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 *referant = 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 referant = 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 /* referant will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referant;
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.
1570 * If the new size is a big power of two, don't bother: we assume the
1571 * caller wanted a nice 2^N sized block and will be annoyed at getting
1573 * Only increment if the allocation isn't MEM_SIZE_MAX,
1574 * otherwise it will wrap to 0.
1576 if ( (newlen < 0x1000 || (newlen & (newlen - 1)))
1577 && newlen != MEM_SIZE_MAX
1582 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1583 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586 if (newlen > SvLEN(sv)) { /* need more room? */
1587 STRLEN minlen = SvCUR(sv);
1588 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1589 if (newlen < minlen)
1591 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1593 /* Don't round up on the first allocation, as odds are pretty good that
1594 * the initial request is accurate as to what is really needed */
1596 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1597 if (rounded > newlen)
1601 if (SvLEN(sv) && s) {
1602 s = (char*)saferealloc(s, newlen);
1605 s = (char*)safemalloc(newlen);
1606 if (SvPVX_const(sv) && SvCUR(sv)) {
1607 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1611 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Do this here, do it once, do it right, and then we will never get
1613 called back into sv_grow() unless there really is some growing
1615 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1617 SvLEN_set(sv, newlen);
1624 =for apidoc sv_setiv
1626 Copies an integer into the given SV, upgrading first if necessary.
1627 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1633 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1635 PERL_ARGS_ASSERT_SV_SETIV;
1637 SV_CHECK_THINKFIRST_COW_DROP(sv);
1638 switch (SvTYPE(sv)) {
1641 sv_upgrade(sv, SVt_IV);
1644 sv_upgrade(sv, SVt_PVIV);
1648 if (!isGV_with_GP(sv))
1655 /* diag_listed_as: Can't coerce %s to %s in %s */
1656 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1661 (void)SvIOK_only(sv); /* validate number */
1667 =for apidoc sv_setiv_mg
1669 Like C<sv_setiv>, but also handles 'set' magic.
1675 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1677 PERL_ARGS_ASSERT_SV_SETIV_MG;
1684 =for apidoc sv_setuv
1686 Copies an unsigned integer into the given SV, upgrading first if necessary.
1687 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1693 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1695 PERL_ARGS_ASSERT_SV_SETUV;
1697 /* With the if statement to ensure that integers are stored as IVs whenever
1699 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1702 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1704 If you wish to remove the following if statement, so that this routine
1705 (and its callers) always return UVs, please benchmark to see what the
1706 effect is. Modern CPUs may be different. Or may not :-)
1708 if (u <= (UV)IV_MAX) {
1709 sv_setiv(sv, (IV)u);
1718 =for apidoc sv_setuv_mg
1720 Like C<sv_setuv>, but also handles 'set' magic.
1726 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1728 PERL_ARGS_ASSERT_SV_SETUV_MG;
1735 =for apidoc sv_setnv
1737 Copies a double into the given SV, upgrading first if necessary.
1738 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1744 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1746 PERL_ARGS_ASSERT_SV_SETNV;
1748 SV_CHECK_THINKFIRST_COW_DROP(sv);
1749 switch (SvTYPE(sv)) {
1752 sv_upgrade(sv, SVt_NV);
1756 sv_upgrade(sv, SVt_PVNV);
1760 if (!isGV_with_GP(sv))
1767 /* diag_listed_as: Can't coerce %s to %s in %s */
1768 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1774 (void)SvNOK_only(sv); /* validate number */
1779 =for apidoc sv_setnv_mg
1781 Like C<sv_setnv>, but also handles 'set' magic.
1787 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1789 PERL_ARGS_ASSERT_SV_SETNV_MG;
1795 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1796 * not incrementable warning display.
1797 * Originally part of S_not_a_number().
1798 * The return value may be != tmpbuf.
1802 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1805 PERL_ARGS_ASSERT_SV_DISPLAY;
1808 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1809 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1812 const char * const limit = tmpbuf + tmpbuf_size - 8;
1813 /* each *s can expand to 4 chars + "...\0",
1814 i.e. need room for 8 chars */
1816 const char *s = SvPVX_const(sv);
1817 const char * const end = s + SvCUR(sv);
1818 for ( ; s < end && d < limit; s++ ) {
1820 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1824 /* Map to ASCII "equivalent" of Latin1 */
1825 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1831 else if (ch == '\r') {
1835 else if (ch == '\f') {
1839 else if (ch == '\\') {
1843 else if (ch == '\0') {
1847 else if (isPRINT_LC(ch))
1866 /* Print an "isn't numeric" warning, using a cleaned-up,
1867 * printable version of the offending string
1871 S_not_a_number(pTHX_ SV *const sv)
1876 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1878 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1881 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1882 /* diag_listed_as: Argument "%s" isn't numeric%s */
1883 "Argument \"%s\" isn't numeric in %s", pv,
1886 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1887 /* diag_listed_as: Argument "%s" isn't numeric%s */
1888 "Argument \"%s\" isn't numeric", pv);
1892 S_not_incrementable(pTHX_ SV *const sv) {
1896 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1898 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1900 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1901 "Argument \"%s\" treated as 0 in increment (++)", pv);
1905 =for apidoc looks_like_number
1907 Test if the content of an SV looks like a number (or is a number).
1908 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1909 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1916 Perl_looks_like_number(pTHX_ SV *const sv)
1922 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1924 if (SvPOK(sv) || SvPOKp(sv)) {
1925 sbegin = SvPV_nomg_const(sv, len);
1928 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1929 numtype = grok_number(sbegin, len, NULL);
1930 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1934 S_glob_2number(pTHX_ GV * const gv)
1936 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1938 /* We know that all GVs stringify to something that is not-a-number,
1939 so no need to test that. */
1940 if (ckWARN(WARN_NUMERIC))
1942 SV *const buffer = sv_newmortal();
1943 gv_efullname3(buffer, gv, "*");
1944 not_a_number(buffer);
1946 /* We just want something true to return, so that S_sv_2iuv_common
1947 can tail call us and return true. */
1951 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1952 until proven guilty, assume that things are not that bad... */
1957 As 64 bit platforms often have an NV that doesn't preserve all bits of
1958 an IV (an assumption perl has been based on to date) it becomes necessary
1959 to remove the assumption that the NV always carries enough precision to
1960 recreate the IV whenever needed, and that the NV is the canonical form.
1961 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1962 precision as a side effect of conversion (which would lead to insanity
1963 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1964 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1965 where precision was lost, and IV/UV/NV slots that have a valid conversion
1966 which has lost no precision
1967 2) to ensure that if a numeric conversion to one form is requested that
1968 would lose precision, the precise conversion (or differently
1969 imprecise conversion) is also performed and cached, to prevent
1970 requests for different numeric formats on the same SV causing
1971 lossy conversion chains. (lossless conversion chains are perfectly
1976 SvIOKp is true if the IV slot contains a valid value
1977 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1978 SvNOKp is true if the NV slot contains a valid value
1979 SvNOK is true only if the NV value is accurate
1982 while converting from PV to NV, check to see if converting that NV to an
1983 IV(or UV) would lose accuracy over a direct conversion from PV to
1984 IV(or UV). If it would, cache both conversions, return NV, but mark
1985 SV as IOK NOKp (ie not NOK).
1987 While converting from PV to IV, check to see if converting that IV to an
1988 NV would lose accuracy over a direct conversion from PV to NV. If it
1989 would, cache both conversions, flag similarly.
1991 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1992 correctly because if IV & NV were set NV *always* overruled.
1993 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1994 changes - now IV and NV together means that the two are interchangeable:
1995 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1997 The benefit of this is that operations such as pp_add know that if
1998 SvIOK is true for both left and right operands, then integer addition
1999 can be used instead of floating point (for cases where the result won't
2000 overflow). Before, floating point was always used, which could lead to
2001 loss of precision compared with integer addition.
2003 * making IV and NV equal status should make maths accurate on 64 bit
2005 * may speed up maths somewhat if pp_add and friends start to use
2006 integers when possible instead of fp. (Hopefully the overhead in
2007 looking for SvIOK and checking for overflow will not outweigh the
2008 fp to integer speedup)
2009 * will slow down integer operations (callers of SvIV) on "inaccurate"
2010 values, as the change from SvIOK to SvIOKp will cause a call into
2011 sv_2iv each time rather than a macro access direct to the IV slot
2012 * should speed up number->string conversion on integers as IV is
2013 favoured when IV and NV are equally accurate
2015 ####################################################################
2016 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2017 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2018 On the other hand, SvUOK is true iff UV.
2019 ####################################################################
2021 Your mileage will vary depending your CPU's relative fp to integer
2025 #ifndef NV_PRESERVES_UV
2026 # define IS_NUMBER_UNDERFLOW_IV 1
2027 # define IS_NUMBER_UNDERFLOW_UV 2
2028 # define IS_NUMBER_IV_AND_UV 2
2029 # define IS_NUMBER_OVERFLOW_IV 4
2030 # define IS_NUMBER_OVERFLOW_UV 5
2032 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2034 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2036 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2042 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2043 PERL_UNUSED_CONTEXT;
2045 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));
2046 if (SvNVX(sv) < (NV)IV_MIN) {
2047 (void)SvIOKp_on(sv);
2049 SvIV_set(sv, IV_MIN);
2050 return IS_NUMBER_UNDERFLOW_IV;
2052 if (SvNVX(sv) > (NV)UV_MAX) {
2053 (void)SvIOKp_on(sv);
2056 SvUV_set(sv, UV_MAX);
2057 return IS_NUMBER_OVERFLOW_UV;
2059 (void)SvIOKp_on(sv);
2061 /* Can't use strtol etc to convert this string. (See truth table in
2063 if (SvNVX(sv) <= (UV)IV_MAX) {
2064 SvIV_set(sv, I_V(SvNVX(sv)));
2065 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2066 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2068 /* Integer is imprecise. NOK, IOKp */
2070 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2073 SvUV_set(sv, U_V(SvNVX(sv)));
2074 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2075 if (SvUVX(sv) == UV_MAX) {
2076 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2077 possibly be preserved by NV. Hence, it must be overflow.
2079 return IS_NUMBER_OVERFLOW_UV;
2081 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2083 /* Integer is imprecise. NOK, IOKp */
2085 return IS_NUMBER_OVERFLOW_IV;
2087 #endif /* !NV_PRESERVES_UV*/
2089 /* If numtype is infnan, set the NV of the sv accordingly.
2090 * If numtype is anything else, try setting the NV using Atof(PV). */
2092 # pragma warning(push)
2093 # pragma warning(disable:4756;disable:4056)
2096 S_sv_setnv(pTHX_ SV* sv, int numtype)
2098 bool pok = cBOOL(SvPOK(sv));
2101 if ((numtype & IS_NUMBER_INFINITY)) {
2102 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2107 if ((numtype & IS_NUMBER_NAN)) {
2108 SvNV_set(sv, NV_NAN);
2113 SvNV_set(sv, Atof(SvPVX_const(sv)));
2114 /* Purposefully no true nok here, since we don't want to blow
2115 * away the possible IOK/UV of an existing sv. */
2118 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2120 SvPOK_on(sv); /* PV is okay, though. */
2124 # pragma warning(pop)
2128 S_sv_2iuv_common(pTHX_ SV *const sv)
2130 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2133 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2134 * without also getting a cached IV/UV from it at the same time
2135 * (ie PV->NV conversion should detect loss of accuracy and cache
2136 * IV or UV at same time to avoid this. */
2137 /* IV-over-UV optimisation - choose to cache IV if possible */
2139 if (SvTYPE(sv) == SVt_NV)
2140 sv_upgrade(sv, SVt_PVNV);
2142 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2143 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2144 certainly cast into the IV range at IV_MAX, whereas the correct
2145 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2147 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2148 if (Perl_isnan(SvNVX(sv))) {
2154 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2155 SvIV_set(sv, I_V(SvNVX(sv)));
2156 if (SvNVX(sv) == (NV) SvIVX(sv)
2157 #ifndef NV_PRESERVES_UV
2158 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2159 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2160 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2161 /* Don't flag it as "accurately an integer" if the number
2162 came from a (by definition imprecise) NV operation, and
2163 we're outside the range of NV integer precision */
2167 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2169 /* scalar has trailing garbage, eg "42a" */
2171 DEBUG_c(PerlIO_printf(Perl_debug_log,
2172 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2178 /* IV not precise. No need to convert from PV, as NV
2179 conversion would already have cached IV if it detected
2180 that PV->IV would be better than PV->NV->IV
2181 flags already correct - don't set public IOK. */
2182 DEBUG_c(PerlIO_printf(Perl_debug_log,
2183 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2188 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2189 but the cast (NV)IV_MIN rounds to a the value less (more
2190 negative) than IV_MIN which happens to be equal to SvNVX ??
2191 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2192 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2193 (NV)UVX == NVX are both true, but the values differ. :-(
2194 Hopefully for 2s complement IV_MIN is something like
2195 0x8000000000000000 which will be exact. NWC */
2198 SvUV_set(sv, U_V(SvNVX(sv)));
2200 (SvNVX(sv) == (NV) SvUVX(sv))
2201 #ifndef NV_PRESERVES_UV
2202 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2203 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2204 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2205 /* Don't flag it as "accurately an integer" if the number
2206 came from a (by definition imprecise) NV operation, and
2207 we're outside the range of NV integer precision */
2213 DEBUG_c(PerlIO_printf(Perl_debug_log,
2214 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2220 else if (SvPOKp(sv)) {
2223 const char *s = SvPVX_const(sv);
2224 const STRLEN cur = SvCUR(sv);
2226 /* short-cut for a single digit string like "1" */
2231 if (SvTYPE(sv) < SVt_PVIV)
2232 sv_upgrade(sv, SVt_PVIV);
2234 SvIV_set(sv, (IV)(c - '0'));
2239 numtype = grok_number(s, cur, &value);
2240 /* We want to avoid a possible problem when we cache an IV/ a UV which
2241 may be later translated to an NV, and the resulting NV is not
2242 the same as the direct translation of the initial string
2243 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2244 be careful to ensure that the value with the .456 is around if the
2245 NV value is requested in the future).
2247 This means that if we cache such an IV/a UV, we need to cache the
2248 NV as well. Moreover, we trade speed for space, and do not
2249 cache the NV if we are sure it's not needed.
2252 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2253 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2254 == IS_NUMBER_IN_UV) {
2255 /* It's definitely an integer, only upgrade to PVIV */
2256 if (SvTYPE(sv) < SVt_PVIV)
2257 sv_upgrade(sv, SVt_PVIV);
2259 } else if (SvTYPE(sv) < SVt_PVNV)
2260 sv_upgrade(sv, SVt_PVNV);
2262 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2263 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2265 S_sv_setnv(aTHX_ sv, numtype);
2269 /* If NVs preserve UVs then we only use the UV value if we know that
2270 we aren't going to call atof() below. If NVs don't preserve UVs
2271 then the value returned may have more precision than atof() will
2272 return, even though value isn't perfectly accurate. */
2273 if ((numtype & (IS_NUMBER_IN_UV
2274 #ifdef NV_PRESERVES_UV
2277 )) == IS_NUMBER_IN_UV) {
2278 /* This won't turn off the public IOK flag if it was set above */
2279 (void)SvIOKp_on(sv);
2281 if (!(numtype & IS_NUMBER_NEG)) {
2283 if (value <= (UV)IV_MAX) {
2284 SvIV_set(sv, (IV)value);
2286 /* it didn't overflow, and it was positive. */
2287 SvUV_set(sv, value);
2291 /* 2s complement assumption */
2292 if (value <= (UV)IV_MIN) {
2293 SvIV_set(sv, value == (UV)IV_MIN
2294 ? IV_MIN : -(IV)value);
2296 /* Too negative for an IV. This is a double upgrade, but
2297 I'm assuming it will be rare. */
2298 if (SvTYPE(sv) < SVt_PVNV)
2299 sv_upgrade(sv, SVt_PVNV);
2303 SvNV_set(sv, -(NV)value);
2304 SvIV_set(sv, IV_MIN);
2308 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2309 will be in the previous block to set the IV slot, and the next
2310 block to set the NV slot. So no else here. */
2312 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2313 != IS_NUMBER_IN_UV) {
2314 /* It wasn't an (integer that doesn't overflow the UV). */
2315 S_sv_setnv(aTHX_ sv, numtype);
2317 if (! numtype && ckWARN(WARN_NUMERIC))
2320 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2321 PTR2UV(sv), SvNVX(sv)));
2323 #ifdef NV_PRESERVES_UV
2324 (void)SvIOKp_on(sv);
2326 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2327 if (Perl_isnan(SvNVX(sv))) {
2333 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2334 SvIV_set(sv, I_V(SvNVX(sv)));
2335 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2338 NOOP; /* Integer is imprecise. NOK, IOKp */
2340 /* UV will not work better than IV */
2342 if (SvNVX(sv) > (NV)UV_MAX) {
2344 /* Integer is inaccurate. NOK, IOKp, is UV */
2345 SvUV_set(sv, UV_MAX);
2347 SvUV_set(sv, U_V(SvNVX(sv)));
2348 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2349 NV preservse UV so can do correct comparison. */
2350 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2353 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2358 #else /* NV_PRESERVES_UV */
2359 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2360 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2361 /* The IV/UV slot will have been set from value returned by
2362 grok_number above. The NV slot has just been set using
2365 assert (SvIOKp(sv));
2367 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2368 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2369 /* Small enough to preserve all bits. */
2370 (void)SvIOKp_on(sv);
2372 SvIV_set(sv, I_V(SvNVX(sv)));
2373 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2375 /* Assumption: first non-preserved integer is < IV_MAX,
2376 this NV is in the preserved range, therefore: */
2377 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2379 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);
2383 0 0 already failed to read UV.
2384 0 1 already failed to read UV.
2385 1 0 you won't get here in this case. IV/UV
2386 slot set, public IOK, Atof() unneeded.
2387 1 1 already read UV.
2388 so there's no point in sv_2iuv_non_preserve() attempting
2389 to use atol, strtol, strtoul etc. */
2391 sv_2iuv_non_preserve (sv, numtype);
2393 sv_2iuv_non_preserve (sv);
2397 #endif /* NV_PRESERVES_UV */
2398 /* It might be more code efficient to go through the entire logic above
2399 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2400 gets complex and potentially buggy, so more programmer efficient
2401 to do it this way, by turning off the public flags: */
2403 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2407 if (isGV_with_GP(sv))
2408 return glob_2number(MUTABLE_GV(sv));
2410 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2412 if (SvTYPE(sv) < SVt_IV)
2413 /* Typically the caller expects that sv_any is not NULL now. */
2414 sv_upgrade(sv, SVt_IV);
2415 /* Return 0 from the caller. */
2422 =for apidoc sv_2iv_flags
2424 Return the integer value of an SV, doing any necessary string
2425 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2426 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2432 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2434 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2436 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2437 && SvTYPE(sv) != SVt_PVFM);
2439 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2445 if (flags & SV_SKIP_OVERLOAD)
2447 tmpstr = AMG_CALLunary(sv, numer_amg);
2448 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2449 return SvIV(tmpstr);
2452 return PTR2IV(SvRV(sv));
2455 if (SvVALID(sv) || isREGEXP(sv)) {
2456 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2457 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2458 In practice they are extremely unlikely to actually get anywhere
2459 accessible by user Perl code - the only way that I'm aware of is when
2460 a constant subroutine which is used as the second argument to index.
2462 Regexps have no SvIVX and SvNVX fields.
2464 assert(isREGEXP(sv) || SvPOKp(sv));
2467 const char * const ptr =
2468 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2470 = grok_number(ptr, SvCUR(sv), &value);
2472 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2473 == IS_NUMBER_IN_UV) {
2474 /* It's definitely an integer */
2475 if (numtype & IS_NUMBER_NEG) {
2476 if (value < (UV)IV_MIN)
2479 if (value < (UV)IV_MAX)
2484 /* Quite wrong but no good choices. */
2485 if ((numtype & IS_NUMBER_INFINITY)) {
2486 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2487 } else if ((numtype & IS_NUMBER_NAN)) {
2488 return 0; /* So wrong. */
2492 if (ckWARN(WARN_NUMERIC))
2495 return I_V(Atof(ptr));
2499 if (SvTHINKFIRST(sv)) {
2500 if (SvREADONLY(sv) && !SvOK(sv)) {
2501 if (ckWARN(WARN_UNINITIALIZED))
2508 if (S_sv_2iuv_common(aTHX_ sv))
2512 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2513 PTR2UV(sv),SvIVX(sv)));
2514 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2518 =for apidoc sv_2uv_flags
2520 Return the unsigned integer value of an SV, doing any necessary string
2521 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2522 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2528 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2530 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2532 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2538 if (flags & SV_SKIP_OVERLOAD)
2540 tmpstr = AMG_CALLunary(sv, numer_amg);
2541 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2542 return SvUV(tmpstr);
2545 return PTR2UV(SvRV(sv));
2548 if (SvVALID(sv) || isREGEXP(sv)) {
2549 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2550 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2551 Regexps have no SvIVX and SvNVX fields. */
2552 assert(isREGEXP(sv) || SvPOKp(sv));
2555 const char * const ptr =
2556 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2558 = grok_number(ptr, SvCUR(sv), &value);
2560 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2561 == IS_NUMBER_IN_UV) {
2562 /* It's definitely an integer */
2563 if (!(numtype & IS_NUMBER_NEG))
2567 /* Quite wrong but no good choices. */
2568 if ((numtype & IS_NUMBER_INFINITY)) {
2569 return UV_MAX; /* So wrong. */
2570 } else if ((numtype & IS_NUMBER_NAN)) {
2571 return 0; /* So wrong. */
2575 if (ckWARN(WARN_NUMERIC))
2578 return U_V(Atof(ptr));
2582 if (SvTHINKFIRST(sv)) {
2583 if (SvREADONLY(sv) && !SvOK(sv)) {
2584 if (ckWARN(WARN_UNINITIALIZED))
2591 if (S_sv_2iuv_common(aTHX_ sv))
2595 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2596 PTR2UV(sv),SvUVX(sv)));
2597 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2601 =for apidoc sv_2nv_flags
2603 Return the num value of an SV, doing any necessary string or integer
2604 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2605 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2611 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2613 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2615 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2616 && SvTYPE(sv) != SVt_PVFM);
2617 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2618 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2619 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2620 Regexps have no SvIVX and SvNVX fields. */
2622 if (flags & SV_GMAGIC)
2626 if (SvPOKp(sv) && !SvIOKp(sv)) {
2627 ptr = SvPVX_const(sv);
2629 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2630 !grok_number(ptr, SvCUR(sv), NULL))
2636 return (NV)SvUVX(sv);
2638 return (NV)SvIVX(sv);
2644 ptr = RX_WRAPPED((REGEXP *)sv);
2647 assert(SvTYPE(sv) >= SVt_PVMG);
2648 /* This falls through to the report_uninit near the end of the
2650 } else if (SvTHINKFIRST(sv)) {
2655 if (flags & SV_SKIP_OVERLOAD)
2657 tmpstr = AMG_CALLunary(sv, numer_amg);
2658 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2659 return SvNV(tmpstr);
2662 return PTR2NV(SvRV(sv));
2664 if (SvREADONLY(sv) && !SvOK(sv)) {
2665 if (ckWARN(WARN_UNINITIALIZED))
2670 if (SvTYPE(sv) < SVt_NV) {
2671 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2672 sv_upgrade(sv, SVt_NV);
2674 STORE_NUMERIC_LOCAL_SET_STANDARD();
2675 PerlIO_printf(Perl_debug_log,
2676 "0x%"UVxf" num(%" NVgf ")\n",
2677 PTR2UV(sv), SvNVX(sv));
2678 RESTORE_NUMERIC_LOCAL();
2681 else if (SvTYPE(sv) < SVt_PVNV)
2682 sv_upgrade(sv, SVt_PVNV);
2687 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2688 #ifdef NV_PRESERVES_UV
2694 /* Only set the public NV OK flag if this NV preserves the IV */
2695 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2697 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2698 : (SvIVX(sv) == I_V(SvNVX(sv))))
2704 else if (SvPOKp(sv)) {
2706 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2707 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2709 #ifdef NV_PRESERVES_UV
2710 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2711 == IS_NUMBER_IN_UV) {
2712 /* It's definitely an integer */
2713 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2715 S_sv_setnv(aTHX_ sv, numtype);
2722 SvNV_set(sv, Atof(SvPVX_const(sv)));
2723 /* Only set the public NV OK flag if this NV preserves the value in
2724 the PV at least as well as an IV/UV would.
2725 Not sure how to do this 100% reliably. */
2726 /* if that shift count is out of range then Configure's test is
2727 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2729 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2730 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2731 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2732 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2733 /* Can't use strtol etc to convert this string, so don't try.
2734 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2737 /* value has been set. It may not be precise. */
2738 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2739 /* 2s complement assumption for (UV)IV_MIN */
2740 SvNOK_on(sv); /* Integer is too negative. */
2745 if (numtype & IS_NUMBER_NEG) {
2746 /* -IV_MIN is undefined, but we should never reach
2747 * this point with both IS_NUMBER_NEG and value ==
2749 assert(value != (UV)IV_MIN);
2750 SvIV_set(sv, -(IV)value);
2751 } else if (value <= (UV)IV_MAX) {
2752 SvIV_set(sv, (IV)value);
2754 SvUV_set(sv, value);
2758 if (numtype & IS_NUMBER_NOT_INT) {
2759 /* I believe that even if the original PV had decimals,
2760 they are lost beyond the limit of the FP precision.
2761 However, neither is canonical, so both only get p
2762 flags. NWC, 2000/11/25 */
2763 /* Both already have p flags, so do nothing */
2765 const NV nv = SvNVX(sv);
2766 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2767 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2768 if (SvIVX(sv) == I_V(nv)) {
2771 /* It had no "." so it must be integer. */
2775 /* between IV_MAX and NV(UV_MAX).
2776 Could be slightly > UV_MAX */
2778 if (numtype & IS_NUMBER_NOT_INT) {
2779 /* UV and NV both imprecise. */
2781 const UV nv_as_uv = U_V(nv);
2783 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2792 /* It might be more code efficient to go through the entire logic above
2793 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2794 gets complex and potentially buggy, so more programmer efficient
2795 to do it this way, by turning off the public flags: */
2797 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2798 #endif /* NV_PRESERVES_UV */
2801 if (isGV_with_GP(sv)) {
2802 glob_2number(MUTABLE_GV(sv));
2806 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2808 assert (SvTYPE(sv) >= SVt_NV);
2809 /* Typically the caller expects that sv_any is not NULL now. */
2810 /* XXX Ilya implies that this is a bug in callers that assume this
2811 and ideally should be fixed. */
2815 STORE_NUMERIC_LOCAL_SET_STANDARD();
2816 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2817 PTR2UV(sv), SvNVX(sv));
2818 RESTORE_NUMERIC_LOCAL();
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2834 Perl_sv_2num(pTHX_ SV *const sv)
2836 PERL_ARGS_ASSERT_SV_2NUM;
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2849 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2850 * UV as a string towards the end of buf, and return pointers to start and
2853 * We assume that buf is at least TYPE_CHARS(UV) long.
2857 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2859 char *ptr = buf + TYPE_CHARS(UV);
2860 char * const ebuf = ptr;
2863 PERL_ARGS_ASSERT_UIV_2BUF;
2871 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2875 *--ptr = '0' + (char)(uv % 10);
2883 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2884 * infinity or a not-a-number, writes the appropriate strings to the
2885 * buffer, including a zero byte. On success returns the written length,
2886 * excluding the zero byte, on failure (not an infinity, not a nan)
2887 * returns zero, assert-fails on maxlen being too short.
2889 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2890 * shared string constants we point to, instead of generating a new
2891 * string for each instance. */
2893 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2895 assert(maxlen >= 4);
2896 if (Perl_isinf(nv)) {
2898 if (maxlen < 5) /* "-Inf\0" */
2908 else if (Perl_isnan(nv)) {
2912 /* XXX optionally output the payload mantissa bits as
2913 * "(unsigned)" (to match the nan("...") C99 function,
2914 * or maybe as "(0xhhh...)" would make more sense...
2915 * provide a format string so that the user can decide?
2916 * NOTE: would affect the maxlen and assert() logic.*/
2921 assert((s == buffer + 3) || (s == buffer + 4));
2923 return s - buffer - 1; /* -1: excluding the zero byte */
2927 =for apidoc sv_2pv_flags
2929 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2930 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2931 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2932 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2938 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2942 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2944 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2945 && SvTYPE(sv) != SVt_PVFM);
2946 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2951 if (flags & SV_SKIP_OVERLOAD)
2953 tmpstr = AMG_CALLunary(sv, string_amg);
2954 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2955 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2957 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2961 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2962 if (flags & SV_CONST_RETURN) {
2963 pv = (char *) SvPVX_const(tmpstr);
2965 pv = (flags & SV_MUTABLE_RETURN)
2966 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2969 *lp = SvCUR(tmpstr);
2971 pv = sv_2pv_flags(tmpstr, lp, flags);
2984 SV *const referent = SvRV(sv);
2988 retval = buffer = savepvn("NULLREF", len);
2989 } else if (SvTYPE(referent) == SVt_REGEXP &&
2990 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2991 amagic_is_enabled(string_amg))) {
2992 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2996 /* If the regex is UTF-8 we want the containing scalar to
2997 have an UTF-8 flag too */
3004 *lp = RX_WRAPLEN(re);
3006 return RX_WRAPPED(re);
3008 const char *const typestr = sv_reftype(referent, 0);
3009 const STRLEN typelen = strlen(typestr);
3010 UV addr = PTR2UV(referent);
3011 const char *stashname = NULL;
3012 STRLEN stashnamelen = 0; /* hush, gcc */
3013 const char *buffer_end;
3015 if (SvOBJECT(referent)) {
3016 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3019 stashname = HEK_KEY(name);
3020 stashnamelen = HEK_LEN(name);
3022 if (HEK_UTF8(name)) {
3028 stashname = "__ANON__";
3031 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3032 + 2 * sizeof(UV) + 2 /* )\0 */;
3034 len = typelen + 3 /* (0x */
3035 + 2 * sizeof(UV) + 2 /* )\0 */;
3038 Newx(buffer, len, char);
3039 buffer_end = retval = buffer + len;
3041 /* Working backwards */
3045 *--retval = PL_hexdigit[addr & 15];
3046 } while (addr >>= 4);
3052 memcpy(retval, typestr, typelen);
3056 retval -= stashnamelen;
3057 memcpy(retval, stashname, stashnamelen);
3059 /* retval may not necessarily have reached the start of the
3061 assert (retval >= buffer);
3063 len = buffer_end - retval - 1; /* -1 for that \0 */
3075 if (flags & SV_MUTABLE_RETURN)
3076 return SvPVX_mutable(sv);
3077 if (flags & SV_CONST_RETURN)
3078 return (char *)SvPVX_const(sv);
3083 /* I'm assuming that if both IV and NV are equally valid then
3084 converting the IV is going to be more efficient */
3085 const U32 isUIOK = SvIsUV(sv);
3086 char buf[TYPE_CHARS(UV)];
3090 if (SvTYPE(sv) < SVt_PVIV)
3091 sv_upgrade(sv, SVt_PVIV);
3092 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3094 /* inlined from sv_setpvn */
3095 s = SvGROW_mutable(sv, len + 1);
3096 Move(ptr, s, len, char);
3101 else if (SvNOK(sv)) {
3102 if (SvTYPE(sv) < SVt_PVNV)
3103 sv_upgrade(sv, SVt_PVNV);
3104 if (SvNVX(sv) == 0.0
3105 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3106 && !Perl_isnan(SvNVX(sv))
3109 s = SvGROW_mutable(sv, 2);
3114 STRLEN size = 5; /* "-Inf\0" */
3116 s = SvGROW_mutable(sv, size);
3117 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3123 /* some Xenix systems wipe out errno here */
3132 5 + /* exponent digits */
3136 s = SvGROW_mutable(sv, size);
3137 #ifndef USE_LOCALE_NUMERIC
3138 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3144 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3145 STORE_LC_NUMERIC_SET_TO_NEEDED();
3147 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3148 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3149 size += SvLEN(PL_numeric_radix_sv) - 1;
3150 s = SvGROW_mutable(sv, size);
3153 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3155 /* If the radix character is UTF-8, and actually is in the
3156 * output, turn on the UTF-8 flag for the scalar */
3158 && SvUTF8(PL_numeric_radix_sv)
3159 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3164 RESTORE_LC_NUMERIC();
3167 /* We don't call SvPOK_on(), because it may come to
3168 * pass that the locale changes so that the
3169 * stringification we just did is no longer correct. We
3170 * will have to re-stringify every time it is needed */
3177 else if (isGV_with_GP(sv)) {
3178 GV *const gv = MUTABLE_GV(sv);
3179 SV *const buffer = sv_newmortal();
3181 gv_efullname3(buffer, gv, "*");
3183 assert(SvPOK(buffer));
3187 *lp = SvCUR(buffer);
3188 return SvPVX(buffer);
3190 else if (isREGEXP(sv)) {
3191 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3192 return RX_WRAPPED((REGEXP *)sv);
3197 if (flags & SV_UNDEF_RETURNS_NULL)
3199 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3201 /* Typically the caller expects that sv_any is not NULL now. */
3202 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3203 sv_upgrade(sv, SVt_PV);
3208 const STRLEN len = s - SvPVX_const(sv);
3213 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3214 PTR2UV(sv),SvPVX_const(sv)));
3215 if (flags & SV_CONST_RETURN)
3216 return (char *)SvPVX_const(sv);
3217 if (flags & SV_MUTABLE_RETURN)
3218 return SvPVX_mutable(sv);
3223 =for apidoc sv_copypv
3225 Copies a stringified representation of the source SV into the
3226 destination SV. Automatically performs any necessary C<mg_get> and
3227 coercion of numeric values into strings. Guaranteed to preserve
3228 C<UTF8> flag even from overloaded objects. Similar in nature to
3229 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3230 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3231 would lose the UTF-8'ness of the PV.
3233 =for apidoc sv_copypv_nomg
3235 Like C<sv_copypv>, but doesn't invoke get magic first.
3237 =for apidoc sv_copypv_flags
3239 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3240 has the C<SV_GMAGIC> bit set.
3246 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3251 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3253 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3254 sv_setpvn(dsv,s,len);
3262 =for apidoc sv_2pvbyte
3264 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3265 to its length. May cause the SV to be downgraded from UTF-8 as a
3268 Usually accessed via the C<SvPVbyte> macro.
3274 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3276 PERL_ARGS_ASSERT_SV_2PVBYTE;
3279 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3280 || isGV_with_GP(sv) || SvROK(sv)) {
3281 SV *sv2 = sv_newmortal();
3282 sv_copypv_nomg(sv2,sv);
3285 sv_utf8_downgrade(sv,0);
3286 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3290 =for apidoc sv_2pvutf8
3292 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3293 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3295 Usually accessed via the C<SvPVutf8> macro.
3301 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3303 PERL_ARGS_ASSERT_SV_2PVUTF8;
3305 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3306 || isGV_with_GP(sv) || SvROK(sv))
3307 sv = sv_mortalcopy(sv);
3310 sv_utf8_upgrade_nomg(sv);
3311 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3316 =for apidoc sv_2bool
3318 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3319 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3320 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3322 =for apidoc sv_2bool_flags
3324 This function is only used by C<sv_true()> and friends, and only if
3325 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3326 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3333 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3335 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3338 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3344 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3345 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3348 if(SvGMAGICAL(sv)) {
3350 goto restart; /* call sv_2bool */
3352 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3353 else if(!SvOK(sv)) {
3356 else if(SvPOK(sv)) {
3357 svb = SvPVXtrue(sv);
3359 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3360 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3361 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3365 goto restart; /* call sv_2bool_nomg */
3370 return SvRV(sv) != 0;
3374 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3375 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3379 =for apidoc sv_utf8_upgrade
3381 Converts the PV of an SV to its UTF-8-encoded form.
3382 Forces the SV to string form if it is not already.
3383 Will C<mg_get> on C<sv> if appropriate.
3384 Always sets the C<SvUTF8> flag to avoid future validity checks even
3385 if the whole string is the same in UTF-8 as not.
3386 Returns the number of bytes in the converted string
3388 This is not a general purpose byte encoding to Unicode interface:
3389 use the Encode extension for that.
3391 =for apidoc sv_utf8_upgrade_nomg
3393 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3395 =for apidoc sv_utf8_upgrade_flags
3397 Converts the PV of an SV to its UTF-8-encoded form.
3398 Forces the SV to string form if it is not already.
3399 Always sets the SvUTF8 flag to avoid future validity checks even
3400 if all the bytes are invariant in UTF-8.
3401 If C<flags> has C<SV_GMAGIC> bit set,
3402 will C<mg_get> on C<sv> if appropriate, else not.
3404 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3405 will expand when converted to UTF-8, and skips the extra work of checking for
3406 that. Typically this flag is used by a routine that has already parsed the
3407 string and found such characters, and passes this information on so that the
3408 work doesn't have to be repeated.
3410 Returns the number of bytes in the converted string.
3412 This is not a general purpose byte encoding to Unicode interface:
3413 use the Encode extension for that.
3415 =for apidoc sv_utf8_upgrade_flags_grow
3417 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3418 the number of unused bytes the string of C<sv> is guaranteed to have free after
3419 it upon return. This allows the caller to reserve extra space that it intends
3420 to fill, to avoid extra grows.
3422 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3423 are implemented in terms of this function.
3425 Returns the number of bytes in the converted string (not including the spares).
3429 (One might think that the calling routine could pass in the position of the
3430 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3431 have to be found again. But that is not the case, because typically when the
3432 caller is likely to use this flag, it won't be calling this routine unless it
3433 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3434 and just use bytes. But some things that do fit into a byte are variants in
3435 utf8, and the caller may not have been keeping track of these.)
3437 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3438 C<NUL> isn't guaranteed due to having other routines do the work in some input
3439 cases, or if the input is already flagged as being in utf8.
3441 The speed of this could perhaps be improved for many cases if someone wanted to
3442 write a fast function that counts the number of variant characters in a string,
3443 especially if it could return the position of the first one.
3448 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3450 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3452 if (sv == &PL_sv_undef)
3454 if (!SvPOK_nog(sv)) {
3456 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3457 (void) sv_2pv_flags(sv,&len, flags);
3459 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3463 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3468 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3473 S_sv_uncow(aTHX_ sv, 0);
3476 if (SvCUR(sv) == 0) {
3477 if (extra) SvGROW(sv, extra);
3478 } else { /* Assume Latin-1/EBCDIC */
3479 /* This function could be much more efficient if we
3480 * had a FLAG in SVs to signal if there are any variant
3481 * chars in the PV. Given that there isn't such a flag
3482 * make the loop as fast as possible (although there are certainly ways
3483 * to speed this up, eg. through vectorization) */
3484 U8 * s = (U8 *) SvPVX_const(sv);
3485 U8 * e = (U8 *) SvEND(sv);
3487 STRLEN two_byte_count = 0;
3489 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3491 /* See if really will need to convert to utf8. We mustn't rely on our
3492 * incoming SV being well formed and having a trailing '\0', as certain
3493 * code in pp_formline can send us partially built SVs. */
3497 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3499 t--; /* t already incremented; re-point to first variant */
3504 /* utf8 conversion not needed because all are invariants. Mark as
3505 * UTF-8 even if no variant - saves scanning loop */
3507 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3512 /* Here, the string should be converted to utf8, either because of an
3513 * input flag (two_byte_count = 0), or because a character that
3514 * requires 2 bytes was found (two_byte_count = 1). t points either to
3515 * the beginning of the string (if we didn't examine anything), or to
3516 * the first variant. In either case, everything from s to t - 1 will
3517 * occupy only 1 byte each on output.
3519 * There are two main ways to convert. One is to create a new string
3520 * and go through the input starting from the beginning, appending each
3521 * converted value onto the new string as we go along. It's probably
3522 * best to allocate enough space in the string for the worst possible
3523 * case rather than possibly running out of space and having to
3524 * reallocate and then copy what we've done so far. Since everything
3525 * from s to t - 1 is invariant, the destination can be initialized
3526 * with these using a fast memory copy
3528 * The other way is to figure out exactly how big the string should be
3529 * by parsing the entire input. Then you don't have to make it big
3530 * enough to handle the worst possible case, and more importantly, if
3531 * the string you already have is large enough, you don't have to
3532 * allocate a new string, you can copy the last character in the input
3533 * string to the final position(s) that will be occupied by the
3534 * converted string and go backwards, stopping at t, since everything
3535 * before that is invariant.
3537 * There are advantages and disadvantages to each method.
3539 * In the first method, we can allocate a new string, do the memory
3540 * copy from the s to t - 1, and then proceed through the rest of the
3541 * string byte-by-byte.
3543 * In the second method, we proceed through the rest of the input
3544 * string just calculating how big the converted string will be. Then
3545 * there are two cases:
3546 * 1) if the string has enough extra space to handle the converted
3547 * value. We go backwards through the string, converting until we
3548 * get to the position we are at now, and then stop. If this
3549 * position is far enough along in the string, this method is
3550 * faster than the other method. If the memory copy were the same
3551 * speed as the byte-by-byte loop, that position would be about
3552 * half-way, as at the half-way mark, parsing to the end and back
3553 * is one complete string's parse, the same amount as starting
3554 * over and going all the way through. Actually, it would be
3555 * somewhat less than half-way, as it's faster to just count bytes
3556 * than to also copy, and we don't have the overhead of allocating
3557 * a new string, changing the scalar to use it, and freeing the
3558 * existing one. But if the memory copy is fast, the break-even
3559 * point is somewhere after half way. The counting loop could be
3560 * sped up by vectorization, etc, to move the break-even point
3561 * further towards the beginning.
3562 * 2) if the string doesn't have enough space to handle the converted
3563 * value. A new string will have to be allocated, and one might
3564 * as well, given that, start from the beginning doing the first
3565 * method. We've spent extra time parsing the string and in
3566 * exchange all we've gotten is that we know precisely how big to
3567 * make the new one. Perl is more optimized for time than space,
3568 * so this case is a loser.
3569 * So what I've decided to do is not use the 2nd method unless it is
3570 * guaranteed that a new string won't have to be allocated, assuming
3571 * the worst case. I also decided not to put any more conditions on it
3572 * than this, for now. It seems likely that, since the worst case is
3573 * twice as big as the unknown portion of the string (plus 1), we won't
3574 * be guaranteed enough space, causing us to go to the first method,
3575 * unless the string is short, or the first variant character is near
3576 * the end of it. In either of these cases, it seems best to use the
3577 * 2nd method. The only circumstance I can think of where this would
3578 * be really slower is if the string had once had much more data in it
3579 * than it does now, but there is still a substantial amount in it */
3582 STRLEN invariant_head = t - s;
3583 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3584 if (SvLEN(sv) < size) {
3586 /* Here, have decided to allocate a new string */
3591 Newx(dst, size, U8);
3593 /* If no known invariants at the beginning of the input string,
3594 * set so starts from there. Otherwise, can use memory copy to
3595 * get up to where we are now, and then start from here */
3597 if (invariant_head == 0) {
3600 Copy(s, dst, invariant_head, char);
3601 d = dst + invariant_head;
3605 append_utf8_from_native_byte(*t, &d);
3609 SvPV_free(sv); /* No longer using pre-existing string */
3610 SvPV_set(sv, (char*)dst);
3611 SvCUR_set(sv, d - dst);
3612 SvLEN_set(sv, size);
3615 /* Here, have decided to get the exact size of the string.
3616 * Currently this happens only when we know that there is
3617 * guaranteed enough space to fit the converted string, so
3618 * don't have to worry about growing. If two_byte_count is 0,
3619 * then t points to the first byte of the string which hasn't
3620 * been examined yet. Otherwise two_byte_count is 1, and t
3621 * points to the first byte in the string that will expand to
3622 * two. Depending on this, start examining at t or 1 after t.
3625 U8 *d = t + two_byte_count;
3628 /* Count up the remaining bytes that expand to two */
3631 const U8 chr = *d++;
3632 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3635 /* The string will expand by just the number of bytes that
3636 * occupy two positions. But we are one afterwards because of
3637 * the increment just above. This is the place to put the
3638 * trailing NUL, and to set the length before we decrement */
3640 d += two_byte_count;
3641 SvCUR_set(sv, d - s);
3645 /* Having decremented d, it points to the position to put the
3646 * very last byte of the expanded string. Go backwards through
3647 * the string, copying and expanding as we go, stopping when we
3648 * get to the part that is invariant the rest of the way down */
3652 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3655 *d-- = UTF8_EIGHT_BIT_LO(*e);
3656 *d-- = UTF8_EIGHT_BIT_HI(*e);
3662 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3663 /* Update pos. We do it at the end rather than during
3664 * the upgrade, to avoid slowing down the common case
3665 * (upgrade without pos).
3666 * pos can be stored as either bytes or characters. Since
3667 * this was previously a byte string we can just turn off
3668 * the bytes flag. */
3669 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3671 mg->mg_flags &= ~MGf_BYTES;
3673 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3674 magic_setutf8(sv,mg); /* clear UTF8 cache */
3679 /* Mark as UTF-8 even if no variant - saves scanning loop */
3685 =for apidoc sv_utf8_downgrade
3687 Attempts to convert the PV of an SV from characters to bytes.
3688 If the PV contains a character that cannot fit
3689 in a byte, this conversion will fail;
3690 in this case, either returns false or, if C<fail_ok> is not
3693 This is not a general purpose Unicode to byte encoding interface:
3694 use the C<Encode> extension for that.
3700 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3702 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3704 if (SvPOKp(sv) && SvUTF8(sv)) {
3708 int mg_flags = SV_GMAGIC;
3711 S_sv_uncow(aTHX_ sv, 0);
3713 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3715 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3716 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3717 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3718 SV_GMAGIC|SV_CONST_RETURN);
3719 mg_flags = 0; /* sv_pos_b2u does get magic */
3721 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3722 magic_setutf8(sv,mg); /* clear UTF8 cache */
3725 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3727 if (!utf8_to_bytes(s, &len)) {
3732 Perl_croak(aTHX_ "Wide character in %s",
3735 Perl_croak(aTHX_ "Wide character");
3746 =for apidoc sv_utf8_encode
3748 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3749 flag off so that it looks like octets again.
3755 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3757 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3759 if (SvREADONLY(sv)) {
3760 sv_force_normal_flags(sv, 0);
3762 (void) sv_utf8_upgrade(sv);
3767 =for apidoc sv_utf8_decode
3769 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3770 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3771 so that it looks like a character. If the PV contains only single-byte
3772 characters, the C<SvUTF8> flag stays off.
3773 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3779 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3781 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3784 const U8 *start, *c;
3786 /* The octets may have got themselves encoded - get them back as
3789 if (!sv_utf8_downgrade(sv, TRUE))
3792 /* it is actually just a matter of turning the utf8 flag on, but
3793 * we want to make sure everything inside is valid utf8 first.
3795 c = start = (const U8 *) SvPVX_const(sv);
3796 if (!is_utf8_string(c, SvCUR(sv)))
3798 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3801 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3802 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3803 after this, clearing pos. Does anything on CPAN
3805 /* adjust pos to the start of a UTF8 char sequence */
3806 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3808 I32 pos = mg->mg_len;
3810 for (c = start + pos; c > start; c--) {
3811 if (UTF8_IS_START(*c))
3814 mg->mg_len = c - start;
3817 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3818 magic_setutf8(sv,mg); /* clear UTF8 cache */
3825 =for apidoc sv_setsv
3827 Copies the contents of the source SV C<ssv> into the destination SV
3828 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3829 function if the source SV needs to be reused. Does not handle 'set' magic on
3830 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3831 performs a copy-by-value, obliterating any previous content of the
3834 You probably want to use one of the assortment of wrappers, such as
3835 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3836 C<SvSetMagicSV_nosteal>.
3838 =for apidoc sv_setsv_flags
3840 Copies the contents of the source SV C<ssv> into the destination SV
3841 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3842 function if the source SV needs to be reused. Does not handle 'set' magic.
3843 Loosely speaking, it performs a copy-by-value, obliterating any previous
3844 content of the destination.
3845 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3846 C<ssv> if appropriate, else not. If the C<flags>
3847 parameter has the C<SV_NOSTEAL> bit set then the
3848 buffers of temps will not be stolen. C<sv_setsv>
3849 and C<sv_setsv_nomg> are implemented in terms of this function.
3851 You probably want to use one of the assortment of wrappers, such as
3852 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3853 C<SvSetMagicSV_nosteal>.
3855 This is the primary function for copying scalars, and most other
3856 copy-ish functions and macros use this underneath.
3862 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3864 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3865 HV *old_stash = NULL;
3867 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3869 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3870 const char * const name = GvNAME(sstr);
3871 const STRLEN len = GvNAMELEN(sstr);
3873 if (dtype >= SVt_PV) {
3879 SvUPGRADE(dstr, SVt_PVGV);
3880 (void)SvOK_off(dstr);
3881 isGV_with_GP_on(dstr);
3883 GvSTASH(dstr) = GvSTASH(sstr);
3885 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3886 gv_name_set(MUTABLE_GV(dstr), name, len,
3887 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3888 SvFAKE_on(dstr); /* can coerce to non-glob */
3891 if(GvGP(MUTABLE_GV(sstr))) {
3892 /* If source has method cache entry, clear it */
3894 SvREFCNT_dec(GvCV(sstr));
3895 GvCV_set(sstr, NULL);
3898 /* If source has a real method, then a method is
3901 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3907 /* If dest already had a real method, that's a change as well */
3909 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3910 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3915 /* We don't need to check the name of the destination if it was not a
3916 glob to begin with. */
3917 if(dtype == SVt_PVGV) {
3918 const char * const name = GvNAME((const GV *)dstr);
3921 /* The stash may have been detached from the symbol table, so
3923 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3927 const STRLEN len = GvNAMELEN(dstr);
3928 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3929 || (len == 1 && name[0] == ':')) {
3932 /* Set aside the old stash, so we can reset isa caches on
3934 if((old_stash = GvHV(dstr)))
3935 /* Make sure we do not lose it early. */
3936 SvREFCNT_inc_simple_void_NN(
3937 sv_2mortal((SV *)old_stash)
3942 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3945 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3946 * so temporarily protect it */
3948 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3949 gp_free(MUTABLE_GV(dstr));
3950 GvINTRO_off(dstr); /* one-shot flag */
3951 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3954 if (SvTAINTED(sstr))
3956 if (GvIMPORTED(dstr) != GVf_IMPORTED
3957 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3959 GvIMPORTED_on(dstr);
3962 if(mro_changes == 2) {
3963 if (GvAV((const GV *)sstr)) {
3965 SV * const sref = (SV *)GvAV((const GV *)dstr);
3966 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3967 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3968 AV * const ary = newAV();
3969 av_push(ary, mg->mg_obj); /* takes the refcount */
3970 mg->mg_obj = (SV *)ary;
3972 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3974 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3976 mro_isa_changed_in(GvSTASH(dstr));
3978 else if(mro_changes == 3) {
3979 HV * const stash = GvHV(dstr);
3980 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3986 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3987 if (GvIO(dstr) && dtype == SVt_PVGV) {
3988 DEBUG_o(Perl_deb(aTHX_
3989 "glob_assign_glob clearing PL_stashcache\n"));
3990 /* It's a cache. It will rebuild itself quite happily.
3991 It's a lot of effort to work out exactly which key (or keys)
3992 might be invalidated by the creation of the this file handle.
3994 hv_clear(PL_stashcache);
4000 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4002 SV * const sref = SvRV(sstr);
4004 const int intro = GvINTRO(dstr);
4007 const U32 stype = SvTYPE(sref);
4009 PERL_ARGS_ASSERT_GV_SETREF;
4012 GvINTRO_off(dstr); /* one-shot flag */
4013 GvLINE(dstr) = CopLINE(PL_curcop);
4014 GvEGV(dstr) = MUTABLE_GV(dstr);
4019 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4020 import_flag = GVf_IMPORTED_CV;
4023 location = (SV **) &GvHV(dstr);
4024 import_flag = GVf_IMPORTED_HV;
4027 location = (SV **) &GvAV(dstr);
4028 import_flag = GVf_IMPORTED_AV;
4031 location = (SV **) &GvIOp(dstr);
4034 location = (SV **) &GvFORM(dstr);
4037 location = &GvSV(dstr);
4038 import_flag = GVf_IMPORTED_SV;
4041 if (stype == SVt_PVCV) {
4042 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4043 if (GvCVGEN(dstr)) {
4044 SvREFCNT_dec(GvCV(dstr));
4045 GvCV_set(dstr, NULL);
4046 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4049 /* SAVEt_GVSLOT takes more room on the savestack and has more
4050 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4051 leave_scope needs access to the GV so it can reset method
4052 caches. We must use SAVEt_GVSLOT whenever the type is
4053 SVt_PVCV, even if the stash is anonymous, as the stash may
4054 gain a name somehow before leave_scope. */
4055 if (stype == SVt_PVCV) {
4056 /* There is no save_pushptrptrptr. Creating it for this
4057 one call site would be overkill. So inline the ss add
4061 SS_ADD_PTR(location);
4062 SS_ADD_PTR(SvREFCNT_inc(*location));
4063 SS_ADD_UV(SAVEt_GVSLOT);
4066 else SAVEGENERICSV(*location);
4069 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4070 CV* const cv = MUTABLE_CV(*location);
4072 if (!GvCVGEN((const GV *)dstr) &&
4073 (CvROOT(cv) || CvXSUB(cv)) &&
4074 /* redundant check that avoids creating the extra SV
4075 most of the time: */
4076 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4078 SV * const new_const_sv =
4079 CvCONST((const CV *)sref)
4080 ? cv_const_sv((const CV *)sref)
4082 HV * const stash = GvSTASH((const GV *)dstr);
4083 report_redefined_cv(
4086 ? Perl_newSVpvf(aTHX_
4088 HEKfARG(HvNAME_HEK(stash)),
4089 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4090 : Perl_newSVpvf(aTHX_
4092 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4095 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4099 cv_ckproto_len_flags(cv, (const GV *)dstr,
4100 SvPOK(sref) ? CvPROTO(sref) : NULL,
4101 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4102 SvPOK(sref) ? SvUTF8(sref) : 0);
4104 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4105 GvASSUMECV_on(dstr);
4106 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4107 if (intro && GvREFCNT(dstr) > 1) {
4108 /* temporary remove extra savestack's ref */
4110 gv_method_changed(dstr);
4113 else gv_method_changed(dstr);
4116 *location = SvREFCNT_inc_simple_NN(sref);
4117 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4118 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4119 GvFLAGS(dstr) |= import_flag;
4122 if (stype == SVt_PVHV) {
4123 const char * const name = GvNAME((GV*)dstr);
4124 const STRLEN len = GvNAMELEN(dstr);
4127 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4128 || (len == 1 && name[0] == ':')
4130 && (!dref || HvENAME_get(dref))
4133 (HV *)sref, (HV *)dref,
4139 stype == SVt_PVAV && sref != dref
4140 && strEQ(GvNAME((GV*)dstr), "ISA")
4141 /* The stash may have been detached from the symbol table, so
4142 check its name before doing anything. */
4143 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4146 MAGIC * const omg = dref && SvSMAGICAL(dref)
4147 ? mg_find(dref, PERL_MAGIC_isa)
4149 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4150 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4151 AV * const ary = newAV();
4152 av_push(ary, mg->mg_obj); /* takes the refcount */
4153 mg->mg_obj = (SV *)ary;
4156 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4157 SV **svp = AvARRAY((AV *)omg->mg_obj);
4158 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4162 SvREFCNT_inc_simple_NN(*svp++)
4168 SvREFCNT_inc_simple_NN(omg->mg_obj)
4172 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4178 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4180 for (i = 0; i <= AvFILL(sref); ++i) {
4181 SV **elem = av_fetch ((AV*)sref, i, 0);
4184 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4188 mg = mg_find(sref, PERL_MAGIC_isa);
4190 /* Since the *ISA assignment could have affected more than
4191 one stash, don't call mro_isa_changed_in directly, but let
4192 magic_clearisa do it for us, as it already has the logic for
4193 dealing with globs vs arrays of globs. */
4195 Perl_magic_clearisa(aTHX_ NULL, mg);
4197 else if (stype == SVt_PVIO) {
4198 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4199 /* It's a cache. It will rebuild itself quite happily.
4200 It's a lot of effort to work out exactly which key (or keys)
4201 might be invalidated by the creation of the this file handle.
4203 hv_clear(PL_stashcache);
4207 if (!intro) SvREFCNT_dec(dref);
4208 if (SvTAINTED(sstr))
4216 #ifdef PERL_DEBUG_READONLY_COW
4217 # include <sys/mman.h>
4219 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4220 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4224 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4226 struct perl_memory_debug_header * const header =
4227 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4228 const MEM_SIZE len = header->size;
4229 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4230 # ifdef PERL_TRACK_MEMPOOL
4231 if (!header->readonly) header->readonly = 1;
4233 if (mprotect(header, len, PROT_READ))
4234 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4235 header, len, errno);
4239 S_sv_buf_to_rw(pTHX_ SV *sv)
4241 struct perl_memory_debug_header * const header =
4242 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4243 const MEM_SIZE len = header->size;
4244 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4245 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4246 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4247 header, len, errno);
4248 # ifdef PERL_TRACK_MEMPOOL
4249 header->readonly = 0;
4254 # define sv_buf_to_ro(sv) NOOP
4255 # define sv_buf_to_rw(sv) NOOP
4259 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4264 unsigned int both_type;
4266 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4268 if (UNLIKELY( sstr == dstr ))
4271 if (UNLIKELY( !sstr ))
4272 sstr = &PL_sv_undef;
4274 stype = SvTYPE(sstr);
4275 dtype = SvTYPE(dstr);
4276 both_type = (stype | dtype);
4278 /* with these values, we can check that both SVs are NULL/IV (and not
4279 * freed) just by testing the or'ed types */
4280 STATIC_ASSERT_STMT(SVt_NULL == 0);
4281 STATIC_ASSERT_STMT(SVt_IV == 1);
4282 if (both_type <= 1) {
4283 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4288 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4289 if (SvREADONLY(dstr))
4290 Perl_croak_no_modify();
4292 sv_unref_flags(dstr, 0);
4294 assert(!SvGMAGICAL(sstr));
4295 assert(!SvGMAGICAL(dstr));
4297 sflags = SvFLAGS(sstr);
4298 if (sflags & (SVf_IOK|SVf_ROK)) {
4299 SET_SVANY_FOR_BODYLESS_IV(dstr);
4300 new_dflags = SVt_IV;
4302 if (sflags & SVf_ROK) {
4303 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4304 new_dflags |= SVf_ROK;
4307 /* both src and dst are <= SVt_IV, so sv_any points to the
4308 * head; so access the head directly
4310 assert( &(sstr->sv_u.svu_iv)
4311 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4312 assert( &(dstr->sv_u.svu_iv)
4313 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4314 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4315 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4319 new_dflags = dtype; /* turn off everything except the type */
4321 SvFLAGS(dstr) = new_dflags;
4326 if (UNLIKELY(both_type == SVTYPEMASK)) {
4327 if (SvIS_FREED(dstr)) {
4328 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4329 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4331 if (SvIS_FREED(sstr)) {
4332 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4333 (void*)sstr, (void*)dstr);
4339 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4340 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4342 /* There's a lot of redundancy below but we're going for speed here */
4347 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4348 (void)SvOK_off(dstr);
4356 /* For performance, we inline promoting to type SVt_IV. */
4357 /* We're starting from SVt_NULL, so provided that define is
4358 * actual 0, we don't have to unset any SV type flags
4359 * to promote to SVt_IV. */
4360 STATIC_ASSERT_STMT(SVt_NULL == 0);
4361 SET_SVANY_FOR_BODYLESS_IV(dstr);
4362 SvFLAGS(dstr) |= SVt_IV;
4366 sv_upgrade(dstr, SVt_PVIV);
4370 goto end_of_first_switch;
4372 (void)SvIOK_only(dstr);
4373 SvIV_set(dstr, SvIVX(sstr));
4376 /* SvTAINTED can only be true if the SV has taint magic, which in
4377 turn means that the SV type is PVMG (or greater). This is the
4378 case statement for SVt_IV, so this cannot be true (whatever gcov
4380 assert(!SvTAINTED(sstr));
4385 if (dtype < SVt_PV && dtype != SVt_IV)
4386 sv_upgrade(dstr, SVt_IV);
4390 if (LIKELY( SvNOK(sstr) )) {
4394 sv_upgrade(dstr, SVt_NV);
4398 sv_upgrade(dstr, SVt_PVNV);
4402 goto end_of_first_switch;
4404 SvNV_set(dstr, SvNVX(sstr));
4405 (void)SvNOK_only(dstr);
4406 /* SvTAINTED can only be true if the SV has taint magic, which in
4407 turn means that the SV type is PVMG (or greater). This is the
4408 case statement for SVt_NV, so this cannot be true (whatever gcov
4410 assert(!SvTAINTED(sstr));
4417 sv_upgrade(dstr, SVt_PV);
4420 if (dtype < SVt_PVIV)
4421 sv_upgrade(dstr, SVt_PVIV);
4424 if (dtype < SVt_PVNV)
4425 sv_upgrade(dstr, SVt_PVNV);
4429 const char * const type = sv_reftype(sstr,0);
4431 /* diag_listed_as: Bizarre copy of %s */
4432 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4434 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4436 NOT_REACHED; /* NOTREACHED */
4440 if (dtype < SVt_REGEXP)
4442 if (dtype >= SVt_PV) {
4448 sv_upgrade(dstr, SVt_REGEXP);
4456 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4458 if (SvTYPE(sstr) != stype)
4459 stype = SvTYPE(sstr);
4461 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4462 glob_assign_glob(dstr, sstr, dtype);
4465 if (stype == SVt_PVLV)
4467 if (isREGEXP(sstr)) goto upgregexp;
4468 SvUPGRADE(dstr, SVt_PVNV);
4471 SvUPGRADE(dstr, (svtype)stype);
4473 end_of_first_switch:
4475 /* dstr may have been upgraded. */
4476 dtype = SvTYPE(dstr);
4477 sflags = SvFLAGS(sstr);
4479 if (UNLIKELY( dtype == SVt_PVCV )) {
4480 /* Assigning to a subroutine sets the prototype. */
4483 const char *const ptr = SvPV_const(sstr, len);
4485 SvGROW(dstr, len + 1);
4486 Copy(ptr, SvPVX(dstr), len + 1, char);
4487 SvCUR_set(dstr, len);
4489 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4490 CvAUTOLOAD_off(dstr);
4495 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4496 || dtype == SVt_PVFM))
4498 const char * const type = sv_reftype(dstr,0);
4500 /* diag_listed_as: Cannot copy to %s */
4501 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4503 Perl_croak(aTHX_ "Cannot copy to %s", type);
4504 } else if (sflags & SVf_ROK) {
4505 if (isGV_with_GP(dstr)
4506 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4509 if (GvIMPORTED(dstr) != GVf_IMPORTED
4510 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4512 GvIMPORTED_on(dstr);
4517 glob_assign_glob(dstr, sstr, dtype);
4521 if (dtype >= SVt_PV) {
4522 if (isGV_with_GP(dstr)) {
4523 gv_setref(dstr, sstr);
4526 if (SvPVX_const(dstr)) {
4532 (void)SvOK_off(dstr);
4533 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4534 SvFLAGS(dstr) |= sflags & SVf_ROK;
4535 assert(!(sflags & SVp_NOK));
4536 assert(!(sflags & SVp_IOK));
4537 assert(!(sflags & SVf_NOK));
4538 assert(!(sflags & SVf_IOK));
4540 else if (isGV_with_GP(dstr)) {
4541 if (!(sflags & SVf_OK)) {
4542 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4543 "Undefined value assigned to typeglob");
4546 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4547 if (dstr != (const SV *)gv) {
4548 const char * const name = GvNAME((const GV *)dstr);
4549 const STRLEN len = GvNAMELEN(dstr);
4550 HV *old_stash = NULL;
4551 bool reset_isa = FALSE;
4552 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4553 || (len == 1 && name[0] == ':')) {
4554 /* Set aside the old stash, so we can reset isa caches
4555 on its subclasses. */
4556 if((old_stash = GvHV(dstr))) {
4557 /* Make sure we do not lose it early. */
4558 SvREFCNT_inc_simple_void_NN(
4559 sv_2mortal((SV *)old_stash)
4566 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4567 gp_free(MUTABLE_GV(dstr));
4569 GvGP_set(dstr, gp_ref(GvGP(gv)));
4572 HV * const stash = GvHV(dstr);
4574 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4584 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4585 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4586 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4588 else if (sflags & SVp_POK) {
4589 const STRLEN cur = SvCUR(sstr);
4590 const STRLEN len = SvLEN(sstr);
4593 * We have three basic ways to copy the string:
4599 * Which we choose is based on various factors. The following
4600 * things are listed in order of speed, fastest to slowest:
4602 * - Copying a short string
4603 * - Copy-on-write bookkeeping
4605 * - Copying a long string
4607 * We swipe the string (steal the string buffer) if the SV on the
4608 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4609 * big win on long strings. It should be a win on short strings if
4610 * SvPVX_const(dstr) has to be allocated. If not, it should not
4611 * slow things down, as SvPVX_const(sstr) would have been freed
4614 * We also steal the buffer from a PADTMP (operator target) if it
4615 * is ‘long enough’. For short strings, a swipe does not help
4616 * here, as it causes more malloc calls the next time the target
4617 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4618 * be allocated it is still not worth swiping PADTMPs for short
4619 * strings, as the savings here are small.
4621 * If swiping is not an option, then we see whether it is
4622 * worth using copy-on-write. If the lhs already has a buf-
4623 * fer big enough and the string is short, we skip it and fall back
4624 * to method 3, since memcpy is faster for short strings than the
4625 * later bookkeeping overhead that copy-on-write entails.
4627 * If the rhs is not a copy-on-write string yet, then we also
4628 * consider whether the buffer is too large relative to the string
4629 * it holds. Some operations such as readline allocate a large
4630 * buffer in the expectation of reusing it. But turning such into
4631 * a COW buffer is counter-productive because it increases memory
4632 * usage by making readline allocate a new large buffer the sec-
4633 * ond time round. So, if the buffer is too large, again, we use
4636 * Finally, if there is no buffer on the left, or the buffer is too
4637 * small, then we use copy-on-write and make both SVs share the
4642 /* Whichever path we take through the next code, we want this true,
4643 and doing it now facilitates the COW check. */
4644 (void)SvPOK_only(dstr);
4648 /* slated for free anyway (and not COW)? */
4649 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4650 /* or a swipable TARG */
4652 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4654 /* whose buffer is worth stealing */
4655 && CHECK_COWBUF_THRESHOLD(cur,len)
4658 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4659 (!(flags & SV_NOSTEAL)) &&
4660 /* and we're allowed to steal temps */
4661 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4662 len) /* and really is a string */
4663 { /* Passes the swipe test. */
4664 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4666 SvPV_set(dstr, SvPVX_mutable(sstr));
4667 SvLEN_set(dstr, SvLEN(sstr));
4668 SvCUR_set(dstr, SvCUR(sstr));
4671 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4672 SvPV_set(sstr, NULL);
4677 else if (flags & SV_COW_SHARED_HASH_KEYS
4679 #ifdef PERL_COPY_ON_WRITE
4682 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4683 /* If this is a regular (non-hek) COW, only so
4684 many COW "copies" are possible. */
4685 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4686 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4687 && !(SvFLAGS(dstr) & SVf_BREAK)
4688 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4689 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4693 && !(SvFLAGS(dstr) & SVf_BREAK)
4696 /* Either it's a shared hash key, or it's suitable for
4699 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4704 if (!(sflags & SVf_IsCOW)) {
4706 CowREFCNT(sstr) = 0;
4709 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4715 if (sflags & SVf_IsCOW) {
4719 SvPV_set(dstr, SvPVX_mutable(sstr));
4724 /* SvIsCOW_shared_hash */
4725 DEBUG_C(PerlIO_printf(Perl_debug_log,
4726 "Copy on write: Sharing hash\n"));
4728 assert (SvTYPE(dstr) >= SVt_PV);
4730 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4732 SvLEN_set(dstr, len);
4733 SvCUR_set(dstr, cur);
4736 /* Failed the swipe test, and we cannot do copy-on-write either.
4737 Have to copy the string. */
4738 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4739 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4740 SvCUR_set(dstr, cur);
4741 *SvEND(dstr) = '\0';
4743 if (sflags & SVp_NOK) {
4744 SvNV_set(dstr, SvNVX(sstr));
4746 if (sflags & SVp_IOK) {
4747 SvIV_set(dstr, SvIVX(sstr));
4748 /* Must do this otherwise some other overloaded use of 0x80000000
4749 gets confused. I guess SVpbm_VALID */
4750 if (sflags & SVf_IVisUV)
4753 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4755 const MAGIC * const smg = SvVSTRING_mg(sstr);
4757 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4758 smg->mg_ptr, smg->mg_len);
4759 SvRMAGICAL_on(dstr);
4763 else if (sflags & (SVp_IOK|SVp_NOK)) {
4764 (void)SvOK_off(dstr);
4765 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4766 if (sflags & SVp_IOK) {
4767 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4768 SvIV_set(dstr, SvIVX(sstr));
4770 if (sflags & SVp_NOK) {
4771 SvNV_set(dstr, SvNVX(sstr));
4775 if (isGV_with_GP(sstr)) {
4776 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4779 (void)SvOK_off(dstr);
4781 if (SvTAINTED(sstr))
4786 =for apidoc sv_setsv_mg
4788 Like C<sv_setsv>, but also handles 'set' magic.
4794 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4796 PERL_ARGS_ASSERT_SV_SETSV_MG;
4798 sv_setsv(dstr,sstr);
4803 # define SVt_COW SVt_PV
4805 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4807 STRLEN cur = SvCUR(sstr);
4808 STRLEN len = SvLEN(sstr);
4810 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4811 const bool already = cBOOL(SvIsCOW(sstr));
4814 PERL_ARGS_ASSERT_SV_SETSV_COW;
4817 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4818 (void*)sstr, (void*)dstr);
4825 if (SvTHINKFIRST(dstr))
4826 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4827 else if (SvPVX_const(dstr))
4828 Safefree(SvPVX_mutable(dstr));
4832 SvUPGRADE(dstr, SVt_COW);
4834 assert (SvPOK(sstr));
4835 assert (SvPOKp(sstr));
4837 if (SvIsCOW(sstr)) {
4839 if (SvLEN(sstr) == 0) {
4840 /* source is a COW shared hash key. */
4841 DEBUG_C(PerlIO_printf(Perl_debug_log,
4842 "Fast copy on write: Sharing hash\n"));
4843 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4846 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4847 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4849 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4850 SvUPGRADE(sstr, SVt_COW);
4852 DEBUG_C(PerlIO_printf(Perl_debug_log,
4853 "Fast copy on write: Converting sstr to COW\n"));
4854 CowREFCNT(sstr) = 0;
4856 # ifdef PERL_DEBUG_READONLY_COW
4857 if (already) sv_buf_to_rw(sstr);
4860 new_pv = SvPVX_mutable(sstr);
4864 SvPV_set(dstr, new_pv);
4865 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4868 SvLEN_set(dstr, len);
4869 SvCUR_set(dstr, cur);
4878 =for apidoc sv_setpv_bufsize
4880 Sets the SV to be a string of cur bytes length, with at least
4881 len bytes available. Ensures that there is a null byte at SvEND.
4882 Returns a char * pointer to the SvPV buffer.
4888 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4892 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4894 SV_CHECK_THINKFIRST_COW_DROP(sv);
4895 SvUPGRADE(sv, SVt_PV);
4896 pv = SvGROW(sv, len + 1);
4899 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4902 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4907 =for apidoc sv_setpvn
4909 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4910 The C<len> parameter indicates the number of
4911 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4912 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4918 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4922 PERL_ARGS_ASSERT_SV_SETPVN;
4924 SV_CHECK_THINKFIRST_COW_DROP(sv);
4930 /* len is STRLEN which is unsigned, need to copy to signed */
4933 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4936 SvUPGRADE(sv, SVt_PV);
4938 dptr = SvGROW(sv, len + 1);
4939 Move(ptr,dptr,len,char);
4942 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4944 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4948 =for apidoc sv_setpvn_mg
4950 Like C<sv_setpvn>, but also handles 'set' magic.
4956 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4958 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4960 sv_setpvn(sv,ptr,len);
4965 =for apidoc sv_setpv
4967 Copies a string into an SV. The string must be terminated with a C<NUL>
4968 character, and not contain embeded C<NUL>'s.
4969 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4975 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4979 PERL_ARGS_ASSERT_SV_SETPV;
4981 SV_CHECK_THINKFIRST_COW_DROP(sv);
4987 SvUPGRADE(sv, SVt_PV);
4989 SvGROW(sv, len + 1);
4990 Move(ptr,SvPVX(sv),len+1,char);
4992 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4994 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4998 =for apidoc sv_setpv_mg
5000 Like C<sv_setpv>, but also handles 'set' magic.
5006 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5008 PERL_ARGS_ASSERT_SV_SETPV_MG;
5015 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5017 PERL_ARGS_ASSERT_SV_SETHEK;
5023 if (HEK_LEN(hek) == HEf_SVKEY) {
5024 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5027 const int flags = HEK_FLAGS(hek);
5028 if (flags & HVhek_WASUTF8) {
5029 STRLEN utf8_len = HEK_LEN(hek);
5030 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5031 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5034 } else if (flags & HVhek_UNSHARED) {
5035 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5038 else SvUTF8_off(sv);
5042 SV_CHECK_THINKFIRST_COW_DROP(sv);
5043 SvUPGRADE(sv, SVt_PV);
5045 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5046 SvCUR_set(sv, HEK_LEN(hek));
5052 else SvUTF8_off(sv);
5060 =for apidoc sv_usepvn_flags
5062 Tells an SV to use C<ptr> to find its string value. Normally the
5063 string is stored inside the SV, but sv_usepvn allows the SV to use an
5064 outside string. C<ptr> should point to memory that was allocated
5065 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5066 the start of a C<Newx>-ed block of memory, and not a pointer to the
5067 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5068 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5069 string length, C<len>, must be supplied. By default this function
5070 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5071 so that pointer should not be freed or used by the programmer after
5072 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5073 that pointer (e.g. ptr + 1) be used.
5075 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5076 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5078 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5079 C<len>, and already meets the requirements for storing in C<SvPVX>).
5085 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5089 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5091 SV_CHECK_THINKFIRST_COW_DROP(sv);
5092 SvUPGRADE(sv, SVt_PV);
5095 if (flags & SV_SMAGIC)
5099 if (SvPVX_const(sv))
5103 if (flags & SV_HAS_TRAILING_NUL)
5104 assert(ptr[len] == '\0');
5107 allocate = (flags & SV_HAS_TRAILING_NUL)
5109 #ifdef Perl_safesysmalloc_size
5112 PERL_STRLEN_ROUNDUP(len + 1);
5114 if (flags & SV_HAS_TRAILING_NUL) {
5115 /* It's long enough - do nothing.
5116 Specifically Perl_newCONSTSUB is relying on this. */
5119 /* Force a move to shake out bugs in callers. */
5120 char *new_ptr = (char*)safemalloc(allocate);
5121 Copy(ptr, new_ptr, len, char);
5122 PoisonFree(ptr,len,char);
5126 ptr = (char*) saferealloc (ptr, allocate);
5129 #ifdef Perl_safesysmalloc_size
5130 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5132 SvLEN_set(sv, allocate);
5136 if (!(flags & SV_HAS_TRAILING_NUL)) {
5139 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5141 if (flags & SV_SMAGIC)
5146 =for apidoc sv_force_normal_flags
5148 Undo various types of fakery on an SV, where fakery means
5149 "more than" a string: if the PV is a shared string, make
5150 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5151 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5152 we do the copy, and is also used locally; if this is a
5153 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5154 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5155 C<SvPOK_off> rather than making a copy. (Used where this
5156 scalar is about to be set to some other value.) In addition,
5157 the C<flags> parameter gets passed to C<sv_unref_flags()>
5158 when unreffing. C<sv_force_normal> calls this function
5159 with flags set to 0.
5161 This function is expected to be used to signal to perl that this SV is
5162 about to be written to, and any extra book-keeping needs to be taken care
5163 of. Hence, it croaks on read-only values.
5169 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5171 assert(SvIsCOW(sv));
5174 const char * const pvx = SvPVX_const(sv);
5175 const STRLEN len = SvLEN(sv);
5176 const STRLEN cur = SvCUR(sv);
5179 PerlIO_printf(Perl_debug_log,
5180 "Copy on write: Force normal %ld\n",
5185 # ifdef PERL_COPY_ON_WRITE
5187 /* Must do this first, since the CowREFCNT uses SvPVX and
5188 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5189 the only owner left of the buffer. */
5190 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5192 U8 cowrefcnt = CowREFCNT(sv);
5193 if(cowrefcnt != 0) {
5195 CowREFCNT(sv) = cowrefcnt;
5200 /* Else we are the only owner of the buffer. */
5205 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5210 if (flags & SV_COW_DROP_PV) {
5211 /* OK, so we don't need to copy our buffer. */
5214 SvGROW(sv, cur + 1);
5215 Move(pvx,SvPVX(sv),cur,char);
5221 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5228 const char * const pvx = SvPVX_const(sv);
5229 const STRLEN len = SvCUR(sv);
5233 if (flags & SV_COW_DROP_PV) {
5234 /* OK, so we don't need to copy our buffer. */
5237 SvGROW(sv, len + 1);
5238 Move(pvx,SvPVX(sv),len,char);
5241 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5247 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5249 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5252 Perl_croak_no_modify();
5253 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5254 S_sv_uncow(aTHX_ sv, flags);
5256 sv_unref_flags(sv, flags);
5257 else if (SvFAKE(sv) && isGV_with_GP(sv))
5258 sv_unglob(sv, flags);
5259 else if (SvFAKE(sv) && isREGEXP(sv)) {
5260 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5261 to sv_unglob. We only need it here, so inline it. */
5262 const bool islv = SvTYPE(sv) == SVt_PVLV;
5263 const svtype new_type =
5264 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5265 SV *const temp = newSV_type(new_type);
5266 regexp *const temp_p = ReANY((REGEXP *)sv);
5268 if (new_type == SVt_PVMG) {
5269 SvMAGIC_set(temp, SvMAGIC(sv));
5270 SvMAGIC_set(sv, NULL);
5271 SvSTASH_set(temp, SvSTASH(sv));
5272 SvSTASH_set(sv, NULL);
5274 if (!islv) SvCUR_set(temp, SvCUR(sv));
5275 /* Remember that SvPVX is in the head, not the body. But
5276 RX_WRAPPED is in the body. */
5277 assert(ReANY((REGEXP *)sv)->mother_re);
5278 /* Their buffer is already owned by someone else. */
5279 if (flags & SV_COW_DROP_PV) {
5280 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5281 zeroed body. For SVt_PVLV, it should have been set to 0
5282 before turning into a regexp. */
5283 assert(!SvLEN(islv ? sv : temp));
5284 sv->sv_u.svu_pv = 0;
5287 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5288 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5292 /* Now swap the rest of the bodies. */
5296 SvFLAGS(sv) &= ~SVTYPEMASK;
5297 SvFLAGS(sv) |= new_type;
5298 SvANY(sv) = SvANY(temp);
5301 SvFLAGS(temp) &= ~(SVTYPEMASK);
5302 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5303 SvANY(temp) = temp_p;
5304 temp->sv_u.svu_rx = (regexp *)temp_p;
5306 SvREFCNT_dec_NN(temp);
5308 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5314 Efficient removal of characters from the beginning of the string buffer.
5315 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5316 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5317 character of the adjusted string. Uses the C<OOK> hack. On return, only
5318 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5320 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5321 refer to the same chunk of data.
5323 The unfortunate similarity of this function's name to that of Perl's C<chop>
5324 operator is strictly coincidental. This function works from the left;
5325 C<chop> works from the right.
5331 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5342 PERL_ARGS_ASSERT_SV_CHOP;
5344 if (!ptr || !SvPOKp(sv))
5346 delta = ptr - SvPVX_const(sv);
5348 /* Nothing to do. */
5351 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5352 if (delta > max_delta)
5353 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5354 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5355 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5356 SV_CHECK_THINKFIRST(sv);
5357 SvPOK_only_UTF8(sv);
5360 if (!SvLEN(sv)) { /* make copy of shared string */
5361 const char *pvx = SvPVX_const(sv);
5362 const STRLEN len = SvCUR(sv);
5363 SvGROW(sv, len + 1);
5364 Move(pvx,SvPVX(sv),len,char);
5370 SvOOK_offset(sv, old_delta);
5372 SvLEN_set(sv, SvLEN(sv) - delta);
5373 SvCUR_set(sv, SvCUR(sv) - delta);
5374 SvPV_set(sv, SvPVX(sv) + delta);
5376 p = (U8 *)SvPVX_const(sv);
5379 /* how many bytes were evacuated? we will fill them with sentinel
5380 bytes, except for the part holding the new offset of course. */
5383 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5385 assert(evacn <= delta + old_delta);
5389 /* This sets 'delta' to the accumulated value of all deltas so far */
5393 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5394 * the string; otherwise store a 0 byte there and store 'delta' just prior
5395 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5396 * portion of the chopped part of the string */
5397 if (delta < 0x100) {
5401 p -= sizeof(STRLEN);
5402 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5406 /* Fill the preceding buffer with sentinals to verify that no-one is
5416 =for apidoc sv_catpvn
5418 Concatenates the string onto the end of the string which is in the SV.
5419 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5420 status set, then the bytes appended should be valid UTF-8.
5421 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5423 =for apidoc sv_catpvn_flags
5425 Concatenates the string onto the end of the string which is in the SV. The
5426 C<len> indicates number of bytes to copy.
5428 By default, the string appended is assumed to be valid UTF-8 if the SV has
5429 the UTF-8 status set, and a string of bytes otherwise. One can force the
5430 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5431 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5432 string appended will be upgraded to UTF-8 if necessary.
5434 If C<flags> has the C<SV_SMAGIC> bit set, will
5435 C<mg_set> on C<dsv> afterwards if appropriate.
5436 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5437 in terms of this function.
5443 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5446 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5448 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5449 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5451 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5452 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5453 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5456 else SvGROW(dsv, dlen + slen + 1);
5458 sstr = SvPVX_const(dsv);
5459 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5460 SvCUR_set(dsv, SvCUR(dsv) + slen);
5463 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5464 const char * const send = sstr + slen;
5467 /* Something this code does not account for, which I think is
5468 impossible; it would require the same pv to be treated as
5469 bytes *and* utf8, which would indicate a bug elsewhere. */
5470 assert(sstr != dstr);
5472 SvGROW(dsv, dlen + slen * 2 + 1);
5473 d = (U8 *)SvPVX(dsv) + dlen;
5475 while (sstr < send) {
5476 append_utf8_from_native_byte(*sstr, &d);
5479 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5482 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5484 if (flags & SV_SMAGIC)
5489 =for apidoc sv_catsv
5491 Concatenates the string from SV C<ssv> onto the end of the string in SV
5492 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5493 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5494 and C<L</sv_catsv_nomg>>.
5496 =for apidoc sv_catsv_flags
5498 Concatenates the string from SV C<ssv> onto the end of the string in SV
5499 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5500 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5501 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5502 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5503 and C<sv_catsv_mg> are implemented in terms of this function.
5508 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5510 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5514 const char *spv = SvPV_flags_const(ssv, slen, flags);
5515 if (flags & SV_GMAGIC)
5517 sv_catpvn_flags(dsv, spv, slen,
5518 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5519 if (flags & SV_SMAGIC)
5525 =for apidoc sv_catpv
5527 Concatenates the C<NUL>-terminated string onto the end of the string which is
5529 If the SV has the UTF-8 status set, then the bytes appended should be
5530 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5536 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5542 PERL_ARGS_ASSERT_SV_CATPV;
5546 junk = SvPV_force(sv, tlen);
5548 SvGROW(sv, tlen + len + 1);
5550 ptr = SvPVX_const(sv);
5551 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5552 SvCUR_set(sv, SvCUR(sv) + len);
5553 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5558 =for apidoc sv_catpv_flags
5560 Concatenates the C<NUL>-terminated string onto the end of the string which is
5562 If the SV has the UTF-8 status set, then the bytes appended should
5563 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5564 on the modified SV if appropriate.
5570 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5572 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5573 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5577 =for apidoc sv_catpv_mg
5579 Like C<sv_catpv>, but also handles 'set' magic.
5585 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5587 PERL_ARGS_ASSERT_SV_CATPV_MG;
5596 Creates a new SV. A non-zero C<len> parameter indicates the number of
5597 bytes of preallocated string space the SV should have. An extra byte for a
5598 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5599 space is allocated.) The reference count for the new SV is set to 1.
5601 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5602 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5603 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5604 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5605 modules supporting older perls.
5611 Perl_newSV(pTHX_ const STRLEN len)
5617 sv_grow(sv, len + 1);
5622 =for apidoc sv_magicext
5624 Adds magic to an SV, upgrading it if necessary. Applies the
5625 supplied C<vtable> and returns a pointer to the magic added.
5627 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5628 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5629 one instance of the same C<how>.
5631 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5632 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5633 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5634 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5636 (This is now used as a subroutine by C<sv_magic>.)
5641 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5642 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5646 PERL_ARGS_ASSERT_SV_MAGICEXT;
5648 SvUPGRADE(sv, SVt_PVMG);
5649 Newxz(mg, 1, MAGIC);
5650 mg->mg_moremagic = SvMAGIC(sv);
5651 SvMAGIC_set(sv, mg);
5653 /* Sometimes a magic contains a reference loop, where the sv and
5654 object refer to each other. To prevent a reference loop that
5655 would prevent such objects being freed, we look for such loops
5656 and if we find one we avoid incrementing the object refcount.
5658 Note we cannot do this to avoid self-tie loops as intervening RV must
5659 have its REFCNT incremented to keep it in existence.
5662 if (!obj || obj == sv ||
5663 how == PERL_MAGIC_arylen ||
5664 how == PERL_MAGIC_symtab ||
5665 (SvTYPE(obj) == SVt_PVGV &&
5666 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5667 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5668 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5673 mg->mg_obj = SvREFCNT_inc_simple(obj);
5674 mg->mg_flags |= MGf_REFCOUNTED;
5677 /* Normal self-ties simply pass a null object, and instead of
5678 using mg_obj directly, use the SvTIED_obj macro to produce a
5679 new RV as needed. For glob "self-ties", we are tieing the PVIO
5680 with an RV obj pointing to the glob containing the PVIO. In
5681 this case, to avoid a reference loop, we need to weaken the
5685 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5686 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5692 mg->mg_len = namlen;
5695 mg->mg_ptr = savepvn(name, namlen);
5696 else if (namlen == HEf_SVKEY) {
5697 /* Yes, this is casting away const. This is only for the case of
5698 HEf_SVKEY. I think we need to document this aberation of the
5699 constness of the API, rather than making name non-const, as
5700 that change propagating outwards a long way. */
5701 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5703 mg->mg_ptr = (char *) name;
5705 mg->mg_virtual = (MGVTBL *) vtable;
5712 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5714 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5715 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5716 /* This sv is only a delegate. //g magic must be attached to
5721 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5722 &PL_vtbl_mglob, 0, 0);
5726 =for apidoc sv_magic
5728 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5729 necessary, then adds a new magic item of type C<how> to the head of the
5732 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5733 handling of the C<name> and C<namlen> arguments.
5735 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5736 to add more than one instance of the same C<how>.
5742 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5743 const char *const name, const I32 namlen)
5745 const MGVTBL *vtable;
5748 unsigned int vtable_index;
5750 PERL_ARGS_ASSERT_SV_MAGIC;
5752 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5753 || ((flags = PL_magic_data[how]),
5754 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5755 > magic_vtable_max))
5756 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5758 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5759 Useful for attaching extension internal data to perl vars.
5760 Note that multiple extensions may clash if magical scalars
5761 etc holding private data from one are passed to another. */
5763 vtable = (vtable_index == magic_vtable_max)
5764 ? NULL : PL_magic_vtables + vtable_index;
5766 if (SvREADONLY(sv)) {
5768 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5771 Perl_croak_no_modify();
5774 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5775 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5776 /* sv_magic() refuses to add a magic of the same 'how' as an
5779 if (how == PERL_MAGIC_taint)
5785 /* Force pos to be stored as characters, not bytes. */
5786 if (SvMAGICAL(sv) && DO_UTF8(sv)
5787 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5789 && mg->mg_flags & MGf_BYTES) {
5790 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5792 mg->mg_flags &= ~MGf_BYTES;
5795 /* Rest of work is done else where */
5796 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5799 case PERL_MAGIC_taint:
5802 case PERL_MAGIC_ext:
5803 case PERL_MAGIC_dbfile:
5810 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5817 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5819 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5820 for (mg = *mgp; mg; mg = *mgp) {
5821 const MGVTBL* const virt = mg->mg_virtual;
5822 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5823 *mgp = mg->mg_moremagic;
5824 if (virt && virt->svt_free)
5825 virt->svt_free(aTHX_ sv, mg);
5826 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5828 Safefree(mg->mg_ptr);
5829 else if (mg->mg_len == HEf_SVKEY)
5830 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5831 else if (mg->mg_type == PERL_MAGIC_utf8)
5832 Safefree(mg->mg_ptr);
5834 if (mg->mg_flags & MGf_REFCOUNTED)
5835 SvREFCNT_dec(mg->mg_obj);
5839 mgp = &mg->mg_moremagic;
5842 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5843 mg_magical(sv); /* else fix the flags now */
5852 =for apidoc sv_unmagic
5854 Removes all magic of type C<type> from an SV.
5860 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5862 PERL_ARGS_ASSERT_SV_UNMAGIC;
5863 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5867 =for apidoc sv_unmagicext
5869 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5875 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5877 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5878 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5882 =for apidoc sv_rvweaken
5884 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5885 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5886 push a back-reference to this RV onto the array of backreferences
5887 associated with that magic. If the RV is magical, set magic will be
5888 called after the RV is cleared.
5894 Perl_sv_rvweaken(pTHX_ SV *const sv)
5898 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5900 if (!SvOK(sv)) /* let undefs pass */
5903 Perl_croak(aTHX_ "Can't weaken a nonreference");
5904 else if (SvWEAKREF(sv)) {
5905 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5908 else if (SvREADONLY(sv)) croak_no_modify();
5910 Perl_sv_add_backref(aTHX_ tsv, sv);
5912 SvREFCNT_dec_NN(tsv);
5917 =for apidoc sv_get_backrefs
5919 If C<sv> is the target of a weak reference then it returns the back
5920 references structure associated with the sv; otherwise return C<NULL>.
5922 When returning a non-null result the type of the return is relevant. If it
5923 is an AV then the elements of the AV are the weak reference RVs which
5924 point at this item. If it is any other type then the item itself is the
5927 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5928 C<Perl_sv_kill_backrefs()>
5934 Perl_sv_get_backrefs(SV *const sv)
5938 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5940 /* find slot to store array or singleton backref */
5942 if (SvTYPE(sv) == SVt_PVHV) {
5944 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5945 backrefs = (SV *)iter->xhv_backreferences;
5947 } else if (SvMAGICAL(sv)) {
5948 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5950 backrefs = mg->mg_obj;
5955 /* Give tsv backref magic if it hasn't already got it, then push a
5956 * back-reference to sv onto the array associated with the backref magic.
5958 * As an optimisation, if there's only one backref and it's not an AV,
5959 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5960 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5964 /* A discussion about the backreferences array and its refcount:
5966 * The AV holding the backreferences is pointed to either as the mg_obj of
5967 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5968 * xhv_backreferences field. The array is created with a refcount
5969 * of 2. This means that if during global destruction the array gets
5970 * picked on before its parent to have its refcount decremented by the
5971 * random zapper, it won't actually be freed, meaning it's still there for
5972 * when its parent gets freed.
5974 * When the parent SV is freed, the extra ref is killed by
5975 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5976 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5978 * When a single backref SV is stored directly, it is not reference
5983 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5989 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5991 /* find slot to store array or singleton backref */
5993 if (SvTYPE(tsv) == SVt_PVHV) {
5994 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5997 mg = mg_find(tsv, PERL_MAGIC_backref);
5999 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6000 svp = &(mg->mg_obj);
6003 /* create or retrieve the array */
6005 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6006 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6010 mg->mg_flags |= MGf_REFCOUNTED;
6013 SvREFCNT_inc_simple_void_NN(av);
6014 /* av now has a refcnt of 2; see discussion above */
6015 av_extend(av, *svp ? 2 : 1);
6017 /* move single existing backref to the array */
6018 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6023 av = MUTABLE_AV(*svp);
6025 /* optimisation: store single backref directly in HvAUX or mg_obj */
6029 assert(SvTYPE(av) == SVt_PVAV);
6030 if (AvFILLp(av) >= AvMAX(av)) {
6031 av_extend(av, AvFILLp(av)+1);
6034 /* push new backref */
6035 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6038 /* delete a back-reference to ourselves from the backref magic associated
6039 * with the SV we point to.
6043 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6047 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6049 if (SvTYPE(tsv) == SVt_PVHV) {
6051 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6053 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6054 /* It's possible for the the last (strong) reference to tsv to have
6055 become freed *before* the last thing holding a weak reference.
6056 If both survive longer than the backreferences array, then when
6057 the referent's reference count drops to 0 and it is freed, it's
6058 not able to chase the backreferences, so they aren't NULLed.
6060 For example, a CV holds a weak reference to its stash. If both the
6061 CV and the stash survive longer than the backreferences array,
6062 and the CV gets picked for the SvBREAK() treatment first,
6063 *and* it turns out that the stash is only being kept alive because
6064 of an our variable in the pad of the CV, then midway during CV
6065 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6066 It ends up pointing to the freed HV. Hence it's chased in here, and
6067 if this block wasn't here, it would hit the !svp panic just below.
6069 I don't believe that "better" destruction ordering is going to help
6070 here - during global destruction there's always going to be the
6071 chance that something goes out of order. We've tried to make it
6072 foolproof before, and it only resulted in evolutionary pressure on
6073 fools. Which made us look foolish for our hubris. :-(
6079 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6080 svp = mg ? &(mg->mg_obj) : NULL;
6084 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6086 /* It's possible that sv is being freed recursively part way through the
6087 freeing of tsv. If this happens, the backreferences array of tsv has
6088 already been freed, and so svp will be NULL. If this is the case,
6089 we should not panic. Instead, nothing needs doing, so return. */
6090 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6092 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6093 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6096 if (SvTYPE(*svp) == SVt_PVAV) {
6100 AV * const av = (AV*)*svp;
6102 assert(!SvIS_FREED(av));
6106 /* for an SV with N weak references to it, if all those
6107 * weak refs are deleted, then sv_del_backref will be called
6108 * N times and O(N^2) compares will be done within the backref
6109 * array. To ameliorate this potential slowness, we:
6110 * 1) make sure this code is as tight as possible;
6111 * 2) when looking for SV, look for it at both the head and tail of the
6112 * array first before searching the rest, since some create/destroy
6113 * patterns will cause the backrefs to be freed in order.
6120 SV **p = &svp[fill];
6121 SV *const topsv = *p;
6128 /* We weren't the last entry.
6129 An unordered list has this property that you
6130 can take the last element off the end to fill
6131 the hole, and it's still an unordered list :-)
6137 break; /* should only be one */
6144 AvFILLp(av) = fill-1;
6146 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6147 /* freed AV; skip */
6150 /* optimisation: only a single backref, stored directly */
6152 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6153 (void*)*svp, (void*)sv);
6160 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6166 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6171 /* after multiple passes through Perl_sv_clean_all() for a thingy
6172 * that has badly leaked, the backref array may have gotten freed,
6173 * since we only protect it against 1 round of cleanup */
6174 if (SvIS_FREED(av)) {
6175 if (PL_in_clean_all) /* All is fair */
6178 "panic: magic_killbackrefs (freed backref AV/SV)");
6182 is_array = (SvTYPE(av) == SVt_PVAV);
6184 assert(!SvIS_FREED(av));
6187 last = svp + AvFILLp(av);
6190 /* optimisation: only a single backref, stored directly */
6196 while (svp <= last) {
6198 SV *const referrer = *svp;
6199 if (SvWEAKREF(referrer)) {
6200 /* XXX Should we check that it hasn't changed? */
6201 assert(SvROK(referrer));
6202 SvRV_set(referrer, 0);
6204 SvWEAKREF_off(referrer);
6205 SvSETMAGIC(referrer);
6206 } else if (SvTYPE(referrer) == SVt_PVGV ||
6207 SvTYPE(referrer) == SVt_PVLV) {
6208 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6209 /* You lookin' at me? */
6210 assert(GvSTASH(referrer));
6211 assert(GvSTASH(referrer) == (const HV *)sv);
6212 GvSTASH(referrer) = 0;
6213 } else if (SvTYPE(referrer) == SVt_PVCV ||
6214 SvTYPE(referrer) == SVt_PVFM) {
6215 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6216 /* You lookin' at me? */
6217 assert(CvSTASH(referrer));
6218 assert(CvSTASH(referrer) == (const HV *)sv);
6219 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6222 assert(SvTYPE(sv) == SVt_PVGV);
6223 /* You lookin' at me? */
6224 assert(CvGV(referrer));
6225 assert(CvGV(referrer) == (const GV *)sv);
6226 anonymise_cv_maybe(MUTABLE_GV(sv),
6227 MUTABLE_CV(referrer));
6232 "panic: magic_killbackrefs (flags=%"UVxf")",
6233 (UV)SvFLAGS(referrer));
6244 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6250 =for apidoc sv_insert
6252 Inserts a string at the specified offset/length within the SV. Similar to
6253 the Perl C<substr()> function. Handles get magic.
6255 =for apidoc sv_insert_flags
6257 Same as C<sv_insert>, but the extra C<flags> are passed to the
6258 C<SvPV_force_flags> that applies to C<bigstr>.
6264 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6270 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6273 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6275 SvPV_force_flags(bigstr, curlen, flags);
6276 (void)SvPOK_only_UTF8(bigstr);
6277 if (offset + len > curlen) {
6278 SvGROW(bigstr, offset+len+1);
6279 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6280 SvCUR_set(bigstr, offset+len);
6284 i = littlelen - len;
6285 if (i > 0) { /* string might grow */
6286 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6287 mid = big + offset + len;
6288 midend = bigend = big + SvCUR(bigstr);
6291 while (midend > mid) /* shove everything down */
6292 *--bigend = *--midend;
6293 Move(little,big+offset,littlelen,char);
6294 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6299 Move(little,SvPVX(bigstr)+offset,len,char);
6304 big = SvPVX(bigstr);
6307 bigend = big + SvCUR(bigstr);
6309 if (midend > bigend)
6310 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6313 if (mid - big > bigend - midend) { /* faster to shorten from end */
6315 Move(little, mid, littlelen,char);
6318 i = bigend - midend;
6320 Move(midend, mid, i,char);
6324 SvCUR_set(bigstr, mid - big);
6326 else if ((i = mid - big)) { /* faster from front */
6327 midend -= littlelen;
6329 Move(big, midend - i, i, char);
6330 sv_chop(bigstr,midend-i);
6332 Move(little, mid, littlelen,char);
6334 else if (littlelen) {
6335 midend -= littlelen;
6336 sv_chop(bigstr,midend);
6337 Move(little,midend,littlelen,char);
6340 sv_chop(bigstr,midend);
6346 =for apidoc sv_replace
6348 Make the first argument a copy of the second, then delete the original.
6349 The target SV physically takes over ownership of the body of the source SV
6350 and inherits its flags; however, the target keeps any magic it owns,
6351 and any magic in the source is discarded.
6352 Note that this is a rather specialist SV copying operation; most of the
6353 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6359 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6361 const U32 refcnt = SvREFCNT(sv);
6363 PERL_ARGS_ASSERT_SV_REPLACE;
6365 SV_CHECK_THINKFIRST_COW_DROP(sv);
6366 if (SvREFCNT(nsv) != 1) {
6367 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6368 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6370 if (SvMAGICAL(sv)) {
6374 sv_upgrade(nsv, SVt_PVMG);
6375 SvMAGIC_set(nsv, SvMAGIC(sv));
6376 SvFLAGS(nsv) |= SvMAGICAL(sv);
6378 SvMAGIC_set(sv, NULL);
6382 assert(!SvREFCNT(sv));
6383 #ifdef DEBUG_LEAKING_SCALARS
6384 sv->sv_flags = nsv->sv_flags;
6385 sv->sv_any = nsv->sv_any;
6386 sv->sv_refcnt = nsv->sv_refcnt;
6387 sv->sv_u = nsv->sv_u;
6389 StructCopy(nsv,sv,SV);
6391 if(SvTYPE(sv) == SVt_IV) {
6392 SET_SVANY_FOR_BODYLESS_IV(sv);
6396 SvREFCNT(sv) = refcnt;
6397 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6402 /* We're about to free a GV which has a CV that refers back to us.
6403 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6407 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6412 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6415 assert(SvREFCNT(gv) == 0);
6416 assert(isGV(gv) && isGV_with_GP(gv));
6418 assert(!CvANON(cv));
6419 assert(CvGV(cv) == gv);
6420 assert(!CvNAMED(cv));
6422 /* will the CV shortly be freed by gp_free() ? */
6423 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6424 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6428 /* if not, anonymise: */
6429 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6430 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6431 : newSVpvn_flags( "__ANON__", 8, 0 );
6432 sv_catpvs(gvname, "::__ANON__");
6433 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6434 SvREFCNT_dec_NN(gvname);
6438 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6443 =for apidoc sv_clear
6445 Clear an SV: call any destructors, free up any memory used by the body,
6446 and free the body itself. The SV's head is I<not> freed, although
6447 its type is set to all 1's so that it won't inadvertently be assumed
6448 to be live during global destruction etc.
6449 This function should only be called when C<REFCNT> is zero. Most of the time
6450 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6457 Perl_sv_clear(pTHX_ SV *const orig_sv)
6462 const struct body_details *sv_type_details;
6466 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6467 Not strictly necessary */
6469 PERL_ARGS_ASSERT_SV_CLEAR;
6471 /* within this loop, sv is the SV currently being freed, and
6472 * iter_sv is the most recent AV or whatever that's being iterated
6473 * over to provide more SVs */
6479 assert(SvREFCNT(sv) == 0);
6480 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6482 if (type <= SVt_IV) {
6483 /* See the comment in sv.h about the collusion between this
6484 * early return and the overloading of the NULL slots in the
6488 SvFLAGS(sv) &= SVf_BREAK;
6489 SvFLAGS(sv) |= SVTYPEMASK;
6493 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6494 for another purpose */
6495 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6497 if (type >= SVt_PVMG) {
6499 if (!curse(sv, 1)) goto get_next_sv;
6500 type = SvTYPE(sv); /* destructor may have changed it */
6502 /* Free back-references before magic, in case the magic calls
6503 * Perl code that has weak references to sv. */
6504 if (type == SVt_PVHV) {
6505 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6509 else if (SvMAGIC(sv)) {
6510 /* Free back-references before other types of magic. */
6511 sv_unmagic(sv, PERL_MAGIC_backref);
6517 /* case SVt_INVLIST: */
6520 IoIFP(sv) != PerlIO_stdin() &&
6521 IoIFP(sv) != PerlIO_stdout() &&
6522 IoIFP(sv) != PerlIO_stderr() &&
6523 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6525 io_close(MUTABLE_IO(sv), NULL, FALSE,
6526 (IoTYPE(sv) == IoTYPE_WRONLY ||
6527 IoTYPE(sv) == IoTYPE_RDWR ||
6528 IoTYPE(sv) == IoTYPE_APPEND));
6530 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6531 PerlDir_close(IoDIRP(sv));
6532 IoDIRP(sv) = (DIR*)NULL;
6533 Safefree(IoTOP_NAME(sv));
6534 Safefree(IoFMT_NAME(sv));
6535 Safefree(IoBOTTOM_NAME(sv));
6536 if ((const GV *)sv == PL_statgv)
6540 /* FIXME for plugins */
6542 pregfree2((REGEXP*) sv);
6546 cv_undef(MUTABLE_CV(sv));
6547 /* If we're in a stash, we don't own a reference to it.
6548 * However it does have a back reference to us, which needs to
6550 if ((stash = CvSTASH(sv)))
6551 sv_del_backref(MUTABLE_SV(stash), sv);
6554 if (PL_last_swash_hv == (const HV *)sv) {
6555 PL_last_swash_hv = NULL;
6557 if (HvTOTALKEYS((HV*)sv) > 0) {
6559 /* this statement should match the one at the beginning of
6560 * hv_undef_flags() */
6561 if ( PL_phase != PERL_PHASE_DESTRUCT
6562 && (hek = HvNAME_HEK((HV*)sv)))
6564 if (PL_stashcache) {
6565 DEBUG_o(Perl_deb(aTHX_
6566 "sv_clear clearing PL_stashcache for '%"HEKf
6569 (void)hv_deletehek(PL_stashcache,
6572 hv_name_set((HV*)sv, NULL, 0, 0);
6575 /* save old iter_sv in unused SvSTASH field */
6576 assert(!SvOBJECT(sv));
6577 SvSTASH(sv) = (HV*)iter_sv;
6580 /* save old hash_index in unused SvMAGIC field */
6581 assert(!SvMAGICAL(sv));
6582 assert(!SvMAGIC(sv));
6583 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6586 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6587 goto get_next_sv; /* process this new sv */
6589 /* free empty hash */
6590 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6591 assert(!HvARRAY((HV*)sv));
6595 AV* av = MUTABLE_AV(sv);
6596 if (PL_comppad == av) {
6600 if (AvREAL(av) && AvFILLp(av) > -1) {
6601 next_sv = AvARRAY(av)[AvFILLp(av)--];
6602 /* save old iter_sv in top-most slot of AV,
6603 * and pray that it doesn't get wiped in the meantime */
6604 AvARRAY(av)[AvMAX(av)] = iter_sv;
6606 goto get_next_sv; /* process this new sv */
6608 Safefree(AvALLOC(av));
6613 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6614 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6615 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6616 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6618 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6619 SvREFCNT_dec(LvTARG(sv));
6620 if (isREGEXP(sv)) goto freeregexp;
6623 if (isGV_with_GP(sv)) {
6624 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6625 && HvENAME_get(stash))
6626 mro_method_changed_in(stash);
6627 gp_free(MUTABLE_GV(sv));
6629 unshare_hek(GvNAME_HEK(sv));
6630 /* If we're in a stash, we don't own a reference to it.
6631 * However it does have a back reference to us, which
6632 * needs to be cleared. */
6633 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6634 sv_del_backref(MUTABLE_SV(stash), sv);
6636 /* FIXME. There are probably more unreferenced pointers to SVs
6637 * in the interpreter struct that we should check and tidy in
6638 * a similar fashion to this: */
6639 /* See also S_sv_unglob, which does the same thing. */
6640 if ((const GV *)sv == PL_last_in_gv)
6641 PL_last_in_gv = NULL;
6642 else if ((const GV *)sv == PL_statgv)
6644 else if ((const GV *)sv == PL_stderrgv)
6653 /* Don't bother with SvOOK_off(sv); as we're only going to
6657 SvOOK_offset(sv, offset);
6658 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6659 /* Don't even bother with turning off the OOK flag. */
6664 SV * const target = SvRV(sv);
6666 sv_del_backref(target, sv);
6672 else if (SvPVX_const(sv)
6673 && !(SvTYPE(sv) == SVt_PVIO
6674 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6678 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6682 if (CowREFCNT(sv)) {
6689 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6694 Safefree(SvPVX_mutable(sv));
6698 else if (SvPVX_const(sv) && SvLEN(sv)
6699 && !(SvTYPE(sv) == SVt_PVIO
6700 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6701 Safefree(SvPVX_mutable(sv));
6702 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6703 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6713 SvFLAGS(sv) &= SVf_BREAK;
6714 SvFLAGS(sv) |= SVTYPEMASK;
6716 sv_type_details = bodies_by_type + type;
6717 if (sv_type_details->arena) {
6718 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6719 &PL_body_roots[type]);
6721 else if (sv_type_details->body_size) {
6722 safefree(SvANY(sv));
6726 /* caller is responsible for freeing the head of the original sv */
6727 if (sv != orig_sv && !SvREFCNT(sv))
6730 /* grab and free next sv, if any */
6738 else if (!iter_sv) {
6740 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6741 AV *const av = (AV*)iter_sv;
6742 if (AvFILLp(av) > -1) {
6743 sv = AvARRAY(av)[AvFILLp(av)--];
6745 else { /* no more elements of current AV to free */
6748 /* restore previous value, squirrelled away */
6749 iter_sv = AvARRAY(av)[AvMAX(av)];
6750 Safefree(AvALLOC(av));
6753 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6754 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6755 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6756 /* no more elements of current HV to free */
6759 /* Restore previous values of iter_sv and hash_index,
6760 * squirrelled away */
6761 assert(!SvOBJECT(sv));
6762 iter_sv = (SV*)SvSTASH(sv);
6763 assert(!SvMAGICAL(sv));
6764 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6766 /* perl -DA does not like rubbish in SvMAGIC. */
6770 /* free any remaining detritus from the hash struct */
6771 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6772 assert(!HvARRAY((HV*)sv));
6777 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6781 if (!SvREFCNT(sv)) {
6785 if (--(SvREFCNT(sv)))
6789 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6790 "Attempt to free temp prematurely: SV 0x%"UVxf
6791 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6795 if (SvIMMORTAL(sv)) {
6796 /* make sure SvREFCNT(sv)==0 happens very seldom */
6797 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6806 /* This routine curses the sv itself, not the object referenced by sv. So
6807 sv does not have to be ROK. */
6810 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6811 PERL_ARGS_ASSERT_CURSE;
6812 assert(SvOBJECT(sv));
6814 if (PL_defstash && /* Still have a symbol table? */
6820 stash = SvSTASH(sv);
6821 assert(SvTYPE(stash) == SVt_PVHV);
6822 if (HvNAME(stash)) {
6823 CV* destructor = NULL;
6824 struct mro_meta *meta;
6826 assert (SvOOK(stash));
6828 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6831 /* don't make this an initialization above the assert, since it needs
6833 meta = HvMROMETA(stash);
6834 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6835 destructor = meta->destroy;
6836 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6837 (void *)destructor, HvNAME(stash)) );
6840 bool autoload = FALSE;
6842 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6844 destructor = GvCV(gv);
6846 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6847 GV_AUTOLOAD_ISMETHOD);
6849 destructor = GvCV(gv);
6853 /* we don't cache AUTOLOAD for DESTROY, since this code
6854 would then need to set $__PACKAGE__::AUTOLOAD, or the
6855 equivalent for XS AUTOLOADs */
6857 meta->destroy_gen = PL_sub_generation;
6858 meta->destroy = destructor;
6860 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6861 (void *)destructor, HvNAME(stash)) );
6864 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6868 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6870 /* A constant subroutine can have no side effects, so
6871 don't bother calling it. */
6872 && !CvCONST(destructor)
6873 /* Don't bother calling an empty destructor or one that
6874 returns immediately. */
6875 && (CvISXSUB(destructor)
6876 || (CvSTART(destructor)
6877 && (CvSTART(destructor)->op_next->op_type
6879 && (CvSTART(destructor)->op_next->op_type
6881 || CvSTART(destructor)->op_next->op_next->op_type
6887 SV* const tmpref = newRV(sv);
6888 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6890 PUSHSTACKi(PERLSI_DESTROY);
6895 call_sv(MUTABLE_SV(destructor),
6896 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6900 if(SvREFCNT(tmpref) < 2) {
6901 /* tmpref is not kept alive! */
6903 SvRV_set(tmpref, NULL);
6906 SvREFCNT_dec_NN(tmpref);
6909 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6912 if (check_refcnt && SvREFCNT(sv)) {
6913 if (PL_in_clean_objs)
6915 "DESTROY created new reference to dead object '%"HEKf"'",
6916 HEKfARG(HvNAME_HEK(stash)));
6917 /* DESTROY gave object new lease on life */
6923 HV * const stash = SvSTASH(sv);
6924 /* Curse before freeing the stash, as freeing the stash could cause
6925 a recursive call into S_curse. */
6926 SvOBJECT_off(sv); /* Curse the object. */
6927 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6928 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6934 =for apidoc sv_newref
6936 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6943 Perl_sv_newref(pTHX_ SV *const sv)
6945 PERL_UNUSED_CONTEXT;
6954 Decrement an SV's reference count, and if it drops to zero, call
6955 C<sv_clear> to invoke destructors and free up any memory used by
6956 the body; finally, deallocating the SV's head itself.
6957 Normally called via a wrapper macro C<SvREFCNT_dec>.
6963 Perl_sv_free(pTHX_ SV *const sv)
6969 /* Private helper function for SvREFCNT_dec().
6970 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6973 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6977 PERL_ARGS_ASSERT_SV_FREE2;
6979 if (LIKELY( rc == 1 )) {
6985 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6986 "Attempt to free temp prematurely: SV 0x%"UVxf
6987 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6991 if (SvIMMORTAL(sv)) {
6992 /* make sure SvREFCNT(sv)==0 happens very seldom */
6993 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6997 if (! SvREFCNT(sv)) /* may have have been resurrected */
7002 /* handle exceptional cases */
7006 if (SvFLAGS(sv) & SVf_BREAK)
7007 /* this SV's refcnt has been artificially decremented to
7008 * trigger cleanup */
7010 if (PL_in_clean_all) /* All is fair */
7012 if (SvIMMORTAL(sv)) {
7013 /* make sure SvREFCNT(sv)==0 happens very seldom */
7014 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7017 if (ckWARN_d(WARN_INTERNAL)) {
7018 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7019 Perl_dump_sv_child(aTHX_ sv);
7021 #ifdef DEBUG_LEAKING_SCALARS
7024 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7025 if (PL_warnhook == PERL_WARNHOOK_FATAL
7026 || ckDEAD(packWARN(WARN_INTERNAL))) {
7027 /* Don't let Perl_warner cause us to escape our fate: */
7031 /* This may not return: */
7032 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7033 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7034 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7037 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7047 Returns the length of the string in the SV. Handles magic and type
7048 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7049 gives raw access to the C<xpv_cur> slot.
7055 Perl_sv_len(pTHX_ SV *const sv)
7062 (void)SvPV_const(sv, len);
7067 =for apidoc sv_len_utf8
7069 Returns the number of characters in the string in an SV, counting wide
7070 UTF-8 bytes as a single character. Handles magic and type coercion.
7076 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7077 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7078 * (Note that the mg_len is not the length of the mg_ptr field.
7079 * This allows the cache to store the character length of the string without
7080 * needing to malloc() extra storage to attach to the mg_ptr.)
7085 Perl_sv_len_utf8(pTHX_ SV *const sv)
7091 return sv_len_utf8_nomg(sv);
7095 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7098 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7100 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7102 if (PL_utf8cache && SvUTF8(sv)) {
7104 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7106 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7107 if (mg->mg_len != -1)
7110 /* We can use the offset cache for a headstart.
7111 The longer value is stored in the first pair. */
7112 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7114 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7118 if (PL_utf8cache < 0) {
7119 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7120 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7124 ulen = Perl_utf8_length(aTHX_ s, s + len);
7125 utf8_mg_len_cache_update(sv, &mg, ulen);
7129 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7132 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7135 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7136 STRLEN *const uoffset_p, bool *const at_end)
7138 const U8 *s = start;
7139 STRLEN uoffset = *uoffset_p;
7141 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7143 while (s < send && uoffset) {
7150 else if (s > send) {
7152 /* This is the existing behaviour. Possibly it should be a croak, as
7153 it's actually a bounds error */
7156 *uoffset_p -= uoffset;
7160 /* Given the length of the string in both bytes and UTF-8 characters, decide
7161 whether to walk forwards or backwards to find the byte corresponding to
7162 the passed in UTF-8 offset. */
7164 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7165 STRLEN uoffset, const STRLEN uend)
7167 STRLEN backw = uend - uoffset;
7169 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7171 if (uoffset < 2 * backw) {
7172 /* The assumption is that going forwards is twice the speed of going
7173 forward (that's where the 2 * backw comes from).
7174 (The real figure of course depends on the UTF-8 data.) */
7175 const U8 *s = start;
7177 while (s < send && uoffset--)
7187 while (UTF8_IS_CONTINUATION(*send))
7190 return send - start;
7193 /* For the string representation of the given scalar, find the byte
7194 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7195 give another position in the string, *before* the sought offset, which
7196 (which is always true, as 0, 0 is a valid pair of positions), which should
7197 help reduce the amount of linear searching.
7198 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7199 will be used to reduce the amount of linear searching. The cache will be
7200 created if necessary, and the found value offered to it for update. */
7202 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7203 const U8 *const send, STRLEN uoffset,
7204 STRLEN uoffset0, STRLEN boffset0)
7206 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7208 bool at_end = FALSE;
7210 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7212 assert (uoffset >= uoffset0);
7217 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7219 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7220 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7221 if ((*mgp)->mg_ptr) {
7222 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7223 if (cache[0] == uoffset) {
7224 /* An exact match. */
7227 if (cache[2] == uoffset) {
7228 /* An exact match. */
7232 if (cache[0] < uoffset) {
7233 /* The cache already knows part of the way. */
7234 if (cache[0] > uoffset0) {
7235 /* The cache knows more than the passed in pair */
7236 uoffset0 = cache[0];
7237 boffset0 = cache[1];
7239 if ((*mgp)->mg_len != -1) {
7240 /* And we know the end too. */
7242 + sv_pos_u2b_midway(start + boffset0, send,
7244 (*mgp)->mg_len - uoffset0);
7246 uoffset -= uoffset0;
7248 + sv_pos_u2b_forwards(start + boffset0,
7249 send, &uoffset, &at_end);
7250 uoffset += uoffset0;
7253 else if (cache[2] < uoffset) {
7254 /* We're between the two cache entries. */
7255 if (cache[2] > uoffset0) {
7256 /* and the cache knows more than the passed in pair */
7257 uoffset0 = cache[2];
7258 boffset0 = cache[3];
7262 + sv_pos_u2b_midway(start + boffset0,
7265 cache[0] - uoffset0);
7268 + sv_pos_u2b_midway(start + boffset0,
7271 cache[2] - uoffset0);
7275 else if ((*mgp)->mg_len != -1) {
7276 /* If we can take advantage of a passed in offset, do so. */
7277 /* In fact, offset0 is either 0, or less than offset, so don't
7278 need to worry about the other possibility. */
7280 + sv_pos_u2b_midway(start + boffset0, send,
7282 (*mgp)->mg_len - uoffset0);
7287 if (!found || PL_utf8cache < 0) {
7288 STRLEN real_boffset;
7289 uoffset -= uoffset0;
7290 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7291 send, &uoffset, &at_end);
7292 uoffset += uoffset0;
7294 if (found && PL_utf8cache < 0)
7295 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7297 boffset = real_boffset;
7300 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7302 utf8_mg_len_cache_update(sv, mgp, uoffset);
7304 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7311 =for apidoc sv_pos_u2b_flags
7313 Converts the offset from a count of UTF-8 chars from
7314 the start of the string, to a count of the equivalent number of bytes; if
7315 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7316 C<offset>, rather than from the start
7317 of the string. Handles type coercion.
7318 C<flags> is passed to C<SvPV_flags>, and usually should be
7319 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7325 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7326 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7327 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7332 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7339 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7341 start = (U8*)SvPV_flags(sv, len, flags);
7343 const U8 * const send = start + len;
7345 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7348 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7349 is 0, and *lenp is already set to that. */) {
7350 /* Convert the relative offset to absolute. */
7351 const STRLEN uoffset2 = uoffset + *lenp;
7352 const STRLEN boffset2
7353 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7354 uoffset, boffset) - boffset;
7368 =for apidoc sv_pos_u2b
7370 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7371 the start of the string, to a count of the equivalent number of bytes; if
7372 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7373 the offset, rather than from the start of the string. Handles magic and
7376 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7383 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7384 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7385 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7389 /* This function is subject to size and sign problems */
7392 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7394 PERL_ARGS_ASSERT_SV_POS_U2B;
7397 STRLEN ulen = (STRLEN)*lenp;
7398 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7399 SV_GMAGIC|SV_CONST_RETURN);
7402 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7403 SV_GMAGIC|SV_CONST_RETURN);
7408 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7411 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7412 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7415 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7416 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7417 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7421 (*mgp)->mg_len = ulen;
7424 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7425 byte length pairing. The (byte) length of the total SV is passed in too,
7426 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7427 may not have updated SvCUR, so we can't rely on reading it directly.
7429 The proffered utf8/byte length pairing isn't used if the cache already has
7430 two pairs, and swapping either for the proffered pair would increase the
7431 RMS of the intervals between known byte offsets.
7433 The cache itself consists of 4 STRLEN values
7434 0: larger UTF-8 offset
7435 1: corresponding byte offset
7436 2: smaller UTF-8 offset
7437 3: corresponding byte offset
7439 Unused cache pairs have the value 0, 0.
7440 Keeping the cache "backwards" means that the invariant of
7441 cache[0] >= cache[2] is maintained even with empty slots, which means that
7442 the code that uses it doesn't need to worry if only 1 entry has actually
7443 been set to non-zero. It also makes the "position beyond the end of the
7444 cache" logic much simpler, as the first slot is always the one to start
7448 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7449 const STRLEN utf8, const STRLEN blen)
7453 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7458 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7459 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7460 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7462 (*mgp)->mg_len = -1;
7466 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7467 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7468 (*mgp)->mg_ptr = (char *) cache;
7472 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7473 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7474 a pointer. Note that we no longer cache utf8 offsets on refer-
7475 ences, but this check is still a good idea, for robustness. */
7476 const U8 *start = (const U8 *) SvPVX_const(sv);
7477 const STRLEN realutf8 = utf8_length(start, start + byte);
7479 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7483 /* Cache is held with the later position first, to simplify the code
7484 that deals with unbounded ends. */
7486 ASSERT_UTF8_CACHE(cache);
7487 if (cache[1] == 0) {
7488 /* Cache is totally empty */
7491 } else if (cache[3] == 0) {
7492 if (byte > cache[1]) {
7493 /* New one is larger, so goes first. */
7494 cache[2] = cache[0];
7495 cache[3] = cache[1];
7503 /* float casts necessary? XXX */
7504 #define THREEWAY_SQUARE(a,b,c,d) \
7505 ((float)((d) - (c))) * ((float)((d) - (c))) \
7506 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7507 + ((float)((b) - (a))) * ((float)((b) - (a)))
7509 /* Cache has 2 slots in use, and we know three potential pairs.
7510 Keep the two that give the lowest RMS distance. Do the
7511 calculation in bytes simply because we always know the byte
7512 length. squareroot has the same ordering as the positive value,
7513 so don't bother with the actual square root. */
7514 if (byte > cache[1]) {
7515 /* New position is after the existing pair of pairs. */
7516 const float keep_earlier
7517 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7518 const float keep_later
7519 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7521 if (keep_later < keep_earlier) {
7522 cache[2] = cache[0];
7523 cache[3] = cache[1];
7529 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7530 float b, c, keep_earlier;
7531 if (byte > cache[3]) {
7532 /* New position is between the existing pair of pairs. */
7533 b = (float)cache[3];
7536 /* New position is before the existing pair of pairs. */
7538 c = (float)cache[3];
7540 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7541 if (byte > cache[3]) {
7542 if (keep_later < keep_earlier) {
7552 if (! (keep_later < keep_earlier)) {
7553 cache[0] = cache[2];
7554 cache[1] = cache[3];
7561 ASSERT_UTF8_CACHE(cache);
7564 /* We already know all of the way, now we may be able to walk back. The same
7565 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7566 backward is half the speed of walking forward. */
7568 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7569 const U8 *end, STRLEN endu)
7571 const STRLEN forw = target - s;
7572 STRLEN backw = end - target;
7574 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7576 if (forw < 2 * backw) {
7577 return utf8_length(s, target);
7580 while (end > target) {
7582 while (UTF8_IS_CONTINUATION(*end)) {
7591 =for apidoc sv_pos_b2u_flags
7593 Converts C<offset> from a count of bytes from the start of the string, to
7594 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7595 C<flags> is passed to C<SvPV_flags>, and usually should be
7596 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7602 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7603 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7608 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7611 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7617 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7619 s = (const U8*)SvPV_flags(sv, blen, flags);
7622 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7623 ", byte=%"UVuf, (UV)blen, (UV)offset);
7629 && SvTYPE(sv) >= SVt_PVMG
7630 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7633 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7634 if (cache[1] == offset) {
7635 /* An exact match. */
7638 if (cache[3] == offset) {
7639 /* An exact match. */
7643 if (cache[1] < offset) {
7644 /* We already know part of the way. */
7645 if (mg->mg_len != -1) {
7646 /* Actually, we know the end too. */
7648 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7649 s + blen, mg->mg_len - cache[0]);
7651 len = cache[0] + utf8_length(s + cache[1], send);
7654 else if (cache[3] < offset) {
7655 /* We're between the two cached pairs, so we do the calculation
7656 offset by the byte/utf-8 positions for the earlier pair,
7657 then add the utf-8 characters from the string start to
7659 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7660 s + cache[1], cache[0] - cache[2])
7664 else { /* cache[3] > offset */
7665 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7669 ASSERT_UTF8_CACHE(cache);
7671 } else if (mg->mg_len != -1) {
7672 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7676 if (!found || PL_utf8cache < 0) {
7677 const STRLEN real_len = utf8_length(s, send);
7679 if (found && PL_utf8cache < 0)
7680 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7686 utf8_mg_len_cache_update(sv, &mg, len);
7688 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7695 =for apidoc sv_pos_b2u
7697 Converts the value pointed to by C<offsetp> from a count of bytes from the
7698 start of the string, to a count of the equivalent number of UTF-8 chars.
7699 Handles magic and type coercion.
7701 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7708 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7709 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7714 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7716 PERL_ARGS_ASSERT_SV_POS_B2U;
7721 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7722 SV_GMAGIC|SV_CONST_RETURN);
7726 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7727 STRLEN real, SV *const sv)
7729 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7731 /* As this is debugging only code, save space by keeping this test here,
7732 rather than inlining it in all the callers. */
7733 if (from_cache == real)
7736 /* Need to turn the assertions off otherwise we may recurse infinitely
7737 while printing error messages. */
7738 SAVEI8(PL_utf8cache);
7740 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7741 func, (UV) from_cache, (UV) real, SVfARG(sv));
7747 Returns a boolean indicating whether the strings in the two SVs are
7748 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7749 coerce its args to strings if necessary.
7751 =for apidoc sv_eq_flags
7753 Returns a boolean indicating whether the strings in the two SVs are
7754 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7755 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7761 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7768 SV* svrecode = NULL;
7775 /* if pv1 and pv2 are the same, second SvPV_const call may
7776 * invalidate pv1 (if we are handling magic), so we may need to
7778 if (sv1 == sv2 && flags & SV_GMAGIC
7779 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7780 pv1 = SvPV_const(sv1, cur1);
7781 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7783 pv1 = SvPV_flags_const(sv1, cur1, flags);
7791 pv2 = SvPV_flags_const(sv2, cur2, flags);
7793 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7794 /* Differing utf8ness. */
7796 /* sv1 is the UTF-8 one */
7797 return bytes_cmp_utf8((const U8*)pv2, cur2,
7798 (const U8*)pv1, cur1) == 0;
7801 /* sv2 is the UTF-8 one */
7802 return bytes_cmp_utf8((const U8*)pv1, cur1,
7803 (const U8*)pv2, cur2) == 0;
7808 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7810 SvREFCNT_dec(svrecode);
7818 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7819 string in C<sv1> is less than, equal to, or greater than the string in
7820 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7821 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7823 =for apidoc sv_cmp_flags
7825 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7826 string in C<sv1> is less than, equal to, or greater than the string in
7827 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7828 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7829 also C<L</sv_cmp_locale_flags>>.
7835 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7837 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7841 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7845 const char *pv1, *pv2;
7847 SV *svrecode = NULL;
7854 pv1 = SvPV_flags_const(sv1, cur1, flags);
7861 pv2 = SvPV_flags_const(sv2, cur2, flags);
7863 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7864 /* Differing utf8ness. */
7866 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7867 (const U8*)pv1, cur1);
7868 return retval ? retval < 0 ? -1 : +1 : 0;
7871 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7872 (const U8*)pv2, cur2);
7873 return retval ? retval < 0 ? -1 : +1 : 0;
7877 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7880 cmp = cur2 ? -1 : 0;
7884 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7887 if (! DO_UTF8(sv1)) {
7889 const I32 retval = memcmp((const void*)pv1,
7893 cmp = retval < 0 ? -1 : 1;
7894 } else if (cur1 == cur2) {
7897 cmp = cur1 < cur2 ? -1 : 1;
7901 else { /* Both are to be treated as UTF-EBCDIC */
7903 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7904 * which remaps code points 0-255. We therefore generally have to
7905 * unmap back to the original values to get an accurate comparison.
7906 * But we don't have to do that for UTF-8 invariants, as by
7907 * definition, they aren't remapped, nor do we have to do it for
7908 * above-latin1 code points, as they also aren't remapped. (This
7909 * code also works on ASCII platforms, but the memcmp() above is
7912 const char *e = pv1 + shortest_len;
7914 /* Find the first bytes that differ between the two strings */
7915 while (pv1 < e && *pv1 == *pv2) {
7921 if (pv1 == e) { /* Are the same all the way to the end */
7925 cmp = cur1 < cur2 ? -1 : 1;
7928 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7929 * in the strings were. The current bytes may or may not be
7930 * at the beginning of a character. But neither or both are
7931 * (or else earlier bytes would have been different). And
7932 * if we are in the middle of a character, the two
7933 * characters are comprised of the same number of bytes
7934 * (because in this case the start bytes are the same, and
7935 * the start bytes encode the character's length). */
7936 if (UTF8_IS_INVARIANT(*pv1))
7938 /* If both are invariants; can just compare directly */
7939 if (UTF8_IS_INVARIANT(*pv2)) {
7940 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7942 else /* Since *pv1 is invariant, it is the whole character,
7943 which means it is at the beginning of a character.
7944 That means pv2 is also at the beginning of a
7945 character (see earlier comment). Since it isn't
7946 invariant, it must be a start byte. If it starts a
7947 character whose code point is above 255, that
7948 character is greater than any single-byte char, which
7950 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7955 /* Here, pv2 points to a character composed of 2 bytes
7956 * whose code point is < 256. Get its code point and
7957 * compare with *pv1 */
7958 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7963 else /* The code point starting at pv1 isn't a single byte */
7964 if (UTF8_IS_INVARIANT(*pv2))
7966 /* But here, the code point starting at *pv2 is a single byte,
7967 * and so *pv1 must begin a character, hence is a start byte.
7968 * If that character is above 255, it is larger than any
7969 * single-byte char, which *pv2 is */
7970 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7974 /* Here, pv1 points to a character composed of 2 bytes
7975 * whose code point is < 256. Get its code point and
7976 * compare with the single byte character *pv2 */
7977 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7982 else /* Here, we've ruled out either *pv1 and *pv2 being
7983 invariant. That means both are part of variants, but not
7984 necessarily at the start of a character */
7985 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7986 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7988 /* Here, at least one is the start of a character, which means
7989 * the other is also a start byte. And the code point of at
7990 * least one of the characters is above 255. It is a
7991 * characteristic of UTF-EBCDIC that all start bytes for
7992 * above-latin1 code points are well behaved as far as code
7993 * point comparisons go, and all are larger than all other
7994 * start bytes, so the comparison with those is also well
7996 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7999 /* Here both *pv1 and *pv2 are part of variant characters.
8000 * They could be both continuations, or both start characters.
8001 * (One or both could even be an illegal start character (for
8002 * an overlong) which for the purposes of sorting we treat as
8004 if (UTF8_IS_CONTINUATION(*pv1)) {
8006 /* If they are continuations for code points above 255,
8007 * then comparing the current byte is sufficient, as there
8008 * is no remapping of these and so the comparison is
8009 * well-behaved. We determine if they are such
8010 * continuations by looking at the preceding byte. It
8011 * could be a start byte, from which we can tell if it is
8012 * for an above 255 code point. Or it could be a
8013 * continuation, which means the character occupies at
8014 * least 3 bytes, so must be above 255. */
8015 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8016 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8018 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8022 /* Here, the continuations are for code points below 256;
8023 * back up one to get to the start byte */
8028 /* We need to get the actual native code point of each of these
8029 * variants in order to compare them */
8030 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8031 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8040 SvREFCNT_dec(svrecode);
8046 =for apidoc sv_cmp_locale
8048 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8049 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8050 if necessary. See also C<L</sv_cmp>>.
8052 =for apidoc sv_cmp_locale_flags
8054 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8055 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8056 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8057 C<L</sv_cmp_flags>>.
8063 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8065 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8069 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8072 #ifdef USE_LOCALE_COLLATE
8078 if (PL_collation_standard)
8083 /* Revert to using raw compare if both operands exist, but either one
8084 * doesn't transform properly for collation */
8086 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8090 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8096 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8097 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8100 if (!pv1 || !len1) {
8111 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8114 return retval < 0 ? -1 : 1;
8117 * When the result of collation is equality, that doesn't mean
8118 * that there are no differences -- some locales exclude some
8119 * characters from consideration. So to avoid false equalities,
8120 * we use the raw string as a tiebreaker.
8127 PERL_UNUSED_ARG(flags);
8128 #endif /* USE_LOCALE_COLLATE */
8130 return sv_cmp(sv1, sv2);
8134 #ifdef USE_LOCALE_COLLATE
8137 =for apidoc sv_collxfrm
8139 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8140 C<L</sv_collxfrm_flags>>.
8142 =for apidoc sv_collxfrm_flags
8144 Add Collate Transform magic to an SV if it doesn't already have it. If the
8145 flags contain C<SV_GMAGIC>, it handles get-magic.
8147 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8148 scalar data of the variable, but transformed to such a format that a normal
8149 memory comparison can be used to compare the data according to the locale
8156 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8160 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8162 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8164 /* If we don't have collation magic on 'sv', or the locale has changed
8165 * since the last time we calculated it, get it and save it now */
8166 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8171 /* Free the old space */
8173 Safefree(mg->mg_ptr);
8175 s = SvPV_flags_const(sv, len, flags);
8176 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8178 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8193 if (mg && mg->mg_ptr) {
8195 return mg->mg_ptr + sizeof(PL_collation_ix);
8203 #endif /* USE_LOCALE_COLLATE */
8206 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8208 SV * const tsv = newSV(0);
8211 sv_gets(tsv, fp, 0);
8212 sv_utf8_upgrade_nomg(tsv);
8213 SvCUR_set(sv,append);
8216 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8220 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8223 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8224 /* Grab the size of the record we're getting */
8225 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8232 /* With a true, record-oriented file on VMS, we need to use read directly
8233 * to ensure that we respect RMS record boundaries. The user is responsible
8234 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8235 * record size) field. N.B. This is likely to produce invalid results on
8236 * varying-width character data when a record ends mid-character.
8238 fd = PerlIO_fileno(fp);
8240 && PerlLIO_fstat(fd, &st) == 0
8241 && (st.st_fab_rfm == FAB$C_VAR
8242 || st.st_fab_rfm == FAB$C_VFC
8243 || st.st_fab_rfm == FAB$C_FIX)) {
8245 bytesread = PerlLIO_read(fd, buffer, recsize);
8247 else /* in-memory file from PerlIO::Scalar
8248 * or not a record-oriented file
8252 bytesread = PerlIO_read(fp, buffer, recsize);
8254 /* At this point, the logic in sv_get() means that sv will
8255 be treated as utf-8 if the handle is utf8.
8257 if (PerlIO_isutf8(fp) && bytesread > 0) {
8258 char *bend = buffer + bytesread;
8259 char *bufp = buffer;
8260 size_t charcount = 0;
8261 bool charstart = TRUE;
8264 while (charcount < recsize) {
8265 /* count accumulated characters */
8266 while (bufp < bend) {
8268 skip = UTF8SKIP(bufp);
8270 if (bufp + skip > bend) {
8271 /* partial at the end */
8282 if (charcount < recsize) {
8284 STRLEN bufp_offset = bufp - buffer;
8285 SSize_t morebytesread;
8287 /* originally I read enough to fill any incomplete
8288 character and the first byte of the next
8289 character if needed, but if there's many
8290 multi-byte encoded characters we're going to be
8291 making a read call for every character beyond
8292 the original read size.
8294 So instead, read the rest of the character if
8295 any, and enough bytes to match at least the
8296 start bytes for each character we're going to
8300 readsize = recsize - charcount;
8302 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8303 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8304 bend = buffer + bytesread;
8305 morebytesread = PerlIO_read(fp, bend, readsize);
8306 if (morebytesread <= 0) {
8307 /* we're done, if we still have incomplete
8308 characters the check code in sv_gets() will
8311 I'd originally considered doing
8312 PerlIO_ungetc() on all but the lead
8313 character of the incomplete character, but
8314 read() doesn't do that, so I don't.
8319 /* prepare to scan some more */
8320 bytesread += morebytesread;
8321 bend = buffer + bytesread;
8322 bufp = buffer + bufp_offset;
8330 SvCUR_set(sv, bytesread + append);
8331 buffer[bytesread] = '\0';
8332 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8338 Get a line from the filehandle and store it into the SV, optionally
8339 appending to the currently-stored string. If C<append> is not 0, the
8340 line is appended to the SV instead of overwriting it. C<append> should
8341 be set to the byte offset that the appended string should start at
8342 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8348 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8358 PERL_ARGS_ASSERT_SV_GETS;
8360 if (SvTHINKFIRST(sv))
8361 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8362 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8364 However, perlbench says it's slower, because the existing swipe code
8365 is faster than copy on write.
8366 Swings and roundabouts. */
8367 SvUPGRADE(sv, SVt_PV);
8370 /* line is going to be appended to the existing buffer in the sv */
8371 if (PerlIO_isutf8(fp)) {
8373 sv_utf8_upgrade_nomg(sv);
8374 sv_pos_u2b(sv,&append,0);
8376 } else if (SvUTF8(sv)) {
8377 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8383 /* not appending - "clear" the string by setting SvCUR to 0,
8384 * the pv is still avaiable. */
8387 if (PerlIO_isutf8(fp))
8390 if (IN_PERL_COMPILETIME) {
8391 /* we always read code in line mode */
8395 else if (RsSNARF(PL_rs)) {
8396 /* If it is a regular disk file use size from stat() as estimate
8397 of amount we are going to read -- may result in mallocing
8398 more memory than we really need if the layers below reduce
8399 the size we read (e.g. CRLF or a gzip layer).
8402 int fd = PerlIO_fileno(fp);
8403 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8404 const Off_t offset = PerlIO_tell(fp);
8405 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8406 #ifdef PERL_COPY_ON_WRITE
8407 /* Add an extra byte for the sake of copy-on-write's
8408 * buffer reference count. */
8409 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8411 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8418 else if (RsRECORD(PL_rs)) {
8419 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8421 else if (RsPARA(PL_rs)) {
8427 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8428 if (PerlIO_isutf8(fp)) {
8429 rsptr = SvPVutf8(PL_rs, rslen);
8432 if (SvUTF8(PL_rs)) {
8433 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8434 Perl_croak(aTHX_ "Wide character in $/");
8437 /* extract the raw pointer to the record separator */
8438 rsptr = SvPV_const(PL_rs, rslen);
8442 /* rslast is the last character in the record separator
8443 * note we don't use rslast except when rslen is true, so the
8444 * null assign is a placeholder. */
8445 rslast = rslen ? rsptr[rslen - 1] : '\0';
8447 if (rspara) { /* have to do this both before and after */
8448 do { /* to make sure file boundaries work right */
8451 i = PerlIO_getc(fp);
8455 PerlIO_ungetc(fp,i);
8461 /* See if we know enough about I/O mechanism to cheat it ! */
8463 /* This used to be #ifdef test - it is made run-time test for ease
8464 of abstracting out stdio interface. One call should be cheap
8465 enough here - and may even be a macro allowing compile
8469 if (PerlIO_fast_gets(fp)) {
8471 * We can do buffer based IO operations on this filehandle.
8473 * This means we can bypass a lot of subcalls and process
8474 * the buffer directly, it also means we know the upper bound
8475 * on the amount of data we might read of the current buffer
8476 * into our sv. Knowing this allows us to preallocate the pv
8477 * to be able to hold that maximum, which allows us to simplify
8478 * a lot of logic. */
8481 * We're going to steal some values from the stdio struct
8482 * and put EVERYTHING in the innermost loop into registers.
8484 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8485 STRLEN bpx; /* length of the data in the target sv
8486 used to fix pointers after a SvGROW */
8487 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8488 of data left in the read-ahead buffer.
8489 If 0 then the pv buffer can hold the full
8490 amount left, otherwise this is the amount it
8493 /* Here is some breathtakingly efficient cheating */
8495 /* When you read the following logic resist the urge to think
8496 * of record separators that are 1 byte long. They are an
8497 * uninteresting special (simple) case.
8499 * Instead think of record separators which are at least 2 bytes
8500 * long, and keep in mind that we need to deal with such
8501 * separators when they cross a read-ahead buffer boundary.
8503 * Also consider that we need to gracefully deal with separators
8504 * that may be longer than a single read ahead buffer.
8506 * Lastly do not forget we want to copy the delimiter as well. We
8507 * are copying all data in the file _up_to_and_including_ the separator
8510 * Now that you have all that in mind here is what is happening below:
8512 * 1. When we first enter the loop we do some memory book keeping to see
8513 * how much free space there is in the target SV. (This sub assumes that
8514 * it is operating on the same SV most of the time via $_ and that it is
8515 * going to be able to reuse the same pv buffer each call.) If there is
8516 * "enough" room then we set "shortbuffered" to how much space there is
8517 * and start reading forward.
8519 * 2. When we scan forward we copy from the read-ahead buffer to the target
8520 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8521 * and the end of the of pv, as well as for the "rslast", which is the last
8522 * char of the separator.
8524 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8525 * (which has a "complete" record up to the point we saw rslast) and check
8526 * it to see if it matches the separator. If it does we are done. If it doesn't
8527 * we continue on with the scan/copy.
8529 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8530 * the IO system to read the next buffer. We do this by doing a getc(), which
8531 * returns a single char read (or EOF), and prefills the buffer, and also
8532 * allows us to find out how full the buffer is. We use this information to
8533 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8534 * the returned single char into the target sv, and then go back into scan
8537 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8538 * remaining space in the read-buffer.
8540 * Note that this code despite its twisty-turny nature is pretty darn slick.
8541 * It manages single byte separators, multi-byte cross boundary separators,
8542 * and cross-read-buffer separators cleanly and efficiently at the cost
8543 * of potentially greatly overallocating the target SV.
8549 /* get the number of bytes remaining in the read-ahead buffer
8550 * on first call on a given fp this will return 0.*/
8551 cnt = PerlIO_get_cnt(fp);
8553 /* make sure we have the room */
8554 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8555 /* Not room for all of it
8556 if we are looking for a separator and room for some
8558 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8559 /* just process what we have room for */
8560 shortbuffered = cnt - SvLEN(sv) + append + 1;
8561 cnt -= shortbuffered;
8564 /* ensure that the target sv has enough room to hold
8565 * the rest of the read-ahead buffer */
8567 /* remember that cnt can be negative */
8568 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8572 /* we have enough room to hold the full buffer, lets scream */
8576 /* extract the pointer to sv's string buffer, offset by append as necessary */
8577 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8578 /* extract the point to the read-ahead buffer */
8579 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8581 /* some trace debug output */
8582 DEBUG_P(PerlIO_printf(Perl_debug_log,
8583 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8584 DEBUG_P(PerlIO_printf(Perl_debug_log,
8585 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8587 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8588 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8592 /* if there is stuff left in the read-ahead buffer */
8594 /* if there is a separator */
8596 /* loop until we hit the end of the read-ahead buffer */
8597 while (cnt > 0) { /* this | eat */
8598 /* scan forward copying and searching for rslast as we go */
8600 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8601 goto thats_all_folks; /* screams | sed :-) */
8605 /* no separator, slurp the full buffer */
8606 Copy(ptr, bp, cnt, char); /* this | eat */
8607 bp += cnt; /* screams | dust */
8608 ptr += cnt; /* louder | sed :-) */
8610 assert (!shortbuffered);
8611 goto cannot_be_shortbuffered;
8615 if (shortbuffered) { /* oh well, must extend */
8616 /* we didnt have enough room to fit the line into the target buffer
8617 * so we must extend the target buffer and keep going */
8618 cnt = shortbuffered;
8620 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8622 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8623 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8624 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8628 cannot_be_shortbuffered:
8629 /* we need to refill the read-ahead buffer if possible */
8631 DEBUG_P(PerlIO_printf(Perl_debug_log,
8632 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8633 PTR2UV(ptr),(IV)cnt));
8634 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8636 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8637 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8638 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8639 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8642 call PerlIO_getc() to let it prefill the lookahead buffer
8644 This used to call 'filbuf' in stdio form, but as that behaves like
8645 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8646 another abstraction.
8648 Note we have to deal with the char in 'i' if we are not at EOF
8650 i = PerlIO_getc(fp); /* get more characters */
8652 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8653 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8654 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8655 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8657 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8658 cnt = PerlIO_get_cnt(fp);
8659 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8660 DEBUG_P(PerlIO_printf(Perl_debug_log,
8661 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8662 PTR2UV(ptr),(IV)cnt));
8664 if (i == EOF) /* all done for ever? */
8665 goto thats_really_all_folks;
8667 /* make sure we have enough space in the target sv */
8668 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8670 SvGROW(sv, bpx + cnt + 2);
8671 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8673 /* copy of the char we got from getc() */
8674 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8676 /* make sure we deal with the i being the last character of a separator */
8677 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8678 goto thats_all_folks;
8682 /* check if we have actually found the separator - only really applies
8684 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8685 memNE((char*)bp - rslen, rsptr, rslen))
8686 goto screamer; /* go back to the fray */
8687 thats_really_all_folks:
8689 cnt += shortbuffered;
8690 DEBUG_P(PerlIO_printf(Perl_debug_log,
8691 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8692 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8693 DEBUG_P(PerlIO_printf(Perl_debug_log,
8694 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8696 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8697 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8699 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8700 DEBUG_P(PerlIO_printf(Perl_debug_log,
8701 "Screamer: done, len=%ld, string=|%.*s|\n",
8702 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8706 /*The big, slow, and stupid way. */
8707 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8708 STDCHAR *buf = NULL;
8709 Newx(buf, 8192, STDCHAR);
8717 const STDCHAR * const bpe = buf + sizeof(buf);
8719 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8720 ; /* keep reading */
8724 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8725 /* Accommodate broken VAXC compiler, which applies U8 cast to
8726 * both args of ?: operator, causing EOF to change into 255
8729 i = (U8)buf[cnt - 1];
8735 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8737 sv_catpvn_nomg(sv, (char *) buf, cnt);
8739 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8741 if (i != EOF && /* joy */
8743 SvCUR(sv) < rslen ||
8744 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8748 * If we're reading from a TTY and we get a short read,
8749 * indicating that the user hit his EOF character, we need
8750 * to notice it now, because if we try to read from the TTY
8751 * again, the EOF condition will disappear.
8753 * The comparison of cnt to sizeof(buf) is an optimization
8754 * that prevents unnecessary calls to feof().
8758 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8762 #ifdef USE_HEAP_INSTEAD_OF_STACK
8767 if (rspara) { /* have to do this both before and after */
8768 while (i != EOF) { /* to make sure file boundaries work right */
8769 i = PerlIO_getc(fp);
8771 PerlIO_ungetc(fp,i);
8777 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8783 Auto-increment of the value in the SV, doing string to numeric conversion
8784 if necessary. Handles 'get' magic and operator overloading.
8790 Perl_sv_inc(pTHX_ SV *const sv)
8799 =for apidoc sv_inc_nomg
8801 Auto-increment of the value in the SV, doing string to numeric conversion
8802 if necessary. Handles operator overloading. Skips handling 'get' magic.
8808 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8815 if (SvTHINKFIRST(sv)) {
8816 if (SvREADONLY(sv)) {
8817 Perl_croak_no_modify();
8821 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8823 i = PTR2IV(SvRV(sv));
8827 else sv_force_normal_flags(sv, 0);
8829 flags = SvFLAGS(sv);
8830 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8831 /* It's (privately or publicly) a float, but not tested as an
8832 integer, so test it to see. */
8834 flags = SvFLAGS(sv);
8836 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8837 /* It's publicly an integer, or privately an integer-not-float */
8838 #ifdef PERL_PRESERVE_IVUV
8842 if (SvUVX(sv) == UV_MAX)
8843 sv_setnv(sv, UV_MAX_P1);
8845 (void)SvIOK_only_UV(sv);
8846 SvUV_set(sv, SvUVX(sv) + 1);
8848 if (SvIVX(sv) == IV_MAX)
8849 sv_setuv(sv, (UV)IV_MAX + 1);
8851 (void)SvIOK_only(sv);
8852 SvIV_set(sv, SvIVX(sv) + 1);
8857 if (flags & SVp_NOK) {
8858 const NV was = SvNVX(sv);
8859 if (LIKELY(!Perl_isinfnan(was)) &&
8860 NV_OVERFLOWS_INTEGERS_AT &&
8861 was >= NV_OVERFLOWS_INTEGERS_AT) {
8862 /* diag_listed_as: Lost precision when %s %f by 1 */
8863 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8864 "Lost precision when incrementing %" NVff " by 1",
8867 (void)SvNOK_only(sv);
8868 SvNV_set(sv, was + 1.0);
8872 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8873 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8874 Perl_croak_no_modify();
8876 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8877 if ((flags & SVTYPEMASK) < SVt_PVIV)
8878 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8879 (void)SvIOK_only(sv);
8884 while (isALPHA(*d)) d++;
8885 while (isDIGIT(*d)) d++;
8886 if (d < SvEND(sv)) {
8887 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8888 #ifdef PERL_PRESERVE_IVUV
8889 /* Got to punt this as an integer if needs be, but we don't issue
8890 warnings. Probably ought to make the sv_iv_please() that does
8891 the conversion if possible, and silently. */
8892 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8893 /* Need to try really hard to see if it's an integer.
8894 9.22337203685478e+18 is an integer.
8895 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8896 so $a="9.22337203685478e+18"; $a+0; $a++
8897 needs to be the same as $a="9.22337203685478e+18"; $a++
8904 /* sv_2iv *should* have made this an NV */
8905 if (flags & SVp_NOK) {
8906 (void)SvNOK_only(sv);
8907 SvNV_set(sv, SvNVX(sv) + 1.0);
8910 /* I don't think we can get here. Maybe I should assert this
8911 And if we do get here I suspect that sv_setnv will croak. NWC
8913 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8914 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8916 #endif /* PERL_PRESERVE_IVUV */
8917 if (!numtype && ckWARN(WARN_NUMERIC))
8918 not_incrementable(sv);
8919 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8923 while (d >= SvPVX_const(sv)) {
8931 /* MKS: The original code here died if letters weren't consecutive.
8932 * at least it didn't have to worry about non-C locales. The
8933 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8934 * arranged in order (although not consecutively) and that only
8935 * [A-Za-z] are accepted by isALPHA in the C locale.
8937 if (isALPHA_FOLD_NE(*d, 'z')) {
8938 do { ++*d; } while (!isALPHA(*d));
8941 *(d--) -= 'z' - 'a';
8946 *(d--) -= 'z' - 'a' + 1;
8950 /* oh,oh, the number grew */
8951 SvGROW(sv, SvCUR(sv) + 2);
8952 SvCUR_set(sv, SvCUR(sv) + 1);
8953 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8964 Auto-decrement of the value in the SV, doing string to numeric conversion
8965 if necessary. Handles 'get' magic and operator overloading.
8971 Perl_sv_dec(pTHX_ SV *const sv)
8980 =for apidoc sv_dec_nomg
8982 Auto-decrement of the value in the SV, doing string to numeric conversion
8983 if necessary. Handles operator overloading. Skips handling 'get' magic.
8989 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8995 if (SvTHINKFIRST(sv)) {
8996 if (SvREADONLY(sv)) {
8997 Perl_croak_no_modify();
9001 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9003 i = PTR2IV(SvRV(sv));
9007 else sv_force_normal_flags(sv, 0);
9009 /* Unlike sv_inc we don't have to worry about string-never-numbers
9010 and keeping them magic. But we mustn't warn on punting */
9011 flags = SvFLAGS(sv);
9012 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9013 /* It's publicly an integer, or privately an integer-not-float */
9014 #ifdef PERL_PRESERVE_IVUV
9018 if (SvUVX(sv) == 0) {
9019 (void)SvIOK_only(sv);
9023 (void)SvIOK_only_UV(sv);
9024 SvUV_set(sv, SvUVX(sv) - 1);
9027 if (SvIVX(sv) == IV_MIN) {
9028 sv_setnv(sv, (NV)IV_MIN);
9032 (void)SvIOK_only(sv);
9033 SvIV_set(sv, SvIVX(sv) - 1);
9038 if (flags & SVp_NOK) {
9041 const NV was = SvNVX(sv);
9042 if (LIKELY(!Perl_isinfnan(was)) &&
9043 NV_OVERFLOWS_INTEGERS_AT &&
9044 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9045 /* diag_listed_as: Lost precision when %s %f by 1 */
9046 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9047 "Lost precision when decrementing %" NVff " by 1",
9050 (void)SvNOK_only(sv);
9051 SvNV_set(sv, was - 1.0);
9056 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9057 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9058 Perl_croak_no_modify();
9060 if (!(flags & SVp_POK)) {
9061 if ((flags & SVTYPEMASK) < SVt_PVIV)
9062 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9064 (void)SvIOK_only(sv);
9067 #ifdef PERL_PRESERVE_IVUV
9069 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9070 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9071 /* Need to try really hard to see if it's an integer.
9072 9.22337203685478e+18 is an integer.
9073 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9074 so $a="9.22337203685478e+18"; $a+0; $a--
9075 needs to be the same as $a="9.22337203685478e+18"; $a--
9082 /* sv_2iv *should* have made this an NV */
9083 if (flags & SVp_NOK) {
9084 (void)SvNOK_only(sv);
9085 SvNV_set(sv, SvNVX(sv) - 1.0);
9088 /* I don't think we can get here. Maybe I should assert this
9089 And if we do get here I suspect that sv_setnv will croak. NWC
9091 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9092 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9095 #endif /* PERL_PRESERVE_IVUV */
9096 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9099 /* this define is used to eliminate a chunk of duplicated but shared logic
9100 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9101 * used anywhere but here - yves
9103 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9105 SSize_t ix = ++PL_tmps_ix; \
9106 if (UNLIKELY(ix >= PL_tmps_max)) \
9107 ix = tmps_grow_p(ix); \
9108 PL_tmps_stack[ix] = (AnSv); \
9112 =for apidoc sv_mortalcopy
9114 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9115 The new SV is marked as mortal. It will be destroyed "soon", either by an
9116 explicit call to C<FREETMPS>, or by an implicit call at places such as
9117 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9122 /* Make a string that will exist for the duration of the expression
9123 * evaluation. Actually, it may have to last longer than that, but
9124 * hopefully we won't free it until it has been assigned to a
9125 * permanent location. */
9128 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9132 if (flags & SV_GMAGIC)
9133 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9135 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9136 PUSH_EXTEND_MORTAL__SV_C(sv);
9142 =for apidoc sv_newmortal
9144 Creates a new null SV which is mortal. The reference count of the SV is
9145 set to 1. It will be destroyed "soon", either by an explicit call to
9146 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9147 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9153 Perl_sv_newmortal(pTHX)
9158 SvFLAGS(sv) = SVs_TEMP;
9159 PUSH_EXTEND_MORTAL__SV_C(sv);
9165 =for apidoc newSVpvn_flags
9167 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9168 characters) into it. The reference count for the
9169 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9170 string. You are responsible for ensuring that the source string is at least
9171 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9172 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9173 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9174 returning. If C<SVf_UTF8> is set, C<s>
9175 is considered to be in UTF-8 and the
9176 C<SVf_UTF8> flag will be set on the new SV.
9177 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9179 #define newSVpvn_utf8(s, len, u) \
9180 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9186 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9190 /* All the flags we don't support must be zero.
9191 And we're new code so I'm going to assert this from the start. */
9192 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9194 sv_setpvn(sv,s,len);
9196 /* This code used to do a sv_2mortal(), however we now unroll the call to
9197 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9198 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9199 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9200 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9201 * means that we eliminate quite a few steps than it looks - Yves
9202 * (explaining patch by gfx) */
9204 SvFLAGS(sv) |= flags;
9206 if(flags & SVs_TEMP){
9207 PUSH_EXTEND_MORTAL__SV_C(sv);
9214 =for apidoc sv_2mortal
9216 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9217 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9218 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9219 string buffer can be "stolen" if this SV is copied. See also
9220 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9226 Perl_sv_2mortal(pTHX_ SV *const sv)
9233 PUSH_EXTEND_MORTAL__SV_C(sv);
9241 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9242 characters) into it. The reference count for the
9243 SV is set to 1. If C<len> is zero, Perl will compute the length using
9244 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9245 C<NUL> characters and has to have a terminating C<NUL> byte).
9247 For efficiency, consider using C<newSVpvn> instead.
9253 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9258 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9263 =for apidoc newSVpvn
9265 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9266 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9267 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9268 are responsible for ensuring that the source buffer is at least
9269 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9276 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9280 sv_setpvn(sv,buffer,len);
9285 =for apidoc newSVhek
9287 Creates a new SV from the hash key structure. It will generate scalars that
9288 point to the shared string table where possible. Returns a new (undefined)
9289 SV if C<hek> is NULL.
9295 Perl_newSVhek(pTHX_ const HEK *const hek)
9304 if (HEK_LEN(hek) == HEf_SVKEY) {
9305 return newSVsv(*(SV**)HEK_KEY(hek));
9307 const int flags = HEK_FLAGS(hek);
9308 if (flags & HVhek_WASUTF8) {
9310 Andreas would like keys he put in as utf8 to come back as utf8
9312 STRLEN utf8_len = HEK_LEN(hek);
9313 SV * const sv = newSV_type(SVt_PV);
9314 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9315 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9316 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9319 } else if (flags & HVhek_UNSHARED) {
9320 /* A hash that isn't using shared hash keys has to have
9321 the flag in every key so that we know not to try to call
9322 share_hek_hek on it. */
9324 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9329 /* This will be overwhelminly the most common case. */
9331 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9332 more efficient than sharepvn(). */
9336 sv_upgrade(sv, SVt_PV);
9337 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9338 SvCUR_set(sv, HEK_LEN(hek));
9350 =for apidoc newSVpvn_share
9352 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9353 table. If the string does not already exist in the table, it is
9354 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9355 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9356 is non-zero, that value is used; otherwise the hash is computed.
9357 The string's hash can later be retrieved from the SV
9358 with the C<SvSHARED_HASH()> macro. The idea here is
9359 that as the string table is used for shared hash keys these strings will have
9360 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9366 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9370 bool is_utf8 = FALSE;
9371 const char *const orig_src = src;
9374 STRLEN tmplen = -len;
9376 /* See the note in hv.c:hv_fetch() --jhi */
9377 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9381 PERL_HASH(hash, src, len);
9383 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9384 changes here, update it there too. */
9385 sv_upgrade(sv, SVt_PV);
9386 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9393 if (src != orig_src)
9399 =for apidoc newSVpv_share
9401 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9408 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9410 return newSVpvn_share(src, strlen(src), hash);
9413 #if defined(PERL_IMPLICIT_CONTEXT)
9415 /* pTHX_ magic can't cope with varargs, so this is a no-context
9416 * version of the main function, (which may itself be aliased to us).
9417 * Don't access this version directly.
9421 Perl_newSVpvf_nocontext(const char *const pat, ...)
9427 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9429 va_start(args, pat);
9430 sv = vnewSVpvf(pat, &args);
9437 =for apidoc newSVpvf
9439 Creates a new SV and initializes it with the string formatted like
9446 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9451 PERL_ARGS_ASSERT_NEWSVPVF;
9453 va_start(args, pat);
9454 sv = vnewSVpvf(pat, &args);
9459 /* backend for newSVpvf() and newSVpvf_nocontext() */
9462 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9466 PERL_ARGS_ASSERT_VNEWSVPVF;
9469 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9476 Creates a new SV and copies a floating point value into it.
9477 The reference count for the SV is set to 1.
9483 Perl_newSVnv(pTHX_ const NV n)
9495 Creates a new SV and copies an integer into it. The reference count for the
9502 Perl_newSViv(pTHX_ const IV i)
9508 /* Inlining ONLY the small relevant subset of sv_setiv here
9509 * for performance. Makes a significant difference. */
9511 /* We're starting from SVt_FIRST, so provided that's
9512 * actual 0, we don't have to unset any SV type flags
9513 * to promote to SVt_IV. */
9514 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9516 SET_SVANY_FOR_BODYLESS_IV(sv);
9517 SvFLAGS(sv) |= SVt_IV;
9529 Creates a new SV and copies an unsigned integer into it.
9530 The reference count for the SV is set to 1.
9536 Perl_newSVuv(pTHX_ const UV u)
9540 /* Inlining ONLY the small relevant subset of sv_setuv here
9541 * for performance. Makes a significant difference. */
9543 /* Using ivs is more efficient than using uvs - see sv_setuv */
9544 if (u <= (UV)IV_MAX) {
9545 return newSViv((IV)u);
9550 /* We're starting from SVt_FIRST, so provided that's
9551 * actual 0, we don't have to unset any SV type flags
9552 * to promote to SVt_IV. */
9553 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9555 SET_SVANY_FOR_BODYLESS_IV(sv);
9556 SvFLAGS(sv) |= SVt_IV;
9558 (void)SvIsUV_on(sv);
9567 =for apidoc newSV_type
9569 Creates a new SV, of the type specified. The reference count for the new SV
9576 Perl_newSV_type(pTHX_ const svtype type)
9581 ASSUME(SvTYPE(sv) == SVt_FIRST);
9582 if(type != SVt_FIRST)
9583 sv_upgrade(sv, type);
9588 =for apidoc newRV_noinc
9590 Creates an RV wrapper for an SV. The reference count for the original
9591 SV is B<not> incremented.
9597 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9601 PERL_ARGS_ASSERT_NEWRV_NOINC;
9605 /* We're starting from SVt_FIRST, so provided that's
9606 * actual 0, we don't have to unset any SV type flags
9607 * to promote to SVt_IV. */
9608 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9610 SET_SVANY_FOR_BODYLESS_IV(sv);
9611 SvFLAGS(sv) |= SVt_IV;
9616 SvRV_set(sv, tmpRef);
9621 /* newRV_inc is the official function name to use now.
9622 * newRV_inc is in fact #defined to newRV in sv.h
9626 Perl_newRV(pTHX_ SV *const sv)
9628 PERL_ARGS_ASSERT_NEWRV;
9630 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9636 Creates a new SV which is an exact duplicate of the original SV.
9643 Perl_newSVsv(pTHX_ SV *const old)
9649 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9650 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9653 /* Do this here, otherwise we leak the new SV if this croaks. */
9656 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9657 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9658 sv_setsv_flags(sv, old, SV_NOSTEAL);
9663 =for apidoc sv_reset
9665 Underlying implementation for the C<reset> Perl function.
9666 Note that the perl-level function is vaguely deprecated.
9672 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9674 PERL_ARGS_ASSERT_SV_RESET;
9676 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9680 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9682 char todo[PERL_UCHAR_MAX+1];
9685 if (!stash || SvTYPE(stash) != SVt_PVHV)
9688 if (!s) { /* reset ?? searches */
9689 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9691 const U32 count = mg->mg_len / sizeof(PMOP**);
9692 PMOP **pmp = (PMOP**) mg->mg_ptr;
9693 PMOP *const *const end = pmp + count;
9697 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9699 (*pmp)->op_pmflags &= ~PMf_USED;
9707 /* reset variables */
9709 if (!HvARRAY(stash))
9712 Zero(todo, 256, char);
9716 I32 i = (unsigned char)*s;
9720 max = (unsigned char)*s++;
9721 for ( ; i <= max; i++) {
9724 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9726 for (entry = HvARRAY(stash)[i];
9728 entry = HeNEXT(entry))
9733 if (!todo[(U8)*HeKEY(entry)])
9735 gv = MUTABLE_GV(HeVAL(entry));
9739 if (sv && !SvREADONLY(sv)) {
9740 SV_CHECK_THINKFIRST_COW_DROP(sv);
9741 if (!isGV(sv)) SvOK_off(sv);
9746 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9757 Using various gambits, try to get an IO from an SV: the IO slot if its a
9758 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9759 named after the PV if we're a string.
9761 'Get' magic is ignored on the C<sv> passed in, but will be called on
9762 C<SvRV(sv)> if C<sv> is an RV.
9768 Perl_sv_2io(pTHX_ SV *const sv)
9773 PERL_ARGS_ASSERT_SV_2IO;
9775 switch (SvTYPE(sv)) {
9777 io = MUTABLE_IO(sv);
9781 if (isGV_with_GP(sv)) {
9782 gv = MUTABLE_GV(sv);
9785 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9786 HEKfARG(GvNAME_HEK(gv)));
9792 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9794 SvGETMAGIC(SvRV(sv));
9795 return sv_2io(SvRV(sv));
9797 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9804 if (SvGMAGICAL(sv)) {
9805 newsv = sv_newmortal();
9806 sv_setsv_nomg(newsv, sv);
9808 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9818 Using various gambits, try to get a CV from an SV; in addition, try if
9819 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9820 The flags in C<lref> are passed to C<gv_fetchsv>.
9826 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9831 PERL_ARGS_ASSERT_SV_2CV;
9838 switch (SvTYPE(sv)) {
9842 return MUTABLE_CV(sv);
9852 sv = amagic_deref_call(sv, to_cv_amg);
9855 if (SvTYPE(sv) == SVt_PVCV) {
9856 cv = MUTABLE_CV(sv);
9861 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9862 gv = MUTABLE_GV(sv);
9864 Perl_croak(aTHX_ "Not a subroutine reference");
9866 else if (isGV_with_GP(sv)) {
9867 gv = MUTABLE_GV(sv);
9870 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9877 /* Some flags to gv_fetchsv mean don't really create the GV */
9878 if (!isGV_with_GP(gv)) {
9883 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9884 /* XXX this is probably not what they think they're getting.
9885 * It has the same effect as "sub name;", i.e. just a forward
9896 Returns true if the SV has a true value by Perl's rules.
9897 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9898 instead use an in-line version.
9904 Perl_sv_true(pTHX_ SV *const sv)
9909 const XPV* const tXpv = (XPV*)SvANY(sv);
9911 (tXpv->xpv_cur > 1 ||
9912 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9919 return SvIVX(sv) != 0;
9922 return SvNVX(sv) != 0.0;
9924 return sv_2bool(sv);
9930 =for apidoc sv_pvn_force
9932 Get a sensible string out of the SV somehow.
9933 A private implementation of the C<SvPV_force> macro for compilers which
9934 can't cope with complex macro expressions. Always use the macro instead.
9936 =for apidoc sv_pvn_force_flags
9938 Get a sensible string out of the SV somehow.
9939 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9940 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9941 implemented in terms of this function.
9942 You normally want to use the various wrapper macros instead: see
9943 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9949 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9951 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9953 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9954 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9955 sv_force_normal_flags(sv, 0);
9965 if (SvTYPE(sv) > SVt_PVLV
9966 || isGV_with_GP(sv))
9967 /* diag_listed_as: Can't coerce %s to %s in %s */
9968 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9970 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9977 if (SvTYPE(sv) < SVt_PV ||
9978 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9981 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9982 SvGROW(sv, len + 1);
9983 Move(s,SvPVX(sv),len,char);
9985 SvPVX(sv)[len] = '\0';
9988 SvPOK_on(sv); /* validate pointer */
9990 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9991 PTR2UV(sv),SvPVX_const(sv)));
9994 (void)SvPOK_only_UTF8(sv);
9995 return SvPVX_mutable(sv);
9999 =for apidoc sv_pvbyten_force
10001 The backend for the C<SvPVbytex_force> macro. Always use the macro
10008 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10010 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10012 sv_pvn_force(sv,lp);
10013 sv_utf8_downgrade(sv,0);
10019 =for apidoc sv_pvutf8n_force
10021 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10028 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10030 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10032 sv_pvn_force(sv,0);
10033 sv_utf8_upgrade_nomg(sv);
10039 =for apidoc sv_reftype
10041 Returns a string describing what the SV is a reference to.
10043 If ob is true and the SV is blessed, the string is the class name,
10044 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10050 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10052 PERL_ARGS_ASSERT_SV_REFTYPE;
10053 if (ob && SvOBJECT(sv)) {
10054 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10057 /* WARNING - There is code, for instance in mg.c, that assumes that
10058 * the only reason that sv_reftype(sv,0) would return a string starting
10059 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10060 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10061 * this routine inside other subs, and it saves time.
10062 * Do not change this assumption without searching for "dodgy type check" in
10065 switch (SvTYPE(sv)) {
10080 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10081 /* tied lvalues should appear to be
10082 * scalars for backwards compatibility */
10083 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10084 ? "SCALAR" : "LVALUE");
10085 case SVt_PVAV: return "ARRAY";
10086 case SVt_PVHV: return "HASH";
10087 case SVt_PVCV: return "CODE";
10088 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10089 ? "GLOB" : "SCALAR");
10090 case SVt_PVFM: return "FORMAT";
10091 case SVt_PVIO: return "IO";
10092 case SVt_INVLIST: return "INVLIST";
10093 case SVt_REGEXP: return "REGEXP";
10094 default: return "UNKNOWN";
10102 Returns a SV describing what the SV passed in is a reference to.
10104 dst can be a SV to be set to the description or NULL, in which case a
10105 mortal SV is returned.
10107 If ob is true and the SV is blessed, the description is the class
10108 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10114 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10116 PERL_ARGS_ASSERT_SV_REF;
10119 dst = sv_newmortal();
10121 if (ob && SvOBJECT(sv)) {
10122 HvNAME_get(SvSTASH(sv))
10123 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10124 : sv_setpvs(dst, "__ANON__");
10127 const char * reftype = sv_reftype(sv, 0);
10128 sv_setpv(dst, reftype);
10134 =for apidoc sv_isobject
10136 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10137 object. If the SV is not an RV, or if the object is not blessed, then this
10144 Perl_sv_isobject(pTHX_ SV *sv)
10160 Returns a boolean indicating whether the SV is blessed into the specified
10161 class. This does not check for subtypes; use C<sv_derived_from> to verify
10162 an inheritance relationship.
10168 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10170 const char *hvname;
10172 PERL_ARGS_ASSERT_SV_ISA;
10182 hvname = HvNAME_get(SvSTASH(sv));
10186 return strEQ(hvname, name);
10190 =for apidoc newSVrv
10192 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10193 RV then it will be upgraded to one. If C<classname> is non-null then the new
10194 SV will be blessed in the specified package. The new SV is returned and its
10195 reference count is 1. The reference count 1 is owned by C<rv>.
10201 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10205 PERL_ARGS_ASSERT_NEWSVRV;
10209 SV_CHECK_THINKFIRST_COW_DROP(rv);
10211 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10212 const U32 refcnt = SvREFCNT(rv);
10216 SvREFCNT(rv) = refcnt;
10218 sv_upgrade(rv, SVt_IV);
10219 } else if (SvROK(rv)) {
10220 SvREFCNT_dec(SvRV(rv));
10222 prepare_SV_for_RV(rv);
10230 HV* const stash = gv_stashpv(classname, GV_ADD);
10231 (void)sv_bless(rv, stash);
10237 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10239 SV * const lv = newSV_type(SVt_PVLV);
10240 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10242 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10243 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10244 LvSTARGOFF(lv) = ix;
10245 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10250 =for apidoc sv_setref_pv
10252 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10253 argument will be upgraded to an RV. That RV will be modified to point to
10254 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10255 into the SV. The C<classname> argument indicates the package for the
10256 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10257 will have a reference count of 1, and the RV will be returned.
10259 Do not use with other Perl types such as HV, AV, SV, CV, because those
10260 objects will become corrupted by the pointer copy process.
10262 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10268 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10270 PERL_ARGS_ASSERT_SV_SETREF_PV;
10273 sv_setsv(rv, &PL_sv_undef);
10277 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10282 =for apidoc sv_setref_iv
10284 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10285 argument will be upgraded to an RV. That RV will be modified to point to
10286 the new SV. The C<classname> argument indicates the package for the
10287 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10288 will have a reference count of 1, and the RV will be returned.
10294 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10296 PERL_ARGS_ASSERT_SV_SETREF_IV;
10298 sv_setiv(newSVrv(rv,classname), iv);
10303 =for apidoc sv_setref_uv
10305 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10306 argument will be upgraded to an RV. That RV will be modified to point to
10307 the new SV. The C<classname> argument indicates the package for the
10308 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10309 will have a reference count of 1, and the RV will be returned.
10315 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10317 PERL_ARGS_ASSERT_SV_SETREF_UV;
10319 sv_setuv(newSVrv(rv,classname), uv);
10324 =for apidoc sv_setref_nv
10326 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10327 argument will be upgraded to an RV. That RV will be modified to point to
10328 the new SV. The C<classname> argument indicates the package for the
10329 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10330 will have a reference count of 1, and the RV will be returned.
10336 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10338 PERL_ARGS_ASSERT_SV_SETREF_NV;
10340 sv_setnv(newSVrv(rv,classname), nv);
10345 =for apidoc sv_setref_pvn
10347 Copies a string into a new SV, optionally blessing the SV. The length of the
10348 string must be specified with C<n>. The C<rv> argument will be upgraded to
10349 an RV. That RV will be modified to point to the new SV. The C<classname>
10350 argument indicates the package for the blessing. Set C<classname> to
10351 C<NULL> to avoid the blessing. The new SV will have a reference count
10352 of 1, and the RV will be returned.
10354 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10360 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10361 const char *const pv, const STRLEN n)
10363 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10365 sv_setpvn(newSVrv(rv,classname), pv, n);
10370 =for apidoc sv_bless
10372 Blesses an SV into a specified package. The SV must be an RV. The package
10373 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10374 of the SV is unaffected.
10380 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10383 HV *oldstash = NULL;
10385 PERL_ARGS_ASSERT_SV_BLESS;
10389 Perl_croak(aTHX_ "Can't bless non-reference value");
10391 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10392 if (SvREADONLY(tmpRef))
10393 Perl_croak_no_modify();
10394 if (SvOBJECT(tmpRef)) {
10395 oldstash = SvSTASH(tmpRef);
10398 SvOBJECT_on(tmpRef);
10399 SvUPGRADE(tmpRef, SVt_PVMG);
10400 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10401 SvREFCNT_dec(oldstash);
10403 if(SvSMAGICAL(tmpRef))
10404 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10412 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10413 * as it is after unglobbing it.
10416 PERL_STATIC_INLINE void
10417 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10421 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10423 PERL_ARGS_ASSERT_SV_UNGLOB;
10425 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10427 if (!(flags & SV_COW_DROP_PV))
10428 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10430 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10432 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10433 && HvNAME_get(stash))
10434 mro_method_changed_in(stash);
10435 gp_free(MUTABLE_GV(sv));
10438 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10439 GvSTASH(sv) = NULL;
10442 if (GvNAME_HEK(sv)) {
10443 unshare_hek(GvNAME_HEK(sv));
10445 isGV_with_GP_off(sv);
10447 if(SvTYPE(sv) == SVt_PVGV) {
10448 /* need to keep SvANY(sv) in the right arena */
10449 xpvmg = new_XPVMG();
10450 StructCopy(SvANY(sv), xpvmg, XPVMG);
10451 del_XPVGV(SvANY(sv));
10454 SvFLAGS(sv) &= ~SVTYPEMASK;
10455 SvFLAGS(sv) |= SVt_PVMG;
10458 /* Intentionally not calling any local SET magic, as this isn't so much a
10459 set operation as merely an internal storage change. */
10460 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10461 else sv_setsv_flags(sv, temp, 0);
10463 if ((const GV *)sv == PL_last_in_gv)
10464 PL_last_in_gv = NULL;
10465 else if ((const GV *)sv == PL_statgv)
10470 =for apidoc sv_unref_flags
10472 Unsets the RV status of the SV, and decrements the reference count of
10473 whatever was being referenced by the RV. This can almost be thought of
10474 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10475 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10476 (otherwise the decrementing is conditional on the reference count being
10477 different from one or the reference being a readonly SV).
10478 See C<L</SvROK_off>>.
10484 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10486 SV* const target = SvRV(ref);
10488 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10490 if (SvWEAKREF(ref)) {
10491 sv_del_backref(target, ref);
10492 SvWEAKREF_off(ref);
10493 SvRV_set(ref, NULL);
10496 SvRV_set(ref, NULL);
10498 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10499 assigned to as BEGIN {$a = \"Foo"} will fail. */
10500 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10501 SvREFCNT_dec_NN(target);
10502 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10503 sv_2mortal(target); /* Schedule for freeing later */
10507 =for apidoc sv_untaint
10509 Untaint an SV. Use C<SvTAINTED_off> instead.
10515 Perl_sv_untaint(pTHX_ SV *const sv)
10517 PERL_ARGS_ASSERT_SV_UNTAINT;
10518 PERL_UNUSED_CONTEXT;
10520 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10521 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10528 =for apidoc sv_tainted
10530 Test an SV for taintedness. Use C<SvTAINTED> instead.
10536 Perl_sv_tainted(pTHX_ SV *const sv)
10538 PERL_ARGS_ASSERT_SV_TAINTED;
10539 PERL_UNUSED_CONTEXT;
10541 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10542 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10543 if (mg && (mg->mg_len & 1) )
10549 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10550 private to this file */
10553 =for apidoc sv_setpviv
10555 Copies an integer into the given SV, also updating its string value.
10556 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10562 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10564 char buf[TYPE_CHARS(UV)];
10566 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10568 PERL_ARGS_ASSERT_SV_SETPVIV;
10570 sv_setpvn(sv, ptr, ebuf - ptr);
10574 =for apidoc sv_setpviv_mg
10576 Like C<sv_setpviv>, but also handles 'set' magic.
10582 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10584 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10586 sv_setpviv(sv, iv);
10590 #endif /* NO_MATHOMS */
10592 #if defined(PERL_IMPLICIT_CONTEXT)
10594 /* pTHX_ magic can't cope with varargs, so this is a no-context
10595 * version of the main function, (which may itself be aliased to us).
10596 * Don't access this version directly.
10600 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10605 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10607 va_start(args, pat);
10608 sv_vsetpvf(sv, pat, &args);
10612 /* pTHX_ magic can't cope with varargs, so this is a no-context
10613 * version of the main function, (which may itself be aliased to us).
10614 * Don't access this version directly.
10618 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10623 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10625 va_start(args, pat);
10626 sv_vsetpvf_mg(sv, pat, &args);
10632 =for apidoc sv_setpvf
10634 Works like C<sv_catpvf> but copies the text into the SV instead of
10635 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10641 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10645 PERL_ARGS_ASSERT_SV_SETPVF;
10647 va_start(args, pat);
10648 sv_vsetpvf(sv, pat, &args);
10653 =for apidoc sv_vsetpvf
10655 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10656 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10658 Usually used via its frontend C<sv_setpvf>.
10664 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10666 PERL_ARGS_ASSERT_SV_VSETPVF;
10668 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10672 =for apidoc sv_setpvf_mg
10674 Like C<sv_setpvf>, but also handles 'set' magic.
10680 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10684 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10686 va_start(args, pat);
10687 sv_vsetpvf_mg(sv, pat, &args);
10692 =for apidoc sv_vsetpvf_mg
10694 Like C<sv_vsetpvf>, but also handles 'set' magic.
10696 Usually used via its frontend C<sv_setpvf_mg>.
10702 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10704 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10706 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10710 #if defined(PERL_IMPLICIT_CONTEXT)
10712 /* pTHX_ magic can't cope with varargs, so this is a no-context
10713 * version of the main function, (which may itself be aliased to us).
10714 * Don't access this version directly.
10718 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10723 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10725 va_start(args, pat);
10726 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10730 /* pTHX_ magic can't cope with varargs, so this is a no-context
10731 * version of the main function, (which may itself be aliased to us).
10732 * Don't access this version directly.
10736 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10741 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10743 va_start(args, pat);
10744 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10751 =for apidoc sv_catpvf
10753 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10754 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10755 variable argument list, argument reordering is not supported.
10756 If the appended data contains "wide" characters
10757 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10758 and characters >255 formatted with C<%c>), the original SV might get
10759 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10760 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10761 valid UTF-8; if the original SV was bytes, the pattern should be too.
10766 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10770 PERL_ARGS_ASSERT_SV_CATPVF;
10772 va_start(args, pat);
10773 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10778 =for apidoc sv_vcatpvf
10780 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10781 variable argument list, and appends the formatted output
10782 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10784 Usually used via its frontend C<sv_catpvf>.
10790 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10792 PERL_ARGS_ASSERT_SV_VCATPVF;
10794 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10798 =for apidoc sv_catpvf_mg
10800 Like C<sv_catpvf>, but also handles 'set' magic.
10806 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10810 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10812 va_start(args, pat);
10813 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10819 =for apidoc sv_vcatpvf_mg
10821 Like C<sv_vcatpvf>, but also handles 'set' magic.
10823 Usually used via its frontend C<sv_catpvf_mg>.
10829 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10831 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10833 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10838 =for apidoc sv_vsetpvfn
10840 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10843 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10849 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10850 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10852 PERL_ARGS_ASSERT_SV_VSETPVFN;
10855 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10860 * Warn of missing argument to sprintf. The value used in place of such
10861 * arguments should be &PL_sv_no; an undefined value would yield
10862 * inappropriate "use of uninit" warnings [perl #71000].
10865 S_warn_vcatpvfn_missing_argument(pTHX) {
10866 if (ckWARN(WARN_MISSING)) {
10867 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10868 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10874 S_expect_number(pTHX_ char **const pattern)
10878 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10880 switch (**pattern) {
10881 case '1': case '2': case '3':
10882 case '4': case '5': case '6':
10883 case '7': case '8': case '9':
10884 var = *(*pattern)++ - '0';
10885 while (isDIGIT(**pattern)) {
10886 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10888 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10896 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10898 const int neg = nv < 0;
10901 PERL_ARGS_ASSERT_F0CONVERT;
10903 if (UNLIKELY(Perl_isinfnan(nv))) {
10904 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10914 if (uv & 1 && uv == nv)
10915 uv--; /* Round to even */
10917 const unsigned dig = uv % 10;
10919 } while (uv /= 10);
10930 =for apidoc sv_vcatpvfn
10932 =for apidoc sv_vcatpvfn_flags
10934 Processes its arguments like C<vsprintf> and appends the formatted output
10935 to an SV. Uses an array of SVs if the C-style variable argument list is
10936 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10937 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10938 C<va_list> argument list with a format string that uses argument reordering
10939 will yield an exception.
10941 When running with taint checks enabled, indicates via
10942 C<maybe_tainted> if results are untrustworthy (often due to the use of
10945 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10947 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10952 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10953 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10954 vec_utf8 = DO_UTF8(vecsv);
10956 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10959 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10960 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10962 PERL_ARGS_ASSERT_SV_VCATPVFN;
10964 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10967 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10968 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10969 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10970 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10971 * after the first 1023 zero bits.
10973 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10974 * of dynamically growing buffer might be better, start at just 16 bytes
10975 * (for example) and grow only when necessary. Or maybe just by looking
10976 * at the exponents of the two doubles? */
10977 # define DOUBLEDOUBLE_MAXBITS 2098
10980 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10981 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10982 * per xdigit. For the double-double case, this can be rather many.
10983 * The non-double-double-long-double overshoots since all bits of NV
10984 * are not mantissa bits, there are also exponent bits. */
10985 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10986 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10988 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10991 /* If we do not have a known long double format, (including not using
10992 * long doubles, or long doubles being equal to doubles) then we will
10993 * fall back to the ldexp/frexp route, with which we can retrieve at
10994 * most as many bits as our widest unsigned integer type is. We try
10995 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10997 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10998 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11000 #if defined(HAS_QUAD) && defined(Uquad_t)
11001 # define MANTISSATYPE Uquad_t
11002 # define MANTISSASIZE 8
11004 # define MANTISSATYPE UV
11005 # define MANTISSASIZE UVSIZE
11008 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11009 # define HEXTRACT_LITTLE_ENDIAN
11010 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11011 # define HEXTRACT_BIG_ENDIAN
11013 # define HEXTRACT_MIX_ENDIAN
11016 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
11017 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11018 * are being extracted from (either directly from the long double in-memory
11019 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11020 * is used to update the exponent. The subnormal is set to true
11021 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11022 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11024 * The tricky part is that S_hextract() needs to be called twice:
11025 * the first time with vend as NULL, and the second time with vend as
11026 * the pointer returned by the first call. What happens is that on
11027 * the first round the output size is computed, and the intended
11028 * extraction sanity checked. On the second round the actual output
11029 * (the extraction of the hexadecimal values) takes place.
11030 * Sanity failures cause fatal failures during both rounds. */
11032 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11033 U8* vhex, U8* vend)
11037 int ixmin = 0, ixmax = 0;
11039 /* XXX Inf/NaN are not handled here, since it is
11040 * assumed they are to be output as "Inf" and "NaN". */
11042 /* These macros are just to reduce typos, they have multiple
11043 * repetitions below, but usually only one (or sometimes two)
11044 * of them is really being used. */
11045 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11046 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11047 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11048 #define HEXTRACT_OUTPUT(ix) \
11050 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11052 #define HEXTRACT_COUNT(ix, c) \
11054 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11056 #define HEXTRACT_BYTE(ix) \
11058 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11060 #define HEXTRACT_LO_NYBBLE(ix) \
11062 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11064 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11065 * to make it look less odd when the top bits of a NV
11066 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11067 * order bits can be in the "low nybble" of a byte. */
11068 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11069 #define HEXTRACT_BYTES_LE(a, b) \
11070 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11071 #define HEXTRACT_BYTES_BE(a, b) \
11072 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11073 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11074 #define HEXTRACT_IMPLICIT_BIT(nv) \
11076 if (!*subnormal) { \
11077 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11081 /* Most formats do. Those which don't should undef this.
11083 * But also note that IEEE 754 subnormals do not have it, or,
11084 * expressed alternatively, their implicit bit is zero. */
11085 #define HEXTRACT_HAS_IMPLICIT_BIT
11087 /* Many formats do. Those which don't should undef this. */
11088 #define HEXTRACT_HAS_TOP_NYBBLE
11090 /* HEXTRACTSIZE is the maximum number of xdigits. */
11091 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11092 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11094 # define HEXTRACTSIZE 2 * NVSIZE
11097 const U8* vmaxend = vhex + HEXTRACTSIZE;
11098 PERL_UNUSED_VAR(ix); /* might happen */
11099 (void)Perl_frexp(PERL_ABS(nv), exponent);
11100 *subnormal = FALSE;
11101 if (vend && (vend <= vhex || vend > vmaxend)) {
11102 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11103 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11106 /* First check if using long doubles. */
11107 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11108 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11109 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11110 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11111 /* The bytes 13..0 are the mantissa/fraction,
11112 * the 15,14 are the sign+exponent. */
11113 const U8* nvp = (const U8*)(&nv);
11114 HEXTRACT_GET_SUBNORMAL(nv);
11115 HEXTRACT_IMPLICIT_BIT(nv);
11116 # undef HEXTRACT_HAS_TOP_NYBBLE
11117 HEXTRACT_BYTES_LE(13, 0);
11118 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11119 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11120 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11121 /* The bytes 2..15 are the mantissa/fraction,
11122 * the 0,1 are the sign+exponent. */
11123 const U8* nvp = (const U8*)(&nv);
11124 HEXTRACT_GET_SUBNORMAL(nv);
11125 HEXTRACT_IMPLICIT_BIT(nv);
11126 # undef HEXTRACT_HAS_TOP_NYBBLE
11127 HEXTRACT_BYTES_BE(2, 15);
11128 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11129 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11130 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11131 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11132 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11133 /* The bytes 0..1 are the sign+exponent,
11134 * the bytes 2..9 are the mantissa/fraction. */
11135 const U8* nvp = (const U8*)(&nv);
11136 # undef HEXTRACT_HAS_IMPLICIT_BIT
11137 # undef HEXTRACT_HAS_TOP_NYBBLE
11138 HEXTRACT_GET_SUBNORMAL(nv);
11139 HEXTRACT_BYTES_LE(7, 0);
11140 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11141 /* Does this format ever happen? (Wikipedia says the Motorola
11142 * 6888x math coprocessors used format _like_ this but padded
11143 * to 96 bits with 16 unused bits between the exponent and the
11145 const U8* nvp = (const U8*)(&nv);
11146 # undef HEXTRACT_HAS_IMPLICIT_BIT
11147 # undef HEXTRACT_HAS_TOP_NYBBLE
11148 HEXTRACT_GET_SUBNORMAL(nv);
11149 HEXTRACT_BYTES_BE(0, 7);
11151 # define HEXTRACT_FALLBACK
11152 /* Double-double format: two doubles next to each other.
11153 * The first double is the high-order one, exactly like
11154 * it would be for a "lone" double. The second double
11155 * is shifted down using the exponent so that that there
11156 * are no common bits. The tricky part is that the value
11157 * of the double-double is the SUM of the two doubles and
11158 * the second one can be also NEGATIVE.
11160 * Because of this tricky construction the bytewise extraction we
11161 * use for the other long double formats doesn't work, we must
11162 * extract the values bit by bit.
11164 * The little-endian double-double is used .. somewhere?
11166 * The big endian double-double is used in e.g. PPC/Power (AIX)
11169 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11170 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11171 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11174 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11175 /* Using normal doubles, not long doubles.
11177 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11178 * bytes, since we might need to handle printf precision, and
11179 * also need to insert the radix. */
11181 # ifdef HEXTRACT_LITTLE_ENDIAN
11182 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11183 const U8* nvp = (const U8*)(&nv);
11184 HEXTRACT_GET_SUBNORMAL(nv);
11185 HEXTRACT_IMPLICIT_BIT(nv);
11186 HEXTRACT_TOP_NYBBLE(6);
11187 HEXTRACT_BYTES_LE(5, 0);
11188 # elif defined(HEXTRACT_BIG_ENDIAN)
11189 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11190 const U8* nvp = (const U8*)(&nv);
11191 HEXTRACT_GET_SUBNORMAL(nv);
11192 HEXTRACT_IMPLICIT_BIT(nv);
11193 HEXTRACT_TOP_NYBBLE(1);
11194 HEXTRACT_BYTES_BE(2, 7);
11195 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11196 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11197 const U8* nvp = (const U8*)(&nv);
11198 HEXTRACT_GET_SUBNORMAL(nv);
11199 HEXTRACT_IMPLICIT_BIT(nv);
11200 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11201 HEXTRACT_BYTE(1); /* 5 */
11202 HEXTRACT_BYTE(0); /* 4 */
11203 HEXTRACT_BYTE(7); /* 3 */
11204 HEXTRACT_BYTE(6); /* 2 */
11205 HEXTRACT_BYTE(5); /* 1 */
11206 HEXTRACT_BYTE(4); /* 0 */
11207 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11208 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11209 const U8* nvp = (const U8*)(&nv);
11210 HEXTRACT_GET_SUBNORMAL(nv);
11211 HEXTRACT_IMPLICIT_BIT(nv);
11212 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11213 HEXTRACT_BYTE(6); /* 5 */
11214 HEXTRACT_BYTE(7); /* 4 */
11215 HEXTRACT_BYTE(0); /* 3 */
11216 HEXTRACT_BYTE(1); /* 2 */
11217 HEXTRACT_BYTE(2); /* 1 */
11218 HEXTRACT_BYTE(3); /* 0 */
11220 # define HEXTRACT_FALLBACK
11223 # define HEXTRACT_FALLBACK
11225 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11226 # ifdef HEXTRACT_FALLBACK
11227 HEXTRACT_GET_SUBNORMAL(nv);
11228 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11229 /* The fallback is used for the double-double format, and
11230 * for unknown long double formats, and for unknown double
11231 * formats, or in general unknown NV formats. */
11232 if (nv == (NV)0.0) {
11240 NV d = nv < 0 ? -nv : nv;
11242 U8 ha = 0x0; /* hexvalue accumulator */
11243 U8 hd = 0x8; /* hexvalue digit */
11245 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11246 * this is essentially manual frexp(). Multiplying by 0.5 and
11247 * doubling should be lossless in binary floating point. */
11257 while (d >= e + e) {
11261 /* Now e <= d < 2*e */
11263 /* First extract the leading hexdigit (the implicit bit). */
11279 /* Then extract the remaining hexdigits. */
11280 while (d > (NV)0.0) {
11286 /* Output or count in groups of four bits,
11287 * that is, when the hexdigit is down to one. */
11292 /* Reset the hexvalue. */
11301 /* Flush possible pending hexvalue. */
11311 /* Croak for various reasons: if the output pointer escaped the
11312 * output buffer, if the extraction index escaped the extraction
11313 * buffer, or if the ending output pointer didn't match the
11314 * previously computed value. */
11315 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11316 /* For double-double the ixmin and ixmax stay at zero,
11317 * which is convenient since the HEXTRACTSIZE is tricky
11318 * for double-double. */
11319 ixmin < 0 || ixmax >= NVSIZE ||
11320 (vend && v != vend)) {
11321 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11322 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11327 /* Helper for sv_vcatpvfn_flags(). */
11328 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11333 (var) = &PL_sv_no; /* [perl #71000] */ \
11334 arg_missing = TRUE; \
11339 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11340 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11345 const char *patend;
11348 static const char nullstr[] = "(null)";
11350 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11351 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11353 /* Times 4: a decimal digit takes more than 3 binary digits.
11354 * NV_DIG: mantissa takes than many decimal digits.
11355 * Plus 32: Playing safe. */
11356 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11357 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11358 bool hexfp = FALSE; /* hexadecimal floating point? */
11360 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11362 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11363 PERL_UNUSED_ARG(maybe_tainted);
11365 if (flags & SV_GMAGIC)
11368 /* no matter what, this is a string now */
11369 (void)SvPV_force_nomg(sv, origlen);
11371 /* special-case "", "%s", and "%-p" (SVf - see below) */
11373 if (svmax && ckWARN(WARN_REDUNDANT))
11374 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11375 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11378 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11379 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11380 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11381 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11384 const char * const s = va_arg(*args, char*);
11385 sv_catpv_nomg(sv, s ? s : nullstr);
11387 else if (svix < svmax) {
11388 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11389 SvGETMAGIC(*svargs);
11390 sv_catsv_nomg(sv, *svargs);
11393 S_warn_vcatpvfn_missing_argument(aTHX);
11396 if (args && patlen == 3 && pat[0] == '%' &&
11397 pat[1] == '-' && pat[2] == 'p') {
11398 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11399 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11400 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11401 argsv = MUTABLE_SV(va_arg(*args, void*));
11402 sv_catsv_nomg(sv, argsv);
11406 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11407 /* special-case "%.<number>[gf]" */
11408 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11409 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11410 unsigned digits = 0;
11414 while (*pp >= '0' && *pp <= '9')
11415 digits = 10 * digits + (*pp++ - '0');
11417 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11418 format the first argument and WARN_REDUNDANT if svmax > 1?
11419 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11420 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11421 const NV nv = SvNV(*svargs);
11422 if (LIKELY(!Perl_isinfnan(nv))) {
11424 /* Add check for digits != 0 because it seems that some
11425 gconverts are buggy in this case, and we don't yet have
11426 a Configure test for this. */
11427 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11428 /* 0, point, slack */
11429 STORE_LC_NUMERIC_SET_TO_NEEDED();
11430 SNPRINTF_G(nv, ebuf, size, digits);
11431 sv_catpv_nomg(sv, ebuf);
11432 if (*ebuf) /* May return an empty string for digits==0 */
11435 } else if (!digits) {
11438 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11439 sv_catpvn_nomg(sv, p, l);
11446 #endif /* !USE_LONG_DOUBLE */
11448 if (!args && svix < svmax && DO_UTF8(*svargs))
11451 patend = (char*)pat + patlen;
11452 for (p = (char*)pat; p < patend; p = q) {
11455 bool vectorize = FALSE;
11456 bool vectorarg = FALSE;
11457 bool vec_utf8 = FALSE;
11463 bool has_precis = FALSE;
11465 const I32 osvix = svix;
11466 bool is_utf8 = FALSE; /* is this item utf8? */
11467 bool used_explicit_ix = FALSE;
11468 bool arg_missing = FALSE;
11469 #ifdef HAS_LDBL_SPRINTF_BUG
11470 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11471 with sfio - Allen <allens@cpan.org> */
11472 bool fix_ldbl_sprintf_bug = FALSE;
11476 U8 utf8buf[UTF8_MAXBYTES+1];
11477 STRLEN esignlen = 0;
11479 const char *eptr = NULL;
11480 const char *fmtstart;
11483 const U8 *vecstr = NULL;
11490 /* We need a long double target in case HAS_LONG_DOUBLE,
11491 * even without USE_LONG_DOUBLE, so that we can printf with
11492 * long double formats, even without NV being long double.
11493 * But we call the target 'fv' instead of 'nv', since most of
11494 * the time it is not (most compilers these days recognize
11495 * "long double", even if only as a synonym for "double").
11497 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11498 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11500 # ifdef Perl_isfinitel
11501 # define FV_ISFINITE(x) Perl_isfinitel(x)
11503 # define FV_GF PERL_PRIgldbl
11504 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11505 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11506 # define NV_TO_FV(nv,fv) STMT_START { \
11508 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11511 # define NV_TO_FV(nv,fv) (fv)=(nv)
11515 # define FV_GF NVgf
11516 # define NV_TO_FV(nv,fv) (fv)=(nv)
11518 #ifndef FV_ISFINITE
11519 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11525 const char *dotstr = ".";
11526 STRLEN dotstrlen = 1;
11527 I32 efix = 0; /* explicit format parameter index */
11528 I32 ewix = 0; /* explicit width index */
11529 I32 epix = 0; /* explicit precision index */
11530 I32 evix = 0; /* explicit vector index */
11531 bool asterisk = FALSE;
11532 bool infnan = FALSE;
11534 /* echo everything up to the next format specification */
11535 for (q = p; q < patend && *q != '%'; ++q) ;
11537 if (has_utf8 && !pat_utf8)
11538 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11540 sv_catpvn_nomg(sv, p, q - p);
11549 We allow format specification elements in this order:
11550 \d+\$ explicit format parameter index
11552 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11553 0 flag (as above): repeated to allow "v02"
11554 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11555 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11557 [%bcdefginopsuxDFOUX] format (mandatory)
11562 As of perl5.9.3, printf format checking is on by default.
11563 Internally, perl uses %p formats to provide an escape to
11564 some extended formatting. This block deals with those
11565 extensions: if it does not match, (char*)q is reset and
11566 the normal format processing code is used.
11568 Currently defined extensions are:
11569 %p include pointer address (standard)
11570 %-p (SVf) include an SV (previously %_)
11571 %-<num>p include an SV with precision <num>
11573 %3p include a HEK with precision of 256
11574 %4p char* preceded by utf8 flag and length
11575 %<num>p (where num is 1 or > 4) reserved for future
11578 Robin Barker 2005-07-14 (but modified since)
11580 %1p (VDf) removed. RMB 2007-10-19
11587 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11588 /* The argument has already gone through cBOOL, so the cast
11590 is_utf8 = (bool)va_arg(*args, int);
11591 elen = va_arg(*args, UV);
11592 /* if utf8 length is larger than 0x7ffff..., then it might
11593 * have been a signed value that wrapped */
11594 if (elen > ((~(STRLEN)0) >> 1)) {
11595 assert(0); /* in DEBUGGING build we want to crash */
11596 elen= 0; /* otherwise we want to treat this as an empty string */
11598 eptr = va_arg(*args, char *);
11599 q += sizeof(UTF8f)-1;
11602 n = expect_number(&q);
11604 if (sv) { /* SVf */
11609 argsv = MUTABLE_SV(va_arg(*args, void*));
11610 eptr = SvPV_const(argsv, elen);
11611 if (DO_UTF8(argsv))
11615 else if (n==2 || n==3) { /* HEKf */
11616 HEK * const hek = va_arg(*args, HEK *);
11617 eptr = HEK_KEY(hek);
11618 elen = HEK_LEN(hek);
11619 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11620 if (n==3) precis = 256, has_precis = TRUE;
11624 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11625 "internal %%<num>p might conflict with future printf extensions");
11631 if ( (width = expect_number(&q)) ) {
11634 Perl_croak_nocontext(
11635 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11638 used_explicit_ix = TRUE;
11650 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11679 if ( (ewix = expect_number(&q)) ) {
11682 Perl_croak_nocontext(
11683 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11684 used_explicit_ix = TRUE;
11694 if ((vectorarg = asterisk)) {
11707 width = expect_number(&q);
11710 if (vectorize && vectorarg) {
11711 /* vectorizing, but not with the default "." */
11713 vecsv = va_arg(*args, SV*);
11715 FETCH_VCATPVFN_ARGUMENT(
11716 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11718 FETCH_VCATPVFN_ARGUMENT(
11719 vecsv, svix < svmax, svargs[svix++]);
11721 dotstr = SvPV_const(vecsv, dotstrlen);
11722 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11723 bad with tied or overloaded values that return UTF8. */
11724 if (DO_UTF8(vecsv))
11726 else if (has_utf8) {
11727 vecsv = sv_mortalcopy(vecsv);
11728 sv_utf8_upgrade(vecsv);
11729 dotstr = SvPV_const(vecsv, dotstrlen);
11736 i = va_arg(*args, int);
11738 i = (ewix ? ewix <= svmax : svix < svmax) ?
11739 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11741 width = (i < 0) ? -i : i;
11751 if ( (epix = expect_number(&q)) ) {
11754 Perl_croak_nocontext(
11755 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11756 used_explicit_ix = TRUE;
11761 i = va_arg(*args, int);
11765 FETCH_VCATPVFN_ARGUMENT(
11766 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11768 FETCH_VCATPVFN_ARGUMENT(
11769 precsv, svix < svmax, svargs[svix++]);
11770 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11773 has_precis = !(i < 0);
11777 while (isDIGIT(*q))
11778 precis = precis * 10 + (*q++ - '0');
11787 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11788 vecsv = svargs[efix ? efix-1 : svix++];
11789 vecstr = (U8*)SvPV_const(vecsv,veclen);
11790 vec_utf8 = DO_UTF8(vecsv);
11792 /* if this is a version object, we need to convert
11793 * back into v-string notation and then let the
11794 * vectorize happen normally
11796 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11797 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11798 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11799 "vector argument not supported with alpha versions");
11802 vecsv = sv_newmortal();
11803 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11805 vecstr = (U8*)SvPV_const(vecsv, veclen);
11806 vec_utf8 = DO_UTF8(vecsv);
11820 case 'I': /* Ix, I32x, and I64x */
11821 # ifdef USE_64_BIT_INT
11822 if (q[1] == '6' && q[2] == '4') {
11828 if (q[1] == '3' && q[2] == '2') {
11832 # ifdef USE_64_BIT_INT
11838 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11839 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11842 # ifdef USE_QUADMATH
11855 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11856 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11857 if (*q == 'l') { /* lld, llf */
11866 if (*++q == 'h') { /* hhd, hhu */
11895 if (!vectorize && !args) {
11897 const I32 i = efix-1;
11898 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11900 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11905 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11906 /* XXX va_arg(*args) case? need peek, use va_copy? */
11908 if (UNLIKELY(SvAMAGIC(argsv)))
11909 argsv = sv_2num(argsv);
11910 infnan = UNLIKELY(isinfnansv(argsv));
11913 switch (c = *q++) {
11921 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11922 /* no va_arg() case */
11923 SvNV_nomg(argsv), (int)c);
11924 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11926 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11928 eptr = (char*)utf8buf;
11929 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11943 eptr = va_arg(*args, char*);
11945 elen = strlen(eptr);
11947 eptr = (char *)nullstr;
11948 elen = sizeof nullstr - 1;
11952 eptr = SvPV_const(argsv, elen);
11953 if (DO_UTF8(argsv)) {
11954 STRLEN old_precis = precis;
11955 if (has_precis && precis < elen) {
11956 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11957 STRLEN p = precis > ulen ? ulen : precis;
11958 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11959 /* sticks at end */
11961 if (width) { /* fudge width (can't fudge elen) */
11962 if (has_precis && precis < elen)
11963 width += precis - old_precis;
11966 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11973 if (has_precis && precis < elen)
11981 goto floating_point;
11983 if (alt || vectorize)
11985 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11999 goto floating_point;
12004 goto donevalidconversion;
12006 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12015 esignbuf[esignlen++] = plus;
12019 case 'c': iv = (char)va_arg(*args, int); break;
12020 case 'h': iv = (short)va_arg(*args, int); break;
12021 case 'l': iv = va_arg(*args, long); break;
12022 case 'V': iv = va_arg(*args, IV); break;
12023 case 'z': iv = va_arg(*args, SSize_t); break;
12024 #ifdef HAS_PTRDIFF_T
12025 case 't': iv = va_arg(*args, ptrdiff_t); break;
12027 default: iv = va_arg(*args, int); break;
12029 case 'j': iv = va_arg(*args, intmax_t); break;
12033 iv = va_arg(*args, Quad_t); break;
12040 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12042 case 'c': iv = (char)tiv; break;
12043 case 'h': iv = (short)tiv; break;
12044 case 'l': iv = (long)tiv; break;
12046 default: iv = tiv; break;
12049 iv = (Quad_t)tiv; break;
12055 if ( !vectorize ) /* we already set uv above */
12060 esignbuf[esignlen++] = plus;
12063 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12064 esignbuf[esignlen++] = '-';
12103 goto floating_point;
12109 goto donevalidconversion;
12111 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12122 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12123 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12124 case 'l': uv = va_arg(*args, unsigned long); break;
12125 case 'V': uv = va_arg(*args, UV); break;
12126 case 'z': uv = va_arg(*args, Size_t); break;
12127 #ifdef HAS_PTRDIFF_T
12128 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12131 case 'j': uv = va_arg(*args, uintmax_t); break;
12133 default: uv = va_arg(*args, unsigned); break;
12136 uv = va_arg(*args, Uquad_t); break;
12143 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12145 case 'c': uv = (unsigned char)tuv; break;
12146 case 'h': uv = (unsigned short)tuv; break;
12147 case 'l': uv = (unsigned long)tuv; break;
12149 default: uv = tuv; break;
12152 uv = (Uquad_t)tuv; break;
12161 char *ptr = ebuf + sizeof ebuf;
12162 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12168 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12172 } while (uv >>= 4);
12174 esignbuf[esignlen++] = '0';
12175 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12181 *--ptr = '0' + dig;
12182 } while (uv >>= 3);
12183 if (alt && *ptr != '0')
12189 *--ptr = '0' + dig;
12190 } while (uv >>= 1);
12192 esignbuf[esignlen++] = '0';
12193 esignbuf[esignlen++] = c;
12196 default: /* it had better be ten or less */
12199 *--ptr = '0' + dig;
12200 } while (uv /= base);
12203 elen = (ebuf + sizeof ebuf) - ptr;
12207 zeros = precis - elen;
12208 else if (precis == 0 && elen == 1 && *eptr == '0'
12209 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12212 /* a precision nullifies the 0 flag. */
12219 /* FLOATING POINT */
12224 c = 'f'; /* maybe %F isn't supported here */
12226 case 'e': case 'E':
12228 case 'g': case 'G':
12229 case 'a': case 'A':
12233 /* This is evil, but floating point is even more evil */
12235 /* for SV-style calling, we can only get NV
12236 for C-style calling, we assume %f is double;
12237 for simplicity we allow any of %Lf, %llf, %qf for long double
12241 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12245 /* [perl #20339] - we should accept and ignore %lf rather than die */
12249 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12250 intsize = args ? 0 : 'q';
12254 #if defined(HAS_LONG_DOUBLE)
12267 /* Now we need (long double) if intsize == 'q', else (double). */
12269 /* Note: do not pull NVs off the va_list with va_arg()
12270 * (pull doubles instead) because if you have a build
12271 * with long doubles, you would always be pulling long
12272 * doubles, which would badly break anyone using only
12273 * doubles (i.e. the majority of builds). In other
12274 * words, you cannot mix doubles and long doubles.
12275 * The only case where you can pull off long doubles
12276 * is when the format specifier explicitly asks so with
12278 #ifdef USE_QUADMATH
12279 fv = intsize == 'q' ?
12280 va_arg(*args, NV) : va_arg(*args, double);
12282 #elif LONG_DOUBLESIZE > DOUBLESIZE
12283 if (intsize == 'q') {
12284 fv = va_arg(*args, long double);
12287 nv = va_arg(*args, double);
12291 nv = va_arg(*args, double);
12297 if (!infnan) SvGETMAGIC(argsv);
12298 nv = SvNV_nomg(argsv);
12303 /* frexp() (or frexpl) has some unspecified behaviour for
12304 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12305 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12307 (void)Perl_frexp((NV)fv, &i);
12308 if (i == PERL_INT_MIN)
12309 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12310 /* Do not set hexfp earlier since we want to printf
12311 * Inf/NaN for Inf/NaN, not their hexfp. */
12312 hexfp = isALPHA_FOLD_EQ(c, 'a');
12313 if (UNLIKELY(hexfp)) {
12314 /* This seriously overshoots in most cases, but
12315 * better the undershooting. Firstly, all bytes
12316 * of the NV are not mantissa, some of them are
12317 * exponent. Secondly, for the reasonably common
12318 * long doubles case, the "80-bit extended", two
12319 * or six bytes of the NV are unused. */
12321 (fv < 0) ? 1 : 0 + /* possible unary minus */
12323 1 + /* the very unlikely carry */
12326 2 * NVSIZE + /* 2 hexdigits for each byte */
12328 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12330 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12331 /* However, for the "double double", we need more.
12332 * Since each double has their own exponent, the
12333 * doubles may float (haha) rather far from each
12334 * other, and the number of required bits is much
12335 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12336 * See the definition of DOUBLEDOUBLE_MAXBITS.
12338 * Need 2 hexdigits for each byte. */
12339 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12340 /* the size for the exponent already added */
12342 #ifdef USE_LOCALE_NUMERIC
12343 STORE_LC_NUMERIC_SET_TO_NEEDED();
12344 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12345 need += SvLEN(PL_numeric_radix_sv);
12346 RESTORE_LC_NUMERIC();
12350 need = BIT_DIGITS(i);
12351 } /* if i < 0, the number of digits is hard to predict. */
12353 need += has_precis ? precis : 6; /* known default */
12358 #ifdef HAS_LDBL_SPRINTF_BUG
12359 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12360 with sfio - Allen <allens@cpan.org> */
12363 # define MY_DBL_MAX DBL_MAX
12364 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12365 # if DOUBLESIZE >= 8
12366 # define MY_DBL_MAX 1.7976931348623157E+308L
12368 # define MY_DBL_MAX 3.40282347E+38L
12372 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12373 # define MY_DBL_MAX_BUG 1L
12375 # define MY_DBL_MAX_BUG MY_DBL_MAX
12379 # define MY_DBL_MIN DBL_MIN
12380 # else /* XXX guessing! -Allen */
12381 # if DOUBLESIZE >= 8
12382 # define MY_DBL_MIN 2.2250738585072014E-308L
12384 # define MY_DBL_MIN 1.17549435E-38L
12388 if ((intsize == 'q') && (c == 'f') &&
12389 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12390 (need < DBL_DIG)) {
12391 /* it's going to be short enough that
12392 * long double precision is not needed */
12394 if ((fv <= 0L) && (fv >= -0L))
12395 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12397 /* would use Perl_fp_class as a double-check but not
12398 * functional on IRIX - see perl.h comments */
12400 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12401 /* It's within the range that a double can represent */
12402 #if defined(DBL_MAX) && !defined(DBL_MIN)
12403 if ((fv >= ((long double)1/DBL_MAX)) ||
12404 (fv <= (-(long double)1/DBL_MAX)))
12406 fix_ldbl_sprintf_bug = TRUE;
12409 if (fix_ldbl_sprintf_bug == TRUE) {
12419 # undef MY_DBL_MAX_BUG
12422 #endif /* HAS_LDBL_SPRINTF_BUG */
12424 need += 20; /* fudge factor */
12425 if (PL_efloatsize < need) {
12426 Safefree(PL_efloatbuf);
12427 PL_efloatsize = need + 20; /* more fudge */
12428 Newx(PL_efloatbuf, PL_efloatsize, char);
12429 PL_efloatbuf[0] = '\0';
12432 if ( !(width || left || plus || alt) && fill != '0'
12433 && has_precis && intsize != 'q' /* Shortcuts */
12434 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12435 /* See earlier comment about buggy Gconvert when digits,
12437 if ( c == 'g' && precis ) {
12438 STORE_LC_NUMERIC_SET_TO_NEEDED();
12439 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12440 /* May return an empty string for digits==0 */
12441 if (*PL_efloatbuf) {
12442 elen = strlen(PL_efloatbuf);
12443 goto float_converted;
12445 } else if ( c == 'f' && !precis ) {
12446 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12451 if (UNLIKELY(hexfp)) {
12452 /* Hexadecimal floating point. */
12453 char* p = PL_efloatbuf;
12454 U8 vhex[VHEX_SIZE];
12455 U8* v = vhex; /* working pointer to vhex */
12456 U8* vend; /* pointer to one beyond last digit of vhex */
12457 U8* vfnz = NULL; /* first non-zero */
12458 U8* vlnz = NULL; /* last non-zero */
12459 U8* v0 = NULL; /* first output */
12460 const bool lower = (c == 'a');
12461 /* At output the values of vhex (up to vend) will
12462 * be mapped through the xdig to get the actual
12463 * human-readable xdigits. */
12464 const char* xdig = PL_hexdigit;
12465 int zerotail = 0; /* how many extra zeros to append */
12466 int exponent = 0; /* exponent of the floating point input */
12467 bool hexradix = FALSE; /* should we output the radix */
12468 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12469 bool negative = FALSE;
12471 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12473 * For example with denormals, (assuming the vanilla
12474 * 64-bit double): the exponent is zero. 1xp-1074 is
12475 * the smallest denormal and the smallest double, it
12476 * could be output also as 0x0.0000000000001p-1022 to
12477 * match its internal structure. */
12479 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12480 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12482 #if NVSIZE > DOUBLESIZE
12483 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12484 /* In this case there is an implicit bit,
12485 * and therefore the exponent is shifted by one. */
12488 # ifdef NV_X86_80_BIT
12490 /* The subnormals of the x86-80 have a base exponent of -16382,
12491 * (while the physical exponent bits are zero) but the frexp()
12492 * returned the scientific-style floating exponent. We want
12493 * to map the last one as:
12494 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12495 * -16835..-16388 -> -16384
12496 * since we want to keep the first hexdigit
12497 * as one of the [8421]. */
12498 exponent = -4 * ( (exponent + 1) / -4) - 2;
12503 /* TBD: other non-implicit-bit platforms than the x86-80. */
12507 negative = fv < 0 || Perl_signbit(nv);
12518 xdig += 16; /* Use uppercase hex. */
12521 /* Find the first non-zero xdigit. */
12522 for (v = vhex; v < vend; v++) {
12530 /* Find the last non-zero xdigit. */
12531 for (v = vend - 1; v >= vhex; v--) {
12538 #if NVSIZE == DOUBLESIZE
12544 #ifndef NV_X86_80_BIT
12546 /* IEEE 754 subnormals (but not the x86 80-bit):
12547 * we want "normalize" the subnormal,
12548 * so we need to right shift the hex nybbles
12549 * so that the output of the subnormal starts
12550 * from the first true bit. (Another, equally
12551 * valid, policy would be to dump the subnormal
12552 * nybbles as-is, to display the "physical" layout.) */
12555 /* Find the ceil(log2(v[0])) of
12556 * the top non-zero nybble. */
12557 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12560 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12561 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12575 U8* ve = (subnormal ? vlnz + 1 : vend);
12576 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12577 if ((SSize_t)(precis + 1) < vn) {
12578 bool overflow = FALSE;
12579 if (v0[precis + 1] < 0x8) {
12580 /* Round down, nothing to do. */
12581 } else if (v0[precis + 1] > 0x8) {
12584 overflow = v0[precis] > 0xF;
12586 } else { /* v0[precis] == 0x8 */
12587 /* Half-point: round towards the one
12588 * with the even least-significant digit:
12596 * 78 -> 8 f8 -> 10 */
12597 if ((v0[precis] & 0x1)) {
12600 overflow = v0[precis] > 0xF;
12605 for (v = v0 + precis - 1; v >= v0; v--) {
12607 overflow = *v > 0xF;
12613 if (v == v0 - 1 && overflow) {
12614 /* If the overflow goes all the
12615 * way to the front, we need to
12616 * insert 0x1 in front, and adjust
12618 Move(v0, v0 + 1, vn, char);
12624 /* The new effective "last non zero". */
12625 vlnz = v0 + precis;
12629 subnormal ? precis - vn + 1 :
12630 precis - (vlnz - vhex);
12637 /* If there are non-zero xdigits, the radix
12638 * is output after the first one. */
12649 /* The radix is always output if precis, or if alt. */
12650 if (precis > 0 || alt) {
12655 #ifndef USE_LOCALE_NUMERIC
12658 STORE_LC_NUMERIC_SET_TO_NEEDED();
12659 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12661 const char* r = SvPV(PL_numeric_radix_sv, n);
12662 Copy(r, p, n, char);
12668 RESTORE_LC_NUMERIC();
12677 if (zerotail > 0) {
12678 while (zerotail--) {
12683 elen = p - PL_efloatbuf;
12684 elen += my_snprintf(p, PL_efloatsize - elen,
12685 "%c%+d", lower ? 'p' : 'P',
12688 if (elen < width) {
12690 /* Pad the back with spaces. */
12691 memset(PL_efloatbuf + elen, ' ', width - elen);
12693 else if (fill == '0') {
12694 /* Insert the zeros after the "0x" and the
12695 * the potential sign, but before the digits,
12696 * otherwise we end up with "0000xH.HHH...",
12697 * when we want "0x000H.HHH..." */
12698 STRLEN nzero = width - elen;
12699 char* zerox = PL_efloatbuf + 2;
12700 STRLEN nmove = elen - 2;
12701 if (negative || plus) {
12705 Move(zerox, zerox + nzero, nmove, char);
12706 memset(zerox, fill, nzero);
12709 /* Move it to the right. */
12710 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12712 /* Pad the front with spaces. */
12713 memset(PL_efloatbuf, ' ', width - elen);
12719 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12721 /* Not affecting infnan output: precision, alt, fill. */
12722 if (elen < width) {
12724 /* Pack the back with spaces. */
12725 memset(PL_efloatbuf + elen, ' ', width - elen);
12727 /* Move it to the right. */
12728 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12730 /* Pad the front with spaces. */
12731 memset(PL_efloatbuf, ' ', width - elen);
12739 char *ptr = ebuf + sizeof ebuf;
12742 #if defined(USE_QUADMATH)
12743 if (intsize == 'q') {
12747 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12748 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12749 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12750 * not USE_LONG_DOUBLE and NVff. In other words,
12751 * this needs to work without USE_LONG_DOUBLE. */
12752 if (intsize == 'q') {
12753 /* Copy the one or more characters in a long double
12754 * format before the 'base' ([efgEFG]) character to
12755 * the format string. */
12756 static char const ldblf[] = PERL_PRIfldbl;
12757 char const *p = ldblf + sizeof(ldblf) - 3;
12758 while (p >= ldblf) { *--ptr = *p--; }
12763 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12768 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12780 /* No taint. Otherwise we are in the strange situation
12781 * where printf() taints but print($float) doesn't.
12784 STORE_LC_NUMERIC_SET_TO_NEEDED();
12786 /* hopefully the above makes ptr a very constrained format
12787 * that is safe to use, even though it's not literal */
12788 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12789 #ifdef USE_QUADMATH
12791 const char* qfmt = quadmath_format_single(ptr);
12793 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12794 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12796 if ((IV)elen == -1)
12797 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12801 #elif defined(HAS_LONG_DOUBLE)
12802 elen = ((intsize == 'q')
12803 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12804 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12806 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12812 eptr = PL_efloatbuf;
12813 assert((IV)elen > 0); /* here zero elen is bad */
12815 #ifdef USE_LOCALE_NUMERIC
12816 /* If the decimal point character in the string is UTF-8, make the
12818 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12819 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12832 i = SvCUR(sv) - origlen;
12835 case 'c': *(va_arg(*args, char*)) = i; break;
12836 case 'h': *(va_arg(*args, short*)) = i; break;
12837 default: *(va_arg(*args, int*)) = i; break;
12838 case 'l': *(va_arg(*args, long*)) = i; break;
12839 case 'V': *(va_arg(*args, IV*)) = i; break;
12840 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12841 #ifdef HAS_PTRDIFF_T
12842 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12845 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12849 *(va_arg(*args, Quad_t*)) = i; break;
12856 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12857 goto donevalidconversion;
12864 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12865 && ckWARN(WARN_PRINTF))
12867 SV * const msg = sv_newmortal();
12868 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12869 (PL_op->op_type == OP_PRTF) ? "" : "s");
12870 if (fmtstart < patend) {
12871 const char * const fmtend = q < patend ? q : patend;
12873 sv_catpvs(msg, "\"%");
12874 for (f = fmtstart; f < fmtend; f++) {
12876 sv_catpvn_nomg(msg, f, 1);
12878 Perl_sv_catpvf(aTHX_ msg,
12879 "\\%03"UVof, (UV)*f & 0xFF);
12882 sv_catpvs(msg, "\"");
12884 sv_catpvs(msg, "end of string");
12886 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12889 /* output mangled stuff ... */
12895 /* ... right here, because formatting flags should not apply */
12896 SvGROW(sv, SvCUR(sv) + elen + 1);
12898 Copy(eptr, p, elen, char);
12901 SvCUR_set(sv, p - SvPVX_const(sv));
12903 continue; /* not "break" */
12906 if (is_utf8 != has_utf8) {
12909 sv_utf8_upgrade(sv);
12912 const STRLEN old_elen = elen;
12913 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12914 sv_utf8_upgrade(nsv);
12915 eptr = SvPVX_const(nsv);
12918 if (width) { /* fudge width (can't fudge elen) */
12919 width += elen - old_elen;
12925 /* signed value that's wrapped? */
12926 assert(elen <= ((~(STRLEN)0) >> 1));
12927 have = esignlen + zeros + elen;
12929 croak_memory_wrap();
12931 need = (have > width ? have : width);
12934 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12935 croak_memory_wrap();
12936 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12938 if (esignlen && fill == '0') {
12940 for (i = 0; i < (int)esignlen; i++)
12941 *p++ = esignbuf[i];
12943 if (gap && !left) {
12944 memset(p, fill, gap);
12947 if (esignlen && fill != '0') {
12949 for (i = 0; i < (int)esignlen; i++)
12950 *p++ = esignbuf[i];
12954 for (i = zeros; i; i--)
12958 Copy(eptr, p, elen, char);
12962 memset(p, ' ', gap);
12967 Copy(dotstr, p, dotstrlen, char);
12971 vectorize = FALSE; /* done iterating over vecstr */
12978 SvCUR_set(sv, p - SvPVX_const(sv));
12984 donevalidconversion:
12985 if (used_explicit_ix)
12986 no_redundant_warning = TRUE;
12988 S_warn_vcatpvfn_missing_argument(aTHX);
12991 /* Now that we've consumed all our printf format arguments (svix)
12992 * do we have things left on the stack that we didn't use?
12994 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12995 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12996 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13001 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13005 /* =========================================================================
13007 =head1 Cloning an interpreter
13011 All the macros and functions in this section are for the private use of
13012 the main function, perl_clone().
13014 The foo_dup() functions make an exact copy of an existing foo thingy.
13015 During the course of a cloning, a hash table is used to map old addresses
13016 to new addresses. The table is created and manipulated with the
13017 ptr_table_* functions.
13019 * =========================================================================*/
13022 #if defined(USE_ITHREADS)
13024 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13025 #ifndef GpREFCNT_inc
13026 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13030 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13031 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13032 If this changes, please unmerge ss_dup.
13033 Likewise, sv_dup_inc_multiple() relies on this fact. */
13034 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13035 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13036 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13037 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13038 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13039 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13040 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13041 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13042 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13043 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13044 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13045 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13046 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13048 /* clone a parser */
13051 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13055 PERL_ARGS_ASSERT_PARSER_DUP;
13060 /* look for it in the table first */
13061 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13065 /* create anew and remember what it is */
13066 Newxz(parser, 1, yy_parser);
13067 ptr_table_store(PL_ptr_table, proto, parser);
13069 /* XXX these not yet duped */
13070 parser->old_parser = NULL;
13071 parser->stack = NULL;
13073 parser->stack_size = 0;
13074 /* XXX parser->stack->state = 0; */
13076 /* XXX eventually, just Copy() most of the parser struct ? */
13078 parser->lex_brackets = proto->lex_brackets;
13079 parser->lex_casemods = proto->lex_casemods;
13080 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13081 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13082 parser->lex_casestack = savepvn(proto->lex_casestack,
13083 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13084 parser->lex_defer = proto->lex_defer;
13085 parser->lex_dojoin = proto->lex_dojoin;
13086 parser->lex_formbrack = proto->lex_formbrack;
13087 parser->lex_inpat = proto->lex_inpat;
13088 parser->lex_inwhat = proto->lex_inwhat;
13089 parser->lex_op = proto->lex_op;
13090 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13091 parser->lex_starts = proto->lex_starts;
13092 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13093 parser->multi_close = proto->multi_close;
13094 parser->multi_open = proto->multi_open;
13095 parser->multi_start = proto->multi_start;
13096 parser->multi_end = proto->multi_end;
13097 parser->preambled = proto->preambled;
13098 parser->lex_super_state = proto->lex_super_state;
13099 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13100 parser->lex_sub_op = proto->lex_sub_op;
13101 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13102 parser->linestr = sv_dup_inc(proto->linestr, param);
13103 parser->expect = proto->expect;
13104 parser->copline = proto->copline;
13105 parser->last_lop_op = proto->last_lop_op;
13106 parser->lex_state = proto->lex_state;
13107 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13108 /* rsfp_filters entries have fake IoDIRP() */
13109 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13110 parser->in_my = proto->in_my;
13111 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13112 parser->error_count = proto->error_count;
13113 parser->sig_elems = proto->sig_elems;
13114 parser->sig_optelems= proto->sig_optelems;
13115 parser->sig_slurpy = proto->sig_slurpy;
13116 parser->linestr = sv_dup_inc(proto->linestr, param);
13119 char * const ols = SvPVX(proto->linestr);
13120 char * const ls = SvPVX(parser->linestr);
13122 parser->bufptr = ls + (proto->bufptr >= ols ?
13123 proto->bufptr - ols : 0);
13124 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13125 proto->oldbufptr - ols : 0);
13126 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13127 proto->oldoldbufptr - ols : 0);
13128 parser->linestart = ls + (proto->linestart >= ols ?
13129 proto->linestart - ols : 0);
13130 parser->last_uni = ls + (proto->last_uni >= ols ?
13131 proto->last_uni - ols : 0);
13132 parser->last_lop = ls + (proto->last_lop >= ols ?
13133 proto->last_lop - ols : 0);
13135 parser->bufend = ls + SvCUR(parser->linestr);
13138 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13141 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13142 Copy(proto->nexttype, parser->nexttype, 5, I32);
13143 parser->nexttoke = proto->nexttoke;
13145 /* XXX should clone saved_curcop here, but we aren't passed
13146 * proto_perl; so do it in perl_clone_using instead */
13152 /* duplicate a file handle */
13155 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13159 PERL_ARGS_ASSERT_FP_DUP;
13160 PERL_UNUSED_ARG(type);
13163 return (PerlIO*)NULL;
13165 /* look for it in the table first */
13166 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13170 /* create anew and remember what it is */
13171 #ifdef __amigaos4__
13172 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13174 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13176 ptr_table_store(PL_ptr_table, fp, ret);
13180 /* duplicate a directory handle */
13183 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13187 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13189 const Direntry_t *dirent;
13190 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13196 PERL_UNUSED_CONTEXT;
13197 PERL_ARGS_ASSERT_DIRP_DUP;
13202 /* look for it in the table first */
13203 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13207 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13209 PERL_UNUSED_ARG(param);
13213 /* open the current directory (so we can switch back) */
13214 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13216 /* chdir to our dir handle and open the present working directory */
13217 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13218 PerlDir_close(pwd);
13219 return (DIR *)NULL;
13221 /* Now we should have two dir handles pointing to the same dir. */
13223 /* Be nice to the calling code and chdir back to where we were. */
13224 /* XXX If this fails, then what? */
13225 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13227 /* We have no need of the pwd handle any more. */
13228 PerlDir_close(pwd);
13231 # define d_namlen(d) (d)->d_namlen
13233 # define d_namlen(d) strlen((d)->d_name)
13235 /* Iterate once through dp, to get the file name at the current posi-
13236 tion. Then step back. */
13237 pos = PerlDir_tell(dp);
13238 if ((dirent = PerlDir_read(dp))) {
13239 len = d_namlen(dirent);
13240 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13241 /* If the len is somehow magically longer than the
13242 * maximum length of the directory entry, even though
13243 * we could fit it in a buffer, we could not copy it
13244 * from the dirent. Bail out. */
13245 PerlDir_close(ret);
13248 if (len <= sizeof smallbuf) name = smallbuf;
13249 else Newx(name, len, char);
13250 Move(dirent->d_name, name, len, char);
13252 PerlDir_seek(dp, pos);
13254 /* Iterate through the new dir handle, till we find a file with the
13256 if (!dirent) /* just before the end */
13258 pos = PerlDir_tell(ret);
13259 if (PerlDir_read(ret)) continue; /* not there yet */
13260 PerlDir_seek(ret, pos); /* step back */
13264 const long pos0 = PerlDir_tell(ret);
13266 pos = PerlDir_tell(ret);
13267 if ((dirent = PerlDir_read(ret))) {
13268 if (len == (STRLEN)d_namlen(dirent)
13269 && memEQ(name, dirent->d_name, len)) {
13271 PerlDir_seek(ret, pos); /* step back */
13274 /* else we are not there yet; keep iterating */
13276 else { /* This is not meant to happen. The best we can do is
13277 reset the iterator to the beginning. */
13278 PerlDir_seek(ret, pos0);
13285 if (name && name != smallbuf)
13290 ret = win32_dirp_dup(dp, param);
13293 /* pop it in the pointer table */
13295 ptr_table_store(PL_ptr_table, dp, ret);
13300 /* duplicate a typeglob */
13303 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13307 PERL_ARGS_ASSERT_GP_DUP;
13311 /* look for it in the table first */
13312 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13316 /* create anew and remember what it is */
13318 ptr_table_store(PL_ptr_table, gp, ret);
13321 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13322 on Newxz() to do this for us. */
13323 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13324 ret->gp_io = io_dup_inc(gp->gp_io, param);
13325 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13326 ret->gp_av = av_dup_inc(gp->gp_av, param);
13327 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13328 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13329 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13330 ret->gp_cvgen = gp->gp_cvgen;
13331 ret->gp_line = gp->gp_line;
13332 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13336 /* duplicate a chain of magic */
13339 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13341 MAGIC *mgret = NULL;
13342 MAGIC **mgprev_p = &mgret;
13344 PERL_ARGS_ASSERT_MG_DUP;
13346 for (; mg; mg = mg->mg_moremagic) {
13349 if ((param->flags & CLONEf_JOIN_IN)
13350 && mg->mg_type == PERL_MAGIC_backref)
13351 /* when joining, we let the individual SVs add themselves to
13352 * backref as needed. */
13355 Newx(nmg, 1, MAGIC);
13357 mgprev_p = &(nmg->mg_moremagic);
13359 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13360 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13361 from the original commit adding Perl_mg_dup() - revision 4538.
13362 Similarly there is the annotation "XXX random ptr?" next to the
13363 assignment to nmg->mg_ptr. */
13366 /* FIXME for plugins
13367 if (nmg->mg_type == PERL_MAGIC_qr) {
13368 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13372 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13373 ? nmg->mg_type == PERL_MAGIC_backref
13374 /* The backref AV has its reference
13375 * count deliberately bumped by 1 */
13376 ? SvREFCNT_inc(av_dup_inc((const AV *)
13377 nmg->mg_obj, param))
13378 : sv_dup_inc(nmg->mg_obj, param)
13379 : sv_dup(nmg->mg_obj, param);
13381 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13382 if (nmg->mg_len > 0) {
13383 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13384 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13385 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13387 AMT * const namtp = (AMT*)nmg->mg_ptr;
13388 sv_dup_inc_multiple((SV**)(namtp->table),
13389 (SV**)(namtp->table), NofAMmeth, param);
13392 else if (nmg->mg_len == HEf_SVKEY)
13393 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13395 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13396 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13402 #endif /* USE_ITHREADS */
13404 struct ptr_tbl_arena {
13405 struct ptr_tbl_arena *next;
13406 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13409 /* create a new pointer-mapping table */
13412 Perl_ptr_table_new(pTHX)
13415 PERL_UNUSED_CONTEXT;
13417 Newx(tbl, 1, PTR_TBL_t);
13418 tbl->tbl_max = 511;
13419 tbl->tbl_items = 0;
13420 tbl->tbl_arena = NULL;
13421 tbl->tbl_arena_next = NULL;
13422 tbl->tbl_arena_end = NULL;
13423 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13427 #define PTR_TABLE_HASH(ptr) \
13428 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13430 /* map an existing pointer using a table */
13432 STATIC PTR_TBL_ENT_t *
13433 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13435 PTR_TBL_ENT_t *tblent;
13436 const UV hash = PTR_TABLE_HASH(sv);
13438 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13440 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13441 for (; tblent; tblent = tblent->next) {
13442 if (tblent->oldval == sv)
13449 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13451 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13453 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13454 PERL_UNUSED_CONTEXT;
13456 return tblent ? tblent->newval : NULL;
13459 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13460 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13461 * the core's typical use of ptr_tables in thread cloning. */
13464 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13466 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13468 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13469 PERL_UNUSED_CONTEXT;
13472 tblent->newval = newsv;
13474 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13476 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13477 struct ptr_tbl_arena *new_arena;
13479 Newx(new_arena, 1, struct ptr_tbl_arena);
13480 new_arena->next = tbl->tbl_arena;
13481 tbl->tbl_arena = new_arena;
13482 tbl->tbl_arena_next = new_arena->array;
13483 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13486 tblent = tbl->tbl_arena_next++;
13488 tblent->oldval = oldsv;
13489 tblent->newval = newsv;
13490 tblent->next = tbl->tbl_ary[entry];
13491 tbl->tbl_ary[entry] = tblent;
13493 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13494 ptr_table_split(tbl);
13498 /* double the hash bucket size of an existing ptr table */
13501 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13503 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13504 const UV oldsize = tbl->tbl_max + 1;
13505 UV newsize = oldsize * 2;
13508 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13509 PERL_UNUSED_CONTEXT;
13511 Renew(ary, newsize, PTR_TBL_ENT_t*);
13512 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13513 tbl->tbl_max = --newsize;
13514 tbl->tbl_ary = ary;
13515 for (i=0; i < oldsize; i++, ary++) {
13516 PTR_TBL_ENT_t **entp = ary;
13517 PTR_TBL_ENT_t *ent = *ary;
13518 PTR_TBL_ENT_t **curentp;
13521 curentp = ary + oldsize;
13523 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13525 ent->next = *curentp;
13535 /* remove all the entries from a ptr table */
13536 /* Deprecated - will be removed post 5.14 */
13539 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13541 PERL_UNUSED_CONTEXT;
13542 if (tbl && tbl->tbl_items) {
13543 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13545 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13548 struct ptr_tbl_arena *next = arena->next;
13554 tbl->tbl_items = 0;
13555 tbl->tbl_arena = NULL;
13556 tbl->tbl_arena_next = NULL;
13557 tbl->tbl_arena_end = NULL;
13561 /* clear and free a ptr table */
13564 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13566 struct ptr_tbl_arena *arena;
13568 PERL_UNUSED_CONTEXT;
13574 arena = tbl->tbl_arena;
13577 struct ptr_tbl_arena *next = arena->next;
13583 Safefree(tbl->tbl_ary);
13587 #if defined(USE_ITHREADS)
13590 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13592 PERL_ARGS_ASSERT_RVPV_DUP;
13594 assert(!isREGEXP(sstr));
13596 if (SvWEAKREF(sstr)) {
13597 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13598 if (param->flags & CLONEf_JOIN_IN) {
13599 /* if joining, we add any back references individually rather
13600 * than copying the whole backref array */
13601 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13605 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13607 else if (SvPVX_const(sstr)) {
13608 /* Has something there */
13610 /* Normal PV - clone whole allocated space */
13611 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13612 /* sstr may not be that normal, but actually copy on write.
13613 But we are a true, independent SV, so: */
13617 /* Special case - not normally malloced for some reason */
13618 if (isGV_with_GP(sstr)) {
13619 /* Don't need to do anything here. */
13621 else if ((SvIsCOW(sstr))) {
13622 /* A "shared" PV - clone it as "shared" PV */
13624 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13628 /* Some other special case - random pointer */
13629 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13634 /* Copy the NULL */
13635 SvPV_set(dstr, NULL);
13639 /* duplicate a list of SVs. source and dest may point to the same memory. */
13641 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13642 SSize_t items, CLONE_PARAMS *const param)
13644 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13646 while (items-- > 0) {
13647 *dest++ = sv_dup_inc(*source++, param);
13653 /* duplicate an SV of any type (including AV, HV etc) */
13656 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13661 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13663 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13664 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13669 /* look for it in the table first */
13670 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13674 if(param->flags & CLONEf_JOIN_IN) {
13675 /** We are joining here so we don't want do clone
13676 something that is bad **/
13677 if (SvTYPE(sstr) == SVt_PVHV) {
13678 const HEK * const hvname = HvNAME_HEK(sstr);
13680 /** don't clone stashes if they already exist **/
13681 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13682 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13683 ptr_table_store(PL_ptr_table, sstr, dstr);
13687 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13688 HV *stash = GvSTASH(sstr);
13689 const HEK * hvname;
13690 if (stash && (hvname = HvNAME_HEK(stash))) {
13691 /** don't clone GVs if they already exist **/
13693 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13694 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13696 stash, GvNAME(sstr),
13702 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13703 ptr_table_store(PL_ptr_table, sstr, *svp);
13710 /* create anew and remember what it is */
13713 #ifdef DEBUG_LEAKING_SCALARS
13714 dstr->sv_debug_optype = sstr->sv_debug_optype;
13715 dstr->sv_debug_line = sstr->sv_debug_line;
13716 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13717 dstr->sv_debug_parent = (SV*)sstr;
13718 FREE_SV_DEBUG_FILE(dstr);
13719 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13722 ptr_table_store(PL_ptr_table, sstr, dstr);
13725 SvFLAGS(dstr) = SvFLAGS(sstr);
13726 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13727 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13730 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13731 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13732 (void*)PL_watch_pvx, SvPVX_const(sstr));
13735 /* don't clone objects whose class has asked us not to */
13737 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13743 switch (SvTYPE(sstr)) {
13745 SvANY(dstr) = NULL;
13748 SET_SVANY_FOR_BODYLESS_IV(dstr);
13750 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13752 SvIV_set(dstr, SvIVX(sstr));
13756 #if NVSIZE <= IVSIZE
13757 SET_SVANY_FOR_BODYLESS_NV(dstr);
13759 SvANY(dstr) = new_XNV();
13761 SvNV_set(dstr, SvNVX(sstr));
13765 /* These are all the types that need complex bodies allocating. */
13767 const svtype sv_type = SvTYPE(sstr);
13768 const struct body_details *const sv_type_details
13769 = bodies_by_type + sv_type;
13773 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13789 assert(sv_type_details->body_size);
13790 if (sv_type_details->arena) {
13791 new_body_inline(new_body, sv_type);
13793 = (void*)((char*)new_body - sv_type_details->offset);
13795 new_body = new_NOARENA(sv_type_details);
13799 SvANY(dstr) = new_body;
13802 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13803 ((char*)SvANY(dstr)) + sv_type_details->offset,
13804 sv_type_details->copy, char);
13806 Copy(((char*)SvANY(sstr)),
13807 ((char*)SvANY(dstr)),
13808 sv_type_details->body_size + sv_type_details->offset, char);
13811 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13812 && !isGV_with_GP(dstr)
13814 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13815 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13817 /* The Copy above means that all the source (unduplicated) pointers
13818 are now in the destination. We can check the flags and the
13819 pointers in either, but it's possible that there's less cache
13820 missing by always going for the destination.
13821 FIXME - instrument and check that assumption */
13822 if (sv_type >= SVt_PVMG) {
13824 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13825 if (SvOBJECT(dstr) && SvSTASH(dstr))
13826 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13827 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13830 /* The cast silences a GCC warning about unhandled types. */
13831 switch ((int)sv_type) {
13842 /* FIXME for plugins */
13843 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13844 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13847 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13848 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13849 LvTARG(dstr) = dstr;
13850 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13851 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13853 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13854 if (isREGEXP(sstr)) goto duprex;
13856 /* non-GP case already handled above */
13857 if(isGV_with_GP(sstr)) {
13858 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13859 /* Don't call sv_add_backref here as it's going to be
13860 created as part of the magic cloning of the symbol
13861 table--unless this is during a join and the stash
13862 is not actually being cloned. */
13863 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13864 at the point of this comment. */
13865 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13866 if (param->flags & CLONEf_JOIN_IN)
13867 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13868 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13869 (void)GpREFCNT_inc(GvGP(dstr));
13873 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13874 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13875 /* I have no idea why fake dirp (rsfps)
13876 should be treated differently but otherwise
13877 we end up with leaks -- sky*/
13878 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13879 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13880 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13882 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13883 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13884 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13885 if (IoDIRP(dstr)) {
13886 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13889 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13891 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13893 if (IoOFP(dstr) == IoIFP(sstr))
13894 IoOFP(dstr) = IoIFP(dstr);
13896 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13897 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13898 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13899 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13902 /* avoid cloning an empty array */
13903 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13904 SV **dst_ary, **src_ary;
13905 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13907 src_ary = AvARRAY((const AV *)sstr);
13908 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13909 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13910 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13911 AvALLOC((const AV *)dstr) = dst_ary;
13912 if (AvREAL((const AV *)sstr)) {
13913 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13917 while (items-- > 0)
13918 *dst_ary++ = sv_dup(*src_ary++, param);
13920 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13921 while (items-- > 0) {
13926 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13927 AvALLOC((const AV *)dstr) = (SV**)NULL;
13928 AvMAX( (const AV *)dstr) = -1;
13929 AvFILLp((const AV *)dstr) = -1;
13933 if (HvARRAY((const HV *)sstr)) {
13935 const bool sharekeys = !!HvSHAREKEYS(sstr);
13936 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13937 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13939 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13940 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13942 HvARRAY(dstr) = (HE**)darray;
13943 while (i <= sxhv->xhv_max) {
13944 const HE * const source = HvARRAY(sstr)[i];
13945 HvARRAY(dstr)[i] = source
13946 ? he_dup(source, sharekeys, param) : 0;
13950 const struct xpvhv_aux * const saux = HvAUX(sstr);
13951 struct xpvhv_aux * const daux = HvAUX(dstr);
13952 /* This flag isn't copied. */
13955 if (saux->xhv_name_count) {
13956 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13958 = saux->xhv_name_count < 0
13959 ? -saux->xhv_name_count
13960 : saux->xhv_name_count;
13961 HEK **shekp = sname + count;
13963 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13964 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13965 while (shekp-- > sname) {
13967 *dhekp = hek_dup(*shekp, param);
13971 daux->xhv_name_u.xhvnameu_name
13972 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13975 daux->xhv_name_count = saux->xhv_name_count;
13977 daux->xhv_aux_flags = saux->xhv_aux_flags;
13978 #ifdef PERL_HASH_RANDOMIZE_KEYS
13979 daux->xhv_rand = saux->xhv_rand;
13980 daux->xhv_last_rand = saux->xhv_last_rand;
13982 daux->xhv_riter = saux->xhv_riter;
13983 daux->xhv_eiter = saux->xhv_eiter
13984 ? he_dup(saux->xhv_eiter,
13985 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13986 /* backref array needs refcnt=2; see sv_add_backref */
13987 daux->xhv_backreferences =
13988 (param->flags & CLONEf_JOIN_IN)
13989 /* when joining, we let the individual GVs and
13990 * CVs add themselves to backref as
13991 * needed. This avoids pulling in stuff
13992 * that isn't required, and simplifies the
13993 * case where stashes aren't cloned back
13994 * if they already exist in the parent
13997 : saux->xhv_backreferences
13998 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13999 ? MUTABLE_AV(SvREFCNT_inc(
14000 sv_dup_inc((const SV *)
14001 saux->xhv_backreferences, param)))
14002 : MUTABLE_AV(sv_dup((const SV *)
14003 saux->xhv_backreferences, param))
14006 daux->xhv_mro_meta = saux->xhv_mro_meta
14007 ? mro_meta_dup(saux->xhv_mro_meta, param)
14010 /* Record stashes for possible cloning in Perl_clone(). */
14012 av_push(param->stashes, dstr);
14016 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14019 if (!(param->flags & CLONEf_COPY_STACKS)) {
14024 /* NOTE: not refcounted */
14025 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14026 hv_dup(CvSTASH(dstr), param);
14027 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14028 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14029 if (!CvISXSUB(dstr)) {
14031 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14033 CvSLABBED_off(dstr);
14034 } else if (CvCONST(dstr)) {
14035 CvXSUBANY(dstr).any_ptr =
14036 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14038 assert(!CvSLABBED(dstr));
14039 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14041 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14042 hek_dup(CvNAME_HEK((CV *)sstr), param);
14043 /* don't dup if copying back - CvGV isn't refcounted, so the
14044 * duped GV may never be freed. A bit of a hack! DAPM */
14046 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14048 ? gv_dup_inc(CvGV(sstr), param)
14049 : (param->flags & CLONEf_JOIN_IN)
14051 : gv_dup(CvGV(sstr), param);
14053 if (!CvISXSUB(sstr)) {
14054 PADLIST * padlist = CvPADLIST(sstr);
14056 padlist = padlist_dup(padlist, param);
14057 CvPADLIST_set(dstr, padlist);
14059 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14060 PoisonPADLIST(dstr);
14063 CvWEAKOUTSIDE(sstr)
14064 ? cv_dup( CvOUTSIDE(dstr), param)
14065 : cv_dup_inc(CvOUTSIDE(dstr), param);
14075 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14077 PERL_ARGS_ASSERT_SV_DUP_INC;
14078 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14082 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14084 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14085 PERL_ARGS_ASSERT_SV_DUP;
14087 /* Track every SV that (at least initially) had a reference count of 0.
14088 We need to do this by holding an actual reference to it in this array.
14089 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14090 (akin to the stashes hash, and the perl stack), we come unstuck if
14091 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14092 thread) is manipulated in a CLONE method, because CLONE runs before the
14093 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14094 (and fix things up by giving each a reference via the temps stack).
14095 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14096 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14097 before the walk of unreferenced happens and a reference to that is SV
14098 added to the temps stack. At which point we have the same SV considered
14099 to be in use, and free to be re-used. Not good.
14101 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14102 assert(param->unreferenced);
14103 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14109 /* duplicate a context */
14112 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14114 PERL_CONTEXT *ncxs;
14116 PERL_ARGS_ASSERT_CX_DUP;
14119 return (PERL_CONTEXT*)NULL;
14121 /* look for it in the table first */
14122 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14126 /* create anew and remember what it is */
14127 Newx(ncxs, max + 1, PERL_CONTEXT);
14128 ptr_table_store(PL_ptr_table, cxs, ncxs);
14129 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14132 PERL_CONTEXT * const ncx = &ncxs[ix];
14133 if (CxTYPE(ncx) == CXt_SUBST) {
14134 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14137 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14138 switch (CxTYPE(ncx)) {
14140 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14141 if(CxHASARGS(ncx)){
14142 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14144 ncx->blk_sub.savearray = NULL;
14146 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14147 ncx->blk_sub.prevcomppad);
14150 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14152 /* XXX should this sv_dup_inc? Or only if SvSCREAM ???? */
14153 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14154 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14155 /* XXX what do do with cur_top_env ???? */
14157 case CXt_LOOP_LAZYSV:
14158 ncx->blk_loop.state_u.lazysv.end
14159 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14160 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14161 duplication code instead.
14162 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14163 actually being the same function, and (2) order
14164 equivalence of the two unions.
14165 We can assert the later [but only at run time :-(] */
14166 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14167 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14170 ncx->blk_loop.state_u.ary.ary
14171 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14173 case CXt_LOOP_LIST:
14174 case CXt_LOOP_LAZYIV:
14175 /* code common to all 'for' CXt_LOOP_* types */
14176 ncx->blk_loop.itersave =
14177 sv_dup_inc(ncx->blk_loop.itersave, param);
14178 if (CxPADLOOP(ncx)) {
14179 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14180 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14181 ncx->blk_loop.oldcomppad =
14182 (PAD*)ptr_table_fetch(PL_ptr_table,
14183 ncx->blk_loop.oldcomppad);
14184 ncx->blk_loop.itervar_u.svp =
14185 &CX_CURPAD_SV(ncx->blk_loop, off);
14188 /* this copies the GV if CXp_FOR_GV, or the SV for an
14189 * alias (for \$x (...)) - relies on gv_dup being the
14190 * same as sv_dup */
14191 ncx->blk_loop.itervar_u.gv
14192 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14196 case CXt_LOOP_PLAIN:
14199 ncx->blk_format.prevcomppad =
14200 (PAD*)ptr_table_fetch(PL_ptr_table,
14201 ncx->blk_format.prevcomppad);
14202 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14203 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14204 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14208 ncx->blk_givwhen.defsv_save =
14209 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14222 /* duplicate a stack info structure */
14225 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14229 PERL_ARGS_ASSERT_SI_DUP;
14232 return (PERL_SI*)NULL;
14234 /* look for it in the table first */
14235 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14239 /* create anew and remember what it is */
14240 Newxz(nsi, 1, PERL_SI);
14241 ptr_table_store(PL_ptr_table, si, nsi);
14243 nsi->si_stack = av_dup_inc(si->si_stack, param);
14244 nsi->si_cxix = si->si_cxix;
14245 nsi->si_cxmax = si->si_cxmax;
14246 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14247 nsi->si_type = si->si_type;
14248 nsi->si_prev = si_dup(si->si_prev, param);
14249 nsi->si_next = si_dup(si->si_next, param);
14250 nsi->si_markoff = si->si_markoff;
14255 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14256 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14257 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14258 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14259 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14260 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14261 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14262 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14263 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14264 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14265 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14266 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14267 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14268 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14269 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14270 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14273 #define pv_dup_inc(p) SAVEPV(p)
14274 #define pv_dup(p) SAVEPV(p)
14275 #define svp_dup_inc(p,pp) any_dup(p,pp)
14277 /* map any object to the new equivent - either something in the
14278 * ptr table, or something in the interpreter structure
14282 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14286 PERL_ARGS_ASSERT_ANY_DUP;
14289 return (void*)NULL;
14291 /* look for it in the table first */
14292 ret = ptr_table_fetch(PL_ptr_table, v);
14296 /* see if it is part of the interpreter structure */
14297 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14298 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14306 /* duplicate the save stack */
14309 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14312 ANY * const ss = proto_perl->Isavestack;
14313 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14314 I32 ix = proto_perl->Isavestack_ix;
14327 void (*dptr) (void*);
14328 void (*dxptr) (pTHX_ void*);
14330 PERL_ARGS_ASSERT_SS_DUP;
14332 Newxz(nss, max, ANY);
14335 const UV uv = POPUV(ss,ix);
14336 const U8 type = (U8)uv & SAVE_MASK;
14338 TOPUV(nss,ix) = uv;
14340 case SAVEt_CLEARSV:
14341 case SAVEt_CLEARPADRANGE:
14343 case SAVEt_HELEM: /* hash element */
14344 case SAVEt_SV: /* scalar reference */
14345 sv = (const SV *)POPPTR(ss,ix);
14346 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14348 case SAVEt_ITEM: /* normal string */
14349 case SAVEt_GVSV: /* scalar slot in GV */
14350 sv = (const SV *)POPPTR(ss,ix);
14351 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14352 if (type == SAVEt_SV)
14356 case SAVEt_MORTALIZESV:
14357 case SAVEt_READONLY_OFF:
14358 sv = (const SV *)POPPTR(ss,ix);
14359 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14361 case SAVEt_FREEPADNAME:
14362 ptr = POPPTR(ss,ix);
14363 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14364 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14366 case SAVEt_SHARED_PVREF: /* char* in shared space */
14367 c = (char*)POPPTR(ss,ix);
14368 TOPPTR(nss,ix) = savesharedpv(c);
14369 ptr = POPPTR(ss,ix);
14370 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14372 case SAVEt_GENERIC_SVREF: /* generic sv */
14373 case SAVEt_SVREF: /* scalar reference */
14374 sv = (const SV *)POPPTR(ss,ix);
14375 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14376 if (type == SAVEt_SVREF)
14377 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14378 ptr = POPPTR(ss,ix);
14379 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14381 case SAVEt_GVSLOT: /* any slot in GV */
14382 sv = (const SV *)POPPTR(ss,ix);
14383 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14384 ptr = POPPTR(ss,ix);
14385 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14386 sv = (const SV *)POPPTR(ss,ix);
14387 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14389 case SAVEt_HV: /* hash reference */
14390 case SAVEt_AV: /* array reference */
14391 sv = (const SV *) POPPTR(ss,ix);
14392 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14394 case SAVEt_COMPPAD:
14396 sv = (const SV *) POPPTR(ss,ix);
14397 TOPPTR(nss,ix) = sv_dup(sv, param);
14399 case SAVEt_INT: /* int reference */
14400 ptr = POPPTR(ss,ix);
14401 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14402 intval = (int)POPINT(ss,ix);
14403 TOPINT(nss,ix) = intval;
14405 case SAVEt_LONG: /* long reference */
14406 ptr = POPPTR(ss,ix);
14407 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14408 longval = (long)POPLONG(ss,ix);
14409 TOPLONG(nss,ix) = longval;
14411 case SAVEt_I32: /* I32 reference */
14412 ptr = POPPTR(ss,ix);
14413 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14415 TOPINT(nss,ix) = i;
14417 case SAVEt_IV: /* IV reference */
14418 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14419 ptr = POPPTR(ss,ix);
14420 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14422 TOPIV(nss,ix) = iv;
14424 case SAVEt_TMPSFLOOR:
14426 TOPIV(nss,ix) = iv;
14428 case SAVEt_HPTR: /* HV* reference */
14429 case SAVEt_APTR: /* AV* reference */
14430 case SAVEt_SPTR: /* SV* reference */
14431 ptr = POPPTR(ss,ix);
14432 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14433 sv = (const SV *)POPPTR(ss,ix);
14434 TOPPTR(nss,ix) = sv_dup(sv, param);
14436 case SAVEt_VPTR: /* random* reference */
14437 ptr = POPPTR(ss,ix);
14438 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14440 case SAVEt_INT_SMALL:
14441 case SAVEt_I32_SMALL:
14442 case SAVEt_I16: /* I16 reference */
14443 case SAVEt_I8: /* I8 reference */
14445 ptr = POPPTR(ss,ix);
14446 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14448 case SAVEt_GENERIC_PVREF: /* generic char* */
14449 case SAVEt_PPTR: /* char* reference */
14450 ptr = POPPTR(ss,ix);
14451 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14452 c = (char*)POPPTR(ss,ix);
14453 TOPPTR(nss,ix) = pv_dup(c);
14455 case SAVEt_GP: /* scalar reference */
14456 gp = (GP*)POPPTR(ss,ix);
14457 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14458 (void)GpREFCNT_inc(gp);
14459 gv = (const GV *)POPPTR(ss,ix);
14460 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14463 ptr = POPPTR(ss,ix);
14464 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14465 /* these are assumed to be refcounted properly */
14467 switch (((OP*)ptr)->op_type) {
14469 case OP_LEAVESUBLV:
14473 case OP_LEAVEWRITE:
14474 TOPPTR(nss,ix) = ptr;
14477 (void) OpREFCNT_inc(o);
14481 TOPPTR(nss,ix) = NULL;
14486 TOPPTR(nss,ix) = NULL;
14488 case SAVEt_FREECOPHH:
14489 ptr = POPPTR(ss,ix);
14490 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14492 case SAVEt_ADELETE:
14493 av = (const AV *)POPPTR(ss,ix);
14494 TOPPTR(nss,ix) = av_dup_inc(av, param);
14496 TOPINT(nss,ix) = i;
14499 hv = (const HV *)POPPTR(ss,ix);
14500 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14502 TOPINT(nss,ix) = i;
14505 c = (char*)POPPTR(ss,ix);
14506 TOPPTR(nss,ix) = pv_dup_inc(c);
14508 case SAVEt_STACK_POS: /* Position on Perl stack */
14510 TOPINT(nss,ix) = i;
14512 case SAVEt_DESTRUCTOR:
14513 ptr = POPPTR(ss,ix);
14514 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14515 dptr = POPDPTR(ss,ix);
14516 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14517 any_dup(FPTR2DPTR(void *, dptr),
14520 case SAVEt_DESTRUCTOR_X:
14521 ptr = POPPTR(ss,ix);
14522 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14523 dxptr = POPDXPTR(ss,ix);
14524 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14525 any_dup(FPTR2DPTR(void *, dxptr),
14528 case SAVEt_REGCONTEXT:
14530 ix -= uv >> SAVE_TIGHT_SHIFT;
14532 case SAVEt_AELEM: /* array element */
14533 sv = (const SV *)POPPTR(ss,ix);
14534 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14536 TOPINT(nss,ix) = i;
14537 av = (const AV *)POPPTR(ss,ix);
14538 TOPPTR(nss,ix) = av_dup_inc(av, param);
14541 ptr = POPPTR(ss,ix);
14542 TOPPTR(nss,ix) = ptr;
14545 ptr = POPPTR(ss,ix);
14546 ptr = cophh_copy((COPHH*)ptr);
14547 TOPPTR(nss,ix) = ptr;
14549 TOPINT(nss,ix) = i;
14550 if (i & HINT_LOCALIZE_HH) {
14551 hv = (const HV *)POPPTR(ss,ix);
14552 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14555 case SAVEt_PADSV_AND_MORTALIZE:
14556 longval = (long)POPLONG(ss,ix);
14557 TOPLONG(nss,ix) = longval;
14558 ptr = POPPTR(ss,ix);
14559 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14560 sv = (const SV *)POPPTR(ss,ix);
14561 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14563 case SAVEt_SET_SVFLAGS:
14565 TOPINT(nss,ix) = i;
14567 TOPINT(nss,ix) = i;
14568 sv = (const SV *)POPPTR(ss,ix);
14569 TOPPTR(nss,ix) = sv_dup(sv, param);
14571 case SAVEt_COMPILE_WARNINGS:
14572 ptr = POPPTR(ss,ix);
14573 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14576 ptr = POPPTR(ss,ix);
14577 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14581 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14589 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14590 * flag to the result. This is done for each stash before cloning starts,
14591 * so we know which stashes want their objects cloned */
14594 do_mark_cloneable_stash(pTHX_ SV *const sv)
14596 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14598 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14599 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14600 if (cloner && GvCV(cloner)) {
14607 mXPUSHs(newSVhek(hvname));
14609 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14616 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14624 =for apidoc perl_clone
14626 Create and return a new interpreter by cloning the current one.
14628 C<perl_clone> takes these flags as parameters:
14630 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14631 without it we only clone the data and zero the stacks,
14632 with it we copy the stacks and the new perl interpreter is
14633 ready to run at the exact same point as the previous one.
14634 The pseudo-fork code uses C<COPY_STACKS> while the
14635 threads->create doesn't.
14637 C<CLONEf_KEEP_PTR_TABLE> -
14638 C<perl_clone> keeps a ptr_table with the pointer of the old
14639 variable as a key and the new variable as a value,
14640 this allows it to check if something has been cloned and not
14641 clone it again but rather just use the value and increase the
14642 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14643 the ptr_table using the function
14644 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14645 reason to keep it around is if you want to dup some of your own
14646 variable who are outside the graph perl scans, an example of this
14647 code is in F<threads.xs> create.
14649 C<CLONEf_CLONE_HOST> -
14650 This is a win32 thing, it is ignored on unix, it tells perls
14651 win32host code (which is c++) to clone itself, this is needed on
14652 win32 if you want to run two threads at the same time,
14653 if you just want to do some stuff in a separate perl interpreter
14654 and then throw it away and return to the original one,
14655 you don't need to do anything.
14660 /* XXX the above needs expanding by someone who actually understands it ! */
14661 EXTERN_C PerlInterpreter *
14662 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14665 perl_clone(PerlInterpreter *proto_perl, UV flags)
14668 #ifdef PERL_IMPLICIT_SYS
14670 PERL_ARGS_ASSERT_PERL_CLONE;
14672 /* perlhost.h so we need to call into it
14673 to clone the host, CPerlHost should have a c interface, sky */
14675 #ifndef __amigaos4__
14676 if (flags & CLONEf_CLONE_HOST) {
14677 return perl_clone_host(proto_perl,flags);
14680 return perl_clone_using(proto_perl, flags,
14682 proto_perl->IMemShared,
14683 proto_perl->IMemParse,
14685 proto_perl->IStdIO,
14689 proto_perl->IProc);
14693 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14694 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14695 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14696 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14697 struct IPerlDir* ipD, struct IPerlSock* ipS,
14698 struct IPerlProc* ipP)
14700 /* XXX many of the string copies here can be optimized if they're
14701 * constants; they need to be allocated as common memory and just
14702 * their pointers copied. */
14705 CLONE_PARAMS clone_params;
14706 CLONE_PARAMS* const param = &clone_params;
14708 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14710 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14711 #else /* !PERL_IMPLICIT_SYS */
14713 CLONE_PARAMS clone_params;
14714 CLONE_PARAMS* param = &clone_params;
14715 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14717 PERL_ARGS_ASSERT_PERL_CLONE;
14718 #endif /* PERL_IMPLICIT_SYS */
14720 /* for each stash, determine whether its objects should be cloned */
14721 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14722 PERL_SET_THX(my_perl);
14725 PoisonNew(my_perl, 1, PerlInterpreter);
14728 PL_defstash = NULL; /* may be used by perl malloc() */
14731 PL_scopestack_name = 0;
14733 PL_savestack_ix = 0;
14734 PL_savestack_max = -1;
14735 PL_sig_pending = 0;
14737 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14738 Zero(&PL_padname_undef, 1, PADNAME);
14739 Zero(&PL_padname_const, 1, PADNAME);
14740 # ifdef DEBUG_LEAKING_SCALARS
14741 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14743 # ifdef PERL_TRACE_OPS
14744 Zero(PL_op_exec_cnt, OP_max+2, UV);
14746 #else /* !DEBUGGING */
14747 Zero(my_perl, 1, PerlInterpreter);
14748 #endif /* DEBUGGING */
14750 #ifdef PERL_IMPLICIT_SYS
14751 /* host pointers */
14753 PL_MemShared = ipMS;
14754 PL_MemParse = ipMP;
14761 #endif /* PERL_IMPLICIT_SYS */
14764 param->flags = flags;
14765 /* Nothing in the core code uses this, but we make it available to
14766 extensions (using mg_dup). */
14767 param->proto_perl = proto_perl;
14768 /* Likely nothing will use this, but it is initialised to be consistent
14769 with Perl_clone_params_new(). */
14770 param->new_perl = my_perl;
14771 param->unreferenced = NULL;
14774 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14776 PL_body_arenas = NULL;
14777 Zero(&PL_body_roots, 1, PL_body_roots);
14781 PL_sv_arenaroot = NULL;
14783 PL_debug = proto_perl->Idebug;
14785 /* dbargs array probably holds garbage */
14788 PL_compiling = proto_perl->Icompiling;
14790 /* pseudo environmental stuff */
14791 PL_origargc = proto_perl->Iorigargc;
14792 PL_origargv = proto_perl->Iorigargv;
14794 #ifndef NO_TAINT_SUPPORT
14795 /* Set tainting stuff before PerlIO_debug can possibly get called */
14796 PL_tainting = proto_perl->Itainting;
14797 PL_taint_warn = proto_perl->Itaint_warn;
14799 PL_tainting = FALSE;
14800 PL_taint_warn = FALSE;
14803 PL_minus_c = proto_perl->Iminus_c;
14805 PL_localpatches = proto_perl->Ilocalpatches;
14806 PL_splitstr = proto_perl->Isplitstr;
14807 PL_minus_n = proto_perl->Iminus_n;
14808 PL_minus_p = proto_perl->Iminus_p;
14809 PL_minus_l = proto_perl->Iminus_l;
14810 PL_minus_a = proto_perl->Iminus_a;
14811 PL_minus_E = proto_perl->Iminus_E;
14812 PL_minus_F = proto_perl->Iminus_F;
14813 PL_doswitches = proto_perl->Idoswitches;
14814 PL_dowarn = proto_perl->Idowarn;
14815 #ifdef PERL_SAWAMPERSAND
14816 PL_sawampersand = proto_perl->Isawampersand;
14818 PL_unsafe = proto_perl->Iunsafe;
14819 PL_perldb = proto_perl->Iperldb;
14820 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14821 PL_exit_flags = proto_perl->Iexit_flags;
14823 /* XXX time(&PL_basetime) when asked for? */
14824 PL_basetime = proto_perl->Ibasetime;
14826 PL_maxsysfd = proto_perl->Imaxsysfd;
14827 PL_statusvalue = proto_perl->Istatusvalue;
14829 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14831 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14834 /* RE engine related */
14835 PL_regmatch_slab = NULL;
14836 PL_reg_curpm = NULL;
14838 PL_sub_generation = proto_perl->Isub_generation;
14840 /* funky return mechanisms */
14841 PL_forkprocess = proto_perl->Iforkprocess;
14843 /* internal state */
14844 PL_main_start = proto_perl->Imain_start;
14845 PL_eval_root = proto_perl->Ieval_root;
14846 PL_eval_start = proto_perl->Ieval_start;
14848 PL_filemode = proto_perl->Ifilemode;
14849 PL_lastfd = proto_perl->Ilastfd;
14850 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14853 PL_gensym = proto_perl->Igensym;
14855 PL_laststatval = proto_perl->Ilaststatval;
14856 PL_laststype = proto_perl->Ilaststype;
14859 PL_profiledata = NULL;
14861 PL_generation = proto_perl->Igeneration;
14863 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14864 PL_in_clean_all = proto_perl->Iin_clean_all;
14866 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14867 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14868 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14869 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14870 PL_nomemok = proto_perl->Inomemok;
14871 PL_an = proto_perl->Ian;
14872 PL_evalseq = proto_perl->Ievalseq;
14873 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14874 PL_origalen = proto_perl->Iorigalen;
14876 PL_sighandlerp = proto_perl->Isighandlerp;
14878 PL_runops = proto_perl->Irunops;
14880 PL_subline = proto_perl->Isubline;
14882 PL_cv_has_eval = proto_perl->Icv_has_eval;
14885 PL_cryptseen = proto_perl->Icryptseen;
14888 #ifdef USE_LOCALE_COLLATE
14889 PL_collation_ix = proto_perl->Icollation_ix;
14890 PL_collation_standard = proto_perl->Icollation_standard;
14891 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14892 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14893 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14894 #endif /* USE_LOCALE_COLLATE */
14896 #ifdef USE_LOCALE_NUMERIC
14897 PL_numeric_standard = proto_perl->Inumeric_standard;
14898 PL_numeric_local = proto_perl->Inumeric_local;
14899 #endif /* !USE_LOCALE_NUMERIC */
14901 /* Did the locale setup indicate UTF-8? */
14902 PL_utf8locale = proto_perl->Iutf8locale;
14903 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14904 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
14905 /* Unicode features (see perlrun/-C) */
14906 PL_unicode = proto_perl->Iunicode;
14908 /* Pre-5.8 signals control */
14909 PL_signals = proto_perl->Isignals;
14911 /* times() ticks per second */
14912 PL_clocktick = proto_perl->Iclocktick;
14914 /* Recursion stopper for PerlIO_find_layer */
14915 PL_in_load_module = proto_perl->Iin_load_module;
14917 /* sort() routine */
14918 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14920 /* Not really needed/useful since the reenrant_retint is "volatile",
14921 * but do it for consistency's sake. */
14922 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14924 /* Hooks to shared SVs and locks. */
14925 PL_sharehook = proto_perl->Isharehook;
14926 PL_lockhook = proto_perl->Ilockhook;
14927 PL_unlockhook = proto_perl->Iunlockhook;
14928 PL_threadhook = proto_perl->Ithreadhook;
14929 PL_destroyhook = proto_perl->Idestroyhook;
14930 PL_signalhook = proto_perl->Isignalhook;
14932 PL_globhook = proto_perl->Iglobhook;
14935 PL_last_swash_hv = NULL; /* reinits on demand */
14936 PL_last_swash_klen = 0;
14937 PL_last_swash_key[0]= '\0';
14938 PL_last_swash_tmps = (U8*)NULL;
14939 PL_last_swash_slen = 0;
14941 PL_srand_called = proto_perl->Isrand_called;
14942 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14944 if (flags & CLONEf_COPY_STACKS) {
14945 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14946 PL_tmps_ix = proto_perl->Itmps_ix;
14947 PL_tmps_max = proto_perl->Itmps_max;
14948 PL_tmps_floor = proto_perl->Itmps_floor;
14950 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14951 * NOTE: unlike the others! */
14952 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14953 PL_scopestack_max = proto_perl->Iscopestack_max;
14955 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14956 * NOTE: unlike the others! */
14957 PL_savestack_ix = proto_perl->Isavestack_ix;
14958 PL_savestack_max = proto_perl->Isavestack_max;
14961 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14962 PL_top_env = &PL_start_env;
14964 PL_op = proto_perl->Iop;
14967 PL_Xpv = (XPV*)NULL;
14968 my_perl->Ina = proto_perl->Ina;
14970 PL_statbuf = proto_perl->Istatbuf;
14971 PL_statcache = proto_perl->Istatcache;
14973 #ifndef NO_TAINT_SUPPORT
14974 PL_tainted = proto_perl->Itainted;
14976 PL_tainted = FALSE;
14978 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14980 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14982 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14983 PL_restartop = proto_perl->Irestartop;
14984 PL_in_eval = proto_perl->Iin_eval;
14985 PL_delaymagic = proto_perl->Idelaymagic;
14986 PL_phase = proto_perl->Iphase;
14987 PL_localizing = proto_perl->Ilocalizing;
14989 PL_hv_fetch_ent_mh = NULL;
14990 PL_modcount = proto_perl->Imodcount;
14991 PL_lastgotoprobe = NULL;
14992 PL_dumpindent = proto_perl->Idumpindent;
14994 PL_efloatbuf = NULL; /* reinits on demand */
14995 PL_efloatsize = 0; /* reinits on demand */
14999 PL_colorset = 0; /* reinits PL_colors[] */
15000 /*PL_colors[6] = {0,0,0,0,0,0};*/
15002 /* Pluggable optimizer */
15003 PL_peepp = proto_perl->Ipeepp;
15004 PL_rpeepp = proto_perl->Irpeepp;
15005 /* op_free() hook */
15006 PL_opfreehook = proto_perl->Iopfreehook;
15008 #ifdef USE_REENTRANT_API
15009 /* XXX: things like -Dm will segfault here in perlio, but doing
15010 * PERL_SET_CONTEXT(proto_perl);
15011 * breaks too many other things
15013 Perl_reentrant_init(aTHX);
15016 /* create SV map for pointer relocation */
15017 PL_ptr_table = ptr_table_new();
15019 /* initialize these special pointers as early as possible */
15021 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15022 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15023 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15024 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15025 &PL_padname_const);
15027 /* create (a non-shared!) shared string table */
15028 PL_strtab = newHV();
15029 HvSHAREKEYS_off(PL_strtab);
15030 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15031 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15033 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15035 /* This PV will be free'd special way so must set it same way op.c does */
15036 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15037 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15039 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15040 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15041 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15042 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15044 param->stashes = newAV(); /* Setup array of objects to call clone on */
15045 /* This makes no difference to the implementation, as it always pushes
15046 and shifts pointers to other SVs without changing their reference
15047 count, with the array becoming empty before it is freed. However, it
15048 makes it conceptually clear what is going on, and will avoid some
15049 work inside av.c, filling slots between AvFILL() and AvMAX() with
15050 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15051 AvREAL_off(param->stashes);
15053 if (!(flags & CLONEf_COPY_STACKS)) {
15054 param->unreferenced = newAV();
15057 #ifdef PERLIO_LAYERS
15058 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15059 PerlIO_clone(aTHX_ proto_perl, param);
15062 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15063 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15064 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15065 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15066 PL_xsubfilename = proto_perl->Ixsubfilename;
15067 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15068 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15071 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15072 PL_inplace = SAVEPV(proto_perl->Iinplace);
15073 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15075 /* magical thingies */
15077 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
15078 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
15079 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
15082 /* Clone the regex array */
15083 /* ORANGE FIXME for plugins, probably in the SV dup code.
15084 newSViv(PTR2IV(CALLREGDUPE(
15085 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15087 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15088 PL_regex_pad = AvARRAY(PL_regex_padav);
15090 PL_stashpadmax = proto_perl->Istashpadmax;
15091 PL_stashpadix = proto_perl->Istashpadix ;
15092 Newx(PL_stashpad, PL_stashpadmax, HV *);
15095 for (; o < PL_stashpadmax; ++o)
15096 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15099 /* shortcuts to various I/O objects */
15100 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15101 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15102 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15103 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15104 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15105 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15106 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15108 /* shortcuts to regexp stuff */
15109 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15111 /* shortcuts to misc objects */
15112 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15114 /* shortcuts to debugging objects */
15115 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15116 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15117 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15118 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15119 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15120 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15121 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15123 /* symbol tables */
15124 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15125 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15126 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15127 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15128 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15130 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15131 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15132 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15133 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15134 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15135 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15136 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15137 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15138 PL_savebegin = proto_perl->Isavebegin;
15140 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15142 /* subprocess state */
15143 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15145 if (proto_perl->Iop_mask)
15146 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15149 /* PL_asserting = proto_perl->Iasserting; */
15151 /* current interpreter roots */
15152 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15154 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15157 /* runtime control stuff */
15158 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15160 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15162 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15164 /* interpreter atexit processing */
15165 PL_exitlistlen = proto_perl->Iexitlistlen;
15166 if (PL_exitlistlen) {
15167 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15168 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15171 PL_exitlist = (PerlExitListEntry*)NULL;
15173 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15174 if (PL_my_cxt_size) {
15175 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15176 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15177 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15178 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15179 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15183 PL_my_cxt_list = (void**)NULL;
15184 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15185 PL_my_cxt_keys = (const char**)NULL;
15188 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15189 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15190 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15191 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15193 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15195 PAD_CLONE_VARS(proto_perl, param);
15197 #ifdef HAVE_INTERP_INTERN
15198 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15201 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15203 #ifdef PERL_USES_PL_PIDSTATUS
15204 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15206 PL_osname = SAVEPV(proto_perl->Iosname);
15207 PL_parser = parser_dup(proto_perl->Iparser, param);
15209 /* XXX this only works if the saved cop has already been cloned */
15210 if (proto_perl->Iparser) {
15211 PL_parser->saved_curcop = (COP*)any_dup(
15212 proto_perl->Iparser->saved_curcop,
15216 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15218 #ifdef USE_LOCALE_CTYPE
15219 /* Should we warn if uses locale? */
15220 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15223 #ifdef USE_LOCALE_COLLATE
15224 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15225 #endif /* USE_LOCALE_COLLATE */
15227 #ifdef USE_LOCALE_NUMERIC
15228 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15229 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15230 #endif /* !USE_LOCALE_NUMERIC */
15232 /* Unicode inversion lists */
15233 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15234 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15235 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15236 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15238 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15239 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15241 /* utf8 character class swashes */
15242 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15243 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15245 for (i = 0; i < POSIX_CC_COUNT; i++) {
15246 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15248 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15249 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15250 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15251 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15252 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15253 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15254 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15255 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15256 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15257 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15258 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15259 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15260 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15261 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15262 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15263 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15264 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15266 if (proto_perl->Ipsig_pend) {
15267 Newxz(PL_psig_pend, SIG_SIZE, int);
15270 PL_psig_pend = (int*)NULL;
15273 if (proto_perl->Ipsig_name) {
15274 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15275 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15277 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15280 PL_psig_ptr = (SV**)NULL;
15281 PL_psig_name = (SV**)NULL;
15284 if (flags & CLONEf_COPY_STACKS) {
15285 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15286 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15287 PL_tmps_ix+1, param);
15289 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15290 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15291 Newxz(PL_markstack, i, I32);
15292 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15293 - proto_perl->Imarkstack);
15294 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15295 - proto_perl->Imarkstack);
15296 Copy(proto_perl->Imarkstack, PL_markstack,
15297 PL_markstack_ptr - PL_markstack + 1, I32);
15299 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15300 * NOTE: unlike the others! */
15301 Newxz(PL_scopestack, PL_scopestack_max, I32);
15302 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15305 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15306 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15308 /* reset stack AV to correct length before its duped via
15309 * PL_curstackinfo */
15310 AvFILLp(proto_perl->Icurstack) =
15311 proto_perl->Istack_sp - proto_perl->Istack_base;
15313 /* NOTE: si_dup() looks at PL_markstack */
15314 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15316 /* PL_curstack = PL_curstackinfo->si_stack; */
15317 PL_curstack = av_dup(proto_perl->Icurstack, param);
15318 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15320 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15321 PL_stack_base = AvARRAY(PL_curstack);
15322 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15323 - proto_perl->Istack_base);
15324 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15326 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15327 PL_savestack = ss_dup(proto_perl, param);
15331 ENTER; /* perl_destruct() wants to LEAVE; */
15334 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15335 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15337 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15338 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15339 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15340 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15341 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15342 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15344 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15346 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15347 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15348 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15350 PL_stashcache = newHV();
15352 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15353 proto_perl->Iwatchaddr);
15354 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15355 if (PL_debug && PL_watchaddr) {
15356 PerlIO_printf(Perl_debug_log,
15357 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15358 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15359 PTR2UV(PL_watchok));
15362 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15363 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15364 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15366 /* Call the ->CLONE method, if it exists, for each of the stashes
15367 identified by sv_dup() above.
15369 while(av_tindex(param->stashes) != -1) {
15370 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15371 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15372 if (cloner && GvCV(cloner)) {
15377 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15379 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15385 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15386 ptr_table_free(PL_ptr_table);
15387 PL_ptr_table = NULL;
15390 if (!(flags & CLONEf_COPY_STACKS)) {
15391 unreferenced_to_tmp_stack(param->unreferenced);
15394 SvREFCNT_dec(param->stashes);
15396 /* orphaned? eg threads->new inside BEGIN or use */
15397 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15398 SvREFCNT_inc_simple_void(PL_compcv);
15399 SAVEFREESV(PL_compcv);
15406 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15408 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15410 if (AvFILLp(unreferenced) > -1) {
15411 SV **svp = AvARRAY(unreferenced);
15412 SV **const last = svp + AvFILLp(unreferenced);
15416 if (SvREFCNT(*svp) == 1)
15418 } while (++svp <= last);
15420 EXTEND_MORTAL(count);
15421 svp = AvARRAY(unreferenced);
15424 if (SvREFCNT(*svp) == 1) {
15425 /* Our reference is the only one to this SV. This means that
15426 in this thread, the scalar effectively has a 0 reference.
15427 That doesn't work (cleanup never happens), so donate our
15428 reference to it onto the save stack. */
15429 PL_tmps_stack[++PL_tmps_ix] = *svp;
15431 /* As an optimisation, because we are already walking the
15432 entire array, instead of above doing either
15433 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15434 release our reference to the scalar, so that at the end of
15435 the array owns zero references to the scalars it happens to
15436 point to. We are effectively converting the array from
15437 AvREAL() on to AvREAL() off. This saves the av_clear()
15438 (triggered by the SvREFCNT_dec(unreferenced) below) from
15439 walking the array a second time. */
15440 SvREFCNT_dec(*svp);
15443 } while (++svp <= last);
15444 AvREAL_off(unreferenced);
15446 SvREFCNT_dec_NN(unreferenced);
15450 Perl_clone_params_del(CLONE_PARAMS *param)
15452 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15454 PerlInterpreter *const to = param->new_perl;
15456 PerlInterpreter *const was = PERL_GET_THX;
15458 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15464 SvREFCNT_dec(param->stashes);
15465 if (param->unreferenced)
15466 unreferenced_to_tmp_stack(param->unreferenced);
15476 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15479 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15480 does a dTHX; to get the context from thread local storage.
15481 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15482 a version that passes in my_perl. */
15483 PerlInterpreter *const was = PERL_GET_THX;
15484 CLONE_PARAMS *param;
15486 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15492 /* Given that we've set the context, we can do this unshared. */
15493 Newx(param, 1, CLONE_PARAMS);
15496 param->proto_perl = from;
15497 param->new_perl = to;
15498 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15499 AvREAL_off(param->stashes);
15500 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15508 #endif /* USE_ITHREADS */
15511 Perl_init_constants(pTHX)
15513 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15514 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15515 SvANY(&PL_sv_undef) = NULL;
15517 SvANY(&PL_sv_no) = new_XPVNV();
15518 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15519 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15520 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15523 SvANY(&PL_sv_yes) = new_XPVNV();
15524 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15525 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15526 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15529 SvPV_set(&PL_sv_no, (char*)PL_No);
15530 SvCUR_set(&PL_sv_no, 0);
15531 SvLEN_set(&PL_sv_no, 0);
15532 SvIV_set(&PL_sv_no, 0);
15533 SvNV_set(&PL_sv_no, 0);
15535 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15536 SvCUR_set(&PL_sv_yes, 1);
15537 SvLEN_set(&PL_sv_yes, 0);
15538 SvIV_set(&PL_sv_yes, 1);
15539 SvNV_set(&PL_sv_yes, 1);
15541 PadnamePV(&PL_padname_const) = (char *)PL_No;
15545 =head1 Unicode Support
15547 =for apidoc sv_recode_to_utf8
15549 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15550 of C<sv> is assumed to be octets in that encoding, and C<sv>
15551 will be converted into Unicode (and UTF-8).
15553 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15554 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15555 an C<Encode::XS> Encoding object, bad things will happen.
15556 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15558 The PV of C<sv> is returned.
15563 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15565 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15567 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15576 if (SvPADTMP(nsv)) {
15577 nsv = sv_newmortal();
15578 SvSetSV_nosteal(nsv, sv);
15587 Passing sv_yes is wrong - it needs to be or'ed set of constants
15588 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15589 remove converted chars from source.
15591 Both will default the value - let them.
15593 XPUSHs(&PL_sv_yes);
15596 call_method("decode", G_SCALAR);
15600 s = SvPV_const(uni, len);
15601 if (s != SvPVX_const(sv)) {
15602 SvGROW(sv, len + 1);
15603 Move(s, SvPVX(sv), len + 1, char);
15604 SvCUR_set(sv, len);
15609 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15610 /* clear pos and any utf8 cache */
15611 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15614 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15615 magic_setutf8(sv,mg); /* clear UTF8 cache */
15620 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15624 =for apidoc sv_cat_decode
15626 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15627 assumed to be octets in that encoding and decoding the input starts
15628 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15629 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15630 when the string C<tstr> appears in decoding output or the input ends on
15631 the PV of C<ssv>. The value which C<offset> points will be modified
15632 to the last input position on C<ssv>.
15634 Returns TRUE if the terminator was found, else returns FALSE.
15639 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15640 SV *ssv, int *offset, char *tstr, int tlen)
15644 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15646 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15657 offsv = newSViv(*offset);
15659 mPUSHp(tstr, tlen);
15661 call_method("cat_decode", G_SCALAR);
15663 ret = SvTRUE(TOPs);
15664 *offset = SvIV(offsv);
15670 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15675 /* ---------------------------------------------------------------------
15677 * support functions for report_uninit()
15680 /* the maxiumum size of array or hash where we will scan looking
15681 * for the undefined element that triggered the warning */
15683 #define FUV_MAX_SEARCH_SIZE 1000
15685 /* Look for an entry in the hash whose value has the same SV as val;
15686 * If so, return a mortal copy of the key. */
15689 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15695 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15697 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15698 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15701 array = HvARRAY(hv);
15703 for (i=HvMAX(hv); i>=0; i--) {
15705 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15706 if (HeVAL(entry) != val)
15708 if ( HeVAL(entry) == &PL_sv_undef ||
15709 HeVAL(entry) == &PL_sv_placeholder)
15713 if (HeKLEN(entry) == HEf_SVKEY)
15714 return sv_mortalcopy(HeKEY_sv(entry));
15715 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15721 /* Look for an entry in the array whose value has the same SV as val;
15722 * If so, return the index, otherwise return -1. */
15725 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15727 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15729 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15730 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15733 if (val != &PL_sv_undef) {
15734 SV ** const svp = AvARRAY(av);
15737 for (i=AvFILLp(av); i>=0; i--)
15744 /* varname(): return the name of a variable, optionally with a subscript.
15745 * If gv is non-zero, use the name of that global, along with gvtype (one
15746 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15747 * targ. Depending on the value of the subscript_type flag, return:
15750 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15751 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15752 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15753 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15756 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15757 const SV *const keyname, SSize_t aindex, int subscript_type)
15760 SV * const name = sv_newmortal();
15761 if (gv && isGV(gv)) {
15763 buffer[0] = gvtype;
15766 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15768 gv_fullname4(name, gv, buffer, 0);
15770 if ((unsigned int)SvPVX(name)[1] <= 26) {
15772 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15774 /* Swap the 1 unprintable control character for the 2 byte pretty
15775 version - ie substr($name, 1, 1) = $buffer; */
15776 sv_insert(name, 1, 1, buffer, 2);
15780 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15783 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15785 if (!cv || !CvPADLIST(cv))
15787 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15788 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15792 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15793 SV * const sv = newSV(0);
15795 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15797 *SvPVX(name) = '$';
15798 Perl_sv_catpvf(aTHX_ name, "{%s}",
15799 pv_pretty(sv, pv, len, 32, NULL, NULL,
15800 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15801 SvREFCNT_dec_NN(sv);
15803 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15804 *SvPVX(name) = '$';
15805 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15807 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15808 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15809 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15817 =for apidoc find_uninit_var
15819 Find the name of the undefined variable (if any) that caused the operator
15820 to issue a "Use of uninitialized value" warning.
15821 If match is true, only return a name if its value matches C<uninit_sv>.
15822 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15823 warning, then following the direct child of the op may yield an
15824 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15825 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15826 the variable name if we get an exact match.
15827 C<desc_p> points to a string pointer holding the description of the op.
15828 This may be updated if needed.
15830 The name is returned as a mortal SV.
15832 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15833 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15839 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15840 bool match, const char **desc_p)
15845 const OP *o, *o2, *kid;
15847 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15849 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15850 uninit_sv == &PL_sv_placeholder)))
15853 switch (obase->op_type) {
15856 /* undef should care if its args are undef - any warnings
15857 * will be from tied/magic vars */
15865 const bool pad = ( obase->op_type == OP_PADAV
15866 || obase->op_type == OP_PADHV
15867 || obase->op_type == OP_PADRANGE
15870 const bool hash = ( obase->op_type == OP_PADHV
15871 || obase->op_type == OP_RV2HV
15872 || (obase->op_type == OP_PADRANGE
15873 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15877 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15879 if (pad) { /* @lex, %lex */
15880 sv = PAD_SVl(obase->op_targ);
15884 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15885 /* @global, %global */
15886 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15889 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15891 else if (obase == PL_op) /* @{expr}, %{expr} */
15892 return find_uninit_var(cUNOPx(obase)->op_first,
15893 uninit_sv, match, desc_p);
15894 else /* @{expr}, %{expr} as a sub-expression */
15898 /* attempt to find a match within the aggregate */
15900 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15902 subscript_type = FUV_SUBSCRIPT_HASH;
15905 index = find_array_subscript((const AV *)sv, uninit_sv);
15907 subscript_type = FUV_SUBSCRIPT_ARRAY;
15910 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15913 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15914 keysv, index, subscript_type);
15918 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15920 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15921 if (!gv || !GvSTASH(gv))
15923 if (match && (GvSV(gv) != uninit_sv))
15925 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15928 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15931 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15933 return varname(NULL, '$', obase->op_targ,
15934 NULL, 0, FUV_SUBSCRIPT_NONE);
15937 gv = cGVOPx_gv(obase);
15938 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15940 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15942 case OP_AELEMFAST_LEX:
15945 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15946 if (!av || SvRMAGICAL(av))
15948 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15949 if (!svp || *svp != uninit_sv)
15952 return varname(NULL, '$', obase->op_targ,
15953 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15956 gv = cGVOPx_gv(obase);
15961 AV *const av = GvAV(gv);
15962 if (!av || SvRMAGICAL(av))
15964 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15965 if (!svp || *svp != uninit_sv)
15968 return varname(gv, '$', 0,
15969 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15971 NOT_REACHED; /* NOTREACHED */
15974 o = cUNOPx(obase)->op_first;
15975 if (!o || o->op_type != OP_NULL ||
15976 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15978 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15983 bool negate = FALSE;
15985 if (PL_op == obase)
15986 /* $a[uninit_expr] or $h{uninit_expr} */
15987 return find_uninit_var(cBINOPx(obase)->op_last,
15988 uninit_sv, match, desc_p);
15991 o = cBINOPx(obase)->op_first;
15992 kid = cBINOPx(obase)->op_last;
15994 /* get the av or hv, and optionally the gv */
15996 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15997 sv = PAD_SV(o->op_targ);
15999 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16000 && cUNOPo->op_first->op_type == OP_GV)
16002 gv = cGVOPx_gv(cUNOPo->op_first);
16006 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16011 if (kid && kid->op_type == OP_NEGATE) {
16013 kid = cUNOPx(kid)->op_first;
16016 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16017 /* index is constant */
16020 kidsv = newSVpvs_flags("-", SVs_TEMP);
16021 sv_catsv(kidsv, cSVOPx_sv(kid));
16024 kidsv = cSVOPx_sv(kid);
16028 if (obase->op_type == OP_HELEM) {
16029 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16030 if (!he || HeVAL(he) != uninit_sv)
16034 SV * const opsv = cSVOPx_sv(kid);
16035 const IV opsviv = SvIV(opsv);
16036 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16037 negate ? - opsviv : opsviv,
16039 if (!svp || *svp != uninit_sv)
16043 if (obase->op_type == OP_HELEM)
16044 return varname(gv, '%', o->op_targ,
16045 kidsv, 0, FUV_SUBSCRIPT_HASH);
16047 return varname(gv, '@', o->op_targ, NULL,
16048 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16049 FUV_SUBSCRIPT_ARRAY);
16052 /* index is an expression;
16053 * attempt to find a match within the aggregate */
16054 if (obase->op_type == OP_HELEM) {
16055 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16057 return varname(gv, '%', o->op_targ,
16058 keysv, 0, FUV_SUBSCRIPT_HASH);
16061 const SSize_t index
16062 = find_array_subscript((const AV *)sv, uninit_sv);
16064 return varname(gv, '@', o->op_targ,
16065 NULL, index, FUV_SUBSCRIPT_ARRAY);
16070 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16072 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16074 NOT_REACHED; /* NOTREACHED */
16077 case OP_MULTIDEREF: {
16078 /* If we were executing OP_MULTIDEREF when the undef warning
16079 * triggered, then it must be one of the index values within
16080 * that triggered it. If not, then the only possibility is that
16081 * the value retrieved by the last aggregate index might be the
16082 * culprit. For the former, we set PL_multideref_pc each time before
16083 * using an index, so work though the item list until we reach
16084 * that point. For the latter, just work through the entire item
16085 * list; the last aggregate retrieved will be the candidate.
16086 * There is a third rare possibility: something triggered
16087 * magic while fetching an array/hash element. Just display
16088 * nothing in this case.
16091 /* the named aggregate, if any */
16092 PADOFFSET agg_targ = 0;
16094 /* the last-seen index */
16096 PADOFFSET index_targ;
16098 IV index_const_iv = 0; /* init for spurious compiler warn */
16099 SV *index_const_sv;
16100 int depth = 0; /* how many array/hash lookups we've done */
16102 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16103 UNOP_AUX_item *last = NULL;
16104 UV actions = items->uv;
16107 if (PL_op == obase) {
16108 last = PL_multideref_pc;
16109 assert(last >= items && last <= items + items[-1].uv);
16116 switch (actions & MDEREF_ACTION_MASK) {
16118 case MDEREF_reload:
16119 actions = (++items)->uv;
16122 case MDEREF_HV_padhv_helem: /* $lex{...} */
16125 case MDEREF_AV_padav_aelem: /* $lex[...] */
16126 agg_targ = (++items)->pad_offset;
16130 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16133 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16135 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16136 assert(isGV_with_GP(agg_gv));
16139 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16140 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16143 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16144 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16150 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16151 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16154 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16155 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16162 index_const_sv = NULL;
16164 index_type = (actions & MDEREF_INDEX_MASK);
16165 switch (index_type) {
16166 case MDEREF_INDEX_none:
16168 case MDEREF_INDEX_const:
16170 index_const_sv = UNOP_AUX_item_sv(++items)
16172 index_const_iv = (++items)->iv;
16174 case MDEREF_INDEX_padsv:
16175 index_targ = (++items)->pad_offset;
16177 case MDEREF_INDEX_gvsv:
16178 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16179 assert(isGV_with_GP(index_gv));
16183 if (index_type != MDEREF_INDEX_none)
16186 if ( index_type == MDEREF_INDEX_none
16187 || (actions & MDEREF_FLAG_last)
16188 || (last && items >= last)
16192 actions >>= MDEREF_SHIFT;
16195 if (PL_op == obase) {
16196 /* most likely index was undef */
16198 *desc_p = ( (actions & MDEREF_FLAG_last)
16199 && (obase->op_private
16200 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16202 (obase->op_private & OPpMULTIDEREF_EXISTS)
16205 : is_hv ? "hash element" : "array element";
16206 assert(index_type != MDEREF_INDEX_none);
16208 if (GvSV(index_gv) == uninit_sv)
16209 return varname(index_gv, '$', 0, NULL, 0,
16210 FUV_SUBSCRIPT_NONE);
16215 if (PL_curpad[index_targ] == uninit_sv)
16216 return varname(NULL, '$', index_targ,
16217 NULL, 0, FUV_SUBSCRIPT_NONE);
16221 /* If we got to this point it was undef on a const subscript,
16222 * so magic probably involved, e.g. $ISA[0]. Give up. */
16226 /* the SV returned by pp_multideref() was undef, if anything was */
16232 sv = PAD_SV(agg_targ);
16234 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16238 if (index_type == MDEREF_INDEX_const) {
16243 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16244 if (!he || HeVAL(he) != uninit_sv)
16248 SV * const * const svp =
16249 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16250 if (!svp || *svp != uninit_sv)
16255 ? varname(agg_gv, '%', agg_targ,
16256 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16257 : varname(agg_gv, '@', agg_targ,
16258 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16261 /* index is an var */
16263 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16265 return varname(agg_gv, '%', agg_targ,
16266 keysv, 0, FUV_SUBSCRIPT_HASH);
16269 const SSize_t index
16270 = find_array_subscript((const AV *)sv, uninit_sv);
16272 return varname(agg_gv, '@', agg_targ,
16273 NULL, index, FUV_SUBSCRIPT_ARRAY);
16277 return varname(agg_gv,
16279 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16281 NOT_REACHED; /* NOTREACHED */
16285 /* only examine RHS */
16286 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16290 o = cUNOPx(obase)->op_first;
16291 if ( o->op_type == OP_PUSHMARK
16292 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16296 if (!OpHAS_SIBLING(o)) {
16297 /* one-arg version of open is highly magical */
16299 if (o->op_type == OP_GV) { /* open FOO; */
16301 if (match && GvSV(gv) != uninit_sv)
16303 return varname(gv, '$', 0,
16304 NULL, 0, FUV_SUBSCRIPT_NONE);
16306 /* other possibilities not handled are:
16307 * open $x; or open my $x; should return '${*$x}'
16308 * open expr; should return '$'.expr ideally
16315 /* ops where $_ may be an implicit arg */
16320 if ( !(obase->op_flags & OPf_STACKED)) {
16321 if (uninit_sv == DEFSV)
16322 return newSVpvs_flags("$_", SVs_TEMP);
16323 else if (obase->op_targ
16324 && uninit_sv == PAD_SVl(obase->op_targ))
16325 return varname(NULL, '$', obase->op_targ, NULL, 0,
16326 FUV_SUBSCRIPT_NONE);
16333 match = 1; /* print etc can return undef on defined args */
16334 /* skip filehandle as it can't produce 'undef' warning */
16335 o = cUNOPx(obase)->op_first;
16336 if ((obase->op_flags & OPf_STACKED)
16338 ( o->op_type == OP_PUSHMARK
16339 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16340 o = OpSIBLING(OpSIBLING(o));
16344 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16345 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16347 /* the following ops are capable of returning PL_sv_undef even for
16348 * defined arg(s) */
16367 case OP_GETPEERNAME:
16414 case OP_SMARTMATCH:
16423 /* XXX tmp hack: these two may call an XS sub, and currently
16424 XS subs don't have a SUB entry on the context stack, so CV and
16425 pad determination goes wrong, and BAD things happen. So, just
16426 don't try to determine the value under those circumstances.
16427 Need a better fix at dome point. DAPM 11/2007 */
16433 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16434 if (gv && GvSV(gv) == uninit_sv)
16435 return newSVpvs_flags("$.", SVs_TEMP);
16440 /* def-ness of rval pos() is independent of the def-ness of its arg */
16441 if ( !(obase->op_flags & OPf_MOD))
16446 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16447 return newSVpvs_flags("${$/}", SVs_TEMP);
16452 if (!(obase->op_flags & OPf_KIDS))
16454 o = cUNOPx(obase)->op_first;
16460 /* This loop checks all the kid ops, skipping any that cannot pos-
16461 * sibly be responsible for the uninitialized value; i.e., defined
16462 * constants and ops that return nothing. If there is only one op
16463 * left that is not skipped, then we *know* it is responsible for
16464 * the uninitialized value. If there is more than one op left, we
16465 * have to look for an exact match in the while() loop below.
16466 * Note that we skip padrange, because the individual pad ops that
16467 * it replaced are still in the tree, so we work on them instead.
16470 for (kid=o; kid; kid = OpSIBLING(kid)) {
16471 const OPCODE type = kid->op_type;
16472 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16473 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16474 || (type == OP_PUSHMARK)
16475 || (type == OP_PADRANGE)
16479 if (o2) { /* more than one found */
16486 return find_uninit_var(o2, uninit_sv, match, desc_p);
16488 /* scan all args */
16490 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16502 =for apidoc report_uninit
16504 Print appropriate "Use of uninitialized variable" warning.
16510 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16512 const char *desc = NULL;
16513 SV* varname = NULL;
16516 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16519 if (uninit_sv && PL_curpad) {
16520 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16522 sv_insert(varname, 0, 0, " ", 1);
16525 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16526 /* we've reached the end of a sort block or sub,
16527 * and the uninit value is probably what that code returned */
16530 /* PL_warn_uninit_sv is constant */
16531 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16533 /* diag_listed_as: Use of uninitialized value%s */
16534 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16535 SVfARG(varname ? varname : &PL_sv_no),
16538 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16544 * ex: set ts=8 sts=4 sw=4 et: