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));
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));
4811 if (SvREADONLY(sv)) {
4812 /* does undeffing PL_sv_undef count as modifying a read-only
4813 * variable? Some XS code does this */
4814 if (sv == &PL_sv_undef)
4816 Perl_croak_no_modify();
4821 sv_unref_flags(sv, 0);
4824 SvFLAGS(sv) = type; /* quickly turn off all flags */
4825 SvREFCNT_dec_NN(rv);
4829 SvFLAGS(sv) = type; /* quickly turn off all flags */
4834 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4837 SV_CHECK_THINKFIRST_COW_DROP(sv);
4839 if (isGV_with_GP(sv))
4840 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4841 "Undefined value assigned to typeglob");
4849 =for apidoc sv_setsv_mg
4851 Like C<sv_setsv>, but also handles 'set' magic.
4857 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4859 PERL_ARGS_ASSERT_SV_SETSV_MG;
4861 sv_setsv(dstr,sstr);
4866 # define SVt_COW SVt_PV
4868 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4870 STRLEN cur = SvCUR(sstr);
4871 STRLEN len = SvLEN(sstr);
4873 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4874 const bool already = cBOOL(SvIsCOW(sstr));
4877 PERL_ARGS_ASSERT_SV_SETSV_COW;
4880 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4881 (void*)sstr, (void*)dstr);
4888 if (SvTHINKFIRST(dstr))
4889 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4890 else if (SvPVX_const(dstr))
4891 Safefree(SvPVX_mutable(dstr));
4895 SvUPGRADE(dstr, SVt_COW);
4897 assert (SvPOK(sstr));
4898 assert (SvPOKp(sstr));
4900 if (SvIsCOW(sstr)) {
4902 if (SvLEN(sstr) == 0) {
4903 /* source is a COW shared hash key. */
4904 DEBUG_C(PerlIO_printf(Perl_debug_log,
4905 "Fast copy on write: Sharing hash\n"));
4906 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4909 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4910 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4912 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4913 SvUPGRADE(sstr, SVt_COW);
4915 DEBUG_C(PerlIO_printf(Perl_debug_log,
4916 "Fast copy on write: Converting sstr to COW\n"));
4917 CowREFCNT(sstr) = 0;
4919 # ifdef PERL_DEBUG_READONLY_COW
4920 if (already) sv_buf_to_rw(sstr);
4923 new_pv = SvPVX_mutable(sstr);
4927 SvPV_set(dstr, new_pv);
4928 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4931 SvLEN_set(dstr, len);
4932 SvCUR_set(dstr, cur);
4941 =for apidoc sv_setpv_bufsize
4943 Sets the SV to be a string of cur bytes length, with at least
4944 len bytes available. Ensures that there is a null byte at SvEND.
4945 Returns a char * pointer to the SvPV buffer.
4951 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4955 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4957 SV_CHECK_THINKFIRST_COW_DROP(sv);
4958 SvUPGRADE(sv, SVt_PV);
4959 pv = SvGROW(sv, len + 1);
4962 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4965 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4970 =for apidoc sv_setpvn
4972 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4973 The C<len> parameter indicates the number of
4974 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4975 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4981 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4985 PERL_ARGS_ASSERT_SV_SETPVN;
4987 SV_CHECK_THINKFIRST_COW_DROP(sv);
4988 if (isGV_with_GP(sv))
4989 Perl_croak_no_modify();
4995 /* len is STRLEN which is unsigned, need to copy to signed */
4998 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
5001 SvUPGRADE(sv, SVt_PV);
5003 dptr = SvGROW(sv, len + 1);
5004 Move(ptr,dptr,len,char);
5007 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5009 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5013 =for apidoc sv_setpvn_mg
5015 Like C<sv_setpvn>, but also handles 'set' magic.
5021 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5023 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5025 sv_setpvn(sv,ptr,len);
5030 =for apidoc sv_setpv
5032 Copies a string into an SV. The string must be terminated with a C<NUL>
5033 character, and not contain embeded C<NUL>'s.
5034 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5040 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5044 PERL_ARGS_ASSERT_SV_SETPV;
5046 SV_CHECK_THINKFIRST_COW_DROP(sv);
5052 SvUPGRADE(sv, SVt_PV);
5054 SvGROW(sv, len + 1);
5055 Move(ptr,SvPVX(sv),len+1,char);
5057 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5059 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5063 =for apidoc sv_setpv_mg
5065 Like C<sv_setpv>, but also handles 'set' magic.
5071 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5073 PERL_ARGS_ASSERT_SV_SETPV_MG;
5080 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5082 PERL_ARGS_ASSERT_SV_SETHEK;
5088 if (HEK_LEN(hek) == HEf_SVKEY) {
5089 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5092 const int flags = HEK_FLAGS(hek);
5093 if (flags & HVhek_WASUTF8) {
5094 STRLEN utf8_len = HEK_LEN(hek);
5095 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5096 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5099 } else if (flags & HVhek_UNSHARED) {
5100 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5103 else SvUTF8_off(sv);
5107 SV_CHECK_THINKFIRST_COW_DROP(sv);
5108 SvUPGRADE(sv, SVt_PV);
5110 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5111 SvCUR_set(sv, HEK_LEN(hek));
5117 else SvUTF8_off(sv);
5125 =for apidoc sv_usepvn_flags
5127 Tells an SV to use C<ptr> to find its string value. Normally the
5128 string is stored inside the SV, but sv_usepvn allows the SV to use an
5129 outside string. C<ptr> should point to memory that was allocated
5130 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5131 the start of a C<Newx>-ed block of memory, and not a pointer to the
5132 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5133 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5134 string length, C<len>, must be supplied. By default this function
5135 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5136 so that pointer should not be freed or used by the programmer after
5137 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5138 that pointer (e.g. ptr + 1) be used.
5140 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5141 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5143 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5144 C<len>, and already meets the requirements for storing in C<SvPVX>).
5150 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5154 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5156 SV_CHECK_THINKFIRST_COW_DROP(sv);
5157 SvUPGRADE(sv, SVt_PV);
5160 if (flags & SV_SMAGIC)
5164 if (SvPVX_const(sv))
5168 if (flags & SV_HAS_TRAILING_NUL)
5169 assert(ptr[len] == '\0');
5172 allocate = (flags & SV_HAS_TRAILING_NUL)
5174 #ifdef Perl_safesysmalloc_size
5177 PERL_STRLEN_ROUNDUP(len + 1);
5179 if (flags & SV_HAS_TRAILING_NUL) {
5180 /* It's long enough - do nothing.
5181 Specifically Perl_newCONSTSUB is relying on this. */
5184 /* Force a move to shake out bugs in callers. */
5185 char *new_ptr = (char*)safemalloc(allocate);
5186 Copy(ptr, new_ptr, len, char);
5187 PoisonFree(ptr,len,char);
5191 ptr = (char*) saferealloc (ptr, allocate);
5194 #ifdef Perl_safesysmalloc_size
5195 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5197 SvLEN_set(sv, allocate);
5201 if (!(flags & SV_HAS_TRAILING_NUL)) {
5204 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5206 if (flags & SV_SMAGIC)
5211 =for apidoc sv_force_normal_flags
5213 Undo various types of fakery on an SV, where fakery means
5214 "more than" a string: if the PV is a shared string, make
5215 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5216 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5217 we do the copy, and is also used locally; if this is a
5218 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5219 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5220 C<SvPOK_off> rather than making a copy. (Used where this
5221 scalar is about to be set to some other value.) In addition,
5222 the C<flags> parameter gets passed to C<sv_unref_flags()>
5223 when unreffing. C<sv_force_normal> calls this function
5224 with flags set to 0.
5226 This function is expected to be used to signal to perl that this SV is
5227 about to be written to, and any extra book-keeping needs to be taken care
5228 of. Hence, it croaks on read-only values.
5234 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5236 assert(SvIsCOW(sv));
5239 const char * const pvx = SvPVX_const(sv);
5240 const STRLEN len = SvLEN(sv);
5241 const STRLEN cur = SvCUR(sv);
5244 PerlIO_printf(Perl_debug_log,
5245 "Copy on write: Force normal %ld\n",
5250 # ifdef PERL_COPY_ON_WRITE
5252 /* Must do this first, since the CowREFCNT uses SvPVX and
5253 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5254 the only owner left of the buffer. */
5255 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5257 U8 cowrefcnt = CowREFCNT(sv);
5258 if(cowrefcnt != 0) {
5260 CowREFCNT(sv) = cowrefcnt;
5265 /* Else we are the only owner of the buffer. */
5270 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5275 if (flags & SV_COW_DROP_PV) {
5276 /* OK, so we don't need to copy our buffer. */
5279 SvGROW(sv, cur + 1);
5280 Move(pvx,SvPVX(sv),cur,char);
5286 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5293 const char * const pvx = SvPVX_const(sv);
5294 const STRLEN len = SvCUR(sv);
5298 if (flags & SV_COW_DROP_PV) {
5299 /* OK, so we don't need to copy our buffer. */
5302 SvGROW(sv, len + 1);
5303 Move(pvx,SvPVX(sv),len,char);
5306 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5312 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5314 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5317 Perl_croak_no_modify();
5318 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5319 S_sv_uncow(aTHX_ sv, flags);
5321 sv_unref_flags(sv, flags);
5322 else if (SvFAKE(sv) && isGV_with_GP(sv))
5323 sv_unglob(sv, flags);
5324 else if (SvFAKE(sv) && isREGEXP(sv)) {
5325 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5326 to sv_unglob. We only need it here, so inline it. */
5327 const bool islv = SvTYPE(sv) == SVt_PVLV;
5328 const svtype new_type =
5329 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5330 SV *const temp = newSV_type(new_type);
5331 regexp *const temp_p = ReANY((REGEXP *)sv);
5333 if (new_type == SVt_PVMG) {
5334 SvMAGIC_set(temp, SvMAGIC(sv));
5335 SvMAGIC_set(sv, NULL);
5336 SvSTASH_set(temp, SvSTASH(sv));
5337 SvSTASH_set(sv, NULL);
5339 if (!islv) SvCUR_set(temp, SvCUR(sv));
5340 /* Remember that SvPVX is in the head, not the body. But
5341 RX_WRAPPED is in the body. */
5342 assert(ReANY((REGEXP *)sv)->mother_re);
5343 /* Their buffer is already owned by someone else. */
5344 if (flags & SV_COW_DROP_PV) {
5345 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5346 zeroed body. For SVt_PVLV, it should have been set to 0
5347 before turning into a regexp. */
5348 assert(!SvLEN(islv ? sv : temp));
5349 sv->sv_u.svu_pv = 0;
5352 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5353 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5357 /* Now swap the rest of the bodies. */
5361 SvFLAGS(sv) &= ~SVTYPEMASK;
5362 SvFLAGS(sv) |= new_type;
5363 SvANY(sv) = SvANY(temp);
5366 SvFLAGS(temp) &= ~(SVTYPEMASK);
5367 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5368 SvANY(temp) = temp_p;
5369 temp->sv_u.svu_rx = (regexp *)temp_p;
5371 SvREFCNT_dec_NN(temp);
5373 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5379 Efficient removal of characters from the beginning of the string buffer.
5380 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5381 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5382 character of the adjusted string. Uses the C<OOK> hack. On return, only
5383 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5385 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5386 refer to the same chunk of data.
5388 The unfortunate similarity of this function's name to that of Perl's C<chop>
5389 operator is strictly coincidental. This function works from the left;
5390 C<chop> works from the right.
5396 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5407 PERL_ARGS_ASSERT_SV_CHOP;
5409 if (!ptr || !SvPOKp(sv))
5411 delta = ptr - SvPVX_const(sv);
5413 /* Nothing to do. */
5416 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5417 if (delta > max_delta)
5418 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5419 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5420 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5421 SV_CHECK_THINKFIRST(sv);
5422 SvPOK_only_UTF8(sv);
5425 if (!SvLEN(sv)) { /* make copy of shared string */
5426 const char *pvx = SvPVX_const(sv);
5427 const STRLEN len = SvCUR(sv);
5428 SvGROW(sv, len + 1);
5429 Move(pvx,SvPVX(sv),len,char);
5435 SvOOK_offset(sv, old_delta);
5437 SvLEN_set(sv, SvLEN(sv) - delta);
5438 SvCUR_set(sv, SvCUR(sv) - delta);
5439 SvPV_set(sv, SvPVX(sv) + delta);
5441 p = (U8 *)SvPVX_const(sv);
5444 /* how many bytes were evacuated? we will fill them with sentinel
5445 bytes, except for the part holding the new offset of course. */
5448 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5450 assert(evacn <= delta + old_delta);
5454 /* This sets 'delta' to the accumulated value of all deltas so far */
5458 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5459 * the string; otherwise store a 0 byte there and store 'delta' just prior
5460 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5461 * portion of the chopped part of the string */
5462 if (delta < 0x100) {
5466 p -= sizeof(STRLEN);
5467 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5471 /* Fill the preceding buffer with sentinals to verify that no-one is
5481 =for apidoc sv_catpvn
5483 Concatenates the string onto the end of the string which is in the SV.
5484 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5485 status set, then the bytes appended should be valid UTF-8.
5486 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5488 =for apidoc sv_catpvn_flags
5490 Concatenates the string onto the end of the string which is in the SV. The
5491 C<len> indicates number of bytes to copy.
5493 By default, the string appended is assumed to be valid UTF-8 if the SV has
5494 the UTF-8 status set, and a string of bytes otherwise. One can force the
5495 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5496 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5497 string appended will be upgraded to UTF-8 if necessary.
5499 If C<flags> has the C<SV_SMAGIC> bit set, will
5500 C<mg_set> on C<dsv> afterwards if appropriate.
5501 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5502 in terms of this function.
5508 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5511 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5513 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5514 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5516 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5517 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5518 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5521 else SvGROW(dsv, dlen + slen + 3);
5523 sstr = SvPVX_const(dsv);
5524 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5525 SvCUR_set(dsv, SvCUR(dsv) + slen);
5528 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5529 const char * const send = sstr + slen;
5532 /* Something this code does not account for, which I think is
5533 impossible; it would require the same pv to be treated as
5534 bytes *and* utf8, which would indicate a bug elsewhere. */
5535 assert(sstr != dstr);
5537 SvGROW(dsv, dlen + slen * 2 + 3);
5538 d = (U8 *)SvPVX(dsv) + dlen;
5540 while (sstr < send) {
5541 append_utf8_from_native_byte(*sstr, &d);
5544 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5547 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5549 if (flags & SV_SMAGIC)
5554 =for apidoc sv_catsv
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 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5559 and C<L</sv_catsv_nomg>>.
5561 =for apidoc sv_catsv_flags
5563 Concatenates the string from SV C<ssv> onto the end of the string in SV
5564 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5565 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5566 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5567 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5568 and C<sv_catsv_mg> are implemented in terms of this function.
5573 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5575 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5579 const char *spv = SvPV_flags_const(ssv, slen, flags);
5580 if (flags & SV_GMAGIC)
5582 sv_catpvn_flags(dsv, spv, slen,
5583 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5584 if (flags & SV_SMAGIC)
5590 =for apidoc sv_catpv
5592 Concatenates the C<NUL>-terminated string onto the end of the string which is
5594 If the SV has the UTF-8 status set, then the bytes appended should be
5595 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5601 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5607 PERL_ARGS_ASSERT_SV_CATPV;
5611 junk = SvPV_force(sv, tlen);
5613 SvGROW(sv, tlen + len + 1);
5615 ptr = SvPVX_const(sv);
5616 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5617 SvCUR_set(sv, SvCUR(sv) + len);
5618 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5623 =for apidoc sv_catpv_flags
5625 Concatenates the C<NUL>-terminated string onto the end of the string which is
5627 If the SV has the UTF-8 status set, then the bytes appended should
5628 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5629 on the modified SV if appropriate.
5635 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5637 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5638 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5642 =for apidoc sv_catpv_mg
5644 Like C<sv_catpv>, but also handles 'set' magic.
5650 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5652 PERL_ARGS_ASSERT_SV_CATPV_MG;
5661 Creates a new SV. A non-zero C<len> parameter indicates the number of
5662 bytes of preallocated string space the SV should have. An extra byte for a
5663 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5664 space is allocated.) The reference count for the new SV is set to 1.
5666 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5667 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5668 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5669 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5670 modules supporting older perls.
5676 Perl_newSV(pTHX_ const STRLEN len)
5682 sv_grow(sv, len + 1);
5687 =for apidoc sv_magicext
5689 Adds magic to an SV, upgrading it if necessary. Applies the
5690 supplied C<vtable> and returns a pointer to the magic added.
5692 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5693 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5694 one instance of the same C<how>.
5696 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5697 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5698 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5699 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5701 (This is now used as a subroutine by C<sv_magic>.)
5706 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5707 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5711 PERL_ARGS_ASSERT_SV_MAGICEXT;
5713 SvUPGRADE(sv, SVt_PVMG);
5714 Newxz(mg, 1, MAGIC);
5715 mg->mg_moremagic = SvMAGIC(sv);
5716 SvMAGIC_set(sv, mg);
5718 /* Sometimes a magic contains a reference loop, where the sv and
5719 object refer to each other. To prevent a reference loop that
5720 would prevent such objects being freed, we look for such loops
5721 and if we find one we avoid incrementing the object refcount.
5723 Note we cannot do this to avoid self-tie loops as intervening RV must
5724 have its REFCNT incremented to keep it in existence.
5727 if (!obj || obj == sv ||
5728 how == PERL_MAGIC_arylen ||
5729 how == PERL_MAGIC_regdata ||
5730 how == PERL_MAGIC_regdatum ||
5731 how == PERL_MAGIC_symtab ||
5732 (SvTYPE(obj) == SVt_PVGV &&
5733 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5734 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5735 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5740 mg->mg_obj = SvREFCNT_inc_simple(obj);
5741 mg->mg_flags |= MGf_REFCOUNTED;
5744 /* Normal self-ties simply pass a null object, and instead of
5745 using mg_obj directly, use the SvTIED_obj macro to produce a
5746 new RV as needed. For glob "self-ties", we are tieing the PVIO
5747 with an RV obj pointing to the glob containing the PVIO. In
5748 this case, to avoid a reference loop, we need to weaken the
5752 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5753 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5759 mg->mg_len = namlen;
5762 mg->mg_ptr = savepvn(name, namlen);
5763 else if (namlen == HEf_SVKEY) {
5764 /* Yes, this is casting away const. This is only for the case of
5765 HEf_SVKEY. I think we need to document this aberation of the
5766 constness of the API, rather than making name non-const, as
5767 that change propagating outwards a long way. */
5768 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5770 mg->mg_ptr = (char *) name;
5772 mg->mg_virtual = (MGVTBL *) vtable;
5779 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5781 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5782 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5783 /* This sv is only a delegate. //g magic must be attached to
5788 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5789 &PL_vtbl_mglob, 0, 0);
5793 =for apidoc sv_magic
5795 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5796 necessary, then adds a new magic item of type C<how> to the head of the
5799 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5800 handling of the C<name> and C<namlen> arguments.
5802 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5803 to add more than one instance of the same C<how>.
5809 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5810 const char *const name, const I32 namlen)
5812 const MGVTBL *vtable;
5815 unsigned int vtable_index;
5817 PERL_ARGS_ASSERT_SV_MAGIC;
5819 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5820 || ((flags = PL_magic_data[how]),
5821 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5822 > magic_vtable_max))
5823 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5825 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5826 Useful for attaching extension internal data to perl vars.
5827 Note that multiple extensions may clash if magical scalars
5828 etc holding private data from one are passed to another. */
5830 vtable = (vtable_index == magic_vtable_max)
5831 ? NULL : PL_magic_vtables + vtable_index;
5833 if (SvREADONLY(sv)) {
5835 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5838 Perl_croak_no_modify();
5841 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5842 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5843 /* sv_magic() refuses to add a magic of the same 'how' as an
5846 if (how == PERL_MAGIC_taint)
5852 /* Force pos to be stored as characters, not bytes. */
5853 if (SvMAGICAL(sv) && DO_UTF8(sv)
5854 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5856 && mg->mg_flags & MGf_BYTES) {
5857 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5859 mg->mg_flags &= ~MGf_BYTES;
5862 /* Rest of work is done else where */
5863 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5866 case PERL_MAGIC_taint:
5869 case PERL_MAGIC_ext:
5870 case PERL_MAGIC_dbfile:
5877 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5884 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5886 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5887 for (mg = *mgp; mg; mg = *mgp) {
5888 const MGVTBL* const virt = mg->mg_virtual;
5889 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5890 *mgp = mg->mg_moremagic;
5891 if (virt && virt->svt_free)
5892 virt->svt_free(aTHX_ sv, mg);
5893 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5895 Safefree(mg->mg_ptr);
5896 else if (mg->mg_len == HEf_SVKEY)
5897 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5898 else if (mg->mg_type == PERL_MAGIC_utf8)
5899 Safefree(mg->mg_ptr);
5901 if (mg->mg_flags & MGf_REFCOUNTED)
5902 SvREFCNT_dec(mg->mg_obj);
5906 mgp = &mg->mg_moremagic;
5909 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5910 mg_magical(sv); /* else fix the flags now */
5919 =for apidoc sv_unmagic
5921 Removes all magic of type C<type> from an SV.
5927 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5929 PERL_ARGS_ASSERT_SV_UNMAGIC;
5930 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5934 =for apidoc sv_unmagicext
5936 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5942 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5944 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5945 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5949 =for apidoc sv_rvweaken
5951 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5952 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5953 push a back-reference to this RV onto the array of backreferences
5954 associated with that magic. If the RV is magical, set magic will be
5955 called after the RV is cleared.
5961 Perl_sv_rvweaken(pTHX_ SV *const sv)
5965 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5967 if (!SvOK(sv)) /* let undefs pass */
5970 Perl_croak(aTHX_ "Can't weaken a nonreference");
5971 else if (SvWEAKREF(sv)) {
5972 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5975 else if (SvREADONLY(sv)) croak_no_modify();
5977 Perl_sv_add_backref(aTHX_ tsv, sv);
5979 SvREFCNT_dec_NN(tsv);
5984 =for apidoc sv_get_backrefs
5986 If C<sv> is the target of a weak reference then it returns the back
5987 references structure associated with the sv; otherwise return C<NULL>.
5989 When returning a non-null result the type of the return is relevant. If it
5990 is an AV then the elements of the AV are the weak reference RVs which
5991 point at this item. If it is any other type then the item itself is the
5994 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5995 C<Perl_sv_kill_backrefs()>
6001 Perl_sv_get_backrefs(SV *const sv)
6005 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6007 /* find slot to store array or singleton backref */
6009 if (SvTYPE(sv) == SVt_PVHV) {
6011 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6012 backrefs = (SV *)iter->xhv_backreferences;
6014 } else if (SvMAGICAL(sv)) {
6015 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6017 backrefs = mg->mg_obj;
6022 /* Give tsv backref magic if it hasn't already got it, then push a
6023 * back-reference to sv onto the array associated with the backref magic.
6025 * As an optimisation, if there's only one backref and it's not an AV,
6026 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6027 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6031 /* A discussion about the backreferences array and its refcount:
6033 * The AV holding the backreferences is pointed to either as the mg_obj of
6034 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6035 * xhv_backreferences field. The array is created with a refcount
6036 * of 2. This means that if during global destruction the array gets
6037 * picked on before its parent to have its refcount decremented by the
6038 * random zapper, it won't actually be freed, meaning it's still there for
6039 * when its parent gets freed.
6041 * When the parent SV is freed, the extra ref is killed by
6042 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6043 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6045 * When a single backref SV is stored directly, it is not reference
6050 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6056 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6058 /* find slot to store array or singleton backref */
6060 if (SvTYPE(tsv) == SVt_PVHV) {
6061 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6064 mg = mg_find(tsv, PERL_MAGIC_backref);
6066 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6067 svp = &(mg->mg_obj);
6070 /* create or retrieve the array */
6072 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6073 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6077 mg->mg_flags |= MGf_REFCOUNTED;
6080 SvREFCNT_inc_simple_void_NN(av);
6081 /* av now has a refcnt of 2; see discussion above */
6082 av_extend(av, *svp ? 2 : 1);
6084 /* move single existing backref to the array */
6085 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6090 av = MUTABLE_AV(*svp);
6092 /* optimisation: store single backref directly in HvAUX or mg_obj */
6096 assert(SvTYPE(av) == SVt_PVAV);
6097 if (AvFILLp(av) >= AvMAX(av)) {
6098 av_extend(av, AvFILLp(av)+1);
6101 /* push new backref */
6102 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6105 /* delete a back-reference to ourselves from the backref magic associated
6106 * with the SV we point to.
6110 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6114 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6116 if (SvTYPE(tsv) == SVt_PVHV) {
6118 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6120 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6121 /* It's possible for the the last (strong) reference to tsv to have
6122 become freed *before* the last thing holding a weak reference.
6123 If both survive longer than the backreferences array, then when
6124 the referent's reference count drops to 0 and it is freed, it's
6125 not able to chase the backreferences, so they aren't NULLed.
6127 For example, a CV holds a weak reference to its stash. If both the
6128 CV and the stash survive longer than the backreferences array,
6129 and the CV gets picked for the SvBREAK() treatment first,
6130 *and* it turns out that the stash is only being kept alive because
6131 of an our variable in the pad of the CV, then midway during CV
6132 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6133 It ends up pointing to the freed HV. Hence it's chased in here, and
6134 if this block wasn't here, it would hit the !svp panic just below.
6136 I don't believe that "better" destruction ordering is going to help
6137 here - during global destruction there's always going to be the
6138 chance that something goes out of order. We've tried to make it
6139 foolproof before, and it only resulted in evolutionary pressure on
6140 fools. Which made us look foolish for our hubris. :-(
6146 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6147 svp = mg ? &(mg->mg_obj) : NULL;
6151 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6153 /* It's possible that sv is being freed recursively part way through the
6154 freeing of tsv. If this happens, the backreferences array of tsv has
6155 already been freed, and so svp will be NULL. If this is the case,
6156 we should not panic. Instead, nothing needs doing, so return. */
6157 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6159 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6160 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6163 if (SvTYPE(*svp) == SVt_PVAV) {
6167 AV * const av = (AV*)*svp;
6169 assert(!SvIS_FREED(av));
6173 /* for an SV with N weak references to it, if all those
6174 * weak refs are deleted, then sv_del_backref will be called
6175 * N times and O(N^2) compares will be done within the backref
6176 * array. To ameliorate this potential slowness, we:
6177 * 1) make sure this code is as tight as possible;
6178 * 2) when looking for SV, look for it at both the head and tail of the
6179 * array first before searching the rest, since some create/destroy
6180 * patterns will cause the backrefs to be freed in order.
6187 SV **p = &svp[fill];
6188 SV *const topsv = *p;
6195 /* We weren't the last entry.
6196 An unordered list has this property that you
6197 can take the last element off the end to fill
6198 the hole, and it's still an unordered list :-)
6204 break; /* should only be one */
6211 AvFILLp(av) = fill-1;
6213 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6214 /* freed AV; skip */
6217 /* optimisation: only a single backref, stored directly */
6219 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6220 (void*)*svp, (void*)sv);
6227 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6233 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6238 /* after multiple passes through Perl_sv_clean_all() for a thingy
6239 * that has badly leaked, the backref array may have gotten freed,
6240 * since we only protect it against 1 round of cleanup */
6241 if (SvIS_FREED(av)) {
6242 if (PL_in_clean_all) /* All is fair */
6245 "panic: magic_killbackrefs (freed backref AV/SV)");
6249 is_array = (SvTYPE(av) == SVt_PVAV);
6251 assert(!SvIS_FREED(av));
6254 last = svp + AvFILLp(av);
6257 /* optimisation: only a single backref, stored directly */
6263 while (svp <= last) {
6265 SV *const referrer = *svp;
6266 if (SvWEAKREF(referrer)) {
6267 /* XXX Should we check that it hasn't changed? */
6268 assert(SvROK(referrer));
6269 SvRV_set(referrer, 0);
6271 SvWEAKREF_off(referrer);
6272 SvSETMAGIC(referrer);
6273 } else if (SvTYPE(referrer) == SVt_PVGV ||
6274 SvTYPE(referrer) == SVt_PVLV) {
6275 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6276 /* You lookin' at me? */
6277 assert(GvSTASH(referrer));
6278 assert(GvSTASH(referrer) == (const HV *)sv);
6279 GvSTASH(referrer) = 0;
6280 } else if (SvTYPE(referrer) == SVt_PVCV ||
6281 SvTYPE(referrer) == SVt_PVFM) {
6282 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6283 /* You lookin' at me? */
6284 assert(CvSTASH(referrer));
6285 assert(CvSTASH(referrer) == (const HV *)sv);
6286 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6289 assert(SvTYPE(sv) == SVt_PVGV);
6290 /* You lookin' at me? */
6291 assert(CvGV(referrer));
6292 assert(CvGV(referrer) == (const GV *)sv);
6293 anonymise_cv_maybe(MUTABLE_GV(sv),
6294 MUTABLE_CV(referrer));
6299 "panic: magic_killbackrefs (flags=%" UVxf ")",
6300 (UV)SvFLAGS(referrer));
6311 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6317 =for apidoc sv_insert
6319 Inserts a string at the specified offset/length within the SV. Similar to
6320 the Perl C<substr()> function. Handles get magic.
6322 =for apidoc sv_insert_flags
6324 Same as C<sv_insert>, but the extra C<flags> are passed to the
6325 C<SvPV_force_flags> that applies to C<bigstr>.
6331 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6337 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6340 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6342 SvPV_force_flags(bigstr, curlen, flags);
6343 (void)SvPOK_only_UTF8(bigstr);
6345 if (little >= SvPVX(bigstr) &&
6346 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6347 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6348 or little...little+littlelen might overlap offset...offset+len we make a copy
6350 little = savepvn(little, littlelen);
6354 if (offset + len > curlen) {
6355 SvGROW(bigstr, offset+len+1);
6356 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6357 SvCUR_set(bigstr, offset+len);
6361 i = littlelen - len;
6362 if (i > 0) { /* string might grow */
6363 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6364 mid = big + offset + len;
6365 midend = bigend = big + SvCUR(bigstr);
6368 while (midend > mid) /* shove everything down */
6369 *--bigend = *--midend;
6370 Move(little,big+offset,littlelen,char);
6371 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6376 Move(little,SvPVX(bigstr)+offset,len,char);
6381 big = SvPVX(bigstr);
6384 bigend = big + SvCUR(bigstr);
6386 if (midend > bigend)
6387 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6390 if (mid - big > bigend - midend) { /* faster to shorten from end */
6392 Move(little, mid, littlelen,char);
6395 i = bigend - midend;
6397 Move(midend, mid, i,char);
6401 SvCUR_set(bigstr, mid - big);
6403 else if ((i = mid - big)) { /* faster from front */
6404 midend -= littlelen;
6406 Move(big, midend - i, i, char);
6407 sv_chop(bigstr,midend-i);
6409 Move(little, mid, littlelen,char);
6411 else if (littlelen) {
6412 midend -= littlelen;
6413 sv_chop(bigstr,midend);
6414 Move(little,midend,littlelen,char);
6417 sv_chop(bigstr,midend);
6423 =for apidoc sv_replace
6425 Make the first argument a copy of the second, then delete the original.
6426 The target SV physically takes over ownership of the body of the source SV
6427 and inherits its flags; however, the target keeps any magic it owns,
6428 and any magic in the source is discarded.
6429 Note that this is a rather specialist SV copying operation; most of the
6430 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6436 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6438 const U32 refcnt = SvREFCNT(sv);
6440 PERL_ARGS_ASSERT_SV_REPLACE;
6442 SV_CHECK_THINKFIRST_COW_DROP(sv);
6443 if (SvREFCNT(nsv) != 1) {
6444 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6445 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6447 if (SvMAGICAL(sv)) {
6451 sv_upgrade(nsv, SVt_PVMG);
6452 SvMAGIC_set(nsv, SvMAGIC(sv));
6453 SvFLAGS(nsv) |= SvMAGICAL(sv);
6455 SvMAGIC_set(sv, NULL);
6459 assert(!SvREFCNT(sv));
6460 #ifdef DEBUG_LEAKING_SCALARS
6461 sv->sv_flags = nsv->sv_flags;
6462 sv->sv_any = nsv->sv_any;
6463 sv->sv_refcnt = nsv->sv_refcnt;
6464 sv->sv_u = nsv->sv_u;
6466 StructCopy(nsv,sv,SV);
6468 if(SvTYPE(sv) == SVt_IV) {
6469 SET_SVANY_FOR_BODYLESS_IV(sv);
6473 SvREFCNT(sv) = refcnt;
6474 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6479 /* We're about to free a GV which has a CV that refers back to us.
6480 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6484 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6489 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6492 assert(SvREFCNT(gv) == 0);
6493 assert(isGV(gv) && isGV_with_GP(gv));
6495 assert(!CvANON(cv));
6496 assert(CvGV(cv) == gv);
6497 assert(!CvNAMED(cv));
6499 /* will the CV shortly be freed by gp_free() ? */
6500 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6501 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6505 /* if not, anonymise: */
6506 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6507 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6508 : newSVpvn_flags( "__ANON__", 8, 0 );
6509 sv_catpvs(gvname, "::__ANON__");
6510 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6511 SvREFCNT_dec_NN(gvname);
6515 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6520 =for apidoc sv_clear
6522 Clear an SV: call any destructors, free up any memory used by the body,
6523 and free the body itself. The SV's head is I<not> freed, although
6524 its type is set to all 1's so that it won't inadvertently be assumed
6525 to be live during global destruction etc.
6526 This function should only be called when C<REFCNT> is zero. Most of the time
6527 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6534 Perl_sv_clear(pTHX_ SV *const orig_sv)
6539 const struct body_details *sv_type_details;
6543 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6544 Not strictly necessary */
6546 PERL_ARGS_ASSERT_SV_CLEAR;
6548 /* within this loop, sv is the SV currently being freed, and
6549 * iter_sv is the most recent AV or whatever that's being iterated
6550 * over to provide more SVs */
6556 assert(SvREFCNT(sv) == 0);
6557 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6559 if (type <= SVt_IV) {
6560 /* See the comment in sv.h about the collusion between this
6561 * early return and the overloading of the NULL slots in the
6565 SvFLAGS(sv) &= SVf_BREAK;
6566 SvFLAGS(sv) |= SVTYPEMASK;
6570 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6571 for another purpose */
6572 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6574 if (type >= SVt_PVMG) {
6576 if (!curse(sv, 1)) goto get_next_sv;
6577 type = SvTYPE(sv); /* destructor may have changed it */
6579 /* Free back-references before magic, in case the magic calls
6580 * Perl code that has weak references to sv. */
6581 if (type == SVt_PVHV) {
6582 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6586 else if (SvMAGIC(sv)) {
6587 /* Free back-references before other types of magic. */
6588 sv_unmagic(sv, PERL_MAGIC_backref);
6594 /* case SVt_INVLIST: */
6597 IoIFP(sv) != PerlIO_stdin() &&
6598 IoIFP(sv) != PerlIO_stdout() &&
6599 IoIFP(sv) != PerlIO_stderr() &&
6600 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6602 io_close(MUTABLE_IO(sv), NULL, FALSE,
6603 (IoTYPE(sv) == IoTYPE_WRONLY ||
6604 IoTYPE(sv) == IoTYPE_RDWR ||
6605 IoTYPE(sv) == IoTYPE_APPEND));
6607 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6608 PerlDir_close(IoDIRP(sv));
6609 IoDIRP(sv) = (DIR*)NULL;
6610 Safefree(IoTOP_NAME(sv));
6611 Safefree(IoFMT_NAME(sv));
6612 Safefree(IoBOTTOM_NAME(sv));
6613 if ((const GV *)sv == PL_statgv)
6617 /* FIXME for plugins */
6619 pregfree2((REGEXP*) sv);
6623 cv_undef(MUTABLE_CV(sv));
6624 /* If we're in a stash, we don't own a reference to it.
6625 * However it does have a back reference to us, which needs to
6627 if ((stash = CvSTASH(sv)))
6628 sv_del_backref(MUTABLE_SV(stash), sv);
6631 if (PL_last_swash_hv == (const HV *)sv) {
6632 PL_last_swash_hv = NULL;
6634 if (HvTOTALKEYS((HV*)sv) > 0) {
6636 /* this statement should match the one at the beginning of
6637 * hv_undef_flags() */
6638 if ( PL_phase != PERL_PHASE_DESTRUCT
6639 && (hek = HvNAME_HEK((HV*)sv)))
6641 if (PL_stashcache) {
6642 DEBUG_o(Perl_deb(aTHX_
6643 "sv_clear clearing PL_stashcache for '%" HEKf
6646 (void)hv_deletehek(PL_stashcache,
6649 hv_name_set((HV*)sv, NULL, 0, 0);
6652 /* save old iter_sv in unused SvSTASH field */
6653 assert(!SvOBJECT(sv));
6654 SvSTASH(sv) = (HV*)iter_sv;
6657 /* save old hash_index in unused SvMAGIC field */
6658 assert(!SvMAGICAL(sv));
6659 assert(!SvMAGIC(sv));
6660 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6663 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6664 goto get_next_sv; /* process this new sv */
6666 /* free empty hash */
6667 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6668 assert(!HvARRAY((HV*)sv));
6672 AV* av = MUTABLE_AV(sv);
6673 if (PL_comppad == av) {
6677 if (AvREAL(av) && AvFILLp(av) > -1) {
6678 next_sv = AvARRAY(av)[AvFILLp(av)--];
6679 /* save old iter_sv in top-most slot of AV,
6680 * and pray that it doesn't get wiped in the meantime */
6681 AvARRAY(av)[AvMAX(av)] = iter_sv;
6683 goto get_next_sv; /* process this new sv */
6685 Safefree(AvALLOC(av));
6690 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6691 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6692 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6693 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6695 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6696 SvREFCNT_dec(LvTARG(sv));
6697 if (isREGEXP(sv)) goto freeregexp;
6700 if (isGV_with_GP(sv)) {
6701 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6702 && HvENAME_get(stash))
6703 mro_method_changed_in(stash);
6704 gp_free(MUTABLE_GV(sv));
6706 unshare_hek(GvNAME_HEK(sv));
6707 /* If we're in a stash, we don't own a reference to it.
6708 * However it does have a back reference to us, which
6709 * needs to be cleared. */
6710 if ((stash = GvSTASH(sv)))
6711 sv_del_backref(MUTABLE_SV(stash), sv);
6713 /* FIXME. There are probably more unreferenced pointers to SVs
6714 * in the interpreter struct that we should check and tidy in
6715 * a similar fashion to this: */
6716 /* See also S_sv_unglob, which does the same thing. */
6717 if ((const GV *)sv == PL_last_in_gv)
6718 PL_last_in_gv = NULL;
6719 else if ((const GV *)sv == PL_statgv)
6721 else if ((const GV *)sv == PL_stderrgv)
6730 /* Don't bother with SvOOK_off(sv); as we're only going to
6734 SvOOK_offset(sv, offset);
6735 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6736 /* Don't even bother with turning off the OOK flag. */
6741 SV * const target = SvRV(sv);
6743 sv_del_backref(target, sv);
6749 else if (SvPVX_const(sv)
6750 && !(SvTYPE(sv) == SVt_PVIO
6751 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6755 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6759 if (CowREFCNT(sv)) {
6766 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6771 Safefree(SvPVX_mutable(sv));
6775 else if (SvPVX_const(sv) && SvLEN(sv)
6776 && !(SvTYPE(sv) == SVt_PVIO
6777 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6778 Safefree(SvPVX_mutable(sv));
6779 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6780 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6790 SvFLAGS(sv) &= SVf_BREAK;
6791 SvFLAGS(sv) |= SVTYPEMASK;
6793 sv_type_details = bodies_by_type + type;
6794 if (sv_type_details->arena) {
6795 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6796 &PL_body_roots[type]);
6798 else if (sv_type_details->body_size) {
6799 safefree(SvANY(sv));
6803 /* caller is responsible for freeing the head of the original sv */
6804 if (sv != orig_sv && !SvREFCNT(sv))
6807 /* grab and free next sv, if any */
6815 else if (!iter_sv) {
6817 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6818 AV *const av = (AV*)iter_sv;
6819 if (AvFILLp(av) > -1) {
6820 sv = AvARRAY(av)[AvFILLp(av)--];
6822 else { /* no more elements of current AV to free */
6825 /* restore previous value, squirrelled away */
6826 iter_sv = AvARRAY(av)[AvMAX(av)];
6827 Safefree(AvALLOC(av));
6830 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6831 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6832 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6833 /* no more elements of current HV to free */
6836 /* Restore previous values of iter_sv and hash_index,
6837 * squirrelled away */
6838 assert(!SvOBJECT(sv));
6839 iter_sv = (SV*)SvSTASH(sv);
6840 assert(!SvMAGICAL(sv));
6841 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6843 /* perl -DA does not like rubbish in SvMAGIC. */
6847 /* free any remaining detritus from the hash struct */
6848 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6849 assert(!HvARRAY((HV*)sv));
6854 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6858 if (!SvREFCNT(sv)) {
6862 if (--(SvREFCNT(sv)))
6866 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6867 "Attempt to free temp prematurely: SV 0x%" UVxf
6868 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6872 if (SvIMMORTAL(sv)) {
6873 /* make sure SvREFCNT(sv)==0 happens very seldom */
6874 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6883 /* This routine curses the sv itself, not the object referenced by sv. So
6884 sv does not have to be ROK. */
6887 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6888 PERL_ARGS_ASSERT_CURSE;
6889 assert(SvOBJECT(sv));
6891 if (PL_defstash && /* Still have a symbol table? */
6897 stash = SvSTASH(sv);
6898 assert(SvTYPE(stash) == SVt_PVHV);
6899 if (HvNAME(stash)) {
6900 CV* destructor = NULL;
6901 struct mro_meta *meta;
6903 assert (SvOOK(stash));
6905 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6908 /* don't make this an initialization above the assert, since it needs
6910 meta = HvMROMETA(stash);
6911 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6912 destructor = meta->destroy;
6913 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6914 (void *)destructor, HvNAME(stash)) );
6917 bool autoload = FALSE;
6919 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6921 destructor = GvCV(gv);
6923 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6924 GV_AUTOLOAD_ISMETHOD);
6926 destructor = GvCV(gv);
6930 /* we don't cache AUTOLOAD for DESTROY, since this code
6931 would then need to set $__PACKAGE__::AUTOLOAD, or the
6932 equivalent for XS AUTOLOADs */
6934 meta->destroy_gen = PL_sub_generation;
6935 meta->destroy = destructor;
6937 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6938 (void *)destructor, HvNAME(stash)) );
6941 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6945 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6947 /* A constant subroutine can have no side effects, so
6948 don't bother calling it. */
6949 && !CvCONST(destructor)
6950 /* Don't bother calling an empty destructor or one that
6951 returns immediately. */
6952 && (CvISXSUB(destructor)
6953 || (CvSTART(destructor)
6954 && (CvSTART(destructor)->op_next->op_type
6956 && (CvSTART(destructor)->op_next->op_type
6958 || CvSTART(destructor)->op_next->op_next->op_type
6964 SV* const tmpref = newRV(sv);
6965 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6967 PUSHSTACKi(PERLSI_DESTROY);
6972 call_sv(MUTABLE_SV(destructor),
6973 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6977 if(SvREFCNT(tmpref) < 2) {
6978 /* tmpref is not kept alive! */
6980 SvRV_set(tmpref, NULL);
6983 SvREFCNT_dec_NN(tmpref);
6986 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6989 if (check_refcnt && SvREFCNT(sv)) {
6990 if (PL_in_clean_objs)
6992 "DESTROY created new reference to dead object '%" HEKf "'",
6993 HEKfARG(HvNAME_HEK(stash)));
6994 /* DESTROY gave object new lease on life */
7000 HV * const stash = SvSTASH(sv);
7001 /* Curse before freeing the stash, as freeing the stash could cause
7002 a recursive call into S_curse. */
7003 SvOBJECT_off(sv); /* Curse the object. */
7004 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7005 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7011 =for apidoc sv_newref
7013 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7020 Perl_sv_newref(pTHX_ SV *const sv)
7022 PERL_UNUSED_CONTEXT;
7031 Decrement an SV's reference count, and if it drops to zero, call
7032 C<sv_clear> to invoke destructors and free up any memory used by
7033 the body; finally, deallocating the SV's head itself.
7034 Normally called via a wrapper macro C<SvREFCNT_dec>.
7040 Perl_sv_free(pTHX_ SV *const sv)
7046 /* Private helper function for SvREFCNT_dec().
7047 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7050 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7054 PERL_ARGS_ASSERT_SV_FREE2;
7056 if (LIKELY( rc == 1 )) {
7062 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7063 "Attempt to free temp prematurely: SV 0x%" UVxf
7064 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7068 if (SvIMMORTAL(sv)) {
7069 /* make sure SvREFCNT(sv)==0 happens very seldom */
7070 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7074 if (! SvREFCNT(sv)) /* may have have been resurrected */
7079 /* handle exceptional cases */
7083 if (SvFLAGS(sv) & SVf_BREAK)
7084 /* this SV's refcnt has been artificially decremented to
7085 * trigger cleanup */
7087 if (PL_in_clean_all) /* All is fair */
7089 if (SvIMMORTAL(sv)) {
7090 /* make sure SvREFCNT(sv)==0 happens very seldom */
7091 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7094 if (ckWARN_d(WARN_INTERNAL)) {
7095 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7096 Perl_dump_sv_child(aTHX_ sv);
7098 #ifdef DEBUG_LEAKING_SCALARS
7101 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7102 if (PL_warnhook == PERL_WARNHOOK_FATAL
7103 || ckDEAD(packWARN(WARN_INTERNAL))) {
7104 /* Don't let Perl_warner cause us to escape our fate: */
7108 /* This may not return: */
7109 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7110 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7111 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7114 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7124 Returns the length of the string in the SV. Handles magic and type
7125 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7126 gives raw access to the C<xpv_cur> slot.
7132 Perl_sv_len(pTHX_ SV *const sv)
7139 (void)SvPV_const(sv, len);
7144 =for apidoc sv_len_utf8
7146 Returns the number of characters in the string in an SV, counting wide
7147 UTF-8 bytes as a single character. Handles magic and type coercion.
7153 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7154 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7155 * (Note that the mg_len is not the length of the mg_ptr field.
7156 * This allows the cache to store the character length of the string without
7157 * needing to malloc() extra storage to attach to the mg_ptr.)
7162 Perl_sv_len_utf8(pTHX_ SV *const sv)
7168 return sv_len_utf8_nomg(sv);
7172 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7175 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7177 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7179 if (PL_utf8cache && SvUTF8(sv)) {
7181 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7183 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7184 if (mg->mg_len != -1)
7187 /* We can use the offset cache for a headstart.
7188 The longer value is stored in the first pair. */
7189 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7191 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7195 if (PL_utf8cache < 0) {
7196 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7197 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7201 ulen = Perl_utf8_length(aTHX_ s, s + len);
7202 utf8_mg_len_cache_update(sv, &mg, ulen);
7206 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7209 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7212 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7213 STRLEN *const uoffset_p, bool *const at_end)
7215 const U8 *s = start;
7216 STRLEN uoffset = *uoffset_p;
7218 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7220 while (s < send && uoffset) {
7227 else if (s > send) {
7229 /* This is the existing behaviour. Possibly it should be a croak, as
7230 it's actually a bounds error */
7233 *uoffset_p -= uoffset;
7237 /* Given the length of the string in both bytes and UTF-8 characters, decide
7238 whether to walk forwards or backwards to find the byte corresponding to
7239 the passed in UTF-8 offset. */
7241 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7242 STRLEN uoffset, const STRLEN uend)
7244 STRLEN backw = uend - uoffset;
7246 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7248 if (uoffset < 2 * backw) {
7249 /* The assumption is that going forwards is twice the speed of going
7250 forward (that's where the 2 * backw comes from).
7251 (The real figure of course depends on the UTF-8 data.) */
7252 const U8 *s = start;
7254 while (s < send && uoffset--)
7264 while (UTF8_IS_CONTINUATION(*send))
7267 return send - start;
7270 /* For the string representation of the given scalar, find the byte
7271 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7272 give another position in the string, *before* the sought offset, which
7273 (which is always true, as 0, 0 is a valid pair of positions), which should
7274 help reduce the amount of linear searching.
7275 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7276 will be used to reduce the amount of linear searching. The cache will be
7277 created if necessary, and the found value offered to it for update. */
7279 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7280 const U8 *const send, STRLEN uoffset,
7281 STRLEN uoffset0, STRLEN boffset0)
7283 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7285 bool at_end = FALSE;
7287 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7289 assert (uoffset >= uoffset0);
7294 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7296 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7297 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7298 if ((*mgp)->mg_ptr) {
7299 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7300 if (cache[0] == uoffset) {
7301 /* An exact match. */
7304 if (cache[2] == uoffset) {
7305 /* An exact match. */
7309 if (cache[0] < uoffset) {
7310 /* The cache already knows part of the way. */
7311 if (cache[0] > uoffset0) {
7312 /* The cache knows more than the passed in pair */
7313 uoffset0 = cache[0];
7314 boffset0 = cache[1];
7316 if ((*mgp)->mg_len != -1) {
7317 /* And we know the end too. */
7319 + sv_pos_u2b_midway(start + boffset0, send,
7321 (*mgp)->mg_len - uoffset0);
7323 uoffset -= uoffset0;
7325 + sv_pos_u2b_forwards(start + boffset0,
7326 send, &uoffset, &at_end);
7327 uoffset += uoffset0;
7330 else if (cache[2] < uoffset) {
7331 /* We're between the two cache entries. */
7332 if (cache[2] > uoffset0) {
7333 /* and the cache knows more than the passed in pair */
7334 uoffset0 = cache[2];
7335 boffset0 = cache[3];
7339 + sv_pos_u2b_midway(start + boffset0,
7342 cache[0] - uoffset0);
7345 + sv_pos_u2b_midway(start + boffset0,
7348 cache[2] - uoffset0);
7352 else if ((*mgp)->mg_len != -1) {
7353 /* If we can take advantage of a passed in offset, do so. */
7354 /* In fact, offset0 is either 0, or less than offset, so don't
7355 need to worry about the other possibility. */
7357 + sv_pos_u2b_midway(start + boffset0, send,
7359 (*mgp)->mg_len - uoffset0);
7364 if (!found || PL_utf8cache < 0) {
7365 STRLEN real_boffset;
7366 uoffset -= uoffset0;
7367 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7368 send, &uoffset, &at_end);
7369 uoffset += uoffset0;
7371 if (found && PL_utf8cache < 0)
7372 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7374 boffset = real_boffset;
7377 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7379 utf8_mg_len_cache_update(sv, mgp, uoffset);
7381 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7388 =for apidoc sv_pos_u2b_flags
7390 Converts the offset from a count of UTF-8 chars from
7391 the start of the string, to a count of the equivalent number of bytes; if
7392 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7393 C<offset>, rather than from the start
7394 of the string. Handles type coercion.
7395 C<flags> is passed to C<SvPV_flags>, and usually should be
7396 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7402 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7403 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7404 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7409 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7416 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7418 start = (U8*)SvPV_flags(sv, len, flags);
7420 const U8 * const send = start + len;
7422 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7425 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7426 is 0, and *lenp is already set to that. */) {
7427 /* Convert the relative offset to absolute. */
7428 const STRLEN uoffset2 = uoffset + *lenp;
7429 const STRLEN boffset2
7430 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7431 uoffset, boffset) - boffset;
7445 =for apidoc sv_pos_u2b
7447 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7448 the start of the string, to a count of the equivalent number of bytes; if
7449 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7450 the offset, rather than from the start of the string. Handles magic and
7453 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7460 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7461 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7462 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7466 /* This function is subject to size and sign problems */
7469 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7471 PERL_ARGS_ASSERT_SV_POS_U2B;
7474 STRLEN ulen = (STRLEN)*lenp;
7475 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7476 SV_GMAGIC|SV_CONST_RETURN);
7479 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7480 SV_GMAGIC|SV_CONST_RETURN);
7485 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7488 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7489 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7492 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7493 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7494 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7498 (*mgp)->mg_len = ulen;
7501 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7502 byte length pairing. The (byte) length of the total SV is passed in too,
7503 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7504 may not have updated SvCUR, so we can't rely on reading it directly.
7506 The proffered utf8/byte length pairing isn't used if the cache already has
7507 two pairs, and swapping either for the proffered pair would increase the
7508 RMS of the intervals between known byte offsets.
7510 The cache itself consists of 4 STRLEN values
7511 0: larger UTF-8 offset
7512 1: corresponding byte offset
7513 2: smaller UTF-8 offset
7514 3: corresponding byte offset
7516 Unused cache pairs have the value 0, 0.
7517 Keeping the cache "backwards" means that the invariant of
7518 cache[0] >= cache[2] is maintained even with empty slots, which means that
7519 the code that uses it doesn't need to worry if only 1 entry has actually
7520 been set to non-zero. It also makes the "position beyond the end of the
7521 cache" logic much simpler, as the first slot is always the one to start
7525 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7526 const STRLEN utf8, const STRLEN blen)
7530 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7535 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7536 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7537 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7539 (*mgp)->mg_len = -1;
7543 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7544 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7545 (*mgp)->mg_ptr = (char *) cache;
7549 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7550 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7551 a pointer. Note that we no longer cache utf8 offsets on refer-
7552 ences, but this check is still a good idea, for robustness. */
7553 const U8 *start = (const U8 *) SvPVX_const(sv);
7554 const STRLEN realutf8 = utf8_length(start, start + byte);
7556 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7560 /* Cache is held with the later position first, to simplify the code
7561 that deals with unbounded ends. */
7563 ASSERT_UTF8_CACHE(cache);
7564 if (cache[1] == 0) {
7565 /* Cache is totally empty */
7568 } else if (cache[3] == 0) {
7569 if (byte > cache[1]) {
7570 /* New one is larger, so goes first. */
7571 cache[2] = cache[0];
7572 cache[3] = cache[1];
7580 /* float casts necessary? XXX */
7581 #define THREEWAY_SQUARE(a,b,c,d) \
7582 ((float)((d) - (c))) * ((float)((d) - (c))) \
7583 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7584 + ((float)((b) - (a))) * ((float)((b) - (a)))
7586 /* Cache has 2 slots in use, and we know three potential pairs.
7587 Keep the two that give the lowest RMS distance. Do the
7588 calculation in bytes simply because we always know the byte
7589 length. squareroot has the same ordering as the positive value,
7590 so don't bother with the actual square root. */
7591 if (byte > cache[1]) {
7592 /* New position is after the existing pair of pairs. */
7593 const float keep_earlier
7594 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7595 const float keep_later
7596 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7598 if (keep_later < keep_earlier) {
7599 cache[2] = cache[0];
7600 cache[3] = cache[1];
7606 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7607 float b, c, keep_earlier;
7608 if (byte > cache[3]) {
7609 /* New position is between the existing pair of pairs. */
7610 b = (float)cache[3];
7613 /* New position is before the existing pair of pairs. */
7615 c = (float)cache[3];
7617 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7618 if (byte > cache[3]) {
7619 if (keep_later < keep_earlier) {
7629 if (! (keep_later < keep_earlier)) {
7630 cache[0] = cache[2];
7631 cache[1] = cache[3];
7638 ASSERT_UTF8_CACHE(cache);
7641 /* We already know all of the way, now we may be able to walk back. The same
7642 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7643 backward is half the speed of walking forward. */
7645 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7646 const U8 *end, STRLEN endu)
7648 const STRLEN forw = target - s;
7649 STRLEN backw = end - target;
7651 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7653 if (forw < 2 * backw) {
7654 return utf8_length(s, target);
7657 while (end > target) {
7659 while (UTF8_IS_CONTINUATION(*end)) {
7668 =for apidoc sv_pos_b2u_flags
7670 Converts C<offset> from a count of bytes from the start of the string, to
7671 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7672 C<flags> is passed to C<SvPV_flags>, and usually should be
7673 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7679 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7680 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7685 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7688 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7694 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7696 s = (const U8*)SvPV_flags(sv, blen, flags);
7699 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7700 ", byte=%" UVuf, (UV)blen, (UV)offset);
7706 && SvTYPE(sv) >= SVt_PVMG
7707 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7710 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7711 if (cache[1] == offset) {
7712 /* An exact match. */
7715 if (cache[3] == offset) {
7716 /* An exact match. */
7720 if (cache[1] < offset) {
7721 /* We already know part of the way. */
7722 if (mg->mg_len != -1) {
7723 /* Actually, we know the end too. */
7725 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7726 s + blen, mg->mg_len - cache[0]);
7728 len = cache[0] + utf8_length(s + cache[1], send);
7731 else if (cache[3] < offset) {
7732 /* We're between the two cached pairs, so we do the calculation
7733 offset by the byte/utf-8 positions for the earlier pair,
7734 then add the utf-8 characters from the string start to
7736 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7737 s + cache[1], cache[0] - cache[2])
7741 else { /* cache[3] > offset */
7742 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7746 ASSERT_UTF8_CACHE(cache);
7748 } else if (mg->mg_len != -1) {
7749 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7753 if (!found || PL_utf8cache < 0) {
7754 const STRLEN real_len = utf8_length(s, send);
7756 if (found && PL_utf8cache < 0)
7757 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7763 utf8_mg_len_cache_update(sv, &mg, len);
7765 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7772 =for apidoc sv_pos_b2u
7774 Converts the value pointed to by C<offsetp> from a count of bytes from the
7775 start of the string, to a count of the equivalent number of UTF-8 chars.
7776 Handles magic and type coercion.
7778 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7785 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7786 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7791 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7793 PERL_ARGS_ASSERT_SV_POS_B2U;
7798 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7799 SV_GMAGIC|SV_CONST_RETURN);
7803 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7804 STRLEN real, SV *const sv)
7806 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7808 /* As this is debugging only code, save space by keeping this test here,
7809 rather than inlining it in all the callers. */
7810 if (from_cache == real)
7813 /* Need to turn the assertions off otherwise we may recurse infinitely
7814 while printing error messages. */
7815 SAVEI8(PL_utf8cache);
7817 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7818 func, (UV) from_cache, (UV) real, SVfARG(sv));
7824 Returns a boolean indicating whether the strings in the two SVs are
7825 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7826 coerce its args to strings if necessary.
7828 =for apidoc sv_eq_flags
7830 Returns a boolean indicating whether the strings in the two SVs are
7831 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7832 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7838 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7845 SV* svrecode = NULL;
7852 /* if pv1 and pv2 are the same, second SvPV_const call may
7853 * invalidate pv1 (if we are handling magic), so we may need to
7855 if (sv1 == sv2 && flags & SV_GMAGIC
7856 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7857 pv1 = SvPV_const(sv1, cur1);
7858 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7860 pv1 = SvPV_flags_const(sv1, cur1, flags);
7868 pv2 = SvPV_flags_const(sv2, cur2, flags);
7870 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7871 /* Differing utf8ness. */
7873 /* sv1 is the UTF-8 one */
7874 return bytes_cmp_utf8((const U8*)pv2, cur2,
7875 (const U8*)pv1, cur1) == 0;
7878 /* sv2 is the UTF-8 one */
7879 return bytes_cmp_utf8((const U8*)pv1, cur1,
7880 (const U8*)pv2, cur2) == 0;
7885 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7887 SvREFCNT_dec(svrecode);
7895 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7896 string in C<sv1> is less than, equal to, or greater than the string in
7897 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7898 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7900 =for apidoc sv_cmp_flags
7902 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7903 string in C<sv1> is less than, equal to, or greater than the string in
7904 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7905 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7906 also C<L</sv_cmp_locale_flags>>.
7912 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7914 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7918 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7922 const char *pv1, *pv2;
7924 SV *svrecode = NULL;
7931 pv1 = SvPV_flags_const(sv1, cur1, flags);
7938 pv2 = SvPV_flags_const(sv2, cur2, flags);
7940 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7941 /* Differing utf8ness. */
7943 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7944 (const U8*)pv1, cur1);
7945 return retval ? retval < 0 ? -1 : +1 : 0;
7948 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7949 (const U8*)pv2, cur2);
7950 return retval ? retval < 0 ? -1 : +1 : 0;
7954 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7957 cmp = cur2 ? -1 : 0;
7961 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7964 if (! DO_UTF8(sv1)) {
7966 const I32 retval = memcmp((const void*)pv1,
7970 cmp = retval < 0 ? -1 : 1;
7971 } else if (cur1 == cur2) {
7974 cmp = cur1 < cur2 ? -1 : 1;
7978 else { /* Both are to be treated as UTF-EBCDIC */
7980 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7981 * which remaps code points 0-255. We therefore generally have to
7982 * unmap back to the original values to get an accurate comparison.
7983 * But we don't have to do that for UTF-8 invariants, as by
7984 * definition, they aren't remapped, nor do we have to do it for
7985 * above-latin1 code points, as they also aren't remapped. (This
7986 * code also works on ASCII platforms, but the memcmp() above is
7989 const char *e = pv1 + shortest_len;
7991 /* Find the first bytes that differ between the two strings */
7992 while (pv1 < e && *pv1 == *pv2) {
7998 if (pv1 == e) { /* Are the same all the way to the end */
8002 cmp = cur1 < cur2 ? -1 : 1;
8005 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8006 * in the strings were. The current bytes may or may not be
8007 * at the beginning of a character. But neither or both are
8008 * (or else earlier bytes would have been different). And
8009 * if we are in the middle of a character, the two
8010 * characters are comprised of the same number of bytes
8011 * (because in this case the start bytes are the same, and
8012 * the start bytes encode the character's length). */
8013 if (UTF8_IS_INVARIANT(*pv1))
8015 /* If both are invariants; can just compare directly */
8016 if (UTF8_IS_INVARIANT(*pv2)) {
8017 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8019 else /* Since *pv1 is invariant, it is the whole character,
8020 which means it is at the beginning of a character.
8021 That means pv2 is also at the beginning of a
8022 character (see earlier comment). Since it isn't
8023 invariant, it must be a start byte. If it starts a
8024 character whose code point is above 255, that
8025 character is greater than any single-byte char, which
8027 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8032 /* Here, pv2 points to a character composed of 2 bytes
8033 * whose code point is < 256. Get its code point and
8034 * compare with *pv1 */
8035 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8040 else /* The code point starting at pv1 isn't a single byte */
8041 if (UTF8_IS_INVARIANT(*pv2))
8043 /* But here, the code point starting at *pv2 is a single byte,
8044 * and so *pv1 must begin a character, hence is a start byte.
8045 * If that character is above 255, it is larger than any
8046 * single-byte char, which *pv2 is */
8047 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8051 /* Here, pv1 points to a character composed of 2 bytes
8052 * whose code point is < 256. Get its code point and
8053 * compare with the single byte character *pv2 */
8054 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8059 else /* Here, we've ruled out either *pv1 and *pv2 being
8060 invariant. That means both are part of variants, but not
8061 necessarily at the start of a character */
8062 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8063 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8065 /* Here, at least one is the start of a character, which means
8066 * the other is also a start byte. And the code point of at
8067 * least one of the characters is above 255. It is a
8068 * characteristic of UTF-EBCDIC that all start bytes for
8069 * above-latin1 code points are well behaved as far as code
8070 * point comparisons go, and all are larger than all other
8071 * start bytes, so the comparison with those is also well
8073 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8076 /* Here both *pv1 and *pv2 are part of variant characters.
8077 * They could be both continuations, or both start characters.
8078 * (One or both could even be an illegal start character (for
8079 * an overlong) which for the purposes of sorting we treat as
8081 if (UTF8_IS_CONTINUATION(*pv1)) {
8083 /* If they are continuations for code points above 255,
8084 * then comparing the current byte is sufficient, as there
8085 * is no remapping of these and so the comparison is
8086 * well-behaved. We determine if they are such
8087 * continuations by looking at the preceding byte. It
8088 * could be a start byte, from which we can tell if it is
8089 * for an above 255 code point. Or it could be a
8090 * continuation, which means the character occupies at
8091 * least 3 bytes, so must be above 255. */
8092 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8093 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8095 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8099 /* Here, the continuations are for code points below 256;
8100 * back up one to get to the start byte */
8105 /* We need to get the actual native code point of each of these
8106 * variants in order to compare them */
8107 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8108 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8117 SvREFCNT_dec(svrecode);
8123 =for apidoc sv_cmp_locale
8125 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8126 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8127 if necessary. See also C<L</sv_cmp>>.
8129 =for apidoc sv_cmp_locale_flags
8131 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8132 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8133 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8134 C<L</sv_cmp_flags>>.
8140 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8142 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8146 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8149 #ifdef USE_LOCALE_COLLATE
8155 if (PL_collation_standard)
8160 /* Revert to using raw compare if both operands exist, but either one
8161 * doesn't transform properly for collation */
8163 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8167 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8173 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8174 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8177 if (!pv1 || !len1) {
8188 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8191 return retval < 0 ? -1 : 1;
8194 * When the result of collation is equality, that doesn't mean
8195 * that there are no differences -- some locales exclude some
8196 * characters from consideration. So to avoid false equalities,
8197 * we use the raw string as a tiebreaker.
8204 PERL_UNUSED_ARG(flags);
8205 #endif /* USE_LOCALE_COLLATE */
8207 return sv_cmp(sv1, sv2);
8211 #ifdef USE_LOCALE_COLLATE
8214 =for apidoc sv_collxfrm
8216 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8217 C<L</sv_collxfrm_flags>>.
8219 =for apidoc sv_collxfrm_flags
8221 Add Collate Transform magic to an SV if it doesn't already have it. If the
8222 flags contain C<SV_GMAGIC>, it handles get-magic.
8224 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8225 scalar data of the variable, but transformed to such a format that a normal
8226 memory comparison can be used to compare the data according to the locale
8233 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8237 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8239 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8241 /* If we don't have collation magic on 'sv', or the locale has changed
8242 * since the last time we calculated it, get it and save it now */
8243 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8248 /* Free the old space */
8250 Safefree(mg->mg_ptr);
8252 s = SvPV_flags_const(sv, len, flags);
8253 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8255 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8270 if (mg && mg->mg_ptr) {
8272 return mg->mg_ptr + sizeof(PL_collation_ix);
8280 #endif /* USE_LOCALE_COLLATE */
8283 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8285 SV * const tsv = newSV(0);
8288 sv_gets(tsv, fp, 0);
8289 sv_utf8_upgrade_nomg(tsv);
8290 SvCUR_set(sv,append);
8293 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8297 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8300 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8301 /* Grab the size of the record we're getting */
8302 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8309 /* With a true, record-oriented file on VMS, we need to use read directly
8310 * to ensure that we respect RMS record boundaries. The user is responsible
8311 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8312 * record size) field. N.B. This is likely to produce invalid results on
8313 * varying-width character data when a record ends mid-character.
8315 fd = PerlIO_fileno(fp);
8317 && PerlLIO_fstat(fd, &st) == 0
8318 && (st.st_fab_rfm == FAB$C_VAR
8319 || st.st_fab_rfm == FAB$C_VFC
8320 || st.st_fab_rfm == FAB$C_FIX)) {
8322 bytesread = PerlLIO_read(fd, buffer, recsize);
8324 else /* in-memory file from PerlIO::Scalar
8325 * or not a record-oriented file
8329 bytesread = PerlIO_read(fp, buffer, recsize);
8331 /* At this point, the logic in sv_get() means that sv will
8332 be treated as utf-8 if the handle is utf8.
8334 if (PerlIO_isutf8(fp) && bytesread > 0) {
8335 char *bend = buffer + bytesread;
8336 char *bufp = buffer;
8337 size_t charcount = 0;
8338 bool charstart = TRUE;
8341 while (charcount < recsize) {
8342 /* count accumulated characters */
8343 while (bufp < bend) {
8345 skip = UTF8SKIP(bufp);
8347 if (bufp + skip > bend) {
8348 /* partial at the end */
8359 if (charcount < recsize) {
8361 STRLEN bufp_offset = bufp - buffer;
8362 SSize_t morebytesread;
8364 /* originally I read enough to fill any incomplete
8365 character and the first byte of the next
8366 character if needed, but if there's many
8367 multi-byte encoded characters we're going to be
8368 making a read call for every character beyond
8369 the original read size.
8371 So instead, read the rest of the character if
8372 any, and enough bytes to match at least the
8373 start bytes for each character we're going to
8377 readsize = recsize - charcount;
8379 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8380 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8381 bend = buffer + bytesread;
8382 morebytesread = PerlIO_read(fp, bend, readsize);
8383 if (morebytesread <= 0) {
8384 /* we're done, if we still have incomplete
8385 characters the check code in sv_gets() will
8388 I'd originally considered doing
8389 PerlIO_ungetc() on all but the lead
8390 character of the incomplete character, but
8391 read() doesn't do that, so I don't.
8396 /* prepare to scan some more */
8397 bytesread += morebytesread;
8398 bend = buffer + bytesread;
8399 bufp = buffer + bufp_offset;
8407 SvCUR_set(sv, bytesread + append);
8408 buffer[bytesread] = '\0';
8409 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8415 Get a line from the filehandle and store it into the SV, optionally
8416 appending to the currently-stored string. If C<append> is not 0, the
8417 line is appended to the SV instead of overwriting it. C<append> should
8418 be set to the byte offset that the appended string should start at
8419 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8425 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8435 PERL_ARGS_ASSERT_SV_GETS;
8437 if (SvTHINKFIRST(sv))
8438 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8439 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8441 However, perlbench says it's slower, because the existing swipe code
8442 is faster than copy on write.
8443 Swings and roundabouts. */
8444 SvUPGRADE(sv, SVt_PV);
8447 /* line is going to be appended to the existing buffer in the sv */
8448 if (PerlIO_isutf8(fp)) {
8450 sv_utf8_upgrade_nomg(sv);
8451 sv_pos_u2b(sv,&append,0);
8453 } else if (SvUTF8(sv)) {
8454 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8460 /* not appending - "clear" the string by setting SvCUR to 0,
8461 * the pv is still avaiable. */
8464 if (PerlIO_isutf8(fp))
8467 if (IN_PERL_COMPILETIME) {
8468 /* we always read code in line mode */
8472 else if (RsSNARF(PL_rs)) {
8473 /* If it is a regular disk file use size from stat() as estimate
8474 of amount we are going to read -- may result in mallocing
8475 more memory than we really need if the layers below reduce
8476 the size we read (e.g. CRLF or a gzip layer).
8479 int fd = PerlIO_fileno(fp);
8480 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8481 const Off_t offset = PerlIO_tell(fp);
8482 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8483 #ifdef PERL_COPY_ON_WRITE
8484 /* Add an extra byte for the sake of copy-on-write's
8485 * buffer reference count. */
8486 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8488 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8495 else if (RsRECORD(PL_rs)) {
8496 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8498 else if (RsPARA(PL_rs)) {
8504 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8505 if (PerlIO_isutf8(fp)) {
8506 rsptr = SvPVutf8(PL_rs, rslen);
8509 if (SvUTF8(PL_rs)) {
8510 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8511 Perl_croak(aTHX_ "Wide character in $/");
8514 /* extract the raw pointer to the record separator */
8515 rsptr = SvPV_const(PL_rs, rslen);
8519 /* rslast is the last character in the record separator
8520 * note we don't use rslast except when rslen is true, so the
8521 * null assign is a placeholder. */
8522 rslast = rslen ? rsptr[rslen - 1] : '\0';
8524 if (rspara) { /* have to do this both before and after */
8525 do { /* to make sure file boundaries work right */
8528 i = PerlIO_getc(fp);
8532 PerlIO_ungetc(fp,i);
8538 /* See if we know enough about I/O mechanism to cheat it ! */
8540 /* This used to be #ifdef test - it is made run-time test for ease
8541 of abstracting out stdio interface. One call should be cheap
8542 enough here - and may even be a macro allowing compile
8546 if (PerlIO_fast_gets(fp)) {
8548 * We can do buffer based IO operations on this filehandle.
8550 * This means we can bypass a lot of subcalls and process
8551 * the buffer directly, it also means we know the upper bound
8552 * on the amount of data we might read of the current buffer
8553 * into our sv. Knowing this allows us to preallocate the pv
8554 * to be able to hold that maximum, which allows us to simplify
8555 * a lot of logic. */
8558 * We're going to steal some values from the stdio struct
8559 * and put EVERYTHING in the innermost loop into registers.
8561 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8562 STRLEN bpx; /* length of the data in the target sv
8563 used to fix pointers after a SvGROW */
8564 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8565 of data left in the read-ahead buffer.
8566 If 0 then the pv buffer can hold the full
8567 amount left, otherwise this is the amount it
8570 /* Here is some breathtakingly efficient cheating */
8572 /* When you read the following logic resist the urge to think
8573 * of record separators that are 1 byte long. They are an
8574 * uninteresting special (simple) case.
8576 * Instead think of record separators which are at least 2 bytes
8577 * long, and keep in mind that we need to deal with such
8578 * separators when they cross a read-ahead buffer boundary.
8580 * Also consider that we need to gracefully deal with separators
8581 * that may be longer than a single read ahead buffer.
8583 * Lastly do not forget we want to copy the delimiter as well. We
8584 * are copying all data in the file _up_to_and_including_ the separator
8587 * Now that you have all that in mind here is what is happening below:
8589 * 1. When we first enter the loop we do some memory book keeping to see
8590 * how much free space there is in the target SV. (This sub assumes that
8591 * it is operating on the same SV most of the time via $_ and that it is
8592 * going to be able to reuse the same pv buffer each call.) If there is
8593 * "enough" room then we set "shortbuffered" to how much space there is
8594 * and start reading forward.
8596 * 2. When we scan forward we copy from the read-ahead buffer to the target
8597 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8598 * and the end of the of pv, as well as for the "rslast", which is the last
8599 * char of the separator.
8601 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8602 * (which has a "complete" record up to the point we saw rslast) and check
8603 * it to see if it matches the separator. If it does we are done. If it doesn't
8604 * we continue on with the scan/copy.
8606 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8607 * the IO system to read the next buffer. We do this by doing a getc(), which
8608 * returns a single char read (or EOF), and prefills the buffer, and also
8609 * allows us to find out how full the buffer is. We use this information to
8610 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8611 * the returned single char into the target sv, and then go back into scan
8614 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8615 * remaining space in the read-buffer.
8617 * Note that this code despite its twisty-turny nature is pretty darn slick.
8618 * It manages single byte separators, multi-byte cross boundary separators,
8619 * and cross-read-buffer separators cleanly and efficiently at the cost
8620 * of potentially greatly overallocating the target SV.
8626 /* get the number of bytes remaining in the read-ahead buffer
8627 * on first call on a given fp this will return 0.*/
8628 cnt = PerlIO_get_cnt(fp);
8630 /* make sure we have the room */
8631 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8632 /* Not room for all of it
8633 if we are looking for a separator and room for some
8635 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8636 /* just process what we have room for */
8637 shortbuffered = cnt - SvLEN(sv) + append + 1;
8638 cnt -= shortbuffered;
8641 /* ensure that the target sv has enough room to hold
8642 * the rest of the read-ahead buffer */
8644 /* remember that cnt can be negative */
8645 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8649 /* we have enough room to hold the full buffer, lets scream */
8653 /* extract the pointer to sv's string buffer, offset by append as necessary */
8654 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8655 /* extract the point to the read-ahead buffer */
8656 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8658 /* some trace debug output */
8659 DEBUG_P(PerlIO_printf(Perl_debug_log,
8660 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8661 DEBUG_P(PerlIO_printf(Perl_debug_log,
8662 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8664 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8665 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8669 /* if there is stuff left in the read-ahead buffer */
8671 /* if there is a separator */
8673 /* find next rslast */
8676 /* shortcut common case of blank line */
8678 if ((*bp++ = *ptr++) == rslast)
8679 goto thats_all_folks;
8681 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8683 SSize_t got = p - ptr + 1;
8684 Copy(ptr, bp, got, STDCHAR);
8688 goto thats_all_folks;
8690 Copy(ptr, bp, cnt, STDCHAR);
8696 /* no separator, slurp the full buffer */
8697 Copy(ptr, bp, cnt, char); /* this | eat */
8698 bp += cnt; /* screams | dust */
8699 ptr += cnt; /* louder | sed :-) */
8701 assert (!shortbuffered);
8702 goto cannot_be_shortbuffered;
8706 if (shortbuffered) { /* oh well, must extend */
8707 /* we didnt have enough room to fit the line into the target buffer
8708 * so we must extend the target buffer and keep going */
8709 cnt = shortbuffered;
8711 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8713 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8714 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8715 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8719 cannot_be_shortbuffered:
8720 /* we need to refill the read-ahead buffer if possible */
8722 DEBUG_P(PerlIO_printf(Perl_debug_log,
8723 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8724 PTR2UV(ptr),(IV)cnt));
8725 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8727 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8728 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8729 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8730 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8733 call PerlIO_getc() to let it prefill the lookahead buffer
8735 This used to call 'filbuf' in stdio form, but as that behaves like
8736 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8737 another abstraction.
8739 Note we have to deal with the char in 'i' if we are not at EOF
8741 i = PerlIO_getc(fp); /* get more characters */
8743 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8744 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8745 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8746 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8748 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8749 cnt = PerlIO_get_cnt(fp);
8750 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8751 DEBUG_P(PerlIO_printf(Perl_debug_log,
8752 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8753 PTR2UV(ptr),(IV)cnt));
8755 if (i == EOF) /* all done for ever? */
8756 goto thats_really_all_folks;
8758 /* make sure we have enough space in the target sv */
8759 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8761 SvGROW(sv, bpx + cnt + 2);
8762 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8764 /* copy of the char we got from getc() */
8765 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8767 /* make sure we deal with the i being the last character of a separator */
8768 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8769 goto thats_all_folks;
8773 /* check if we have actually found the separator - only really applies
8775 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8776 memNE((char*)bp - rslen, rsptr, rslen))
8777 goto screamer; /* go back to the fray */
8778 thats_really_all_folks:
8780 cnt += shortbuffered;
8781 DEBUG_P(PerlIO_printf(Perl_debug_log,
8782 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8783 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8784 DEBUG_P(PerlIO_printf(Perl_debug_log,
8785 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8787 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8788 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8790 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8791 DEBUG_P(PerlIO_printf(Perl_debug_log,
8792 "Screamer: done, len=%ld, string=|%.*s|\n",
8793 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8797 /*The big, slow, and stupid way. */
8798 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8799 STDCHAR *buf = NULL;
8800 Newx(buf, 8192, STDCHAR);
8808 const STDCHAR * const bpe = buf + sizeof(buf);
8810 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8811 ; /* keep reading */
8815 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8816 /* Accommodate broken VAXC compiler, which applies U8 cast to
8817 * both args of ?: operator, causing EOF to change into 255
8820 i = (U8)buf[cnt - 1];
8826 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8828 sv_catpvn_nomg(sv, (char *) buf, cnt);
8830 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8832 if (i != EOF && /* joy */
8834 SvCUR(sv) < rslen ||
8835 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8839 * If we're reading from a TTY and we get a short read,
8840 * indicating that the user hit his EOF character, we need
8841 * to notice it now, because if we try to read from the TTY
8842 * again, the EOF condition will disappear.
8844 * The comparison of cnt to sizeof(buf) is an optimization
8845 * that prevents unnecessary calls to feof().
8849 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8853 #ifdef USE_HEAP_INSTEAD_OF_STACK
8858 if (rspara) { /* have to do this both before and after */
8859 while (i != EOF) { /* to make sure file boundaries work right */
8860 i = PerlIO_getc(fp);
8862 PerlIO_ungetc(fp,i);
8868 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8874 Auto-increment of the value in the SV, doing string to numeric conversion
8875 if necessary. Handles 'get' magic and operator overloading.
8881 Perl_sv_inc(pTHX_ SV *const sv)
8890 =for apidoc sv_inc_nomg
8892 Auto-increment of the value in the SV, doing string to numeric conversion
8893 if necessary. Handles operator overloading. Skips handling 'get' magic.
8899 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8906 if (SvTHINKFIRST(sv)) {
8907 if (SvREADONLY(sv)) {
8908 Perl_croak_no_modify();
8912 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8914 i = PTR2IV(SvRV(sv));
8918 else sv_force_normal_flags(sv, 0);
8920 flags = SvFLAGS(sv);
8921 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8922 /* It's (privately or publicly) a float, but not tested as an
8923 integer, so test it to see. */
8925 flags = SvFLAGS(sv);
8927 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8928 /* It's publicly an integer, or privately an integer-not-float */
8929 #ifdef PERL_PRESERVE_IVUV
8933 if (SvUVX(sv) == UV_MAX)
8934 sv_setnv(sv, UV_MAX_P1);
8936 (void)SvIOK_only_UV(sv);
8937 SvUV_set(sv, SvUVX(sv) + 1);
8939 if (SvIVX(sv) == IV_MAX)
8940 sv_setuv(sv, (UV)IV_MAX + 1);
8942 (void)SvIOK_only(sv);
8943 SvIV_set(sv, SvIVX(sv) + 1);
8948 if (flags & SVp_NOK) {
8949 const NV was = SvNVX(sv);
8950 if (LIKELY(!Perl_isinfnan(was)) &&
8951 NV_OVERFLOWS_INTEGERS_AT &&
8952 was >= NV_OVERFLOWS_INTEGERS_AT) {
8953 /* diag_listed_as: Lost precision when %s %f by 1 */
8954 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8955 "Lost precision when incrementing %" NVff " by 1",
8958 (void)SvNOK_only(sv);
8959 SvNV_set(sv, was + 1.0);
8963 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8964 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8965 Perl_croak_no_modify();
8967 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8968 if ((flags & SVTYPEMASK) < SVt_PVIV)
8969 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8970 (void)SvIOK_only(sv);
8975 while (isALPHA(*d)) d++;
8976 while (isDIGIT(*d)) d++;
8977 if (d < SvEND(sv)) {
8978 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8979 #ifdef PERL_PRESERVE_IVUV
8980 /* Got to punt this as an integer if needs be, but we don't issue
8981 warnings. Probably ought to make the sv_iv_please() that does
8982 the conversion if possible, and silently. */
8983 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8984 /* Need to try really hard to see if it's an integer.
8985 9.22337203685478e+18 is an integer.
8986 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8987 so $a="9.22337203685478e+18"; $a+0; $a++
8988 needs to be the same as $a="9.22337203685478e+18"; $a++
8995 /* sv_2iv *should* have made this an NV */
8996 if (flags & SVp_NOK) {
8997 (void)SvNOK_only(sv);
8998 SvNV_set(sv, SvNVX(sv) + 1.0);
9001 /* I don't think we can get here. Maybe I should assert this
9002 And if we do get here I suspect that sv_setnv will croak. NWC
9004 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9005 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9007 #endif /* PERL_PRESERVE_IVUV */
9008 if (!numtype && ckWARN(WARN_NUMERIC))
9009 not_incrementable(sv);
9010 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9014 while (d >= SvPVX_const(sv)) {
9022 /* MKS: The original code here died if letters weren't consecutive.
9023 * at least it didn't have to worry about non-C locales. The
9024 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9025 * arranged in order (although not consecutively) and that only
9026 * [A-Za-z] are accepted by isALPHA in the C locale.
9028 if (isALPHA_FOLD_NE(*d, 'z')) {
9029 do { ++*d; } while (!isALPHA(*d));
9032 *(d--) -= 'z' - 'a';
9037 *(d--) -= 'z' - 'a' + 1;
9041 /* oh,oh, the number grew */
9042 SvGROW(sv, SvCUR(sv) + 2);
9043 SvCUR_set(sv, SvCUR(sv) + 1);
9044 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9055 Auto-decrement of the value in the SV, doing string to numeric conversion
9056 if necessary. Handles 'get' magic and operator overloading.
9062 Perl_sv_dec(pTHX_ SV *const sv)
9071 =for apidoc sv_dec_nomg
9073 Auto-decrement of the value in the SV, doing string to numeric conversion
9074 if necessary. Handles operator overloading. Skips handling 'get' magic.
9080 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9086 if (SvTHINKFIRST(sv)) {
9087 if (SvREADONLY(sv)) {
9088 Perl_croak_no_modify();
9092 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9094 i = PTR2IV(SvRV(sv));
9098 else sv_force_normal_flags(sv, 0);
9100 /* Unlike sv_inc we don't have to worry about string-never-numbers
9101 and keeping them magic. But we mustn't warn on punting */
9102 flags = SvFLAGS(sv);
9103 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9104 /* It's publicly an integer, or privately an integer-not-float */
9105 #ifdef PERL_PRESERVE_IVUV
9109 if (SvUVX(sv) == 0) {
9110 (void)SvIOK_only(sv);
9114 (void)SvIOK_only_UV(sv);
9115 SvUV_set(sv, SvUVX(sv) - 1);
9118 if (SvIVX(sv) == IV_MIN) {
9119 sv_setnv(sv, (NV)IV_MIN);
9123 (void)SvIOK_only(sv);
9124 SvIV_set(sv, SvIVX(sv) - 1);
9129 if (flags & SVp_NOK) {
9132 const NV was = SvNVX(sv);
9133 if (LIKELY(!Perl_isinfnan(was)) &&
9134 NV_OVERFLOWS_INTEGERS_AT &&
9135 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9136 /* diag_listed_as: Lost precision when %s %f by 1 */
9137 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9138 "Lost precision when decrementing %" NVff " by 1",
9141 (void)SvNOK_only(sv);
9142 SvNV_set(sv, was - 1.0);
9147 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9148 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9149 Perl_croak_no_modify();
9151 if (!(flags & SVp_POK)) {
9152 if ((flags & SVTYPEMASK) < SVt_PVIV)
9153 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9155 (void)SvIOK_only(sv);
9158 #ifdef PERL_PRESERVE_IVUV
9160 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9161 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9162 /* Need to try really hard to see if it's an integer.
9163 9.22337203685478e+18 is an integer.
9164 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9165 so $a="9.22337203685478e+18"; $a+0; $a--
9166 needs to be the same as $a="9.22337203685478e+18"; $a--
9173 /* sv_2iv *should* have made this an NV */
9174 if (flags & SVp_NOK) {
9175 (void)SvNOK_only(sv);
9176 SvNV_set(sv, SvNVX(sv) - 1.0);
9179 /* I don't think we can get here. Maybe I should assert this
9180 And if we do get here I suspect that sv_setnv will croak. NWC
9182 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9183 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9186 #endif /* PERL_PRESERVE_IVUV */
9187 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9190 /* this define is used to eliminate a chunk of duplicated but shared logic
9191 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9192 * used anywhere but here - yves
9194 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9196 SSize_t ix = ++PL_tmps_ix; \
9197 if (UNLIKELY(ix >= PL_tmps_max)) \
9198 ix = tmps_grow_p(ix); \
9199 PL_tmps_stack[ix] = (AnSv); \
9203 =for apidoc sv_mortalcopy
9205 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9206 The new SV is marked as mortal. It will be destroyed "soon", either by an
9207 explicit call to C<FREETMPS>, or by an implicit call at places such as
9208 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9213 /* Make a string that will exist for the duration of the expression
9214 * evaluation. Actually, it may have to last longer than that, but
9215 * hopefully we won't free it until it has been assigned to a
9216 * permanent location. */
9219 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9223 if (flags & SV_GMAGIC)
9224 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9226 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9227 PUSH_EXTEND_MORTAL__SV_C(sv);
9233 =for apidoc sv_newmortal
9235 Creates a new null SV which is mortal. The reference count of the SV is
9236 set to 1. It will be destroyed "soon", either by an explicit call to
9237 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9238 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9244 Perl_sv_newmortal(pTHX)
9249 SvFLAGS(sv) = SVs_TEMP;
9250 PUSH_EXTEND_MORTAL__SV_C(sv);
9256 =for apidoc newSVpvn_flags
9258 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9259 characters) into it. The reference count for the
9260 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9261 string. You are responsible for ensuring that the source string is at least
9262 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9263 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9264 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9265 returning. If C<SVf_UTF8> is set, C<s>
9266 is considered to be in UTF-8 and the
9267 C<SVf_UTF8> flag will be set on the new SV.
9268 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9270 #define newSVpvn_utf8(s, len, u) \
9271 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9277 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9281 /* All the flags we don't support must be zero.
9282 And we're new code so I'm going to assert this from the start. */
9283 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9285 sv_setpvn(sv,s,len);
9287 /* This code used to do a sv_2mortal(), however we now unroll the call to
9288 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9289 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9290 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9291 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9292 * means that we eliminate quite a few steps than it looks - Yves
9293 * (explaining patch by gfx) */
9295 SvFLAGS(sv) |= flags;
9297 if(flags & SVs_TEMP){
9298 PUSH_EXTEND_MORTAL__SV_C(sv);
9305 =for apidoc sv_2mortal
9307 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9308 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9309 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9310 string buffer can be "stolen" if this SV is copied. See also
9311 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9317 Perl_sv_2mortal(pTHX_ SV *const sv)
9324 PUSH_EXTEND_MORTAL__SV_C(sv);
9332 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9333 characters) into it. The reference count for the
9334 SV is set to 1. If C<len> is zero, Perl will compute the length using
9335 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9336 C<NUL> characters and has to have a terminating C<NUL> byte).
9338 This function can cause reliability issues if you are likely to pass in
9339 empty strings that are not null terminated, because it will run
9340 strlen on the string and potentially run past valid memory.
9342 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9343 For string literals use L</newSVpvs> instead. This function will work fine for
9344 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9345 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9351 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9356 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9361 =for apidoc newSVpvn
9363 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9364 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9365 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9366 are responsible for ensuring that the source buffer is at least
9367 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9374 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9378 sv_setpvn(sv,buffer,len);
9383 =for apidoc newSVhek
9385 Creates a new SV from the hash key structure. It will generate scalars that
9386 point to the shared string table where possible. Returns a new (undefined)
9387 SV if C<hek> is NULL.
9393 Perl_newSVhek(pTHX_ const HEK *const hek)
9402 if (HEK_LEN(hek) == HEf_SVKEY) {
9403 return newSVsv(*(SV**)HEK_KEY(hek));
9405 const int flags = HEK_FLAGS(hek);
9406 if (flags & HVhek_WASUTF8) {
9408 Andreas would like keys he put in as utf8 to come back as utf8
9410 STRLEN utf8_len = HEK_LEN(hek);
9411 SV * const sv = newSV_type(SVt_PV);
9412 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9413 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9414 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9417 } else if (flags & HVhek_UNSHARED) {
9418 /* A hash that isn't using shared hash keys has to have
9419 the flag in every key so that we know not to try to call
9420 share_hek_hek on it. */
9422 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9427 /* This will be overwhelminly the most common case. */
9429 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9430 more efficient than sharepvn(). */
9434 sv_upgrade(sv, SVt_PV);
9435 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9436 SvCUR_set(sv, HEK_LEN(hek));
9448 =for apidoc newSVpvn_share
9450 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9451 table. If the string does not already exist in the table, it is
9452 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9453 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9454 is non-zero, that value is used; otherwise the hash is computed.
9455 The string's hash can later be retrieved from the SV
9456 with the C<SvSHARED_HASH()> macro. The idea here is
9457 that as the string table is used for shared hash keys these strings will have
9458 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9464 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9468 bool is_utf8 = FALSE;
9469 const char *const orig_src = src;
9472 STRLEN tmplen = -len;
9474 /* See the note in hv.c:hv_fetch() --jhi */
9475 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9479 PERL_HASH(hash, src, len);
9481 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9482 changes here, update it there too. */
9483 sv_upgrade(sv, SVt_PV);
9484 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9491 if (src != orig_src)
9497 =for apidoc newSVpv_share
9499 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9506 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9508 return newSVpvn_share(src, strlen(src), hash);
9511 #if defined(PERL_IMPLICIT_CONTEXT)
9513 /* pTHX_ magic can't cope with varargs, so this is a no-context
9514 * version of the main function, (which may itself be aliased to us).
9515 * Don't access this version directly.
9519 Perl_newSVpvf_nocontext(const char *const pat, ...)
9525 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9527 va_start(args, pat);
9528 sv = vnewSVpvf(pat, &args);
9535 =for apidoc newSVpvf
9537 Creates a new SV and initializes it with the string formatted like
9544 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9549 PERL_ARGS_ASSERT_NEWSVPVF;
9551 va_start(args, pat);
9552 sv = vnewSVpvf(pat, &args);
9557 /* backend for newSVpvf() and newSVpvf_nocontext() */
9560 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9564 PERL_ARGS_ASSERT_VNEWSVPVF;
9567 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9574 Creates a new SV and copies a floating point value into it.
9575 The reference count for the SV is set to 1.
9581 Perl_newSVnv(pTHX_ const NV n)
9593 Creates a new SV and copies an integer into it. The reference count for the
9600 Perl_newSViv(pTHX_ const IV i)
9606 /* Inlining ONLY the small relevant subset of sv_setiv here
9607 * for performance. Makes a significant difference. */
9609 /* We're starting from SVt_FIRST, so provided that's
9610 * actual 0, we don't have to unset any SV type flags
9611 * to promote to SVt_IV. */
9612 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9614 SET_SVANY_FOR_BODYLESS_IV(sv);
9615 SvFLAGS(sv) |= SVt_IV;
9627 Creates a new SV and copies an unsigned integer into it.
9628 The reference count for the SV is set to 1.
9634 Perl_newSVuv(pTHX_ const UV u)
9638 /* Inlining ONLY the small relevant subset of sv_setuv here
9639 * for performance. Makes a significant difference. */
9641 /* Using ivs is more efficient than using uvs - see sv_setuv */
9642 if (u <= (UV)IV_MAX) {
9643 return newSViv((IV)u);
9648 /* We're starting from SVt_FIRST, so provided that's
9649 * actual 0, we don't have to unset any SV type flags
9650 * to promote to SVt_IV. */
9651 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9653 SET_SVANY_FOR_BODYLESS_IV(sv);
9654 SvFLAGS(sv) |= SVt_IV;
9656 (void)SvIsUV_on(sv);
9665 =for apidoc newSV_type
9667 Creates a new SV, of the type specified. The reference count for the new SV
9674 Perl_newSV_type(pTHX_ const svtype type)
9679 ASSUME(SvTYPE(sv) == SVt_FIRST);
9680 if(type != SVt_FIRST)
9681 sv_upgrade(sv, type);
9686 =for apidoc newRV_noinc
9688 Creates an RV wrapper for an SV. The reference count for the original
9689 SV is B<not> incremented.
9695 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9699 PERL_ARGS_ASSERT_NEWRV_NOINC;
9703 /* We're starting from SVt_FIRST, so provided that's
9704 * actual 0, we don't have to unset any SV type flags
9705 * to promote to SVt_IV. */
9706 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9708 SET_SVANY_FOR_BODYLESS_IV(sv);
9709 SvFLAGS(sv) |= SVt_IV;
9714 SvRV_set(sv, tmpRef);
9719 /* newRV_inc is the official function name to use now.
9720 * newRV_inc is in fact #defined to newRV in sv.h
9724 Perl_newRV(pTHX_ SV *const sv)
9726 PERL_ARGS_ASSERT_NEWRV;
9728 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9734 Creates a new SV which is an exact duplicate of the original SV.
9741 Perl_newSVsv(pTHX_ SV *const old)
9747 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9748 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9751 /* Do this here, otherwise we leak the new SV if this croaks. */
9754 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9755 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9756 sv_setsv_flags(sv, old, SV_NOSTEAL);
9761 =for apidoc sv_reset
9763 Underlying implementation for the C<reset> Perl function.
9764 Note that the perl-level function is vaguely deprecated.
9770 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9772 PERL_ARGS_ASSERT_SV_RESET;
9774 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9778 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9780 char todo[PERL_UCHAR_MAX+1];
9783 if (!stash || SvTYPE(stash) != SVt_PVHV)
9786 if (!s) { /* reset ?? searches */
9787 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9789 const U32 count = mg->mg_len / sizeof(PMOP**);
9790 PMOP **pmp = (PMOP**) mg->mg_ptr;
9791 PMOP *const *const end = pmp + count;
9795 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9797 (*pmp)->op_pmflags &= ~PMf_USED;
9805 /* reset variables */
9807 if (!HvARRAY(stash))
9810 Zero(todo, 256, char);
9814 I32 i = (unsigned char)*s;
9818 max = (unsigned char)*s++;
9819 for ( ; i <= max; i++) {
9822 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9824 for (entry = HvARRAY(stash)[i];
9826 entry = HeNEXT(entry))
9831 if (!todo[(U8)*HeKEY(entry)])
9833 gv = MUTABLE_GV(HeVAL(entry));
9837 if (sv && !SvREADONLY(sv)) {
9838 SV_CHECK_THINKFIRST_COW_DROP(sv);
9839 if (!isGV(sv)) SvOK_off(sv);
9844 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9855 Using various gambits, try to get an IO from an SV: the IO slot if its a
9856 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9857 named after the PV if we're a string.
9859 'Get' magic is ignored on the C<sv> passed in, but will be called on
9860 C<SvRV(sv)> if C<sv> is an RV.
9866 Perl_sv_2io(pTHX_ SV *const sv)
9871 PERL_ARGS_ASSERT_SV_2IO;
9873 switch (SvTYPE(sv)) {
9875 io = MUTABLE_IO(sv);
9879 if (isGV_with_GP(sv)) {
9880 gv = MUTABLE_GV(sv);
9883 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9884 HEKfARG(GvNAME_HEK(gv)));
9890 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9892 SvGETMAGIC(SvRV(sv));
9893 return sv_2io(SvRV(sv));
9895 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9902 if (SvGMAGICAL(sv)) {
9903 newsv = sv_newmortal();
9904 sv_setsv_nomg(newsv, sv);
9906 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9916 Using various gambits, try to get a CV from an SV; in addition, try if
9917 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9918 The flags in C<lref> are passed to C<gv_fetchsv>.
9924 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9929 PERL_ARGS_ASSERT_SV_2CV;
9936 switch (SvTYPE(sv)) {
9940 return MUTABLE_CV(sv);
9950 sv = amagic_deref_call(sv, to_cv_amg);
9953 if (SvTYPE(sv) == SVt_PVCV) {
9954 cv = MUTABLE_CV(sv);
9959 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9960 gv = MUTABLE_GV(sv);
9962 Perl_croak(aTHX_ "Not a subroutine reference");
9964 else if (isGV_with_GP(sv)) {
9965 gv = MUTABLE_GV(sv);
9968 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9975 /* Some flags to gv_fetchsv mean don't really create the GV */
9976 if (!isGV_with_GP(gv)) {
9981 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9982 /* XXX this is probably not what they think they're getting.
9983 * It has the same effect as "sub name;", i.e. just a forward
9994 Returns true if the SV has a true value by Perl's rules.
9995 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9996 instead use an in-line version.
10002 Perl_sv_true(pTHX_ SV *const sv)
10007 const XPV* const tXpv = (XPV*)SvANY(sv);
10009 (tXpv->xpv_cur > 1 ||
10010 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10017 return SvIVX(sv) != 0;
10020 return SvNVX(sv) != 0.0;
10022 return sv_2bool(sv);
10028 =for apidoc sv_pvn_force
10030 Get a sensible string out of the SV somehow.
10031 A private implementation of the C<SvPV_force> macro for compilers which
10032 can't cope with complex macro expressions. Always use the macro instead.
10034 =for apidoc sv_pvn_force_flags
10036 Get a sensible string out of the SV somehow.
10037 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10038 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10039 implemented in terms of this function.
10040 You normally want to use the various wrapper macros instead: see
10041 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10047 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10049 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10051 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10052 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10053 sv_force_normal_flags(sv, 0);
10063 if (SvTYPE(sv) > SVt_PVLV
10064 || isGV_with_GP(sv))
10065 /* diag_listed_as: Can't coerce %s to %s in %s */
10066 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10068 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10075 if (SvTYPE(sv) < SVt_PV ||
10076 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10079 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10080 SvGROW(sv, len + 1);
10081 Move(s,SvPVX(sv),len,char);
10082 SvCUR_set(sv, len);
10083 SvPVX(sv)[len] = '\0';
10086 SvPOK_on(sv); /* validate pointer */
10088 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10089 PTR2UV(sv),SvPVX_const(sv)));
10092 (void)SvPOK_only_UTF8(sv);
10093 return SvPVX_mutable(sv);
10097 =for apidoc sv_pvbyten_force
10099 The backend for the C<SvPVbytex_force> macro. Always use the macro
10106 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10108 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10110 sv_pvn_force(sv,lp);
10111 sv_utf8_downgrade(sv,0);
10117 =for apidoc sv_pvutf8n_force
10119 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10126 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10128 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10130 sv_pvn_force(sv,0);
10131 sv_utf8_upgrade_nomg(sv);
10137 =for apidoc sv_reftype
10139 Returns a string describing what the SV is a reference to.
10141 If ob is true and the SV is blessed, the string is the class name,
10142 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10148 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10150 PERL_ARGS_ASSERT_SV_REFTYPE;
10151 if (ob && SvOBJECT(sv)) {
10152 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10155 /* WARNING - There is code, for instance in mg.c, that assumes that
10156 * the only reason that sv_reftype(sv,0) would return a string starting
10157 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10158 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10159 * this routine inside other subs, and it saves time.
10160 * Do not change this assumption without searching for "dodgy type check" in
10163 switch (SvTYPE(sv)) {
10178 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10179 /* tied lvalues should appear to be
10180 * scalars for backwards compatibility */
10181 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10182 ? "SCALAR" : "LVALUE");
10183 case SVt_PVAV: return "ARRAY";
10184 case SVt_PVHV: return "HASH";
10185 case SVt_PVCV: return "CODE";
10186 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10187 ? "GLOB" : "SCALAR");
10188 case SVt_PVFM: return "FORMAT";
10189 case SVt_PVIO: return "IO";
10190 case SVt_INVLIST: return "INVLIST";
10191 case SVt_REGEXP: return "REGEXP";
10192 default: return "UNKNOWN";
10200 Returns a SV describing what the SV passed in is a reference to.
10202 dst can be a SV to be set to the description or NULL, in which case a
10203 mortal SV is returned.
10205 If ob is true and the SV is blessed, the description is the class
10206 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10212 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10214 PERL_ARGS_ASSERT_SV_REF;
10217 dst = sv_newmortal();
10219 if (ob && SvOBJECT(sv)) {
10220 HvNAME_get(SvSTASH(sv))
10221 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10222 : sv_setpvs(dst, "__ANON__");
10225 const char * reftype = sv_reftype(sv, 0);
10226 sv_setpv(dst, reftype);
10232 =for apidoc sv_isobject
10234 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10235 object. If the SV is not an RV, or if the object is not blessed, then this
10242 Perl_sv_isobject(pTHX_ SV *sv)
10258 Returns a boolean indicating whether the SV is blessed into the specified
10259 class. This does not check for subtypes; use C<sv_derived_from> to verify
10260 an inheritance relationship.
10266 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10268 const char *hvname;
10270 PERL_ARGS_ASSERT_SV_ISA;
10280 hvname = HvNAME_get(SvSTASH(sv));
10284 return strEQ(hvname, name);
10288 =for apidoc newSVrv
10290 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10291 RV then it will be upgraded to one. If C<classname> is non-null then the new
10292 SV will be blessed in the specified package. The new SV is returned and its
10293 reference count is 1. The reference count 1 is owned by C<rv>.
10299 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10303 PERL_ARGS_ASSERT_NEWSVRV;
10307 SV_CHECK_THINKFIRST_COW_DROP(rv);
10309 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10310 const U32 refcnt = SvREFCNT(rv);
10314 SvREFCNT(rv) = refcnt;
10316 sv_upgrade(rv, SVt_IV);
10317 } else if (SvROK(rv)) {
10318 SvREFCNT_dec(SvRV(rv));
10320 prepare_SV_for_RV(rv);
10328 HV* const stash = gv_stashpv(classname, GV_ADD);
10329 (void)sv_bless(rv, stash);
10335 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10337 SV * const lv = newSV_type(SVt_PVLV);
10338 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10340 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10341 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10342 LvSTARGOFF(lv) = ix;
10343 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10348 =for apidoc sv_setref_pv
10350 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10351 argument will be upgraded to an RV. That RV will be modified to point to
10352 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10353 into the SV. The C<classname> argument indicates the package for the
10354 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10355 will have a reference count of 1, and the RV will be returned.
10357 Do not use with other Perl types such as HV, AV, SV, CV, because those
10358 objects will become corrupted by the pointer copy process.
10360 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10366 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10368 PERL_ARGS_ASSERT_SV_SETREF_PV;
10375 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10380 =for apidoc sv_setref_iv
10382 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10383 argument will be upgraded to an RV. That RV will be modified to point to
10384 the new SV. The C<classname> argument indicates the package for the
10385 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10386 will have a reference count of 1, and the RV will be returned.
10392 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10394 PERL_ARGS_ASSERT_SV_SETREF_IV;
10396 sv_setiv(newSVrv(rv,classname), iv);
10401 =for apidoc sv_setref_uv
10403 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10404 argument will be upgraded to an RV. That RV will be modified to point to
10405 the new SV. The C<classname> argument indicates the package for the
10406 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10407 will have a reference count of 1, and the RV will be returned.
10413 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10415 PERL_ARGS_ASSERT_SV_SETREF_UV;
10417 sv_setuv(newSVrv(rv,classname), uv);
10422 =for apidoc sv_setref_nv
10424 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10425 argument will be upgraded to an RV. That RV will be modified to point to
10426 the new SV. The C<classname> argument indicates the package for the
10427 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10428 will have a reference count of 1, and the RV will be returned.
10434 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10436 PERL_ARGS_ASSERT_SV_SETREF_NV;
10438 sv_setnv(newSVrv(rv,classname), nv);
10443 =for apidoc sv_setref_pvn
10445 Copies a string into a new SV, optionally blessing the SV. The length of the
10446 string must be specified with C<n>. The C<rv> argument will be upgraded to
10447 an RV. That RV will be modified to point to the new SV. The C<classname>
10448 argument indicates the package for the blessing. Set C<classname> to
10449 C<NULL> to avoid the blessing. The new SV will have a reference count
10450 of 1, and the RV will be returned.
10452 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10458 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10459 const char *const pv, const STRLEN n)
10461 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10463 sv_setpvn(newSVrv(rv,classname), pv, n);
10468 =for apidoc sv_bless
10470 Blesses an SV into a specified package. The SV must be an RV. The package
10471 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10472 of the SV is unaffected.
10478 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10481 HV *oldstash = NULL;
10483 PERL_ARGS_ASSERT_SV_BLESS;
10487 Perl_croak(aTHX_ "Can't bless non-reference value");
10489 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10490 if (SvREADONLY(tmpRef))
10491 Perl_croak_no_modify();
10492 if (SvOBJECT(tmpRef)) {
10493 oldstash = SvSTASH(tmpRef);
10496 SvOBJECT_on(tmpRef);
10497 SvUPGRADE(tmpRef, SVt_PVMG);
10498 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10499 SvREFCNT_dec(oldstash);
10501 if(SvSMAGICAL(tmpRef))
10502 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10510 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10511 * as it is after unglobbing it.
10514 PERL_STATIC_INLINE void
10515 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10519 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10521 PERL_ARGS_ASSERT_SV_UNGLOB;
10523 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10525 if (!(flags & SV_COW_DROP_PV))
10526 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10528 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10530 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10531 && HvNAME_get(stash))
10532 mro_method_changed_in(stash);
10533 gp_free(MUTABLE_GV(sv));
10536 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10537 GvSTASH(sv) = NULL;
10540 if (GvNAME_HEK(sv)) {
10541 unshare_hek(GvNAME_HEK(sv));
10543 isGV_with_GP_off(sv);
10545 if(SvTYPE(sv) == SVt_PVGV) {
10546 /* need to keep SvANY(sv) in the right arena */
10547 xpvmg = new_XPVMG();
10548 StructCopy(SvANY(sv), xpvmg, XPVMG);
10549 del_XPVGV(SvANY(sv));
10552 SvFLAGS(sv) &= ~SVTYPEMASK;
10553 SvFLAGS(sv) |= SVt_PVMG;
10556 /* Intentionally not calling any local SET magic, as this isn't so much a
10557 set operation as merely an internal storage change. */
10558 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10559 else sv_setsv_flags(sv, temp, 0);
10561 if ((const GV *)sv == PL_last_in_gv)
10562 PL_last_in_gv = NULL;
10563 else if ((const GV *)sv == PL_statgv)
10568 =for apidoc sv_unref_flags
10570 Unsets the RV status of the SV, and decrements the reference count of
10571 whatever was being referenced by the RV. This can almost be thought of
10572 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10573 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10574 (otherwise the decrementing is conditional on the reference count being
10575 different from one or the reference being a readonly SV).
10576 See C<L</SvROK_off>>.
10582 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10584 SV* const target = SvRV(ref);
10586 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10588 if (SvWEAKREF(ref)) {
10589 sv_del_backref(target, ref);
10590 SvWEAKREF_off(ref);
10591 SvRV_set(ref, NULL);
10594 SvRV_set(ref, NULL);
10596 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10597 assigned to as BEGIN {$a = \"Foo"} will fail. */
10598 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10599 SvREFCNT_dec_NN(target);
10600 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10601 sv_2mortal(target); /* Schedule for freeing later */
10605 =for apidoc sv_untaint
10607 Untaint an SV. Use C<SvTAINTED_off> instead.
10613 Perl_sv_untaint(pTHX_ SV *const sv)
10615 PERL_ARGS_ASSERT_SV_UNTAINT;
10616 PERL_UNUSED_CONTEXT;
10618 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10619 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10626 =for apidoc sv_tainted
10628 Test an SV for taintedness. Use C<SvTAINTED> instead.
10634 Perl_sv_tainted(pTHX_ SV *const sv)
10636 PERL_ARGS_ASSERT_SV_TAINTED;
10637 PERL_UNUSED_CONTEXT;
10639 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10640 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10641 if (mg && (mg->mg_len & 1) )
10647 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10648 private to this file */
10651 =for apidoc sv_setpviv
10653 Copies an integer into the given SV, also updating its string value.
10654 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10660 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10662 char buf[TYPE_CHARS(UV)];
10664 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10666 PERL_ARGS_ASSERT_SV_SETPVIV;
10668 sv_setpvn(sv, ptr, ebuf - ptr);
10672 =for apidoc sv_setpviv_mg
10674 Like C<sv_setpviv>, but also handles 'set' magic.
10680 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10682 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10684 sv_setpviv(sv, iv);
10688 #endif /* NO_MATHOMS */
10690 #if defined(PERL_IMPLICIT_CONTEXT)
10692 /* pTHX_ magic can't cope with varargs, so this is a no-context
10693 * version of the main function, (which may itself be aliased to us).
10694 * Don't access this version directly.
10698 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10703 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10705 va_start(args, pat);
10706 sv_vsetpvf(sv, pat, &args);
10710 /* pTHX_ magic can't cope with varargs, so this is a no-context
10711 * version of the main function, (which may itself be aliased to us).
10712 * Don't access this version directly.
10716 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10721 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10723 va_start(args, pat);
10724 sv_vsetpvf_mg(sv, pat, &args);
10730 =for apidoc sv_setpvf
10732 Works like C<sv_catpvf> but copies the text into the SV instead of
10733 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10739 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10743 PERL_ARGS_ASSERT_SV_SETPVF;
10745 va_start(args, pat);
10746 sv_vsetpvf(sv, pat, &args);
10751 =for apidoc sv_vsetpvf
10753 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10754 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10756 Usually used via its frontend C<sv_setpvf>.
10762 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10764 PERL_ARGS_ASSERT_SV_VSETPVF;
10766 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10770 =for apidoc sv_setpvf_mg
10772 Like C<sv_setpvf>, but also handles 'set' magic.
10778 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10782 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10784 va_start(args, pat);
10785 sv_vsetpvf_mg(sv, pat, &args);
10790 =for apidoc sv_vsetpvf_mg
10792 Like C<sv_vsetpvf>, but also handles 'set' magic.
10794 Usually used via its frontend C<sv_setpvf_mg>.
10800 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10802 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10804 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10808 #if defined(PERL_IMPLICIT_CONTEXT)
10810 /* pTHX_ magic can't cope with varargs, so this is a no-context
10811 * version of the main function, (which may itself be aliased to us).
10812 * Don't access this version directly.
10816 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10821 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10823 va_start(args, pat);
10824 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10828 /* pTHX_ magic can't cope with varargs, so this is a no-context
10829 * version of the main function, (which may itself be aliased to us).
10830 * Don't access this version directly.
10834 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10839 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10841 va_start(args, pat);
10842 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10849 =for apidoc sv_catpvf
10851 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10852 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10853 variable argument list, argument reordering is not supported.
10854 If the appended data contains "wide" characters
10855 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10856 and characters >255 formatted with C<%c>), the original SV might get
10857 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10858 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10859 valid UTF-8; if the original SV was bytes, the pattern should be too.
10864 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10868 PERL_ARGS_ASSERT_SV_CATPVF;
10870 va_start(args, pat);
10871 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10876 =for apidoc sv_vcatpvf
10878 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10879 variable argument list, and appends the formatted output
10880 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10882 Usually used via its frontend C<sv_catpvf>.
10888 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10890 PERL_ARGS_ASSERT_SV_VCATPVF;
10892 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10896 =for apidoc sv_catpvf_mg
10898 Like C<sv_catpvf>, but also handles 'set' magic.
10904 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10908 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10910 va_start(args, pat);
10911 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10917 =for apidoc sv_vcatpvf_mg
10919 Like C<sv_vcatpvf>, but also handles 'set' magic.
10921 Usually used via its frontend C<sv_catpvf_mg>.
10927 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10929 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10931 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10936 =for apidoc sv_vsetpvfn
10938 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10941 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10947 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10948 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10950 PERL_ARGS_ASSERT_SV_VSETPVFN;
10953 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10958 * Warn of missing argument to sprintf. The value used in place of such
10959 * arguments should be &PL_sv_no; an undefined value would yield
10960 * inappropriate "use of uninit" warnings [perl #71000].
10963 S_warn_vcatpvfn_missing_argument(pTHX) {
10964 if (ckWARN(WARN_MISSING)) {
10965 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10966 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10972 S_expect_number(pTHX_ char **const pattern)
10976 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10978 switch (**pattern) {
10979 case '1': case '2': case '3':
10980 case '4': case '5': case '6':
10981 case '7': case '8': case '9':
10982 var = *(*pattern)++ - '0';
10983 while (isDIGIT(**pattern)) {
10984 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10986 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10994 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10996 const int neg = nv < 0;
10999 PERL_ARGS_ASSERT_F0CONVERT;
11001 if (UNLIKELY(Perl_isinfnan(nv))) {
11002 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
11012 if (uv & 1 && uv == nv)
11013 uv--; /* Round to even */
11015 const unsigned dig = uv % 10;
11017 } while (uv /= 10);
11028 =for apidoc sv_vcatpvfn
11030 =for apidoc sv_vcatpvfn_flags
11032 Processes its arguments like C<vsprintf> and appends the formatted output
11033 to an SV. Uses an array of SVs if the C-style variable argument list is
11034 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11035 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11036 C<va_list> argument list with a format string that uses argument reordering
11037 will yield an exception.
11039 When running with taint checks enabled, indicates via
11040 C<maybe_tainted> if results are untrustworthy (often due to the use of
11043 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11045 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11050 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
11051 vecstr = (U8*)SvPV_const(vecsv,veclen);\
11052 vec_utf8 = DO_UTF8(vecsv);
11054 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11057 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11058 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
11060 PERL_ARGS_ASSERT_SV_VCATPVFN;
11062 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11065 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11066 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11067 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11068 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11069 * after the first 1023 zero bits.
11071 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11072 * of dynamically growing buffer might be better, start at just 16 bytes
11073 * (for example) and grow only when necessary. Or maybe just by looking
11074 * at the exponents of the two doubles? */
11075 # define DOUBLEDOUBLE_MAXBITS 2098
11078 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11079 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11080 * per xdigit. For the double-double case, this can be rather many.
11081 * The non-double-double-long-double overshoots since all bits of NV
11082 * are not mantissa bits, there are also exponent bits. */
11083 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11084 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11086 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11089 /* If we do not have a known long double format, (including not using
11090 * long doubles, or long doubles being equal to doubles) then we will
11091 * fall back to the ldexp/frexp route, with which we can retrieve at
11092 * most as many bits as our widest unsigned integer type is. We try
11093 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11095 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11096 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11098 #if defined(HAS_QUAD) && defined(Uquad_t)
11099 # define MANTISSATYPE Uquad_t
11100 # define MANTISSASIZE 8
11102 # define MANTISSATYPE UV
11103 # define MANTISSASIZE UVSIZE
11106 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11107 # define HEXTRACT_LITTLE_ENDIAN
11108 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11109 # define HEXTRACT_BIG_ENDIAN
11111 # define HEXTRACT_MIX_ENDIAN
11114 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
11115 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11116 * are being extracted from (either directly from the long double in-memory
11117 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11118 * is used to update the exponent. The subnormal is set to true
11119 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11120 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11122 * The tricky part is that S_hextract() needs to be called twice:
11123 * the first time with vend as NULL, and the second time with vend as
11124 * the pointer returned by the first call. What happens is that on
11125 * the first round the output size is computed, and the intended
11126 * extraction sanity checked. On the second round the actual output
11127 * (the extraction of the hexadecimal values) takes place.
11128 * Sanity failures cause fatal failures during both rounds. */
11130 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11131 U8* vhex, U8* vend)
11135 int ixmin = 0, ixmax = 0;
11137 /* XXX Inf/NaN are not handled here, since it is
11138 * assumed they are to be output as "Inf" and "NaN". */
11140 /* These macros are just to reduce typos, they have multiple
11141 * repetitions below, but usually only one (or sometimes two)
11142 * of them is really being used. */
11143 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11144 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11145 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11146 #define HEXTRACT_OUTPUT(ix) \
11148 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11150 #define HEXTRACT_COUNT(ix, c) \
11152 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11154 #define HEXTRACT_BYTE(ix) \
11156 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11158 #define HEXTRACT_LO_NYBBLE(ix) \
11160 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11162 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11163 * to make it look less odd when the top bits of a NV
11164 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11165 * order bits can be in the "low nybble" of a byte. */
11166 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11167 #define HEXTRACT_BYTES_LE(a, b) \
11168 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11169 #define HEXTRACT_BYTES_BE(a, b) \
11170 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11171 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11172 #define HEXTRACT_IMPLICIT_BIT(nv) \
11174 if (!*subnormal) { \
11175 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11179 /* Most formats do. Those which don't should undef this.
11181 * But also note that IEEE 754 subnormals do not have it, or,
11182 * expressed alternatively, their implicit bit is zero. */
11183 #define HEXTRACT_HAS_IMPLICIT_BIT
11185 /* Many formats do. Those which don't should undef this. */
11186 #define HEXTRACT_HAS_TOP_NYBBLE
11188 /* HEXTRACTSIZE is the maximum number of xdigits. */
11189 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11190 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11192 # define HEXTRACTSIZE 2 * NVSIZE
11195 const U8* vmaxend = vhex + HEXTRACTSIZE;
11196 PERL_UNUSED_VAR(ix); /* might happen */
11197 (void)Perl_frexp(PERL_ABS(nv), exponent);
11198 *subnormal = FALSE;
11199 if (vend && (vend <= vhex || vend > vmaxend)) {
11200 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11201 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11204 /* First check if using long doubles. */
11205 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11206 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11207 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11208 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11209 /* The bytes 13..0 are the mantissa/fraction,
11210 * the 15,14 are the sign+exponent. */
11211 const U8* nvp = (const U8*)(&nv);
11212 HEXTRACT_GET_SUBNORMAL(nv);
11213 HEXTRACT_IMPLICIT_BIT(nv);
11214 # undef HEXTRACT_HAS_TOP_NYBBLE
11215 HEXTRACT_BYTES_LE(13, 0);
11216 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11217 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11218 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11219 /* The bytes 2..15 are the mantissa/fraction,
11220 * the 0,1 are the sign+exponent. */
11221 const U8* nvp = (const U8*)(&nv);
11222 HEXTRACT_GET_SUBNORMAL(nv);
11223 HEXTRACT_IMPLICIT_BIT(nv);
11224 # undef HEXTRACT_HAS_TOP_NYBBLE
11225 HEXTRACT_BYTES_BE(2, 15);
11226 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11227 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11228 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11229 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11230 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11231 /* The bytes 0..1 are the sign+exponent,
11232 * the bytes 2..9 are the mantissa/fraction. */
11233 const U8* nvp = (const U8*)(&nv);
11234 # undef HEXTRACT_HAS_IMPLICIT_BIT
11235 # undef HEXTRACT_HAS_TOP_NYBBLE
11236 HEXTRACT_GET_SUBNORMAL(nv);
11237 HEXTRACT_BYTES_LE(7, 0);
11238 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11239 /* Does this format ever happen? (Wikipedia says the Motorola
11240 * 6888x math coprocessors used format _like_ this but padded
11241 * to 96 bits with 16 unused bits between the exponent and the
11243 const U8* nvp = (const U8*)(&nv);
11244 # undef HEXTRACT_HAS_IMPLICIT_BIT
11245 # undef HEXTRACT_HAS_TOP_NYBBLE
11246 HEXTRACT_GET_SUBNORMAL(nv);
11247 HEXTRACT_BYTES_BE(0, 7);
11249 # define HEXTRACT_FALLBACK
11250 /* Double-double format: two doubles next to each other.
11251 * The first double is the high-order one, exactly like
11252 * it would be for a "lone" double. The second double
11253 * is shifted down using the exponent so that that there
11254 * are no common bits. The tricky part is that the value
11255 * of the double-double is the SUM of the two doubles and
11256 * the second one can be also NEGATIVE.
11258 * Because of this tricky construction the bytewise extraction we
11259 * use for the other long double formats doesn't work, we must
11260 * extract the values bit by bit.
11262 * The little-endian double-double is used .. somewhere?
11264 * The big endian double-double is used in e.g. PPC/Power (AIX)
11267 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11268 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11269 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11272 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11273 /* Using normal doubles, not long doubles.
11275 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11276 * bytes, since we might need to handle printf precision, and
11277 * also need to insert the radix. */
11279 # ifdef HEXTRACT_LITTLE_ENDIAN
11280 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11281 const U8* nvp = (const U8*)(&nv);
11282 HEXTRACT_GET_SUBNORMAL(nv);
11283 HEXTRACT_IMPLICIT_BIT(nv);
11284 HEXTRACT_TOP_NYBBLE(6);
11285 HEXTRACT_BYTES_LE(5, 0);
11286 # elif defined(HEXTRACT_BIG_ENDIAN)
11287 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11288 const U8* nvp = (const U8*)(&nv);
11289 HEXTRACT_GET_SUBNORMAL(nv);
11290 HEXTRACT_IMPLICIT_BIT(nv);
11291 HEXTRACT_TOP_NYBBLE(1);
11292 HEXTRACT_BYTES_BE(2, 7);
11293 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11294 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11295 const U8* nvp = (const U8*)(&nv);
11296 HEXTRACT_GET_SUBNORMAL(nv);
11297 HEXTRACT_IMPLICIT_BIT(nv);
11298 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11299 HEXTRACT_BYTE(1); /* 5 */
11300 HEXTRACT_BYTE(0); /* 4 */
11301 HEXTRACT_BYTE(7); /* 3 */
11302 HEXTRACT_BYTE(6); /* 2 */
11303 HEXTRACT_BYTE(5); /* 1 */
11304 HEXTRACT_BYTE(4); /* 0 */
11305 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11306 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11307 const U8* nvp = (const U8*)(&nv);
11308 HEXTRACT_GET_SUBNORMAL(nv);
11309 HEXTRACT_IMPLICIT_BIT(nv);
11310 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11311 HEXTRACT_BYTE(6); /* 5 */
11312 HEXTRACT_BYTE(7); /* 4 */
11313 HEXTRACT_BYTE(0); /* 3 */
11314 HEXTRACT_BYTE(1); /* 2 */
11315 HEXTRACT_BYTE(2); /* 1 */
11316 HEXTRACT_BYTE(3); /* 0 */
11318 # define HEXTRACT_FALLBACK
11321 # define HEXTRACT_FALLBACK
11323 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11324 # ifdef HEXTRACT_FALLBACK
11325 HEXTRACT_GET_SUBNORMAL(nv);
11326 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11327 /* The fallback is used for the double-double format, and
11328 * for unknown long double formats, and for unknown double
11329 * formats, or in general unknown NV formats. */
11330 if (nv == (NV)0.0) {
11338 NV d = nv < 0 ? -nv : nv;
11340 U8 ha = 0x0; /* hexvalue accumulator */
11341 U8 hd = 0x8; /* hexvalue digit */
11343 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11344 * this is essentially manual frexp(). Multiplying by 0.5 and
11345 * doubling should be lossless in binary floating point. */
11355 while (d >= e + e) {
11359 /* Now e <= d < 2*e */
11361 /* First extract the leading hexdigit (the implicit bit). */
11377 /* Then extract the remaining hexdigits. */
11378 while (d > (NV)0.0) {
11384 /* Output or count in groups of four bits,
11385 * that is, when the hexdigit is down to one. */
11390 /* Reset the hexvalue. */
11399 /* Flush possible pending hexvalue. */
11409 /* Croak for various reasons: if the output pointer escaped the
11410 * output buffer, if the extraction index escaped the extraction
11411 * buffer, or if the ending output pointer didn't match the
11412 * previously computed value. */
11413 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11414 /* For double-double the ixmin and ixmax stay at zero,
11415 * which is convenient since the HEXTRACTSIZE is tricky
11416 * for double-double. */
11417 ixmin < 0 || ixmax >= NVSIZE ||
11418 (vend && v != vend)) {
11419 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11420 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11425 /* Helper for sv_vcatpvfn_flags(). */
11426 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11431 (var) = &PL_sv_no; /* [perl #71000] */ \
11432 arg_missing = TRUE; \
11437 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11438 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11443 const char *patend;
11446 static const char nullstr[] = "(null)";
11448 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11449 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11451 /* Times 4: a decimal digit takes more than 3 binary digits.
11452 * NV_DIG: mantissa takes than many decimal digits.
11453 * Plus 32: Playing safe. */
11454 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11455 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11456 bool hexfp = FALSE; /* hexadecimal floating point? */
11458 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11460 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11461 PERL_UNUSED_ARG(maybe_tainted);
11463 if (flags & SV_GMAGIC)
11466 /* no matter what, this is a string now */
11467 (void)SvPV_force_nomg(sv, origlen);
11469 /* special-case "", "%s", and "%-p" (SVf - see below) */
11471 if (svmax && ckWARN(WARN_REDUNDANT))
11472 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11473 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11476 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11477 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11478 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11479 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11482 const char * const s = va_arg(*args, char*);
11483 sv_catpv_nomg(sv, s ? s : nullstr);
11485 else if (svix < svmax) {
11486 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11487 SvGETMAGIC(*svargs);
11488 sv_catsv_nomg(sv, *svargs);
11491 S_warn_vcatpvfn_missing_argument(aTHX);
11494 if (args && patlen == 3 && pat[0] == '%' &&
11495 pat[1] == '-' && pat[2] == 'p') {
11496 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11497 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11498 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11499 argsv = MUTABLE_SV(va_arg(*args, void*));
11500 sv_catsv_nomg(sv, argsv);
11504 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11505 /* special-case "%.<number>[gf]" */
11506 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11507 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11508 unsigned digits = 0;
11512 while (*pp >= '0' && *pp <= '9')
11513 digits = 10 * digits + (*pp++ - '0');
11515 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11516 format the first argument and WARN_REDUNDANT if svmax > 1?
11517 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11518 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11519 const NV nv = SvNV(*svargs);
11520 if (LIKELY(!Perl_isinfnan(nv))) {
11522 /* Add check for digits != 0 because it seems that some
11523 gconverts are buggy in this case, and we don't yet have
11524 a Configure test for this. */
11525 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11526 /* 0, point, slack */
11527 STORE_LC_NUMERIC_SET_TO_NEEDED();
11528 SNPRINTF_G(nv, ebuf, size, digits);
11529 sv_catpv_nomg(sv, ebuf);
11530 if (*ebuf) /* May return an empty string for digits==0 */
11533 } else if (!digits) {
11536 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11537 sv_catpvn_nomg(sv, p, l);
11544 #endif /* !USE_LONG_DOUBLE */
11546 if (!args && svix < svmax && DO_UTF8(*svargs))
11549 patend = (char*)pat + patlen;
11550 for (p = (char*)pat; p < patend; p = q) {
11553 bool vectorize = FALSE;
11554 bool vectorarg = FALSE;
11555 bool vec_utf8 = FALSE;
11561 bool has_precis = FALSE;
11563 const I32 osvix = svix;
11564 bool is_utf8 = FALSE; /* is this item utf8? */
11565 bool used_explicit_ix = FALSE;
11566 bool arg_missing = FALSE;
11567 #ifdef HAS_LDBL_SPRINTF_BUG
11568 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11569 with sfio - Allen <allens@cpan.org> */
11570 bool fix_ldbl_sprintf_bug = FALSE;
11574 U8 utf8buf[UTF8_MAXBYTES+1];
11575 STRLEN esignlen = 0;
11577 const char *eptr = NULL;
11578 const char *fmtstart;
11581 const U8 *vecstr = NULL;
11588 /* We need a long double target in case HAS_LONG_DOUBLE,
11589 * even without USE_LONG_DOUBLE, so that we can printf with
11590 * long double formats, even without NV being long double.
11591 * But we call the target 'fv' instead of 'nv', since most of
11592 * the time it is not (most compilers these days recognize
11593 * "long double", even if only as a synonym for "double").
11595 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11596 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11598 # ifdef Perl_isfinitel
11599 # define FV_ISFINITE(x) Perl_isfinitel(x)
11601 # define FV_GF PERL_PRIgldbl
11602 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11603 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11604 # define NV_TO_FV(nv,fv) STMT_START { \
11606 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11609 # define NV_TO_FV(nv,fv) (fv)=(nv)
11613 # define FV_GF NVgf
11614 # define NV_TO_FV(nv,fv) (fv)=(nv)
11616 #ifndef FV_ISFINITE
11617 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11623 const char *dotstr = ".";
11624 STRLEN dotstrlen = 1;
11625 I32 efix = 0; /* explicit format parameter index */
11626 I32 ewix = 0; /* explicit width index */
11627 I32 epix = 0; /* explicit precision index */
11628 I32 evix = 0; /* explicit vector index */
11629 bool asterisk = FALSE;
11630 bool infnan = FALSE;
11632 /* echo everything up to the next format specification */
11633 for (q = p; q < patend && *q != '%'; ++q) ;
11635 if (has_utf8 && !pat_utf8)
11636 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11638 sv_catpvn_nomg(sv, p, q - p);
11647 We allow format specification elements in this order:
11648 \d+\$ explicit format parameter index
11650 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11651 0 flag (as above): repeated to allow "v02"
11652 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11653 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11655 [%bcdefginopsuxDFOUX] format (mandatory)
11660 As of perl5.9.3, printf format checking is on by default.
11661 Internally, perl uses %p formats to provide an escape to
11662 some extended formatting. This block deals with those
11663 extensions: if it does not match, (char*)q is reset and
11664 the normal format processing code is used.
11666 Currently defined extensions are:
11667 %p include pointer address (standard)
11668 %-p (SVf) include an SV (previously %_)
11669 %-<num>p include an SV with precision <num>
11671 %3p include a HEK with precision of 256
11672 %4p char* preceded by utf8 flag and length
11673 %<num>p (where num is 1 or > 4) reserved for future
11676 Robin Barker 2005-07-14 (but modified since)
11678 %1p (VDf) removed. RMB 2007-10-19
11685 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11686 /* The argument has already gone through cBOOL, so the cast
11688 is_utf8 = (bool)va_arg(*args, int);
11689 elen = va_arg(*args, UV);
11690 /* if utf8 length is larger than 0x7ffff..., then it might
11691 * have been a signed value that wrapped */
11692 if (elen > ((~(STRLEN)0) >> 1)) {
11693 assert(0); /* in DEBUGGING build we want to crash */
11694 elen= 0; /* otherwise we want to treat this as an empty string */
11696 eptr = va_arg(*args, char *);
11697 q += sizeof(UTF8f)-1;
11700 n = expect_number(&q);
11702 if (sv) { /* SVf */
11707 argsv = MUTABLE_SV(va_arg(*args, void*));
11708 eptr = SvPV_const(argsv, elen);
11709 if (DO_UTF8(argsv))
11713 else if (n==2 || n==3) { /* HEKf */
11714 HEK * const hek = va_arg(*args, HEK *);
11715 eptr = HEK_KEY(hek);
11716 elen = HEK_LEN(hek);
11717 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11718 if (n==3) precis = 256, has_precis = TRUE;
11722 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11723 "internal %%<num>p might conflict with future printf extensions");
11729 if ( (width = expect_number(&q)) ) {
11732 Perl_croak_nocontext(
11733 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11736 used_explicit_ix = TRUE;
11748 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11777 if ( (ewix = expect_number(&q)) ) {
11780 Perl_croak_nocontext(
11781 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11782 used_explicit_ix = TRUE;
11792 if ((vectorarg = asterisk)) {
11805 width = expect_number(&q);
11808 if (vectorize && vectorarg) {
11809 /* vectorizing, but not with the default "." */
11811 vecsv = va_arg(*args, SV*);
11813 FETCH_VCATPVFN_ARGUMENT(
11814 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11816 FETCH_VCATPVFN_ARGUMENT(
11817 vecsv, svix < svmax, svargs[svix++]);
11819 dotstr = SvPV_const(vecsv, dotstrlen);
11820 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11821 bad with tied or overloaded values that return UTF8. */
11822 if (DO_UTF8(vecsv))
11824 else if (has_utf8) {
11825 vecsv = sv_mortalcopy(vecsv);
11826 sv_utf8_upgrade(vecsv);
11827 dotstr = SvPV_const(vecsv, dotstrlen);
11834 i = va_arg(*args, int);
11836 i = (ewix ? ewix <= svmax : svix < svmax) ?
11837 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11839 width = (i < 0) ? -i : i;
11849 if ( (epix = expect_number(&q)) ) {
11852 Perl_croak_nocontext(
11853 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11854 used_explicit_ix = TRUE;
11859 i = va_arg(*args, int);
11863 FETCH_VCATPVFN_ARGUMENT(
11864 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11866 FETCH_VCATPVFN_ARGUMENT(
11867 precsv, svix < svmax, svargs[svix++]);
11868 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11871 has_precis = !(i < 0);
11875 while (isDIGIT(*q))
11876 precis = precis * 10 + (*q++ - '0');
11885 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11886 vecsv = svargs[efix ? efix-1 : svix++];
11887 vecstr = (U8*)SvPV_const(vecsv,veclen);
11888 vec_utf8 = DO_UTF8(vecsv);
11890 /* if this is a version object, we need to convert
11891 * back into v-string notation and then let the
11892 * vectorize happen normally
11894 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11895 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
11896 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11897 "vector argument not supported with alpha versions");
11900 vecsv = sv_newmortal();
11901 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11903 vecstr = (U8*)SvPV_const(vecsv, veclen);
11904 vec_utf8 = DO_UTF8(vecsv);
11918 case 'I': /* Ix, I32x, and I64x */
11919 # ifdef USE_64_BIT_INT
11920 if (q[1] == '6' && q[2] == '4') {
11926 if (q[1] == '3' && q[2] == '2') {
11930 # ifdef USE_64_BIT_INT
11936 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11937 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11940 # ifdef USE_QUADMATH
11953 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11954 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11955 if (*q == 'l') { /* lld, llf */
11964 if (*++q == 'h') { /* hhd, hhu */
11993 if (!vectorize && !args) {
11995 const I32 i = efix-1;
11996 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11998 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
12003 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
12004 /* XXX va_arg(*args) case? need peek, use va_copy? */
12006 if (UNLIKELY(SvAMAGIC(argsv)))
12007 argsv = sv_2num(argsv);
12008 infnan = UNLIKELY(isinfnansv(argsv));
12011 switch (c = *q++) {
12019 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12020 /* no va_arg() case */
12021 SvNV_nomg(argsv), (int)c);
12022 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
12024 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12026 eptr = (char*)utf8buf;
12027 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
12041 eptr = va_arg(*args, char*);
12043 elen = strlen(eptr);
12045 eptr = (char *)nullstr;
12046 elen = sizeof nullstr - 1;
12050 eptr = SvPV_const(argsv, elen);
12051 if (DO_UTF8(argsv)) {
12052 STRLEN old_precis = precis;
12053 if (has_precis && precis < elen) {
12054 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12055 STRLEN p = precis > ulen ? ulen : precis;
12056 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12057 /* sticks at end */
12059 if (width) { /* fudge width (can't fudge elen) */
12060 if (has_precis && precis < elen)
12061 width += precis - old_precis;
12064 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12071 if (has_precis && precis < elen)
12079 goto floating_point;
12081 if (alt || vectorize)
12083 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12097 goto floating_point;
12102 goto donevalidconversion;
12104 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12113 esignbuf[esignlen++] = plus;
12117 case 'c': iv = (char)va_arg(*args, int); break;
12118 case 'h': iv = (short)va_arg(*args, int); break;
12119 case 'l': iv = va_arg(*args, long); break;
12120 case 'V': iv = va_arg(*args, IV); break;
12121 case 'z': iv = va_arg(*args, SSize_t); break;
12122 #ifdef HAS_PTRDIFF_T
12123 case 't': iv = va_arg(*args, ptrdiff_t); break;
12125 default: iv = va_arg(*args, int); break;
12127 case 'j': iv = va_arg(*args, intmax_t); break;
12131 iv = va_arg(*args, Quad_t); break;
12138 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12140 case 'c': iv = (char)tiv; break;
12141 case 'h': iv = (short)tiv; break;
12142 case 'l': iv = (long)tiv; break;
12144 default: iv = tiv; break;
12147 iv = (Quad_t)tiv; break;
12153 if ( !vectorize ) /* we already set uv above */
12158 esignbuf[esignlen++] = plus;
12161 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12162 esignbuf[esignlen++] = '-';
12201 goto floating_point;
12207 goto donevalidconversion;
12209 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12220 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12221 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12222 case 'l': uv = va_arg(*args, unsigned long); break;
12223 case 'V': uv = va_arg(*args, UV); break;
12224 case 'z': uv = va_arg(*args, Size_t); break;
12225 #ifdef HAS_PTRDIFF_T
12226 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12229 case 'j': uv = va_arg(*args, uintmax_t); break;
12231 default: uv = va_arg(*args, unsigned); break;
12234 uv = va_arg(*args, Uquad_t); break;
12241 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12243 case 'c': uv = (unsigned char)tuv; break;
12244 case 'h': uv = (unsigned short)tuv; break;
12245 case 'l': uv = (unsigned long)tuv; break;
12247 default: uv = tuv; break;
12250 uv = (Uquad_t)tuv; break;
12259 char *ptr = ebuf + sizeof ebuf;
12260 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12266 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12270 } while (uv >>= 4);
12272 esignbuf[esignlen++] = '0';
12273 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12279 *--ptr = '0' + dig;
12280 } while (uv >>= 3);
12281 if (alt && *ptr != '0')
12287 *--ptr = '0' + dig;
12288 } while (uv >>= 1);
12290 esignbuf[esignlen++] = '0';
12291 esignbuf[esignlen++] = c;
12294 default: /* it had better be ten or less */
12297 *--ptr = '0' + dig;
12298 } while (uv /= base);
12301 elen = (ebuf + sizeof ebuf) - ptr;
12305 zeros = precis - elen;
12306 else if (precis == 0 && elen == 1 && *eptr == '0'
12307 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12310 /* a precision nullifies the 0 flag. */
12317 /* FLOATING POINT */
12322 c = 'f'; /* maybe %F isn't supported here */
12324 case 'e': case 'E':
12326 case 'g': case 'G':
12327 case 'a': case 'A':
12331 /* This is evil, but floating point is even more evil */
12333 /* for SV-style calling, we can only get NV
12334 for C-style calling, we assume %f is double;
12335 for simplicity we allow any of %Lf, %llf, %qf for long double
12339 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12343 /* [perl #20339] - we should accept and ignore %lf rather than die */
12347 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12348 intsize = args ? 0 : 'q';
12352 #if defined(HAS_LONG_DOUBLE)
12365 /* Now we need (long double) if intsize == 'q', else (double). */
12367 /* Note: do not pull NVs off the va_list with va_arg()
12368 * (pull doubles instead) because if you have a build
12369 * with long doubles, you would always be pulling long
12370 * doubles, which would badly break anyone using only
12371 * doubles (i.e. the majority of builds). In other
12372 * words, you cannot mix doubles and long doubles.
12373 * The only case where you can pull off long doubles
12374 * is when the format specifier explicitly asks so with
12376 #ifdef USE_QUADMATH
12377 fv = intsize == 'q' ?
12378 va_arg(*args, NV) : va_arg(*args, double);
12380 #elif LONG_DOUBLESIZE > DOUBLESIZE
12381 if (intsize == 'q') {
12382 fv = va_arg(*args, long double);
12385 nv = va_arg(*args, double);
12389 nv = va_arg(*args, double);
12395 if (!infnan) SvGETMAGIC(argsv);
12396 nv = SvNV_nomg(argsv);
12401 /* frexp() (or frexpl) has some unspecified behaviour for
12402 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12403 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12405 (void)Perl_frexp((NV)fv, &i);
12406 if (i == PERL_INT_MIN)
12407 Perl_die(aTHX_ "panic: frexp: %" FV_GF, fv);
12408 /* Do not set hexfp earlier since we want to printf
12409 * Inf/NaN for Inf/NaN, not their hexfp. */
12410 hexfp = isALPHA_FOLD_EQ(c, 'a');
12411 if (UNLIKELY(hexfp)) {
12412 /* This seriously overshoots in most cases, but
12413 * better the undershooting. Firstly, all bytes
12414 * of the NV are not mantissa, some of them are
12415 * exponent. Secondly, for the reasonably common
12416 * long doubles case, the "80-bit extended", two
12417 * or six bytes of the NV are unused. */
12419 (fv < 0) ? 1 : 0 + /* possible unary minus */
12421 1 + /* the very unlikely carry */
12424 2 * NVSIZE + /* 2 hexdigits for each byte */
12426 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12428 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12429 /* However, for the "double double", we need more.
12430 * Since each double has their own exponent, the
12431 * doubles may float (haha) rather far from each
12432 * other, and the number of required bits is much
12433 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12434 * See the definition of DOUBLEDOUBLE_MAXBITS.
12436 * Need 2 hexdigits for each byte. */
12437 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12438 /* the size for the exponent already added */
12440 #ifdef USE_LOCALE_NUMERIC
12441 STORE_LC_NUMERIC_SET_TO_NEEDED();
12442 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12443 need += SvLEN(PL_numeric_radix_sv);
12444 RESTORE_LC_NUMERIC();
12448 need = BIT_DIGITS(i);
12449 } /* if i < 0, the number of digits is hard to predict. */
12451 need += has_precis ? precis : 6; /* known default */
12456 #ifdef HAS_LDBL_SPRINTF_BUG
12457 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12458 with sfio - Allen <allens@cpan.org> */
12461 # define MY_DBL_MAX DBL_MAX
12462 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12463 # if DOUBLESIZE >= 8
12464 # define MY_DBL_MAX 1.7976931348623157E+308L
12466 # define MY_DBL_MAX 3.40282347E+38L
12470 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12471 # define MY_DBL_MAX_BUG 1L
12473 # define MY_DBL_MAX_BUG MY_DBL_MAX
12477 # define MY_DBL_MIN DBL_MIN
12478 # else /* XXX guessing! -Allen */
12479 # if DOUBLESIZE >= 8
12480 # define MY_DBL_MIN 2.2250738585072014E-308L
12482 # define MY_DBL_MIN 1.17549435E-38L
12486 if ((intsize == 'q') && (c == 'f') &&
12487 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12488 (need < DBL_DIG)) {
12489 /* it's going to be short enough that
12490 * long double precision is not needed */
12492 if ((fv <= 0L) && (fv >= -0L))
12493 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12495 /* would use Perl_fp_class as a double-check but not
12496 * functional on IRIX - see perl.h comments */
12498 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12499 /* It's within the range that a double can represent */
12500 #if defined(DBL_MAX) && !defined(DBL_MIN)
12501 if ((fv >= ((long double)1/DBL_MAX)) ||
12502 (fv <= (-(long double)1/DBL_MAX)))
12504 fix_ldbl_sprintf_bug = TRUE;
12507 if (fix_ldbl_sprintf_bug == TRUE) {
12517 # undef MY_DBL_MAX_BUG
12520 #endif /* HAS_LDBL_SPRINTF_BUG */
12522 need += 20; /* fudge factor */
12523 if (PL_efloatsize < need) {
12524 Safefree(PL_efloatbuf);
12525 PL_efloatsize = need + 20; /* more fudge */
12526 Newx(PL_efloatbuf, PL_efloatsize, char);
12527 PL_efloatbuf[0] = '\0';
12530 if ( !(width || left || plus || alt) && fill != '0'
12531 && has_precis && intsize != 'q' /* Shortcuts */
12532 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12533 /* See earlier comment about buggy Gconvert when digits,
12535 if ( c == 'g' && precis ) {
12536 STORE_LC_NUMERIC_SET_TO_NEEDED();
12537 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12538 /* May return an empty string for digits==0 */
12539 if (*PL_efloatbuf) {
12540 elen = strlen(PL_efloatbuf);
12541 goto float_converted;
12543 } else if ( c == 'f' && !precis ) {
12544 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12549 if (UNLIKELY(hexfp)) {
12550 /* Hexadecimal floating point. */
12551 char* p = PL_efloatbuf;
12552 U8 vhex[VHEX_SIZE];
12553 U8* v = vhex; /* working pointer to vhex */
12554 U8* vend; /* pointer to one beyond last digit of vhex */
12555 U8* vfnz = NULL; /* first non-zero */
12556 U8* vlnz = NULL; /* last non-zero */
12557 U8* v0 = NULL; /* first output */
12558 const bool lower = (c == 'a');
12559 /* At output the values of vhex (up to vend) will
12560 * be mapped through the xdig to get the actual
12561 * human-readable xdigits. */
12562 const char* xdig = PL_hexdigit;
12563 int zerotail = 0; /* how many extra zeros to append */
12564 int exponent = 0; /* exponent of the floating point input */
12565 bool hexradix = FALSE; /* should we output the radix */
12566 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12567 bool negative = FALSE;
12569 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12571 * For example with denormals, (assuming the vanilla
12572 * 64-bit double): the exponent is zero. 1xp-1074 is
12573 * the smallest denormal and the smallest double, it
12574 * could be output also as 0x0.0000000000001p-1022 to
12575 * match its internal structure. */
12577 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12578 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12580 #if NVSIZE > DOUBLESIZE
12581 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12582 /* In this case there is an implicit bit,
12583 * and therefore the exponent is shifted by one. */
12586 # ifdef NV_X86_80_BIT
12588 /* The subnormals of the x86-80 have a base exponent of -16382,
12589 * (while the physical exponent bits are zero) but the frexp()
12590 * returned the scientific-style floating exponent. We want
12591 * to map the last one as:
12592 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12593 * -16835..-16388 -> -16384
12594 * since we want to keep the first hexdigit
12595 * as one of the [8421]. */
12596 exponent = -4 * ( (exponent + 1) / -4) - 2;
12601 /* TBD: other non-implicit-bit platforms than the x86-80. */
12605 negative = fv < 0 || Perl_signbit(nv);
12616 xdig += 16; /* Use uppercase hex. */
12619 /* Find the first non-zero xdigit. */
12620 for (v = vhex; v < vend; v++) {
12628 /* Find the last non-zero xdigit. */
12629 for (v = vend - 1; v >= vhex; v--) {
12636 #if NVSIZE == DOUBLESIZE
12642 #ifndef NV_X86_80_BIT
12644 /* IEEE 754 subnormals (but not the x86 80-bit):
12645 * we want "normalize" the subnormal,
12646 * so we need to right shift the hex nybbles
12647 * so that the output of the subnormal starts
12648 * from the first true bit. (Another, equally
12649 * valid, policy would be to dump the subnormal
12650 * nybbles as-is, to display the "physical" layout.) */
12653 /* Find the ceil(log2(v[0])) of
12654 * the top non-zero nybble. */
12655 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12658 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12659 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12673 U8* ve = (subnormal ? vlnz + 1 : vend);
12674 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12675 if ((SSize_t)(precis + 1) < vn) {
12676 bool overflow = FALSE;
12677 if (v0[precis + 1] < 0x8) {
12678 /* Round down, nothing to do. */
12679 } else if (v0[precis + 1] > 0x8) {
12682 overflow = v0[precis] > 0xF;
12684 } else { /* v0[precis] == 0x8 */
12685 /* Half-point: round towards the one
12686 * with the even least-significant digit:
12694 * 78 -> 8 f8 -> 10 */
12695 if ((v0[precis] & 0x1)) {
12698 overflow = v0[precis] > 0xF;
12703 for (v = v0 + precis - 1; v >= v0; v--) {
12705 overflow = *v > 0xF;
12711 if (v == v0 - 1 && overflow) {
12712 /* If the overflow goes all the
12713 * way to the front, we need to
12714 * insert 0x1 in front, and adjust
12716 Move(v0, v0 + 1, vn, char);
12722 /* The new effective "last non zero". */
12723 vlnz = v0 + precis;
12727 subnormal ? precis - vn + 1 :
12728 precis - (vlnz - vhex);
12735 /* If there are non-zero xdigits, the radix
12736 * is output after the first one. */
12747 /* The radix is always output if precis, or if alt. */
12748 if (precis > 0 || alt) {
12753 #ifndef USE_LOCALE_NUMERIC
12756 STORE_LC_NUMERIC_SET_TO_NEEDED();
12757 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12759 const char* r = SvPV(PL_numeric_radix_sv, n);
12760 Copy(r, p, n, char);
12766 RESTORE_LC_NUMERIC();
12775 if (zerotail > 0) {
12776 while (zerotail--) {
12781 elen = p - PL_efloatbuf;
12782 elen += my_snprintf(p, PL_efloatsize - elen,
12783 "%c%+d", lower ? 'p' : 'P',
12786 if (elen < width) {
12788 /* Pad the back with spaces. */
12789 memset(PL_efloatbuf + elen, ' ', width - elen);
12791 else if (fill == '0') {
12792 /* Insert the zeros after the "0x" and the
12793 * the potential sign, but before the digits,
12794 * otherwise we end up with "0000xH.HHH...",
12795 * when we want "0x000H.HHH..." */
12796 STRLEN nzero = width - elen;
12797 char* zerox = PL_efloatbuf + 2;
12798 STRLEN nmove = elen - 2;
12799 if (negative || plus) {
12803 Move(zerox, zerox + nzero, nmove, char);
12804 memset(zerox, fill, nzero);
12807 /* Move it to the right. */
12808 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12810 /* Pad the front with spaces. */
12811 memset(PL_efloatbuf, ' ', width - elen);
12817 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12819 /* Not affecting infnan output: precision, alt, fill. */
12820 if (elen < width) {
12822 /* Pack the back with spaces. */
12823 memset(PL_efloatbuf + elen, ' ', width - elen);
12825 /* Move it to the right. */
12826 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12828 /* Pad the front with spaces. */
12829 memset(PL_efloatbuf, ' ', width - elen);
12837 char *ptr = ebuf + sizeof ebuf;
12840 #if defined(USE_QUADMATH)
12841 if (intsize == 'q') {
12845 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12846 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12847 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12848 * not USE_LONG_DOUBLE and NVff. In other words,
12849 * this needs to work without USE_LONG_DOUBLE. */
12850 if (intsize == 'q') {
12851 /* Copy the one or more characters in a long double
12852 * format before the 'base' ([efgEFG]) character to
12853 * the format string. */
12854 static char const ldblf[] = PERL_PRIfldbl;
12855 char const *p = ldblf + sizeof(ldblf) - 3;
12856 while (p >= ldblf) { *--ptr = *p--; }
12861 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12866 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12878 /* No taint. Otherwise we are in the strange situation
12879 * where printf() taints but print($float) doesn't.
12882 STORE_LC_NUMERIC_SET_TO_NEEDED();
12884 /* hopefully the above makes ptr a very constrained format
12885 * that is safe to use, even though it's not literal */
12886 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12887 #ifdef USE_QUADMATH
12889 const char* qfmt = quadmath_format_single(ptr);
12891 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12892 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12894 if ((IV)elen == -1) {
12897 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12902 #elif defined(HAS_LONG_DOUBLE)
12903 elen = ((intsize == 'q')
12904 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12905 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12907 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12913 eptr = PL_efloatbuf;
12914 assert((IV)elen > 0); /* here zero elen is bad */
12916 #ifdef USE_LOCALE_NUMERIC
12917 /* If the decimal point character in the string is UTF-8, make the
12919 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12920 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12933 i = SvCUR(sv) - origlen;
12936 case 'c': *(va_arg(*args, char*)) = i; break;
12937 case 'h': *(va_arg(*args, short*)) = i; break;
12938 default: *(va_arg(*args, int*)) = i; break;
12939 case 'l': *(va_arg(*args, long*)) = i; break;
12940 case 'V': *(va_arg(*args, IV*)) = i; break;
12941 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12942 #ifdef HAS_PTRDIFF_T
12943 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12946 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12950 *(va_arg(*args, Quad_t*)) = i; break;
12957 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12958 goto donevalidconversion;
12965 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12966 && ckWARN(WARN_PRINTF))
12968 SV * const msg = sv_newmortal();
12969 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12970 (PL_op->op_type == OP_PRTF) ? "" : "s");
12971 if (fmtstart < patend) {
12972 const char * const fmtend = q < patend ? q : patend;
12974 sv_catpvs(msg, "\"%");
12975 for (f = fmtstart; f < fmtend; f++) {
12977 sv_catpvn_nomg(msg, f, 1);
12979 Perl_sv_catpvf(aTHX_ msg,
12980 "\\%03" UVof, (UV)*f & 0xFF);
12983 sv_catpvs(msg, "\"");
12985 sv_catpvs(msg, "end of string");
12987 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
12990 /* output mangled stuff ... */
12996 /* ... right here, because formatting flags should not apply */
12997 SvGROW(sv, SvCUR(sv) + elen + 1);
12999 Copy(eptr, p, elen, char);
13002 SvCUR_set(sv, p - SvPVX_const(sv));
13004 continue; /* not "break" */
13007 if (is_utf8 != has_utf8) {
13010 sv_utf8_upgrade(sv);
13013 const STRLEN old_elen = elen;
13014 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13015 sv_utf8_upgrade(nsv);
13016 eptr = SvPVX_const(nsv);
13019 if (width) { /* fudge width (can't fudge elen) */
13020 width += elen - old_elen;
13026 /* signed value that's wrapped? */
13027 assert(elen <= ((~(STRLEN)0) >> 1));
13028 have = esignlen + zeros + elen;
13030 croak_memory_wrap();
13032 need = (have > width ? have : width);
13035 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
13036 croak_memory_wrap();
13037 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
13039 if (esignlen && fill == '0') {
13041 for (i = 0; i < (int)esignlen; i++)
13042 *p++ = esignbuf[i];
13044 if (gap && !left) {
13045 memset(p, fill, gap);
13048 if (esignlen && fill != '0') {
13050 for (i = 0; i < (int)esignlen; i++)
13051 *p++ = esignbuf[i];
13055 for (i = zeros; i; i--)
13059 Copy(eptr, p, elen, char);
13063 memset(p, ' ', gap);
13068 Copy(dotstr, p, dotstrlen, char);
13072 vectorize = FALSE; /* done iterating over vecstr */
13079 SvCUR_set(sv, p - SvPVX_const(sv));
13085 donevalidconversion:
13086 if (used_explicit_ix)
13087 no_redundant_warning = TRUE;
13089 S_warn_vcatpvfn_missing_argument(aTHX);
13092 /* Now that we've consumed all our printf format arguments (svix)
13093 * do we have things left on the stack that we didn't use?
13095 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13096 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13097 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13102 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13106 /* =========================================================================
13108 =head1 Cloning an interpreter
13112 All the macros and functions in this section are for the private use of
13113 the main function, perl_clone().
13115 The foo_dup() functions make an exact copy of an existing foo thingy.
13116 During the course of a cloning, a hash table is used to map old addresses
13117 to new addresses. The table is created and manipulated with the
13118 ptr_table_* functions.
13120 * =========================================================================*/
13123 #if defined(USE_ITHREADS)
13125 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13126 #ifndef GpREFCNT_inc
13127 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13131 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13132 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13133 If this changes, please unmerge ss_dup.
13134 Likewise, sv_dup_inc_multiple() relies on this fact. */
13135 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13136 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13137 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13138 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13139 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13140 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13141 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13142 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13143 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13144 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13145 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13146 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13147 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13149 /* clone a parser */
13152 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13156 PERL_ARGS_ASSERT_PARSER_DUP;
13161 /* look for it in the table first */
13162 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13166 /* create anew and remember what it is */
13167 Newxz(parser, 1, yy_parser);
13168 ptr_table_store(PL_ptr_table, proto, parser);
13170 /* XXX these not yet duped */
13171 parser->old_parser = NULL;
13172 parser->stack = NULL;
13174 parser->stack_max1 = 0;
13175 /* XXX parser->stack->state = 0; */
13177 /* XXX eventually, just Copy() most of the parser struct ? */
13179 parser->lex_brackets = proto->lex_brackets;
13180 parser->lex_casemods = proto->lex_casemods;
13181 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13182 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13183 parser->lex_casestack = savepvn(proto->lex_casestack,
13184 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13185 parser->lex_defer = proto->lex_defer;
13186 parser->lex_dojoin = proto->lex_dojoin;
13187 parser->lex_formbrack = proto->lex_formbrack;
13188 parser->lex_inpat = proto->lex_inpat;
13189 parser->lex_inwhat = proto->lex_inwhat;
13190 parser->lex_op = proto->lex_op;
13191 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13192 parser->lex_starts = proto->lex_starts;
13193 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13194 parser->multi_close = proto->multi_close;
13195 parser->multi_open = proto->multi_open;
13196 parser->multi_start = proto->multi_start;
13197 parser->multi_end = proto->multi_end;
13198 parser->preambled = proto->preambled;
13199 parser->lex_super_state = proto->lex_super_state;
13200 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13201 parser->lex_sub_op = proto->lex_sub_op;
13202 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13203 parser->linestr = sv_dup_inc(proto->linestr, param);
13204 parser->expect = proto->expect;
13205 parser->copline = proto->copline;
13206 parser->last_lop_op = proto->last_lop_op;
13207 parser->lex_state = proto->lex_state;
13208 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13209 /* rsfp_filters entries have fake IoDIRP() */
13210 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13211 parser->in_my = proto->in_my;
13212 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13213 parser->error_count = proto->error_count;
13214 parser->sig_elems = proto->sig_elems;
13215 parser->sig_optelems= proto->sig_optelems;
13216 parser->sig_slurpy = proto->sig_slurpy;
13217 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13218 parser->linestr = sv_dup_inc(proto->linestr, param);
13221 char * const ols = SvPVX(proto->linestr);
13222 char * const ls = SvPVX(parser->linestr);
13224 parser->bufptr = ls + (proto->bufptr >= ols ?
13225 proto->bufptr - ols : 0);
13226 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13227 proto->oldbufptr - ols : 0);
13228 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13229 proto->oldoldbufptr - ols : 0);
13230 parser->linestart = ls + (proto->linestart >= ols ?
13231 proto->linestart - ols : 0);
13232 parser->last_uni = ls + (proto->last_uni >= ols ?
13233 proto->last_uni - ols : 0);
13234 parser->last_lop = ls + (proto->last_lop >= ols ?
13235 proto->last_lop - ols : 0);
13237 parser->bufend = ls + SvCUR(parser->linestr);
13240 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13243 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13244 Copy(proto->nexttype, parser->nexttype, 5, I32);
13245 parser->nexttoke = proto->nexttoke;
13247 /* XXX should clone saved_curcop here, but we aren't passed
13248 * proto_perl; so do it in perl_clone_using instead */
13254 /* duplicate a file handle */
13257 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13261 PERL_ARGS_ASSERT_FP_DUP;
13262 PERL_UNUSED_ARG(type);
13265 return (PerlIO*)NULL;
13267 /* look for it in the table first */
13268 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13272 /* create anew and remember what it is */
13273 #ifdef __amigaos4__
13274 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13276 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13278 ptr_table_store(PL_ptr_table, fp, ret);
13282 /* duplicate a directory handle */
13285 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13289 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13291 const Direntry_t *dirent;
13292 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13298 PERL_UNUSED_CONTEXT;
13299 PERL_ARGS_ASSERT_DIRP_DUP;
13304 /* look for it in the table first */
13305 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13309 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13311 PERL_UNUSED_ARG(param);
13315 /* open the current directory (so we can switch back) */
13316 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13318 /* chdir to our dir handle and open the present working directory */
13319 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13320 PerlDir_close(pwd);
13321 return (DIR *)NULL;
13323 /* Now we should have two dir handles pointing to the same dir. */
13325 /* Be nice to the calling code and chdir back to where we were. */
13326 /* XXX If this fails, then what? */
13327 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13329 /* We have no need of the pwd handle any more. */
13330 PerlDir_close(pwd);
13333 # define d_namlen(d) (d)->d_namlen
13335 # define d_namlen(d) strlen((d)->d_name)
13337 /* Iterate once through dp, to get the file name at the current posi-
13338 tion. Then step back. */
13339 pos = PerlDir_tell(dp);
13340 if ((dirent = PerlDir_read(dp))) {
13341 len = d_namlen(dirent);
13342 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13343 /* If the len is somehow magically longer than the
13344 * maximum length of the directory entry, even though
13345 * we could fit it in a buffer, we could not copy it
13346 * from the dirent. Bail out. */
13347 PerlDir_close(ret);
13350 if (len <= sizeof smallbuf) name = smallbuf;
13351 else Newx(name, len, char);
13352 Move(dirent->d_name, name, len, char);
13354 PerlDir_seek(dp, pos);
13356 /* Iterate through the new dir handle, till we find a file with the
13358 if (!dirent) /* just before the end */
13360 pos = PerlDir_tell(ret);
13361 if (PerlDir_read(ret)) continue; /* not there yet */
13362 PerlDir_seek(ret, pos); /* step back */
13366 const long pos0 = PerlDir_tell(ret);
13368 pos = PerlDir_tell(ret);
13369 if ((dirent = PerlDir_read(ret))) {
13370 if (len == (STRLEN)d_namlen(dirent)
13371 && memEQ(name, dirent->d_name, len)) {
13373 PerlDir_seek(ret, pos); /* step back */
13376 /* else we are not there yet; keep iterating */
13378 else { /* This is not meant to happen. The best we can do is
13379 reset the iterator to the beginning. */
13380 PerlDir_seek(ret, pos0);
13387 if (name && name != smallbuf)
13392 ret = win32_dirp_dup(dp, param);
13395 /* pop it in the pointer table */
13397 ptr_table_store(PL_ptr_table, dp, ret);
13402 /* duplicate a typeglob */
13405 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13409 PERL_ARGS_ASSERT_GP_DUP;
13413 /* look for it in the table first */
13414 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13418 /* create anew and remember what it is */
13420 ptr_table_store(PL_ptr_table, gp, ret);
13423 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13424 on Newxz() to do this for us. */
13425 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13426 ret->gp_io = io_dup_inc(gp->gp_io, param);
13427 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13428 ret->gp_av = av_dup_inc(gp->gp_av, param);
13429 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13430 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13431 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13432 ret->gp_cvgen = gp->gp_cvgen;
13433 ret->gp_line = gp->gp_line;
13434 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13438 /* duplicate a chain of magic */
13441 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13443 MAGIC *mgret = NULL;
13444 MAGIC **mgprev_p = &mgret;
13446 PERL_ARGS_ASSERT_MG_DUP;
13448 for (; mg; mg = mg->mg_moremagic) {
13451 if ((param->flags & CLONEf_JOIN_IN)
13452 && mg->mg_type == PERL_MAGIC_backref)
13453 /* when joining, we let the individual SVs add themselves to
13454 * backref as needed. */
13457 Newx(nmg, 1, MAGIC);
13459 mgprev_p = &(nmg->mg_moremagic);
13461 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13462 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13463 from the original commit adding Perl_mg_dup() - revision 4538.
13464 Similarly there is the annotation "XXX random ptr?" next to the
13465 assignment to nmg->mg_ptr. */
13468 /* FIXME for plugins
13469 if (nmg->mg_type == PERL_MAGIC_qr) {
13470 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13474 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13475 ? nmg->mg_type == PERL_MAGIC_backref
13476 /* The backref AV has its reference
13477 * count deliberately bumped by 1 */
13478 ? SvREFCNT_inc(av_dup_inc((const AV *)
13479 nmg->mg_obj, param))
13480 : sv_dup_inc(nmg->mg_obj, param)
13481 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13482 nmg->mg_type == PERL_MAGIC_regdata)
13484 : sv_dup(nmg->mg_obj, param);
13486 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13487 if (nmg->mg_len > 0) {
13488 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13489 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13490 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13492 AMT * const namtp = (AMT*)nmg->mg_ptr;
13493 sv_dup_inc_multiple((SV**)(namtp->table),
13494 (SV**)(namtp->table), NofAMmeth, param);
13497 else if (nmg->mg_len == HEf_SVKEY)
13498 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13500 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13501 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13507 #endif /* USE_ITHREADS */
13509 struct ptr_tbl_arena {
13510 struct ptr_tbl_arena *next;
13511 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13514 /* create a new pointer-mapping table */
13517 Perl_ptr_table_new(pTHX)
13520 PERL_UNUSED_CONTEXT;
13522 Newx(tbl, 1, PTR_TBL_t);
13523 tbl->tbl_max = 511;
13524 tbl->tbl_items = 0;
13525 tbl->tbl_arena = NULL;
13526 tbl->tbl_arena_next = NULL;
13527 tbl->tbl_arena_end = NULL;
13528 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13532 #define PTR_TABLE_HASH(ptr) \
13533 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13535 /* map an existing pointer using a table */
13537 STATIC PTR_TBL_ENT_t *
13538 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13540 PTR_TBL_ENT_t *tblent;
13541 const UV hash = PTR_TABLE_HASH(sv);
13543 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13545 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13546 for (; tblent; tblent = tblent->next) {
13547 if (tblent->oldval == sv)
13554 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13556 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13558 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13559 PERL_UNUSED_CONTEXT;
13561 return tblent ? tblent->newval : NULL;
13564 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13565 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13566 * the core's typical use of ptr_tables in thread cloning. */
13569 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13571 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13573 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13574 PERL_UNUSED_CONTEXT;
13577 tblent->newval = newsv;
13579 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13581 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13582 struct ptr_tbl_arena *new_arena;
13584 Newx(new_arena, 1, struct ptr_tbl_arena);
13585 new_arena->next = tbl->tbl_arena;
13586 tbl->tbl_arena = new_arena;
13587 tbl->tbl_arena_next = new_arena->array;
13588 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13591 tblent = tbl->tbl_arena_next++;
13593 tblent->oldval = oldsv;
13594 tblent->newval = newsv;
13595 tblent->next = tbl->tbl_ary[entry];
13596 tbl->tbl_ary[entry] = tblent;
13598 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13599 ptr_table_split(tbl);
13603 /* double the hash bucket size of an existing ptr table */
13606 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13608 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13609 const UV oldsize = tbl->tbl_max + 1;
13610 UV newsize = oldsize * 2;
13613 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13614 PERL_UNUSED_CONTEXT;
13616 Renew(ary, newsize, PTR_TBL_ENT_t*);
13617 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13618 tbl->tbl_max = --newsize;
13619 tbl->tbl_ary = ary;
13620 for (i=0; i < oldsize; i++, ary++) {
13621 PTR_TBL_ENT_t **entp = ary;
13622 PTR_TBL_ENT_t *ent = *ary;
13623 PTR_TBL_ENT_t **curentp;
13626 curentp = ary + oldsize;
13628 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13630 ent->next = *curentp;
13640 /* remove all the entries from a ptr table */
13641 /* Deprecated - will be removed post 5.14 */
13644 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13646 PERL_UNUSED_CONTEXT;
13647 if (tbl && tbl->tbl_items) {
13648 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13650 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13653 struct ptr_tbl_arena *next = arena->next;
13659 tbl->tbl_items = 0;
13660 tbl->tbl_arena = NULL;
13661 tbl->tbl_arena_next = NULL;
13662 tbl->tbl_arena_end = NULL;
13666 /* clear and free a ptr table */
13669 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13671 struct ptr_tbl_arena *arena;
13673 PERL_UNUSED_CONTEXT;
13679 arena = tbl->tbl_arena;
13682 struct ptr_tbl_arena *next = arena->next;
13688 Safefree(tbl->tbl_ary);
13692 #if defined(USE_ITHREADS)
13695 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13697 PERL_ARGS_ASSERT_RVPV_DUP;
13699 assert(!isREGEXP(sstr));
13701 if (SvWEAKREF(sstr)) {
13702 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13703 if (param->flags & CLONEf_JOIN_IN) {
13704 /* if joining, we add any back references individually rather
13705 * than copying the whole backref array */
13706 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13710 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13712 else if (SvPVX_const(sstr)) {
13713 /* Has something there */
13715 /* Normal PV - clone whole allocated space */
13716 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13717 /* sstr may not be that normal, but actually copy on write.
13718 But we are a true, independent SV, so: */
13722 /* Special case - not normally malloced for some reason */
13723 if (isGV_with_GP(sstr)) {
13724 /* Don't need to do anything here. */
13726 else if ((SvIsCOW(sstr))) {
13727 /* A "shared" PV - clone it as "shared" PV */
13729 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13733 /* Some other special case - random pointer */
13734 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13739 /* Copy the NULL */
13740 SvPV_set(dstr, NULL);
13744 /* duplicate a list of SVs. source and dest may point to the same memory. */
13746 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13747 SSize_t items, CLONE_PARAMS *const param)
13749 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13751 while (items-- > 0) {
13752 *dest++ = sv_dup_inc(*source++, param);
13758 /* duplicate an SV of any type (including AV, HV etc) */
13761 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13766 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13768 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13769 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13774 /* look for it in the table first */
13775 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13779 if(param->flags & CLONEf_JOIN_IN) {
13780 /** We are joining here so we don't want do clone
13781 something that is bad **/
13782 if (SvTYPE(sstr) == SVt_PVHV) {
13783 const HEK * const hvname = HvNAME_HEK(sstr);
13785 /** don't clone stashes if they already exist **/
13786 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13787 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13788 ptr_table_store(PL_ptr_table, sstr, dstr);
13792 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13793 HV *stash = GvSTASH(sstr);
13794 const HEK * hvname;
13795 if (stash && (hvname = HvNAME_HEK(stash))) {
13796 /** don't clone GVs if they already exist **/
13798 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13799 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13801 stash, GvNAME(sstr),
13807 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13808 ptr_table_store(PL_ptr_table, sstr, *svp);
13815 /* create anew and remember what it is */
13818 #ifdef DEBUG_LEAKING_SCALARS
13819 dstr->sv_debug_optype = sstr->sv_debug_optype;
13820 dstr->sv_debug_line = sstr->sv_debug_line;
13821 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13822 dstr->sv_debug_parent = (SV*)sstr;
13823 FREE_SV_DEBUG_FILE(dstr);
13824 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13827 ptr_table_store(PL_ptr_table, sstr, dstr);
13830 SvFLAGS(dstr) = SvFLAGS(sstr);
13831 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13832 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13835 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13836 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13837 (void*)PL_watch_pvx, SvPVX_const(sstr));
13840 /* don't clone objects whose class has asked us not to */
13842 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13848 switch (SvTYPE(sstr)) {
13850 SvANY(dstr) = NULL;
13853 SET_SVANY_FOR_BODYLESS_IV(dstr);
13855 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13857 SvIV_set(dstr, SvIVX(sstr));
13861 #if NVSIZE <= IVSIZE
13862 SET_SVANY_FOR_BODYLESS_NV(dstr);
13864 SvANY(dstr) = new_XNV();
13866 SvNV_set(dstr, SvNVX(sstr));
13870 /* These are all the types that need complex bodies allocating. */
13872 const svtype sv_type = SvTYPE(sstr);
13873 const struct body_details *const sv_type_details
13874 = bodies_by_type + sv_type;
13878 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13894 assert(sv_type_details->body_size);
13895 if (sv_type_details->arena) {
13896 new_body_inline(new_body, sv_type);
13898 = (void*)((char*)new_body - sv_type_details->offset);
13900 new_body = new_NOARENA(sv_type_details);
13904 SvANY(dstr) = new_body;
13907 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13908 ((char*)SvANY(dstr)) + sv_type_details->offset,
13909 sv_type_details->copy, char);
13911 Copy(((char*)SvANY(sstr)),
13912 ((char*)SvANY(dstr)),
13913 sv_type_details->body_size + sv_type_details->offset, char);
13916 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13917 && !isGV_with_GP(dstr)
13919 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13920 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13922 /* The Copy above means that all the source (unduplicated) pointers
13923 are now in the destination. We can check the flags and the
13924 pointers in either, but it's possible that there's less cache
13925 missing by always going for the destination.
13926 FIXME - instrument and check that assumption */
13927 if (sv_type >= SVt_PVMG) {
13929 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13930 if (SvOBJECT(dstr) && SvSTASH(dstr))
13931 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13932 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13935 /* The cast silences a GCC warning about unhandled types. */
13936 switch ((int)sv_type) {
13947 /* FIXME for plugins */
13948 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13949 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13952 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13953 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13954 LvTARG(dstr) = dstr;
13955 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13956 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13958 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13959 if (isREGEXP(sstr)) goto duprex;
13961 /* non-GP case already handled above */
13962 if(isGV_with_GP(sstr)) {
13963 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13964 /* Don't call sv_add_backref here as it's going to be
13965 created as part of the magic cloning of the symbol
13966 table--unless this is during a join and the stash
13967 is not actually being cloned. */
13968 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13969 at the point of this comment. */
13970 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13971 if (param->flags & CLONEf_JOIN_IN)
13972 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13973 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13974 (void)GpREFCNT_inc(GvGP(dstr));
13978 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13979 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13980 /* I have no idea why fake dirp (rsfps)
13981 should be treated differently but otherwise
13982 we end up with leaks -- sky*/
13983 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13984 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13985 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13987 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13988 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13989 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13990 if (IoDIRP(dstr)) {
13991 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13994 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13996 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13998 if (IoOFP(dstr) == IoIFP(sstr))
13999 IoOFP(dstr) = IoIFP(dstr);
14001 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14002 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14003 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14004 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14007 /* avoid cloning an empty array */
14008 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14009 SV **dst_ary, **src_ary;
14010 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14012 src_ary = AvARRAY((const AV *)sstr);
14013 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14014 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14015 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14016 AvALLOC((const AV *)dstr) = dst_ary;
14017 if (AvREAL((const AV *)sstr)) {
14018 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14022 while (items-- > 0)
14023 *dst_ary++ = sv_dup(*src_ary++, param);
14025 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14026 while (items-- > 0) {
14031 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14032 AvALLOC((const AV *)dstr) = (SV**)NULL;
14033 AvMAX( (const AV *)dstr) = -1;
14034 AvFILLp((const AV *)dstr) = -1;
14038 if (HvARRAY((const HV *)sstr)) {
14040 const bool sharekeys = !!HvSHAREKEYS(sstr);
14041 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14042 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14044 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14045 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14047 HvARRAY(dstr) = (HE**)darray;
14048 while (i <= sxhv->xhv_max) {
14049 const HE * const source = HvARRAY(sstr)[i];
14050 HvARRAY(dstr)[i] = source
14051 ? he_dup(source, sharekeys, param) : 0;
14055 const struct xpvhv_aux * const saux = HvAUX(sstr);
14056 struct xpvhv_aux * const daux = HvAUX(dstr);
14057 /* This flag isn't copied. */
14060 if (saux->xhv_name_count) {
14061 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14063 = saux->xhv_name_count < 0
14064 ? -saux->xhv_name_count
14065 : saux->xhv_name_count;
14066 HEK **shekp = sname + count;
14068 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14069 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14070 while (shekp-- > sname) {
14072 *dhekp = hek_dup(*shekp, param);
14076 daux->xhv_name_u.xhvnameu_name
14077 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14080 daux->xhv_name_count = saux->xhv_name_count;
14082 daux->xhv_aux_flags = saux->xhv_aux_flags;
14083 #ifdef PERL_HASH_RANDOMIZE_KEYS
14084 daux->xhv_rand = saux->xhv_rand;
14085 daux->xhv_last_rand = saux->xhv_last_rand;
14087 daux->xhv_riter = saux->xhv_riter;
14088 daux->xhv_eiter = saux->xhv_eiter
14089 ? he_dup(saux->xhv_eiter,
14090 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14091 /* backref array needs refcnt=2; see sv_add_backref */
14092 daux->xhv_backreferences =
14093 (param->flags & CLONEf_JOIN_IN)
14094 /* when joining, we let the individual GVs and
14095 * CVs add themselves to backref as
14096 * needed. This avoids pulling in stuff
14097 * that isn't required, and simplifies the
14098 * case where stashes aren't cloned back
14099 * if they already exist in the parent
14102 : saux->xhv_backreferences
14103 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14104 ? MUTABLE_AV(SvREFCNT_inc(
14105 sv_dup_inc((const SV *)
14106 saux->xhv_backreferences, param)))
14107 : MUTABLE_AV(sv_dup((const SV *)
14108 saux->xhv_backreferences, param))
14111 daux->xhv_mro_meta = saux->xhv_mro_meta
14112 ? mro_meta_dup(saux->xhv_mro_meta, param)
14115 /* Record stashes for possible cloning in Perl_clone(). */
14117 av_push(param->stashes, dstr);
14121 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14124 if (!(param->flags & CLONEf_COPY_STACKS)) {
14129 /* NOTE: not refcounted */
14130 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14131 hv_dup(CvSTASH(dstr), param);
14132 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14133 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14134 if (!CvISXSUB(dstr)) {
14136 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14138 CvSLABBED_off(dstr);
14139 } else if (CvCONST(dstr)) {
14140 CvXSUBANY(dstr).any_ptr =
14141 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14143 assert(!CvSLABBED(dstr));
14144 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14146 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14147 hek_dup(CvNAME_HEK((CV *)sstr), param);
14148 /* don't dup if copying back - CvGV isn't refcounted, so the
14149 * duped GV may never be freed. A bit of a hack! DAPM */
14151 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14153 ? gv_dup_inc(CvGV(sstr), param)
14154 : (param->flags & CLONEf_JOIN_IN)
14156 : gv_dup(CvGV(sstr), param);
14158 if (!CvISXSUB(sstr)) {
14159 PADLIST * padlist = CvPADLIST(sstr);
14161 padlist = padlist_dup(padlist, param);
14162 CvPADLIST_set(dstr, padlist);
14164 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14165 PoisonPADLIST(dstr);
14168 CvWEAKOUTSIDE(sstr)
14169 ? cv_dup( CvOUTSIDE(dstr), param)
14170 : cv_dup_inc(CvOUTSIDE(dstr), param);
14180 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14182 PERL_ARGS_ASSERT_SV_DUP_INC;
14183 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14187 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14189 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14190 PERL_ARGS_ASSERT_SV_DUP;
14192 /* Track every SV that (at least initially) had a reference count of 0.
14193 We need to do this by holding an actual reference to it in this array.
14194 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14195 (akin to the stashes hash, and the perl stack), we come unstuck if
14196 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14197 thread) is manipulated in a CLONE method, because CLONE runs before the
14198 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14199 (and fix things up by giving each a reference via the temps stack).
14200 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14201 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14202 before the walk of unreferenced happens and a reference to that is SV
14203 added to the temps stack. At which point we have the same SV considered
14204 to be in use, and free to be re-used. Not good.
14206 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14207 assert(param->unreferenced);
14208 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14214 /* duplicate a context */
14217 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14219 PERL_CONTEXT *ncxs;
14221 PERL_ARGS_ASSERT_CX_DUP;
14224 return (PERL_CONTEXT*)NULL;
14226 /* look for it in the table first */
14227 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14231 /* create anew and remember what it is */
14232 Newx(ncxs, max + 1, PERL_CONTEXT);
14233 ptr_table_store(PL_ptr_table, cxs, ncxs);
14234 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14237 PERL_CONTEXT * const ncx = &ncxs[ix];
14238 if (CxTYPE(ncx) == CXt_SUBST) {
14239 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14242 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14243 switch (CxTYPE(ncx)) {
14245 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14246 if(CxHASARGS(ncx)){
14247 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14249 ncx->blk_sub.savearray = NULL;
14251 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14252 ncx->blk_sub.prevcomppad);
14255 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14257 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14258 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14259 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14260 /* XXX what do do with cur_top_env ???? */
14262 case CXt_LOOP_LAZYSV:
14263 ncx->blk_loop.state_u.lazysv.end
14264 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14265 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14266 duplication code instead.
14267 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14268 actually being the same function, and (2) order
14269 equivalence of the two unions.
14270 We can assert the later [but only at run time :-(] */
14271 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14272 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14275 ncx->blk_loop.state_u.ary.ary
14276 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14278 case CXt_LOOP_LIST:
14279 case CXt_LOOP_LAZYIV:
14280 /* code common to all 'for' CXt_LOOP_* types */
14281 ncx->blk_loop.itersave =
14282 sv_dup_inc(ncx->blk_loop.itersave, param);
14283 if (CxPADLOOP(ncx)) {
14284 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14285 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14286 ncx->blk_loop.oldcomppad =
14287 (PAD*)ptr_table_fetch(PL_ptr_table,
14288 ncx->blk_loop.oldcomppad);
14289 ncx->blk_loop.itervar_u.svp =
14290 &CX_CURPAD_SV(ncx->blk_loop, off);
14293 /* this copies the GV if CXp_FOR_GV, or the SV for an
14294 * alias (for \$x (...)) - relies on gv_dup being the
14295 * same as sv_dup */
14296 ncx->blk_loop.itervar_u.gv
14297 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14301 case CXt_LOOP_PLAIN:
14304 ncx->blk_format.prevcomppad =
14305 (PAD*)ptr_table_fetch(PL_ptr_table,
14306 ncx->blk_format.prevcomppad);
14307 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14308 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14309 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14313 ncx->blk_givwhen.defsv_save =
14314 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14327 /* duplicate a stack info structure */
14330 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14334 PERL_ARGS_ASSERT_SI_DUP;
14337 return (PERL_SI*)NULL;
14339 /* look for it in the table first */
14340 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14344 /* create anew and remember what it is */
14345 Newxz(nsi, 1, PERL_SI);
14346 ptr_table_store(PL_ptr_table, si, nsi);
14348 nsi->si_stack = av_dup_inc(si->si_stack, param);
14349 nsi->si_cxix = si->si_cxix;
14350 nsi->si_cxmax = si->si_cxmax;
14351 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14352 nsi->si_type = si->si_type;
14353 nsi->si_prev = si_dup(si->si_prev, param);
14354 nsi->si_next = si_dup(si->si_next, param);
14355 nsi->si_markoff = si->si_markoff;
14360 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14361 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14362 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14363 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14364 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14365 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14366 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14367 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14368 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14369 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14370 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14371 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14372 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14373 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14374 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14375 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14378 #define pv_dup_inc(p) SAVEPV(p)
14379 #define pv_dup(p) SAVEPV(p)
14380 #define svp_dup_inc(p,pp) any_dup(p,pp)
14382 /* map any object to the new equivent - either something in the
14383 * ptr table, or something in the interpreter structure
14387 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14391 PERL_ARGS_ASSERT_ANY_DUP;
14394 return (void*)NULL;
14396 /* look for it in the table first */
14397 ret = ptr_table_fetch(PL_ptr_table, v);
14401 /* see if it is part of the interpreter structure */
14402 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14403 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14411 /* duplicate the save stack */
14414 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14417 ANY * const ss = proto_perl->Isavestack;
14418 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14419 I32 ix = proto_perl->Isavestack_ix;
14432 void (*dptr) (void*);
14433 void (*dxptr) (pTHX_ void*);
14435 PERL_ARGS_ASSERT_SS_DUP;
14437 Newxz(nss, max, ANY);
14440 const UV uv = POPUV(ss,ix);
14441 const U8 type = (U8)uv & SAVE_MASK;
14443 TOPUV(nss,ix) = uv;
14445 case SAVEt_CLEARSV:
14446 case SAVEt_CLEARPADRANGE:
14448 case SAVEt_HELEM: /* hash element */
14449 case SAVEt_SV: /* scalar reference */
14450 sv = (const SV *)POPPTR(ss,ix);
14451 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14453 case SAVEt_ITEM: /* normal string */
14454 case SAVEt_GVSV: /* scalar slot in GV */
14455 sv = (const SV *)POPPTR(ss,ix);
14456 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14457 if (type == SAVEt_SV)
14461 case SAVEt_MORTALIZESV:
14462 case SAVEt_READONLY_OFF:
14463 sv = (const SV *)POPPTR(ss,ix);
14464 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14466 case SAVEt_FREEPADNAME:
14467 ptr = POPPTR(ss,ix);
14468 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14469 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14471 case SAVEt_SHARED_PVREF: /* char* in shared space */
14472 c = (char*)POPPTR(ss,ix);
14473 TOPPTR(nss,ix) = savesharedpv(c);
14474 ptr = POPPTR(ss,ix);
14475 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14477 case SAVEt_GENERIC_SVREF: /* generic sv */
14478 case SAVEt_SVREF: /* scalar reference */
14479 sv = (const SV *)POPPTR(ss,ix);
14480 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14481 if (type == SAVEt_SVREF)
14482 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14483 ptr = POPPTR(ss,ix);
14484 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14486 case SAVEt_GVSLOT: /* any slot in GV */
14487 sv = (const SV *)POPPTR(ss,ix);
14488 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14489 ptr = POPPTR(ss,ix);
14490 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14491 sv = (const SV *)POPPTR(ss,ix);
14492 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14494 case SAVEt_HV: /* hash reference */
14495 case SAVEt_AV: /* array reference */
14496 sv = (const SV *) POPPTR(ss,ix);
14497 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14499 case SAVEt_COMPPAD:
14501 sv = (const SV *) POPPTR(ss,ix);
14502 TOPPTR(nss,ix) = sv_dup(sv, param);
14504 case SAVEt_INT: /* int reference */
14505 ptr = POPPTR(ss,ix);
14506 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14507 intval = (int)POPINT(ss,ix);
14508 TOPINT(nss,ix) = intval;
14510 case SAVEt_LONG: /* long reference */
14511 ptr = POPPTR(ss,ix);
14512 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14513 longval = (long)POPLONG(ss,ix);
14514 TOPLONG(nss,ix) = longval;
14516 case SAVEt_I32: /* I32 reference */
14517 ptr = POPPTR(ss,ix);
14518 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14520 TOPINT(nss,ix) = i;
14522 case SAVEt_IV: /* IV reference */
14523 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14524 ptr = POPPTR(ss,ix);
14525 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14527 TOPIV(nss,ix) = iv;
14529 case SAVEt_TMPSFLOOR:
14531 TOPIV(nss,ix) = iv;
14533 case SAVEt_HPTR: /* HV* reference */
14534 case SAVEt_APTR: /* AV* reference */
14535 case SAVEt_SPTR: /* SV* reference */
14536 ptr = POPPTR(ss,ix);
14537 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14538 sv = (const SV *)POPPTR(ss,ix);
14539 TOPPTR(nss,ix) = sv_dup(sv, param);
14541 case SAVEt_VPTR: /* random* reference */
14542 ptr = POPPTR(ss,ix);
14543 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14545 case SAVEt_INT_SMALL:
14546 case SAVEt_I32_SMALL:
14547 case SAVEt_I16: /* I16 reference */
14548 case SAVEt_I8: /* I8 reference */
14550 ptr = POPPTR(ss,ix);
14551 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14553 case SAVEt_GENERIC_PVREF: /* generic char* */
14554 case SAVEt_PPTR: /* char* reference */
14555 ptr = POPPTR(ss,ix);
14556 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14557 c = (char*)POPPTR(ss,ix);
14558 TOPPTR(nss,ix) = pv_dup(c);
14560 case SAVEt_GP: /* scalar reference */
14561 gp = (GP*)POPPTR(ss,ix);
14562 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14563 (void)GpREFCNT_inc(gp);
14564 gv = (const GV *)POPPTR(ss,ix);
14565 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14568 ptr = POPPTR(ss,ix);
14569 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14570 /* these are assumed to be refcounted properly */
14572 switch (((OP*)ptr)->op_type) {
14574 case OP_LEAVESUBLV:
14578 case OP_LEAVEWRITE:
14579 TOPPTR(nss,ix) = ptr;
14582 (void) OpREFCNT_inc(o);
14586 TOPPTR(nss,ix) = NULL;
14591 TOPPTR(nss,ix) = NULL;
14593 case SAVEt_FREECOPHH:
14594 ptr = POPPTR(ss,ix);
14595 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14597 case SAVEt_ADELETE:
14598 av = (const AV *)POPPTR(ss,ix);
14599 TOPPTR(nss,ix) = av_dup_inc(av, param);
14601 TOPINT(nss,ix) = i;
14604 hv = (const HV *)POPPTR(ss,ix);
14605 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14607 TOPINT(nss,ix) = i;
14610 c = (char*)POPPTR(ss,ix);
14611 TOPPTR(nss,ix) = pv_dup_inc(c);
14613 case SAVEt_STACK_POS: /* Position on Perl stack */
14615 TOPINT(nss,ix) = i;
14617 case SAVEt_DESTRUCTOR:
14618 ptr = POPPTR(ss,ix);
14619 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14620 dptr = POPDPTR(ss,ix);
14621 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14622 any_dup(FPTR2DPTR(void *, dptr),
14625 case SAVEt_DESTRUCTOR_X:
14626 ptr = POPPTR(ss,ix);
14627 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14628 dxptr = POPDXPTR(ss,ix);
14629 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14630 any_dup(FPTR2DPTR(void *, dxptr),
14633 case SAVEt_REGCONTEXT:
14635 ix -= uv >> SAVE_TIGHT_SHIFT;
14637 case SAVEt_AELEM: /* array element */
14638 sv = (const SV *)POPPTR(ss,ix);
14639 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14641 TOPINT(nss,ix) = i;
14642 av = (const AV *)POPPTR(ss,ix);
14643 TOPPTR(nss,ix) = av_dup_inc(av, param);
14646 ptr = POPPTR(ss,ix);
14647 TOPPTR(nss,ix) = ptr;
14650 ptr = POPPTR(ss,ix);
14651 ptr = cophh_copy((COPHH*)ptr);
14652 TOPPTR(nss,ix) = ptr;
14654 TOPINT(nss,ix) = i;
14655 if (i & HINT_LOCALIZE_HH) {
14656 hv = (const HV *)POPPTR(ss,ix);
14657 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14660 case SAVEt_PADSV_AND_MORTALIZE:
14661 longval = (long)POPLONG(ss,ix);
14662 TOPLONG(nss,ix) = longval;
14663 ptr = POPPTR(ss,ix);
14664 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14665 sv = (const SV *)POPPTR(ss,ix);
14666 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14668 case SAVEt_SET_SVFLAGS:
14670 TOPINT(nss,ix) = i;
14672 TOPINT(nss,ix) = i;
14673 sv = (const SV *)POPPTR(ss,ix);
14674 TOPPTR(nss,ix) = sv_dup(sv, param);
14676 case SAVEt_COMPILE_WARNINGS:
14677 ptr = POPPTR(ss,ix);
14678 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14681 ptr = POPPTR(ss,ix);
14682 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14686 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14694 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14695 * flag to the result. This is done for each stash before cloning starts,
14696 * so we know which stashes want their objects cloned */
14699 do_mark_cloneable_stash(pTHX_ SV *const sv)
14701 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14703 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14704 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14705 if (cloner && GvCV(cloner)) {
14712 mXPUSHs(newSVhek(hvname));
14714 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14721 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14729 =for apidoc perl_clone
14731 Create and return a new interpreter by cloning the current one.
14733 C<perl_clone> takes these flags as parameters:
14735 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14736 without it we only clone the data and zero the stacks,
14737 with it we copy the stacks and the new perl interpreter is
14738 ready to run at the exact same point as the previous one.
14739 The pseudo-fork code uses C<COPY_STACKS> while the
14740 threads->create doesn't.
14742 C<CLONEf_KEEP_PTR_TABLE> -
14743 C<perl_clone> keeps a ptr_table with the pointer of the old
14744 variable as a key and the new variable as a value,
14745 this allows it to check if something has been cloned and not
14746 clone it again but rather just use the value and increase the
14747 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14748 the ptr_table using the function
14749 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14750 reason to keep it around is if you want to dup some of your own
14751 variable who are outside the graph perl scans, an example of this
14752 code is in F<threads.xs> create.
14754 C<CLONEf_CLONE_HOST> -
14755 This is a win32 thing, it is ignored on unix, it tells perls
14756 win32host code (which is c++) to clone itself, this is needed on
14757 win32 if you want to run two threads at the same time,
14758 if you just want to do some stuff in a separate perl interpreter
14759 and then throw it away and return to the original one,
14760 you don't need to do anything.
14765 /* XXX the above needs expanding by someone who actually understands it ! */
14766 EXTERN_C PerlInterpreter *
14767 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14770 perl_clone(PerlInterpreter *proto_perl, UV flags)
14773 #ifdef PERL_IMPLICIT_SYS
14775 PERL_ARGS_ASSERT_PERL_CLONE;
14777 /* perlhost.h so we need to call into it
14778 to clone the host, CPerlHost should have a c interface, sky */
14780 #ifndef __amigaos4__
14781 if (flags & CLONEf_CLONE_HOST) {
14782 return perl_clone_host(proto_perl,flags);
14785 return perl_clone_using(proto_perl, flags,
14787 proto_perl->IMemShared,
14788 proto_perl->IMemParse,
14790 proto_perl->IStdIO,
14794 proto_perl->IProc);
14798 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14799 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14800 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14801 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14802 struct IPerlDir* ipD, struct IPerlSock* ipS,
14803 struct IPerlProc* ipP)
14805 /* XXX many of the string copies here can be optimized if they're
14806 * constants; they need to be allocated as common memory and just
14807 * their pointers copied. */
14810 CLONE_PARAMS clone_params;
14811 CLONE_PARAMS* const param = &clone_params;
14813 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14815 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14816 #else /* !PERL_IMPLICIT_SYS */
14818 CLONE_PARAMS clone_params;
14819 CLONE_PARAMS* param = &clone_params;
14820 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14822 PERL_ARGS_ASSERT_PERL_CLONE;
14823 #endif /* PERL_IMPLICIT_SYS */
14825 /* for each stash, determine whether its objects should be cloned */
14826 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14827 PERL_SET_THX(my_perl);
14830 PoisonNew(my_perl, 1, PerlInterpreter);
14833 PL_defstash = NULL; /* may be used by perl malloc() */
14836 PL_scopestack_name = 0;
14838 PL_savestack_ix = 0;
14839 PL_savestack_max = -1;
14840 PL_sig_pending = 0;
14842 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14843 Zero(&PL_padname_undef, 1, PADNAME);
14844 Zero(&PL_padname_const, 1, PADNAME);
14845 # ifdef DEBUG_LEAKING_SCALARS
14846 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14848 # ifdef PERL_TRACE_OPS
14849 Zero(PL_op_exec_cnt, OP_max+2, UV);
14851 #else /* !DEBUGGING */
14852 Zero(my_perl, 1, PerlInterpreter);
14853 #endif /* DEBUGGING */
14855 #ifdef PERL_IMPLICIT_SYS
14856 /* host pointers */
14858 PL_MemShared = ipMS;
14859 PL_MemParse = ipMP;
14866 #endif /* PERL_IMPLICIT_SYS */
14869 param->flags = flags;
14870 /* Nothing in the core code uses this, but we make it available to
14871 extensions (using mg_dup). */
14872 param->proto_perl = proto_perl;
14873 /* Likely nothing will use this, but it is initialised to be consistent
14874 with Perl_clone_params_new(). */
14875 param->new_perl = my_perl;
14876 param->unreferenced = NULL;
14879 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14881 PL_body_arenas = NULL;
14882 Zero(&PL_body_roots, 1, PL_body_roots);
14886 PL_sv_arenaroot = NULL;
14888 PL_debug = proto_perl->Idebug;
14890 /* dbargs array probably holds garbage */
14893 PL_compiling = proto_perl->Icompiling;
14895 /* pseudo environmental stuff */
14896 PL_origargc = proto_perl->Iorigargc;
14897 PL_origargv = proto_perl->Iorigargv;
14899 #ifndef NO_TAINT_SUPPORT
14900 /* Set tainting stuff before PerlIO_debug can possibly get called */
14901 PL_tainting = proto_perl->Itainting;
14902 PL_taint_warn = proto_perl->Itaint_warn;
14904 PL_tainting = FALSE;
14905 PL_taint_warn = FALSE;
14908 PL_minus_c = proto_perl->Iminus_c;
14910 PL_localpatches = proto_perl->Ilocalpatches;
14911 PL_splitstr = proto_perl->Isplitstr;
14912 PL_minus_n = proto_perl->Iminus_n;
14913 PL_minus_p = proto_perl->Iminus_p;
14914 PL_minus_l = proto_perl->Iminus_l;
14915 PL_minus_a = proto_perl->Iminus_a;
14916 PL_minus_E = proto_perl->Iminus_E;
14917 PL_minus_F = proto_perl->Iminus_F;
14918 PL_doswitches = proto_perl->Idoswitches;
14919 PL_dowarn = proto_perl->Idowarn;
14920 #ifdef PERL_SAWAMPERSAND
14921 PL_sawampersand = proto_perl->Isawampersand;
14923 PL_unsafe = proto_perl->Iunsafe;
14924 PL_perldb = proto_perl->Iperldb;
14925 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14926 PL_exit_flags = proto_perl->Iexit_flags;
14928 /* XXX time(&PL_basetime) when asked for? */
14929 PL_basetime = proto_perl->Ibasetime;
14931 PL_maxsysfd = proto_perl->Imaxsysfd;
14932 PL_statusvalue = proto_perl->Istatusvalue;
14934 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14936 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14939 /* RE engine related */
14940 PL_regmatch_slab = NULL;
14941 PL_reg_curpm = NULL;
14943 PL_sub_generation = proto_perl->Isub_generation;
14945 /* funky return mechanisms */
14946 PL_forkprocess = proto_perl->Iforkprocess;
14948 /* internal state */
14949 PL_main_start = proto_perl->Imain_start;
14950 PL_eval_root = proto_perl->Ieval_root;
14951 PL_eval_start = proto_perl->Ieval_start;
14953 PL_filemode = proto_perl->Ifilemode;
14954 PL_lastfd = proto_perl->Ilastfd;
14955 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14958 PL_gensym = proto_perl->Igensym;
14960 PL_laststatval = proto_perl->Ilaststatval;
14961 PL_laststype = proto_perl->Ilaststype;
14964 PL_profiledata = NULL;
14966 PL_generation = proto_perl->Igeneration;
14968 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14969 PL_in_clean_all = proto_perl->Iin_clean_all;
14971 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14972 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14973 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14974 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14975 PL_nomemok = proto_perl->Inomemok;
14976 PL_an = proto_perl->Ian;
14977 PL_evalseq = proto_perl->Ievalseq;
14978 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14979 PL_origalen = proto_perl->Iorigalen;
14981 PL_sighandlerp = proto_perl->Isighandlerp;
14983 PL_runops = proto_perl->Irunops;
14985 PL_subline = proto_perl->Isubline;
14987 PL_cv_has_eval = proto_perl->Icv_has_eval;
14990 PL_cryptseen = proto_perl->Icryptseen;
14993 #ifdef USE_LOCALE_COLLATE
14994 PL_collation_ix = proto_perl->Icollation_ix;
14995 PL_collation_standard = proto_perl->Icollation_standard;
14996 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14997 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14998 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14999 #endif /* USE_LOCALE_COLLATE */
15001 #ifdef USE_LOCALE_NUMERIC
15002 PL_numeric_standard = proto_perl->Inumeric_standard;
15003 PL_numeric_local = proto_perl->Inumeric_local;
15004 #endif /* !USE_LOCALE_NUMERIC */
15006 /* Did the locale setup indicate UTF-8? */
15007 PL_utf8locale = proto_perl->Iutf8locale;
15008 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15009 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15010 /* Unicode features (see perlrun/-C) */
15011 PL_unicode = proto_perl->Iunicode;
15013 /* Pre-5.8 signals control */
15014 PL_signals = proto_perl->Isignals;
15016 /* times() ticks per second */
15017 PL_clocktick = proto_perl->Iclocktick;
15019 /* Recursion stopper for PerlIO_find_layer */
15020 PL_in_load_module = proto_perl->Iin_load_module;
15022 /* sort() routine */
15023 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15025 /* Not really needed/useful since the reenrant_retint is "volatile",
15026 * but do it for consistency's sake. */
15027 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15029 /* Hooks to shared SVs and locks. */
15030 PL_sharehook = proto_perl->Isharehook;
15031 PL_lockhook = proto_perl->Ilockhook;
15032 PL_unlockhook = proto_perl->Iunlockhook;
15033 PL_threadhook = proto_perl->Ithreadhook;
15034 PL_destroyhook = proto_perl->Idestroyhook;
15035 PL_signalhook = proto_perl->Isignalhook;
15037 PL_globhook = proto_perl->Iglobhook;
15040 PL_last_swash_hv = NULL; /* reinits on demand */
15041 PL_last_swash_klen = 0;
15042 PL_last_swash_key[0]= '\0';
15043 PL_last_swash_tmps = (U8*)NULL;
15044 PL_last_swash_slen = 0;
15046 PL_srand_called = proto_perl->Isrand_called;
15047 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15049 if (flags & CLONEf_COPY_STACKS) {
15050 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15051 PL_tmps_ix = proto_perl->Itmps_ix;
15052 PL_tmps_max = proto_perl->Itmps_max;
15053 PL_tmps_floor = proto_perl->Itmps_floor;
15055 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15056 * NOTE: unlike the others! */
15057 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15058 PL_scopestack_max = proto_perl->Iscopestack_max;
15060 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15061 * NOTE: unlike the others! */
15062 PL_savestack_ix = proto_perl->Isavestack_ix;
15063 PL_savestack_max = proto_perl->Isavestack_max;
15066 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15067 PL_top_env = &PL_start_env;
15069 PL_op = proto_perl->Iop;
15072 PL_Xpv = (XPV*)NULL;
15073 my_perl->Ina = proto_perl->Ina;
15075 PL_statbuf = proto_perl->Istatbuf;
15076 PL_statcache = proto_perl->Istatcache;
15078 #ifndef NO_TAINT_SUPPORT
15079 PL_tainted = proto_perl->Itainted;
15081 PL_tainted = FALSE;
15083 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15085 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15087 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15088 PL_restartop = proto_perl->Irestartop;
15089 PL_in_eval = proto_perl->Iin_eval;
15090 PL_delaymagic = proto_perl->Idelaymagic;
15091 PL_phase = proto_perl->Iphase;
15092 PL_localizing = proto_perl->Ilocalizing;
15094 PL_hv_fetch_ent_mh = NULL;
15095 PL_modcount = proto_perl->Imodcount;
15096 PL_lastgotoprobe = NULL;
15097 PL_dumpindent = proto_perl->Idumpindent;
15099 PL_efloatbuf = NULL; /* reinits on demand */
15100 PL_efloatsize = 0; /* reinits on demand */
15104 PL_colorset = 0; /* reinits PL_colors[] */
15105 /*PL_colors[6] = {0,0,0,0,0,0};*/
15107 /* Pluggable optimizer */
15108 PL_peepp = proto_perl->Ipeepp;
15109 PL_rpeepp = proto_perl->Irpeepp;
15110 /* op_free() hook */
15111 PL_opfreehook = proto_perl->Iopfreehook;
15113 #ifdef USE_REENTRANT_API
15114 /* XXX: things like -Dm will segfault here in perlio, but doing
15115 * PERL_SET_CONTEXT(proto_perl);
15116 * breaks too many other things
15118 Perl_reentrant_init(aTHX);
15121 /* create SV map for pointer relocation */
15122 PL_ptr_table = ptr_table_new();
15124 /* initialize these special pointers as early as possible */
15126 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15127 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15128 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15129 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15130 &PL_padname_const);
15132 /* create (a non-shared!) shared string table */
15133 PL_strtab = newHV();
15134 HvSHAREKEYS_off(PL_strtab);
15135 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15136 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15138 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15140 /* This PV will be free'd special way so must set it same way op.c does */
15141 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15142 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15144 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15145 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15146 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15147 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15149 param->stashes = newAV(); /* Setup array of objects to call clone on */
15150 /* This makes no difference to the implementation, as it always pushes
15151 and shifts pointers to other SVs without changing their reference
15152 count, with the array becoming empty before it is freed. However, it
15153 makes it conceptually clear what is going on, and will avoid some
15154 work inside av.c, filling slots between AvFILL() and AvMAX() with
15155 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15156 AvREAL_off(param->stashes);
15158 if (!(flags & CLONEf_COPY_STACKS)) {
15159 param->unreferenced = newAV();
15162 #ifdef PERLIO_LAYERS
15163 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15164 PerlIO_clone(aTHX_ proto_perl, param);
15167 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15168 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15169 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15170 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15171 PL_xsubfilename = proto_perl->Ixsubfilename;
15172 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15173 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15176 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15177 PL_inplace = SAVEPV(proto_perl->Iinplace);
15178 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15180 /* magical thingies */
15182 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15183 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15184 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15187 /* Clone the regex array */
15188 /* ORANGE FIXME for plugins, probably in the SV dup code.
15189 newSViv(PTR2IV(CALLREGDUPE(
15190 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15192 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15193 PL_regex_pad = AvARRAY(PL_regex_padav);
15195 PL_stashpadmax = proto_perl->Istashpadmax;
15196 PL_stashpadix = proto_perl->Istashpadix ;
15197 Newx(PL_stashpad, PL_stashpadmax, HV *);
15200 for (; o < PL_stashpadmax; ++o)
15201 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15204 /* shortcuts to various I/O objects */
15205 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15206 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15207 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15208 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15209 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15210 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15211 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15213 /* shortcuts to regexp stuff */
15214 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15216 /* shortcuts to misc objects */
15217 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15219 /* shortcuts to debugging objects */
15220 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15221 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15222 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15223 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15224 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15225 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15226 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15228 /* symbol tables */
15229 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15230 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15231 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15232 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15233 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15235 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15236 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15237 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15238 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15239 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15240 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15241 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15242 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15243 PL_savebegin = proto_perl->Isavebegin;
15245 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15247 /* subprocess state */
15248 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15250 if (proto_perl->Iop_mask)
15251 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15254 /* PL_asserting = proto_perl->Iasserting; */
15256 /* current interpreter roots */
15257 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15259 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15262 /* runtime control stuff */
15263 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15265 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15267 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15269 /* interpreter atexit processing */
15270 PL_exitlistlen = proto_perl->Iexitlistlen;
15271 if (PL_exitlistlen) {
15272 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15273 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15276 PL_exitlist = (PerlExitListEntry*)NULL;
15278 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15279 if (PL_my_cxt_size) {
15280 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15281 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15282 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15283 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15284 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15288 PL_my_cxt_list = (void**)NULL;
15289 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15290 PL_my_cxt_keys = (const char**)NULL;
15293 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15294 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15295 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15296 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15298 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15300 PAD_CLONE_VARS(proto_perl, param);
15302 #ifdef HAVE_INTERP_INTERN
15303 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15306 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15308 #ifdef PERL_USES_PL_PIDSTATUS
15309 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15311 PL_osname = SAVEPV(proto_perl->Iosname);
15312 PL_parser = parser_dup(proto_perl->Iparser, param);
15314 /* XXX this only works if the saved cop has already been cloned */
15315 if (proto_perl->Iparser) {
15316 PL_parser->saved_curcop = (COP*)any_dup(
15317 proto_perl->Iparser->saved_curcop,
15321 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15323 #ifdef USE_LOCALE_CTYPE
15324 /* Should we warn if uses locale? */
15325 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15328 #ifdef USE_LOCALE_COLLATE
15329 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15330 #endif /* USE_LOCALE_COLLATE */
15332 #ifdef USE_LOCALE_NUMERIC
15333 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15334 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15335 #endif /* !USE_LOCALE_NUMERIC */
15337 /* Unicode inversion lists */
15338 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15339 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15340 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15341 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15343 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15344 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15346 /* utf8 character class swashes */
15347 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15348 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15350 for (i = 0; i < POSIX_CC_COUNT; i++) {
15351 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15353 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15354 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15355 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15356 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15357 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15358 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15359 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15360 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15361 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15362 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15363 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15364 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15365 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15366 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15367 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15368 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15369 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15370 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15372 if (proto_perl->Ipsig_pend) {
15373 Newxz(PL_psig_pend, SIG_SIZE, int);
15376 PL_psig_pend = (int*)NULL;
15379 if (proto_perl->Ipsig_name) {
15380 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15381 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15383 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15386 PL_psig_ptr = (SV**)NULL;
15387 PL_psig_name = (SV**)NULL;
15390 if (flags & CLONEf_COPY_STACKS) {
15391 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15392 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15393 PL_tmps_ix+1, param);
15395 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15396 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15397 Newxz(PL_markstack, i, I32);
15398 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15399 - proto_perl->Imarkstack);
15400 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15401 - proto_perl->Imarkstack);
15402 Copy(proto_perl->Imarkstack, PL_markstack,
15403 PL_markstack_ptr - PL_markstack + 1, I32);
15405 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15406 * NOTE: unlike the others! */
15407 Newxz(PL_scopestack, PL_scopestack_max, I32);
15408 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15411 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15412 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15414 /* reset stack AV to correct length before its duped via
15415 * PL_curstackinfo */
15416 AvFILLp(proto_perl->Icurstack) =
15417 proto_perl->Istack_sp - proto_perl->Istack_base;
15419 /* NOTE: si_dup() looks at PL_markstack */
15420 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15422 /* PL_curstack = PL_curstackinfo->si_stack; */
15423 PL_curstack = av_dup(proto_perl->Icurstack, param);
15424 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15426 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15427 PL_stack_base = AvARRAY(PL_curstack);
15428 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15429 - proto_perl->Istack_base);
15430 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15432 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15433 PL_savestack = ss_dup(proto_perl, param);
15437 ENTER; /* perl_destruct() wants to LEAVE; */
15440 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15441 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15443 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15444 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15445 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15446 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15447 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15448 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15450 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15452 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15453 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15454 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15456 PL_stashcache = newHV();
15458 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15459 proto_perl->Iwatchaddr);
15460 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15461 if (PL_debug && PL_watchaddr) {
15462 PerlIO_printf(Perl_debug_log,
15463 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15464 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15465 PTR2UV(PL_watchok));
15468 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15469 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15470 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15472 /* Call the ->CLONE method, if it exists, for each of the stashes
15473 identified by sv_dup() above.
15475 while(av_tindex(param->stashes) != -1) {
15476 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15477 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15478 if (cloner && GvCV(cloner)) {
15483 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15485 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15491 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15492 ptr_table_free(PL_ptr_table);
15493 PL_ptr_table = NULL;
15496 if (!(flags & CLONEf_COPY_STACKS)) {
15497 unreferenced_to_tmp_stack(param->unreferenced);
15500 SvREFCNT_dec(param->stashes);
15502 /* orphaned? eg threads->new inside BEGIN or use */
15503 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15504 SvREFCNT_inc_simple_void(PL_compcv);
15505 SAVEFREESV(PL_compcv);
15512 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15514 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15516 if (AvFILLp(unreferenced) > -1) {
15517 SV **svp = AvARRAY(unreferenced);
15518 SV **const last = svp + AvFILLp(unreferenced);
15522 if (SvREFCNT(*svp) == 1)
15524 } while (++svp <= last);
15526 EXTEND_MORTAL(count);
15527 svp = AvARRAY(unreferenced);
15530 if (SvREFCNT(*svp) == 1) {
15531 /* Our reference is the only one to this SV. This means that
15532 in this thread, the scalar effectively has a 0 reference.
15533 That doesn't work (cleanup never happens), so donate our
15534 reference to it onto the save stack. */
15535 PL_tmps_stack[++PL_tmps_ix] = *svp;
15537 /* As an optimisation, because we are already walking the
15538 entire array, instead of above doing either
15539 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15540 release our reference to the scalar, so that at the end of
15541 the array owns zero references to the scalars it happens to
15542 point to. We are effectively converting the array from
15543 AvREAL() on to AvREAL() off. This saves the av_clear()
15544 (triggered by the SvREFCNT_dec(unreferenced) below) from
15545 walking the array a second time. */
15546 SvREFCNT_dec(*svp);
15549 } while (++svp <= last);
15550 AvREAL_off(unreferenced);
15552 SvREFCNT_dec_NN(unreferenced);
15556 Perl_clone_params_del(CLONE_PARAMS *param)
15558 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15560 PerlInterpreter *const to = param->new_perl;
15562 PerlInterpreter *const was = PERL_GET_THX;
15564 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15570 SvREFCNT_dec(param->stashes);
15571 if (param->unreferenced)
15572 unreferenced_to_tmp_stack(param->unreferenced);
15582 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15585 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15586 does a dTHX; to get the context from thread local storage.
15587 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15588 a version that passes in my_perl. */
15589 PerlInterpreter *const was = PERL_GET_THX;
15590 CLONE_PARAMS *param;
15592 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15598 /* Given that we've set the context, we can do this unshared. */
15599 Newx(param, 1, CLONE_PARAMS);
15602 param->proto_perl = from;
15603 param->new_perl = to;
15604 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15605 AvREAL_off(param->stashes);
15606 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15614 #endif /* USE_ITHREADS */
15617 Perl_init_constants(pTHX)
15619 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15620 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15621 SvANY(&PL_sv_undef) = NULL;
15623 SvANY(&PL_sv_no) = new_XPVNV();
15624 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15625 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15626 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15629 SvANY(&PL_sv_yes) = new_XPVNV();
15630 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15631 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15632 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15635 SvPV_set(&PL_sv_no, (char*)PL_No);
15636 SvCUR_set(&PL_sv_no, 0);
15637 SvLEN_set(&PL_sv_no, 0);
15638 SvIV_set(&PL_sv_no, 0);
15639 SvNV_set(&PL_sv_no, 0);
15641 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15642 SvCUR_set(&PL_sv_yes, 1);
15643 SvLEN_set(&PL_sv_yes, 0);
15644 SvIV_set(&PL_sv_yes, 1);
15645 SvNV_set(&PL_sv_yes, 1);
15647 PadnamePV(&PL_padname_const) = (char *)PL_No;
15651 =head1 Unicode Support
15653 =for apidoc sv_recode_to_utf8
15655 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15656 of C<sv> is assumed to be octets in that encoding, and C<sv>
15657 will be converted into Unicode (and UTF-8).
15659 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15660 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15661 an C<Encode::XS> Encoding object, bad things will happen.
15662 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15664 The PV of C<sv> is returned.
15669 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15671 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15673 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15682 if (SvPADTMP(nsv)) {
15683 nsv = sv_newmortal();
15684 SvSetSV_nosteal(nsv, sv);
15693 Passing sv_yes is wrong - it needs to be or'ed set of constants
15694 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15695 remove converted chars from source.
15697 Both will default the value - let them.
15699 XPUSHs(&PL_sv_yes);
15702 call_method("decode", G_SCALAR);
15706 s = SvPV_const(uni, len);
15707 if (s != SvPVX_const(sv)) {
15708 SvGROW(sv, len + 1);
15709 Move(s, SvPVX(sv), len + 1, char);
15710 SvCUR_set(sv, len);
15715 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15716 /* clear pos and any utf8 cache */
15717 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15720 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15721 magic_setutf8(sv,mg); /* clear UTF8 cache */
15726 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15730 =for apidoc sv_cat_decode
15732 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15733 assumed to be octets in that encoding and decoding the input starts
15734 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15735 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15736 when the string C<tstr> appears in decoding output or the input ends on
15737 the PV of C<ssv>. The value which C<offset> points will be modified
15738 to the last input position on C<ssv>.
15740 Returns TRUE if the terminator was found, else returns FALSE.
15745 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15746 SV *ssv, int *offset, char *tstr, int tlen)
15750 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15752 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15763 offsv = newSViv(*offset);
15765 mPUSHp(tstr, tlen);
15767 call_method("cat_decode", G_SCALAR);
15769 ret = SvTRUE(TOPs);
15770 *offset = SvIV(offsv);
15776 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15781 /* ---------------------------------------------------------------------
15783 * support functions for report_uninit()
15786 /* the maxiumum size of array or hash where we will scan looking
15787 * for the undefined element that triggered the warning */
15789 #define FUV_MAX_SEARCH_SIZE 1000
15791 /* Look for an entry in the hash whose value has the same SV as val;
15792 * If so, return a mortal copy of the key. */
15795 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15801 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15803 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15804 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15807 array = HvARRAY(hv);
15809 for (i=HvMAX(hv); i>=0; i--) {
15811 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15812 if (HeVAL(entry) != val)
15814 if ( HeVAL(entry) == &PL_sv_undef ||
15815 HeVAL(entry) == &PL_sv_placeholder)
15819 if (HeKLEN(entry) == HEf_SVKEY)
15820 return sv_mortalcopy(HeKEY_sv(entry));
15821 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15827 /* Look for an entry in the array whose value has the same SV as val;
15828 * If so, return the index, otherwise return -1. */
15831 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15833 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15835 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15836 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15839 if (val != &PL_sv_undef) {
15840 SV ** const svp = AvARRAY(av);
15843 for (i=AvFILLp(av); i>=0; i--)
15850 /* varname(): return the name of a variable, optionally with a subscript.
15851 * If gv is non-zero, use the name of that global, along with gvtype (one
15852 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15853 * targ. Depending on the value of the subscript_type flag, return:
15856 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15857 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15858 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15859 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15862 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15863 const SV *const keyname, SSize_t aindex, int subscript_type)
15866 SV * const name = sv_newmortal();
15867 if (gv && isGV(gv)) {
15869 buffer[0] = gvtype;
15872 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15874 gv_fullname4(name, gv, buffer, 0);
15876 if ((unsigned int)SvPVX(name)[1] <= 26) {
15878 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15880 /* Swap the 1 unprintable control character for the 2 byte pretty
15881 version - ie substr($name, 1, 1) = $buffer; */
15882 sv_insert(name, 1, 1, buffer, 2);
15886 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15889 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15891 if (!cv || !CvPADLIST(cv))
15893 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15894 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15898 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15899 SV * const sv = newSV(0);
15901 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15903 *SvPVX(name) = '$';
15904 Perl_sv_catpvf(aTHX_ name, "{%s}",
15905 pv_pretty(sv, pv, len, 32, NULL, NULL,
15906 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15907 SvREFCNT_dec_NN(sv);
15909 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15910 *SvPVX(name) = '$';
15911 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
15913 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15914 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15915 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15923 =for apidoc find_uninit_var
15925 Find the name of the undefined variable (if any) that caused the operator
15926 to issue a "Use of uninitialized value" warning.
15927 If match is true, only return a name if its value matches C<uninit_sv>.
15928 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15929 warning, then following the direct child of the op may yield an
15930 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15931 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15932 the variable name if we get an exact match.
15933 C<desc_p> points to a string pointer holding the description of the op.
15934 This may be updated if needed.
15936 The name is returned as a mortal SV.
15938 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15939 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15945 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15946 bool match, const char **desc_p)
15951 const OP *o, *o2, *kid;
15953 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15955 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15956 uninit_sv == &PL_sv_placeholder)))
15959 switch (obase->op_type) {
15962 /* undef should care if its args are undef - any warnings
15963 * will be from tied/magic vars */
15971 const bool pad = ( obase->op_type == OP_PADAV
15972 || obase->op_type == OP_PADHV
15973 || obase->op_type == OP_PADRANGE
15976 const bool hash = ( obase->op_type == OP_PADHV
15977 || obase->op_type == OP_RV2HV
15978 || (obase->op_type == OP_PADRANGE
15979 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15983 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15985 if (pad) { /* @lex, %lex */
15986 sv = PAD_SVl(obase->op_targ);
15990 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15991 /* @global, %global */
15992 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15995 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15997 else if (obase == PL_op) /* @{expr}, %{expr} */
15998 return find_uninit_var(cUNOPx(obase)->op_first,
15999 uninit_sv, match, desc_p);
16000 else /* @{expr}, %{expr} as a sub-expression */
16004 /* attempt to find a match within the aggregate */
16006 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16008 subscript_type = FUV_SUBSCRIPT_HASH;
16011 index = find_array_subscript((const AV *)sv, uninit_sv);
16013 subscript_type = FUV_SUBSCRIPT_ARRAY;
16016 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16019 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16020 keysv, index, subscript_type);
16024 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16026 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16027 if (!gv || !GvSTASH(gv))
16029 if (match && (GvSV(gv) != uninit_sv))
16031 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16034 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16037 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16039 return varname(NULL, '$', obase->op_targ,
16040 NULL, 0, FUV_SUBSCRIPT_NONE);
16043 gv = cGVOPx_gv(obase);
16044 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16046 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16048 case OP_AELEMFAST_LEX:
16051 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16052 if (!av || SvRMAGICAL(av))
16054 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16055 if (!svp || *svp != uninit_sv)
16058 return varname(NULL, '$', obase->op_targ,
16059 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16062 gv = cGVOPx_gv(obase);
16067 AV *const av = GvAV(gv);
16068 if (!av || SvRMAGICAL(av))
16070 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16071 if (!svp || *svp != uninit_sv)
16074 return varname(gv, '$', 0,
16075 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16077 NOT_REACHED; /* NOTREACHED */
16080 o = cUNOPx(obase)->op_first;
16081 if (!o || o->op_type != OP_NULL ||
16082 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16084 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16089 bool negate = FALSE;
16091 if (PL_op == obase)
16092 /* $a[uninit_expr] or $h{uninit_expr} */
16093 return find_uninit_var(cBINOPx(obase)->op_last,
16094 uninit_sv, match, desc_p);
16097 o = cBINOPx(obase)->op_first;
16098 kid = cBINOPx(obase)->op_last;
16100 /* get the av or hv, and optionally the gv */
16102 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16103 sv = PAD_SV(o->op_targ);
16105 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16106 && cUNOPo->op_first->op_type == OP_GV)
16108 gv = cGVOPx_gv(cUNOPo->op_first);
16112 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16117 if (kid && kid->op_type == OP_NEGATE) {
16119 kid = cUNOPx(kid)->op_first;
16122 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16123 /* index is constant */
16126 kidsv = newSVpvs_flags("-", SVs_TEMP);
16127 sv_catsv(kidsv, cSVOPx_sv(kid));
16130 kidsv = cSVOPx_sv(kid);
16134 if (obase->op_type == OP_HELEM) {
16135 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16136 if (!he || HeVAL(he) != uninit_sv)
16140 SV * const opsv = cSVOPx_sv(kid);
16141 const IV opsviv = SvIV(opsv);
16142 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16143 negate ? - opsviv : opsviv,
16145 if (!svp || *svp != uninit_sv)
16149 if (obase->op_type == OP_HELEM)
16150 return varname(gv, '%', o->op_targ,
16151 kidsv, 0, FUV_SUBSCRIPT_HASH);
16153 return varname(gv, '@', o->op_targ, NULL,
16154 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16155 FUV_SUBSCRIPT_ARRAY);
16158 /* index is an expression;
16159 * attempt to find a match within the aggregate */
16160 if (obase->op_type == OP_HELEM) {
16161 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16163 return varname(gv, '%', o->op_targ,
16164 keysv, 0, FUV_SUBSCRIPT_HASH);
16167 const SSize_t index
16168 = find_array_subscript((const AV *)sv, uninit_sv);
16170 return varname(gv, '@', o->op_targ,
16171 NULL, index, FUV_SUBSCRIPT_ARRAY);
16176 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16178 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16180 NOT_REACHED; /* NOTREACHED */
16183 case OP_MULTIDEREF: {
16184 /* If we were executing OP_MULTIDEREF when the undef warning
16185 * triggered, then it must be one of the index values within
16186 * that triggered it. If not, then the only possibility is that
16187 * the value retrieved by the last aggregate index might be the
16188 * culprit. For the former, we set PL_multideref_pc each time before
16189 * using an index, so work though the item list until we reach
16190 * that point. For the latter, just work through the entire item
16191 * list; the last aggregate retrieved will be the candidate.
16192 * There is a third rare possibility: something triggered
16193 * magic while fetching an array/hash element. Just display
16194 * nothing in this case.
16197 /* the named aggregate, if any */
16198 PADOFFSET agg_targ = 0;
16200 /* the last-seen index */
16202 PADOFFSET index_targ;
16204 IV index_const_iv = 0; /* init for spurious compiler warn */
16205 SV *index_const_sv;
16206 int depth = 0; /* how many array/hash lookups we've done */
16208 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16209 UNOP_AUX_item *last = NULL;
16210 UV actions = items->uv;
16213 if (PL_op == obase) {
16214 last = PL_multideref_pc;
16215 assert(last >= items && last <= items + items[-1].uv);
16222 switch (actions & MDEREF_ACTION_MASK) {
16224 case MDEREF_reload:
16225 actions = (++items)->uv;
16228 case MDEREF_HV_padhv_helem: /* $lex{...} */
16231 case MDEREF_AV_padav_aelem: /* $lex[...] */
16232 agg_targ = (++items)->pad_offset;
16236 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16239 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16241 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16242 assert(isGV_with_GP(agg_gv));
16245 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16246 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16249 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16250 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16256 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16257 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16260 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16261 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16268 index_const_sv = NULL;
16270 index_type = (actions & MDEREF_INDEX_MASK);
16271 switch (index_type) {
16272 case MDEREF_INDEX_none:
16274 case MDEREF_INDEX_const:
16276 index_const_sv = UNOP_AUX_item_sv(++items)
16278 index_const_iv = (++items)->iv;
16280 case MDEREF_INDEX_padsv:
16281 index_targ = (++items)->pad_offset;
16283 case MDEREF_INDEX_gvsv:
16284 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16285 assert(isGV_with_GP(index_gv));
16289 if (index_type != MDEREF_INDEX_none)
16292 if ( index_type == MDEREF_INDEX_none
16293 || (actions & MDEREF_FLAG_last)
16294 || (last && items >= last)
16298 actions >>= MDEREF_SHIFT;
16301 if (PL_op == obase) {
16302 /* most likely index was undef */
16304 *desc_p = ( (actions & MDEREF_FLAG_last)
16305 && (obase->op_private
16306 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16308 (obase->op_private & OPpMULTIDEREF_EXISTS)
16311 : is_hv ? "hash element" : "array element";
16312 assert(index_type != MDEREF_INDEX_none);
16314 if (GvSV(index_gv) == uninit_sv)
16315 return varname(index_gv, '$', 0, NULL, 0,
16316 FUV_SUBSCRIPT_NONE);
16321 if (PL_curpad[index_targ] == uninit_sv)
16322 return varname(NULL, '$', index_targ,
16323 NULL, 0, FUV_SUBSCRIPT_NONE);
16327 /* If we got to this point it was undef on a const subscript,
16328 * so magic probably involved, e.g. $ISA[0]. Give up. */
16332 /* the SV returned by pp_multideref() was undef, if anything was */
16338 sv = PAD_SV(agg_targ);
16340 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16344 if (index_type == MDEREF_INDEX_const) {
16349 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16350 if (!he || HeVAL(he) != uninit_sv)
16354 SV * const * const svp =
16355 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16356 if (!svp || *svp != uninit_sv)
16361 ? varname(agg_gv, '%', agg_targ,
16362 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16363 : varname(agg_gv, '@', agg_targ,
16364 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16367 /* index is an var */
16369 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16371 return varname(agg_gv, '%', agg_targ,
16372 keysv, 0, FUV_SUBSCRIPT_HASH);
16375 const SSize_t index
16376 = find_array_subscript((const AV *)sv, uninit_sv);
16378 return varname(agg_gv, '@', agg_targ,
16379 NULL, index, FUV_SUBSCRIPT_ARRAY);
16383 return varname(agg_gv,
16385 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16387 NOT_REACHED; /* NOTREACHED */
16391 /* only examine RHS */
16392 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16396 o = cUNOPx(obase)->op_first;
16397 if ( o->op_type == OP_PUSHMARK
16398 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16402 if (!OpHAS_SIBLING(o)) {
16403 /* one-arg version of open is highly magical */
16405 if (o->op_type == OP_GV) { /* open FOO; */
16407 if (match && GvSV(gv) != uninit_sv)
16409 return varname(gv, '$', 0,
16410 NULL, 0, FUV_SUBSCRIPT_NONE);
16412 /* other possibilities not handled are:
16413 * open $x; or open my $x; should return '${*$x}'
16414 * open expr; should return '$'.expr ideally
16421 /* ops where $_ may be an implicit arg */
16426 if ( !(obase->op_flags & OPf_STACKED)) {
16427 if (uninit_sv == DEFSV)
16428 return newSVpvs_flags("$_", SVs_TEMP);
16429 else if (obase->op_targ
16430 && uninit_sv == PAD_SVl(obase->op_targ))
16431 return varname(NULL, '$', obase->op_targ, NULL, 0,
16432 FUV_SUBSCRIPT_NONE);
16439 match = 1; /* print etc can return undef on defined args */
16440 /* skip filehandle as it can't produce 'undef' warning */
16441 o = cUNOPx(obase)->op_first;
16442 if ((obase->op_flags & OPf_STACKED)
16444 ( o->op_type == OP_PUSHMARK
16445 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16446 o = OpSIBLING(OpSIBLING(o));
16450 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16451 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16453 /* the following ops are capable of returning PL_sv_undef even for
16454 * defined arg(s) */
16473 case OP_GETPEERNAME:
16520 case OP_SMARTMATCH:
16529 /* XXX tmp hack: these two may call an XS sub, and currently
16530 XS subs don't have a SUB entry on the context stack, so CV and
16531 pad determination goes wrong, and BAD things happen. So, just
16532 don't try to determine the value under those circumstances.
16533 Need a better fix at dome point. DAPM 11/2007 */
16539 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16540 if (gv && GvSV(gv) == uninit_sv)
16541 return newSVpvs_flags("$.", SVs_TEMP);
16546 /* def-ness of rval pos() is independent of the def-ness of its arg */
16547 if ( !(obase->op_flags & OPf_MOD))
16552 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16553 return newSVpvs_flags("${$/}", SVs_TEMP);
16558 if (!(obase->op_flags & OPf_KIDS))
16560 o = cUNOPx(obase)->op_first;
16566 /* This loop checks all the kid ops, skipping any that cannot pos-
16567 * sibly be responsible for the uninitialized value; i.e., defined
16568 * constants and ops that return nothing. If there is only one op
16569 * left that is not skipped, then we *know* it is responsible for
16570 * the uninitialized value. If there is more than one op left, we
16571 * have to look for an exact match in the while() loop below.
16572 * Note that we skip padrange, because the individual pad ops that
16573 * it replaced are still in the tree, so we work on them instead.
16576 for (kid=o; kid; kid = OpSIBLING(kid)) {
16577 const OPCODE type = kid->op_type;
16578 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16579 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16580 || (type == OP_PUSHMARK)
16581 || (type == OP_PADRANGE)
16585 if (o2) { /* more than one found */
16592 return find_uninit_var(o2, uninit_sv, match, desc_p);
16594 /* scan all args */
16596 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16608 =for apidoc report_uninit
16610 Print appropriate "Use of uninitialized variable" warning.
16616 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16618 const char *desc = NULL;
16619 SV* varname = NULL;
16622 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16625 if (uninit_sv && PL_curpad) {
16626 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16628 sv_insert(varname, 0, 0, " ", 1);
16631 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16632 /* we've reached the end of a sort block or sub,
16633 * and the uninit value is probably what that code returned */
16636 /* PL_warn_uninit_sv is constant */
16637 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16639 /* diag_listed_as: Use of uninitialized value%s */
16640 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16641 SVfARG(varname ? varname : &PL_sv_no),
16644 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16650 * ex: set ts=8 sts=4 sw=4 et: