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
36 # if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L && !defined(VMS)
45 /* Missing proto on LynxOS */
46 char *gconvert(double, int, int, char *);
49 /* void Gconvert: on Linux at least, gcvt (which Gconvert gets deffed to),
50 * has a mandatory return value, even though that value is just the same
53 #define V_Gconvert(x,n,t,b) \
55 char *rc = (char *)Gconvert(x,n,t,b); \
56 PERL_UNUSED_VAR(rc); \
60 #ifdef PERL_UTF8_CACHE_ASSERT
61 /* if adding more checks watch out for the following tests:
62 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
63 * lib/utf8.t lib/Unicode/Collate/t/index.t
66 # define ASSERT_UTF8_CACHE(cache) \
67 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
68 assert((cache)[2] <= (cache)[3]); \
69 assert((cache)[3] <= (cache)[1]);} \
72 # define ASSERT_UTF8_CACHE(cache) NOOP
75 #ifdef PERL_OLD_COPY_ON_WRITE
76 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
77 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
80 /* ============================================================================
82 =head1 Allocation and deallocation of SVs.
84 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
85 sv, av, hv...) contains type and reference count information, and for
86 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
87 contains fields specific to each type. Some types store all they need
88 in the head, so don't have a body.
90 In all but the most memory-paranoid configurations (ex: PURIFY), heads
91 and bodies are allocated out of arenas, which by default are
92 approximately 4K chunks of memory parcelled up into N heads or bodies.
93 Sv-bodies are allocated by their sv-type, guaranteeing size
94 consistency needed to allocate safely from arrays.
96 For SV-heads, the first slot in each arena is reserved, and holds a
97 link to the next arena, some flags, and a note of the number of slots.
98 Snaked through each arena chain is a linked list of free items; when
99 this becomes empty, an extra arena is allocated and divided up into N
100 items which are threaded into the free list.
102 SV-bodies are similar, but they use arena-sets by default, which
103 separate the link and info from the arena itself, and reclaim the 1st
104 slot in the arena. SV-bodies are further described later.
106 The following global variables are associated with arenas:
108 PL_sv_arenaroot pointer to list of SV arenas
109 PL_sv_root pointer to list of free SV structures
111 PL_body_arenas head of linked-list of body arenas
112 PL_body_roots[] array of pointers to list of free bodies of svtype
113 arrays are indexed by the svtype needed
115 A few special SV heads are not allocated from an arena, but are
116 instead directly created in the interpreter structure, eg PL_sv_undef.
117 The size of arenas can be changed from the default by setting
118 PERL_ARENA_SIZE appropriately at compile time.
120 The SV arena serves the secondary purpose of allowing still-live SVs
121 to be located and destroyed during final cleanup.
123 At the lowest level, the macros new_SV() and del_SV() grab and free
124 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
125 to return the SV to the free list with error checking.) new_SV() calls
126 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
127 SVs in the free list have their SvTYPE field set to all ones.
129 At the time of very final cleanup, sv_free_arenas() is called from
130 perl_destruct() to physically free all the arenas allocated since the
131 start of the interpreter.
133 The function visit() scans the SV arenas list, and calls a specified
134 function for each SV it finds which is still live - ie which has an SvTYPE
135 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
136 following functions (specified as [function that calls visit()] / [function
137 called by visit() for each SV]):
139 sv_report_used() / do_report_used()
140 dump all remaining SVs (debugging aid)
142 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
143 do_clean_named_io_objs(),do_curse()
144 Attempt to free all objects pointed to by RVs,
145 try to do the same for all objects indir-
146 ectly referenced by typeglobs too, and
147 then do a final sweep, cursing any
148 objects that remain. Called once from
149 perl_destruct(), prior to calling sv_clean_all()
152 sv_clean_all() / do_clean_all()
153 SvREFCNT_dec(sv) each remaining SV, possibly
154 triggering an sv_free(). It also sets the
155 SVf_BREAK flag on the SV to indicate that the
156 refcnt has been artificially lowered, and thus
157 stopping sv_free() from giving spurious warnings
158 about SVs which unexpectedly have a refcnt
159 of zero. called repeatedly from perl_destruct()
160 until there are no SVs left.
162 =head2 Arena allocator API Summary
164 Private API to rest of sv.c
168 new_XPVNV(), del_XPVGV(),
173 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
177 * ========================================================================= */
180 * "A time to plant, and a time to uproot what was planted..."
184 # define MEM_LOG_NEW_SV(sv, file, line, func) \
185 Perl_mem_log_new_sv(sv, file, line, func)
186 # define MEM_LOG_DEL_SV(sv, file, line, func) \
187 Perl_mem_log_del_sv(sv, file, line, func)
189 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
190 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
193 #ifdef DEBUG_LEAKING_SCALARS
194 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
195 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
197 # define DEBUG_SV_SERIAL(sv) \
198 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
199 PTR2UV(sv), (long)(sv)->sv_debug_serial))
201 # define FREE_SV_DEBUG_FILE(sv)
202 # define DEBUG_SV_SERIAL(sv) NOOP
206 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
207 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
208 /* Whilst I'd love to do this, it seems that things like to check on
210 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
212 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
213 PoisonNew(&SvREFCNT(sv), 1, U32)
215 # define SvARENA_CHAIN(sv) SvANY(sv)
216 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
217 # define POSION_SV_HEAD(sv)
220 /* Mark an SV head as unused, and add to free list.
222 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
223 * its refcount artificially decremented during global destruction, so
224 * there may be dangling pointers to it. The last thing we want in that
225 * case is for it to be reused. */
227 #define plant_SV(p) \
229 const U32 old_flags = SvFLAGS(p); \
230 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
231 DEBUG_SV_SERIAL(p); \
232 FREE_SV_DEBUG_FILE(p); \
234 SvFLAGS(p) = SVTYPEMASK; \
235 if (!(old_flags & SVf_BREAK)) { \
236 SvARENA_CHAIN_SET(p, PL_sv_root); \
242 #define uproot_SV(p) \
245 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
250 /* make some more SVs by adding another arena */
257 char *chunk; /* must use New here to match call to */
258 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
259 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
264 /* new_SV(): return a new, empty SV head */
266 #ifdef DEBUG_LEAKING_SCALARS
267 /* provide a real function for a debugger to play with */
269 S_new_SV(pTHX_ const char *file, int line, const char *func)
276 sv = S_more_sv(aTHX);
280 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
281 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
287 sv->sv_debug_inpad = 0;
288 sv->sv_debug_parent = NULL;
289 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
291 sv->sv_debug_serial = PL_sv_serial++;
293 MEM_LOG_NEW_SV(sv, file, line, func);
294 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
295 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
299 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
307 (p) = S_more_sv(aTHX); \
311 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
316 /* del_SV(): return an empty SV head to the free list */
329 S_del_sv(pTHX_ SV *p)
333 PERL_ARGS_ASSERT_DEL_SV;
338 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
339 const SV * const sv = sva + 1;
340 const SV * const svend = &sva[SvREFCNT(sva)];
341 if (p >= sv && p < svend) {
347 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
348 "Attempt to free non-arena SV: 0x%"UVxf
349 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
356 #else /* ! DEBUGGING */
358 #define del_SV(p) plant_SV(p)
360 #endif /* DEBUGGING */
364 =head1 SV Manipulation Functions
366 =for apidoc sv_add_arena
368 Given a chunk of memory, link it to the head of the list of arenas,
369 and split it into a list of free SVs.
375 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
378 SV *const sva = MUTABLE_SV(ptr);
382 PERL_ARGS_ASSERT_SV_ADD_ARENA;
384 /* The first SV in an arena isn't an SV. */
385 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
386 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
387 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
389 PL_sv_arenaroot = sva;
390 PL_sv_root = sva + 1;
392 svend = &sva[SvREFCNT(sva) - 1];
395 SvARENA_CHAIN_SET(sv, (sv + 1));
399 /* Must always set typemask because it's always checked in on cleanup
400 when the arenas are walked looking for objects. */
401 SvFLAGS(sv) = SVTYPEMASK;
404 SvARENA_CHAIN_SET(sv, 0);
408 SvFLAGS(sv) = SVTYPEMASK;
411 /* visit(): call the named function for each non-free SV in the arenas
412 * whose flags field matches the flags/mask args. */
415 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
421 PERL_ARGS_ASSERT_VISIT;
423 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
424 const SV * const svend = &sva[SvREFCNT(sva)];
426 for (sv = sva + 1; sv < svend; ++sv) {
427 if (SvTYPE(sv) != (svtype)SVTYPEMASK
428 && (sv->sv_flags & mask) == flags
441 /* called by sv_report_used() for each live SV */
444 do_report_used(pTHX_ SV *const sv)
446 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
447 PerlIO_printf(Perl_debug_log, "****\n");
454 =for apidoc sv_report_used
456 Dump the contents of all SVs not yet freed (debugging aid).
462 Perl_sv_report_used(pTHX)
465 visit(do_report_used, 0, 0);
471 /* called by sv_clean_objs() for each live SV */
474 do_clean_objs(pTHX_ SV *const ref)
479 SV * const target = SvRV(ref);
480 if (SvOBJECT(target)) {
481 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
482 if (SvWEAKREF(ref)) {
483 sv_del_backref(target, ref);
489 SvREFCNT_dec_NN(target);
496 /* clear any slots in a GV which hold objects - except IO;
497 * called by sv_clean_objs() for each live GV */
500 do_clean_named_objs(pTHX_ SV *const sv)
504 assert(SvTYPE(sv) == SVt_PVGV);
505 assert(isGV_with_GP(sv));
509 /* freeing GP entries may indirectly free the current GV;
510 * hold onto it while we mess with the GP slots */
513 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
514 DEBUG_D((PerlIO_printf(Perl_debug_log,
515 "Cleaning named glob SV object:\n "), sv_dump(obj)));
517 SvREFCNT_dec_NN(obj);
519 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
520 DEBUG_D((PerlIO_printf(Perl_debug_log,
521 "Cleaning named glob AV object:\n "), sv_dump(obj)));
523 SvREFCNT_dec_NN(obj);
525 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log,
527 "Cleaning named glob HV object:\n "), sv_dump(obj)));
529 SvREFCNT_dec_NN(obj);
531 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
532 DEBUG_D((PerlIO_printf(Perl_debug_log,
533 "Cleaning named glob CV object:\n "), sv_dump(obj)));
535 SvREFCNT_dec_NN(obj);
537 SvREFCNT_dec_NN(sv); /* undo the inc above */
540 /* clear any IO slots in a GV which hold objects (except stderr, defout);
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_io_objs(pTHX_ SV *const sv)
548 assert(SvTYPE(sv) == SVt_PVGV);
549 assert(isGV_with_GP(sv));
550 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
554 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
555 DEBUG_D((PerlIO_printf(Perl_debug_log,
556 "Cleaning named glob IO object:\n "), sv_dump(obj)));
558 SvREFCNT_dec_NN(obj);
560 SvREFCNT_dec_NN(sv); /* undo the inc above */
563 /* Void wrapper to pass to visit() */
565 do_curse(pTHX_ SV * const sv) {
566 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
567 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
573 =for apidoc sv_clean_objs
575 Attempt to destroy all objects not yet freed.
581 Perl_sv_clean_objs(pTHX)
585 PL_in_clean_objs = TRUE;
586 visit(do_clean_objs, SVf_ROK, SVf_ROK);
587 /* Some barnacles may yet remain, clinging to typeglobs.
588 * Run the non-IO destructors first: they may want to output
589 * error messages, close files etc */
590 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
591 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
592 /* And if there are some very tenacious barnacles clinging to arrays,
593 closures, or what have you.... */
594 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
595 olddef = PL_defoutgv;
596 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
597 if (olddef && isGV_with_GP(olddef))
598 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
599 olderr = PL_stderrgv;
600 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
601 if (olderr && isGV_with_GP(olderr))
602 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
603 SvREFCNT_dec(olddef);
604 PL_in_clean_objs = FALSE;
607 /* called by sv_clean_all() for each live SV */
610 do_clean_all(pTHX_ SV *const sv)
613 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
614 /* don't clean pid table and strtab */
617 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
618 SvFLAGS(sv) |= SVf_BREAK;
623 =for apidoc sv_clean_all
625 Decrement the refcnt of each remaining SV, possibly triggering a
626 cleanup. This function may have to be called multiple times to free
627 SVs which are in complex self-referential hierarchies.
633 Perl_sv_clean_all(pTHX)
637 PL_in_clean_all = TRUE;
638 cleaned = visit(do_clean_all, 0,0);
643 ARENASETS: a meta-arena implementation which separates arena-info
644 into struct arena_set, which contains an array of struct
645 arena_descs, each holding info for a single arena. By separating
646 the meta-info from the arena, we recover the 1st slot, formerly
647 borrowed for list management. The arena_set is about the size of an
648 arena, avoiding the needless malloc overhead of a naive linked-list.
650 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
651 memory in the last arena-set (1/2 on average). In trade, we get
652 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
653 smaller types). The recovery of the wasted space allows use of
654 small arenas for large, rare body types, by changing array* fields
655 in body_details_by_type[] below.
658 char *arena; /* the raw storage, allocated aligned */
659 size_t size; /* its size ~4k typ */
660 svtype utype; /* bodytype stored in arena */
665 /* Get the maximum number of elements in set[] such that struct arena_set
666 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
667 therefore likely to be 1 aligned memory page. */
669 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
670 - 2 * sizeof(int)) / sizeof (struct arena_desc))
673 struct arena_set* next;
674 unsigned int set_size; /* ie ARENAS_PER_SET */
675 unsigned int curr; /* index of next available arena-desc */
676 struct arena_desc set[ARENAS_PER_SET];
680 =for apidoc sv_free_arenas
682 Deallocate the memory used by all arenas. Note that all the individual SV
683 heads and bodies within the arenas must already have been freed.
688 Perl_sv_free_arenas(pTHX)
695 /* Free arenas here, but be careful about fake ones. (We assume
696 contiguity of the fake ones with the corresponding real ones.) */
698 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
699 svanext = MUTABLE_SV(SvANY(sva));
700 while (svanext && SvFAKE(svanext))
701 svanext = MUTABLE_SV(SvANY(svanext));
708 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
711 struct arena_set *current = aroot;
714 assert(aroot->set[i].arena);
715 Safefree(aroot->set[i].arena);
723 i = PERL_ARENA_ROOTS_SIZE;
725 PL_body_roots[i] = 0;
732 Here are mid-level routines that manage the allocation of bodies out
733 of the various arenas. There are 5 kinds of arenas:
735 1. SV-head arenas, which are discussed and handled above
736 2. regular body arenas
737 3. arenas for reduced-size bodies
740 Arena types 2 & 3 are chained by body-type off an array of
741 arena-root pointers, which is indexed by svtype. Some of the
742 larger/less used body types are malloced singly, since a large
743 unused block of them is wasteful. Also, several svtypes dont have
744 bodies; the data fits into the sv-head itself. The arena-root
745 pointer thus has a few unused root-pointers (which may be hijacked
746 later for arena types 4,5)
748 3 differs from 2 as an optimization; some body types have several
749 unused fields in the front of the structure (which are kept in-place
750 for consistency). These bodies can be allocated in smaller chunks,
751 because the leading fields arent accessed. Pointers to such bodies
752 are decremented to point at the unused 'ghost' memory, knowing that
753 the pointers are used with offsets to the real memory.
756 =head1 SV-Body Allocation
758 Allocation of SV-bodies is similar to SV-heads, differing as follows;
759 the allocation mechanism is used for many body types, so is somewhat
760 more complicated, it uses arena-sets, and has no need for still-live
763 At the outermost level, (new|del)_X*V macros return bodies of the
764 appropriate type. These macros call either (new|del)_body_type or
765 (new|del)_body_allocated macro pairs, depending on specifics of the
766 type. Most body types use the former pair, the latter pair is used to
767 allocate body types with "ghost fields".
769 "ghost fields" are fields that are unused in certain types, and
770 consequently don't need to actually exist. They are declared because
771 they're part of a "base type", which allows use of functions as
772 methods. The simplest examples are AVs and HVs, 2 aggregate types
773 which don't use the fields which support SCALAR semantics.
775 For these types, the arenas are carved up into appropriately sized
776 chunks, we thus avoid wasted memory for those unaccessed members.
777 When bodies are allocated, we adjust the pointer back in memory by the
778 size of the part not allocated, so it's as if we allocated the full
779 structure. (But things will all go boom if you write to the part that
780 is "not there", because you'll be overwriting the last members of the
781 preceding structure in memory.)
783 We calculate the correction using the STRUCT_OFFSET macro on the first
784 member present. If the allocated structure is smaller (no initial NV
785 actually allocated) then the net effect is to subtract the size of the NV
786 from the pointer, to return a new pointer as if an initial NV were actually
787 allocated. (We were using structures named *_allocated for this, but
788 this turned out to be a subtle bug, because a structure without an NV
789 could have a lower alignment constraint, but the compiler is allowed to
790 optimised accesses based on the alignment constraint of the actual pointer
791 to the full structure, for example, using a single 64 bit load instruction
792 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
794 This is the same trick as was used for NV and IV bodies. Ironically it
795 doesn't need to be used for NV bodies any more, because NV is now at
796 the start of the structure. IV bodies don't need it either, because
797 they are no longer allocated.
799 In turn, the new_body_* allocators call S_new_body(), which invokes
800 new_body_inline macro, which takes a lock, and takes a body off the
801 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
802 necessary to refresh an empty list. Then the lock is released, and
803 the body is returned.
805 Perl_more_bodies allocates a new arena, and carves it up into an array of N
806 bodies, which it strings into a linked list. It looks up arena-size
807 and body-size from the body_details table described below, thus
808 supporting the multiple body-types.
810 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
811 the (new|del)_X*V macros are mapped directly to malloc/free.
813 For each sv-type, struct body_details bodies_by_type[] carries
814 parameters which control these aspects of SV handling:
816 Arena_size determines whether arenas are used for this body type, and if
817 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
818 zero, forcing individual mallocs and frees.
820 Body_size determines how big a body is, and therefore how many fit into
821 each arena. Offset carries the body-pointer adjustment needed for
822 "ghost fields", and is used in *_allocated macros.
824 But its main purpose is to parameterize info needed in
825 Perl_sv_upgrade(). The info here dramatically simplifies the function
826 vs the implementation in 5.8.8, making it table-driven. All fields
827 are used for this, except for arena_size.
829 For the sv-types that have no bodies, arenas are not used, so those
830 PL_body_roots[sv_type] are unused, and can be overloaded. In
831 something of a special case, SVt_NULL is borrowed for HE arenas;
832 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
833 bodies_by_type[SVt_NULL] slot is not used, as the table is not
838 struct body_details {
839 U8 body_size; /* Size to allocate */
840 U8 copy; /* Size of structure to copy (may be shorter) */
842 unsigned int type : 4; /* We have space for a sanity check. */
843 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
844 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
845 unsigned int arena : 1; /* Allocated from an arena */
846 size_t arena_size; /* Size of arena to allocate */
854 /* With -DPURFIY we allocate everything directly, and don't use arenas.
855 This seems a rather elegant way to simplify some of the code below. */
856 #define HASARENA FALSE
858 #define HASARENA TRUE
860 #define NOARENA FALSE
862 /* Size the arenas to exactly fit a given number of bodies. A count
863 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
864 simplifying the default. If count > 0, the arena is sized to fit
865 only that many bodies, allowing arenas to be used for large, rare
866 bodies (XPVFM, XPVIO) without undue waste. The arena size is
867 limited by PERL_ARENA_SIZE, so we can safely oversize the
870 #define FIT_ARENA0(body_size) \
871 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
872 #define FIT_ARENAn(count,body_size) \
873 ( count * body_size <= PERL_ARENA_SIZE) \
874 ? count * body_size \
875 : FIT_ARENA0 (body_size)
876 #define FIT_ARENA(count,body_size) \
878 ? FIT_ARENAn (count, body_size) \
879 : FIT_ARENA0 (body_size)
881 /* Calculate the length to copy. Specifically work out the length less any
882 final padding the compiler needed to add. See the comment in sv_upgrade
883 for why copying the padding proved to be a bug. */
885 #define copy_length(type, last_member) \
886 STRUCT_OFFSET(type, last_member) \
887 + sizeof (((type*)SvANY((const SV *)0))->last_member)
889 static const struct body_details bodies_by_type[] = {
890 /* HEs use this offset for their arena. */
891 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
893 /* IVs are in the head, so the allocation size is 0. */
895 sizeof(IV), /* This is used to copy out the IV body. */
896 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
897 NOARENA /* IVS don't need an arena */, 0
900 { sizeof(NV), sizeof(NV),
901 STRUCT_OFFSET(XPVNV, xnv_u),
902 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
904 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
905 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
906 + STRUCT_OFFSET(XPV, xpv_cur),
907 SVt_PV, FALSE, NONV, HASARENA,
908 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
910 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
911 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
912 + STRUCT_OFFSET(XPV, xpv_cur),
913 SVt_INVLIST, TRUE, NONV, HASARENA,
914 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
916 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
917 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
918 + STRUCT_OFFSET(XPV, xpv_cur),
919 SVt_PVIV, FALSE, NONV, HASARENA,
920 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
922 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
923 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
924 + STRUCT_OFFSET(XPV, xpv_cur),
925 SVt_PVNV, FALSE, HADNV, HASARENA,
926 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
928 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
929 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
934 SVt_REGEXP, TRUE, NONV, HASARENA,
935 FIT_ARENA(0, sizeof(regexp))
938 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
939 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
941 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
942 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
945 copy_length(XPVAV, xav_alloc),
947 SVt_PVAV, TRUE, NONV, HASARENA,
948 FIT_ARENA(0, sizeof(XPVAV)) },
951 copy_length(XPVHV, xhv_max),
953 SVt_PVHV, TRUE, NONV, HASARENA,
954 FIT_ARENA(0, sizeof(XPVHV)) },
959 SVt_PVCV, TRUE, NONV, HASARENA,
960 FIT_ARENA(0, sizeof(XPVCV)) },
965 SVt_PVFM, TRUE, NONV, NOARENA,
966 FIT_ARENA(20, sizeof(XPVFM)) },
971 SVt_PVIO, TRUE, NONV, HASARENA,
972 FIT_ARENA(24, sizeof(XPVIO)) },
975 #define new_body_allocated(sv_type) \
976 (void *)((char *)S_new_body(aTHX_ sv_type) \
977 - bodies_by_type[sv_type].offset)
979 /* return a thing to the free list */
981 #define del_body(thing, root) \
983 void ** const thing_copy = (void **)thing; \
984 *thing_copy = *root; \
985 *root = (void*)thing_copy; \
990 #define new_XNV() safemalloc(sizeof(XPVNV))
991 #define new_XPVNV() safemalloc(sizeof(XPVNV))
992 #define new_XPVMG() safemalloc(sizeof(XPVMG))
994 #define del_XPVGV(p) safefree(p)
998 #define new_XNV() new_body_allocated(SVt_NV)
999 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1000 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1002 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1003 &PL_body_roots[SVt_PVGV])
1007 /* no arena for you! */
1009 #define new_NOARENA(details) \
1010 safemalloc((details)->body_size + (details)->offset)
1011 #define new_NOARENAZ(details) \
1012 safecalloc((details)->body_size + (details)->offset, 1)
1015 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1016 const size_t arena_size)
1019 void ** const root = &PL_body_roots[sv_type];
1020 struct arena_desc *adesc;
1021 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1025 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1026 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1027 static bool done_sanity_check;
1029 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1030 * variables like done_sanity_check. */
1031 if (!done_sanity_check) {
1032 unsigned int i = SVt_LAST;
1034 done_sanity_check = TRUE;
1037 assert (bodies_by_type[i].type == i);
1043 /* may need new arena-set to hold new arena */
1044 if (!aroot || aroot->curr >= aroot->set_size) {
1045 struct arena_set *newroot;
1046 Newxz(newroot, 1, struct arena_set);
1047 newroot->set_size = ARENAS_PER_SET;
1048 newroot->next = aroot;
1050 PL_body_arenas = (void *) newroot;
1051 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1054 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1055 curr = aroot->curr++;
1056 adesc = &(aroot->set[curr]);
1057 assert(!adesc->arena);
1059 Newx(adesc->arena, good_arena_size, char);
1060 adesc->size = good_arena_size;
1061 adesc->utype = sv_type;
1062 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1063 curr, (void*)adesc->arena, (UV)good_arena_size));
1065 start = (char *) adesc->arena;
1067 /* Get the address of the byte after the end of the last body we can fit.
1068 Remember, this is integer division: */
1069 end = start + good_arena_size / body_size * body_size;
1071 /* computed count doesn't reflect the 1st slot reservation */
1072 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1073 DEBUG_m(PerlIO_printf(Perl_debug_log,
1074 "arena %p end %p arena-size %d (from %d) type %d "
1076 (void*)start, (void*)end, (int)good_arena_size,
1077 (int)arena_size, sv_type, (int)body_size,
1078 (int)good_arena_size / (int)body_size));
1080 DEBUG_m(PerlIO_printf(Perl_debug_log,
1081 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1082 (void*)start, (void*)end,
1083 (int)arena_size, sv_type, (int)body_size,
1084 (int)good_arena_size / (int)body_size));
1086 *root = (void *)start;
1089 /* Where the next body would start: */
1090 char * const next = start + body_size;
1093 /* This is the last body: */
1094 assert(next == end);
1096 *(void **)start = 0;
1100 *(void**) start = (void *)next;
1105 /* grab a new thing from the free list, allocating more if necessary.
1106 The inline version is used for speed in hot routines, and the
1107 function using it serves the rest (unless PURIFY).
1109 #define new_body_inline(xpv, sv_type) \
1111 void ** const r3wt = &PL_body_roots[sv_type]; \
1112 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1113 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1114 bodies_by_type[sv_type].body_size,\
1115 bodies_by_type[sv_type].arena_size)); \
1116 *(r3wt) = *(void**)(xpv); \
1122 S_new_body(pTHX_ const svtype sv_type)
1126 new_body_inline(xpv, sv_type);
1132 static const struct body_details fake_rv =
1133 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1136 =for apidoc sv_upgrade
1138 Upgrade an SV to a more complex form. Generally adds a new body type to the
1139 SV, then copies across as much information as possible from the old body.
1140 It croaks if the SV is already in a more complex form than requested. You
1141 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1142 before calling C<sv_upgrade>, and hence does not croak. See also
1149 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1154 const svtype old_type = SvTYPE(sv);
1155 const struct body_details *new_type_details;
1156 const struct body_details *old_type_details
1157 = bodies_by_type + old_type;
1158 SV *referant = NULL;
1160 PERL_ARGS_ASSERT_SV_UPGRADE;
1162 if (old_type == new_type)
1165 /* This clause was purposefully added ahead of the early return above to
1166 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1167 inference by Nick I-S that it would fix other troublesome cases. See
1168 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1170 Given that shared hash key scalars are no longer PVIV, but PV, there is
1171 no longer need to unshare so as to free up the IVX slot for its proper
1172 purpose. So it's safe to move the early return earlier. */
1174 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1175 sv_force_normal_flags(sv, 0);
1178 old_body = SvANY(sv);
1180 /* Copying structures onto other structures that have been neatly zeroed
1181 has a subtle gotcha. Consider XPVMG
1183 +------+------+------+------+------+-------+-------+
1184 | NV | CUR | LEN | IV | MAGIC | STASH |
1185 +------+------+------+------+------+-------+-------+
1186 0 4 8 12 16 20 24 28
1188 where NVs are aligned to 8 bytes, so that sizeof that structure is
1189 actually 32 bytes long, with 4 bytes of padding at the end:
1191 +------+------+------+------+------+-------+-------+------+
1192 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1193 +------+------+------+------+------+-------+-------+------+
1194 0 4 8 12 16 20 24 28 32
1196 so what happens if you allocate memory for this structure:
1198 +------+------+------+------+------+-------+-------+------+------+...
1199 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1200 +------+------+------+------+------+-------+-------+------+------+...
1201 0 4 8 12 16 20 24 28 32 36
1203 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1204 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1205 started out as zero once, but it's quite possible that it isn't. So now,
1206 rather than a nicely zeroed GP, you have it pointing somewhere random.
1209 (In fact, GP ends up pointing at a previous GP structure, because the
1210 principle cause of the padding in XPVMG getting garbage is a copy of
1211 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1212 this happens to be moot because XPVGV has been re-ordered, with GP
1213 no longer after STASH)
1215 So we are careful and work out the size of used parts of all the
1223 referant = SvRV(sv);
1224 old_type_details = &fake_rv;
1225 if (new_type == SVt_NV)
1226 new_type = SVt_PVNV;
1228 if (new_type < SVt_PVIV) {
1229 new_type = (new_type == SVt_NV)
1230 ? SVt_PVNV : SVt_PVIV;
1235 if (new_type < SVt_PVNV) {
1236 new_type = SVt_PVNV;
1240 assert(new_type > SVt_PV);
1241 assert(SVt_IV < SVt_PV);
1242 assert(SVt_NV < SVt_PV);
1249 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1250 there's no way that it can be safely upgraded, because perl.c
1251 expects to Safefree(SvANY(PL_mess_sv)) */
1252 assert(sv != PL_mess_sv);
1253 /* This flag bit is used to mean other things in other scalar types.
1254 Given that it only has meaning inside the pad, it shouldn't be set
1255 on anything that can get upgraded. */
1256 assert(!SvPAD_TYPED(sv));
1259 if (UNLIKELY(old_type_details->cant_upgrade))
1260 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1261 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1264 if (UNLIKELY(old_type > new_type))
1265 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1266 (int)old_type, (int)new_type);
1268 new_type_details = bodies_by_type + new_type;
1270 SvFLAGS(sv) &= ~SVTYPEMASK;
1271 SvFLAGS(sv) |= new_type;
1273 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1274 the return statements above will have triggered. */
1275 assert (new_type != SVt_NULL);
1278 assert(old_type == SVt_NULL);
1279 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1283 assert(old_type == SVt_NULL);
1284 SvANY(sv) = new_XNV();
1289 assert(new_type_details->body_size);
1292 assert(new_type_details->arena);
1293 assert(new_type_details->arena_size);
1294 /* This points to the start of the allocated area. */
1295 new_body_inline(new_body, new_type);
1296 Zero(new_body, new_type_details->body_size, char);
1297 new_body = ((char *)new_body) - new_type_details->offset;
1299 /* We always allocated the full length item with PURIFY. To do this
1300 we fake things so that arena is false for all 16 types.. */
1301 new_body = new_NOARENAZ(new_type_details);
1303 SvANY(sv) = new_body;
1304 if (new_type == SVt_PVAV) {
1308 if (old_type_details->body_size) {
1311 /* It will have been zeroed when the new body was allocated.
1312 Lets not write to it, in case it confuses a write-back
1318 #ifndef NODEFAULT_SHAREKEYS
1319 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1321 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1322 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1325 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1326 The target created by newSVrv also is, and it can have magic.
1327 However, it never has SvPVX set.
1329 if (old_type == SVt_IV) {
1331 } else if (old_type >= SVt_PV) {
1332 assert(SvPVX_const(sv) == 0);
1335 if (old_type >= SVt_PVMG) {
1336 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1337 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1339 sv->sv_u.svu_array = NULL; /* or svu_hash */
1344 /* XXX Is this still needed? Was it ever needed? Surely as there is
1345 no route from NV to PVIV, NOK can never be true */
1346 assert(!SvNOKp(sv));
1359 assert(new_type_details->body_size);
1360 /* We always allocated the full length item with PURIFY. To do this
1361 we fake things so that arena is false for all 16 types.. */
1362 if(new_type_details->arena) {
1363 /* This points to the start of the allocated area. */
1364 new_body_inline(new_body, new_type);
1365 Zero(new_body, new_type_details->body_size, char);
1366 new_body = ((char *)new_body) - new_type_details->offset;
1368 new_body = new_NOARENAZ(new_type_details);
1370 SvANY(sv) = new_body;
1372 if (old_type_details->copy) {
1373 /* There is now the potential for an upgrade from something without
1374 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1375 int offset = old_type_details->offset;
1376 int length = old_type_details->copy;
1378 if (new_type_details->offset > old_type_details->offset) {
1379 const int difference
1380 = new_type_details->offset - old_type_details->offset;
1381 offset += difference;
1382 length -= difference;
1384 assert (length >= 0);
1386 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1390 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1391 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1392 * correct 0.0 for us. Otherwise, if the old body didn't have an
1393 * NV slot, but the new one does, then we need to initialise the
1394 * freshly created NV slot with whatever the correct bit pattern is
1396 if (old_type_details->zero_nv && !new_type_details->zero_nv
1397 && !isGV_with_GP(sv))
1401 if (UNLIKELY(new_type == SVt_PVIO)) {
1402 IO * const io = MUTABLE_IO(sv);
1403 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1406 /* Clear the stashcache because a new IO could overrule a package
1408 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1409 hv_clear(PL_stashcache);
1411 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1412 IoPAGE_LEN(sv) = 60;
1414 if (UNLIKELY(new_type == SVt_REGEXP))
1415 sv->sv_u.svu_rx = (regexp *)new_body;
1416 else if (old_type < SVt_PV) {
1417 /* referant will be NULL unless the old type was SVt_IV emulating
1419 sv->sv_u.svu_rv = referant;
1423 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1424 (unsigned long)new_type);
1427 if (old_type > SVt_IV) {
1431 /* Note that there is an assumption that all bodies of types that
1432 can be upgraded came from arenas. Only the more complex non-
1433 upgradable types are allowed to be directly malloc()ed. */
1434 assert(old_type_details->arena);
1435 del_body((void*)((char*)old_body + old_type_details->offset),
1436 &PL_body_roots[old_type]);
1442 =for apidoc sv_backoff
1444 Remove any string offset. You should normally use the C<SvOOK_off> macro
1451 Perl_sv_backoff(pTHX_ SV *const sv)
1454 const char * const s = SvPVX_const(sv);
1456 PERL_ARGS_ASSERT_SV_BACKOFF;
1457 PERL_UNUSED_CONTEXT;
1460 assert(SvTYPE(sv) != SVt_PVHV);
1461 assert(SvTYPE(sv) != SVt_PVAV);
1463 SvOOK_offset(sv, delta);
1465 SvLEN_set(sv, SvLEN(sv) + delta);
1466 SvPV_set(sv, SvPVX(sv) - delta);
1467 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1468 SvFLAGS(sv) &= ~SVf_OOK;
1475 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1476 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1477 Use the C<SvGROW> wrapper instead.
1482 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1485 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1489 PERL_ARGS_ASSERT_SV_GROW;
1493 if (SvTYPE(sv) < SVt_PV) {
1494 sv_upgrade(sv, SVt_PV);
1495 s = SvPVX_mutable(sv);
1497 else if (SvOOK(sv)) { /* pv is offset? */
1499 s = SvPVX_mutable(sv);
1500 if (newlen > SvLEN(sv))
1501 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1505 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1506 s = SvPVX_mutable(sv);
1509 #ifdef PERL_NEW_COPY_ON_WRITE
1510 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1511 * to store the COW count. So in general, allocate one more byte than
1512 * asked for, to make it likely this byte is always spare: and thus
1513 * make more strings COW-able.
1514 * If the new size is a big power of two, don't bother: we assume the
1515 * caller wanted a nice 2^N sized block and will be annoyed at getting
1521 if (newlen > SvLEN(sv)) { /* need more room? */
1522 STRLEN minlen = SvCUR(sv);
1523 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1524 if (newlen < minlen)
1526 #ifndef Perl_safesysmalloc_size
1527 newlen = PERL_STRLEN_ROUNDUP(newlen);
1529 if (SvLEN(sv) && s) {
1530 s = (char*)saferealloc(s, newlen);
1533 s = (char*)safemalloc(newlen);
1534 if (SvPVX_const(sv) && SvCUR(sv)) {
1535 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1539 #ifdef Perl_safesysmalloc_size
1540 /* Do this here, do it once, do it right, and then we will never get
1541 called back into sv_grow() unless there really is some growing
1543 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1545 SvLEN_set(sv, newlen);
1552 =for apidoc sv_setiv
1554 Copies an integer into the given SV, upgrading first if necessary.
1555 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1561 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1565 PERL_ARGS_ASSERT_SV_SETIV;
1567 SV_CHECK_THINKFIRST_COW_DROP(sv);
1568 switch (SvTYPE(sv)) {
1571 sv_upgrade(sv, SVt_IV);
1574 sv_upgrade(sv, SVt_PVIV);
1578 if (!isGV_with_GP(sv))
1585 /* diag_listed_as: Can't coerce %s to %s in %s */
1586 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1590 (void)SvIOK_only(sv); /* validate number */
1596 =for apidoc sv_setiv_mg
1598 Like C<sv_setiv>, but also handles 'set' magic.
1604 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1606 PERL_ARGS_ASSERT_SV_SETIV_MG;
1613 =for apidoc sv_setuv
1615 Copies an unsigned integer into the given SV, upgrading first if necessary.
1616 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1622 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1624 PERL_ARGS_ASSERT_SV_SETUV;
1626 /* With the if statement to ensure that integers are stored as IVs whenever
1628 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1631 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1633 If you wish to remove the following if statement, so that this routine
1634 (and its callers) always return UVs, please benchmark to see what the
1635 effect is. Modern CPUs may be different. Or may not :-)
1637 if (u <= (UV)IV_MAX) {
1638 sv_setiv(sv, (IV)u);
1647 =for apidoc sv_setuv_mg
1649 Like C<sv_setuv>, but also handles 'set' magic.
1655 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1657 PERL_ARGS_ASSERT_SV_SETUV_MG;
1664 =for apidoc sv_setnv
1666 Copies a double into the given SV, upgrading first if necessary.
1667 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1673 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1677 PERL_ARGS_ASSERT_SV_SETNV;
1679 SV_CHECK_THINKFIRST_COW_DROP(sv);
1680 switch (SvTYPE(sv)) {
1683 sv_upgrade(sv, SVt_NV);
1687 sv_upgrade(sv, SVt_PVNV);
1691 if (!isGV_with_GP(sv))
1698 /* diag_listed_as: Can't coerce %s to %s in %s */
1699 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1704 (void)SvNOK_only(sv); /* validate number */
1709 =for apidoc sv_setnv_mg
1711 Like C<sv_setnv>, but also handles 'set' magic.
1717 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1719 PERL_ARGS_ASSERT_SV_SETNV_MG;
1725 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1726 * not incrementable warning display.
1727 * Originally part of S_not_a_number().
1728 * The return value may be != tmpbuf.
1732 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1735 PERL_ARGS_ASSERT_SV_DISPLAY;
1738 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1739 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1742 const char * const limit = tmpbuf + tmpbuf_size - 8;
1743 /* each *s can expand to 4 chars + "...\0",
1744 i.e. need room for 8 chars */
1746 const char *s = SvPVX_const(sv);
1747 const char * const end = s + SvCUR(sv);
1748 for ( ; s < end && d < limit; s++ ) {
1750 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1754 /* Map to ASCII "equivalent" of Latin1 */
1755 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1761 else if (ch == '\r') {
1765 else if (ch == '\f') {
1769 else if (ch == '\\') {
1773 else if (ch == '\0') {
1777 else if (isPRINT_LC(ch))
1796 /* Print an "isn't numeric" warning, using a cleaned-up,
1797 * printable version of the offending string
1801 S_not_a_number(pTHX_ SV *const sv)
1807 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1809 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1812 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1813 /* diag_listed_as: Argument "%s" isn't numeric%s */
1814 "Argument \"%s\" isn't numeric in %s", pv,
1817 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1818 /* diag_listed_as: Argument "%s" isn't numeric%s */
1819 "Argument \"%s\" isn't numeric", pv);
1823 S_not_incrementable(pTHX_ SV *const sv) {
1828 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1830 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1832 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1833 "Argument \"%s\" treated as 0 in increment (++)", pv);
1837 =for apidoc looks_like_number
1839 Test if the content of an SV looks like a number (or is a number).
1840 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1841 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1848 Perl_looks_like_number(pTHX_ SV *const sv)
1853 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1855 if (SvPOK(sv) || SvPOKp(sv)) {
1856 sbegin = SvPV_nomg_const(sv, len);
1859 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1860 return grok_number(sbegin, len, NULL);
1864 S_glob_2number(pTHX_ GV * const gv)
1866 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1868 /* We know that all GVs stringify to something that is not-a-number,
1869 so no need to test that. */
1870 if (ckWARN(WARN_NUMERIC))
1872 SV *const buffer = sv_newmortal();
1873 gv_efullname3(buffer, gv, "*");
1874 not_a_number(buffer);
1876 /* We just want something true to return, so that S_sv_2iuv_common
1877 can tail call us and return true. */
1881 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1882 until proven guilty, assume that things are not that bad... */
1887 As 64 bit platforms often have an NV that doesn't preserve all bits of
1888 an IV (an assumption perl has been based on to date) it becomes necessary
1889 to remove the assumption that the NV always carries enough precision to
1890 recreate the IV whenever needed, and that the NV is the canonical form.
1891 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1892 precision as a side effect of conversion (which would lead to insanity
1893 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1894 1) to distinguish between IV/UV/NV slots that have cached a valid
1895 conversion where precision was lost and IV/UV/NV slots that have a
1896 valid conversion which has lost no precision
1897 2) to ensure that if a numeric conversion to one form is requested that
1898 would lose precision, the precise conversion (or differently
1899 imprecise conversion) is also performed and cached, to prevent
1900 requests for different numeric formats on the same SV causing
1901 lossy conversion chains. (lossless conversion chains are perfectly
1906 SvIOKp is true if the IV slot contains a valid value
1907 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1908 SvNOKp is true if the NV slot contains a valid value
1909 SvNOK is true only if the NV value is accurate
1912 while converting from PV to NV, check to see if converting that NV to an
1913 IV(or UV) would lose accuracy over a direct conversion from PV to
1914 IV(or UV). If it would, cache both conversions, return NV, but mark
1915 SV as IOK NOKp (ie not NOK).
1917 While converting from PV to IV, check to see if converting that IV to an
1918 NV would lose accuracy over a direct conversion from PV to NV. If it
1919 would, cache both conversions, flag similarly.
1921 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1922 correctly because if IV & NV were set NV *always* overruled.
1923 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1924 changes - now IV and NV together means that the two are interchangeable:
1925 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1927 The benefit of this is that operations such as pp_add know that if
1928 SvIOK is true for both left and right operands, then integer addition
1929 can be used instead of floating point (for cases where the result won't
1930 overflow). Before, floating point was always used, which could lead to
1931 loss of precision compared with integer addition.
1933 * making IV and NV equal status should make maths accurate on 64 bit
1935 * may speed up maths somewhat if pp_add and friends start to use
1936 integers when possible instead of fp. (Hopefully the overhead in
1937 looking for SvIOK and checking for overflow will not outweigh the
1938 fp to integer speedup)
1939 * will slow down integer operations (callers of SvIV) on "inaccurate"
1940 values, as the change from SvIOK to SvIOKp will cause a call into
1941 sv_2iv each time rather than a macro access direct to the IV slot
1942 * should speed up number->string conversion on integers as IV is
1943 favoured when IV and NV are equally accurate
1945 ####################################################################
1946 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1947 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1948 On the other hand, SvUOK is true iff UV.
1949 ####################################################################
1951 Your mileage will vary depending your CPU's relative fp to integer
1955 #ifndef NV_PRESERVES_UV
1956 # define IS_NUMBER_UNDERFLOW_IV 1
1957 # define IS_NUMBER_UNDERFLOW_UV 2
1958 # define IS_NUMBER_IV_AND_UV 2
1959 # define IS_NUMBER_OVERFLOW_IV 4
1960 # define IS_NUMBER_OVERFLOW_UV 5
1962 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
1964 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1966 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
1974 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1976 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));
1977 if (SvNVX(sv) < (NV)IV_MIN) {
1978 (void)SvIOKp_on(sv);
1980 SvIV_set(sv, IV_MIN);
1981 return IS_NUMBER_UNDERFLOW_IV;
1983 if (SvNVX(sv) > (NV)UV_MAX) {
1984 (void)SvIOKp_on(sv);
1987 SvUV_set(sv, UV_MAX);
1988 return IS_NUMBER_OVERFLOW_UV;
1990 (void)SvIOKp_on(sv);
1992 /* Can't use strtol etc to convert this string. (See truth table in
1994 if (SvNVX(sv) <= (UV)IV_MAX) {
1995 SvIV_set(sv, I_V(SvNVX(sv)));
1996 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1997 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1999 /* Integer is imprecise. NOK, IOKp */
2001 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2004 SvUV_set(sv, U_V(SvNVX(sv)));
2005 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2006 if (SvUVX(sv) == UV_MAX) {
2007 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2008 possibly be preserved by NV. Hence, it must be overflow.
2010 return IS_NUMBER_OVERFLOW_UV;
2012 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2014 /* Integer is imprecise. NOK, IOKp */
2016 return IS_NUMBER_OVERFLOW_IV;
2018 #endif /* !NV_PRESERVES_UV*/
2021 S_sv_2iuv_common(pTHX_ SV *const sv)
2025 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2028 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2029 * without also getting a cached IV/UV from it at the same time
2030 * (ie PV->NV conversion should detect loss of accuracy and cache
2031 * IV or UV at same time to avoid this. */
2032 /* IV-over-UV optimisation - choose to cache IV if possible */
2034 if (SvTYPE(sv) == SVt_NV)
2035 sv_upgrade(sv, SVt_PVNV);
2037 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2038 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2039 certainly cast into the IV range at IV_MAX, whereas the correct
2040 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2042 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2043 if (Perl_isnan(SvNVX(sv))) {
2049 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2050 SvIV_set(sv, I_V(SvNVX(sv)));
2051 if (SvNVX(sv) == (NV) SvIVX(sv)
2052 #ifndef NV_PRESERVES_UV
2053 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2054 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2055 /* Don't flag it as "accurately an integer" if the number
2056 came from a (by definition imprecise) NV operation, and
2057 we're outside the range of NV integer precision */
2061 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2063 /* scalar has trailing garbage, eg "42a" */
2065 DEBUG_c(PerlIO_printf(Perl_debug_log,
2066 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2072 /* IV not precise. No need to convert from PV, as NV
2073 conversion would already have cached IV if it detected
2074 that PV->IV would be better than PV->NV->IV
2075 flags already correct - don't set public IOK. */
2076 DEBUG_c(PerlIO_printf(Perl_debug_log,
2077 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2082 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2083 but the cast (NV)IV_MIN rounds to a the value less (more
2084 negative) than IV_MIN which happens to be equal to SvNVX ??
2085 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2086 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2087 (NV)UVX == NVX are both true, but the values differ. :-(
2088 Hopefully for 2s complement IV_MIN is something like
2089 0x8000000000000000 which will be exact. NWC */
2092 SvUV_set(sv, U_V(SvNVX(sv)));
2094 (SvNVX(sv) == (NV) SvUVX(sv))
2095 #ifndef NV_PRESERVES_UV
2096 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2097 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2098 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2099 /* Don't flag it as "accurately an integer" if the number
2100 came from a (by definition imprecise) NV operation, and
2101 we're outside the range of NV integer precision */
2107 DEBUG_c(PerlIO_printf(Perl_debug_log,
2108 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2114 else if (SvPOKp(sv)) {
2116 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2117 /* We want to avoid a possible problem when we cache an IV/ a UV which
2118 may be later translated to an NV, and the resulting NV is not
2119 the same as the direct translation of the initial string
2120 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2121 be careful to ensure that the value with the .456 is around if the
2122 NV value is requested in the future).
2124 This means that if we cache such an IV/a UV, we need to cache the
2125 NV as well. Moreover, we trade speed for space, and do not
2126 cache the NV if we are sure it's not needed.
2129 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2130 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2131 == IS_NUMBER_IN_UV) {
2132 /* It's definitely an integer, only upgrade to PVIV */
2133 if (SvTYPE(sv) < SVt_PVIV)
2134 sv_upgrade(sv, SVt_PVIV);
2136 } else if (SvTYPE(sv) < SVt_PVNV)
2137 sv_upgrade(sv, SVt_PVNV);
2139 /* If NVs preserve UVs then we only use the UV value if we know that
2140 we aren't going to call atof() below. If NVs don't preserve UVs
2141 then the value returned may have more precision than atof() will
2142 return, even though value isn't perfectly accurate. */
2143 if ((numtype & (IS_NUMBER_IN_UV
2144 #ifdef NV_PRESERVES_UV
2147 )) == IS_NUMBER_IN_UV) {
2148 /* This won't turn off the public IOK flag if it was set above */
2149 (void)SvIOKp_on(sv);
2151 if (!(numtype & IS_NUMBER_NEG)) {
2153 if (value <= (UV)IV_MAX) {
2154 SvIV_set(sv, (IV)value);
2156 /* it didn't overflow, and it was positive. */
2157 SvUV_set(sv, value);
2161 /* 2s complement assumption */
2162 if (value <= (UV)IV_MIN) {
2163 SvIV_set(sv, -(IV)value);
2165 /* Too negative for an IV. This is a double upgrade, but
2166 I'm assuming it will be rare. */
2167 if (SvTYPE(sv) < SVt_PVNV)
2168 sv_upgrade(sv, SVt_PVNV);
2172 SvNV_set(sv, -(NV)value);
2173 SvIV_set(sv, IV_MIN);
2177 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2178 will be in the previous block to set the IV slot, and the next
2179 block to set the NV slot. So no else here. */
2181 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2182 != IS_NUMBER_IN_UV) {
2183 /* It wasn't an (integer that doesn't overflow the UV). */
2184 SvNV_set(sv, Atof(SvPVX_const(sv)));
2186 if (! numtype && ckWARN(WARN_NUMERIC))
2189 #if defined(USE_LONG_DOUBLE)
2190 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2191 PTR2UV(sv), SvNVX(sv)));
2193 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2194 PTR2UV(sv), SvNVX(sv)));
2197 #ifdef NV_PRESERVES_UV
2198 (void)SvIOKp_on(sv);
2200 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2201 SvIV_set(sv, I_V(SvNVX(sv)));
2202 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2205 NOOP; /* Integer is imprecise. NOK, IOKp */
2207 /* UV will not work better than IV */
2209 if (SvNVX(sv) > (NV)UV_MAX) {
2211 /* Integer is inaccurate. NOK, IOKp, is UV */
2212 SvUV_set(sv, UV_MAX);
2214 SvUV_set(sv, U_V(SvNVX(sv)));
2215 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2216 NV preservse UV so can do correct comparison. */
2217 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2220 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2225 #else /* NV_PRESERVES_UV */
2226 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2227 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2228 /* The IV/UV slot will have been set from value returned by
2229 grok_number above. The NV slot has just been set using
2232 assert (SvIOKp(sv));
2234 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2235 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2236 /* Small enough to preserve all bits. */
2237 (void)SvIOKp_on(sv);
2239 SvIV_set(sv, I_V(SvNVX(sv)));
2240 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2242 /* Assumption: first non-preserved integer is < IV_MAX,
2243 this NV is in the preserved range, therefore: */
2244 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2246 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);
2250 0 0 already failed to read UV.
2251 0 1 already failed to read UV.
2252 1 0 you won't get here in this case. IV/UV
2253 slot set, public IOK, Atof() unneeded.
2254 1 1 already read UV.
2255 so there's no point in sv_2iuv_non_preserve() attempting
2256 to use atol, strtol, strtoul etc. */
2258 sv_2iuv_non_preserve (sv, numtype);
2260 sv_2iuv_non_preserve (sv);
2264 #endif /* NV_PRESERVES_UV */
2265 /* It might be more code efficient to go through the entire logic above
2266 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2267 gets complex and potentially buggy, so more programmer efficient
2268 to do it this way, by turning off the public flags: */
2270 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2274 if (isGV_with_GP(sv))
2275 return glob_2number(MUTABLE_GV(sv));
2277 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2279 if (SvTYPE(sv) < SVt_IV)
2280 /* Typically the caller expects that sv_any is not NULL now. */
2281 sv_upgrade(sv, SVt_IV);
2282 /* Return 0 from the caller. */
2289 =for apidoc sv_2iv_flags
2291 Return the integer value of an SV, doing any necessary string
2292 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2293 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2299 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2303 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2305 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2306 && SvTYPE(sv) != SVt_PVFM);
2308 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2314 if (flags & SV_SKIP_OVERLOAD)
2316 tmpstr = AMG_CALLunary(sv, numer_amg);
2317 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2318 return SvIV(tmpstr);
2321 return PTR2IV(SvRV(sv));
2324 if (SvVALID(sv) || isREGEXP(sv)) {
2325 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2326 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2327 In practice they are extremely unlikely to actually get anywhere
2328 accessible by user Perl code - the only way that I'm aware of is when
2329 a constant subroutine which is used as the second argument to index.
2331 Regexps have no SvIVX and SvNVX fields.
2333 assert(isREGEXP(sv) || SvPOKp(sv));
2336 const char * const ptr =
2337 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2339 = grok_number(ptr, SvCUR(sv), &value);
2341 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2342 == IS_NUMBER_IN_UV) {
2343 /* It's definitely an integer */
2344 if (numtype & IS_NUMBER_NEG) {
2345 if (value < (UV)IV_MIN)
2348 if (value < (UV)IV_MAX)
2353 if (ckWARN(WARN_NUMERIC))
2356 return I_V(Atof(ptr));
2360 if (SvTHINKFIRST(sv)) {
2361 #ifdef PERL_OLD_COPY_ON_WRITE
2363 sv_force_normal_flags(sv, 0);
2366 if (SvREADONLY(sv) && !SvOK(sv)) {
2367 if (ckWARN(WARN_UNINITIALIZED))
2374 if (S_sv_2iuv_common(aTHX_ sv))
2378 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2379 PTR2UV(sv),SvIVX(sv)));
2380 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2384 =for apidoc sv_2uv_flags
2386 Return the unsigned integer value of an SV, doing any necessary string
2387 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2388 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2394 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2398 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2400 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2406 if (flags & SV_SKIP_OVERLOAD)
2408 tmpstr = AMG_CALLunary(sv, numer_amg);
2409 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2410 return SvUV(tmpstr);
2413 return PTR2UV(SvRV(sv));
2416 if (SvVALID(sv) || isREGEXP(sv)) {
2417 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2418 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2419 Regexps have no SvIVX and SvNVX fields. */
2420 assert(isREGEXP(sv) || SvPOKp(sv));
2423 const char * const ptr =
2424 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2426 = grok_number(ptr, SvCUR(sv), &value);
2428 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2429 == IS_NUMBER_IN_UV) {
2430 /* It's definitely an integer */
2431 if (!(numtype & IS_NUMBER_NEG))
2435 if (ckWARN(WARN_NUMERIC))
2438 return U_V(Atof(ptr));
2442 if (SvTHINKFIRST(sv)) {
2443 #ifdef PERL_OLD_COPY_ON_WRITE
2445 sv_force_normal_flags(sv, 0);
2448 if (SvREADONLY(sv) && !SvOK(sv)) {
2449 if (ckWARN(WARN_UNINITIALIZED))
2456 if (S_sv_2iuv_common(aTHX_ sv))
2460 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2461 PTR2UV(sv),SvUVX(sv)));
2462 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2466 =for apidoc sv_2nv_flags
2468 Return the num value of an SV, doing any necessary string or integer
2469 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2470 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2476 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2480 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2482 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2483 && SvTYPE(sv) != SVt_PVFM);
2484 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2485 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2486 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2487 Regexps have no SvIVX and SvNVX fields. */
2489 if (flags & SV_GMAGIC)
2493 if (SvPOKp(sv) && !SvIOKp(sv)) {
2494 ptr = SvPVX_const(sv);
2496 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2497 !grok_number(ptr, SvCUR(sv), NULL))
2503 return (NV)SvUVX(sv);
2505 return (NV)SvIVX(sv);
2511 ptr = RX_WRAPPED((REGEXP *)sv);
2514 assert(SvTYPE(sv) >= SVt_PVMG);
2515 /* This falls through to the report_uninit near the end of the
2517 } else if (SvTHINKFIRST(sv)) {
2522 if (flags & SV_SKIP_OVERLOAD)
2524 tmpstr = AMG_CALLunary(sv, numer_amg);
2525 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2526 return SvNV(tmpstr);
2529 return PTR2NV(SvRV(sv));
2531 #ifdef PERL_OLD_COPY_ON_WRITE
2533 sv_force_normal_flags(sv, 0);
2536 if (SvREADONLY(sv) && !SvOK(sv)) {
2537 if (ckWARN(WARN_UNINITIALIZED))
2542 if (SvTYPE(sv) < SVt_NV) {
2543 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2544 sv_upgrade(sv, SVt_NV);
2545 #ifdef USE_LONG_DOUBLE
2547 STORE_NUMERIC_LOCAL_SET_STANDARD();
2548 PerlIO_printf(Perl_debug_log,
2549 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2550 PTR2UV(sv), SvNVX(sv));
2551 RESTORE_NUMERIC_LOCAL();
2555 STORE_NUMERIC_LOCAL_SET_STANDARD();
2556 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2557 PTR2UV(sv), SvNVX(sv));
2558 RESTORE_NUMERIC_LOCAL();
2562 else if (SvTYPE(sv) < SVt_PVNV)
2563 sv_upgrade(sv, SVt_PVNV);
2568 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2569 #ifdef NV_PRESERVES_UV
2575 /* Only set the public NV OK flag if this NV preserves the IV */
2576 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2578 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2579 : (SvIVX(sv) == I_V(SvNVX(sv))))
2585 else if (SvPOKp(sv)) {
2587 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2588 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2590 #ifdef NV_PRESERVES_UV
2591 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2592 == IS_NUMBER_IN_UV) {
2593 /* It's definitely an integer */
2594 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2596 SvNV_set(sv, Atof(SvPVX_const(sv)));
2602 SvNV_set(sv, Atof(SvPVX_const(sv)));
2603 /* Only set the public NV OK flag if this NV preserves the value in
2604 the PV at least as well as an IV/UV would.
2605 Not sure how to do this 100% reliably. */
2606 /* if that shift count is out of range then Configure's test is
2607 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2609 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2610 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2611 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2612 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2613 /* Can't use strtol etc to convert this string, so don't try.
2614 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2617 /* value has been set. It may not be precise. */
2618 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2619 /* 2s complement assumption for (UV)IV_MIN */
2620 SvNOK_on(sv); /* Integer is too negative. */
2625 if (numtype & IS_NUMBER_NEG) {
2626 SvIV_set(sv, -(IV)value);
2627 } else if (value <= (UV)IV_MAX) {
2628 SvIV_set(sv, (IV)value);
2630 SvUV_set(sv, value);
2634 if (numtype & IS_NUMBER_NOT_INT) {
2635 /* I believe that even if the original PV had decimals,
2636 they are lost beyond the limit of the FP precision.
2637 However, neither is canonical, so both only get p
2638 flags. NWC, 2000/11/25 */
2639 /* Both already have p flags, so do nothing */
2641 const NV nv = SvNVX(sv);
2642 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2643 if (SvIVX(sv) == I_V(nv)) {
2646 /* It had no "." so it must be integer. */
2650 /* between IV_MAX and NV(UV_MAX).
2651 Could be slightly > UV_MAX */
2653 if (numtype & IS_NUMBER_NOT_INT) {
2654 /* UV and NV both imprecise. */
2656 const UV nv_as_uv = U_V(nv);
2658 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2667 /* It might be more code efficient to go through the entire logic above
2668 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2669 gets complex and potentially buggy, so more programmer efficient
2670 to do it this way, by turning off the public flags: */
2672 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2673 #endif /* NV_PRESERVES_UV */
2676 if (isGV_with_GP(sv)) {
2677 glob_2number(MUTABLE_GV(sv));
2681 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2683 assert (SvTYPE(sv) >= SVt_NV);
2684 /* Typically the caller expects that sv_any is not NULL now. */
2685 /* XXX Ilya implies that this is a bug in callers that assume this
2686 and ideally should be fixed. */
2689 #if defined(USE_LONG_DOUBLE)
2691 STORE_NUMERIC_LOCAL_SET_STANDARD();
2692 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2693 PTR2UV(sv), SvNVX(sv));
2694 RESTORE_NUMERIC_LOCAL();
2698 STORE_NUMERIC_LOCAL_SET_STANDARD();
2699 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2700 PTR2UV(sv), SvNVX(sv));
2701 RESTORE_NUMERIC_LOCAL();
2710 Return an SV with the numeric value of the source SV, doing any necessary
2711 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2712 access this function.
2718 Perl_sv_2num(pTHX_ SV *const sv)
2720 PERL_ARGS_ASSERT_SV_2NUM;
2725 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2726 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2727 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2728 return sv_2num(tmpsv);
2730 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2733 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2734 * UV as a string towards the end of buf, and return pointers to start and
2737 * We assume that buf is at least TYPE_CHARS(UV) long.
2741 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2743 char *ptr = buf + TYPE_CHARS(UV);
2744 char * const ebuf = ptr;
2747 PERL_ARGS_ASSERT_UIV_2BUF;
2759 *--ptr = '0' + (char)(uv % 10);
2768 =for apidoc sv_2pv_flags
2770 Returns a pointer to the string value of an SV, and sets *lp to its length.
2771 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2772 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2773 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2779 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2784 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2786 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2787 && SvTYPE(sv) != SVt_PVFM);
2788 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2793 if (flags & SV_SKIP_OVERLOAD)
2795 tmpstr = AMG_CALLunary(sv, string_amg);
2796 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2797 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2799 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2803 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2804 if (flags & SV_CONST_RETURN) {
2805 pv = (char *) SvPVX_const(tmpstr);
2807 pv = (flags & SV_MUTABLE_RETURN)
2808 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2811 *lp = SvCUR(tmpstr);
2813 pv = sv_2pv_flags(tmpstr, lp, flags);
2826 SV *const referent = SvRV(sv);
2830 retval = buffer = savepvn("NULLREF", len);
2831 } else if (SvTYPE(referent) == SVt_REGEXP &&
2832 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2833 amagic_is_enabled(string_amg))) {
2834 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2838 /* If the regex is UTF-8 we want the containing scalar to
2839 have an UTF-8 flag too */
2846 *lp = RX_WRAPLEN(re);
2848 return RX_WRAPPED(re);
2850 const char *const typestr = sv_reftype(referent, 0);
2851 const STRLEN typelen = strlen(typestr);
2852 UV addr = PTR2UV(referent);
2853 const char *stashname = NULL;
2854 STRLEN stashnamelen = 0; /* hush, gcc */
2855 const char *buffer_end;
2857 if (SvOBJECT(referent)) {
2858 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2861 stashname = HEK_KEY(name);
2862 stashnamelen = HEK_LEN(name);
2864 if (HEK_UTF8(name)) {
2870 stashname = "__ANON__";
2873 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2874 + 2 * sizeof(UV) + 2 /* )\0 */;
2876 len = typelen + 3 /* (0x */
2877 + 2 * sizeof(UV) + 2 /* )\0 */;
2880 Newx(buffer, len, char);
2881 buffer_end = retval = buffer + len;
2883 /* Working backwards */
2887 *--retval = PL_hexdigit[addr & 15];
2888 } while (addr >>= 4);
2894 memcpy(retval, typestr, typelen);
2898 retval -= stashnamelen;
2899 memcpy(retval, stashname, stashnamelen);
2901 /* retval may not necessarily have reached the start of the
2903 assert (retval >= buffer);
2905 len = buffer_end - retval - 1; /* -1 for that \0 */
2917 if (flags & SV_MUTABLE_RETURN)
2918 return SvPVX_mutable(sv);
2919 if (flags & SV_CONST_RETURN)
2920 return (char *)SvPVX_const(sv);
2925 /* I'm assuming that if both IV and NV are equally valid then
2926 converting the IV is going to be more efficient */
2927 const U32 isUIOK = SvIsUV(sv);
2928 char buf[TYPE_CHARS(UV)];
2932 if (SvTYPE(sv) < SVt_PVIV)
2933 sv_upgrade(sv, SVt_PVIV);
2934 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2936 /* inlined from sv_setpvn */
2937 s = SvGROW_mutable(sv, len + 1);
2938 Move(ptr, s, len, char);
2943 else if (SvNOK(sv)) {
2944 if (SvTYPE(sv) < SVt_PVNV)
2945 sv_upgrade(sv, SVt_PVNV);
2946 if (SvNVX(sv) == 0.0) {
2947 s = SvGROW_mutable(sv, 2);
2952 /* The +20 is pure guesswork. Configure test needed. --jhi */
2953 s = SvGROW_mutable(sv, NV_DIG + 20);
2954 /* some Xenix systems wipe out errno here */
2956 #ifndef USE_LOCALE_NUMERIC
2957 V_Gconvert(SvNVX(sv), NV_DIG, 0, s);
2961 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
2962 V_Gconvert(SvNVX(sv), NV_DIG, 0, s);
2964 /* If the radix character is UTF-8, and actually is in the
2965 * output, turn on the UTF-8 flag for the scalar */
2966 if (PL_numeric_local
2967 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
2968 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
2972 RESTORE_LC_NUMERIC();
2975 /* We don't call SvPOK_on(), because it may come to pass that the
2976 * locale changes so that the stringification we just did is no
2977 * longer correct. We will have to re-stringify every time it is
2984 else if (isGV_with_GP(sv)) {
2985 GV *const gv = MUTABLE_GV(sv);
2986 SV *const buffer = sv_newmortal();
2988 gv_efullname3(buffer, gv, "*");
2990 assert(SvPOK(buffer));
2994 *lp = SvCUR(buffer);
2995 return SvPVX(buffer);
2997 else if (isREGEXP(sv)) {
2998 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2999 return RX_WRAPPED((REGEXP *)sv);
3004 if (flags & SV_UNDEF_RETURNS_NULL)
3006 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3008 /* Typically the caller expects that sv_any is not NULL now. */
3009 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3010 sv_upgrade(sv, SVt_PV);
3015 const STRLEN len = s - SvPVX_const(sv);
3020 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3021 PTR2UV(sv),SvPVX_const(sv)));
3022 if (flags & SV_CONST_RETURN)
3023 return (char *)SvPVX_const(sv);
3024 if (flags & SV_MUTABLE_RETURN)
3025 return SvPVX_mutable(sv);
3030 =for apidoc sv_copypv
3032 Copies a stringified representation of the source SV into the
3033 destination SV. Automatically performs any necessary mg_get and
3034 coercion of numeric values into strings. Guaranteed to preserve
3035 UTF8 flag even from overloaded objects. Similar in nature to
3036 sv_2pv[_flags] but operates directly on an SV instead of just the
3037 string. Mostly uses sv_2pv_flags to do its work, except when that
3038 would lose the UTF-8'ness of the PV.
3040 =for apidoc sv_copypv_nomg
3042 Like sv_copypv, but doesn't invoke get magic first.
3044 =for apidoc sv_copypv_flags
3046 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3053 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3055 PERL_ARGS_ASSERT_SV_COPYPV;
3057 sv_copypv_flags(dsv, ssv, 0);
3061 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3066 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3068 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3070 s = SvPV_nomg_const(ssv,len);
3071 sv_setpvn(dsv,s,len);
3079 =for apidoc sv_2pvbyte
3081 Return a pointer to the byte-encoded representation of the SV, and set *lp
3082 to its length. May cause the SV to be downgraded from UTF-8 as a
3085 Usually accessed via the C<SvPVbyte> macro.
3091 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3093 PERL_ARGS_ASSERT_SV_2PVBYTE;
3096 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3097 || isGV_with_GP(sv) || SvROK(sv)) {
3098 SV *sv2 = sv_newmortal();
3099 sv_copypv_nomg(sv2,sv);
3102 sv_utf8_downgrade(sv,0);
3103 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3107 =for apidoc sv_2pvutf8
3109 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3110 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3112 Usually accessed via the C<SvPVutf8> macro.
3118 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3120 PERL_ARGS_ASSERT_SV_2PVUTF8;
3122 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3123 || isGV_with_GP(sv) || SvROK(sv))
3124 sv = sv_mortalcopy(sv);
3127 sv_utf8_upgrade_nomg(sv);
3128 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3133 =for apidoc sv_2bool
3135 This macro is only used by sv_true() or its macro equivalent, and only if
3136 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3137 It calls sv_2bool_flags with the SV_GMAGIC flag.
3139 =for apidoc sv_2bool_flags
3141 This function is only used by sv_true() and friends, and only if
3142 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3143 contain SV_GMAGIC, then it does an mg_get() first.
3150 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3154 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3157 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3163 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3164 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3167 if(SvGMAGICAL(sv)) {
3169 goto restart; /* call sv_2bool */
3171 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3172 else if(!SvOK(sv)) {
3175 else if(SvPOK(sv)) {
3176 svb = SvPVXtrue(sv);
3178 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3179 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3180 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3184 goto restart; /* call sv_2bool_nomg */
3189 return SvRV(sv) != 0;
3193 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3194 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3198 =for apidoc sv_utf8_upgrade
3200 Converts the PV of an SV to its UTF-8-encoded form.
3201 Forces the SV to string form if it is not already.
3202 Will C<mg_get> on C<sv> if appropriate.
3203 Always sets the SvUTF8 flag to avoid future validity checks even
3204 if the whole string is the same in UTF-8 as not.
3205 Returns the number of bytes in the converted string
3207 This is not a general purpose byte encoding to Unicode interface:
3208 use the Encode extension for that.
3210 =for apidoc sv_utf8_upgrade_nomg
3212 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3214 =for apidoc sv_utf8_upgrade_flags
3216 Converts the PV of an SV to its UTF-8-encoded form.
3217 Forces the SV to string form if it is not already.
3218 Always sets the SvUTF8 flag to avoid future validity checks even
3219 if all the bytes are invariant in UTF-8.
3220 If C<flags> has C<SV_GMAGIC> bit set,
3221 will C<mg_get> on C<sv> if appropriate, else not.
3223 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3224 will expand when converted to UTF-8, and skips the extra work of checking for
3225 that. Typically this flag is used by a routine that has already parsed the
3226 string and found such characters, and passes this information on so that the
3227 work doesn't have to be repeated.
3229 Returns the number of bytes in the converted string.
3231 This is not a general purpose byte encoding to Unicode interface:
3232 use the Encode extension for that.
3234 =for apidoc sv_utf8_upgrade_flags_grow
3236 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3237 the number of unused bytes the string of 'sv' is guaranteed to have free after
3238 it upon return. This allows the caller to reserve extra space that it intends
3239 to fill, to avoid extra grows.
3241 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3242 are implemented in terms of this function.
3244 Returns the number of bytes in the converted string (not including the spares).
3248 (One might think that the calling routine could pass in the position of the
3249 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3250 have to be found again. But that is not the case, because typically when the
3251 caller is likely to use this flag, it won't be calling this routine unless it
3252 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3253 and just use bytes. But some things that do fit into a byte are variants in
3254 utf8, and the caller may not have been keeping track of these.)
3256 If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3257 isn't guaranteed due to having other routines do the work in some input cases,
3258 or if the input is already flagged as being in utf8.
3260 The speed of this could perhaps be improved for many cases if someone wanted to
3261 write a fast function that counts the number of variant characters in a string,
3262 especially if it could return the position of the first one.
3267 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3271 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3273 if (sv == &PL_sv_undef)
3275 if (!SvPOK_nog(sv)) {
3277 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3278 (void) sv_2pv_flags(sv,&len, flags);
3280 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3284 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3289 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3294 S_sv_uncow(aTHX_ sv, 0);
3297 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3298 sv_recode_to_utf8(sv, PL_encoding);
3299 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3303 if (SvCUR(sv) == 0) {
3304 if (extra) SvGROW(sv, extra);
3305 } else { /* Assume Latin-1/EBCDIC */
3306 /* This function could be much more efficient if we
3307 * had a FLAG in SVs to signal if there are any variant
3308 * chars in the PV. Given that there isn't such a flag
3309 * make the loop as fast as possible (although there are certainly ways
3310 * to speed this up, eg. through vectorization) */
3311 U8 * s = (U8 *) SvPVX_const(sv);
3312 U8 * e = (U8 *) SvEND(sv);
3314 STRLEN two_byte_count = 0;
3316 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3318 /* See if really will need to convert to utf8. We mustn't rely on our
3319 * incoming SV being well formed and having a trailing '\0', as certain
3320 * code in pp_formline can send us partially built SVs. */
3324 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3326 t--; /* t already incremented; re-point to first variant */
3331 /* utf8 conversion not needed because all are invariants. Mark as
3332 * UTF-8 even if no variant - saves scanning loop */
3334 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3339 /* Here, the string should be converted to utf8, either because of an
3340 * input flag (two_byte_count = 0), or because a character that
3341 * requires 2 bytes was found (two_byte_count = 1). t points either to
3342 * the beginning of the string (if we didn't examine anything), or to
3343 * the first variant. In either case, everything from s to t - 1 will
3344 * occupy only 1 byte each on output.
3346 * There are two main ways to convert. One is to create a new string
3347 * and go through the input starting from the beginning, appending each
3348 * converted value onto the new string as we go along. It's probably
3349 * best to allocate enough space in the string for the worst possible
3350 * case rather than possibly running out of space and having to
3351 * reallocate and then copy what we've done so far. Since everything
3352 * from s to t - 1 is invariant, the destination can be initialized
3353 * with these using a fast memory copy
3355 * The other way is to figure out exactly how big the string should be
3356 * by parsing the entire input. Then you don't have to make it big
3357 * enough to handle the worst possible case, and more importantly, if
3358 * the string you already have is large enough, you don't have to
3359 * allocate a new string, you can copy the last character in the input
3360 * string to the final position(s) that will be occupied by the
3361 * converted string and go backwards, stopping at t, since everything
3362 * before that is invariant.
3364 * There are advantages and disadvantages to each method.
3366 * In the first method, we can allocate a new string, do the memory
3367 * copy from the s to t - 1, and then proceed through the rest of the
3368 * string byte-by-byte.
3370 * In the second method, we proceed through the rest of the input
3371 * string just calculating how big the converted string will be. Then
3372 * there are two cases:
3373 * 1) if the string has enough extra space to handle the converted
3374 * value. We go backwards through the string, converting until we
3375 * get to the position we are at now, and then stop. If this
3376 * position is far enough along in the string, this method is
3377 * faster than the other method. If the memory copy were the same
3378 * speed as the byte-by-byte loop, that position would be about
3379 * half-way, as at the half-way mark, parsing to the end and back
3380 * is one complete string's parse, the same amount as starting
3381 * over and going all the way through. Actually, it would be
3382 * somewhat less than half-way, as it's faster to just count bytes
3383 * than to also copy, and we don't have the overhead of allocating
3384 * a new string, changing the scalar to use it, and freeing the
3385 * existing one. But if the memory copy is fast, the break-even
3386 * point is somewhere after half way. The counting loop could be
3387 * sped up by vectorization, etc, to move the break-even point
3388 * further towards the beginning.
3389 * 2) if the string doesn't have enough space to handle the converted
3390 * value. A new string will have to be allocated, and one might
3391 * as well, given that, start from the beginning doing the first
3392 * method. We've spent extra time parsing the string and in
3393 * exchange all we've gotten is that we know precisely how big to
3394 * make the new one. Perl is more optimized for time than space,
3395 * so this case is a loser.
3396 * So what I've decided to do is not use the 2nd method unless it is
3397 * guaranteed that a new string won't have to be allocated, assuming
3398 * the worst case. I also decided not to put any more conditions on it
3399 * than this, for now. It seems likely that, since the worst case is
3400 * twice as big as the unknown portion of the string (plus 1), we won't
3401 * be guaranteed enough space, causing us to go to the first method,
3402 * unless the string is short, or the first variant character is near
3403 * the end of it. In either of these cases, it seems best to use the
3404 * 2nd method. The only circumstance I can think of where this would
3405 * be really slower is if the string had once had much more data in it
3406 * than it does now, but there is still a substantial amount in it */
3409 STRLEN invariant_head = t - s;
3410 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3411 if (SvLEN(sv) < size) {
3413 /* Here, have decided to allocate a new string */
3418 Newx(dst, size, U8);
3420 /* If no known invariants at the beginning of the input string,
3421 * set so starts from there. Otherwise, can use memory copy to
3422 * get up to where we are now, and then start from here */
3424 if (invariant_head <= 0) {
3427 Copy(s, dst, invariant_head, char);
3428 d = dst + invariant_head;
3432 append_utf8_from_native_byte(*t, &d);
3436 SvPV_free(sv); /* No longer using pre-existing string */
3437 SvPV_set(sv, (char*)dst);
3438 SvCUR_set(sv, d - dst);
3439 SvLEN_set(sv, size);
3442 /* Here, have decided to get the exact size of the string.
3443 * Currently this happens only when we know that there is
3444 * guaranteed enough space to fit the converted string, so
3445 * don't have to worry about growing. If two_byte_count is 0,
3446 * then t points to the first byte of the string which hasn't
3447 * been examined yet. Otherwise two_byte_count is 1, and t
3448 * points to the first byte in the string that will expand to
3449 * two. Depending on this, start examining at t or 1 after t.
3452 U8 *d = t + two_byte_count;
3455 /* Count up the remaining bytes that expand to two */
3458 const U8 chr = *d++;
3459 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3462 /* The string will expand by just the number of bytes that
3463 * occupy two positions. But we are one afterwards because of
3464 * the increment just above. This is the place to put the
3465 * trailing NUL, and to set the length before we decrement */
3467 d += two_byte_count;
3468 SvCUR_set(sv, d - s);
3472 /* Having decremented d, it points to the position to put the
3473 * very last byte of the expanded string. Go backwards through
3474 * the string, copying and expanding as we go, stopping when we
3475 * get to the part that is invariant the rest of the way down */
3479 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3482 *d-- = UTF8_EIGHT_BIT_LO(*e);
3483 *d-- = UTF8_EIGHT_BIT_HI(*e);
3489 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3490 /* Update pos. We do it at the end rather than during
3491 * the upgrade, to avoid slowing down the common case
3492 * (upgrade without pos).
3493 * pos can be stored as either bytes or characters. Since
3494 * this was previously a byte string we can just turn off
3495 * the bytes flag. */
3496 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3498 mg->mg_flags &= ~MGf_BYTES;
3500 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3501 magic_setutf8(sv,mg); /* clear UTF8 cache */
3506 /* Mark as UTF-8 even if no variant - saves scanning loop */
3512 =for apidoc sv_utf8_downgrade
3514 Attempts to convert the PV of an SV from characters to bytes.
3515 If the PV contains a character that cannot fit
3516 in a byte, this conversion will fail;
3517 in this case, either returns false or, if C<fail_ok> is not
3520 This is not a general purpose Unicode to byte encoding interface:
3521 use the Encode extension for that.
3527 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3531 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3533 if (SvPOKp(sv) && SvUTF8(sv)) {
3537 int mg_flags = SV_GMAGIC;
3540 S_sv_uncow(aTHX_ sv, 0);
3542 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3544 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3545 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3546 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3547 SV_GMAGIC|SV_CONST_RETURN);
3548 mg_flags = 0; /* sv_pos_b2u does get magic */
3550 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3551 magic_setutf8(sv,mg); /* clear UTF8 cache */
3554 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3556 if (!utf8_to_bytes(s, &len)) {
3561 Perl_croak(aTHX_ "Wide character in %s",
3564 Perl_croak(aTHX_ "Wide character");
3575 =for apidoc sv_utf8_encode
3577 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3578 flag off so that it looks like octets again.
3584 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3586 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3588 if (SvREADONLY(sv)) {
3589 sv_force_normal_flags(sv, 0);
3591 (void) sv_utf8_upgrade(sv);
3596 =for apidoc sv_utf8_decode
3598 If the PV of the SV is an octet sequence in UTF-8
3599 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3600 so that it looks like a character. If the PV contains only single-byte
3601 characters, the C<SvUTF8> flag stays off.
3602 Scans PV for validity and returns false if the PV is invalid UTF-8.
3608 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3610 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3613 const U8 *start, *c;
3616 /* The octets may have got themselves encoded - get them back as
3619 if (!sv_utf8_downgrade(sv, TRUE))
3622 /* it is actually just a matter of turning the utf8 flag on, but
3623 * we want to make sure everything inside is valid utf8 first.
3625 c = start = (const U8 *) SvPVX_const(sv);
3626 if (!is_utf8_string(c, SvCUR(sv)))
3628 e = (const U8 *) SvEND(sv);
3631 if (!UTF8_IS_INVARIANT(ch)) {
3636 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3637 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3638 after this, clearing pos. Does anything on CPAN
3640 /* adjust pos to the start of a UTF8 char sequence */
3641 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3643 I32 pos = mg->mg_len;
3645 for (c = start + pos; c > start; c--) {
3646 if (UTF8_IS_START(*c))
3649 mg->mg_len = c - start;
3652 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3653 magic_setutf8(sv,mg); /* clear UTF8 cache */
3660 =for apidoc sv_setsv
3662 Copies the contents of the source SV C<ssv> into the destination SV
3663 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3664 function if the source SV needs to be reused. Does not handle 'set' magic on
3665 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3666 performs a copy-by-value, obliterating any previous content of the
3669 You probably want to use one of the assortment of wrappers, such as
3670 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3671 C<SvSetMagicSV_nosteal>.
3673 =for apidoc sv_setsv_flags
3675 Copies the contents of the source SV C<ssv> into the destination SV
3676 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3677 function if the source SV needs to be reused. Does not handle 'set' magic.
3678 Loosely speaking, it performs a copy-by-value, obliterating any previous
3679 content of the destination.
3680 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3681 C<ssv> if appropriate, else not. If the C<flags>
3682 parameter has the C<SV_NOSTEAL> bit set then the
3683 buffers of temps will not be stolen. <sv_setsv>
3684 and C<sv_setsv_nomg> are implemented in terms of this function.
3686 You probably want to use one of the assortment of wrappers, such as
3687 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3688 C<SvSetMagicSV_nosteal>.
3690 This is the primary function for copying scalars, and most other
3691 copy-ish functions and macros use this underneath.
3697 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3699 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3700 HV *old_stash = NULL;
3702 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3704 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3705 const char * const name = GvNAME(sstr);
3706 const STRLEN len = GvNAMELEN(sstr);
3708 if (dtype >= SVt_PV) {
3714 SvUPGRADE(dstr, SVt_PVGV);
3715 (void)SvOK_off(dstr);
3716 isGV_with_GP_on(dstr);
3718 GvSTASH(dstr) = GvSTASH(sstr);
3720 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3721 gv_name_set(MUTABLE_GV(dstr), name, len,
3722 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3723 SvFAKE_on(dstr); /* can coerce to non-glob */
3726 if(GvGP(MUTABLE_GV(sstr))) {
3727 /* If source has method cache entry, clear it */
3729 SvREFCNT_dec(GvCV(sstr));
3730 GvCV_set(sstr, NULL);
3733 /* If source has a real method, then a method is
3736 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3742 /* If dest already had a real method, that's a change as well */
3744 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3745 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3750 /* We don't need to check the name of the destination if it was not a
3751 glob to begin with. */
3752 if(dtype == SVt_PVGV) {
3753 const char * const name = GvNAME((const GV *)dstr);
3756 /* The stash may have been detached from the symbol table, so
3758 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3762 const STRLEN len = GvNAMELEN(dstr);
3763 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3764 || (len == 1 && name[0] == ':')) {
3767 /* Set aside the old stash, so we can reset isa caches on
3769 if((old_stash = GvHV(dstr)))
3770 /* Make sure we do not lose it early. */
3771 SvREFCNT_inc_simple_void_NN(
3772 sv_2mortal((SV *)old_stash)
3777 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3780 gp_free(MUTABLE_GV(dstr));
3781 GvINTRO_off(dstr); /* one-shot flag */
3782 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3783 if (SvTAINTED(sstr))
3785 if (GvIMPORTED(dstr) != GVf_IMPORTED
3786 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3788 GvIMPORTED_on(dstr);
3791 if(mro_changes == 2) {
3792 if (GvAV((const GV *)sstr)) {
3794 SV * const sref = (SV *)GvAV((const GV *)dstr);
3795 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3796 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3797 AV * const ary = newAV();
3798 av_push(ary, mg->mg_obj); /* takes the refcount */
3799 mg->mg_obj = (SV *)ary;
3801 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3803 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3805 mro_isa_changed_in(GvSTASH(dstr));
3807 else if(mro_changes == 3) {
3808 HV * const stash = GvHV(dstr);
3809 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3815 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3816 if (GvIO(dstr) && dtype == SVt_PVGV) {
3817 DEBUG_o(Perl_deb(aTHX_
3818 "glob_assign_glob clearing PL_stashcache\n"));
3819 /* It's a cache. It will rebuild itself quite happily.
3820 It's a lot of effort to work out exactly which key (or keys)
3821 might be invalidated by the creation of the this file handle.
3823 hv_clear(PL_stashcache);
3829 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3831 SV * const sref = SvRV(sstr);
3833 const int intro = GvINTRO(dstr);
3836 const U32 stype = SvTYPE(sref);
3838 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3841 GvINTRO_off(dstr); /* one-shot flag */
3842 GvLINE(dstr) = CopLINE(PL_curcop);
3843 GvEGV(dstr) = MUTABLE_GV(dstr);
3848 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3849 import_flag = GVf_IMPORTED_CV;
3852 location = (SV **) &GvHV(dstr);
3853 import_flag = GVf_IMPORTED_HV;
3856 location = (SV **) &GvAV(dstr);
3857 import_flag = GVf_IMPORTED_AV;
3860 location = (SV **) &GvIOp(dstr);
3863 location = (SV **) &GvFORM(dstr);
3866 location = &GvSV(dstr);
3867 import_flag = GVf_IMPORTED_SV;
3870 if (stype == SVt_PVCV) {
3871 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3872 if (GvCVGEN(dstr)) {
3873 SvREFCNT_dec(GvCV(dstr));
3874 GvCV_set(dstr, NULL);
3875 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3878 /* SAVEt_GVSLOT takes more room on the savestack and has more
3879 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3880 leave_scope needs access to the GV so it can reset method
3881 caches. We must use SAVEt_GVSLOT whenever the type is
3882 SVt_PVCV, even if the stash is anonymous, as the stash may
3883 gain a name somehow before leave_scope. */
3884 if (stype == SVt_PVCV) {
3885 /* There is no save_pushptrptrptr. Creating it for this
3886 one call site would be overkill. So inline the ss add
3890 SS_ADD_PTR(location);
3891 SS_ADD_PTR(SvREFCNT_inc(*location));
3892 SS_ADD_UV(SAVEt_GVSLOT);
3895 else SAVEGENERICSV(*location);
3898 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3899 CV* const cv = MUTABLE_CV(*location);
3901 if (!GvCVGEN((const GV *)dstr) &&
3902 (CvROOT(cv) || CvXSUB(cv)) &&
3903 /* redundant check that avoids creating the extra SV
3904 most of the time: */
3905 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3907 SV * const new_const_sv =
3908 CvCONST((const CV *)sref)
3909 ? cv_const_sv((const CV *)sref)
3911 report_redefined_cv(
3912 sv_2mortal(Perl_newSVpvf(aTHX_
3915 HvNAME_HEK(GvSTASH((const GV *)dstr))
3917 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3920 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3924 cv_ckproto_len_flags(cv, (const GV *)dstr,
3925 SvPOK(sref) ? CvPROTO(sref) : NULL,
3926 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3927 SvPOK(sref) ? SvUTF8(sref) : 0);
3929 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3930 GvASSUMECV_on(dstr);
3931 if(GvSTASH(dstr)) gv_method_changed(dstr); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3933 *location = SvREFCNT_inc_simple_NN(sref);
3934 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3935 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3936 GvFLAGS(dstr) |= import_flag;
3938 if (stype == SVt_PVHV) {
3939 const char * const name = GvNAME((GV*)dstr);
3940 const STRLEN len = GvNAMELEN(dstr);
3943 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3944 || (len == 1 && name[0] == ':')
3946 && (!dref || HvENAME_get(dref))
3949 (HV *)sref, (HV *)dref,
3955 stype == SVt_PVAV && sref != dref
3956 && strEQ(GvNAME((GV*)dstr), "ISA")
3957 /* The stash may have been detached from the symbol table, so
3958 check its name before doing anything. */
3959 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3962 MAGIC * const omg = dref && SvSMAGICAL(dref)
3963 ? mg_find(dref, PERL_MAGIC_isa)
3965 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3966 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3967 AV * const ary = newAV();
3968 av_push(ary, mg->mg_obj); /* takes the refcount */
3969 mg->mg_obj = (SV *)ary;
3972 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
3973 SV **svp = AvARRAY((AV *)omg->mg_obj);
3974 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
3978 SvREFCNT_inc_simple_NN(*svp++)
3984 SvREFCNT_inc_simple_NN(omg->mg_obj)
3988 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
3993 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
3995 mg = mg_find(sref, PERL_MAGIC_isa);
3997 /* Since the *ISA assignment could have affected more than
3998 one stash, don't call mro_isa_changed_in directly, but let
3999 magic_clearisa do it for us, as it already has the logic for
4000 dealing with globs vs arrays of globs. */
4002 Perl_magic_clearisa(aTHX_ NULL, mg);
4004 else if (stype == SVt_PVIO) {
4005 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4006 /* It's a cache. It will rebuild itself quite happily.
4007 It's a lot of effort to work out exactly which key (or keys)
4008 might be invalidated by the creation of the this file handle.
4010 hv_clear(PL_stashcache);
4014 if (!intro) SvREFCNT_dec(dref);
4015 if (SvTAINTED(sstr))
4020 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
4022 #if SV_COW_THRESHOLD
4023 # define GE_COW_THRESHOLD(len) ((len) >= SV_COW_THRESHOLD)
4025 # define GE_COW_THRESHOLD(len) 1
4027 #if SV_COWBUF_THRESHOLD
4028 # define GE_COWBUF_THRESHOLD(len) ((len) >= SV_COWBUF_THRESHOLD)
4030 # define GE_COWBUF_THRESHOLD(len) 1
4033 #ifdef PERL_DEBUG_READONLY_COW
4034 # include <sys/mman.h>
4036 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4037 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4041 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4043 struct perl_memory_debug_header * const header =
4044 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4045 const MEM_SIZE len = header->size;
4046 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4047 # ifdef PERL_TRACK_MEMPOOL
4048 if (!header->readonly) header->readonly = 1;
4050 if (mprotect(header, len, PROT_READ))
4051 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4052 header, len, errno);
4056 S_sv_buf_to_rw(pTHX_ SV *sv)
4058 struct perl_memory_debug_header * const header =
4059 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4060 const MEM_SIZE len = header->size;
4061 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4062 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4063 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4064 header, len, errno);
4065 # ifdef PERL_TRACK_MEMPOOL
4066 header->readonly = 0;
4071 # define sv_buf_to_ro(sv) NOOP
4072 # define sv_buf_to_rw(sv) NOOP
4076 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4083 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4088 if (SvIS_FREED(dstr)) {
4089 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4090 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4092 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4094 sstr = &PL_sv_undef;
4095 if (SvIS_FREED(sstr)) {
4096 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4097 (void*)sstr, (void*)dstr);
4099 stype = SvTYPE(sstr);
4100 dtype = SvTYPE(dstr);
4102 /* There's a lot of redundancy below but we're going for speed here */
4107 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4108 (void)SvOK_off(dstr);
4116 sv_upgrade(dstr, SVt_IV);
4120 sv_upgrade(dstr, SVt_PVIV);
4124 goto end_of_first_switch;
4126 (void)SvIOK_only(dstr);
4127 SvIV_set(dstr, SvIVX(sstr));
4130 /* SvTAINTED can only be true if the SV has taint magic, which in
4131 turn means that the SV type is PVMG (or greater). This is the
4132 case statement for SVt_IV, so this cannot be true (whatever gcov
4134 assert(!SvTAINTED(sstr));
4139 if (dtype < SVt_PV && dtype != SVt_IV)
4140 sv_upgrade(dstr, SVt_IV);
4148 sv_upgrade(dstr, SVt_NV);
4152 sv_upgrade(dstr, SVt_PVNV);
4156 goto end_of_first_switch;
4158 SvNV_set(dstr, SvNVX(sstr));
4159 (void)SvNOK_only(dstr);
4160 /* SvTAINTED can only be true if the SV has taint magic, which in
4161 turn means that the SV type is PVMG (or greater). This is the
4162 case statement for SVt_NV, so this cannot be true (whatever gcov
4164 assert(!SvTAINTED(sstr));
4171 sv_upgrade(dstr, SVt_PV);
4174 if (dtype < SVt_PVIV)
4175 sv_upgrade(dstr, SVt_PVIV);
4178 if (dtype < SVt_PVNV)
4179 sv_upgrade(dstr, SVt_PVNV);
4183 const char * const type = sv_reftype(sstr,0);
4185 /* diag_listed_as: Bizarre copy of %s */
4186 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4188 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4194 if (dtype < SVt_REGEXP)
4196 if (dtype >= SVt_PV) {
4202 sv_upgrade(dstr, SVt_REGEXP);
4210 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4212 if (SvTYPE(sstr) != stype)
4213 stype = SvTYPE(sstr);
4215 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4216 glob_assign_glob(dstr, sstr, dtype);
4219 if (stype == SVt_PVLV)
4221 if (isREGEXP(sstr)) goto upgregexp;
4222 SvUPGRADE(dstr, SVt_PVNV);
4225 SvUPGRADE(dstr, (svtype)stype);
4227 end_of_first_switch:
4229 /* dstr may have been upgraded. */
4230 dtype = SvTYPE(dstr);
4231 sflags = SvFLAGS(sstr);
4233 if (dtype == SVt_PVCV) {
4234 /* Assigning to a subroutine sets the prototype. */
4237 const char *const ptr = SvPV_const(sstr, len);
4239 SvGROW(dstr, len + 1);
4240 Copy(ptr, SvPVX(dstr), len + 1, char);
4241 SvCUR_set(dstr, len);
4243 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4244 CvAUTOLOAD_off(dstr);
4249 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4250 const char * const type = sv_reftype(dstr,0);
4252 /* diag_listed_as: Cannot copy to %s */
4253 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4255 Perl_croak(aTHX_ "Cannot copy to %s", type);
4256 } else if (sflags & SVf_ROK) {
4257 if (isGV_with_GP(dstr)
4258 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4261 if (GvIMPORTED(dstr) != GVf_IMPORTED
4262 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4264 GvIMPORTED_on(dstr);
4269 glob_assign_glob(dstr, sstr, dtype);
4273 if (dtype >= SVt_PV) {
4274 if (isGV_with_GP(dstr)) {
4275 glob_assign_ref(dstr, sstr);
4278 if (SvPVX_const(dstr)) {
4284 (void)SvOK_off(dstr);
4285 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4286 SvFLAGS(dstr) |= sflags & SVf_ROK;
4287 assert(!(sflags & SVp_NOK));
4288 assert(!(sflags & SVp_IOK));
4289 assert(!(sflags & SVf_NOK));
4290 assert(!(sflags & SVf_IOK));
4292 else if (isGV_with_GP(dstr)) {
4293 if (!(sflags & SVf_OK)) {
4294 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4295 "Undefined value assigned to typeglob");
4298 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4299 if (dstr != (const SV *)gv) {
4300 const char * const name = GvNAME((const GV *)dstr);
4301 const STRLEN len = GvNAMELEN(dstr);
4302 HV *old_stash = NULL;
4303 bool reset_isa = FALSE;
4304 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4305 || (len == 1 && name[0] == ':')) {
4306 /* Set aside the old stash, so we can reset isa caches
4307 on its subclasses. */
4308 if((old_stash = GvHV(dstr))) {
4309 /* Make sure we do not lose it early. */
4310 SvREFCNT_inc_simple_void_NN(
4311 sv_2mortal((SV *)old_stash)
4318 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4319 gp_free(MUTABLE_GV(dstr));
4321 GvGP_set(dstr, gp_ref(GvGP(gv)));
4324 HV * const stash = GvHV(dstr);
4326 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4336 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4337 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4338 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4340 else if (sflags & SVp_POK) {
4341 const STRLEN cur = SvCUR(sstr);
4342 const STRLEN len = SvLEN(sstr);
4345 * We have three basic ways to copy the string:
4351 * Which we choose is based on various factors. The following
4352 * things are listed in order of speed, fastest to slowest:
4354 * - Copying a short string
4355 * - Copy-on-write bookkeeping
4357 * - Copying a long string
4359 * We swipe the string (steal the string buffer) if the SV on the
4360 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4361 * big win on long strings. It should be a win on short strings if
4362 * SvPVX_const(dstr) has to be allocated. If not, it should not
4363 * slow things down, as SvPVX_const(sstr) would have been freed
4366 * We also steal the buffer from a PADTMP (operator target) if it
4367 * is ‘long enough’. For short strings, a swipe does not help
4368 * here, as it causes more malloc calls the next time the target
4369 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4370 * be allocated it is still not worth swiping PADTMPs for short
4371 * strings, as the savings here are small.
4373 * If the rhs is already flagged as a copy-on-write string and COW
4374 * is possible here, we use copy-on-write and make both SVs share
4375 * the string buffer.
4377 * If the rhs is not flagged as copy-on-write, then we see whether
4378 * it is worth upgrading it to such. If the lhs already has a buf-
4379 * fer big enough and the string is short, we skip it and fall back
4380 * to method 3, since memcpy is faster for short strings than the
4381 * later bookkeeping overhead that copy-on-write entails.
4383 * If there is no buffer on the left, or the buffer is too small,
4384 * then we use copy-on-write.
4387 /* Whichever path we take through the next code, we want this true,
4388 and doing it now facilitates the COW check. */
4389 (void)SvPOK_only(dstr);
4393 /* slated for free anyway (and not COW)? */
4394 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4395 /* or a swipable TARG */
4396 || ((sflags & (SVs_PADTMP|SVf_READONLY|SVf_IsCOW))
4398 /* whose buffer is worth stealing */
4399 && GE_COWBUF_THRESHOLD(cur)
4402 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4403 (!(flags & SV_NOSTEAL)) &&
4404 /* and we're allowed to steal temps */
4405 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4406 len) /* and really is a string */
4407 { /* Passes the swipe test. */
4408 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4410 SvPV_set(dstr, SvPVX_mutable(sstr));
4411 SvLEN_set(dstr, SvLEN(sstr));
4412 SvCUR_set(dstr, SvCUR(sstr));
4415 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4416 SvPV_set(sstr, NULL);
4421 else if (flags & SV_COW_SHARED_HASH_KEYS
4423 #ifdef PERL_OLD_COPY_ON_WRITE
4424 ( sflags & SVf_IsCOW
4425 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4426 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4427 && SvTYPE(sstr) >= SVt_PVIV && len
4430 #elif defined(PERL_NEW_COPY_ON_WRITE)
4433 ( (GE_COWBUF_THRESHOLD(cur) || SvLEN(dstr) < cur+1)
4434 /* If this is a regular (non-hek) COW, only so
4435 many COW "copies" are possible. */
4436 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4437 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4438 && !(SvFLAGS(dstr) & SVf_BREAK)
4439 && GE_COW_THRESHOLD(cur) && cur+1 < len
4440 && (GE_COWBUF_THRESHOLD(cur) || SvLEN(dstr) < cur+1)
4444 && !(SvFLAGS(dstr) & SVf_BREAK)
4447 /* Either it's a shared hash key, or it's suitable for
4450 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4455 if (!(sflags & SVf_IsCOW)) {
4457 # ifdef PERL_OLD_COPY_ON_WRITE
4458 /* Make the source SV into a loop of 1.
4459 (about to become 2) */
4460 SV_COW_NEXT_SV_SET(sstr, sstr);
4462 CowREFCNT(sstr) = 0;
4466 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4472 # ifdef PERL_OLD_COPY_ON_WRITE
4473 assert (SvTYPE(dstr) >= SVt_PVIV);
4474 /* SvIsCOW_normal */
4475 /* splice us in between source and next-after-source. */
4476 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4477 SV_COW_NEXT_SV_SET(sstr, dstr);
4479 if (sflags & SVf_IsCOW) {
4484 SvPV_set(dstr, SvPVX_mutable(sstr));
4489 /* SvIsCOW_shared_hash */
4490 DEBUG_C(PerlIO_printf(Perl_debug_log,
4491 "Copy on write: Sharing hash\n"));
4493 assert (SvTYPE(dstr) >= SVt_PV);
4495 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4497 SvLEN_set(dstr, len);
4498 SvCUR_set(dstr, cur);
4501 /* Failed the swipe test, and we cannot do copy-on-write either.
4502 Have to copy the string. */
4503 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4504 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4505 SvCUR_set(dstr, cur);
4506 *SvEND(dstr) = '\0';
4508 if (sflags & SVp_NOK) {
4509 SvNV_set(dstr, SvNVX(sstr));
4511 if (sflags & SVp_IOK) {
4512 SvIV_set(dstr, SvIVX(sstr));
4513 /* Must do this otherwise some other overloaded use of 0x80000000
4514 gets confused. I guess SVpbm_VALID */
4515 if (sflags & SVf_IVisUV)
4518 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4520 const MAGIC * const smg = SvVSTRING_mg(sstr);
4522 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4523 smg->mg_ptr, smg->mg_len);
4524 SvRMAGICAL_on(dstr);
4528 else if (sflags & (SVp_IOK|SVp_NOK)) {
4529 (void)SvOK_off(dstr);
4530 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4531 if (sflags & SVp_IOK) {
4532 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4533 SvIV_set(dstr, SvIVX(sstr));
4535 if (sflags & SVp_NOK) {
4536 SvNV_set(dstr, SvNVX(sstr));
4540 if (isGV_with_GP(sstr)) {
4541 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4544 (void)SvOK_off(dstr);
4546 if (SvTAINTED(sstr))
4551 =for apidoc sv_setsv_mg
4553 Like C<sv_setsv>, but also handles 'set' magic.
4559 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4561 PERL_ARGS_ASSERT_SV_SETSV_MG;
4563 sv_setsv(dstr,sstr);
4568 # ifdef PERL_OLD_COPY_ON_WRITE
4569 # define SVt_COW SVt_PVIV
4571 # define SVt_COW SVt_PV
4574 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4576 STRLEN cur = SvCUR(sstr);
4577 STRLEN len = SvLEN(sstr);
4579 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4580 const bool already = cBOOL(SvIsCOW(sstr));
4583 PERL_ARGS_ASSERT_SV_SETSV_COW;
4586 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4587 (void*)sstr, (void*)dstr);
4594 if (SvTHINKFIRST(dstr))
4595 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4596 else if (SvPVX_const(dstr))
4597 Safefree(SvPVX_mutable(dstr));
4601 SvUPGRADE(dstr, SVt_COW);
4603 assert (SvPOK(sstr));
4604 assert (SvPOKp(sstr));
4605 # ifdef PERL_OLD_COPY_ON_WRITE
4606 assert (!SvIOK(sstr));
4607 assert (!SvIOKp(sstr));
4608 assert (!SvNOK(sstr));
4609 assert (!SvNOKp(sstr));
4612 if (SvIsCOW(sstr)) {
4614 if (SvLEN(sstr) == 0) {
4615 /* source is a COW shared hash key. */
4616 DEBUG_C(PerlIO_printf(Perl_debug_log,
4617 "Fast copy on write: Sharing hash\n"));
4618 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4621 # ifdef PERL_OLD_COPY_ON_WRITE
4622 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4624 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4625 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4628 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4629 SvUPGRADE(sstr, SVt_COW);
4631 DEBUG_C(PerlIO_printf(Perl_debug_log,
4632 "Fast copy on write: Converting sstr to COW\n"));
4633 # ifdef PERL_OLD_COPY_ON_WRITE
4634 SV_COW_NEXT_SV_SET(dstr, sstr);
4636 CowREFCNT(sstr) = 0;
4639 # ifdef PERL_OLD_COPY_ON_WRITE
4640 SV_COW_NEXT_SV_SET(sstr, dstr);
4642 # ifdef PERL_DEBUG_READONLY_COW
4643 if (already) sv_buf_to_rw(sstr);
4647 new_pv = SvPVX_mutable(sstr);
4651 SvPV_set(dstr, new_pv);
4652 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4655 SvLEN_set(dstr, len);
4656 SvCUR_set(dstr, cur);
4665 =for apidoc sv_setpvn
4667 Copies a string into an SV. The C<len> parameter indicates the number of
4668 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4669 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4675 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4680 PERL_ARGS_ASSERT_SV_SETPVN;
4682 SV_CHECK_THINKFIRST_COW_DROP(sv);
4688 /* len is STRLEN which is unsigned, need to copy to signed */
4691 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4694 SvUPGRADE(sv, SVt_PV);
4696 dptr = SvGROW(sv, len + 1);
4697 Move(ptr,dptr,len,char);
4700 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4702 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4706 =for apidoc sv_setpvn_mg
4708 Like C<sv_setpvn>, but also handles 'set' magic.
4714 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4716 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4718 sv_setpvn(sv,ptr,len);
4723 =for apidoc sv_setpv
4725 Copies a string into an SV. The string must be null-terminated. Does not
4726 handle 'set' magic. See C<sv_setpv_mg>.
4732 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4737 PERL_ARGS_ASSERT_SV_SETPV;
4739 SV_CHECK_THINKFIRST_COW_DROP(sv);
4745 SvUPGRADE(sv, SVt_PV);
4747 SvGROW(sv, len + 1);
4748 Move(ptr,SvPVX(sv),len+1,char);
4750 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4752 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4756 =for apidoc sv_setpv_mg
4758 Like C<sv_setpv>, but also handles 'set' magic.
4764 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4766 PERL_ARGS_ASSERT_SV_SETPV_MG;
4773 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4777 PERL_ARGS_ASSERT_SV_SETHEK;
4783 if (HEK_LEN(hek) == HEf_SVKEY) {
4784 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4787 const int flags = HEK_FLAGS(hek);
4788 if (flags & HVhek_WASUTF8) {
4789 STRLEN utf8_len = HEK_LEN(hek);
4790 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4791 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4794 } else if (flags & HVhek_UNSHARED) {
4795 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4798 else SvUTF8_off(sv);
4802 SV_CHECK_THINKFIRST_COW_DROP(sv);
4803 SvUPGRADE(sv, SVt_PV);
4805 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4806 SvCUR_set(sv, HEK_LEN(hek));
4812 else SvUTF8_off(sv);
4820 =for apidoc sv_usepvn_flags
4822 Tells an SV to use C<ptr> to find its string value. Normally the
4823 string is stored inside the SV but sv_usepvn allows the SV to use an
4824 outside string. The C<ptr> should point to memory that was allocated
4825 by C<malloc>. It must be the start of a mallocked block
4826 of memory, and not a pointer to the middle of it. The
4827 string length, C<len>, must be supplied. By default
4828 this function will realloc (i.e. move) the memory pointed to by C<ptr>,
4829 so that pointer should not be freed or used by the programmer after
4830 giving it to sv_usepvn, and neither should any pointers from "behind"
4831 that pointer (e.g. ptr + 1) be used.
4833 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4834 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be NUL, and the realloc
4835 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4836 C<len>, and already meets the requirements for storing in C<SvPVX>).
4842 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4847 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4849 SV_CHECK_THINKFIRST_COW_DROP(sv);
4850 SvUPGRADE(sv, SVt_PV);
4853 if (flags & SV_SMAGIC)
4857 if (SvPVX_const(sv))
4861 if (flags & SV_HAS_TRAILING_NUL)
4862 assert(ptr[len] == '\0');
4865 allocate = (flags & SV_HAS_TRAILING_NUL)
4867 #ifdef Perl_safesysmalloc_size
4870 PERL_STRLEN_ROUNDUP(len + 1);
4872 if (flags & SV_HAS_TRAILING_NUL) {
4873 /* It's long enough - do nothing.
4874 Specifically Perl_newCONSTSUB is relying on this. */
4877 /* Force a move to shake out bugs in callers. */
4878 char *new_ptr = (char*)safemalloc(allocate);
4879 Copy(ptr, new_ptr, len, char);
4880 PoisonFree(ptr,len,char);
4884 ptr = (char*) saferealloc (ptr, allocate);
4887 #ifdef Perl_safesysmalloc_size
4888 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4890 SvLEN_set(sv, allocate);
4894 if (!(flags & SV_HAS_TRAILING_NUL)) {
4897 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4899 if (flags & SV_SMAGIC)
4903 #ifdef PERL_OLD_COPY_ON_WRITE
4904 /* Need to do this *after* making the SV normal, as we need the buffer
4905 pointer to remain valid until after we've copied it. If we let go too early,
4906 another thread could invalidate it by unsharing last of the same hash key
4907 (which it can do by means other than releasing copy-on-write Svs)
4908 or by changing the other copy-on-write SVs in the loop. */
4910 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
4912 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4914 { /* this SV was SvIsCOW_normal(sv) */
4915 /* we need to find the SV pointing to us. */
4916 SV *current = SV_COW_NEXT_SV(after);
4918 if (current == sv) {
4919 /* The SV we point to points back to us (there were only two of us
4921 Hence other SV is no longer copy on write either. */
4923 sv_buf_to_rw(after);
4925 /* We need to follow the pointers around the loop. */
4927 while ((next = SV_COW_NEXT_SV(current)) != sv) {
4930 /* don't loop forever if the structure is bust, and we have
4931 a pointer into a closed loop. */
4932 assert (current != after);
4933 assert (SvPVX_const(current) == pvx);
4935 /* Make the SV before us point to the SV after us. */
4936 SV_COW_NEXT_SV_SET(current, after);
4942 =for apidoc sv_force_normal_flags
4944 Undo various types of fakery on an SV, where fakery means
4945 "more than" a string: if the PV is a shared string, make
4946 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
4947 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
4948 we do the copy, and is also used locally; if this is a
4949 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
4950 then a copy-on-write scalar drops its PV buffer (if any) and becomes
4951 SvPOK_off rather than making a copy. (Used where this
4952 scalar is about to be set to some other value.) In addition,
4953 the C<flags> parameter gets passed to C<sv_unref_flags()>
4954 when unreffing. C<sv_force_normal> calls this function
4955 with flags set to 0.
4957 This function is expected to be used to signal to perl that this SV is
4958 about to be written to, and any extra book-keeping needs to be taken care
4959 of. Hence, it croaks on read-only values.
4965 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
4969 assert(SvIsCOW(sv));
4972 const char * const pvx = SvPVX_const(sv);
4973 const STRLEN len = SvLEN(sv);
4974 const STRLEN cur = SvCUR(sv);
4975 # ifdef PERL_OLD_COPY_ON_WRITE
4976 /* next COW sv in the loop. If len is 0 then this is a shared-hash
4977 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
4978 we'll fail an assertion. */
4979 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
4983 PerlIO_printf(Perl_debug_log,
4984 "Copy on write: Force normal %ld\n",
4989 # ifdef PERL_NEW_COPY_ON_WRITE
4990 if (len && CowREFCNT(sv) == 0)
4991 /* We own the buffer ourselves. */
4997 /* This SV doesn't own the buffer, so need to Newx() a new one: */
4998 # ifdef PERL_NEW_COPY_ON_WRITE
4999 /* Must do this first, since the macro uses SvPVX. */
5008 if (flags & SV_COW_DROP_PV) {
5009 /* OK, so we don't need to copy our buffer. */
5012 SvGROW(sv, cur + 1);
5013 Move(pvx,SvPVX(sv),cur,char);
5018 # ifdef PERL_OLD_COPY_ON_WRITE
5019 sv_release_COW(sv, pvx, next);
5022 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5029 const char * const pvx = SvPVX_const(sv);
5030 const STRLEN len = SvCUR(sv);
5034 if (flags & SV_COW_DROP_PV) {
5035 /* OK, so we don't need to copy our buffer. */
5038 SvGROW(sv, len + 1);
5039 Move(pvx,SvPVX(sv),len,char);
5042 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5048 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5050 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5053 Perl_croak_no_modify();
5054 else if (SvIsCOW(sv))
5055 S_sv_uncow(aTHX_ sv, flags);
5057 sv_unref_flags(sv, flags);
5058 else if (SvFAKE(sv) && isGV_with_GP(sv))
5059 sv_unglob(sv, flags);
5060 else if (SvFAKE(sv) && isREGEXP(sv)) {
5061 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5062 to sv_unglob. We only need it here, so inline it. */
5063 const bool islv = SvTYPE(sv) == SVt_PVLV;
5064 const svtype new_type =
5065 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5066 SV *const temp = newSV_type(new_type);
5067 regexp *const temp_p = ReANY((REGEXP *)sv);
5069 if (new_type == SVt_PVMG) {
5070 SvMAGIC_set(temp, SvMAGIC(sv));
5071 SvMAGIC_set(sv, NULL);
5072 SvSTASH_set(temp, SvSTASH(sv));
5073 SvSTASH_set(sv, NULL);
5075 if (!islv) SvCUR_set(temp, SvCUR(sv));
5076 /* Remember that SvPVX is in the head, not the body. But
5077 RX_WRAPPED is in the body. */
5078 assert(ReANY((REGEXP *)sv)->mother_re);
5079 /* Their buffer is already owned by someone else. */
5080 if (flags & SV_COW_DROP_PV) {
5081 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5082 zeroed body. For SVt_PVLV, it should have been set to 0
5083 before turning into a regexp. */
5084 assert(!SvLEN(islv ? sv : temp));
5085 sv->sv_u.svu_pv = 0;
5088 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5089 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5093 /* Now swap the rest of the bodies. */
5097 SvFLAGS(sv) &= ~SVTYPEMASK;
5098 SvFLAGS(sv) |= new_type;
5099 SvANY(sv) = SvANY(temp);
5102 SvFLAGS(temp) &= ~(SVTYPEMASK);
5103 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5104 SvANY(temp) = temp_p;
5105 temp->sv_u.svu_rx = (regexp *)temp_p;
5107 SvREFCNT_dec_NN(temp);
5109 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5115 Efficient removal of characters from the beginning of the string buffer.
5116 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5117 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5118 character of the adjusted string. Uses the "OOK hack". On return, only
5119 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5121 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5122 refer to the same chunk of data.
5124 The unfortunate similarity of this function's name to that of Perl's C<chop>
5125 operator is strictly coincidental. This function works from the left;
5126 C<chop> works from the right.
5132 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5143 PERL_ARGS_ASSERT_SV_CHOP;
5145 if (!ptr || !SvPOKp(sv))
5147 delta = ptr - SvPVX_const(sv);
5149 /* Nothing to do. */
5152 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5153 if (delta > max_delta)
5154 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5155 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5156 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5157 SV_CHECK_THINKFIRST(sv);
5158 SvPOK_only_UTF8(sv);
5161 if (!SvLEN(sv)) { /* make copy of shared string */
5162 const char *pvx = SvPVX_const(sv);
5163 const STRLEN len = SvCUR(sv);
5164 SvGROW(sv, len + 1);
5165 Move(pvx,SvPVX(sv),len,char);
5171 SvOOK_offset(sv, old_delta);
5173 SvLEN_set(sv, SvLEN(sv) - delta);
5174 SvCUR_set(sv, SvCUR(sv) - delta);
5175 SvPV_set(sv, SvPVX(sv) + delta);
5177 p = (U8 *)SvPVX_const(sv);
5180 /* how many bytes were evacuated? we will fill them with sentinel
5181 bytes, except for the part holding the new offset of course. */
5184 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5186 assert(evacn <= delta + old_delta);
5190 /* This sets 'delta' to the accumulated value of all deltas so far */
5194 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5195 * the string; otherwise store a 0 byte there and store 'delta' just prior
5196 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5197 * portion of the chopped part of the string */
5198 if (delta < 0x100) {
5202 p -= sizeof(STRLEN);
5203 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5207 /* Fill the preceding buffer with sentinals to verify that no-one is
5217 =for apidoc sv_catpvn
5219 Concatenates the string onto the end of the string which is in the SV. The
5220 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5221 status set, then the bytes appended should be valid UTF-8.
5222 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5224 =for apidoc sv_catpvn_flags
5226 Concatenates the string onto the end of the string which is in the SV. The
5227 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5228 status set, then the bytes appended should be valid UTF-8.
5229 If C<flags> has the C<SV_SMAGIC> bit set, will
5230 C<mg_set> on C<dsv> afterwards if appropriate.
5231 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5232 in terms of this function.
5238 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5242 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5244 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5245 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5247 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5248 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5249 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5252 else SvGROW(dsv, dlen + slen + 1);
5254 sstr = SvPVX_const(dsv);
5255 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5256 SvCUR_set(dsv, SvCUR(dsv) + slen);
5259 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5260 const char * const send = sstr + slen;
5263 /* Something this code does not account for, which I think is
5264 impossible; it would require the same pv to be treated as
5265 bytes *and* utf8, which would indicate a bug elsewhere. */
5266 assert(sstr != dstr);
5268 SvGROW(dsv, dlen + slen * 2 + 1);
5269 d = (U8 *)SvPVX(dsv) + dlen;
5271 while (sstr < send) {
5272 append_utf8_from_native_byte(*sstr, &d);
5275 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5278 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5280 if (flags & SV_SMAGIC)
5285 =for apidoc sv_catsv
5287 Concatenates the string from SV C<ssv> onto the end of the string in SV
5288 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5289 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5292 =for apidoc sv_catsv_flags
5294 Concatenates the string from SV C<ssv> onto the end of the string in SV
5295 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5296 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5297 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5298 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5299 and C<sv_catsv_mg> are implemented in terms of this function.
5304 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5308 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5312 const char *spv = SvPV_flags_const(ssv, slen, flags);
5314 if (flags & SV_GMAGIC)
5316 sv_catpvn_flags(dsv, spv, slen,
5317 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5318 if (flags & SV_SMAGIC)
5325 =for apidoc sv_catpv
5327 Concatenates the string onto the end of the string which is in the SV.
5328 If the SV has the UTF-8 status set, then the bytes appended should be
5329 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5334 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5341 PERL_ARGS_ASSERT_SV_CATPV;
5345 junk = SvPV_force(sv, tlen);
5347 SvGROW(sv, tlen + len + 1);
5349 ptr = SvPVX_const(sv);
5350 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5351 SvCUR_set(sv, SvCUR(sv) + len);
5352 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5357 =for apidoc sv_catpv_flags
5359 Concatenates the string onto the end of the string which is in the SV.
5360 If the SV has the UTF-8 status set, then the bytes appended should
5361 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5362 on the modified SV if appropriate.
5368 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5370 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5371 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5375 =for apidoc sv_catpv_mg
5377 Like C<sv_catpv>, but also handles 'set' magic.
5383 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5385 PERL_ARGS_ASSERT_SV_CATPV_MG;
5394 Creates a new SV. A non-zero C<len> parameter indicates the number of
5395 bytes of preallocated string space the SV should have. An extra byte for a
5396 trailing NUL is also reserved. (SvPOK is not set for the SV even if string
5397 space is allocated.) The reference count for the new SV is set to 1.
5399 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5400 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5401 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5402 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5403 modules supporting older perls.
5409 Perl_newSV(pTHX_ const STRLEN len)
5416 sv_upgrade(sv, SVt_PV);
5417 SvGROW(sv, len + 1);
5422 =for apidoc sv_magicext
5424 Adds magic to an SV, upgrading it if necessary. Applies the
5425 supplied vtable and returns a pointer to the magic added.
5427 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5428 In particular, you can add magic to SvREADONLY SVs, and add more than
5429 one instance of the same 'how'.
5431 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5432 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5433 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5434 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5436 (This is now used as a subroutine by C<sv_magic>.)
5441 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5442 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5447 PERL_ARGS_ASSERT_SV_MAGICEXT;
5449 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5451 SvUPGRADE(sv, SVt_PVMG);
5452 Newxz(mg, 1, MAGIC);
5453 mg->mg_moremagic = SvMAGIC(sv);
5454 SvMAGIC_set(sv, mg);
5456 /* Sometimes a magic contains a reference loop, where the sv and
5457 object refer to each other. To prevent a reference loop that
5458 would prevent such objects being freed, we look for such loops
5459 and if we find one we avoid incrementing the object refcount.
5461 Note we cannot do this to avoid self-tie loops as intervening RV must
5462 have its REFCNT incremented to keep it in existence.
5465 if (!obj || obj == sv ||
5466 how == PERL_MAGIC_arylen ||
5467 how == PERL_MAGIC_symtab ||
5468 (SvTYPE(obj) == SVt_PVGV &&
5469 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5470 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5471 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5476 mg->mg_obj = SvREFCNT_inc_simple(obj);
5477 mg->mg_flags |= MGf_REFCOUNTED;
5480 /* Normal self-ties simply pass a null object, and instead of
5481 using mg_obj directly, use the SvTIED_obj macro to produce a
5482 new RV as needed. For glob "self-ties", we are tieing the PVIO
5483 with an RV obj pointing to the glob containing the PVIO. In
5484 this case, to avoid a reference loop, we need to weaken the
5488 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5489 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5495 mg->mg_len = namlen;
5498 mg->mg_ptr = savepvn(name, namlen);
5499 else if (namlen == HEf_SVKEY) {
5500 /* Yes, this is casting away const. This is only for the case of
5501 HEf_SVKEY. I think we need to document this aberation of the
5502 constness of the API, rather than making name non-const, as
5503 that change propagating outwards a long way. */
5504 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5506 mg->mg_ptr = (char *) name;
5508 mg->mg_virtual = (MGVTBL *) vtable;
5515 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5517 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5518 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5519 /* This sv is only a delegate. //g magic must be attached to
5524 #ifdef PERL_OLD_COPY_ON_WRITE
5526 sv_force_normal_flags(sv, 0);
5528 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5529 &PL_vtbl_mglob, 0, 0);
5533 =for apidoc sv_magic
5535 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5536 necessary, then adds a new magic item of type C<how> to the head of the
5539 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5540 handling of the C<name> and C<namlen> arguments.
5542 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5543 to add more than one instance of the same 'how'.
5549 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5550 const char *const name, const I32 namlen)
5553 const MGVTBL *vtable;
5556 unsigned int vtable_index;
5558 PERL_ARGS_ASSERT_SV_MAGIC;
5560 if (how < 0 || (unsigned)how > C_ARRAY_LENGTH(PL_magic_data)
5561 || ((flags = PL_magic_data[how]),
5562 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5563 > magic_vtable_max))
5564 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5566 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5567 Useful for attaching extension internal data to perl vars.
5568 Note that multiple extensions may clash if magical scalars
5569 etc holding private data from one are passed to another. */
5571 vtable = (vtable_index == magic_vtable_max)
5572 ? NULL : PL_magic_vtables + vtable_index;
5574 #ifdef PERL_OLD_COPY_ON_WRITE
5576 sv_force_normal_flags(sv, 0);
5578 if (SvREADONLY(sv)) {
5580 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5583 Perl_croak_no_modify();
5586 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5587 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5588 /* sv_magic() refuses to add a magic of the same 'how' as an
5591 if (how == PERL_MAGIC_taint)
5597 /* Force pos to be stored as characters, not bytes. */
5598 if (SvMAGICAL(sv) && DO_UTF8(sv)
5599 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5601 && mg->mg_flags & MGf_BYTES) {
5602 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5604 mg->mg_flags &= ~MGf_BYTES;
5607 /* Rest of work is done else where */
5608 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5611 case PERL_MAGIC_taint:
5614 case PERL_MAGIC_ext:
5615 case PERL_MAGIC_dbfile:
5622 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5629 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5631 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5632 for (mg = *mgp; mg; mg = *mgp) {
5633 const MGVTBL* const virt = mg->mg_virtual;
5634 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5635 *mgp = mg->mg_moremagic;
5636 if (virt && virt->svt_free)
5637 virt->svt_free(aTHX_ sv, mg);
5638 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5640 Safefree(mg->mg_ptr);
5641 else if (mg->mg_len == HEf_SVKEY)
5642 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5643 else if (mg->mg_type == PERL_MAGIC_utf8)
5644 Safefree(mg->mg_ptr);
5646 if (mg->mg_flags & MGf_REFCOUNTED)
5647 SvREFCNT_dec(mg->mg_obj);
5651 mgp = &mg->mg_moremagic;
5654 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5655 mg_magical(sv); /* else fix the flags now */
5659 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5665 =for apidoc sv_unmagic
5667 Removes all magic of type C<type> from an SV.
5673 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5675 PERL_ARGS_ASSERT_SV_UNMAGIC;
5676 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5680 =for apidoc sv_unmagicext
5682 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5688 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5690 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5691 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5695 =for apidoc sv_rvweaken
5697 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5698 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5699 push a back-reference to this RV onto the array of backreferences
5700 associated with that magic. If the RV is magical, set magic will be
5701 called after the RV is cleared.
5707 Perl_sv_rvweaken(pTHX_ SV *const sv)
5711 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5713 if (!SvOK(sv)) /* let undefs pass */
5716 Perl_croak(aTHX_ "Can't weaken a nonreference");
5717 else if (SvWEAKREF(sv)) {
5718 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5721 else if (SvREADONLY(sv)) croak_no_modify();
5723 Perl_sv_add_backref(aTHX_ tsv, sv);
5725 SvREFCNT_dec_NN(tsv);
5729 /* Give tsv backref magic if it hasn't already got it, then push a
5730 * back-reference to sv onto the array associated with the backref magic.
5732 * As an optimisation, if there's only one backref and it's not an AV,
5733 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5734 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5738 /* A discussion about the backreferences array and its refcount:
5740 * The AV holding the backreferences is pointed to either as the mg_obj of
5741 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5742 * xhv_backreferences field. The array is created with a refcount
5743 * of 2. This means that if during global destruction the array gets
5744 * picked on before its parent to have its refcount decremented by the
5745 * random zapper, it won't actually be freed, meaning it's still there for
5746 * when its parent gets freed.
5748 * When the parent SV is freed, the extra ref is killed by
5749 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5750 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5752 * When a single backref SV is stored directly, it is not reference
5757 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5764 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5766 /* find slot to store array or singleton backref */
5768 if (SvTYPE(tsv) == SVt_PVHV) {
5769 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5772 mg = mg_find(tsv, PERL_MAGIC_backref);
5774 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5775 svp = &(mg->mg_obj);
5778 /* create or retrieve the array */
5780 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5781 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5785 mg->mg_flags |= MGf_REFCOUNTED;
5788 SvREFCNT_inc_simple_void_NN(av);
5789 /* av now has a refcnt of 2; see discussion above */
5790 av_extend(av, *svp ? 2 : 1);
5792 /* move single existing backref to the array */
5793 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5798 av = MUTABLE_AV(*svp);
5800 /* optimisation: store single backref directly in HvAUX or mg_obj */
5804 assert(SvTYPE(av) == SVt_PVAV);
5805 if (AvFILLp(av) >= AvMAX(av)) {
5806 av_extend(av, AvFILLp(av)+1);
5809 /* push new backref */
5810 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5813 /* delete a back-reference to ourselves from the backref magic associated
5814 * with the SV we point to.
5818 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5823 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5825 if (SvTYPE(tsv) == SVt_PVHV) {
5827 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5829 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5830 /* It's possible for the the last (strong) reference to tsv to have
5831 become freed *before* the last thing holding a weak reference.
5832 If both survive longer than the backreferences array, then when
5833 the referent's reference count drops to 0 and it is freed, it's
5834 not able to chase the backreferences, so they aren't NULLed.
5836 For example, a CV holds a weak reference to its stash. If both the
5837 CV and the stash survive longer than the backreferences array,
5838 and the CV gets picked for the SvBREAK() treatment first,
5839 *and* it turns out that the stash is only being kept alive because
5840 of an our variable in the pad of the CV, then midway during CV
5841 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5842 It ends up pointing to the freed HV. Hence it's chased in here, and
5843 if this block wasn't here, it would hit the !svp panic just below.
5845 I don't believe that "better" destruction ordering is going to help
5846 here - during global destruction there's always going to be the
5847 chance that something goes out of order. We've tried to make it
5848 foolproof before, and it only resulted in evolutionary pressure on
5849 fools. Which made us look foolish for our hubris. :-(
5855 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5856 svp = mg ? &(mg->mg_obj) : NULL;
5860 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5862 /* It's possible that sv is being freed recursively part way through the
5863 freeing of tsv. If this happens, the backreferences array of tsv has
5864 already been freed, and so svp will be NULL. If this is the case,
5865 we should not panic. Instead, nothing needs doing, so return. */
5866 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
5868 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
5869 *svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
5872 if (SvTYPE(*svp) == SVt_PVAV) {
5876 AV * const av = (AV*)*svp;
5878 assert(!SvIS_FREED(av));
5882 /* for an SV with N weak references to it, if all those
5883 * weak refs are deleted, then sv_del_backref will be called
5884 * N times and O(N^2) compares will be done within the backref
5885 * array. To ameliorate this potential slowness, we:
5886 * 1) make sure this code is as tight as possible;
5887 * 2) when looking for SV, look for it at both the head and tail of the
5888 * array first before searching the rest, since some create/destroy
5889 * patterns will cause the backrefs to be freed in order.
5896 SV **p = &svp[fill];
5897 SV *const topsv = *p;
5904 /* We weren't the last entry.
5905 An unordered list has this property that you
5906 can take the last element off the end to fill
5907 the hole, and it's still an unordered list :-)
5913 break; /* should only be one */
5920 AvFILLp(av) = fill-1;
5922 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
5923 /* freed AV; skip */
5926 /* optimisation: only a single backref, stored directly */
5928 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p", *svp, sv);
5935 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
5941 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
5946 /* after multiple passes through Perl_sv_clean_all() for a thingy
5947 * that has badly leaked, the backref array may have gotten freed,
5948 * since we only protect it against 1 round of cleanup */
5949 if (SvIS_FREED(av)) {
5950 if (PL_in_clean_all) /* All is fair */
5953 "panic: magic_killbackrefs (freed backref AV/SV)");
5957 is_array = (SvTYPE(av) == SVt_PVAV);
5959 assert(!SvIS_FREED(av));
5962 last = svp + AvFILLp(av);
5965 /* optimisation: only a single backref, stored directly */
5971 while (svp <= last) {
5973 SV *const referrer = *svp;
5974 if (SvWEAKREF(referrer)) {
5975 /* XXX Should we check that it hasn't changed? */
5976 assert(SvROK(referrer));
5977 SvRV_set(referrer, 0);
5979 SvWEAKREF_off(referrer);
5980 SvSETMAGIC(referrer);
5981 } else if (SvTYPE(referrer) == SVt_PVGV ||
5982 SvTYPE(referrer) == SVt_PVLV) {
5983 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
5984 /* You lookin' at me? */
5985 assert(GvSTASH(referrer));
5986 assert(GvSTASH(referrer) == (const HV *)sv);
5987 GvSTASH(referrer) = 0;
5988 } else if (SvTYPE(referrer) == SVt_PVCV ||
5989 SvTYPE(referrer) == SVt_PVFM) {
5990 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
5991 /* You lookin' at me? */
5992 assert(CvSTASH(referrer));
5993 assert(CvSTASH(referrer) == (const HV *)sv);
5994 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
5997 assert(SvTYPE(sv) == SVt_PVGV);
5998 /* You lookin' at me? */
5999 assert(CvGV(referrer));
6000 assert(CvGV(referrer) == (const GV *)sv);
6001 anonymise_cv_maybe(MUTABLE_GV(sv),
6002 MUTABLE_CV(referrer));
6007 "panic: magic_killbackrefs (flags=%"UVxf")",
6008 (UV)SvFLAGS(referrer));
6019 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6025 =for apidoc sv_insert
6027 Inserts a string at the specified offset/length within the SV. Similar to
6028 the Perl substr() function. Handles get magic.
6030 =for apidoc sv_insert_flags
6032 Same as C<sv_insert>, but the extra C<flags> are passed to the
6033 C<SvPV_force_flags> that applies to C<bigstr>.
6039 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6046 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6049 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6052 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6053 SvPV_force_flags(bigstr, curlen, flags);
6054 (void)SvPOK_only_UTF8(bigstr);
6055 if (offset + len > curlen) {
6056 SvGROW(bigstr, offset+len+1);
6057 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6058 SvCUR_set(bigstr, offset+len);
6062 i = littlelen - len;
6063 if (i > 0) { /* string might grow */
6064 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6065 mid = big + offset + len;
6066 midend = bigend = big + SvCUR(bigstr);
6069 while (midend > mid) /* shove everything down */
6070 *--bigend = *--midend;
6071 Move(little,big+offset,littlelen,char);
6072 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6077 Move(little,SvPVX(bigstr)+offset,len,char);
6082 big = SvPVX(bigstr);
6085 bigend = big + SvCUR(bigstr);
6087 if (midend > bigend)
6088 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6091 if (mid - big > bigend - midend) { /* faster to shorten from end */
6093 Move(little, mid, littlelen,char);
6096 i = bigend - midend;
6098 Move(midend, mid, i,char);
6102 SvCUR_set(bigstr, mid - big);
6104 else if ((i = mid - big)) { /* faster from front */
6105 midend -= littlelen;
6107 Move(big, midend - i, i, char);
6108 sv_chop(bigstr,midend-i);
6110 Move(little, mid, littlelen,char);
6112 else if (littlelen) {
6113 midend -= littlelen;
6114 sv_chop(bigstr,midend);
6115 Move(little,midend,littlelen,char);
6118 sv_chop(bigstr,midend);
6124 =for apidoc sv_replace
6126 Make the first argument a copy of the second, then delete the original.
6127 The target SV physically takes over ownership of the body of the source SV
6128 and inherits its flags; however, the target keeps any magic it owns,
6129 and any magic in the source is discarded.
6130 Note that this is a rather specialist SV copying operation; most of the
6131 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6137 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6140 const U32 refcnt = SvREFCNT(sv);
6142 PERL_ARGS_ASSERT_SV_REPLACE;
6144 SV_CHECK_THINKFIRST_COW_DROP(sv);
6145 if (SvREFCNT(nsv) != 1) {
6146 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6147 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6149 if (SvMAGICAL(sv)) {
6153 sv_upgrade(nsv, SVt_PVMG);
6154 SvMAGIC_set(nsv, SvMAGIC(sv));
6155 SvFLAGS(nsv) |= SvMAGICAL(sv);
6157 SvMAGIC_set(sv, NULL);
6161 assert(!SvREFCNT(sv));
6162 #ifdef DEBUG_LEAKING_SCALARS
6163 sv->sv_flags = nsv->sv_flags;
6164 sv->sv_any = nsv->sv_any;
6165 sv->sv_refcnt = nsv->sv_refcnt;
6166 sv->sv_u = nsv->sv_u;
6168 StructCopy(nsv,sv,SV);
6170 if(SvTYPE(sv) == SVt_IV) {
6172 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6176 #ifdef PERL_OLD_COPY_ON_WRITE
6177 if (SvIsCOW_normal(nsv)) {
6178 /* We need to follow the pointers around the loop to make the
6179 previous SV point to sv, rather than nsv. */
6182 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6185 assert(SvPVX_const(current) == SvPVX_const(nsv));
6187 /* Make the SV before us point to the SV after us. */
6189 PerlIO_printf(Perl_debug_log, "previous is\n");
6191 PerlIO_printf(Perl_debug_log,
6192 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6193 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6195 SV_COW_NEXT_SV_SET(current, sv);
6198 SvREFCNT(sv) = refcnt;
6199 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6204 /* We're about to free a GV which has a CV that refers back to us.
6205 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6209 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6214 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6217 assert(SvREFCNT(gv) == 0);
6218 assert(isGV(gv) && isGV_with_GP(gv));
6220 assert(!CvANON(cv));
6221 assert(CvGV(cv) == gv);
6222 assert(!CvNAMED(cv));
6224 /* will the CV shortly be freed by gp_free() ? */
6225 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6226 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6230 /* if not, anonymise: */
6231 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6232 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6233 : newSVpvn_flags( "__ANON__", 8, 0 );
6234 sv_catpvs(gvname, "::__ANON__");
6235 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6236 SvREFCNT_dec_NN(gvname);
6240 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6245 =for apidoc sv_clear
6247 Clear an SV: call any destructors, free up any memory used by the body,
6248 and free the body itself. The SV's head is I<not> freed, although
6249 its type is set to all 1's so that it won't inadvertently be assumed
6250 to be live during global destruction etc.
6251 This function should only be called when REFCNT is zero. Most of the time
6252 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6259 Perl_sv_clear(pTHX_ SV *const orig_sv)
6264 const struct body_details *sv_type_details;
6270 PERL_ARGS_ASSERT_SV_CLEAR;
6272 /* within this loop, sv is the SV currently being freed, and
6273 * iter_sv is the most recent AV or whatever that's being iterated
6274 * over to provide more SVs */
6280 assert(SvREFCNT(sv) == 0);
6281 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6283 if (type <= SVt_IV) {
6284 /* See the comment in sv.h about the collusion between this
6285 * early return and the overloading of the NULL slots in the
6289 SvFLAGS(sv) &= SVf_BREAK;
6290 SvFLAGS(sv) |= SVTYPEMASK;
6294 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6296 if (type >= SVt_PVMG) {
6298 if (!curse(sv, 1)) goto get_next_sv;
6299 type = SvTYPE(sv); /* destructor may have changed it */
6301 /* Free back-references before magic, in case the magic calls
6302 * Perl code that has weak references to sv. */
6303 if (type == SVt_PVHV) {
6304 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6308 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6309 SvREFCNT_dec(SvOURSTASH(sv));
6311 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6312 assert(!SvMAGICAL(sv));
6313 } else if (SvMAGIC(sv)) {
6314 /* Free back-references before other types of magic. */
6315 sv_unmagic(sv, PERL_MAGIC_backref);
6319 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6320 SvREFCNT_dec(SvSTASH(sv));
6323 /* case SVt_INVLIST: */
6326 IoIFP(sv) != PerlIO_stdin() &&
6327 IoIFP(sv) != PerlIO_stdout() &&
6328 IoIFP(sv) != PerlIO_stderr() &&
6329 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6331 io_close(MUTABLE_IO(sv), FALSE);
6333 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6334 PerlDir_close(IoDIRP(sv));
6335 IoDIRP(sv) = (DIR*)NULL;
6336 Safefree(IoTOP_NAME(sv));
6337 Safefree(IoFMT_NAME(sv));
6338 Safefree(IoBOTTOM_NAME(sv));
6339 if ((const GV *)sv == PL_statgv)
6343 /* FIXME for plugins */
6345 pregfree2((REGEXP*) sv);
6349 cv_undef(MUTABLE_CV(sv));
6350 /* If we're in a stash, we don't own a reference to it.
6351 * However it does have a back reference to us, which needs to
6353 if ((stash = CvSTASH(sv)))
6354 sv_del_backref(MUTABLE_SV(stash), sv);
6357 if (PL_last_swash_hv == (const HV *)sv) {
6358 PL_last_swash_hv = NULL;
6360 if (HvTOTALKEYS((HV*)sv) > 0) {
6362 /* this statement should match the one at the beginning of
6363 * hv_undef_flags() */
6364 if ( PL_phase != PERL_PHASE_DESTRUCT
6365 && (name = HvNAME((HV*)sv)))
6367 if (PL_stashcache) {
6368 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6370 (void)hv_deletehek(PL_stashcache,
6371 HvNAME_HEK((HV*)sv), G_DISCARD);
6373 hv_name_set((HV*)sv, NULL, 0, 0);
6376 /* save old iter_sv in unused SvSTASH field */
6377 assert(!SvOBJECT(sv));
6378 SvSTASH(sv) = (HV*)iter_sv;
6381 /* save old hash_index in unused SvMAGIC field */
6382 assert(!SvMAGICAL(sv));
6383 assert(!SvMAGIC(sv));
6384 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6387 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6388 goto get_next_sv; /* process this new sv */
6390 /* free empty hash */
6391 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6392 assert(!HvARRAY((HV*)sv));
6396 AV* av = MUTABLE_AV(sv);
6397 if (PL_comppad == av) {
6401 if (AvREAL(av) && AvFILLp(av) > -1) {
6402 next_sv = AvARRAY(av)[AvFILLp(av)--];
6403 /* save old iter_sv in top-most slot of AV,
6404 * and pray that it doesn't get wiped in the meantime */
6405 AvARRAY(av)[AvMAX(av)] = iter_sv;
6407 goto get_next_sv; /* process this new sv */
6409 Safefree(AvALLOC(av));
6414 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6415 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6416 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6417 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6419 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6420 SvREFCNT_dec(LvTARG(sv));
6421 if (isREGEXP(sv)) goto freeregexp;
6423 if (isGV_with_GP(sv)) {
6424 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6425 && HvENAME_get(stash))
6426 mro_method_changed_in(stash);
6427 gp_free(MUTABLE_GV(sv));
6429 unshare_hek(GvNAME_HEK(sv));
6430 /* If we're in a stash, we don't own a reference to it.
6431 * However it does have a back reference to us, which
6432 * needs to be cleared. */
6433 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6434 sv_del_backref(MUTABLE_SV(stash), sv);
6436 /* FIXME. There are probably more unreferenced pointers to SVs
6437 * in the interpreter struct that we should check and tidy in
6438 * a similar fashion to this: */
6439 /* See also S_sv_unglob, which does the same thing. */
6440 if ((const GV *)sv == PL_last_in_gv)
6441 PL_last_in_gv = NULL;
6442 else if ((const GV *)sv == PL_statgv)
6444 else if ((const GV *)sv == PL_stderrgv)
6452 /* Don't bother with SvOOK_off(sv); as we're only going to
6456 SvOOK_offset(sv, offset);
6457 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6458 /* Don't even bother with turning off the OOK flag. */
6463 SV * const target = SvRV(sv);
6465 sv_del_backref(target, sv);
6471 else if (SvPVX_const(sv)
6472 && !(SvTYPE(sv) == SVt_PVIO
6473 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6477 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6481 # ifdef PERL_OLD_COPY_ON_WRITE
6482 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6484 if (CowREFCNT(sv)) {
6492 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6496 # ifdef PERL_OLD_COPY_ON_WRITE
6500 Safefree(SvPVX_mutable(sv));
6504 else if (SvPVX_const(sv) && SvLEN(sv)
6505 && !(SvTYPE(sv) == SVt_PVIO
6506 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6507 Safefree(SvPVX_mutable(sv));
6508 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6509 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6519 SvFLAGS(sv) &= SVf_BREAK;
6520 SvFLAGS(sv) |= SVTYPEMASK;
6522 sv_type_details = bodies_by_type + type;
6523 if (sv_type_details->arena) {
6524 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6525 &PL_body_roots[type]);
6527 else if (sv_type_details->body_size) {
6528 safefree(SvANY(sv));
6532 /* caller is responsible for freeing the head of the original sv */
6533 if (sv != orig_sv && !SvREFCNT(sv))
6536 /* grab and free next sv, if any */
6544 else if (!iter_sv) {
6546 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6547 AV *const av = (AV*)iter_sv;
6548 if (AvFILLp(av) > -1) {
6549 sv = AvARRAY(av)[AvFILLp(av)--];
6551 else { /* no more elements of current AV to free */
6554 /* restore previous value, squirrelled away */
6555 iter_sv = AvARRAY(av)[AvMAX(av)];
6556 Safefree(AvALLOC(av));
6559 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6560 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6561 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6562 /* no more elements of current HV to free */
6565 /* Restore previous values of iter_sv and hash_index,
6566 * squirrelled away */
6567 assert(!SvOBJECT(sv));
6568 iter_sv = (SV*)SvSTASH(sv);
6569 assert(!SvMAGICAL(sv));
6570 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6572 /* perl -DA does not like rubbish in SvMAGIC. */
6576 /* free any remaining detritus from the hash struct */
6577 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6578 assert(!HvARRAY((HV*)sv));
6583 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6587 if (!SvREFCNT(sv)) {
6591 if (--(SvREFCNT(sv)))
6595 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6596 "Attempt to free temp prematurely: SV 0x%"UVxf
6597 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6601 if (SvIMMORTAL(sv)) {
6602 /* make sure SvREFCNT(sv)==0 happens very seldom */
6603 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6612 /* This routine curses the sv itself, not the object referenced by sv. So
6613 sv does not have to be ROK. */
6616 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6619 PERL_ARGS_ASSERT_CURSE;
6620 assert(SvOBJECT(sv));
6622 if (PL_defstash && /* Still have a symbol table? */
6628 stash = SvSTASH(sv);
6629 assert(SvTYPE(stash) == SVt_PVHV);
6630 if (HvNAME(stash)) {
6631 CV* destructor = NULL;
6632 assert (SvOOK(stash));
6633 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6634 if (!destructor || HvMROMETA(stash)->destroy_gen
6635 != PL_sub_generation)
6638 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6639 if (gv) destructor = GvCV(gv);
6640 if (!SvOBJECT(stash))
6643 destructor ? (HV *)destructor : ((HV *)0)+1;
6644 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6648 assert(!destructor || destructor == ((CV *)0)+1
6649 || SvTYPE(destructor) == SVt_PVCV);
6650 if (destructor && destructor != ((CV *)0)+1
6651 /* A constant subroutine can have no side effects, so
6652 don't bother calling it. */
6653 && !CvCONST(destructor)
6654 /* Don't bother calling an empty destructor or one that
6655 returns immediately. */
6656 && (CvISXSUB(destructor)
6657 || (CvSTART(destructor)
6658 && (CvSTART(destructor)->op_next->op_type
6660 && (CvSTART(destructor)->op_next->op_type
6662 || CvSTART(destructor)->op_next->op_next->op_type
6668 SV* const tmpref = newRV(sv);
6669 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6671 PUSHSTACKi(PERLSI_DESTROY);
6676 call_sv(MUTABLE_SV(destructor),
6677 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6681 if(SvREFCNT(tmpref) < 2) {
6682 /* tmpref is not kept alive! */
6684 SvRV_set(tmpref, NULL);
6687 SvREFCNT_dec_NN(tmpref);
6690 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6693 if (check_refcnt && SvREFCNT(sv)) {
6694 if (PL_in_clean_objs)
6696 "DESTROY created new reference to dead object '%"HEKf"'",
6697 HEKfARG(HvNAME_HEK(stash)));
6698 /* DESTROY gave object new lease on life */
6704 HV * const stash = SvSTASH(sv);
6705 /* Curse before freeing the stash, as freeing the stash could cause
6706 a recursive call into S_curse. */
6707 SvOBJECT_off(sv); /* Curse the object. */
6708 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6709 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6715 =for apidoc sv_newref
6717 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6724 Perl_sv_newref(pTHX_ SV *const sv)
6726 PERL_UNUSED_CONTEXT;
6735 Decrement an SV's reference count, and if it drops to zero, call
6736 C<sv_clear> to invoke destructors and free up any memory used by
6737 the body; finally, deallocate the SV's head itself.
6738 Normally called via a wrapper macro C<SvREFCNT_dec>.
6744 Perl_sv_free(pTHX_ SV *const sv)
6750 /* Private helper function for SvREFCNT_dec().
6751 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6754 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6758 PERL_ARGS_ASSERT_SV_FREE2;
6760 if (LIKELY( rc == 1 )) {
6766 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6767 "Attempt to free temp prematurely: SV 0x%"UVxf
6768 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6772 if (SvIMMORTAL(sv)) {
6773 /* make sure SvREFCNT(sv)==0 happens very seldom */
6774 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6778 if (! SvREFCNT(sv)) /* may have have been resurrected */
6783 /* handle exceptional cases */
6787 if (SvFLAGS(sv) & SVf_BREAK)
6788 /* this SV's refcnt has been artificially decremented to
6789 * trigger cleanup */
6791 if (PL_in_clean_all) /* All is fair */
6793 if (SvIMMORTAL(sv)) {
6794 /* make sure SvREFCNT(sv)==0 happens very seldom */
6795 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6798 if (ckWARN_d(WARN_INTERNAL)) {
6799 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6800 Perl_dump_sv_child(aTHX_ sv);
6802 #ifdef DEBUG_LEAKING_SCALARS
6805 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6806 if (PL_warnhook == PERL_WARNHOOK_FATAL
6807 || ckDEAD(packWARN(WARN_INTERNAL))) {
6808 /* Don't let Perl_warner cause us to escape our fate: */
6812 /* This may not return: */
6813 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6814 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6815 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6818 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6828 Returns the length of the string in the SV. Handles magic and type
6829 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6830 gives raw access to the xpv_cur slot.
6836 Perl_sv_len(pTHX_ SV *const sv)
6843 (void)SvPV_const(sv, len);
6848 =for apidoc sv_len_utf8
6850 Returns the number of characters in the string in an SV, counting wide
6851 UTF-8 bytes as a single character. Handles magic and type coercion.
6857 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6858 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6859 * (Note that the mg_len is not the length of the mg_ptr field.
6860 * This allows the cache to store the character length of the string without
6861 * needing to malloc() extra storage to attach to the mg_ptr.)
6866 Perl_sv_len_utf8(pTHX_ SV *const sv)
6872 return sv_len_utf8_nomg(sv);
6876 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6880 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6882 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6884 if (PL_utf8cache && SvUTF8(sv)) {
6886 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6888 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6889 if (mg->mg_len != -1)
6892 /* We can use the offset cache for a headstart.
6893 The longer value is stored in the first pair. */
6894 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6896 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
6900 if (PL_utf8cache < 0) {
6901 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
6902 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
6906 ulen = Perl_utf8_length(aTHX_ s, s + len);
6907 utf8_mg_len_cache_update(sv, &mg, ulen);
6911 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
6914 /* Walk forwards to find the byte corresponding to the passed in UTF-8
6917 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
6918 STRLEN *const uoffset_p, bool *const at_end)
6920 const U8 *s = start;
6921 STRLEN uoffset = *uoffset_p;
6923 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
6925 while (s < send && uoffset) {
6932 else if (s > send) {
6934 /* This is the existing behaviour. Possibly it should be a croak, as
6935 it's actually a bounds error */
6938 *uoffset_p -= uoffset;
6942 /* Given the length of the string in both bytes and UTF-8 characters, decide
6943 whether to walk forwards or backwards to find the byte corresponding to
6944 the passed in UTF-8 offset. */
6946 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
6947 STRLEN uoffset, const STRLEN uend)
6949 STRLEN backw = uend - uoffset;
6951 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
6953 if (uoffset < 2 * backw) {
6954 /* The assumption is that going forwards is twice the speed of going
6955 forward (that's where the 2 * backw comes from).
6956 (The real figure of course depends on the UTF-8 data.) */
6957 const U8 *s = start;
6959 while (s < send && uoffset--)
6969 while (UTF8_IS_CONTINUATION(*send))
6972 return send - start;
6975 /* For the string representation of the given scalar, find the byte
6976 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
6977 give another position in the string, *before* the sought offset, which
6978 (which is always true, as 0, 0 is a valid pair of positions), which should
6979 help reduce the amount of linear searching.
6980 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
6981 will be used to reduce the amount of linear searching. The cache will be
6982 created if necessary, and the found value offered to it for update. */
6984 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
6985 const U8 *const send, STRLEN uoffset,
6986 STRLEN uoffset0, STRLEN boffset0)
6988 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
6990 bool at_end = FALSE;
6992 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
6994 assert (uoffset >= uoffset0);
6999 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7001 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7002 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7003 if ((*mgp)->mg_ptr) {
7004 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7005 if (cache[0] == uoffset) {
7006 /* An exact match. */
7009 if (cache[2] == uoffset) {
7010 /* An exact match. */
7014 if (cache[0] < uoffset) {
7015 /* The cache already knows part of the way. */
7016 if (cache[0] > uoffset0) {
7017 /* The cache knows more than the passed in pair */
7018 uoffset0 = cache[0];
7019 boffset0 = cache[1];
7021 if ((*mgp)->mg_len != -1) {
7022 /* And we know the end too. */
7024 + sv_pos_u2b_midway(start + boffset0, send,
7026 (*mgp)->mg_len - uoffset0);
7028 uoffset -= uoffset0;
7030 + sv_pos_u2b_forwards(start + boffset0,
7031 send, &uoffset, &at_end);
7032 uoffset += uoffset0;
7035 else if (cache[2] < uoffset) {
7036 /* We're between the two cache entries. */
7037 if (cache[2] > uoffset0) {
7038 /* and the cache knows more than the passed in pair */
7039 uoffset0 = cache[2];
7040 boffset0 = cache[3];
7044 + sv_pos_u2b_midway(start + boffset0,
7047 cache[0] - uoffset0);
7050 + sv_pos_u2b_midway(start + boffset0,
7053 cache[2] - uoffset0);
7057 else if ((*mgp)->mg_len != -1) {
7058 /* If we can take advantage of a passed in offset, do so. */
7059 /* In fact, offset0 is either 0, or less than offset, so don't
7060 need to worry about the other possibility. */
7062 + sv_pos_u2b_midway(start + boffset0, send,
7064 (*mgp)->mg_len - uoffset0);
7069 if (!found || PL_utf8cache < 0) {
7070 STRLEN real_boffset;
7071 uoffset -= uoffset0;
7072 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7073 send, &uoffset, &at_end);
7074 uoffset += uoffset0;
7076 if (found && PL_utf8cache < 0)
7077 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7079 boffset = real_boffset;
7082 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7084 utf8_mg_len_cache_update(sv, mgp, uoffset);
7086 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7093 =for apidoc sv_pos_u2b_flags
7095 Converts the offset from a count of UTF-8 chars from
7096 the start of the string, to a count of the equivalent number of bytes; if
7097 lenp is non-zero, it does the same to lenp, but this time starting from
7098 the offset, rather than from the start
7099 of the string. Handles type coercion.
7100 I<flags> is passed to C<SvPV_flags>, and usually should be
7101 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7107 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7108 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7109 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7114 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7121 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7123 start = (U8*)SvPV_flags(sv, len, flags);
7125 const U8 * const send = start + len;
7127 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7130 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7131 is 0, and *lenp is already set to that. */) {
7132 /* Convert the relative offset to absolute. */
7133 const STRLEN uoffset2 = uoffset + *lenp;
7134 const STRLEN boffset2
7135 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7136 uoffset, boffset) - boffset;
7150 =for apidoc sv_pos_u2b
7152 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7153 the start of the string, to a count of the equivalent number of bytes; if
7154 lenp is non-zero, it does the same to lenp, but this time starting from
7155 the offset, rather than from the start of the string. Handles magic and
7158 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7165 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7166 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7167 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7171 /* This function is subject to size and sign problems */
7174 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7176 PERL_ARGS_ASSERT_SV_POS_U2B;
7179 STRLEN ulen = (STRLEN)*lenp;
7180 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7181 SV_GMAGIC|SV_CONST_RETURN);
7184 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7185 SV_GMAGIC|SV_CONST_RETURN);
7190 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7193 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7194 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7197 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7198 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7199 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7203 (*mgp)->mg_len = ulen;
7206 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7207 byte length pairing. The (byte) length of the total SV is passed in too,
7208 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7209 may not have updated SvCUR, so we can't rely on reading it directly.
7211 The proffered utf8/byte length pairing isn't used if the cache already has
7212 two pairs, and swapping either for the proffered pair would increase the
7213 RMS of the intervals between known byte offsets.
7215 The cache itself consists of 4 STRLEN values
7216 0: larger UTF-8 offset
7217 1: corresponding byte offset
7218 2: smaller UTF-8 offset
7219 3: corresponding byte offset
7221 Unused cache pairs have the value 0, 0.
7222 Keeping the cache "backwards" means that the invariant of
7223 cache[0] >= cache[2] is maintained even with empty slots, which means that
7224 the code that uses it doesn't need to worry if only 1 entry has actually
7225 been set to non-zero. It also makes the "position beyond the end of the
7226 cache" logic much simpler, as the first slot is always the one to start
7230 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7231 const STRLEN utf8, const STRLEN blen)
7235 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7240 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7241 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7242 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7244 (*mgp)->mg_len = -1;
7248 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7249 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7250 (*mgp)->mg_ptr = (char *) cache;
7254 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7255 /* SvPOKp() because it's possible that sv has string overloading, and
7256 therefore is a reference, hence SvPVX() is actually a pointer.
7257 This cures the (very real) symptoms of RT 69422, but I'm not actually
7258 sure whether we should even be caching the results of UTF-8
7259 operations on overloading, given that nothing stops overloading
7260 returning a different value every time it's called. */
7261 const U8 *start = (const U8 *) SvPVX_const(sv);
7262 const STRLEN realutf8 = utf8_length(start, start + byte);
7264 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7268 /* Cache is held with the later position first, to simplify the code
7269 that deals with unbounded ends. */
7271 ASSERT_UTF8_CACHE(cache);
7272 if (cache[1] == 0) {
7273 /* Cache is totally empty */
7276 } else if (cache[3] == 0) {
7277 if (byte > cache[1]) {
7278 /* New one is larger, so goes first. */
7279 cache[2] = cache[0];
7280 cache[3] = cache[1];
7288 #define THREEWAY_SQUARE(a,b,c,d) \
7289 ((float)((d) - (c))) * ((float)((d) - (c))) \
7290 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7291 + ((float)((b) - (a))) * ((float)((b) - (a)))
7293 /* Cache has 2 slots in use, and we know three potential pairs.
7294 Keep the two that give the lowest RMS distance. Do the
7295 calculation in bytes simply because we always know the byte
7296 length. squareroot has the same ordering as the positive value,
7297 so don't bother with the actual square root. */
7298 if (byte > cache[1]) {
7299 /* New position is after the existing pair of pairs. */
7300 const float keep_earlier
7301 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7302 const float keep_later
7303 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7305 if (keep_later < keep_earlier) {
7306 cache[2] = cache[0];
7307 cache[3] = cache[1];
7316 else if (byte > cache[3]) {
7317 /* New position is between the existing pair of pairs. */
7318 const float keep_earlier
7319 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7320 const float keep_later
7321 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7323 if (keep_later < keep_earlier) {
7333 /* New position is before the existing pair of pairs. */
7334 const float keep_earlier
7335 = THREEWAY_SQUARE(0, byte, cache[3], blen);
7336 const float keep_later
7337 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7339 if (keep_later < keep_earlier) {
7344 cache[0] = cache[2];
7345 cache[1] = cache[3];
7351 ASSERT_UTF8_CACHE(cache);
7354 /* We already know all of the way, now we may be able to walk back. The same
7355 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7356 backward is half the speed of walking forward. */
7358 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7359 const U8 *end, STRLEN endu)
7361 const STRLEN forw = target - s;
7362 STRLEN backw = end - target;
7364 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7366 if (forw < 2 * backw) {
7367 return utf8_length(s, target);
7370 while (end > target) {
7372 while (UTF8_IS_CONTINUATION(*end)) {
7381 =for apidoc sv_pos_b2u_flags
7383 Converts the offset from a count of bytes from the start of the string, to
7384 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7385 I<flags> is passed to C<SvPV_flags>, and usually should be
7386 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7392 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7393 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7398 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7401 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7407 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7409 s = (const U8*)SvPV_flags(sv, blen, flags);
7412 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7413 ", byte=%"UVuf, (UV)blen, (UV)offset);
7419 && SvTYPE(sv) >= SVt_PVMG
7420 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7423 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7424 if (cache[1] == offset) {
7425 /* An exact match. */
7428 if (cache[3] == offset) {
7429 /* An exact match. */
7433 if (cache[1] < offset) {
7434 /* We already know part of the way. */
7435 if (mg->mg_len != -1) {
7436 /* Actually, we know the end too. */
7438 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7439 s + blen, mg->mg_len - cache[0]);
7441 len = cache[0] + utf8_length(s + cache[1], send);
7444 else if (cache[3] < offset) {
7445 /* We're between the two cached pairs, so we do the calculation
7446 offset by the byte/utf-8 positions for the earlier pair,
7447 then add the utf-8 characters from the string start to
7449 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7450 s + cache[1], cache[0] - cache[2])
7454 else { /* cache[3] > offset */
7455 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7459 ASSERT_UTF8_CACHE(cache);
7461 } else if (mg->mg_len != -1) {
7462 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7466 if (!found || PL_utf8cache < 0) {
7467 const STRLEN real_len = utf8_length(s, send);
7469 if (found && PL_utf8cache < 0)
7470 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7476 utf8_mg_len_cache_update(sv, &mg, len);
7478 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7485 =for apidoc sv_pos_b2u
7487 Converts the value pointed to by offsetp from a count of bytes from the
7488 start of the string, to a count of the equivalent number of UTF-8 chars.
7489 Handles magic and type coercion.
7491 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7498 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7499 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7504 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7506 PERL_ARGS_ASSERT_SV_POS_B2U;
7511 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7512 SV_GMAGIC|SV_CONST_RETURN);
7516 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7517 STRLEN real, SV *const sv)
7519 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7521 /* As this is debugging only code, save space by keeping this test here,
7522 rather than inlining it in all the callers. */
7523 if (from_cache == real)
7526 /* Need to turn the assertions off otherwise we may recurse infinitely
7527 while printing error messages. */
7528 SAVEI8(PL_utf8cache);
7530 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7531 func, (UV) from_cache, (UV) real, SVfARG(sv));
7537 Returns a boolean indicating whether the strings in the two SVs are
7538 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7539 coerce its args to strings if necessary.
7541 =for apidoc sv_eq_flags
7543 Returns a boolean indicating whether the strings in the two SVs are
7544 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7545 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7551 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7559 SV* svrecode = NULL;
7566 /* if pv1 and pv2 are the same, second SvPV_const call may
7567 * invalidate pv1 (if we are handling magic), so we may need to
7569 if (sv1 == sv2 && flags & SV_GMAGIC
7570 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7571 pv1 = SvPV_const(sv1, cur1);
7572 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7574 pv1 = SvPV_flags_const(sv1, cur1, flags);
7582 pv2 = SvPV_flags_const(sv2, cur2, flags);
7584 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7585 /* Differing utf8ness.
7586 * Do not UTF8size the comparands as a side-effect. */
7589 svrecode = newSVpvn(pv2, cur2);
7590 sv_recode_to_utf8(svrecode, PL_encoding);
7591 pv2 = SvPV_const(svrecode, cur2);
7594 svrecode = newSVpvn(pv1, cur1);
7595 sv_recode_to_utf8(svrecode, PL_encoding);
7596 pv1 = SvPV_const(svrecode, cur1);
7598 /* Now both are in UTF-8. */
7600 SvREFCNT_dec_NN(svrecode);
7606 /* sv1 is the UTF-8 one */
7607 return bytes_cmp_utf8((const U8*)pv2, cur2,
7608 (const U8*)pv1, cur1) == 0;
7611 /* sv2 is the UTF-8 one */
7612 return bytes_cmp_utf8((const U8*)pv1, cur1,
7613 (const U8*)pv2, cur2) == 0;
7619 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7621 SvREFCNT_dec(svrecode);
7629 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7630 string in C<sv1> is less than, equal to, or greater than the string in
7631 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7632 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7634 =for apidoc sv_cmp_flags
7636 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7637 string in C<sv1> is less than, equal to, or greater than the string in
7638 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7639 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7640 also C<sv_cmp_locale_flags>.
7646 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7648 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7652 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7657 const char *pv1, *pv2;
7659 SV *svrecode = NULL;
7666 pv1 = SvPV_flags_const(sv1, cur1, flags);
7673 pv2 = SvPV_flags_const(sv2, cur2, flags);
7675 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7676 /* Differing utf8ness.
7677 * Do not UTF8size the comparands as a side-effect. */
7680 svrecode = newSVpvn(pv2, cur2);
7681 sv_recode_to_utf8(svrecode, PL_encoding);
7682 pv2 = SvPV_const(svrecode, cur2);
7685 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7686 (const U8*)pv1, cur1);
7687 return retval ? retval < 0 ? -1 : +1 : 0;
7692 svrecode = newSVpvn(pv1, cur1);
7693 sv_recode_to_utf8(svrecode, PL_encoding);
7694 pv1 = SvPV_const(svrecode, cur1);
7697 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7698 (const U8*)pv2, cur2);
7699 return retval ? retval < 0 ? -1 : +1 : 0;
7705 cmp = cur2 ? -1 : 0;
7709 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7712 cmp = retval < 0 ? -1 : 1;
7713 } else if (cur1 == cur2) {
7716 cmp = cur1 < cur2 ? -1 : 1;
7720 SvREFCNT_dec(svrecode);
7726 =for apidoc sv_cmp_locale
7728 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7729 'use bytes' aware, handles get magic, and will coerce its args to strings
7730 if necessary. See also C<sv_cmp>.
7732 =for apidoc sv_cmp_locale_flags
7734 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7735 'use bytes' aware and will coerce its args to strings if necessary. If the
7736 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7742 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7744 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7748 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7752 #ifdef USE_LOCALE_COLLATE
7758 if (PL_collation_standard)
7762 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7764 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7766 if (!pv1 || !len1) {
7777 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7780 return retval < 0 ? -1 : 1;
7783 * When the result of collation is equality, that doesn't mean
7784 * that there are no differences -- some locales exclude some
7785 * characters from consideration. So to avoid false equalities,
7786 * we use the raw string as a tiebreaker.
7793 PERL_UNUSED_ARG(flags);
7794 #endif /* USE_LOCALE_COLLATE */
7796 return sv_cmp(sv1, sv2);
7800 #ifdef USE_LOCALE_COLLATE
7803 =for apidoc sv_collxfrm
7805 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7806 C<sv_collxfrm_flags>.
7808 =for apidoc sv_collxfrm_flags
7810 Add Collate Transform magic to an SV if it doesn't already have it. If the
7811 flags contain SV_GMAGIC, it handles get-magic.
7813 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7814 scalar data of the variable, but transformed to such a format that a normal
7815 memory comparison can be used to compare the data according to the locale
7822 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7827 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7829 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7830 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7836 Safefree(mg->mg_ptr);
7837 s = SvPV_flags_const(sv, len, flags);
7838 if ((xf = mem_collxfrm(s, len, &xlen))) {
7840 #ifdef PERL_OLD_COPY_ON_WRITE
7842 sv_force_normal_flags(sv, 0);
7844 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
7858 if (mg && mg->mg_ptr) {
7860 return mg->mg_ptr + sizeof(PL_collation_ix);
7868 #endif /* USE_LOCALE_COLLATE */
7871 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7873 SV * const tsv = newSV(0);
7876 sv_gets(tsv, fp, 0);
7877 sv_utf8_upgrade_nomg(tsv);
7878 SvCUR_set(sv,append);
7881 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
7885 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7888 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
7889 /* Grab the size of the record we're getting */
7890 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
7898 /* With a true, record-oriented file on VMS, we need to use read directly
7899 * to ensure that we respect RMS record boundaries. The user is responsible
7900 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
7901 * record size) field. N.B. This is likely to produce invalid results on
7902 * varying-width character data when a record ends mid-character.
7904 fd = PerlIO_fileno(fp);
7906 && PerlLIO_fstat(fd, &st) == 0
7907 && (st.st_fab_rfm == FAB$C_VAR
7908 || st.st_fab_rfm == FAB$C_VFC
7909 || st.st_fab_rfm == FAB$C_FIX)) {
7911 bytesread = PerlLIO_read(fd, buffer, recsize);
7913 else /* in-memory file from PerlIO::Scalar
7914 * or not a record-oriented file
7918 bytesread = PerlIO_read(fp, buffer, recsize);
7920 /* At this point, the logic in sv_get() means that sv will
7921 be treated as utf-8 if the handle is utf8.
7923 if (PerlIO_isutf8(fp) && bytesread > 0) {
7924 char *bend = buffer + bytesread;
7925 char *bufp = buffer;
7926 size_t charcount = 0;
7927 bool charstart = TRUE;
7930 while (charcount < recsize) {
7931 /* count accumulated characters */
7932 while (bufp < bend) {
7934 skip = UTF8SKIP(bufp);
7936 if (bufp + skip > bend) {
7937 /* partial at the end */
7948 if (charcount < recsize) {
7950 STRLEN bufp_offset = bufp - buffer;
7951 SSize_t morebytesread;
7953 /* originally I read enough to fill any incomplete
7954 character and the first byte of the next
7955 character if needed, but if there's many
7956 multi-byte encoded characters we're going to be
7957 making a read call for every character beyond
7958 the original read size.
7960 So instead, read the rest of the character if
7961 any, and enough bytes to match at least the
7962 start bytes for each character we're going to
7966 readsize = recsize - charcount;
7968 readsize = skip - (bend - bufp) + recsize - charcount - 1;
7969 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
7970 bend = buffer + bytesread;
7971 morebytesread = PerlIO_read(fp, bend, readsize);
7972 if (morebytesread <= 0) {
7973 /* we're done, if we still have incomplete
7974 characters the check code in sv_gets() will
7977 I'd originally considered doing
7978 PerlIO_ungetc() on all but the lead
7979 character of the incomplete character, but
7980 read() doesn't do that, so I don't.
7985 /* prepare to scan some more */
7986 bytesread += morebytesread;
7987 bend = buffer + bytesread;
7988 bufp = buffer + bufp_offset;
7996 SvCUR_set(sv, bytesread + append);
7997 buffer[bytesread] = '\0';
7998 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8004 Get a line from the filehandle and store it into the SV, optionally
8005 appending to the currently-stored string. If C<append> is not 0, the
8006 line is appended to the SV instead of overwriting it. C<append> should
8007 be set to the byte offset that the appended string should start at
8008 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8014 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8025 PERL_ARGS_ASSERT_SV_GETS;
8027 if (SvTHINKFIRST(sv))
8028 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8029 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8031 However, perlbench says it's slower, because the existing swipe code
8032 is faster than copy on write.
8033 Swings and roundabouts. */
8034 SvUPGRADE(sv, SVt_PV);
8037 if (PerlIO_isutf8(fp)) {
8039 sv_utf8_upgrade_nomg(sv);
8040 sv_pos_u2b(sv,&append,0);
8042 } else if (SvUTF8(sv)) {
8043 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8051 if (PerlIO_isutf8(fp))
8054 if (IN_PERL_COMPILETIME) {
8055 /* we always read code in line mode */
8059 else if (RsSNARF(PL_rs)) {
8060 /* If it is a regular disk file use size from stat() as estimate
8061 of amount we are going to read -- may result in mallocing
8062 more memory than we really need if the layers below reduce
8063 the size we read (e.g. CRLF or a gzip layer).
8066 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8067 const Off_t offset = PerlIO_tell(fp);
8068 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8069 #ifdef PERL_NEW_COPY_ON_WRITE
8070 /* Add an extra byte for the sake of copy-on-write's
8071 * buffer reference count. */
8072 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8074 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8081 else if (RsRECORD(PL_rs)) {
8082 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8084 else if (RsPARA(PL_rs)) {
8090 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8091 if (PerlIO_isutf8(fp)) {
8092 rsptr = SvPVutf8(PL_rs, rslen);
8095 if (SvUTF8(PL_rs)) {
8096 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8097 Perl_croak(aTHX_ "Wide character in $/");
8100 rsptr = SvPV_const(PL_rs, rslen);
8104 rslast = rslen ? rsptr[rslen - 1] : '\0';
8106 if (rspara) { /* have to do this both before and after */
8107 do { /* to make sure file boundaries work right */
8110 i = PerlIO_getc(fp);
8114 PerlIO_ungetc(fp,i);
8120 /* See if we know enough about I/O mechanism to cheat it ! */
8122 /* This used to be #ifdef test - it is made run-time test for ease
8123 of abstracting out stdio interface. One call should be cheap
8124 enough here - and may even be a macro allowing compile
8128 if (PerlIO_fast_gets(fp)) {
8131 * We're going to steal some values from the stdio struct
8132 * and put EVERYTHING in the innermost loop into registers.
8138 #if defined(VMS) && defined(PERLIO_IS_STDIO)
8139 /* An ungetc()d char is handled separately from the regular
8140 * buffer, so we getc() it back out and stuff it in the buffer.
8142 i = PerlIO_getc(fp);
8143 if (i == EOF) return 0;
8144 *(--((*fp)->_ptr)) = (unsigned char) i;
8148 /* Here is some breathtakingly efficient cheating */
8150 cnt = PerlIO_get_cnt(fp); /* get count into register */
8151 /* make sure we have the room */
8152 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8153 /* Not room for all of it
8154 if we are looking for a separator and room for some
8156 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8157 /* just process what we have room for */
8158 shortbuffered = cnt - SvLEN(sv) + append + 1;
8159 cnt -= shortbuffered;
8163 /* remember that cnt can be negative */
8164 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8169 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8170 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8171 DEBUG_P(PerlIO_printf(Perl_debug_log,
8172 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8173 DEBUG_P(PerlIO_printf(Perl_debug_log,
8174 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%zd, base=%"
8176 PTR2UV(PerlIO_get_ptr(fp)), PerlIO_get_cnt(fp),
8177 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8182 while (cnt > 0) { /* this | eat */
8184 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8185 goto thats_all_folks; /* screams | sed :-) */
8189 Copy(ptr, bp, cnt, char); /* this | eat */
8190 bp += cnt; /* screams | dust */
8191 ptr += cnt; /* louder | sed :-) */
8193 assert (!shortbuffered);
8194 goto cannot_be_shortbuffered;
8198 if (shortbuffered) { /* oh well, must extend */
8199 cnt = shortbuffered;
8201 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8203 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8204 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8208 cannot_be_shortbuffered:
8209 DEBUG_P(PerlIO_printf(Perl_debug_log,
8210 "Screamer: going to getc, ptr=%"UVuf", cnt=%zd\n",
8212 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8214 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8215 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%zd, base=%"UVuf"\n",
8216 PTR2UV(PerlIO_get_ptr(fp)), PerlIO_get_cnt(fp),
8217 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8219 /* This used to call 'filbuf' in stdio form, but as that behaves like
8220 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8221 another abstraction. */
8222 i = PerlIO_getc(fp); /* get more characters */
8224 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8225 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%zd, base=%"UVuf"\n",
8226 PTR2UV(PerlIO_get_ptr(fp)), PerlIO_get_cnt(fp),
8227 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8229 cnt = PerlIO_get_cnt(fp);
8230 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8231 DEBUG_P(PerlIO_printf(Perl_debug_log,
8232 "Screamer: after getc, ptr=%"UVuf", cnt=%zd\n",
8235 if (i == EOF) /* all done for ever? */
8236 goto thats_really_all_folks;
8238 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8240 SvGROW(sv, bpx + cnt + 2);
8241 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8243 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8245 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8246 goto thats_all_folks;
8250 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8251 memNE((char*)bp - rslen, rsptr, rslen))
8252 goto screamer; /* go back to the fray */
8253 thats_really_all_folks:
8255 cnt += shortbuffered;
8256 DEBUG_P(PerlIO_printf(Perl_debug_log,
8257 "Screamer: quitting, ptr=%"UVuf", cnt=%zd\n",PTR2UV(ptr),cnt));
8258 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8259 DEBUG_P(PerlIO_printf(Perl_debug_log,
8260 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%zd, base=%"UVuf
8262 PTR2UV(PerlIO_get_ptr(fp)), PerlIO_get_cnt(fp),
8263 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8265 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8266 DEBUG_P(PerlIO_printf(Perl_debug_log,
8267 "Screamer: done, len=%ld, string=|%.*s|\n",
8268 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8272 /*The big, slow, and stupid way. */
8273 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8274 STDCHAR *buf = NULL;
8275 Newx(buf, 8192, STDCHAR);
8283 const STDCHAR * const bpe = buf + sizeof(buf);
8285 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8286 ; /* keep reading */
8290 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8291 /* Accommodate broken VAXC compiler, which applies U8 cast to
8292 * both args of ?: operator, causing EOF to change into 255
8295 i = (U8)buf[cnt - 1];
8301 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8303 sv_catpvn_nomg(sv, (char *) buf, cnt);
8305 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8307 if (i != EOF && /* joy */
8309 SvCUR(sv) < rslen ||
8310 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8314 * If we're reading from a TTY and we get a short read,
8315 * indicating that the user hit his EOF character, we need
8316 * to notice it now, because if we try to read from the TTY
8317 * again, the EOF condition will disappear.
8319 * The comparison of cnt to sizeof(buf) is an optimization
8320 * that prevents unnecessary calls to feof().
8324 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8328 #ifdef USE_HEAP_INSTEAD_OF_STACK
8333 if (rspara) { /* have to do this both before and after */
8334 while (i != EOF) { /* to make sure file boundaries work right */
8335 i = PerlIO_getc(fp);
8337 PerlIO_ungetc(fp,i);
8343 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8349 Auto-increment of the value in the SV, doing string to numeric conversion
8350 if necessary. Handles 'get' magic and operator overloading.
8356 Perl_sv_inc(pTHX_ SV *const sv)
8365 =for apidoc sv_inc_nomg
8367 Auto-increment of the value in the SV, doing string to numeric conversion
8368 if necessary. Handles operator overloading. Skips handling 'get' magic.
8374 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8382 if (SvTHINKFIRST(sv)) {
8383 if (SvREADONLY(sv)) {
8384 Perl_croak_no_modify();
8388 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8390 i = PTR2IV(SvRV(sv));
8394 else sv_force_normal_flags(sv, 0);
8396 flags = SvFLAGS(sv);
8397 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8398 /* It's (privately or publicly) a float, but not tested as an
8399 integer, so test it to see. */
8401 flags = SvFLAGS(sv);
8403 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8404 /* It's publicly an integer, or privately an integer-not-float */
8405 #ifdef PERL_PRESERVE_IVUV
8409 if (SvUVX(sv) == UV_MAX)
8410 sv_setnv(sv, UV_MAX_P1);
8412 (void)SvIOK_only_UV(sv);
8413 SvUV_set(sv, SvUVX(sv) + 1);
8415 if (SvIVX(sv) == IV_MAX)
8416 sv_setuv(sv, (UV)IV_MAX + 1);
8418 (void)SvIOK_only(sv);
8419 SvIV_set(sv, SvIVX(sv) + 1);
8424 if (flags & SVp_NOK) {
8425 const NV was = SvNVX(sv);
8426 if (NV_OVERFLOWS_INTEGERS_AT &&
8427 was >= NV_OVERFLOWS_INTEGERS_AT) {
8428 /* diag_listed_as: Lost precision when %s %f by 1 */
8429 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8430 "Lost precision when incrementing %" NVff " by 1",
8433 (void)SvNOK_only(sv);
8434 SvNV_set(sv, was + 1.0);
8438 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8439 if ((flags & SVTYPEMASK) < SVt_PVIV)
8440 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8441 (void)SvIOK_only(sv);
8446 while (isALPHA(*d)) d++;
8447 while (isDIGIT(*d)) d++;
8448 if (d < SvEND(sv)) {
8449 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8450 #ifdef PERL_PRESERVE_IVUV
8451 /* Got to punt this as an integer if needs be, but we don't issue
8452 warnings. Probably ought to make the sv_iv_please() that does
8453 the conversion if possible, and silently. */
8454 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8455 /* Need to try really hard to see if it's an integer.
8456 9.22337203685478e+18 is an integer.
8457 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8458 so $a="9.22337203685478e+18"; $a+0; $a++
8459 needs to be the same as $a="9.22337203685478e+18"; $a++
8466 /* sv_2iv *should* have made this an NV */
8467 if (flags & SVp_NOK) {
8468 (void)SvNOK_only(sv);
8469 SvNV_set(sv, SvNVX(sv) + 1.0);
8472 /* I don't think we can get here. Maybe I should assert this
8473 And if we do get here I suspect that sv_setnv will croak. NWC
8475 #if defined(USE_LONG_DOUBLE)
8476 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"PERL_PRIgldbl"\n",
8477 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8479 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8480 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8483 #endif /* PERL_PRESERVE_IVUV */
8484 if (!numtype && ckWARN(WARN_NUMERIC))
8485 not_incrementable(sv);
8486 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8490 while (d >= SvPVX_const(sv)) {
8498 /* MKS: The original code here died if letters weren't consecutive.
8499 * at least it didn't have to worry about non-C locales. The
8500 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8501 * arranged in order (although not consecutively) and that only
8502 * [A-Za-z] are accepted by isALPHA in the C locale.
8504 if (*d != 'z' && *d != 'Z') {
8505 do { ++*d; } while (!isALPHA(*d));
8508 *(d--) -= 'z' - 'a';
8513 *(d--) -= 'z' - 'a' + 1;
8517 /* oh,oh, the number grew */
8518 SvGROW(sv, SvCUR(sv) + 2);
8519 SvCUR_set(sv, SvCUR(sv) + 1);
8520 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8531 Auto-decrement of the value in the SV, doing string to numeric conversion
8532 if necessary. Handles 'get' magic and operator overloading.
8538 Perl_sv_dec(pTHX_ SV *const sv)
8548 =for apidoc sv_dec_nomg
8550 Auto-decrement of the value in the SV, doing string to numeric conversion
8551 if necessary. Handles operator overloading. Skips handling 'get' magic.
8557 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8564 if (SvTHINKFIRST(sv)) {
8565 if (SvREADONLY(sv)) {
8566 Perl_croak_no_modify();
8570 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8572 i = PTR2IV(SvRV(sv));
8576 else sv_force_normal_flags(sv, 0);
8578 /* Unlike sv_inc we don't have to worry about string-never-numbers
8579 and keeping them magic. But we mustn't warn on punting */
8580 flags = SvFLAGS(sv);
8581 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8582 /* It's publicly an integer, or privately an integer-not-float */
8583 #ifdef PERL_PRESERVE_IVUV
8587 if (SvUVX(sv) == 0) {
8588 (void)SvIOK_only(sv);
8592 (void)SvIOK_only_UV(sv);
8593 SvUV_set(sv, SvUVX(sv) - 1);
8596 if (SvIVX(sv) == IV_MIN) {
8597 sv_setnv(sv, (NV)IV_MIN);
8601 (void)SvIOK_only(sv);
8602 SvIV_set(sv, SvIVX(sv) - 1);
8607 if (flags & SVp_NOK) {
8610 const NV was = SvNVX(sv);
8611 if (NV_OVERFLOWS_INTEGERS_AT &&
8612 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8613 /* diag_listed_as: Lost precision when %s %f by 1 */
8614 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8615 "Lost precision when decrementing %" NVff " by 1",
8618 (void)SvNOK_only(sv);
8619 SvNV_set(sv, was - 1.0);
8623 if (!(flags & SVp_POK)) {
8624 if ((flags & SVTYPEMASK) < SVt_PVIV)
8625 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8627 (void)SvIOK_only(sv);
8630 #ifdef PERL_PRESERVE_IVUV
8632 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8633 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8634 /* Need to try really hard to see if it's an integer.
8635 9.22337203685478e+18 is an integer.
8636 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8637 so $a="9.22337203685478e+18"; $a+0; $a--
8638 needs to be the same as $a="9.22337203685478e+18"; $a--
8645 /* sv_2iv *should* have made this an NV */
8646 if (flags & SVp_NOK) {
8647 (void)SvNOK_only(sv);
8648 SvNV_set(sv, SvNVX(sv) - 1.0);
8651 /* I don't think we can get here. Maybe I should assert this
8652 And if we do get here I suspect that sv_setnv will croak. NWC
8654 #if defined(USE_LONG_DOUBLE)
8655 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"PERL_PRIgldbl"\n",
8656 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8658 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8659 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8663 #endif /* PERL_PRESERVE_IVUV */
8664 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8667 /* this define is used to eliminate a chunk of duplicated but shared logic
8668 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8669 * used anywhere but here - yves
8671 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8674 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8678 =for apidoc sv_mortalcopy
8680 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8681 The new SV is marked as mortal. It will be destroyed "soon", either by an
8682 explicit call to FREETMPS, or by an implicit call at places such as
8683 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8688 /* Make a string that will exist for the duration of the expression
8689 * evaluation. Actually, it may have to last longer than that, but
8690 * hopefully we won't free it until it has been assigned to a
8691 * permanent location. */
8694 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8699 if (flags & SV_GMAGIC)
8700 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8702 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8703 PUSH_EXTEND_MORTAL__SV_C(sv);
8709 =for apidoc sv_newmortal
8711 Creates a new null SV which is mortal. The reference count of the SV is
8712 set to 1. It will be destroyed "soon", either by an explicit call to
8713 FREETMPS, or by an implicit call at places such as statement boundaries.
8714 See also C<sv_mortalcopy> and C<sv_2mortal>.
8720 Perl_sv_newmortal(pTHX)
8726 SvFLAGS(sv) = SVs_TEMP;
8727 PUSH_EXTEND_MORTAL__SV_C(sv);
8733 =for apidoc newSVpvn_flags
8735 Creates a new SV and copies a string into it. The reference count for the
8736 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8737 string. You are responsible for ensuring that the source string is at least
8738 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8739 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8740 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8741 returning. If C<SVf_UTF8> is set, C<s>
8742 is considered to be in UTF-8 and the
8743 C<SVf_UTF8> flag will be set on the new SV.
8744 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8746 #define newSVpvn_utf8(s, len, u) \
8747 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8753 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8758 /* All the flags we don't support must be zero.
8759 And we're new code so I'm going to assert this from the start. */
8760 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
8762 sv_setpvn(sv,s,len);
8764 /* This code used to do a sv_2mortal(), however we now unroll the call to
8765 * sv_2mortal() and do what it does ourselves here. Since we have asserted
8766 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
8767 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
8768 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
8769 * means that we eliminate quite a few steps than it looks - Yves
8770 * (explaining patch by gfx) */
8772 SvFLAGS(sv) |= flags;
8774 if(flags & SVs_TEMP){
8775 PUSH_EXTEND_MORTAL__SV_C(sv);
8782 =for apidoc sv_2mortal
8784 Marks an existing SV as mortal. The SV will be destroyed "soon", either
8785 by an explicit call to FREETMPS, or by an implicit call at places such as
8786 statement boundaries. SvTEMP() is turned on which means that the SV's
8787 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
8788 and C<sv_mortalcopy>.
8794 Perl_sv_2mortal(pTHX_ SV *const sv)
8801 PUSH_EXTEND_MORTAL__SV_C(sv);
8809 Creates a new SV and copies a string into it. The reference count for the
8810 SV is set to 1. If C<len> is zero, Perl will compute the length using
8811 strlen(). For efficiency, consider using C<newSVpvn> instead.
8817 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
8823 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
8828 =for apidoc newSVpvn
8830 Creates a new SV and copies a buffer into it, which may contain NUL characters
8831 (C<\0>) and other binary data. The reference count for the SV is set to 1.
8832 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
8833 are responsible for ensuring that the source buffer is at least
8834 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
8841 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
8847 sv_setpvn(sv,buffer,len);
8852 =for apidoc newSVhek
8854 Creates a new SV from the hash key structure. It will generate scalars that
8855 point to the shared string table where possible. Returns a new (undefined)
8856 SV if the hek is NULL.
8862 Perl_newSVhek(pTHX_ const HEK *const hek)
8872 if (HEK_LEN(hek) == HEf_SVKEY) {
8873 return newSVsv(*(SV**)HEK_KEY(hek));
8875 const int flags = HEK_FLAGS(hek);
8876 if (flags & HVhek_WASUTF8) {
8878 Andreas would like keys he put in as utf8 to come back as utf8
8880 STRLEN utf8_len = HEK_LEN(hek);
8881 SV * const sv = newSV_type(SVt_PV);
8882 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
8883 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
8884 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
8887 } else if (flags & HVhek_UNSHARED) {
8888 /* A hash that isn't using shared hash keys has to have
8889 the flag in every key so that we know not to try to call
8890 share_hek_hek on it. */
8892 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
8897 /* This will be overwhelminly the most common case. */
8899 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
8900 more efficient than sharepvn(). */
8904 sv_upgrade(sv, SVt_PV);
8905 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
8906 SvCUR_set(sv, HEK_LEN(hek));
8918 =for apidoc newSVpvn_share
8920 Creates a new SV with its SvPVX_const pointing to a shared string in the string
8921 table. If the string does not already exist in the table, it is
8922 created first. Turns on the SvIsCOW flag (or READONLY
8923 and FAKE in 5.16 and earlier). If the C<hash> parameter
8924 is non-zero, that value is used; otherwise the hash is computed.
8925 The string's hash can later be retrieved from the SV
8926 with the C<SvSHARED_HASH()> macro. The idea here is
8927 that as the string table is used for shared hash keys these strings will have
8928 SvPVX_const == HeKEY and hash lookup will avoid string compare.
8934 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
8938 bool is_utf8 = FALSE;
8939 const char *const orig_src = src;
8942 STRLEN tmplen = -len;
8944 /* See the note in hv.c:hv_fetch() --jhi */
8945 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
8949 PERL_HASH(hash, src, len);
8951 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
8952 changes here, update it there too. */
8953 sv_upgrade(sv, SVt_PV);
8954 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
8961 if (src != orig_src)
8967 =for apidoc newSVpv_share
8969 Like C<newSVpvn_share>, but takes a nul-terminated string instead of a
8976 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
8978 return newSVpvn_share(src, strlen(src), hash);
8981 #if defined(PERL_IMPLICIT_CONTEXT)
8983 /* pTHX_ magic can't cope with varargs, so this is a no-context
8984 * version of the main function, (which may itself be aliased to us).
8985 * Don't access this version directly.
8989 Perl_newSVpvf_nocontext(const char *const pat, ...)
8995 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
8997 va_start(args, pat);
8998 sv = vnewSVpvf(pat, &args);
9005 =for apidoc newSVpvf
9007 Creates a new SV and initializes it with the string formatted like
9014 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9019 PERL_ARGS_ASSERT_NEWSVPVF;
9021 va_start(args, pat);
9022 sv = vnewSVpvf(pat, &args);
9027 /* backend for newSVpvf() and newSVpvf_nocontext() */
9030 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9035 PERL_ARGS_ASSERT_VNEWSVPVF;
9038 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9045 Creates a new SV and copies a floating point value into it.
9046 The reference count for the SV is set to 1.
9052 Perl_newSVnv(pTHX_ const NV n)
9065 Creates a new SV and copies an integer into it. The reference count for the
9072 Perl_newSViv(pTHX_ const IV i)
9085 Creates a new SV and copies an unsigned integer into it.
9086 The reference count for the SV is set to 1.
9092 Perl_newSVuv(pTHX_ const UV u)
9103 =for apidoc newSV_type
9105 Creates a new SV, of the type specified. The reference count for the new SV
9112 Perl_newSV_type(pTHX_ const svtype type)
9117 sv_upgrade(sv, type);
9122 =for apidoc newRV_noinc
9124 Creates an RV wrapper for an SV. The reference count for the original
9125 SV is B<not> incremented.
9131 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9134 SV *sv = newSV_type(SVt_IV);
9136 PERL_ARGS_ASSERT_NEWRV_NOINC;
9139 SvRV_set(sv, tmpRef);
9144 /* newRV_inc is the official function name to use now.
9145 * newRV_inc is in fact #defined to newRV in sv.h
9149 Perl_newRV(pTHX_ SV *const sv)
9153 PERL_ARGS_ASSERT_NEWRV;
9155 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9161 Creates a new SV which is an exact duplicate of the original SV.
9168 Perl_newSVsv(pTHX_ SV *const old)
9175 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9176 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9179 /* Do this here, otherwise we leak the new SV if this croaks. */
9182 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9183 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9184 sv_setsv_flags(sv, old, SV_NOSTEAL);
9189 =for apidoc sv_reset
9191 Underlying implementation for the C<reset> Perl function.
9192 Note that the perl-level function is vaguely deprecated.
9198 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9200 PERL_ARGS_ASSERT_SV_RESET;
9202 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9206 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9209 char todo[PERL_UCHAR_MAX+1];
9212 if (!stash || SvTYPE(stash) != SVt_PVHV)
9215 if (!s) { /* reset ?? searches */
9216 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9218 const U32 count = mg->mg_len / sizeof(PMOP**);
9219 PMOP **pmp = (PMOP**) mg->mg_ptr;
9220 PMOP *const *const end = pmp + count;
9224 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9226 (*pmp)->op_pmflags &= ~PMf_USED;
9234 /* reset variables */
9236 if (!HvARRAY(stash))
9239 Zero(todo, 256, char);
9243 I32 i = (unsigned char)*s;
9247 max = (unsigned char)*s++;
9248 for ( ; i <= max; i++) {
9251 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9253 for (entry = HvARRAY(stash)[i];
9255 entry = HeNEXT(entry))
9260 if (!todo[(U8)*HeKEY(entry)])
9262 gv = MUTABLE_GV(HeVAL(entry));
9264 if (sv && !SvREADONLY(sv)) {
9265 SV_CHECK_THINKFIRST_COW_DROP(sv);
9266 if (!isGV(sv)) SvOK_off(sv);
9271 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9282 Using various gambits, try to get an IO from an SV: the IO slot if its a
9283 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9284 named after the PV if we're a string.
9286 'Get' magic is ignored on the sv passed in, but will be called on
9287 C<SvRV(sv)> if sv is an RV.
9293 Perl_sv_2io(pTHX_ SV *const sv)
9298 PERL_ARGS_ASSERT_SV_2IO;
9300 switch (SvTYPE(sv)) {
9302 io = MUTABLE_IO(sv);
9306 if (isGV_with_GP(sv)) {
9307 gv = MUTABLE_GV(sv);
9310 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9311 HEKfARG(GvNAME_HEK(gv)));
9317 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9319 SvGETMAGIC(SvRV(sv));
9320 return sv_2io(SvRV(sv));
9322 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9329 if (SvGMAGICAL(sv)) {
9330 newsv = sv_newmortal();
9331 sv_setsv_nomg(newsv, sv);
9333 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9343 Using various gambits, try to get a CV from an SV; in addition, try if
9344 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9345 The flags in C<lref> are passed to gv_fetchsv.
9351 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9357 PERL_ARGS_ASSERT_SV_2CV;
9364 switch (SvTYPE(sv)) {
9368 return MUTABLE_CV(sv);
9378 sv = amagic_deref_call(sv, to_cv_amg);
9381 if (SvTYPE(sv) == SVt_PVCV) {
9382 cv = MUTABLE_CV(sv);
9387 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9388 gv = MUTABLE_GV(sv);
9390 Perl_croak(aTHX_ "Not a subroutine reference");
9392 else if (isGV_with_GP(sv)) {
9393 gv = MUTABLE_GV(sv);
9396 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9403 /* Some flags to gv_fetchsv mean don't really create the GV */
9404 if (!isGV_with_GP(gv)) {
9409 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9410 /* XXX this is probably not what they think they're getting.
9411 * It has the same effect as "sub name;", i.e. just a forward
9422 Returns true if the SV has a true value by Perl's rules.
9423 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9424 instead use an in-line version.
9430 Perl_sv_true(pTHX_ SV *const sv)
9435 const XPV* const tXpv = (XPV*)SvANY(sv);
9437 (tXpv->xpv_cur > 1 ||
9438 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9445 return SvIVX(sv) != 0;
9448 return SvNVX(sv) != 0.0;
9450 return sv_2bool(sv);
9456 =for apidoc sv_pvn_force
9458 Get a sensible string out of the SV somehow.
9459 A private implementation of the C<SvPV_force> macro for compilers which
9460 can't cope with complex macro expressions. Always use the macro instead.
9462 =for apidoc sv_pvn_force_flags
9464 Get a sensible string out of the SV somehow.
9465 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9466 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9467 implemented in terms of this function.
9468 You normally want to use the various wrapper macros instead: see
9469 C<SvPV_force> and C<SvPV_force_nomg>
9475 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9479 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9481 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9482 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9483 sv_force_normal_flags(sv, 0);
9493 if (SvTYPE(sv) > SVt_PVLV
9494 || isGV_with_GP(sv))
9495 /* diag_listed_as: Can't coerce %s to %s in %s */
9496 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9498 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9505 if (SvTYPE(sv) < SVt_PV ||
9506 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9509 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9510 SvGROW(sv, len + 1);
9511 Move(s,SvPVX(sv),len,char);
9513 SvPVX(sv)[len] = '\0';
9516 SvPOK_on(sv); /* validate pointer */
9518 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9519 PTR2UV(sv),SvPVX_const(sv)));
9522 (void)SvPOK_only_UTF8(sv);
9523 return SvPVX_mutable(sv);
9527 =for apidoc sv_pvbyten_force
9529 The backend for the C<SvPVbytex_force> macro. Always use the macro
9536 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9538 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9540 sv_pvn_force(sv,lp);
9541 sv_utf8_downgrade(sv,0);
9547 =for apidoc sv_pvutf8n_force
9549 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9556 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9558 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9561 sv_utf8_upgrade_nomg(sv);
9567 =for apidoc sv_reftype
9569 Returns a string describing what the SV is a reference to.
9575 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9577 PERL_ARGS_ASSERT_SV_REFTYPE;
9578 if (ob && SvOBJECT(sv)) {
9579 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9582 /* WARNING - There is code, for instance in mg.c, that assumes that
9583 * the only reason that sv_reftype(sv,0) would return a string starting
9584 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9585 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9586 * this routine inside other subs, and it saves time.
9587 * Do not change this assumption without searching for "dodgy type check" in
9590 switch (SvTYPE(sv)) {
9605 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9606 /* tied lvalues should appear to be
9607 * scalars for backwards compatibility */
9608 : (LvTYPE(sv) == 't' || LvTYPE(sv) == 'T')
9609 ? "SCALAR" : "LVALUE");
9610 case SVt_PVAV: return "ARRAY";
9611 case SVt_PVHV: return "HASH";
9612 case SVt_PVCV: return "CODE";
9613 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9614 ? "GLOB" : "SCALAR");
9615 case SVt_PVFM: return "FORMAT";
9616 case SVt_PVIO: return "IO";
9617 case SVt_INVLIST: return "INVLIST";
9618 case SVt_REGEXP: return "REGEXP";
9619 default: return "UNKNOWN";
9627 Returns a SV describing what the SV passed in is a reference to.
9633 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9635 PERL_ARGS_ASSERT_SV_REF;
9638 dst = sv_newmortal();
9640 if (ob && SvOBJECT(sv)) {
9641 HvNAME_get(SvSTASH(sv))
9642 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9643 : sv_setpvn(dst, "__ANON__", 8);
9646 const char * reftype = sv_reftype(sv, 0);
9647 sv_setpv(dst, reftype);
9653 =for apidoc sv_isobject
9655 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9656 object. If the SV is not an RV, or if the object is not blessed, then this
9663 Perl_sv_isobject(pTHX_ SV *sv)
9679 Returns a boolean indicating whether the SV is blessed into the specified
9680 class. This does not check for subtypes; use C<sv_derived_from> to verify
9681 an inheritance relationship.
9687 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9691 PERL_ARGS_ASSERT_SV_ISA;
9701 hvname = HvNAME_get(SvSTASH(sv));
9705 return strEQ(hvname, name);
9711 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9712 RV then it will be upgraded to one. If C<classname> is non-null then the new
9713 SV will be blessed in the specified package. The new SV is returned and its
9714 reference count is 1. The reference count 1 is owned by C<rv>.
9720 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9725 PERL_ARGS_ASSERT_NEWSVRV;
9729 SV_CHECK_THINKFIRST_COW_DROP(rv);
9731 if (SvTYPE(rv) >= SVt_PVMG) {
9732 const U32 refcnt = SvREFCNT(rv);
9736 SvREFCNT(rv) = refcnt;
9738 sv_upgrade(rv, SVt_IV);
9739 } else if (SvROK(rv)) {
9740 SvREFCNT_dec(SvRV(rv));
9742 prepare_SV_for_RV(rv);
9750 HV* const stash = gv_stashpv(classname, GV_ADD);
9751 (void)sv_bless(rv, stash);
9757 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9759 SV * const lv = newSV_type(SVt_PVLV);
9760 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9762 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9763 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9764 LvSTARGOFF(lv) = ix;
9765 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9770 =for apidoc sv_setref_pv
9772 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9773 argument will be upgraded to an RV. That RV will be modified to point to
9774 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
9775 into the SV. The C<classname> argument indicates the package for the
9776 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9777 will have a reference count of 1, and the RV will be returned.
9779 Do not use with other Perl types such as HV, AV, SV, CV, because those
9780 objects will become corrupted by the pointer copy process.
9782 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
9788 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
9792 PERL_ARGS_ASSERT_SV_SETREF_PV;
9795 sv_setsv(rv, &PL_sv_undef);
9799 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
9804 =for apidoc sv_setref_iv
9806 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
9807 argument will be upgraded to an RV. That RV will be modified to point to
9808 the new SV. The C<classname> argument indicates the package for the
9809 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9810 will have a reference count of 1, and the RV will be returned.
9816 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
9818 PERL_ARGS_ASSERT_SV_SETREF_IV;
9820 sv_setiv(newSVrv(rv,classname), iv);
9825 =for apidoc sv_setref_uv
9827 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
9828 argument will be upgraded to an RV. That RV will be modified to point to
9829 the new SV. The C<classname> argument indicates the package for the
9830 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9831 will have a reference count of 1, and the RV will be returned.
9837 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
9839 PERL_ARGS_ASSERT_SV_SETREF_UV;
9841 sv_setuv(newSVrv(rv,classname), uv);
9846 =for apidoc sv_setref_nv
9848 Copies a double into a new SV, optionally blessing the SV. The C<rv>
9849 argument will be upgraded to an RV. That RV will be modified to point to
9850 the new SV. The C<classname> argument indicates the package for the
9851 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9852 will have a reference count of 1, and the RV will be returned.
9858 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
9860 PERL_ARGS_ASSERT_SV_SETREF_NV;
9862 sv_setnv(newSVrv(rv,classname), nv);
9867 =for apidoc sv_setref_pvn
9869 Copies a string into a new SV, optionally blessing the SV. The length of the
9870 string must be specified with C<n>. The C<rv> argument will be upgraded to
9871 an RV. That RV will be modified to point to the new SV. The C<classname>
9872 argument indicates the package for the blessing. Set C<classname> to
9873 C<NULL> to avoid the blessing. The new SV will have a reference count
9874 of 1, and the RV will be returned.
9876 Note that C<sv_setref_pv> copies the pointer while this copies the string.
9882 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
9883 const char *const pv, const STRLEN n)
9885 PERL_ARGS_ASSERT_SV_SETREF_PVN;
9887 sv_setpvn(newSVrv(rv,classname), pv, n);
9892 =for apidoc sv_bless
9894 Blesses an SV into a specified package. The SV must be an RV. The package
9895 must be designated by its stash (see C<gv_stashpv()>). The reference count
9896 of the SV is unaffected.
9902 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
9906 HV *oldstash = NULL;
9908 PERL_ARGS_ASSERT_SV_BLESS;
9912 Perl_croak(aTHX_ "Can't bless non-reference value");
9914 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY)) {
9915 if (SvREADONLY(tmpRef))
9916 Perl_croak_no_modify();
9917 if (SvOBJECT(tmpRef)) {
9918 oldstash = SvSTASH(tmpRef);
9921 SvOBJECT_on(tmpRef);
9922 SvUPGRADE(tmpRef, SVt_PVMG);
9923 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
9924 SvREFCNT_dec(oldstash);
9926 if(SvSMAGICAL(tmpRef))
9927 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
9935 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
9936 * as it is after unglobbing it.
9939 PERL_STATIC_INLINE void
9940 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
9945 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
9947 PERL_ARGS_ASSERT_SV_UNGLOB;
9949 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
9951 if (!(flags & SV_COW_DROP_PV))
9952 gv_efullname3(temp, MUTABLE_GV(sv), "*");
9954 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
9956 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
9957 && HvNAME_get(stash))
9958 mro_method_changed_in(stash);
9959 gp_free(MUTABLE_GV(sv));
9962 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
9966 if (GvNAME_HEK(sv)) {
9967 unshare_hek(GvNAME_HEK(sv));
9969 isGV_with_GP_off(sv);
9971 if(SvTYPE(sv) == SVt_PVGV) {
9972 /* need to keep SvANY(sv) in the right arena */
9973 xpvmg = new_XPVMG();
9974 StructCopy(SvANY(sv), xpvmg, XPVMG);
9975 del_XPVGV(SvANY(sv));
9978 SvFLAGS(sv) &= ~SVTYPEMASK;
9979 SvFLAGS(sv) |= SVt_PVMG;
9982 /* Intentionally not calling any local SET magic, as this isn't so much a
9983 set operation as merely an internal storage change. */
9984 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
9985 else sv_setsv_flags(sv, temp, 0);
9987 if ((const GV *)sv == PL_last_in_gv)
9988 PL_last_in_gv = NULL;
9989 else if ((const GV *)sv == PL_statgv)
9994 =for apidoc sv_unref_flags
9996 Unsets the RV status of the SV, and decrements the reference count of
9997 whatever was being referenced by the RV. This can almost be thought of
9998 as a reversal of C<newSVrv>. The C<cflags> argument can contain
9999 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10000 (otherwise the decrementing is conditional on the reference count being
10001 different from one or the reference being a readonly SV).
10008 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10010 SV* const target = SvRV(ref);
10012 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10014 if (SvWEAKREF(ref)) {
10015 sv_del_backref(target, ref);
10016 SvWEAKREF_off(ref);
10017 SvRV_set(ref, NULL);
10020 SvRV_set(ref, NULL);
10022 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10023 assigned to as BEGIN {$a = \"Foo"} will fail. */
10024 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10025 SvREFCNT_dec_NN(target);
10026 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10027 sv_2mortal(target); /* Schedule for freeing later */
10031 =for apidoc sv_untaint
10033 Untaint an SV. Use C<SvTAINTED_off> instead.
10039 Perl_sv_untaint(pTHX_ SV *const sv)
10041 PERL_ARGS_ASSERT_SV_UNTAINT;
10043 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10044 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10051 =for apidoc sv_tainted
10053 Test an SV for taintedness. Use C<SvTAINTED> instead.
10059 Perl_sv_tainted(pTHX_ SV *const sv)
10061 PERL_ARGS_ASSERT_SV_TAINTED;
10063 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10064 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10065 if (mg && (mg->mg_len & 1) )
10072 =for apidoc sv_setpviv
10074 Copies an integer into the given SV, also updating its string value.
10075 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10081 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10083 char buf[TYPE_CHARS(UV)];
10085 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10087 PERL_ARGS_ASSERT_SV_SETPVIV;
10089 sv_setpvn(sv, ptr, ebuf - ptr);
10093 =for apidoc sv_setpviv_mg
10095 Like C<sv_setpviv>, but also handles 'set' magic.
10101 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10103 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10105 sv_setpviv(sv, iv);
10109 #if defined(PERL_IMPLICIT_CONTEXT)
10111 /* pTHX_ magic can't cope with varargs, so this is a no-context
10112 * version of the main function, (which may itself be aliased to us).
10113 * Don't access this version directly.
10117 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10122 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10124 va_start(args, pat);
10125 sv_vsetpvf(sv, pat, &args);
10129 /* pTHX_ magic can't cope with varargs, so this is a no-context
10130 * version of the main function, (which may itself be aliased to us).
10131 * Don't access this version directly.
10135 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10140 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10142 va_start(args, pat);
10143 sv_vsetpvf_mg(sv, pat, &args);
10149 =for apidoc sv_setpvf
10151 Works like C<sv_catpvf> but copies the text into the SV instead of
10152 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10158 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10162 PERL_ARGS_ASSERT_SV_SETPVF;
10164 va_start(args, pat);
10165 sv_vsetpvf(sv, pat, &args);
10170 =for apidoc sv_vsetpvf
10172 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10173 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10175 Usually used via its frontend C<sv_setpvf>.
10181 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10183 PERL_ARGS_ASSERT_SV_VSETPVF;
10185 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10189 =for apidoc sv_setpvf_mg
10191 Like C<sv_setpvf>, but also handles 'set' magic.
10197 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10201 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10203 va_start(args, pat);
10204 sv_vsetpvf_mg(sv, pat, &args);
10209 =for apidoc sv_vsetpvf_mg
10211 Like C<sv_vsetpvf>, but also handles 'set' magic.
10213 Usually used via its frontend C<sv_setpvf_mg>.
10219 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10221 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10223 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10227 #if defined(PERL_IMPLICIT_CONTEXT)
10229 /* pTHX_ magic can't cope with varargs, so this is a no-context
10230 * version of the main function, (which may itself be aliased to us).
10231 * Don't access this version directly.
10235 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10240 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10242 va_start(args, pat);
10243 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10247 /* pTHX_ magic can't cope with varargs, so this is a no-context
10248 * version of the main function, (which may itself be aliased to us).
10249 * Don't access this version directly.
10253 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10258 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10260 va_start(args, pat);
10261 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10268 =for apidoc sv_catpvf
10270 Processes its arguments like C<sprintf> and appends the formatted
10271 output to an SV. If the appended data contains "wide" characters
10272 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10273 and characters >255 formatted with %c), the original SV might get
10274 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10275 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10276 valid UTF-8; if the original SV was bytes, the pattern should be too.
10281 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10285 PERL_ARGS_ASSERT_SV_CATPVF;
10287 va_start(args, pat);
10288 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10293 =for apidoc sv_vcatpvf
10295 Processes its arguments like C<vsprintf> and appends the formatted output
10296 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10298 Usually used via its frontend C<sv_catpvf>.
10304 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10306 PERL_ARGS_ASSERT_SV_VCATPVF;
10308 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10312 =for apidoc sv_catpvf_mg
10314 Like C<sv_catpvf>, but also handles 'set' magic.
10320 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10324 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10326 va_start(args, pat);
10327 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10333 =for apidoc sv_vcatpvf_mg
10335 Like C<sv_vcatpvf>, but also handles 'set' magic.
10337 Usually used via its frontend C<sv_catpvf_mg>.
10343 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10345 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10347 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10352 =for apidoc sv_vsetpvfn
10354 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10357 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10363 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10364 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10366 PERL_ARGS_ASSERT_SV_VSETPVFN;
10369 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10374 * Warn of missing argument to sprintf, and then return a defined value
10375 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10377 #define WARN_MISSING WARN_UNINITIALIZED /* Not sure we want a new category */
10379 S_vcatpvfn_missing_argument(pTHX) {
10380 if (ckWARN(WARN_MISSING)) {
10381 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10382 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10389 S_expect_number(pTHX_ char **const pattern)
10394 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10396 switch (**pattern) {
10397 case '1': case '2': case '3':
10398 case '4': case '5': case '6':
10399 case '7': case '8': case '9':
10400 var = *(*pattern)++ - '0';
10401 while (isDIGIT(**pattern)) {
10402 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10404 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10412 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10414 const int neg = nv < 0;
10417 PERL_ARGS_ASSERT_F0CONVERT;
10425 if (uv & 1 && uv == nv)
10426 uv--; /* Round to even */
10428 const unsigned dig = uv % 10;
10430 } while (uv /= 10);
10441 =for apidoc sv_vcatpvfn
10443 =for apidoc sv_vcatpvfn_flags
10445 Processes its arguments like C<vsprintf> and appends the formatted output
10446 to an SV. Uses an array of SVs if the C style variable argument list is
10447 missing (NULL). When running with taint checks enabled, indicates via
10448 C<maybe_tainted> if results are untrustworthy (often due to the use of
10451 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10453 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10458 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10459 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10460 vec_utf8 = DO_UTF8(vecsv);
10462 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10465 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10466 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10468 PERL_ARGS_ASSERT_SV_VCATPVFN;
10470 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10474 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10475 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
10481 const char *patend;
10484 static const char nullstr[] = "(null)";
10486 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
10487 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
10489 /* Times 4: a decimal digit takes more than 3 binary digits.
10490 * NV_DIG: mantissa takes than many decimal digits.
10491 * Plus 32: Playing safe. */
10492 char ebuf[IV_DIG * 4 + NV_DIG + 32];
10493 /* large enough for "%#.#f" --chip */
10494 /* what about long double NVs? --jhi */
10496 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
10498 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
10499 PERL_UNUSED_ARG(maybe_tainted);
10501 if (flags & SV_GMAGIC)
10504 /* no matter what, this is a string now */
10505 (void)SvPV_force_nomg(sv, origlen);
10507 /* special-case "", "%s", and "%-p" (SVf - see below) */
10510 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
10512 const char * const s = va_arg(*args, char*);
10513 sv_catpv_nomg(sv, s ? s : nullstr);
10515 else if (svix < svmax) {
10516 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
10517 SvGETMAGIC(*svargs);
10518 sv_catsv_nomg(sv, *svargs);
10521 S_vcatpvfn_missing_argument(aTHX);
10524 if (args && patlen == 3 && pat[0] == '%' &&
10525 pat[1] == '-' && pat[2] == 'p') {
10526 argsv = MUTABLE_SV(va_arg(*args, void*));
10527 sv_catsv_nomg(sv, argsv);
10531 #ifndef USE_LONG_DOUBLE
10532 /* special-case "%.<number>[gf]" */
10533 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
10534 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
10535 unsigned digits = 0;
10539 while (*pp >= '0' && *pp <= '9')
10540 digits = 10 * digits + (*pp++ - '0');
10541 if (pp - pat == (int)patlen - 1 && svix < svmax) {
10542 const NV nv = SvNV(*svargs);
10544 /* Add check for digits != 0 because it seems that some
10545 gconverts are buggy in this case, and we don't yet have
10546 a Configure test for this. */
10547 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
10548 /* 0, point, slack */
10549 STORE_LC_NUMERIC_SET_TO_NEEDED();
10550 V_Gconvert(nv, (int)digits, 0, ebuf);
10551 sv_catpv_nomg(sv, ebuf);
10552 if (*ebuf) /* May return an empty string for digits==0 */
10555 } else if (!digits) {
10558 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
10559 sv_catpvn_nomg(sv, p, l);
10565 #endif /* !USE_LONG_DOUBLE */
10567 if (!args && svix < svmax && DO_UTF8(*svargs))
10570 patend = (char*)pat + patlen;
10571 for (p = (char*)pat; p < patend; p = q) {
10574 bool vectorize = FALSE;
10575 bool vectorarg = FALSE;
10576 bool vec_utf8 = FALSE;
10582 bool has_precis = FALSE;
10584 const I32 osvix = svix;
10585 bool is_utf8 = FALSE; /* is this item utf8? */
10586 #ifdef HAS_LDBL_SPRINTF_BUG
10587 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
10588 with sfio - Allen <allens@cpan.org> */
10589 bool fix_ldbl_sprintf_bug = FALSE;
10593 U8 utf8buf[UTF8_MAXBYTES+1];
10594 STRLEN esignlen = 0;
10596 const char *eptr = NULL;
10597 const char *fmtstart;
10600 const U8 *vecstr = NULL;
10607 /* we need a long double target in case HAS_LONG_DOUBLE but
10608 not USE_LONG_DOUBLE
10610 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE
10618 const char *dotstr = ".";
10619 STRLEN dotstrlen = 1;
10620 I32 efix = 0; /* explicit format parameter index */
10621 I32 ewix = 0; /* explicit width index */
10622 I32 epix = 0; /* explicit precision index */
10623 I32 evix = 0; /* explicit vector index */
10624 bool asterisk = FALSE;
10626 /* echo everything up to the next format specification */
10627 for (q = p; q < patend && *q != '%'; ++q) ;
10629 if (has_utf8 && !pat_utf8)
10630 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
10632 sv_catpvn_nomg(sv, p, q - p);
10641 We allow format specification elements in this order:
10642 \d+\$ explicit format parameter index
10644 v|\*(\d+\$)?v vector with optional (optionally specified) arg
10645 0 flag (as above): repeated to allow "v02"
10646 \d+|\*(\d+\$)? width using optional (optionally specified) arg
10647 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
10649 [%bcdefginopsuxDFOUX] format (mandatory)
10654 As of perl5.9.3, printf format checking is on by default.
10655 Internally, perl uses %p formats to provide an escape to
10656 some extended formatting. This block deals with those
10657 extensions: if it does not match, (char*)q is reset and
10658 the normal format processing code is used.
10660 Currently defined extensions are:
10661 %p include pointer address (standard)
10662 %-p (SVf) include an SV (previously %_)
10663 %-<num>p include an SV with precision <num>
10665 %3p include a HEK with precision of 256
10666 %4p char* preceded by utf8 flag and length
10667 %<num>p (where num is 1 or > 4) reserved for future
10670 Robin Barker 2005-07-14 (but modified since)
10672 %1p (VDf) removed. RMB 2007-10-19
10679 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
10680 /* The argument has already gone through cBOOL, so the cast
10682 is_utf8 = (bool)va_arg(*args, int);
10683 elen = va_arg(*args, UV);
10684 eptr = va_arg(*args, char *);
10685 q += sizeof(UTF8f)-1;
10688 n = expect_number(&q);
10690 if (sv) { /* SVf */
10695 argsv = MUTABLE_SV(va_arg(*args, void*));
10696 eptr = SvPV_const(argsv, elen);
10697 if (DO_UTF8(argsv))
10701 else if (n==2 || n==3) { /* HEKf */
10702 HEK * const hek = va_arg(*args, HEK *);
10703 eptr = HEK_KEY(hek);
10704 elen = HEK_LEN(hek);
10705 if (HEK_UTF8(hek)) is_utf8 = TRUE;
10706 if (n==3) precis = 256, has_precis = TRUE;
10710 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
10711 "internal %%<num>p might conflict with future printf extensions");
10717 if ( (width = expect_number(&q)) ) {
10732 if (plus == '+' && *q == ' ') /* '+' over ' ' */
10761 if ( (ewix = expect_number(&q)) )
10770 if ((vectorarg = asterisk)) {
10783 width = expect_number(&q);
10786 if (vectorize && vectorarg) {
10787 /* vectorizing, but not with the default "." */
10789 vecsv = va_arg(*args, SV*);
10791 vecsv = (evix > 0 && evix <= svmax)
10792 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
10794 vecsv = svix < svmax
10795 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
10797 dotstr = SvPV_const(vecsv, dotstrlen);
10798 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
10799 bad with tied or overloaded values that return UTF8. */
10800 if (DO_UTF8(vecsv))
10802 else if (has_utf8) {
10803 vecsv = sv_mortalcopy(vecsv);
10804 sv_utf8_upgrade(vecsv);
10805 dotstr = SvPV_const(vecsv, dotstrlen);
10812 i = va_arg(*args, int);
10814 i = (ewix ? ewix <= svmax : svix < svmax) ?
10815 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
10817 width = (i < 0) ? -i : i;
10827 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
10829 /* XXX: todo, support specified precision parameter */
10833 i = va_arg(*args, int);
10835 i = (ewix ? ewix <= svmax : svix < svmax)
10836 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
10838 has_precis = !(i < 0);
10842 while (isDIGIT(*q))
10843 precis = precis * 10 + (*q++ - '0');
10852 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
10853 vecsv = svargs[efix ? efix-1 : svix++];
10854 vecstr = (U8*)SvPV_const(vecsv,veclen);
10855 vec_utf8 = DO_UTF8(vecsv);
10857 /* if this is a version object, we need to convert
10858 * back into v-string notation and then let the
10859 * vectorize happen normally
10861 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
10862 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
10863 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
10864 "vector argument not supported with alpha versions");
10867 vecsv = sv_newmortal();
10868 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
10870 vecstr = (U8*)SvPV_const(vecsv, veclen);
10871 vec_utf8 = DO_UTF8(vecsv);
10885 case 'I': /* Ix, I32x, and I64x */
10886 # ifdef USE_64_BIT_INT
10887 if (q[1] == '6' && q[2] == '4') {
10893 if (q[1] == '3' && q[2] == '2') {
10897 # ifdef USE_64_BIT_INT
10903 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
10915 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
10916 if (*q == 'l') { /* lld, llf */
10925 if (*++q == 'h') { /* hhd, hhu */
10954 if (!vectorize && !args) {
10956 const I32 i = efix-1;
10957 argsv = (i >= 0 && i < svmax)
10958 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
10960 argsv = (svix >= 0 && svix < svmax)
10961 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
10965 switch (c = *q++) {
10972 uv = (args) ? va_arg(*args, int) : SvIV(argsv);
10974 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
10976 eptr = (char*)utf8buf;
10977 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
10991 eptr = va_arg(*args, char*);
10993 elen = strlen(eptr);
10995 eptr = (char *)nullstr;
10996 elen = sizeof nullstr - 1;
11000 eptr = SvPV_const(argsv, elen);
11001 if (DO_UTF8(argsv)) {
11002 STRLEN old_precis = precis;
11003 if (has_precis && precis < elen) {
11004 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11005 STRLEN p = precis > ulen ? ulen : precis;
11006 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11007 /* sticks at end */
11009 if (width) { /* fudge width (can't fudge elen) */
11010 if (has_precis && precis < elen)
11011 width += precis - old_precis;
11014 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11021 if (has_precis && precis < elen)
11028 if (alt || vectorize)
11030 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11048 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11057 esignbuf[esignlen++] = plus;
11061 case 'c': iv = (char)va_arg(*args, int); break;
11062 case 'h': iv = (short)va_arg(*args, int); break;
11063 case 'l': iv = va_arg(*args, long); break;
11064 case 'V': iv = va_arg(*args, IV); break;
11065 case 'z': iv = va_arg(*args, SSize_t); break;
11066 case 't': iv = va_arg(*args, ptrdiff_t); break;
11067 default: iv = va_arg(*args, int); break;
11069 case 'j': iv = va_arg(*args, intmax_t); break;
11073 iv = va_arg(*args, Quad_t); break;
11080 IV tiv = SvIV(argsv); /* work around GCC bug #13488 */
11082 case 'c': iv = (char)tiv; break;
11083 case 'h': iv = (short)tiv; break;
11084 case 'l': iv = (long)tiv; break;
11086 default: iv = tiv; break;
11089 iv = (Quad_t)tiv; break;
11095 if ( !vectorize ) /* we already set uv above */
11100 esignbuf[esignlen++] = plus;
11104 esignbuf[esignlen++] = '-';
11148 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11159 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11160 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11161 case 'l': uv = va_arg(*args, unsigned long); break;
11162 case 'V': uv = va_arg(*args, UV); break;
11163 case 'z': uv = va_arg(*args, Size_t); break;
11164 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11166 case 'j': uv = va_arg(*args, uintmax_t); break;
11168 default: uv = va_arg(*args, unsigned); break;
11171 uv = va_arg(*args, Uquad_t); break;
11178 UV tuv = SvUV(argsv); /* work around GCC bug #13488 */
11180 case 'c': uv = (unsigned char)tuv; break;
11181 case 'h': uv = (unsigned short)tuv; break;
11182 case 'l': uv = (unsigned long)tuv; break;
11184 default: uv = tuv; break;
11187 uv = (Uquad_t)tuv; break;
11196 char *ptr = ebuf + sizeof ebuf;
11197 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11203 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11207 } while (uv >>= 4);
11209 esignbuf[esignlen++] = '0';
11210 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11216 *--ptr = '0' + dig;
11217 } while (uv >>= 3);
11218 if (alt && *ptr != '0')
11224 *--ptr = '0' + dig;
11225 } while (uv >>= 1);
11227 esignbuf[esignlen++] = '0';
11228 esignbuf[esignlen++] = c;
11231 default: /* it had better be ten or less */
11234 *--ptr = '0' + dig;
11235 } while (uv /= base);
11238 elen = (ebuf + sizeof ebuf) - ptr;
11242 zeros = precis - elen;
11243 else if (precis == 0 && elen == 1 && *eptr == '0'
11244 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11247 /* a precision nullifies the 0 flag. */
11254 /* FLOATING POINT */
11257 c = 'f'; /* maybe %F isn't supported here */
11259 case 'e': case 'E':
11261 case 'g': case 'G':
11265 /* This is evil, but floating point is even more evil */
11267 /* for SV-style calling, we can only get NV
11268 for C-style calling, we assume %f is double;
11269 for simplicity we allow any of %Lf, %llf, %qf for long double
11273 #if defined(USE_LONG_DOUBLE)
11277 /* [perl #20339] - we should accept and ignore %lf rather than die */
11281 #if defined(USE_LONG_DOUBLE)
11282 intsize = args ? 0 : 'q';
11286 #if defined(HAS_LONG_DOUBLE)
11299 /* now we need (long double) if intsize == 'q', else (double) */
11301 #if LONG_DOUBLESIZE > DOUBLESIZE
11303 va_arg(*args, long double) :
11304 va_arg(*args, double)
11306 va_arg(*args, double)
11311 /* nv * 0 will be NaN for NaN, +Inf and -Inf, and 0 for anything
11312 else. frexp() has some unspecified behaviour for those three */
11313 if (c != 'e' && c != 'E' && (nv * 0) == 0) {
11315 /* FIXME: if HAS_LONG_DOUBLE but not USE_LONG_DOUBLE this
11316 will cast our (long double) to (double) */
11317 (void)Perl_frexp(nv, &i);
11318 if (i == PERL_INT_MIN)
11319 Perl_die(aTHX_ "panic: frexp");
11321 need = BIT_DIGITS(i);
11323 need += has_precis ? precis : 6; /* known default */
11328 #ifdef HAS_LDBL_SPRINTF_BUG
11329 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11330 with sfio - Allen <allens@cpan.org> */
11333 # define MY_DBL_MAX DBL_MAX
11334 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11335 # if DOUBLESIZE >= 8
11336 # define MY_DBL_MAX 1.7976931348623157E+308L
11338 # define MY_DBL_MAX 3.40282347E+38L
11342 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
11343 # define MY_DBL_MAX_BUG 1L
11345 # define MY_DBL_MAX_BUG MY_DBL_MAX
11349 # define MY_DBL_MIN DBL_MIN
11350 # else /* XXX guessing! -Allen */
11351 # if DOUBLESIZE >= 8
11352 # define MY_DBL_MIN 2.2250738585072014E-308L
11354 # define MY_DBL_MIN 1.17549435E-38L
11358 if ((intsize == 'q') && (c == 'f') &&
11359 ((nv < MY_DBL_MAX_BUG) && (nv > -MY_DBL_MAX_BUG)) &&
11360 (need < DBL_DIG)) {
11361 /* it's going to be short enough that
11362 * long double precision is not needed */
11364 if ((nv <= 0L) && (nv >= -0L))
11365 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
11367 /* would use Perl_fp_class as a double-check but not
11368 * functional on IRIX - see perl.h comments */
11370 if ((nv >= MY_DBL_MIN) || (nv <= -MY_DBL_MIN)) {
11371 /* It's within the range that a double can represent */
11372 #if defined(DBL_MAX) && !defined(DBL_MIN)
11373 if ((nv >= ((long double)1/DBL_MAX)) ||
11374 (nv <= (-(long double)1/DBL_MAX)))
11376 fix_ldbl_sprintf_bug = TRUE;
11379 if (fix_ldbl_sprintf_bug == TRUE) {
11389 # undef MY_DBL_MAX_BUG
11392 #endif /* HAS_LDBL_SPRINTF_BUG */
11394 need += 20; /* fudge factor */
11395 if (PL_efloatsize < need) {
11396 Safefree(PL_efloatbuf);
11397 PL_efloatsize = need + 20; /* more fudge */
11398 Newx(PL_efloatbuf, PL_efloatsize, char);
11399 PL_efloatbuf[0] = '\0';
11402 if ( !(width || left || plus || alt) && fill != '0'
11403 && has_precis && intsize != 'q' ) { /* Shortcuts */
11404 /* See earlier comment about buggy Gconvert when digits,
11406 if ( c == 'g' && precis) {
11407 STORE_LC_NUMERIC_SET_TO_NEEDED();
11408 V_Gconvert((NV)nv, (int)precis, 0, PL_efloatbuf);
11409 /* May return an empty string for digits==0 */
11410 if (*PL_efloatbuf) {
11411 elen = strlen(PL_efloatbuf);
11412 goto float_converted;
11414 } else if ( c == 'f' && !precis) {
11415 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
11420 char *ptr = ebuf + sizeof ebuf;
11423 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
11424 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
11425 if (intsize == 'q') {
11426 /* Copy the one or more characters in a long double
11427 * format before the 'base' ([efgEFG]) character to
11428 * the format string. */
11429 static char const prifldbl[] = PERL_PRIfldbl;
11430 char const *p = prifldbl + sizeof(prifldbl) - 3;
11431 while (p >= prifldbl) { *--ptr = *p--; }
11436 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11441 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11453 /* No taint. Otherwise we are in the strange situation
11454 * where printf() taints but print($float) doesn't.
11457 STORE_LC_NUMERIC_SET_TO_NEEDED();
11459 /* hopefully the above makes ptr a very constrained format
11460 * that is safe to use, even though it's not literal */
11461 GCC_DIAG_IGNORE(-Wformat-nonliteral);
11462 #if defined(HAS_LONG_DOUBLE)
11463 elen = ((intsize == 'q')
11464 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, nv)
11465 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)nv));
11467 elen = my_sprintf(PL_efloatbuf, ptr, nv);
11472 eptr = PL_efloatbuf;
11474 #ifdef USE_LOCALE_NUMERIC
11475 /* If the decimal point character in the string is UTF-8, make the
11477 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
11478 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
11491 i = SvCUR(sv) - origlen;
11494 case 'c': *(va_arg(*args, char*)) = i; break;
11495 case 'h': *(va_arg(*args, short*)) = i; break;
11496 default: *(va_arg(*args, int*)) = i; break;
11497 case 'l': *(va_arg(*args, long*)) = i; break;
11498 case 'V': *(va_arg(*args, IV*)) = i; break;
11499 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
11500 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
11502 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
11506 *(va_arg(*args, Quad_t*)) = i; break;
11513 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
11514 continue; /* not "break" */
11521 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
11522 && ckWARN(WARN_PRINTF))
11524 SV * const msg = sv_newmortal();
11525 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
11526 (PL_op->op_type == OP_PRTF) ? "" : "s");
11527 if (fmtstart < patend) {
11528 const char * const fmtend = q < patend ? q : patend;
11530 sv_catpvs(msg, "\"%");
11531 for (f = fmtstart; f < fmtend; f++) {
11533 sv_catpvn_nomg(msg, f, 1);
11535 Perl_sv_catpvf(aTHX_ msg,
11536 "\\%03"UVof, (UV)*f & 0xFF);
11539 sv_catpvs(msg, "\"");
11541 sv_catpvs(msg, "end of string");
11543 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
11546 /* output mangled stuff ... */
11552 /* ... right here, because formatting flags should not apply */
11553 SvGROW(sv, SvCUR(sv) + elen + 1);
11555 Copy(eptr, p, elen, char);
11558 SvCUR_set(sv, p - SvPVX_const(sv));
11560 continue; /* not "break" */
11563 if (is_utf8 != has_utf8) {
11566 sv_utf8_upgrade(sv);
11569 const STRLEN old_elen = elen;
11570 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
11571 sv_utf8_upgrade(nsv);
11572 eptr = SvPVX_const(nsv);
11575 if (width) { /* fudge width (can't fudge elen) */
11576 width += elen - old_elen;
11582 have = esignlen + zeros + elen;
11584 croak_memory_wrap();
11586 need = (have > width ? have : width);
11589 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
11590 croak_memory_wrap();
11591 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
11593 if (esignlen && fill == '0') {
11595 for (i = 0; i < (int)esignlen; i++)
11596 *p++ = esignbuf[i];
11598 if (gap && !left) {
11599 memset(p, fill, gap);
11602 if (esignlen && fill != '0') {
11604 for (i = 0; i < (int)esignlen; i++)
11605 *p++ = esignbuf[i];
11609 for (i = zeros; i; i--)
11613 Copy(eptr, p, elen, char);
11617 memset(p, ' ', gap);
11622 Copy(dotstr, p, dotstrlen, char);
11626 vectorize = FALSE; /* done iterating over vecstr */
11633 SvCUR_set(sv, p - SvPVX_const(sv));
11641 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
11645 /* =========================================================================
11647 =head1 Cloning an interpreter
11649 All the macros and functions in this section are for the private use of
11650 the main function, perl_clone().
11652 The foo_dup() functions make an exact copy of an existing foo thingy.
11653 During the course of a cloning, a hash table is used to map old addresses
11654 to new addresses. The table is created and manipulated with the
11655 ptr_table_* functions.
11659 * =========================================================================*/
11662 #if defined(USE_ITHREADS)
11664 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
11665 #ifndef GpREFCNT_inc
11666 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
11670 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
11671 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
11672 If this changes, please unmerge ss_dup.
11673 Likewise, sv_dup_inc_multiple() relies on this fact. */
11674 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
11675 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
11676 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11677 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
11678 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11679 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
11680 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
11681 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
11682 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
11683 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
11684 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
11685 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
11686 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11688 /* clone a parser */
11691 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
11695 PERL_ARGS_ASSERT_PARSER_DUP;
11700 /* look for it in the table first */
11701 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
11705 /* create anew and remember what it is */
11706 Newxz(parser, 1, yy_parser);
11707 ptr_table_store(PL_ptr_table, proto, parser);
11709 /* XXX these not yet duped */
11710 parser->old_parser = NULL;
11711 parser->stack = NULL;
11713 parser->stack_size = 0;
11714 /* XXX parser->stack->state = 0; */
11716 /* XXX eventually, just Copy() most of the parser struct ? */
11718 parser->lex_brackets = proto->lex_brackets;
11719 parser->lex_casemods = proto->lex_casemods;
11720 parser->lex_brackstack = savepvn(proto->lex_brackstack,
11721 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
11722 parser->lex_casestack = savepvn(proto->lex_casestack,
11723 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
11724 parser->lex_defer = proto->lex_defer;
11725 parser->lex_dojoin = proto->lex_dojoin;
11726 parser->lex_expect = proto->lex_expect;
11727 parser->lex_formbrack = proto->lex_formbrack;
11728 parser->lex_inpat = proto->lex_inpat;
11729 parser->lex_inwhat = proto->lex_inwhat;
11730 parser->lex_op = proto->lex_op;
11731 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
11732 parser->lex_starts = proto->lex_starts;
11733 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
11734 parser->multi_close = proto->multi_close;
11735 parser->multi_open = proto->multi_open;
11736 parser->multi_start = proto->multi_start;
11737 parser->multi_end = proto->multi_end;
11738 parser->preambled = proto->preambled;
11739 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
11740 parser->linestr = sv_dup_inc(proto->linestr, param);
11741 parser->expect = proto->expect;
11742 parser->copline = proto->copline;
11743 parser->last_lop_op = proto->last_lop_op;
11744 parser->lex_state = proto->lex_state;
11745 parser->rsfp = fp_dup(proto->rsfp, '<', param);
11746 /* rsfp_filters entries have fake IoDIRP() */
11747 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
11748 parser->in_my = proto->in_my;
11749 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
11750 parser->error_count = proto->error_count;
11753 parser->linestr = sv_dup_inc(proto->linestr, param);
11756 char * const ols = SvPVX(proto->linestr);
11757 char * const ls = SvPVX(parser->linestr);
11759 parser->bufptr = ls + (proto->bufptr >= ols ?
11760 proto->bufptr - ols : 0);
11761 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
11762 proto->oldbufptr - ols : 0);
11763 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
11764 proto->oldoldbufptr - ols : 0);
11765 parser->linestart = ls + (proto->linestart >= ols ?
11766 proto->linestart - ols : 0);
11767 parser->last_uni = ls + (proto->last_uni >= ols ?
11768 proto->last_uni - ols : 0);
11769 parser->last_lop = ls + (proto->last_lop >= ols ?
11770 proto->last_lop - ols : 0);
11772 parser->bufend = ls + SvCUR(parser->linestr);
11775 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
11779 parser->endwhite = proto->endwhite;
11780 parser->faketokens = proto->faketokens;
11781 parser->lasttoke = proto->lasttoke;
11782 parser->nextwhite = proto->nextwhite;
11783 parser->realtokenstart = proto->realtokenstart;
11784 parser->skipwhite = proto->skipwhite;
11785 parser->thisclose = proto->thisclose;
11786 parser->thismad = proto->thismad;
11787 parser->thisopen = proto->thisopen;
11788 parser->thisstuff = proto->thisstuff;
11789 parser->thistoken = proto->thistoken;
11790 parser->thiswhite = proto->thiswhite;
11792 Copy(proto->nexttoke, parser->nexttoke, 5, NEXTTOKE);
11793 parser->curforce = proto->curforce;
11795 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
11796 Copy(proto->nexttype, parser->nexttype, 5, I32);
11797 parser->nexttoke = proto->nexttoke;
11800 /* XXX should clone saved_curcop here, but we aren't passed
11801 * proto_perl; so do it in perl_clone_using instead */
11807 /* duplicate a file handle */
11810 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
11814 PERL_ARGS_ASSERT_FP_DUP;
11815 PERL_UNUSED_ARG(type);
11818 return (PerlIO*)NULL;
11820 /* look for it in the table first */
11821 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
11825 /* create anew and remember what it is */
11826 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
11827 ptr_table_store(PL_ptr_table, fp, ret);
11831 /* duplicate a directory handle */
11834 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
11838 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
11841 const Direntry_t *dirent;
11842 char smallbuf[256];
11848 PERL_UNUSED_CONTEXT;
11849 PERL_ARGS_ASSERT_DIRP_DUP;
11854 /* look for it in the table first */
11855 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
11859 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
11861 PERL_UNUSED_ARG(param);
11865 /* open the current directory (so we can switch back) */
11866 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
11868 /* chdir to our dir handle and open the present working directory */
11869 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
11870 PerlDir_close(pwd);
11871 return (DIR *)NULL;
11873 /* Now we should have two dir handles pointing to the same dir. */
11875 /* Be nice to the calling code and chdir back to where we were. */
11876 rc = fchdir(my_dirfd(pwd));
11877 /* XXX If this fails, then what? */
11878 PERL_UNUSED_VAR(rc);
11880 /* We have no need of the pwd handle any more. */
11881 PerlDir_close(pwd);
11884 # define d_namlen(d) (d)->d_namlen
11886 # define d_namlen(d) strlen((d)->d_name)
11888 /* Iterate once through dp, to get the file name at the current posi-
11889 tion. Then step back. */
11890 pos = PerlDir_tell(dp);
11891 if ((dirent = PerlDir_read(dp))) {
11892 len = d_namlen(dirent);
11893 if (len <= sizeof smallbuf) name = smallbuf;
11894 else Newx(name, len, char);
11895 Move(dirent->d_name, name, len, char);
11897 PerlDir_seek(dp, pos);
11899 /* Iterate through the new dir handle, till we find a file with the
11901 if (!dirent) /* just before the end */
11903 pos = PerlDir_tell(ret);
11904 if (PerlDir_read(ret)) continue; /* not there yet */
11905 PerlDir_seek(ret, pos); /* step back */
11909 const long pos0 = PerlDir_tell(ret);
11911 pos = PerlDir_tell(ret);
11912 if ((dirent = PerlDir_read(ret))) {
11913 if (len == d_namlen(dirent)
11914 && memEQ(name, dirent->d_name, len)) {
11916 PerlDir_seek(ret, pos); /* step back */
11919 /* else we are not there yet; keep iterating */
11921 else { /* This is not meant to happen. The best we can do is
11922 reset the iterator to the beginning. */
11923 PerlDir_seek(ret, pos0);
11930 if (name && name != smallbuf)
11935 ret = win32_dirp_dup(dp, param);
11938 /* pop it in the pointer table */
11940 ptr_table_store(PL_ptr_table, dp, ret);
11945 /* duplicate a typeglob */
11948 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
11952 PERL_ARGS_ASSERT_GP_DUP;
11956 /* look for it in the table first */
11957 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
11961 /* create anew and remember what it is */
11963 ptr_table_store(PL_ptr_table, gp, ret);
11966 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
11967 on Newxz() to do this for us. */
11968 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
11969 ret->gp_io = io_dup_inc(gp->gp_io, param);
11970 ret->gp_form = cv_dup_inc(gp->gp_form, param);
11971 ret->gp_av = av_dup_inc(gp->gp_av, param);
11972 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
11973 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
11974 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
11975 ret->gp_cvgen = gp->gp_cvgen;
11976 ret->gp_line = gp->gp_line;
11977 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
11981 /* duplicate a chain of magic */
11984 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
11986 MAGIC *mgret = NULL;
11987 MAGIC **mgprev_p = &mgret;
11989 PERL_ARGS_ASSERT_MG_DUP;
11991 for (; mg; mg = mg->mg_moremagic) {
11994 if ((param->flags & CLONEf_JOIN_IN)
11995 && mg->mg_type == PERL_MAGIC_backref)
11996 /* when joining, we let the individual SVs add themselves to
11997 * backref as needed. */
12000 Newx(nmg, 1, MAGIC);
12002 mgprev_p = &(nmg->mg_moremagic);
12004 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12005 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12006 from the original commit adding Perl_mg_dup() - revision 4538.
12007 Similarly there is the annotation "XXX random ptr?" next to the
12008 assignment to nmg->mg_ptr. */
12011 /* FIXME for plugins
12012 if (nmg->mg_type == PERL_MAGIC_qr) {
12013 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12017 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12018 ? nmg->mg_type == PERL_MAGIC_backref
12019 /* The backref AV has its reference
12020 * count deliberately bumped by 1 */
12021 ? SvREFCNT_inc(av_dup_inc((const AV *)
12022 nmg->mg_obj, param))
12023 : sv_dup_inc(nmg->mg_obj, param)
12024 : sv_dup(nmg->mg_obj, param);
12026 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12027 if (nmg->mg_len > 0) {
12028 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12029 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12030 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12032 AMT * const namtp = (AMT*)nmg->mg_ptr;
12033 sv_dup_inc_multiple((SV**)(namtp->table),
12034 (SV**)(namtp->table), NofAMmeth, param);
12037 else if (nmg->mg_len == HEf_SVKEY)
12038 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12040 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12041 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12047 #endif /* USE_ITHREADS */
12049 struct ptr_tbl_arena {
12050 struct ptr_tbl_arena *next;
12051 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12054 /* create a new pointer-mapping table */
12057 Perl_ptr_table_new(pTHX)
12060 PERL_UNUSED_CONTEXT;
12062 Newx(tbl, 1, PTR_TBL_t);
12063 tbl->tbl_max = 511;
12064 tbl->tbl_items = 0;
12065 tbl->tbl_arena = NULL;
12066 tbl->tbl_arena_next = NULL;
12067 tbl->tbl_arena_end = NULL;
12068 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12072 #define PTR_TABLE_HASH(ptr) \
12073 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12075 /* map an existing pointer using a table */
12077 STATIC PTR_TBL_ENT_t *
12078 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12080 PTR_TBL_ENT_t *tblent;
12081 const UV hash = PTR_TABLE_HASH(sv);
12083 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12085 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12086 for (; tblent; tblent = tblent->next) {
12087 if (tblent->oldval == sv)
12094 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12096 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12098 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12099 PERL_UNUSED_CONTEXT;
12101 return tblent ? tblent->newval : NULL;
12104 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12105 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12106 * the core's typical use of ptr_tables in thread cloning. */
12109 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12111 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12113 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12114 PERL_UNUSED_CONTEXT;
12117 tblent->newval = newsv;
12119 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12121 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12122 struct ptr_tbl_arena *new_arena;
12124 Newx(new_arena, 1, struct ptr_tbl_arena);
12125 new_arena->next = tbl->tbl_arena;
12126 tbl->tbl_arena = new_arena;
12127 tbl->tbl_arena_next = new_arena->array;
12128 tbl->tbl_arena_end = new_arena->array
12129 + sizeof(new_arena->array) / sizeof(new_arena->array[0]);
12132 tblent = tbl->tbl_arena_next++;
12134 tblent->oldval = oldsv;
12135 tblent->newval = newsv;
12136 tblent->next = tbl->tbl_ary[entry];
12137 tbl->tbl_ary[entry] = tblent;
12139 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12140 ptr_table_split(tbl);
12144 /* double the hash bucket size of an existing ptr table */
12147 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
12149 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
12150 const UV oldsize = tbl->tbl_max + 1;
12151 UV newsize = oldsize * 2;
12154 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
12155 PERL_UNUSED_CONTEXT;
12157 Renew(ary, newsize, PTR_TBL_ENT_t*);
12158 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
12159 tbl->tbl_max = --newsize;
12160 tbl->tbl_ary = ary;
12161 for (i=0; i < oldsize; i++, ary++) {
12162 PTR_TBL_ENT_t **entp = ary;
12163 PTR_TBL_ENT_t *ent = *ary;
12164 PTR_TBL_ENT_t **curentp;
12167 curentp = ary + oldsize;
12169 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
12171 ent->next = *curentp;
12181 /* remove all the entries from a ptr table */
12182 /* Deprecated - will be removed post 5.14 */
12185 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
12187 if (tbl && tbl->tbl_items) {
12188 struct ptr_tbl_arena *arena = tbl->tbl_arena;
12190 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
12193 struct ptr_tbl_arena *next = arena->next;
12199 tbl->tbl_items = 0;
12200 tbl->tbl_arena = NULL;
12201 tbl->tbl_arena_next = NULL;
12202 tbl->tbl_arena_end = NULL;
12206 /* clear and free a ptr table */
12209 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
12211 struct ptr_tbl_arena *arena;
12217 arena = tbl->tbl_arena;
12220 struct ptr_tbl_arena *next = arena->next;
12226 Safefree(tbl->tbl_ary);
12230 #if defined(USE_ITHREADS)
12233 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
12235 PERL_ARGS_ASSERT_RVPV_DUP;
12237 assert(!isREGEXP(sstr));
12239 if (SvWEAKREF(sstr)) {
12240 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
12241 if (param->flags & CLONEf_JOIN_IN) {
12242 /* if joining, we add any back references individually rather
12243 * than copying the whole backref array */
12244 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
12248 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
12250 else if (SvPVX_const(sstr)) {
12251 /* Has something there */
12253 /* Normal PV - clone whole allocated space */
12254 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
12255 /* sstr may not be that normal, but actually copy on write.
12256 But we are a true, independent SV, so: */
12260 /* Special case - not normally malloced for some reason */
12261 if (isGV_with_GP(sstr)) {
12262 /* Don't need to do anything here. */
12264 else if ((SvIsCOW(sstr))) {
12265 /* A "shared" PV - clone it as "shared" PV */
12267 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
12271 /* Some other special case - random pointer */
12272 SvPV_set(dstr, (char *) SvPVX_const(sstr));
12277 /* Copy the NULL */
12278 SvPV_set(dstr, NULL);
12282 /* duplicate a list of SVs. source and dest may point to the same memory. */
12284 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
12285 SSize_t items, CLONE_PARAMS *const param)
12287 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
12289 while (items-- > 0) {
12290 *dest++ = sv_dup_inc(*source++, param);
12296 /* duplicate an SV of any type (including AV, HV etc) */
12299 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12304 PERL_ARGS_ASSERT_SV_DUP_COMMON;
12306 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
12307 #ifdef DEBUG_LEAKING_SCALARS_ABORT
12312 /* look for it in the table first */
12313 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
12317 if(param->flags & CLONEf_JOIN_IN) {
12318 /** We are joining here so we don't want do clone
12319 something that is bad **/
12320 if (SvTYPE(sstr) == SVt_PVHV) {
12321 const HEK * const hvname = HvNAME_HEK(sstr);
12323 /** don't clone stashes if they already exist **/
12324 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12325 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
12326 ptr_table_store(PL_ptr_table, sstr, dstr);
12330 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
12331 HV *stash = GvSTASH(sstr);
12332 const HEK * hvname;
12333 if (stash && (hvname = HvNAME_HEK(stash))) {
12334 /** don't clone GVs if they already exist **/
12336 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12337 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
12339 stash, GvNAME(sstr),
12345 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
12346 ptr_table_store(PL_ptr_table, sstr, *svp);
12353 /* create anew and remember what it is */
12356 #ifdef DEBUG_LEAKING_SCALARS
12357 dstr->sv_debug_optype = sstr->sv_debug_optype;
12358 dstr->sv_debug_line = sstr->sv_debug_line;
12359 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
12360 dstr->sv_debug_parent = (SV*)sstr;
12361 FREE_SV_DEBUG_FILE(dstr);
12362 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
12365 ptr_table_store(PL_ptr_table, sstr, dstr);
12368 SvFLAGS(dstr) = SvFLAGS(sstr);
12369 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
12370 SvREFCNT(dstr) = 0; /* must be before any other dups! */
12373 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
12374 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
12375 (void*)PL_watch_pvx, SvPVX_const(sstr));
12378 /* don't clone objects whose class has asked us not to */
12379 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
12384 switch (SvTYPE(sstr)) {
12386 SvANY(dstr) = NULL;
12389 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
12391 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12393 SvIV_set(dstr, SvIVX(sstr));
12397 SvANY(dstr) = new_XNV();
12398 SvNV_set(dstr, SvNVX(sstr));
12402 /* These are all the types that need complex bodies allocating. */
12404 const svtype sv_type = SvTYPE(sstr);
12405 const struct body_details *const sv_type_details
12406 = bodies_by_type + sv_type;
12410 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
12426 assert(sv_type_details->body_size);
12427 if (sv_type_details->arena) {
12428 new_body_inline(new_body, sv_type);
12430 = (void*)((char*)new_body - sv_type_details->offset);
12432 new_body = new_NOARENA(sv_type_details);
12436 SvANY(dstr) = new_body;
12439 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
12440 ((char*)SvANY(dstr)) + sv_type_details->offset,
12441 sv_type_details->copy, char);
12443 Copy(((char*)SvANY(sstr)),
12444 ((char*)SvANY(dstr)),
12445 sv_type_details->body_size + sv_type_details->offset, char);
12448 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
12449 && !isGV_with_GP(dstr)
12451 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
12452 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12454 /* The Copy above means that all the source (unduplicated) pointers
12455 are now in the destination. We can check the flags and the
12456 pointers in either, but it's possible that there's less cache
12457 missing by always going for the destination.
12458 FIXME - instrument and check that assumption */
12459 if (sv_type >= SVt_PVMG) {
12460 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
12461 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
12462 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
12464 } else if (SvMAGIC(dstr))
12465 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
12466 if (SvOBJECT(dstr) && SvSTASH(dstr))
12467 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
12468 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
12471 /* The cast silences a GCC warning about unhandled types. */
12472 switch ((int)sv_type) {
12483 /* FIXME for plugins */
12484 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
12485 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
12488 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
12489 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
12490 LvTARG(dstr) = dstr;
12491 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
12492 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
12494 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
12495 if (isREGEXP(sstr)) goto duprex;
12497 /* non-GP case already handled above */
12498 if(isGV_with_GP(sstr)) {
12499 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
12500 /* Don't call sv_add_backref here as it's going to be
12501 created as part of the magic cloning of the symbol
12502 table--unless this is during a join and the stash
12503 is not actually being cloned. */
12504 /* Danger Will Robinson - GvGP(dstr) isn't initialised
12505 at the point of this comment. */
12506 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
12507 if (param->flags & CLONEf_JOIN_IN)
12508 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
12509 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
12510 (void)GpREFCNT_inc(GvGP(dstr));
12514 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
12515 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
12516 /* I have no idea why fake dirp (rsfps)
12517 should be treated differently but otherwise
12518 we end up with leaks -- sky*/
12519 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
12520 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
12521 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
12523 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
12524 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
12525 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
12526 if (IoDIRP(dstr)) {
12527 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
12530 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
12532 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
12534 if (IoOFP(dstr) == IoIFP(sstr))
12535 IoOFP(dstr) = IoIFP(dstr);
12537 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
12538 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
12539 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
12540 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
12543 /* avoid cloning an empty array */
12544 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
12545 SV **dst_ary, **src_ary;
12546 SSize_t items = AvFILLp((const AV *)sstr) + 1;
12548 src_ary = AvARRAY((const AV *)sstr);
12549 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
12550 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
12551 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
12552 AvALLOC((const AV *)dstr) = dst_ary;
12553 if (AvREAL((const AV *)sstr)) {
12554 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
12558 while (items-- > 0)
12559 *dst_ary++ = sv_dup(*src_ary++, param);
12561 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
12562 while (items-- > 0) {
12563 *dst_ary++ = &PL_sv_undef;
12567 AvARRAY(MUTABLE_AV(dstr)) = NULL;
12568 AvALLOC((const AV *)dstr) = (SV**)NULL;
12569 AvMAX( (const AV *)dstr) = -1;
12570 AvFILLp((const AV *)dstr) = -1;
12574 if (HvARRAY((const HV *)sstr)) {
12576 const bool sharekeys = !!HvSHAREKEYS(sstr);
12577 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
12578 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
12580 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
12581 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
12583 HvARRAY(dstr) = (HE**)darray;
12584 while (i <= sxhv->xhv_max) {
12585 const HE * const source = HvARRAY(sstr)[i];
12586 HvARRAY(dstr)[i] = source
12587 ? he_dup(source, sharekeys, param) : 0;
12591 const struct xpvhv_aux * const saux = HvAUX(sstr);
12592 struct xpvhv_aux * const daux = HvAUX(dstr);
12593 /* This flag isn't copied. */
12596 if (saux->xhv_name_count) {
12597 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
12599 = saux->xhv_name_count < 0
12600 ? -saux->xhv_name_count
12601 : saux->xhv_name_count;
12602 HEK **shekp = sname + count;
12604 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
12605 dhekp = daux->xhv_name_u.xhvnameu_names + count;
12606 while (shekp-- > sname) {
12608 *dhekp = hek_dup(*shekp, param);
12612 daux->xhv_name_u.xhvnameu_name
12613 = hek_dup(saux->xhv_name_u.xhvnameu_name,
12616 daux->xhv_name_count = saux->xhv_name_count;
12618 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
12619 daux->xhv_aux_flags = saux->xhv_aux_flags;
12620 #ifdef PERL_HASH_RANDOMIZE_KEYS
12621 daux->xhv_rand = saux->xhv_rand;
12622 daux->xhv_last_rand = saux->xhv_last_rand;
12624 daux->xhv_riter = saux->xhv_riter;
12625 daux->xhv_eiter = saux->xhv_eiter
12626 ? he_dup(saux->xhv_eiter,
12627 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
12628 /* backref array needs refcnt=2; see sv_add_backref */
12629 daux->xhv_backreferences =
12630 (param->flags & CLONEf_JOIN_IN)
12631 /* when joining, we let the individual GVs and
12632 * CVs add themselves to backref as
12633 * needed. This avoids pulling in stuff
12634 * that isn't required, and simplifies the
12635 * case where stashes aren't cloned back
12636 * if they already exist in the parent
12639 : saux->xhv_backreferences
12640 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
12641 ? MUTABLE_AV(SvREFCNT_inc(
12642 sv_dup_inc((const SV *)
12643 saux->xhv_backreferences, param)))
12644 : MUTABLE_AV(sv_dup((const SV *)
12645 saux->xhv_backreferences, param))
12648 daux->xhv_mro_meta = saux->xhv_mro_meta
12649 ? mro_meta_dup(saux->xhv_mro_meta, param)
12652 /* Record stashes for possible cloning in Perl_clone(). */
12654 av_push(param->stashes, dstr);
12658 HvARRAY(MUTABLE_HV(dstr)) = NULL;
12661 if (!(param->flags & CLONEf_COPY_STACKS)) {
12666 /* NOTE: not refcounted */
12667 SvANY(MUTABLE_CV(dstr))->xcv_stash =
12668 hv_dup(CvSTASH(dstr), param);
12669 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
12670 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
12671 if (!CvISXSUB(dstr)) {
12673 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
12675 CvSLABBED_off(dstr);
12676 } else if (CvCONST(dstr)) {
12677 CvXSUBANY(dstr).any_ptr =
12678 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
12680 assert(!CvSLABBED(dstr));
12681 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
12683 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
12684 share_hek_hek(CvNAME_HEK((CV *)sstr));
12685 /* don't dup if copying back - CvGV isn't refcounted, so the
12686 * duped GV may never be freed. A bit of a hack! DAPM */
12688 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
12690 ? gv_dup_inc(CvGV(sstr), param)
12691 : (param->flags & CLONEf_JOIN_IN)
12693 : gv_dup(CvGV(sstr), param);
12695 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
12697 CvWEAKOUTSIDE(sstr)
12698 ? cv_dup( CvOUTSIDE(dstr), param)
12699 : cv_dup_inc(CvOUTSIDE(dstr), param);
12709 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12711 PERL_ARGS_ASSERT_SV_DUP_INC;
12712 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
12716 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12718 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
12719 PERL_ARGS_ASSERT_SV_DUP;
12721 /* Track every SV that (at least initially) had a reference count of 0.
12722 We need to do this by holding an actual reference to it in this array.
12723 If we attempt to cheat, turn AvREAL_off(), and store only pointers
12724 (akin to the stashes hash, and the perl stack), we come unstuck if
12725 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
12726 thread) is manipulated in a CLONE method, because CLONE runs before the
12727 unreferenced array is walked to find SVs still with SvREFCNT() == 0
12728 (and fix things up by giving each a reference via the temps stack).
12729 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
12730 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
12731 before the walk of unreferenced happens and a reference to that is SV
12732 added to the temps stack. At which point we have the same SV considered
12733 to be in use, and free to be re-used. Not good.
12735 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
12736 assert(param->unreferenced);
12737 av_push(param->unreferenced, SvREFCNT_inc(dstr));
12743 /* duplicate a context */
12746 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
12748 PERL_CONTEXT *ncxs;
12750 PERL_ARGS_ASSERT_CX_DUP;
12753 return (PERL_CONTEXT*)NULL;
12755 /* look for it in the table first */
12756 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
12760 /* create anew and remember what it is */
12761 Newx(ncxs, max + 1, PERL_CONTEXT);
12762 ptr_table_store(PL_ptr_table, cxs, ncxs);
12763 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
12766 PERL_CONTEXT * const ncx = &ncxs[ix];
12767 if (CxTYPE(ncx) == CXt_SUBST) {
12768 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
12771 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
12772 switch (CxTYPE(ncx)) {
12774 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
12775 ? cv_dup_inc(ncx->blk_sub.cv, param)
12776 : cv_dup(ncx->blk_sub.cv,param));
12777 ncx->blk_sub.argarray = (CxHASARGS(ncx)
12778 ? av_dup_inc(ncx->blk_sub.argarray,
12781 ncx->blk_sub.savearray = (CxHASARGS(ncx)
12782 ? av_dup_inc(ncx->blk_sub.savearray,
12785 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
12786 ncx->blk_sub.oldcomppad);
12789 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
12791 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
12792 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
12794 case CXt_LOOP_LAZYSV:
12795 ncx->blk_loop.state_u.lazysv.end
12796 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
12797 /* We are taking advantage of av_dup_inc and sv_dup_inc
12798 actually being the same function, and order equivalence of
12800 We can assert the later [but only at run time :-(] */
12801 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
12802 (void *) &ncx->blk_loop.state_u.lazysv.cur);
12804 ncx->blk_loop.state_u.ary.ary
12805 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
12806 case CXt_LOOP_LAZYIV:
12807 case CXt_LOOP_PLAIN:
12808 if (CxPADLOOP(ncx)) {
12809 ncx->blk_loop.itervar_u.oldcomppad
12810 = (PAD*)ptr_table_fetch(PL_ptr_table,
12811 ncx->blk_loop.itervar_u.oldcomppad);
12813 ncx->blk_loop.itervar_u.gv
12814 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
12819 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
12820 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
12821 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
12836 /* duplicate a stack info structure */
12839 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
12843 PERL_ARGS_ASSERT_SI_DUP;
12846 return (PERL_SI*)NULL;
12848 /* look for it in the table first */
12849 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
12853 /* create anew and remember what it is */
12854 Newxz(nsi, 1, PERL_SI);
12855 ptr_table_store(PL_ptr_table, si, nsi);
12857 nsi->si_stack = av_dup_inc(si->si_stack, param);
12858 nsi->si_cxix = si->si_cxix;
12859 nsi->si_cxmax = si->si_cxmax;
12860 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
12861 nsi->si_type = si->si_type;
12862 nsi->si_prev = si_dup(si->si_prev, param);
12863 nsi->si_next = si_dup(si->si_next, param);
12864 nsi->si_markoff = si->si_markoff;
12869 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
12870 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
12871 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
12872 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
12873 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
12874 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
12875 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
12876 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
12877 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
12878 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
12879 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
12880 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
12881 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
12882 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
12883 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
12884 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
12887 #define pv_dup_inc(p) SAVEPV(p)
12888 #define pv_dup(p) SAVEPV(p)
12889 #define svp_dup_inc(p,pp) any_dup(p,pp)
12891 /* map any object to the new equivent - either something in the
12892 * ptr table, or something in the interpreter structure
12896 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
12900 PERL_ARGS_ASSERT_ANY_DUP;
12903 return (void*)NULL;
12905 /* look for it in the table first */
12906 ret = ptr_table_fetch(PL_ptr_table, v);
12910 /* see if it is part of the interpreter structure */
12911 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
12912 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
12920 /* duplicate the save stack */
12923 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
12926 ANY * const ss = proto_perl->Isavestack;
12927 const I32 max = proto_perl->Isavestack_max;
12928 I32 ix = proto_perl->Isavestack_ix;
12941 void (*dptr) (void*);
12942 void (*dxptr) (pTHX_ void*);
12944 PERL_ARGS_ASSERT_SS_DUP;
12946 Newxz(nss, max, ANY);
12949 const UV uv = POPUV(ss,ix);
12950 const U8 type = (U8)uv & SAVE_MASK;
12952 TOPUV(nss,ix) = uv;
12954 case SAVEt_CLEARSV:
12955 case SAVEt_CLEARPADRANGE:
12957 case SAVEt_HELEM: /* hash element */
12958 sv = (const SV *)POPPTR(ss,ix);
12959 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12961 case SAVEt_ITEM: /* normal string */
12962 case SAVEt_GVSV: /* scalar slot in GV */
12963 case SAVEt_SV: /* scalar reference */
12964 sv = (const SV *)POPPTR(ss,ix);
12965 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12968 case SAVEt_MORTALIZESV:
12969 case SAVEt_READONLY_OFF:
12970 sv = (const SV *)POPPTR(ss,ix);
12971 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12973 case SAVEt_SHARED_PVREF: /* char* in shared space */
12974 c = (char*)POPPTR(ss,ix);
12975 TOPPTR(nss,ix) = savesharedpv(c);
12976 ptr = POPPTR(ss,ix);
12977 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
12979 case SAVEt_GENERIC_SVREF: /* generic sv */
12980 case SAVEt_SVREF: /* scalar reference */
12981 sv = (const SV *)POPPTR(ss,ix);
12982 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12983 ptr = POPPTR(ss,ix);
12984 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
12986 case SAVEt_GVSLOT: /* any slot in GV */
12987 sv = (const SV *)POPPTR(ss,ix);
12988 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12989 ptr = POPPTR(ss,ix);
12990 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
12991 sv = (const SV *)POPPTR(ss,ix);
12992 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12994 case SAVEt_HV: /* hash reference */
12995 case SAVEt_AV: /* array reference */
12996 sv = (const SV *) POPPTR(ss,ix);
12997 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
12999 case SAVEt_COMPPAD:
13001 sv = (const SV *) POPPTR(ss,ix);
13002 TOPPTR(nss,ix) = sv_dup(sv, param);
13004 case SAVEt_INT: /* int reference */
13005 ptr = POPPTR(ss,ix);
13006 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13007 intval = (int)POPINT(ss,ix);
13008 TOPINT(nss,ix) = intval;
13010 case SAVEt_LONG: /* long reference */
13011 ptr = POPPTR(ss,ix);
13012 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13013 longval = (long)POPLONG(ss,ix);
13014 TOPLONG(nss,ix) = longval;
13016 case SAVEt_I32: /* I32 reference */
13017 ptr = POPPTR(ss,ix);
13018 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13020 TOPINT(nss,ix) = i;
13022 case SAVEt_IV: /* IV reference */
13023 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13024 ptr = POPPTR(ss,ix);
13025 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13027 TOPIV(nss,ix) = iv;
13029 case SAVEt_HPTR: /* HV* reference */
13030 case SAVEt_APTR: /* AV* reference */
13031 case SAVEt_SPTR: /* SV* reference */
13032 ptr = POPPTR(ss,ix);
13033 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13034 sv = (const SV *)POPPTR(ss,ix);
13035 TOPPTR(nss,ix) = sv_dup(sv, param);
13037 case SAVEt_VPTR: /* random* reference */
13038 ptr = POPPTR(ss,ix);
13039 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13041 case SAVEt_INT_SMALL:
13042 case SAVEt_I32_SMALL:
13043 case SAVEt_I16: /* I16 reference */
13044 case SAVEt_I8: /* I8 reference */
13046 ptr = POPPTR(ss,ix);
13047 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13049 case SAVEt_GENERIC_PVREF: /* generic char* */
13050 case SAVEt_PPTR: /* char* reference */
13051 ptr = POPPTR(ss,ix);
13052 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13053 c = (char*)POPPTR(ss,ix);
13054 TOPPTR(nss,ix) = pv_dup(c);
13056 case SAVEt_GP: /* scalar reference */
13057 gp = (GP*)POPPTR(ss,ix);
13058 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13059 (void)GpREFCNT_inc(gp);
13060 gv = (const GV *)POPPTR(ss,ix);
13061 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13064 ptr = POPPTR(ss,ix);
13065 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13066 /* these are assumed to be refcounted properly */
13068 switch (((OP*)ptr)->op_type) {
13070 case OP_LEAVESUBLV:
13074 case OP_LEAVEWRITE:
13075 TOPPTR(nss,ix) = ptr;
13078 (void) OpREFCNT_inc(o);
13082 TOPPTR(nss,ix) = NULL;
13087 TOPPTR(nss,ix) = NULL;
13089 case SAVEt_FREECOPHH:
13090 ptr = POPPTR(ss,ix);
13091 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13093 case SAVEt_ADELETE:
13094 av = (const AV *)POPPTR(ss,ix);
13095 TOPPTR(nss,ix) = av_dup_inc(av, param);
13097 TOPINT(nss,ix) = i;
13100 hv = (const HV *)POPPTR(ss,ix);
13101 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13103 TOPINT(nss,ix) = i;
13106 c = (char*)POPPTR(ss,ix);
13107 TOPPTR(nss,ix) = pv_dup_inc(c);
13109 case SAVEt_STACK_POS: /* Position on Perl stack */
13111 TOPINT(nss,ix) = i;
13113 case SAVEt_DESTRUCTOR:
13114 ptr = POPPTR(ss,ix);
13115 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13116 dptr = POPDPTR(ss,ix);
13117 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13118 any_dup(FPTR2DPTR(void *, dptr),
13121 case SAVEt_DESTRUCTOR_X:
13122 ptr = POPPTR(ss,ix);
13123 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13124 dxptr = POPDXPTR(ss,ix);
13125 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13126 any_dup(FPTR2DPTR(void *, dxptr),
13129 case SAVEt_REGCONTEXT:
13131 ix -= uv >> SAVE_TIGHT_SHIFT;
13133 case SAVEt_AELEM: /* array element */
13134 sv = (const SV *)POPPTR(ss,ix);
13135 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13137 TOPINT(nss,ix) = i;
13138 av = (const AV *)POPPTR(ss,ix);
13139 TOPPTR(nss,ix) = av_dup_inc(av, param);
13142 ptr = POPPTR(ss,ix);
13143 TOPPTR(nss,ix) = ptr;
13146 ptr = POPPTR(ss,ix);
13147 ptr = cophh_copy((COPHH*)ptr);
13148 TOPPTR(nss,ix) = ptr;
13150 TOPINT(nss,ix) = i;
13151 if (i & HINT_LOCALIZE_HH) {
13152 hv = (const HV *)POPPTR(ss,ix);
13153 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13156 case SAVEt_PADSV_AND_MORTALIZE:
13157 longval = (long)POPLONG(ss,ix);
13158 TOPLONG(nss,ix) = longval;
13159 ptr = POPPTR(ss,ix);
13160 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13161 sv = (const SV *)POPPTR(ss,ix);
13162 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13164 case SAVEt_SET_SVFLAGS:
13166 TOPINT(nss,ix) = i;
13168 TOPINT(nss,ix) = i;
13169 sv = (const SV *)POPPTR(ss,ix);
13170 TOPPTR(nss,ix) = sv_dup(sv, param);
13172 case SAVEt_COMPILE_WARNINGS:
13173 ptr = POPPTR(ss,ix);
13174 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
13177 ptr = POPPTR(ss,ix);
13178 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
13182 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
13190 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
13191 * flag to the result. This is done for each stash before cloning starts,
13192 * so we know which stashes want their objects cloned */
13195 do_mark_cloneable_stash(pTHX_ SV *const sv)
13197 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
13199 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
13200 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
13201 if (cloner && GvCV(cloner)) {
13208 mXPUSHs(newSVhek(hvname));
13210 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
13217 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
13225 =for apidoc perl_clone
13227 Create and return a new interpreter by cloning the current one.
13229 perl_clone takes these flags as parameters:
13231 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
13232 without it we only clone the data and zero the stacks,
13233 with it we copy the stacks and the new perl interpreter is
13234 ready to run at the exact same point as the previous one.
13235 The pseudo-fork code uses COPY_STACKS while the
13236 threads->create doesn't.
13238 CLONEf_KEEP_PTR_TABLE -
13239 perl_clone keeps a ptr_table with the pointer of the old
13240 variable as a key and the new variable as a value,
13241 this allows it to check if something has been cloned and not
13242 clone it again but rather just use the value and increase the
13243 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
13244 the ptr_table using the function
13245 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
13246 reason to keep it around is if you want to dup some of your own
13247 variable who are outside the graph perl scans, example of this
13248 code is in threads.xs create.
13250 CLONEf_CLONE_HOST -
13251 This is a win32 thing, it is ignored on unix, it tells perls
13252 win32host code (which is c++) to clone itself, this is needed on
13253 win32 if you want to run two threads at the same time,
13254 if you just want to do some stuff in a separate perl interpreter
13255 and then throw it away and return to the original one,
13256 you don't need to do anything.
13261 /* XXX the above needs expanding by someone who actually understands it ! */
13262 EXTERN_C PerlInterpreter *
13263 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
13266 perl_clone(PerlInterpreter *proto_perl, UV flags)
13269 #ifdef PERL_IMPLICIT_SYS
13271 PERL_ARGS_ASSERT_PERL_CLONE;
13273 /* perlhost.h so we need to call into it
13274 to clone the host, CPerlHost should have a c interface, sky */
13276 if (flags & CLONEf_CLONE_HOST) {
13277 return perl_clone_host(proto_perl,flags);
13279 return perl_clone_using(proto_perl, flags,
13281 proto_perl->IMemShared,
13282 proto_perl->IMemParse,
13284 proto_perl->IStdIO,
13288 proto_perl->IProc);
13292 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
13293 struct IPerlMem* ipM, struct IPerlMem* ipMS,
13294 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
13295 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
13296 struct IPerlDir* ipD, struct IPerlSock* ipS,
13297 struct IPerlProc* ipP)
13299 /* XXX many of the string copies here can be optimized if they're
13300 * constants; they need to be allocated as common memory and just
13301 * their pointers copied. */
13304 CLONE_PARAMS clone_params;
13305 CLONE_PARAMS* const param = &clone_params;
13307 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
13309 PERL_ARGS_ASSERT_PERL_CLONE_USING;
13310 #else /* !PERL_IMPLICIT_SYS */
13312 CLONE_PARAMS clone_params;
13313 CLONE_PARAMS* param = &clone_params;
13314 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
13316 PERL_ARGS_ASSERT_PERL_CLONE;
13317 #endif /* PERL_IMPLICIT_SYS */
13319 /* for each stash, determine whether its objects should be cloned */
13320 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
13321 PERL_SET_THX(my_perl);
13324 PoisonNew(my_perl, 1, PerlInterpreter);
13327 PL_defstash = NULL; /* may be used by perl malloc() */
13330 PL_scopestack_name = 0;
13332 PL_savestack_ix = 0;
13333 PL_savestack_max = -1;
13334 PL_sig_pending = 0;
13336 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
13337 # ifdef DEBUG_LEAKING_SCALARS
13338 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
13340 #else /* !DEBUGGING */
13341 Zero(my_perl, 1, PerlInterpreter);
13342 #endif /* DEBUGGING */
13344 #ifdef PERL_IMPLICIT_SYS
13345 /* host pointers */
13347 PL_MemShared = ipMS;
13348 PL_MemParse = ipMP;
13355 #endif /* PERL_IMPLICIT_SYS */
13358 param->flags = flags;
13359 /* Nothing in the core code uses this, but we make it available to
13360 extensions (using mg_dup). */
13361 param->proto_perl = proto_perl;
13362 /* Likely nothing will use this, but it is initialised to be consistent
13363 with Perl_clone_params_new(). */
13364 param->new_perl = my_perl;
13365 param->unreferenced = NULL;
13368 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
13370 PL_body_arenas = NULL;
13371 Zero(&PL_body_roots, 1, PL_body_roots);
13375 PL_sv_arenaroot = NULL;
13377 PL_debug = proto_perl->Idebug;
13379 /* dbargs array probably holds garbage */
13382 PL_compiling = proto_perl->Icompiling;
13384 /* pseudo environmental stuff */
13385 PL_origargc = proto_perl->Iorigargc;
13386 PL_origargv = proto_perl->Iorigargv;
13388 #ifndef NO_TAINT_SUPPORT
13389 /* Set tainting stuff before PerlIO_debug can possibly get called */
13390 PL_tainting = proto_perl->Itainting;
13391 PL_taint_warn = proto_perl->Itaint_warn;
13393 PL_tainting = FALSE;
13394 PL_taint_warn = FALSE;
13397 PL_minus_c = proto_perl->Iminus_c;
13399 PL_localpatches = proto_perl->Ilocalpatches;
13400 PL_splitstr = proto_perl->Isplitstr;
13401 PL_minus_n = proto_perl->Iminus_n;
13402 PL_minus_p = proto_perl->Iminus_p;
13403 PL_minus_l = proto_perl->Iminus_l;
13404 PL_minus_a = proto_perl->Iminus_a;
13405 PL_minus_E = proto_perl->Iminus_E;
13406 PL_minus_F = proto_perl->Iminus_F;
13407 PL_doswitches = proto_perl->Idoswitches;
13408 PL_dowarn = proto_perl->Idowarn;
13409 #ifdef PERL_SAWAMPERSAND
13410 PL_sawampersand = proto_perl->Isawampersand;
13412 PL_unsafe = proto_perl->Iunsafe;
13413 PL_perldb = proto_perl->Iperldb;
13414 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
13415 PL_exit_flags = proto_perl->Iexit_flags;
13417 /* XXX time(&PL_basetime) when asked for? */
13418 PL_basetime = proto_perl->Ibasetime;
13420 PL_maxsysfd = proto_perl->Imaxsysfd;
13421 PL_statusvalue = proto_perl->Istatusvalue;
13423 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
13425 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
13428 /* RE engine related */
13429 PL_regmatch_slab = NULL;
13430 PL_reg_curpm = NULL;
13432 PL_sub_generation = proto_perl->Isub_generation;
13434 /* funky return mechanisms */
13435 PL_forkprocess = proto_perl->Iforkprocess;
13437 /* internal state */
13438 PL_maxo = proto_perl->Imaxo;
13440 PL_main_start = proto_perl->Imain_start;
13441 PL_eval_root = proto_perl->Ieval_root;
13442 PL_eval_start = proto_perl->Ieval_start;
13444 PL_filemode = proto_perl->Ifilemode;
13445 PL_lastfd = proto_perl->Ilastfd;
13446 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
13449 PL_gensym = proto_perl->Igensym;
13451 PL_laststatval = proto_perl->Ilaststatval;
13452 PL_laststype = proto_perl->Ilaststype;
13455 PL_profiledata = NULL;
13457 PL_generation = proto_perl->Igeneration;
13459 PL_in_clean_objs = proto_perl->Iin_clean_objs;
13460 PL_in_clean_all = proto_perl->Iin_clean_all;
13462 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
13463 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
13464 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
13465 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
13466 PL_nomemok = proto_perl->Inomemok;
13467 PL_an = proto_perl->Ian;
13468 PL_evalseq = proto_perl->Ievalseq;
13469 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
13470 PL_origalen = proto_perl->Iorigalen;
13472 PL_sighandlerp = proto_perl->Isighandlerp;
13474 PL_runops = proto_perl->Irunops;
13476 PL_subline = proto_perl->Isubline;
13479 PL_cryptseen = proto_perl->Icryptseen;
13482 #ifdef USE_LOCALE_COLLATE
13483 PL_collation_ix = proto_perl->Icollation_ix;
13484 PL_collation_standard = proto_perl->Icollation_standard;
13485 PL_collxfrm_base = proto_perl->Icollxfrm_base;
13486 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
13487 #endif /* USE_LOCALE_COLLATE */
13489 #ifdef USE_LOCALE_NUMERIC
13490 PL_numeric_standard = proto_perl->Inumeric_standard;
13491 PL_numeric_local = proto_perl->Inumeric_local;
13492 #endif /* !USE_LOCALE_NUMERIC */
13494 /* Did the locale setup indicate UTF-8? */
13495 PL_utf8locale = proto_perl->Iutf8locale;
13496 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
13497 /* Unicode features (see perlrun/-C) */
13498 PL_unicode = proto_perl->Iunicode;
13500 /* Pre-5.8 signals control */
13501 PL_signals = proto_perl->Isignals;
13503 /* times() ticks per second */
13504 PL_clocktick = proto_perl->Iclocktick;
13506 /* Recursion stopper for PerlIO_find_layer */
13507 PL_in_load_module = proto_perl->Iin_load_module;
13509 /* sort() routine */
13510 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
13512 /* Not really needed/useful since the reenrant_retint is "volatile",
13513 * but do it for consistency's sake. */
13514 PL_reentrant_retint = proto_perl->Ireentrant_retint;
13516 /* Hooks to shared SVs and locks. */
13517 PL_sharehook = proto_perl->Isharehook;
13518 PL_lockhook = proto_perl->Ilockhook;
13519 PL_unlockhook = proto_perl->Iunlockhook;
13520 PL_threadhook = proto_perl->Ithreadhook;
13521 PL_destroyhook = proto_perl->Idestroyhook;
13522 PL_signalhook = proto_perl->Isignalhook;
13524 PL_globhook = proto_perl->Iglobhook;
13527 PL_last_swash_hv = NULL; /* reinits on demand */
13528 PL_last_swash_klen = 0;
13529 PL_last_swash_key[0]= '\0';
13530 PL_last_swash_tmps = (U8*)NULL;
13531 PL_last_swash_slen = 0;
13533 PL_srand_called = proto_perl->Isrand_called;
13534 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
13536 if (flags & CLONEf_COPY_STACKS) {
13537 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
13538 PL_tmps_ix = proto_perl->Itmps_ix;
13539 PL_tmps_max = proto_perl->Itmps_max;
13540 PL_tmps_floor = proto_perl->Itmps_floor;
13542 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
13543 * NOTE: unlike the others! */
13544 PL_scopestack_ix = proto_perl->Iscopestack_ix;
13545 PL_scopestack_max = proto_perl->Iscopestack_max;
13547 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
13548 * NOTE: unlike the others! */
13549 PL_savestack_ix = proto_perl->Isavestack_ix;
13550 PL_savestack_max = proto_perl->Isavestack_max;
13553 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
13554 PL_top_env = &PL_start_env;
13556 PL_op = proto_perl->Iop;
13559 PL_Xpv = (XPV*)NULL;
13560 my_perl->Ina = proto_perl->Ina;
13562 PL_statbuf = proto_perl->Istatbuf;
13563 PL_statcache = proto_perl->Istatcache;
13565 #ifndef NO_TAINT_SUPPORT
13566 PL_tainted = proto_perl->Itainted;
13568 PL_tainted = FALSE;
13570 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
13572 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
13574 PL_restartjmpenv = proto_perl->Irestartjmpenv;
13575 PL_restartop = proto_perl->Irestartop;
13576 PL_in_eval = proto_perl->Iin_eval;
13577 PL_delaymagic = proto_perl->Idelaymagic;
13578 PL_phase = proto_perl->Iphase;
13579 PL_localizing = proto_perl->Ilocalizing;
13581 PL_hv_fetch_ent_mh = NULL;
13582 PL_modcount = proto_perl->Imodcount;
13583 PL_lastgotoprobe = NULL;
13584 PL_dumpindent = proto_perl->Idumpindent;
13586 PL_efloatbuf = NULL; /* reinits on demand */
13587 PL_efloatsize = 0; /* reinits on demand */
13591 PL_colorset = 0; /* reinits PL_colors[] */
13592 /*PL_colors[6] = {0,0,0,0,0,0};*/
13594 /* Pluggable optimizer */
13595 PL_peepp = proto_perl->Ipeepp;
13596 PL_rpeepp = proto_perl->Irpeepp;
13597 /* op_free() hook */
13598 PL_opfreehook = proto_perl->Iopfreehook;
13600 #ifdef USE_REENTRANT_API
13601 /* XXX: things like -Dm will segfault here in perlio, but doing
13602 * PERL_SET_CONTEXT(proto_perl);
13603 * breaks too many other things
13605 Perl_reentrant_init(aTHX);
13608 /* create SV map for pointer relocation */
13609 PL_ptr_table = ptr_table_new();
13611 /* initialize these special pointers as early as possible */
13613 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
13614 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
13615 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
13617 /* create (a non-shared!) shared string table */
13618 PL_strtab = newHV();
13619 HvSHAREKEYS_off(PL_strtab);
13620 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
13621 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
13623 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
13625 /* This PV will be free'd special way so must set it same way op.c does */
13626 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
13627 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
13629 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
13630 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
13631 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
13632 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
13634 param->stashes = newAV(); /* Setup array of objects to call clone on */
13635 /* This makes no difference to the implementation, as it always pushes
13636 and shifts pointers to other SVs without changing their reference
13637 count, with the array becoming empty before it is freed. However, it
13638 makes it conceptually clear what is going on, and will avoid some
13639 work inside av.c, filling slots between AvFILL() and AvMAX() with
13640 &PL_sv_undef, and SvREFCNT_dec()ing those. */
13641 AvREAL_off(param->stashes);
13643 if (!(flags & CLONEf_COPY_STACKS)) {
13644 param->unreferenced = newAV();
13647 #ifdef PERLIO_LAYERS
13648 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
13649 PerlIO_clone(aTHX_ proto_perl, param);
13652 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
13653 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
13654 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
13655 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
13656 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
13657 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
13660 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
13661 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
13662 PL_inplace = SAVEPV(proto_perl->Iinplace);
13663 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
13665 /* magical thingies */
13667 PL_encoding = sv_dup(proto_perl->Iencoding, param);
13669 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
13670 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
13671 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
13674 /* Clone the regex array */
13675 /* ORANGE FIXME for plugins, probably in the SV dup code.
13676 newSViv(PTR2IV(CALLREGDUPE(
13677 INT2PTR(REGEXP *, SvIVX(regex)), param))))
13679 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
13680 PL_regex_pad = AvARRAY(PL_regex_padav);
13682 PL_stashpadmax = proto_perl->Istashpadmax;
13683 PL_stashpadix = proto_perl->Istashpadix ;
13684 Newx(PL_stashpad, PL_stashpadmax, HV *);
13687 for (; o < PL_stashpadmax; ++o)
13688 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
13691 /* shortcuts to various I/O objects */
13692 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
13693 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
13694 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
13695 PL_defgv = gv_dup(proto_perl->Idefgv, param);
13696 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
13697 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
13698 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
13700 /* shortcuts to regexp stuff */
13701 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
13703 /* shortcuts to misc objects */
13704 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
13706 /* shortcuts to debugging objects */
13707 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
13708 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
13709 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
13710 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
13711 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
13712 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
13714 /* symbol tables */
13715 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
13716 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
13717 PL_debstash = hv_dup(proto_perl->Idebstash, param);
13718 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
13719 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
13721 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
13722 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
13723 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
13724 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
13725 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
13726 PL_endav = av_dup_inc(proto_perl->Iendav, param);
13727 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
13728 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
13730 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
13732 /* subprocess state */
13733 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
13735 if (proto_perl->Iop_mask)
13736 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
13739 /* PL_asserting = proto_perl->Iasserting; */
13741 /* current interpreter roots */
13742 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
13744 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
13747 /* runtime control stuff */
13748 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
13750 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
13752 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
13754 /* interpreter atexit processing */
13755 PL_exitlistlen = proto_perl->Iexitlistlen;
13756 if (PL_exitlistlen) {
13757 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13758 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13761 PL_exitlist = (PerlExitListEntry*)NULL;
13763 PL_my_cxt_size = proto_perl->Imy_cxt_size;
13764 if (PL_my_cxt_size) {
13765 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
13766 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
13767 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13768 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
13769 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
13773 PL_my_cxt_list = (void**)NULL;
13774 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13775 PL_my_cxt_keys = (const char**)NULL;
13778 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
13779 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
13780 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
13781 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
13783 PL_compcv = cv_dup(proto_perl->Icompcv, param);
13785 PAD_CLONE_VARS(proto_perl, param);
13787 #ifdef HAVE_INTERP_INTERN
13788 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
13791 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
13793 #ifdef PERL_USES_PL_PIDSTATUS
13794 PL_pidstatus = newHV(); /* XXX flag for cloning? */
13796 PL_osname = SAVEPV(proto_perl->Iosname);
13797 PL_parser = parser_dup(proto_perl->Iparser, param);
13799 /* XXX this only works if the saved cop has already been cloned */
13800 if (proto_perl->Iparser) {
13801 PL_parser->saved_curcop = (COP*)any_dup(
13802 proto_perl->Iparser->saved_curcop,
13806 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
13808 #ifdef USE_LOCALE_COLLATE
13809 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
13810 #endif /* USE_LOCALE_COLLATE */
13812 #ifdef USE_LOCALE_NUMERIC
13813 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
13814 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
13815 #endif /* !USE_LOCALE_NUMERIC */
13817 /* Unicode inversion lists */
13818 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
13819 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
13820 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
13822 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
13823 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
13825 /* utf8 character class swashes */
13826 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
13827 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
13829 for (i = 0; i < POSIX_CC_COUNT; i++) {
13830 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
13832 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
13833 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
13834 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
13835 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
13836 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
13837 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
13838 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
13839 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
13840 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
13841 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
13842 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
13843 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
13844 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
13845 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
13846 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
13847 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
13849 if (proto_perl->Ipsig_pend) {
13850 Newxz(PL_psig_pend, SIG_SIZE, int);
13853 PL_psig_pend = (int*)NULL;
13856 if (proto_perl->Ipsig_name) {
13857 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
13858 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
13860 PL_psig_ptr = PL_psig_name + SIG_SIZE;
13863 PL_psig_ptr = (SV**)NULL;
13864 PL_psig_name = (SV**)NULL;
13867 if (flags & CLONEf_COPY_STACKS) {
13868 Newx(PL_tmps_stack, PL_tmps_max, SV*);
13869 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
13870 PL_tmps_ix+1, param);
13872 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
13873 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
13874 Newxz(PL_markstack, i, I32);
13875 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
13876 - proto_perl->Imarkstack);
13877 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
13878 - proto_perl->Imarkstack);
13879 Copy(proto_perl->Imarkstack, PL_markstack,
13880 PL_markstack_ptr - PL_markstack + 1, I32);
13882 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
13883 * NOTE: unlike the others! */
13884 Newxz(PL_scopestack, PL_scopestack_max, I32);
13885 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
13888 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
13889 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
13891 /* reset stack AV to correct length before its duped via
13892 * PL_curstackinfo */
13893 AvFILLp(proto_perl->Icurstack) =
13894 proto_perl->Istack_sp - proto_perl->Istack_base;
13896 /* NOTE: si_dup() looks at PL_markstack */
13897 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
13899 /* PL_curstack = PL_curstackinfo->si_stack; */
13900 PL_curstack = av_dup(proto_perl->Icurstack, param);
13901 PL_mainstack = av_dup(proto_perl->Imainstack, param);
13903 /* next PUSHs() etc. set *(PL_stack_sp+1) */
13904 PL_stack_base = AvARRAY(PL_curstack);
13905 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
13906 - proto_perl->Istack_base);
13907 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
13909 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
13910 PL_savestack = ss_dup(proto_perl, param);
13914 ENTER; /* perl_destruct() wants to LEAVE; */
13917 PL_statgv = gv_dup(proto_perl->Istatgv, param);
13918 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
13920 PL_rs = sv_dup_inc(proto_perl->Irs, param);
13921 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
13922 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
13923 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
13924 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
13925 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
13927 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
13929 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
13930 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
13931 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
13933 PL_stashcache = newHV();
13935 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
13936 proto_perl->Iwatchaddr);
13937 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
13938 if (PL_debug && PL_watchaddr) {
13939 PerlIO_printf(Perl_debug_log,
13940 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
13941 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
13942 PTR2UV(PL_watchok));
13945 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
13946 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
13947 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
13949 /* Call the ->CLONE method, if it exists, for each of the stashes
13950 identified by sv_dup() above.
13952 while(av_tindex(param->stashes) != -1) {
13953 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
13954 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
13955 if (cloner && GvCV(cloner)) {
13960 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
13962 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
13968 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
13969 ptr_table_free(PL_ptr_table);
13970 PL_ptr_table = NULL;
13973 if (!(flags & CLONEf_COPY_STACKS)) {
13974 unreferenced_to_tmp_stack(param->unreferenced);
13977 SvREFCNT_dec(param->stashes);
13979 /* orphaned? eg threads->new inside BEGIN or use */
13980 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
13981 SvREFCNT_inc_simple_void(PL_compcv);
13982 SAVEFREESV(PL_compcv);
13989 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
13991 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
13993 if (AvFILLp(unreferenced) > -1) {
13994 SV **svp = AvARRAY(unreferenced);
13995 SV **const last = svp + AvFILLp(unreferenced);
13999 if (SvREFCNT(*svp) == 1)
14001 } while (++svp <= last);
14003 EXTEND_MORTAL(count);
14004 svp = AvARRAY(unreferenced);
14007 if (SvREFCNT(*svp) == 1) {
14008 /* Our reference is the only one to this SV. This means that
14009 in this thread, the scalar effectively has a 0 reference.
14010 That doesn't work (cleanup never happens), so donate our
14011 reference to it onto the save stack. */
14012 PL_tmps_stack[++PL_tmps_ix] = *svp;
14014 /* As an optimisation, because we are already walking the
14015 entire array, instead of above doing either
14016 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14017 release our reference to the scalar, so that at the end of
14018 the array owns zero references to the scalars it happens to
14019 point to. We are effectively converting the array from
14020 AvREAL() on to AvREAL() off. This saves the av_clear()
14021 (triggered by the SvREFCNT_dec(unreferenced) below) from
14022 walking the array a second time. */
14023 SvREFCNT_dec(*svp);
14026 } while (++svp <= last);
14027 AvREAL_off(unreferenced);
14029 SvREFCNT_dec_NN(unreferenced);
14033 Perl_clone_params_del(CLONE_PARAMS *param)
14035 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14037 PerlInterpreter *const to = param->new_perl;
14039 PerlInterpreter *const was = PERL_GET_THX;
14041 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14047 SvREFCNT_dec(param->stashes);
14048 if (param->unreferenced)
14049 unreferenced_to_tmp_stack(param->unreferenced);
14059 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14062 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14063 does a dTHX; to get the context from thread local storage.
14064 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14065 a version that passes in my_perl. */
14066 PerlInterpreter *const was = PERL_GET_THX;
14067 CLONE_PARAMS *param;
14069 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14075 /* Given that we've set the context, we can do this unshared. */
14076 Newx(param, 1, CLONE_PARAMS);
14079 param->proto_perl = from;
14080 param->new_perl = to;
14081 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14082 AvREAL_off(param->stashes);
14083 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14091 #endif /* USE_ITHREADS */
14094 Perl_init_constants(pTHX)
14096 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14097 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVt_NULL;
14098 SvANY(&PL_sv_undef) = NULL;
14100 SvANY(&PL_sv_no) = new_XPVNV();
14101 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14102 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY
14103 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14106 SvANY(&PL_sv_yes) = new_XPVNV();
14107 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14108 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY
14109 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14112 SvPV_set(&PL_sv_no, (char*)PL_No);
14113 SvCUR_set(&PL_sv_no, 0);
14114 SvLEN_set(&PL_sv_no, 0);
14115 SvIV_set(&PL_sv_no, 0);
14116 SvNV_set(&PL_sv_no, 0);
14118 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14119 SvCUR_set(&PL_sv_yes, 1);
14120 SvLEN_set(&PL_sv_yes, 0);
14121 SvIV_set(&PL_sv_yes, 1);
14122 SvNV_set(&PL_sv_yes, 1);
14126 =head1 Unicode Support
14128 =for apidoc sv_recode_to_utf8
14130 The encoding is assumed to be an Encode object, on entry the PV
14131 of the sv is assumed to be octets in that encoding, and the sv
14132 will be converted into Unicode (and UTF-8).
14134 If the sv already is UTF-8 (or if it is not POK), or if the encoding
14135 is not a reference, nothing is done to the sv. If the encoding is not
14136 an C<Encode::XS> Encoding object, bad things will happen.
14137 (See F<lib/encoding.pm> and L<Encode>.)
14139 The PV of the sv is returned.
14144 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
14148 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
14150 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
14159 if (SvPADTMP(nsv)) {
14160 nsv = sv_newmortal();
14161 SvSetSV_nosteal(nsv, sv);
14170 Passing sv_yes is wrong - it needs to be or'ed set of constants
14171 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
14172 remove converted chars from source.
14174 Both will default the value - let them.
14176 XPUSHs(&PL_sv_yes);
14179 call_method("decode", G_SCALAR);
14183 s = SvPV_const(uni, len);
14184 if (s != SvPVX_const(sv)) {
14185 SvGROW(sv, len + 1);
14186 Move(s, SvPVX(sv), len + 1, char);
14187 SvCUR_set(sv, len);
14192 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
14193 /* clear pos and any utf8 cache */
14194 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
14197 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
14198 magic_setutf8(sv,mg); /* clear UTF8 cache */
14203 return SvPOKp(sv) ? SvPVX(sv) : NULL;
14207 =for apidoc sv_cat_decode
14209 The encoding is assumed to be an Encode object, the PV of the ssv is
14210 assumed to be octets in that encoding and decoding the input starts
14211 from the position which (PV + *offset) pointed to. The dsv will be
14212 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
14213 when the string tstr appears in decoding output or the input ends on
14214 the PV of the ssv. The value which the offset points will be modified
14215 to the last input position on the ssv.
14217 Returns TRUE if the terminator was found, else returns FALSE.
14222 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
14223 SV *ssv, int *offset, char *tstr, int tlen)
14228 PERL_ARGS_ASSERT_SV_CAT_DECODE;
14230 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
14241 offsv = newSViv(*offset);
14243 mPUSHp(tstr, tlen);
14245 call_method("cat_decode", G_SCALAR);
14247 ret = SvTRUE(TOPs);
14248 *offset = SvIV(offsv);
14254 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
14259 /* ---------------------------------------------------------------------
14261 * support functions for report_uninit()
14264 /* the maxiumum size of array or hash where we will scan looking
14265 * for the undefined element that triggered the warning */
14267 #define FUV_MAX_SEARCH_SIZE 1000
14269 /* Look for an entry in the hash whose value has the same SV as val;
14270 * If so, return a mortal copy of the key. */
14273 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
14279 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
14281 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
14282 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
14285 array = HvARRAY(hv);
14287 for (i=HvMAX(hv); i>=0; i--) {
14289 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
14290 if (HeVAL(entry) != val)
14292 if ( HeVAL(entry) == &PL_sv_undef ||
14293 HeVAL(entry) == &PL_sv_placeholder)
14297 if (HeKLEN(entry) == HEf_SVKEY)
14298 return sv_mortalcopy(HeKEY_sv(entry));
14299 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
14305 /* Look for an entry in the array whose value has the same SV as val;
14306 * If so, return the index, otherwise return -1. */
14309 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
14313 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
14315 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
14316 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
14319 if (val != &PL_sv_undef) {
14320 SV ** const svp = AvARRAY(av);
14323 for (i=AvFILLp(av); i>=0; i--)
14330 /* varname(): return the name of a variable, optionally with a subscript.
14331 * If gv is non-zero, use the name of that global, along with gvtype (one
14332 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
14333 * targ. Depending on the value of the subscript_type flag, return:
14336 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
14337 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
14338 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
14339 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
14342 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
14343 const SV *const keyname, I32 aindex, int subscript_type)
14346 SV * const name = sv_newmortal();
14347 if (gv && isGV(gv)) {
14349 buffer[0] = gvtype;
14352 /* as gv_fullname4(), but add literal '^' for $^FOO names */
14354 gv_fullname4(name, gv, buffer, 0);
14356 if ((unsigned int)SvPVX(name)[1] <= 26) {
14358 buffer[1] = SvPVX(name)[1] + 'A' - 1;
14360 /* Swap the 1 unprintable control character for the 2 byte pretty
14361 version - ie substr($name, 1, 1) = $buffer; */
14362 sv_insert(name, 1, 1, buffer, 2);
14366 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
14370 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
14372 if (!cv || !CvPADLIST(cv))
14374 av = *PadlistARRAY(CvPADLIST(cv));
14375 sv = *av_fetch(av, targ, FALSE);
14376 sv_setsv_flags(name, sv, 0);
14379 if (subscript_type == FUV_SUBSCRIPT_HASH) {
14380 SV * const sv = newSV(0);
14381 *SvPVX(name) = '$';
14382 Perl_sv_catpvf(aTHX_ name, "{%s}",
14383 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
14384 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
14385 SvREFCNT_dec_NN(sv);
14387 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
14388 *SvPVX(name) = '$';
14389 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
14391 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
14392 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
14393 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
14401 =for apidoc find_uninit_var
14403 Find the name of the undefined variable (if any) that caused the operator
14404 to issue a "Use of uninitialized value" warning.
14405 If match is true, only return a name if its value matches uninit_sv.
14406 So roughly speaking, if a unary operator (such as OP_COS) generates a
14407 warning, then following the direct child of the op may yield an
14408 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
14409 other hand, with OP_ADD there are two branches to follow, so we only print
14410 the variable name if we get an exact match.
14412 The name is returned as a mortal SV.
14414 Assumes that PL_op is the op that originally triggered the error, and that
14415 PL_comppad/PL_curpad points to the currently executing pad.
14421 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
14427 const OP *o, *o2, *kid;
14429 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
14430 uninit_sv == &PL_sv_placeholder)))
14433 switch (obase->op_type) {
14440 const bool pad = ( obase->op_type == OP_PADAV
14441 || obase->op_type == OP_PADHV
14442 || obase->op_type == OP_PADRANGE
14445 const bool hash = ( obase->op_type == OP_PADHV
14446 || obase->op_type == OP_RV2HV
14447 || (obase->op_type == OP_PADRANGE
14448 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
14452 int subscript_type = FUV_SUBSCRIPT_WITHIN;
14454 if (pad) { /* @lex, %lex */
14455 sv = PAD_SVl(obase->op_targ);
14459 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14460 /* @global, %global */
14461 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14464 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
14466 else if (obase == PL_op) /* @{expr}, %{expr} */
14467 return find_uninit_var(cUNOPx(obase)->op_first,
14469 else /* @{expr}, %{expr} as a sub-expression */
14473 /* attempt to find a match within the aggregate */
14475 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14477 subscript_type = FUV_SUBSCRIPT_HASH;
14480 index = find_array_subscript((const AV *)sv, uninit_sv);
14482 subscript_type = FUV_SUBSCRIPT_ARRAY;
14485 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
14488 return varname(gv, hash ? '%' : '@', obase->op_targ,
14489 keysv, index, subscript_type);
14493 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14495 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14496 if (!gv || !GvSTASH(gv))
14498 if (match && (GvSV(gv) != uninit_sv))
14500 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14503 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
14506 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
14508 return varname(NULL, '$', obase->op_targ,
14509 NULL, 0, FUV_SUBSCRIPT_NONE);
14512 gv = cGVOPx_gv(obase);
14513 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
14515 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14517 case OP_AELEMFAST_LEX:
14520 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
14521 if (!av || SvRMAGICAL(av))
14523 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14524 if (!svp || *svp != uninit_sv)
14527 return varname(NULL, '$', obase->op_targ,
14528 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14531 gv = cGVOPx_gv(obase);
14536 AV *const av = GvAV(gv);
14537 if (!av || SvRMAGICAL(av))
14539 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14540 if (!svp || *svp != uninit_sv)
14543 return varname(gv, '$', 0,
14544 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14549 o = cUNOPx(obase)->op_first;
14550 if (!o || o->op_type != OP_NULL ||
14551 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
14553 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
14558 bool negate = FALSE;
14560 if (PL_op == obase)
14561 /* $a[uninit_expr] or $h{uninit_expr} */
14562 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
14565 o = cBINOPx(obase)->op_first;
14566 kid = cBINOPx(obase)->op_last;
14568 /* get the av or hv, and optionally the gv */
14570 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
14571 sv = PAD_SV(o->op_targ);
14573 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
14574 && cUNOPo->op_first->op_type == OP_GV)
14576 gv = cGVOPx_gv(cUNOPo->op_first);
14580 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
14585 if (kid && kid->op_type == OP_NEGATE) {
14587 kid = cUNOPx(kid)->op_first;
14590 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
14591 /* index is constant */
14594 kidsv = sv_2mortal(newSVpvs("-"));
14595 sv_catsv(kidsv, cSVOPx_sv(kid));
14598 kidsv = cSVOPx_sv(kid);
14602 if (obase->op_type == OP_HELEM) {
14603 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
14604 if (!he || HeVAL(he) != uninit_sv)
14608 SV * const opsv = cSVOPx_sv(kid);
14609 const IV opsviv = SvIV(opsv);
14610 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
14611 negate ? - opsviv : opsviv,
14613 if (!svp || *svp != uninit_sv)
14617 if (obase->op_type == OP_HELEM)
14618 return varname(gv, '%', o->op_targ,
14619 kidsv, 0, FUV_SUBSCRIPT_HASH);
14621 return varname(gv, '@', o->op_targ, NULL,
14622 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
14623 FUV_SUBSCRIPT_ARRAY);
14626 /* index is an expression;
14627 * attempt to find a match within the aggregate */
14628 if (obase->op_type == OP_HELEM) {
14629 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14631 return varname(gv, '%', o->op_targ,
14632 keysv, 0, FUV_SUBSCRIPT_HASH);
14636 = find_array_subscript((const AV *)sv, uninit_sv);
14638 return varname(gv, '@', o->op_targ,
14639 NULL, index, FUV_SUBSCRIPT_ARRAY);
14644 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
14646 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
14652 /* only examine RHS */
14653 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
14656 o = cUNOPx(obase)->op_first;
14657 if ( o->op_type == OP_PUSHMARK
14658 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
14662 if (!o->op_sibling) {
14663 /* one-arg version of open is highly magical */
14665 if (o->op_type == OP_GV) { /* open FOO; */
14667 if (match && GvSV(gv) != uninit_sv)
14669 return varname(gv, '$', 0,
14670 NULL, 0, FUV_SUBSCRIPT_NONE);
14672 /* other possibilities not handled are:
14673 * open $x; or open my $x; should return '${*$x}'
14674 * open expr; should return '$'.expr ideally
14680 /* ops where $_ may be an implicit arg */
14685 if ( !(obase->op_flags & OPf_STACKED)) {
14686 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
14687 ? PAD_SVl(obase->op_targ)
14690 sv = sv_newmortal();
14691 sv_setpvs(sv, "$_");
14700 match = 1; /* print etc can return undef on defined args */
14701 /* skip filehandle as it can't produce 'undef' warning */
14702 o = cUNOPx(obase)->op_first;
14703 if ((obase->op_flags & OPf_STACKED)
14705 ( o->op_type == OP_PUSHMARK
14706 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
14707 o = o->op_sibling->op_sibling;
14711 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
14712 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
14714 /* the following ops are capable of returning PL_sv_undef even for
14715 * defined arg(s) */
14734 case OP_GETPEERNAME:
14782 case OP_SMARTMATCH:
14791 /* XXX tmp hack: these two may call an XS sub, and currently
14792 XS subs don't have a SUB entry on the context stack, so CV and
14793 pad determination goes wrong, and BAD things happen. So, just
14794 don't try to determine the value under those circumstances.
14795 Need a better fix at dome point. DAPM 11/2007 */
14801 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
14802 if (gv && GvSV(gv) == uninit_sv)
14803 return newSVpvs_flags("$.", SVs_TEMP);
14808 /* def-ness of rval pos() is independent of the def-ness of its arg */
14809 if ( !(obase->op_flags & OPf_MOD))
14814 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
14815 return newSVpvs_flags("${$/}", SVs_TEMP);
14820 if (!(obase->op_flags & OPf_KIDS))
14822 o = cUNOPx(obase)->op_first;
14828 /* This loop checks all the kid ops, skipping any that cannot pos-
14829 * sibly be responsible for the uninitialized value; i.e., defined
14830 * constants and ops that return nothing. If there is only one op
14831 * left that is not skipped, then we *know* it is responsible for
14832 * the uninitialized value. If there is more than one op left, we
14833 * have to look for an exact match in the while() loop below.
14834 * Note that we skip padrange, because the individual pad ops that
14835 * it replaced are still in the tree, so we work on them instead.
14838 for (kid=o; kid; kid = kid->op_sibling) {
14840 const OPCODE type = kid->op_type;
14841 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
14842 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
14843 || (type == OP_PUSHMARK)
14844 || (type == OP_PADRANGE)
14848 if (o2) { /* more than one found */
14855 return find_uninit_var(o2, uninit_sv, match);
14857 /* scan all args */
14859 sv = find_uninit_var(o, uninit_sv, 1);
14871 =for apidoc report_uninit
14873 Print appropriate "Use of uninitialized variable" warning.
14879 Perl_report_uninit(pTHX_ const SV *uninit_sv)
14883 SV* varname = NULL;
14884 if (uninit_sv && PL_curpad) {
14885 varname = find_uninit_var(PL_op, uninit_sv,0);
14887 sv_insert(varname, 0, 0, " ", 1);
14889 /* PL_warn_uninit_sv is constant */
14890 GCC_DIAG_IGNORE(-Wformat-nonliteral);
14891 /* diag_listed_as: Use of uninitialized value%s */
14892 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
14893 SVfARG(varname ? varname : &PL_sv_no),
14894 " in ", OP_DESC(PL_op));
14898 /* PL_warn_uninit is constant */
14899 GCC_DIAG_IGNORE(-Wformat-nonliteral);
14900 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
14908 * c-indentation-style: bsd
14909 * c-basic-offset: 4
14910 * indent-tabs-mode: nil
14913 * ex: set ts=8 sts=4 sw=4 et: