, record-oriented file on VMS, we need to use read directly
8137 * to ensure that we respect RMS record boundaries. The user is responsible
8138 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8139 * record size) field. N.B. This is likely to produce invalid results on
8140 * varying-width character data when a record ends mid-character.
8142 fd = PerlIO_fileno(fp);
8144 && PerlLIO_fstat(fd, &st) == 0
8145 && (st.st_fab_rfm == FAB$C_VAR
8146 || st.st_fab_rfm == FAB$C_VFC
8147 || st.st_fab_rfm == FAB$C_FIX)) {
8149 bytesread = PerlLIO_read(fd, buffer, recsize);
8151 else /* in-memory file from PerlIO::Scalar
8152 * or not a record-oriented file
8156 bytesread = PerlIO_read(fp, buffer, recsize);
8158 /* At this point, the logic in sv_get() means that sv will
8159 be treated as utf-8 if the handle is utf8.
8161 if (PerlIO_isutf8(fp) && bytesread > 0) {
8162 char *bend = buffer + bytesread;
8163 char *bufp = buffer;
8164 size_t charcount = 0;
8165 bool charstart = TRUE;
8168 while (charcount < recsize) {
8169 /* count accumulated characters */
8170 while (bufp < bend) {
8172 skip = UTF8SKIP(bufp);
8174 if (bufp + skip > bend) {
8175 /* partial at the end */
8186 if (charcount < recsize) {
8188 STRLEN bufp_offset = bufp - buffer;
8189 SSize_t morebytesread;
8191 /* originally I read enough to fill any incomplete
8192 character and the first byte of the next
8193 character if needed, but if there's many
8194 multi-byte encoded characters we're going to be
8195 making a read call for every character beyond
8196 the original read size.
8198 So instead, read the rest of the character if
8199 any, and enough bytes to match at least the
8200 start bytes for each character we're going to
8204 readsize = recsize - charcount;
8206 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8207 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8208 bend = buffer + bytesread;
8209 morebytesread = PerlIO_read(fp, bend, readsize);
8210 if (morebytesread <= 0) {
8211 /* we're done, if we still have incomplete
8212 characters the check code in sv_gets() will
8215 I'd originally considered doing
8216 PerlIO_ungetc() on all but the lead
8217 character of the incomplete character, but
8218 read() doesn't do that, so I don't.
8223 /* prepare to scan some more */
8224 bytesread += morebytesread;
8225 bend = buffer + bytesread;
8226 bufp = buffer + bufp_offset;
8234 SvCUR_set(sv, bytesread + append);
8235 buffer[bytesread] = '\0';
8236 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8242 Get a line from the filehandle and store it into the SV, optionally
8243 appending to the currently-stored string. If C<append> is not 0, the
8244 line is appended to the SV instead of overwriting it. C<append> should
8245 be set to the byte offset that the appended string should start at
8246 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8252 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8262 PERL_ARGS_ASSERT_SV_GETS;
8264 if (SvTHINKFIRST(sv))
8265 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8266 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8268 However, perlbench says it's slower, because the existing swipe code
8269 is faster than copy on write.
8270 Swings and roundabouts. */
8271 SvUPGRADE(sv, SVt_PV);
8274 /* line is going to be appended to the existing buffer in the sv */
8275 if (PerlIO_isutf8(fp)) {
8277 sv_utf8_upgrade_nomg(sv);
8278 sv_pos_u2b(sv,&append,0);
8280 } else if (SvUTF8(sv)) {
8281 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8287 /* not appending - "clear" the string by setting SvCUR to 0,
8288 * the pv is still avaiable. */
8291 if (PerlIO_isutf8(fp))
8294 if (IN_PERL_COMPILETIME) {
8295 /* we always read code in line mode */
8299 else if (RsSNARF(PL_rs)) {
8300 /* If it is a regular disk file use size from stat() as estimate
8301 of amount we are going to read -- may result in mallocing
8302 more memory than we really need if the layers below reduce
8303 the size we read (e.g. CRLF or a gzip layer).
8306 int fd = PerlIO_fileno(fp);
8307 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8308 const Off_t offset = PerlIO_tell(fp);
8309 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8310 #ifdef PERL_NEW_COPY_ON_WRITE
8311 /* Add an extra byte for the sake of copy-on-write's
8312 * buffer reference count. */
8313 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8315 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8322 else if (RsRECORD(PL_rs)) {
8323 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8325 else if (RsPARA(PL_rs)) {
8331 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8332 if (PerlIO_isutf8(fp)) {
8333 rsptr = SvPVutf8(PL_rs, rslen);
8336 if (SvUTF8(PL_rs)) {
8337 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8338 Perl_croak(aTHX_ "Wide character in $/");
8341 /* extract the raw pointer to the record separator */
8342 rsptr = SvPV_const(PL_rs, rslen);
8346 /* rslast is the last character in the record separator
8347 * note we don't use rslast except when rslen is true, so the
8348 * null assign is a placeholder. */
8349 rslast = rslen ? rsptr[rslen - 1] : '\0';
8351 if (rspara) { /* have to do this both before and after */
8352 do { /* to make sure file boundaries work right */
8355 i = PerlIO_getc(fp);
8359 PerlIO_ungetc(fp,i);
8365 /* See if we know enough about I/O mechanism to cheat it ! */
8367 /* This used to be #ifdef test - it is made run-time test for ease
8368 of abstracting out stdio interface. One call should be cheap
8369 enough here - and may even be a macro allowing compile
8373 if (PerlIO_fast_gets(fp)) {
8375 * We can do buffer based IO operations on this filehandle.
8377 * This means we can bypass a lot of subcalls and process
8378 * the buffer directly, it also means we know the upper bound
8379 * on the amount of data we might read of the current buffer
8380 * into our sv. Knowing this allows us to preallocate the pv
8381 * to be able to hold that maximum, which allows us to simplify
8382 * a lot of logic. */
8385 * We're going to steal some values from the stdio struct
8386 * and put EVERYTHING in the innermost loop into registers.
8388 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8389 STRLEN bpx; /* length of the data in the target sv
8390 used to fix pointers after a SvGROW */
8391 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8392 of data left in the read-ahead buffer.
8393 If 0 then the pv buffer can hold the full
8394 amount left, otherwise this is the amount it
8397 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8398 /* An ungetc()d char is handled separately from the regular
8399 * buffer, so we getc() it back out and stuff it in the buffer.
8401 i = PerlIO_getc(fp);
8402 if (i == EOF) return 0;
8403 *(--((*fp)->_ptr)) = (unsigned char) i;
8407 /* Here is some breathtakingly efficient cheating */
8409 /* When you read the following logic resist the urge to think
8410 * of record separators that are 1 byte long. They are an
8411 * uninteresting special (simple) case.
8413 * Instead think of record separators which are at least 2 bytes
8414 * long, and keep in mind that we need to deal with such
8415 * separators when they cross a read-ahead buffer boundary.
8417 * Also consider that we need to gracefully deal with separators
8418 * that may be longer than a single read ahead buffer.
8420 * Lastly do not forget we want to copy the delimiter as well. We
8421 * are copying all data in the file _up_to_and_including_ the separator
8424 * Now that you have all that in mind here is what is happening below:
8426 * 1. When we first enter the loop we do some memory book keeping to see
8427 * how much free space there is in the target SV. (This sub assumes that
8428 * it is operating on the same SV most of the time via $_ and that it is
8429 * going to be able to reuse the same pv buffer each call.) If there is
8430 * "enough" room then we set "shortbuffered" to how much space there is
8431 * and start reading forward.
8433 * 2. When we scan forward we copy from the read-ahead buffer to the target
8434 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8435 * and the end of the of pv, as well as for the "rslast", which is the last
8436 * char of the separator.
8438 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8439 * (which has a "complete" record up to the point we saw rslast) and check
8440 * it to see if it matches the separator. If it does we are done. If it doesn't
8441 * we continue on with the scan/copy.
8443 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8444 * the IO system to read the next buffer. We do this by doing a getc(), which
8445 * returns a single char read (or EOF), and prefills the buffer, and also
8446 * allows us to find out how full the buffer is. We use this information to
8447 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8448 * the returned single char into the target sv, and then go back into scan
8451 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8452 * remaining space in the read-buffer.
8454 * Note that this code despite its twisty-turny nature is pretty darn slick.
8455 * It manages single byte separators, multi-byte cross boundary separators,
8456 * and cross-read-buffer separators cleanly and efficiently at the cost
8457 * of potentially greatly overallocating the target SV.
8463 /* get the number of bytes remaining in the read-ahead buffer
8464 * on first call on a given fp this will return 0.*/
8465 cnt = PerlIO_get_cnt(fp);
8467 /* make sure we have the room */
8468 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8469 /* Not room for all of it
8470 if we are looking for a separator and room for some
8472 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8473 /* just process what we have room for */
8474 shortbuffered = cnt - SvLEN(sv) + append + 1;
8475 cnt -= shortbuffered;
8478 /* ensure that the target sv has enough room to hold
8479 * the rest of the read-ahead buffer */
8481 /* remember that cnt can be negative */
8482 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8486 /* we have enough room to hold the full buffer, lets scream */
8490 /* extract the pointer to sv's string buffer, offset by append as necessary */
8491 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8492 /* extract the point to the read-ahead buffer */
8493 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8495 /* some trace debug output */
8496 DEBUG_P(PerlIO_printf(Perl_debug_log,
8497 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8498 DEBUG_P(PerlIO_printf(Perl_debug_log,
8499 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8501 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8502 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8506 /* if there is stuff left in the read-ahead buffer */
8508 /* if there is a separator */
8510 /* loop until we hit the end of the read-ahead buffer */
8511 while (cnt > 0) { /* this | eat */
8512 /* scan forward copying and searching for rslast as we go */
8514 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8515 goto thats_all_folks; /* screams | sed :-) */
8519 /* no separator, slurp the full buffer */
8520 Copy(ptr, bp, cnt, char); /* this | eat */
8521 bp += cnt; /* screams | dust */
8522 ptr += cnt; /* louder | sed :-) */
8524 assert (!shortbuffered);
8525 goto cannot_be_shortbuffered;
8529 if (shortbuffered) { /* oh well, must extend */
8530 /* we didnt have enough room to fit the line into the target buffer
8531 * so we must extend the target buffer and keep going */
8532 cnt = shortbuffered;
8534 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8536 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8537 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8538 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8542 cannot_be_shortbuffered:
8543 /* we need to refill the read-ahead buffer if possible */
8545 DEBUG_P(PerlIO_printf(Perl_debug_log,
8546 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8547 PTR2UV(ptr),(IV)cnt));
8548 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8550 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8551 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8552 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8553 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8556 call PerlIO_getc() to let it prefill the lookahead buffer
8558 This used to call 'filbuf' in stdio form, but as that behaves like
8559 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8560 another abstraction.
8562 Note we have to deal with the char in 'i' if we are not at EOF
8564 i = PerlIO_getc(fp); /* get more characters */
8566 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8567 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8568 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8569 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8571 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8572 cnt = PerlIO_get_cnt(fp);
8573 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8574 DEBUG_P(PerlIO_printf(Perl_debug_log,
8575 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8576 PTR2UV(ptr),(IV)cnt));
8578 if (i == EOF) /* all done for ever? */
8579 goto thats_really_all_folks;
8581 /* make sure we have enough space in the target sv */
8582 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8584 SvGROW(sv, bpx + cnt + 2);
8585 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8587 /* copy of the char we got from getc() */
8588 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8590 /* make sure we deal with the i being the last character of a separator */
8591 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8592 goto thats_all_folks;
8596 /* check if we have actually found the separator - only really applies
8598 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8599 memNE((char*)bp - rslen, rsptr, rslen))
8600 goto screamer; /* go back to the fray */
8601 thats_really_all_folks:
8603 cnt += shortbuffered;
8604 DEBUG_P(PerlIO_printf(Perl_debug_log,
8605 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8606 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8607 DEBUG_P(PerlIO_printf(Perl_debug_log,
8608 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8610 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8611 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8613 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8614 DEBUG_P(PerlIO_printf(Perl_debug_log,
8615 "Screamer: done, len=%ld, string=|%.*s|\n",
8616 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8620 /*The big, slow, and stupid way. */
8621 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8622 STDCHAR *buf = NULL;
8623 Newx(buf, 8192, STDCHAR);
8631 const STDCHAR * const bpe = buf + sizeof(buf);
8633 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8634 ; /* keep reading */
8638 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8639 /* Accommodate broken VAXC compiler, which applies U8 cast to
8640 * both args of ?: operator, causing EOF to change into 255
8643 i = (U8)buf[cnt - 1];
8649 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8651 sv_catpvn_nomg(sv, (char *) buf, cnt);
8653 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8655 if (i != EOF && /* joy */
8657 SvCUR(sv) < rslen ||
8658 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8662 * If we're reading from a TTY and we get a short read,
8663 * indicating that the user hit his EOF character, we need
8664 * to notice it now, because if we try to read from the TTY
8665 * again, the EOF condition will disappear.
8667 * The comparison of cnt to sizeof(buf) is an optimization
8668 * that prevents unnecessary calls to feof().
8672 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8676 #ifdef USE_HEAP_INSTEAD_OF_STACK
8681 if (rspara) { /* have to do this both before and after */
8682 while (i != EOF) { /* to make sure file boundaries work right */
8683 i = PerlIO_getc(fp);
8685 PerlIO_ungetc(fp,i);
8691 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8697 Auto-increment of the value in the SV, doing string to numeric conversion
8698 if necessary. Handles 'get' magic and operator overloading.
8704 Perl_sv_inc(pTHX_ SV *const sv)
8713 =for apidoc sv_inc_nomg
8715 Auto-increment of the value in the SV, doing string to numeric conversion
8716 if necessary. Handles operator overloading. Skips handling 'get' magic.
8722 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8729 if (SvTHINKFIRST(sv)) {
8730 if (SvREADONLY(sv)) {
8731 Perl_croak_no_modify();
8735 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8737 i = PTR2IV(SvRV(sv));
8741 else sv_force_normal_flags(sv, 0);
8743 flags = SvFLAGS(sv);
8744 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8745 /* It's (privately or publicly) a float, but not tested as an
8746 integer, so test it to see. */
8748 flags = SvFLAGS(sv);
8750 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8751 /* It's publicly an integer, or privately an integer-not-float */
8752 #ifdef PERL_PRESERVE_IVUV
8756 if (SvUVX(sv) == UV_MAX)
8757 sv_setnv(sv, UV_MAX_P1);
8759 (void)SvIOK_only_UV(sv);
8760 SvUV_set(sv, SvUVX(sv) + 1);
8762 if (SvIVX(sv) == IV_MAX)
8763 sv_setuv(sv, (UV)IV_MAX + 1);
8765 (void)SvIOK_only(sv);
8766 SvIV_set(sv, SvIVX(sv) + 1);
8771 if (flags & SVp_NOK) {
8772 const NV was = SvNVX(sv);
8773 if (LIKELY(!Perl_isinfnan(was)) &&
8774 NV_OVERFLOWS_INTEGERS_AT &&
8775 was >= NV_OVERFLOWS_INTEGERS_AT) {
8776 /* diag_listed_as: Lost precision when %s %f by 1 */
8777 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8778 "Lost precision when incrementing %" NVff " by 1",
8781 (void)SvNOK_only(sv);
8782 SvNV_set(sv, was + 1.0);
8786 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8787 if ((flags & SVTYPEMASK) < SVt_PVIV)
8788 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8789 (void)SvIOK_only(sv);
8794 while (isALPHA(*d)) d++;
8795 while (isDIGIT(*d)) d++;
8796 if (d < SvEND(sv)) {
8797 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8798 #ifdef PERL_PRESERVE_IVUV
8799 /* Got to punt this as an integer if needs be, but we don't issue
8800 warnings. Probably ought to make the sv_iv_please() that does
8801 the conversion if possible, and silently. */
8802 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8803 /* Need to try really hard to see if it's an integer.
8804 9.22337203685478e+18 is an integer.
8805 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8806 so $a="9.22337203685478e+18"; $a+0; $a++
8807 needs to be the same as $a="9.22337203685478e+18"; $a++
8814 /* sv_2iv *should* have made this an NV */
8815 if (flags & SVp_NOK) {
8816 (void)SvNOK_only(sv);
8817 SvNV_set(sv, SvNVX(sv) + 1.0);
8820 /* I don't think we can get here. Maybe I should assert this
8821 And if we do get here I suspect that sv_setnv will croak. NWC
8823 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8824 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8826 #endif /* PERL_PRESERVE_IVUV */
8827 if (!numtype && ckWARN(WARN_NUMERIC))
8828 not_incrementable(sv);
8829 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8833 while (d >= SvPVX_const(sv)) {
8841 /* MKS: The original code here died if letters weren't consecutive.
8842 * at least it didn't have to worry about non-C locales. The
8843 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8844 * arranged in order (although not consecutively) and that only
8845 * [A-Za-z] are accepted by isALPHA in the C locale.
8847 if (isALPHA_FOLD_NE(*d, 'z')) {
8848 do { ++*d; } while (!isALPHA(*d));
8851 *(d--) -= 'z' - 'a';
8856 *(d--) -= 'z' - 'a' + 1;
8860 /* oh,oh, the number grew */
8861 SvGROW(sv, SvCUR(sv) + 2);
8862 SvCUR_set(sv, SvCUR(sv) + 1);
8863 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8874 Auto-decrement of the value in the SV, doing string to numeric conversion
8875 if necessary. Handles 'get' magic and operator overloading.
8881 Perl_sv_dec(pTHX_ SV *const sv)
8890 =for apidoc sv_dec_nomg
8892 Auto-decrement of the value in the SV, doing string to numeric conversion
8893 if necessary. Handles operator overloading. Skips handling 'get' magic.
8899 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8905 if (SvTHINKFIRST(sv)) {
8906 if (SvREADONLY(sv)) {
8907 Perl_croak_no_modify();
8911 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8913 i = PTR2IV(SvRV(sv));
8917 else sv_force_normal_flags(sv, 0);
8919 /* Unlike sv_inc we don't have to worry about string-never-numbers
8920 and keeping them magic. But we mustn't warn on punting */
8921 flags = SvFLAGS(sv);
8922 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8923 /* It's publicly an integer, or privately an integer-not-float */
8924 #ifdef PERL_PRESERVE_IVUV
8928 if (SvUVX(sv) == 0) {
8929 (void)SvIOK_only(sv);
8933 (void)SvIOK_only_UV(sv);
8934 SvUV_set(sv, SvUVX(sv) - 1);
8937 if (SvIVX(sv) == IV_MIN) {
8938 sv_setnv(sv, (NV)IV_MIN);
8942 (void)SvIOK_only(sv);
8943 SvIV_set(sv, SvIVX(sv) - 1);
8948 if (flags & SVp_NOK) {
8951 const NV was = SvNVX(sv);
8952 if (LIKELY(!Perl_isinfnan(was)) &&
8953 NV_OVERFLOWS_INTEGERS_AT &&
8954 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8955 /* diag_listed_as: Lost precision when %s %f by 1 */
8956 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8957 "Lost precision when decrementing %" NVff " by 1",
8960 (void)SvNOK_only(sv);
8961 SvNV_set(sv, was - 1.0);
8965 if (!(flags & SVp_POK)) {
8966 if ((flags & SVTYPEMASK) < SVt_PVIV)
8967 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8969 (void)SvIOK_only(sv);
8972 #ifdef PERL_PRESERVE_IVUV
8974 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8975 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8976 /* Need to try really hard to see if it's an integer.
8977 9.22337203685478e+18 is an integer.
8978 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8979 so $a="9.22337203685478e+18"; $a+0; $a--
8980 needs to be the same as $a="9.22337203685478e+18"; $a--
8987 /* sv_2iv *should* have made this an NV */
8988 if (flags & SVp_NOK) {
8989 (void)SvNOK_only(sv);
8990 SvNV_set(sv, SvNVX(sv) - 1.0);
8993 /* I don't think we can get here. Maybe I should assert this
8994 And if we do get here I suspect that sv_setnv will croak. NWC
8996 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8997 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9000 #endif /* PERL_PRESERVE_IVUV */
9001 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9004 /* this define is used to eliminate a chunk of duplicated but shared logic
9005 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9006 * used anywhere but here - yves
9008 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9010 SSize_t ix = ++PL_tmps_ix; \
9011 if (UNLIKELY(ix >= PL_tmps_max)) \
9012 ix = tmps_grow_p(ix); \
9013 PL_tmps_stack[ix] = (AnSv); \
9017 =for apidoc sv_mortalcopy
9019 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9020 The new SV is marked as mortal. It will be destroyed "soon", either by an
9021 explicit call to FREETMPS, or by an implicit call at places such as
9022 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
9027 /* Make a string that will exist for the duration of the expression
9028 * evaluation. Actually, it may have to last longer than that, but
9029 * hopefully we won't free it until it has been assigned to a
9030 * permanent location. */
9033 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9037 if (flags & SV_GMAGIC)
9038 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9040 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9041 PUSH_EXTEND_MORTAL__SV_C(sv);
9047 =for apidoc sv_newmortal
9049 Creates a new null SV which is mortal. The reference count of the SV is
9050 set to 1. It will be destroyed "soon", either by an explicit call to
9051 FREETMPS, or by an implicit call at places such as statement boundaries.
9052 See also C<sv_mortalcopy> and C<sv_2mortal>.
9058 Perl_sv_newmortal(pTHX)
9063 SvFLAGS(sv) = SVs_TEMP;
9064 PUSH_EXTEND_MORTAL__SV_C(sv);
9070 =for apidoc newSVpvn_flags
9072 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9073 characters) into it. The reference count for the
9074 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9075 string. You are responsible for ensuring that the source string is at least
9076 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9077 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9078 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9079 returning. If C<SVf_UTF8> is set, C<s>
9080 is considered to be in UTF-8 and the
9081 C<SVf_UTF8> flag will be set on the new SV.
9082 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9084 #define newSVpvn_utf8(s, len, u) \
9085 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9091 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9095 /* All the flags we don't support must be zero.
9096 And we're new code so I'm going to assert this from the start. */
9097 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9099 sv_setpvn(sv,s,len);
9101 /* This code used to do a sv_2mortal(), however we now unroll the call to
9102 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9103 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9104 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9105 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9106 * means that we eliminate quite a few steps than it looks - Yves
9107 * (explaining patch by gfx) */
9109 SvFLAGS(sv) |= flags;
9111 if(flags & SVs_TEMP){
9112 PUSH_EXTEND_MORTAL__SV_C(sv);
9119 =for apidoc sv_2mortal
9121 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9122 by an explicit call to FREETMPS, or by an implicit call at places such as
9123 statement boundaries. SvTEMP() is turned on which means that the SV's
9124 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9125 and C<sv_mortalcopy>.
9131 Perl_sv_2mortal(pTHX_ SV *const sv)
9138 PUSH_EXTEND_MORTAL__SV_C(sv);
9146 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9147 characters) into it. The reference count for the
9148 SV is set to 1. If C<len> is zero, Perl will compute the length using
9149 strlen(), (which means if you use this option, that C<s> can't have embedded
9150 C<NUL> characters and has to have a terminating C<NUL> byte).
9152 For efficiency, consider using C<newSVpvn> instead.
9158 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9163 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9168 =for apidoc newSVpvn
9170 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9171 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9172 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9173 are responsible for ensuring that the source buffer is at least
9174 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9181 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9185 sv_setpvn(sv,buffer,len);
9190 =for apidoc newSVhek
9192 Creates a new SV from the hash key structure. It will generate scalars that
9193 point to the shared string table where possible. Returns a new (undefined)
9194 SV if the hek is NULL.
9200 Perl_newSVhek(pTHX_ const HEK *const hek)
9209 if (HEK_LEN(hek) == HEf_SVKEY) {
9210 return newSVsv(*(SV**)HEK_KEY(hek));
9212 const int flags = HEK_FLAGS(hek);
9213 if (flags & HVhek_WASUTF8) {
9215 Andreas would like keys he put in as utf8 to come back as utf8
9217 STRLEN utf8_len = HEK_LEN(hek);
9218 SV * const sv = newSV_type(SVt_PV);
9219 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9220 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9221 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9224 } else if (flags & HVhek_UNSHARED) {
9225 /* A hash that isn't using shared hash keys has to have
9226 the flag in every key so that we know not to try to call
9227 share_hek_hek on it. */
9229 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9234 /* This will be overwhelminly the most common case. */
9236 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9237 more efficient than sharepvn(). */
9241 sv_upgrade(sv, SVt_PV);
9242 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9243 SvCUR_set(sv, HEK_LEN(hek));
9255 =for apidoc newSVpvn_share
9257 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9258 table. If the string does not already exist in the table, it is
9259 created first. Turns on the SvIsCOW flag (or READONLY
9260 and FAKE in 5.16 and earlier). If the C<hash> parameter
9261 is non-zero, that value is used; otherwise the hash is computed.
9262 The string's hash can later be retrieved from the SV
9263 with the C<SvSHARED_HASH()> macro. The idea here is
9264 that as the string table is used for shared hash keys these strings will have
9265 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9271 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9275 bool is_utf8 = FALSE;
9276 const char *const orig_src = src;
9279 STRLEN tmplen = -len;
9281 /* See the note in hv.c:hv_fetch() --jhi */
9282 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9286 PERL_HASH(hash, src, len);
9288 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9289 changes here, update it there too. */
9290 sv_upgrade(sv, SVt_PV);
9291 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9298 if (src != orig_src)
9304 =for apidoc newSVpv_share
9306 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9313 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9315 return newSVpvn_share(src, strlen(src), hash);
9318 #if defined(PERL_IMPLICIT_CONTEXT)
9320 /* pTHX_ magic can't cope with varargs, so this is a no-context
9321 * version of the main function, (which may itself be aliased to us).
9322 * Don't access this version directly.
9326 Perl_newSVpvf_nocontext(const char *const pat, ...)
9332 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9334 va_start(args, pat);
9335 sv = vnewSVpvf(pat, &args);
9342 =for apidoc newSVpvf
9344 Creates a new SV and initializes it with the string formatted like
9351 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9356 PERL_ARGS_ASSERT_NEWSVPVF;
9358 va_start(args, pat);
9359 sv = vnewSVpvf(pat, &args);
9364 /* backend for newSVpvf() and newSVpvf_nocontext() */
9367 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9371 PERL_ARGS_ASSERT_VNEWSVPVF;
9374 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9381 Creates a new SV and copies a floating point value into it.
9382 The reference count for the SV is set to 1.
9388 Perl_newSVnv(pTHX_ const NV n)
9400 Creates a new SV and copies an integer into it. The reference count for the
9407 Perl_newSViv(pTHX_ const IV i)
9413 /* Inlining ONLY the small relevant subset of sv_setiv here
9414 * for performance. Makes a significant difference. */
9416 /* We're starting from SVt_FIRST, so provided that's
9417 * actual 0, we don't have to unset any SV type flags
9418 * to promote to SVt_IV. */
9419 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9421 SET_SVANY_FOR_BODYLESS_IV(sv);
9422 SvFLAGS(sv) |= SVt_IV;
9434 Creates a new SV and copies an unsigned integer into it.
9435 The reference count for the SV is set to 1.
9441 Perl_newSVuv(pTHX_ const UV u)
9445 /* Inlining ONLY the small relevant subset of sv_setuv here
9446 * for performance. Makes a significant difference. */
9448 /* Using ivs is more efficient than using uvs - see sv_setuv */
9449 if (u <= (UV)IV_MAX) {
9450 return newSViv((IV)u);
9455 /* We're starting from SVt_FIRST, so provided that's
9456 * actual 0, we don't have to unset any SV type flags
9457 * to promote to SVt_IV. */
9458 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9460 SET_SVANY_FOR_BODYLESS_IV(sv);
9461 SvFLAGS(sv) |= SVt_IV;
9463 (void)SvIsUV_on(sv);
9472 =for apidoc newSV_type
9474 Creates a new SV, of the type specified. The reference count for the new SV
9481 Perl_newSV_type(pTHX_ const svtype type)
9486 ASSUME(SvTYPE(sv) == SVt_FIRST);
9487 if(type != SVt_FIRST)
9488 sv_upgrade(sv, type);
9493 =for apidoc newRV_noinc
9495 Creates an RV wrapper for an SV. The reference count for the original
9496 SV is B<not> incremented.
9502 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9506 PERL_ARGS_ASSERT_NEWRV_NOINC;
9510 /* We're starting from SVt_FIRST, so provided that's
9511 * actual 0, we don't have to unset any SV type flags
9512 * to promote to SVt_IV. */
9513 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9515 SET_SVANY_FOR_BODYLESS_IV(sv);
9516 SvFLAGS(sv) |= SVt_IV;
9521 SvRV_set(sv, tmpRef);
9526 /* newRV_inc is the official function name to use now.
9527 * newRV_inc is in fact #defined to newRV in sv.h
9531 Perl_newRV(pTHX_ SV *const sv)
9533 PERL_ARGS_ASSERT_NEWRV;
9535 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9541 Creates a new SV which is an exact duplicate of the original SV.
9548 Perl_newSVsv(pTHX_ SV *const old)
9554 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9555 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9558 /* Do this here, otherwise we leak the new SV if this croaks. */
9561 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9562 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9563 sv_setsv_flags(sv, old, SV_NOSTEAL);
9568 =for apidoc sv_reset
9570 Underlying implementation for the C<reset> Perl function.
9571 Note that the perl-level function is vaguely deprecated.
9577 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9579 PERL_ARGS_ASSERT_SV_RESET;
9581 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9585 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9587 char todo[PERL_UCHAR_MAX+1];
9590 if (!stash || SvTYPE(stash) != SVt_PVHV)
9593 if (!s) { /* reset ?? searches */
9594 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9596 const U32 count = mg->mg_len / sizeof(PMOP**);
9597 PMOP **pmp = (PMOP**) mg->mg_ptr;
9598 PMOP *const *const end = pmp + count;
9602 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9604 (*pmp)->op_pmflags &= ~PMf_USED;
9612 /* reset variables */
9614 if (!HvARRAY(stash))
9617 Zero(todo, 256, char);
9621 I32 i = (unsigned char)*s;
9625 max = (unsigned char)*s++;
9626 for ( ; i <= max; i++) {
9629 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9631 for (entry = HvARRAY(stash)[i];
9633 entry = HeNEXT(entry))
9638 if (!todo[(U8)*HeKEY(entry)])
9640 gv = MUTABLE_GV(HeVAL(entry));
9642 if (sv && !SvREADONLY(sv)) {
9643 SV_CHECK_THINKFIRST_COW_DROP(sv);
9644 if (!isGV(sv)) SvOK_off(sv);
9649 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9660 Using various gambits, try to get an IO from an SV: the IO slot if its a
9661 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9662 named after the PV if we're a string.
9664 'Get' magic is ignored on the sv passed in, but will be called on
9665 C<SvRV(sv)> if sv is an RV.
9671 Perl_sv_2io(pTHX_ SV *const sv)
9676 PERL_ARGS_ASSERT_SV_2IO;
9678 switch (SvTYPE(sv)) {
9680 io = MUTABLE_IO(sv);
9684 if (isGV_with_GP(sv)) {
9685 gv = MUTABLE_GV(sv);
9688 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9689 HEKfARG(GvNAME_HEK(gv)));
9695 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9697 SvGETMAGIC(SvRV(sv));
9698 return sv_2io(SvRV(sv));
9700 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9707 if (SvGMAGICAL(sv)) {
9708 newsv = sv_newmortal();
9709 sv_setsv_nomg(newsv, sv);
9711 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9721 Using various gambits, try to get a CV from an SV; in addition, try if
9722 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9723 The flags in C<lref> are passed to gv_fetchsv.
9729 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9734 PERL_ARGS_ASSERT_SV_2CV;
9741 switch (SvTYPE(sv)) {
9745 return MUTABLE_CV(sv);
9755 sv = amagic_deref_call(sv, to_cv_amg);
9758 if (SvTYPE(sv) == SVt_PVCV) {
9759 cv = MUTABLE_CV(sv);
9764 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9765 gv = MUTABLE_GV(sv);
9767 Perl_croak(aTHX_ "Not a subroutine reference");
9769 else if (isGV_with_GP(sv)) {
9770 gv = MUTABLE_GV(sv);
9773 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9780 /* Some flags to gv_fetchsv mean don't really create the GV */
9781 if (!isGV_with_GP(gv)) {
9786 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9787 /* XXX this is probably not what they think they're getting.
9788 * It has the same effect as "sub name;", i.e. just a forward
9799 Returns true if the SV has a true value by Perl's rules.
9800 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9801 instead use an in-line version.
9807 Perl_sv_true(pTHX_ SV *const sv)
9812 const XPV* const tXpv = (XPV*)SvANY(sv);
9814 (tXpv->xpv_cur > 1 ||
9815 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9822 return SvIVX(sv) != 0;
9825 return SvNVX(sv) != 0.0;
9827 return sv_2bool(sv);
9833 =for apidoc sv_pvn_force
9835 Get a sensible string out of the SV somehow.
9836 A private implementation of the C<SvPV_force> macro for compilers which
9837 can't cope with complex macro expressions. Always use the macro instead.
9839 =for apidoc sv_pvn_force_flags
9841 Get a sensible string out of the SV somehow.
9842 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9843 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9844 implemented in terms of this function.
9845 You normally want to use the various wrapper macros instead: see
9846 C<SvPV_force> and C<SvPV_force_nomg>
9852 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9854 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9856 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9857 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9858 sv_force_normal_flags(sv, 0);
9868 if (SvTYPE(sv) > SVt_PVLV
9869 || isGV_with_GP(sv))
9870 /* diag_listed_as: Can't coerce %s to %s in %s */
9871 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9873 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9880 if (SvTYPE(sv) < SVt_PV ||
9881 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9884 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9885 SvGROW(sv, len + 1);
9886 Move(s,SvPVX(sv),len,char);
9888 SvPVX(sv)[len] = '\0';
9891 SvPOK_on(sv); /* validate pointer */
9893 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9894 PTR2UV(sv),SvPVX_const(sv)));
9897 (void)SvPOK_only_UTF8(sv);
9898 return SvPVX_mutable(sv);
9902 =for apidoc sv_pvbyten_force
9904 The backend for the C<SvPVbytex_force> macro. Always use the macro
9911 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9913 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9915 sv_pvn_force(sv,lp);
9916 sv_utf8_downgrade(sv,0);
9922 =for apidoc sv_pvutf8n_force
9924 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9931 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9933 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9936 sv_utf8_upgrade_nomg(sv);
9942 =for apidoc sv_reftype
9944 Returns a string describing what the SV is a reference to.
9950 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9952 PERL_ARGS_ASSERT_SV_REFTYPE;
9953 if (ob && SvOBJECT(sv)) {
9954 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9957 /* WARNING - There is code, for instance in mg.c, that assumes that
9958 * the only reason that sv_reftype(sv,0) would return a string starting
9959 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9960 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9961 * this routine inside other subs, and it saves time.
9962 * Do not change this assumption without searching for "dodgy type check" in
9965 switch (SvTYPE(sv)) {
9980 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9981 /* tied lvalues should appear to be
9982 * scalars for backwards compatibility */
9983 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9984 ? "SCALAR" : "LVALUE");
9985 case SVt_PVAV: return "ARRAY";
9986 case SVt_PVHV: return "HASH";
9987 case SVt_PVCV: return "CODE";
9988 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9989 ? "GLOB" : "SCALAR");
9990 case SVt_PVFM: return "FORMAT";
9991 case SVt_PVIO: return "IO";
9992 case SVt_INVLIST: return "INVLIST";
9993 case SVt_REGEXP: return "REGEXP";
9994 default: return "UNKNOWN";
10002 Returns a SV describing what the SV passed in is a reference to.
10008 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10010 PERL_ARGS_ASSERT_SV_REF;
10013 dst = sv_newmortal();
10015 if (ob && SvOBJECT(sv)) {
10016 HvNAME_get(SvSTASH(sv))
10017 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10018 : sv_setpvn(dst, "__ANON__", 8);
10021 const char * reftype = sv_reftype(sv, 0);
10022 sv_setpv(dst, reftype);
10028 =for apidoc sv_isobject
10030 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10031 object. If the SV is not an RV, or if the object is not blessed, then this
10038 Perl_sv_isobject(pTHX_ SV *sv)
10054 Returns a boolean indicating whether the SV is blessed into the specified
10055 class. This does not check for subtypes; use C<sv_derived_from> to verify
10056 an inheritance relationship.
10062 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10064 const char *hvname;
10066 PERL_ARGS_ASSERT_SV_ISA;
10076 hvname = HvNAME_get(SvSTASH(sv));
10080 return strEQ(hvname, name);
10084 =for apidoc newSVrv
10086 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10087 RV then it will be upgraded to one. If C<classname> is non-null then the new
10088 SV will be blessed in the specified package. The new SV is returned and its
10089 reference count is 1. The reference count 1 is owned by C<rv>.
10095 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10099 PERL_ARGS_ASSERT_NEWSVRV;
10103 SV_CHECK_THINKFIRST_COW_DROP(rv);
10105 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10106 const U32 refcnt = SvREFCNT(rv);
10110 SvREFCNT(rv) = refcnt;
10112 sv_upgrade(rv, SVt_IV);
10113 } else if (SvROK(rv)) {
10114 SvREFCNT_dec(SvRV(rv));
10116 prepare_SV_for_RV(rv);
10124 HV* const stash = gv_stashpv(classname, GV_ADD);
10125 (void)sv_bless(rv, stash);
10131 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10133 SV * const lv = newSV_type(SVt_PVLV);
10134 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10136 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10137 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10138 LvSTARGOFF(lv) = ix;
10139 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10144 =for apidoc sv_setref_pv
10146 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10147 argument will be upgraded to an RV. That RV will be modified to point to
10148 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10149 into the SV. The C<classname> argument indicates the package for the
10150 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10151 will have a reference count of 1, and the RV will be returned.
10153 Do not use with other Perl types such as HV, AV, SV, CV, because those
10154 objects will become corrupted by the pointer copy process.
10156 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10162 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10164 PERL_ARGS_ASSERT_SV_SETREF_PV;
10167 sv_setsv(rv, &PL_sv_undef);
10171 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10176 =for apidoc sv_setref_iv
10178 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10179 argument will be upgraded to an RV. That RV will be modified to point to
10180 the new SV. The C<classname> argument indicates the package for the
10181 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10182 will have a reference count of 1, and the RV will be returned.
10188 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10190 PERL_ARGS_ASSERT_SV_SETREF_IV;
10192 sv_setiv(newSVrv(rv,classname), iv);
10197 =for apidoc sv_setref_uv
10199 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10200 argument will be upgraded to an RV. That RV will be modified to point to
10201 the new SV. The C<classname> argument indicates the package for the
10202 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10203 will have a reference count of 1, and the RV will be returned.
10209 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10211 PERL_ARGS_ASSERT_SV_SETREF_UV;
10213 sv_setuv(newSVrv(rv,classname), uv);
10218 =for apidoc sv_setref_nv
10220 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10221 argument will be upgraded to an RV. That RV will be modified to point to
10222 the new SV. The C<classname> argument indicates the package for the
10223 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10224 will have a reference count of 1, and the RV will be returned.
10230 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10232 PERL_ARGS_ASSERT_SV_SETREF_NV;
10234 sv_setnv(newSVrv(rv,classname), nv);
10239 =for apidoc sv_setref_pvn
10241 Copies a string into a new SV, optionally blessing the SV. The length of the
10242 string must be specified with C<n>. The C<rv> argument will be upgraded to
10243 an RV. That RV will be modified to point to the new SV. The C<classname>
10244 argument indicates the package for the blessing. Set C<classname> to
10245 C<NULL> to avoid the blessing. The new SV will have a reference count
10246 of 1, and the RV will be returned.
10248 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10254 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10255 const char *const pv, const STRLEN n)
10257 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10259 sv_setpvn(newSVrv(rv,classname), pv, n);
10264 =for apidoc sv_bless
10266 Blesses an SV into a specified package. The SV must be an RV. The package
10267 must be designated by its stash (see C<gv_stashpv()>). The reference count
10268 of the SV is unaffected.
10274 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10277 HV *oldstash = NULL;
10279 PERL_ARGS_ASSERT_SV_BLESS;
10283 Perl_croak(aTHX_ "Can't bless non-reference value");
10285 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10286 if (SvREADONLY(tmpRef))
10287 Perl_croak_no_modify();
10288 if (SvOBJECT(tmpRef)) {
10289 oldstash = SvSTASH(tmpRef);
10292 SvOBJECT_on(tmpRef);
10293 SvUPGRADE(tmpRef, SVt_PVMG);
10294 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10295 SvREFCNT_dec(oldstash);
10297 if(SvSMAGICAL(tmpRef))
10298 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10306 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10307 * as it is after unglobbing it.
10310 PERL_STATIC_INLINE void
10311 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10315 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10317 PERL_ARGS_ASSERT_SV_UNGLOB;
10319 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10321 if (!(flags & SV_COW_DROP_PV))
10322 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10324 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10326 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10327 && HvNAME_get(stash))
10328 mro_method_changed_in(stash);
10329 gp_free(MUTABLE_GV(sv));
10332 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10333 GvSTASH(sv) = NULL;
10336 if (GvNAME_HEK(sv)) {
10337 unshare_hek(GvNAME_HEK(sv));
10339 isGV_with_GP_off(sv);
10341 if(SvTYPE(sv) == SVt_PVGV) {
10342 /* need to keep SvANY(sv) in the right arena */
10343 xpvmg = new_XPVMG();
10344 StructCopy(SvANY(sv), xpvmg, XPVMG);
10345 del_XPVGV(SvANY(sv));
10348 SvFLAGS(sv) &= ~SVTYPEMASK;
10349 SvFLAGS(sv) |= SVt_PVMG;
10352 /* Intentionally not calling any local SET magic, as this isn't so much a
10353 set operation as merely an internal storage change. */
10354 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10355 else sv_setsv_flags(sv, temp, 0);
10357 if ((const GV *)sv == PL_last_in_gv)
10358 PL_last_in_gv = NULL;
10359 else if ((const GV *)sv == PL_statgv)
10364 =for apidoc sv_unref_flags
10366 Unsets the RV status of the SV, and decrements the reference count of
10367 whatever was being referenced by the RV. This can almost be thought of
10368 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10369 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10370 (otherwise the decrementing is conditional on the reference count being
10371 different from one or the reference being a readonly SV).
10378 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10380 SV* const target = SvRV(ref);
10382 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10384 if (SvWEAKREF(ref)) {
10385 sv_del_backref(target, ref);
10386 SvWEAKREF_off(ref);
10387 SvRV_set(ref, NULL);
10390 SvRV_set(ref, NULL);
10392 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10393 assigned to as BEGIN {$a = \"Foo"} will fail. */
10394 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10395 SvREFCNT_dec_NN(target);
10396 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10397 sv_2mortal(target); /* Schedule for freeing later */
10401 =for apidoc sv_untaint
10403 Untaint an SV. Use C<SvTAINTED_off> instead.
10409 Perl_sv_untaint(pTHX_ SV *const sv)
10411 PERL_ARGS_ASSERT_SV_UNTAINT;
10412 PERL_UNUSED_CONTEXT;
10414 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10415 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10422 =for apidoc sv_tainted
10424 Test an SV for taintedness. Use C<SvTAINTED> instead.
10430 Perl_sv_tainted(pTHX_ SV *const sv)
10432 PERL_ARGS_ASSERT_SV_TAINTED;
10433 PERL_UNUSED_CONTEXT;
10435 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10436 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10437 if (mg && (mg->mg_len & 1) )
10444 =for apidoc sv_setpviv
10446 Copies an integer into the given SV, also updating its string value.
10447 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10453 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10455 char buf[TYPE_CHARS(UV)];
10457 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10459 PERL_ARGS_ASSERT_SV_SETPVIV;
10461 sv_setpvn(sv, ptr, ebuf - ptr);
10465 =for apidoc sv_setpviv_mg
10467 Like C<sv_setpviv>, but also handles 'set' magic.
10473 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10475 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10477 sv_setpviv(sv, iv);
10481 #if defined(PERL_IMPLICIT_CONTEXT)
10483 /* pTHX_ magic can't cope with varargs, so this is a no-context
10484 * version of the main function, (which may itself be aliased to us).
10485 * Don't access this version directly.
10489 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10494 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10496 va_start(args, pat);
10497 sv_vsetpvf(sv, pat, &args);
10501 /* pTHX_ magic can't cope with varargs, so this is a no-context
10502 * version of the main function, (which may itself be aliased to us).
10503 * Don't access this version directly.
10507 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10512 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10514 va_start(args, pat);
10515 sv_vsetpvf_mg(sv, pat, &args);
10521 =for apidoc sv_setpvf
10523 Works like C<sv_catpvf> but copies the text into the SV instead of
10524 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10530 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10534 PERL_ARGS_ASSERT_SV_SETPVF;
10536 va_start(args, pat);
10537 sv_vsetpvf(sv, pat, &args);
10542 =for apidoc sv_vsetpvf
10544 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10545 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10547 Usually used via its frontend C<sv_setpvf>.
10553 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10555 PERL_ARGS_ASSERT_SV_VSETPVF;
10557 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10561 =for apidoc sv_setpvf_mg
10563 Like C<sv_setpvf>, but also handles 'set' magic.
10569 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10573 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10575 va_start(args, pat);
10576 sv_vsetpvf_mg(sv, pat, &args);
10581 =for apidoc sv_vsetpvf_mg
10583 Like C<sv_vsetpvf>, but also handles 'set' magic.
10585 Usually used via its frontend C<sv_setpvf_mg>.
10591 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10593 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10595 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10599 #if defined(PERL_IMPLICIT_CONTEXT)
10601 /* pTHX_ magic can't cope with varargs, so this is a no-context
10602 * version of the main function, (which may itself be aliased to us).
10603 * Don't access this version directly.
10607 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10612 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10614 va_start(args, pat);
10615 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10619 /* pTHX_ magic can't cope with varargs, so this is a no-context
10620 * version of the main function, (which may itself be aliased to us).
10621 * Don't access this version directly.
10625 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10630 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10632 va_start(args, pat);
10633 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10640 =for apidoc sv_catpvf
10642 Processes its arguments like C<sprintf> and appends the formatted
10643 output to an SV. If the appended data contains "wide" characters
10644 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10645 and characters >255 formatted with %c), the original SV might get
10646 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10647 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10648 valid UTF-8; if the original SV was bytes, the pattern should be too.
10653 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10657 PERL_ARGS_ASSERT_SV_CATPVF;
10659 va_start(args, pat);
10660 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10665 =for apidoc sv_vcatpvf
10667 Processes its arguments like C<vsprintf> and appends the formatted output
10668 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10670 Usually used via its frontend C<sv_catpvf>.
10676 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10678 PERL_ARGS_ASSERT_SV_VCATPVF;
10680 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10684 =for apidoc sv_catpvf_mg
10686 Like C<sv_catpvf>, but also handles 'set' magic.
10692 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10696 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10698 va_start(args, pat);
10699 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10705 =for apidoc sv_vcatpvf_mg
10707 Like C<sv_vcatpvf>, but also handles 'set' magic.
10709 Usually used via its frontend C<sv_catpvf_mg>.
10715 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10717 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10719 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10724 =for apidoc sv_vsetpvfn
10726 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10729 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10735 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10736 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10738 PERL_ARGS_ASSERT_SV_VSETPVFN;
10741 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10746 * Warn of missing argument to sprintf, and then return a defined value
10747 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10750 S_vcatpvfn_missing_argument(pTHX) {
10751 if (ckWARN(WARN_MISSING)) {
10752 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10753 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10760 S_expect_number(pTHX_ char **const pattern)
10764 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10766 switch (**pattern) {
10767 case '1': case '2': case '3':
10768 case '4': case '5': case '6':
10769 case '7': case '8': case '9':
10770 var = *(*pattern)++ - '0';
10771 while (isDIGIT(**pattern)) {
10772 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10774 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10782 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10784 const int neg = nv < 0;
10787 PERL_ARGS_ASSERT_F0CONVERT;
10789 if (UNLIKELY(Perl_isinfnan(nv))) {
10790 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10800 if (uv & 1 && uv == nv)
10801 uv--; /* Round to even */
10803 const unsigned dig = uv % 10;
10805 } while (uv /= 10);
10816 =for apidoc sv_vcatpvfn
10818 =for apidoc sv_vcatpvfn_flags
10820 Processes its arguments like C<vsprintf> and appends the formatted output
10821 to an SV. Uses an array of SVs if the C style variable argument list is
10822 missing (NULL). When running with taint checks enabled, indicates via
10823 C<maybe_tainted> if results are untrustworthy (often due to the use of
10826 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10828 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10833 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10834 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10835 vec_utf8 = DO_UTF8(vecsv);
10837 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10840 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10841 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10843 PERL_ARGS_ASSERT_SV_VCATPVFN;
10845 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10848 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10849 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10850 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10851 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10852 * after the first 1023 zero bits.
10854 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10855 * of dynamically growing buffer might be better, start at just 16 bytes
10856 * (for example) and grow only when necessary. Or maybe just by looking
10857 * at the exponents of the two doubles? */
10858 # define DOUBLEDOUBLE_MAXBITS 2098
10861 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10862 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10863 * per xdigit. For the double-double case, this can be rather many.
10864 * The non-double-double-long-double overshoots since all bits of NV
10865 * are not mantissa bits, there are also exponent bits. */
10866 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10867 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10869 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10872 /* If we do not have a known long double format, (including not using
10873 * long doubles, or long doubles being equal to doubles) then we will
10874 * fall back to the ldexp/frexp route, with which we can retrieve at
10875 * most as many bits as our widest unsigned integer type is. We try
10876 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10878 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10879 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10881 #if defined(HAS_QUAD) && defined(Uquad_t)
10882 # define MANTISSATYPE Uquad_t
10883 # define MANTISSASIZE 8
10885 # define MANTISSATYPE UV
10886 # define MANTISSASIZE UVSIZE
10889 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10890 # define HEXTRACT_LITTLE_ENDIAN
10891 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10892 # define HEXTRACT_BIG_ENDIAN
10894 # define HEXTRACT_MIX_ENDIAN
10897 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10898 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10899 * are being extracted from (either directly from the long double in-memory
10900 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10901 * is used to update the exponent. vhex is the pointer to the beginning
10902 * of the output buffer (of VHEX_SIZE).
10904 * The tricky part is that S_hextract() needs to be called twice:
10905 * the first time with vend as NULL, and the second time with vend as
10906 * the pointer returned by the first call. What happens is that on
10907 * the first round the output size is computed, and the intended
10908 * extraction sanity checked. On the second round the actual output
10909 * (the extraction of the hexadecimal values) takes place.
10910 * Sanity failures cause fatal failures during both rounds. */
10912 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10916 int ixmin = 0, ixmax = 0;
10918 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10919 * and elsewhere. */
10921 /* These macros are just to reduce typos, they have multiple
10922 * repetitions below, but usually only one (or sometimes two)
10923 * of them is really being used. */
10924 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10925 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10926 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10927 #define HEXTRACT_OUTPUT(ix) \
10929 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10931 #define HEXTRACT_COUNT(ix, c) \
10933 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10935 #define HEXTRACT_BYTE(ix) \
10937 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10939 #define HEXTRACT_LO_NYBBLE(ix) \
10941 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10943 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10944 * to make it look less odd when the top bits of a NV
10945 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10946 * order bits can be in the "low nybble" of a byte. */
10947 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10948 #define HEXTRACT_BYTES_LE(a, b) \
10949 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10950 #define HEXTRACT_BYTES_BE(a, b) \
10951 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10952 #define HEXTRACT_IMPLICIT_BIT(nv) \
10954 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10957 /* Most formats do. Those which don't should undef this. */
10958 #define HEXTRACT_HAS_IMPLICIT_BIT
10959 /* Many formats do. Those which don't should undef this. */
10960 #define HEXTRACT_HAS_TOP_NYBBLE
10962 /* HEXTRACTSIZE is the maximum number of xdigits. */
10963 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10964 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/4)
10966 # define HEXTRACTSIZE 2 * NVSIZE
10969 const U8* vmaxend = vhex + HEXTRACTSIZE;
10970 PERL_UNUSED_VAR(ix); /* might happen */
10971 (void)Perl_frexp(PERL_ABS(nv), exponent);
10972 if (vend && (vend <= vhex || vend > vmaxend))
10973 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10975 /* First check if using long doubles. */
10976 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10977 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10978 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10979 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10980 /* The bytes 13..0 are the mantissa/fraction,
10981 * the 15,14 are the sign+exponent. */
10982 const U8* nvp = (const U8*)(&nv);
10983 HEXTRACT_IMPLICIT_BIT(nv);
10984 # undef HEXTRACT_HAS_TOP_NYBBLE
10985 HEXTRACT_BYTES_LE(13, 0);
10986 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10987 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10988 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10989 /* The bytes 2..15 are the mantissa/fraction,
10990 * the 0,1 are the sign+exponent. */
10991 const U8* nvp = (const U8*)(&nv);
10992 HEXTRACT_IMPLICIT_BIT(nv);
10993 # undef HEXTRACT_HAS_TOP_NYBBLE
10994 HEXTRACT_BYTES_BE(2, 15);
10995 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10996 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10997 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10998 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10999 * meaning that 2 or 6 bytes are empty padding. */
11000 /* The bytes 7..0 are the mantissa/fraction */
11001 const U8* nvp = (const U8*)(&nv);
11002 # undef HEXTRACT_HAS_IMPLICIT_BIT
11003 # undef HEXTRACT_HAS_TOP_NYBBLE
11004 HEXTRACT_BYTES_LE(7, 0);
11005 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11006 /* Does this format ever happen? (Wikipedia says the Motorola
11007 * 6888x math coprocessors used format _like_ this but padded
11008 * to 96 bits with 16 unused bits between the exponent and the
11010 const U8* nvp = (const U8*)(&nv);
11011 # undef HEXTRACT_HAS_IMPLICIT_BIT
11012 # undef HEXTRACT_HAS_TOP_NYBBLE
11013 HEXTRACT_BYTES_BE(0, 7);
11015 # define HEXTRACT_FALLBACK
11016 /* Double-double format: two doubles next to each other.
11017 * The first double is the high-order one, exactly like
11018 * it would be for a "lone" double. The second double
11019 * is shifted down using the exponent so that that there
11020 * are no common bits. The tricky part is that the value
11021 * of the double-double is the SUM of the two doubles and
11022 * the second one can be also NEGATIVE.
11024 * Because of this tricky construction the bytewise extraction we
11025 * use for the other long double formats doesn't work, we must
11026 * extract the values bit by bit.
11028 * The little-endian double-double is used .. somewhere?
11030 * The big endian double-double is used in e.g. PPC/Power (AIX)
11033 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11034 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11035 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11038 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11039 /* Using normal doubles, not long doubles.
11041 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11042 * bytes, since we might need to handle printf precision, and
11043 * also need to insert the radix. */
11045 # ifdef HEXTRACT_LITTLE_ENDIAN
11046 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11047 const U8* nvp = (const U8*)(&nv);
11048 HEXTRACT_IMPLICIT_BIT(nv);
11049 HEXTRACT_TOP_NYBBLE(6);
11050 HEXTRACT_BYTES_LE(5, 0);
11051 # elif defined(HEXTRACT_BIG_ENDIAN)
11052 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11053 const U8* nvp = (const U8*)(&nv);
11054 HEXTRACT_IMPLICIT_BIT(nv);
11055 HEXTRACT_TOP_NYBBLE(1);
11056 HEXTRACT_BYTES_BE(2, 7);
11057 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11058 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11059 const U8* nvp = (const U8*)(&nv);
11060 HEXTRACT_IMPLICIT_BIT(nv);
11061 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11062 HEXTRACT_BYTE(1); /* 5 */
11063 HEXTRACT_BYTE(0); /* 4 */
11064 HEXTRACT_BYTE(7); /* 3 */
11065 HEXTRACT_BYTE(6); /* 2 */
11066 HEXTRACT_BYTE(5); /* 1 */
11067 HEXTRACT_BYTE(4); /* 0 */
11068 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11069 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11070 const U8* nvp = (const U8*)(&nv);
11071 HEXTRACT_IMPLICIT_BIT(nv);
11072 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11073 HEXTRACT_BYTE(6); /* 5 */
11074 HEXTRACT_BYTE(7); /* 4 */
11075 HEXTRACT_BYTE(0); /* 3 */
11076 HEXTRACT_BYTE(1); /* 2 */
11077 HEXTRACT_BYTE(2); /* 1 */
11078 HEXTRACT_BYTE(3); /* 0 */
11080 # define HEXTRACT_FALLBACK
11083 # define HEXTRACT_FALLBACK
11085 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11086 # ifdef HEXTRACT_FALLBACK
11087 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11088 /* The fallback is used for the double-double format, and
11089 * for unknown long double formats, and for unknown double
11090 * formats, or in general unknown NV formats. */
11091 if (nv == (NV)0.0) {
11099 NV d = nv < 0 ? -nv : nv;
11101 U8 ha = 0x0; /* hexvalue accumulator */
11102 U8 hd = 0x8; /* hexvalue digit */
11104 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11105 * this is essentially manual frexp(). Multiplying by 0.5 and
11106 * doubling should be lossless in binary floating point. */
11116 while (d >= e + e) {
11120 /* Now e <= d < 2*e */
11122 /* First extract the leading hexdigit (the implicit bit). */
11138 /* Then extract the remaining hexdigits. */
11139 while (d > (NV)0.0) {
11145 /* Output or count in groups of four bits,
11146 * that is, when the hexdigit is down to one. */
11151 /* Reset the hexvalue. */
11160 /* Flush possible pending hexvalue. */
11170 /* Croak for various reasons: if the output pointer escaped the
11171 * output buffer, if the extraction index escaped the extraction
11172 * buffer, or if the ending output pointer didn't match the
11173 * previously computed value. */
11174 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11175 /* For double-double the ixmin and ixmax stay at zero,
11176 * which is convenient since the HEXTRACTSIZE is tricky
11177 * for double-double. */
11178 ixmin < 0 || ixmax >= NVSIZE ||
11179 (vend && v != vend))
11180 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11185 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11186 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11191 const char *patend;
11194 static const char nullstr[] = "(null)";
11196 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11197 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11199 /* Times 4: a decimal digit takes more than 3 binary digits.
11200 * NV_DIG: mantissa takes than many decimal digits.
11201 * Plus 32: Playing safe. */
11202 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11203 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11204 bool hexfp = FALSE; /* hexadecimal floating point? */
11206 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11208 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11209 PERL_UNUSED_ARG(maybe_tainted);
11211 if (flags & SV_GMAGIC)
11214 /* no matter what, this is a string now */
11215 (void)SvPV_force_nomg(sv, origlen);
11217 /* special-case "", "%s", and "%-p" (SVf - see below) */
11219 if (svmax && ckWARN(WARN_REDUNDANT))
11220 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11221 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11224 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11225 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11226 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11227 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11230 const char * const s = va_arg(*args, char*);
11231 sv_catpv_nomg(sv, s ? s : nullstr);
11233 else if (svix < svmax) {
11234 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11235 SvGETMAGIC(*svargs);
11236 sv_catsv_nomg(sv, *svargs);
11239 S_vcatpvfn_missing_argument(aTHX);
11242 if (args && patlen == 3 && pat[0] == '%' &&
11243 pat[1] == '-' && pat[2] == 'p') {
11244 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11245 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11246 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11247 argsv = MUTABLE_SV(va_arg(*args, void*));
11248 sv_catsv_nomg(sv, argsv);
11252 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11253 /* special-case "%.<number>[gf]" */
11254 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11255 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11256 unsigned digits = 0;
11260 while (*pp >= '0' && *pp <= '9')
11261 digits = 10 * digits + (*pp++ - '0');
11263 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11264 format the first argument and WARN_REDUNDANT if svmax > 1?
11265 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11266 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11267 const NV nv = SvNV(*svargs);
11268 if (LIKELY(!Perl_isinfnan(nv))) {
11270 /* Add check for digits != 0 because it seems that some
11271 gconverts are buggy in this case, and we don't yet have
11272 a Configure test for this. */
11273 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11274 /* 0, point, slack */
11275 STORE_LC_NUMERIC_SET_TO_NEEDED();
11276 SNPRINTF_G(nv, ebuf, size, digits);
11277 sv_catpv_nomg(sv, ebuf);
11278 if (*ebuf) /* May return an empty string for digits==0 */
11281 } else if (!digits) {
11284 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11285 sv_catpvn_nomg(sv, p, l);
11292 #endif /* !USE_LONG_DOUBLE */
11294 if (!args && svix < svmax && DO_UTF8(*svargs))
11297 patend = (char*)pat + patlen;
11298 for (p = (char*)pat; p < patend; p = q) {
11301 bool vectorize = FALSE;
11302 bool vectorarg = FALSE;
11303 bool vec_utf8 = FALSE;
11309 bool has_precis = FALSE;
11311 const I32 osvix = svix;
11312 bool is_utf8 = FALSE; /* is this item utf8? */
11313 #ifdef HAS_LDBL_SPRINTF_BUG
11314 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11315 with sfio - Allen <allens@cpan.org> */
11316 bool fix_ldbl_sprintf_bug = FALSE;
11320 U8 utf8buf[UTF8_MAXBYTES+1];
11321 STRLEN esignlen = 0;
11323 const char *eptr = NULL;
11324 const char *fmtstart;
11327 const U8 *vecstr = NULL;
11334 /* We need a long double target in case HAS_LONG_DOUBLE,
11335 * even without USE_LONG_DOUBLE, so that we can printf with
11336 * long double formats, even without NV being long double.
11337 * But we call the target 'fv' instead of 'nv', since most of
11338 * the time it is not (most compilers these days recognize
11339 * "long double", even if only as a synonym for "double").
11341 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11342 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11344 # ifdef Perl_isfinitel
11345 # define FV_ISFINITE(x) Perl_isfinitel(x)
11347 # define FV_GF PERL_PRIgldbl
11348 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11349 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11350 # define NV_TO_FV(nv,fv) STMT_START { \
11352 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11355 # define NV_TO_FV(nv,fv) (fv)=(nv)
11359 # define FV_GF NVgf
11360 # define NV_TO_FV(nv,fv) (fv)=(nv)
11362 #ifndef FV_ISFINITE
11363 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11369 const char *dotstr = ".";
11370 STRLEN dotstrlen = 1;
11371 I32 efix = 0; /* explicit format parameter index */
11372 I32 ewix = 0; /* explicit width index */
11373 I32 epix = 0; /* explicit precision index */
11374 I32 evix = 0; /* explicit vector index */
11375 bool asterisk = FALSE;
11376 bool infnan = FALSE;
11378 /* echo everything up to the next format specification */
11379 for (q = p; q < patend && *q != '%'; ++q) ;
11381 if (has_utf8 && !pat_utf8)
11382 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11384 sv_catpvn_nomg(sv, p, q - p);
11393 We allow format specification elements in this order:
11394 \d+\$ explicit format parameter index
11396 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11397 0 flag (as above): repeated to allow "v02"
11398 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11399 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11401 [%bcdefginopsuxDFOUX] format (mandatory)
11406 As of perl5.9.3, printf format checking is on by default.
11407 Internally, perl uses %p formats to provide an escape to
11408 some extended formatting. This block deals with those
11409 extensions: if it does not match, (char*)q is reset and
11410 the normal format processing code is used.
11412 Currently defined extensions are:
11413 %p include pointer address (standard)
11414 %-p (SVf) include an SV (previously %_)
11415 %-<num>p include an SV with precision <num>
11417 %3p include a HEK with precision of 256
11418 %4p char* preceded by utf8 flag and length
11419 %<num>p (where num is 1 or > 4) reserved for future
11422 Robin Barker 2005-07-14 (but modified since)
11424 %1p (VDf) removed. RMB 2007-10-19
11431 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11432 /* The argument has already gone through cBOOL, so the cast
11434 is_utf8 = (bool)va_arg(*args, int);
11435 elen = va_arg(*args, UV);
11436 if ((IV)elen < 0) {
11437 /* check if utf8 length is larger than 0 when cast to IV */
11438 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11439 elen= 0; /* otherwise we want to treat this as an empty string */
11441 eptr = va_arg(*args, char *);
11442 q += sizeof(UTF8f)-1;
11445 n = expect_number(&q);
11447 if (sv) { /* SVf */
11452 argsv = MUTABLE_SV(va_arg(*args, void*));
11453 eptr = SvPV_const(argsv, elen);
11454 if (DO_UTF8(argsv))
11458 else if (n==2 || n==3) { /* HEKf */
11459 HEK * const hek = va_arg(*args, HEK *);
11460 eptr = HEK_KEY(hek);
11461 elen = HEK_LEN(hek);
11462 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11463 if (n==3) precis = 256, has_precis = TRUE;
11467 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11468 "internal %%<num>p might conflict with future printf extensions");
11474 if ( (width = expect_number(&q)) ) {
11478 if (!no_redundant_warning)
11479 /* I've forgotten if it's a better
11480 micro-optimization to always set this or to
11481 only set it if it's unset */
11482 no_redundant_warning = TRUE;
11494 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11523 if ( (ewix = expect_number(&q)) )
11532 if ((vectorarg = asterisk)) {
11545 width = expect_number(&q);
11548 if (vectorize && vectorarg) {
11549 /* vectorizing, but not with the default "." */
11551 vecsv = va_arg(*args, SV*);
11553 vecsv = (evix > 0 && evix <= svmax)
11554 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11556 vecsv = svix < svmax
11557 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11559 dotstr = SvPV_const(vecsv, dotstrlen);
11560 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11561 bad with tied or overloaded values that return UTF8. */
11562 if (DO_UTF8(vecsv))
11564 else if (has_utf8) {
11565 vecsv = sv_mortalcopy(vecsv);
11566 sv_utf8_upgrade(vecsv);
11567 dotstr = SvPV_const(vecsv, dotstrlen);
11574 i = va_arg(*args, int);
11576 i = (ewix ? ewix <= svmax : svix < svmax) ?
11577 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11579 width = (i < 0) ? -i : i;
11589 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11591 /* XXX: todo, support specified precision parameter */
11595 i = va_arg(*args, int);
11597 i = (ewix ? ewix <= svmax : svix < svmax)
11598 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11600 has_precis = !(i < 0);
11604 while (isDIGIT(*q))
11605 precis = precis * 10 + (*q++ - '0');
11614 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11615 vecsv = svargs[efix ? efix-1 : svix++];
11616 vecstr = (U8*)SvPV_const(vecsv,veclen);
11617 vec_utf8 = DO_UTF8(vecsv);
11619 /* if this is a version object, we need to convert
11620 * back into v-string notation and then let the
11621 * vectorize happen normally
11623 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11624 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11625 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11626 "vector argument not supported with alpha versions");
11629 vecsv = sv_newmortal();
11630 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11632 vecstr = (U8*)SvPV_const(vecsv, veclen);
11633 vec_utf8 = DO_UTF8(vecsv);
11647 case 'I': /* Ix, I32x, and I64x */
11648 # ifdef USE_64_BIT_INT
11649 if (q[1] == '6' && q[2] == '4') {
11655 if (q[1] == '3' && q[2] == '2') {
11659 # ifdef USE_64_BIT_INT
11665 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11666 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11669 # ifdef USE_QUADMATH
11682 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11683 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11684 if (*q == 'l') { /* lld, llf */
11693 if (*++q == 'h') { /* hhd, hhu */
11722 if (!vectorize && !args) {
11724 const I32 i = efix-1;
11725 argsv = (i >= 0 && i < svmax)
11726 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11728 argsv = (svix >= 0 && svix < svmax)
11729 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11733 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11734 /* XXX va_arg(*args) case? need peek, use va_copy? */
11736 if (UNLIKELY(SvAMAGIC(argsv)))
11737 argsv = sv_2num(argsv);
11738 infnan = UNLIKELY(isinfnansv(argsv));
11741 switch (c = *q++) {
11749 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11750 /* no va_arg() case */
11751 SvNV_nomg(argsv), (int)c);
11752 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11754 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11756 eptr = (char*)utf8buf;
11757 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11771 eptr = va_arg(*args, char*);
11773 elen = strlen(eptr);
11775 eptr = (char *)nullstr;
11776 elen = sizeof nullstr - 1;
11780 eptr = SvPV_const(argsv, elen);
11781 if (DO_UTF8(argsv)) {
11782 STRLEN old_precis = precis;
11783 if (has_precis && precis < elen) {
11784 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11785 STRLEN p = precis > ulen ? ulen : precis;
11786 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11787 /* sticks at end */
11789 if (width) { /* fudge width (can't fudge elen) */
11790 if (has_precis && precis < elen)
11791 width += precis - old_precis;
11794 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11801 if (has_precis && precis < elen)
11809 goto floating_point;
11811 if (alt || vectorize)
11813 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11827 goto floating_point;
11834 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11843 esignbuf[esignlen++] = plus;
11847 case 'c': iv = (char)va_arg(*args, int); break;
11848 case 'h': iv = (short)va_arg(*args, int); break;
11849 case 'l': iv = va_arg(*args, long); break;
11850 case 'V': iv = va_arg(*args, IV); break;
11851 case 'z': iv = va_arg(*args, SSize_t); break;
11852 #ifdef HAS_PTRDIFF_T
11853 case 't': iv = va_arg(*args, ptrdiff_t); break;
11855 default: iv = va_arg(*args, int); break;
11857 case 'j': iv = va_arg(*args, intmax_t); break;
11861 iv = va_arg(*args, Quad_t); break;
11868 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11870 case 'c': iv = (char)tiv; break;
11871 case 'h': iv = (short)tiv; break;
11872 case 'l': iv = (long)tiv; break;
11874 default: iv = tiv; break;
11877 iv = (Quad_t)tiv; break;
11883 if ( !vectorize ) /* we already set uv above */
11888 esignbuf[esignlen++] = plus;
11891 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
11892 esignbuf[esignlen++] = '-';
11931 goto floating_point;
11939 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11950 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11951 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11952 case 'l': uv = va_arg(*args, unsigned long); break;
11953 case 'V': uv = va_arg(*args, UV); break;
11954 case 'z': uv = va_arg(*args, Size_t); break;
11955 #ifdef HAS_PTRDIFF_T
11956 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11959 case 'j': uv = va_arg(*args, uintmax_t); break;
11961 default: uv = va_arg(*args, unsigned); break;
11964 uv = va_arg(*args, Uquad_t); break;
11971 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11973 case 'c': uv = (unsigned char)tuv; break;
11974 case 'h': uv = (unsigned short)tuv; break;
11975 case 'l': uv = (unsigned long)tuv; break;
11977 default: uv = tuv; break;
11980 uv = (Uquad_t)tuv; break;
11989 char *ptr = ebuf + sizeof ebuf;
11990 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11996 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12000 } while (uv >>= 4);
12002 esignbuf[esignlen++] = '0';
12003 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12009 *--ptr = '0' + dig;
12010 } while (uv >>= 3);
12011 if (alt && *ptr != '0')
12017 *--ptr = '0' + dig;
12018 } while (uv >>= 1);
12020 esignbuf[esignlen++] = '0';
12021 esignbuf[esignlen++] = c;
12024 default: /* it had better be ten or less */
12027 *--ptr = '0' + dig;
12028 } while (uv /= base);
12031 elen = (ebuf + sizeof ebuf) - ptr;
12035 zeros = precis - elen;
12036 else if (precis == 0 && elen == 1 && *eptr == '0'
12037 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12040 /* a precision nullifies the 0 flag. */
12047 /* FLOATING POINT */
12052 c = 'f'; /* maybe %F isn't supported here */
12054 case 'e': case 'E':
12056 case 'g': case 'G':
12057 case 'a': case 'A':
12061 /* This is evil, but floating point is even more evil */
12063 /* for SV-style calling, we can only get NV
12064 for C-style calling, we assume %f is double;
12065 for simplicity we allow any of %Lf, %llf, %qf for long double
12069 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12073 /* [perl #20339] - we should accept and ignore %lf rather than die */
12077 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12078 intsize = args ? 0 : 'q';
12082 #if defined(HAS_LONG_DOUBLE)
12095 /* Now we need (long double) if intsize == 'q', else (double). */
12097 /* Note: do not pull NVs off the va_list with va_arg()
12098 * (pull doubles instead) because if you have a build
12099 * with long doubles, you would always be pulling long
12100 * doubles, which would badly break anyone using only
12101 * doubles (i.e. the majority of builds). In other
12102 * words, you cannot mix doubles and long doubles.
12103 * The only case where you can pull off long doubles
12104 * is when the format specifier explicitly asks so with
12106 #ifdef USE_QUADMATH
12107 fv = intsize == 'q' ?
12108 va_arg(*args, NV) : va_arg(*args, double);
12110 #elif LONG_DOUBLESIZE > DOUBLESIZE
12111 if (intsize == 'q') {
12112 fv = va_arg(*args, long double);
12115 nv = va_arg(*args, double);
12119 nv = va_arg(*args, double);
12125 if (!infnan) SvGETMAGIC(argsv);
12126 nv = SvNV_nomg(argsv);
12131 /* frexp() (or frexpl) has some unspecified behaviour for
12132 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12133 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12135 (void)Perl_frexp((NV)fv, &i);
12136 if (i == PERL_INT_MIN)
12137 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12138 /* Do not set hexfp earlier since we want to printf
12139 * Inf/NaN for Inf/NaN, not their hexfp. */
12140 hexfp = isALPHA_FOLD_EQ(c, 'a');
12141 if (UNLIKELY(hexfp)) {
12142 /* This seriously overshoots in most cases, but
12143 * better the undershooting. Firstly, all bytes
12144 * of the NV are not mantissa, some of them are
12145 * exponent. Secondly, for the reasonably common
12146 * long doubles case, the "80-bit extended", two
12147 * or six bytes of the NV are unused. */
12149 (fv < 0) ? 1 : 0 + /* possible unary minus */
12151 1 + /* the very unlikely carry */
12154 2 * NVSIZE + /* 2 hexdigits for each byte */
12156 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12158 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12159 /* However, for the "double double", we need more.
12160 * Since each double has their own exponent, the
12161 * doubles may float (haha) rather far from each
12162 * other, and the number of required bits is much
12163 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12164 * See the definition of DOUBLEDOUBLE_MAXBITS.
12166 * Need 2 hexdigits for each byte. */
12167 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12168 /* the size for the exponent already added */
12170 #ifdef USE_LOCALE_NUMERIC
12171 STORE_LC_NUMERIC_SET_TO_NEEDED();
12172 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12173 need += SvLEN(PL_numeric_radix_sv);
12174 RESTORE_LC_NUMERIC();
12178 need = BIT_DIGITS(i);
12179 } /* if i < 0, the number of digits is hard to predict. */
12181 need += has_precis ? precis : 6; /* known default */
12186 #ifdef HAS_LDBL_SPRINTF_BUG
12187 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12188 with sfio - Allen <allens@cpan.org> */
12191 # define MY_DBL_MAX DBL_MAX
12192 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12193 # if DOUBLESIZE >= 8
12194 # define MY_DBL_MAX 1.7976931348623157E+308L
12196 # define MY_DBL_MAX 3.40282347E+38L
12200 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12201 # define MY_DBL_MAX_BUG 1L
12203 # define MY_DBL_MAX_BUG MY_DBL_MAX
12207 # define MY_DBL_MIN DBL_MIN
12208 # else /* XXX guessing! -Allen */
12209 # if DOUBLESIZE >= 8
12210 # define MY_DBL_MIN 2.2250738585072014E-308L
12212 # define MY_DBL_MIN 1.17549435E-38L
12216 if ((intsize == 'q') && (c == 'f') &&
12217 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12218 (need < DBL_DIG)) {
12219 /* it's going to be short enough that
12220 * long double precision is not needed */
12222 if ((fv <= 0L) && (fv >= -0L))
12223 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12225 /* would use Perl_fp_class as a double-check but not
12226 * functional on IRIX - see perl.h comments */
12228 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12229 /* It's within the range that a double can represent */
12230 #if defined(DBL_MAX) && !defined(DBL_MIN)
12231 if ((fv >= ((long double)1/DBL_MAX)) ||
12232 (fv <= (-(long double)1/DBL_MAX)))
12234 fix_ldbl_sprintf_bug = TRUE;
12237 if (fix_ldbl_sprintf_bug == TRUE) {
12247 # undef MY_DBL_MAX_BUG
12250 #endif /* HAS_LDBL_SPRINTF_BUG */
12252 need += 20; /* fudge factor */
12253 if (PL_efloatsize < need) {
12254 Safefree(PL_efloatbuf);
12255 PL_efloatsize = need + 20; /* more fudge */
12256 Newx(PL_efloatbuf, PL_efloatsize, char);
12257 PL_efloatbuf[0] = '\0';
12260 if ( !(width || left || plus || alt) && fill != '0'
12261 && has_precis && intsize != 'q' /* Shortcuts */
12262 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12263 /* See earlier comment about buggy Gconvert when digits,
12265 if ( c == 'g' && precis ) {
12266 STORE_LC_NUMERIC_SET_TO_NEEDED();
12267 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12268 /* May return an empty string for digits==0 */
12269 if (*PL_efloatbuf) {
12270 elen = strlen(PL_efloatbuf);
12271 goto float_converted;
12273 } else if ( c == 'f' && !precis ) {
12274 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12279 if (UNLIKELY(hexfp)) {
12280 /* Hexadecimal floating point. */
12281 char* p = PL_efloatbuf;
12282 U8 vhex[VHEX_SIZE];
12283 U8* v = vhex; /* working pointer to vhex */
12284 U8* vend; /* pointer to one beyond last digit of vhex */
12285 U8* vfnz = NULL; /* first non-zero */
12286 const bool lower = (c == 'a');
12287 /* At output the values of vhex (up to vend) will
12288 * be mapped through the xdig to get the actual
12289 * human-readable xdigits. */
12290 const char* xdig = PL_hexdigit;
12291 int zerotail = 0; /* how many extra zeros to append */
12292 int exponent = 0; /* exponent of the floating point input */
12294 /* XXX: denormals, NaN, Inf.
12296 * For example with denormals, (assuming the vanilla
12297 * 64-bit double): the exponent is zero. 1xp-1074 is
12298 * the smallest denormal and the smallest double, it
12299 * should be output as 0x0.0000000000001p-1022 to
12300 * match its internal structure. */
12302 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
12303 S_hextract(aTHX_ nv, &exponent, vhex, vend);
12305 #if NVSIZE > DOUBLESIZE
12306 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12307 /* In this case there is an implicit bit,
12308 * and therefore the exponent is shifted shift by one. */
12311 /* In this case there is no implicit bit,
12312 * and the exponent is shifted by the first xdigit. */
12327 xdig += 16; /* Use uppercase hex. */
12330 /* Find the first non-zero xdigit. */
12331 for (v = vhex; v < vend; v++) {
12339 U8* vlnz = NULL; /* The last non-zero. */
12341 /* Find the last non-zero xdigit. */
12342 for (v = vend - 1; v >= vhex; v--) {
12349 #if NVSIZE == DOUBLESIZE
12355 if ((SSize_t)(precis + 1) < vend - vhex) {
12358 v = vhex + precis + 1;
12359 /* Round away from zero: if the tail
12360 * beyond the precis xdigits is equal to
12361 * or greater than 0x8000... */
12363 if (!round && *v == 0x8) {
12364 for (v++; v < vend; v++) {
12372 for (v = vhex + precis; v >= vhex; v--) {
12379 /* If the carry goes all the way to
12380 * the front, we need to output
12381 * a single '1'. This goes against
12382 * the "xdigit and then radix"
12383 * but since this is "cannot happen"
12384 * category, that is probably good. */
12389 /* The new effective "last non zero". */
12390 vlnz = vhex + precis;
12393 zerotail = precis - (vlnz - vhex);
12400 /* The radix is always output after the first
12401 * non-zero xdigit, or if alt. */
12402 if (vfnz < vlnz || alt) {
12403 #ifndef USE_LOCALE_NUMERIC
12406 STORE_LC_NUMERIC_SET_TO_NEEDED();
12407 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12409 const char* r = SvPV(PL_numeric_radix_sv, n);
12410 Copy(r, p, n, char);
12416 RESTORE_LC_NUMERIC();
12431 elen = p - PL_efloatbuf;
12432 elen += my_snprintf(p, PL_efloatsize - elen,
12433 "%c%+d", lower ? 'p' : 'P',
12436 if (elen < width) {
12438 /* Pad the back with spaces. */
12439 memset(PL_efloatbuf + elen, ' ', width - elen);
12441 else if (fill == '0') {
12442 /* Insert the zeros between the "0x" and
12443 * the digits, otherwise we end up with
12445 STRLEN nzero = width - elen;
12446 char* zerox = PL_efloatbuf + 2;
12447 Move(zerox, zerox + nzero, elen - 2, char);
12448 memset(zerox, fill, nzero);
12451 /* Move it to the right. */
12452 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12454 /* Pad the front with spaces. */
12455 memset(PL_efloatbuf, ' ', width - elen);
12461 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12463 /* Not affecting infnan output: precision, alt, fill. */
12464 if (elen < width) {
12466 /* Pack the back with spaces. */
12467 memset(PL_efloatbuf + elen, ' ', width - elen);
12469 /* Move it to the right. */
12470 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12472 /* Pad the front with spaces. */
12473 memset(PL_efloatbuf, ' ', width - elen);
12481 char *ptr = ebuf + sizeof ebuf;
12484 #if defined(USE_QUADMATH)
12485 if (intsize == 'q') {
12489 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12490 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12491 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12492 * not USE_LONG_DOUBLE and NVff. In other words,
12493 * this needs to work without USE_LONG_DOUBLE. */
12494 if (intsize == 'q') {
12495 /* Copy the one or more characters in a long double
12496 * format before the 'base' ([efgEFG]) character to
12497 * the format string. */
12498 static char const ldblf[] = PERL_PRIfldbl;
12499 char const *p = ldblf + sizeof(ldblf) - 3;
12500 while (p >= ldblf) { *--ptr = *p--; }
12505 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12510 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12522 /* No taint. Otherwise we are in the strange situation
12523 * where printf() taints but print($float) doesn't.
12526 STORE_LC_NUMERIC_SET_TO_NEEDED();
12528 /* hopefully the above makes ptr a very constrained format
12529 * that is safe to use, even though it's not literal */
12530 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12531 #ifdef USE_QUADMATH
12533 const char* qfmt = quadmath_format_single(ptr);
12535 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12536 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12538 if ((IV)elen == -1)
12539 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12543 #elif defined(HAS_LONG_DOUBLE)
12544 elen = ((intsize == 'q')
12545 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12546 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12548 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12554 eptr = PL_efloatbuf;
12555 assert((IV)elen > 0); /* here zero elen is bad */
12557 #ifdef USE_LOCALE_NUMERIC
12558 /* If the decimal point character in the string is UTF-8, make the
12560 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12561 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12574 i = SvCUR(sv) - origlen;
12577 case 'c': *(va_arg(*args, char*)) = i; break;
12578 case 'h': *(va_arg(*args, short*)) = i; break;
12579 default: *(va_arg(*args, int*)) = i; break;
12580 case 'l': *(va_arg(*args, long*)) = i; break;
12581 case 'V': *(va_arg(*args, IV*)) = i; break;
12582 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12583 #ifdef HAS_PTRDIFF_T
12584 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12587 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12591 *(va_arg(*args, Quad_t*)) = i; break;
12598 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12599 continue; /* not "break" */
12606 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12607 && ckWARN(WARN_PRINTF))
12609 SV * const msg = sv_newmortal();
12610 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12611 (PL_op->op_type == OP_PRTF) ? "" : "s");
12612 if (fmtstart < patend) {
12613 const char * const fmtend = q < patend ? q : patend;
12615 sv_catpvs(msg, "\"%");
12616 for (f = fmtstart; f < fmtend; f++) {
12618 sv_catpvn_nomg(msg, f, 1);
12620 Perl_sv_catpvf(aTHX_ msg,
12621 "\\%03"UVof, (UV)*f & 0xFF);
12624 sv_catpvs(msg, "\"");
12626 sv_catpvs(msg, "end of string");
12628 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12631 /* output mangled stuff ... */
12637 /* ... right here, because formatting flags should not apply */
12638 SvGROW(sv, SvCUR(sv) + elen + 1);
12640 Copy(eptr, p, elen, char);
12643 SvCUR_set(sv, p - SvPVX_const(sv));
12645 continue; /* not "break" */
12648 if (is_utf8 != has_utf8) {
12651 sv_utf8_upgrade(sv);
12654 const STRLEN old_elen = elen;
12655 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12656 sv_utf8_upgrade(nsv);
12657 eptr = SvPVX_const(nsv);
12660 if (width) { /* fudge width (can't fudge elen) */
12661 width += elen - old_elen;
12667 assert((IV)elen >= 0); /* here zero elen is fine */
12668 have = esignlen + zeros + elen;
12670 croak_memory_wrap();
12672 need = (have > width ? have : width);
12675 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12676 croak_memory_wrap();
12677 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12679 if (esignlen && fill == '0') {
12681 for (i = 0; i < (int)esignlen; i++)
12682 *p++ = esignbuf[i];
12684 if (gap && !left) {
12685 memset(p, fill, gap);
12688 if (esignlen && fill != '0') {
12690 for (i = 0; i < (int)esignlen; i++)
12691 *p++ = esignbuf[i];
12695 for (i = zeros; i; i--)
12699 Copy(eptr, p, elen, char);
12703 memset(p, ' ', gap);
12708 Copy(dotstr, p, dotstrlen, char);
12712 vectorize = FALSE; /* done iterating over vecstr */
12719 SvCUR_set(sv, p - SvPVX_const(sv));
12726 /* Now that we've consumed all our printf format arguments (svix)
12727 * do we have things left on the stack that we didn't use?
12729 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12730 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12731 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12736 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12740 /* =========================================================================
12742 =head1 Cloning an interpreter
12746 All the macros and functions in this section are for the private use of
12747 the main function, perl_clone().
12749 The foo_dup() functions make an exact copy of an existing foo thingy.
12750 During the course of a cloning, a hash table is used to map old addresses
12751 to new addresses. The table is created and manipulated with the
12752 ptr_table_* functions.
12754 * =========================================================================*/
12757 #if defined(USE_ITHREADS)
12759 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12760 #ifndef GpREFCNT_inc
12761 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12765 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12766 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12767 If this changes, please unmerge ss_dup.
12768 Likewise, sv_dup_inc_multiple() relies on this fact. */
12769 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12770 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12771 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12772 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12773 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12774 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12775 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12776 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12777 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12778 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12779 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12780 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12781 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12783 /* clone a parser */
12786 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12790 PERL_ARGS_ASSERT_PARSER_DUP;
12795 /* look for it in the table first */
12796 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12800 /* create anew and remember what it is */
12801 Newxz(parser, 1, yy_parser);
12802 ptr_table_store(PL_ptr_table, proto, parser);
12804 /* XXX these not yet duped */
12805 parser->old_parser = NULL;
12806 parser->stack = NULL;
12808 parser->stack_size = 0;
12809 /* XXX parser->stack->state = 0; */
12811 /* XXX eventually, just Copy() most of the parser struct ? */
12813 parser->lex_brackets = proto->lex_brackets;
12814 parser->lex_casemods = proto->lex_casemods;
12815 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12816 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12817 parser->lex_casestack = savepvn(proto->lex_casestack,
12818 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12819 parser->lex_defer = proto->lex_defer;
12820 parser->lex_dojoin = proto->lex_dojoin;
12821 parser->lex_formbrack = proto->lex_formbrack;
12822 parser->lex_inpat = proto->lex_inpat;
12823 parser->lex_inwhat = proto->lex_inwhat;
12824 parser->lex_op = proto->lex_op;
12825 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12826 parser->lex_starts = proto->lex_starts;
12827 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12828 parser->multi_close = proto->multi_close;
12829 parser->multi_open = proto->multi_open;
12830 parser->multi_start = proto->multi_start;
12831 parser->multi_end = proto->multi_end;
12832 parser->preambled = proto->preambled;
12833 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12834 parser->linestr = sv_dup_inc(proto->linestr, param);
12835 parser->expect = proto->expect;
12836 parser->copline = proto->copline;
12837 parser->last_lop_op = proto->last_lop_op;
12838 parser->lex_state = proto->lex_state;
12839 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12840 /* rsfp_filters entries have fake IoDIRP() */
12841 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12842 parser->in_my = proto->in_my;
12843 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12844 parser->error_count = proto->error_count;
12847 parser->linestr = sv_dup_inc(proto->linestr, param);
12850 char * const ols = SvPVX(proto->linestr);
12851 char * const ls = SvPVX(parser->linestr);
12853 parser->bufptr = ls + (proto->bufptr >= ols ?
12854 proto->bufptr - ols : 0);
12855 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12856 proto->oldbufptr - ols : 0);
12857 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12858 proto->oldoldbufptr - ols : 0);
12859 parser->linestart = ls + (proto->linestart >= ols ?
12860 proto->linestart - ols : 0);
12861 parser->last_uni = ls + (proto->last_uni >= ols ?
12862 proto->last_uni - ols : 0);
12863 parser->last_lop = ls + (proto->last_lop >= ols ?
12864 proto->last_lop - ols : 0);
12866 parser->bufend = ls + SvCUR(parser->linestr);
12869 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12872 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12873 Copy(proto->nexttype, parser->nexttype, 5, I32);
12874 parser->nexttoke = proto->nexttoke;
12876 /* XXX should clone saved_curcop here, but we aren't passed
12877 * proto_perl; so do it in perl_clone_using instead */
12883 /* duplicate a file handle */
12886 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12890 PERL_ARGS_ASSERT_FP_DUP;
12891 PERL_UNUSED_ARG(type);
12894 return (PerlIO*)NULL;
12896 /* look for it in the table first */
12897 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12901 /* create anew and remember what it is */
12902 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12903 ptr_table_store(PL_ptr_table, fp, ret);
12907 /* duplicate a directory handle */
12910 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12914 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12916 const Direntry_t *dirent;
12917 char smallbuf[256];
12923 PERL_UNUSED_CONTEXT;
12924 PERL_ARGS_ASSERT_DIRP_DUP;
12929 /* look for it in the table first */
12930 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12934 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12936 PERL_UNUSED_ARG(param);
12940 /* open the current directory (so we can switch back) */
12941 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12943 /* chdir to our dir handle and open the present working directory */
12944 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12945 PerlDir_close(pwd);
12946 return (DIR *)NULL;
12948 /* Now we should have two dir handles pointing to the same dir. */
12950 /* Be nice to the calling code and chdir back to where we were. */
12951 /* XXX If this fails, then what? */
12952 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12954 /* We have no need of the pwd handle any more. */
12955 PerlDir_close(pwd);
12958 # define d_namlen(d) (d)->d_namlen
12960 # define d_namlen(d) strlen((d)->d_name)
12962 /* Iterate once through dp, to get the file name at the current posi-
12963 tion. Then step back. */
12964 pos = PerlDir_tell(dp);
12965 if ((dirent = PerlDir_read(dp))) {
12966 len = d_namlen(dirent);
12967 if (len <= sizeof smallbuf) name = smallbuf;
12968 else Newx(name, len, char);
12969 Move(dirent->d_name, name, len, char);
12971 PerlDir_seek(dp, pos);
12973 /* Iterate through the new dir handle, till we find a file with the
12975 if (!dirent) /* just before the end */
12977 pos = PerlDir_tell(ret);
12978 if (PerlDir_read(ret)) continue; /* not there yet */
12979 PerlDir_seek(ret, pos); /* step back */
12983 const long pos0 = PerlDir_tell(ret);
12985 pos = PerlDir_tell(ret);
12986 if ((dirent = PerlDir_read(ret))) {
12987 if (len == (STRLEN)d_namlen(dirent)
12988 && memEQ(name, dirent->d_name, len)) {
12990 PerlDir_seek(ret, pos); /* step back */
12993 /* else we are not there yet; keep iterating */
12995 else { /* This is not meant to happen. The best we can do is
12996 reset the iterator to the beginning. */
12997 PerlDir_seek(ret, pos0);
13004 if (name && name != smallbuf)
13009 ret = win32_dirp_dup(dp, param);
13012 /* pop it in the pointer table */
13014 ptr_table_store(PL_ptr_table, dp, ret);
13019 /* duplicate a typeglob */
13022 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13026 PERL_ARGS_ASSERT_GP_DUP;
13030 /* look for it in the table first */
13031 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13035 /* create anew and remember what it is */
13037 ptr_table_store(PL_ptr_table, gp, ret);
13040 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13041 on Newxz() to do this for us. */
13042 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13043 ret->gp_io = io_dup_inc(gp->gp_io, param);
13044 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13045 ret->gp_av = av_dup_inc(gp->gp_av, param);
13046 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13047 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13048 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13049 ret->gp_cvgen = gp->gp_cvgen;
13050 ret->gp_line = gp->gp_line;
13051 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13055 /* duplicate a chain of magic */
13058 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13060 MAGIC *mgret = NULL;
13061 MAGIC **mgprev_p = &mgret;
13063 PERL_ARGS_ASSERT_MG_DUP;
13065 for (; mg; mg = mg->mg_moremagic) {
13068 if ((param->flags & CLONEf_JOIN_IN)
13069 && mg->mg_type == PERL_MAGIC_backref)
13070 /* when joining, we let the individual SVs add themselves to
13071 * backref as needed. */
13074 Newx(nmg, 1, MAGIC);
13076 mgprev_p = &(nmg->mg_moremagic);
13078 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13079 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13080 from the original commit adding Perl_mg_dup() - revision 4538.
13081 Similarly there is the annotation "XXX random ptr?" next to the
13082 assignment to nmg->mg_ptr. */
13085 /* FIXME for plugins
13086 if (nmg->mg_type == PERL_MAGIC_qr) {
13087 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13091 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13092 ? nmg->mg_type == PERL_MAGIC_backref
13093 /* The backref AV has its reference
13094 * count deliberately bumped by 1 */
13095 ? SvREFCNT_inc(av_dup_inc((const AV *)
13096 nmg->mg_obj, param))
13097 : sv_dup_inc(nmg->mg_obj, param)
13098 : sv_dup(nmg->mg_obj, param);
13100 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13101 if (nmg->mg_len > 0) {
13102 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13103 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13104 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13106 AMT * const namtp = (AMT*)nmg->mg_ptr;
13107 sv_dup_inc_multiple((SV**)(namtp->table),
13108 (SV**)(namtp->table), NofAMmeth, param);
13111 else if (nmg->mg_len == HEf_SVKEY)
13112 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13114 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13115 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13121 #endif /* USE_ITHREADS */
13123 struct ptr_tbl_arena {
13124 struct ptr_tbl_arena *next;
13125 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13128 /* create a new pointer-mapping table */
13131 Perl_ptr_table_new(pTHX)
13134 PERL_UNUSED_CONTEXT;
13136 Newx(tbl, 1, PTR_TBL_t);
13137 tbl->tbl_max = 511;
13138 tbl->tbl_items = 0;
13139 tbl->tbl_arena = NULL;
13140 tbl->tbl_arena_next = NULL;
13141 tbl->tbl_arena_end = NULL;
13142 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13146 #define PTR_TABLE_HASH(ptr) \
13147 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13149 /* map an existing pointer using a table */
13151 STATIC PTR_TBL_ENT_t *
13152 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13154 PTR_TBL_ENT_t *tblent;
13155 const UV hash = PTR_TABLE_HASH(sv);
13157 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13159 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13160 for (; tblent; tblent = tblent->next) {
13161 if (tblent->oldval == sv)
13168 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13170 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13172 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13173 PERL_UNUSED_CONTEXT;
13175 return tblent ? tblent->newval : NULL;
13178 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13179 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13180 * the core's typical use of ptr_tables in thread cloning. */
13183 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13185 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13187 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13188 PERL_UNUSED_CONTEXT;
13191 tblent->newval = newsv;
13193 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13195 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13196 struct ptr_tbl_arena *new_arena;
13198 Newx(new_arena, 1, struct ptr_tbl_arena);
13199 new_arena->next = tbl->tbl_arena;
13200 tbl->tbl_arena = new_arena;
13201 tbl->tbl_arena_next = new_arena->array;
13202 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13205 tblent = tbl->tbl_arena_next++;
13207 tblent->oldval = oldsv;
13208 tblent->newval = newsv;
13209 tblent->next = tbl->tbl_ary[entry];
13210 tbl->tbl_ary[entry] = tblent;
13212 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13213 ptr_table_split(tbl);
13217 /* double the hash bucket size of an existing ptr table */
13220 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13222 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13223 const UV oldsize = tbl->tbl_max + 1;
13224 UV newsize = oldsize * 2;
13227 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13228 PERL_UNUSED_CONTEXT;
13230 Renew(ary, newsize, PTR_TBL_ENT_t*);
13231 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13232 tbl->tbl_max = --newsize;
13233 tbl->tbl_ary = ary;
13234 for (i=0; i < oldsize; i++, ary++) {
13235 PTR_TBL_ENT_t **entp = ary;
13236 PTR_TBL_ENT_t *ent = *ary;
13237 PTR_TBL_ENT_t **curentp;
13240 curentp = ary + oldsize;
13242 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13244 ent->next = *curentp;
13254 /* remove all the entries from a ptr table */
13255 /* Deprecated - will be removed post 5.14 */
13258 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13260 PERL_UNUSED_CONTEXT;
13261 if (tbl && tbl->tbl_items) {
13262 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13264 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13267 struct ptr_tbl_arena *next = arena->next;
13273 tbl->tbl_items = 0;
13274 tbl->tbl_arena = NULL;
13275 tbl->tbl_arena_next = NULL;
13276 tbl->tbl_arena_end = NULL;
13280 /* clear and free a ptr table */
13283 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13285 struct ptr_tbl_arena *arena;
13287 PERL_UNUSED_CONTEXT;
13293 arena = tbl->tbl_arena;
13296 struct ptr_tbl_arena *next = arena->next;
13302 Safefree(tbl->tbl_ary);
13306 #if defined(USE_ITHREADS)
13309 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13311 PERL_ARGS_ASSERT_RVPV_DUP;
13313 assert(!isREGEXP(sstr));
13315 if (SvWEAKREF(sstr)) {
13316 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13317 if (param->flags & CLONEf_JOIN_IN) {
13318 /* if joining, we add any back references individually rather
13319 * than copying the whole backref array */
13320 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13324 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13326 else if (SvPVX_const(sstr)) {
13327 /* Has something there */
13329 /* Normal PV - clone whole allocated space */
13330 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13331 /* sstr may not be that normal, but actually copy on write.
13332 But we are a true, independent SV, so: */
13336 /* Special case - not normally malloced for some reason */
13337 if (isGV_with_GP(sstr)) {
13338 /* Don't need to do anything here. */
13340 else if ((SvIsCOW(sstr))) {
13341 /* A "shared" PV - clone it as "shared" PV */
13343 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13347 /* Some other special case - random pointer */
13348 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13353 /* Copy the NULL */
13354 SvPV_set(dstr, NULL);
13358 /* duplicate a list of SVs. source and dest may point to the same memory. */
13360 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13361 SSize_t items, CLONE_PARAMS *const param)
13363 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13365 while (items-- > 0) {
13366 *dest++ = sv_dup_inc(*source++, param);
13372 /* duplicate an SV of any type (including AV, HV etc) */
13375 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13380 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13382 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13383 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13388 /* look for it in the table first */
13389 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13393 if(param->flags & CLONEf_JOIN_IN) {
13394 /** We are joining here so we don't want do clone
13395 something that is bad **/
13396 if (SvTYPE(sstr) == SVt_PVHV) {
13397 const HEK * const hvname = HvNAME_HEK(sstr);
13399 /** don't clone stashes if they already exist **/
13400 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13401 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13402 ptr_table_store(PL_ptr_table, sstr, dstr);
13406 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13407 HV *stash = GvSTASH(sstr);
13408 const HEK * hvname;
13409 if (stash && (hvname = HvNAME_HEK(stash))) {
13410 /** don't clone GVs if they already exist **/
13412 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13413 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13415 stash, GvNAME(sstr),
13421 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13422 ptr_table_store(PL_ptr_table, sstr, *svp);
13429 /* create anew and remember what it is */
13432 #ifdef DEBUG_LEAKING_SCALARS
13433 dstr->sv_debug_optype = sstr->sv_debug_optype;
13434 dstr->sv_debug_line = sstr->sv_debug_line;
13435 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13436 dstr->sv_debug_parent = (SV*)sstr;
13437 FREE_SV_DEBUG_FILE(dstr);
13438 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13441 ptr_table_store(PL_ptr_table, sstr, dstr);
13444 SvFLAGS(dstr) = SvFLAGS(sstr);
13445 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13446 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13449 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13450 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13451 (void*)PL_watch_pvx, SvPVX_const(sstr));
13454 /* don't clone objects whose class has asked us not to */
13456 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13462 switch (SvTYPE(sstr)) {
13464 SvANY(dstr) = NULL;
13467 SET_SVANY_FOR_BODYLESS_IV(dstr);
13469 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13471 SvIV_set(dstr, SvIVX(sstr));
13475 #if NVSIZE <= IVSIZE
13476 SET_SVANY_FOR_BODYLESS_NV(dstr);
13478 SvANY(dstr) = new_XNV();
13480 SvNV_set(dstr, SvNVX(sstr));
13484 /* These are all the types that need complex bodies allocating. */
13486 const svtype sv_type = SvTYPE(sstr);
13487 const struct body_details *const sv_type_details
13488 = bodies_by_type + sv_type;
13492 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13508 assert(sv_type_details->body_size);
13509 if (sv_type_details->arena) {
13510 new_body_inline(new_body, sv_type);
13512 = (void*)((char*)new_body - sv_type_details->offset);
13514 new_body = new_NOARENA(sv_type_details);
13518 SvANY(dstr) = new_body;
13521 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13522 ((char*)SvANY(dstr)) + sv_type_details->offset,
13523 sv_type_details->copy, char);
13525 Copy(((char*)SvANY(sstr)),
13526 ((char*)SvANY(dstr)),
13527 sv_type_details->body_size + sv_type_details->offset, char);
13530 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13531 && !isGV_with_GP(dstr)
13533 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13534 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13536 /* The Copy above means that all the source (unduplicated) pointers
13537 are now in the destination. We can check the flags and the
13538 pointers in either, but it's possible that there's less cache
13539 missing by always going for the destination.
13540 FIXME - instrument and check that assumption */
13541 if (sv_type >= SVt_PVMG) {
13543 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13544 if (SvOBJECT(dstr) && SvSTASH(dstr))
13545 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13546 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13549 /* The cast silences a GCC warning about unhandled types. */
13550 switch ((int)sv_type) {
13561 /* FIXME for plugins */
13562 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13563 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13566 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13567 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13568 LvTARG(dstr) = dstr;
13569 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13570 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13572 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13573 if (isREGEXP(sstr)) goto duprex;
13575 /* non-GP case already handled above */
13576 if(isGV_with_GP(sstr)) {
13577 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13578 /* Don't call sv_add_backref here as it's going to be
13579 created as part of the magic cloning of the symbol
13580 table--unless this is during a join and the stash
13581 is not actually being cloned. */
13582 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13583 at the point of this comment. */
13584 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13585 if (param->flags & CLONEf_JOIN_IN)
13586 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13587 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13588 (void)GpREFCNT_inc(GvGP(dstr));
13592 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13593 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13594 /* I have no idea why fake dirp (rsfps)
13595 should be treated differently but otherwise
13596 we end up with leaks -- sky*/
13597 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13598 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13599 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13601 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13602 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13603 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13604 if (IoDIRP(dstr)) {
13605 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13608 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13610 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13612 if (IoOFP(dstr) == IoIFP(sstr))
13613 IoOFP(dstr) = IoIFP(dstr);
13615 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13616 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13617 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13618 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13621 /* avoid cloning an empty array */
13622 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13623 SV **dst_ary, **src_ary;
13624 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13626 src_ary = AvARRAY((const AV *)sstr);
13627 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13628 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13629 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13630 AvALLOC((const AV *)dstr) = dst_ary;
13631 if (AvREAL((const AV *)sstr)) {
13632 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13636 while (items-- > 0)
13637 *dst_ary++ = sv_dup(*src_ary++, param);
13639 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13640 while (items-- > 0) {
13645 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13646 AvALLOC((const AV *)dstr) = (SV**)NULL;
13647 AvMAX( (const AV *)dstr) = -1;
13648 AvFILLp((const AV *)dstr) = -1;
13652 if (HvARRAY((const HV *)sstr)) {
13654 const bool sharekeys = !!HvSHAREKEYS(sstr);
13655 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13656 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13658 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13659 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13661 HvARRAY(dstr) = (HE**)darray;
13662 while (i <= sxhv->xhv_max) {
13663 const HE * const source = HvARRAY(sstr)[i];
13664 HvARRAY(dstr)[i] = source
13665 ? he_dup(source, sharekeys, param) : 0;
13669 const struct xpvhv_aux * const saux = HvAUX(sstr);
13670 struct xpvhv_aux * const daux = HvAUX(dstr);
13671 /* This flag isn't copied. */
13674 if (saux->xhv_name_count) {
13675 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13677 = saux->xhv_name_count < 0
13678 ? -saux->xhv_name_count
13679 : saux->xhv_name_count;
13680 HEK **shekp = sname + count;
13682 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13683 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13684 while (shekp-- > sname) {
13686 *dhekp = hek_dup(*shekp, param);
13690 daux->xhv_name_u.xhvnameu_name
13691 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13694 daux->xhv_name_count = saux->xhv_name_count;
13696 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13697 daux->xhv_aux_flags = saux->xhv_aux_flags;
13698 #ifdef PERL_HASH_RANDOMIZE_KEYS
13699 daux->xhv_rand = saux->xhv_rand;
13700 daux->xhv_last_rand = saux->xhv_last_rand;
13702 daux->xhv_riter = saux->xhv_riter;
13703 daux->xhv_eiter = saux->xhv_eiter
13704 ? he_dup(saux->xhv_eiter,
13705 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13706 /* backref array needs refcnt=2; see sv_add_backref */
13707 daux->xhv_backreferences =
13708 (param->flags & CLONEf_JOIN_IN)
13709 /* when joining, we let the individual GVs and
13710 * CVs add themselves to backref as
13711 * needed. This avoids pulling in stuff
13712 * that isn't required, and simplifies the
13713 * case where stashes aren't cloned back
13714 * if they already exist in the parent
13717 : saux->xhv_backreferences
13718 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13719 ? MUTABLE_AV(SvREFCNT_inc(
13720 sv_dup_inc((const SV *)
13721 saux->xhv_backreferences, param)))
13722 : MUTABLE_AV(sv_dup((const SV *)
13723 saux->xhv_backreferences, param))
13726 daux->xhv_mro_meta = saux->xhv_mro_meta
13727 ? mro_meta_dup(saux->xhv_mro_meta, param)
13730 /* Record stashes for possible cloning in Perl_clone(). */
13732 av_push(param->stashes, dstr);
13736 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13739 if (!(param->flags & CLONEf_COPY_STACKS)) {
13744 /* NOTE: not refcounted */
13745 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13746 hv_dup(CvSTASH(dstr), param);
13747 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13748 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13749 if (!CvISXSUB(dstr)) {
13751 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13753 CvSLABBED_off(dstr);
13754 } else if (CvCONST(dstr)) {
13755 CvXSUBANY(dstr).any_ptr =
13756 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13758 assert(!CvSLABBED(dstr));
13759 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13761 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13762 hek_dup(CvNAME_HEK((CV *)sstr), param);
13763 /* don't dup if copying back - CvGV isn't refcounted, so the
13764 * duped GV may never be freed. A bit of a hack! DAPM */
13766 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13768 ? gv_dup_inc(CvGV(sstr), param)
13769 : (param->flags & CLONEf_JOIN_IN)
13771 : gv_dup(CvGV(sstr), param);
13773 if (!CvISXSUB(sstr)) {
13774 PADLIST * padlist = CvPADLIST(sstr);
13776 padlist = padlist_dup(padlist, param);
13777 CvPADLIST_set(dstr, padlist);
13779 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13780 PoisonPADLIST(dstr);
13783 CvWEAKOUTSIDE(sstr)
13784 ? cv_dup( CvOUTSIDE(dstr), param)
13785 : cv_dup_inc(CvOUTSIDE(dstr), param);
13795 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13797 PERL_ARGS_ASSERT_SV_DUP_INC;
13798 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13802 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13804 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13805 PERL_ARGS_ASSERT_SV_DUP;
13807 /* Track every SV that (at least initially) had a reference count of 0.
13808 We need to do this by holding an actual reference to it in this array.
13809 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13810 (akin to the stashes hash, and the perl stack), we come unstuck if
13811 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13812 thread) is manipulated in a CLONE method, because CLONE runs before the
13813 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13814 (and fix things up by giving each a reference via the temps stack).
13815 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13816 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13817 before the walk of unreferenced happens and a reference to that is SV
13818 added to the temps stack. At which point we have the same SV considered
13819 to be in use, and free to be re-used. Not good.
13821 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13822 assert(param->unreferenced);
13823 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13829 /* duplicate a context */
13832 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13834 PERL_CONTEXT *ncxs;
13836 PERL_ARGS_ASSERT_CX_DUP;
13839 return (PERL_CONTEXT*)NULL;
13841 /* look for it in the table first */
13842 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13846 /* create anew and remember what it is */
13847 Newx(ncxs, max + 1, PERL_CONTEXT);
13848 ptr_table_store(PL_ptr_table, cxs, ncxs);
13849 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13852 PERL_CONTEXT * const ncx = &ncxs[ix];
13853 if (CxTYPE(ncx) == CXt_SUBST) {
13854 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13857 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13858 switch (CxTYPE(ncx)) {
13860 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13861 ? cv_dup_inc(ncx->blk_sub.cv, param)
13862 : cv_dup(ncx->blk_sub.cv,param));
13863 if(CxHASARGS(ncx)){
13864 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13865 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13867 ncx->blk_sub.argarray = NULL;
13868 ncx->blk_sub.savearray = NULL;
13870 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13871 ncx->blk_sub.oldcomppad);
13874 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13876 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13877 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13879 case CXt_LOOP_LAZYSV:
13880 ncx->blk_loop.state_u.lazysv.end
13881 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13882 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
13883 duplication code instead.
13884 We are taking advantage of (1) av_dup_inc and sv_dup_inc
13885 actually being the same function, and (2) order
13886 equivalence of the two unions.
13887 We can assert the later [but only at run time :-(] */
13888 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13889 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13892 ncx->blk_loop.state_u.ary.ary
13893 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13895 case CXt_LOOP_LAZYIV:
13896 case CXt_LOOP_PLAIN:
13897 /* code common to all CXt_LOOP_* types */
13898 if (CxPADLOOP(ncx)) {
13899 ncx->blk_loop.itervar_u.oldcomppad
13900 = (PAD*)ptr_table_fetch(PL_ptr_table,
13901 ncx->blk_loop.itervar_u.oldcomppad);
13903 ncx->blk_loop.itervar_u.gv
13904 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13909 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13910 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13911 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13926 /* duplicate a stack info structure */
13929 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13933 PERL_ARGS_ASSERT_SI_DUP;
13936 return (PERL_SI*)NULL;
13938 /* look for it in the table first */
13939 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13943 /* create anew and remember what it is */
13944 Newxz(nsi, 1, PERL_SI);
13945 ptr_table_store(PL_ptr_table, si, nsi);
13947 nsi->si_stack = av_dup_inc(si->si_stack, param);
13948 nsi->si_cxix = si->si_cxix;
13949 nsi->si_cxmax = si->si_cxmax;
13950 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13951 nsi->si_type = si->si_type;
13952 nsi->si_prev = si_dup(si->si_prev, param);
13953 nsi->si_next = si_dup(si->si_next, param);
13954 nsi->si_markoff = si->si_markoff;
13959 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13960 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13961 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13962 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13963 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13964 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13965 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13966 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13967 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13968 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13969 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13970 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13971 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13972 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13973 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13974 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13977 #define pv_dup_inc(p) SAVEPV(p)
13978 #define pv_dup(p) SAVEPV(p)
13979 #define svp_dup_inc(p,pp) any_dup(p,pp)
13981 /* map any object to the new equivent - either something in the
13982 * ptr table, or something in the interpreter structure
13986 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13990 PERL_ARGS_ASSERT_ANY_DUP;
13993 return (void*)NULL;
13995 /* look for it in the table first */
13996 ret = ptr_table_fetch(PL_ptr_table, v);
14000 /* see if it is part of the interpreter structure */
14001 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14002 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14010 /* duplicate the save stack */
14013 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14016 ANY * const ss = proto_perl->Isavestack;
14017 const I32 max = proto_perl->Isavestack_max;
14018 I32 ix = proto_perl->Isavestack_ix;
14031 void (*dptr) (void*);
14032 void (*dxptr) (pTHX_ void*);
14034 PERL_ARGS_ASSERT_SS_DUP;
14036 Newxz(nss, max, ANY);
14039 const UV uv = POPUV(ss,ix);
14040 const U8 type = (U8)uv & SAVE_MASK;
14042 TOPUV(nss,ix) = uv;
14044 case SAVEt_CLEARSV:
14045 case SAVEt_CLEARPADRANGE:
14047 case SAVEt_HELEM: /* hash element */
14048 case SAVEt_SV: /* scalar reference */
14049 sv = (const SV *)POPPTR(ss,ix);
14050 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14052 case SAVEt_ITEM: /* normal string */
14053 case SAVEt_GVSV: /* scalar slot in GV */
14054 sv = (const SV *)POPPTR(ss,ix);
14055 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14056 if (type == SAVEt_SV)
14060 case SAVEt_MORTALIZESV:
14061 case SAVEt_READONLY_OFF:
14062 sv = (const SV *)POPPTR(ss,ix);
14063 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14065 case SAVEt_FREEPADNAME:
14066 ptr = POPPTR(ss,ix);
14067 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14068 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14070 case SAVEt_SHARED_PVREF: /* char* in shared space */
14071 c = (char*)POPPTR(ss,ix);
14072 TOPPTR(nss,ix) = savesharedpv(c);
14073 ptr = POPPTR(ss,ix);
14074 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14076 case SAVEt_GENERIC_SVREF: /* generic sv */
14077 case SAVEt_SVREF: /* scalar reference */
14078 sv = (const SV *)POPPTR(ss,ix);
14079 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14080 if (type == SAVEt_SVREF)
14081 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14082 ptr = POPPTR(ss,ix);
14083 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14085 case SAVEt_GVSLOT: /* any slot in GV */
14086 sv = (const SV *)POPPTR(ss,ix);
14087 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14088 ptr = POPPTR(ss,ix);
14089 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14090 sv = (const SV *)POPPTR(ss,ix);
14091 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14093 case SAVEt_HV: /* hash reference */
14094 case SAVEt_AV: /* array reference */
14095 sv = (const SV *) POPPTR(ss,ix);
14096 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14098 case SAVEt_COMPPAD:
14100 sv = (const SV *) POPPTR(ss,ix);
14101 TOPPTR(nss,ix) = sv_dup(sv, param);
14103 case SAVEt_INT: /* int reference */
14104 ptr = POPPTR(ss,ix);
14105 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14106 intval = (int)POPINT(ss,ix);
14107 TOPINT(nss,ix) = intval;
14109 case SAVEt_LONG: /* long reference */
14110 ptr = POPPTR(ss,ix);
14111 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14112 longval = (long)POPLONG(ss,ix);
14113 TOPLONG(nss,ix) = longval;
14115 case SAVEt_I32: /* I32 reference */
14116 ptr = POPPTR(ss,ix);
14117 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14119 TOPINT(nss,ix) = i;
14121 case SAVEt_IV: /* IV reference */
14122 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14123 ptr = POPPTR(ss,ix);
14124 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14126 TOPIV(nss,ix) = iv;
14128 case SAVEt_HPTR: /* HV* reference */
14129 case SAVEt_APTR: /* AV* reference */
14130 case SAVEt_SPTR: /* SV* reference */
14131 ptr = POPPTR(ss,ix);
14132 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14133 sv = (const SV *)POPPTR(ss,ix);
14134 TOPPTR(nss,ix) = sv_dup(sv, param);
14136 case SAVEt_VPTR: /* random* reference */
14137 ptr = POPPTR(ss,ix);
14138 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14140 case SAVEt_INT_SMALL:
14141 case SAVEt_I32_SMALL:
14142 case SAVEt_I16: /* I16 reference */
14143 case SAVEt_I8: /* I8 reference */
14145 ptr = POPPTR(ss,ix);
14146 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14148 case SAVEt_GENERIC_PVREF: /* generic char* */
14149 case SAVEt_PPTR: /* char* reference */
14150 ptr = POPPTR(ss,ix);
14151 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14152 c = (char*)POPPTR(ss,ix);
14153 TOPPTR(nss,ix) = pv_dup(c);
14155 case SAVEt_GP: /* scalar reference */
14156 gp = (GP*)POPPTR(ss,ix);
14157 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14158 (void)GpREFCNT_inc(gp);
14159 gv = (const GV *)POPPTR(ss,ix);
14160 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14163 ptr = POPPTR(ss,ix);
14164 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14165 /* these are assumed to be refcounted properly */
14167 switch (((OP*)ptr)->op_type) {
14169 case OP_LEAVESUBLV:
14173 case OP_LEAVEWRITE:
14174 TOPPTR(nss,ix) = ptr;
14177 (void) OpREFCNT_inc(o);
14181 TOPPTR(nss,ix) = NULL;
14186 TOPPTR(nss,ix) = NULL;
14188 case SAVEt_FREECOPHH:
14189 ptr = POPPTR(ss,ix);
14190 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14192 case SAVEt_ADELETE:
14193 av = (const AV *)POPPTR(ss,ix);
14194 TOPPTR(nss,ix) = av_dup_inc(av, param);
14196 TOPINT(nss,ix) = i;
14199 hv = (const HV *)POPPTR(ss,ix);
14200 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14202 TOPINT(nss,ix) = i;
14205 c = (char*)POPPTR(ss,ix);
14206 TOPPTR(nss,ix) = pv_dup_inc(c);
14208 case SAVEt_STACK_POS: /* Position on Perl stack */
14210 TOPINT(nss,ix) = i;
14212 case SAVEt_DESTRUCTOR:
14213 ptr = POPPTR(ss,ix);
14214 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14215 dptr = POPDPTR(ss,ix);
14216 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14217 any_dup(FPTR2DPTR(void *, dptr),
14220 case SAVEt_DESTRUCTOR_X:
14221 ptr = POPPTR(ss,ix);
14222 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14223 dxptr = POPDXPTR(ss,ix);
14224 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14225 any_dup(FPTR2DPTR(void *, dxptr),
14228 case SAVEt_REGCONTEXT:
14230 ix -= uv >> SAVE_TIGHT_SHIFT;
14232 case SAVEt_AELEM: /* array element */
14233 sv = (const SV *)POPPTR(ss,ix);
14234 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14236 TOPINT(nss,ix) = i;
14237 av = (const AV *)POPPTR(ss,ix);
14238 TOPPTR(nss,ix) = av_dup_inc(av, param);
14241 ptr = POPPTR(ss,ix);
14242 TOPPTR(nss,ix) = ptr;
14245 ptr = POPPTR(ss,ix);
14246 ptr = cophh_copy((COPHH*)ptr);
14247 TOPPTR(nss,ix) = ptr;
14249 TOPINT(nss,ix) = i;
14250 if (i & HINT_LOCALIZE_HH) {
14251 hv = (const HV *)POPPTR(ss,ix);
14252 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14255 case SAVEt_PADSV_AND_MORTALIZE:
14256 longval = (long)POPLONG(ss,ix);
14257 TOPLONG(nss,ix) = longval;
14258 ptr = POPPTR(ss,ix);
14259 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14260 sv = (const SV *)POPPTR(ss,ix);
14261 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14263 case SAVEt_SET_SVFLAGS:
14265 TOPINT(nss,ix) = i;
14267 TOPINT(nss,ix) = i;
14268 sv = (const SV *)POPPTR(ss,ix);
14269 TOPPTR(nss,ix) = sv_dup(sv, param);
14271 case SAVEt_COMPILE_WARNINGS:
14272 ptr = POPPTR(ss,ix);
14273 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14276 ptr = POPPTR(ss,ix);
14277 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14279 case SAVEt_GP_ALIASED_SV: {
14280 GP * gp_ptr = (GP *)POPPTR(ss,ix);
14281 GP * new_gp_ptr = gp_dup(gp_ptr, param);
14282 TOPPTR(nss,ix) = new_gp_ptr;
14283 new_gp_ptr->gp_refcnt++;
14288 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14296 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14297 * flag to the result. This is done for each stash before cloning starts,
14298 * so we know which stashes want their objects cloned */
14301 do_mark_cloneable_stash(pTHX_ SV *const sv)
14303 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14305 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14306 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14307 if (cloner && GvCV(cloner)) {
14314 mXPUSHs(newSVhek(hvname));
14316 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14323 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14331 =for apidoc perl_clone
14333 Create and return a new interpreter by cloning the current one.
14335 perl_clone takes these flags as parameters:
14337 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14338 without it we only clone the data and zero the stacks,
14339 with it we copy the stacks and the new perl interpreter is
14340 ready to run at the exact same point as the previous one.
14341 The pseudo-fork code uses COPY_STACKS while the
14342 threads->create doesn't.
14344 CLONEf_KEEP_PTR_TABLE -
14345 perl_clone keeps a ptr_table with the pointer of the old
14346 variable as a key and the new variable as a value,
14347 this allows it to check if something has been cloned and not
14348 clone it again but rather just use the value and increase the
14349 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14350 the ptr_table using the function
14351 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14352 reason to keep it around is if you want to dup some of your own
14353 variable who are outside the graph perl scans, example of this
14354 code is in threads.xs create.
14356 CLONEf_CLONE_HOST -
14357 This is a win32 thing, it is ignored on unix, it tells perls
14358 win32host code (which is c++) to clone itself, this is needed on
14359 win32 if you want to run two threads at the same time,
14360 if you just want to do some stuff in a separate perl interpreter
14361 and then throw it away and return to the original one,
14362 you don't need to do anything.
14367 /* XXX the above needs expanding by someone who actually understands it ! */
14368 EXTERN_C PerlInterpreter *
14369 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14372 perl_clone(PerlInterpreter *proto_perl, UV flags)
14375 #ifdef PERL_IMPLICIT_SYS
14377 PERL_ARGS_ASSERT_PERL_CLONE;
14379 /* perlhost.h so we need to call into it
14380 to clone the host, CPerlHost should have a c interface, sky */
14382 if (flags & CLONEf_CLONE_HOST) {
14383 return perl_clone_host(proto_perl,flags);
14385 return perl_clone_using(proto_perl, flags,
14387 proto_perl->IMemShared,
14388 proto_perl->IMemParse,
14390 proto_perl->IStdIO,
14394 proto_perl->IProc);
14398 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14399 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14400 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14401 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14402 struct IPerlDir* ipD, struct IPerlSock* ipS,
14403 struct IPerlProc* ipP)
14405 /* XXX many of the string copies here can be optimized if they're
14406 * constants; they need to be allocated as common memory and just
14407 * their pointers copied. */
14410 CLONE_PARAMS clone_params;
14411 CLONE_PARAMS* const param = &clone_params;
14413 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14415 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14416 #else /* !PERL_IMPLICIT_SYS */
14418 CLONE_PARAMS clone_params;
14419 CLONE_PARAMS* param = &clone_params;
14420 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14422 PERL_ARGS_ASSERT_PERL_CLONE;
14423 #endif /* PERL_IMPLICIT_SYS */
14425 /* for each stash, determine whether its objects should be cloned */
14426 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14427 PERL_SET_THX(my_perl);
14430 PoisonNew(my_perl, 1, PerlInterpreter);
14433 PL_defstash = NULL; /* may be used by perl malloc() */
14436 PL_scopestack_name = 0;
14438 PL_savestack_ix = 0;
14439 PL_savestack_max = -1;
14440 PL_sig_pending = 0;
14442 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14443 Zero(&PL_padname_undef, 1, PADNAME);
14444 Zero(&PL_padname_const, 1, PADNAME);
14445 # ifdef DEBUG_LEAKING_SCALARS
14446 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14448 # ifdef PERL_TRACE_OPS
14449 Zero(PL_op_exec_cnt, OP_max+2, UV);
14451 #else /* !DEBUGGING */
14452 Zero(my_perl, 1, PerlInterpreter);
14453 #endif /* DEBUGGING */
14455 #ifdef PERL_IMPLICIT_SYS
14456 /* host pointers */
14458 PL_MemShared = ipMS;
14459 PL_MemParse = ipMP;
14466 #endif /* PERL_IMPLICIT_SYS */
14469 param->flags = flags;
14470 /* Nothing in the core code uses this, but we make it available to
14471 extensions (using mg_dup). */
14472 param->proto_perl = proto_perl;
14473 /* Likely nothing will use this, but it is initialised to be consistent
14474 with Perl_clone_params_new(). */
14475 param->new_perl = my_perl;
14476 param->unreferenced = NULL;
14479 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14481 PL_body_arenas = NULL;
14482 Zero(&PL_body_roots, 1, PL_body_roots);
14486 PL_sv_arenaroot = NULL;
14488 PL_debug = proto_perl->Idebug;
14490 /* dbargs array probably holds garbage */
14493 PL_compiling = proto_perl->Icompiling;
14495 /* pseudo environmental stuff */
14496 PL_origargc = proto_perl->Iorigargc;
14497 PL_origargv = proto_perl->Iorigargv;
14499 #ifndef NO_TAINT_SUPPORT
14500 /* Set tainting stuff before PerlIO_debug can possibly get called */
14501 PL_tainting = proto_perl->Itainting;
14502 PL_taint_warn = proto_perl->Itaint_warn;
14504 PL_tainting = FALSE;
14505 PL_taint_warn = FALSE;
14508 PL_minus_c = proto_perl->Iminus_c;
14510 PL_localpatches = proto_perl->Ilocalpatches;
14511 PL_splitstr = proto_perl->Isplitstr;
14512 PL_minus_n = proto_perl->Iminus_n;
14513 PL_minus_p = proto_perl->Iminus_p;
14514 PL_minus_l = proto_perl->Iminus_l;
14515 PL_minus_a = proto_perl->Iminus_a;
14516 PL_minus_E = proto_perl->Iminus_E;
14517 PL_minus_F = proto_perl->Iminus_F;
14518 PL_doswitches = proto_perl->Idoswitches;
14519 PL_dowarn = proto_perl->Idowarn;
14520 PL_sawalias = proto_perl->Isawalias;
14521 #ifdef PERL_SAWAMPERSAND
14522 PL_sawampersand = proto_perl->Isawampersand;
14524 PL_unsafe = proto_perl->Iunsafe;
14525 PL_perldb = proto_perl->Iperldb;
14526 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14527 PL_exit_flags = proto_perl->Iexit_flags;
14529 /* XXX time(&PL_basetime) when asked for? */
14530 PL_basetime = proto_perl->Ibasetime;
14532 PL_maxsysfd = proto_perl->Imaxsysfd;
14533 PL_statusvalue = proto_perl->Istatusvalue;
14535 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14537 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14540 /* RE engine related */
14541 PL_regmatch_slab = NULL;
14542 PL_reg_curpm = NULL;
14544 PL_sub_generation = proto_perl->Isub_generation;
14546 /* funky return mechanisms */
14547 PL_forkprocess = proto_perl->Iforkprocess;
14549 /* internal state */
14550 PL_maxo = proto_perl->Imaxo;
14552 PL_main_start = proto_perl->Imain_start;
14553 PL_eval_root = proto_perl->Ieval_root;
14554 PL_eval_start = proto_perl->Ieval_start;
14556 PL_filemode = proto_perl->Ifilemode;
14557 PL_lastfd = proto_perl->Ilastfd;
14558 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14561 PL_gensym = proto_perl->Igensym;
14563 PL_laststatval = proto_perl->Ilaststatval;
14564 PL_laststype = proto_perl->Ilaststype;
14567 PL_profiledata = NULL;
14569 PL_generation = proto_perl->Igeneration;
14571 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14572 PL_in_clean_all = proto_perl->Iin_clean_all;
14574 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14575 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14576 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14577 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14578 PL_nomemok = proto_perl->Inomemok;
14579 PL_an = proto_perl->Ian;
14580 PL_evalseq = proto_perl->Ievalseq;
14581 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14582 PL_origalen = proto_perl->Iorigalen;
14584 PL_sighandlerp = proto_perl->Isighandlerp;
14586 PL_runops = proto_perl->Irunops;
14588 PL_subline = proto_perl->Isubline;
14590 PL_cv_has_eval = proto_perl->Icv_has_eval;
14593 PL_cryptseen = proto_perl->Icryptseen;
14596 #ifdef USE_LOCALE_COLLATE
14597 PL_collation_ix = proto_perl->Icollation_ix;
14598 PL_collation_standard = proto_perl->Icollation_standard;
14599 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14600 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14601 #endif /* USE_LOCALE_COLLATE */
14603 #ifdef USE_LOCALE_NUMERIC
14604 PL_numeric_standard = proto_perl->Inumeric_standard;
14605 PL_numeric_local = proto_perl->Inumeric_local;
14606 #endif /* !USE_LOCALE_NUMERIC */
14608 /* Did the locale setup indicate UTF-8? */
14609 PL_utf8locale = proto_perl->Iutf8locale;
14610 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14611 /* Unicode features (see perlrun/-C) */
14612 PL_unicode = proto_perl->Iunicode;
14614 /* Pre-5.8 signals control */
14615 PL_signals = proto_perl->Isignals;
14617 /* times() ticks per second */
14618 PL_clocktick = proto_perl->Iclocktick;
14620 /* Recursion stopper for PerlIO_find_layer */
14621 PL_in_load_module = proto_perl->Iin_load_module;
14623 /* sort() routine */
14624 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14626 /* Not really needed/useful since the reenrant_retint is "volatile",
14627 * but do it for consistency's sake. */
14628 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14630 /* Hooks to shared SVs and locks. */
14631 PL_sharehook = proto_perl->Isharehook;
14632 PL_lockhook = proto_perl->Ilockhook;
14633 PL_unlockhook = proto_perl->Iunlockhook;
14634 PL_threadhook = proto_perl->Ithreadhook;
14635 PL_destroyhook = proto_perl->Idestroyhook;
14636 PL_signalhook = proto_perl->Isignalhook;
14638 PL_globhook = proto_perl->Iglobhook;
14641 PL_last_swash_hv = NULL; /* reinits on demand */
14642 PL_last_swash_klen = 0;
14643 PL_last_swash_key[0]= '\0';
14644 PL_last_swash_tmps = (U8*)NULL;
14645 PL_last_swash_slen = 0;
14647 PL_srand_called = proto_perl->Isrand_called;
14648 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14650 if (flags & CLONEf_COPY_STACKS) {
14651 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14652 PL_tmps_ix = proto_perl->Itmps_ix;
14653 PL_tmps_max = proto_perl->Itmps_max;
14654 PL_tmps_floor = proto_perl->Itmps_floor;
14656 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14657 * NOTE: unlike the others! */
14658 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14659 PL_scopestack_max = proto_perl->Iscopestack_max;
14661 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14662 * NOTE: unlike the others! */
14663 PL_savestack_ix = proto_perl->Isavestack_ix;
14664 PL_savestack_max = proto_perl->Isavestack_max;
14667 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14668 PL_top_env = &PL_start_env;
14670 PL_op = proto_perl->Iop;
14673 PL_Xpv = (XPV*)NULL;
14674 my_perl->Ina = proto_perl->Ina;
14676 PL_statbuf = proto_perl->Istatbuf;
14677 PL_statcache = proto_perl->Istatcache;
14679 #ifndef NO_TAINT_SUPPORT
14680 PL_tainted = proto_perl->Itainted;
14682 PL_tainted = FALSE;
14684 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14686 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14688 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14689 PL_restartop = proto_perl->Irestartop;
14690 PL_in_eval = proto_perl->Iin_eval;
14691 PL_delaymagic = proto_perl->Idelaymagic;
14692 PL_phase = proto_perl->Iphase;
14693 PL_localizing = proto_perl->Ilocalizing;
14695 PL_hv_fetch_ent_mh = NULL;
14696 PL_modcount = proto_perl->Imodcount;
14697 PL_lastgotoprobe = NULL;
14698 PL_dumpindent = proto_perl->Idumpindent;
14700 PL_efloatbuf = NULL; /* reinits on demand */
14701 PL_efloatsize = 0; /* reinits on demand */
14705 PL_colorset = 0; /* reinits PL_colors[] */
14706 /*PL_colors[6] = {0,0,0,0,0,0};*/
14708 /* Pluggable optimizer */
14709 PL_peepp = proto_perl->Ipeepp;
14710 PL_rpeepp = proto_perl->Irpeepp;
14711 /* op_free() hook */
14712 PL_opfreehook = proto_perl->Iopfreehook;
14714 #ifdef USE_REENTRANT_API
14715 /* XXX: things like -Dm will segfault here in perlio, but doing
14716 * PERL_SET_CONTEXT(proto_perl);
14717 * breaks too many other things
14719 Perl_reentrant_init(aTHX);
14722 /* create SV map for pointer relocation */
14723 PL_ptr_table = ptr_table_new();
14725 /* initialize these special pointers as early as possible */
14727 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14728 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14729 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14730 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14731 &PL_padname_const);
14733 /* create (a non-shared!) shared string table */
14734 PL_strtab = newHV();
14735 HvSHAREKEYS_off(PL_strtab);
14736 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14737 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14739 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14741 /* This PV will be free'd special way so must set it same way op.c does */
14742 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14743 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14745 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14746 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14747 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14748 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14750 param->stashes = newAV(); /* Setup array of objects to call clone on */
14751 /* This makes no difference to the implementation, as it always pushes
14752 and shifts pointers to other SVs without changing their reference
14753 count, with the array becoming empty before it is freed. However, it
14754 makes it conceptually clear what is going on, and will avoid some
14755 work inside av.c, filling slots between AvFILL() and AvMAX() with
14756 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14757 AvREAL_off(param->stashes);
14759 if (!(flags & CLONEf_COPY_STACKS)) {
14760 param->unreferenced = newAV();
14763 #ifdef PERLIO_LAYERS
14764 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14765 PerlIO_clone(aTHX_ proto_perl, param);
14768 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14769 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14770 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14771 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14772 PL_xsubfilename = proto_perl->Ixsubfilename;
14773 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14774 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14777 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14778 PL_inplace = SAVEPV(proto_perl->Iinplace);
14779 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14781 /* magical thingies */
14783 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14784 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14786 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14787 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14788 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14791 /* Clone the regex array */
14792 /* ORANGE FIXME for plugins, probably in the SV dup code.
14793 newSViv(PTR2IV(CALLREGDUPE(
14794 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14796 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14797 PL_regex_pad = AvARRAY(PL_regex_padav);
14799 PL_stashpadmax = proto_perl->Istashpadmax;
14800 PL_stashpadix = proto_perl->Istashpadix ;
14801 Newx(PL_stashpad, PL_stashpadmax, HV *);
14804 for (; o < PL_stashpadmax; ++o)
14805 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14808 /* shortcuts to various I/O objects */
14809 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14810 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14811 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14812 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14813 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14814 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14815 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14817 /* shortcuts to regexp stuff */
14818 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14820 /* shortcuts to misc objects */
14821 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14823 /* shortcuts to debugging objects */
14824 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14825 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14826 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14827 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14828 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14829 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14830 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14832 /* symbol tables */
14833 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14834 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14835 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14836 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14837 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14839 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14840 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14841 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14842 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14843 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14844 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14845 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14846 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14847 PL_savebegin = proto_perl->Isavebegin;
14849 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14851 /* subprocess state */
14852 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14854 if (proto_perl->Iop_mask)
14855 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14858 /* PL_asserting = proto_perl->Iasserting; */
14860 /* current interpreter roots */
14861 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14863 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14866 /* runtime control stuff */
14867 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14869 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14871 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14873 /* interpreter atexit processing */
14874 PL_exitlistlen = proto_perl->Iexitlistlen;
14875 if (PL_exitlistlen) {
14876 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14877 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14880 PL_exitlist = (PerlExitListEntry*)NULL;
14882 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14883 if (PL_my_cxt_size) {
14884 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14885 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14886 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14887 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14888 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14892 PL_my_cxt_list = (void**)NULL;
14893 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14894 PL_my_cxt_keys = (const char**)NULL;
14897 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14898 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14899 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14900 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14902 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14904 PAD_CLONE_VARS(proto_perl, param);
14906 #ifdef HAVE_INTERP_INTERN
14907 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14910 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14912 #ifdef PERL_USES_PL_PIDSTATUS
14913 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14915 PL_osname = SAVEPV(proto_perl->Iosname);
14916 PL_parser = parser_dup(proto_perl->Iparser, param);
14918 /* XXX this only works if the saved cop has already been cloned */
14919 if (proto_perl->Iparser) {
14920 PL_parser->saved_curcop = (COP*)any_dup(
14921 proto_perl->Iparser->saved_curcop,
14925 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14927 #ifdef USE_LOCALE_CTYPE
14928 /* Should we warn if uses locale? */
14929 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
14932 #ifdef USE_LOCALE_COLLATE
14933 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14934 #endif /* USE_LOCALE_COLLATE */
14936 #ifdef USE_LOCALE_NUMERIC
14937 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14938 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14939 #endif /* !USE_LOCALE_NUMERIC */
14941 /* Unicode inversion lists */
14942 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14943 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14944 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14945 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14947 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14948 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14950 /* utf8 character class swashes */
14951 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14952 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14954 for (i = 0; i < POSIX_CC_COUNT; i++) {
14955 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14957 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
14958 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
14959 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
14960 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14961 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14962 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14963 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14964 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14965 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14966 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14967 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14968 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14969 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14970 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14971 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14972 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14973 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14975 if (proto_perl->Ipsig_pend) {
14976 Newxz(PL_psig_pend, SIG_SIZE, int);
14979 PL_psig_pend = (int*)NULL;
14982 if (proto_perl->Ipsig_name) {
14983 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14984 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14986 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14989 PL_psig_ptr = (SV**)NULL;
14990 PL_psig_name = (SV**)NULL;
14993 if (flags & CLONEf_COPY_STACKS) {
14994 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14995 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14996 PL_tmps_ix+1, param);
14998 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14999 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15000 Newxz(PL_markstack, i, I32);
15001 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15002 - proto_perl->Imarkstack);
15003 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15004 - proto_perl->Imarkstack);
15005 Copy(proto_perl->Imarkstack, PL_markstack,
15006 PL_markstack_ptr - PL_markstack + 1, I32);
15008 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15009 * NOTE: unlike the others! */
15010 Newxz(PL_scopestack, PL_scopestack_max, I32);
15011 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15014 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15015 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15017 /* reset stack AV to correct length before its duped via
15018 * PL_curstackinfo */
15019 AvFILLp(proto_perl->Icurstack) =
15020 proto_perl->Istack_sp - proto_perl->Istack_base;
15022 /* NOTE: si_dup() looks at PL_markstack */
15023 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15025 /* PL_curstack = PL_curstackinfo->si_stack; */
15026 PL_curstack = av_dup(proto_perl->Icurstack, param);
15027 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15029 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15030 PL_stack_base = AvARRAY(PL_curstack);
15031 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15032 - proto_perl->Istack_base);
15033 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15035 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15036 PL_savestack = ss_dup(proto_perl, param);
15040 ENTER; /* perl_destruct() wants to LEAVE; */
15043 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15044 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15046 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15047 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15048 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15049 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15050 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15051 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15053 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15055 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15056 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15057 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15059 PL_stashcache = newHV();
15061 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15062 proto_perl->Iwatchaddr);
15063 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15064 if (PL_debug && PL_watchaddr) {
15065 PerlIO_printf(Perl_debug_log,
15066 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15067 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15068 PTR2UV(PL_watchok));
15071 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15072 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15073 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15075 /* Call the ->CLONE method, if it exists, for each of the stashes
15076 identified by sv_dup() above.
15078 while(av_tindex(param->stashes) != -1) {
15079 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15080 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15081 if (cloner && GvCV(cloner)) {
15086 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15088 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15094 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15095 ptr_table_free(PL_ptr_table);
15096 PL_ptr_table = NULL;
15099 if (!(flags & CLONEf_COPY_STACKS)) {
15100 unreferenced_to_tmp_stack(param->unreferenced);
15103 SvREFCNT_dec(param->stashes);
15105 /* orphaned? eg threads->new inside BEGIN or use */
15106 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15107 SvREFCNT_inc_simple_void(PL_compcv);
15108 SAVEFREESV(PL_compcv);
15115 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15117 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15119 if (AvFILLp(unreferenced) > -1) {
15120 SV **svp = AvARRAY(unreferenced);
15121 SV **const last = svp + AvFILLp(unreferenced);
15125 if (SvREFCNT(*svp) == 1)
15127 } while (++svp <= last);
15129 EXTEND_MORTAL(count);
15130 svp = AvARRAY(unreferenced);
15133 if (SvREFCNT(*svp) == 1) {
15134 /* Our reference is the only one to this SV. This means that
15135 in this thread, the scalar effectively has a 0 reference.
15136 That doesn't work (cleanup never happens), so donate our
15137 reference to it onto the save stack. */
15138 PL_tmps_stack[++PL_tmps_ix] = *svp;
15140 /* As an optimisation, because we are already walking the
15141 entire array, instead of above doing either
15142 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15143 release our reference to the scalar, so that at the end of
15144 the array owns zero references to the scalars it happens to
15145 point to. We are effectively converting the array from
15146 AvREAL() on to AvREAL() off. This saves the av_clear()
15147 (triggered by the SvREFCNT_dec(unreferenced) below) from
15148 walking the array a second time. */
15149 SvREFCNT_dec(*svp);
15152 } while (++svp <= last);
15153 AvREAL_off(unreferenced);
15155 SvREFCNT_dec_NN(unreferenced);
15159 Perl_clone_params_del(CLONE_PARAMS *param)
15161 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15163 PerlInterpreter *const to = param->new_perl;
15165 PerlInterpreter *const was = PERL_GET_THX;
15167 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15173 SvREFCNT_dec(param->stashes);
15174 if (param->unreferenced)
15175 unreferenced_to_tmp_stack(param->unreferenced);
15185 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15188 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15189 does a dTHX; to get the context from thread local storage.
15190 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15191 a version that passes in my_perl. */
15192 PerlInterpreter *const was = PERL_GET_THX;
15193 CLONE_PARAMS *param;
15195 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15201 /* Given that we've set the context, we can do this unshared. */
15202 Newx(param, 1, CLONE_PARAMS);
15205 param->proto_perl = from;
15206 param->new_perl = to;
15207 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15208 AvREAL_off(param->stashes);
15209 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15217 #endif /* USE_ITHREADS */
15220 Perl_init_constants(pTHX)
15222 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15223 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15224 SvANY(&PL_sv_undef) = NULL;
15226 SvANY(&PL_sv_no) = new_XPVNV();
15227 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15228 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15229 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15232 SvANY(&PL_sv_yes) = new_XPVNV();
15233 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15234 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15235 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15238 SvPV_set(&PL_sv_no, (char*)PL_No);
15239 SvCUR_set(&PL_sv_no, 0);
15240 SvLEN_set(&PL_sv_no, 0);
15241 SvIV_set(&PL_sv_no, 0);
15242 SvNV_set(&PL_sv_no, 0);
15244 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15245 SvCUR_set(&PL_sv_yes, 1);
15246 SvLEN_set(&PL_sv_yes, 0);
15247 SvIV_set(&PL_sv_yes, 1);
15248 SvNV_set(&PL_sv_yes, 1);
15250 PadnamePV(&PL_padname_const) = (char *)PL_No;
15254 =head1 Unicode Support
15256 =for apidoc sv_recode_to_utf8
15258 The encoding is assumed to be an Encode object, on entry the PV
15259 of the sv is assumed to be octets in that encoding, and the sv
15260 will be converted into Unicode (and UTF-8).
15262 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15263 is not a reference, nothing is done to the sv. If the encoding is not
15264 an C<Encode::XS> Encoding object, bad things will happen.
15265 (See F<lib/encoding.pm> and L<Encode>.)
15267 The PV of the sv is returned.
15272 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15274 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15276 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15285 if (SvPADTMP(nsv)) {
15286 nsv = sv_newmortal();
15287 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)) {
15366 offsv = newSViv(*offset);
15368 mPUSHp(tstr, tlen);
15370 call_method("cat_decode", G_SCALAR);
15372 ret = SvTRUE(TOPs);
15373 *offset = SvIV(offsv);
15379 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15384 /* ---------------------------------------------------------------------
15386 * support functions for report_uninit()
15389 /* the maxiumum size of array or hash where we will scan looking
15390 * for the undefined element that triggered the warning */
15392 #define FUV_MAX_SEARCH_SIZE 1000
15394 /* Look for an entry in the hash whose value has the same SV as val;
15395 * If so, return a mortal copy of the key. */
15398 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15404 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15406 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15407 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15410 array = HvARRAY(hv);
15412 for (i=HvMAX(hv); i>=0; i--) {
15414 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15415 if (HeVAL(entry) != val)
15417 if ( HeVAL(entry) == &PL_sv_undef ||
15418 HeVAL(entry) == &PL_sv_placeholder)
15422 if (HeKLEN(entry) == HEf_SVKEY)
15423 return sv_mortalcopy(HeKEY_sv(entry));
15424 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15430 /* Look for an entry in the array whose value has the same SV as val;
15431 * If so, return the index, otherwise return -1. */
15434 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15436 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15438 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15439 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15442 if (val != &PL_sv_undef) {
15443 SV ** const svp = AvARRAY(av);
15446 for (i=AvFILLp(av); i>=0; i--)
15453 /* varname(): return the name of a variable, optionally with a subscript.
15454 * If gv is non-zero, use the name of that global, along with gvtype (one
15455 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15456 * targ. Depending on the value of the subscript_type flag, return:
15459 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15460 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15461 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15462 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15465 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15466 const SV *const keyname, I32 aindex, int subscript_type)
15469 SV * const name = sv_newmortal();
15470 if (gv && isGV(gv)) {
15472 buffer[0] = gvtype;
15475 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15477 gv_fullname4(name, gv, buffer, 0);
15479 if ((unsigned int)SvPVX(name)[1] <= 26) {
15481 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15483 /* Swap the 1 unprintable control character for the 2 byte pretty
15484 version - ie substr($name, 1, 1) = $buffer; */
15485 sv_insert(name, 1, 1, buffer, 2);
15489 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15492 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15494 if (!cv || !CvPADLIST(cv))
15496 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15497 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15501 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15502 SV * const sv = newSV(0);
15503 *SvPVX(name) = '$';
15504 Perl_sv_catpvf(aTHX_ name, "{%s}",
15505 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15506 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15507 SvREFCNT_dec_NN(sv);
15509 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15510 *SvPVX(name) = '$';
15511 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15513 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15514 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15515 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15523 =for apidoc find_uninit_var
15525 Find the name of the undefined variable (if any) that caused the operator
15526 to issue a "Use of uninitialized value" warning.
15527 If match is true, only return a name if its value matches uninit_sv.
15528 So roughly speaking, if a unary operator (such as OP_COS) generates a
15529 warning, then following the direct child of the op may yield an
15530 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15531 other hand, with OP_ADD there are two branches to follow, so we only print
15532 the variable name if we get an exact match.
15533 desc_p points to a string pointer holding the description of the op.
15534 This may be updated if needed.
15536 The name is returned as a mortal SV.
15538 Assumes that PL_op is the op that originally triggered the error, and that
15539 PL_comppad/PL_curpad points to the currently executing pad.
15545 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15546 bool match, const char **desc_p)
15551 const OP *o, *o2, *kid;
15553 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15555 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15556 uninit_sv == &PL_sv_placeholder)))
15559 switch (obase->op_type) {
15566 const bool pad = ( obase->op_type == OP_PADAV
15567 || obase->op_type == OP_PADHV
15568 || obase->op_type == OP_PADRANGE
15571 const bool hash = ( obase->op_type == OP_PADHV
15572 || obase->op_type == OP_RV2HV
15573 || (obase->op_type == OP_PADRANGE
15574 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15578 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15580 if (pad) { /* @lex, %lex */
15581 sv = PAD_SVl(obase->op_targ);
15585 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15586 /* @global, %global */
15587 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15590 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15592 else if (obase == PL_op) /* @{expr}, %{expr} */
15593 return find_uninit_var(cUNOPx(obase)->op_first,
15594 uninit_sv, match, desc_p);
15595 else /* @{expr}, %{expr} as a sub-expression */
15599 /* attempt to find a match within the aggregate */
15601 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15603 subscript_type = FUV_SUBSCRIPT_HASH;
15606 index = find_array_subscript((const AV *)sv, uninit_sv);
15608 subscript_type = FUV_SUBSCRIPT_ARRAY;
15611 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15614 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15615 keysv, index, subscript_type);
15619 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15621 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15622 if (!gv || !GvSTASH(gv))
15624 if (match && (GvSV(gv) != uninit_sv))
15626 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15629 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15632 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15634 return varname(NULL, '$', obase->op_targ,
15635 NULL, 0, FUV_SUBSCRIPT_NONE);
15638 gv = cGVOPx_gv(obase);
15639 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15641 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15643 case OP_AELEMFAST_LEX:
15646 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15647 if (!av || SvRMAGICAL(av))
15649 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15650 if (!svp || *svp != uninit_sv)
15653 return varname(NULL, '$', obase->op_targ,
15654 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15657 gv = cGVOPx_gv(obase);
15662 AV *const av = GvAV(gv);
15663 if (!av || SvRMAGICAL(av))
15665 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15666 if (!svp || *svp != uninit_sv)
15669 return varname(gv, '$', 0,
15670 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15672 NOT_REACHED; /* NOTREACHED */
15675 o = cUNOPx(obase)->op_first;
15676 if (!o || o->op_type != OP_NULL ||
15677 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15679 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15684 bool negate = FALSE;
15686 if (PL_op == obase)
15687 /* $a[uninit_expr] or $h{uninit_expr} */
15688 return find_uninit_var(cBINOPx(obase)->op_last,
15689 uninit_sv, match, desc_p);
15692 o = cBINOPx(obase)->op_first;
15693 kid = cBINOPx(obase)->op_last;
15695 /* get the av or hv, and optionally the gv */
15697 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15698 sv = PAD_SV(o->op_targ);
15700 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15701 && cUNOPo->op_first->op_type == OP_GV)
15703 gv = cGVOPx_gv(cUNOPo->op_first);
15707 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15712 if (kid && kid->op_type == OP_NEGATE) {
15714 kid = cUNOPx(kid)->op_first;
15717 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15718 /* index is constant */
15721 kidsv = newSVpvs_flags("-", SVs_TEMP);
15722 sv_catsv(kidsv, cSVOPx_sv(kid));
15725 kidsv = cSVOPx_sv(kid);
15729 if (obase->op_type == OP_HELEM) {
15730 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15731 if (!he || HeVAL(he) != uninit_sv)
15735 SV * const opsv = cSVOPx_sv(kid);
15736 const IV opsviv = SvIV(opsv);
15737 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15738 negate ? - opsviv : opsviv,
15740 if (!svp || *svp != uninit_sv)
15744 if (obase->op_type == OP_HELEM)
15745 return varname(gv, '%', o->op_targ,
15746 kidsv, 0, FUV_SUBSCRIPT_HASH);
15748 return varname(gv, '@', o->op_targ, NULL,
15749 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15750 FUV_SUBSCRIPT_ARRAY);
15753 /* index is an expression;
15754 * attempt to find a match within the aggregate */
15755 if (obase->op_type == OP_HELEM) {
15756 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15758 return varname(gv, '%', o->op_targ,
15759 keysv, 0, FUV_SUBSCRIPT_HASH);
15763 = find_array_subscript((const AV *)sv, uninit_sv);
15765 return varname(gv, '@', o->op_targ,
15766 NULL, index, FUV_SUBSCRIPT_ARRAY);
15771 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15773 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15775 NOT_REACHED; /* NOTREACHED */
15778 case OP_MULTIDEREF: {
15779 /* If we were executing OP_MULTIDEREF when the undef warning
15780 * triggered, then it must be one of the index values within
15781 * that triggered it. If not, then the only possibility is that
15782 * the value retrieved by the last aggregate lookup might be the
15783 * culprit. For the former, we set PL_multideref_pc each time before
15784 * using an index, so work though the item list until we reach
15785 * that point. For the latter, just work through the entire item
15786 * list; the last aggregate retrieved will be the candidate.
15789 /* the named aggregate, if any */
15790 PADOFFSET agg_targ = 0;
15792 /* the last-seen index */
15794 PADOFFSET index_targ;
15796 IV index_const_iv = 0; /* init for spurious compiler warn */
15797 SV *index_const_sv;
15798 int depth = 0; /* how many array/hash lookups we've done */
15800 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15801 UNOP_AUX_item *last = NULL;
15802 UV actions = items->uv;
15805 if (PL_op == obase) {
15806 last = PL_multideref_pc;
15807 assert(last >= items && last <= items + items[-1].uv);
15814 switch (actions & MDEREF_ACTION_MASK) {
15816 case MDEREF_reload:
15817 actions = (++items)->uv;
15820 case MDEREF_HV_padhv_helem: /* $lex{...} */
15823 case MDEREF_AV_padav_aelem: /* $lex[...] */
15824 agg_targ = (++items)->pad_offset;
15828 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15831 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15833 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15834 assert(isGV_with_GP(agg_gv));
15837 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15838 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15841 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15842 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15848 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15849 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15852 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15853 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15860 index_const_sv = NULL;
15862 index_type = (actions & MDEREF_INDEX_MASK);
15863 switch (index_type) {
15864 case MDEREF_INDEX_none:
15866 case MDEREF_INDEX_const:
15868 index_const_sv = UNOP_AUX_item_sv(++items)
15870 index_const_iv = (++items)->iv;
15872 case MDEREF_INDEX_padsv:
15873 index_targ = (++items)->pad_offset;
15875 case MDEREF_INDEX_gvsv:
15876 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15877 assert(isGV_with_GP(index_gv));
15881 if (index_type != MDEREF_INDEX_none)
15884 if ( index_type == MDEREF_INDEX_none
15885 || (actions & MDEREF_FLAG_last)
15886 || (last && items == last)
15890 actions >>= MDEREF_SHIFT;
15893 if (PL_op == obase) {
15894 /* index was undef */
15896 *desc_p = ( (actions & MDEREF_FLAG_last)
15897 && (obase->op_private
15898 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15900 (obase->op_private & OPpMULTIDEREF_EXISTS)
15903 : is_hv ? "hash element" : "array element";
15904 assert(index_type != MDEREF_INDEX_none);
15906 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15908 return varname(NULL, '$', index_targ,
15909 NULL, 0, FUV_SUBSCRIPT_NONE);
15910 assert(is_hv); /* AV index is an IV and can't be undef */
15911 /* can a const HV index ever be undef? */
15915 /* the SV returned by pp_multideref() was undef, if anything was */
15921 sv = PAD_SV(agg_targ);
15923 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15927 if (index_type == MDEREF_INDEX_const) {
15932 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15933 if (!he || HeVAL(he) != uninit_sv)
15937 SV * const * const svp =
15938 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15939 if (!svp || *svp != uninit_sv)
15944 ? varname(agg_gv, '%', agg_targ,
15945 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15946 : varname(agg_gv, '@', agg_targ,
15947 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15950 /* index is an var */
15952 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15954 return varname(agg_gv, '%', agg_targ,
15955 keysv, 0, FUV_SUBSCRIPT_HASH);
15959 = find_array_subscript((const AV *)sv, uninit_sv);
15961 return varname(agg_gv, '@', agg_targ,
15962 NULL, index, FUV_SUBSCRIPT_ARRAY);
15966 return varname(agg_gv,
15968 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15970 NOT_REACHED; /* NOTREACHED */
15974 /* only examine RHS */
15975 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
15979 o = cUNOPx(obase)->op_first;
15980 if ( o->op_type == OP_PUSHMARK
15981 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15985 if (!OpHAS_SIBLING(o)) {
15986 /* one-arg version of open is highly magical */
15988 if (o->op_type == OP_GV) { /* open FOO; */
15990 if (match && GvSV(gv) != uninit_sv)
15992 return varname(gv, '$', 0,
15993 NULL, 0, FUV_SUBSCRIPT_NONE);
15995 /* other possibilities not handled are:
15996 * open $x; or open my $x; should return '${*$x}'
15997 * open expr; should return '$'.expr ideally
16003 /* ops where $_ may be an implicit arg */
16008 if ( !(obase->op_flags & OPf_STACKED)) {
16009 if (uninit_sv == DEFSV)
16010 return newSVpvs_flags("$_", SVs_TEMP);
16011 else if (obase->op_targ
16012 && uninit_sv == PAD_SVl(obase->op_targ))
16013 return varname(NULL, '$', obase->op_targ, NULL, 0,
16014 FUV_SUBSCRIPT_NONE);
16021 match = 1; /* print etc can return undef on defined args */
16022 /* skip filehandle as it can't produce 'undef' warning */
16023 o = cUNOPx(obase)->op_first;
16024 if ((obase->op_flags & OPf_STACKED)
16026 ( o->op_type == OP_PUSHMARK
16027 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16028 o = OpSIBLING(OpSIBLING(o));
16032 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16033 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16035 /* the following ops are capable of returning PL_sv_undef even for
16036 * defined arg(s) */
16055 case OP_GETPEERNAME:
16103 case OP_SMARTMATCH:
16112 /* XXX tmp hack: these two may call an XS sub, and currently
16113 XS subs don't have a SUB entry on the context stack, so CV and
16114 pad determination goes wrong, and BAD things happen. So, just
16115 don't try to determine the value under those circumstances.
16116 Need a better fix at dome point. DAPM 11/2007 */
16122 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16123 if (gv && GvSV(gv) == uninit_sv)
16124 return newSVpvs_flags("$.", SVs_TEMP);
16129 /* def-ness of rval pos() is independent of the def-ness of its arg */
16130 if ( !(obase->op_flags & OPf_MOD))
16135 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16136 return newSVpvs_flags("${$/}", SVs_TEMP);
16141 if (!(obase->op_flags & OPf_KIDS))
16143 o = cUNOPx(obase)->op_first;
16149 /* This loop checks all the kid ops, skipping any that cannot pos-
16150 * sibly be responsible for the uninitialized value; i.e., defined
16151 * constants and ops that return nothing. If there is only one op
16152 * left that is not skipped, then we *know* it is responsible for
16153 * the uninitialized value. If there is more than one op left, we
16154 * have to look for an exact match in the while() loop below.
16155 * Note that we skip padrange, because the individual pad ops that
16156 * it replaced are still in the tree, so we work on them instead.
16159 for (kid=o; kid; kid = OpSIBLING(kid)) {
16160 const OPCODE type = kid->op_type;
16161 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16162 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16163 || (type == OP_PUSHMARK)
16164 || (type == OP_PADRANGE)
16168 if (o2) { /* more than one found */
16175 return find_uninit_var(o2, uninit_sv, match, desc_p);
16177 /* scan all args */
16179 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16191 =for apidoc report_uninit
16193 Print appropriate "Use of uninitialized variable" warning.
16199 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16201 const char *desc = NULL;
16202 SV* varname = NULL;
16205 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16208 if (uninit_sv && PL_curpad) {
16209 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16211 sv_insert(varname, 0, 0, " ", 1);
16214 else if (PL_curstackinfo->si_type == PERLSI_SORT
16215 && CxMULTICALL(&cxstack[cxstack_ix]))
16217 /* we've reached the end of a sort block or sub,
16218 * and the uninit value is probably what that code returned */
16222 /* PL_warn_uninit_sv is constant */
16223 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16225 /* diag_listed_as: Use of uninitialized value%s */
16226 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16227 SVfARG(varname ? varname : &PL_sv_no),
16230 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16236 * ex: set ts=8 sts=4 sw=4 et:
https://perl5.git.perl.org / perl5.git / blob