3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%"UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referant = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referant = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referant will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referant;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1531 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533 Use the C<SvGROW> wrapper instead.
1538 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1541 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1545 PERL_ARGS_ASSERT_SV_GROW;
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1553 else if (SvOOK(sv)) { /* pv is offset? */
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1562 s = SvPVX_mutable(sv);
1565 #ifdef PERL_COPY_ON_WRITE
1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1567 * to store the COW count. So in general, allocate one more byte than
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
1570 * If the new size is a big power of two, don't bother: we assume the
1571 * caller wanted a nice 2^N sized block and will be annoyed at getting
1573 * Only increment if the allocation isn't MEM_SIZE_MAX,
1574 * otherwise it will wrap to 0.
1576 if ( (newlen < 0x1000 || (newlen & (newlen - 1)))
1577 && newlen != MEM_SIZE_MAX
1582 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1583 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586 if (newlen > SvLEN(sv)) { /* need more room? */
1587 STRLEN minlen = SvCUR(sv);
1588 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1589 if (newlen < minlen)
1591 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1593 /* Don't round up on the first allocation, as odds are pretty good that
1594 * the initial request is accurate as to what is really needed */
1596 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1597 if (rounded > newlen)
1601 if (SvLEN(sv) && s) {
1602 s = (char*)saferealloc(s, newlen);
1605 s = (char*)safemalloc(newlen);
1606 if (SvPVX_const(sv) && SvCUR(sv)) {
1607 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1611 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Do this here, do it once, do it right, and then we will never get
1613 called back into sv_grow() unless there really is some growing
1615 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1617 SvLEN_set(sv, newlen);
1624 =for apidoc sv_setiv
1626 Copies an integer into the given SV, upgrading first if necessary.
1627 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1633 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1635 PERL_ARGS_ASSERT_SV_SETIV;
1637 SV_CHECK_THINKFIRST_COW_DROP(sv);
1638 switch (SvTYPE(sv)) {
1641 sv_upgrade(sv, SVt_IV);
1644 sv_upgrade(sv, SVt_PVIV);
1648 if (!isGV_with_GP(sv))
1655 /* diag_listed_as: Can't coerce %s to %s in %s */
1656 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1661 (void)SvIOK_only(sv); /* validate number */
1667 =for apidoc sv_setiv_mg
1669 Like C<sv_setiv>, but also handles 'set' magic.
1675 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1677 PERL_ARGS_ASSERT_SV_SETIV_MG;
1684 =for apidoc sv_setuv
1686 Copies an unsigned integer into the given SV, upgrading first if necessary.
1687 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1693 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1695 PERL_ARGS_ASSERT_SV_SETUV;
1697 /* With the if statement to ensure that integers are stored as IVs whenever
1699 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1702 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1704 If you wish to remove the following if statement, so that this routine
1705 (and its callers) always return UVs, please benchmark to see what the
1706 effect is. Modern CPUs may be different. Or may not :-)
1708 if (u <= (UV)IV_MAX) {
1709 sv_setiv(sv, (IV)u);
1718 =for apidoc sv_setuv_mg
1720 Like C<sv_setuv>, but also handles 'set' magic.
1726 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1728 PERL_ARGS_ASSERT_SV_SETUV_MG;
1735 =for apidoc sv_setnv
1737 Copies a double into the given SV, upgrading first if necessary.
1738 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1744 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1746 PERL_ARGS_ASSERT_SV_SETNV;
1748 SV_CHECK_THINKFIRST_COW_DROP(sv);
1749 switch (SvTYPE(sv)) {
1752 sv_upgrade(sv, SVt_NV);
1756 sv_upgrade(sv, SVt_PVNV);
1760 if (!isGV_with_GP(sv))
1767 /* diag_listed_as: Can't coerce %s to %s in %s */
1768 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1774 (void)SvNOK_only(sv); /* validate number */
1779 =for apidoc sv_setnv_mg
1781 Like C<sv_setnv>, but also handles 'set' magic.
1787 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1789 PERL_ARGS_ASSERT_SV_SETNV_MG;
1795 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1796 * not incrementable warning display.
1797 * Originally part of S_not_a_number().
1798 * The return value may be != tmpbuf.
1802 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1805 PERL_ARGS_ASSERT_SV_DISPLAY;
1808 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1809 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1812 const char * const limit = tmpbuf + tmpbuf_size - 8;
1813 /* each *s can expand to 4 chars + "...\0",
1814 i.e. need room for 8 chars */
1816 const char *s = SvPVX_const(sv);
1817 const char * const end = s + SvCUR(sv);
1818 for ( ; s < end && d < limit; s++ ) {
1820 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1824 /* Map to ASCII "equivalent" of Latin1 */
1825 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1831 else if (ch == '\r') {
1835 else if (ch == '\f') {
1839 else if (ch == '\\') {
1843 else if (ch == '\0') {
1847 else if (isPRINT_LC(ch))
1866 /* Print an "isn't numeric" warning, using a cleaned-up,
1867 * printable version of the offending string
1871 S_not_a_number(pTHX_ SV *const sv)
1876 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1878 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1881 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1882 /* diag_listed_as: Argument "%s" isn't numeric%s */
1883 "Argument \"%s\" isn't numeric in %s", pv,
1886 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1887 /* diag_listed_as: Argument "%s" isn't numeric%s */
1888 "Argument \"%s\" isn't numeric", pv);
1892 S_not_incrementable(pTHX_ SV *const sv) {
1896 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1898 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1900 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1901 "Argument \"%s\" treated as 0 in increment (++)", pv);
1905 =for apidoc looks_like_number
1907 Test if the content of an SV looks like a number (or is a number).
1908 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1909 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1916 Perl_looks_like_number(pTHX_ SV *const sv)
1922 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1924 if (SvPOK(sv) || SvPOKp(sv)) {
1925 sbegin = SvPV_nomg_const(sv, len);
1928 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1929 numtype = grok_number(sbegin, len, NULL);
1930 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1934 S_glob_2number(pTHX_ GV * const gv)
1936 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1938 /* We know that all GVs stringify to something that is not-a-number,
1939 so no need to test that. */
1940 if (ckWARN(WARN_NUMERIC))
1942 SV *const buffer = sv_newmortal();
1943 gv_efullname3(buffer, gv, "*");
1944 not_a_number(buffer);
1946 /* We just want something true to return, so that S_sv_2iuv_common
1947 can tail call us and return true. */
1951 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1952 until proven guilty, assume that things are not that bad... */
1957 As 64 bit platforms often have an NV that doesn't preserve all bits of
1958 an IV (an assumption perl has been based on to date) it becomes necessary
1959 to remove the assumption that the NV always carries enough precision to
1960 recreate the IV whenever needed, and that the NV is the canonical form.
1961 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1962 precision as a side effect of conversion (which would lead to insanity
1963 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1964 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1965 where precision was lost, and IV/UV/NV slots that have a valid conversion
1966 which has lost no precision
1967 2) to ensure that if a numeric conversion to one form is requested that
1968 would lose precision, the precise conversion (or differently
1969 imprecise conversion) is also performed and cached, to prevent
1970 requests for different numeric formats on the same SV causing
1971 lossy conversion chains. (lossless conversion chains are perfectly
1976 SvIOKp is true if the IV slot contains a valid value
1977 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1978 SvNOKp is true if the NV slot contains a valid value
1979 SvNOK is true only if the NV value is accurate
1982 while converting from PV to NV, check to see if converting that NV to an
1983 IV(or UV) would lose accuracy over a direct conversion from PV to
1984 IV(or UV). If it would, cache both conversions, return NV, but mark
1985 SV as IOK NOKp (ie not NOK).
1987 While converting from PV to IV, check to see if converting that IV to an
1988 NV would lose accuracy over a direct conversion from PV to NV. If it
1989 would, cache both conversions, flag similarly.
1991 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1992 correctly because if IV & NV were set NV *always* overruled.
1993 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1994 changes - now IV and NV together means that the two are interchangeable:
1995 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1997 The benefit of this is that operations such as pp_add know that if
1998 SvIOK is true for both left and right operands, then integer addition
1999 can be used instead of floating point (for cases where the result won't
2000 overflow). Before, floating point was always used, which could lead to
2001 loss of precision compared with integer addition.
2003 * making IV and NV equal status should make maths accurate on 64 bit
2005 * may speed up maths somewhat if pp_add and friends start to use
2006 integers when possible instead of fp. (Hopefully the overhead in
2007 looking for SvIOK and checking for overflow will not outweigh the
2008 fp to integer speedup)
2009 * will slow down integer operations (callers of SvIV) on "inaccurate"
2010 values, as the change from SvIOK to SvIOKp will cause a call into
2011 sv_2iv each time rather than a macro access direct to the IV slot
2012 * should speed up number->string conversion on integers as IV is
2013 favoured when IV and NV are equally accurate
2015 ####################################################################
2016 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2017 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2018 On the other hand, SvUOK is true iff UV.
2019 ####################################################################
2021 Your mileage will vary depending your CPU's relative fp to integer
2025 #ifndef NV_PRESERVES_UV
2026 # define IS_NUMBER_UNDERFLOW_IV 1
2027 # define IS_NUMBER_UNDERFLOW_UV 2
2028 # define IS_NUMBER_IV_AND_UV 2
2029 # define IS_NUMBER_OVERFLOW_IV 4
2030 # define IS_NUMBER_OVERFLOW_UV 5
2032 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2034 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2036 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2042 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2043 PERL_UNUSED_CONTEXT;
2045 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%"UVxf" NV=%"NVgf" inttype=%"UVXf"\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2046 if (SvNVX(sv) < (NV)IV_MIN) {
2047 (void)SvIOKp_on(sv);
2049 SvIV_set(sv, IV_MIN);
2050 return IS_NUMBER_UNDERFLOW_IV;
2052 if (SvNVX(sv) > (NV)UV_MAX) {
2053 (void)SvIOKp_on(sv);
2056 SvUV_set(sv, UV_MAX);
2057 return IS_NUMBER_OVERFLOW_UV;
2059 (void)SvIOKp_on(sv);
2061 /* Can't use strtol etc to convert this string. (See truth table in
2063 if (SvNVX(sv) <= (UV)IV_MAX) {
2064 SvIV_set(sv, I_V(SvNVX(sv)));
2065 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2066 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2068 /* Integer is imprecise. NOK, IOKp */
2070 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2073 SvUV_set(sv, U_V(SvNVX(sv)));
2074 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2075 if (SvUVX(sv) == UV_MAX) {
2076 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2077 possibly be preserved by NV. Hence, it must be overflow.
2079 return IS_NUMBER_OVERFLOW_UV;
2081 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2083 /* Integer is imprecise. NOK, IOKp */
2085 return IS_NUMBER_OVERFLOW_IV;
2087 #endif /* !NV_PRESERVES_UV*/
2089 /* If numtype is infnan, set the NV of the sv accordingly.
2090 * If numtype is anything else, try setting the NV using Atof(PV). */
2092 # pragma warning(push)
2093 # pragma warning(disable:4756;disable:4056)
2096 S_sv_setnv(pTHX_ SV* sv, int numtype)
2098 bool pok = cBOOL(SvPOK(sv));
2101 if ((numtype & IS_NUMBER_INFINITY)) {
2102 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2107 if ((numtype & IS_NUMBER_NAN)) {
2108 SvNV_set(sv, NV_NAN);
2113 SvNV_set(sv, Atof(SvPVX_const(sv)));
2114 /* Purposefully no true nok here, since we don't want to blow
2115 * away the possible IOK/UV of an existing sv. */
2118 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2120 SvPOK_on(sv); /* PV is okay, though. */
2124 # pragma warning(pop)
2128 S_sv_2iuv_common(pTHX_ SV *const sv)
2130 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2133 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2134 * without also getting a cached IV/UV from it at the same time
2135 * (ie PV->NV conversion should detect loss of accuracy and cache
2136 * IV or UV at same time to avoid this. */
2137 /* IV-over-UV optimisation - choose to cache IV if possible */
2139 if (SvTYPE(sv) == SVt_NV)
2140 sv_upgrade(sv, SVt_PVNV);
2142 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2143 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2144 certainly cast into the IV range at IV_MAX, whereas the correct
2145 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2147 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2148 if (Perl_isnan(SvNVX(sv))) {
2154 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2155 SvIV_set(sv, I_V(SvNVX(sv)));
2156 if (SvNVX(sv) == (NV) SvIVX(sv)
2157 #ifndef NV_PRESERVES_UV
2158 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2159 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2160 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2161 /* Don't flag it as "accurately an integer" if the number
2162 came from a (by definition imprecise) NV operation, and
2163 we're outside the range of NV integer precision */
2167 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2169 /* scalar has trailing garbage, eg "42a" */
2171 DEBUG_c(PerlIO_printf(Perl_debug_log,
2172 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2178 /* IV not precise. No need to convert from PV, as NV
2179 conversion would already have cached IV if it detected
2180 that PV->IV would be better than PV->NV->IV
2181 flags already correct - don't set public IOK. */
2182 DEBUG_c(PerlIO_printf(Perl_debug_log,
2183 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2188 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2189 but the cast (NV)IV_MIN rounds to a the value less (more
2190 negative) than IV_MIN which happens to be equal to SvNVX ??
2191 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2192 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2193 (NV)UVX == NVX are both true, but the values differ. :-(
2194 Hopefully for 2s complement IV_MIN is something like
2195 0x8000000000000000 which will be exact. NWC */
2198 SvUV_set(sv, U_V(SvNVX(sv)));
2200 (SvNVX(sv) == (NV) SvUVX(sv))
2201 #ifndef NV_PRESERVES_UV
2202 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2203 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2204 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2205 /* Don't flag it as "accurately an integer" if the number
2206 came from a (by definition imprecise) NV operation, and
2207 we're outside the range of NV integer precision */
2213 DEBUG_c(PerlIO_printf(Perl_debug_log,
2214 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2220 else if (SvPOKp(sv)) {
2223 const char *s = SvPVX_const(sv);
2224 const STRLEN cur = SvCUR(sv);
2226 /* short-cut for a single digit string like "1" */
2231 if (SvTYPE(sv) < SVt_PVIV)
2232 sv_upgrade(sv, SVt_PVIV);
2234 SvIV_set(sv, (IV)(c - '0'));
2239 numtype = grok_number(s, cur, &value);
2240 /* We want to avoid a possible problem when we cache an IV/ a UV which
2241 may be later translated to an NV, and the resulting NV is not
2242 the same as the direct translation of the initial string
2243 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2244 be careful to ensure that the value with the .456 is around if the
2245 NV value is requested in the future).
2247 This means that if we cache such an IV/a UV, we need to cache the
2248 NV as well. Moreover, we trade speed for space, and do not
2249 cache the NV if we are sure it's not needed.
2252 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2253 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2254 == IS_NUMBER_IN_UV) {
2255 /* It's definitely an integer, only upgrade to PVIV */
2256 if (SvTYPE(sv) < SVt_PVIV)
2257 sv_upgrade(sv, SVt_PVIV);
2259 } else if (SvTYPE(sv) < SVt_PVNV)
2260 sv_upgrade(sv, SVt_PVNV);
2262 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2263 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2265 S_sv_setnv(aTHX_ sv, numtype);
2269 /* If NVs preserve UVs then we only use the UV value if we know that
2270 we aren't going to call atof() below. If NVs don't preserve UVs
2271 then the value returned may have more precision than atof() will
2272 return, even though value isn't perfectly accurate. */
2273 if ((numtype & (IS_NUMBER_IN_UV
2274 #ifdef NV_PRESERVES_UV
2277 )) == IS_NUMBER_IN_UV) {
2278 /* This won't turn off the public IOK flag if it was set above */
2279 (void)SvIOKp_on(sv);
2281 if (!(numtype & IS_NUMBER_NEG)) {
2283 if (value <= (UV)IV_MAX) {
2284 SvIV_set(sv, (IV)value);
2286 /* it didn't overflow, and it was positive. */
2287 SvUV_set(sv, value);
2291 /* 2s complement assumption */
2292 if (value <= (UV)IV_MIN) {
2293 SvIV_set(sv, value == (UV)IV_MIN
2294 ? IV_MIN : -(IV)value);
2296 /* Too negative for an IV. This is a double upgrade, but
2297 I'm assuming it will be rare. */
2298 if (SvTYPE(sv) < SVt_PVNV)
2299 sv_upgrade(sv, SVt_PVNV);
2303 SvNV_set(sv, -(NV)value);
2304 SvIV_set(sv, IV_MIN);
2308 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2309 will be in the previous block to set the IV slot, and the next
2310 block to set the NV slot. So no else here. */
2312 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2313 != IS_NUMBER_IN_UV) {
2314 /* It wasn't an (integer that doesn't overflow the UV). */
2315 S_sv_setnv(aTHX_ sv, numtype);
2317 if (! numtype && ckWARN(WARN_NUMERIC))
2320 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2321 PTR2UV(sv), SvNVX(sv)));
2323 #ifdef NV_PRESERVES_UV
2324 (void)SvIOKp_on(sv);
2326 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2327 if (Perl_isnan(SvNVX(sv))) {
2333 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2334 SvIV_set(sv, I_V(SvNVX(sv)));
2335 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2338 NOOP; /* Integer is imprecise. NOK, IOKp */
2340 /* UV will not work better than IV */
2342 if (SvNVX(sv) > (NV)UV_MAX) {
2344 /* Integer is inaccurate. NOK, IOKp, is UV */
2345 SvUV_set(sv, UV_MAX);
2347 SvUV_set(sv, U_V(SvNVX(sv)));
2348 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2349 NV preservse UV so can do correct comparison. */
2350 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2353 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2358 #else /* NV_PRESERVES_UV */
2359 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2360 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2361 /* The IV/UV slot will have been set from value returned by
2362 grok_number above. The NV slot has just been set using
2365 assert (SvIOKp(sv));
2367 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2368 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2369 /* Small enough to preserve all bits. */
2370 (void)SvIOKp_on(sv);
2372 SvIV_set(sv, I_V(SvNVX(sv)));
2373 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2375 /* Assumption: first non-preserved integer is < IV_MAX,
2376 this NV is in the preserved range, therefore: */
2377 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2379 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2383 0 0 already failed to read UV.
2384 0 1 already failed to read UV.
2385 1 0 you won't get here in this case. IV/UV
2386 slot set, public IOK, Atof() unneeded.
2387 1 1 already read UV.
2388 so there's no point in sv_2iuv_non_preserve() attempting
2389 to use atol, strtol, strtoul etc. */
2391 sv_2iuv_non_preserve (sv, numtype);
2393 sv_2iuv_non_preserve (sv);
2397 #endif /* NV_PRESERVES_UV */
2398 /* It might be more code efficient to go through the entire logic above
2399 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2400 gets complex and potentially buggy, so more programmer efficient
2401 to do it this way, by turning off the public flags: */
2403 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2407 if (isGV_with_GP(sv))
2408 return glob_2number(MUTABLE_GV(sv));
2410 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2412 if (SvTYPE(sv) < SVt_IV)
2413 /* Typically the caller expects that sv_any is not NULL now. */
2414 sv_upgrade(sv, SVt_IV);
2415 /* Return 0 from the caller. */
2422 =for apidoc sv_2iv_flags
2424 Return the integer value of an SV, doing any necessary string
2425 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2426 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2432 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2434 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2436 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2437 && SvTYPE(sv) != SVt_PVFM);
2439 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2445 if (flags & SV_SKIP_OVERLOAD)
2447 tmpstr = AMG_CALLunary(sv, numer_amg);
2448 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2449 return SvIV(tmpstr);
2452 return PTR2IV(SvRV(sv));
2455 if (SvVALID(sv) || isREGEXP(sv)) {
2456 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2457 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2458 In practice they are extremely unlikely to actually get anywhere
2459 accessible by user Perl code - the only way that I'm aware of is when
2460 a constant subroutine which is used as the second argument to index.
2462 Regexps have no SvIVX and SvNVX fields.
2464 assert(isREGEXP(sv) || SvPOKp(sv));
2467 const char * const ptr =
2468 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2470 = grok_number(ptr, SvCUR(sv), &value);
2472 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2473 == IS_NUMBER_IN_UV) {
2474 /* It's definitely an integer */
2475 if (numtype & IS_NUMBER_NEG) {
2476 if (value < (UV)IV_MIN)
2479 if (value < (UV)IV_MAX)
2484 /* Quite wrong but no good choices. */
2485 if ((numtype & IS_NUMBER_INFINITY)) {
2486 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2487 } else if ((numtype & IS_NUMBER_NAN)) {
2488 return 0; /* So wrong. */
2492 if (ckWARN(WARN_NUMERIC))
2495 return I_V(Atof(ptr));
2499 if (SvTHINKFIRST(sv)) {
2500 if (SvREADONLY(sv) && !SvOK(sv)) {
2501 if (ckWARN(WARN_UNINITIALIZED))
2508 if (S_sv_2iuv_common(aTHX_ sv))
2512 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2513 PTR2UV(sv),SvIVX(sv)));
2514 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2518 =for apidoc sv_2uv_flags
2520 Return the unsigned integer value of an SV, doing any necessary string
2521 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2522 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2528 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2530 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2532 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2538 if (flags & SV_SKIP_OVERLOAD)
2540 tmpstr = AMG_CALLunary(sv, numer_amg);
2541 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2542 return SvUV(tmpstr);
2545 return PTR2UV(SvRV(sv));
2548 if (SvVALID(sv) || isREGEXP(sv)) {
2549 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2550 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2551 Regexps have no SvIVX and SvNVX fields. */
2552 assert(isREGEXP(sv) || SvPOKp(sv));
2555 const char * const ptr =
2556 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2558 = grok_number(ptr, SvCUR(sv), &value);
2560 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2561 == IS_NUMBER_IN_UV) {
2562 /* It's definitely an integer */
2563 if (!(numtype & IS_NUMBER_NEG))
2567 /* Quite wrong but no good choices. */
2568 if ((numtype & IS_NUMBER_INFINITY)) {
2569 return UV_MAX; /* So wrong. */
2570 } else if ((numtype & IS_NUMBER_NAN)) {
2571 return 0; /* So wrong. */
2575 if (ckWARN(WARN_NUMERIC))
2578 return U_V(Atof(ptr));
2582 if (SvTHINKFIRST(sv)) {
2583 if (SvREADONLY(sv) && !SvOK(sv)) {
2584 if (ckWARN(WARN_UNINITIALIZED))
2591 if (S_sv_2iuv_common(aTHX_ sv))
2595 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2596 PTR2UV(sv),SvUVX(sv)));
2597 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2601 =for apidoc sv_2nv_flags
2603 Return the num value of an SV, doing any necessary string or integer
2604 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2605 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2611 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2613 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2615 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2616 && SvTYPE(sv) != SVt_PVFM);
2617 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2618 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2619 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2620 Regexps have no SvIVX and SvNVX fields. */
2622 if (flags & SV_GMAGIC)
2626 if (SvPOKp(sv) && !SvIOKp(sv)) {
2627 ptr = SvPVX_const(sv);
2629 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2630 !grok_number(ptr, SvCUR(sv), NULL))
2636 return (NV)SvUVX(sv);
2638 return (NV)SvIVX(sv);
2644 ptr = RX_WRAPPED((REGEXP *)sv);
2647 assert(SvTYPE(sv) >= SVt_PVMG);
2648 /* This falls through to the report_uninit near the end of the
2650 } else if (SvTHINKFIRST(sv)) {
2655 if (flags & SV_SKIP_OVERLOAD)
2657 tmpstr = AMG_CALLunary(sv, numer_amg);
2658 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2659 return SvNV(tmpstr);
2662 return PTR2NV(SvRV(sv));
2664 if (SvREADONLY(sv) && !SvOK(sv)) {
2665 if (ckWARN(WARN_UNINITIALIZED))
2670 if (SvTYPE(sv) < SVt_NV) {
2671 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2672 sv_upgrade(sv, SVt_NV);
2674 STORE_NUMERIC_LOCAL_SET_STANDARD();
2675 PerlIO_printf(Perl_debug_log,
2676 "0x%"UVxf" num(%" NVgf ")\n",
2677 PTR2UV(sv), SvNVX(sv));
2678 RESTORE_NUMERIC_LOCAL();
2681 else if (SvTYPE(sv) < SVt_PVNV)
2682 sv_upgrade(sv, SVt_PVNV);
2687 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2688 #ifdef NV_PRESERVES_UV
2694 /* Only set the public NV OK flag if this NV preserves the IV */
2695 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2697 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2698 : (SvIVX(sv) == I_V(SvNVX(sv))))
2704 else if (SvPOKp(sv)) {
2706 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2707 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2709 #ifdef NV_PRESERVES_UV
2710 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2711 == IS_NUMBER_IN_UV) {
2712 /* It's definitely an integer */
2713 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2715 S_sv_setnv(aTHX_ sv, numtype);
2722 SvNV_set(sv, Atof(SvPVX_const(sv)));
2723 /* Only set the public NV OK flag if this NV preserves the value in
2724 the PV at least as well as an IV/UV would.
2725 Not sure how to do this 100% reliably. */
2726 /* if that shift count is out of range then Configure's test is
2727 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2729 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2730 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2731 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2732 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2733 /* Can't use strtol etc to convert this string, so don't try.
2734 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2737 /* value has been set. It may not be precise. */
2738 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2739 /* 2s complement assumption for (UV)IV_MIN */
2740 SvNOK_on(sv); /* Integer is too negative. */
2745 if (numtype & IS_NUMBER_NEG) {
2746 /* -IV_MIN is undefined, but we should never reach
2747 * this point with both IS_NUMBER_NEG and value ==
2749 assert(value != (UV)IV_MIN);
2750 SvIV_set(sv, -(IV)value);
2751 } else if (value <= (UV)IV_MAX) {
2752 SvIV_set(sv, (IV)value);
2754 SvUV_set(sv, value);
2758 if (numtype & IS_NUMBER_NOT_INT) {
2759 /* I believe that even if the original PV had decimals,
2760 they are lost beyond the limit of the FP precision.
2761 However, neither is canonical, so both only get p
2762 flags. NWC, 2000/11/25 */
2763 /* Both already have p flags, so do nothing */
2765 const NV nv = SvNVX(sv);
2766 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2767 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2768 if (SvIVX(sv) == I_V(nv)) {
2771 /* It had no "." so it must be integer. */
2775 /* between IV_MAX and NV(UV_MAX).
2776 Could be slightly > UV_MAX */
2778 if (numtype & IS_NUMBER_NOT_INT) {
2779 /* UV and NV both imprecise. */
2781 const UV nv_as_uv = U_V(nv);
2783 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2792 /* It might be more code efficient to go through the entire logic above
2793 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2794 gets complex and potentially buggy, so more programmer efficient
2795 to do it this way, by turning off the public flags: */
2797 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2798 #endif /* NV_PRESERVES_UV */
2801 if (isGV_with_GP(sv)) {
2802 glob_2number(MUTABLE_GV(sv));
2806 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2808 assert (SvTYPE(sv) >= SVt_NV);
2809 /* Typically the caller expects that sv_any is not NULL now. */
2810 /* XXX Ilya implies that this is a bug in callers that assume this
2811 and ideally should be fixed. */
2815 STORE_NUMERIC_LOCAL_SET_STANDARD();
2816 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2817 PTR2UV(sv), SvNVX(sv));
2818 RESTORE_NUMERIC_LOCAL();
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2834 Perl_sv_2num(pTHX_ SV *const sv)
2836 PERL_ARGS_ASSERT_SV_2NUM;
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2849 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2850 * UV as a string towards the end of buf, and return pointers to start and
2853 * We assume that buf is at least TYPE_CHARS(UV) long.
2857 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2859 char *ptr = buf + TYPE_CHARS(UV);
2860 char * const ebuf = ptr;
2863 PERL_ARGS_ASSERT_UIV_2BUF;
2871 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2875 *--ptr = '0' + (char)(uv % 10);
2883 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2884 * infinity or a not-a-number, writes the appropriate strings to the
2885 * buffer, including a zero byte. On success returns the written length,
2886 * excluding the zero byte, on failure (not an infinity, not a nan)
2887 * returns zero, assert-fails on maxlen being too short.
2889 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2890 * shared string constants we point to, instead of generating a new
2891 * string for each instance. */
2893 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2895 assert(maxlen >= 4);
2896 if (Perl_isinf(nv)) {
2898 if (maxlen < 5) /* "-Inf\0" */
2908 else if (Perl_isnan(nv)) {
2912 /* XXX optionally output the payload mantissa bits as
2913 * "(unsigned)" (to match the nan("...") C99 function,
2914 * or maybe as "(0xhhh...)" would make more sense...
2915 * provide a format string so that the user can decide?
2916 * NOTE: would affect the maxlen and assert() logic.*/
2921 assert((s == buffer + 3) || (s == buffer + 4));
2923 return s - buffer - 1; /* -1: excluding the zero byte */
2927 =for apidoc sv_2pv_flags
2929 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2930 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2931 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2932 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2938 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2942 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2944 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2945 && SvTYPE(sv) != SVt_PVFM);
2946 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2951 if (flags & SV_SKIP_OVERLOAD)
2953 tmpstr = AMG_CALLunary(sv, string_amg);
2954 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2955 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2957 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2961 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2962 if (flags & SV_CONST_RETURN) {
2963 pv = (char *) SvPVX_const(tmpstr);
2965 pv = (flags & SV_MUTABLE_RETURN)
2966 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2969 *lp = SvCUR(tmpstr);
2971 pv = sv_2pv_flags(tmpstr, lp, flags);
2984 SV *const referent = SvRV(sv);
2988 retval = buffer = savepvn("NULLREF", len);
2989 } else if (SvTYPE(referent) == SVt_REGEXP &&
2990 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2991 amagic_is_enabled(string_amg))) {
2992 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2996 /* If the regex is UTF-8 we want the containing scalar to
2997 have an UTF-8 flag too */
3004 *lp = RX_WRAPLEN(re);
3006 return RX_WRAPPED(re);
3008 const char *const typestr = sv_reftype(referent, 0);
3009 const STRLEN typelen = strlen(typestr);
3010 UV addr = PTR2UV(referent);
3011 const char *stashname = NULL;
3012 STRLEN stashnamelen = 0; /* hush, gcc */
3013 const char *buffer_end;
3015 if (SvOBJECT(referent)) {
3016 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3019 stashname = HEK_KEY(name);
3020 stashnamelen = HEK_LEN(name);
3022 if (HEK_UTF8(name)) {
3028 stashname = "__ANON__";
3031 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3032 + 2 * sizeof(UV) + 2 /* )\0 */;
3034 len = typelen + 3 /* (0x */
3035 + 2 * sizeof(UV) + 2 /* )\0 */;
3038 Newx(buffer, len, char);
3039 buffer_end = retval = buffer + len;
3041 /* Working backwards */
3045 *--retval = PL_hexdigit[addr & 15];
3046 } while (addr >>= 4);
3052 memcpy(retval, typestr, typelen);
3056 retval -= stashnamelen;
3057 memcpy(retval, stashname, stashnamelen);
3059 /* retval may not necessarily have reached the start of the
3061 assert (retval >= buffer);
3063 len = buffer_end - retval - 1; /* -1 for that \0 */
3075 if (flags & SV_MUTABLE_RETURN)
3076 return SvPVX_mutable(sv);
3077 if (flags & SV_CONST_RETURN)
3078 return (char *)SvPVX_const(sv);
3083 /* I'm assuming that if both IV and NV are equally valid then
3084 converting the IV is going to be more efficient */
3085 const U32 isUIOK = SvIsUV(sv);
3086 char buf[TYPE_CHARS(UV)];
3090 if (SvTYPE(sv) < SVt_PVIV)
3091 sv_upgrade(sv, SVt_PVIV);
3092 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3094 /* inlined from sv_setpvn */
3095 s = SvGROW_mutable(sv, len + 1);
3096 Move(ptr, s, len, char);
3101 else if (SvNOK(sv)) {
3102 if (SvTYPE(sv) < SVt_PVNV)
3103 sv_upgrade(sv, SVt_PVNV);
3104 if (SvNVX(sv) == 0.0
3105 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3106 && !Perl_isnan(SvNVX(sv))
3109 s = SvGROW_mutable(sv, 2);
3114 STRLEN size = 5; /* "-Inf\0" */
3116 s = SvGROW_mutable(sv, size);
3117 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3123 /* some Xenix systems wipe out errno here */
3132 5 + /* exponent digits */
3136 s = SvGROW_mutable(sv, size);
3137 #ifndef USE_LOCALE_NUMERIC
3138 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3144 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3145 STORE_LC_NUMERIC_SET_TO_NEEDED();
3147 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3148 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3149 size += SvLEN(PL_numeric_radix_sv) - 1;
3150 s = SvGROW_mutable(sv, size);
3153 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3155 /* If the radix character is UTF-8, and actually is in the
3156 * output, turn on the UTF-8 flag for the scalar */
3158 && SvUTF8(PL_numeric_radix_sv)
3159 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3164 RESTORE_LC_NUMERIC();
3167 /* We don't call SvPOK_on(), because it may come to
3168 * pass that the locale changes so that the
3169 * stringification we just did is no longer correct. We
3170 * will have to re-stringify every time it is needed */
3177 else if (isGV_with_GP(sv)) {
3178 GV *const gv = MUTABLE_GV(sv);
3179 SV *const buffer = sv_newmortal();
3181 gv_efullname3(buffer, gv, "*");
3183 assert(SvPOK(buffer));
3187 *lp = SvCUR(buffer);
3188 return SvPVX(buffer);
3190 else if (isREGEXP(sv)) {
3191 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3192 return RX_WRAPPED((REGEXP *)sv);
3197 if (flags & SV_UNDEF_RETURNS_NULL)
3199 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3201 /* Typically the caller expects that sv_any is not NULL now. */
3202 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3203 sv_upgrade(sv, SVt_PV);
3208 const STRLEN len = s - SvPVX_const(sv);
3213 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3214 PTR2UV(sv),SvPVX_const(sv)));
3215 if (flags & SV_CONST_RETURN)
3216 return (char *)SvPVX_const(sv);
3217 if (flags & SV_MUTABLE_RETURN)
3218 return SvPVX_mutable(sv);
3223 =for apidoc sv_copypv
3225 Copies a stringified representation of the source SV into the
3226 destination SV. Automatically performs any necessary C<mg_get> and
3227 coercion of numeric values into strings. Guaranteed to preserve
3228 C<UTF8> flag even from overloaded objects. Similar in nature to
3229 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3230 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3231 would lose the UTF-8'ness of the PV.
3233 =for apidoc sv_copypv_nomg
3235 Like C<sv_copypv>, but doesn't invoke get magic first.
3237 =for apidoc sv_copypv_flags
3239 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3240 has the C<SV_GMAGIC> bit set.
3246 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3251 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3253 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3254 sv_setpvn(dsv,s,len);
3262 =for apidoc sv_2pvbyte
3264 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3265 to its length. May cause the SV to be downgraded from UTF-8 as a
3268 Usually accessed via the C<SvPVbyte> macro.
3274 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3276 PERL_ARGS_ASSERT_SV_2PVBYTE;
3279 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3280 || isGV_with_GP(sv) || SvROK(sv)) {
3281 SV *sv2 = sv_newmortal();
3282 sv_copypv_nomg(sv2,sv);
3285 sv_utf8_downgrade(sv,0);
3286 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3290 =for apidoc sv_2pvutf8
3292 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3293 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3295 Usually accessed via the C<SvPVutf8> macro.
3301 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3303 PERL_ARGS_ASSERT_SV_2PVUTF8;
3305 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3306 || isGV_with_GP(sv) || SvROK(sv))
3307 sv = sv_mortalcopy(sv);
3310 sv_utf8_upgrade_nomg(sv);
3311 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3316 =for apidoc sv_2bool
3318 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3319 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3320 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3322 =for apidoc sv_2bool_flags
3324 This function is only used by C<sv_true()> and friends, and only if
3325 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3326 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3333 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3335 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3338 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3344 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3345 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3348 if(SvGMAGICAL(sv)) {
3350 goto restart; /* call sv_2bool */
3352 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3353 else if(!SvOK(sv)) {
3356 else if(SvPOK(sv)) {
3357 svb = SvPVXtrue(sv);
3359 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3360 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3361 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3365 goto restart; /* call sv_2bool_nomg */
3370 return SvRV(sv) != 0;
3374 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3375 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3379 =for apidoc sv_utf8_upgrade
3381 Converts the PV of an SV to its UTF-8-encoded form.
3382 Forces the SV to string form if it is not already.
3383 Will C<mg_get> on C<sv> if appropriate.
3384 Always sets the C<SvUTF8> flag to avoid future validity checks even
3385 if the whole string is the same in UTF-8 as not.
3386 Returns the number of bytes in the converted string
3388 This is not a general purpose byte encoding to Unicode interface:
3389 use the Encode extension for that.
3391 =for apidoc sv_utf8_upgrade_nomg
3393 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3395 =for apidoc sv_utf8_upgrade_flags
3397 Converts the PV of an SV to its UTF-8-encoded form.
3398 Forces the SV to string form if it is not already.
3399 Always sets the SvUTF8 flag to avoid future validity checks even
3400 if all the bytes are invariant in UTF-8.
3401 If C<flags> has C<SV_GMAGIC> bit set,
3402 will C<mg_get> on C<sv> if appropriate, else not.
3404 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3405 will expand when converted to UTF-8, and skips the extra work of checking for
3406 that. Typically this flag is used by a routine that has already parsed the
3407 string and found such characters, and passes this information on so that the
3408 work doesn't have to be repeated.
3410 Returns the number of bytes in the converted string.
3412 This is not a general purpose byte encoding to Unicode interface:
3413 use the Encode extension for that.
3415 =for apidoc sv_utf8_upgrade_flags_grow
3417 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3418 the number of unused bytes the string of C<sv> is guaranteed to have free after
3419 it upon return. This allows the caller to reserve extra space that it intends
3420 to fill, to avoid extra grows.
3422 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3423 are implemented in terms of this function.
3425 Returns the number of bytes in the converted string (not including the spares).
3429 (One might think that the calling routine could pass in the position of the
3430 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3431 have to be found again. But that is not the case, because typically when the
3432 caller is likely to use this flag, it won't be calling this routine unless it
3433 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3434 and just use bytes. But some things that do fit into a byte are variants in
3435 utf8, and the caller may not have been keeping track of these.)
3437 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3438 C<NUL> isn't guaranteed due to having other routines do the work in some input
3439 cases, or if the input is already flagged as being in utf8.
3441 The speed of this could perhaps be improved for many cases if someone wanted to
3442 write a fast function that counts the number of variant characters in a string,
3443 especially if it could return the position of the first one.
3448 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3450 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3452 if (sv == &PL_sv_undef)
3454 if (!SvPOK_nog(sv)) {
3456 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3457 (void) sv_2pv_flags(sv,&len, flags);
3459 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3463 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3468 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3473 S_sv_uncow(aTHX_ sv, 0);
3476 if (SvCUR(sv) == 0) {
3477 if (extra) SvGROW(sv, extra);
3478 } else { /* Assume Latin-1/EBCDIC */
3479 /* This function could be much more efficient if we
3480 * had a FLAG in SVs to signal if there are any variant
3481 * chars in the PV. Given that there isn't such a flag
3482 * make the loop as fast as possible (although there are certainly ways
3483 * to speed this up, eg. through vectorization) */
3484 U8 * s = (U8 *) SvPVX_const(sv);
3485 U8 * e = (U8 *) SvEND(sv);
3487 STRLEN two_byte_count = 0;
3489 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3491 /* See if really will need to convert to utf8. We mustn't rely on our
3492 * incoming SV being well formed and having a trailing '\0', as certain
3493 * code in pp_formline can send us partially built SVs. */
3497 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3499 t--; /* t already incremented; re-point to first variant */
3504 /* utf8 conversion not needed because all are invariants. Mark as
3505 * UTF-8 even if no variant - saves scanning loop */
3507 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3512 /* Here, the string should be converted to utf8, either because of an
3513 * input flag (two_byte_count = 0), or because a character that
3514 * requires 2 bytes was found (two_byte_count = 1). t points either to
3515 * the beginning of the string (if we didn't examine anything), or to
3516 * the first variant. In either case, everything from s to t - 1 will
3517 * occupy only 1 byte each on output.
3519 * There are two main ways to convert. One is to create a new string
3520 * and go through the input starting from the beginning, appending each
3521 * converted value onto the new string as we go along. It's probably
3522 * best to allocate enough space in the string for the worst possible
3523 * case rather than possibly running out of space and having to
3524 * reallocate and then copy what we've done so far. Since everything
3525 * from s to t - 1 is invariant, the destination can be initialized
3526 * with these using a fast memory copy
3528 * The other way is to figure out exactly how big the string should be
3529 * by parsing the entire input. Then you don't have to make it big
3530 * enough to handle the worst possible case, and more importantly, if
3531 * the string you already have is large enough, you don't have to
3532 * allocate a new string, you can copy the last character in the input
3533 * string to the final position(s) that will be occupied by the
3534 * converted string and go backwards, stopping at t, since everything
3535 * before that is invariant.
3537 * There are advantages and disadvantages to each method.
3539 * In the first method, we can allocate a new string, do the memory
3540 * copy from the s to t - 1, and then proceed through the rest of the
3541 * string byte-by-byte.
3543 * In the second method, we proceed through the rest of the input
3544 * string just calculating how big the converted string will be. Then
3545 * there are two cases:
3546 * 1) if the string has enough extra space to handle the converted
3547 * value. We go backwards through the string, converting until we
3548 * get to the position we are at now, and then stop. If this
3549 * position is far enough along in the string, this method is
3550 * faster than the other method. If the memory copy were the same
3551 * speed as the byte-by-byte loop, that position would be about
3552 * half-way, as at the half-way mark, parsing to the end and back
3553 * is one complete string's parse, the same amount as starting
3554 * over and going all the way through. Actually, it would be
3555 * somewhat less than half-way, as it's faster to just count bytes
3556 * than to also copy, and we don't have the overhead of allocating
3557 * a new string, changing the scalar to use it, and freeing the
3558 * existing one. But if the memory copy is fast, the break-even
3559 * point is somewhere after half way. The counting loop could be
3560 * sped up by vectorization, etc, to move the break-even point
3561 * further towards the beginning.
3562 * 2) if the string doesn't have enough space to handle the converted
3563 * value. A new string will have to be allocated, and one might
3564 * as well, given that, start from the beginning doing the first
3565 * method. We've spent extra time parsing the string and in
3566 * exchange all we've gotten is that we know precisely how big to
3567 * make the new one. Perl is more optimized for time than space,
3568 * so this case is a loser.
3569 * So what I've decided to do is not use the 2nd method unless it is
3570 * guaranteed that a new string won't have to be allocated, assuming
3571 * the worst case. I also decided not to put any more conditions on it
3572 * than this, for now. It seems likely that, since the worst case is
3573 * twice as big as the unknown portion of the string (plus 1), we won't
3574 * be guaranteed enough space, causing us to go to the first method,
3575 * unless the string is short, or the first variant character is near
3576 * the end of it. In either of these cases, it seems best to use the
3577 * 2nd method. The only circumstance I can think of where this would
3578 * be really slower is if the string had once had much more data in it
3579 * than it does now, but there is still a substantial amount in it */
3582 STRLEN invariant_head = t - s;
3583 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3584 if (SvLEN(sv) < size) {
3586 /* Here, have decided to allocate a new string */
3591 Newx(dst, size, U8);
3593 /* If no known invariants at the beginning of the input string,
3594 * set so starts from there. Otherwise, can use memory copy to
3595 * get up to where we are now, and then start from here */
3597 if (invariant_head == 0) {
3600 Copy(s, dst, invariant_head, char);
3601 d = dst + invariant_head;
3605 append_utf8_from_native_byte(*t, &d);
3609 SvPV_free(sv); /* No longer using pre-existing string */
3610 SvPV_set(sv, (char*)dst);
3611 SvCUR_set(sv, d - dst);
3612 SvLEN_set(sv, size);
3615 /* Here, have decided to get the exact size of the string.
3616 * Currently this happens only when we know that there is
3617 * guaranteed enough space to fit the converted string, so
3618 * don't have to worry about growing. If two_byte_count is 0,
3619 * then t points to the first byte of the string which hasn't
3620 * been examined yet. Otherwise two_byte_count is 1, and t
3621 * points to the first byte in the string that will expand to
3622 * two. Depending on this, start examining at t or 1 after t.
3625 U8 *d = t + two_byte_count;
3628 /* Count up the remaining bytes that expand to two */
3631 const U8 chr = *d++;
3632 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3635 /* The string will expand by just the number of bytes that
3636 * occupy two positions. But we are one afterwards because of
3637 * the increment just above. This is the place to put the
3638 * trailing NUL, and to set the length before we decrement */
3640 d += two_byte_count;
3641 SvCUR_set(sv, d - s);
3645 /* Having decremented d, it points to the position to put the
3646 * very last byte of the expanded string. Go backwards through
3647 * the string, copying and expanding as we go, stopping when we
3648 * get to the part that is invariant the rest of the way down */
3652 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3655 *d-- = UTF8_EIGHT_BIT_LO(*e);
3656 *d-- = UTF8_EIGHT_BIT_HI(*e);
3662 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3663 /* Update pos. We do it at the end rather than during
3664 * the upgrade, to avoid slowing down the common case
3665 * (upgrade without pos).
3666 * pos can be stored as either bytes or characters. Since
3667 * this was previously a byte string we can just turn off
3668 * the bytes flag. */
3669 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3671 mg->mg_flags &= ~MGf_BYTES;
3673 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3674 magic_setutf8(sv,mg); /* clear UTF8 cache */
3679 /* Mark as UTF-8 even if no variant - saves scanning loop */
3685 =for apidoc sv_utf8_downgrade
3687 Attempts to convert the PV of an SV from characters to bytes.
3688 If the PV contains a character that cannot fit
3689 in a byte, this conversion will fail;
3690 in this case, either returns false or, if C<fail_ok> is not
3693 This is not a general purpose Unicode to byte encoding interface:
3694 use the C<Encode> extension for that.
3700 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3702 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3704 if (SvPOKp(sv) && SvUTF8(sv)) {
3708 int mg_flags = SV_GMAGIC;
3711 S_sv_uncow(aTHX_ sv, 0);
3713 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3715 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3716 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3717 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3718 SV_GMAGIC|SV_CONST_RETURN);
3719 mg_flags = 0; /* sv_pos_b2u does get magic */
3721 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3722 magic_setutf8(sv,mg); /* clear UTF8 cache */
3725 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3727 if (!utf8_to_bytes(s, &len)) {
3732 Perl_croak(aTHX_ "Wide character in %s",
3735 Perl_croak(aTHX_ "Wide character");
3746 =for apidoc sv_utf8_encode
3748 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3749 flag off so that it looks like octets again.
3755 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3757 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3759 if (SvREADONLY(sv)) {
3760 sv_force_normal_flags(sv, 0);
3762 (void) sv_utf8_upgrade(sv);
3767 =for apidoc sv_utf8_decode
3769 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3770 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3771 so that it looks like a character. If the PV contains only single-byte
3772 characters, the C<SvUTF8> flag stays off.
3773 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3779 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3781 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3784 const U8 *start, *c;
3786 /* The octets may have got themselves encoded - get them back as
3789 if (!sv_utf8_downgrade(sv, TRUE))
3792 /* it is actually just a matter of turning the utf8 flag on, but
3793 * we want to make sure everything inside is valid utf8 first.
3795 c = start = (const U8 *) SvPVX_const(sv);
3796 if (!is_utf8_string(c, SvCUR(sv)))
3798 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3801 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3802 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3803 after this, clearing pos. Does anything on CPAN
3805 /* adjust pos to the start of a UTF8 char sequence */
3806 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3808 I32 pos = mg->mg_len;
3810 for (c = start + pos; c > start; c--) {
3811 if (UTF8_IS_START(*c))
3814 mg->mg_len = c - start;
3817 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3818 magic_setutf8(sv,mg); /* clear UTF8 cache */
3825 =for apidoc sv_setsv
3827 Copies the contents of the source SV C<ssv> into the destination SV
3828 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3829 function if the source SV needs to be reused. Does not handle 'set' magic on
3830 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3831 performs a copy-by-value, obliterating any previous content of the
3834 You probably want to use one of the assortment of wrappers, such as
3835 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3836 C<SvSetMagicSV_nosteal>.
3838 =for apidoc sv_setsv_flags
3840 Copies the contents of the source SV C<ssv> into the destination SV
3841 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3842 function if the source SV needs to be reused. Does not handle 'set' magic.
3843 Loosely speaking, it performs a copy-by-value, obliterating any previous
3844 content of the destination.
3845 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3846 C<ssv> if appropriate, else not. If the C<flags>
3847 parameter has the C<SV_NOSTEAL> bit set then the
3848 buffers of temps will not be stolen. C<sv_setsv>
3849 and C<sv_setsv_nomg> are implemented in terms of this function.
3851 You probably want to use one of the assortment of wrappers, such as
3852 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3853 C<SvSetMagicSV_nosteal>.
3855 This is the primary function for copying scalars, and most other
3856 copy-ish functions and macros use this underneath.
3862 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3864 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3865 HV *old_stash = NULL;
3867 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3869 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3870 const char * const name = GvNAME(sstr);
3871 const STRLEN len = GvNAMELEN(sstr);
3873 if (dtype >= SVt_PV) {
3879 SvUPGRADE(dstr, SVt_PVGV);
3880 (void)SvOK_off(dstr);
3881 isGV_with_GP_on(dstr);
3883 GvSTASH(dstr) = GvSTASH(sstr);
3885 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3886 gv_name_set(MUTABLE_GV(dstr), name, len,
3887 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3888 SvFAKE_on(dstr); /* can coerce to non-glob */
3891 if(GvGP(MUTABLE_GV(sstr))) {
3892 /* If source has method cache entry, clear it */
3894 SvREFCNT_dec(GvCV(sstr));
3895 GvCV_set(sstr, NULL);
3898 /* If source has a real method, then a method is
3901 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3907 /* If dest already had a real method, that's a change as well */
3909 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3910 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3915 /* We don't need to check the name of the destination if it was not a
3916 glob to begin with. */
3917 if(dtype == SVt_PVGV) {
3918 const char * const name = GvNAME((const GV *)dstr);
3921 /* The stash may have been detached from the symbol table, so
3923 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3927 const STRLEN len = GvNAMELEN(dstr);
3928 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3929 || (len == 1 && name[0] == ':')) {
3932 /* Set aside the old stash, so we can reset isa caches on
3934 if((old_stash = GvHV(dstr)))
3935 /* Make sure we do not lose it early. */
3936 SvREFCNT_inc_simple_void_NN(
3937 sv_2mortal((SV *)old_stash)
3942 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3945 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3946 * so temporarily protect it */
3948 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3949 gp_free(MUTABLE_GV(dstr));
3950 GvINTRO_off(dstr); /* one-shot flag */
3951 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3954 if (SvTAINTED(sstr))
3956 if (GvIMPORTED(dstr) != GVf_IMPORTED
3957 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3959 GvIMPORTED_on(dstr);
3962 if(mro_changes == 2) {
3963 if (GvAV((const GV *)sstr)) {
3965 SV * const sref = (SV *)GvAV((const GV *)dstr);
3966 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3967 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3968 AV * const ary = newAV();
3969 av_push(ary, mg->mg_obj); /* takes the refcount */
3970 mg->mg_obj = (SV *)ary;
3972 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3974 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3976 mro_isa_changed_in(GvSTASH(dstr));
3978 else if(mro_changes == 3) {
3979 HV * const stash = GvHV(dstr);
3980 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3986 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3987 if (GvIO(dstr) && dtype == SVt_PVGV) {
3988 DEBUG_o(Perl_deb(aTHX_
3989 "glob_assign_glob clearing PL_stashcache\n"));
3990 /* It's a cache. It will rebuild itself quite happily.
3991 It's a lot of effort to work out exactly which key (or keys)
3992 might be invalidated by the creation of the this file handle.
3994 hv_clear(PL_stashcache);
4000 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4002 SV * const sref = SvRV(sstr);
4004 const int intro = GvINTRO(dstr);
4007 const U32 stype = SvTYPE(sref);
4009 PERL_ARGS_ASSERT_GV_SETREF;
4012 GvINTRO_off(dstr); /* one-shot flag */
4013 GvLINE(dstr) = CopLINE(PL_curcop);
4014 GvEGV(dstr) = MUTABLE_GV(dstr);
4019 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4020 import_flag = GVf_IMPORTED_CV;
4023 location = (SV **) &GvHV(dstr);
4024 import_flag = GVf_IMPORTED_HV;
4027 location = (SV **) &GvAV(dstr);
4028 import_flag = GVf_IMPORTED_AV;
4031 location = (SV **) &GvIOp(dstr);
4034 location = (SV **) &GvFORM(dstr);
4037 location = &GvSV(dstr);
4038 import_flag = GVf_IMPORTED_SV;
4041 if (stype == SVt_PVCV) {
4042 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4043 if (GvCVGEN(dstr)) {
4044 SvREFCNT_dec(GvCV(dstr));
4045 GvCV_set(dstr, NULL);
4046 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4049 /* SAVEt_GVSLOT takes more room on the savestack and has more
4050 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4051 leave_scope needs access to the GV so it can reset method
4052 caches. We must use SAVEt_GVSLOT whenever the type is
4053 SVt_PVCV, even if the stash is anonymous, as the stash may
4054 gain a name somehow before leave_scope. */
4055 if (stype == SVt_PVCV) {
4056 /* There is no save_pushptrptrptr. Creating it for this
4057 one call site would be overkill. So inline the ss add
4061 SS_ADD_PTR(location);
4062 SS_ADD_PTR(SvREFCNT_inc(*location));
4063 SS_ADD_UV(SAVEt_GVSLOT);
4066 else SAVEGENERICSV(*location);
4069 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4070 CV* const cv = MUTABLE_CV(*location);
4072 if (!GvCVGEN((const GV *)dstr) &&
4073 (CvROOT(cv) || CvXSUB(cv)) &&
4074 /* redundant check that avoids creating the extra SV
4075 most of the time: */
4076 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4078 SV * const new_const_sv =
4079 CvCONST((const CV *)sref)
4080 ? cv_const_sv((const CV *)sref)
4082 HV * const stash = GvSTASH((const GV *)dstr);
4083 report_redefined_cv(
4086 ? Perl_newSVpvf(aTHX_
4088 HEKfARG(HvNAME_HEK(stash)),
4089 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4090 : Perl_newSVpvf(aTHX_
4092 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4095 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4099 cv_ckproto_len_flags(cv, (const GV *)dstr,
4100 SvPOK(sref) ? CvPROTO(sref) : NULL,
4101 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4102 SvPOK(sref) ? SvUTF8(sref) : 0);
4104 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4105 GvASSUMECV_on(dstr);
4106 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4107 if (intro && GvREFCNT(dstr) > 1) {
4108 /* temporary remove extra savestack's ref */
4110 gv_method_changed(dstr);
4113 else gv_method_changed(dstr);
4116 *location = SvREFCNT_inc_simple_NN(sref);
4117 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4118 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4119 GvFLAGS(dstr) |= import_flag;
4122 if (stype == SVt_PVHV) {
4123 const char * const name = GvNAME((GV*)dstr);
4124 const STRLEN len = GvNAMELEN(dstr);
4127 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4128 || (len == 1 && name[0] == ':')
4130 && (!dref || HvENAME_get(dref))
4133 (HV *)sref, (HV *)dref,
4139 stype == SVt_PVAV && sref != dref
4140 && strEQ(GvNAME((GV*)dstr), "ISA")
4141 /* The stash may have been detached from the symbol table, so
4142 check its name before doing anything. */
4143 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4146 MAGIC * const omg = dref && SvSMAGICAL(dref)
4147 ? mg_find(dref, PERL_MAGIC_isa)
4149 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4150 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4151 AV * const ary = newAV();
4152 av_push(ary, mg->mg_obj); /* takes the refcount */
4153 mg->mg_obj = (SV *)ary;
4156 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4157 SV **svp = AvARRAY((AV *)omg->mg_obj);
4158 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4162 SvREFCNT_inc_simple_NN(*svp++)
4168 SvREFCNT_inc_simple_NN(omg->mg_obj)
4172 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4178 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4180 for (i = 0; i <= AvFILL(sref); ++i) {
4181 SV **elem = av_fetch ((AV*)sref, i, 0);
4184 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4188 mg = mg_find(sref, PERL_MAGIC_isa);
4190 /* Since the *ISA assignment could have affected more than
4191 one stash, don't call mro_isa_changed_in directly, but let
4192 magic_clearisa do it for us, as it already has the logic for
4193 dealing with globs vs arrays of globs. */
4195 Perl_magic_clearisa(aTHX_ NULL, mg);
4197 else if (stype == SVt_PVIO) {
4198 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4199 /* It's a cache. It will rebuild itself quite happily.
4200 It's a lot of effort to work out exactly which key (or keys)
4201 might be invalidated by the creation of the this file handle.
4203 hv_clear(PL_stashcache);
4207 if (!intro) SvREFCNT_dec(dref);
4208 if (SvTAINTED(sstr))
4216 #ifdef PERL_DEBUG_READONLY_COW
4217 # include <sys/mman.h>
4219 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4220 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4224 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4226 struct perl_memory_debug_header * const header =
4227 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4228 const MEM_SIZE len = header->size;
4229 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4230 # ifdef PERL_TRACK_MEMPOOL
4231 if (!header->readonly) header->readonly = 1;
4233 if (mprotect(header, len, PROT_READ))
4234 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4235 header, len, errno);
4239 S_sv_buf_to_rw(pTHX_ SV *sv)
4241 struct perl_memory_debug_header * const header =
4242 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4243 const MEM_SIZE len = header->size;
4244 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4245 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4246 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4247 header, len, errno);
4248 # ifdef PERL_TRACK_MEMPOOL
4249 header->readonly = 0;
4254 # define sv_buf_to_ro(sv) NOOP
4255 # define sv_buf_to_rw(sv) NOOP
4259 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4264 unsigned int both_type;
4266 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4268 if (UNLIKELY( sstr == dstr ))
4271 if (UNLIKELY( !sstr ))
4272 sstr = &PL_sv_undef;
4274 stype = SvTYPE(sstr);
4275 dtype = SvTYPE(dstr);
4276 both_type = (stype | dtype);
4278 /* with these values, we can check that both SVs are NULL/IV (and not
4279 * freed) just by testing the or'ed types */
4280 STATIC_ASSERT_STMT(SVt_NULL == 0);
4281 STATIC_ASSERT_STMT(SVt_IV == 1);
4282 if (both_type <= 1) {
4283 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4288 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4289 if (SvREADONLY(dstr))
4290 Perl_croak_no_modify();
4292 sv_unref_flags(dstr, 0);
4294 assert(!SvGMAGICAL(sstr));
4295 assert(!SvGMAGICAL(dstr));
4297 sflags = SvFLAGS(sstr);
4298 if (sflags & (SVf_IOK|SVf_ROK)) {
4299 SET_SVANY_FOR_BODYLESS_IV(dstr);
4300 new_dflags = SVt_IV;
4302 if (sflags & SVf_ROK) {
4303 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4304 new_dflags |= SVf_ROK;
4307 /* both src and dst are <= SVt_IV, so sv_any points to the
4308 * head; so access the head directly
4310 assert( &(sstr->sv_u.svu_iv)
4311 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4312 assert( &(dstr->sv_u.svu_iv)
4313 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4314 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4315 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4319 new_dflags = dtype; /* turn off everything except the type */
4321 SvFLAGS(dstr) = new_dflags;
4326 if (UNLIKELY(both_type == SVTYPEMASK)) {
4327 if (SvIS_FREED(dstr)) {
4328 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4329 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4331 if (SvIS_FREED(sstr)) {
4332 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4333 (void*)sstr, (void*)dstr);
4339 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4340 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4342 /* There's a lot of redundancy below but we're going for speed here */
4347 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4348 (void)SvOK_off(dstr);
4356 /* For performance, we inline promoting to type SVt_IV. */
4357 /* We're starting from SVt_NULL, so provided that define is
4358 * actual 0, we don't have to unset any SV type flags
4359 * to promote to SVt_IV. */
4360 STATIC_ASSERT_STMT(SVt_NULL == 0);
4361 SET_SVANY_FOR_BODYLESS_IV(dstr);
4362 SvFLAGS(dstr) |= SVt_IV;
4366 sv_upgrade(dstr, SVt_PVIV);
4370 goto end_of_first_switch;
4372 (void)SvIOK_only(dstr);
4373 SvIV_set(dstr, SvIVX(sstr));
4376 /* SvTAINTED can only be true if the SV has taint magic, which in
4377 turn means that the SV type is PVMG (or greater). This is the
4378 case statement for SVt_IV, so this cannot be true (whatever gcov
4380 assert(!SvTAINTED(sstr));
4385 if (dtype < SVt_PV && dtype != SVt_IV)
4386 sv_upgrade(dstr, SVt_IV);
4390 if (LIKELY( SvNOK(sstr) )) {
4394 sv_upgrade(dstr, SVt_NV);
4398 sv_upgrade(dstr, SVt_PVNV);
4402 goto end_of_first_switch;
4404 SvNV_set(dstr, SvNVX(sstr));
4405 (void)SvNOK_only(dstr);
4406 /* SvTAINTED can only be true if the SV has taint magic, which in
4407 turn means that the SV type is PVMG (or greater). This is the
4408 case statement for SVt_NV, so this cannot be true (whatever gcov
4410 assert(!SvTAINTED(sstr));
4417 sv_upgrade(dstr, SVt_PV);
4420 if (dtype < SVt_PVIV)
4421 sv_upgrade(dstr, SVt_PVIV);
4424 if (dtype < SVt_PVNV)
4425 sv_upgrade(dstr, SVt_PVNV);
4429 const char * const type = sv_reftype(sstr,0);
4431 /* diag_listed_as: Bizarre copy of %s */
4432 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4434 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4436 NOT_REACHED; /* NOTREACHED */
4440 if (dtype < SVt_REGEXP)
4442 if (dtype >= SVt_PV) {
4448 sv_upgrade(dstr, SVt_REGEXP);
4456 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4458 if (SvTYPE(sstr) != stype)
4459 stype = SvTYPE(sstr);
4461 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4462 glob_assign_glob(dstr, sstr, dtype);
4465 if (stype == SVt_PVLV)
4467 if (isREGEXP(sstr)) goto upgregexp;
4468 SvUPGRADE(dstr, SVt_PVNV);
4471 SvUPGRADE(dstr, (svtype)stype);
4473 end_of_first_switch:
4475 /* dstr may have been upgraded. */
4476 dtype = SvTYPE(dstr);
4477 sflags = SvFLAGS(sstr);
4479 if (UNLIKELY( dtype == SVt_PVCV )) {
4480 /* Assigning to a subroutine sets the prototype. */
4483 const char *const ptr = SvPV_const(sstr, len);
4485 SvGROW(dstr, len + 1);
4486 Copy(ptr, SvPVX(dstr), len + 1, char);
4487 SvCUR_set(dstr, len);
4489 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4490 CvAUTOLOAD_off(dstr);
4495 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4496 || dtype == SVt_PVFM))
4498 const char * const type = sv_reftype(dstr,0);
4500 /* diag_listed_as: Cannot copy to %s */
4501 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4503 Perl_croak(aTHX_ "Cannot copy to %s", type);
4504 } else if (sflags & SVf_ROK) {
4505 if (isGV_with_GP(dstr)
4506 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4509 if (GvIMPORTED(dstr) != GVf_IMPORTED
4510 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4512 GvIMPORTED_on(dstr);
4517 glob_assign_glob(dstr, sstr, dtype);
4521 if (dtype >= SVt_PV) {
4522 if (isGV_with_GP(dstr)) {
4523 gv_setref(dstr, sstr);
4526 if (SvPVX_const(dstr)) {
4532 (void)SvOK_off(dstr);
4533 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4534 SvFLAGS(dstr) |= sflags & SVf_ROK;
4535 assert(!(sflags & SVp_NOK));
4536 assert(!(sflags & SVp_IOK));
4537 assert(!(sflags & SVf_NOK));
4538 assert(!(sflags & SVf_IOK));
4540 else if (isGV_with_GP(dstr)) {
4541 if (!(sflags & SVf_OK)) {
4542 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4543 "Undefined value assigned to typeglob");
4546 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4547 if (dstr != (const SV *)gv) {
4548 const char * const name = GvNAME((const GV *)dstr);
4549 const STRLEN len = GvNAMELEN(dstr);
4550 HV *old_stash = NULL;
4551 bool reset_isa = FALSE;
4552 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4553 || (len == 1 && name[0] == ':')) {
4554 /* Set aside the old stash, so we can reset isa caches
4555 on its subclasses. */
4556 if((old_stash = GvHV(dstr))) {
4557 /* Make sure we do not lose it early. */
4558 SvREFCNT_inc_simple_void_NN(
4559 sv_2mortal((SV *)old_stash)
4566 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4567 gp_free(MUTABLE_GV(dstr));
4569 GvGP_set(dstr, gp_ref(GvGP(gv)));
4572 HV * const stash = GvHV(dstr);
4574 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4584 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4585 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4586 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4588 else if (sflags & SVp_POK) {
4589 const STRLEN cur = SvCUR(sstr);
4590 const STRLEN len = SvLEN(sstr);
4593 * We have three basic ways to copy the string:
4599 * Which we choose is based on various factors. The following
4600 * things are listed in order of speed, fastest to slowest:
4602 * - Copying a short string
4603 * - Copy-on-write bookkeeping
4605 * - Copying a long string
4607 * We swipe the string (steal the string buffer) if the SV on the
4608 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4609 * big win on long strings. It should be a win on short strings if
4610 * SvPVX_const(dstr) has to be allocated. If not, it should not
4611 * slow things down, as SvPVX_const(sstr) would have been freed
4614 * We also steal the buffer from a PADTMP (operator target) if it
4615 * is ‘long enough’. For short strings, a swipe does not help
4616 * here, as it causes more malloc calls the next time the target
4617 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4618 * be allocated it is still not worth swiping PADTMPs for short
4619 * strings, as the savings here are small.
4621 * If swiping is not an option, then we see whether it is
4622 * worth using copy-on-write. If the lhs already has a buf-
4623 * fer big enough and the string is short, we skip it and fall back
4624 * to method 3, since memcpy is faster for short strings than the
4625 * later bookkeeping overhead that copy-on-write entails.
4627 * If the rhs is not a copy-on-write string yet, then we also
4628 * consider whether the buffer is too large relative to the string
4629 * it holds. Some operations such as readline allocate a large
4630 * buffer in the expectation of reusing it. But turning such into
4631 * a COW buffer is counter-productive because it increases memory
4632 * usage by making readline allocate a new large buffer the sec-
4633 * ond time round. So, if the buffer is too large, again, we use
4636 * Finally, if there is no buffer on the left, or the buffer is too
4637 * small, then we use copy-on-write and make both SVs share the
4642 /* Whichever path we take through the next code, we want this true,
4643 and doing it now facilitates the COW check. */
4644 (void)SvPOK_only(dstr);
4648 /* slated for free anyway (and not COW)? */
4649 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4650 /* or a swipable TARG */
4652 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4654 /* whose buffer is worth stealing */
4655 && CHECK_COWBUF_THRESHOLD(cur,len)
4658 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4659 (!(flags & SV_NOSTEAL)) &&
4660 /* and we're allowed to steal temps */
4661 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4662 len) /* and really is a string */
4663 { /* Passes the swipe test. */
4664 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4666 SvPV_set(dstr, SvPVX_mutable(sstr));
4667 SvLEN_set(dstr, SvLEN(sstr));
4668 SvCUR_set(dstr, SvCUR(sstr));
4671 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4672 SvPV_set(sstr, NULL);
4677 else if (flags & SV_COW_SHARED_HASH_KEYS
4679 #ifdef PERL_COPY_ON_WRITE
4682 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4683 /* If this is a regular (non-hek) COW, only so
4684 many COW "copies" are possible. */
4685 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4686 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4687 && !(SvFLAGS(dstr) & SVf_BREAK)
4688 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4689 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4693 && !(SvFLAGS(dstr) & SVf_BREAK)
4696 /* Either it's a shared hash key, or it's suitable for
4699 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4704 if (!(sflags & SVf_IsCOW)) {
4706 CowREFCNT(sstr) = 0;
4709 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4715 if (sflags & SVf_IsCOW) {
4719 SvPV_set(dstr, SvPVX_mutable(sstr));
4724 /* SvIsCOW_shared_hash */
4725 DEBUG_C(PerlIO_printf(Perl_debug_log,
4726 "Copy on write: Sharing hash\n"));
4728 assert (SvTYPE(dstr) >= SVt_PV);
4730 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4732 SvLEN_set(dstr, len);
4733 SvCUR_set(dstr, cur);
4736 /* Failed the swipe test, and we cannot do copy-on-write either.
4737 Have to copy the string. */
4738 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4739 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4740 SvCUR_set(dstr, cur);
4741 *SvEND(dstr) = '\0';
4743 if (sflags & SVp_NOK) {
4744 SvNV_set(dstr, SvNVX(sstr));
4746 if (sflags & SVp_IOK) {
4747 SvIV_set(dstr, SvIVX(sstr));
4748 /* Must do this otherwise some other overloaded use of 0x80000000
4749 gets confused. I guess SVpbm_VALID */
4750 if (sflags & SVf_IVisUV)
4753 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4755 const MAGIC * const smg = SvVSTRING_mg(sstr);
4757 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4758 smg->mg_ptr, smg->mg_len);
4759 SvRMAGICAL_on(dstr);
4763 else if (sflags & (SVp_IOK|SVp_NOK)) {
4764 (void)SvOK_off(dstr);
4765 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4766 if (sflags & SVp_IOK) {
4767 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4768 SvIV_set(dstr, SvIVX(sstr));
4770 if (sflags & SVp_NOK) {
4771 SvNV_set(dstr, SvNVX(sstr));
4775 if (isGV_with_GP(sstr)) {
4776 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4779 (void)SvOK_off(dstr);
4781 if (SvTAINTED(sstr))
4786 =for apidoc sv_setsv_mg
4788 Like C<sv_setsv>, but also handles 'set' magic.
4794 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4796 PERL_ARGS_ASSERT_SV_SETSV_MG;
4798 sv_setsv(dstr,sstr);
4803 # define SVt_COW SVt_PV
4805 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4807 STRLEN cur = SvCUR(sstr);
4808 STRLEN len = SvLEN(sstr);
4810 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4811 const bool already = cBOOL(SvIsCOW(sstr));
4814 PERL_ARGS_ASSERT_SV_SETSV_COW;
4817 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4818 (void*)sstr, (void*)dstr);
4825 if (SvTHINKFIRST(dstr))
4826 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4827 else if (SvPVX_const(dstr))
4828 Safefree(SvPVX_mutable(dstr));
4832 SvUPGRADE(dstr, SVt_COW);
4834 assert (SvPOK(sstr));
4835 assert (SvPOKp(sstr));
4837 if (SvIsCOW(sstr)) {
4839 if (SvLEN(sstr) == 0) {
4840 /* source is a COW shared hash key. */
4841 DEBUG_C(PerlIO_printf(Perl_debug_log,
4842 "Fast copy on write: Sharing hash\n"));
4843 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4846 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4847 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4849 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4850 SvUPGRADE(sstr, SVt_COW);
4852 DEBUG_C(PerlIO_printf(Perl_debug_log,
4853 "Fast copy on write: Converting sstr to COW\n"));
4854 CowREFCNT(sstr) = 0;
4856 # ifdef PERL_DEBUG_READONLY_COW
4857 if (already) sv_buf_to_rw(sstr);
4860 new_pv = SvPVX_mutable(sstr);
4864 SvPV_set(dstr, new_pv);
4865 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4868 SvLEN_set(dstr, len);
4869 SvCUR_set(dstr, cur);
4878 =for apidoc sv_setpvn
4880 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4881 The C<len> parameter indicates the number of
4882 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4883 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4889 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4893 PERL_ARGS_ASSERT_SV_SETPVN;
4895 SV_CHECK_THINKFIRST_COW_DROP(sv);
4901 /* len is STRLEN which is unsigned, need to copy to signed */
4904 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4907 SvUPGRADE(sv, SVt_PV);
4909 dptr = SvGROW(sv, len + 1);
4910 Move(ptr,dptr,len,char);
4913 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4915 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4919 =for apidoc sv_setpvn_mg
4921 Like C<sv_setpvn>, but also handles 'set' magic.
4927 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4929 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4931 sv_setpvn(sv,ptr,len);
4936 =for apidoc sv_setpv
4938 Copies a string into an SV. The string must be terminated with a C<NUL>
4939 character, and not contain embeded C<NUL>'s.
4940 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4946 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4950 PERL_ARGS_ASSERT_SV_SETPV;
4952 SV_CHECK_THINKFIRST_COW_DROP(sv);
4958 SvUPGRADE(sv, SVt_PV);
4960 SvGROW(sv, len + 1);
4961 Move(ptr,SvPVX(sv),len+1,char);
4963 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4965 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4969 =for apidoc sv_setpv_mg
4971 Like C<sv_setpv>, but also handles 'set' magic.
4977 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4979 PERL_ARGS_ASSERT_SV_SETPV_MG;
4986 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4988 PERL_ARGS_ASSERT_SV_SETHEK;
4994 if (HEK_LEN(hek) == HEf_SVKEY) {
4995 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4998 const int flags = HEK_FLAGS(hek);
4999 if (flags & HVhek_WASUTF8) {
5000 STRLEN utf8_len = HEK_LEN(hek);
5001 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5002 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5005 } else if (flags & HVhek_UNSHARED) {
5006 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5009 else SvUTF8_off(sv);
5013 SV_CHECK_THINKFIRST_COW_DROP(sv);
5014 SvUPGRADE(sv, SVt_PV);
5016 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5017 SvCUR_set(sv, HEK_LEN(hek));
5023 else SvUTF8_off(sv);
5031 =for apidoc sv_usepvn_flags
5033 Tells an SV to use C<ptr> to find its string value. Normally the
5034 string is stored inside the SV, but sv_usepvn allows the SV to use an
5035 outside string. C<ptr> should point to memory that was allocated
5036 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5037 the start of a C<Newx>-ed block of memory, and not a pointer to the
5038 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5039 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5040 string length, C<len>, must be supplied. By default this function
5041 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5042 so that pointer should not be freed or used by the programmer after
5043 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5044 that pointer (e.g. ptr + 1) be used.
5046 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5047 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5049 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5050 C<len>, and already meets the requirements for storing in C<SvPVX>).
5056 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5060 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5062 SV_CHECK_THINKFIRST_COW_DROP(sv);
5063 SvUPGRADE(sv, SVt_PV);
5066 if (flags & SV_SMAGIC)
5070 if (SvPVX_const(sv))
5074 if (flags & SV_HAS_TRAILING_NUL)
5075 assert(ptr[len] == '\0');
5078 allocate = (flags & SV_HAS_TRAILING_NUL)
5080 #ifdef Perl_safesysmalloc_size
5083 PERL_STRLEN_ROUNDUP(len + 1);
5085 if (flags & SV_HAS_TRAILING_NUL) {
5086 /* It's long enough - do nothing.
5087 Specifically Perl_newCONSTSUB is relying on this. */
5090 /* Force a move to shake out bugs in callers. */
5091 char *new_ptr = (char*)safemalloc(allocate);
5092 Copy(ptr, new_ptr, len, char);
5093 PoisonFree(ptr,len,char);
5097 ptr = (char*) saferealloc (ptr, allocate);
5100 #ifdef Perl_safesysmalloc_size
5101 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5103 SvLEN_set(sv, allocate);
5107 if (!(flags & SV_HAS_TRAILING_NUL)) {
5110 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5112 if (flags & SV_SMAGIC)
5117 =for apidoc sv_force_normal_flags
5119 Undo various types of fakery on an SV, where fakery means
5120 "more than" a string: if the PV is a shared string, make
5121 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5122 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5123 we do the copy, and is also used locally; if this is a
5124 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5125 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5126 C<SvPOK_off> rather than making a copy. (Used where this
5127 scalar is about to be set to some other value.) In addition,
5128 the C<flags> parameter gets passed to C<sv_unref_flags()>
5129 when unreffing. C<sv_force_normal> calls this function
5130 with flags set to 0.
5132 This function is expected to be used to signal to perl that this SV is
5133 about to be written to, and any extra book-keeping needs to be taken care
5134 of. Hence, it croaks on read-only values.
5140 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5142 assert(SvIsCOW(sv));
5145 const char * const pvx = SvPVX_const(sv);
5146 const STRLEN len = SvLEN(sv);
5147 const STRLEN cur = SvCUR(sv);
5150 PerlIO_printf(Perl_debug_log,
5151 "Copy on write: Force normal %ld\n",
5156 # ifdef PERL_COPY_ON_WRITE
5158 /* Must do this first, since the CowREFCNT uses SvPVX and
5159 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5160 the only owner left of the buffer. */
5161 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5163 U8 cowrefcnt = CowREFCNT(sv);
5164 if(cowrefcnt != 0) {
5166 CowREFCNT(sv) = cowrefcnt;
5171 /* Else we are the only owner of the buffer. */
5176 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5181 if (flags & SV_COW_DROP_PV) {
5182 /* OK, so we don't need to copy our buffer. */
5185 SvGROW(sv, cur + 1);
5186 Move(pvx,SvPVX(sv),cur,char);
5192 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5199 const char * const pvx = SvPVX_const(sv);
5200 const STRLEN len = SvCUR(sv);
5204 if (flags & SV_COW_DROP_PV) {
5205 /* OK, so we don't need to copy our buffer. */
5208 SvGROW(sv, len + 1);
5209 Move(pvx,SvPVX(sv),len,char);
5212 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5218 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5220 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5223 Perl_croak_no_modify();
5224 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5225 S_sv_uncow(aTHX_ sv, flags);
5227 sv_unref_flags(sv, flags);
5228 else if (SvFAKE(sv) && isGV_with_GP(sv))
5229 sv_unglob(sv, flags);
5230 else if (SvFAKE(sv) && isREGEXP(sv)) {
5231 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5232 to sv_unglob. We only need it here, so inline it. */
5233 const bool islv = SvTYPE(sv) == SVt_PVLV;
5234 const svtype new_type =
5235 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5236 SV *const temp = newSV_type(new_type);
5237 regexp *const temp_p = ReANY((REGEXP *)sv);
5239 if (new_type == SVt_PVMG) {
5240 SvMAGIC_set(temp, SvMAGIC(sv));
5241 SvMAGIC_set(sv, NULL);
5242 SvSTASH_set(temp, SvSTASH(sv));
5243 SvSTASH_set(sv, NULL);
5245 if (!islv) SvCUR_set(temp, SvCUR(sv));
5246 /* Remember that SvPVX is in the head, not the body. But
5247 RX_WRAPPED is in the body. */
5248 assert(ReANY((REGEXP *)sv)->mother_re);
5249 /* Their buffer is already owned by someone else. */
5250 if (flags & SV_COW_DROP_PV) {
5251 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5252 zeroed body. For SVt_PVLV, it should have been set to 0
5253 before turning into a regexp. */
5254 assert(!SvLEN(islv ? sv : temp));
5255 sv->sv_u.svu_pv = 0;
5258 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5259 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5263 /* Now swap the rest of the bodies. */
5267 SvFLAGS(sv) &= ~SVTYPEMASK;
5268 SvFLAGS(sv) |= new_type;
5269 SvANY(sv) = SvANY(temp);
5272 SvFLAGS(temp) &= ~(SVTYPEMASK);
5273 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5274 SvANY(temp) = temp_p;
5275 temp->sv_u.svu_rx = (regexp *)temp_p;
5277 SvREFCNT_dec_NN(temp);
5279 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5285 Efficient removal of characters from the beginning of the string buffer.
5286 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5287 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5288 character of the adjusted string. Uses the C<OOK> hack. On return, only
5289 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5291 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5292 refer to the same chunk of data.
5294 The unfortunate similarity of this function's name to that of Perl's C<chop>
5295 operator is strictly coincidental. This function works from the left;
5296 C<chop> works from the right.
5302 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5313 PERL_ARGS_ASSERT_SV_CHOP;
5315 if (!ptr || !SvPOKp(sv))
5317 delta = ptr - SvPVX_const(sv);
5319 /* Nothing to do. */
5322 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5323 if (delta > max_delta)
5324 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5325 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5326 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5327 SV_CHECK_THINKFIRST(sv);
5328 SvPOK_only_UTF8(sv);
5331 if (!SvLEN(sv)) { /* make copy of shared string */
5332 const char *pvx = SvPVX_const(sv);
5333 const STRLEN len = SvCUR(sv);
5334 SvGROW(sv, len + 1);
5335 Move(pvx,SvPVX(sv),len,char);
5341 SvOOK_offset(sv, old_delta);
5343 SvLEN_set(sv, SvLEN(sv) - delta);
5344 SvCUR_set(sv, SvCUR(sv) - delta);
5345 SvPV_set(sv, SvPVX(sv) + delta);
5347 p = (U8 *)SvPVX_const(sv);
5350 /* how many bytes were evacuated? we will fill them with sentinel
5351 bytes, except for the part holding the new offset of course. */
5354 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5356 assert(evacn <= delta + old_delta);
5360 /* This sets 'delta' to the accumulated value of all deltas so far */
5364 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5365 * the string; otherwise store a 0 byte there and store 'delta' just prior
5366 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5367 * portion of the chopped part of the string */
5368 if (delta < 0x100) {
5372 p -= sizeof(STRLEN);
5373 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5377 /* Fill the preceding buffer with sentinals to verify that no-one is
5387 =for apidoc sv_catpvn
5389 Concatenates the string onto the end of the string which is in the SV.
5390 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5391 status set, then the bytes appended should be valid UTF-8.
5392 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5394 =for apidoc sv_catpvn_flags
5396 Concatenates the string onto the end of the string which is in the SV. The
5397 C<len> indicates number of bytes to copy.
5399 By default, the string appended is assumed to be valid UTF-8 if the SV has
5400 the UTF-8 status set, and a string of bytes otherwise. One can force the
5401 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5402 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5403 string appended will be upgraded to UTF-8 if necessary.
5405 If C<flags> has the C<SV_SMAGIC> bit set, will
5406 C<mg_set> on C<dsv> afterwards if appropriate.
5407 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5408 in terms of this function.
5414 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5417 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5419 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5420 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5422 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5423 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5424 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5427 else SvGROW(dsv, dlen + slen + 1);
5429 sstr = SvPVX_const(dsv);
5430 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5431 SvCUR_set(dsv, SvCUR(dsv) + slen);
5434 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5435 const char * const send = sstr + slen;
5438 /* Something this code does not account for, which I think is
5439 impossible; it would require the same pv to be treated as
5440 bytes *and* utf8, which would indicate a bug elsewhere. */
5441 assert(sstr != dstr);
5443 SvGROW(dsv, dlen + slen * 2 + 1);
5444 d = (U8 *)SvPVX(dsv) + dlen;
5446 while (sstr < send) {
5447 append_utf8_from_native_byte(*sstr, &d);
5450 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5453 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5455 if (flags & SV_SMAGIC)
5460 =for apidoc sv_catsv
5462 Concatenates the string from SV C<ssv> onto the end of the string in SV
5463 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5464 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5465 and C<L</sv_catsv_nomg>>.
5467 =for apidoc sv_catsv_flags
5469 Concatenates the string from SV C<ssv> onto the end of the string in SV
5470 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5471 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5472 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5473 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5474 and C<sv_catsv_mg> are implemented in terms of this function.
5479 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5481 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5485 const char *spv = SvPV_flags_const(ssv, slen, flags);
5486 if (flags & SV_GMAGIC)
5488 sv_catpvn_flags(dsv, spv, slen,
5489 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5490 if (flags & SV_SMAGIC)
5496 =for apidoc sv_catpv
5498 Concatenates the C<NUL>-terminated string onto the end of the string which is
5500 If the SV has the UTF-8 status set, then the bytes appended should be
5501 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5507 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5513 PERL_ARGS_ASSERT_SV_CATPV;
5517 junk = SvPV_force(sv, tlen);
5519 SvGROW(sv, tlen + len + 1);
5521 ptr = SvPVX_const(sv);
5522 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5523 SvCUR_set(sv, SvCUR(sv) + len);
5524 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5529 =for apidoc sv_catpv_flags
5531 Concatenates the C<NUL>-terminated string onto the end of the string which is
5533 If the SV has the UTF-8 status set, then the bytes appended should
5534 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5535 on the modified SV if appropriate.
5541 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5543 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5544 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5548 =for apidoc sv_catpv_mg
5550 Like C<sv_catpv>, but also handles 'set' magic.
5556 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5558 PERL_ARGS_ASSERT_SV_CATPV_MG;
5567 Creates a new SV. A non-zero C<len> parameter indicates the number of
5568 bytes of preallocated string space the SV should have. An extra byte for a
5569 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5570 space is allocated.) The reference count for the new SV is set to 1.
5572 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5573 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5574 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5575 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5576 modules supporting older perls.
5582 Perl_newSV(pTHX_ const STRLEN len)
5588 sv_grow(sv, len + 1);
5593 =for apidoc sv_magicext
5595 Adds magic to an SV, upgrading it if necessary. Applies the
5596 supplied C<vtable> and returns a pointer to the magic added.
5598 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5599 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5600 one instance of the same C<how>.
5602 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5603 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5604 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5605 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5607 (This is now used as a subroutine by C<sv_magic>.)
5612 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5613 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5617 PERL_ARGS_ASSERT_SV_MAGICEXT;
5619 SvUPGRADE(sv, SVt_PVMG);
5620 Newxz(mg, 1, MAGIC);
5621 mg->mg_moremagic = SvMAGIC(sv);
5622 SvMAGIC_set(sv, mg);
5624 /* Sometimes a magic contains a reference loop, where the sv and
5625 object refer to each other. To prevent a reference loop that
5626 would prevent such objects being freed, we look for such loops
5627 and if we find one we avoid incrementing the object refcount.
5629 Note we cannot do this to avoid self-tie loops as intervening RV must
5630 have its REFCNT incremented to keep it in existence.
5633 if (!obj || obj == sv ||
5634 how == PERL_MAGIC_arylen ||
5635 how == PERL_MAGIC_symtab ||
5636 (SvTYPE(obj) == SVt_PVGV &&
5637 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5638 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5639 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5644 mg->mg_obj = SvREFCNT_inc_simple(obj);
5645 mg->mg_flags |= MGf_REFCOUNTED;
5648 /* Normal self-ties simply pass a null object, and instead of
5649 using mg_obj directly, use the SvTIED_obj macro to produce a
5650 new RV as needed. For glob "self-ties", we are tieing the PVIO
5651 with an RV obj pointing to the glob containing the PVIO. In
5652 this case, to avoid a reference loop, we need to weaken the
5656 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5657 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5663 mg->mg_len = namlen;
5666 mg->mg_ptr = savepvn(name, namlen);
5667 else if (namlen == HEf_SVKEY) {
5668 /* Yes, this is casting away const. This is only for the case of
5669 HEf_SVKEY. I think we need to document this aberation of the
5670 constness of the API, rather than making name non-const, as
5671 that change propagating outwards a long way. */
5672 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5674 mg->mg_ptr = (char *) name;
5676 mg->mg_virtual = (MGVTBL *) vtable;
5683 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5685 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5686 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5687 /* This sv is only a delegate. //g magic must be attached to
5692 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5693 &PL_vtbl_mglob, 0, 0);
5697 =for apidoc sv_magic
5699 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5700 necessary, then adds a new magic item of type C<how> to the head of the
5703 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5704 handling of the C<name> and C<namlen> arguments.
5706 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5707 to add more than one instance of the same C<how>.
5713 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5714 const char *const name, const I32 namlen)
5716 const MGVTBL *vtable;
5719 unsigned int vtable_index;
5721 PERL_ARGS_ASSERT_SV_MAGIC;
5723 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5724 || ((flags = PL_magic_data[how]),
5725 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5726 > magic_vtable_max))
5727 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5729 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5730 Useful for attaching extension internal data to perl vars.
5731 Note that multiple extensions may clash if magical scalars
5732 etc holding private data from one are passed to another. */
5734 vtable = (vtable_index == magic_vtable_max)
5735 ? NULL : PL_magic_vtables + vtable_index;
5737 if (SvREADONLY(sv)) {
5739 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5742 Perl_croak_no_modify();
5745 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5746 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5747 /* sv_magic() refuses to add a magic of the same 'how' as an
5750 if (how == PERL_MAGIC_taint)
5756 /* Force pos to be stored as characters, not bytes. */
5757 if (SvMAGICAL(sv) && DO_UTF8(sv)
5758 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5760 && mg->mg_flags & MGf_BYTES) {
5761 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5763 mg->mg_flags &= ~MGf_BYTES;
5766 /* Rest of work is done else where */
5767 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5770 case PERL_MAGIC_taint:
5773 case PERL_MAGIC_ext:
5774 case PERL_MAGIC_dbfile:
5781 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5788 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5790 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5791 for (mg = *mgp; mg; mg = *mgp) {
5792 const MGVTBL* const virt = mg->mg_virtual;
5793 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5794 *mgp = mg->mg_moremagic;
5795 if (virt && virt->svt_free)
5796 virt->svt_free(aTHX_ sv, mg);
5797 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5799 Safefree(mg->mg_ptr);
5800 else if (mg->mg_len == HEf_SVKEY)
5801 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5802 else if (mg->mg_type == PERL_MAGIC_utf8)
5803 Safefree(mg->mg_ptr);
5805 if (mg->mg_flags & MGf_REFCOUNTED)
5806 SvREFCNT_dec(mg->mg_obj);
5810 mgp = &mg->mg_moremagic;
5813 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5814 mg_magical(sv); /* else fix the flags now */
5823 =for apidoc sv_unmagic
5825 Removes all magic of type C<type> from an SV.
5831 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5833 PERL_ARGS_ASSERT_SV_UNMAGIC;
5834 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5838 =for apidoc sv_unmagicext
5840 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5846 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5848 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5849 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5853 =for apidoc sv_rvweaken
5855 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5856 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5857 push a back-reference to this RV onto the array of backreferences
5858 associated with that magic. If the RV is magical, set magic will be
5859 called after the RV is cleared.
5865 Perl_sv_rvweaken(pTHX_ SV *const sv)
5869 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5871 if (!SvOK(sv)) /* let undefs pass */
5874 Perl_croak(aTHX_ "Can't weaken a nonreference");
5875 else if (SvWEAKREF(sv)) {
5876 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5879 else if (SvREADONLY(sv)) croak_no_modify();
5881 Perl_sv_add_backref(aTHX_ tsv, sv);
5883 SvREFCNT_dec_NN(tsv);
5888 =for apidoc sv_get_backrefs
5890 If C<sv> is the target of a weak reference then it returns the back
5891 references structure associated with the sv; otherwise return C<NULL>.
5893 When returning a non-null result the type of the return is relevant. If it
5894 is an AV then the elements of the AV are the weak reference RVs which
5895 point at this item. If it is any other type then the item itself is the
5898 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5899 C<Perl_sv_kill_backrefs()>
5905 Perl_sv_get_backrefs(SV *const sv)
5909 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5911 /* find slot to store array or singleton backref */
5913 if (SvTYPE(sv) == SVt_PVHV) {
5915 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5916 backrefs = (SV *)iter->xhv_backreferences;
5918 } else if (SvMAGICAL(sv)) {
5919 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5921 backrefs = mg->mg_obj;
5926 /* Give tsv backref magic if it hasn't already got it, then push a
5927 * back-reference to sv onto the array associated with the backref magic.
5929 * As an optimisation, if there's only one backref and it's not an AV,
5930 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5931 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5935 /* A discussion about the backreferences array and its refcount:
5937 * The AV holding the backreferences is pointed to either as the mg_obj of
5938 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5939 * xhv_backreferences field. The array is created with a refcount
5940 * of 2. This means that if during global destruction the array gets
5941 * picked on before its parent to have its refcount decremented by the
5942 * random zapper, it won't actually be freed, meaning it's still there for
5943 * when its parent gets freed.
5945 * When the parent SV is freed, the extra ref is killed by
5946 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5947 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5949 * When a single backref SV is stored directly, it is not reference
5954 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5960 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5962 /* find slot to store array or singleton backref */
5964 if (SvTYPE(tsv) == SVt_PVHV) {
5965 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5968 mg = mg_find(tsv, PERL_MAGIC_backref);
5970 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5971 svp = &(mg->mg_obj);
5974 /* create or retrieve the array */
5976 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5977 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5981 mg->mg_flags |= MGf_REFCOUNTED;
5984 SvREFCNT_inc_simple_void_NN(av);
5985 /* av now has a refcnt of 2; see discussion above */
5986 av_extend(av, *svp ? 2 : 1);
5988 /* move single existing backref to the array */
5989 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5994 av = MUTABLE_AV(*svp);
5996 /* optimisation: store single backref directly in HvAUX or mg_obj */
6000 assert(SvTYPE(av) == SVt_PVAV);
6001 if (AvFILLp(av) >= AvMAX(av)) {
6002 av_extend(av, AvFILLp(av)+1);
6005 /* push new backref */
6006 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6009 /* delete a back-reference to ourselves from the backref magic associated
6010 * with the SV we point to.
6014 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6018 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6020 if (SvTYPE(tsv) == SVt_PVHV) {
6022 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6024 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6025 /* It's possible for the the last (strong) reference to tsv to have
6026 become freed *before* the last thing holding a weak reference.
6027 If both survive longer than the backreferences array, then when
6028 the referent's reference count drops to 0 and it is freed, it's
6029 not able to chase the backreferences, so they aren't NULLed.
6031 For example, a CV holds a weak reference to its stash. If both the
6032 CV and the stash survive longer than the backreferences array,
6033 and the CV gets picked for the SvBREAK() treatment first,
6034 *and* it turns out that the stash is only being kept alive because
6035 of an our variable in the pad of the CV, then midway during CV
6036 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6037 It ends up pointing to the freed HV. Hence it's chased in here, and
6038 if this block wasn't here, it would hit the !svp panic just below.
6040 I don't believe that "better" destruction ordering is going to help
6041 here - during global destruction there's always going to be the
6042 chance that something goes out of order. We've tried to make it
6043 foolproof before, and it only resulted in evolutionary pressure on
6044 fools. Which made us look foolish for our hubris. :-(
6050 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6051 svp = mg ? &(mg->mg_obj) : NULL;
6055 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6057 /* It's possible that sv is being freed recursively part way through the
6058 freeing of tsv. If this happens, the backreferences array of tsv has
6059 already been freed, and so svp will be NULL. If this is the case,
6060 we should not panic. Instead, nothing needs doing, so return. */
6061 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6063 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6064 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6067 if (SvTYPE(*svp) == SVt_PVAV) {
6071 AV * const av = (AV*)*svp;
6073 assert(!SvIS_FREED(av));
6077 /* for an SV with N weak references to it, if all those
6078 * weak refs are deleted, then sv_del_backref will be called
6079 * N times and O(N^2) compares will be done within the backref
6080 * array. To ameliorate this potential slowness, we:
6081 * 1) make sure this code is as tight as possible;
6082 * 2) when looking for SV, look for it at both the head and tail of the
6083 * array first before searching the rest, since some create/destroy
6084 * patterns will cause the backrefs to be freed in order.
6091 SV **p = &svp[fill];
6092 SV *const topsv = *p;
6099 /* We weren't the last entry.
6100 An unordered list has this property that you
6101 can take the last element off the end to fill
6102 the hole, and it's still an unordered list :-)
6108 break; /* should only be one */
6115 AvFILLp(av) = fill-1;
6117 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6118 /* freed AV; skip */
6121 /* optimisation: only a single backref, stored directly */
6123 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6124 (void*)*svp, (void*)sv);
6131 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6137 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6142 /* after multiple passes through Perl_sv_clean_all() for a thingy
6143 * that has badly leaked, the backref array may have gotten freed,
6144 * since we only protect it against 1 round of cleanup */
6145 if (SvIS_FREED(av)) {
6146 if (PL_in_clean_all) /* All is fair */
6149 "panic: magic_killbackrefs (freed backref AV/SV)");
6153 is_array = (SvTYPE(av) == SVt_PVAV);
6155 assert(!SvIS_FREED(av));
6158 last = svp + AvFILLp(av);
6161 /* optimisation: only a single backref, stored directly */
6167 while (svp <= last) {
6169 SV *const referrer = *svp;
6170 if (SvWEAKREF(referrer)) {
6171 /* XXX Should we check that it hasn't changed? */
6172 assert(SvROK(referrer));
6173 SvRV_set(referrer, 0);
6175 SvWEAKREF_off(referrer);
6176 SvSETMAGIC(referrer);
6177 } else if (SvTYPE(referrer) == SVt_PVGV ||
6178 SvTYPE(referrer) == SVt_PVLV) {
6179 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6180 /* You lookin' at me? */
6181 assert(GvSTASH(referrer));
6182 assert(GvSTASH(referrer) == (const HV *)sv);
6183 GvSTASH(referrer) = 0;
6184 } else if (SvTYPE(referrer) == SVt_PVCV ||
6185 SvTYPE(referrer) == SVt_PVFM) {
6186 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6187 /* You lookin' at me? */
6188 assert(CvSTASH(referrer));
6189 assert(CvSTASH(referrer) == (const HV *)sv);
6190 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6193 assert(SvTYPE(sv) == SVt_PVGV);
6194 /* You lookin' at me? */
6195 assert(CvGV(referrer));
6196 assert(CvGV(referrer) == (const GV *)sv);
6197 anonymise_cv_maybe(MUTABLE_GV(sv),
6198 MUTABLE_CV(referrer));
6203 "panic: magic_killbackrefs (flags=%"UVxf")",
6204 (UV)SvFLAGS(referrer));
6215 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6221 =for apidoc sv_insert
6223 Inserts a string at the specified offset/length within the SV. Similar to
6224 the Perl C<substr()> function. Handles get magic.
6226 =for apidoc sv_insert_flags
6228 Same as C<sv_insert>, but the extra C<flags> are passed to the
6229 C<SvPV_force_flags> that applies to C<bigstr>.
6235 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6241 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6244 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6246 SvPV_force_flags(bigstr, curlen, flags);
6247 (void)SvPOK_only_UTF8(bigstr);
6248 if (offset + len > curlen) {
6249 SvGROW(bigstr, offset+len+1);
6250 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6251 SvCUR_set(bigstr, offset+len);
6255 i = littlelen - len;
6256 if (i > 0) { /* string might grow */
6257 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6258 mid = big + offset + len;
6259 midend = bigend = big + SvCUR(bigstr);
6262 while (midend > mid) /* shove everything down */
6263 *--bigend = *--midend;
6264 Move(little,big+offset,littlelen,char);
6265 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6270 Move(little,SvPVX(bigstr)+offset,len,char);
6275 big = SvPVX(bigstr);
6278 bigend = big + SvCUR(bigstr);
6280 if (midend > bigend)
6281 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6284 if (mid - big > bigend - midend) { /* faster to shorten from end */
6286 Move(little, mid, littlelen,char);
6289 i = bigend - midend;
6291 Move(midend, mid, i,char);
6295 SvCUR_set(bigstr, mid - big);
6297 else if ((i = mid - big)) { /* faster from front */
6298 midend -= littlelen;
6300 Move(big, midend - i, i, char);
6301 sv_chop(bigstr,midend-i);
6303 Move(little, mid, littlelen,char);
6305 else if (littlelen) {
6306 midend -= littlelen;
6307 sv_chop(bigstr,midend);
6308 Move(little,midend,littlelen,char);
6311 sv_chop(bigstr,midend);
6317 =for apidoc sv_replace
6319 Make the first argument a copy of the second, then delete the original.
6320 The target SV physically takes over ownership of the body of the source SV
6321 and inherits its flags; however, the target keeps any magic it owns,
6322 and any magic in the source is discarded.
6323 Note that this is a rather specialist SV copying operation; most of the
6324 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6330 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6332 const U32 refcnt = SvREFCNT(sv);
6334 PERL_ARGS_ASSERT_SV_REPLACE;
6336 SV_CHECK_THINKFIRST_COW_DROP(sv);
6337 if (SvREFCNT(nsv) != 1) {
6338 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6339 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6341 if (SvMAGICAL(sv)) {
6345 sv_upgrade(nsv, SVt_PVMG);
6346 SvMAGIC_set(nsv, SvMAGIC(sv));
6347 SvFLAGS(nsv) |= SvMAGICAL(sv);
6349 SvMAGIC_set(sv, NULL);
6353 assert(!SvREFCNT(sv));
6354 #ifdef DEBUG_LEAKING_SCALARS
6355 sv->sv_flags = nsv->sv_flags;
6356 sv->sv_any = nsv->sv_any;
6357 sv->sv_refcnt = nsv->sv_refcnt;
6358 sv->sv_u = nsv->sv_u;
6360 StructCopy(nsv,sv,SV);
6362 if(SvTYPE(sv) == SVt_IV) {
6363 SET_SVANY_FOR_BODYLESS_IV(sv);
6367 SvREFCNT(sv) = refcnt;
6368 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6373 /* We're about to free a GV which has a CV that refers back to us.
6374 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6378 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6383 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6386 assert(SvREFCNT(gv) == 0);
6387 assert(isGV(gv) && isGV_with_GP(gv));
6389 assert(!CvANON(cv));
6390 assert(CvGV(cv) == gv);
6391 assert(!CvNAMED(cv));
6393 /* will the CV shortly be freed by gp_free() ? */
6394 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6395 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6399 /* if not, anonymise: */
6400 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6401 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6402 : newSVpvn_flags( "__ANON__", 8, 0 );
6403 sv_catpvs(gvname, "::__ANON__");
6404 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6405 SvREFCNT_dec_NN(gvname);
6409 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6414 =for apidoc sv_clear
6416 Clear an SV: call any destructors, free up any memory used by the body,
6417 and free the body itself. The SV's head is I<not> freed, although
6418 its type is set to all 1's so that it won't inadvertently be assumed
6419 to be live during global destruction etc.
6420 This function should only be called when C<REFCNT> is zero. Most of the time
6421 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6428 Perl_sv_clear(pTHX_ SV *const orig_sv)
6433 const struct body_details *sv_type_details;
6437 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6438 Not strictly necessary */
6440 PERL_ARGS_ASSERT_SV_CLEAR;
6442 /* within this loop, sv is the SV currently being freed, and
6443 * iter_sv is the most recent AV or whatever that's being iterated
6444 * over to provide more SVs */
6450 assert(SvREFCNT(sv) == 0);
6451 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6453 if (type <= SVt_IV) {
6454 /* See the comment in sv.h about the collusion between this
6455 * early return and the overloading of the NULL slots in the
6459 SvFLAGS(sv) &= SVf_BREAK;
6460 SvFLAGS(sv) |= SVTYPEMASK;
6464 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6465 for another purpose */
6466 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6468 if (type >= SVt_PVMG) {
6470 if (!curse(sv, 1)) goto get_next_sv;
6471 type = SvTYPE(sv); /* destructor may have changed it */
6473 /* Free back-references before magic, in case the magic calls
6474 * Perl code that has weak references to sv. */
6475 if (type == SVt_PVHV) {
6476 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6480 else if (SvMAGIC(sv)) {
6481 /* Free back-references before other types of magic. */
6482 sv_unmagic(sv, PERL_MAGIC_backref);
6488 /* case SVt_INVLIST: */
6491 IoIFP(sv) != PerlIO_stdin() &&
6492 IoIFP(sv) != PerlIO_stdout() &&
6493 IoIFP(sv) != PerlIO_stderr() &&
6494 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6496 io_close(MUTABLE_IO(sv), NULL, FALSE,
6497 (IoTYPE(sv) == IoTYPE_WRONLY ||
6498 IoTYPE(sv) == IoTYPE_RDWR ||
6499 IoTYPE(sv) == IoTYPE_APPEND));
6501 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6502 PerlDir_close(IoDIRP(sv));
6503 IoDIRP(sv) = (DIR*)NULL;
6504 Safefree(IoTOP_NAME(sv));
6505 Safefree(IoFMT_NAME(sv));
6506 Safefree(IoBOTTOM_NAME(sv));
6507 if ((const GV *)sv == PL_statgv)
6511 /* FIXME for plugins */
6513 pregfree2((REGEXP*) sv);
6517 cv_undef(MUTABLE_CV(sv));
6518 /* If we're in a stash, we don't own a reference to it.
6519 * However it does have a back reference to us, which needs to
6521 if ((stash = CvSTASH(sv)))
6522 sv_del_backref(MUTABLE_SV(stash), sv);
6525 if (PL_last_swash_hv == (const HV *)sv) {
6526 PL_last_swash_hv = NULL;
6528 if (HvTOTALKEYS((HV*)sv) > 0) {
6530 /* this statement should match the one at the beginning of
6531 * hv_undef_flags() */
6532 if ( PL_phase != PERL_PHASE_DESTRUCT
6533 && (hek = HvNAME_HEK((HV*)sv)))
6535 if (PL_stashcache) {
6536 DEBUG_o(Perl_deb(aTHX_
6537 "sv_clear clearing PL_stashcache for '%"HEKf
6540 (void)hv_deletehek(PL_stashcache,
6543 hv_name_set((HV*)sv, NULL, 0, 0);
6546 /* save old iter_sv in unused SvSTASH field */
6547 assert(!SvOBJECT(sv));
6548 SvSTASH(sv) = (HV*)iter_sv;
6551 /* save old hash_index in unused SvMAGIC field */
6552 assert(!SvMAGICAL(sv));
6553 assert(!SvMAGIC(sv));
6554 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6557 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6558 goto get_next_sv; /* process this new sv */
6560 /* free empty hash */
6561 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6562 assert(!HvARRAY((HV*)sv));
6566 AV* av = MUTABLE_AV(sv);
6567 if (PL_comppad == av) {
6571 if (AvREAL(av) && AvFILLp(av) > -1) {
6572 next_sv = AvARRAY(av)[AvFILLp(av)--];
6573 /* save old iter_sv in top-most slot of AV,
6574 * and pray that it doesn't get wiped in the meantime */
6575 AvARRAY(av)[AvMAX(av)] = iter_sv;
6577 goto get_next_sv; /* process this new sv */
6579 Safefree(AvALLOC(av));
6584 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6585 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6586 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6587 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6589 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6590 SvREFCNT_dec(LvTARG(sv));
6591 if (isREGEXP(sv)) goto freeregexp;
6594 if (isGV_with_GP(sv)) {
6595 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6596 && HvENAME_get(stash))
6597 mro_method_changed_in(stash);
6598 gp_free(MUTABLE_GV(sv));
6600 unshare_hek(GvNAME_HEK(sv));
6601 /* If we're in a stash, we don't own a reference to it.
6602 * However it does have a back reference to us, which
6603 * needs to be cleared. */
6604 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6605 sv_del_backref(MUTABLE_SV(stash), sv);
6607 /* FIXME. There are probably more unreferenced pointers to SVs
6608 * in the interpreter struct that we should check and tidy in
6609 * a similar fashion to this: */
6610 /* See also S_sv_unglob, which does the same thing. */
6611 if ((const GV *)sv == PL_last_in_gv)
6612 PL_last_in_gv = NULL;
6613 else if ((const GV *)sv == PL_statgv)
6615 else if ((const GV *)sv == PL_stderrgv)
6624 /* Don't bother with SvOOK_off(sv); as we're only going to
6628 SvOOK_offset(sv, offset);
6629 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6630 /* Don't even bother with turning off the OOK flag. */
6635 SV * const target = SvRV(sv);
6637 sv_del_backref(target, sv);
6643 else if (SvPVX_const(sv)
6644 && !(SvTYPE(sv) == SVt_PVIO
6645 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6649 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6653 if (CowREFCNT(sv)) {
6660 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6665 Safefree(SvPVX_mutable(sv));
6669 else if (SvPVX_const(sv) && SvLEN(sv)
6670 && !(SvTYPE(sv) == SVt_PVIO
6671 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6672 Safefree(SvPVX_mutable(sv));
6673 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6674 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6684 SvFLAGS(sv) &= SVf_BREAK;
6685 SvFLAGS(sv) |= SVTYPEMASK;
6687 sv_type_details = bodies_by_type + type;
6688 if (sv_type_details->arena) {
6689 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6690 &PL_body_roots[type]);
6692 else if (sv_type_details->body_size) {
6693 safefree(SvANY(sv));
6697 /* caller is responsible for freeing the head of the original sv */
6698 if (sv != orig_sv && !SvREFCNT(sv))
6701 /* grab and free next sv, if any */
6709 else if (!iter_sv) {
6711 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6712 AV *const av = (AV*)iter_sv;
6713 if (AvFILLp(av) > -1) {
6714 sv = AvARRAY(av)[AvFILLp(av)--];
6716 else { /* no more elements of current AV to free */
6719 /* restore previous value, squirrelled away */
6720 iter_sv = AvARRAY(av)[AvMAX(av)];
6721 Safefree(AvALLOC(av));
6724 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6725 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6726 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6727 /* no more elements of current HV to free */
6730 /* Restore previous values of iter_sv and hash_index,
6731 * squirrelled away */
6732 assert(!SvOBJECT(sv));
6733 iter_sv = (SV*)SvSTASH(sv);
6734 assert(!SvMAGICAL(sv));
6735 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6737 /* perl -DA does not like rubbish in SvMAGIC. */
6741 /* free any remaining detritus from the hash struct */
6742 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6743 assert(!HvARRAY((HV*)sv));
6748 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6752 if (!SvREFCNT(sv)) {
6756 if (--(SvREFCNT(sv)))
6760 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6761 "Attempt to free temp prematurely: SV 0x%"UVxf
6762 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6766 if (SvIMMORTAL(sv)) {
6767 /* make sure SvREFCNT(sv)==0 happens very seldom */
6768 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6777 /* This routine curses the sv itself, not the object referenced by sv. So
6778 sv does not have to be ROK. */
6781 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6782 PERL_ARGS_ASSERT_CURSE;
6783 assert(SvOBJECT(sv));
6785 if (PL_defstash && /* Still have a symbol table? */
6791 stash = SvSTASH(sv);
6792 assert(SvTYPE(stash) == SVt_PVHV);
6793 if (HvNAME(stash)) {
6794 CV* destructor = NULL;
6795 struct mro_meta *meta;
6797 assert (SvOOK(stash));
6799 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6802 /* don't make this an initialization above the assert, since it needs
6804 meta = HvMROMETA(stash);
6805 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6806 destructor = meta->destroy;
6807 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6808 (void *)destructor, HvNAME(stash)) );
6811 bool autoload = FALSE;
6813 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6815 destructor = GvCV(gv);
6817 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6818 GV_AUTOLOAD_ISMETHOD);
6820 destructor = GvCV(gv);
6824 /* we don't cache AUTOLOAD for DESTROY, since this code
6825 would then need to set $__PACKAGE__::AUTOLOAD, or the
6826 equivalent for XS AUTOLOADs */
6828 meta->destroy_gen = PL_sub_generation;
6829 meta->destroy = destructor;
6831 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6832 (void *)destructor, HvNAME(stash)) );
6835 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6839 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6841 /* A constant subroutine can have no side effects, so
6842 don't bother calling it. */
6843 && !CvCONST(destructor)
6844 /* Don't bother calling an empty destructor or one that
6845 returns immediately. */
6846 && (CvISXSUB(destructor)
6847 || (CvSTART(destructor)
6848 && (CvSTART(destructor)->op_next->op_type
6850 && (CvSTART(destructor)->op_next->op_type
6852 || CvSTART(destructor)->op_next->op_next->op_type
6858 SV* const tmpref = newRV(sv);
6859 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6861 PUSHSTACKi(PERLSI_DESTROY);
6866 call_sv(MUTABLE_SV(destructor),
6867 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6871 if(SvREFCNT(tmpref) < 2) {
6872 /* tmpref is not kept alive! */
6874 SvRV_set(tmpref, NULL);
6877 SvREFCNT_dec_NN(tmpref);
6880 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6883 if (check_refcnt && SvREFCNT(sv)) {
6884 if (PL_in_clean_objs)
6886 "DESTROY created new reference to dead object '%"HEKf"'",
6887 HEKfARG(HvNAME_HEK(stash)));
6888 /* DESTROY gave object new lease on life */
6894 HV * const stash = SvSTASH(sv);
6895 /* Curse before freeing the stash, as freeing the stash could cause
6896 a recursive call into S_curse. */
6897 SvOBJECT_off(sv); /* Curse the object. */
6898 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6899 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6905 =for apidoc sv_newref
6907 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6914 Perl_sv_newref(pTHX_ SV *const sv)
6916 PERL_UNUSED_CONTEXT;
6925 Decrement an SV's reference count, and if it drops to zero, call
6926 C<sv_clear> to invoke destructors and free up any memory used by
6927 the body; finally, deallocating the SV's head itself.
6928 Normally called via a wrapper macro C<SvREFCNT_dec>.
6934 Perl_sv_free(pTHX_ SV *const sv)
6940 /* Private helper function for SvREFCNT_dec().
6941 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6944 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6948 PERL_ARGS_ASSERT_SV_FREE2;
6950 if (LIKELY( rc == 1 )) {
6956 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6957 "Attempt to free temp prematurely: SV 0x%"UVxf
6958 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6962 if (SvIMMORTAL(sv)) {
6963 /* make sure SvREFCNT(sv)==0 happens very seldom */
6964 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6968 if (! SvREFCNT(sv)) /* may have have been resurrected */
6973 /* handle exceptional cases */
6977 if (SvFLAGS(sv) & SVf_BREAK)
6978 /* this SV's refcnt has been artificially decremented to
6979 * trigger cleanup */
6981 if (PL_in_clean_all) /* All is fair */
6983 if (SvIMMORTAL(sv)) {
6984 /* make sure SvREFCNT(sv)==0 happens very seldom */
6985 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6988 if (ckWARN_d(WARN_INTERNAL)) {
6989 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6990 Perl_dump_sv_child(aTHX_ sv);
6992 #ifdef DEBUG_LEAKING_SCALARS
6995 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6996 if (PL_warnhook == PERL_WARNHOOK_FATAL
6997 || ckDEAD(packWARN(WARN_INTERNAL))) {
6998 /* Don't let Perl_warner cause us to escape our fate: */
7002 /* This may not return: */
7003 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7004 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7005 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7008 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7018 Returns the length of the string in the SV. Handles magic and type
7019 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7020 gives raw access to the C<xpv_cur> slot.
7026 Perl_sv_len(pTHX_ SV *const sv)
7033 (void)SvPV_const(sv, len);
7038 =for apidoc sv_len_utf8
7040 Returns the number of characters in the string in an SV, counting wide
7041 UTF-8 bytes as a single character. Handles magic and type coercion.
7047 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7048 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7049 * (Note that the mg_len is not the length of the mg_ptr field.
7050 * This allows the cache to store the character length of the string without
7051 * needing to malloc() extra storage to attach to the mg_ptr.)
7056 Perl_sv_len_utf8(pTHX_ SV *const sv)
7062 return sv_len_utf8_nomg(sv);
7066 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7069 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7071 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7073 if (PL_utf8cache && SvUTF8(sv)) {
7075 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7077 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7078 if (mg->mg_len != -1)
7081 /* We can use the offset cache for a headstart.
7082 The longer value is stored in the first pair. */
7083 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7085 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7089 if (PL_utf8cache < 0) {
7090 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7091 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7095 ulen = Perl_utf8_length(aTHX_ s, s + len);
7096 utf8_mg_len_cache_update(sv, &mg, ulen);
7100 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7103 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7106 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7107 STRLEN *const uoffset_p, bool *const at_end)
7109 const U8 *s = start;
7110 STRLEN uoffset = *uoffset_p;
7112 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7114 while (s < send && uoffset) {
7121 else if (s > send) {
7123 /* This is the existing behaviour. Possibly it should be a croak, as
7124 it's actually a bounds error */
7127 *uoffset_p -= uoffset;
7131 /* Given the length of the string in both bytes and UTF-8 characters, decide
7132 whether to walk forwards or backwards to find the byte corresponding to
7133 the passed in UTF-8 offset. */
7135 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7136 STRLEN uoffset, const STRLEN uend)
7138 STRLEN backw = uend - uoffset;
7140 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7142 if (uoffset < 2 * backw) {
7143 /* The assumption is that going forwards is twice the speed of going
7144 forward (that's where the 2 * backw comes from).
7145 (The real figure of course depends on the UTF-8 data.) */
7146 const U8 *s = start;
7148 while (s < send && uoffset--)
7158 while (UTF8_IS_CONTINUATION(*send))
7161 return send - start;
7164 /* For the string representation of the given scalar, find the byte
7165 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7166 give another position in the string, *before* the sought offset, which
7167 (which is always true, as 0, 0 is a valid pair of positions), which should
7168 help reduce the amount of linear searching.
7169 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7170 will be used to reduce the amount of linear searching. The cache will be
7171 created if necessary, and the found value offered to it for update. */
7173 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7174 const U8 *const send, STRLEN uoffset,
7175 STRLEN uoffset0, STRLEN boffset0)
7177 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7179 bool at_end = FALSE;
7181 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7183 assert (uoffset >= uoffset0);
7188 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7190 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7191 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7192 if ((*mgp)->mg_ptr) {
7193 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7194 if (cache[0] == uoffset) {
7195 /* An exact match. */
7198 if (cache[2] == uoffset) {
7199 /* An exact match. */
7203 if (cache[0] < uoffset) {
7204 /* The cache already knows part of the way. */
7205 if (cache[0] > uoffset0) {
7206 /* The cache knows more than the passed in pair */
7207 uoffset0 = cache[0];
7208 boffset0 = cache[1];
7210 if ((*mgp)->mg_len != -1) {
7211 /* And we know the end too. */
7213 + sv_pos_u2b_midway(start + boffset0, send,
7215 (*mgp)->mg_len - uoffset0);
7217 uoffset -= uoffset0;
7219 + sv_pos_u2b_forwards(start + boffset0,
7220 send, &uoffset, &at_end);
7221 uoffset += uoffset0;
7224 else if (cache[2] < uoffset) {
7225 /* We're between the two cache entries. */
7226 if (cache[2] > uoffset0) {
7227 /* and the cache knows more than the passed in pair */
7228 uoffset0 = cache[2];
7229 boffset0 = cache[3];
7233 + sv_pos_u2b_midway(start + boffset0,
7236 cache[0] - uoffset0);
7239 + sv_pos_u2b_midway(start + boffset0,
7242 cache[2] - uoffset0);
7246 else if ((*mgp)->mg_len != -1) {
7247 /* If we can take advantage of a passed in offset, do so. */
7248 /* In fact, offset0 is either 0, or less than offset, so don't
7249 need to worry about the other possibility. */
7251 + sv_pos_u2b_midway(start + boffset0, send,
7253 (*mgp)->mg_len - uoffset0);
7258 if (!found || PL_utf8cache < 0) {
7259 STRLEN real_boffset;
7260 uoffset -= uoffset0;
7261 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7262 send, &uoffset, &at_end);
7263 uoffset += uoffset0;
7265 if (found && PL_utf8cache < 0)
7266 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7268 boffset = real_boffset;
7271 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7273 utf8_mg_len_cache_update(sv, mgp, uoffset);
7275 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7282 =for apidoc sv_pos_u2b_flags
7284 Converts the offset from a count of UTF-8 chars from
7285 the start of the string, to a count of the equivalent number of bytes; if
7286 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7287 C<offset>, rather than from the start
7288 of the string. Handles type coercion.
7289 C<flags> is passed to C<SvPV_flags>, and usually should be
7290 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7296 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7297 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7298 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7303 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7310 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7312 start = (U8*)SvPV_flags(sv, len, flags);
7314 const U8 * const send = start + len;
7316 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7319 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7320 is 0, and *lenp is already set to that. */) {
7321 /* Convert the relative offset to absolute. */
7322 const STRLEN uoffset2 = uoffset + *lenp;
7323 const STRLEN boffset2
7324 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7325 uoffset, boffset) - boffset;
7339 =for apidoc sv_pos_u2b
7341 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7342 the start of the string, to a count of the equivalent number of bytes; if
7343 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7344 the offset, rather than from the start of the string. Handles magic and
7347 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7354 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7355 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7356 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7360 /* This function is subject to size and sign problems */
7363 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7365 PERL_ARGS_ASSERT_SV_POS_U2B;
7368 STRLEN ulen = (STRLEN)*lenp;
7369 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7370 SV_GMAGIC|SV_CONST_RETURN);
7373 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7374 SV_GMAGIC|SV_CONST_RETURN);
7379 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7382 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7383 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7386 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7387 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7388 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7392 (*mgp)->mg_len = ulen;
7395 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7396 byte length pairing. The (byte) length of the total SV is passed in too,
7397 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7398 may not have updated SvCUR, so we can't rely on reading it directly.
7400 The proffered utf8/byte length pairing isn't used if the cache already has
7401 two pairs, and swapping either for the proffered pair would increase the
7402 RMS of the intervals between known byte offsets.
7404 The cache itself consists of 4 STRLEN values
7405 0: larger UTF-8 offset
7406 1: corresponding byte offset
7407 2: smaller UTF-8 offset
7408 3: corresponding byte offset
7410 Unused cache pairs have the value 0, 0.
7411 Keeping the cache "backwards" means that the invariant of
7412 cache[0] >= cache[2] is maintained even with empty slots, which means that
7413 the code that uses it doesn't need to worry if only 1 entry has actually
7414 been set to non-zero. It also makes the "position beyond the end of the
7415 cache" logic much simpler, as the first slot is always the one to start
7419 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7420 const STRLEN utf8, const STRLEN blen)
7424 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7429 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7430 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7431 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7433 (*mgp)->mg_len = -1;
7437 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7438 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7439 (*mgp)->mg_ptr = (char *) cache;
7443 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7444 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7445 a pointer. Note that we no longer cache utf8 offsets on refer-
7446 ences, but this check is still a good idea, for robustness. */
7447 const U8 *start = (const U8 *) SvPVX_const(sv);
7448 const STRLEN realutf8 = utf8_length(start, start + byte);
7450 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7454 /* Cache is held with the later position first, to simplify the code
7455 that deals with unbounded ends. */
7457 ASSERT_UTF8_CACHE(cache);
7458 if (cache[1] == 0) {
7459 /* Cache is totally empty */
7462 } else if (cache[3] == 0) {
7463 if (byte > cache[1]) {
7464 /* New one is larger, so goes first. */
7465 cache[2] = cache[0];
7466 cache[3] = cache[1];
7474 /* float casts necessary? XXX */
7475 #define THREEWAY_SQUARE(a,b,c,d) \
7476 ((float)((d) - (c))) * ((float)((d) - (c))) \
7477 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7478 + ((float)((b) - (a))) * ((float)((b) - (a)))
7480 /* Cache has 2 slots in use, and we know three potential pairs.
7481 Keep the two that give the lowest RMS distance. Do the
7482 calculation in bytes simply because we always know the byte
7483 length. squareroot has the same ordering as the positive value,
7484 so don't bother with the actual square root. */
7485 if (byte > cache[1]) {
7486 /* New position is after the existing pair of pairs. */
7487 const float keep_earlier
7488 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7489 const float keep_later
7490 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7492 if (keep_later < keep_earlier) {
7493 cache[2] = cache[0];
7494 cache[3] = cache[1];
7500 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7501 float b, c, keep_earlier;
7502 if (byte > cache[3]) {
7503 /* New position is between the existing pair of pairs. */
7504 b = (float)cache[3];
7507 /* New position is before the existing pair of pairs. */
7509 c = (float)cache[3];
7511 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7512 if (byte > cache[3]) {
7513 if (keep_later < keep_earlier) {
7523 if (! (keep_later < keep_earlier)) {
7524 cache[0] = cache[2];
7525 cache[1] = cache[3];
7532 ASSERT_UTF8_CACHE(cache);
7535 /* We already know all of the way, now we may be able to walk back. The same
7536 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7537 backward is half the speed of walking forward. */
7539 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7540 const U8 *end, STRLEN endu)
7542 const STRLEN forw = target - s;
7543 STRLEN backw = end - target;
7545 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7547 if (forw < 2 * backw) {
7548 return utf8_length(s, target);
7551 while (end > target) {
7553 while (UTF8_IS_CONTINUATION(*end)) {
7562 =for apidoc sv_pos_b2u_flags
7564 Converts C<offset> from a count of bytes from the start of the string, to
7565 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7566 C<flags> is passed to C<SvPV_flags>, and usually should be
7567 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7573 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7574 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7579 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7582 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7588 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7590 s = (const U8*)SvPV_flags(sv, blen, flags);
7593 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7594 ", byte=%"UVuf, (UV)blen, (UV)offset);
7600 && SvTYPE(sv) >= SVt_PVMG
7601 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7604 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7605 if (cache[1] == offset) {
7606 /* An exact match. */
7609 if (cache[3] == offset) {
7610 /* An exact match. */
7614 if (cache[1] < offset) {
7615 /* We already know part of the way. */
7616 if (mg->mg_len != -1) {
7617 /* Actually, we know the end too. */
7619 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7620 s + blen, mg->mg_len - cache[0]);
7622 len = cache[0] + utf8_length(s + cache[1], send);
7625 else if (cache[3] < offset) {
7626 /* We're between the two cached pairs, so we do the calculation
7627 offset by the byte/utf-8 positions for the earlier pair,
7628 then add the utf-8 characters from the string start to
7630 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7631 s + cache[1], cache[0] - cache[2])
7635 else { /* cache[3] > offset */
7636 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7640 ASSERT_UTF8_CACHE(cache);
7642 } else if (mg->mg_len != -1) {
7643 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7647 if (!found || PL_utf8cache < 0) {
7648 const STRLEN real_len = utf8_length(s, send);
7650 if (found && PL_utf8cache < 0)
7651 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7657 utf8_mg_len_cache_update(sv, &mg, len);
7659 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7666 =for apidoc sv_pos_b2u
7668 Converts the value pointed to by C<offsetp> from a count of bytes from the
7669 start of the string, to a count of the equivalent number of UTF-8 chars.
7670 Handles magic and type coercion.
7672 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7679 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7680 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7685 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7687 PERL_ARGS_ASSERT_SV_POS_B2U;
7692 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7693 SV_GMAGIC|SV_CONST_RETURN);
7697 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7698 STRLEN real, SV *const sv)
7700 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7702 /* As this is debugging only code, save space by keeping this test here,
7703 rather than inlining it in all the callers. */
7704 if (from_cache == real)
7707 /* Need to turn the assertions off otherwise we may recurse infinitely
7708 while printing error messages. */
7709 SAVEI8(PL_utf8cache);
7711 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7712 func, (UV) from_cache, (UV) real, SVfARG(sv));
7718 Returns a boolean indicating whether the strings in the two SVs are
7719 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7720 coerce its args to strings if necessary.
7722 =for apidoc sv_eq_flags
7724 Returns a boolean indicating whether the strings in the two SVs are
7725 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7726 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7732 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7739 SV* svrecode = NULL;
7746 /* if pv1 and pv2 are the same, second SvPV_const call may
7747 * invalidate pv1 (if we are handling magic), so we may need to
7749 if (sv1 == sv2 && flags & SV_GMAGIC
7750 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7751 pv1 = SvPV_const(sv1, cur1);
7752 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7754 pv1 = SvPV_flags_const(sv1, cur1, flags);
7762 pv2 = SvPV_flags_const(sv2, cur2, flags);
7764 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7765 /* Differing utf8ness. */
7767 /* sv1 is the UTF-8 one */
7768 return bytes_cmp_utf8((const U8*)pv2, cur2,
7769 (const U8*)pv1, cur1) == 0;
7772 /* sv2 is the UTF-8 one */
7773 return bytes_cmp_utf8((const U8*)pv1, cur1,
7774 (const U8*)pv2, cur2) == 0;
7779 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7781 SvREFCNT_dec(svrecode);
7789 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7790 string in C<sv1> is less than, equal to, or greater than the string in
7791 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7792 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7794 =for apidoc sv_cmp_flags
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 and will coerce its args to strings
7799 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7800 also C<L</sv_cmp_locale_flags>>.
7806 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7808 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7812 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7816 const char *pv1, *pv2;
7818 SV *svrecode = NULL;
7825 pv1 = SvPV_flags_const(sv1, cur1, flags);
7832 pv2 = SvPV_flags_const(sv2, cur2, flags);
7834 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7835 /* Differing utf8ness. */
7837 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7838 (const U8*)pv1, cur1);
7839 return retval ? retval < 0 ? -1 : +1 : 0;
7842 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7843 (const U8*)pv2, cur2);
7844 return retval ? retval < 0 ? -1 : +1 : 0;
7848 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7851 cmp = cur2 ? -1 : 0;
7855 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7858 if (! DO_UTF8(sv1)) {
7860 const I32 retval = memcmp((const void*)pv1,
7864 cmp = retval < 0 ? -1 : 1;
7865 } else if (cur1 == cur2) {
7868 cmp = cur1 < cur2 ? -1 : 1;
7872 else { /* Both are to be treated as UTF-EBCDIC */
7874 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7875 * which remaps code points 0-255. We therefore generally have to
7876 * unmap back to the original values to get an accurate comparison.
7877 * But we don't have to do that for UTF-8 invariants, as by
7878 * definition, they aren't remapped, nor do we have to do it for
7879 * above-latin1 code points, as they also aren't remapped. (This
7880 * code also works on ASCII platforms, but the memcmp() above is
7883 const char *e = pv1 + shortest_len;
7885 /* Find the first bytes that differ between the two strings */
7886 while (pv1 < e && *pv1 == *pv2) {
7892 if (pv1 == e) { /* Are the same all the way to the end */
7896 cmp = cur1 < cur2 ? -1 : 1;
7899 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7900 * in the strings were. The current bytes may or may not be
7901 * at the beginning of a character. But neither or both are
7902 * (or else earlier bytes would have been different). And
7903 * if we are in the middle of a character, the two
7904 * characters are comprised of the same number of bytes
7905 * (because in this case the start bytes are the same, and
7906 * the start bytes encode the character's length). */
7907 if (UTF8_IS_INVARIANT(*pv1))
7909 /* If both are invariants; can just compare directly */
7910 if (UTF8_IS_INVARIANT(*pv2)) {
7911 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7913 else /* Since *pv1 is invariant, it is the whole character,
7914 which means it is at the beginning of a character.
7915 That means pv2 is also at the beginning of a
7916 character (see earlier comment). Since it isn't
7917 invariant, it must be a start byte. If it starts a
7918 character whose code point is above 255, that
7919 character is greater than any single-byte char, which
7921 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7926 /* Here, pv2 points to a character composed of 2 bytes
7927 * whose code point is < 256. Get its code point and
7928 * compare with *pv1 */
7929 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7934 else /* The code point starting at pv1 isn't a single byte */
7935 if (UTF8_IS_INVARIANT(*pv2))
7937 /* But here, the code point starting at *pv2 is a single byte,
7938 * and so *pv1 must begin a character, hence is a start byte.
7939 * If that character is above 255, it is larger than any
7940 * single-byte char, which *pv2 is */
7941 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7945 /* Here, pv1 points to a character composed of 2 bytes
7946 * whose code point is < 256. Get its code point and
7947 * compare with the single byte character *pv2 */
7948 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7953 else /* Here, we've ruled out either *pv1 and *pv2 being
7954 invariant. That means both are part of variants, but not
7955 necessarily at the start of a character */
7956 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7957 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7959 /* Here, at least one is the start of a character, which means
7960 * the other is also a start byte. And the code point of at
7961 * least one of the characters is above 255. It is a
7962 * characteristic of UTF-EBCDIC that all start bytes for
7963 * above-latin1 code points are well behaved as far as code
7964 * point comparisons go, and all are larger than all other
7965 * start bytes, so the comparison with those is also well
7967 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7970 /* Here both *pv1 and *pv2 are part of variant characters.
7971 * They could be both continuations, or both start characters.
7972 * (One or both could even be an illegal start character (for
7973 * an overlong) which for the purposes of sorting we treat as
7975 if (UTF8_IS_CONTINUATION(*pv1)) {
7977 /* If they are continuations for code points above 255,
7978 * then comparing the current byte is sufficient, as there
7979 * is no remapping of these and so the comparison is
7980 * well-behaved. We determine if they are such
7981 * continuations by looking at the preceding byte. It
7982 * could be a start byte, from which we can tell if it is
7983 * for an above 255 code point. Or it could be a
7984 * continuation, which means the character occupies at
7985 * least 3 bytes, so must be above 255. */
7986 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
7987 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
7989 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7993 /* Here, the continuations are for code points below 256;
7994 * back up one to get to the start byte */
7999 /* We need to get the actual native code point of each of these
8000 * variants in order to compare them */
8001 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8002 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8011 SvREFCNT_dec(svrecode);
8017 =for apidoc sv_cmp_locale
8019 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8020 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8021 if necessary. See also C<L</sv_cmp>>.
8023 =for apidoc sv_cmp_locale_flags
8025 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8026 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8027 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8028 C<L</sv_cmp_flags>>.
8034 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8036 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8040 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8043 #ifdef USE_LOCALE_COLLATE
8049 if (PL_collation_standard)
8054 /* Revert to using raw compare if both operands exist, but either one
8055 * doesn't transform properly for collation */
8057 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8061 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8067 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8068 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8071 if (!pv1 || !len1) {
8082 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8085 return retval < 0 ? -1 : 1;
8088 * When the result of collation is equality, that doesn't mean
8089 * that there are no differences -- some locales exclude some
8090 * characters from consideration. So to avoid false equalities,
8091 * we use the raw string as a tiebreaker.
8098 PERL_UNUSED_ARG(flags);
8099 #endif /* USE_LOCALE_COLLATE */
8101 return sv_cmp(sv1, sv2);
8105 #ifdef USE_LOCALE_COLLATE
8108 =for apidoc sv_collxfrm
8110 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8111 C<L</sv_collxfrm_flags>>.
8113 =for apidoc sv_collxfrm_flags
8115 Add Collate Transform magic to an SV if it doesn't already have it. If the
8116 flags contain C<SV_GMAGIC>, it handles get-magic.
8118 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8119 scalar data of the variable, but transformed to such a format that a normal
8120 memory comparison can be used to compare the data according to the locale
8127 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8131 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8133 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8135 /* If we don't have collation magic on 'sv', or the locale has changed
8136 * since the last time we calculated it, get it and save it now */
8137 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8142 /* Free the old space */
8144 Safefree(mg->mg_ptr);
8146 s = SvPV_flags_const(sv, len, flags);
8147 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8149 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8164 if (mg && mg->mg_ptr) {
8166 return mg->mg_ptr + sizeof(PL_collation_ix);
8174 #endif /* USE_LOCALE_COLLATE */
8177 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8179 SV * const tsv = newSV(0);
8182 sv_gets(tsv, fp, 0);
8183 sv_utf8_upgrade_nomg(tsv);
8184 SvCUR_set(sv,append);
8187 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8191 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8194 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8195 /* Grab the size of the record we're getting */
8196 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8203 /* With a true, record-oriented file on VMS, we need to use read directly
8204 * to ensure that we respect RMS record boundaries. The user is responsible
8205 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8206 * record size) field. N.B. This is likely to produce invalid results on
8207 * varying-width character data when a record ends mid-character.
8209 fd = PerlIO_fileno(fp);
8211 && PerlLIO_fstat(fd, &st) == 0
8212 && (st.st_fab_rfm == FAB$C_VAR
8213 || st.st_fab_rfm == FAB$C_VFC
8214 || st.st_fab_rfm == FAB$C_FIX)) {
8216 bytesread = PerlLIO_read(fd, buffer, recsize);
8218 else /* in-memory file from PerlIO::Scalar
8219 * or not a record-oriented file
8223 bytesread = PerlIO_read(fp, buffer, recsize);
8225 /* At this point, the logic in sv_get() means that sv will
8226 be treated as utf-8 if the handle is utf8.
8228 if (PerlIO_isutf8(fp) && bytesread > 0) {
8229 char *bend = buffer + bytesread;
8230 char *bufp = buffer;
8231 size_t charcount = 0;
8232 bool charstart = TRUE;
8235 while (charcount < recsize) {
8236 /* count accumulated characters */
8237 while (bufp < bend) {
8239 skip = UTF8SKIP(bufp);
8241 if (bufp + skip > bend) {
8242 /* partial at the end */
8253 if (charcount < recsize) {
8255 STRLEN bufp_offset = bufp - buffer;
8256 SSize_t morebytesread;
8258 /* originally I read enough to fill any incomplete
8259 character and the first byte of the next
8260 character if needed, but if there's many
8261 multi-byte encoded characters we're going to be
8262 making a read call for every character beyond
8263 the original read size.
8265 So instead, read the rest of the character if
8266 any, and enough bytes to match at least the
8267 start bytes for each character we're going to
8271 readsize = recsize - charcount;
8273 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8274 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8275 bend = buffer + bytesread;
8276 morebytesread = PerlIO_read(fp, bend, readsize);
8277 if (morebytesread <= 0) {
8278 /* we're done, if we still have incomplete
8279 characters the check code in sv_gets() will
8282 I'd originally considered doing
8283 PerlIO_ungetc() on all but the lead
8284 character of the incomplete character, but
8285 read() doesn't do that, so I don't.
8290 /* prepare to scan some more */
8291 bytesread += morebytesread;
8292 bend = buffer + bytesread;
8293 bufp = buffer + bufp_offset;
8301 SvCUR_set(sv, bytesread + append);
8302 buffer[bytesread] = '\0';
8303 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8309 Get a line from the filehandle and store it into the SV, optionally
8310 appending to the currently-stored string. If C<append> is not 0, the
8311 line is appended to the SV instead of overwriting it. C<append> should
8312 be set to the byte offset that the appended string should start at
8313 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8319 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8329 PERL_ARGS_ASSERT_SV_GETS;
8331 if (SvTHINKFIRST(sv))
8332 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8333 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8335 However, perlbench says it's slower, because the existing swipe code
8336 is faster than copy on write.
8337 Swings and roundabouts. */
8338 SvUPGRADE(sv, SVt_PV);
8341 /* line is going to be appended to the existing buffer in the sv */
8342 if (PerlIO_isutf8(fp)) {
8344 sv_utf8_upgrade_nomg(sv);
8345 sv_pos_u2b(sv,&append,0);
8347 } else if (SvUTF8(sv)) {
8348 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8354 /* not appending - "clear" the string by setting SvCUR to 0,
8355 * the pv is still avaiable. */
8358 if (PerlIO_isutf8(fp))
8361 if (IN_PERL_COMPILETIME) {
8362 /* we always read code in line mode */
8366 else if (RsSNARF(PL_rs)) {
8367 /* If it is a regular disk file use size from stat() as estimate
8368 of amount we are going to read -- may result in mallocing
8369 more memory than we really need if the layers below reduce
8370 the size we read (e.g. CRLF or a gzip layer).
8373 int fd = PerlIO_fileno(fp);
8374 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8375 const Off_t offset = PerlIO_tell(fp);
8376 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8377 #ifdef PERL_COPY_ON_WRITE
8378 /* Add an extra byte for the sake of copy-on-write's
8379 * buffer reference count. */
8380 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8382 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8389 else if (RsRECORD(PL_rs)) {
8390 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8392 else if (RsPARA(PL_rs)) {
8398 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8399 if (PerlIO_isutf8(fp)) {
8400 rsptr = SvPVutf8(PL_rs, rslen);
8403 if (SvUTF8(PL_rs)) {
8404 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8405 Perl_croak(aTHX_ "Wide character in $/");
8408 /* extract the raw pointer to the record separator */
8409 rsptr = SvPV_const(PL_rs, rslen);
8413 /* rslast is the last character in the record separator
8414 * note we don't use rslast except when rslen is true, so the
8415 * null assign is a placeholder. */
8416 rslast = rslen ? rsptr[rslen - 1] : '\0';
8418 if (rspara) { /* have to do this both before and after */
8419 do { /* to make sure file boundaries work right */
8422 i = PerlIO_getc(fp);
8426 PerlIO_ungetc(fp,i);
8432 /* See if we know enough about I/O mechanism to cheat it ! */
8434 /* This used to be #ifdef test - it is made run-time test for ease
8435 of abstracting out stdio interface. One call should be cheap
8436 enough here - and may even be a macro allowing compile
8440 if (PerlIO_fast_gets(fp)) {
8442 * We can do buffer based IO operations on this filehandle.
8444 * This means we can bypass a lot of subcalls and process
8445 * the buffer directly, it also means we know the upper bound
8446 * on the amount of data we might read of the current buffer
8447 * into our sv. Knowing this allows us to preallocate the pv
8448 * to be able to hold that maximum, which allows us to simplify
8449 * a lot of logic. */
8452 * We're going to steal some values from the stdio struct
8453 * and put EVERYTHING in the innermost loop into registers.
8455 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8456 STRLEN bpx; /* length of the data in the target sv
8457 used to fix pointers after a SvGROW */
8458 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8459 of data left in the read-ahead buffer.
8460 If 0 then the pv buffer can hold the full
8461 amount left, otherwise this is the amount it
8464 /* Here is some breathtakingly efficient cheating */
8466 /* When you read the following logic resist the urge to think
8467 * of record separators that are 1 byte long. They are an
8468 * uninteresting special (simple) case.
8470 * Instead think of record separators which are at least 2 bytes
8471 * long, and keep in mind that we need to deal with such
8472 * separators when they cross a read-ahead buffer boundary.
8474 * Also consider that we need to gracefully deal with separators
8475 * that may be longer than a single read ahead buffer.
8477 * Lastly do not forget we want to copy the delimiter as well. We
8478 * are copying all data in the file _up_to_and_including_ the separator
8481 * Now that you have all that in mind here is what is happening below:
8483 * 1. When we first enter the loop we do some memory book keeping to see
8484 * how much free space there is in the target SV. (This sub assumes that
8485 * it is operating on the same SV most of the time via $_ and that it is
8486 * going to be able to reuse the same pv buffer each call.) If there is
8487 * "enough" room then we set "shortbuffered" to how much space there is
8488 * and start reading forward.
8490 * 2. When we scan forward we copy from the read-ahead buffer to the target
8491 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8492 * and the end of the of pv, as well as for the "rslast", which is the last
8493 * char of the separator.
8495 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8496 * (which has a "complete" record up to the point we saw rslast) and check
8497 * it to see if it matches the separator. If it does we are done. If it doesn't
8498 * we continue on with the scan/copy.
8500 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8501 * the IO system to read the next buffer. We do this by doing a getc(), which
8502 * returns a single char read (or EOF), and prefills the buffer, and also
8503 * allows us to find out how full the buffer is. We use this information to
8504 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8505 * the returned single char into the target sv, and then go back into scan
8508 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8509 * remaining space in the read-buffer.
8511 * Note that this code despite its twisty-turny nature is pretty darn slick.
8512 * It manages single byte separators, multi-byte cross boundary separators,
8513 * and cross-read-buffer separators cleanly and efficiently at the cost
8514 * of potentially greatly overallocating the target SV.
8520 /* get the number of bytes remaining in the read-ahead buffer
8521 * on first call on a given fp this will return 0.*/
8522 cnt = PerlIO_get_cnt(fp);
8524 /* make sure we have the room */
8525 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8526 /* Not room for all of it
8527 if we are looking for a separator and room for some
8529 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8530 /* just process what we have room for */
8531 shortbuffered = cnt - SvLEN(sv) + append + 1;
8532 cnt -= shortbuffered;
8535 /* ensure that the target sv has enough room to hold
8536 * the rest of the read-ahead buffer */
8538 /* remember that cnt can be negative */
8539 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8543 /* we have enough room to hold the full buffer, lets scream */
8547 /* extract the pointer to sv's string buffer, offset by append as necessary */
8548 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8549 /* extract the point to the read-ahead buffer */
8550 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8552 /* some trace debug output */
8553 DEBUG_P(PerlIO_printf(Perl_debug_log,
8554 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8555 DEBUG_P(PerlIO_printf(Perl_debug_log,
8556 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8558 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8559 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8563 /* if there is stuff left in the read-ahead buffer */
8565 /* if there is a separator */
8567 /* loop until we hit the end of the read-ahead buffer */
8568 while (cnt > 0) { /* this | eat */
8569 /* scan forward copying and searching for rslast as we go */
8571 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8572 goto thats_all_folks; /* screams | sed :-) */
8576 /* no separator, slurp the full buffer */
8577 Copy(ptr, bp, cnt, char); /* this | eat */
8578 bp += cnt; /* screams | dust */
8579 ptr += cnt; /* louder | sed :-) */
8581 assert (!shortbuffered);
8582 goto cannot_be_shortbuffered;
8586 if (shortbuffered) { /* oh well, must extend */
8587 /* we didnt have enough room to fit the line into the target buffer
8588 * so we must extend the target buffer and keep going */
8589 cnt = shortbuffered;
8591 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8593 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8594 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8595 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8599 cannot_be_shortbuffered:
8600 /* we need to refill the read-ahead buffer if possible */
8602 DEBUG_P(PerlIO_printf(Perl_debug_log,
8603 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8604 PTR2UV(ptr),(IV)cnt));
8605 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8607 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8608 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8609 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8610 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8613 call PerlIO_getc() to let it prefill the lookahead buffer
8615 This used to call 'filbuf' in stdio form, but as that behaves like
8616 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8617 another abstraction.
8619 Note we have to deal with the char in 'i' if we are not at EOF
8621 i = PerlIO_getc(fp); /* get more characters */
8623 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8624 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8625 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8626 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8628 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8629 cnt = PerlIO_get_cnt(fp);
8630 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8631 DEBUG_P(PerlIO_printf(Perl_debug_log,
8632 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8633 PTR2UV(ptr),(IV)cnt));
8635 if (i == EOF) /* all done for ever? */
8636 goto thats_really_all_folks;
8638 /* make sure we have enough space in the target sv */
8639 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8641 SvGROW(sv, bpx + cnt + 2);
8642 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8644 /* copy of the char we got from getc() */
8645 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8647 /* make sure we deal with the i being the last character of a separator */
8648 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8649 goto thats_all_folks;
8653 /* check if we have actually found the separator - only really applies
8655 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8656 memNE((char*)bp - rslen, rsptr, rslen))
8657 goto screamer; /* go back to the fray */
8658 thats_really_all_folks:
8660 cnt += shortbuffered;
8661 DEBUG_P(PerlIO_printf(Perl_debug_log,
8662 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8663 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8664 DEBUG_P(PerlIO_printf(Perl_debug_log,
8665 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8667 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8668 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8670 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8671 DEBUG_P(PerlIO_printf(Perl_debug_log,
8672 "Screamer: done, len=%ld, string=|%.*s|\n",
8673 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8677 /*The big, slow, and stupid way. */
8678 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8679 STDCHAR *buf = NULL;
8680 Newx(buf, 8192, STDCHAR);
8688 const STDCHAR * const bpe = buf + sizeof(buf);
8690 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8691 ; /* keep reading */
8695 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8696 /* Accommodate broken VAXC compiler, which applies U8 cast to
8697 * both args of ?: operator, causing EOF to change into 255
8700 i = (U8)buf[cnt - 1];
8706 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8708 sv_catpvn_nomg(sv, (char *) buf, cnt);
8710 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8712 if (i != EOF && /* joy */
8714 SvCUR(sv) < rslen ||
8715 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8719 * If we're reading from a TTY and we get a short read,
8720 * indicating that the user hit his EOF character, we need
8721 * to notice it now, because if we try to read from the TTY
8722 * again, the EOF condition will disappear.
8724 * The comparison of cnt to sizeof(buf) is an optimization
8725 * that prevents unnecessary calls to feof().
8729 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8733 #ifdef USE_HEAP_INSTEAD_OF_STACK
8738 if (rspara) { /* have to do this both before and after */
8739 while (i != EOF) { /* to make sure file boundaries work right */
8740 i = PerlIO_getc(fp);
8742 PerlIO_ungetc(fp,i);
8748 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8754 Auto-increment of the value in the SV, doing string to numeric conversion
8755 if necessary. Handles 'get' magic and operator overloading.
8761 Perl_sv_inc(pTHX_ SV *const sv)
8770 =for apidoc sv_inc_nomg
8772 Auto-increment of the value in the SV, doing string to numeric conversion
8773 if necessary. Handles operator overloading. Skips handling 'get' magic.
8779 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8786 if (SvTHINKFIRST(sv)) {
8787 if (SvREADONLY(sv)) {
8788 Perl_croak_no_modify();
8792 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8794 i = PTR2IV(SvRV(sv));
8798 else sv_force_normal_flags(sv, 0);
8800 flags = SvFLAGS(sv);
8801 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8802 /* It's (privately or publicly) a float, but not tested as an
8803 integer, so test it to see. */
8805 flags = SvFLAGS(sv);
8807 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8808 /* It's publicly an integer, or privately an integer-not-float */
8809 #ifdef PERL_PRESERVE_IVUV
8813 if (SvUVX(sv) == UV_MAX)
8814 sv_setnv(sv, UV_MAX_P1);
8816 (void)SvIOK_only_UV(sv);
8817 SvUV_set(sv, SvUVX(sv) + 1);
8819 if (SvIVX(sv) == IV_MAX)
8820 sv_setuv(sv, (UV)IV_MAX + 1);
8822 (void)SvIOK_only(sv);
8823 SvIV_set(sv, SvIVX(sv) + 1);
8828 if (flags & SVp_NOK) {
8829 const NV was = SvNVX(sv);
8830 if (LIKELY(!Perl_isinfnan(was)) &&
8831 NV_OVERFLOWS_INTEGERS_AT &&
8832 was >= NV_OVERFLOWS_INTEGERS_AT) {
8833 /* diag_listed_as: Lost precision when %s %f by 1 */
8834 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8835 "Lost precision when incrementing %" NVff " by 1",
8838 (void)SvNOK_only(sv);
8839 SvNV_set(sv, was + 1.0);
8843 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8844 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8845 Perl_croak_no_modify();
8847 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8848 if ((flags & SVTYPEMASK) < SVt_PVIV)
8849 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8850 (void)SvIOK_only(sv);
8855 while (isALPHA(*d)) d++;
8856 while (isDIGIT(*d)) d++;
8857 if (d < SvEND(sv)) {
8858 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8859 #ifdef PERL_PRESERVE_IVUV
8860 /* Got to punt this as an integer if needs be, but we don't issue
8861 warnings. Probably ought to make the sv_iv_please() that does
8862 the conversion if possible, and silently. */
8863 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8864 /* Need to try really hard to see if it's an integer.
8865 9.22337203685478e+18 is an integer.
8866 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8867 so $a="9.22337203685478e+18"; $a+0; $a++
8868 needs to be the same as $a="9.22337203685478e+18"; $a++
8875 /* sv_2iv *should* have made this an NV */
8876 if (flags & SVp_NOK) {
8877 (void)SvNOK_only(sv);
8878 SvNV_set(sv, SvNVX(sv) + 1.0);
8881 /* I don't think we can get here. Maybe I should assert this
8882 And if we do get here I suspect that sv_setnv will croak. NWC
8884 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8885 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8887 #endif /* PERL_PRESERVE_IVUV */
8888 if (!numtype && ckWARN(WARN_NUMERIC))
8889 not_incrementable(sv);
8890 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8894 while (d >= SvPVX_const(sv)) {
8902 /* MKS: The original code here died if letters weren't consecutive.
8903 * at least it didn't have to worry about non-C locales. The
8904 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8905 * arranged in order (although not consecutively) and that only
8906 * [A-Za-z] are accepted by isALPHA in the C locale.
8908 if (isALPHA_FOLD_NE(*d, 'z')) {
8909 do { ++*d; } while (!isALPHA(*d));
8912 *(d--) -= 'z' - 'a';
8917 *(d--) -= 'z' - 'a' + 1;
8921 /* oh,oh, the number grew */
8922 SvGROW(sv, SvCUR(sv) + 2);
8923 SvCUR_set(sv, SvCUR(sv) + 1);
8924 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8935 Auto-decrement of the value in the SV, doing string to numeric conversion
8936 if necessary. Handles 'get' magic and operator overloading.
8942 Perl_sv_dec(pTHX_ SV *const sv)
8951 =for apidoc sv_dec_nomg
8953 Auto-decrement of the value in the SV, doing string to numeric conversion
8954 if necessary. Handles operator overloading. Skips handling 'get' magic.
8960 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8966 if (SvTHINKFIRST(sv)) {
8967 if (SvREADONLY(sv)) {
8968 Perl_croak_no_modify();
8972 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8974 i = PTR2IV(SvRV(sv));
8978 else sv_force_normal_flags(sv, 0);
8980 /* Unlike sv_inc we don't have to worry about string-never-numbers
8981 and keeping them magic. But we mustn't warn on punting */
8982 flags = SvFLAGS(sv);
8983 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8984 /* It's publicly an integer, or privately an integer-not-float */
8985 #ifdef PERL_PRESERVE_IVUV
8989 if (SvUVX(sv) == 0) {
8990 (void)SvIOK_only(sv);
8994 (void)SvIOK_only_UV(sv);
8995 SvUV_set(sv, SvUVX(sv) - 1);
8998 if (SvIVX(sv) == IV_MIN) {
8999 sv_setnv(sv, (NV)IV_MIN);
9003 (void)SvIOK_only(sv);
9004 SvIV_set(sv, SvIVX(sv) - 1);
9009 if (flags & SVp_NOK) {
9012 const NV was = SvNVX(sv);
9013 if (LIKELY(!Perl_isinfnan(was)) &&
9014 NV_OVERFLOWS_INTEGERS_AT &&
9015 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9016 /* diag_listed_as: Lost precision when %s %f by 1 */
9017 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9018 "Lost precision when decrementing %" NVff " by 1",
9021 (void)SvNOK_only(sv);
9022 SvNV_set(sv, was - 1.0);
9027 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9028 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9029 Perl_croak_no_modify();
9031 if (!(flags & SVp_POK)) {
9032 if ((flags & SVTYPEMASK) < SVt_PVIV)
9033 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9035 (void)SvIOK_only(sv);
9038 #ifdef PERL_PRESERVE_IVUV
9040 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9041 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9042 /* Need to try really hard to see if it's an integer.
9043 9.22337203685478e+18 is an integer.
9044 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9045 so $a="9.22337203685478e+18"; $a+0; $a--
9046 needs to be the same as $a="9.22337203685478e+18"; $a--
9053 /* sv_2iv *should* have made this an NV */
9054 if (flags & SVp_NOK) {
9055 (void)SvNOK_only(sv);
9056 SvNV_set(sv, SvNVX(sv) - 1.0);
9059 /* I don't think we can get here. Maybe I should assert this
9060 And if we do get here I suspect that sv_setnv will croak. NWC
9062 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9063 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9066 #endif /* PERL_PRESERVE_IVUV */
9067 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9070 /* this define is used to eliminate a chunk of duplicated but shared logic
9071 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9072 * used anywhere but here - yves
9074 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9076 SSize_t ix = ++PL_tmps_ix; \
9077 if (UNLIKELY(ix >= PL_tmps_max)) \
9078 ix = tmps_grow_p(ix); \
9079 PL_tmps_stack[ix] = (AnSv); \
9083 =for apidoc sv_mortalcopy
9085 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9086 The new SV is marked as mortal. It will be destroyed "soon", either by an
9087 explicit call to C<FREETMPS>, or by an implicit call at places such as
9088 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9093 /* Make a string that will exist for the duration of the expression
9094 * evaluation. Actually, it may have to last longer than that, but
9095 * hopefully we won't free it until it has been assigned to a
9096 * permanent location. */
9099 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9103 if (flags & SV_GMAGIC)
9104 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9106 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9107 PUSH_EXTEND_MORTAL__SV_C(sv);
9113 =for apidoc sv_newmortal
9115 Creates a new null SV which is mortal. The reference count of the SV is
9116 set to 1. It will be destroyed "soon", either by an explicit call to
9117 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9118 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9124 Perl_sv_newmortal(pTHX)
9129 SvFLAGS(sv) = SVs_TEMP;
9130 PUSH_EXTEND_MORTAL__SV_C(sv);
9136 =for apidoc newSVpvn_flags
9138 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9139 characters) into it. The reference count for the
9140 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9141 string. You are responsible for ensuring that the source string is at least
9142 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9143 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9144 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9145 returning. If C<SVf_UTF8> is set, C<s>
9146 is considered to be in UTF-8 and the
9147 C<SVf_UTF8> flag will be set on the new SV.
9148 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9150 #define newSVpvn_utf8(s, len, u) \
9151 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9157 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9161 /* All the flags we don't support must be zero.
9162 And we're new code so I'm going to assert this from the start. */
9163 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9165 sv_setpvn(sv,s,len);
9167 /* This code used to do a sv_2mortal(), however we now unroll the call to
9168 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9169 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9170 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9171 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9172 * means that we eliminate quite a few steps than it looks - Yves
9173 * (explaining patch by gfx) */
9175 SvFLAGS(sv) |= flags;
9177 if(flags & SVs_TEMP){
9178 PUSH_EXTEND_MORTAL__SV_C(sv);
9185 =for apidoc sv_2mortal
9187 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9188 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9189 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9190 string buffer can be "stolen" if this SV is copied. See also
9191 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9197 Perl_sv_2mortal(pTHX_ SV *const sv)
9204 PUSH_EXTEND_MORTAL__SV_C(sv);
9212 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9213 characters) into it. The reference count for the
9214 SV is set to 1. If C<len> is zero, Perl will compute the length using
9215 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9216 C<NUL> characters and has to have a terminating C<NUL> byte).
9218 For efficiency, consider using C<newSVpvn> instead.
9224 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9229 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9234 =for apidoc newSVpvn
9236 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9237 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9238 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9239 are responsible for ensuring that the source buffer is at least
9240 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9247 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9251 sv_setpvn(sv,buffer,len);
9256 =for apidoc newSVhek
9258 Creates a new SV from the hash key structure. It will generate scalars that
9259 point to the shared string table where possible. Returns a new (undefined)
9260 SV if C<hek> is NULL.
9266 Perl_newSVhek(pTHX_ const HEK *const hek)
9275 if (HEK_LEN(hek) == HEf_SVKEY) {
9276 return newSVsv(*(SV**)HEK_KEY(hek));
9278 const int flags = HEK_FLAGS(hek);
9279 if (flags & HVhek_WASUTF8) {
9281 Andreas would like keys he put in as utf8 to come back as utf8
9283 STRLEN utf8_len = HEK_LEN(hek);
9284 SV * const sv = newSV_type(SVt_PV);
9285 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9286 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9287 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9290 } else if (flags & HVhek_UNSHARED) {
9291 /* A hash that isn't using shared hash keys has to have
9292 the flag in every key so that we know not to try to call
9293 share_hek_hek on it. */
9295 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9300 /* This will be overwhelminly the most common case. */
9302 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9303 more efficient than sharepvn(). */
9307 sv_upgrade(sv, SVt_PV);
9308 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9309 SvCUR_set(sv, HEK_LEN(hek));
9321 =for apidoc newSVpvn_share
9323 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9324 table. If the string does not already exist in the table, it is
9325 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9326 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9327 is non-zero, that value is used; otherwise the hash is computed.
9328 The string's hash can later be retrieved from the SV
9329 with the C<SvSHARED_HASH()> macro. The idea here is
9330 that as the string table is used for shared hash keys these strings will have
9331 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9337 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9341 bool is_utf8 = FALSE;
9342 const char *const orig_src = src;
9345 STRLEN tmplen = -len;
9347 /* See the note in hv.c:hv_fetch() --jhi */
9348 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9352 PERL_HASH(hash, src, len);
9354 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9355 changes here, update it there too. */
9356 sv_upgrade(sv, SVt_PV);
9357 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9364 if (src != orig_src)
9370 =for apidoc newSVpv_share
9372 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9379 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9381 return newSVpvn_share(src, strlen(src), hash);
9384 #if defined(PERL_IMPLICIT_CONTEXT)
9386 /* pTHX_ magic can't cope with varargs, so this is a no-context
9387 * version of the main function, (which may itself be aliased to us).
9388 * Don't access this version directly.
9392 Perl_newSVpvf_nocontext(const char *const pat, ...)
9398 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9400 va_start(args, pat);
9401 sv = vnewSVpvf(pat, &args);
9408 =for apidoc newSVpvf
9410 Creates a new SV and initializes it with the string formatted like
9417 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9422 PERL_ARGS_ASSERT_NEWSVPVF;
9424 va_start(args, pat);
9425 sv = vnewSVpvf(pat, &args);
9430 /* backend for newSVpvf() and newSVpvf_nocontext() */
9433 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9437 PERL_ARGS_ASSERT_VNEWSVPVF;
9440 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9447 Creates a new SV and copies a floating point value into it.
9448 The reference count for the SV is set to 1.
9454 Perl_newSVnv(pTHX_ const NV n)
9466 Creates a new SV and copies an integer into it. The reference count for the
9473 Perl_newSViv(pTHX_ const IV i)
9479 /* Inlining ONLY the small relevant subset of sv_setiv here
9480 * for performance. Makes a significant difference. */
9482 /* We're starting from SVt_FIRST, so provided that's
9483 * actual 0, we don't have to unset any SV type flags
9484 * to promote to SVt_IV. */
9485 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9487 SET_SVANY_FOR_BODYLESS_IV(sv);
9488 SvFLAGS(sv) |= SVt_IV;
9500 Creates a new SV and copies an unsigned integer into it.
9501 The reference count for the SV is set to 1.
9507 Perl_newSVuv(pTHX_ const UV u)
9511 /* Inlining ONLY the small relevant subset of sv_setuv here
9512 * for performance. Makes a significant difference. */
9514 /* Using ivs is more efficient than using uvs - see sv_setuv */
9515 if (u <= (UV)IV_MAX) {
9516 return newSViv((IV)u);
9521 /* We're starting from SVt_FIRST, so provided that's
9522 * actual 0, we don't have to unset any SV type flags
9523 * to promote to SVt_IV. */
9524 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9526 SET_SVANY_FOR_BODYLESS_IV(sv);
9527 SvFLAGS(sv) |= SVt_IV;
9529 (void)SvIsUV_on(sv);
9538 =for apidoc newSV_type
9540 Creates a new SV, of the type specified. The reference count for the new SV
9547 Perl_newSV_type(pTHX_ const svtype type)
9552 ASSUME(SvTYPE(sv) == SVt_FIRST);
9553 if(type != SVt_FIRST)
9554 sv_upgrade(sv, type);
9559 =for apidoc newRV_noinc
9561 Creates an RV wrapper for an SV. The reference count for the original
9562 SV is B<not> incremented.
9568 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9572 PERL_ARGS_ASSERT_NEWRV_NOINC;
9576 /* We're starting from SVt_FIRST, so provided that's
9577 * actual 0, we don't have to unset any SV type flags
9578 * to promote to SVt_IV. */
9579 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9581 SET_SVANY_FOR_BODYLESS_IV(sv);
9582 SvFLAGS(sv) |= SVt_IV;
9587 SvRV_set(sv, tmpRef);
9592 /* newRV_inc is the official function name to use now.
9593 * newRV_inc is in fact #defined to newRV in sv.h
9597 Perl_newRV(pTHX_ SV *const sv)
9599 PERL_ARGS_ASSERT_NEWRV;
9601 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9607 Creates a new SV which is an exact duplicate of the original SV.
9614 Perl_newSVsv(pTHX_ SV *const old)
9620 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9621 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9624 /* Do this here, otherwise we leak the new SV if this croaks. */
9627 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9628 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9629 sv_setsv_flags(sv, old, SV_NOSTEAL);
9634 =for apidoc sv_reset
9636 Underlying implementation for the C<reset> Perl function.
9637 Note that the perl-level function is vaguely deprecated.
9643 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9645 PERL_ARGS_ASSERT_SV_RESET;
9647 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9651 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9653 char todo[PERL_UCHAR_MAX+1];
9656 if (!stash || SvTYPE(stash) != SVt_PVHV)
9659 if (!s) { /* reset ?? searches */
9660 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9662 const U32 count = mg->mg_len / sizeof(PMOP**);
9663 PMOP **pmp = (PMOP**) mg->mg_ptr;
9664 PMOP *const *const end = pmp + count;
9668 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9670 (*pmp)->op_pmflags &= ~PMf_USED;
9678 /* reset variables */
9680 if (!HvARRAY(stash))
9683 Zero(todo, 256, char);
9687 I32 i = (unsigned char)*s;
9691 max = (unsigned char)*s++;
9692 for ( ; i <= max; i++) {
9695 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9697 for (entry = HvARRAY(stash)[i];
9699 entry = HeNEXT(entry))
9704 if (!todo[(U8)*HeKEY(entry)])
9706 gv = MUTABLE_GV(HeVAL(entry));
9710 if (sv && !SvREADONLY(sv)) {
9711 SV_CHECK_THINKFIRST_COW_DROP(sv);
9712 if (!isGV(sv)) SvOK_off(sv);
9717 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9728 Using various gambits, try to get an IO from an SV: the IO slot if its a
9729 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9730 named after the PV if we're a string.
9732 'Get' magic is ignored on the C<sv> passed in, but will be called on
9733 C<SvRV(sv)> if C<sv> is an RV.
9739 Perl_sv_2io(pTHX_ SV *const sv)
9744 PERL_ARGS_ASSERT_SV_2IO;
9746 switch (SvTYPE(sv)) {
9748 io = MUTABLE_IO(sv);
9752 if (isGV_with_GP(sv)) {
9753 gv = MUTABLE_GV(sv);
9756 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9757 HEKfARG(GvNAME_HEK(gv)));
9763 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9765 SvGETMAGIC(SvRV(sv));
9766 return sv_2io(SvRV(sv));
9768 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9775 if (SvGMAGICAL(sv)) {
9776 newsv = sv_newmortal();
9777 sv_setsv_nomg(newsv, sv);
9779 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9789 Using various gambits, try to get a CV from an SV; in addition, try if
9790 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9791 The flags in C<lref> are passed to C<gv_fetchsv>.
9797 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9802 PERL_ARGS_ASSERT_SV_2CV;
9809 switch (SvTYPE(sv)) {
9813 return MUTABLE_CV(sv);
9823 sv = amagic_deref_call(sv, to_cv_amg);
9826 if (SvTYPE(sv) == SVt_PVCV) {
9827 cv = MUTABLE_CV(sv);
9832 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9833 gv = MUTABLE_GV(sv);
9835 Perl_croak(aTHX_ "Not a subroutine reference");
9837 else if (isGV_with_GP(sv)) {
9838 gv = MUTABLE_GV(sv);
9841 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9848 /* Some flags to gv_fetchsv mean don't really create the GV */
9849 if (!isGV_with_GP(gv)) {
9854 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9855 /* XXX this is probably not what they think they're getting.
9856 * It has the same effect as "sub name;", i.e. just a forward
9867 Returns true if the SV has a true value by Perl's rules.
9868 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9869 instead use an in-line version.
9875 Perl_sv_true(pTHX_ SV *const sv)
9880 const XPV* const tXpv = (XPV*)SvANY(sv);
9882 (tXpv->xpv_cur > 1 ||
9883 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9890 return SvIVX(sv) != 0;
9893 return SvNVX(sv) != 0.0;
9895 return sv_2bool(sv);
9901 =for apidoc sv_pvn_force
9903 Get a sensible string out of the SV somehow.
9904 A private implementation of the C<SvPV_force> macro for compilers which
9905 can't cope with complex macro expressions. Always use the macro instead.
9907 =for apidoc sv_pvn_force_flags
9909 Get a sensible string out of the SV somehow.
9910 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9911 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9912 implemented in terms of this function.
9913 You normally want to use the various wrapper macros instead: see
9914 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9920 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9922 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9924 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9925 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9926 sv_force_normal_flags(sv, 0);
9936 if (SvTYPE(sv) > SVt_PVLV
9937 || isGV_with_GP(sv))
9938 /* diag_listed_as: Can't coerce %s to %s in %s */
9939 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9941 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9948 if (SvTYPE(sv) < SVt_PV ||
9949 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9952 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9953 SvGROW(sv, len + 1);
9954 Move(s,SvPVX(sv),len,char);
9956 SvPVX(sv)[len] = '\0';
9959 SvPOK_on(sv); /* validate pointer */
9961 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9962 PTR2UV(sv),SvPVX_const(sv)));
9965 (void)SvPOK_only_UTF8(sv);
9966 return SvPVX_mutable(sv);
9970 =for apidoc sv_pvbyten_force
9972 The backend for the C<SvPVbytex_force> macro. Always use the macro
9979 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9981 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9983 sv_pvn_force(sv,lp);
9984 sv_utf8_downgrade(sv,0);
9990 =for apidoc sv_pvutf8n_force
9992 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9999 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10001 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10003 sv_pvn_force(sv,0);
10004 sv_utf8_upgrade_nomg(sv);
10010 =for apidoc sv_reftype
10012 Returns a string describing what the SV is a reference to.
10014 If ob is true and the SV is blessed, the string is the class name,
10015 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10021 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10023 PERL_ARGS_ASSERT_SV_REFTYPE;
10024 if (ob && SvOBJECT(sv)) {
10025 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10028 /* WARNING - There is code, for instance in mg.c, that assumes that
10029 * the only reason that sv_reftype(sv,0) would return a string starting
10030 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10031 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10032 * this routine inside other subs, and it saves time.
10033 * Do not change this assumption without searching for "dodgy type check" in
10036 switch (SvTYPE(sv)) {
10051 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10052 /* tied lvalues should appear to be
10053 * scalars for backwards compatibility */
10054 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10055 ? "SCALAR" : "LVALUE");
10056 case SVt_PVAV: return "ARRAY";
10057 case SVt_PVHV: return "HASH";
10058 case SVt_PVCV: return "CODE";
10059 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10060 ? "GLOB" : "SCALAR");
10061 case SVt_PVFM: return "FORMAT";
10062 case SVt_PVIO: return "IO";
10063 case SVt_INVLIST: return "INVLIST";
10064 case SVt_REGEXP: return "REGEXP";
10065 default: return "UNKNOWN";
10073 Returns a SV describing what the SV passed in is a reference to.
10075 dst can be a SV to be set to the description or NULL, in which case a
10076 mortal SV is returned.
10078 If ob is true and the SV is blessed, the description is the class
10079 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10085 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10087 PERL_ARGS_ASSERT_SV_REF;
10090 dst = sv_newmortal();
10092 if (ob && SvOBJECT(sv)) {
10093 HvNAME_get(SvSTASH(sv))
10094 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10095 : sv_setpvs(dst, "__ANON__");
10098 const char * reftype = sv_reftype(sv, 0);
10099 sv_setpv(dst, reftype);
10105 =for apidoc sv_isobject
10107 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10108 object. If the SV is not an RV, or if the object is not blessed, then this
10115 Perl_sv_isobject(pTHX_ SV *sv)
10131 Returns a boolean indicating whether the SV is blessed into the specified
10132 class. This does not check for subtypes; use C<sv_derived_from> to verify
10133 an inheritance relationship.
10139 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10141 const char *hvname;
10143 PERL_ARGS_ASSERT_SV_ISA;
10153 hvname = HvNAME_get(SvSTASH(sv));
10157 return strEQ(hvname, name);
10161 =for apidoc newSVrv
10163 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10164 RV then it will be upgraded to one. If C<classname> is non-null then the new
10165 SV will be blessed in the specified package. The new SV is returned and its
10166 reference count is 1. The reference count 1 is owned by C<rv>.
10172 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10176 PERL_ARGS_ASSERT_NEWSVRV;
10180 SV_CHECK_THINKFIRST_COW_DROP(rv);
10182 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10183 const U32 refcnt = SvREFCNT(rv);
10187 SvREFCNT(rv) = refcnt;
10189 sv_upgrade(rv, SVt_IV);
10190 } else if (SvROK(rv)) {
10191 SvREFCNT_dec(SvRV(rv));
10193 prepare_SV_for_RV(rv);
10201 HV* const stash = gv_stashpv(classname, GV_ADD);
10202 (void)sv_bless(rv, stash);
10208 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10210 SV * const lv = newSV_type(SVt_PVLV);
10211 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10213 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10214 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10215 LvSTARGOFF(lv) = ix;
10216 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10221 =for apidoc sv_setref_pv
10223 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10224 argument will be upgraded to an RV. That RV will be modified to point to
10225 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10226 into the SV. The C<classname> argument indicates the package for the
10227 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10228 will have a reference count of 1, and the RV will be returned.
10230 Do not use with other Perl types such as HV, AV, SV, CV, because those
10231 objects will become corrupted by the pointer copy process.
10233 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10239 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10241 PERL_ARGS_ASSERT_SV_SETREF_PV;
10244 sv_setsv(rv, &PL_sv_undef);
10248 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10253 =for apidoc sv_setref_iv
10255 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10256 argument will be upgraded to an RV. That RV will be modified to point to
10257 the new SV. The C<classname> argument indicates the package for the
10258 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10259 will have a reference count of 1, and the RV will be returned.
10265 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10267 PERL_ARGS_ASSERT_SV_SETREF_IV;
10269 sv_setiv(newSVrv(rv,classname), iv);
10274 =for apidoc sv_setref_uv
10276 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10277 argument will be upgraded to an RV. That RV will be modified to point to
10278 the new SV. The C<classname> argument indicates the package for the
10279 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10280 will have a reference count of 1, and the RV will be returned.
10286 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10288 PERL_ARGS_ASSERT_SV_SETREF_UV;
10290 sv_setuv(newSVrv(rv,classname), uv);
10295 =for apidoc sv_setref_nv
10297 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10298 argument will be upgraded to an RV. That RV will be modified to point to
10299 the new SV. The C<classname> argument indicates the package for the
10300 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10301 will have a reference count of 1, and the RV will be returned.
10307 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10309 PERL_ARGS_ASSERT_SV_SETREF_NV;
10311 sv_setnv(newSVrv(rv,classname), nv);
10316 =for apidoc sv_setref_pvn
10318 Copies a string into a new SV, optionally blessing the SV. The length of the
10319 string must be specified with C<n>. The C<rv> argument will be upgraded to
10320 an RV. That RV will be modified to point to the new SV. The C<classname>
10321 argument indicates the package for the blessing. Set C<classname> to
10322 C<NULL> to avoid the blessing. The new SV will have a reference count
10323 of 1, and the RV will be returned.
10325 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10331 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10332 const char *const pv, const STRLEN n)
10334 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10336 sv_setpvn(newSVrv(rv,classname), pv, n);
10341 =for apidoc sv_bless
10343 Blesses an SV into a specified package. The SV must be an RV. The package
10344 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10345 of the SV is unaffected.
10351 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10354 HV *oldstash = NULL;
10356 PERL_ARGS_ASSERT_SV_BLESS;
10360 Perl_croak(aTHX_ "Can't bless non-reference value");
10362 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10363 if (SvREADONLY(tmpRef))
10364 Perl_croak_no_modify();
10365 if (SvOBJECT(tmpRef)) {
10366 oldstash = SvSTASH(tmpRef);
10369 SvOBJECT_on(tmpRef);
10370 SvUPGRADE(tmpRef, SVt_PVMG);
10371 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10372 SvREFCNT_dec(oldstash);
10374 if(SvSMAGICAL(tmpRef))
10375 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10383 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10384 * as it is after unglobbing it.
10387 PERL_STATIC_INLINE void
10388 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10392 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10394 PERL_ARGS_ASSERT_SV_UNGLOB;
10396 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10398 if (!(flags & SV_COW_DROP_PV))
10399 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10401 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10403 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10404 && HvNAME_get(stash))
10405 mro_method_changed_in(stash);
10406 gp_free(MUTABLE_GV(sv));
10409 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10410 GvSTASH(sv) = NULL;
10413 if (GvNAME_HEK(sv)) {
10414 unshare_hek(GvNAME_HEK(sv));
10416 isGV_with_GP_off(sv);
10418 if(SvTYPE(sv) == SVt_PVGV) {
10419 /* need to keep SvANY(sv) in the right arena */
10420 xpvmg = new_XPVMG();
10421 StructCopy(SvANY(sv), xpvmg, XPVMG);
10422 del_XPVGV(SvANY(sv));
10425 SvFLAGS(sv) &= ~SVTYPEMASK;
10426 SvFLAGS(sv) |= SVt_PVMG;
10429 /* Intentionally not calling any local SET magic, as this isn't so much a
10430 set operation as merely an internal storage change. */
10431 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10432 else sv_setsv_flags(sv, temp, 0);
10434 if ((const GV *)sv == PL_last_in_gv)
10435 PL_last_in_gv = NULL;
10436 else if ((const GV *)sv == PL_statgv)
10441 =for apidoc sv_unref_flags
10443 Unsets the RV status of the SV, and decrements the reference count of
10444 whatever was being referenced by the RV. This can almost be thought of
10445 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10446 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10447 (otherwise the decrementing is conditional on the reference count being
10448 different from one or the reference being a readonly SV).
10449 See C<L</SvROK_off>>.
10455 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10457 SV* const target = SvRV(ref);
10459 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10461 if (SvWEAKREF(ref)) {
10462 sv_del_backref(target, ref);
10463 SvWEAKREF_off(ref);
10464 SvRV_set(ref, NULL);
10467 SvRV_set(ref, NULL);
10469 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10470 assigned to as BEGIN {$a = \"Foo"} will fail. */
10471 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10472 SvREFCNT_dec_NN(target);
10473 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10474 sv_2mortal(target); /* Schedule for freeing later */
10478 =for apidoc sv_untaint
10480 Untaint an SV. Use C<SvTAINTED_off> instead.
10486 Perl_sv_untaint(pTHX_ SV *const sv)
10488 PERL_ARGS_ASSERT_SV_UNTAINT;
10489 PERL_UNUSED_CONTEXT;
10491 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10492 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10499 =for apidoc sv_tainted
10501 Test an SV for taintedness. Use C<SvTAINTED> instead.
10507 Perl_sv_tainted(pTHX_ SV *const sv)
10509 PERL_ARGS_ASSERT_SV_TAINTED;
10510 PERL_UNUSED_CONTEXT;
10512 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10513 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10514 if (mg && (mg->mg_len & 1) )
10520 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10521 private to this file */
10524 =for apidoc sv_setpviv
10526 Copies an integer into the given SV, also updating its string value.
10527 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10533 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10535 char buf[TYPE_CHARS(UV)];
10537 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10539 PERL_ARGS_ASSERT_SV_SETPVIV;
10541 sv_setpvn(sv, ptr, ebuf - ptr);
10545 =for apidoc sv_setpviv_mg
10547 Like C<sv_setpviv>, but also handles 'set' magic.
10553 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10555 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10557 sv_setpviv(sv, iv);
10561 #endif /* NO_MATHOMS */
10563 #if defined(PERL_IMPLICIT_CONTEXT)
10565 /* pTHX_ magic can't cope with varargs, so this is a no-context
10566 * version of the main function, (which may itself be aliased to us).
10567 * Don't access this version directly.
10571 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10576 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10578 va_start(args, pat);
10579 sv_vsetpvf(sv, pat, &args);
10583 /* pTHX_ magic can't cope with varargs, so this is a no-context
10584 * version of the main function, (which may itself be aliased to us).
10585 * Don't access this version directly.
10589 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10594 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10596 va_start(args, pat);
10597 sv_vsetpvf_mg(sv, pat, &args);
10603 =for apidoc sv_setpvf
10605 Works like C<sv_catpvf> but copies the text into the SV instead of
10606 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10612 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10616 PERL_ARGS_ASSERT_SV_SETPVF;
10618 va_start(args, pat);
10619 sv_vsetpvf(sv, pat, &args);
10624 =for apidoc sv_vsetpvf
10626 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10627 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10629 Usually used via its frontend C<sv_setpvf>.
10635 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10637 PERL_ARGS_ASSERT_SV_VSETPVF;
10639 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10643 =for apidoc sv_setpvf_mg
10645 Like C<sv_setpvf>, but also handles 'set' magic.
10651 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10655 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10657 va_start(args, pat);
10658 sv_vsetpvf_mg(sv, pat, &args);
10663 =for apidoc sv_vsetpvf_mg
10665 Like C<sv_vsetpvf>, but also handles 'set' magic.
10667 Usually used via its frontend C<sv_setpvf_mg>.
10673 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10675 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10677 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10681 #if defined(PERL_IMPLICIT_CONTEXT)
10683 /* pTHX_ magic can't cope with varargs, so this is a no-context
10684 * version of the main function, (which may itself be aliased to us).
10685 * Don't access this version directly.
10689 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10694 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10696 va_start(args, pat);
10697 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10701 /* pTHX_ magic can't cope with varargs, so this is a no-context
10702 * version of the main function, (which may itself be aliased to us).
10703 * Don't access this version directly.
10707 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10712 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10714 va_start(args, pat);
10715 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10722 =for apidoc sv_catpvf
10724 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10725 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10726 variable argument list, argument reordering is not supported.
10727 If the appended data contains "wide" characters
10728 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10729 and characters >255 formatted with C<%c>), the original SV might get
10730 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10731 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10732 valid UTF-8; if the original SV was bytes, the pattern should be too.
10737 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10741 PERL_ARGS_ASSERT_SV_CATPVF;
10743 va_start(args, pat);
10744 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10749 =for apidoc sv_vcatpvf
10751 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10752 variable argument list, and appends the formatted output
10753 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10755 Usually used via its frontend C<sv_catpvf>.
10761 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10763 PERL_ARGS_ASSERT_SV_VCATPVF;
10765 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10769 =for apidoc sv_catpvf_mg
10771 Like C<sv_catpvf>, but also handles 'set' magic.
10777 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10781 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10783 va_start(args, pat);
10784 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10790 =for apidoc sv_vcatpvf_mg
10792 Like C<sv_vcatpvf>, but also handles 'set' magic.
10794 Usually used via its frontend C<sv_catpvf_mg>.
10800 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10802 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10804 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10809 =for apidoc sv_vsetpvfn
10811 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10814 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10820 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10821 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10823 PERL_ARGS_ASSERT_SV_VSETPVFN;
10826 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10831 * Warn of missing argument to sprintf. The value used in place of such
10832 * arguments should be &PL_sv_no; an undefined value would yield
10833 * inappropriate "use of uninit" warnings [perl #71000].
10836 S_warn_vcatpvfn_missing_argument(pTHX) {
10837 if (ckWARN(WARN_MISSING)) {
10838 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10839 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10845 S_expect_number(pTHX_ char **const pattern)
10849 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10851 switch (**pattern) {
10852 case '1': case '2': case '3':
10853 case '4': case '5': case '6':
10854 case '7': case '8': case '9':
10855 var = *(*pattern)++ - '0';
10856 while (isDIGIT(**pattern)) {
10857 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10859 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10867 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10869 const int neg = nv < 0;
10872 PERL_ARGS_ASSERT_F0CONVERT;
10874 if (UNLIKELY(Perl_isinfnan(nv))) {
10875 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10885 if (uv & 1 && uv == nv)
10886 uv--; /* Round to even */
10888 const unsigned dig = uv % 10;
10890 } while (uv /= 10);
10901 =for apidoc sv_vcatpvfn
10903 =for apidoc sv_vcatpvfn_flags
10905 Processes its arguments like C<vsprintf> and appends the formatted output
10906 to an SV. Uses an array of SVs if the C-style variable argument list is
10907 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10908 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10909 C<va_list> argument list with a format string that uses argument reordering
10910 will yield an exception.
10912 When running with taint checks enabled, indicates via
10913 C<maybe_tainted> if results are untrustworthy (often due to the use of
10916 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10918 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10923 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10924 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10925 vec_utf8 = DO_UTF8(vecsv);
10927 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10930 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10931 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10933 PERL_ARGS_ASSERT_SV_VCATPVFN;
10935 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10938 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10939 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10940 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10941 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10942 * after the first 1023 zero bits.
10944 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10945 * of dynamically growing buffer might be better, start at just 16 bytes
10946 * (for example) and grow only when necessary. Or maybe just by looking
10947 * at the exponents of the two doubles? */
10948 # define DOUBLEDOUBLE_MAXBITS 2098
10951 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10952 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10953 * per xdigit. For the double-double case, this can be rather many.
10954 * The non-double-double-long-double overshoots since all bits of NV
10955 * are not mantissa bits, there are also exponent bits. */
10956 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10957 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10959 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10962 /* If we do not have a known long double format, (including not using
10963 * long doubles, or long doubles being equal to doubles) then we will
10964 * fall back to the ldexp/frexp route, with which we can retrieve at
10965 * most as many bits as our widest unsigned integer type is. We try
10966 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10968 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10969 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10971 #if defined(HAS_QUAD) && defined(Uquad_t)
10972 # define MANTISSATYPE Uquad_t
10973 # define MANTISSASIZE 8
10975 # define MANTISSATYPE UV
10976 # define MANTISSASIZE UVSIZE
10979 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10980 # define HEXTRACT_LITTLE_ENDIAN
10981 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10982 # define HEXTRACT_BIG_ENDIAN
10984 # define HEXTRACT_MIX_ENDIAN
10987 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10988 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10989 * are being extracted from (either directly from the long double in-memory
10990 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10991 * is used to update the exponent. The subnormal is set to true
10992 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
10993 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
10995 * The tricky part is that S_hextract() needs to be called twice:
10996 * the first time with vend as NULL, and the second time with vend as
10997 * the pointer returned by the first call. What happens is that on
10998 * the first round the output size is computed, and the intended
10999 * extraction sanity checked. On the second round the actual output
11000 * (the extraction of the hexadecimal values) takes place.
11001 * Sanity failures cause fatal failures during both rounds. */
11003 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11004 U8* vhex, U8* vend)
11008 int ixmin = 0, ixmax = 0;
11010 /* XXX Inf/NaN are not handled here, since it is
11011 * assumed they are to be output as "Inf" and "NaN". */
11013 /* These macros are just to reduce typos, they have multiple
11014 * repetitions below, but usually only one (or sometimes two)
11015 * of them is really being used. */
11016 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11017 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11018 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11019 #define HEXTRACT_OUTPUT(ix) \
11021 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11023 #define HEXTRACT_COUNT(ix, c) \
11025 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11027 #define HEXTRACT_BYTE(ix) \
11029 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11031 #define HEXTRACT_LO_NYBBLE(ix) \
11033 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11035 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11036 * to make it look less odd when the top bits of a NV
11037 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11038 * order bits can be in the "low nybble" of a byte. */
11039 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11040 #define HEXTRACT_BYTES_LE(a, b) \
11041 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11042 #define HEXTRACT_BYTES_BE(a, b) \
11043 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11044 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11045 #define HEXTRACT_IMPLICIT_BIT(nv) \
11047 if (!*subnormal) { \
11048 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11052 /* Most formats do. Those which don't should undef this.
11054 * But also note that IEEE 754 subnormals do not have it, or,
11055 * expressed alternatively, their implicit bit is zero. */
11056 #define HEXTRACT_HAS_IMPLICIT_BIT
11058 /* Many formats do. Those which don't should undef this. */
11059 #define HEXTRACT_HAS_TOP_NYBBLE
11061 /* HEXTRACTSIZE is the maximum number of xdigits. */
11062 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11063 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11065 # define HEXTRACTSIZE 2 * NVSIZE
11068 const U8* vmaxend = vhex + HEXTRACTSIZE;
11069 PERL_UNUSED_VAR(ix); /* might happen */
11070 (void)Perl_frexp(PERL_ABS(nv), exponent);
11071 *subnormal = FALSE;
11072 if (vend && (vend <= vhex || vend > vmaxend)) {
11073 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11074 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11077 /* First check if using long doubles. */
11078 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11079 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11080 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11081 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11082 /* The bytes 13..0 are the mantissa/fraction,
11083 * the 15,14 are the sign+exponent. */
11084 const U8* nvp = (const U8*)(&nv);
11085 HEXTRACT_GET_SUBNORMAL(nv);
11086 HEXTRACT_IMPLICIT_BIT(nv);
11087 # undef HEXTRACT_HAS_TOP_NYBBLE
11088 HEXTRACT_BYTES_LE(13, 0);
11089 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11090 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11091 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11092 /* The bytes 2..15 are the mantissa/fraction,
11093 * the 0,1 are the sign+exponent. */
11094 const U8* nvp = (const U8*)(&nv);
11095 HEXTRACT_GET_SUBNORMAL(nv);
11096 HEXTRACT_IMPLICIT_BIT(nv);
11097 # undef HEXTRACT_HAS_TOP_NYBBLE
11098 HEXTRACT_BYTES_BE(2, 15);
11099 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11100 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11101 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11102 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11103 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11104 /* The bytes 0..1 are the sign+exponent,
11105 * the bytes 2..9 are the mantissa/fraction. */
11106 const U8* nvp = (const U8*)(&nv);
11107 # undef HEXTRACT_HAS_IMPLICIT_BIT
11108 # undef HEXTRACT_HAS_TOP_NYBBLE
11109 HEXTRACT_GET_SUBNORMAL(nv);
11110 HEXTRACT_BYTES_LE(7, 0);
11111 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11112 /* Does this format ever happen? (Wikipedia says the Motorola
11113 * 6888x math coprocessors used format _like_ this but padded
11114 * to 96 bits with 16 unused bits between the exponent and the
11116 const U8* nvp = (const U8*)(&nv);
11117 # undef HEXTRACT_HAS_IMPLICIT_BIT
11118 # undef HEXTRACT_HAS_TOP_NYBBLE
11119 HEXTRACT_GET_SUBNORMAL(nv);
11120 HEXTRACT_BYTES_BE(0, 7);
11122 # define HEXTRACT_FALLBACK
11123 /* Double-double format: two doubles next to each other.
11124 * The first double is the high-order one, exactly like
11125 * it would be for a "lone" double. The second double
11126 * is shifted down using the exponent so that that there
11127 * are no common bits. The tricky part is that the value
11128 * of the double-double is the SUM of the two doubles and
11129 * the second one can be also NEGATIVE.
11131 * Because of this tricky construction the bytewise extraction we
11132 * use for the other long double formats doesn't work, we must
11133 * extract the values bit by bit.
11135 * The little-endian double-double is used .. somewhere?
11137 * The big endian double-double is used in e.g. PPC/Power (AIX)
11140 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11141 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11142 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11145 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11146 /* Using normal doubles, not long doubles.
11148 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11149 * bytes, since we might need to handle printf precision, and
11150 * also need to insert the radix. */
11152 # ifdef HEXTRACT_LITTLE_ENDIAN
11153 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11154 const U8* nvp = (const U8*)(&nv);
11155 HEXTRACT_GET_SUBNORMAL(nv);
11156 HEXTRACT_IMPLICIT_BIT(nv);
11157 HEXTRACT_TOP_NYBBLE(6);
11158 HEXTRACT_BYTES_LE(5, 0);
11159 # elif defined(HEXTRACT_BIG_ENDIAN)
11160 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11161 const U8* nvp = (const U8*)(&nv);
11162 HEXTRACT_GET_SUBNORMAL(nv);
11163 HEXTRACT_IMPLICIT_BIT(nv);
11164 HEXTRACT_TOP_NYBBLE(1);
11165 HEXTRACT_BYTES_BE(2, 7);
11166 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11167 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11168 const U8* nvp = (const U8*)(&nv);
11169 HEXTRACT_GET_SUBNORMAL(nv);
11170 HEXTRACT_IMPLICIT_BIT(nv);
11171 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11172 HEXTRACT_BYTE(1); /* 5 */
11173 HEXTRACT_BYTE(0); /* 4 */
11174 HEXTRACT_BYTE(7); /* 3 */
11175 HEXTRACT_BYTE(6); /* 2 */
11176 HEXTRACT_BYTE(5); /* 1 */
11177 HEXTRACT_BYTE(4); /* 0 */
11178 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11179 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11180 const U8* nvp = (const U8*)(&nv);
11181 HEXTRACT_GET_SUBNORMAL(nv);
11182 HEXTRACT_IMPLICIT_BIT(nv);
11183 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11184 HEXTRACT_BYTE(6); /* 5 */
11185 HEXTRACT_BYTE(7); /* 4 */
11186 HEXTRACT_BYTE(0); /* 3 */
11187 HEXTRACT_BYTE(1); /* 2 */
11188 HEXTRACT_BYTE(2); /* 1 */
11189 HEXTRACT_BYTE(3); /* 0 */
11191 # define HEXTRACT_FALLBACK
11194 # define HEXTRACT_FALLBACK
11196 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11197 # ifdef HEXTRACT_FALLBACK
11198 HEXTRACT_GET_SUBNORMAL(nv);
11199 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11200 /* The fallback is used for the double-double format, and
11201 * for unknown long double formats, and for unknown double
11202 * formats, or in general unknown NV formats. */
11203 if (nv == (NV)0.0) {
11211 NV d = nv < 0 ? -nv : nv;
11213 U8 ha = 0x0; /* hexvalue accumulator */
11214 U8 hd = 0x8; /* hexvalue digit */
11216 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11217 * this is essentially manual frexp(). Multiplying by 0.5 and
11218 * doubling should be lossless in binary floating point. */
11228 while (d >= e + e) {
11232 /* Now e <= d < 2*e */
11234 /* First extract the leading hexdigit (the implicit bit). */
11250 /* Then extract the remaining hexdigits. */
11251 while (d > (NV)0.0) {
11257 /* Output or count in groups of four bits,
11258 * that is, when the hexdigit is down to one. */
11263 /* Reset the hexvalue. */
11272 /* Flush possible pending hexvalue. */
11282 /* Croak for various reasons: if the output pointer escaped the
11283 * output buffer, if the extraction index escaped the extraction
11284 * buffer, or if the ending output pointer didn't match the
11285 * previously computed value. */
11286 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11287 /* For double-double the ixmin and ixmax stay at zero,
11288 * which is convenient since the HEXTRACTSIZE is tricky
11289 * for double-double. */
11290 ixmin < 0 || ixmax >= NVSIZE ||
11291 (vend && v != vend)) {
11292 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11293 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11298 /* Helper for sv_vcatpvfn_flags(). */
11299 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11304 (var) = &PL_sv_no; /* [perl #71000] */ \
11305 arg_missing = TRUE; \
11310 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11311 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11316 const char *patend;
11319 static const char nullstr[] = "(null)";
11321 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11322 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11324 /* Times 4: a decimal digit takes more than 3 binary digits.
11325 * NV_DIG: mantissa takes than many decimal digits.
11326 * Plus 32: Playing safe. */
11327 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11328 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11329 bool hexfp = FALSE; /* hexadecimal floating point? */
11331 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11333 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11334 PERL_UNUSED_ARG(maybe_tainted);
11336 if (flags & SV_GMAGIC)
11339 /* no matter what, this is a string now */
11340 (void)SvPV_force_nomg(sv, origlen);
11342 /* special-case "", "%s", and "%-p" (SVf - see below) */
11344 if (svmax && ckWARN(WARN_REDUNDANT))
11345 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11346 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11349 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11350 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11351 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11352 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11355 const char * const s = va_arg(*args, char*);
11356 sv_catpv_nomg(sv, s ? s : nullstr);
11358 else if (svix < svmax) {
11359 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11360 SvGETMAGIC(*svargs);
11361 sv_catsv_nomg(sv, *svargs);
11364 S_warn_vcatpvfn_missing_argument(aTHX);
11367 if (args && patlen == 3 && pat[0] == '%' &&
11368 pat[1] == '-' && pat[2] == 'p') {
11369 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11370 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11371 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11372 argsv = MUTABLE_SV(va_arg(*args, void*));
11373 sv_catsv_nomg(sv, argsv);
11377 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11378 /* special-case "%.<number>[gf]" */
11379 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11380 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11381 unsigned digits = 0;
11385 while (*pp >= '0' && *pp <= '9')
11386 digits = 10 * digits + (*pp++ - '0');
11388 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11389 format the first argument and WARN_REDUNDANT if svmax > 1?
11390 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11391 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11392 const NV nv = SvNV(*svargs);
11393 if (LIKELY(!Perl_isinfnan(nv))) {
11395 /* Add check for digits != 0 because it seems that some
11396 gconverts are buggy in this case, and we don't yet have
11397 a Configure test for this. */
11398 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11399 /* 0, point, slack */
11400 STORE_LC_NUMERIC_SET_TO_NEEDED();
11401 SNPRINTF_G(nv, ebuf, size, digits);
11402 sv_catpv_nomg(sv, ebuf);
11403 if (*ebuf) /* May return an empty string for digits==0 */
11406 } else if (!digits) {
11409 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11410 sv_catpvn_nomg(sv, p, l);
11417 #endif /* !USE_LONG_DOUBLE */
11419 if (!args && svix < svmax && DO_UTF8(*svargs))
11422 patend = (char*)pat + patlen;
11423 for (p = (char*)pat; p < patend; p = q) {
11426 bool vectorize = FALSE;
11427 bool vectorarg = FALSE;
11428 bool vec_utf8 = FALSE;
11434 bool has_precis = FALSE;
11436 const I32 osvix = svix;
11437 bool is_utf8 = FALSE; /* is this item utf8? */
11438 bool used_explicit_ix = FALSE;
11439 bool arg_missing = FALSE;
11440 #ifdef HAS_LDBL_SPRINTF_BUG
11441 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11442 with sfio - Allen <allens@cpan.org> */
11443 bool fix_ldbl_sprintf_bug = FALSE;
11447 U8 utf8buf[UTF8_MAXBYTES+1];
11448 STRLEN esignlen = 0;
11450 const char *eptr = NULL;
11451 const char *fmtstart;
11454 const U8 *vecstr = NULL;
11461 /* We need a long double target in case HAS_LONG_DOUBLE,
11462 * even without USE_LONG_DOUBLE, so that we can printf with
11463 * long double formats, even without NV being long double.
11464 * But we call the target 'fv' instead of 'nv', since most of
11465 * the time it is not (most compilers these days recognize
11466 * "long double", even if only as a synonym for "double").
11468 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11469 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11471 # ifdef Perl_isfinitel
11472 # define FV_ISFINITE(x) Perl_isfinitel(x)
11474 # define FV_GF PERL_PRIgldbl
11475 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11476 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11477 # define NV_TO_FV(nv,fv) STMT_START { \
11479 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11482 # define NV_TO_FV(nv,fv) (fv)=(nv)
11486 # define FV_GF NVgf
11487 # define NV_TO_FV(nv,fv) (fv)=(nv)
11489 #ifndef FV_ISFINITE
11490 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11496 const char *dotstr = ".";
11497 STRLEN dotstrlen = 1;
11498 I32 efix = 0; /* explicit format parameter index */
11499 I32 ewix = 0; /* explicit width index */
11500 I32 epix = 0; /* explicit precision index */
11501 I32 evix = 0; /* explicit vector index */
11502 bool asterisk = FALSE;
11503 bool infnan = FALSE;
11505 /* echo everything up to the next format specification */
11506 for (q = p; q < patend && *q != '%'; ++q) ;
11508 if (has_utf8 && !pat_utf8)
11509 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11511 sv_catpvn_nomg(sv, p, q - p);
11520 We allow format specification elements in this order:
11521 \d+\$ explicit format parameter index
11523 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11524 0 flag (as above): repeated to allow "v02"
11525 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11526 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11528 [%bcdefginopsuxDFOUX] format (mandatory)
11533 As of perl5.9.3, printf format checking is on by default.
11534 Internally, perl uses %p formats to provide an escape to
11535 some extended formatting. This block deals with those
11536 extensions: if it does not match, (char*)q is reset and
11537 the normal format processing code is used.
11539 Currently defined extensions are:
11540 %p include pointer address (standard)
11541 %-p (SVf) include an SV (previously %_)
11542 %-<num>p include an SV with precision <num>
11544 %3p include a HEK with precision of 256
11545 %4p char* preceded by utf8 flag and length
11546 %<num>p (where num is 1 or > 4) reserved for future
11549 Robin Barker 2005-07-14 (but modified since)
11551 %1p (VDf) removed. RMB 2007-10-19
11558 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11559 /* The argument has already gone through cBOOL, so the cast
11561 is_utf8 = (bool)va_arg(*args, int);
11562 elen = va_arg(*args, UV);
11563 /* if utf8 length is larger than 0x7ffff..., then it might
11564 * have been a signed value that wrapped */
11565 if (elen > ((~(STRLEN)0) >> 1)) {
11566 assert(0); /* in DEBUGGING build we want to crash */
11567 elen= 0; /* otherwise we want to treat this as an empty string */
11569 eptr = va_arg(*args, char *);
11570 q += sizeof(UTF8f)-1;
11573 n = expect_number(&q);
11575 if (sv) { /* SVf */
11580 argsv = MUTABLE_SV(va_arg(*args, void*));
11581 eptr = SvPV_const(argsv, elen);
11582 if (DO_UTF8(argsv))
11586 else if (n==2 || n==3) { /* HEKf */
11587 HEK * const hek = va_arg(*args, HEK *);
11588 eptr = HEK_KEY(hek);
11589 elen = HEK_LEN(hek);
11590 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11591 if (n==3) precis = 256, has_precis = TRUE;
11595 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11596 "internal %%<num>p might conflict with future printf extensions");
11602 if ( (width = expect_number(&q)) ) {
11605 Perl_croak_nocontext(
11606 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11609 used_explicit_ix = TRUE;
11621 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11650 if ( (ewix = expect_number(&q)) ) {
11653 Perl_croak_nocontext(
11654 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11655 used_explicit_ix = TRUE;
11665 if ((vectorarg = asterisk)) {
11678 width = expect_number(&q);
11681 if (vectorize && vectorarg) {
11682 /* vectorizing, but not with the default "." */
11684 vecsv = va_arg(*args, SV*);
11686 FETCH_VCATPVFN_ARGUMENT(
11687 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11689 FETCH_VCATPVFN_ARGUMENT(
11690 vecsv, svix < svmax, svargs[svix++]);
11692 dotstr = SvPV_const(vecsv, dotstrlen);
11693 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11694 bad with tied or overloaded values that return UTF8. */
11695 if (DO_UTF8(vecsv))
11697 else if (has_utf8) {
11698 vecsv = sv_mortalcopy(vecsv);
11699 sv_utf8_upgrade(vecsv);
11700 dotstr = SvPV_const(vecsv, dotstrlen);
11707 i = va_arg(*args, int);
11709 i = (ewix ? ewix <= svmax : svix < svmax) ?
11710 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11712 width = (i < 0) ? -i : i;
11722 if ( (epix = expect_number(&q)) ) {
11725 Perl_croak_nocontext(
11726 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11727 used_explicit_ix = TRUE;
11732 i = va_arg(*args, int);
11736 FETCH_VCATPVFN_ARGUMENT(
11737 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11739 FETCH_VCATPVFN_ARGUMENT(
11740 precsv, svix < svmax, svargs[svix++]);
11741 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11744 has_precis = !(i < 0);
11748 while (isDIGIT(*q))
11749 precis = precis * 10 + (*q++ - '0');
11758 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11759 vecsv = svargs[efix ? efix-1 : svix++];
11760 vecstr = (U8*)SvPV_const(vecsv,veclen);
11761 vec_utf8 = DO_UTF8(vecsv);
11763 /* if this is a version object, we need to convert
11764 * back into v-string notation and then let the
11765 * vectorize happen normally
11767 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11768 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11769 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11770 "vector argument not supported with alpha versions");
11773 vecsv = sv_newmortal();
11774 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11776 vecstr = (U8*)SvPV_const(vecsv, veclen);
11777 vec_utf8 = DO_UTF8(vecsv);
11791 case 'I': /* Ix, I32x, and I64x */
11792 # ifdef USE_64_BIT_INT
11793 if (q[1] == '6' && q[2] == '4') {
11799 if (q[1] == '3' && q[2] == '2') {
11803 # ifdef USE_64_BIT_INT
11809 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11810 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11813 # ifdef USE_QUADMATH
11826 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11827 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11828 if (*q == 'l') { /* lld, llf */
11837 if (*++q == 'h') { /* hhd, hhu */
11866 if (!vectorize && !args) {
11868 const I32 i = efix-1;
11869 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11871 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11876 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11877 /* XXX va_arg(*args) case? need peek, use va_copy? */
11879 if (UNLIKELY(SvAMAGIC(argsv)))
11880 argsv = sv_2num(argsv);
11881 infnan = UNLIKELY(isinfnansv(argsv));
11884 switch (c = *q++) {
11892 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11893 /* no va_arg() case */
11894 SvNV_nomg(argsv), (int)c);
11895 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11897 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11899 eptr = (char*)utf8buf;
11900 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11914 eptr = va_arg(*args, char*);
11916 elen = strlen(eptr);
11918 eptr = (char *)nullstr;
11919 elen = sizeof nullstr - 1;
11923 eptr = SvPV_const(argsv, elen);
11924 if (DO_UTF8(argsv)) {
11925 STRLEN old_precis = precis;
11926 if (has_precis && precis < elen) {
11927 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11928 STRLEN p = precis > ulen ? ulen : precis;
11929 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11930 /* sticks at end */
11932 if (width) { /* fudge width (can't fudge elen) */
11933 if (has_precis && precis < elen)
11934 width += precis - old_precis;
11937 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11944 if (has_precis && precis < elen)
11952 goto floating_point;
11954 if (alt || vectorize)
11956 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11970 goto floating_point;
11975 goto donevalidconversion;
11977 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11986 esignbuf[esignlen++] = plus;
11990 case 'c': iv = (char)va_arg(*args, int); break;
11991 case 'h': iv = (short)va_arg(*args, int); break;
11992 case 'l': iv = va_arg(*args, long); break;
11993 case 'V': iv = va_arg(*args, IV); break;
11994 case 'z': iv = va_arg(*args, SSize_t); break;
11995 #ifdef HAS_PTRDIFF_T
11996 case 't': iv = va_arg(*args, ptrdiff_t); break;
11998 default: iv = va_arg(*args, int); break;
12000 case 'j': iv = va_arg(*args, intmax_t); break;
12004 iv = va_arg(*args, Quad_t); break;
12011 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
12013 case 'c': iv = (char)tiv; break;
12014 case 'h': iv = (short)tiv; break;
12015 case 'l': iv = (long)tiv; break;
12017 default: iv = tiv; break;
12020 iv = (Quad_t)tiv; break;
12026 if ( !vectorize ) /* we already set uv above */
12031 esignbuf[esignlen++] = plus;
12034 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12035 esignbuf[esignlen++] = '-';
12074 goto floating_point;
12080 goto donevalidconversion;
12082 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12093 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12094 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12095 case 'l': uv = va_arg(*args, unsigned long); break;
12096 case 'V': uv = va_arg(*args, UV); break;
12097 case 'z': uv = va_arg(*args, Size_t); break;
12098 #ifdef HAS_PTRDIFF_T
12099 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12102 case 'j': uv = va_arg(*args, uintmax_t); break;
12104 default: uv = va_arg(*args, unsigned); break;
12107 uv = va_arg(*args, Uquad_t); break;
12114 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12116 case 'c': uv = (unsigned char)tuv; break;
12117 case 'h': uv = (unsigned short)tuv; break;
12118 case 'l': uv = (unsigned long)tuv; break;
12120 default: uv = tuv; break;
12123 uv = (Uquad_t)tuv; break;
12132 char *ptr = ebuf + sizeof ebuf;
12133 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12139 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12143 } while (uv >>= 4);
12145 esignbuf[esignlen++] = '0';
12146 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12152 *--ptr = '0' + dig;
12153 } while (uv >>= 3);
12154 if (alt && *ptr != '0')
12160 *--ptr = '0' + dig;
12161 } while (uv >>= 1);
12163 esignbuf[esignlen++] = '0';
12164 esignbuf[esignlen++] = c;
12167 default: /* it had better be ten or less */
12170 *--ptr = '0' + dig;
12171 } while (uv /= base);
12174 elen = (ebuf + sizeof ebuf) - ptr;
12178 zeros = precis - elen;
12179 else if (precis == 0 && elen == 1 && *eptr == '0'
12180 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12183 /* a precision nullifies the 0 flag. */
12190 /* FLOATING POINT */
12195 c = 'f'; /* maybe %F isn't supported here */
12197 case 'e': case 'E':
12199 case 'g': case 'G':
12200 case 'a': case 'A':
12204 /* This is evil, but floating point is even more evil */
12206 /* for SV-style calling, we can only get NV
12207 for C-style calling, we assume %f is double;
12208 for simplicity we allow any of %Lf, %llf, %qf for long double
12212 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12216 /* [perl #20339] - we should accept and ignore %lf rather than die */
12220 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12221 intsize = args ? 0 : 'q';
12225 #if defined(HAS_LONG_DOUBLE)
12238 /* Now we need (long double) if intsize == 'q', else (double). */
12240 /* Note: do not pull NVs off the va_list with va_arg()
12241 * (pull doubles instead) because if you have a build
12242 * with long doubles, you would always be pulling long
12243 * doubles, which would badly break anyone using only
12244 * doubles (i.e. the majority of builds). In other
12245 * words, you cannot mix doubles and long doubles.
12246 * The only case where you can pull off long doubles
12247 * is when the format specifier explicitly asks so with
12249 #ifdef USE_QUADMATH
12250 fv = intsize == 'q' ?
12251 va_arg(*args, NV) : va_arg(*args, double);
12253 #elif LONG_DOUBLESIZE > DOUBLESIZE
12254 if (intsize == 'q') {
12255 fv = va_arg(*args, long double);
12258 nv = va_arg(*args, double);
12262 nv = va_arg(*args, double);
12268 if (!infnan) SvGETMAGIC(argsv);
12269 nv = SvNV_nomg(argsv);
12274 /* frexp() (or frexpl) has some unspecified behaviour for
12275 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12276 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12278 (void)Perl_frexp((NV)fv, &i);
12279 if (i == PERL_INT_MIN)
12280 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12281 /* Do not set hexfp earlier since we want to printf
12282 * Inf/NaN for Inf/NaN, not their hexfp. */
12283 hexfp = isALPHA_FOLD_EQ(c, 'a');
12284 if (UNLIKELY(hexfp)) {
12285 /* This seriously overshoots in most cases, but
12286 * better the undershooting. Firstly, all bytes
12287 * of the NV are not mantissa, some of them are
12288 * exponent. Secondly, for the reasonably common
12289 * long doubles case, the "80-bit extended", two
12290 * or six bytes of the NV are unused. */
12292 (fv < 0) ? 1 : 0 + /* possible unary minus */
12294 1 + /* the very unlikely carry */
12297 2 * NVSIZE + /* 2 hexdigits for each byte */
12299 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12301 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12302 /* However, for the "double double", we need more.
12303 * Since each double has their own exponent, the
12304 * doubles may float (haha) rather far from each
12305 * other, and the number of required bits is much
12306 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12307 * See the definition of DOUBLEDOUBLE_MAXBITS.
12309 * Need 2 hexdigits for each byte. */
12310 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12311 /* the size for the exponent already added */
12313 #ifdef USE_LOCALE_NUMERIC
12314 STORE_LC_NUMERIC_SET_TO_NEEDED();
12315 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12316 need += SvLEN(PL_numeric_radix_sv);
12317 RESTORE_LC_NUMERIC();
12321 need = BIT_DIGITS(i);
12322 } /* if i < 0, the number of digits is hard to predict. */
12324 need += has_precis ? precis : 6; /* known default */
12329 #ifdef HAS_LDBL_SPRINTF_BUG
12330 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12331 with sfio - Allen <allens@cpan.org> */
12334 # define MY_DBL_MAX DBL_MAX
12335 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12336 # if DOUBLESIZE >= 8
12337 # define MY_DBL_MAX 1.7976931348623157E+308L
12339 # define MY_DBL_MAX 3.40282347E+38L
12343 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12344 # define MY_DBL_MAX_BUG 1L
12346 # define MY_DBL_MAX_BUG MY_DBL_MAX
12350 # define MY_DBL_MIN DBL_MIN
12351 # else /* XXX guessing! -Allen */
12352 # if DOUBLESIZE >= 8
12353 # define MY_DBL_MIN 2.2250738585072014E-308L
12355 # define MY_DBL_MIN 1.17549435E-38L
12359 if ((intsize == 'q') && (c == 'f') &&
12360 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12361 (need < DBL_DIG)) {
12362 /* it's going to be short enough that
12363 * long double precision is not needed */
12365 if ((fv <= 0L) && (fv >= -0L))
12366 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12368 /* would use Perl_fp_class as a double-check but not
12369 * functional on IRIX - see perl.h comments */
12371 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12372 /* It's within the range that a double can represent */
12373 #if defined(DBL_MAX) && !defined(DBL_MIN)
12374 if ((fv >= ((long double)1/DBL_MAX)) ||
12375 (fv <= (-(long double)1/DBL_MAX)))
12377 fix_ldbl_sprintf_bug = TRUE;
12380 if (fix_ldbl_sprintf_bug == TRUE) {
12390 # undef MY_DBL_MAX_BUG
12393 #endif /* HAS_LDBL_SPRINTF_BUG */
12395 need += 20; /* fudge factor */
12396 if (PL_efloatsize < need) {
12397 Safefree(PL_efloatbuf);
12398 PL_efloatsize = need + 20; /* more fudge */
12399 Newx(PL_efloatbuf, PL_efloatsize, char);
12400 PL_efloatbuf[0] = '\0';
12403 if ( !(width || left || plus || alt) && fill != '0'
12404 && has_precis && intsize != 'q' /* Shortcuts */
12405 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12406 /* See earlier comment about buggy Gconvert when digits,
12408 if ( c == 'g' && precis ) {
12409 STORE_LC_NUMERIC_SET_TO_NEEDED();
12410 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12411 /* May return an empty string for digits==0 */
12412 if (*PL_efloatbuf) {
12413 elen = strlen(PL_efloatbuf);
12414 goto float_converted;
12416 } else if ( c == 'f' && !precis ) {
12417 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12422 if (UNLIKELY(hexfp)) {
12423 /* Hexadecimal floating point. */
12424 char* p = PL_efloatbuf;
12425 U8 vhex[VHEX_SIZE];
12426 U8* v = vhex; /* working pointer to vhex */
12427 U8* vend; /* pointer to one beyond last digit of vhex */
12428 U8* vfnz = NULL; /* first non-zero */
12429 U8* vlnz = NULL; /* last non-zero */
12430 U8* v0 = NULL; /* first output */
12431 const bool lower = (c == 'a');
12432 /* At output the values of vhex (up to vend) will
12433 * be mapped through the xdig to get the actual
12434 * human-readable xdigits. */
12435 const char* xdig = PL_hexdigit;
12436 int zerotail = 0; /* how many extra zeros to append */
12437 int exponent = 0; /* exponent of the floating point input */
12438 bool hexradix = FALSE; /* should we output the radix */
12439 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12440 bool negative = FALSE;
12442 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12444 * For example with denormals, (assuming the vanilla
12445 * 64-bit double): the exponent is zero. 1xp-1074 is
12446 * the smallest denormal and the smallest double, it
12447 * could be output also as 0x0.0000000000001p-1022 to
12448 * match its internal structure. */
12450 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12451 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12453 #if NVSIZE > DOUBLESIZE
12454 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12455 /* In this case there is an implicit bit,
12456 * and therefore the exponent is shifted by one. */
12459 # ifdef NV_X86_80_BIT
12461 /* The subnormals of the x86-80 have a base exponent of -16382,
12462 * (while the physical exponent bits are zero) but the frexp()
12463 * returned the scientific-style floating exponent. We want
12464 * to map the last one as:
12465 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12466 * -16835..-16388 -> -16384
12467 * since we want to keep the first hexdigit
12468 * as one of the [8421]. */
12469 exponent = -4 * ( (exponent + 1) / -4) - 2;
12474 /* TBD: other non-implicit-bit platforms than the x86-80. */
12478 negative = fv < 0 || Perl_signbit(nv);
12489 xdig += 16; /* Use uppercase hex. */
12492 /* Find the first non-zero xdigit. */
12493 for (v = vhex; v < vend; v++) {
12501 /* Find the last non-zero xdigit. */
12502 for (v = vend - 1; v >= vhex; v--) {
12509 #if NVSIZE == DOUBLESIZE
12515 #ifndef NV_X86_80_BIT
12517 /* IEEE 754 subnormals (but not the x86 80-bit):
12518 * we want "normalize" the subnormal,
12519 * so we need to right shift the hex nybbles
12520 * so that the output of the subnormal starts
12521 * from the first true bit. (Another, equally
12522 * valid, policy would be to dump the subnormal
12523 * nybbles as-is, to display the "physical" layout.) */
12526 /* Find the ceil(log2(v[0])) of
12527 * the top non-zero nybble. */
12528 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12531 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12532 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12546 U8* ve = (subnormal ? vlnz + 1 : vend);
12547 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12548 if ((SSize_t)(precis + 1) < vn) {
12549 bool overflow = FALSE;
12550 if (v0[precis + 1] < 0x8) {
12551 /* Round down, nothing to do. */
12552 } else if (v0[precis + 1] > 0x8) {
12555 overflow = v0[precis] > 0xF;
12557 } else { /* v0[precis] == 0x8 */
12558 /* Half-point: round towards the one
12559 * with the even least-significant digit:
12567 * 78 -> 8 f8 -> 10 */
12568 if ((v0[precis] & 0x1)) {
12571 overflow = v0[precis] > 0xF;
12576 for (v = v0 + precis - 1; v >= v0; v--) {
12578 overflow = *v > 0xF;
12584 if (v == v0 - 1 && overflow) {
12585 /* If the overflow goes all the
12586 * way to the front, we need to
12587 * insert 0x1 in front, and adjust
12589 Move(v0, v0 + 1, vn, char);
12595 /* The new effective "last non zero". */
12596 vlnz = v0 + precis;
12600 subnormal ? precis - vn + 1 :
12601 precis - (vlnz - vhex);
12608 /* If there are non-zero xdigits, the radix
12609 * is output after the first one. */
12620 /* The radix is always output if precis, or if alt. */
12621 if (precis > 0 || alt) {
12626 #ifndef USE_LOCALE_NUMERIC
12629 STORE_LC_NUMERIC_SET_TO_NEEDED();
12630 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12632 const char* r = SvPV(PL_numeric_radix_sv, n);
12633 Copy(r, p, n, char);
12639 RESTORE_LC_NUMERIC();
12648 if (zerotail > 0) {
12649 while (zerotail--) {
12654 elen = p - PL_efloatbuf;
12655 elen += my_snprintf(p, PL_efloatsize - elen,
12656 "%c%+d", lower ? 'p' : 'P',
12659 if (elen < width) {
12661 /* Pad the back with spaces. */
12662 memset(PL_efloatbuf + elen, ' ', width - elen);
12664 else if (fill == '0') {
12665 /* Insert the zeros after the "0x" and the
12666 * the potential sign, but before the digits,
12667 * otherwise we end up with "0000xH.HHH...",
12668 * when we want "0x000H.HHH..." */
12669 STRLEN nzero = width - elen;
12670 char* zerox = PL_efloatbuf + 2;
12671 STRLEN nmove = elen - 2;
12672 if (negative || plus) {
12676 Move(zerox, zerox + nzero, nmove, char);
12677 memset(zerox, fill, nzero);
12680 /* Move it to the right. */
12681 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12683 /* Pad the front with spaces. */
12684 memset(PL_efloatbuf, ' ', width - elen);
12690 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12692 /* Not affecting infnan output: precision, alt, fill. */
12693 if (elen < width) {
12695 /* Pack the back with spaces. */
12696 memset(PL_efloatbuf + elen, ' ', width - elen);
12698 /* Move it to the right. */
12699 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12701 /* Pad the front with spaces. */
12702 memset(PL_efloatbuf, ' ', width - elen);
12710 char *ptr = ebuf + sizeof ebuf;
12713 #if defined(USE_QUADMATH)
12714 if (intsize == 'q') {
12718 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12719 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12720 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12721 * not USE_LONG_DOUBLE and NVff. In other words,
12722 * this needs to work without USE_LONG_DOUBLE. */
12723 if (intsize == 'q') {
12724 /* Copy the one or more characters in a long double
12725 * format before the 'base' ([efgEFG]) character to
12726 * the format string. */
12727 static char const ldblf[] = PERL_PRIfldbl;
12728 char const *p = ldblf + sizeof(ldblf) - 3;
12729 while (p >= ldblf) { *--ptr = *p--; }
12734 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12739 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12751 /* No taint. Otherwise we are in the strange situation
12752 * where printf() taints but print($float) doesn't.
12755 STORE_LC_NUMERIC_SET_TO_NEEDED();
12757 /* hopefully the above makes ptr a very constrained format
12758 * that is safe to use, even though it's not literal */
12759 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12760 #ifdef USE_QUADMATH
12762 const char* qfmt = quadmath_format_single(ptr);
12764 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12765 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12767 if ((IV)elen == -1)
12768 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12772 #elif defined(HAS_LONG_DOUBLE)
12773 elen = ((intsize == 'q')
12774 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12775 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12777 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12783 eptr = PL_efloatbuf;
12784 assert((IV)elen > 0); /* here zero elen is bad */
12786 #ifdef USE_LOCALE_NUMERIC
12787 /* If the decimal point character in the string is UTF-8, make the
12789 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12790 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12803 i = SvCUR(sv) - origlen;
12806 case 'c': *(va_arg(*args, char*)) = i; break;
12807 case 'h': *(va_arg(*args, short*)) = i; break;
12808 default: *(va_arg(*args, int*)) = i; break;
12809 case 'l': *(va_arg(*args, long*)) = i; break;
12810 case 'V': *(va_arg(*args, IV*)) = i; break;
12811 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12812 #ifdef HAS_PTRDIFF_T
12813 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12816 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12820 *(va_arg(*args, Quad_t*)) = i; break;
12827 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12828 goto donevalidconversion;
12835 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12836 && ckWARN(WARN_PRINTF))
12838 SV * const msg = sv_newmortal();
12839 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12840 (PL_op->op_type == OP_PRTF) ? "" : "s");
12841 if (fmtstart < patend) {
12842 const char * const fmtend = q < patend ? q : patend;
12844 sv_catpvs(msg, "\"%");
12845 for (f = fmtstart; f < fmtend; f++) {
12847 sv_catpvn_nomg(msg, f, 1);
12849 Perl_sv_catpvf(aTHX_ msg,
12850 "\\%03"UVof, (UV)*f & 0xFF);
12853 sv_catpvs(msg, "\"");
12855 sv_catpvs(msg, "end of string");
12857 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12860 /* output mangled stuff ... */
12866 /* ... right here, because formatting flags should not apply */
12867 SvGROW(sv, SvCUR(sv) + elen + 1);
12869 Copy(eptr, p, elen, char);
12872 SvCUR_set(sv, p - SvPVX_const(sv));
12874 continue; /* not "break" */
12877 if (is_utf8 != has_utf8) {
12880 sv_utf8_upgrade(sv);
12883 const STRLEN old_elen = elen;
12884 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12885 sv_utf8_upgrade(nsv);
12886 eptr = SvPVX_const(nsv);
12889 if (width) { /* fudge width (can't fudge elen) */
12890 width += elen - old_elen;
12896 /* signed value that's wrapped? */
12897 assert(elen <= ((~(STRLEN)0) >> 1));
12898 have = esignlen + zeros + elen;
12900 croak_memory_wrap();
12902 need = (have > width ? have : width);
12905 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12906 croak_memory_wrap();
12907 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12909 if (esignlen && fill == '0') {
12911 for (i = 0; i < (int)esignlen; i++)
12912 *p++ = esignbuf[i];
12914 if (gap && !left) {
12915 memset(p, fill, gap);
12918 if (esignlen && fill != '0') {
12920 for (i = 0; i < (int)esignlen; i++)
12921 *p++ = esignbuf[i];
12925 for (i = zeros; i; i--)
12929 Copy(eptr, p, elen, char);
12933 memset(p, ' ', gap);
12938 Copy(dotstr, p, dotstrlen, char);
12942 vectorize = FALSE; /* done iterating over vecstr */
12949 SvCUR_set(sv, p - SvPVX_const(sv));
12955 donevalidconversion:
12956 if (used_explicit_ix)
12957 no_redundant_warning = TRUE;
12959 S_warn_vcatpvfn_missing_argument(aTHX);
12962 /* Now that we've consumed all our printf format arguments (svix)
12963 * do we have things left on the stack that we didn't use?
12965 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12966 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12967 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12972 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12976 /* =========================================================================
12978 =head1 Cloning an interpreter
12982 All the macros and functions in this section are for the private use of
12983 the main function, perl_clone().
12985 The foo_dup() functions make an exact copy of an existing foo thingy.
12986 During the course of a cloning, a hash table is used to map old addresses
12987 to new addresses. The table is created and manipulated with the
12988 ptr_table_* functions.
12990 * =========================================================================*/
12993 #if defined(USE_ITHREADS)
12995 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12996 #ifndef GpREFCNT_inc
12997 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13001 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13002 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13003 If this changes, please unmerge ss_dup.
13004 Likewise, sv_dup_inc_multiple() relies on this fact. */
13005 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13006 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13007 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13008 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13009 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13010 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13011 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13012 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13013 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13014 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13015 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13016 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13017 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13019 /* clone a parser */
13022 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13026 PERL_ARGS_ASSERT_PARSER_DUP;
13031 /* look for it in the table first */
13032 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13036 /* create anew and remember what it is */
13037 Newxz(parser, 1, yy_parser);
13038 ptr_table_store(PL_ptr_table, proto, parser);
13040 /* XXX these not yet duped */
13041 parser->old_parser = NULL;
13042 parser->stack = NULL;
13044 parser->stack_size = 0;
13045 /* XXX parser->stack->state = 0; */
13047 /* XXX eventually, just Copy() most of the parser struct ? */
13049 parser->lex_brackets = proto->lex_brackets;
13050 parser->lex_casemods = proto->lex_casemods;
13051 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13052 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13053 parser->lex_casestack = savepvn(proto->lex_casestack,
13054 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13055 parser->lex_defer = proto->lex_defer;
13056 parser->lex_dojoin = proto->lex_dojoin;
13057 parser->lex_formbrack = proto->lex_formbrack;
13058 parser->lex_inpat = proto->lex_inpat;
13059 parser->lex_inwhat = proto->lex_inwhat;
13060 parser->lex_op = proto->lex_op;
13061 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13062 parser->lex_starts = proto->lex_starts;
13063 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13064 parser->multi_close = proto->multi_close;
13065 parser->multi_open = proto->multi_open;
13066 parser->multi_start = proto->multi_start;
13067 parser->multi_end = proto->multi_end;
13068 parser->preambled = proto->preambled;
13069 parser->lex_super_state = proto->lex_super_state;
13070 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13071 parser->lex_sub_op = proto->lex_sub_op;
13072 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13073 parser->linestr = sv_dup_inc(proto->linestr, param);
13074 parser->expect = proto->expect;
13075 parser->copline = proto->copline;
13076 parser->last_lop_op = proto->last_lop_op;
13077 parser->lex_state = proto->lex_state;
13078 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13079 /* rsfp_filters entries have fake IoDIRP() */
13080 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13081 parser->in_my = proto->in_my;
13082 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13083 parser->error_count = proto->error_count;
13084 parser->sig_elems = proto->sig_elems;
13085 parser->sig_optelems= proto->sig_optelems;
13086 parser->sig_slurpy = proto->sig_slurpy;
13087 parser->linestr = sv_dup_inc(proto->linestr, param);
13090 char * const ols = SvPVX(proto->linestr);
13091 char * const ls = SvPVX(parser->linestr);
13093 parser->bufptr = ls + (proto->bufptr >= ols ?
13094 proto->bufptr - ols : 0);
13095 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13096 proto->oldbufptr - ols : 0);
13097 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13098 proto->oldoldbufptr - ols : 0);
13099 parser->linestart = ls + (proto->linestart >= ols ?
13100 proto->linestart - ols : 0);
13101 parser->last_uni = ls + (proto->last_uni >= ols ?
13102 proto->last_uni - ols : 0);
13103 parser->last_lop = ls + (proto->last_lop >= ols ?
13104 proto->last_lop - ols : 0);
13106 parser->bufend = ls + SvCUR(parser->linestr);
13109 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13112 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13113 Copy(proto->nexttype, parser->nexttype, 5, I32);
13114 parser->nexttoke = proto->nexttoke;
13116 /* XXX should clone saved_curcop here, but we aren't passed
13117 * proto_perl; so do it in perl_clone_using instead */
13123 /* duplicate a file handle */
13126 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13130 PERL_ARGS_ASSERT_FP_DUP;
13131 PERL_UNUSED_ARG(type);
13134 return (PerlIO*)NULL;
13136 /* look for it in the table first */
13137 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13141 /* create anew and remember what it is */
13142 #ifdef __amigaos4__
13143 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13145 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13147 ptr_table_store(PL_ptr_table, fp, ret);
13151 /* duplicate a directory handle */
13154 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13158 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13160 const Direntry_t *dirent;
13161 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13167 PERL_UNUSED_CONTEXT;
13168 PERL_ARGS_ASSERT_DIRP_DUP;
13173 /* look for it in the table first */
13174 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13178 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13180 PERL_UNUSED_ARG(param);
13184 /* open the current directory (so we can switch back) */
13185 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13187 /* chdir to our dir handle and open the present working directory */
13188 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13189 PerlDir_close(pwd);
13190 return (DIR *)NULL;
13192 /* Now we should have two dir handles pointing to the same dir. */
13194 /* Be nice to the calling code and chdir back to where we were. */
13195 /* XXX If this fails, then what? */
13196 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13198 /* We have no need of the pwd handle any more. */
13199 PerlDir_close(pwd);
13202 # define d_namlen(d) (d)->d_namlen
13204 # define d_namlen(d) strlen((d)->d_name)
13206 /* Iterate once through dp, to get the file name at the current posi-
13207 tion. Then step back. */
13208 pos = PerlDir_tell(dp);
13209 if ((dirent = PerlDir_read(dp))) {
13210 len = d_namlen(dirent);
13211 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13212 /* If the len is somehow magically longer than the
13213 * maximum length of the directory entry, even though
13214 * we could fit it in a buffer, we could not copy it
13215 * from the dirent. Bail out. */
13216 PerlDir_close(ret);
13219 if (len <= sizeof smallbuf) name = smallbuf;
13220 else Newx(name, len, char);
13221 Move(dirent->d_name, name, len, char);
13223 PerlDir_seek(dp, pos);
13225 /* Iterate through the new dir handle, till we find a file with the
13227 if (!dirent) /* just before the end */
13229 pos = PerlDir_tell(ret);
13230 if (PerlDir_read(ret)) continue; /* not there yet */
13231 PerlDir_seek(ret, pos); /* step back */
13235 const long pos0 = PerlDir_tell(ret);
13237 pos = PerlDir_tell(ret);
13238 if ((dirent = PerlDir_read(ret))) {
13239 if (len == (STRLEN)d_namlen(dirent)
13240 && memEQ(name, dirent->d_name, len)) {
13242 PerlDir_seek(ret, pos); /* step back */
13245 /* else we are not there yet; keep iterating */
13247 else { /* This is not meant to happen. The best we can do is
13248 reset the iterator to the beginning. */
13249 PerlDir_seek(ret, pos0);
13256 if (name && name != smallbuf)
13261 ret = win32_dirp_dup(dp, param);
13264 /* pop it in the pointer table */
13266 ptr_table_store(PL_ptr_table, dp, ret);
13271 /* duplicate a typeglob */
13274 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13278 PERL_ARGS_ASSERT_GP_DUP;
13282 /* look for it in the table first */
13283 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13287 /* create anew and remember what it is */
13289 ptr_table_store(PL_ptr_table, gp, ret);
13292 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13293 on Newxz() to do this for us. */
13294 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13295 ret->gp_io = io_dup_inc(gp->gp_io, param);
13296 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13297 ret->gp_av = av_dup_inc(gp->gp_av, param);
13298 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13299 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13300 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13301 ret->gp_cvgen = gp->gp_cvgen;
13302 ret->gp_line = gp->gp_line;
13303 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13307 /* duplicate a chain of magic */
13310 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13312 MAGIC *mgret = NULL;
13313 MAGIC **mgprev_p = &mgret;
13315 PERL_ARGS_ASSERT_MG_DUP;
13317 for (; mg; mg = mg->mg_moremagic) {
13320 if ((param->flags & CLONEf_JOIN_IN)
13321 && mg->mg_type == PERL_MAGIC_backref)
13322 /* when joining, we let the individual SVs add themselves to
13323 * backref as needed. */
13326 Newx(nmg, 1, MAGIC);
13328 mgprev_p = &(nmg->mg_moremagic);
13330 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13331 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13332 from the original commit adding Perl_mg_dup() - revision 4538.
13333 Similarly there is the annotation "XXX random ptr?" next to the
13334 assignment to nmg->mg_ptr. */
13337 /* FIXME for plugins
13338 if (nmg->mg_type == PERL_MAGIC_qr) {
13339 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13343 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13344 ? nmg->mg_type == PERL_MAGIC_backref
13345 /* The backref AV has its reference
13346 * count deliberately bumped by 1 */
13347 ? SvREFCNT_inc(av_dup_inc((const AV *)
13348 nmg->mg_obj, param))
13349 : sv_dup_inc(nmg->mg_obj, param)
13350 : sv_dup(nmg->mg_obj, param);
13352 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13353 if (nmg->mg_len > 0) {
13354 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13355 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13356 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13358 AMT * const namtp = (AMT*)nmg->mg_ptr;
13359 sv_dup_inc_multiple((SV**)(namtp->table),
13360 (SV**)(namtp->table), NofAMmeth, param);
13363 else if (nmg->mg_len == HEf_SVKEY)
13364 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13366 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13367 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13373 #endif /* USE_ITHREADS */
13375 struct ptr_tbl_arena {
13376 struct ptr_tbl_arena *next;
13377 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13380 /* create a new pointer-mapping table */
13383 Perl_ptr_table_new(pTHX)
13386 PERL_UNUSED_CONTEXT;
13388 Newx(tbl, 1, PTR_TBL_t);
13389 tbl->tbl_max = 511;
13390 tbl->tbl_items = 0;
13391 tbl->tbl_arena = NULL;
13392 tbl->tbl_arena_next = NULL;
13393 tbl->tbl_arena_end = NULL;
13394 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13398 #define PTR_TABLE_HASH(ptr) \
13399 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13401 /* map an existing pointer using a table */
13403 STATIC PTR_TBL_ENT_t *
13404 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13406 PTR_TBL_ENT_t *tblent;
13407 const UV hash = PTR_TABLE_HASH(sv);
13409 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13411 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13412 for (; tblent; tblent = tblent->next) {
13413 if (tblent->oldval == sv)
13420 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13422 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13424 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13425 PERL_UNUSED_CONTEXT;
13427 return tblent ? tblent->newval : NULL;
13430 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13431 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13432 * the core's typical use of ptr_tables in thread cloning. */
13435 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13437 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13439 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13440 PERL_UNUSED_CONTEXT;
13443 tblent->newval = newsv;
13445 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13447 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13448 struct ptr_tbl_arena *new_arena;
13450 Newx(new_arena, 1, struct ptr_tbl_arena);
13451 new_arena->next = tbl->tbl_arena;
13452 tbl->tbl_arena = new_arena;
13453 tbl->tbl_arena_next = new_arena->array;
13454 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13457 tblent = tbl->tbl_arena_next++;
13459 tblent->oldval = oldsv;
13460 tblent->newval = newsv;
13461 tblent->next = tbl->tbl_ary[entry];
13462 tbl->tbl_ary[entry] = tblent;
13464 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13465 ptr_table_split(tbl);
13469 /* double the hash bucket size of an existing ptr table */
13472 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13474 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13475 const UV oldsize = tbl->tbl_max + 1;
13476 UV newsize = oldsize * 2;
13479 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13480 PERL_UNUSED_CONTEXT;
13482 Renew(ary, newsize, PTR_TBL_ENT_t*);
13483 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13484 tbl->tbl_max = --newsize;
13485 tbl->tbl_ary = ary;
13486 for (i=0; i < oldsize; i++, ary++) {
13487 PTR_TBL_ENT_t **entp = ary;
13488 PTR_TBL_ENT_t *ent = *ary;
13489 PTR_TBL_ENT_t **curentp;
13492 curentp = ary + oldsize;
13494 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13496 ent->next = *curentp;
13506 /* remove all the entries from a ptr table */
13507 /* Deprecated - will be removed post 5.14 */
13510 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13512 PERL_UNUSED_CONTEXT;
13513 if (tbl && tbl->tbl_items) {
13514 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13516 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13519 struct ptr_tbl_arena *next = arena->next;
13525 tbl->tbl_items = 0;
13526 tbl->tbl_arena = NULL;
13527 tbl->tbl_arena_next = NULL;
13528 tbl->tbl_arena_end = NULL;
13532 /* clear and free a ptr table */
13535 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13537 struct ptr_tbl_arena *arena;
13539 PERL_UNUSED_CONTEXT;
13545 arena = tbl->tbl_arena;
13548 struct ptr_tbl_arena *next = arena->next;
13554 Safefree(tbl->tbl_ary);
13558 #if defined(USE_ITHREADS)
13561 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13563 PERL_ARGS_ASSERT_RVPV_DUP;
13565 assert(!isREGEXP(sstr));
13567 if (SvWEAKREF(sstr)) {
13568 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13569 if (param->flags & CLONEf_JOIN_IN) {
13570 /* if joining, we add any back references individually rather
13571 * than copying the whole backref array */
13572 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13576 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13578 else if (SvPVX_const(sstr)) {
13579 /* Has something there */
13581 /* Normal PV - clone whole allocated space */
13582 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13583 /* sstr may not be that normal, but actually copy on write.
13584 But we are a true, independent SV, so: */
13588 /* Special case - not normally malloced for some reason */
13589 if (isGV_with_GP(sstr)) {
13590 /* Don't need to do anything here. */
13592 else if ((SvIsCOW(sstr))) {
13593 /* A "shared" PV - clone it as "shared" PV */
13595 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13599 /* Some other special case - random pointer */
13600 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13605 /* Copy the NULL */
13606 SvPV_set(dstr, NULL);
13610 /* duplicate a list of SVs. source and dest may point to the same memory. */
13612 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13613 SSize_t items, CLONE_PARAMS *const param)
13615 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13617 while (items-- > 0) {
13618 *dest++ = sv_dup_inc(*source++, param);
13624 /* duplicate an SV of any type (including AV, HV etc) */
13627 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13632 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13634 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13635 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13640 /* look for it in the table first */
13641 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13645 if(param->flags & CLONEf_JOIN_IN) {
13646 /** We are joining here so we don't want do clone
13647 something that is bad **/
13648 if (SvTYPE(sstr) == SVt_PVHV) {
13649 const HEK * const hvname = HvNAME_HEK(sstr);
13651 /** don't clone stashes if they already exist **/
13652 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13653 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13654 ptr_table_store(PL_ptr_table, sstr, dstr);
13658 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13659 HV *stash = GvSTASH(sstr);
13660 const HEK * hvname;
13661 if (stash && (hvname = HvNAME_HEK(stash))) {
13662 /** don't clone GVs if they already exist **/
13664 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13665 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13667 stash, GvNAME(sstr),
13673 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13674 ptr_table_store(PL_ptr_table, sstr, *svp);
13681 /* create anew and remember what it is */
13684 #ifdef DEBUG_LEAKING_SCALARS
13685 dstr->sv_debug_optype = sstr->sv_debug_optype;
13686 dstr->sv_debug_line = sstr->sv_debug_line;
13687 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13688 dstr->sv_debug_parent = (SV*)sstr;
13689 FREE_SV_DEBUG_FILE(dstr);
13690 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13693 ptr_table_store(PL_ptr_table, sstr, dstr);
13696 SvFLAGS(dstr) = SvFLAGS(sstr);
13697 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13698 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13701 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13702 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13703 (void*)PL_watch_pvx, SvPVX_const(sstr));
13706 /* don't clone objects whose class has asked us not to */
13708 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13714 switch (SvTYPE(sstr)) {
13716 SvANY(dstr) = NULL;
13719 SET_SVANY_FOR_BODYLESS_IV(dstr);
13721 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13723 SvIV_set(dstr, SvIVX(sstr));
13727 #if NVSIZE <= IVSIZE
13728 SET_SVANY_FOR_BODYLESS_NV(dstr);
13730 SvANY(dstr) = new_XNV();
13732 SvNV_set(dstr, SvNVX(sstr));
13736 /* These are all the types that need complex bodies allocating. */
13738 const svtype sv_type = SvTYPE(sstr);
13739 const struct body_details *const sv_type_details
13740 = bodies_by_type + sv_type;
13744 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13760 assert(sv_type_details->body_size);
13761 if (sv_type_details->arena) {
13762 new_body_inline(new_body, sv_type);
13764 = (void*)((char*)new_body - sv_type_details->offset);
13766 new_body = new_NOARENA(sv_type_details);
13770 SvANY(dstr) = new_body;
13773 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13774 ((char*)SvANY(dstr)) + sv_type_details->offset,
13775 sv_type_details->copy, char);
13777 Copy(((char*)SvANY(sstr)),
13778 ((char*)SvANY(dstr)),
13779 sv_type_details->body_size + sv_type_details->offset, char);
13782 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13783 && !isGV_with_GP(dstr)
13785 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13786 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13788 /* The Copy above means that all the source (unduplicated) pointers
13789 are now in the destination. We can check the flags and the
13790 pointers in either, but it's possible that there's less cache
13791 missing by always going for the destination.
13792 FIXME - instrument and check that assumption */
13793 if (sv_type >= SVt_PVMG) {
13795 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13796 if (SvOBJECT(dstr) && SvSTASH(dstr))
13797 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13798 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13801 /* The cast silences a GCC warning about unhandled types. */
13802 switch ((int)sv_type) {
13813 /* FIXME for plugins */
13814 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13815 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13818 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13819 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13820 LvTARG(dstr) = dstr;
13821 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13822 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13824 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13825 if (isREGEXP(sstr)) goto duprex;
13827 /* non-GP case already handled above */
13828 if(isGV_with_GP(sstr)) {
13829 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13830 /* Don't call sv_add_backref here as it's going to be
13831 created as part of the magic cloning of the symbol
13832 table--unless this is during a join and the stash
13833 is not actually being cloned. */
13834 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13835 at the point of this comment. */
13836 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13837 if (param->flags & CLONEf_JOIN_IN)
13838 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13839 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13840 (void)GpREFCNT_inc(GvGP(dstr));
13844 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13845 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13846 /* I have no idea why fake dirp (rsfps)
13847 should be treated differently but otherwise
13848 we end up with leaks -- sky*/
13849 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13850 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13851 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13853 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13854 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13855 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13856 if (IoDIRP(dstr)) {
13857 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13860 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13862 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13864 if (IoOFP(dstr) == IoIFP(sstr))
13865 IoOFP(dstr) = IoIFP(dstr);
13867 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13868 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13869 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13870 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13873 /* avoid cloning an empty array */
13874 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13875 SV **dst_ary, **src_ary;
13876 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13878 src_ary = AvARRAY((const AV *)sstr);
13879 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13880 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13881 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13882 AvALLOC((const AV *)dstr) = dst_ary;
13883 if (AvREAL((const AV *)sstr)) {
13884 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13888 while (items-- > 0)
13889 *dst_ary++ = sv_dup(*src_ary++, param);
13891 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13892 while (items-- > 0) {
13897 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13898 AvALLOC((const AV *)dstr) = (SV**)NULL;
13899 AvMAX( (const AV *)dstr) = -1;
13900 AvFILLp((const AV *)dstr) = -1;
13904 if (HvARRAY((const HV *)sstr)) {
13906 const bool sharekeys = !!HvSHAREKEYS(sstr);
13907 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13908 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13910 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13911 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13913 HvARRAY(dstr) = (HE**)darray;
13914 while (i <= sxhv->xhv_max) {
13915 const HE * const source = HvARRAY(sstr)[i];
13916 HvARRAY(dstr)[i] = source
13917 ? he_dup(source, sharekeys, param) : 0;
13921 const struct xpvhv_aux * const saux = HvAUX(sstr);
13922 struct xpvhv_aux * const daux = HvAUX(dstr);
13923 /* This flag isn't copied. */
13926 if (saux->xhv_name_count) {
13927 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13929 = saux->xhv_name_count < 0
13930 ? -saux->xhv_name_count
13931 : saux->xhv_name_count;
13932 HEK **shekp = sname + count;
13934 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13935 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13936 while (shekp-- > sname) {
13938 *dhekp = hek_dup(*shekp, param);
13942 daux->xhv_name_u.xhvnameu_name
13943 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13946 daux->xhv_name_count = saux->xhv_name_count;
13948 daux->xhv_aux_flags = saux->xhv_aux_flags;
13949 #ifdef PERL_HASH_RANDOMIZE_KEYS
13950 daux->xhv_rand = saux->xhv_rand;
13951 daux->xhv_last_rand = saux->xhv_last_rand;
13953 daux->xhv_riter = saux->xhv_riter;
13954 daux->xhv_eiter = saux->xhv_eiter
13955 ? he_dup(saux->xhv_eiter,
13956 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13957 /* backref array needs refcnt=2; see sv_add_backref */
13958 daux->xhv_backreferences =
13959 (param->flags & CLONEf_JOIN_IN)
13960 /* when joining, we let the individual GVs and
13961 * CVs add themselves to backref as
13962 * needed. This avoids pulling in stuff
13963 * that isn't required, and simplifies the
13964 * case where stashes aren't cloned back
13965 * if they already exist in the parent
13968 : saux->xhv_backreferences
13969 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13970 ? MUTABLE_AV(SvREFCNT_inc(
13971 sv_dup_inc((const SV *)
13972 saux->xhv_backreferences, param)))
13973 : MUTABLE_AV(sv_dup((const SV *)
13974 saux->xhv_backreferences, param))
13977 daux->xhv_mro_meta = saux->xhv_mro_meta
13978 ? mro_meta_dup(saux->xhv_mro_meta, param)
13981 /* Record stashes for possible cloning in Perl_clone(). */
13983 av_push(param->stashes, dstr);
13987 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13990 if (!(param->flags & CLONEf_COPY_STACKS)) {
13995 /* NOTE: not refcounted */
13996 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13997 hv_dup(CvSTASH(dstr), param);
13998 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13999 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14000 if (!CvISXSUB(dstr)) {
14002 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14004 CvSLABBED_off(dstr);
14005 } else if (CvCONST(dstr)) {
14006 CvXSUBANY(dstr).any_ptr =
14007 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14009 assert(!CvSLABBED(dstr));
14010 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14012 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14013 hek_dup(CvNAME_HEK((CV *)sstr), param);
14014 /* don't dup if copying back - CvGV isn't refcounted, so the
14015 * duped GV may never be freed. A bit of a hack! DAPM */
14017 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14019 ? gv_dup_inc(CvGV(sstr), param)
14020 : (param->flags & CLONEf_JOIN_IN)
14022 : gv_dup(CvGV(sstr), param);
14024 if (!CvISXSUB(sstr)) {
14025 PADLIST * padlist = CvPADLIST(sstr);
14027 padlist = padlist_dup(padlist, param);
14028 CvPADLIST_set(dstr, padlist);
14030 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14031 PoisonPADLIST(dstr);
14034 CvWEAKOUTSIDE(sstr)
14035 ? cv_dup( CvOUTSIDE(dstr), param)
14036 : cv_dup_inc(CvOUTSIDE(dstr), param);
14046 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14048 PERL_ARGS_ASSERT_SV_DUP_INC;
14049 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14053 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14055 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14056 PERL_ARGS_ASSERT_SV_DUP;
14058 /* Track every SV that (at least initially) had a reference count of 0.
14059 We need to do this by holding an actual reference to it in this array.
14060 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14061 (akin to the stashes hash, and the perl stack), we come unstuck if
14062 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14063 thread) is manipulated in a CLONE method, because CLONE runs before the
14064 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14065 (and fix things up by giving each a reference via the temps stack).
14066 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14067 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14068 before the walk of unreferenced happens and a reference to that is SV
14069 added to the temps stack. At which point we have the same SV considered
14070 to be in use, and free to be re-used. Not good.
14072 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14073 assert(param->unreferenced);
14074 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14080 /* duplicate a context */
14083 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14085 PERL_CONTEXT *ncxs;
14087 PERL_ARGS_ASSERT_CX_DUP;
14090 return (PERL_CONTEXT*)NULL;
14092 /* look for it in the table first */
14093 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14097 /* create anew and remember what it is */
14098 Newx(ncxs, max + 1, PERL_CONTEXT);
14099 ptr_table_store(PL_ptr_table, cxs, ncxs);
14100 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14103 PERL_CONTEXT * const ncx = &ncxs[ix];
14104 if (CxTYPE(ncx) == CXt_SUBST) {
14105 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14108 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14109 switch (CxTYPE(ncx)) {
14111 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14112 if(CxHASARGS(ncx)){
14113 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14115 ncx->blk_sub.savearray = NULL;
14117 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14118 ncx->blk_sub.prevcomppad);
14121 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14123 /* XXX should this sv_dup_inc? Or only if SvSCREAM ???? */
14124 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14125 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14126 /* XXX what do do with cur_top_env ???? */
14128 case CXt_LOOP_LAZYSV:
14129 ncx->blk_loop.state_u.lazysv.end
14130 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14131 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14132 duplication code instead.
14133 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14134 actually being the same function, and (2) order
14135 equivalence of the two unions.
14136 We can assert the later [but only at run time :-(] */
14137 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14138 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14141 ncx->blk_loop.state_u.ary.ary
14142 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14144 case CXt_LOOP_LIST:
14145 case CXt_LOOP_LAZYIV:
14146 /* code common to all 'for' CXt_LOOP_* types */
14147 ncx->blk_loop.itersave =
14148 sv_dup_inc(ncx->blk_loop.itersave, param);
14149 if (CxPADLOOP(ncx)) {
14150 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14151 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14152 ncx->blk_loop.oldcomppad =
14153 (PAD*)ptr_table_fetch(PL_ptr_table,
14154 ncx->blk_loop.oldcomppad);
14155 ncx->blk_loop.itervar_u.svp =
14156 &CX_CURPAD_SV(ncx->blk_loop, off);
14159 /* this copies the GV if CXp_FOR_GV, or the SV for an
14160 * alias (for \$x (...)) - relies on gv_dup being the
14161 * same as sv_dup */
14162 ncx->blk_loop.itervar_u.gv
14163 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14167 case CXt_LOOP_PLAIN:
14170 ncx->blk_format.prevcomppad =
14171 (PAD*)ptr_table_fetch(PL_ptr_table,
14172 ncx->blk_format.prevcomppad);
14173 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14174 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14175 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14179 ncx->blk_givwhen.defsv_save =
14180 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14193 /* duplicate a stack info structure */
14196 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14200 PERL_ARGS_ASSERT_SI_DUP;
14203 return (PERL_SI*)NULL;
14205 /* look for it in the table first */
14206 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14210 /* create anew and remember what it is */
14211 Newxz(nsi, 1, PERL_SI);
14212 ptr_table_store(PL_ptr_table, si, nsi);
14214 nsi->si_stack = av_dup_inc(si->si_stack, param);
14215 nsi->si_cxix = si->si_cxix;
14216 nsi->si_cxmax = si->si_cxmax;
14217 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14218 nsi->si_type = si->si_type;
14219 nsi->si_prev = si_dup(si->si_prev, param);
14220 nsi->si_next = si_dup(si->si_next, param);
14221 nsi->si_markoff = si->si_markoff;
14226 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14227 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14228 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14229 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14230 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14231 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14232 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14233 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14234 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14235 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14236 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14237 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14238 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14239 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14240 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14241 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14244 #define pv_dup_inc(p) SAVEPV(p)
14245 #define pv_dup(p) SAVEPV(p)
14246 #define svp_dup_inc(p,pp) any_dup(p,pp)
14248 /* map any object to the new equivent - either something in the
14249 * ptr table, or something in the interpreter structure
14253 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14257 PERL_ARGS_ASSERT_ANY_DUP;
14260 return (void*)NULL;
14262 /* look for it in the table first */
14263 ret = ptr_table_fetch(PL_ptr_table, v);
14267 /* see if it is part of the interpreter structure */
14268 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14269 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14277 /* duplicate the save stack */
14280 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14283 ANY * const ss = proto_perl->Isavestack;
14284 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14285 I32 ix = proto_perl->Isavestack_ix;
14298 void (*dptr) (void*);
14299 void (*dxptr) (pTHX_ void*);
14301 PERL_ARGS_ASSERT_SS_DUP;
14303 Newxz(nss, max, ANY);
14306 const UV uv = POPUV(ss,ix);
14307 const U8 type = (U8)uv & SAVE_MASK;
14309 TOPUV(nss,ix) = uv;
14311 case SAVEt_CLEARSV:
14312 case SAVEt_CLEARPADRANGE:
14314 case SAVEt_HELEM: /* hash element */
14315 case SAVEt_SV: /* scalar reference */
14316 sv = (const SV *)POPPTR(ss,ix);
14317 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14319 case SAVEt_ITEM: /* normal string */
14320 case SAVEt_GVSV: /* scalar slot in GV */
14321 sv = (const SV *)POPPTR(ss,ix);
14322 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14323 if (type == SAVEt_SV)
14327 case SAVEt_MORTALIZESV:
14328 case SAVEt_READONLY_OFF:
14329 sv = (const SV *)POPPTR(ss,ix);
14330 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14332 case SAVEt_FREEPADNAME:
14333 ptr = POPPTR(ss,ix);
14334 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14335 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14337 case SAVEt_SHARED_PVREF: /* char* in shared space */
14338 c = (char*)POPPTR(ss,ix);
14339 TOPPTR(nss,ix) = savesharedpv(c);
14340 ptr = POPPTR(ss,ix);
14341 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14343 case SAVEt_GENERIC_SVREF: /* generic sv */
14344 case SAVEt_SVREF: /* scalar reference */
14345 sv = (const SV *)POPPTR(ss,ix);
14346 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14347 if (type == SAVEt_SVREF)
14348 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14349 ptr = POPPTR(ss,ix);
14350 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14352 case SAVEt_GVSLOT: /* any slot in GV */
14353 sv = (const SV *)POPPTR(ss,ix);
14354 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14355 ptr = POPPTR(ss,ix);
14356 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14357 sv = (const SV *)POPPTR(ss,ix);
14358 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14360 case SAVEt_HV: /* hash reference */
14361 case SAVEt_AV: /* array reference */
14362 sv = (const SV *) POPPTR(ss,ix);
14363 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14365 case SAVEt_COMPPAD:
14367 sv = (const SV *) POPPTR(ss,ix);
14368 TOPPTR(nss,ix) = sv_dup(sv, param);
14370 case SAVEt_INT: /* int reference */
14371 ptr = POPPTR(ss,ix);
14372 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14373 intval = (int)POPINT(ss,ix);
14374 TOPINT(nss,ix) = intval;
14376 case SAVEt_LONG: /* long reference */
14377 ptr = POPPTR(ss,ix);
14378 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14379 longval = (long)POPLONG(ss,ix);
14380 TOPLONG(nss,ix) = longval;
14382 case SAVEt_I32: /* I32 reference */
14383 ptr = POPPTR(ss,ix);
14384 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14386 TOPINT(nss,ix) = i;
14388 case SAVEt_IV: /* IV reference */
14389 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14390 ptr = POPPTR(ss,ix);
14391 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14393 TOPIV(nss,ix) = iv;
14395 case SAVEt_TMPSFLOOR:
14397 TOPIV(nss,ix) = iv;
14399 case SAVEt_HPTR: /* HV* reference */
14400 case SAVEt_APTR: /* AV* reference */
14401 case SAVEt_SPTR: /* SV* reference */
14402 ptr = POPPTR(ss,ix);
14403 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14404 sv = (const SV *)POPPTR(ss,ix);
14405 TOPPTR(nss,ix) = sv_dup(sv, param);
14407 case SAVEt_VPTR: /* random* reference */
14408 ptr = POPPTR(ss,ix);
14409 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14411 case SAVEt_INT_SMALL:
14412 case SAVEt_I32_SMALL:
14413 case SAVEt_I16: /* I16 reference */
14414 case SAVEt_I8: /* I8 reference */
14416 ptr = POPPTR(ss,ix);
14417 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14419 case SAVEt_GENERIC_PVREF: /* generic char* */
14420 case SAVEt_PPTR: /* char* reference */
14421 ptr = POPPTR(ss,ix);
14422 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14423 c = (char*)POPPTR(ss,ix);
14424 TOPPTR(nss,ix) = pv_dup(c);
14426 case SAVEt_GP: /* scalar reference */
14427 gp = (GP*)POPPTR(ss,ix);
14428 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14429 (void)GpREFCNT_inc(gp);
14430 gv = (const GV *)POPPTR(ss,ix);
14431 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14434 ptr = POPPTR(ss,ix);
14435 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14436 /* these are assumed to be refcounted properly */
14438 switch (((OP*)ptr)->op_type) {
14440 case OP_LEAVESUBLV:
14444 case OP_LEAVEWRITE:
14445 TOPPTR(nss,ix) = ptr;
14448 (void) OpREFCNT_inc(o);
14452 TOPPTR(nss,ix) = NULL;
14457 TOPPTR(nss,ix) = NULL;
14459 case SAVEt_FREECOPHH:
14460 ptr = POPPTR(ss,ix);
14461 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14463 case SAVEt_ADELETE:
14464 av = (const AV *)POPPTR(ss,ix);
14465 TOPPTR(nss,ix) = av_dup_inc(av, param);
14467 TOPINT(nss,ix) = i;
14470 hv = (const HV *)POPPTR(ss,ix);
14471 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14473 TOPINT(nss,ix) = i;
14476 c = (char*)POPPTR(ss,ix);
14477 TOPPTR(nss,ix) = pv_dup_inc(c);
14479 case SAVEt_STACK_POS: /* Position on Perl stack */
14481 TOPINT(nss,ix) = i;
14483 case SAVEt_DESTRUCTOR:
14484 ptr = POPPTR(ss,ix);
14485 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14486 dptr = POPDPTR(ss,ix);
14487 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14488 any_dup(FPTR2DPTR(void *, dptr),
14491 case SAVEt_DESTRUCTOR_X:
14492 ptr = POPPTR(ss,ix);
14493 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14494 dxptr = POPDXPTR(ss,ix);
14495 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14496 any_dup(FPTR2DPTR(void *, dxptr),
14499 case SAVEt_REGCONTEXT:
14501 ix -= uv >> SAVE_TIGHT_SHIFT;
14503 case SAVEt_AELEM: /* array element */
14504 sv = (const SV *)POPPTR(ss,ix);
14505 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14507 TOPINT(nss,ix) = i;
14508 av = (const AV *)POPPTR(ss,ix);
14509 TOPPTR(nss,ix) = av_dup_inc(av, param);
14512 ptr = POPPTR(ss,ix);
14513 TOPPTR(nss,ix) = ptr;
14516 ptr = POPPTR(ss,ix);
14517 ptr = cophh_copy((COPHH*)ptr);
14518 TOPPTR(nss,ix) = ptr;
14520 TOPINT(nss,ix) = i;
14521 if (i & HINT_LOCALIZE_HH) {
14522 hv = (const HV *)POPPTR(ss,ix);
14523 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14526 case SAVEt_PADSV_AND_MORTALIZE:
14527 longval = (long)POPLONG(ss,ix);
14528 TOPLONG(nss,ix) = longval;
14529 ptr = POPPTR(ss,ix);
14530 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14531 sv = (const SV *)POPPTR(ss,ix);
14532 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14534 case SAVEt_SET_SVFLAGS:
14536 TOPINT(nss,ix) = i;
14538 TOPINT(nss,ix) = i;
14539 sv = (const SV *)POPPTR(ss,ix);
14540 TOPPTR(nss,ix) = sv_dup(sv, param);
14542 case SAVEt_COMPILE_WARNINGS:
14543 ptr = POPPTR(ss,ix);
14544 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14547 ptr = POPPTR(ss,ix);
14548 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14552 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14560 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14561 * flag to the result. This is done for each stash before cloning starts,
14562 * so we know which stashes want their objects cloned */
14565 do_mark_cloneable_stash(pTHX_ SV *const sv)
14567 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14569 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14570 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14571 if (cloner && GvCV(cloner)) {
14578 mXPUSHs(newSVhek(hvname));
14580 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14587 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14595 =for apidoc perl_clone
14597 Create and return a new interpreter by cloning the current one.
14599 C<perl_clone> takes these flags as parameters:
14601 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14602 without it we only clone the data and zero the stacks,
14603 with it we copy the stacks and the new perl interpreter is
14604 ready to run at the exact same point as the previous one.
14605 The pseudo-fork code uses C<COPY_STACKS> while the
14606 threads->create doesn't.
14608 C<CLONEf_KEEP_PTR_TABLE> -
14609 C<perl_clone> keeps a ptr_table with the pointer of the old
14610 variable as a key and the new variable as a value,
14611 this allows it to check if something has been cloned and not
14612 clone it again but rather just use the value and increase the
14613 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14614 the ptr_table using the function
14615 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14616 reason to keep it around is if you want to dup some of your own
14617 variable who are outside the graph perl scans, an example of this
14618 code is in F<threads.xs> create.
14620 C<CLONEf_CLONE_HOST> -
14621 This is a win32 thing, it is ignored on unix, it tells perls
14622 win32host code (which is c++) to clone itself, this is needed on
14623 win32 if you want to run two threads at the same time,
14624 if you just want to do some stuff in a separate perl interpreter
14625 and then throw it away and return to the original one,
14626 you don't need to do anything.
14631 /* XXX the above needs expanding by someone who actually understands it ! */
14632 EXTERN_C PerlInterpreter *
14633 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14636 perl_clone(PerlInterpreter *proto_perl, UV flags)
14639 #ifdef PERL_IMPLICIT_SYS
14641 PERL_ARGS_ASSERT_PERL_CLONE;
14643 /* perlhost.h so we need to call into it
14644 to clone the host, CPerlHost should have a c interface, sky */
14646 #ifndef __amigaos4__
14647 if (flags & CLONEf_CLONE_HOST) {
14648 return perl_clone_host(proto_perl,flags);
14651 return perl_clone_using(proto_perl, flags,
14653 proto_perl->IMemShared,
14654 proto_perl->IMemParse,
14656 proto_perl->IStdIO,
14660 proto_perl->IProc);
14664 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14665 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14666 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14667 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14668 struct IPerlDir* ipD, struct IPerlSock* ipS,
14669 struct IPerlProc* ipP)
14671 /* XXX many of the string copies here can be optimized if they're
14672 * constants; they need to be allocated as common memory and just
14673 * their pointers copied. */
14676 CLONE_PARAMS clone_params;
14677 CLONE_PARAMS* const param = &clone_params;
14679 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14681 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14682 #else /* !PERL_IMPLICIT_SYS */
14684 CLONE_PARAMS clone_params;
14685 CLONE_PARAMS* param = &clone_params;
14686 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14688 PERL_ARGS_ASSERT_PERL_CLONE;
14689 #endif /* PERL_IMPLICIT_SYS */
14691 /* for each stash, determine whether its objects should be cloned */
14692 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14693 PERL_SET_THX(my_perl);
14696 PoisonNew(my_perl, 1, PerlInterpreter);
14699 PL_defstash = NULL; /* may be used by perl malloc() */
14702 PL_scopestack_name = 0;
14704 PL_savestack_ix = 0;
14705 PL_savestack_max = -1;
14706 PL_sig_pending = 0;
14708 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14709 Zero(&PL_padname_undef, 1, PADNAME);
14710 Zero(&PL_padname_const, 1, PADNAME);
14711 # ifdef DEBUG_LEAKING_SCALARS
14712 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14714 # ifdef PERL_TRACE_OPS
14715 Zero(PL_op_exec_cnt, OP_max+2, UV);
14717 #else /* !DEBUGGING */
14718 Zero(my_perl, 1, PerlInterpreter);
14719 #endif /* DEBUGGING */
14721 #ifdef PERL_IMPLICIT_SYS
14722 /* host pointers */
14724 PL_MemShared = ipMS;
14725 PL_MemParse = ipMP;
14732 #endif /* PERL_IMPLICIT_SYS */
14735 param->flags = flags;
14736 /* Nothing in the core code uses this, but we make it available to
14737 extensions (using mg_dup). */
14738 param->proto_perl = proto_perl;
14739 /* Likely nothing will use this, but it is initialised to be consistent
14740 with Perl_clone_params_new(). */
14741 param->new_perl = my_perl;
14742 param->unreferenced = NULL;
14745 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14747 PL_body_arenas = NULL;
14748 Zero(&PL_body_roots, 1, PL_body_roots);
14752 PL_sv_arenaroot = NULL;
14754 PL_debug = proto_perl->Idebug;
14756 /* dbargs array probably holds garbage */
14759 PL_compiling = proto_perl->Icompiling;
14761 /* pseudo environmental stuff */
14762 PL_origargc = proto_perl->Iorigargc;
14763 PL_origargv = proto_perl->Iorigargv;
14765 #ifndef NO_TAINT_SUPPORT
14766 /* Set tainting stuff before PerlIO_debug can possibly get called */
14767 PL_tainting = proto_perl->Itainting;
14768 PL_taint_warn = proto_perl->Itaint_warn;
14770 PL_tainting = FALSE;
14771 PL_taint_warn = FALSE;
14774 PL_minus_c = proto_perl->Iminus_c;
14776 PL_localpatches = proto_perl->Ilocalpatches;
14777 PL_splitstr = proto_perl->Isplitstr;
14778 PL_minus_n = proto_perl->Iminus_n;
14779 PL_minus_p = proto_perl->Iminus_p;
14780 PL_minus_l = proto_perl->Iminus_l;
14781 PL_minus_a = proto_perl->Iminus_a;
14782 PL_minus_E = proto_perl->Iminus_E;
14783 PL_minus_F = proto_perl->Iminus_F;
14784 PL_doswitches = proto_perl->Idoswitches;
14785 PL_dowarn = proto_perl->Idowarn;
14786 #ifdef PERL_SAWAMPERSAND
14787 PL_sawampersand = proto_perl->Isawampersand;
14789 PL_unsafe = proto_perl->Iunsafe;
14790 PL_perldb = proto_perl->Iperldb;
14791 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14792 PL_exit_flags = proto_perl->Iexit_flags;
14794 /* XXX time(&PL_basetime) when asked for? */
14795 PL_basetime = proto_perl->Ibasetime;
14797 PL_maxsysfd = proto_perl->Imaxsysfd;
14798 PL_statusvalue = proto_perl->Istatusvalue;
14800 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14802 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14805 /* RE engine related */
14806 PL_regmatch_slab = NULL;
14807 PL_reg_curpm = NULL;
14809 PL_sub_generation = proto_perl->Isub_generation;
14811 /* funky return mechanisms */
14812 PL_forkprocess = proto_perl->Iforkprocess;
14814 /* internal state */
14815 PL_main_start = proto_perl->Imain_start;
14816 PL_eval_root = proto_perl->Ieval_root;
14817 PL_eval_start = proto_perl->Ieval_start;
14819 PL_filemode = proto_perl->Ifilemode;
14820 PL_lastfd = proto_perl->Ilastfd;
14821 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14824 PL_gensym = proto_perl->Igensym;
14826 PL_laststatval = proto_perl->Ilaststatval;
14827 PL_laststype = proto_perl->Ilaststype;
14830 PL_profiledata = NULL;
14832 PL_generation = proto_perl->Igeneration;
14834 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14835 PL_in_clean_all = proto_perl->Iin_clean_all;
14837 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14838 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14839 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14840 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14841 PL_nomemok = proto_perl->Inomemok;
14842 PL_an = proto_perl->Ian;
14843 PL_evalseq = proto_perl->Ievalseq;
14844 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14845 PL_origalen = proto_perl->Iorigalen;
14847 PL_sighandlerp = proto_perl->Isighandlerp;
14849 PL_runops = proto_perl->Irunops;
14851 PL_subline = proto_perl->Isubline;
14853 PL_cv_has_eval = proto_perl->Icv_has_eval;
14856 PL_cryptseen = proto_perl->Icryptseen;
14859 #ifdef USE_LOCALE_COLLATE
14860 PL_collation_ix = proto_perl->Icollation_ix;
14861 PL_collation_standard = proto_perl->Icollation_standard;
14862 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14863 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14864 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14865 #endif /* USE_LOCALE_COLLATE */
14867 #ifdef USE_LOCALE_NUMERIC
14868 PL_numeric_standard = proto_perl->Inumeric_standard;
14869 PL_numeric_local = proto_perl->Inumeric_local;
14870 #endif /* !USE_LOCALE_NUMERIC */
14872 /* Did the locale setup indicate UTF-8? */
14873 PL_utf8locale = proto_perl->Iutf8locale;
14874 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14875 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
14876 /* Unicode features (see perlrun/-C) */
14877 PL_unicode = proto_perl->Iunicode;
14879 /* Pre-5.8 signals control */
14880 PL_signals = proto_perl->Isignals;
14882 /* times() ticks per second */
14883 PL_clocktick = proto_perl->Iclocktick;
14885 /* Recursion stopper for PerlIO_find_layer */
14886 PL_in_load_module = proto_perl->Iin_load_module;
14888 /* sort() routine */
14889 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14891 /* Not really needed/useful since the reenrant_retint is "volatile",
14892 * but do it for consistency's sake. */
14893 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14895 /* Hooks to shared SVs and locks. */
14896 PL_sharehook = proto_perl->Isharehook;
14897 PL_lockhook = proto_perl->Ilockhook;
14898 PL_unlockhook = proto_perl->Iunlockhook;
14899 PL_threadhook = proto_perl->Ithreadhook;
14900 PL_destroyhook = proto_perl->Idestroyhook;
14901 PL_signalhook = proto_perl->Isignalhook;
14903 PL_globhook = proto_perl->Iglobhook;
14906 PL_last_swash_hv = NULL; /* reinits on demand */
14907 PL_last_swash_klen = 0;
14908 PL_last_swash_key[0]= '\0';
14909 PL_last_swash_tmps = (U8*)NULL;
14910 PL_last_swash_slen = 0;
14912 PL_srand_called = proto_perl->Isrand_called;
14913 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14915 if (flags & CLONEf_COPY_STACKS) {
14916 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14917 PL_tmps_ix = proto_perl->Itmps_ix;
14918 PL_tmps_max = proto_perl->Itmps_max;
14919 PL_tmps_floor = proto_perl->Itmps_floor;
14921 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14922 * NOTE: unlike the others! */
14923 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14924 PL_scopestack_max = proto_perl->Iscopestack_max;
14926 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14927 * NOTE: unlike the others! */
14928 PL_savestack_ix = proto_perl->Isavestack_ix;
14929 PL_savestack_max = proto_perl->Isavestack_max;
14932 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14933 PL_top_env = &PL_start_env;
14935 PL_op = proto_perl->Iop;
14938 PL_Xpv = (XPV*)NULL;
14939 my_perl->Ina = proto_perl->Ina;
14941 PL_statbuf = proto_perl->Istatbuf;
14942 PL_statcache = proto_perl->Istatcache;
14944 #ifndef NO_TAINT_SUPPORT
14945 PL_tainted = proto_perl->Itainted;
14947 PL_tainted = FALSE;
14949 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14951 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14953 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14954 PL_restartop = proto_perl->Irestartop;
14955 PL_in_eval = proto_perl->Iin_eval;
14956 PL_delaymagic = proto_perl->Idelaymagic;
14957 PL_phase = proto_perl->Iphase;
14958 PL_localizing = proto_perl->Ilocalizing;
14960 PL_hv_fetch_ent_mh = NULL;
14961 PL_modcount = proto_perl->Imodcount;
14962 PL_lastgotoprobe = NULL;
14963 PL_dumpindent = proto_perl->Idumpindent;
14965 PL_efloatbuf = NULL; /* reinits on demand */
14966 PL_efloatsize = 0; /* reinits on demand */
14970 PL_colorset = 0; /* reinits PL_colors[] */
14971 /*PL_colors[6] = {0,0,0,0,0,0};*/
14973 /* Pluggable optimizer */
14974 PL_peepp = proto_perl->Ipeepp;
14975 PL_rpeepp = proto_perl->Irpeepp;
14976 /* op_free() hook */
14977 PL_opfreehook = proto_perl->Iopfreehook;
14979 #ifdef USE_REENTRANT_API
14980 /* XXX: things like -Dm will segfault here in perlio, but doing
14981 * PERL_SET_CONTEXT(proto_perl);
14982 * breaks too many other things
14984 Perl_reentrant_init(aTHX);
14987 /* create SV map for pointer relocation */
14988 PL_ptr_table = ptr_table_new();
14990 /* initialize these special pointers as early as possible */
14992 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14993 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14994 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14995 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14996 &PL_padname_const);
14998 /* create (a non-shared!) shared string table */
14999 PL_strtab = newHV();
15000 HvSHAREKEYS_off(PL_strtab);
15001 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15002 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15004 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15006 /* This PV will be free'd special way so must set it same way op.c does */
15007 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15008 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15010 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15011 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15012 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15013 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15015 param->stashes = newAV(); /* Setup array of objects to call clone on */
15016 /* This makes no difference to the implementation, as it always pushes
15017 and shifts pointers to other SVs without changing their reference
15018 count, with the array becoming empty before it is freed. However, it
15019 makes it conceptually clear what is going on, and will avoid some
15020 work inside av.c, filling slots between AvFILL() and AvMAX() with
15021 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15022 AvREAL_off(param->stashes);
15024 if (!(flags & CLONEf_COPY_STACKS)) {
15025 param->unreferenced = newAV();
15028 #ifdef PERLIO_LAYERS
15029 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15030 PerlIO_clone(aTHX_ proto_perl, param);
15033 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15034 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15035 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15036 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15037 PL_xsubfilename = proto_perl->Ixsubfilename;
15038 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15039 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15042 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15043 PL_inplace = SAVEPV(proto_perl->Iinplace);
15044 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15046 /* magical thingies */
15048 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
15049 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
15050 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
15053 /* Clone the regex array */
15054 /* ORANGE FIXME for plugins, probably in the SV dup code.
15055 newSViv(PTR2IV(CALLREGDUPE(
15056 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15058 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15059 PL_regex_pad = AvARRAY(PL_regex_padav);
15061 PL_stashpadmax = proto_perl->Istashpadmax;
15062 PL_stashpadix = proto_perl->Istashpadix ;
15063 Newx(PL_stashpad, PL_stashpadmax, HV *);
15066 for (; o < PL_stashpadmax; ++o)
15067 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15070 /* shortcuts to various I/O objects */
15071 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15072 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15073 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15074 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15075 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15076 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15077 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15079 /* shortcuts to regexp stuff */
15080 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15082 /* shortcuts to misc objects */
15083 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15085 /* shortcuts to debugging objects */
15086 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15087 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15088 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15089 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15090 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15091 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15092 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15094 /* symbol tables */
15095 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15096 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15097 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15098 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15099 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15101 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15102 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15103 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15104 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15105 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15106 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15107 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15108 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15109 PL_savebegin = proto_perl->Isavebegin;
15111 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15113 /* subprocess state */
15114 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15116 if (proto_perl->Iop_mask)
15117 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15120 /* PL_asserting = proto_perl->Iasserting; */
15122 /* current interpreter roots */
15123 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15125 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15128 /* runtime control stuff */
15129 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15131 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15133 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15135 /* interpreter atexit processing */
15136 PL_exitlistlen = proto_perl->Iexitlistlen;
15137 if (PL_exitlistlen) {
15138 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15139 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15142 PL_exitlist = (PerlExitListEntry*)NULL;
15144 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15145 if (PL_my_cxt_size) {
15146 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15147 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15148 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15149 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15150 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15154 PL_my_cxt_list = (void**)NULL;
15155 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15156 PL_my_cxt_keys = (const char**)NULL;
15159 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15160 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15161 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15162 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15164 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15166 PAD_CLONE_VARS(proto_perl, param);
15168 #ifdef HAVE_INTERP_INTERN
15169 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15172 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15174 #ifdef PERL_USES_PL_PIDSTATUS
15175 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15177 PL_osname = SAVEPV(proto_perl->Iosname);
15178 PL_parser = parser_dup(proto_perl->Iparser, param);
15180 /* XXX this only works if the saved cop has already been cloned */
15181 if (proto_perl->Iparser) {
15182 PL_parser->saved_curcop = (COP*)any_dup(
15183 proto_perl->Iparser->saved_curcop,
15187 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15189 #ifdef USE_LOCALE_CTYPE
15190 /* Should we warn if uses locale? */
15191 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15194 #ifdef USE_LOCALE_COLLATE
15195 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15196 #endif /* USE_LOCALE_COLLATE */
15198 #ifdef USE_LOCALE_NUMERIC
15199 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15200 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15201 #endif /* !USE_LOCALE_NUMERIC */
15203 /* Unicode inversion lists */
15204 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15205 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15206 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15207 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15209 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15210 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15212 /* utf8 character class swashes */
15213 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15214 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15216 for (i = 0; i < POSIX_CC_COUNT; i++) {
15217 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15219 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15220 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15221 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15222 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15223 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15224 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15225 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15226 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15227 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15228 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15229 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15230 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15231 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15232 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15233 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15234 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15235 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15237 if (proto_perl->Ipsig_pend) {
15238 Newxz(PL_psig_pend, SIG_SIZE, int);
15241 PL_psig_pend = (int*)NULL;
15244 if (proto_perl->Ipsig_name) {
15245 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15246 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15248 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15251 PL_psig_ptr = (SV**)NULL;
15252 PL_psig_name = (SV**)NULL;
15255 if (flags & CLONEf_COPY_STACKS) {
15256 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15257 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15258 PL_tmps_ix+1, param);
15260 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15261 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15262 Newxz(PL_markstack, i, I32);
15263 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15264 - proto_perl->Imarkstack);
15265 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15266 - proto_perl->Imarkstack);
15267 Copy(proto_perl->Imarkstack, PL_markstack,
15268 PL_markstack_ptr - PL_markstack + 1, I32);
15270 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15271 * NOTE: unlike the others! */
15272 Newxz(PL_scopestack, PL_scopestack_max, I32);
15273 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15276 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15277 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15279 /* reset stack AV to correct length before its duped via
15280 * PL_curstackinfo */
15281 AvFILLp(proto_perl->Icurstack) =
15282 proto_perl->Istack_sp - proto_perl->Istack_base;
15284 /* NOTE: si_dup() looks at PL_markstack */
15285 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15287 /* PL_curstack = PL_curstackinfo->si_stack; */
15288 PL_curstack = av_dup(proto_perl->Icurstack, param);
15289 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15291 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15292 PL_stack_base = AvARRAY(PL_curstack);
15293 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15294 - proto_perl->Istack_base);
15295 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15297 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15298 PL_savestack = ss_dup(proto_perl, param);
15302 ENTER; /* perl_destruct() wants to LEAVE; */
15305 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15306 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15308 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15309 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15310 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15311 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15312 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15313 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15315 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15317 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15318 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15319 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15321 PL_stashcache = newHV();
15323 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15324 proto_perl->Iwatchaddr);
15325 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15326 if (PL_debug && PL_watchaddr) {
15327 PerlIO_printf(Perl_debug_log,
15328 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15329 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15330 PTR2UV(PL_watchok));
15333 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15334 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15335 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15337 /* Call the ->CLONE method, if it exists, for each of the stashes
15338 identified by sv_dup() above.
15340 while(av_tindex(param->stashes) != -1) {
15341 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15342 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15343 if (cloner && GvCV(cloner)) {
15348 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15350 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15356 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15357 ptr_table_free(PL_ptr_table);
15358 PL_ptr_table = NULL;
15361 if (!(flags & CLONEf_COPY_STACKS)) {
15362 unreferenced_to_tmp_stack(param->unreferenced);
15365 SvREFCNT_dec(param->stashes);
15367 /* orphaned? eg threads->new inside BEGIN or use */
15368 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15369 SvREFCNT_inc_simple_void(PL_compcv);
15370 SAVEFREESV(PL_compcv);
15377 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15379 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15381 if (AvFILLp(unreferenced) > -1) {
15382 SV **svp = AvARRAY(unreferenced);
15383 SV **const last = svp + AvFILLp(unreferenced);
15387 if (SvREFCNT(*svp) == 1)
15389 } while (++svp <= last);
15391 EXTEND_MORTAL(count);
15392 svp = AvARRAY(unreferenced);
15395 if (SvREFCNT(*svp) == 1) {
15396 /* Our reference is the only one to this SV. This means that
15397 in this thread, the scalar effectively has a 0 reference.
15398 That doesn't work (cleanup never happens), so donate our
15399 reference to it onto the save stack. */
15400 PL_tmps_stack[++PL_tmps_ix] = *svp;
15402 /* As an optimisation, because we are already walking the
15403 entire array, instead of above doing either
15404 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15405 release our reference to the scalar, so that at the end of
15406 the array owns zero references to the scalars it happens to
15407 point to. We are effectively converting the array from
15408 AvREAL() on to AvREAL() off. This saves the av_clear()
15409 (triggered by the SvREFCNT_dec(unreferenced) below) from
15410 walking the array a second time. */
15411 SvREFCNT_dec(*svp);
15414 } while (++svp <= last);
15415 AvREAL_off(unreferenced);
15417 SvREFCNT_dec_NN(unreferenced);
15421 Perl_clone_params_del(CLONE_PARAMS *param)
15423 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15425 PerlInterpreter *const to = param->new_perl;
15427 PerlInterpreter *const was = PERL_GET_THX;
15429 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15435 SvREFCNT_dec(param->stashes);
15436 if (param->unreferenced)
15437 unreferenced_to_tmp_stack(param->unreferenced);
15447 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15450 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15451 does a dTHX; to get the context from thread local storage.
15452 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15453 a version that passes in my_perl. */
15454 PerlInterpreter *const was = PERL_GET_THX;
15455 CLONE_PARAMS *param;
15457 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15463 /* Given that we've set the context, we can do this unshared. */
15464 Newx(param, 1, CLONE_PARAMS);
15467 param->proto_perl = from;
15468 param->new_perl = to;
15469 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15470 AvREAL_off(param->stashes);
15471 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15479 #endif /* USE_ITHREADS */
15482 Perl_init_constants(pTHX)
15484 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15485 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15486 SvANY(&PL_sv_undef) = NULL;
15488 SvANY(&PL_sv_no) = new_XPVNV();
15489 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15490 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15491 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15494 SvANY(&PL_sv_yes) = new_XPVNV();
15495 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15496 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15497 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15500 SvPV_set(&PL_sv_no, (char*)PL_No);
15501 SvCUR_set(&PL_sv_no, 0);
15502 SvLEN_set(&PL_sv_no, 0);
15503 SvIV_set(&PL_sv_no, 0);
15504 SvNV_set(&PL_sv_no, 0);
15506 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15507 SvCUR_set(&PL_sv_yes, 1);
15508 SvLEN_set(&PL_sv_yes, 0);
15509 SvIV_set(&PL_sv_yes, 1);
15510 SvNV_set(&PL_sv_yes, 1);
15512 PadnamePV(&PL_padname_const) = (char *)PL_No;
15516 =head1 Unicode Support
15518 =for apidoc sv_recode_to_utf8
15520 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15521 of C<sv> is assumed to be octets in that encoding, and C<sv>
15522 will be converted into Unicode (and UTF-8).
15524 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15525 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15526 an C<Encode::XS> Encoding object, bad things will happen.
15527 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15529 The PV of C<sv> is returned.
15534 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15536 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15538 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15547 if (SvPADTMP(nsv)) {
15548 nsv = sv_newmortal();
15549 SvSetSV_nosteal(nsv, sv);
15558 Passing sv_yes is wrong - it needs to be or'ed set of constants
15559 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15560 remove converted chars from source.
15562 Both will default the value - let them.
15564 XPUSHs(&PL_sv_yes);
15567 call_method("decode", G_SCALAR);
15571 s = SvPV_const(uni, len);
15572 if (s != SvPVX_const(sv)) {
15573 SvGROW(sv, len + 1);
15574 Move(s, SvPVX(sv), len + 1, char);
15575 SvCUR_set(sv, len);
15580 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15581 /* clear pos and any utf8 cache */
15582 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15585 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15586 magic_setutf8(sv,mg); /* clear UTF8 cache */
15591 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15595 =for apidoc sv_cat_decode
15597 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15598 assumed to be octets in that encoding and decoding the input starts
15599 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15600 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15601 when the string C<tstr> appears in decoding output or the input ends on
15602 the PV of C<ssv>. The value which C<offset> points will be modified
15603 to the last input position on C<ssv>.
15605 Returns TRUE if the terminator was found, else returns FALSE.
15610 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15611 SV *ssv, int *offset, char *tstr, int tlen)
15615 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15617 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15628 offsv = newSViv(*offset);
15630 mPUSHp(tstr, tlen);
15632 call_method("cat_decode", G_SCALAR);
15634 ret = SvTRUE(TOPs);
15635 *offset = SvIV(offsv);
15641 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15646 /* ---------------------------------------------------------------------
15648 * support functions for report_uninit()
15651 /* the maxiumum size of array or hash where we will scan looking
15652 * for the undefined element that triggered the warning */
15654 #define FUV_MAX_SEARCH_SIZE 1000
15656 /* Look for an entry in the hash whose value has the same SV as val;
15657 * If so, return a mortal copy of the key. */
15660 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15666 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15668 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15669 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15672 array = HvARRAY(hv);
15674 for (i=HvMAX(hv); i>=0; i--) {
15676 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15677 if (HeVAL(entry) != val)
15679 if ( HeVAL(entry) == &PL_sv_undef ||
15680 HeVAL(entry) == &PL_sv_placeholder)
15684 if (HeKLEN(entry) == HEf_SVKEY)
15685 return sv_mortalcopy(HeKEY_sv(entry));
15686 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15692 /* Look for an entry in the array whose value has the same SV as val;
15693 * If so, return the index, otherwise return -1. */
15696 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15698 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15700 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15701 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15704 if (val != &PL_sv_undef) {
15705 SV ** const svp = AvARRAY(av);
15708 for (i=AvFILLp(av); i>=0; i--)
15715 /* varname(): return the name of a variable, optionally with a subscript.
15716 * If gv is non-zero, use the name of that global, along with gvtype (one
15717 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15718 * targ. Depending on the value of the subscript_type flag, return:
15721 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15722 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15723 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15724 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15727 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15728 const SV *const keyname, SSize_t aindex, int subscript_type)
15731 SV * const name = sv_newmortal();
15732 if (gv && isGV(gv)) {
15734 buffer[0] = gvtype;
15737 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15739 gv_fullname4(name, gv, buffer, 0);
15741 if ((unsigned int)SvPVX(name)[1] <= 26) {
15743 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15745 /* Swap the 1 unprintable control character for the 2 byte pretty
15746 version - ie substr($name, 1, 1) = $buffer; */
15747 sv_insert(name, 1, 1, buffer, 2);
15751 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15754 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15756 if (!cv || !CvPADLIST(cv))
15758 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15759 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15763 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15764 SV * const sv = newSV(0);
15766 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15768 *SvPVX(name) = '$';
15769 Perl_sv_catpvf(aTHX_ name, "{%s}",
15770 pv_pretty(sv, pv, len, 32, NULL, NULL,
15771 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15772 SvREFCNT_dec_NN(sv);
15774 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15775 *SvPVX(name) = '$';
15776 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15778 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15779 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15780 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15788 =for apidoc find_uninit_var
15790 Find the name of the undefined variable (if any) that caused the operator
15791 to issue a "Use of uninitialized value" warning.
15792 If match is true, only return a name if its value matches C<uninit_sv>.
15793 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15794 warning, then following the direct child of the op may yield an
15795 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15796 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15797 the variable name if we get an exact match.
15798 C<desc_p> points to a string pointer holding the description of the op.
15799 This may be updated if needed.
15801 The name is returned as a mortal SV.
15803 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15804 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15810 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15811 bool match, const char **desc_p)
15816 const OP *o, *o2, *kid;
15818 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15820 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15821 uninit_sv == &PL_sv_placeholder)))
15824 switch (obase->op_type) {
15827 /* undef should care if its args are undef - any warnings
15828 * will be from tied/magic vars */
15836 const bool pad = ( obase->op_type == OP_PADAV
15837 || obase->op_type == OP_PADHV
15838 || obase->op_type == OP_PADRANGE
15841 const bool hash = ( obase->op_type == OP_PADHV
15842 || obase->op_type == OP_RV2HV
15843 || (obase->op_type == OP_PADRANGE
15844 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15848 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15850 if (pad) { /* @lex, %lex */
15851 sv = PAD_SVl(obase->op_targ);
15855 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15856 /* @global, %global */
15857 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15860 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15862 else if (obase == PL_op) /* @{expr}, %{expr} */
15863 return find_uninit_var(cUNOPx(obase)->op_first,
15864 uninit_sv, match, desc_p);
15865 else /* @{expr}, %{expr} as a sub-expression */
15869 /* attempt to find a match within the aggregate */
15871 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15873 subscript_type = FUV_SUBSCRIPT_HASH;
15876 index = find_array_subscript((const AV *)sv, uninit_sv);
15878 subscript_type = FUV_SUBSCRIPT_ARRAY;
15881 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15884 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15885 keysv, index, subscript_type);
15889 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15891 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15892 if (!gv || !GvSTASH(gv))
15894 if (match && (GvSV(gv) != uninit_sv))
15896 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15899 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15902 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15904 return varname(NULL, '$', obase->op_targ,
15905 NULL, 0, FUV_SUBSCRIPT_NONE);
15908 gv = cGVOPx_gv(obase);
15909 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15911 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15913 case OP_AELEMFAST_LEX:
15916 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15917 if (!av || SvRMAGICAL(av))
15919 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15920 if (!svp || *svp != uninit_sv)
15923 return varname(NULL, '$', obase->op_targ,
15924 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15927 gv = cGVOPx_gv(obase);
15932 AV *const av = GvAV(gv);
15933 if (!av || SvRMAGICAL(av))
15935 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15936 if (!svp || *svp != uninit_sv)
15939 return varname(gv, '$', 0,
15940 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15942 NOT_REACHED; /* NOTREACHED */
15945 o = cUNOPx(obase)->op_first;
15946 if (!o || o->op_type != OP_NULL ||
15947 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15949 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15954 bool negate = FALSE;
15956 if (PL_op == obase)
15957 /* $a[uninit_expr] or $h{uninit_expr} */
15958 return find_uninit_var(cBINOPx(obase)->op_last,
15959 uninit_sv, match, desc_p);
15962 o = cBINOPx(obase)->op_first;
15963 kid = cBINOPx(obase)->op_last;
15965 /* get the av or hv, and optionally the gv */
15967 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15968 sv = PAD_SV(o->op_targ);
15970 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15971 && cUNOPo->op_first->op_type == OP_GV)
15973 gv = cGVOPx_gv(cUNOPo->op_first);
15977 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15982 if (kid && kid->op_type == OP_NEGATE) {
15984 kid = cUNOPx(kid)->op_first;
15987 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15988 /* index is constant */
15991 kidsv = newSVpvs_flags("-", SVs_TEMP);
15992 sv_catsv(kidsv, cSVOPx_sv(kid));
15995 kidsv = cSVOPx_sv(kid);
15999 if (obase->op_type == OP_HELEM) {
16000 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16001 if (!he || HeVAL(he) != uninit_sv)
16005 SV * const opsv = cSVOPx_sv(kid);
16006 const IV opsviv = SvIV(opsv);
16007 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16008 negate ? - opsviv : opsviv,
16010 if (!svp || *svp != uninit_sv)
16014 if (obase->op_type == OP_HELEM)
16015 return varname(gv, '%', o->op_targ,
16016 kidsv, 0, FUV_SUBSCRIPT_HASH);
16018 return varname(gv, '@', o->op_targ, NULL,
16019 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16020 FUV_SUBSCRIPT_ARRAY);
16023 /* index is an expression;
16024 * attempt to find a match within the aggregate */
16025 if (obase->op_type == OP_HELEM) {
16026 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16028 return varname(gv, '%', o->op_targ,
16029 keysv, 0, FUV_SUBSCRIPT_HASH);
16032 const SSize_t index
16033 = find_array_subscript((const AV *)sv, uninit_sv);
16035 return varname(gv, '@', o->op_targ,
16036 NULL, index, FUV_SUBSCRIPT_ARRAY);
16041 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16043 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16045 NOT_REACHED; /* NOTREACHED */
16048 case OP_MULTIDEREF: {
16049 /* If we were executing OP_MULTIDEREF when the undef warning
16050 * triggered, then it must be one of the index values within
16051 * that triggered it. If not, then the only possibility is that
16052 * the value retrieved by the last aggregate index might be the
16053 * culprit. For the former, we set PL_multideref_pc each time before
16054 * using an index, so work though the item list until we reach
16055 * that point. For the latter, just work through the entire item
16056 * list; the last aggregate retrieved will be the candidate.
16057 * There is a third rare possibility: something triggered
16058 * magic while fetching an array/hash element. Just display
16059 * nothing in this case.
16062 /* the named aggregate, if any */
16063 PADOFFSET agg_targ = 0;
16065 /* the last-seen index */
16067 PADOFFSET index_targ;
16069 IV index_const_iv = 0; /* init for spurious compiler warn */
16070 SV *index_const_sv;
16071 int depth = 0; /* how many array/hash lookups we've done */
16073 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16074 UNOP_AUX_item *last = NULL;
16075 UV actions = items->uv;
16078 if (PL_op == obase) {
16079 last = PL_multideref_pc;
16080 assert(last >= items && last <= items + items[-1].uv);
16087 switch (actions & MDEREF_ACTION_MASK) {
16089 case MDEREF_reload:
16090 actions = (++items)->uv;
16093 case MDEREF_HV_padhv_helem: /* $lex{...} */
16096 case MDEREF_AV_padav_aelem: /* $lex[...] */
16097 agg_targ = (++items)->pad_offset;
16101 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16104 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16106 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16107 assert(isGV_with_GP(agg_gv));
16110 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16111 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16114 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16115 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16121 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16122 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16125 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16126 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16133 index_const_sv = NULL;
16135 index_type = (actions & MDEREF_INDEX_MASK);
16136 switch (index_type) {
16137 case MDEREF_INDEX_none:
16139 case MDEREF_INDEX_const:
16141 index_const_sv = UNOP_AUX_item_sv(++items)
16143 index_const_iv = (++items)->iv;
16145 case MDEREF_INDEX_padsv:
16146 index_targ = (++items)->pad_offset;
16148 case MDEREF_INDEX_gvsv:
16149 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16150 assert(isGV_with_GP(index_gv));
16154 if (index_type != MDEREF_INDEX_none)
16157 if ( index_type == MDEREF_INDEX_none
16158 || (actions & MDEREF_FLAG_last)
16159 || (last && items >= last)
16163 actions >>= MDEREF_SHIFT;
16166 if (PL_op == obase) {
16167 /* most likely index was undef */
16169 *desc_p = ( (actions & MDEREF_FLAG_last)
16170 && (obase->op_private
16171 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16173 (obase->op_private & OPpMULTIDEREF_EXISTS)
16176 : is_hv ? "hash element" : "array element";
16177 assert(index_type != MDEREF_INDEX_none);
16179 if (GvSV(index_gv) == uninit_sv)
16180 return varname(index_gv, '$', 0, NULL, 0,
16181 FUV_SUBSCRIPT_NONE);
16186 if (PL_curpad[index_targ] == uninit_sv)
16187 return varname(NULL, '$', index_targ,
16188 NULL, 0, FUV_SUBSCRIPT_NONE);
16192 /* If we got to this point it was undef on a const subscript,
16193 * so magic probably involved, e.g. $ISA[0]. Give up. */
16197 /* the SV returned by pp_multideref() was undef, if anything was */
16203 sv = PAD_SV(agg_targ);
16205 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16209 if (index_type == MDEREF_INDEX_const) {
16214 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16215 if (!he || HeVAL(he) != uninit_sv)
16219 SV * const * const svp =
16220 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16221 if (!svp || *svp != uninit_sv)
16226 ? varname(agg_gv, '%', agg_targ,
16227 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16228 : varname(agg_gv, '@', agg_targ,
16229 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16232 /* index is an var */
16234 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16236 return varname(agg_gv, '%', agg_targ,
16237 keysv, 0, FUV_SUBSCRIPT_HASH);
16240 const SSize_t index
16241 = find_array_subscript((const AV *)sv, uninit_sv);
16243 return varname(agg_gv, '@', agg_targ,
16244 NULL, index, FUV_SUBSCRIPT_ARRAY);
16248 return varname(agg_gv,
16250 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16252 NOT_REACHED; /* NOTREACHED */
16256 /* only examine RHS */
16257 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16261 o = cUNOPx(obase)->op_first;
16262 if ( o->op_type == OP_PUSHMARK
16263 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16267 if (!OpHAS_SIBLING(o)) {
16268 /* one-arg version of open is highly magical */
16270 if (o->op_type == OP_GV) { /* open FOO; */
16272 if (match && GvSV(gv) != uninit_sv)
16274 return varname(gv, '$', 0,
16275 NULL, 0, FUV_SUBSCRIPT_NONE);
16277 /* other possibilities not handled are:
16278 * open $x; or open my $x; should return '${*$x}'
16279 * open expr; should return '$'.expr ideally
16286 /* ops where $_ may be an implicit arg */
16291 if ( !(obase->op_flags & OPf_STACKED)) {
16292 if (uninit_sv == DEFSV)
16293 return newSVpvs_flags("$_", SVs_TEMP);
16294 else if (obase->op_targ
16295 && uninit_sv == PAD_SVl(obase->op_targ))
16296 return varname(NULL, '$', obase->op_targ, NULL, 0,
16297 FUV_SUBSCRIPT_NONE);
16304 match = 1; /* print etc can return undef on defined args */
16305 /* skip filehandle as it can't produce 'undef' warning */
16306 o = cUNOPx(obase)->op_first;
16307 if ((obase->op_flags & OPf_STACKED)
16309 ( o->op_type == OP_PUSHMARK
16310 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16311 o = OpSIBLING(OpSIBLING(o));
16315 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16316 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16318 /* the following ops are capable of returning PL_sv_undef even for
16319 * defined arg(s) */
16338 case OP_GETPEERNAME:
16385 case OP_SMARTMATCH:
16394 /* XXX tmp hack: these two may call an XS sub, and currently
16395 XS subs don't have a SUB entry on the context stack, so CV and
16396 pad determination goes wrong, and BAD things happen. So, just
16397 don't try to determine the value under those circumstances.
16398 Need a better fix at dome point. DAPM 11/2007 */
16404 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16405 if (gv && GvSV(gv) == uninit_sv)
16406 return newSVpvs_flags("$.", SVs_TEMP);
16411 /* def-ness of rval pos() is independent of the def-ness of its arg */
16412 if ( !(obase->op_flags & OPf_MOD))
16417 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16418 return newSVpvs_flags("${$/}", SVs_TEMP);
16423 if (!(obase->op_flags & OPf_KIDS))
16425 o = cUNOPx(obase)->op_first;
16431 /* This loop checks all the kid ops, skipping any that cannot pos-
16432 * sibly be responsible for the uninitialized value; i.e., defined
16433 * constants and ops that return nothing. If there is only one op
16434 * left that is not skipped, then we *know* it is responsible for
16435 * the uninitialized value. If there is more than one op left, we
16436 * have to look for an exact match in the while() loop below.
16437 * Note that we skip padrange, because the individual pad ops that
16438 * it replaced are still in the tree, so we work on them instead.
16441 for (kid=o; kid; kid = OpSIBLING(kid)) {
16442 const OPCODE type = kid->op_type;
16443 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16444 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16445 || (type == OP_PUSHMARK)
16446 || (type == OP_PADRANGE)
16450 if (o2) { /* more than one found */
16457 return find_uninit_var(o2, uninit_sv, match, desc_p);
16459 /* scan all args */
16461 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16473 =for apidoc report_uninit
16475 Print appropriate "Use of uninitialized variable" warning.
16481 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16483 const char *desc = NULL;
16484 SV* varname = NULL;
16487 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16490 if (uninit_sv && PL_curpad) {
16491 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16493 sv_insert(varname, 0, 0, " ", 1);
16496 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16497 /* we've reached the end of a sort block or sub,
16498 * and the uninit value is probably what that code returned */
16501 /* PL_warn_uninit_sv is constant */
16502 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16504 /* diag_listed_as: Use of uninitialized value%s */
16505 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16506 SVfARG(varname ? varname : &PL_sv_no),
16509 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16515 * ex: set ts=8 sts=4 sw=4 et: