3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%" UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referent = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referent = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referent will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referent;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1531 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533 Use the C<SvGROW> wrapper instead.
1538 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1541 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1545 PERL_ARGS_ASSERT_SV_GROW;
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1553 else if (SvOOK(sv)) { /* pv is offset? */
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1562 s = SvPVX_mutable(sv);
1565 #ifdef PERL_COPY_ON_WRITE
1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1567 * to store the COW count. So in general, allocate one more byte than
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
1571 * Only increment if the allocation isn't MEM_SIZE_MAX,
1572 * otherwise it will wrap to 0.
1574 if ( newlen != MEM_SIZE_MAX )
1578 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1579 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1582 if (newlen > SvLEN(sv)) { /* need more room? */
1583 STRLEN minlen = SvCUR(sv);
1584 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1585 if (newlen < minlen)
1587 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1589 /* Don't round up on the first allocation, as odds are pretty good that
1590 * the initial request is accurate as to what is really needed */
1592 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1593 if (rounded > newlen)
1597 if (SvLEN(sv) && s) {
1598 s = (char*)saferealloc(s, newlen);
1601 s = (char*)safemalloc(newlen);
1602 if (SvPVX_const(sv) && SvCUR(sv)) {
1603 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1607 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1608 /* Do this here, do it once, do it right, and then we will never get
1609 called back into sv_grow() unless there really is some growing
1611 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1613 SvLEN_set(sv, newlen);
1620 =for apidoc sv_setiv
1622 Copies an integer into the given SV, upgrading first if necessary.
1623 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1629 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1631 PERL_ARGS_ASSERT_SV_SETIV;
1633 SV_CHECK_THINKFIRST_COW_DROP(sv);
1634 switch (SvTYPE(sv)) {
1637 sv_upgrade(sv, SVt_IV);
1640 sv_upgrade(sv, SVt_PVIV);
1644 if (!isGV_with_GP(sv))
1651 /* diag_listed_as: Can't coerce %s to %s in %s */
1652 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1657 (void)SvIOK_only(sv); /* validate number */
1663 =for apidoc sv_setiv_mg
1665 Like C<sv_setiv>, but also handles 'set' magic.
1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1680 =for apidoc sv_setuv
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1691 PERL_ARGS_ASSERT_SV_SETUV;
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1714 =for apidoc sv_setuv_mg
1716 Like C<sv_setuv>, but also handles 'set' magic.
1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1731 =for apidoc sv_setnv
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1742 PERL_ARGS_ASSERT_SV_SETNV;
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1748 sv_upgrade(sv, SVt_NV);
1752 sv_upgrade(sv, SVt_PVNV);
1756 if (!isGV_with_GP(sv))
1763 /* diag_listed_as: Can't coerce %s to %s in %s */
1764 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1770 (void)SvNOK_only(sv); /* validate number */
1775 =for apidoc sv_setnv_mg
1777 Like C<sv_setnv>, but also handles 'set' magic.
1783 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1785 PERL_ARGS_ASSERT_SV_SETNV_MG;
1791 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1792 * not incrementable warning display.
1793 * Originally part of S_not_a_number().
1794 * The return value may be != tmpbuf.
1798 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1801 PERL_ARGS_ASSERT_SV_DISPLAY;
1804 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1805 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1808 const char * const limit = tmpbuf + tmpbuf_size - 8;
1809 /* each *s can expand to 4 chars + "...\0",
1810 i.e. need room for 8 chars */
1812 const char *s = SvPVX_const(sv);
1813 const char * const end = s + SvCUR(sv);
1814 for ( ; s < end && d < limit; s++ ) {
1816 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1820 /* Map to ASCII "equivalent" of Latin1 */
1821 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1827 else if (ch == '\r') {
1831 else if (ch == '\f') {
1835 else if (ch == '\\') {
1839 else if (ch == '\0') {
1843 else if (isPRINT_LC(ch))
1862 /* Print an "isn't numeric" warning, using a cleaned-up,
1863 * printable version of the offending string
1867 S_not_a_number(pTHX_ SV *const sv)
1872 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1874 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1877 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1878 /* diag_listed_as: Argument "%s" isn't numeric%s */
1879 "Argument \"%s\" isn't numeric in %s", pv,
1882 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1883 /* diag_listed_as: Argument "%s" isn't numeric%s */
1884 "Argument \"%s\" isn't numeric", pv);
1888 S_not_incrementable(pTHX_ SV *const sv) {
1892 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1894 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1896 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1897 "Argument \"%s\" treated as 0 in increment (++)", pv);
1901 =for apidoc looks_like_number
1903 Test if the content of an SV looks like a number (or is a number).
1904 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1905 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1912 Perl_looks_like_number(pTHX_ SV *const sv)
1918 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1920 if (SvPOK(sv) || SvPOKp(sv)) {
1921 sbegin = SvPV_nomg_const(sv, len);
1924 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1925 numtype = grok_number(sbegin, len, NULL);
1926 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1930 S_glob_2number(pTHX_ GV * const gv)
1932 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1934 /* We know that all GVs stringify to something that is not-a-number,
1935 so no need to test that. */
1936 if (ckWARN(WARN_NUMERIC))
1938 SV *const buffer = sv_newmortal();
1939 gv_efullname3(buffer, gv, "*");
1940 not_a_number(buffer);
1942 /* We just want something true to return, so that S_sv_2iuv_common
1943 can tail call us and return true. */
1947 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1948 until proven guilty, assume that things are not that bad... */
1953 As 64 bit platforms often have an NV that doesn't preserve all bits of
1954 an IV (an assumption perl has been based on to date) it becomes necessary
1955 to remove the assumption that the NV always carries enough precision to
1956 recreate the IV whenever needed, and that the NV is the canonical form.
1957 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1958 precision as a side effect of conversion (which would lead to insanity
1959 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1960 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1961 where precision was lost, and IV/UV/NV slots that have a valid conversion
1962 which has lost no precision
1963 2) to ensure that if a numeric conversion to one form is requested that
1964 would lose precision, the precise conversion (or differently
1965 imprecise conversion) is also performed and cached, to prevent
1966 requests for different numeric formats on the same SV causing
1967 lossy conversion chains. (lossless conversion chains are perfectly
1972 SvIOKp is true if the IV slot contains a valid value
1973 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1974 SvNOKp is true if the NV slot contains a valid value
1975 SvNOK is true only if the NV value is accurate
1978 while converting from PV to NV, check to see if converting that NV to an
1979 IV(or UV) would lose accuracy over a direct conversion from PV to
1980 IV(or UV). If it would, cache both conversions, return NV, but mark
1981 SV as IOK NOKp (ie not NOK).
1983 While converting from PV to IV, check to see if converting that IV to an
1984 NV would lose accuracy over a direct conversion from PV to NV. If it
1985 would, cache both conversions, flag similarly.
1987 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1988 correctly because if IV & NV were set NV *always* overruled.
1989 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1990 changes - now IV and NV together means that the two are interchangeable:
1991 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1993 The benefit of this is that operations such as pp_add know that if
1994 SvIOK is true for both left and right operands, then integer addition
1995 can be used instead of floating point (for cases where the result won't
1996 overflow). Before, floating point was always used, which could lead to
1997 loss of precision compared with integer addition.
1999 * making IV and NV equal status should make maths accurate on 64 bit
2001 * may speed up maths somewhat if pp_add and friends start to use
2002 integers when possible instead of fp. (Hopefully the overhead in
2003 looking for SvIOK and checking for overflow will not outweigh the
2004 fp to integer speedup)
2005 * will slow down integer operations (callers of SvIV) on "inaccurate"
2006 values, as the change from SvIOK to SvIOKp will cause a call into
2007 sv_2iv each time rather than a macro access direct to the IV slot
2008 * should speed up number->string conversion on integers as IV is
2009 favoured when IV and NV are equally accurate
2011 ####################################################################
2012 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2013 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2014 On the other hand, SvUOK is true iff UV.
2015 ####################################################################
2017 Your mileage will vary depending your CPU's relative fp to integer
2021 #ifndef NV_PRESERVES_UV
2022 # define IS_NUMBER_UNDERFLOW_IV 1
2023 # define IS_NUMBER_UNDERFLOW_UV 2
2024 # define IS_NUMBER_IV_AND_UV 2
2025 # define IS_NUMBER_OVERFLOW_IV 4
2026 # define IS_NUMBER_OVERFLOW_UV 5
2028 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2030 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2032 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2038 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2039 PERL_UNUSED_CONTEXT;
2041 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%" UVxf " NV=%" NVgf " inttype=%" UVXf "\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2042 if (SvNVX(sv) < (NV)IV_MIN) {
2043 (void)SvIOKp_on(sv);
2045 SvIV_set(sv, IV_MIN);
2046 return IS_NUMBER_UNDERFLOW_IV;
2048 if (SvNVX(sv) > (NV)UV_MAX) {
2049 (void)SvIOKp_on(sv);
2052 SvUV_set(sv, UV_MAX);
2053 return IS_NUMBER_OVERFLOW_UV;
2055 (void)SvIOKp_on(sv);
2057 /* Can't use strtol etc to convert this string. (See truth table in
2059 if (SvNVX(sv) <= (UV)IV_MAX) {
2060 SvIV_set(sv, I_V(SvNVX(sv)));
2061 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2062 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2064 /* Integer is imprecise. NOK, IOKp */
2066 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2069 SvUV_set(sv, U_V(SvNVX(sv)));
2070 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2071 if (SvUVX(sv) == UV_MAX) {
2072 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2073 possibly be preserved by NV. Hence, it must be overflow.
2075 return IS_NUMBER_OVERFLOW_UV;
2077 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2079 /* Integer is imprecise. NOK, IOKp */
2081 return IS_NUMBER_OVERFLOW_IV;
2083 #endif /* !NV_PRESERVES_UV*/
2085 /* If numtype is infnan, set the NV of the sv accordingly.
2086 * If numtype is anything else, try setting the NV using Atof(PV). */
2088 # pragma warning(push)
2089 # pragma warning(disable:4756;disable:4056)
2092 S_sv_setnv(pTHX_ SV* sv, int numtype)
2094 bool pok = cBOOL(SvPOK(sv));
2097 if ((numtype & IS_NUMBER_INFINITY)) {
2098 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2103 if ((numtype & IS_NUMBER_NAN)) {
2104 SvNV_set(sv, NV_NAN);
2109 SvNV_set(sv, Atof(SvPVX_const(sv)));
2110 /* Purposefully no true nok here, since we don't want to blow
2111 * away the possible IOK/UV of an existing sv. */
2114 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2116 SvPOK_on(sv); /* PV is okay, though. */
2120 # pragma warning(pop)
2124 S_sv_2iuv_common(pTHX_ SV *const sv)
2126 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2129 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2130 * without also getting a cached IV/UV from it at the same time
2131 * (ie PV->NV conversion should detect loss of accuracy and cache
2132 * IV or UV at same time to avoid this. */
2133 /* IV-over-UV optimisation - choose to cache IV if possible */
2135 if (SvTYPE(sv) == SVt_NV)
2136 sv_upgrade(sv, SVt_PVNV);
2138 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2139 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2140 certainly cast into the IV range at IV_MAX, whereas the correct
2141 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2143 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2144 if (Perl_isnan(SvNVX(sv))) {
2150 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2151 SvIV_set(sv, I_V(SvNVX(sv)));
2152 if (SvNVX(sv) == (NV) SvIVX(sv)
2153 #ifndef NV_PRESERVES_UV
2154 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2155 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2156 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2157 /* Don't flag it as "accurately an integer" if the number
2158 came from a (by definition imprecise) NV operation, and
2159 we're outside the range of NV integer precision */
2163 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2165 /* scalar has trailing garbage, eg "42a" */
2167 DEBUG_c(PerlIO_printf(Perl_debug_log,
2168 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2174 /* IV not precise. No need to convert from PV, as NV
2175 conversion would already have cached IV if it detected
2176 that PV->IV would be better than PV->NV->IV
2177 flags already correct - don't set public IOK. */
2178 DEBUG_c(PerlIO_printf(Perl_debug_log,
2179 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2184 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2185 but the cast (NV)IV_MIN rounds to a the value less (more
2186 negative) than IV_MIN which happens to be equal to SvNVX ??
2187 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2188 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2189 (NV)UVX == NVX are both true, but the values differ. :-(
2190 Hopefully for 2s complement IV_MIN is something like
2191 0x8000000000000000 which will be exact. NWC */
2194 SvUV_set(sv, U_V(SvNVX(sv)));
2196 (SvNVX(sv) == (NV) SvUVX(sv))
2197 #ifndef NV_PRESERVES_UV
2198 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2199 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2200 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2201 /* Don't flag it as "accurately an integer" if the number
2202 came from a (by definition imprecise) NV operation, and
2203 we're outside the range of NV integer precision */
2209 DEBUG_c(PerlIO_printf(Perl_debug_log,
2210 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2216 else if (SvPOKp(sv)) {
2219 const char *s = SvPVX_const(sv);
2220 const STRLEN cur = SvCUR(sv);
2222 /* short-cut for a single digit string like "1" */
2227 if (SvTYPE(sv) < SVt_PVIV)
2228 sv_upgrade(sv, SVt_PVIV);
2230 SvIV_set(sv, (IV)(c - '0'));
2235 numtype = grok_number(s, cur, &value);
2236 /* We want to avoid a possible problem when we cache an IV/ a UV which
2237 may be later translated to an NV, and the resulting NV is not
2238 the same as the direct translation of the initial string
2239 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2240 be careful to ensure that the value with the .456 is around if the
2241 NV value is requested in the future).
2243 This means that if we cache such an IV/a UV, we need to cache the
2244 NV as well. Moreover, we trade speed for space, and do not
2245 cache the NV if we are sure it's not needed.
2248 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2249 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2250 == IS_NUMBER_IN_UV) {
2251 /* It's definitely an integer, only upgrade to PVIV */
2252 if (SvTYPE(sv) < SVt_PVIV)
2253 sv_upgrade(sv, SVt_PVIV);
2255 } else if (SvTYPE(sv) < SVt_PVNV)
2256 sv_upgrade(sv, SVt_PVNV);
2258 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2259 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2261 S_sv_setnv(aTHX_ sv, numtype);
2265 /* If NVs preserve UVs then we only use the UV value if we know that
2266 we aren't going to call atof() below. If NVs don't preserve UVs
2267 then the value returned may have more precision than atof() will
2268 return, even though value isn't perfectly accurate. */
2269 if ((numtype & (IS_NUMBER_IN_UV
2270 #ifdef NV_PRESERVES_UV
2273 )) == IS_NUMBER_IN_UV) {
2274 /* This won't turn off the public IOK flag if it was set above */
2275 (void)SvIOKp_on(sv);
2277 if (!(numtype & IS_NUMBER_NEG)) {
2279 if (value <= (UV)IV_MAX) {
2280 SvIV_set(sv, (IV)value);
2282 /* it didn't overflow, and it was positive. */
2283 SvUV_set(sv, value);
2287 /* 2s complement assumption */
2288 if (value <= (UV)IV_MIN) {
2289 SvIV_set(sv, value == (UV)IV_MIN
2290 ? IV_MIN : -(IV)value);
2292 /* Too negative for an IV. This is a double upgrade, but
2293 I'm assuming it will be rare. */
2294 if (SvTYPE(sv) < SVt_PVNV)
2295 sv_upgrade(sv, SVt_PVNV);
2299 SvNV_set(sv, -(NV)value);
2300 SvIV_set(sv, IV_MIN);
2304 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2305 will be in the previous block to set the IV slot, and the next
2306 block to set the NV slot. So no else here. */
2308 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2309 != IS_NUMBER_IN_UV) {
2310 /* It wasn't an (integer that doesn't overflow the UV). */
2311 S_sv_setnv(aTHX_ sv, numtype);
2313 if (! numtype && ckWARN(WARN_NUMERIC))
2316 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2317 PTR2UV(sv), SvNVX(sv)));
2319 #ifdef NV_PRESERVES_UV
2320 (void)SvIOKp_on(sv);
2322 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2323 if (Perl_isnan(SvNVX(sv))) {
2329 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2330 SvIV_set(sv, I_V(SvNVX(sv)));
2331 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2334 NOOP; /* Integer is imprecise. NOK, IOKp */
2336 /* UV will not work better than IV */
2338 if (SvNVX(sv) > (NV)UV_MAX) {
2340 /* Integer is inaccurate. NOK, IOKp, is UV */
2341 SvUV_set(sv, UV_MAX);
2343 SvUV_set(sv, U_V(SvNVX(sv)));
2344 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2345 NV preservse UV so can do correct comparison. */
2346 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2349 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2354 #else /* NV_PRESERVES_UV */
2355 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2356 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2357 /* The IV/UV slot will have been set from value returned by
2358 grok_number above. The NV slot has just been set using
2361 assert (SvIOKp(sv));
2363 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2364 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2365 /* Small enough to preserve all bits. */
2366 (void)SvIOKp_on(sv);
2368 SvIV_set(sv, I_V(SvNVX(sv)));
2369 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2371 /* Assumption: first non-preserved integer is < IV_MAX,
2372 this NV is in the preserved range, therefore: */
2373 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2375 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%" NVgf " U_V is 0x%" UVxf ", IV_MAX is 0x%" UVxf "\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2379 0 0 already failed to read UV.
2380 0 1 already failed to read UV.
2381 1 0 you won't get here in this case. IV/UV
2382 slot set, public IOK, Atof() unneeded.
2383 1 1 already read UV.
2384 so there's no point in sv_2iuv_non_preserve() attempting
2385 to use atol, strtol, strtoul etc. */
2387 sv_2iuv_non_preserve (sv, numtype);
2389 sv_2iuv_non_preserve (sv);
2393 #endif /* NV_PRESERVES_UV */
2394 /* It might be more code efficient to go through the entire logic above
2395 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2396 gets complex and potentially buggy, so more programmer efficient
2397 to do it this way, by turning off the public flags: */
2399 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2403 if (isGV_with_GP(sv))
2404 return glob_2number(MUTABLE_GV(sv));
2406 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2408 if (SvTYPE(sv) < SVt_IV)
2409 /* Typically the caller expects that sv_any is not NULL now. */
2410 sv_upgrade(sv, SVt_IV);
2411 /* Return 0 from the caller. */
2418 =for apidoc sv_2iv_flags
2420 Return the integer value of an SV, doing any necessary string
2421 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2422 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2428 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2430 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2432 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2433 && SvTYPE(sv) != SVt_PVFM);
2435 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2441 if (flags & SV_SKIP_OVERLOAD)
2443 tmpstr = AMG_CALLunary(sv, numer_amg);
2444 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2445 return SvIV(tmpstr);
2448 return PTR2IV(SvRV(sv));
2451 if (SvVALID(sv) || isREGEXP(sv)) {
2452 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2453 must not let them cache IVs.
2454 In practice they are extremely unlikely to actually get anywhere
2455 accessible by user Perl code - the only way that I'm aware of is when
2456 a constant subroutine which is used as the second argument to index.
2458 Regexps have no SvIVX and SvNVX fields.
2460 assert(isREGEXP(sv) || SvPOKp(sv));
2463 const char * const ptr =
2464 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2466 = grok_number(ptr, SvCUR(sv), &value);
2468 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2469 == IS_NUMBER_IN_UV) {
2470 /* It's definitely an integer */
2471 if (numtype & IS_NUMBER_NEG) {
2472 if (value < (UV)IV_MIN)
2475 if (value < (UV)IV_MAX)
2480 /* Quite wrong but no good choices. */
2481 if ((numtype & IS_NUMBER_INFINITY)) {
2482 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2483 } else if ((numtype & IS_NUMBER_NAN)) {
2484 return 0; /* So wrong. */
2488 if (ckWARN(WARN_NUMERIC))
2491 return I_V(Atof(ptr));
2495 if (SvTHINKFIRST(sv)) {
2496 if (SvREADONLY(sv) && !SvOK(sv)) {
2497 if (ckWARN(WARN_UNINITIALIZED))
2504 if (S_sv_2iuv_common(aTHX_ sv))
2508 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2509 PTR2UV(sv),SvIVX(sv)));
2510 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2514 =for apidoc sv_2uv_flags
2516 Return the unsigned integer value of an SV, doing any necessary string
2517 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2518 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2524 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2526 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2528 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2534 if (flags & SV_SKIP_OVERLOAD)
2536 tmpstr = AMG_CALLunary(sv, numer_amg);
2537 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2538 return SvUV(tmpstr);
2541 return PTR2UV(SvRV(sv));
2544 if (SvVALID(sv) || isREGEXP(sv)) {
2545 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2546 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2547 Regexps have no SvIVX and SvNVX fields. */
2548 assert(isREGEXP(sv) || SvPOKp(sv));
2551 const char * const ptr =
2552 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2554 = grok_number(ptr, SvCUR(sv), &value);
2556 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2557 == IS_NUMBER_IN_UV) {
2558 /* It's definitely an integer */
2559 if (!(numtype & IS_NUMBER_NEG))
2563 /* Quite wrong but no good choices. */
2564 if ((numtype & IS_NUMBER_INFINITY)) {
2565 return UV_MAX; /* So wrong. */
2566 } else if ((numtype & IS_NUMBER_NAN)) {
2567 return 0; /* So wrong. */
2571 if (ckWARN(WARN_NUMERIC))
2574 return U_V(Atof(ptr));
2578 if (SvTHINKFIRST(sv)) {
2579 if (SvREADONLY(sv) && !SvOK(sv)) {
2580 if (ckWARN(WARN_UNINITIALIZED))
2587 if (S_sv_2iuv_common(aTHX_ sv))
2591 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2592 PTR2UV(sv),SvUVX(sv)));
2593 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2597 =for apidoc sv_2nv_flags
2599 Return the num value of an SV, doing any necessary string or integer
2600 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2601 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2607 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2609 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2611 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2612 && SvTYPE(sv) != SVt_PVFM);
2613 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2614 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2615 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2616 Regexps have no SvIVX and SvNVX fields. */
2618 if (flags & SV_GMAGIC)
2622 if (SvPOKp(sv) && !SvIOKp(sv)) {
2623 ptr = SvPVX_const(sv);
2625 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2626 !grok_number(ptr, SvCUR(sv), NULL))
2632 return (NV)SvUVX(sv);
2634 return (NV)SvIVX(sv);
2640 ptr = RX_WRAPPED((REGEXP *)sv);
2643 assert(SvTYPE(sv) >= SVt_PVMG);
2644 /* This falls through to the report_uninit near the end of the
2646 } else if (SvTHINKFIRST(sv)) {
2651 if (flags & SV_SKIP_OVERLOAD)
2653 tmpstr = AMG_CALLunary(sv, numer_amg);
2654 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2655 return SvNV(tmpstr);
2658 return PTR2NV(SvRV(sv));
2660 if (SvREADONLY(sv) && !SvOK(sv)) {
2661 if (ckWARN(WARN_UNINITIALIZED))
2666 if (SvTYPE(sv) < SVt_NV) {
2667 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2668 sv_upgrade(sv, SVt_NV);
2670 STORE_NUMERIC_LOCAL_SET_STANDARD();
2671 PerlIO_printf(Perl_debug_log,
2672 "0x%" UVxf " num(%" NVgf ")\n",
2673 PTR2UV(sv), SvNVX(sv));
2674 RESTORE_NUMERIC_LOCAL();
2677 else if (SvTYPE(sv) < SVt_PVNV)
2678 sv_upgrade(sv, SVt_PVNV);
2683 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2684 #ifdef NV_PRESERVES_UV
2690 /* Only set the public NV OK flag if this NV preserves the IV */
2691 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2693 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2694 : (SvIVX(sv) == I_V(SvNVX(sv))))
2700 else if (SvPOKp(sv)) {
2702 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2703 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2705 #ifdef NV_PRESERVES_UV
2706 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2707 == IS_NUMBER_IN_UV) {
2708 /* It's definitely an integer */
2709 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2711 S_sv_setnv(aTHX_ sv, numtype);
2718 SvNV_set(sv, Atof(SvPVX_const(sv)));
2719 /* Only set the public NV OK flag if this NV preserves the value in
2720 the PV at least as well as an IV/UV would.
2721 Not sure how to do this 100% reliably. */
2722 /* if that shift count is out of range then Configure's test is
2723 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2725 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2726 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2727 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2728 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2729 /* Can't use strtol etc to convert this string, so don't try.
2730 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2733 /* value has been set. It may not be precise. */
2734 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2735 /* 2s complement assumption for (UV)IV_MIN */
2736 SvNOK_on(sv); /* Integer is too negative. */
2741 if (numtype & IS_NUMBER_NEG) {
2742 /* -IV_MIN is undefined, but we should never reach
2743 * this point with both IS_NUMBER_NEG and value ==
2745 assert(value != (UV)IV_MIN);
2746 SvIV_set(sv, -(IV)value);
2747 } else if (value <= (UV)IV_MAX) {
2748 SvIV_set(sv, (IV)value);
2750 SvUV_set(sv, value);
2754 if (numtype & IS_NUMBER_NOT_INT) {
2755 /* I believe that even if the original PV had decimals,
2756 they are lost beyond the limit of the FP precision.
2757 However, neither is canonical, so both only get p
2758 flags. NWC, 2000/11/25 */
2759 /* Both already have p flags, so do nothing */
2761 const NV nv = SvNVX(sv);
2762 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2763 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2764 if (SvIVX(sv) == I_V(nv)) {
2767 /* It had no "." so it must be integer. */
2771 /* between IV_MAX and NV(UV_MAX).
2772 Could be slightly > UV_MAX */
2774 if (numtype & IS_NUMBER_NOT_INT) {
2775 /* UV and NV both imprecise. */
2777 const UV nv_as_uv = U_V(nv);
2779 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2788 /* It might be more code efficient to go through the entire logic above
2789 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2790 gets complex and potentially buggy, so more programmer efficient
2791 to do it this way, by turning off the public flags: */
2793 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2794 #endif /* NV_PRESERVES_UV */
2797 if (isGV_with_GP(sv)) {
2798 glob_2number(MUTABLE_GV(sv));
2802 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2804 assert (SvTYPE(sv) >= SVt_NV);
2805 /* Typically the caller expects that sv_any is not NULL now. */
2806 /* XXX Ilya implies that this is a bug in callers that assume this
2807 and ideally should be fixed. */
2811 STORE_NUMERIC_LOCAL_SET_STANDARD();
2812 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2813 PTR2UV(sv), SvNVX(sv));
2814 RESTORE_NUMERIC_LOCAL();
2822 Return an SV with the numeric value of the source SV, doing any necessary
2823 reference or overload conversion. The caller is expected to have handled
2830 Perl_sv_2num(pTHX_ SV *const sv)
2832 PERL_ARGS_ASSERT_SV_2NUM;
2837 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2838 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2839 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2840 return sv_2num(tmpsv);
2842 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2845 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2846 * UV as a string towards the end of buf, and return pointers to start and
2849 * We assume that buf is at least TYPE_CHARS(UV) long.
2853 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2855 char *ptr = buf + TYPE_CHARS(UV);
2856 char * const ebuf = ptr;
2859 PERL_ARGS_ASSERT_UIV_2BUF;
2867 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2871 *--ptr = '0' + (char)(uv % 10);
2879 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2880 * infinity or a not-a-number, writes the appropriate strings to the
2881 * buffer, including a zero byte. On success returns the written length,
2882 * excluding the zero byte, on failure (not an infinity, not a nan)
2883 * returns zero, assert-fails on maxlen being too short.
2885 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2886 * shared string constants we point to, instead of generating a new
2887 * string for each instance. */
2889 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2891 assert(maxlen >= 4);
2892 if (Perl_isinf(nv)) {
2894 if (maxlen < 5) /* "-Inf\0" */
2904 else if (Perl_isnan(nv)) {
2908 /* XXX optionally output the payload mantissa bits as
2909 * "(unsigned)" (to match the nan("...") C99 function,
2910 * or maybe as "(0xhhh...)" would make more sense...
2911 * provide a format string so that the user can decide?
2912 * NOTE: would affect the maxlen and assert() logic.*/
2917 assert((s == buffer + 3) || (s == buffer + 4));
2919 return s - buffer - 1; /* -1: excluding the zero byte */
2923 =for apidoc sv_2pv_flags
2925 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2926 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2927 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2928 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2934 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2938 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2940 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2941 && SvTYPE(sv) != SVt_PVFM);
2942 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2947 if (flags & SV_SKIP_OVERLOAD)
2949 tmpstr = AMG_CALLunary(sv, string_amg);
2950 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2951 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2953 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2957 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2958 if (flags & SV_CONST_RETURN) {
2959 pv = (char *) SvPVX_const(tmpstr);
2961 pv = (flags & SV_MUTABLE_RETURN)
2962 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2965 *lp = SvCUR(tmpstr);
2967 pv = sv_2pv_flags(tmpstr, lp, flags);
2980 SV *const referent = SvRV(sv);
2984 retval = buffer = savepvn("NULLREF", len);
2985 } else if (SvTYPE(referent) == SVt_REGEXP &&
2986 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2987 amagic_is_enabled(string_amg))) {
2988 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2992 /* If the regex is UTF-8 we want the containing scalar to
2993 have an UTF-8 flag too */
3000 *lp = RX_WRAPLEN(re);
3002 return RX_WRAPPED(re);
3004 const char *const typestr = sv_reftype(referent, 0);
3005 const STRLEN typelen = strlen(typestr);
3006 UV addr = PTR2UV(referent);
3007 const char *stashname = NULL;
3008 STRLEN stashnamelen = 0; /* hush, gcc */
3009 const char *buffer_end;
3011 if (SvOBJECT(referent)) {
3012 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3015 stashname = HEK_KEY(name);
3016 stashnamelen = HEK_LEN(name);
3018 if (HEK_UTF8(name)) {
3024 stashname = "__ANON__";
3027 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3028 + 2 * sizeof(UV) + 2 /* )\0 */;
3030 len = typelen + 3 /* (0x */
3031 + 2 * sizeof(UV) + 2 /* )\0 */;
3034 Newx(buffer, len, char);
3035 buffer_end = retval = buffer + len;
3037 /* Working backwards */
3041 *--retval = PL_hexdigit[addr & 15];
3042 } while (addr >>= 4);
3048 memcpy(retval, typestr, typelen);
3052 retval -= stashnamelen;
3053 memcpy(retval, stashname, stashnamelen);
3055 /* retval may not necessarily have reached the start of the
3057 assert (retval >= buffer);
3059 len = buffer_end - retval - 1; /* -1 for that \0 */
3071 if (flags & SV_MUTABLE_RETURN)
3072 return SvPVX_mutable(sv);
3073 if (flags & SV_CONST_RETURN)
3074 return (char *)SvPVX_const(sv);
3079 /* I'm assuming that if both IV and NV are equally valid then
3080 converting the IV is going to be more efficient */
3081 const U32 isUIOK = SvIsUV(sv);
3082 char buf[TYPE_CHARS(UV)];
3086 if (SvTYPE(sv) < SVt_PVIV)
3087 sv_upgrade(sv, SVt_PVIV);
3088 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3090 /* inlined from sv_setpvn */
3091 s = SvGROW_mutable(sv, len + 1);
3092 Move(ptr, s, len, char);
3097 else if (SvNOK(sv)) {
3098 if (SvTYPE(sv) < SVt_PVNV)
3099 sv_upgrade(sv, SVt_PVNV);
3100 if (SvNVX(sv) == 0.0
3101 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3102 && !Perl_isnan(SvNVX(sv))
3105 s = SvGROW_mutable(sv, 2);
3110 STRLEN size = 5; /* "-Inf\0" */
3112 s = SvGROW_mutable(sv, size);
3113 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3119 /* some Xenix systems wipe out errno here */
3128 5 + /* exponent digits */
3132 s = SvGROW_mutable(sv, size);
3133 #ifndef USE_LOCALE_NUMERIC
3134 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3140 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3141 STORE_LC_NUMERIC_SET_TO_NEEDED();
3143 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3144 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3145 size += SvLEN(PL_numeric_radix_sv) - 1;
3146 s = SvGROW_mutable(sv, size);
3149 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3151 /* If the radix character is UTF-8, and actually is in the
3152 * output, turn on the UTF-8 flag for the scalar */
3154 && SvUTF8(PL_numeric_radix_sv)
3155 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3160 RESTORE_LC_NUMERIC();
3163 /* We don't call SvPOK_on(), because it may come to
3164 * pass that the locale changes so that the
3165 * stringification we just did is no longer correct. We
3166 * will have to re-stringify every time it is needed */
3173 else if (isGV_with_GP(sv)) {
3174 GV *const gv = MUTABLE_GV(sv);
3175 SV *const buffer = sv_newmortal();
3177 gv_efullname3(buffer, gv, "*");
3179 assert(SvPOK(buffer));
3183 *lp = SvCUR(buffer);
3184 return SvPVX(buffer);
3186 else if (isREGEXP(sv)) {
3187 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3188 return RX_WRAPPED((REGEXP *)sv);
3193 if (flags & SV_UNDEF_RETURNS_NULL)
3195 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3197 /* Typically the caller expects that sv_any is not NULL now. */
3198 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3199 sv_upgrade(sv, SVt_PV);
3204 const STRLEN len = s - SvPVX_const(sv);
3209 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3210 PTR2UV(sv),SvPVX_const(sv)));
3211 if (flags & SV_CONST_RETURN)
3212 return (char *)SvPVX_const(sv);
3213 if (flags & SV_MUTABLE_RETURN)
3214 return SvPVX_mutable(sv);
3219 =for apidoc sv_copypv
3221 Copies a stringified representation of the source SV into the
3222 destination SV. Automatically performs any necessary C<mg_get> and
3223 coercion of numeric values into strings. Guaranteed to preserve
3224 C<UTF8> flag even from overloaded objects. Similar in nature to
3225 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3226 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3227 would lose the UTF-8'ness of the PV.
3229 =for apidoc sv_copypv_nomg
3231 Like C<sv_copypv>, but doesn't invoke get magic first.
3233 =for apidoc sv_copypv_flags
3235 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3236 has the C<SV_GMAGIC> bit set.
3242 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3247 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3249 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3250 sv_setpvn(dsv,s,len);
3258 =for apidoc sv_2pvbyte
3260 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3261 to its length. May cause the SV to be downgraded from UTF-8 as a
3264 Usually accessed via the C<SvPVbyte> macro.
3270 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3272 PERL_ARGS_ASSERT_SV_2PVBYTE;
3275 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3276 || isGV_with_GP(sv) || SvROK(sv)) {
3277 SV *sv2 = sv_newmortal();
3278 sv_copypv_nomg(sv2,sv);
3281 sv_utf8_downgrade(sv,0);
3282 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3286 =for apidoc sv_2pvutf8
3288 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3289 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3291 Usually accessed via the C<SvPVutf8> macro.
3297 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3299 PERL_ARGS_ASSERT_SV_2PVUTF8;
3301 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3302 || isGV_with_GP(sv) || SvROK(sv))
3303 sv = sv_mortalcopy(sv);
3306 sv_utf8_upgrade_nomg(sv);
3307 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3312 =for apidoc sv_2bool
3314 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3315 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3316 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3318 =for apidoc sv_2bool_flags
3320 This function is only used by C<sv_true()> and friends, and only if
3321 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3322 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3329 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3331 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3334 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3340 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3341 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3344 if(SvGMAGICAL(sv)) {
3346 goto restart; /* call sv_2bool */
3348 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3349 else if(!SvOK(sv)) {
3352 else if(SvPOK(sv)) {
3353 svb = SvPVXtrue(sv);
3355 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3356 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3357 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3361 goto restart; /* call sv_2bool_nomg */
3366 return SvRV(sv) != 0;
3370 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3371 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3375 =for apidoc sv_utf8_upgrade
3377 Converts the PV of an SV to its UTF-8-encoded form.
3378 Forces the SV to string form if it is not already.
3379 Will C<mg_get> on C<sv> if appropriate.
3380 Always sets the C<SvUTF8> flag to avoid future validity checks even
3381 if the whole string is the same in UTF-8 as not.
3382 Returns the number of bytes in the converted string
3384 This is not a general purpose byte encoding to Unicode interface:
3385 use the Encode extension for that.
3387 =for apidoc sv_utf8_upgrade_nomg
3389 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3391 =for apidoc sv_utf8_upgrade_flags
3393 Converts the PV of an SV to its UTF-8-encoded form.
3394 Forces the SV to string form if it is not already.
3395 Always sets the SvUTF8 flag to avoid future validity checks even
3396 if all the bytes are invariant in UTF-8.
3397 If C<flags> has C<SV_GMAGIC> bit set,
3398 will C<mg_get> on C<sv> if appropriate, else not.
3400 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3401 will expand when converted to UTF-8, and skips the extra work of checking for
3402 that. Typically this flag is used by a routine that has already parsed the
3403 string and found such characters, and passes this information on so that the
3404 work doesn't have to be repeated.
3406 Returns the number of bytes in the converted string.
3408 This is not a general purpose byte encoding to Unicode interface:
3409 use the Encode extension for that.
3411 =for apidoc sv_utf8_upgrade_flags_grow
3413 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3414 the number of unused bytes the string of C<sv> is guaranteed to have free after
3415 it upon return. This allows the caller to reserve extra space that it intends
3416 to fill, to avoid extra grows.
3418 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3419 are implemented in terms of this function.
3421 Returns the number of bytes in the converted string (not including the spares).
3425 (One might think that the calling routine could pass in the position of the
3426 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3427 have to be found again. But that is not the case, because typically when the
3428 caller is likely to use this flag, it won't be calling this routine unless it
3429 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3430 and just use bytes. But some things that do fit into a byte are variants in
3431 utf8, and the caller may not have been keeping track of these.)
3433 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3434 C<NUL> isn't guaranteed due to having other routines do the work in some input
3435 cases, or if the input is already flagged as being in utf8.
3437 The speed of this could perhaps be improved for many cases if someone wanted to
3438 write a fast function that counts the number of variant characters in a string,
3439 especially if it could return the position of the first one.
3444 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3446 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3448 if (sv == &PL_sv_undef)
3450 if (!SvPOK_nog(sv)) {
3452 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3453 (void) sv_2pv_flags(sv,&len, flags);
3455 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3459 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3464 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3469 S_sv_uncow(aTHX_ sv, 0);
3472 if (SvCUR(sv) == 0) {
3473 if (extra) SvGROW(sv, extra);
3474 } else { /* Assume Latin-1/EBCDIC */
3475 /* This function could be much more efficient if we
3476 * had a FLAG in SVs to signal if there are any variant
3477 * chars in the PV. Given that there isn't such a flag
3478 * make the loop as fast as possible (although there are certainly ways
3479 * to speed this up, eg. through vectorization) */
3480 U8 * s = (U8 *) SvPVX_const(sv);
3481 U8 * e = (U8 *) SvEND(sv);
3483 STRLEN two_byte_count = 0;
3485 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3487 /* See if really will need to convert to utf8. We mustn't rely on our
3488 * incoming SV being well formed and having a trailing '\0', as certain
3489 * code in pp_formline can send us partially built SVs. */
3493 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3495 t--; /* t already incremented; re-point to first variant */
3500 /* utf8 conversion not needed because all are invariants. Mark as
3501 * UTF-8 even if no variant - saves scanning loop */
3503 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3508 /* Here, the string should be converted to utf8, either because of an
3509 * input flag (two_byte_count = 0), or because a character that
3510 * requires 2 bytes was found (two_byte_count = 1). t points either to
3511 * the beginning of the string (if we didn't examine anything), or to
3512 * the first variant. In either case, everything from s to t - 1 will
3513 * occupy only 1 byte each on output.
3515 * There are two main ways to convert. One is to create a new string
3516 * and go through the input starting from the beginning, appending each
3517 * converted value onto the new string as we go along. It's probably
3518 * best to allocate enough space in the string for the worst possible
3519 * case rather than possibly running out of space and having to
3520 * reallocate and then copy what we've done so far. Since everything
3521 * from s to t - 1 is invariant, the destination can be initialized
3522 * with these using a fast memory copy
3524 * The other way is to figure out exactly how big the string should be
3525 * by parsing the entire input. Then you don't have to make it big
3526 * enough to handle the worst possible case, and more importantly, if
3527 * the string you already have is large enough, you don't have to
3528 * allocate a new string, you can copy the last character in the input
3529 * string to the final position(s) that will be occupied by the
3530 * converted string and go backwards, stopping at t, since everything
3531 * before that is invariant.
3533 * There are advantages and disadvantages to each method.
3535 * In the first method, we can allocate a new string, do the memory
3536 * copy from the s to t - 1, and then proceed through the rest of the
3537 * string byte-by-byte.
3539 * In the second method, we proceed through the rest of the input
3540 * string just calculating how big the converted string will be. Then
3541 * there are two cases:
3542 * 1) if the string has enough extra space to handle the converted
3543 * value. We go backwards through the string, converting until we
3544 * get to the position we are at now, and then stop. If this
3545 * position is far enough along in the string, this method is
3546 * faster than the other method. If the memory copy were the same
3547 * speed as the byte-by-byte loop, that position would be about
3548 * half-way, as at the half-way mark, parsing to the end and back
3549 * is one complete string's parse, the same amount as starting
3550 * over and going all the way through. Actually, it would be
3551 * somewhat less than half-way, as it's faster to just count bytes
3552 * than to also copy, and we don't have the overhead of allocating
3553 * a new string, changing the scalar to use it, and freeing the
3554 * existing one. But if the memory copy is fast, the break-even
3555 * point is somewhere after half way. The counting loop could be
3556 * sped up by vectorization, etc, to move the break-even point
3557 * further towards the beginning.
3558 * 2) if the string doesn't have enough space to handle the converted
3559 * value. A new string will have to be allocated, and one might
3560 * as well, given that, start from the beginning doing the first
3561 * method. We've spent extra time parsing the string and in
3562 * exchange all we've gotten is that we know precisely how big to
3563 * make the new one. Perl is more optimized for time than space,
3564 * so this case is a loser.
3565 * So what I've decided to do is not use the 2nd method unless it is
3566 * guaranteed that a new string won't have to be allocated, assuming
3567 * the worst case. I also decided not to put any more conditions on it
3568 * than this, for now. It seems likely that, since the worst case is
3569 * twice as big as the unknown portion of the string (plus 1), we won't
3570 * be guaranteed enough space, causing us to go to the first method,
3571 * unless the string is short, or the first variant character is near
3572 * the end of it. In either of these cases, it seems best to use the
3573 * 2nd method. The only circumstance I can think of where this would
3574 * be really slower is if the string had once had much more data in it
3575 * than it does now, but there is still a substantial amount in it */
3578 STRLEN invariant_head = t - s;
3579 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3580 if (SvLEN(sv) < size) {
3582 /* Here, have decided to allocate a new string */
3587 Newx(dst, size, U8);
3589 /* If no known invariants at the beginning of the input string,
3590 * set so starts from there. Otherwise, can use memory copy to
3591 * get up to where we are now, and then start from here */
3593 if (invariant_head == 0) {
3596 Copy(s, dst, invariant_head, char);
3597 d = dst + invariant_head;
3601 append_utf8_from_native_byte(*t, &d);
3605 SvPV_free(sv); /* No longer using pre-existing string */
3606 SvPV_set(sv, (char*)dst);
3607 SvCUR_set(sv, d - dst);
3608 SvLEN_set(sv, size);
3611 /* Here, have decided to get the exact size of the string.
3612 * Currently this happens only when we know that there is
3613 * guaranteed enough space to fit the converted string, so
3614 * don't have to worry about growing. If two_byte_count is 0,
3615 * then t points to the first byte of the string which hasn't
3616 * been examined yet. Otherwise two_byte_count is 1, and t
3617 * points to the first byte in the string that will expand to
3618 * two. Depending on this, start examining at t or 1 after t.
3621 U8 *d = t + two_byte_count;
3624 /* Count up the remaining bytes that expand to two */
3627 const U8 chr = *d++;
3628 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3631 /* The string will expand by just the number of bytes that
3632 * occupy two positions. But we are one afterwards because of
3633 * the increment just above. This is the place to put the
3634 * trailing NUL, and to set the length before we decrement */
3636 d += two_byte_count;
3637 SvCUR_set(sv, d - s);
3641 /* Having decremented d, it points to the position to put the
3642 * very last byte of the expanded string. Go backwards through
3643 * the string, copying and expanding as we go, stopping when we
3644 * get to the part that is invariant the rest of the way down */
3648 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3651 *d-- = UTF8_EIGHT_BIT_LO(*e);
3652 *d-- = UTF8_EIGHT_BIT_HI(*e);
3658 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3659 /* Update pos. We do it at the end rather than during
3660 * the upgrade, to avoid slowing down the common case
3661 * (upgrade without pos).
3662 * pos can be stored as either bytes or characters. Since
3663 * this was previously a byte string we can just turn off
3664 * the bytes flag. */
3665 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3667 mg->mg_flags &= ~MGf_BYTES;
3669 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3670 magic_setutf8(sv,mg); /* clear UTF8 cache */
3675 /* Mark as UTF-8 even if no variant - saves scanning loop */
3681 =for apidoc sv_utf8_downgrade
3683 Attempts to convert the PV of an SV from characters to bytes.
3684 If the PV contains a character that cannot fit
3685 in a byte, this conversion will fail;
3686 in this case, either returns false or, if C<fail_ok> is not
3689 This is not a general purpose Unicode to byte encoding interface:
3690 use the C<Encode> extension for that.
3696 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3698 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3700 if (SvPOKp(sv) && SvUTF8(sv)) {
3704 int mg_flags = SV_GMAGIC;
3707 S_sv_uncow(aTHX_ sv, 0);
3709 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3711 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3712 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3713 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3714 SV_GMAGIC|SV_CONST_RETURN);
3715 mg_flags = 0; /* sv_pos_b2u does get magic */
3717 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3718 magic_setutf8(sv,mg); /* clear UTF8 cache */
3721 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3723 if (!utf8_to_bytes(s, &len)) {
3728 Perl_croak(aTHX_ "Wide character in %s",
3731 Perl_croak(aTHX_ "Wide character");
3742 =for apidoc sv_utf8_encode
3744 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3745 flag off so that it looks like octets again.
3751 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3753 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3755 if (SvREADONLY(sv)) {
3756 sv_force_normal_flags(sv, 0);
3758 (void) sv_utf8_upgrade(sv);
3763 =for apidoc sv_utf8_decode
3765 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3766 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3767 so that it looks like a character. If the PV contains only single-byte
3768 characters, the C<SvUTF8> flag stays off.
3769 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3775 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3777 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3780 const U8 *start, *c;
3782 /* The octets may have got themselves encoded - get them back as
3785 if (!sv_utf8_downgrade(sv, TRUE))
3788 /* it is actually just a matter of turning the utf8 flag on, but
3789 * we want to make sure everything inside is valid utf8 first.
3791 c = start = (const U8 *) SvPVX_const(sv);
3792 if (!is_utf8_string(c, SvCUR(sv)))
3794 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3797 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3798 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3799 after this, clearing pos. Does anything on CPAN
3801 /* adjust pos to the start of a UTF8 char sequence */
3802 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3804 I32 pos = mg->mg_len;
3806 for (c = start + pos; c > start; c--) {
3807 if (UTF8_IS_START(*c))
3810 mg->mg_len = c - start;
3813 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3814 magic_setutf8(sv,mg); /* clear UTF8 cache */
3821 =for apidoc sv_setsv
3823 Copies the contents of the source SV C<ssv> into the destination SV
3824 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3825 function if the source SV needs to be reused. Does not handle 'set' magic on
3826 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3827 performs a copy-by-value, obliterating any previous content of the
3830 You probably want to use one of the assortment of wrappers, such as
3831 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3832 C<SvSetMagicSV_nosteal>.
3834 =for apidoc sv_setsv_flags
3836 Copies the contents of the source SV C<ssv> into the destination SV
3837 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3838 function if the source SV needs to be reused. Does not handle 'set' magic.
3839 Loosely speaking, it performs a copy-by-value, obliterating any previous
3840 content of the destination.
3841 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3842 C<ssv> if appropriate, else not. If the C<flags>
3843 parameter has the C<SV_NOSTEAL> bit set then the
3844 buffers of temps will not be stolen. C<sv_setsv>
3845 and C<sv_setsv_nomg> are implemented in terms of this function.
3847 You probably want to use one of the assortment of wrappers, such as
3848 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3849 C<SvSetMagicSV_nosteal>.
3851 This is the primary function for copying scalars, and most other
3852 copy-ish functions and macros use this underneath.
3858 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3860 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3861 HV *old_stash = NULL;
3863 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3865 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3866 const char * const name = GvNAME(sstr);
3867 const STRLEN len = GvNAMELEN(sstr);
3869 if (dtype >= SVt_PV) {
3875 SvUPGRADE(dstr, SVt_PVGV);
3876 (void)SvOK_off(dstr);
3877 isGV_with_GP_on(dstr);
3879 GvSTASH(dstr) = GvSTASH(sstr);
3881 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3882 gv_name_set(MUTABLE_GV(dstr), name, len,
3883 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3884 SvFAKE_on(dstr); /* can coerce to non-glob */
3887 if(GvGP(MUTABLE_GV(sstr))) {
3888 /* If source has method cache entry, clear it */
3890 SvREFCNT_dec(GvCV(sstr));
3891 GvCV_set(sstr, NULL);
3894 /* If source has a real method, then a method is
3897 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3903 /* If dest already had a real method, that's a change as well */
3905 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3906 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3911 /* We don't need to check the name of the destination if it was not a
3912 glob to begin with. */
3913 if(dtype == SVt_PVGV) {
3914 const char * const name = GvNAME((const GV *)dstr);
3917 /* The stash may have been detached from the symbol table, so
3919 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3923 const STRLEN len = GvNAMELEN(dstr);
3924 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3925 || (len == 1 && name[0] == ':')) {
3928 /* Set aside the old stash, so we can reset isa caches on
3930 if((old_stash = GvHV(dstr)))
3931 /* Make sure we do not lose it early. */
3932 SvREFCNT_inc_simple_void_NN(
3933 sv_2mortal((SV *)old_stash)
3938 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3941 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3942 * so temporarily protect it */
3944 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3945 gp_free(MUTABLE_GV(dstr));
3946 GvINTRO_off(dstr); /* one-shot flag */
3947 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3950 if (SvTAINTED(sstr))
3952 if (GvIMPORTED(dstr) != GVf_IMPORTED
3953 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3955 GvIMPORTED_on(dstr);
3958 if(mro_changes == 2) {
3959 if (GvAV((const GV *)sstr)) {
3961 SV * const sref = (SV *)GvAV((const GV *)dstr);
3962 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3963 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3964 AV * const ary = newAV();
3965 av_push(ary, mg->mg_obj); /* takes the refcount */
3966 mg->mg_obj = (SV *)ary;
3968 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3970 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3972 mro_isa_changed_in(GvSTASH(dstr));
3974 else if(mro_changes == 3) {
3975 HV * const stash = GvHV(dstr);
3976 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3982 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3983 if (GvIO(dstr) && dtype == SVt_PVGV) {
3984 DEBUG_o(Perl_deb(aTHX_
3985 "glob_assign_glob clearing PL_stashcache\n"));
3986 /* It's a cache. It will rebuild itself quite happily.
3987 It's a lot of effort to work out exactly which key (or keys)
3988 might be invalidated by the creation of the this file handle.
3990 hv_clear(PL_stashcache);
3996 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3998 SV * const sref = SvRV(sstr);
4000 const int intro = GvINTRO(dstr);
4003 const U32 stype = SvTYPE(sref);
4005 PERL_ARGS_ASSERT_GV_SETREF;
4008 GvINTRO_off(dstr); /* one-shot flag */
4009 GvLINE(dstr) = CopLINE(PL_curcop);
4010 GvEGV(dstr) = MUTABLE_GV(dstr);
4015 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4016 import_flag = GVf_IMPORTED_CV;
4019 location = (SV **) &GvHV(dstr);
4020 import_flag = GVf_IMPORTED_HV;
4023 location = (SV **) &GvAV(dstr);
4024 import_flag = GVf_IMPORTED_AV;
4027 location = (SV **) &GvIOp(dstr);
4030 location = (SV **) &GvFORM(dstr);
4033 location = &GvSV(dstr);
4034 import_flag = GVf_IMPORTED_SV;
4037 if (stype == SVt_PVCV) {
4038 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4039 if (GvCVGEN(dstr)) {
4040 SvREFCNT_dec(GvCV(dstr));
4041 GvCV_set(dstr, NULL);
4042 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4045 /* SAVEt_GVSLOT takes more room on the savestack and has more
4046 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4047 leave_scope needs access to the GV so it can reset method
4048 caches. We must use SAVEt_GVSLOT whenever the type is
4049 SVt_PVCV, even if the stash is anonymous, as the stash may
4050 gain a name somehow before leave_scope. */
4051 if (stype == SVt_PVCV) {
4052 /* There is no save_pushptrptrptr. Creating it for this
4053 one call site would be overkill. So inline the ss add
4057 SS_ADD_PTR(location);
4058 SS_ADD_PTR(SvREFCNT_inc(*location));
4059 SS_ADD_UV(SAVEt_GVSLOT);
4062 else SAVEGENERICSV(*location);
4065 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4066 CV* const cv = MUTABLE_CV(*location);
4068 if (!GvCVGEN((const GV *)dstr) &&
4069 (CvROOT(cv) || CvXSUB(cv)) &&
4070 /* redundant check that avoids creating the extra SV
4071 most of the time: */
4072 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4074 SV * const new_const_sv =
4075 CvCONST((const CV *)sref)
4076 ? cv_const_sv((const CV *)sref)
4078 HV * const stash = GvSTASH((const GV *)dstr);
4079 report_redefined_cv(
4082 ? Perl_newSVpvf(aTHX_
4083 "%" HEKf "::%" HEKf,
4084 HEKfARG(HvNAME_HEK(stash)),
4085 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4086 : Perl_newSVpvf(aTHX_
4088 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4091 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4095 cv_ckproto_len_flags(cv, (const GV *)dstr,
4096 SvPOK(sref) ? CvPROTO(sref) : NULL,
4097 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4098 SvPOK(sref) ? SvUTF8(sref) : 0);
4100 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4101 GvASSUMECV_on(dstr);
4102 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4103 if (intro && GvREFCNT(dstr) > 1) {
4104 /* temporary remove extra savestack's ref */
4106 gv_method_changed(dstr);
4109 else gv_method_changed(dstr);
4112 *location = SvREFCNT_inc_simple_NN(sref);
4113 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4114 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4115 GvFLAGS(dstr) |= import_flag;
4118 if (stype == SVt_PVHV) {
4119 const char * const name = GvNAME((GV*)dstr);
4120 const STRLEN len = GvNAMELEN(dstr);
4123 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4124 || (len == 1 && name[0] == ':')
4126 && (!dref || HvENAME_get(dref))
4129 (HV *)sref, (HV *)dref,
4135 stype == SVt_PVAV && sref != dref
4136 && strEQ(GvNAME((GV*)dstr), "ISA")
4137 /* The stash may have been detached from the symbol table, so
4138 check its name before doing anything. */
4139 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4142 MAGIC * const omg = dref && SvSMAGICAL(dref)
4143 ? mg_find(dref, PERL_MAGIC_isa)
4145 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4146 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4147 AV * const ary = newAV();
4148 av_push(ary, mg->mg_obj); /* takes the refcount */
4149 mg->mg_obj = (SV *)ary;
4152 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4153 SV **svp = AvARRAY((AV *)omg->mg_obj);
4154 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4158 SvREFCNT_inc_simple_NN(*svp++)
4164 SvREFCNT_inc_simple_NN(omg->mg_obj)
4168 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4174 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4176 for (i = 0; i <= AvFILL(sref); ++i) {
4177 SV **elem = av_fetch ((AV*)sref, i, 0);
4180 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4184 mg = mg_find(sref, PERL_MAGIC_isa);
4186 /* Since the *ISA assignment could have affected more than
4187 one stash, don't call mro_isa_changed_in directly, but let
4188 magic_clearisa do it for us, as it already has the logic for
4189 dealing with globs vs arrays of globs. */
4191 Perl_magic_clearisa(aTHX_ NULL, mg);
4193 else if (stype == SVt_PVIO) {
4194 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4195 /* It's a cache. It will rebuild itself quite happily.
4196 It's a lot of effort to work out exactly which key (or keys)
4197 might be invalidated by the creation of the this file handle.
4199 hv_clear(PL_stashcache);
4203 if (!intro) SvREFCNT_dec(dref);
4204 if (SvTAINTED(sstr))
4212 #ifdef PERL_DEBUG_READONLY_COW
4213 # include <sys/mman.h>
4215 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4216 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4220 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4222 struct perl_memory_debug_header * const header =
4223 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4224 const MEM_SIZE len = header->size;
4225 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4226 # ifdef PERL_TRACK_MEMPOOL
4227 if (!header->readonly) header->readonly = 1;
4229 if (mprotect(header, len, PROT_READ))
4230 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4231 header, len, errno);
4235 S_sv_buf_to_rw(pTHX_ SV *sv)
4237 struct perl_memory_debug_header * const header =
4238 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4239 const MEM_SIZE len = header->size;
4240 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4241 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4242 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4243 header, len, errno);
4244 # ifdef PERL_TRACK_MEMPOOL
4245 header->readonly = 0;
4250 # define sv_buf_to_ro(sv) NOOP
4251 # define sv_buf_to_rw(sv) NOOP
4255 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4260 unsigned int both_type;
4262 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4264 if (UNLIKELY( sstr == dstr ))
4267 if (UNLIKELY( !sstr ))
4268 sstr = &PL_sv_undef;
4270 stype = SvTYPE(sstr);
4271 dtype = SvTYPE(dstr);
4272 both_type = (stype | dtype);
4274 /* with these values, we can check that both SVs are NULL/IV (and not
4275 * freed) just by testing the or'ed types */
4276 STATIC_ASSERT_STMT(SVt_NULL == 0);
4277 STATIC_ASSERT_STMT(SVt_IV == 1);
4278 if (both_type <= 1) {
4279 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4285 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4286 if (SvREADONLY(dstr))
4287 Perl_croak_no_modify();
4289 if (SvWEAKREF(dstr))
4290 sv_unref_flags(dstr, 0);
4292 old_rv = SvRV(dstr);
4295 assert(!SvGMAGICAL(sstr));
4296 assert(!SvGMAGICAL(dstr));
4298 sflags = SvFLAGS(sstr);
4299 if (sflags & (SVf_IOK|SVf_ROK)) {
4300 SET_SVANY_FOR_BODYLESS_IV(dstr);
4301 new_dflags = SVt_IV;
4303 if (sflags & SVf_ROK) {
4304 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4305 new_dflags |= SVf_ROK;
4308 /* both src and dst are <= SVt_IV, so sv_any points to the
4309 * head; so access the head directly
4311 assert( &(sstr->sv_u.svu_iv)
4312 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4313 assert( &(dstr->sv_u.svu_iv)
4314 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4315 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4316 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4320 new_dflags = dtype; /* turn off everything except the type */
4322 SvFLAGS(dstr) = new_dflags;
4323 SvREFCNT_dec(old_rv);
4328 if (UNLIKELY(both_type == SVTYPEMASK)) {
4329 if (SvIS_FREED(dstr)) {
4330 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4331 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4333 if (SvIS_FREED(sstr)) {
4334 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4335 (void*)sstr, (void*)dstr);
4341 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4342 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4344 /* There's a lot of redundancy below but we're going for speed here */
4349 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4350 (void)SvOK_off(dstr);
4358 /* For performance, we inline promoting to type SVt_IV. */
4359 /* We're starting from SVt_NULL, so provided that define is
4360 * actual 0, we don't have to unset any SV type flags
4361 * to promote to SVt_IV. */
4362 STATIC_ASSERT_STMT(SVt_NULL == 0);
4363 SET_SVANY_FOR_BODYLESS_IV(dstr);
4364 SvFLAGS(dstr) |= SVt_IV;
4368 sv_upgrade(dstr, SVt_PVIV);
4372 goto end_of_first_switch;
4374 (void)SvIOK_only(dstr);
4375 SvIV_set(dstr, SvIVX(sstr));
4378 /* SvTAINTED can only be true if the SV has taint magic, which in
4379 turn means that the SV type is PVMG (or greater). This is the
4380 case statement for SVt_IV, so this cannot be true (whatever gcov
4382 assert(!SvTAINTED(sstr));
4387 if (dtype < SVt_PV && dtype != SVt_IV)
4388 sv_upgrade(dstr, SVt_IV);
4392 if (LIKELY( SvNOK(sstr) )) {
4396 sv_upgrade(dstr, SVt_NV);
4400 sv_upgrade(dstr, SVt_PVNV);
4404 goto end_of_first_switch;
4406 SvNV_set(dstr, SvNVX(sstr));
4407 (void)SvNOK_only(dstr);
4408 /* SvTAINTED can only be true if the SV has taint magic, which in
4409 turn means that the SV type is PVMG (or greater). This is the
4410 case statement for SVt_NV, so this cannot be true (whatever gcov
4412 assert(!SvTAINTED(sstr));
4419 sv_upgrade(dstr, SVt_PV);
4422 if (dtype < SVt_PVIV)
4423 sv_upgrade(dstr, SVt_PVIV);
4426 if (dtype < SVt_PVNV)
4427 sv_upgrade(dstr, SVt_PVNV);
4431 const char * const type = sv_reftype(sstr,0);
4433 /* diag_listed_as: Bizarre copy of %s */
4434 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4436 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4438 NOT_REACHED; /* NOTREACHED */
4442 if (dtype < SVt_REGEXP)
4444 if (dtype >= SVt_PV) {
4450 sv_upgrade(dstr, SVt_REGEXP);
4458 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4460 if (SvTYPE(sstr) != stype)
4461 stype = SvTYPE(sstr);
4463 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4464 glob_assign_glob(dstr, sstr, dtype);
4467 if (stype == SVt_PVLV)
4469 if (isREGEXP(sstr)) goto upgregexp;
4470 SvUPGRADE(dstr, SVt_PVNV);
4473 SvUPGRADE(dstr, (svtype)stype);
4475 end_of_first_switch:
4477 /* dstr may have been upgraded. */
4478 dtype = SvTYPE(dstr);
4479 sflags = SvFLAGS(sstr);
4481 if (UNLIKELY( dtype == SVt_PVCV )) {
4482 /* Assigning to a subroutine sets the prototype. */
4485 const char *const ptr = SvPV_const(sstr, len);
4487 SvGROW(dstr, len + 1);
4488 Copy(ptr, SvPVX(dstr), len + 1, char);
4489 SvCUR_set(dstr, len);
4491 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4492 CvAUTOLOAD_off(dstr);
4497 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4498 || dtype == SVt_PVFM))
4500 const char * const type = sv_reftype(dstr,0);
4502 /* diag_listed_as: Cannot copy to %s */
4503 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4505 Perl_croak(aTHX_ "Cannot copy to %s", type);
4506 } else if (sflags & SVf_ROK) {
4507 if (isGV_with_GP(dstr)
4508 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4511 if (GvIMPORTED(dstr) != GVf_IMPORTED
4512 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4514 GvIMPORTED_on(dstr);
4519 glob_assign_glob(dstr, sstr, dtype);
4523 if (dtype >= SVt_PV) {
4524 if (isGV_with_GP(dstr)) {
4525 gv_setref(dstr, sstr);
4528 if (SvPVX_const(dstr)) {
4534 (void)SvOK_off(dstr);
4535 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4536 SvFLAGS(dstr) |= sflags & SVf_ROK;
4537 assert(!(sflags & SVp_NOK));
4538 assert(!(sflags & SVp_IOK));
4539 assert(!(sflags & SVf_NOK));
4540 assert(!(sflags & SVf_IOK));
4542 else if (isGV_with_GP(dstr)) {
4543 if (!(sflags & SVf_OK)) {
4544 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4545 "Undefined value assigned to typeglob");
4548 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4549 if (dstr != (const SV *)gv) {
4550 const char * const name = GvNAME((const GV *)dstr);
4551 const STRLEN len = GvNAMELEN(dstr);
4552 HV *old_stash = NULL;
4553 bool reset_isa = FALSE;
4554 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4555 || (len == 1 && name[0] == ':')) {
4556 /* Set aside the old stash, so we can reset isa caches
4557 on its subclasses. */
4558 if((old_stash = GvHV(dstr))) {
4559 /* Make sure we do not lose it early. */
4560 SvREFCNT_inc_simple_void_NN(
4561 sv_2mortal((SV *)old_stash)
4568 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4569 gp_free(MUTABLE_GV(dstr));
4571 GvGP_set(dstr, gp_ref(GvGP(gv)));
4574 HV * const stash = GvHV(dstr);
4576 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4586 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4587 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4588 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4590 else if (sflags & SVp_POK) {
4591 const STRLEN cur = SvCUR(sstr);
4592 const STRLEN len = SvLEN(sstr);
4595 * We have three basic ways to copy the string:
4601 * Which we choose is based on various factors. The following
4602 * things are listed in order of speed, fastest to slowest:
4604 * - Copying a short string
4605 * - Copy-on-write bookkeeping
4607 * - Copying a long string
4609 * We swipe the string (steal the string buffer) if the SV on the
4610 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4611 * big win on long strings. It should be a win on short strings if
4612 * SvPVX_const(dstr) has to be allocated. If not, it should not
4613 * slow things down, as SvPVX_const(sstr) would have been freed
4616 * We also steal the buffer from a PADTMP (operator target) if it
4617 * is ‘long enough’. For short strings, a swipe does not help
4618 * here, as it causes more malloc calls the next time the target
4619 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4620 * be allocated it is still not worth swiping PADTMPs for short
4621 * strings, as the savings here are small.
4623 * If swiping is not an option, then we see whether it is
4624 * worth using copy-on-write. If the lhs already has a buf-
4625 * fer big enough and the string is short, we skip it and fall back
4626 * to method 3, since memcpy is faster for short strings than the
4627 * later bookkeeping overhead that copy-on-write entails.
4629 * If the rhs is not a copy-on-write string yet, then we also
4630 * consider whether the buffer is too large relative to the string
4631 * it holds. Some operations such as readline allocate a large
4632 * buffer in the expectation of reusing it. But turning such into
4633 * a COW buffer is counter-productive because it increases memory
4634 * usage by making readline allocate a new large buffer the sec-
4635 * ond time round. So, if the buffer is too large, again, we use
4638 * Finally, if there is no buffer on the left, or the buffer is too
4639 * small, then we use copy-on-write and make both SVs share the
4644 /* Whichever path we take through the next code, we want this true,
4645 and doing it now facilitates the COW check. */
4646 (void)SvPOK_only(dstr);
4650 /* slated for free anyway (and not COW)? */
4651 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4652 /* or a swipable TARG */
4654 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4656 /* whose buffer is worth stealing */
4657 && CHECK_COWBUF_THRESHOLD(cur,len)
4660 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4661 (!(flags & SV_NOSTEAL)) &&
4662 /* and we're allowed to steal temps */
4663 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4664 len) /* and really is a string */
4665 { /* Passes the swipe test. */
4666 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4668 SvPV_set(dstr, SvPVX_mutable(sstr));
4669 SvLEN_set(dstr, SvLEN(sstr));
4670 SvCUR_set(dstr, SvCUR(sstr));
4673 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4674 SvPV_set(sstr, NULL);
4679 else if (flags & SV_COW_SHARED_HASH_KEYS
4681 #ifdef PERL_COPY_ON_WRITE
4684 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4685 /* If this is a regular (non-hek) COW, only so
4686 many COW "copies" are possible. */
4687 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4688 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4689 && !(SvFLAGS(dstr) & SVf_BREAK)
4690 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4691 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4695 && !(SvFLAGS(dstr) & SVf_BREAK)
4698 /* Either it's a shared hash key, or it's suitable for
4701 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4706 if (!(sflags & SVf_IsCOW)) {
4708 CowREFCNT(sstr) = 0;
4711 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4717 if (sflags & SVf_IsCOW) {
4721 SvPV_set(dstr, SvPVX_mutable(sstr));
4726 /* SvIsCOW_shared_hash */
4727 DEBUG_C(PerlIO_printf(Perl_debug_log,
4728 "Copy on write: Sharing hash\n"));
4730 assert (SvTYPE(dstr) >= SVt_PV);
4732 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4734 SvLEN_set(dstr, len);
4735 SvCUR_set(dstr, cur);
4738 /* Failed the swipe test, and we cannot do copy-on-write either.
4739 Have to copy the string. */
4740 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4741 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4742 SvCUR_set(dstr, cur);
4743 *SvEND(dstr) = '\0';
4745 if (sflags & SVp_NOK) {
4746 SvNV_set(dstr, SvNVX(sstr));
4748 if (sflags & SVp_IOK) {
4749 SvIV_set(dstr, SvIVX(sstr));
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))
4787 =for apidoc sv_set_undef
4789 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4790 Doesn't handle set magic.
4792 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4793 buffer, unlike C<undef $sv>.
4795 Introduced in perl 5.26.0.
4801 Perl_sv_set_undef(pTHX_ SV *sv)
4803 U32 type = SvTYPE(sv);
4805 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4807 /* shortcut, NULL, IV, RV */
4809 if (type <= SVt_IV) {
4810 assert(!SvGMAGICAL(sv));
4812 Perl_croak_no_modify();
4816 sv_unref_flags(sv, 0);
4819 SvFLAGS(sv) = type; /* quickly turn off all flags */
4820 SvREFCNT_dec_NN(rv);
4824 SvFLAGS(sv) = type; /* quickly turn off all flags */
4829 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4832 SV_CHECK_THINKFIRST_COW_DROP(sv);
4834 if (isGV_with_GP(sv))
4835 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4836 "Undefined value assigned to typeglob");
4844 =for apidoc sv_setsv_mg
4846 Like C<sv_setsv>, but also handles 'set' magic.
4852 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4854 PERL_ARGS_ASSERT_SV_SETSV_MG;
4856 sv_setsv(dstr,sstr);
4861 # define SVt_COW SVt_PV
4863 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4865 STRLEN cur = SvCUR(sstr);
4866 STRLEN len = SvLEN(sstr);
4868 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4869 const bool already = cBOOL(SvIsCOW(sstr));
4872 PERL_ARGS_ASSERT_SV_SETSV_COW;
4875 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4876 (void*)sstr, (void*)dstr);
4883 if (SvTHINKFIRST(dstr))
4884 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4885 else if (SvPVX_const(dstr))
4886 Safefree(SvPVX_mutable(dstr));
4890 SvUPGRADE(dstr, SVt_COW);
4892 assert (SvPOK(sstr));
4893 assert (SvPOKp(sstr));
4895 if (SvIsCOW(sstr)) {
4897 if (SvLEN(sstr) == 0) {
4898 /* source is a COW shared hash key. */
4899 DEBUG_C(PerlIO_printf(Perl_debug_log,
4900 "Fast copy on write: Sharing hash\n"));
4901 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4904 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4905 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4907 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4908 SvUPGRADE(sstr, SVt_COW);
4910 DEBUG_C(PerlIO_printf(Perl_debug_log,
4911 "Fast copy on write: Converting sstr to COW\n"));
4912 CowREFCNT(sstr) = 0;
4914 # ifdef PERL_DEBUG_READONLY_COW
4915 if (already) sv_buf_to_rw(sstr);
4918 new_pv = SvPVX_mutable(sstr);
4922 SvPV_set(dstr, new_pv);
4923 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4926 SvLEN_set(dstr, len);
4927 SvCUR_set(dstr, cur);
4936 =for apidoc sv_setpv_bufsize
4938 Sets the SV to be a string of cur bytes length, with at least
4939 len bytes available. Ensures that there is a null byte at SvEND.
4940 Returns a char * pointer to the SvPV buffer.
4946 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4950 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4952 SV_CHECK_THINKFIRST_COW_DROP(sv);
4953 SvUPGRADE(sv, SVt_PV);
4954 pv = SvGROW(sv, len + 1);
4957 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4960 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4965 =for apidoc sv_setpvn
4967 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4968 The C<len> parameter indicates the number of
4969 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4970 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4976 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4980 PERL_ARGS_ASSERT_SV_SETPVN;
4982 SV_CHECK_THINKFIRST_COW_DROP(sv);
4988 /* len is STRLEN which is unsigned, need to copy to signed */
4991 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4994 SvUPGRADE(sv, SVt_PV);
4996 dptr = SvGROW(sv, len + 1);
4997 Move(ptr,dptr,len,char);
5000 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5002 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5006 =for apidoc sv_setpvn_mg
5008 Like C<sv_setpvn>, but also handles 'set' magic.
5014 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5016 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5018 sv_setpvn(sv,ptr,len);
5023 =for apidoc sv_setpv
5025 Copies a string into an SV. The string must be terminated with a C<NUL>
5026 character, and not contain embeded C<NUL>'s.
5027 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5033 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5037 PERL_ARGS_ASSERT_SV_SETPV;
5039 SV_CHECK_THINKFIRST_COW_DROP(sv);
5045 SvUPGRADE(sv, SVt_PV);
5047 SvGROW(sv, len + 1);
5048 Move(ptr,SvPVX(sv),len+1,char);
5050 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5052 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5056 =for apidoc sv_setpv_mg
5058 Like C<sv_setpv>, but also handles 'set' magic.
5064 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5066 PERL_ARGS_ASSERT_SV_SETPV_MG;
5073 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5075 PERL_ARGS_ASSERT_SV_SETHEK;
5081 if (HEK_LEN(hek) == HEf_SVKEY) {
5082 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5085 const int flags = HEK_FLAGS(hek);
5086 if (flags & HVhek_WASUTF8) {
5087 STRLEN utf8_len = HEK_LEN(hek);
5088 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5089 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5092 } else if (flags & HVhek_UNSHARED) {
5093 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5096 else SvUTF8_off(sv);
5100 SV_CHECK_THINKFIRST_COW_DROP(sv);
5101 SvUPGRADE(sv, SVt_PV);
5103 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5104 SvCUR_set(sv, HEK_LEN(hek));
5110 else SvUTF8_off(sv);
5118 =for apidoc sv_usepvn_flags
5120 Tells an SV to use C<ptr> to find its string value. Normally the
5121 string is stored inside the SV, but sv_usepvn allows the SV to use an
5122 outside string. C<ptr> should point to memory that was allocated
5123 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5124 the start of a C<Newx>-ed block of memory, and not a pointer to the
5125 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5126 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5127 string length, C<len>, must be supplied. By default this function
5128 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5129 so that pointer should not be freed or used by the programmer after
5130 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5131 that pointer (e.g. ptr + 1) be used.
5133 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5134 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5136 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5137 C<len>, and already meets the requirements for storing in C<SvPVX>).
5143 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5147 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5149 SV_CHECK_THINKFIRST_COW_DROP(sv);
5150 SvUPGRADE(sv, SVt_PV);
5153 if (flags & SV_SMAGIC)
5157 if (SvPVX_const(sv))
5161 if (flags & SV_HAS_TRAILING_NUL)
5162 assert(ptr[len] == '\0');
5165 allocate = (flags & SV_HAS_TRAILING_NUL)
5167 #ifdef Perl_safesysmalloc_size
5170 PERL_STRLEN_ROUNDUP(len + 1);
5172 if (flags & SV_HAS_TRAILING_NUL) {
5173 /* It's long enough - do nothing.
5174 Specifically Perl_newCONSTSUB is relying on this. */
5177 /* Force a move to shake out bugs in callers. */
5178 char *new_ptr = (char*)safemalloc(allocate);
5179 Copy(ptr, new_ptr, len, char);
5180 PoisonFree(ptr,len,char);
5184 ptr = (char*) saferealloc (ptr, allocate);
5187 #ifdef Perl_safesysmalloc_size
5188 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5190 SvLEN_set(sv, allocate);
5194 if (!(flags & SV_HAS_TRAILING_NUL)) {
5197 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5199 if (flags & SV_SMAGIC)
5204 =for apidoc sv_force_normal_flags
5206 Undo various types of fakery on an SV, where fakery means
5207 "more than" a string: if the PV is a shared string, make
5208 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5209 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5210 we do the copy, and is also used locally; if this is a
5211 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5212 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5213 C<SvPOK_off> rather than making a copy. (Used where this
5214 scalar is about to be set to some other value.) In addition,
5215 the C<flags> parameter gets passed to C<sv_unref_flags()>
5216 when unreffing. C<sv_force_normal> calls this function
5217 with flags set to 0.
5219 This function is expected to be used to signal to perl that this SV is
5220 about to be written to, and any extra book-keeping needs to be taken care
5221 of. Hence, it croaks on read-only values.
5227 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5229 assert(SvIsCOW(sv));
5232 const char * const pvx = SvPVX_const(sv);
5233 const STRLEN len = SvLEN(sv);
5234 const STRLEN cur = SvCUR(sv);
5237 PerlIO_printf(Perl_debug_log,
5238 "Copy on write: Force normal %ld\n",
5243 # ifdef PERL_COPY_ON_WRITE
5245 /* Must do this first, since the CowREFCNT uses SvPVX and
5246 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5247 the only owner left of the buffer. */
5248 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5250 U8 cowrefcnt = CowREFCNT(sv);
5251 if(cowrefcnt != 0) {
5253 CowREFCNT(sv) = cowrefcnt;
5258 /* Else we are the only owner of the buffer. */
5263 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5268 if (flags & SV_COW_DROP_PV) {
5269 /* OK, so we don't need to copy our buffer. */
5272 SvGROW(sv, cur + 1);
5273 Move(pvx,SvPVX(sv),cur,char);
5279 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5286 const char * const pvx = SvPVX_const(sv);
5287 const STRLEN len = SvCUR(sv);
5291 if (flags & SV_COW_DROP_PV) {
5292 /* OK, so we don't need to copy our buffer. */
5295 SvGROW(sv, len + 1);
5296 Move(pvx,SvPVX(sv),len,char);
5299 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5305 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5307 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5310 Perl_croak_no_modify();
5311 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5312 S_sv_uncow(aTHX_ sv, flags);
5314 sv_unref_flags(sv, flags);
5315 else if (SvFAKE(sv) && isGV_with_GP(sv))
5316 sv_unglob(sv, flags);
5317 else if (SvFAKE(sv) && isREGEXP(sv)) {
5318 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5319 to sv_unglob. We only need it here, so inline it. */
5320 const bool islv = SvTYPE(sv) == SVt_PVLV;
5321 const svtype new_type =
5322 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5323 SV *const temp = newSV_type(new_type);
5324 regexp *const temp_p = ReANY((REGEXP *)sv);
5326 if (new_type == SVt_PVMG) {
5327 SvMAGIC_set(temp, SvMAGIC(sv));
5328 SvMAGIC_set(sv, NULL);
5329 SvSTASH_set(temp, SvSTASH(sv));
5330 SvSTASH_set(sv, NULL);
5332 if (!islv) SvCUR_set(temp, SvCUR(sv));
5333 /* Remember that SvPVX is in the head, not the body. But
5334 RX_WRAPPED is in the body. */
5335 assert(ReANY((REGEXP *)sv)->mother_re);
5336 /* Their buffer is already owned by someone else. */
5337 if (flags & SV_COW_DROP_PV) {
5338 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5339 zeroed body. For SVt_PVLV, it should have been set to 0
5340 before turning into a regexp. */
5341 assert(!SvLEN(islv ? sv : temp));
5342 sv->sv_u.svu_pv = 0;
5345 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5346 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5350 /* Now swap the rest of the bodies. */
5354 SvFLAGS(sv) &= ~SVTYPEMASK;
5355 SvFLAGS(sv) |= new_type;
5356 SvANY(sv) = SvANY(temp);
5359 SvFLAGS(temp) &= ~(SVTYPEMASK);
5360 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5361 SvANY(temp) = temp_p;
5362 temp->sv_u.svu_rx = (regexp *)temp_p;
5364 SvREFCNT_dec_NN(temp);
5366 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5372 Efficient removal of characters from the beginning of the string buffer.
5373 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5374 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5375 character of the adjusted string. Uses the C<OOK> hack. On return, only
5376 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5378 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5379 refer to the same chunk of data.
5381 The unfortunate similarity of this function's name to that of Perl's C<chop>
5382 operator is strictly coincidental. This function works from the left;
5383 C<chop> works from the right.
5389 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5400 PERL_ARGS_ASSERT_SV_CHOP;
5402 if (!ptr || !SvPOKp(sv))
5404 delta = ptr - SvPVX_const(sv);
5406 /* Nothing to do. */
5409 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5410 if (delta > max_delta)
5411 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5412 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5413 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5414 SV_CHECK_THINKFIRST(sv);
5415 SvPOK_only_UTF8(sv);
5418 if (!SvLEN(sv)) { /* make copy of shared string */
5419 const char *pvx = SvPVX_const(sv);
5420 const STRLEN len = SvCUR(sv);
5421 SvGROW(sv, len + 1);
5422 Move(pvx,SvPVX(sv),len,char);
5428 SvOOK_offset(sv, old_delta);
5430 SvLEN_set(sv, SvLEN(sv) - delta);
5431 SvCUR_set(sv, SvCUR(sv) - delta);
5432 SvPV_set(sv, SvPVX(sv) + delta);
5434 p = (U8 *)SvPVX_const(sv);
5437 /* how many bytes were evacuated? we will fill them with sentinel
5438 bytes, except for the part holding the new offset of course. */
5441 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5443 assert(evacn <= delta + old_delta);
5447 /* This sets 'delta' to the accumulated value of all deltas so far */
5451 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5452 * the string; otherwise store a 0 byte there and store 'delta' just prior
5453 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5454 * portion of the chopped part of the string */
5455 if (delta < 0x100) {
5459 p -= sizeof(STRLEN);
5460 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5464 /* Fill the preceding buffer with sentinals to verify that no-one is
5474 =for apidoc sv_catpvn
5476 Concatenates the string onto the end of the string which is in the SV.
5477 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5478 status set, then the bytes appended should be valid UTF-8.
5479 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5481 =for apidoc sv_catpvn_flags
5483 Concatenates the string onto the end of the string which is in the SV. The
5484 C<len> indicates number of bytes to copy.
5486 By default, the string appended is assumed to be valid UTF-8 if the SV has
5487 the UTF-8 status set, and a string of bytes otherwise. One can force the
5488 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5489 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5490 string appended will be upgraded to UTF-8 if necessary.
5492 If C<flags> has the C<SV_SMAGIC> bit set, will
5493 C<mg_set> on C<dsv> afterwards if appropriate.
5494 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5495 in terms of this function.
5501 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5504 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5506 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5507 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5509 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5510 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5511 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5514 else SvGROW(dsv, dlen + slen + 3);
5516 sstr = SvPVX_const(dsv);
5517 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5518 SvCUR_set(dsv, SvCUR(dsv) + slen);
5521 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5522 const char * const send = sstr + slen;
5525 /* Something this code does not account for, which I think is
5526 impossible; it would require the same pv to be treated as
5527 bytes *and* utf8, which would indicate a bug elsewhere. */
5528 assert(sstr != dstr);
5530 SvGROW(dsv, dlen + slen * 2 + 3);
5531 d = (U8 *)SvPVX(dsv) + dlen;
5533 while (sstr < send) {
5534 append_utf8_from_native_byte(*sstr, &d);
5537 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5540 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5542 if (flags & SV_SMAGIC)
5547 =for apidoc sv_catsv
5549 Concatenates the string from SV C<ssv> onto the end of the string in SV
5550 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5551 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5552 and C<L</sv_catsv_nomg>>.
5554 =for apidoc sv_catsv_flags
5556 Concatenates the string from SV C<ssv> onto the end of the string in SV
5557 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5558 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5559 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5560 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5561 and C<sv_catsv_mg> are implemented in terms of this function.
5566 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5568 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5572 const char *spv = SvPV_flags_const(ssv, slen, flags);
5573 if (flags & SV_GMAGIC)
5575 sv_catpvn_flags(dsv, spv, slen,
5576 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5577 if (flags & SV_SMAGIC)
5583 =for apidoc sv_catpv
5585 Concatenates the C<NUL>-terminated string onto the end of the string which is
5587 If the SV has the UTF-8 status set, then the bytes appended should be
5588 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5594 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5600 PERL_ARGS_ASSERT_SV_CATPV;
5604 junk = SvPV_force(sv, tlen);
5606 SvGROW(sv, tlen + len + 1);
5608 ptr = SvPVX_const(sv);
5609 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5610 SvCUR_set(sv, SvCUR(sv) + len);
5611 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5616 =for apidoc sv_catpv_flags
5618 Concatenates the C<NUL>-terminated string onto the end of the string which is
5620 If the SV has the UTF-8 status set, then the bytes appended should
5621 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5622 on the modified SV if appropriate.
5628 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5630 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5631 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5635 =for apidoc sv_catpv_mg
5637 Like C<sv_catpv>, but also handles 'set' magic.
5643 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5645 PERL_ARGS_ASSERT_SV_CATPV_MG;
5654 Creates a new SV. A non-zero C<len> parameter indicates the number of
5655 bytes of preallocated string space the SV should have. An extra byte for a
5656 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5657 space is allocated.) The reference count for the new SV is set to 1.
5659 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5660 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5661 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5662 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5663 modules supporting older perls.
5669 Perl_newSV(pTHX_ const STRLEN len)
5675 sv_grow(sv, len + 1);
5680 =for apidoc sv_magicext
5682 Adds magic to an SV, upgrading it if necessary. Applies the
5683 supplied C<vtable> and returns a pointer to the magic added.
5685 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5686 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5687 one instance of the same C<how>.
5689 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5690 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5691 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5692 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5694 (This is now used as a subroutine by C<sv_magic>.)
5699 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5700 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5704 PERL_ARGS_ASSERT_SV_MAGICEXT;
5706 SvUPGRADE(sv, SVt_PVMG);
5707 Newxz(mg, 1, MAGIC);
5708 mg->mg_moremagic = SvMAGIC(sv);
5709 SvMAGIC_set(sv, mg);
5711 /* Sometimes a magic contains a reference loop, where the sv and
5712 object refer to each other. To prevent a reference loop that
5713 would prevent such objects being freed, we look for such loops
5714 and if we find one we avoid incrementing the object refcount.
5716 Note we cannot do this to avoid self-tie loops as intervening RV must
5717 have its REFCNT incremented to keep it in existence.
5720 if (!obj || obj == sv ||
5721 how == PERL_MAGIC_arylen ||
5722 how == PERL_MAGIC_regdata ||
5723 how == PERL_MAGIC_regdatum ||
5724 how == PERL_MAGIC_symtab ||
5725 (SvTYPE(obj) == SVt_PVGV &&
5726 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5727 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5728 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5733 mg->mg_obj = SvREFCNT_inc_simple(obj);
5734 mg->mg_flags |= MGf_REFCOUNTED;
5737 /* Normal self-ties simply pass a null object, and instead of
5738 using mg_obj directly, use the SvTIED_obj macro to produce a
5739 new RV as needed. For glob "self-ties", we are tieing the PVIO
5740 with an RV obj pointing to the glob containing the PVIO. In
5741 this case, to avoid a reference loop, we need to weaken the
5745 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5746 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5752 mg->mg_len = namlen;
5755 mg->mg_ptr = savepvn(name, namlen);
5756 else if (namlen == HEf_SVKEY) {
5757 /* Yes, this is casting away const. This is only for the case of
5758 HEf_SVKEY. I think we need to document this aberation of the
5759 constness of the API, rather than making name non-const, as
5760 that change propagating outwards a long way. */
5761 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5763 mg->mg_ptr = (char *) name;
5765 mg->mg_virtual = (MGVTBL *) vtable;
5772 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5774 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5775 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5776 /* This sv is only a delegate. //g magic must be attached to
5781 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5782 &PL_vtbl_mglob, 0, 0);
5786 =for apidoc sv_magic
5788 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5789 necessary, then adds a new magic item of type C<how> to the head of the
5792 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5793 handling of the C<name> and C<namlen> arguments.
5795 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5796 to add more than one instance of the same C<how>.
5802 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5803 const char *const name, const I32 namlen)
5805 const MGVTBL *vtable;
5808 unsigned int vtable_index;
5810 PERL_ARGS_ASSERT_SV_MAGIC;
5812 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5813 || ((flags = PL_magic_data[how]),
5814 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5815 > magic_vtable_max))
5816 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5818 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5819 Useful for attaching extension internal data to perl vars.
5820 Note that multiple extensions may clash if magical scalars
5821 etc holding private data from one are passed to another. */
5823 vtable = (vtable_index == magic_vtable_max)
5824 ? NULL : PL_magic_vtables + vtable_index;
5826 if (SvREADONLY(sv)) {
5828 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5831 Perl_croak_no_modify();
5834 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5835 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5836 /* sv_magic() refuses to add a magic of the same 'how' as an
5839 if (how == PERL_MAGIC_taint)
5845 /* Force pos to be stored as characters, not bytes. */
5846 if (SvMAGICAL(sv) && DO_UTF8(sv)
5847 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5849 && mg->mg_flags & MGf_BYTES) {
5850 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5852 mg->mg_flags &= ~MGf_BYTES;
5855 /* Rest of work is done else where */
5856 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5859 case PERL_MAGIC_taint:
5862 case PERL_MAGIC_ext:
5863 case PERL_MAGIC_dbfile:
5870 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5877 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5879 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5880 for (mg = *mgp; mg; mg = *mgp) {
5881 const MGVTBL* const virt = mg->mg_virtual;
5882 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5883 *mgp = mg->mg_moremagic;
5884 if (virt && virt->svt_free)
5885 virt->svt_free(aTHX_ sv, mg);
5886 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5888 Safefree(mg->mg_ptr);
5889 else if (mg->mg_len == HEf_SVKEY)
5890 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5891 else if (mg->mg_type == PERL_MAGIC_utf8)
5892 Safefree(mg->mg_ptr);
5894 if (mg->mg_flags & MGf_REFCOUNTED)
5895 SvREFCNT_dec(mg->mg_obj);
5899 mgp = &mg->mg_moremagic;
5902 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5903 mg_magical(sv); /* else fix the flags now */
5912 =for apidoc sv_unmagic
5914 Removes all magic of type C<type> from an SV.
5920 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5922 PERL_ARGS_ASSERT_SV_UNMAGIC;
5923 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5927 =for apidoc sv_unmagicext
5929 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5935 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5937 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5938 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5942 =for apidoc sv_rvweaken
5944 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5945 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5946 push a back-reference to this RV onto the array of backreferences
5947 associated with that magic. If the RV is magical, set magic will be
5948 called after the RV is cleared.
5954 Perl_sv_rvweaken(pTHX_ SV *const sv)
5958 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5960 if (!SvOK(sv)) /* let undefs pass */
5963 Perl_croak(aTHX_ "Can't weaken a nonreference");
5964 else if (SvWEAKREF(sv)) {
5965 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5968 else if (SvREADONLY(sv)) croak_no_modify();
5970 Perl_sv_add_backref(aTHX_ tsv, sv);
5972 SvREFCNT_dec_NN(tsv);
5977 =for apidoc sv_get_backrefs
5979 If C<sv> is the target of a weak reference then it returns the back
5980 references structure associated with the sv; otherwise return C<NULL>.
5982 When returning a non-null result the type of the return is relevant. If it
5983 is an AV then the elements of the AV are the weak reference RVs which
5984 point at this item. If it is any other type then the item itself is the
5987 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5988 C<Perl_sv_kill_backrefs()>
5994 Perl_sv_get_backrefs(SV *const sv)
5998 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6000 /* find slot to store array or singleton backref */
6002 if (SvTYPE(sv) == SVt_PVHV) {
6004 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6005 backrefs = (SV *)iter->xhv_backreferences;
6007 } else if (SvMAGICAL(sv)) {
6008 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6010 backrefs = mg->mg_obj;
6015 /* Give tsv backref magic if it hasn't already got it, then push a
6016 * back-reference to sv onto the array associated with the backref magic.
6018 * As an optimisation, if there's only one backref and it's not an AV,
6019 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6020 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6024 /* A discussion about the backreferences array and its refcount:
6026 * The AV holding the backreferences is pointed to either as the mg_obj of
6027 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6028 * xhv_backreferences field. The array is created with a refcount
6029 * of 2. This means that if during global destruction the array gets
6030 * picked on before its parent to have its refcount decremented by the
6031 * random zapper, it won't actually be freed, meaning it's still there for
6032 * when its parent gets freed.
6034 * When the parent SV is freed, the extra ref is killed by
6035 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6036 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6038 * When a single backref SV is stored directly, it is not reference
6043 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6049 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6051 /* find slot to store array or singleton backref */
6053 if (SvTYPE(tsv) == SVt_PVHV) {
6054 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6057 mg = mg_find(tsv, PERL_MAGIC_backref);
6059 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6060 svp = &(mg->mg_obj);
6063 /* create or retrieve the array */
6065 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6066 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6070 mg->mg_flags |= MGf_REFCOUNTED;
6073 SvREFCNT_inc_simple_void_NN(av);
6074 /* av now has a refcnt of 2; see discussion above */
6075 av_extend(av, *svp ? 2 : 1);
6077 /* move single existing backref to the array */
6078 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6083 av = MUTABLE_AV(*svp);
6085 /* optimisation: store single backref directly in HvAUX or mg_obj */
6089 assert(SvTYPE(av) == SVt_PVAV);
6090 if (AvFILLp(av) >= AvMAX(av)) {
6091 av_extend(av, AvFILLp(av)+1);
6094 /* push new backref */
6095 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6098 /* delete a back-reference to ourselves from the backref magic associated
6099 * with the SV we point to.
6103 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6107 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6109 if (SvTYPE(tsv) == SVt_PVHV) {
6111 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6113 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6114 /* It's possible for the the last (strong) reference to tsv to have
6115 become freed *before* the last thing holding a weak reference.
6116 If both survive longer than the backreferences array, then when
6117 the referent's reference count drops to 0 and it is freed, it's
6118 not able to chase the backreferences, so they aren't NULLed.
6120 For example, a CV holds a weak reference to its stash. If both the
6121 CV and the stash survive longer than the backreferences array,
6122 and the CV gets picked for the SvBREAK() treatment first,
6123 *and* it turns out that the stash is only being kept alive because
6124 of an our variable in the pad of the CV, then midway during CV
6125 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6126 It ends up pointing to the freed HV. Hence it's chased in here, and
6127 if this block wasn't here, it would hit the !svp panic just below.
6129 I don't believe that "better" destruction ordering is going to help
6130 here - during global destruction there's always going to be the
6131 chance that something goes out of order. We've tried to make it
6132 foolproof before, and it only resulted in evolutionary pressure on
6133 fools. Which made us look foolish for our hubris. :-(
6139 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6140 svp = mg ? &(mg->mg_obj) : NULL;
6144 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6146 /* It's possible that sv is being freed recursively part way through the
6147 freeing of tsv. If this happens, the backreferences array of tsv has
6148 already been freed, and so svp will be NULL. If this is the case,
6149 we should not panic. Instead, nothing needs doing, so return. */
6150 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6152 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6153 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6156 if (SvTYPE(*svp) == SVt_PVAV) {
6160 AV * const av = (AV*)*svp;
6162 assert(!SvIS_FREED(av));
6166 /* for an SV with N weak references to it, if all those
6167 * weak refs are deleted, then sv_del_backref will be called
6168 * N times and O(N^2) compares will be done within the backref
6169 * array. To ameliorate this potential slowness, we:
6170 * 1) make sure this code is as tight as possible;
6171 * 2) when looking for SV, look for it at both the head and tail of the
6172 * array first before searching the rest, since some create/destroy
6173 * patterns will cause the backrefs to be freed in order.
6180 SV **p = &svp[fill];
6181 SV *const topsv = *p;
6188 /* We weren't the last entry.
6189 An unordered list has this property that you
6190 can take the last element off the end to fill
6191 the hole, and it's still an unordered list :-)
6197 break; /* should only be one */
6204 AvFILLp(av) = fill-1;
6206 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6207 /* freed AV; skip */
6210 /* optimisation: only a single backref, stored directly */
6212 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6213 (void*)*svp, (void*)sv);
6220 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6226 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6231 /* after multiple passes through Perl_sv_clean_all() for a thingy
6232 * that has badly leaked, the backref array may have gotten freed,
6233 * since we only protect it against 1 round of cleanup */
6234 if (SvIS_FREED(av)) {
6235 if (PL_in_clean_all) /* All is fair */
6238 "panic: magic_killbackrefs (freed backref AV/SV)");
6242 is_array = (SvTYPE(av) == SVt_PVAV);
6244 assert(!SvIS_FREED(av));
6247 last = svp + AvFILLp(av);
6250 /* optimisation: only a single backref, stored directly */
6256 while (svp <= last) {
6258 SV *const referrer = *svp;
6259 if (SvWEAKREF(referrer)) {
6260 /* XXX Should we check that it hasn't changed? */
6261 assert(SvROK(referrer));
6262 SvRV_set(referrer, 0);
6264 SvWEAKREF_off(referrer);
6265 SvSETMAGIC(referrer);
6266 } else if (SvTYPE(referrer) == SVt_PVGV ||
6267 SvTYPE(referrer) == SVt_PVLV) {
6268 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6269 /* You lookin' at me? */
6270 assert(GvSTASH(referrer));
6271 assert(GvSTASH(referrer) == (const HV *)sv);
6272 GvSTASH(referrer) = 0;
6273 } else if (SvTYPE(referrer) == SVt_PVCV ||
6274 SvTYPE(referrer) == SVt_PVFM) {
6275 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6276 /* You lookin' at me? */
6277 assert(CvSTASH(referrer));
6278 assert(CvSTASH(referrer) == (const HV *)sv);
6279 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6282 assert(SvTYPE(sv) == SVt_PVGV);
6283 /* You lookin' at me? */
6284 assert(CvGV(referrer));
6285 assert(CvGV(referrer) == (const GV *)sv);
6286 anonymise_cv_maybe(MUTABLE_GV(sv),
6287 MUTABLE_CV(referrer));
6292 "panic: magic_killbackrefs (flags=%" UVxf ")",
6293 (UV)SvFLAGS(referrer));
6304 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6310 =for apidoc sv_insert
6312 Inserts a string at the specified offset/length within the SV. Similar to
6313 the Perl C<substr()> function. Handles get magic.
6315 =for apidoc sv_insert_flags
6317 Same as C<sv_insert>, but the extra C<flags> are passed to the
6318 C<SvPV_force_flags> that applies to C<bigstr>.
6324 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6330 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6333 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6335 SvPV_force_flags(bigstr, curlen, flags);
6336 (void)SvPOK_only_UTF8(bigstr);
6337 if (offset + len > curlen) {
6338 SvGROW(bigstr, offset+len+1);
6339 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6340 SvCUR_set(bigstr, offset+len);
6344 i = littlelen - len;
6345 if (i > 0) { /* string might grow */
6346 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6347 mid = big + offset + len;
6348 midend = bigend = big + SvCUR(bigstr);
6351 while (midend > mid) /* shove everything down */
6352 *--bigend = *--midend;
6353 Move(little,big+offset,littlelen,char);
6354 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6359 Move(little,SvPVX(bigstr)+offset,len,char);
6364 big = SvPVX(bigstr);
6367 bigend = big + SvCUR(bigstr);
6369 if (midend > bigend)
6370 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6373 if (mid - big > bigend - midend) { /* faster to shorten from end */
6375 Move(little, mid, littlelen,char);
6378 i = bigend - midend;
6380 Move(midend, mid, i,char);
6384 SvCUR_set(bigstr, mid - big);
6386 else if ((i = mid - big)) { /* faster from front */
6387 midend -= littlelen;
6389 Move(big, midend - i, i, char);
6390 sv_chop(bigstr,midend-i);
6392 Move(little, mid, littlelen,char);
6394 else if (littlelen) {
6395 midend -= littlelen;
6396 sv_chop(bigstr,midend);
6397 Move(little,midend,littlelen,char);
6400 sv_chop(bigstr,midend);
6406 =for apidoc sv_replace
6408 Make the first argument a copy of the second, then delete the original.
6409 The target SV physically takes over ownership of the body of the source SV
6410 and inherits its flags; however, the target keeps any magic it owns,
6411 and any magic in the source is discarded.
6412 Note that this is a rather specialist SV copying operation; most of the
6413 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6419 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6421 const U32 refcnt = SvREFCNT(sv);
6423 PERL_ARGS_ASSERT_SV_REPLACE;
6425 SV_CHECK_THINKFIRST_COW_DROP(sv);
6426 if (SvREFCNT(nsv) != 1) {
6427 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6428 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6430 if (SvMAGICAL(sv)) {
6434 sv_upgrade(nsv, SVt_PVMG);
6435 SvMAGIC_set(nsv, SvMAGIC(sv));
6436 SvFLAGS(nsv) |= SvMAGICAL(sv);
6438 SvMAGIC_set(sv, NULL);
6442 assert(!SvREFCNT(sv));
6443 #ifdef DEBUG_LEAKING_SCALARS
6444 sv->sv_flags = nsv->sv_flags;
6445 sv->sv_any = nsv->sv_any;
6446 sv->sv_refcnt = nsv->sv_refcnt;
6447 sv->sv_u = nsv->sv_u;
6449 StructCopy(nsv,sv,SV);
6451 if(SvTYPE(sv) == SVt_IV) {
6452 SET_SVANY_FOR_BODYLESS_IV(sv);
6456 SvREFCNT(sv) = refcnt;
6457 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6462 /* We're about to free a GV which has a CV that refers back to us.
6463 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6467 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6472 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6475 assert(SvREFCNT(gv) == 0);
6476 assert(isGV(gv) && isGV_with_GP(gv));
6478 assert(!CvANON(cv));
6479 assert(CvGV(cv) == gv);
6480 assert(!CvNAMED(cv));
6482 /* will the CV shortly be freed by gp_free() ? */
6483 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6484 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6488 /* if not, anonymise: */
6489 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6490 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6491 : newSVpvn_flags( "__ANON__", 8, 0 );
6492 sv_catpvs(gvname, "::__ANON__");
6493 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6494 SvREFCNT_dec_NN(gvname);
6498 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6503 =for apidoc sv_clear
6505 Clear an SV: call any destructors, free up any memory used by the body,
6506 and free the body itself. The SV's head is I<not> freed, although
6507 its type is set to all 1's so that it won't inadvertently be assumed
6508 to be live during global destruction etc.
6509 This function should only be called when C<REFCNT> is zero. Most of the time
6510 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6517 Perl_sv_clear(pTHX_ SV *const orig_sv)
6522 const struct body_details *sv_type_details;
6526 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6527 Not strictly necessary */
6529 PERL_ARGS_ASSERT_SV_CLEAR;
6531 /* within this loop, sv is the SV currently being freed, and
6532 * iter_sv is the most recent AV or whatever that's being iterated
6533 * over to provide more SVs */
6539 assert(SvREFCNT(sv) == 0);
6540 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6542 if (type <= SVt_IV) {
6543 /* See the comment in sv.h about the collusion between this
6544 * early return and the overloading of the NULL slots in the
6548 SvFLAGS(sv) &= SVf_BREAK;
6549 SvFLAGS(sv) |= SVTYPEMASK;
6553 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6554 for another purpose */
6555 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6557 if (type >= SVt_PVMG) {
6559 if (!curse(sv, 1)) goto get_next_sv;
6560 type = SvTYPE(sv); /* destructor may have changed it */
6562 /* Free back-references before magic, in case the magic calls
6563 * Perl code that has weak references to sv. */
6564 if (type == SVt_PVHV) {
6565 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6569 else if (SvMAGIC(sv)) {
6570 /* Free back-references before other types of magic. */
6571 sv_unmagic(sv, PERL_MAGIC_backref);
6577 /* case SVt_INVLIST: */
6580 IoIFP(sv) != PerlIO_stdin() &&
6581 IoIFP(sv) != PerlIO_stdout() &&
6582 IoIFP(sv) != PerlIO_stderr() &&
6583 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6585 io_close(MUTABLE_IO(sv), NULL, FALSE,
6586 (IoTYPE(sv) == IoTYPE_WRONLY ||
6587 IoTYPE(sv) == IoTYPE_RDWR ||
6588 IoTYPE(sv) == IoTYPE_APPEND));
6590 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6591 PerlDir_close(IoDIRP(sv));
6592 IoDIRP(sv) = (DIR*)NULL;
6593 Safefree(IoTOP_NAME(sv));
6594 Safefree(IoFMT_NAME(sv));
6595 Safefree(IoBOTTOM_NAME(sv));
6596 if ((const GV *)sv == PL_statgv)
6600 /* FIXME for plugins */
6602 pregfree2((REGEXP*) sv);
6606 cv_undef(MUTABLE_CV(sv));
6607 /* If we're in a stash, we don't own a reference to it.
6608 * However it does have a back reference to us, which needs to
6610 if ((stash = CvSTASH(sv)))
6611 sv_del_backref(MUTABLE_SV(stash), sv);
6614 if (PL_last_swash_hv == (const HV *)sv) {
6615 PL_last_swash_hv = NULL;
6617 if (HvTOTALKEYS((HV*)sv) > 0) {
6619 /* this statement should match the one at the beginning of
6620 * hv_undef_flags() */
6621 if ( PL_phase != PERL_PHASE_DESTRUCT
6622 && (hek = HvNAME_HEK((HV*)sv)))
6624 if (PL_stashcache) {
6625 DEBUG_o(Perl_deb(aTHX_
6626 "sv_clear clearing PL_stashcache for '%" HEKf
6629 (void)hv_deletehek(PL_stashcache,
6632 hv_name_set((HV*)sv, NULL, 0, 0);
6635 /* save old iter_sv in unused SvSTASH field */
6636 assert(!SvOBJECT(sv));
6637 SvSTASH(sv) = (HV*)iter_sv;
6640 /* save old hash_index in unused SvMAGIC field */
6641 assert(!SvMAGICAL(sv));
6642 assert(!SvMAGIC(sv));
6643 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6646 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6647 goto get_next_sv; /* process this new sv */
6649 /* free empty hash */
6650 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6651 assert(!HvARRAY((HV*)sv));
6655 AV* av = MUTABLE_AV(sv);
6656 if (PL_comppad == av) {
6660 if (AvREAL(av) && AvFILLp(av) > -1) {
6661 next_sv = AvARRAY(av)[AvFILLp(av)--];
6662 /* save old iter_sv in top-most slot of AV,
6663 * and pray that it doesn't get wiped in the meantime */
6664 AvARRAY(av)[AvMAX(av)] = iter_sv;
6666 goto get_next_sv; /* process this new sv */
6668 Safefree(AvALLOC(av));
6673 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6674 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6675 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6676 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6678 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6679 SvREFCNT_dec(LvTARG(sv));
6680 if (isREGEXP(sv)) goto freeregexp;
6683 if (isGV_with_GP(sv)) {
6684 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6685 && HvENAME_get(stash))
6686 mro_method_changed_in(stash);
6687 gp_free(MUTABLE_GV(sv));
6689 unshare_hek(GvNAME_HEK(sv));
6690 /* If we're in a stash, we don't own a reference to it.
6691 * However it does have a back reference to us, which
6692 * needs to be cleared. */
6693 if ((stash = GvSTASH(sv)))
6694 sv_del_backref(MUTABLE_SV(stash), sv);
6696 /* FIXME. There are probably more unreferenced pointers to SVs
6697 * in the interpreter struct that we should check and tidy in
6698 * a similar fashion to this: */
6699 /* See also S_sv_unglob, which does the same thing. */
6700 if ((const GV *)sv == PL_last_in_gv)
6701 PL_last_in_gv = NULL;
6702 else if ((const GV *)sv == PL_statgv)
6704 else if ((const GV *)sv == PL_stderrgv)
6713 /* Don't bother with SvOOK_off(sv); as we're only going to
6717 SvOOK_offset(sv, offset);
6718 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6719 /* Don't even bother with turning off the OOK flag. */
6724 SV * const target = SvRV(sv);
6726 sv_del_backref(target, sv);
6732 else if (SvPVX_const(sv)
6733 && !(SvTYPE(sv) == SVt_PVIO
6734 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6738 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6742 if (CowREFCNT(sv)) {
6749 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6754 Safefree(SvPVX_mutable(sv));
6758 else if (SvPVX_const(sv) && SvLEN(sv)
6759 && !(SvTYPE(sv) == SVt_PVIO
6760 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6761 Safefree(SvPVX_mutable(sv));
6762 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6763 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6773 SvFLAGS(sv) &= SVf_BREAK;
6774 SvFLAGS(sv) |= SVTYPEMASK;
6776 sv_type_details = bodies_by_type + type;
6777 if (sv_type_details->arena) {
6778 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6779 &PL_body_roots[type]);
6781 else if (sv_type_details->body_size) {
6782 safefree(SvANY(sv));
6786 /* caller is responsible for freeing the head of the original sv */
6787 if (sv != orig_sv && !SvREFCNT(sv))
6790 /* grab and free next sv, if any */
6798 else if (!iter_sv) {
6800 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6801 AV *const av = (AV*)iter_sv;
6802 if (AvFILLp(av) > -1) {
6803 sv = AvARRAY(av)[AvFILLp(av)--];
6805 else { /* no more elements of current AV to free */
6808 /* restore previous value, squirrelled away */
6809 iter_sv = AvARRAY(av)[AvMAX(av)];
6810 Safefree(AvALLOC(av));
6813 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6814 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6815 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6816 /* no more elements of current HV to free */
6819 /* Restore previous values of iter_sv and hash_index,
6820 * squirrelled away */
6821 assert(!SvOBJECT(sv));
6822 iter_sv = (SV*)SvSTASH(sv);
6823 assert(!SvMAGICAL(sv));
6824 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6826 /* perl -DA does not like rubbish in SvMAGIC. */
6830 /* free any remaining detritus from the hash struct */
6831 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6832 assert(!HvARRAY((HV*)sv));
6837 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6841 if (!SvREFCNT(sv)) {
6845 if (--(SvREFCNT(sv)))
6849 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6850 "Attempt to free temp prematurely: SV 0x%" UVxf
6851 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6855 if (SvIMMORTAL(sv)) {
6856 /* make sure SvREFCNT(sv)==0 happens very seldom */
6857 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6866 /* This routine curses the sv itself, not the object referenced by sv. So
6867 sv does not have to be ROK. */
6870 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6871 PERL_ARGS_ASSERT_CURSE;
6872 assert(SvOBJECT(sv));
6874 if (PL_defstash && /* Still have a symbol table? */
6880 stash = SvSTASH(sv);
6881 assert(SvTYPE(stash) == SVt_PVHV);
6882 if (HvNAME(stash)) {
6883 CV* destructor = NULL;
6884 struct mro_meta *meta;
6886 assert (SvOOK(stash));
6888 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6891 /* don't make this an initialization above the assert, since it needs
6893 meta = HvMROMETA(stash);
6894 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6895 destructor = meta->destroy;
6896 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6897 (void *)destructor, HvNAME(stash)) );
6900 bool autoload = FALSE;
6902 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6904 destructor = GvCV(gv);
6906 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6907 GV_AUTOLOAD_ISMETHOD);
6909 destructor = GvCV(gv);
6913 /* we don't cache AUTOLOAD for DESTROY, since this code
6914 would then need to set $__PACKAGE__::AUTOLOAD, or the
6915 equivalent for XS AUTOLOADs */
6917 meta->destroy_gen = PL_sub_generation;
6918 meta->destroy = destructor;
6920 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6921 (void *)destructor, HvNAME(stash)) );
6924 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6928 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6930 /* A constant subroutine can have no side effects, so
6931 don't bother calling it. */
6932 && !CvCONST(destructor)
6933 /* Don't bother calling an empty destructor or one that
6934 returns immediately. */
6935 && (CvISXSUB(destructor)
6936 || (CvSTART(destructor)
6937 && (CvSTART(destructor)->op_next->op_type
6939 && (CvSTART(destructor)->op_next->op_type
6941 || CvSTART(destructor)->op_next->op_next->op_type
6947 SV* const tmpref = newRV(sv);
6948 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6950 PUSHSTACKi(PERLSI_DESTROY);
6955 call_sv(MUTABLE_SV(destructor),
6956 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6960 if(SvREFCNT(tmpref) < 2) {
6961 /* tmpref is not kept alive! */
6963 SvRV_set(tmpref, NULL);
6966 SvREFCNT_dec_NN(tmpref);
6969 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6972 if (check_refcnt && SvREFCNT(sv)) {
6973 if (PL_in_clean_objs)
6975 "DESTROY created new reference to dead object '%" HEKf "'",
6976 HEKfARG(HvNAME_HEK(stash)));
6977 /* DESTROY gave object new lease on life */
6983 HV * const stash = SvSTASH(sv);
6984 /* Curse before freeing the stash, as freeing the stash could cause
6985 a recursive call into S_curse. */
6986 SvOBJECT_off(sv); /* Curse the object. */
6987 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6988 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6994 =for apidoc sv_newref
6996 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7003 Perl_sv_newref(pTHX_ SV *const sv)
7005 PERL_UNUSED_CONTEXT;
7014 Decrement an SV's reference count, and if it drops to zero, call
7015 C<sv_clear> to invoke destructors and free up any memory used by
7016 the body; finally, deallocating the SV's head itself.
7017 Normally called via a wrapper macro C<SvREFCNT_dec>.
7023 Perl_sv_free(pTHX_ SV *const sv)
7029 /* Private helper function for SvREFCNT_dec().
7030 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7033 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7037 PERL_ARGS_ASSERT_SV_FREE2;
7039 if (LIKELY( rc == 1 )) {
7045 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7046 "Attempt to free temp prematurely: SV 0x%" UVxf
7047 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7051 if (SvIMMORTAL(sv)) {
7052 /* make sure SvREFCNT(sv)==0 happens very seldom */
7053 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7057 if (! SvREFCNT(sv)) /* may have have been resurrected */
7062 /* handle exceptional cases */
7066 if (SvFLAGS(sv) & SVf_BREAK)
7067 /* this SV's refcnt has been artificially decremented to
7068 * trigger cleanup */
7070 if (PL_in_clean_all) /* All is fair */
7072 if (SvIMMORTAL(sv)) {
7073 /* make sure SvREFCNT(sv)==0 happens very seldom */
7074 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7077 if (ckWARN_d(WARN_INTERNAL)) {
7078 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7079 Perl_dump_sv_child(aTHX_ sv);
7081 #ifdef DEBUG_LEAKING_SCALARS
7084 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7085 if (PL_warnhook == PERL_WARNHOOK_FATAL
7086 || ckDEAD(packWARN(WARN_INTERNAL))) {
7087 /* Don't let Perl_warner cause us to escape our fate: */
7091 /* This may not return: */
7092 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7093 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7094 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7097 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7107 Returns the length of the string in the SV. Handles magic and type
7108 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7109 gives raw access to the C<xpv_cur> slot.
7115 Perl_sv_len(pTHX_ SV *const sv)
7122 (void)SvPV_const(sv, len);
7127 =for apidoc sv_len_utf8
7129 Returns the number of characters in the string in an SV, counting wide
7130 UTF-8 bytes as a single character. Handles magic and type coercion.
7136 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7137 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7138 * (Note that the mg_len is not the length of the mg_ptr field.
7139 * This allows the cache to store the character length of the string without
7140 * needing to malloc() extra storage to attach to the mg_ptr.)
7145 Perl_sv_len_utf8(pTHX_ SV *const sv)
7151 return sv_len_utf8_nomg(sv);
7155 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7158 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7160 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7162 if (PL_utf8cache && SvUTF8(sv)) {
7164 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7166 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7167 if (mg->mg_len != -1)
7170 /* We can use the offset cache for a headstart.
7171 The longer value is stored in the first pair. */
7172 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7174 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7178 if (PL_utf8cache < 0) {
7179 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7180 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7184 ulen = Perl_utf8_length(aTHX_ s, s + len);
7185 utf8_mg_len_cache_update(sv, &mg, ulen);
7189 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7192 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7195 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7196 STRLEN *const uoffset_p, bool *const at_end)
7198 const U8 *s = start;
7199 STRLEN uoffset = *uoffset_p;
7201 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7203 while (s < send && uoffset) {
7210 else if (s > send) {
7212 /* This is the existing behaviour. Possibly it should be a croak, as
7213 it's actually a bounds error */
7216 *uoffset_p -= uoffset;
7220 /* Given the length of the string in both bytes and UTF-8 characters, decide
7221 whether to walk forwards or backwards to find the byte corresponding to
7222 the passed in UTF-8 offset. */
7224 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7225 STRLEN uoffset, const STRLEN uend)
7227 STRLEN backw = uend - uoffset;
7229 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7231 if (uoffset < 2 * backw) {
7232 /* The assumption is that going forwards is twice the speed of going
7233 forward (that's where the 2 * backw comes from).
7234 (The real figure of course depends on the UTF-8 data.) */
7235 const U8 *s = start;
7237 while (s < send && uoffset--)
7247 while (UTF8_IS_CONTINUATION(*send))
7250 return send - start;
7253 /* For the string representation of the given scalar, find the byte
7254 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7255 give another position in the string, *before* the sought offset, which
7256 (which is always true, as 0, 0 is a valid pair of positions), which should
7257 help reduce the amount of linear searching.
7258 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7259 will be used to reduce the amount of linear searching. The cache will be
7260 created if necessary, and the found value offered to it for update. */
7262 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7263 const U8 *const send, STRLEN uoffset,
7264 STRLEN uoffset0, STRLEN boffset0)
7266 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7268 bool at_end = FALSE;
7270 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7272 assert (uoffset >= uoffset0);
7277 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7279 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7280 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7281 if ((*mgp)->mg_ptr) {
7282 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7283 if (cache[0] == uoffset) {
7284 /* An exact match. */
7287 if (cache[2] == uoffset) {
7288 /* An exact match. */
7292 if (cache[0] < uoffset) {
7293 /* The cache already knows part of the way. */
7294 if (cache[0] > uoffset0) {
7295 /* The cache knows more than the passed in pair */
7296 uoffset0 = cache[0];
7297 boffset0 = cache[1];
7299 if ((*mgp)->mg_len != -1) {
7300 /* And we know the end too. */
7302 + sv_pos_u2b_midway(start + boffset0, send,
7304 (*mgp)->mg_len - uoffset0);
7306 uoffset -= uoffset0;
7308 + sv_pos_u2b_forwards(start + boffset0,
7309 send, &uoffset, &at_end);
7310 uoffset += uoffset0;
7313 else if (cache[2] < uoffset) {
7314 /* We're between the two cache entries. */
7315 if (cache[2] > uoffset0) {
7316 /* and the cache knows more than the passed in pair */
7317 uoffset0 = cache[2];
7318 boffset0 = cache[3];
7322 + sv_pos_u2b_midway(start + boffset0,
7325 cache[0] - uoffset0);
7328 + sv_pos_u2b_midway(start + boffset0,
7331 cache[2] - uoffset0);
7335 else if ((*mgp)->mg_len != -1) {
7336 /* If we can take advantage of a passed in offset, do so. */
7337 /* In fact, offset0 is either 0, or less than offset, so don't
7338 need to worry about the other possibility. */
7340 + sv_pos_u2b_midway(start + boffset0, send,
7342 (*mgp)->mg_len - uoffset0);
7347 if (!found || PL_utf8cache < 0) {
7348 STRLEN real_boffset;
7349 uoffset -= uoffset0;
7350 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7351 send, &uoffset, &at_end);
7352 uoffset += uoffset0;
7354 if (found && PL_utf8cache < 0)
7355 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7357 boffset = real_boffset;
7360 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7362 utf8_mg_len_cache_update(sv, mgp, uoffset);
7364 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7371 =for apidoc sv_pos_u2b_flags
7373 Converts the offset from a count of UTF-8 chars from
7374 the start of the string, to a count of the equivalent number of bytes; if
7375 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7376 C<offset>, rather than from the start
7377 of the string. Handles type coercion.
7378 C<flags> is passed to C<SvPV_flags>, and usually should be
7379 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7385 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7386 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7387 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7392 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7399 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7401 start = (U8*)SvPV_flags(sv, len, flags);
7403 const U8 * const send = start + len;
7405 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7408 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7409 is 0, and *lenp is already set to that. */) {
7410 /* Convert the relative offset to absolute. */
7411 const STRLEN uoffset2 = uoffset + *lenp;
7412 const STRLEN boffset2
7413 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7414 uoffset, boffset) - boffset;
7428 =for apidoc sv_pos_u2b
7430 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7431 the start of the string, to a count of the equivalent number of bytes; if
7432 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7433 the offset, rather than from the start of the string. Handles magic and
7436 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7443 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7444 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7445 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7449 /* This function is subject to size and sign problems */
7452 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7454 PERL_ARGS_ASSERT_SV_POS_U2B;
7457 STRLEN ulen = (STRLEN)*lenp;
7458 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7459 SV_GMAGIC|SV_CONST_RETURN);
7462 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7463 SV_GMAGIC|SV_CONST_RETURN);
7468 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7471 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7472 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7475 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7476 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7477 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7481 (*mgp)->mg_len = ulen;
7484 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7485 byte length pairing. The (byte) length of the total SV is passed in too,
7486 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7487 may not have updated SvCUR, so we can't rely on reading it directly.
7489 The proffered utf8/byte length pairing isn't used if the cache already has
7490 two pairs, and swapping either for the proffered pair would increase the
7491 RMS of the intervals between known byte offsets.
7493 The cache itself consists of 4 STRLEN values
7494 0: larger UTF-8 offset
7495 1: corresponding byte offset
7496 2: smaller UTF-8 offset
7497 3: corresponding byte offset
7499 Unused cache pairs have the value 0, 0.
7500 Keeping the cache "backwards" means that the invariant of
7501 cache[0] >= cache[2] is maintained even with empty slots, which means that
7502 the code that uses it doesn't need to worry if only 1 entry has actually
7503 been set to non-zero. It also makes the "position beyond the end of the
7504 cache" logic much simpler, as the first slot is always the one to start
7508 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7509 const STRLEN utf8, const STRLEN blen)
7513 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7518 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7519 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7520 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7522 (*mgp)->mg_len = -1;
7526 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7527 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7528 (*mgp)->mg_ptr = (char *) cache;
7532 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7533 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7534 a pointer. Note that we no longer cache utf8 offsets on refer-
7535 ences, but this check is still a good idea, for robustness. */
7536 const U8 *start = (const U8 *) SvPVX_const(sv);
7537 const STRLEN realutf8 = utf8_length(start, start + byte);
7539 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7543 /* Cache is held with the later position first, to simplify the code
7544 that deals with unbounded ends. */
7546 ASSERT_UTF8_CACHE(cache);
7547 if (cache[1] == 0) {
7548 /* Cache is totally empty */
7551 } else if (cache[3] == 0) {
7552 if (byte > cache[1]) {
7553 /* New one is larger, so goes first. */
7554 cache[2] = cache[0];
7555 cache[3] = cache[1];
7563 /* float casts necessary? XXX */
7564 #define THREEWAY_SQUARE(a,b,c,d) \
7565 ((float)((d) - (c))) * ((float)((d) - (c))) \
7566 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7567 + ((float)((b) - (a))) * ((float)((b) - (a)))
7569 /* Cache has 2 slots in use, and we know three potential pairs.
7570 Keep the two that give the lowest RMS distance. Do the
7571 calculation in bytes simply because we always know the byte
7572 length. squareroot has the same ordering as the positive value,
7573 so don't bother with the actual square root. */
7574 if (byte > cache[1]) {
7575 /* New position is after the existing pair of pairs. */
7576 const float keep_earlier
7577 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7578 const float keep_later
7579 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7581 if (keep_later < keep_earlier) {
7582 cache[2] = cache[0];
7583 cache[3] = cache[1];
7589 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7590 float b, c, keep_earlier;
7591 if (byte > cache[3]) {
7592 /* New position is between the existing pair of pairs. */
7593 b = (float)cache[3];
7596 /* New position is before the existing pair of pairs. */
7598 c = (float)cache[3];
7600 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7601 if (byte > cache[3]) {
7602 if (keep_later < keep_earlier) {
7612 if (! (keep_later < keep_earlier)) {
7613 cache[0] = cache[2];
7614 cache[1] = cache[3];
7621 ASSERT_UTF8_CACHE(cache);
7624 /* We already know all of the way, now we may be able to walk back. The same
7625 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7626 backward is half the speed of walking forward. */
7628 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7629 const U8 *end, STRLEN endu)
7631 const STRLEN forw = target - s;
7632 STRLEN backw = end - target;
7634 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7636 if (forw < 2 * backw) {
7637 return utf8_length(s, target);
7640 while (end > target) {
7642 while (UTF8_IS_CONTINUATION(*end)) {
7651 =for apidoc sv_pos_b2u_flags
7653 Converts C<offset> from a count of bytes from the start of the string, to
7654 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7655 C<flags> is passed to C<SvPV_flags>, and usually should be
7656 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7662 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7663 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7668 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7671 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7677 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7679 s = (const U8*)SvPV_flags(sv, blen, flags);
7682 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7683 ", byte=%" UVuf, (UV)blen, (UV)offset);
7689 && SvTYPE(sv) >= SVt_PVMG
7690 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7693 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7694 if (cache[1] == offset) {
7695 /* An exact match. */
7698 if (cache[3] == offset) {
7699 /* An exact match. */
7703 if (cache[1] < offset) {
7704 /* We already know part of the way. */
7705 if (mg->mg_len != -1) {
7706 /* Actually, we know the end too. */
7708 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7709 s + blen, mg->mg_len - cache[0]);
7711 len = cache[0] + utf8_length(s + cache[1], send);
7714 else if (cache[3] < offset) {
7715 /* We're between the two cached pairs, so we do the calculation
7716 offset by the byte/utf-8 positions for the earlier pair,
7717 then add the utf-8 characters from the string start to
7719 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7720 s + cache[1], cache[0] - cache[2])
7724 else { /* cache[3] > offset */
7725 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7729 ASSERT_UTF8_CACHE(cache);
7731 } else if (mg->mg_len != -1) {
7732 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7736 if (!found || PL_utf8cache < 0) {
7737 const STRLEN real_len = utf8_length(s, send);
7739 if (found && PL_utf8cache < 0)
7740 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7746 utf8_mg_len_cache_update(sv, &mg, len);
7748 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7755 =for apidoc sv_pos_b2u
7757 Converts the value pointed to by C<offsetp> from a count of bytes from the
7758 start of the string, to a count of the equivalent number of UTF-8 chars.
7759 Handles magic and type coercion.
7761 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7768 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7769 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7774 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7776 PERL_ARGS_ASSERT_SV_POS_B2U;
7781 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7782 SV_GMAGIC|SV_CONST_RETURN);
7786 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7787 STRLEN real, SV *const sv)
7789 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7791 /* As this is debugging only code, save space by keeping this test here,
7792 rather than inlining it in all the callers. */
7793 if (from_cache == real)
7796 /* Need to turn the assertions off otherwise we may recurse infinitely
7797 while printing error messages. */
7798 SAVEI8(PL_utf8cache);
7800 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7801 func, (UV) from_cache, (UV) real, SVfARG(sv));
7807 Returns a boolean indicating whether the strings in the two SVs are
7808 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7809 coerce its args to strings if necessary.
7811 =for apidoc sv_eq_flags
7813 Returns a boolean indicating whether the strings in the two SVs are
7814 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7815 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7821 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7828 SV* svrecode = NULL;
7835 /* if pv1 and pv2 are the same, second SvPV_const call may
7836 * invalidate pv1 (if we are handling magic), so we may need to
7838 if (sv1 == sv2 && flags & SV_GMAGIC
7839 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7840 pv1 = SvPV_const(sv1, cur1);
7841 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7843 pv1 = SvPV_flags_const(sv1, cur1, flags);
7851 pv2 = SvPV_flags_const(sv2, cur2, flags);
7853 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7854 /* Differing utf8ness. */
7856 /* sv1 is the UTF-8 one */
7857 return bytes_cmp_utf8((const U8*)pv2, cur2,
7858 (const U8*)pv1, cur1) == 0;
7861 /* sv2 is the UTF-8 one */
7862 return bytes_cmp_utf8((const U8*)pv1, cur1,
7863 (const U8*)pv2, cur2) == 0;
7868 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7870 SvREFCNT_dec(svrecode);
7878 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7879 string in C<sv1> is less than, equal to, or greater than the string in
7880 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7881 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7883 =for apidoc sv_cmp_flags
7885 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7886 string in C<sv1> is less than, equal to, or greater than the string in
7887 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7888 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7889 also C<L</sv_cmp_locale_flags>>.
7895 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7897 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7901 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7905 const char *pv1, *pv2;
7907 SV *svrecode = NULL;
7914 pv1 = SvPV_flags_const(sv1, cur1, flags);
7921 pv2 = SvPV_flags_const(sv2, cur2, flags);
7923 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7924 /* Differing utf8ness. */
7926 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7927 (const U8*)pv1, cur1);
7928 return retval ? retval < 0 ? -1 : +1 : 0;
7931 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7932 (const U8*)pv2, cur2);
7933 return retval ? retval < 0 ? -1 : +1 : 0;
7937 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7940 cmp = cur2 ? -1 : 0;
7944 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7947 if (! DO_UTF8(sv1)) {
7949 const I32 retval = memcmp((const void*)pv1,
7953 cmp = retval < 0 ? -1 : 1;
7954 } else if (cur1 == cur2) {
7957 cmp = cur1 < cur2 ? -1 : 1;
7961 else { /* Both are to be treated as UTF-EBCDIC */
7963 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7964 * which remaps code points 0-255. We therefore generally have to
7965 * unmap back to the original values to get an accurate comparison.
7966 * But we don't have to do that for UTF-8 invariants, as by
7967 * definition, they aren't remapped, nor do we have to do it for
7968 * above-latin1 code points, as they also aren't remapped. (This
7969 * code also works on ASCII platforms, but the memcmp() above is
7972 const char *e = pv1 + shortest_len;
7974 /* Find the first bytes that differ between the two strings */
7975 while (pv1 < e && *pv1 == *pv2) {
7981 if (pv1 == e) { /* Are the same all the way to the end */
7985 cmp = cur1 < cur2 ? -1 : 1;
7988 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7989 * in the strings were. The current bytes may or may not be
7990 * at the beginning of a character. But neither or both are
7991 * (or else earlier bytes would have been different). And
7992 * if we are in the middle of a character, the two
7993 * characters are comprised of the same number of bytes
7994 * (because in this case the start bytes are the same, and
7995 * the start bytes encode the character's length). */
7996 if (UTF8_IS_INVARIANT(*pv1))
7998 /* If both are invariants; can just compare directly */
7999 if (UTF8_IS_INVARIANT(*pv2)) {
8000 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8002 else /* Since *pv1 is invariant, it is the whole character,
8003 which means it is at the beginning of a character.
8004 That means pv2 is also at the beginning of a
8005 character (see earlier comment). Since it isn't
8006 invariant, it must be a start byte. If it starts a
8007 character whose code point is above 255, that
8008 character is greater than any single-byte char, which
8010 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8015 /* Here, pv2 points to a character composed of 2 bytes
8016 * whose code point is < 256. Get its code point and
8017 * compare with *pv1 */
8018 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8023 else /* The code point starting at pv1 isn't a single byte */
8024 if (UTF8_IS_INVARIANT(*pv2))
8026 /* But here, the code point starting at *pv2 is a single byte,
8027 * and so *pv1 must begin a character, hence is a start byte.
8028 * If that character is above 255, it is larger than any
8029 * single-byte char, which *pv2 is */
8030 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8034 /* Here, pv1 points to a character composed of 2 bytes
8035 * whose code point is < 256. Get its code point and
8036 * compare with the single byte character *pv2 */
8037 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8042 else /* Here, we've ruled out either *pv1 and *pv2 being
8043 invariant. That means both are part of variants, but not
8044 necessarily at the start of a character */
8045 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8046 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8048 /* Here, at least one is the start of a character, which means
8049 * the other is also a start byte. And the code point of at
8050 * least one of the characters is above 255. It is a
8051 * characteristic of UTF-EBCDIC that all start bytes for
8052 * above-latin1 code points are well behaved as far as code
8053 * point comparisons go, and all are larger than all other
8054 * start bytes, so the comparison with those is also well
8056 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8059 /* Here both *pv1 and *pv2 are part of variant characters.
8060 * They could be both continuations, or both start characters.
8061 * (One or both could even be an illegal start character (for
8062 * an overlong) which for the purposes of sorting we treat as
8064 if (UTF8_IS_CONTINUATION(*pv1)) {
8066 /* If they are continuations for code points above 255,
8067 * then comparing the current byte is sufficient, as there
8068 * is no remapping of these and so the comparison is
8069 * well-behaved. We determine if they are such
8070 * continuations by looking at the preceding byte. It
8071 * could be a start byte, from which we can tell if it is
8072 * for an above 255 code point. Or it could be a
8073 * continuation, which means the character occupies at
8074 * least 3 bytes, so must be above 255. */
8075 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8076 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8078 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8082 /* Here, the continuations are for code points below 256;
8083 * back up one to get to the start byte */
8088 /* We need to get the actual native code point of each of these
8089 * variants in order to compare them */
8090 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8091 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8100 SvREFCNT_dec(svrecode);
8106 =for apidoc sv_cmp_locale
8108 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8109 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8110 if necessary. See also C<L</sv_cmp>>.
8112 =for apidoc sv_cmp_locale_flags
8114 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8115 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8116 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8117 C<L</sv_cmp_flags>>.
8123 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8125 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8129 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8132 #ifdef USE_LOCALE_COLLATE
8138 if (PL_collation_standard)
8143 /* Revert to using raw compare if both operands exist, but either one
8144 * doesn't transform properly for collation */
8146 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8150 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8156 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8157 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8160 if (!pv1 || !len1) {
8171 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8174 return retval < 0 ? -1 : 1;
8177 * When the result of collation is equality, that doesn't mean
8178 * that there are no differences -- some locales exclude some
8179 * characters from consideration. So to avoid false equalities,
8180 * we use the raw string as a tiebreaker.
8187 PERL_UNUSED_ARG(flags);
8188 #endif /* USE_LOCALE_COLLATE */
8190 return sv_cmp(sv1, sv2);
8194 #ifdef USE_LOCALE_COLLATE
8197 =for apidoc sv_collxfrm
8199 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8200 C<L</sv_collxfrm_flags>>.
8202 =for apidoc sv_collxfrm_flags
8204 Add Collate Transform magic to an SV if it doesn't already have it. If the
8205 flags contain C<SV_GMAGIC>, it handles get-magic.
8207 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8208 scalar data of the variable, but transformed to such a format that a normal
8209 memory comparison can be used to compare the data according to the locale
8216 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8220 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8222 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8224 /* If we don't have collation magic on 'sv', or the locale has changed
8225 * since the last time we calculated it, get it and save it now */
8226 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8231 /* Free the old space */
8233 Safefree(mg->mg_ptr);
8235 s = SvPV_flags_const(sv, len, flags);
8236 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8238 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8253 if (mg && mg->mg_ptr) {
8255 return mg->mg_ptr + sizeof(PL_collation_ix);
8263 #endif /* USE_LOCALE_COLLATE */
8266 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8268 SV * const tsv = newSV(0);
8271 sv_gets(tsv, fp, 0);
8272 sv_utf8_upgrade_nomg(tsv);
8273 SvCUR_set(sv,append);
8276 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8280 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8283 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8284 /* Grab the size of the record we're getting */
8285 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8292 /* With a true, record-oriented file on VMS, we need to use read directly
8293 * to ensure that we respect RMS record boundaries. The user is responsible
8294 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8295 * record size) field. N.B. This is likely to produce invalid results on
8296 * varying-width character data when a record ends mid-character.
8298 fd = PerlIO_fileno(fp);
8300 && PerlLIO_fstat(fd, &st) == 0
8301 && (st.st_fab_rfm == FAB$C_VAR
8302 || st.st_fab_rfm == FAB$C_VFC
8303 || st.st_fab_rfm == FAB$C_FIX)) {
8305 bytesread = PerlLIO_read(fd, buffer, recsize);
8307 else /* in-memory file from PerlIO::Scalar
8308 * or not a record-oriented file
8312 bytesread = PerlIO_read(fp, buffer, recsize);
8314 /* At this point, the logic in sv_get() means that sv will
8315 be treated as utf-8 if the handle is utf8.
8317 if (PerlIO_isutf8(fp) && bytesread > 0) {
8318 char *bend = buffer + bytesread;
8319 char *bufp = buffer;
8320 size_t charcount = 0;
8321 bool charstart = TRUE;
8324 while (charcount < recsize) {
8325 /* count accumulated characters */
8326 while (bufp < bend) {
8328 skip = UTF8SKIP(bufp);
8330 if (bufp + skip > bend) {
8331 /* partial at the end */
8342 if (charcount < recsize) {
8344 STRLEN bufp_offset = bufp - buffer;
8345 SSize_t morebytesread;
8347 /* originally I read enough to fill any incomplete
8348 character and the first byte of the next
8349 character if needed, but if there's many
8350 multi-byte encoded characters we're going to be
8351 making a read call for every character beyond
8352 the original read size.
8354 So instead, read the rest of the character if
8355 any, and enough bytes to match at least the
8356 start bytes for each character we're going to
8360 readsize = recsize - charcount;
8362 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8363 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8364 bend = buffer + bytesread;
8365 morebytesread = PerlIO_read(fp, bend, readsize);
8366 if (morebytesread <= 0) {
8367 /* we're done, if we still have incomplete
8368 characters the check code in sv_gets() will
8371 I'd originally considered doing
8372 PerlIO_ungetc() on all but the lead
8373 character of the incomplete character, but
8374 read() doesn't do that, so I don't.
8379 /* prepare to scan some more */
8380 bytesread += morebytesread;
8381 bend = buffer + bytesread;
8382 bufp = buffer + bufp_offset;
8390 SvCUR_set(sv, bytesread + append);
8391 buffer[bytesread] = '\0';
8392 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8398 Get a line from the filehandle and store it into the SV, optionally
8399 appending to the currently-stored string. If C<append> is not 0, the
8400 line is appended to the SV instead of overwriting it. C<append> should
8401 be set to the byte offset that the appended string should start at
8402 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8408 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8418 PERL_ARGS_ASSERT_SV_GETS;
8420 if (SvTHINKFIRST(sv))
8421 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8422 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8424 However, perlbench says it's slower, because the existing swipe code
8425 is faster than copy on write.
8426 Swings and roundabouts. */
8427 SvUPGRADE(sv, SVt_PV);
8430 /* line is going to be appended to the existing buffer in the sv */
8431 if (PerlIO_isutf8(fp)) {
8433 sv_utf8_upgrade_nomg(sv);
8434 sv_pos_u2b(sv,&append,0);
8436 } else if (SvUTF8(sv)) {
8437 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8443 /* not appending - "clear" the string by setting SvCUR to 0,
8444 * the pv is still avaiable. */
8447 if (PerlIO_isutf8(fp))
8450 if (IN_PERL_COMPILETIME) {
8451 /* we always read code in line mode */
8455 else if (RsSNARF(PL_rs)) {
8456 /* If it is a regular disk file use size from stat() as estimate
8457 of amount we are going to read -- may result in mallocing
8458 more memory than we really need if the layers below reduce
8459 the size we read (e.g. CRLF or a gzip layer).
8462 int fd = PerlIO_fileno(fp);
8463 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8464 const Off_t offset = PerlIO_tell(fp);
8465 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8466 #ifdef PERL_COPY_ON_WRITE
8467 /* Add an extra byte for the sake of copy-on-write's
8468 * buffer reference count. */
8469 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8471 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8478 else if (RsRECORD(PL_rs)) {
8479 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8481 else if (RsPARA(PL_rs)) {
8487 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8488 if (PerlIO_isutf8(fp)) {
8489 rsptr = SvPVutf8(PL_rs, rslen);
8492 if (SvUTF8(PL_rs)) {
8493 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8494 Perl_croak(aTHX_ "Wide character in $/");
8497 /* extract the raw pointer to the record separator */
8498 rsptr = SvPV_const(PL_rs, rslen);
8502 /* rslast is the last character in the record separator
8503 * note we don't use rslast except when rslen is true, so the
8504 * null assign is a placeholder. */
8505 rslast = rslen ? rsptr[rslen - 1] : '\0';
8507 if (rspara) { /* have to do this both before and after */
8508 do { /* to make sure file boundaries work right */
8511 i = PerlIO_getc(fp);
8515 PerlIO_ungetc(fp,i);
8521 /* See if we know enough about I/O mechanism to cheat it ! */
8523 /* This used to be #ifdef test - it is made run-time test for ease
8524 of abstracting out stdio interface. One call should be cheap
8525 enough here - and may even be a macro allowing compile
8529 if (PerlIO_fast_gets(fp)) {
8531 * We can do buffer based IO operations on this filehandle.
8533 * This means we can bypass a lot of subcalls and process
8534 * the buffer directly, it also means we know the upper bound
8535 * on the amount of data we might read of the current buffer
8536 * into our sv. Knowing this allows us to preallocate the pv
8537 * to be able to hold that maximum, which allows us to simplify
8538 * a lot of logic. */
8541 * We're going to steal some values from the stdio struct
8542 * and put EVERYTHING in the innermost loop into registers.
8544 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8545 STRLEN bpx; /* length of the data in the target sv
8546 used to fix pointers after a SvGROW */
8547 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8548 of data left in the read-ahead buffer.
8549 If 0 then the pv buffer can hold the full
8550 amount left, otherwise this is the amount it
8553 /* Here is some breathtakingly efficient cheating */
8555 /* When you read the following logic resist the urge to think
8556 * of record separators that are 1 byte long. They are an
8557 * uninteresting special (simple) case.
8559 * Instead think of record separators which are at least 2 bytes
8560 * long, and keep in mind that we need to deal with such
8561 * separators when they cross a read-ahead buffer boundary.
8563 * Also consider that we need to gracefully deal with separators
8564 * that may be longer than a single read ahead buffer.
8566 * Lastly do not forget we want to copy the delimiter as well. We
8567 * are copying all data in the file _up_to_and_including_ the separator
8570 * Now that you have all that in mind here is what is happening below:
8572 * 1. When we first enter the loop we do some memory book keeping to see
8573 * how much free space there is in the target SV. (This sub assumes that
8574 * it is operating on the same SV most of the time via $_ and that it is
8575 * going to be able to reuse the same pv buffer each call.) If there is
8576 * "enough" room then we set "shortbuffered" to how much space there is
8577 * and start reading forward.
8579 * 2. When we scan forward we copy from the read-ahead buffer to the target
8580 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8581 * and the end of the of pv, as well as for the "rslast", which is the last
8582 * char of the separator.
8584 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8585 * (which has a "complete" record up to the point we saw rslast) and check
8586 * it to see if it matches the separator. If it does we are done. If it doesn't
8587 * we continue on with the scan/copy.
8589 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8590 * the IO system to read the next buffer. We do this by doing a getc(), which
8591 * returns a single char read (or EOF), and prefills the buffer, and also
8592 * allows us to find out how full the buffer is. We use this information to
8593 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8594 * the returned single char into the target sv, and then go back into scan
8597 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8598 * remaining space in the read-buffer.
8600 * Note that this code despite its twisty-turny nature is pretty darn slick.
8601 * It manages single byte separators, multi-byte cross boundary separators,
8602 * and cross-read-buffer separators cleanly and efficiently at the cost
8603 * of potentially greatly overallocating the target SV.
8609 /* get the number of bytes remaining in the read-ahead buffer
8610 * on first call on a given fp this will return 0.*/
8611 cnt = PerlIO_get_cnt(fp);
8613 /* make sure we have the room */
8614 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8615 /* Not room for all of it
8616 if we are looking for a separator and room for some
8618 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8619 /* just process what we have room for */
8620 shortbuffered = cnt - SvLEN(sv) + append + 1;
8621 cnt -= shortbuffered;
8624 /* ensure that the target sv has enough room to hold
8625 * the rest of the read-ahead buffer */
8627 /* remember that cnt can be negative */
8628 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8632 /* we have enough room to hold the full buffer, lets scream */
8636 /* extract the pointer to sv's string buffer, offset by append as necessary */
8637 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8638 /* extract the point to the read-ahead buffer */
8639 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8641 /* some trace debug output */
8642 DEBUG_P(PerlIO_printf(Perl_debug_log,
8643 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8644 DEBUG_P(PerlIO_printf(Perl_debug_log,
8645 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8647 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8648 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8652 /* if there is stuff left in the read-ahead buffer */
8654 /* if there is a separator */
8656 /* find next rslast */
8659 /* shortcut common case of blank line */
8661 if ((*bp++ = *ptr++) == rslast)
8662 goto thats_all_folks;
8664 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8666 SSize_t got = p - ptr + 1;
8667 Copy(ptr, bp, got, STDCHAR);
8671 goto thats_all_folks;
8673 Copy(ptr, bp, cnt, STDCHAR);
8679 /* no separator, slurp the full buffer */
8680 Copy(ptr, bp, cnt, char); /* this | eat */
8681 bp += cnt; /* screams | dust */
8682 ptr += cnt; /* louder | sed :-) */
8684 assert (!shortbuffered);
8685 goto cannot_be_shortbuffered;
8689 if (shortbuffered) { /* oh well, must extend */
8690 /* we didnt have enough room to fit the line into the target buffer
8691 * so we must extend the target buffer and keep going */
8692 cnt = shortbuffered;
8694 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8696 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8697 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8698 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8702 cannot_be_shortbuffered:
8703 /* we need to refill the read-ahead buffer if possible */
8705 DEBUG_P(PerlIO_printf(Perl_debug_log,
8706 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8707 PTR2UV(ptr),(IV)cnt));
8708 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8710 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8711 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8712 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8713 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8716 call PerlIO_getc() to let it prefill the lookahead buffer
8718 This used to call 'filbuf' in stdio form, but as that behaves like
8719 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8720 another abstraction.
8722 Note we have to deal with the char in 'i' if we are not at EOF
8724 i = PerlIO_getc(fp); /* get more characters */
8726 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8727 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8728 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8729 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8731 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8732 cnt = PerlIO_get_cnt(fp);
8733 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8734 DEBUG_P(PerlIO_printf(Perl_debug_log,
8735 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8736 PTR2UV(ptr),(IV)cnt));
8738 if (i == EOF) /* all done for ever? */
8739 goto thats_really_all_folks;
8741 /* make sure we have enough space in the target sv */
8742 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8744 SvGROW(sv, bpx + cnt + 2);
8745 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8747 /* copy of the char we got from getc() */
8748 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8750 /* make sure we deal with the i being the last character of a separator */
8751 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8752 goto thats_all_folks;
8756 /* check if we have actually found the separator - only really applies
8758 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8759 memNE((char*)bp - rslen, rsptr, rslen))
8760 goto screamer; /* go back to the fray */
8761 thats_really_all_folks:
8763 cnt += shortbuffered;
8764 DEBUG_P(PerlIO_printf(Perl_debug_log,
8765 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8766 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8767 DEBUG_P(PerlIO_printf(Perl_debug_log,
8768 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8770 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8771 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8773 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8774 DEBUG_P(PerlIO_printf(Perl_debug_log,
8775 "Screamer: done, len=%ld, string=|%.*s|\n",
8776 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8780 /*The big, slow, and stupid way. */
8781 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8782 STDCHAR *buf = NULL;
8783 Newx(buf, 8192, STDCHAR);
8791 const STDCHAR * const bpe = buf + sizeof(buf);
8793 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8794 ; /* keep reading */
8798 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8799 /* Accommodate broken VAXC compiler, which applies U8 cast to
8800 * both args of ?: operator, causing EOF to change into 255
8803 i = (U8)buf[cnt - 1];
8809 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8811 sv_catpvn_nomg(sv, (char *) buf, cnt);
8813 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8815 if (i != EOF && /* joy */
8817 SvCUR(sv) < rslen ||
8818 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8822 * If we're reading from a TTY and we get a short read,
8823 * indicating that the user hit his EOF character, we need
8824 * to notice it now, because if we try to read from the TTY
8825 * again, the EOF condition will disappear.
8827 * The comparison of cnt to sizeof(buf) is an optimization
8828 * that prevents unnecessary calls to feof().
8832 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8836 #ifdef USE_HEAP_INSTEAD_OF_STACK
8841 if (rspara) { /* have to do this both before and after */
8842 while (i != EOF) { /* to make sure file boundaries work right */
8843 i = PerlIO_getc(fp);
8845 PerlIO_ungetc(fp,i);
8851 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8857 Auto-increment of the value in the SV, doing string to numeric conversion
8858 if necessary. Handles 'get' magic and operator overloading.
8864 Perl_sv_inc(pTHX_ SV *const sv)
8873 =for apidoc sv_inc_nomg
8875 Auto-increment of the value in the SV, doing string to numeric conversion
8876 if necessary. Handles operator overloading. Skips handling 'get' magic.
8882 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8889 if (SvTHINKFIRST(sv)) {
8890 if (SvREADONLY(sv)) {
8891 Perl_croak_no_modify();
8895 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8897 i = PTR2IV(SvRV(sv));
8901 else sv_force_normal_flags(sv, 0);
8903 flags = SvFLAGS(sv);
8904 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8905 /* It's (privately or publicly) a float, but not tested as an
8906 integer, so test it to see. */
8908 flags = SvFLAGS(sv);
8910 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8911 /* It's publicly an integer, or privately an integer-not-float */
8912 #ifdef PERL_PRESERVE_IVUV
8916 if (SvUVX(sv) == UV_MAX)
8917 sv_setnv(sv, UV_MAX_P1);
8919 (void)SvIOK_only_UV(sv);
8920 SvUV_set(sv, SvUVX(sv) + 1);
8922 if (SvIVX(sv) == IV_MAX)
8923 sv_setuv(sv, (UV)IV_MAX + 1);
8925 (void)SvIOK_only(sv);
8926 SvIV_set(sv, SvIVX(sv) + 1);
8931 if (flags & SVp_NOK) {
8932 const NV was = SvNVX(sv);
8933 if (LIKELY(!Perl_isinfnan(was)) &&
8934 NV_OVERFLOWS_INTEGERS_AT &&
8935 was >= NV_OVERFLOWS_INTEGERS_AT) {
8936 /* diag_listed_as: Lost precision when %s %f by 1 */
8937 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8938 "Lost precision when incrementing %" NVff " by 1",
8941 (void)SvNOK_only(sv);
8942 SvNV_set(sv, was + 1.0);
8946 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8947 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8948 Perl_croak_no_modify();
8950 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8951 if ((flags & SVTYPEMASK) < SVt_PVIV)
8952 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8953 (void)SvIOK_only(sv);
8958 while (isALPHA(*d)) d++;
8959 while (isDIGIT(*d)) d++;
8960 if (d < SvEND(sv)) {
8961 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8962 #ifdef PERL_PRESERVE_IVUV
8963 /* Got to punt this as an integer if needs be, but we don't issue
8964 warnings. Probably ought to make the sv_iv_please() that does
8965 the conversion if possible, and silently. */
8966 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8967 /* Need to try really hard to see if it's an integer.
8968 9.22337203685478e+18 is an integer.
8969 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8970 so $a="9.22337203685478e+18"; $a+0; $a++
8971 needs to be the same as $a="9.22337203685478e+18"; $a++
8978 /* sv_2iv *should* have made this an NV */
8979 if (flags & SVp_NOK) {
8980 (void)SvNOK_only(sv);
8981 SvNV_set(sv, SvNVX(sv) + 1.0);
8984 /* I don't think we can get here. Maybe I should assert this
8985 And if we do get here I suspect that sv_setnv will croak. NWC
8987 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
8988 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8990 #endif /* PERL_PRESERVE_IVUV */
8991 if (!numtype && ckWARN(WARN_NUMERIC))
8992 not_incrementable(sv);
8993 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8997 while (d >= SvPVX_const(sv)) {
9005 /* MKS: The original code here died if letters weren't consecutive.
9006 * at least it didn't have to worry about non-C locales. The
9007 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9008 * arranged in order (although not consecutively) and that only
9009 * [A-Za-z] are accepted by isALPHA in the C locale.
9011 if (isALPHA_FOLD_NE(*d, 'z')) {
9012 do { ++*d; } while (!isALPHA(*d));
9015 *(d--) -= 'z' - 'a';
9020 *(d--) -= 'z' - 'a' + 1;
9024 /* oh,oh, the number grew */
9025 SvGROW(sv, SvCUR(sv) + 2);
9026 SvCUR_set(sv, SvCUR(sv) + 1);
9027 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9038 Auto-decrement of the value in the SV, doing string to numeric conversion
9039 if necessary. Handles 'get' magic and operator overloading.
9045 Perl_sv_dec(pTHX_ SV *const sv)
9054 =for apidoc sv_dec_nomg
9056 Auto-decrement of the value in the SV, doing string to numeric conversion
9057 if necessary. Handles operator overloading. Skips handling 'get' magic.
9063 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9069 if (SvTHINKFIRST(sv)) {
9070 if (SvREADONLY(sv)) {
9071 Perl_croak_no_modify();
9075 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9077 i = PTR2IV(SvRV(sv));
9081 else sv_force_normal_flags(sv, 0);
9083 /* Unlike sv_inc we don't have to worry about string-never-numbers
9084 and keeping them magic. But we mustn't warn on punting */
9085 flags = SvFLAGS(sv);
9086 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9087 /* It's publicly an integer, or privately an integer-not-float */
9088 #ifdef PERL_PRESERVE_IVUV
9092 if (SvUVX(sv) == 0) {
9093 (void)SvIOK_only(sv);
9097 (void)SvIOK_only_UV(sv);
9098 SvUV_set(sv, SvUVX(sv) - 1);
9101 if (SvIVX(sv) == IV_MIN) {
9102 sv_setnv(sv, (NV)IV_MIN);
9106 (void)SvIOK_only(sv);
9107 SvIV_set(sv, SvIVX(sv) - 1);
9112 if (flags & SVp_NOK) {
9115 const NV was = SvNVX(sv);
9116 if (LIKELY(!Perl_isinfnan(was)) &&
9117 NV_OVERFLOWS_INTEGERS_AT &&
9118 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9119 /* diag_listed_as: Lost precision when %s %f by 1 */
9120 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9121 "Lost precision when decrementing %" NVff " by 1",
9124 (void)SvNOK_only(sv);
9125 SvNV_set(sv, was - 1.0);
9130 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9131 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9132 Perl_croak_no_modify();
9134 if (!(flags & SVp_POK)) {
9135 if ((flags & SVTYPEMASK) < SVt_PVIV)
9136 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9138 (void)SvIOK_only(sv);
9141 #ifdef PERL_PRESERVE_IVUV
9143 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9144 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9145 /* Need to try really hard to see if it's an integer.
9146 9.22337203685478e+18 is an integer.
9147 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9148 so $a="9.22337203685478e+18"; $a+0; $a--
9149 needs to be the same as $a="9.22337203685478e+18"; $a--
9156 /* sv_2iv *should* have made this an NV */
9157 if (flags & SVp_NOK) {
9158 (void)SvNOK_only(sv);
9159 SvNV_set(sv, SvNVX(sv) - 1.0);
9162 /* I don't think we can get here. Maybe I should assert this
9163 And if we do get here I suspect that sv_setnv will croak. NWC
9165 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9166 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9169 #endif /* PERL_PRESERVE_IVUV */
9170 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9173 /* this define is used to eliminate a chunk of duplicated but shared logic
9174 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9175 * used anywhere but here - yves
9177 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9179 SSize_t ix = ++PL_tmps_ix; \
9180 if (UNLIKELY(ix >= PL_tmps_max)) \
9181 ix = tmps_grow_p(ix); \
9182 PL_tmps_stack[ix] = (AnSv); \
9186 =for apidoc sv_mortalcopy
9188 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9189 The new SV is marked as mortal. It will be destroyed "soon", either by an
9190 explicit call to C<FREETMPS>, or by an implicit call at places such as
9191 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9196 /* Make a string that will exist for the duration of the expression
9197 * evaluation. Actually, it may have to last longer than that, but
9198 * hopefully we won't free it until it has been assigned to a
9199 * permanent location. */
9202 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9206 if (flags & SV_GMAGIC)
9207 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9209 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9210 PUSH_EXTEND_MORTAL__SV_C(sv);
9216 =for apidoc sv_newmortal
9218 Creates a new null SV which is mortal. The reference count of the SV is
9219 set to 1. It will be destroyed "soon", either by an explicit call to
9220 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9221 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9227 Perl_sv_newmortal(pTHX)
9232 SvFLAGS(sv) = SVs_TEMP;
9233 PUSH_EXTEND_MORTAL__SV_C(sv);
9239 =for apidoc newSVpvn_flags
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. Note that if C<len> is zero, Perl will create a zero length
9244 string. You are responsible for ensuring that the source string is at least
9245 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9246 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9247 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9248 returning. If C<SVf_UTF8> is set, C<s>
9249 is considered to be in UTF-8 and the
9250 C<SVf_UTF8> flag will be set on the new SV.
9251 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9253 #define newSVpvn_utf8(s, len, u) \
9254 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9260 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9264 /* All the flags we don't support must be zero.
9265 And we're new code so I'm going to assert this from the start. */
9266 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9268 sv_setpvn(sv,s,len);
9270 /* This code used to do a sv_2mortal(), however we now unroll the call to
9271 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9272 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9273 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9274 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9275 * means that we eliminate quite a few steps than it looks - Yves
9276 * (explaining patch by gfx) */
9278 SvFLAGS(sv) |= flags;
9280 if(flags & SVs_TEMP){
9281 PUSH_EXTEND_MORTAL__SV_C(sv);
9288 =for apidoc sv_2mortal
9290 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9291 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9292 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9293 string buffer can be "stolen" if this SV is copied. See also
9294 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9300 Perl_sv_2mortal(pTHX_ SV *const sv)
9307 PUSH_EXTEND_MORTAL__SV_C(sv);
9315 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9316 characters) into it. The reference count for the
9317 SV is set to 1. If C<len> is zero, Perl will compute the length using
9318 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9319 C<NUL> characters and has to have a terminating C<NUL> byte).
9321 For efficiency, consider using C<newSVpvn> instead.
9327 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9332 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9337 =for apidoc newSVpvn
9339 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9340 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9341 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9342 are responsible for ensuring that the source buffer is at least
9343 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9350 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9354 sv_setpvn(sv,buffer,len);
9359 =for apidoc newSVhek
9361 Creates a new SV from the hash key structure. It will generate scalars that
9362 point to the shared string table where possible. Returns a new (undefined)
9363 SV if C<hek> is NULL.
9369 Perl_newSVhek(pTHX_ const HEK *const hek)
9378 if (HEK_LEN(hek) == HEf_SVKEY) {
9379 return newSVsv(*(SV**)HEK_KEY(hek));
9381 const int flags = HEK_FLAGS(hek);
9382 if (flags & HVhek_WASUTF8) {
9384 Andreas would like keys he put in as utf8 to come back as utf8
9386 STRLEN utf8_len = HEK_LEN(hek);
9387 SV * const sv = newSV_type(SVt_PV);
9388 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9389 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9390 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9393 } else if (flags & HVhek_UNSHARED) {
9394 /* A hash that isn't using shared hash keys has to have
9395 the flag in every key so that we know not to try to call
9396 share_hek_hek on it. */
9398 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9403 /* This will be overwhelminly the most common case. */
9405 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9406 more efficient than sharepvn(). */
9410 sv_upgrade(sv, SVt_PV);
9411 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9412 SvCUR_set(sv, HEK_LEN(hek));
9424 =for apidoc newSVpvn_share
9426 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9427 table. If the string does not already exist in the table, it is
9428 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9429 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9430 is non-zero, that value is used; otherwise the hash is computed.
9431 The string's hash can later be retrieved from the SV
9432 with the C<SvSHARED_HASH()> macro. The idea here is
9433 that as the string table is used for shared hash keys these strings will have
9434 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9440 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9444 bool is_utf8 = FALSE;
9445 const char *const orig_src = src;
9448 STRLEN tmplen = -len;
9450 /* See the note in hv.c:hv_fetch() --jhi */
9451 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9455 PERL_HASH(hash, src, len);
9457 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9458 changes here, update it there too. */
9459 sv_upgrade(sv, SVt_PV);
9460 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9467 if (src != orig_src)
9473 =for apidoc newSVpv_share
9475 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9482 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9484 return newSVpvn_share(src, strlen(src), hash);
9487 #if defined(PERL_IMPLICIT_CONTEXT)
9489 /* pTHX_ magic can't cope with varargs, so this is a no-context
9490 * version of the main function, (which may itself be aliased to us).
9491 * Don't access this version directly.
9495 Perl_newSVpvf_nocontext(const char *const pat, ...)
9501 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9503 va_start(args, pat);
9504 sv = vnewSVpvf(pat, &args);
9511 =for apidoc newSVpvf
9513 Creates a new SV and initializes it with the string formatted like
9520 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9525 PERL_ARGS_ASSERT_NEWSVPVF;
9527 va_start(args, pat);
9528 sv = vnewSVpvf(pat, &args);
9533 /* backend for newSVpvf() and newSVpvf_nocontext() */
9536 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9540 PERL_ARGS_ASSERT_VNEWSVPVF;
9543 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9550 Creates a new SV and copies a floating point value into it.
9551 The reference count for the SV is set to 1.
9557 Perl_newSVnv(pTHX_ const NV n)
9569 Creates a new SV and copies an integer into it. The reference count for the
9576 Perl_newSViv(pTHX_ const IV i)
9582 /* Inlining ONLY the small relevant subset of sv_setiv here
9583 * for performance. Makes a significant difference. */
9585 /* We're starting from SVt_FIRST, so provided that's
9586 * actual 0, we don't have to unset any SV type flags
9587 * to promote to SVt_IV. */
9588 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9590 SET_SVANY_FOR_BODYLESS_IV(sv);
9591 SvFLAGS(sv) |= SVt_IV;
9603 Creates a new SV and copies an unsigned integer into it.
9604 The reference count for the SV is set to 1.
9610 Perl_newSVuv(pTHX_ const UV u)
9614 /* Inlining ONLY the small relevant subset of sv_setuv here
9615 * for performance. Makes a significant difference. */
9617 /* Using ivs is more efficient than using uvs - see sv_setuv */
9618 if (u <= (UV)IV_MAX) {
9619 return newSViv((IV)u);
9624 /* We're starting from SVt_FIRST, so provided that's
9625 * actual 0, we don't have to unset any SV type flags
9626 * to promote to SVt_IV. */
9627 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9629 SET_SVANY_FOR_BODYLESS_IV(sv);
9630 SvFLAGS(sv) |= SVt_IV;
9632 (void)SvIsUV_on(sv);
9641 =for apidoc newSV_type
9643 Creates a new SV, of the type specified. The reference count for the new SV
9650 Perl_newSV_type(pTHX_ const svtype type)
9655 ASSUME(SvTYPE(sv) == SVt_FIRST);
9656 if(type != SVt_FIRST)
9657 sv_upgrade(sv, type);
9662 =for apidoc newRV_noinc
9664 Creates an RV wrapper for an SV. The reference count for the original
9665 SV is B<not> incremented.
9671 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9675 PERL_ARGS_ASSERT_NEWRV_NOINC;
9679 /* We're starting from SVt_FIRST, so provided that's
9680 * actual 0, we don't have to unset any SV type flags
9681 * to promote to SVt_IV. */
9682 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9684 SET_SVANY_FOR_BODYLESS_IV(sv);
9685 SvFLAGS(sv) |= SVt_IV;
9690 SvRV_set(sv, tmpRef);
9695 /* newRV_inc is the official function name to use now.
9696 * newRV_inc is in fact #defined to newRV in sv.h
9700 Perl_newRV(pTHX_ SV *const sv)
9702 PERL_ARGS_ASSERT_NEWRV;
9704 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9710 Creates a new SV which is an exact duplicate of the original SV.
9717 Perl_newSVsv(pTHX_ SV *const old)
9723 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9724 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9727 /* Do this here, otherwise we leak the new SV if this croaks. */
9730 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9731 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9732 sv_setsv_flags(sv, old, SV_NOSTEAL);
9737 =for apidoc sv_reset
9739 Underlying implementation for the C<reset> Perl function.
9740 Note that the perl-level function is vaguely deprecated.
9746 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9748 PERL_ARGS_ASSERT_SV_RESET;
9750 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9754 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9756 char todo[PERL_UCHAR_MAX+1];
9759 if (!stash || SvTYPE(stash) != SVt_PVHV)
9762 if (!s) { /* reset ?? searches */
9763 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9765 const U32 count = mg->mg_len / sizeof(PMOP**);
9766 PMOP **pmp = (PMOP**) mg->mg_ptr;
9767 PMOP *const *const end = pmp + count;
9771 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9773 (*pmp)->op_pmflags &= ~PMf_USED;
9781 /* reset variables */
9783 if (!HvARRAY(stash))
9786 Zero(todo, 256, char);
9790 I32 i = (unsigned char)*s;
9794 max = (unsigned char)*s++;
9795 for ( ; i <= max; i++) {
9798 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9800 for (entry = HvARRAY(stash)[i];
9802 entry = HeNEXT(entry))
9807 if (!todo[(U8)*HeKEY(entry)])
9809 gv = MUTABLE_GV(HeVAL(entry));
9813 if (sv && !SvREADONLY(sv)) {
9814 SV_CHECK_THINKFIRST_COW_DROP(sv);
9815 if (!isGV(sv)) SvOK_off(sv);
9820 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9831 Using various gambits, try to get an IO from an SV: the IO slot if its a
9832 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9833 named after the PV if we're a string.
9835 'Get' magic is ignored on the C<sv> passed in, but will be called on
9836 C<SvRV(sv)> if C<sv> is an RV.
9842 Perl_sv_2io(pTHX_ SV *const sv)
9847 PERL_ARGS_ASSERT_SV_2IO;
9849 switch (SvTYPE(sv)) {
9851 io = MUTABLE_IO(sv);
9855 if (isGV_with_GP(sv)) {
9856 gv = MUTABLE_GV(sv);
9859 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9860 HEKfARG(GvNAME_HEK(gv)));
9866 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9868 SvGETMAGIC(SvRV(sv));
9869 return sv_2io(SvRV(sv));
9871 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9878 if (SvGMAGICAL(sv)) {
9879 newsv = sv_newmortal();
9880 sv_setsv_nomg(newsv, sv);
9882 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9892 Using various gambits, try to get a CV from an SV; in addition, try if
9893 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9894 The flags in C<lref> are passed to C<gv_fetchsv>.
9900 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9905 PERL_ARGS_ASSERT_SV_2CV;
9912 switch (SvTYPE(sv)) {
9916 return MUTABLE_CV(sv);
9926 sv = amagic_deref_call(sv, to_cv_amg);
9929 if (SvTYPE(sv) == SVt_PVCV) {
9930 cv = MUTABLE_CV(sv);
9935 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9936 gv = MUTABLE_GV(sv);
9938 Perl_croak(aTHX_ "Not a subroutine reference");
9940 else if (isGV_with_GP(sv)) {
9941 gv = MUTABLE_GV(sv);
9944 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9951 /* Some flags to gv_fetchsv mean don't really create the GV */
9952 if (!isGV_with_GP(gv)) {
9957 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9958 /* XXX this is probably not what they think they're getting.
9959 * It has the same effect as "sub name;", i.e. just a forward
9970 Returns true if the SV has a true value by Perl's rules.
9971 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9972 instead use an in-line version.
9978 Perl_sv_true(pTHX_ SV *const sv)
9983 const XPV* const tXpv = (XPV*)SvANY(sv);
9985 (tXpv->xpv_cur > 1 ||
9986 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9993 return SvIVX(sv) != 0;
9996 return SvNVX(sv) != 0.0;
9998 return sv_2bool(sv);
10004 =for apidoc sv_pvn_force
10006 Get a sensible string out of the SV somehow.
10007 A private implementation of the C<SvPV_force> macro for compilers which
10008 can't cope with complex macro expressions. Always use the macro instead.
10010 =for apidoc sv_pvn_force_flags
10012 Get a sensible string out of the SV somehow.
10013 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10014 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10015 implemented in terms of this function.
10016 You normally want to use the various wrapper macros instead: see
10017 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10023 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10025 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10027 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10028 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10029 sv_force_normal_flags(sv, 0);
10039 if (SvTYPE(sv) > SVt_PVLV
10040 || isGV_with_GP(sv))
10041 /* diag_listed_as: Can't coerce %s to %s in %s */
10042 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10044 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10051 if (SvTYPE(sv) < SVt_PV ||
10052 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10055 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10056 SvGROW(sv, len + 1);
10057 Move(s,SvPVX(sv),len,char);
10058 SvCUR_set(sv, len);
10059 SvPVX(sv)[len] = '\0';
10062 SvPOK_on(sv); /* validate pointer */
10064 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10065 PTR2UV(sv),SvPVX_const(sv)));
10068 (void)SvPOK_only_UTF8(sv);
10069 return SvPVX_mutable(sv);
10073 =for apidoc sv_pvbyten_force
10075 The backend for the C<SvPVbytex_force> macro. Always use the macro
10082 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10084 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10086 sv_pvn_force(sv,lp);
10087 sv_utf8_downgrade(sv,0);
10093 =for apidoc sv_pvutf8n_force
10095 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10102 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10104 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10106 sv_pvn_force(sv,0);
10107 sv_utf8_upgrade_nomg(sv);
10113 =for apidoc sv_reftype
10115 Returns a string describing what the SV is a reference to.
10117 If ob is true and the SV is blessed, the string is the class name,
10118 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10124 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10126 PERL_ARGS_ASSERT_SV_REFTYPE;
10127 if (ob && SvOBJECT(sv)) {
10128 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10131 /* WARNING - There is code, for instance in mg.c, that assumes that
10132 * the only reason that sv_reftype(sv,0) would return a string starting
10133 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10134 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10135 * this routine inside other subs, and it saves time.
10136 * Do not change this assumption without searching for "dodgy type check" in
10139 switch (SvTYPE(sv)) {
10154 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10155 /* tied lvalues should appear to be
10156 * scalars for backwards compatibility */
10157 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10158 ? "SCALAR" : "LVALUE");
10159 case SVt_PVAV: return "ARRAY";
10160 case SVt_PVHV: return "HASH";
10161 case SVt_PVCV: return "CODE";
10162 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10163 ? "GLOB" : "SCALAR");
10164 case SVt_PVFM: return "FORMAT";
10165 case SVt_PVIO: return "IO";
10166 case SVt_INVLIST: return "INVLIST";
10167 case SVt_REGEXP: return "REGEXP";
10168 default: return "UNKNOWN";
10176 Returns a SV describing what the SV passed in is a reference to.
10178 dst can be a SV to be set to the description or NULL, in which case a
10179 mortal SV is returned.
10181 If ob is true and the SV is blessed, the description is the class
10182 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10188 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10190 PERL_ARGS_ASSERT_SV_REF;
10193 dst = sv_newmortal();
10195 if (ob && SvOBJECT(sv)) {
10196 HvNAME_get(SvSTASH(sv))
10197 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10198 : sv_setpvs(dst, "__ANON__");
10201 const char * reftype = sv_reftype(sv, 0);
10202 sv_setpv(dst, reftype);
10208 =for apidoc sv_isobject
10210 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10211 object. If the SV is not an RV, or if the object is not blessed, then this
10218 Perl_sv_isobject(pTHX_ SV *sv)
10234 Returns a boolean indicating whether the SV is blessed into the specified
10235 class. This does not check for subtypes; use C<sv_derived_from> to verify
10236 an inheritance relationship.
10242 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10244 const char *hvname;
10246 PERL_ARGS_ASSERT_SV_ISA;
10256 hvname = HvNAME_get(SvSTASH(sv));
10260 return strEQ(hvname, name);
10264 =for apidoc newSVrv
10266 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10267 RV then it will be upgraded to one. If C<classname> is non-null then the new
10268 SV will be blessed in the specified package. The new SV is returned and its
10269 reference count is 1. The reference count 1 is owned by C<rv>.
10275 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10279 PERL_ARGS_ASSERT_NEWSVRV;
10283 SV_CHECK_THINKFIRST_COW_DROP(rv);
10285 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10286 const U32 refcnt = SvREFCNT(rv);
10290 SvREFCNT(rv) = refcnt;
10292 sv_upgrade(rv, SVt_IV);
10293 } else if (SvROK(rv)) {
10294 SvREFCNT_dec(SvRV(rv));
10296 prepare_SV_for_RV(rv);
10304 HV* const stash = gv_stashpv(classname, GV_ADD);
10305 (void)sv_bless(rv, stash);
10311 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10313 SV * const lv = newSV_type(SVt_PVLV);
10314 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10316 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10317 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10318 LvSTARGOFF(lv) = ix;
10319 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10324 =for apidoc sv_setref_pv
10326 Copies a pointer 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. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10329 into the SV. The C<classname> argument indicates the package for the
10330 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10331 will have a reference count of 1, and the RV will be returned.
10333 Do not use with other Perl types such as HV, AV, SV, CV, because those
10334 objects will become corrupted by the pointer copy process.
10336 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10342 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10344 PERL_ARGS_ASSERT_SV_SETREF_PV;
10351 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10356 =for apidoc sv_setref_iv
10358 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10359 argument will be upgraded to an RV. That RV will be modified to point to
10360 the new SV. The C<classname> argument indicates the package for the
10361 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10362 will have a reference count of 1, and the RV will be returned.
10368 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10370 PERL_ARGS_ASSERT_SV_SETREF_IV;
10372 sv_setiv(newSVrv(rv,classname), iv);
10377 =for apidoc sv_setref_uv
10379 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10380 argument will be upgraded to an RV. That RV will be modified to point to
10381 the new SV. The C<classname> argument indicates the package for the
10382 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10383 will have a reference count of 1, and the RV will be returned.
10389 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10391 PERL_ARGS_ASSERT_SV_SETREF_UV;
10393 sv_setuv(newSVrv(rv,classname), uv);
10398 =for apidoc sv_setref_nv
10400 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10401 argument will be upgraded to an RV. That RV will be modified to point to
10402 the new SV. The C<classname> argument indicates the package for the
10403 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10404 will have a reference count of 1, and the RV will be returned.
10410 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10412 PERL_ARGS_ASSERT_SV_SETREF_NV;
10414 sv_setnv(newSVrv(rv,classname), nv);
10419 =for apidoc sv_setref_pvn
10421 Copies a string into a new SV, optionally blessing the SV. The length of the
10422 string must be specified with C<n>. The C<rv> argument will be upgraded to
10423 an RV. That RV will be modified to point to the new SV. The C<classname>
10424 argument indicates the package for the blessing. Set C<classname> to
10425 C<NULL> to avoid the blessing. The new SV will have a reference count
10426 of 1, and the RV will be returned.
10428 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10434 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10435 const char *const pv, const STRLEN n)
10437 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10439 sv_setpvn(newSVrv(rv,classname), pv, n);
10444 =for apidoc sv_bless
10446 Blesses an SV into a specified package. The SV must be an RV. The package
10447 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10448 of the SV is unaffected.
10454 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10457 HV *oldstash = NULL;
10459 PERL_ARGS_ASSERT_SV_BLESS;
10463 Perl_croak(aTHX_ "Can't bless non-reference value");
10465 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10466 if (SvREADONLY(tmpRef))
10467 Perl_croak_no_modify();
10468 if (SvOBJECT(tmpRef)) {
10469 oldstash = SvSTASH(tmpRef);
10472 SvOBJECT_on(tmpRef);
10473 SvUPGRADE(tmpRef, SVt_PVMG);
10474 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10475 SvREFCNT_dec(oldstash);
10477 if(SvSMAGICAL(tmpRef))
10478 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10486 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10487 * as it is after unglobbing it.
10490 PERL_STATIC_INLINE void
10491 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10495 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10497 PERL_ARGS_ASSERT_SV_UNGLOB;
10499 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10501 if (!(flags & SV_COW_DROP_PV))
10502 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10504 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10506 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10507 && HvNAME_get(stash))
10508 mro_method_changed_in(stash);
10509 gp_free(MUTABLE_GV(sv));
10512 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10513 GvSTASH(sv) = NULL;
10516 if (GvNAME_HEK(sv)) {
10517 unshare_hek(GvNAME_HEK(sv));
10519 isGV_with_GP_off(sv);
10521 if(SvTYPE(sv) == SVt_PVGV) {
10522 /* need to keep SvANY(sv) in the right arena */
10523 xpvmg = new_XPVMG();
10524 StructCopy(SvANY(sv), xpvmg, XPVMG);
10525 del_XPVGV(SvANY(sv));
10528 SvFLAGS(sv) &= ~SVTYPEMASK;
10529 SvFLAGS(sv) |= SVt_PVMG;
10532 /* Intentionally not calling any local SET magic, as this isn't so much a
10533 set operation as merely an internal storage change. */
10534 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10535 else sv_setsv_flags(sv, temp, 0);
10537 if ((const GV *)sv == PL_last_in_gv)
10538 PL_last_in_gv = NULL;
10539 else if ((const GV *)sv == PL_statgv)
10544 =for apidoc sv_unref_flags
10546 Unsets the RV status of the SV, and decrements the reference count of
10547 whatever was being referenced by the RV. This can almost be thought of
10548 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10549 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10550 (otherwise the decrementing is conditional on the reference count being
10551 different from one or the reference being a readonly SV).
10552 See C<L</SvROK_off>>.
10558 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10560 SV* const target = SvRV(ref);
10562 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10564 if (SvWEAKREF(ref)) {
10565 sv_del_backref(target, ref);
10566 SvWEAKREF_off(ref);
10567 SvRV_set(ref, NULL);
10570 SvRV_set(ref, NULL);
10572 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10573 assigned to as BEGIN {$a = \"Foo"} will fail. */
10574 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10575 SvREFCNT_dec_NN(target);
10576 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10577 sv_2mortal(target); /* Schedule for freeing later */
10581 =for apidoc sv_untaint
10583 Untaint an SV. Use C<SvTAINTED_off> instead.
10589 Perl_sv_untaint(pTHX_ SV *const sv)
10591 PERL_ARGS_ASSERT_SV_UNTAINT;
10592 PERL_UNUSED_CONTEXT;
10594 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10595 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10602 =for apidoc sv_tainted
10604 Test an SV for taintedness. Use C<SvTAINTED> instead.
10610 Perl_sv_tainted(pTHX_ SV *const sv)
10612 PERL_ARGS_ASSERT_SV_TAINTED;
10613 PERL_UNUSED_CONTEXT;
10615 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10616 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10617 if (mg && (mg->mg_len & 1) )
10623 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10624 private to this file */
10627 =for apidoc sv_setpviv
10629 Copies an integer into the given SV, also updating its string value.
10630 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10636 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10638 char buf[TYPE_CHARS(UV)];
10640 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10642 PERL_ARGS_ASSERT_SV_SETPVIV;
10644 sv_setpvn(sv, ptr, ebuf - ptr);
10648 =for apidoc sv_setpviv_mg
10650 Like C<sv_setpviv>, but also handles 'set' magic.
10656 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10658 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10660 sv_setpviv(sv, iv);
10664 #endif /* NO_MATHOMS */
10666 #if defined(PERL_IMPLICIT_CONTEXT)
10668 /* pTHX_ magic can't cope with varargs, so this is a no-context
10669 * version of the main function, (which may itself be aliased to us).
10670 * Don't access this version directly.
10674 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10679 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10681 va_start(args, pat);
10682 sv_vsetpvf(sv, pat, &args);
10686 /* pTHX_ magic can't cope with varargs, so this is a no-context
10687 * version of the main function, (which may itself be aliased to us).
10688 * Don't access this version directly.
10692 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10697 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10699 va_start(args, pat);
10700 sv_vsetpvf_mg(sv, pat, &args);
10706 =for apidoc sv_setpvf
10708 Works like C<sv_catpvf> but copies the text into the SV instead of
10709 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10715 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10719 PERL_ARGS_ASSERT_SV_SETPVF;
10721 va_start(args, pat);
10722 sv_vsetpvf(sv, pat, &args);
10727 =for apidoc sv_vsetpvf
10729 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10730 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10732 Usually used via its frontend C<sv_setpvf>.
10738 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10740 PERL_ARGS_ASSERT_SV_VSETPVF;
10742 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10746 =for apidoc sv_setpvf_mg
10748 Like C<sv_setpvf>, but also handles 'set' magic.
10754 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10758 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10760 va_start(args, pat);
10761 sv_vsetpvf_mg(sv, pat, &args);
10766 =for apidoc sv_vsetpvf_mg
10768 Like C<sv_vsetpvf>, but also handles 'set' magic.
10770 Usually used via its frontend C<sv_setpvf_mg>.
10776 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10778 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10780 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10784 #if defined(PERL_IMPLICIT_CONTEXT)
10786 /* pTHX_ magic can't cope with varargs, so this is a no-context
10787 * version of the main function, (which may itself be aliased to us).
10788 * Don't access this version directly.
10792 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10797 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10799 va_start(args, pat);
10800 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10804 /* pTHX_ magic can't cope with varargs, so this is a no-context
10805 * version of the main function, (which may itself be aliased to us).
10806 * Don't access this version directly.
10810 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10815 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10817 va_start(args, pat);
10818 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10825 =for apidoc sv_catpvf
10827 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10828 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10829 variable argument list, argument reordering is not supported.
10830 If the appended data contains "wide" characters
10831 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10832 and characters >255 formatted with C<%c>), the original SV might get
10833 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10834 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10835 valid UTF-8; if the original SV was bytes, the pattern should be too.
10840 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10844 PERL_ARGS_ASSERT_SV_CATPVF;
10846 va_start(args, pat);
10847 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10852 =for apidoc sv_vcatpvf
10854 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10855 variable argument list, and appends the formatted output
10856 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10858 Usually used via its frontend C<sv_catpvf>.
10864 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10866 PERL_ARGS_ASSERT_SV_VCATPVF;
10868 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10872 =for apidoc sv_catpvf_mg
10874 Like C<sv_catpvf>, but also handles 'set' magic.
10880 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10884 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10886 va_start(args, pat);
10887 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10893 =for apidoc sv_vcatpvf_mg
10895 Like C<sv_vcatpvf>, but also handles 'set' magic.
10897 Usually used via its frontend C<sv_catpvf_mg>.
10903 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10905 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10907 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10912 =for apidoc sv_vsetpvfn
10914 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10917 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10923 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10924 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10926 PERL_ARGS_ASSERT_SV_VSETPVFN;
10929 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10934 * Warn of missing argument to sprintf. The value used in place of such
10935 * arguments should be &PL_sv_no; an undefined value would yield
10936 * inappropriate "use of uninit" warnings [perl #71000].
10939 S_warn_vcatpvfn_missing_argument(pTHX) {
10940 if (ckWARN(WARN_MISSING)) {
10941 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10942 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10948 S_expect_number(pTHX_ char **const pattern)
10952 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10954 switch (**pattern) {
10955 case '1': case '2': case '3':
10956 case '4': case '5': case '6':
10957 case '7': case '8': case '9':
10958 var = *(*pattern)++ - '0';
10959 while (isDIGIT(**pattern)) {
10960 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10962 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10970 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10972 const int neg = nv < 0;
10975 PERL_ARGS_ASSERT_F0CONVERT;
10977 if (UNLIKELY(Perl_isinfnan(nv))) {
10978 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10988 if (uv & 1 && uv == nv)
10989 uv--; /* Round to even */
10991 const unsigned dig = uv % 10;
10993 } while (uv /= 10);
11004 =for apidoc sv_vcatpvfn
11006 =for apidoc sv_vcatpvfn_flags
11008 Processes its arguments like C<vsprintf> and appends the formatted output
11009 to an SV. Uses an array of SVs if the C-style variable argument list is
11010 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11011 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11012 C<va_list> argument list with a format string that uses argument reordering
11013 will yield an exception.
11015 When running with taint checks enabled, indicates via
11016 C<maybe_tainted> if results are untrustworthy (often due to the use of
11019 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11021 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11026 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
11027 vecstr = (U8*)SvPV_const(vecsv,veclen);\
11028 vec_utf8 = DO_UTF8(vecsv);
11030 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11033 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11034 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
11036 PERL_ARGS_ASSERT_SV_VCATPVFN;
11038 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11041 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11042 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11043 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11044 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11045 * after the first 1023 zero bits.
11047 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11048 * of dynamically growing buffer might be better, start at just 16 bytes
11049 * (for example) and grow only when necessary. Or maybe just by looking
11050 * at the exponents of the two doubles? */
11051 # define DOUBLEDOUBLE_MAXBITS 2098
11054 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11055 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11056 * per xdigit. For the double-double case, this can be rather many.
11057 * The non-double-double-long-double overshoots since all bits of NV
11058 * are not mantissa bits, there are also exponent bits. */
11059 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11060 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11062 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11065 /* If we do not have a known long double format, (including not using
11066 * long doubles, or long doubles being equal to doubles) then we will
11067 * fall back to the ldexp/frexp route, with which we can retrieve at
11068 * most as many bits as our widest unsigned integer type is. We try
11069 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11071 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11072 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11074 #if defined(HAS_QUAD) && defined(Uquad_t)
11075 # define MANTISSATYPE Uquad_t
11076 # define MANTISSASIZE 8
11078 # define MANTISSATYPE UV
11079 # define MANTISSASIZE UVSIZE
11082 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11083 # define HEXTRACT_LITTLE_ENDIAN
11084 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11085 # define HEXTRACT_BIG_ENDIAN
11087 # define HEXTRACT_MIX_ENDIAN
11090 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
11091 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11092 * are being extracted from (either directly from the long double in-memory
11093 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11094 * is used to update the exponent. The subnormal is set to true
11095 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11096 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11098 * The tricky part is that S_hextract() needs to be called twice:
11099 * the first time with vend as NULL, and the second time with vend as
11100 * the pointer returned by the first call. What happens is that on
11101 * the first round the output size is computed, and the intended
11102 * extraction sanity checked. On the second round the actual output
11103 * (the extraction of the hexadecimal values) takes place.
11104 * Sanity failures cause fatal failures during both rounds. */
11106 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11107 U8* vhex, U8* vend)
11111 int ixmin = 0, ixmax = 0;
11113 /* XXX Inf/NaN are not handled here, since it is
11114 * assumed they are to be output as "Inf" and "NaN". */
11116 /* These macros are just to reduce typos, they have multiple
11117 * repetitions below, but usually only one (or sometimes two)
11118 * of them is really being used. */
11119 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11120 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11121 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11122 #define HEXTRACT_OUTPUT(ix) \
11124 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11126 #define HEXTRACT_COUNT(ix, c) \
11128 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11130 #define HEXTRACT_BYTE(ix) \
11132 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11134 #define HEXTRACT_LO_NYBBLE(ix) \
11136 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11138 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11139 * to make it look less odd when the top bits of a NV
11140 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11141 * order bits can be in the "low nybble" of a byte. */
11142 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11143 #define HEXTRACT_BYTES_LE(a, b) \
11144 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11145 #define HEXTRACT_BYTES_BE(a, b) \
11146 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11147 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11148 #define HEXTRACT_IMPLICIT_BIT(nv) \
11150 if (!*subnormal) { \
11151 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11155 /* Most formats do. Those which don't should undef this.
11157 * But also note that IEEE 754 subnormals do not have it, or,
11158 * expressed alternatively, their implicit bit is zero. */
11159 #define HEXTRACT_HAS_IMPLICIT_BIT
11161 /* Many formats do. Those which don't should undef this. */
11162 #define HEXTRACT_HAS_TOP_NYBBLE
11164 /* HEXTRACTSIZE is the maximum number of xdigits. */
11165 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11166 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11168 # define HEXTRACTSIZE 2 * NVSIZE
11171 const U8* vmaxend = vhex + HEXTRACTSIZE;
11172 PERL_UNUSED_VAR(ix); /* might happen */
11173 (void)Perl_frexp(PERL_ABS(nv), exponent);
11174 *subnormal = FALSE;
11175 if (vend && (vend <= vhex || vend > vmaxend)) {
11176 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11177 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11180 /* First check if using long doubles. */
11181 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11182 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11183 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11184 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11185 /* The bytes 13..0 are the mantissa/fraction,
11186 * the 15,14 are the sign+exponent. */
11187 const U8* nvp = (const U8*)(&nv);
11188 HEXTRACT_GET_SUBNORMAL(nv);
11189 HEXTRACT_IMPLICIT_BIT(nv);
11190 # undef HEXTRACT_HAS_TOP_NYBBLE
11191 HEXTRACT_BYTES_LE(13, 0);
11192 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11193 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11194 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11195 /* The bytes 2..15 are the mantissa/fraction,
11196 * the 0,1 are the sign+exponent. */
11197 const U8* nvp = (const U8*)(&nv);
11198 HEXTRACT_GET_SUBNORMAL(nv);
11199 HEXTRACT_IMPLICIT_BIT(nv);
11200 # undef HEXTRACT_HAS_TOP_NYBBLE
11201 HEXTRACT_BYTES_BE(2, 15);
11202 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11203 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11204 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11205 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11206 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11207 /* The bytes 0..1 are the sign+exponent,
11208 * the bytes 2..9 are the mantissa/fraction. */
11209 const U8* nvp = (const U8*)(&nv);
11210 # undef HEXTRACT_HAS_IMPLICIT_BIT
11211 # undef HEXTRACT_HAS_TOP_NYBBLE
11212 HEXTRACT_GET_SUBNORMAL(nv);
11213 HEXTRACT_BYTES_LE(7, 0);
11214 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11215 /* Does this format ever happen? (Wikipedia says the Motorola
11216 * 6888x math coprocessors used format _like_ this but padded
11217 * to 96 bits with 16 unused bits between the exponent and the
11219 const U8* nvp = (const U8*)(&nv);
11220 # undef HEXTRACT_HAS_IMPLICIT_BIT
11221 # undef HEXTRACT_HAS_TOP_NYBBLE
11222 HEXTRACT_GET_SUBNORMAL(nv);
11223 HEXTRACT_BYTES_BE(0, 7);
11225 # define HEXTRACT_FALLBACK
11226 /* Double-double format: two doubles next to each other.
11227 * The first double is the high-order one, exactly like
11228 * it would be for a "lone" double. The second double
11229 * is shifted down using the exponent so that that there
11230 * are no common bits. The tricky part is that the value
11231 * of the double-double is the SUM of the two doubles and
11232 * the second one can be also NEGATIVE.
11234 * Because of this tricky construction the bytewise extraction we
11235 * use for the other long double formats doesn't work, we must
11236 * extract the values bit by bit.
11238 * The little-endian double-double is used .. somewhere?
11240 * The big endian double-double is used in e.g. PPC/Power (AIX)
11243 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11244 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11245 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11248 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11249 /* Using normal doubles, not long doubles.
11251 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11252 * bytes, since we might need to handle printf precision, and
11253 * also need to insert the radix. */
11255 # ifdef HEXTRACT_LITTLE_ENDIAN
11256 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11257 const U8* nvp = (const U8*)(&nv);
11258 HEXTRACT_GET_SUBNORMAL(nv);
11259 HEXTRACT_IMPLICIT_BIT(nv);
11260 HEXTRACT_TOP_NYBBLE(6);
11261 HEXTRACT_BYTES_LE(5, 0);
11262 # elif defined(HEXTRACT_BIG_ENDIAN)
11263 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11264 const U8* nvp = (const U8*)(&nv);
11265 HEXTRACT_GET_SUBNORMAL(nv);
11266 HEXTRACT_IMPLICIT_BIT(nv);
11267 HEXTRACT_TOP_NYBBLE(1);
11268 HEXTRACT_BYTES_BE(2, 7);
11269 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11270 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11271 const U8* nvp = (const U8*)(&nv);
11272 HEXTRACT_GET_SUBNORMAL(nv);
11273 HEXTRACT_IMPLICIT_BIT(nv);
11274 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11275 HEXTRACT_BYTE(1); /* 5 */
11276 HEXTRACT_BYTE(0); /* 4 */
11277 HEXTRACT_BYTE(7); /* 3 */
11278 HEXTRACT_BYTE(6); /* 2 */
11279 HEXTRACT_BYTE(5); /* 1 */
11280 HEXTRACT_BYTE(4); /* 0 */
11281 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11282 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11283 const U8* nvp = (const U8*)(&nv);
11284 HEXTRACT_GET_SUBNORMAL(nv);
11285 HEXTRACT_IMPLICIT_BIT(nv);
11286 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11287 HEXTRACT_BYTE(6); /* 5 */
11288 HEXTRACT_BYTE(7); /* 4 */
11289 HEXTRACT_BYTE(0); /* 3 */
11290 HEXTRACT_BYTE(1); /* 2 */
11291 HEXTRACT_BYTE(2); /* 1 */
11292 HEXTRACT_BYTE(3); /* 0 */
11294 # define HEXTRACT_FALLBACK
11297 # define HEXTRACT_FALLBACK
11299 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11300 # ifdef HEXTRACT_FALLBACK
11301 HEXTRACT_GET_SUBNORMAL(nv);
11302 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11303 /* The fallback is used for the double-double format, and
11304 * for unknown long double formats, and for unknown double
11305 * formats, or in general unknown NV formats. */
11306 if (nv == (NV)0.0) {
11314 NV d = nv < 0 ? -nv : nv;
11316 U8 ha = 0x0; /* hexvalue accumulator */
11317 U8 hd = 0x8; /* hexvalue digit */
11319 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11320 * this is essentially manual frexp(). Multiplying by 0.5 and
11321 * doubling should be lossless in binary floating point. */
11331 while (d >= e + e) {
11335 /* Now e <= d < 2*e */
11337 /* First extract the leading hexdigit (the implicit bit). */
11353 /* Then extract the remaining hexdigits. */
11354 while (d > (NV)0.0) {
11360 /* Output or count in groups of four bits,
11361 * that is, when the hexdigit is down to one. */
11366 /* Reset the hexvalue. */
11375 /* Flush possible pending hexvalue. */
11385 /* Croak for various reasons: if the output pointer escaped the
11386 * output buffer, if the extraction index escaped the extraction
11387 * buffer, or if the ending output pointer didn't match the
11388 * previously computed value. */
11389 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11390 /* For double-double the ixmin and ixmax stay at zero,
11391 * which is convenient since the HEXTRACTSIZE is tricky
11392 * for double-double. */
11393 ixmin < 0 || ixmax >= NVSIZE ||
11394 (vend && v != vend)) {
11395 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11396 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11401 /* Helper for sv_vcatpvfn_flags(). */
11402 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11407 (var) = &PL_sv_no; /* [perl #71000] */ \
11408 arg_missing = TRUE; \
11413 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11414 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11419 const char *patend;
11422 static const char nullstr[] = "(null)";
11424 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11425 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11427 /* Times 4: a decimal digit takes more than 3 binary digits.
11428 * NV_DIG: mantissa takes than many decimal digits.
11429 * Plus 32: Playing safe. */
11430 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11431 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11432 bool hexfp = FALSE; /* hexadecimal floating point? */
11434 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11436 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11437 PERL_UNUSED_ARG(maybe_tainted);
11439 if (flags & SV_GMAGIC)
11442 /* no matter what, this is a string now */
11443 (void)SvPV_force_nomg(sv, origlen);
11445 /* special-case "", "%s", and "%-p" (SVf - see below) */
11447 if (svmax && ckWARN(WARN_REDUNDANT))
11448 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11449 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11452 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11453 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11454 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11455 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11458 const char * const s = va_arg(*args, char*);
11459 sv_catpv_nomg(sv, s ? s : nullstr);
11461 else if (svix < svmax) {
11462 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11463 SvGETMAGIC(*svargs);
11464 sv_catsv_nomg(sv, *svargs);
11467 S_warn_vcatpvfn_missing_argument(aTHX);
11470 if (args && patlen == 3 && pat[0] == '%' &&
11471 pat[1] == '-' && pat[2] == 'p') {
11472 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11473 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11474 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11475 argsv = MUTABLE_SV(va_arg(*args, void*));
11476 sv_catsv_nomg(sv, argsv);
11480 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11481 /* special-case "%.<number>[gf]" */
11482 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11483 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11484 unsigned digits = 0;
11488 while (*pp >= '0' && *pp <= '9')
11489 digits = 10 * digits + (*pp++ - '0');
11491 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11492 format the first argument and WARN_REDUNDANT if svmax > 1?
11493 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11494 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11495 const NV nv = SvNV(*svargs);
11496 if (LIKELY(!Perl_isinfnan(nv))) {
11498 /* Add check for digits != 0 because it seems that some
11499 gconverts are buggy in this case, and we don't yet have
11500 a Configure test for this. */
11501 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11502 /* 0, point, slack */
11503 STORE_LC_NUMERIC_SET_TO_NEEDED();
11504 SNPRINTF_G(nv, ebuf, size, digits);
11505 sv_catpv_nomg(sv, ebuf);
11506 if (*ebuf) /* May return an empty string for digits==0 */
11509 } else if (!digits) {
11512 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11513 sv_catpvn_nomg(sv, p, l);
11520 #endif /* !USE_LONG_DOUBLE */
11522 if (!args && svix < svmax && DO_UTF8(*svargs))
11525 patend = (char*)pat + patlen;
11526 for (p = (char*)pat; p < patend; p = q) {
11529 bool vectorize = FALSE;
11530 bool vectorarg = FALSE;
11531 bool vec_utf8 = FALSE;
11537 bool has_precis = FALSE;
11539 const I32 osvix = svix;
11540 bool is_utf8 = FALSE; /* is this item utf8? */
11541 bool used_explicit_ix = FALSE;
11542 bool arg_missing = FALSE;
11543 #ifdef HAS_LDBL_SPRINTF_BUG
11544 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11545 with sfio - Allen <allens@cpan.org> */
11546 bool fix_ldbl_sprintf_bug = FALSE;
11550 U8 utf8buf[UTF8_MAXBYTES+1];
11551 STRLEN esignlen = 0;
11553 const char *eptr = NULL;
11554 const char *fmtstart;
11557 const U8 *vecstr = NULL;
11564 /* We need a long double target in case HAS_LONG_DOUBLE,
11565 * even without USE_LONG_DOUBLE, so that we can printf with
11566 * long double formats, even without NV being long double.
11567 * But we call the target 'fv' instead of 'nv', since most of
11568 * the time it is not (most compilers these days recognize
11569 * "long double", even if only as a synonym for "double").
11571 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11572 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11574 # ifdef Perl_isfinitel
11575 # define FV_ISFINITE(x) Perl_isfinitel(x)
11577 # define FV_GF PERL_PRIgldbl
11578 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11579 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11580 # define NV_TO_FV(nv,fv) STMT_START { \
11582 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11585 # define NV_TO_FV(nv,fv) (fv)=(nv)
11589 # define FV_GF NVgf
11590 # define NV_TO_FV(nv,fv) (fv)=(nv)
11592 #ifndef FV_ISFINITE
11593 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11599 const char *dotstr = ".";
11600 STRLEN dotstrlen = 1;
11601 I32 efix = 0; /* explicit format parameter index */
11602 I32 ewix = 0; /* explicit width index */
11603 I32 epix = 0; /* explicit precision index */
11604 I32 evix = 0; /* explicit vector index */
11605 bool asterisk = FALSE;
11606 bool infnan = FALSE;
11608 /* echo everything up to the next format specification */
11609 for (q = p; q < patend && *q != '%'; ++q) ;
11611 if (has_utf8 && !pat_utf8)
11612 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11614 sv_catpvn_nomg(sv, p, q - p);
11623 We allow format specification elements in this order:
11624 \d+\$ explicit format parameter index
11626 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11627 0 flag (as above): repeated to allow "v02"
11628 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11629 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11631 [%bcdefginopsuxDFOUX] format (mandatory)
11636 As of perl5.9.3, printf format checking is on by default.
11637 Internally, perl uses %p formats to provide an escape to
11638 some extended formatting. This block deals with those
11639 extensions: if it does not match, (char*)q is reset and
11640 the normal format processing code is used.
11642 Currently defined extensions are:
11643 %p include pointer address (standard)
11644 %-p (SVf) include an SV (previously %_)
11645 %-<num>p include an SV with precision <num>
11647 %3p include a HEK with precision of 256
11648 %4p char* preceded by utf8 flag and length
11649 %<num>p (where num is 1 or > 4) reserved for future
11652 Robin Barker 2005-07-14 (but modified since)
11654 %1p (VDf) removed. RMB 2007-10-19
11661 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11662 /* The argument has already gone through cBOOL, so the cast
11664 is_utf8 = (bool)va_arg(*args, int);
11665 elen = va_arg(*args, UV);
11666 /* if utf8 length is larger than 0x7ffff..., then it might
11667 * have been a signed value that wrapped */
11668 if (elen > ((~(STRLEN)0) >> 1)) {
11669 assert(0); /* in DEBUGGING build we want to crash */
11670 elen= 0; /* otherwise we want to treat this as an empty string */
11672 eptr = va_arg(*args, char *);
11673 q += sizeof(UTF8f)-1;
11676 n = expect_number(&q);
11678 if (sv) { /* SVf */
11683 argsv = MUTABLE_SV(va_arg(*args, void*));
11684 eptr = SvPV_const(argsv, elen);
11685 if (DO_UTF8(argsv))
11689 else if (n==2 || n==3) { /* HEKf */
11690 HEK * const hek = va_arg(*args, HEK *);
11691 eptr = HEK_KEY(hek);
11692 elen = HEK_LEN(hek);
11693 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11694 if (n==3) precis = 256, has_precis = TRUE;
11698 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11699 "internal %%<num>p might conflict with future printf extensions");
11705 if ( (width = expect_number(&q)) ) {
11708 Perl_croak_nocontext(
11709 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11712 used_explicit_ix = TRUE;
11724 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11753 if ( (ewix = expect_number(&q)) ) {
11756 Perl_croak_nocontext(
11757 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11758 used_explicit_ix = TRUE;
11768 if ((vectorarg = asterisk)) {
11781 width = expect_number(&q);
11784 if (vectorize && vectorarg) {
11785 /* vectorizing, but not with the default "." */
11787 vecsv = va_arg(*args, SV*);
11789 FETCH_VCATPVFN_ARGUMENT(
11790 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11792 FETCH_VCATPVFN_ARGUMENT(
11793 vecsv, svix < svmax, svargs[svix++]);
11795 dotstr = SvPV_const(vecsv, dotstrlen);
11796 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11797 bad with tied or overloaded values that return UTF8. */
11798 if (DO_UTF8(vecsv))
11800 else if (has_utf8) {
11801 vecsv = sv_mortalcopy(vecsv);
11802 sv_utf8_upgrade(vecsv);
11803 dotstr = SvPV_const(vecsv, dotstrlen);
11810 i = va_arg(*args, int);
11812 i = (ewix ? ewix <= svmax : svix < svmax) ?
11813 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11815 width = (i < 0) ? -i : i;
11825 if ( (epix = expect_number(&q)) ) {
11828 Perl_croak_nocontext(
11829 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11830 used_explicit_ix = TRUE;
11835 i = va_arg(*args, int);
11839 FETCH_VCATPVFN_ARGUMENT(
11840 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11842 FETCH_VCATPVFN_ARGUMENT(
11843 precsv, svix < svmax, svargs[svix++]);
11844 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11847 has_precis = !(i < 0);
11851 while (isDIGIT(*q))
11852 precis = precis * 10 + (*q++ - '0');
11861 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11862 vecsv = svargs[efix ? efix-1 : svix++];
11863 vecstr = (U8*)SvPV_const(vecsv,veclen);
11864 vec_utf8 = DO_UTF8(vecsv);
11866 /* if this is a version object, we need to convert
11867 * back into v-string notation and then let the
11868 * vectorize happen normally
11870 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11871 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
11872 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11873 "vector argument not supported with alpha versions");
11876 vecsv = sv_newmortal();
11877 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11879 vecstr = (U8*)SvPV_const(vecsv, veclen);
11880 vec_utf8 = DO_UTF8(vecsv);
11894 case 'I': /* Ix, I32x, and I64x */
11895 # ifdef USE_64_BIT_INT
11896 if (q[1] == '6' && q[2] == '4') {
11902 if (q[1] == '3' && q[2] == '2') {
11906 # ifdef USE_64_BIT_INT
11912 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11913 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11916 # ifdef USE_QUADMATH
11929 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11930 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11931 if (*q == 'l') { /* lld, llf */
11940 if (*++q == 'h') { /* hhd, hhu */
11969 if (!vectorize && !args) {
11971 const I32 i = efix-1;
11972 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11974 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11979 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11980 /* XXX va_arg(*args) case? need peek, use va_copy? */
11982 if (UNLIKELY(SvAMAGIC(argsv)))
11983 argsv = sv_2num(argsv);
11984 infnan = UNLIKELY(isinfnansv(argsv));
11987 switch (c = *q++) {
11995 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
11996 /* no va_arg() case */
11997 SvNV_nomg(argsv), (int)c);
11998 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
12000 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12002 eptr = (char*)utf8buf;
12003 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
12017 eptr = va_arg(*args, char*);
12019 elen = strlen(eptr);
12021 eptr = (char *)nullstr;
12022 elen = sizeof nullstr - 1;
12026 eptr = SvPV_const(argsv, elen);
12027 if (DO_UTF8(argsv)) {
12028 STRLEN old_precis = precis;
12029 if (has_precis && precis < elen) {
12030 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12031 STRLEN p = precis > ulen ? ulen : precis;
12032 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12033 /* sticks at end */
12035 if (width) { /* fudge width (can't fudge elen) */
12036 if (has_precis && precis < elen)
12037 width += precis - old_precis;
12040 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12047 if (has_precis && precis < elen)
12055 goto floating_point;
12057 if (alt || vectorize)
12059 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12073 goto floating_point;
12078 goto donevalidconversion;
12080 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12089 esignbuf[esignlen++] = plus;
12093 case 'c': iv = (char)va_arg(*args, int); break;
12094 case 'h': iv = (short)va_arg(*args, int); break;
12095 case 'l': iv = va_arg(*args, long); break;
12096 case 'V': iv = va_arg(*args, IV); break;
12097 case 'z': iv = va_arg(*args, SSize_t); break;
12098 #ifdef HAS_PTRDIFF_T
12099 case 't': iv = va_arg(*args, ptrdiff_t); break;
12101 default: iv = va_arg(*args, int); break;
12103 case 'j': iv = va_arg(*args, intmax_t); break;
12107 iv = va_arg(*args, Quad_t); break;
12114 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12116 case 'c': iv = (char)tiv; break;
12117 case 'h': iv = (short)tiv; break;
12118 case 'l': iv = (long)tiv; break;
12120 default: iv = tiv; break;
12123 iv = (Quad_t)tiv; break;
12129 if ( !vectorize ) /* we already set uv above */
12134 esignbuf[esignlen++] = plus;
12137 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12138 esignbuf[esignlen++] = '-';
12177 goto floating_point;
12183 goto donevalidconversion;
12185 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12196 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12197 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12198 case 'l': uv = va_arg(*args, unsigned long); break;
12199 case 'V': uv = va_arg(*args, UV); break;
12200 case 'z': uv = va_arg(*args, Size_t); break;
12201 #ifdef HAS_PTRDIFF_T
12202 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12205 case 'j': uv = va_arg(*args, uintmax_t); break;
12207 default: uv = va_arg(*args, unsigned); break;
12210 uv = va_arg(*args, Uquad_t); break;
12217 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12219 case 'c': uv = (unsigned char)tuv; break;
12220 case 'h': uv = (unsigned short)tuv; break;
12221 case 'l': uv = (unsigned long)tuv; break;
12223 default: uv = tuv; break;
12226 uv = (Uquad_t)tuv; break;
12235 char *ptr = ebuf + sizeof ebuf;
12236 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12242 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12246 } while (uv >>= 4);
12248 esignbuf[esignlen++] = '0';
12249 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12255 *--ptr = '0' + dig;
12256 } while (uv >>= 3);
12257 if (alt && *ptr != '0')
12263 *--ptr = '0' + dig;
12264 } while (uv >>= 1);
12266 esignbuf[esignlen++] = '0';
12267 esignbuf[esignlen++] = c;
12270 default: /* it had better be ten or less */
12273 *--ptr = '0' + dig;
12274 } while (uv /= base);
12277 elen = (ebuf + sizeof ebuf) - ptr;
12281 zeros = precis - elen;
12282 else if (precis == 0 && elen == 1 && *eptr == '0'
12283 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12286 /* a precision nullifies the 0 flag. */
12293 /* FLOATING POINT */
12298 c = 'f'; /* maybe %F isn't supported here */
12300 case 'e': case 'E':
12302 case 'g': case 'G':
12303 case 'a': case 'A':
12307 /* This is evil, but floating point is even more evil */
12309 /* for SV-style calling, we can only get NV
12310 for C-style calling, we assume %f is double;
12311 for simplicity we allow any of %Lf, %llf, %qf for long double
12315 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12319 /* [perl #20339] - we should accept and ignore %lf rather than die */
12323 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12324 intsize = args ? 0 : 'q';
12328 #if defined(HAS_LONG_DOUBLE)
12341 /* Now we need (long double) if intsize == 'q', else (double). */
12343 /* Note: do not pull NVs off the va_list with va_arg()
12344 * (pull doubles instead) because if you have a build
12345 * with long doubles, you would always be pulling long
12346 * doubles, which would badly break anyone using only
12347 * doubles (i.e. the majority of builds). In other
12348 * words, you cannot mix doubles and long doubles.
12349 * The only case where you can pull off long doubles
12350 * is when the format specifier explicitly asks so with
12352 #ifdef USE_QUADMATH
12353 fv = intsize == 'q' ?
12354 va_arg(*args, NV) : va_arg(*args, double);
12356 #elif LONG_DOUBLESIZE > DOUBLESIZE
12357 if (intsize == 'q') {
12358 fv = va_arg(*args, long double);
12361 nv = va_arg(*args, double);
12365 nv = va_arg(*args, double);
12371 if (!infnan) SvGETMAGIC(argsv);
12372 nv = SvNV_nomg(argsv);
12377 /* frexp() (or frexpl) has some unspecified behaviour for
12378 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12379 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12381 (void)Perl_frexp((NV)fv, &i);
12382 if (i == PERL_INT_MIN)
12383 Perl_die(aTHX_ "panic: frexp: %" FV_GF, fv);
12384 /* Do not set hexfp earlier since we want to printf
12385 * Inf/NaN for Inf/NaN, not their hexfp. */
12386 hexfp = isALPHA_FOLD_EQ(c, 'a');
12387 if (UNLIKELY(hexfp)) {
12388 /* This seriously overshoots in most cases, but
12389 * better the undershooting. Firstly, all bytes
12390 * of the NV are not mantissa, some of them are
12391 * exponent. Secondly, for the reasonably common
12392 * long doubles case, the "80-bit extended", two
12393 * or six bytes of the NV are unused. */
12395 (fv < 0) ? 1 : 0 + /* possible unary minus */
12397 1 + /* the very unlikely carry */
12400 2 * NVSIZE + /* 2 hexdigits for each byte */
12402 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12404 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12405 /* However, for the "double double", we need more.
12406 * Since each double has their own exponent, the
12407 * doubles may float (haha) rather far from each
12408 * other, and the number of required bits is much
12409 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12410 * See the definition of DOUBLEDOUBLE_MAXBITS.
12412 * Need 2 hexdigits for each byte. */
12413 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12414 /* the size for the exponent already added */
12416 #ifdef USE_LOCALE_NUMERIC
12417 STORE_LC_NUMERIC_SET_TO_NEEDED();
12418 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12419 need += SvLEN(PL_numeric_radix_sv);
12420 RESTORE_LC_NUMERIC();
12424 need = BIT_DIGITS(i);
12425 } /* if i < 0, the number of digits is hard to predict. */
12427 need += has_precis ? precis : 6; /* known default */
12432 #ifdef HAS_LDBL_SPRINTF_BUG
12433 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12434 with sfio - Allen <allens@cpan.org> */
12437 # define MY_DBL_MAX DBL_MAX
12438 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12439 # if DOUBLESIZE >= 8
12440 # define MY_DBL_MAX 1.7976931348623157E+308L
12442 # define MY_DBL_MAX 3.40282347E+38L
12446 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12447 # define MY_DBL_MAX_BUG 1L
12449 # define MY_DBL_MAX_BUG MY_DBL_MAX
12453 # define MY_DBL_MIN DBL_MIN
12454 # else /* XXX guessing! -Allen */
12455 # if DOUBLESIZE >= 8
12456 # define MY_DBL_MIN 2.2250738585072014E-308L
12458 # define MY_DBL_MIN 1.17549435E-38L
12462 if ((intsize == 'q') && (c == 'f') &&
12463 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12464 (need < DBL_DIG)) {
12465 /* it's going to be short enough that
12466 * long double precision is not needed */
12468 if ((fv <= 0L) && (fv >= -0L))
12469 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12471 /* would use Perl_fp_class as a double-check but not
12472 * functional on IRIX - see perl.h comments */
12474 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12475 /* It's within the range that a double can represent */
12476 #if defined(DBL_MAX) && !defined(DBL_MIN)
12477 if ((fv >= ((long double)1/DBL_MAX)) ||
12478 (fv <= (-(long double)1/DBL_MAX)))
12480 fix_ldbl_sprintf_bug = TRUE;
12483 if (fix_ldbl_sprintf_bug == TRUE) {
12493 # undef MY_DBL_MAX_BUG
12496 #endif /* HAS_LDBL_SPRINTF_BUG */
12498 need += 20; /* fudge factor */
12499 if (PL_efloatsize < need) {
12500 Safefree(PL_efloatbuf);
12501 PL_efloatsize = need + 20; /* more fudge */
12502 Newx(PL_efloatbuf, PL_efloatsize, char);
12503 PL_efloatbuf[0] = '\0';
12506 if ( !(width || left || plus || alt) && fill != '0'
12507 && has_precis && intsize != 'q' /* Shortcuts */
12508 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12509 /* See earlier comment about buggy Gconvert when digits,
12511 if ( c == 'g' && precis ) {
12512 STORE_LC_NUMERIC_SET_TO_NEEDED();
12513 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12514 /* May return an empty string for digits==0 */
12515 if (*PL_efloatbuf) {
12516 elen = strlen(PL_efloatbuf);
12517 goto float_converted;
12519 } else if ( c == 'f' && !precis ) {
12520 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12525 if (UNLIKELY(hexfp)) {
12526 /* Hexadecimal floating point. */
12527 char* p = PL_efloatbuf;
12528 U8 vhex[VHEX_SIZE];
12529 U8* v = vhex; /* working pointer to vhex */
12530 U8* vend; /* pointer to one beyond last digit of vhex */
12531 U8* vfnz = NULL; /* first non-zero */
12532 U8* vlnz = NULL; /* last non-zero */
12533 U8* v0 = NULL; /* first output */
12534 const bool lower = (c == 'a');
12535 /* At output the values of vhex (up to vend) will
12536 * be mapped through the xdig to get the actual
12537 * human-readable xdigits. */
12538 const char* xdig = PL_hexdigit;
12539 int zerotail = 0; /* how many extra zeros to append */
12540 int exponent = 0; /* exponent of the floating point input */
12541 bool hexradix = FALSE; /* should we output the radix */
12542 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12543 bool negative = FALSE;
12545 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12547 * For example with denormals, (assuming the vanilla
12548 * 64-bit double): the exponent is zero. 1xp-1074 is
12549 * the smallest denormal and the smallest double, it
12550 * could be output also as 0x0.0000000000001p-1022 to
12551 * match its internal structure. */
12553 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12554 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12556 #if NVSIZE > DOUBLESIZE
12557 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12558 /* In this case there is an implicit bit,
12559 * and therefore the exponent is shifted by one. */
12562 # ifdef NV_X86_80_BIT
12564 /* The subnormals of the x86-80 have a base exponent of -16382,
12565 * (while the physical exponent bits are zero) but the frexp()
12566 * returned the scientific-style floating exponent. We want
12567 * to map the last one as:
12568 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12569 * -16835..-16388 -> -16384
12570 * since we want to keep the first hexdigit
12571 * as one of the [8421]. */
12572 exponent = -4 * ( (exponent + 1) / -4) - 2;
12577 /* TBD: other non-implicit-bit platforms than the x86-80. */
12581 negative = fv < 0 || Perl_signbit(nv);
12592 xdig += 16; /* Use uppercase hex. */
12595 /* Find the first non-zero xdigit. */
12596 for (v = vhex; v < vend; v++) {
12604 /* Find the last non-zero xdigit. */
12605 for (v = vend - 1; v >= vhex; v--) {
12612 #if NVSIZE == DOUBLESIZE
12618 #ifndef NV_X86_80_BIT
12620 /* IEEE 754 subnormals (but not the x86 80-bit):
12621 * we want "normalize" the subnormal,
12622 * so we need to right shift the hex nybbles
12623 * so that the output of the subnormal starts
12624 * from the first true bit. (Another, equally
12625 * valid, policy would be to dump the subnormal
12626 * nybbles as-is, to display the "physical" layout.) */
12629 /* Find the ceil(log2(v[0])) of
12630 * the top non-zero nybble. */
12631 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12634 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12635 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12649 U8* ve = (subnormal ? vlnz + 1 : vend);
12650 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12651 if ((SSize_t)(precis + 1) < vn) {
12652 bool overflow = FALSE;
12653 if (v0[precis + 1] < 0x8) {
12654 /* Round down, nothing to do. */
12655 } else if (v0[precis + 1] > 0x8) {
12658 overflow = v0[precis] > 0xF;
12660 } else { /* v0[precis] == 0x8 */
12661 /* Half-point: round towards the one
12662 * with the even least-significant digit:
12670 * 78 -> 8 f8 -> 10 */
12671 if ((v0[precis] & 0x1)) {
12674 overflow = v0[precis] > 0xF;
12679 for (v = v0 + precis - 1; v >= v0; v--) {
12681 overflow = *v > 0xF;
12687 if (v == v0 - 1 && overflow) {
12688 /* If the overflow goes all the
12689 * way to the front, we need to
12690 * insert 0x1 in front, and adjust
12692 Move(v0, v0 + 1, vn, char);
12698 /* The new effective "last non zero". */
12699 vlnz = v0 + precis;
12703 subnormal ? precis - vn + 1 :
12704 precis - (vlnz - vhex);
12711 /* If there are non-zero xdigits, the radix
12712 * is output after the first one. */
12723 /* The radix is always output if precis, or if alt. */
12724 if (precis > 0 || alt) {
12729 #ifndef USE_LOCALE_NUMERIC
12732 STORE_LC_NUMERIC_SET_TO_NEEDED();
12733 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12735 const char* r = SvPV(PL_numeric_radix_sv, n);
12736 Copy(r, p, n, char);
12742 RESTORE_LC_NUMERIC();
12751 if (zerotail > 0) {
12752 while (zerotail--) {
12757 elen = p - PL_efloatbuf;
12758 elen += my_snprintf(p, PL_efloatsize - elen,
12759 "%c%+d", lower ? 'p' : 'P',
12762 if (elen < width) {
12764 /* Pad the back with spaces. */
12765 memset(PL_efloatbuf + elen, ' ', width - elen);
12767 else if (fill == '0') {
12768 /* Insert the zeros after the "0x" and the
12769 * the potential sign, but before the digits,
12770 * otherwise we end up with "0000xH.HHH...",
12771 * when we want "0x000H.HHH..." */
12772 STRLEN nzero = width - elen;
12773 char* zerox = PL_efloatbuf + 2;
12774 STRLEN nmove = elen - 2;
12775 if (negative || plus) {
12779 Move(zerox, zerox + nzero, nmove, char);
12780 memset(zerox, fill, nzero);
12783 /* Move it to the right. */
12784 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12786 /* Pad the front with spaces. */
12787 memset(PL_efloatbuf, ' ', width - elen);
12793 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12795 /* Not affecting infnan output: precision, alt, fill. */
12796 if (elen < width) {
12798 /* Pack the back with spaces. */
12799 memset(PL_efloatbuf + elen, ' ', width - elen);
12801 /* Move it to the right. */
12802 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12804 /* Pad the front with spaces. */
12805 memset(PL_efloatbuf, ' ', width - elen);
12813 char *ptr = ebuf + sizeof ebuf;
12816 #if defined(USE_QUADMATH)
12817 if (intsize == 'q') {
12821 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12822 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12823 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12824 * not USE_LONG_DOUBLE and NVff. In other words,
12825 * this needs to work without USE_LONG_DOUBLE. */
12826 if (intsize == 'q') {
12827 /* Copy the one or more characters in a long double
12828 * format before the 'base' ([efgEFG]) character to
12829 * the format string. */
12830 static char const ldblf[] = PERL_PRIfldbl;
12831 char const *p = ldblf + sizeof(ldblf) - 3;
12832 while (p >= ldblf) { *--ptr = *p--; }
12837 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12842 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12854 /* No taint. Otherwise we are in the strange situation
12855 * where printf() taints but print($float) doesn't.
12858 STORE_LC_NUMERIC_SET_TO_NEEDED();
12860 /* hopefully the above makes ptr a very constrained format
12861 * that is safe to use, even though it's not literal */
12862 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12863 #ifdef USE_QUADMATH
12865 const char* qfmt = quadmath_format_single(ptr);
12867 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12868 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12870 if ((IV)elen == -1)
12871 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12875 #elif defined(HAS_LONG_DOUBLE)
12876 elen = ((intsize == 'q')
12877 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12878 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12880 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12886 eptr = PL_efloatbuf;
12887 assert((IV)elen > 0); /* here zero elen is bad */
12889 #ifdef USE_LOCALE_NUMERIC
12890 /* If the decimal point character in the string is UTF-8, make the
12892 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12893 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12906 i = SvCUR(sv) - origlen;
12909 case 'c': *(va_arg(*args, char*)) = i; break;
12910 case 'h': *(va_arg(*args, short*)) = i; break;
12911 default: *(va_arg(*args, int*)) = i; break;
12912 case 'l': *(va_arg(*args, long*)) = i; break;
12913 case 'V': *(va_arg(*args, IV*)) = i; break;
12914 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12915 #ifdef HAS_PTRDIFF_T
12916 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12919 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12923 *(va_arg(*args, Quad_t*)) = i; break;
12930 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12931 goto donevalidconversion;
12938 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12939 && ckWARN(WARN_PRINTF))
12941 SV * const msg = sv_newmortal();
12942 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12943 (PL_op->op_type == OP_PRTF) ? "" : "s");
12944 if (fmtstart < patend) {
12945 const char * const fmtend = q < patend ? q : patend;
12947 sv_catpvs(msg, "\"%");
12948 for (f = fmtstart; f < fmtend; f++) {
12950 sv_catpvn_nomg(msg, f, 1);
12952 Perl_sv_catpvf(aTHX_ msg,
12953 "\\%03" UVof, (UV)*f & 0xFF);
12956 sv_catpvs(msg, "\"");
12958 sv_catpvs(msg, "end of string");
12960 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
12963 /* output mangled stuff ... */
12969 /* ... right here, because formatting flags should not apply */
12970 SvGROW(sv, SvCUR(sv) + elen + 1);
12972 Copy(eptr, p, elen, char);
12975 SvCUR_set(sv, p - SvPVX_const(sv));
12977 continue; /* not "break" */
12980 if (is_utf8 != has_utf8) {
12983 sv_utf8_upgrade(sv);
12986 const STRLEN old_elen = elen;
12987 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12988 sv_utf8_upgrade(nsv);
12989 eptr = SvPVX_const(nsv);
12992 if (width) { /* fudge width (can't fudge elen) */
12993 width += elen - old_elen;
12999 /* signed value that's wrapped? */
13000 assert(elen <= ((~(STRLEN)0) >> 1));
13001 have = esignlen + zeros + elen;
13003 croak_memory_wrap();
13005 need = (have > width ? have : width);
13008 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
13009 croak_memory_wrap();
13010 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
13012 if (esignlen && fill == '0') {
13014 for (i = 0; i < (int)esignlen; i++)
13015 *p++ = esignbuf[i];
13017 if (gap && !left) {
13018 memset(p, fill, gap);
13021 if (esignlen && fill != '0') {
13023 for (i = 0; i < (int)esignlen; i++)
13024 *p++ = esignbuf[i];
13028 for (i = zeros; i; i--)
13032 Copy(eptr, p, elen, char);
13036 memset(p, ' ', gap);
13041 Copy(dotstr, p, dotstrlen, char);
13045 vectorize = FALSE; /* done iterating over vecstr */
13052 SvCUR_set(sv, p - SvPVX_const(sv));
13058 donevalidconversion:
13059 if (used_explicit_ix)
13060 no_redundant_warning = TRUE;
13062 S_warn_vcatpvfn_missing_argument(aTHX);
13065 /* Now that we've consumed all our printf format arguments (svix)
13066 * do we have things left on the stack that we didn't use?
13068 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13069 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13070 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13075 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13079 /* =========================================================================
13081 =head1 Cloning an interpreter
13085 All the macros and functions in this section are for the private use of
13086 the main function, perl_clone().
13088 The foo_dup() functions make an exact copy of an existing foo thingy.
13089 During the course of a cloning, a hash table is used to map old addresses
13090 to new addresses. The table is created and manipulated with the
13091 ptr_table_* functions.
13093 * =========================================================================*/
13096 #if defined(USE_ITHREADS)
13098 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13099 #ifndef GpREFCNT_inc
13100 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13104 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13105 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13106 If this changes, please unmerge ss_dup.
13107 Likewise, sv_dup_inc_multiple() relies on this fact. */
13108 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13109 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13110 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13111 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13112 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13113 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13114 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13115 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13116 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13117 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13118 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13119 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13120 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13122 /* clone a parser */
13125 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13129 PERL_ARGS_ASSERT_PARSER_DUP;
13134 /* look for it in the table first */
13135 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13139 /* create anew and remember what it is */
13140 Newxz(parser, 1, yy_parser);
13141 ptr_table_store(PL_ptr_table, proto, parser);
13143 /* XXX these not yet duped */
13144 parser->old_parser = NULL;
13145 parser->stack = NULL;
13147 parser->stack_max1 = 0;
13148 /* XXX parser->stack->state = 0; */
13150 /* XXX eventually, just Copy() most of the parser struct ? */
13152 parser->lex_brackets = proto->lex_brackets;
13153 parser->lex_casemods = proto->lex_casemods;
13154 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13155 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13156 parser->lex_casestack = savepvn(proto->lex_casestack,
13157 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13158 parser->lex_defer = proto->lex_defer;
13159 parser->lex_dojoin = proto->lex_dojoin;
13160 parser->lex_formbrack = proto->lex_formbrack;
13161 parser->lex_inpat = proto->lex_inpat;
13162 parser->lex_inwhat = proto->lex_inwhat;
13163 parser->lex_op = proto->lex_op;
13164 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13165 parser->lex_starts = proto->lex_starts;
13166 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13167 parser->multi_close = proto->multi_close;
13168 parser->multi_open = proto->multi_open;
13169 parser->multi_start = proto->multi_start;
13170 parser->multi_end = proto->multi_end;
13171 parser->preambled = proto->preambled;
13172 parser->lex_super_state = proto->lex_super_state;
13173 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13174 parser->lex_sub_op = proto->lex_sub_op;
13175 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13176 parser->linestr = sv_dup_inc(proto->linestr, param);
13177 parser->expect = proto->expect;
13178 parser->copline = proto->copline;
13179 parser->last_lop_op = proto->last_lop_op;
13180 parser->lex_state = proto->lex_state;
13181 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13182 /* rsfp_filters entries have fake IoDIRP() */
13183 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13184 parser->in_my = proto->in_my;
13185 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13186 parser->error_count = proto->error_count;
13187 parser->sig_elems = proto->sig_elems;
13188 parser->sig_optelems= proto->sig_optelems;
13189 parser->sig_slurpy = proto->sig_slurpy;
13190 parser->linestr = sv_dup_inc(proto->linestr, param);
13193 char * const ols = SvPVX(proto->linestr);
13194 char * const ls = SvPVX(parser->linestr);
13196 parser->bufptr = ls + (proto->bufptr >= ols ?
13197 proto->bufptr - ols : 0);
13198 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13199 proto->oldbufptr - ols : 0);
13200 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13201 proto->oldoldbufptr - ols : 0);
13202 parser->linestart = ls + (proto->linestart >= ols ?
13203 proto->linestart - ols : 0);
13204 parser->last_uni = ls + (proto->last_uni >= ols ?
13205 proto->last_uni - ols : 0);
13206 parser->last_lop = ls + (proto->last_lop >= ols ?
13207 proto->last_lop - ols : 0);
13209 parser->bufend = ls + SvCUR(parser->linestr);
13212 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13215 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13216 Copy(proto->nexttype, parser->nexttype, 5, I32);
13217 parser->nexttoke = proto->nexttoke;
13219 /* XXX should clone saved_curcop here, but we aren't passed
13220 * proto_perl; so do it in perl_clone_using instead */
13226 /* duplicate a file handle */
13229 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13233 PERL_ARGS_ASSERT_FP_DUP;
13234 PERL_UNUSED_ARG(type);
13237 return (PerlIO*)NULL;
13239 /* look for it in the table first */
13240 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13244 /* create anew and remember what it is */
13245 #ifdef __amigaos4__
13246 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13248 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13250 ptr_table_store(PL_ptr_table, fp, ret);
13254 /* duplicate a directory handle */
13257 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13261 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13263 const Direntry_t *dirent;
13264 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13270 PERL_UNUSED_CONTEXT;
13271 PERL_ARGS_ASSERT_DIRP_DUP;
13276 /* look for it in the table first */
13277 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13281 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13283 PERL_UNUSED_ARG(param);
13287 /* open the current directory (so we can switch back) */
13288 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13290 /* chdir to our dir handle and open the present working directory */
13291 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13292 PerlDir_close(pwd);
13293 return (DIR *)NULL;
13295 /* Now we should have two dir handles pointing to the same dir. */
13297 /* Be nice to the calling code and chdir back to where we were. */
13298 /* XXX If this fails, then what? */
13299 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13301 /* We have no need of the pwd handle any more. */
13302 PerlDir_close(pwd);
13305 # define d_namlen(d) (d)->d_namlen
13307 # define d_namlen(d) strlen((d)->d_name)
13309 /* Iterate once through dp, to get the file name at the current posi-
13310 tion. Then step back. */
13311 pos = PerlDir_tell(dp);
13312 if ((dirent = PerlDir_read(dp))) {
13313 len = d_namlen(dirent);
13314 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13315 /* If the len is somehow magically longer than the
13316 * maximum length of the directory entry, even though
13317 * we could fit it in a buffer, we could not copy it
13318 * from the dirent. Bail out. */
13319 PerlDir_close(ret);
13322 if (len <= sizeof smallbuf) name = smallbuf;
13323 else Newx(name, len, char);
13324 Move(dirent->d_name, name, len, char);
13326 PerlDir_seek(dp, pos);
13328 /* Iterate through the new dir handle, till we find a file with the
13330 if (!dirent) /* just before the end */
13332 pos = PerlDir_tell(ret);
13333 if (PerlDir_read(ret)) continue; /* not there yet */
13334 PerlDir_seek(ret, pos); /* step back */
13338 const long pos0 = PerlDir_tell(ret);
13340 pos = PerlDir_tell(ret);
13341 if ((dirent = PerlDir_read(ret))) {
13342 if (len == (STRLEN)d_namlen(dirent)
13343 && memEQ(name, dirent->d_name, len)) {
13345 PerlDir_seek(ret, pos); /* step back */
13348 /* else we are not there yet; keep iterating */
13350 else { /* This is not meant to happen. The best we can do is
13351 reset the iterator to the beginning. */
13352 PerlDir_seek(ret, pos0);
13359 if (name && name != smallbuf)
13364 ret = win32_dirp_dup(dp, param);
13367 /* pop it in the pointer table */
13369 ptr_table_store(PL_ptr_table, dp, ret);
13374 /* duplicate a typeglob */
13377 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13381 PERL_ARGS_ASSERT_GP_DUP;
13385 /* look for it in the table first */
13386 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13390 /* create anew and remember what it is */
13392 ptr_table_store(PL_ptr_table, gp, ret);
13395 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13396 on Newxz() to do this for us. */
13397 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13398 ret->gp_io = io_dup_inc(gp->gp_io, param);
13399 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13400 ret->gp_av = av_dup_inc(gp->gp_av, param);
13401 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13402 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13403 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13404 ret->gp_cvgen = gp->gp_cvgen;
13405 ret->gp_line = gp->gp_line;
13406 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13410 /* duplicate a chain of magic */
13413 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13415 MAGIC *mgret = NULL;
13416 MAGIC **mgprev_p = &mgret;
13418 PERL_ARGS_ASSERT_MG_DUP;
13420 for (; mg; mg = mg->mg_moremagic) {
13423 if ((param->flags & CLONEf_JOIN_IN)
13424 && mg->mg_type == PERL_MAGIC_backref)
13425 /* when joining, we let the individual SVs add themselves to
13426 * backref as needed. */
13429 Newx(nmg, 1, MAGIC);
13431 mgprev_p = &(nmg->mg_moremagic);
13433 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13434 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13435 from the original commit adding Perl_mg_dup() - revision 4538.
13436 Similarly there is the annotation "XXX random ptr?" next to the
13437 assignment to nmg->mg_ptr. */
13440 /* FIXME for plugins
13441 if (nmg->mg_type == PERL_MAGIC_qr) {
13442 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13446 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13447 ? nmg->mg_type == PERL_MAGIC_backref
13448 /* The backref AV has its reference
13449 * count deliberately bumped by 1 */
13450 ? SvREFCNT_inc(av_dup_inc((const AV *)
13451 nmg->mg_obj, param))
13452 : sv_dup_inc(nmg->mg_obj, param)
13453 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13454 nmg->mg_type == PERL_MAGIC_regdata)
13456 : sv_dup(nmg->mg_obj, param);
13458 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13459 if (nmg->mg_len > 0) {
13460 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13461 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13462 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13464 AMT * const namtp = (AMT*)nmg->mg_ptr;
13465 sv_dup_inc_multiple((SV**)(namtp->table),
13466 (SV**)(namtp->table), NofAMmeth, param);
13469 else if (nmg->mg_len == HEf_SVKEY)
13470 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13472 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13473 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13479 #endif /* USE_ITHREADS */
13481 struct ptr_tbl_arena {
13482 struct ptr_tbl_arena *next;
13483 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13486 /* create a new pointer-mapping table */
13489 Perl_ptr_table_new(pTHX)
13492 PERL_UNUSED_CONTEXT;
13494 Newx(tbl, 1, PTR_TBL_t);
13495 tbl->tbl_max = 511;
13496 tbl->tbl_items = 0;
13497 tbl->tbl_arena = NULL;
13498 tbl->tbl_arena_next = NULL;
13499 tbl->tbl_arena_end = NULL;
13500 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13504 #define PTR_TABLE_HASH(ptr) \
13505 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13507 /* map an existing pointer using a table */
13509 STATIC PTR_TBL_ENT_t *
13510 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13512 PTR_TBL_ENT_t *tblent;
13513 const UV hash = PTR_TABLE_HASH(sv);
13515 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13517 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13518 for (; tblent; tblent = tblent->next) {
13519 if (tblent->oldval == sv)
13526 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13528 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13530 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13531 PERL_UNUSED_CONTEXT;
13533 return tblent ? tblent->newval : NULL;
13536 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13537 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13538 * the core's typical use of ptr_tables in thread cloning. */
13541 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13543 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13545 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13546 PERL_UNUSED_CONTEXT;
13549 tblent->newval = newsv;
13551 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13553 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13554 struct ptr_tbl_arena *new_arena;
13556 Newx(new_arena, 1, struct ptr_tbl_arena);
13557 new_arena->next = tbl->tbl_arena;
13558 tbl->tbl_arena = new_arena;
13559 tbl->tbl_arena_next = new_arena->array;
13560 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13563 tblent = tbl->tbl_arena_next++;
13565 tblent->oldval = oldsv;
13566 tblent->newval = newsv;
13567 tblent->next = tbl->tbl_ary[entry];
13568 tbl->tbl_ary[entry] = tblent;
13570 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13571 ptr_table_split(tbl);
13575 /* double the hash bucket size of an existing ptr table */
13578 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13580 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13581 const UV oldsize = tbl->tbl_max + 1;
13582 UV newsize = oldsize * 2;
13585 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13586 PERL_UNUSED_CONTEXT;
13588 Renew(ary, newsize, PTR_TBL_ENT_t*);
13589 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13590 tbl->tbl_max = --newsize;
13591 tbl->tbl_ary = ary;
13592 for (i=0; i < oldsize; i++, ary++) {
13593 PTR_TBL_ENT_t **entp = ary;
13594 PTR_TBL_ENT_t *ent = *ary;
13595 PTR_TBL_ENT_t **curentp;
13598 curentp = ary + oldsize;
13600 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13602 ent->next = *curentp;
13612 /* remove all the entries from a ptr table */
13613 /* Deprecated - will be removed post 5.14 */
13616 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13618 PERL_UNUSED_CONTEXT;
13619 if (tbl && tbl->tbl_items) {
13620 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13622 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13625 struct ptr_tbl_arena *next = arena->next;
13631 tbl->tbl_items = 0;
13632 tbl->tbl_arena = NULL;
13633 tbl->tbl_arena_next = NULL;
13634 tbl->tbl_arena_end = NULL;
13638 /* clear and free a ptr table */
13641 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13643 struct ptr_tbl_arena *arena;
13645 PERL_UNUSED_CONTEXT;
13651 arena = tbl->tbl_arena;
13654 struct ptr_tbl_arena *next = arena->next;
13660 Safefree(tbl->tbl_ary);
13664 #if defined(USE_ITHREADS)
13667 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13669 PERL_ARGS_ASSERT_RVPV_DUP;
13671 assert(!isREGEXP(sstr));
13673 if (SvWEAKREF(sstr)) {
13674 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13675 if (param->flags & CLONEf_JOIN_IN) {
13676 /* if joining, we add any back references individually rather
13677 * than copying the whole backref array */
13678 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13682 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13684 else if (SvPVX_const(sstr)) {
13685 /* Has something there */
13687 /* Normal PV - clone whole allocated space */
13688 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13689 /* sstr may not be that normal, but actually copy on write.
13690 But we are a true, independent SV, so: */
13694 /* Special case - not normally malloced for some reason */
13695 if (isGV_with_GP(sstr)) {
13696 /* Don't need to do anything here. */
13698 else if ((SvIsCOW(sstr))) {
13699 /* A "shared" PV - clone it as "shared" PV */
13701 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13705 /* Some other special case - random pointer */
13706 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13711 /* Copy the NULL */
13712 SvPV_set(dstr, NULL);
13716 /* duplicate a list of SVs. source and dest may point to the same memory. */
13718 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13719 SSize_t items, CLONE_PARAMS *const param)
13721 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13723 while (items-- > 0) {
13724 *dest++ = sv_dup_inc(*source++, param);
13730 /* duplicate an SV of any type (including AV, HV etc) */
13733 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13738 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13740 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13741 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13746 /* look for it in the table first */
13747 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13751 if(param->flags & CLONEf_JOIN_IN) {
13752 /** We are joining here so we don't want do clone
13753 something that is bad **/
13754 if (SvTYPE(sstr) == SVt_PVHV) {
13755 const HEK * const hvname = HvNAME_HEK(sstr);
13757 /** don't clone stashes if they already exist **/
13758 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13759 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13760 ptr_table_store(PL_ptr_table, sstr, dstr);
13764 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13765 HV *stash = GvSTASH(sstr);
13766 const HEK * hvname;
13767 if (stash && (hvname = HvNAME_HEK(stash))) {
13768 /** don't clone GVs if they already exist **/
13770 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13771 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13773 stash, GvNAME(sstr),
13779 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13780 ptr_table_store(PL_ptr_table, sstr, *svp);
13787 /* create anew and remember what it is */
13790 #ifdef DEBUG_LEAKING_SCALARS
13791 dstr->sv_debug_optype = sstr->sv_debug_optype;
13792 dstr->sv_debug_line = sstr->sv_debug_line;
13793 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13794 dstr->sv_debug_parent = (SV*)sstr;
13795 FREE_SV_DEBUG_FILE(dstr);
13796 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13799 ptr_table_store(PL_ptr_table, sstr, dstr);
13802 SvFLAGS(dstr) = SvFLAGS(sstr);
13803 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13804 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13807 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13808 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13809 (void*)PL_watch_pvx, SvPVX_const(sstr));
13812 /* don't clone objects whose class has asked us not to */
13814 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13820 switch (SvTYPE(sstr)) {
13822 SvANY(dstr) = NULL;
13825 SET_SVANY_FOR_BODYLESS_IV(dstr);
13827 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13829 SvIV_set(dstr, SvIVX(sstr));
13833 #if NVSIZE <= IVSIZE
13834 SET_SVANY_FOR_BODYLESS_NV(dstr);
13836 SvANY(dstr) = new_XNV();
13838 SvNV_set(dstr, SvNVX(sstr));
13842 /* These are all the types that need complex bodies allocating. */
13844 const svtype sv_type = SvTYPE(sstr);
13845 const struct body_details *const sv_type_details
13846 = bodies_by_type + sv_type;
13850 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13866 assert(sv_type_details->body_size);
13867 if (sv_type_details->arena) {
13868 new_body_inline(new_body, sv_type);
13870 = (void*)((char*)new_body - sv_type_details->offset);
13872 new_body = new_NOARENA(sv_type_details);
13876 SvANY(dstr) = new_body;
13879 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13880 ((char*)SvANY(dstr)) + sv_type_details->offset,
13881 sv_type_details->copy, char);
13883 Copy(((char*)SvANY(sstr)),
13884 ((char*)SvANY(dstr)),
13885 sv_type_details->body_size + sv_type_details->offset, char);
13888 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13889 && !isGV_with_GP(dstr)
13891 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13892 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13894 /* The Copy above means that all the source (unduplicated) pointers
13895 are now in the destination. We can check the flags and the
13896 pointers in either, but it's possible that there's less cache
13897 missing by always going for the destination.
13898 FIXME - instrument and check that assumption */
13899 if (sv_type >= SVt_PVMG) {
13901 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13902 if (SvOBJECT(dstr) && SvSTASH(dstr))
13903 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13904 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13907 /* The cast silences a GCC warning about unhandled types. */
13908 switch ((int)sv_type) {
13919 /* FIXME for plugins */
13920 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13921 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13924 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13925 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13926 LvTARG(dstr) = dstr;
13927 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13928 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13930 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13931 if (isREGEXP(sstr)) goto duprex;
13933 /* non-GP case already handled above */
13934 if(isGV_with_GP(sstr)) {
13935 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13936 /* Don't call sv_add_backref here as it's going to be
13937 created as part of the magic cloning of the symbol
13938 table--unless this is during a join and the stash
13939 is not actually being cloned. */
13940 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13941 at the point of this comment. */
13942 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13943 if (param->flags & CLONEf_JOIN_IN)
13944 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13945 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13946 (void)GpREFCNT_inc(GvGP(dstr));
13950 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13951 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13952 /* I have no idea why fake dirp (rsfps)
13953 should be treated differently but otherwise
13954 we end up with leaks -- sky*/
13955 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13956 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13957 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13959 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13960 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13961 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13962 if (IoDIRP(dstr)) {
13963 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13966 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13968 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13970 if (IoOFP(dstr) == IoIFP(sstr))
13971 IoOFP(dstr) = IoIFP(dstr);
13973 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13974 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13975 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13976 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13979 /* avoid cloning an empty array */
13980 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13981 SV **dst_ary, **src_ary;
13982 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13984 src_ary = AvARRAY((const AV *)sstr);
13985 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13986 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13987 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13988 AvALLOC((const AV *)dstr) = dst_ary;
13989 if (AvREAL((const AV *)sstr)) {
13990 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13994 while (items-- > 0)
13995 *dst_ary++ = sv_dup(*src_ary++, param);
13997 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13998 while (items-- > 0) {
14003 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14004 AvALLOC((const AV *)dstr) = (SV**)NULL;
14005 AvMAX( (const AV *)dstr) = -1;
14006 AvFILLp((const AV *)dstr) = -1;
14010 if (HvARRAY((const HV *)sstr)) {
14012 const bool sharekeys = !!HvSHAREKEYS(sstr);
14013 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14014 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14016 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14017 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14019 HvARRAY(dstr) = (HE**)darray;
14020 while (i <= sxhv->xhv_max) {
14021 const HE * const source = HvARRAY(sstr)[i];
14022 HvARRAY(dstr)[i] = source
14023 ? he_dup(source, sharekeys, param) : 0;
14027 const struct xpvhv_aux * const saux = HvAUX(sstr);
14028 struct xpvhv_aux * const daux = HvAUX(dstr);
14029 /* This flag isn't copied. */
14032 if (saux->xhv_name_count) {
14033 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14035 = saux->xhv_name_count < 0
14036 ? -saux->xhv_name_count
14037 : saux->xhv_name_count;
14038 HEK **shekp = sname + count;
14040 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14041 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14042 while (shekp-- > sname) {
14044 *dhekp = hek_dup(*shekp, param);
14048 daux->xhv_name_u.xhvnameu_name
14049 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14052 daux->xhv_name_count = saux->xhv_name_count;
14054 daux->xhv_aux_flags = saux->xhv_aux_flags;
14055 #ifdef PERL_HASH_RANDOMIZE_KEYS
14056 daux->xhv_rand = saux->xhv_rand;
14057 daux->xhv_last_rand = saux->xhv_last_rand;
14059 daux->xhv_riter = saux->xhv_riter;
14060 daux->xhv_eiter = saux->xhv_eiter
14061 ? he_dup(saux->xhv_eiter,
14062 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14063 /* backref array needs refcnt=2; see sv_add_backref */
14064 daux->xhv_backreferences =
14065 (param->flags & CLONEf_JOIN_IN)
14066 /* when joining, we let the individual GVs and
14067 * CVs add themselves to backref as
14068 * needed. This avoids pulling in stuff
14069 * that isn't required, and simplifies the
14070 * case where stashes aren't cloned back
14071 * if they already exist in the parent
14074 : saux->xhv_backreferences
14075 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14076 ? MUTABLE_AV(SvREFCNT_inc(
14077 sv_dup_inc((const SV *)
14078 saux->xhv_backreferences, param)))
14079 : MUTABLE_AV(sv_dup((const SV *)
14080 saux->xhv_backreferences, param))
14083 daux->xhv_mro_meta = saux->xhv_mro_meta
14084 ? mro_meta_dup(saux->xhv_mro_meta, param)
14087 /* Record stashes for possible cloning in Perl_clone(). */
14089 av_push(param->stashes, dstr);
14093 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14096 if (!(param->flags & CLONEf_COPY_STACKS)) {
14101 /* NOTE: not refcounted */
14102 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14103 hv_dup(CvSTASH(dstr), param);
14104 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14105 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14106 if (!CvISXSUB(dstr)) {
14108 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14110 CvSLABBED_off(dstr);
14111 } else if (CvCONST(dstr)) {
14112 CvXSUBANY(dstr).any_ptr =
14113 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14115 assert(!CvSLABBED(dstr));
14116 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14118 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14119 hek_dup(CvNAME_HEK((CV *)sstr), param);
14120 /* don't dup if copying back - CvGV isn't refcounted, so the
14121 * duped GV may never be freed. A bit of a hack! DAPM */
14123 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14125 ? gv_dup_inc(CvGV(sstr), param)
14126 : (param->flags & CLONEf_JOIN_IN)
14128 : gv_dup(CvGV(sstr), param);
14130 if (!CvISXSUB(sstr)) {
14131 PADLIST * padlist = CvPADLIST(sstr);
14133 padlist = padlist_dup(padlist, param);
14134 CvPADLIST_set(dstr, padlist);
14136 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14137 PoisonPADLIST(dstr);
14140 CvWEAKOUTSIDE(sstr)
14141 ? cv_dup( CvOUTSIDE(dstr), param)
14142 : cv_dup_inc(CvOUTSIDE(dstr), param);
14152 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14154 PERL_ARGS_ASSERT_SV_DUP_INC;
14155 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14159 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14161 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14162 PERL_ARGS_ASSERT_SV_DUP;
14164 /* Track every SV that (at least initially) had a reference count of 0.
14165 We need to do this by holding an actual reference to it in this array.
14166 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14167 (akin to the stashes hash, and the perl stack), we come unstuck if
14168 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14169 thread) is manipulated in a CLONE method, because CLONE runs before the
14170 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14171 (and fix things up by giving each a reference via the temps stack).
14172 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14173 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14174 before the walk of unreferenced happens and a reference to that is SV
14175 added to the temps stack. At which point we have the same SV considered
14176 to be in use, and free to be re-used. Not good.
14178 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14179 assert(param->unreferenced);
14180 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14186 /* duplicate a context */
14189 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14191 PERL_CONTEXT *ncxs;
14193 PERL_ARGS_ASSERT_CX_DUP;
14196 return (PERL_CONTEXT*)NULL;
14198 /* look for it in the table first */
14199 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14203 /* create anew and remember what it is */
14204 Newx(ncxs, max + 1, PERL_CONTEXT);
14205 ptr_table_store(PL_ptr_table, cxs, ncxs);
14206 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14209 PERL_CONTEXT * const ncx = &ncxs[ix];
14210 if (CxTYPE(ncx) == CXt_SUBST) {
14211 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14214 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14215 switch (CxTYPE(ncx)) {
14217 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14218 if(CxHASARGS(ncx)){
14219 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14221 ncx->blk_sub.savearray = NULL;
14223 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14224 ncx->blk_sub.prevcomppad);
14227 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14229 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14230 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14231 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14232 /* XXX what do do with cur_top_env ???? */
14234 case CXt_LOOP_LAZYSV:
14235 ncx->blk_loop.state_u.lazysv.end
14236 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14237 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14238 duplication code instead.
14239 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14240 actually being the same function, and (2) order
14241 equivalence of the two unions.
14242 We can assert the later [but only at run time :-(] */
14243 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14244 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14247 ncx->blk_loop.state_u.ary.ary
14248 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14250 case CXt_LOOP_LIST:
14251 case CXt_LOOP_LAZYIV:
14252 /* code common to all 'for' CXt_LOOP_* types */
14253 ncx->blk_loop.itersave =
14254 sv_dup_inc(ncx->blk_loop.itersave, param);
14255 if (CxPADLOOP(ncx)) {
14256 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14257 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14258 ncx->blk_loop.oldcomppad =
14259 (PAD*)ptr_table_fetch(PL_ptr_table,
14260 ncx->blk_loop.oldcomppad);
14261 ncx->blk_loop.itervar_u.svp =
14262 &CX_CURPAD_SV(ncx->blk_loop, off);
14265 /* this copies the GV if CXp_FOR_GV, or the SV for an
14266 * alias (for \$x (...)) - relies on gv_dup being the
14267 * same as sv_dup */
14268 ncx->blk_loop.itervar_u.gv
14269 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14273 case CXt_LOOP_PLAIN:
14276 ncx->blk_format.prevcomppad =
14277 (PAD*)ptr_table_fetch(PL_ptr_table,
14278 ncx->blk_format.prevcomppad);
14279 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14280 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14281 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14285 ncx->blk_givwhen.defsv_save =
14286 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14299 /* duplicate a stack info structure */
14302 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14306 PERL_ARGS_ASSERT_SI_DUP;
14309 return (PERL_SI*)NULL;
14311 /* look for it in the table first */
14312 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14316 /* create anew and remember what it is */
14317 Newxz(nsi, 1, PERL_SI);
14318 ptr_table_store(PL_ptr_table, si, nsi);
14320 nsi->si_stack = av_dup_inc(si->si_stack, param);
14321 nsi->si_cxix = si->si_cxix;
14322 nsi->si_cxmax = si->si_cxmax;
14323 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14324 nsi->si_type = si->si_type;
14325 nsi->si_prev = si_dup(si->si_prev, param);
14326 nsi->si_next = si_dup(si->si_next, param);
14327 nsi->si_markoff = si->si_markoff;
14332 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14333 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14334 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14335 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14336 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14337 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14338 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14339 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14340 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14341 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14342 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14343 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14344 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14345 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14346 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14347 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14350 #define pv_dup_inc(p) SAVEPV(p)
14351 #define pv_dup(p) SAVEPV(p)
14352 #define svp_dup_inc(p,pp) any_dup(p,pp)
14354 /* map any object to the new equivent - either something in the
14355 * ptr table, or something in the interpreter structure
14359 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14363 PERL_ARGS_ASSERT_ANY_DUP;
14366 return (void*)NULL;
14368 /* look for it in the table first */
14369 ret = ptr_table_fetch(PL_ptr_table, v);
14373 /* see if it is part of the interpreter structure */
14374 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14375 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14383 /* duplicate the save stack */
14386 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14389 ANY * const ss = proto_perl->Isavestack;
14390 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14391 I32 ix = proto_perl->Isavestack_ix;
14404 void (*dptr) (void*);
14405 void (*dxptr) (pTHX_ void*);
14407 PERL_ARGS_ASSERT_SS_DUP;
14409 Newxz(nss, max, ANY);
14412 const UV uv = POPUV(ss,ix);
14413 const U8 type = (U8)uv & SAVE_MASK;
14415 TOPUV(nss,ix) = uv;
14417 case SAVEt_CLEARSV:
14418 case SAVEt_CLEARPADRANGE:
14420 case SAVEt_HELEM: /* hash element */
14421 case SAVEt_SV: /* scalar reference */
14422 sv = (const SV *)POPPTR(ss,ix);
14423 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14425 case SAVEt_ITEM: /* normal string */
14426 case SAVEt_GVSV: /* scalar slot in GV */
14427 sv = (const SV *)POPPTR(ss,ix);
14428 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14429 if (type == SAVEt_SV)
14433 case SAVEt_MORTALIZESV:
14434 case SAVEt_READONLY_OFF:
14435 sv = (const SV *)POPPTR(ss,ix);
14436 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14438 case SAVEt_FREEPADNAME:
14439 ptr = POPPTR(ss,ix);
14440 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14441 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14443 case SAVEt_SHARED_PVREF: /* char* in shared space */
14444 c = (char*)POPPTR(ss,ix);
14445 TOPPTR(nss,ix) = savesharedpv(c);
14446 ptr = POPPTR(ss,ix);
14447 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14449 case SAVEt_GENERIC_SVREF: /* generic sv */
14450 case SAVEt_SVREF: /* scalar reference */
14451 sv = (const SV *)POPPTR(ss,ix);
14452 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14453 if (type == SAVEt_SVREF)
14454 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14455 ptr = POPPTR(ss,ix);
14456 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14458 case SAVEt_GVSLOT: /* any slot in GV */
14459 sv = (const SV *)POPPTR(ss,ix);
14460 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14461 ptr = POPPTR(ss,ix);
14462 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14463 sv = (const SV *)POPPTR(ss,ix);
14464 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14466 case SAVEt_HV: /* hash reference */
14467 case SAVEt_AV: /* array reference */
14468 sv = (const SV *) POPPTR(ss,ix);
14469 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14471 case SAVEt_COMPPAD:
14473 sv = (const SV *) POPPTR(ss,ix);
14474 TOPPTR(nss,ix) = sv_dup(sv, param);
14476 case SAVEt_INT: /* int reference */
14477 ptr = POPPTR(ss,ix);
14478 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14479 intval = (int)POPINT(ss,ix);
14480 TOPINT(nss,ix) = intval;
14482 case SAVEt_LONG: /* long reference */
14483 ptr = POPPTR(ss,ix);
14484 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14485 longval = (long)POPLONG(ss,ix);
14486 TOPLONG(nss,ix) = longval;
14488 case SAVEt_I32: /* I32 reference */
14489 ptr = POPPTR(ss,ix);
14490 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14492 TOPINT(nss,ix) = i;
14494 case SAVEt_IV: /* IV reference */
14495 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14496 ptr = POPPTR(ss,ix);
14497 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14499 TOPIV(nss,ix) = iv;
14501 case SAVEt_TMPSFLOOR:
14503 TOPIV(nss,ix) = iv;
14505 case SAVEt_HPTR: /* HV* reference */
14506 case SAVEt_APTR: /* AV* reference */
14507 case SAVEt_SPTR: /* SV* reference */
14508 ptr = POPPTR(ss,ix);
14509 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14510 sv = (const SV *)POPPTR(ss,ix);
14511 TOPPTR(nss,ix) = sv_dup(sv, param);
14513 case SAVEt_VPTR: /* random* reference */
14514 ptr = POPPTR(ss,ix);
14515 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14517 case SAVEt_INT_SMALL:
14518 case SAVEt_I32_SMALL:
14519 case SAVEt_I16: /* I16 reference */
14520 case SAVEt_I8: /* I8 reference */
14522 ptr = POPPTR(ss,ix);
14523 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14525 case SAVEt_GENERIC_PVREF: /* generic char* */
14526 case SAVEt_PPTR: /* char* reference */
14527 ptr = POPPTR(ss,ix);
14528 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14529 c = (char*)POPPTR(ss,ix);
14530 TOPPTR(nss,ix) = pv_dup(c);
14532 case SAVEt_GP: /* scalar reference */
14533 gp = (GP*)POPPTR(ss,ix);
14534 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14535 (void)GpREFCNT_inc(gp);
14536 gv = (const GV *)POPPTR(ss,ix);
14537 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14540 ptr = POPPTR(ss,ix);
14541 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14542 /* these are assumed to be refcounted properly */
14544 switch (((OP*)ptr)->op_type) {
14546 case OP_LEAVESUBLV:
14550 case OP_LEAVEWRITE:
14551 TOPPTR(nss,ix) = ptr;
14554 (void) OpREFCNT_inc(o);
14558 TOPPTR(nss,ix) = NULL;
14563 TOPPTR(nss,ix) = NULL;
14565 case SAVEt_FREECOPHH:
14566 ptr = POPPTR(ss,ix);
14567 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14569 case SAVEt_ADELETE:
14570 av = (const AV *)POPPTR(ss,ix);
14571 TOPPTR(nss,ix) = av_dup_inc(av, param);
14573 TOPINT(nss,ix) = i;
14576 hv = (const HV *)POPPTR(ss,ix);
14577 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14579 TOPINT(nss,ix) = i;
14582 c = (char*)POPPTR(ss,ix);
14583 TOPPTR(nss,ix) = pv_dup_inc(c);
14585 case SAVEt_STACK_POS: /* Position on Perl stack */
14587 TOPINT(nss,ix) = i;
14589 case SAVEt_DESTRUCTOR:
14590 ptr = POPPTR(ss,ix);
14591 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14592 dptr = POPDPTR(ss,ix);
14593 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14594 any_dup(FPTR2DPTR(void *, dptr),
14597 case SAVEt_DESTRUCTOR_X:
14598 ptr = POPPTR(ss,ix);
14599 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14600 dxptr = POPDXPTR(ss,ix);
14601 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14602 any_dup(FPTR2DPTR(void *, dxptr),
14605 case SAVEt_REGCONTEXT:
14607 ix -= uv >> SAVE_TIGHT_SHIFT;
14609 case SAVEt_AELEM: /* array element */
14610 sv = (const SV *)POPPTR(ss,ix);
14611 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14613 TOPINT(nss,ix) = i;
14614 av = (const AV *)POPPTR(ss,ix);
14615 TOPPTR(nss,ix) = av_dup_inc(av, param);
14618 ptr = POPPTR(ss,ix);
14619 TOPPTR(nss,ix) = ptr;
14622 ptr = POPPTR(ss,ix);
14623 ptr = cophh_copy((COPHH*)ptr);
14624 TOPPTR(nss,ix) = ptr;
14626 TOPINT(nss,ix) = i;
14627 if (i & HINT_LOCALIZE_HH) {
14628 hv = (const HV *)POPPTR(ss,ix);
14629 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14632 case SAVEt_PADSV_AND_MORTALIZE:
14633 longval = (long)POPLONG(ss,ix);
14634 TOPLONG(nss,ix) = longval;
14635 ptr = POPPTR(ss,ix);
14636 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14637 sv = (const SV *)POPPTR(ss,ix);
14638 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14640 case SAVEt_SET_SVFLAGS:
14642 TOPINT(nss,ix) = i;
14644 TOPINT(nss,ix) = i;
14645 sv = (const SV *)POPPTR(ss,ix);
14646 TOPPTR(nss,ix) = sv_dup(sv, param);
14648 case SAVEt_COMPILE_WARNINGS:
14649 ptr = POPPTR(ss,ix);
14650 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14653 ptr = POPPTR(ss,ix);
14654 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14658 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14666 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14667 * flag to the result. This is done for each stash before cloning starts,
14668 * so we know which stashes want their objects cloned */
14671 do_mark_cloneable_stash(pTHX_ SV *const sv)
14673 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14675 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14676 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14677 if (cloner && GvCV(cloner)) {
14684 mXPUSHs(newSVhek(hvname));
14686 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14693 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14701 =for apidoc perl_clone
14703 Create and return a new interpreter by cloning the current one.
14705 C<perl_clone> takes these flags as parameters:
14707 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14708 without it we only clone the data and zero the stacks,
14709 with it we copy the stacks and the new perl interpreter is
14710 ready to run at the exact same point as the previous one.
14711 The pseudo-fork code uses C<COPY_STACKS> while the
14712 threads->create doesn't.
14714 C<CLONEf_KEEP_PTR_TABLE> -
14715 C<perl_clone> keeps a ptr_table with the pointer of the old
14716 variable as a key and the new variable as a value,
14717 this allows it to check if something has been cloned and not
14718 clone it again but rather just use the value and increase the
14719 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14720 the ptr_table using the function
14721 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14722 reason to keep it around is if you want to dup some of your own
14723 variable who are outside the graph perl scans, an example of this
14724 code is in F<threads.xs> create.
14726 C<CLONEf_CLONE_HOST> -
14727 This is a win32 thing, it is ignored on unix, it tells perls
14728 win32host code (which is c++) to clone itself, this is needed on
14729 win32 if you want to run two threads at the same time,
14730 if you just want to do some stuff in a separate perl interpreter
14731 and then throw it away and return to the original one,
14732 you don't need to do anything.
14737 /* XXX the above needs expanding by someone who actually understands it ! */
14738 EXTERN_C PerlInterpreter *
14739 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14742 perl_clone(PerlInterpreter *proto_perl, UV flags)
14745 #ifdef PERL_IMPLICIT_SYS
14747 PERL_ARGS_ASSERT_PERL_CLONE;
14749 /* perlhost.h so we need to call into it
14750 to clone the host, CPerlHost should have a c interface, sky */
14752 #ifndef __amigaos4__
14753 if (flags & CLONEf_CLONE_HOST) {
14754 return perl_clone_host(proto_perl,flags);
14757 return perl_clone_using(proto_perl, flags,
14759 proto_perl->IMemShared,
14760 proto_perl->IMemParse,
14762 proto_perl->IStdIO,
14766 proto_perl->IProc);
14770 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14771 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14772 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14773 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14774 struct IPerlDir* ipD, struct IPerlSock* ipS,
14775 struct IPerlProc* ipP)
14777 /* XXX many of the string copies here can be optimized if they're
14778 * constants; they need to be allocated as common memory and just
14779 * their pointers copied. */
14782 CLONE_PARAMS clone_params;
14783 CLONE_PARAMS* const param = &clone_params;
14785 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14787 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14788 #else /* !PERL_IMPLICIT_SYS */
14790 CLONE_PARAMS clone_params;
14791 CLONE_PARAMS* param = &clone_params;
14792 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14794 PERL_ARGS_ASSERT_PERL_CLONE;
14795 #endif /* PERL_IMPLICIT_SYS */
14797 /* for each stash, determine whether its objects should be cloned */
14798 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14799 PERL_SET_THX(my_perl);
14802 PoisonNew(my_perl, 1, PerlInterpreter);
14805 PL_defstash = NULL; /* may be used by perl malloc() */
14808 PL_scopestack_name = 0;
14810 PL_savestack_ix = 0;
14811 PL_savestack_max = -1;
14812 PL_sig_pending = 0;
14814 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14815 Zero(&PL_padname_undef, 1, PADNAME);
14816 Zero(&PL_padname_const, 1, PADNAME);
14817 # ifdef DEBUG_LEAKING_SCALARS
14818 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14820 # ifdef PERL_TRACE_OPS
14821 Zero(PL_op_exec_cnt, OP_max+2, UV);
14823 #else /* !DEBUGGING */
14824 Zero(my_perl, 1, PerlInterpreter);
14825 #endif /* DEBUGGING */
14827 #ifdef PERL_IMPLICIT_SYS
14828 /* host pointers */
14830 PL_MemShared = ipMS;
14831 PL_MemParse = ipMP;
14838 #endif /* PERL_IMPLICIT_SYS */
14841 param->flags = flags;
14842 /* Nothing in the core code uses this, but we make it available to
14843 extensions (using mg_dup). */
14844 param->proto_perl = proto_perl;
14845 /* Likely nothing will use this, but it is initialised to be consistent
14846 with Perl_clone_params_new(). */
14847 param->new_perl = my_perl;
14848 param->unreferenced = NULL;
14851 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14853 PL_body_arenas = NULL;
14854 Zero(&PL_body_roots, 1, PL_body_roots);
14858 PL_sv_arenaroot = NULL;
14860 PL_debug = proto_perl->Idebug;
14862 /* dbargs array probably holds garbage */
14865 PL_compiling = proto_perl->Icompiling;
14867 /* pseudo environmental stuff */
14868 PL_origargc = proto_perl->Iorigargc;
14869 PL_origargv = proto_perl->Iorigargv;
14871 #ifndef NO_TAINT_SUPPORT
14872 /* Set tainting stuff before PerlIO_debug can possibly get called */
14873 PL_tainting = proto_perl->Itainting;
14874 PL_taint_warn = proto_perl->Itaint_warn;
14876 PL_tainting = FALSE;
14877 PL_taint_warn = FALSE;
14880 PL_minus_c = proto_perl->Iminus_c;
14882 PL_localpatches = proto_perl->Ilocalpatches;
14883 PL_splitstr = proto_perl->Isplitstr;
14884 PL_minus_n = proto_perl->Iminus_n;
14885 PL_minus_p = proto_perl->Iminus_p;
14886 PL_minus_l = proto_perl->Iminus_l;
14887 PL_minus_a = proto_perl->Iminus_a;
14888 PL_minus_E = proto_perl->Iminus_E;
14889 PL_minus_F = proto_perl->Iminus_F;
14890 PL_doswitches = proto_perl->Idoswitches;
14891 PL_dowarn = proto_perl->Idowarn;
14892 #ifdef PERL_SAWAMPERSAND
14893 PL_sawampersand = proto_perl->Isawampersand;
14895 PL_unsafe = proto_perl->Iunsafe;
14896 PL_perldb = proto_perl->Iperldb;
14897 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14898 PL_exit_flags = proto_perl->Iexit_flags;
14900 /* XXX time(&PL_basetime) when asked for? */
14901 PL_basetime = proto_perl->Ibasetime;
14903 PL_maxsysfd = proto_perl->Imaxsysfd;
14904 PL_statusvalue = proto_perl->Istatusvalue;
14906 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14908 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14911 /* RE engine related */
14912 PL_regmatch_slab = NULL;
14913 PL_reg_curpm = NULL;
14915 PL_sub_generation = proto_perl->Isub_generation;
14917 /* funky return mechanisms */
14918 PL_forkprocess = proto_perl->Iforkprocess;
14920 /* internal state */
14921 PL_main_start = proto_perl->Imain_start;
14922 PL_eval_root = proto_perl->Ieval_root;
14923 PL_eval_start = proto_perl->Ieval_start;
14925 PL_filemode = proto_perl->Ifilemode;
14926 PL_lastfd = proto_perl->Ilastfd;
14927 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14930 PL_gensym = proto_perl->Igensym;
14932 PL_laststatval = proto_perl->Ilaststatval;
14933 PL_laststype = proto_perl->Ilaststype;
14936 PL_profiledata = NULL;
14938 PL_generation = proto_perl->Igeneration;
14940 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14941 PL_in_clean_all = proto_perl->Iin_clean_all;
14943 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14944 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14945 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14946 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14947 PL_nomemok = proto_perl->Inomemok;
14948 PL_an = proto_perl->Ian;
14949 PL_evalseq = proto_perl->Ievalseq;
14950 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14951 PL_origalen = proto_perl->Iorigalen;
14953 PL_sighandlerp = proto_perl->Isighandlerp;
14955 PL_runops = proto_perl->Irunops;
14957 PL_subline = proto_perl->Isubline;
14959 PL_cv_has_eval = proto_perl->Icv_has_eval;
14962 PL_cryptseen = proto_perl->Icryptseen;
14965 #ifdef USE_LOCALE_COLLATE
14966 PL_collation_ix = proto_perl->Icollation_ix;
14967 PL_collation_standard = proto_perl->Icollation_standard;
14968 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14969 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14970 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14971 #endif /* USE_LOCALE_COLLATE */
14973 #ifdef USE_LOCALE_NUMERIC
14974 PL_numeric_standard = proto_perl->Inumeric_standard;
14975 PL_numeric_local = proto_perl->Inumeric_local;
14976 #endif /* !USE_LOCALE_NUMERIC */
14978 /* Did the locale setup indicate UTF-8? */
14979 PL_utf8locale = proto_perl->Iutf8locale;
14980 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14981 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
14982 /* Unicode features (see perlrun/-C) */
14983 PL_unicode = proto_perl->Iunicode;
14985 /* Pre-5.8 signals control */
14986 PL_signals = proto_perl->Isignals;
14988 /* times() ticks per second */
14989 PL_clocktick = proto_perl->Iclocktick;
14991 /* Recursion stopper for PerlIO_find_layer */
14992 PL_in_load_module = proto_perl->Iin_load_module;
14994 /* sort() routine */
14995 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14997 /* Not really needed/useful since the reenrant_retint is "volatile",
14998 * but do it for consistency's sake. */
14999 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15001 /* Hooks to shared SVs and locks. */
15002 PL_sharehook = proto_perl->Isharehook;
15003 PL_lockhook = proto_perl->Ilockhook;
15004 PL_unlockhook = proto_perl->Iunlockhook;
15005 PL_threadhook = proto_perl->Ithreadhook;
15006 PL_destroyhook = proto_perl->Idestroyhook;
15007 PL_signalhook = proto_perl->Isignalhook;
15009 PL_globhook = proto_perl->Iglobhook;
15012 PL_last_swash_hv = NULL; /* reinits on demand */
15013 PL_last_swash_klen = 0;
15014 PL_last_swash_key[0]= '\0';
15015 PL_last_swash_tmps = (U8*)NULL;
15016 PL_last_swash_slen = 0;
15018 PL_srand_called = proto_perl->Isrand_called;
15019 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15021 if (flags & CLONEf_COPY_STACKS) {
15022 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15023 PL_tmps_ix = proto_perl->Itmps_ix;
15024 PL_tmps_max = proto_perl->Itmps_max;
15025 PL_tmps_floor = proto_perl->Itmps_floor;
15027 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15028 * NOTE: unlike the others! */
15029 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15030 PL_scopestack_max = proto_perl->Iscopestack_max;
15032 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15033 * NOTE: unlike the others! */
15034 PL_savestack_ix = proto_perl->Isavestack_ix;
15035 PL_savestack_max = proto_perl->Isavestack_max;
15038 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15039 PL_top_env = &PL_start_env;
15041 PL_op = proto_perl->Iop;
15044 PL_Xpv = (XPV*)NULL;
15045 my_perl->Ina = proto_perl->Ina;
15047 PL_statbuf = proto_perl->Istatbuf;
15048 PL_statcache = proto_perl->Istatcache;
15050 #ifndef NO_TAINT_SUPPORT
15051 PL_tainted = proto_perl->Itainted;
15053 PL_tainted = FALSE;
15055 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15057 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15059 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15060 PL_restartop = proto_perl->Irestartop;
15061 PL_in_eval = proto_perl->Iin_eval;
15062 PL_delaymagic = proto_perl->Idelaymagic;
15063 PL_phase = proto_perl->Iphase;
15064 PL_localizing = proto_perl->Ilocalizing;
15066 PL_hv_fetch_ent_mh = NULL;
15067 PL_modcount = proto_perl->Imodcount;
15068 PL_lastgotoprobe = NULL;
15069 PL_dumpindent = proto_perl->Idumpindent;
15071 PL_efloatbuf = NULL; /* reinits on demand */
15072 PL_efloatsize = 0; /* reinits on demand */
15076 PL_colorset = 0; /* reinits PL_colors[] */
15077 /*PL_colors[6] = {0,0,0,0,0,0};*/
15079 /* Pluggable optimizer */
15080 PL_peepp = proto_perl->Ipeepp;
15081 PL_rpeepp = proto_perl->Irpeepp;
15082 /* op_free() hook */
15083 PL_opfreehook = proto_perl->Iopfreehook;
15085 #ifdef USE_REENTRANT_API
15086 /* XXX: things like -Dm will segfault here in perlio, but doing
15087 * PERL_SET_CONTEXT(proto_perl);
15088 * breaks too many other things
15090 Perl_reentrant_init(aTHX);
15093 /* create SV map for pointer relocation */
15094 PL_ptr_table = ptr_table_new();
15096 /* initialize these special pointers as early as possible */
15098 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15099 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15100 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15101 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15102 &PL_padname_const);
15104 /* create (a non-shared!) shared string table */
15105 PL_strtab = newHV();
15106 HvSHAREKEYS_off(PL_strtab);
15107 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15108 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15110 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15112 /* This PV will be free'd special way so must set it same way op.c does */
15113 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15114 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15116 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15117 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15118 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15119 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15121 param->stashes = newAV(); /* Setup array of objects to call clone on */
15122 /* This makes no difference to the implementation, as it always pushes
15123 and shifts pointers to other SVs without changing their reference
15124 count, with the array becoming empty before it is freed. However, it
15125 makes it conceptually clear what is going on, and will avoid some
15126 work inside av.c, filling slots between AvFILL() and AvMAX() with
15127 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15128 AvREAL_off(param->stashes);
15130 if (!(flags & CLONEf_COPY_STACKS)) {
15131 param->unreferenced = newAV();
15134 #ifdef PERLIO_LAYERS
15135 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15136 PerlIO_clone(aTHX_ proto_perl, param);
15139 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15140 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15141 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15142 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15143 PL_xsubfilename = proto_perl->Ixsubfilename;
15144 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15145 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15148 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15149 PL_inplace = SAVEPV(proto_perl->Iinplace);
15150 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15152 /* magical thingies */
15154 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15155 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15156 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15159 /* Clone the regex array */
15160 /* ORANGE FIXME for plugins, probably in the SV dup code.
15161 newSViv(PTR2IV(CALLREGDUPE(
15162 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15164 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15165 PL_regex_pad = AvARRAY(PL_regex_padav);
15167 PL_stashpadmax = proto_perl->Istashpadmax;
15168 PL_stashpadix = proto_perl->Istashpadix ;
15169 Newx(PL_stashpad, PL_stashpadmax, HV *);
15172 for (; o < PL_stashpadmax; ++o)
15173 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15176 /* shortcuts to various I/O objects */
15177 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15178 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15179 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15180 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15181 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15182 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15183 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15185 /* shortcuts to regexp stuff */
15186 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15188 /* shortcuts to misc objects */
15189 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15191 /* shortcuts to debugging objects */
15192 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15193 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15194 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15195 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15196 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15197 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15198 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15200 /* symbol tables */
15201 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15202 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15203 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15204 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15205 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15207 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15208 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15209 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15210 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15211 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15212 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15213 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15214 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15215 PL_savebegin = proto_perl->Isavebegin;
15217 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15219 /* subprocess state */
15220 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15222 if (proto_perl->Iop_mask)
15223 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15226 /* PL_asserting = proto_perl->Iasserting; */
15228 /* current interpreter roots */
15229 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15231 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15234 /* runtime control stuff */
15235 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15237 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15239 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15241 /* interpreter atexit processing */
15242 PL_exitlistlen = proto_perl->Iexitlistlen;
15243 if (PL_exitlistlen) {
15244 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15245 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15248 PL_exitlist = (PerlExitListEntry*)NULL;
15250 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15251 if (PL_my_cxt_size) {
15252 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15253 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15254 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15255 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15256 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15260 PL_my_cxt_list = (void**)NULL;
15261 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15262 PL_my_cxt_keys = (const char**)NULL;
15265 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15266 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15267 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15268 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15270 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15272 PAD_CLONE_VARS(proto_perl, param);
15274 #ifdef HAVE_INTERP_INTERN
15275 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15278 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15280 #ifdef PERL_USES_PL_PIDSTATUS
15281 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15283 PL_osname = SAVEPV(proto_perl->Iosname);
15284 PL_parser = parser_dup(proto_perl->Iparser, param);
15286 /* XXX this only works if the saved cop has already been cloned */
15287 if (proto_perl->Iparser) {
15288 PL_parser->saved_curcop = (COP*)any_dup(
15289 proto_perl->Iparser->saved_curcop,
15293 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15295 #ifdef USE_LOCALE_CTYPE
15296 /* Should we warn if uses locale? */
15297 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15300 #ifdef USE_LOCALE_COLLATE
15301 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15302 #endif /* USE_LOCALE_COLLATE */
15304 #ifdef USE_LOCALE_NUMERIC
15305 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15306 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15307 #endif /* !USE_LOCALE_NUMERIC */
15309 /* Unicode inversion lists */
15310 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15311 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15312 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15313 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15315 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15316 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15318 /* utf8 character class swashes */
15319 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15320 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15322 for (i = 0; i < POSIX_CC_COUNT; i++) {
15323 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15325 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15326 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15327 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15328 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15329 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15330 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15331 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15332 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15333 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15334 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15335 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15336 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15337 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15338 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15339 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15340 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15341 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15343 if (proto_perl->Ipsig_pend) {
15344 Newxz(PL_psig_pend, SIG_SIZE, int);
15347 PL_psig_pend = (int*)NULL;
15350 if (proto_perl->Ipsig_name) {
15351 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15352 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15354 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15357 PL_psig_ptr = (SV**)NULL;
15358 PL_psig_name = (SV**)NULL;
15361 if (flags & CLONEf_COPY_STACKS) {
15362 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15363 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15364 PL_tmps_ix+1, param);
15366 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15367 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15368 Newxz(PL_markstack, i, I32);
15369 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15370 - proto_perl->Imarkstack);
15371 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15372 - proto_perl->Imarkstack);
15373 Copy(proto_perl->Imarkstack, PL_markstack,
15374 PL_markstack_ptr - PL_markstack + 1, I32);
15376 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15377 * NOTE: unlike the others! */
15378 Newxz(PL_scopestack, PL_scopestack_max, I32);
15379 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15382 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15383 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15385 /* reset stack AV to correct length before its duped via
15386 * PL_curstackinfo */
15387 AvFILLp(proto_perl->Icurstack) =
15388 proto_perl->Istack_sp - proto_perl->Istack_base;
15390 /* NOTE: si_dup() looks at PL_markstack */
15391 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15393 /* PL_curstack = PL_curstackinfo->si_stack; */
15394 PL_curstack = av_dup(proto_perl->Icurstack, param);
15395 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15397 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15398 PL_stack_base = AvARRAY(PL_curstack);
15399 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15400 - proto_perl->Istack_base);
15401 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15403 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15404 PL_savestack = ss_dup(proto_perl, param);
15408 ENTER; /* perl_destruct() wants to LEAVE; */
15411 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15412 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15414 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15415 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15416 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15417 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15418 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15419 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15421 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15423 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15424 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15425 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15427 PL_stashcache = newHV();
15429 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15430 proto_perl->Iwatchaddr);
15431 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15432 if (PL_debug && PL_watchaddr) {
15433 PerlIO_printf(Perl_debug_log,
15434 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15435 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15436 PTR2UV(PL_watchok));
15439 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15440 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15441 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15443 /* Call the ->CLONE method, if it exists, for each of the stashes
15444 identified by sv_dup() above.
15446 while(av_tindex(param->stashes) != -1) {
15447 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15448 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15449 if (cloner && GvCV(cloner)) {
15454 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15456 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15462 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15463 ptr_table_free(PL_ptr_table);
15464 PL_ptr_table = NULL;
15467 if (!(flags & CLONEf_COPY_STACKS)) {
15468 unreferenced_to_tmp_stack(param->unreferenced);
15471 SvREFCNT_dec(param->stashes);
15473 /* orphaned? eg threads->new inside BEGIN or use */
15474 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15475 SvREFCNT_inc_simple_void(PL_compcv);
15476 SAVEFREESV(PL_compcv);
15483 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15485 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15487 if (AvFILLp(unreferenced) > -1) {
15488 SV **svp = AvARRAY(unreferenced);
15489 SV **const last = svp + AvFILLp(unreferenced);
15493 if (SvREFCNT(*svp) == 1)
15495 } while (++svp <= last);
15497 EXTEND_MORTAL(count);
15498 svp = AvARRAY(unreferenced);
15501 if (SvREFCNT(*svp) == 1) {
15502 /* Our reference is the only one to this SV. This means that
15503 in this thread, the scalar effectively has a 0 reference.
15504 That doesn't work (cleanup never happens), so donate our
15505 reference to it onto the save stack. */
15506 PL_tmps_stack[++PL_tmps_ix] = *svp;
15508 /* As an optimisation, because we are already walking the
15509 entire array, instead of above doing either
15510 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15511 release our reference to the scalar, so that at the end of
15512 the array owns zero references to the scalars it happens to
15513 point to. We are effectively converting the array from
15514 AvREAL() on to AvREAL() off. This saves the av_clear()
15515 (triggered by the SvREFCNT_dec(unreferenced) below) from
15516 walking the array a second time. */
15517 SvREFCNT_dec(*svp);
15520 } while (++svp <= last);
15521 AvREAL_off(unreferenced);
15523 SvREFCNT_dec_NN(unreferenced);
15527 Perl_clone_params_del(CLONE_PARAMS *param)
15529 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15531 PerlInterpreter *const to = param->new_perl;
15533 PerlInterpreter *const was = PERL_GET_THX;
15535 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15541 SvREFCNT_dec(param->stashes);
15542 if (param->unreferenced)
15543 unreferenced_to_tmp_stack(param->unreferenced);
15553 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15556 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15557 does a dTHX; to get the context from thread local storage.
15558 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15559 a version that passes in my_perl. */
15560 PerlInterpreter *const was = PERL_GET_THX;
15561 CLONE_PARAMS *param;
15563 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15569 /* Given that we've set the context, we can do this unshared. */
15570 Newx(param, 1, CLONE_PARAMS);
15573 param->proto_perl = from;
15574 param->new_perl = to;
15575 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15576 AvREAL_off(param->stashes);
15577 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15585 #endif /* USE_ITHREADS */
15588 Perl_init_constants(pTHX)
15590 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15591 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15592 SvANY(&PL_sv_undef) = NULL;
15594 SvANY(&PL_sv_no) = new_XPVNV();
15595 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15596 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15597 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15600 SvANY(&PL_sv_yes) = new_XPVNV();
15601 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15602 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15603 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15606 SvPV_set(&PL_sv_no, (char*)PL_No);
15607 SvCUR_set(&PL_sv_no, 0);
15608 SvLEN_set(&PL_sv_no, 0);
15609 SvIV_set(&PL_sv_no, 0);
15610 SvNV_set(&PL_sv_no, 0);
15612 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15613 SvCUR_set(&PL_sv_yes, 1);
15614 SvLEN_set(&PL_sv_yes, 0);
15615 SvIV_set(&PL_sv_yes, 1);
15616 SvNV_set(&PL_sv_yes, 1);
15618 PadnamePV(&PL_padname_const) = (char *)PL_No;
15622 =head1 Unicode Support
15624 =for apidoc sv_recode_to_utf8
15626 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15627 of C<sv> is assumed to be octets in that encoding, and C<sv>
15628 will be converted into Unicode (and UTF-8).
15630 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15631 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15632 an C<Encode::XS> Encoding object, bad things will happen.
15633 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15635 The PV of C<sv> is returned.
15640 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15642 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15644 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15653 if (SvPADTMP(nsv)) {
15654 nsv = sv_newmortal();
15655 SvSetSV_nosteal(nsv, sv);
15664 Passing sv_yes is wrong - it needs to be or'ed set of constants
15665 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15666 remove converted chars from source.
15668 Both will default the value - let them.
15670 XPUSHs(&PL_sv_yes);
15673 call_method("decode", G_SCALAR);
15677 s = SvPV_const(uni, len);
15678 if (s != SvPVX_const(sv)) {
15679 SvGROW(sv, len + 1);
15680 Move(s, SvPVX(sv), len + 1, char);
15681 SvCUR_set(sv, len);
15686 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15687 /* clear pos and any utf8 cache */
15688 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15691 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15692 magic_setutf8(sv,mg); /* clear UTF8 cache */
15697 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15701 =for apidoc sv_cat_decode
15703 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15704 assumed to be octets in that encoding and decoding the input starts
15705 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15706 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15707 when the string C<tstr> appears in decoding output or the input ends on
15708 the PV of C<ssv>. The value which C<offset> points will be modified
15709 to the last input position on C<ssv>.
15711 Returns TRUE if the terminator was found, else returns FALSE.
15716 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15717 SV *ssv, int *offset, char *tstr, int tlen)
15721 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15723 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15734 offsv = newSViv(*offset);
15736 mPUSHp(tstr, tlen);
15738 call_method("cat_decode", G_SCALAR);
15740 ret = SvTRUE(TOPs);
15741 *offset = SvIV(offsv);
15747 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15752 /* ---------------------------------------------------------------------
15754 * support functions for report_uninit()
15757 /* the maxiumum size of array or hash where we will scan looking
15758 * for the undefined element that triggered the warning */
15760 #define FUV_MAX_SEARCH_SIZE 1000
15762 /* Look for an entry in the hash whose value has the same SV as val;
15763 * If so, return a mortal copy of the key. */
15766 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15772 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15774 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15775 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15778 array = HvARRAY(hv);
15780 for (i=HvMAX(hv); i>=0; i--) {
15782 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15783 if (HeVAL(entry) != val)
15785 if ( HeVAL(entry) == &PL_sv_undef ||
15786 HeVAL(entry) == &PL_sv_placeholder)
15790 if (HeKLEN(entry) == HEf_SVKEY)
15791 return sv_mortalcopy(HeKEY_sv(entry));
15792 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15798 /* Look for an entry in the array whose value has the same SV as val;
15799 * If so, return the index, otherwise return -1. */
15802 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15804 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15806 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15807 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15810 if (val != &PL_sv_undef) {
15811 SV ** const svp = AvARRAY(av);
15814 for (i=AvFILLp(av); i>=0; i--)
15821 /* varname(): return the name of a variable, optionally with a subscript.
15822 * If gv is non-zero, use the name of that global, along with gvtype (one
15823 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15824 * targ. Depending on the value of the subscript_type flag, return:
15827 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15828 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15829 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15830 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15833 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15834 const SV *const keyname, SSize_t aindex, int subscript_type)
15837 SV * const name = sv_newmortal();
15838 if (gv && isGV(gv)) {
15840 buffer[0] = gvtype;
15843 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15845 gv_fullname4(name, gv, buffer, 0);
15847 if ((unsigned int)SvPVX(name)[1] <= 26) {
15849 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15851 /* Swap the 1 unprintable control character for the 2 byte pretty
15852 version - ie substr($name, 1, 1) = $buffer; */
15853 sv_insert(name, 1, 1, buffer, 2);
15857 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15860 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15862 if (!cv || !CvPADLIST(cv))
15864 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15865 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15869 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15870 SV * const sv = newSV(0);
15872 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15874 *SvPVX(name) = '$';
15875 Perl_sv_catpvf(aTHX_ name, "{%s}",
15876 pv_pretty(sv, pv, len, 32, NULL, NULL,
15877 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15878 SvREFCNT_dec_NN(sv);
15880 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15881 *SvPVX(name) = '$';
15882 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
15884 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15885 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15886 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15894 =for apidoc find_uninit_var
15896 Find the name of the undefined variable (if any) that caused the operator
15897 to issue a "Use of uninitialized value" warning.
15898 If match is true, only return a name if its value matches C<uninit_sv>.
15899 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15900 warning, then following the direct child of the op may yield an
15901 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15902 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15903 the variable name if we get an exact match.
15904 C<desc_p> points to a string pointer holding the description of the op.
15905 This may be updated if needed.
15907 The name is returned as a mortal SV.
15909 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15910 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15916 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15917 bool match, const char **desc_p)
15922 const OP *o, *o2, *kid;
15924 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15926 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15927 uninit_sv == &PL_sv_placeholder)))
15930 switch (obase->op_type) {
15933 /* undef should care if its args are undef - any warnings
15934 * will be from tied/magic vars */
15942 const bool pad = ( obase->op_type == OP_PADAV
15943 || obase->op_type == OP_PADHV
15944 || obase->op_type == OP_PADRANGE
15947 const bool hash = ( obase->op_type == OP_PADHV
15948 || obase->op_type == OP_RV2HV
15949 || (obase->op_type == OP_PADRANGE
15950 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15954 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15956 if (pad) { /* @lex, %lex */
15957 sv = PAD_SVl(obase->op_targ);
15961 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15962 /* @global, %global */
15963 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15966 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15968 else if (obase == PL_op) /* @{expr}, %{expr} */
15969 return find_uninit_var(cUNOPx(obase)->op_first,
15970 uninit_sv, match, desc_p);
15971 else /* @{expr}, %{expr} as a sub-expression */
15975 /* attempt to find a match within the aggregate */
15977 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15979 subscript_type = FUV_SUBSCRIPT_HASH;
15982 index = find_array_subscript((const AV *)sv, uninit_sv);
15984 subscript_type = FUV_SUBSCRIPT_ARRAY;
15987 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15990 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15991 keysv, index, subscript_type);
15995 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15997 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15998 if (!gv || !GvSTASH(gv))
16000 if (match && (GvSV(gv) != uninit_sv))
16002 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16005 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16008 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16010 return varname(NULL, '$', obase->op_targ,
16011 NULL, 0, FUV_SUBSCRIPT_NONE);
16014 gv = cGVOPx_gv(obase);
16015 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16017 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16019 case OP_AELEMFAST_LEX:
16022 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16023 if (!av || SvRMAGICAL(av))
16025 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16026 if (!svp || *svp != uninit_sv)
16029 return varname(NULL, '$', obase->op_targ,
16030 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16033 gv = cGVOPx_gv(obase);
16038 AV *const av = GvAV(gv);
16039 if (!av || SvRMAGICAL(av))
16041 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16042 if (!svp || *svp != uninit_sv)
16045 return varname(gv, '$', 0,
16046 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16048 NOT_REACHED; /* NOTREACHED */
16051 o = cUNOPx(obase)->op_first;
16052 if (!o || o->op_type != OP_NULL ||
16053 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16055 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16060 bool negate = FALSE;
16062 if (PL_op == obase)
16063 /* $a[uninit_expr] or $h{uninit_expr} */
16064 return find_uninit_var(cBINOPx(obase)->op_last,
16065 uninit_sv, match, desc_p);
16068 o = cBINOPx(obase)->op_first;
16069 kid = cBINOPx(obase)->op_last;
16071 /* get the av or hv, and optionally the gv */
16073 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16074 sv = PAD_SV(o->op_targ);
16076 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16077 && cUNOPo->op_first->op_type == OP_GV)
16079 gv = cGVOPx_gv(cUNOPo->op_first);
16083 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16088 if (kid && kid->op_type == OP_NEGATE) {
16090 kid = cUNOPx(kid)->op_first;
16093 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16094 /* index is constant */
16097 kidsv = newSVpvs_flags("-", SVs_TEMP);
16098 sv_catsv(kidsv, cSVOPx_sv(kid));
16101 kidsv = cSVOPx_sv(kid);
16105 if (obase->op_type == OP_HELEM) {
16106 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16107 if (!he || HeVAL(he) != uninit_sv)
16111 SV * const opsv = cSVOPx_sv(kid);
16112 const IV opsviv = SvIV(opsv);
16113 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16114 negate ? - opsviv : opsviv,
16116 if (!svp || *svp != uninit_sv)
16120 if (obase->op_type == OP_HELEM)
16121 return varname(gv, '%', o->op_targ,
16122 kidsv, 0, FUV_SUBSCRIPT_HASH);
16124 return varname(gv, '@', o->op_targ, NULL,
16125 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16126 FUV_SUBSCRIPT_ARRAY);
16129 /* index is an expression;
16130 * attempt to find a match within the aggregate */
16131 if (obase->op_type == OP_HELEM) {
16132 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16134 return varname(gv, '%', o->op_targ,
16135 keysv, 0, FUV_SUBSCRIPT_HASH);
16138 const SSize_t index
16139 = find_array_subscript((const AV *)sv, uninit_sv);
16141 return varname(gv, '@', o->op_targ,
16142 NULL, index, FUV_SUBSCRIPT_ARRAY);
16147 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16149 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16151 NOT_REACHED; /* NOTREACHED */
16154 case OP_MULTIDEREF: {
16155 /* If we were executing OP_MULTIDEREF when the undef warning
16156 * triggered, then it must be one of the index values within
16157 * that triggered it. If not, then the only possibility is that
16158 * the value retrieved by the last aggregate index might be the
16159 * culprit. For the former, we set PL_multideref_pc each time before
16160 * using an index, so work though the item list until we reach
16161 * that point. For the latter, just work through the entire item
16162 * list; the last aggregate retrieved will be the candidate.
16163 * There is a third rare possibility: something triggered
16164 * magic while fetching an array/hash element. Just display
16165 * nothing in this case.
16168 /* the named aggregate, if any */
16169 PADOFFSET agg_targ = 0;
16171 /* the last-seen index */
16173 PADOFFSET index_targ;
16175 IV index_const_iv = 0; /* init for spurious compiler warn */
16176 SV *index_const_sv;
16177 int depth = 0; /* how many array/hash lookups we've done */
16179 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16180 UNOP_AUX_item *last = NULL;
16181 UV actions = items->uv;
16184 if (PL_op == obase) {
16185 last = PL_multideref_pc;
16186 assert(last >= items && last <= items + items[-1].uv);
16193 switch (actions & MDEREF_ACTION_MASK) {
16195 case MDEREF_reload:
16196 actions = (++items)->uv;
16199 case MDEREF_HV_padhv_helem: /* $lex{...} */
16202 case MDEREF_AV_padav_aelem: /* $lex[...] */
16203 agg_targ = (++items)->pad_offset;
16207 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16210 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16212 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16213 assert(isGV_with_GP(agg_gv));
16216 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16217 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16220 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16221 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16227 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16228 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16231 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16232 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16239 index_const_sv = NULL;
16241 index_type = (actions & MDEREF_INDEX_MASK);
16242 switch (index_type) {
16243 case MDEREF_INDEX_none:
16245 case MDEREF_INDEX_const:
16247 index_const_sv = UNOP_AUX_item_sv(++items)
16249 index_const_iv = (++items)->iv;
16251 case MDEREF_INDEX_padsv:
16252 index_targ = (++items)->pad_offset;
16254 case MDEREF_INDEX_gvsv:
16255 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16256 assert(isGV_with_GP(index_gv));
16260 if (index_type != MDEREF_INDEX_none)
16263 if ( index_type == MDEREF_INDEX_none
16264 || (actions & MDEREF_FLAG_last)
16265 || (last && items >= last)
16269 actions >>= MDEREF_SHIFT;
16272 if (PL_op == obase) {
16273 /* most likely index was undef */
16275 *desc_p = ( (actions & MDEREF_FLAG_last)
16276 && (obase->op_private
16277 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16279 (obase->op_private & OPpMULTIDEREF_EXISTS)
16282 : is_hv ? "hash element" : "array element";
16283 assert(index_type != MDEREF_INDEX_none);
16285 if (GvSV(index_gv) == uninit_sv)
16286 return varname(index_gv, '$', 0, NULL, 0,
16287 FUV_SUBSCRIPT_NONE);
16292 if (PL_curpad[index_targ] == uninit_sv)
16293 return varname(NULL, '$', index_targ,
16294 NULL, 0, FUV_SUBSCRIPT_NONE);
16298 /* If we got to this point it was undef on a const subscript,
16299 * so magic probably involved, e.g. $ISA[0]. Give up. */
16303 /* the SV returned by pp_multideref() was undef, if anything was */
16309 sv = PAD_SV(agg_targ);
16311 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16315 if (index_type == MDEREF_INDEX_const) {
16320 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16321 if (!he || HeVAL(he) != uninit_sv)
16325 SV * const * const svp =
16326 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16327 if (!svp || *svp != uninit_sv)
16332 ? varname(agg_gv, '%', agg_targ,
16333 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16334 : varname(agg_gv, '@', agg_targ,
16335 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16338 /* index is an var */
16340 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16342 return varname(agg_gv, '%', agg_targ,
16343 keysv, 0, FUV_SUBSCRIPT_HASH);
16346 const SSize_t index
16347 = find_array_subscript((const AV *)sv, uninit_sv);
16349 return varname(agg_gv, '@', agg_targ,
16350 NULL, index, FUV_SUBSCRIPT_ARRAY);
16354 return varname(agg_gv,
16356 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16358 NOT_REACHED; /* NOTREACHED */
16362 /* only examine RHS */
16363 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16367 o = cUNOPx(obase)->op_first;
16368 if ( o->op_type == OP_PUSHMARK
16369 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16373 if (!OpHAS_SIBLING(o)) {
16374 /* one-arg version of open is highly magical */
16376 if (o->op_type == OP_GV) { /* open FOO; */
16378 if (match && GvSV(gv) != uninit_sv)
16380 return varname(gv, '$', 0,
16381 NULL, 0, FUV_SUBSCRIPT_NONE);
16383 /* other possibilities not handled are:
16384 * open $x; or open my $x; should return '${*$x}'
16385 * open expr; should return '$'.expr ideally
16392 /* ops where $_ may be an implicit arg */
16397 if ( !(obase->op_flags & OPf_STACKED)) {
16398 if (uninit_sv == DEFSV)
16399 return newSVpvs_flags("$_", SVs_TEMP);
16400 else if (obase->op_targ
16401 && uninit_sv == PAD_SVl(obase->op_targ))
16402 return varname(NULL, '$', obase->op_targ, NULL, 0,
16403 FUV_SUBSCRIPT_NONE);
16410 match = 1; /* print etc can return undef on defined args */
16411 /* skip filehandle as it can't produce 'undef' warning */
16412 o = cUNOPx(obase)->op_first;
16413 if ((obase->op_flags & OPf_STACKED)
16415 ( o->op_type == OP_PUSHMARK
16416 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16417 o = OpSIBLING(OpSIBLING(o));
16421 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16422 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16424 /* the following ops are capable of returning PL_sv_undef even for
16425 * defined arg(s) */
16444 case OP_GETPEERNAME:
16491 case OP_SMARTMATCH:
16500 /* XXX tmp hack: these two may call an XS sub, and currently
16501 XS subs don't have a SUB entry on the context stack, so CV and
16502 pad determination goes wrong, and BAD things happen. So, just
16503 don't try to determine the value under those circumstances.
16504 Need a better fix at dome point. DAPM 11/2007 */
16510 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16511 if (gv && GvSV(gv) == uninit_sv)
16512 return newSVpvs_flags("$.", SVs_TEMP);
16517 /* def-ness of rval pos() is independent of the def-ness of its arg */
16518 if ( !(obase->op_flags & OPf_MOD))
16523 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16524 return newSVpvs_flags("${$/}", SVs_TEMP);
16529 if (!(obase->op_flags & OPf_KIDS))
16531 o = cUNOPx(obase)->op_first;
16537 /* This loop checks all the kid ops, skipping any that cannot pos-
16538 * sibly be responsible for the uninitialized value; i.e., defined
16539 * constants and ops that return nothing. If there is only one op
16540 * left that is not skipped, then we *know* it is responsible for
16541 * the uninitialized value. If there is more than one op left, we
16542 * have to look for an exact match in the while() loop below.
16543 * Note that we skip padrange, because the individual pad ops that
16544 * it replaced are still in the tree, so we work on them instead.
16547 for (kid=o; kid; kid = OpSIBLING(kid)) {
16548 const OPCODE type = kid->op_type;
16549 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16550 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16551 || (type == OP_PUSHMARK)
16552 || (type == OP_PADRANGE)
16556 if (o2) { /* more than one found */
16563 return find_uninit_var(o2, uninit_sv, match, desc_p);
16565 /* scan all args */
16567 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16579 =for apidoc report_uninit
16581 Print appropriate "Use of uninitialized variable" warning.
16587 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16589 const char *desc = NULL;
16590 SV* varname = NULL;
16593 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16596 if (uninit_sv && PL_curpad) {
16597 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16599 sv_insert(varname, 0, 0, " ", 1);
16602 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16603 /* we've reached the end of a sort block or sub,
16604 * and the uninit value is probably what that code returned */
16607 /* PL_warn_uninit_sv is constant */
16608 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16610 /* diag_listed_as: Use of uninitialized value%s */
16611 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16612 SVfARG(varname ? varname : &PL_sv_no),
16615 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16621 * ex: set ts=8 sts=4 sw=4 et: