3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%"UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referant = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referant = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referant will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referant;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1531 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533 Use the C<SvGROW> wrapper instead.
1538 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1541 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1545 PERL_ARGS_ASSERT_SV_GROW;
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1553 else if (SvOOK(sv)) { /* pv is offset? */
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1562 s = SvPVX_mutable(sv);
1565 #ifdef PERL_COPY_ON_WRITE
1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1567 * to store the COW count. So in general, allocate one more byte than
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
1570 * If the new size is a big power of two, don't bother: we assume the
1571 * caller wanted a nice 2^N sized block and will be annoyed at getting
1573 * Only increment if the allocation isn't MEM_SIZE_MAX,
1574 * otherwise it will wrap to 0.
1576 if ( (newlen < 0x1000 || (newlen & (newlen - 1)))
1577 && newlen != MEM_SIZE_MAX
1582 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1583 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586 if (newlen > SvLEN(sv)) { /* need more room? */
1587 STRLEN minlen = SvCUR(sv);
1588 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1589 if (newlen < minlen)
1591 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1593 /* Don't round up on the first allocation, as odds are pretty good that
1594 * the initial request is accurate as to what is really needed */
1596 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1597 if (rounded > newlen)
1601 if (SvLEN(sv) && s) {
1602 s = (char*)saferealloc(s, newlen);
1605 s = (char*)safemalloc(newlen);
1606 if (SvPVX_const(sv) && SvCUR(sv)) {
1607 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1611 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Do this here, do it once, do it right, and then we will never get
1613 called back into sv_grow() unless there really is some growing
1615 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1617 SvLEN_set(sv, newlen);
1624 =for apidoc sv_setiv
1626 Copies an integer into the given SV, upgrading first if necessary.
1627 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1633 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1635 PERL_ARGS_ASSERT_SV_SETIV;
1637 SV_CHECK_THINKFIRST_COW_DROP(sv);
1638 switch (SvTYPE(sv)) {
1641 sv_upgrade(sv, SVt_IV);
1644 sv_upgrade(sv, SVt_PVIV);
1648 if (!isGV_with_GP(sv))
1655 /* diag_listed_as: Can't coerce %s to %s in %s */
1656 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1661 (void)SvIOK_only(sv); /* validate number */
1667 =for apidoc sv_setiv_mg
1669 Like C<sv_setiv>, but also handles 'set' magic.
1675 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1677 PERL_ARGS_ASSERT_SV_SETIV_MG;
1684 =for apidoc sv_setuv
1686 Copies an unsigned integer into the given SV, upgrading first if necessary.
1687 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1693 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1695 PERL_ARGS_ASSERT_SV_SETUV;
1697 /* With the if statement to ensure that integers are stored as IVs whenever
1699 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1702 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1704 If you wish to remove the following if statement, so that this routine
1705 (and its callers) always return UVs, please benchmark to see what the
1706 effect is. Modern CPUs may be different. Or may not :-)
1708 if (u <= (UV)IV_MAX) {
1709 sv_setiv(sv, (IV)u);
1718 =for apidoc sv_setuv_mg
1720 Like C<sv_setuv>, but also handles 'set' magic.
1726 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1728 PERL_ARGS_ASSERT_SV_SETUV_MG;
1735 =for apidoc sv_setnv
1737 Copies a double into the given SV, upgrading first if necessary.
1738 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1744 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1746 PERL_ARGS_ASSERT_SV_SETNV;
1748 SV_CHECK_THINKFIRST_COW_DROP(sv);
1749 switch (SvTYPE(sv)) {
1752 sv_upgrade(sv, SVt_NV);
1756 sv_upgrade(sv, SVt_PVNV);
1760 if (!isGV_with_GP(sv))
1767 /* diag_listed_as: Can't coerce %s to %s in %s */
1768 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1774 (void)SvNOK_only(sv); /* validate number */
1779 =for apidoc sv_setnv_mg
1781 Like C<sv_setnv>, but also handles 'set' magic.
1787 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1789 PERL_ARGS_ASSERT_SV_SETNV_MG;
1795 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1796 * not incrementable warning display.
1797 * Originally part of S_not_a_number().
1798 * The return value may be != tmpbuf.
1802 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1805 PERL_ARGS_ASSERT_SV_DISPLAY;
1808 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1809 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1812 const char * const limit = tmpbuf + tmpbuf_size - 8;
1813 /* each *s can expand to 4 chars + "...\0",
1814 i.e. need room for 8 chars */
1816 const char *s = SvPVX_const(sv);
1817 const char * const end = s + SvCUR(sv);
1818 for ( ; s < end && d < limit; s++ ) {
1820 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1824 /* Map to ASCII "equivalent" of Latin1 */
1825 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1831 else if (ch == '\r') {
1835 else if (ch == '\f') {
1839 else if (ch == '\\') {
1843 else if (ch == '\0') {
1847 else if (isPRINT_LC(ch))
1866 /* Print an "isn't numeric" warning, using a cleaned-up,
1867 * printable version of the offending string
1871 S_not_a_number(pTHX_ SV *const sv)
1876 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1878 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1881 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1882 /* diag_listed_as: Argument "%s" isn't numeric%s */
1883 "Argument \"%s\" isn't numeric in %s", pv,
1886 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1887 /* diag_listed_as: Argument "%s" isn't numeric%s */
1888 "Argument \"%s\" isn't numeric", pv);
1892 S_not_incrementable(pTHX_ SV *const sv) {
1896 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1898 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1900 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1901 "Argument \"%s\" treated as 0 in increment (++)", pv);
1905 =for apidoc looks_like_number
1907 Test if the content of an SV looks like a number (or is a number).
1908 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1909 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1916 Perl_looks_like_number(pTHX_ SV *const sv)
1922 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1924 if (SvPOK(sv) || SvPOKp(sv)) {
1925 sbegin = SvPV_nomg_const(sv, len);
1928 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1929 numtype = grok_number(sbegin, len, NULL);
1930 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1934 S_glob_2number(pTHX_ GV * const gv)
1936 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1938 /* We know that all GVs stringify to something that is not-a-number,
1939 so no need to test that. */
1940 if (ckWARN(WARN_NUMERIC))
1942 SV *const buffer = sv_newmortal();
1943 gv_efullname3(buffer, gv, "*");
1944 not_a_number(buffer);
1946 /* We just want something true to return, so that S_sv_2iuv_common
1947 can tail call us and return true. */
1951 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1952 until proven guilty, assume that things are not that bad... */
1957 As 64 bit platforms often have an NV that doesn't preserve all bits of
1958 an IV (an assumption perl has been based on to date) it becomes necessary
1959 to remove the assumption that the NV always carries enough precision to
1960 recreate the IV whenever needed, and that the NV is the canonical form.
1961 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1962 precision as a side effect of conversion (which would lead to insanity
1963 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1964 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1965 where precision was lost, and IV/UV/NV slots that have a valid conversion
1966 which has lost no precision
1967 2) to ensure that if a numeric conversion to one form is requested that
1968 would lose precision, the precise conversion (or differently
1969 imprecise conversion) is also performed and cached, to prevent
1970 requests for different numeric formats on the same SV causing
1971 lossy conversion chains. (lossless conversion chains are perfectly
1976 SvIOKp is true if the IV slot contains a valid value
1977 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1978 SvNOKp is true if the NV slot contains a valid value
1979 SvNOK is true only if the NV value is accurate
1982 while converting from PV to NV, check to see if converting that NV to an
1983 IV(or UV) would lose accuracy over a direct conversion from PV to
1984 IV(or UV). If it would, cache both conversions, return NV, but mark
1985 SV as IOK NOKp (ie not NOK).
1987 While converting from PV to IV, check to see if converting that IV to an
1988 NV would lose accuracy over a direct conversion from PV to NV. If it
1989 would, cache both conversions, flag similarly.
1991 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1992 correctly because if IV & NV were set NV *always* overruled.
1993 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1994 changes - now IV and NV together means that the two are interchangeable:
1995 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1997 The benefit of this is that operations such as pp_add know that if
1998 SvIOK is true for both left and right operands, then integer addition
1999 can be used instead of floating point (for cases where the result won't
2000 overflow). Before, floating point was always used, which could lead to
2001 loss of precision compared with integer addition.
2003 * making IV and NV equal status should make maths accurate on 64 bit
2005 * may speed up maths somewhat if pp_add and friends start to use
2006 integers when possible instead of fp. (Hopefully the overhead in
2007 looking for SvIOK and checking for overflow will not outweigh the
2008 fp to integer speedup)
2009 * will slow down integer operations (callers of SvIV) on "inaccurate"
2010 values, as the change from SvIOK to SvIOKp will cause a call into
2011 sv_2iv each time rather than a macro access direct to the IV slot
2012 * should speed up number->string conversion on integers as IV is
2013 favoured when IV and NV are equally accurate
2015 ####################################################################
2016 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2017 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2018 On the other hand, SvUOK is true iff UV.
2019 ####################################################################
2021 Your mileage will vary depending your CPU's relative fp to integer
2025 #ifndef NV_PRESERVES_UV
2026 # define IS_NUMBER_UNDERFLOW_IV 1
2027 # define IS_NUMBER_UNDERFLOW_UV 2
2028 # define IS_NUMBER_IV_AND_UV 2
2029 # define IS_NUMBER_OVERFLOW_IV 4
2030 # define IS_NUMBER_OVERFLOW_UV 5
2032 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2034 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2036 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2042 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2043 PERL_UNUSED_CONTEXT;
2045 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%"UVxf" NV=%"NVgf" inttype=%"UVXf"\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2046 if (SvNVX(sv) < (NV)IV_MIN) {
2047 (void)SvIOKp_on(sv);
2049 SvIV_set(sv, IV_MIN);
2050 return IS_NUMBER_UNDERFLOW_IV;
2052 if (SvNVX(sv) > (NV)UV_MAX) {
2053 (void)SvIOKp_on(sv);
2056 SvUV_set(sv, UV_MAX);
2057 return IS_NUMBER_OVERFLOW_UV;
2059 (void)SvIOKp_on(sv);
2061 /* Can't use strtol etc to convert this string. (See truth table in
2063 if (SvNVX(sv) <= (UV)IV_MAX) {
2064 SvIV_set(sv, I_V(SvNVX(sv)));
2065 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2066 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2068 /* Integer is imprecise. NOK, IOKp */
2070 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2073 SvUV_set(sv, U_V(SvNVX(sv)));
2074 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2075 if (SvUVX(sv) == UV_MAX) {
2076 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2077 possibly be preserved by NV. Hence, it must be overflow.
2079 return IS_NUMBER_OVERFLOW_UV;
2081 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2083 /* Integer is imprecise. NOK, IOKp */
2085 return IS_NUMBER_OVERFLOW_IV;
2087 #endif /* !NV_PRESERVES_UV*/
2089 /* If numtype is infnan, set the NV of the sv accordingly.
2090 * If numtype is anything else, try setting the NV using Atof(PV). */
2092 # pragma warning(push)
2093 # pragma warning(disable:4756;disable:4056)
2096 S_sv_setnv(pTHX_ SV* sv, int numtype)
2098 bool pok = cBOOL(SvPOK(sv));
2100 if ((numtype & IS_NUMBER_INFINITY)) {
2101 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2104 else if ((numtype & IS_NUMBER_NAN)) {
2105 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)) {
2218 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2219 /* We want to avoid a possible problem when we cache an IV/ a UV which
2220 may be later translated to an NV, and the resulting NV is not
2221 the same as the direct translation of the initial string
2222 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2223 be careful to ensure that the value with the .456 is around if the
2224 NV value is requested in the future).
2226 This means that if we cache such an IV/a UV, we need to cache the
2227 NV as well. Moreover, we trade speed for space, and do not
2228 cache the NV if we are sure it's not needed.
2231 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2232 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2233 == IS_NUMBER_IN_UV) {
2234 /* It's definitely an integer, only upgrade to PVIV */
2235 if (SvTYPE(sv) < SVt_PVIV)
2236 sv_upgrade(sv, SVt_PVIV);
2238 } else if (SvTYPE(sv) < SVt_PVNV)
2239 sv_upgrade(sv, SVt_PVNV);
2241 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2242 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2244 S_sv_setnv(aTHX_ sv, numtype);
2248 /* If NVs preserve UVs then we only use the UV value if we know that
2249 we aren't going to call atof() below. If NVs don't preserve UVs
2250 then the value returned may have more precision than atof() will
2251 return, even though value isn't perfectly accurate. */
2252 if ((numtype & (IS_NUMBER_IN_UV
2253 #ifdef NV_PRESERVES_UV
2256 )) == IS_NUMBER_IN_UV) {
2257 /* This won't turn off the public IOK flag if it was set above */
2258 (void)SvIOKp_on(sv);
2260 if (!(numtype & IS_NUMBER_NEG)) {
2262 if (value <= (UV)IV_MAX) {
2263 SvIV_set(sv, (IV)value);
2265 /* it didn't overflow, and it was positive. */
2266 SvUV_set(sv, value);
2270 /* 2s complement assumption */
2271 if (value <= (UV)IV_MIN) {
2272 SvIV_set(sv, value == (UV)IV_MIN
2273 ? IV_MIN : -(IV)value);
2275 /* Too negative for an IV. This is a double upgrade, but
2276 I'm assuming it will be rare. */
2277 if (SvTYPE(sv) < SVt_PVNV)
2278 sv_upgrade(sv, SVt_PVNV);
2282 SvNV_set(sv, -(NV)value);
2283 SvIV_set(sv, IV_MIN);
2287 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2288 will be in the previous block to set the IV slot, and the next
2289 block to set the NV slot. So no else here. */
2291 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2292 != IS_NUMBER_IN_UV) {
2293 /* It wasn't an (integer that doesn't overflow the UV). */
2294 S_sv_setnv(aTHX_ sv, numtype);
2296 if (! numtype && ckWARN(WARN_NUMERIC))
2299 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2300 PTR2UV(sv), SvNVX(sv)));
2302 #ifdef NV_PRESERVES_UV
2303 (void)SvIOKp_on(sv);
2305 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2306 if (Perl_isnan(SvNVX(sv))) {
2312 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2313 SvIV_set(sv, I_V(SvNVX(sv)));
2314 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2317 NOOP; /* Integer is imprecise. NOK, IOKp */
2319 /* UV will not work better than IV */
2321 if (SvNVX(sv) > (NV)UV_MAX) {
2323 /* Integer is inaccurate. NOK, IOKp, is UV */
2324 SvUV_set(sv, UV_MAX);
2326 SvUV_set(sv, U_V(SvNVX(sv)));
2327 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2328 NV preservse UV so can do correct comparison. */
2329 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2332 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2337 #else /* NV_PRESERVES_UV */
2338 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2339 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2340 /* The IV/UV slot will have been set from value returned by
2341 grok_number above. The NV slot has just been set using
2344 assert (SvIOKp(sv));
2346 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2347 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2348 /* Small enough to preserve all bits. */
2349 (void)SvIOKp_on(sv);
2351 SvIV_set(sv, I_V(SvNVX(sv)));
2352 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2354 /* Assumption: first non-preserved integer is < IV_MAX,
2355 this NV is in the preserved range, therefore: */
2356 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2358 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);
2362 0 0 already failed to read UV.
2363 0 1 already failed to read UV.
2364 1 0 you won't get here in this case. IV/UV
2365 slot set, public IOK, Atof() unneeded.
2366 1 1 already read UV.
2367 so there's no point in sv_2iuv_non_preserve() attempting
2368 to use atol, strtol, strtoul etc. */
2370 sv_2iuv_non_preserve (sv, numtype);
2372 sv_2iuv_non_preserve (sv);
2376 #endif /* NV_PRESERVES_UV */
2377 /* It might be more code efficient to go through the entire logic above
2378 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2379 gets complex and potentially buggy, so more programmer efficient
2380 to do it this way, by turning off the public flags: */
2382 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2386 if (isGV_with_GP(sv))
2387 return glob_2number(MUTABLE_GV(sv));
2389 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2391 if (SvTYPE(sv) < SVt_IV)
2392 /* Typically the caller expects that sv_any is not NULL now. */
2393 sv_upgrade(sv, SVt_IV);
2394 /* Return 0 from the caller. */
2401 =for apidoc sv_2iv_flags
2403 Return the integer value of an SV, doing any necessary string
2404 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2405 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2411 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2413 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2415 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2416 && SvTYPE(sv) != SVt_PVFM);
2418 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2424 if (flags & SV_SKIP_OVERLOAD)
2426 tmpstr = AMG_CALLunary(sv, numer_amg);
2427 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2428 return SvIV(tmpstr);
2431 return PTR2IV(SvRV(sv));
2434 if (SvVALID(sv) || isREGEXP(sv)) {
2435 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2436 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2437 In practice they are extremely unlikely to actually get anywhere
2438 accessible by user Perl code - the only way that I'm aware of is when
2439 a constant subroutine which is used as the second argument to index.
2441 Regexps have no SvIVX and SvNVX fields.
2443 assert(isREGEXP(sv) || SvPOKp(sv));
2446 const char * const ptr =
2447 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2449 = grok_number(ptr, SvCUR(sv), &value);
2451 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2452 == IS_NUMBER_IN_UV) {
2453 /* It's definitely an integer */
2454 if (numtype & IS_NUMBER_NEG) {
2455 if (value < (UV)IV_MIN)
2458 if (value < (UV)IV_MAX)
2463 /* Quite wrong but no good choices. */
2464 if ((numtype & IS_NUMBER_INFINITY)) {
2465 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2466 } else if ((numtype & IS_NUMBER_NAN)) {
2467 return 0; /* So wrong. */
2471 if (ckWARN(WARN_NUMERIC))
2474 return I_V(Atof(ptr));
2478 if (SvTHINKFIRST(sv)) {
2479 if (SvREADONLY(sv) && !SvOK(sv)) {
2480 if (ckWARN(WARN_UNINITIALIZED))
2487 if (S_sv_2iuv_common(aTHX_ sv))
2491 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2492 PTR2UV(sv),SvIVX(sv)));
2493 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2497 =for apidoc sv_2uv_flags
2499 Return the unsigned integer value of an SV, doing any necessary string
2500 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2501 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2507 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2509 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2511 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2517 if (flags & SV_SKIP_OVERLOAD)
2519 tmpstr = AMG_CALLunary(sv, numer_amg);
2520 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2521 return SvUV(tmpstr);
2524 return PTR2UV(SvRV(sv));
2527 if (SvVALID(sv) || isREGEXP(sv)) {
2528 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2529 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2530 Regexps have no SvIVX and SvNVX fields. */
2531 assert(isREGEXP(sv) || SvPOKp(sv));
2534 const char * const ptr =
2535 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2537 = grok_number(ptr, SvCUR(sv), &value);
2539 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2540 == IS_NUMBER_IN_UV) {
2541 /* It's definitely an integer */
2542 if (!(numtype & IS_NUMBER_NEG))
2546 /* Quite wrong but no good choices. */
2547 if ((numtype & IS_NUMBER_INFINITY)) {
2548 return UV_MAX; /* So wrong. */
2549 } else if ((numtype & IS_NUMBER_NAN)) {
2550 return 0; /* So wrong. */
2554 if (ckWARN(WARN_NUMERIC))
2557 return U_V(Atof(ptr));
2561 if (SvTHINKFIRST(sv)) {
2562 if (SvREADONLY(sv) && !SvOK(sv)) {
2563 if (ckWARN(WARN_UNINITIALIZED))
2570 if (S_sv_2iuv_common(aTHX_ sv))
2574 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2575 PTR2UV(sv),SvUVX(sv)));
2576 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2580 =for apidoc sv_2nv_flags
2582 Return the num value of an SV, doing any necessary string or integer
2583 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2584 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2590 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2592 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2594 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2595 && SvTYPE(sv) != SVt_PVFM);
2596 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2597 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2598 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2599 Regexps have no SvIVX and SvNVX fields. */
2601 if (flags & SV_GMAGIC)
2605 if (SvPOKp(sv) && !SvIOKp(sv)) {
2606 ptr = SvPVX_const(sv);
2608 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2609 !grok_number(ptr, SvCUR(sv), NULL))
2615 return (NV)SvUVX(sv);
2617 return (NV)SvIVX(sv);
2623 ptr = RX_WRAPPED((REGEXP *)sv);
2626 assert(SvTYPE(sv) >= SVt_PVMG);
2627 /* This falls through to the report_uninit near the end of the
2629 } else if (SvTHINKFIRST(sv)) {
2634 if (flags & SV_SKIP_OVERLOAD)
2636 tmpstr = AMG_CALLunary(sv, numer_amg);
2637 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2638 return SvNV(tmpstr);
2641 return PTR2NV(SvRV(sv));
2643 if (SvREADONLY(sv) && !SvOK(sv)) {
2644 if (ckWARN(WARN_UNINITIALIZED))
2649 if (SvTYPE(sv) < SVt_NV) {
2650 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2651 sv_upgrade(sv, SVt_NV);
2653 STORE_NUMERIC_LOCAL_SET_STANDARD();
2654 PerlIO_printf(Perl_debug_log,
2655 "0x%"UVxf" num(%" NVgf ")\n",
2656 PTR2UV(sv), SvNVX(sv));
2657 RESTORE_NUMERIC_LOCAL();
2660 else if (SvTYPE(sv) < SVt_PVNV)
2661 sv_upgrade(sv, SVt_PVNV);
2666 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2667 #ifdef NV_PRESERVES_UV
2673 /* Only set the public NV OK flag if this NV preserves the IV */
2674 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2676 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2677 : (SvIVX(sv) == I_V(SvNVX(sv))))
2683 else if (SvPOKp(sv)) {
2685 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2686 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2688 #ifdef NV_PRESERVES_UV
2689 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2690 == IS_NUMBER_IN_UV) {
2691 /* It's definitely an integer */
2692 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2694 S_sv_setnv(aTHX_ sv, numtype);
2701 SvNV_set(sv, Atof(SvPVX_const(sv)));
2702 /* Only set the public NV OK flag if this NV preserves the value in
2703 the PV at least as well as an IV/UV would.
2704 Not sure how to do this 100% reliably. */
2705 /* if that shift count is out of range then Configure's test is
2706 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2708 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2709 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2710 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2711 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2712 /* Can't use strtol etc to convert this string, so don't try.
2713 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2716 /* value has been set. It may not be precise. */
2717 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2718 /* 2s complement assumption for (UV)IV_MIN */
2719 SvNOK_on(sv); /* Integer is too negative. */
2724 if (numtype & IS_NUMBER_NEG) {
2725 /* -IV_MIN is undefined, but we should never reach
2726 * this point with both IS_NUMBER_NEG and value ==
2728 assert(value != (UV)IV_MIN);
2729 SvIV_set(sv, -(IV)value);
2730 } else if (value <= (UV)IV_MAX) {
2731 SvIV_set(sv, (IV)value);
2733 SvUV_set(sv, value);
2737 if (numtype & IS_NUMBER_NOT_INT) {
2738 /* I believe that even if the original PV had decimals,
2739 they are lost beyond the limit of the FP precision.
2740 However, neither is canonical, so both only get p
2741 flags. NWC, 2000/11/25 */
2742 /* Both already have p flags, so do nothing */
2744 const NV nv = SvNVX(sv);
2745 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2746 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2747 if (SvIVX(sv) == I_V(nv)) {
2750 /* It had no "." so it must be integer. */
2754 /* between IV_MAX and NV(UV_MAX).
2755 Could be slightly > UV_MAX */
2757 if (numtype & IS_NUMBER_NOT_INT) {
2758 /* UV and NV both imprecise. */
2760 const UV nv_as_uv = U_V(nv);
2762 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2771 /* It might be more code efficient to go through the entire logic above
2772 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2773 gets complex and potentially buggy, so more programmer efficient
2774 to do it this way, by turning off the public flags: */
2776 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2777 #endif /* NV_PRESERVES_UV */
2780 if (isGV_with_GP(sv)) {
2781 glob_2number(MUTABLE_GV(sv));
2785 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2787 assert (SvTYPE(sv) >= SVt_NV);
2788 /* Typically the caller expects that sv_any is not NULL now. */
2789 /* XXX Ilya implies that this is a bug in callers that assume this
2790 and ideally should be fixed. */
2794 STORE_NUMERIC_LOCAL_SET_STANDARD();
2795 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2796 PTR2UV(sv), SvNVX(sv));
2797 RESTORE_NUMERIC_LOCAL();
2805 Return an SV with the numeric value of the source SV, doing any necessary
2806 reference or overload conversion. The caller is expected to have handled
2813 Perl_sv_2num(pTHX_ SV *const sv)
2815 PERL_ARGS_ASSERT_SV_2NUM;
2820 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2821 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2822 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2823 return sv_2num(tmpsv);
2825 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2828 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2829 * UV as a string towards the end of buf, and return pointers to start and
2832 * We assume that buf is at least TYPE_CHARS(UV) long.
2836 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2838 char *ptr = buf + TYPE_CHARS(UV);
2839 char * const ebuf = ptr;
2842 PERL_ARGS_ASSERT_UIV_2BUF;
2850 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2854 *--ptr = '0' + (char)(uv % 10);
2862 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2863 * infinity or a not-a-number, writes the appropriate strings to the
2864 * buffer, including a zero byte. On success returns the written length,
2865 * excluding the zero byte, on failure (not an infinity, not a nan)
2866 * returns zero, assert-fails on maxlen being too short.
2868 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2869 * shared string constants we point to, instead of generating a new
2870 * string for each instance. */
2872 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2874 assert(maxlen >= 4);
2875 if (Perl_isinf(nv)) {
2877 if (maxlen < 5) /* "-Inf\0" */
2887 else if (Perl_isnan(nv)) {
2891 /* XXX optionally output the payload mantissa bits as
2892 * "(unsigned)" (to match the nan("...") C99 function,
2893 * or maybe as "(0xhhh...)" would make more sense...
2894 * provide a format string so that the user can decide?
2895 * NOTE: would affect the maxlen and assert() logic.*/
2900 assert((s == buffer + 3) || (s == buffer + 4));
2902 return s - buffer - 1; /* -1: excluding the zero byte */
2906 =for apidoc sv_2pv_flags
2908 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2909 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2910 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2911 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2917 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2921 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2923 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2924 && SvTYPE(sv) != SVt_PVFM);
2925 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2930 if (flags & SV_SKIP_OVERLOAD)
2932 tmpstr = AMG_CALLunary(sv, string_amg);
2933 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2934 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2936 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2940 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2941 if (flags & SV_CONST_RETURN) {
2942 pv = (char *) SvPVX_const(tmpstr);
2944 pv = (flags & SV_MUTABLE_RETURN)
2945 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2948 *lp = SvCUR(tmpstr);
2950 pv = sv_2pv_flags(tmpstr, lp, flags);
2963 SV *const referent = SvRV(sv);
2967 retval = buffer = savepvn("NULLREF", len);
2968 } else if (SvTYPE(referent) == SVt_REGEXP &&
2969 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2970 amagic_is_enabled(string_amg))) {
2971 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2975 /* If the regex is UTF-8 we want the containing scalar to
2976 have an UTF-8 flag too */
2983 *lp = RX_WRAPLEN(re);
2985 return RX_WRAPPED(re);
2987 const char *const typestr = sv_reftype(referent, 0);
2988 const STRLEN typelen = strlen(typestr);
2989 UV addr = PTR2UV(referent);
2990 const char *stashname = NULL;
2991 STRLEN stashnamelen = 0; /* hush, gcc */
2992 const char *buffer_end;
2994 if (SvOBJECT(referent)) {
2995 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2998 stashname = HEK_KEY(name);
2999 stashnamelen = HEK_LEN(name);
3001 if (HEK_UTF8(name)) {
3007 stashname = "__ANON__";
3010 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3011 + 2 * sizeof(UV) + 2 /* )\0 */;
3013 len = typelen + 3 /* (0x */
3014 + 2 * sizeof(UV) + 2 /* )\0 */;
3017 Newx(buffer, len, char);
3018 buffer_end = retval = buffer + len;
3020 /* Working backwards */
3024 *--retval = PL_hexdigit[addr & 15];
3025 } while (addr >>= 4);
3031 memcpy(retval, typestr, typelen);
3035 retval -= stashnamelen;
3036 memcpy(retval, stashname, stashnamelen);
3038 /* retval may not necessarily have reached the start of the
3040 assert (retval >= buffer);
3042 len = buffer_end - retval - 1; /* -1 for that \0 */
3054 if (flags & SV_MUTABLE_RETURN)
3055 return SvPVX_mutable(sv);
3056 if (flags & SV_CONST_RETURN)
3057 return (char *)SvPVX_const(sv);
3062 /* I'm assuming that if both IV and NV are equally valid then
3063 converting the IV is going to be more efficient */
3064 const U32 isUIOK = SvIsUV(sv);
3065 char buf[TYPE_CHARS(UV)];
3069 if (SvTYPE(sv) < SVt_PVIV)
3070 sv_upgrade(sv, SVt_PVIV);
3071 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3073 /* inlined from sv_setpvn */
3074 s = SvGROW_mutable(sv, len + 1);
3075 Move(ptr, s, len, char);
3080 else if (SvNOK(sv)) {
3081 if (SvTYPE(sv) < SVt_PVNV)
3082 sv_upgrade(sv, SVt_PVNV);
3083 if (SvNVX(sv) == 0.0
3084 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3085 && !Perl_isnan(SvNVX(sv))
3088 s = SvGROW_mutable(sv, 2);
3093 STRLEN size = 5; /* "-Inf\0" */
3095 s = SvGROW_mutable(sv, size);
3096 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3102 /* some Xenix systems wipe out errno here */
3111 5 + /* exponent digits */
3115 s = SvGROW_mutable(sv, size);
3116 #ifndef USE_LOCALE_NUMERIC
3117 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3123 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3124 STORE_LC_NUMERIC_SET_TO_NEEDED();
3126 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3127 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3128 size += SvLEN(PL_numeric_radix_sv) - 1;
3129 s = SvGROW_mutable(sv, size);
3132 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3134 /* If the radix character is UTF-8, and actually is in the
3135 * output, turn on the UTF-8 flag for the scalar */
3137 && SvUTF8(PL_numeric_radix_sv)
3138 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3143 RESTORE_LC_NUMERIC();
3146 /* We don't call SvPOK_on(), because it may come to
3147 * pass that the locale changes so that the
3148 * stringification we just did is no longer correct. We
3149 * will have to re-stringify every time it is needed */
3156 else if (isGV_with_GP(sv)) {
3157 GV *const gv = MUTABLE_GV(sv);
3158 SV *const buffer = sv_newmortal();
3160 gv_efullname3(buffer, gv, "*");
3162 assert(SvPOK(buffer));
3166 *lp = SvCUR(buffer);
3167 return SvPVX(buffer);
3169 else if (isREGEXP(sv)) {
3170 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3171 return RX_WRAPPED((REGEXP *)sv);
3176 if (flags & SV_UNDEF_RETURNS_NULL)
3178 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3180 /* Typically the caller expects that sv_any is not NULL now. */
3181 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3182 sv_upgrade(sv, SVt_PV);
3187 const STRLEN len = s - SvPVX_const(sv);
3192 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3193 PTR2UV(sv),SvPVX_const(sv)));
3194 if (flags & SV_CONST_RETURN)
3195 return (char *)SvPVX_const(sv);
3196 if (flags & SV_MUTABLE_RETURN)
3197 return SvPVX_mutable(sv);
3202 =for apidoc sv_copypv
3204 Copies a stringified representation of the source SV into the
3205 destination SV. Automatically performs any necessary C<mg_get> and
3206 coercion of numeric values into strings. Guaranteed to preserve
3207 C<UTF8> flag even from overloaded objects. Similar in nature to
3208 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3209 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3210 would lose the UTF-8'ness of the PV.
3212 =for apidoc sv_copypv_nomg
3214 Like C<sv_copypv>, but doesn't invoke get magic first.
3216 =for apidoc sv_copypv_flags
3218 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3219 has the C<SV_GMAGIC> bit set.
3225 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3230 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3232 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3233 sv_setpvn(dsv,s,len);
3241 =for apidoc sv_2pvbyte
3243 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3244 to its length. May cause the SV to be downgraded from UTF-8 as a
3247 Usually accessed via the C<SvPVbyte> macro.
3253 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3255 PERL_ARGS_ASSERT_SV_2PVBYTE;
3258 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3259 || isGV_with_GP(sv) || SvROK(sv)) {
3260 SV *sv2 = sv_newmortal();
3261 sv_copypv_nomg(sv2,sv);
3264 sv_utf8_downgrade(sv,0);
3265 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3269 =for apidoc sv_2pvutf8
3271 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3272 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3274 Usually accessed via the C<SvPVutf8> macro.
3280 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3282 PERL_ARGS_ASSERT_SV_2PVUTF8;
3284 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3285 || isGV_with_GP(sv) || SvROK(sv))
3286 sv = sv_mortalcopy(sv);
3289 sv_utf8_upgrade_nomg(sv);
3290 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3295 =for apidoc sv_2bool
3297 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3298 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3299 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3301 =for apidoc sv_2bool_flags
3303 This function is only used by C<sv_true()> and friends, and only if
3304 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3305 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3312 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3314 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3317 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3323 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3324 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3327 if(SvGMAGICAL(sv)) {
3329 goto restart; /* call sv_2bool */
3331 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3332 else if(!SvOK(sv)) {
3335 else if(SvPOK(sv)) {
3336 svb = SvPVXtrue(sv);
3338 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3339 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3340 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3344 goto restart; /* call sv_2bool_nomg */
3349 return SvRV(sv) != 0;
3353 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3354 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3358 =for apidoc sv_utf8_upgrade
3360 Converts the PV of an SV to its UTF-8-encoded form.
3361 Forces the SV to string form if it is not already.
3362 Will C<mg_get> on C<sv> if appropriate.
3363 Always sets the C<SvUTF8> flag to avoid future validity checks even
3364 if the whole string is the same in UTF-8 as not.
3365 Returns the number of bytes in the converted string
3367 This is not a general purpose byte encoding to Unicode interface:
3368 use the Encode extension for that.
3370 =for apidoc sv_utf8_upgrade_nomg
3372 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3374 =for apidoc sv_utf8_upgrade_flags
3376 Converts the PV of an SV to its UTF-8-encoded form.
3377 Forces the SV to string form if it is not already.
3378 Always sets the SvUTF8 flag to avoid future validity checks even
3379 if all the bytes are invariant in UTF-8.
3380 If C<flags> has C<SV_GMAGIC> bit set,
3381 will C<mg_get> on C<sv> if appropriate, else not.
3383 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3384 will expand when converted to UTF-8, and skips the extra work of checking for
3385 that. Typically this flag is used by a routine that has already parsed the
3386 string and found such characters, and passes this information on so that the
3387 work doesn't have to be repeated.
3389 Returns the number of bytes in the converted string.
3391 This is not a general purpose byte encoding to Unicode interface:
3392 use the Encode extension for that.
3394 =for apidoc sv_utf8_upgrade_flags_grow
3396 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3397 the number of unused bytes the string of C<sv> is guaranteed to have free after
3398 it upon return. This allows the caller to reserve extra space that it intends
3399 to fill, to avoid extra grows.
3401 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3402 are implemented in terms of this function.
3404 Returns the number of bytes in the converted string (not including the spares).
3408 (One might think that the calling routine could pass in the position of the
3409 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3410 have to be found again. But that is not the case, because typically when the
3411 caller is likely to use this flag, it won't be calling this routine unless it
3412 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3413 and just use bytes. But some things that do fit into a byte are variants in
3414 utf8, and the caller may not have been keeping track of these.)
3416 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3417 C<NUL> isn't guaranteed due to having other routines do the work in some input
3418 cases, or if the input is already flagged as being in utf8.
3420 The speed of this could perhaps be improved for many cases if someone wanted to
3421 write a fast function that counts the number of variant characters in a string,
3422 especially if it could return the position of the first one.
3427 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3429 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3431 if (sv == &PL_sv_undef)
3433 if (!SvPOK_nog(sv)) {
3435 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3436 (void) sv_2pv_flags(sv,&len, flags);
3438 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3442 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3447 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3452 S_sv_uncow(aTHX_ sv, 0);
3455 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3456 sv_recode_to_utf8(sv, _get_encoding());
3457 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3461 if (SvCUR(sv) == 0) {
3462 if (extra) SvGROW(sv, extra);
3463 } else { /* Assume Latin-1/EBCDIC */
3464 /* This function could be much more efficient if we
3465 * had a FLAG in SVs to signal if there are any variant
3466 * chars in the PV. Given that there isn't such a flag
3467 * make the loop as fast as possible (although there are certainly ways
3468 * to speed this up, eg. through vectorization) */
3469 U8 * s = (U8 *) SvPVX_const(sv);
3470 U8 * e = (U8 *) SvEND(sv);
3472 STRLEN two_byte_count = 0;
3474 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3476 /* See if really will need to convert to utf8. We mustn't rely on our
3477 * incoming SV being well formed and having a trailing '\0', as certain
3478 * code in pp_formline can send us partially built SVs. */
3482 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3484 t--; /* t already incremented; re-point to first variant */
3489 /* utf8 conversion not needed because all are invariants. Mark as
3490 * UTF-8 even if no variant - saves scanning loop */
3492 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3497 /* Here, the string should be converted to utf8, either because of an
3498 * input flag (two_byte_count = 0), or because a character that
3499 * requires 2 bytes was found (two_byte_count = 1). t points either to
3500 * the beginning of the string (if we didn't examine anything), or to
3501 * the first variant. In either case, everything from s to t - 1 will
3502 * occupy only 1 byte each on output.
3504 * There are two main ways to convert. One is to create a new string
3505 * and go through the input starting from the beginning, appending each
3506 * converted value onto the new string as we go along. It's probably
3507 * best to allocate enough space in the string for the worst possible
3508 * case rather than possibly running out of space and having to
3509 * reallocate and then copy what we've done so far. Since everything
3510 * from s to t - 1 is invariant, the destination can be initialized
3511 * with these using a fast memory copy
3513 * The other way is to figure out exactly how big the string should be
3514 * by parsing the entire input. Then you don't have to make it big
3515 * enough to handle the worst possible case, and more importantly, if
3516 * the string you already have is large enough, you don't have to
3517 * allocate a new string, you can copy the last character in the input
3518 * string to the final position(s) that will be occupied by the
3519 * converted string and go backwards, stopping at t, since everything
3520 * before that is invariant.
3522 * There are advantages and disadvantages to each method.
3524 * In the first method, we can allocate a new string, do the memory
3525 * copy from the s to t - 1, and then proceed through the rest of the
3526 * string byte-by-byte.
3528 * In the second method, we proceed through the rest of the input
3529 * string just calculating how big the converted string will be. Then
3530 * there are two cases:
3531 * 1) if the string has enough extra space to handle the converted
3532 * value. We go backwards through the string, converting until we
3533 * get to the position we are at now, and then stop. If this
3534 * position is far enough along in the string, this method is
3535 * faster than the other method. If the memory copy were the same
3536 * speed as the byte-by-byte loop, that position would be about
3537 * half-way, as at the half-way mark, parsing to the end and back
3538 * is one complete string's parse, the same amount as starting
3539 * over and going all the way through. Actually, it would be
3540 * somewhat less than half-way, as it's faster to just count bytes
3541 * than to also copy, and we don't have the overhead of allocating
3542 * a new string, changing the scalar to use it, and freeing the
3543 * existing one. But if the memory copy is fast, the break-even
3544 * point is somewhere after half way. The counting loop could be
3545 * sped up by vectorization, etc, to move the break-even point
3546 * further towards the beginning.
3547 * 2) if the string doesn't have enough space to handle the converted
3548 * value. A new string will have to be allocated, and one might
3549 * as well, given that, start from the beginning doing the first
3550 * method. We've spent extra time parsing the string and in
3551 * exchange all we've gotten is that we know precisely how big to
3552 * make the new one. Perl is more optimized for time than space,
3553 * so this case is a loser.
3554 * So what I've decided to do is not use the 2nd method unless it is
3555 * guaranteed that a new string won't have to be allocated, assuming
3556 * the worst case. I also decided not to put any more conditions on it
3557 * than this, for now. It seems likely that, since the worst case is
3558 * twice as big as the unknown portion of the string (plus 1), we won't
3559 * be guaranteed enough space, causing us to go to the first method,
3560 * unless the string is short, or the first variant character is near
3561 * the end of it. In either of these cases, it seems best to use the
3562 * 2nd method. The only circumstance I can think of where this would
3563 * be really slower is if the string had once had much more data in it
3564 * than it does now, but there is still a substantial amount in it */
3567 STRLEN invariant_head = t - s;
3568 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3569 if (SvLEN(sv) < size) {
3571 /* Here, have decided to allocate a new string */
3576 Newx(dst, size, U8);
3578 /* If no known invariants at the beginning of the input string,
3579 * set so starts from there. Otherwise, can use memory copy to
3580 * get up to where we are now, and then start from here */
3582 if (invariant_head == 0) {
3585 Copy(s, dst, invariant_head, char);
3586 d = dst + invariant_head;
3590 append_utf8_from_native_byte(*t, &d);
3594 SvPV_free(sv); /* No longer using pre-existing string */
3595 SvPV_set(sv, (char*)dst);
3596 SvCUR_set(sv, d - dst);
3597 SvLEN_set(sv, size);
3600 /* Here, have decided to get the exact size of the string.
3601 * Currently this happens only when we know that there is
3602 * guaranteed enough space to fit the converted string, so
3603 * don't have to worry about growing. If two_byte_count is 0,
3604 * then t points to the first byte of the string which hasn't
3605 * been examined yet. Otherwise two_byte_count is 1, and t
3606 * points to the first byte in the string that will expand to
3607 * two. Depending on this, start examining at t or 1 after t.
3610 U8 *d = t + two_byte_count;
3613 /* Count up the remaining bytes that expand to two */
3616 const U8 chr = *d++;
3617 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3620 /* The string will expand by just the number of bytes that
3621 * occupy two positions. But we are one afterwards because of
3622 * the increment just above. This is the place to put the
3623 * trailing NUL, and to set the length before we decrement */
3625 d += two_byte_count;
3626 SvCUR_set(sv, d - s);
3630 /* Having decremented d, it points to the position to put the
3631 * very last byte of the expanded string. Go backwards through
3632 * the string, copying and expanding as we go, stopping when we
3633 * get to the part that is invariant the rest of the way down */
3637 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3640 *d-- = UTF8_EIGHT_BIT_LO(*e);
3641 *d-- = UTF8_EIGHT_BIT_HI(*e);
3647 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3648 /* Update pos. We do it at the end rather than during
3649 * the upgrade, to avoid slowing down the common case
3650 * (upgrade without pos).
3651 * pos can be stored as either bytes or characters. Since
3652 * this was previously a byte string we can just turn off
3653 * the bytes flag. */
3654 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3656 mg->mg_flags &= ~MGf_BYTES;
3658 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3659 magic_setutf8(sv,mg); /* clear UTF8 cache */
3664 /* Mark as UTF-8 even if no variant - saves scanning loop */
3670 =for apidoc sv_utf8_downgrade
3672 Attempts to convert the PV of an SV from characters to bytes.
3673 If the PV contains a character that cannot fit
3674 in a byte, this conversion will fail;
3675 in this case, either returns false or, if C<fail_ok> is not
3678 This is not a general purpose Unicode to byte encoding interface:
3679 use the C<Encode> extension for that.
3685 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3687 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3689 if (SvPOKp(sv) && SvUTF8(sv)) {
3693 int mg_flags = SV_GMAGIC;
3696 S_sv_uncow(aTHX_ sv, 0);
3698 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3700 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3701 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3702 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3703 SV_GMAGIC|SV_CONST_RETURN);
3704 mg_flags = 0; /* sv_pos_b2u does get magic */
3706 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3707 magic_setutf8(sv,mg); /* clear UTF8 cache */
3710 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3712 if (!utf8_to_bytes(s, &len)) {
3717 Perl_croak(aTHX_ "Wide character in %s",
3720 Perl_croak(aTHX_ "Wide character");
3731 =for apidoc sv_utf8_encode
3733 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3734 flag off so that it looks like octets again.
3740 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3742 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3744 if (SvREADONLY(sv)) {
3745 sv_force_normal_flags(sv, 0);
3747 (void) sv_utf8_upgrade(sv);
3752 =for apidoc sv_utf8_decode
3754 If the PV of the SV is an octet sequence in UTF-8
3755 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3756 so that it looks like a character. If the PV contains only single-byte
3757 characters, the C<SvUTF8> flag stays off.
3758 Scans PV for validity and returns false if the PV is invalid UTF-8.
3764 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3766 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3769 const U8 *start, *c;
3772 /* The octets may have got themselves encoded - get them back as
3775 if (!sv_utf8_downgrade(sv, TRUE))
3778 /* it is actually just a matter of turning the utf8 flag on, but
3779 * we want to make sure everything inside is valid utf8 first.
3781 c = start = (const U8 *) SvPVX_const(sv);
3782 if (!is_utf8_string(c, SvCUR(sv)))
3784 e = (const U8 *) SvEND(sv);
3787 if (!UTF8_IS_INVARIANT(ch)) {
3792 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3793 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3794 after this, clearing pos. Does anything on CPAN
3796 /* adjust pos to the start of a UTF8 char sequence */
3797 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3799 I32 pos = mg->mg_len;
3801 for (c = start + pos; c > start; c--) {
3802 if (UTF8_IS_START(*c))
3805 mg->mg_len = c - start;
3808 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3809 magic_setutf8(sv,mg); /* clear UTF8 cache */
3816 =for apidoc sv_setsv
3818 Copies the contents of the source SV C<ssv> into the destination SV
3819 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3820 function if the source SV needs to be reused. Does not handle 'set' magic on
3821 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3822 performs a copy-by-value, obliterating any previous content of the
3825 You probably want to use one of the assortment of wrappers, such as
3826 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3827 C<SvSetMagicSV_nosteal>.
3829 =for apidoc sv_setsv_flags
3831 Copies the contents of the source SV C<ssv> into the destination SV
3832 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3833 function if the source SV needs to be reused. Does not handle 'set' magic.
3834 Loosely speaking, it performs a copy-by-value, obliterating any previous
3835 content of the destination.
3836 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3837 C<ssv> if appropriate, else not. If the C<flags>
3838 parameter has the C<SV_NOSTEAL> bit set then the
3839 buffers of temps will not be stolen. C<sv_setsv>
3840 and C<sv_setsv_nomg> are implemented in terms of this function.
3842 You probably want to use one of the assortment of wrappers, such as
3843 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3844 C<SvSetMagicSV_nosteal>.
3846 This is the primary function for copying scalars, and most other
3847 copy-ish functions and macros use this underneath.
3853 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3855 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3856 HV *old_stash = NULL;
3858 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3860 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3861 const char * const name = GvNAME(sstr);
3862 const STRLEN len = GvNAMELEN(sstr);
3864 if (dtype >= SVt_PV) {
3870 SvUPGRADE(dstr, SVt_PVGV);
3871 (void)SvOK_off(dstr);
3872 isGV_with_GP_on(dstr);
3874 GvSTASH(dstr) = GvSTASH(sstr);
3876 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3877 gv_name_set(MUTABLE_GV(dstr), name, len,
3878 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3879 SvFAKE_on(dstr); /* can coerce to non-glob */
3882 if(GvGP(MUTABLE_GV(sstr))) {
3883 /* If source has method cache entry, clear it */
3885 SvREFCNT_dec(GvCV(sstr));
3886 GvCV_set(sstr, NULL);
3889 /* If source has a real method, then a method is
3892 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3898 /* If dest already had a real method, that's a change as well */
3900 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3901 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3906 /* We don't need to check the name of the destination if it was not a
3907 glob to begin with. */
3908 if(dtype == SVt_PVGV) {
3909 const char * const name = GvNAME((const GV *)dstr);
3912 /* The stash may have been detached from the symbol table, so
3914 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3918 const STRLEN len = GvNAMELEN(dstr);
3919 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3920 || (len == 1 && name[0] == ':')) {
3923 /* Set aside the old stash, so we can reset isa caches on
3925 if((old_stash = GvHV(dstr)))
3926 /* Make sure we do not lose it early. */
3927 SvREFCNT_inc_simple_void_NN(
3928 sv_2mortal((SV *)old_stash)
3933 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3936 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3937 * so temporarily protect it */
3939 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3940 gp_free(MUTABLE_GV(dstr));
3941 GvINTRO_off(dstr); /* one-shot flag */
3942 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3945 if (SvTAINTED(sstr))
3947 if (GvIMPORTED(dstr) != GVf_IMPORTED
3948 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3950 GvIMPORTED_on(dstr);
3953 if(mro_changes == 2) {
3954 if (GvAV((const GV *)sstr)) {
3956 SV * const sref = (SV *)GvAV((const GV *)dstr);
3957 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3958 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3959 AV * const ary = newAV();
3960 av_push(ary, mg->mg_obj); /* takes the refcount */
3961 mg->mg_obj = (SV *)ary;
3963 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3965 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3967 mro_isa_changed_in(GvSTASH(dstr));
3969 else if(mro_changes == 3) {
3970 HV * const stash = GvHV(dstr);
3971 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3977 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3978 if (GvIO(dstr) && dtype == SVt_PVGV) {
3979 DEBUG_o(Perl_deb(aTHX_
3980 "glob_assign_glob clearing PL_stashcache\n"));
3981 /* It's a cache. It will rebuild itself quite happily.
3982 It's a lot of effort to work out exactly which key (or keys)
3983 might be invalidated by the creation of the this file handle.
3985 hv_clear(PL_stashcache);
3991 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3993 SV * const sref = SvRV(sstr);
3995 const int intro = GvINTRO(dstr);
3998 const U32 stype = SvTYPE(sref);
4000 PERL_ARGS_ASSERT_GV_SETREF;
4003 GvINTRO_off(dstr); /* one-shot flag */
4004 GvLINE(dstr) = CopLINE(PL_curcop);
4005 GvEGV(dstr) = MUTABLE_GV(dstr);
4010 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4011 import_flag = GVf_IMPORTED_CV;
4014 location = (SV **) &GvHV(dstr);
4015 import_flag = GVf_IMPORTED_HV;
4018 location = (SV **) &GvAV(dstr);
4019 import_flag = GVf_IMPORTED_AV;
4022 location = (SV **) &GvIOp(dstr);
4025 location = (SV **) &GvFORM(dstr);
4028 location = &GvSV(dstr);
4029 import_flag = GVf_IMPORTED_SV;
4032 if (stype == SVt_PVCV) {
4033 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4034 if (GvCVGEN(dstr)) {
4035 SvREFCNT_dec(GvCV(dstr));
4036 GvCV_set(dstr, NULL);
4037 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4040 /* SAVEt_GVSLOT takes more room on the savestack and has more
4041 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4042 leave_scope needs access to the GV so it can reset method
4043 caches. We must use SAVEt_GVSLOT whenever the type is
4044 SVt_PVCV, even if the stash is anonymous, as the stash may
4045 gain a name somehow before leave_scope. */
4046 if (stype == SVt_PVCV) {
4047 /* There is no save_pushptrptrptr. Creating it for this
4048 one call site would be overkill. So inline the ss add
4052 SS_ADD_PTR(location);
4053 SS_ADD_PTR(SvREFCNT_inc(*location));
4054 SS_ADD_UV(SAVEt_GVSLOT);
4057 else SAVEGENERICSV(*location);
4060 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4061 CV* const cv = MUTABLE_CV(*location);
4063 if (!GvCVGEN((const GV *)dstr) &&
4064 (CvROOT(cv) || CvXSUB(cv)) &&
4065 /* redundant check that avoids creating the extra SV
4066 most of the time: */
4067 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4069 SV * const new_const_sv =
4070 CvCONST((const CV *)sref)
4071 ? cv_const_sv((const CV *)sref)
4073 report_redefined_cv(
4074 sv_2mortal(Perl_newSVpvf(aTHX_
4077 HvNAME_HEK(GvSTASH((const GV *)dstr))
4079 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4082 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4086 cv_ckproto_len_flags(cv, (const GV *)dstr,
4087 SvPOK(sref) ? CvPROTO(sref) : NULL,
4088 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4089 SvPOK(sref) ? SvUTF8(sref) : 0);
4091 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4092 GvASSUMECV_on(dstr);
4093 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4094 if (intro && GvREFCNT(dstr) > 1) {
4095 /* temporary remove extra savestack's ref */
4097 gv_method_changed(dstr);
4100 else gv_method_changed(dstr);
4103 *location = SvREFCNT_inc_simple_NN(sref);
4104 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4105 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4106 GvFLAGS(dstr) |= import_flag;
4109 if (stype == SVt_PVHV) {
4110 const char * const name = GvNAME((GV*)dstr);
4111 const STRLEN len = GvNAMELEN(dstr);
4114 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4115 || (len == 1 && name[0] == ':')
4117 && (!dref || HvENAME_get(dref))
4120 (HV *)sref, (HV *)dref,
4126 stype == SVt_PVAV && sref != dref
4127 && strEQ(GvNAME((GV*)dstr), "ISA")
4128 /* The stash may have been detached from the symbol table, so
4129 check its name before doing anything. */
4130 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4133 MAGIC * const omg = dref && SvSMAGICAL(dref)
4134 ? mg_find(dref, PERL_MAGIC_isa)
4136 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4137 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4138 AV * const ary = newAV();
4139 av_push(ary, mg->mg_obj); /* takes the refcount */
4140 mg->mg_obj = (SV *)ary;
4143 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4144 SV **svp = AvARRAY((AV *)omg->mg_obj);
4145 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4149 SvREFCNT_inc_simple_NN(*svp++)
4155 SvREFCNT_inc_simple_NN(omg->mg_obj)
4159 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4165 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4167 for (i = 0; i <= AvFILL(sref); ++i) {
4168 SV **elem = av_fetch ((AV*)sref, i, 0);
4171 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4175 mg = mg_find(sref, PERL_MAGIC_isa);
4177 /* Since the *ISA assignment could have affected more than
4178 one stash, don't call mro_isa_changed_in directly, but let
4179 magic_clearisa do it for us, as it already has the logic for
4180 dealing with globs vs arrays of globs. */
4182 Perl_magic_clearisa(aTHX_ NULL, mg);
4184 else if (stype == SVt_PVIO) {
4185 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4186 /* It's a cache. It will rebuild itself quite happily.
4187 It's a lot of effort to work out exactly which key (or keys)
4188 might be invalidated by the creation of the this file handle.
4190 hv_clear(PL_stashcache);
4194 if (!intro) SvREFCNT_dec(dref);
4195 if (SvTAINTED(sstr))
4203 #ifdef PERL_DEBUG_READONLY_COW
4204 # include <sys/mman.h>
4206 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4207 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4211 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4213 struct perl_memory_debug_header * const header =
4214 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4215 const MEM_SIZE len = header->size;
4216 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4217 # ifdef PERL_TRACK_MEMPOOL
4218 if (!header->readonly) header->readonly = 1;
4220 if (mprotect(header, len, PROT_READ))
4221 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4222 header, len, errno);
4226 S_sv_buf_to_rw(pTHX_ SV *sv)
4228 struct perl_memory_debug_header * const header =
4229 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4230 const MEM_SIZE len = header->size;
4231 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4232 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4233 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4234 header, len, errno);
4235 # ifdef PERL_TRACK_MEMPOOL
4236 header->readonly = 0;
4241 # define sv_buf_to_ro(sv) NOOP
4242 # define sv_buf_to_rw(sv) NOOP
4246 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4251 unsigned int both_type;
4253 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4255 if (UNLIKELY( sstr == dstr ))
4258 if (UNLIKELY( !sstr ))
4259 sstr = &PL_sv_undef;
4261 stype = SvTYPE(sstr);
4262 dtype = SvTYPE(dstr);
4263 both_type = (stype | dtype);
4265 /* with these values, we can check that both SVs are NULL/IV (and not
4266 * freed) just by testing the or'ed types */
4267 STATIC_ASSERT_STMT(SVt_NULL == 0);
4268 STATIC_ASSERT_STMT(SVt_IV == 1);
4269 if (both_type <= 1) {
4270 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4275 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4276 if (SvREADONLY(dstr))
4277 Perl_croak_no_modify();
4279 sv_unref_flags(dstr, 0);
4281 assert(!SvGMAGICAL(sstr));
4282 assert(!SvGMAGICAL(dstr));
4284 sflags = SvFLAGS(sstr);
4285 if (sflags & (SVf_IOK|SVf_ROK)) {
4286 SET_SVANY_FOR_BODYLESS_IV(dstr);
4287 new_dflags = SVt_IV;
4289 if (sflags & SVf_ROK) {
4290 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4291 new_dflags |= SVf_ROK;
4294 /* both src and dst are <= SVt_IV, so sv_any points to the
4295 * head; so access the head directly
4297 assert( &(sstr->sv_u.svu_iv)
4298 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4299 assert( &(dstr->sv_u.svu_iv)
4300 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4301 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4302 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4306 new_dflags = dtype; /* turn off everything except the type */
4308 SvFLAGS(dstr) = new_dflags;
4313 if (UNLIKELY(both_type == SVTYPEMASK)) {
4314 if (SvIS_FREED(dstr)) {
4315 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4316 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4318 if (SvIS_FREED(sstr)) {
4319 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4320 (void*)sstr, (void*)dstr);
4326 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4327 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4329 /* There's a lot of redundancy below but we're going for speed here */
4334 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4335 (void)SvOK_off(dstr);
4343 /* For performance, we inline promoting to type SVt_IV. */
4344 /* We're starting from SVt_NULL, so provided that define is
4345 * actual 0, we don't have to unset any SV type flags
4346 * to promote to SVt_IV. */
4347 STATIC_ASSERT_STMT(SVt_NULL == 0);
4348 SET_SVANY_FOR_BODYLESS_IV(dstr);
4349 SvFLAGS(dstr) |= SVt_IV;
4353 sv_upgrade(dstr, SVt_PVIV);
4357 goto end_of_first_switch;
4359 (void)SvIOK_only(dstr);
4360 SvIV_set(dstr, SvIVX(sstr));
4363 /* SvTAINTED can only be true if the SV has taint magic, which in
4364 turn means that the SV type is PVMG (or greater). This is the
4365 case statement for SVt_IV, so this cannot be true (whatever gcov
4367 assert(!SvTAINTED(sstr));
4372 if (dtype < SVt_PV && dtype != SVt_IV)
4373 sv_upgrade(dstr, SVt_IV);
4377 if (LIKELY( SvNOK(sstr) )) {
4381 sv_upgrade(dstr, SVt_NV);
4385 sv_upgrade(dstr, SVt_PVNV);
4389 goto end_of_first_switch;
4391 SvNV_set(dstr, SvNVX(sstr));
4392 (void)SvNOK_only(dstr);
4393 /* SvTAINTED can only be true if the SV has taint magic, which in
4394 turn means that the SV type is PVMG (or greater). This is the
4395 case statement for SVt_NV, so this cannot be true (whatever gcov
4397 assert(!SvTAINTED(sstr));
4404 sv_upgrade(dstr, SVt_PV);
4407 if (dtype < SVt_PVIV)
4408 sv_upgrade(dstr, SVt_PVIV);
4411 if (dtype < SVt_PVNV)
4412 sv_upgrade(dstr, SVt_PVNV);
4416 const char * const type = sv_reftype(sstr,0);
4418 /* diag_listed_as: Bizarre copy of %s */
4419 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4421 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4423 NOT_REACHED; /* NOTREACHED */
4427 if (dtype < SVt_REGEXP)
4429 if (dtype >= SVt_PV) {
4435 sv_upgrade(dstr, SVt_REGEXP);
4443 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4445 if (SvTYPE(sstr) != stype)
4446 stype = SvTYPE(sstr);
4448 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4449 glob_assign_glob(dstr, sstr, dtype);
4452 if (stype == SVt_PVLV)
4454 if (isREGEXP(sstr)) goto upgregexp;
4455 SvUPGRADE(dstr, SVt_PVNV);
4458 SvUPGRADE(dstr, (svtype)stype);
4460 end_of_first_switch:
4462 /* dstr may have been upgraded. */
4463 dtype = SvTYPE(dstr);
4464 sflags = SvFLAGS(sstr);
4466 if (UNLIKELY( dtype == SVt_PVCV )) {
4467 /* Assigning to a subroutine sets the prototype. */
4470 const char *const ptr = SvPV_const(sstr, len);
4472 SvGROW(dstr, len + 1);
4473 Copy(ptr, SvPVX(dstr), len + 1, char);
4474 SvCUR_set(dstr, len);
4476 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4477 CvAUTOLOAD_off(dstr);
4482 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4483 || dtype == SVt_PVFM))
4485 const char * const type = sv_reftype(dstr,0);
4487 /* diag_listed_as: Cannot copy to %s */
4488 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4490 Perl_croak(aTHX_ "Cannot copy to %s", type);
4491 } else if (sflags & SVf_ROK) {
4492 if (isGV_with_GP(dstr)
4493 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4496 if (GvIMPORTED(dstr) != GVf_IMPORTED
4497 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4499 GvIMPORTED_on(dstr);
4504 glob_assign_glob(dstr, sstr, dtype);
4508 if (dtype >= SVt_PV) {
4509 if (isGV_with_GP(dstr)) {
4510 gv_setref(dstr, sstr);
4513 if (SvPVX_const(dstr)) {
4519 (void)SvOK_off(dstr);
4520 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4521 SvFLAGS(dstr) |= sflags & SVf_ROK;
4522 assert(!(sflags & SVp_NOK));
4523 assert(!(sflags & SVp_IOK));
4524 assert(!(sflags & SVf_NOK));
4525 assert(!(sflags & SVf_IOK));
4527 else if (isGV_with_GP(dstr)) {
4528 if (!(sflags & SVf_OK)) {
4529 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4530 "Undefined value assigned to typeglob");
4533 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4534 if (dstr != (const SV *)gv) {
4535 const char * const name = GvNAME((const GV *)dstr);
4536 const STRLEN len = GvNAMELEN(dstr);
4537 HV *old_stash = NULL;
4538 bool reset_isa = FALSE;
4539 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4540 || (len == 1 && name[0] == ':')) {
4541 /* Set aside the old stash, so we can reset isa caches
4542 on its subclasses. */
4543 if((old_stash = GvHV(dstr))) {
4544 /* Make sure we do not lose it early. */
4545 SvREFCNT_inc_simple_void_NN(
4546 sv_2mortal((SV *)old_stash)
4553 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4554 gp_free(MUTABLE_GV(dstr));
4556 GvGP_set(dstr, gp_ref(GvGP(gv)));
4559 HV * const stash = GvHV(dstr);
4561 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4571 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4572 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4573 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4575 else if (sflags & SVp_POK) {
4576 const STRLEN cur = SvCUR(sstr);
4577 const STRLEN len = SvLEN(sstr);
4580 * We have three basic ways to copy the string:
4586 * Which we choose is based on various factors. The following
4587 * things are listed in order of speed, fastest to slowest:
4589 * - Copying a short string
4590 * - Copy-on-write bookkeeping
4592 * - Copying a long string
4594 * We swipe the string (steal the string buffer) if the SV on the
4595 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4596 * big win on long strings. It should be a win on short strings if
4597 * SvPVX_const(dstr) has to be allocated. If not, it should not
4598 * slow things down, as SvPVX_const(sstr) would have been freed
4601 * We also steal the buffer from a PADTMP (operator target) if it
4602 * is ‘long enough’. For short strings, a swipe does not help
4603 * here, as it causes more malloc calls the next time the target
4604 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4605 * be allocated it is still not worth swiping PADTMPs for short
4606 * strings, as the savings here are small.
4608 * If swiping is not an option, then we see whether it is
4609 * worth using copy-on-write. If the lhs already has a buf-
4610 * fer big enough and the string is short, we skip it and fall back
4611 * to method 3, since memcpy is faster for short strings than the
4612 * later bookkeeping overhead that copy-on-write entails.
4614 * If the rhs is not a copy-on-write string yet, then we also
4615 * consider whether the buffer is too large relative to the string
4616 * it holds. Some operations such as readline allocate a large
4617 * buffer in the expectation of reusing it. But turning such into
4618 * a COW buffer is counter-productive because it increases memory
4619 * usage by making readline allocate a new large buffer the sec-
4620 * ond time round. So, if the buffer is too large, again, we use
4623 * Finally, if there is no buffer on the left, or the buffer is too
4624 * small, then we use copy-on-write and make both SVs share the
4629 /* Whichever path we take through the next code, we want this true,
4630 and doing it now facilitates the COW check. */
4631 (void)SvPOK_only(dstr);
4635 /* slated for free anyway (and not COW)? */
4636 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4637 /* or a swipable TARG */
4639 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4641 /* whose buffer is worth stealing */
4642 && CHECK_COWBUF_THRESHOLD(cur,len)
4645 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4646 (!(flags & SV_NOSTEAL)) &&
4647 /* and we're allowed to steal temps */
4648 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4649 len) /* and really is a string */
4650 { /* Passes the swipe test. */
4651 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4653 SvPV_set(dstr, SvPVX_mutable(sstr));
4654 SvLEN_set(dstr, SvLEN(sstr));
4655 SvCUR_set(dstr, SvCUR(sstr));
4658 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4659 SvPV_set(sstr, NULL);
4664 else if (flags & SV_COW_SHARED_HASH_KEYS
4666 #ifdef PERL_COPY_ON_WRITE
4669 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4670 /* If this is a regular (non-hek) COW, only so
4671 many COW "copies" are possible. */
4672 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4673 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4674 && !(SvFLAGS(dstr) & SVf_BREAK)
4675 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4676 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4680 && !(SvFLAGS(dstr) & SVf_BREAK)
4683 /* Either it's a shared hash key, or it's suitable for
4686 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4691 if (!(sflags & SVf_IsCOW)) {
4693 CowREFCNT(sstr) = 0;
4696 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4702 if (sflags & SVf_IsCOW) {
4706 SvPV_set(dstr, SvPVX_mutable(sstr));
4711 /* SvIsCOW_shared_hash */
4712 DEBUG_C(PerlIO_printf(Perl_debug_log,
4713 "Copy on write: Sharing hash\n"));
4715 assert (SvTYPE(dstr) >= SVt_PV);
4717 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4719 SvLEN_set(dstr, len);
4720 SvCUR_set(dstr, cur);
4723 /* Failed the swipe test, and we cannot do copy-on-write either.
4724 Have to copy the string. */
4725 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4726 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4727 SvCUR_set(dstr, cur);
4728 *SvEND(dstr) = '\0';
4730 if (sflags & SVp_NOK) {
4731 SvNV_set(dstr, SvNVX(sstr));
4733 if (sflags & SVp_IOK) {
4734 SvIV_set(dstr, SvIVX(sstr));
4735 /* Must do this otherwise some other overloaded use of 0x80000000
4736 gets confused. I guess SVpbm_VALID */
4737 if (sflags & SVf_IVisUV)
4740 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4742 const MAGIC * const smg = SvVSTRING_mg(sstr);
4744 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4745 smg->mg_ptr, smg->mg_len);
4746 SvRMAGICAL_on(dstr);
4750 else if (sflags & (SVp_IOK|SVp_NOK)) {
4751 (void)SvOK_off(dstr);
4752 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4753 if (sflags & SVp_IOK) {
4754 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4755 SvIV_set(dstr, SvIVX(sstr));
4757 if (sflags & SVp_NOK) {
4758 SvNV_set(dstr, SvNVX(sstr));
4762 if (isGV_with_GP(sstr)) {
4763 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4766 (void)SvOK_off(dstr);
4768 if (SvTAINTED(sstr))
4773 =for apidoc sv_setsv_mg
4775 Like C<sv_setsv>, but also handles 'set' magic.
4781 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4783 PERL_ARGS_ASSERT_SV_SETSV_MG;
4785 sv_setsv(dstr,sstr);
4790 # define SVt_COW SVt_PV
4792 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4794 STRLEN cur = SvCUR(sstr);
4795 STRLEN len = SvLEN(sstr);
4797 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4798 const bool already = cBOOL(SvIsCOW(sstr));
4801 PERL_ARGS_ASSERT_SV_SETSV_COW;
4804 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4805 (void*)sstr, (void*)dstr);
4812 if (SvTHINKFIRST(dstr))
4813 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4814 else if (SvPVX_const(dstr))
4815 Safefree(SvPVX_mutable(dstr));
4819 SvUPGRADE(dstr, SVt_COW);
4821 assert (SvPOK(sstr));
4822 assert (SvPOKp(sstr));
4824 if (SvIsCOW(sstr)) {
4826 if (SvLEN(sstr) == 0) {
4827 /* source is a COW shared hash key. */
4828 DEBUG_C(PerlIO_printf(Perl_debug_log,
4829 "Fast copy on write: Sharing hash\n"));
4830 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4833 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4834 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4836 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4837 SvUPGRADE(sstr, SVt_COW);
4839 DEBUG_C(PerlIO_printf(Perl_debug_log,
4840 "Fast copy on write: Converting sstr to COW\n"));
4841 CowREFCNT(sstr) = 0;
4843 # ifdef PERL_DEBUG_READONLY_COW
4844 if (already) sv_buf_to_rw(sstr);
4847 new_pv = SvPVX_mutable(sstr);
4851 SvPV_set(dstr, new_pv);
4852 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4855 SvLEN_set(dstr, len);
4856 SvCUR_set(dstr, cur);
4865 =for apidoc sv_setpvn
4867 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4868 The C<len> parameter indicates the number of
4869 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4870 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4876 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4880 PERL_ARGS_ASSERT_SV_SETPVN;
4882 SV_CHECK_THINKFIRST_COW_DROP(sv);
4888 /* len is STRLEN which is unsigned, need to copy to signed */
4891 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4894 SvUPGRADE(sv, SVt_PV);
4896 dptr = SvGROW(sv, len + 1);
4897 Move(ptr,dptr,len,char);
4900 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4902 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4906 =for apidoc sv_setpvn_mg
4908 Like C<sv_setpvn>, but also handles 'set' magic.
4914 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4916 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4918 sv_setpvn(sv,ptr,len);
4923 =for apidoc sv_setpv
4925 Copies a string into an SV. The string must be terminated with a C<NUL>
4927 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4933 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4937 PERL_ARGS_ASSERT_SV_SETPV;
4939 SV_CHECK_THINKFIRST_COW_DROP(sv);
4945 SvUPGRADE(sv, SVt_PV);
4947 SvGROW(sv, len + 1);
4948 Move(ptr,SvPVX(sv),len+1,char);
4950 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4952 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4956 =for apidoc sv_setpv_mg
4958 Like C<sv_setpv>, but also handles 'set' magic.
4964 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4966 PERL_ARGS_ASSERT_SV_SETPV_MG;
4973 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4975 PERL_ARGS_ASSERT_SV_SETHEK;
4981 if (HEK_LEN(hek) == HEf_SVKEY) {
4982 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4985 const int flags = HEK_FLAGS(hek);
4986 if (flags & HVhek_WASUTF8) {
4987 STRLEN utf8_len = HEK_LEN(hek);
4988 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4989 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4992 } else if (flags & HVhek_UNSHARED) {
4993 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4996 else SvUTF8_off(sv);
5000 SV_CHECK_THINKFIRST_COW_DROP(sv);
5001 SvUPGRADE(sv, SVt_PV);
5003 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5004 SvCUR_set(sv, HEK_LEN(hek));
5010 else SvUTF8_off(sv);
5018 =for apidoc sv_usepvn_flags
5020 Tells an SV to use C<ptr> to find its string value. Normally the
5021 string is stored inside the SV, but sv_usepvn allows the SV to use an
5022 outside string. C<ptr> should point to memory that was allocated
5023 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5024 the start of a C<Newx>-ed block of memory, and not a pointer to the
5025 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5026 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5027 string length, C<len>, must be supplied. By default this function
5028 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5029 so that pointer should not be freed or used by the programmer after
5030 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5031 that pointer (e.g. ptr + 1) be used.
5033 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5034 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5036 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5037 C<len>, and already meets the requirements for storing in C<SvPVX>).
5043 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5047 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5049 SV_CHECK_THINKFIRST_COW_DROP(sv);
5050 SvUPGRADE(sv, SVt_PV);
5053 if (flags & SV_SMAGIC)
5057 if (SvPVX_const(sv))
5061 if (flags & SV_HAS_TRAILING_NUL)
5062 assert(ptr[len] == '\0');
5065 allocate = (flags & SV_HAS_TRAILING_NUL)
5067 #ifdef Perl_safesysmalloc_size
5070 PERL_STRLEN_ROUNDUP(len + 1);
5072 if (flags & SV_HAS_TRAILING_NUL) {
5073 /* It's long enough - do nothing.
5074 Specifically Perl_newCONSTSUB is relying on this. */
5077 /* Force a move to shake out bugs in callers. */
5078 char *new_ptr = (char*)safemalloc(allocate);
5079 Copy(ptr, new_ptr, len, char);
5080 PoisonFree(ptr,len,char);
5084 ptr = (char*) saferealloc (ptr, allocate);
5087 #ifdef Perl_safesysmalloc_size
5088 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5090 SvLEN_set(sv, allocate);
5094 if (!(flags & SV_HAS_TRAILING_NUL)) {
5097 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5099 if (flags & SV_SMAGIC)
5104 =for apidoc sv_force_normal_flags
5106 Undo various types of fakery on an SV, where fakery means
5107 "more than" a string: if the PV is a shared string, make
5108 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5109 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5110 we do the copy, and is also used locally; if this is a
5111 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5112 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5113 C<SvPOK_off> rather than making a copy. (Used where this
5114 scalar is about to be set to some other value.) In addition,
5115 the C<flags> parameter gets passed to C<sv_unref_flags()>
5116 when unreffing. C<sv_force_normal> calls this function
5117 with flags set to 0.
5119 This function is expected to be used to signal to perl that this SV is
5120 about to be written to, and any extra book-keeping needs to be taken care
5121 of. Hence, it croaks on read-only values.
5127 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5129 assert(SvIsCOW(sv));
5132 const char * const pvx = SvPVX_const(sv);
5133 const STRLEN len = SvLEN(sv);
5134 const STRLEN cur = SvCUR(sv);
5137 PerlIO_printf(Perl_debug_log,
5138 "Copy on write: Force normal %ld\n",
5143 # ifdef PERL_COPY_ON_WRITE
5145 /* Must do this first, since the CowREFCNT uses SvPVX and
5146 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5147 the only owner left of the buffer. */
5148 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5150 U8 cowrefcnt = CowREFCNT(sv);
5151 if(cowrefcnt != 0) {
5153 CowREFCNT(sv) = cowrefcnt;
5158 /* Else we are the only owner of the buffer. */
5163 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5168 if (flags & SV_COW_DROP_PV) {
5169 /* OK, so we don't need to copy our buffer. */
5172 SvGROW(sv, cur + 1);
5173 Move(pvx,SvPVX(sv),cur,char);
5179 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5186 const char * const pvx = SvPVX_const(sv);
5187 const STRLEN len = SvCUR(sv);
5191 if (flags & SV_COW_DROP_PV) {
5192 /* OK, so we don't need to copy our buffer. */
5195 SvGROW(sv, len + 1);
5196 Move(pvx,SvPVX(sv),len,char);
5199 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5205 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5207 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5210 Perl_croak_no_modify();
5211 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5212 S_sv_uncow(aTHX_ sv, flags);
5214 sv_unref_flags(sv, flags);
5215 else if (SvFAKE(sv) && isGV_with_GP(sv))
5216 sv_unglob(sv, flags);
5217 else if (SvFAKE(sv) && isREGEXP(sv)) {
5218 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5219 to sv_unglob. We only need it here, so inline it. */
5220 const bool islv = SvTYPE(sv) == SVt_PVLV;
5221 const svtype new_type =
5222 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5223 SV *const temp = newSV_type(new_type);
5224 regexp *const temp_p = ReANY((REGEXP *)sv);
5226 if (new_type == SVt_PVMG) {
5227 SvMAGIC_set(temp, SvMAGIC(sv));
5228 SvMAGIC_set(sv, NULL);
5229 SvSTASH_set(temp, SvSTASH(sv));
5230 SvSTASH_set(sv, NULL);
5232 if (!islv) SvCUR_set(temp, SvCUR(sv));
5233 /* Remember that SvPVX is in the head, not the body. But
5234 RX_WRAPPED is in the body. */
5235 assert(ReANY((REGEXP *)sv)->mother_re);
5236 /* Their buffer is already owned by someone else. */
5237 if (flags & SV_COW_DROP_PV) {
5238 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5239 zeroed body. For SVt_PVLV, it should have been set to 0
5240 before turning into a regexp. */
5241 assert(!SvLEN(islv ? sv : temp));
5242 sv->sv_u.svu_pv = 0;
5245 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5246 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5250 /* Now swap the rest of the bodies. */
5254 SvFLAGS(sv) &= ~SVTYPEMASK;
5255 SvFLAGS(sv) |= new_type;
5256 SvANY(sv) = SvANY(temp);
5259 SvFLAGS(temp) &= ~(SVTYPEMASK);
5260 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5261 SvANY(temp) = temp_p;
5262 temp->sv_u.svu_rx = (regexp *)temp_p;
5264 SvREFCNT_dec_NN(temp);
5266 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5272 Efficient removal of characters from the beginning of the string buffer.
5273 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5274 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5275 character of the adjusted string. Uses the C<OOK> hack. On return, only
5276 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5278 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5279 refer to the same chunk of data.
5281 The unfortunate similarity of this function's name to that of Perl's C<chop>
5282 operator is strictly coincidental. This function works from the left;
5283 C<chop> works from the right.
5289 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5300 PERL_ARGS_ASSERT_SV_CHOP;
5302 if (!ptr || !SvPOKp(sv))
5304 delta = ptr - SvPVX_const(sv);
5306 /* Nothing to do. */
5309 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5310 if (delta > max_delta)
5311 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5312 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5313 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5314 SV_CHECK_THINKFIRST(sv);
5315 SvPOK_only_UTF8(sv);
5318 if (!SvLEN(sv)) { /* make copy of shared string */
5319 const char *pvx = SvPVX_const(sv);
5320 const STRLEN len = SvCUR(sv);
5321 SvGROW(sv, len + 1);
5322 Move(pvx,SvPVX(sv),len,char);
5328 SvOOK_offset(sv, old_delta);
5330 SvLEN_set(sv, SvLEN(sv) - delta);
5331 SvCUR_set(sv, SvCUR(sv) - delta);
5332 SvPV_set(sv, SvPVX(sv) + delta);
5334 p = (U8 *)SvPVX_const(sv);
5337 /* how many bytes were evacuated? we will fill them with sentinel
5338 bytes, except for the part holding the new offset of course. */
5341 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5343 assert(evacn <= delta + old_delta);
5347 /* This sets 'delta' to the accumulated value of all deltas so far */
5351 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5352 * the string; otherwise store a 0 byte there and store 'delta' just prior
5353 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5354 * portion of the chopped part of the string */
5355 if (delta < 0x100) {
5359 p -= sizeof(STRLEN);
5360 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5364 /* Fill the preceding buffer with sentinals to verify that no-one is
5374 =for apidoc sv_catpvn
5376 Concatenates the string onto the end of the string which is in the SV.
5377 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5378 status set, then the bytes appended should be valid UTF-8.
5379 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5381 =for apidoc sv_catpvn_flags
5383 Concatenates the string onto the end of the string which is in the SV. The
5384 C<len> indicates number of bytes to copy.
5386 By default, the string appended is assumed to be valid UTF-8 if the SV has
5387 the UTF-8 status set, and a string of bytes otherwise. One can force the
5388 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5389 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5390 string appended will be upgraded to UTF-8 if necessary.
5392 If C<flags> has the C<SV_SMAGIC> bit set, will
5393 C<mg_set> on C<dsv> afterwards if appropriate.
5394 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5395 in terms of this function.
5401 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5404 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5406 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5407 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5409 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5410 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5411 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5414 else SvGROW(dsv, dlen + slen + 1);
5416 sstr = SvPVX_const(dsv);
5417 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5418 SvCUR_set(dsv, SvCUR(dsv) + slen);
5421 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5422 const char * const send = sstr + slen;
5425 /* Something this code does not account for, which I think is
5426 impossible; it would require the same pv to be treated as
5427 bytes *and* utf8, which would indicate a bug elsewhere. */
5428 assert(sstr != dstr);
5430 SvGROW(dsv, dlen + slen * 2 + 1);
5431 d = (U8 *)SvPVX(dsv) + dlen;
5433 while (sstr < send) {
5434 append_utf8_from_native_byte(*sstr, &d);
5437 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5440 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5442 if (flags & SV_SMAGIC)
5447 =for apidoc sv_catsv
5449 Concatenates the string from SV C<ssv> onto the end of the string in SV
5450 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5451 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5452 and C<L</sv_catsv_nomg>>.
5454 =for apidoc sv_catsv_flags
5456 Concatenates the string from SV C<ssv> onto the end of the string in SV
5457 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5458 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5459 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5460 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5461 and C<sv_catsv_mg> are implemented in terms of this function.
5466 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5468 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5472 const char *spv = SvPV_flags_const(ssv, slen, flags);
5473 if (flags & SV_GMAGIC)
5475 sv_catpvn_flags(dsv, spv, slen,
5476 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5477 if (flags & SV_SMAGIC)
5483 =for apidoc sv_catpv
5485 Concatenates the C<NUL>-terminated string onto the end of the string which is
5487 If the SV has the UTF-8 status set, then the bytes appended should be
5488 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5494 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5500 PERL_ARGS_ASSERT_SV_CATPV;
5504 junk = SvPV_force(sv, tlen);
5506 SvGROW(sv, tlen + len + 1);
5508 ptr = SvPVX_const(sv);
5509 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5510 SvCUR_set(sv, SvCUR(sv) + len);
5511 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5516 =for apidoc sv_catpv_flags
5518 Concatenates the C<NUL>-terminated string onto the end of the string which is
5520 If the SV has the UTF-8 status set, then the bytes appended should
5521 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5522 on the modified SV if appropriate.
5528 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5530 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5531 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5535 =for apidoc sv_catpv_mg
5537 Like C<sv_catpv>, but also handles 'set' magic.
5543 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5545 PERL_ARGS_ASSERT_SV_CATPV_MG;
5554 Creates a new SV. A non-zero C<len> parameter indicates the number of
5555 bytes of preallocated string space the SV should have. An extra byte for a
5556 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5557 space is allocated.) The reference count for the new SV is set to 1.
5559 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5560 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5561 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5562 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5563 modules supporting older perls.
5569 Perl_newSV(pTHX_ const STRLEN len)
5575 sv_grow(sv, len + 1);
5580 =for apidoc sv_magicext
5582 Adds magic to an SV, upgrading it if necessary. Applies the
5583 supplied C<vtable> and returns a pointer to the magic added.
5585 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5586 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5587 one instance of the same C<how>.
5589 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5590 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5591 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5592 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5594 (This is now used as a subroutine by C<sv_magic>.)
5599 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5600 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5604 PERL_ARGS_ASSERT_SV_MAGICEXT;
5606 SvUPGRADE(sv, SVt_PVMG);
5607 Newxz(mg, 1, MAGIC);
5608 mg->mg_moremagic = SvMAGIC(sv);
5609 SvMAGIC_set(sv, mg);
5611 /* Sometimes a magic contains a reference loop, where the sv and
5612 object refer to each other. To prevent a reference loop that
5613 would prevent such objects being freed, we look for such loops
5614 and if we find one we avoid incrementing the object refcount.
5616 Note we cannot do this to avoid self-tie loops as intervening RV must
5617 have its REFCNT incremented to keep it in existence.
5620 if (!obj || obj == sv ||
5621 how == PERL_MAGIC_arylen ||
5622 how == PERL_MAGIC_symtab ||
5623 (SvTYPE(obj) == SVt_PVGV &&
5624 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5625 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5626 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5631 mg->mg_obj = SvREFCNT_inc_simple(obj);
5632 mg->mg_flags |= MGf_REFCOUNTED;
5635 /* Normal self-ties simply pass a null object, and instead of
5636 using mg_obj directly, use the SvTIED_obj macro to produce a
5637 new RV as needed. For glob "self-ties", we are tieing the PVIO
5638 with an RV obj pointing to the glob containing the PVIO. In
5639 this case, to avoid a reference loop, we need to weaken the
5643 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5644 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5650 mg->mg_len = namlen;
5653 mg->mg_ptr = savepvn(name, namlen);
5654 else if (namlen == HEf_SVKEY) {
5655 /* Yes, this is casting away const. This is only for the case of
5656 HEf_SVKEY. I think we need to document this aberation of the
5657 constness of the API, rather than making name non-const, as
5658 that change propagating outwards a long way. */
5659 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5661 mg->mg_ptr = (char *) name;
5663 mg->mg_virtual = (MGVTBL *) vtable;
5670 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5672 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5673 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5674 /* This sv is only a delegate. //g magic must be attached to
5679 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5680 &PL_vtbl_mglob, 0, 0);
5684 =for apidoc sv_magic
5686 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5687 necessary, then adds a new magic item of type C<how> to the head of the
5690 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5691 handling of the C<name> and C<namlen> arguments.
5693 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5694 to add more than one instance of the same C<how>.
5700 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5701 const char *const name, const I32 namlen)
5703 const MGVTBL *vtable;
5706 unsigned int vtable_index;
5708 PERL_ARGS_ASSERT_SV_MAGIC;
5710 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5711 || ((flags = PL_magic_data[how]),
5712 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5713 > magic_vtable_max))
5714 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5716 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5717 Useful for attaching extension internal data to perl vars.
5718 Note that multiple extensions may clash if magical scalars
5719 etc holding private data from one are passed to another. */
5721 vtable = (vtable_index == magic_vtable_max)
5722 ? NULL : PL_magic_vtables + vtable_index;
5724 if (SvREADONLY(sv)) {
5726 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5729 Perl_croak_no_modify();
5732 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5733 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5734 /* sv_magic() refuses to add a magic of the same 'how' as an
5737 if (how == PERL_MAGIC_taint)
5743 /* Force pos to be stored as characters, not bytes. */
5744 if (SvMAGICAL(sv) && DO_UTF8(sv)
5745 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5747 && mg->mg_flags & MGf_BYTES) {
5748 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5750 mg->mg_flags &= ~MGf_BYTES;
5753 /* Rest of work is done else where */
5754 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5757 case PERL_MAGIC_taint:
5760 case PERL_MAGIC_ext:
5761 case PERL_MAGIC_dbfile:
5768 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5775 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5777 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5778 for (mg = *mgp; mg; mg = *mgp) {
5779 const MGVTBL* const virt = mg->mg_virtual;
5780 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5781 *mgp = mg->mg_moremagic;
5782 if (virt && virt->svt_free)
5783 virt->svt_free(aTHX_ sv, mg);
5784 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5786 Safefree(mg->mg_ptr);
5787 else if (mg->mg_len == HEf_SVKEY)
5788 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5789 else if (mg->mg_type == PERL_MAGIC_utf8)
5790 Safefree(mg->mg_ptr);
5792 if (mg->mg_flags & MGf_REFCOUNTED)
5793 SvREFCNT_dec(mg->mg_obj);
5797 mgp = &mg->mg_moremagic;
5800 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5801 mg_magical(sv); /* else fix the flags now */
5810 =for apidoc sv_unmagic
5812 Removes all magic of type C<type> from an SV.
5818 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5820 PERL_ARGS_ASSERT_SV_UNMAGIC;
5821 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5825 =for apidoc sv_unmagicext
5827 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5833 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5835 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5836 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5840 =for apidoc sv_rvweaken
5842 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5843 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5844 push a back-reference to this RV onto the array of backreferences
5845 associated with that magic. If the RV is magical, set magic will be
5846 called after the RV is cleared.
5852 Perl_sv_rvweaken(pTHX_ SV *const sv)
5856 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5858 if (!SvOK(sv)) /* let undefs pass */
5861 Perl_croak(aTHX_ "Can't weaken a nonreference");
5862 else if (SvWEAKREF(sv)) {
5863 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5866 else if (SvREADONLY(sv)) croak_no_modify();
5868 Perl_sv_add_backref(aTHX_ tsv, sv);
5870 SvREFCNT_dec_NN(tsv);
5875 =for apidoc sv_get_backrefs
5877 If C<sv> is the target of a weak reference then it returns the back
5878 references structure associated with the sv; otherwise return C<NULL>.
5880 When returning a non-null result the type of the return is relevant. If it
5881 is an AV then the elements of the AV are the weak reference RVs which
5882 point at this item. If it is any other type then the item itself is the
5885 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5886 C<Perl_sv_kill_backrefs()>
5892 Perl_sv_get_backrefs(SV *const sv)
5896 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5898 /* find slot to store array or singleton backref */
5900 if (SvTYPE(sv) == SVt_PVHV) {
5902 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5903 backrefs = (SV *)iter->xhv_backreferences;
5905 } else if (SvMAGICAL(sv)) {
5906 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5908 backrefs = mg->mg_obj;
5913 /* Give tsv backref magic if it hasn't already got it, then push a
5914 * back-reference to sv onto the array associated with the backref magic.
5916 * As an optimisation, if there's only one backref and it's not an AV,
5917 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5918 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5922 /* A discussion about the backreferences array and its refcount:
5924 * The AV holding the backreferences is pointed to either as the mg_obj of
5925 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5926 * xhv_backreferences field. The array is created with a refcount
5927 * of 2. This means that if during global destruction the array gets
5928 * picked on before its parent to have its refcount decremented by the
5929 * random zapper, it won't actually be freed, meaning it's still there for
5930 * when its parent gets freed.
5932 * When the parent SV is freed, the extra ref is killed by
5933 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5934 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5936 * When a single backref SV is stored directly, it is not reference
5941 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5947 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5949 /* find slot to store array or singleton backref */
5951 if (SvTYPE(tsv) == SVt_PVHV) {
5952 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5955 mg = mg_find(tsv, PERL_MAGIC_backref);
5957 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5958 svp = &(mg->mg_obj);
5961 /* create or retrieve the array */
5963 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5964 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5968 mg->mg_flags |= MGf_REFCOUNTED;
5971 SvREFCNT_inc_simple_void_NN(av);
5972 /* av now has a refcnt of 2; see discussion above */
5973 av_extend(av, *svp ? 2 : 1);
5975 /* move single existing backref to the array */
5976 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5981 av = MUTABLE_AV(*svp);
5983 /* optimisation: store single backref directly in HvAUX or mg_obj */
5987 assert(SvTYPE(av) == SVt_PVAV);
5988 if (AvFILLp(av) >= AvMAX(av)) {
5989 av_extend(av, AvFILLp(av)+1);
5992 /* push new backref */
5993 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5996 /* delete a back-reference to ourselves from the backref magic associated
5997 * with the SV we point to.
6001 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6005 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6007 if (SvTYPE(tsv) == SVt_PVHV) {
6009 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6011 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6012 /* It's possible for the the last (strong) reference to tsv to have
6013 become freed *before* the last thing holding a weak reference.
6014 If both survive longer than the backreferences array, then when
6015 the referent's reference count drops to 0 and it is freed, it's
6016 not able to chase the backreferences, so they aren't NULLed.
6018 For example, a CV holds a weak reference to its stash. If both the
6019 CV and the stash survive longer than the backreferences array,
6020 and the CV gets picked for the SvBREAK() treatment first,
6021 *and* it turns out that the stash is only being kept alive because
6022 of an our variable in the pad of the CV, then midway during CV
6023 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6024 It ends up pointing to the freed HV. Hence it's chased in here, and
6025 if this block wasn't here, it would hit the !svp panic just below.
6027 I don't believe that "better" destruction ordering is going to help
6028 here - during global destruction there's always going to be the
6029 chance that something goes out of order. We've tried to make it
6030 foolproof before, and it only resulted in evolutionary pressure on
6031 fools. Which made us look foolish for our hubris. :-(
6037 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6038 svp = mg ? &(mg->mg_obj) : NULL;
6042 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6044 /* It's possible that sv is being freed recursively part way through the
6045 freeing of tsv. If this happens, the backreferences array of tsv has
6046 already been freed, and so svp will be NULL. If this is the case,
6047 we should not panic. Instead, nothing needs doing, so return. */
6048 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6050 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6051 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6054 if (SvTYPE(*svp) == SVt_PVAV) {
6058 AV * const av = (AV*)*svp;
6060 assert(!SvIS_FREED(av));
6064 /* for an SV with N weak references to it, if all those
6065 * weak refs are deleted, then sv_del_backref will be called
6066 * N times and O(N^2) compares will be done within the backref
6067 * array. To ameliorate this potential slowness, we:
6068 * 1) make sure this code is as tight as possible;
6069 * 2) when looking for SV, look for it at both the head and tail of the
6070 * array first before searching the rest, since some create/destroy
6071 * patterns will cause the backrefs to be freed in order.
6078 SV **p = &svp[fill];
6079 SV *const topsv = *p;
6086 /* We weren't the last entry.
6087 An unordered list has this property that you
6088 can take the last element off the end to fill
6089 the hole, and it's still an unordered list :-)
6095 break; /* should only be one */
6102 AvFILLp(av) = fill-1;
6104 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6105 /* freed AV; skip */
6108 /* optimisation: only a single backref, stored directly */
6110 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6111 (void*)*svp, (void*)sv);
6118 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6124 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6129 /* after multiple passes through Perl_sv_clean_all() for a thingy
6130 * that has badly leaked, the backref array may have gotten freed,
6131 * since we only protect it against 1 round of cleanup */
6132 if (SvIS_FREED(av)) {
6133 if (PL_in_clean_all) /* All is fair */
6136 "panic: magic_killbackrefs (freed backref AV/SV)");
6140 is_array = (SvTYPE(av) == SVt_PVAV);
6142 assert(!SvIS_FREED(av));
6145 last = svp + AvFILLp(av);
6148 /* optimisation: only a single backref, stored directly */
6154 while (svp <= last) {
6156 SV *const referrer = *svp;
6157 if (SvWEAKREF(referrer)) {
6158 /* XXX Should we check that it hasn't changed? */
6159 assert(SvROK(referrer));
6160 SvRV_set(referrer, 0);
6162 SvWEAKREF_off(referrer);
6163 SvSETMAGIC(referrer);
6164 } else if (SvTYPE(referrer) == SVt_PVGV ||
6165 SvTYPE(referrer) == SVt_PVLV) {
6166 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6167 /* You lookin' at me? */
6168 assert(GvSTASH(referrer));
6169 assert(GvSTASH(referrer) == (const HV *)sv);
6170 GvSTASH(referrer) = 0;
6171 } else if (SvTYPE(referrer) == SVt_PVCV ||
6172 SvTYPE(referrer) == SVt_PVFM) {
6173 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6174 /* You lookin' at me? */
6175 assert(CvSTASH(referrer));
6176 assert(CvSTASH(referrer) == (const HV *)sv);
6177 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6180 assert(SvTYPE(sv) == SVt_PVGV);
6181 /* You lookin' at me? */
6182 assert(CvGV(referrer));
6183 assert(CvGV(referrer) == (const GV *)sv);
6184 anonymise_cv_maybe(MUTABLE_GV(sv),
6185 MUTABLE_CV(referrer));
6190 "panic: magic_killbackrefs (flags=%"UVxf")",
6191 (UV)SvFLAGS(referrer));
6202 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6208 =for apidoc sv_insert
6210 Inserts a string at the specified offset/length within the SV. Similar to
6211 the Perl C<substr()> function. Handles get magic.
6213 =for apidoc sv_insert_flags
6215 Same as C<sv_insert>, but the extra C<flags> are passed to the
6216 C<SvPV_force_flags> that applies to C<bigstr>.
6222 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6228 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6231 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6233 SvPV_force_flags(bigstr, curlen, flags);
6234 (void)SvPOK_only_UTF8(bigstr);
6235 if (offset + len > curlen) {
6236 SvGROW(bigstr, offset+len+1);
6237 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6238 SvCUR_set(bigstr, offset+len);
6242 i = littlelen - len;
6243 if (i > 0) { /* string might grow */
6244 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6245 mid = big + offset + len;
6246 midend = bigend = big + SvCUR(bigstr);
6249 while (midend > mid) /* shove everything down */
6250 *--bigend = *--midend;
6251 Move(little,big+offset,littlelen,char);
6252 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6257 Move(little,SvPVX(bigstr)+offset,len,char);
6262 big = SvPVX(bigstr);
6265 bigend = big + SvCUR(bigstr);
6267 if (midend > bigend)
6268 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6271 if (mid - big > bigend - midend) { /* faster to shorten from end */
6273 Move(little, mid, littlelen,char);
6276 i = bigend - midend;
6278 Move(midend, mid, i,char);
6282 SvCUR_set(bigstr, mid - big);
6284 else if ((i = mid - big)) { /* faster from front */
6285 midend -= littlelen;
6287 Move(big, midend - i, i, char);
6288 sv_chop(bigstr,midend-i);
6290 Move(little, mid, littlelen,char);
6292 else if (littlelen) {
6293 midend -= littlelen;
6294 sv_chop(bigstr,midend);
6295 Move(little,midend,littlelen,char);
6298 sv_chop(bigstr,midend);
6304 =for apidoc sv_replace
6306 Make the first argument a copy of the second, then delete the original.
6307 The target SV physically takes over ownership of the body of the source SV
6308 and inherits its flags; however, the target keeps any magic it owns,
6309 and any magic in the source is discarded.
6310 Note that this is a rather specialist SV copying operation; most of the
6311 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6317 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6319 const U32 refcnt = SvREFCNT(sv);
6321 PERL_ARGS_ASSERT_SV_REPLACE;
6323 SV_CHECK_THINKFIRST_COW_DROP(sv);
6324 if (SvREFCNT(nsv) != 1) {
6325 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6326 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6328 if (SvMAGICAL(sv)) {
6332 sv_upgrade(nsv, SVt_PVMG);
6333 SvMAGIC_set(nsv, SvMAGIC(sv));
6334 SvFLAGS(nsv) |= SvMAGICAL(sv);
6336 SvMAGIC_set(sv, NULL);
6340 assert(!SvREFCNT(sv));
6341 #ifdef DEBUG_LEAKING_SCALARS
6342 sv->sv_flags = nsv->sv_flags;
6343 sv->sv_any = nsv->sv_any;
6344 sv->sv_refcnt = nsv->sv_refcnt;
6345 sv->sv_u = nsv->sv_u;
6347 StructCopy(nsv,sv,SV);
6349 if(SvTYPE(sv) == SVt_IV) {
6350 SET_SVANY_FOR_BODYLESS_IV(sv);
6354 SvREFCNT(sv) = refcnt;
6355 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6360 /* We're about to free a GV which has a CV that refers back to us.
6361 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6365 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6370 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6373 assert(SvREFCNT(gv) == 0);
6374 assert(isGV(gv) && isGV_with_GP(gv));
6376 assert(!CvANON(cv));
6377 assert(CvGV(cv) == gv);
6378 assert(!CvNAMED(cv));
6380 /* will the CV shortly be freed by gp_free() ? */
6381 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6382 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6386 /* if not, anonymise: */
6387 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6388 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6389 : newSVpvn_flags( "__ANON__", 8, 0 );
6390 sv_catpvs(gvname, "::__ANON__");
6391 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6392 SvREFCNT_dec_NN(gvname);
6396 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6401 =for apidoc sv_clear
6403 Clear an SV: call any destructors, free up any memory used by the body,
6404 and free the body itself. The SV's head is I<not> freed, although
6405 its type is set to all 1's so that it won't inadvertently be assumed
6406 to be live during global destruction etc.
6407 This function should only be called when C<REFCNT> is zero. Most of the time
6408 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6415 Perl_sv_clear(pTHX_ SV *const orig_sv)
6420 const struct body_details *sv_type_details;
6424 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6425 Not strictly necessary */
6427 PERL_ARGS_ASSERT_SV_CLEAR;
6429 /* within this loop, sv is the SV currently being freed, and
6430 * iter_sv is the most recent AV or whatever that's being iterated
6431 * over to provide more SVs */
6437 assert(SvREFCNT(sv) == 0);
6438 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6440 if (type <= SVt_IV) {
6441 /* See the comment in sv.h about the collusion between this
6442 * early return and the overloading of the NULL slots in the
6446 SvFLAGS(sv) &= SVf_BREAK;
6447 SvFLAGS(sv) |= SVTYPEMASK;
6451 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6452 for another purpose */
6453 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6455 if (type >= SVt_PVMG) {
6457 if (!curse(sv, 1)) goto get_next_sv;
6458 type = SvTYPE(sv); /* destructor may have changed it */
6460 /* Free back-references before magic, in case the magic calls
6461 * Perl code that has weak references to sv. */
6462 if (type == SVt_PVHV) {
6463 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6467 else if (SvMAGIC(sv)) {
6468 /* Free back-references before other types of magic. */
6469 sv_unmagic(sv, PERL_MAGIC_backref);
6475 /* case SVt_INVLIST: */
6478 IoIFP(sv) != PerlIO_stdin() &&
6479 IoIFP(sv) != PerlIO_stdout() &&
6480 IoIFP(sv) != PerlIO_stderr() &&
6481 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6483 io_close(MUTABLE_IO(sv), NULL, FALSE,
6484 (IoTYPE(sv) == IoTYPE_WRONLY ||
6485 IoTYPE(sv) == IoTYPE_RDWR ||
6486 IoTYPE(sv) == IoTYPE_APPEND));
6488 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6489 PerlDir_close(IoDIRP(sv));
6490 IoDIRP(sv) = (DIR*)NULL;
6491 Safefree(IoTOP_NAME(sv));
6492 Safefree(IoFMT_NAME(sv));
6493 Safefree(IoBOTTOM_NAME(sv));
6494 if ((const GV *)sv == PL_statgv)
6498 /* FIXME for plugins */
6500 pregfree2((REGEXP*) sv);
6504 cv_undef(MUTABLE_CV(sv));
6505 /* If we're in a stash, we don't own a reference to it.
6506 * However it does have a back reference to us, which needs to
6508 if ((stash = CvSTASH(sv)))
6509 sv_del_backref(MUTABLE_SV(stash), sv);
6512 if (PL_last_swash_hv == (const HV *)sv) {
6513 PL_last_swash_hv = NULL;
6515 if (HvTOTALKEYS((HV*)sv) > 0) {
6517 /* this statement should match the one at the beginning of
6518 * hv_undef_flags() */
6519 if ( PL_phase != PERL_PHASE_DESTRUCT
6520 && (hek = HvNAME_HEK((HV*)sv)))
6522 if (PL_stashcache) {
6523 DEBUG_o(Perl_deb(aTHX_
6524 "sv_clear clearing PL_stashcache for '%"HEKf
6527 (void)hv_deletehek(PL_stashcache,
6530 hv_name_set((HV*)sv, NULL, 0, 0);
6533 /* save old iter_sv in unused SvSTASH field */
6534 assert(!SvOBJECT(sv));
6535 SvSTASH(sv) = (HV*)iter_sv;
6538 /* save old hash_index in unused SvMAGIC field */
6539 assert(!SvMAGICAL(sv));
6540 assert(!SvMAGIC(sv));
6541 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6544 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6545 goto get_next_sv; /* process this new sv */
6547 /* free empty hash */
6548 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6549 assert(!HvARRAY((HV*)sv));
6553 AV* av = MUTABLE_AV(sv);
6554 if (PL_comppad == av) {
6558 if (AvREAL(av) && AvFILLp(av) > -1) {
6559 next_sv = AvARRAY(av)[AvFILLp(av)--];
6560 /* save old iter_sv in top-most slot of AV,
6561 * and pray that it doesn't get wiped in the meantime */
6562 AvARRAY(av)[AvMAX(av)] = iter_sv;
6564 goto get_next_sv; /* process this new sv */
6566 Safefree(AvALLOC(av));
6571 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6572 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6573 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6574 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6576 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6577 SvREFCNT_dec(LvTARG(sv));
6578 if (isREGEXP(sv)) goto freeregexp;
6581 if (isGV_with_GP(sv)) {
6582 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6583 && HvENAME_get(stash))
6584 mro_method_changed_in(stash);
6585 gp_free(MUTABLE_GV(sv));
6587 unshare_hek(GvNAME_HEK(sv));
6588 /* If we're in a stash, we don't own a reference to it.
6589 * However it does have a back reference to us, which
6590 * needs to be cleared. */
6591 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6592 sv_del_backref(MUTABLE_SV(stash), sv);
6594 /* FIXME. There are probably more unreferenced pointers to SVs
6595 * in the interpreter struct that we should check and tidy in
6596 * a similar fashion to this: */
6597 /* See also S_sv_unglob, which does the same thing. */
6598 if ((const GV *)sv == PL_last_in_gv)
6599 PL_last_in_gv = NULL;
6600 else if ((const GV *)sv == PL_statgv)
6602 else if ((const GV *)sv == PL_stderrgv)
6611 /* Don't bother with SvOOK_off(sv); as we're only going to
6615 SvOOK_offset(sv, offset);
6616 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6617 /* Don't even bother with turning off the OOK flag. */
6622 SV * const target = SvRV(sv);
6624 sv_del_backref(target, sv);
6630 else if (SvPVX_const(sv)
6631 && !(SvTYPE(sv) == SVt_PVIO
6632 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6636 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6640 if (CowREFCNT(sv)) {
6647 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6652 Safefree(SvPVX_mutable(sv));
6656 else if (SvPVX_const(sv) && SvLEN(sv)
6657 && !(SvTYPE(sv) == SVt_PVIO
6658 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6659 Safefree(SvPVX_mutable(sv));
6660 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6661 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6671 SvFLAGS(sv) &= SVf_BREAK;
6672 SvFLAGS(sv) |= SVTYPEMASK;
6674 sv_type_details = bodies_by_type + type;
6675 if (sv_type_details->arena) {
6676 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6677 &PL_body_roots[type]);
6679 else if (sv_type_details->body_size) {
6680 safefree(SvANY(sv));
6684 /* caller is responsible for freeing the head of the original sv */
6685 if (sv != orig_sv && !SvREFCNT(sv))
6688 /* grab and free next sv, if any */
6696 else if (!iter_sv) {
6698 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6699 AV *const av = (AV*)iter_sv;
6700 if (AvFILLp(av) > -1) {
6701 sv = AvARRAY(av)[AvFILLp(av)--];
6703 else { /* no more elements of current AV to free */
6706 /* restore previous value, squirrelled away */
6707 iter_sv = AvARRAY(av)[AvMAX(av)];
6708 Safefree(AvALLOC(av));
6711 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6712 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6713 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6714 /* no more elements of current HV to free */
6717 /* Restore previous values of iter_sv and hash_index,
6718 * squirrelled away */
6719 assert(!SvOBJECT(sv));
6720 iter_sv = (SV*)SvSTASH(sv);
6721 assert(!SvMAGICAL(sv));
6722 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6724 /* perl -DA does not like rubbish in SvMAGIC. */
6728 /* free any remaining detritus from the hash struct */
6729 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6730 assert(!HvARRAY((HV*)sv));
6735 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6739 if (!SvREFCNT(sv)) {
6743 if (--(SvREFCNT(sv)))
6747 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6748 "Attempt to free temp prematurely: SV 0x%"UVxf
6749 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6753 if (SvIMMORTAL(sv)) {
6754 /* make sure SvREFCNT(sv)==0 happens very seldom */
6755 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6764 /* This routine curses the sv itself, not the object referenced by sv. So
6765 sv does not have to be ROK. */
6768 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6769 PERL_ARGS_ASSERT_CURSE;
6770 assert(SvOBJECT(sv));
6772 if (PL_defstash && /* Still have a symbol table? */
6778 stash = SvSTASH(sv);
6779 assert(SvTYPE(stash) == SVt_PVHV);
6780 if (HvNAME(stash)) {
6781 CV* destructor = NULL;
6782 struct mro_meta *meta;
6784 assert (SvOOK(stash));
6786 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6789 /* don't make this an initialization above the assert, since it needs
6791 meta = HvMROMETA(stash);
6792 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6793 destructor = meta->destroy;
6794 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6795 (void *)destructor, HvNAME(stash)) );
6798 bool autoload = FALSE;
6800 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6802 destructor = GvCV(gv);
6804 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6805 GV_AUTOLOAD_ISMETHOD);
6807 destructor = GvCV(gv);
6811 /* we don't cache AUTOLOAD for DESTROY, since this code
6812 would then need to set $__PACKAGE__::AUTOLOAD, or the
6813 equivalent for XS AUTOLOADs */
6815 meta->destroy_gen = PL_sub_generation;
6816 meta->destroy = destructor;
6818 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6819 (void *)destructor, HvNAME(stash)) );
6822 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6826 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6828 /* A constant subroutine can have no side effects, so
6829 don't bother calling it. */
6830 && !CvCONST(destructor)
6831 /* Don't bother calling an empty destructor or one that
6832 returns immediately. */
6833 && (CvISXSUB(destructor)
6834 || (CvSTART(destructor)
6835 && (CvSTART(destructor)->op_next->op_type
6837 && (CvSTART(destructor)->op_next->op_type
6839 || CvSTART(destructor)->op_next->op_next->op_type
6845 SV* const tmpref = newRV(sv);
6846 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6848 PUSHSTACKi(PERLSI_DESTROY);
6853 call_sv(MUTABLE_SV(destructor),
6854 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6858 if(SvREFCNT(tmpref) < 2) {
6859 /* tmpref is not kept alive! */
6861 SvRV_set(tmpref, NULL);
6864 SvREFCNT_dec_NN(tmpref);
6867 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6870 if (check_refcnt && SvREFCNT(sv)) {
6871 if (PL_in_clean_objs)
6873 "DESTROY created new reference to dead object '%"HEKf"'",
6874 HEKfARG(HvNAME_HEK(stash)));
6875 /* DESTROY gave object new lease on life */
6881 HV * const stash = SvSTASH(sv);
6882 /* Curse before freeing the stash, as freeing the stash could cause
6883 a recursive call into S_curse. */
6884 SvOBJECT_off(sv); /* Curse the object. */
6885 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6886 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6892 =for apidoc sv_newref
6894 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6901 Perl_sv_newref(pTHX_ SV *const sv)
6903 PERL_UNUSED_CONTEXT;
6912 Decrement an SV's reference count, and if it drops to zero, call
6913 C<sv_clear> to invoke destructors and free up any memory used by
6914 the body; finally, deallocating the SV's head itself.
6915 Normally called via a wrapper macro C<SvREFCNT_dec>.
6921 Perl_sv_free(pTHX_ SV *const sv)
6927 /* Private helper function for SvREFCNT_dec().
6928 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6931 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6935 PERL_ARGS_ASSERT_SV_FREE2;
6937 if (LIKELY( rc == 1 )) {
6943 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6944 "Attempt to free temp prematurely: SV 0x%"UVxf
6945 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6949 if (SvIMMORTAL(sv)) {
6950 /* make sure SvREFCNT(sv)==0 happens very seldom */
6951 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6955 if (! SvREFCNT(sv)) /* may have have been resurrected */
6960 /* handle exceptional cases */
6964 if (SvFLAGS(sv) & SVf_BREAK)
6965 /* this SV's refcnt has been artificially decremented to
6966 * trigger cleanup */
6968 if (PL_in_clean_all) /* All is fair */
6970 if (SvIMMORTAL(sv)) {
6971 /* make sure SvREFCNT(sv)==0 happens very seldom */
6972 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6975 if (ckWARN_d(WARN_INTERNAL)) {
6976 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6977 Perl_dump_sv_child(aTHX_ sv);
6979 #ifdef DEBUG_LEAKING_SCALARS
6982 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6983 if (PL_warnhook == PERL_WARNHOOK_FATAL
6984 || ckDEAD(packWARN(WARN_INTERNAL))) {
6985 /* Don't let Perl_warner cause us to escape our fate: */
6989 /* This may not return: */
6990 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6991 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6992 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6995 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7005 Returns the length of the string in the SV. Handles magic and type
7006 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7007 gives raw access to the C<xpv_cur> slot.
7013 Perl_sv_len(pTHX_ SV *const sv)
7020 (void)SvPV_const(sv, len);
7025 =for apidoc sv_len_utf8
7027 Returns the number of characters in the string in an SV, counting wide
7028 UTF-8 bytes as a single character. Handles magic and type coercion.
7034 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7035 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7036 * (Note that the mg_len is not the length of the mg_ptr field.
7037 * This allows the cache to store the character length of the string without
7038 * needing to malloc() extra storage to attach to the mg_ptr.)
7043 Perl_sv_len_utf8(pTHX_ SV *const sv)
7049 return sv_len_utf8_nomg(sv);
7053 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7056 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7058 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7060 if (PL_utf8cache && SvUTF8(sv)) {
7062 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7064 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7065 if (mg->mg_len != -1)
7068 /* We can use the offset cache for a headstart.
7069 The longer value is stored in the first pair. */
7070 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7072 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7076 if (PL_utf8cache < 0) {
7077 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7078 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7082 ulen = Perl_utf8_length(aTHX_ s, s + len);
7083 utf8_mg_len_cache_update(sv, &mg, ulen);
7087 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7090 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7093 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7094 STRLEN *const uoffset_p, bool *const at_end)
7096 const U8 *s = start;
7097 STRLEN uoffset = *uoffset_p;
7099 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7101 while (s < send && uoffset) {
7108 else if (s > send) {
7110 /* This is the existing behaviour. Possibly it should be a croak, as
7111 it's actually a bounds error */
7114 *uoffset_p -= uoffset;
7118 /* Given the length of the string in both bytes and UTF-8 characters, decide
7119 whether to walk forwards or backwards to find the byte corresponding to
7120 the passed in UTF-8 offset. */
7122 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7123 STRLEN uoffset, const STRLEN uend)
7125 STRLEN backw = uend - uoffset;
7127 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7129 if (uoffset < 2 * backw) {
7130 /* The assumption is that going forwards is twice the speed of going
7131 forward (that's where the 2 * backw comes from).
7132 (The real figure of course depends on the UTF-8 data.) */
7133 const U8 *s = start;
7135 while (s < send && uoffset--)
7145 while (UTF8_IS_CONTINUATION(*send))
7148 return send - start;
7151 /* For the string representation of the given scalar, find the byte
7152 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7153 give another position in the string, *before* the sought offset, which
7154 (which is always true, as 0, 0 is a valid pair of positions), which should
7155 help reduce the amount of linear searching.
7156 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7157 will be used to reduce the amount of linear searching. The cache will be
7158 created if necessary, and the found value offered to it for update. */
7160 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7161 const U8 *const send, STRLEN uoffset,
7162 STRLEN uoffset0, STRLEN boffset0)
7164 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7166 bool at_end = FALSE;
7168 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7170 assert (uoffset >= uoffset0);
7175 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7177 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7178 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7179 if ((*mgp)->mg_ptr) {
7180 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7181 if (cache[0] == uoffset) {
7182 /* An exact match. */
7185 if (cache[2] == uoffset) {
7186 /* An exact match. */
7190 if (cache[0] < uoffset) {
7191 /* The cache already knows part of the way. */
7192 if (cache[0] > uoffset0) {
7193 /* The cache knows more than the passed in pair */
7194 uoffset0 = cache[0];
7195 boffset0 = cache[1];
7197 if ((*mgp)->mg_len != -1) {
7198 /* And we know the end too. */
7200 + sv_pos_u2b_midway(start + boffset0, send,
7202 (*mgp)->mg_len - uoffset0);
7204 uoffset -= uoffset0;
7206 + sv_pos_u2b_forwards(start + boffset0,
7207 send, &uoffset, &at_end);
7208 uoffset += uoffset0;
7211 else if (cache[2] < uoffset) {
7212 /* We're between the two cache entries. */
7213 if (cache[2] > uoffset0) {
7214 /* and the cache knows more than the passed in pair */
7215 uoffset0 = cache[2];
7216 boffset0 = cache[3];
7220 + sv_pos_u2b_midway(start + boffset0,
7223 cache[0] - uoffset0);
7226 + sv_pos_u2b_midway(start + boffset0,
7229 cache[2] - uoffset0);
7233 else if ((*mgp)->mg_len != -1) {
7234 /* If we can take advantage of a passed in offset, do so. */
7235 /* In fact, offset0 is either 0, or less than offset, so don't
7236 need to worry about the other possibility. */
7238 + sv_pos_u2b_midway(start + boffset0, send,
7240 (*mgp)->mg_len - uoffset0);
7245 if (!found || PL_utf8cache < 0) {
7246 STRLEN real_boffset;
7247 uoffset -= uoffset0;
7248 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7249 send, &uoffset, &at_end);
7250 uoffset += uoffset0;
7252 if (found && PL_utf8cache < 0)
7253 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7255 boffset = real_boffset;
7258 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7260 utf8_mg_len_cache_update(sv, mgp, uoffset);
7262 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7269 =for apidoc sv_pos_u2b_flags
7271 Converts the offset from a count of UTF-8 chars from
7272 the start of the string, to a count of the equivalent number of bytes; if
7273 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7274 C<offset>, rather than from the start
7275 of the string. Handles type coercion.
7276 C<flags> is passed to C<SvPV_flags>, and usually should be
7277 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7283 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7284 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7285 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7290 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7297 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7299 start = (U8*)SvPV_flags(sv, len, flags);
7301 const U8 * const send = start + len;
7303 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7306 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7307 is 0, and *lenp is already set to that. */) {
7308 /* Convert the relative offset to absolute. */
7309 const STRLEN uoffset2 = uoffset + *lenp;
7310 const STRLEN boffset2
7311 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7312 uoffset, boffset) - boffset;
7326 =for apidoc sv_pos_u2b
7328 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7329 the start of the string, to a count of the equivalent number of bytes; if
7330 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7331 the offset, rather than from the start of the string. Handles magic and
7334 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7341 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7342 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7343 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7347 /* This function is subject to size and sign problems */
7350 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7352 PERL_ARGS_ASSERT_SV_POS_U2B;
7355 STRLEN ulen = (STRLEN)*lenp;
7356 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7357 SV_GMAGIC|SV_CONST_RETURN);
7360 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7361 SV_GMAGIC|SV_CONST_RETURN);
7366 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7369 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7370 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7373 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7374 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7375 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7379 (*mgp)->mg_len = ulen;
7382 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7383 byte length pairing. The (byte) length of the total SV is passed in too,
7384 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7385 may not have updated SvCUR, so we can't rely on reading it directly.
7387 The proffered utf8/byte length pairing isn't used if the cache already has
7388 two pairs, and swapping either for the proffered pair would increase the
7389 RMS of the intervals between known byte offsets.
7391 The cache itself consists of 4 STRLEN values
7392 0: larger UTF-8 offset
7393 1: corresponding byte offset
7394 2: smaller UTF-8 offset
7395 3: corresponding byte offset
7397 Unused cache pairs have the value 0, 0.
7398 Keeping the cache "backwards" means that the invariant of
7399 cache[0] >= cache[2] is maintained even with empty slots, which means that
7400 the code that uses it doesn't need to worry if only 1 entry has actually
7401 been set to non-zero. It also makes the "position beyond the end of the
7402 cache" logic much simpler, as the first slot is always the one to start
7406 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7407 const STRLEN utf8, const STRLEN blen)
7411 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7416 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7417 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7418 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7420 (*mgp)->mg_len = -1;
7424 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7425 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7426 (*mgp)->mg_ptr = (char *) cache;
7430 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7431 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7432 a pointer. Note that we no longer cache utf8 offsets on refer-
7433 ences, but this check is still a good idea, for robustness. */
7434 const U8 *start = (const U8 *) SvPVX_const(sv);
7435 const STRLEN realutf8 = utf8_length(start, start + byte);
7437 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7441 /* Cache is held with the later position first, to simplify the code
7442 that deals with unbounded ends. */
7444 ASSERT_UTF8_CACHE(cache);
7445 if (cache[1] == 0) {
7446 /* Cache is totally empty */
7449 } else if (cache[3] == 0) {
7450 if (byte > cache[1]) {
7451 /* New one is larger, so goes first. */
7452 cache[2] = cache[0];
7453 cache[3] = cache[1];
7461 /* float casts necessary? XXX */
7462 #define THREEWAY_SQUARE(a,b,c,d) \
7463 ((float)((d) - (c))) * ((float)((d) - (c))) \
7464 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7465 + ((float)((b) - (a))) * ((float)((b) - (a)))
7467 /* Cache has 2 slots in use, and we know three potential pairs.
7468 Keep the two that give the lowest RMS distance. Do the
7469 calculation in bytes simply because we always know the byte
7470 length. squareroot has the same ordering as the positive value,
7471 so don't bother with the actual square root. */
7472 if (byte > cache[1]) {
7473 /* New position is after the existing pair of pairs. */
7474 const float keep_earlier
7475 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7476 const float keep_later
7477 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7479 if (keep_later < keep_earlier) {
7480 cache[2] = cache[0];
7481 cache[3] = cache[1];
7487 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7488 float b, c, keep_earlier;
7489 if (byte > cache[3]) {
7490 /* New position is between the existing pair of pairs. */
7491 b = (float)cache[3];
7494 /* New position is before the existing pair of pairs. */
7496 c = (float)cache[3];
7498 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7499 if (byte > cache[3]) {
7500 if (keep_later < keep_earlier) {
7510 if (! (keep_later < keep_earlier)) {
7511 cache[0] = cache[2];
7512 cache[1] = cache[3];
7519 ASSERT_UTF8_CACHE(cache);
7522 /* We already know all of the way, now we may be able to walk back. The same
7523 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7524 backward is half the speed of walking forward. */
7526 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7527 const U8 *end, STRLEN endu)
7529 const STRLEN forw = target - s;
7530 STRLEN backw = end - target;
7532 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7534 if (forw < 2 * backw) {
7535 return utf8_length(s, target);
7538 while (end > target) {
7540 while (UTF8_IS_CONTINUATION(*end)) {
7549 =for apidoc sv_pos_b2u_flags
7551 Converts C<offset> from a count of bytes from the start of the string, to
7552 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7553 C<flags> is passed to C<SvPV_flags>, and usually should be
7554 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7560 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7561 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7566 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7569 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7575 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7577 s = (const U8*)SvPV_flags(sv, blen, flags);
7580 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7581 ", byte=%"UVuf, (UV)blen, (UV)offset);
7587 && SvTYPE(sv) >= SVt_PVMG
7588 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7591 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7592 if (cache[1] == offset) {
7593 /* An exact match. */
7596 if (cache[3] == offset) {
7597 /* An exact match. */
7601 if (cache[1] < offset) {
7602 /* We already know part of the way. */
7603 if (mg->mg_len != -1) {
7604 /* Actually, we know the end too. */
7606 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7607 s + blen, mg->mg_len - cache[0]);
7609 len = cache[0] + utf8_length(s + cache[1], send);
7612 else if (cache[3] < offset) {
7613 /* We're between the two cached pairs, so we do the calculation
7614 offset by the byte/utf-8 positions for the earlier pair,
7615 then add the utf-8 characters from the string start to
7617 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7618 s + cache[1], cache[0] - cache[2])
7622 else { /* cache[3] > offset */
7623 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7627 ASSERT_UTF8_CACHE(cache);
7629 } else if (mg->mg_len != -1) {
7630 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7634 if (!found || PL_utf8cache < 0) {
7635 const STRLEN real_len = utf8_length(s, send);
7637 if (found && PL_utf8cache < 0)
7638 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7644 utf8_mg_len_cache_update(sv, &mg, len);
7646 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7653 =for apidoc sv_pos_b2u
7655 Converts the value pointed to by C<offsetp> from a count of bytes from the
7656 start of the string, to a count of the equivalent number of UTF-8 chars.
7657 Handles magic and type coercion.
7659 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7666 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7667 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7672 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7674 PERL_ARGS_ASSERT_SV_POS_B2U;
7679 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7680 SV_GMAGIC|SV_CONST_RETURN);
7684 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7685 STRLEN real, SV *const sv)
7687 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7689 /* As this is debugging only code, save space by keeping this test here,
7690 rather than inlining it in all the callers. */
7691 if (from_cache == real)
7694 /* Need to turn the assertions off otherwise we may recurse infinitely
7695 while printing error messages. */
7696 SAVEI8(PL_utf8cache);
7698 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7699 func, (UV) from_cache, (UV) real, SVfARG(sv));
7705 Returns a boolean indicating whether the strings in the two SVs are
7706 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7707 coerce its args to strings if necessary.
7709 =for apidoc sv_eq_flags
7711 Returns a boolean indicating whether the strings in the two SVs are
7712 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7713 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7719 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7726 SV* svrecode = NULL;
7733 /* if pv1 and pv2 are the same, second SvPV_const call may
7734 * invalidate pv1 (if we are handling magic), so we may need to
7736 if (sv1 == sv2 && flags & SV_GMAGIC
7737 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7738 pv1 = SvPV_const(sv1, cur1);
7739 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7741 pv1 = SvPV_flags_const(sv1, cur1, flags);
7749 pv2 = SvPV_flags_const(sv2, cur2, flags);
7751 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7752 /* Differing utf8ness.
7753 * Do not UTF8size the comparands as a side-effect. */
7756 svrecode = newSVpvn(pv2, cur2);
7757 sv_recode_to_utf8(svrecode, _get_encoding());
7758 pv2 = SvPV_const(svrecode, cur2);
7761 svrecode = newSVpvn(pv1, cur1);
7762 sv_recode_to_utf8(svrecode, _get_encoding());
7763 pv1 = SvPV_const(svrecode, cur1);
7765 /* Now both are in UTF-8. */
7767 SvREFCNT_dec_NN(svrecode);
7773 /* sv1 is the UTF-8 one */
7774 return bytes_cmp_utf8((const U8*)pv2, cur2,
7775 (const U8*)pv1, cur1) == 0;
7778 /* sv2 is the UTF-8 one */
7779 return bytes_cmp_utf8((const U8*)pv1, cur1,
7780 (const U8*)pv2, cur2) == 0;
7786 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7788 SvREFCNT_dec(svrecode);
7796 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7797 string in C<sv1> is less than, equal to, or greater than the string in
7798 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7799 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7801 =for apidoc sv_cmp_flags
7803 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7804 string in C<sv1> is less than, equal to, or greater than the string in
7805 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7806 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7807 also C<L</sv_cmp_locale_flags>>.
7813 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7815 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7819 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7823 const char *pv1, *pv2;
7825 SV *svrecode = NULL;
7832 pv1 = SvPV_flags_const(sv1, cur1, flags);
7839 pv2 = SvPV_flags_const(sv2, cur2, flags);
7841 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7842 /* Differing utf8ness.
7843 * Do not UTF8size the comparands as a side-effect. */
7846 svrecode = newSVpvn(pv2, cur2);
7847 sv_recode_to_utf8(svrecode, _get_encoding());
7848 pv2 = SvPV_const(svrecode, cur2);
7851 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7852 (const U8*)pv1, cur1);
7853 return retval ? retval < 0 ? -1 : +1 : 0;
7858 svrecode = newSVpvn(pv1, cur1);
7859 sv_recode_to_utf8(svrecode, _get_encoding());
7860 pv1 = SvPV_const(svrecode, cur1);
7863 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7864 (const U8*)pv2, cur2);
7865 return retval ? retval < 0 ? -1 : +1 : 0;
7870 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7873 cmp = cur2 ? -1 : 0;
7877 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7880 if (! DO_UTF8(sv1)) {
7882 const I32 retval = memcmp((const void*)pv1,
7886 cmp = retval < 0 ? -1 : 1;
7887 } else if (cur1 == cur2) {
7890 cmp = cur1 < cur2 ? -1 : 1;
7894 else { /* Both are to be treated as UTF-EBCDIC */
7896 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7897 * which remaps code points 0-255. We therefore generally have to
7898 * unmap back to the original values to get an accurate comparison.
7899 * But we don't have to do that for UTF-8 invariants, as by
7900 * definition, they aren't remapped, nor do we have to do it for
7901 * above-latin1 code points, as they also aren't remapped. (This
7902 * code also works on ASCII platforms, but the memcmp() above is
7905 const char *e = pv1 + shortest_len;
7907 /* Find the first bytes that differ between the two strings */
7908 while (pv1 < e && *pv1 == *pv2) {
7914 if (pv1 == e) { /* Are the same all the way to the end */
7918 cmp = cur1 < cur2 ? -1 : 1;
7921 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7922 * in the strings were. The current bytes may or may not be
7923 * at the beginning of a character. But neither or both are
7924 * (or else earlier bytes would have been different). And
7925 * if we are in the middle of a character, the two
7926 * characters are comprised of the same number of bytes
7927 * (because in this case the start bytes are the same, and
7928 * the start bytes encode the character's length). */
7929 if (UTF8_IS_INVARIANT(*pv1))
7931 /* If both are invariants; can just compare directly */
7932 if (UTF8_IS_INVARIANT(*pv2)) {
7933 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7935 else /* Since *pv1 is invariant, it is the whole character,
7936 which means it is at the beginning of a character.
7937 That means pv2 is also at the beginning of a
7938 character (see earlier comment). Since it isn't
7939 invariant, it must be a start byte. If it starts a
7940 character whose code point is above 255, that
7941 character is greater than any single-byte char, which
7943 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7948 /* Here, pv2 points to a character composed of 2 bytes
7949 * whose code point is < 256. Get its code point and
7950 * compare with *pv1 */
7951 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7956 else /* The code point starting at pv1 isn't a single byte */
7957 if (UTF8_IS_INVARIANT(*pv2))
7959 /* But here, the code point starting at *pv2 is a single byte,
7960 * and so *pv1 must begin a character, hence is a start byte.
7961 * If that character is above 255, it is larger than any
7962 * single-byte char, which *pv2 is */
7963 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7967 /* Here, pv1 points to a character composed of 2 bytes
7968 * whose code point is < 256. Get its code point and
7969 * compare with the single byte character *pv2 */
7970 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7975 else /* Here, we've ruled out either *pv1 and *pv2 being
7976 invariant. That means both are part of variants, but not
7977 necessarily at the start of a character */
7978 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7979 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7981 /* Here, at least one is the start of a character, which means
7982 * the other is also a start byte. And the code point of at
7983 * least one of the characters is above 255. It is a
7984 * characteristic of UTF-EBCDIC that all start bytes for
7985 * above-latin1 code points are well behaved as far as code
7986 * point comparisons go, and all are larger than all other
7987 * start bytes, so the comparison with those is also well
7989 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7992 /* Here both *pv1 and *pv2 are part of variant characters.
7993 * They could be both continuations, or both start characters.
7994 * (One or both could even be an illegal start character (for
7995 * an overlong) which for the purposes of sorting we treat as
7997 if (UTF8_IS_CONTINUATION(*pv1)) {
7999 /* If they are continuations for code points above 255,
8000 * then comparing the current byte is sufficient, as there
8001 * is no remapping of these and so the comparison is
8002 * well-behaved. We determine if they are such
8003 * continuations by looking at the preceding byte. It
8004 * could be a start byte, from which we can tell if it is
8005 * for an above 255 code point. Or it could be a
8006 * continuation, which means the character occupies at
8007 * least 3 bytes, so must be above 255. */
8008 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8009 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8011 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8015 /* Here, the continuations are for code points below 256;
8016 * back up one to get to the start byte */
8021 /* We need to get the actual native code point of each of these
8022 * variants in order to compare them */
8023 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8024 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8033 SvREFCNT_dec(svrecode);
8039 =for apidoc sv_cmp_locale
8041 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8042 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8043 if necessary. See also C<L</sv_cmp>>.
8045 =for apidoc sv_cmp_locale_flags
8047 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8048 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8049 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8050 C<L</sv_cmp_flags>>.
8056 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8058 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8062 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8065 #ifdef USE_LOCALE_COLLATE
8071 if (PL_collation_standard)
8075 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8077 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8079 if (!pv1 || !len1) {
8090 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8093 return retval < 0 ? -1 : 1;
8096 * When the result of collation is equality, that doesn't mean
8097 * that there are no differences -- some locales exclude some
8098 * characters from consideration. So to avoid false equalities,
8099 * we use the raw string as a tiebreaker.
8106 PERL_UNUSED_ARG(flags);
8107 #endif /* USE_LOCALE_COLLATE */
8109 return sv_cmp(sv1, sv2);
8113 #ifdef USE_LOCALE_COLLATE
8116 =for apidoc sv_collxfrm
8118 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8119 C<L</sv_collxfrm_flags>>.
8121 =for apidoc sv_collxfrm_flags
8123 Add Collate Transform magic to an SV if it doesn't already have it. If the
8124 flags contain C<SV_GMAGIC>, it handles get-magic.
8126 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8127 scalar data of the variable, but transformed to such a format that a normal
8128 memory comparison can be used to compare the data according to the locale
8135 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8139 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8141 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8143 /* If we don't have collation magic on 'sv', or the locale has changed
8144 * since the last time we calculated it, get it and save it now */
8145 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8150 /* Free the old space */
8152 Safefree(mg->mg_ptr);
8154 s = SvPV_flags_const(sv, len, flags);
8155 if ((xf = mem_collxfrm(s, len, &xlen))) {
8157 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8172 if (mg && mg->mg_ptr) {
8174 return mg->mg_ptr + sizeof(PL_collation_ix);
8182 #endif /* USE_LOCALE_COLLATE */
8185 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8187 SV * const tsv = newSV(0);
8190 sv_gets(tsv, fp, 0);
8191 sv_utf8_upgrade_nomg(tsv);
8192 SvCUR_set(sv,append);
8195 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8199 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8202 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8203 /* Grab the size of the record we're getting */
8204 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8211 /* With a true, record-oriented file on VMS, we need to use read directly
8212 * to ensure that we respect RMS record boundaries. The user is responsible
8213 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8214 * record size) field. N.B. This is likely to produce invalid results on
8215 * varying-width character data when a record ends mid-character.
8217 fd = PerlIO_fileno(fp);
8219 && PerlLIO_fstat(fd, &st) == 0
8220 && (st.st_fab_rfm == FAB$C_VAR
8221 || st.st_fab_rfm == FAB$C_VFC
8222 || st.st_fab_rfm == FAB$C_FIX)) {
8224 bytesread = PerlLIO_read(fd, buffer, recsize);
8226 else /* in-memory file from PerlIO::Scalar
8227 * or not a record-oriented file
8231 bytesread = PerlIO_read(fp, buffer, recsize);
8233 /* At this point, the logic in sv_get() means that sv will
8234 be treated as utf-8 if the handle is utf8.
8236 if (PerlIO_isutf8(fp) && bytesread > 0) {
8237 char *bend = buffer + bytesread;
8238 char *bufp = buffer;
8239 size_t charcount = 0;
8240 bool charstart = TRUE;
8243 while (charcount < recsize) {
8244 /* count accumulated characters */
8245 while (bufp < bend) {
8247 skip = UTF8SKIP(bufp);
8249 if (bufp + skip > bend) {
8250 /* partial at the end */
8261 if (charcount < recsize) {
8263 STRLEN bufp_offset = bufp - buffer;
8264 SSize_t morebytesread;
8266 /* originally I read enough to fill any incomplete
8267 character and the first byte of the next
8268 character if needed, but if there's many
8269 multi-byte encoded characters we're going to be
8270 making a read call for every character beyond
8271 the original read size.
8273 So instead, read the rest of the character if
8274 any, and enough bytes to match at least the
8275 start bytes for each character we're going to
8279 readsize = recsize - charcount;
8281 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8282 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8283 bend = buffer + bytesread;
8284 morebytesread = PerlIO_read(fp, bend, readsize);
8285 if (morebytesread <= 0) {
8286 /* we're done, if we still have incomplete
8287 characters the check code in sv_gets() will
8290 I'd originally considered doing
8291 PerlIO_ungetc() on all but the lead
8292 character of the incomplete character, but
8293 read() doesn't do that, so I don't.
8298 /* prepare to scan some more */
8299 bytesread += morebytesread;
8300 bend = buffer + bytesread;
8301 bufp = buffer + bufp_offset;
8309 SvCUR_set(sv, bytesread + append);
8310 buffer[bytesread] = '\0';
8311 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8317 Get a line from the filehandle and store it into the SV, optionally
8318 appending to the currently-stored string. If C<append> is not 0, the
8319 line is appended to the SV instead of overwriting it. C<append> should
8320 be set to the byte offset that the appended string should start at
8321 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8327 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8337 PERL_ARGS_ASSERT_SV_GETS;
8339 if (SvTHINKFIRST(sv))
8340 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8341 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8343 However, perlbench says it's slower, because the existing swipe code
8344 is faster than copy on write.
8345 Swings and roundabouts. */
8346 SvUPGRADE(sv, SVt_PV);
8349 /* line is going to be appended to the existing buffer in the sv */
8350 if (PerlIO_isutf8(fp)) {
8352 sv_utf8_upgrade_nomg(sv);
8353 sv_pos_u2b(sv,&append,0);
8355 } else if (SvUTF8(sv)) {
8356 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8362 /* not appending - "clear" the string by setting SvCUR to 0,
8363 * the pv is still avaiable. */
8366 if (PerlIO_isutf8(fp))
8369 if (IN_PERL_COMPILETIME) {
8370 /* we always read code in line mode */
8374 else if (RsSNARF(PL_rs)) {
8375 /* If it is a regular disk file use size from stat() as estimate
8376 of amount we are going to read -- may result in mallocing
8377 more memory than we really need if the layers below reduce
8378 the size we read (e.g. CRLF or a gzip layer).
8381 int fd = PerlIO_fileno(fp);
8382 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8383 const Off_t offset = PerlIO_tell(fp);
8384 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8385 #ifdef PERL_COPY_ON_WRITE
8386 /* Add an extra byte for the sake of copy-on-write's
8387 * buffer reference count. */
8388 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8390 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8397 else if (RsRECORD(PL_rs)) {
8398 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8400 else if (RsPARA(PL_rs)) {
8406 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8407 if (PerlIO_isutf8(fp)) {
8408 rsptr = SvPVutf8(PL_rs, rslen);
8411 if (SvUTF8(PL_rs)) {
8412 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8413 Perl_croak(aTHX_ "Wide character in $/");
8416 /* extract the raw pointer to the record separator */
8417 rsptr = SvPV_const(PL_rs, rslen);
8421 /* rslast is the last character in the record separator
8422 * note we don't use rslast except when rslen is true, so the
8423 * null assign is a placeholder. */
8424 rslast = rslen ? rsptr[rslen - 1] : '\0';
8426 if (rspara) { /* have to do this both before and after */
8427 do { /* to make sure file boundaries work right */
8430 i = PerlIO_getc(fp);
8434 PerlIO_ungetc(fp,i);
8440 /* See if we know enough about I/O mechanism to cheat it ! */
8442 /* This used to be #ifdef test - it is made run-time test for ease
8443 of abstracting out stdio interface. One call should be cheap
8444 enough here - and may even be a macro allowing compile
8448 if (PerlIO_fast_gets(fp)) {
8450 * We can do buffer based IO operations on this filehandle.
8452 * This means we can bypass a lot of subcalls and process
8453 * the buffer directly, it also means we know the upper bound
8454 * on the amount of data we might read of the current buffer
8455 * into our sv. Knowing this allows us to preallocate the pv
8456 * to be able to hold that maximum, which allows us to simplify
8457 * a lot of logic. */
8460 * We're going to steal some values from the stdio struct
8461 * and put EVERYTHING in the innermost loop into registers.
8463 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8464 STRLEN bpx; /* length of the data in the target sv
8465 used to fix pointers after a SvGROW */
8466 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8467 of data left in the read-ahead buffer.
8468 If 0 then the pv buffer can hold the full
8469 amount left, otherwise this is the amount it
8472 /* Here is some breathtakingly efficient cheating */
8474 /* When you read the following logic resist the urge to think
8475 * of record separators that are 1 byte long. They are an
8476 * uninteresting special (simple) case.
8478 * Instead think of record separators which are at least 2 bytes
8479 * long, and keep in mind that we need to deal with such
8480 * separators when they cross a read-ahead buffer boundary.
8482 * Also consider that we need to gracefully deal with separators
8483 * that may be longer than a single read ahead buffer.
8485 * Lastly do not forget we want to copy the delimiter as well. We
8486 * are copying all data in the file _up_to_and_including_ the separator
8489 * Now that you have all that in mind here is what is happening below:
8491 * 1. When we first enter the loop we do some memory book keeping to see
8492 * how much free space there is in the target SV. (This sub assumes that
8493 * it is operating on the same SV most of the time via $_ and that it is
8494 * going to be able to reuse the same pv buffer each call.) If there is
8495 * "enough" room then we set "shortbuffered" to how much space there is
8496 * and start reading forward.
8498 * 2. When we scan forward we copy from the read-ahead buffer to the target
8499 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8500 * and the end of the of pv, as well as for the "rslast", which is the last
8501 * char of the separator.
8503 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8504 * (which has a "complete" record up to the point we saw rslast) and check
8505 * it to see if it matches the separator. If it does we are done. If it doesn't
8506 * we continue on with the scan/copy.
8508 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8509 * the IO system to read the next buffer. We do this by doing a getc(), which
8510 * returns a single char read (or EOF), and prefills the buffer, and also
8511 * allows us to find out how full the buffer is. We use this information to
8512 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8513 * the returned single char into the target sv, and then go back into scan
8516 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8517 * remaining space in the read-buffer.
8519 * Note that this code despite its twisty-turny nature is pretty darn slick.
8520 * It manages single byte separators, multi-byte cross boundary separators,
8521 * and cross-read-buffer separators cleanly and efficiently at the cost
8522 * of potentially greatly overallocating the target SV.
8528 /* get the number of bytes remaining in the read-ahead buffer
8529 * on first call on a given fp this will return 0.*/
8530 cnt = PerlIO_get_cnt(fp);
8532 /* make sure we have the room */
8533 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8534 /* Not room for all of it
8535 if we are looking for a separator and room for some
8537 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8538 /* just process what we have room for */
8539 shortbuffered = cnt - SvLEN(sv) + append + 1;
8540 cnt -= shortbuffered;
8543 /* ensure that the target sv has enough room to hold
8544 * the rest of the read-ahead buffer */
8546 /* remember that cnt can be negative */
8547 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8551 /* we have enough room to hold the full buffer, lets scream */
8555 /* extract the pointer to sv's string buffer, offset by append as necessary */
8556 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8557 /* extract the point to the read-ahead buffer */
8558 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8560 /* some trace debug output */
8561 DEBUG_P(PerlIO_printf(Perl_debug_log,
8562 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8563 DEBUG_P(PerlIO_printf(Perl_debug_log,
8564 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8566 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8567 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8571 /* if there is stuff left in the read-ahead buffer */
8573 /* if there is a separator */
8575 /* loop until we hit the end of the read-ahead buffer */
8576 while (cnt > 0) { /* this | eat */
8577 /* scan forward copying and searching for rslast as we go */
8579 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8580 goto thats_all_folks; /* screams | sed :-) */
8584 /* no separator, slurp the full buffer */
8585 Copy(ptr, bp, cnt, char); /* this | eat */
8586 bp += cnt; /* screams | dust */
8587 ptr += cnt; /* louder | sed :-) */
8589 assert (!shortbuffered);
8590 goto cannot_be_shortbuffered;
8594 if (shortbuffered) { /* oh well, must extend */
8595 /* we didnt have enough room to fit the line into the target buffer
8596 * so we must extend the target buffer and keep going */
8597 cnt = shortbuffered;
8599 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8601 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8602 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8603 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8607 cannot_be_shortbuffered:
8608 /* we need to refill the read-ahead buffer if possible */
8610 DEBUG_P(PerlIO_printf(Perl_debug_log,
8611 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8612 PTR2UV(ptr),(IV)cnt));
8613 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8615 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8616 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8617 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8618 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8621 call PerlIO_getc() to let it prefill the lookahead buffer
8623 This used to call 'filbuf' in stdio form, but as that behaves like
8624 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8625 another abstraction.
8627 Note we have to deal with the char in 'i' if we are not at EOF
8629 i = PerlIO_getc(fp); /* get more characters */
8631 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8632 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8633 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8634 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8636 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8637 cnt = PerlIO_get_cnt(fp);
8638 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8639 DEBUG_P(PerlIO_printf(Perl_debug_log,
8640 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8641 PTR2UV(ptr),(IV)cnt));
8643 if (i == EOF) /* all done for ever? */
8644 goto thats_really_all_folks;
8646 /* make sure we have enough space in the target sv */
8647 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8649 SvGROW(sv, bpx + cnt + 2);
8650 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8652 /* copy of the char we got from getc() */
8653 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8655 /* make sure we deal with the i being the last character of a separator */
8656 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8657 goto thats_all_folks;
8661 /* check if we have actually found the separator - only really applies
8663 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8664 memNE((char*)bp - rslen, rsptr, rslen))
8665 goto screamer; /* go back to the fray */
8666 thats_really_all_folks:
8668 cnt += shortbuffered;
8669 DEBUG_P(PerlIO_printf(Perl_debug_log,
8670 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8671 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8672 DEBUG_P(PerlIO_printf(Perl_debug_log,
8673 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8675 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8676 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8678 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8679 DEBUG_P(PerlIO_printf(Perl_debug_log,
8680 "Screamer: done, len=%ld, string=|%.*s|\n",
8681 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8685 /*The big, slow, and stupid way. */
8686 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8687 STDCHAR *buf = NULL;
8688 Newx(buf, 8192, STDCHAR);
8696 const STDCHAR * const bpe = buf + sizeof(buf);
8698 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8699 ; /* keep reading */
8703 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8704 /* Accommodate broken VAXC compiler, which applies U8 cast to
8705 * both args of ?: operator, causing EOF to change into 255
8708 i = (U8)buf[cnt - 1];
8714 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8716 sv_catpvn_nomg(sv, (char *) buf, cnt);
8718 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8720 if (i != EOF && /* joy */
8722 SvCUR(sv) < rslen ||
8723 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8727 * If we're reading from a TTY and we get a short read,
8728 * indicating that the user hit his EOF character, we need
8729 * to notice it now, because if we try to read from the TTY
8730 * again, the EOF condition will disappear.
8732 * The comparison of cnt to sizeof(buf) is an optimization
8733 * that prevents unnecessary calls to feof().
8737 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8741 #ifdef USE_HEAP_INSTEAD_OF_STACK
8746 if (rspara) { /* have to do this both before and after */
8747 while (i != EOF) { /* to make sure file boundaries work right */
8748 i = PerlIO_getc(fp);
8750 PerlIO_ungetc(fp,i);
8756 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8762 Auto-increment of the value in the SV, doing string to numeric conversion
8763 if necessary. Handles 'get' magic and operator overloading.
8769 Perl_sv_inc(pTHX_ SV *const sv)
8778 =for apidoc sv_inc_nomg
8780 Auto-increment of the value in the SV, doing string to numeric conversion
8781 if necessary. Handles operator overloading. Skips handling 'get' magic.
8787 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8794 if (SvTHINKFIRST(sv)) {
8795 if (SvREADONLY(sv)) {
8796 Perl_croak_no_modify();
8800 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8802 i = PTR2IV(SvRV(sv));
8806 else sv_force_normal_flags(sv, 0);
8808 flags = SvFLAGS(sv);
8809 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8810 /* It's (privately or publicly) a float, but not tested as an
8811 integer, so test it to see. */
8813 flags = SvFLAGS(sv);
8815 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8816 /* It's publicly an integer, or privately an integer-not-float */
8817 #ifdef PERL_PRESERVE_IVUV
8821 if (SvUVX(sv) == UV_MAX)
8822 sv_setnv(sv, UV_MAX_P1);
8824 (void)SvIOK_only_UV(sv);
8825 SvUV_set(sv, SvUVX(sv) + 1);
8827 if (SvIVX(sv) == IV_MAX)
8828 sv_setuv(sv, (UV)IV_MAX + 1);
8830 (void)SvIOK_only(sv);
8831 SvIV_set(sv, SvIVX(sv) + 1);
8836 if (flags & SVp_NOK) {
8837 const NV was = SvNVX(sv);
8838 if (LIKELY(!Perl_isinfnan(was)) &&
8839 NV_OVERFLOWS_INTEGERS_AT &&
8840 was >= NV_OVERFLOWS_INTEGERS_AT) {
8841 /* diag_listed_as: Lost precision when %s %f by 1 */
8842 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8843 "Lost precision when incrementing %" NVff " by 1",
8846 (void)SvNOK_only(sv);
8847 SvNV_set(sv, was + 1.0);
8851 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8852 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8853 Perl_croak_no_modify();
8855 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8856 if ((flags & SVTYPEMASK) < SVt_PVIV)
8857 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8858 (void)SvIOK_only(sv);
8863 while (isALPHA(*d)) d++;
8864 while (isDIGIT(*d)) d++;
8865 if (d < SvEND(sv)) {
8866 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8867 #ifdef PERL_PRESERVE_IVUV
8868 /* Got to punt this as an integer if needs be, but we don't issue
8869 warnings. Probably ought to make the sv_iv_please() that does
8870 the conversion if possible, and silently. */
8871 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8872 /* Need to try really hard to see if it's an integer.
8873 9.22337203685478e+18 is an integer.
8874 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8875 so $a="9.22337203685478e+18"; $a+0; $a++
8876 needs to be the same as $a="9.22337203685478e+18"; $a++
8883 /* sv_2iv *should* have made this an NV */
8884 if (flags & SVp_NOK) {
8885 (void)SvNOK_only(sv);
8886 SvNV_set(sv, SvNVX(sv) + 1.0);
8889 /* I don't think we can get here. Maybe I should assert this
8890 And if we do get here I suspect that sv_setnv will croak. NWC
8892 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8893 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8895 #endif /* PERL_PRESERVE_IVUV */
8896 if (!numtype && ckWARN(WARN_NUMERIC))
8897 not_incrementable(sv);
8898 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8902 while (d >= SvPVX_const(sv)) {
8910 /* MKS: The original code here died if letters weren't consecutive.
8911 * at least it didn't have to worry about non-C locales. The
8912 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8913 * arranged in order (although not consecutively) and that only
8914 * [A-Za-z] are accepted by isALPHA in the C locale.
8916 if (isALPHA_FOLD_NE(*d, 'z')) {
8917 do { ++*d; } while (!isALPHA(*d));
8920 *(d--) -= 'z' - 'a';
8925 *(d--) -= 'z' - 'a' + 1;
8929 /* oh,oh, the number grew */
8930 SvGROW(sv, SvCUR(sv) + 2);
8931 SvCUR_set(sv, SvCUR(sv) + 1);
8932 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8943 Auto-decrement of the value in the SV, doing string to numeric conversion
8944 if necessary. Handles 'get' magic and operator overloading.
8950 Perl_sv_dec(pTHX_ SV *const sv)
8959 =for apidoc sv_dec_nomg
8961 Auto-decrement of the value in the SV, doing string to numeric conversion
8962 if necessary. Handles operator overloading. Skips handling 'get' magic.
8968 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8974 if (SvTHINKFIRST(sv)) {
8975 if (SvREADONLY(sv)) {
8976 Perl_croak_no_modify();
8980 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8982 i = PTR2IV(SvRV(sv));
8986 else sv_force_normal_flags(sv, 0);
8988 /* Unlike sv_inc we don't have to worry about string-never-numbers
8989 and keeping them magic. But we mustn't warn on punting */
8990 flags = SvFLAGS(sv);
8991 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8992 /* It's publicly an integer, or privately an integer-not-float */
8993 #ifdef PERL_PRESERVE_IVUV
8997 if (SvUVX(sv) == 0) {
8998 (void)SvIOK_only(sv);
9002 (void)SvIOK_only_UV(sv);
9003 SvUV_set(sv, SvUVX(sv) - 1);
9006 if (SvIVX(sv) == IV_MIN) {
9007 sv_setnv(sv, (NV)IV_MIN);
9011 (void)SvIOK_only(sv);
9012 SvIV_set(sv, SvIVX(sv) - 1);
9017 if (flags & SVp_NOK) {
9020 const NV was = SvNVX(sv);
9021 if (LIKELY(!Perl_isinfnan(was)) &&
9022 NV_OVERFLOWS_INTEGERS_AT &&
9023 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9024 /* diag_listed_as: Lost precision when %s %f by 1 */
9025 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9026 "Lost precision when decrementing %" NVff " by 1",
9029 (void)SvNOK_only(sv);
9030 SvNV_set(sv, was - 1.0);
9035 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9036 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9037 Perl_croak_no_modify();
9039 if (!(flags & SVp_POK)) {
9040 if ((flags & SVTYPEMASK) < SVt_PVIV)
9041 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9043 (void)SvIOK_only(sv);
9046 #ifdef PERL_PRESERVE_IVUV
9048 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9049 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9050 /* Need to try really hard to see if it's an integer.
9051 9.22337203685478e+18 is an integer.
9052 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9053 so $a="9.22337203685478e+18"; $a+0; $a--
9054 needs to be the same as $a="9.22337203685478e+18"; $a--
9061 /* sv_2iv *should* have made this an NV */
9062 if (flags & SVp_NOK) {
9063 (void)SvNOK_only(sv);
9064 SvNV_set(sv, SvNVX(sv) - 1.0);
9067 /* I don't think we can get here. Maybe I should assert this
9068 And if we do get here I suspect that sv_setnv will croak. NWC
9070 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9071 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9074 #endif /* PERL_PRESERVE_IVUV */
9075 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9078 /* this define is used to eliminate a chunk of duplicated but shared logic
9079 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9080 * used anywhere but here - yves
9082 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9084 SSize_t ix = ++PL_tmps_ix; \
9085 if (UNLIKELY(ix >= PL_tmps_max)) \
9086 ix = tmps_grow_p(ix); \
9087 PL_tmps_stack[ix] = (AnSv); \
9091 =for apidoc sv_mortalcopy
9093 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9094 The new SV is marked as mortal. It will be destroyed "soon", either by an
9095 explicit call to C<FREETMPS>, or by an implicit call at places such as
9096 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9101 /* Make a string that will exist for the duration of the expression
9102 * evaluation. Actually, it may have to last longer than that, but
9103 * hopefully we won't free it until it has been assigned to a
9104 * permanent location. */
9107 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9111 if (flags & SV_GMAGIC)
9112 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9114 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9115 PUSH_EXTEND_MORTAL__SV_C(sv);
9121 =for apidoc sv_newmortal
9123 Creates a new null SV which is mortal. The reference count of the SV is
9124 set to 1. It will be destroyed "soon", either by an explicit call to
9125 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9126 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9132 Perl_sv_newmortal(pTHX)
9137 SvFLAGS(sv) = SVs_TEMP;
9138 PUSH_EXTEND_MORTAL__SV_C(sv);
9144 =for apidoc newSVpvn_flags
9146 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9147 characters) into it. The reference count for the
9148 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9149 string. You are responsible for ensuring that the source string is at least
9150 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9151 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9152 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9153 returning. If C<SVf_UTF8> is set, C<s>
9154 is considered to be in UTF-8 and the
9155 C<SVf_UTF8> flag will be set on the new SV.
9156 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9158 #define newSVpvn_utf8(s, len, u) \
9159 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9165 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9169 /* All the flags we don't support must be zero.
9170 And we're new code so I'm going to assert this from the start. */
9171 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9173 sv_setpvn(sv,s,len);
9175 /* This code used to do a sv_2mortal(), however we now unroll the call to
9176 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9177 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9178 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9179 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9180 * means that we eliminate quite a few steps than it looks - Yves
9181 * (explaining patch by gfx) */
9183 SvFLAGS(sv) |= flags;
9185 if(flags & SVs_TEMP){
9186 PUSH_EXTEND_MORTAL__SV_C(sv);
9193 =for apidoc sv_2mortal
9195 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9196 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9197 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9198 string buffer can be "stolen" if this SV is copied. See also
9199 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9205 Perl_sv_2mortal(pTHX_ SV *const sv)
9212 PUSH_EXTEND_MORTAL__SV_C(sv);
9220 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9221 characters) into it. The reference count for the
9222 SV is set to 1. If C<len> is zero, Perl will compute the length using
9223 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9224 C<NUL> characters and has to have a terminating C<NUL> byte).
9226 For efficiency, consider using C<newSVpvn> instead.
9232 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9237 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9242 =for apidoc newSVpvn
9244 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9245 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9246 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9247 are responsible for ensuring that the source buffer is at least
9248 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9255 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9259 sv_setpvn(sv,buffer,len);
9264 =for apidoc newSVhek
9266 Creates a new SV from the hash key structure. It will generate scalars that
9267 point to the shared string table where possible. Returns a new (undefined)
9268 SV if C<hek> is NULL.
9274 Perl_newSVhek(pTHX_ const HEK *const hek)
9283 if (HEK_LEN(hek) == HEf_SVKEY) {
9284 return newSVsv(*(SV**)HEK_KEY(hek));
9286 const int flags = HEK_FLAGS(hek);
9287 if (flags & HVhek_WASUTF8) {
9289 Andreas would like keys he put in as utf8 to come back as utf8
9291 STRLEN utf8_len = HEK_LEN(hek);
9292 SV * const sv = newSV_type(SVt_PV);
9293 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9294 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9295 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9298 } else if (flags & HVhek_UNSHARED) {
9299 /* A hash that isn't using shared hash keys has to have
9300 the flag in every key so that we know not to try to call
9301 share_hek_hek on it. */
9303 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9308 /* This will be overwhelminly the most common case. */
9310 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9311 more efficient than sharepvn(). */
9315 sv_upgrade(sv, SVt_PV);
9316 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9317 SvCUR_set(sv, HEK_LEN(hek));
9329 =for apidoc newSVpvn_share
9331 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9332 table. If the string does not already exist in the table, it is
9333 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9334 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9335 is non-zero, that value is used; otherwise the hash is computed.
9336 The string's hash can later be retrieved from the SV
9337 with the C<SvSHARED_HASH()> macro. The idea here is
9338 that as the string table is used for shared hash keys these strings will have
9339 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9345 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9349 bool is_utf8 = FALSE;
9350 const char *const orig_src = src;
9353 STRLEN tmplen = -len;
9355 /* See the note in hv.c:hv_fetch() --jhi */
9356 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9360 PERL_HASH(hash, src, len);
9362 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9363 changes here, update it there too. */
9364 sv_upgrade(sv, SVt_PV);
9365 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9372 if (src != orig_src)
9378 =for apidoc newSVpv_share
9380 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9387 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9389 return newSVpvn_share(src, strlen(src), hash);
9392 #if defined(PERL_IMPLICIT_CONTEXT)
9394 /* pTHX_ magic can't cope with varargs, so this is a no-context
9395 * version of the main function, (which may itself be aliased to us).
9396 * Don't access this version directly.
9400 Perl_newSVpvf_nocontext(const char *const pat, ...)
9406 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9408 va_start(args, pat);
9409 sv = vnewSVpvf(pat, &args);
9416 =for apidoc newSVpvf
9418 Creates a new SV and initializes it with the string formatted like
9425 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9430 PERL_ARGS_ASSERT_NEWSVPVF;
9432 va_start(args, pat);
9433 sv = vnewSVpvf(pat, &args);
9438 /* backend for newSVpvf() and newSVpvf_nocontext() */
9441 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9445 PERL_ARGS_ASSERT_VNEWSVPVF;
9448 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9455 Creates a new SV and copies a floating point value into it.
9456 The reference count for the SV is set to 1.
9462 Perl_newSVnv(pTHX_ const NV n)
9474 Creates a new SV and copies an integer into it. The reference count for the
9481 Perl_newSViv(pTHX_ const IV i)
9487 /* Inlining ONLY the small relevant subset of sv_setiv here
9488 * for performance. Makes a significant difference. */
9490 /* We're starting from SVt_FIRST, so provided that's
9491 * actual 0, we don't have to unset any SV type flags
9492 * to promote to SVt_IV. */
9493 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9495 SET_SVANY_FOR_BODYLESS_IV(sv);
9496 SvFLAGS(sv) |= SVt_IV;
9508 Creates a new SV and copies an unsigned integer into it.
9509 The reference count for the SV is set to 1.
9515 Perl_newSVuv(pTHX_ const UV u)
9519 /* Inlining ONLY the small relevant subset of sv_setuv here
9520 * for performance. Makes a significant difference. */
9522 /* Using ivs is more efficient than using uvs - see sv_setuv */
9523 if (u <= (UV)IV_MAX) {
9524 return newSViv((IV)u);
9529 /* We're starting from SVt_FIRST, so provided that's
9530 * actual 0, we don't have to unset any SV type flags
9531 * to promote to SVt_IV. */
9532 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9534 SET_SVANY_FOR_BODYLESS_IV(sv);
9535 SvFLAGS(sv) |= SVt_IV;
9537 (void)SvIsUV_on(sv);
9546 =for apidoc newSV_type
9548 Creates a new SV, of the type specified. The reference count for the new SV
9555 Perl_newSV_type(pTHX_ const svtype type)
9560 ASSUME(SvTYPE(sv) == SVt_FIRST);
9561 if(type != SVt_FIRST)
9562 sv_upgrade(sv, type);
9567 =for apidoc newRV_noinc
9569 Creates an RV wrapper for an SV. The reference count for the original
9570 SV is B<not> incremented.
9576 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9580 PERL_ARGS_ASSERT_NEWRV_NOINC;
9584 /* We're starting from SVt_FIRST, so provided that's
9585 * actual 0, we don't have to unset any SV type flags
9586 * to promote to SVt_IV. */
9587 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9589 SET_SVANY_FOR_BODYLESS_IV(sv);
9590 SvFLAGS(sv) |= SVt_IV;
9595 SvRV_set(sv, tmpRef);
9600 /* newRV_inc is the official function name to use now.
9601 * newRV_inc is in fact #defined to newRV in sv.h
9605 Perl_newRV(pTHX_ SV *const sv)
9607 PERL_ARGS_ASSERT_NEWRV;
9609 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9615 Creates a new SV which is an exact duplicate of the original SV.
9622 Perl_newSVsv(pTHX_ SV *const old)
9628 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9629 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9632 /* Do this here, otherwise we leak the new SV if this croaks. */
9635 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9636 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9637 sv_setsv_flags(sv, old, SV_NOSTEAL);
9642 =for apidoc sv_reset
9644 Underlying implementation for the C<reset> Perl function.
9645 Note that the perl-level function is vaguely deprecated.
9651 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9653 PERL_ARGS_ASSERT_SV_RESET;
9655 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9659 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9661 char todo[PERL_UCHAR_MAX+1];
9664 if (!stash || SvTYPE(stash) != SVt_PVHV)
9667 if (!s) { /* reset ?? searches */
9668 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9670 const U32 count = mg->mg_len / sizeof(PMOP**);
9671 PMOP **pmp = (PMOP**) mg->mg_ptr;
9672 PMOP *const *const end = pmp + count;
9676 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9678 (*pmp)->op_pmflags &= ~PMf_USED;
9686 /* reset variables */
9688 if (!HvARRAY(stash))
9691 Zero(todo, 256, char);
9695 I32 i = (unsigned char)*s;
9699 max = (unsigned char)*s++;
9700 for ( ; i <= max; i++) {
9703 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9705 for (entry = HvARRAY(stash)[i];
9707 entry = HeNEXT(entry))
9712 if (!todo[(U8)*HeKEY(entry)])
9714 gv = MUTABLE_GV(HeVAL(entry));
9718 if (sv && !SvREADONLY(sv)) {
9719 SV_CHECK_THINKFIRST_COW_DROP(sv);
9720 if (!isGV(sv)) SvOK_off(sv);
9725 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9736 Using various gambits, try to get an IO from an SV: the IO slot if its a
9737 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9738 named after the PV if we're a string.
9740 'Get' magic is ignored on the C<sv> passed in, but will be called on
9741 C<SvRV(sv)> if C<sv> is an RV.
9747 Perl_sv_2io(pTHX_ SV *const sv)
9752 PERL_ARGS_ASSERT_SV_2IO;
9754 switch (SvTYPE(sv)) {
9756 io = MUTABLE_IO(sv);
9760 if (isGV_with_GP(sv)) {
9761 gv = MUTABLE_GV(sv);
9764 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9765 HEKfARG(GvNAME_HEK(gv)));
9771 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9773 SvGETMAGIC(SvRV(sv));
9774 return sv_2io(SvRV(sv));
9776 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9783 if (SvGMAGICAL(sv)) {
9784 newsv = sv_newmortal();
9785 sv_setsv_nomg(newsv, sv);
9787 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9797 Using various gambits, try to get a CV from an SV; in addition, try if
9798 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9799 The flags in C<lref> are passed to C<gv_fetchsv>.
9805 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9810 PERL_ARGS_ASSERT_SV_2CV;
9817 switch (SvTYPE(sv)) {
9821 return MUTABLE_CV(sv);
9831 sv = amagic_deref_call(sv, to_cv_amg);
9834 if (SvTYPE(sv) == SVt_PVCV) {
9835 cv = MUTABLE_CV(sv);
9840 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9841 gv = MUTABLE_GV(sv);
9843 Perl_croak(aTHX_ "Not a subroutine reference");
9845 else if (isGV_with_GP(sv)) {
9846 gv = MUTABLE_GV(sv);
9849 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9856 /* Some flags to gv_fetchsv mean don't really create the GV */
9857 if (!isGV_with_GP(gv)) {
9862 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9863 /* XXX this is probably not what they think they're getting.
9864 * It has the same effect as "sub name;", i.e. just a forward
9875 Returns true if the SV has a true value by Perl's rules.
9876 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9877 instead use an in-line version.
9883 Perl_sv_true(pTHX_ SV *const sv)
9888 const XPV* const tXpv = (XPV*)SvANY(sv);
9890 (tXpv->xpv_cur > 1 ||
9891 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9898 return SvIVX(sv) != 0;
9901 return SvNVX(sv) != 0.0;
9903 return sv_2bool(sv);
9909 =for apidoc sv_pvn_force
9911 Get a sensible string out of the SV somehow.
9912 A private implementation of the C<SvPV_force> macro for compilers which
9913 can't cope with complex macro expressions. Always use the macro instead.
9915 =for apidoc sv_pvn_force_flags
9917 Get a sensible string out of the SV somehow.
9918 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9919 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9920 implemented in terms of this function.
9921 You normally want to use the various wrapper macros instead: see
9922 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9928 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9930 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9932 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9933 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9934 sv_force_normal_flags(sv, 0);
9944 if (SvTYPE(sv) > SVt_PVLV
9945 || isGV_with_GP(sv))
9946 /* diag_listed_as: Can't coerce %s to %s in %s */
9947 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9949 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9956 if (SvTYPE(sv) < SVt_PV ||
9957 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9960 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9961 SvGROW(sv, len + 1);
9962 Move(s,SvPVX(sv),len,char);
9964 SvPVX(sv)[len] = '\0';
9967 SvPOK_on(sv); /* validate pointer */
9969 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9970 PTR2UV(sv),SvPVX_const(sv)));
9973 (void)SvPOK_only_UTF8(sv);
9974 return SvPVX_mutable(sv);
9978 =for apidoc sv_pvbyten_force
9980 The backend for the C<SvPVbytex_force> macro. Always use the macro
9987 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9989 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9991 sv_pvn_force(sv,lp);
9992 sv_utf8_downgrade(sv,0);
9998 =for apidoc sv_pvutf8n_force
10000 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10007 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10009 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10011 sv_pvn_force(sv,0);
10012 sv_utf8_upgrade_nomg(sv);
10018 =for apidoc sv_reftype
10020 Returns a string describing what the SV is a reference to.
10022 If ob is true and the SV is blessed, the string is the class name,
10023 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10029 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10031 PERL_ARGS_ASSERT_SV_REFTYPE;
10032 if (ob && SvOBJECT(sv)) {
10033 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10036 /* WARNING - There is code, for instance in mg.c, that assumes that
10037 * the only reason that sv_reftype(sv,0) would return a string starting
10038 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10039 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10040 * this routine inside other subs, and it saves time.
10041 * Do not change this assumption without searching for "dodgy type check" in
10044 switch (SvTYPE(sv)) {
10059 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10060 /* tied lvalues should appear to be
10061 * scalars for backwards compatibility */
10062 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10063 ? "SCALAR" : "LVALUE");
10064 case SVt_PVAV: return "ARRAY";
10065 case SVt_PVHV: return "HASH";
10066 case SVt_PVCV: return "CODE";
10067 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10068 ? "GLOB" : "SCALAR");
10069 case SVt_PVFM: return "FORMAT";
10070 case SVt_PVIO: return "IO";
10071 case SVt_INVLIST: return "INVLIST";
10072 case SVt_REGEXP: return "REGEXP";
10073 default: return "UNKNOWN";
10081 Returns a SV describing what the SV passed in is a reference to.
10083 dst can be a SV to be set to the description or NULL, in which case a
10084 mortal SV is returned.
10086 If ob is true and the SV is blessed, the description is the class
10087 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10093 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10095 PERL_ARGS_ASSERT_SV_REF;
10098 dst = sv_newmortal();
10100 if (ob && SvOBJECT(sv)) {
10101 HvNAME_get(SvSTASH(sv))
10102 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10103 : sv_setpvn(dst, "__ANON__", 8);
10106 const char * reftype = sv_reftype(sv, 0);
10107 sv_setpv(dst, reftype);
10113 =for apidoc sv_isobject
10115 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10116 object. If the SV is not an RV, or if the object is not blessed, then this
10123 Perl_sv_isobject(pTHX_ SV *sv)
10139 Returns a boolean indicating whether the SV is blessed into the specified
10140 class. This does not check for subtypes; use C<sv_derived_from> to verify
10141 an inheritance relationship.
10147 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10149 const char *hvname;
10151 PERL_ARGS_ASSERT_SV_ISA;
10161 hvname = HvNAME_get(SvSTASH(sv));
10165 return strEQ(hvname, name);
10169 =for apidoc newSVrv
10171 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10172 RV then it will be upgraded to one. If C<classname> is non-null then the new
10173 SV will be blessed in the specified package. The new SV is returned and its
10174 reference count is 1. The reference count 1 is owned by C<rv>.
10180 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10184 PERL_ARGS_ASSERT_NEWSVRV;
10188 SV_CHECK_THINKFIRST_COW_DROP(rv);
10190 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10191 const U32 refcnt = SvREFCNT(rv);
10195 SvREFCNT(rv) = refcnt;
10197 sv_upgrade(rv, SVt_IV);
10198 } else if (SvROK(rv)) {
10199 SvREFCNT_dec(SvRV(rv));
10201 prepare_SV_for_RV(rv);
10209 HV* const stash = gv_stashpv(classname, GV_ADD);
10210 (void)sv_bless(rv, stash);
10216 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10218 SV * const lv = newSV_type(SVt_PVLV);
10219 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10221 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10222 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10223 LvSTARGOFF(lv) = ix;
10224 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10229 =for apidoc sv_setref_pv
10231 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10232 argument will be upgraded to an RV. That RV will be modified to point to
10233 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10234 into the SV. The C<classname> argument indicates the package for the
10235 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10236 will have a reference count of 1, and the RV will be returned.
10238 Do not use with other Perl types such as HV, AV, SV, CV, because those
10239 objects will become corrupted by the pointer copy process.
10241 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10247 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10249 PERL_ARGS_ASSERT_SV_SETREF_PV;
10252 sv_setsv(rv, &PL_sv_undef);
10256 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10261 =for apidoc sv_setref_iv
10263 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10264 argument will be upgraded to an RV. That RV will be modified to point to
10265 the new SV. The C<classname> argument indicates the package for the
10266 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10267 will have a reference count of 1, and the RV will be returned.
10273 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10275 PERL_ARGS_ASSERT_SV_SETREF_IV;
10277 sv_setiv(newSVrv(rv,classname), iv);
10282 =for apidoc sv_setref_uv
10284 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10285 argument will be upgraded to an RV. That RV will be modified to point to
10286 the new SV. The C<classname> argument indicates the package for the
10287 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10288 will have a reference count of 1, and the RV will be returned.
10294 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10296 PERL_ARGS_ASSERT_SV_SETREF_UV;
10298 sv_setuv(newSVrv(rv,classname), uv);
10303 =for apidoc sv_setref_nv
10305 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10306 argument will be upgraded to an RV. That RV will be modified to point to
10307 the new SV. The C<classname> argument indicates the package for the
10308 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10309 will have a reference count of 1, and the RV will be returned.
10315 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10317 PERL_ARGS_ASSERT_SV_SETREF_NV;
10319 sv_setnv(newSVrv(rv,classname), nv);
10324 =for apidoc sv_setref_pvn
10326 Copies a string into a new SV, optionally blessing the SV. The length of the
10327 string must be specified with C<n>. The C<rv> argument will be upgraded to
10328 an RV. That RV will be modified to point to the new SV. The C<classname>
10329 argument indicates the package for the blessing. Set C<classname> to
10330 C<NULL> to avoid the blessing. The new SV will have a reference count
10331 of 1, and the RV will be returned.
10333 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10339 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10340 const char *const pv, const STRLEN n)
10342 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10344 sv_setpvn(newSVrv(rv,classname), pv, n);
10349 =for apidoc sv_bless
10351 Blesses an SV into a specified package. The SV must be an RV. The package
10352 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10353 of the SV is unaffected.
10359 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10362 HV *oldstash = NULL;
10364 PERL_ARGS_ASSERT_SV_BLESS;
10368 Perl_croak(aTHX_ "Can't bless non-reference value");
10370 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10371 if (SvREADONLY(tmpRef))
10372 Perl_croak_no_modify();
10373 if (SvOBJECT(tmpRef)) {
10374 oldstash = SvSTASH(tmpRef);
10377 SvOBJECT_on(tmpRef);
10378 SvUPGRADE(tmpRef, SVt_PVMG);
10379 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10380 SvREFCNT_dec(oldstash);
10382 if(SvSMAGICAL(tmpRef))
10383 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10391 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10392 * as it is after unglobbing it.
10395 PERL_STATIC_INLINE void
10396 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10400 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10402 PERL_ARGS_ASSERT_SV_UNGLOB;
10404 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10406 if (!(flags & SV_COW_DROP_PV))
10407 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10409 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10411 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10412 && HvNAME_get(stash))
10413 mro_method_changed_in(stash);
10414 gp_free(MUTABLE_GV(sv));
10417 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10418 GvSTASH(sv) = NULL;
10421 if (GvNAME_HEK(sv)) {
10422 unshare_hek(GvNAME_HEK(sv));
10424 isGV_with_GP_off(sv);
10426 if(SvTYPE(sv) == SVt_PVGV) {
10427 /* need to keep SvANY(sv) in the right arena */
10428 xpvmg = new_XPVMG();
10429 StructCopy(SvANY(sv), xpvmg, XPVMG);
10430 del_XPVGV(SvANY(sv));
10433 SvFLAGS(sv) &= ~SVTYPEMASK;
10434 SvFLAGS(sv) |= SVt_PVMG;
10437 /* Intentionally not calling any local SET magic, as this isn't so much a
10438 set operation as merely an internal storage change. */
10439 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10440 else sv_setsv_flags(sv, temp, 0);
10442 if ((const GV *)sv == PL_last_in_gv)
10443 PL_last_in_gv = NULL;
10444 else if ((const GV *)sv == PL_statgv)
10449 =for apidoc sv_unref_flags
10451 Unsets the RV status of the SV, and decrements the reference count of
10452 whatever was being referenced by the RV. This can almost be thought of
10453 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10454 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10455 (otherwise the decrementing is conditional on the reference count being
10456 different from one or the reference being a readonly SV).
10457 See C<L</SvROK_off>>.
10463 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10465 SV* const target = SvRV(ref);
10467 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10469 if (SvWEAKREF(ref)) {
10470 sv_del_backref(target, ref);
10471 SvWEAKREF_off(ref);
10472 SvRV_set(ref, NULL);
10475 SvRV_set(ref, NULL);
10477 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10478 assigned to as BEGIN {$a = \"Foo"} will fail. */
10479 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10480 SvREFCNT_dec_NN(target);
10481 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10482 sv_2mortal(target); /* Schedule for freeing later */
10486 =for apidoc sv_untaint
10488 Untaint an SV. Use C<SvTAINTED_off> instead.
10494 Perl_sv_untaint(pTHX_ SV *const sv)
10496 PERL_ARGS_ASSERT_SV_UNTAINT;
10497 PERL_UNUSED_CONTEXT;
10499 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10500 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10507 =for apidoc sv_tainted
10509 Test an SV for taintedness. Use C<SvTAINTED> instead.
10515 Perl_sv_tainted(pTHX_ SV *const sv)
10517 PERL_ARGS_ASSERT_SV_TAINTED;
10518 PERL_UNUSED_CONTEXT;
10520 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10521 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10522 if (mg && (mg->mg_len & 1) )
10528 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10529 private to this file */
10532 =for apidoc sv_setpviv
10534 Copies an integer into the given SV, also updating its string value.
10535 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10541 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10543 char buf[TYPE_CHARS(UV)];
10545 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10547 PERL_ARGS_ASSERT_SV_SETPVIV;
10549 sv_setpvn(sv, ptr, ebuf - ptr);
10553 =for apidoc sv_setpviv_mg
10555 Like C<sv_setpviv>, but also handles 'set' magic.
10561 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10563 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10565 sv_setpviv(sv, iv);
10569 #endif /* NO_MATHOMS */
10571 #if defined(PERL_IMPLICIT_CONTEXT)
10573 /* pTHX_ magic can't cope with varargs, so this is a no-context
10574 * version of the main function, (which may itself be aliased to us).
10575 * Don't access this version directly.
10579 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10584 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10586 va_start(args, pat);
10587 sv_vsetpvf(sv, pat, &args);
10591 /* pTHX_ magic can't cope with varargs, so this is a no-context
10592 * version of the main function, (which may itself be aliased to us).
10593 * Don't access this version directly.
10597 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10602 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10604 va_start(args, pat);
10605 sv_vsetpvf_mg(sv, pat, &args);
10611 =for apidoc sv_setpvf
10613 Works like C<sv_catpvf> but copies the text into the SV instead of
10614 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10620 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10624 PERL_ARGS_ASSERT_SV_SETPVF;
10626 va_start(args, pat);
10627 sv_vsetpvf(sv, pat, &args);
10632 =for apidoc sv_vsetpvf
10634 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10635 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10637 Usually used via its frontend C<sv_setpvf>.
10643 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10645 PERL_ARGS_ASSERT_SV_VSETPVF;
10647 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10651 =for apidoc sv_setpvf_mg
10653 Like C<sv_setpvf>, but also handles 'set' magic.
10659 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10663 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10665 va_start(args, pat);
10666 sv_vsetpvf_mg(sv, pat, &args);
10671 =for apidoc sv_vsetpvf_mg
10673 Like C<sv_vsetpvf>, but also handles 'set' magic.
10675 Usually used via its frontend C<sv_setpvf_mg>.
10681 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10683 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10685 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10689 #if defined(PERL_IMPLICIT_CONTEXT)
10691 /* pTHX_ magic can't cope with varargs, so this is a no-context
10692 * version of the main function, (which may itself be aliased to us).
10693 * Don't access this version directly.
10697 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10702 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10704 va_start(args, pat);
10705 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10709 /* pTHX_ magic can't cope with varargs, so this is a no-context
10710 * version of the main function, (which may itself be aliased to us).
10711 * Don't access this version directly.
10715 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10720 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10722 va_start(args, pat);
10723 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10730 =for apidoc sv_catpvf
10732 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10733 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10734 variable argument list, argument reordering is not supported.
10735 If the appended data contains "wide" characters
10736 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10737 and characters >255 formatted with C<%c>), the original SV might get
10738 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10739 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10740 valid UTF-8; if the original SV was bytes, the pattern should be too.
10745 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10749 PERL_ARGS_ASSERT_SV_CATPVF;
10751 va_start(args, pat);
10752 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10757 =for apidoc sv_vcatpvf
10759 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10760 variable argument list, and appends the formatted output
10761 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10763 Usually used via its frontend C<sv_catpvf>.
10769 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10771 PERL_ARGS_ASSERT_SV_VCATPVF;
10773 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10777 =for apidoc sv_catpvf_mg
10779 Like C<sv_catpvf>, but also handles 'set' magic.
10785 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10789 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10791 va_start(args, pat);
10792 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10798 =for apidoc sv_vcatpvf_mg
10800 Like C<sv_vcatpvf>, but also handles 'set' magic.
10802 Usually used via its frontend C<sv_catpvf_mg>.
10808 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10810 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10812 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10817 =for apidoc sv_vsetpvfn
10819 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10822 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10828 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10829 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10831 PERL_ARGS_ASSERT_SV_VSETPVFN;
10834 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10839 * Warn of missing argument to sprintf. The value used in place of such
10840 * arguments should be &PL_sv_no; an undefined value would yield
10841 * inappropriate "use of uninit" warnings [perl #71000].
10844 S_warn_vcatpvfn_missing_argument(pTHX) {
10845 if (ckWARN(WARN_MISSING)) {
10846 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10847 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10853 S_expect_number(pTHX_ char **const pattern)
10857 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10859 switch (**pattern) {
10860 case '1': case '2': case '3':
10861 case '4': case '5': case '6':
10862 case '7': case '8': case '9':
10863 var = *(*pattern)++ - '0';
10864 while (isDIGIT(**pattern)) {
10865 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10867 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10875 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10877 const int neg = nv < 0;
10880 PERL_ARGS_ASSERT_F0CONVERT;
10882 if (UNLIKELY(Perl_isinfnan(nv))) {
10883 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10893 if (uv & 1 && uv == nv)
10894 uv--; /* Round to even */
10896 const unsigned dig = uv % 10;
10898 } while (uv /= 10);
10909 =for apidoc sv_vcatpvfn
10911 =for apidoc sv_vcatpvfn_flags
10913 Processes its arguments like C<vsprintf> and appends the formatted output
10914 to an SV. Uses an array of SVs if the C-style variable argument list is
10915 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10916 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10917 C<va_list> argument list with a format string that uses argument reordering
10918 will yield an exception.
10920 When running with taint checks enabled, indicates via
10921 C<maybe_tainted> if results are untrustworthy (often due to the use of
10924 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10926 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10931 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10932 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10933 vec_utf8 = DO_UTF8(vecsv);
10935 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10938 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10939 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10941 PERL_ARGS_ASSERT_SV_VCATPVFN;
10943 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10946 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10947 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10948 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10949 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10950 * after the first 1023 zero bits.
10952 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10953 * of dynamically growing buffer might be better, start at just 16 bytes
10954 * (for example) and grow only when necessary. Or maybe just by looking
10955 * at the exponents of the two doubles? */
10956 # define DOUBLEDOUBLE_MAXBITS 2098
10959 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10960 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10961 * per xdigit. For the double-double case, this can be rather many.
10962 * The non-double-double-long-double overshoots since all bits of NV
10963 * are not mantissa bits, there are also exponent bits. */
10964 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10965 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10967 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10970 /* If we do not have a known long double format, (including not using
10971 * long doubles, or long doubles being equal to doubles) then we will
10972 * fall back to the ldexp/frexp route, with which we can retrieve at
10973 * most as many bits as our widest unsigned integer type is. We try
10974 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10976 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10977 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10979 #if defined(HAS_QUAD) && defined(Uquad_t)
10980 # define MANTISSATYPE Uquad_t
10981 # define MANTISSASIZE 8
10983 # define MANTISSATYPE UV
10984 # define MANTISSASIZE UVSIZE
10987 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10988 # define HEXTRACT_LITTLE_ENDIAN
10989 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10990 # define HEXTRACT_BIG_ENDIAN
10992 # define HEXTRACT_MIX_ENDIAN
10995 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10996 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10997 * are being extracted from (either directly from the long double in-memory
10998 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10999 * is used to update the exponent. vhex is the pointer to the beginning
11000 * of the output buffer (of VHEX_SIZE).
11002 * The tricky part is that S_hextract() needs to be called twice:
11003 * the first time with vend as NULL, and the second time with vend as
11004 * the pointer returned by the first call. What happens is that on
11005 * the first round the output size is computed, and the intended
11006 * extraction sanity checked. On the second round the actual output
11007 * (the extraction of the hexadecimal values) takes place.
11008 * Sanity failures cause fatal failures during both rounds. */
11010 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
11014 int ixmin = 0, ixmax = 0;
11016 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
11017 * and elsewhere. */
11019 /* These macros are just to reduce typos, they have multiple
11020 * repetitions below, but usually only one (or sometimes two)
11021 * of them is really being used. */
11022 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11023 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11024 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11025 #define HEXTRACT_OUTPUT(ix) \
11027 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11029 #define HEXTRACT_COUNT(ix, c) \
11031 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11033 #define HEXTRACT_BYTE(ix) \
11035 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11037 #define HEXTRACT_LO_NYBBLE(ix) \
11039 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11041 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11042 * to make it look less odd when the top bits of a NV
11043 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11044 * order bits can be in the "low nybble" of a byte. */
11045 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11046 #define HEXTRACT_BYTES_LE(a, b) \
11047 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11048 #define HEXTRACT_BYTES_BE(a, b) \
11049 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11050 #define HEXTRACT_IMPLICIT_BIT(nv) \
11052 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11055 /* Most formats do. Those which don't should undef this. */
11056 #define HEXTRACT_HAS_IMPLICIT_BIT
11057 /* Many formats do. Those which don't should undef this. */
11058 #define HEXTRACT_HAS_TOP_NYBBLE
11060 /* HEXTRACTSIZE is the maximum number of xdigits. */
11061 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11062 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11064 # define HEXTRACTSIZE 2 * NVSIZE
11067 const U8* vmaxend = vhex + HEXTRACTSIZE;
11068 PERL_UNUSED_VAR(ix); /* might happen */
11069 (void)Perl_frexp(PERL_ABS(nv), exponent);
11070 if (vend && (vend <= vhex || vend > vmaxend)) {
11071 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11072 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11075 /* First check if using long doubles. */
11076 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11077 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11078 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11079 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
11080 /* The bytes 13..0 are the mantissa/fraction,
11081 * the 15,14 are the sign+exponent. */
11082 const U8* nvp = (const U8*)(&nv);
11083 HEXTRACT_IMPLICIT_BIT(nv);
11084 # undef HEXTRACT_HAS_TOP_NYBBLE
11085 HEXTRACT_BYTES_LE(13, 0);
11086 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11087 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11088 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11089 /* The bytes 2..15 are the mantissa/fraction,
11090 * the 0,1 are the sign+exponent. */
11091 const U8* nvp = (const U8*)(&nv);
11092 HEXTRACT_IMPLICIT_BIT(nv);
11093 # undef HEXTRACT_HAS_TOP_NYBBLE
11094 HEXTRACT_BYTES_BE(2, 15);
11095 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11096 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11097 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
11098 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
11099 * meaning that 2 or 6 bytes are empty padding. */
11100 /* The bytes 7..0 are the mantissa/fraction */
11101 const U8* nvp = (const U8*)(&nv);
11102 # undef HEXTRACT_HAS_IMPLICIT_BIT
11103 # undef HEXTRACT_HAS_TOP_NYBBLE
11104 HEXTRACT_BYTES_LE(7, 0);
11105 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11106 /* Does this format ever happen? (Wikipedia says the Motorola
11107 * 6888x math coprocessors used format _like_ this but padded
11108 * to 96 bits with 16 unused bits between the exponent and the
11110 const U8* nvp = (const U8*)(&nv);
11111 # undef HEXTRACT_HAS_IMPLICIT_BIT
11112 # undef HEXTRACT_HAS_TOP_NYBBLE
11113 HEXTRACT_BYTES_BE(0, 7);
11115 # define HEXTRACT_FALLBACK
11116 /* Double-double format: two doubles next to each other.
11117 * The first double is the high-order one, exactly like
11118 * it would be for a "lone" double. The second double
11119 * is shifted down using the exponent so that that there
11120 * are no common bits. The tricky part is that the value
11121 * of the double-double is the SUM of the two doubles and
11122 * the second one can be also NEGATIVE.
11124 * Because of this tricky construction the bytewise extraction we
11125 * use for the other long double formats doesn't work, we must
11126 * extract the values bit by bit.
11128 * The little-endian double-double is used .. somewhere?
11130 * The big endian double-double is used in e.g. PPC/Power (AIX)
11133 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11134 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11135 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11138 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11139 /* Using normal doubles, not long doubles.
11141 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11142 * bytes, since we might need to handle printf precision, and
11143 * also need to insert the radix. */
11145 # ifdef HEXTRACT_LITTLE_ENDIAN
11146 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11147 const U8* nvp = (const U8*)(&nv);
11148 HEXTRACT_IMPLICIT_BIT(nv);
11149 HEXTRACT_TOP_NYBBLE(6);
11150 HEXTRACT_BYTES_LE(5, 0);
11151 # elif defined(HEXTRACT_BIG_ENDIAN)
11152 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11153 const U8* nvp = (const U8*)(&nv);
11154 HEXTRACT_IMPLICIT_BIT(nv);
11155 HEXTRACT_TOP_NYBBLE(1);
11156 HEXTRACT_BYTES_BE(2, 7);
11157 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11158 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11159 const U8* nvp = (const U8*)(&nv);
11160 HEXTRACT_IMPLICIT_BIT(nv);
11161 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11162 HEXTRACT_BYTE(1); /* 5 */
11163 HEXTRACT_BYTE(0); /* 4 */
11164 HEXTRACT_BYTE(7); /* 3 */
11165 HEXTRACT_BYTE(6); /* 2 */
11166 HEXTRACT_BYTE(5); /* 1 */
11167 HEXTRACT_BYTE(4); /* 0 */
11168 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11169 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11170 const U8* nvp = (const U8*)(&nv);
11171 HEXTRACT_IMPLICIT_BIT(nv);
11172 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11173 HEXTRACT_BYTE(6); /* 5 */
11174 HEXTRACT_BYTE(7); /* 4 */
11175 HEXTRACT_BYTE(0); /* 3 */
11176 HEXTRACT_BYTE(1); /* 2 */
11177 HEXTRACT_BYTE(2); /* 1 */
11178 HEXTRACT_BYTE(3); /* 0 */
11180 # define HEXTRACT_FALLBACK
11183 # define HEXTRACT_FALLBACK
11185 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11186 # ifdef HEXTRACT_FALLBACK
11187 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11188 /* The fallback is used for the double-double format, and
11189 * for unknown long double formats, and for unknown double
11190 * formats, or in general unknown NV formats. */
11191 if (nv == (NV)0.0) {
11199 NV d = nv < 0 ? -nv : nv;
11201 U8 ha = 0x0; /* hexvalue accumulator */
11202 U8 hd = 0x8; /* hexvalue digit */
11204 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11205 * this is essentially manual frexp(). Multiplying by 0.5 and
11206 * doubling should be lossless in binary floating point. */
11216 while (d >= e + e) {
11220 /* Now e <= d < 2*e */
11222 /* First extract the leading hexdigit (the implicit bit). */
11238 /* Then extract the remaining hexdigits. */
11239 while (d > (NV)0.0) {
11245 /* Output or count in groups of four bits,
11246 * that is, when the hexdigit is down to one. */
11251 /* Reset the hexvalue. */
11260 /* Flush possible pending hexvalue. */
11270 /* Croak for various reasons: if the output pointer escaped the
11271 * output buffer, if the extraction index escaped the extraction
11272 * buffer, or if the ending output pointer didn't match the
11273 * previously computed value. */
11274 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11275 /* For double-double the ixmin and ixmax stay at zero,
11276 * which is convenient since the HEXTRACTSIZE is tricky
11277 * for double-double. */
11278 ixmin < 0 || ixmax >= NVSIZE ||
11279 (vend && v != vend)) {
11280 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11281 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11286 /* Helper for sv_vcatpvfn_flags(). */
11287 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11292 (var) = &PL_sv_no; /* [perl #71000] */ \
11293 arg_missing = TRUE; \
11298 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11299 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11304 const char *patend;
11307 static const char nullstr[] = "(null)";
11309 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11310 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11312 /* Times 4: a decimal digit takes more than 3 binary digits.
11313 * NV_DIG: mantissa takes than many decimal digits.
11314 * Plus 32: Playing safe. */
11315 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11316 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11317 bool hexfp = FALSE; /* hexadecimal floating point? */
11319 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11321 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11322 PERL_UNUSED_ARG(maybe_tainted);
11324 if (flags & SV_GMAGIC)
11327 /* no matter what, this is a string now */
11328 (void)SvPV_force_nomg(sv, origlen);
11330 /* special-case "", "%s", and "%-p" (SVf - see below) */
11332 if (svmax && ckWARN(WARN_REDUNDANT))
11333 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11334 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11337 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11338 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11339 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11340 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11343 const char * const s = va_arg(*args, char*);
11344 sv_catpv_nomg(sv, s ? s : nullstr);
11346 else if (svix < svmax) {
11347 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11348 SvGETMAGIC(*svargs);
11349 sv_catsv_nomg(sv, *svargs);
11352 S_warn_vcatpvfn_missing_argument(aTHX);
11355 if (args && patlen == 3 && pat[0] == '%' &&
11356 pat[1] == '-' && pat[2] == 'p') {
11357 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11358 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11359 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11360 argsv = MUTABLE_SV(va_arg(*args, void*));
11361 sv_catsv_nomg(sv, argsv);
11365 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11366 /* special-case "%.<number>[gf]" */
11367 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11368 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11369 unsigned digits = 0;
11373 while (*pp >= '0' && *pp <= '9')
11374 digits = 10 * digits + (*pp++ - '0');
11376 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11377 format the first argument and WARN_REDUNDANT if svmax > 1?
11378 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11379 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11380 const NV nv = SvNV(*svargs);
11381 if (LIKELY(!Perl_isinfnan(nv))) {
11383 /* Add check for digits != 0 because it seems that some
11384 gconverts are buggy in this case, and we don't yet have
11385 a Configure test for this. */
11386 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11387 /* 0, point, slack */
11388 STORE_LC_NUMERIC_SET_TO_NEEDED();
11389 SNPRINTF_G(nv, ebuf, size, digits);
11390 sv_catpv_nomg(sv, ebuf);
11391 if (*ebuf) /* May return an empty string for digits==0 */
11394 } else if (!digits) {
11397 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11398 sv_catpvn_nomg(sv, p, l);
11405 #endif /* !USE_LONG_DOUBLE */
11407 if (!args && svix < svmax && DO_UTF8(*svargs))
11410 patend = (char*)pat + patlen;
11411 for (p = (char*)pat; p < patend; p = q) {
11414 bool vectorize = FALSE;
11415 bool vectorarg = FALSE;
11416 bool vec_utf8 = FALSE;
11422 bool has_precis = FALSE;
11424 const I32 osvix = svix;
11425 bool is_utf8 = FALSE; /* is this item utf8? */
11426 bool used_explicit_ix = FALSE;
11427 bool arg_missing = FALSE;
11428 #ifdef HAS_LDBL_SPRINTF_BUG
11429 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11430 with sfio - Allen <allens@cpan.org> */
11431 bool fix_ldbl_sprintf_bug = FALSE;
11435 U8 utf8buf[UTF8_MAXBYTES+1];
11436 STRLEN esignlen = 0;
11438 const char *eptr = NULL;
11439 const char *fmtstart;
11442 const U8 *vecstr = NULL;
11449 /* We need a long double target in case HAS_LONG_DOUBLE,
11450 * even without USE_LONG_DOUBLE, so that we can printf with
11451 * long double formats, even without NV being long double.
11452 * But we call the target 'fv' instead of 'nv', since most of
11453 * the time it is not (most compilers these days recognize
11454 * "long double", even if only as a synonym for "double").
11456 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11457 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11459 # ifdef Perl_isfinitel
11460 # define FV_ISFINITE(x) Perl_isfinitel(x)
11462 # define FV_GF PERL_PRIgldbl
11463 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11464 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11465 # define NV_TO_FV(nv,fv) STMT_START { \
11467 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11470 # define NV_TO_FV(nv,fv) (fv)=(nv)
11474 # define FV_GF NVgf
11475 # define NV_TO_FV(nv,fv) (fv)=(nv)
11477 #ifndef FV_ISFINITE
11478 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11484 const char *dotstr = ".";
11485 STRLEN dotstrlen = 1;
11486 I32 efix = 0; /* explicit format parameter index */
11487 I32 ewix = 0; /* explicit width index */
11488 I32 epix = 0; /* explicit precision index */
11489 I32 evix = 0; /* explicit vector index */
11490 bool asterisk = FALSE;
11491 bool infnan = FALSE;
11493 /* echo everything up to the next format specification */
11494 for (q = p; q < patend && *q != '%'; ++q) ;
11496 if (has_utf8 && !pat_utf8)
11497 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11499 sv_catpvn_nomg(sv, p, q - p);
11508 We allow format specification elements in this order:
11509 \d+\$ explicit format parameter index
11511 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11512 0 flag (as above): repeated to allow "v02"
11513 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11514 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11516 [%bcdefginopsuxDFOUX] format (mandatory)
11521 As of perl5.9.3, printf format checking is on by default.
11522 Internally, perl uses %p formats to provide an escape to
11523 some extended formatting. This block deals with those
11524 extensions: if it does not match, (char*)q is reset and
11525 the normal format processing code is used.
11527 Currently defined extensions are:
11528 %p include pointer address (standard)
11529 %-p (SVf) include an SV (previously %_)
11530 %-<num>p include an SV with precision <num>
11532 %3p include a HEK with precision of 256
11533 %4p char* preceded by utf8 flag and length
11534 %<num>p (where num is 1 or > 4) reserved for future
11537 Robin Barker 2005-07-14 (but modified since)
11539 %1p (VDf) removed. RMB 2007-10-19
11546 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11547 /* The argument has already gone through cBOOL, so the cast
11549 is_utf8 = (bool)va_arg(*args, int);
11550 elen = va_arg(*args, UV);
11551 /* if utf8 length is larger than 0x7ffff..., then it might
11552 * have been a signed value that wrapped */
11553 if (elen > ((~(STRLEN)0) >> 1)) {
11554 assert(0); /* in DEBUGGING build we want to crash */
11555 elen= 0; /* otherwise we want to treat this as an empty string */
11557 eptr = va_arg(*args, char *);
11558 q += sizeof(UTF8f)-1;
11561 n = expect_number(&q);
11563 if (sv) { /* SVf */
11568 argsv = MUTABLE_SV(va_arg(*args, void*));
11569 eptr = SvPV_const(argsv, elen);
11570 if (DO_UTF8(argsv))
11574 else if (n==2 || n==3) { /* HEKf */
11575 HEK * const hek = va_arg(*args, HEK *);
11576 eptr = HEK_KEY(hek);
11577 elen = HEK_LEN(hek);
11578 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11579 if (n==3) precis = 256, has_precis = TRUE;
11583 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11584 "internal %%<num>p might conflict with future printf extensions");
11590 if ( (width = expect_number(&q)) ) {
11593 Perl_croak_nocontext(
11594 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11597 used_explicit_ix = TRUE;
11609 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11638 if ( (ewix = expect_number(&q)) ) {
11641 Perl_croak_nocontext(
11642 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11643 used_explicit_ix = TRUE;
11653 if ((vectorarg = asterisk)) {
11666 width = expect_number(&q);
11669 if (vectorize && vectorarg) {
11670 /* vectorizing, but not with the default "." */
11672 vecsv = va_arg(*args, SV*);
11674 FETCH_VCATPVFN_ARGUMENT(
11675 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11677 FETCH_VCATPVFN_ARGUMENT(
11678 vecsv, svix < svmax, svargs[svix++]);
11680 dotstr = SvPV_const(vecsv, dotstrlen);
11681 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11682 bad with tied or overloaded values that return UTF8. */
11683 if (DO_UTF8(vecsv))
11685 else if (has_utf8) {
11686 vecsv = sv_mortalcopy(vecsv);
11687 sv_utf8_upgrade(vecsv);
11688 dotstr = SvPV_const(vecsv, dotstrlen);
11695 i = va_arg(*args, int);
11697 i = (ewix ? ewix <= svmax : svix < svmax) ?
11698 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11700 width = (i < 0) ? -i : i;
11710 if ( (epix = expect_number(&q)) ) {
11713 Perl_croak_nocontext(
11714 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11715 used_explicit_ix = TRUE;
11720 i = va_arg(*args, int);
11724 FETCH_VCATPVFN_ARGUMENT(
11725 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11727 FETCH_VCATPVFN_ARGUMENT(
11728 precsv, svix < svmax, svargs[svix++]);
11729 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11732 has_precis = !(i < 0);
11736 while (isDIGIT(*q))
11737 precis = precis * 10 + (*q++ - '0');
11746 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11747 vecsv = svargs[efix ? efix-1 : svix++];
11748 vecstr = (U8*)SvPV_const(vecsv,veclen);
11749 vec_utf8 = DO_UTF8(vecsv);
11751 /* if this is a version object, we need to convert
11752 * back into v-string notation and then let the
11753 * vectorize happen normally
11755 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11756 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11757 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11758 "vector argument not supported with alpha versions");
11761 vecsv = sv_newmortal();
11762 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11764 vecstr = (U8*)SvPV_const(vecsv, veclen);
11765 vec_utf8 = DO_UTF8(vecsv);
11779 case 'I': /* Ix, I32x, and I64x */
11780 # ifdef USE_64_BIT_INT
11781 if (q[1] == '6' && q[2] == '4') {
11787 if (q[1] == '3' && q[2] == '2') {
11791 # ifdef USE_64_BIT_INT
11797 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11798 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11801 # ifdef USE_QUADMATH
11814 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11815 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11816 if (*q == 'l') { /* lld, llf */
11825 if (*++q == 'h') { /* hhd, hhu */
11854 if (!vectorize && !args) {
11856 const I32 i = efix-1;
11857 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11859 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11864 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11865 /* XXX va_arg(*args) case? need peek, use va_copy? */
11867 if (UNLIKELY(SvAMAGIC(argsv)))
11868 argsv = sv_2num(argsv);
11869 infnan = UNLIKELY(isinfnansv(argsv));
11872 switch (c = *q++) {
11880 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11881 /* no va_arg() case */
11882 SvNV_nomg(argsv), (int)c);
11883 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11885 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11887 eptr = (char*)utf8buf;
11888 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11902 eptr = va_arg(*args, char*);
11904 elen = strlen(eptr);
11906 eptr = (char *)nullstr;
11907 elen = sizeof nullstr - 1;
11911 eptr = SvPV_const(argsv, elen);
11912 if (DO_UTF8(argsv)) {
11913 STRLEN old_precis = precis;
11914 if (has_precis && precis < elen) {
11915 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11916 STRLEN p = precis > ulen ? ulen : precis;
11917 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11918 /* sticks at end */
11920 if (width) { /* fudge width (can't fudge elen) */
11921 if (has_precis && precis < elen)
11922 width += precis - old_precis;
11925 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11932 if (has_precis && precis < elen)
11940 goto floating_point;
11942 if (alt || vectorize)
11944 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11958 goto floating_point;
11963 goto donevalidconversion;
11965 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11974 esignbuf[esignlen++] = plus;
11978 case 'c': iv = (char)va_arg(*args, int); break;
11979 case 'h': iv = (short)va_arg(*args, int); break;
11980 case 'l': iv = va_arg(*args, long); break;
11981 case 'V': iv = va_arg(*args, IV); break;
11982 case 'z': iv = va_arg(*args, SSize_t); break;
11983 #ifdef HAS_PTRDIFF_T
11984 case 't': iv = va_arg(*args, ptrdiff_t); break;
11986 default: iv = va_arg(*args, int); break;
11988 case 'j': iv = va_arg(*args, intmax_t); break;
11992 iv = va_arg(*args, Quad_t); break;
11999 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12001 case 'c': iv = (char)tiv; break;
12002 case 'h': iv = (short)tiv; break;
12003 case 'l': iv = (long)tiv; break;
12005 default: iv = tiv; break;
12008 iv = (Quad_t)tiv; break;
12014 if ( !vectorize ) /* we already set uv above */
12019 esignbuf[esignlen++] = plus;
12022 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12023 esignbuf[esignlen++] = '-';
12062 goto floating_point;
12068 goto donevalidconversion;
12070 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12081 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12082 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12083 case 'l': uv = va_arg(*args, unsigned long); break;
12084 case 'V': uv = va_arg(*args, UV); break;
12085 case 'z': uv = va_arg(*args, Size_t); break;
12086 #ifdef HAS_PTRDIFF_T
12087 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12090 case 'j': uv = va_arg(*args, uintmax_t); break;
12092 default: uv = va_arg(*args, unsigned); break;
12095 uv = va_arg(*args, Uquad_t); break;
12102 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12104 case 'c': uv = (unsigned char)tuv; break;
12105 case 'h': uv = (unsigned short)tuv; break;
12106 case 'l': uv = (unsigned long)tuv; break;
12108 default: uv = tuv; break;
12111 uv = (Uquad_t)tuv; break;
12120 char *ptr = ebuf + sizeof ebuf;
12121 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12127 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12131 } while (uv >>= 4);
12133 esignbuf[esignlen++] = '0';
12134 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12140 *--ptr = '0' + dig;
12141 } while (uv >>= 3);
12142 if (alt && *ptr != '0')
12148 *--ptr = '0' + dig;
12149 } while (uv >>= 1);
12151 esignbuf[esignlen++] = '0';
12152 esignbuf[esignlen++] = c;
12155 default: /* it had better be ten or less */
12158 *--ptr = '0' + dig;
12159 } while (uv /= base);
12162 elen = (ebuf + sizeof ebuf) - ptr;
12166 zeros = precis - elen;
12167 else if (precis == 0 && elen == 1 && *eptr == '0'
12168 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12171 /* a precision nullifies the 0 flag. */
12178 /* FLOATING POINT */
12183 c = 'f'; /* maybe %F isn't supported here */
12185 case 'e': case 'E':
12187 case 'g': case 'G':
12188 case 'a': case 'A':
12192 /* This is evil, but floating point is even more evil */
12194 /* for SV-style calling, we can only get NV
12195 for C-style calling, we assume %f is double;
12196 for simplicity we allow any of %Lf, %llf, %qf for long double
12200 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12204 /* [perl #20339] - we should accept and ignore %lf rather than die */
12208 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12209 intsize = args ? 0 : 'q';
12213 #if defined(HAS_LONG_DOUBLE)
12226 /* Now we need (long double) if intsize == 'q', else (double). */
12228 /* Note: do not pull NVs off the va_list with va_arg()
12229 * (pull doubles instead) because if you have a build
12230 * with long doubles, you would always be pulling long
12231 * doubles, which would badly break anyone using only
12232 * doubles (i.e. the majority of builds). In other
12233 * words, you cannot mix doubles and long doubles.
12234 * The only case where you can pull off long doubles
12235 * is when the format specifier explicitly asks so with
12237 #ifdef USE_QUADMATH
12238 fv = intsize == 'q' ?
12239 va_arg(*args, NV) : va_arg(*args, double);
12241 #elif LONG_DOUBLESIZE > DOUBLESIZE
12242 if (intsize == 'q') {
12243 fv = va_arg(*args, long double);
12246 nv = va_arg(*args, double);
12250 nv = va_arg(*args, double);
12256 if (!infnan) SvGETMAGIC(argsv);
12257 nv = SvNV_nomg(argsv);
12262 /* frexp() (or frexpl) has some unspecified behaviour for
12263 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12264 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12266 (void)Perl_frexp((NV)fv, &i);
12267 if (i == PERL_INT_MIN)
12268 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12269 /* Do not set hexfp earlier since we want to printf
12270 * Inf/NaN for Inf/NaN, not their hexfp. */
12271 hexfp = isALPHA_FOLD_EQ(c, 'a');
12272 if (UNLIKELY(hexfp)) {
12273 /* This seriously overshoots in most cases, but
12274 * better the undershooting. Firstly, all bytes
12275 * of the NV are not mantissa, some of them are
12276 * exponent. Secondly, for the reasonably common
12277 * long doubles case, the "80-bit extended", two
12278 * or six bytes of the NV are unused. */
12280 (fv < 0) ? 1 : 0 + /* possible unary minus */
12282 1 + /* the very unlikely carry */
12285 2 * NVSIZE + /* 2 hexdigits for each byte */
12287 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12289 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12290 /* However, for the "double double", we need more.
12291 * Since each double has their own exponent, the
12292 * doubles may float (haha) rather far from each
12293 * other, and the number of required bits is much
12294 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12295 * See the definition of DOUBLEDOUBLE_MAXBITS.
12297 * Need 2 hexdigits for each byte. */
12298 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12299 /* the size for the exponent already added */
12301 #ifdef USE_LOCALE_NUMERIC
12302 STORE_LC_NUMERIC_SET_TO_NEEDED();
12303 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12304 need += SvLEN(PL_numeric_radix_sv);
12305 RESTORE_LC_NUMERIC();
12309 need = BIT_DIGITS(i);
12310 } /* if i < 0, the number of digits is hard to predict. */
12312 need += has_precis ? precis : 6; /* known default */
12317 #ifdef HAS_LDBL_SPRINTF_BUG
12318 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12319 with sfio - Allen <allens@cpan.org> */
12322 # define MY_DBL_MAX DBL_MAX
12323 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12324 # if DOUBLESIZE >= 8
12325 # define MY_DBL_MAX 1.7976931348623157E+308L
12327 # define MY_DBL_MAX 3.40282347E+38L
12331 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12332 # define MY_DBL_MAX_BUG 1L
12334 # define MY_DBL_MAX_BUG MY_DBL_MAX
12338 # define MY_DBL_MIN DBL_MIN
12339 # else /* XXX guessing! -Allen */
12340 # if DOUBLESIZE >= 8
12341 # define MY_DBL_MIN 2.2250738585072014E-308L
12343 # define MY_DBL_MIN 1.17549435E-38L
12347 if ((intsize == 'q') && (c == 'f') &&
12348 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12349 (need < DBL_DIG)) {
12350 /* it's going to be short enough that
12351 * long double precision is not needed */
12353 if ((fv <= 0L) && (fv >= -0L))
12354 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12356 /* would use Perl_fp_class as a double-check but not
12357 * functional on IRIX - see perl.h comments */
12359 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12360 /* It's within the range that a double can represent */
12361 #if defined(DBL_MAX) && !defined(DBL_MIN)
12362 if ((fv >= ((long double)1/DBL_MAX)) ||
12363 (fv <= (-(long double)1/DBL_MAX)))
12365 fix_ldbl_sprintf_bug = TRUE;
12368 if (fix_ldbl_sprintf_bug == TRUE) {
12378 # undef MY_DBL_MAX_BUG
12381 #endif /* HAS_LDBL_SPRINTF_BUG */
12383 need += 20; /* fudge factor */
12384 if (PL_efloatsize < need) {
12385 Safefree(PL_efloatbuf);
12386 PL_efloatsize = need + 20; /* more fudge */
12387 Newx(PL_efloatbuf, PL_efloatsize, char);
12388 PL_efloatbuf[0] = '\0';
12391 if ( !(width || left || plus || alt) && fill != '0'
12392 && has_precis && intsize != 'q' /* Shortcuts */
12393 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12394 /* See earlier comment about buggy Gconvert when digits,
12396 if ( c == 'g' && precis ) {
12397 STORE_LC_NUMERIC_SET_TO_NEEDED();
12398 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12399 /* May return an empty string for digits==0 */
12400 if (*PL_efloatbuf) {
12401 elen = strlen(PL_efloatbuf);
12402 goto float_converted;
12404 } else if ( c == 'f' && !precis ) {
12405 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12410 if (UNLIKELY(hexfp)) {
12411 /* Hexadecimal floating point. */
12412 char* p = PL_efloatbuf;
12413 U8 vhex[VHEX_SIZE];
12414 U8* v = vhex; /* working pointer to vhex */
12415 U8* vend; /* pointer to one beyond last digit of vhex */
12416 U8* vfnz = NULL; /* first non-zero */
12417 U8* vlnz = NULL; /* last non-zero */
12418 const bool lower = (c == 'a');
12419 /* At output the values of vhex (up to vend) will
12420 * be mapped through the xdig to get the actual
12421 * human-readable xdigits. */
12422 const char* xdig = PL_hexdigit;
12423 int zerotail = 0; /* how many extra zeros to append */
12424 int exponent = 0; /* exponent of the floating point input */
12425 bool hexradix = FALSE; /* should we output the radix */
12427 /* XXX: denormals, NaN, Inf.
12429 * For example with denormals, (assuming the vanilla
12430 * 64-bit double): the exponent is zero. 1xp-1074 is
12431 * the smallest denormal and the smallest double, it
12432 * should be output as 0x0.0000000000001p-1022 to
12433 * match its internal structure. */
12435 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
12436 S_hextract(aTHX_ nv, &exponent, vhex, vend);
12438 #if NVSIZE > DOUBLESIZE
12439 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12440 /* In this case there is an implicit bit,
12441 * and therefore the exponent is shifted shift by one. */
12444 /* In this case there is no implicit bit,
12445 * and the exponent is shifted by the first xdigit. */
12451 || Perl_signbit(nv)
12462 xdig += 16; /* Use uppercase hex. */
12465 /* Find the first non-zero xdigit. */
12466 for (v = vhex; v < vend; v++) {
12474 /* Find the last non-zero xdigit. */
12475 for (v = vend - 1; v >= vhex; v--) {
12482 #if NVSIZE == DOUBLESIZE
12488 if ((SSize_t)(precis + 1) < vend - vhex) {
12491 v = vhex + precis + 1;
12492 /* Round away from zero: if the tail
12493 * beyond the precis xdigits is equal to
12494 * or greater than 0x8000... */
12496 if (!round && *v == 0x8) {
12497 for (v++; v < vend; v++) {
12505 for (v = vhex + precis; v >= vhex; v--) {
12512 /* If the carry goes all the way to
12513 * the front, we need to output
12514 * a single '1'. This goes against
12515 * the "xdigit and then radix"
12516 * but since this is "cannot happen"
12517 * category, that is probably good. */
12522 /* The new effective "last non zero". */
12523 vlnz = vhex + precis;
12526 zerotail = precis - (vlnz - vhex);
12533 /* If there are non-zero xdigits, the radix
12534 * is output after the first one. */
12545 /* The radix is always output if precis, or if alt. */
12546 if (precis > 0 || alt) {
12551 #ifndef USE_LOCALE_NUMERIC
12554 STORE_LC_NUMERIC_SET_TO_NEEDED();
12555 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12557 const char* r = SvPV(PL_numeric_radix_sv, n);
12558 Copy(r, p, n, char);
12564 RESTORE_LC_NUMERIC();
12573 if (zerotail > 0) {
12574 while (zerotail--) {
12579 elen = p - PL_efloatbuf;
12580 elen += my_snprintf(p, PL_efloatsize - elen,
12581 "%c%+d", lower ? 'p' : 'P',
12584 if (elen < width) {
12586 /* Pad the back with spaces. */
12587 memset(PL_efloatbuf + elen, ' ', width - elen);
12589 else if (fill == '0') {
12590 /* Insert the zeros between the "0x" and
12591 * the digits, otherwise we end up with
12593 STRLEN nzero = width - elen;
12594 char* zerox = PL_efloatbuf + 2;
12595 Move(zerox, zerox + nzero, elen - 2, char);
12596 memset(zerox, fill, nzero);
12599 /* Move it to the right. */
12600 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12602 /* Pad the front with spaces. */
12603 memset(PL_efloatbuf, ' ', width - elen);
12609 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12611 /* Not affecting infnan output: precision, alt, fill. */
12612 if (elen < width) {
12614 /* Pack the back with spaces. */
12615 memset(PL_efloatbuf + elen, ' ', width - elen);
12617 /* Move it to the right. */
12618 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12620 /* Pad the front with spaces. */
12621 memset(PL_efloatbuf, ' ', width - elen);
12629 char *ptr = ebuf + sizeof ebuf;
12632 #if defined(USE_QUADMATH)
12633 if (intsize == 'q') {
12637 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12638 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12639 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12640 * not USE_LONG_DOUBLE and NVff. In other words,
12641 * this needs to work without USE_LONG_DOUBLE. */
12642 if (intsize == 'q') {
12643 /* Copy the one or more characters in a long double
12644 * format before the 'base' ([efgEFG]) character to
12645 * the format string. */
12646 static char const ldblf[] = PERL_PRIfldbl;
12647 char const *p = ldblf + sizeof(ldblf) - 3;
12648 while (p >= ldblf) { *--ptr = *p--; }
12653 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12658 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12670 /* No taint. Otherwise we are in the strange situation
12671 * where printf() taints but print($float) doesn't.
12674 STORE_LC_NUMERIC_SET_TO_NEEDED();
12676 /* hopefully the above makes ptr a very constrained format
12677 * that is safe to use, even though it's not literal */
12678 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12679 #ifdef USE_QUADMATH
12681 const char* qfmt = quadmath_format_single(ptr);
12683 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12684 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12686 if ((IV)elen == -1)
12687 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12691 #elif defined(HAS_LONG_DOUBLE)
12692 elen = ((intsize == 'q')
12693 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12694 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12696 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12702 eptr = PL_efloatbuf;
12703 assert((IV)elen > 0); /* here zero elen is bad */
12705 #ifdef USE_LOCALE_NUMERIC
12706 /* If the decimal point character in the string is UTF-8, make the
12708 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12709 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12722 i = SvCUR(sv) - origlen;
12725 case 'c': *(va_arg(*args, char*)) = i; break;
12726 case 'h': *(va_arg(*args, short*)) = i; break;
12727 default: *(va_arg(*args, int*)) = i; break;
12728 case 'l': *(va_arg(*args, long*)) = i; break;
12729 case 'V': *(va_arg(*args, IV*)) = i; break;
12730 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12731 #ifdef HAS_PTRDIFF_T
12732 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12735 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12739 *(va_arg(*args, Quad_t*)) = i; break;
12746 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12747 goto donevalidconversion;
12754 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12755 && ckWARN(WARN_PRINTF))
12757 SV * const msg = sv_newmortal();
12758 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12759 (PL_op->op_type == OP_PRTF) ? "" : "s");
12760 if (fmtstart < patend) {
12761 const char * const fmtend = q < patend ? q : patend;
12763 sv_catpvs(msg, "\"%");
12764 for (f = fmtstart; f < fmtend; f++) {
12766 sv_catpvn_nomg(msg, f, 1);
12768 Perl_sv_catpvf(aTHX_ msg,
12769 "\\%03"UVof, (UV)*f & 0xFF);
12772 sv_catpvs(msg, "\"");
12774 sv_catpvs(msg, "end of string");
12776 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12779 /* output mangled stuff ... */
12785 /* ... right here, because formatting flags should not apply */
12786 SvGROW(sv, SvCUR(sv) + elen + 1);
12788 Copy(eptr, p, elen, char);
12791 SvCUR_set(sv, p - SvPVX_const(sv));
12793 continue; /* not "break" */
12796 if (is_utf8 != has_utf8) {
12799 sv_utf8_upgrade(sv);
12802 const STRLEN old_elen = elen;
12803 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12804 sv_utf8_upgrade(nsv);
12805 eptr = SvPVX_const(nsv);
12808 if (width) { /* fudge width (can't fudge elen) */
12809 width += elen - old_elen;
12815 /* signed value that's wrapped? */
12816 assert(elen <= ((~(STRLEN)0) >> 1));
12817 have = esignlen + zeros + elen;
12819 croak_memory_wrap();
12821 need = (have > width ? have : width);
12824 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12825 croak_memory_wrap();
12826 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12828 if (esignlen && fill == '0') {
12830 for (i = 0; i < (int)esignlen; i++)
12831 *p++ = esignbuf[i];
12833 if (gap && !left) {
12834 memset(p, fill, gap);
12837 if (esignlen && fill != '0') {
12839 for (i = 0; i < (int)esignlen; i++)
12840 *p++ = esignbuf[i];
12844 for (i = zeros; i; i--)
12848 Copy(eptr, p, elen, char);
12852 memset(p, ' ', gap);
12857 Copy(dotstr, p, dotstrlen, char);
12861 vectorize = FALSE; /* done iterating over vecstr */
12868 SvCUR_set(sv, p - SvPVX_const(sv));
12874 donevalidconversion:
12875 if (used_explicit_ix)
12876 no_redundant_warning = TRUE;
12878 S_warn_vcatpvfn_missing_argument(aTHX);
12881 /* Now that we've consumed all our printf format arguments (svix)
12882 * do we have things left on the stack that we didn't use?
12884 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12885 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12886 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12891 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12895 /* =========================================================================
12897 =head1 Cloning an interpreter
12901 All the macros and functions in this section are for the private use of
12902 the main function, perl_clone().
12904 The foo_dup() functions make an exact copy of an existing foo thingy.
12905 During the course of a cloning, a hash table is used to map old addresses
12906 to new addresses. The table is created and manipulated with the
12907 ptr_table_* functions.
12909 * =========================================================================*/
12912 #if defined(USE_ITHREADS)
12914 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12915 #ifndef GpREFCNT_inc
12916 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12920 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12921 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12922 If this changes, please unmerge ss_dup.
12923 Likewise, sv_dup_inc_multiple() relies on this fact. */
12924 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12925 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12926 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12927 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12928 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12929 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12930 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12931 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12932 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12933 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12934 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12935 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12936 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12938 /* clone a parser */
12941 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12945 PERL_ARGS_ASSERT_PARSER_DUP;
12950 /* look for it in the table first */
12951 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12955 /* create anew and remember what it is */
12956 Newxz(parser, 1, yy_parser);
12957 ptr_table_store(PL_ptr_table, proto, parser);
12959 /* XXX these not yet duped */
12960 parser->old_parser = NULL;
12961 parser->stack = NULL;
12963 parser->stack_size = 0;
12964 /* XXX parser->stack->state = 0; */
12966 /* XXX eventually, just Copy() most of the parser struct ? */
12968 parser->lex_brackets = proto->lex_brackets;
12969 parser->lex_casemods = proto->lex_casemods;
12970 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12971 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12972 parser->lex_casestack = savepvn(proto->lex_casestack,
12973 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12974 parser->lex_defer = proto->lex_defer;
12975 parser->lex_dojoin = proto->lex_dojoin;
12976 parser->lex_formbrack = proto->lex_formbrack;
12977 parser->lex_inpat = proto->lex_inpat;
12978 parser->lex_inwhat = proto->lex_inwhat;
12979 parser->lex_op = proto->lex_op;
12980 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12981 parser->lex_starts = proto->lex_starts;
12982 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12983 parser->multi_close = proto->multi_close;
12984 parser->multi_open = proto->multi_open;
12985 parser->multi_start = proto->multi_start;
12986 parser->multi_end = proto->multi_end;
12987 parser->preambled = proto->preambled;
12988 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12989 parser->linestr = sv_dup_inc(proto->linestr, param);
12990 parser->expect = proto->expect;
12991 parser->copline = proto->copline;
12992 parser->last_lop_op = proto->last_lop_op;
12993 parser->lex_state = proto->lex_state;
12994 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12995 /* rsfp_filters entries have fake IoDIRP() */
12996 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12997 parser->in_my = proto->in_my;
12998 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12999 parser->error_count = proto->error_count;
13002 parser->linestr = sv_dup_inc(proto->linestr, param);
13005 char * const ols = SvPVX(proto->linestr);
13006 char * const ls = SvPVX(parser->linestr);
13008 parser->bufptr = ls + (proto->bufptr >= ols ?
13009 proto->bufptr - ols : 0);
13010 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13011 proto->oldbufptr - ols : 0);
13012 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13013 proto->oldoldbufptr - ols : 0);
13014 parser->linestart = ls + (proto->linestart >= ols ?
13015 proto->linestart - ols : 0);
13016 parser->last_uni = ls + (proto->last_uni >= ols ?
13017 proto->last_uni - ols : 0);
13018 parser->last_lop = ls + (proto->last_lop >= ols ?
13019 proto->last_lop - ols : 0);
13021 parser->bufend = ls + SvCUR(parser->linestr);
13024 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13027 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13028 Copy(proto->nexttype, parser->nexttype, 5, I32);
13029 parser->nexttoke = proto->nexttoke;
13031 /* XXX should clone saved_curcop here, but we aren't passed
13032 * proto_perl; so do it in perl_clone_using instead */
13038 /* duplicate a file handle */
13041 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13045 PERL_ARGS_ASSERT_FP_DUP;
13046 PERL_UNUSED_ARG(type);
13049 return (PerlIO*)NULL;
13051 /* look for it in the table first */
13052 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13056 /* create anew and remember what it is */
13057 #ifdef __amigaos4__
13058 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13060 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13062 ptr_table_store(PL_ptr_table, fp, ret);
13066 /* duplicate a directory handle */
13069 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13073 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13075 const Direntry_t *dirent;
13076 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13082 PERL_UNUSED_CONTEXT;
13083 PERL_ARGS_ASSERT_DIRP_DUP;
13088 /* look for it in the table first */
13089 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13093 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13095 PERL_UNUSED_ARG(param);
13099 /* open the current directory (so we can switch back) */
13100 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13102 /* chdir to our dir handle and open the present working directory */
13103 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13104 PerlDir_close(pwd);
13105 return (DIR *)NULL;
13107 /* Now we should have two dir handles pointing to the same dir. */
13109 /* Be nice to the calling code and chdir back to where we were. */
13110 /* XXX If this fails, then what? */
13111 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13113 /* We have no need of the pwd handle any more. */
13114 PerlDir_close(pwd);
13117 # define d_namlen(d) (d)->d_namlen
13119 # define d_namlen(d) strlen((d)->d_name)
13121 /* Iterate once through dp, to get the file name at the current posi-
13122 tion. Then step back. */
13123 pos = PerlDir_tell(dp);
13124 if ((dirent = PerlDir_read(dp))) {
13125 len = d_namlen(dirent);
13126 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13127 /* If the len is somehow magically longer than the
13128 * maximum length of the directory entry, even though
13129 * we could fit it in a buffer, we could not copy it
13130 * from the dirent. Bail out. */
13131 PerlDir_close(ret);
13134 if (len <= sizeof smallbuf) name = smallbuf;
13135 else Newx(name, len, char);
13136 Move(dirent->d_name, name, len, char);
13138 PerlDir_seek(dp, pos);
13140 /* Iterate through the new dir handle, till we find a file with the
13142 if (!dirent) /* just before the end */
13144 pos = PerlDir_tell(ret);
13145 if (PerlDir_read(ret)) continue; /* not there yet */
13146 PerlDir_seek(ret, pos); /* step back */
13150 const long pos0 = PerlDir_tell(ret);
13152 pos = PerlDir_tell(ret);
13153 if ((dirent = PerlDir_read(ret))) {
13154 if (len == (STRLEN)d_namlen(dirent)
13155 && memEQ(name, dirent->d_name, len)) {
13157 PerlDir_seek(ret, pos); /* step back */
13160 /* else we are not there yet; keep iterating */
13162 else { /* This is not meant to happen. The best we can do is
13163 reset the iterator to the beginning. */
13164 PerlDir_seek(ret, pos0);
13171 if (name && name != smallbuf)
13176 ret = win32_dirp_dup(dp, param);
13179 /* pop it in the pointer table */
13181 ptr_table_store(PL_ptr_table, dp, ret);
13186 /* duplicate a typeglob */
13189 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13193 PERL_ARGS_ASSERT_GP_DUP;
13197 /* look for it in the table first */
13198 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13202 /* create anew and remember what it is */
13204 ptr_table_store(PL_ptr_table, gp, ret);
13207 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13208 on Newxz() to do this for us. */
13209 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13210 ret->gp_io = io_dup_inc(gp->gp_io, param);
13211 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13212 ret->gp_av = av_dup_inc(gp->gp_av, param);
13213 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13214 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13215 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13216 ret->gp_cvgen = gp->gp_cvgen;
13217 ret->gp_line = gp->gp_line;
13218 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13222 /* duplicate a chain of magic */
13225 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13227 MAGIC *mgret = NULL;
13228 MAGIC **mgprev_p = &mgret;
13230 PERL_ARGS_ASSERT_MG_DUP;
13232 for (; mg; mg = mg->mg_moremagic) {
13235 if ((param->flags & CLONEf_JOIN_IN)
13236 && mg->mg_type == PERL_MAGIC_backref)
13237 /* when joining, we let the individual SVs add themselves to
13238 * backref as needed. */
13241 Newx(nmg, 1, MAGIC);
13243 mgprev_p = &(nmg->mg_moremagic);
13245 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13246 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13247 from the original commit adding Perl_mg_dup() - revision 4538.
13248 Similarly there is the annotation "XXX random ptr?" next to the
13249 assignment to nmg->mg_ptr. */
13252 /* FIXME for plugins
13253 if (nmg->mg_type == PERL_MAGIC_qr) {
13254 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13258 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13259 ? nmg->mg_type == PERL_MAGIC_backref
13260 /* The backref AV has its reference
13261 * count deliberately bumped by 1 */
13262 ? SvREFCNT_inc(av_dup_inc((const AV *)
13263 nmg->mg_obj, param))
13264 : sv_dup_inc(nmg->mg_obj, param)
13265 : sv_dup(nmg->mg_obj, param);
13267 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13268 if (nmg->mg_len > 0) {
13269 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13270 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13271 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13273 AMT * const namtp = (AMT*)nmg->mg_ptr;
13274 sv_dup_inc_multiple((SV**)(namtp->table),
13275 (SV**)(namtp->table), NofAMmeth, param);
13278 else if (nmg->mg_len == HEf_SVKEY)
13279 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13281 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13282 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13288 #endif /* USE_ITHREADS */
13290 struct ptr_tbl_arena {
13291 struct ptr_tbl_arena *next;
13292 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13295 /* create a new pointer-mapping table */
13298 Perl_ptr_table_new(pTHX)
13301 PERL_UNUSED_CONTEXT;
13303 Newx(tbl, 1, PTR_TBL_t);
13304 tbl->tbl_max = 511;
13305 tbl->tbl_items = 0;
13306 tbl->tbl_arena = NULL;
13307 tbl->tbl_arena_next = NULL;
13308 tbl->tbl_arena_end = NULL;
13309 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13313 #define PTR_TABLE_HASH(ptr) \
13314 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13316 /* map an existing pointer using a table */
13318 STATIC PTR_TBL_ENT_t *
13319 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13321 PTR_TBL_ENT_t *tblent;
13322 const UV hash = PTR_TABLE_HASH(sv);
13324 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13326 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13327 for (; tblent; tblent = tblent->next) {
13328 if (tblent->oldval == sv)
13335 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13337 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13339 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13340 PERL_UNUSED_CONTEXT;
13342 return tblent ? tblent->newval : NULL;
13345 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13346 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13347 * the core's typical use of ptr_tables in thread cloning. */
13350 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13352 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13354 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13355 PERL_UNUSED_CONTEXT;
13358 tblent->newval = newsv;
13360 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13362 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13363 struct ptr_tbl_arena *new_arena;
13365 Newx(new_arena, 1, struct ptr_tbl_arena);
13366 new_arena->next = tbl->tbl_arena;
13367 tbl->tbl_arena = new_arena;
13368 tbl->tbl_arena_next = new_arena->array;
13369 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13372 tblent = tbl->tbl_arena_next++;
13374 tblent->oldval = oldsv;
13375 tblent->newval = newsv;
13376 tblent->next = tbl->tbl_ary[entry];
13377 tbl->tbl_ary[entry] = tblent;
13379 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13380 ptr_table_split(tbl);
13384 /* double the hash bucket size of an existing ptr table */
13387 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13389 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13390 const UV oldsize = tbl->tbl_max + 1;
13391 UV newsize = oldsize * 2;
13394 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13395 PERL_UNUSED_CONTEXT;
13397 Renew(ary, newsize, PTR_TBL_ENT_t*);
13398 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13399 tbl->tbl_max = --newsize;
13400 tbl->tbl_ary = ary;
13401 for (i=0; i < oldsize; i++, ary++) {
13402 PTR_TBL_ENT_t **entp = ary;
13403 PTR_TBL_ENT_t *ent = *ary;
13404 PTR_TBL_ENT_t **curentp;
13407 curentp = ary + oldsize;
13409 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13411 ent->next = *curentp;
13421 /* remove all the entries from a ptr table */
13422 /* Deprecated - will be removed post 5.14 */
13425 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13427 PERL_UNUSED_CONTEXT;
13428 if (tbl && tbl->tbl_items) {
13429 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13431 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13434 struct ptr_tbl_arena *next = arena->next;
13440 tbl->tbl_items = 0;
13441 tbl->tbl_arena = NULL;
13442 tbl->tbl_arena_next = NULL;
13443 tbl->tbl_arena_end = NULL;
13447 /* clear and free a ptr table */
13450 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13452 struct ptr_tbl_arena *arena;
13454 PERL_UNUSED_CONTEXT;
13460 arena = tbl->tbl_arena;
13463 struct ptr_tbl_arena *next = arena->next;
13469 Safefree(tbl->tbl_ary);
13473 #if defined(USE_ITHREADS)
13476 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13478 PERL_ARGS_ASSERT_RVPV_DUP;
13480 assert(!isREGEXP(sstr));
13482 if (SvWEAKREF(sstr)) {
13483 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13484 if (param->flags & CLONEf_JOIN_IN) {
13485 /* if joining, we add any back references individually rather
13486 * than copying the whole backref array */
13487 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13491 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13493 else if (SvPVX_const(sstr)) {
13494 /* Has something there */
13496 /* Normal PV - clone whole allocated space */
13497 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13498 /* sstr may not be that normal, but actually copy on write.
13499 But we are a true, independent SV, so: */
13503 /* Special case - not normally malloced for some reason */
13504 if (isGV_with_GP(sstr)) {
13505 /* Don't need to do anything here. */
13507 else if ((SvIsCOW(sstr))) {
13508 /* A "shared" PV - clone it as "shared" PV */
13510 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13514 /* Some other special case - random pointer */
13515 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13520 /* Copy the NULL */
13521 SvPV_set(dstr, NULL);
13525 /* duplicate a list of SVs. source and dest may point to the same memory. */
13527 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13528 SSize_t items, CLONE_PARAMS *const param)
13530 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13532 while (items-- > 0) {
13533 *dest++ = sv_dup_inc(*source++, param);
13539 /* duplicate an SV of any type (including AV, HV etc) */
13542 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13547 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13549 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13550 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13555 /* look for it in the table first */
13556 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13560 if(param->flags & CLONEf_JOIN_IN) {
13561 /** We are joining here so we don't want do clone
13562 something that is bad **/
13563 if (SvTYPE(sstr) == SVt_PVHV) {
13564 const HEK * const hvname = HvNAME_HEK(sstr);
13566 /** don't clone stashes if they already exist **/
13567 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13568 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13569 ptr_table_store(PL_ptr_table, sstr, dstr);
13573 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13574 HV *stash = GvSTASH(sstr);
13575 const HEK * hvname;
13576 if (stash && (hvname = HvNAME_HEK(stash))) {
13577 /** don't clone GVs if they already exist **/
13579 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13580 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13582 stash, GvNAME(sstr),
13588 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13589 ptr_table_store(PL_ptr_table, sstr, *svp);
13596 /* create anew and remember what it is */
13599 #ifdef DEBUG_LEAKING_SCALARS
13600 dstr->sv_debug_optype = sstr->sv_debug_optype;
13601 dstr->sv_debug_line = sstr->sv_debug_line;
13602 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13603 dstr->sv_debug_parent = (SV*)sstr;
13604 FREE_SV_DEBUG_FILE(dstr);
13605 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13608 ptr_table_store(PL_ptr_table, sstr, dstr);
13611 SvFLAGS(dstr) = SvFLAGS(sstr);
13612 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13613 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13616 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13617 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13618 (void*)PL_watch_pvx, SvPVX_const(sstr));
13621 /* don't clone objects whose class has asked us not to */
13623 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13629 switch (SvTYPE(sstr)) {
13631 SvANY(dstr) = NULL;
13634 SET_SVANY_FOR_BODYLESS_IV(dstr);
13636 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13638 SvIV_set(dstr, SvIVX(sstr));
13642 #if NVSIZE <= IVSIZE
13643 SET_SVANY_FOR_BODYLESS_NV(dstr);
13645 SvANY(dstr) = new_XNV();
13647 SvNV_set(dstr, SvNVX(sstr));
13651 /* These are all the types that need complex bodies allocating. */
13653 const svtype sv_type = SvTYPE(sstr);
13654 const struct body_details *const sv_type_details
13655 = bodies_by_type + sv_type;
13659 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13675 assert(sv_type_details->body_size);
13676 if (sv_type_details->arena) {
13677 new_body_inline(new_body, sv_type);
13679 = (void*)((char*)new_body - sv_type_details->offset);
13681 new_body = new_NOARENA(sv_type_details);
13685 SvANY(dstr) = new_body;
13688 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13689 ((char*)SvANY(dstr)) + sv_type_details->offset,
13690 sv_type_details->copy, char);
13692 Copy(((char*)SvANY(sstr)),
13693 ((char*)SvANY(dstr)),
13694 sv_type_details->body_size + sv_type_details->offset, char);
13697 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13698 && !isGV_with_GP(dstr)
13700 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13701 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13703 /* The Copy above means that all the source (unduplicated) pointers
13704 are now in the destination. We can check the flags and the
13705 pointers in either, but it's possible that there's less cache
13706 missing by always going for the destination.
13707 FIXME - instrument and check that assumption */
13708 if (sv_type >= SVt_PVMG) {
13710 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13711 if (SvOBJECT(dstr) && SvSTASH(dstr))
13712 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13713 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13716 /* The cast silences a GCC warning about unhandled types. */
13717 switch ((int)sv_type) {
13728 /* FIXME for plugins */
13729 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13730 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13733 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13734 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13735 LvTARG(dstr) = dstr;
13736 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13737 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13739 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13740 if (isREGEXP(sstr)) goto duprex;
13742 /* non-GP case already handled above */
13743 if(isGV_with_GP(sstr)) {
13744 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13745 /* Don't call sv_add_backref here as it's going to be
13746 created as part of the magic cloning of the symbol
13747 table--unless this is during a join and the stash
13748 is not actually being cloned. */
13749 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13750 at the point of this comment. */
13751 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13752 if (param->flags & CLONEf_JOIN_IN)
13753 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13754 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13755 (void)GpREFCNT_inc(GvGP(dstr));
13759 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13760 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13761 /* I have no idea why fake dirp (rsfps)
13762 should be treated differently but otherwise
13763 we end up with leaks -- sky*/
13764 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13765 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13766 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13768 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13769 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13770 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13771 if (IoDIRP(dstr)) {
13772 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13775 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13777 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13779 if (IoOFP(dstr) == IoIFP(sstr))
13780 IoOFP(dstr) = IoIFP(dstr);
13782 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13783 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13784 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13785 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13788 /* avoid cloning an empty array */
13789 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13790 SV **dst_ary, **src_ary;
13791 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13793 src_ary = AvARRAY((const AV *)sstr);
13794 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13795 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13796 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13797 AvALLOC((const AV *)dstr) = dst_ary;
13798 if (AvREAL((const AV *)sstr)) {
13799 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13803 while (items-- > 0)
13804 *dst_ary++ = sv_dup(*src_ary++, param);
13806 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13807 while (items-- > 0) {
13812 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13813 AvALLOC((const AV *)dstr) = (SV**)NULL;
13814 AvMAX( (const AV *)dstr) = -1;
13815 AvFILLp((const AV *)dstr) = -1;
13819 if (HvARRAY((const HV *)sstr)) {
13821 const bool sharekeys = !!HvSHAREKEYS(sstr);
13822 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13823 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13825 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13826 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13828 HvARRAY(dstr) = (HE**)darray;
13829 while (i <= sxhv->xhv_max) {
13830 const HE * const source = HvARRAY(sstr)[i];
13831 HvARRAY(dstr)[i] = source
13832 ? he_dup(source, sharekeys, param) : 0;
13836 const struct xpvhv_aux * const saux = HvAUX(sstr);
13837 struct xpvhv_aux * const daux = HvAUX(dstr);
13838 /* This flag isn't copied. */
13841 if (saux->xhv_name_count) {
13842 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13844 = saux->xhv_name_count < 0
13845 ? -saux->xhv_name_count
13846 : saux->xhv_name_count;
13847 HEK **shekp = sname + count;
13849 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13850 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13851 while (shekp-- > sname) {
13853 *dhekp = hek_dup(*shekp, param);
13857 daux->xhv_name_u.xhvnameu_name
13858 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13861 daux->xhv_name_count = saux->xhv_name_count;
13863 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13864 daux->xhv_aux_flags = saux->xhv_aux_flags;
13865 #ifdef PERL_HASH_RANDOMIZE_KEYS
13866 daux->xhv_rand = saux->xhv_rand;
13867 daux->xhv_last_rand = saux->xhv_last_rand;
13869 daux->xhv_riter = saux->xhv_riter;
13870 daux->xhv_eiter = saux->xhv_eiter
13871 ? he_dup(saux->xhv_eiter,
13872 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13873 /* backref array needs refcnt=2; see sv_add_backref */
13874 daux->xhv_backreferences =
13875 (param->flags & CLONEf_JOIN_IN)
13876 /* when joining, we let the individual GVs and
13877 * CVs add themselves to backref as
13878 * needed. This avoids pulling in stuff
13879 * that isn't required, and simplifies the
13880 * case where stashes aren't cloned back
13881 * if they already exist in the parent
13884 : saux->xhv_backreferences
13885 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13886 ? MUTABLE_AV(SvREFCNT_inc(
13887 sv_dup_inc((const SV *)
13888 saux->xhv_backreferences, param)))
13889 : MUTABLE_AV(sv_dup((const SV *)
13890 saux->xhv_backreferences, param))
13893 daux->xhv_mro_meta = saux->xhv_mro_meta
13894 ? mro_meta_dup(saux->xhv_mro_meta, param)
13897 /* Record stashes for possible cloning in Perl_clone(). */
13899 av_push(param->stashes, dstr);
13903 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13906 if (!(param->flags & CLONEf_COPY_STACKS)) {
13911 /* NOTE: not refcounted */
13912 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13913 hv_dup(CvSTASH(dstr), param);
13914 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13915 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13916 if (!CvISXSUB(dstr)) {
13918 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13920 CvSLABBED_off(dstr);
13921 } else if (CvCONST(dstr)) {
13922 CvXSUBANY(dstr).any_ptr =
13923 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13925 assert(!CvSLABBED(dstr));
13926 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13928 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13929 hek_dup(CvNAME_HEK((CV *)sstr), param);
13930 /* don't dup if copying back - CvGV isn't refcounted, so the
13931 * duped GV may never be freed. A bit of a hack! DAPM */
13933 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13935 ? gv_dup_inc(CvGV(sstr), param)
13936 : (param->flags & CLONEf_JOIN_IN)
13938 : gv_dup(CvGV(sstr), param);
13940 if (!CvISXSUB(sstr)) {
13941 PADLIST * padlist = CvPADLIST(sstr);
13943 padlist = padlist_dup(padlist, param);
13944 CvPADLIST_set(dstr, padlist);
13946 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13947 PoisonPADLIST(dstr);
13950 CvWEAKOUTSIDE(sstr)
13951 ? cv_dup( CvOUTSIDE(dstr), param)
13952 : cv_dup_inc(CvOUTSIDE(dstr), param);
13962 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13964 PERL_ARGS_ASSERT_SV_DUP_INC;
13965 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13969 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13971 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13972 PERL_ARGS_ASSERT_SV_DUP;
13974 /* Track every SV that (at least initially) had a reference count of 0.
13975 We need to do this by holding an actual reference to it in this array.
13976 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13977 (akin to the stashes hash, and the perl stack), we come unstuck if
13978 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13979 thread) is manipulated in a CLONE method, because CLONE runs before the
13980 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13981 (and fix things up by giving each a reference via the temps stack).
13982 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13983 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13984 before the walk of unreferenced happens and a reference to that is SV
13985 added to the temps stack. At which point we have the same SV considered
13986 to be in use, and free to be re-used. Not good.
13988 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13989 assert(param->unreferenced);
13990 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13996 /* duplicate a context */
13999 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14001 PERL_CONTEXT *ncxs;
14003 PERL_ARGS_ASSERT_CX_DUP;
14006 return (PERL_CONTEXT*)NULL;
14008 /* look for it in the table first */
14009 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14013 /* create anew and remember what it is */
14014 Newx(ncxs, max + 1, PERL_CONTEXT);
14015 ptr_table_store(PL_ptr_table, cxs, ncxs);
14016 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14019 PERL_CONTEXT * const ncx = &ncxs[ix];
14020 if (CxTYPE(ncx) == CXt_SUBST) {
14021 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14024 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14025 switch (CxTYPE(ncx)) {
14027 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14028 if(CxHASARGS(ncx)){
14029 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14031 ncx->blk_sub.savearray = NULL;
14033 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14034 ncx->blk_sub.prevcomppad);
14037 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14039 /* XXX should this sv_dup_inc? Or only if SvSCREAM ???? */
14040 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14041 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14042 /* XXX what do do with cur_top_env ???? */
14044 case CXt_LOOP_LAZYSV:
14045 ncx->blk_loop.state_u.lazysv.end
14046 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14047 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14048 duplication code instead.
14049 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14050 actually being the same function, and (2) order
14051 equivalence of the two unions.
14052 We can assert the later [but only at run time :-(] */
14053 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14054 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14057 ncx->blk_loop.state_u.ary.ary
14058 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14060 case CXt_LOOP_LIST:
14061 case CXt_LOOP_LAZYIV:
14062 /* code common to all 'for' CXt_LOOP_* types */
14063 ncx->blk_loop.itersave =
14064 sv_dup_inc(ncx->blk_loop.itersave, param);
14065 if (CxPADLOOP(ncx)) {
14066 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14067 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14068 ncx->blk_loop.oldcomppad =
14069 (PAD*)ptr_table_fetch(PL_ptr_table,
14070 ncx->blk_loop.oldcomppad);
14071 ncx->blk_loop.itervar_u.svp =
14072 &CX_CURPAD_SV(ncx->blk_loop, off);
14075 /* this copies the GV if CXp_FOR_GV, or the SV for an
14076 * alias (for \$x (...)) - relies on gv_dup being the
14077 * same as sv_dup */
14078 ncx->blk_loop.itervar_u.gv
14079 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14083 case CXt_LOOP_PLAIN:
14086 ncx->blk_format.prevcomppad =
14087 (PAD*)ptr_table_fetch(PL_ptr_table,
14088 ncx->blk_format.prevcomppad);
14089 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14090 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14091 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14095 ncx->blk_givwhen.defsv_save =
14096 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14109 /* duplicate a stack info structure */
14112 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14116 PERL_ARGS_ASSERT_SI_DUP;
14119 return (PERL_SI*)NULL;
14121 /* look for it in the table first */
14122 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14126 /* create anew and remember what it is */
14127 Newxz(nsi, 1, PERL_SI);
14128 ptr_table_store(PL_ptr_table, si, nsi);
14130 nsi->si_stack = av_dup_inc(si->si_stack, param);
14131 nsi->si_cxix = si->si_cxix;
14132 nsi->si_cxmax = si->si_cxmax;
14133 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14134 nsi->si_type = si->si_type;
14135 nsi->si_prev = si_dup(si->si_prev, param);
14136 nsi->si_next = si_dup(si->si_next, param);
14137 nsi->si_markoff = si->si_markoff;
14142 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14143 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14144 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14145 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14146 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14147 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14148 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14149 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14150 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14151 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14152 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14153 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14154 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14155 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14156 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14157 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14160 #define pv_dup_inc(p) SAVEPV(p)
14161 #define pv_dup(p) SAVEPV(p)
14162 #define svp_dup_inc(p,pp) any_dup(p,pp)
14164 /* map any object to the new equivent - either something in the
14165 * ptr table, or something in the interpreter structure
14169 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14173 PERL_ARGS_ASSERT_ANY_DUP;
14176 return (void*)NULL;
14178 /* look for it in the table first */
14179 ret = ptr_table_fetch(PL_ptr_table, v);
14183 /* see if it is part of the interpreter structure */
14184 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14185 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14193 /* duplicate the save stack */
14196 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14199 ANY * const ss = proto_perl->Isavestack;
14200 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14201 I32 ix = proto_perl->Isavestack_ix;
14214 void (*dptr) (void*);
14215 void (*dxptr) (pTHX_ void*);
14217 PERL_ARGS_ASSERT_SS_DUP;
14219 Newxz(nss, max, ANY);
14222 const UV uv = POPUV(ss,ix);
14223 const U8 type = (U8)uv & SAVE_MASK;
14225 TOPUV(nss,ix) = uv;
14227 case SAVEt_CLEARSV:
14228 case SAVEt_CLEARPADRANGE:
14230 case SAVEt_HELEM: /* hash element */
14231 case SAVEt_SV: /* scalar reference */
14232 sv = (const SV *)POPPTR(ss,ix);
14233 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14235 case SAVEt_ITEM: /* normal string */
14236 case SAVEt_GVSV: /* scalar slot in GV */
14237 sv = (const SV *)POPPTR(ss,ix);
14238 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14239 if (type == SAVEt_SV)
14243 case SAVEt_MORTALIZESV:
14244 case SAVEt_READONLY_OFF:
14245 sv = (const SV *)POPPTR(ss,ix);
14246 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14248 case SAVEt_FREEPADNAME:
14249 ptr = POPPTR(ss,ix);
14250 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14251 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14253 case SAVEt_SHARED_PVREF: /* char* in shared space */
14254 c = (char*)POPPTR(ss,ix);
14255 TOPPTR(nss,ix) = savesharedpv(c);
14256 ptr = POPPTR(ss,ix);
14257 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14259 case SAVEt_GENERIC_SVREF: /* generic sv */
14260 case SAVEt_SVREF: /* scalar reference */
14261 sv = (const SV *)POPPTR(ss,ix);
14262 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14263 if (type == SAVEt_SVREF)
14264 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14265 ptr = POPPTR(ss,ix);
14266 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14268 case SAVEt_GVSLOT: /* any slot in GV */
14269 sv = (const SV *)POPPTR(ss,ix);
14270 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14271 ptr = POPPTR(ss,ix);
14272 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14273 sv = (const SV *)POPPTR(ss,ix);
14274 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14276 case SAVEt_HV: /* hash reference */
14277 case SAVEt_AV: /* array reference */
14278 sv = (const SV *) POPPTR(ss,ix);
14279 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14281 case SAVEt_COMPPAD:
14283 sv = (const SV *) POPPTR(ss,ix);
14284 TOPPTR(nss,ix) = sv_dup(sv, param);
14286 case SAVEt_INT: /* int reference */
14287 ptr = POPPTR(ss,ix);
14288 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14289 intval = (int)POPINT(ss,ix);
14290 TOPINT(nss,ix) = intval;
14292 case SAVEt_LONG: /* long reference */
14293 ptr = POPPTR(ss,ix);
14294 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14295 longval = (long)POPLONG(ss,ix);
14296 TOPLONG(nss,ix) = longval;
14298 case SAVEt_I32: /* I32 reference */
14299 ptr = POPPTR(ss,ix);
14300 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14302 TOPINT(nss,ix) = i;
14304 case SAVEt_IV: /* IV reference */
14305 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14306 ptr = POPPTR(ss,ix);
14307 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14309 TOPIV(nss,ix) = iv;
14311 case SAVEt_TMPSFLOOR:
14313 TOPIV(nss,ix) = iv;
14315 case SAVEt_HPTR: /* HV* reference */
14316 case SAVEt_APTR: /* AV* reference */
14317 case SAVEt_SPTR: /* SV* reference */
14318 ptr = POPPTR(ss,ix);
14319 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14320 sv = (const SV *)POPPTR(ss,ix);
14321 TOPPTR(nss,ix) = sv_dup(sv, param);
14323 case SAVEt_VPTR: /* random* reference */
14324 ptr = POPPTR(ss,ix);
14325 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14327 case SAVEt_INT_SMALL:
14328 case SAVEt_I32_SMALL:
14329 case SAVEt_I16: /* I16 reference */
14330 case SAVEt_I8: /* I8 reference */
14332 ptr = POPPTR(ss,ix);
14333 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14335 case SAVEt_GENERIC_PVREF: /* generic char* */
14336 case SAVEt_PPTR: /* char* reference */
14337 ptr = POPPTR(ss,ix);
14338 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14339 c = (char*)POPPTR(ss,ix);
14340 TOPPTR(nss,ix) = pv_dup(c);
14342 case SAVEt_GP: /* scalar reference */
14343 gp = (GP*)POPPTR(ss,ix);
14344 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14345 (void)GpREFCNT_inc(gp);
14346 gv = (const GV *)POPPTR(ss,ix);
14347 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14350 ptr = POPPTR(ss,ix);
14351 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14352 /* these are assumed to be refcounted properly */
14354 switch (((OP*)ptr)->op_type) {
14356 case OP_LEAVESUBLV:
14360 case OP_LEAVEWRITE:
14361 TOPPTR(nss,ix) = ptr;
14364 (void) OpREFCNT_inc(o);
14368 TOPPTR(nss,ix) = NULL;
14373 TOPPTR(nss,ix) = NULL;
14375 case SAVEt_FREECOPHH:
14376 ptr = POPPTR(ss,ix);
14377 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14379 case SAVEt_ADELETE:
14380 av = (const AV *)POPPTR(ss,ix);
14381 TOPPTR(nss,ix) = av_dup_inc(av, param);
14383 TOPINT(nss,ix) = i;
14386 hv = (const HV *)POPPTR(ss,ix);
14387 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14389 TOPINT(nss,ix) = i;
14392 c = (char*)POPPTR(ss,ix);
14393 TOPPTR(nss,ix) = pv_dup_inc(c);
14395 case SAVEt_STACK_POS: /* Position on Perl stack */
14397 TOPINT(nss,ix) = i;
14399 case SAVEt_DESTRUCTOR:
14400 ptr = POPPTR(ss,ix);
14401 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14402 dptr = POPDPTR(ss,ix);
14403 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14404 any_dup(FPTR2DPTR(void *, dptr),
14407 case SAVEt_DESTRUCTOR_X:
14408 ptr = POPPTR(ss,ix);
14409 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14410 dxptr = POPDXPTR(ss,ix);
14411 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14412 any_dup(FPTR2DPTR(void *, dxptr),
14415 case SAVEt_REGCONTEXT:
14417 ix -= uv >> SAVE_TIGHT_SHIFT;
14419 case SAVEt_AELEM: /* array element */
14420 sv = (const SV *)POPPTR(ss,ix);
14421 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14423 TOPINT(nss,ix) = i;
14424 av = (const AV *)POPPTR(ss,ix);
14425 TOPPTR(nss,ix) = av_dup_inc(av, param);
14428 ptr = POPPTR(ss,ix);
14429 TOPPTR(nss,ix) = ptr;
14432 ptr = POPPTR(ss,ix);
14433 ptr = cophh_copy((COPHH*)ptr);
14434 TOPPTR(nss,ix) = ptr;
14436 TOPINT(nss,ix) = i;
14437 if (i & HINT_LOCALIZE_HH) {
14438 hv = (const HV *)POPPTR(ss,ix);
14439 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14442 case SAVEt_PADSV_AND_MORTALIZE:
14443 longval = (long)POPLONG(ss,ix);
14444 TOPLONG(nss,ix) = longval;
14445 ptr = POPPTR(ss,ix);
14446 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14447 sv = (const SV *)POPPTR(ss,ix);
14448 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14450 case SAVEt_SET_SVFLAGS:
14452 TOPINT(nss,ix) = i;
14454 TOPINT(nss,ix) = i;
14455 sv = (const SV *)POPPTR(ss,ix);
14456 TOPPTR(nss,ix) = sv_dup(sv, param);
14458 case SAVEt_COMPILE_WARNINGS:
14459 ptr = POPPTR(ss,ix);
14460 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14463 ptr = POPPTR(ss,ix);
14464 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14468 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14476 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14477 * flag to the result. This is done for each stash before cloning starts,
14478 * so we know which stashes want their objects cloned */
14481 do_mark_cloneable_stash(pTHX_ SV *const sv)
14483 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14485 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14486 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14487 if (cloner && GvCV(cloner)) {
14494 mXPUSHs(newSVhek(hvname));
14496 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14503 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14511 =for apidoc perl_clone
14513 Create and return a new interpreter by cloning the current one.
14515 C<perl_clone> takes these flags as parameters:
14517 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14518 without it we only clone the data and zero the stacks,
14519 with it we copy the stacks and the new perl interpreter is
14520 ready to run at the exact same point as the previous one.
14521 The pseudo-fork code uses C<COPY_STACKS> while the
14522 threads->create doesn't.
14524 C<CLONEf_KEEP_PTR_TABLE> -
14525 C<perl_clone> keeps a ptr_table with the pointer of the old
14526 variable as a key and the new variable as a value,
14527 this allows it to check if something has been cloned and not
14528 clone it again but rather just use the value and increase the
14529 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14530 the ptr_table using the function
14531 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14532 reason to keep it around is if you want to dup some of your own
14533 variable who are outside the graph perl scans, an example of this
14534 code is in F<threads.xs> create.
14536 C<CLONEf_CLONE_HOST> -
14537 This is a win32 thing, it is ignored on unix, it tells perls
14538 win32host code (which is c++) to clone itself, this is needed on
14539 win32 if you want to run two threads at the same time,
14540 if you just want to do some stuff in a separate perl interpreter
14541 and then throw it away and return to the original one,
14542 you don't need to do anything.
14547 /* XXX the above needs expanding by someone who actually understands it ! */
14548 EXTERN_C PerlInterpreter *
14549 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14552 perl_clone(PerlInterpreter *proto_perl, UV flags)
14555 #ifdef PERL_IMPLICIT_SYS
14557 PERL_ARGS_ASSERT_PERL_CLONE;
14559 /* perlhost.h so we need to call into it
14560 to clone the host, CPerlHost should have a c interface, sky */
14562 #ifndef __amigaos4__
14563 if (flags & CLONEf_CLONE_HOST) {
14564 return perl_clone_host(proto_perl,flags);
14567 return perl_clone_using(proto_perl, flags,
14569 proto_perl->IMemShared,
14570 proto_perl->IMemParse,
14572 proto_perl->IStdIO,
14576 proto_perl->IProc);
14580 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14581 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14582 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14583 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14584 struct IPerlDir* ipD, struct IPerlSock* ipS,
14585 struct IPerlProc* ipP)
14587 /* XXX many of the string copies here can be optimized if they're
14588 * constants; they need to be allocated as common memory and just
14589 * their pointers copied. */
14592 CLONE_PARAMS clone_params;
14593 CLONE_PARAMS* const param = &clone_params;
14595 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14597 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14598 #else /* !PERL_IMPLICIT_SYS */
14600 CLONE_PARAMS clone_params;
14601 CLONE_PARAMS* param = &clone_params;
14602 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14604 PERL_ARGS_ASSERT_PERL_CLONE;
14605 #endif /* PERL_IMPLICIT_SYS */
14607 /* for each stash, determine whether its objects should be cloned */
14608 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14609 PERL_SET_THX(my_perl);
14612 PoisonNew(my_perl, 1, PerlInterpreter);
14615 PL_defstash = NULL; /* may be used by perl malloc() */
14618 PL_scopestack_name = 0;
14620 PL_savestack_ix = 0;
14621 PL_savestack_max = -1;
14622 PL_sig_pending = 0;
14624 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14625 Zero(&PL_padname_undef, 1, PADNAME);
14626 Zero(&PL_padname_const, 1, PADNAME);
14627 # ifdef DEBUG_LEAKING_SCALARS
14628 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14630 # ifdef PERL_TRACE_OPS
14631 Zero(PL_op_exec_cnt, OP_max+2, UV);
14633 #else /* !DEBUGGING */
14634 Zero(my_perl, 1, PerlInterpreter);
14635 #endif /* DEBUGGING */
14637 #ifdef PERL_IMPLICIT_SYS
14638 /* host pointers */
14640 PL_MemShared = ipMS;
14641 PL_MemParse = ipMP;
14648 #endif /* PERL_IMPLICIT_SYS */
14651 param->flags = flags;
14652 /* Nothing in the core code uses this, but we make it available to
14653 extensions (using mg_dup). */
14654 param->proto_perl = proto_perl;
14655 /* Likely nothing will use this, but it is initialised to be consistent
14656 with Perl_clone_params_new(). */
14657 param->new_perl = my_perl;
14658 param->unreferenced = NULL;
14661 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14663 PL_body_arenas = NULL;
14664 Zero(&PL_body_roots, 1, PL_body_roots);
14668 PL_sv_arenaroot = NULL;
14670 PL_debug = proto_perl->Idebug;
14672 /* dbargs array probably holds garbage */
14675 PL_compiling = proto_perl->Icompiling;
14677 /* pseudo environmental stuff */
14678 PL_origargc = proto_perl->Iorigargc;
14679 PL_origargv = proto_perl->Iorigargv;
14681 #ifndef NO_TAINT_SUPPORT
14682 /* Set tainting stuff before PerlIO_debug can possibly get called */
14683 PL_tainting = proto_perl->Itainting;
14684 PL_taint_warn = proto_perl->Itaint_warn;
14686 PL_tainting = FALSE;
14687 PL_taint_warn = FALSE;
14690 PL_minus_c = proto_perl->Iminus_c;
14692 PL_localpatches = proto_perl->Ilocalpatches;
14693 PL_splitstr = proto_perl->Isplitstr;
14694 PL_minus_n = proto_perl->Iminus_n;
14695 PL_minus_p = proto_perl->Iminus_p;
14696 PL_minus_l = proto_perl->Iminus_l;
14697 PL_minus_a = proto_perl->Iminus_a;
14698 PL_minus_E = proto_perl->Iminus_E;
14699 PL_minus_F = proto_perl->Iminus_F;
14700 PL_doswitches = proto_perl->Idoswitches;
14701 PL_dowarn = proto_perl->Idowarn;
14702 #ifdef PERL_SAWAMPERSAND
14703 PL_sawampersand = proto_perl->Isawampersand;
14705 PL_unsafe = proto_perl->Iunsafe;
14706 PL_perldb = proto_perl->Iperldb;
14707 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14708 PL_exit_flags = proto_perl->Iexit_flags;
14710 /* XXX time(&PL_basetime) when asked for? */
14711 PL_basetime = proto_perl->Ibasetime;
14713 PL_maxsysfd = proto_perl->Imaxsysfd;
14714 PL_statusvalue = proto_perl->Istatusvalue;
14716 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14718 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14721 /* RE engine related */
14722 PL_regmatch_slab = NULL;
14723 PL_reg_curpm = NULL;
14725 PL_sub_generation = proto_perl->Isub_generation;
14727 /* funky return mechanisms */
14728 PL_forkprocess = proto_perl->Iforkprocess;
14730 /* internal state */
14731 PL_maxo = proto_perl->Imaxo;
14733 PL_main_start = proto_perl->Imain_start;
14734 PL_eval_root = proto_perl->Ieval_root;
14735 PL_eval_start = proto_perl->Ieval_start;
14737 PL_filemode = proto_perl->Ifilemode;
14738 PL_lastfd = proto_perl->Ilastfd;
14739 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14742 PL_gensym = proto_perl->Igensym;
14744 PL_laststatval = proto_perl->Ilaststatval;
14745 PL_laststype = proto_perl->Ilaststype;
14748 PL_profiledata = NULL;
14750 PL_generation = proto_perl->Igeneration;
14752 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14753 PL_in_clean_all = proto_perl->Iin_clean_all;
14755 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14756 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14757 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14758 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14759 PL_nomemok = proto_perl->Inomemok;
14760 PL_an = proto_perl->Ian;
14761 PL_evalseq = proto_perl->Ievalseq;
14762 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14763 PL_origalen = proto_perl->Iorigalen;
14765 PL_sighandlerp = proto_perl->Isighandlerp;
14767 PL_runops = proto_perl->Irunops;
14769 PL_subline = proto_perl->Isubline;
14771 PL_cv_has_eval = proto_perl->Icv_has_eval;
14774 PL_cryptseen = proto_perl->Icryptseen;
14777 #ifdef USE_LOCALE_COLLATE
14778 PL_collation_ix = proto_perl->Icollation_ix;
14779 PL_collation_standard = proto_perl->Icollation_standard;
14780 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14781 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14782 #endif /* USE_LOCALE_COLLATE */
14784 #ifdef USE_LOCALE_NUMERIC
14785 PL_numeric_standard = proto_perl->Inumeric_standard;
14786 PL_numeric_local = proto_perl->Inumeric_local;
14787 #endif /* !USE_LOCALE_NUMERIC */
14789 /* Did the locale setup indicate UTF-8? */
14790 PL_utf8locale = proto_perl->Iutf8locale;
14791 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14792 /* Unicode features (see perlrun/-C) */
14793 PL_unicode = proto_perl->Iunicode;
14795 /* Pre-5.8 signals control */
14796 PL_signals = proto_perl->Isignals;
14798 /* times() ticks per second */
14799 PL_clocktick = proto_perl->Iclocktick;
14801 /* Recursion stopper for PerlIO_find_layer */
14802 PL_in_load_module = proto_perl->Iin_load_module;
14804 /* sort() routine */
14805 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14807 /* Not really needed/useful since the reenrant_retint is "volatile",
14808 * but do it for consistency's sake. */
14809 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14811 /* Hooks to shared SVs and locks. */
14812 PL_sharehook = proto_perl->Isharehook;
14813 PL_lockhook = proto_perl->Ilockhook;
14814 PL_unlockhook = proto_perl->Iunlockhook;
14815 PL_threadhook = proto_perl->Ithreadhook;
14816 PL_destroyhook = proto_perl->Idestroyhook;
14817 PL_signalhook = proto_perl->Isignalhook;
14819 PL_globhook = proto_perl->Iglobhook;
14822 PL_last_swash_hv = NULL; /* reinits on demand */
14823 PL_last_swash_klen = 0;
14824 PL_last_swash_key[0]= '\0';
14825 PL_last_swash_tmps = (U8*)NULL;
14826 PL_last_swash_slen = 0;
14828 PL_srand_called = proto_perl->Isrand_called;
14829 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14831 if (flags & CLONEf_COPY_STACKS) {
14832 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14833 PL_tmps_ix = proto_perl->Itmps_ix;
14834 PL_tmps_max = proto_perl->Itmps_max;
14835 PL_tmps_floor = proto_perl->Itmps_floor;
14837 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14838 * NOTE: unlike the others! */
14839 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14840 PL_scopestack_max = proto_perl->Iscopestack_max;
14842 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14843 * NOTE: unlike the others! */
14844 PL_savestack_ix = proto_perl->Isavestack_ix;
14845 PL_savestack_max = proto_perl->Isavestack_max;
14848 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14849 PL_top_env = &PL_start_env;
14851 PL_op = proto_perl->Iop;
14854 PL_Xpv = (XPV*)NULL;
14855 my_perl->Ina = proto_perl->Ina;
14857 PL_statbuf = proto_perl->Istatbuf;
14858 PL_statcache = proto_perl->Istatcache;
14860 #ifndef NO_TAINT_SUPPORT
14861 PL_tainted = proto_perl->Itainted;
14863 PL_tainted = FALSE;
14865 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14867 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14869 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14870 PL_restartop = proto_perl->Irestartop;
14871 PL_in_eval = proto_perl->Iin_eval;
14872 PL_delaymagic = proto_perl->Idelaymagic;
14873 PL_phase = proto_perl->Iphase;
14874 PL_localizing = proto_perl->Ilocalizing;
14876 PL_hv_fetch_ent_mh = NULL;
14877 PL_modcount = proto_perl->Imodcount;
14878 PL_lastgotoprobe = NULL;
14879 PL_dumpindent = proto_perl->Idumpindent;
14881 PL_efloatbuf = NULL; /* reinits on demand */
14882 PL_efloatsize = 0; /* reinits on demand */
14886 PL_colorset = 0; /* reinits PL_colors[] */
14887 /*PL_colors[6] = {0,0,0,0,0,0};*/
14889 /* Pluggable optimizer */
14890 PL_peepp = proto_perl->Ipeepp;
14891 PL_rpeepp = proto_perl->Irpeepp;
14892 /* op_free() hook */
14893 PL_opfreehook = proto_perl->Iopfreehook;
14895 #ifdef USE_REENTRANT_API
14896 /* XXX: things like -Dm will segfault here in perlio, but doing
14897 * PERL_SET_CONTEXT(proto_perl);
14898 * breaks too many other things
14900 Perl_reentrant_init(aTHX);
14903 /* create SV map for pointer relocation */
14904 PL_ptr_table = ptr_table_new();
14906 /* initialize these special pointers as early as possible */
14908 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14909 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14910 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14911 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14912 &PL_padname_const);
14914 /* create (a non-shared!) shared string table */
14915 PL_strtab = newHV();
14916 HvSHAREKEYS_off(PL_strtab);
14917 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14918 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14920 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14922 /* This PV will be free'd special way so must set it same way op.c does */
14923 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14924 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14926 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14927 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14928 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14929 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14931 param->stashes = newAV(); /* Setup array of objects to call clone on */
14932 /* This makes no difference to the implementation, as it always pushes
14933 and shifts pointers to other SVs without changing their reference
14934 count, with the array becoming empty before it is freed. However, it
14935 makes it conceptually clear what is going on, and will avoid some
14936 work inside av.c, filling slots between AvFILL() and AvMAX() with
14937 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14938 AvREAL_off(param->stashes);
14940 if (!(flags & CLONEf_COPY_STACKS)) {
14941 param->unreferenced = newAV();
14944 #ifdef PERLIO_LAYERS
14945 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14946 PerlIO_clone(aTHX_ proto_perl, param);
14949 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14950 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14951 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14952 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14953 PL_xsubfilename = proto_perl->Ixsubfilename;
14954 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14955 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14958 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14959 PL_inplace = SAVEPV(proto_perl->Iinplace);
14960 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14962 /* magical thingies */
14964 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14965 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14967 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14968 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14969 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14972 /* Clone the regex array */
14973 /* ORANGE FIXME for plugins, probably in the SV dup code.
14974 newSViv(PTR2IV(CALLREGDUPE(
14975 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14977 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14978 PL_regex_pad = AvARRAY(PL_regex_padav);
14980 PL_stashpadmax = proto_perl->Istashpadmax;
14981 PL_stashpadix = proto_perl->Istashpadix ;
14982 Newx(PL_stashpad, PL_stashpadmax, HV *);
14985 for (; o < PL_stashpadmax; ++o)
14986 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14989 /* shortcuts to various I/O objects */
14990 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14991 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14992 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14993 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14994 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14995 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14996 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14998 /* shortcuts to regexp stuff */
14999 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15001 /* shortcuts to misc objects */
15002 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15004 /* shortcuts to debugging objects */
15005 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15006 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15007 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15008 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15009 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15010 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15011 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15013 /* symbol tables */
15014 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15015 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15016 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15017 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15018 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15020 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15021 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15022 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15023 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15024 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15025 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15026 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15027 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15028 PL_savebegin = proto_perl->Isavebegin;
15030 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15032 /* subprocess state */
15033 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15035 if (proto_perl->Iop_mask)
15036 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15039 /* PL_asserting = proto_perl->Iasserting; */
15041 /* current interpreter roots */
15042 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15044 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15047 /* runtime control stuff */
15048 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15050 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15052 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15054 /* interpreter atexit processing */
15055 PL_exitlistlen = proto_perl->Iexitlistlen;
15056 if (PL_exitlistlen) {
15057 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15058 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15061 PL_exitlist = (PerlExitListEntry*)NULL;
15063 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15064 if (PL_my_cxt_size) {
15065 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15066 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15067 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15068 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15069 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15073 PL_my_cxt_list = (void**)NULL;
15074 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15075 PL_my_cxt_keys = (const char**)NULL;
15078 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15079 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15080 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15081 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15083 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15085 PAD_CLONE_VARS(proto_perl, param);
15087 #ifdef HAVE_INTERP_INTERN
15088 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15091 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15093 #ifdef PERL_USES_PL_PIDSTATUS
15094 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15096 PL_osname = SAVEPV(proto_perl->Iosname);
15097 PL_parser = parser_dup(proto_perl->Iparser, param);
15099 /* XXX this only works if the saved cop has already been cloned */
15100 if (proto_perl->Iparser) {
15101 PL_parser->saved_curcop = (COP*)any_dup(
15102 proto_perl->Iparser->saved_curcop,
15106 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15108 #ifdef USE_LOCALE_CTYPE
15109 /* Should we warn if uses locale? */
15110 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15113 #ifdef USE_LOCALE_COLLATE
15114 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15115 #endif /* USE_LOCALE_COLLATE */
15117 #ifdef USE_LOCALE_NUMERIC
15118 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15119 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15120 #endif /* !USE_LOCALE_NUMERIC */
15122 /* Unicode inversion lists */
15123 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15124 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15125 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15126 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15128 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15129 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15131 /* utf8 character class swashes */
15132 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15133 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15135 for (i = 0; i < POSIX_CC_COUNT; i++) {
15136 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15138 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15139 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15140 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15141 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15142 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15143 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15144 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15145 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15146 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15147 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15148 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15149 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15150 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15151 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15152 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15153 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15154 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15156 if (proto_perl->Ipsig_pend) {
15157 Newxz(PL_psig_pend, SIG_SIZE, int);
15160 PL_psig_pend = (int*)NULL;
15163 if (proto_perl->Ipsig_name) {
15164 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15165 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15167 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15170 PL_psig_ptr = (SV**)NULL;
15171 PL_psig_name = (SV**)NULL;
15174 if (flags & CLONEf_COPY_STACKS) {
15175 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15176 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15177 PL_tmps_ix+1, param);
15179 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15180 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15181 Newxz(PL_markstack, i, I32);
15182 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15183 - proto_perl->Imarkstack);
15184 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15185 - proto_perl->Imarkstack);
15186 Copy(proto_perl->Imarkstack, PL_markstack,
15187 PL_markstack_ptr - PL_markstack + 1, I32);
15189 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15190 * NOTE: unlike the others! */
15191 Newxz(PL_scopestack, PL_scopestack_max, I32);
15192 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15195 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15196 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15198 /* reset stack AV to correct length before its duped via
15199 * PL_curstackinfo */
15200 AvFILLp(proto_perl->Icurstack) =
15201 proto_perl->Istack_sp - proto_perl->Istack_base;
15203 /* NOTE: si_dup() looks at PL_markstack */
15204 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15206 /* PL_curstack = PL_curstackinfo->si_stack; */
15207 PL_curstack = av_dup(proto_perl->Icurstack, param);
15208 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15210 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15211 PL_stack_base = AvARRAY(PL_curstack);
15212 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15213 - proto_perl->Istack_base);
15214 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15216 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15217 PL_savestack = ss_dup(proto_perl, param);
15221 ENTER; /* perl_destruct() wants to LEAVE; */
15224 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15225 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15227 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15228 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15229 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15230 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15231 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15232 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15234 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15236 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15237 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15238 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15240 PL_stashcache = newHV();
15242 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15243 proto_perl->Iwatchaddr);
15244 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15245 if (PL_debug && PL_watchaddr) {
15246 PerlIO_printf(Perl_debug_log,
15247 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15248 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15249 PTR2UV(PL_watchok));
15252 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15253 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15254 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15256 /* Call the ->CLONE method, if it exists, for each of the stashes
15257 identified by sv_dup() above.
15259 while(av_tindex(param->stashes) != -1) {
15260 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15261 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15262 if (cloner && GvCV(cloner)) {
15267 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15269 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15275 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15276 ptr_table_free(PL_ptr_table);
15277 PL_ptr_table = NULL;
15280 if (!(flags & CLONEf_COPY_STACKS)) {
15281 unreferenced_to_tmp_stack(param->unreferenced);
15284 SvREFCNT_dec(param->stashes);
15286 /* orphaned? eg threads->new inside BEGIN or use */
15287 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15288 SvREFCNT_inc_simple_void(PL_compcv);
15289 SAVEFREESV(PL_compcv);
15296 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15298 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15300 if (AvFILLp(unreferenced) > -1) {
15301 SV **svp = AvARRAY(unreferenced);
15302 SV **const last = svp + AvFILLp(unreferenced);
15306 if (SvREFCNT(*svp) == 1)
15308 } while (++svp <= last);
15310 EXTEND_MORTAL(count);
15311 svp = AvARRAY(unreferenced);
15314 if (SvREFCNT(*svp) == 1) {
15315 /* Our reference is the only one to this SV. This means that
15316 in this thread, the scalar effectively has a 0 reference.
15317 That doesn't work (cleanup never happens), so donate our
15318 reference to it onto the save stack. */
15319 PL_tmps_stack[++PL_tmps_ix] = *svp;
15321 /* As an optimisation, because we are already walking the
15322 entire array, instead of above doing either
15323 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15324 release our reference to the scalar, so that at the end of
15325 the array owns zero references to the scalars it happens to
15326 point to. We are effectively converting the array from
15327 AvREAL() on to AvREAL() off. This saves the av_clear()
15328 (triggered by the SvREFCNT_dec(unreferenced) below) from
15329 walking the array a second time. */
15330 SvREFCNT_dec(*svp);
15333 } while (++svp <= last);
15334 AvREAL_off(unreferenced);
15336 SvREFCNT_dec_NN(unreferenced);
15340 Perl_clone_params_del(CLONE_PARAMS *param)
15342 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15344 PerlInterpreter *const to = param->new_perl;
15346 PerlInterpreter *const was = PERL_GET_THX;
15348 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15354 SvREFCNT_dec(param->stashes);
15355 if (param->unreferenced)
15356 unreferenced_to_tmp_stack(param->unreferenced);
15366 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15369 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15370 does a dTHX; to get the context from thread local storage.
15371 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15372 a version that passes in my_perl. */
15373 PerlInterpreter *const was = PERL_GET_THX;
15374 CLONE_PARAMS *param;
15376 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15382 /* Given that we've set the context, we can do this unshared. */
15383 Newx(param, 1, CLONE_PARAMS);
15386 param->proto_perl = from;
15387 param->new_perl = to;
15388 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15389 AvREAL_off(param->stashes);
15390 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15398 #endif /* USE_ITHREADS */
15401 Perl_init_constants(pTHX)
15403 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15404 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15405 SvANY(&PL_sv_undef) = NULL;
15407 SvANY(&PL_sv_no) = new_XPVNV();
15408 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15409 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15410 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15413 SvANY(&PL_sv_yes) = new_XPVNV();
15414 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15415 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15416 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15419 SvPV_set(&PL_sv_no, (char*)PL_No);
15420 SvCUR_set(&PL_sv_no, 0);
15421 SvLEN_set(&PL_sv_no, 0);
15422 SvIV_set(&PL_sv_no, 0);
15423 SvNV_set(&PL_sv_no, 0);
15425 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15426 SvCUR_set(&PL_sv_yes, 1);
15427 SvLEN_set(&PL_sv_yes, 0);
15428 SvIV_set(&PL_sv_yes, 1);
15429 SvNV_set(&PL_sv_yes, 1);
15431 PadnamePV(&PL_padname_const) = (char *)PL_No;
15435 =head1 Unicode Support
15437 =for apidoc sv_recode_to_utf8
15439 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15440 of C<sv> is assumed to be octets in that encoding, and C<sv>
15441 will be converted into Unicode (and UTF-8).
15443 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15444 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15445 an C<Encode::XS> Encoding object, bad things will happen.
15446 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15448 The PV of C<sv> is returned.
15453 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15455 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15457 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15466 if (SvPADTMP(nsv)) {
15467 nsv = sv_newmortal();
15468 SvSetSV_nosteal(nsv, sv);
15477 Passing sv_yes is wrong - it needs to be or'ed set of constants
15478 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15479 remove converted chars from source.
15481 Both will default the value - let them.
15483 XPUSHs(&PL_sv_yes);
15486 call_method("decode", G_SCALAR);
15490 s = SvPV_const(uni, len);
15491 if (s != SvPVX_const(sv)) {
15492 SvGROW(sv, len + 1);
15493 Move(s, SvPVX(sv), len + 1, char);
15494 SvCUR_set(sv, len);
15499 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15500 /* clear pos and any utf8 cache */
15501 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15504 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15505 magic_setutf8(sv,mg); /* clear UTF8 cache */
15510 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15514 =for apidoc sv_cat_decode
15516 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15517 assumed to be octets in that encoding and decoding the input starts
15518 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15519 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15520 when the string C<tstr> appears in decoding output or the input ends on
15521 the PV of C<ssv>. The value which C<offset> points will be modified
15522 to the last input position on C<ssv>.
15524 Returns TRUE if the terminator was found, else returns FALSE.
15529 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15530 SV *ssv, int *offset, char *tstr, int tlen)
15534 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15536 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15547 offsv = newSViv(*offset);
15549 mPUSHp(tstr, tlen);
15551 call_method("cat_decode", G_SCALAR);
15553 ret = SvTRUE(TOPs);
15554 *offset = SvIV(offsv);
15560 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15565 /* ---------------------------------------------------------------------
15567 * support functions for report_uninit()
15570 /* the maxiumum size of array or hash where we will scan looking
15571 * for the undefined element that triggered the warning */
15573 #define FUV_MAX_SEARCH_SIZE 1000
15575 /* Look for an entry in the hash whose value has the same SV as val;
15576 * If so, return a mortal copy of the key. */
15579 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15585 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15587 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15588 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15591 array = HvARRAY(hv);
15593 for (i=HvMAX(hv); i>=0; i--) {
15595 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15596 if (HeVAL(entry) != val)
15598 if ( HeVAL(entry) == &PL_sv_undef ||
15599 HeVAL(entry) == &PL_sv_placeholder)
15603 if (HeKLEN(entry) == HEf_SVKEY)
15604 return sv_mortalcopy(HeKEY_sv(entry));
15605 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15611 /* Look for an entry in the array whose value has the same SV as val;
15612 * If so, return the index, otherwise return -1. */
15615 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15617 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15619 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15620 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15623 if (val != &PL_sv_undef) {
15624 SV ** const svp = AvARRAY(av);
15627 for (i=AvFILLp(av); i>=0; i--)
15634 /* varname(): return the name of a variable, optionally with a subscript.
15635 * If gv is non-zero, use the name of that global, along with gvtype (one
15636 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15637 * targ. Depending on the value of the subscript_type flag, return:
15640 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15641 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15642 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15643 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15646 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15647 const SV *const keyname, I32 aindex, int subscript_type)
15650 SV * const name = sv_newmortal();
15651 if (gv && isGV(gv)) {
15653 buffer[0] = gvtype;
15656 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15658 gv_fullname4(name, gv, buffer, 0);
15660 if ((unsigned int)SvPVX(name)[1] <= 26) {
15662 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15664 /* Swap the 1 unprintable control character for the 2 byte pretty
15665 version - ie substr($name, 1, 1) = $buffer; */
15666 sv_insert(name, 1, 1, buffer, 2);
15670 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15673 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15675 if (!cv || !CvPADLIST(cv))
15677 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15678 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15682 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15683 SV * const sv = newSV(0);
15684 *SvPVX(name) = '$';
15685 Perl_sv_catpvf(aTHX_ name, "{%s}",
15686 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15687 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15688 SvREFCNT_dec_NN(sv);
15690 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15691 *SvPVX(name) = '$';
15692 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15694 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15695 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15696 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15704 =for apidoc find_uninit_var
15706 Find the name of the undefined variable (if any) that caused the operator
15707 to issue a "Use of uninitialized value" warning.
15708 If match is true, only return a name if its value matches C<uninit_sv>.
15709 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15710 warning, then following the direct child of the op may yield an
15711 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15712 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15713 the variable name if we get an exact match.
15714 C<desc_p> points to a string pointer holding the description of the op.
15715 This may be updated if needed.
15717 The name is returned as a mortal SV.
15719 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15720 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15726 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15727 bool match, const char **desc_p)
15732 const OP *o, *o2, *kid;
15734 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15736 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15737 uninit_sv == &PL_sv_placeholder)))
15740 switch (obase->op_type) {
15747 const bool pad = ( obase->op_type == OP_PADAV
15748 || obase->op_type == OP_PADHV
15749 || obase->op_type == OP_PADRANGE
15752 const bool hash = ( obase->op_type == OP_PADHV
15753 || obase->op_type == OP_RV2HV
15754 || (obase->op_type == OP_PADRANGE
15755 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15759 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15761 if (pad) { /* @lex, %lex */
15762 sv = PAD_SVl(obase->op_targ);
15766 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15767 /* @global, %global */
15768 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15771 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15773 else if (obase == PL_op) /* @{expr}, %{expr} */
15774 return find_uninit_var(cUNOPx(obase)->op_first,
15775 uninit_sv, match, desc_p);
15776 else /* @{expr}, %{expr} as a sub-expression */
15780 /* attempt to find a match within the aggregate */
15782 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15784 subscript_type = FUV_SUBSCRIPT_HASH;
15787 index = find_array_subscript((const AV *)sv, uninit_sv);
15789 subscript_type = FUV_SUBSCRIPT_ARRAY;
15792 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15795 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15796 keysv, index, subscript_type);
15800 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15802 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15803 if (!gv || !GvSTASH(gv))
15805 if (match && (GvSV(gv) != uninit_sv))
15807 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15810 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15813 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15815 return varname(NULL, '$', obase->op_targ,
15816 NULL, 0, FUV_SUBSCRIPT_NONE);
15819 gv = cGVOPx_gv(obase);
15820 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15822 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15824 case OP_AELEMFAST_LEX:
15827 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15828 if (!av || SvRMAGICAL(av))
15830 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15831 if (!svp || *svp != uninit_sv)
15834 return varname(NULL, '$', obase->op_targ,
15835 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15838 gv = cGVOPx_gv(obase);
15843 AV *const av = GvAV(gv);
15844 if (!av || SvRMAGICAL(av))
15846 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15847 if (!svp || *svp != uninit_sv)
15850 return varname(gv, '$', 0,
15851 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15853 NOT_REACHED; /* NOTREACHED */
15856 o = cUNOPx(obase)->op_first;
15857 if (!o || o->op_type != OP_NULL ||
15858 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15860 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15865 bool negate = FALSE;
15867 if (PL_op == obase)
15868 /* $a[uninit_expr] or $h{uninit_expr} */
15869 return find_uninit_var(cBINOPx(obase)->op_last,
15870 uninit_sv, match, desc_p);
15873 o = cBINOPx(obase)->op_first;
15874 kid = cBINOPx(obase)->op_last;
15876 /* get the av or hv, and optionally the gv */
15878 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15879 sv = PAD_SV(o->op_targ);
15881 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15882 && cUNOPo->op_first->op_type == OP_GV)
15884 gv = cGVOPx_gv(cUNOPo->op_first);
15888 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15893 if (kid && kid->op_type == OP_NEGATE) {
15895 kid = cUNOPx(kid)->op_first;
15898 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15899 /* index is constant */
15902 kidsv = newSVpvs_flags("-", SVs_TEMP);
15903 sv_catsv(kidsv, cSVOPx_sv(kid));
15906 kidsv = cSVOPx_sv(kid);
15910 if (obase->op_type == OP_HELEM) {
15911 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15912 if (!he || HeVAL(he) != uninit_sv)
15916 SV * const opsv = cSVOPx_sv(kid);
15917 const IV opsviv = SvIV(opsv);
15918 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15919 negate ? - opsviv : opsviv,
15921 if (!svp || *svp != uninit_sv)
15925 if (obase->op_type == OP_HELEM)
15926 return varname(gv, '%', o->op_targ,
15927 kidsv, 0, FUV_SUBSCRIPT_HASH);
15929 return varname(gv, '@', o->op_targ, NULL,
15930 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15931 FUV_SUBSCRIPT_ARRAY);
15934 /* index is an expression;
15935 * attempt to find a match within the aggregate */
15936 if (obase->op_type == OP_HELEM) {
15937 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15939 return varname(gv, '%', o->op_targ,
15940 keysv, 0, FUV_SUBSCRIPT_HASH);
15944 = find_array_subscript((const AV *)sv, uninit_sv);
15946 return varname(gv, '@', o->op_targ,
15947 NULL, index, FUV_SUBSCRIPT_ARRAY);
15952 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15954 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15956 NOT_REACHED; /* NOTREACHED */
15959 case OP_MULTIDEREF: {
15960 /* If we were executing OP_MULTIDEREF when the undef warning
15961 * triggered, then it must be one of the index values within
15962 * that triggered it. If not, then the only possibility is that
15963 * the value retrieved by the last aggregate lookup might be the
15964 * culprit. For the former, we set PL_multideref_pc each time before
15965 * using an index, so work though the item list until we reach
15966 * that point. For the latter, just work through the entire item
15967 * list; the last aggregate retrieved will be the candidate.
15970 /* the named aggregate, if any */
15971 PADOFFSET agg_targ = 0;
15973 /* the last-seen index */
15975 PADOFFSET index_targ;
15977 IV index_const_iv = 0; /* init for spurious compiler warn */
15978 SV *index_const_sv;
15979 int depth = 0; /* how many array/hash lookups we've done */
15981 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15982 UNOP_AUX_item *last = NULL;
15983 UV actions = items->uv;
15986 if (PL_op == obase) {
15987 last = PL_multideref_pc;
15988 assert(last >= items && last <= items + items[-1].uv);
15995 switch (actions & MDEREF_ACTION_MASK) {
15997 case MDEREF_reload:
15998 actions = (++items)->uv;
16001 case MDEREF_HV_padhv_helem: /* $lex{...} */
16004 case MDEREF_AV_padav_aelem: /* $lex[...] */
16005 agg_targ = (++items)->pad_offset;
16009 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16012 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16014 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16015 assert(isGV_with_GP(agg_gv));
16018 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16019 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16022 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16023 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16029 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16030 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16033 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16034 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16041 index_const_sv = NULL;
16043 index_type = (actions & MDEREF_INDEX_MASK);
16044 switch (index_type) {
16045 case MDEREF_INDEX_none:
16047 case MDEREF_INDEX_const:
16049 index_const_sv = UNOP_AUX_item_sv(++items)
16051 index_const_iv = (++items)->iv;
16053 case MDEREF_INDEX_padsv:
16054 index_targ = (++items)->pad_offset;
16056 case MDEREF_INDEX_gvsv:
16057 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16058 assert(isGV_with_GP(index_gv));
16062 if (index_type != MDEREF_INDEX_none)
16065 if ( index_type == MDEREF_INDEX_none
16066 || (actions & MDEREF_FLAG_last)
16067 || (last && items == last)
16071 actions >>= MDEREF_SHIFT;
16074 if (PL_op == obase) {
16075 /* index was undef */
16077 *desc_p = ( (actions & MDEREF_FLAG_last)
16078 && (obase->op_private
16079 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16081 (obase->op_private & OPpMULTIDEREF_EXISTS)
16084 : is_hv ? "hash element" : "array element";
16085 assert(index_type != MDEREF_INDEX_none);
16087 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16089 return varname(NULL, '$', index_targ,
16090 NULL, 0, FUV_SUBSCRIPT_NONE);
16091 assert(is_hv); /* AV index is an IV and can't be undef */
16092 /* can a const HV index ever be undef? */
16096 /* the SV returned by pp_multideref() was undef, if anything was */
16102 sv = PAD_SV(agg_targ);
16104 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16108 if (index_type == MDEREF_INDEX_const) {
16113 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16114 if (!he || HeVAL(he) != uninit_sv)
16118 SV * const * const svp =
16119 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16120 if (!svp || *svp != uninit_sv)
16125 ? varname(agg_gv, '%', agg_targ,
16126 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16127 : varname(agg_gv, '@', agg_targ,
16128 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16131 /* index is an var */
16133 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16135 return varname(agg_gv, '%', agg_targ,
16136 keysv, 0, FUV_SUBSCRIPT_HASH);
16140 = find_array_subscript((const AV *)sv, uninit_sv);
16142 return varname(agg_gv, '@', agg_targ,
16143 NULL, index, FUV_SUBSCRIPT_ARRAY);
16147 return varname(agg_gv,
16149 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16151 NOT_REACHED; /* NOTREACHED */
16155 /* only examine RHS */
16156 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16160 o = cUNOPx(obase)->op_first;
16161 if ( o->op_type == OP_PUSHMARK
16162 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16166 if (!OpHAS_SIBLING(o)) {
16167 /* one-arg version of open is highly magical */
16169 if (o->op_type == OP_GV) { /* open FOO; */
16171 if (match && GvSV(gv) != uninit_sv)
16173 return varname(gv, '$', 0,
16174 NULL, 0, FUV_SUBSCRIPT_NONE);
16176 /* other possibilities not handled are:
16177 * open $x; or open my $x; should return '${*$x}'
16178 * open expr; should return '$'.expr ideally
16184 /* ops where $_ may be an implicit arg */
16189 if ( !(obase->op_flags & OPf_STACKED)) {
16190 if (uninit_sv == DEFSV)
16191 return newSVpvs_flags("$_", SVs_TEMP);
16192 else if (obase->op_targ
16193 && uninit_sv == PAD_SVl(obase->op_targ))
16194 return varname(NULL, '$', obase->op_targ, NULL, 0,
16195 FUV_SUBSCRIPT_NONE);
16202 match = 1; /* print etc can return undef on defined args */
16203 /* skip filehandle as it can't produce 'undef' warning */
16204 o = cUNOPx(obase)->op_first;
16205 if ((obase->op_flags & OPf_STACKED)
16207 ( o->op_type == OP_PUSHMARK
16208 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16209 o = OpSIBLING(OpSIBLING(o));
16213 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16214 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16216 /* the following ops are capable of returning PL_sv_undef even for
16217 * defined arg(s) */
16236 case OP_GETPEERNAME:
16284 case OP_SMARTMATCH:
16293 /* XXX tmp hack: these two may call an XS sub, and currently
16294 XS subs don't have a SUB entry on the context stack, so CV and
16295 pad determination goes wrong, and BAD things happen. So, just
16296 don't try to determine the value under those circumstances.
16297 Need a better fix at dome point. DAPM 11/2007 */
16303 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16304 if (gv && GvSV(gv) == uninit_sv)
16305 return newSVpvs_flags("$.", SVs_TEMP);
16310 /* def-ness of rval pos() is independent of the def-ness of its arg */
16311 if ( !(obase->op_flags & OPf_MOD))
16316 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16317 return newSVpvs_flags("${$/}", SVs_TEMP);
16322 if (!(obase->op_flags & OPf_KIDS))
16324 o = cUNOPx(obase)->op_first;
16330 /* This loop checks all the kid ops, skipping any that cannot pos-
16331 * sibly be responsible for the uninitialized value; i.e., defined
16332 * constants and ops that return nothing. If there is only one op
16333 * left that is not skipped, then we *know* it is responsible for
16334 * the uninitialized value. If there is more than one op left, we
16335 * have to look for an exact match in the while() loop below.
16336 * Note that we skip padrange, because the individual pad ops that
16337 * it replaced are still in the tree, so we work on them instead.
16340 for (kid=o; kid; kid = OpSIBLING(kid)) {
16341 const OPCODE type = kid->op_type;
16342 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16343 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16344 || (type == OP_PUSHMARK)
16345 || (type == OP_PADRANGE)
16349 if (o2) { /* more than one found */
16356 return find_uninit_var(o2, uninit_sv, match, desc_p);
16358 /* scan all args */
16360 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16372 =for apidoc report_uninit
16374 Print appropriate "Use of uninitialized variable" warning.
16380 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16382 const char *desc = NULL;
16383 SV* varname = NULL;
16386 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16389 if (uninit_sv && PL_curpad) {
16390 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16392 sv_insert(varname, 0, 0, " ", 1);
16395 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16396 /* we've reached the end of a sort block or sub,
16397 * and the uninit value is probably what that code returned */
16400 /* PL_warn_uninit_sv is constant */
16401 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16403 /* diag_listed_as: Use of uninitialized value%s */
16404 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16405 SVfARG(varname ? varname : &PL_sv_no),
16408 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16414 * ex: set ts=8 sts=4 sw=4 et: