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 # if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L && !defined(__VMS)
48 /* Missing proto on LynxOS */
49 char *gconvert(double, int, int, char *);
52 #ifdef PERL_NEW_COPY_ON_WRITE
53 # ifndef SV_COW_THRESHOLD
54 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
56 # ifndef SV_COWBUF_THRESHOLD
57 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
59 # ifndef SV_COW_MAX_WASTE_THRESHOLD
60 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
62 # ifndef SV_COWBUF_WASTE_THRESHOLD
63 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
65 # ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
66 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
68 # ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
69 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
72 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
75 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
77 # define GE_COW_THRESHOLD(cur) 1
79 #if SV_COWBUF_THRESHOLD
80 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
82 # define GE_COWBUF_THRESHOLD(cur) 1
84 #if SV_COW_MAX_WASTE_THRESHOLD
85 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
87 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
89 #if SV_COWBUF_WASTE_THRESHOLD
90 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
92 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
94 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
95 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
97 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
99 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
100 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
102 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
105 #define CHECK_COW_THRESHOLD(cur,len) (\
106 GE_COW_THRESHOLD((cur)) && \
107 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
108 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
110 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
111 GE_COWBUF_THRESHOLD((cur)) && \
112 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
113 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
115 /* void Gconvert: on Linux at least, gcvt (which Gconvert gets deffed to),
116 * has a mandatory return value, even though that value is just the same
119 #ifdef PERL_UTF8_CACHE_ASSERT
120 /* if adding more checks watch out for the following tests:
121 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
122 * lib/utf8.t lib/Unicode/Collate/t/index.t
125 # define ASSERT_UTF8_CACHE(cache) \
126 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
127 assert((cache)[2] <= (cache)[3]); \
128 assert((cache)[3] <= (cache)[1]);} \
131 # define ASSERT_UTF8_CACHE(cache) NOOP
134 #ifdef PERL_OLD_COPY_ON_WRITE
135 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
136 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
139 /* ============================================================================
141 =head1 Allocation and deallocation of SVs.
142 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
143 sv, av, hv...) contains type and reference count information, and for
144 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
145 contains fields specific to each type. Some types store all they need
146 in the head, so don't have a body.
148 In all but the most memory-paranoid configurations (ex: PURIFY), heads
149 and bodies are allocated out of arenas, which by default are
150 approximately 4K chunks of memory parcelled up into N heads or bodies.
151 Sv-bodies are allocated by their sv-type, guaranteeing size
152 consistency needed to allocate safely from arrays.
154 For SV-heads, the first slot in each arena is reserved, and holds a
155 link to the next arena, some flags, and a note of the number of slots.
156 Snaked through each arena chain is a linked list of free items; when
157 this becomes empty, an extra arena is allocated and divided up into N
158 items which are threaded into the free list.
160 SV-bodies are similar, but they use arena-sets by default, which
161 separate the link and info from the arena itself, and reclaim the 1st
162 slot in the arena. SV-bodies are further described later.
164 The following global variables are associated with arenas:
166 PL_sv_arenaroot pointer to list of SV arenas
167 PL_sv_root pointer to list of free SV structures
169 PL_body_arenas head of linked-list of body arenas
170 PL_body_roots[] array of pointers to list of free bodies of svtype
171 arrays are indexed by the svtype needed
173 A few special SV heads are not allocated from an arena, but are
174 instead directly created in the interpreter structure, eg PL_sv_undef.
175 The size of arenas can be changed from the default by setting
176 PERL_ARENA_SIZE appropriately at compile time.
178 The SV arena serves the secondary purpose of allowing still-live SVs
179 to be located and destroyed during final cleanup.
181 At the lowest level, the macros new_SV() and del_SV() grab and free
182 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
183 to return the SV to the free list with error checking.) new_SV() calls
184 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
185 SVs in the free list have their SvTYPE field set to all ones.
187 At the time of very final cleanup, sv_free_arenas() is called from
188 perl_destruct() to physically free all the arenas allocated since the
189 start of the interpreter.
191 The function visit() scans the SV arenas list, and calls a specified
192 function for each SV it finds which is still live - ie which has an SvTYPE
193 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
194 following functions (specified as [function that calls visit()] / [function
195 called by visit() for each SV]):
197 sv_report_used() / do_report_used()
198 dump all remaining SVs (debugging aid)
200 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
201 do_clean_named_io_objs(),do_curse()
202 Attempt to free all objects pointed to by RVs,
203 try to do the same for all objects indir-
204 ectly referenced by typeglobs too, and
205 then do a final sweep, cursing any
206 objects that remain. Called once from
207 perl_destruct(), prior to calling sv_clean_all()
210 sv_clean_all() / do_clean_all()
211 SvREFCNT_dec(sv) each remaining SV, possibly
212 triggering an sv_free(). It also sets the
213 SVf_BREAK flag on the SV to indicate that the
214 refcnt has been artificially lowered, and thus
215 stopping sv_free() from giving spurious warnings
216 about SVs which unexpectedly have a refcnt
217 of zero. called repeatedly from perl_destruct()
218 until there are no SVs left.
220 =head2 Arena allocator API Summary
222 Private API to rest of sv.c
226 new_XPVNV(), del_XPVGV(),
231 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
235 * ========================================================================= */
238 * "A time to plant, and a time to uproot what was planted..."
242 # define MEM_LOG_NEW_SV(sv, file, line, func) \
243 Perl_mem_log_new_sv(sv, file, line, func)
244 # define MEM_LOG_DEL_SV(sv, file, line, func) \
245 Perl_mem_log_del_sv(sv, file, line, func)
247 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
248 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
251 #ifdef DEBUG_LEAKING_SCALARS
252 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
253 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
255 # define DEBUG_SV_SERIAL(sv) \
256 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
257 PTR2UV(sv), (long)(sv)->sv_debug_serial))
259 # define FREE_SV_DEBUG_FILE(sv)
260 # define DEBUG_SV_SERIAL(sv) NOOP
264 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
265 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
266 /* Whilst I'd love to do this, it seems that things like to check on
268 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
270 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
271 PoisonNew(&SvREFCNT(sv), 1, U32)
273 # define SvARENA_CHAIN(sv) SvANY(sv)
274 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
275 # define POSION_SV_HEAD(sv)
278 /* Mark an SV head as unused, and add to free list.
280 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
281 * its refcount artificially decremented during global destruction, so
282 * there may be dangling pointers to it. The last thing we want in that
283 * case is for it to be reused. */
285 #define plant_SV(p) \
287 const U32 old_flags = SvFLAGS(p); \
288 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
289 DEBUG_SV_SERIAL(p); \
290 FREE_SV_DEBUG_FILE(p); \
292 SvFLAGS(p) = SVTYPEMASK; \
293 if (!(old_flags & SVf_BREAK)) { \
294 SvARENA_CHAIN_SET(p, PL_sv_root); \
300 #define uproot_SV(p) \
303 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
308 /* make some more SVs by adding another arena */
314 char *chunk; /* must use New here to match call to */
315 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
316 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
321 /* new_SV(): return a new, empty SV head */
323 #ifdef DEBUG_LEAKING_SCALARS
324 /* provide a real function for a debugger to play with */
326 S_new_SV(pTHX_ const char *file, int line, const char *func)
333 sv = S_more_sv(aTHX);
337 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
338 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
344 sv->sv_debug_inpad = 0;
345 sv->sv_debug_parent = NULL;
346 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
348 sv->sv_debug_serial = PL_sv_serial++;
350 MEM_LOG_NEW_SV(sv, file, line, func);
351 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
352 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
356 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
364 (p) = S_more_sv(aTHX); \
368 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
373 /* del_SV(): return an empty SV head to the free list */
386 S_del_sv(pTHX_ SV *p)
390 PERL_ARGS_ASSERT_DEL_SV;
395 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
396 const SV * const sv = sva + 1;
397 const SV * const svend = &sva[SvREFCNT(sva)];
398 if (p >= sv && p < svend) {
404 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
405 "Attempt to free non-arena SV: 0x%"UVxf
406 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
413 #else /* ! DEBUGGING */
415 #define del_SV(p) plant_SV(p)
417 #endif /* DEBUGGING */
421 =head1 SV Manipulation Functions
423 =for apidoc sv_add_arena
425 Given a chunk of memory, link it to the head of the list of arenas,
426 and split it into a list of free SVs.
432 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
434 SV *const sva = MUTABLE_SV(ptr);
438 PERL_ARGS_ASSERT_SV_ADD_ARENA;
440 /* The first SV in an arena isn't an SV. */
441 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
442 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
443 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
445 PL_sv_arenaroot = sva;
446 PL_sv_root = sva + 1;
448 svend = &sva[SvREFCNT(sva) - 1];
451 SvARENA_CHAIN_SET(sv, (sv + 1));
455 /* Must always set typemask because it's always checked in on cleanup
456 when the arenas are walked looking for objects. */
457 SvFLAGS(sv) = SVTYPEMASK;
460 SvARENA_CHAIN_SET(sv, 0);
464 SvFLAGS(sv) = SVTYPEMASK;
467 /* visit(): call the named function for each non-free SV in the arenas
468 * whose flags field matches the flags/mask args. */
471 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
476 PERL_ARGS_ASSERT_VISIT;
478 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
479 const SV * const svend = &sva[SvREFCNT(sva)];
481 for (sv = sva + 1; sv < svend; ++sv) {
482 if (SvTYPE(sv) != (svtype)SVTYPEMASK
483 && (sv->sv_flags & mask) == flags
496 /* called by sv_report_used() for each live SV */
499 do_report_used(pTHX_ SV *const sv)
501 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
502 PerlIO_printf(Perl_debug_log, "****\n");
509 =for apidoc sv_report_used
511 Dump the contents of all SVs not yet freed (debugging aid).
517 Perl_sv_report_used(pTHX)
520 visit(do_report_used, 0, 0);
526 /* called by sv_clean_objs() for each live SV */
529 do_clean_objs(pTHX_ SV *const ref)
533 SV * const target = SvRV(ref);
534 if (SvOBJECT(target)) {
535 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
536 if (SvWEAKREF(ref)) {
537 sv_del_backref(target, ref);
543 SvREFCNT_dec_NN(target);
550 /* clear any slots in a GV which hold objects - except IO;
551 * called by sv_clean_objs() for each live GV */
554 do_clean_named_objs(pTHX_ SV *const sv)
557 assert(SvTYPE(sv) == SVt_PVGV);
558 assert(isGV_with_GP(sv));
562 /* freeing GP entries may indirectly free the current GV;
563 * hold onto it while we mess with the GP slots */
566 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
567 DEBUG_D((PerlIO_printf(Perl_debug_log,
568 "Cleaning named glob SV object:\n "), sv_dump(obj)));
570 SvREFCNT_dec_NN(obj);
572 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
573 DEBUG_D((PerlIO_printf(Perl_debug_log,
574 "Cleaning named glob AV object:\n "), sv_dump(obj)));
576 SvREFCNT_dec_NN(obj);
578 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
579 DEBUG_D((PerlIO_printf(Perl_debug_log,
580 "Cleaning named glob HV object:\n "), sv_dump(obj)));
582 SvREFCNT_dec_NN(obj);
584 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
585 DEBUG_D((PerlIO_printf(Perl_debug_log,
586 "Cleaning named glob CV object:\n "), sv_dump(obj)));
588 SvREFCNT_dec_NN(obj);
590 SvREFCNT_dec_NN(sv); /* undo the inc above */
593 /* clear any IO slots in a GV which hold objects (except stderr, defout);
594 * called by sv_clean_objs() for each live GV */
597 do_clean_named_io_objs(pTHX_ SV *const sv)
600 assert(SvTYPE(sv) == SVt_PVGV);
601 assert(isGV_with_GP(sv));
602 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
606 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
607 DEBUG_D((PerlIO_printf(Perl_debug_log,
608 "Cleaning named glob IO object:\n "), sv_dump(obj)));
610 SvREFCNT_dec_NN(obj);
612 SvREFCNT_dec_NN(sv); /* undo the inc above */
615 /* Void wrapper to pass to visit() */
617 do_curse(pTHX_ SV * const sv) {
618 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
619 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
625 =for apidoc sv_clean_objs
627 Attempt to destroy all objects not yet freed.
633 Perl_sv_clean_objs(pTHX)
636 PL_in_clean_objs = TRUE;
637 visit(do_clean_objs, SVf_ROK, SVf_ROK);
638 /* Some barnacles may yet remain, clinging to typeglobs.
639 * Run the non-IO destructors first: they may want to output
640 * error messages, close files etc */
641 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
642 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
643 /* And if there are some very tenacious barnacles clinging to arrays,
644 closures, or what have you.... */
645 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
646 olddef = PL_defoutgv;
647 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
648 if (olddef && isGV_with_GP(olddef))
649 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
650 olderr = PL_stderrgv;
651 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
652 if (olderr && isGV_with_GP(olderr))
653 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
654 SvREFCNT_dec(olddef);
655 PL_in_clean_objs = FALSE;
658 /* called by sv_clean_all() for each live SV */
661 do_clean_all(pTHX_ SV *const sv)
663 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
664 /* don't clean pid table and strtab */
667 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
668 SvFLAGS(sv) |= SVf_BREAK;
673 =for apidoc sv_clean_all
675 Decrement the refcnt of each remaining SV, possibly triggering a
676 cleanup. This function may have to be called multiple times to free
677 SVs which are in complex self-referential hierarchies.
683 Perl_sv_clean_all(pTHX)
686 PL_in_clean_all = TRUE;
687 cleaned = visit(do_clean_all, 0,0);
692 ARENASETS: a meta-arena implementation which separates arena-info
693 into struct arena_set, which contains an array of struct
694 arena_descs, each holding info for a single arena. By separating
695 the meta-info from the arena, we recover the 1st slot, formerly
696 borrowed for list management. The arena_set is about the size of an
697 arena, avoiding the needless malloc overhead of a naive linked-list.
699 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
700 memory in the last arena-set (1/2 on average). In trade, we get
701 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
702 smaller types). The recovery of the wasted space allows use of
703 small arenas for large, rare body types, by changing array* fields
704 in body_details_by_type[] below.
707 char *arena; /* the raw storage, allocated aligned */
708 size_t size; /* its size ~4k typ */
709 svtype utype; /* bodytype stored in arena */
714 /* Get the maximum number of elements in set[] such that struct arena_set
715 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
716 therefore likely to be 1 aligned memory page. */
718 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
719 - 2 * sizeof(int)) / sizeof (struct arena_desc))
722 struct arena_set* next;
723 unsigned int set_size; /* ie ARENAS_PER_SET */
724 unsigned int curr; /* index of next available arena-desc */
725 struct arena_desc set[ARENAS_PER_SET];
729 =for apidoc sv_free_arenas
731 Deallocate the memory used by all arenas. Note that all the individual SV
732 heads and bodies within the arenas must already have been freed.
738 Perl_sv_free_arenas(pTHX)
744 /* Free arenas here, but be careful about fake ones. (We assume
745 contiguity of the fake ones with the corresponding real ones.) */
747 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
748 svanext = MUTABLE_SV(SvANY(sva));
749 while (svanext && SvFAKE(svanext))
750 svanext = MUTABLE_SV(SvANY(svanext));
757 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
760 struct arena_set *current = aroot;
763 assert(aroot->set[i].arena);
764 Safefree(aroot->set[i].arena);
772 i = PERL_ARENA_ROOTS_SIZE;
774 PL_body_roots[i] = 0;
781 Here are mid-level routines that manage the allocation of bodies out
782 of the various arenas. There are 5 kinds of arenas:
784 1. SV-head arenas, which are discussed and handled above
785 2. regular body arenas
786 3. arenas for reduced-size bodies
789 Arena types 2 & 3 are chained by body-type off an array of
790 arena-root pointers, which is indexed by svtype. Some of the
791 larger/less used body types are malloced singly, since a large
792 unused block of them is wasteful. Also, several svtypes dont have
793 bodies; the data fits into the sv-head itself. The arena-root
794 pointer thus has a few unused root-pointers (which may be hijacked
795 later for arena types 4,5)
797 3 differs from 2 as an optimization; some body types have several
798 unused fields in the front of the structure (which are kept in-place
799 for consistency). These bodies can be allocated in smaller chunks,
800 because the leading fields arent accessed. Pointers to such bodies
801 are decremented to point at the unused 'ghost' memory, knowing that
802 the pointers are used with offsets to the real memory.
805 =head1 SV-Body Allocation
809 Allocation of SV-bodies is similar to SV-heads, differing as follows;
810 the allocation mechanism is used for many body types, so is somewhat
811 more complicated, it uses arena-sets, and has no need for still-live
814 At the outermost level, (new|del)_X*V macros return bodies of the
815 appropriate type. These macros call either (new|del)_body_type or
816 (new|del)_body_allocated macro pairs, depending on specifics of the
817 type. Most body types use the former pair, the latter pair is used to
818 allocate body types with "ghost fields".
820 "ghost fields" are fields that are unused in certain types, and
821 consequently don't need to actually exist. They are declared because
822 they're part of a "base type", which allows use of functions as
823 methods. The simplest examples are AVs and HVs, 2 aggregate types
824 which don't use the fields which support SCALAR semantics.
826 For these types, the arenas are carved up into appropriately sized
827 chunks, we thus avoid wasted memory for those unaccessed members.
828 When bodies are allocated, we adjust the pointer back in memory by the
829 size of the part not allocated, so it's as if we allocated the full
830 structure. (But things will all go boom if you write to the part that
831 is "not there", because you'll be overwriting the last members of the
832 preceding structure in memory.)
834 We calculate the correction using the STRUCT_OFFSET macro on the first
835 member present. If the allocated structure is smaller (no initial NV
836 actually allocated) then the net effect is to subtract the size of the NV
837 from the pointer, to return a new pointer as if an initial NV were actually
838 allocated. (We were using structures named *_allocated for this, but
839 this turned out to be a subtle bug, because a structure without an NV
840 could have a lower alignment constraint, but the compiler is allowed to
841 optimised accesses based on the alignment constraint of the actual pointer
842 to the full structure, for example, using a single 64 bit load instruction
843 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
845 This is the same trick as was used for NV and IV bodies. Ironically it
846 doesn't need to be used for NV bodies any more, because NV is now at
847 the start of the structure. IV bodies don't need it either, because
848 they are no longer allocated.
850 In turn, the new_body_* allocators call S_new_body(), which invokes
851 new_body_inline macro, which takes a lock, and takes a body off the
852 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
853 necessary to refresh an empty list. Then the lock is released, and
854 the body is returned.
856 Perl_more_bodies allocates a new arena, and carves it up into an array of N
857 bodies, which it strings into a linked list. It looks up arena-size
858 and body-size from the body_details table described below, thus
859 supporting the multiple body-types.
861 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
862 the (new|del)_X*V macros are mapped directly to malloc/free.
864 For each sv-type, struct body_details bodies_by_type[] carries
865 parameters which control these aspects of SV handling:
867 Arena_size determines whether arenas are used for this body type, and if
868 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
869 zero, forcing individual mallocs and frees.
871 Body_size determines how big a body is, and therefore how many fit into
872 each arena. Offset carries the body-pointer adjustment needed for
873 "ghost fields", and is used in *_allocated macros.
875 But its main purpose is to parameterize info needed in
876 Perl_sv_upgrade(). The info here dramatically simplifies the function
877 vs the implementation in 5.8.8, making it table-driven. All fields
878 are used for this, except for arena_size.
880 For the sv-types that have no bodies, arenas are not used, so those
881 PL_body_roots[sv_type] are unused, and can be overloaded. In
882 something of a special case, SVt_NULL is borrowed for HE arenas;
883 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
884 bodies_by_type[SVt_NULL] slot is not used, as the table is not
889 struct body_details {
890 U8 body_size; /* Size to allocate */
891 U8 copy; /* Size of structure to copy (may be shorter) */
893 unsigned int type : 4; /* We have space for a sanity check. */
894 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
895 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
896 unsigned int arena : 1; /* Allocated from an arena */
897 size_t arena_size; /* Size of arena to allocate */
905 /* With -DPURFIY we allocate everything directly, and don't use arenas.
906 This seems a rather elegant way to simplify some of the code below. */
907 #define HASARENA FALSE
909 #define HASARENA TRUE
911 #define NOARENA FALSE
913 /* Size the arenas to exactly fit a given number of bodies. A count
914 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
915 simplifying the default. If count > 0, the arena is sized to fit
916 only that many bodies, allowing arenas to be used for large, rare
917 bodies (XPVFM, XPVIO) without undue waste. The arena size is
918 limited by PERL_ARENA_SIZE, so we can safely oversize the
921 #define FIT_ARENA0(body_size) \
922 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
923 #define FIT_ARENAn(count,body_size) \
924 ( count * body_size <= PERL_ARENA_SIZE) \
925 ? count * body_size \
926 : FIT_ARENA0 (body_size)
927 #define FIT_ARENA(count,body_size) \
929 ? FIT_ARENAn (count, body_size) \
930 : FIT_ARENA0 (body_size)
932 /* Calculate the length to copy. Specifically work out the length less any
933 final padding the compiler needed to add. See the comment in sv_upgrade
934 for why copying the padding proved to be a bug. */
936 #define copy_length(type, last_member) \
937 STRUCT_OFFSET(type, last_member) \
938 + sizeof (((type*)SvANY((const SV *)0))->last_member)
940 static const struct body_details bodies_by_type[] = {
941 /* HEs use this offset for their arena. */
942 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
944 /* IVs are in the head, so the allocation size is 0. */
946 sizeof(IV), /* This is used to copy out the IV body. */
947 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
948 NOARENA /* IVS don't need an arena */, 0
951 { sizeof(NV), sizeof(NV),
952 STRUCT_OFFSET(XPVNV, xnv_u),
953 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
955 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
956 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
957 + STRUCT_OFFSET(XPV, xpv_cur),
958 SVt_PV, FALSE, NONV, HASARENA,
959 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
961 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
962 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
963 + STRUCT_OFFSET(XPV, xpv_cur),
964 SVt_INVLIST, TRUE, NONV, HASARENA,
965 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
967 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
968 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
969 + STRUCT_OFFSET(XPV, xpv_cur),
970 SVt_PVIV, FALSE, NONV, HASARENA,
971 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
973 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
974 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
975 + STRUCT_OFFSET(XPV, xpv_cur),
976 SVt_PVNV, FALSE, HADNV, HASARENA,
977 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
979 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
980 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
985 SVt_REGEXP, TRUE, NONV, HASARENA,
986 FIT_ARENA(0, sizeof(regexp))
989 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
990 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
992 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
993 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
996 copy_length(XPVAV, xav_alloc),
998 SVt_PVAV, TRUE, NONV, HASARENA,
999 FIT_ARENA(0, sizeof(XPVAV)) },
1002 copy_length(XPVHV, xhv_max),
1004 SVt_PVHV, TRUE, NONV, HASARENA,
1005 FIT_ARENA(0, sizeof(XPVHV)) },
1010 SVt_PVCV, TRUE, NONV, HASARENA,
1011 FIT_ARENA(0, sizeof(XPVCV)) },
1016 SVt_PVFM, TRUE, NONV, NOARENA,
1017 FIT_ARENA(20, sizeof(XPVFM)) },
1022 SVt_PVIO, TRUE, NONV, HASARENA,
1023 FIT_ARENA(24, sizeof(XPVIO)) },
1026 #define new_body_allocated(sv_type) \
1027 (void *)((char *)S_new_body(aTHX_ sv_type) \
1028 - bodies_by_type[sv_type].offset)
1030 /* return a thing to the free list */
1032 #define del_body(thing, root) \
1034 void ** const thing_copy = (void **)thing; \
1035 *thing_copy = *root; \
1036 *root = (void*)thing_copy; \
1041 #define new_XNV() safemalloc(sizeof(XPVNV))
1042 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1043 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1045 #define del_XPVGV(p) safefree(p)
1049 #define new_XNV() new_body_allocated(SVt_NV)
1050 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1051 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1053 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1054 &PL_body_roots[SVt_PVGV])
1058 /* no arena for you! */
1060 #define new_NOARENA(details) \
1061 safemalloc((details)->body_size + (details)->offset)
1062 #define new_NOARENAZ(details) \
1063 safecalloc((details)->body_size + (details)->offset, 1)
1066 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1067 const size_t arena_size)
1069 void ** const root = &PL_body_roots[sv_type];
1070 struct arena_desc *adesc;
1071 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1075 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1076 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1079 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1080 static bool done_sanity_check;
1082 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1083 * variables like done_sanity_check. */
1084 if (!done_sanity_check) {
1085 unsigned int i = SVt_LAST;
1087 done_sanity_check = TRUE;
1090 assert (bodies_by_type[i].type == i);
1096 /* may need new arena-set to hold new arena */
1097 if (!aroot || aroot->curr >= aroot->set_size) {
1098 struct arena_set *newroot;
1099 Newxz(newroot, 1, struct arena_set);
1100 newroot->set_size = ARENAS_PER_SET;
1101 newroot->next = aroot;
1103 PL_body_arenas = (void *) newroot;
1104 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1107 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1108 curr = aroot->curr++;
1109 adesc = &(aroot->set[curr]);
1110 assert(!adesc->arena);
1112 Newx(adesc->arena, good_arena_size, char);
1113 adesc->size = good_arena_size;
1114 adesc->utype = sv_type;
1115 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1116 curr, (void*)adesc->arena, (UV)good_arena_size));
1118 start = (char *) adesc->arena;
1120 /* Get the address of the byte after the end of the last body we can fit.
1121 Remember, this is integer division: */
1122 end = start + good_arena_size / body_size * body_size;
1124 /* computed count doesn't reflect the 1st slot reservation */
1125 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1126 DEBUG_m(PerlIO_printf(Perl_debug_log,
1127 "arena %p end %p arena-size %d (from %d) type %d "
1129 (void*)start, (void*)end, (int)good_arena_size,
1130 (int)arena_size, sv_type, (int)body_size,
1131 (int)good_arena_size / (int)body_size));
1133 DEBUG_m(PerlIO_printf(Perl_debug_log,
1134 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1135 (void*)start, (void*)end,
1136 (int)arena_size, sv_type, (int)body_size,
1137 (int)good_arena_size / (int)body_size));
1139 *root = (void *)start;
1142 /* Where the next body would start: */
1143 char * const next = start + body_size;
1146 /* This is the last body: */
1147 assert(next == end);
1149 *(void **)start = 0;
1153 *(void**) start = (void *)next;
1158 /* grab a new thing from the free list, allocating more if necessary.
1159 The inline version is used for speed in hot routines, and the
1160 function using it serves the rest (unless PURIFY).
1162 #define new_body_inline(xpv, sv_type) \
1164 void ** const r3wt = &PL_body_roots[sv_type]; \
1165 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1166 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1167 bodies_by_type[sv_type].body_size,\
1168 bodies_by_type[sv_type].arena_size)); \
1169 *(r3wt) = *(void**)(xpv); \
1175 S_new_body(pTHX_ const svtype sv_type)
1178 new_body_inline(xpv, sv_type);
1184 static const struct body_details fake_rv =
1185 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1188 =for apidoc sv_upgrade
1190 Upgrade an SV to a more complex form. Generally adds a new body type to the
1191 SV, then copies across as much information as possible from the old body.
1192 It croaks if the SV is already in a more complex form than requested. You
1193 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1194 before calling C<sv_upgrade>, and hence does not croak. See also
1201 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1205 const svtype old_type = SvTYPE(sv);
1206 const struct body_details *new_type_details;
1207 const struct body_details *old_type_details
1208 = bodies_by_type + old_type;
1209 SV *referant = NULL;
1211 PERL_ARGS_ASSERT_SV_UPGRADE;
1213 if (old_type == new_type)
1216 /* This clause was purposefully added ahead of the early return above to
1217 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1218 inference by Nick I-S that it would fix other troublesome cases. See
1219 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1221 Given that shared hash key scalars are no longer PVIV, but PV, there is
1222 no longer need to unshare so as to free up the IVX slot for its proper
1223 purpose. So it's safe to move the early return earlier. */
1225 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1226 sv_force_normal_flags(sv, 0);
1229 old_body = SvANY(sv);
1231 /* Copying structures onto other structures that have been neatly zeroed
1232 has a subtle gotcha. Consider XPVMG
1234 +------+------+------+------+------+-------+-------+
1235 | NV | CUR | LEN | IV | MAGIC | STASH |
1236 +------+------+------+------+------+-------+-------+
1237 0 4 8 12 16 20 24 28
1239 where NVs are aligned to 8 bytes, so that sizeof that structure is
1240 actually 32 bytes long, with 4 bytes of padding at the end:
1242 +------+------+------+------+------+-------+-------+------+
1243 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1244 +------+------+------+------+------+-------+-------+------+
1245 0 4 8 12 16 20 24 28 32
1247 so what happens if you allocate memory for this structure:
1249 +------+------+------+------+------+-------+-------+------+------+...
1250 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1251 +------+------+------+------+------+-------+-------+------+------+...
1252 0 4 8 12 16 20 24 28 32 36
1254 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1255 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1256 started out as zero once, but it's quite possible that it isn't. So now,
1257 rather than a nicely zeroed GP, you have it pointing somewhere random.
1260 (In fact, GP ends up pointing at a previous GP structure, because the
1261 principle cause of the padding in XPVMG getting garbage is a copy of
1262 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1263 this happens to be moot because XPVGV has been re-ordered, with GP
1264 no longer after STASH)
1266 So we are careful and work out the size of used parts of all the
1274 referant = SvRV(sv);
1275 old_type_details = &fake_rv;
1276 if (new_type == SVt_NV)
1277 new_type = SVt_PVNV;
1279 if (new_type < SVt_PVIV) {
1280 new_type = (new_type == SVt_NV)
1281 ? SVt_PVNV : SVt_PVIV;
1286 if (new_type < SVt_PVNV) {
1287 new_type = SVt_PVNV;
1291 assert(new_type > SVt_PV);
1292 assert(SVt_IV < SVt_PV);
1293 assert(SVt_NV < SVt_PV);
1300 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1301 there's no way that it can be safely upgraded, because perl.c
1302 expects to Safefree(SvANY(PL_mess_sv)) */
1303 assert(sv != PL_mess_sv);
1304 /* This flag bit is used to mean other things in other scalar types.
1305 Given that it only has meaning inside the pad, it shouldn't be set
1306 on anything that can get upgraded. */
1307 assert(!SvPAD_TYPED(sv));
1310 if (UNLIKELY(old_type_details->cant_upgrade))
1311 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1312 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1315 if (UNLIKELY(old_type > new_type))
1316 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1317 (int)old_type, (int)new_type);
1319 new_type_details = bodies_by_type + new_type;
1321 SvFLAGS(sv) &= ~SVTYPEMASK;
1322 SvFLAGS(sv) |= new_type;
1324 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1325 the return statements above will have triggered. */
1326 assert (new_type != SVt_NULL);
1329 assert(old_type == SVt_NULL);
1330 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1334 assert(old_type == SVt_NULL);
1335 SvANY(sv) = new_XNV();
1340 assert(new_type_details->body_size);
1343 assert(new_type_details->arena);
1344 assert(new_type_details->arena_size);
1345 /* This points to the start of the allocated area. */
1346 new_body_inline(new_body, new_type);
1347 Zero(new_body, new_type_details->body_size, char);
1348 new_body = ((char *)new_body) - new_type_details->offset;
1350 /* We always allocated the full length item with PURIFY. To do this
1351 we fake things so that arena is false for all 16 types.. */
1352 new_body = new_NOARENAZ(new_type_details);
1354 SvANY(sv) = new_body;
1355 if (new_type == SVt_PVAV) {
1359 if (old_type_details->body_size) {
1362 /* It will have been zeroed when the new body was allocated.
1363 Lets not write to it, in case it confuses a write-back
1369 #ifndef NODEFAULT_SHAREKEYS
1370 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1372 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1373 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1376 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1377 The target created by newSVrv also is, and it can have magic.
1378 However, it never has SvPVX set.
1380 if (old_type == SVt_IV) {
1382 } else if (old_type >= SVt_PV) {
1383 assert(SvPVX_const(sv) == 0);
1386 if (old_type >= SVt_PVMG) {
1387 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1388 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1390 sv->sv_u.svu_array = NULL; /* or svu_hash */
1395 /* XXX Is this still needed? Was it ever needed? Surely as there is
1396 no route from NV to PVIV, NOK can never be true */
1397 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 (old_type > SVt_IV) {
1482 /* Note that there is an assumption that all bodies of types that
1483 can be upgraded came from arenas. Only the more complex non-
1484 upgradable types are allowed to be directly malloc()ed. */
1485 assert(old_type_details->arena);
1486 del_body((void*)((char*)old_body + old_type_details->offset),
1487 &PL_body_roots[old_type]);
1493 =for apidoc sv_backoff
1495 Remove any string offset. You should normally use the C<SvOOK_off> macro
1502 Perl_sv_backoff(SV *const sv)
1505 const char * const s = SvPVX_const(sv);
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
1513 SvOOK_offset(sv, delta);
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
1517 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1518 SvFLAGS(sv) &= ~SVf_OOK;
1525 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1526 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1527 Use the C<SvGROW> wrapper instead.
1532 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1535 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1539 PERL_ARGS_ASSERT_SV_GROW;
1543 if (SvTYPE(sv) < SVt_PV) {
1544 sv_upgrade(sv, SVt_PV);
1545 s = SvPVX_mutable(sv);
1547 else if (SvOOK(sv)) { /* pv is offset? */
1549 s = SvPVX_mutable(sv);
1550 if (newlen > SvLEN(sv))
1551 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1555 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1556 s = SvPVX_mutable(sv);
1559 #ifdef PERL_NEW_COPY_ON_WRITE
1560 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1561 * to store the COW count. So in general, allocate one more byte than
1562 * asked for, to make it likely this byte is always spare: and thus
1563 * make more strings COW-able.
1564 * If the new size is a big power of two, don't bother: we assume the
1565 * caller wanted a nice 2^N sized block and will be annoyed at getting
1571 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1572 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1575 if (newlen > SvLEN(sv)) { /* need more room? */
1576 STRLEN minlen = SvCUR(sv);
1577 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1578 if (newlen < minlen)
1580 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1582 /* Don't round up on the first allocation, as odds are pretty good that
1583 * the initial request is accurate as to what is really needed */
1585 newlen = PERL_STRLEN_ROUNDUP(newlen);
1588 if (SvLEN(sv) && s) {
1589 s = (char*)saferealloc(s, newlen);
1592 s = (char*)safemalloc(newlen);
1593 if (SvPVX_const(sv) && SvCUR(sv)) {
1594 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1598 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1599 /* Do this here, do it once, do it right, and then we will never get
1600 called back into sv_grow() unless there really is some growing
1602 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1604 SvLEN_set(sv, newlen);
1611 =for apidoc sv_setiv
1613 Copies an integer into the given SV, upgrading first if necessary.
1614 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1620 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1622 PERL_ARGS_ASSERT_SV_SETIV;
1624 SV_CHECK_THINKFIRST_COW_DROP(sv);
1625 switch (SvTYPE(sv)) {
1628 sv_upgrade(sv, SVt_IV);
1631 sv_upgrade(sv, SVt_PVIV);
1635 if (!isGV_with_GP(sv))
1642 /* diag_listed_as: Can't coerce %s to %s in %s */
1643 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1647 (void)SvIOK_only(sv); /* validate number */
1653 =for apidoc sv_setiv_mg
1655 Like C<sv_setiv>, but also handles 'set' magic.
1661 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1663 PERL_ARGS_ASSERT_SV_SETIV_MG;
1670 =for apidoc sv_setuv
1672 Copies an unsigned integer into the given SV, upgrading first if necessary.
1673 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1679 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1681 PERL_ARGS_ASSERT_SV_SETUV;
1683 /* With the if statement to ensure that integers are stored as IVs whenever
1685 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1688 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1690 If you wish to remove the following if statement, so that this routine
1691 (and its callers) always return UVs, please benchmark to see what the
1692 effect is. Modern CPUs may be different. Or may not :-)
1694 if (u <= (UV)IV_MAX) {
1695 sv_setiv(sv, (IV)u);
1704 =for apidoc sv_setuv_mg
1706 Like C<sv_setuv>, but also handles 'set' magic.
1712 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1714 PERL_ARGS_ASSERT_SV_SETUV_MG;
1721 =for apidoc sv_setnv
1723 Copies a double into the given SV, upgrading first if necessary.
1724 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1730 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1732 PERL_ARGS_ASSERT_SV_SETNV;
1734 SV_CHECK_THINKFIRST_COW_DROP(sv);
1735 switch (SvTYPE(sv)) {
1738 sv_upgrade(sv, SVt_NV);
1742 sv_upgrade(sv, SVt_PVNV);
1746 if (!isGV_with_GP(sv))
1753 /* diag_listed_as: Can't coerce %s to %s in %s */
1754 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1759 (void)SvNOK_only(sv); /* validate number */
1764 =for apidoc sv_setnv_mg
1766 Like C<sv_setnv>, but also handles 'set' magic.
1772 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1774 PERL_ARGS_ASSERT_SV_SETNV_MG;
1780 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1781 * not incrementable warning display.
1782 * Originally part of S_not_a_number().
1783 * The return value may be != tmpbuf.
1787 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1790 PERL_ARGS_ASSERT_SV_DISPLAY;
1793 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1794 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1797 const char * const limit = tmpbuf + tmpbuf_size - 8;
1798 /* each *s can expand to 4 chars + "...\0",
1799 i.e. need room for 8 chars */
1801 const char *s = SvPVX_const(sv);
1802 const char * const end = s + SvCUR(sv);
1803 for ( ; s < end && d < limit; s++ ) {
1805 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1809 /* Map to ASCII "equivalent" of Latin1 */
1810 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1816 else if (ch == '\r') {
1820 else if (ch == '\f') {
1824 else if (ch == '\\') {
1828 else if (ch == '\0') {
1832 else if (isPRINT_LC(ch))
1851 /* Print an "isn't numeric" warning, using a cleaned-up,
1852 * printable version of the offending string
1856 S_not_a_number(pTHX_ SV *const sv)
1862 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1864 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1867 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1868 /* diag_listed_as: Argument "%s" isn't numeric%s */
1869 "Argument \"%s\" isn't numeric in %s", pv,
1872 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1873 /* diag_listed_as: Argument "%s" isn't numeric%s */
1874 "Argument \"%s\" isn't numeric", pv);
1878 S_not_incrementable(pTHX_ SV *const sv) {
1883 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1885 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1887 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1888 "Argument \"%s\" treated as 0 in increment (++)", pv);
1892 =for apidoc looks_like_number
1894 Test if the content of an SV looks like a number (or is a number).
1895 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1896 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1903 Perl_looks_like_number(pTHX_ SV *const sv)
1908 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1910 if (SvPOK(sv) || SvPOKp(sv)) {
1911 sbegin = SvPV_nomg_const(sv, len);
1914 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1915 return grok_number(sbegin, len, NULL);
1919 S_glob_2number(pTHX_ GV * const gv)
1921 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1923 /* We know that all GVs stringify to something that is not-a-number,
1924 so no need to test that. */
1925 if (ckWARN(WARN_NUMERIC))
1927 SV *const buffer = sv_newmortal();
1928 gv_efullname3(buffer, gv, "*");
1929 not_a_number(buffer);
1931 /* We just want something true to return, so that S_sv_2iuv_common
1932 can tail call us and return true. */
1936 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1937 until proven guilty, assume that things are not that bad... */
1942 As 64 bit platforms often have an NV that doesn't preserve all bits of
1943 an IV (an assumption perl has been based on to date) it becomes necessary
1944 to remove the assumption that the NV always carries enough precision to
1945 recreate the IV whenever needed, and that the NV is the canonical form.
1946 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1947 precision as a side effect of conversion (which would lead to insanity
1948 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1949 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1950 where precision was lost, and IV/UV/NV slots that have a valid conversion
1951 which has lost no precision
1952 2) to ensure that if a numeric conversion to one form is requested that
1953 would lose precision, the precise conversion (or differently
1954 imprecise conversion) is also performed and cached, to prevent
1955 requests for different numeric formats on the same SV causing
1956 lossy conversion chains. (lossless conversion chains are perfectly
1961 SvIOKp is true if the IV slot contains a valid value
1962 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1963 SvNOKp is true if the NV slot contains a valid value
1964 SvNOK is true only if the NV value is accurate
1967 while converting from PV to NV, check to see if converting that NV to an
1968 IV(or UV) would lose accuracy over a direct conversion from PV to
1969 IV(or UV). If it would, cache both conversions, return NV, but mark
1970 SV as IOK NOKp (ie not NOK).
1972 While converting from PV to IV, check to see if converting that IV to an
1973 NV would lose accuracy over a direct conversion from PV to NV. If it
1974 would, cache both conversions, flag similarly.
1976 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1977 correctly because if IV & NV were set NV *always* overruled.
1978 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1979 changes - now IV and NV together means that the two are interchangeable:
1980 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1982 The benefit of this is that operations such as pp_add know that if
1983 SvIOK is true for both left and right operands, then integer addition
1984 can be used instead of floating point (for cases where the result won't
1985 overflow). Before, floating point was always used, which could lead to
1986 loss of precision compared with integer addition.
1988 * making IV and NV equal status should make maths accurate on 64 bit
1990 * may speed up maths somewhat if pp_add and friends start to use
1991 integers when possible instead of fp. (Hopefully the overhead in
1992 looking for SvIOK and checking for overflow will not outweigh the
1993 fp to integer speedup)
1994 * will slow down integer operations (callers of SvIV) on "inaccurate"
1995 values, as the change from SvIOK to SvIOKp will cause a call into
1996 sv_2iv each time rather than a macro access direct to the IV slot
1997 * should speed up number->string conversion on integers as IV is
1998 favoured when IV and NV are equally accurate
2000 ####################################################################
2001 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2002 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2003 On the other hand, SvUOK is true iff UV.
2004 ####################################################################
2006 Your mileage will vary depending your CPU's relative fp to integer
2010 #ifndef NV_PRESERVES_UV
2011 # define IS_NUMBER_UNDERFLOW_IV 1
2012 # define IS_NUMBER_UNDERFLOW_UV 2
2013 # define IS_NUMBER_IV_AND_UV 2
2014 # define IS_NUMBER_OVERFLOW_IV 4
2015 # define IS_NUMBER_OVERFLOW_UV 5
2017 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2019 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2021 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2027 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2028 PERL_UNUSED_CONTEXT;
2030 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));
2031 if (SvNVX(sv) < (NV)IV_MIN) {
2032 (void)SvIOKp_on(sv);
2034 SvIV_set(sv, IV_MIN);
2035 return IS_NUMBER_UNDERFLOW_IV;
2037 if (SvNVX(sv) > (NV)UV_MAX) {
2038 (void)SvIOKp_on(sv);
2041 SvUV_set(sv, UV_MAX);
2042 return IS_NUMBER_OVERFLOW_UV;
2044 (void)SvIOKp_on(sv);
2046 /* Can't use strtol etc to convert this string. (See truth table in
2048 if (SvNVX(sv) <= (UV)IV_MAX) {
2049 SvIV_set(sv, I_V(SvNVX(sv)));
2050 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2051 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2053 /* Integer is imprecise. NOK, IOKp */
2055 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2058 SvUV_set(sv, U_V(SvNVX(sv)));
2059 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2060 if (SvUVX(sv) == UV_MAX) {
2061 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2062 possibly be preserved by NV. Hence, it must be overflow.
2064 return IS_NUMBER_OVERFLOW_UV;
2066 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2068 /* Integer is imprecise. NOK, IOKp */
2070 return IS_NUMBER_OVERFLOW_IV;
2072 #endif /* !NV_PRESERVES_UV*/
2075 S_sv_2iuv_common(pTHX_ SV *const sv)
2077 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2080 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2081 * without also getting a cached IV/UV from it at the same time
2082 * (ie PV->NV conversion should detect loss of accuracy and cache
2083 * IV or UV at same time to avoid this. */
2084 /* IV-over-UV optimisation - choose to cache IV if possible */
2086 if (SvTYPE(sv) == SVt_NV)
2087 sv_upgrade(sv, SVt_PVNV);
2089 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2090 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2091 certainly cast into the IV range at IV_MAX, whereas the correct
2092 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2094 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2095 if (Perl_isnan(SvNVX(sv))) {
2101 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2102 SvIV_set(sv, I_V(SvNVX(sv)));
2103 if (SvNVX(sv) == (NV) SvIVX(sv)
2104 #ifndef NV_PRESERVES_UV
2105 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2106 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2107 /* Don't flag it as "accurately an integer" if the number
2108 came from a (by definition imprecise) NV operation, and
2109 we're outside the range of NV integer precision */
2113 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2115 /* scalar has trailing garbage, eg "42a" */
2117 DEBUG_c(PerlIO_printf(Perl_debug_log,
2118 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2124 /* IV not precise. No need to convert from PV, as NV
2125 conversion would already have cached IV if it detected
2126 that PV->IV would be better than PV->NV->IV
2127 flags already correct - don't set public IOK. */
2128 DEBUG_c(PerlIO_printf(Perl_debug_log,
2129 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2134 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2135 but the cast (NV)IV_MIN rounds to a the value less (more
2136 negative) than IV_MIN which happens to be equal to SvNVX ??
2137 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2138 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2139 (NV)UVX == NVX are both true, but the values differ. :-(
2140 Hopefully for 2s complement IV_MIN is something like
2141 0x8000000000000000 which will be exact. NWC */
2144 SvUV_set(sv, U_V(SvNVX(sv)));
2146 (SvNVX(sv) == (NV) SvUVX(sv))
2147 #ifndef NV_PRESERVES_UV
2148 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2149 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2150 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2151 /* Don't flag it as "accurately an integer" if the number
2152 came from a (by definition imprecise) NV operation, and
2153 we're outside the range of NV integer precision */
2159 DEBUG_c(PerlIO_printf(Perl_debug_log,
2160 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2166 else if (SvPOKp(sv)) {
2168 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2169 /* We want to avoid a possible problem when we cache an IV/ a UV which
2170 may be later translated to an NV, and the resulting NV is not
2171 the same as the direct translation of the initial string
2172 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2173 be careful to ensure that the value with the .456 is around if the
2174 NV value is requested in the future).
2176 This means that if we cache such an IV/a UV, we need to cache the
2177 NV as well. Moreover, we trade speed for space, and do not
2178 cache the NV if we are sure it's not needed.
2181 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2182 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2183 == IS_NUMBER_IN_UV) {
2184 /* It's definitely an integer, only upgrade to PVIV */
2185 if (SvTYPE(sv) < SVt_PVIV)
2186 sv_upgrade(sv, SVt_PVIV);
2188 } else if (SvTYPE(sv) < SVt_PVNV)
2189 sv_upgrade(sv, SVt_PVNV);
2191 /* If NVs preserve UVs then we only use the UV value if we know that
2192 we aren't going to call atof() below. If NVs don't preserve UVs
2193 then the value returned may have more precision than atof() will
2194 return, even though value isn't perfectly accurate. */
2195 if ((numtype & (IS_NUMBER_IN_UV
2196 #ifdef NV_PRESERVES_UV
2199 )) == IS_NUMBER_IN_UV) {
2200 /* This won't turn off the public IOK flag if it was set above */
2201 (void)SvIOKp_on(sv);
2203 if (!(numtype & IS_NUMBER_NEG)) {
2205 if (value <= (UV)IV_MAX) {
2206 SvIV_set(sv, (IV)value);
2208 /* it didn't overflow, and it was positive. */
2209 SvUV_set(sv, value);
2213 /* 2s complement assumption */
2214 if (value <= (UV)IV_MIN) {
2215 SvIV_set(sv, -(IV)value);
2217 /* Too negative for an IV. This is a double upgrade, but
2218 I'm assuming it will be rare. */
2219 if (SvTYPE(sv) < SVt_PVNV)
2220 sv_upgrade(sv, SVt_PVNV);
2224 SvNV_set(sv, -(NV)value);
2225 SvIV_set(sv, IV_MIN);
2229 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2230 will be in the previous block to set the IV slot, and the next
2231 block to set the NV slot. So no else here. */
2233 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2234 != IS_NUMBER_IN_UV) {
2235 /* It wasn't an (integer that doesn't overflow the UV). */
2236 SvNV_set(sv, Atof(SvPVX_const(sv)));
2238 if (! numtype && ckWARN(WARN_NUMERIC))
2241 #if defined(USE_LONG_DOUBLE)
2242 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2243 PTR2UV(sv), SvNVX(sv)));
2245 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2246 PTR2UV(sv), SvNVX(sv)));
2249 #ifdef NV_PRESERVES_UV
2250 (void)SvIOKp_on(sv);
2252 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2253 SvIV_set(sv, I_V(SvNVX(sv)));
2254 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2257 NOOP; /* Integer is imprecise. NOK, IOKp */
2259 /* UV will not work better than IV */
2261 if (SvNVX(sv) > (NV)UV_MAX) {
2263 /* Integer is inaccurate. NOK, IOKp, is UV */
2264 SvUV_set(sv, UV_MAX);
2266 SvUV_set(sv, U_V(SvNVX(sv)));
2267 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2268 NV preservse UV so can do correct comparison. */
2269 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2272 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2277 #else /* NV_PRESERVES_UV */
2278 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2279 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2280 /* The IV/UV slot will have been set from value returned by
2281 grok_number above. The NV slot has just been set using
2284 assert (SvIOKp(sv));
2286 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2287 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2288 /* Small enough to preserve all bits. */
2289 (void)SvIOKp_on(sv);
2291 SvIV_set(sv, I_V(SvNVX(sv)));
2292 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2294 /* Assumption: first non-preserved integer is < IV_MAX,
2295 this NV is in the preserved range, therefore: */
2296 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2298 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);
2302 0 0 already failed to read UV.
2303 0 1 already failed to read UV.
2304 1 0 you won't get here in this case. IV/UV
2305 slot set, public IOK, Atof() unneeded.
2306 1 1 already read UV.
2307 so there's no point in sv_2iuv_non_preserve() attempting
2308 to use atol, strtol, strtoul etc. */
2310 sv_2iuv_non_preserve (sv, numtype);
2312 sv_2iuv_non_preserve (sv);
2316 #endif /* NV_PRESERVES_UV */
2317 /* It might be more code efficient to go through the entire logic above
2318 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2319 gets complex and potentially buggy, so more programmer efficient
2320 to do it this way, by turning off the public flags: */
2322 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2326 if (isGV_with_GP(sv))
2327 return glob_2number(MUTABLE_GV(sv));
2329 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2331 if (SvTYPE(sv) < SVt_IV)
2332 /* Typically the caller expects that sv_any is not NULL now. */
2333 sv_upgrade(sv, SVt_IV);
2334 /* Return 0 from the caller. */
2341 =for apidoc sv_2iv_flags
2343 Return the integer value of an SV, doing any necessary string
2344 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2345 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2351 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2353 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2355 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2356 && SvTYPE(sv) != SVt_PVFM);
2358 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2364 if (flags & SV_SKIP_OVERLOAD)
2366 tmpstr = AMG_CALLunary(sv, numer_amg);
2367 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2368 return SvIV(tmpstr);
2371 return PTR2IV(SvRV(sv));
2374 if (SvVALID(sv) || isREGEXP(sv)) {
2375 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2376 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2377 In practice they are extremely unlikely to actually get anywhere
2378 accessible by user Perl code - the only way that I'm aware of is when
2379 a constant subroutine which is used as the second argument to index.
2381 Regexps have no SvIVX and SvNVX fields.
2383 assert(isREGEXP(sv) || SvPOKp(sv));
2386 const char * const ptr =
2387 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2389 = grok_number(ptr, SvCUR(sv), &value);
2391 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2392 == IS_NUMBER_IN_UV) {
2393 /* It's definitely an integer */
2394 if (numtype & IS_NUMBER_NEG) {
2395 if (value < (UV)IV_MIN)
2398 if (value < (UV)IV_MAX)
2403 if (ckWARN(WARN_NUMERIC))
2406 return I_V(Atof(ptr));
2410 if (SvTHINKFIRST(sv)) {
2411 #ifdef PERL_OLD_COPY_ON_WRITE
2413 sv_force_normal_flags(sv, 0);
2416 if (SvREADONLY(sv) && !SvOK(sv)) {
2417 if (ckWARN(WARN_UNINITIALIZED))
2424 if (S_sv_2iuv_common(aTHX_ sv))
2428 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2429 PTR2UV(sv),SvIVX(sv)));
2430 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2434 =for apidoc sv_2uv_flags
2436 Return the unsigned integer value of an SV, doing any necessary string
2437 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2438 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2444 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2446 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2448 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2454 if (flags & SV_SKIP_OVERLOAD)
2456 tmpstr = AMG_CALLunary(sv, numer_amg);
2457 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2458 return SvUV(tmpstr);
2461 return PTR2UV(SvRV(sv));
2464 if (SvVALID(sv) || isREGEXP(sv)) {
2465 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2466 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2467 Regexps have no SvIVX and SvNVX fields. */
2468 assert(isREGEXP(sv) || SvPOKp(sv));
2471 const char * const ptr =
2472 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2474 = grok_number(ptr, SvCUR(sv), &value);
2476 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2477 == IS_NUMBER_IN_UV) {
2478 /* It's definitely an integer */
2479 if (!(numtype & IS_NUMBER_NEG))
2483 if (ckWARN(WARN_NUMERIC))
2486 return U_V(Atof(ptr));
2490 if (SvTHINKFIRST(sv)) {
2491 #ifdef PERL_OLD_COPY_ON_WRITE
2493 sv_force_normal_flags(sv, 0);
2496 if (SvREADONLY(sv) && !SvOK(sv)) {
2497 if (ckWARN(WARN_UNINITIALIZED))
2504 if (S_sv_2iuv_common(aTHX_ sv))
2508 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2509 PTR2UV(sv),SvUVX(sv)));
2510 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2514 =for apidoc sv_2nv_flags
2516 Return the num value of an SV, doing any necessary string or integer
2517 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2518 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2524 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2526 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2528 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2529 && SvTYPE(sv) != SVt_PVFM);
2530 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2531 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2532 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2533 Regexps have no SvIVX and SvNVX fields. */
2535 if (flags & SV_GMAGIC)
2539 if (SvPOKp(sv) && !SvIOKp(sv)) {
2540 ptr = SvPVX_const(sv);
2542 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2543 !grok_number(ptr, SvCUR(sv), NULL))
2549 return (NV)SvUVX(sv);
2551 return (NV)SvIVX(sv);
2557 ptr = RX_WRAPPED((REGEXP *)sv);
2560 assert(SvTYPE(sv) >= SVt_PVMG);
2561 /* This falls through to the report_uninit near the end of the
2563 } else if (SvTHINKFIRST(sv)) {
2568 if (flags & SV_SKIP_OVERLOAD)
2570 tmpstr = AMG_CALLunary(sv, numer_amg);
2571 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2572 return SvNV(tmpstr);
2575 return PTR2NV(SvRV(sv));
2577 #ifdef PERL_OLD_COPY_ON_WRITE
2579 sv_force_normal_flags(sv, 0);
2582 if (SvREADONLY(sv) && !SvOK(sv)) {
2583 if (ckWARN(WARN_UNINITIALIZED))
2588 if (SvTYPE(sv) < SVt_NV) {
2589 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2590 sv_upgrade(sv, SVt_NV);
2591 #ifdef USE_LONG_DOUBLE
2593 STORE_NUMERIC_LOCAL_SET_STANDARD();
2594 PerlIO_printf(Perl_debug_log,
2595 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2596 PTR2UV(sv), SvNVX(sv));
2597 RESTORE_NUMERIC_LOCAL();
2601 STORE_NUMERIC_LOCAL_SET_STANDARD();
2602 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2603 PTR2UV(sv), SvNVX(sv));
2604 RESTORE_NUMERIC_LOCAL();
2608 else if (SvTYPE(sv) < SVt_PVNV)
2609 sv_upgrade(sv, SVt_PVNV);
2614 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2615 #ifdef NV_PRESERVES_UV
2621 /* Only set the public NV OK flag if this NV preserves the IV */
2622 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2624 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2625 : (SvIVX(sv) == I_V(SvNVX(sv))))
2631 else if (SvPOKp(sv)) {
2633 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2634 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2636 #ifdef NV_PRESERVES_UV
2637 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2638 == IS_NUMBER_IN_UV) {
2639 /* It's definitely an integer */
2640 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2642 SvNV_set(sv, Atof(SvPVX_const(sv)));
2648 SvNV_set(sv, Atof(SvPVX_const(sv)));
2649 /* Only set the public NV OK flag if this NV preserves the value in
2650 the PV at least as well as an IV/UV would.
2651 Not sure how to do this 100% reliably. */
2652 /* if that shift count is out of range then Configure's test is
2653 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2655 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2656 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2657 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2658 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2659 /* Can't use strtol etc to convert this string, so don't try.
2660 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2663 /* value has been set. It may not be precise. */
2664 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2665 /* 2s complement assumption for (UV)IV_MIN */
2666 SvNOK_on(sv); /* Integer is too negative. */
2671 if (numtype & IS_NUMBER_NEG) {
2672 SvIV_set(sv, -(IV)value);
2673 } else if (value <= (UV)IV_MAX) {
2674 SvIV_set(sv, (IV)value);
2676 SvUV_set(sv, value);
2680 if (numtype & IS_NUMBER_NOT_INT) {
2681 /* I believe that even if the original PV had decimals,
2682 they are lost beyond the limit of the FP precision.
2683 However, neither is canonical, so both only get p
2684 flags. NWC, 2000/11/25 */
2685 /* Both already have p flags, so do nothing */
2687 const NV nv = SvNVX(sv);
2688 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2689 if (SvIVX(sv) == I_V(nv)) {
2692 /* It had no "." so it must be integer. */
2696 /* between IV_MAX and NV(UV_MAX).
2697 Could be slightly > UV_MAX */
2699 if (numtype & IS_NUMBER_NOT_INT) {
2700 /* UV and NV both imprecise. */
2702 const UV nv_as_uv = U_V(nv);
2704 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2713 /* It might be more code efficient to go through the entire logic above
2714 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2715 gets complex and potentially buggy, so more programmer efficient
2716 to do it this way, by turning off the public flags: */
2718 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2719 #endif /* NV_PRESERVES_UV */
2722 if (isGV_with_GP(sv)) {
2723 glob_2number(MUTABLE_GV(sv));
2727 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2729 assert (SvTYPE(sv) >= SVt_NV);
2730 /* Typically the caller expects that sv_any is not NULL now. */
2731 /* XXX Ilya implies that this is a bug in callers that assume this
2732 and ideally should be fixed. */
2735 #if defined(USE_LONG_DOUBLE)
2737 STORE_NUMERIC_LOCAL_SET_STANDARD();
2738 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2739 PTR2UV(sv), SvNVX(sv));
2740 RESTORE_NUMERIC_LOCAL();
2744 STORE_NUMERIC_LOCAL_SET_STANDARD();
2745 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2746 PTR2UV(sv), SvNVX(sv));
2747 RESTORE_NUMERIC_LOCAL();
2756 Return an SV with the numeric value of the source SV, doing any necessary
2757 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2758 access this function.
2764 Perl_sv_2num(pTHX_ SV *const sv)
2766 PERL_ARGS_ASSERT_SV_2NUM;
2771 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2772 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2773 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2774 return sv_2num(tmpsv);
2776 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2779 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2780 * UV as a string towards the end of buf, and return pointers to start and
2783 * We assume that buf is at least TYPE_CHARS(UV) long.
2787 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2789 char *ptr = buf + TYPE_CHARS(UV);
2790 char * const ebuf = ptr;
2793 PERL_ARGS_ASSERT_UIV_2BUF;
2805 *--ptr = '0' + (char)(uv % 10);
2814 =for apidoc sv_2pv_flags
2816 Returns a pointer to the string value of an SV, and sets *lp to its length.
2817 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2818 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2819 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2825 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2829 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2831 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2832 && SvTYPE(sv) != SVt_PVFM);
2833 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2838 if (flags & SV_SKIP_OVERLOAD)
2840 tmpstr = AMG_CALLunary(sv, string_amg);
2841 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2842 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2844 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2848 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2849 if (flags & SV_CONST_RETURN) {
2850 pv = (char *) SvPVX_const(tmpstr);
2852 pv = (flags & SV_MUTABLE_RETURN)
2853 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2856 *lp = SvCUR(tmpstr);
2858 pv = sv_2pv_flags(tmpstr, lp, flags);
2871 SV *const referent = SvRV(sv);
2875 retval = buffer = savepvn("NULLREF", len);
2876 } else if (SvTYPE(referent) == SVt_REGEXP &&
2877 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2878 amagic_is_enabled(string_amg))) {
2879 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2883 /* If the regex is UTF-8 we want the containing scalar to
2884 have an UTF-8 flag too */
2891 *lp = RX_WRAPLEN(re);
2893 return RX_WRAPPED(re);
2895 const char *const typestr = sv_reftype(referent, 0);
2896 const STRLEN typelen = strlen(typestr);
2897 UV addr = PTR2UV(referent);
2898 const char *stashname = NULL;
2899 STRLEN stashnamelen = 0; /* hush, gcc */
2900 const char *buffer_end;
2902 if (SvOBJECT(referent)) {
2903 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2906 stashname = HEK_KEY(name);
2907 stashnamelen = HEK_LEN(name);
2909 if (HEK_UTF8(name)) {
2915 stashname = "__ANON__";
2918 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2919 + 2 * sizeof(UV) + 2 /* )\0 */;
2921 len = typelen + 3 /* (0x */
2922 + 2 * sizeof(UV) + 2 /* )\0 */;
2925 Newx(buffer, len, char);
2926 buffer_end = retval = buffer + len;
2928 /* Working backwards */
2932 *--retval = PL_hexdigit[addr & 15];
2933 } while (addr >>= 4);
2939 memcpy(retval, typestr, typelen);
2943 retval -= stashnamelen;
2944 memcpy(retval, stashname, stashnamelen);
2946 /* retval may not necessarily have reached the start of the
2948 assert (retval >= buffer);
2950 len = buffer_end - retval - 1; /* -1 for that \0 */
2962 if (flags & SV_MUTABLE_RETURN)
2963 return SvPVX_mutable(sv);
2964 if (flags & SV_CONST_RETURN)
2965 return (char *)SvPVX_const(sv);
2970 /* I'm assuming that if both IV and NV are equally valid then
2971 converting the IV is going to be more efficient */
2972 const U32 isUIOK = SvIsUV(sv);
2973 char buf[TYPE_CHARS(UV)];
2977 if (SvTYPE(sv) < SVt_PVIV)
2978 sv_upgrade(sv, SVt_PVIV);
2979 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2981 /* inlined from sv_setpvn */
2982 s = SvGROW_mutable(sv, len + 1);
2983 Move(ptr, s, len, char);
2988 else if (SvNOK(sv)) {
2989 if (SvTYPE(sv) < SVt_PVNV)
2990 sv_upgrade(sv, SVt_PVNV);
2991 if (SvNVX(sv) == 0.0) {
2992 s = SvGROW_mutable(sv, 2);
2997 /* The +20 is pure guesswork. Configure test needed. --jhi */
2998 s = SvGROW_mutable(sv, NV_DIG + 20);
2999 /* some Xenix systems wipe out errno here */
3001 #ifndef USE_LOCALE_NUMERIC
3002 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3006 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3007 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3009 /* If the radix character is UTF-8, and actually is in the
3010 * output, turn on the UTF-8 flag for the scalar */
3011 if (PL_numeric_local
3012 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
3013 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3017 RESTORE_LC_NUMERIC();
3020 /* We don't call SvPOK_on(), because it may come to pass that the
3021 * locale changes so that the stringification we just did is no
3022 * longer correct. We will have to re-stringify every time it is
3029 else if (isGV_with_GP(sv)) {
3030 GV *const gv = MUTABLE_GV(sv);
3031 SV *const buffer = sv_newmortal();
3033 gv_efullname3(buffer, gv, "*");
3035 assert(SvPOK(buffer));
3039 *lp = SvCUR(buffer);
3040 return SvPVX(buffer);
3042 else if (isREGEXP(sv)) {
3043 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3044 return RX_WRAPPED((REGEXP *)sv);
3049 if (flags & SV_UNDEF_RETURNS_NULL)
3051 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3053 /* Typically the caller expects that sv_any is not NULL now. */
3054 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3055 sv_upgrade(sv, SVt_PV);
3060 const STRLEN len = s - SvPVX_const(sv);
3065 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3066 PTR2UV(sv),SvPVX_const(sv)));
3067 if (flags & SV_CONST_RETURN)
3068 return (char *)SvPVX_const(sv);
3069 if (flags & SV_MUTABLE_RETURN)
3070 return SvPVX_mutable(sv);
3075 =for apidoc sv_copypv
3077 Copies a stringified representation of the source SV into the
3078 destination SV. Automatically performs any necessary mg_get and
3079 coercion of numeric values into strings. Guaranteed to preserve
3080 UTF8 flag even from overloaded objects. Similar in nature to
3081 sv_2pv[_flags] but operates directly on an SV instead of just the
3082 string. Mostly uses sv_2pv_flags to do its work, except when that
3083 would lose the UTF-8'ness of the PV.
3085 =for apidoc sv_copypv_nomg
3087 Like sv_copypv, but doesn't invoke get magic first.
3089 =for apidoc sv_copypv_flags
3091 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3098 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3100 PERL_ARGS_ASSERT_SV_COPYPV;
3102 sv_copypv_flags(dsv, ssv, 0);
3106 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3111 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3113 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3115 s = SvPV_nomg_const(ssv,len);
3116 sv_setpvn(dsv,s,len);
3124 =for apidoc sv_2pvbyte
3126 Return a pointer to the byte-encoded representation of the SV, and set *lp
3127 to its length. May cause the SV to be downgraded from UTF-8 as a
3130 Usually accessed via the C<SvPVbyte> macro.
3136 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3138 PERL_ARGS_ASSERT_SV_2PVBYTE;
3141 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3142 || isGV_with_GP(sv) || SvROK(sv)) {
3143 SV *sv2 = sv_newmortal();
3144 sv_copypv_nomg(sv2,sv);
3147 sv_utf8_downgrade(sv,0);
3148 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3152 =for apidoc sv_2pvutf8
3154 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3155 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3157 Usually accessed via the C<SvPVutf8> macro.
3163 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3165 PERL_ARGS_ASSERT_SV_2PVUTF8;
3167 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3168 || isGV_with_GP(sv) || SvROK(sv))
3169 sv = sv_mortalcopy(sv);
3172 sv_utf8_upgrade_nomg(sv);
3173 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3178 =for apidoc sv_2bool
3180 This macro is only used by sv_true() or its macro equivalent, and only if
3181 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3182 It calls sv_2bool_flags with the SV_GMAGIC flag.
3184 =for apidoc sv_2bool_flags
3186 This function is only used by sv_true() and friends, and only if
3187 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3188 contain SV_GMAGIC, then it does an mg_get() first.
3195 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3197 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3200 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3206 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3207 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3210 if(SvGMAGICAL(sv)) {
3212 goto restart; /* call sv_2bool */
3214 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3215 else if(!SvOK(sv)) {
3218 else if(SvPOK(sv)) {
3219 svb = SvPVXtrue(sv);
3221 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3222 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3223 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3227 goto restart; /* call sv_2bool_nomg */
3232 return SvRV(sv) != 0;
3236 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3237 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3241 =for apidoc sv_utf8_upgrade
3243 Converts the PV of an SV to its UTF-8-encoded form.
3244 Forces the SV to string form if it is not already.
3245 Will C<mg_get> on C<sv> if appropriate.
3246 Always sets the SvUTF8 flag to avoid future validity checks even
3247 if the whole string is the same in UTF-8 as not.
3248 Returns the number of bytes in the converted string
3250 This is not a general purpose byte encoding to Unicode interface:
3251 use the Encode extension for that.
3253 =for apidoc sv_utf8_upgrade_nomg
3255 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3257 =for apidoc sv_utf8_upgrade_flags
3259 Converts the PV of an SV to its UTF-8-encoded form.
3260 Forces the SV to string form if it is not already.
3261 Always sets the SvUTF8 flag to avoid future validity checks even
3262 if all the bytes are invariant in UTF-8.
3263 If C<flags> has C<SV_GMAGIC> bit set,
3264 will C<mg_get> on C<sv> if appropriate, else not.
3266 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3267 will expand when converted to UTF-8, and skips the extra work of checking for
3268 that. Typically this flag is used by a routine that has already parsed the
3269 string and found such characters, and passes this information on so that the
3270 work doesn't have to be repeated.
3272 Returns the number of bytes in the converted string.
3274 This is not a general purpose byte encoding to Unicode interface:
3275 use the Encode extension for that.
3277 =for apidoc sv_utf8_upgrade_flags_grow
3279 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3280 the number of unused bytes the string of 'sv' is guaranteed to have free after
3281 it upon return. This allows the caller to reserve extra space that it intends
3282 to fill, to avoid extra grows.
3284 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3285 are implemented in terms of this function.
3287 Returns the number of bytes in the converted string (not including the spares).
3291 (One might think that the calling routine could pass in the position of the
3292 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3293 have to be found again. But that is not the case, because typically when the
3294 caller is likely to use this flag, it won't be calling this routine unless it
3295 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3296 and just use bytes. But some things that do fit into a byte are variants in
3297 utf8, and the caller may not have been keeping track of these.)
3299 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3300 C<NUL> isn't guaranteed due to having other routines do the work in some input
3301 cases, or if the input is already flagged as being in utf8.
3303 The speed of this could perhaps be improved for many cases if someone wanted to
3304 write a fast function that counts the number of variant characters in a string,
3305 especially if it could return the position of the first one.
3310 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3312 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3314 if (sv == &PL_sv_undef)
3316 if (!SvPOK_nog(sv)) {
3318 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3319 (void) sv_2pv_flags(sv,&len, flags);
3321 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3325 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3330 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3335 S_sv_uncow(aTHX_ sv, 0);
3338 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3339 sv_recode_to_utf8(sv, PL_encoding);
3340 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3344 if (SvCUR(sv) == 0) {
3345 if (extra) SvGROW(sv, extra);
3346 } else { /* Assume Latin-1/EBCDIC */
3347 /* This function could be much more efficient if we
3348 * had a FLAG in SVs to signal if there are any variant
3349 * chars in the PV. Given that there isn't such a flag
3350 * make the loop as fast as possible (although there are certainly ways
3351 * to speed this up, eg. through vectorization) */
3352 U8 * s = (U8 *) SvPVX_const(sv);
3353 U8 * e = (U8 *) SvEND(sv);
3355 STRLEN two_byte_count = 0;
3357 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3359 /* See if really will need to convert to utf8. We mustn't rely on our
3360 * incoming SV being well formed and having a trailing '\0', as certain
3361 * code in pp_formline can send us partially built SVs. */
3365 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3367 t--; /* t already incremented; re-point to first variant */
3372 /* utf8 conversion not needed because all are invariants. Mark as
3373 * UTF-8 even if no variant - saves scanning loop */
3375 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3380 /* Here, the string should be converted to utf8, either because of an
3381 * input flag (two_byte_count = 0), or because a character that
3382 * requires 2 bytes was found (two_byte_count = 1). t points either to
3383 * the beginning of the string (if we didn't examine anything), or to
3384 * the first variant. In either case, everything from s to t - 1 will
3385 * occupy only 1 byte each on output.
3387 * There are two main ways to convert. One is to create a new string
3388 * and go through the input starting from the beginning, appending each
3389 * converted value onto the new string as we go along. It's probably
3390 * best to allocate enough space in the string for the worst possible
3391 * case rather than possibly running out of space and having to
3392 * reallocate and then copy what we've done so far. Since everything
3393 * from s to t - 1 is invariant, the destination can be initialized
3394 * with these using a fast memory copy
3396 * The other way is to figure out exactly how big the string should be
3397 * by parsing the entire input. Then you don't have to make it big
3398 * enough to handle the worst possible case, and more importantly, if
3399 * the string you already have is large enough, you don't have to
3400 * allocate a new string, you can copy the last character in the input
3401 * string to the final position(s) that will be occupied by the
3402 * converted string and go backwards, stopping at t, since everything
3403 * before that is invariant.
3405 * There are advantages and disadvantages to each method.
3407 * In the first method, we can allocate a new string, do the memory
3408 * copy from the s to t - 1, and then proceed through the rest of the
3409 * string byte-by-byte.
3411 * In the second method, we proceed through the rest of the input
3412 * string just calculating how big the converted string will be. Then
3413 * there are two cases:
3414 * 1) if the string has enough extra space to handle the converted
3415 * value. We go backwards through the string, converting until we
3416 * get to the position we are at now, and then stop. If this
3417 * position is far enough along in the string, this method is
3418 * faster than the other method. If the memory copy were the same
3419 * speed as the byte-by-byte loop, that position would be about
3420 * half-way, as at the half-way mark, parsing to the end and back
3421 * is one complete string's parse, the same amount as starting
3422 * over and going all the way through. Actually, it would be
3423 * somewhat less than half-way, as it's faster to just count bytes
3424 * than to also copy, and we don't have the overhead of allocating
3425 * a new string, changing the scalar to use it, and freeing the
3426 * existing one. But if the memory copy is fast, the break-even
3427 * point is somewhere after half way. The counting loop could be
3428 * sped up by vectorization, etc, to move the break-even point
3429 * further towards the beginning.
3430 * 2) if the string doesn't have enough space to handle the converted
3431 * value. A new string will have to be allocated, and one might
3432 * as well, given that, start from the beginning doing the first
3433 * method. We've spent extra time parsing the string and in
3434 * exchange all we've gotten is that we know precisely how big to
3435 * make the new one. Perl is more optimized for time than space,
3436 * so this case is a loser.
3437 * So what I've decided to do is not use the 2nd method unless it is
3438 * guaranteed that a new string won't have to be allocated, assuming
3439 * the worst case. I also decided not to put any more conditions on it
3440 * than this, for now. It seems likely that, since the worst case is
3441 * twice as big as the unknown portion of the string (plus 1), we won't
3442 * be guaranteed enough space, causing us to go to the first method,
3443 * unless the string is short, or the first variant character is near
3444 * the end of it. In either of these cases, it seems best to use the
3445 * 2nd method. The only circumstance I can think of where this would
3446 * be really slower is if the string had once had much more data in it
3447 * than it does now, but there is still a substantial amount in it */
3450 STRLEN invariant_head = t - s;
3451 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3452 if (SvLEN(sv) < size) {
3454 /* Here, have decided to allocate a new string */
3459 Newx(dst, size, U8);
3461 /* If no known invariants at the beginning of the input string,
3462 * set so starts from there. Otherwise, can use memory copy to
3463 * get up to where we are now, and then start from here */
3465 if (invariant_head <= 0) {
3468 Copy(s, dst, invariant_head, char);
3469 d = dst + invariant_head;
3473 append_utf8_from_native_byte(*t, &d);
3477 SvPV_free(sv); /* No longer using pre-existing string */
3478 SvPV_set(sv, (char*)dst);
3479 SvCUR_set(sv, d - dst);
3480 SvLEN_set(sv, size);
3483 /* Here, have decided to get the exact size of the string.
3484 * Currently this happens only when we know that there is
3485 * guaranteed enough space to fit the converted string, so
3486 * don't have to worry about growing. If two_byte_count is 0,
3487 * then t points to the first byte of the string which hasn't
3488 * been examined yet. Otherwise two_byte_count is 1, and t
3489 * points to the first byte in the string that will expand to
3490 * two. Depending on this, start examining at t or 1 after t.
3493 U8 *d = t + two_byte_count;
3496 /* Count up the remaining bytes that expand to two */
3499 const U8 chr = *d++;
3500 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3503 /* The string will expand by just the number of bytes that
3504 * occupy two positions. But we are one afterwards because of
3505 * the increment just above. This is the place to put the
3506 * trailing NUL, and to set the length before we decrement */
3508 d += two_byte_count;
3509 SvCUR_set(sv, d - s);
3513 /* Having decremented d, it points to the position to put the
3514 * very last byte of the expanded string. Go backwards through
3515 * the string, copying and expanding as we go, stopping when we
3516 * get to the part that is invariant the rest of the way down */
3520 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3523 *d-- = UTF8_EIGHT_BIT_LO(*e);
3524 *d-- = UTF8_EIGHT_BIT_HI(*e);
3530 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3531 /* Update pos. We do it at the end rather than during
3532 * the upgrade, to avoid slowing down the common case
3533 * (upgrade without pos).
3534 * pos can be stored as either bytes or characters. Since
3535 * this was previously a byte string we can just turn off
3536 * the bytes flag. */
3537 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3539 mg->mg_flags &= ~MGf_BYTES;
3541 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3542 magic_setutf8(sv,mg); /* clear UTF8 cache */
3547 /* Mark as UTF-8 even if no variant - saves scanning loop */
3553 =for apidoc sv_utf8_downgrade
3555 Attempts to convert the PV of an SV from characters to bytes.
3556 If the PV contains a character that cannot fit
3557 in a byte, this conversion will fail;
3558 in this case, either returns false or, if C<fail_ok> is not
3561 This is not a general purpose Unicode to byte encoding interface:
3562 use the Encode extension for that.
3568 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3570 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3572 if (SvPOKp(sv) && SvUTF8(sv)) {
3576 int mg_flags = SV_GMAGIC;
3579 S_sv_uncow(aTHX_ sv, 0);
3581 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3583 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3584 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3585 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3586 SV_GMAGIC|SV_CONST_RETURN);
3587 mg_flags = 0; /* sv_pos_b2u does get magic */
3589 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3590 magic_setutf8(sv,mg); /* clear UTF8 cache */
3593 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3595 if (!utf8_to_bytes(s, &len)) {
3600 Perl_croak(aTHX_ "Wide character in %s",
3603 Perl_croak(aTHX_ "Wide character");
3614 =for apidoc sv_utf8_encode
3616 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3617 flag off so that it looks like octets again.
3623 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3625 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3627 if (SvREADONLY(sv)) {
3628 sv_force_normal_flags(sv, 0);
3630 (void) sv_utf8_upgrade(sv);
3635 =for apidoc sv_utf8_decode
3637 If the PV of the SV is an octet sequence in UTF-8
3638 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3639 so that it looks like a character. If the PV contains only single-byte
3640 characters, the C<SvUTF8> flag stays off.
3641 Scans PV for validity and returns false if the PV is invalid UTF-8.
3647 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3649 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3652 const U8 *start, *c;
3655 /* The octets may have got themselves encoded - get them back as
3658 if (!sv_utf8_downgrade(sv, TRUE))
3661 /* it is actually just a matter of turning the utf8 flag on, but
3662 * we want to make sure everything inside is valid utf8 first.
3664 c = start = (const U8 *) SvPVX_const(sv);
3665 if (!is_utf8_string(c, SvCUR(sv)))
3667 e = (const U8 *) SvEND(sv);
3670 if (!UTF8_IS_INVARIANT(ch)) {
3675 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3676 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3677 after this, clearing pos. Does anything on CPAN
3679 /* adjust pos to the start of a UTF8 char sequence */
3680 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3682 I32 pos = mg->mg_len;
3684 for (c = start + pos; c > start; c--) {
3685 if (UTF8_IS_START(*c))
3688 mg->mg_len = c - start;
3691 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3692 magic_setutf8(sv,mg); /* clear UTF8 cache */
3699 =for apidoc sv_setsv
3701 Copies the contents of the source SV C<ssv> into the destination SV
3702 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3703 function if the source SV needs to be reused. Does not handle 'set' magic on
3704 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3705 performs a copy-by-value, obliterating any previous content of the
3708 You probably want to use one of the assortment of wrappers, such as
3709 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3710 C<SvSetMagicSV_nosteal>.
3712 =for apidoc sv_setsv_flags
3714 Copies the contents of the source SV C<ssv> into the destination SV
3715 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3716 function if the source SV needs to be reused. Does not handle 'set' magic.
3717 Loosely speaking, it performs a copy-by-value, obliterating any previous
3718 content of the destination.
3719 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3720 C<ssv> if appropriate, else not. If the C<flags>
3721 parameter has the C<SV_NOSTEAL> bit set then the
3722 buffers of temps will not be stolen. <sv_setsv>
3723 and C<sv_setsv_nomg> are implemented in terms of this function.
3725 You probably want to use one of the assortment of wrappers, such as
3726 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3727 C<SvSetMagicSV_nosteal>.
3729 This is the primary function for copying scalars, and most other
3730 copy-ish functions and macros use this underneath.
3736 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3738 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3739 HV *old_stash = NULL;
3741 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3743 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3744 const char * const name = GvNAME(sstr);
3745 const STRLEN len = GvNAMELEN(sstr);
3747 if (dtype >= SVt_PV) {
3753 SvUPGRADE(dstr, SVt_PVGV);
3754 (void)SvOK_off(dstr);
3755 isGV_with_GP_on(dstr);
3757 GvSTASH(dstr) = GvSTASH(sstr);
3759 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3760 gv_name_set(MUTABLE_GV(dstr), name, len,
3761 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3762 SvFAKE_on(dstr); /* can coerce to non-glob */
3765 if(GvGP(MUTABLE_GV(sstr))) {
3766 /* If source has method cache entry, clear it */
3768 SvREFCNT_dec(GvCV(sstr));
3769 GvCV_set(sstr, NULL);
3772 /* If source has a real method, then a method is
3775 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3781 /* If dest already had a real method, that's a change as well */
3783 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3784 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3789 /* We don't need to check the name of the destination if it was not a
3790 glob to begin with. */
3791 if(dtype == SVt_PVGV) {
3792 const char * const name = GvNAME((const GV *)dstr);
3795 /* The stash may have been detached from the symbol table, so
3797 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3801 const STRLEN len = GvNAMELEN(dstr);
3802 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3803 || (len == 1 && name[0] == ':')) {
3806 /* Set aside the old stash, so we can reset isa caches on
3808 if((old_stash = GvHV(dstr)))
3809 /* Make sure we do not lose it early. */
3810 SvREFCNT_inc_simple_void_NN(
3811 sv_2mortal((SV *)old_stash)
3816 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3819 gp_free(MUTABLE_GV(dstr));
3820 GvINTRO_off(dstr); /* one-shot flag */
3821 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3822 if (SvTAINTED(sstr))
3824 if (GvIMPORTED(dstr) != GVf_IMPORTED
3825 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3827 GvIMPORTED_on(dstr);
3830 if(mro_changes == 2) {
3831 if (GvAV((const GV *)sstr)) {
3833 SV * const sref = (SV *)GvAV((const GV *)dstr);
3834 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3835 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3836 AV * const ary = newAV();
3837 av_push(ary, mg->mg_obj); /* takes the refcount */
3838 mg->mg_obj = (SV *)ary;
3840 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3842 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3844 mro_isa_changed_in(GvSTASH(dstr));
3846 else if(mro_changes == 3) {
3847 HV * const stash = GvHV(dstr);
3848 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3854 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3855 if (GvIO(dstr) && dtype == SVt_PVGV) {
3856 DEBUG_o(Perl_deb(aTHX_
3857 "glob_assign_glob clearing PL_stashcache\n"));
3858 /* It's a cache. It will rebuild itself quite happily.
3859 It's a lot of effort to work out exactly which key (or keys)
3860 might be invalidated by the creation of the this file handle.
3862 hv_clear(PL_stashcache);
3868 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3870 SV * const sref = SvRV(sstr);
3872 const int intro = GvINTRO(dstr);
3875 const U32 stype = SvTYPE(sref);
3877 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3880 GvINTRO_off(dstr); /* one-shot flag */
3881 GvLINE(dstr) = CopLINE(PL_curcop);
3882 GvEGV(dstr) = MUTABLE_GV(dstr);
3887 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3888 import_flag = GVf_IMPORTED_CV;
3891 location = (SV **) &GvHV(dstr);
3892 import_flag = GVf_IMPORTED_HV;
3895 location = (SV **) &GvAV(dstr);
3896 import_flag = GVf_IMPORTED_AV;
3899 location = (SV **) &GvIOp(dstr);
3902 location = (SV **) &GvFORM(dstr);
3905 location = &GvSV(dstr);
3906 import_flag = GVf_IMPORTED_SV;
3909 if (stype == SVt_PVCV) {
3910 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3911 if (GvCVGEN(dstr)) {
3912 SvREFCNT_dec(GvCV(dstr));
3913 GvCV_set(dstr, NULL);
3914 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3917 /* SAVEt_GVSLOT takes more room on the savestack and has more
3918 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3919 leave_scope needs access to the GV so it can reset method
3920 caches. We must use SAVEt_GVSLOT whenever the type is
3921 SVt_PVCV, even if the stash is anonymous, as the stash may
3922 gain a name somehow before leave_scope. */
3923 if (stype == SVt_PVCV) {
3924 /* There is no save_pushptrptrptr. Creating it for this
3925 one call site would be overkill. So inline the ss add
3929 SS_ADD_PTR(location);
3930 SS_ADD_PTR(SvREFCNT_inc(*location));
3931 SS_ADD_UV(SAVEt_GVSLOT);
3934 else SAVEGENERICSV(*location);
3937 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3938 CV* const cv = MUTABLE_CV(*location);
3940 if (!GvCVGEN((const GV *)dstr) &&
3941 (CvROOT(cv) || CvXSUB(cv)) &&
3942 /* redundant check that avoids creating the extra SV
3943 most of the time: */
3944 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3946 SV * const new_const_sv =
3947 CvCONST((const CV *)sref)
3948 ? cv_const_sv((const CV *)sref)
3950 report_redefined_cv(
3951 sv_2mortal(Perl_newSVpvf(aTHX_
3954 HvNAME_HEK(GvSTASH((const GV *)dstr))
3956 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3959 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3963 cv_ckproto_len_flags(cv, (const GV *)dstr,
3964 SvPOK(sref) ? CvPROTO(sref) : NULL,
3965 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3966 SvPOK(sref) ? SvUTF8(sref) : 0);
3968 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3969 GvASSUMECV_on(dstr);
3970 if(GvSTASH(dstr)) gv_method_changed(dstr); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3972 *location = SvREFCNT_inc_simple_NN(sref);
3973 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3974 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3975 GvFLAGS(dstr) |= import_flag;
3977 if (stype == SVt_PVHV) {
3978 const char * const name = GvNAME((GV*)dstr);
3979 const STRLEN len = GvNAMELEN(dstr);
3982 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3983 || (len == 1 && name[0] == ':')
3985 && (!dref || HvENAME_get(dref))
3988 (HV *)sref, (HV *)dref,
3994 stype == SVt_PVAV && sref != dref
3995 && strEQ(GvNAME((GV*)dstr), "ISA")
3996 /* The stash may have been detached from the symbol table, so
3997 check its name before doing anything. */
3998 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4001 MAGIC * const omg = dref && SvSMAGICAL(dref)
4002 ? mg_find(dref, PERL_MAGIC_isa)
4004 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4005 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4006 AV * const ary = newAV();
4007 av_push(ary, mg->mg_obj); /* takes the refcount */
4008 mg->mg_obj = (SV *)ary;
4011 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4012 SV **svp = AvARRAY((AV *)omg->mg_obj);
4013 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4017 SvREFCNT_inc_simple_NN(*svp++)
4023 SvREFCNT_inc_simple_NN(omg->mg_obj)
4027 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4032 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4034 mg = mg_find(sref, PERL_MAGIC_isa);
4036 /* Since the *ISA assignment could have affected more than
4037 one stash, don't call mro_isa_changed_in directly, but let
4038 magic_clearisa do it for us, as it already has the logic for
4039 dealing with globs vs arrays of globs. */
4041 Perl_magic_clearisa(aTHX_ NULL, mg);
4043 else if (stype == SVt_PVIO) {
4044 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4045 /* It's a cache. It will rebuild itself quite happily.
4046 It's a lot of effort to work out exactly which key (or keys)
4047 might be invalidated by the creation of the this file handle.
4049 hv_clear(PL_stashcache);
4053 if (!intro) SvREFCNT_dec(dref);
4054 if (SvTAINTED(sstr))
4062 #ifdef PERL_DEBUG_READONLY_COW
4063 # include <sys/mman.h>
4065 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4066 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4070 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4072 struct perl_memory_debug_header * const header =
4073 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4074 const MEM_SIZE len = header->size;
4075 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4076 # ifdef PERL_TRACK_MEMPOOL
4077 if (!header->readonly) header->readonly = 1;
4079 if (mprotect(header, len, PROT_READ))
4080 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4081 header, len, errno);
4085 S_sv_buf_to_rw(pTHX_ SV *sv)
4087 struct perl_memory_debug_header * const header =
4088 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4089 const MEM_SIZE len = header->size;
4090 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4091 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4092 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4093 header, len, errno);
4094 # ifdef PERL_TRACK_MEMPOOL
4095 header->readonly = 0;
4100 # define sv_buf_to_ro(sv) NOOP
4101 # define sv_buf_to_rw(sv) NOOP
4105 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4111 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4116 if (SvIS_FREED(dstr)) {
4117 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4118 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4120 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4122 sstr = &PL_sv_undef;
4123 if (SvIS_FREED(sstr)) {
4124 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4125 (void*)sstr, (void*)dstr);
4127 stype = SvTYPE(sstr);
4128 dtype = SvTYPE(dstr);
4130 /* There's a lot of redundancy below but we're going for speed here */
4135 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4136 (void)SvOK_off(dstr);
4144 sv_upgrade(dstr, SVt_IV);
4148 sv_upgrade(dstr, SVt_PVIV);
4152 goto end_of_first_switch;
4154 (void)SvIOK_only(dstr);
4155 SvIV_set(dstr, SvIVX(sstr));
4158 /* SvTAINTED can only be true if the SV has taint magic, which in
4159 turn means that the SV type is PVMG (or greater). This is the
4160 case statement for SVt_IV, so this cannot be true (whatever gcov
4162 assert(!SvTAINTED(sstr));
4167 if (dtype < SVt_PV && dtype != SVt_IV)
4168 sv_upgrade(dstr, SVt_IV);
4176 sv_upgrade(dstr, SVt_NV);
4180 sv_upgrade(dstr, SVt_PVNV);
4184 goto end_of_first_switch;
4186 SvNV_set(dstr, SvNVX(sstr));
4187 (void)SvNOK_only(dstr);
4188 /* SvTAINTED can only be true if the SV has taint magic, which in
4189 turn means that the SV type is PVMG (or greater). This is the
4190 case statement for SVt_NV, so this cannot be true (whatever gcov
4192 assert(!SvTAINTED(sstr));
4199 sv_upgrade(dstr, SVt_PV);
4202 if (dtype < SVt_PVIV)
4203 sv_upgrade(dstr, SVt_PVIV);
4206 if (dtype < SVt_PVNV)
4207 sv_upgrade(dstr, SVt_PVNV);
4211 const char * const type = sv_reftype(sstr,0);
4213 /* diag_listed_as: Bizarre copy of %s */
4214 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4216 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4218 NOT_REACHED; /* NOTREACHED */
4222 if (dtype < SVt_REGEXP)
4224 if (dtype >= SVt_PV) {
4230 sv_upgrade(dstr, SVt_REGEXP);
4238 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4240 if (SvTYPE(sstr) != stype)
4241 stype = SvTYPE(sstr);
4243 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4244 glob_assign_glob(dstr, sstr, dtype);
4247 if (stype == SVt_PVLV)
4249 if (isREGEXP(sstr)) goto upgregexp;
4250 SvUPGRADE(dstr, SVt_PVNV);
4253 SvUPGRADE(dstr, (svtype)stype);
4255 end_of_first_switch:
4257 /* dstr may have been upgraded. */
4258 dtype = SvTYPE(dstr);
4259 sflags = SvFLAGS(sstr);
4261 if (dtype == SVt_PVCV) {
4262 /* Assigning to a subroutine sets the prototype. */
4265 const char *const ptr = SvPV_const(sstr, len);
4267 SvGROW(dstr, len + 1);
4268 Copy(ptr, SvPVX(dstr), len + 1, char);
4269 SvCUR_set(dstr, len);
4271 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4272 CvAUTOLOAD_off(dstr);
4277 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4278 const char * const type = sv_reftype(dstr,0);
4280 /* diag_listed_as: Cannot copy to %s */
4281 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4283 Perl_croak(aTHX_ "Cannot copy to %s", type);
4284 } else if (sflags & SVf_ROK) {
4285 if (isGV_with_GP(dstr)
4286 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4289 if (GvIMPORTED(dstr) != GVf_IMPORTED
4290 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4292 GvIMPORTED_on(dstr);
4297 glob_assign_glob(dstr, sstr, dtype);
4301 if (dtype >= SVt_PV) {
4302 if (isGV_with_GP(dstr)) {
4303 glob_assign_ref(dstr, sstr);
4306 if (SvPVX_const(dstr)) {
4312 (void)SvOK_off(dstr);
4313 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4314 SvFLAGS(dstr) |= sflags & SVf_ROK;
4315 assert(!(sflags & SVp_NOK));
4316 assert(!(sflags & SVp_IOK));
4317 assert(!(sflags & SVf_NOK));
4318 assert(!(sflags & SVf_IOK));
4320 else if (isGV_with_GP(dstr)) {
4321 if (!(sflags & SVf_OK)) {
4322 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4323 "Undefined value assigned to typeglob");
4326 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4327 if (dstr != (const SV *)gv) {
4328 const char * const name = GvNAME((const GV *)dstr);
4329 const STRLEN len = GvNAMELEN(dstr);
4330 HV *old_stash = NULL;
4331 bool reset_isa = FALSE;
4332 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4333 || (len == 1 && name[0] == ':')) {
4334 /* Set aside the old stash, so we can reset isa caches
4335 on its subclasses. */
4336 if((old_stash = GvHV(dstr))) {
4337 /* Make sure we do not lose it early. */
4338 SvREFCNT_inc_simple_void_NN(
4339 sv_2mortal((SV *)old_stash)
4346 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4347 gp_free(MUTABLE_GV(dstr));
4349 GvGP_set(dstr, gp_ref(GvGP(gv)));
4352 HV * const stash = GvHV(dstr);
4354 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4364 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4365 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4366 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4368 else if (sflags & SVp_POK) {
4369 const STRLEN cur = SvCUR(sstr);
4370 const STRLEN len = SvLEN(sstr);
4373 * We have three basic ways to copy the string:
4379 * Which we choose is based on various factors. The following
4380 * things are listed in order of speed, fastest to slowest:
4382 * - Copying a short string
4383 * - Copy-on-write bookkeeping
4385 * - Copying a long string
4387 * We swipe the string (steal the string buffer) if the SV on the
4388 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4389 * big win on long strings. It should be a win on short strings if
4390 * SvPVX_const(dstr) has to be allocated. If not, it should not
4391 * slow things down, as SvPVX_const(sstr) would have been freed
4394 * We also steal the buffer from a PADTMP (operator target) if it
4395 * is ‘long enough’. For short strings, a swipe does not help
4396 * here, as it causes more malloc calls the next time the target
4397 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4398 * be allocated it is still not worth swiping PADTMPs for short
4399 * strings, as the savings here are small.
4401 * If the rhs is already flagged as a copy-on-write string and COW
4402 * is possible here, we use copy-on-write and make both SVs share
4403 * the string buffer.
4405 * If the rhs is not flagged as copy-on-write, then we see whether
4406 * it is worth upgrading it to such. If the lhs already has a buf-
4407 * fer big enough and the string is short, we skip it and fall back
4408 * to method 3, since memcpy is faster for short strings than the
4409 * later bookkeeping overhead that copy-on-write entails.
4411 * If there is no buffer on the left, or the buffer is too small,
4412 * then we use copy-on-write.
4415 /* Whichever path we take through the next code, we want this true,
4416 and doing it now facilitates the COW check. */
4417 (void)SvPOK_only(dstr);
4421 /* slated for free anyway (and not COW)? */
4422 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4423 /* or a swipable TARG */
4424 || ((sflags & (SVs_PADTMP|SVf_READONLY|SVf_IsCOW))
4426 /* whose buffer is worth stealing */
4427 && CHECK_COWBUF_THRESHOLD(cur,len)
4430 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4431 (!(flags & SV_NOSTEAL)) &&
4432 /* and we're allowed to steal temps */
4433 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4434 len) /* and really is a string */
4435 { /* Passes the swipe test. */
4436 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4438 SvPV_set(dstr, SvPVX_mutable(sstr));
4439 SvLEN_set(dstr, SvLEN(sstr));
4440 SvCUR_set(dstr, SvCUR(sstr));
4443 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4444 SvPV_set(sstr, NULL);
4449 else if (flags & SV_COW_SHARED_HASH_KEYS
4451 #ifdef PERL_OLD_COPY_ON_WRITE
4452 ( sflags & SVf_IsCOW
4453 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4454 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4455 && SvTYPE(sstr) >= SVt_PVIV && len
4458 #elif defined(PERL_NEW_COPY_ON_WRITE)
4461 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4462 /* If this is a regular (non-hek) COW, only so
4463 many COW "copies" are possible. */
4464 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4465 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4466 && !(SvFLAGS(dstr) & SVf_BREAK)
4467 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4468 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4472 && !(SvFLAGS(dstr) & SVf_BREAK)
4475 /* Either it's a shared hash key, or it's suitable for
4478 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4483 if (!(sflags & SVf_IsCOW)) {
4485 # ifdef PERL_OLD_COPY_ON_WRITE
4486 /* Make the source SV into a loop of 1.
4487 (about to become 2) */
4488 SV_COW_NEXT_SV_SET(sstr, sstr);
4490 CowREFCNT(sstr) = 0;
4494 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4500 # ifdef PERL_OLD_COPY_ON_WRITE
4501 assert (SvTYPE(dstr) >= SVt_PVIV);
4502 /* SvIsCOW_normal */
4503 /* splice us in between source and next-after-source. */
4504 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4505 SV_COW_NEXT_SV_SET(sstr, dstr);
4507 if (sflags & SVf_IsCOW) {
4512 SvPV_set(dstr, SvPVX_mutable(sstr));
4517 /* SvIsCOW_shared_hash */
4518 DEBUG_C(PerlIO_printf(Perl_debug_log,
4519 "Copy on write: Sharing hash\n"));
4521 assert (SvTYPE(dstr) >= SVt_PV);
4523 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4525 SvLEN_set(dstr, len);
4526 SvCUR_set(dstr, cur);
4529 /* Failed the swipe test, and we cannot do copy-on-write either.
4530 Have to copy the string. */
4531 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4532 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4533 SvCUR_set(dstr, cur);
4534 *SvEND(dstr) = '\0';
4536 if (sflags & SVp_NOK) {
4537 SvNV_set(dstr, SvNVX(sstr));
4539 if (sflags & SVp_IOK) {
4540 SvIV_set(dstr, SvIVX(sstr));
4541 /* Must do this otherwise some other overloaded use of 0x80000000
4542 gets confused. I guess SVpbm_VALID */
4543 if (sflags & SVf_IVisUV)
4546 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4548 const MAGIC * const smg = SvVSTRING_mg(sstr);
4550 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4551 smg->mg_ptr, smg->mg_len);
4552 SvRMAGICAL_on(dstr);
4556 else if (sflags & (SVp_IOK|SVp_NOK)) {
4557 (void)SvOK_off(dstr);
4558 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4559 if (sflags & SVp_IOK) {
4560 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4561 SvIV_set(dstr, SvIVX(sstr));
4563 if (sflags & SVp_NOK) {
4564 SvNV_set(dstr, SvNVX(sstr));
4568 if (isGV_with_GP(sstr)) {
4569 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4572 (void)SvOK_off(dstr);
4574 if (SvTAINTED(sstr))
4579 =for apidoc sv_setsv_mg
4581 Like C<sv_setsv>, but also handles 'set' magic.
4587 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4589 PERL_ARGS_ASSERT_SV_SETSV_MG;
4591 sv_setsv(dstr,sstr);
4596 # ifdef PERL_OLD_COPY_ON_WRITE
4597 # define SVt_COW SVt_PVIV
4599 # define SVt_COW SVt_PV
4602 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4604 STRLEN cur = SvCUR(sstr);
4605 STRLEN len = SvLEN(sstr);
4607 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4608 const bool already = cBOOL(SvIsCOW(sstr));
4611 PERL_ARGS_ASSERT_SV_SETSV_COW;
4614 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4615 (void*)sstr, (void*)dstr);
4622 if (SvTHINKFIRST(dstr))
4623 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4624 else if (SvPVX_const(dstr))
4625 Safefree(SvPVX_mutable(dstr));
4629 SvUPGRADE(dstr, SVt_COW);
4631 assert (SvPOK(sstr));
4632 assert (SvPOKp(sstr));
4633 # ifdef PERL_OLD_COPY_ON_WRITE
4634 assert (!SvIOK(sstr));
4635 assert (!SvIOKp(sstr));
4636 assert (!SvNOK(sstr));
4637 assert (!SvNOKp(sstr));
4640 if (SvIsCOW(sstr)) {
4642 if (SvLEN(sstr) == 0) {
4643 /* source is a COW shared hash key. */
4644 DEBUG_C(PerlIO_printf(Perl_debug_log,
4645 "Fast copy on write: Sharing hash\n"));
4646 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4649 # ifdef PERL_OLD_COPY_ON_WRITE
4650 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4652 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4653 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4656 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4657 SvUPGRADE(sstr, SVt_COW);
4659 DEBUG_C(PerlIO_printf(Perl_debug_log,
4660 "Fast copy on write: Converting sstr to COW\n"));
4661 # ifdef PERL_OLD_COPY_ON_WRITE
4662 SV_COW_NEXT_SV_SET(dstr, sstr);
4664 CowREFCNT(sstr) = 0;
4667 # ifdef PERL_OLD_COPY_ON_WRITE
4668 SV_COW_NEXT_SV_SET(sstr, dstr);
4670 # ifdef PERL_DEBUG_READONLY_COW
4671 if (already) sv_buf_to_rw(sstr);
4675 new_pv = SvPVX_mutable(sstr);
4679 SvPV_set(dstr, new_pv);
4680 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4683 SvLEN_set(dstr, len);
4684 SvCUR_set(dstr, cur);
4693 =for apidoc sv_setpvn
4695 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4696 The C<len> parameter indicates the number of
4697 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4698 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4704 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4708 PERL_ARGS_ASSERT_SV_SETPVN;
4710 SV_CHECK_THINKFIRST_COW_DROP(sv);
4716 /* len is STRLEN which is unsigned, need to copy to signed */
4719 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4722 SvUPGRADE(sv, SVt_PV);
4724 dptr = SvGROW(sv, len + 1);
4725 Move(ptr,dptr,len,char);
4728 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4730 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4734 =for apidoc sv_setpvn_mg
4736 Like C<sv_setpvn>, but also handles 'set' magic.
4742 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4744 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4746 sv_setpvn(sv,ptr,len);
4751 =for apidoc sv_setpv
4753 Copies a string into an SV. The string must be terminated with a C<NUL>
4755 Does not handle 'set' magic. See C<sv_setpv_mg>.
4761 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4765 PERL_ARGS_ASSERT_SV_SETPV;
4767 SV_CHECK_THINKFIRST_COW_DROP(sv);
4773 SvUPGRADE(sv, SVt_PV);
4775 SvGROW(sv, len + 1);
4776 Move(ptr,SvPVX(sv),len+1,char);
4778 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4780 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4784 =for apidoc sv_setpv_mg
4786 Like C<sv_setpv>, but also handles 'set' magic.
4792 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4794 PERL_ARGS_ASSERT_SV_SETPV_MG;
4801 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4803 PERL_ARGS_ASSERT_SV_SETHEK;
4809 if (HEK_LEN(hek) == HEf_SVKEY) {
4810 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4813 const int flags = HEK_FLAGS(hek);
4814 if (flags & HVhek_WASUTF8) {
4815 STRLEN utf8_len = HEK_LEN(hek);
4816 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4817 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4820 } else if (flags & HVhek_UNSHARED) {
4821 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4824 else SvUTF8_off(sv);
4828 SV_CHECK_THINKFIRST_COW_DROP(sv);
4829 SvUPGRADE(sv, SVt_PV);
4831 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4832 SvCUR_set(sv, HEK_LEN(hek));
4838 else SvUTF8_off(sv);
4846 =for apidoc sv_usepvn_flags
4848 Tells an SV to use C<ptr> to find its string value. Normally the
4849 string is stored inside the SV, but sv_usepvn allows the SV to use an
4850 outside string. The C<ptr> should point to memory that was allocated
4851 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4852 the start of a Newx-ed block of memory, and not a pointer to the
4853 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4854 and not be from a non-Newx memory allocator like C<malloc>. The
4855 string length, C<len>, must be supplied. By default this function
4856 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4857 so that pointer should not be freed or used by the programmer after
4858 giving it to sv_usepvn, and neither should any pointers from "behind"
4859 that pointer (e.g. ptr + 1) be used.
4861 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4862 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
4863 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4864 C<len>, and already meets the requirements for storing in C<SvPVX>).
4870 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4874 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4876 SV_CHECK_THINKFIRST_COW_DROP(sv);
4877 SvUPGRADE(sv, SVt_PV);
4880 if (flags & SV_SMAGIC)
4884 if (SvPVX_const(sv))
4888 if (flags & SV_HAS_TRAILING_NUL)
4889 assert(ptr[len] == '\0');
4892 allocate = (flags & SV_HAS_TRAILING_NUL)
4894 #ifdef Perl_safesysmalloc_size
4897 PERL_STRLEN_ROUNDUP(len + 1);
4899 if (flags & SV_HAS_TRAILING_NUL) {
4900 /* It's long enough - do nothing.
4901 Specifically Perl_newCONSTSUB is relying on this. */
4904 /* Force a move to shake out bugs in callers. */
4905 char *new_ptr = (char*)safemalloc(allocate);
4906 Copy(ptr, new_ptr, len, char);
4907 PoisonFree(ptr,len,char);
4911 ptr = (char*) saferealloc (ptr, allocate);
4914 #ifdef Perl_safesysmalloc_size
4915 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4917 SvLEN_set(sv, allocate);
4921 if (!(flags & SV_HAS_TRAILING_NUL)) {
4924 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4926 if (flags & SV_SMAGIC)
4930 #ifdef PERL_OLD_COPY_ON_WRITE
4931 /* Need to do this *after* making the SV normal, as we need the buffer
4932 pointer to remain valid until after we've copied it. If we let go too early,
4933 another thread could invalidate it by unsharing last of the same hash key
4934 (which it can do by means other than releasing copy-on-write Svs)
4935 or by changing the other copy-on-write SVs in the loop. */
4937 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
4939 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4941 { /* this SV was SvIsCOW_normal(sv) */
4942 /* we need to find the SV pointing to us. */
4943 SV *current = SV_COW_NEXT_SV(after);
4945 if (current == sv) {
4946 /* The SV we point to points back to us (there were only two of us
4948 Hence other SV is no longer copy on write either. */
4950 sv_buf_to_rw(after);
4952 /* We need to follow the pointers around the loop. */
4954 while ((next = SV_COW_NEXT_SV(current)) != sv) {
4957 /* don't loop forever if the structure is bust, and we have
4958 a pointer into a closed loop. */
4959 assert (current != after);
4960 assert (SvPVX_const(current) == pvx);
4962 /* Make the SV before us point to the SV after us. */
4963 SV_COW_NEXT_SV_SET(current, after);
4969 =for apidoc sv_force_normal_flags
4971 Undo various types of fakery on an SV, where fakery means
4972 "more than" a string: if the PV is a shared string, make
4973 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
4974 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
4975 we do the copy, and is also used locally; if this is a
4976 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
4977 then a copy-on-write scalar drops its PV buffer (if any) and becomes
4978 SvPOK_off rather than making a copy. (Used where this
4979 scalar is about to be set to some other value.) In addition,
4980 the C<flags> parameter gets passed to C<sv_unref_flags()>
4981 when unreffing. C<sv_force_normal> calls this function
4982 with flags set to 0.
4984 This function is expected to be used to signal to perl that this SV is
4985 about to be written to, and any extra book-keeping needs to be taken care
4986 of. Hence, it croaks on read-only values.
4992 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
4994 assert(SvIsCOW(sv));
4997 const char * const pvx = SvPVX_const(sv);
4998 const STRLEN len = SvLEN(sv);
4999 const STRLEN cur = SvCUR(sv);
5000 # ifdef PERL_OLD_COPY_ON_WRITE
5001 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5002 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5003 we'll fail an assertion. */
5004 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5008 PerlIO_printf(Perl_debug_log,
5009 "Copy on write: Force normal %ld\n",
5014 # ifdef PERL_NEW_COPY_ON_WRITE
5015 if (len && CowREFCNT(sv) == 0)
5016 /* We own the buffer ourselves. */
5022 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5023 # ifdef PERL_NEW_COPY_ON_WRITE
5024 /* Must do this first, since the macro uses SvPVX. */
5034 if (flags & SV_COW_DROP_PV) {
5035 /* OK, so we don't need to copy our buffer. */
5038 SvGROW(sv, cur + 1);
5039 Move(pvx,SvPVX(sv),cur,char);
5044 # ifdef PERL_OLD_COPY_ON_WRITE
5045 sv_release_COW(sv, pvx, next);
5048 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5055 const char * const pvx = SvPVX_const(sv);
5056 const STRLEN len = SvCUR(sv);
5060 if (flags & SV_COW_DROP_PV) {
5061 /* OK, so we don't need to copy our buffer. */
5064 SvGROW(sv, len + 1);
5065 Move(pvx,SvPVX(sv),len,char);
5068 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5074 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5076 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5079 Perl_croak_no_modify();
5080 else if (SvIsCOW(sv))
5081 S_sv_uncow(aTHX_ sv, flags);
5083 sv_unref_flags(sv, flags);
5084 else if (SvFAKE(sv) && isGV_with_GP(sv))
5085 sv_unglob(sv, flags);
5086 else if (SvFAKE(sv) && isREGEXP(sv)) {
5087 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5088 to sv_unglob. We only need it here, so inline it. */
5089 const bool islv = SvTYPE(sv) == SVt_PVLV;
5090 const svtype new_type =
5091 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5092 SV *const temp = newSV_type(new_type);
5093 regexp *const temp_p = ReANY((REGEXP *)sv);
5095 if (new_type == SVt_PVMG) {
5096 SvMAGIC_set(temp, SvMAGIC(sv));
5097 SvMAGIC_set(sv, NULL);
5098 SvSTASH_set(temp, SvSTASH(sv));
5099 SvSTASH_set(sv, NULL);
5101 if (!islv) SvCUR_set(temp, SvCUR(sv));
5102 /* Remember that SvPVX is in the head, not the body. But
5103 RX_WRAPPED is in the body. */
5104 assert(ReANY((REGEXP *)sv)->mother_re);
5105 /* Their buffer is already owned by someone else. */
5106 if (flags & SV_COW_DROP_PV) {
5107 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5108 zeroed body. For SVt_PVLV, it should have been set to 0
5109 before turning into a regexp. */
5110 assert(!SvLEN(islv ? sv : temp));
5111 sv->sv_u.svu_pv = 0;
5114 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5115 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5119 /* Now swap the rest of the bodies. */
5123 SvFLAGS(sv) &= ~SVTYPEMASK;
5124 SvFLAGS(sv) |= new_type;
5125 SvANY(sv) = SvANY(temp);
5128 SvFLAGS(temp) &= ~(SVTYPEMASK);
5129 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5130 SvANY(temp) = temp_p;
5131 temp->sv_u.svu_rx = (regexp *)temp_p;
5133 SvREFCNT_dec_NN(temp);
5135 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5141 Efficient removal of characters from the beginning of the string buffer.
5142 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5143 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5144 character of the adjusted string. Uses the "OOK hack". On return, only
5145 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5147 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5148 refer to the same chunk of data.
5150 The unfortunate similarity of this function's name to that of Perl's C<chop>
5151 operator is strictly coincidental. This function works from the left;
5152 C<chop> works from the right.
5158 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5169 PERL_ARGS_ASSERT_SV_CHOP;
5171 if (!ptr || !SvPOKp(sv))
5173 delta = ptr - SvPVX_const(sv);
5175 /* Nothing to do. */
5178 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5179 if (delta > max_delta)
5180 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5181 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5182 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5183 SV_CHECK_THINKFIRST(sv);
5184 SvPOK_only_UTF8(sv);
5187 if (!SvLEN(sv)) { /* make copy of shared string */
5188 const char *pvx = SvPVX_const(sv);
5189 const STRLEN len = SvCUR(sv);
5190 SvGROW(sv, len + 1);
5191 Move(pvx,SvPVX(sv),len,char);
5197 SvOOK_offset(sv, old_delta);
5199 SvLEN_set(sv, SvLEN(sv) - delta);
5200 SvCUR_set(sv, SvCUR(sv) - delta);
5201 SvPV_set(sv, SvPVX(sv) + delta);
5203 p = (U8 *)SvPVX_const(sv);
5206 /* how many bytes were evacuated? we will fill them with sentinel
5207 bytes, except for the part holding the new offset of course. */
5210 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5212 assert(evacn <= delta + old_delta);
5216 /* This sets 'delta' to the accumulated value of all deltas so far */
5220 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5221 * the string; otherwise store a 0 byte there and store 'delta' just prior
5222 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5223 * portion of the chopped part of the string */
5224 if (delta < 0x100) {
5228 p -= sizeof(STRLEN);
5229 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5233 /* Fill the preceding buffer with sentinals to verify that no-one is
5243 =for apidoc sv_catpvn
5245 Concatenates the string onto the end of the string which is in the SV. The
5246 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5247 status set, then the bytes appended should be valid UTF-8.
5248 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5250 =for apidoc sv_catpvn_flags
5252 Concatenates the string onto the end of the string which is in the SV. The
5253 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5254 status set, then the bytes appended should be valid UTF-8.
5255 If C<flags> has the C<SV_SMAGIC> bit set, will
5256 C<mg_set> on C<dsv> afterwards if appropriate.
5257 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5258 in terms of this function.
5264 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5267 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5269 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5270 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5272 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5273 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5274 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5277 else SvGROW(dsv, dlen + slen + 1);
5279 sstr = SvPVX_const(dsv);
5280 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5281 SvCUR_set(dsv, SvCUR(dsv) + slen);
5284 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5285 const char * const send = sstr + slen;
5288 /* Something this code does not account for, which I think is
5289 impossible; it would require the same pv to be treated as
5290 bytes *and* utf8, which would indicate a bug elsewhere. */
5291 assert(sstr != dstr);
5293 SvGROW(dsv, dlen + slen * 2 + 1);
5294 d = (U8 *)SvPVX(dsv) + dlen;
5296 while (sstr < send) {
5297 append_utf8_from_native_byte(*sstr, &d);
5300 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5303 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5305 if (flags & SV_SMAGIC)
5310 =for apidoc sv_catsv
5312 Concatenates the string from SV C<ssv> onto the end of the string in SV
5313 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5314 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5317 =for apidoc sv_catsv_flags
5319 Concatenates the string from SV C<ssv> onto the end of the string in SV
5320 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5321 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5322 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5323 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5324 and C<sv_catsv_mg> are implemented in terms of this function.
5329 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5331 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5335 const char *spv = SvPV_flags_const(ssv, slen, flags);
5337 if (flags & SV_GMAGIC)
5339 sv_catpvn_flags(dsv, spv, slen,
5340 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5341 if (flags & SV_SMAGIC)
5348 =for apidoc sv_catpv
5350 Concatenates the C<NUL>-terminated string onto the end of the string which is
5352 If the SV has the UTF-8 status set, then the bytes appended should be
5353 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5358 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5364 PERL_ARGS_ASSERT_SV_CATPV;
5368 junk = SvPV_force(sv, tlen);
5370 SvGROW(sv, tlen + len + 1);
5372 ptr = SvPVX_const(sv);
5373 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5374 SvCUR_set(sv, SvCUR(sv) + len);
5375 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5380 =for apidoc sv_catpv_flags
5382 Concatenates the C<NUL>-terminated string onto the end of the string which is
5384 If the SV has the UTF-8 status set, then the bytes appended should
5385 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5386 on the modified SV if appropriate.
5392 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5394 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5395 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5399 =for apidoc sv_catpv_mg
5401 Like C<sv_catpv>, but also handles 'set' magic.
5407 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5409 PERL_ARGS_ASSERT_SV_CATPV_MG;
5418 Creates a new SV. A non-zero C<len> parameter indicates the number of
5419 bytes of preallocated string space the SV should have. An extra byte for a
5420 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5421 space is allocated.) The reference count for the new SV is set to 1.
5423 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5424 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5425 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5426 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5427 modules supporting older perls.
5433 Perl_newSV(pTHX_ const STRLEN len)
5439 sv_upgrade(sv, SVt_PV);
5440 SvGROW(sv, len + 1);
5445 =for apidoc sv_magicext
5447 Adds magic to an SV, upgrading it if necessary. Applies the
5448 supplied vtable and returns a pointer to the magic added.
5450 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5451 In particular, you can add magic to SvREADONLY SVs, and add more than
5452 one instance of the same 'how'.
5454 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5455 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5456 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5457 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5459 (This is now used as a subroutine by C<sv_magic>.)
5464 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5465 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5469 PERL_ARGS_ASSERT_SV_MAGICEXT;
5471 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5473 SvUPGRADE(sv, SVt_PVMG);
5474 Newxz(mg, 1, MAGIC);
5475 mg->mg_moremagic = SvMAGIC(sv);
5476 SvMAGIC_set(sv, mg);
5478 /* Sometimes a magic contains a reference loop, where the sv and
5479 object refer to each other. To prevent a reference loop that
5480 would prevent such objects being freed, we look for such loops
5481 and if we find one we avoid incrementing the object refcount.
5483 Note we cannot do this to avoid self-tie loops as intervening RV must
5484 have its REFCNT incremented to keep it in existence.
5487 if (!obj || obj == sv ||
5488 how == PERL_MAGIC_arylen ||
5489 how == PERL_MAGIC_symtab ||
5490 (SvTYPE(obj) == SVt_PVGV &&
5491 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5492 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5493 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5498 mg->mg_obj = SvREFCNT_inc_simple(obj);
5499 mg->mg_flags |= MGf_REFCOUNTED;
5502 /* Normal self-ties simply pass a null object, and instead of
5503 using mg_obj directly, use the SvTIED_obj macro to produce a
5504 new RV as needed. For glob "self-ties", we are tieing the PVIO
5505 with an RV obj pointing to the glob containing the PVIO. In
5506 this case, to avoid a reference loop, we need to weaken the
5510 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5511 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5517 mg->mg_len = namlen;
5520 mg->mg_ptr = savepvn(name, namlen);
5521 else if (namlen == HEf_SVKEY) {
5522 /* Yes, this is casting away const. This is only for the case of
5523 HEf_SVKEY. I think we need to document this aberation of the
5524 constness of the API, rather than making name non-const, as
5525 that change propagating outwards a long way. */
5526 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5528 mg->mg_ptr = (char *) name;
5530 mg->mg_virtual = (MGVTBL *) vtable;
5537 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5539 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5540 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5541 /* This sv is only a delegate. //g magic must be attached to
5546 #ifdef PERL_OLD_COPY_ON_WRITE
5548 sv_force_normal_flags(sv, 0);
5550 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5551 &PL_vtbl_mglob, 0, 0);
5555 =for apidoc sv_magic
5557 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5558 necessary, then adds a new magic item of type C<how> to the head of the
5561 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5562 handling of the C<name> and C<namlen> arguments.
5564 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5565 to add more than one instance of the same 'how'.
5571 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5572 const char *const name, const I32 namlen)
5574 const MGVTBL *vtable;
5577 unsigned int vtable_index;
5579 PERL_ARGS_ASSERT_SV_MAGIC;
5581 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5582 || ((flags = PL_magic_data[how]),
5583 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5584 > magic_vtable_max))
5585 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5587 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5588 Useful for attaching extension internal data to perl vars.
5589 Note that multiple extensions may clash if magical scalars
5590 etc holding private data from one are passed to another. */
5592 vtable = (vtable_index == magic_vtable_max)
5593 ? NULL : PL_magic_vtables + vtable_index;
5595 #ifdef PERL_OLD_COPY_ON_WRITE
5597 sv_force_normal_flags(sv, 0);
5599 if (SvREADONLY(sv)) {
5601 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5604 Perl_croak_no_modify();
5607 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5608 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5609 /* sv_magic() refuses to add a magic of the same 'how' as an
5612 if (how == PERL_MAGIC_taint)
5618 /* Force pos to be stored as characters, not bytes. */
5619 if (SvMAGICAL(sv) && DO_UTF8(sv)
5620 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5622 && mg->mg_flags & MGf_BYTES) {
5623 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5625 mg->mg_flags &= ~MGf_BYTES;
5628 /* Rest of work is done else where */
5629 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5632 case PERL_MAGIC_taint:
5635 case PERL_MAGIC_ext:
5636 case PERL_MAGIC_dbfile:
5643 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5650 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5652 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5653 for (mg = *mgp; mg; mg = *mgp) {
5654 const MGVTBL* const virt = mg->mg_virtual;
5655 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5656 *mgp = mg->mg_moremagic;
5657 if (virt && virt->svt_free)
5658 virt->svt_free(aTHX_ sv, mg);
5659 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5661 Safefree(mg->mg_ptr);
5662 else if (mg->mg_len == HEf_SVKEY)
5663 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5664 else if (mg->mg_type == PERL_MAGIC_utf8)
5665 Safefree(mg->mg_ptr);
5667 if (mg->mg_flags & MGf_REFCOUNTED)
5668 SvREFCNT_dec(mg->mg_obj);
5672 mgp = &mg->mg_moremagic;
5675 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5676 mg_magical(sv); /* else fix the flags now */
5680 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5686 =for apidoc sv_unmagic
5688 Removes all magic of type C<type> from an SV.
5694 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5696 PERL_ARGS_ASSERT_SV_UNMAGIC;
5697 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5701 =for apidoc sv_unmagicext
5703 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5709 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5711 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5712 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5716 =for apidoc sv_rvweaken
5718 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5719 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5720 push a back-reference to this RV onto the array of backreferences
5721 associated with that magic. If the RV is magical, set magic will be
5722 called after the RV is cleared.
5728 Perl_sv_rvweaken(pTHX_ SV *const sv)
5732 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5734 if (!SvOK(sv)) /* let undefs pass */
5737 Perl_croak(aTHX_ "Can't weaken a nonreference");
5738 else if (SvWEAKREF(sv)) {
5739 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5742 else if (SvREADONLY(sv)) croak_no_modify();
5744 Perl_sv_add_backref(aTHX_ tsv, sv);
5746 SvREFCNT_dec_NN(tsv);
5750 /* Give tsv backref magic if it hasn't already got it, then push a
5751 * back-reference to sv onto the array associated with the backref magic.
5753 * As an optimisation, if there's only one backref and it's not an AV,
5754 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5755 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5759 /* A discussion about the backreferences array and its refcount:
5761 * The AV holding the backreferences is pointed to either as the mg_obj of
5762 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5763 * xhv_backreferences field. The array is created with a refcount
5764 * of 2. This means that if during global destruction the array gets
5765 * picked on before its parent to have its refcount decremented by the
5766 * random zapper, it won't actually be freed, meaning it's still there for
5767 * when its parent gets freed.
5769 * When the parent SV is freed, the extra ref is killed by
5770 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5771 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5773 * When a single backref SV is stored directly, it is not reference
5778 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5784 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5786 /* find slot to store array or singleton backref */
5788 if (SvTYPE(tsv) == SVt_PVHV) {
5789 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5792 mg = mg_find(tsv, PERL_MAGIC_backref);
5794 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5795 svp = &(mg->mg_obj);
5798 /* create or retrieve the array */
5800 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5801 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5805 mg->mg_flags |= MGf_REFCOUNTED;
5808 SvREFCNT_inc_simple_void_NN(av);
5809 /* av now has a refcnt of 2; see discussion above */
5810 av_extend(av, *svp ? 2 : 1);
5812 /* move single existing backref to the array */
5813 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5818 av = MUTABLE_AV(*svp);
5820 /* optimisation: store single backref directly in HvAUX or mg_obj */
5824 assert(SvTYPE(av) == SVt_PVAV);
5825 if (AvFILLp(av) >= AvMAX(av)) {
5826 av_extend(av, AvFILLp(av)+1);
5829 /* push new backref */
5830 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5833 /* delete a back-reference to ourselves from the backref magic associated
5834 * with the SV we point to.
5838 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5842 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5844 if (SvTYPE(tsv) == SVt_PVHV) {
5846 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5848 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5849 /* It's possible for the the last (strong) reference to tsv to have
5850 become freed *before* the last thing holding a weak reference.
5851 If both survive longer than the backreferences array, then when
5852 the referent's reference count drops to 0 and it is freed, it's
5853 not able to chase the backreferences, so they aren't NULLed.
5855 For example, a CV holds a weak reference to its stash. If both the
5856 CV and the stash survive longer than the backreferences array,
5857 and the CV gets picked for the SvBREAK() treatment first,
5858 *and* it turns out that the stash is only being kept alive because
5859 of an our variable in the pad of the CV, then midway during CV
5860 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5861 It ends up pointing to the freed HV. Hence it's chased in here, and
5862 if this block wasn't here, it would hit the !svp panic just below.
5864 I don't believe that "better" destruction ordering is going to help
5865 here - during global destruction there's always going to be the
5866 chance that something goes out of order. We've tried to make it
5867 foolproof before, and it only resulted in evolutionary pressure on
5868 fools. Which made us look foolish for our hubris. :-(
5874 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5875 svp = mg ? &(mg->mg_obj) : NULL;
5879 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5881 /* It's possible that sv is being freed recursively part way through the
5882 freeing of tsv. If this happens, the backreferences array of tsv has
5883 already been freed, and so svp will be NULL. If this is the case,
5884 we should not panic. Instead, nothing needs doing, so return. */
5885 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
5887 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
5888 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
5891 if (SvTYPE(*svp) == SVt_PVAV) {
5895 AV * const av = (AV*)*svp;
5897 assert(!SvIS_FREED(av));
5901 /* for an SV with N weak references to it, if all those
5902 * weak refs are deleted, then sv_del_backref will be called
5903 * N times and O(N^2) compares will be done within the backref
5904 * array. To ameliorate this potential slowness, we:
5905 * 1) make sure this code is as tight as possible;
5906 * 2) when looking for SV, look for it at both the head and tail of the
5907 * array first before searching the rest, since some create/destroy
5908 * patterns will cause the backrefs to be freed in order.
5915 SV **p = &svp[fill];
5916 SV *const topsv = *p;
5923 /* We weren't the last entry.
5924 An unordered list has this property that you
5925 can take the last element off the end to fill
5926 the hole, and it's still an unordered list :-)
5932 break; /* should only be one */
5939 AvFILLp(av) = fill-1;
5941 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
5942 /* freed AV; skip */
5945 /* optimisation: only a single backref, stored directly */
5947 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
5948 (void*)*svp, (void*)sv);
5955 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
5961 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
5966 /* after multiple passes through Perl_sv_clean_all() for a thingy
5967 * that has badly leaked, the backref array may have gotten freed,
5968 * since we only protect it against 1 round of cleanup */
5969 if (SvIS_FREED(av)) {
5970 if (PL_in_clean_all) /* All is fair */
5973 "panic: magic_killbackrefs (freed backref AV/SV)");
5977 is_array = (SvTYPE(av) == SVt_PVAV);
5979 assert(!SvIS_FREED(av));
5982 last = svp + AvFILLp(av);
5985 /* optimisation: only a single backref, stored directly */
5991 while (svp <= last) {
5993 SV *const referrer = *svp;
5994 if (SvWEAKREF(referrer)) {
5995 /* XXX Should we check that it hasn't changed? */
5996 assert(SvROK(referrer));
5997 SvRV_set(referrer, 0);
5999 SvWEAKREF_off(referrer);
6000 SvSETMAGIC(referrer);
6001 } else if (SvTYPE(referrer) == SVt_PVGV ||
6002 SvTYPE(referrer) == SVt_PVLV) {
6003 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6004 /* You lookin' at me? */
6005 assert(GvSTASH(referrer));
6006 assert(GvSTASH(referrer) == (const HV *)sv);
6007 GvSTASH(referrer) = 0;
6008 } else if (SvTYPE(referrer) == SVt_PVCV ||
6009 SvTYPE(referrer) == SVt_PVFM) {
6010 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6011 /* You lookin' at me? */
6012 assert(CvSTASH(referrer));
6013 assert(CvSTASH(referrer) == (const HV *)sv);
6014 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6017 assert(SvTYPE(sv) == SVt_PVGV);
6018 /* You lookin' at me? */
6019 assert(CvGV(referrer));
6020 assert(CvGV(referrer) == (const GV *)sv);
6021 anonymise_cv_maybe(MUTABLE_GV(sv),
6022 MUTABLE_CV(referrer));
6027 "panic: magic_killbackrefs (flags=%"UVxf")",
6028 (UV)SvFLAGS(referrer));
6039 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6045 =for apidoc sv_insert
6047 Inserts a string at the specified offset/length within the SV. Similar to
6048 the Perl substr() function. Handles get magic.
6050 =for apidoc sv_insert_flags
6052 Same as C<sv_insert>, but the extra C<flags> are passed to the
6053 C<SvPV_force_flags> that applies to C<bigstr>.
6059 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6065 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6068 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6071 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6072 SvPV_force_flags(bigstr, curlen, flags);
6073 (void)SvPOK_only_UTF8(bigstr);
6074 if (offset + len > curlen) {
6075 SvGROW(bigstr, offset+len+1);
6076 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6077 SvCUR_set(bigstr, offset+len);
6081 i = littlelen - len;
6082 if (i > 0) { /* string might grow */
6083 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6084 mid = big + offset + len;
6085 midend = bigend = big + SvCUR(bigstr);
6088 while (midend > mid) /* shove everything down */
6089 *--bigend = *--midend;
6090 Move(little,big+offset,littlelen,char);
6091 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6096 Move(little,SvPVX(bigstr)+offset,len,char);
6101 big = SvPVX(bigstr);
6104 bigend = big + SvCUR(bigstr);
6106 if (midend > bigend)
6107 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6110 if (mid - big > bigend - midend) { /* faster to shorten from end */
6112 Move(little, mid, littlelen,char);
6115 i = bigend - midend;
6117 Move(midend, mid, i,char);
6121 SvCUR_set(bigstr, mid - big);
6123 else if ((i = mid - big)) { /* faster from front */
6124 midend -= littlelen;
6126 Move(big, midend - i, i, char);
6127 sv_chop(bigstr,midend-i);
6129 Move(little, mid, littlelen,char);
6131 else if (littlelen) {
6132 midend -= littlelen;
6133 sv_chop(bigstr,midend);
6134 Move(little,midend,littlelen,char);
6137 sv_chop(bigstr,midend);
6143 =for apidoc sv_replace
6145 Make the first argument a copy of the second, then delete the original.
6146 The target SV physically takes over ownership of the body of the source SV
6147 and inherits its flags; however, the target keeps any magic it owns,
6148 and any magic in the source is discarded.
6149 Note that this is a rather specialist SV copying operation; most of the
6150 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6156 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6158 const U32 refcnt = SvREFCNT(sv);
6160 PERL_ARGS_ASSERT_SV_REPLACE;
6162 SV_CHECK_THINKFIRST_COW_DROP(sv);
6163 if (SvREFCNT(nsv) != 1) {
6164 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6165 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6167 if (SvMAGICAL(sv)) {
6171 sv_upgrade(nsv, SVt_PVMG);
6172 SvMAGIC_set(nsv, SvMAGIC(sv));
6173 SvFLAGS(nsv) |= SvMAGICAL(sv);
6175 SvMAGIC_set(sv, NULL);
6179 assert(!SvREFCNT(sv));
6180 #ifdef DEBUG_LEAKING_SCALARS
6181 sv->sv_flags = nsv->sv_flags;
6182 sv->sv_any = nsv->sv_any;
6183 sv->sv_refcnt = nsv->sv_refcnt;
6184 sv->sv_u = nsv->sv_u;
6186 StructCopy(nsv,sv,SV);
6188 if(SvTYPE(sv) == SVt_IV) {
6190 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6194 #ifdef PERL_OLD_COPY_ON_WRITE
6195 if (SvIsCOW_normal(nsv)) {
6196 /* We need to follow the pointers around the loop to make the
6197 previous SV point to sv, rather than nsv. */
6200 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6203 assert(SvPVX_const(current) == SvPVX_const(nsv));
6205 /* Make the SV before us point to the SV after us. */
6207 PerlIO_printf(Perl_debug_log, "previous is\n");
6209 PerlIO_printf(Perl_debug_log,
6210 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6211 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6213 SV_COW_NEXT_SV_SET(current, sv);
6216 SvREFCNT(sv) = refcnt;
6217 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6222 /* We're about to free a GV which has a CV that refers back to us.
6223 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6227 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6232 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6235 assert(SvREFCNT(gv) == 0);
6236 assert(isGV(gv) && isGV_with_GP(gv));
6238 assert(!CvANON(cv));
6239 assert(CvGV(cv) == gv);
6240 assert(!CvNAMED(cv));
6242 /* will the CV shortly be freed by gp_free() ? */
6243 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6244 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6248 /* if not, anonymise: */
6249 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6250 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6251 : newSVpvn_flags( "__ANON__", 8, 0 );
6252 sv_catpvs(gvname, "::__ANON__");
6253 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6254 SvREFCNT_dec_NN(gvname);
6258 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6263 =for apidoc sv_clear
6265 Clear an SV: call any destructors, free up any memory used by the body,
6266 and free the body itself. The SV's head is I<not> freed, although
6267 its type is set to all 1's so that it won't inadvertently be assumed
6268 to be live during global destruction etc.
6269 This function should only be called when REFCNT is zero. Most of the time
6270 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6277 Perl_sv_clear(pTHX_ SV *const orig_sv)
6282 const struct body_details *sv_type_details;
6288 PERL_ARGS_ASSERT_SV_CLEAR;
6290 /* within this loop, sv is the SV currently being freed, and
6291 * iter_sv is the most recent AV or whatever that's being iterated
6292 * over to provide more SVs */
6298 assert(SvREFCNT(sv) == 0);
6299 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6301 if (type <= SVt_IV) {
6302 /* See the comment in sv.h about the collusion between this
6303 * early return and the overloading of the NULL slots in the
6307 SvFLAGS(sv) &= SVf_BREAK;
6308 SvFLAGS(sv) |= SVTYPEMASK;
6312 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6314 if (type >= SVt_PVMG) {
6316 if (!curse(sv, 1)) goto get_next_sv;
6317 type = SvTYPE(sv); /* destructor may have changed it */
6319 /* Free back-references before magic, in case the magic calls
6320 * Perl code that has weak references to sv. */
6321 if (type == SVt_PVHV) {
6322 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6326 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6327 SvREFCNT_dec(SvOURSTASH(sv));
6329 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6330 assert(!SvMAGICAL(sv));
6331 } else if (SvMAGIC(sv)) {
6332 /* Free back-references before other types of magic. */
6333 sv_unmagic(sv, PERL_MAGIC_backref);
6337 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6338 SvREFCNT_dec(SvSTASH(sv));
6341 /* case SVt_INVLIST: */
6344 IoIFP(sv) != PerlIO_stdin() &&
6345 IoIFP(sv) != PerlIO_stdout() &&
6346 IoIFP(sv) != PerlIO_stderr() &&
6347 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6349 io_close(MUTABLE_IO(sv), FALSE);
6351 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6352 PerlDir_close(IoDIRP(sv));
6353 IoDIRP(sv) = (DIR*)NULL;
6354 Safefree(IoTOP_NAME(sv));
6355 Safefree(IoFMT_NAME(sv));
6356 Safefree(IoBOTTOM_NAME(sv));
6357 if ((const GV *)sv == PL_statgv)
6361 /* FIXME for plugins */
6363 pregfree2((REGEXP*) sv);
6367 cv_undef(MUTABLE_CV(sv));
6368 /* If we're in a stash, we don't own a reference to it.
6369 * However it does have a back reference to us, which needs to
6371 if ((stash = CvSTASH(sv)))
6372 sv_del_backref(MUTABLE_SV(stash), sv);
6375 if (PL_last_swash_hv == (const HV *)sv) {
6376 PL_last_swash_hv = NULL;
6378 if (HvTOTALKEYS((HV*)sv) > 0) {
6380 /* this statement should match the one at the beginning of
6381 * hv_undef_flags() */
6382 if ( PL_phase != PERL_PHASE_DESTRUCT
6383 && (name = HvNAME((HV*)sv)))
6385 if (PL_stashcache) {
6386 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6388 (void)hv_deletehek(PL_stashcache,
6389 HvNAME_HEK((HV*)sv), G_DISCARD);
6391 hv_name_set((HV*)sv, NULL, 0, 0);
6394 /* save old iter_sv in unused SvSTASH field */
6395 assert(!SvOBJECT(sv));
6396 SvSTASH(sv) = (HV*)iter_sv;
6399 /* save old hash_index in unused SvMAGIC field */
6400 assert(!SvMAGICAL(sv));
6401 assert(!SvMAGIC(sv));
6402 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6405 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6406 goto get_next_sv; /* process this new sv */
6408 /* free empty hash */
6409 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6410 assert(!HvARRAY((HV*)sv));
6414 AV* av = MUTABLE_AV(sv);
6415 if (PL_comppad == av) {
6419 if (AvREAL(av) && AvFILLp(av) > -1) {
6420 next_sv = AvARRAY(av)[AvFILLp(av)--];
6421 /* save old iter_sv in top-most slot of AV,
6422 * and pray that it doesn't get wiped in the meantime */
6423 AvARRAY(av)[AvMAX(av)] = iter_sv;
6425 goto get_next_sv; /* process this new sv */
6427 Safefree(AvALLOC(av));
6432 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6433 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6434 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6435 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6437 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6438 SvREFCNT_dec(LvTARG(sv));
6439 if (isREGEXP(sv)) goto freeregexp;
6441 if (isGV_with_GP(sv)) {
6442 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6443 && HvENAME_get(stash))
6444 mro_method_changed_in(stash);
6445 gp_free(MUTABLE_GV(sv));
6447 unshare_hek(GvNAME_HEK(sv));
6448 /* If we're in a stash, we don't own a reference to it.
6449 * However it does have a back reference to us, which
6450 * needs to be cleared. */
6451 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6452 sv_del_backref(MUTABLE_SV(stash), sv);
6454 /* FIXME. There are probably more unreferenced pointers to SVs
6455 * in the interpreter struct that we should check and tidy in
6456 * a similar fashion to this: */
6457 /* See also S_sv_unglob, which does the same thing. */
6458 if ((const GV *)sv == PL_last_in_gv)
6459 PL_last_in_gv = NULL;
6460 else if ((const GV *)sv == PL_statgv)
6462 else if ((const GV *)sv == PL_stderrgv)
6470 /* Don't bother with SvOOK_off(sv); as we're only going to
6474 SvOOK_offset(sv, offset);
6475 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6476 /* Don't even bother with turning off the OOK flag. */
6481 SV * const target = SvRV(sv);
6483 sv_del_backref(target, sv);
6489 else if (SvPVX_const(sv)
6490 && !(SvTYPE(sv) == SVt_PVIO
6491 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6495 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6499 # ifdef PERL_OLD_COPY_ON_WRITE
6500 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6502 if (CowREFCNT(sv)) {
6510 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6514 # ifdef PERL_OLD_COPY_ON_WRITE
6518 Safefree(SvPVX_mutable(sv));
6522 else if (SvPVX_const(sv) && SvLEN(sv)
6523 && !(SvTYPE(sv) == SVt_PVIO
6524 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6525 Safefree(SvPVX_mutable(sv));
6526 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6527 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6537 SvFLAGS(sv) &= SVf_BREAK;
6538 SvFLAGS(sv) |= SVTYPEMASK;
6540 sv_type_details = bodies_by_type + type;
6541 if (sv_type_details->arena) {
6542 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6543 &PL_body_roots[type]);
6545 else if (sv_type_details->body_size) {
6546 safefree(SvANY(sv));
6550 /* caller is responsible for freeing the head of the original sv */
6551 if (sv != orig_sv && !SvREFCNT(sv))
6554 /* grab and free next sv, if any */
6562 else if (!iter_sv) {
6564 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6565 AV *const av = (AV*)iter_sv;
6566 if (AvFILLp(av) > -1) {
6567 sv = AvARRAY(av)[AvFILLp(av)--];
6569 else { /* no more elements of current AV to free */
6572 /* restore previous value, squirrelled away */
6573 iter_sv = AvARRAY(av)[AvMAX(av)];
6574 Safefree(AvALLOC(av));
6577 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6578 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6579 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6580 /* no more elements of current HV to free */
6583 /* Restore previous values of iter_sv and hash_index,
6584 * squirrelled away */
6585 assert(!SvOBJECT(sv));
6586 iter_sv = (SV*)SvSTASH(sv);
6587 assert(!SvMAGICAL(sv));
6588 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6590 /* perl -DA does not like rubbish in SvMAGIC. */
6594 /* free any remaining detritus from the hash struct */
6595 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6596 assert(!HvARRAY((HV*)sv));
6601 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6605 if (!SvREFCNT(sv)) {
6609 if (--(SvREFCNT(sv)))
6613 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6614 "Attempt to free temp prematurely: SV 0x%"UVxf
6615 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6619 if (SvIMMORTAL(sv)) {
6620 /* make sure SvREFCNT(sv)==0 happens very seldom */
6621 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6630 /* This routine curses the sv itself, not the object referenced by sv. So
6631 sv does not have to be ROK. */
6634 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6635 PERL_ARGS_ASSERT_CURSE;
6636 assert(SvOBJECT(sv));
6638 if (PL_defstash && /* Still have a symbol table? */
6644 stash = SvSTASH(sv);
6645 assert(SvTYPE(stash) == SVt_PVHV);
6646 if (HvNAME(stash)) {
6647 CV* destructor = NULL;
6648 assert (SvOOK(stash));
6649 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6650 if (!destructor || HvMROMETA(stash)->destroy_gen
6651 != PL_sub_generation)
6654 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6655 if (gv) destructor = GvCV(gv);
6656 if (!SvOBJECT(stash))
6659 destructor ? (HV *)destructor : ((HV *)0)+1;
6660 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6664 assert(!destructor || destructor == ((CV *)0)+1
6665 || SvTYPE(destructor) == SVt_PVCV);
6666 if (destructor && destructor != ((CV *)0)+1
6667 /* A constant subroutine can have no side effects, so
6668 don't bother calling it. */
6669 && !CvCONST(destructor)
6670 /* Don't bother calling an empty destructor or one that
6671 returns immediately. */
6672 && (CvISXSUB(destructor)
6673 || (CvSTART(destructor)
6674 && (CvSTART(destructor)->op_next->op_type
6676 && (CvSTART(destructor)->op_next->op_type
6678 || CvSTART(destructor)->op_next->op_next->op_type
6684 SV* const tmpref = newRV(sv);
6685 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6687 PUSHSTACKi(PERLSI_DESTROY);
6692 call_sv(MUTABLE_SV(destructor),
6693 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6697 if(SvREFCNT(tmpref) < 2) {
6698 /* tmpref is not kept alive! */
6700 SvRV_set(tmpref, NULL);
6703 SvREFCNT_dec_NN(tmpref);
6706 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6709 if (check_refcnt && SvREFCNT(sv)) {
6710 if (PL_in_clean_objs)
6712 "DESTROY created new reference to dead object '%"HEKf"'",
6713 HEKfARG(HvNAME_HEK(stash)));
6714 /* DESTROY gave object new lease on life */
6720 HV * const stash = SvSTASH(sv);
6721 /* Curse before freeing the stash, as freeing the stash could cause
6722 a recursive call into S_curse. */
6723 SvOBJECT_off(sv); /* Curse the object. */
6724 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6725 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6731 =for apidoc sv_newref
6733 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6740 Perl_sv_newref(pTHX_ SV *const sv)
6742 PERL_UNUSED_CONTEXT;
6751 Decrement an SV's reference count, and if it drops to zero, call
6752 C<sv_clear> to invoke destructors and free up any memory used by
6753 the body; finally, deallocate the SV's head itself.
6754 Normally called via a wrapper macro C<SvREFCNT_dec>.
6760 Perl_sv_free(pTHX_ SV *const sv)
6766 /* Private helper function for SvREFCNT_dec().
6767 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6770 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6774 PERL_ARGS_ASSERT_SV_FREE2;
6776 if (LIKELY( rc == 1 )) {
6782 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6783 "Attempt to free temp prematurely: SV 0x%"UVxf
6784 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6788 if (SvIMMORTAL(sv)) {
6789 /* make sure SvREFCNT(sv)==0 happens very seldom */
6790 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6794 if (! SvREFCNT(sv)) /* may have have been resurrected */
6799 /* handle exceptional cases */
6803 if (SvFLAGS(sv) & SVf_BREAK)
6804 /* this SV's refcnt has been artificially decremented to
6805 * trigger cleanup */
6807 if (PL_in_clean_all) /* All is fair */
6809 if (SvIMMORTAL(sv)) {
6810 /* make sure SvREFCNT(sv)==0 happens very seldom */
6811 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6814 if (ckWARN_d(WARN_INTERNAL)) {
6815 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6816 Perl_dump_sv_child(aTHX_ sv);
6818 #ifdef DEBUG_LEAKING_SCALARS
6821 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6822 if (PL_warnhook == PERL_WARNHOOK_FATAL
6823 || ckDEAD(packWARN(WARN_INTERNAL))) {
6824 /* Don't let Perl_warner cause us to escape our fate: */
6828 /* This may not return: */
6829 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6830 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6831 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6834 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6844 Returns the length of the string in the SV. Handles magic and type
6845 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6846 gives raw access to the xpv_cur slot.
6852 Perl_sv_len(pTHX_ SV *const sv)
6859 (void)SvPV_const(sv, len);
6864 =for apidoc sv_len_utf8
6866 Returns the number of characters in the string in an SV, counting wide
6867 UTF-8 bytes as a single character. Handles magic and type coercion.
6873 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6874 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6875 * (Note that the mg_len is not the length of the mg_ptr field.
6876 * This allows the cache to store the character length of the string without
6877 * needing to malloc() extra storage to attach to the mg_ptr.)
6882 Perl_sv_len_utf8(pTHX_ SV *const sv)
6888 return sv_len_utf8_nomg(sv);
6892 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6895 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6897 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6899 if (PL_utf8cache && SvUTF8(sv)) {
6901 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6903 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6904 if (mg->mg_len != -1)
6907 /* We can use the offset cache for a headstart.
6908 The longer value is stored in the first pair. */
6909 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6911 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
6915 if (PL_utf8cache < 0) {
6916 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
6917 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
6921 ulen = Perl_utf8_length(aTHX_ s, s + len);
6922 utf8_mg_len_cache_update(sv, &mg, ulen);
6926 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
6929 /* Walk forwards to find the byte corresponding to the passed in UTF-8
6932 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
6933 STRLEN *const uoffset_p, bool *const at_end)
6935 const U8 *s = start;
6936 STRLEN uoffset = *uoffset_p;
6938 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
6940 while (s < send && uoffset) {
6947 else if (s > send) {
6949 /* This is the existing behaviour. Possibly it should be a croak, as
6950 it's actually a bounds error */
6953 *uoffset_p -= uoffset;
6957 /* Given the length of the string in both bytes and UTF-8 characters, decide
6958 whether to walk forwards or backwards to find the byte corresponding to
6959 the passed in UTF-8 offset. */
6961 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
6962 STRLEN uoffset, const STRLEN uend)
6964 STRLEN backw = uend - uoffset;
6966 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
6968 if (uoffset < 2 * backw) {
6969 /* The assumption is that going forwards is twice the speed of going
6970 forward (that's where the 2 * backw comes from).
6971 (The real figure of course depends on the UTF-8 data.) */
6972 const U8 *s = start;
6974 while (s < send && uoffset--)
6984 while (UTF8_IS_CONTINUATION(*send))
6987 return send - start;
6990 /* For the string representation of the given scalar, find the byte
6991 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
6992 give another position in the string, *before* the sought offset, which
6993 (which is always true, as 0, 0 is a valid pair of positions), which should
6994 help reduce the amount of linear searching.
6995 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
6996 will be used to reduce the amount of linear searching. The cache will be
6997 created if necessary, and the found value offered to it for update. */
6999 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7000 const U8 *const send, STRLEN uoffset,
7001 STRLEN uoffset0, STRLEN boffset0)
7003 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7005 bool at_end = FALSE;
7007 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7009 assert (uoffset >= uoffset0);
7014 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7016 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7017 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7018 if ((*mgp)->mg_ptr) {
7019 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7020 if (cache[0] == uoffset) {
7021 /* An exact match. */
7024 if (cache[2] == uoffset) {
7025 /* An exact match. */
7029 if (cache[0] < uoffset) {
7030 /* The cache already knows part of the way. */
7031 if (cache[0] > uoffset0) {
7032 /* The cache knows more than the passed in pair */
7033 uoffset0 = cache[0];
7034 boffset0 = cache[1];
7036 if ((*mgp)->mg_len != -1) {
7037 /* And we know the end too. */
7039 + sv_pos_u2b_midway(start + boffset0, send,
7041 (*mgp)->mg_len - uoffset0);
7043 uoffset -= uoffset0;
7045 + sv_pos_u2b_forwards(start + boffset0,
7046 send, &uoffset, &at_end);
7047 uoffset += uoffset0;
7050 else if (cache[2] < uoffset) {
7051 /* We're between the two cache entries. */
7052 if (cache[2] > uoffset0) {
7053 /* and the cache knows more than the passed in pair */
7054 uoffset0 = cache[2];
7055 boffset0 = cache[3];
7059 + sv_pos_u2b_midway(start + boffset0,
7062 cache[0] - uoffset0);
7065 + sv_pos_u2b_midway(start + boffset0,
7068 cache[2] - uoffset0);
7072 else if ((*mgp)->mg_len != -1) {
7073 /* If we can take advantage of a passed in offset, do so. */
7074 /* In fact, offset0 is either 0, or less than offset, so don't
7075 need to worry about the other possibility. */
7077 + sv_pos_u2b_midway(start + boffset0, send,
7079 (*mgp)->mg_len - uoffset0);
7084 if (!found || PL_utf8cache < 0) {
7085 STRLEN real_boffset;
7086 uoffset -= uoffset0;
7087 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7088 send, &uoffset, &at_end);
7089 uoffset += uoffset0;
7091 if (found && PL_utf8cache < 0)
7092 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7094 boffset = real_boffset;
7097 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7099 utf8_mg_len_cache_update(sv, mgp, uoffset);
7101 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7108 =for apidoc sv_pos_u2b_flags
7110 Converts the offset from a count of UTF-8 chars from
7111 the start of the string, to a count of the equivalent number of bytes; if
7112 lenp is non-zero, it does the same to lenp, but this time starting from
7113 the offset, rather than from the start
7114 of the string. Handles type coercion.
7115 I<flags> is passed to C<SvPV_flags>, and usually should be
7116 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7122 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7123 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7124 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7129 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7136 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7138 start = (U8*)SvPV_flags(sv, len, flags);
7140 const U8 * const send = start + len;
7142 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7145 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7146 is 0, and *lenp is already set to that. */) {
7147 /* Convert the relative offset to absolute. */
7148 const STRLEN uoffset2 = uoffset + *lenp;
7149 const STRLEN boffset2
7150 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7151 uoffset, boffset) - boffset;
7165 =for apidoc sv_pos_u2b
7167 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7168 the start of the string, to a count of the equivalent number of bytes; if
7169 lenp is non-zero, it does the same to lenp, but this time starting from
7170 the offset, rather than from the start of the string. Handles magic and
7173 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7180 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7181 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7182 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7186 /* This function is subject to size and sign problems */
7189 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7191 PERL_ARGS_ASSERT_SV_POS_U2B;
7194 STRLEN ulen = (STRLEN)*lenp;
7195 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7196 SV_GMAGIC|SV_CONST_RETURN);
7199 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7200 SV_GMAGIC|SV_CONST_RETURN);
7205 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7208 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7209 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7212 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7213 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7214 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7218 (*mgp)->mg_len = ulen;
7221 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7222 byte length pairing. The (byte) length of the total SV is passed in too,
7223 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7224 may not have updated SvCUR, so we can't rely on reading it directly.
7226 The proffered utf8/byte length pairing isn't used if the cache already has
7227 two pairs, and swapping either for the proffered pair would increase the
7228 RMS of the intervals between known byte offsets.
7230 The cache itself consists of 4 STRLEN values
7231 0: larger UTF-8 offset
7232 1: corresponding byte offset
7233 2: smaller UTF-8 offset
7234 3: corresponding byte offset
7236 Unused cache pairs have the value 0, 0.
7237 Keeping the cache "backwards" means that the invariant of
7238 cache[0] >= cache[2] is maintained even with empty slots, which means that
7239 the code that uses it doesn't need to worry if only 1 entry has actually
7240 been set to non-zero. It also makes the "position beyond the end of the
7241 cache" logic much simpler, as the first slot is always the one to start
7245 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7246 const STRLEN utf8, const STRLEN blen)
7250 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7255 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7256 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7257 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7259 (*mgp)->mg_len = -1;
7263 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7264 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7265 (*mgp)->mg_ptr = (char *) cache;
7269 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7270 /* SvPOKp() because it's possible that sv has string overloading, and
7271 therefore is a reference, hence SvPVX() is actually a pointer.
7272 This cures the (very real) symptoms of RT 69422, but I'm not actually
7273 sure whether we should even be caching the results of UTF-8
7274 operations on overloading, given that nothing stops overloading
7275 returning a different value every time it's called. */
7276 const U8 *start = (const U8 *) SvPVX_const(sv);
7277 const STRLEN realutf8 = utf8_length(start, start + byte);
7279 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7283 /* Cache is held with the later position first, to simplify the code
7284 that deals with unbounded ends. */
7286 ASSERT_UTF8_CACHE(cache);
7287 if (cache[1] == 0) {
7288 /* Cache is totally empty */
7291 } else if (cache[3] == 0) {
7292 if (byte > cache[1]) {
7293 /* New one is larger, so goes first. */
7294 cache[2] = cache[0];
7295 cache[3] = cache[1];
7303 #define THREEWAY_SQUARE(a,b,c,d) \
7304 ((float)((d) - (c))) * ((float)((d) - (c))) \
7305 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7306 + ((float)((b) - (a))) * ((float)((b) - (a)))
7308 /* Cache has 2 slots in use, and we know three potential pairs.
7309 Keep the two that give the lowest RMS distance. Do the
7310 calculation in bytes simply because we always know the byte
7311 length. squareroot has the same ordering as the positive value,
7312 so don't bother with the actual square root. */
7313 if (byte > cache[1]) {
7314 /* New position is after the existing pair of pairs. */
7315 const float keep_earlier
7316 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7317 const float keep_later
7318 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7320 if (keep_later < keep_earlier) {
7321 cache[2] = cache[0];
7322 cache[3] = cache[1];
7331 else if (byte > cache[3]) {
7332 /* New position is between the existing pair of pairs. */
7333 const float keep_earlier
7334 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7335 const float keep_later
7336 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7338 if (keep_later < keep_earlier) {
7348 /* New position is before the existing pair of pairs. */
7349 const float keep_earlier
7350 = THREEWAY_SQUARE(0, byte, cache[3], blen);
7351 const float keep_later
7352 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7354 if (keep_later < keep_earlier) {
7359 cache[0] = cache[2];
7360 cache[1] = cache[3];
7366 ASSERT_UTF8_CACHE(cache);
7369 /* We already know all of the way, now we may be able to walk back. The same
7370 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7371 backward is half the speed of walking forward. */
7373 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7374 const U8 *end, STRLEN endu)
7376 const STRLEN forw = target - s;
7377 STRLEN backw = end - target;
7379 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7381 if (forw < 2 * backw) {
7382 return utf8_length(s, target);
7385 while (end > target) {
7387 while (UTF8_IS_CONTINUATION(*end)) {
7396 =for apidoc sv_pos_b2u_flags
7398 Converts the offset from a count of bytes from the start of the string, to
7399 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7400 I<flags> is passed to C<SvPV_flags>, and usually should be
7401 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7407 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7408 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7413 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7416 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7422 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7424 s = (const U8*)SvPV_flags(sv, blen, flags);
7427 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7428 ", byte=%"UVuf, (UV)blen, (UV)offset);
7434 && SvTYPE(sv) >= SVt_PVMG
7435 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7438 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7439 if (cache[1] == offset) {
7440 /* An exact match. */
7443 if (cache[3] == offset) {
7444 /* An exact match. */
7448 if (cache[1] < offset) {
7449 /* We already know part of the way. */
7450 if (mg->mg_len != -1) {
7451 /* Actually, we know the end too. */
7453 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7454 s + blen, mg->mg_len - cache[0]);
7456 len = cache[0] + utf8_length(s + cache[1], send);
7459 else if (cache[3] < offset) {
7460 /* We're between the two cached pairs, so we do the calculation
7461 offset by the byte/utf-8 positions for the earlier pair,
7462 then add the utf-8 characters from the string start to
7464 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7465 s + cache[1], cache[0] - cache[2])
7469 else { /* cache[3] > offset */
7470 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7474 ASSERT_UTF8_CACHE(cache);
7476 } else if (mg->mg_len != -1) {
7477 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7481 if (!found || PL_utf8cache < 0) {
7482 const STRLEN real_len = utf8_length(s, send);
7484 if (found && PL_utf8cache < 0)
7485 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7491 utf8_mg_len_cache_update(sv, &mg, len);
7493 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7500 =for apidoc sv_pos_b2u
7502 Converts the value pointed to by offsetp from a count of bytes from the
7503 start of the string, to a count of the equivalent number of UTF-8 chars.
7504 Handles magic and type coercion.
7506 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7513 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7514 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7519 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7521 PERL_ARGS_ASSERT_SV_POS_B2U;
7526 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7527 SV_GMAGIC|SV_CONST_RETURN);
7531 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7532 STRLEN real, SV *const sv)
7534 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7536 /* As this is debugging only code, save space by keeping this test here,
7537 rather than inlining it in all the callers. */
7538 if (from_cache == real)
7541 /* Need to turn the assertions off otherwise we may recurse infinitely
7542 while printing error messages. */
7543 SAVEI8(PL_utf8cache);
7545 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7546 func, (UV) from_cache, (UV) real, SVfARG(sv));
7552 Returns a boolean indicating whether the strings in the two SVs are
7553 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7554 coerce its args to strings if necessary.
7556 =for apidoc sv_eq_flags
7558 Returns a boolean indicating whether the strings in the two SVs are
7559 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7560 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7566 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7573 SV* svrecode = NULL;
7580 /* if pv1 and pv2 are the same, second SvPV_const call may
7581 * invalidate pv1 (if we are handling magic), so we may need to
7583 if (sv1 == sv2 && flags & SV_GMAGIC
7584 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7585 pv1 = SvPV_const(sv1, cur1);
7586 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7588 pv1 = SvPV_flags_const(sv1, cur1, flags);
7596 pv2 = SvPV_flags_const(sv2, cur2, flags);
7598 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7599 /* Differing utf8ness.
7600 * Do not UTF8size the comparands as a side-effect. */
7603 svrecode = newSVpvn(pv2, cur2);
7604 sv_recode_to_utf8(svrecode, PL_encoding);
7605 pv2 = SvPV_const(svrecode, cur2);
7608 svrecode = newSVpvn(pv1, cur1);
7609 sv_recode_to_utf8(svrecode, PL_encoding);
7610 pv1 = SvPV_const(svrecode, cur1);
7612 /* Now both are in UTF-8. */
7614 SvREFCNT_dec_NN(svrecode);
7620 /* sv1 is the UTF-8 one */
7621 return bytes_cmp_utf8((const U8*)pv2, cur2,
7622 (const U8*)pv1, cur1) == 0;
7625 /* sv2 is the UTF-8 one */
7626 return bytes_cmp_utf8((const U8*)pv1, cur1,
7627 (const U8*)pv2, cur2) == 0;
7633 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7635 SvREFCNT_dec(svrecode);
7643 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7644 string in C<sv1> is less than, equal to, or greater than the string in
7645 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7646 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7648 =for apidoc sv_cmp_flags
7650 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7651 string in C<sv1> is less than, equal to, or greater than the string in
7652 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7653 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7654 also C<sv_cmp_locale_flags>.
7660 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7662 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7666 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7670 const char *pv1, *pv2;
7672 SV *svrecode = NULL;
7679 pv1 = SvPV_flags_const(sv1, cur1, flags);
7686 pv2 = SvPV_flags_const(sv2, cur2, flags);
7688 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7689 /* Differing utf8ness.
7690 * Do not UTF8size the comparands as a side-effect. */
7693 svrecode = newSVpvn(pv2, cur2);
7694 sv_recode_to_utf8(svrecode, PL_encoding);
7695 pv2 = SvPV_const(svrecode, cur2);
7698 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7699 (const U8*)pv1, cur1);
7700 return retval ? retval < 0 ? -1 : +1 : 0;
7705 svrecode = newSVpvn(pv1, cur1);
7706 sv_recode_to_utf8(svrecode, PL_encoding);
7707 pv1 = SvPV_const(svrecode, cur1);
7710 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7711 (const U8*)pv2, cur2);
7712 return retval ? retval < 0 ? -1 : +1 : 0;
7718 cmp = cur2 ? -1 : 0;
7722 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7725 cmp = retval < 0 ? -1 : 1;
7726 } else if (cur1 == cur2) {
7729 cmp = cur1 < cur2 ? -1 : 1;
7733 SvREFCNT_dec(svrecode);
7739 =for apidoc sv_cmp_locale
7741 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7742 'use bytes' aware, handles get magic, and will coerce its args to strings
7743 if necessary. See also C<sv_cmp>.
7745 =for apidoc sv_cmp_locale_flags
7747 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7748 'use bytes' aware and will coerce its args to strings if necessary. If the
7749 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7755 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7757 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7761 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7764 #ifdef USE_LOCALE_COLLATE
7770 if (PL_collation_standard)
7774 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7776 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7778 if (!pv1 || !len1) {
7789 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7792 return retval < 0 ? -1 : 1;
7795 * When the result of collation is equality, that doesn't mean
7796 * that there are no differences -- some locales exclude some
7797 * characters from consideration. So to avoid false equalities,
7798 * we use the raw string as a tiebreaker.
7805 PERL_UNUSED_ARG(flags);
7806 #endif /* USE_LOCALE_COLLATE */
7808 return sv_cmp(sv1, sv2);
7812 #ifdef USE_LOCALE_COLLATE
7815 =for apidoc sv_collxfrm
7817 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7818 C<sv_collxfrm_flags>.
7820 =for apidoc sv_collxfrm_flags
7822 Add Collate Transform magic to an SV if it doesn't already have it. If the
7823 flags contain SV_GMAGIC, it handles get-magic.
7825 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7826 scalar data of the variable, but transformed to such a format that a normal
7827 memory comparison can be used to compare the data according to the locale
7834 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7838 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7840 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7841 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7847 Safefree(mg->mg_ptr);
7848 s = SvPV_flags_const(sv, len, flags);
7849 if ((xf = mem_collxfrm(s, len, &xlen))) {
7851 #ifdef PERL_OLD_COPY_ON_WRITE
7853 sv_force_normal_flags(sv, 0);
7855 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
7869 if (mg && mg->mg_ptr) {
7871 return mg->mg_ptr + sizeof(PL_collation_ix);
7879 #endif /* USE_LOCALE_COLLATE */
7882 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7884 SV * const tsv = newSV(0);
7887 sv_gets(tsv, fp, 0);
7888 sv_utf8_upgrade_nomg(tsv);
7889 SvCUR_set(sv,append);
7892 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
7896 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7899 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
7900 /* Grab the size of the record we're getting */
7901 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
7908 /* With a true, record-oriented file on VMS, we need to use read directly
7909 * to ensure that we respect RMS record boundaries. The user is responsible
7910 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
7911 * record size) field. N.B. This is likely to produce invalid results on
7912 * varying-width character data when a record ends mid-character.
7914 fd = PerlIO_fileno(fp);
7916 && PerlLIO_fstat(fd, &st) == 0
7917 && (st.st_fab_rfm == FAB$C_VAR
7918 || st.st_fab_rfm == FAB$C_VFC
7919 || st.st_fab_rfm == FAB$C_FIX)) {
7921 bytesread = PerlLIO_read(fd, buffer, recsize);
7923 else /* in-memory file from PerlIO::Scalar
7924 * or not a record-oriented file
7928 bytesread = PerlIO_read(fp, buffer, recsize);
7930 /* At this point, the logic in sv_get() means that sv will
7931 be treated as utf-8 if the handle is utf8.
7933 if (PerlIO_isutf8(fp) && bytesread > 0) {
7934 char *bend = buffer + bytesread;
7935 char *bufp = buffer;
7936 size_t charcount = 0;
7937 bool charstart = TRUE;
7940 while (charcount < recsize) {
7941 /* count accumulated characters */
7942 while (bufp < bend) {
7944 skip = UTF8SKIP(bufp);
7946 if (bufp + skip > bend) {
7947 /* partial at the end */
7958 if (charcount < recsize) {
7960 STRLEN bufp_offset = bufp - buffer;
7961 SSize_t morebytesread;
7963 /* originally I read enough to fill any incomplete
7964 character and the first byte of the next
7965 character if needed, but if there's many
7966 multi-byte encoded characters we're going to be
7967 making a read call for every character beyond
7968 the original read size.
7970 So instead, read the rest of the character if
7971 any, and enough bytes to match at least the
7972 start bytes for each character we're going to
7976 readsize = recsize - charcount;
7978 readsize = skip - (bend - bufp) + recsize - charcount - 1;
7979 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
7980 bend = buffer + bytesread;
7981 morebytesread = PerlIO_read(fp, bend, readsize);
7982 if (morebytesread <= 0) {
7983 /* we're done, if we still have incomplete
7984 characters the check code in sv_gets() will
7987 I'd originally considered doing
7988 PerlIO_ungetc() on all but the lead
7989 character of the incomplete character, but
7990 read() doesn't do that, so I don't.
7995 /* prepare to scan some more */
7996 bytesread += morebytesread;
7997 bend = buffer + bytesread;
7998 bufp = buffer + bufp_offset;
8006 SvCUR_set(sv, bytesread + append);
8007 buffer[bytesread] = '\0';
8008 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8014 Get a line from the filehandle and store it into the SV, optionally
8015 appending to the currently-stored string. If C<append> is not 0, the
8016 line is appended to the SV instead of overwriting it. C<append> should
8017 be set to the byte offset that the appended string should start at
8018 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8024 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8034 PERL_ARGS_ASSERT_SV_GETS;
8036 if (SvTHINKFIRST(sv))
8037 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8038 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8040 However, perlbench says it's slower, because the existing swipe code
8041 is faster than copy on write.
8042 Swings and roundabouts. */
8043 SvUPGRADE(sv, SVt_PV);
8046 /* line is going to be appended to the existing buffer in the sv */
8047 if (PerlIO_isutf8(fp)) {
8049 sv_utf8_upgrade_nomg(sv);
8050 sv_pos_u2b(sv,&append,0);
8052 } else if (SvUTF8(sv)) {
8053 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8059 /* not appending - "clear" the string by setting SvCUR to 0,
8060 * the pv is still avaiable. */
8063 if (PerlIO_isutf8(fp))
8066 if (IN_PERL_COMPILETIME) {
8067 /* we always read code in line mode */
8071 else if (RsSNARF(PL_rs)) {
8072 /* If it is a regular disk file use size from stat() as estimate
8073 of amount we are going to read -- may result in mallocing
8074 more memory than we really need if the layers below reduce
8075 the size we read (e.g. CRLF or a gzip layer).
8078 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8079 const Off_t offset = PerlIO_tell(fp);
8080 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8081 #ifdef PERL_NEW_COPY_ON_WRITE
8082 /* Add an extra byte for the sake of copy-on-write's
8083 * buffer reference count. */
8084 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8086 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8093 else if (RsRECORD(PL_rs)) {
8094 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8096 else if (RsPARA(PL_rs)) {
8102 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8103 if (PerlIO_isutf8(fp)) {
8104 rsptr = SvPVutf8(PL_rs, rslen);
8107 if (SvUTF8(PL_rs)) {
8108 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8109 Perl_croak(aTHX_ "Wide character in $/");
8112 /* extract the raw pointer to the record separator */
8113 rsptr = SvPV_const(PL_rs, rslen);
8117 /* rslast is the last character in the record separator
8118 * note we don't use rslast except when rslen is true, so the
8119 * null assign is a placeholder. */
8120 rslast = rslen ? rsptr[rslen - 1] : '\0';
8122 if (rspara) { /* have to do this both before and after */
8123 do { /* to make sure file boundaries work right */
8126 i = PerlIO_getc(fp);
8130 PerlIO_ungetc(fp,i);
8136 /* See if we know enough about I/O mechanism to cheat it ! */
8138 /* This used to be #ifdef test - it is made run-time test for ease
8139 of abstracting out stdio interface. One call should be cheap
8140 enough here - and may even be a macro allowing compile
8144 if (PerlIO_fast_gets(fp)) {
8146 * We can do buffer based IO operations on this filehandle.
8148 * This means we can bypass a lot of subcalls and process
8149 * the buffer directly, it also means we know the upper bound
8150 * on the amount of data we might read of the current buffer
8151 * into our sv. Knowing this allows us to preallocate the pv
8152 * to be able to hold that maximum, which allows us to simplify
8153 * a lot of logic. */
8156 * We're going to steal some values from the stdio struct
8157 * and put EVERYTHING in the innermost loop into registers.
8159 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8160 STRLEN bpx; /* length of the data in the target sv
8161 used to fix pointers after a SvGROW */
8162 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8163 of data left in the read-ahead buffer.
8164 If 0 then the pv buffer can hold the full
8165 amount left, otherwise this is the amount it
8168 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8169 /* An ungetc()d char is handled separately from the regular
8170 * buffer, so we getc() it back out and stuff it in the buffer.
8172 i = PerlIO_getc(fp);
8173 if (i == EOF) return 0;
8174 *(--((*fp)->_ptr)) = (unsigned char) i;
8178 /* Here is some breathtakingly efficient cheating */
8180 /* When you read the following logic resist the urge to think
8181 * of record separators that are 1 byte long. They are an
8182 * uninteresting special (simple) case.
8184 * Instead think of record separators which are at least 2 bytes
8185 * long, and keep in mind that we need to deal with such
8186 * separators when they cross a read-ahead buffer boundary.
8188 * Also consider that we need to gracefully deal with separators
8189 * that may be longer than a single read ahead buffer.
8191 * Lastly do not forget we want to copy the delimiter as well. We
8192 * are copying all data in the file _up_to_and_including_ the separator
8195 * Now that you have all that in mind here is what is happening below:
8197 * 1. When we first enter the loop we do some memory book keeping to see
8198 * how much free space there is in the target SV. (This sub assumes that
8199 * it is operating on the same SV most of the time via $_ and that it is
8200 * going to be able to reuse the same pv buffer each call.) If there is
8201 * "enough" room then we set "shortbuffered" to how much space there is
8202 * and start reading forward.
8204 * 2. When we scan forward we copy from the read-ahead buffer to the target
8205 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8206 * and the end of the of pv, as well as for the "rslast", which is the last
8207 * char of the separator.
8209 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8210 * (which has a "complete" record up to the point we saw rslast) and check
8211 * it to see if it matches the separator. If it does we are done. If it doesn't
8212 * we continue on with the scan/copy.
8214 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8215 * the IO system to read the next buffer. We do this by doing a getc(), which
8216 * returns a single char read (or EOF), and prefills the buffer, and also
8217 * allows us to find out how full the buffer is. We use this information to
8218 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8219 * the returned single char into the target sv, and then go back into scan
8222 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8223 * remaining space in the read-buffer.
8225 * Note that this code despite its twisty-turny nature is pretty darn slick.
8226 * It manages single byte separators, multi-byte cross boundary separators,
8227 * and cross-read-buffer separators cleanly and efficiently at the cost
8228 * of potentially greatly overallocating the target SV.
8234 /* get the number of bytes remaining in the read-ahead buffer
8235 * on first call on a given fp this will return 0.*/
8236 cnt = PerlIO_get_cnt(fp);
8238 /* make sure we have the room */
8239 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8240 /* Not room for all of it
8241 if we are looking for a separator and room for some
8243 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8244 /* just process what we have room for */
8245 shortbuffered = cnt - SvLEN(sv) + append + 1;
8246 cnt -= shortbuffered;
8249 /* ensure that the target sv has enough room to hold
8250 * the rest of the read-ahead buffer */
8252 /* remember that cnt can be negative */
8253 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8257 /* we have enough room to hold the full buffer, lets scream */
8261 /* extract the pointer to sv's string buffer, offset by append as necessary */
8262 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8263 /* extract the point to the read-ahead buffer */
8264 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8266 /* some trace debug output */
8267 DEBUG_P(PerlIO_printf(Perl_debug_log,
8268 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8269 DEBUG_P(PerlIO_printf(Perl_debug_log,
8270 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8272 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8273 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8277 /* if there is stuff left in the read-ahead buffer */
8279 /* if there is a separator */
8281 /* loop until we hit the end of the read-ahead buffer */
8282 while (cnt > 0) { /* this | eat */
8283 /* scan forward copying and searching for rslast as we go */
8285 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8286 goto thats_all_folks; /* screams | sed :-) */
8290 /* no separator, slurp the full buffer */
8291 Copy(ptr, bp, cnt, char); /* this | eat */
8292 bp += cnt; /* screams | dust */
8293 ptr += cnt; /* louder | sed :-) */
8295 assert (!shortbuffered);
8296 goto cannot_be_shortbuffered;
8300 if (shortbuffered) { /* oh well, must extend */
8301 /* we didnt have enough room to fit the line into the target buffer
8302 * so we must extend the target buffer and keep going */
8303 cnt = shortbuffered;
8305 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8307 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8308 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8309 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8313 cannot_be_shortbuffered:
8314 /* we need to refill the read-ahead buffer if possible */
8316 DEBUG_P(PerlIO_printf(Perl_debug_log,
8317 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8318 PTR2UV(ptr),(IV)cnt));
8319 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8321 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8322 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8323 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8324 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8327 call PerlIO_getc() to let it prefill the lookahead buffer
8329 This used to call 'filbuf' in stdio form, but as that behaves like
8330 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8331 another abstraction.
8333 Note we have to deal with the char in 'i' if we are not at EOF
8335 i = PerlIO_getc(fp); /* get more characters */
8337 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8338 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8339 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8340 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8342 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8343 cnt = PerlIO_get_cnt(fp);
8344 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8345 DEBUG_P(PerlIO_printf(Perl_debug_log,
8346 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8347 PTR2UV(ptr),(IV)cnt));
8349 if (i == EOF) /* all done for ever? */
8350 goto thats_really_all_folks;
8352 /* make sure we have enough space in the target sv */
8353 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8355 SvGROW(sv, bpx + cnt + 2);
8356 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8358 /* copy of the char we got from getc() */
8359 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8361 /* make sure we deal with the i being the last character of a separator */
8362 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8363 goto thats_all_folks;
8367 /* check if we have actually found the separator - only really applies
8369 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8370 memNE((char*)bp - rslen, rsptr, rslen))
8371 goto screamer; /* go back to the fray */
8372 thats_really_all_folks:
8374 cnt += shortbuffered;
8375 DEBUG_P(PerlIO_printf(Perl_debug_log,
8376 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8377 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8378 DEBUG_P(PerlIO_printf(Perl_debug_log,
8379 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8381 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8382 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8384 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8385 DEBUG_P(PerlIO_printf(Perl_debug_log,
8386 "Screamer: done, len=%ld, string=|%.*s|\n",
8387 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8391 /*The big, slow, and stupid way. */
8392 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8393 STDCHAR *buf = NULL;
8394 Newx(buf, 8192, STDCHAR);
8402 const STDCHAR * const bpe = buf + sizeof(buf);
8404 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8405 ; /* keep reading */
8409 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8410 /* Accommodate broken VAXC compiler, which applies U8 cast to
8411 * both args of ?: operator, causing EOF to change into 255
8414 i = (U8)buf[cnt - 1];
8420 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8422 sv_catpvn_nomg(sv, (char *) buf, cnt);
8424 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8426 if (i != EOF && /* joy */
8428 SvCUR(sv) < rslen ||
8429 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8433 * If we're reading from a TTY and we get a short read,
8434 * indicating that the user hit his EOF character, we need
8435 * to notice it now, because if we try to read from the TTY
8436 * again, the EOF condition will disappear.
8438 * The comparison of cnt to sizeof(buf) is an optimization
8439 * that prevents unnecessary calls to feof().
8443 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8447 #ifdef USE_HEAP_INSTEAD_OF_STACK
8452 if (rspara) { /* have to do this both before and after */
8453 while (i != EOF) { /* to make sure file boundaries work right */
8454 i = PerlIO_getc(fp);
8456 PerlIO_ungetc(fp,i);
8462 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8468 Auto-increment of the value in the SV, doing string to numeric conversion
8469 if necessary. Handles 'get' magic and operator overloading.
8475 Perl_sv_inc(pTHX_ SV *const sv)
8484 =for apidoc sv_inc_nomg
8486 Auto-increment of the value in the SV, doing string to numeric conversion
8487 if necessary. Handles operator overloading. Skips handling 'get' magic.
8493 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8500 if (SvTHINKFIRST(sv)) {
8501 if (SvREADONLY(sv)) {
8502 Perl_croak_no_modify();
8506 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8508 i = PTR2IV(SvRV(sv));
8512 else sv_force_normal_flags(sv, 0);
8514 flags = SvFLAGS(sv);
8515 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8516 /* It's (privately or publicly) a float, but not tested as an
8517 integer, so test it to see. */
8519 flags = SvFLAGS(sv);
8521 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8522 /* It's publicly an integer, or privately an integer-not-float */
8523 #ifdef PERL_PRESERVE_IVUV
8527 if (SvUVX(sv) == UV_MAX)
8528 sv_setnv(sv, UV_MAX_P1);
8530 (void)SvIOK_only_UV(sv);
8531 SvUV_set(sv, SvUVX(sv) + 1);
8533 if (SvIVX(sv) == IV_MAX)
8534 sv_setuv(sv, (UV)IV_MAX + 1);
8536 (void)SvIOK_only(sv);
8537 SvIV_set(sv, SvIVX(sv) + 1);
8542 if (flags & SVp_NOK) {
8543 const NV was = SvNVX(sv);
8544 if (NV_OVERFLOWS_INTEGERS_AT &&
8545 was >= NV_OVERFLOWS_INTEGERS_AT) {
8546 /* diag_listed_as: Lost precision when %s %f by 1 */
8547 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8548 "Lost precision when incrementing %" NVff " by 1",
8551 (void)SvNOK_only(sv);
8552 SvNV_set(sv, was + 1.0);
8556 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8557 if ((flags & SVTYPEMASK) < SVt_PVIV)
8558 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8559 (void)SvIOK_only(sv);
8564 while (isALPHA(*d)) d++;
8565 while (isDIGIT(*d)) d++;
8566 if (d < SvEND(sv)) {
8567 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8568 #ifdef PERL_PRESERVE_IVUV
8569 /* Got to punt this as an integer if needs be, but we don't issue
8570 warnings. Probably ought to make the sv_iv_please() that does
8571 the conversion if possible, and silently. */
8572 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8573 /* Need to try really hard to see if it's an integer.
8574 9.22337203685478e+18 is an integer.
8575 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8576 so $a="9.22337203685478e+18"; $a+0; $a++
8577 needs to be the same as $a="9.22337203685478e+18"; $a++
8584 /* sv_2iv *should* have made this an NV */
8585 if (flags & SVp_NOK) {
8586 (void)SvNOK_only(sv);
8587 SvNV_set(sv, SvNVX(sv) + 1.0);
8590 /* I don't think we can get here. Maybe I should assert this
8591 And if we do get here I suspect that sv_setnv will croak. NWC
8593 #if defined(USE_LONG_DOUBLE)
8594 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"PERL_PRIgldbl"\n",
8595 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8597 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8598 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8601 #endif /* PERL_PRESERVE_IVUV */
8602 if (!numtype && ckWARN(WARN_NUMERIC))
8603 not_incrementable(sv);
8604 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8608 while (d >= SvPVX_const(sv)) {
8616 /* MKS: The original code here died if letters weren't consecutive.
8617 * at least it didn't have to worry about non-C locales. The
8618 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8619 * arranged in order (although not consecutively) and that only
8620 * [A-Za-z] are accepted by isALPHA in the C locale.
8622 if (*d != 'z' && *d != 'Z') {
8623 do { ++*d; } while (!isALPHA(*d));
8626 *(d--) -= 'z' - 'a';
8631 *(d--) -= 'z' - 'a' + 1;
8635 /* oh,oh, the number grew */
8636 SvGROW(sv, SvCUR(sv) + 2);
8637 SvCUR_set(sv, SvCUR(sv) + 1);
8638 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8649 Auto-decrement of the value in the SV, doing string to numeric conversion
8650 if necessary. Handles 'get' magic and operator overloading.
8656 Perl_sv_dec(pTHX_ SV *const sv)
8665 =for apidoc sv_dec_nomg
8667 Auto-decrement of the value in the SV, doing string to numeric conversion
8668 if necessary. Handles operator overloading. Skips handling 'get' magic.
8674 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8680 if (SvTHINKFIRST(sv)) {
8681 if (SvREADONLY(sv)) {
8682 Perl_croak_no_modify();
8686 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8688 i = PTR2IV(SvRV(sv));
8692 else sv_force_normal_flags(sv, 0);
8694 /* Unlike sv_inc we don't have to worry about string-never-numbers
8695 and keeping them magic. But we mustn't warn on punting */
8696 flags = SvFLAGS(sv);
8697 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8698 /* It's publicly an integer, or privately an integer-not-float */
8699 #ifdef PERL_PRESERVE_IVUV
8703 if (SvUVX(sv) == 0) {
8704 (void)SvIOK_only(sv);
8708 (void)SvIOK_only_UV(sv);
8709 SvUV_set(sv, SvUVX(sv) - 1);
8712 if (SvIVX(sv) == IV_MIN) {
8713 sv_setnv(sv, (NV)IV_MIN);
8717 (void)SvIOK_only(sv);
8718 SvIV_set(sv, SvIVX(sv) - 1);
8723 if (flags & SVp_NOK) {
8726 const NV was = SvNVX(sv);
8727 if (NV_OVERFLOWS_INTEGERS_AT &&
8728 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8729 /* diag_listed_as: Lost precision when %s %f by 1 */
8730 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8731 "Lost precision when decrementing %" NVff " by 1",
8734 (void)SvNOK_only(sv);
8735 SvNV_set(sv, was - 1.0);
8739 if (!(flags & SVp_POK)) {
8740 if ((flags & SVTYPEMASK) < SVt_PVIV)
8741 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8743 (void)SvIOK_only(sv);
8746 #ifdef PERL_PRESERVE_IVUV
8748 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8749 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8750 /* Need to try really hard to see if it's an integer.
8751 9.22337203685478e+18 is an integer.
8752 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8753 so $a="9.22337203685478e+18"; $a+0; $a--
8754 needs to be the same as $a="9.22337203685478e+18"; $a--
8761 /* sv_2iv *should* have made this an NV */
8762 if (flags & SVp_NOK) {
8763 (void)SvNOK_only(sv);
8764 SvNV_set(sv, SvNVX(sv) - 1.0);
8767 /* I don't think we can get here. Maybe I should assert this
8768 And if we do get here I suspect that sv_setnv will croak. NWC
8770 #if defined(USE_LONG_DOUBLE)
8771 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"PERL_PRIgldbl"\n",
8772 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8774 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8775 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8779 #endif /* PERL_PRESERVE_IVUV */
8780 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8783 /* this define is used to eliminate a chunk of duplicated but shared logic
8784 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8785 * used anywhere but here - yves
8787 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8790 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8794 =for apidoc sv_mortalcopy
8796 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8797 The new SV is marked as mortal. It will be destroyed "soon", either by an
8798 explicit call to FREETMPS, or by an implicit call at places such as
8799 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8804 /* Make a string that will exist for the duration of the expression
8805 * evaluation. Actually, it may have to last longer than that, but
8806 * hopefully we won't free it until it has been assigned to a
8807 * permanent location. */
8810 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8814 if (flags & SV_GMAGIC)
8815 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8817 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8818 PUSH_EXTEND_MORTAL__SV_C(sv);
8824 =for apidoc sv_newmortal
8826 Creates a new null SV which is mortal. The reference count of the SV is
8827 set to 1. It will be destroyed "soon", either by an explicit call to
8828 FREETMPS, or by an implicit call at places such as statement boundaries.
8829 See also C<sv_mortalcopy> and C<sv_2mortal>.
8835 Perl_sv_newmortal(pTHX)
8840 SvFLAGS(sv) = SVs_TEMP;
8841 PUSH_EXTEND_MORTAL__SV_C(sv);
8847 =for apidoc newSVpvn_flags
8849 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8850 characters) into it. The reference count for the
8851 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8852 string. You are responsible for ensuring that the source string is at least
8853 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8854 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8855 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8856 returning. If C<SVf_UTF8> is set, C<s>
8857 is considered to be in UTF-8 and the
8858 C<SVf_UTF8> flag will be set on the new SV.
8859 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8861 #define newSVpvn_utf8(s, len, u) \
8862 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8868 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8872 /* All the flags we don't support must be zero.
8873 And we're new code so I'm going to assert this from the start. */
8874 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
8876 sv_setpvn(sv,s,len);
8878 /* This code used to do a sv_2mortal(), however we now unroll the call to
8879 * sv_2mortal() and do what it does ourselves here. Since we have asserted
8880 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
8881 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
8882 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
8883 * means that we eliminate quite a few steps than it looks - Yves
8884 * (explaining patch by gfx) */
8886 SvFLAGS(sv) |= flags;
8888 if(flags & SVs_TEMP){
8889 PUSH_EXTEND_MORTAL__SV_C(sv);
8896 =for apidoc sv_2mortal
8898 Marks an existing SV as mortal. The SV will be destroyed "soon", either
8899 by an explicit call to FREETMPS, or by an implicit call at places such as
8900 statement boundaries. SvTEMP() is turned on which means that the SV's
8901 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
8902 and C<sv_mortalcopy>.
8908 Perl_sv_2mortal(pTHX_ SV *const sv)
8915 PUSH_EXTEND_MORTAL__SV_C(sv);
8923 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8924 characters) into it. The reference count for the
8925 SV is set to 1. If C<len> is zero, Perl will compute the length using
8926 strlen(), (which means if you use this option, that C<s> can't have embedded
8927 C<NUL> characters and has to have a terminating C<NUL> byte).
8929 For efficiency, consider using C<newSVpvn> instead.
8935 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
8940 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
8945 =for apidoc newSVpvn
8947 Creates a new SV and copies a string into it, which may contain C<NUL> characters
8948 (C<\0>) and other binary data. The reference count for the SV is set to 1.
8949 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
8950 are responsible for ensuring that the source buffer is at least
8951 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
8958 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
8962 sv_setpvn(sv,buffer,len);
8967 =for apidoc newSVhek
8969 Creates a new SV from the hash key structure. It will generate scalars that
8970 point to the shared string table where possible. Returns a new (undefined)
8971 SV if the hek is NULL.
8977 Perl_newSVhek(pTHX_ const HEK *const hek)
8986 if (HEK_LEN(hek) == HEf_SVKEY) {
8987 return newSVsv(*(SV**)HEK_KEY(hek));
8989 const int flags = HEK_FLAGS(hek);
8990 if (flags & HVhek_WASUTF8) {
8992 Andreas would like keys he put in as utf8 to come back as utf8
8994 STRLEN utf8_len = HEK_LEN(hek);
8995 SV * const sv = newSV_type(SVt_PV);
8996 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
8997 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
8998 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9001 } else if (flags & HVhek_UNSHARED) {
9002 /* A hash that isn't using shared hash keys has to have
9003 the flag in every key so that we know not to try to call
9004 share_hek_hek on it. */
9006 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9011 /* This will be overwhelminly the most common case. */
9013 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9014 more efficient than sharepvn(). */
9018 sv_upgrade(sv, SVt_PV);
9019 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9020 SvCUR_set(sv, HEK_LEN(hek));
9032 =for apidoc newSVpvn_share
9034 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9035 table. If the string does not already exist in the table, it is
9036 created first. Turns on the SvIsCOW flag (or READONLY
9037 and FAKE in 5.16 and earlier). If the C<hash> parameter
9038 is non-zero, that value is used; otherwise the hash is computed.
9039 The string's hash can later be retrieved from the SV
9040 with the C<SvSHARED_HASH()> macro. The idea here is
9041 that as the string table is used for shared hash keys these strings will have
9042 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9048 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9052 bool is_utf8 = FALSE;
9053 const char *const orig_src = src;
9056 STRLEN tmplen = -len;
9058 /* See the note in hv.c:hv_fetch() --jhi */
9059 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9063 PERL_HASH(hash, src, len);
9065 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9066 changes here, update it there too. */
9067 sv_upgrade(sv, SVt_PV);
9068 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9075 if (src != orig_src)
9081 =for apidoc newSVpv_share
9083 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9090 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9092 return newSVpvn_share(src, strlen(src), hash);
9095 #if defined(PERL_IMPLICIT_CONTEXT)
9097 /* pTHX_ magic can't cope with varargs, so this is a no-context
9098 * version of the main function, (which may itself be aliased to us).
9099 * Don't access this version directly.
9103 Perl_newSVpvf_nocontext(const char *const pat, ...)
9109 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9111 va_start(args, pat);
9112 sv = vnewSVpvf(pat, &args);
9119 =for apidoc newSVpvf
9121 Creates a new SV and initializes it with the string formatted like
9128 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9133 PERL_ARGS_ASSERT_NEWSVPVF;
9135 va_start(args, pat);
9136 sv = vnewSVpvf(pat, &args);
9141 /* backend for newSVpvf() and newSVpvf_nocontext() */
9144 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9148 PERL_ARGS_ASSERT_VNEWSVPVF;
9151 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9158 Creates a new SV and copies a floating point value into it.
9159 The reference count for the SV is set to 1.
9165 Perl_newSVnv(pTHX_ const NV n)
9177 Creates a new SV and copies an integer into it. The reference count for the
9184 Perl_newSViv(pTHX_ const IV i)
9196 Creates a new SV and copies an unsigned integer into it.
9197 The reference count for the SV is set to 1.
9203 Perl_newSVuv(pTHX_ const UV u)
9213 =for apidoc newSV_type
9215 Creates a new SV, of the type specified. The reference count for the new SV
9222 Perl_newSV_type(pTHX_ const svtype type)
9227 sv_upgrade(sv, type);
9232 =for apidoc newRV_noinc
9234 Creates an RV wrapper for an SV. The reference count for the original
9235 SV is B<not> incremented.
9241 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9243 SV *sv = newSV_type(SVt_IV);
9245 PERL_ARGS_ASSERT_NEWRV_NOINC;
9248 SvRV_set(sv, tmpRef);
9253 /* newRV_inc is the official function name to use now.
9254 * newRV_inc is in fact #defined to newRV in sv.h
9258 Perl_newRV(pTHX_ SV *const sv)
9260 PERL_ARGS_ASSERT_NEWRV;
9262 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9268 Creates a new SV which is an exact duplicate of the original SV.
9275 Perl_newSVsv(pTHX_ SV *const old)
9281 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9282 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9285 /* Do this here, otherwise we leak the new SV if this croaks. */
9288 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9289 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9290 sv_setsv_flags(sv, old, SV_NOSTEAL);
9295 =for apidoc sv_reset
9297 Underlying implementation for the C<reset> Perl function.
9298 Note that the perl-level function is vaguely deprecated.
9304 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9306 PERL_ARGS_ASSERT_SV_RESET;
9308 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9312 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9314 char todo[PERL_UCHAR_MAX+1];
9317 if (!stash || SvTYPE(stash) != SVt_PVHV)
9320 if (!s) { /* reset ?? searches */
9321 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9323 const U32 count = mg->mg_len / sizeof(PMOP**);
9324 PMOP **pmp = (PMOP**) mg->mg_ptr;
9325 PMOP *const *const end = pmp + count;
9329 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9331 (*pmp)->op_pmflags &= ~PMf_USED;
9339 /* reset variables */
9341 if (!HvARRAY(stash))
9344 Zero(todo, 256, char);
9348 I32 i = (unsigned char)*s;
9352 max = (unsigned char)*s++;
9353 for ( ; i <= max; i++) {
9356 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9358 for (entry = HvARRAY(stash)[i];
9360 entry = HeNEXT(entry))
9365 if (!todo[(U8)*HeKEY(entry)])
9367 gv = MUTABLE_GV(HeVAL(entry));
9369 if (sv && !SvREADONLY(sv)) {
9370 SV_CHECK_THINKFIRST_COW_DROP(sv);
9371 if (!isGV(sv)) SvOK_off(sv);
9376 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9387 Using various gambits, try to get an IO from an SV: the IO slot if its a
9388 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9389 named after the PV if we're a string.
9391 'Get' magic is ignored on the sv passed in, but will be called on
9392 C<SvRV(sv)> if sv is an RV.
9398 Perl_sv_2io(pTHX_ SV *const sv)
9403 PERL_ARGS_ASSERT_SV_2IO;
9405 switch (SvTYPE(sv)) {
9407 io = MUTABLE_IO(sv);
9411 if (isGV_with_GP(sv)) {
9412 gv = MUTABLE_GV(sv);
9415 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9416 HEKfARG(GvNAME_HEK(gv)));
9422 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9424 SvGETMAGIC(SvRV(sv));
9425 return sv_2io(SvRV(sv));
9427 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9434 if (SvGMAGICAL(sv)) {
9435 newsv = sv_newmortal();
9436 sv_setsv_nomg(newsv, sv);
9438 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9448 Using various gambits, try to get a CV from an SV; in addition, try if
9449 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9450 The flags in C<lref> are passed to gv_fetchsv.
9456 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9461 PERL_ARGS_ASSERT_SV_2CV;
9468 switch (SvTYPE(sv)) {
9472 return MUTABLE_CV(sv);
9482 sv = amagic_deref_call(sv, to_cv_amg);
9485 if (SvTYPE(sv) == SVt_PVCV) {
9486 cv = MUTABLE_CV(sv);
9491 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9492 gv = MUTABLE_GV(sv);
9494 Perl_croak(aTHX_ "Not a subroutine reference");
9496 else if (isGV_with_GP(sv)) {
9497 gv = MUTABLE_GV(sv);
9500 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9507 /* Some flags to gv_fetchsv mean don't really create the GV */
9508 if (!isGV_with_GP(gv)) {
9513 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9514 /* XXX this is probably not what they think they're getting.
9515 * It has the same effect as "sub name;", i.e. just a forward
9526 Returns true if the SV has a true value by Perl's rules.
9527 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9528 instead use an in-line version.
9534 Perl_sv_true(pTHX_ SV *const sv)
9539 const XPV* const tXpv = (XPV*)SvANY(sv);
9541 (tXpv->xpv_cur > 1 ||
9542 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9549 return SvIVX(sv) != 0;
9552 return SvNVX(sv) != 0.0;
9554 return sv_2bool(sv);
9560 =for apidoc sv_pvn_force
9562 Get a sensible string out of the SV somehow.
9563 A private implementation of the C<SvPV_force> macro for compilers which
9564 can't cope with complex macro expressions. Always use the macro instead.
9566 =for apidoc sv_pvn_force_flags
9568 Get a sensible string out of the SV somehow.
9569 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9570 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9571 implemented in terms of this function.
9572 You normally want to use the various wrapper macros instead: see
9573 C<SvPV_force> and C<SvPV_force_nomg>
9579 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9581 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9583 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9584 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9585 sv_force_normal_flags(sv, 0);
9595 if (SvTYPE(sv) > SVt_PVLV
9596 || isGV_with_GP(sv))
9597 /* diag_listed_as: Can't coerce %s to %s in %s */
9598 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9600 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9607 if (SvTYPE(sv) < SVt_PV ||
9608 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9611 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9612 SvGROW(sv, len + 1);
9613 Move(s,SvPVX(sv),len,char);
9615 SvPVX(sv)[len] = '\0';
9618 SvPOK_on(sv); /* validate pointer */
9620 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9621 PTR2UV(sv),SvPVX_const(sv)));
9624 (void)SvPOK_only_UTF8(sv);
9625 return SvPVX_mutable(sv);
9629 =for apidoc sv_pvbyten_force
9631 The backend for the C<SvPVbytex_force> macro. Always use the macro
9638 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9640 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9642 sv_pvn_force(sv,lp);
9643 sv_utf8_downgrade(sv,0);
9649 =for apidoc sv_pvutf8n_force
9651 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9658 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9660 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9663 sv_utf8_upgrade_nomg(sv);
9669 =for apidoc sv_reftype
9671 Returns a string describing what the SV is a reference to.
9677 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9679 PERL_ARGS_ASSERT_SV_REFTYPE;
9680 if (ob && SvOBJECT(sv)) {
9681 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9684 /* WARNING - There is code, for instance in mg.c, that assumes that
9685 * the only reason that sv_reftype(sv,0) would return a string starting
9686 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9687 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9688 * this routine inside other subs, and it saves time.
9689 * Do not change this assumption without searching for "dodgy type check" in
9692 switch (SvTYPE(sv)) {
9707 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9708 /* tied lvalues should appear to be
9709 * scalars for backwards compatibility */
9710 : (LvTYPE(sv) == 't' || LvTYPE(sv) == 'T')
9711 ? "SCALAR" : "LVALUE");
9712 case SVt_PVAV: return "ARRAY";
9713 case SVt_PVHV: return "HASH";
9714 case SVt_PVCV: return "CODE";
9715 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9716 ? "GLOB" : "SCALAR");
9717 case SVt_PVFM: return "FORMAT";
9718 case SVt_PVIO: return "IO";
9719 case SVt_INVLIST: return "INVLIST";
9720 case SVt_REGEXP: return "REGEXP";
9721 default: return "UNKNOWN";
9729 Returns a SV describing what the SV passed in is a reference to.
9735 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9737 PERL_ARGS_ASSERT_SV_REF;
9740 dst = sv_newmortal();
9742 if (ob && SvOBJECT(sv)) {
9743 HvNAME_get(SvSTASH(sv))
9744 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9745 : sv_setpvn(dst, "__ANON__", 8);
9748 const char * reftype = sv_reftype(sv, 0);
9749 sv_setpv(dst, reftype);
9755 =for apidoc sv_isobject
9757 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9758 object. If the SV is not an RV, or if the object is not blessed, then this
9765 Perl_sv_isobject(pTHX_ SV *sv)
9781 Returns a boolean indicating whether the SV is blessed into the specified
9782 class. This does not check for subtypes; use C<sv_derived_from> to verify
9783 an inheritance relationship.
9789 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9793 PERL_ARGS_ASSERT_SV_ISA;
9803 hvname = HvNAME_get(SvSTASH(sv));
9807 return strEQ(hvname, name);
9813 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9814 RV then it will be upgraded to one. If C<classname> is non-null then the new
9815 SV will be blessed in the specified package. The new SV is returned and its
9816 reference count is 1. The reference count 1 is owned by C<rv>.
9822 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9826 PERL_ARGS_ASSERT_NEWSVRV;
9830 SV_CHECK_THINKFIRST_COW_DROP(rv);
9832 if (SvTYPE(rv) >= SVt_PVMG) {
9833 const U32 refcnt = SvREFCNT(rv);
9837 SvREFCNT(rv) = refcnt;
9839 sv_upgrade(rv, SVt_IV);
9840 } else if (SvROK(rv)) {
9841 SvREFCNT_dec(SvRV(rv));
9843 prepare_SV_for_RV(rv);
9851 HV* const stash = gv_stashpv(classname, GV_ADD);
9852 (void)sv_bless(rv, stash);
9858 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9860 SV * const lv = newSV_type(SVt_PVLV);
9861 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9863 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9864 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9865 LvSTARGOFF(lv) = ix;
9866 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9871 =for apidoc sv_setref_pv
9873 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9874 argument will be upgraded to an RV. That RV will be modified to point to
9875 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
9876 into the SV. The C<classname> argument indicates the package for the
9877 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9878 will have a reference count of 1, and the RV will be returned.
9880 Do not use with other Perl types such as HV, AV, SV, CV, because those
9881 objects will become corrupted by the pointer copy process.
9883 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
9889 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
9891 PERL_ARGS_ASSERT_SV_SETREF_PV;
9894 sv_setsv(rv, &PL_sv_undef);
9898 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
9903 =for apidoc sv_setref_iv
9905 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
9906 argument will be upgraded to an RV. That RV will be modified to point to
9907 the new SV. The C<classname> argument indicates the package for the
9908 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9909 will have a reference count of 1, and the RV will be returned.
9915 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
9917 PERL_ARGS_ASSERT_SV_SETREF_IV;
9919 sv_setiv(newSVrv(rv,classname), iv);
9924 =for apidoc sv_setref_uv
9926 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
9927 argument will be upgraded to an RV. That RV will be modified to point to
9928 the new SV. The C<classname> argument indicates the package for the
9929 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9930 will have a reference count of 1, and the RV will be returned.
9936 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
9938 PERL_ARGS_ASSERT_SV_SETREF_UV;
9940 sv_setuv(newSVrv(rv,classname), uv);
9945 =for apidoc sv_setref_nv
9947 Copies a double into a new SV, optionally blessing the SV. The C<rv>
9948 argument will be upgraded to an RV. That RV will be modified to point to
9949 the new SV. The C<classname> argument indicates the package for the
9950 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9951 will have a reference count of 1, and the RV will be returned.
9957 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
9959 PERL_ARGS_ASSERT_SV_SETREF_NV;
9961 sv_setnv(newSVrv(rv,classname), nv);
9966 =for apidoc sv_setref_pvn
9968 Copies a string into a new SV, optionally blessing the SV. The length of the
9969 string must be specified with C<n>. The C<rv> argument will be upgraded to
9970 an RV. That RV will be modified to point to the new SV. The C<classname>
9971 argument indicates the package for the blessing. Set C<classname> to
9972 C<NULL> to avoid the blessing. The new SV will have a reference count
9973 of 1, and the RV will be returned.
9975 Note that C<sv_setref_pv> copies the pointer while this copies the string.
9981 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
9982 const char *const pv, const STRLEN n)
9984 PERL_ARGS_ASSERT_SV_SETREF_PVN;
9986 sv_setpvn(newSVrv(rv,classname), pv, n);
9991 =for apidoc sv_bless
9993 Blesses an SV into a specified package. The SV must be an RV. The package
9994 must be designated by its stash (see C<gv_stashpv()>). The reference count
9995 of the SV is unaffected.
10001 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10004 HV *oldstash = NULL;
10006 PERL_ARGS_ASSERT_SV_BLESS;
10010 Perl_croak(aTHX_ "Can't bless non-reference value");
10012 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY)) {
10013 if (SvREADONLY(tmpRef))
10014 Perl_croak_no_modify();
10015 if (SvOBJECT(tmpRef)) {
10016 oldstash = SvSTASH(tmpRef);
10019 SvOBJECT_on(tmpRef);
10020 SvUPGRADE(tmpRef, SVt_PVMG);
10021 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10022 SvREFCNT_dec(oldstash);
10024 if(SvSMAGICAL(tmpRef))
10025 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10033 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10034 * as it is after unglobbing it.
10037 PERL_STATIC_INLINE void
10038 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10042 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10044 PERL_ARGS_ASSERT_SV_UNGLOB;
10046 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10048 if (!(flags & SV_COW_DROP_PV))
10049 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10051 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10053 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10054 && HvNAME_get(stash))
10055 mro_method_changed_in(stash);
10056 gp_free(MUTABLE_GV(sv));
10059 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10060 GvSTASH(sv) = NULL;
10063 if (GvNAME_HEK(sv)) {
10064 unshare_hek(GvNAME_HEK(sv));
10066 isGV_with_GP_off(sv);
10068 if(SvTYPE(sv) == SVt_PVGV) {
10069 /* need to keep SvANY(sv) in the right arena */
10070 xpvmg = new_XPVMG();
10071 StructCopy(SvANY(sv), xpvmg, XPVMG);
10072 del_XPVGV(SvANY(sv));
10075 SvFLAGS(sv) &= ~SVTYPEMASK;
10076 SvFLAGS(sv) |= SVt_PVMG;
10079 /* Intentionally not calling any local SET magic, as this isn't so much a
10080 set operation as merely an internal storage change. */
10081 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10082 else sv_setsv_flags(sv, temp, 0);
10084 if ((const GV *)sv == PL_last_in_gv)
10085 PL_last_in_gv = NULL;
10086 else if ((const GV *)sv == PL_statgv)
10091 =for apidoc sv_unref_flags
10093 Unsets the RV status of the SV, and decrements the reference count of
10094 whatever was being referenced by the RV. This can almost be thought of
10095 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10096 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10097 (otherwise the decrementing is conditional on the reference count being
10098 different from one or the reference being a readonly SV).
10105 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10107 SV* const target = SvRV(ref);
10109 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10111 if (SvWEAKREF(ref)) {
10112 sv_del_backref(target, ref);
10113 SvWEAKREF_off(ref);
10114 SvRV_set(ref, NULL);
10117 SvRV_set(ref, NULL);
10119 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10120 assigned to as BEGIN {$a = \"Foo"} will fail. */
10121 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10122 SvREFCNT_dec_NN(target);
10123 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10124 sv_2mortal(target); /* Schedule for freeing later */
10128 =for apidoc sv_untaint
10130 Untaint an SV. Use C<SvTAINTED_off> instead.
10136 Perl_sv_untaint(pTHX_ SV *const sv)
10138 PERL_ARGS_ASSERT_SV_UNTAINT;
10139 PERL_UNUSED_CONTEXT;
10141 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10142 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10149 =for apidoc sv_tainted
10151 Test an SV for taintedness. Use C<SvTAINTED> instead.
10157 Perl_sv_tainted(pTHX_ SV *const sv)
10159 PERL_ARGS_ASSERT_SV_TAINTED;
10160 PERL_UNUSED_CONTEXT;
10162 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10163 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10164 if (mg && (mg->mg_len & 1) )
10171 =for apidoc sv_setpviv
10173 Copies an integer into the given SV, also updating its string value.
10174 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10180 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10182 char buf[TYPE_CHARS(UV)];
10184 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10186 PERL_ARGS_ASSERT_SV_SETPVIV;
10188 sv_setpvn(sv, ptr, ebuf - ptr);
10192 =for apidoc sv_setpviv_mg
10194 Like C<sv_setpviv>, but also handles 'set' magic.
10200 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10202 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10204 sv_setpviv(sv, iv);
10208 #if defined(PERL_IMPLICIT_CONTEXT)
10210 /* pTHX_ magic can't cope with varargs, so this is a no-context
10211 * version of the main function, (which may itself be aliased to us).
10212 * Don't access this version directly.
10216 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10221 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10223 va_start(args, pat);
10224 sv_vsetpvf(sv, pat, &args);
10228 /* pTHX_ magic can't cope with varargs, so this is a no-context
10229 * version of the main function, (which may itself be aliased to us).
10230 * Don't access this version directly.
10234 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10239 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10241 va_start(args, pat);
10242 sv_vsetpvf_mg(sv, pat, &args);
10248 =for apidoc sv_setpvf
10250 Works like C<sv_catpvf> but copies the text into the SV instead of
10251 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10257 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10261 PERL_ARGS_ASSERT_SV_SETPVF;
10263 va_start(args, pat);
10264 sv_vsetpvf(sv, pat, &args);
10269 =for apidoc sv_vsetpvf
10271 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10272 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10274 Usually used via its frontend C<sv_setpvf>.
10280 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10282 PERL_ARGS_ASSERT_SV_VSETPVF;
10284 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10288 =for apidoc sv_setpvf_mg
10290 Like C<sv_setpvf>, but also handles 'set' magic.
10296 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10300 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10302 va_start(args, pat);
10303 sv_vsetpvf_mg(sv, pat, &args);
10308 =for apidoc sv_vsetpvf_mg
10310 Like C<sv_vsetpvf>, but also handles 'set' magic.
10312 Usually used via its frontend C<sv_setpvf_mg>.
10318 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10320 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10322 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10326 #if defined(PERL_IMPLICIT_CONTEXT)
10328 /* pTHX_ magic can't cope with varargs, so this is a no-context
10329 * version of the main function, (which may itself be aliased to us).
10330 * Don't access this version directly.
10334 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10339 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10341 va_start(args, pat);
10342 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10346 /* pTHX_ magic can't cope with varargs, so this is a no-context
10347 * version of the main function, (which may itself be aliased to us).
10348 * Don't access this version directly.
10352 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10357 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10359 va_start(args, pat);
10360 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10367 =for apidoc sv_catpvf
10369 Processes its arguments like C<sprintf> and appends the formatted
10370 output to an SV. If the appended data contains "wide" characters
10371 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10372 and characters >255 formatted with %c), the original SV might get
10373 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10374 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10375 valid UTF-8; if the original SV was bytes, the pattern should be too.
10380 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10384 PERL_ARGS_ASSERT_SV_CATPVF;
10386 va_start(args, pat);
10387 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10392 =for apidoc sv_vcatpvf
10394 Processes its arguments like C<vsprintf> and appends the formatted output
10395 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10397 Usually used via its frontend C<sv_catpvf>.
10403 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10405 PERL_ARGS_ASSERT_SV_VCATPVF;
10407 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10411 =for apidoc sv_catpvf_mg
10413 Like C<sv_catpvf>, but also handles 'set' magic.
10419 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10423 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10425 va_start(args, pat);
10426 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10432 =for apidoc sv_vcatpvf_mg
10434 Like C<sv_vcatpvf>, but also handles 'set' magic.
10436 Usually used via its frontend C<sv_catpvf_mg>.
10442 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10444 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10446 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10451 =for apidoc sv_vsetpvfn
10453 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10456 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10462 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10463 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10465 PERL_ARGS_ASSERT_SV_VSETPVFN;
10468 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10473 * Warn of missing argument to sprintf, and then return a defined value
10474 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10477 S_vcatpvfn_missing_argument(pTHX) {
10478 if (ckWARN(WARN_MISSING)) {
10479 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10480 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10487 S_expect_number(pTHX_ char **const pattern)
10491 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10493 switch (**pattern) {
10494 case '1': case '2': case '3':
10495 case '4': case '5': case '6':
10496 case '7': case '8': case '9':
10497 var = *(*pattern)++ - '0';
10498 while (isDIGIT(**pattern)) {
10499 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10501 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10509 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10511 const int neg = nv < 0;
10514 PERL_ARGS_ASSERT_F0CONVERT;
10522 if (uv & 1 && uv == nv)
10523 uv--; /* Round to even */
10525 const unsigned dig = uv % 10;
10527 } while (uv /= 10);
10538 =for apidoc sv_vcatpvfn
10540 =for apidoc sv_vcatpvfn_flags
10542 Processes its arguments like C<vsprintf> and appends the formatted output
10543 to an SV. Uses an array of SVs if the C style variable argument list is
10544 missing (NULL). When running with taint checks enabled, indicates via
10545 C<maybe_tainted> if results are untrustworthy (often due to the use of
10548 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10550 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10555 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10556 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10557 vec_utf8 = DO_UTF8(vecsv);
10559 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10562 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10563 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10565 PERL_ARGS_ASSERT_SV_VCATPVFN;
10567 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10571 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10572 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
10577 const char *patend;
10580 static const char nullstr[] = "(null)";
10582 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
10583 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
10585 /* Times 4: a decimal digit takes more than 3 binary digits.
10586 * NV_DIG: mantissa takes than many decimal digits.
10587 * Plus 32: Playing safe. */
10588 char ebuf[IV_DIG * 4 + NV_DIG + 32];
10589 /* large enough for "%#.#f" --chip */
10590 /* what about long double NVs? --jhi */
10591 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
10593 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
10595 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
10596 PERL_UNUSED_ARG(maybe_tainted);
10598 if (flags & SV_GMAGIC)
10601 /* no matter what, this is a string now */
10602 (void)SvPV_force_nomg(sv, origlen);
10604 /* special-case "", "%s", and "%-p" (SVf - see below) */
10606 if (svmax && ckWARN(WARN_REDUNDANT))
10607 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10608 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10611 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
10612 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10613 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10614 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10617 const char * const s = va_arg(*args, char*);
10618 sv_catpv_nomg(sv, s ? s : nullstr);
10620 else if (svix < svmax) {
10621 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
10622 SvGETMAGIC(*svargs);
10623 sv_catsv_nomg(sv, *svargs);
10626 S_vcatpvfn_missing_argument(aTHX);
10629 if (args && patlen == 3 && pat[0] == '%' &&
10630 pat[1] == '-' && pat[2] == 'p') {
10631 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10632 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10633 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10634 argsv = MUTABLE_SV(va_arg(*args, void*));
10635 sv_catsv_nomg(sv, argsv);
10639 #ifndef USE_LONG_DOUBLE
10640 /* special-case "%.<number>[gf]" */
10641 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
10642 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
10643 unsigned digits = 0;
10647 while (*pp >= '0' && *pp <= '9')
10648 digits = 10 * digits + (*pp++ - '0');
10650 /* XXX: Why do this `svix < svmax` test? Couldn't we just
10651 format the first argument and WARN_REDUNDANT if svmax > 1?
10652 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
10653 if (pp - pat == (int)patlen - 1 && svix < svmax) {
10654 const NV nv = SvNV(*svargs);
10656 /* Add check for digits != 0 because it seems that some
10657 gconverts are buggy in this case, and we don't yet have
10658 a Configure test for this. */
10659 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
10660 /* 0, point, slack */
10661 STORE_LC_NUMERIC_SET_TO_NEEDED();
10662 PERL_UNUSED_RESULT(Gconvert(nv, (int)digits, 0, ebuf));
10663 sv_catpv_nomg(sv, ebuf);
10664 if (*ebuf) /* May return an empty string for digits==0 */
10667 } else if (!digits) {
10670 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
10671 sv_catpvn_nomg(sv, p, l);
10677 #endif /* !USE_LONG_DOUBLE */
10679 if (!args && svix < svmax && DO_UTF8(*svargs))
10682 patend = (char*)pat + patlen;
10683 for (p = (char*)pat; p < patend; p = q) {
10686 bool vectorize = FALSE;
10687 bool vectorarg = FALSE;
10688 bool vec_utf8 = FALSE;
10694 bool has_precis = FALSE;
10696 const I32 osvix = svix;
10697 bool is_utf8 = FALSE; /* is this item utf8? */
10698 #ifdef HAS_LDBL_SPRINTF_BUG
10699 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
10700 with sfio - Allen <allens@cpan.org> */
10701 bool fix_ldbl_sprintf_bug = FALSE;
10705 U8 utf8buf[UTF8_MAXBYTES+1];
10706 STRLEN esignlen = 0;
10708 const char *eptr = NULL;
10709 const char *fmtstart;
10712 const U8 *vecstr = NULL;
10719 /* we need a long double target in case HAS_LONG_DOUBLE but
10720 not USE_LONG_DOUBLE
10722 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE
10730 const char *dotstr = ".";
10731 STRLEN dotstrlen = 1;
10732 I32 efix = 0; /* explicit format parameter index */
10733 I32 ewix = 0; /* explicit width index */
10734 I32 epix = 0; /* explicit precision index */
10735 I32 evix = 0; /* explicit vector index */
10736 bool asterisk = FALSE;
10738 /* echo everything up to the next format specification */
10739 for (q = p; q < patend && *q != '%'; ++q) ;
10741 if (has_utf8 && !pat_utf8)
10742 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
10744 sv_catpvn_nomg(sv, p, q - p);
10753 We allow format specification elements in this order:
10754 \d+\$ explicit format parameter index
10756 v|\*(\d+\$)?v vector with optional (optionally specified) arg
10757 0 flag (as above): repeated to allow "v02"
10758 \d+|\*(\d+\$)? width using optional (optionally specified) arg
10759 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
10761 [%bcdefginopsuxDFOUX] format (mandatory)
10766 As of perl5.9.3, printf format checking is on by default.
10767 Internally, perl uses %p formats to provide an escape to
10768 some extended formatting. This block deals with those
10769 extensions: if it does not match, (char*)q is reset and
10770 the normal format processing code is used.
10772 Currently defined extensions are:
10773 %p include pointer address (standard)
10774 %-p (SVf) include an SV (previously %_)
10775 %-<num>p include an SV with precision <num>
10777 %3p include a HEK with precision of 256
10778 %4p char* preceded by utf8 flag and length
10779 %<num>p (where num is 1 or > 4) reserved for future
10782 Robin Barker 2005-07-14 (but modified since)
10784 %1p (VDf) removed. RMB 2007-10-19
10791 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
10792 /* The argument has already gone through cBOOL, so the cast
10794 is_utf8 = (bool)va_arg(*args, int);
10795 elen = va_arg(*args, UV);
10796 eptr = va_arg(*args, char *);
10797 q += sizeof(UTF8f)-1;
10800 n = expect_number(&q);
10802 if (sv) { /* SVf */
10807 argsv = MUTABLE_SV(va_arg(*args, void*));
10808 eptr = SvPV_const(argsv, elen);
10809 if (DO_UTF8(argsv))
10813 else if (n==2 || n==3) { /* HEKf */
10814 HEK * const hek = va_arg(*args, HEK *);
10815 eptr = HEK_KEY(hek);
10816 elen = HEK_LEN(hek);
10817 if (HEK_UTF8(hek)) is_utf8 = TRUE;
10818 if (n==3) precis = 256, has_precis = TRUE;
10822 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
10823 "internal %%<num>p might conflict with future printf extensions");
10829 if ( (width = expect_number(&q)) ) {
10833 if (!no_redundant_warning)
10834 /* I've forgotten if it's a better
10835 micro-optimization to always set this or to
10836 only set it if it's unset */
10837 no_redundant_warning = TRUE;
10849 if (plus == '+' && *q == ' ') /* '+' over ' ' */
10878 if ( (ewix = expect_number(&q)) )
10887 if ((vectorarg = asterisk)) {
10900 width = expect_number(&q);
10903 if (vectorize && vectorarg) {
10904 /* vectorizing, but not with the default "." */
10906 vecsv = va_arg(*args, SV*);
10908 vecsv = (evix > 0 && evix <= svmax)
10909 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
10911 vecsv = svix < svmax
10912 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
10914 dotstr = SvPV_const(vecsv, dotstrlen);
10915 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
10916 bad with tied or overloaded values that return UTF8. */
10917 if (DO_UTF8(vecsv))
10919 else if (has_utf8) {
10920 vecsv = sv_mortalcopy(vecsv);
10921 sv_utf8_upgrade(vecsv);
10922 dotstr = SvPV_const(vecsv, dotstrlen);
10929 i = va_arg(*args, int);
10931 i = (ewix ? ewix <= svmax : svix < svmax) ?
10932 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
10934 width = (i < 0) ? -i : i;
10944 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
10946 /* XXX: todo, support specified precision parameter */
10950 i = va_arg(*args, int);
10952 i = (ewix ? ewix <= svmax : svix < svmax)
10953 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
10955 has_precis = !(i < 0);
10959 while (isDIGIT(*q))
10960 precis = precis * 10 + (*q++ - '0');
10969 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
10970 vecsv = svargs[efix ? efix-1 : svix++];
10971 vecstr = (U8*)SvPV_const(vecsv,veclen);
10972 vec_utf8 = DO_UTF8(vecsv);
10974 /* if this is a version object, we need to convert
10975 * back into v-string notation and then let the
10976 * vectorize happen normally
10978 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
10979 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
10980 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
10981 "vector argument not supported with alpha versions");
10984 vecsv = sv_newmortal();
10985 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
10987 vecstr = (U8*)SvPV_const(vecsv, veclen);
10988 vec_utf8 = DO_UTF8(vecsv);
11002 case 'I': /* Ix, I32x, and I64x */
11003 # ifdef USE_64_BIT_INT
11004 if (q[1] == '6' && q[2] == '4') {
11010 if (q[1] == '3' && q[2] == '2') {
11014 # ifdef USE_64_BIT_INT
11020 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11032 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11033 if (*q == 'l') { /* lld, llf */
11042 if (*++q == 'h') { /* hhd, hhu */
11071 if (!vectorize && !args) {
11073 const I32 i = efix-1;
11074 argsv = (i >= 0 && i < svmax)
11075 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11077 argsv = (svix >= 0 && svix < svmax)
11078 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11082 switch (c = *q++) {
11089 uv = (args) ? va_arg(*args, int) : SvIV(argsv);
11091 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11093 eptr = (char*)utf8buf;
11094 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11108 eptr = va_arg(*args, char*);
11110 elen = strlen(eptr);
11112 eptr = (char *)nullstr;
11113 elen = sizeof nullstr - 1;
11117 eptr = SvPV_const(argsv, elen);
11118 if (DO_UTF8(argsv)) {
11119 STRLEN old_precis = precis;
11120 if (has_precis && precis < elen) {
11121 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11122 STRLEN p = precis > ulen ? ulen : precis;
11123 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11124 /* sticks at end */
11126 if (width) { /* fudge width (can't fudge elen) */
11127 if (has_precis && precis < elen)
11128 width += precis - old_precis;
11131 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11138 if (has_precis && precis < elen)
11145 if (alt || vectorize)
11147 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11165 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11174 esignbuf[esignlen++] = plus;
11178 case 'c': iv = (char)va_arg(*args, int); break;
11179 case 'h': iv = (short)va_arg(*args, int); break;
11180 case 'l': iv = va_arg(*args, long); break;
11181 case 'V': iv = va_arg(*args, IV); break;
11182 case 'z': iv = va_arg(*args, SSize_t); break;
11183 #ifdef HAS_PTRDIFF_T
11184 case 't': iv = va_arg(*args, ptrdiff_t); break;
11186 default: iv = va_arg(*args, int); break;
11188 case 'j': iv = va_arg(*args, intmax_t); break;
11192 iv = va_arg(*args, Quad_t); break;
11199 IV tiv = SvIV(argsv); /* work around GCC bug #13488 */
11201 case 'c': iv = (char)tiv; break;
11202 case 'h': iv = (short)tiv; break;
11203 case 'l': iv = (long)tiv; break;
11205 default: iv = tiv; break;
11208 iv = (Quad_t)tiv; break;
11214 if ( !vectorize ) /* we already set uv above */
11219 esignbuf[esignlen++] = plus;
11223 esignbuf[esignlen++] = '-';
11267 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11278 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11279 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11280 case 'l': uv = va_arg(*args, unsigned long); break;
11281 case 'V': uv = va_arg(*args, UV); break;
11282 case 'z': uv = va_arg(*args, Size_t); break;
11283 #ifdef HAS_PTRDIFF_T
11284 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11287 case 'j': uv = va_arg(*args, uintmax_t); break;
11289 default: uv = va_arg(*args, unsigned); break;
11292 uv = va_arg(*args, Uquad_t); break;
11299 UV tuv = SvUV(argsv); /* work around GCC bug #13488 */
11301 case 'c': uv = (unsigned char)tuv; break;
11302 case 'h': uv = (unsigned short)tuv; break;
11303 case 'l': uv = (unsigned long)tuv; break;
11305 default: uv = tuv; break;
11308 uv = (Uquad_t)tuv; break;
11317 char *ptr = ebuf + sizeof ebuf;
11318 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11324 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11328 } while (uv >>= 4);
11330 esignbuf[esignlen++] = '0';
11331 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11337 *--ptr = '0' + dig;
11338 } while (uv >>= 3);
11339 if (alt && *ptr != '0')
11345 *--ptr = '0' + dig;
11346 } while (uv >>= 1);
11348 esignbuf[esignlen++] = '0';
11349 esignbuf[esignlen++] = c;
11352 default: /* it had better be ten or less */
11355 *--ptr = '0' + dig;
11356 } while (uv /= base);
11359 elen = (ebuf + sizeof ebuf) - ptr;
11363 zeros = precis - elen;
11364 else if (precis == 0 && elen == 1 && *eptr == '0'
11365 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11368 /* a precision nullifies the 0 flag. */
11375 /* FLOATING POINT */
11378 c = 'f'; /* maybe %F isn't supported here */
11380 case 'e': case 'E':
11382 case 'g': case 'G':
11386 /* This is evil, but floating point is even more evil */
11388 /* for SV-style calling, we can only get NV
11389 for C-style calling, we assume %f is double;
11390 for simplicity we allow any of %Lf, %llf, %qf for long double
11394 #if defined(USE_LONG_DOUBLE)
11398 /* [perl #20339] - we should accept and ignore %lf rather than die */
11402 #if defined(USE_LONG_DOUBLE)
11403 intsize = args ? 0 : 'q';
11407 #if defined(HAS_LONG_DOUBLE)
11420 /* now we need (long double) if intsize == 'q', else (double) */
11422 #if LONG_DOUBLESIZE > DOUBLESIZE
11424 va_arg(*args, long double) :
11425 va_arg(*args, double)
11427 va_arg(*args, double)
11432 /* nv * 0 will be NaN for NaN, +Inf and -Inf, and 0 for anything
11433 else. frexp() has some unspecified behaviour for those three */
11434 if (c != 'e' && c != 'E' && (nv * 0) == 0) {
11436 /* FIXME: if HAS_LONG_DOUBLE but not USE_LONG_DOUBLE this
11437 will cast our (long double) to (double) */
11438 (void)Perl_frexp(nv, &i);
11439 if (i == PERL_INT_MIN)
11440 Perl_die(aTHX_ "panic: frexp");
11442 need = BIT_DIGITS(i);
11444 need += has_precis ? precis : 6; /* known default */
11449 #ifdef HAS_LDBL_SPRINTF_BUG
11450 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11451 with sfio - Allen <allens@cpan.org> */
11454 # define MY_DBL_MAX DBL_MAX
11455 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11456 # if DOUBLESIZE >= 8
11457 # define MY_DBL_MAX 1.7976931348623157E+308L
11459 # define MY_DBL_MAX 3.40282347E+38L
11463 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
11464 # define MY_DBL_MAX_BUG 1L
11466 # define MY_DBL_MAX_BUG MY_DBL_MAX
11470 # define MY_DBL_MIN DBL_MIN
11471 # else /* XXX guessing! -Allen */
11472 # if DOUBLESIZE >= 8
11473 # define MY_DBL_MIN 2.2250738585072014E-308L
11475 # define MY_DBL_MIN 1.17549435E-38L
11479 if ((intsize == 'q') && (c == 'f') &&
11480 ((nv < MY_DBL_MAX_BUG) && (nv > -MY_DBL_MAX_BUG)) &&
11481 (need < DBL_DIG)) {
11482 /* it's going to be short enough that
11483 * long double precision is not needed */
11485 if ((nv <= 0L) && (nv >= -0L))
11486 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
11488 /* would use Perl_fp_class as a double-check but not
11489 * functional on IRIX - see perl.h comments */
11491 if ((nv >= MY_DBL_MIN) || (nv <= -MY_DBL_MIN)) {
11492 /* It's within the range that a double can represent */
11493 #if defined(DBL_MAX) && !defined(DBL_MIN)
11494 if ((nv >= ((long double)1/DBL_MAX)) ||
11495 (nv <= (-(long double)1/DBL_MAX)))
11497 fix_ldbl_sprintf_bug = TRUE;
11500 if (fix_ldbl_sprintf_bug == TRUE) {
11510 # undef MY_DBL_MAX_BUG
11513 #endif /* HAS_LDBL_SPRINTF_BUG */
11515 need += 20; /* fudge factor */
11516 if (PL_efloatsize < need) {
11517 Safefree(PL_efloatbuf);
11518 PL_efloatsize = need + 20; /* more fudge */
11519 Newx(PL_efloatbuf, PL_efloatsize, char);
11520 PL_efloatbuf[0] = '\0';
11523 if ( !(width || left || plus || alt) && fill != '0'
11524 && has_precis && intsize != 'q' ) { /* Shortcuts */
11525 /* See earlier comment about buggy Gconvert when digits,
11527 if ( c == 'g' && precis) {
11528 STORE_LC_NUMERIC_SET_TO_NEEDED();
11529 PERL_UNUSED_RESULT(Gconvert((NV)nv, (int)precis, 0, PL_efloatbuf));
11530 /* May return an empty string for digits==0 */
11531 if (*PL_efloatbuf) {
11532 elen = strlen(PL_efloatbuf);
11533 goto float_converted;
11535 } else if ( c == 'f' && !precis) {
11536 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
11541 char *ptr = ebuf + sizeof ebuf;
11544 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
11545 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
11546 if (intsize == 'q') {
11547 /* Copy the one or more characters in a long double
11548 * format before the 'base' ([efgEFG]) character to
11549 * the format string. */
11550 static char const prifldbl[] = PERL_PRIfldbl;
11551 char const *p = prifldbl + sizeof(prifldbl) - 3;
11552 while (p >= prifldbl) { *--ptr = *p--; }
11557 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11562 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11574 /* No taint. Otherwise we are in the strange situation
11575 * where printf() taints but print($float) doesn't.
11578 STORE_LC_NUMERIC_SET_TO_NEEDED();
11580 /* hopefully the above makes ptr a very constrained format
11581 * that is safe to use, even though it's not literal */
11582 GCC_DIAG_IGNORE(-Wformat-nonliteral);
11583 #if defined(HAS_LONG_DOUBLE)
11584 elen = ((intsize == 'q')
11585 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, nv)
11586 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)nv));
11588 elen = my_sprintf(PL_efloatbuf, ptr, nv);
11593 eptr = PL_efloatbuf;
11595 #ifdef USE_LOCALE_NUMERIC
11596 /* If the decimal point character in the string is UTF-8, make the
11598 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
11599 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
11612 i = SvCUR(sv) - origlen;
11615 case 'c': *(va_arg(*args, char*)) = i; break;
11616 case 'h': *(va_arg(*args, short*)) = i; break;
11617 default: *(va_arg(*args, int*)) = i; break;
11618 case 'l': *(va_arg(*args, long*)) = i; break;
11619 case 'V': *(va_arg(*args, IV*)) = i; break;
11620 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
11621 #ifdef HAS_PTRDIFF_T
11622 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
11625 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
11629 *(va_arg(*args, Quad_t*)) = i; break;
11636 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
11637 continue; /* not "break" */
11644 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
11645 && ckWARN(WARN_PRINTF))
11647 SV * const msg = sv_newmortal();
11648 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
11649 (PL_op->op_type == OP_PRTF) ? "" : "s");
11650 if (fmtstart < patend) {
11651 const char * const fmtend = q < patend ? q : patend;
11653 sv_catpvs(msg, "\"%");
11654 for (f = fmtstart; f < fmtend; f++) {
11656 sv_catpvn_nomg(msg, f, 1);
11658 Perl_sv_catpvf(aTHX_ msg,
11659 "\\%03"UVof, (UV)*f & 0xFF);
11662 sv_catpvs(msg, "\"");
11664 sv_catpvs(msg, "end of string");
11666 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
11669 /* output mangled stuff ... */
11675 /* ... right here, because formatting flags should not apply */
11676 SvGROW(sv, SvCUR(sv) + elen + 1);
11678 Copy(eptr, p, elen, char);
11681 SvCUR_set(sv, p - SvPVX_const(sv));
11683 continue; /* not "break" */
11686 if (is_utf8 != has_utf8) {
11689 sv_utf8_upgrade(sv);
11692 const STRLEN old_elen = elen;
11693 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
11694 sv_utf8_upgrade(nsv);
11695 eptr = SvPVX_const(nsv);
11698 if (width) { /* fudge width (can't fudge elen) */
11699 width += elen - old_elen;
11705 have = esignlen + zeros + elen;
11707 croak_memory_wrap();
11709 need = (have > width ? have : width);
11712 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
11713 croak_memory_wrap();
11714 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
11716 if (esignlen && fill == '0') {
11718 for (i = 0; i < (int)esignlen; i++)
11719 *p++ = esignbuf[i];
11721 if (gap && !left) {
11722 memset(p, fill, gap);
11725 if (esignlen && fill != '0') {
11727 for (i = 0; i < (int)esignlen; i++)
11728 *p++ = esignbuf[i];
11732 for (i = zeros; i; i--)
11736 Copy(eptr, p, elen, char);
11740 memset(p, ' ', gap);
11745 Copy(dotstr, p, dotstrlen, char);
11749 vectorize = FALSE; /* done iterating over vecstr */
11756 SvCUR_set(sv, p - SvPVX_const(sv));
11763 /* Now that we've consumed all our printf format arguments (svix)
11764 * do we have things left on the stack that we didn't use?
11766 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
11767 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11768 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11773 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
11777 /* =========================================================================
11779 =head1 Cloning an interpreter
11783 All the macros and functions in this section are for the private use of
11784 the main function, perl_clone().
11786 The foo_dup() functions make an exact copy of an existing foo thingy.
11787 During the course of a cloning, a hash table is used to map old addresses
11788 to new addresses. The table is created and manipulated with the
11789 ptr_table_* functions.
11791 * =========================================================================*/
11794 #if defined(USE_ITHREADS)
11796 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
11797 #ifndef GpREFCNT_inc
11798 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
11802 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
11803 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
11804 If this changes, please unmerge ss_dup.
11805 Likewise, sv_dup_inc_multiple() relies on this fact. */
11806 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
11807 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
11808 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11809 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
11810 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11811 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
11812 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
11813 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
11814 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
11815 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
11816 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
11817 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
11818 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11820 /* clone a parser */
11823 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
11827 PERL_ARGS_ASSERT_PARSER_DUP;
11832 /* look for it in the table first */
11833 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
11837 /* create anew and remember what it is */
11838 Newxz(parser, 1, yy_parser);
11839 ptr_table_store(PL_ptr_table, proto, parser);
11841 /* XXX these not yet duped */
11842 parser->old_parser = NULL;
11843 parser->stack = NULL;
11845 parser->stack_size = 0;
11846 /* XXX parser->stack->state = 0; */
11848 /* XXX eventually, just Copy() most of the parser struct ? */
11850 parser->lex_brackets = proto->lex_brackets;
11851 parser->lex_casemods = proto->lex_casemods;
11852 parser->lex_brackstack = savepvn(proto->lex_brackstack,
11853 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
11854 parser->lex_casestack = savepvn(proto->lex_casestack,
11855 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
11856 parser->lex_defer = proto->lex_defer;
11857 parser->lex_dojoin = proto->lex_dojoin;
11858 parser->lex_expect = proto->lex_expect;
11859 parser->lex_formbrack = proto->lex_formbrack;
11860 parser->lex_inpat = proto->lex_inpat;
11861 parser->lex_inwhat = proto->lex_inwhat;
11862 parser->lex_op = proto->lex_op;
11863 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
11864 parser->lex_starts = proto->lex_starts;
11865 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
11866 parser->multi_close = proto->multi_close;
11867 parser->multi_open = proto->multi_open;
11868 parser->multi_start = proto->multi_start;
11869 parser->multi_end = proto->multi_end;
11870 parser->preambled = proto->preambled;
11871 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
11872 parser->linestr = sv_dup_inc(proto->linestr, param);
11873 parser->expect = proto->expect;
11874 parser->copline = proto->copline;
11875 parser->last_lop_op = proto->last_lop_op;
11876 parser->lex_state = proto->lex_state;
11877 parser->rsfp = fp_dup(proto->rsfp, '<', param);
11878 /* rsfp_filters entries have fake IoDIRP() */
11879 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
11880 parser->in_my = proto->in_my;
11881 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
11882 parser->error_count = proto->error_count;
11885 parser->linestr = sv_dup_inc(proto->linestr, param);
11888 char * const ols = SvPVX(proto->linestr);
11889 char * const ls = SvPVX(parser->linestr);
11891 parser->bufptr = ls + (proto->bufptr >= ols ?
11892 proto->bufptr - ols : 0);
11893 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
11894 proto->oldbufptr - ols : 0);
11895 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
11896 proto->oldoldbufptr - ols : 0);
11897 parser->linestart = ls + (proto->linestart >= ols ?
11898 proto->linestart - ols : 0);
11899 parser->last_uni = ls + (proto->last_uni >= ols ?
11900 proto->last_uni - ols : 0);
11901 parser->last_lop = ls + (proto->last_lop >= ols ?
11902 proto->last_lop - ols : 0);
11904 parser->bufend = ls + SvCUR(parser->linestr);
11907 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
11910 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
11911 Copy(proto->nexttype, parser->nexttype, 5, I32);
11912 parser->nexttoke = proto->nexttoke;
11914 /* XXX should clone saved_curcop here, but we aren't passed
11915 * proto_perl; so do it in perl_clone_using instead */
11921 /* duplicate a file handle */
11924 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
11928 PERL_ARGS_ASSERT_FP_DUP;
11929 PERL_UNUSED_ARG(type);
11932 return (PerlIO*)NULL;
11934 /* look for it in the table first */
11935 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
11939 /* create anew and remember what it is */
11940 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
11941 ptr_table_store(PL_ptr_table, fp, ret);
11945 /* duplicate a directory handle */
11948 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
11952 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
11954 const Direntry_t *dirent;
11955 char smallbuf[256];
11961 PERL_UNUSED_CONTEXT;
11962 PERL_ARGS_ASSERT_DIRP_DUP;
11967 /* look for it in the table first */
11968 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
11972 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
11974 PERL_UNUSED_ARG(param);
11978 /* open the current directory (so we can switch back) */
11979 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
11981 /* chdir to our dir handle and open the present working directory */
11982 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
11983 PerlDir_close(pwd);
11984 return (DIR *)NULL;
11986 /* Now we should have two dir handles pointing to the same dir. */
11988 /* Be nice to the calling code and chdir back to where we were. */
11989 /* XXX If this fails, then what? */
11990 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
11992 /* We have no need of the pwd handle any more. */
11993 PerlDir_close(pwd);
11996 # define d_namlen(d) (d)->d_namlen
11998 # define d_namlen(d) strlen((d)->d_name)
12000 /* Iterate once through dp, to get the file name at the current posi-
12001 tion. Then step back. */
12002 pos = PerlDir_tell(dp);
12003 if ((dirent = PerlDir_read(dp))) {
12004 len = d_namlen(dirent);
12005 if (len <= sizeof smallbuf) name = smallbuf;
12006 else Newx(name, len, char);
12007 Move(dirent->d_name, name, len, char);
12009 PerlDir_seek(dp, pos);
12011 /* Iterate through the new dir handle, till we find a file with the
12013 if (!dirent) /* just before the end */
12015 pos = PerlDir_tell(ret);
12016 if (PerlDir_read(ret)) continue; /* not there yet */
12017 PerlDir_seek(ret, pos); /* step back */
12021 const long pos0 = PerlDir_tell(ret);
12023 pos = PerlDir_tell(ret);
12024 if ((dirent = PerlDir_read(ret))) {
12025 if (len == (STRLEN)d_namlen(dirent)
12026 && memEQ(name, dirent->d_name, len)) {
12028 PerlDir_seek(ret, pos); /* step back */
12031 /* else we are not there yet; keep iterating */
12033 else { /* This is not meant to happen. The best we can do is
12034 reset the iterator to the beginning. */
12035 PerlDir_seek(ret, pos0);
12042 if (name && name != smallbuf)
12047 ret = win32_dirp_dup(dp, param);
12050 /* pop it in the pointer table */
12052 ptr_table_store(PL_ptr_table, dp, ret);
12057 /* duplicate a typeglob */
12060 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12064 PERL_ARGS_ASSERT_GP_DUP;
12068 /* look for it in the table first */
12069 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12073 /* create anew and remember what it is */
12075 ptr_table_store(PL_ptr_table, gp, ret);
12078 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12079 on Newxz() to do this for us. */
12080 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12081 ret->gp_io = io_dup_inc(gp->gp_io, param);
12082 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12083 ret->gp_av = av_dup_inc(gp->gp_av, param);
12084 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12085 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12086 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12087 ret->gp_cvgen = gp->gp_cvgen;
12088 ret->gp_line = gp->gp_line;
12089 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12093 /* duplicate a chain of magic */
12096 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12098 MAGIC *mgret = NULL;
12099 MAGIC **mgprev_p = &mgret;
12101 PERL_ARGS_ASSERT_MG_DUP;
12103 for (; mg; mg = mg->mg_moremagic) {
12106 if ((param->flags & CLONEf_JOIN_IN)
12107 && mg->mg_type == PERL_MAGIC_backref)
12108 /* when joining, we let the individual SVs add themselves to
12109 * backref as needed. */
12112 Newx(nmg, 1, MAGIC);
12114 mgprev_p = &(nmg->mg_moremagic);
12116 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12117 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12118 from the original commit adding Perl_mg_dup() - revision 4538.
12119 Similarly there is the annotation "XXX random ptr?" next to the
12120 assignment to nmg->mg_ptr. */
12123 /* FIXME for plugins
12124 if (nmg->mg_type == PERL_MAGIC_qr) {
12125 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12129 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12130 ? nmg->mg_type == PERL_MAGIC_backref
12131 /* The backref AV has its reference
12132 * count deliberately bumped by 1 */
12133 ? SvREFCNT_inc(av_dup_inc((const AV *)
12134 nmg->mg_obj, param))
12135 : sv_dup_inc(nmg->mg_obj, param)
12136 : sv_dup(nmg->mg_obj, param);
12138 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12139 if (nmg->mg_len > 0) {
12140 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12141 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12142 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12144 AMT * const namtp = (AMT*)nmg->mg_ptr;
12145 sv_dup_inc_multiple((SV**)(namtp->table),
12146 (SV**)(namtp->table), NofAMmeth, param);
12149 else if (nmg->mg_len == HEf_SVKEY)
12150 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12152 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12153 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12159 #endif /* USE_ITHREADS */
12161 struct ptr_tbl_arena {
12162 struct ptr_tbl_arena *next;
12163 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12166 /* create a new pointer-mapping table */
12169 Perl_ptr_table_new(pTHX)
12172 PERL_UNUSED_CONTEXT;
12174 Newx(tbl, 1, PTR_TBL_t);
12175 tbl->tbl_max = 511;
12176 tbl->tbl_items = 0;
12177 tbl->tbl_arena = NULL;
12178 tbl->tbl_arena_next = NULL;
12179 tbl->tbl_arena_end = NULL;
12180 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12184 #define PTR_TABLE_HASH(ptr) \
12185 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12187 /* map an existing pointer using a table */
12189 STATIC PTR_TBL_ENT_t *
12190 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12192 PTR_TBL_ENT_t *tblent;
12193 const UV hash = PTR_TABLE_HASH(sv);
12195 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12197 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12198 for (; tblent; tblent = tblent->next) {
12199 if (tblent->oldval == sv)
12206 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12208 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12210 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12211 PERL_UNUSED_CONTEXT;
12213 return tblent ? tblent->newval : NULL;
12216 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12217 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12218 * the core's typical use of ptr_tables in thread cloning. */
12221 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12223 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12225 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12226 PERL_UNUSED_CONTEXT;
12229 tblent->newval = newsv;
12231 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12233 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12234 struct ptr_tbl_arena *new_arena;
12236 Newx(new_arena, 1, struct ptr_tbl_arena);
12237 new_arena->next = tbl->tbl_arena;
12238 tbl->tbl_arena = new_arena;
12239 tbl->tbl_arena_next = new_arena->array;
12240 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
12243 tblent = tbl->tbl_arena_next++;
12245 tblent->oldval = oldsv;
12246 tblent->newval = newsv;
12247 tblent->next = tbl->tbl_ary[entry];
12248 tbl->tbl_ary[entry] = tblent;
12250 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12251 ptr_table_split(tbl);
12255 /* double the hash bucket size of an existing ptr table */
12258 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
12260 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
12261 const UV oldsize = tbl->tbl_max + 1;
12262 UV newsize = oldsize * 2;
12265 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
12266 PERL_UNUSED_CONTEXT;
12268 Renew(ary, newsize, PTR_TBL_ENT_t*);
12269 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
12270 tbl->tbl_max = --newsize;
12271 tbl->tbl_ary = ary;
12272 for (i=0; i < oldsize; i++, ary++) {
12273 PTR_TBL_ENT_t **entp = ary;
12274 PTR_TBL_ENT_t *ent = *ary;
12275 PTR_TBL_ENT_t **curentp;
12278 curentp = ary + oldsize;
12280 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
12282 ent->next = *curentp;
12292 /* remove all the entries from a ptr table */
12293 /* Deprecated - will be removed post 5.14 */
12296 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
12298 PERL_UNUSED_CONTEXT;
12299 if (tbl && tbl->tbl_items) {
12300 struct ptr_tbl_arena *arena = tbl->tbl_arena;
12302 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
12305 struct ptr_tbl_arena *next = arena->next;
12311 tbl->tbl_items = 0;
12312 tbl->tbl_arena = NULL;
12313 tbl->tbl_arena_next = NULL;
12314 tbl->tbl_arena_end = NULL;
12318 /* clear and free a ptr table */
12321 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
12323 struct ptr_tbl_arena *arena;
12325 PERL_UNUSED_CONTEXT;
12331 arena = tbl->tbl_arena;
12334 struct ptr_tbl_arena *next = arena->next;
12340 Safefree(tbl->tbl_ary);
12344 #if defined(USE_ITHREADS)
12347 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
12349 PERL_ARGS_ASSERT_RVPV_DUP;
12351 assert(!isREGEXP(sstr));
12353 if (SvWEAKREF(sstr)) {
12354 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
12355 if (param->flags & CLONEf_JOIN_IN) {
12356 /* if joining, we add any back references individually rather
12357 * than copying the whole backref array */
12358 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
12362 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
12364 else if (SvPVX_const(sstr)) {
12365 /* Has something there */
12367 /* Normal PV - clone whole allocated space */
12368 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
12369 /* sstr may not be that normal, but actually copy on write.
12370 But we are a true, independent SV, so: */
12374 /* Special case - not normally malloced for some reason */
12375 if (isGV_with_GP(sstr)) {
12376 /* Don't need to do anything here. */
12378 else if ((SvIsCOW(sstr))) {
12379 /* A "shared" PV - clone it as "shared" PV */
12381 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
12385 /* Some other special case - random pointer */
12386 SvPV_set(dstr, (char *) SvPVX_const(sstr));
12391 /* Copy the NULL */
12392 SvPV_set(dstr, NULL);
12396 /* duplicate a list of SVs. source and dest may point to the same memory. */
12398 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
12399 SSize_t items, CLONE_PARAMS *const param)
12401 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
12403 while (items-- > 0) {
12404 *dest++ = sv_dup_inc(*source++, param);
12410 /* duplicate an SV of any type (including AV, HV etc) */
12413 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12418 PERL_ARGS_ASSERT_SV_DUP_COMMON;
12420 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
12421 #ifdef DEBUG_LEAKING_SCALARS_ABORT
12426 /* look for it in the table first */
12427 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
12431 if(param->flags & CLONEf_JOIN_IN) {
12432 /** We are joining here so we don't want do clone
12433 something that is bad **/
12434 if (SvTYPE(sstr) == SVt_PVHV) {
12435 const HEK * const hvname = HvNAME_HEK(sstr);
12437 /** don't clone stashes if they already exist **/
12438 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12439 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
12440 ptr_table_store(PL_ptr_table, sstr, dstr);
12444 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
12445 HV *stash = GvSTASH(sstr);
12446 const HEK * hvname;
12447 if (stash && (hvname = HvNAME_HEK(stash))) {
12448 /** don't clone GVs if they already exist **/
12450 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12451 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
12453 stash, GvNAME(sstr),
12459 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
12460 ptr_table_store(PL_ptr_table, sstr, *svp);
12467 /* create anew and remember what it is */
12470 #ifdef DEBUG_LEAKING_SCALARS
12471 dstr->sv_debug_optype = sstr->sv_debug_optype;
12472 dstr->sv_debug_line = sstr->sv_debug_line;
12473 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
12474 dstr->sv_debug_parent = (SV*)sstr;
12475 FREE_SV_DEBUG_FILE(dstr);
12476 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
12479 ptr_table_store(PL_ptr_table, sstr, dstr);
12482 SvFLAGS(dstr) = SvFLAGS(sstr);
12483 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
12484 SvREFCNT(dstr) = 0; /* must be before any other dups! */
12487 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
12488 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
12489 (void*)PL_watch_pvx, SvPVX_const(sstr));
12492 /* don't clone objects whose class has asked us not to */
12493 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
12498 switch (SvTYPE(sstr)) {
12500 SvANY(dstr) = NULL;
12503 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
12505 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12507 SvIV_set(dstr, SvIVX(sstr));
12511 SvANY(dstr) = new_XNV();
12512 SvNV_set(dstr, SvNVX(sstr));
12516 /* These are all the types that need complex bodies allocating. */
12518 const svtype sv_type = SvTYPE(sstr);
12519 const struct body_details *const sv_type_details
12520 = bodies_by_type + sv_type;
12524 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
12540 assert(sv_type_details->body_size);
12541 if (sv_type_details->arena) {
12542 new_body_inline(new_body, sv_type);
12544 = (void*)((char*)new_body - sv_type_details->offset);
12546 new_body = new_NOARENA(sv_type_details);
12550 SvANY(dstr) = new_body;
12553 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
12554 ((char*)SvANY(dstr)) + sv_type_details->offset,
12555 sv_type_details->copy, char);
12557 Copy(((char*)SvANY(sstr)),
12558 ((char*)SvANY(dstr)),
12559 sv_type_details->body_size + sv_type_details->offset, char);
12562 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
12563 && !isGV_with_GP(dstr)
12565 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
12566 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12568 /* The Copy above means that all the source (unduplicated) pointers
12569 are now in the destination. We can check the flags and the
12570 pointers in either, but it's possible that there's less cache
12571 missing by always going for the destination.
12572 FIXME - instrument and check that assumption */
12573 if (sv_type >= SVt_PVMG) {
12574 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
12575 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
12576 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
12578 } else if (SvMAGIC(dstr))
12579 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
12580 if (SvOBJECT(dstr) && SvSTASH(dstr))
12581 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
12582 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
12585 /* The cast silences a GCC warning about unhandled types. */
12586 switch ((int)sv_type) {
12597 /* FIXME for plugins */
12598 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
12599 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
12602 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
12603 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
12604 LvTARG(dstr) = dstr;
12605 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
12606 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
12608 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
12609 if (isREGEXP(sstr)) goto duprex;
12611 /* non-GP case already handled above */
12612 if(isGV_with_GP(sstr)) {
12613 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
12614 /* Don't call sv_add_backref here as it's going to be
12615 created as part of the magic cloning of the symbol
12616 table--unless this is during a join and the stash
12617 is not actually being cloned. */
12618 /* Danger Will Robinson - GvGP(dstr) isn't initialised
12619 at the point of this comment. */
12620 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
12621 if (param->flags & CLONEf_JOIN_IN)
12622 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
12623 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
12624 (void)GpREFCNT_inc(GvGP(dstr));
12628 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
12629 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
12630 /* I have no idea why fake dirp (rsfps)
12631 should be treated differently but otherwise
12632 we end up with leaks -- sky*/
12633 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
12634 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
12635 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
12637 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
12638 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
12639 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
12640 if (IoDIRP(dstr)) {
12641 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
12644 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
12646 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
12648 if (IoOFP(dstr) == IoIFP(sstr))
12649 IoOFP(dstr) = IoIFP(dstr);
12651 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
12652 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
12653 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
12654 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
12657 /* avoid cloning an empty array */
12658 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
12659 SV **dst_ary, **src_ary;
12660 SSize_t items = AvFILLp((const AV *)sstr) + 1;
12662 src_ary = AvARRAY((const AV *)sstr);
12663 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
12664 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
12665 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
12666 AvALLOC((const AV *)dstr) = dst_ary;
12667 if (AvREAL((const AV *)sstr)) {
12668 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
12672 while (items-- > 0)
12673 *dst_ary++ = sv_dup(*src_ary++, param);
12675 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
12676 while (items-- > 0) {
12677 *dst_ary++ = &PL_sv_undef;
12681 AvARRAY(MUTABLE_AV(dstr)) = NULL;
12682 AvALLOC((const AV *)dstr) = (SV**)NULL;
12683 AvMAX( (const AV *)dstr) = -1;
12684 AvFILLp((const AV *)dstr) = -1;
12688 if (HvARRAY((const HV *)sstr)) {
12690 const bool sharekeys = !!HvSHAREKEYS(sstr);
12691 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
12692 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
12694 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
12695 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
12697 HvARRAY(dstr) = (HE**)darray;
12698 while (i <= sxhv->xhv_max) {
12699 const HE * const source = HvARRAY(sstr)[i];
12700 HvARRAY(dstr)[i] = source
12701 ? he_dup(source, sharekeys, param) : 0;
12705 const struct xpvhv_aux * const saux = HvAUX(sstr);
12706 struct xpvhv_aux * const daux = HvAUX(dstr);
12707 /* This flag isn't copied. */
12710 if (saux->xhv_name_count) {
12711 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
12713 = saux->xhv_name_count < 0
12714 ? -saux->xhv_name_count
12715 : saux->xhv_name_count;
12716 HEK **shekp = sname + count;
12718 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
12719 dhekp = daux->xhv_name_u.xhvnameu_names + count;
12720 while (shekp-- > sname) {
12722 *dhekp = hek_dup(*shekp, param);
12726 daux->xhv_name_u.xhvnameu_name
12727 = hek_dup(saux->xhv_name_u.xhvnameu_name,
12730 daux->xhv_name_count = saux->xhv_name_count;
12732 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
12733 daux->xhv_aux_flags = saux->xhv_aux_flags;
12734 #ifdef PERL_HASH_RANDOMIZE_KEYS
12735 daux->xhv_rand = saux->xhv_rand;
12736 daux->xhv_last_rand = saux->xhv_last_rand;
12738 daux->xhv_riter = saux->xhv_riter;
12739 daux->xhv_eiter = saux->xhv_eiter
12740 ? he_dup(saux->xhv_eiter,
12741 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
12742 /* backref array needs refcnt=2; see sv_add_backref */
12743 daux->xhv_backreferences =
12744 (param->flags & CLONEf_JOIN_IN)
12745 /* when joining, we let the individual GVs and
12746 * CVs add themselves to backref as
12747 * needed. This avoids pulling in stuff
12748 * that isn't required, and simplifies the
12749 * case where stashes aren't cloned back
12750 * if they already exist in the parent
12753 : saux->xhv_backreferences
12754 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
12755 ? MUTABLE_AV(SvREFCNT_inc(
12756 sv_dup_inc((const SV *)
12757 saux->xhv_backreferences, param)))
12758 : MUTABLE_AV(sv_dup((const SV *)
12759 saux->xhv_backreferences, param))
12762 daux->xhv_mro_meta = saux->xhv_mro_meta
12763 ? mro_meta_dup(saux->xhv_mro_meta, param)
12766 /* Record stashes for possible cloning in Perl_clone(). */
12768 av_push(param->stashes, dstr);
12772 HvARRAY(MUTABLE_HV(dstr)) = NULL;
12775 if (!(param->flags & CLONEf_COPY_STACKS)) {
12780 /* NOTE: not refcounted */
12781 SvANY(MUTABLE_CV(dstr))->xcv_stash =
12782 hv_dup(CvSTASH(dstr), param);
12783 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
12784 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
12785 if (!CvISXSUB(dstr)) {
12787 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
12789 CvSLABBED_off(dstr);
12790 } else if (CvCONST(dstr)) {
12791 CvXSUBANY(dstr).any_ptr =
12792 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
12794 assert(!CvSLABBED(dstr));
12795 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
12797 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
12798 share_hek_hek(CvNAME_HEK((CV *)sstr));
12799 /* don't dup if copying back - CvGV isn't refcounted, so the
12800 * duped GV may never be freed. A bit of a hack! DAPM */
12802 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
12804 ? gv_dup_inc(CvGV(sstr), param)
12805 : (param->flags & CLONEf_JOIN_IN)
12807 : gv_dup(CvGV(sstr), param);
12809 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
12811 CvWEAKOUTSIDE(sstr)
12812 ? cv_dup( CvOUTSIDE(dstr), param)
12813 : cv_dup_inc(CvOUTSIDE(dstr), param);
12823 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12825 PERL_ARGS_ASSERT_SV_DUP_INC;
12826 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
12830 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12832 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
12833 PERL_ARGS_ASSERT_SV_DUP;
12835 /* Track every SV that (at least initially) had a reference count of 0.
12836 We need to do this by holding an actual reference to it in this array.
12837 If we attempt to cheat, turn AvREAL_off(), and store only pointers
12838 (akin to the stashes hash, and the perl stack), we come unstuck if
12839 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
12840 thread) is manipulated in a CLONE method, because CLONE runs before the
12841 unreferenced array is walked to find SVs still with SvREFCNT() == 0
12842 (and fix things up by giving each a reference via the temps stack).
12843 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
12844 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
12845 before the walk of unreferenced happens and a reference to that is SV
12846 added to the temps stack. At which point we have the same SV considered
12847 to be in use, and free to be re-used. Not good.
12849 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
12850 assert(param->unreferenced);
12851 av_push(param->unreferenced, SvREFCNT_inc(dstr));
12857 /* duplicate a context */
12860 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
12862 PERL_CONTEXT *ncxs;
12864 PERL_ARGS_ASSERT_CX_DUP;
12867 return (PERL_CONTEXT*)NULL;
12869 /* look for it in the table first */
12870 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
12874 /* create anew and remember what it is */
12875 Newx(ncxs, max + 1, PERL_CONTEXT);
12876 ptr_table_store(PL_ptr_table, cxs, ncxs);
12877 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
12880 PERL_CONTEXT * const ncx = &ncxs[ix];
12881 if (CxTYPE(ncx) == CXt_SUBST) {
12882 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
12885 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
12886 switch (CxTYPE(ncx)) {
12888 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
12889 ? cv_dup_inc(ncx->blk_sub.cv, param)
12890 : cv_dup(ncx->blk_sub.cv,param));
12891 if(CxHASARGS(ncx)){
12892 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
12893 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
12895 ncx->blk_sub.argarray = NULL;
12896 ncx->blk_sub.savearray = NULL;
12898 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
12899 ncx->blk_sub.oldcomppad);
12902 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
12904 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
12905 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
12907 case CXt_LOOP_LAZYSV:
12908 ncx->blk_loop.state_u.lazysv.end
12909 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
12910 /* We are taking advantage of av_dup_inc and sv_dup_inc
12911 actually being the same function, and order equivalence of
12913 We can assert the later [but only at run time :-(] */
12914 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
12915 (void *) &ncx->blk_loop.state_u.lazysv.cur);
12917 ncx->blk_loop.state_u.ary.ary
12918 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
12919 case CXt_LOOP_LAZYIV:
12920 case CXt_LOOP_PLAIN:
12921 if (CxPADLOOP(ncx)) {
12922 ncx->blk_loop.itervar_u.oldcomppad
12923 = (PAD*)ptr_table_fetch(PL_ptr_table,
12924 ncx->blk_loop.itervar_u.oldcomppad);
12926 ncx->blk_loop.itervar_u.gv
12927 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
12932 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
12933 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
12934 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
12949 /* duplicate a stack info structure */
12952 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
12956 PERL_ARGS_ASSERT_SI_DUP;
12959 return (PERL_SI*)NULL;
12961 /* look for it in the table first */
12962 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
12966 /* create anew and remember what it is */
12967 Newxz(nsi, 1, PERL_SI);
12968 ptr_table_store(PL_ptr_table, si, nsi);
12970 nsi->si_stack = av_dup_inc(si->si_stack, param);
12971 nsi->si_cxix = si->si_cxix;
12972 nsi->si_cxmax = si->si_cxmax;
12973 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
12974 nsi->si_type = si->si_type;
12975 nsi->si_prev = si_dup(si->si_prev, param);
12976 nsi->si_next = si_dup(si->si_next, param);
12977 nsi->si_markoff = si->si_markoff;
12982 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
12983 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
12984 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
12985 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
12986 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
12987 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
12988 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
12989 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
12990 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
12991 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
12992 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
12993 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
12994 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
12995 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
12996 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
12997 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13000 #define pv_dup_inc(p) SAVEPV(p)
13001 #define pv_dup(p) SAVEPV(p)
13002 #define svp_dup_inc(p,pp) any_dup(p,pp)
13004 /* map any object to the new equivent - either something in the
13005 * ptr table, or something in the interpreter structure
13009 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13013 PERL_ARGS_ASSERT_ANY_DUP;
13016 return (void*)NULL;
13018 /* look for it in the table first */
13019 ret = ptr_table_fetch(PL_ptr_table, v);
13023 /* see if it is part of the interpreter structure */
13024 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13025 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13033 /* duplicate the save stack */
13036 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13039 ANY * const ss = proto_perl->Isavestack;
13040 const I32 max = proto_perl->Isavestack_max;
13041 I32 ix = proto_perl->Isavestack_ix;
13054 void (*dptr) (void*);
13055 void (*dxptr) (pTHX_ void*);
13057 PERL_ARGS_ASSERT_SS_DUP;
13059 Newxz(nss, max, ANY);
13062 const UV uv = POPUV(ss,ix);
13063 const U8 type = (U8)uv & SAVE_MASK;
13065 TOPUV(nss,ix) = uv;
13067 case SAVEt_CLEARSV:
13068 case SAVEt_CLEARPADRANGE:
13070 case SAVEt_HELEM: /* hash element */
13071 sv = (const SV *)POPPTR(ss,ix);
13072 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13074 case SAVEt_ITEM: /* normal string */
13075 case SAVEt_GVSV: /* scalar slot in GV */
13076 case SAVEt_SV: /* scalar reference */
13077 sv = (const SV *)POPPTR(ss,ix);
13078 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13081 case SAVEt_MORTALIZESV:
13082 case SAVEt_READONLY_OFF:
13083 sv = (const SV *)POPPTR(ss,ix);
13084 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13086 case SAVEt_SHARED_PVREF: /* char* in shared space */
13087 c = (char*)POPPTR(ss,ix);
13088 TOPPTR(nss,ix) = savesharedpv(c);
13089 ptr = POPPTR(ss,ix);
13090 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13092 case SAVEt_GENERIC_SVREF: /* generic sv */
13093 case SAVEt_SVREF: /* scalar reference */
13094 sv = (const SV *)POPPTR(ss,ix);
13095 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13096 ptr = POPPTR(ss,ix);
13097 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13099 case SAVEt_GVSLOT: /* any slot in GV */
13100 sv = (const SV *)POPPTR(ss,ix);
13101 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13102 ptr = POPPTR(ss,ix);
13103 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13104 sv = (const SV *)POPPTR(ss,ix);
13105 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13107 case SAVEt_HV: /* hash reference */
13108 case SAVEt_AV: /* array reference */
13109 sv = (const SV *) POPPTR(ss,ix);
13110 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13112 case SAVEt_COMPPAD:
13114 sv = (const SV *) POPPTR(ss,ix);
13115 TOPPTR(nss,ix) = sv_dup(sv, param);
13117 case SAVEt_INT: /* int reference */
13118 ptr = POPPTR(ss,ix);
13119 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13120 intval = (int)POPINT(ss,ix);
13121 TOPINT(nss,ix) = intval;
13123 case SAVEt_LONG: /* long reference */
13124 ptr = POPPTR(ss,ix);
13125 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13126 longval = (long)POPLONG(ss,ix);
13127 TOPLONG(nss,ix) = longval;
13129 case SAVEt_I32: /* I32 reference */
13130 ptr = POPPTR(ss,ix);
13131 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13133 TOPINT(nss,ix) = i;
13135 case SAVEt_IV: /* IV reference */
13136 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13137 ptr = POPPTR(ss,ix);
13138 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13140 TOPIV(nss,ix) = iv;
13142 case SAVEt_HPTR: /* HV* reference */
13143 case SAVEt_APTR: /* AV* reference */
13144 case SAVEt_SPTR: /* SV* reference */
13145 ptr = POPPTR(ss,ix);
13146 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13147 sv = (const SV *)POPPTR(ss,ix);
13148 TOPPTR(nss,ix) = sv_dup(sv, param);
13150 case SAVEt_VPTR: /* random* reference */
13151 ptr = POPPTR(ss,ix);
13152 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13154 case SAVEt_INT_SMALL:
13155 case SAVEt_I32_SMALL:
13156 case SAVEt_I16: /* I16 reference */
13157 case SAVEt_I8: /* I8 reference */
13159 ptr = POPPTR(ss,ix);
13160 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13162 case SAVEt_GENERIC_PVREF: /* generic char* */
13163 case SAVEt_PPTR: /* char* reference */
13164 ptr = POPPTR(ss,ix);
13165 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13166 c = (char*)POPPTR(ss,ix);
13167 TOPPTR(nss,ix) = pv_dup(c);
13169 case SAVEt_GP: /* scalar reference */
13170 gp = (GP*)POPPTR(ss,ix);
13171 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13172 (void)GpREFCNT_inc(gp);
13173 gv = (const GV *)POPPTR(ss,ix);
13174 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13177 ptr = POPPTR(ss,ix);
13178 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13179 /* these are assumed to be refcounted properly */
13181 switch (((OP*)ptr)->op_type) {
13183 case OP_LEAVESUBLV:
13187 case OP_LEAVEWRITE:
13188 TOPPTR(nss,ix) = ptr;
13191 (void) OpREFCNT_inc(o);
13195 TOPPTR(nss,ix) = NULL;
13200 TOPPTR(nss,ix) = NULL;
13202 case SAVEt_FREECOPHH:
13203 ptr = POPPTR(ss,ix);
13204 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13206 case SAVEt_ADELETE:
13207 av = (const AV *)POPPTR(ss,ix);
13208 TOPPTR(nss,ix) = av_dup_inc(av, param);
13210 TOPINT(nss,ix) = i;
13213 hv = (const HV *)POPPTR(ss,ix);
13214 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13216 TOPINT(nss,ix) = i;
13219 c = (char*)POPPTR(ss,ix);
13220 TOPPTR(nss,ix) = pv_dup_inc(c);
13222 case SAVEt_STACK_POS: /* Position on Perl stack */
13224 TOPINT(nss,ix) = i;
13226 case SAVEt_DESTRUCTOR:
13227 ptr = POPPTR(ss,ix);
13228 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13229 dptr = POPDPTR(ss,ix);
13230 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13231 any_dup(FPTR2DPTR(void *, dptr),
13234 case SAVEt_DESTRUCTOR_X:
13235 ptr = POPPTR(ss,ix);
13236 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13237 dxptr = POPDXPTR(ss,ix);
13238 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13239 any_dup(FPTR2DPTR(void *, dxptr),
13242 case SAVEt_REGCONTEXT:
13244 ix -= uv >> SAVE_TIGHT_SHIFT;
13246 case SAVEt_AELEM: /* array element */
13247 sv = (const SV *)POPPTR(ss,ix);
13248 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13250 TOPINT(nss,ix) = i;
13251 av = (const AV *)POPPTR(ss,ix);
13252 TOPPTR(nss,ix) = av_dup_inc(av, param);
13255 ptr = POPPTR(ss,ix);
13256 TOPPTR(nss,ix) = ptr;
13259 ptr = POPPTR(ss,ix);
13260 ptr = cophh_copy((COPHH*)ptr);
13261 TOPPTR(nss,ix) = ptr;
13263 TOPINT(nss,ix) = i;
13264 if (i & HINT_LOCALIZE_HH) {
13265 hv = (const HV *)POPPTR(ss,ix);
13266 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13269 case SAVEt_PADSV_AND_MORTALIZE:
13270 longval = (long)POPLONG(ss,ix);
13271 TOPLONG(nss,ix) = longval;
13272 ptr = POPPTR(ss,ix);
13273 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13274 sv = (const SV *)POPPTR(ss,ix);
13275 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13277 case SAVEt_SET_SVFLAGS:
13279 TOPINT(nss,ix) = i;
13281 TOPINT(nss,ix) = i;
13282 sv = (const SV *)POPPTR(ss,ix);
13283 TOPPTR(nss,ix) = sv_dup(sv, param);
13285 case SAVEt_COMPILE_WARNINGS:
13286 ptr = POPPTR(ss,ix);
13287 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
13290 ptr = POPPTR(ss,ix);
13291 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
13295 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
13303 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
13304 * flag to the result. This is done for each stash before cloning starts,
13305 * so we know which stashes want their objects cloned */
13308 do_mark_cloneable_stash(pTHX_ SV *const sv)
13310 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
13312 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
13313 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
13314 if (cloner && GvCV(cloner)) {
13321 mXPUSHs(newSVhek(hvname));
13323 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
13330 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
13338 =for apidoc perl_clone
13340 Create and return a new interpreter by cloning the current one.
13342 perl_clone takes these flags as parameters:
13344 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
13345 without it we only clone the data and zero the stacks,
13346 with it we copy the stacks and the new perl interpreter is
13347 ready to run at the exact same point as the previous one.
13348 The pseudo-fork code uses COPY_STACKS while the
13349 threads->create doesn't.
13351 CLONEf_KEEP_PTR_TABLE -
13352 perl_clone keeps a ptr_table with the pointer of the old
13353 variable as a key and the new variable as a value,
13354 this allows it to check if something has been cloned and not
13355 clone it again but rather just use the value and increase the
13356 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
13357 the ptr_table using the function
13358 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
13359 reason to keep it around is if you want to dup some of your own
13360 variable who are outside the graph perl scans, example of this
13361 code is in threads.xs create.
13363 CLONEf_CLONE_HOST -
13364 This is a win32 thing, it is ignored on unix, it tells perls
13365 win32host code (which is c++) to clone itself, this is needed on
13366 win32 if you want to run two threads at the same time,
13367 if you just want to do some stuff in a separate perl interpreter
13368 and then throw it away and return to the original one,
13369 you don't need to do anything.
13374 /* XXX the above needs expanding by someone who actually understands it ! */
13375 EXTERN_C PerlInterpreter *
13376 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
13379 perl_clone(PerlInterpreter *proto_perl, UV flags)
13382 #ifdef PERL_IMPLICIT_SYS
13384 PERL_ARGS_ASSERT_PERL_CLONE;
13386 /* perlhost.h so we need to call into it
13387 to clone the host, CPerlHost should have a c interface, sky */
13389 if (flags & CLONEf_CLONE_HOST) {
13390 return perl_clone_host(proto_perl,flags);
13392 return perl_clone_using(proto_perl, flags,
13394 proto_perl->IMemShared,
13395 proto_perl->IMemParse,
13397 proto_perl->IStdIO,
13401 proto_perl->IProc);
13405 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
13406 struct IPerlMem* ipM, struct IPerlMem* ipMS,
13407 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
13408 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
13409 struct IPerlDir* ipD, struct IPerlSock* ipS,
13410 struct IPerlProc* ipP)
13412 /* XXX many of the string copies here can be optimized if they're
13413 * constants; they need to be allocated as common memory and just
13414 * their pointers copied. */
13417 CLONE_PARAMS clone_params;
13418 CLONE_PARAMS* const param = &clone_params;
13420 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
13422 PERL_ARGS_ASSERT_PERL_CLONE_USING;
13423 #else /* !PERL_IMPLICIT_SYS */
13425 CLONE_PARAMS clone_params;
13426 CLONE_PARAMS* param = &clone_params;
13427 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
13429 PERL_ARGS_ASSERT_PERL_CLONE;
13430 #endif /* PERL_IMPLICIT_SYS */
13432 /* for each stash, determine whether its objects should be cloned */
13433 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
13434 PERL_SET_THX(my_perl);
13437 PoisonNew(my_perl, 1, PerlInterpreter);
13440 PL_defstash = NULL; /* may be used by perl malloc() */
13443 PL_scopestack_name = 0;
13445 PL_savestack_ix = 0;
13446 PL_savestack_max = -1;
13447 PL_sig_pending = 0;
13449 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
13450 # ifdef DEBUG_LEAKING_SCALARS
13451 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
13453 #else /* !DEBUGGING */
13454 Zero(my_perl, 1, PerlInterpreter);
13455 #endif /* DEBUGGING */
13457 #ifdef PERL_IMPLICIT_SYS
13458 /* host pointers */
13460 PL_MemShared = ipMS;
13461 PL_MemParse = ipMP;
13468 #endif /* PERL_IMPLICIT_SYS */
13471 param->flags = flags;
13472 /* Nothing in the core code uses this, but we make it available to
13473 extensions (using mg_dup). */
13474 param->proto_perl = proto_perl;
13475 /* Likely nothing will use this, but it is initialised to be consistent
13476 with Perl_clone_params_new(). */
13477 param->new_perl = my_perl;
13478 param->unreferenced = NULL;
13481 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
13483 PL_body_arenas = NULL;
13484 Zero(&PL_body_roots, 1, PL_body_roots);
13488 PL_sv_arenaroot = NULL;
13490 PL_debug = proto_perl->Idebug;
13492 /* dbargs array probably holds garbage */
13495 PL_compiling = proto_perl->Icompiling;
13497 /* pseudo environmental stuff */
13498 PL_origargc = proto_perl->Iorigargc;
13499 PL_origargv = proto_perl->Iorigargv;
13501 #ifndef NO_TAINT_SUPPORT
13502 /* Set tainting stuff before PerlIO_debug can possibly get called */
13503 PL_tainting = proto_perl->Itainting;
13504 PL_taint_warn = proto_perl->Itaint_warn;
13506 PL_tainting = FALSE;
13507 PL_taint_warn = FALSE;
13510 PL_minus_c = proto_perl->Iminus_c;
13512 PL_localpatches = proto_perl->Ilocalpatches;
13513 PL_splitstr = proto_perl->Isplitstr;
13514 PL_minus_n = proto_perl->Iminus_n;
13515 PL_minus_p = proto_perl->Iminus_p;
13516 PL_minus_l = proto_perl->Iminus_l;
13517 PL_minus_a = proto_perl->Iminus_a;
13518 PL_minus_E = proto_perl->Iminus_E;
13519 PL_minus_F = proto_perl->Iminus_F;
13520 PL_doswitches = proto_perl->Idoswitches;
13521 PL_dowarn = proto_perl->Idowarn;
13522 #ifdef PERL_SAWAMPERSAND
13523 PL_sawampersand = proto_perl->Isawampersand;
13525 PL_unsafe = proto_perl->Iunsafe;
13526 PL_perldb = proto_perl->Iperldb;
13527 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
13528 PL_exit_flags = proto_perl->Iexit_flags;
13530 /* XXX time(&PL_basetime) when asked for? */
13531 PL_basetime = proto_perl->Ibasetime;
13533 PL_maxsysfd = proto_perl->Imaxsysfd;
13534 PL_statusvalue = proto_perl->Istatusvalue;
13536 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
13538 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
13541 /* RE engine related */
13542 PL_regmatch_slab = NULL;
13543 PL_reg_curpm = NULL;
13545 PL_sub_generation = proto_perl->Isub_generation;
13547 /* funky return mechanisms */
13548 PL_forkprocess = proto_perl->Iforkprocess;
13550 /* internal state */
13551 PL_maxo = proto_perl->Imaxo;
13553 PL_main_start = proto_perl->Imain_start;
13554 PL_eval_root = proto_perl->Ieval_root;
13555 PL_eval_start = proto_perl->Ieval_start;
13557 PL_filemode = proto_perl->Ifilemode;
13558 PL_lastfd = proto_perl->Ilastfd;
13559 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
13562 PL_gensym = proto_perl->Igensym;
13564 PL_laststatval = proto_perl->Ilaststatval;
13565 PL_laststype = proto_perl->Ilaststype;
13568 PL_profiledata = NULL;
13570 PL_generation = proto_perl->Igeneration;
13572 PL_in_clean_objs = proto_perl->Iin_clean_objs;
13573 PL_in_clean_all = proto_perl->Iin_clean_all;
13575 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
13576 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
13577 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
13578 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
13579 PL_nomemok = proto_perl->Inomemok;
13580 PL_an = proto_perl->Ian;
13581 PL_evalseq = proto_perl->Ievalseq;
13582 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
13583 PL_origalen = proto_perl->Iorigalen;
13585 PL_sighandlerp = proto_perl->Isighandlerp;
13587 PL_runops = proto_perl->Irunops;
13589 PL_subline = proto_perl->Isubline;
13592 PL_cryptseen = proto_perl->Icryptseen;
13595 #ifdef USE_LOCALE_COLLATE
13596 PL_collation_ix = proto_perl->Icollation_ix;
13597 PL_collation_standard = proto_perl->Icollation_standard;
13598 PL_collxfrm_base = proto_perl->Icollxfrm_base;
13599 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
13600 #endif /* USE_LOCALE_COLLATE */
13602 #ifdef USE_LOCALE_NUMERIC
13603 PL_numeric_standard = proto_perl->Inumeric_standard;
13604 PL_numeric_local = proto_perl->Inumeric_local;
13605 #endif /* !USE_LOCALE_NUMERIC */
13607 /* Did the locale setup indicate UTF-8? */
13608 PL_utf8locale = proto_perl->Iutf8locale;
13609 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
13610 /* Unicode features (see perlrun/-C) */
13611 PL_unicode = proto_perl->Iunicode;
13613 /* Pre-5.8 signals control */
13614 PL_signals = proto_perl->Isignals;
13616 /* times() ticks per second */
13617 PL_clocktick = proto_perl->Iclocktick;
13619 /* Recursion stopper for PerlIO_find_layer */
13620 PL_in_load_module = proto_perl->Iin_load_module;
13622 /* sort() routine */
13623 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
13625 /* Not really needed/useful since the reenrant_retint is "volatile",
13626 * but do it for consistency's sake. */
13627 PL_reentrant_retint = proto_perl->Ireentrant_retint;
13629 /* Hooks to shared SVs and locks. */
13630 PL_sharehook = proto_perl->Isharehook;
13631 PL_lockhook = proto_perl->Ilockhook;
13632 PL_unlockhook = proto_perl->Iunlockhook;
13633 PL_threadhook = proto_perl->Ithreadhook;
13634 PL_destroyhook = proto_perl->Idestroyhook;
13635 PL_signalhook = proto_perl->Isignalhook;
13637 PL_globhook = proto_perl->Iglobhook;
13640 PL_last_swash_hv = NULL; /* reinits on demand */
13641 PL_last_swash_klen = 0;
13642 PL_last_swash_key[0]= '\0';
13643 PL_last_swash_tmps = (U8*)NULL;
13644 PL_last_swash_slen = 0;
13646 PL_srand_called = proto_perl->Isrand_called;
13647 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
13649 if (flags & CLONEf_COPY_STACKS) {
13650 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
13651 PL_tmps_ix = proto_perl->Itmps_ix;
13652 PL_tmps_max = proto_perl->Itmps_max;
13653 PL_tmps_floor = proto_perl->Itmps_floor;
13655 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
13656 * NOTE: unlike the others! */
13657 PL_scopestack_ix = proto_perl->Iscopestack_ix;
13658 PL_scopestack_max = proto_perl->Iscopestack_max;
13660 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
13661 * NOTE: unlike the others! */
13662 PL_savestack_ix = proto_perl->Isavestack_ix;
13663 PL_savestack_max = proto_perl->Isavestack_max;
13666 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
13667 PL_top_env = &PL_start_env;
13669 PL_op = proto_perl->Iop;
13672 PL_Xpv = (XPV*)NULL;
13673 my_perl->Ina = proto_perl->Ina;
13675 PL_statbuf = proto_perl->Istatbuf;
13676 PL_statcache = proto_perl->Istatcache;
13678 #ifndef NO_TAINT_SUPPORT
13679 PL_tainted = proto_perl->Itainted;
13681 PL_tainted = FALSE;
13683 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
13685 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
13687 PL_restartjmpenv = proto_perl->Irestartjmpenv;
13688 PL_restartop = proto_perl->Irestartop;
13689 PL_in_eval = proto_perl->Iin_eval;
13690 PL_delaymagic = proto_perl->Idelaymagic;
13691 PL_phase = proto_perl->Iphase;
13692 PL_localizing = proto_perl->Ilocalizing;
13694 PL_hv_fetch_ent_mh = NULL;
13695 PL_modcount = proto_perl->Imodcount;
13696 PL_lastgotoprobe = NULL;
13697 PL_dumpindent = proto_perl->Idumpindent;
13699 PL_efloatbuf = NULL; /* reinits on demand */
13700 PL_efloatsize = 0; /* reinits on demand */
13704 PL_colorset = 0; /* reinits PL_colors[] */
13705 /*PL_colors[6] = {0,0,0,0,0,0};*/
13707 /* Pluggable optimizer */
13708 PL_peepp = proto_perl->Ipeepp;
13709 PL_rpeepp = proto_perl->Irpeepp;
13710 /* op_free() hook */
13711 PL_opfreehook = proto_perl->Iopfreehook;
13713 #ifdef USE_REENTRANT_API
13714 /* XXX: things like -Dm will segfault here in perlio, but doing
13715 * PERL_SET_CONTEXT(proto_perl);
13716 * breaks too many other things
13718 Perl_reentrant_init(aTHX);
13721 /* create SV map for pointer relocation */
13722 PL_ptr_table = ptr_table_new();
13724 /* initialize these special pointers as early as possible */
13726 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
13727 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
13728 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
13730 /* create (a non-shared!) shared string table */
13731 PL_strtab = newHV();
13732 HvSHAREKEYS_off(PL_strtab);
13733 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
13734 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
13736 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
13738 /* This PV will be free'd special way so must set it same way op.c does */
13739 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
13740 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
13742 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
13743 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
13744 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
13745 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
13747 param->stashes = newAV(); /* Setup array of objects to call clone on */
13748 /* This makes no difference to the implementation, as it always pushes
13749 and shifts pointers to other SVs without changing their reference
13750 count, with the array becoming empty before it is freed. However, it
13751 makes it conceptually clear what is going on, and will avoid some
13752 work inside av.c, filling slots between AvFILL() and AvMAX() with
13753 &PL_sv_undef, and SvREFCNT_dec()ing those. */
13754 AvREAL_off(param->stashes);
13756 if (!(flags & CLONEf_COPY_STACKS)) {
13757 param->unreferenced = newAV();
13760 #ifdef PERLIO_LAYERS
13761 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
13762 PerlIO_clone(aTHX_ proto_perl, param);
13765 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
13766 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
13767 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
13768 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
13769 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
13770 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
13773 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
13774 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
13775 PL_inplace = SAVEPV(proto_perl->Iinplace);
13776 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
13778 /* magical thingies */
13780 PL_encoding = sv_dup(proto_perl->Iencoding, param);
13782 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
13783 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
13784 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
13787 /* Clone the regex array */
13788 /* ORANGE FIXME for plugins, probably in the SV dup code.
13789 newSViv(PTR2IV(CALLREGDUPE(
13790 INT2PTR(REGEXP *, SvIVX(regex)), param))))
13792 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
13793 PL_regex_pad = AvARRAY(PL_regex_padav);
13795 PL_stashpadmax = proto_perl->Istashpadmax;
13796 PL_stashpadix = proto_perl->Istashpadix ;
13797 Newx(PL_stashpad, PL_stashpadmax, HV *);
13800 for (; o < PL_stashpadmax; ++o)
13801 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
13804 /* shortcuts to various I/O objects */
13805 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
13806 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
13807 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
13808 PL_defgv = gv_dup(proto_perl->Idefgv, param);
13809 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
13810 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
13811 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
13813 /* shortcuts to regexp stuff */
13814 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
13816 /* shortcuts to misc objects */
13817 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
13819 /* shortcuts to debugging objects */
13820 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
13821 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
13822 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
13823 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
13824 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
13825 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
13827 /* symbol tables */
13828 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
13829 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
13830 PL_debstash = hv_dup(proto_perl->Idebstash, param);
13831 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
13832 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
13834 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
13835 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
13836 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
13837 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
13838 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
13839 PL_endav = av_dup_inc(proto_perl->Iendav, param);
13840 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
13841 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
13843 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
13845 /* subprocess state */
13846 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
13848 if (proto_perl->Iop_mask)
13849 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
13852 /* PL_asserting = proto_perl->Iasserting; */
13854 /* current interpreter roots */
13855 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
13857 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
13860 /* runtime control stuff */
13861 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
13863 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
13865 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
13867 /* interpreter atexit processing */
13868 PL_exitlistlen = proto_perl->Iexitlistlen;
13869 if (PL_exitlistlen) {
13870 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13871 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13874 PL_exitlist = (PerlExitListEntry*)NULL;
13876 PL_my_cxt_size = proto_perl->Imy_cxt_size;
13877 if (PL_my_cxt_size) {
13878 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
13879 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
13880 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13881 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
13882 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
13886 PL_my_cxt_list = (void**)NULL;
13887 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13888 PL_my_cxt_keys = (const char**)NULL;
13891 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
13892 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
13893 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
13894 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
13896 PL_compcv = cv_dup(proto_perl->Icompcv, param);
13898 PAD_CLONE_VARS(proto_perl, param);
13900 #ifdef HAVE_INTERP_INTERN
13901 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
13904 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
13906 #ifdef PERL_USES_PL_PIDSTATUS
13907 PL_pidstatus = newHV(); /* XXX flag for cloning? */
13909 PL_osname = SAVEPV(proto_perl->Iosname);
13910 PL_parser = parser_dup(proto_perl->Iparser, param);
13912 /* XXX this only works if the saved cop has already been cloned */
13913 if (proto_perl->Iparser) {
13914 PL_parser->saved_curcop = (COP*)any_dup(
13915 proto_perl->Iparser->saved_curcop,
13919 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
13921 #ifdef USE_LOCALE_COLLATE
13922 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
13923 #endif /* USE_LOCALE_COLLATE */
13925 #ifdef USE_LOCALE_NUMERIC
13926 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
13927 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
13928 #endif /* !USE_LOCALE_NUMERIC */
13930 /* Unicode inversion lists */
13931 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
13932 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
13933 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
13935 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
13936 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
13938 /* utf8 character class swashes */
13939 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
13940 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
13942 for (i = 0; i < POSIX_CC_COUNT; i++) {
13943 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
13945 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
13946 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
13947 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
13948 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
13949 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
13950 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
13951 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
13952 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
13953 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
13954 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
13955 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
13956 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
13957 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
13958 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
13959 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
13960 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
13962 if (proto_perl->Ipsig_pend) {
13963 Newxz(PL_psig_pend, SIG_SIZE, int);
13966 PL_psig_pend = (int*)NULL;
13969 if (proto_perl->Ipsig_name) {
13970 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
13971 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
13973 PL_psig_ptr = PL_psig_name + SIG_SIZE;
13976 PL_psig_ptr = (SV**)NULL;
13977 PL_psig_name = (SV**)NULL;
13980 if (flags & CLONEf_COPY_STACKS) {
13981 Newx(PL_tmps_stack, PL_tmps_max, SV*);
13982 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
13983 PL_tmps_ix+1, param);
13985 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
13986 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
13987 Newxz(PL_markstack, i, I32);
13988 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
13989 - proto_perl->Imarkstack);
13990 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
13991 - proto_perl->Imarkstack);
13992 Copy(proto_perl->Imarkstack, PL_markstack,
13993 PL_markstack_ptr - PL_markstack + 1, I32);
13995 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
13996 * NOTE: unlike the others! */
13997 Newxz(PL_scopestack, PL_scopestack_max, I32);
13998 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14001 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14002 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14004 /* reset stack AV to correct length before its duped via
14005 * PL_curstackinfo */
14006 AvFILLp(proto_perl->Icurstack) =
14007 proto_perl->Istack_sp - proto_perl->Istack_base;
14009 /* NOTE: si_dup() looks at PL_markstack */
14010 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14012 /* PL_curstack = PL_curstackinfo->si_stack; */
14013 PL_curstack = av_dup(proto_perl->Icurstack, param);
14014 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14016 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14017 PL_stack_base = AvARRAY(PL_curstack);
14018 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14019 - proto_perl->Istack_base);
14020 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14022 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14023 PL_savestack = ss_dup(proto_perl, param);
14027 ENTER; /* perl_destruct() wants to LEAVE; */
14030 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14031 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14033 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14034 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14035 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14036 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14037 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14038 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14040 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14042 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14043 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14044 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14046 PL_stashcache = newHV();
14048 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14049 proto_perl->Iwatchaddr);
14050 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14051 if (PL_debug && PL_watchaddr) {
14052 PerlIO_printf(Perl_debug_log,
14053 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14054 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14055 PTR2UV(PL_watchok));
14058 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14059 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14060 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14062 /* Call the ->CLONE method, if it exists, for each of the stashes
14063 identified by sv_dup() above.
14065 while(av_tindex(param->stashes) != -1) {
14066 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14067 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14068 if (cloner && GvCV(cloner)) {
14073 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14075 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14081 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14082 ptr_table_free(PL_ptr_table);
14083 PL_ptr_table = NULL;
14086 if (!(flags & CLONEf_COPY_STACKS)) {
14087 unreferenced_to_tmp_stack(param->unreferenced);
14090 SvREFCNT_dec(param->stashes);
14092 /* orphaned? eg threads->new inside BEGIN or use */
14093 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14094 SvREFCNT_inc_simple_void(PL_compcv);
14095 SAVEFREESV(PL_compcv);
14102 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14104 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14106 if (AvFILLp(unreferenced) > -1) {
14107 SV **svp = AvARRAY(unreferenced);
14108 SV **const last = svp + AvFILLp(unreferenced);
14112 if (SvREFCNT(*svp) == 1)
14114 } while (++svp <= last);
14116 EXTEND_MORTAL(count);
14117 svp = AvARRAY(unreferenced);
14120 if (SvREFCNT(*svp) == 1) {
14121 /* Our reference is the only one to this SV. This means that
14122 in this thread, the scalar effectively has a 0 reference.
14123 That doesn't work (cleanup never happens), so donate our
14124 reference to it onto the save stack. */
14125 PL_tmps_stack[++PL_tmps_ix] = *svp;
14127 /* As an optimisation, because we are already walking the
14128 entire array, instead of above doing either
14129 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14130 release our reference to the scalar, so that at the end of
14131 the array owns zero references to the scalars it happens to
14132 point to. We are effectively converting the array from
14133 AvREAL() on to AvREAL() off. This saves the av_clear()
14134 (triggered by the SvREFCNT_dec(unreferenced) below) from
14135 walking the array a second time. */
14136 SvREFCNT_dec(*svp);
14139 } while (++svp <= last);
14140 AvREAL_off(unreferenced);
14142 SvREFCNT_dec_NN(unreferenced);
14146 Perl_clone_params_del(CLONE_PARAMS *param)
14148 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14150 PerlInterpreter *const to = param->new_perl;
14152 PerlInterpreter *const was = PERL_GET_THX;
14154 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14160 SvREFCNT_dec(param->stashes);
14161 if (param->unreferenced)
14162 unreferenced_to_tmp_stack(param->unreferenced);
14172 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14175 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14176 does a dTHX; to get the context from thread local storage.
14177 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14178 a version that passes in my_perl. */
14179 PerlInterpreter *const was = PERL_GET_THX;
14180 CLONE_PARAMS *param;
14182 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14188 /* Given that we've set the context, we can do this unshared. */
14189 Newx(param, 1, CLONE_PARAMS);
14192 param->proto_perl = from;
14193 param->new_perl = to;
14194 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14195 AvREAL_off(param->stashes);
14196 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14204 #endif /* USE_ITHREADS */
14207 Perl_init_constants(pTHX)
14209 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14210 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVt_NULL;
14211 SvANY(&PL_sv_undef) = NULL;
14213 SvANY(&PL_sv_no) = new_XPVNV();
14214 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14215 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY
14216 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14219 SvANY(&PL_sv_yes) = new_XPVNV();
14220 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14221 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY
14222 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14225 SvPV_set(&PL_sv_no, (char*)PL_No);
14226 SvCUR_set(&PL_sv_no, 0);
14227 SvLEN_set(&PL_sv_no, 0);
14228 SvIV_set(&PL_sv_no, 0);
14229 SvNV_set(&PL_sv_no, 0);
14231 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14232 SvCUR_set(&PL_sv_yes, 1);
14233 SvLEN_set(&PL_sv_yes, 0);
14234 SvIV_set(&PL_sv_yes, 1);
14235 SvNV_set(&PL_sv_yes, 1);
14239 =head1 Unicode Support
14241 =for apidoc sv_recode_to_utf8
14243 The encoding is assumed to be an Encode object, on entry the PV
14244 of the sv is assumed to be octets in that encoding, and the sv
14245 will be converted into Unicode (and UTF-8).
14247 If the sv already is UTF-8 (or if it is not POK), or if the encoding
14248 is not a reference, nothing is done to the sv. If the encoding is not
14249 an C<Encode::XS> Encoding object, bad things will happen.
14250 (See F<lib/encoding.pm> and L<Encode>.)
14252 The PV of the sv is returned.
14257 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
14259 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
14261 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
14270 if (SvPADTMP(nsv)) {
14271 nsv = sv_newmortal();
14272 SvSetSV_nosteal(nsv, sv);
14281 Passing sv_yes is wrong - it needs to be or'ed set of constants
14282 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
14283 remove converted chars from source.
14285 Both will default the value - let them.
14287 XPUSHs(&PL_sv_yes);
14290 call_method("decode", G_SCALAR);
14294 s = SvPV_const(uni, len);
14295 if (s != SvPVX_const(sv)) {
14296 SvGROW(sv, len + 1);
14297 Move(s, SvPVX(sv), len + 1, char);
14298 SvCUR_set(sv, len);
14303 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
14304 /* clear pos and any utf8 cache */
14305 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
14308 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
14309 magic_setutf8(sv,mg); /* clear UTF8 cache */
14314 return SvPOKp(sv) ? SvPVX(sv) : NULL;
14318 =for apidoc sv_cat_decode
14320 The encoding is assumed to be an Encode object, the PV of the ssv is
14321 assumed to be octets in that encoding and decoding the input starts
14322 from the position which (PV + *offset) pointed to. The dsv will be
14323 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
14324 when the string tstr appears in decoding output or the input ends on
14325 the PV of the ssv. The value which the offset points will be modified
14326 to the last input position on the ssv.
14328 Returns TRUE if the terminator was found, else returns FALSE.
14333 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
14334 SV *ssv, int *offset, char *tstr, int tlen)
14338 PERL_ARGS_ASSERT_SV_CAT_DECODE;
14340 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
14351 offsv = newSViv(*offset);
14353 mPUSHp(tstr, tlen);
14355 call_method("cat_decode", G_SCALAR);
14357 ret = SvTRUE(TOPs);
14358 *offset = SvIV(offsv);
14364 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
14369 /* ---------------------------------------------------------------------
14371 * support functions for report_uninit()
14374 /* the maxiumum size of array or hash where we will scan looking
14375 * for the undefined element that triggered the warning */
14377 #define FUV_MAX_SEARCH_SIZE 1000
14379 /* Look for an entry in the hash whose value has the same SV as val;
14380 * If so, return a mortal copy of the key. */
14383 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
14389 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
14391 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
14392 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
14395 array = HvARRAY(hv);
14397 for (i=HvMAX(hv); i>=0; i--) {
14399 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
14400 if (HeVAL(entry) != val)
14402 if ( HeVAL(entry) == &PL_sv_undef ||
14403 HeVAL(entry) == &PL_sv_placeholder)
14407 if (HeKLEN(entry) == HEf_SVKEY)
14408 return sv_mortalcopy(HeKEY_sv(entry));
14409 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
14415 /* Look for an entry in the array whose value has the same SV as val;
14416 * If so, return the index, otherwise return -1. */
14419 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
14421 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
14423 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
14424 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
14427 if (val != &PL_sv_undef) {
14428 SV ** const svp = AvARRAY(av);
14431 for (i=AvFILLp(av); i>=0; i--)
14438 /* varname(): return the name of a variable, optionally with a subscript.
14439 * If gv is non-zero, use the name of that global, along with gvtype (one
14440 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
14441 * targ. Depending on the value of the subscript_type flag, return:
14444 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
14445 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
14446 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
14447 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
14450 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
14451 const SV *const keyname, I32 aindex, int subscript_type)
14454 SV * const name = sv_newmortal();
14455 if (gv && isGV(gv)) {
14457 buffer[0] = gvtype;
14460 /* as gv_fullname4(), but add literal '^' for $^FOO names */
14462 gv_fullname4(name, gv, buffer, 0);
14464 if ((unsigned int)SvPVX(name)[1] <= 26) {
14466 buffer[1] = SvPVX(name)[1] + 'A' - 1;
14468 /* Swap the 1 unprintable control character for the 2 byte pretty
14469 version - ie substr($name, 1, 1) = $buffer; */
14470 sv_insert(name, 1, 1, buffer, 2);
14474 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
14478 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
14480 if (!cv || !CvPADLIST(cv))
14482 av = *PadlistARRAY(CvPADLIST(cv));
14483 sv = *av_fetch(av, targ, FALSE);
14484 sv_setsv_flags(name, sv, 0);
14487 if (subscript_type == FUV_SUBSCRIPT_HASH) {
14488 SV * const sv = newSV(0);
14489 *SvPVX(name) = '$';
14490 Perl_sv_catpvf(aTHX_ name, "{%s}",
14491 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
14492 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
14493 SvREFCNT_dec_NN(sv);
14495 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
14496 *SvPVX(name) = '$';
14497 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
14499 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
14500 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
14501 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
14509 =for apidoc find_uninit_var
14511 Find the name of the undefined variable (if any) that caused the operator
14512 to issue a "Use of uninitialized value" warning.
14513 If match is true, only return a name if its value matches uninit_sv.
14514 So roughly speaking, if a unary operator (such as OP_COS) generates a
14515 warning, then following the direct child of the op may yield an
14516 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
14517 other hand, with OP_ADD there are two branches to follow, so we only print
14518 the variable name if we get an exact match.
14520 The name is returned as a mortal SV.
14522 Assumes that PL_op is the op that originally triggered the error, and that
14523 PL_comppad/PL_curpad points to the currently executing pad.
14529 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
14535 const OP *o, *o2, *kid;
14537 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
14538 uninit_sv == &PL_sv_placeholder)))
14541 switch (obase->op_type) {
14548 const bool pad = ( obase->op_type == OP_PADAV
14549 || obase->op_type == OP_PADHV
14550 || obase->op_type == OP_PADRANGE
14553 const bool hash = ( obase->op_type == OP_PADHV
14554 || obase->op_type == OP_RV2HV
14555 || (obase->op_type == OP_PADRANGE
14556 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
14560 int subscript_type = FUV_SUBSCRIPT_WITHIN;
14562 if (pad) { /* @lex, %lex */
14563 sv = PAD_SVl(obase->op_targ);
14567 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14568 /* @global, %global */
14569 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14572 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
14574 else if (obase == PL_op) /* @{expr}, %{expr} */
14575 return find_uninit_var(cUNOPx(obase)->op_first,
14577 else /* @{expr}, %{expr} as a sub-expression */
14581 /* attempt to find a match within the aggregate */
14583 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14585 subscript_type = FUV_SUBSCRIPT_HASH;
14588 index = find_array_subscript((const AV *)sv, uninit_sv);
14590 subscript_type = FUV_SUBSCRIPT_ARRAY;
14593 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
14596 return varname(gv, hash ? '%' : '@', obase->op_targ,
14597 keysv, index, subscript_type);
14601 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14603 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14604 if (!gv || !GvSTASH(gv))
14606 if (match && (GvSV(gv) != uninit_sv))
14608 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14611 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
14614 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
14616 return varname(NULL, '$', obase->op_targ,
14617 NULL, 0, FUV_SUBSCRIPT_NONE);
14620 gv = cGVOPx_gv(obase);
14621 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
14623 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14625 case OP_AELEMFAST_LEX:
14628 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
14629 if (!av || SvRMAGICAL(av))
14631 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14632 if (!svp || *svp != uninit_sv)
14635 return varname(NULL, '$', obase->op_targ,
14636 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14639 gv = cGVOPx_gv(obase);
14644 AV *const av = GvAV(gv);
14645 if (!av || SvRMAGICAL(av))
14647 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14648 if (!svp || *svp != uninit_sv)
14651 return varname(gv, '$', 0,
14652 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14654 NOT_REACHED; /* NOTREACHED */
14657 o = cUNOPx(obase)->op_first;
14658 if (!o || o->op_type != OP_NULL ||
14659 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
14661 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
14666 bool negate = FALSE;
14668 if (PL_op == obase)
14669 /* $a[uninit_expr] or $h{uninit_expr} */
14670 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
14673 o = cBINOPx(obase)->op_first;
14674 kid = cBINOPx(obase)->op_last;
14676 /* get the av or hv, and optionally the gv */
14678 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
14679 sv = PAD_SV(o->op_targ);
14681 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
14682 && cUNOPo->op_first->op_type == OP_GV)
14684 gv = cGVOPx_gv(cUNOPo->op_first);
14688 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
14693 if (kid && kid->op_type == OP_NEGATE) {
14695 kid = cUNOPx(kid)->op_first;
14698 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
14699 /* index is constant */
14702 kidsv = sv_2mortal(newSVpvs("-"));
14703 sv_catsv(kidsv, cSVOPx_sv(kid));
14706 kidsv = cSVOPx_sv(kid);
14710 if (obase->op_type == OP_HELEM) {
14711 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
14712 if (!he || HeVAL(he) != uninit_sv)
14716 SV * const opsv = cSVOPx_sv(kid);
14717 const IV opsviv = SvIV(opsv);
14718 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
14719 negate ? - opsviv : opsviv,
14721 if (!svp || *svp != uninit_sv)
14725 if (obase->op_type == OP_HELEM)
14726 return varname(gv, '%', o->op_targ,
14727 kidsv, 0, FUV_SUBSCRIPT_HASH);
14729 return varname(gv, '@', o->op_targ, NULL,
14730 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
14731 FUV_SUBSCRIPT_ARRAY);
14734 /* index is an expression;
14735 * attempt to find a match within the aggregate */
14736 if (obase->op_type == OP_HELEM) {
14737 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14739 return varname(gv, '%', o->op_targ,
14740 keysv, 0, FUV_SUBSCRIPT_HASH);
14744 = find_array_subscript((const AV *)sv, uninit_sv);
14746 return varname(gv, '@', o->op_targ,
14747 NULL, index, FUV_SUBSCRIPT_ARRAY);
14752 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
14754 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
14756 NOT_REACHED; /* NOTREACHED */
14760 /* only examine RHS */
14761 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
14764 o = cUNOPx(obase)->op_first;
14765 if ( o->op_type == OP_PUSHMARK
14766 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
14770 if (!OP_HAS_SIBLING(o)) {
14771 /* one-arg version of open is highly magical */
14773 if (o->op_type == OP_GV) { /* open FOO; */
14775 if (match && GvSV(gv) != uninit_sv)
14777 return varname(gv, '$', 0,
14778 NULL, 0, FUV_SUBSCRIPT_NONE);
14780 /* other possibilities not handled are:
14781 * open $x; or open my $x; should return '${*$x}'
14782 * open expr; should return '$'.expr ideally
14788 /* ops where $_ may be an implicit arg */
14793 if ( !(obase->op_flags & OPf_STACKED)) {
14794 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
14795 ? PAD_SVl(obase->op_targ)
14798 sv = sv_newmortal();
14799 sv_setpvs(sv, "$_");
14808 match = 1; /* print etc can return undef on defined args */
14809 /* skip filehandle as it can't produce 'undef' warning */
14810 o = cUNOPx(obase)->op_first;
14811 if ((obase->op_flags & OPf_STACKED)
14813 ( o->op_type == OP_PUSHMARK
14814 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
14815 o = OP_SIBLING(OP_SIBLING(o));
14819 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
14820 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
14822 /* the following ops are capable of returning PL_sv_undef even for
14823 * defined arg(s) */
14842 case OP_GETPEERNAME:
14890 case OP_SMARTMATCH:
14899 /* XXX tmp hack: these two may call an XS sub, and currently
14900 XS subs don't have a SUB entry on the context stack, so CV and
14901 pad determination goes wrong, and BAD things happen. So, just
14902 don't try to determine the value under those circumstances.
14903 Need a better fix at dome point. DAPM 11/2007 */
14909 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
14910 if (gv && GvSV(gv) == uninit_sv)
14911 return newSVpvs_flags("$.", SVs_TEMP);
14916 /* def-ness of rval pos() is independent of the def-ness of its arg */
14917 if ( !(obase->op_flags & OPf_MOD))
14922 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
14923 return newSVpvs_flags("${$/}", SVs_TEMP);
14928 if (!(obase->op_flags & OPf_KIDS))
14930 o = cUNOPx(obase)->op_first;
14936 /* This loop checks all the kid ops, skipping any that cannot pos-
14937 * sibly be responsible for the uninitialized value; i.e., defined
14938 * constants and ops that return nothing. If there is only one op
14939 * left that is not skipped, then we *know* it is responsible for
14940 * the uninitialized value. If there is more than one op left, we
14941 * have to look for an exact match in the while() loop below.
14942 * Note that we skip padrange, because the individual pad ops that
14943 * it replaced are still in the tree, so we work on them instead.
14946 for (kid=o; kid; kid = OP_SIBLING(kid)) {
14947 const OPCODE type = kid->op_type;
14948 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
14949 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
14950 || (type == OP_PUSHMARK)
14951 || (type == OP_PADRANGE)
14955 if (o2) { /* more than one found */
14962 return find_uninit_var(o2, uninit_sv, match);
14964 /* scan all args */
14966 sv = find_uninit_var(o, uninit_sv, 1);
14978 =for apidoc report_uninit
14980 Print appropriate "Use of uninitialized variable" warning.
14986 Perl_report_uninit(pTHX_ const SV *uninit_sv)
14989 SV* varname = NULL;
14990 if (uninit_sv && PL_curpad) {
14991 varname = find_uninit_var(PL_op, uninit_sv,0);
14993 sv_insert(varname, 0, 0, " ", 1);
14995 /* PL_warn_uninit_sv is constant */
14996 GCC_DIAG_IGNORE(-Wformat-nonliteral);
14997 /* diag_listed_as: Use of uninitialized value%s */
14998 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
14999 SVfARG(varname ? varname : &PL_sv_no),
15000 " in ", OP_DESC(PL_op));
15004 /* PL_warn_uninit is constant */
15005 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15006 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15014 * c-indentation-style: bsd
15015 * c-basic-offset: 4
15016 * indent-tabs-mode: nil
15019 * ex: set ts=8 sts=4 sw=4 et: