3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%"UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referent = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referent = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referent will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referent;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1531 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533 Use the C<SvGROW> wrapper instead.
1538 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1541 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1545 PERL_ARGS_ASSERT_SV_GROW;
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1553 else if (SvOOK(sv)) { /* pv is offset? */
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1562 s = SvPVX_mutable(sv);
1565 #ifdef PERL_COPY_ON_WRITE
1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1567 * to store the COW count. So in general, allocate one more byte than
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
1571 * Only increment if the allocation isn't MEM_SIZE_MAX,
1572 * otherwise it will wrap to 0.
1574 if ( newlen != MEM_SIZE_MAX )
1578 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1579 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1582 if (newlen > SvLEN(sv)) { /* need more room? */
1583 STRLEN minlen = SvCUR(sv);
1584 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1585 if (newlen < minlen)
1587 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1589 /* Don't round up on the first allocation, as odds are pretty good that
1590 * the initial request is accurate as to what is really needed */
1592 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1593 if (rounded > newlen)
1597 if (SvLEN(sv) && s) {
1598 s = (char*)saferealloc(s, newlen);
1601 s = (char*)safemalloc(newlen);
1602 if (SvPVX_const(sv) && SvCUR(sv)) {
1603 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1607 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1608 /* Do this here, do it once, do it right, and then we will never get
1609 called back into sv_grow() unless there really is some growing
1611 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1613 SvLEN_set(sv, newlen);
1620 =for apidoc sv_setiv
1622 Copies an integer into the given SV, upgrading first if necessary.
1623 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1629 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1631 PERL_ARGS_ASSERT_SV_SETIV;
1633 SV_CHECK_THINKFIRST_COW_DROP(sv);
1634 switch (SvTYPE(sv)) {
1637 sv_upgrade(sv, SVt_IV);
1640 sv_upgrade(sv, SVt_PVIV);
1644 if (!isGV_with_GP(sv))
1651 /* diag_listed_as: Can't coerce %s to %s in %s */
1652 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1657 (void)SvIOK_only(sv); /* validate number */
1663 =for apidoc sv_setiv_mg
1665 Like C<sv_setiv>, but also handles 'set' magic.
1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1680 =for apidoc sv_setuv
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1691 PERL_ARGS_ASSERT_SV_SETUV;
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1714 =for apidoc sv_setuv_mg
1716 Like C<sv_setuv>, but also handles 'set' magic.
1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1731 =for apidoc sv_setnv
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1742 PERL_ARGS_ASSERT_SV_SETNV;
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1748 sv_upgrade(sv, SVt_NV);
1752 sv_upgrade(sv, SVt_PVNV);
1756 if (!isGV_with_GP(sv))
1763 /* diag_listed_as: Can't coerce %s to %s in %s */
1764 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1770 (void)SvNOK_only(sv); /* validate number */
1775 =for apidoc sv_setnv_mg
1777 Like C<sv_setnv>, but also handles 'set' magic.
1783 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1785 PERL_ARGS_ASSERT_SV_SETNV_MG;
1791 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1792 * not incrementable warning display.
1793 * Originally part of S_not_a_number().
1794 * The return value may be != tmpbuf.
1798 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1801 PERL_ARGS_ASSERT_SV_DISPLAY;
1804 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1805 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1808 const char * const limit = tmpbuf + tmpbuf_size - 8;
1809 /* each *s can expand to 4 chars + "...\0",
1810 i.e. need room for 8 chars */
1812 const char *s = SvPVX_const(sv);
1813 const char * const end = s + SvCUR(sv);
1814 for ( ; s < end && d < limit; s++ ) {
1816 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1820 /* Map to ASCII "equivalent" of Latin1 */
1821 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1827 else if (ch == '\r') {
1831 else if (ch == '\f') {
1835 else if (ch == '\\') {
1839 else if (ch == '\0') {
1843 else if (isPRINT_LC(ch))
1862 /* Print an "isn't numeric" warning, using a cleaned-up,
1863 * printable version of the offending string
1867 S_not_a_number(pTHX_ SV *const sv)
1872 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1874 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1877 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1878 /* diag_listed_as: Argument "%s" isn't numeric%s */
1879 "Argument \"%s\" isn't numeric in %s", pv,
1882 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1883 /* diag_listed_as: Argument "%s" isn't numeric%s */
1884 "Argument \"%s\" isn't numeric", pv);
1888 S_not_incrementable(pTHX_ SV *const sv) {
1892 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1894 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1896 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1897 "Argument \"%s\" treated as 0 in increment (++)", pv);
1901 =for apidoc looks_like_number
1903 Test if the content of an SV looks like a number (or is a number).
1904 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1905 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1912 Perl_looks_like_number(pTHX_ SV *const sv)
1918 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1920 if (SvPOK(sv) || SvPOKp(sv)) {
1921 sbegin = SvPV_nomg_const(sv, len);
1924 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1925 numtype = grok_number(sbegin, len, NULL);
1926 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1930 S_glob_2number(pTHX_ GV * const gv)
1932 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1934 /* We know that all GVs stringify to something that is not-a-number,
1935 so no need to test that. */
1936 if (ckWARN(WARN_NUMERIC))
1938 SV *const buffer = sv_newmortal();
1939 gv_efullname3(buffer, gv, "*");
1940 not_a_number(buffer);
1942 /* We just want something true to return, so that S_sv_2iuv_common
1943 can tail call us and return true. */
1947 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1948 until proven guilty, assume that things are not that bad... */
1953 As 64 bit platforms often have an NV that doesn't preserve all bits of
1954 an IV (an assumption perl has been based on to date) it becomes necessary
1955 to remove the assumption that the NV always carries enough precision to
1956 recreate the IV whenever needed, and that the NV is the canonical form.
1957 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1958 precision as a side effect of conversion (which would lead to insanity
1959 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1960 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1961 where precision was lost, and IV/UV/NV slots that have a valid conversion
1962 which has lost no precision
1963 2) to ensure that if a numeric conversion to one form is requested that
1964 would lose precision, the precise conversion (or differently
1965 imprecise conversion) is also performed and cached, to prevent
1966 requests for different numeric formats on the same SV causing
1967 lossy conversion chains. (lossless conversion chains are perfectly
1972 SvIOKp is true if the IV slot contains a valid value
1973 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1974 SvNOKp is true if the NV slot contains a valid value
1975 SvNOK is true only if the NV value is accurate
1978 while converting from PV to NV, check to see if converting that NV to an
1979 IV(or UV) would lose accuracy over a direct conversion from PV to
1980 IV(or UV). If it would, cache both conversions, return NV, but mark
1981 SV as IOK NOKp (ie not NOK).
1983 While converting from PV to IV, check to see if converting that IV to an
1984 NV would lose accuracy over a direct conversion from PV to NV. If it
1985 would, cache both conversions, flag similarly.
1987 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1988 correctly because if IV & NV were set NV *always* overruled.
1989 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1990 changes - now IV and NV together means that the two are interchangeable:
1991 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1993 The benefit of this is that operations such as pp_add know that if
1994 SvIOK is true for both left and right operands, then integer addition
1995 can be used instead of floating point (for cases where the result won't
1996 overflow). Before, floating point was always used, which could lead to
1997 loss of precision compared with integer addition.
1999 * making IV and NV equal status should make maths accurate on 64 bit
2001 * may speed up maths somewhat if pp_add and friends start to use
2002 integers when possible instead of fp. (Hopefully the overhead in
2003 looking for SvIOK and checking for overflow will not outweigh the
2004 fp to integer speedup)
2005 * will slow down integer operations (callers of SvIV) on "inaccurate"
2006 values, as the change from SvIOK to SvIOKp will cause a call into
2007 sv_2iv each time rather than a macro access direct to the IV slot
2008 * should speed up number->string conversion on integers as IV is
2009 favoured when IV and NV are equally accurate
2011 ####################################################################
2012 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2013 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2014 On the other hand, SvUOK is true iff UV.
2015 ####################################################################
2017 Your mileage will vary depending your CPU's relative fp to integer
2021 #ifndef NV_PRESERVES_UV
2022 # define IS_NUMBER_UNDERFLOW_IV 1
2023 # define IS_NUMBER_UNDERFLOW_UV 2
2024 # define IS_NUMBER_IV_AND_UV 2
2025 # define IS_NUMBER_OVERFLOW_IV 4
2026 # define IS_NUMBER_OVERFLOW_UV 5
2028 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2030 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2032 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2038 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2039 PERL_UNUSED_CONTEXT;
2041 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));
2042 if (SvNVX(sv) < (NV)IV_MIN) {
2043 (void)SvIOKp_on(sv);
2045 SvIV_set(sv, IV_MIN);
2046 return IS_NUMBER_UNDERFLOW_IV;
2048 if (SvNVX(sv) > (NV)UV_MAX) {
2049 (void)SvIOKp_on(sv);
2052 SvUV_set(sv, UV_MAX);
2053 return IS_NUMBER_OVERFLOW_UV;
2055 (void)SvIOKp_on(sv);
2057 /* Can't use strtol etc to convert this string. (See truth table in
2059 if (SvNVX(sv) <= (UV)IV_MAX) {
2060 SvIV_set(sv, I_V(SvNVX(sv)));
2061 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2062 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2064 /* Integer is imprecise. NOK, IOKp */
2066 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2069 SvUV_set(sv, U_V(SvNVX(sv)));
2070 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2071 if (SvUVX(sv) == UV_MAX) {
2072 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2073 possibly be preserved by NV. Hence, it must be overflow.
2075 return IS_NUMBER_OVERFLOW_UV;
2077 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2079 /* Integer is imprecise. NOK, IOKp */
2081 return IS_NUMBER_OVERFLOW_IV;
2083 #endif /* !NV_PRESERVES_UV*/
2085 /* If numtype is infnan, set the NV of the sv accordingly.
2086 * If numtype is anything else, try setting the NV using Atof(PV). */
2088 # pragma warning(push)
2089 # pragma warning(disable:4756;disable:4056)
2092 S_sv_setnv(pTHX_ SV* sv, int numtype)
2094 bool pok = cBOOL(SvPOK(sv));
2097 if ((numtype & IS_NUMBER_INFINITY)) {
2098 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2103 if ((numtype & IS_NUMBER_NAN)) {
2104 SvNV_set(sv, NV_NAN);
2109 SvNV_set(sv, Atof(SvPVX_const(sv)));
2110 /* Purposefully no true nok here, since we don't want to blow
2111 * away the possible IOK/UV of an existing sv. */
2114 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2116 SvPOK_on(sv); /* PV is okay, though. */
2120 # pragma warning(pop)
2124 S_sv_2iuv_common(pTHX_ SV *const sv)
2126 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2129 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2130 * without also getting a cached IV/UV from it at the same time
2131 * (ie PV->NV conversion should detect loss of accuracy and cache
2132 * IV or UV at same time to avoid this. */
2133 /* IV-over-UV optimisation - choose to cache IV if possible */
2135 if (SvTYPE(sv) == SVt_NV)
2136 sv_upgrade(sv, SVt_PVNV);
2138 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2139 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2140 certainly cast into the IV range at IV_MAX, whereas the correct
2141 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2143 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2144 if (Perl_isnan(SvNVX(sv))) {
2150 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2151 SvIV_set(sv, I_V(SvNVX(sv)));
2152 if (SvNVX(sv) == (NV) SvIVX(sv)
2153 #ifndef NV_PRESERVES_UV
2154 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2155 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2156 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2157 /* Don't flag it as "accurately an integer" if the number
2158 came from a (by definition imprecise) NV operation, and
2159 we're outside the range of NV integer precision */
2163 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2165 /* scalar has trailing garbage, eg "42a" */
2167 DEBUG_c(PerlIO_printf(Perl_debug_log,
2168 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2174 /* IV not precise. No need to convert from PV, as NV
2175 conversion would already have cached IV if it detected
2176 that PV->IV would be better than PV->NV->IV
2177 flags already correct - don't set public IOK. */
2178 DEBUG_c(PerlIO_printf(Perl_debug_log,
2179 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2184 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2185 but the cast (NV)IV_MIN rounds to a the value less (more
2186 negative) than IV_MIN which happens to be equal to SvNVX ??
2187 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2188 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2189 (NV)UVX == NVX are both true, but the values differ. :-(
2190 Hopefully for 2s complement IV_MIN is something like
2191 0x8000000000000000 which will be exact. NWC */
2194 SvUV_set(sv, U_V(SvNVX(sv)));
2196 (SvNVX(sv) == (NV) SvUVX(sv))
2197 #ifndef NV_PRESERVES_UV
2198 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2199 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2200 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2201 /* Don't flag it as "accurately an integer" if the number
2202 came from a (by definition imprecise) NV operation, and
2203 we're outside the range of NV integer precision */
2209 DEBUG_c(PerlIO_printf(Perl_debug_log,
2210 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2216 else if (SvPOKp(sv)) {
2219 const char *s = SvPVX_const(sv);
2220 const STRLEN cur = SvCUR(sv);
2222 /* short-cut for a single digit string like "1" */
2227 if (SvTYPE(sv) < SVt_PVIV)
2228 sv_upgrade(sv, SVt_PVIV);
2230 SvIV_set(sv, (IV)(c - '0'));
2235 numtype = grok_number(s, cur, &value);
2236 /* We want to avoid a possible problem when we cache an IV/ a UV which
2237 may be later translated to an NV, and the resulting NV is not
2238 the same as the direct translation of the initial string
2239 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2240 be careful to ensure that the value with the .456 is around if the
2241 NV value is requested in the future).
2243 This means that if we cache such an IV/a UV, we need to cache the
2244 NV as well. Moreover, we trade speed for space, and do not
2245 cache the NV if we are sure it's not needed.
2248 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2249 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2250 == IS_NUMBER_IN_UV) {
2251 /* It's definitely an integer, only upgrade to PVIV */
2252 if (SvTYPE(sv) < SVt_PVIV)
2253 sv_upgrade(sv, SVt_PVIV);
2255 } else if (SvTYPE(sv) < SVt_PVNV)
2256 sv_upgrade(sv, SVt_PVNV);
2258 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2259 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2261 S_sv_setnv(aTHX_ sv, numtype);
2265 /* If NVs preserve UVs then we only use the UV value if we know that
2266 we aren't going to call atof() below. If NVs don't preserve UVs
2267 then the value returned may have more precision than atof() will
2268 return, even though value isn't perfectly accurate. */
2269 if ((numtype & (IS_NUMBER_IN_UV
2270 #ifdef NV_PRESERVES_UV
2273 )) == IS_NUMBER_IN_UV) {
2274 /* This won't turn off the public IOK flag if it was set above */
2275 (void)SvIOKp_on(sv);
2277 if (!(numtype & IS_NUMBER_NEG)) {
2279 if (value <= (UV)IV_MAX) {
2280 SvIV_set(sv, (IV)value);
2282 /* it didn't overflow, and it was positive. */
2283 SvUV_set(sv, value);
2287 /* 2s complement assumption */
2288 if (value <= (UV)IV_MIN) {
2289 SvIV_set(sv, value == (UV)IV_MIN
2290 ? IV_MIN : -(IV)value);
2292 /* Too negative for an IV. This is a double upgrade, but
2293 I'm assuming it will be rare. */
2294 if (SvTYPE(sv) < SVt_PVNV)
2295 sv_upgrade(sv, SVt_PVNV);
2299 SvNV_set(sv, -(NV)value);
2300 SvIV_set(sv, IV_MIN);
2304 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2305 will be in the previous block to set the IV slot, and the next
2306 block to set the NV slot. So no else here. */
2308 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2309 != IS_NUMBER_IN_UV) {
2310 /* It wasn't an (integer that doesn't overflow the UV). */
2311 S_sv_setnv(aTHX_ sv, numtype);
2313 if (! numtype && ckWARN(WARN_NUMERIC))
2316 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2317 PTR2UV(sv), SvNVX(sv)));
2319 #ifdef NV_PRESERVES_UV
2320 (void)SvIOKp_on(sv);
2322 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2323 if (Perl_isnan(SvNVX(sv))) {
2329 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2330 SvIV_set(sv, I_V(SvNVX(sv)));
2331 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2334 NOOP; /* Integer is imprecise. NOK, IOKp */
2336 /* UV will not work better than IV */
2338 if (SvNVX(sv) > (NV)UV_MAX) {
2340 /* Integer is inaccurate. NOK, IOKp, is UV */
2341 SvUV_set(sv, UV_MAX);
2343 SvUV_set(sv, U_V(SvNVX(sv)));
2344 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2345 NV preservse UV so can do correct comparison. */
2346 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2349 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2354 #else /* NV_PRESERVES_UV */
2355 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2356 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2357 /* The IV/UV slot will have been set from value returned by
2358 grok_number above. The NV slot has just been set using
2361 assert (SvIOKp(sv));
2363 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2364 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2365 /* Small enough to preserve all bits. */
2366 (void)SvIOKp_on(sv);
2368 SvIV_set(sv, I_V(SvNVX(sv)));
2369 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2371 /* Assumption: first non-preserved integer is < IV_MAX,
2372 this NV is in the preserved range, therefore: */
2373 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2375 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);
2379 0 0 already failed to read UV.
2380 0 1 already failed to read UV.
2381 1 0 you won't get here in this case. IV/UV
2382 slot set, public IOK, Atof() unneeded.
2383 1 1 already read UV.
2384 so there's no point in sv_2iuv_non_preserve() attempting
2385 to use atol, strtol, strtoul etc. */
2387 sv_2iuv_non_preserve (sv, numtype);
2389 sv_2iuv_non_preserve (sv);
2393 #endif /* NV_PRESERVES_UV */
2394 /* It might be more code efficient to go through the entire logic above
2395 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2396 gets complex and potentially buggy, so more programmer efficient
2397 to do it this way, by turning off the public flags: */
2399 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2403 if (isGV_with_GP(sv))
2404 return glob_2number(MUTABLE_GV(sv));
2406 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2408 if (SvTYPE(sv) < SVt_IV)
2409 /* Typically the caller expects that sv_any is not NULL now. */
2410 sv_upgrade(sv, SVt_IV);
2411 /* Return 0 from the caller. */
2418 =for apidoc sv_2iv_flags
2420 Return the integer value of an SV, doing any necessary string
2421 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2422 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2428 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2430 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2432 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2433 && SvTYPE(sv) != SVt_PVFM);
2435 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2441 if (flags & SV_SKIP_OVERLOAD)
2443 tmpstr = AMG_CALLunary(sv, numer_amg);
2444 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2445 return SvIV(tmpstr);
2448 return PTR2IV(SvRV(sv));
2451 if (SvVALID(sv) || isREGEXP(sv)) {
2452 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2453 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2454 In practice they are extremely unlikely to actually get anywhere
2455 accessible by user Perl code - the only way that I'm aware of is when
2456 a constant subroutine which is used as the second argument to index.
2458 Regexps have no SvIVX and SvNVX fields.
2460 assert(isREGEXP(sv) || SvPOKp(sv));
2463 const char * const ptr =
2464 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2466 = grok_number(ptr, SvCUR(sv), &value);
2468 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2469 == IS_NUMBER_IN_UV) {
2470 /* It's definitely an integer */
2471 if (numtype & IS_NUMBER_NEG) {
2472 if (value < (UV)IV_MIN)
2475 if (value < (UV)IV_MAX)
2480 /* Quite wrong but no good choices. */
2481 if ((numtype & IS_NUMBER_INFINITY)) {
2482 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2483 } else if ((numtype & IS_NUMBER_NAN)) {
2484 return 0; /* So wrong. */
2488 if (ckWARN(WARN_NUMERIC))
2491 return I_V(Atof(ptr));
2495 if (SvTHINKFIRST(sv)) {
2496 if (SvREADONLY(sv) && !SvOK(sv)) {
2497 if (ckWARN(WARN_UNINITIALIZED))
2504 if (S_sv_2iuv_common(aTHX_ sv))
2508 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2509 PTR2UV(sv),SvIVX(sv)));
2510 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2514 =for apidoc sv_2uv_flags
2516 Return the unsigned integer value of an SV, doing any necessary string
2517 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2518 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2524 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2526 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2528 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2534 if (flags & SV_SKIP_OVERLOAD)
2536 tmpstr = AMG_CALLunary(sv, numer_amg);
2537 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2538 return SvUV(tmpstr);
2541 return PTR2UV(SvRV(sv));
2544 if (SvVALID(sv) || isREGEXP(sv)) {
2545 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2546 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2547 Regexps have no SvIVX and SvNVX fields. */
2548 assert(isREGEXP(sv) || SvPOKp(sv));
2551 const char * const ptr =
2552 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2554 = grok_number(ptr, SvCUR(sv), &value);
2556 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2557 == IS_NUMBER_IN_UV) {
2558 /* It's definitely an integer */
2559 if (!(numtype & IS_NUMBER_NEG))
2563 /* Quite wrong but no good choices. */
2564 if ((numtype & IS_NUMBER_INFINITY)) {
2565 return UV_MAX; /* So wrong. */
2566 } else if ((numtype & IS_NUMBER_NAN)) {
2567 return 0; /* So wrong. */
2571 if (ckWARN(WARN_NUMERIC))
2574 return U_V(Atof(ptr));
2578 if (SvTHINKFIRST(sv)) {
2579 if (SvREADONLY(sv) && !SvOK(sv)) {
2580 if (ckWARN(WARN_UNINITIALIZED))
2587 if (S_sv_2iuv_common(aTHX_ sv))
2591 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2592 PTR2UV(sv),SvUVX(sv)));
2593 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2597 =for apidoc sv_2nv_flags
2599 Return the num value of an SV, doing any necessary string or integer
2600 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2601 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2607 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2609 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2611 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2612 && SvTYPE(sv) != SVt_PVFM);
2613 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2614 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2615 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2616 Regexps have no SvIVX and SvNVX fields. */
2618 if (flags & SV_GMAGIC)
2622 if (SvPOKp(sv) && !SvIOKp(sv)) {
2623 ptr = SvPVX_const(sv);
2625 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2626 !grok_number(ptr, SvCUR(sv), NULL))
2632 return (NV)SvUVX(sv);
2634 return (NV)SvIVX(sv);
2640 ptr = RX_WRAPPED((REGEXP *)sv);
2643 assert(SvTYPE(sv) >= SVt_PVMG);
2644 /* This falls through to the report_uninit near the end of the
2646 } else if (SvTHINKFIRST(sv)) {
2651 if (flags & SV_SKIP_OVERLOAD)
2653 tmpstr = AMG_CALLunary(sv, numer_amg);
2654 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2655 return SvNV(tmpstr);
2658 return PTR2NV(SvRV(sv));
2660 if (SvREADONLY(sv) && !SvOK(sv)) {
2661 if (ckWARN(WARN_UNINITIALIZED))
2666 if (SvTYPE(sv) < SVt_NV) {
2667 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2668 sv_upgrade(sv, SVt_NV);
2670 STORE_NUMERIC_LOCAL_SET_STANDARD();
2671 PerlIO_printf(Perl_debug_log,
2672 "0x%"UVxf" num(%" NVgf ")\n",
2673 PTR2UV(sv), SvNVX(sv));
2674 RESTORE_NUMERIC_LOCAL();
2677 else if (SvTYPE(sv) < SVt_PVNV)
2678 sv_upgrade(sv, SVt_PVNV);
2683 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2684 #ifdef NV_PRESERVES_UV
2690 /* Only set the public NV OK flag if this NV preserves the IV */
2691 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2693 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2694 : (SvIVX(sv) == I_V(SvNVX(sv))))
2700 else if (SvPOKp(sv)) {
2702 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2703 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2705 #ifdef NV_PRESERVES_UV
2706 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2707 == IS_NUMBER_IN_UV) {
2708 /* It's definitely an integer */
2709 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2711 S_sv_setnv(aTHX_ sv, numtype);
2718 SvNV_set(sv, Atof(SvPVX_const(sv)));
2719 /* Only set the public NV OK flag if this NV preserves the value in
2720 the PV at least as well as an IV/UV would.
2721 Not sure how to do this 100% reliably. */
2722 /* if that shift count is out of range then Configure's test is
2723 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2725 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2726 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2727 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2728 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2729 /* Can't use strtol etc to convert this string, so don't try.
2730 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2733 /* value has been set. It may not be precise. */
2734 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2735 /* 2s complement assumption for (UV)IV_MIN */
2736 SvNOK_on(sv); /* Integer is too negative. */
2741 if (numtype & IS_NUMBER_NEG) {
2742 /* -IV_MIN is undefined, but we should never reach
2743 * this point with both IS_NUMBER_NEG and value ==
2745 assert(value != (UV)IV_MIN);
2746 SvIV_set(sv, -(IV)value);
2747 } else if (value <= (UV)IV_MAX) {
2748 SvIV_set(sv, (IV)value);
2750 SvUV_set(sv, value);
2754 if (numtype & IS_NUMBER_NOT_INT) {
2755 /* I believe that even if the original PV had decimals,
2756 they are lost beyond the limit of the FP precision.
2757 However, neither is canonical, so both only get p
2758 flags. NWC, 2000/11/25 */
2759 /* Both already have p flags, so do nothing */
2761 const NV nv = SvNVX(sv);
2762 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2763 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2764 if (SvIVX(sv) == I_V(nv)) {
2767 /* It had no "." so it must be integer. */
2771 /* between IV_MAX and NV(UV_MAX).
2772 Could be slightly > UV_MAX */
2774 if (numtype & IS_NUMBER_NOT_INT) {
2775 /* UV and NV both imprecise. */
2777 const UV nv_as_uv = U_V(nv);
2779 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2788 /* It might be more code efficient to go through the entire logic above
2789 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2790 gets complex and potentially buggy, so more programmer efficient
2791 to do it this way, by turning off the public flags: */
2793 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2794 #endif /* NV_PRESERVES_UV */
2797 if (isGV_with_GP(sv)) {
2798 glob_2number(MUTABLE_GV(sv));
2802 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2804 assert (SvTYPE(sv) >= SVt_NV);
2805 /* Typically the caller expects that sv_any is not NULL now. */
2806 /* XXX Ilya implies that this is a bug in callers that assume this
2807 and ideally should be fixed. */
2811 STORE_NUMERIC_LOCAL_SET_STANDARD();
2812 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2813 PTR2UV(sv), SvNVX(sv));
2814 RESTORE_NUMERIC_LOCAL();
2822 Return an SV with the numeric value of the source SV, doing any necessary
2823 reference or overload conversion. The caller is expected to have handled
2830 Perl_sv_2num(pTHX_ SV *const sv)
2832 PERL_ARGS_ASSERT_SV_2NUM;
2837 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2838 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2839 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2840 return sv_2num(tmpsv);
2842 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2845 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2846 * UV as a string towards the end of buf, and return pointers to start and
2849 * We assume that buf is at least TYPE_CHARS(UV) long.
2853 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2855 char *ptr = buf + TYPE_CHARS(UV);
2856 char * const ebuf = ptr;
2859 PERL_ARGS_ASSERT_UIV_2BUF;
2867 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2871 *--ptr = '0' + (char)(uv % 10);
2879 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2880 * infinity or a not-a-number, writes the appropriate strings to the
2881 * buffer, including a zero byte. On success returns the written length,
2882 * excluding the zero byte, on failure (not an infinity, not a nan)
2883 * returns zero, assert-fails on maxlen being too short.
2885 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2886 * shared string constants we point to, instead of generating a new
2887 * string for each instance. */
2889 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2891 assert(maxlen >= 4);
2892 if (Perl_isinf(nv)) {
2894 if (maxlen < 5) /* "-Inf\0" */
2904 else if (Perl_isnan(nv)) {
2908 /* XXX optionally output the payload mantissa bits as
2909 * "(unsigned)" (to match the nan("...") C99 function,
2910 * or maybe as "(0xhhh...)" would make more sense...
2911 * provide a format string so that the user can decide?
2912 * NOTE: would affect the maxlen and assert() logic.*/
2917 assert((s == buffer + 3) || (s == buffer + 4));
2919 return s - buffer - 1; /* -1: excluding the zero byte */
2923 =for apidoc sv_2pv_flags
2925 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2926 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2927 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2928 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2934 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2938 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2940 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2941 && SvTYPE(sv) != SVt_PVFM);
2942 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2947 if (flags & SV_SKIP_OVERLOAD)
2949 tmpstr = AMG_CALLunary(sv, string_amg);
2950 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2951 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2953 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2957 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2958 if (flags & SV_CONST_RETURN) {
2959 pv = (char *) SvPVX_const(tmpstr);
2961 pv = (flags & SV_MUTABLE_RETURN)
2962 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2965 *lp = SvCUR(tmpstr);
2967 pv = sv_2pv_flags(tmpstr, lp, flags);
2980 SV *const referent = SvRV(sv);
2984 retval = buffer = savepvn("NULLREF", len);
2985 } else if (SvTYPE(referent) == SVt_REGEXP &&
2986 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2987 amagic_is_enabled(string_amg))) {
2988 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2992 /* If the regex is UTF-8 we want the containing scalar to
2993 have an UTF-8 flag too */
3000 *lp = RX_WRAPLEN(re);
3002 return RX_WRAPPED(re);
3004 const char *const typestr = sv_reftype(referent, 0);
3005 const STRLEN typelen = strlen(typestr);
3006 UV addr = PTR2UV(referent);
3007 const char *stashname = NULL;
3008 STRLEN stashnamelen = 0; /* hush, gcc */
3009 const char *buffer_end;
3011 if (SvOBJECT(referent)) {
3012 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3015 stashname = HEK_KEY(name);
3016 stashnamelen = HEK_LEN(name);
3018 if (HEK_UTF8(name)) {
3024 stashname = "__ANON__";
3027 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3028 + 2 * sizeof(UV) + 2 /* )\0 */;
3030 len = typelen + 3 /* (0x */
3031 + 2 * sizeof(UV) + 2 /* )\0 */;
3034 Newx(buffer, len, char);
3035 buffer_end = retval = buffer + len;
3037 /* Working backwards */
3041 *--retval = PL_hexdigit[addr & 15];
3042 } while (addr >>= 4);
3048 memcpy(retval, typestr, typelen);
3052 retval -= stashnamelen;
3053 memcpy(retval, stashname, stashnamelen);
3055 /* retval may not necessarily have reached the start of the
3057 assert (retval >= buffer);
3059 len = buffer_end - retval - 1; /* -1 for that \0 */
3071 if (flags & SV_MUTABLE_RETURN)
3072 return SvPVX_mutable(sv);
3073 if (flags & SV_CONST_RETURN)
3074 return (char *)SvPVX_const(sv);
3079 /* I'm assuming that if both IV and NV are equally valid then
3080 converting the IV is going to be more efficient */
3081 const U32 isUIOK = SvIsUV(sv);
3082 char buf[TYPE_CHARS(UV)];
3086 if (SvTYPE(sv) < SVt_PVIV)
3087 sv_upgrade(sv, SVt_PVIV);
3088 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3090 /* inlined from sv_setpvn */
3091 s = SvGROW_mutable(sv, len + 1);
3092 Move(ptr, s, len, char);
3097 else if (SvNOK(sv)) {
3098 if (SvTYPE(sv) < SVt_PVNV)
3099 sv_upgrade(sv, SVt_PVNV);
3100 if (SvNVX(sv) == 0.0
3101 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3102 && !Perl_isnan(SvNVX(sv))
3105 s = SvGROW_mutable(sv, 2);
3110 STRLEN size = 5; /* "-Inf\0" */
3112 s = SvGROW_mutable(sv, size);
3113 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3119 /* some Xenix systems wipe out errno here */
3128 5 + /* exponent digits */
3132 s = SvGROW_mutable(sv, size);
3133 #ifndef USE_LOCALE_NUMERIC
3134 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3140 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3141 STORE_LC_NUMERIC_SET_TO_NEEDED();
3143 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3144 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3145 size += SvLEN(PL_numeric_radix_sv) - 1;
3146 s = SvGROW_mutable(sv, size);
3149 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3151 /* If the radix character is UTF-8, and actually is in the
3152 * output, turn on the UTF-8 flag for the scalar */
3154 && SvUTF8(PL_numeric_radix_sv)
3155 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3160 RESTORE_LC_NUMERIC();
3163 /* We don't call SvPOK_on(), because it may come to
3164 * pass that the locale changes so that the
3165 * stringification we just did is no longer correct. We
3166 * will have to re-stringify every time it is needed */
3173 else if (isGV_with_GP(sv)) {
3174 GV *const gv = MUTABLE_GV(sv);
3175 SV *const buffer = sv_newmortal();
3177 gv_efullname3(buffer, gv, "*");
3179 assert(SvPOK(buffer));
3183 *lp = SvCUR(buffer);
3184 return SvPVX(buffer);
3186 else if (isREGEXP(sv)) {
3187 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3188 return RX_WRAPPED((REGEXP *)sv);
3193 if (flags & SV_UNDEF_RETURNS_NULL)
3195 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3197 /* Typically the caller expects that sv_any is not NULL now. */
3198 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3199 sv_upgrade(sv, SVt_PV);
3204 const STRLEN len = s - SvPVX_const(sv);
3209 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3210 PTR2UV(sv),SvPVX_const(sv)));
3211 if (flags & SV_CONST_RETURN)
3212 return (char *)SvPVX_const(sv);
3213 if (flags & SV_MUTABLE_RETURN)
3214 return SvPVX_mutable(sv);
3219 =for apidoc sv_copypv
3221 Copies a stringified representation of the source SV into the
3222 destination SV. Automatically performs any necessary C<mg_get> and
3223 coercion of numeric values into strings. Guaranteed to preserve
3224 C<UTF8> flag even from overloaded objects. Similar in nature to
3225 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3226 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3227 would lose the UTF-8'ness of the PV.
3229 =for apidoc sv_copypv_nomg
3231 Like C<sv_copypv>, but doesn't invoke get magic first.
3233 =for apidoc sv_copypv_flags
3235 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3236 has the C<SV_GMAGIC> bit set.
3242 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3247 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3249 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3250 sv_setpvn(dsv,s,len);
3258 =for apidoc sv_2pvbyte
3260 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3261 to its length. May cause the SV to be downgraded from UTF-8 as a
3264 Usually accessed via the C<SvPVbyte> macro.
3270 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3272 PERL_ARGS_ASSERT_SV_2PVBYTE;
3275 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3276 || isGV_with_GP(sv) || SvROK(sv)) {
3277 SV *sv2 = sv_newmortal();
3278 sv_copypv_nomg(sv2,sv);
3281 sv_utf8_downgrade(sv,0);
3282 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3286 =for apidoc sv_2pvutf8
3288 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3289 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3291 Usually accessed via the C<SvPVutf8> macro.
3297 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3299 PERL_ARGS_ASSERT_SV_2PVUTF8;
3301 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3302 || isGV_with_GP(sv) || SvROK(sv))
3303 sv = sv_mortalcopy(sv);
3306 sv_utf8_upgrade_nomg(sv);
3307 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3312 =for apidoc sv_2bool
3314 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3315 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3316 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3318 =for apidoc sv_2bool_flags
3320 This function is only used by C<sv_true()> and friends, and only if
3321 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3322 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3329 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3331 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3334 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3340 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3341 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3344 if(SvGMAGICAL(sv)) {
3346 goto restart; /* call sv_2bool */
3348 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3349 else if(!SvOK(sv)) {
3352 else if(SvPOK(sv)) {
3353 svb = SvPVXtrue(sv);
3355 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3356 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3357 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3361 goto restart; /* call sv_2bool_nomg */
3366 return SvRV(sv) != 0;
3370 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3371 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3375 =for apidoc sv_utf8_upgrade
3377 Converts the PV of an SV to its UTF-8-encoded form.
3378 Forces the SV to string form if it is not already.
3379 Will C<mg_get> on C<sv> if appropriate.
3380 Always sets the C<SvUTF8> flag to avoid future validity checks even
3381 if the whole string is the same in UTF-8 as not.
3382 Returns the number of bytes in the converted string
3384 This is not a general purpose byte encoding to Unicode interface:
3385 use the Encode extension for that.
3387 =for apidoc sv_utf8_upgrade_nomg
3389 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3391 =for apidoc sv_utf8_upgrade_flags
3393 Converts the PV of an SV to its UTF-8-encoded form.
3394 Forces the SV to string form if it is not already.
3395 Always sets the SvUTF8 flag to avoid future validity checks even
3396 if all the bytes are invariant in UTF-8.
3397 If C<flags> has C<SV_GMAGIC> bit set,
3398 will C<mg_get> on C<sv> if appropriate, else not.
3400 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3401 will expand when converted to UTF-8, and skips the extra work of checking for
3402 that. Typically this flag is used by a routine that has already parsed the
3403 string and found such characters, and passes this information on so that the
3404 work doesn't have to be repeated.
3406 Returns the number of bytes in the converted string.
3408 This is not a general purpose byte encoding to Unicode interface:
3409 use the Encode extension for that.
3411 =for apidoc sv_utf8_upgrade_flags_grow
3413 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3414 the number of unused bytes the string of C<sv> is guaranteed to have free after
3415 it upon return. This allows the caller to reserve extra space that it intends
3416 to fill, to avoid extra grows.
3418 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3419 are implemented in terms of this function.
3421 Returns the number of bytes in the converted string (not including the spares).
3425 (One might think that the calling routine could pass in the position of the
3426 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3427 have to be found again. But that is not the case, because typically when the
3428 caller is likely to use this flag, it won't be calling this routine unless it
3429 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3430 and just use bytes. But some things that do fit into a byte are variants in
3431 utf8, and the caller may not have been keeping track of these.)
3433 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3434 C<NUL> isn't guaranteed due to having other routines do the work in some input
3435 cases, or if the input is already flagged as being in utf8.
3437 The speed of this could perhaps be improved for many cases if someone wanted to
3438 write a fast function that counts the number of variant characters in a string,
3439 especially if it could return the position of the first one.
3444 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3446 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3448 if (sv == &PL_sv_undef)
3450 if (!SvPOK_nog(sv)) {
3452 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3453 (void) sv_2pv_flags(sv,&len, flags);
3455 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3459 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3464 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3469 S_sv_uncow(aTHX_ sv, 0);
3472 if (SvCUR(sv) == 0) {
3473 if (extra) SvGROW(sv, extra);
3474 } else { /* Assume Latin-1/EBCDIC */
3475 /* This function could be much more efficient if we
3476 * had a FLAG in SVs to signal if there are any variant
3477 * chars in the PV. Given that there isn't such a flag
3478 * make the loop as fast as possible (although there are certainly ways
3479 * to speed this up, eg. through vectorization) */
3480 U8 * s = (U8 *) SvPVX_const(sv);
3481 U8 * e = (U8 *) SvEND(sv);
3483 STRLEN two_byte_count = 0;
3485 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3487 /* See if really will need to convert to utf8. We mustn't rely on our
3488 * incoming SV being well formed and having a trailing '\0', as certain
3489 * code in pp_formline can send us partially built SVs. */
3493 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3495 t--; /* t already incremented; re-point to first variant */
3500 /* utf8 conversion not needed because all are invariants. Mark as
3501 * UTF-8 even if no variant - saves scanning loop */
3503 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3508 /* Here, the string should be converted to utf8, either because of an
3509 * input flag (two_byte_count = 0), or because a character that
3510 * requires 2 bytes was found (two_byte_count = 1). t points either to
3511 * the beginning of the string (if we didn't examine anything), or to
3512 * the first variant. In either case, everything from s to t - 1 will
3513 * occupy only 1 byte each on output.
3515 * There are two main ways to convert. One is to create a new string
3516 * and go through the input starting from the beginning, appending each
3517 * converted value onto the new string as we go along. It's probably
3518 * best to allocate enough space in the string for the worst possible
3519 * case rather than possibly running out of space and having to
3520 * reallocate and then copy what we've done so far. Since everything
3521 * from s to t - 1 is invariant, the destination can be initialized
3522 * with these using a fast memory copy
3524 * The other way is to figure out exactly how big the string should be
3525 * by parsing the entire input. Then you don't have to make it big
3526 * enough to handle the worst possible case, and more importantly, if
3527 * the string you already have is large enough, you don't have to
3528 * allocate a new string, you can copy the last character in the input
3529 * string to the final position(s) that will be occupied by the
3530 * converted string and go backwards, stopping at t, since everything
3531 * before that is invariant.
3533 * There are advantages and disadvantages to each method.
3535 * In the first method, we can allocate a new string, do the memory
3536 * copy from the s to t - 1, and then proceed through the rest of the
3537 * string byte-by-byte.
3539 * In the second method, we proceed through the rest of the input
3540 * string just calculating how big the converted string will be. Then
3541 * there are two cases:
3542 * 1) if the string has enough extra space to handle the converted
3543 * value. We go backwards through the string, converting until we
3544 * get to the position we are at now, and then stop. If this
3545 * position is far enough along in the string, this method is
3546 * faster than the other method. If the memory copy were the same
3547 * speed as the byte-by-byte loop, that position would be about
3548 * half-way, as at the half-way mark, parsing to the end and back
3549 * is one complete string's parse, the same amount as starting
3550 * over and going all the way through. Actually, it would be
3551 * somewhat less than half-way, as it's faster to just count bytes
3552 * than to also copy, and we don't have the overhead of allocating
3553 * a new string, changing the scalar to use it, and freeing the
3554 * existing one. But if the memory copy is fast, the break-even
3555 * point is somewhere after half way. The counting loop could be
3556 * sped up by vectorization, etc, to move the break-even point
3557 * further towards the beginning.
3558 * 2) if the string doesn't have enough space to handle the converted
3559 * value. A new string will have to be allocated, and one might
3560 * as well, given that, start from the beginning doing the first
3561 * method. We've spent extra time parsing the string and in
3562 * exchange all we've gotten is that we know precisely how big to
3563 * make the new one. Perl is more optimized for time than space,
3564 * so this case is a loser.
3565 * So what I've decided to do is not use the 2nd method unless it is
3566 * guaranteed that a new string won't have to be allocated, assuming
3567 * the worst case. I also decided not to put any more conditions on it
3568 * than this, for now. It seems likely that, since the worst case is
3569 * twice as big as the unknown portion of the string (plus 1), we won't
3570 * be guaranteed enough space, causing us to go to the first method,
3571 * unless the string is short, or the first variant character is near
3572 * the end of it. In either of these cases, it seems best to use the
3573 * 2nd method. The only circumstance I can think of where this would
3574 * be really slower is if the string had once had much more data in it
3575 * than it does now, but there is still a substantial amount in it */
3578 STRLEN invariant_head = t - s;
3579 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3580 if (SvLEN(sv) < size) {
3582 /* Here, have decided to allocate a new string */
3587 Newx(dst, size, U8);
3589 /* If no known invariants at the beginning of the input string,
3590 * set so starts from there. Otherwise, can use memory copy to
3591 * get up to where we are now, and then start from here */
3593 if (invariant_head == 0) {
3596 Copy(s, dst, invariant_head, char);
3597 d = dst + invariant_head;
3601 append_utf8_from_native_byte(*t, &d);
3605 SvPV_free(sv); /* No longer using pre-existing string */
3606 SvPV_set(sv, (char*)dst);
3607 SvCUR_set(sv, d - dst);
3608 SvLEN_set(sv, size);
3611 /* Here, have decided to get the exact size of the string.
3612 * Currently this happens only when we know that there is
3613 * guaranteed enough space to fit the converted string, so
3614 * don't have to worry about growing. If two_byte_count is 0,
3615 * then t points to the first byte of the string which hasn't
3616 * been examined yet. Otherwise two_byte_count is 1, and t
3617 * points to the first byte in the string that will expand to
3618 * two. Depending on this, start examining at t or 1 after t.
3621 U8 *d = t + two_byte_count;
3624 /* Count up the remaining bytes that expand to two */
3627 const U8 chr = *d++;
3628 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3631 /* The string will expand by just the number of bytes that
3632 * occupy two positions. But we are one afterwards because of
3633 * the increment just above. This is the place to put the
3634 * trailing NUL, and to set the length before we decrement */
3636 d += two_byte_count;
3637 SvCUR_set(sv, d - s);
3641 /* Having decremented d, it points to the position to put the
3642 * very last byte of the expanded string. Go backwards through
3643 * the string, copying and expanding as we go, stopping when we
3644 * get to the part that is invariant the rest of the way down */
3648 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3651 *d-- = UTF8_EIGHT_BIT_LO(*e);
3652 *d-- = UTF8_EIGHT_BIT_HI(*e);
3658 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3659 /* Update pos. We do it at the end rather than during
3660 * the upgrade, to avoid slowing down the common case
3661 * (upgrade without pos).
3662 * pos can be stored as either bytes or characters. Since
3663 * this was previously a byte string we can just turn off
3664 * the bytes flag. */
3665 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3667 mg->mg_flags &= ~MGf_BYTES;
3669 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3670 magic_setutf8(sv,mg); /* clear UTF8 cache */
3675 /* Mark as UTF-8 even if no variant - saves scanning loop */
3681 =for apidoc sv_utf8_downgrade
3683 Attempts to convert the PV of an SV from characters to bytes.
3684 If the PV contains a character that cannot fit
3685 in a byte, this conversion will fail;
3686 in this case, either returns false or, if C<fail_ok> is not
3689 This is not a general purpose Unicode to byte encoding interface:
3690 use the C<Encode> extension for that.
3696 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3698 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3700 if (SvPOKp(sv) && SvUTF8(sv)) {
3704 int mg_flags = SV_GMAGIC;
3707 S_sv_uncow(aTHX_ sv, 0);
3709 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3711 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3712 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3713 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3714 SV_GMAGIC|SV_CONST_RETURN);
3715 mg_flags = 0; /* sv_pos_b2u does get magic */
3717 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3718 magic_setutf8(sv,mg); /* clear UTF8 cache */
3721 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3723 if (!utf8_to_bytes(s, &len)) {
3728 Perl_croak(aTHX_ "Wide character in %s",
3731 Perl_croak(aTHX_ "Wide character");
3742 =for apidoc sv_utf8_encode
3744 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3745 flag off so that it looks like octets again.
3751 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3753 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3755 if (SvREADONLY(sv)) {
3756 sv_force_normal_flags(sv, 0);
3758 (void) sv_utf8_upgrade(sv);
3763 =for apidoc sv_utf8_decode
3765 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3766 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3767 so that it looks like a character. If the PV contains only single-byte
3768 characters, the C<SvUTF8> flag stays off.
3769 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3775 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3777 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3780 const U8 *start, *c;
3782 /* The octets may have got themselves encoded - get them back as
3785 if (!sv_utf8_downgrade(sv, TRUE))
3788 /* it is actually just a matter of turning the utf8 flag on, but
3789 * we want to make sure everything inside is valid utf8 first.
3791 c = start = (const U8 *) SvPVX_const(sv);
3792 if (!is_utf8_string(c, SvCUR(sv)))
3794 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3797 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3798 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3799 after this, clearing pos. Does anything on CPAN
3801 /* adjust pos to the start of a UTF8 char sequence */
3802 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3804 I32 pos = mg->mg_len;
3806 for (c = start + pos; c > start; c--) {
3807 if (UTF8_IS_START(*c))
3810 mg->mg_len = c - start;
3813 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3814 magic_setutf8(sv,mg); /* clear UTF8 cache */
3821 =for apidoc sv_setsv
3823 Copies the contents of the source SV C<ssv> into the destination SV
3824 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3825 function if the source SV needs to be reused. Does not handle 'set' magic on
3826 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3827 performs a copy-by-value, obliterating any previous content of the
3830 You probably want to use one of the assortment of wrappers, such as
3831 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3832 C<SvSetMagicSV_nosteal>.
3834 =for apidoc sv_setsv_flags
3836 Copies the contents of the source SV C<ssv> into the destination SV
3837 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3838 function if the source SV needs to be reused. Does not handle 'set' magic.
3839 Loosely speaking, it performs a copy-by-value, obliterating any previous
3840 content of the destination.
3841 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3842 C<ssv> if appropriate, else not. If the C<flags>
3843 parameter has the C<SV_NOSTEAL> bit set then the
3844 buffers of temps will not be stolen. C<sv_setsv>
3845 and C<sv_setsv_nomg> are implemented in terms of this function.
3847 You probably want to use one of the assortment of wrappers, such as
3848 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3849 C<SvSetMagicSV_nosteal>.
3851 This is the primary function for copying scalars, and most other
3852 copy-ish functions and macros use this underneath.
3858 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3860 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3861 HV *old_stash = NULL;
3863 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3865 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3866 const char * const name = GvNAME(sstr);
3867 const STRLEN len = GvNAMELEN(sstr);
3869 if (dtype >= SVt_PV) {
3875 SvUPGRADE(dstr, SVt_PVGV);
3876 (void)SvOK_off(dstr);
3877 isGV_with_GP_on(dstr);
3879 GvSTASH(dstr) = GvSTASH(sstr);
3881 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3882 gv_name_set(MUTABLE_GV(dstr), name, len,
3883 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3884 SvFAKE_on(dstr); /* can coerce to non-glob */
3887 if(GvGP(MUTABLE_GV(sstr))) {
3888 /* If source has method cache entry, clear it */
3890 SvREFCNT_dec(GvCV(sstr));
3891 GvCV_set(sstr, NULL);
3894 /* If source has a real method, then a method is
3897 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3903 /* If dest already had a real method, that's a change as well */
3905 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3906 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3911 /* We don't need to check the name of the destination if it was not a
3912 glob to begin with. */
3913 if(dtype == SVt_PVGV) {
3914 const char * const name = GvNAME((const GV *)dstr);
3917 /* The stash may have been detached from the symbol table, so
3919 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3923 const STRLEN len = GvNAMELEN(dstr);
3924 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3925 || (len == 1 && name[0] == ':')) {
3928 /* Set aside the old stash, so we can reset isa caches on
3930 if((old_stash = GvHV(dstr)))
3931 /* Make sure we do not lose it early. */
3932 SvREFCNT_inc_simple_void_NN(
3933 sv_2mortal((SV *)old_stash)
3938 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3941 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3942 * so temporarily protect it */
3944 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3945 gp_free(MUTABLE_GV(dstr));
3946 GvINTRO_off(dstr); /* one-shot flag */
3947 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3950 if (SvTAINTED(sstr))
3952 if (GvIMPORTED(dstr) != GVf_IMPORTED
3953 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3955 GvIMPORTED_on(dstr);
3958 if(mro_changes == 2) {
3959 if (GvAV((const GV *)sstr)) {
3961 SV * const sref = (SV *)GvAV((const GV *)dstr);
3962 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3963 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3964 AV * const ary = newAV();
3965 av_push(ary, mg->mg_obj); /* takes the refcount */
3966 mg->mg_obj = (SV *)ary;
3968 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3970 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3972 mro_isa_changed_in(GvSTASH(dstr));
3974 else if(mro_changes == 3) {
3975 HV * const stash = GvHV(dstr);
3976 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3982 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3983 if (GvIO(dstr) && dtype == SVt_PVGV) {
3984 DEBUG_o(Perl_deb(aTHX_
3985 "glob_assign_glob clearing PL_stashcache\n"));
3986 /* It's a cache. It will rebuild itself quite happily.
3987 It's a lot of effort to work out exactly which key (or keys)
3988 might be invalidated by the creation of the this file handle.
3990 hv_clear(PL_stashcache);
3996 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3998 SV * const sref = SvRV(sstr);
4000 const int intro = GvINTRO(dstr);
4003 const U32 stype = SvTYPE(sref);
4005 PERL_ARGS_ASSERT_GV_SETREF;
4008 GvINTRO_off(dstr); /* one-shot flag */
4009 GvLINE(dstr) = CopLINE(PL_curcop);
4010 GvEGV(dstr) = MUTABLE_GV(dstr);
4015 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4016 import_flag = GVf_IMPORTED_CV;
4019 location = (SV **) &GvHV(dstr);
4020 import_flag = GVf_IMPORTED_HV;
4023 location = (SV **) &GvAV(dstr);
4024 import_flag = GVf_IMPORTED_AV;
4027 location = (SV **) &GvIOp(dstr);
4030 location = (SV **) &GvFORM(dstr);
4033 location = &GvSV(dstr);
4034 import_flag = GVf_IMPORTED_SV;
4037 if (stype == SVt_PVCV) {
4038 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4039 if (GvCVGEN(dstr)) {
4040 SvREFCNT_dec(GvCV(dstr));
4041 GvCV_set(dstr, NULL);
4042 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4045 /* SAVEt_GVSLOT takes more room on the savestack and has more
4046 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4047 leave_scope needs access to the GV so it can reset method
4048 caches. We must use SAVEt_GVSLOT whenever the type is
4049 SVt_PVCV, even if the stash is anonymous, as the stash may
4050 gain a name somehow before leave_scope. */
4051 if (stype == SVt_PVCV) {
4052 /* There is no save_pushptrptrptr. Creating it for this
4053 one call site would be overkill. So inline the ss add
4057 SS_ADD_PTR(location);
4058 SS_ADD_PTR(SvREFCNT_inc(*location));
4059 SS_ADD_UV(SAVEt_GVSLOT);
4062 else SAVEGENERICSV(*location);
4065 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4066 CV* const cv = MUTABLE_CV(*location);
4068 if (!GvCVGEN((const GV *)dstr) &&
4069 (CvROOT(cv) || CvXSUB(cv)) &&
4070 /* redundant check that avoids creating the extra SV
4071 most of the time: */
4072 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4074 SV * const new_const_sv =
4075 CvCONST((const CV *)sref)
4076 ? cv_const_sv((const CV *)sref)
4078 HV * const stash = GvSTASH((const GV *)dstr);
4079 report_redefined_cv(
4082 ? Perl_newSVpvf(aTHX_
4084 HEKfARG(HvNAME_HEK(stash)),
4085 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4086 : Perl_newSVpvf(aTHX_
4088 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4091 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4095 cv_ckproto_len_flags(cv, (const GV *)dstr,
4096 SvPOK(sref) ? CvPROTO(sref) : NULL,
4097 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4098 SvPOK(sref) ? SvUTF8(sref) : 0);
4100 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4101 GvASSUMECV_on(dstr);
4102 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4103 if (intro && GvREFCNT(dstr) > 1) {
4104 /* temporary remove extra savestack's ref */
4106 gv_method_changed(dstr);
4109 else gv_method_changed(dstr);
4112 *location = SvREFCNT_inc_simple_NN(sref);
4113 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4114 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4115 GvFLAGS(dstr) |= import_flag;
4118 if (stype == SVt_PVHV) {
4119 const char * const name = GvNAME((GV*)dstr);
4120 const STRLEN len = GvNAMELEN(dstr);
4123 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4124 || (len == 1 && name[0] == ':')
4126 && (!dref || HvENAME_get(dref))
4129 (HV *)sref, (HV *)dref,
4135 stype == SVt_PVAV && sref != dref
4136 && strEQ(GvNAME((GV*)dstr), "ISA")
4137 /* The stash may have been detached from the symbol table, so
4138 check its name before doing anything. */
4139 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4142 MAGIC * const omg = dref && SvSMAGICAL(dref)
4143 ? mg_find(dref, PERL_MAGIC_isa)
4145 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4146 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4147 AV * const ary = newAV();
4148 av_push(ary, mg->mg_obj); /* takes the refcount */
4149 mg->mg_obj = (SV *)ary;
4152 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4153 SV **svp = AvARRAY((AV *)omg->mg_obj);
4154 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4158 SvREFCNT_inc_simple_NN(*svp++)
4164 SvREFCNT_inc_simple_NN(omg->mg_obj)
4168 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4174 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4176 for (i = 0; i <= AvFILL(sref); ++i) {
4177 SV **elem = av_fetch ((AV*)sref, i, 0);
4180 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4184 mg = mg_find(sref, PERL_MAGIC_isa);
4186 /* Since the *ISA assignment could have affected more than
4187 one stash, don't call mro_isa_changed_in directly, but let
4188 magic_clearisa do it for us, as it already has the logic for
4189 dealing with globs vs arrays of globs. */
4191 Perl_magic_clearisa(aTHX_ NULL, mg);
4193 else if (stype == SVt_PVIO) {
4194 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4195 /* It's a cache. It will rebuild itself quite happily.
4196 It's a lot of effort to work out exactly which key (or keys)
4197 might be invalidated by the creation of the this file handle.
4199 hv_clear(PL_stashcache);
4203 if (!intro) SvREFCNT_dec(dref);
4204 if (SvTAINTED(sstr))
4212 #ifdef PERL_DEBUG_READONLY_COW
4213 # include <sys/mman.h>
4215 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4216 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4220 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4222 struct perl_memory_debug_header * const header =
4223 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4224 const MEM_SIZE len = header->size;
4225 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4226 # ifdef PERL_TRACK_MEMPOOL
4227 if (!header->readonly) header->readonly = 1;
4229 if (mprotect(header, len, PROT_READ))
4230 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4231 header, len, errno);
4235 S_sv_buf_to_rw(pTHX_ SV *sv)
4237 struct perl_memory_debug_header * const header =
4238 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4239 const MEM_SIZE len = header->size;
4240 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4241 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4242 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4243 header, len, errno);
4244 # ifdef PERL_TRACK_MEMPOOL
4245 header->readonly = 0;
4250 # define sv_buf_to_ro(sv) NOOP
4251 # define sv_buf_to_rw(sv) NOOP
4255 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4260 unsigned int both_type;
4262 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4264 if (UNLIKELY( sstr == dstr ))
4267 if (UNLIKELY( !sstr ))
4268 sstr = &PL_sv_undef;
4270 stype = SvTYPE(sstr);
4271 dtype = SvTYPE(dstr);
4272 both_type = (stype | dtype);
4274 /* with these values, we can check that both SVs are NULL/IV (and not
4275 * freed) just by testing the or'ed types */
4276 STATIC_ASSERT_STMT(SVt_NULL == 0);
4277 STATIC_ASSERT_STMT(SVt_IV == 1);
4278 if (both_type <= 1) {
4279 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4284 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4285 if (SvREADONLY(dstr))
4286 Perl_croak_no_modify();
4288 sv_unref_flags(dstr, 0);
4290 assert(!SvGMAGICAL(sstr));
4291 assert(!SvGMAGICAL(dstr));
4293 sflags = SvFLAGS(sstr);
4294 if (sflags & (SVf_IOK|SVf_ROK)) {
4295 SET_SVANY_FOR_BODYLESS_IV(dstr);
4296 new_dflags = SVt_IV;
4298 if (sflags & SVf_ROK) {
4299 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4300 new_dflags |= SVf_ROK;
4303 /* both src and dst are <= SVt_IV, so sv_any points to the
4304 * head; so access the head directly
4306 assert( &(sstr->sv_u.svu_iv)
4307 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4308 assert( &(dstr->sv_u.svu_iv)
4309 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4310 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4311 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4315 new_dflags = dtype; /* turn off everything except the type */
4317 SvFLAGS(dstr) = new_dflags;
4322 if (UNLIKELY(both_type == SVTYPEMASK)) {
4323 if (SvIS_FREED(dstr)) {
4324 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4325 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4327 if (SvIS_FREED(sstr)) {
4328 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4329 (void*)sstr, (void*)dstr);
4335 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4336 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4338 /* There's a lot of redundancy below but we're going for speed here */
4343 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4344 (void)SvOK_off(dstr);
4352 /* For performance, we inline promoting to type SVt_IV. */
4353 /* We're starting from SVt_NULL, so provided that define is
4354 * actual 0, we don't have to unset any SV type flags
4355 * to promote to SVt_IV. */
4356 STATIC_ASSERT_STMT(SVt_NULL == 0);
4357 SET_SVANY_FOR_BODYLESS_IV(dstr);
4358 SvFLAGS(dstr) |= SVt_IV;
4362 sv_upgrade(dstr, SVt_PVIV);
4366 goto end_of_first_switch;
4368 (void)SvIOK_only(dstr);
4369 SvIV_set(dstr, SvIVX(sstr));
4372 /* SvTAINTED can only be true if the SV has taint magic, which in
4373 turn means that the SV type is PVMG (or greater). This is the
4374 case statement for SVt_IV, so this cannot be true (whatever gcov
4376 assert(!SvTAINTED(sstr));
4381 if (dtype < SVt_PV && dtype != SVt_IV)
4382 sv_upgrade(dstr, SVt_IV);
4386 if (LIKELY( SvNOK(sstr) )) {
4390 sv_upgrade(dstr, SVt_NV);
4394 sv_upgrade(dstr, SVt_PVNV);
4398 goto end_of_first_switch;
4400 SvNV_set(dstr, SvNVX(sstr));
4401 (void)SvNOK_only(dstr);
4402 /* SvTAINTED can only be true if the SV has taint magic, which in
4403 turn means that the SV type is PVMG (or greater). This is the
4404 case statement for SVt_NV, so this cannot be true (whatever gcov
4406 assert(!SvTAINTED(sstr));
4413 sv_upgrade(dstr, SVt_PV);
4416 if (dtype < SVt_PVIV)
4417 sv_upgrade(dstr, SVt_PVIV);
4420 if (dtype < SVt_PVNV)
4421 sv_upgrade(dstr, SVt_PVNV);
4425 const char * const type = sv_reftype(sstr,0);
4427 /* diag_listed_as: Bizarre copy of %s */
4428 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4430 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4432 NOT_REACHED; /* NOTREACHED */
4436 if (dtype < SVt_REGEXP)
4438 if (dtype >= SVt_PV) {
4444 sv_upgrade(dstr, SVt_REGEXP);
4452 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4454 if (SvTYPE(sstr) != stype)
4455 stype = SvTYPE(sstr);
4457 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4458 glob_assign_glob(dstr, sstr, dtype);
4461 if (stype == SVt_PVLV)
4463 if (isREGEXP(sstr)) goto upgregexp;
4464 SvUPGRADE(dstr, SVt_PVNV);
4467 SvUPGRADE(dstr, (svtype)stype);
4469 end_of_first_switch:
4471 /* dstr may have been upgraded. */
4472 dtype = SvTYPE(dstr);
4473 sflags = SvFLAGS(sstr);
4475 if (UNLIKELY( dtype == SVt_PVCV )) {
4476 /* Assigning to a subroutine sets the prototype. */
4479 const char *const ptr = SvPV_const(sstr, len);
4481 SvGROW(dstr, len + 1);
4482 Copy(ptr, SvPVX(dstr), len + 1, char);
4483 SvCUR_set(dstr, len);
4485 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4486 CvAUTOLOAD_off(dstr);
4491 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4492 || dtype == SVt_PVFM))
4494 const char * const type = sv_reftype(dstr,0);
4496 /* diag_listed_as: Cannot copy to %s */
4497 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4499 Perl_croak(aTHX_ "Cannot copy to %s", type);
4500 } else if (sflags & SVf_ROK) {
4501 if (isGV_with_GP(dstr)
4502 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4505 if (GvIMPORTED(dstr) != GVf_IMPORTED
4506 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4508 GvIMPORTED_on(dstr);
4513 glob_assign_glob(dstr, sstr, dtype);
4517 if (dtype >= SVt_PV) {
4518 if (isGV_with_GP(dstr)) {
4519 gv_setref(dstr, sstr);
4522 if (SvPVX_const(dstr)) {
4528 (void)SvOK_off(dstr);
4529 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4530 SvFLAGS(dstr) |= sflags & SVf_ROK;
4531 assert(!(sflags & SVp_NOK));
4532 assert(!(sflags & SVp_IOK));
4533 assert(!(sflags & SVf_NOK));
4534 assert(!(sflags & SVf_IOK));
4536 else if (isGV_with_GP(dstr)) {
4537 if (!(sflags & SVf_OK)) {
4538 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4539 "Undefined value assigned to typeglob");
4542 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4543 if (dstr != (const SV *)gv) {
4544 const char * const name = GvNAME((const GV *)dstr);
4545 const STRLEN len = GvNAMELEN(dstr);
4546 HV *old_stash = NULL;
4547 bool reset_isa = FALSE;
4548 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4549 || (len == 1 && name[0] == ':')) {
4550 /* Set aside the old stash, so we can reset isa caches
4551 on its subclasses. */
4552 if((old_stash = GvHV(dstr))) {
4553 /* Make sure we do not lose it early. */
4554 SvREFCNT_inc_simple_void_NN(
4555 sv_2mortal((SV *)old_stash)
4562 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4563 gp_free(MUTABLE_GV(dstr));
4565 GvGP_set(dstr, gp_ref(GvGP(gv)));
4568 HV * const stash = GvHV(dstr);
4570 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4580 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4581 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4582 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4584 else if (sflags & SVp_POK) {
4585 const STRLEN cur = SvCUR(sstr);
4586 const STRLEN len = SvLEN(sstr);
4589 * We have three basic ways to copy the string:
4595 * Which we choose is based on various factors. The following
4596 * things are listed in order of speed, fastest to slowest:
4598 * - Copying a short string
4599 * - Copy-on-write bookkeeping
4601 * - Copying a long string
4603 * We swipe the string (steal the string buffer) if the SV on the
4604 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4605 * big win on long strings. It should be a win on short strings if
4606 * SvPVX_const(dstr) has to be allocated. If not, it should not
4607 * slow things down, as SvPVX_const(sstr) would have been freed
4610 * We also steal the buffer from a PADTMP (operator target) if it
4611 * is ‘long enough’. For short strings, a swipe does not help
4612 * here, as it causes more malloc calls the next time the target
4613 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4614 * be allocated it is still not worth swiping PADTMPs for short
4615 * strings, as the savings here are small.
4617 * If swiping is not an option, then we see whether it is
4618 * worth using copy-on-write. If the lhs already has a buf-
4619 * fer big enough and the string is short, we skip it and fall back
4620 * to method 3, since memcpy is faster for short strings than the
4621 * later bookkeeping overhead that copy-on-write entails.
4623 * If the rhs is not a copy-on-write string yet, then we also
4624 * consider whether the buffer is too large relative to the string
4625 * it holds. Some operations such as readline allocate a large
4626 * buffer in the expectation of reusing it. But turning such into
4627 * a COW buffer is counter-productive because it increases memory
4628 * usage by making readline allocate a new large buffer the sec-
4629 * ond time round. So, if the buffer is too large, again, we use
4632 * Finally, if there is no buffer on the left, or the buffer is too
4633 * small, then we use copy-on-write and make both SVs share the
4638 /* Whichever path we take through the next code, we want this true,
4639 and doing it now facilitates the COW check. */
4640 (void)SvPOK_only(dstr);
4644 /* slated for free anyway (and not COW)? */
4645 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4646 /* or a swipable TARG */
4648 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4650 /* whose buffer is worth stealing */
4651 && CHECK_COWBUF_THRESHOLD(cur,len)
4654 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4655 (!(flags & SV_NOSTEAL)) &&
4656 /* and we're allowed to steal temps */
4657 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4658 len) /* and really is a string */
4659 { /* Passes the swipe test. */
4660 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4662 SvPV_set(dstr, SvPVX_mutable(sstr));
4663 SvLEN_set(dstr, SvLEN(sstr));
4664 SvCUR_set(dstr, SvCUR(sstr));
4667 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4668 SvPV_set(sstr, NULL);
4673 else if (flags & SV_COW_SHARED_HASH_KEYS
4675 #ifdef PERL_COPY_ON_WRITE
4678 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4679 /* If this is a regular (non-hek) COW, only so
4680 many COW "copies" are possible. */
4681 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4682 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4683 && !(SvFLAGS(dstr) & SVf_BREAK)
4684 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4685 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4689 && !(SvFLAGS(dstr) & SVf_BREAK)
4692 /* Either it's a shared hash key, or it's suitable for
4695 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4700 if (!(sflags & SVf_IsCOW)) {
4702 CowREFCNT(sstr) = 0;
4705 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4711 if (sflags & SVf_IsCOW) {
4715 SvPV_set(dstr, SvPVX_mutable(sstr));
4720 /* SvIsCOW_shared_hash */
4721 DEBUG_C(PerlIO_printf(Perl_debug_log,
4722 "Copy on write: Sharing hash\n"));
4724 assert (SvTYPE(dstr) >= SVt_PV);
4726 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4728 SvLEN_set(dstr, len);
4729 SvCUR_set(dstr, cur);
4732 /* Failed the swipe test, and we cannot do copy-on-write either.
4733 Have to copy the string. */
4734 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4735 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4736 SvCUR_set(dstr, cur);
4737 *SvEND(dstr) = '\0';
4739 if (sflags & SVp_NOK) {
4740 SvNV_set(dstr, SvNVX(sstr));
4742 if (sflags & SVp_IOK) {
4743 SvIV_set(dstr, SvIVX(sstr));
4744 /* Must do this otherwise some other overloaded use of 0x80000000
4745 gets confused. I guess SVpbm_VALID */
4746 if (sflags & SVf_IVisUV)
4749 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4751 const MAGIC * const smg = SvVSTRING_mg(sstr);
4753 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4754 smg->mg_ptr, smg->mg_len);
4755 SvRMAGICAL_on(dstr);
4759 else if (sflags & (SVp_IOK|SVp_NOK)) {
4760 (void)SvOK_off(dstr);
4761 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4762 if (sflags & SVp_IOK) {
4763 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4764 SvIV_set(dstr, SvIVX(sstr));
4766 if (sflags & SVp_NOK) {
4767 SvNV_set(dstr, SvNVX(sstr));
4771 if (isGV_with_GP(sstr)) {
4772 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4775 (void)SvOK_off(dstr);
4777 if (SvTAINTED(sstr))
4782 =for apidoc sv_setsv_mg
4784 Like C<sv_setsv>, but also handles 'set' magic.
4790 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4792 PERL_ARGS_ASSERT_SV_SETSV_MG;
4794 sv_setsv(dstr,sstr);
4799 # define SVt_COW SVt_PV
4801 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4803 STRLEN cur = SvCUR(sstr);
4804 STRLEN len = SvLEN(sstr);
4806 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4807 const bool already = cBOOL(SvIsCOW(sstr));
4810 PERL_ARGS_ASSERT_SV_SETSV_COW;
4813 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4814 (void*)sstr, (void*)dstr);
4821 if (SvTHINKFIRST(dstr))
4822 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4823 else if (SvPVX_const(dstr))
4824 Safefree(SvPVX_mutable(dstr));
4828 SvUPGRADE(dstr, SVt_COW);
4830 assert (SvPOK(sstr));
4831 assert (SvPOKp(sstr));
4833 if (SvIsCOW(sstr)) {
4835 if (SvLEN(sstr) == 0) {
4836 /* source is a COW shared hash key. */
4837 DEBUG_C(PerlIO_printf(Perl_debug_log,
4838 "Fast copy on write: Sharing hash\n"));
4839 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4842 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4843 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4845 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4846 SvUPGRADE(sstr, SVt_COW);
4848 DEBUG_C(PerlIO_printf(Perl_debug_log,
4849 "Fast copy on write: Converting sstr to COW\n"));
4850 CowREFCNT(sstr) = 0;
4852 # ifdef PERL_DEBUG_READONLY_COW
4853 if (already) sv_buf_to_rw(sstr);
4856 new_pv = SvPVX_mutable(sstr);
4860 SvPV_set(dstr, new_pv);
4861 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4864 SvLEN_set(dstr, len);
4865 SvCUR_set(dstr, cur);
4874 =for apidoc sv_setpv_bufsize
4876 Sets the SV to be a string of cur bytes length, with at least
4877 len bytes available. Ensures that there is a null byte at SvEND.
4878 Returns a char * pointer to the SvPV buffer.
4884 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4888 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4890 SV_CHECK_THINKFIRST_COW_DROP(sv);
4891 SvUPGRADE(sv, SVt_PV);
4892 pv = SvGROW(sv, len + 1);
4895 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4898 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4903 =for apidoc sv_setpvn
4905 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4906 The C<len> parameter indicates the number of
4907 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4908 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4914 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4918 PERL_ARGS_ASSERT_SV_SETPVN;
4920 SV_CHECK_THINKFIRST_COW_DROP(sv);
4926 /* len is STRLEN which is unsigned, need to copy to signed */
4929 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4932 SvUPGRADE(sv, SVt_PV);
4934 dptr = SvGROW(sv, len + 1);
4935 Move(ptr,dptr,len,char);
4938 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4940 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4944 =for apidoc sv_setpvn_mg
4946 Like C<sv_setpvn>, but also handles 'set' magic.
4952 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4954 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4956 sv_setpvn(sv,ptr,len);
4961 =for apidoc sv_setpv
4963 Copies a string into an SV. The string must be terminated with a C<NUL>
4964 character, and not contain embeded C<NUL>'s.
4965 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4971 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4975 PERL_ARGS_ASSERT_SV_SETPV;
4977 SV_CHECK_THINKFIRST_COW_DROP(sv);
4983 SvUPGRADE(sv, SVt_PV);
4985 SvGROW(sv, len + 1);
4986 Move(ptr,SvPVX(sv),len+1,char);
4988 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4990 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4994 =for apidoc sv_setpv_mg
4996 Like C<sv_setpv>, but also handles 'set' magic.
5002 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5004 PERL_ARGS_ASSERT_SV_SETPV_MG;
5011 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5013 PERL_ARGS_ASSERT_SV_SETHEK;
5019 if (HEK_LEN(hek) == HEf_SVKEY) {
5020 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5023 const int flags = HEK_FLAGS(hek);
5024 if (flags & HVhek_WASUTF8) {
5025 STRLEN utf8_len = HEK_LEN(hek);
5026 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5027 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5030 } else if (flags & HVhek_UNSHARED) {
5031 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5034 else SvUTF8_off(sv);
5038 SV_CHECK_THINKFIRST_COW_DROP(sv);
5039 SvUPGRADE(sv, SVt_PV);
5041 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5042 SvCUR_set(sv, HEK_LEN(hek));
5048 else SvUTF8_off(sv);
5056 =for apidoc sv_usepvn_flags
5058 Tells an SV to use C<ptr> to find its string value. Normally the
5059 string is stored inside the SV, but sv_usepvn allows the SV to use an
5060 outside string. C<ptr> should point to memory that was allocated
5061 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5062 the start of a C<Newx>-ed block of memory, and not a pointer to the
5063 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5064 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5065 string length, C<len>, must be supplied. By default this function
5066 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5067 so that pointer should not be freed or used by the programmer after
5068 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5069 that pointer (e.g. ptr + 1) be used.
5071 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5072 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5074 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5075 C<len>, and already meets the requirements for storing in C<SvPVX>).
5081 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5085 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5087 SV_CHECK_THINKFIRST_COW_DROP(sv);
5088 SvUPGRADE(sv, SVt_PV);
5091 if (flags & SV_SMAGIC)
5095 if (SvPVX_const(sv))
5099 if (flags & SV_HAS_TRAILING_NUL)
5100 assert(ptr[len] == '\0');
5103 allocate = (flags & SV_HAS_TRAILING_NUL)
5105 #ifdef Perl_safesysmalloc_size
5108 PERL_STRLEN_ROUNDUP(len + 1);
5110 if (flags & SV_HAS_TRAILING_NUL) {
5111 /* It's long enough - do nothing.
5112 Specifically Perl_newCONSTSUB is relying on this. */
5115 /* Force a move to shake out bugs in callers. */
5116 char *new_ptr = (char*)safemalloc(allocate);
5117 Copy(ptr, new_ptr, len, char);
5118 PoisonFree(ptr,len,char);
5122 ptr = (char*) saferealloc (ptr, allocate);
5125 #ifdef Perl_safesysmalloc_size
5126 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5128 SvLEN_set(sv, allocate);
5132 if (!(flags & SV_HAS_TRAILING_NUL)) {
5135 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5137 if (flags & SV_SMAGIC)
5142 =for apidoc sv_force_normal_flags
5144 Undo various types of fakery on an SV, where fakery means
5145 "more than" a string: if the PV is a shared string, make
5146 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5147 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5148 we do the copy, and is also used locally; if this is a
5149 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5150 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5151 C<SvPOK_off> rather than making a copy. (Used where this
5152 scalar is about to be set to some other value.) In addition,
5153 the C<flags> parameter gets passed to C<sv_unref_flags()>
5154 when unreffing. C<sv_force_normal> calls this function
5155 with flags set to 0.
5157 This function is expected to be used to signal to perl that this SV is
5158 about to be written to, and any extra book-keeping needs to be taken care
5159 of. Hence, it croaks on read-only values.
5165 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5167 assert(SvIsCOW(sv));
5170 const char * const pvx = SvPVX_const(sv);
5171 const STRLEN len = SvLEN(sv);
5172 const STRLEN cur = SvCUR(sv);
5175 PerlIO_printf(Perl_debug_log,
5176 "Copy on write: Force normal %ld\n",
5181 # ifdef PERL_COPY_ON_WRITE
5183 /* Must do this first, since the CowREFCNT uses SvPVX and
5184 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5185 the only owner left of the buffer. */
5186 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5188 U8 cowrefcnt = CowREFCNT(sv);
5189 if(cowrefcnt != 0) {
5191 CowREFCNT(sv) = cowrefcnt;
5196 /* Else we are the only owner of the buffer. */
5201 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5206 if (flags & SV_COW_DROP_PV) {
5207 /* OK, so we don't need to copy our buffer. */
5210 SvGROW(sv, cur + 1);
5211 Move(pvx,SvPVX(sv),cur,char);
5217 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5224 const char * const pvx = SvPVX_const(sv);
5225 const STRLEN len = SvCUR(sv);
5229 if (flags & SV_COW_DROP_PV) {
5230 /* OK, so we don't need to copy our buffer. */
5233 SvGROW(sv, len + 1);
5234 Move(pvx,SvPVX(sv),len,char);
5237 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5243 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5245 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5248 Perl_croak_no_modify();
5249 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5250 S_sv_uncow(aTHX_ sv, flags);
5252 sv_unref_flags(sv, flags);
5253 else if (SvFAKE(sv) && isGV_with_GP(sv))
5254 sv_unglob(sv, flags);
5255 else if (SvFAKE(sv) && isREGEXP(sv)) {
5256 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5257 to sv_unglob. We only need it here, so inline it. */
5258 const bool islv = SvTYPE(sv) == SVt_PVLV;
5259 const svtype new_type =
5260 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5261 SV *const temp = newSV_type(new_type);
5262 regexp *const temp_p = ReANY((REGEXP *)sv);
5264 if (new_type == SVt_PVMG) {
5265 SvMAGIC_set(temp, SvMAGIC(sv));
5266 SvMAGIC_set(sv, NULL);
5267 SvSTASH_set(temp, SvSTASH(sv));
5268 SvSTASH_set(sv, NULL);
5270 if (!islv) SvCUR_set(temp, SvCUR(sv));
5271 /* Remember that SvPVX is in the head, not the body. But
5272 RX_WRAPPED is in the body. */
5273 assert(ReANY((REGEXP *)sv)->mother_re);
5274 /* Their buffer is already owned by someone else. */
5275 if (flags & SV_COW_DROP_PV) {
5276 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5277 zeroed body. For SVt_PVLV, it should have been set to 0
5278 before turning into a regexp. */
5279 assert(!SvLEN(islv ? sv : temp));
5280 sv->sv_u.svu_pv = 0;
5283 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5284 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5288 /* Now swap the rest of the bodies. */
5292 SvFLAGS(sv) &= ~SVTYPEMASK;
5293 SvFLAGS(sv) |= new_type;
5294 SvANY(sv) = SvANY(temp);
5297 SvFLAGS(temp) &= ~(SVTYPEMASK);
5298 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5299 SvANY(temp) = temp_p;
5300 temp->sv_u.svu_rx = (regexp *)temp_p;
5302 SvREFCNT_dec_NN(temp);
5304 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5310 Efficient removal of characters from the beginning of the string buffer.
5311 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5312 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5313 character of the adjusted string. Uses the C<OOK> hack. On return, only
5314 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5316 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5317 refer to the same chunk of data.
5319 The unfortunate similarity of this function's name to that of Perl's C<chop>
5320 operator is strictly coincidental. This function works from the left;
5321 C<chop> works from the right.
5327 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5338 PERL_ARGS_ASSERT_SV_CHOP;
5340 if (!ptr || !SvPOKp(sv))
5342 delta = ptr - SvPVX_const(sv);
5344 /* Nothing to do. */
5347 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5348 if (delta > max_delta)
5349 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5350 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5351 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5352 SV_CHECK_THINKFIRST(sv);
5353 SvPOK_only_UTF8(sv);
5356 if (!SvLEN(sv)) { /* make copy of shared string */
5357 const char *pvx = SvPVX_const(sv);
5358 const STRLEN len = SvCUR(sv);
5359 SvGROW(sv, len + 1);
5360 Move(pvx,SvPVX(sv),len,char);
5366 SvOOK_offset(sv, old_delta);
5368 SvLEN_set(sv, SvLEN(sv) - delta);
5369 SvCUR_set(sv, SvCUR(sv) - delta);
5370 SvPV_set(sv, SvPVX(sv) + delta);
5372 p = (U8 *)SvPVX_const(sv);
5375 /* how many bytes were evacuated? we will fill them with sentinel
5376 bytes, except for the part holding the new offset of course. */
5379 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5381 assert(evacn <= delta + old_delta);
5385 /* This sets 'delta' to the accumulated value of all deltas so far */
5389 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5390 * the string; otherwise store a 0 byte there and store 'delta' just prior
5391 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5392 * portion of the chopped part of the string */
5393 if (delta < 0x100) {
5397 p -= sizeof(STRLEN);
5398 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5402 /* Fill the preceding buffer with sentinals to verify that no-one is
5412 =for apidoc sv_catpvn
5414 Concatenates the string onto the end of the string which is in the SV.
5415 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5416 status set, then the bytes appended should be valid UTF-8.
5417 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5419 =for apidoc sv_catpvn_flags
5421 Concatenates the string onto the end of the string which is in the SV. The
5422 C<len> indicates number of bytes to copy.
5424 By default, the string appended is assumed to be valid UTF-8 if the SV has
5425 the UTF-8 status set, and a string of bytes otherwise. One can force the
5426 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5427 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5428 string appended will be upgraded to UTF-8 if necessary.
5430 If C<flags> has the C<SV_SMAGIC> bit set, will
5431 C<mg_set> on C<dsv> afterwards if appropriate.
5432 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5433 in terms of this function.
5439 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5442 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5444 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5445 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5447 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5448 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5449 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5452 else SvGROW(dsv, dlen + slen + 3);
5454 sstr = SvPVX_const(dsv);
5455 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5456 SvCUR_set(dsv, SvCUR(dsv) + slen);
5459 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5460 const char * const send = sstr + slen;
5463 /* Something this code does not account for, which I think is
5464 impossible; it would require the same pv to be treated as
5465 bytes *and* utf8, which would indicate a bug elsewhere. */
5466 assert(sstr != dstr);
5468 SvGROW(dsv, dlen + slen * 2 + 3);
5469 d = (U8 *)SvPVX(dsv) + dlen;
5471 while (sstr < send) {
5472 append_utf8_from_native_byte(*sstr, &d);
5475 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5478 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5480 if (flags & SV_SMAGIC)
5485 =for apidoc sv_catsv
5487 Concatenates the string from SV C<ssv> onto the end of the string in SV
5488 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5489 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5490 and C<L</sv_catsv_nomg>>.
5492 =for apidoc sv_catsv_flags
5494 Concatenates the string from SV C<ssv> onto the end of the string in SV
5495 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5496 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5497 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5498 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5499 and C<sv_catsv_mg> are implemented in terms of this function.
5504 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5506 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5510 const char *spv = SvPV_flags_const(ssv, slen, flags);
5511 if (flags & SV_GMAGIC)
5513 sv_catpvn_flags(dsv, spv, slen,
5514 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5515 if (flags & SV_SMAGIC)
5521 =for apidoc sv_catpv
5523 Concatenates the C<NUL>-terminated string onto the end of the string which is
5525 If the SV has the UTF-8 status set, then the bytes appended should be
5526 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5532 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5538 PERL_ARGS_ASSERT_SV_CATPV;
5542 junk = SvPV_force(sv, tlen);
5544 SvGROW(sv, tlen + len + 1);
5546 ptr = SvPVX_const(sv);
5547 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5548 SvCUR_set(sv, SvCUR(sv) + len);
5549 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5554 =for apidoc sv_catpv_flags
5556 Concatenates the C<NUL>-terminated string onto the end of the string which is
5558 If the SV has the UTF-8 status set, then the bytes appended should
5559 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5560 on the modified SV if appropriate.
5566 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5568 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5569 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5573 =for apidoc sv_catpv_mg
5575 Like C<sv_catpv>, but also handles 'set' magic.
5581 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5583 PERL_ARGS_ASSERT_SV_CATPV_MG;
5592 Creates a new SV. A non-zero C<len> parameter indicates the number of
5593 bytes of preallocated string space the SV should have. An extra byte for a
5594 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5595 space is allocated.) The reference count for the new SV is set to 1.
5597 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5598 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5599 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5600 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5601 modules supporting older perls.
5607 Perl_newSV(pTHX_ const STRLEN len)
5613 sv_grow(sv, len + 1);
5618 =for apidoc sv_magicext
5620 Adds magic to an SV, upgrading it if necessary. Applies the
5621 supplied C<vtable> and returns a pointer to the magic added.
5623 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5624 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5625 one instance of the same C<how>.
5627 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5628 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5629 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5630 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5632 (This is now used as a subroutine by C<sv_magic>.)
5637 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5638 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5642 PERL_ARGS_ASSERT_SV_MAGICEXT;
5644 SvUPGRADE(sv, SVt_PVMG);
5645 Newxz(mg, 1, MAGIC);
5646 mg->mg_moremagic = SvMAGIC(sv);
5647 SvMAGIC_set(sv, mg);
5649 /* Sometimes a magic contains a reference loop, where the sv and
5650 object refer to each other. To prevent a reference loop that
5651 would prevent such objects being freed, we look for such loops
5652 and if we find one we avoid incrementing the object refcount.
5654 Note we cannot do this to avoid self-tie loops as intervening RV must
5655 have its REFCNT incremented to keep it in existence.
5658 if (!obj || obj == sv ||
5659 how == PERL_MAGIC_arylen ||
5660 how == PERL_MAGIC_symtab ||
5661 (SvTYPE(obj) == SVt_PVGV &&
5662 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5663 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5664 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5669 mg->mg_obj = SvREFCNT_inc_simple(obj);
5670 mg->mg_flags |= MGf_REFCOUNTED;
5673 /* Normal self-ties simply pass a null object, and instead of
5674 using mg_obj directly, use the SvTIED_obj macro to produce a
5675 new RV as needed. For glob "self-ties", we are tieing the PVIO
5676 with an RV obj pointing to the glob containing the PVIO. In
5677 this case, to avoid a reference loop, we need to weaken the
5681 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5682 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5688 mg->mg_len = namlen;
5691 mg->mg_ptr = savepvn(name, namlen);
5692 else if (namlen == HEf_SVKEY) {
5693 /* Yes, this is casting away const. This is only for the case of
5694 HEf_SVKEY. I think we need to document this aberation of the
5695 constness of the API, rather than making name non-const, as
5696 that change propagating outwards a long way. */
5697 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5699 mg->mg_ptr = (char *) name;
5701 mg->mg_virtual = (MGVTBL *) vtable;
5708 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5710 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5711 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5712 /* This sv is only a delegate. //g magic must be attached to
5717 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5718 &PL_vtbl_mglob, 0, 0);
5722 =for apidoc sv_magic
5724 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5725 necessary, then adds a new magic item of type C<how> to the head of the
5728 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5729 handling of the C<name> and C<namlen> arguments.
5731 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5732 to add more than one instance of the same C<how>.
5738 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5739 const char *const name, const I32 namlen)
5741 const MGVTBL *vtable;
5744 unsigned int vtable_index;
5746 PERL_ARGS_ASSERT_SV_MAGIC;
5748 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5749 || ((flags = PL_magic_data[how]),
5750 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5751 > magic_vtable_max))
5752 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5754 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5755 Useful for attaching extension internal data to perl vars.
5756 Note that multiple extensions may clash if magical scalars
5757 etc holding private data from one are passed to another. */
5759 vtable = (vtable_index == magic_vtable_max)
5760 ? NULL : PL_magic_vtables + vtable_index;
5762 if (SvREADONLY(sv)) {
5764 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5767 Perl_croak_no_modify();
5770 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5771 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5772 /* sv_magic() refuses to add a magic of the same 'how' as an
5775 if (how == PERL_MAGIC_taint)
5781 /* Force pos to be stored as characters, not bytes. */
5782 if (SvMAGICAL(sv) && DO_UTF8(sv)
5783 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5785 && mg->mg_flags & MGf_BYTES) {
5786 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5788 mg->mg_flags &= ~MGf_BYTES;
5791 /* Rest of work is done else where */
5792 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5795 case PERL_MAGIC_taint:
5798 case PERL_MAGIC_ext:
5799 case PERL_MAGIC_dbfile:
5806 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5813 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5815 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5816 for (mg = *mgp; mg; mg = *mgp) {
5817 const MGVTBL* const virt = mg->mg_virtual;
5818 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5819 *mgp = mg->mg_moremagic;
5820 if (virt && virt->svt_free)
5821 virt->svt_free(aTHX_ sv, mg);
5822 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5824 Safefree(mg->mg_ptr);
5825 else if (mg->mg_len == HEf_SVKEY)
5826 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5827 else if (mg->mg_type == PERL_MAGIC_utf8)
5828 Safefree(mg->mg_ptr);
5830 if (mg->mg_flags & MGf_REFCOUNTED)
5831 SvREFCNT_dec(mg->mg_obj);
5835 mgp = &mg->mg_moremagic;
5838 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5839 mg_magical(sv); /* else fix the flags now */
5848 =for apidoc sv_unmagic
5850 Removes all magic of type C<type> from an SV.
5856 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5858 PERL_ARGS_ASSERT_SV_UNMAGIC;
5859 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5863 =for apidoc sv_unmagicext
5865 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5871 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5873 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5874 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5878 =for apidoc sv_rvweaken
5880 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5881 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5882 push a back-reference to this RV onto the array of backreferences
5883 associated with that magic. If the RV is magical, set magic will be
5884 called after the RV is cleared.
5890 Perl_sv_rvweaken(pTHX_ SV *const sv)
5894 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5896 if (!SvOK(sv)) /* let undefs pass */
5899 Perl_croak(aTHX_ "Can't weaken a nonreference");
5900 else if (SvWEAKREF(sv)) {
5901 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5904 else if (SvREADONLY(sv)) croak_no_modify();
5906 Perl_sv_add_backref(aTHX_ tsv, sv);
5908 SvREFCNT_dec_NN(tsv);
5913 =for apidoc sv_get_backrefs
5915 If C<sv> is the target of a weak reference then it returns the back
5916 references structure associated with the sv; otherwise return C<NULL>.
5918 When returning a non-null result the type of the return is relevant. If it
5919 is an AV then the elements of the AV are the weak reference RVs which
5920 point at this item. If it is any other type then the item itself is the
5923 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5924 C<Perl_sv_kill_backrefs()>
5930 Perl_sv_get_backrefs(SV *const sv)
5934 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5936 /* find slot to store array or singleton backref */
5938 if (SvTYPE(sv) == SVt_PVHV) {
5940 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5941 backrefs = (SV *)iter->xhv_backreferences;
5943 } else if (SvMAGICAL(sv)) {
5944 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5946 backrefs = mg->mg_obj;
5951 /* Give tsv backref magic if it hasn't already got it, then push a
5952 * back-reference to sv onto the array associated with the backref magic.
5954 * As an optimisation, if there's only one backref and it's not an AV,
5955 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5956 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5960 /* A discussion about the backreferences array and its refcount:
5962 * The AV holding the backreferences is pointed to either as the mg_obj of
5963 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5964 * xhv_backreferences field. The array is created with a refcount
5965 * of 2. This means that if during global destruction the array gets
5966 * picked on before its parent to have its refcount decremented by the
5967 * random zapper, it won't actually be freed, meaning it's still there for
5968 * when its parent gets freed.
5970 * When the parent SV is freed, the extra ref is killed by
5971 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5972 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5974 * When a single backref SV is stored directly, it is not reference
5979 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5985 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5987 /* find slot to store array or singleton backref */
5989 if (SvTYPE(tsv) == SVt_PVHV) {
5990 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5993 mg = mg_find(tsv, PERL_MAGIC_backref);
5995 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5996 svp = &(mg->mg_obj);
5999 /* create or retrieve the array */
6001 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6002 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6006 mg->mg_flags |= MGf_REFCOUNTED;
6009 SvREFCNT_inc_simple_void_NN(av);
6010 /* av now has a refcnt of 2; see discussion above */
6011 av_extend(av, *svp ? 2 : 1);
6013 /* move single existing backref to the array */
6014 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6019 av = MUTABLE_AV(*svp);
6021 /* optimisation: store single backref directly in HvAUX or mg_obj */
6025 assert(SvTYPE(av) == SVt_PVAV);
6026 if (AvFILLp(av) >= AvMAX(av)) {
6027 av_extend(av, AvFILLp(av)+1);
6030 /* push new backref */
6031 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6034 /* delete a back-reference to ourselves from the backref magic associated
6035 * with the SV we point to.
6039 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6043 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6045 if (SvTYPE(tsv) == SVt_PVHV) {
6047 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6049 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6050 /* It's possible for the the last (strong) reference to tsv to have
6051 become freed *before* the last thing holding a weak reference.
6052 If both survive longer than the backreferences array, then when
6053 the referent's reference count drops to 0 and it is freed, it's
6054 not able to chase the backreferences, so they aren't NULLed.
6056 For example, a CV holds a weak reference to its stash. If both the
6057 CV and the stash survive longer than the backreferences array,
6058 and the CV gets picked for the SvBREAK() treatment first,
6059 *and* it turns out that the stash is only being kept alive because
6060 of an our variable in the pad of the CV, then midway during CV
6061 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6062 It ends up pointing to the freed HV. Hence it's chased in here, and
6063 if this block wasn't here, it would hit the !svp panic just below.
6065 I don't believe that "better" destruction ordering is going to help
6066 here - during global destruction there's always going to be the
6067 chance that something goes out of order. We've tried to make it
6068 foolproof before, and it only resulted in evolutionary pressure on
6069 fools. Which made us look foolish for our hubris. :-(
6075 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6076 svp = mg ? &(mg->mg_obj) : NULL;
6080 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6082 /* It's possible that sv is being freed recursively part way through the
6083 freeing of tsv. If this happens, the backreferences array of tsv has
6084 already been freed, and so svp will be NULL. If this is the case,
6085 we should not panic. Instead, nothing needs doing, so return. */
6086 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6088 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6089 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6092 if (SvTYPE(*svp) == SVt_PVAV) {
6096 AV * const av = (AV*)*svp;
6098 assert(!SvIS_FREED(av));
6102 /* for an SV with N weak references to it, if all those
6103 * weak refs are deleted, then sv_del_backref will be called
6104 * N times and O(N^2) compares will be done within the backref
6105 * array. To ameliorate this potential slowness, we:
6106 * 1) make sure this code is as tight as possible;
6107 * 2) when looking for SV, look for it at both the head and tail of the
6108 * array first before searching the rest, since some create/destroy
6109 * patterns will cause the backrefs to be freed in order.
6116 SV **p = &svp[fill];
6117 SV *const topsv = *p;
6124 /* We weren't the last entry.
6125 An unordered list has this property that you
6126 can take the last element off the end to fill
6127 the hole, and it's still an unordered list :-)
6133 break; /* should only be one */
6140 AvFILLp(av) = fill-1;
6142 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6143 /* freed AV; skip */
6146 /* optimisation: only a single backref, stored directly */
6148 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6149 (void*)*svp, (void*)sv);
6156 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6162 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6167 /* after multiple passes through Perl_sv_clean_all() for a thingy
6168 * that has badly leaked, the backref array may have gotten freed,
6169 * since we only protect it against 1 round of cleanup */
6170 if (SvIS_FREED(av)) {
6171 if (PL_in_clean_all) /* All is fair */
6174 "panic: magic_killbackrefs (freed backref AV/SV)");
6178 is_array = (SvTYPE(av) == SVt_PVAV);
6180 assert(!SvIS_FREED(av));
6183 last = svp + AvFILLp(av);
6186 /* optimisation: only a single backref, stored directly */
6192 while (svp <= last) {
6194 SV *const referrer = *svp;
6195 if (SvWEAKREF(referrer)) {
6196 /* XXX Should we check that it hasn't changed? */
6197 assert(SvROK(referrer));
6198 SvRV_set(referrer, 0);
6200 SvWEAKREF_off(referrer);
6201 SvSETMAGIC(referrer);
6202 } else if (SvTYPE(referrer) == SVt_PVGV ||
6203 SvTYPE(referrer) == SVt_PVLV) {
6204 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6205 /* You lookin' at me? */
6206 assert(GvSTASH(referrer));
6207 assert(GvSTASH(referrer) == (const HV *)sv);
6208 GvSTASH(referrer) = 0;
6209 } else if (SvTYPE(referrer) == SVt_PVCV ||
6210 SvTYPE(referrer) == SVt_PVFM) {
6211 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6212 /* You lookin' at me? */
6213 assert(CvSTASH(referrer));
6214 assert(CvSTASH(referrer) == (const HV *)sv);
6215 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6218 assert(SvTYPE(sv) == SVt_PVGV);
6219 /* You lookin' at me? */
6220 assert(CvGV(referrer));
6221 assert(CvGV(referrer) == (const GV *)sv);
6222 anonymise_cv_maybe(MUTABLE_GV(sv),
6223 MUTABLE_CV(referrer));
6228 "panic: magic_killbackrefs (flags=%"UVxf")",
6229 (UV)SvFLAGS(referrer));
6240 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6246 =for apidoc sv_insert
6248 Inserts a string at the specified offset/length within the SV. Similar to
6249 the Perl C<substr()> function. Handles get magic.
6251 =for apidoc sv_insert_flags
6253 Same as C<sv_insert>, but the extra C<flags> are passed to the
6254 C<SvPV_force_flags> that applies to C<bigstr>.
6260 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6266 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6269 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6271 SvPV_force_flags(bigstr, curlen, flags);
6272 (void)SvPOK_only_UTF8(bigstr);
6273 if (offset + len > curlen) {
6274 SvGROW(bigstr, offset+len+1);
6275 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6276 SvCUR_set(bigstr, offset+len);
6280 i = littlelen - len;
6281 if (i > 0) { /* string might grow */
6282 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6283 mid = big + offset + len;
6284 midend = bigend = big + SvCUR(bigstr);
6287 while (midend > mid) /* shove everything down */
6288 *--bigend = *--midend;
6289 Move(little,big+offset,littlelen,char);
6290 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6295 Move(little,SvPVX(bigstr)+offset,len,char);
6300 big = SvPVX(bigstr);
6303 bigend = big + SvCUR(bigstr);
6305 if (midend > bigend)
6306 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6309 if (mid - big > bigend - midend) { /* faster to shorten from end */
6311 Move(little, mid, littlelen,char);
6314 i = bigend - midend;
6316 Move(midend, mid, i,char);
6320 SvCUR_set(bigstr, mid - big);
6322 else if ((i = mid - big)) { /* faster from front */
6323 midend -= littlelen;
6325 Move(big, midend - i, i, char);
6326 sv_chop(bigstr,midend-i);
6328 Move(little, mid, littlelen,char);
6330 else if (littlelen) {
6331 midend -= littlelen;
6332 sv_chop(bigstr,midend);
6333 Move(little,midend,littlelen,char);
6336 sv_chop(bigstr,midend);
6342 =for apidoc sv_replace
6344 Make the first argument a copy of the second, then delete the original.
6345 The target SV physically takes over ownership of the body of the source SV
6346 and inherits its flags; however, the target keeps any magic it owns,
6347 and any magic in the source is discarded.
6348 Note that this is a rather specialist SV copying operation; most of the
6349 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6355 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6357 const U32 refcnt = SvREFCNT(sv);
6359 PERL_ARGS_ASSERT_SV_REPLACE;
6361 SV_CHECK_THINKFIRST_COW_DROP(sv);
6362 if (SvREFCNT(nsv) != 1) {
6363 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6364 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6366 if (SvMAGICAL(sv)) {
6370 sv_upgrade(nsv, SVt_PVMG);
6371 SvMAGIC_set(nsv, SvMAGIC(sv));
6372 SvFLAGS(nsv) |= SvMAGICAL(sv);
6374 SvMAGIC_set(sv, NULL);
6378 assert(!SvREFCNT(sv));
6379 #ifdef DEBUG_LEAKING_SCALARS
6380 sv->sv_flags = nsv->sv_flags;
6381 sv->sv_any = nsv->sv_any;
6382 sv->sv_refcnt = nsv->sv_refcnt;
6383 sv->sv_u = nsv->sv_u;
6385 StructCopy(nsv,sv,SV);
6387 if(SvTYPE(sv) == SVt_IV) {
6388 SET_SVANY_FOR_BODYLESS_IV(sv);
6392 SvREFCNT(sv) = refcnt;
6393 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6398 /* We're about to free a GV which has a CV that refers back to us.
6399 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6403 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6408 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6411 assert(SvREFCNT(gv) == 0);
6412 assert(isGV(gv) && isGV_with_GP(gv));
6414 assert(!CvANON(cv));
6415 assert(CvGV(cv) == gv);
6416 assert(!CvNAMED(cv));
6418 /* will the CV shortly be freed by gp_free() ? */
6419 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6420 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6424 /* if not, anonymise: */
6425 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6426 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6427 : newSVpvn_flags( "__ANON__", 8, 0 );
6428 sv_catpvs(gvname, "::__ANON__");
6429 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6430 SvREFCNT_dec_NN(gvname);
6434 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6439 =for apidoc sv_clear
6441 Clear an SV: call any destructors, free up any memory used by the body,
6442 and free the body itself. The SV's head is I<not> freed, although
6443 its type is set to all 1's so that it won't inadvertently be assumed
6444 to be live during global destruction etc.
6445 This function should only be called when C<REFCNT> is zero. Most of the time
6446 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6453 Perl_sv_clear(pTHX_ SV *const orig_sv)
6458 const struct body_details *sv_type_details;
6462 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6463 Not strictly necessary */
6465 PERL_ARGS_ASSERT_SV_CLEAR;
6467 /* within this loop, sv is the SV currently being freed, and
6468 * iter_sv is the most recent AV or whatever that's being iterated
6469 * over to provide more SVs */
6475 assert(SvREFCNT(sv) == 0);
6476 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6478 if (type <= SVt_IV) {
6479 /* See the comment in sv.h about the collusion between this
6480 * early return and the overloading of the NULL slots in the
6484 SvFLAGS(sv) &= SVf_BREAK;
6485 SvFLAGS(sv) |= SVTYPEMASK;
6489 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6490 for another purpose */
6491 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6493 if (type >= SVt_PVMG) {
6495 if (!curse(sv, 1)) goto get_next_sv;
6496 type = SvTYPE(sv); /* destructor may have changed it */
6498 /* Free back-references before magic, in case the magic calls
6499 * Perl code that has weak references to sv. */
6500 if (type == SVt_PVHV) {
6501 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6505 else if (SvMAGIC(sv)) {
6506 /* Free back-references before other types of magic. */
6507 sv_unmagic(sv, PERL_MAGIC_backref);
6513 /* case SVt_INVLIST: */
6516 IoIFP(sv) != PerlIO_stdin() &&
6517 IoIFP(sv) != PerlIO_stdout() &&
6518 IoIFP(sv) != PerlIO_stderr() &&
6519 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6521 io_close(MUTABLE_IO(sv), NULL, FALSE,
6522 (IoTYPE(sv) == IoTYPE_WRONLY ||
6523 IoTYPE(sv) == IoTYPE_RDWR ||
6524 IoTYPE(sv) == IoTYPE_APPEND));
6526 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6527 PerlDir_close(IoDIRP(sv));
6528 IoDIRP(sv) = (DIR*)NULL;
6529 Safefree(IoTOP_NAME(sv));
6530 Safefree(IoFMT_NAME(sv));
6531 Safefree(IoBOTTOM_NAME(sv));
6532 if ((const GV *)sv == PL_statgv)
6536 /* FIXME for plugins */
6538 pregfree2((REGEXP*) sv);
6542 cv_undef(MUTABLE_CV(sv));
6543 /* If we're in a stash, we don't own a reference to it.
6544 * However it does have a back reference to us, which needs to
6546 if ((stash = CvSTASH(sv)))
6547 sv_del_backref(MUTABLE_SV(stash), sv);
6550 if (PL_last_swash_hv == (const HV *)sv) {
6551 PL_last_swash_hv = NULL;
6553 if (HvTOTALKEYS((HV*)sv) > 0) {
6555 /* this statement should match the one at the beginning of
6556 * hv_undef_flags() */
6557 if ( PL_phase != PERL_PHASE_DESTRUCT
6558 && (hek = HvNAME_HEK((HV*)sv)))
6560 if (PL_stashcache) {
6561 DEBUG_o(Perl_deb(aTHX_
6562 "sv_clear clearing PL_stashcache for '%"HEKf
6565 (void)hv_deletehek(PL_stashcache,
6568 hv_name_set((HV*)sv, NULL, 0, 0);
6571 /* save old iter_sv in unused SvSTASH field */
6572 assert(!SvOBJECT(sv));
6573 SvSTASH(sv) = (HV*)iter_sv;
6576 /* save old hash_index in unused SvMAGIC field */
6577 assert(!SvMAGICAL(sv));
6578 assert(!SvMAGIC(sv));
6579 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6582 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6583 goto get_next_sv; /* process this new sv */
6585 /* free empty hash */
6586 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6587 assert(!HvARRAY((HV*)sv));
6591 AV* av = MUTABLE_AV(sv);
6592 if (PL_comppad == av) {
6596 if (AvREAL(av) && AvFILLp(av) > -1) {
6597 next_sv = AvARRAY(av)[AvFILLp(av)--];
6598 /* save old iter_sv in top-most slot of AV,
6599 * and pray that it doesn't get wiped in the meantime */
6600 AvARRAY(av)[AvMAX(av)] = iter_sv;
6602 goto get_next_sv; /* process this new sv */
6604 Safefree(AvALLOC(av));
6609 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6610 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6611 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6612 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6614 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6615 SvREFCNT_dec(LvTARG(sv));
6616 if (isREGEXP(sv)) goto freeregexp;
6619 if (isGV_with_GP(sv)) {
6620 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6621 && HvENAME_get(stash))
6622 mro_method_changed_in(stash);
6623 gp_free(MUTABLE_GV(sv));
6625 unshare_hek(GvNAME_HEK(sv));
6626 /* If we're in a stash, we don't own a reference to it.
6627 * However it does have a back reference to us, which
6628 * needs to be cleared. */
6629 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6630 sv_del_backref(MUTABLE_SV(stash), sv);
6632 /* FIXME. There are probably more unreferenced pointers to SVs
6633 * in the interpreter struct that we should check and tidy in
6634 * a similar fashion to this: */
6635 /* See also S_sv_unglob, which does the same thing. */
6636 if ((const GV *)sv == PL_last_in_gv)
6637 PL_last_in_gv = NULL;
6638 else if ((const GV *)sv == PL_statgv)
6640 else if ((const GV *)sv == PL_stderrgv)
6649 /* Don't bother with SvOOK_off(sv); as we're only going to
6653 SvOOK_offset(sv, offset);
6654 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6655 /* Don't even bother with turning off the OOK flag. */
6660 SV * const target = SvRV(sv);
6662 sv_del_backref(target, sv);
6668 else if (SvPVX_const(sv)
6669 && !(SvTYPE(sv) == SVt_PVIO
6670 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6674 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6678 if (CowREFCNT(sv)) {
6685 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6690 Safefree(SvPVX_mutable(sv));
6694 else if (SvPVX_const(sv) && SvLEN(sv)
6695 && !(SvTYPE(sv) == SVt_PVIO
6696 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6697 Safefree(SvPVX_mutable(sv));
6698 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6699 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6709 SvFLAGS(sv) &= SVf_BREAK;
6710 SvFLAGS(sv) |= SVTYPEMASK;
6712 sv_type_details = bodies_by_type + type;
6713 if (sv_type_details->arena) {
6714 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6715 &PL_body_roots[type]);
6717 else if (sv_type_details->body_size) {
6718 safefree(SvANY(sv));
6722 /* caller is responsible for freeing the head of the original sv */
6723 if (sv != orig_sv && !SvREFCNT(sv))
6726 /* grab and free next sv, if any */
6734 else if (!iter_sv) {
6736 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6737 AV *const av = (AV*)iter_sv;
6738 if (AvFILLp(av) > -1) {
6739 sv = AvARRAY(av)[AvFILLp(av)--];
6741 else { /* no more elements of current AV to free */
6744 /* restore previous value, squirrelled away */
6745 iter_sv = AvARRAY(av)[AvMAX(av)];
6746 Safefree(AvALLOC(av));
6749 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6750 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6751 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6752 /* no more elements of current HV to free */
6755 /* Restore previous values of iter_sv and hash_index,
6756 * squirrelled away */
6757 assert(!SvOBJECT(sv));
6758 iter_sv = (SV*)SvSTASH(sv);
6759 assert(!SvMAGICAL(sv));
6760 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6762 /* perl -DA does not like rubbish in SvMAGIC. */
6766 /* free any remaining detritus from the hash struct */
6767 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6768 assert(!HvARRAY((HV*)sv));
6773 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6777 if (!SvREFCNT(sv)) {
6781 if (--(SvREFCNT(sv)))
6785 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6786 "Attempt to free temp prematurely: SV 0x%"UVxf
6787 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6791 if (SvIMMORTAL(sv)) {
6792 /* make sure SvREFCNT(sv)==0 happens very seldom */
6793 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6802 /* This routine curses the sv itself, not the object referenced by sv. So
6803 sv does not have to be ROK. */
6806 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6807 PERL_ARGS_ASSERT_CURSE;
6808 assert(SvOBJECT(sv));
6810 if (PL_defstash && /* Still have a symbol table? */
6816 stash = SvSTASH(sv);
6817 assert(SvTYPE(stash) == SVt_PVHV);
6818 if (HvNAME(stash)) {
6819 CV* destructor = NULL;
6820 struct mro_meta *meta;
6822 assert (SvOOK(stash));
6824 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6827 /* don't make this an initialization above the assert, since it needs
6829 meta = HvMROMETA(stash);
6830 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6831 destructor = meta->destroy;
6832 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6833 (void *)destructor, HvNAME(stash)) );
6836 bool autoload = FALSE;
6838 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6840 destructor = GvCV(gv);
6842 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6843 GV_AUTOLOAD_ISMETHOD);
6845 destructor = GvCV(gv);
6849 /* we don't cache AUTOLOAD for DESTROY, since this code
6850 would then need to set $__PACKAGE__::AUTOLOAD, or the
6851 equivalent for XS AUTOLOADs */
6853 meta->destroy_gen = PL_sub_generation;
6854 meta->destroy = destructor;
6856 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6857 (void *)destructor, HvNAME(stash)) );
6860 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6864 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6866 /* A constant subroutine can have no side effects, so
6867 don't bother calling it. */
6868 && !CvCONST(destructor)
6869 /* Don't bother calling an empty destructor or one that
6870 returns immediately. */
6871 && (CvISXSUB(destructor)
6872 || (CvSTART(destructor)
6873 && (CvSTART(destructor)->op_next->op_type
6875 && (CvSTART(destructor)->op_next->op_type
6877 || CvSTART(destructor)->op_next->op_next->op_type
6883 SV* const tmpref = newRV(sv);
6884 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6886 PUSHSTACKi(PERLSI_DESTROY);
6891 call_sv(MUTABLE_SV(destructor),
6892 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6896 if(SvREFCNT(tmpref) < 2) {
6897 /* tmpref is not kept alive! */
6899 SvRV_set(tmpref, NULL);
6902 SvREFCNT_dec_NN(tmpref);
6905 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6908 if (check_refcnt && SvREFCNT(sv)) {
6909 if (PL_in_clean_objs)
6911 "DESTROY created new reference to dead object '%"HEKf"'",
6912 HEKfARG(HvNAME_HEK(stash)));
6913 /* DESTROY gave object new lease on life */
6919 HV * const stash = SvSTASH(sv);
6920 /* Curse before freeing the stash, as freeing the stash could cause
6921 a recursive call into S_curse. */
6922 SvOBJECT_off(sv); /* Curse the object. */
6923 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6924 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6930 =for apidoc sv_newref
6932 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6939 Perl_sv_newref(pTHX_ SV *const sv)
6941 PERL_UNUSED_CONTEXT;
6950 Decrement an SV's reference count, and if it drops to zero, call
6951 C<sv_clear> to invoke destructors and free up any memory used by
6952 the body; finally, deallocating the SV's head itself.
6953 Normally called via a wrapper macro C<SvREFCNT_dec>.
6959 Perl_sv_free(pTHX_ SV *const sv)
6965 /* Private helper function for SvREFCNT_dec().
6966 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6969 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6973 PERL_ARGS_ASSERT_SV_FREE2;
6975 if (LIKELY( rc == 1 )) {
6981 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6982 "Attempt to free temp prematurely: SV 0x%"UVxf
6983 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6987 if (SvIMMORTAL(sv)) {
6988 /* make sure SvREFCNT(sv)==0 happens very seldom */
6989 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6993 if (! SvREFCNT(sv)) /* may have have been resurrected */
6998 /* handle exceptional cases */
7002 if (SvFLAGS(sv) & SVf_BREAK)
7003 /* this SV's refcnt has been artificially decremented to
7004 * trigger cleanup */
7006 if (PL_in_clean_all) /* All is fair */
7008 if (SvIMMORTAL(sv)) {
7009 /* make sure SvREFCNT(sv)==0 happens very seldom */
7010 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7013 if (ckWARN_d(WARN_INTERNAL)) {
7014 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7015 Perl_dump_sv_child(aTHX_ sv);
7017 #ifdef DEBUG_LEAKING_SCALARS
7020 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7021 if (PL_warnhook == PERL_WARNHOOK_FATAL
7022 || ckDEAD(packWARN(WARN_INTERNAL))) {
7023 /* Don't let Perl_warner cause us to escape our fate: */
7027 /* This may not return: */
7028 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7029 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7030 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7033 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7043 Returns the length of the string in the SV. Handles magic and type
7044 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7045 gives raw access to the C<xpv_cur> slot.
7051 Perl_sv_len(pTHX_ SV *const sv)
7058 (void)SvPV_const(sv, len);
7063 =for apidoc sv_len_utf8
7065 Returns the number of characters in the string in an SV, counting wide
7066 UTF-8 bytes as a single character. Handles magic and type coercion.
7072 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7073 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7074 * (Note that the mg_len is not the length of the mg_ptr field.
7075 * This allows the cache to store the character length of the string without
7076 * needing to malloc() extra storage to attach to the mg_ptr.)
7081 Perl_sv_len_utf8(pTHX_ SV *const sv)
7087 return sv_len_utf8_nomg(sv);
7091 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7094 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7096 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7098 if (PL_utf8cache && SvUTF8(sv)) {
7100 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7102 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7103 if (mg->mg_len != -1)
7106 /* We can use the offset cache for a headstart.
7107 The longer value is stored in the first pair. */
7108 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7110 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7114 if (PL_utf8cache < 0) {
7115 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7116 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7120 ulen = Perl_utf8_length(aTHX_ s, s + len);
7121 utf8_mg_len_cache_update(sv, &mg, ulen);
7125 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7128 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7131 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7132 STRLEN *const uoffset_p, bool *const at_end)
7134 const U8 *s = start;
7135 STRLEN uoffset = *uoffset_p;
7137 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7139 while (s < send && uoffset) {
7146 else if (s > send) {
7148 /* This is the existing behaviour. Possibly it should be a croak, as
7149 it's actually a bounds error */
7152 *uoffset_p -= uoffset;
7156 /* Given the length of the string in both bytes and UTF-8 characters, decide
7157 whether to walk forwards or backwards to find the byte corresponding to
7158 the passed in UTF-8 offset. */
7160 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7161 STRLEN uoffset, const STRLEN uend)
7163 STRLEN backw = uend - uoffset;
7165 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7167 if (uoffset < 2 * backw) {
7168 /* The assumption is that going forwards is twice the speed of going
7169 forward (that's where the 2 * backw comes from).
7170 (The real figure of course depends on the UTF-8 data.) */
7171 const U8 *s = start;
7173 while (s < send && uoffset--)
7183 while (UTF8_IS_CONTINUATION(*send))
7186 return send - start;
7189 /* For the string representation of the given scalar, find the byte
7190 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7191 give another position in the string, *before* the sought offset, which
7192 (which is always true, as 0, 0 is a valid pair of positions), which should
7193 help reduce the amount of linear searching.
7194 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7195 will be used to reduce the amount of linear searching. The cache will be
7196 created if necessary, and the found value offered to it for update. */
7198 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7199 const U8 *const send, STRLEN uoffset,
7200 STRLEN uoffset0, STRLEN boffset0)
7202 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7204 bool at_end = FALSE;
7206 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7208 assert (uoffset >= uoffset0);
7213 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7215 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7216 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7217 if ((*mgp)->mg_ptr) {
7218 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7219 if (cache[0] == uoffset) {
7220 /* An exact match. */
7223 if (cache[2] == uoffset) {
7224 /* An exact match. */
7228 if (cache[0] < uoffset) {
7229 /* The cache already knows part of the way. */
7230 if (cache[0] > uoffset0) {
7231 /* The cache knows more than the passed in pair */
7232 uoffset0 = cache[0];
7233 boffset0 = cache[1];
7235 if ((*mgp)->mg_len != -1) {
7236 /* And we know the end too. */
7238 + sv_pos_u2b_midway(start + boffset0, send,
7240 (*mgp)->mg_len - uoffset0);
7242 uoffset -= uoffset0;
7244 + sv_pos_u2b_forwards(start + boffset0,
7245 send, &uoffset, &at_end);
7246 uoffset += uoffset0;
7249 else if (cache[2] < uoffset) {
7250 /* We're between the two cache entries. */
7251 if (cache[2] > uoffset0) {
7252 /* and the cache knows more than the passed in pair */
7253 uoffset0 = cache[2];
7254 boffset0 = cache[3];
7258 + sv_pos_u2b_midway(start + boffset0,
7261 cache[0] - uoffset0);
7264 + sv_pos_u2b_midway(start + boffset0,
7267 cache[2] - uoffset0);
7271 else if ((*mgp)->mg_len != -1) {
7272 /* If we can take advantage of a passed in offset, do so. */
7273 /* In fact, offset0 is either 0, or less than offset, so don't
7274 need to worry about the other possibility. */
7276 + sv_pos_u2b_midway(start + boffset0, send,
7278 (*mgp)->mg_len - uoffset0);
7283 if (!found || PL_utf8cache < 0) {
7284 STRLEN real_boffset;
7285 uoffset -= uoffset0;
7286 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7287 send, &uoffset, &at_end);
7288 uoffset += uoffset0;
7290 if (found && PL_utf8cache < 0)
7291 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7293 boffset = real_boffset;
7296 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7298 utf8_mg_len_cache_update(sv, mgp, uoffset);
7300 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7307 =for apidoc sv_pos_u2b_flags
7309 Converts the offset from a count of UTF-8 chars from
7310 the start of the string, to a count of the equivalent number of bytes; if
7311 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7312 C<offset>, rather than from the start
7313 of the string. Handles type coercion.
7314 C<flags> is passed to C<SvPV_flags>, and usually should be
7315 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7321 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7322 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7323 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7328 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7335 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7337 start = (U8*)SvPV_flags(sv, len, flags);
7339 const U8 * const send = start + len;
7341 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7344 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7345 is 0, and *lenp is already set to that. */) {
7346 /* Convert the relative offset to absolute. */
7347 const STRLEN uoffset2 = uoffset + *lenp;
7348 const STRLEN boffset2
7349 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7350 uoffset, boffset) - boffset;
7364 =for apidoc sv_pos_u2b
7366 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7367 the start of the string, to a count of the equivalent number of bytes; if
7368 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7369 the offset, rather than from the start of the string. Handles magic and
7372 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7379 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7380 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7381 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7385 /* This function is subject to size and sign problems */
7388 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7390 PERL_ARGS_ASSERT_SV_POS_U2B;
7393 STRLEN ulen = (STRLEN)*lenp;
7394 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7395 SV_GMAGIC|SV_CONST_RETURN);
7398 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7399 SV_GMAGIC|SV_CONST_RETURN);
7404 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7407 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7408 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7411 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7412 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7413 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7417 (*mgp)->mg_len = ulen;
7420 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7421 byte length pairing. The (byte) length of the total SV is passed in too,
7422 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7423 may not have updated SvCUR, so we can't rely on reading it directly.
7425 The proffered utf8/byte length pairing isn't used if the cache already has
7426 two pairs, and swapping either for the proffered pair would increase the
7427 RMS of the intervals between known byte offsets.
7429 The cache itself consists of 4 STRLEN values
7430 0: larger UTF-8 offset
7431 1: corresponding byte offset
7432 2: smaller UTF-8 offset
7433 3: corresponding byte offset
7435 Unused cache pairs have the value 0, 0.
7436 Keeping the cache "backwards" means that the invariant of
7437 cache[0] >= cache[2] is maintained even with empty slots, which means that
7438 the code that uses it doesn't need to worry if only 1 entry has actually
7439 been set to non-zero. It also makes the "position beyond the end of the
7440 cache" logic much simpler, as the first slot is always the one to start
7444 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7445 const STRLEN utf8, const STRLEN blen)
7449 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7454 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7455 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7456 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7458 (*mgp)->mg_len = -1;
7462 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7463 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7464 (*mgp)->mg_ptr = (char *) cache;
7468 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7469 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7470 a pointer. Note that we no longer cache utf8 offsets on refer-
7471 ences, but this check is still a good idea, for robustness. */
7472 const U8 *start = (const U8 *) SvPVX_const(sv);
7473 const STRLEN realutf8 = utf8_length(start, start + byte);
7475 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7479 /* Cache is held with the later position first, to simplify the code
7480 that deals with unbounded ends. */
7482 ASSERT_UTF8_CACHE(cache);
7483 if (cache[1] == 0) {
7484 /* Cache is totally empty */
7487 } else if (cache[3] == 0) {
7488 if (byte > cache[1]) {
7489 /* New one is larger, so goes first. */
7490 cache[2] = cache[0];
7491 cache[3] = cache[1];
7499 /* float casts necessary? XXX */
7500 #define THREEWAY_SQUARE(a,b,c,d) \
7501 ((float)((d) - (c))) * ((float)((d) - (c))) \
7502 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7503 + ((float)((b) - (a))) * ((float)((b) - (a)))
7505 /* Cache has 2 slots in use, and we know three potential pairs.
7506 Keep the two that give the lowest RMS distance. Do the
7507 calculation in bytes simply because we always know the byte
7508 length. squareroot has the same ordering as the positive value,
7509 so don't bother with the actual square root. */
7510 if (byte > cache[1]) {
7511 /* New position is after the existing pair of pairs. */
7512 const float keep_earlier
7513 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7514 const float keep_later
7515 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7517 if (keep_later < keep_earlier) {
7518 cache[2] = cache[0];
7519 cache[3] = cache[1];
7525 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7526 float b, c, keep_earlier;
7527 if (byte > cache[3]) {
7528 /* New position is between the existing pair of pairs. */
7529 b = (float)cache[3];
7532 /* New position is before the existing pair of pairs. */
7534 c = (float)cache[3];
7536 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7537 if (byte > cache[3]) {
7538 if (keep_later < keep_earlier) {
7548 if (! (keep_later < keep_earlier)) {
7549 cache[0] = cache[2];
7550 cache[1] = cache[3];
7557 ASSERT_UTF8_CACHE(cache);
7560 /* We already know all of the way, now we may be able to walk back. The same
7561 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7562 backward is half the speed of walking forward. */
7564 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7565 const U8 *end, STRLEN endu)
7567 const STRLEN forw = target - s;
7568 STRLEN backw = end - target;
7570 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7572 if (forw < 2 * backw) {
7573 return utf8_length(s, target);
7576 while (end > target) {
7578 while (UTF8_IS_CONTINUATION(*end)) {
7587 =for apidoc sv_pos_b2u_flags
7589 Converts C<offset> from a count of bytes from the start of the string, to
7590 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7591 C<flags> is passed to C<SvPV_flags>, and usually should be
7592 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7598 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7599 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7604 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7607 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7613 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7615 s = (const U8*)SvPV_flags(sv, blen, flags);
7618 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7619 ", byte=%"UVuf, (UV)blen, (UV)offset);
7625 && SvTYPE(sv) >= SVt_PVMG
7626 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7629 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7630 if (cache[1] == offset) {
7631 /* An exact match. */
7634 if (cache[3] == offset) {
7635 /* An exact match. */
7639 if (cache[1] < offset) {
7640 /* We already know part of the way. */
7641 if (mg->mg_len != -1) {
7642 /* Actually, we know the end too. */
7644 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7645 s + blen, mg->mg_len - cache[0]);
7647 len = cache[0] + utf8_length(s + cache[1], send);
7650 else if (cache[3] < offset) {
7651 /* We're between the two cached pairs, so we do the calculation
7652 offset by the byte/utf-8 positions for the earlier pair,
7653 then add the utf-8 characters from the string start to
7655 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7656 s + cache[1], cache[0] - cache[2])
7660 else { /* cache[3] > offset */
7661 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7665 ASSERT_UTF8_CACHE(cache);
7667 } else if (mg->mg_len != -1) {
7668 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7672 if (!found || PL_utf8cache < 0) {
7673 const STRLEN real_len = utf8_length(s, send);
7675 if (found && PL_utf8cache < 0)
7676 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7682 utf8_mg_len_cache_update(sv, &mg, len);
7684 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7691 =for apidoc sv_pos_b2u
7693 Converts the value pointed to by C<offsetp> from a count of bytes from the
7694 start of the string, to a count of the equivalent number of UTF-8 chars.
7695 Handles magic and type coercion.
7697 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7704 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7705 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7710 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7712 PERL_ARGS_ASSERT_SV_POS_B2U;
7717 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7718 SV_GMAGIC|SV_CONST_RETURN);
7722 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7723 STRLEN real, SV *const sv)
7725 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7727 /* As this is debugging only code, save space by keeping this test here,
7728 rather than inlining it in all the callers. */
7729 if (from_cache == real)
7732 /* Need to turn the assertions off otherwise we may recurse infinitely
7733 while printing error messages. */
7734 SAVEI8(PL_utf8cache);
7736 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7737 func, (UV) from_cache, (UV) real, SVfARG(sv));
7743 Returns a boolean indicating whether the strings in the two SVs are
7744 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7745 coerce its args to strings if necessary.
7747 =for apidoc sv_eq_flags
7749 Returns a boolean indicating whether the strings in the two SVs are
7750 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7751 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7757 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7764 SV* svrecode = NULL;
7771 /* if pv1 and pv2 are the same, second SvPV_const call may
7772 * invalidate pv1 (if we are handling magic), so we may need to
7774 if (sv1 == sv2 && flags & SV_GMAGIC
7775 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7776 pv1 = SvPV_const(sv1, cur1);
7777 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7779 pv1 = SvPV_flags_const(sv1, cur1, flags);
7787 pv2 = SvPV_flags_const(sv2, cur2, flags);
7789 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7790 /* Differing utf8ness. */
7792 /* sv1 is the UTF-8 one */
7793 return bytes_cmp_utf8((const U8*)pv2, cur2,
7794 (const U8*)pv1, cur1) == 0;
7797 /* sv2 is the UTF-8 one */
7798 return bytes_cmp_utf8((const U8*)pv1, cur1,
7799 (const U8*)pv2, cur2) == 0;
7804 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7806 SvREFCNT_dec(svrecode);
7814 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7815 string in C<sv1> is less than, equal to, or greater than the string in
7816 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7817 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7819 =for apidoc sv_cmp_flags
7821 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7822 string in C<sv1> is less than, equal to, or greater than the string in
7823 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7824 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7825 also C<L</sv_cmp_locale_flags>>.
7831 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7833 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7837 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7841 const char *pv1, *pv2;
7843 SV *svrecode = NULL;
7850 pv1 = SvPV_flags_const(sv1, cur1, flags);
7857 pv2 = SvPV_flags_const(sv2, cur2, flags);
7859 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7860 /* Differing utf8ness. */
7862 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7863 (const U8*)pv1, cur1);
7864 return retval ? retval < 0 ? -1 : +1 : 0;
7867 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7868 (const U8*)pv2, cur2);
7869 return retval ? retval < 0 ? -1 : +1 : 0;
7873 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7876 cmp = cur2 ? -1 : 0;
7880 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7883 if (! DO_UTF8(sv1)) {
7885 const I32 retval = memcmp((const void*)pv1,
7889 cmp = retval < 0 ? -1 : 1;
7890 } else if (cur1 == cur2) {
7893 cmp = cur1 < cur2 ? -1 : 1;
7897 else { /* Both are to be treated as UTF-EBCDIC */
7899 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7900 * which remaps code points 0-255. We therefore generally have to
7901 * unmap back to the original values to get an accurate comparison.
7902 * But we don't have to do that for UTF-8 invariants, as by
7903 * definition, they aren't remapped, nor do we have to do it for
7904 * above-latin1 code points, as they also aren't remapped. (This
7905 * code also works on ASCII platforms, but the memcmp() above is
7908 const char *e = pv1 + shortest_len;
7910 /* Find the first bytes that differ between the two strings */
7911 while (pv1 < e && *pv1 == *pv2) {
7917 if (pv1 == e) { /* Are the same all the way to the end */
7921 cmp = cur1 < cur2 ? -1 : 1;
7924 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7925 * in the strings were. The current bytes may or may not be
7926 * at the beginning of a character. But neither or both are
7927 * (or else earlier bytes would have been different). And
7928 * if we are in the middle of a character, the two
7929 * characters are comprised of the same number of bytes
7930 * (because in this case the start bytes are the same, and
7931 * the start bytes encode the character's length). */
7932 if (UTF8_IS_INVARIANT(*pv1))
7934 /* If both are invariants; can just compare directly */
7935 if (UTF8_IS_INVARIANT(*pv2)) {
7936 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7938 else /* Since *pv1 is invariant, it is the whole character,
7939 which means it is at the beginning of a character.
7940 That means pv2 is also at the beginning of a
7941 character (see earlier comment). Since it isn't
7942 invariant, it must be a start byte. If it starts a
7943 character whose code point is above 255, that
7944 character is greater than any single-byte char, which
7946 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7951 /* Here, pv2 points to a character composed of 2 bytes
7952 * whose code point is < 256. Get its code point and
7953 * compare with *pv1 */
7954 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7959 else /* The code point starting at pv1 isn't a single byte */
7960 if (UTF8_IS_INVARIANT(*pv2))
7962 /* But here, the code point starting at *pv2 is a single byte,
7963 * and so *pv1 must begin a character, hence is a start byte.
7964 * If that character is above 255, it is larger than any
7965 * single-byte char, which *pv2 is */
7966 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7970 /* Here, pv1 points to a character composed of 2 bytes
7971 * whose code point is < 256. Get its code point and
7972 * compare with the single byte character *pv2 */
7973 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7978 else /* Here, we've ruled out either *pv1 and *pv2 being
7979 invariant. That means both are part of variants, but not
7980 necessarily at the start of a character */
7981 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7982 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7984 /* Here, at least one is the start of a character, which means
7985 * the other is also a start byte. And the code point of at
7986 * least one of the characters is above 255. It is a
7987 * characteristic of UTF-EBCDIC that all start bytes for
7988 * above-latin1 code points are well behaved as far as code
7989 * point comparisons go, and all are larger than all other
7990 * start bytes, so the comparison with those is also well
7992 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7995 /* Here both *pv1 and *pv2 are part of variant characters.
7996 * They could be both continuations, or both start characters.
7997 * (One or both could even be an illegal start character (for
7998 * an overlong) which for the purposes of sorting we treat as
8000 if (UTF8_IS_CONTINUATION(*pv1)) {
8002 /* If they are continuations for code points above 255,
8003 * then comparing the current byte is sufficient, as there
8004 * is no remapping of these and so the comparison is
8005 * well-behaved. We determine if they are such
8006 * continuations by looking at the preceding byte. It
8007 * could be a start byte, from which we can tell if it is
8008 * for an above 255 code point. Or it could be a
8009 * continuation, which means the character occupies at
8010 * least 3 bytes, so must be above 255. */
8011 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8012 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8014 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8018 /* Here, the continuations are for code points below 256;
8019 * back up one to get to the start byte */
8024 /* We need to get the actual native code point of each of these
8025 * variants in order to compare them */
8026 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8027 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8036 SvREFCNT_dec(svrecode);
8042 =for apidoc sv_cmp_locale
8044 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8045 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8046 if necessary. See also C<L</sv_cmp>>.
8048 =for apidoc sv_cmp_locale_flags
8050 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8051 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8052 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8053 C<L</sv_cmp_flags>>.
8059 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8061 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8065 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8068 #ifdef USE_LOCALE_COLLATE
8074 if (PL_collation_standard)
8079 /* Revert to using raw compare if both operands exist, but either one
8080 * doesn't transform properly for collation */
8082 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8086 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8092 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8093 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8096 if (!pv1 || !len1) {
8107 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8110 return retval < 0 ? -1 : 1;
8113 * When the result of collation is equality, that doesn't mean
8114 * that there are no differences -- some locales exclude some
8115 * characters from consideration. So to avoid false equalities,
8116 * we use the raw string as a tiebreaker.
8123 PERL_UNUSED_ARG(flags);
8124 #endif /* USE_LOCALE_COLLATE */
8126 return sv_cmp(sv1, sv2);
8130 #ifdef USE_LOCALE_COLLATE
8133 =for apidoc sv_collxfrm
8135 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8136 C<L</sv_collxfrm_flags>>.
8138 =for apidoc sv_collxfrm_flags
8140 Add Collate Transform magic to an SV if it doesn't already have it. If the
8141 flags contain C<SV_GMAGIC>, it handles get-magic.
8143 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8144 scalar data of the variable, but transformed to such a format that a normal
8145 memory comparison can be used to compare the data according to the locale
8152 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8156 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8158 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8160 /* If we don't have collation magic on 'sv', or the locale has changed
8161 * since the last time we calculated it, get it and save it now */
8162 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8167 /* Free the old space */
8169 Safefree(mg->mg_ptr);
8171 s = SvPV_flags_const(sv, len, flags);
8172 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8174 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8189 if (mg && mg->mg_ptr) {
8191 return mg->mg_ptr + sizeof(PL_collation_ix);
8199 #endif /* USE_LOCALE_COLLATE */
8202 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8204 SV * const tsv = newSV(0);
8207 sv_gets(tsv, fp, 0);
8208 sv_utf8_upgrade_nomg(tsv);
8209 SvCUR_set(sv,append);
8212 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8216 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8219 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8220 /* Grab the size of the record we're getting */
8221 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8228 /* With a true, record-oriented file on VMS, we need to use read directly
8229 * to ensure that we respect RMS record boundaries. The user is responsible
8230 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8231 * record size) field. N.B. This is likely to produce invalid results on
8232 * varying-width character data when a record ends mid-character.
8234 fd = PerlIO_fileno(fp);
8236 && PerlLIO_fstat(fd, &st) == 0
8237 && (st.st_fab_rfm == FAB$C_VAR
8238 || st.st_fab_rfm == FAB$C_VFC
8239 || st.st_fab_rfm == FAB$C_FIX)) {
8241 bytesread = PerlLIO_read(fd, buffer, recsize);
8243 else /* in-memory file from PerlIO::Scalar
8244 * or not a record-oriented file
8248 bytesread = PerlIO_read(fp, buffer, recsize);
8250 /* At this point, the logic in sv_get() means that sv will
8251 be treated as utf-8 if the handle is utf8.
8253 if (PerlIO_isutf8(fp) && bytesread > 0) {
8254 char *bend = buffer + bytesread;
8255 char *bufp = buffer;
8256 size_t charcount = 0;
8257 bool charstart = TRUE;
8260 while (charcount < recsize) {
8261 /* count accumulated characters */
8262 while (bufp < bend) {
8264 skip = UTF8SKIP(bufp);
8266 if (bufp + skip > bend) {
8267 /* partial at the end */
8278 if (charcount < recsize) {
8280 STRLEN bufp_offset = bufp - buffer;
8281 SSize_t morebytesread;
8283 /* originally I read enough to fill any incomplete
8284 character and the first byte of the next
8285 character if needed, but if there's many
8286 multi-byte encoded characters we're going to be
8287 making a read call for every character beyond
8288 the original read size.
8290 So instead, read the rest of the character if
8291 any, and enough bytes to match at least the
8292 start bytes for each character we're going to
8296 readsize = recsize - charcount;
8298 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8299 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8300 bend = buffer + bytesread;
8301 morebytesread = PerlIO_read(fp, bend, readsize);
8302 if (morebytesread <= 0) {
8303 /* we're done, if we still have incomplete
8304 characters the check code in sv_gets() will
8307 I'd originally considered doing
8308 PerlIO_ungetc() on all but the lead
8309 character of the incomplete character, but
8310 read() doesn't do that, so I don't.
8315 /* prepare to scan some more */
8316 bytesread += morebytesread;
8317 bend = buffer + bytesread;
8318 bufp = buffer + bufp_offset;
8326 SvCUR_set(sv, bytesread + append);
8327 buffer[bytesread] = '\0';
8328 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8334 Get a line from the filehandle and store it into the SV, optionally
8335 appending to the currently-stored string. If C<append> is not 0, the
8336 line is appended to the SV instead of overwriting it. C<append> should
8337 be set to the byte offset that the appended string should start at
8338 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8344 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8354 PERL_ARGS_ASSERT_SV_GETS;
8356 if (SvTHINKFIRST(sv))
8357 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8358 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8360 However, perlbench says it's slower, because the existing swipe code
8361 is faster than copy on write.
8362 Swings and roundabouts. */
8363 SvUPGRADE(sv, SVt_PV);
8366 /* line is going to be appended to the existing buffer in the sv */
8367 if (PerlIO_isutf8(fp)) {
8369 sv_utf8_upgrade_nomg(sv);
8370 sv_pos_u2b(sv,&append,0);
8372 } else if (SvUTF8(sv)) {
8373 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8379 /* not appending - "clear" the string by setting SvCUR to 0,
8380 * the pv is still avaiable. */
8383 if (PerlIO_isutf8(fp))
8386 if (IN_PERL_COMPILETIME) {
8387 /* we always read code in line mode */
8391 else if (RsSNARF(PL_rs)) {
8392 /* If it is a regular disk file use size from stat() as estimate
8393 of amount we are going to read -- may result in mallocing
8394 more memory than we really need if the layers below reduce
8395 the size we read (e.g. CRLF or a gzip layer).
8398 int fd = PerlIO_fileno(fp);
8399 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8400 const Off_t offset = PerlIO_tell(fp);
8401 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8402 #ifdef PERL_COPY_ON_WRITE
8403 /* Add an extra byte for the sake of copy-on-write's
8404 * buffer reference count. */
8405 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8407 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8414 else if (RsRECORD(PL_rs)) {
8415 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8417 else if (RsPARA(PL_rs)) {
8423 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8424 if (PerlIO_isutf8(fp)) {
8425 rsptr = SvPVutf8(PL_rs, rslen);
8428 if (SvUTF8(PL_rs)) {
8429 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8430 Perl_croak(aTHX_ "Wide character in $/");
8433 /* extract the raw pointer to the record separator */
8434 rsptr = SvPV_const(PL_rs, rslen);
8438 /* rslast is the last character in the record separator
8439 * note we don't use rslast except when rslen is true, so the
8440 * null assign is a placeholder. */
8441 rslast = rslen ? rsptr[rslen - 1] : '\0';
8443 if (rspara) { /* have to do this both before and after */
8444 do { /* to make sure file boundaries work right */
8447 i = PerlIO_getc(fp);
8451 PerlIO_ungetc(fp,i);
8457 /* See if we know enough about I/O mechanism to cheat it ! */
8459 /* This used to be #ifdef test - it is made run-time test for ease
8460 of abstracting out stdio interface. One call should be cheap
8461 enough here - and may even be a macro allowing compile
8465 if (PerlIO_fast_gets(fp)) {
8467 * We can do buffer based IO operations on this filehandle.
8469 * This means we can bypass a lot of subcalls and process
8470 * the buffer directly, it also means we know the upper bound
8471 * on the amount of data we might read of the current buffer
8472 * into our sv. Knowing this allows us to preallocate the pv
8473 * to be able to hold that maximum, which allows us to simplify
8474 * a lot of logic. */
8477 * We're going to steal some values from the stdio struct
8478 * and put EVERYTHING in the innermost loop into registers.
8480 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8481 STRLEN bpx; /* length of the data in the target sv
8482 used to fix pointers after a SvGROW */
8483 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8484 of data left in the read-ahead buffer.
8485 If 0 then the pv buffer can hold the full
8486 amount left, otherwise this is the amount it
8489 /* Here is some breathtakingly efficient cheating */
8491 /* When you read the following logic resist the urge to think
8492 * of record separators that are 1 byte long. They are an
8493 * uninteresting special (simple) case.
8495 * Instead think of record separators which are at least 2 bytes
8496 * long, and keep in mind that we need to deal with such
8497 * separators when they cross a read-ahead buffer boundary.
8499 * Also consider that we need to gracefully deal with separators
8500 * that may be longer than a single read ahead buffer.
8502 * Lastly do not forget we want to copy the delimiter as well. We
8503 * are copying all data in the file _up_to_and_including_ the separator
8506 * Now that you have all that in mind here is what is happening below:
8508 * 1. When we first enter the loop we do some memory book keeping to see
8509 * how much free space there is in the target SV. (This sub assumes that
8510 * it is operating on the same SV most of the time via $_ and that it is
8511 * going to be able to reuse the same pv buffer each call.) If there is
8512 * "enough" room then we set "shortbuffered" to how much space there is
8513 * and start reading forward.
8515 * 2. When we scan forward we copy from the read-ahead buffer to the target
8516 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8517 * and the end of the of pv, as well as for the "rslast", which is the last
8518 * char of the separator.
8520 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8521 * (which has a "complete" record up to the point we saw rslast) and check
8522 * it to see if it matches the separator. If it does we are done. If it doesn't
8523 * we continue on with the scan/copy.
8525 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8526 * the IO system to read the next buffer. We do this by doing a getc(), which
8527 * returns a single char read (or EOF), and prefills the buffer, and also
8528 * allows us to find out how full the buffer is. We use this information to
8529 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8530 * the returned single char into the target sv, and then go back into scan
8533 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8534 * remaining space in the read-buffer.
8536 * Note that this code despite its twisty-turny nature is pretty darn slick.
8537 * It manages single byte separators, multi-byte cross boundary separators,
8538 * and cross-read-buffer separators cleanly and efficiently at the cost
8539 * of potentially greatly overallocating the target SV.
8545 /* get the number of bytes remaining in the read-ahead buffer
8546 * on first call on a given fp this will return 0.*/
8547 cnt = PerlIO_get_cnt(fp);
8549 /* make sure we have the room */
8550 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8551 /* Not room for all of it
8552 if we are looking for a separator and room for some
8554 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8555 /* just process what we have room for */
8556 shortbuffered = cnt - SvLEN(sv) + append + 1;
8557 cnt -= shortbuffered;
8560 /* ensure that the target sv has enough room to hold
8561 * the rest of the read-ahead buffer */
8563 /* remember that cnt can be negative */
8564 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8568 /* we have enough room to hold the full buffer, lets scream */
8572 /* extract the pointer to sv's string buffer, offset by append as necessary */
8573 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8574 /* extract the point to the read-ahead buffer */
8575 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8577 /* some trace debug output */
8578 DEBUG_P(PerlIO_printf(Perl_debug_log,
8579 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8580 DEBUG_P(PerlIO_printf(Perl_debug_log,
8581 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8583 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8584 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8588 /* if there is stuff left in the read-ahead buffer */
8590 /* if there is a separator */
8592 /* loop until we hit the end of the read-ahead buffer */
8593 while (cnt > 0) { /* this | eat */
8594 /* scan forward copying and searching for rslast as we go */
8596 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8597 goto thats_all_folks; /* screams | sed :-) */
8601 /* no separator, slurp the full buffer */
8602 Copy(ptr, bp, cnt, char); /* this | eat */
8603 bp += cnt; /* screams | dust */
8604 ptr += cnt; /* louder | sed :-) */
8606 assert (!shortbuffered);
8607 goto cannot_be_shortbuffered;
8611 if (shortbuffered) { /* oh well, must extend */
8612 /* we didnt have enough room to fit the line into the target buffer
8613 * so we must extend the target buffer and keep going */
8614 cnt = shortbuffered;
8616 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8618 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8619 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8620 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8624 cannot_be_shortbuffered:
8625 /* we need to refill the read-ahead buffer if possible */
8627 DEBUG_P(PerlIO_printf(Perl_debug_log,
8628 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8629 PTR2UV(ptr),(IV)cnt));
8630 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8632 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8633 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8634 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8635 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8638 call PerlIO_getc() to let it prefill the lookahead buffer
8640 This used to call 'filbuf' in stdio form, but as that behaves like
8641 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8642 another abstraction.
8644 Note we have to deal with the char in 'i' if we are not at EOF
8646 i = PerlIO_getc(fp); /* get more characters */
8648 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8649 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8650 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8651 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8653 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8654 cnt = PerlIO_get_cnt(fp);
8655 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8656 DEBUG_P(PerlIO_printf(Perl_debug_log,
8657 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8658 PTR2UV(ptr),(IV)cnt));
8660 if (i == EOF) /* all done for ever? */
8661 goto thats_really_all_folks;
8663 /* make sure we have enough space in the target sv */
8664 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8666 SvGROW(sv, bpx + cnt + 2);
8667 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8669 /* copy of the char we got from getc() */
8670 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8672 /* make sure we deal with the i being the last character of a separator */
8673 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8674 goto thats_all_folks;
8678 /* check if we have actually found the separator - only really applies
8680 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8681 memNE((char*)bp - rslen, rsptr, rslen))
8682 goto screamer; /* go back to the fray */
8683 thats_really_all_folks:
8685 cnt += shortbuffered;
8686 DEBUG_P(PerlIO_printf(Perl_debug_log,
8687 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8688 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8689 DEBUG_P(PerlIO_printf(Perl_debug_log,
8690 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8692 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8693 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8695 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8696 DEBUG_P(PerlIO_printf(Perl_debug_log,
8697 "Screamer: done, len=%ld, string=|%.*s|\n",
8698 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8702 /*The big, slow, and stupid way. */
8703 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8704 STDCHAR *buf = NULL;
8705 Newx(buf, 8192, STDCHAR);
8713 const STDCHAR * const bpe = buf + sizeof(buf);
8715 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8716 ; /* keep reading */
8720 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8721 /* Accommodate broken VAXC compiler, which applies U8 cast to
8722 * both args of ?: operator, causing EOF to change into 255
8725 i = (U8)buf[cnt - 1];
8731 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8733 sv_catpvn_nomg(sv, (char *) buf, cnt);
8735 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8737 if (i != EOF && /* joy */
8739 SvCUR(sv) < rslen ||
8740 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8744 * If we're reading from a TTY and we get a short read,
8745 * indicating that the user hit his EOF character, we need
8746 * to notice it now, because if we try to read from the TTY
8747 * again, the EOF condition will disappear.
8749 * The comparison of cnt to sizeof(buf) is an optimization
8750 * that prevents unnecessary calls to feof().
8754 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8758 #ifdef USE_HEAP_INSTEAD_OF_STACK
8763 if (rspara) { /* have to do this both before and after */
8764 while (i != EOF) { /* to make sure file boundaries work right */
8765 i = PerlIO_getc(fp);
8767 PerlIO_ungetc(fp,i);
8773 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8779 Auto-increment of the value in the SV, doing string to numeric conversion
8780 if necessary. Handles 'get' magic and operator overloading.
8786 Perl_sv_inc(pTHX_ SV *const sv)
8795 =for apidoc sv_inc_nomg
8797 Auto-increment of the value in the SV, doing string to numeric conversion
8798 if necessary. Handles operator overloading. Skips handling 'get' magic.
8804 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8811 if (SvTHINKFIRST(sv)) {
8812 if (SvREADONLY(sv)) {
8813 Perl_croak_no_modify();
8817 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8819 i = PTR2IV(SvRV(sv));
8823 else sv_force_normal_flags(sv, 0);
8825 flags = SvFLAGS(sv);
8826 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8827 /* It's (privately or publicly) a float, but not tested as an
8828 integer, so test it to see. */
8830 flags = SvFLAGS(sv);
8832 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8833 /* It's publicly an integer, or privately an integer-not-float */
8834 #ifdef PERL_PRESERVE_IVUV
8838 if (SvUVX(sv) == UV_MAX)
8839 sv_setnv(sv, UV_MAX_P1);
8841 (void)SvIOK_only_UV(sv);
8842 SvUV_set(sv, SvUVX(sv) + 1);
8844 if (SvIVX(sv) == IV_MAX)
8845 sv_setuv(sv, (UV)IV_MAX + 1);
8847 (void)SvIOK_only(sv);
8848 SvIV_set(sv, SvIVX(sv) + 1);
8853 if (flags & SVp_NOK) {
8854 const NV was = SvNVX(sv);
8855 if (LIKELY(!Perl_isinfnan(was)) &&
8856 NV_OVERFLOWS_INTEGERS_AT &&
8857 was >= NV_OVERFLOWS_INTEGERS_AT) {
8858 /* diag_listed_as: Lost precision when %s %f by 1 */
8859 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8860 "Lost precision when incrementing %" NVff " by 1",
8863 (void)SvNOK_only(sv);
8864 SvNV_set(sv, was + 1.0);
8868 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8869 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8870 Perl_croak_no_modify();
8872 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8873 if ((flags & SVTYPEMASK) < SVt_PVIV)
8874 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8875 (void)SvIOK_only(sv);
8880 while (isALPHA(*d)) d++;
8881 while (isDIGIT(*d)) d++;
8882 if (d < SvEND(sv)) {
8883 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8884 #ifdef PERL_PRESERVE_IVUV
8885 /* Got to punt this as an integer if needs be, but we don't issue
8886 warnings. Probably ought to make the sv_iv_please() that does
8887 the conversion if possible, and silently. */
8888 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8889 /* Need to try really hard to see if it's an integer.
8890 9.22337203685478e+18 is an integer.
8891 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8892 so $a="9.22337203685478e+18"; $a+0; $a++
8893 needs to be the same as $a="9.22337203685478e+18"; $a++
8900 /* sv_2iv *should* have made this an NV */
8901 if (flags & SVp_NOK) {
8902 (void)SvNOK_only(sv);
8903 SvNV_set(sv, SvNVX(sv) + 1.0);
8906 /* I don't think we can get here. Maybe I should assert this
8907 And if we do get here I suspect that sv_setnv will croak. NWC
8909 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8910 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8912 #endif /* PERL_PRESERVE_IVUV */
8913 if (!numtype && ckWARN(WARN_NUMERIC))
8914 not_incrementable(sv);
8915 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8919 while (d >= SvPVX_const(sv)) {
8927 /* MKS: The original code here died if letters weren't consecutive.
8928 * at least it didn't have to worry about non-C locales. The
8929 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8930 * arranged in order (although not consecutively) and that only
8931 * [A-Za-z] are accepted by isALPHA in the C locale.
8933 if (isALPHA_FOLD_NE(*d, 'z')) {
8934 do { ++*d; } while (!isALPHA(*d));
8937 *(d--) -= 'z' - 'a';
8942 *(d--) -= 'z' - 'a' + 1;
8946 /* oh,oh, the number grew */
8947 SvGROW(sv, SvCUR(sv) + 2);
8948 SvCUR_set(sv, SvCUR(sv) + 1);
8949 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8960 Auto-decrement of the value in the SV, doing string to numeric conversion
8961 if necessary. Handles 'get' magic and operator overloading.
8967 Perl_sv_dec(pTHX_ SV *const sv)
8976 =for apidoc sv_dec_nomg
8978 Auto-decrement of the value in the SV, doing string to numeric conversion
8979 if necessary. Handles operator overloading. Skips handling 'get' magic.
8985 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8991 if (SvTHINKFIRST(sv)) {
8992 if (SvREADONLY(sv)) {
8993 Perl_croak_no_modify();
8997 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8999 i = PTR2IV(SvRV(sv));
9003 else sv_force_normal_flags(sv, 0);
9005 /* Unlike sv_inc we don't have to worry about string-never-numbers
9006 and keeping them magic. But we mustn't warn on punting */
9007 flags = SvFLAGS(sv);
9008 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9009 /* It's publicly an integer, or privately an integer-not-float */
9010 #ifdef PERL_PRESERVE_IVUV
9014 if (SvUVX(sv) == 0) {
9015 (void)SvIOK_only(sv);
9019 (void)SvIOK_only_UV(sv);
9020 SvUV_set(sv, SvUVX(sv) - 1);
9023 if (SvIVX(sv) == IV_MIN) {
9024 sv_setnv(sv, (NV)IV_MIN);
9028 (void)SvIOK_only(sv);
9029 SvIV_set(sv, SvIVX(sv) - 1);
9034 if (flags & SVp_NOK) {
9037 const NV was = SvNVX(sv);
9038 if (LIKELY(!Perl_isinfnan(was)) &&
9039 NV_OVERFLOWS_INTEGERS_AT &&
9040 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9041 /* diag_listed_as: Lost precision when %s %f by 1 */
9042 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9043 "Lost precision when decrementing %" NVff " by 1",
9046 (void)SvNOK_only(sv);
9047 SvNV_set(sv, was - 1.0);
9052 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9053 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9054 Perl_croak_no_modify();
9056 if (!(flags & SVp_POK)) {
9057 if ((flags & SVTYPEMASK) < SVt_PVIV)
9058 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9060 (void)SvIOK_only(sv);
9063 #ifdef PERL_PRESERVE_IVUV
9065 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9066 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9067 /* Need to try really hard to see if it's an integer.
9068 9.22337203685478e+18 is an integer.
9069 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9070 so $a="9.22337203685478e+18"; $a+0; $a--
9071 needs to be the same as $a="9.22337203685478e+18"; $a--
9078 /* sv_2iv *should* have made this an NV */
9079 if (flags & SVp_NOK) {
9080 (void)SvNOK_only(sv);
9081 SvNV_set(sv, SvNVX(sv) - 1.0);
9084 /* I don't think we can get here. Maybe I should assert this
9085 And if we do get here I suspect that sv_setnv will croak. NWC
9087 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9088 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9091 #endif /* PERL_PRESERVE_IVUV */
9092 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9095 /* this define is used to eliminate a chunk of duplicated but shared logic
9096 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9097 * used anywhere but here - yves
9099 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9101 SSize_t ix = ++PL_tmps_ix; \
9102 if (UNLIKELY(ix >= PL_tmps_max)) \
9103 ix = tmps_grow_p(ix); \
9104 PL_tmps_stack[ix] = (AnSv); \
9108 =for apidoc sv_mortalcopy
9110 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9111 The new SV is marked as mortal. It will be destroyed "soon", either by an
9112 explicit call to C<FREETMPS>, or by an implicit call at places such as
9113 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9118 /* Make a string that will exist for the duration of the expression
9119 * evaluation. Actually, it may have to last longer than that, but
9120 * hopefully we won't free it until it has been assigned to a
9121 * permanent location. */
9124 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9128 if (flags & SV_GMAGIC)
9129 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9131 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9132 PUSH_EXTEND_MORTAL__SV_C(sv);
9138 =for apidoc sv_newmortal
9140 Creates a new null SV which is mortal. The reference count of the SV is
9141 set to 1. It will be destroyed "soon", either by an explicit call to
9142 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9143 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9149 Perl_sv_newmortal(pTHX)
9154 SvFLAGS(sv) = SVs_TEMP;
9155 PUSH_EXTEND_MORTAL__SV_C(sv);
9161 =for apidoc newSVpvn_flags
9163 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9164 characters) into it. The reference count for the
9165 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9166 string. You are responsible for ensuring that the source string is at least
9167 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9168 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9169 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9170 returning. If C<SVf_UTF8> is set, C<s>
9171 is considered to be in UTF-8 and the
9172 C<SVf_UTF8> flag will be set on the new SV.
9173 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9175 #define newSVpvn_utf8(s, len, u) \
9176 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9182 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9186 /* All the flags we don't support must be zero.
9187 And we're new code so I'm going to assert this from the start. */
9188 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9190 sv_setpvn(sv,s,len);
9192 /* This code used to do a sv_2mortal(), however we now unroll the call to
9193 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9194 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9195 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9196 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9197 * means that we eliminate quite a few steps than it looks - Yves
9198 * (explaining patch by gfx) */
9200 SvFLAGS(sv) |= flags;
9202 if(flags & SVs_TEMP){
9203 PUSH_EXTEND_MORTAL__SV_C(sv);
9210 =for apidoc sv_2mortal
9212 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9213 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9214 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9215 string buffer can be "stolen" if this SV is copied. See also
9216 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9222 Perl_sv_2mortal(pTHX_ SV *const sv)
9229 PUSH_EXTEND_MORTAL__SV_C(sv);
9237 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9238 characters) into it. The reference count for the
9239 SV is set to 1. If C<len> is zero, Perl will compute the length using
9240 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9241 C<NUL> characters and has to have a terminating C<NUL> byte).
9243 For efficiency, consider using C<newSVpvn> instead.
9249 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9254 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9259 =for apidoc newSVpvn
9261 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9262 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9263 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9264 are responsible for ensuring that the source buffer is at least
9265 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9272 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9276 sv_setpvn(sv,buffer,len);
9281 =for apidoc newSVhek
9283 Creates a new SV from the hash key structure. It will generate scalars that
9284 point to the shared string table where possible. Returns a new (undefined)
9285 SV if C<hek> is NULL.
9291 Perl_newSVhek(pTHX_ const HEK *const hek)
9300 if (HEK_LEN(hek) == HEf_SVKEY) {
9301 return newSVsv(*(SV**)HEK_KEY(hek));
9303 const int flags = HEK_FLAGS(hek);
9304 if (flags & HVhek_WASUTF8) {
9306 Andreas would like keys he put in as utf8 to come back as utf8
9308 STRLEN utf8_len = HEK_LEN(hek);
9309 SV * const sv = newSV_type(SVt_PV);
9310 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9311 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9312 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9315 } else if (flags & HVhek_UNSHARED) {
9316 /* A hash that isn't using shared hash keys has to have
9317 the flag in every key so that we know not to try to call
9318 share_hek_hek on it. */
9320 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9325 /* This will be overwhelminly the most common case. */
9327 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9328 more efficient than sharepvn(). */
9332 sv_upgrade(sv, SVt_PV);
9333 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9334 SvCUR_set(sv, HEK_LEN(hek));
9346 =for apidoc newSVpvn_share
9348 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9349 table. If the string does not already exist in the table, it is
9350 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9351 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9352 is non-zero, that value is used; otherwise the hash is computed.
9353 The string's hash can later be retrieved from the SV
9354 with the C<SvSHARED_HASH()> macro. The idea here is
9355 that as the string table is used for shared hash keys these strings will have
9356 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9362 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9366 bool is_utf8 = FALSE;
9367 const char *const orig_src = src;
9370 STRLEN tmplen = -len;
9372 /* See the note in hv.c:hv_fetch() --jhi */
9373 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9377 PERL_HASH(hash, src, len);
9379 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9380 changes here, update it there too. */
9381 sv_upgrade(sv, SVt_PV);
9382 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9389 if (src != orig_src)
9395 =for apidoc newSVpv_share
9397 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9404 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9406 return newSVpvn_share(src, strlen(src), hash);
9409 #if defined(PERL_IMPLICIT_CONTEXT)
9411 /* pTHX_ magic can't cope with varargs, so this is a no-context
9412 * version of the main function, (which may itself be aliased to us).
9413 * Don't access this version directly.
9417 Perl_newSVpvf_nocontext(const char *const pat, ...)
9423 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9425 va_start(args, pat);
9426 sv = vnewSVpvf(pat, &args);
9433 =for apidoc newSVpvf
9435 Creates a new SV and initializes it with the string formatted like
9442 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9447 PERL_ARGS_ASSERT_NEWSVPVF;
9449 va_start(args, pat);
9450 sv = vnewSVpvf(pat, &args);
9455 /* backend for newSVpvf() and newSVpvf_nocontext() */
9458 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9462 PERL_ARGS_ASSERT_VNEWSVPVF;
9465 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9472 Creates a new SV and copies a floating point value into it.
9473 The reference count for the SV is set to 1.
9479 Perl_newSVnv(pTHX_ const NV n)
9491 Creates a new SV and copies an integer into it. The reference count for the
9498 Perl_newSViv(pTHX_ const IV i)
9504 /* Inlining ONLY the small relevant subset of sv_setiv here
9505 * for performance. Makes a significant difference. */
9507 /* We're starting from SVt_FIRST, so provided that's
9508 * actual 0, we don't have to unset any SV type flags
9509 * to promote to SVt_IV. */
9510 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9512 SET_SVANY_FOR_BODYLESS_IV(sv);
9513 SvFLAGS(sv) |= SVt_IV;
9525 Creates a new SV and copies an unsigned integer into it.
9526 The reference count for the SV is set to 1.
9532 Perl_newSVuv(pTHX_ const UV u)
9536 /* Inlining ONLY the small relevant subset of sv_setuv here
9537 * for performance. Makes a significant difference. */
9539 /* Using ivs is more efficient than using uvs - see sv_setuv */
9540 if (u <= (UV)IV_MAX) {
9541 return newSViv((IV)u);
9546 /* We're starting from SVt_FIRST, so provided that's
9547 * actual 0, we don't have to unset any SV type flags
9548 * to promote to SVt_IV. */
9549 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9551 SET_SVANY_FOR_BODYLESS_IV(sv);
9552 SvFLAGS(sv) |= SVt_IV;
9554 (void)SvIsUV_on(sv);
9563 =for apidoc newSV_type
9565 Creates a new SV, of the type specified. The reference count for the new SV
9572 Perl_newSV_type(pTHX_ const svtype type)
9577 ASSUME(SvTYPE(sv) == SVt_FIRST);
9578 if(type != SVt_FIRST)
9579 sv_upgrade(sv, type);
9584 =for apidoc newRV_noinc
9586 Creates an RV wrapper for an SV. The reference count for the original
9587 SV is B<not> incremented.
9593 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9597 PERL_ARGS_ASSERT_NEWRV_NOINC;
9601 /* We're starting from SVt_FIRST, so provided that's
9602 * actual 0, we don't have to unset any SV type flags
9603 * to promote to SVt_IV. */
9604 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9606 SET_SVANY_FOR_BODYLESS_IV(sv);
9607 SvFLAGS(sv) |= SVt_IV;
9612 SvRV_set(sv, tmpRef);
9617 /* newRV_inc is the official function name to use now.
9618 * newRV_inc is in fact #defined to newRV in sv.h
9622 Perl_newRV(pTHX_ SV *const sv)
9624 PERL_ARGS_ASSERT_NEWRV;
9626 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9632 Creates a new SV which is an exact duplicate of the original SV.
9639 Perl_newSVsv(pTHX_ SV *const old)
9645 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9646 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9649 /* Do this here, otherwise we leak the new SV if this croaks. */
9652 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9653 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9654 sv_setsv_flags(sv, old, SV_NOSTEAL);
9659 =for apidoc sv_reset
9661 Underlying implementation for the C<reset> Perl function.
9662 Note that the perl-level function is vaguely deprecated.
9668 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9670 PERL_ARGS_ASSERT_SV_RESET;
9672 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9676 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9678 char todo[PERL_UCHAR_MAX+1];
9681 if (!stash || SvTYPE(stash) != SVt_PVHV)
9684 if (!s) { /* reset ?? searches */
9685 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9687 const U32 count = mg->mg_len / sizeof(PMOP**);
9688 PMOP **pmp = (PMOP**) mg->mg_ptr;
9689 PMOP *const *const end = pmp + count;
9693 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9695 (*pmp)->op_pmflags &= ~PMf_USED;
9703 /* reset variables */
9705 if (!HvARRAY(stash))
9708 Zero(todo, 256, char);
9712 I32 i = (unsigned char)*s;
9716 max = (unsigned char)*s++;
9717 for ( ; i <= max; i++) {
9720 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9722 for (entry = HvARRAY(stash)[i];
9724 entry = HeNEXT(entry))
9729 if (!todo[(U8)*HeKEY(entry)])
9731 gv = MUTABLE_GV(HeVAL(entry));
9735 if (sv && !SvREADONLY(sv)) {
9736 SV_CHECK_THINKFIRST_COW_DROP(sv);
9737 if (!isGV(sv)) SvOK_off(sv);
9742 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9753 Using various gambits, try to get an IO from an SV: the IO slot if its a
9754 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9755 named after the PV if we're a string.
9757 'Get' magic is ignored on the C<sv> passed in, but will be called on
9758 C<SvRV(sv)> if C<sv> is an RV.
9764 Perl_sv_2io(pTHX_ SV *const sv)
9769 PERL_ARGS_ASSERT_SV_2IO;
9771 switch (SvTYPE(sv)) {
9773 io = MUTABLE_IO(sv);
9777 if (isGV_with_GP(sv)) {
9778 gv = MUTABLE_GV(sv);
9781 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9782 HEKfARG(GvNAME_HEK(gv)));
9788 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9790 SvGETMAGIC(SvRV(sv));
9791 return sv_2io(SvRV(sv));
9793 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9800 if (SvGMAGICAL(sv)) {
9801 newsv = sv_newmortal();
9802 sv_setsv_nomg(newsv, sv);
9804 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9814 Using various gambits, try to get a CV from an SV; in addition, try if
9815 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9816 The flags in C<lref> are passed to C<gv_fetchsv>.
9822 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9827 PERL_ARGS_ASSERT_SV_2CV;
9834 switch (SvTYPE(sv)) {
9838 return MUTABLE_CV(sv);
9848 sv = amagic_deref_call(sv, to_cv_amg);
9851 if (SvTYPE(sv) == SVt_PVCV) {
9852 cv = MUTABLE_CV(sv);
9857 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9858 gv = MUTABLE_GV(sv);
9860 Perl_croak(aTHX_ "Not a subroutine reference");
9862 else if (isGV_with_GP(sv)) {
9863 gv = MUTABLE_GV(sv);
9866 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9873 /* Some flags to gv_fetchsv mean don't really create the GV */
9874 if (!isGV_with_GP(gv)) {
9879 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9880 /* XXX this is probably not what they think they're getting.
9881 * It has the same effect as "sub name;", i.e. just a forward
9892 Returns true if the SV has a true value by Perl's rules.
9893 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9894 instead use an in-line version.
9900 Perl_sv_true(pTHX_ SV *const sv)
9905 const XPV* const tXpv = (XPV*)SvANY(sv);
9907 (tXpv->xpv_cur > 1 ||
9908 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9915 return SvIVX(sv) != 0;
9918 return SvNVX(sv) != 0.0;
9920 return sv_2bool(sv);
9926 =for apidoc sv_pvn_force
9928 Get a sensible string out of the SV somehow.
9929 A private implementation of the C<SvPV_force> macro for compilers which
9930 can't cope with complex macro expressions. Always use the macro instead.
9932 =for apidoc sv_pvn_force_flags
9934 Get a sensible string out of the SV somehow.
9935 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9936 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9937 implemented in terms of this function.
9938 You normally want to use the various wrapper macros instead: see
9939 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9945 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9947 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9949 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9950 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9951 sv_force_normal_flags(sv, 0);
9961 if (SvTYPE(sv) > SVt_PVLV
9962 || isGV_with_GP(sv))
9963 /* diag_listed_as: Can't coerce %s to %s in %s */
9964 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9966 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9973 if (SvTYPE(sv) < SVt_PV ||
9974 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9977 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9978 SvGROW(sv, len + 1);
9979 Move(s,SvPVX(sv),len,char);
9981 SvPVX(sv)[len] = '\0';
9984 SvPOK_on(sv); /* validate pointer */
9986 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9987 PTR2UV(sv),SvPVX_const(sv)));
9990 (void)SvPOK_only_UTF8(sv);
9991 return SvPVX_mutable(sv);
9995 =for apidoc sv_pvbyten_force
9997 The backend for the C<SvPVbytex_force> macro. Always use the macro
10004 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10006 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10008 sv_pvn_force(sv,lp);
10009 sv_utf8_downgrade(sv,0);
10015 =for apidoc sv_pvutf8n_force
10017 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10024 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10026 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10028 sv_pvn_force(sv,0);
10029 sv_utf8_upgrade_nomg(sv);
10035 =for apidoc sv_reftype
10037 Returns a string describing what the SV is a reference to.
10039 If ob is true and the SV is blessed, the string is the class name,
10040 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10046 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10048 PERL_ARGS_ASSERT_SV_REFTYPE;
10049 if (ob && SvOBJECT(sv)) {
10050 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10053 /* WARNING - There is code, for instance in mg.c, that assumes that
10054 * the only reason that sv_reftype(sv,0) would return a string starting
10055 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10056 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10057 * this routine inside other subs, and it saves time.
10058 * Do not change this assumption without searching for "dodgy type check" in
10061 switch (SvTYPE(sv)) {
10076 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10077 /* tied lvalues should appear to be
10078 * scalars for backwards compatibility */
10079 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10080 ? "SCALAR" : "LVALUE");
10081 case SVt_PVAV: return "ARRAY";
10082 case SVt_PVHV: return "HASH";
10083 case SVt_PVCV: return "CODE";
10084 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10085 ? "GLOB" : "SCALAR");
10086 case SVt_PVFM: return "FORMAT";
10087 case SVt_PVIO: return "IO";
10088 case SVt_INVLIST: return "INVLIST";
10089 case SVt_REGEXP: return "REGEXP";
10090 default: return "UNKNOWN";
10098 Returns a SV describing what the SV passed in is a reference to.
10100 dst can be a SV to be set to the description or NULL, in which case a
10101 mortal SV is returned.
10103 If ob is true and the SV is blessed, the description is the class
10104 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10110 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10112 PERL_ARGS_ASSERT_SV_REF;
10115 dst = sv_newmortal();
10117 if (ob && SvOBJECT(sv)) {
10118 HvNAME_get(SvSTASH(sv))
10119 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10120 : sv_setpvs(dst, "__ANON__");
10123 const char * reftype = sv_reftype(sv, 0);
10124 sv_setpv(dst, reftype);
10130 =for apidoc sv_isobject
10132 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10133 object. If the SV is not an RV, or if the object is not blessed, then this
10140 Perl_sv_isobject(pTHX_ SV *sv)
10156 Returns a boolean indicating whether the SV is blessed into the specified
10157 class. This does not check for subtypes; use C<sv_derived_from> to verify
10158 an inheritance relationship.
10164 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10166 const char *hvname;
10168 PERL_ARGS_ASSERT_SV_ISA;
10178 hvname = HvNAME_get(SvSTASH(sv));
10182 return strEQ(hvname, name);
10186 =for apidoc newSVrv
10188 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10189 RV then it will be upgraded to one. If C<classname> is non-null then the new
10190 SV will be blessed in the specified package. The new SV is returned and its
10191 reference count is 1. The reference count 1 is owned by C<rv>.
10197 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10201 PERL_ARGS_ASSERT_NEWSVRV;
10205 SV_CHECK_THINKFIRST_COW_DROP(rv);
10207 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10208 const U32 refcnt = SvREFCNT(rv);
10212 SvREFCNT(rv) = refcnt;
10214 sv_upgrade(rv, SVt_IV);
10215 } else if (SvROK(rv)) {
10216 SvREFCNT_dec(SvRV(rv));
10218 prepare_SV_for_RV(rv);
10226 HV* const stash = gv_stashpv(classname, GV_ADD);
10227 (void)sv_bless(rv, stash);
10233 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10235 SV * const lv = newSV_type(SVt_PVLV);
10236 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10238 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10239 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10240 LvSTARGOFF(lv) = ix;
10241 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10246 =for apidoc sv_setref_pv
10248 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10249 argument will be upgraded to an RV. That RV will be modified to point to
10250 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10251 into the SV. The C<classname> argument indicates the package for the
10252 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10253 will have a reference count of 1, and the RV will be returned.
10255 Do not use with other Perl types such as HV, AV, SV, CV, because those
10256 objects will become corrupted by the pointer copy process.
10258 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10264 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10266 PERL_ARGS_ASSERT_SV_SETREF_PV;
10269 sv_setsv(rv, &PL_sv_undef);
10273 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10278 =for apidoc sv_setref_iv
10280 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10281 argument will be upgraded to an RV. That RV will be modified to point to
10282 the new SV. The C<classname> argument indicates the package for the
10283 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10284 will have a reference count of 1, and the RV will be returned.
10290 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10292 PERL_ARGS_ASSERT_SV_SETREF_IV;
10294 sv_setiv(newSVrv(rv,classname), iv);
10299 =for apidoc sv_setref_uv
10301 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10302 argument will be upgraded to an RV. That RV will be modified to point to
10303 the new SV. The C<classname> argument indicates the package for the
10304 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10305 will have a reference count of 1, and the RV will be returned.
10311 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10313 PERL_ARGS_ASSERT_SV_SETREF_UV;
10315 sv_setuv(newSVrv(rv,classname), uv);
10320 =for apidoc sv_setref_nv
10322 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10323 argument will be upgraded to an RV. That RV will be modified to point to
10324 the new SV. The C<classname> argument indicates the package for the
10325 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10326 will have a reference count of 1, and the RV will be returned.
10332 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10334 PERL_ARGS_ASSERT_SV_SETREF_NV;
10336 sv_setnv(newSVrv(rv,classname), nv);
10341 =for apidoc sv_setref_pvn
10343 Copies a string into a new SV, optionally blessing the SV. The length of the
10344 string must be specified with C<n>. The C<rv> argument will be upgraded to
10345 an RV. That RV will be modified to point to the new SV. The C<classname>
10346 argument indicates the package for the blessing. Set C<classname> to
10347 C<NULL> to avoid the blessing. The new SV will have a reference count
10348 of 1, and the RV will be returned.
10350 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10356 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10357 const char *const pv, const STRLEN n)
10359 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10361 sv_setpvn(newSVrv(rv,classname), pv, n);
10366 =for apidoc sv_bless
10368 Blesses an SV into a specified package. The SV must be an RV. The package
10369 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10370 of the SV is unaffected.
10376 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10379 HV *oldstash = NULL;
10381 PERL_ARGS_ASSERT_SV_BLESS;
10385 Perl_croak(aTHX_ "Can't bless non-reference value");
10387 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10388 if (SvREADONLY(tmpRef))
10389 Perl_croak_no_modify();
10390 if (SvOBJECT(tmpRef)) {
10391 oldstash = SvSTASH(tmpRef);
10394 SvOBJECT_on(tmpRef);
10395 SvUPGRADE(tmpRef, SVt_PVMG);
10396 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10397 SvREFCNT_dec(oldstash);
10399 if(SvSMAGICAL(tmpRef))
10400 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10408 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10409 * as it is after unglobbing it.
10412 PERL_STATIC_INLINE void
10413 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10417 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10419 PERL_ARGS_ASSERT_SV_UNGLOB;
10421 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10423 if (!(flags & SV_COW_DROP_PV))
10424 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10426 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10428 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10429 && HvNAME_get(stash))
10430 mro_method_changed_in(stash);
10431 gp_free(MUTABLE_GV(sv));
10434 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10435 GvSTASH(sv) = NULL;
10438 if (GvNAME_HEK(sv)) {
10439 unshare_hek(GvNAME_HEK(sv));
10441 isGV_with_GP_off(sv);
10443 if(SvTYPE(sv) == SVt_PVGV) {
10444 /* need to keep SvANY(sv) in the right arena */
10445 xpvmg = new_XPVMG();
10446 StructCopy(SvANY(sv), xpvmg, XPVMG);
10447 del_XPVGV(SvANY(sv));
10450 SvFLAGS(sv) &= ~SVTYPEMASK;
10451 SvFLAGS(sv) |= SVt_PVMG;
10454 /* Intentionally not calling any local SET magic, as this isn't so much a
10455 set operation as merely an internal storage change. */
10456 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10457 else sv_setsv_flags(sv, temp, 0);
10459 if ((const GV *)sv == PL_last_in_gv)
10460 PL_last_in_gv = NULL;
10461 else if ((const GV *)sv == PL_statgv)
10466 =for apidoc sv_unref_flags
10468 Unsets the RV status of the SV, and decrements the reference count of
10469 whatever was being referenced by the RV. This can almost be thought of
10470 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10471 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10472 (otherwise the decrementing is conditional on the reference count being
10473 different from one or the reference being a readonly SV).
10474 See C<L</SvROK_off>>.
10480 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10482 SV* const target = SvRV(ref);
10484 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10486 if (SvWEAKREF(ref)) {
10487 sv_del_backref(target, ref);
10488 SvWEAKREF_off(ref);
10489 SvRV_set(ref, NULL);
10492 SvRV_set(ref, NULL);
10494 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10495 assigned to as BEGIN {$a = \"Foo"} will fail. */
10496 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10497 SvREFCNT_dec_NN(target);
10498 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10499 sv_2mortal(target); /* Schedule for freeing later */
10503 =for apidoc sv_untaint
10505 Untaint an SV. Use C<SvTAINTED_off> instead.
10511 Perl_sv_untaint(pTHX_ SV *const sv)
10513 PERL_ARGS_ASSERT_SV_UNTAINT;
10514 PERL_UNUSED_CONTEXT;
10516 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10517 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10524 =for apidoc sv_tainted
10526 Test an SV for taintedness. Use C<SvTAINTED> instead.
10532 Perl_sv_tainted(pTHX_ SV *const sv)
10534 PERL_ARGS_ASSERT_SV_TAINTED;
10535 PERL_UNUSED_CONTEXT;
10537 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10538 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10539 if (mg && (mg->mg_len & 1) )
10545 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10546 private to this file */
10549 =for apidoc sv_setpviv
10551 Copies an integer into the given SV, also updating its string value.
10552 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10558 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10560 char buf[TYPE_CHARS(UV)];
10562 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10564 PERL_ARGS_ASSERT_SV_SETPVIV;
10566 sv_setpvn(sv, ptr, ebuf - ptr);
10570 =for apidoc sv_setpviv_mg
10572 Like C<sv_setpviv>, but also handles 'set' magic.
10578 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10580 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10582 sv_setpviv(sv, iv);
10586 #endif /* NO_MATHOMS */
10588 #if defined(PERL_IMPLICIT_CONTEXT)
10590 /* pTHX_ magic can't cope with varargs, so this is a no-context
10591 * version of the main function, (which may itself be aliased to us).
10592 * Don't access this version directly.
10596 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10601 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10603 va_start(args, pat);
10604 sv_vsetpvf(sv, pat, &args);
10608 /* pTHX_ magic can't cope with varargs, so this is a no-context
10609 * version of the main function, (which may itself be aliased to us).
10610 * Don't access this version directly.
10614 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10619 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10621 va_start(args, pat);
10622 sv_vsetpvf_mg(sv, pat, &args);
10628 =for apidoc sv_setpvf
10630 Works like C<sv_catpvf> but copies the text into the SV instead of
10631 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10637 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10641 PERL_ARGS_ASSERT_SV_SETPVF;
10643 va_start(args, pat);
10644 sv_vsetpvf(sv, pat, &args);
10649 =for apidoc sv_vsetpvf
10651 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10652 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10654 Usually used via its frontend C<sv_setpvf>.
10660 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10662 PERL_ARGS_ASSERT_SV_VSETPVF;
10664 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10668 =for apidoc sv_setpvf_mg
10670 Like C<sv_setpvf>, but also handles 'set' magic.
10676 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10680 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10682 va_start(args, pat);
10683 sv_vsetpvf_mg(sv, pat, &args);
10688 =for apidoc sv_vsetpvf_mg
10690 Like C<sv_vsetpvf>, but also handles 'set' magic.
10692 Usually used via its frontend C<sv_setpvf_mg>.
10698 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10700 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10702 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10706 #if defined(PERL_IMPLICIT_CONTEXT)
10708 /* pTHX_ magic can't cope with varargs, so this is a no-context
10709 * version of the main function, (which may itself be aliased to us).
10710 * Don't access this version directly.
10714 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10719 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10721 va_start(args, pat);
10722 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10726 /* pTHX_ magic can't cope with varargs, so this is a no-context
10727 * version of the main function, (which may itself be aliased to us).
10728 * Don't access this version directly.
10732 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10737 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10739 va_start(args, pat);
10740 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10747 =for apidoc sv_catpvf
10749 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10750 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10751 variable argument list, argument reordering is not supported.
10752 If the appended data contains "wide" characters
10753 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10754 and characters >255 formatted with C<%c>), the original SV might get
10755 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10756 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10757 valid UTF-8; if the original SV was bytes, the pattern should be too.
10762 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10766 PERL_ARGS_ASSERT_SV_CATPVF;
10768 va_start(args, pat);
10769 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10774 =for apidoc sv_vcatpvf
10776 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10777 variable argument list, and appends the formatted output
10778 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10780 Usually used via its frontend C<sv_catpvf>.
10786 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10788 PERL_ARGS_ASSERT_SV_VCATPVF;
10790 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10794 =for apidoc sv_catpvf_mg
10796 Like C<sv_catpvf>, but also handles 'set' magic.
10802 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10806 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10808 va_start(args, pat);
10809 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10815 =for apidoc sv_vcatpvf_mg
10817 Like C<sv_vcatpvf>, but also handles 'set' magic.
10819 Usually used via its frontend C<sv_catpvf_mg>.
10825 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10827 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10829 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10834 =for apidoc sv_vsetpvfn
10836 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10839 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10845 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10846 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10848 PERL_ARGS_ASSERT_SV_VSETPVFN;
10851 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10856 * Warn of missing argument to sprintf. The value used in place of such
10857 * arguments should be &PL_sv_no; an undefined value would yield
10858 * inappropriate "use of uninit" warnings [perl #71000].
10861 S_warn_vcatpvfn_missing_argument(pTHX) {
10862 if (ckWARN(WARN_MISSING)) {
10863 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10864 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10870 S_expect_number(pTHX_ char **const pattern)
10874 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10876 switch (**pattern) {
10877 case '1': case '2': case '3':
10878 case '4': case '5': case '6':
10879 case '7': case '8': case '9':
10880 var = *(*pattern)++ - '0';
10881 while (isDIGIT(**pattern)) {
10882 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10884 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10892 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10894 const int neg = nv < 0;
10897 PERL_ARGS_ASSERT_F0CONVERT;
10899 if (UNLIKELY(Perl_isinfnan(nv))) {
10900 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10910 if (uv & 1 && uv == nv)
10911 uv--; /* Round to even */
10913 const unsigned dig = uv % 10;
10915 } while (uv /= 10);
10926 =for apidoc sv_vcatpvfn
10928 =for apidoc sv_vcatpvfn_flags
10930 Processes its arguments like C<vsprintf> and appends the formatted output
10931 to an SV. Uses an array of SVs if the C-style variable argument list is
10932 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10933 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10934 C<va_list> argument list with a format string that uses argument reordering
10935 will yield an exception.
10937 When running with taint checks enabled, indicates via
10938 C<maybe_tainted> if results are untrustworthy (often due to the use of
10941 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10943 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10948 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10949 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10950 vec_utf8 = DO_UTF8(vecsv);
10952 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10955 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10956 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10958 PERL_ARGS_ASSERT_SV_VCATPVFN;
10960 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10963 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10964 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10965 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10966 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10967 * after the first 1023 zero bits.
10969 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10970 * of dynamically growing buffer might be better, start at just 16 bytes
10971 * (for example) and grow only when necessary. Or maybe just by looking
10972 * at the exponents of the two doubles? */
10973 # define DOUBLEDOUBLE_MAXBITS 2098
10976 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10977 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10978 * per xdigit. For the double-double case, this can be rather many.
10979 * The non-double-double-long-double overshoots since all bits of NV
10980 * are not mantissa bits, there are also exponent bits. */
10981 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10982 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10984 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10987 /* If we do not have a known long double format, (including not using
10988 * long doubles, or long doubles being equal to doubles) then we will
10989 * fall back to the ldexp/frexp route, with which we can retrieve at
10990 * most as many bits as our widest unsigned integer type is. We try
10991 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10993 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10994 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10996 #if defined(HAS_QUAD) && defined(Uquad_t)
10997 # define MANTISSATYPE Uquad_t
10998 # define MANTISSASIZE 8
11000 # define MANTISSATYPE UV
11001 # define MANTISSASIZE UVSIZE
11004 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11005 # define HEXTRACT_LITTLE_ENDIAN
11006 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11007 # define HEXTRACT_BIG_ENDIAN
11009 # define HEXTRACT_MIX_ENDIAN
11012 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
11013 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11014 * are being extracted from (either directly from the long double in-memory
11015 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11016 * is used to update the exponent. The subnormal is set to true
11017 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11018 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11020 * The tricky part is that S_hextract() needs to be called twice:
11021 * the first time with vend as NULL, and the second time with vend as
11022 * the pointer returned by the first call. What happens is that on
11023 * the first round the output size is computed, and the intended
11024 * extraction sanity checked. On the second round the actual output
11025 * (the extraction of the hexadecimal values) takes place.
11026 * Sanity failures cause fatal failures during both rounds. */
11028 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11029 U8* vhex, U8* vend)
11033 int ixmin = 0, ixmax = 0;
11035 /* XXX Inf/NaN are not handled here, since it is
11036 * assumed they are to be output as "Inf" and "NaN". */
11038 /* These macros are just to reduce typos, they have multiple
11039 * repetitions below, but usually only one (or sometimes two)
11040 * of them is really being used. */
11041 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11042 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11043 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11044 #define HEXTRACT_OUTPUT(ix) \
11046 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11048 #define HEXTRACT_COUNT(ix, c) \
11050 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11052 #define HEXTRACT_BYTE(ix) \
11054 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11056 #define HEXTRACT_LO_NYBBLE(ix) \
11058 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11060 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11061 * to make it look less odd when the top bits of a NV
11062 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11063 * order bits can be in the "low nybble" of a byte. */
11064 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11065 #define HEXTRACT_BYTES_LE(a, b) \
11066 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11067 #define HEXTRACT_BYTES_BE(a, b) \
11068 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11069 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11070 #define HEXTRACT_IMPLICIT_BIT(nv) \
11072 if (!*subnormal) { \
11073 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11077 /* Most formats do. Those which don't should undef this.
11079 * But also note that IEEE 754 subnormals do not have it, or,
11080 * expressed alternatively, their implicit bit is zero. */
11081 #define HEXTRACT_HAS_IMPLICIT_BIT
11083 /* Many formats do. Those which don't should undef this. */
11084 #define HEXTRACT_HAS_TOP_NYBBLE
11086 /* HEXTRACTSIZE is the maximum number of xdigits. */
11087 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11088 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11090 # define HEXTRACTSIZE 2 * NVSIZE
11093 const U8* vmaxend = vhex + HEXTRACTSIZE;
11094 PERL_UNUSED_VAR(ix); /* might happen */
11095 (void)Perl_frexp(PERL_ABS(nv), exponent);
11096 *subnormal = FALSE;
11097 if (vend && (vend <= vhex || vend > vmaxend)) {
11098 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11099 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11102 /* First check if using long doubles. */
11103 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11104 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11105 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11106 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11107 /* The bytes 13..0 are the mantissa/fraction,
11108 * the 15,14 are the sign+exponent. */
11109 const U8* nvp = (const U8*)(&nv);
11110 HEXTRACT_GET_SUBNORMAL(nv);
11111 HEXTRACT_IMPLICIT_BIT(nv);
11112 # undef HEXTRACT_HAS_TOP_NYBBLE
11113 HEXTRACT_BYTES_LE(13, 0);
11114 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11115 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11116 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11117 /* The bytes 2..15 are the mantissa/fraction,
11118 * the 0,1 are the sign+exponent. */
11119 const U8* nvp = (const U8*)(&nv);
11120 HEXTRACT_GET_SUBNORMAL(nv);
11121 HEXTRACT_IMPLICIT_BIT(nv);
11122 # undef HEXTRACT_HAS_TOP_NYBBLE
11123 HEXTRACT_BYTES_BE(2, 15);
11124 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11125 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11126 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11127 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11128 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11129 /* The bytes 0..1 are the sign+exponent,
11130 * the bytes 2..9 are the mantissa/fraction. */
11131 const U8* nvp = (const U8*)(&nv);
11132 # undef HEXTRACT_HAS_IMPLICIT_BIT
11133 # undef HEXTRACT_HAS_TOP_NYBBLE
11134 HEXTRACT_GET_SUBNORMAL(nv);
11135 HEXTRACT_BYTES_LE(7, 0);
11136 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11137 /* Does this format ever happen? (Wikipedia says the Motorola
11138 * 6888x math coprocessors used format _like_ this but padded
11139 * to 96 bits with 16 unused bits between the exponent and the
11141 const U8* nvp = (const U8*)(&nv);
11142 # undef HEXTRACT_HAS_IMPLICIT_BIT
11143 # undef HEXTRACT_HAS_TOP_NYBBLE
11144 HEXTRACT_GET_SUBNORMAL(nv);
11145 HEXTRACT_BYTES_BE(0, 7);
11147 # define HEXTRACT_FALLBACK
11148 /* Double-double format: two doubles next to each other.
11149 * The first double is the high-order one, exactly like
11150 * it would be for a "lone" double. The second double
11151 * is shifted down using the exponent so that that there
11152 * are no common bits. The tricky part is that the value
11153 * of the double-double is the SUM of the two doubles and
11154 * the second one can be also NEGATIVE.
11156 * Because of this tricky construction the bytewise extraction we
11157 * use for the other long double formats doesn't work, we must
11158 * extract the values bit by bit.
11160 * The little-endian double-double is used .. somewhere?
11162 * The big endian double-double is used in e.g. PPC/Power (AIX)
11165 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11166 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11167 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11170 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11171 /* Using normal doubles, not long doubles.
11173 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11174 * bytes, since we might need to handle printf precision, and
11175 * also need to insert the radix. */
11177 # ifdef HEXTRACT_LITTLE_ENDIAN
11178 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11179 const U8* nvp = (const U8*)(&nv);
11180 HEXTRACT_GET_SUBNORMAL(nv);
11181 HEXTRACT_IMPLICIT_BIT(nv);
11182 HEXTRACT_TOP_NYBBLE(6);
11183 HEXTRACT_BYTES_LE(5, 0);
11184 # elif defined(HEXTRACT_BIG_ENDIAN)
11185 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11186 const U8* nvp = (const U8*)(&nv);
11187 HEXTRACT_GET_SUBNORMAL(nv);
11188 HEXTRACT_IMPLICIT_BIT(nv);
11189 HEXTRACT_TOP_NYBBLE(1);
11190 HEXTRACT_BYTES_BE(2, 7);
11191 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11192 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11193 const U8* nvp = (const U8*)(&nv);
11194 HEXTRACT_GET_SUBNORMAL(nv);
11195 HEXTRACT_IMPLICIT_BIT(nv);
11196 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11197 HEXTRACT_BYTE(1); /* 5 */
11198 HEXTRACT_BYTE(0); /* 4 */
11199 HEXTRACT_BYTE(7); /* 3 */
11200 HEXTRACT_BYTE(6); /* 2 */
11201 HEXTRACT_BYTE(5); /* 1 */
11202 HEXTRACT_BYTE(4); /* 0 */
11203 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11204 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11205 const U8* nvp = (const U8*)(&nv);
11206 HEXTRACT_GET_SUBNORMAL(nv);
11207 HEXTRACT_IMPLICIT_BIT(nv);
11208 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11209 HEXTRACT_BYTE(6); /* 5 */
11210 HEXTRACT_BYTE(7); /* 4 */
11211 HEXTRACT_BYTE(0); /* 3 */
11212 HEXTRACT_BYTE(1); /* 2 */
11213 HEXTRACT_BYTE(2); /* 1 */
11214 HEXTRACT_BYTE(3); /* 0 */
11216 # define HEXTRACT_FALLBACK
11219 # define HEXTRACT_FALLBACK
11221 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11222 # ifdef HEXTRACT_FALLBACK
11223 HEXTRACT_GET_SUBNORMAL(nv);
11224 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11225 /* The fallback is used for the double-double format, and
11226 * for unknown long double formats, and for unknown double
11227 * formats, or in general unknown NV formats. */
11228 if (nv == (NV)0.0) {
11236 NV d = nv < 0 ? -nv : nv;
11238 U8 ha = 0x0; /* hexvalue accumulator */
11239 U8 hd = 0x8; /* hexvalue digit */
11241 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11242 * this is essentially manual frexp(). Multiplying by 0.5 and
11243 * doubling should be lossless in binary floating point. */
11253 while (d >= e + e) {
11257 /* Now e <= d < 2*e */
11259 /* First extract the leading hexdigit (the implicit bit). */
11275 /* Then extract the remaining hexdigits. */
11276 while (d > (NV)0.0) {
11282 /* Output or count in groups of four bits,
11283 * that is, when the hexdigit is down to one. */
11288 /* Reset the hexvalue. */
11297 /* Flush possible pending hexvalue. */
11307 /* Croak for various reasons: if the output pointer escaped the
11308 * output buffer, if the extraction index escaped the extraction
11309 * buffer, or if the ending output pointer didn't match the
11310 * previously computed value. */
11311 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11312 /* For double-double the ixmin and ixmax stay at zero,
11313 * which is convenient since the HEXTRACTSIZE is tricky
11314 * for double-double. */
11315 ixmin < 0 || ixmax >= NVSIZE ||
11316 (vend && v != vend)) {
11317 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11318 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11323 /* Helper for sv_vcatpvfn_flags(). */
11324 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11329 (var) = &PL_sv_no; /* [perl #71000] */ \
11330 arg_missing = TRUE; \
11335 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11336 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11341 const char *patend;
11344 static const char nullstr[] = "(null)";
11346 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11347 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11349 /* Times 4: a decimal digit takes more than 3 binary digits.
11350 * NV_DIG: mantissa takes than many decimal digits.
11351 * Plus 32: Playing safe. */
11352 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11353 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11354 bool hexfp = FALSE; /* hexadecimal floating point? */
11356 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11358 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11359 PERL_UNUSED_ARG(maybe_tainted);
11361 if (flags & SV_GMAGIC)
11364 /* no matter what, this is a string now */
11365 (void)SvPV_force_nomg(sv, origlen);
11367 /* special-case "", "%s", and "%-p" (SVf - see below) */
11369 if (svmax && ckWARN(WARN_REDUNDANT))
11370 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11371 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11374 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11375 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11376 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11377 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11380 const char * const s = va_arg(*args, char*);
11381 sv_catpv_nomg(sv, s ? s : nullstr);
11383 else if (svix < svmax) {
11384 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11385 SvGETMAGIC(*svargs);
11386 sv_catsv_nomg(sv, *svargs);
11389 S_warn_vcatpvfn_missing_argument(aTHX);
11392 if (args && patlen == 3 && pat[0] == '%' &&
11393 pat[1] == '-' && pat[2] == 'p') {
11394 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11395 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11396 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11397 argsv = MUTABLE_SV(va_arg(*args, void*));
11398 sv_catsv_nomg(sv, argsv);
11402 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11403 /* special-case "%.<number>[gf]" */
11404 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11405 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11406 unsigned digits = 0;
11410 while (*pp >= '0' && *pp <= '9')
11411 digits = 10 * digits + (*pp++ - '0');
11413 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11414 format the first argument and WARN_REDUNDANT if svmax > 1?
11415 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11416 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11417 const NV nv = SvNV(*svargs);
11418 if (LIKELY(!Perl_isinfnan(nv))) {
11420 /* Add check for digits != 0 because it seems that some
11421 gconverts are buggy in this case, and we don't yet have
11422 a Configure test for this. */
11423 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11424 /* 0, point, slack */
11425 STORE_LC_NUMERIC_SET_TO_NEEDED();
11426 SNPRINTF_G(nv, ebuf, size, digits);
11427 sv_catpv_nomg(sv, ebuf);
11428 if (*ebuf) /* May return an empty string for digits==0 */
11431 } else if (!digits) {
11434 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11435 sv_catpvn_nomg(sv, p, l);
11442 #endif /* !USE_LONG_DOUBLE */
11444 if (!args && svix < svmax && DO_UTF8(*svargs))
11447 patend = (char*)pat + patlen;
11448 for (p = (char*)pat; p < patend; p = q) {
11451 bool vectorize = FALSE;
11452 bool vectorarg = FALSE;
11453 bool vec_utf8 = FALSE;
11459 bool has_precis = FALSE;
11461 const I32 osvix = svix;
11462 bool is_utf8 = FALSE; /* is this item utf8? */
11463 bool used_explicit_ix = FALSE;
11464 bool arg_missing = FALSE;
11465 #ifdef HAS_LDBL_SPRINTF_BUG
11466 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11467 with sfio - Allen <allens@cpan.org> */
11468 bool fix_ldbl_sprintf_bug = FALSE;
11472 U8 utf8buf[UTF8_MAXBYTES+1];
11473 STRLEN esignlen = 0;
11475 const char *eptr = NULL;
11476 const char *fmtstart;
11479 const U8 *vecstr = NULL;
11486 /* We need a long double target in case HAS_LONG_DOUBLE,
11487 * even without USE_LONG_DOUBLE, so that we can printf with
11488 * long double formats, even without NV being long double.
11489 * But we call the target 'fv' instead of 'nv', since most of
11490 * the time it is not (most compilers these days recognize
11491 * "long double", even if only as a synonym for "double").
11493 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11494 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11496 # ifdef Perl_isfinitel
11497 # define FV_ISFINITE(x) Perl_isfinitel(x)
11499 # define FV_GF PERL_PRIgldbl
11500 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11501 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11502 # define NV_TO_FV(nv,fv) STMT_START { \
11504 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11507 # define NV_TO_FV(nv,fv) (fv)=(nv)
11511 # define FV_GF NVgf
11512 # define NV_TO_FV(nv,fv) (fv)=(nv)
11514 #ifndef FV_ISFINITE
11515 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11521 const char *dotstr = ".";
11522 STRLEN dotstrlen = 1;
11523 I32 efix = 0; /* explicit format parameter index */
11524 I32 ewix = 0; /* explicit width index */
11525 I32 epix = 0; /* explicit precision index */
11526 I32 evix = 0; /* explicit vector index */
11527 bool asterisk = FALSE;
11528 bool infnan = FALSE;
11530 /* echo everything up to the next format specification */
11531 for (q = p; q < patend && *q != '%'; ++q) ;
11533 if (has_utf8 && !pat_utf8)
11534 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11536 sv_catpvn_nomg(sv, p, q - p);
11545 We allow format specification elements in this order:
11546 \d+\$ explicit format parameter index
11548 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11549 0 flag (as above): repeated to allow "v02"
11550 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11551 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11553 [%bcdefginopsuxDFOUX] format (mandatory)
11558 As of perl5.9.3, printf format checking is on by default.
11559 Internally, perl uses %p formats to provide an escape to
11560 some extended formatting. This block deals with those
11561 extensions: if it does not match, (char*)q is reset and
11562 the normal format processing code is used.
11564 Currently defined extensions are:
11565 %p include pointer address (standard)
11566 %-p (SVf) include an SV (previously %_)
11567 %-<num>p include an SV with precision <num>
11569 %3p include a HEK with precision of 256
11570 %4p char* preceded by utf8 flag and length
11571 %<num>p (where num is 1 or > 4) reserved for future
11574 Robin Barker 2005-07-14 (but modified since)
11576 %1p (VDf) removed. RMB 2007-10-19
11583 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11584 /* The argument has already gone through cBOOL, so the cast
11586 is_utf8 = (bool)va_arg(*args, int);
11587 elen = va_arg(*args, UV);
11588 /* if utf8 length is larger than 0x7ffff..., then it might
11589 * have been a signed value that wrapped */
11590 if (elen > ((~(STRLEN)0) >> 1)) {
11591 assert(0); /* in DEBUGGING build we want to crash */
11592 elen= 0; /* otherwise we want to treat this as an empty string */
11594 eptr = va_arg(*args, char *);
11595 q += sizeof(UTF8f)-1;
11598 n = expect_number(&q);
11600 if (sv) { /* SVf */
11605 argsv = MUTABLE_SV(va_arg(*args, void*));
11606 eptr = SvPV_const(argsv, elen);
11607 if (DO_UTF8(argsv))
11611 else if (n==2 || n==3) { /* HEKf */
11612 HEK * const hek = va_arg(*args, HEK *);
11613 eptr = HEK_KEY(hek);
11614 elen = HEK_LEN(hek);
11615 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11616 if (n==3) precis = 256, has_precis = TRUE;
11620 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11621 "internal %%<num>p might conflict with future printf extensions");
11627 if ( (width = expect_number(&q)) ) {
11630 Perl_croak_nocontext(
11631 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11634 used_explicit_ix = TRUE;
11646 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11675 if ( (ewix = expect_number(&q)) ) {
11678 Perl_croak_nocontext(
11679 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11680 used_explicit_ix = TRUE;
11690 if ((vectorarg = asterisk)) {
11703 width = expect_number(&q);
11706 if (vectorize && vectorarg) {
11707 /* vectorizing, but not with the default "." */
11709 vecsv = va_arg(*args, SV*);
11711 FETCH_VCATPVFN_ARGUMENT(
11712 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11714 FETCH_VCATPVFN_ARGUMENT(
11715 vecsv, svix < svmax, svargs[svix++]);
11717 dotstr = SvPV_const(vecsv, dotstrlen);
11718 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11719 bad with tied or overloaded values that return UTF8. */
11720 if (DO_UTF8(vecsv))
11722 else if (has_utf8) {
11723 vecsv = sv_mortalcopy(vecsv);
11724 sv_utf8_upgrade(vecsv);
11725 dotstr = SvPV_const(vecsv, dotstrlen);
11732 i = va_arg(*args, int);
11734 i = (ewix ? ewix <= svmax : svix < svmax) ?
11735 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11737 width = (i < 0) ? -i : i;
11747 if ( (epix = expect_number(&q)) ) {
11750 Perl_croak_nocontext(
11751 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11752 used_explicit_ix = TRUE;
11757 i = va_arg(*args, int);
11761 FETCH_VCATPVFN_ARGUMENT(
11762 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11764 FETCH_VCATPVFN_ARGUMENT(
11765 precsv, svix < svmax, svargs[svix++]);
11766 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11769 has_precis = !(i < 0);
11773 while (isDIGIT(*q))
11774 precis = precis * 10 + (*q++ - '0');
11783 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11784 vecsv = svargs[efix ? efix-1 : svix++];
11785 vecstr = (U8*)SvPV_const(vecsv,veclen);
11786 vec_utf8 = DO_UTF8(vecsv);
11788 /* if this is a version object, we need to convert
11789 * back into v-string notation and then let the
11790 * vectorize happen normally
11792 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11793 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
11794 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11795 "vector argument not supported with alpha versions");
11798 vecsv = sv_newmortal();
11799 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11801 vecstr = (U8*)SvPV_const(vecsv, veclen);
11802 vec_utf8 = DO_UTF8(vecsv);
11816 case 'I': /* Ix, I32x, and I64x */
11817 # ifdef USE_64_BIT_INT
11818 if (q[1] == '6' && q[2] == '4') {
11824 if (q[1] == '3' && q[2] == '2') {
11828 # ifdef USE_64_BIT_INT
11834 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11835 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11838 # ifdef USE_QUADMATH
11851 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11852 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11853 if (*q == 'l') { /* lld, llf */
11862 if (*++q == 'h') { /* hhd, hhu */
11891 if (!vectorize && !args) {
11893 const I32 i = efix-1;
11894 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11896 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11901 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11902 /* XXX va_arg(*args) case? need peek, use va_copy? */
11904 if (UNLIKELY(SvAMAGIC(argsv)))
11905 argsv = sv_2num(argsv);
11906 infnan = UNLIKELY(isinfnansv(argsv));
11909 switch (c = *q++) {
11917 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11918 /* no va_arg() case */
11919 SvNV_nomg(argsv), (int)c);
11920 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11922 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11924 eptr = (char*)utf8buf;
11925 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11939 eptr = va_arg(*args, char*);
11941 elen = strlen(eptr);
11943 eptr = (char *)nullstr;
11944 elen = sizeof nullstr - 1;
11948 eptr = SvPV_const(argsv, elen);
11949 if (DO_UTF8(argsv)) {
11950 STRLEN old_precis = precis;
11951 if (has_precis && precis < elen) {
11952 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11953 STRLEN p = precis > ulen ? ulen : precis;
11954 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11955 /* sticks at end */
11957 if (width) { /* fudge width (can't fudge elen) */
11958 if (has_precis && precis < elen)
11959 width += precis - old_precis;
11962 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11969 if (has_precis && precis < elen)
11977 goto floating_point;
11979 if (alt || vectorize)
11981 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11995 goto floating_point;
12000 goto donevalidconversion;
12002 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12011 esignbuf[esignlen++] = plus;
12015 case 'c': iv = (char)va_arg(*args, int); break;
12016 case 'h': iv = (short)va_arg(*args, int); break;
12017 case 'l': iv = va_arg(*args, long); break;
12018 case 'V': iv = va_arg(*args, IV); break;
12019 case 'z': iv = va_arg(*args, SSize_t); break;
12020 #ifdef HAS_PTRDIFF_T
12021 case 't': iv = va_arg(*args, ptrdiff_t); break;
12023 default: iv = va_arg(*args, int); break;
12025 case 'j': iv = va_arg(*args, intmax_t); break;
12029 iv = va_arg(*args, Quad_t); break;
12036 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12038 case 'c': iv = (char)tiv; break;
12039 case 'h': iv = (short)tiv; break;
12040 case 'l': iv = (long)tiv; break;
12042 default: iv = tiv; break;
12045 iv = (Quad_t)tiv; break;
12051 if ( !vectorize ) /* we already set uv above */
12056 esignbuf[esignlen++] = plus;
12059 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12060 esignbuf[esignlen++] = '-';
12099 goto floating_point;
12105 goto donevalidconversion;
12107 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12118 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12119 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12120 case 'l': uv = va_arg(*args, unsigned long); break;
12121 case 'V': uv = va_arg(*args, UV); break;
12122 case 'z': uv = va_arg(*args, Size_t); break;
12123 #ifdef HAS_PTRDIFF_T
12124 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12127 case 'j': uv = va_arg(*args, uintmax_t); break;
12129 default: uv = va_arg(*args, unsigned); break;
12132 uv = va_arg(*args, Uquad_t); break;
12139 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12141 case 'c': uv = (unsigned char)tuv; break;
12142 case 'h': uv = (unsigned short)tuv; break;
12143 case 'l': uv = (unsigned long)tuv; break;
12145 default: uv = tuv; break;
12148 uv = (Uquad_t)tuv; break;
12157 char *ptr = ebuf + sizeof ebuf;
12158 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12164 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12168 } while (uv >>= 4);
12170 esignbuf[esignlen++] = '0';
12171 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12177 *--ptr = '0' + dig;
12178 } while (uv >>= 3);
12179 if (alt && *ptr != '0')
12185 *--ptr = '0' + dig;
12186 } while (uv >>= 1);
12188 esignbuf[esignlen++] = '0';
12189 esignbuf[esignlen++] = c;
12192 default: /* it had better be ten or less */
12195 *--ptr = '0' + dig;
12196 } while (uv /= base);
12199 elen = (ebuf + sizeof ebuf) - ptr;
12203 zeros = precis - elen;
12204 else if (precis == 0 && elen == 1 && *eptr == '0'
12205 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12208 /* a precision nullifies the 0 flag. */
12215 /* FLOATING POINT */
12220 c = 'f'; /* maybe %F isn't supported here */
12222 case 'e': case 'E':
12224 case 'g': case 'G':
12225 case 'a': case 'A':
12229 /* This is evil, but floating point is even more evil */
12231 /* for SV-style calling, we can only get NV
12232 for C-style calling, we assume %f is double;
12233 for simplicity we allow any of %Lf, %llf, %qf for long double
12237 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12241 /* [perl #20339] - we should accept and ignore %lf rather than die */
12245 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12246 intsize = args ? 0 : 'q';
12250 #if defined(HAS_LONG_DOUBLE)
12263 /* Now we need (long double) if intsize == 'q', else (double). */
12265 /* Note: do not pull NVs off the va_list with va_arg()
12266 * (pull doubles instead) because if you have a build
12267 * with long doubles, you would always be pulling long
12268 * doubles, which would badly break anyone using only
12269 * doubles (i.e. the majority of builds). In other
12270 * words, you cannot mix doubles and long doubles.
12271 * The only case where you can pull off long doubles
12272 * is when the format specifier explicitly asks so with
12274 #ifdef USE_QUADMATH
12275 fv = intsize == 'q' ?
12276 va_arg(*args, NV) : va_arg(*args, double);
12278 #elif LONG_DOUBLESIZE > DOUBLESIZE
12279 if (intsize == 'q') {
12280 fv = va_arg(*args, long double);
12283 nv = va_arg(*args, double);
12287 nv = va_arg(*args, double);
12293 if (!infnan) SvGETMAGIC(argsv);
12294 nv = SvNV_nomg(argsv);
12299 /* frexp() (or frexpl) has some unspecified behaviour for
12300 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12301 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12303 (void)Perl_frexp((NV)fv, &i);
12304 if (i == PERL_INT_MIN)
12305 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12306 /* Do not set hexfp earlier since we want to printf
12307 * Inf/NaN for Inf/NaN, not their hexfp. */
12308 hexfp = isALPHA_FOLD_EQ(c, 'a');
12309 if (UNLIKELY(hexfp)) {
12310 /* This seriously overshoots in most cases, but
12311 * better the undershooting. Firstly, all bytes
12312 * of the NV are not mantissa, some of them are
12313 * exponent. Secondly, for the reasonably common
12314 * long doubles case, the "80-bit extended", two
12315 * or six bytes of the NV are unused. */
12317 (fv < 0) ? 1 : 0 + /* possible unary minus */
12319 1 + /* the very unlikely carry */
12322 2 * NVSIZE + /* 2 hexdigits for each byte */
12324 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12326 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12327 /* However, for the "double double", we need more.
12328 * Since each double has their own exponent, the
12329 * doubles may float (haha) rather far from each
12330 * other, and the number of required bits is much
12331 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12332 * See the definition of DOUBLEDOUBLE_MAXBITS.
12334 * Need 2 hexdigits for each byte. */
12335 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12336 /* the size for the exponent already added */
12338 #ifdef USE_LOCALE_NUMERIC
12339 STORE_LC_NUMERIC_SET_TO_NEEDED();
12340 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12341 need += SvLEN(PL_numeric_radix_sv);
12342 RESTORE_LC_NUMERIC();
12346 need = BIT_DIGITS(i);
12347 } /* if i < 0, the number of digits is hard to predict. */
12349 need += has_precis ? precis : 6; /* known default */
12354 #ifdef HAS_LDBL_SPRINTF_BUG
12355 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12356 with sfio - Allen <allens@cpan.org> */
12359 # define MY_DBL_MAX DBL_MAX
12360 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12361 # if DOUBLESIZE >= 8
12362 # define MY_DBL_MAX 1.7976931348623157E+308L
12364 # define MY_DBL_MAX 3.40282347E+38L
12368 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12369 # define MY_DBL_MAX_BUG 1L
12371 # define MY_DBL_MAX_BUG MY_DBL_MAX
12375 # define MY_DBL_MIN DBL_MIN
12376 # else /* XXX guessing! -Allen */
12377 # if DOUBLESIZE >= 8
12378 # define MY_DBL_MIN 2.2250738585072014E-308L
12380 # define MY_DBL_MIN 1.17549435E-38L
12384 if ((intsize == 'q') && (c == 'f') &&
12385 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12386 (need < DBL_DIG)) {
12387 /* it's going to be short enough that
12388 * long double precision is not needed */
12390 if ((fv <= 0L) && (fv >= -0L))
12391 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12393 /* would use Perl_fp_class as a double-check but not
12394 * functional on IRIX - see perl.h comments */
12396 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12397 /* It's within the range that a double can represent */
12398 #if defined(DBL_MAX) && !defined(DBL_MIN)
12399 if ((fv >= ((long double)1/DBL_MAX)) ||
12400 (fv <= (-(long double)1/DBL_MAX)))
12402 fix_ldbl_sprintf_bug = TRUE;
12405 if (fix_ldbl_sprintf_bug == TRUE) {
12415 # undef MY_DBL_MAX_BUG
12418 #endif /* HAS_LDBL_SPRINTF_BUG */
12420 need += 20; /* fudge factor */
12421 if (PL_efloatsize < need) {
12422 Safefree(PL_efloatbuf);
12423 PL_efloatsize = need + 20; /* more fudge */
12424 Newx(PL_efloatbuf, PL_efloatsize, char);
12425 PL_efloatbuf[0] = '\0';
12428 if ( !(width || left || plus || alt) && fill != '0'
12429 && has_precis && intsize != 'q' /* Shortcuts */
12430 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12431 /* See earlier comment about buggy Gconvert when digits,
12433 if ( c == 'g' && precis ) {
12434 STORE_LC_NUMERIC_SET_TO_NEEDED();
12435 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12436 /* May return an empty string for digits==0 */
12437 if (*PL_efloatbuf) {
12438 elen = strlen(PL_efloatbuf);
12439 goto float_converted;
12441 } else if ( c == 'f' && !precis ) {
12442 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12447 if (UNLIKELY(hexfp)) {
12448 /* Hexadecimal floating point. */
12449 char* p = PL_efloatbuf;
12450 U8 vhex[VHEX_SIZE];
12451 U8* v = vhex; /* working pointer to vhex */
12452 U8* vend; /* pointer to one beyond last digit of vhex */
12453 U8* vfnz = NULL; /* first non-zero */
12454 U8* vlnz = NULL; /* last non-zero */
12455 U8* v0 = NULL; /* first output */
12456 const bool lower = (c == 'a');
12457 /* At output the values of vhex (up to vend) will
12458 * be mapped through the xdig to get the actual
12459 * human-readable xdigits. */
12460 const char* xdig = PL_hexdigit;
12461 int zerotail = 0; /* how many extra zeros to append */
12462 int exponent = 0; /* exponent of the floating point input */
12463 bool hexradix = FALSE; /* should we output the radix */
12464 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12465 bool negative = FALSE;
12467 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12469 * For example with denormals, (assuming the vanilla
12470 * 64-bit double): the exponent is zero. 1xp-1074 is
12471 * the smallest denormal and the smallest double, it
12472 * could be output also as 0x0.0000000000001p-1022 to
12473 * match its internal structure. */
12475 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12476 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12478 #if NVSIZE > DOUBLESIZE
12479 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12480 /* In this case there is an implicit bit,
12481 * and therefore the exponent is shifted by one. */
12484 # ifdef NV_X86_80_BIT
12486 /* The subnormals of the x86-80 have a base exponent of -16382,
12487 * (while the physical exponent bits are zero) but the frexp()
12488 * returned the scientific-style floating exponent. We want
12489 * to map the last one as:
12490 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12491 * -16835..-16388 -> -16384
12492 * since we want to keep the first hexdigit
12493 * as one of the [8421]. */
12494 exponent = -4 * ( (exponent + 1) / -4) - 2;
12499 /* TBD: other non-implicit-bit platforms than the x86-80. */
12503 negative = fv < 0 || Perl_signbit(nv);
12514 xdig += 16; /* Use uppercase hex. */
12517 /* Find the first non-zero xdigit. */
12518 for (v = vhex; v < vend; v++) {
12526 /* Find the last non-zero xdigit. */
12527 for (v = vend - 1; v >= vhex; v--) {
12534 #if NVSIZE == DOUBLESIZE
12540 #ifndef NV_X86_80_BIT
12542 /* IEEE 754 subnormals (but not the x86 80-bit):
12543 * we want "normalize" the subnormal,
12544 * so we need to right shift the hex nybbles
12545 * so that the output of the subnormal starts
12546 * from the first true bit. (Another, equally
12547 * valid, policy would be to dump the subnormal
12548 * nybbles as-is, to display the "physical" layout.) */
12551 /* Find the ceil(log2(v[0])) of
12552 * the top non-zero nybble. */
12553 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12556 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12557 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12571 U8* ve = (subnormal ? vlnz + 1 : vend);
12572 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12573 if ((SSize_t)(precis + 1) < vn) {
12574 bool overflow = FALSE;
12575 if (v0[precis + 1] < 0x8) {
12576 /* Round down, nothing to do. */
12577 } else if (v0[precis + 1] > 0x8) {
12580 overflow = v0[precis] > 0xF;
12582 } else { /* v0[precis] == 0x8 */
12583 /* Half-point: round towards the one
12584 * with the even least-significant digit:
12592 * 78 -> 8 f8 -> 10 */
12593 if ((v0[precis] & 0x1)) {
12596 overflow = v0[precis] > 0xF;
12601 for (v = v0 + precis - 1; v >= v0; v--) {
12603 overflow = *v > 0xF;
12609 if (v == v0 - 1 && overflow) {
12610 /* If the overflow goes all the
12611 * way to the front, we need to
12612 * insert 0x1 in front, and adjust
12614 Move(v0, v0 + 1, vn, char);
12620 /* The new effective "last non zero". */
12621 vlnz = v0 + precis;
12625 subnormal ? precis - vn + 1 :
12626 precis - (vlnz - vhex);
12633 /* If there are non-zero xdigits, the radix
12634 * is output after the first one. */
12645 /* The radix is always output if precis, or if alt. */
12646 if (precis > 0 || alt) {
12651 #ifndef USE_LOCALE_NUMERIC
12654 STORE_LC_NUMERIC_SET_TO_NEEDED();
12655 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12657 const char* r = SvPV(PL_numeric_radix_sv, n);
12658 Copy(r, p, n, char);
12664 RESTORE_LC_NUMERIC();
12673 if (zerotail > 0) {
12674 while (zerotail--) {
12679 elen = p - PL_efloatbuf;
12680 elen += my_snprintf(p, PL_efloatsize - elen,
12681 "%c%+d", lower ? 'p' : 'P',
12684 if (elen < width) {
12686 /* Pad the back with spaces. */
12687 memset(PL_efloatbuf + elen, ' ', width - elen);
12689 else if (fill == '0') {
12690 /* Insert the zeros after the "0x" and the
12691 * the potential sign, but before the digits,
12692 * otherwise we end up with "0000xH.HHH...",
12693 * when we want "0x000H.HHH..." */
12694 STRLEN nzero = width - elen;
12695 char* zerox = PL_efloatbuf + 2;
12696 STRLEN nmove = elen - 2;
12697 if (negative || plus) {
12701 Move(zerox, zerox + nzero, nmove, char);
12702 memset(zerox, fill, nzero);
12705 /* Move it to the right. */
12706 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12708 /* Pad the front with spaces. */
12709 memset(PL_efloatbuf, ' ', width - elen);
12715 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12717 /* Not affecting infnan output: precision, alt, fill. */
12718 if (elen < width) {
12720 /* Pack the back with spaces. */
12721 memset(PL_efloatbuf + elen, ' ', width - elen);
12723 /* Move it to the right. */
12724 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12726 /* Pad the front with spaces. */
12727 memset(PL_efloatbuf, ' ', width - elen);
12735 char *ptr = ebuf + sizeof ebuf;
12738 #if defined(USE_QUADMATH)
12739 if (intsize == 'q') {
12743 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12744 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12745 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12746 * not USE_LONG_DOUBLE and NVff. In other words,
12747 * this needs to work without USE_LONG_DOUBLE. */
12748 if (intsize == 'q') {
12749 /* Copy the one or more characters in a long double
12750 * format before the 'base' ([efgEFG]) character to
12751 * the format string. */
12752 static char const ldblf[] = PERL_PRIfldbl;
12753 char const *p = ldblf + sizeof(ldblf) - 3;
12754 while (p >= ldblf) { *--ptr = *p--; }
12759 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12764 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12776 /* No taint. Otherwise we are in the strange situation
12777 * where printf() taints but print($float) doesn't.
12780 STORE_LC_NUMERIC_SET_TO_NEEDED();
12782 /* hopefully the above makes ptr a very constrained format
12783 * that is safe to use, even though it's not literal */
12784 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12785 #ifdef USE_QUADMATH
12787 const char* qfmt = quadmath_format_single(ptr);
12789 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12790 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12792 if ((IV)elen == -1)
12793 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12797 #elif defined(HAS_LONG_DOUBLE)
12798 elen = ((intsize == 'q')
12799 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12800 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12802 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12808 eptr = PL_efloatbuf;
12809 assert((IV)elen > 0); /* here zero elen is bad */
12811 #ifdef USE_LOCALE_NUMERIC
12812 /* If the decimal point character in the string is UTF-8, make the
12814 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12815 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12828 i = SvCUR(sv) - origlen;
12831 case 'c': *(va_arg(*args, char*)) = i; break;
12832 case 'h': *(va_arg(*args, short*)) = i; break;
12833 default: *(va_arg(*args, int*)) = i; break;
12834 case 'l': *(va_arg(*args, long*)) = i; break;
12835 case 'V': *(va_arg(*args, IV*)) = i; break;
12836 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12837 #ifdef HAS_PTRDIFF_T
12838 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12841 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12845 *(va_arg(*args, Quad_t*)) = i; break;
12852 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12853 goto donevalidconversion;
12860 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12861 && ckWARN(WARN_PRINTF))
12863 SV * const msg = sv_newmortal();
12864 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12865 (PL_op->op_type == OP_PRTF) ? "" : "s");
12866 if (fmtstart < patend) {
12867 const char * const fmtend = q < patend ? q : patend;
12869 sv_catpvs(msg, "\"%");
12870 for (f = fmtstart; f < fmtend; f++) {
12872 sv_catpvn_nomg(msg, f, 1);
12874 Perl_sv_catpvf(aTHX_ msg,
12875 "\\%03"UVof, (UV)*f & 0xFF);
12878 sv_catpvs(msg, "\"");
12880 sv_catpvs(msg, "end of string");
12882 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12885 /* output mangled stuff ... */
12891 /* ... right here, because formatting flags should not apply */
12892 SvGROW(sv, SvCUR(sv) + elen + 1);
12894 Copy(eptr, p, elen, char);
12897 SvCUR_set(sv, p - SvPVX_const(sv));
12899 continue; /* not "break" */
12902 if (is_utf8 != has_utf8) {
12905 sv_utf8_upgrade(sv);
12908 const STRLEN old_elen = elen;
12909 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12910 sv_utf8_upgrade(nsv);
12911 eptr = SvPVX_const(nsv);
12914 if (width) { /* fudge width (can't fudge elen) */
12915 width += elen - old_elen;
12921 /* signed value that's wrapped? */
12922 assert(elen <= ((~(STRLEN)0) >> 1));
12923 have = esignlen + zeros + elen;
12925 croak_memory_wrap();
12927 need = (have > width ? have : width);
12930 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12931 croak_memory_wrap();
12932 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12934 if (esignlen && fill == '0') {
12936 for (i = 0; i < (int)esignlen; i++)
12937 *p++ = esignbuf[i];
12939 if (gap && !left) {
12940 memset(p, fill, gap);
12943 if (esignlen && fill != '0') {
12945 for (i = 0; i < (int)esignlen; i++)
12946 *p++ = esignbuf[i];
12950 for (i = zeros; i; i--)
12954 Copy(eptr, p, elen, char);
12958 memset(p, ' ', gap);
12963 Copy(dotstr, p, dotstrlen, char);
12967 vectorize = FALSE; /* done iterating over vecstr */
12974 SvCUR_set(sv, p - SvPVX_const(sv));
12980 donevalidconversion:
12981 if (used_explicit_ix)
12982 no_redundant_warning = TRUE;
12984 S_warn_vcatpvfn_missing_argument(aTHX);
12987 /* Now that we've consumed all our printf format arguments (svix)
12988 * do we have things left on the stack that we didn't use?
12990 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12991 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12992 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12997 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13001 /* =========================================================================
13003 =head1 Cloning an interpreter
13007 All the macros and functions in this section are for the private use of
13008 the main function, perl_clone().
13010 The foo_dup() functions make an exact copy of an existing foo thingy.
13011 During the course of a cloning, a hash table is used to map old addresses
13012 to new addresses. The table is created and manipulated with the
13013 ptr_table_* functions.
13015 * =========================================================================*/
13018 #if defined(USE_ITHREADS)
13020 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13021 #ifndef GpREFCNT_inc
13022 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13026 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13027 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13028 If this changes, please unmerge ss_dup.
13029 Likewise, sv_dup_inc_multiple() relies on this fact. */
13030 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13031 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13032 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13033 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13034 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13035 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13036 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13037 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13038 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13039 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13040 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13041 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13042 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13044 /* clone a parser */
13047 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13051 PERL_ARGS_ASSERT_PARSER_DUP;
13056 /* look for it in the table first */
13057 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13061 /* create anew and remember what it is */
13062 Newxz(parser, 1, yy_parser);
13063 ptr_table_store(PL_ptr_table, proto, parser);
13065 /* XXX these not yet duped */
13066 parser->old_parser = NULL;
13067 parser->stack = NULL;
13069 parser->stack_size = 0;
13070 /* XXX parser->stack->state = 0; */
13072 /* XXX eventually, just Copy() most of the parser struct ? */
13074 parser->lex_brackets = proto->lex_brackets;
13075 parser->lex_casemods = proto->lex_casemods;
13076 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13077 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13078 parser->lex_casestack = savepvn(proto->lex_casestack,
13079 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13080 parser->lex_defer = proto->lex_defer;
13081 parser->lex_dojoin = proto->lex_dojoin;
13082 parser->lex_formbrack = proto->lex_formbrack;
13083 parser->lex_inpat = proto->lex_inpat;
13084 parser->lex_inwhat = proto->lex_inwhat;
13085 parser->lex_op = proto->lex_op;
13086 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13087 parser->lex_starts = proto->lex_starts;
13088 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13089 parser->multi_close = proto->multi_close;
13090 parser->multi_open = proto->multi_open;
13091 parser->multi_start = proto->multi_start;
13092 parser->multi_end = proto->multi_end;
13093 parser->preambled = proto->preambled;
13094 parser->lex_super_state = proto->lex_super_state;
13095 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13096 parser->lex_sub_op = proto->lex_sub_op;
13097 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13098 parser->linestr = sv_dup_inc(proto->linestr, param);
13099 parser->expect = proto->expect;
13100 parser->copline = proto->copline;
13101 parser->last_lop_op = proto->last_lop_op;
13102 parser->lex_state = proto->lex_state;
13103 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13104 /* rsfp_filters entries have fake IoDIRP() */
13105 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13106 parser->in_my = proto->in_my;
13107 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13108 parser->error_count = proto->error_count;
13109 parser->sig_elems = proto->sig_elems;
13110 parser->sig_optelems= proto->sig_optelems;
13111 parser->sig_slurpy = proto->sig_slurpy;
13112 parser->linestr = sv_dup_inc(proto->linestr, param);
13115 char * const ols = SvPVX(proto->linestr);
13116 char * const ls = SvPVX(parser->linestr);
13118 parser->bufptr = ls + (proto->bufptr >= ols ?
13119 proto->bufptr - ols : 0);
13120 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13121 proto->oldbufptr - ols : 0);
13122 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13123 proto->oldoldbufptr - ols : 0);
13124 parser->linestart = ls + (proto->linestart >= ols ?
13125 proto->linestart - ols : 0);
13126 parser->last_uni = ls + (proto->last_uni >= ols ?
13127 proto->last_uni - ols : 0);
13128 parser->last_lop = ls + (proto->last_lop >= ols ?
13129 proto->last_lop - ols : 0);
13131 parser->bufend = ls + SvCUR(parser->linestr);
13134 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13137 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13138 Copy(proto->nexttype, parser->nexttype, 5, I32);
13139 parser->nexttoke = proto->nexttoke;
13141 /* XXX should clone saved_curcop here, but we aren't passed
13142 * proto_perl; so do it in perl_clone_using instead */
13148 /* duplicate a file handle */
13151 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13155 PERL_ARGS_ASSERT_FP_DUP;
13156 PERL_UNUSED_ARG(type);
13159 return (PerlIO*)NULL;
13161 /* look for it in the table first */
13162 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13166 /* create anew and remember what it is */
13167 #ifdef __amigaos4__
13168 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13170 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13172 ptr_table_store(PL_ptr_table, fp, ret);
13176 /* duplicate a directory handle */
13179 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13183 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13185 const Direntry_t *dirent;
13186 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13192 PERL_UNUSED_CONTEXT;
13193 PERL_ARGS_ASSERT_DIRP_DUP;
13198 /* look for it in the table first */
13199 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13203 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13205 PERL_UNUSED_ARG(param);
13209 /* open the current directory (so we can switch back) */
13210 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13212 /* chdir to our dir handle and open the present working directory */
13213 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13214 PerlDir_close(pwd);
13215 return (DIR *)NULL;
13217 /* Now we should have two dir handles pointing to the same dir. */
13219 /* Be nice to the calling code and chdir back to where we were. */
13220 /* XXX If this fails, then what? */
13221 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13223 /* We have no need of the pwd handle any more. */
13224 PerlDir_close(pwd);
13227 # define d_namlen(d) (d)->d_namlen
13229 # define d_namlen(d) strlen((d)->d_name)
13231 /* Iterate once through dp, to get the file name at the current posi-
13232 tion. Then step back. */
13233 pos = PerlDir_tell(dp);
13234 if ((dirent = PerlDir_read(dp))) {
13235 len = d_namlen(dirent);
13236 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13237 /* If the len is somehow magically longer than the
13238 * maximum length of the directory entry, even though
13239 * we could fit it in a buffer, we could not copy it
13240 * from the dirent. Bail out. */
13241 PerlDir_close(ret);
13244 if (len <= sizeof smallbuf) name = smallbuf;
13245 else Newx(name, len, char);
13246 Move(dirent->d_name, name, len, char);
13248 PerlDir_seek(dp, pos);
13250 /* Iterate through the new dir handle, till we find a file with the
13252 if (!dirent) /* just before the end */
13254 pos = PerlDir_tell(ret);
13255 if (PerlDir_read(ret)) continue; /* not there yet */
13256 PerlDir_seek(ret, pos); /* step back */
13260 const long pos0 = PerlDir_tell(ret);
13262 pos = PerlDir_tell(ret);
13263 if ((dirent = PerlDir_read(ret))) {
13264 if (len == (STRLEN)d_namlen(dirent)
13265 && memEQ(name, dirent->d_name, len)) {
13267 PerlDir_seek(ret, pos); /* step back */
13270 /* else we are not there yet; keep iterating */
13272 else { /* This is not meant to happen. The best we can do is
13273 reset the iterator to the beginning. */
13274 PerlDir_seek(ret, pos0);
13281 if (name && name != smallbuf)
13286 ret = win32_dirp_dup(dp, param);
13289 /* pop it in the pointer table */
13291 ptr_table_store(PL_ptr_table, dp, ret);
13296 /* duplicate a typeglob */
13299 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13303 PERL_ARGS_ASSERT_GP_DUP;
13307 /* look for it in the table first */
13308 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13312 /* create anew and remember what it is */
13314 ptr_table_store(PL_ptr_table, gp, ret);
13317 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13318 on Newxz() to do this for us. */
13319 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13320 ret->gp_io = io_dup_inc(gp->gp_io, param);
13321 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13322 ret->gp_av = av_dup_inc(gp->gp_av, param);
13323 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13324 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13325 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13326 ret->gp_cvgen = gp->gp_cvgen;
13327 ret->gp_line = gp->gp_line;
13328 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13332 /* duplicate a chain of magic */
13335 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13337 MAGIC *mgret = NULL;
13338 MAGIC **mgprev_p = &mgret;
13340 PERL_ARGS_ASSERT_MG_DUP;
13342 for (; mg; mg = mg->mg_moremagic) {
13345 if ((param->flags & CLONEf_JOIN_IN)
13346 && mg->mg_type == PERL_MAGIC_backref)
13347 /* when joining, we let the individual SVs add themselves to
13348 * backref as needed. */
13351 Newx(nmg, 1, MAGIC);
13353 mgprev_p = &(nmg->mg_moremagic);
13355 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13356 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13357 from the original commit adding Perl_mg_dup() - revision 4538.
13358 Similarly there is the annotation "XXX random ptr?" next to the
13359 assignment to nmg->mg_ptr. */
13362 /* FIXME for plugins
13363 if (nmg->mg_type == PERL_MAGIC_qr) {
13364 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13368 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13369 ? nmg->mg_type == PERL_MAGIC_backref
13370 /* The backref AV has its reference
13371 * count deliberately bumped by 1 */
13372 ? SvREFCNT_inc(av_dup_inc((const AV *)
13373 nmg->mg_obj, param))
13374 : sv_dup_inc(nmg->mg_obj, param)
13375 : sv_dup(nmg->mg_obj, param);
13377 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13378 if (nmg->mg_len > 0) {
13379 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13380 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13381 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13383 AMT * const namtp = (AMT*)nmg->mg_ptr;
13384 sv_dup_inc_multiple((SV**)(namtp->table),
13385 (SV**)(namtp->table), NofAMmeth, param);
13388 else if (nmg->mg_len == HEf_SVKEY)
13389 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13391 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13392 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13398 #endif /* USE_ITHREADS */
13400 struct ptr_tbl_arena {
13401 struct ptr_tbl_arena *next;
13402 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13405 /* create a new pointer-mapping table */
13408 Perl_ptr_table_new(pTHX)
13411 PERL_UNUSED_CONTEXT;
13413 Newx(tbl, 1, PTR_TBL_t);
13414 tbl->tbl_max = 511;
13415 tbl->tbl_items = 0;
13416 tbl->tbl_arena = NULL;
13417 tbl->tbl_arena_next = NULL;
13418 tbl->tbl_arena_end = NULL;
13419 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13423 #define PTR_TABLE_HASH(ptr) \
13424 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13426 /* map an existing pointer using a table */
13428 STATIC PTR_TBL_ENT_t *
13429 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13431 PTR_TBL_ENT_t *tblent;
13432 const UV hash = PTR_TABLE_HASH(sv);
13434 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13436 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13437 for (; tblent; tblent = tblent->next) {
13438 if (tblent->oldval == sv)
13445 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13447 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13449 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13450 PERL_UNUSED_CONTEXT;
13452 return tblent ? tblent->newval : NULL;
13455 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13456 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13457 * the core's typical use of ptr_tables in thread cloning. */
13460 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13462 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13464 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13465 PERL_UNUSED_CONTEXT;
13468 tblent->newval = newsv;
13470 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13472 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13473 struct ptr_tbl_arena *new_arena;
13475 Newx(new_arena, 1, struct ptr_tbl_arena);
13476 new_arena->next = tbl->tbl_arena;
13477 tbl->tbl_arena = new_arena;
13478 tbl->tbl_arena_next = new_arena->array;
13479 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13482 tblent = tbl->tbl_arena_next++;
13484 tblent->oldval = oldsv;
13485 tblent->newval = newsv;
13486 tblent->next = tbl->tbl_ary[entry];
13487 tbl->tbl_ary[entry] = tblent;
13489 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13490 ptr_table_split(tbl);
13494 /* double the hash bucket size of an existing ptr table */
13497 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13499 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13500 const UV oldsize = tbl->tbl_max + 1;
13501 UV newsize = oldsize * 2;
13504 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13505 PERL_UNUSED_CONTEXT;
13507 Renew(ary, newsize, PTR_TBL_ENT_t*);
13508 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13509 tbl->tbl_max = --newsize;
13510 tbl->tbl_ary = ary;
13511 for (i=0; i < oldsize; i++, ary++) {
13512 PTR_TBL_ENT_t **entp = ary;
13513 PTR_TBL_ENT_t *ent = *ary;
13514 PTR_TBL_ENT_t **curentp;
13517 curentp = ary + oldsize;
13519 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13521 ent->next = *curentp;
13531 /* remove all the entries from a ptr table */
13532 /* Deprecated - will be removed post 5.14 */
13535 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13537 PERL_UNUSED_CONTEXT;
13538 if (tbl && tbl->tbl_items) {
13539 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13541 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13544 struct ptr_tbl_arena *next = arena->next;
13550 tbl->tbl_items = 0;
13551 tbl->tbl_arena = NULL;
13552 tbl->tbl_arena_next = NULL;
13553 tbl->tbl_arena_end = NULL;
13557 /* clear and free a ptr table */
13560 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13562 struct ptr_tbl_arena *arena;
13564 PERL_UNUSED_CONTEXT;
13570 arena = tbl->tbl_arena;
13573 struct ptr_tbl_arena *next = arena->next;
13579 Safefree(tbl->tbl_ary);
13583 #if defined(USE_ITHREADS)
13586 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13588 PERL_ARGS_ASSERT_RVPV_DUP;
13590 assert(!isREGEXP(sstr));
13592 if (SvWEAKREF(sstr)) {
13593 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13594 if (param->flags & CLONEf_JOIN_IN) {
13595 /* if joining, we add any back references individually rather
13596 * than copying the whole backref array */
13597 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13601 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13603 else if (SvPVX_const(sstr)) {
13604 /* Has something there */
13606 /* Normal PV - clone whole allocated space */
13607 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13608 /* sstr may not be that normal, but actually copy on write.
13609 But we are a true, independent SV, so: */
13613 /* Special case - not normally malloced for some reason */
13614 if (isGV_with_GP(sstr)) {
13615 /* Don't need to do anything here. */
13617 else if ((SvIsCOW(sstr))) {
13618 /* A "shared" PV - clone it as "shared" PV */
13620 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13624 /* Some other special case - random pointer */
13625 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13630 /* Copy the NULL */
13631 SvPV_set(dstr, NULL);
13635 /* duplicate a list of SVs. source and dest may point to the same memory. */
13637 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13638 SSize_t items, CLONE_PARAMS *const param)
13640 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13642 while (items-- > 0) {
13643 *dest++ = sv_dup_inc(*source++, param);
13649 /* duplicate an SV of any type (including AV, HV etc) */
13652 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13657 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13659 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13660 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13665 /* look for it in the table first */
13666 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13670 if(param->flags & CLONEf_JOIN_IN) {
13671 /** We are joining here so we don't want do clone
13672 something that is bad **/
13673 if (SvTYPE(sstr) == SVt_PVHV) {
13674 const HEK * const hvname = HvNAME_HEK(sstr);
13676 /** don't clone stashes if they already exist **/
13677 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13678 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13679 ptr_table_store(PL_ptr_table, sstr, dstr);
13683 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13684 HV *stash = GvSTASH(sstr);
13685 const HEK * hvname;
13686 if (stash && (hvname = HvNAME_HEK(stash))) {
13687 /** don't clone GVs if they already exist **/
13689 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13690 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13692 stash, GvNAME(sstr),
13698 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13699 ptr_table_store(PL_ptr_table, sstr, *svp);
13706 /* create anew and remember what it is */
13709 #ifdef DEBUG_LEAKING_SCALARS
13710 dstr->sv_debug_optype = sstr->sv_debug_optype;
13711 dstr->sv_debug_line = sstr->sv_debug_line;
13712 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13713 dstr->sv_debug_parent = (SV*)sstr;
13714 FREE_SV_DEBUG_FILE(dstr);
13715 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13718 ptr_table_store(PL_ptr_table, sstr, dstr);
13721 SvFLAGS(dstr) = SvFLAGS(sstr);
13722 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13723 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13726 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13727 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13728 (void*)PL_watch_pvx, SvPVX_const(sstr));
13731 /* don't clone objects whose class has asked us not to */
13733 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13739 switch (SvTYPE(sstr)) {
13741 SvANY(dstr) = NULL;
13744 SET_SVANY_FOR_BODYLESS_IV(dstr);
13746 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13748 SvIV_set(dstr, SvIVX(sstr));
13752 #if NVSIZE <= IVSIZE
13753 SET_SVANY_FOR_BODYLESS_NV(dstr);
13755 SvANY(dstr) = new_XNV();
13757 SvNV_set(dstr, SvNVX(sstr));
13761 /* These are all the types that need complex bodies allocating. */
13763 const svtype sv_type = SvTYPE(sstr);
13764 const struct body_details *const sv_type_details
13765 = bodies_by_type + sv_type;
13769 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13785 assert(sv_type_details->body_size);
13786 if (sv_type_details->arena) {
13787 new_body_inline(new_body, sv_type);
13789 = (void*)((char*)new_body - sv_type_details->offset);
13791 new_body = new_NOARENA(sv_type_details);
13795 SvANY(dstr) = new_body;
13798 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13799 ((char*)SvANY(dstr)) + sv_type_details->offset,
13800 sv_type_details->copy, char);
13802 Copy(((char*)SvANY(sstr)),
13803 ((char*)SvANY(dstr)),
13804 sv_type_details->body_size + sv_type_details->offset, char);
13807 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13808 && !isGV_with_GP(dstr)
13810 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13811 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13813 /* The Copy above means that all the source (unduplicated) pointers
13814 are now in the destination. We can check the flags and the
13815 pointers in either, but it's possible that there's less cache
13816 missing by always going for the destination.
13817 FIXME - instrument and check that assumption */
13818 if (sv_type >= SVt_PVMG) {
13820 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13821 if (SvOBJECT(dstr) && SvSTASH(dstr))
13822 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13823 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13826 /* The cast silences a GCC warning about unhandled types. */
13827 switch ((int)sv_type) {
13838 /* FIXME for plugins */
13839 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13840 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13843 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13844 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13845 LvTARG(dstr) = dstr;
13846 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13847 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13849 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13850 if (isREGEXP(sstr)) goto duprex;
13852 /* non-GP case already handled above */
13853 if(isGV_with_GP(sstr)) {
13854 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13855 /* Don't call sv_add_backref here as it's going to be
13856 created as part of the magic cloning of the symbol
13857 table--unless this is during a join and the stash
13858 is not actually being cloned. */
13859 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13860 at the point of this comment. */
13861 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13862 if (param->flags & CLONEf_JOIN_IN)
13863 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13864 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13865 (void)GpREFCNT_inc(GvGP(dstr));
13869 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13870 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13871 /* I have no idea why fake dirp (rsfps)
13872 should be treated differently but otherwise
13873 we end up with leaks -- sky*/
13874 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13875 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13876 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13878 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13879 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13880 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13881 if (IoDIRP(dstr)) {
13882 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13885 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13887 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13889 if (IoOFP(dstr) == IoIFP(sstr))
13890 IoOFP(dstr) = IoIFP(dstr);
13892 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13893 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13894 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13895 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13898 /* avoid cloning an empty array */
13899 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13900 SV **dst_ary, **src_ary;
13901 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13903 src_ary = AvARRAY((const AV *)sstr);
13904 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13905 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13906 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13907 AvALLOC((const AV *)dstr) = dst_ary;
13908 if (AvREAL((const AV *)sstr)) {
13909 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13913 while (items-- > 0)
13914 *dst_ary++ = sv_dup(*src_ary++, param);
13916 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13917 while (items-- > 0) {
13922 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13923 AvALLOC((const AV *)dstr) = (SV**)NULL;
13924 AvMAX( (const AV *)dstr) = -1;
13925 AvFILLp((const AV *)dstr) = -1;
13929 if (HvARRAY((const HV *)sstr)) {
13931 const bool sharekeys = !!HvSHAREKEYS(sstr);
13932 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13933 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13935 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13936 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13938 HvARRAY(dstr) = (HE**)darray;
13939 while (i <= sxhv->xhv_max) {
13940 const HE * const source = HvARRAY(sstr)[i];
13941 HvARRAY(dstr)[i] = source
13942 ? he_dup(source, sharekeys, param) : 0;
13946 const struct xpvhv_aux * const saux = HvAUX(sstr);
13947 struct xpvhv_aux * const daux = HvAUX(dstr);
13948 /* This flag isn't copied. */
13951 if (saux->xhv_name_count) {
13952 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13954 = saux->xhv_name_count < 0
13955 ? -saux->xhv_name_count
13956 : saux->xhv_name_count;
13957 HEK **shekp = sname + count;
13959 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13960 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13961 while (shekp-- > sname) {
13963 *dhekp = hek_dup(*shekp, param);
13967 daux->xhv_name_u.xhvnameu_name
13968 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13971 daux->xhv_name_count = saux->xhv_name_count;
13973 daux->xhv_aux_flags = saux->xhv_aux_flags;
13974 #ifdef PERL_HASH_RANDOMIZE_KEYS
13975 daux->xhv_rand = saux->xhv_rand;
13976 daux->xhv_last_rand = saux->xhv_last_rand;
13978 daux->xhv_riter = saux->xhv_riter;
13979 daux->xhv_eiter = saux->xhv_eiter
13980 ? he_dup(saux->xhv_eiter,
13981 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13982 /* backref array needs refcnt=2; see sv_add_backref */
13983 daux->xhv_backreferences =
13984 (param->flags & CLONEf_JOIN_IN)
13985 /* when joining, we let the individual GVs and
13986 * CVs add themselves to backref as
13987 * needed. This avoids pulling in stuff
13988 * that isn't required, and simplifies the
13989 * case where stashes aren't cloned back
13990 * if they already exist in the parent
13993 : saux->xhv_backreferences
13994 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13995 ? MUTABLE_AV(SvREFCNT_inc(
13996 sv_dup_inc((const SV *)
13997 saux->xhv_backreferences, param)))
13998 : MUTABLE_AV(sv_dup((const SV *)
13999 saux->xhv_backreferences, param))
14002 daux->xhv_mro_meta = saux->xhv_mro_meta
14003 ? mro_meta_dup(saux->xhv_mro_meta, param)
14006 /* Record stashes for possible cloning in Perl_clone(). */
14008 av_push(param->stashes, dstr);
14012 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14015 if (!(param->flags & CLONEf_COPY_STACKS)) {
14020 /* NOTE: not refcounted */
14021 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14022 hv_dup(CvSTASH(dstr), param);
14023 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14024 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14025 if (!CvISXSUB(dstr)) {
14027 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14029 CvSLABBED_off(dstr);
14030 } else if (CvCONST(dstr)) {
14031 CvXSUBANY(dstr).any_ptr =
14032 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14034 assert(!CvSLABBED(dstr));
14035 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14037 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14038 hek_dup(CvNAME_HEK((CV *)sstr), param);
14039 /* don't dup if copying back - CvGV isn't refcounted, so the
14040 * duped GV may never be freed. A bit of a hack! DAPM */
14042 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14044 ? gv_dup_inc(CvGV(sstr), param)
14045 : (param->flags & CLONEf_JOIN_IN)
14047 : gv_dup(CvGV(sstr), param);
14049 if (!CvISXSUB(sstr)) {
14050 PADLIST * padlist = CvPADLIST(sstr);
14052 padlist = padlist_dup(padlist, param);
14053 CvPADLIST_set(dstr, padlist);
14055 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14056 PoisonPADLIST(dstr);
14059 CvWEAKOUTSIDE(sstr)
14060 ? cv_dup( CvOUTSIDE(dstr), param)
14061 : cv_dup_inc(CvOUTSIDE(dstr), param);
14071 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14073 PERL_ARGS_ASSERT_SV_DUP_INC;
14074 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14078 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14080 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14081 PERL_ARGS_ASSERT_SV_DUP;
14083 /* Track every SV that (at least initially) had a reference count of 0.
14084 We need to do this by holding an actual reference to it in this array.
14085 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14086 (akin to the stashes hash, and the perl stack), we come unstuck if
14087 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14088 thread) is manipulated in a CLONE method, because CLONE runs before the
14089 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14090 (and fix things up by giving each a reference via the temps stack).
14091 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14092 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14093 before the walk of unreferenced happens and a reference to that is SV
14094 added to the temps stack. At which point we have the same SV considered
14095 to be in use, and free to be re-used. Not good.
14097 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14098 assert(param->unreferenced);
14099 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14105 /* duplicate a context */
14108 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14110 PERL_CONTEXT *ncxs;
14112 PERL_ARGS_ASSERT_CX_DUP;
14115 return (PERL_CONTEXT*)NULL;
14117 /* look for it in the table first */
14118 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14122 /* create anew and remember what it is */
14123 Newx(ncxs, max + 1, PERL_CONTEXT);
14124 ptr_table_store(PL_ptr_table, cxs, ncxs);
14125 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14128 PERL_CONTEXT * const ncx = &ncxs[ix];
14129 if (CxTYPE(ncx) == CXt_SUBST) {
14130 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14133 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14134 switch (CxTYPE(ncx)) {
14136 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14137 if(CxHASARGS(ncx)){
14138 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14140 ncx->blk_sub.savearray = NULL;
14142 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14143 ncx->blk_sub.prevcomppad);
14146 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14148 /* XXX should this sv_dup_inc? Or only if SvSCREAM ???? */
14149 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14150 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14151 /* XXX what do do with cur_top_env ???? */
14153 case CXt_LOOP_LAZYSV:
14154 ncx->blk_loop.state_u.lazysv.end
14155 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14156 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14157 duplication code instead.
14158 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14159 actually being the same function, and (2) order
14160 equivalence of the two unions.
14161 We can assert the later [but only at run time :-(] */
14162 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14163 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14166 ncx->blk_loop.state_u.ary.ary
14167 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14169 case CXt_LOOP_LIST:
14170 case CXt_LOOP_LAZYIV:
14171 /* code common to all 'for' CXt_LOOP_* types */
14172 ncx->blk_loop.itersave =
14173 sv_dup_inc(ncx->blk_loop.itersave, param);
14174 if (CxPADLOOP(ncx)) {
14175 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14176 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14177 ncx->blk_loop.oldcomppad =
14178 (PAD*)ptr_table_fetch(PL_ptr_table,
14179 ncx->blk_loop.oldcomppad);
14180 ncx->blk_loop.itervar_u.svp =
14181 &CX_CURPAD_SV(ncx->blk_loop, off);
14184 /* this copies the GV if CXp_FOR_GV, or the SV for an
14185 * alias (for \$x (...)) - relies on gv_dup being the
14186 * same as sv_dup */
14187 ncx->blk_loop.itervar_u.gv
14188 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14192 case CXt_LOOP_PLAIN:
14195 ncx->blk_format.prevcomppad =
14196 (PAD*)ptr_table_fetch(PL_ptr_table,
14197 ncx->blk_format.prevcomppad);
14198 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14199 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14200 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14204 ncx->blk_givwhen.defsv_save =
14205 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14218 /* duplicate a stack info structure */
14221 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14225 PERL_ARGS_ASSERT_SI_DUP;
14228 return (PERL_SI*)NULL;
14230 /* look for it in the table first */
14231 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14235 /* create anew and remember what it is */
14236 Newxz(nsi, 1, PERL_SI);
14237 ptr_table_store(PL_ptr_table, si, nsi);
14239 nsi->si_stack = av_dup_inc(si->si_stack, param);
14240 nsi->si_cxix = si->si_cxix;
14241 nsi->si_cxmax = si->si_cxmax;
14242 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14243 nsi->si_type = si->si_type;
14244 nsi->si_prev = si_dup(si->si_prev, param);
14245 nsi->si_next = si_dup(si->si_next, param);
14246 nsi->si_markoff = si->si_markoff;
14251 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14252 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14253 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14254 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14255 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14256 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14257 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14258 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14259 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14260 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14261 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14262 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14263 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14264 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14265 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14266 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14269 #define pv_dup_inc(p) SAVEPV(p)
14270 #define pv_dup(p) SAVEPV(p)
14271 #define svp_dup_inc(p,pp) any_dup(p,pp)
14273 /* map any object to the new equivent - either something in the
14274 * ptr table, or something in the interpreter structure
14278 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14282 PERL_ARGS_ASSERT_ANY_DUP;
14285 return (void*)NULL;
14287 /* look for it in the table first */
14288 ret = ptr_table_fetch(PL_ptr_table, v);
14292 /* see if it is part of the interpreter structure */
14293 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14294 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14302 /* duplicate the save stack */
14305 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14308 ANY * const ss = proto_perl->Isavestack;
14309 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14310 I32 ix = proto_perl->Isavestack_ix;
14323 void (*dptr) (void*);
14324 void (*dxptr) (pTHX_ void*);
14326 PERL_ARGS_ASSERT_SS_DUP;
14328 Newxz(nss, max, ANY);
14331 const UV uv = POPUV(ss,ix);
14332 const U8 type = (U8)uv & SAVE_MASK;
14334 TOPUV(nss,ix) = uv;
14336 case SAVEt_CLEARSV:
14337 case SAVEt_CLEARPADRANGE:
14339 case SAVEt_HELEM: /* hash element */
14340 case SAVEt_SV: /* scalar reference */
14341 sv = (const SV *)POPPTR(ss,ix);
14342 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14344 case SAVEt_ITEM: /* normal string */
14345 case SAVEt_GVSV: /* scalar slot in GV */
14346 sv = (const SV *)POPPTR(ss,ix);
14347 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14348 if (type == SAVEt_SV)
14352 case SAVEt_MORTALIZESV:
14353 case SAVEt_READONLY_OFF:
14354 sv = (const SV *)POPPTR(ss,ix);
14355 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14357 case SAVEt_FREEPADNAME:
14358 ptr = POPPTR(ss,ix);
14359 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14360 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14362 case SAVEt_SHARED_PVREF: /* char* in shared space */
14363 c = (char*)POPPTR(ss,ix);
14364 TOPPTR(nss,ix) = savesharedpv(c);
14365 ptr = POPPTR(ss,ix);
14366 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14368 case SAVEt_GENERIC_SVREF: /* generic sv */
14369 case SAVEt_SVREF: /* scalar reference */
14370 sv = (const SV *)POPPTR(ss,ix);
14371 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14372 if (type == SAVEt_SVREF)
14373 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14374 ptr = POPPTR(ss,ix);
14375 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14377 case SAVEt_GVSLOT: /* any slot in GV */
14378 sv = (const SV *)POPPTR(ss,ix);
14379 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14380 ptr = POPPTR(ss,ix);
14381 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14382 sv = (const SV *)POPPTR(ss,ix);
14383 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14385 case SAVEt_HV: /* hash reference */
14386 case SAVEt_AV: /* array reference */
14387 sv = (const SV *) POPPTR(ss,ix);
14388 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14390 case SAVEt_COMPPAD:
14392 sv = (const SV *) POPPTR(ss,ix);
14393 TOPPTR(nss,ix) = sv_dup(sv, param);
14395 case SAVEt_INT: /* int reference */
14396 ptr = POPPTR(ss,ix);
14397 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14398 intval = (int)POPINT(ss,ix);
14399 TOPINT(nss,ix) = intval;
14401 case SAVEt_LONG: /* long reference */
14402 ptr = POPPTR(ss,ix);
14403 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14404 longval = (long)POPLONG(ss,ix);
14405 TOPLONG(nss,ix) = longval;
14407 case SAVEt_I32: /* I32 reference */
14408 ptr = POPPTR(ss,ix);
14409 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14411 TOPINT(nss,ix) = i;
14413 case SAVEt_IV: /* IV reference */
14414 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14415 ptr = POPPTR(ss,ix);
14416 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14418 TOPIV(nss,ix) = iv;
14420 case SAVEt_TMPSFLOOR:
14422 TOPIV(nss,ix) = iv;
14424 case SAVEt_HPTR: /* HV* reference */
14425 case SAVEt_APTR: /* AV* reference */
14426 case SAVEt_SPTR: /* SV* reference */
14427 ptr = POPPTR(ss,ix);
14428 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14429 sv = (const SV *)POPPTR(ss,ix);
14430 TOPPTR(nss,ix) = sv_dup(sv, param);
14432 case SAVEt_VPTR: /* random* reference */
14433 ptr = POPPTR(ss,ix);
14434 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14436 case SAVEt_INT_SMALL:
14437 case SAVEt_I32_SMALL:
14438 case SAVEt_I16: /* I16 reference */
14439 case SAVEt_I8: /* I8 reference */
14441 ptr = POPPTR(ss,ix);
14442 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14444 case SAVEt_GENERIC_PVREF: /* generic char* */
14445 case SAVEt_PPTR: /* char* reference */
14446 ptr = POPPTR(ss,ix);
14447 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14448 c = (char*)POPPTR(ss,ix);
14449 TOPPTR(nss,ix) = pv_dup(c);
14451 case SAVEt_GP: /* scalar reference */
14452 gp = (GP*)POPPTR(ss,ix);
14453 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14454 (void)GpREFCNT_inc(gp);
14455 gv = (const GV *)POPPTR(ss,ix);
14456 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14459 ptr = POPPTR(ss,ix);
14460 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14461 /* these are assumed to be refcounted properly */
14463 switch (((OP*)ptr)->op_type) {
14465 case OP_LEAVESUBLV:
14469 case OP_LEAVEWRITE:
14470 TOPPTR(nss,ix) = ptr;
14473 (void) OpREFCNT_inc(o);
14477 TOPPTR(nss,ix) = NULL;
14482 TOPPTR(nss,ix) = NULL;
14484 case SAVEt_FREECOPHH:
14485 ptr = POPPTR(ss,ix);
14486 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14488 case SAVEt_ADELETE:
14489 av = (const AV *)POPPTR(ss,ix);
14490 TOPPTR(nss,ix) = av_dup_inc(av, param);
14492 TOPINT(nss,ix) = i;
14495 hv = (const HV *)POPPTR(ss,ix);
14496 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14498 TOPINT(nss,ix) = i;
14501 c = (char*)POPPTR(ss,ix);
14502 TOPPTR(nss,ix) = pv_dup_inc(c);
14504 case SAVEt_STACK_POS: /* Position on Perl stack */
14506 TOPINT(nss,ix) = i;
14508 case SAVEt_DESTRUCTOR:
14509 ptr = POPPTR(ss,ix);
14510 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14511 dptr = POPDPTR(ss,ix);
14512 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14513 any_dup(FPTR2DPTR(void *, dptr),
14516 case SAVEt_DESTRUCTOR_X:
14517 ptr = POPPTR(ss,ix);
14518 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14519 dxptr = POPDXPTR(ss,ix);
14520 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14521 any_dup(FPTR2DPTR(void *, dxptr),
14524 case SAVEt_REGCONTEXT:
14526 ix -= uv >> SAVE_TIGHT_SHIFT;
14528 case SAVEt_AELEM: /* array element */
14529 sv = (const SV *)POPPTR(ss,ix);
14530 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14532 TOPINT(nss,ix) = i;
14533 av = (const AV *)POPPTR(ss,ix);
14534 TOPPTR(nss,ix) = av_dup_inc(av, param);
14537 ptr = POPPTR(ss,ix);
14538 TOPPTR(nss,ix) = ptr;
14541 ptr = POPPTR(ss,ix);
14542 ptr = cophh_copy((COPHH*)ptr);
14543 TOPPTR(nss,ix) = ptr;
14545 TOPINT(nss,ix) = i;
14546 if (i & HINT_LOCALIZE_HH) {
14547 hv = (const HV *)POPPTR(ss,ix);
14548 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14551 case SAVEt_PADSV_AND_MORTALIZE:
14552 longval = (long)POPLONG(ss,ix);
14553 TOPLONG(nss,ix) = longval;
14554 ptr = POPPTR(ss,ix);
14555 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14556 sv = (const SV *)POPPTR(ss,ix);
14557 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14559 case SAVEt_SET_SVFLAGS:
14561 TOPINT(nss,ix) = i;
14563 TOPINT(nss,ix) = i;
14564 sv = (const SV *)POPPTR(ss,ix);
14565 TOPPTR(nss,ix) = sv_dup(sv, param);
14567 case SAVEt_COMPILE_WARNINGS:
14568 ptr = POPPTR(ss,ix);
14569 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14572 ptr = POPPTR(ss,ix);
14573 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14577 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14585 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14586 * flag to the result. This is done for each stash before cloning starts,
14587 * so we know which stashes want their objects cloned */
14590 do_mark_cloneable_stash(pTHX_ SV *const sv)
14592 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14594 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14595 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14596 if (cloner && GvCV(cloner)) {
14603 mXPUSHs(newSVhek(hvname));
14605 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14612 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14620 =for apidoc perl_clone
14622 Create and return a new interpreter by cloning the current one.
14624 C<perl_clone> takes these flags as parameters:
14626 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14627 without it we only clone the data and zero the stacks,
14628 with it we copy the stacks and the new perl interpreter is
14629 ready to run at the exact same point as the previous one.
14630 The pseudo-fork code uses C<COPY_STACKS> while the
14631 threads->create doesn't.
14633 C<CLONEf_KEEP_PTR_TABLE> -
14634 C<perl_clone> keeps a ptr_table with the pointer of the old
14635 variable as a key and the new variable as a value,
14636 this allows it to check if something has been cloned and not
14637 clone it again but rather just use the value and increase the
14638 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14639 the ptr_table using the function
14640 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14641 reason to keep it around is if you want to dup some of your own
14642 variable who are outside the graph perl scans, an example of this
14643 code is in F<threads.xs> create.
14645 C<CLONEf_CLONE_HOST> -
14646 This is a win32 thing, it is ignored on unix, it tells perls
14647 win32host code (which is c++) to clone itself, this is needed on
14648 win32 if you want to run two threads at the same time,
14649 if you just want to do some stuff in a separate perl interpreter
14650 and then throw it away and return to the original one,
14651 you don't need to do anything.
14656 /* XXX the above needs expanding by someone who actually understands it ! */
14657 EXTERN_C PerlInterpreter *
14658 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14661 perl_clone(PerlInterpreter *proto_perl, UV flags)
14664 #ifdef PERL_IMPLICIT_SYS
14666 PERL_ARGS_ASSERT_PERL_CLONE;
14668 /* perlhost.h so we need to call into it
14669 to clone the host, CPerlHost should have a c interface, sky */
14671 #ifndef __amigaos4__
14672 if (flags & CLONEf_CLONE_HOST) {
14673 return perl_clone_host(proto_perl,flags);
14676 return perl_clone_using(proto_perl, flags,
14678 proto_perl->IMemShared,
14679 proto_perl->IMemParse,
14681 proto_perl->IStdIO,
14685 proto_perl->IProc);
14689 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14690 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14691 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14692 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14693 struct IPerlDir* ipD, struct IPerlSock* ipS,
14694 struct IPerlProc* ipP)
14696 /* XXX many of the string copies here can be optimized if they're
14697 * constants; they need to be allocated as common memory and just
14698 * their pointers copied. */
14701 CLONE_PARAMS clone_params;
14702 CLONE_PARAMS* const param = &clone_params;
14704 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14706 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14707 #else /* !PERL_IMPLICIT_SYS */
14709 CLONE_PARAMS clone_params;
14710 CLONE_PARAMS* param = &clone_params;
14711 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14713 PERL_ARGS_ASSERT_PERL_CLONE;
14714 #endif /* PERL_IMPLICIT_SYS */
14716 /* for each stash, determine whether its objects should be cloned */
14717 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14718 PERL_SET_THX(my_perl);
14721 PoisonNew(my_perl, 1, PerlInterpreter);
14724 PL_defstash = NULL; /* may be used by perl malloc() */
14727 PL_scopestack_name = 0;
14729 PL_savestack_ix = 0;
14730 PL_savestack_max = -1;
14731 PL_sig_pending = 0;
14733 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14734 Zero(&PL_padname_undef, 1, PADNAME);
14735 Zero(&PL_padname_const, 1, PADNAME);
14736 # ifdef DEBUG_LEAKING_SCALARS
14737 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14739 # ifdef PERL_TRACE_OPS
14740 Zero(PL_op_exec_cnt, OP_max+2, UV);
14742 #else /* !DEBUGGING */
14743 Zero(my_perl, 1, PerlInterpreter);
14744 #endif /* DEBUGGING */
14746 #ifdef PERL_IMPLICIT_SYS
14747 /* host pointers */
14749 PL_MemShared = ipMS;
14750 PL_MemParse = ipMP;
14757 #endif /* PERL_IMPLICIT_SYS */
14760 param->flags = flags;
14761 /* Nothing in the core code uses this, but we make it available to
14762 extensions (using mg_dup). */
14763 param->proto_perl = proto_perl;
14764 /* Likely nothing will use this, but it is initialised to be consistent
14765 with Perl_clone_params_new(). */
14766 param->new_perl = my_perl;
14767 param->unreferenced = NULL;
14770 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14772 PL_body_arenas = NULL;
14773 Zero(&PL_body_roots, 1, PL_body_roots);
14777 PL_sv_arenaroot = NULL;
14779 PL_debug = proto_perl->Idebug;
14781 /* dbargs array probably holds garbage */
14784 PL_compiling = proto_perl->Icompiling;
14786 /* pseudo environmental stuff */
14787 PL_origargc = proto_perl->Iorigargc;
14788 PL_origargv = proto_perl->Iorigargv;
14790 #ifndef NO_TAINT_SUPPORT
14791 /* Set tainting stuff before PerlIO_debug can possibly get called */
14792 PL_tainting = proto_perl->Itainting;
14793 PL_taint_warn = proto_perl->Itaint_warn;
14795 PL_tainting = FALSE;
14796 PL_taint_warn = FALSE;
14799 PL_minus_c = proto_perl->Iminus_c;
14801 PL_localpatches = proto_perl->Ilocalpatches;
14802 PL_splitstr = proto_perl->Isplitstr;
14803 PL_minus_n = proto_perl->Iminus_n;
14804 PL_minus_p = proto_perl->Iminus_p;
14805 PL_minus_l = proto_perl->Iminus_l;
14806 PL_minus_a = proto_perl->Iminus_a;
14807 PL_minus_E = proto_perl->Iminus_E;
14808 PL_minus_F = proto_perl->Iminus_F;
14809 PL_doswitches = proto_perl->Idoswitches;
14810 PL_dowarn = proto_perl->Idowarn;
14811 #ifdef PERL_SAWAMPERSAND
14812 PL_sawampersand = proto_perl->Isawampersand;
14814 PL_unsafe = proto_perl->Iunsafe;
14815 PL_perldb = proto_perl->Iperldb;
14816 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14817 PL_exit_flags = proto_perl->Iexit_flags;
14819 /* XXX time(&PL_basetime) when asked for? */
14820 PL_basetime = proto_perl->Ibasetime;
14822 PL_maxsysfd = proto_perl->Imaxsysfd;
14823 PL_statusvalue = proto_perl->Istatusvalue;
14825 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14827 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14830 /* RE engine related */
14831 PL_regmatch_slab = NULL;
14832 PL_reg_curpm = NULL;
14834 PL_sub_generation = proto_perl->Isub_generation;
14836 /* funky return mechanisms */
14837 PL_forkprocess = proto_perl->Iforkprocess;
14839 /* internal state */
14840 PL_main_start = proto_perl->Imain_start;
14841 PL_eval_root = proto_perl->Ieval_root;
14842 PL_eval_start = proto_perl->Ieval_start;
14844 PL_filemode = proto_perl->Ifilemode;
14845 PL_lastfd = proto_perl->Ilastfd;
14846 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14849 PL_gensym = proto_perl->Igensym;
14851 PL_laststatval = proto_perl->Ilaststatval;
14852 PL_laststype = proto_perl->Ilaststype;
14855 PL_profiledata = NULL;
14857 PL_generation = proto_perl->Igeneration;
14859 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14860 PL_in_clean_all = proto_perl->Iin_clean_all;
14862 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14863 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14864 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14865 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14866 PL_nomemok = proto_perl->Inomemok;
14867 PL_an = proto_perl->Ian;
14868 PL_evalseq = proto_perl->Ievalseq;
14869 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14870 PL_origalen = proto_perl->Iorigalen;
14872 PL_sighandlerp = proto_perl->Isighandlerp;
14874 PL_runops = proto_perl->Irunops;
14876 PL_subline = proto_perl->Isubline;
14878 PL_cv_has_eval = proto_perl->Icv_has_eval;
14881 PL_cryptseen = proto_perl->Icryptseen;
14884 #ifdef USE_LOCALE_COLLATE
14885 PL_collation_ix = proto_perl->Icollation_ix;
14886 PL_collation_standard = proto_perl->Icollation_standard;
14887 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14888 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14889 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14890 #endif /* USE_LOCALE_COLLATE */
14892 #ifdef USE_LOCALE_NUMERIC
14893 PL_numeric_standard = proto_perl->Inumeric_standard;
14894 PL_numeric_local = proto_perl->Inumeric_local;
14895 #endif /* !USE_LOCALE_NUMERIC */
14897 /* Did the locale setup indicate UTF-8? */
14898 PL_utf8locale = proto_perl->Iutf8locale;
14899 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14900 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
14901 /* Unicode features (see perlrun/-C) */
14902 PL_unicode = proto_perl->Iunicode;
14904 /* Pre-5.8 signals control */
14905 PL_signals = proto_perl->Isignals;
14907 /* times() ticks per second */
14908 PL_clocktick = proto_perl->Iclocktick;
14910 /* Recursion stopper for PerlIO_find_layer */
14911 PL_in_load_module = proto_perl->Iin_load_module;
14913 /* sort() routine */
14914 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14916 /* Not really needed/useful since the reenrant_retint is "volatile",
14917 * but do it for consistency's sake. */
14918 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14920 /* Hooks to shared SVs and locks. */
14921 PL_sharehook = proto_perl->Isharehook;
14922 PL_lockhook = proto_perl->Ilockhook;
14923 PL_unlockhook = proto_perl->Iunlockhook;
14924 PL_threadhook = proto_perl->Ithreadhook;
14925 PL_destroyhook = proto_perl->Idestroyhook;
14926 PL_signalhook = proto_perl->Isignalhook;
14928 PL_globhook = proto_perl->Iglobhook;
14931 PL_last_swash_hv = NULL; /* reinits on demand */
14932 PL_last_swash_klen = 0;
14933 PL_last_swash_key[0]= '\0';
14934 PL_last_swash_tmps = (U8*)NULL;
14935 PL_last_swash_slen = 0;
14937 PL_srand_called = proto_perl->Isrand_called;
14938 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14940 if (flags & CLONEf_COPY_STACKS) {
14941 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14942 PL_tmps_ix = proto_perl->Itmps_ix;
14943 PL_tmps_max = proto_perl->Itmps_max;
14944 PL_tmps_floor = proto_perl->Itmps_floor;
14946 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14947 * NOTE: unlike the others! */
14948 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14949 PL_scopestack_max = proto_perl->Iscopestack_max;
14951 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14952 * NOTE: unlike the others! */
14953 PL_savestack_ix = proto_perl->Isavestack_ix;
14954 PL_savestack_max = proto_perl->Isavestack_max;
14957 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14958 PL_top_env = &PL_start_env;
14960 PL_op = proto_perl->Iop;
14963 PL_Xpv = (XPV*)NULL;
14964 my_perl->Ina = proto_perl->Ina;
14966 PL_statbuf = proto_perl->Istatbuf;
14967 PL_statcache = proto_perl->Istatcache;
14969 #ifndef NO_TAINT_SUPPORT
14970 PL_tainted = proto_perl->Itainted;
14972 PL_tainted = FALSE;
14974 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14976 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14978 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14979 PL_restartop = proto_perl->Irestartop;
14980 PL_in_eval = proto_perl->Iin_eval;
14981 PL_delaymagic = proto_perl->Idelaymagic;
14982 PL_phase = proto_perl->Iphase;
14983 PL_localizing = proto_perl->Ilocalizing;
14985 PL_hv_fetch_ent_mh = NULL;
14986 PL_modcount = proto_perl->Imodcount;
14987 PL_lastgotoprobe = NULL;
14988 PL_dumpindent = proto_perl->Idumpindent;
14990 PL_efloatbuf = NULL; /* reinits on demand */
14991 PL_efloatsize = 0; /* reinits on demand */
14995 PL_colorset = 0; /* reinits PL_colors[] */
14996 /*PL_colors[6] = {0,0,0,0,0,0};*/
14998 /* Pluggable optimizer */
14999 PL_peepp = proto_perl->Ipeepp;
15000 PL_rpeepp = proto_perl->Irpeepp;
15001 /* op_free() hook */
15002 PL_opfreehook = proto_perl->Iopfreehook;
15004 #ifdef USE_REENTRANT_API
15005 /* XXX: things like -Dm will segfault here in perlio, but doing
15006 * PERL_SET_CONTEXT(proto_perl);
15007 * breaks too many other things
15009 Perl_reentrant_init(aTHX);
15012 /* create SV map for pointer relocation */
15013 PL_ptr_table = ptr_table_new();
15015 /* initialize these special pointers as early as possible */
15017 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15018 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15019 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15020 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15021 &PL_padname_const);
15023 /* create (a non-shared!) shared string table */
15024 PL_strtab = newHV();
15025 HvSHAREKEYS_off(PL_strtab);
15026 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15027 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15029 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15031 /* This PV will be free'd special way so must set it same way op.c does */
15032 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15033 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15035 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15036 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15037 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15038 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15040 param->stashes = newAV(); /* Setup array of objects to call clone on */
15041 /* This makes no difference to the implementation, as it always pushes
15042 and shifts pointers to other SVs without changing their reference
15043 count, with the array becoming empty before it is freed. However, it
15044 makes it conceptually clear what is going on, and will avoid some
15045 work inside av.c, filling slots between AvFILL() and AvMAX() with
15046 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15047 AvREAL_off(param->stashes);
15049 if (!(flags & CLONEf_COPY_STACKS)) {
15050 param->unreferenced = newAV();
15053 #ifdef PERLIO_LAYERS
15054 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15055 PerlIO_clone(aTHX_ proto_perl, param);
15058 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15059 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15060 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15061 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15062 PL_xsubfilename = proto_perl->Ixsubfilename;
15063 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15064 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15067 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15068 PL_inplace = SAVEPV(proto_perl->Iinplace);
15069 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15071 /* magical thingies */
15073 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15074 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15075 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15078 /* Clone the regex array */
15079 /* ORANGE FIXME for plugins, probably in the SV dup code.
15080 newSViv(PTR2IV(CALLREGDUPE(
15081 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15083 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15084 PL_regex_pad = AvARRAY(PL_regex_padav);
15086 PL_stashpadmax = proto_perl->Istashpadmax;
15087 PL_stashpadix = proto_perl->Istashpadix ;
15088 Newx(PL_stashpad, PL_stashpadmax, HV *);
15091 for (; o < PL_stashpadmax; ++o)
15092 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15095 /* shortcuts to various I/O objects */
15096 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15097 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15098 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15099 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15100 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15101 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15102 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15104 /* shortcuts to regexp stuff */
15105 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15107 /* shortcuts to misc objects */
15108 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15110 /* shortcuts to debugging objects */
15111 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15112 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15113 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15114 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15115 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15116 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15117 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15119 /* symbol tables */
15120 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15121 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15122 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15123 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15124 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15126 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15127 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15128 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15129 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15130 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15131 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15132 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15133 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15134 PL_savebegin = proto_perl->Isavebegin;
15136 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15138 /* subprocess state */
15139 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15141 if (proto_perl->Iop_mask)
15142 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15145 /* PL_asserting = proto_perl->Iasserting; */
15147 /* current interpreter roots */
15148 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15150 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15153 /* runtime control stuff */
15154 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15156 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15158 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15160 /* interpreter atexit processing */
15161 PL_exitlistlen = proto_perl->Iexitlistlen;
15162 if (PL_exitlistlen) {
15163 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15164 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15167 PL_exitlist = (PerlExitListEntry*)NULL;
15169 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15170 if (PL_my_cxt_size) {
15171 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15172 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15173 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15174 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15175 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15179 PL_my_cxt_list = (void**)NULL;
15180 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15181 PL_my_cxt_keys = (const char**)NULL;
15184 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15185 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15186 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15187 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15189 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15191 PAD_CLONE_VARS(proto_perl, param);
15193 #ifdef HAVE_INTERP_INTERN
15194 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15197 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15199 #ifdef PERL_USES_PL_PIDSTATUS
15200 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15202 PL_osname = SAVEPV(proto_perl->Iosname);
15203 PL_parser = parser_dup(proto_perl->Iparser, param);
15205 /* XXX this only works if the saved cop has already been cloned */
15206 if (proto_perl->Iparser) {
15207 PL_parser->saved_curcop = (COP*)any_dup(
15208 proto_perl->Iparser->saved_curcop,
15212 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15214 #ifdef USE_LOCALE_CTYPE
15215 /* Should we warn if uses locale? */
15216 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15219 #ifdef USE_LOCALE_COLLATE
15220 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15221 #endif /* USE_LOCALE_COLLATE */
15223 #ifdef USE_LOCALE_NUMERIC
15224 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15225 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15226 #endif /* !USE_LOCALE_NUMERIC */
15228 /* Unicode inversion lists */
15229 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15230 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15231 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15232 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15234 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15235 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15237 /* utf8 character class swashes */
15238 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15239 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15241 for (i = 0; i < POSIX_CC_COUNT; i++) {
15242 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15244 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15245 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15246 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15247 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15248 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15249 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15250 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15251 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15252 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15253 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15254 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15255 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15256 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15257 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15258 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15259 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15260 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15262 if (proto_perl->Ipsig_pend) {
15263 Newxz(PL_psig_pend, SIG_SIZE, int);
15266 PL_psig_pend = (int*)NULL;
15269 if (proto_perl->Ipsig_name) {
15270 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15271 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15273 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15276 PL_psig_ptr = (SV**)NULL;
15277 PL_psig_name = (SV**)NULL;
15280 if (flags & CLONEf_COPY_STACKS) {
15281 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15282 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15283 PL_tmps_ix+1, param);
15285 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15286 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15287 Newxz(PL_markstack, i, I32);
15288 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15289 - proto_perl->Imarkstack);
15290 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15291 - proto_perl->Imarkstack);
15292 Copy(proto_perl->Imarkstack, PL_markstack,
15293 PL_markstack_ptr - PL_markstack + 1, I32);
15295 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15296 * NOTE: unlike the others! */
15297 Newxz(PL_scopestack, PL_scopestack_max, I32);
15298 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15301 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15302 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15304 /* reset stack AV to correct length before its duped via
15305 * PL_curstackinfo */
15306 AvFILLp(proto_perl->Icurstack) =
15307 proto_perl->Istack_sp - proto_perl->Istack_base;
15309 /* NOTE: si_dup() looks at PL_markstack */
15310 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15312 /* PL_curstack = PL_curstackinfo->si_stack; */
15313 PL_curstack = av_dup(proto_perl->Icurstack, param);
15314 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15316 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15317 PL_stack_base = AvARRAY(PL_curstack);
15318 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15319 - proto_perl->Istack_base);
15320 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15322 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15323 PL_savestack = ss_dup(proto_perl, param);
15327 ENTER; /* perl_destruct() wants to LEAVE; */
15330 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15331 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15333 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15334 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15335 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15336 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15337 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15338 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15340 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15342 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15343 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15344 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15346 PL_stashcache = newHV();
15348 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15349 proto_perl->Iwatchaddr);
15350 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15351 if (PL_debug && PL_watchaddr) {
15352 PerlIO_printf(Perl_debug_log,
15353 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15354 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15355 PTR2UV(PL_watchok));
15358 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15359 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15360 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15362 /* Call the ->CLONE method, if it exists, for each of the stashes
15363 identified by sv_dup() above.
15365 while(av_tindex(param->stashes) != -1) {
15366 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15367 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15368 if (cloner && GvCV(cloner)) {
15373 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15375 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15381 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15382 ptr_table_free(PL_ptr_table);
15383 PL_ptr_table = NULL;
15386 if (!(flags & CLONEf_COPY_STACKS)) {
15387 unreferenced_to_tmp_stack(param->unreferenced);
15390 SvREFCNT_dec(param->stashes);
15392 /* orphaned? eg threads->new inside BEGIN or use */
15393 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15394 SvREFCNT_inc_simple_void(PL_compcv);
15395 SAVEFREESV(PL_compcv);
15402 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15404 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15406 if (AvFILLp(unreferenced) > -1) {
15407 SV **svp = AvARRAY(unreferenced);
15408 SV **const last = svp + AvFILLp(unreferenced);
15412 if (SvREFCNT(*svp) == 1)
15414 } while (++svp <= last);
15416 EXTEND_MORTAL(count);
15417 svp = AvARRAY(unreferenced);
15420 if (SvREFCNT(*svp) == 1) {
15421 /* Our reference is the only one to this SV. This means that
15422 in this thread, the scalar effectively has a 0 reference.
15423 That doesn't work (cleanup never happens), so donate our
15424 reference to it onto the save stack. */
15425 PL_tmps_stack[++PL_tmps_ix] = *svp;
15427 /* As an optimisation, because we are already walking the
15428 entire array, instead of above doing either
15429 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15430 release our reference to the scalar, so that at the end of
15431 the array owns zero references to the scalars it happens to
15432 point to. We are effectively converting the array from
15433 AvREAL() on to AvREAL() off. This saves the av_clear()
15434 (triggered by the SvREFCNT_dec(unreferenced) below) from
15435 walking the array a second time. */
15436 SvREFCNT_dec(*svp);
15439 } while (++svp <= last);
15440 AvREAL_off(unreferenced);
15442 SvREFCNT_dec_NN(unreferenced);
15446 Perl_clone_params_del(CLONE_PARAMS *param)
15448 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15450 PerlInterpreter *const to = param->new_perl;
15452 PerlInterpreter *const was = PERL_GET_THX;
15454 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15460 SvREFCNT_dec(param->stashes);
15461 if (param->unreferenced)
15462 unreferenced_to_tmp_stack(param->unreferenced);
15472 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15475 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15476 does a dTHX; to get the context from thread local storage.
15477 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15478 a version that passes in my_perl. */
15479 PerlInterpreter *const was = PERL_GET_THX;
15480 CLONE_PARAMS *param;
15482 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15488 /* Given that we've set the context, we can do this unshared. */
15489 Newx(param, 1, CLONE_PARAMS);
15492 param->proto_perl = from;
15493 param->new_perl = to;
15494 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15495 AvREAL_off(param->stashes);
15496 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15504 #endif /* USE_ITHREADS */
15507 Perl_init_constants(pTHX)
15509 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15510 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15511 SvANY(&PL_sv_undef) = NULL;
15513 SvANY(&PL_sv_no) = new_XPVNV();
15514 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15515 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15516 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15519 SvANY(&PL_sv_yes) = new_XPVNV();
15520 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15521 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15522 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15525 SvPV_set(&PL_sv_no, (char*)PL_No);
15526 SvCUR_set(&PL_sv_no, 0);
15527 SvLEN_set(&PL_sv_no, 0);
15528 SvIV_set(&PL_sv_no, 0);
15529 SvNV_set(&PL_sv_no, 0);
15531 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15532 SvCUR_set(&PL_sv_yes, 1);
15533 SvLEN_set(&PL_sv_yes, 0);
15534 SvIV_set(&PL_sv_yes, 1);
15535 SvNV_set(&PL_sv_yes, 1);
15537 PadnamePV(&PL_padname_const) = (char *)PL_No;
15541 =head1 Unicode Support
15543 =for apidoc sv_recode_to_utf8
15545 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15546 of C<sv> is assumed to be octets in that encoding, and C<sv>
15547 will be converted into Unicode (and UTF-8).
15549 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15550 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15551 an C<Encode::XS> Encoding object, bad things will happen.
15552 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15554 The PV of C<sv> is returned.
15559 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15561 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15563 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15572 if (SvPADTMP(nsv)) {
15573 nsv = sv_newmortal();
15574 SvSetSV_nosteal(nsv, sv);
15583 Passing sv_yes is wrong - it needs to be or'ed set of constants
15584 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15585 remove converted chars from source.
15587 Both will default the value - let them.
15589 XPUSHs(&PL_sv_yes);
15592 call_method("decode", G_SCALAR);
15596 s = SvPV_const(uni, len);
15597 if (s != SvPVX_const(sv)) {
15598 SvGROW(sv, len + 1);
15599 Move(s, SvPVX(sv), len + 1, char);
15600 SvCUR_set(sv, len);
15605 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15606 /* clear pos and any utf8 cache */
15607 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15610 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15611 magic_setutf8(sv,mg); /* clear UTF8 cache */
15616 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15620 =for apidoc sv_cat_decode
15622 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15623 assumed to be octets in that encoding and decoding the input starts
15624 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15625 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15626 when the string C<tstr> appears in decoding output or the input ends on
15627 the PV of C<ssv>. The value which C<offset> points will be modified
15628 to the last input position on C<ssv>.
15630 Returns TRUE if the terminator was found, else returns FALSE.
15635 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15636 SV *ssv, int *offset, char *tstr, int tlen)
15640 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15642 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15653 offsv = newSViv(*offset);
15655 mPUSHp(tstr, tlen);
15657 call_method("cat_decode", G_SCALAR);
15659 ret = SvTRUE(TOPs);
15660 *offset = SvIV(offsv);
15666 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15671 /* ---------------------------------------------------------------------
15673 * support functions for report_uninit()
15676 /* the maxiumum size of array or hash where we will scan looking
15677 * for the undefined element that triggered the warning */
15679 #define FUV_MAX_SEARCH_SIZE 1000
15681 /* Look for an entry in the hash whose value has the same SV as val;
15682 * If so, return a mortal copy of the key. */
15685 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15691 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15693 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15694 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15697 array = HvARRAY(hv);
15699 for (i=HvMAX(hv); i>=0; i--) {
15701 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15702 if (HeVAL(entry) != val)
15704 if ( HeVAL(entry) == &PL_sv_undef ||
15705 HeVAL(entry) == &PL_sv_placeholder)
15709 if (HeKLEN(entry) == HEf_SVKEY)
15710 return sv_mortalcopy(HeKEY_sv(entry));
15711 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15717 /* Look for an entry in the array whose value has the same SV as val;
15718 * If so, return the index, otherwise return -1. */
15721 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15723 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15725 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15726 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15729 if (val != &PL_sv_undef) {
15730 SV ** const svp = AvARRAY(av);
15733 for (i=AvFILLp(av); i>=0; i--)
15740 /* varname(): return the name of a variable, optionally with a subscript.
15741 * If gv is non-zero, use the name of that global, along with gvtype (one
15742 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15743 * targ. Depending on the value of the subscript_type flag, return:
15746 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15747 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15748 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15749 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15752 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15753 const SV *const keyname, SSize_t aindex, int subscript_type)
15756 SV * const name = sv_newmortal();
15757 if (gv && isGV(gv)) {
15759 buffer[0] = gvtype;
15762 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15764 gv_fullname4(name, gv, buffer, 0);
15766 if ((unsigned int)SvPVX(name)[1] <= 26) {
15768 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15770 /* Swap the 1 unprintable control character for the 2 byte pretty
15771 version - ie substr($name, 1, 1) = $buffer; */
15772 sv_insert(name, 1, 1, buffer, 2);
15776 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15779 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15781 if (!cv || !CvPADLIST(cv))
15783 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15784 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15788 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15789 SV * const sv = newSV(0);
15791 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15793 *SvPVX(name) = '$';
15794 Perl_sv_catpvf(aTHX_ name, "{%s}",
15795 pv_pretty(sv, pv, len, 32, NULL, NULL,
15796 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15797 SvREFCNT_dec_NN(sv);
15799 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15800 *SvPVX(name) = '$';
15801 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15803 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15804 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15805 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15813 =for apidoc find_uninit_var
15815 Find the name of the undefined variable (if any) that caused the operator
15816 to issue a "Use of uninitialized value" warning.
15817 If match is true, only return a name if its value matches C<uninit_sv>.
15818 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15819 warning, then following the direct child of the op may yield an
15820 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15821 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15822 the variable name if we get an exact match.
15823 C<desc_p> points to a string pointer holding the description of the op.
15824 This may be updated if needed.
15826 The name is returned as a mortal SV.
15828 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15829 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15835 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15836 bool match, const char **desc_p)
15841 const OP *o, *o2, *kid;
15843 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15845 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15846 uninit_sv == &PL_sv_placeholder)))
15849 switch (obase->op_type) {
15852 /* undef should care if its args are undef - any warnings
15853 * will be from tied/magic vars */
15861 const bool pad = ( obase->op_type == OP_PADAV
15862 || obase->op_type == OP_PADHV
15863 || obase->op_type == OP_PADRANGE
15866 const bool hash = ( obase->op_type == OP_PADHV
15867 || obase->op_type == OP_RV2HV
15868 || (obase->op_type == OP_PADRANGE
15869 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15873 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15875 if (pad) { /* @lex, %lex */
15876 sv = PAD_SVl(obase->op_targ);
15880 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15881 /* @global, %global */
15882 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15885 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15887 else if (obase == PL_op) /* @{expr}, %{expr} */
15888 return find_uninit_var(cUNOPx(obase)->op_first,
15889 uninit_sv, match, desc_p);
15890 else /* @{expr}, %{expr} as a sub-expression */
15894 /* attempt to find a match within the aggregate */
15896 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15898 subscript_type = FUV_SUBSCRIPT_HASH;
15901 index = find_array_subscript((const AV *)sv, uninit_sv);
15903 subscript_type = FUV_SUBSCRIPT_ARRAY;
15906 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15909 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15910 keysv, index, subscript_type);
15914 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15916 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15917 if (!gv || !GvSTASH(gv))
15919 if (match && (GvSV(gv) != uninit_sv))
15921 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15924 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15927 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15929 return varname(NULL, '$', obase->op_targ,
15930 NULL, 0, FUV_SUBSCRIPT_NONE);
15933 gv = cGVOPx_gv(obase);
15934 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15936 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15938 case OP_AELEMFAST_LEX:
15941 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15942 if (!av || SvRMAGICAL(av))
15944 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15945 if (!svp || *svp != uninit_sv)
15948 return varname(NULL, '$', obase->op_targ,
15949 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15952 gv = cGVOPx_gv(obase);
15957 AV *const av = GvAV(gv);
15958 if (!av || SvRMAGICAL(av))
15960 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15961 if (!svp || *svp != uninit_sv)
15964 return varname(gv, '$', 0,
15965 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15967 NOT_REACHED; /* NOTREACHED */
15970 o = cUNOPx(obase)->op_first;
15971 if (!o || o->op_type != OP_NULL ||
15972 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15974 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15979 bool negate = FALSE;
15981 if (PL_op == obase)
15982 /* $a[uninit_expr] or $h{uninit_expr} */
15983 return find_uninit_var(cBINOPx(obase)->op_last,
15984 uninit_sv, match, desc_p);
15987 o = cBINOPx(obase)->op_first;
15988 kid = cBINOPx(obase)->op_last;
15990 /* get the av or hv, and optionally the gv */
15992 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15993 sv = PAD_SV(o->op_targ);
15995 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15996 && cUNOPo->op_first->op_type == OP_GV)
15998 gv = cGVOPx_gv(cUNOPo->op_first);
16002 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16007 if (kid && kid->op_type == OP_NEGATE) {
16009 kid = cUNOPx(kid)->op_first;
16012 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16013 /* index is constant */
16016 kidsv = newSVpvs_flags("-", SVs_TEMP);
16017 sv_catsv(kidsv, cSVOPx_sv(kid));
16020 kidsv = cSVOPx_sv(kid);
16024 if (obase->op_type == OP_HELEM) {
16025 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16026 if (!he || HeVAL(he) != uninit_sv)
16030 SV * const opsv = cSVOPx_sv(kid);
16031 const IV opsviv = SvIV(opsv);
16032 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16033 negate ? - opsviv : opsviv,
16035 if (!svp || *svp != uninit_sv)
16039 if (obase->op_type == OP_HELEM)
16040 return varname(gv, '%', o->op_targ,
16041 kidsv, 0, FUV_SUBSCRIPT_HASH);
16043 return varname(gv, '@', o->op_targ, NULL,
16044 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16045 FUV_SUBSCRIPT_ARRAY);
16048 /* index is an expression;
16049 * attempt to find a match within the aggregate */
16050 if (obase->op_type == OP_HELEM) {
16051 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16053 return varname(gv, '%', o->op_targ,
16054 keysv, 0, FUV_SUBSCRIPT_HASH);
16057 const SSize_t index
16058 = find_array_subscript((const AV *)sv, uninit_sv);
16060 return varname(gv, '@', o->op_targ,
16061 NULL, index, FUV_SUBSCRIPT_ARRAY);
16066 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16068 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16070 NOT_REACHED; /* NOTREACHED */
16073 case OP_MULTIDEREF: {
16074 /* If we were executing OP_MULTIDEREF when the undef warning
16075 * triggered, then it must be one of the index values within
16076 * that triggered it. If not, then the only possibility is that
16077 * the value retrieved by the last aggregate index might be the
16078 * culprit. For the former, we set PL_multideref_pc each time before
16079 * using an index, so work though the item list until we reach
16080 * that point. For the latter, just work through the entire item
16081 * list; the last aggregate retrieved will be the candidate.
16082 * There is a third rare possibility: something triggered
16083 * magic while fetching an array/hash element. Just display
16084 * nothing in this case.
16087 /* the named aggregate, if any */
16088 PADOFFSET agg_targ = 0;
16090 /* the last-seen index */
16092 PADOFFSET index_targ;
16094 IV index_const_iv = 0; /* init for spurious compiler warn */
16095 SV *index_const_sv;
16096 int depth = 0; /* how many array/hash lookups we've done */
16098 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16099 UNOP_AUX_item *last = NULL;
16100 UV actions = items->uv;
16103 if (PL_op == obase) {
16104 last = PL_multideref_pc;
16105 assert(last >= items && last <= items + items[-1].uv);
16112 switch (actions & MDEREF_ACTION_MASK) {
16114 case MDEREF_reload:
16115 actions = (++items)->uv;
16118 case MDEREF_HV_padhv_helem: /* $lex{...} */
16121 case MDEREF_AV_padav_aelem: /* $lex[...] */
16122 agg_targ = (++items)->pad_offset;
16126 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16129 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16131 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16132 assert(isGV_with_GP(agg_gv));
16135 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16136 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16139 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16140 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16146 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16147 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16150 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16151 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16158 index_const_sv = NULL;
16160 index_type = (actions & MDEREF_INDEX_MASK);
16161 switch (index_type) {
16162 case MDEREF_INDEX_none:
16164 case MDEREF_INDEX_const:
16166 index_const_sv = UNOP_AUX_item_sv(++items)
16168 index_const_iv = (++items)->iv;
16170 case MDEREF_INDEX_padsv:
16171 index_targ = (++items)->pad_offset;
16173 case MDEREF_INDEX_gvsv:
16174 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16175 assert(isGV_with_GP(index_gv));
16179 if (index_type != MDEREF_INDEX_none)
16182 if ( index_type == MDEREF_INDEX_none
16183 || (actions & MDEREF_FLAG_last)
16184 || (last && items >= last)
16188 actions >>= MDEREF_SHIFT;
16191 if (PL_op == obase) {
16192 /* most likely index was undef */
16194 *desc_p = ( (actions & MDEREF_FLAG_last)
16195 && (obase->op_private
16196 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16198 (obase->op_private & OPpMULTIDEREF_EXISTS)
16201 : is_hv ? "hash element" : "array element";
16202 assert(index_type != MDEREF_INDEX_none);
16204 if (GvSV(index_gv) == uninit_sv)
16205 return varname(index_gv, '$', 0, NULL, 0,
16206 FUV_SUBSCRIPT_NONE);
16211 if (PL_curpad[index_targ] == uninit_sv)
16212 return varname(NULL, '$', index_targ,
16213 NULL, 0, FUV_SUBSCRIPT_NONE);
16217 /* If we got to this point it was undef on a const subscript,
16218 * so magic probably involved, e.g. $ISA[0]. Give up. */
16222 /* the SV returned by pp_multideref() was undef, if anything was */
16228 sv = PAD_SV(agg_targ);
16230 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16234 if (index_type == MDEREF_INDEX_const) {
16239 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16240 if (!he || HeVAL(he) != uninit_sv)
16244 SV * const * const svp =
16245 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16246 if (!svp || *svp != uninit_sv)
16251 ? varname(agg_gv, '%', agg_targ,
16252 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16253 : varname(agg_gv, '@', agg_targ,
16254 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16257 /* index is an var */
16259 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16261 return varname(agg_gv, '%', agg_targ,
16262 keysv, 0, FUV_SUBSCRIPT_HASH);
16265 const SSize_t index
16266 = find_array_subscript((const AV *)sv, uninit_sv);
16268 return varname(agg_gv, '@', agg_targ,
16269 NULL, index, FUV_SUBSCRIPT_ARRAY);
16273 return varname(agg_gv,
16275 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16277 NOT_REACHED; /* NOTREACHED */
16281 /* only examine RHS */
16282 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16286 o = cUNOPx(obase)->op_first;
16287 if ( o->op_type == OP_PUSHMARK
16288 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16292 if (!OpHAS_SIBLING(o)) {
16293 /* one-arg version of open is highly magical */
16295 if (o->op_type == OP_GV) { /* open FOO; */
16297 if (match && GvSV(gv) != uninit_sv)
16299 return varname(gv, '$', 0,
16300 NULL, 0, FUV_SUBSCRIPT_NONE);
16302 /* other possibilities not handled are:
16303 * open $x; or open my $x; should return '${*$x}'
16304 * open expr; should return '$'.expr ideally
16311 /* ops where $_ may be an implicit arg */
16316 if ( !(obase->op_flags & OPf_STACKED)) {
16317 if (uninit_sv == DEFSV)
16318 return newSVpvs_flags("$_", SVs_TEMP);
16319 else if (obase->op_targ
16320 && uninit_sv == PAD_SVl(obase->op_targ))
16321 return varname(NULL, '$', obase->op_targ, NULL, 0,
16322 FUV_SUBSCRIPT_NONE);
16329 match = 1; /* print etc can return undef on defined args */
16330 /* skip filehandle as it can't produce 'undef' warning */
16331 o = cUNOPx(obase)->op_first;
16332 if ((obase->op_flags & OPf_STACKED)
16334 ( o->op_type == OP_PUSHMARK
16335 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16336 o = OpSIBLING(OpSIBLING(o));
16340 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16341 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16343 /* the following ops are capable of returning PL_sv_undef even for
16344 * defined arg(s) */
16363 case OP_GETPEERNAME:
16410 case OP_SMARTMATCH:
16419 /* XXX tmp hack: these two may call an XS sub, and currently
16420 XS subs don't have a SUB entry on the context stack, so CV and
16421 pad determination goes wrong, and BAD things happen. So, just
16422 don't try to determine the value under those circumstances.
16423 Need a better fix at dome point. DAPM 11/2007 */
16429 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16430 if (gv && GvSV(gv) == uninit_sv)
16431 return newSVpvs_flags("$.", SVs_TEMP);
16436 /* def-ness of rval pos() is independent of the def-ness of its arg */
16437 if ( !(obase->op_flags & OPf_MOD))
16442 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16443 return newSVpvs_flags("${$/}", SVs_TEMP);
16448 if (!(obase->op_flags & OPf_KIDS))
16450 o = cUNOPx(obase)->op_first;
16456 /* This loop checks all the kid ops, skipping any that cannot pos-
16457 * sibly be responsible for the uninitialized value; i.e., defined
16458 * constants and ops that return nothing. If there is only one op
16459 * left that is not skipped, then we *know* it is responsible for
16460 * the uninitialized value. If there is more than one op left, we
16461 * have to look for an exact match in the while() loop below.
16462 * Note that we skip padrange, because the individual pad ops that
16463 * it replaced are still in the tree, so we work on them instead.
16466 for (kid=o; kid; kid = OpSIBLING(kid)) {
16467 const OPCODE type = kid->op_type;
16468 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16469 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16470 || (type == OP_PUSHMARK)
16471 || (type == OP_PADRANGE)
16475 if (o2) { /* more than one found */
16482 return find_uninit_var(o2, uninit_sv, match, desc_p);
16484 /* scan all args */
16486 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16498 =for apidoc report_uninit
16500 Print appropriate "Use of uninitialized variable" warning.
16506 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16508 const char *desc = NULL;
16509 SV* varname = NULL;
16512 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16515 if (uninit_sv && PL_curpad) {
16516 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16518 sv_insert(varname, 0, 0, " ", 1);
16521 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16522 /* we've reached the end of a sort block or sub,
16523 * and the uninit value is probably what that code returned */
16526 /* PL_warn_uninit_sv is constant */
16527 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16529 /* diag_listed_as: Use of uninitialized value%s */
16530 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16531 SVfARG(varname ? varname : &PL_sv_no),
16534 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16540 * ex: set ts=8 sts=4 sw=4 et: