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 */
315 char *chunk; /* must use New here to match call to */
316 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
317 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
322 /* new_SV(): return a new, empty SV head */
324 #ifdef DEBUG_LEAKING_SCALARS
325 /* provide a real function for a debugger to play with */
327 S_new_SV(pTHX_ const char *file, int line, const char *func)
334 sv = S_more_sv(aTHX);
338 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
339 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
345 sv->sv_debug_inpad = 0;
346 sv->sv_debug_parent = NULL;
347 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
349 sv->sv_debug_serial = PL_sv_serial++;
351 MEM_LOG_NEW_SV(sv, file, line, func);
352 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
353 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
357 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
365 (p) = S_more_sv(aTHX); \
369 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
374 /* del_SV(): return an empty SV head to the free list */
387 S_del_sv(pTHX_ SV *p)
391 PERL_ARGS_ASSERT_DEL_SV;
396 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
397 const SV * const sv = sva + 1;
398 const SV * const svend = &sva[SvREFCNT(sva)];
399 if (p >= sv && p < svend) {
405 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
406 "Attempt to free non-arena SV: 0x%"UVxf
407 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
414 #else /* ! DEBUGGING */
416 #define del_SV(p) plant_SV(p)
418 #endif /* DEBUGGING */
422 =head1 SV Manipulation Functions
424 =for apidoc sv_add_arena
426 Given a chunk of memory, link it to the head of the list of arenas,
427 and split it into a list of free SVs.
433 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
436 SV *const sva = MUTABLE_SV(ptr);
440 PERL_ARGS_ASSERT_SV_ADD_ARENA;
442 /* The first SV in an arena isn't an SV. */
443 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
444 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
445 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
447 PL_sv_arenaroot = sva;
448 PL_sv_root = sva + 1;
450 svend = &sva[SvREFCNT(sva) - 1];
453 SvARENA_CHAIN_SET(sv, (sv + 1));
457 /* Must always set typemask because it's always checked in on cleanup
458 when the arenas are walked looking for objects. */
459 SvFLAGS(sv) = SVTYPEMASK;
462 SvARENA_CHAIN_SET(sv, 0);
466 SvFLAGS(sv) = SVTYPEMASK;
469 /* visit(): call the named function for each non-free SV in the arenas
470 * whose flags field matches the flags/mask args. */
473 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
479 PERL_ARGS_ASSERT_VISIT;
481 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
482 const SV * const svend = &sva[SvREFCNT(sva)];
484 for (sv = sva + 1; sv < svend; ++sv) {
485 if (SvTYPE(sv) != (svtype)SVTYPEMASK
486 && (sv->sv_flags & mask) == flags
499 /* called by sv_report_used() for each live SV */
502 do_report_used(pTHX_ SV *const sv)
504 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
505 PerlIO_printf(Perl_debug_log, "****\n");
512 =for apidoc sv_report_used
514 Dump the contents of all SVs not yet freed (debugging aid).
520 Perl_sv_report_used(pTHX)
523 visit(do_report_used, 0, 0);
529 /* called by sv_clean_objs() for each live SV */
532 do_clean_objs(pTHX_ SV *const ref)
537 SV * const target = SvRV(ref);
538 if (SvOBJECT(target)) {
539 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
540 if (SvWEAKREF(ref)) {
541 sv_del_backref(target, ref);
547 SvREFCNT_dec_NN(target);
554 /* clear any slots in a GV which hold objects - except IO;
555 * called by sv_clean_objs() for each live GV */
558 do_clean_named_objs(pTHX_ SV *const sv)
562 assert(SvTYPE(sv) == SVt_PVGV);
563 assert(isGV_with_GP(sv));
567 /* freeing GP entries may indirectly free the current GV;
568 * hold onto it while we mess with the GP slots */
571 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
572 DEBUG_D((PerlIO_printf(Perl_debug_log,
573 "Cleaning named glob SV object:\n "), sv_dump(obj)));
575 SvREFCNT_dec_NN(obj);
577 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
578 DEBUG_D((PerlIO_printf(Perl_debug_log,
579 "Cleaning named glob AV object:\n "), sv_dump(obj)));
581 SvREFCNT_dec_NN(obj);
583 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
584 DEBUG_D((PerlIO_printf(Perl_debug_log,
585 "Cleaning named glob HV object:\n "), sv_dump(obj)));
587 SvREFCNT_dec_NN(obj);
589 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
590 DEBUG_D((PerlIO_printf(Perl_debug_log,
591 "Cleaning named glob CV object:\n "), sv_dump(obj)));
593 SvREFCNT_dec_NN(obj);
595 SvREFCNT_dec_NN(sv); /* undo the inc above */
598 /* clear any IO slots in a GV which hold objects (except stderr, defout);
599 * called by sv_clean_objs() for each live GV */
602 do_clean_named_io_objs(pTHX_ SV *const sv)
606 assert(SvTYPE(sv) == SVt_PVGV);
607 assert(isGV_with_GP(sv));
608 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
612 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
613 DEBUG_D((PerlIO_printf(Perl_debug_log,
614 "Cleaning named glob IO object:\n "), sv_dump(obj)));
616 SvREFCNT_dec_NN(obj);
618 SvREFCNT_dec_NN(sv); /* undo the inc above */
621 /* Void wrapper to pass to visit() */
623 do_curse(pTHX_ SV * const sv) {
624 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
625 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
631 =for apidoc sv_clean_objs
633 Attempt to destroy all objects not yet freed.
639 Perl_sv_clean_objs(pTHX)
643 PL_in_clean_objs = TRUE;
644 visit(do_clean_objs, SVf_ROK, SVf_ROK);
645 /* Some barnacles may yet remain, clinging to typeglobs.
646 * Run the non-IO destructors first: they may want to output
647 * error messages, close files etc */
648 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
649 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
650 /* And if there are some very tenacious barnacles clinging to arrays,
651 closures, or what have you.... */
652 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
653 olddef = PL_defoutgv;
654 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
655 if (olddef && isGV_with_GP(olddef))
656 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
657 olderr = PL_stderrgv;
658 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
659 if (olderr && isGV_with_GP(olderr))
660 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
661 SvREFCNT_dec(olddef);
662 PL_in_clean_objs = FALSE;
665 /* called by sv_clean_all() for each live SV */
668 do_clean_all(pTHX_ SV *const sv)
671 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
672 /* don't clean pid table and strtab */
675 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
676 SvFLAGS(sv) |= SVf_BREAK;
681 =for apidoc sv_clean_all
683 Decrement the refcnt of each remaining SV, possibly triggering a
684 cleanup. This function may have to be called multiple times to free
685 SVs which are in complex self-referential hierarchies.
691 Perl_sv_clean_all(pTHX)
695 PL_in_clean_all = TRUE;
696 cleaned = visit(do_clean_all, 0,0);
701 ARENASETS: a meta-arena implementation which separates arena-info
702 into struct arena_set, which contains an array of struct
703 arena_descs, each holding info for a single arena. By separating
704 the meta-info from the arena, we recover the 1st slot, formerly
705 borrowed for list management. The arena_set is about the size of an
706 arena, avoiding the needless malloc overhead of a naive linked-list.
708 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
709 memory in the last arena-set (1/2 on average). In trade, we get
710 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
711 smaller types). The recovery of the wasted space allows use of
712 small arenas for large, rare body types, by changing array* fields
713 in body_details_by_type[] below.
716 char *arena; /* the raw storage, allocated aligned */
717 size_t size; /* its size ~4k typ */
718 svtype utype; /* bodytype stored in arena */
723 /* Get the maximum number of elements in set[] such that struct arena_set
724 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
725 therefore likely to be 1 aligned memory page. */
727 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
728 - 2 * sizeof(int)) / sizeof (struct arena_desc))
731 struct arena_set* next;
732 unsigned int set_size; /* ie ARENAS_PER_SET */
733 unsigned int curr; /* index of next available arena-desc */
734 struct arena_desc set[ARENAS_PER_SET];
738 =for apidoc sv_free_arenas
740 Deallocate the memory used by all arenas. Note that all the individual SV
741 heads and bodies within the arenas must already have been freed.
747 Perl_sv_free_arenas(pTHX)
754 /* Free arenas here, but be careful about fake ones. (We assume
755 contiguity of the fake ones with the corresponding real ones.) */
757 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
758 svanext = MUTABLE_SV(SvANY(sva));
759 while (svanext && SvFAKE(svanext))
760 svanext = MUTABLE_SV(SvANY(svanext));
767 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
770 struct arena_set *current = aroot;
773 assert(aroot->set[i].arena);
774 Safefree(aroot->set[i].arena);
782 i = PERL_ARENA_ROOTS_SIZE;
784 PL_body_roots[i] = 0;
791 Here are mid-level routines that manage the allocation of bodies out
792 of the various arenas. There are 5 kinds of arenas:
794 1. SV-head arenas, which are discussed and handled above
795 2. regular body arenas
796 3. arenas for reduced-size bodies
799 Arena types 2 & 3 are chained by body-type off an array of
800 arena-root pointers, which is indexed by svtype. Some of the
801 larger/less used body types are malloced singly, since a large
802 unused block of them is wasteful. Also, several svtypes dont have
803 bodies; the data fits into the sv-head itself. The arena-root
804 pointer thus has a few unused root-pointers (which may be hijacked
805 later for arena types 4,5)
807 3 differs from 2 as an optimization; some body types have several
808 unused fields in the front of the structure (which are kept in-place
809 for consistency). These bodies can be allocated in smaller chunks,
810 because the leading fields arent accessed. Pointers to such bodies
811 are decremented to point at the unused 'ghost' memory, knowing that
812 the pointers are used with offsets to the real memory.
815 =head1 SV-Body Allocation
817 Allocation of SV-bodies is similar to SV-heads, differing as follows;
818 the allocation mechanism is used for many body types, so is somewhat
819 more complicated, it uses arena-sets, and has no need for still-live
822 At the outermost level, (new|del)_X*V macros return bodies of the
823 appropriate type. These macros call either (new|del)_body_type or
824 (new|del)_body_allocated macro pairs, depending on specifics of the
825 type. Most body types use the former pair, the latter pair is used to
826 allocate body types with "ghost fields".
828 "ghost fields" are fields that are unused in certain types, and
829 consequently don't need to actually exist. They are declared because
830 they're part of a "base type", which allows use of functions as
831 methods. The simplest examples are AVs and HVs, 2 aggregate types
832 which don't use the fields which support SCALAR semantics.
834 For these types, the arenas are carved up into appropriately sized
835 chunks, we thus avoid wasted memory for those unaccessed members.
836 When bodies are allocated, we adjust the pointer back in memory by the
837 size of the part not allocated, so it's as if we allocated the full
838 structure. (But things will all go boom if you write to the part that
839 is "not there", because you'll be overwriting the last members of the
840 preceding structure in memory.)
842 We calculate the correction using the STRUCT_OFFSET macro on the first
843 member present. If the allocated structure is smaller (no initial NV
844 actually allocated) then the net effect is to subtract the size of the NV
845 from the pointer, to return a new pointer as if an initial NV were actually
846 allocated. (We were using structures named *_allocated for this, but
847 this turned out to be a subtle bug, because a structure without an NV
848 could have a lower alignment constraint, but the compiler is allowed to
849 optimised accesses based on the alignment constraint of the actual pointer
850 to the full structure, for example, using a single 64 bit load instruction
851 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
853 This is the same trick as was used for NV and IV bodies. Ironically it
854 doesn't need to be used for NV bodies any more, because NV is now at
855 the start of the structure. IV bodies don't need it either, because
856 they are no longer allocated.
858 In turn, the new_body_* allocators call S_new_body(), which invokes
859 new_body_inline macro, which takes a lock, and takes a body off the
860 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
861 necessary to refresh an empty list. Then the lock is released, and
862 the body is returned.
864 Perl_more_bodies allocates a new arena, and carves it up into an array of N
865 bodies, which it strings into a linked list. It looks up arena-size
866 and body-size from the body_details table described below, thus
867 supporting the multiple body-types.
869 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
870 the (new|del)_X*V macros are mapped directly to malloc/free.
872 For each sv-type, struct body_details bodies_by_type[] carries
873 parameters which control these aspects of SV handling:
875 Arena_size determines whether arenas are used for this body type, and if
876 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
877 zero, forcing individual mallocs and frees.
879 Body_size determines how big a body is, and therefore how many fit into
880 each arena. Offset carries the body-pointer adjustment needed for
881 "ghost fields", and is used in *_allocated macros.
883 But its main purpose is to parameterize info needed in
884 Perl_sv_upgrade(). The info here dramatically simplifies the function
885 vs the implementation in 5.8.8, making it table-driven. All fields
886 are used for this, except for arena_size.
888 For the sv-types that have no bodies, arenas are not used, so those
889 PL_body_roots[sv_type] are unused, and can be overloaded. In
890 something of a special case, SVt_NULL is borrowed for HE arenas;
891 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
892 bodies_by_type[SVt_NULL] slot is not used, as the table is not
897 struct body_details {
898 U8 body_size; /* Size to allocate */
899 U8 copy; /* Size of structure to copy (may be shorter) */
901 unsigned int type : 4; /* We have space for a sanity check. */
902 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
903 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
904 unsigned int arena : 1; /* Allocated from an arena */
905 size_t arena_size; /* Size of arena to allocate */
913 /* With -DPURFIY we allocate everything directly, and don't use arenas.
914 This seems a rather elegant way to simplify some of the code below. */
915 #define HASARENA FALSE
917 #define HASARENA TRUE
919 #define NOARENA FALSE
921 /* Size the arenas to exactly fit a given number of bodies. A count
922 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
923 simplifying the default. If count > 0, the arena is sized to fit
924 only that many bodies, allowing arenas to be used for large, rare
925 bodies (XPVFM, XPVIO) without undue waste. The arena size is
926 limited by PERL_ARENA_SIZE, so we can safely oversize the
929 #define FIT_ARENA0(body_size) \
930 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
931 #define FIT_ARENAn(count,body_size) \
932 ( count * body_size <= PERL_ARENA_SIZE) \
933 ? count * body_size \
934 : FIT_ARENA0 (body_size)
935 #define FIT_ARENA(count,body_size) \
937 ? FIT_ARENAn (count, body_size) \
938 : FIT_ARENA0 (body_size)
940 /* Calculate the length to copy. Specifically work out the length less any
941 final padding the compiler needed to add. See the comment in sv_upgrade
942 for why copying the padding proved to be a bug. */
944 #define copy_length(type, last_member) \
945 STRUCT_OFFSET(type, last_member) \
946 + sizeof (((type*)SvANY((const SV *)0))->last_member)
948 static const struct body_details bodies_by_type[] = {
949 /* HEs use this offset for their arena. */
950 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
952 /* IVs are in the head, so the allocation size is 0. */
954 sizeof(IV), /* This is used to copy out the IV body. */
955 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
956 NOARENA /* IVS don't need an arena */, 0
959 { sizeof(NV), sizeof(NV),
960 STRUCT_OFFSET(XPVNV, xnv_u),
961 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
963 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_INVLIST, TRUE, NONV, HASARENA,
973 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
976 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
977 + STRUCT_OFFSET(XPV, xpv_cur),
978 SVt_PVIV, FALSE, NONV, HASARENA,
979 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
981 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
982 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
983 + STRUCT_OFFSET(XPV, xpv_cur),
984 SVt_PVNV, FALSE, HADNV, HASARENA,
985 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
987 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
988 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
993 SVt_REGEXP, TRUE, NONV, HASARENA,
994 FIT_ARENA(0, sizeof(regexp))
997 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
998 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
1000 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1001 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
1004 copy_length(XPVAV, xav_alloc),
1006 SVt_PVAV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVAV)) },
1010 copy_length(XPVHV, xhv_max),
1012 SVt_PVHV, TRUE, NONV, HASARENA,
1013 FIT_ARENA(0, sizeof(XPVHV)) },
1018 SVt_PVCV, TRUE, NONV, HASARENA,
1019 FIT_ARENA(0, sizeof(XPVCV)) },
1024 SVt_PVFM, TRUE, NONV, NOARENA,
1025 FIT_ARENA(20, sizeof(XPVFM)) },
1030 SVt_PVIO, TRUE, NONV, HASARENA,
1031 FIT_ARENA(24, sizeof(XPVIO)) },
1034 #define new_body_allocated(sv_type) \
1035 (void *)((char *)S_new_body(aTHX_ sv_type) \
1036 - bodies_by_type[sv_type].offset)
1038 /* return a thing to the free list */
1040 #define del_body(thing, root) \
1042 void ** const thing_copy = (void **)thing; \
1043 *thing_copy = *root; \
1044 *root = (void*)thing_copy; \
1049 #define new_XNV() safemalloc(sizeof(XPVNV))
1050 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1051 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1053 #define del_XPVGV(p) safefree(p)
1057 #define new_XNV() new_body_allocated(SVt_NV)
1058 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1059 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1061 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1062 &PL_body_roots[SVt_PVGV])
1066 /* no arena for you! */
1068 #define new_NOARENA(details) \
1069 safemalloc((details)->body_size + (details)->offset)
1070 #define new_NOARENAZ(details) \
1071 safecalloc((details)->body_size + (details)->offset, 1)
1074 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1075 const size_t arena_size)
1078 void ** const root = &PL_body_roots[sv_type];
1079 struct arena_desc *adesc;
1080 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1084 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1085 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1086 static bool done_sanity_check;
1088 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1089 * variables like done_sanity_check. */
1090 if (!done_sanity_check) {
1091 unsigned int i = SVt_LAST;
1093 done_sanity_check = TRUE;
1096 assert (bodies_by_type[i].type == i);
1102 /* may need new arena-set to hold new arena */
1103 if (!aroot || aroot->curr >= aroot->set_size) {
1104 struct arena_set *newroot;
1105 Newxz(newroot, 1, struct arena_set);
1106 newroot->set_size = ARENAS_PER_SET;
1107 newroot->next = aroot;
1109 PL_body_arenas = (void *) newroot;
1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1113 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1114 curr = aroot->curr++;
1115 adesc = &(aroot->set[curr]);
1116 assert(!adesc->arena);
1118 Newx(adesc->arena, good_arena_size, char);
1119 adesc->size = good_arena_size;
1120 adesc->utype = sv_type;
1121 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1122 curr, (void*)adesc->arena, (UV)good_arena_size));
1124 start = (char *) adesc->arena;
1126 /* Get the address of the byte after the end of the last body we can fit.
1127 Remember, this is integer division: */
1128 end = start + good_arena_size / body_size * body_size;
1130 /* computed count doesn't reflect the 1st slot reservation */
1131 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d (from %d) type %d "
1135 (void*)start, (void*)end, (int)good_arena_size,
1136 (int)arena_size, sv_type, (int)body_size,
1137 (int)good_arena_size / (int)body_size));
1139 DEBUG_m(PerlIO_printf(Perl_debug_log,
1140 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1141 (void*)start, (void*)end,
1142 (int)arena_size, sv_type, (int)body_size,
1143 (int)good_arena_size / (int)body_size));
1145 *root = (void *)start;
1148 /* Where the next body would start: */
1149 char * const next = start + body_size;
1152 /* This is the last body: */
1153 assert(next == end);
1155 *(void **)start = 0;
1159 *(void**) start = (void *)next;
1164 /* grab a new thing from the free list, allocating more if necessary.
1165 The inline version is used for speed in hot routines, and the
1166 function using it serves the rest (unless PURIFY).
1168 #define new_body_inline(xpv, sv_type) \
1170 void ** const r3wt = &PL_body_roots[sv_type]; \
1171 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1172 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1173 bodies_by_type[sv_type].body_size,\
1174 bodies_by_type[sv_type].arena_size)); \
1175 *(r3wt) = *(void**)(xpv); \
1181 S_new_body(pTHX_ const svtype sv_type)
1185 new_body_inline(xpv, sv_type);
1191 static const struct body_details fake_rv =
1192 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1195 =for apidoc sv_upgrade
1197 Upgrade an SV to a more complex form. Generally adds a new body type to the
1198 SV, then copies across as much information as possible from the old body.
1199 It croaks if the SV is already in a more complex form than requested. You
1200 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1201 before calling C<sv_upgrade>, and hence does not croak. See also
1208 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1213 const svtype old_type = SvTYPE(sv);
1214 const struct body_details *new_type_details;
1215 const struct body_details *old_type_details
1216 = bodies_by_type + old_type;
1217 SV *referant = NULL;
1219 PERL_ARGS_ASSERT_SV_UPGRADE;
1221 if (old_type == new_type)
1224 /* This clause was purposefully added ahead of the early return above to
1225 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1226 inference by Nick I-S that it would fix other troublesome cases. See
1227 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1229 Given that shared hash key scalars are no longer PVIV, but PV, there is
1230 no longer need to unshare so as to free up the IVX slot for its proper
1231 purpose. So it's safe to move the early return earlier. */
1233 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1234 sv_force_normal_flags(sv, 0);
1237 old_body = SvANY(sv);
1239 /* Copying structures onto other structures that have been neatly zeroed
1240 has a subtle gotcha. Consider XPVMG
1242 +------+------+------+------+------+-------+-------+
1243 | NV | CUR | LEN | IV | MAGIC | STASH |
1244 +------+------+------+------+------+-------+-------+
1245 0 4 8 12 16 20 24 28
1247 where NVs are aligned to 8 bytes, so that sizeof that structure is
1248 actually 32 bytes long, with 4 bytes of padding at the end:
1250 +------+------+------+------+------+-------+-------+------+
1251 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1252 +------+------+------+------+------+-------+-------+------+
1253 0 4 8 12 16 20 24 28 32
1255 so what happens if you allocate memory for this structure:
1257 +------+------+------+------+------+-------+-------+------+------+...
1258 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1259 +------+------+------+------+------+-------+-------+------+------+...
1260 0 4 8 12 16 20 24 28 32 36
1262 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1263 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1264 started out as zero once, but it's quite possible that it isn't. So now,
1265 rather than a nicely zeroed GP, you have it pointing somewhere random.
1268 (In fact, GP ends up pointing at a previous GP structure, because the
1269 principle cause of the padding in XPVMG getting garbage is a copy of
1270 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1271 this happens to be moot because XPVGV has been re-ordered, with GP
1272 no longer after STASH)
1274 So we are careful and work out the size of used parts of all the
1282 referant = SvRV(sv);
1283 old_type_details = &fake_rv;
1284 if (new_type == SVt_NV)
1285 new_type = SVt_PVNV;
1287 if (new_type < SVt_PVIV) {
1288 new_type = (new_type == SVt_NV)
1289 ? SVt_PVNV : SVt_PVIV;
1294 if (new_type < SVt_PVNV) {
1295 new_type = SVt_PVNV;
1299 assert(new_type > SVt_PV);
1300 assert(SVt_IV < SVt_PV);
1301 assert(SVt_NV < SVt_PV);
1308 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1309 there's no way that it can be safely upgraded, because perl.c
1310 expects to Safefree(SvANY(PL_mess_sv)) */
1311 assert(sv != PL_mess_sv);
1312 /* This flag bit is used to mean other things in other scalar types.
1313 Given that it only has meaning inside the pad, it shouldn't be set
1314 on anything that can get upgraded. */
1315 assert(!SvPAD_TYPED(sv));
1318 if (UNLIKELY(old_type_details->cant_upgrade))
1319 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1320 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1323 if (UNLIKELY(old_type > new_type))
1324 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1325 (int)old_type, (int)new_type);
1327 new_type_details = bodies_by_type + new_type;
1329 SvFLAGS(sv) &= ~SVTYPEMASK;
1330 SvFLAGS(sv) |= new_type;
1332 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1333 the return statements above will have triggered. */
1334 assert (new_type != SVt_NULL);
1337 assert(old_type == SVt_NULL);
1338 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1342 assert(old_type == SVt_NULL);
1343 SvANY(sv) = new_XNV();
1348 assert(new_type_details->body_size);
1351 assert(new_type_details->arena);
1352 assert(new_type_details->arena_size);
1353 /* This points to the start of the allocated area. */
1354 new_body_inline(new_body, new_type);
1355 Zero(new_body, new_type_details->body_size, char);
1356 new_body = ((char *)new_body) - new_type_details->offset;
1358 /* We always allocated the full length item with PURIFY. To do this
1359 we fake things so that arena is false for all 16 types.. */
1360 new_body = new_NOARENAZ(new_type_details);
1362 SvANY(sv) = new_body;
1363 if (new_type == SVt_PVAV) {
1367 if (old_type_details->body_size) {
1370 /* It will have been zeroed when the new body was allocated.
1371 Lets not write to it, in case it confuses a write-back
1377 #ifndef NODEFAULT_SHAREKEYS
1378 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1380 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1381 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1384 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1385 The target created by newSVrv also is, and it can have magic.
1386 However, it never has SvPVX set.
1388 if (old_type == SVt_IV) {
1390 } else if (old_type >= SVt_PV) {
1391 assert(SvPVX_const(sv) == 0);
1394 if (old_type >= SVt_PVMG) {
1395 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1396 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1398 sv->sv_u.svu_array = NULL; /* or svu_hash */
1403 /* XXX Is this still needed? Was it ever needed? Surely as there is
1404 no route from NV to PVIV, NOK can never be true */
1405 assert(!SvNOKp(sv));
1418 assert(new_type_details->body_size);
1419 /* We always allocated the full length item with PURIFY. To do this
1420 we fake things so that arena is false for all 16 types.. */
1421 if(new_type_details->arena) {
1422 /* This points to the start of the allocated area. */
1423 new_body_inline(new_body, new_type);
1424 Zero(new_body, new_type_details->body_size, char);
1425 new_body = ((char *)new_body) - new_type_details->offset;
1427 new_body = new_NOARENAZ(new_type_details);
1429 SvANY(sv) = new_body;
1431 if (old_type_details->copy) {
1432 /* There is now the potential for an upgrade from something without
1433 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1434 int offset = old_type_details->offset;
1435 int length = old_type_details->copy;
1437 if (new_type_details->offset > old_type_details->offset) {
1438 const int difference
1439 = new_type_details->offset - old_type_details->offset;
1440 offset += difference;
1441 length -= difference;
1443 assert (length >= 0);
1445 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1449 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1450 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1451 * correct 0.0 for us. Otherwise, if the old body didn't have an
1452 * NV slot, but the new one does, then we need to initialise the
1453 * freshly created NV slot with whatever the correct bit pattern is
1455 if (old_type_details->zero_nv && !new_type_details->zero_nv
1456 && !isGV_with_GP(sv))
1460 if (UNLIKELY(new_type == SVt_PVIO)) {
1461 IO * const io = MUTABLE_IO(sv);
1462 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1465 /* Clear the stashcache because a new IO could overrule a package
1467 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1468 hv_clear(PL_stashcache);
1470 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1471 IoPAGE_LEN(sv) = 60;
1473 if (UNLIKELY(new_type == SVt_REGEXP))
1474 sv->sv_u.svu_rx = (regexp *)new_body;
1475 else if (old_type < SVt_PV) {
1476 /* referant will be NULL unless the old type was SVt_IV emulating
1478 sv->sv_u.svu_rv = referant;
1482 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1483 (unsigned long)new_type);
1486 if (old_type > SVt_IV) {
1490 /* Note that there is an assumption that all bodies of types that
1491 can be upgraded came from arenas. Only the more complex non-
1492 upgradable types are allowed to be directly malloc()ed. */
1493 assert(old_type_details->arena);
1494 del_body((void*)((char*)old_body + old_type_details->offset),
1495 &PL_body_roots[old_type]);
1501 =for apidoc sv_backoff
1503 Remove any string offset. You should normally use the C<SvOOK_off> macro
1510 Perl_sv_backoff(pTHX_ SV *const sv)
1513 const char * const s = SvPVX_const(sv);
1515 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 PERL_UNUSED_CONTEXT;
1519 assert(SvTYPE(sv) != SVt_PVHV);
1520 assert(SvTYPE(sv) != SVt_PVAV);
1522 SvOOK_offset(sv, delta);
1524 SvLEN_set(sv, SvLEN(sv) + delta);
1525 SvPV_set(sv, SvPVX(sv) - delta);
1526 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1527 SvFLAGS(sv) &= ~SVf_OOK;
1534 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1535 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1536 Use the C<SvGROW> wrapper instead.
1541 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1544 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1548 PERL_ARGS_ASSERT_SV_GROW;
1552 if (SvTYPE(sv) < SVt_PV) {
1553 sv_upgrade(sv, SVt_PV);
1554 s = SvPVX_mutable(sv);
1556 else if (SvOOK(sv)) { /* pv is offset? */
1558 s = SvPVX_mutable(sv);
1559 if (newlen > SvLEN(sv))
1560 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1564 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1565 s = SvPVX_mutable(sv);
1568 #ifdef PERL_NEW_COPY_ON_WRITE
1569 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1570 * to store the COW count. So in general, allocate one more byte than
1571 * asked for, to make it likely this byte is always spare: and thus
1572 * make more strings COW-able.
1573 * If the new size is a big power of two, don't bother: we assume the
1574 * caller wanted a nice 2^N sized block and will be annoyed at getting
1580 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1581 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1584 if (newlen > SvLEN(sv)) { /* need more room? */
1585 STRLEN minlen = SvCUR(sv);
1586 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1587 if (newlen < minlen)
1589 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1591 /* Don't round up on the first allocation, as odds are pretty good that
1592 * the initial request is accurate as to what is really needed */
1594 newlen = PERL_STRLEN_ROUNDUP(newlen);
1597 if (SvLEN(sv) && s) {
1598 s = (char*)saferealloc(s, newlen);
1601 s = (char*)safemalloc(newlen);
1602 if (SvPVX_const(sv) && SvCUR(sv)) {
1603 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1607 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1608 /* Do this here, do it once, do it right, and then we will never get
1609 called back into sv_grow() unless there really is some growing
1611 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1613 SvLEN_set(sv, newlen);
1620 =for apidoc sv_setiv
1622 Copies an integer into the given SV, upgrading first if necessary.
1623 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1629 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1633 PERL_ARGS_ASSERT_SV_SETIV;
1635 SV_CHECK_THINKFIRST_COW_DROP(sv);
1636 switch (SvTYPE(sv)) {
1639 sv_upgrade(sv, SVt_IV);
1642 sv_upgrade(sv, SVt_PVIV);
1646 if (!isGV_with_GP(sv))
1653 /* diag_listed_as: Can't coerce %s to %s in %s */
1654 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1658 (void)SvIOK_only(sv); /* validate number */
1664 =for apidoc sv_setiv_mg
1666 Like C<sv_setiv>, but also handles 'set' magic.
1672 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1674 PERL_ARGS_ASSERT_SV_SETIV_MG;
1681 =for apidoc sv_setuv
1683 Copies an unsigned integer into the given SV, upgrading first if necessary.
1684 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1690 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1692 PERL_ARGS_ASSERT_SV_SETUV;
1694 /* With the if statement to ensure that integers are stored as IVs whenever
1696 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1699 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1701 If you wish to remove the following if statement, so that this routine
1702 (and its callers) always return UVs, please benchmark to see what the
1703 effect is. Modern CPUs may be different. Or may not :-)
1705 if (u <= (UV)IV_MAX) {
1706 sv_setiv(sv, (IV)u);
1715 =for apidoc sv_setuv_mg
1717 Like C<sv_setuv>, but also handles 'set' magic.
1723 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1725 PERL_ARGS_ASSERT_SV_SETUV_MG;
1732 =for apidoc sv_setnv
1734 Copies a double into the given SV, upgrading first if necessary.
1735 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1741 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1745 PERL_ARGS_ASSERT_SV_SETNV;
1747 SV_CHECK_THINKFIRST_COW_DROP(sv);
1748 switch (SvTYPE(sv)) {
1751 sv_upgrade(sv, SVt_NV);
1755 sv_upgrade(sv, SVt_PVNV);
1759 if (!isGV_with_GP(sv))
1766 /* diag_listed_as: Can't coerce %s to %s in %s */
1767 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1772 (void)SvNOK_only(sv); /* validate number */
1777 =for apidoc sv_setnv_mg
1779 Like C<sv_setnv>, but also handles 'set' magic.
1785 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1793 /* Print an "isn't numeric" warning, using a cleaned-up,
1794 * printable version of the offending string
1798 S_not_a_number(pTHX_ SV *const sv)
1805 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1808 dsv = newSVpvs_flags("", SVs_TEMP);
1809 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1812 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
1813 /* each *s can expand to 4 chars + "...\0",
1814 i.e. need room for 8 chars */
1816 const char *s = SvPVX_const(sv);
1817 const char * const end = s + SvCUR(sv);
1818 for ( ; s < end && d < limit; s++ ) {
1820 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1824 /* Map to ASCII "equivalent" of Latin1 */
1825 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1831 else if (ch == '\r') {
1835 else if (ch == '\f') {
1839 else if (ch == '\\') {
1843 else if (ch == '\0') {
1847 else if (isPRINT_LC(ch))
1864 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1865 /* diag_listed_as: Argument "%s" isn't numeric%s */
1866 "Argument \"%s\" isn't numeric in %s", pv,
1869 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1870 /* diag_listed_as: Argument "%s" isn't numeric%s */
1871 "Argument \"%s\" isn't numeric", pv);
1875 =for apidoc looks_like_number
1877 Test if the content of an SV looks like a number (or is a number).
1878 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1879 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1886 Perl_looks_like_number(pTHX_ SV *const sv)
1891 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1893 if (SvPOK(sv) || SvPOKp(sv)) {
1894 sbegin = SvPV_nomg_const(sv, len);
1897 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1898 return grok_number(sbegin, len, NULL);
1902 S_glob_2number(pTHX_ GV * const gv)
1904 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1906 /* We know that all GVs stringify to something that is not-a-number,
1907 so no need to test that. */
1908 if (ckWARN(WARN_NUMERIC))
1910 SV *const buffer = sv_newmortal();
1911 gv_efullname3(buffer, gv, "*");
1912 not_a_number(buffer);
1914 /* We just want something true to return, so that S_sv_2iuv_common
1915 can tail call us and return true. */
1919 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1920 until proven guilty, assume that things are not that bad... */
1925 As 64 bit platforms often have an NV that doesn't preserve all bits of
1926 an IV (an assumption perl has been based on to date) it becomes necessary
1927 to remove the assumption that the NV always carries enough precision to
1928 recreate the IV whenever needed, and that the NV is the canonical form.
1929 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1930 precision as a side effect of conversion (which would lead to insanity
1931 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1932 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1933 where precision was lost, and IV/UV/NV slots that have a valid conversion
1934 which has lost no precision
1935 2) to ensure that if a numeric conversion to one form is requested that
1936 would lose precision, the precise conversion (or differently
1937 imprecise conversion) is also performed and cached, to prevent
1938 requests for different numeric formats on the same SV causing
1939 lossy conversion chains. (lossless conversion chains are perfectly
1944 SvIOKp is true if the IV slot contains a valid value
1945 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1946 SvNOKp is true if the NV slot contains a valid value
1947 SvNOK is true only if the NV value is accurate
1950 while converting from PV to NV, check to see if converting that NV to an
1951 IV(or UV) would lose accuracy over a direct conversion from PV to
1952 IV(or UV). If it would, cache both conversions, return NV, but mark
1953 SV as IOK NOKp (ie not NOK).
1955 While converting from PV to IV, check to see if converting that IV to an
1956 NV would lose accuracy over a direct conversion from PV to NV. If it
1957 would, cache both conversions, flag similarly.
1959 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1960 correctly because if IV & NV were set NV *always* overruled.
1961 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1962 changes - now IV and NV together means that the two are interchangeable:
1963 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1965 The benefit of this is that operations such as pp_add know that if
1966 SvIOK is true for both left and right operands, then integer addition
1967 can be used instead of floating point (for cases where the result won't
1968 overflow). Before, floating point was always used, which could lead to
1969 loss of precision compared with integer addition.
1971 * making IV and NV equal status should make maths accurate on 64 bit
1973 * may speed up maths somewhat if pp_add and friends start to use
1974 integers when possible instead of fp. (Hopefully the overhead in
1975 looking for SvIOK and checking for overflow will not outweigh the
1976 fp to integer speedup)
1977 * will slow down integer operations (callers of SvIV) on "inaccurate"
1978 values, as the change from SvIOK to SvIOKp will cause a call into
1979 sv_2iv each time rather than a macro access direct to the IV slot
1980 * should speed up number->string conversion on integers as IV is
1981 favoured when IV and NV are equally accurate
1983 ####################################################################
1984 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1985 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1986 On the other hand, SvUOK is true iff UV.
1987 ####################################################################
1989 Your mileage will vary depending your CPU's relative fp to integer
1993 #ifndef NV_PRESERVES_UV
1994 # define IS_NUMBER_UNDERFLOW_IV 1
1995 # define IS_NUMBER_UNDERFLOW_UV 2
1996 # define IS_NUMBER_IV_AND_UV 2
1997 # define IS_NUMBER_OVERFLOW_IV 4
1998 # define IS_NUMBER_OVERFLOW_UV 5
2000 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2002 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2004 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2012 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2014 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));
2015 if (SvNVX(sv) < (NV)IV_MIN) {
2016 (void)SvIOKp_on(sv);
2018 SvIV_set(sv, IV_MIN);
2019 return IS_NUMBER_UNDERFLOW_IV;
2021 if (SvNVX(sv) > (NV)UV_MAX) {
2022 (void)SvIOKp_on(sv);
2025 SvUV_set(sv, UV_MAX);
2026 return IS_NUMBER_OVERFLOW_UV;
2028 (void)SvIOKp_on(sv);
2030 /* Can't use strtol etc to convert this string. (See truth table in
2032 if (SvNVX(sv) <= (UV)IV_MAX) {
2033 SvIV_set(sv, I_V(SvNVX(sv)));
2034 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2035 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2037 /* Integer is imprecise. NOK, IOKp */
2039 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2042 SvUV_set(sv, U_V(SvNVX(sv)));
2043 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2044 if (SvUVX(sv) == UV_MAX) {
2045 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2046 possibly be preserved by NV. Hence, it must be overflow.
2048 return IS_NUMBER_OVERFLOW_UV;
2050 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2052 /* Integer is imprecise. NOK, IOKp */
2054 return IS_NUMBER_OVERFLOW_IV;
2056 #endif /* !NV_PRESERVES_UV*/
2059 S_sv_2iuv_common(pTHX_ SV *const sv)
2063 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2066 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2067 * without also getting a cached IV/UV from it at the same time
2068 * (ie PV->NV conversion should detect loss of accuracy and cache
2069 * IV or UV at same time to avoid this. */
2070 /* IV-over-UV optimisation - choose to cache IV if possible */
2072 if (SvTYPE(sv) == SVt_NV)
2073 sv_upgrade(sv, SVt_PVNV);
2075 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2076 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2077 certainly cast into the IV range at IV_MAX, whereas the correct
2078 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2080 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2081 if (Perl_isnan(SvNVX(sv))) {
2087 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2088 SvIV_set(sv, I_V(SvNVX(sv)));
2089 if (SvNVX(sv) == (NV) SvIVX(sv)
2090 #ifndef NV_PRESERVES_UV
2091 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2092 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2093 /* Don't flag it as "accurately an integer" if the number
2094 came from a (by definition imprecise) NV operation, and
2095 we're outside the range of NV integer precision */
2099 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2101 /* scalar has trailing garbage, eg "42a" */
2103 DEBUG_c(PerlIO_printf(Perl_debug_log,
2104 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2110 /* IV not precise. No need to convert from PV, as NV
2111 conversion would already have cached IV if it detected
2112 that PV->IV would be better than PV->NV->IV
2113 flags already correct - don't set public IOK. */
2114 DEBUG_c(PerlIO_printf(Perl_debug_log,
2115 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2120 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2121 but the cast (NV)IV_MIN rounds to a the value less (more
2122 negative) than IV_MIN which happens to be equal to SvNVX ??
2123 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2124 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2125 (NV)UVX == NVX are both true, but the values differ. :-(
2126 Hopefully for 2s complement IV_MIN is something like
2127 0x8000000000000000 which will be exact. NWC */
2130 SvUV_set(sv, U_V(SvNVX(sv)));
2132 (SvNVX(sv) == (NV) SvUVX(sv))
2133 #ifndef NV_PRESERVES_UV
2134 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2135 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2136 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2137 /* Don't flag it as "accurately an integer" if the number
2138 came from a (by definition imprecise) NV operation, and
2139 we're outside the range of NV integer precision */
2145 DEBUG_c(PerlIO_printf(Perl_debug_log,
2146 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2152 else if (SvPOKp(sv)) {
2154 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2155 /* We want to avoid a possible problem when we cache an IV/ a UV which
2156 may be later translated to an NV, and the resulting NV is not
2157 the same as the direct translation of the initial string
2158 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2159 be careful to ensure that the value with the .456 is around if the
2160 NV value is requested in the future).
2162 This means that if we cache such an IV/a UV, we need to cache the
2163 NV as well. Moreover, we trade speed for space, and do not
2164 cache the NV if we are sure it's not needed.
2167 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2168 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2169 == IS_NUMBER_IN_UV) {
2170 /* It's definitely an integer, only upgrade to PVIV */
2171 if (SvTYPE(sv) < SVt_PVIV)
2172 sv_upgrade(sv, SVt_PVIV);
2174 } else if (SvTYPE(sv) < SVt_PVNV)
2175 sv_upgrade(sv, SVt_PVNV);
2177 /* If NVs preserve UVs then we only use the UV value if we know that
2178 we aren't going to call atof() below. If NVs don't preserve UVs
2179 then the value returned may have more precision than atof() will
2180 return, even though value isn't perfectly accurate. */
2181 if ((numtype & (IS_NUMBER_IN_UV
2182 #ifdef NV_PRESERVES_UV
2185 )) == IS_NUMBER_IN_UV) {
2186 /* This won't turn off the public IOK flag if it was set above */
2187 (void)SvIOKp_on(sv);
2189 if (!(numtype & IS_NUMBER_NEG)) {
2191 if (value <= (UV)IV_MAX) {
2192 SvIV_set(sv, (IV)value);
2194 /* it didn't overflow, and it was positive. */
2195 SvUV_set(sv, value);
2199 /* 2s complement assumption */
2200 if (value <= (UV)IV_MIN) {
2201 SvIV_set(sv, -(IV)value);
2203 /* Too negative for an IV. This is a double upgrade, but
2204 I'm assuming it will be rare. */
2205 if (SvTYPE(sv) < SVt_PVNV)
2206 sv_upgrade(sv, SVt_PVNV);
2210 SvNV_set(sv, -(NV)value);
2211 SvIV_set(sv, IV_MIN);
2215 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2216 will be in the previous block to set the IV slot, and the next
2217 block to set the NV slot. So no else here. */
2219 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2220 != IS_NUMBER_IN_UV) {
2221 /* It wasn't an (integer that doesn't overflow the UV). */
2222 SvNV_set(sv, Atof(SvPVX_const(sv)));
2224 if (! numtype && ckWARN(WARN_NUMERIC))
2227 #if defined(USE_LONG_DOUBLE)
2228 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2229 PTR2UV(sv), SvNVX(sv)));
2231 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2232 PTR2UV(sv), SvNVX(sv)));
2235 #ifdef NV_PRESERVES_UV
2236 (void)SvIOKp_on(sv);
2238 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2239 SvIV_set(sv, I_V(SvNVX(sv)));
2240 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2243 NOOP; /* Integer is imprecise. NOK, IOKp */
2245 /* UV will not work better than IV */
2247 if (SvNVX(sv) > (NV)UV_MAX) {
2249 /* Integer is inaccurate. NOK, IOKp, is UV */
2250 SvUV_set(sv, UV_MAX);
2252 SvUV_set(sv, U_V(SvNVX(sv)));
2253 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2254 NV preservse UV so can do correct comparison. */
2255 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2258 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2263 #else /* NV_PRESERVES_UV */
2264 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2265 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2266 /* The IV/UV slot will have been set from value returned by
2267 grok_number above. The NV slot has just been set using
2270 assert (SvIOKp(sv));
2272 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2273 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2274 /* Small enough to preserve all bits. */
2275 (void)SvIOKp_on(sv);
2277 SvIV_set(sv, I_V(SvNVX(sv)));
2278 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2280 /* Assumption: first non-preserved integer is < IV_MAX,
2281 this NV is in the preserved range, therefore: */
2282 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2284 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);
2288 0 0 already failed to read UV.
2289 0 1 already failed to read UV.
2290 1 0 you won't get here in this case. IV/UV
2291 slot set, public IOK, Atof() unneeded.
2292 1 1 already read UV.
2293 so there's no point in sv_2iuv_non_preserve() attempting
2294 to use atol, strtol, strtoul etc. */
2296 sv_2iuv_non_preserve (sv, numtype);
2298 sv_2iuv_non_preserve (sv);
2302 #endif /* NV_PRESERVES_UV */
2303 /* It might be more code efficient to go through the entire logic above
2304 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2305 gets complex and potentially buggy, so more programmer efficient
2306 to do it this way, by turning off the public flags: */
2308 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2312 if (isGV_with_GP(sv))
2313 return glob_2number(MUTABLE_GV(sv));
2315 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2317 if (SvTYPE(sv) < SVt_IV)
2318 /* Typically the caller expects that sv_any is not NULL now. */
2319 sv_upgrade(sv, SVt_IV);
2320 /* Return 0 from the caller. */
2327 =for apidoc sv_2iv_flags
2329 Return the integer value of an SV, doing any necessary string
2330 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2331 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2337 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2341 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2343 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2344 && SvTYPE(sv) != SVt_PVFM);
2346 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2352 if (flags & SV_SKIP_OVERLOAD)
2354 tmpstr = AMG_CALLunary(sv, numer_amg);
2355 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2356 return SvIV(tmpstr);
2359 return PTR2IV(SvRV(sv));
2362 if (SvVALID(sv) || isREGEXP(sv)) {
2363 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2364 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2365 In practice they are extremely unlikely to actually get anywhere
2366 accessible by user Perl code - the only way that I'm aware of is when
2367 a constant subroutine which is used as the second argument to index.
2369 Regexps have no SvIVX and SvNVX fields.
2371 assert(isREGEXP(sv) || SvPOKp(sv));
2374 const char * const ptr =
2375 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2377 = grok_number(ptr, SvCUR(sv), &value);
2379 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2380 == IS_NUMBER_IN_UV) {
2381 /* It's definitely an integer */
2382 if (numtype & IS_NUMBER_NEG) {
2383 if (value < (UV)IV_MIN)
2386 if (value < (UV)IV_MAX)
2391 if (ckWARN(WARN_NUMERIC))
2394 return I_V(Atof(ptr));
2398 if (SvTHINKFIRST(sv)) {
2399 #ifdef PERL_OLD_COPY_ON_WRITE
2401 sv_force_normal_flags(sv, 0);
2404 if (SvREADONLY(sv) && !SvOK(sv)) {
2405 if (ckWARN(WARN_UNINITIALIZED))
2412 if (S_sv_2iuv_common(aTHX_ sv))
2416 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2417 PTR2UV(sv),SvIVX(sv)));
2418 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2422 =for apidoc sv_2uv_flags
2424 Return the unsigned integer value of an SV, doing any necessary string
2425 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2426 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2432 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2436 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2438 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2444 if (flags & SV_SKIP_OVERLOAD)
2446 tmpstr = AMG_CALLunary(sv, numer_amg);
2447 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2448 return SvUV(tmpstr);
2451 return PTR2UV(SvRV(sv));
2454 if (SvVALID(sv) || isREGEXP(sv)) {
2455 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2456 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2457 Regexps have no SvIVX and SvNVX fields. */
2458 assert(isREGEXP(sv) || SvPOKp(sv));
2461 const char * const ptr =
2462 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2464 = grok_number(ptr, SvCUR(sv), &value);
2466 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2467 == IS_NUMBER_IN_UV) {
2468 /* It's definitely an integer */
2469 if (!(numtype & IS_NUMBER_NEG))
2473 if (ckWARN(WARN_NUMERIC))
2476 return U_V(Atof(ptr));
2480 if (SvTHINKFIRST(sv)) {
2481 #ifdef PERL_OLD_COPY_ON_WRITE
2483 sv_force_normal_flags(sv, 0);
2486 if (SvREADONLY(sv) && !SvOK(sv)) {
2487 if (ckWARN(WARN_UNINITIALIZED))
2494 if (S_sv_2iuv_common(aTHX_ sv))
2498 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2499 PTR2UV(sv),SvUVX(sv)));
2500 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2504 =for apidoc sv_2nv_flags
2506 Return the num value of an SV, doing any necessary string or integer
2507 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2508 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2514 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2518 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2520 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2521 && SvTYPE(sv) != SVt_PVFM);
2522 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2523 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2524 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2525 Regexps have no SvIVX and SvNVX fields. */
2527 if (flags & SV_GMAGIC)
2531 if (SvPOKp(sv) && !SvIOKp(sv)) {
2532 ptr = SvPVX_const(sv);
2534 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2535 !grok_number(ptr, SvCUR(sv), NULL))
2541 return (NV)SvUVX(sv);
2543 return (NV)SvIVX(sv);
2549 ptr = RX_WRAPPED((REGEXP *)sv);
2552 assert(SvTYPE(sv) >= SVt_PVMG);
2553 /* This falls through to the report_uninit near the end of the
2555 } else if (SvTHINKFIRST(sv)) {
2560 if (flags & SV_SKIP_OVERLOAD)
2562 tmpstr = AMG_CALLunary(sv, numer_amg);
2563 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2564 return SvNV(tmpstr);
2567 return PTR2NV(SvRV(sv));
2569 #ifdef PERL_OLD_COPY_ON_WRITE
2571 sv_force_normal_flags(sv, 0);
2574 if (SvREADONLY(sv) && !SvOK(sv)) {
2575 if (ckWARN(WARN_UNINITIALIZED))
2580 if (SvTYPE(sv) < SVt_NV) {
2581 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2582 sv_upgrade(sv, SVt_NV);
2583 #ifdef USE_LONG_DOUBLE
2585 STORE_NUMERIC_LOCAL_SET_STANDARD();
2586 PerlIO_printf(Perl_debug_log,
2587 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2588 PTR2UV(sv), SvNVX(sv));
2589 RESTORE_NUMERIC_LOCAL();
2593 STORE_NUMERIC_LOCAL_SET_STANDARD();
2594 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2595 PTR2UV(sv), SvNVX(sv));
2596 RESTORE_NUMERIC_LOCAL();
2600 else if (SvTYPE(sv) < SVt_PVNV)
2601 sv_upgrade(sv, SVt_PVNV);
2606 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2607 #ifdef NV_PRESERVES_UV
2613 /* Only set the public NV OK flag if this NV preserves the IV */
2614 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2616 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2617 : (SvIVX(sv) == I_V(SvNVX(sv))))
2623 else if (SvPOKp(sv)) {
2625 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2626 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2628 #ifdef NV_PRESERVES_UV
2629 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2630 == IS_NUMBER_IN_UV) {
2631 /* It's definitely an integer */
2632 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2634 SvNV_set(sv, Atof(SvPVX_const(sv)));
2640 SvNV_set(sv, Atof(SvPVX_const(sv)));
2641 /* Only set the public NV OK flag if this NV preserves the value in
2642 the PV at least as well as an IV/UV would.
2643 Not sure how to do this 100% reliably. */
2644 /* if that shift count is out of range then Configure's test is
2645 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2647 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2648 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2649 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2650 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2651 /* Can't use strtol etc to convert this string, so don't try.
2652 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2655 /* value has been set. It may not be precise. */
2656 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2657 /* 2s complement assumption for (UV)IV_MIN */
2658 SvNOK_on(sv); /* Integer is too negative. */
2663 if (numtype & IS_NUMBER_NEG) {
2664 SvIV_set(sv, -(IV)value);
2665 } else if (value <= (UV)IV_MAX) {
2666 SvIV_set(sv, (IV)value);
2668 SvUV_set(sv, value);
2672 if (numtype & IS_NUMBER_NOT_INT) {
2673 /* I believe that even if the original PV had decimals,
2674 they are lost beyond the limit of the FP precision.
2675 However, neither is canonical, so both only get p
2676 flags. NWC, 2000/11/25 */
2677 /* Both already have p flags, so do nothing */
2679 const NV nv = SvNVX(sv);
2680 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2681 if (SvIVX(sv) == I_V(nv)) {
2684 /* It had no "." so it must be integer. */
2688 /* between IV_MAX and NV(UV_MAX).
2689 Could be slightly > UV_MAX */
2691 if (numtype & IS_NUMBER_NOT_INT) {
2692 /* UV and NV both imprecise. */
2694 const UV nv_as_uv = U_V(nv);
2696 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2705 /* It might be more code efficient to go through the entire logic above
2706 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2707 gets complex and potentially buggy, so more programmer efficient
2708 to do it this way, by turning off the public flags: */
2710 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2711 #endif /* NV_PRESERVES_UV */
2714 if (isGV_with_GP(sv)) {
2715 glob_2number(MUTABLE_GV(sv));
2719 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2721 assert (SvTYPE(sv) >= SVt_NV);
2722 /* Typically the caller expects that sv_any is not NULL now. */
2723 /* XXX Ilya implies that this is a bug in callers that assume this
2724 and ideally should be fixed. */
2727 #if defined(USE_LONG_DOUBLE)
2729 STORE_NUMERIC_LOCAL_SET_STANDARD();
2730 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2731 PTR2UV(sv), SvNVX(sv));
2732 RESTORE_NUMERIC_LOCAL();
2736 STORE_NUMERIC_LOCAL_SET_STANDARD();
2737 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2738 PTR2UV(sv), SvNVX(sv));
2739 RESTORE_NUMERIC_LOCAL();
2748 Return an SV with the numeric value of the source SV, doing any necessary
2749 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2750 access this function.
2756 Perl_sv_2num(pTHX_ SV *const sv)
2758 PERL_ARGS_ASSERT_SV_2NUM;
2763 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2764 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2765 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2766 return sv_2num(tmpsv);
2768 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2771 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2772 * UV as a string towards the end of buf, and return pointers to start and
2775 * We assume that buf is at least TYPE_CHARS(UV) long.
2779 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2781 char *ptr = buf + TYPE_CHARS(UV);
2782 char * const ebuf = ptr;
2785 PERL_ARGS_ASSERT_UIV_2BUF;
2797 *--ptr = '0' + (char)(uv % 10);
2806 =for apidoc sv_2pv_flags
2808 Returns a pointer to the string value of an SV, and sets *lp to its length.
2809 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2810 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2811 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2817 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2822 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2824 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2825 && SvTYPE(sv) != SVt_PVFM);
2826 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2831 if (flags & SV_SKIP_OVERLOAD)
2833 tmpstr = AMG_CALLunary(sv, string_amg);
2834 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2835 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2837 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2841 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2842 if (flags & SV_CONST_RETURN) {
2843 pv = (char *) SvPVX_const(tmpstr);
2845 pv = (flags & SV_MUTABLE_RETURN)
2846 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2849 *lp = SvCUR(tmpstr);
2851 pv = sv_2pv_flags(tmpstr, lp, flags);
2864 SV *const referent = SvRV(sv);
2868 retval = buffer = savepvn("NULLREF", len);
2869 } else if (SvTYPE(referent) == SVt_REGEXP &&
2870 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2871 amagic_is_enabled(string_amg))) {
2872 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2876 /* If the regex is UTF-8 we want the containing scalar to
2877 have an UTF-8 flag too */
2884 *lp = RX_WRAPLEN(re);
2886 return RX_WRAPPED(re);
2888 const char *const typestr = sv_reftype(referent, 0);
2889 const STRLEN typelen = strlen(typestr);
2890 UV addr = PTR2UV(referent);
2891 const char *stashname = NULL;
2892 STRLEN stashnamelen = 0; /* hush, gcc */
2893 const char *buffer_end;
2895 if (SvOBJECT(referent)) {
2896 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2899 stashname = HEK_KEY(name);
2900 stashnamelen = HEK_LEN(name);
2902 if (HEK_UTF8(name)) {
2908 stashname = "__ANON__";
2911 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2912 + 2 * sizeof(UV) + 2 /* )\0 */;
2914 len = typelen + 3 /* (0x */
2915 + 2 * sizeof(UV) + 2 /* )\0 */;
2918 Newx(buffer, len, char);
2919 buffer_end = retval = buffer + len;
2921 /* Working backwards */
2925 *--retval = PL_hexdigit[addr & 15];
2926 } while (addr >>= 4);
2932 memcpy(retval, typestr, typelen);
2936 retval -= stashnamelen;
2937 memcpy(retval, stashname, stashnamelen);
2939 /* retval may not necessarily have reached the start of the
2941 assert (retval >= buffer);
2943 len = buffer_end - retval - 1; /* -1 for that \0 */
2955 if (flags & SV_MUTABLE_RETURN)
2956 return SvPVX_mutable(sv);
2957 if (flags & SV_CONST_RETURN)
2958 return (char *)SvPVX_const(sv);
2963 /* I'm assuming that if both IV and NV are equally valid then
2964 converting the IV is going to be more efficient */
2965 const U32 isUIOK = SvIsUV(sv);
2966 char buf[TYPE_CHARS(UV)];
2970 if (SvTYPE(sv) < SVt_PVIV)
2971 sv_upgrade(sv, SVt_PVIV);
2972 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2974 /* inlined from sv_setpvn */
2975 s = SvGROW_mutable(sv, len + 1);
2976 Move(ptr, s, len, char);
2981 else if (SvNOK(sv)) {
2982 if (SvTYPE(sv) < SVt_PVNV)
2983 sv_upgrade(sv, SVt_PVNV);
2984 if (SvNVX(sv) == 0.0) {
2985 s = SvGROW_mutable(sv, 2);
2990 /* The +20 is pure guesswork. Configure test needed. --jhi */
2991 s = SvGROW_mutable(sv, NV_DIG + 20);
2992 /* some Xenix systems wipe out errno here */
2994 #ifndef USE_LOCALE_NUMERIC
2995 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
2999 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3000 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3002 /* If the radix character is UTF-8, and actually is in the
3003 * output, turn on the UTF-8 flag for the scalar */
3004 if (PL_numeric_local
3005 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
3006 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3010 RESTORE_LC_NUMERIC();
3013 /* We don't call SvPOK_on(), because it may come to pass that the
3014 * locale changes so that the stringification we just did is no
3015 * longer correct. We will have to re-stringify every time it is
3022 else if (isGV_with_GP(sv)) {
3023 GV *const gv = MUTABLE_GV(sv);
3024 SV *const buffer = sv_newmortal();
3026 gv_efullname3(buffer, gv, "*");
3028 assert(SvPOK(buffer));
3032 *lp = SvCUR(buffer);
3033 return SvPVX(buffer);
3035 else if (isREGEXP(sv)) {
3036 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3037 return RX_WRAPPED((REGEXP *)sv);
3042 if (flags & SV_UNDEF_RETURNS_NULL)
3044 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3046 /* Typically the caller expects that sv_any is not NULL now. */
3047 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3048 sv_upgrade(sv, SVt_PV);
3053 const STRLEN len = s - SvPVX_const(sv);
3058 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3059 PTR2UV(sv),SvPVX_const(sv)));
3060 if (flags & SV_CONST_RETURN)
3061 return (char *)SvPVX_const(sv);
3062 if (flags & SV_MUTABLE_RETURN)
3063 return SvPVX_mutable(sv);
3068 =for apidoc sv_copypv
3070 Copies a stringified representation of the source SV into the
3071 destination SV. Automatically performs any necessary mg_get and
3072 coercion of numeric values into strings. Guaranteed to preserve
3073 UTF8 flag even from overloaded objects. Similar in nature to
3074 sv_2pv[_flags] but operates directly on an SV instead of just the
3075 string. Mostly uses sv_2pv_flags to do its work, except when that
3076 would lose the UTF-8'ness of the PV.
3078 =for apidoc sv_copypv_nomg
3080 Like sv_copypv, but doesn't invoke get magic first.
3082 =for apidoc sv_copypv_flags
3084 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3091 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3093 PERL_ARGS_ASSERT_SV_COPYPV;
3095 sv_copypv_flags(dsv, ssv, 0);
3099 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3104 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3106 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3108 s = SvPV_nomg_const(ssv,len);
3109 sv_setpvn(dsv,s,len);
3117 =for apidoc sv_2pvbyte
3119 Return a pointer to the byte-encoded representation of the SV, and set *lp
3120 to its length. May cause the SV to be downgraded from UTF-8 as a
3123 Usually accessed via the C<SvPVbyte> macro.
3129 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3131 PERL_ARGS_ASSERT_SV_2PVBYTE;
3134 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3135 || isGV_with_GP(sv) || SvROK(sv)) {
3136 SV *sv2 = sv_newmortal();
3137 sv_copypv_nomg(sv2,sv);
3140 sv_utf8_downgrade(sv,0);
3141 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3145 =for apidoc sv_2pvutf8
3147 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3148 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3150 Usually accessed via the C<SvPVutf8> macro.
3156 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3158 PERL_ARGS_ASSERT_SV_2PVUTF8;
3160 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3161 || isGV_with_GP(sv) || SvROK(sv))
3162 sv = sv_mortalcopy(sv);
3165 sv_utf8_upgrade_nomg(sv);
3166 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3171 =for apidoc sv_2bool
3173 This macro is only used by sv_true() or its macro equivalent, and only if
3174 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3175 It calls sv_2bool_flags with the SV_GMAGIC flag.
3177 =for apidoc sv_2bool_flags
3179 This function is only used by sv_true() and friends, and only if
3180 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3181 contain SV_GMAGIC, then it does an mg_get() first.
3188 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3192 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3195 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3201 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3202 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3205 if(SvGMAGICAL(sv)) {
3207 goto restart; /* call sv_2bool */
3209 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3210 else if(!SvOK(sv)) {
3213 else if(SvPOK(sv)) {
3214 svb = SvPVXtrue(sv);
3216 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3217 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3218 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3222 goto restart; /* call sv_2bool_nomg */
3227 return SvRV(sv) != 0;
3231 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3232 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3236 =for apidoc sv_utf8_upgrade
3238 Converts the PV of an SV to its UTF-8-encoded form.
3239 Forces the SV to string form if it is not already.
3240 Will C<mg_get> on C<sv> if appropriate.
3241 Always sets the SvUTF8 flag to avoid future validity checks even
3242 if the whole string is the same in UTF-8 as not.
3243 Returns the number of bytes in the converted string
3245 This is not a general purpose byte encoding to Unicode interface:
3246 use the Encode extension for that.
3248 =for apidoc sv_utf8_upgrade_nomg
3250 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3252 =for apidoc sv_utf8_upgrade_flags
3254 Converts the PV of an SV to its UTF-8-encoded form.
3255 Forces the SV to string form if it is not already.
3256 Always sets the SvUTF8 flag to avoid future validity checks even
3257 if all the bytes are invariant in UTF-8.
3258 If C<flags> has C<SV_GMAGIC> bit set,
3259 will C<mg_get> on C<sv> if appropriate, else not.
3261 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3262 will expand when converted to UTF-8, and skips the extra work of checking for
3263 that. Typically this flag is used by a routine that has already parsed the
3264 string and found such characters, and passes this information on so that the
3265 work doesn't have to be repeated.
3267 Returns the number of bytes in the converted string.
3269 This is not a general purpose byte encoding to Unicode interface:
3270 use the Encode extension for that.
3272 =for apidoc sv_utf8_upgrade_flags_grow
3274 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3275 the number of unused bytes the string of 'sv' is guaranteed to have free after
3276 it upon return. This allows the caller to reserve extra space that it intends
3277 to fill, to avoid extra grows.
3279 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3280 are implemented in terms of this function.
3282 Returns the number of bytes in the converted string (not including the spares).
3286 (One might think that the calling routine could pass in the position of the
3287 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3288 have to be found again. But that is not the case, because typically when the
3289 caller is likely to use this flag, it won't be calling this routine unless it
3290 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3291 and just use bytes. But some things that do fit into a byte are variants in
3292 utf8, and the caller may not have been keeping track of these.)
3294 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3295 C<NUL> isn't guaranteed due to having other routines do the work in some input
3296 cases, or if the input is already flagged as being in utf8.
3298 The speed of this could perhaps be improved for many cases if someone wanted to
3299 write a fast function that counts the number of variant characters in a string,
3300 especially if it could return the position of the first one.
3305 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3309 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3311 if (sv == &PL_sv_undef)
3313 if (!SvPOK_nog(sv)) {
3315 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3316 (void) sv_2pv_flags(sv,&len, flags);
3318 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3322 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3327 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3332 S_sv_uncow(aTHX_ sv, 0);
3335 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3336 sv_recode_to_utf8(sv, PL_encoding);
3337 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3341 if (SvCUR(sv) == 0) {
3342 if (extra) SvGROW(sv, extra);
3343 } else { /* Assume Latin-1/EBCDIC */
3344 /* This function could be much more efficient if we
3345 * had a FLAG in SVs to signal if there are any variant
3346 * chars in the PV. Given that there isn't such a flag
3347 * make the loop as fast as possible (although there are certainly ways
3348 * to speed this up, eg. through vectorization) */
3349 U8 * s = (U8 *) SvPVX_const(sv);
3350 U8 * e = (U8 *) SvEND(sv);
3352 STRLEN two_byte_count = 0;
3354 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3356 /* See if really will need to convert to utf8. We mustn't rely on our
3357 * incoming SV being well formed and having a trailing '\0', as certain
3358 * code in pp_formline can send us partially built SVs. */
3362 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3364 t--; /* t already incremented; re-point to first variant */
3369 /* utf8 conversion not needed because all are invariants. Mark as
3370 * UTF-8 even if no variant - saves scanning loop */
3372 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3377 /* Here, the string should be converted to utf8, either because of an
3378 * input flag (two_byte_count = 0), or because a character that
3379 * requires 2 bytes was found (two_byte_count = 1). t points either to
3380 * the beginning of the string (if we didn't examine anything), or to
3381 * the first variant. In either case, everything from s to t - 1 will
3382 * occupy only 1 byte each on output.
3384 * There are two main ways to convert. One is to create a new string
3385 * and go through the input starting from the beginning, appending each
3386 * converted value onto the new string as we go along. It's probably
3387 * best to allocate enough space in the string for the worst possible
3388 * case rather than possibly running out of space and having to
3389 * reallocate and then copy what we've done so far. Since everything
3390 * from s to t - 1 is invariant, the destination can be initialized
3391 * with these using a fast memory copy
3393 * The other way is to figure out exactly how big the string should be
3394 * by parsing the entire input. Then you don't have to make it big
3395 * enough to handle the worst possible case, and more importantly, if
3396 * the string you already have is large enough, you don't have to
3397 * allocate a new string, you can copy the last character in the input
3398 * string to the final position(s) that will be occupied by the
3399 * converted string and go backwards, stopping at t, since everything
3400 * before that is invariant.
3402 * There are advantages and disadvantages to each method.
3404 * In the first method, we can allocate a new string, do the memory
3405 * copy from the s to t - 1, and then proceed through the rest of the
3406 * string byte-by-byte.
3408 * In the second method, we proceed through the rest of the input
3409 * string just calculating how big the converted string will be. Then
3410 * there are two cases:
3411 * 1) if the string has enough extra space to handle the converted
3412 * value. We go backwards through the string, converting until we
3413 * get to the position we are at now, and then stop. If this
3414 * position is far enough along in the string, this method is
3415 * faster than the other method. If the memory copy were the same
3416 * speed as the byte-by-byte loop, that position would be about
3417 * half-way, as at the half-way mark, parsing to the end and back
3418 * is one complete string's parse, the same amount as starting
3419 * over and going all the way through. Actually, it would be
3420 * somewhat less than half-way, as it's faster to just count bytes
3421 * than to also copy, and we don't have the overhead of allocating
3422 * a new string, changing the scalar to use it, and freeing the
3423 * existing one. But if the memory copy is fast, the break-even
3424 * point is somewhere after half way. The counting loop could be
3425 * sped up by vectorization, etc, to move the break-even point
3426 * further towards the beginning.
3427 * 2) if the string doesn't have enough space to handle the converted
3428 * value. A new string will have to be allocated, and one might
3429 * as well, given that, start from the beginning doing the first
3430 * method. We've spent extra time parsing the string and in
3431 * exchange all we've gotten is that we know precisely how big to
3432 * make the new one. Perl is more optimized for time than space,
3433 * so this case is a loser.
3434 * So what I've decided to do is not use the 2nd method unless it is
3435 * guaranteed that a new string won't have to be allocated, assuming
3436 * the worst case. I also decided not to put any more conditions on it
3437 * than this, for now. It seems likely that, since the worst case is
3438 * twice as big as the unknown portion of the string (plus 1), we won't
3439 * be guaranteed enough space, causing us to go to the first method,
3440 * unless the string is short, or the first variant character is near
3441 * the end of it. In either of these cases, it seems best to use the
3442 * 2nd method. The only circumstance I can think of where this would
3443 * be really slower is if the string had once had much more data in it
3444 * than it does now, but there is still a substantial amount in it */
3447 STRLEN invariant_head = t - s;
3448 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3449 if (SvLEN(sv) < size) {
3451 /* Here, have decided to allocate a new string */
3456 Newx(dst, size, U8);
3458 /* If no known invariants at the beginning of the input string,
3459 * set so starts from there. Otherwise, can use memory copy to
3460 * get up to where we are now, and then start from here */
3462 if (invariant_head <= 0) {
3465 Copy(s, dst, invariant_head, char);
3466 d = dst + invariant_head;
3470 append_utf8_from_native_byte(*t, &d);
3474 SvPV_free(sv); /* No longer using pre-existing string */
3475 SvPV_set(sv, (char*)dst);
3476 SvCUR_set(sv, d - dst);
3477 SvLEN_set(sv, size);
3480 /* Here, have decided to get the exact size of the string.
3481 * Currently this happens only when we know that there is
3482 * guaranteed enough space to fit the converted string, so
3483 * don't have to worry about growing. If two_byte_count is 0,
3484 * then t points to the first byte of the string which hasn't
3485 * been examined yet. Otherwise two_byte_count is 1, and t
3486 * points to the first byte in the string that will expand to
3487 * two. Depending on this, start examining at t or 1 after t.
3490 U8 *d = t + two_byte_count;
3493 /* Count up the remaining bytes that expand to two */
3496 const U8 chr = *d++;
3497 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3500 /* The string will expand by just the number of bytes that
3501 * occupy two positions. But we are one afterwards because of
3502 * the increment just above. This is the place to put the
3503 * trailing NUL, and to set the length before we decrement */
3505 d += two_byte_count;
3506 SvCUR_set(sv, d - s);
3510 /* Having decremented d, it points to the position to put the
3511 * very last byte of the expanded string. Go backwards through
3512 * the string, copying and expanding as we go, stopping when we
3513 * get to the part that is invariant the rest of the way down */
3517 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3520 *d-- = UTF8_EIGHT_BIT_LO(*e);
3521 *d-- = UTF8_EIGHT_BIT_HI(*e);
3527 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3528 /* Update pos. We do it at the end rather than during
3529 * the upgrade, to avoid slowing down the common case
3530 * (upgrade without pos).
3531 * pos can be stored as either bytes or characters. Since
3532 * this was previously a byte string we can just turn off
3533 * the bytes flag. */
3534 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3536 mg->mg_flags &= ~MGf_BYTES;
3538 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3539 magic_setutf8(sv,mg); /* clear UTF8 cache */
3544 /* Mark as UTF-8 even if no variant - saves scanning loop */
3550 =for apidoc sv_utf8_downgrade
3552 Attempts to convert the PV of an SV from characters to bytes.
3553 If the PV contains a character that cannot fit
3554 in a byte, this conversion will fail;
3555 in this case, either returns false or, if C<fail_ok> is not
3558 This is not a general purpose Unicode to byte encoding interface:
3559 use the Encode extension for that.
3565 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3569 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3571 if (SvPOKp(sv) && SvUTF8(sv)) {
3575 int mg_flags = SV_GMAGIC;
3578 S_sv_uncow(aTHX_ sv, 0);
3580 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3582 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3583 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3584 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3585 SV_GMAGIC|SV_CONST_RETURN);
3586 mg_flags = 0; /* sv_pos_b2u does get magic */
3588 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3589 magic_setutf8(sv,mg); /* clear UTF8 cache */
3592 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3594 if (!utf8_to_bytes(s, &len)) {
3599 Perl_croak(aTHX_ "Wide character in %s",
3602 Perl_croak(aTHX_ "Wide character");
3613 =for apidoc sv_utf8_encode
3615 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3616 flag off so that it looks like octets again.
3622 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3624 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3626 if (SvREADONLY(sv)) {
3627 sv_force_normal_flags(sv, 0);
3629 (void) sv_utf8_upgrade(sv);
3634 =for apidoc sv_utf8_decode
3636 If the PV of the SV is an octet sequence in UTF-8
3637 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3638 so that it looks like a character. If the PV contains only single-byte
3639 characters, the C<SvUTF8> flag stays off.
3640 Scans PV for validity and returns false if the PV is invalid UTF-8.
3646 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3648 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3651 const U8 *start, *c;
3654 /* The octets may have got themselves encoded - get them back as
3657 if (!sv_utf8_downgrade(sv, TRUE))
3660 /* it is actually just a matter of turning the utf8 flag on, but
3661 * we want to make sure everything inside is valid utf8 first.
3663 c = start = (const U8 *) SvPVX_const(sv);
3664 if (!is_utf8_string(c, SvCUR(sv)))
3666 e = (const U8 *) SvEND(sv);
3669 if (!UTF8_IS_INVARIANT(ch)) {
3674 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3675 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3676 after this, clearing pos. Does anything on CPAN
3678 /* adjust pos to the start of a UTF8 char sequence */
3679 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3681 I32 pos = mg->mg_len;
3683 for (c = start + pos; c > start; c--) {
3684 if (UTF8_IS_START(*c))
3687 mg->mg_len = c - start;
3690 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3691 magic_setutf8(sv,mg); /* clear UTF8 cache */
3698 =for apidoc sv_setsv
3700 Copies the contents of the source SV C<ssv> into the destination SV
3701 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3702 function if the source SV needs to be reused. Does not handle 'set' magic on
3703 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3704 performs a copy-by-value, obliterating any previous content of the
3707 You probably want to use one of the assortment of wrappers, such as
3708 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3709 C<SvSetMagicSV_nosteal>.
3711 =for apidoc sv_setsv_flags
3713 Copies the contents of the source SV C<ssv> into the destination SV
3714 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3715 function if the source SV needs to be reused. Does not handle 'set' magic.
3716 Loosely speaking, it performs a copy-by-value, obliterating any previous
3717 content of the destination.
3718 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3719 C<ssv> if appropriate, else not. If the C<flags>
3720 parameter has the C<SV_NOSTEAL> bit set then the
3721 buffers of temps will not be stolen. <sv_setsv>
3722 and C<sv_setsv_nomg> are implemented in terms of this function.
3724 You probably want to use one of the assortment of wrappers, such as
3725 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3726 C<SvSetMagicSV_nosteal>.
3728 This is the primary function for copying scalars, and most other
3729 copy-ish functions and macros use this underneath.
3735 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3737 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3738 HV *old_stash = NULL;
3740 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3742 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3743 const char * const name = GvNAME(sstr);
3744 const STRLEN len = GvNAMELEN(sstr);
3746 if (dtype >= SVt_PV) {
3752 SvUPGRADE(dstr, SVt_PVGV);
3753 (void)SvOK_off(dstr);
3754 isGV_with_GP_on(dstr);
3756 GvSTASH(dstr) = GvSTASH(sstr);
3758 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3759 gv_name_set(MUTABLE_GV(dstr), name, len,
3760 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3761 SvFAKE_on(dstr); /* can coerce to non-glob */
3764 if(GvGP(MUTABLE_GV(sstr))) {
3765 /* If source has method cache entry, clear it */
3767 SvREFCNT_dec(GvCV(sstr));
3768 GvCV_set(sstr, NULL);
3771 /* If source has a real method, then a method is
3774 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3780 /* If dest already had a real method, that's a change as well */
3782 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3783 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3788 /* We don't need to check the name of the destination if it was not a
3789 glob to begin with. */
3790 if(dtype == SVt_PVGV) {
3791 const char * const name = GvNAME((const GV *)dstr);
3794 /* The stash may have been detached from the symbol table, so
3796 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3800 const STRLEN len = GvNAMELEN(dstr);
3801 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3802 || (len == 1 && name[0] == ':')) {
3805 /* Set aside the old stash, so we can reset isa caches on
3807 if((old_stash = GvHV(dstr)))
3808 /* Make sure we do not lose it early. */
3809 SvREFCNT_inc_simple_void_NN(
3810 sv_2mortal((SV *)old_stash)
3815 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3818 gp_free(MUTABLE_GV(dstr));
3819 GvINTRO_off(dstr); /* one-shot flag */
3820 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3821 if (SvTAINTED(sstr))
3823 if (GvIMPORTED(dstr) != GVf_IMPORTED
3824 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3826 GvIMPORTED_on(dstr);
3829 if(mro_changes == 2) {
3830 if (GvAV((const GV *)sstr)) {
3832 SV * const sref = (SV *)GvAV((const GV *)dstr);
3833 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3834 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3835 AV * const ary = newAV();
3836 av_push(ary, mg->mg_obj); /* takes the refcount */
3837 mg->mg_obj = (SV *)ary;
3839 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3841 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3843 mro_isa_changed_in(GvSTASH(dstr));
3845 else if(mro_changes == 3) {
3846 HV * const stash = GvHV(dstr);
3847 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3853 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3854 if (GvIO(dstr) && dtype == SVt_PVGV) {
3855 DEBUG_o(Perl_deb(aTHX_
3856 "glob_assign_glob clearing PL_stashcache\n"));
3857 /* It's a cache. It will rebuild itself quite happily.
3858 It's a lot of effort to work out exactly which key (or keys)
3859 might be invalidated by the creation of the this file handle.
3861 hv_clear(PL_stashcache);
3867 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3869 SV * const sref = SvRV(sstr);
3871 const int intro = GvINTRO(dstr);
3874 const U32 stype = SvTYPE(sref);
3876 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3879 GvINTRO_off(dstr); /* one-shot flag */
3880 GvLINE(dstr) = CopLINE(PL_curcop);
3881 GvEGV(dstr) = MUTABLE_GV(dstr);
3886 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3887 import_flag = GVf_IMPORTED_CV;
3890 location = (SV **) &GvHV(dstr);
3891 import_flag = GVf_IMPORTED_HV;
3894 location = (SV **) &GvAV(dstr);
3895 import_flag = GVf_IMPORTED_AV;
3898 location = (SV **) &GvIOp(dstr);
3901 location = (SV **) &GvFORM(dstr);
3904 location = &GvSV(dstr);
3905 import_flag = GVf_IMPORTED_SV;
3908 if (stype == SVt_PVCV) {
3909 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3910 if (GvCVGEN(dstr)) {
3911 SvREFCNT_dec(GvCV(dstr));
3912 GvCV_set(dstr, NULL);
3913 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3916 /* SAVEt_GVSLOT takes more room on the savestack and has more
3917 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3918 leave_scope needs access to the GV so it can reset method
3919 caches. We must use SAVEt_GVSLOT whenever the type is
3920 SVt_PVCV, even if the stash is anonymous, as the stash may
3921 gain a name somehow before leave_scope. */
3922 if (stype == SVt_PVCV) {
3923 /* There is no save_pushptrptrptr. Creating it for this
3924 one call site would be overkill. So inline the ss add
3928 SS_ADD_PTR(location);
3929 SS_ADD_PTR(SvREFCNT_inc(*location));
3930 SS_ADD_UV(SAVEt_GVSLOT);
3933 else SAVEGENERICSV(*location);
3936 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3937 CV* const cv = MUTABLE_CV(*location);
3939 if (!GvCVGEN((const GV *)dstr) &&
3940 (CvROOT(cv) || CvXSUB(cv)) &&
3941 /* redundant check that avoids creating the extra SV
3942 most of the time: */
3943 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3945 SV * const new_const_sv =
3946 CvCONST((const CV *)sref)
3947 ? cv_const_sv((const CV *)sref)
3949 report_redefined_cv(
3950 sv_2mortal(Perl_newSVpvf(aTHX_
3953 HvNAME_HEK(GvSTASH((const GV *)dstr))
3955 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3958 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3962 cv_ckproto_len_flags(cv, (const GV *)dstr,
3963 SvPOK(sref) ? CvPROTO(sref) : NULL,
3964 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3965 SvPOK(sref) ? SvUTF8(sref) : 0);
3967 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3968 GvASSUMECV_on(dstr);
3969 if(GvSTASH(dstr)) gv_method_changed(dstr); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3971 *location = SvREFCNT_inc_simple_NN(sref);
3972 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3973 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3974 GvFLAGS(dstr) |= import_flag;
3976 if (stype == SVt_PVHV) {
3977 const char * const name = GvNAME((GV*)dstr);
3978 const STRLEN len = GvNAMELEN(dstr);
3981 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3982 || (len == 1 && name[0] == ':')
3984 && (!dref || HvENAME_get(dref))
3987 (HV *)sref, (HV *)dref,
3993 stype == SVt_PVAV && sref != dref
3994 && strEQ(GvNAME((GV*)dstr), "ISA")
3995 /* The stash may have been detached from the symbol table, so
3996 check its name before doing anything. */
3997 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4000 MAGIC * const omg = dref && SvSMAGICAL(dref)
4001 ? mg_find(dref, PERL_MAGIC_isa)
4003 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4004 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4005 AV * const ary = newAV();
4006 av_push(ary, mg->mg_obj); /* takes the refcount */
4007 mg->mg_obj = (SV *)ary;
4010 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4011 SV **svp = AvARRAY((AV *)omg->mg_obj);
4012 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4016 SvREFCNT_inc_simple_NN(*svp++)
4022 SvREFCNT_inc_simple_NN(omg->mg_obj)
4026 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4031 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4033 mg = mg_find(sref, PERL_MAGIC_isa);
4035 /* Since the *ISA assignment could have affected more than
4036 one stash, don't call mro_isa_changed_in directly, but let
4037 magic_clearisa do it for us, as it already has the logic for
4038 dealing with globs vs arrays of globs. */
4040 Perl_magic_clearisa(aTHX_ NULL, mg);
4042 else if (stype == SVt_PVIO) {
4043 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4044 /* It's a cache. It will rebuild itself quite happily.
4045 It's a lot of effort to work out exactly which key (or keys)
4046 might be invalidated by the creation of the this file handle.
4048 hv_clear(PL_stashcache);
4052 if (!intro) SvREFCNT_dec(dref);
4053 if (SvTAINTED(sstr))
4061 #ifdef PERL_DEBUG_READONLY_COW
4062 # include <sys/mman.h>
4064 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4065 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4069 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4071 struct perl_memory_debug_header * const header =
4072 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4073 const MEM_SIZE len = header->size;
4074 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4075 # ifdef PERL_TRACK_MEMPOOL
4076 if (!header->readonly) header->readonly = 1;
4078 if (mprotect(header, len, PROT_READ))
4079 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4080 header, len, errno);
4084 S_sv_buf_to_rw(pTHX_ SV *sv)
4086 struct perl_memory_debug_header * const header =
4087 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4088 const MEM_SIZE len = header->size;
4089 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4090 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4091 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4092 header, len, errno);
4093 # ifdef PERL_TRACK_MEMPOOL
4094 header->readonly = 0;
4099 # define sv_buf_to_ro(sv) NOOP
4100 # define sv_buf_to_rw(sv) NOOP
4104 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)
4709 PERL_ARGS_ASSERT_SV_SETPVN;
4711 SV_CHECK_THINKFIRST_COW_DROP(sv);
4717 /* len is STRLEN which is unsigned, need to copy to signed */
4720 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4723 SvUPGRADE(sv, SVt_PV);
4725 dptr = SvGROW(sv, len + 1);
4726 Move(ptr,dptr,len,char);
4729 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4731 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4735 =for apidoc sv_setpvn_mg
4737 Like C<sv_setpvn>, but also handles 'set' magic.
4743 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4745 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4747 sv_setpvn(sv,ptr,len);
4752 =for apidoc sv_setpv
4754 Copies a string into an SV. The string must be terminated with a C<NUL>
4756 Does not handle 'set' magic. See C<sv_setpv_mg>.
4762 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4767 PERL_ARGS_ASSERT_SV_SETPV;
4769 SV_CHECK_THINKFIRST_COW_DROP(sv);
4775 SvUPGRADE(sv, SVt_PV);
4777 SvGROW(sv, len + 1);
4778 Move(ptr,SvPVX(sv),len+1,char);
4780 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4782 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4786 =for apidoc sv_setpv_mg
4788 Like C<sv_setpv>, but also handles 'set' magic.
4794 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4796 PERL_ARGS_ASSERT_SV_SETPV_MG;
4803 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4807 PERL_ARGS_ASSERT_SV_SETHEK;
4813 if (HEK_LEN(hek) == HEf_SVKEY) {
4814 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4817 const int flags = HEK_FLAGS(hek);
4818 if (flags & HVhek_WASUTF8) {
4819 STRLEN utf8_len = HEK_LEN(hek);
4820 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4821 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4824 } else if (flags & HVhek_UNSHARED) {
4825 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4828 else SvUTF8_off(sv);
4832 SV_CHECK_THINKFIRST_COW_DROP(sv);
4833 SvUPGRADE(sv, SVt_PV);
4835 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4836 SvCUR_set(sv, HEK_LEN(hek));
4842 else SvUTF8_off(sv);
4850 =for apidoc sv_usepvn_flags
4852 Tells an SV to use C<ptr> to find its string value. Normally the
4853 string is stored inside the SV, but sv_usepvn allows the SV to use an
4854 outside string. The C<ptr> should point to memory that was allocated
4855 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4856 the start of a Newx-ed block of memory, and not a pointer to the
4857 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4858 and not be from a non-Newx memory allocator like C<malloc>. The
4859 string length, C<len>, must be supplied. By default this function
4860 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4861 so that pointer should not be freed or used by the programmer after
4862 giving it to sv_usepvn, and neither should any pointers from "behind"
4863 that pointer (e.g. ptr + 1) be used.
4865 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4866 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
4867 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4868 C<len>, and already meets the requirements for storing in C<SvPVX>).
4874 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4879 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4881 SV_CHECK_THINKFIRST_COW_DROP(sv);
4882 SvUPGRADE(sv, SVt_PV);
4885 if (flags & SV_SMAGIC)
4889 if (SvPVX_const(sv))
4893 if (flags & SV_HAS_TRAILING_NUL)
4894 assert(ptr[len] == '\0');
4897 allocate = (flags & SV_HAS_TRAILING_NUL)
4899 #ifdef Perl_safesysmalloc_size
4902 PERL_STRLEN_ROUNDUP(len + 1);
4904 if (flags & SV_HAS_TRAILING_NUL) {
4905 /* It's long enough - do nothing.
4906 Specifically Perl_newCONSTSUB is relying on this. */
4909 /* Force a move to shake out bugs in callers. */
4910 char *new_ptr = (char*)safemalloc(allocate);
4911 Copy(ptr, new_ptr, len, char);
4912 PoisonFree(ptr,len,char);
4916 ptr = (char*) saferealloc (ptr, allocate);
4919 #ifdef Perl_safesysmalloc_size
4920 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4922 SvLEN_set(sv, allocate);
4926 if (!(flags & SV_HAS_TRAILING_NUL)) {
4929 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4931 if (flags & SV_SMAGIC)
4935 #ifdef PERL_OLD_COPY_ON_WRITE
4936 /* Need to do this *after* making the SV normal, as we need the buffer
4937 pointer to remain valid until after we've copied it. If we let go too early,
4938 another thread could invalidate it by unsharing last of the same hash key
4939 (which it can do by means other than releasing copy-on-write Svs)
4940 or by changing the other copy-on-write SVs in the loop. */
4942 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
4944 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4946 { /* this SV was SvIsCOW_normal(sv) */
4947 /* we need to find the SV pointing to us. */
4948 SV *current = SV_COW_NEXT_SV(after);
4950 if (current == sv) {
4951 /* The SV we point to points back to us (there were only two of us
4953 Hence other SV is no longer copy on write either. */
4955 sv_buf_to_rw(after);
4957 /* We need to follow the pointers around the loop. */
4959 while ((next = SV_COW_NEXT_SV(current)) != sv) {
4962 /* don't loop forever if the structure is bust, and we have
4963 a pointer into a closed loop. */
4964 assert (current != after);
4965 assert (SvPVX_const(current) == pvx);
4967 /* Make the SV before us point to the SV after us. */
4968 SV_COW_NEXT_SV_SET(current, after);
4974 =for apidoc sv_force_normal_flags
4976 Undo various types of fakery on an SV, where fakery means
4977 "more than" a string: if the PV is a shared string, make
4978 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
4979 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
4980 we do the copy, and is also used locally; if this is a
4981 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
4982 then a copy-on-write scalar drops its PV buffer (if any) and becomes
4983 SvPOK_off rather than making a copy. (Used where this
4984 scalar is about to be set to some other value.) In addition,
4985 the C<flags> parameter gets passed to C<sv_unref_flags()>
4986 when unreffing. C<sv_force_normal> calls this function
4987 with flags set to 0.
4989 This function is expected to be used to signal to perl that this SV is
4990 about to be written to, and any extra book-keeping needs to be taken care
4991 of. Hence, it croaks on read-only values.
4997 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5001 assert(SvIsCOW(sv));
5004 const char * const pvx = SvPVX_const(sv);
5005 const STRLEN len = SvLEN(sv);
5006 const STRLEN cur = SvCUR(sv);
5007 # ifdef PERL_OLD_COPY_ON_WRITE
5008 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5009 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5010 we'll fail an assertion. */
5011 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5015 PerlIO_printf(Perl_debug_log,
5016 "Copy on write: Force normal %ld\n",
5021 # ifdef PERL_NEW_COPY_ON_WRITE
5022 if (len && CowREFCNT(sv) == 0)
5023 /* We own the buffer ourselves. */
5029 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5030 # ifdef PERL_NEW_COPY_ON_WRITE
5031 /* Must do this first, since the macro uses SvPVX. */
5041 if (flags & SV_COW_DROP_PV) {
5042 /* OK, so we don't need to copy our buffer. */
5045 SvGROW(sv, cur + 1);
5046 Move(pvx,SvPVX(sv),cur,char);
5051 # ifdef PERL_OLD_COPY_ON_WRITE
5052 sv_release_COW(sv, pvx, next);
5055 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5062 const char * const pvx = SvPVX_const(sv);
5063 const STRLEN len = SvCUR(sv);
5067 if (flags & SV_COW_DROP_PV) {
5068 /* OK, so we don't need to copy our buffer. */
5071 SvGROW(sv, len + 1);
5072 Move(pvx,SvPVX(sv),len,char);
5075 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5081 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5083 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5086 Perl_croak_no_modify();
5087 else if (SvIsCOW(sv))
5088 S_sv_uncow(aTHX_ sv, flags);
5090 sv_unref_flags(sv, flags);
5091 else if (SvFAKE(sv) && isGV_with_GP(sv))
5092 sv_unglob(sv, flags);
5093 else if (SvFAKE(sv) && isREGEXP(sv)) {
5094 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5095 to sv_unglob. We only need it here, so inline it. */
5096 const bool islv = SvTYPE(sv) == SVt_PVLV;
5097 const svtype new_type =
5098 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5099 SV *const temp = newSV_type(new_type);
5100 regexp *const temp_p = ReANY((REGEXP *)sv);
5102 if (new_type == SVt_PVMG) {
5103 SvMAGIC_set(temp, SvMAGIC(sv));
5104 SvMAGIC_set(sv, NULL);
5105 SvSTASH_set(temp, SvSTASH(sv));
5106 SvSTASH_set(sv, NULL);
5108 if (!islv) SvCUR_set(temp, SvCUR(sv));
5109 /* Remember that SvPVX is in the head, not the body. But
5110 RX_WRAPPED is in the body. */
5111 assert(ReANY((REGEXP *)sv)->mother_re);
5112 /* Their buffer is already owned by someone else. */
5113 if (flags & SV_COW_DROP_PV) {
5114 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5115 zeroed body. For SVt_PVLV, it should have been set to 0
5116 before turning into a regexp. */
5117 assert(!SvLEN(islv ? sv : temp));
5118 sv->sv_u.svu_pv = 0;
5121 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5122 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5126 /* Now swap the rest of the bodies. */
5130 SvFLAGS(sv) &= ~SVTYPEMASK;
5131 SvFLAGS(sv) |= new_type;
5132 SvANY(sv) = SvANY(temp);
5135 SvFLAGS(temp) &= ~(SVTYPEMASK);
5136 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5137 SvANY(temp) = temp_p;
5138 temp->sv_u.svu_rx = (regexp *)temp_p;
5140 SvREFCNT_dec_NN(temp);
5142 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5148 Efficient removal of characters from the beginning of the string buffer.
5149 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5150 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5151 character of the adjusted string. Uses the "OOK hack". On return, only
5152 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5154 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5155 refer to the same chunk of data.
5157 The unfortunate similarity of this function's name to that of Perl's C<chop>
5158 operator is strictly coincidental. This function works from the left;
5159 C<chop> works from the right.
5165 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5176 PERL_ARGS_ASSERT_SV_CHOP;
5178 if (!ptr || !SvPOKp(sv))
5180 delta = ptr - SvPVX_const(sv);
5182 /* Nothing to do. */
5185 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5186 if (delta > max_delta)
5187 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5188 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5189 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5190 SV_CHECK_THINKFIRST(sv);
5191 SvPOK_only_UTF8(sv);
5194 if (!SvLEN(sv)) { /* make copy of shared string */
5195 const char *pvx = SvPVX_const(sv);
5196 const STRLEN len = SvCUR(sv);
5197 SvGROW(sv, len + 1);
5198 Move(pvx,SvPVX(sv),len,char);
5204 SvOOK_offset(sv, old_delta);
5206 SvLEN_set(sv, SvLEN(sv) - delta);
5207 SvCUR_set(sv, SvCUR(sv) - delta);
5208 SvPV_set(sv, SvPVX(sv) + delta);
5210 p = (U8 *)SvPVX_const(sv);
5213 /* how many bytes were evacuated? we will fill them with sentinel
5214 bytes, except for the part holding the new offset of course. */
5217 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5219 assert(evacn <= delta + old_delta);
5223 /* This sets 'delta' to the accumulated value of all deltas so far */
5227 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5228 * the string; otherwise store a 0 byte there and store 'delta' just prior
5229 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5230 * portion of the chopped part of the string */
5231 if (delta < 0x100) {
5235 p -= sizeof(STRLEN);
5236 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5240 /* Fill the preceding buffer with sentinals to verify that no-one is
5250 =for apidoc sv_catpvn
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 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5257 =for apidoc sv_catpvn_flags
5259 Concatenates the string onto the end of the string which is in the SV. The
5260 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5261 status set, then the bytes appended should be valid UTF-8.
5262 If C<flags> has the C<SV_SMAGIC> bit set, will
5263 C<mg_set> on C<dsv> afterwards if appropriate.
5264 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5265 in terms of this function.
5271 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5275 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5277 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5278 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5280 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5281 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5282 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5285 else SvGROW(dsv, dlen + slen + 1);
5287 sstr = SvPVX_const(dsv);
5288 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5289 SvCUR_set(dsv, SvCUR(dsv) + slen);
5292 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5293 const char * const send = sstr + slen;
5296 /* Something this code does not account for, which I think is
5297 impossible; it would require the same pv to be treated as
5298 bytes *and* utf8, which would indicate a bug elsewhere. */
5299 assert(sstr != dstr);
5301 SvGROW(dsv, dlen + slen * 2 + 1);
5302 d = (U8 *)SvPVX(dsv) + dlen;
5304 while (sstr < send) {
5305 append_utf8_from_native_byte(*sstr, &d);
5308 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5311 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5313 if (flags & SV_SMAGIC)
5318 =for apidoc sv_catsv
5320 Concatenates the string from SV C<ssv> onto the end of the string in SV
5321 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5322 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5325 =for apidoc sv_catsv_flags
5327 Concatenates the string from SV C<ssv> onto the end of the string in SV
5328 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5329 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5330 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5331 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5332 and C<sv_catsv_mg> are implemented in terms of this function.
5337 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5341 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5345 const char *spv = SvPV_flags_const(ssv, slen, flags);
5347 if (flags & SV_GMAGIC)
5349 sv_catpvn_flags(dsv, spv, slen,
5350 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5351 if (flags & SV_SMAGIC)
5358 =for apidoc sv_catpv
5360 Concatenates the C<NUL>-terminated string onto the end of the string which is
5362 If the SV has the UTF-8 status set, then the bytes appended should be
5363 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5368 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5375 PERL_ARGS_ASSERT_SV_CATPV;
5379 junk = SvPV_force(sv, tlen);
5381 SvGROW(sv, tlen + len + 1);
5383 ptr = SvPVX_const(sv);
5384 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5385 SvCUR_set(sv, SvCUR(sv) + len);
5386 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5391 =for apidoc sv_catpv_flags
5393 Concatenates the C<NUL>-terminated string onto the end of the string which is
5395 If the SV has the UTF-8 status set, then the bytes appended should
5396 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5397 on the modified SV if appropriate.
5403 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5405 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5406 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5410 =for apidoc sv_catpv_mg
5412 Like C<sv_catpv>, but also handles 'set' magic.
5418 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5420 PERL_ARGS_ASSERT_SV_CATPV_MG;
5429 Creates a new SV. A non-zero C<len> parameter indicates the number of
5430 bytes of preallocated string space the SV should have. An extra byte for a
5431 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5432 space is allocated.) The reference count for the new SV is set to 1.
5434 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5435 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5436 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5437 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5438 modules supporting older perls.
5444 Perl_newSV(pTHX_ const STRLEN len)
5451 sv_upgrade(sv, SVt_PV);
5452 SvGROW(sv, len + 1);
5457 =for apidoc sv_magicext
5459 Adds magic to an SV, upgrading it if necessary. Applies the
5460 supplied vtable and returns a pointer to the magic added.
5462 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5463 In particular, you can add magic to SvREADONLY SVs, and add more than
5464 one instance of the same 'how'.
5466 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5467 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5468 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5469 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5471 (This is now used as a subroutine by C<sv_magic>.)
5476 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5477 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5482 PERL_ARGS_ASSERT_SV_MAGICEXT;
5484 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5486 SvUPGRADE(sv, SVt_PVMG);
5487 Newxz(mg, 1, MAGIC);
5488 mg->mg_moremagic = SvMAGIC(sv);
5489 SvMAGIC_set(sv, mg);
5491 /* Sometimes a magic contains a reference loop, where the sv and
5492 object refer to each other. To prevent a reference loop that
5493 would prevent such objects being freed, we look for such loops
5494 and if we find one we avoid incrementing the object refcount.
5496 Note we cannot do this to avoid self-tie loops as intervening RV must
5497 have its REFCNT incremented to keep it in existence.
5500 if (!obj || obj == sv ||
5501 how == PERL_MAGIC_arylen ||
5502 how == PERL_MAGIC_symtab ||
5503 (SvTYPE(obj) == SVt_PVGV &&
5504 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5505 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5506 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5511 mg->mg_obj = SvREFCNT_inc_simple(obj);
5512 mg->mg_flags |= MGf_REFCOUNTED;
5515 /* Normal self-ties simply pass a null object, and instead of
5516 using mg_obj directly, use the SvTIED_obj macro to produce a
5517 new RV as needed. For glob "self-ties", we are tieing the PVIO
5518 with an RV obj pointing to the glob containing the PVIO. In
5519 this case, to avoid a reference loop, we need to weaken the
5523 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5524 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5530 mg->mg_len = namlen;
5533 mg->mg_ptr = savepvn(name, namlen);
5534 else if (namlen == HEf_SVKEY) {
5535 /* Yes, this is casting away const. This is only for the case of
5536 HEf_SVKEY. I think we need to document this aberation of the
5537 constness of the API, rather than making name non-const, as
5538 that change propagating outwards a long way. */
5539 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5541 mg->mg_ptr = (char *) name;
5543 mg->mg_virtual = (MGVTBL *) vtable;
5550 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5552 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5553 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5554 /* This sv is only a delegate. //g magic must be attached to
5559 #ifdef PERL_OLD_COPY_ON_WRITE
5561 sv_force_normal_flags(sv, 0);
5563 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5564 &PL_vtbl_mglob, 0, 0);
5568 =for apidoc sv_magic
5570 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5571 necessary, then adds a new magic item of type C<how> to the head of the
5574 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5575 handling of the C<name> and C<namlen> arguments.
5577 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5578 to add more than one instance of the same 'how'.
5584 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5585 const char *const name, const I32 namlen)
5588 const MGVTBL *vtable;
5591 unsigned int vtable_index;
5593 PERL_ARGS_ASSERT_SV_MAGIC;
5595 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5596 || ((flags = PL_magic_data[how]),
5597 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5598 > magic_vtable_max))
5599 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5601 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5602 Useful for attaching extension internal data to perl vars.
5603 Note that multiple extensions may clash if magical scalars
5604 etc holding private data from one are passed to another. */
5606 vtable = (vtable_index == magic_vtable_max)
5607 ? NULL : PL_magic_vtables + vtable_index;
5609 #ifdef PERL_OLD_COPY_ON_WRITE
5611 sv_force_normal_flags(sv, 0);
5613 if (SvREADONLY(sv)) {
5615 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5618 Perl_croak_no_modify();
5621 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5622 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5623 /* sv_magic() refuses to add a magic of the same 'how' as an
5626 if (how == PERL_MAGIC_taint)
5632 /* Force pos to be stored as characters, not bytes. */
5633 if (SvMAGICAL(sv) && DO_UTF8(sv)
5634 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5636 && mg->mg_flags & MGf_BYTES) {
5637 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5639 mg->mg_flags &= ~MGf_BYTES;
5642 /* Rest of work is done else where */
5643 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5646 case PERL_MAGIC_taint:
5649 case PERL_MAGIC_ext:
5650 case PERL_MAGIC_dbfile:
5657 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5664 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5666 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5667 for (mg = *mgp; mg; mg = *mgp) {
5668 const MGVTBL* const virt = mg->mg_virtual;
5669 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5670 *mgp = mg->mg_moremagic;
5671 if (virt && virt->svt_free)
5672 virt->svt_free(aTHX_ sv, mg);
5673 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5675 Safefree(mg->mg_ptr);
5676 else if (mg->mg_len == HEf_SVKEY)
5677 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5678 else if (mg->mg_type == PERL_MAGIC_utf8)
5679 Safefree(mg->mg_ptr);
5681 if (mg->mg_flags & MGf_REFCOUNTED)
5682 SvREFCNT_dec(mg->mg_obj);
5686 mgp = &mg->mg_moremagic;
5689 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5690 mg_magical(sv); /* else fix the flags now */
5694 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5700 =for apidoc sv_unmagic
5702 Removes all magic of type C<type> from an SV.
5708 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5710 PERL_ARGS_ASSERT_SV_UNMAGIC;
5711 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5715 =for apidoc sv_unmagicext
5717 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5723 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5725 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5726 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5730 =for apidoc sv_rvweaken
5732 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5733 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5734 push a back-reference to this RV onto the array of backreferences
5735 associated with that magic. If the RV is magical, set magic will be
5736 called after the RV is cleared.
5742 Perl_sv_rvweaken(pTHX_ SV *const sv)
5746 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5748 if (!SvOK(sv)) /* let undefs pass */
5751 Perl_croak(aTHX_ "Can't weaken a nonreference");
5752 else if (SvWEAKREF(sv)) {
5753 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5756 else if (SvREADONLY(sv)) croak_no_modify();
5758 Perl_sv_add_backref(aTHX_ tsv, sv);
5760 SvREFCNT_dec_NN(tsv);
5764 /* Give tsv backref magic if it hasn't already got it, then push a
5765 * back-reference to sv onto the array associated with the backref magic.
5767 * As an optimisation, if there's only one backref and it's not an AV,
5768 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5769 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5773 /* A discussion about the backreferences array and its refcount:
5775 * The AV holding the backreferences is pointed to either as the mg_obj of
5776 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5777 * xhv_backreferences field. The array is created with a refcount
5778 * of 2. This means that if during global destruction the array gets
5779 * picked on before its parent to have its refcount decremented by the
5780 * random zapper, it won't actually be freed, meaning it's still there for
5781 * when its parent gets freed.
5783 * When the parent SV is freed, the extra ref is killed by
5784 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5785 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5787 * When a single backref SV is stored directly, it is not reference
5792 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5799 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5801 /* find slot to store array or singleton backref */
5803 if (SvTYPE(tsv) == SVt_PVHV) {
5804 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5807 mg = mg_find(tsv, PERL_MAGIC_backref);
5809 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5810 svp = &(mg->mg_obj);
5813 /* create or retrieve the array */
5815 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5816 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5820 mg->mg_flags |= MGf_REFCOUNTED;
5823 SvREFCNT_inc_simple_void_NN(av);
5824 /* av now has a refcnt of 2; see discussion above */
5825 av_extend(av, *svp ? 2 : 1);
5827 /* move single existing backref to the array */
5828 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5833 av = MUTABLE_AV(*svp);
5835 /* optimisation: store single backref directly in HvAUX or mg_obj */
5839 assert(SvTYPE(av) == SVt_PVAV);
5840 if (AvFILLp(av) >= AvMAX(av)) {
5841 av_extend(av, AvFILLp(av)+1);
5844 /* push new backref */
5845 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5848 /* delete a back-reference to ourselves from the backref magic associated
5849 * with the SV we point to.
5853 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5858 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5860 if (SvTYPE(tsv) == SVt_PVHV) {
5862 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5864 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5865 /* It's possible for the the last (strong) reference to tsv to have
5866 become freed *before* the last thing holding a weak reference.
5867 If both survive longer than the backreferences array, then when
5868 the referent's reference count drops to 0 and it is freed, it's
5869 not able to chase the backreferences, so they aren't NULLed.
5871 For example, a CV holds a weak reference to its stash. If both the
5872 CV and the stash survive longer than the backreferences array,
5873 and the CV gets picked for the SvBREAK() treatment first,
5874 *and* it turns out that the stash is only being kept alive because
5875 of an our variable in the pad of the CV, then midway during CV
5876 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5877 It ends up pointing to the freed HV. Hence it's chased in here, and
5878 if this block wasn't here, it would hit the !svp panic just below.
5880 I don't believe that "better" destruction ordering is going to help
5881 here - during global destruction there's always going to be the
5882 chance that something goes out of order. We've tried to make it
5883 foolproof before, and it only resulted in evolutionary pressure on
5884 fools. Which made us look foolish for our hubris. :-(
5890 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5891 svp = mg ? &(mg->mg_obj) : NULL;
5895 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5897 /* It's possible that sv is being freed recursively part way through the
5898 freeing of tsv. If this happens, the backreferences array of tsv has
5899 already been freed, and so svp will be NULL. If this is the case,
5900 we should not panic. Instead, nothing needs doing, so return. */
5901 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
5903 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
5904 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
5907 if (SvTYPE(*svp) == SVt_PVAV) {
5911 AV * const av = (AV*)*svp;
5913 assert(!SvIS_FREED(av));
5917 /* for an SV with N weak references to it, if all those
5918 * weak refs are deleted, then sv_del_backref will be called
5919 * N times and O(N^2) compares will be done within the backref
5920 * array. To ameliorate this potential slowness, we:
5921 * 1) make sure this code is as tight as possible;
5922 * 2) when looking for SV, look for it at both the head and tail of the
5923 * array first before searching the rest, since some create/destroy
5924 * patterns will cause the backrefs to be freed in order.
5931 SV **p = &svp[fill];
5932 SV *const topsv = *p;
5939 /* We weren't the last entry.
5940 An unordered list has this property that you
5941 can take the last element off the end to fill
5942 the hole, and it's still an unordered list :-)
5948 break; /* should only be one */
5955 AvFILLp(av) = fill-1;
5957 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
5958 /* freed AV; skip */
5961 /* optimisation: only a single backref, stored directly */
5963 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
5964 (void*)*svp, (void*)sv);
5971 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
5977 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
5982 /* after multiple passes through Perl_sv_clean_all() for a thingy
5983 * that has badly leaked, the backref array may have gotten freed,
5984 * since we only protect it against 1 round of cleanup */
5985 if (SvIS_FREED(av)) {
5986 if (PL_in_clean_all) /* All is fair */
5989 "panic: magic_killbackrefs (freed backref AV/SV)");
5993 is_array = (SvTYPE(av) == SVt_PVAV);
5995 assert(!SvIS_FREED(av));
5998 last = svp + AvFILLp(av);
6001 /* optimisation: only a single backref, stored directly */
6007 while (svp <= last) {
6009 SV *const referrer = *svp;
6010 if (SvWEAKREF(referrer)) {
6011 /* XXX Should we check that it hasn't changed? */
6012 assert(SvROK(referrer));
6013 SvRV_set(referrer, 0);
6015 SvWEAKREF_off(referrer);
6016 SvSETMAGIC(referrer);
6017 } else if (SvTYPE(referrer) == SVt_PVGV ||
6018 SvTYPE(referrer) == SVt_PVLV) {
6019 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6020 /* You lookin' at me? */
6021 assert(GvSTASH(referrer));
6022 assert(GvSTASH(referrer) == (const HV *)sv);
6023 GvSTASH(referrer) = 0;
6024 } else if (SvTYPE(referrer) == SVt_PVCV ||
6025 SvTYPE(referrer) == SVt_PVFM) {
6026 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6027 /* You lookin' at me? */
6028 assert(CvSTASH(referrer));
6029 assert(CvSTASH(referrer) == (const HV *)sv);
6030 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6033 assert(SvTYPE(sv) == SVt_PVGV);
6034 /* You lookin' at me? */
6035 assert(CvGV(referrer));
6036 assert(CvGV(referrer) == (const GV *)sv);
6037 anonymise_cv_maybe(MUTABLE_GV(sv),
6038 MUTABLE_CV(referrer));
6043 "panic: magic_killbackrefs (flags=%"UVxf")",
6044 (UV)SvFLAGS(referrer));
6055 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6061 =for apidoc sv_insert
6063 Inserts a string at the specified offset/length within the SV. Similar to
6064 the Perl substr() function. Handles get magic.
6066 =for apidoc sv_insert_flags
6068 Same as C<sv_insert>, but the extra C<flags> are passed to the
6069 C<SvPV_force_flags> that applies to C<bigstr>.
6075 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6082 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6085 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6088 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6089 SvPV_force_flags(bigstr, curlen, flags);
6090 (void)SvPOK_only_UTF8(bigstr);
6091 if (offset + len > curlen) {
6092 SvGROW(bigstr, offset+len+1);
6093 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6094 SvCUR_set(bigstr, offset+len);
6098 i = littlelen - len;
6099 if (i > 0) { /* string might grow */
6100 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6101 mid = big + offset + len;
6102 midend = bigend = big + SvCUR(bigstr);
6105 while (midend > mid) /* shove everything down */
6106 *--bigend = *--midend;
6107 Move(little,big+offset,littlelen,char);
6108 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6113 Move(little,SvPVX(bigstr)+offset,len,char);
6118 big = SvPVX(bigstr);
6121 bigend = big + SvCUR(bigstr);
6123 if (midend > bigend)
6124 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6127 if (mid - big > bigend - midend) { /* faster to shorten from end */
6129 Move(little, mid, littlelen,char);
6132 i = bigend - midend;
6134 Move(midend, mid, i,char);
6138 SvCUR_set(bigstr, mid - big);
6140 else if ((i = mid - big)) { /* faster from front */
6141 midend -= littlelen;
6143 Move(big, midend - i, i, char);
6144 sv_chop(bigstr,midend-i);
6146 Move(little, mid, littlelen,char);
6148 else if (littlelen) {
6149 midend -= littlelen;
6150 sv_chop(bigstr,midend);
6151 Move(little,midend,littlelen,char);
6154 sv_chop(bigstr,midend);
6160 =for apidoc sv_replace
6162 Make the first argument a copy of the second, then delete the original.
6163 The target SV physically takes over ownership of the body of the source SV
6164 and inherits its flags; however, the target keeps any magic it owns,
6165 and any magic in the source is discarded.
6166 Note that this is a rather specialist SV copying operation; most of the
6167 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6173 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6176 const U32 refcnt = SvREFCNT(sv);
6178 PERL_ARGS_ASSERT_SV_REPLACE;
6180 SV_CHECK_THINKFIRST_COW_DROP(sv);
6181 if (SvREFCNT(nsv) != 1) {
6182 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6183 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6185 if (SvMAGICAL(sv)) {
6189 sv_upgrade(nsv, SVt_PVMG);
6190 SvMAGIC_set(nsv, SvMAGIC(sv));
6191 SvFLAGS(nsv) |= SvMAGICAL(sv);
6193 SvMAGIC_set(sv, NULL);
6197 assert(!SvREFCNT(sv));
6198 #ifdef DEBUG_LEAKING_SCALARS
6199 sv->sv_flags = nsv->sv_flags;
6200 sv->sv_any = nsv->sv_any;
6201 sv->sv_refcnt = nsv->sv_refcnt;
6202 sv->sv_u = nsv->sv_u;
6204 StructCopy(nsv,sv,SV);
6206 if(SvTYPE(sv) == SVt_IV) {
6208 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6212 #ifdef PERL_OLD_COPY_ON_WRITE
6213 if (SvIsCOW_normal(nsv)) {
6214 /* We need to follow the pointers around the loop to make the
6215 previous SV point to sv, rather than nsv. */
6218 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6221 assert(SvPVX_const(current) == SvPVX_const(nsv));
6223 /* Make the SV before us point to the SV after us. */
6225 PerlIO_printf(Perl_debug_log, "previous is\n");
6227 PerlIO_printf(Perl_debug_log,
6228 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6229 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6231 SV_COW_NEXT_SV_SET(current, sv);
6234 SvREFCNT(sv) = refcnt;
6235 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6240 /* We're about to free a GV which has a CV that refers back to us.
6241 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6245 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6250 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6253 assert(SvREFCNT(gv) == 0);
6254 assert(isGV(gv) && isGV_with_GP(gv));
6256 assert(!CvANON(cv));
6257 assert(CvGV(cv) == gv);
6258 assert(!CvNAMED(cv));
6260 /* will the CV shortly be freed by gp_free() ? */
6261 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6262 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6266 /* if not, anonymise: */
6267 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6268 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6269 : newSVpvn_flags( "__ANON__", 8, 0 );
6270 sv_catpvs(gvname, "::__ANON__");
6271 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6272 SvREFCNT_dec_NN(gvname);
6276 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6281 =for apidoc sv_clear
6283 Clear an SV: call any destructors, free up any memory used by the body,
6284 and free the body itself. The SV's head is I<not> freed, although
6285 its type is set to all 1's so that it won't inadvertently be assumed
6286 to be live during global destruction etc.
6287 This function should only be called when REFCNT is zero. Most of the time
6288 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6295 Perl_sv_clear(pTHX_ SV *const orig_sv)
6300 const struct body_details *sv_type_details;
6306 PERL_ARGS_ASSERT_SV_CLEAR;
6308 /* within this loop, sv is the SV currently being freed, and
6309 * iter_sv is the most recent AV or whatever that's being iterated
6310 * over to provide more SVs */
6316 assert(SvREFCNT(sv) == 0);
6317 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6319 if (type <= SVt_IV) {
6320 /* See the comment in sv.h about the collusion between this
6321 * early return and the overloading of the NULL slots in the
6325 SvFLAGS(sv) &= SVf_BREAK;
6326 SvFLAGS(sv) |= SVTYPEMASK;
6330 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6332 if (type >= SVt_PVMG) {
6334 if (!curse(sv, 1)) goto get_next_sv;
6335 type = SvTYPE(sv); /* destructor may have changed it */
6337 /* Free back-references before magic, in case the magic calls
6338 * Perl code that has weak references to sv. */
6339 if (type == SVt_PVHV) {
6340 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6344 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6345 SvREFCNT_dec(SvOURSTASH(sv));
6347 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6348 assert(!SvMAGICAL(sv));
6349 } else if (SvMAGIC(sv)) {
6350 /* Free back-references before other types of magic. */
6351 sv_unmagic(sv, PERL_MAGIC_backref);
6355 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6356 SvREFCNT_dec(SvSTASH(sv));
6359 /* case SVt_INVLIST: */
6362 IoIFP(sv) != PerlIO_stdin() &&
6363 IoIFP(sv) != PerlIO_stdout() &&
6364 IoIFP(sv) != PerlIO_stderr() &&
6365 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6367 io_close(MUTABLE_IO(sv), FALSE);
6369 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6370 PerlDir_close(IoDIRP(sv));
6371 IoDIRP(sv) = (DIR*)NULL;
6372 Safefree(IoTOP_NAME(sv));
6373 Safefree(IoFMT_NAME(sv));
6374 Safefree(IoBOTTOM_NAME(sv));
6375 if ((const GV *)sv == PL_statgv)
6379 /* FIXME for plugins */
6381 pregfree2((REGEXP*) sv);
6385 cv_undef(MUTABLE_CV(sv));
6386 /* If we're in a stash, we don't own a reference to it.
6387 * However it does have a back reference to us, which needs to
6389 if ((stash = CvSTASH(sv)))
6390 sv_del_backref(MUTABLE_SV(stash), sv);
6393 if (PL_last_swash_hv == (const HV *)sv) {
6394 PL_last_swash_hv = NULL;
6396 if (HvTOTALKEYS((HV*)sv) > 0) {
6398 /* this statement should match the one at the beginning of
6399 * hv_undef_flags() */
6400 if ( PL_phase != PERL_PHASE_DESTRUCT
6401 && (name = HvNAME((HV*)sv)))
6403 if (PL_stashcache) {
6404 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6406 (void)hv_deletehek(PL_stashcache,
6407 HvNAME_HEK((HV*)sv), G_DISCARD);
6409 hv_name_set((HV*)sv, NULL, 0, 0);
6412 /* save old iter_sv in unused SvSTASH field */
6413 assert(!SvOBJECT(sv));
6414 SvSTASH(sv) = (HV*)iter_sv;
6417 /* save old hash_index in unused SvMAGIC field */
6418 assert(!SvMAGICAL(sv));
6419 assert(!SvMAGIC(sv));
6420 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6423 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6424 goto get_next_sv; /* process this new sv */
6426 /* free empty hash */
6427 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6428 assert(!HvARRAY((HV*)sv));
6432 AV* av = MUTABLE_AV(sv);
6433 if (PL_comppad == av) {
6437 if (AvREAL(av) && AvFILLp(av) > -1) {
6438 next_sv = AvARRAY(av)[AvFILLp(av)--];
6439 /* save old iter_sv in top-most slot of AV,
6440 * and pray that it doesn't get wiped in the meantime */
6441 AvARRAY(av)[AvMAX(av)] = iter_sv;
6443 goto get_next_sv; /* process this new sv */
6445 Safefree(AvALLOC(av));
6450 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6451 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6452 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6453 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6455 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6456 SvREFCNT_dec(LvTARG(sv));
6457 if (isREGEXP(sv)) goto freeregexp;
6459 if (isGV_with_GP(sv)) {
6460 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6461 && HvENAME_get(stash))
6462 mro_method_changed_in(stash);
6463 gp_free(MUTABLE_GV(sv));
6465 unshare_hek(GvNAME_HEK(sv));
6466 /* If we're in a stash, we don't own a reference to it.
6467 * However it does have a back reference to us, which
6468 * needs to be cleared. */
6469 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6470 sv_del_backref(MUTABLE_SV(stash), sv);
6472 /* FIXME. There are probably more unreferenced pointers to SVs
6473 * in the interpreter struct that we should check and tidy in
6474 * a similar fashion to this: */
6475 /* See also S_sv_unglob, which does the same thing. */
6476 if ((const GV *)sv == PL_last_in_gv)
6477 PL_last_in_gv = NULL;
6478 else if ((const GV *)sv == PL_statgv)
6480 else if ((const GV *)sv == PL_stderrgv)
6488 /* Don't bother with SvOOK_off(sv); as we're only going to
6492 SvOOK_offset(sv, offset);
6493 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6494 /* Don't even bother with turning off the OOK flag. */
6499 SV * const target = SvRV(sv);
6501 sv_del_backref(target, sv);
6507 else if (SvPVX_const(sv)
6508 && !(SvTYPE(sv) == SVt_PVIO
6509 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6513 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6517 # ifdef PERL_OLD_COPY_ON_WRITE
6518 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6520 if (CowREFCNT(sv)) {
6528 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6532 # ifdef PERL_OLD_COPY_ON_WRITE
6536 Safefree(SvPVX_mutable(sv));
6540 else if (SvPVX_const(sv) && SvLEN(sv)
6541 && !(SvTYPE(sv) == SVt_PVIO
6542 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6543 Safefree(SvPVX_mutable(sv));
6544 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6545 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6555 SvFLAGS(sv) &= SVf_BREAK;
6556 SvFLAGS(sv) |= SVTYPEMASK;
6558 sv_type_details = bodies_by_type + type;
6559 if (sv_type_details->arena) {
6560 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6561 &PL_body_roots[type]);
6563 else if (sv_type_details->body_size) {
6564 safefree(SvANY(sv));
6568 /* caller is responsible for freeing the head of the original sv */
6569 if (sv != orig_sv && !SvREFCNT(sv))
6572 /* grab and free next sv, if any */
6580 else if (!iter_sv) {
6582 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6583 AV *const av = (AV*)iter_sv;
6584 if (AvFILLp(av) > -1) {
6585 sv = AvARRAY(av)[AvFILLp(av)--];
6587 else { /* no more elements of current AV to free */
6590 /* restore previous value, squirrelled away */
6591 iter_sv = AvARRAY(av)[AvMAX(av)];
6592 Safefree(AvALLOC(av));
6595 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6596 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6597 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6598 /* no more elements of current HV to free */
6601 /* Restore previous values of iter_sv and hash_index,
6602 * squirrelled away */
6603 assert(!SvOBJECT(sv));
6604 iter_sv = (SV*)SvSTASH(sv);
6605 assert(!SvMAGICAL(sv));
6606 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6608 /* perl -DA does not like rubbish in SvMAGIC. */
6612 /* free any remaining detritus from the hash struct */
6613 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6614 assert(!HvARRAY((HV*)sv));
6619 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6623 if (!SvREFCNT(sv)) {
6627 if (--(SvREFCNT(sv)))
6631 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6632 "Attempt to free temp prematurely: SV 0x%"UVxf
6633 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6637 if (SvIMMORTAL(sv)) {
6638 /* make sure SvREFCNT(sv)==0 happens very seldom */
6639 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6648 /* This routine curses the sv itself, not the object referenced by sv. So
6649 sv does not have to be ROK. */
6652 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6655 PERL_ARGS_ASSERT_CURSE;
6656 assert(SvOBJECT(sv));
6658 if (PL_defstash && /* Still have a symbol table? */
6664 stash = SvSTASH(sv);
6665 assert(SvTYPE(stash) == SVt_PVHV);
6666 if (HvNAME(stash)) {
6667 CV* destructor = NULL;
6668 assert (SvOOK(stash));
6669 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6670 if (!destructor || HvMROMETA(stash)->destroy_gen
6671 != PL_sub_generation)
6674 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6675 if (gv) destructor = GvCV(gv);
6676 if (!SvOBJECT(stash))
6679 destructor ? (HV *)destructor : ((HV *)0)+1;
6680 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6684 assert(!destructor || destructor == ((CV *)0)+1
6685 || SvTYPE(destructor) == SVt_PVCV);
6686 if (destructor && destructor != ((CV *)0)+1
6687 /* A constant subroutine can have no side effects, so
6688 don't bother calling it. */
6689 && !CvCONST(destructor)
6690 /* Don't bother calling an empty destructor or one that
6691 returns immediately. */
6692 && (CvISXSUB(destructor)
6693 || (CvSTART(destructor)
6694 && (CvSTART(destructor)->op_next->op_type
6696 && (CvSTART(destructor)->op_next->op_type
6698 || CvSTART(destructor)->op_next->op_next->op_type
6704 SV* const tmpref = newRV(sv);
6705 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6707 PUSHSTACKi(PERLSI_DESTROY);
6712 call_sv(MUTABLE_SV(destructor),
6713 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6717 if(SvREFCNT(tmpref) < 2) {
6718 /* tmpref is not kept alive! */
6720 SvRV_set(tmpref, NULL);
6723 SvREFCNT_dec_NN(tmpref);
6726 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6729 if (check_refcnt && SvREFCNT(sv)) {
6730 if (PL_in_clean_objs)
6732 "DESTROY created new reference to dead object '%"HEKf"'",
6733 HEKfARG(HvNAME_HEK(stash)));
6734 /* DESTROY gave object new lease on life */
6740 HV * const stash = SvSTASH(sv);
6741 /* Curse before freeing the stash, as freeing the stash could cause
6742 a recursive call into S_curse. */
6743 SvOBJECT_off(sv); /* Curse the object. */
6744 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6745 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6751 =for apidoc sv_newref
6753 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6760 Perl_sv_newref(pTHX_ SV *const sv)
6762 PERL_UNUSED_CONTEXT;
6771 Decrement an SV's reference count, and if it drops to zero, call
6772 C<sv_clear> to invoke destructors and free up any memory used by
6773 the body; finally, deallocate the SV's head itself.
6774 Normally called via a wrapper macro C<SvREFCNT_dec>.
6780 Perl_sv_free(pTHX_ SV *const sv)
6786 /* Private helper function for SvREFCNT_dec().
6787 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6790 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6794 PERL_ARGS_ASSERT_SV_FREE2;
6796 if (LIKELY( rc == 1 )) {
6802 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6803 "Attempt to free temp prematurely: SV 0x%"UVxf
6804 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6808 if (SvIMMORTAL(sv)) {
6809 /* make sure SvREFCNT(sv)==0 happens very seldom */
6810 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6814 if (! SvREFCNT(sv)) /* may have have been resurrected */
6819 /* handle exceptional cases */
6823 if (SvFLAGS(sv) & SVf_BREAK)
6824 /* this SV's refcnt has been artificially decremented to
6825 * trigger cleanup */
6827 if (PL_in_clean_all) /* All is fair */
6829 if (SvIMMORTAL(sv)) {
6830 /* make sure SvREFCNT(sv)==0 happens very seldom */
6831 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6834 if (ckWARN_d(WARN_INTERNAL)) {
6835 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6836 Perl_dump_sv_child(aTHX_ sv);
6838 #ifdef DEBUG_LEAKING_SCALARS
6841 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6842 if (PL_warnhook == PERL_WARNHOOK_FATAL
6843 || ckDEAD(packWARN(WARN_INTERNAL))) {
6844 /* Don't let Perl_warner cause us to escape our fate: */
6848 /* This may not return: */
6849 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6850 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6851 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6854 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6864 Returns the length of the string in the SV. Handles magic and type
6865 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6866 gives raw access to the xpv_cur slot.
6872 Perl_sv_len(pTHX_ SV *const sv)
6879 (void)SvPV_const(sv, len);
6884 =for apidoc sv_len_utf8
6886 Returns the number of characters in the string in an SV, counting wide
6887 UTF-8 bytes as a single character. Handles magic and type coercion.
6893 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6894 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6895 * (Note that the mg_len is not the length of the mg_ptr field.
6896 * This allows the cache to store the character length of the string without
6897 * needing to malloc() extra storage to attach to the mg_ptr.)
6902 Perl_sv_len_utf8(pTHX_ SV *const sv)
6908 return sv_len_utf8_nomg(sv);
6912 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6916 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6918 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6920 if (PL_utf8cache && SvUTF8(sv)) {
6922 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6924 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6925 if (mg->mg_len != -1)
6928 /* We can use the offset cache for a headstart.
6929 The longer value is stored in the first pair. */
6930 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6932 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
6936 if (PL_utf8cache < 0) {
6937 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
6938 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
6942 ulen = Perl_utf8_length(aTHX_ s, s + len);
6943 utf8_mg_len_cache_update(sv, &mg, ulen);
6947 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
6950 /* Walk forwards to find the byte corresponding to the passed in UTF-8
6953 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
6954 STRLEN *const uoffset_p, bool *const at_end)
6956 const U8 *s = start;
6957 STRLEN uoffset = *uoffset_p;
6959 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
6961 while (s < send && uoffset) {
6968 else if (s > send) {
6970 /* This is the existing behaviour. Possibly it should be a croak, as
6971 it's actually a bounds error */
6974 *uoffset_p -= uoffset;
6978 /* Given the length of the string in both bytes and UTF-8 characters, decide
6979 whether to walk forwards or backwards to find the byte corresponding to
6980 the passed in UTF-8 offset. */
6982 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
6983 STRLEN uoffset, const STRLEN uend)
6985 STRLEN backw = uend - uoffset;
6987 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
6989 if (uoffset < 2 * backw) {
6990 /* The assumption is that going forwards is twice the speed of going
6991 forward (that's where the 2 * backw comes from).
6992 (The real figure of course depends on the UTF-8 data.) */
6993 const U8 *s = start;
6995 while (s < send && uoffset--)
7005 while (UTF8_IS_CONTINUATION(*send))
7008 return send - start;
7011 /* For the string representation of the given scalar, find the byte
7012 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7013 give another position in the string, *before* the sought offset, which
7014 (which is always true, as 0, 0 is a valid pair of positions), which should
7015 help reduce the amount of linear searching.
7016 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7017 will be used to reduce the amount of linear searching. The cache will be
7018 created if necessary, and the found value offered to it for update. */
7020 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7021 const U8 *const send, STRLEN uoffset,
7022 STRLEN uoffset0, STRLEN boffset0)
7024 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7026 bool at_end = FALSE;
7028 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7030 assert (uoffset >= uoffset0);
7035 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7037 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7038 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7039 if ((*mgp)->mg_ptr) {
7040 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7041 if (cache[0] == uoffset) {
7042 /* An exact match. */
7045 if (cache[2] == uoffset) {
7046 /* An exact match. */
7050 if (cache[0] < uoffset) {
7051 /* The cache already knows part of the way. */
7052 if (cache[0] > uoffset0) {
7053 /* The cache knows more than the passed in pair */
7054 uoffset0 = cache[0];
7055 boffset0 = cache[1];
7057 if ((*mgp)->mg_len != -1) {
7058 /* And we know the end too. */
7060 + sv_pos_u2b_midway(start + boffset0, send,
7062 (*mgp)->mg_len - uoffset0);
7064 uoffset -= uoffset0;
7066 + sv_pos_u2b_forwards(start + boffset0,
7067 send, &uoffset, &at_end);
7068 uoffset += uoffset0;
7071 else if (cache[2] < uoffset) {
7072 /* We're between the two cache entries. */
7073 if (cache[2] > uoffset0) {
7074 /* and the cache knows more than the passed in pair */
7075 uoffset0 = cache[2];
7076 boffset0 = cache[3];
7080 + sv_pos_u2b_midway(start + boffset0,
7083 cache[0] - uoffset0);
7086 + sv_pos_u2b_midway(start + boffset0,
7089 cache[2] - uoffset0);
7093 else if ((*mgp)->mg_len != -1) {
7094 /* If we can take advantage of a passed in offset, do so. */
7095 /* In fact, offset0 is either 0, or less than offset, so don't
7096 need to worry about the other possibility. */
7098 + sv_pos_u2b_midway(start + boffset0, send,
7100 (*mgp)->mg_len - uoffset0);
7105 if (!found || PL_utf8cache < 0) {
7106 STRLEN real_boffset;
7107 uoffset -= uoffset0;
7108 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7109 send, &uoffset, &at_end);
7110 uoffset += uoffset0;
7112 if (found && PL_utf8cache < 0)
7113 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7115 boffset = real_boffset;
7118 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7120 utf8_mg_len_cache_update(sv, mgp, uoffset);
7122 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7129 =for apidoc sv_pos_u2b_flags
7131 Converts the offset from a count of UTF-8 chars from
7132 the start of the string, to a count of the equivalent number of bytes; if
7133 lenp is non-zero, it does the same to lenp, but this time starting from
7134 the offset, rather than from the start
7135 of the string. Handles type coercion.
7136 I<flags> is passed to C<SvPV_flags>, and usually should be
7137 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7143 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7144 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7145 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7150 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7157 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7159 start = (U8*)SvPV_flags(sv, len, flags);
7161 const U8 * const send = start + len;
7163 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7166 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7167 is 0, and *lenp is already set to that. */) {
7168 /* Convert the relative offset to absolute. */
7169 const STRLEN uoffset2 = uoffset + *lenp;
7170 const STRLEN boffset2
7171 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7172 uoffset, boffset) - boffset;
7186 =for apidoc sv_pos_u2b
7188 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7189 the start of the string, to a count of the equivalent number of bytes; if
7190 lenp is non-zero, it does the same to lenp, but this time starting from
7191 the offset, rather than from the start of the string. Handles magic and
7194 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7201 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7202 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7203 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7207 /* This function is subject to size and sign problems */
7210 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7212 PERL_ARGS_ASSERT_SV_POS_U2B;
7215 STRLEN ulen = (STRLEN)*lenp;
7216 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7217 SV_GMAGIC|SV_CONST_RETURN);
7220 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7221 SV_GMAGIC|SV_CONST_RETURN);
7226 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7229 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7230 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7233 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7234 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7235 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7239 (*mgp)->mg_len = ulen;
7242 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7243 byte length pairing. The (byte) length of the total SV is passed in too,
7244 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7245 may not have updated SvCUR, so we can't rely on reading it directly.
7247 The proffered utf8/byte length pairing isn't used if the cache already has
7248 two pairs, and swapping either for the proffered pair would increase the
7249 RMS of the intervals between known byte offsets.
7251 The cache itself consists of 4 STRLEN values
7252 0: larger UTF-8 offset
7253 1: corresponding byte offset
7254 2: smaller UTF-8 offset
7255 3: corresponding byte offset
7257 Unused cache pairs have the value 0, 0.
7258 Keeping the cache "backwards" means that the invariant of
7259 cache[0] >= cache[2] is maintained even with empty slots, which means that
7260 the code that uses it doesn't need to worry if only 1 entry has actually
7261 been set to non-zero. It also makes the "position beyond the end of the
7262 cache" logic much simpler, as the first slot is always the one to start
7266 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7267 const STRLEN utf8, const STRLEN blen)
7271 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7276 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7277 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7278 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7280 (*mgp)->mg_len = -1;
7284 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7285 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7286 (*mgp)->mg_ptr = (char *) cache;
7290 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7291 /* SvPOKp() because it's possible that sv has string overloading, and
7292 therefore is a reference, hence SvPVX() is actually a pointer.
7293 This cures the (very real) symptoms of RT 69422, but I'm not actually
7294 sure whether we should even be caching the results of UTF-8
7295 operations on overloading, given that nothing stops overloading
7296 returning a different value every time it's called. */
7297 const U8 *start = (const U8 *) SvPVX_const(sv);
7298 const STRLEN realutf8 = utf8_length(start, start + byte);
7300 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7304 /* Cache is held with the later position first, to simplify the code
7305 that deals with unbounded ends. */
7307 ASSERT_UTF8_CACHE(cache);
7308 if (cache[1] == 0) {
7309 /* Cache is totally empty */
7312 } else if (cache[3] == 0) {
7313 if (byte > cache[1]) {
7314 /* New one is larger, so goes first. */
7315 cache[2] = cache[0];
7316 cache[3] = cache[1];
7324 #define THREEWAY_SQUARE(a,b,c,d) \
7325 ((float)((d) - (c))) * ((float)((d) - (c))) \
7326 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7327 + ((float)((b) - (a))) * ((float)((b) - (a)))
7329 /* Cache has 2 slots in use, and we know three potential pairs.
7330 Keep the two that give the lowest RMS distance. Do the
7331 calculation in bytes simply because we always know the byte
7332 length. squareroot has the same ordering as the positive value,
7333 so don't bother with the actual square root. */
7334 if (byte > cache[1]) {
7335 /* New position is after the existing pair of pairs. */
7336 const float keep_earlier
7337 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7338 const float keep_later
7339 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7341 if (keep_later < keep_earlier) {
7342 cache[2] = cache[0];
7343 cache[3] = cache[1];
7352 else if (byte > cache[3]) {
7353 /* New position is between the existing pair of pairs. */
7354 const float keep_earlier
7355 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7356 const float keep_later
7357 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7359 if (keep_later < keep_earlier) {
7369 /* New position is before the existing pair of pairs. */
7370 const float keep_earlier
7371 = THREEWAY_SQUARE(0, byte, cache[3], blen);
7372 const float keep_later
7373 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7375 if (keep_later < keep_earlier) {
7380 cache[0] = cache[2];
7381 cache[1] = cache[3];
7387 ASSERT_UTF8_CACHE(cache);
7390 /* We already know all of the way, now we may be able to walk back. The same
7391 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7392 backward is half the speed of walking forward. */
7394 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7395 const U8 *end, STRLEN endu)
7397 const STRLEN forw = target - s;
7398 STRLEN backw = end - target;
7400 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7402 if (forw < 2 * backw) {
7403 return utf8_length(s, target);
7406 while (end > target) {
7408 while (UTF8_IS_CONTINUATION(*end)) {
7417 =for apidoc sv_pos_b2u_flags
7419 Converts the offset from a count of bytes from the start of the string, to
7420 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7421 I<flags> is passed to C<SvPV_flags>, and usually should be
7422 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7428 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7429 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7434 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7437 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7443 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7445 s = (const U8*)SvPV_flags(sv, blen, flags);
7448 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7449 ", byte=%"UVuf, (UV)blen, (UV)offset);
7455 && SvTYPE(sv) >= SVt_PVMG
7456 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7459 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7460 if (cache[1] == offset) {
7461 /* An exact match. */
7464 if (cache[3] == offset) {
7465 /* An exact match. */
7469 if (cache[1] < offset) {
7470 /* We already know part of the way. */
7471 if (mg->mg_len != -1) {
7472 /* Actually, we know the end too. */
7474 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7475 s + blen, mg->mg_len - cache[0]);
7477 len = cache[0] + utf8_length(s + cache[1], send);
7480 else if (cache[3] < offset) {
7481 /* We're between the two cached pairs, so we do the calculation
7482 offset by the byte/utf-8 positions for the earlier pair,
7483 then add the utf-8 characters from the string start to
7485 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7486 s + cache[1], cache[0] - cache[2])
7490 else { /* cache[3] > offset */
7491 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7495 ASSERT_UTF8_CACHE(cache);
7497 } else if (mg->mg_len != -1) {
7498 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7502 if (!found || PL_utf8cache < 0) {
7503 const STRLEN real_len = utf8_length(s, send);
7505 if (found && PL_utf8cache < 0)
7506 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7512 utf8_mg_len_cache_update(sv, &mg, len);
7514 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7521 =for apidoc sv_pos_b2u
7523 Converts the value pointed to by offsetp from a count of bytes from the
7524 start of the string, to a count of the equivalent number of UTF-8 chars.
7525 Handles magic and type coercion.
7527 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7534 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7535 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7540 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7542 PERL_ARGS_ASSERT_SV_POS_B2U;
7547 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7548 SV_GMAGIC|SV_CONST_RETURN);
7552 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7553 STRLEN real, SV *const sv)
7555 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7557 /* As this is debugging only code, save space by keeping this test here,
7558 rather than inlining it in all the callers. */
7559 if (from_cache == real)
7562 /* Need to turn the assertions off otherwise we may recurse infinitely
7563 while printing error messages. */
7564 SAVEI8(PL_utf8cache);
7566 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7567 func, (UV) from_cache, (UV) real, SVfARG(sv));
7573 Returns a boolean indicating whether the strings in the two SVs are
7574 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7575 coerce its args to strings if necessary.
7577 =for apidoc sv_eq_flags
7579 Returns a boolean indicating whether the strings in the two SVs are
7580 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7581 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7587 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7595 SV* svrecode = NULL;
7602 /* if pv1 and pv2 are the same, second SvPV_const call may
7603 * invalidate pv1 (if we are handling magic), so we may need to
7605 if (sv1 == sv2 && flags & SV_GMAGIC
7606 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7607 pv1 = SvPV_const(sv1, cur1);
7608 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7610 pv1 = SvPV_flags_const(sv1, cur1, flags);
7618 pv2 = SvPV_flags_const(sv2, cur2, flags);
7620 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7621 /* Differing utf8ness.
7622 * Do not UTF8size the comparands as a side-effect. */
7625 svrecode = newSVpvn(pv2, cur2);
7626 sv_recode_to_utf8(svrecode, PL_encoding);
7627 pv2 = SvPV_const(svrecode, cur2);
7630 svrecode = newSVpvn(pv1, cur1);
7631 sv_recode_to_utf8(svrecode, PL_encoding);
7632 pv1 = SvPV_const(svrecode, cur1);
7634 /* Now both are in UTF-8. */
7636 SvREFCNT_dec_NN(svrecode);
7642 /* sv1 is the UTF-8 one */
7643 return bytes_cmp_utf8((const U8*)pv2, cur2,
7644 (const U8*)pv1, cur1) == 0;
7647 /* sv2 is the UTF-8 one */
7648 return bytes_cmp_utf8((const U8*)pv1, cur1,
7649 (const U8*)pv2, cur2) == 0;
7655 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7657 SvREFCNT_dec(svrecode);
7665 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7666 string in C<sv1> is less than, equal to, or greater than the string in
7667 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7668 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7670 =for apidoc sv_cmp_flags
7672 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7673 string in C<sv1> is less than, equal to, or greater than the string in
7674 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7675 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7676 also C<sv_cmp_locale_flags>.
7682 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7684 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7688 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7693 const char *pv1, *pv2;
7695 SV *svrecode = NULL;
7702 pv1 = SvPV_flags_const(sv1, cur1, flags);
7709 pv2 = SvPV_flags_const(sv2, cur2, flags);
7711 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7712 /* Differing utf8ness.
7713 * Do not UTF8size the comparands as a side-effect. */
7716 svrecode = newSVpvn(pv2, cur2);
7717 sv_recode_to_utf8(svrecode, PL_encoding);
7718 pv2 = SvPV_const(svrecode, cur2);
7721 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7722 (const U8*)pv1, cur1);
7723 return retval ? retval < 0 ? -1 : +1 : 0;
7728 svrecode = newSVpvn(pv1, cur1);
7729 sv_recode_to_utf8(svrecode, PL_encoding);
7730 pv1 = SvPV_const(svrecode, cur1);
7733 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7734 (const U8*)pv2, cur2);
7735 return retval ? retval < 0 ? -1 : +1 : 0;
7741 cmp = cur2 ? -1 : 0;
7745 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7748 cmp = retval < 0 ? -1 : 1;
7749 } else if (cur1 == cur2) {
7752 cmp = cur1 < cur2 ? -1 : 1;
7756 SvREFCNT_dec(svrecode);
7762 =for apidoc sv_cmp_locale
7764 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7765 'use bytes' aware, handles get magic, and will coerce its args to strings
7766 if necessary. See also C<sv_cmp>.
7768 =for apidoc sv_cmp_locale_flags
7770 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7771 'use bytes' aware and will coerce its args to strings if necessary. If the
7772 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7778 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7780 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7784 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7788 #ifdef USE_LOCALE_COLLATE
7794 if (PL_collation_standard)
7798 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7800 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7802 if (!pv1 || !len1) {
7813 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7816 return retval < 0 ? -1 : 1;
7819 * When the result of collation is equality, that doesn't mean
7820 * that there are no differences -- some locales exclude some
7821 * characters from consideration. So to avoid false equalities,
7822 * we use the raw string as a tiebreaker.
7829 PERL_UNUSED_ARG(flags);
7830 #endif /* USE_LOCALE_COLLATE */
7832 return sv_cmp(sv1, sv2);
7836 #ifdef USE_LOCALE_COLLATE
7839 =for apidoc sv_collxfrm
7841 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7842 C<sv_collxfrm_flags>.
7844 =for apidoc sv_collxfrm_flags
7846 Add Collate Transform magic to an SV if it doesn't already have it. If the
7847 flags contain SV_GMAGIC, it handles get-magic.
7849 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7850 scalar data of the variable, but transformed to such a format that a normal
7851 memory comparison can be used to compare the data according to the locale
7858 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7863 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7865 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7866 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7872 Safefree(mg->mg_ptr);
7873 s = SvPV_flags_const(sv, len, flags);
7874 if ((xf = mem_collxfrm(s, len, &xlen))) {
7876 #ifdef PERL_OLD_COPY_ON_WRITE
7878 sv_force_normal_flags(sv, 0);
7880 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
7894 if (mg && mg->mg_ptr) {
7896 return mg->mg_ptr + sizeof(PL_collation_ix);
7904 #endif /* USE_LOCALE_COLLATE */
7907 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7909 SV * const tsv = newSV(0);
7912 sv_gets(tsv, fp, 0);
7913 sv_utf8_upgrade_nomg(tsv);
7914 SvCUR_set(sv,append);
7917 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
7921 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7924 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
7925 /* Grab the size of the record we're getting */
7926 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
7933 /* With a true, record-oriented file on VMS, we need to use read directly
7934 * to ensure that we respect RMS record boundaries. The user is responsible
7935 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
7936 * record size) field. N.B. This is likely to produce invalid results on
7937 * varying-width character data when a record ends mid-character.
7939 fd = PerlIO_fileno(fp);
7941 && PerlLIO_fstat(fd, &st) == 0
7942 && (st.st_fab_rfm == FAB$C_VAR
7943 || st.st_fab_rfm == FAB$C_VFC
7944 || st.st_fab_rfm == FAB$C_FIX)) {
7946 bytesread = PerlLIO_read(fd, buffer, recsize);
7948 else /* in-memory file from PerlIO::Scalar
7949 * or not a record-oriented file
7953 bytesread = PerlIO_read(fp, buffer, recsize);
7955 /* At this point, the logic in sv_get() means that sv will
7956 be treated as utf-8 if the handle is utf8.
7958 if (PerlIO_isutf8(fp) && bytesread > 0) {
7959 char *bend = buffer + bytesread;
7960 char *bufp = buffer;
7961 size_t charcount = 0;
7962 bool charstart = TRUE;
7965 while (charcount < recsize) {
7966 /* count accumulated characters */
7967 while (bufp < bend) {
7969 skip = UTF8SKIP(bufp);
7971 if (bufp + skip > bend) {
7972 /* partial at the end */
7983 if (charcount < recsize) {
7985 STRLEN bufp_offset = bufp - buffer;
7986 SSize_t morebytesread;
7988 /* originally I read enough to fill any incomplete
7989 character and the first byte of the next
7990 character if needed, but if there's many
7991 multi-byte encoded characters we're going to be
7992 making a read call for every character beyond
7993 the original read size.
7995 So instead, read the rest of the character if
7996 any, and enough bytes to match at least the
7997 start bytes for each character we're going to
8001 readsize = recsize - charcount;
8003 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8004 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8005 bend = buffer + bytesread;
8006 morebytesread = PerlIO_read(fp, bend, readsize);
8007 if (morebytesread <= 0) {
8008 /* we're done, if we still have incomplete
8009 characters the check code in sv_gets() will
8012 I'd originally considered doing
8013 PerlIO_ungetc() on all but the lead
8014 character of the incomplete character, but
8015 read() doesn't do that, so I don't.
8020 /* prepare to scan some more */
8021 bytesread += morebytesread;
8022 bend = buffer + bytesread;
8023 bufp = buffer + bufp_offset;
8031 SvCUR_set(sv, bytesread + append);
8032 buffer[bytesread] = '\0';
8033 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8039 Get a line from the filehandle and store it into the SV, optionally
8040 appending to the currently-stored string. If C<append> is not 0, the
8041 line is appended to the SV instead of overwriting it. C<append> should
8042 be set to the byte offset that the appended string should start at
8043 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8049 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8060 PERL_ARGS_ASSERT_SV_GETS;
8062 if (SvTHINKFIRST(sv))
8063 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8064 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8066 However, perlbench says it's slower, because the existing swipe code
8067 is faster than copy on write.
8068 Swings and roundabouts. */
8069 SvUPGRADE(sv, SVt_PV);
8072 /* line is going to be appended to the existing buffer in the sv */
8073 if (PerlIO_isutf8(fp)) {
8075 sv_utf8_upgrade_nomg(sv);
8076 sv_pos_u2b(sv,&append,0);
8078 } else if (SvUTF8(sv)) {
8079 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8085 /* not appending - "clear" the string by setting SvCUR to 0,
8086 * the pv is still avaiable. */
8089 if (PerlIO_isutf8(fp))
8092 if (IN_PERL_COMPILETIME) {
8093 /* we always read code in line mode */
8097 else if (RsSNARF(PL_rs)) {
8098 /* If it is a regular disk file use size from stat() as estimate
8099 of amount we are going to read -- may result in mallocing
8100 more memory than we really need if the layers below reduce
8101 the size we read (e.g. CRLF or a gzip layer).
8104 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8105 const Off_t offset = PerlIO_tell(fp);
8106 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8107 #ifdef PERL_NEW_COPY_ON_WRITE
8108 /* Add an extra byte for the sake of copy-on-write's
8109 * buffer reference count. */
8110 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8112 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8119 else if (RsRECORD(PL_rs)) {
8120 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8122 else if (RsPARA(PL_rs)) {
8128 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8129 if (PerlIO_isutf8(fp)) {
8130 rsptr = SvPVutf8(PL_rs, rslen);
8133 if (SvUTF8(PL_rs)) {
8134 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8135 Perl_croak(aTHX_ "Wide character in $/");
8138 /* extract the raw pointer to the record separator */
8139 rsptr = SvPV_const(PL_rs, rslen);
8143 /* rslast is the last character in the record separator
8144 * note we don't use rslast except when rslen is true, so the
8145 * null assign is a placeholder. */
8146 rslast = rslen ? rsptr[rslen - 1] : '\0';
8148 if (rspara) { /* have to do this both before and after */
8149 do { /* to make sure file boundaries work right */
8152 i = PerlIO_getc(fp);
8156 PerlIO_ungetc(fp,i);
8162 /* See if we know enough about I/O mechanism to cheat it ! */
8164 /* This used to be #ifdef test - it is made run-time test for ease
8165 of abstracting out stdio interface. One call should be cheap
8166 enough here - and may even be a macro allowing compile
8170 if (PerlIO_fast_gets(fp)) {
8172 * We can do buffer based IO operations on this filehandle.
8174 * This means we can bypass a lot of subcalls and process
8175 * the buffer directly, it also means we know the upper bound
8176 * on the amount of data we might read of the current buffer
8177 * into our sv. Knowing this allows us to preallocate the pv
8178 * to be able to hold that maximum, which allows us to simplify
8179 * a lot of logic. */
8182 * We're going to steal some values from the stdio struct
8183 * and put EVERYTHING in the innermost loop into registers.
8185 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8186 STRLEN bpx; /* length of the data in the target sv
8187 used to fix pointers after a SvGROW */
8188 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8189 of data left in the read-ahead buffer.
8190 If 0 then the pv buffer can hold the full
8191 amount left, otherwise this is the amount it
8194 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8195 /* An ungetc()d char is handled separately from the regular
8196 * buffer, so we getc() it back out and stuff it in the buffer.
8198 i = PerlIO_getc(fp);
8199 if (i == EOF) return 0;
8200 *(--((*fp)->_ptr)) = (unsigned char) i;
8204 /* Here is some breathtakingly efficient cheating */
8206 /* When you read the following logic resist the urge to think
8207 * of record separators that are 1 byte long. They are an
8208 * uninteresting special (simple) case.
8210 * Instead think of record separators which are at least 2 bytes
8211 * long, and keep in mind that we need to deal with such
8212 * separators when they cross a read-ahead buffer boundary.
8214 * Also consider that we need to gracefully deal with separators
8215 * that may be longer than a single read ahead buffer.
8217 * Lastly do not forget we want to copy the delimiter as well. We
8218 * are copying all data in the file _up_to_and_including_ the separator
8221 * Now that you have all that in mind here is what is happening below:
8223 * 1. When we first enter the loop we do some memory book keeping to see
8224 * how much free space there is in the target SV. (This sub assumes that
8225 * it is operating on the same SV most of the time via $_ and that it is
8226 * going to be able to reuse the same pv buffer each call.) If there is
8227 * "enough" room then we set "shortbuffered" to how much space there is
8228 * and start reading forward.
8230 * 2. When we scan forward we copy from the read-ahead buffer to the target
8231 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8232 * and the end of the of pv, as well as for the "rslast", which is the last
8233 * char of the separator.
8235 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8236 * (which has a "complete" record up to the point we saw rslast) and check
8237 * it to see if it matches the separator. If it does we are done. If it doesn't
8238 * we continue on with the scan/copy.
8240 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8241 * the IO system to read the next buffer. We do this by doing a getc(), which
8242 * returns a single char read (or EOF), and prefills the buffer, and also
8243 * allows us to find out how full the buffer is. We use this information to
8244 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8245 * the returned single char into the target sv, and then go back into scan
8248 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8249 * remaining space in the read-buffer.
8251 * Note that this code despite its twisty-turny nature is pretty darn slick.
8252 * It manages single byte separators, multi-byte cross boundary separators,
8253 * and cross-read-buffer separators cleanly and efficiently at the cost
8254 * of potentially greatly overallocating the target SV.
8260 /* get the number of bytes remaining in the read-ahead buffer
8261 * on first call on a given fp this will return 0.*/
8262 cnt = PerlIO_get_cnt(fp);
8264 /* make sure we have the room */
8265 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8266 /* Not room for all of it
8267 if we are looking for a separator and room for some
8269 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8270 /* just process what we have room for */
8271 shortbuffered = cnt - SvLEN(sv) + append + 1;
8272 cnt -= shortbuffered;
8275 /* ensure that the target sv has enough room to hold
8276 * the rest of the read-ahead buffer */
8278 /* remember that cnt can be negative */
8279 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8283 /* we have enough room to hold the full buffer, lets scream */
8287 /* extract the pointer to sv's string buffer, offset by append as necessary */
8288 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8289 /* extract the point to the read-ahead buffer */
8290 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8292 /* some trace debug output */
8293 DEBUG_P(PerlIO_printf(Perl_debug_log,
8294 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8295 DEBUG_P(PerlIO_printf(Perl_debug_log,
8296 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8298 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8299 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8303 /* if there is stuff left in the read-ahead buffer */
8305 /* if there is a separator */
8307 /* loop until we hit the end of the read-ahead buffer */
8308 while (cnt > 0) { /* this | eat */
8309 /* scan forward copying and searching for rslast as we go */
8311 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8312 goto thats_all_folks; /* screams | sed :-) */
8316 /* no separator, slurp the full buffer */
8317 Copy(ptr, bp, cnt, char); /* this | eat */
8318 bp += cnt; /* screams | dust */
8319 ptr += cnt; /* louder | sed :-) */
8321 assert (!shortbuffered);
8322 goto cannot_be_shortbuffered;
8326 if (shortbuffered) { /* oh well, must extend */
8327 /* we didnt have enough room to fit the line into the target buffer
8328 * so we must extend the target buffer and keep going */
8329 cnt = shortbuffered;
8331 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8333 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8334 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8335 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8339 cannot_be_shortbuffered:
8340 /* we need to refill the read-ahead buffer if possible */
8342 DEBUG_P(PerlIO_printf(Perl_debug_log,
8343 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8344 PTR2UV(ptr),(IV)cnt));
8345 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8347 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8348 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8349 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8350 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8353 call PerlIO_getc() to let it prefill the lookahead buffer
8355 This used to call 'filbuf' in stdio form, but as that behaves like
8356 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8357 another abstraction.
8359 Note we have to deal with the char in 'i' if we are not at EOF
8361 i = PerlIO_getc(fp); /* get more characters */
8363 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8364 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8365 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8366 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8368 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8369 cnt = PerlIO_get_cnt(fp);
8370 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8371 DEBUG_P(PerlIO_printf(Perl_debug_log,
8372 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8373 PTR2UV(ptr),(IV)cnt));
8375 if (i == EOF) /* all done for ever? */
8376 goto thats_really_all_folks;
8378 /* make sure we have enough space in the target sv */
8379 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8381 SvGROW(sv, bpx + cnt + 2);
8382 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8384 /* copy of the char we got from getc() */
8385 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8387 /* make sure we deal with the i being the last character of a separator */
8388 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8389 goto thats_all_folks;
8393 /* check if we have actually found the separator - only really applies
8395 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8396 memNE((char*)bp - rslen, rsptr, rslen))
8397 goto screamer; /* go back to the fray */
8398 thats_really_all_folks:
8400 cnt += shortbuffered;
8401 DEBUG_P(PerlIO_printf(Perl_debug_log,
8402 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8403 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8404 DEBUG_P(PerlIO_printf(Perl_debug_log,
8405 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8407 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8408 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8410 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8411 DEBUG_P(PerlIO_printf(Perl_debug_log,
8412 "Screamer: done, len=%ld, string=|%.*s|\n",
8413 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8417 /*The big, slow, and stupid way. */
8418 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8419 STDCHAR *buf = NULL;
8420 Newx(buf, 8192, STDCHAR);
8428 const STDCHAR * const bpe = buf + sizeof(buf);
8430 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8431 ; /* keep reading */
8435 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8436 /* Accommodate broken VAXC compiler, which applies U8 cast to
8437 * both args of ?: operator, causing EOF to change into 255
8440 i = (U8)buf[cnt - 1];
8446 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8448 sv_catpvn_nomg(sv, (char *) buf, cnt);
8450 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8452 if (i != EOF && /* joy */
8454 SvCUR(sv) < rslen ||
8455 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8459 * If we're reading from a TTY and we get a short read,
8460 * indicating that the user hit his EOF character, we need
8461 * to notice it now, because if we try to read from the TTY
8462 * again, the EOF condition will disappear.
8464 * The comparison of cnt to sizeof(buf) is an optimization
8465 * that prevents unnecessary calls to feof().
8469 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8473 #ifdef USE_HEAP_INSTEAD_OF_STACK
8478 if (rspara) { /* have to do this both before and after */
8479 while (i != EOF) { /* to make sure file boundaries work right */
8480 i = PerlIO_getc(fp);
8482 PerlIO_ungetc(fp,i);
8488 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8494 Auto-increment of the value in the SV, doing string to numeric conversion
8495 if necessary. Handles 'get' magic and operator overloading.
8501 Perl_sv_inc(pTHX_ SV *const sv)
8510 =for apidoc sv_inc_nomg
8512 Auto-increment of the value in the SV, doing string to numeric conversion
8513 if necessary. Handles operator overloading. Skips handling 'get' magic.
8519 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8527 if (SvTHINKFIRST(sv)) {
8528 if (SvREADONLY(sv)) {
8529 Perl_croak_no_modify();
8533 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8535 i = PTR2IV(SvRV(sv));
8539 else sv_force_normal_flags(sv, 0);
8541 flags = SvFLAGS(sv);
8542 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8543 /* It's (privately or publicly) a float, but not tested as an
8544 integer, so test it to see. */
8546 flags = SvFLAGS(sv);
8548 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8549 /* It's publicly an integer, or privately an integer-not-float */
8550 #ifdef PERL_PRESERVE_IVUV
8554 if (SvUVX(sv) == UV_MAX)
8555 sv_setnv(sv, UV_MAX_P1);
8557 (void)SvIOK_only_UV(sv);
8558 SvUV_set(sv, SvUVX(sv) + 1);
8560 if (SvIVX(sv) == IV_MAX)
8561 sv_setuv(sv, (UV)IV_MAX + 1);
8563 (void)SvIOK_only(sv);
8564 SvIV_set(sv, SvIVX(sv) + 1);
8569 if (flags & SVp_NOK) {
8570 const NV was = SvNVX(sv);
8571 if (NV_OVERFLOWS_INTEGERS_AT &&
8572 was >= NV_OVERFLOWS_INTEGERS_AT) {
8573 /* diag_listed_as: Lost precision when %s %f by 1 */
8574 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8575 "Lost precision when incrementing %" NVff " by 1",
8578 (void)SvNOK_only(sv);
8579 SvNV_set(sv, was + 1.0);
8583 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8584 if ((flags & SVTYPEMASK) < SVt_PVIV)
8585 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8586 (void)SvIOK_only(sv);
8591 while (isALPHA(*d)) d++;
8592 while (isDIGIT(*d)) d++;
8593 if (d < SvEND(sv)) {
8594 #ifdef PERL_PRESERVE_IVUV
8595 /* Got to punt this as an integer if needs be, but we don't issue
8596 warnings. Probably ought to make the sv_iv_please() that does
8597 the conversion if possible, and silently. */
8598 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8599 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8600 /* Need to try really hard to see if it's an integer.
8601 9.22337203685478e+18 is an integer.
8602 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8603 so $a="9.22337203685478e+18"; $a+0; $a++
8604 needs to be the same as $a="9.22337203685478e+18"; $a++
8611 /* sv_2iv *should* have made this an NV */
8612 if (flags & SVp_NOK) {
8613 (void)SvNOK_only(sv);
8614 SvNV_set(sv, SvNVX(sv) + 1.0);
8617 /* I don't think we can get here. Maybe I should assert this
8618 And if we do get here I suspect that sv_setnv will croak. NWC
8620 #if defined(USE_LONG_DOUBLE)
8621 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",
8622 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8624 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8625 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8628 #endif /* PERL_PRESERVE_IVUV */
8629 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8633 while (d >= SvPVX_const(sv)) {
8641 /* MKS: The original code here died if letters weren't consecutive.
8642 * at least it didn't have to worry about non-C locales. The
8643 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8644 * arranged in order (although not consecutively) and that only
8645 * [A-Za-z] are accepted by isALPHA in the C locale.
8647 if (*d != 'z' && *d != 'Z') {
8648 do { ++*d; } while (!isALPHA(*d));
8651 *(d--) -= 'z' - 'a';
8656 *(d--) -= 'z' - 'a' + 1;
8660 /* oh,oh, the number grew */
8661 SvGROW(sv, SvCUR(sv) + 2);
8662 SvCUR_set(sv, SvCUR(sv) + 1);
8663 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8674 Auto-decrement of the value in the SV, doing string to numeric conversion
8675 if necessary. Handles 'get' magic and operator overloading.
8681 Perl_sv_dec(pTHX_ SV *const sv)
8691 =for apidoc sv_dec_nomg
8693 Auto-decrement of the value in the SV, doing string to numeric conversion
8694 if necessary. Handles operator overloading. Skips handling 'get' magic.
8700 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8707 if (SvTHINKFIRST(sv)) {
8708 if (SvREADONLY(sv)) {
8709 Perl_croak_no_modify();
8713 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8715 i = PTR2IV(SvRV(sv));
8719 else sv_force_normal_flags(sv, 0);
8721 /* Unlike sv_inc we don't have to worry about string-never-numbers
8722 and keeping them magic. But we mustn't warn on punting */
8723 flags = SvFLAGS(sv);
8724 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8725 /* It's publicly an integer, or privately an integer-not-float */
8726 #ifdef PERL_PRESERVE_IVUV
8730 if (SvUVX(sv) == 0) {
8731 (void)SvIOK_only(sv);
8735 (void)SvIOK_only_UV(sv);
8736 SvUV_set(sv, SvUVX(sv) - 1);
8739 if (SvIVX(sv) == IV_MIN) {
8740 sv_setnv(sv, (NV)IV_MIN);
8744 (void)SvIOK_only(sv);
8745 SvIV_set(sv, SvIVX(sv) - 1);
8750 if (flags & SVp_NOK) {
8753 const NV was = SvNVX(sv);
8754 if (NV_OVERFLOWS_INTEGERS_AT &&
8755 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8756 /* diag_listed_as: Lost precision when %s %f by 1 */
8757 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8758 "Lost precision when decrementing %" NVff " by 1",
8761 (void)SvNOK_only(sv);
8762 SvNV_set(sv, was - 1.0);
8766 if (!(flags & SVp_POK)) {
8767 if ((flags & SVTYPEMASK) < SVt_PVIV)
8768 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8770 (void)SvIOK_only(sv);
8773 #ifdef PERL_PRESERVE_IVUV
8775 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8776 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8777 /* Need to try really hard to see if it's an integer.
8778 9.22337203685478e+18 is an integer.
8779 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8780 so $a="9.22337203685478e+18"; $a+0; $a--
8781 needs to be the same as $a="9.22337203685478e+18"; $a--
8788 /* sv_2iv *should* have made this an NV */
8789 if (flags & SVp_NOK) {
8790 (void)SvNOK_only(sv);
8791 SvNV_set(sv, SvNVX(sv) - 1.0);
8794 /* I don't think we can get here. Maybe I should assert this
8795 And if we do get here I suspect that sv_setnv will croak. NWC
8797 #if defined(USE_LONG_DOUBLE)
8798 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",
8799 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8801 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8802 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8806 #endif /* PERL_PRESERVE_IVUV */
8807 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8810 /* this define is used to eliminate a chunk of duplicated but shared logic
8811 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8812 * used anywhere but here - yves
8814 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8817 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8821 =for apidoc sv_mortalcopy
8823 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8824 The new SV is marked as mortal. It will be destroyed "soon", either by an
8825 explicit call to FREETMPS, or by an implicit call at places such as
8826 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8831 /* Make a string that will exist for the duration of the expression
8832 * evaluation. Actually, it may have to last longer than that, but
8833 * hopefully we won't free it until it has been assigned to a
8834 * permanent location. */
8837 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8842 if (flags & SV_GMAGIC)
8843 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8845 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8846 PUSH_EXTEND_MORTAL__SV_C(sv);
8852 =for apidoc sv_newmortal
8854 Creates a new null SV which is mortal. The reference count of the SV is
8855 set to 1. It will be destroyed "soon", either by an explicit call to
8856 FREETMPS, or by an implicit call at places such as statement boundaries.
8857 See also C<sv_mortalcopy> and C<sv_2mortal>.
8863 Perl_sv_newmortal(pTHX)
8869 SvFLAGS(sv) = SVs_TEMP;
8870 PUSH_EXTEND_MORTAL__SV_C(sv);
8876 =for apidoc newSVpvn_flags
8878 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8879 characters) into it. The reference count for the
8880 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8881 string. You are responsible for ensuring that the source string is at least
8882 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8883 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8884 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8885 returning. If C<SVf_UTF8> is set, C<s>
8886 is considered to be in UTF-8 and the
8887 C<SVf_UTF8> flag will be set on the new SV.
8888 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8890 #define newSVpvn_utf8(s, len, u) \
8891 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8897 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8902 /* All the flags we don't support must be zero.
8903 And we're new code so I'm going to assert this from the start. */
8904 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
8906 sv_setpvn(sv,s,len);
8908 /* This code used to do a sv_2mortal(), however we now unroll the call to
8909 * sv_2mortal() and do what it does ourselves here. Since we have asserted
8910 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
8911 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
8912 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
8913 * means that we eliminate quite a few steps than it looks - Yves
8914 * (explaining patch by gfx) */
8916 SvFLAGS(sv) |= flags;
8918 if(flags & SVs_TEMP){
8919 PUSH_EXTEND_MORTAL__SV_C(sv);
8926 =for apidoc sv_2mortal
8928 Marks an existing SV as mortal. The SV will be destroyed "soon", either
8929 by an explicit call to FREETMPS, or by an implicit call at places such as
8930 statement boundaries. SvTEMP() is turned on which means that the SV's
8931 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
8932 and C<sv_mortalcopy>.
8938 Perl_sv_2mortal(pTHX_ SV *const sv)
8945 PUSH_EXTEND_MORTAL__SV_C(sv);
8953 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8954 characters) into it. The reference count for the
8955 SV is set to 1. If C<len> is zero, Perl will compute the length using
8956 strlen(), (which means if you use this option, that C<s> can't have embedded
8957 C<NUL> characters and has to have a terminating C<NUL> byte).
8959 For efficiency, consider using C<newSVpvn> instead.
8965 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
8971 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
8976 =for apidoc newSVpvn
8978 Creates a new SV and copies a string into it, which may contain C<NUL> characters
8979 (C<\0>) and other binary data. The reference count for the SV is set to 1.
8980 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
8981 are responsible for ensuring that the source buffer is at least
8982 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
8989 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
8995 sv_setpvn(sv,buffer,len);
9000 =for apidoc newSVhek
9002 Creates a new SV from the hash key structure. It will generate scalars that
9003 point to the shared string table where possible. Returns a new (undefined)
9004 SV if the hek is NULL.
9010 Perl_newSVhek(pTHX_ const HEK *const hek)
9020 if (HEK_LEN(hek) == HEf_SVKEY) {
9021 return newSVsv(*(SV**)HEK_KEY(hek));
9023 const int flags = HEK_FLAGS(hek);
9024 if (flags & HVhek_WASUTF8) {
9026 Andreas would like keys he put in as utf8 to come back as utf8
9028 STRLEN utf8_len = HEK_LEN(hek);
9029 SV * const sv = newSV_type(SVt_PV);
9030 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9031 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9032 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9035 } else if (flags & HVhek_UNSHARED) {
9036 /* A hash that isn't using shared hash keys has to have
9037 the flag in every key so that we know not to try to call
9038 share_hek_hek on it. */
9040 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9045 /* This will be overwhelminly the most common case. */
9047 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9048 more efficient than sharepvn(). */
9052 sv_upgrade(sv, SVt_PV);
9053 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9054 SvCUR_set(sv, HEK_LEN(hek));
9066 =for apidoc newSVpvn_share
9068 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9069 table. If the string does not already exist in the table, it is
9070 created first. Turns on the SvIsCOW flag (or READONLY
9071 and FAKE in 5.16 and earlier). If the C<hash> parameter
9072 is non-zero, that value is used; otherwise the hash is computed.
9073 The string's hash can later be retrieved from the SV
9074 with the C<SvSHARED_HASH()> macro. The idea here is
9075 that as the string table is used for shared hash keys these strings will have
9076 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9082 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9086 bool is_utf8 = FALSE;
9087 const char *const orig_src = src;
9090 STRLEN tmplen = -len;
9092 /* See the note in hv.c:hv_fetch() --jhi */
9093 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9097 PERL_HASH(hash, src, len);
9099 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9100 changes here, update it there too. */
9101 sv_upgrade(sv, SVt_PV);
9102 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9109 if (src != orig_src)
9115 =for apidoc newSVpv_share
9117 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9124 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9126 return newSVpvn_share(src, strlen(src), hash);
9129 #if defined(PERL_IMPLICIT_CONTEXT)
9131 /* pTHX_ magic can't cope with varargs, so this is a no-context
9132 * version of the main function, (which may itself be aliased to us).
9133 * Don't access this version directly.
9137 Perl_newSVpvf_nocontext(const char *const pat, ...)
9143 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9145 va_start(args, pat);
9146 sv = vnewSVpvf(pat, &args);
9153 =for apidoc newSVpvf
9155 Creates a new SV and initializes it with the string formatted like
9162 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9167 PERL_ARGS_ASSERT_NEWSVPVF;
9169 va_start(args, pat);
9170 sv = vnewSVpvf(pat, &args);
9175 /* backend for newSVpvf() and newSVpvf_nocontext() */
9178 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9183 PERL_ARGS_ASSERT_VNEWSVPVF;
9186 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9193 Creates a new SV and copies a floating point value into it.
9194 The reference count for the SV is set to 1.
9200 Perl_newSVnv(pTHX_ const NV n)
9213 Creates a new SV and copies an integer into it. The reference count for the
9220 Perl_newSViv(pTHX_ const IV i)
9233 Creates a new SV and copies an unsigned integer into it.
9234 The reference count for the SV is set to 1.
9240 Perl_newSVuv(pTHX_ const UV u)
9251 =for apidoc newSV_type
9253 Creates a new SV, of the type specified. The reference count for the new SV
9260 Perl_newSV_type(pTHX_ const svtype type)
9265 sv_upgrade(sv, type);
9270 =for apidoc newRV_noinc
9272 Creates an RV wrapper for an SV. The reference count for the original
9273 SV is B<not> incremented.
9279 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9282 SV *sv = newSV_type(SVt_IV);
9284 PERL_ARGS_ASSERT_NEWRV_NOINC;
9287 SvRV_set(sv, tmpRef);
9292 /* newRV_inc is the official function name to use now.
9293 * newRV_inc is in fact #defined to newRV in sv.h
9297 Perl_newRV(pTHX_ SV *const sv)
9301 PERL_ARGS_ASSERT_NEWRV;
9303 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9309 Creates a new SV which is an exact duplicate of the original SV.
9316 Perl_newSVsv(pTHX_ SV *const old)
9323 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9324 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9327 /* Do this here, otherwise we leak the new SV if this croaks. */
9330 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9331 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9332 sv_setsv_flags(sv, old, SV_NOSTEAL);
9337 =for apidoc sv_reset
9339 Underlying implementation for the C<reset> Perl function.
9340 Note that the perl-level function is vaguely deprecated.
9346 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9348 PERL_ARGS_ASSERT_SV_RESET;
9350 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9354 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9357 char todo[PERL_UCHAR_MAX+1];
9360 if (!stash || SvTYPE(stash) != SVt_PVHV)
9363 if (!s) { /* reset ?? searches */
9364 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9366 const U32 count = mg->mg_len / sizeof(PMOP**);
9367 PMOP **pmp = (PMOP**) mg->mg_ptr;
9368 PMOP *const *const end = pmp + count;
9372 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9374 (*pmp)->op_pmflags &= ~PMf_USED;
9382 /* reset variables */
9384 if (!HvARRAY(stash))
9387 Zero(todo, 256, char);
9391 I32 i = (unsigned char)*s;
9395 max = (unsigned char)*s++;
9396 for ( ; i <= max; i++) {
9399 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9401 for (entry = HvARRAY(stash)[i];
9403 entry = HeNEXT(entry))
9408 if (!todo[(U8)*HeKEY(entry)])
9410 gv = MUTABLE_GV(HeVAL(entry));
9412 if (sv && !SvREADONLY(sv)) {
9413 SV_CHECK_THINKFIRST_COW_DROP(sv);
9414 if (!isGV(sv)) SvOK_off(sv);
9419 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9430 Using various gambits, try to get an IO from an SV: the IO slot if its a
9431 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9432 named after the PV if we're a string.
9434 'Get' magic is ignored on the sv passed in, but will be called on
9435 C<SvRV(sv)> if sv is an RV.
9441 Perl_sv_2io(pTHX_ SV *const sv)
9446 PERL_ARGS_ASSERT_SV_2IO;
9448 switch (SvTYPE(sv)) {
9450 io = MUTABLE_IO(sv);
9454 if (isGV_with_GP(sv)) {
9455 gv = MUTABLE_GV(sv);
9458 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9459 HEKfARG(GvNAME_HEK(gv)));
9465 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9467 SvGETMAGIC(SvRV(sv));
9468 return sv_2io(SvRV(sv));
9470 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9477 if (SvGMAGICAL(sv)) {
9478 newsv = sv_newmortal();
9479 sv_setsv_nomg(newsv, sv);
9481 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9491 Using various gambits, try to get a CV from an SV; in addition, try if
9492 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9493 The flags in C<lref> are passed to gv_fetchsv.
9499 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9505 PERL_ARGS_ASSERT_SV_2CV;
9512 switch (SvTYPE(sv)) {
9516 return MUTABLE_CV(sv);
9526 sv = amagic_deref_call(sv, to_cv_amg);
9529 if (SvTYPE(sv) == SVt_PVCV) {
9530 cv = MUTABLE_CV(sv);
9535 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9536 gv = MUTABLE_GV(sv);
9538 Perl_croak(aTHX_ "Not a subroutine reference");
9540 else if (isGV_with_GP(sv)) {
9541 gv = MUTABLE_GV(sv);
9544 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9551 /* Some flags to gv_fetchsv mean don't really create the GV */
9552 if (!isGV_with_GP(gv)) {
9557 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9558 /* XXX this is probably not what they think they're getting.
9559 * It has the same effect as "sub name;", i.e. just a forward
9570 Returns true if the SV has a true value by Perl's rules.
9571 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9572 instead use an in-line version.
9578 Perl_sv_true(pTHX_ SV *const sv)
9583 const XPV* const tXpv = (XPV*)SvANY(sv);
9585 (tXpv->xpv_cur > 1 ||
9586 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9593 return SvIVX(sv) != 0;
9596 return SvNVX(sv) != 0.0;
9598 return sv_2bool(sv);
9604 =for apidoc sv_pvn_force
9606 Get a sensible string out of the SV somehow.
9607 A private implementation of the C<SvPV_force> macro for compilers which
9608 can't cope with complex macro expressions. Always use the macro instead.
9610 =for apidoc sv_pvn_force_flags
9612 Get a sensible string out of the SV somehow.
9613 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9614 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9615 implemented in terms of this function.
9616 You normally want to use the various wrapper macros instead: see
9617 C<SvPV_force> and C<SvPV_force_nomg>
9623 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9627 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9629 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9630 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9631 sv_force_normal_flags(sv, 0);
9641 if (SvTYPE(sv) > SVt_PVLV
9642 || isGV_with_GP(sv))
9643 /* diag_listed_as: Can't coerce %s to %s in %s */
9644 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9646 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9653 if (SvTYPE(sv) < SVt_PV ||
9654 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9657 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9658 SvGROW(sv, len + 1);
9659 Move(s,SvPVX(sv),len,char);
9661 SvPVX(sv)[len] = '\0';
9664 SvPOK_on(sv); /* validate pointer */
9666 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9667 PTR2UV(sv),SvPVX_const(sv)));
9670 (void)SvPOK_only_UTF8(sv);
9671 return SvPVX_mutable(sv);
9675 =for apidoc sv_pvbyten_force
9677 The backend for the C<SvPVbytex_force> macro. Always use the macro
9684 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9686 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9688 sv_pvn_force(sv,lp);
9689 sv_utf8_downgrade(sv,0);
9695 =for apidoc sv_pvutf8n_force
9697 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9704 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9706 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9709 sv_utf8_upgrade_nomg(sv);
9715 =for apidoc sv_reftype
9717 Returns a string describing what the SV is a reference to.
9723 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9725 PERL_ARGS_ASSERT_SV_REFTYPE;
9726 if (ob && SvOBJECT(sv)) {
9727 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9730 /* WARNING - There is code, for instance in mg.c, that assumes that
9731 * the only reason that sv_reftype(sv,0) would return a string starting
9732 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9733 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9734 * this routine inside other subs, and it saves time.
9735 * Do not change this assumption without searching for "dodgy type check" in
9738 switch (SvTYPE(sv)) {
9753 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9754 /* tied lvalues should appear to be
9755 * scalars for backwards compatibility */
9756 : (LvTYPE(sv) == 't' || LvTYPE(sv) == 'T')
9757 ? "SCALAR" : "LVALUE");
9758 case SVt_PVAV: return "ARRAY";
9759 case SVt_PVHV: return "HASH";
9760 case SVt_PVCV: return "CODE";
9761 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9762 ? "GLOB" : "SCALAR");
9763 case SVt_PVFM: return "FORMAT";
9764 case SVt_PVIO: return "IO";
9765 case SVt_INVLIST: return "INVLIST";
9766 case SVt_REGEXP: return "REGEXP";
9767 default: return "UNKNOWN";
9775 Returns a SV describing what the SV passed in is a reference to.
9781 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9783 PERL_ARGS_ASSERT_SV_REF;
9786 dst = sv_newmortal();
9788 if (ob && SvOBJECT(sv)) {
9789 HvNAME_get(SvSTASH(sv))
9790 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9791 : sv_setpvn(dst, "__ANON__", 8);
9794 const char * reftype = sv_reftype(sv, 0);
9795 sv_setpv(dst, reftype);
9801 =for apidoc sv_isobject
9803 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9804 object. If the SV is not an RV, or if the object is not blessed, then this
9811 Perl_sv_isobject(pTHX_ SV *sv)
9827 Returns a boolean indicating whether the SV is blessed into the specified
9828 class. This does not check for subtypes; use C<sv_derived_from> to verify
9829 an inheritance relationship.
9835 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9839 PERL_ARGS_ASSERT_SV_ISA;
9849 hvname = HvNAME_get(SvSTASH(sv));
9853 return strEQ(hvname, name);
9859 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9860 RV then it will be upgraded to one. If C<classname> is non-null then the new
9861 SV will be blessed in the specified package. The new SV is returned and its
9862 reference count is 1. The reference count 1 is owned by C<rv>.
9868 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9873 PERL_ARGS_ASSERT_NEWSVRV;
9877 SV_CHECK_THINKFIRST_COW_DROP(rv);
9879 if (SvTYPE(rv) >= SVt_PVMG) {
9880 const U32 refcnt = SvREFCNT(rv);
9884 SvREFCNT(rv) = refcnt;
9886 sv_upgrade(rv, SVt_IV);
9887 } else if (SvROK(rv)) {
9888 SvREFCNT_dec(SvRV(rv));
9890 prepare_SV_for_RV(rv);
9898 HV* const stash = gv_stashpv(classname, GV_ADD);
9899 (void)sv_bless(rv, stash);
9905 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9907 SV * const lv = newSV_type(SVt_PVLV);
9908 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9910 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9911 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9912 LvSTARGOFF(lv) = ix;
9913 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9918 =for apidoc sv_setref_pv
9920 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9921 argument will be upgraded to an RV. That RV will be modified to point to
9922 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
9923 into the SV. The C<classname> argument indicates the package for the
9924 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9925 will have a reference count of 1, and the RV will be returned.
9927 Do not use with other Perl types such as HV, AV, SV, CV, because those
9928 objects will become corrupted by the pointer copy process.
9930 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
9936 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
9940 PERL_ARGS_ASSERT_SV_SETREF_PV;
9943 sv_setsv(rv, &PL_sv_undef);
9947 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
9952 =for apidoc sv_setref_iv
9954 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
9955 argument will be upgraded to an RV. That RV will be modified to point to
9956 the new SV. The C<classname> argument indicates the package for the
9957 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9958 will have a reference count of 1, and the RV will be returned.
9964 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
9966 PERL_ARGS_ASSERT_SV_SETREF_IV;
9968 sv_setiv(newSVrv(rv,classname), iv);
9973 =for apidoc sv_setref_uv
9975 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
9976 argument will be upgraded to an RV. That RV will be modified to point to
9977 the new SV. The C<classname> argument indicates the package for the
9978 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9979 will have a reference count of 1, and the RV will be returned.
9985 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
9987 PERL_ARGS_ASSERT_SV_SETREF_UV;
9989 sv_setuv(newSVrv(rv,classname), uv);
9994 =for apidoc sv_setref_nv
9996 Copies a double into a new SV, optionally blessing the SV. The C<rv>
9997 argument will be upgraded to an RV. That RV will be modified to point to
9998 the new SV. The C<classname> argument indicates the package for the
9999 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10000 will have a reference count of 1, and the RV will be returned.
10006 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10008 PERL_ARGS_ASSERT_SV_SETREF_NV;
10010 sv_setnv(newSVrv(rv,classname), nv);
10015 =for apidoc sv_setref_pvn
10017 Copies a string into a new SV, optionally blessing the SV. The length of the
10018 string must be specified with C<n>. The C<rv> argument will be upgraded to
10019 an RV. That RV will be modified to point to the new SV. The C<classname>
10020 argument indicates the package for the blessing. Set C<classname> to
10021 C<NULL> to avoid the blessing. The new SV will have a reference count
10022 of 1, and the RV will be returned.
10024 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10030 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10031 const char *const pv, const STRLEN n)
10033 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10035 sv_setpvn(newSVrv(rv,classname), pv, n);
10040 =for apidoc sv_bless
10042 Blesses an SV into a specified package. The SV must be an RV. The package
10043 must be designated by its stash (see C<gv_stashpv()>). The reference count
10044 of the SV is unaffected.
10050 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10054 HV *oldstash = NULL;
10056 PERL_ARGS_ASSERT_SV_BLESS;
10060 Perl_croak(aTHX_ "Can't bless non-reference value");
10062 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY)) {
10063 if (SvREADONLY(tmpRef))
10064 Perl_croak_no_modify();
10065 if (SvOBJECT(tmpRef)) {
10066 oldstash = SvSTASH(tmpRef);
10069 SvOBJECT_on(tmpRef);
10070 SvUPGRADE(tmpRef, SVt_PVMG);
10071 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10072 SvREFCNT_dec(oldstash);
10074 if(SvSMAGICAL(tmpRef))
10075 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10083 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10084 * as it is after unglobbing it.
10087 PERL_STATIC_INLINE void
10088 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10093 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10095 PERL_ARGS_ASSERT_SV_UNGLOB;
10097 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10099 if (!(flags & SV_COW_DROP_PV))
10100 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10102 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10104 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10105 && HvNAME_get(stash))
10106 mro_method_changed_in(stash);
10107 gp_free(MUTABLE_GV(sv));
10110 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10111 GvSTASH(sv) = NULL;
10114 if (GvNAME_HEK(sv)) {
10115 unshare_hek(GvNAME_HEK(sv));
10117 isGV_with_GP_off(sv);
10119 if(SvTYPE(sv) == SVt_PVGV) {
10120 /* need to keep SvANY(sv) in the right arena */
10121 xpvmg = new_XPVMG();
10122 StructCopy(SvANY(sv), xpvmg, XPVMG);
10123 del_XPVGV(SvANY(sv));
10126 SvFLAGS(sv) &= ~SVTYPEMASK;
10127 SvFLAGS(sv) |= SVt_PVMG;
10130 /* Intentionally not calling any local SET magic, as this isn't so much a
10131 set operation as merely an internal storage change. */
10132 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10133 else sv_setsv_flags(sv, temp, 0);
10135 if ((const GV *)sv == PL_last_in_gv)
10136 PL_last_in_gv = NULL;
10137 else if ((const GV *)sv == PL_statgv)
10142 =for apidoc sv_unref_flags
10144 Unsets the RV status of the SV, and decrements the reference count of
10145 whatever was being referenced by the RV. This can almost be thought of
10146 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10147 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10148 (otherwise the decrementing is conditional on the reference count being
10149 different from one or the reference being a readonly SV).
10156 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10158 SV* const target = SvRV(ref);
10160 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10162 if (SvWEAKREF(ref)) {
10163 sv_del_backref(target, ref);
10164 SvWEAKREF_off(ref);
10165 SvRV_set(ref, NULL);
10168 SvRV_set(ref, NULL);
10170 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10171 assigned to as BEGIN {$a = \"Foo"} will fail. */
10172 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10173 SvREFCNT_dec_NN(target);
10174 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10175 sv_2mortal(target); /* Schedule for freeing later */
10179 =for apidoc sv_untaint
10181 Untaint an SV. Use C<SvTAINTED_off> instead.
10187 Perl_sv_untaint(pTHX_ SV *const sv)
10189 PERL_ARGS_ASSERT_SV_UNTAINT;
10191 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10192 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10199 =for apidoc sv_tainted
10201 Test an SV for taintedness. Use C<SvTAINTED> instead.
10207 Perl_sv_tainted(pTHX_ SV *const sv)
10209 PERL_ARGS_ASSERT_SV_TAINTED;
10211 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10212 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10213 if (mg && (mg->mg_len & 1) )
10220 =for apidoc sv_setpviv
10222 Copies an integer into the given SV, also updating its string value.
10223 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10229 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10231 char buf[TYPE_CHARS(UV)];
10233 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10235 PERL_ARGS_ASSERT_SV_SETPVIV;
10237 sv_setpvn(sv, ptr, ebuf - ptr);
10241 =for apidoc sv_setpviv_mg
10243 Like C<sv_setpviv>, but also handles 'set' magic.
10249 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10251 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10253 sv_setpviv(sv, iv);
10257 #if defined(PERL_IMPLICIT_CONTEXT)
10259 /* pTHX_ magic can't cope with varargs, so this is a no-context
10260 * version of the main function, (which may itself be aliased to us).
10261 * Don't access this version directly.
10265 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10270 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10272 va_start(args, pat);
10273 sv_vsetpvf(sv, pat, &args);
10277 /* pTHX_ magic can't cope with varargs, so this is a no-context
10278 * version of the main function, (which may itself be aliased to us).
10279 * Don't access this version directly.
10283 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10288 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10290 va_start(args, pat);
10291 sv_vsetpvf_mg(sv, pat, &args);
10297 =for apidoc sv_setpvf
10299 Works like C<sv_catpvf> but copies the text into the SV instead of
10300 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10306 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10310 PERL_ARGS_ASSERT_SV_SETPVF;
10312 va_start(args, pat);
10313 sv_vsetpvf(sv, pat, &args);
10318 =for apidoc sv_vsetpvf
10320 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10321 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10323 Usually used via its frontend C<sv_setpvf>.
10329 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10331 PERL_ARGS_ASSERT_SV_VSETPVF;
10333 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10337 =for apidoc sv_setpvf_mg
10339 Like C<sv_setpvf>, but also handles 'set' magic.
10345 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10349 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10351 va_start(args, pat);
10352 sv_vsetpvf_mg(sv, pat, &args);
10357 =for apidoc sv_vsetpvf_mg
10359 Like C<sv_vsetpvf>, but also handles 'set' magic.
10361 Usually used via its frontend C<sv_setpvf_mg>.
10367 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10369 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10371 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10375 #if defined(PERL_IMPLICIT_CONTEXT)
10377 /* pTHX_ magic can't cope with varargs, so this is a no-context
10378 * version of the main function, (which may itself be aliased to us).
10379 * Don't access this version directly.
10383 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10388 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10390 va_start(args, pat);
10391 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10395 /* pTHX_ magic can't cope with varargs, so this is a no-context
10396 * version of the main function, (which may itself be aliased to us).
10397 * Don't access this version directly.
10401 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10406 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10408 va_start(args, pat);
10409 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10416 =for apidoc sv_catpvf
10418 Processes its arguments like C<sprintf> and appends the formatted
10419 output to an SV. If the appended data contains "wide" characters
10420 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10421 and characters >255 formatted with %c), the original SV might get
10422 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10423 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10424 valid UTF-8; if the original SV was bytes, the pattern should be too.
10429 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10433 PERL_ARGS_ASSERT_SV_CATPVF;
10435 va_start(args, pat);
10436 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10441 =for apidoc sv_vcatpvf
10443 Processes its arguments like C<vsprintf> and appends the formatted output
10444 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10446 Usually used via its frontend C<sv_catpvf>.
10452 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10454 PERL_ARGS_ASSERT_SV_VCATPVF;
10456 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10460 =for apidoc sv_catpvf_mg
10462 Like C<sv_catpvf>, but also handles 'set' magic.
10468 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10472 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10474 va_start(args, pat);
10475 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10481 =for apidoc sv_vcatpvf_mg
10483 Like C<sv_vcatpvf>, but also handles 'set' magic.
10485 Usually used via its frontend C<sv_catpvf_mg>.
10491 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10493 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10495 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10500 =for apidoc sv_vsetpvfn
10502 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10505 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10511 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10512 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10514 PERL_ARGS_ASSERT_SV_VSETPVFN;
10517 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10522 * Warn of missing argument to sprintf, and then return a defined value
10523 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10526 S_vcatpvfn_missing_argument(pTHX) {
10527 if (ckWARN(WARN_MISSING)) {
10528 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10529 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10536 S_expect_number(pTHX_ char **const pattern)
10541 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10543 switch (**pattern) {
10544 case '1': case '2': case '3':
10545 case '4': case '5': case '6':
10546 case '7': case '8': case '9':
10547 var = *(*pattern)++ - '0';
10548 while (isDIGIT(**pattern)) {
10549 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10551 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10559 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10561 const int neg = nv < 0;
10564 PERL_ARGS_ASSERT_F0CONVERT;
10572 if (uv & 1 && uv == nv)
10573 uv--; /* Round to even */
10575 const unsigned dig = uv % 10;
10577 } while (uv /= 10);
10588 =for apidoc sv_vcatpvfn
10590 =for apidoc sv_vcatpvfn_flags
10592 Processes its arguments like C<vsprintf> and appends the formatted output
10593 to an SV. Uses an array of SVs if the C style variable argument list is
10594 missing (NULL). When running with taint checks enabled, indicates via
10595 C<maybe_tainted> if results are untrustworthy (often due to the use of
10598 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10600 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10605 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10606 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10607 vec_utf8 = DO_UTF8(vecsv);
10609 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10612 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10613 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10615 PERL_ARGS_ASSERT_SV_VCATPVFN;
10617 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10621 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10622 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
10628 const char *patend;
10631 static const char nullstr[] = "(null)";
10633 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
10634 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
10636 /* Times 4: a decimal digit takes more than 3 binary digits.
10637 * NV_DIG: mantissa takes than many decimal digits.
10638 * Plus 32: Playing safe. */
10639 char ebuf[IV_DIG * 4 + NV_DIG + 32];
10640 /* large enough for "%#.#f" --chip */
10641 /* what about long double NVs? --jhi */
10642 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
10644 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
10646 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
10647 PERL_UNUSED_ARG(maybe_tainted);
10649 if (flags & SV_GMAGIC)
10652 /* no matter what, this is a string now */
10653 (void)SvPV_force_nomg(sv, origlen);
10655 /* special-case "", "%s", and "%-p" (SVf - see below) */
10657 if (svmax && ckWARN(WARN_REDUNDANT))
10658 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10659 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10662 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
10663 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10664 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10665 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10668 const char * const s = va_arg(*args, char*);
10669 sv_catpv_nomg(sv, s ? s : nullstr);
10671 else if (svix < svmax) {
10672 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
10673 SvGETMAGIC(*svargs);
10674 sv_catsv_nomg(sv, *svargs);
10677 S_vcatpvfn_missing_argument(aTHX);
10680 if (args && patlen == 3 && pat[0] == '%' &&
10681 pat[1] == '-' && pat[2] == 'p') {
10682 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10683 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10684 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10685 argsv = MUTABLE_SV(va_arg(*args, void*));
10686 sv_catsv_nomg(sv, argsv);
10690 #ifndef USE_LONG_DOUBLE
10691 /* special-case "%.<number>[gf]" */
10692 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
10693 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
10694 unsigned digits = 0;
10698 while (*pp >= '0' && *pp <= '9')
10699 digits = 10 * digits + (*pp++ - '0');
10701 /* XXX: Why do this `svix < svmax` test? Couldn't we just
10702 format the first argument and WARN_REDUNDANT if svmax > 1?
10703 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
10704 if (pp - pat == (int)patlen - 1 && svix < svmax) {
10705 const NV nv = SvNV(*svargs);
10707 /* Add check for digits != 0 because it seems that some
10708 gconverts are buggy in this case, and we don't yet have
10709 a Configure test for this. */
10710 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
10711 /* 0, point, slack */
10712 STORE_LC_NUMERIC_SET_TO_NEEDED();
10713 PERL_UNUSED_RESULT(Gconvert(nv, (int)digits, 0, ebuf));
10714 sv_catpv_nomg(sv, ebuf);
10715 if (*ebuf) /* May return an empty string for digits==0 */
10718 } else if (!digits) {
10721 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
10722 sv_catpvn_nomg(sv, p, l);
10728 #endif /* !USE_LONG_DOUBLE */
10730 if (!args && svix < svmax && DO_UTF8(*svargs))
10733 patend = (char*)pat + patlen;
10734 for (p = (char*)pat; p < patend; p = q) {
10737 bool vectorize = FALSE;
10738 bool vectorarg = FALSE;
10739 bool vec_utf8 = FALSE;
10745 bool has_precis = FALSE;
10747 const I32 osvix = svix;
10748 bool is_utf8 = FALSE; /* is this item utf8? */
10749 #ifdef HAS_LDBL_SPRINTF_BUG
10750 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
10751 with sfio - Allen <allens@cpan.org> */
10752 bool fix_ldbl_sprintf_bug = FALSE;
10756 U8 utf8buf[UTF8_MAXBYTES+1];
10757 STRLEN esignlen = 0;
10759 const char *eptr = NULL;
10760 const char *fmtstart;
10763 const U8 *vecstr = NULL;
10770 /* we need a long double target in case HAS_LONG_DOUBLE but
10771 not USE_LONG_DOUBLE
10773 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE
10781 const char *dotstr = ".";
10782 STRLEN dotstrlen = 1;
10783 I32 efix = 0; /* explicit format parameter index */
10784 I32 ewix = 0; /* explicit width index */
10785 I32 epix = 0; /* explicit precision index */
10786 I32 evix = 0; /* explicit vector index */
10787 bool asterisk = FALSE;
10789 /* echo everything up to the next format specification */
10790 for (q = p; q < patend && *q != '%'; ++q) ;
10792 if (has_utf8 && !pat_utf8)
10793 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
10795 sv_catpvn_nomg(sv, p, q - p);
10804 We allow format specification elements in this order:
10805 \d+\$ explicit format parameter index
10807 v|\*(\d+\$)?v vector with optional (optionally specified) arg
10808 0 flag (as above): repeated to allow "v02"
10809 \d+|\*(\d+\$)? width using optional (optionally specified) arg
10810 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
10812 [%bcdefginopsuxDFOUX] format (mandatory)
10817 As of perl5.9.3, printf format checking is on by default.
10818 Internally, perl uses %p formats to provide an escape to
10819 some extended formatting. This block deals with those
10820 extensions: if it does not match, (char*)q is reset and
10821 the normal format processing code is used.
10823 Currently defined extensions are:
10824 %p include pointer address (standard)
10825 %-p (SVf) include an SV (previously %_)
10826 %-<num>p include an SV with precision <num>
10828 %3p include a HEK with precision of 256
10829 %4p char* preceded by utf8 flag and length
10830 %<num>p (where num is 1 or > 4) reserved for future
10833 Robin Barker 2005-07-14 (but modified since)
10835 %1p (VDf) removed. RMB 2007-10-19
10842 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
10843 /* The argument has already gone through cBOOL, so the cast
10845 is_utf8 = (bool)va_arg(*args, int);
10846 elen = va_arg(*args, UV);
10847 eptr = va_arg(*args, char *);
10848 q += sizeof(UTF8f)-1;
10851 n = expect_number(&q);
10853 if (sv) { /* SVf */
10858 argsv = MUTABLE_SV(va_arg(*args, void*));
10859 eptr = SvPV_const(argsv, elen);
10860 if (DO_UTF8(argsv))
10864 else if (n==2 || n==3) { /* HEKf */
10865 HEK * const hek = va_arg(*args, HEK *);
10866 eptr = HEK_KEY(hek);
10867 elen = HEK_LEN(hek);
10868 if (HEK_UTF8(hek)) is_utf8 = TRUE;
10869 if (n==3) precis = 256, has_precis = TRUE;
10873 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
10874 "internal %%<num>p might conflict with future printf extensions");
10880 if ( (width = expect_number(&q)) ) {
10884 if (!no_redundant_warning)
10885 /* I've forgotten if it's a better
10886 micro-optimization to always set this or to
10887 only set it if it's unset */
10888 no_redundant_warning = TRUE;
10900 if (plus == '+' && *q == ' ') /* '+' over ' ' */
10929 if ( (ewix = expect_number(&q)) )
10938 if ((vectorarg = asterisk)) {
10951 width = expect_number(&q);
10954 if (vectorize && vectorarg) {
10955 /* vectorizing, but not with the default "." */
10957 vecsv = va_arg(*args, SV*);
10959 vecsv = (evix > 0 && evix <= svmax)
10960 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
10962 vecsv = svix < svmax
10963 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
10965 dotstr = SvPV_const(vecsv, dotstrlen);
10966 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
10967 bad with tied or overloaded values that return UTF8. */
10968 if (DO_UTF8(vecsv))
10970 else if (has_utf8) {
10971 vecsv = sv_mortalcopy(vecsv);
10972 sv_utf8_upgrade(vecsv);
10973 dotstr = SvPV_const(vecsv, dotstrlen);
10980 i = va_arg(*args, int);
10982 i = (ewix ? ewix <= svmax : svix < svmax) ?
10983 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
10985 width = (i < 0) ? -i : i;
10995 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
10997 /* XXX: todo, support specified precision parameter */
11001 i = va_arg(*args, int);
11003 i = (ewix ? ewix <= svmax : svix < svmax)
11004 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11006 has_precis = !(i < 0);
11010 while (isDIGIT(*q))
11011 precis = precis * 10 + (*q++ - '0');
11020 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11021 vecsv = svargs[efix ? efix-1 : svix++];
11022 vecstr = (U8*)SvPV_const(vecsv,veclen);
11023 vec_utf8 = DO_UTF8(vecsv);
11025 /* if this is a version object, we need to convert
11026 * back into v-string notation and then let the
11027 * vectorize happen normally
11029 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11030 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11031 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11032 "vector argument not supported with alpha versions");
11035 vecsv = sv_newmortal();
11036 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11038 vecstr = (U8*)SvPV_const(vecsv, veclen);
11039 vec_utf8 = DO_UTF8(vecsv);
11053 case 'I': /* Ix, I32x, and I64x */
11054 # ifdef USE_64_BIT_INT
11055 if (q[1] == '6' && q[2] == '4') {
11061 if (q[1] == '3' && q[2] == '2') {
11065 # ifdef USE_64_BIT_INT
11071 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11083 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11084 if (*q == 'l') { /* lld, llf */
11093 if (*++q == 'h') { /* hhd, hhu */
11122 if (!vectorize && !args) {
11124 const I32 i = efix-1;
11125 argsv = (i >= 0 && i < svmax)
11126 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11128 argsv = (svix >= 0 && svix < svmax)
11129 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11133 switch (c = *q++) {
11140 uv = (args) ? va_arg(*args, int) : SvIV(argsv);
11142 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11144 eptr = (char*)utf8buf;
11145 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11159 eptr = va_arg(*args, char*);
11161 elen = strlen(eptr);
11163 eptr = (char *)nullstr;
11164 elen = sizeof nullstr - 1;
11168 eptr = SvPV_const(argsv, elen);
11169 if (DO_UTF8(argsv)) {
11170 STRLEN old_precis = precis;
11171 if (has_precis && precis < elen) {
11172 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11173 STRLEN p = precis > ulen ? ulen : precis;
11174 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11175 /* sticks at end */
11177 if (width) { /* fudge width (can't fudge elen) */
11178 if (has_precis && precis < elen)
11179 width += precis - old_precis;
11182 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11189 if (has_precis && precis < elen)
11196 if (alt || vectorize)
11198 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11216 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11225 esignbuf[esignlen++] = plus;
11229 case 'c': iv = (char)va_arg(*args, int); break;
11230 case 'h': iv = (short)va_arg(*args, int); break;
11231 case 'l': iv = va_arg(*args, long); break;
11232 case 'V': iv = va_arg(*args, IV); break;
11233 case 'z': iv = va_arg(*args, SSize_t); break;
11234 #ifdef HAS_PTRDIFF_T
11235 case 't': iv = va_arg(*args, ptrdiff_t); break;
11237 default: iv = va_arg(*args, int); break;
11239 case 'j': iv = va_arg(*args, intmax_t); break;
11243 iv = va_arg(*args, Quad_t); break;
11250 IV tiv = SvIV(argsv); /* work around GCC bug #13488 */
11252 case 'c': iv = (char)tiv; break;
11253 case 'h': iv = (short)tiv; break;
11254 case 'l': iv = (long)tiv; break;
11256 default: iv = tiv; break;
11259 iv = (Quad_t)tiv; break;
11265 if ( !vectorize ) /* we already set uv above */
11270 esignbuf[esignlen++] = plus;
11274 esignbuf[esignlen++] = '-';
11318 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11329 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11330 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11331 case 'l': uv = va_arg(*args, unsigned long); break;
11332 case 'V': uv = va_arg(*args, UV); break;
11333 case 'z': uv = va_arg(*args, Size_t); break;
11334 #ifdef HAS_PTRDIFF_T
11335 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11338 case 'j': uv = va_arg(*args, uintmax_t); break;
11340 default: uv = va_arg(*args, unsigned); break;
11343 uv = va_arg(*args, Uquad_t); break;
11350 UV tuv = SvUV(argsv); /* work around GCC bug #13488 */
11352 case 'c': uv = (unsigned char)tuv; break;
11353 case 'h': uv = (unsigned short)tuv; break;
11354 case 'l': uv = (unsigned long)tuv; break;
11356 default: uv = tuv; break;
11359 uv = (Uquad_t)tuv; break;
11368 char *ptr = ebuf + sizeof ebuf;
11369 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11375 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11379 } while (uv >>= 4);
11381 esignbuf[esignlen++] = '0';
11382 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11388 *--ptr = '0' + dig;
11389 } while (uv >>= 3);
11390 if (alt && *ptr != '0')
11396 *--ptr = '0' + dig;
11397 } while (uv >>= 1);
11399 esignbuf[esignlen++] = '0';
11400 esignbuf[esignlen++] = c;
11403 default: /* it had better be ten or less */
11406 *--ptr = '0' + dig;
11407 } while (uv /= base);
11410 elen = (ebuf + sizeof ebuf) - ptr;
11414 zeros = precis - elen;
11415 else if (precis == 0 && elen == 1 && *eptr == '0'
11416 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11419 /* a precision nullifies the 0 flag. */
11426 /* FLOATING POINT */
11429 c = 'f'; /* maybe %F isn't supported here */
11431 case 'e': case 'E':
11433 case 'g': case 'G':
11437 /* This is evil, but floating point is even more evil */
11439 /* for SV-style calling, we can only get NV
11440 for C-style calling, we assume %f is double;
11441 for simplicity we allow any of %Lf, %llf, %qf for long double
11445 #if defined(USE_LONG_DOUBLE)
11449 /* [perl #20339] - we should accept and ignore %lf rather than die */
11453 #if defined(USE_LONG_DOUBLE)
11454 intsize = args ? 0 : 'q';
11458 #if defined(HAS_LONG_DOUBLE)
11471 /* now we need (long double) if intsize == 'q', else (double) */
11473 #if LONG_DOUBLESIZE > DOUBLESIZE
11475 va_arg(*args, long double) :
11476 va_arg(*args, double)
11478 va_arg(*args, double)
11483 /* nv * 0 will be NaN for NaN, +Inf and -Inf, and 0 for anything
11484 else. frexp() has some unspecified behaviour for those three */
11485 if (c != 'e' && c != 'E' && (nv * 0) == 0) {
11487 /* FIXME: if HAS_LONG_DOUBLE but not USE_LONG_DOUBLE this
11488 will cast our (long double) to (double) */
11489 (void)Perl_frexp(nv, &i);
11490 if (i == PERL_INT_MIN)
11491 Perl_die(aTHX_ "panic: frexp");
11493 need = BIT_DIGITS(i);
11495 need += has_precis ? precis : 6; /* known default */
11500 #ifdef HAS_LDBL_SPRINTF_BUG
11501 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11502 with sfio - Allen <allens@cpan.org> */
11505 # define MY_DBL_MAX DBL_MAX
11506 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11507 # if DOUBLESIZE >= 8
11508 # define MY_DBL_MAX 1.7976931348623157E+308L
11510 # define MY_DBL_MAX 3.40282347E+38L
11514 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
11515 # define MY_DBL_MAX_BUG 1L
11517 # define MY_DBL_MAX_BUG MY_DBL_MAX
11521 # define MY_DBL_MIN DBL_MIN
11522 # else /* XXX guessing! -Allen */
11523 # if DOUBLESIZE >= 8
11524 # define MY_DBL_MIN 2.2250738585072014E-308L
11526 # define MY_DBL_MIN 1.17549435E-38L
11530 if ((intsize == 'q') && (c == 'f') &&
11531 ((nv < MY_DBL_MAX_BUG) && (nv > -MY_DBL_MAX_BUG)) &&
11532 (need < DBL_DIG)) {
11533 /* it's going to be short enough that
11534 * long double precision is not needed */
11536 if ((nv <= 0L) && (nv >= -0L))
11537 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
11539 /* would use Perl_fp_class as a double-check but not
11540 * functional on IRIX - see perl.h comments */
11542 if ((nv >= MY_DBL_MIN) || (nv <= -MY_DBL_MIN)) {
11543 /* It's within the range that a double can represent */
11544 #if defined(DBL_MAX) && !defined(DBL_MIN)
11545 if ((nv >= ((long double)1/DBL_MAX)) ||
11546 (nv <= (-(long double)1/DBL_MAX)))
11548 fix_ldbl_sprintf_bug = TRUE;
11551 if (fix_ldbl_sprintf_bug == TRUE) {
11561 # undef MY_DBL_MAX_BUG
11564 #endif /* HAS_LDBL_SPRINTF_BUG */
11566 need += 20; /* fudge factor */
11567 if (PL_efloatsize < need) {
11568 Safefree(PL_efloatbuf);
11569 PL_efloatsize = need + 20; /* more fudge */
11570 Newx(PL_efloatbuf, PL_efloatsize, char);
11571 PL_efloatbuf[0] = '\0';
11574 if ( !(width || left || plus || alt) && fill != '0'
11575 && has_precis && intsize != 'q' ) { /* Shortcuts */
11576 /* See earlier comment about buggy Gconvert when digits,
11578 if ( c == 'g' && precis) {
11579 STORE_LC_NUMERIC_SET_TO_NEEDED();
11580 PERL_UNUSED_RESULT(Gconvert((NV)nv, (int)precis, 0, PL_efloatbuf));
11581 /* May return an empty string for digits==0 */
11582 if (*PL_efloatbuf) {
11583 elen = strlen(PL_efloatbuf);
11584 goto float_converted;
11586 } else if ( c == 'f' && !precis) {
11587 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
11592 char *ptr = ebuf + sizeof ebuf;
11595 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
11596 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
11597 if (intsize == 'q') {
11598 /* Copy the one or more characters in a long double
11599 * format before the 'base' ([efgEFG]) character to
11600 * the format string. */
11601 static char const prifldbl[] = PERL_PRIfldbl;
11602 char const *p = prifldbl + sizeof(prifldbl) - 3;
11603 while (p >= prifldbl) { *--ptr = *p--; }
11608 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11613 do { *--ptr = '0' + (base % 10); } while (base /= 10);
11625 /* No taint. Otherwise we are in the strange situation
11626 * where printf() taints but print($float) doesn't.
11629 STORE_LC_NUMERIC_SET_TO_NEEDED();
11631 /* hopefully the above makes ptr a very constrained format
11632 * that is safe to use, even though it's not literal */
11633 GCC_DIAG_IGNORE(-Wformat-nonliteral);
11634 #if defined(HAS_LONG_DOUBLE)
11635 elen = ((intsize == 'q')
11636 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, nv)
11637 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)nv));
11639 elen = my_sprintf(PL_efloatbuf, ptr, nv);
11644 eptr = PL_efloatbuf;
11646 #ifdef USE_LOCALE_NUMERIC
11647 /* If the decimal point character in the string is UTF-8, make the
11649 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
11650 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
11663 i = SvCUR(sv) - origlen;
11666 case 'c': *(va_arg(*args, char*)) = i; break;
11667 case 'h': *(va_arg(*args, short*)) = i; break;
11668 default: *(va_arg(*args, int*)) = i; break;
11669 case 'l': *(va_arg(*args, long*)) = i; break;
11670 case 'V': *(va_arg(*args, IV*)) = i; break;
11671 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
11672 #ifdef HAS_PTRDIFF_T
11673 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
11676 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
11680 *(va_arg(*args, Quad_t*)) = i; break;
11687 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
11688 continue; /* not "break" */
11695 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
11696 && ckWARN(WARN_PRINTF))
11698 SV * const msg = sv_newmortal();
11699 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
11700 (PL_op->op_type == OP_PRTF) ? "" : "s");
11701 if (fmtstart < patend) {
11702 const char * const fmtend = q < patend ? q : patend;
11704 sv_catpvs(msg, "\"%");
11705 for (f = fmtstart; f < fmtend; f++) {
11707 sv_catpvn_nomg(msg, f, 1);
11709 Perl_sv_catpvf(aTHX_ msg,
11710 "\\%03"UVof, (UV)*f & 0xFF);
11713 sv_catpvs(msg, "\"");
11715 sv_catpvs(msg, "end of string");
11717 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
11720 /* output mangled stuff ... */
11726 /* ... right here, because formatting flags should not apply */
11727 SvGROW(sv, SvCUR(sv) + elen + 1);
11729 Copy(eptr, p, elen, char);
11732 SvCUR_set(sv, p - SvPVX_const(sv));
11734 continue; /* not "break" */
11737 if (is_utf8 != has_utf8) {
11740 sv_utf8_upgrade(sv);
11743 const STRLEN old_elen = elen;
11744 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
11745 sv_utf8_upgrade(nsv);
11746 eptr = SvPVX_const(nsv);
11749 if (width) { /* fudge width (can't fudge elen) */
11750 width += elen - old_elen;
11756 have = esignlen + zeros + elen;
11758 croak_memory_wrap();
11760 need = (have > width ? have : width);
11763 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
11764 croak_memory_wrap();
11765 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
11767 if (esignlen && fill == '0') {
11769 for (i = 0; i < (int)esignlen; i++)
11770 *p++ = esignbuf[i];
11772 if (gap && !left) {
11773 memset(p, fill, gap);
11776 if (esignlen && fill != '0') {
11778 for (i = 0; i < (int)esignlen; i++)
11779 *p++ = esignbuf[i];
11783 for (i = zeros; i; i--)
11787 Copy(eptr, p, elen, char);
11791 memset(p, ' ', gap);
11796 Copy(dotstr, p, dotstrlen, char);
11800 vectorize = FALSE; /* done iterating over vecstr */
11807 SvCUR_set(sv, p - SvPVX_const(sv));
11814 /* Now that we've consumed all our printf format arguments (svix)
11815 * do we have things left on the stack that we didn't use?
11817 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
11818 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11819 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11824 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
11828 /* =========================================================================
11830 =head1 Cloning an interpreter
11834 All the macros and functions in this section are for the private use of
11835 the main function, perl_clone().
11837 The foo_dup() functions make an exact copy of an existing foo thingy.
11838 During the course of a cloning, a hash table is used to map old addresses
11839 to new addresses. The table is created and manipulated with the
11840 ptr_table_* functions.
11842 * =========================================================================*/
11845 #if defined(USE_ITHREADS)
11847 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
11848 #ifndef GpREFCNT_inc
11849 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
11853 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
11854 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
11855 If this changes, please unmerge ss_dup.
11856 Likewise, sv_dup_inc_multiple() relies on this fact. */
11857 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
11858 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
11859 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11860 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
11861 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11862 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
11863 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
11864 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
11865 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
11866 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
11867 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
11868 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
11869 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11871 /* clone a parser */
11874 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
11878 PERL_ARGS_ASSERT_PARSER_DUP;
11883 /* look for it in the table first */
11884 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
11888 /* create anew and remember what it is */
11889 Newxz(parser, 1, yy_parser);
11890 ptr_table_store(PL_ptr_table, proto, parser);
11892 /* XXX these not yet duped */
11893 parser->old_parser = NULL;
11894 parser->stack = NULL;
11896 parser->stack_size = 0;
11897 /* XXX parser->stack->state = 0; */
11899 /* XXX eventually, just Copy() most of the parser struct ? */
11901 parser->lex_brackets = proto->lex_brackets;
11902 parser->lex_casemods = proto->lex_casemods;
11903 parser->lex_brackstack = savepvn(proto->lex_brackstack,
11904 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
11905 parser->lex_casestack = savepvn(proto->lex_casestack,
11906 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
11907 parser->lex_defer = proto->lex_defer;
11908 parser->lex_dojoin = proto->lex_dojoin;
11909 parser->lex_expect = proto->lex_expect;
11910 parser->lex_formbrack = proto->lex_formbrack;
11911 parser->lex_inpat = proto->lex_inpat;
11912 parser->lex_inwhat = proto->lex_inwhat;
11913 parser->lex_op = proto->lex_op;
11914 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
11915 parser->lex_starts = proto->lex_starts;
11916 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
11917 parser->multi_close = proto->multi_close;
11918 parser->multi_open = proto->multi_open;
11919 parser->multi_start = proto->multi_start;
11920 parser->multi_end = proto->multi_end;
11921 parser->preambled = proto->preambled;
11922 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
11923 parser->linestr = sv_dup_inc(proto->linestr, param);
11924 parser->expect = proto->expect;
11925 parser->copline = proto->copline;
11926 parser->last_lop_op = proto->last_lop_op;
11927 parser->lex_state = proto->lex_state;
11928 parser->rsfp = fp_dup(proto->rsfp, '<', param);
11929 /* rsfp_filters entries have fake IoDIRP() */
11930 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
11931 parser->in_my = proto->in_my;
11932 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
11933 parser->error_count = proto->error_count;
11936 parser->linestr = sv_dup_inc(proto->linestr, param);
11939 char * const ols = SvPVX(proto->linestr);
11940 char * const ls = SvPVX(parser->linestr);
11942 parser->bufptr = ls + (proto->bufptr >= ols ?
11943 proto->bufptr - ols : 0);
11944 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
11945 proto->oldbufptr - ols : 0);
11946 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
11947 proto->oldoldbufptr - ols : 0);
11948 parser->linestart = ls + (proto->linestart >= ols ?
11949 proto->linestart - ols : 0);
11950 parser->last_uni = ls + (proto->last_uni >= ols ?
11951 proto->last_uni - ols : 0);
11952 parser->last_lop = ls + (proto->last_lop >= ols ?
11953 proto->last_lop - ols : 0);
11955 parser->bufend = ls + SvCUR(parser->linestr);
11958 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
11961 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
11962 Copy(proto->nexttype, parser->nexttype, 5, I32);
11963 parser->nexttoke = proto->nexttoke;
11965 /* XXX should clone saved_curcop here, but we aren't passed
11966 * proto_perl; so do it in perl_clone_using instead */
11972 /* duplicate a file handle */
11975 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
11979 PERL_ARGS_ASSERT_FP_DUP;
11980 PERL_UNUSED_ARG(type);
11983 return (PerlIO*)NULL;
11985 /* look for it in the table first */
11986 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
11990 /* create anew and remember what it is */
11991 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
11992 ptr_table_store(PL_ptr_table, fp, ret);
11996 /* duplicate a directory handle */
11999 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12003 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12005 const Direntry_t *dirent;
12006 char smallbuf[256];
12012 PERL_UNUSED_CONTEXT;
12013 PERL_ARGS_ASSERT_DIRP_DUP;
12018 /* look for it in the table first */
12019 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12023 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12025 PERL_UNUSED_ARG(param);
12029 /* open the current directory (so we can switch back) */
12030 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12032 /* chdir to our dir handle and open the present working directory */
12033 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12034 PerlDir_close(pwd);
12035 return (DIR *)NULL;
12037 /* Now we should have two dir handles pointing to the same dir. */
12039 /* Be nice to the calling code and chdir back to where we were. */
12040 /* XXX If this fails, then what? */
12041 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12043 /* We have no need of the pwd handle any more. */
12044 PerlDir_close(pwd);
12047 # define d_namlen(d) (d)->d_namlen
12049 # define d_namlen(d) strlen((d)->d_name)
12051 /* Iterate once through dp, to get the file name at the current posi-
12052 tion. Then step back. */
12053 pos = PerlDir_tell(dp);
12054 if ((dirent = PerlDir_read(dp))) {
12055 len = d_namlen(dirent);
12056 if (len <= sizeof smallbuf) name = smallbuf;
12057 else Newx(name, len, char);
12058 Move(dirent->d_name, name, len, char);
12060 PerlDir_seek(dp, pos);
12062 /* Iterate through the new dir handle, till we find a file with the
12064 if (!dirent) /* just before the end */
12066 pos = PerlDir_tell(ret);
12067 if (PerlDir_read(ret)) continue; /* not there yet */
12068 PerlDir_seek(ret, pos); /* step back */
12072 const long pos0 = PerlDir_tell(ret);
12074 pos = PerlDir_tell(ret);
12075 if ((dirent = PerlDir_read(ret))) {
12076 if (len == (STRLEN)d_namlen(dirent)
12077 && memEQ(name, dirent->d_name, len)) {
12079 PerlDir_seek(ret, pos); /* step back */
12082 /* else we are not there yet; keep iterating */
12084 else { /* This is not meant to happen. The best we can do is
12085 reset the iterator to the beginning. */
12086 PerlDir_seek(ret, pos0);
12093 if (name && name != smallbuf)
12098 ret = win32_dirp_dup(dp, param);
12101 /* pop it in the pointer table */
12103 ptr_table_store(PL_ptr_table, dp, ret);
12108 /* duplicate a typeglob */
12111 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12115 PERL_ARGS_ASSERT_GP_DUP;
12119 /* look for it in the table first */
12120 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12124 /* create anew and remember what it is */
12126 ptr_table_store(PL_ptr_table, gp, ret);
12129 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12130 on Newxz() to do this for us. */
12131 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12132 ret->gp_io = io_dup_inc(gp->gp_io, param);
12133 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12134 ret->gp_av = av_dup_inc(gp->gp_av, param);
12135 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12136 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12137 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12138 ret->gp_cvgen = gp->gp_cvgen;
12139 ret->gp_line = gp->gp_line;
12140 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12144 /* duplicate a chain of magic */
12147 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12149 MAGIC *mgret = NULL;
12150 MAGIC **mgprev_p = &mgret;
12152 PERL_ARGS_ASSERT_MG_DUP;
12154 for (; mg; mg = mg->mg_moremagic) {
12157 if ((param->flags & CLONEf_JOIN_IN)
12158 && mg->mg_type == PERL_MAGIC_backref)
12159 /* when joining, we let the individual SVs add themselves to
12160 * backref as needed. */
12163 Newx(nmg, 1, MAGIC);
12165 mgprev_p = &(nmg->mg_moremagic);
12167 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12168 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12169 from the original commit adding Perl_mg_dup() - revision 4538.
12170 Similarly there is the annotation "XXX random ptr?" next to the
12171 assignment to nmg->mg_ptr. */
12174 /* FIXME for plugins
12175 if (nmg->mg_type == PERL_MAGIC_qr) {
12176 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12180 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12181 ? nmg->mg_type == PERL_MAGIC_backref
12182 /* The backref AV has its reference
12183 * count deliberately bumped by 1 */
12184 ? SvREFCNT_inc(av_dup_inc((const AV *)
12185 nmg->mg_obj, param))
12186 : sv_dup_inc(nmg->mg_obj, param)
12187 : sv_dup(nmg->mg_obj, param);
12189 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12190 if (nmg->mg_len > 0) {
12191 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12192 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12193 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12195 AMT * const namtp = (AMT*)nmg->mg_ptr;
12196 sv_dup_inc_multiple((SV**)(namtp->table),
12197 (SV**)(namtp->table), NofAMmeth, param);
12200 else if (nmg->mg_len == HEf_SVKEY)
12201 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12203 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12204 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12210 #endif /* USE_ITHREADS */
12212 struct ptr_tbl_arena {
12213 struct ptr_tbl_arena *next;
12214 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12217 /* create a new pointer-mapping table */
12220 Perl_ptr_table_new(pTHX)
12223 PERL_UNUSED_CONTEXT;
12225 Newx(tbl, 1, PTR_TBL_t);
12226 tbl->tbl_max = 511;
12227 tbl->tbl_items = 0;
12228 tbl->tbl_arena = NULL;
12229 tbl->tbl_arena_next = NULL;
12230 tbl->tbl_arena_end = NULL;
12231 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12235 #define PTR_TABLE_HASH(ptr) \
12236 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12238 /* map an existing pointer using a table */
12240 STATIC PTR_TBL_ENT_t *
12241 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12243 PTR_TBL_ENT_t *tblent;
12244 const UV hash = PTR_TABLE_HASH(sv);
12246 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12248 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12249 for (; tblent; tblent = tblent->next) {
12250 if (tblent->oldval == sv)
12257 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12259 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12261 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12262 PERL_UNUSED_CONTEXT;
12264 return tblent ? tblent->newval : NULL;
12267 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12268 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12269 * the core's typical use of ptr_tables in thread cloning. */
12272 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12274 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12276 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12277 PERL_UNUSED_CONTEXT;
12280 tblent->newval = newsv;
12282 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12284 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12285 struct ptr_tbl_arena *new_arena;
12287 Newx(new_arena, 1, struct ptr_tbl_arena);
12288 new_arena->next = tbl->tbl_arena;
12289 tbl->tbl_arena = new_arena;
12290 tbl->tbl_arena_next = new_arena->array;
12291 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
12294 tblent = tbl->tbl_arena_next++;
12296 tblent->oldval = oldsv;
12297 tblent->newval = newsv;
12298 tblent->next = tbl->tbl_ary[entry];
12299 tbl->tbl_ary[entry] = tblent;
12301 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12302 ptr_table_split(tbl);
12306 /* double the hash bucket size of an existing ptr table */
12309 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
12311 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
12312 const UV oldsize = tbl->tbl_max + 1;
12313 UV newsize = oldsize * 2;
12316 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
12317 PERL_UNUSED_CONTEXT;
12319 Renew(ary, newsize, PTR_TBL_ENT_t*);
12320 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
12321 tbl->tbl_max = --newsize;
12322 tbl->tbl_ary = ary;
12323 for (i=0; i < oldsize; i++, ary++) {
12324 PTR_TBL_ENT_t **entp = ary;
12325 PTR_TBL_ENT_t *ent = *ary;
12326 PTR_TBL_ENT_t **curentp;
12329 curentp = ary + oldsize;
12331 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
12333 ent->next = *curentp;
12343 /* remove all the entries from a ptr table */
12344 /* Deprecated - will be removed post 5.14 */
12347 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
12349 PERL_UNUSED_CONTEXT;
12350 if (tbl && tbl->tbl_items) {
12351 struct ptr_tbl_arena *arena = tbl->tbl_arena;
12353 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
12356 struct ptr_tbl_arena *next = arena->next;
12362 tbl->tbl_items = 0;
12363 tbl->tbl_arena = NULL;
12364 tbl->tbl_arena_next = NULL;
12365 tbl->tbl_arena_end = NULL;
12369 /* clear and free a ptr table */
12372 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
12374 struct ptr_tbl_arena *arena;
12376 PERL_UNUSED_CONTEXT;
12382 arena = tbl->tbl_arena;
12385 struct ptr_tbl_arena *next = arena->next;
12391 Safefree(tbl->tbl_ary);
12395 #if defined(USE_ITHREADS)
12398 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
12400 PERL_ARGS_ASSERT_RVPV_DUP;
12402 assert(!isREGEXP(sstr));
12404 if (SvWEAKREF(sstr)) {
12405 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
12406 if (param->flags & CLONEf_JOIN_IN) {
12407 /* if joining, we add any back references individually rather
12408 * than copying the whole backref array */
12409 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
12413 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
12415 else if (SvPVX_const(sstr)) {
12416 /* Has something there */
12418 /* Normal PV - clone whole allocated space */
12419 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
12420 /* sstr may not be that normal, but actually copy on write.
12421 But we are a true, independent SV, so: */
12425 /* Special case - not normally malloced for some reason */
12426 if (isGV_with_GP(sstr)) {
12427 /* Don't need to do anything here. */
12429 else if ((SvIsCOW(sstr))) {
12430 /* A "shared" PV - clone it as "shared" PV */
12432 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
12436 /* Some other special case - random pointer */
12437 SvPV_set(dstr, (char *) SvPVX_const(sstr));
12442 /* Copy the NULL */
12443 SvPV_set(dstr, NULL);
12447 /* duplicate a list of SVs. source and dest may point to the same memory. */
12449 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
12450 SSize_t items, CLONE_PARAMS *const param)
12452 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
12454 while (items-- > 0) {
12455 *dest++ = sv_dup_inc(*source++, param);
12461 /* duplicate an SV of any type (including AV, HV etc) */
12464 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12469 PERL_ARGS_ASSERT_SV_DUP_COMMON;
12471 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
12472 #ifdef DEBUG_LEAKING_SCALARS_ABORT
12477 /* look for it in the table first */
12478 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
12482 if(param->flags & CLONEf_JOIN_IN) {
12483 /** We are joining here so we don't want do clone
12484 something that is bad **/
12485 if (SvTYPE(sstr) == SVt_PVHV) {
12486 const HEK * const hvname = HvNAME_HEK(sstr);
12488 /** don't clone stashes if they already exist **/
12489 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12490 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
12491 ptr_table_store(PL_ptr_table, sstr, dstr);
12495 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
12496 HV *stash = GvSTASH(sstr);
12497 const HEK * hvname;
12498 if (stash && (hvname = HvNAME_HEK(stash))) {
12499 /** don't clone GVs if they already exist **/
12501 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12502 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
12504 stash, GvNAME(sstr),
12510 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
12511 ptr_table_store(PL_ptr_table, sstr, *svp);
12518 /* create anew and remember what it is */
12521 #ifdef DEBUG_LEAKING_SCALARS
12522 dstr->sv_debug_optype = sstr->sv_debug_optype;
12523 dstr->sv_debug_line = sstr->sv_debug_line;
12524 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
12525 dstr->sv_debug_parent = (SV*)sstr;
12526 FREE_SV_DEBUG_FILE(dstr);
12527 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
12530 ptr_table_store(PL_ptr_table, sstr, dstr);
12533 SvFLAGS(dstr) = SvFLAGS(sstr);
12534 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
12535 SvREFCNT(dstr) = 0; /* must be before any other dups! */
12538 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
12539 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
12540 (void*)PL_watch_pvx, SvPVX_const(sstr));
12543 /* don't clone objects whose class has asked us not to */
12544 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
12549 switch (SvTYPE(sstr)) {
12551 SvANY(dstr) = NULL;
12554 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
12556 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12558 SvIV_set(dstr, SvIVX(sstr));
12562 SvANY(dstr) = new_XNV();
12563 SvNV_set(dstr, SvNVX(sstr));
12567 /* These are all the types that need complex bodies allocating. */
12569 const svtype sv_type = SvTYPE(sstr);
12570 const struct body_details *const sv_type_details
12571 = bodies_by_type + sv_type;
12575 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
12591 assert(sv_type_details->body_size);
12592 if (sv_type_details->arena) {
12593 new_body_inline(new_body, sv_type);
12595 = (void*)((char*)new_body - sv_type_details->offset);
12597 new_body = new_NOARENA(sv_type_details);
12601 SvANY(dstr) = new_body;
12604 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
12605 ((char*)SvANY(dstr)) + sv_type_details->offset,
12606 sv_type_details->copy, char);
12608 Copy(((char*)SvANY(sstr)),
12609 ((char*)SvANY(dstr)),
12610 sv_type_details->body_size + sv_type_details->offset, char);
12613 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
12614 && !isGV_with_GP(dstr)
12616 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
12617 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
12619 /* The Copy above means that all the source (unduplicated) pointers
12620 are now in the destination. We can check the flags and the
12621 pointers in either, but it's possible that there's less cache
12622 missing by always going for the destination.
12623 FIXME - instrument and check that assumption */
12624 if (sv_type >= SVt_PVMG) {
12625 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
12626 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
12627 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
12629 } else if (SvMAGIC(dstr))
12630 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
12631 if (SvOBJECT(dstr) && SvSTASH(dstr))
12632 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
12633 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
12636 /* The cast silences a GCC warning about unhandled types. */
12637 switch ((int)sv_type) {
12648 /* FIXME for plugins */
12649 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
12650 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
12653 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
12654 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
12655 LvTARG(dstr) = dstr;
12656 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
12657 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
12659 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
12660 if (isREGEXP(sstr)) goto duprex;
12662 /* non-GP case already handled above */
12663 if(isGV_with_GP(sstr)) {
12664 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
12665 /* Don't call sv_add_backref here as it's going to be
12666 created as part of the magic cloning of the symbol
12667 table--unless this is during a join and the stash
12668 is not actually being cloned. */
12669 /* Danger Will Robinson - GvGP(dstr) isn't initialised
12670 at the point of this comment. */
12671 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
12672 if (param->flags & CLONEf_JOIN_IN)
12673 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
12674 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
12675 (void)GpREFCNT_inc(GvGP(dstr));
12679 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
12680 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
12681 /* I have no idea why fake dirp (rsfps)
12682 should be treated differently but otherwise
12683 we end up with leaks -- sky*/
12684 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
12685 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
12686 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
12688 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
12689 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
12690 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
12691 if (IoDIRP(dstr)) {
12692 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
12695 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
12697 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
12699 if (IoOFP(dstr) == IoIFP(sstr))
12700 IoOFP(dstr) = IoIFP(dstr);
12702 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
12703 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
12704 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
12705 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
12708 /* avoid cloning an empty array */
12709 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
12710 SV **dst_ary, **src_ary;
12711 SSize_t items = AvFILLp((const AV *)sstr) + 1;
12713 src_ary = AvARRAY((const AV *)sstr);
12714 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
12715 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
12716 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
12717 AvALLOC((const AV *)dstr) = dst_ary;
12718 if (AvREAL((const AV *)sstr)) {
12719 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
12723 while (items-- > 0)
12724 *dst_ary++ = sv_dup(*src_ary++, param);
12726 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
12727 while (items-- > 0) {
12728 *dst_ary++ = &PL_sv_undef;
12732 AvARRAY(MUTABLE_AV(dstr)) = NULL;
12733 AvALLOC((const AV *)dstr) = (SV**)NULL;
12734 AvMAX( (const AV *)dstr) = -1;
12735 AvFILLp((const AV *)dstr) = -1;
12739 if (HvARRAY((const HV *)sstr)) {
12741 const bool sharekeys = !!HvSHAREKEYS(sstr);
12742 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
12743 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
12745 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
12746 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
12748 HvARRAY(dstr) = (HE**)darray;
12749 while (i <= sxhv->xhv_max) {
12750 const HE * const source = HvARRAY(sstr)[i];
12751 HvARRAY(dstr)[i] = source
12752 ? he_dup(source, sharekeys, param) : 0;
12756 const struct xpvhv_aux * const saux = HvAUX(sstr);
12757 struct xpvhv_aux * const daux = HvAUX(dstr);
12758 /* This flag isn't copied. */
12761 if (saux->xhv_name_count) {
12762 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
12764 = saux->xhv_name_count < 0
12765 ? -saux->xhv_name_count
12766 : saux->xhv_name_count;
12767 HEK **shekp = sname + count;
12769 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
12770 dhekp = daux->xhv_name_u.xhvnameu_names + count;
12771 while (shekp-- > sname) {
12773 *dhekp = hek_dup(*shekp, param);
12777 daux->xhv_name_u.xhvnameu_name
12778 = hek_dup(saux->xhv_name_u.xhvnameu_name,
12781 daux->xhv_name_count = saux->xhv_name_count;
12783 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
12784 daux->xhv_aux_flags = saux->xhv_aux_flags;
12785 #ifdef PERL_HASH_RANDOMIZE_KEYS
12786 daux->xhv_rand = saux->xhv_rand;
12787 daux->xhv_last_rand = saux->xhv_last_rand;
12789 daux->xhv_riter = saux->xhv_riter;
12790 daux->xhv_eiter = saux->xhv_eiter
12791 ? he_dup(saux->xhv_eiter,
12792 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
12793 /* backref array needs refcnt=2; see sv_add_backref */
12794 daux->xhv_backreferences =
12795 (param->flags & CLONEf_JOIN_IN)
12796 /* when joining, we let the individual GVs and
12797 * CVs add themselves to backref as
12798 * needed. This avoids pulling in stuff
12799 * that isn't required, and simplifies the
12800 * case where stashes aren't cloned back
12801 * if they already exist in the parent
12804 : saux->xhv_backreferences
12805 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
12806 ? MUTABLE_AV(SvREFCNT_inc(
12807 sv_dup_inc((const SV *)
12808 saux->xhv_backreferences, param)))
12809 : MUTABLE_AV(sv_dup((const SV *)
12810 saux->xhv_backreferences, param))
12813 daux->xhv_mro_meta = saux->xhv_mro_meta
12814 ? mro_meta_dup(saux->xhv_mro_meta, param)
12817 /* Record stashes for possible cloning in Perl_clone(). */
12819 av_push(param->stashes, dstr);
12823 HvARRAY(MUTABLE_HV(dstr)) = NULL;
12826 if (!(param->flags & CLONEf_COPY_STACKS)) {
12831 /* NOTE: not refcounted */
12832 SvANY(MUTABLE_CV(dstr))->xcv_stash =
12833 hv_dup(CvSTASH(dstr), param);
12834 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
12835 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
12836 if (!CvISXSUB(dstr)) {
12838 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
12840 CvSLABBED_off(dstr);
12841 } else if (CvCONST(dstr)) {
12842 CvXSUBANY(dstr).any_ptr =
12843 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
12845 assert(!CvSLABBED(dstr));
12846 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
12848 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
12849 share_hek_hek(CvNAME_HEK((CV *)sstr));
12850 /* don't dup if copying back - CvGV isn't refcounted, so the
12851 * duped GV may never be freed. A bit of a hack! DAPM */
12853 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
12855 ? gv_dup_inc(CvGV(sstr), param)
12856 : (param->flags & CLONEf_JOIN_IN)
12858 : gv_dup(CvGV(sstr), param);
12860 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
12862 CvWEAKOUTSIDE(sstr)
12863 ? cv_dup( CvOUTSIDE(dstr), param)
12864 : cv_dup_inc(CvOUTSIDE(dstr), param);
12874 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12876 PERL_ARGS_ASSERT_SV_DUP_INC;
12877 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
12881 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12883 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
12884 PERL_ARGS_ASSERT_SV_DUP;
12886 /* Track every SV that (at least initially) had a reference count of 0.
12887 We need to do this by holding an actual reference to it in this array.
12888 If we attempt to cheat, turn AvREAL_off(), and store only pointers
12889 (akin to the stashes hash, and the perl stack), we come unstuck if
12890 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
12891 thread) is manipulated in a CLONE method, because CLONE runs before the
12892 unreferenced array is walked to find SVs still with SvREFCNT() == 0
12893 (and fix things up by giving each a reference via the temps stack).
12894 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
12895 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
12896 before the walk of unreferenced happens and a reference to that is SV
12897 added to the temps stack. At which point we have the same SV considered
12898 to be in use, and free to be re-used. Not good.
12900 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
12901 assert(param->unreferenced);
12902 av_push(param->unreferenced, SvREFCNT_inc(dstr));
12908 /* duplicate a context */
12911 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
12913 PERL_CONTEXT *ncxs;
12915 PERL_ARGS_ASSERT_CX_DUP;
12918 return (PERL_CONTEXT*)NULL;
12920 /* look for it in the table first */
12921 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
12925 /* create anew and remember what it is */
12926 Newx(ncxs, max + 1, PERL_CONTEXT);
12927 ptr_table_store(PL_ptr_table, cxs, ncxs);
12928 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
12931 PERL_CONTEXT * const ncx = &ncxs[ix];
12932 if (CxTYPE(ncx) == CXt_SUBST) {
12933 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
12936 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
12937 switch (CxTYPE(ncx)) {
12939 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
12940 ? cv_dup_inc(ncx->blk_sub.cv, param)
12941 : cv_dup(ncx->blk_sub.cv,param));
12942 if(CxHASARGS(ncx)){
12943 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
12944 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
12946 ncx->blk_sub.argarray = NULL;
12947 ncx->blk_sub.savearray = NULL;
12949 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
12950 ncx->blk_sub.oldcomppad);
12953 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
12955 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
12956 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
12958 case CXt_LOOP_LAZYSV:
12959 ncx->blk_loop.state_u.lazysv.end
12960 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
12961 /* We are taking advantage of av_dup_inc and sv_dup_inc
12962 actually being the same function, and order equivalence of
12964 We can assert the later [but only at run time :-(] */
12965 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
12966 (void *) &ncx->blk_loop.state_u.lazysv.cur);
12968 ncx->blk_loop.state_u.ary.ary
12969 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
12970 case CXt_LOOP_LAZYIV:
12971 case CXt_LOOP_PLAIN:
12972 if (CxPADLOOP(ncx)) {
12973 ncx->blk_loop.itervar_u.oldcomppad
12974 = (PAD*)ptr_table_fetch(PL_ptr_table,
12975 ncx->blk_loop.itervar_u.oldcomppad);
12977 ncx->blk_loop.itervar_u.gv
12978 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
12983 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
12984 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
12985 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13000 /* duplicate a stack info structure */
13003 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13007 PERL_ARGS_ASSERT_SI_DUP;
13010 return (PERL_SI*)NULL;
13012 /* look for it in the table first */
13013 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13017 /* create anew and remember what it is */
13018 Newxz(nsi, 1, PERL_SI);
13019 ptr_table_store(PL_ptr_table, si, nsi);
13021 nsi->si_stack = av_dup_inc(si->si_stack, param);
13022 nsi->si_cxix = si->si_cxix;
13023 nsi->si_cxmax = si->si_cxmax;
13024 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13025 nsi->si_type = si->si_type;
13026 nsi->si_prev = si_dup(si->si_prev, param);
13027 nsi->si_next = si_dup(si->si_next, param);
13028 nsi->si_markoff = si->si_markoff;
13033 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13034 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13035 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13036 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13037 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13038 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13039 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13040 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13041 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13042 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13043 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13044 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13045 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13046 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13047 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13048 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13051 #define pv_dup_inc(p) SAVEPV(p)
13052 #define pv_dup(p) SAVEPV(p)
13053 #define svp_dup_inc(p,pp) any_dup(p,pp)
13055 /* map any object to the new equivent - either something in the
13056 * ptr table, or something in the interpreter structure
13060 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13064 PERL_ARGS_ASSERT_ANY_DUP;
13067 return (void*)NULL;
13069 /* look for it in the table first */
13070 ret = ptr_table_fetch(PL_ptr_table, v);
13074 /* see if it is part of the interpreter structure */
13075 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13076 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13084 /* duplicate the save stack */
13087 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13090 ANY * const ss = proto_perl->Isavestack;
13091 const I32 max = proto_perl->Isavestack_max;
13092 I32 ix = proto_perl->Isavestack_ix;
13105 void (*dptr) (void*);
13106 void (*dxptr) (pTHX_ void*);
13108 PERL_ARGS_ASSERT_SS_DUP;
13110 Newxz(nss, max, ANY);
13113 const UV uv = POPUV(ss,ix);
13114 const U8 type = (U8)uv & SAVE_MASK;
13116 TOPUV(nss,ix) = uv;
13118 case SAVEt_CLEARSV:
13119 case SAVEt_CLEARPADRANGE:
13121 case SAVEt_HELEM: /* hash element */
13122 sv = (const SV *)POPPTR(ss,ix);
13123 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13125 case SAVEt_ITEM: /* normal string */
13126 case SAVEt_GVSV: /* scalar slot in GV */
13127 case SAVEt_SV: /* scalar reference */
13128 sv = (const SV *)POPPTR(ss,ix);
13129 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13132 case SAVEt_MORTALIZESV:
13133 case SAVEt_READONLY_OFF:
13134 sv = (const SV *)POPPTR(ss,ix);
13135 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13137 case SAVEt_SHARED_PVREF: /* char* in shared space */
13138 c = (char*)POPPTR(ss,ix);
13139 TOPPTR(nss,ix) = savesharedpv(c);
13140 ptr = POPPTR(ss,ix);
13141 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13143 case SAVEt_GENERIC_SVREF: /* generic sv */
13144 case SAVEt_SVREF: /* scalar reference */
13145 sv = (const SV *)POPPTR(ss,ix);
13146 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13147 ptr = POPPTR(ss,ix);
13148 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13150 case SAVEt_GVSLOT: /* any slot in GV */
13151 sv = (const SV *)POPPTR(ss,ix);
13152 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13153 ptr = POPPTR(ss,ix);
13154 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13155 sv = (const SV *)POPPTR(ss,ix);
13156 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13158 case SAVEt_HV: /* hash reference */
13159 case SAVEt_AV: /* array reference */
13160 sv = (const SV *) POPPTR(ss,ix);
13161 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13163 case SAVEt_COMPPAD:
13165 sv = (const SV *) POPPTR(ss,ix);
13166 TOPPTR(nss,ix) = sv_dup(sv, param);
13168 case SAVEt_INT: /* int reference */
13169 ptr = POPPTR(ss,ix);
13170 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13171 intval = (int)POPINT(ss,ix);
13172 TOPINT(nss,ix) = intval;
13174 case SAVEt_LONG: /* long reference */
13175 ptr = POPPTR(ss,ix);
13176 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13177 longval = (long)POPLONG(ss,ix);
13178 TOPLONG(nss,ix) = longval;
13180 case SAVEt_I32: /* I32 reference */
13181 ptr = POPPTR(ss,ix);
13182 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13184 TOPINT(nss,ix) = i;
13186 case SAVEt_IV: /* IV reference */
13187 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13188 ptr = POPPTR(ss,ix);
13189 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13191 TOPIV(nss,ix) = iv;
13193 case SAVEt_HPTR: /* HV* reference */
13194 case SAVEt_APTR: /* AV* reference */
13195 case SAVEt_SPTR: /* SV* reference */
13196 ptr = POPPTR(ss,ix);
13197 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13198 sv = (const SV *)POPPTR(ss,ix);
13199 TOPPTR(nss,ix) = sv_dup(sv, param);
13201 case SAVEt_VPTR: /* random* reference */
13202 ptr = POPPTR(ss,ix);
13203 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13205 case SAVEt_INT_SMALL:
13206 case SAVEt_I32_SMALL:
13207 case SAVEt_I16: /* I16 reference */
13208 case SAVEt_I8: /* I8 reference */
13210 ptr = POPPTR(ss,ix);
13211 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13213 case SAVEt_GENERIC_PVREF: /* generic char* */
13214 case SAVEt_PPTR: /* char* reference */
13215 ptr = POPPTR(ss,ix);
13216 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13217 c = (char*)POPPTR(ss,ix);
13218 TOPPTR(nss,ix) = pv_dup(c);
13220 case SAVEt_GP: /* scalar reference */
13221 gp = (GP*)POPPTR(ss,ix);
13222 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13223 (void)GpREFCNT_inc(gp);
13224 gv = (const GV *)POPPTR(ss,ix);
13225 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13228 ptr = POPPTR(ss,ix);
13229 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13230 /* these are assumed to be refcounted properly */
13232 switch (((OP*)ptr)->op_type) {
13234 case OP_LEAVESUBLV:
13238 case OP_LEAVEWRITE:
13239 TOPPTR(nss,ix) = ptr;
13242 (void) OpREFCNT_inc(o);
13246 TOPPTR(nss,ix) = NULL;
13251 TOPPTR(nss,ix) = NULL;
13253 case SAVEt_FREECOPHH:
13254 ptr = POPPTR(ss,ix);
13255 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13257 case SAVEt_ADELETE:
13258 av = (const AV *)POPPTR(ss,ix);
13259 TOPPTR(nss,ix) = av_dup_inc(av, param);
13261 TOPINT(nss,ix) = i;
13264 hv = (const HV *)POPPTR(ss,ix);
13265 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13267 TOPINT(nss,ix) = i;
13270 c = (char*)POPPTR(ss,ix);
13271 TOPPTR(nss,ix) = pv_dup_inc(c);
13273 case SAVEt_STACK_POS: /* Position on Perl stack */
13275 TOPINT(nss,ix) = i;
13277 case SAVEt_DESTRUCTOR:
13278 ptr = POPPTR(ss,ix);
13279 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13280 dptr = POPDPTR(ss,ix);
13281 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13282 any_dup(FPTR2DPTR(void *, dptr),
13285 case SAVEt_DESTRUCTOR_X:
13286 ptr = POPPTR(ss,ix);
13287 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13288 dxptr = POPDXPTR(ss,ix);
13289 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13290 any_dup(FPTR2DPTR(void *, dxptr),
13293 case SAVEt_REGCONTEXT:
13295 ix -= uv >> SAVE_TIGHT_SHIFT;
13297 case SAVEt_AELEM: /* array element */
13298 sv = (const SV *)POPPTR(ss,ix);
13299 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13301 TOPINT(nss,ix) = i;
13302 av = (const AV *)POPPTR(ss,ix);
13303 TOPPTR(nss,ix) = av_dup_inc(av, param);
13306 ptr = POPPTR(ss,ix);
13307 TOPPTR(nss,ix) = ptr;
13310 ptr = POPPTR(ss,ix);
13311 ptr = cophh_copy((COPHH*)ptr);
13312 TOPPTR(nss,ix) = ptr;
13314 TOPINT(nss,ix) = i;
13315 if (i & HINT_LOCALIZE_HH) {
13316 hv = (const HV *)POPPTR(ss,ix);
13317 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13320 case SAVEt_PADSV_AND_MORTALIZE:
13321 longval = (long)POPLONG(ss,ix);
13322 TOPLONG(nss,ix) = longval;
13323 ptr = POPPTR(ss,ix);
13324 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13325 sv = (const SV *)POPPTR(ss,ix);
13326 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13328 case SAVEt_SET_SVFLAGS:
13330 TOPINT(nss,ix) = i;
13332 TOPINT(nss,ix) = i;
13333 sv = (const SV *)POPPTR(ss,ix);
13334 TOPPTR(nss,ix) = sv_dup(sv, param);
13336 case SAVEt_COMPILE_WARNINGS:
13337 ptr = POPPTR(ss,ix);
13338 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
13341 ptr = POPPTR(ss,ix);
13342 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
13346 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
13354 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
13355 * flag to the result. This is done for each stash before cloning starts,
13356 * so we know which stashes want their objects cloned */
13359 do_mark_cloneable_stash(pTHX_ SV *const sv)
13361 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
13363 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
13364 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
13365 if (cloner && GvCV(cloner)) {
13372 mXPUSHs(newSVhek(hvname));
13374 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
13381 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
13389 =for apidoc perl_clone
13391 Create and return a new interpreter by cloning the current one.
13393 perl_clone takes these flags as parameters:
13395 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
13396 without it we only clone the data and zero the stacks,
13397 with it we copy the stacks and the new perl interpreter is
13398 ready to run at the exact same point as the previous one.
13399 The pseudo-fork code uses COPY_STACKS while the
13400 threads->create doesn't.
13402 CLONEf_KEEP_PTR_TABLE -
13403 perl_clone keeps a ptr_table with the pointer of the old
13404 variable as a key and the new variable as a value,
13405 this allows it to check if something has been cloned and not
13406 clone it again but rather just use the value and increase the
13407 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
13408 the ptr_table using the function
13409 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
13410 reason to keep it around is if you want to dup some of your own
13411 variable who are outside the graph perl scans, example of this
13412 code is in threads.xs create.
13414 CLONEf_CLONE_HOST -
13415 This is a win32 thing, it is ignored on unix, it tells perls
13416 win32host code (which is c++) to clone itself, this is needed on
13417 win32 if you want to run two threads at the same time,
13418 if you just want to do some stuff in a separate perl interpreter
13419 and then throw it away and return to the original one,
13420 you don't need to do anything.
13425 /* XXX the above needs expanding by someone who actually understands it ! */
13426 EXTERN_C PerlInterpreter *
13427 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
13430 perl_clone(PerlInterpreter *proto_perl, UV flags)
13433 #ifdef PERL_IMPLICIT_SYS
13435 PERL_ARGS_ASSERT_PERL_CLONE;
13437 /* perlhost.h so we need to call into it
13438 to clone the host, CPerlHost should have a c interface, sky */
13440 if (flags & CLONEf_CLONE_HOST) {
13441 return perl_clone_host(proto_perl,flags);
13443 return perl_clone_using(proto_perl, flags,
13445 proto_perl->IMemShared,
13446 proto_perl->IMemParse,
13448 proto_perl->IStdIO,
13452 proto_perl->IProc);
13456 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
13457 struct IPerlMem* ipM, struct IPerlMem* ipMS,
13458 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
13459 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
13460 struct IPerlDir* ipD, struct IPerlSock* ipS,
13461 struct IPerlProc* ipP)
13463 /* XXX many of the string copies here can be optimized if they're
13464 * constants; they need to be allocated as common memory and just
13465 * their pointers copied. */
13468 CLONE_PARAMS clone_params;
13469 CLONE_PARAMS* const param = &clone_params;
13471 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
13473 PERL_ARGS_ASSERT_PERL_CLONE_USING;
13474 #else /* !PERL_IMPLICIT_SYS */
13476 CLONE_PARAMS clone_params;
13477 CLONE_PARAMS* param = &clone_params;
13478 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
13480 PERL_ARGS_ASSERT_PERL_CLONE;
13481 #endif /* PERL_IMPLICIT_SYS */
13483 /* for each stash, determine whether its objects should be cloned */
13484 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
13485 PERL_SET_THX(my_perl);
13488 PoisonNew(my_perl, 1, PerlInterpreter);
13491 PL_defstash = NULL; /* may be used by perl malloc() */
13494 PL_scopestack_name = 0;
13496 PL_savestack_ix = 0;
13497 PL_savestack_max = -1;
13498 PL_sig_pending = 0;
13500 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
13501 # ifdef DEBUG_LEAKING_SCALARS
13502 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
13504 #else /* !DEBUGGING */
13505 Zero(my_perl, 1, PerlInterpreter);
13506 #endif /* DEBUGGING */
13508 #ifdef PERL_IMPLICIT_SYS
13509 /* host pointers */
13511 PL_MemShared = ipMS;
13512 PL_MemParse = ipMP;
13519 #endif /* PERL_IMPLICIT_SYS */
13522 param->flags = flags;
13523 /* Nothing in the core code uses this, but we make it available to
13524 extensions (using mg_dup). */
13525 param->proto_perl = proto_perl;
13526 /* Likely nothing will use this, but it is initialised to be consistent
13527 with Perl_clone_params_new(). */
13528 param->new_perl = my_perl;
13529 param->unreferenced = NULL;
13532 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
13534 PL_body_arenas = NULL;
13535 Zero(&PL_body_roots, 1, PL_body_roots);
13539 PL_sv_arenaroot = NULL;
13541 PL_debug = proto_perl->Idebug;
13543 /* dbargs array probably holds garbage */
13546 PL_compiling = proto_perl->Icompiling;
13548 /* pseudo environmental stuff */
13549 PL_origargc = proto_perl->Iorigargc;
13550 PL_origargv = proto_perl->Iorigargv;
13552 #ifndef NO_TAINT_SUPPORT
13553 /* Set tainting stuff before PerlIO_debug can possibly get called */
13554 PL_tainting = proto_perl->Itainting;
13555 PL_taint_warn = proto_perl->Itaint_warn;
13557 PL_tainting = FALSE;
13558 PL_taint_warn = FALSE;
13561 PL_minus_c = proto_perl->Iminus_c;
13563 PL_localpatches = proto_perl->Ilocalpatches;
13564 PL_splitstr = proto_perl->Isplitstr;
13565 PL_minus_n = proto_perl->Iminus_n;
13566 PL_minus_p = proto_perl->Iminus_p;
13567 PL_minus_l = proto_perl->Iminus_l;
13568 PL_minus_a = proto_perl->Iminus_a;
13569 PL_minus_E = proto_perl->Iminus_E;
13570 PL_minus_F = proto_perl->Iminus_F;
13571 PL_doswitches = proto_perl->Idoswitches;
13572 PL_dowarn = proto_perl->Idowarn;
13573 #ifdef PERL_SAWAMPERSAND
13574 PL_sawampersand = proto_perl->Isawampersand;
13576 PL_unsafe = proto_perl->Iunsafe;
13577 PL_perldb = proto_perl->Iperldb;
13578 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
13579 PL_exit_flags = proto_perl->Iexit_flags;
13581 /* XXX time(&PL_basetime) when asked for? */
13582 PL_basetime = proto_perl->Ibasetime;
13584 PL_maxsysfd = proto_perl->Imaxsysfd;
13585 PL_statusvalue = proto_perl->Istatusvalue;
13587 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
13589 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
13592 /* RE engine related */
13593 PL_regmatch_slab = NULL;
13594 PL_reg_curpm = NULL;
13596 PL_sub_generation = proto_perl->Isub_generation;
13598 /* funky return mechanisms */
13599 PL_forkprocess = proto_perl->Iforkprocess;
13601 /* internal state */
13602 PL_maxo = proto_perl->Imaxo;
13604 PL_main_start = proto_perl->Imain_start;
13605 PL_eval_root = proto_perl->Ieval_root;
13606 PL_eval_start = proto_perl->Ieval_start;
13608 PL_filemode = proto_perl->Ifilemode;
13609 PL_lastfd = proto_perl->Ilastfd;
13610 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
13613 PL_gensym = proto_perl->Igensym;
13615 PL_laststatval = proto_perl->Ilaststatval;
13616 PL_laststype = proto_perl->Ilaststype;
13619 PL_profiledata = NULL;
13621 PL_generation = proto_perl->Igeneration;
13623 PL_in_clean_objs = proto_perl->Iin_clean_objs;
13624 PL_in_clean_all = proto_perl->Iin_clean_all;
13626 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
13627 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
13628 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
13629 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
13630 PL_nomemok = proto_perl->Inomemok;
13631 PL_an = proto_perl->Ian;
13632 PL_evalseq = proto_perl->Ievalseq;
13633 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
13634 PL_origalen = proto_perl->Iorigalen;
13636 PL_sighandlerp = proto_perl->Isighandlerp;
13638 PL_runops = proto_perl->Irunops;
13640 PL_subline = proto_perl->Isubline;
13643 PL_cryptseen = proto_perl->Icryptseen;
13646 #ifdef USE_LOCALE_COLLATE
13647 PL_collation_ix = proto_perl->Icollation_ix;
13648 PL_collation_standard = proto_perl->Icollation_standard;
13649 PL_collxfrm_base = proto_perl->Icollxfrm_base;
13650 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
13651 #endif /* USE_LOCALE_COLLATE */
13653 #ifdef USE_LOCALE_NUMERIC
13654 PL_numeric_standard = proto_perl->Inumeric_standard;
13655 PL_numeric_local = proto_perl->Inumeric_local;
13656 #endif /* !USE_LOCALE_NUMERIC */
13658 /* Did the locale setup indicate UTF-8? */
13659 PL_utf8locale = proto_perl->Iutf8locale;
13660 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
13661 /* Unicode features (see perlrun/-C) */
13662 PL_unicode = proto_perl->Iunicode;
13664 /* Pre-5.8 signals control */
13665 PL_signals = proto_perl->Isignals;
13667 /* times() ticks per second */
13668 PL_clocktick = proto_perl->Iclocktick;
13670 /* Recursion stopper for PerlIO_find_layer */
13671 PL_in_load_module = proto_perl->Iin_load_module;
13673 /* sort() routine */
13674 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
13676 /* Not really needed/useful since the reenrant_retint is "volatile",
13677 * but do it for consistency's sake. */
13678 PL_reentrant_retint = proto_perl->Ireentrant_retint;
13680 /* Hooks to shared SVs and locks. */
13681 PL_sharehook = proto_perl->Isharehook;
13682 PL_lockhook = proto_perl->Ilockhook;
13683 PL_unlockhook = proto_perl->Iunlockhook;
13684 PL_threadhook = proto_perl->Ithreadhook;
13685 PL_destroyhook = proto_perl->Idestroyhook;
13686 PL_signalhook = proto_perl->Isignalhook;
13688 PL_globhook = proto_perl->Iglobhook;
13691 PL_last_swash_hv = NULL; /* reinits on demand */
13692 PL_last_swash_klen = 0;
13693 PL_last_swash_key[0]= '\0';
13694 PL_last_swash_tmps = (U8*)NULL;
13695 PL_last_swash_slen = 0;
13697 PL_srand_called = proto_perl->Isrand_called;
13698 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
13700 if (flags & CLONEf_COPY_STACKS) {
13701 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
13702 PL_tmps_ix = proto_perl->Itmps_ix;
13703 PL_tmps_max = proto_perl->Itmps_max;
13704 PL_tmps_floor = proto_perl->Itmps_floor;
13706 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
13707 * NOTE: unlike the others! */
13708 PL_scopestack_ix = proto_perl->Iscopestack_ix;
13709 PL_scopestack_max = proto_perl->Iscopestack_max;
13711 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
13712 * NOTE: unlike the others! */
13713 PL_savestack_ix = proto_perl->Isavestack_ix;
13714 PL_savestack_max = proto_perl->Isavestack_max;
13717 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
13718 PL_top_env = &PL_start_env;
13720 PL_op = proto_perl->Iop;
13723 PL_Xpv = (XPV*)NULL;
13724 my_perl->Ina = proto_perl->Ina;
13726 PL_statbuf = proto_perl->Istatbuf;
13727 PL_statcache = proto_perl->Istatcache;
13729 #ifndef NO_TAINT_SUPPORT
13730 PL_tainted = proto_perl->Itainted;
13732 PL_tainted = FALSE;
13734 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
13736 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
13738 PL_restartjmpenv = proto_perl->Irestartjmpenv;
13739 PL_restartop = proto_perl->Irestartop;
13740 PL_in_eval = proto_perl->Iin_eval;
13741 PL_delaymagic = proto_perl->Idelaymagic;
13742 PL_phase = proto_perl->Iphase;
13743 PL_localizing = proto_perl->Ilocalizing;
13745 PL_hv_fetch_ent_mh = NULL;
13746 PL_modcount = proto_perl->Imodcount;
13747 PL_lastgotoprobe = NULL;
13748 PL_dumpindent = proto_perl->Idumpindent;
13750 PL_efloatbuf = NULL; /* reinits on demand */
13751 PL_efloatsize = 0; /* reinits on demand */
13755 PL_colorset = 0; /* reinits PL_colors[] */
13756 /*PL_colors[6] = {0,0,0,0,0,0};*/
13758 /* Pluggable optimizer */
13759 PL_peepp = proto_perl->Ipeepp;
13760 PL_rpeepp = proto_perl->Irpeepp;
13761 /* op_free() hook */
13762 PL_opfreehook = proto_perl->Iopfreehook;
13764 #ifdef USE_REENTRANT_API
13765 /* XXX: things like -Dm will segfault here in perlio, but doing
13766 * PERL_SET_CONTEXT(proto_perl);
13767 * breaks too many other things
13769 Perl_reentrant_init(aTHX);
13772 /* create SV map for pointer relocation */
13773 PL_ptr_table = ptr_table_new();
13775 /* initialize these special pointers as early as possible */
13777 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
13778 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
13779 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
13781 /* create (a non-shared!) shared string table */
13782 PL_strtab = newHV();
13783 HvSHAREKEYS_off(PL_strtab);
13784 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
13785 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
13787 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
13789 /* This PV will be free'd special way so must set it same way op.c does */
13790 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
13791 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
13793 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
13794 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
13795 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
13796 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
13798 param->stashes = newAV(); /* Setup array of objects to call clone on */
13799 /* This makes no difference to the implementation, as it always pushes
13800 and shifts pointers to other SVs without changing their reference
13801 count, with the array becoming empty before it is freed. However, it
13802 makes it conceptually clear what is going on, and will avoid some
13803 work inside av.c, filling slots between AvFILL() and AvMAX() with
13804 &PL_sv_undef, and SvREFCNT_dec()ing those. */
13805 AvREAL_off(param->stashes);
13807 if (!(flags & CLONEf_COPY_STACKS)) {
13808 param->unreferenced = newAV();
13811 #ifdef PERLIO_LAYERS
13812 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
13813 PerlIO_clone(aTHX_ proto_perl, param);
13816 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
13817 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
13818 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
13819 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
13820 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
13821 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
13824 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
13825 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
13826 PL_inplace = SAVEPV(proto_perl->Iinplace);
13827 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
13829 /* magical thingies */
13831 PL_encoding = sv_dup(proto_perl->Iencoding, param);
13833 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
13834 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
13835 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
13838 /* Clone the regex array */
13839 /* ORANGE FIXME for plugins, probably in the SV dup code.
13840 newSViv(PTR2IV(CALLREGDUPE(
13841 INT2PTR(REGEXP *, SvIVX(regex)), param))))
13843 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
13844 PL_regex_pad = AvARRAY(PL_regex_padav);
13846 PL_stashpadmax = proto_perl->Istashpadmax;
13847 PL_stashpadix = proto_perl->Istashpadix ;
13848 Newx(PL_stashpad, PL_stashpadmax, HV *);
13851 for (; o < PL_stashpadmax; ++o)
13852 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
13855 /* shortcuts to various I/O objects */
13856 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
13857 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
13858 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
13859 PL_defgv = gv_dup(proto_perl->Idefgv, param);
13860 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
13861 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
13862 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
13864 /* shortcuts to regexp stuff */
13865 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
13867 /* shortcuts to misc objects */
13868 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
13870 /* shortcuts to debugging objects */
13871 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
13872 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
13873 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
13874 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
13875 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
13876 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
13878 /* symbol tables */
13879 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
13880 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
13881 PL_debstash = hv_dup(proto_perl->Idebstash, param);
13882 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
13883 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
13885 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
13886 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
13887 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
13888 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
13889 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
13890 PL_endav = av_dup_inc(proto_perl->Iendav, param);
13891 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
13892 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
13894 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
13896 /* subprocess state */
13897 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
13899 if (proto_perl->Iop_mask)
13900 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
13903 /* PL_asserting = proto_perl->Iasserting; */
13905 /* current interpreter roots */
13906 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
13908 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
13911 /* runtime control stuff */
13912 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
13914 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
13916 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
13918 /* interpreter atexit processing */
13919 PL_exitlistlen = proto_perl->Iexitlistlen;
13920 if (PL_exitlistlen) {
13921 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13922 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
13925 PL_exitlist = (PerlExitListEntry*)NULL;
13927 PL_my_cxt_size = proto_perl->Imy_cxt_size;
13928 if (PL_my_cxt_size) {
13929 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
13930 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
13931 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13932 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
13933 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
13937 PL_my_cxt_list = (void**)NULL;
13938 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
13939 PL_my_cxt_keys = (const char**)NULL;
13942 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
13943 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
13944 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
13945 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
13947 PL_compcv = cv_dup(proto_perl->Icompcv, param);
13949 PAD_CLONE_VARS(proto_perl, param);
13951 #ifdef HAVE_INTERP_INTERN
13952 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
13955 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
13957 #ifdef PERL_USES_PL_PIDSTATUS
13958 PL_pidstatus = newHV(); /* XXX flag for cloning? */
13960 PL_osname = SAVEPV(proto_perl->Iosname);
13961 PL_parser = parser_dup(proto_perl->Iparser, param);
13963 /* XXX this only works if the saved cop has already been cloned */
13964 if (proto_perl->Iparser) {
13965 PL_parser->saved_curcop = (COP*)any_dup(
13966 proto_perl->Iparser->saved_curcop,
13970 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
13972 #ifdef USE_LOCALE_COLLATE
13973 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
13974 #endif /* USE_LOCALE_COLLATE */
13976 #ifdef USE_LOCALE_NUMERIC
13977 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
13978 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
13979 #endif /* !USE_LOCALE_NUMERIC */
13981 /* Unicode inversion lists */
13982 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
13983 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
13984 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
13986 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
13987 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
13989 /* utf8 character class swashes */
13990 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
13991 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
13993 for (i = 0; i < POSIX_CC_COUNT; i++) {
13994 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
13996 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
13997 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
13998 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
13999 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14000 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14001 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14002 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14003 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14004 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14005 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14006 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14007 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14008 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14009 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14010 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14011 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14013 if (proto_perl->Ipsig_pend) {
14014 Newxz(PL_psig_pend, SIG_SIZE, int);
14017 PL_psig_pend = (int*)NULL;
14020 if (proto_perl->Ipsig_name) {
14021 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14022 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14024 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14027 PL_psig_ptr = (SV**)NULL;
14028 PL_psig_name = (SV**)NULL;
14031 if (flags & CLONEf_COPY_STACKS) {
14032 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14033 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14034 PL_tmps_ix+1, param);
14036 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14037 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14038 Newxz(PL_markstack, i, I32);
14039 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14040 - proto_perl->Imarkstack);
14041 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14042 - proto_perl->Imarkstack);
14043 Copy(proto_perl->Imarkstack, PL_markstack,
14044 PL_markstack_ptr - PL_markstack + 1, I32);
14046 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14047 * NOTE: unlike the others! */
14048 Newxz(PL_scopestack, PL_scopestack_max, I32);
14049 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14052 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14053 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14055 /* reset stack AV to correct length before its duped via
14056 * PL_curstackinfo */
14057 AvFILLp(proto_perl->Icurstack) =
14058 proto_perl->Istack_sp - proto_perl->Istack_base;
14060 /* NOTE: si_dup() looks at PL_markstack */
14061 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14063 /* PL_curstack = PL_curstackinfo->si_stack; */
14064 PL_curstack = av_dup(proto_perl->Icurstack, param);
14065 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14067 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14068 PL_stack_base = AvARRAY(PL_curstack);
14069 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14070 - proto_perl->Istack_base);
14071 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14073 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14074 PL_savestack = ss_dup(proto_perl, param);
14078 ENTER; /* perl_destruct() wants to LEAVE; */
14081 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14082 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14084 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14085 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14086 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14087 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14088 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14089 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14091 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14093 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14094 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14095 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14097 PL_stashcache = newHV();
14099 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14100 proto_perl->Iwatchaddr);
14101 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14102 if (PL_debug && PL_watchaddr) {
14103 PerlIO_printf(Perl_debug_log,
14104 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14105 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14106 PTR2UV(PL_watchok));
14109 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14110 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14111 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14113 /* Call the ->CLONE method, if it exists, for each of the stashes
14114 identified by sv_dup() above.
14116 while(av_tindex(param->stashes) != -1) {
14117 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14118 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14119 if (cloner && GvCV(cloner)) {
14124 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14126 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14132 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14133 ptr_table_free(PL_ptr_table);
14134 PL_ptr_table = NULL;
14137 if (!(flags & CLONEf_COPY_STACKS)) {
14138 unreferenced_to_tmp_stack(param->unreferenced);
14141 SvREFCNT_dec(param->stashes);
14143 /* orphaned? eg threads->new inside BEGIN or use */
14144 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14145 SvREFCNT_inc_simple_void(PL_compcv);
14146 SAVEFREESV(PL_compcv);
14153 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14155 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14157 if (AvFILLp(unreferenced) > -1) {
14158 SV **svp = AvARRAY(unreferenced);
14159 SV **const last = svp + AvFILLp(unreferenced);
14163 if (SvREFCNT(*svp) == 1)
14165 } while (++svp <= last);
14167 EXTEND_MORTAL(count);
14168 svp = AvARRAY(unreferenced);
14171 if (SvREFCNT(*svp) == 1) {
14172 /* Our reference is the only one to this SV. This means that
14173 in this thread, the scalar effectively has a 0 reference.
14174 That doesn't work (cleanup never happens), so donate our
14175 reference to it onto the save stack. */
14176 PL_tmps_stack[++PL_tmps_ix] = *svp;
14178 /* As an optimisation, because we are already walking the
14179 entire array, instead of above doing either
14180 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14181 release our reference to the scalar, so that at the end of
14182 the array owns zero references to the scalars it happens to
14183 point to. We are effectively converting the array from
14184 AvREAL() on to AvREAL() off. This saves the av_clear()
14185 (triggered by the SvREFCNT_dec(unreferenced) below) from
14186 walking the array a second time. */
14187 SvREFCNT_dec(*svp);
14190 } while (++svp <= last);
14191 AvREAL_off(unreferenced);
14193 SvREFCNT_dec_NN(unreferenced);
14197 Perl_clone_params_del(CLONE_PARAMS *param)
14199 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14201 PerlInterpreter *const to = param->new_perl;
14203 PerlInterpreter *const was = PERL_GET_THX;
14205 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14211 SvREFCNT_dec(param->stashes);
14212 if (param->unreferenced)
14213 unreferenced_to_tmp_stack(param->unreferenced);
14223 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14226 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14227 does a dTHX; to get the context from thread local storage.
14228 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14229 a version that passes in my_perl. */
14230 PerlInterpreter *const was = PERL_GET_THX;
14231 CLONE_PARAMS *param;
14233 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14239 /* Given that we've set the context, we can do this unshared. */
14240 Newx(param, 1, CLONE_PARAMS);
14243 param->proto_perl = from;
14244 param->new_perl = to;
14245 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14246 AvREAL_off(param->stashes);
14247 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14255 #endif /* USE_ITHREADS */
14258 Perl_init_constants(pTHX)
14260 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14261 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVt_NULL;
14262 SvANY(&PL_sv_undef) = NULL;
14264 SvANY(&PL_sv_no) = new_XPVNV();
14265 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14266 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY
14267 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14270 SvANY(&PL_sv_yes) = new_XPVNV();
14271 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14272 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY
14273 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14276 SvPV_set(&PL_sv_no, (char*)PL_No);
14277 SvCUR_set(&PL_sv_no, 0);
14278 SvLEN_set(&PL_sv_no, 0);
14279 SvIV_set(&PL_sv_no, 0);
14280 SvNV_set(&PL_sv_no, 0);
14282 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14283 SvCUR_set(&PL_sv_yes, 1);
14284 SvLEN_set(&PL_sv_yes, 0);
14285 SvIV_set(&PL_sv_yes, 1);
14286 SvNV_set(&PL_sv_yes, 1);
14290 =head1 Unicode Support
14292 =for apidoc sv_recode_to_utf8
14294 The encoding is assumed to be an Encode object, on entry the PV
14295 of the sv is assumed to be octets in that encoding, and the sv
14296 will be converted into Unicode (and UTF-8).
14298 If the sv already is UTF-8 (or if it is not POK), or if the encoding
14299 is not a reference, nothing is done to the sv. If the encoding is not
14300 an C<Encode::XS> Encoding object, bad things will happen.
14301 (See F<lib/encoding.pm> and L<Encode>.)
14303 The PV of the sv is returned.
14308 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
14312 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
14314 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
14323 if (SvPADTMP(nsv)) {
14324 nsv = sv_newmortal();
14325 SvSetSV_nosteal(nsv, sv);
14334 Passing sv_yes is wrong - it needs to be or'ed set of constants
14335 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
14336 remove converted chars from source.
14338 Both will default the value - let them.
14340 XPUSHs(&PL_sv_yes);
14343 call_method("decode", G_SCALAR);
14347 s = SvPV_const(uni, len);
14348 if (s != SvPVX_const(sv)) {
14349 SvGROW(sv, len + 1);
14350 Move(s, SvPVX(sv), len + 1, char);
14351 SvCUR_set(sv, len);
14356 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
14357 /* clear pos and any utf8 cache */
14358 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
14361 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
14362 magic_setutf8(sv,mg); /* clear UTF8 cache */
14367 return SvPOKp(sv) ? SvPVX(sv) : NULL;
14371 =for apidoc sv_cat_decode
14373 The encoding is assumed to be an Encode object, the PV of the ssv is
14374 assumed to be octets in that encoding and decoding the input starts
14375 from the position which (PV + *offset) pointed to. The dsv will be
14376 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
14377 when the string tstr appears in decoding output or the input ends on
14378 the PV of the ssv. The value which the offset points will be modified
14379 to the last input position on the ssv.
14381 Returns TRUE if the terminator was found, else returns FALSE.
14386 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
14387 SV *ssv, int *offset, char *tstr, int tlen)
14392 PERL_ARGS_ASSERT_SV_CAT_DECODE;
14394 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
14405 offsv = newSViv(*offset);
14407 mPUSHp(tstr, tlen);
14409 call_method("cat_decode", G_SCALAR);
14411 ret = SvTRUE(TOPs);
14412 *offset = SvIV(offsv);
14418 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
14423 /* ---------------------------------------------------------------------
14425 * support functions for report_uninit()
14428 /* the maxiumum size of array or hash where we will scan looking
14429 * for the undefined element that triggered the warning */
14431 #define FUV_MAX_SEARCH_SIZE 1000
14433 /* Look for an entry in the hash whose value has the same SV as val;
14434 * If so, return a mortal copy of the key. */
14437 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
14443 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
14445 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
14446 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
14449 array = HvARRAY(hv);
14451 for (i=HvMAX(hv); i>=0; i--) {
14453 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
14454 if (HeVAL(entry) != val)
14456 if ( HeVAL(entry) == &PL_sv_undef ||
14457 HeVAL(entry) == &PL_sv_placeholder)
14461 if (HeKLEN(entry) == HEf_SVKEY)
14462 return sv_mortalcopy(HeKEY_sv(entry));
14463 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
14469 /* Look for an entry in the array whose value has the same SV as val;
14470 * If so, return the index, otherwise return -1. */
14473 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
14477 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
14479 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
14480 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
14483 if (val != &PL_sv_undef) {
14484 SV ** const svp = AvARRAY(av);
14487 for (i=AvFILLp(av); i>=0; i--)
14494 /* varname(): return the name of a variable, optionally with a subscript.
14495 * If gv is non-zero, use the name of that global, along with gvtype (one
14496 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
14497 * targ. Depending on the value of the subscript_type flag, return:
14500 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
14501 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
14502 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
14503 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
14506 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
14507 const SV *const keyname, I32 aindex, int subscript_type)
14510 SV * const name = sv_newmortal();
14511 if (gv && isGV(gv)) {
14513 buffer[0] = gvtype;
14516 /* as gv_fullname4(), but add literal '^' for $^FOO names */
14518 gv_fullname4(name, gv, buffer, 0);
14520 if ((unsigned int)SvPVX(name)[1] <= 26) {
14522 buffer[1] = SvPVX(name)[1] + 'A' - 1;
14524 /* Swap the 1 unprintable control character for the 2 byte pretty
14525 version - ie substr($name, 1, 1) = $buffer; */
14526 sv_insert(name, 1, 1, buffer, 2);
14530 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
14534 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
14536 if (!cv || !CvPADLIST(cv))
14538 av = *PadlistARRAY(CvPADLIST(cv));
14539 sv = *av_fetch(av, targ, FALSE);
14540 sv_setsv_flags(name, sv, 0);
14543 if (subscript_type == FUV_SUBSCRIPT_HASH) {
14544 SV * const sv = newSV(0);
14545 *SvPVX(name) = '$';
14546 Perl_sv_catpvf(aTHX_ name, "{%s}",
14547 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
14548 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
14549 SvREFCNT_dec_NN(sv);
14551 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
14552 *SvPVX(name) = '$';
14553 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
14555 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
14556 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
14557 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
14565 =for apidoc find_uninit_var
14567 Find the name of the undefined variable (if any) that caused the operator
14568 to issue a "Use of uninitialized value" warning.
14569 If match is true, only return a name if its value matches uninit_sv.
14570 So roughly speaking, if a unary operator (such as OP_COS) generates a
14571 warning, then following the direct child of the op may yield an
14572 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
14573 other hand, with OP_ADD there are two branches to follow, so we only print
14574 the variable name if we get an exact match.
14576 The name is returned as a mortal SV.
14578 Assumes that PL_op is the op that originally triggered the error, and that
14579 PL_comppad/PL_curpad points to the currently executing pad.
14585 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
14591 const OP *o, *o2, *kid;
14593 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
14594 uninit_sv == &PL_sv_placeholder)))
14597 switch (obase->op_type) {
14604 const bool pad = ( obase->op_type == OP_PADAV
14605 || obase->op_type == OP_PADHV
14606 || obase->op_type == OP_PADRANGE
14609 const bool hash = ( obase->op_type == OP_PADHV
14610 || obase->op_type == OP_RV2HV
14611 || (obase->op_type == OP_PADRANGE
14612 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
14616 int subscript_type = FUV_SUBSCRIPT_WITHIN;
14618 if (pad) { /* @lex, %lex */
14619 sv = PAD_SVl(obase->op_targ);
14623 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14624 /* @global, %global */
14625 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14628 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
14630 else if (obase == PL_op) /* @{expr}, %{expr} */
14631 return find_uninit_var(cUNOPx(obase)->op_first,
14633 else /* @{expr}, %{expr} as a sub-expression */
14637 /* attempt to find a match within the aggregate */
14639 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14641 subscript_type = FUV_SUBSCRIPT_HASH;
14644 index = find_array_subscript((const AV *)sv, uninit_sv);
14646 subscript_type = FUV_SUBSCRIPT_ARRAY;
14649 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
14652 return varname(gv, hash ? '%' : '@', obase->op_targ,
14653 keysv, index, subscript_type);
14657 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
14659 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
14660 if (!gv || !GvSTASH(gv))
14662 if (match && (GvSV(gv) != uninit_sv))
14664 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14667 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
14670 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
14672 return varname(NULL, '$', obase->op_targ,
14673 NULL, 0, FUV_SUBSCRIPT_NONE);
14676 gv = cGVOPx_gv(obase);
14677 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
14679 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
14681 case OP_AELEMFAST_LEX:
14684 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
14685 if (!av || SvRMAGICAL(av))
14687 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14688 if (!svp || *svp != uninit_sv)
14691 return varname(NULL, '$', obase->op_targ,
14692 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14695 gv = cGVOPx_gv(obase);
14700 AV *const av = GvAV(gv);
14701 if (!av || SvRMAGICAL(av))
14703 svp = av_fetch(av, (I8)obase->op_private, FALSE);
14704 if (!svp || *svp != uninit_sv)
14707 return varname(gv, '$', 0,
14708 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
14710 NOT_REACHED; /* NOTREACHED */
14713 o = cUNOPx(obase)->op_first;
14714 if (!o || o->op_type != OP_NULL ||
14715 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
14717 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
14722 bool negate = FALSE;
14724 if (PL_op == obase)
14725 /* $a[uninit_expr] or $h{uninit_expr} */
14726 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
14729 o = cBINOPx(obase)->op_first;
14730 kid = cBINOPx(obase)->op_last;
14732 /* get the av or hv, and optionally the gv */
14734 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
14735 sv = PAD_SV(o->op_targ);
14737 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
14738 && cUNOPo->op_first->op_type == OP_GV)
14740 gv = cGVOPx_gv(cUNOPo->op_first);
14744 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
14749 if (kid && kid->op_type == OP_NEGATE) {
14751 kid = cUNOPx(kid)->op_first;
14754 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
14755 /* index is constant */
14758 kidsv = sv_2mortal(newSVpvs("-"));
14759 sv_catsv(kidsv, cSVOPx_sv(kid));
14762 kidsv = cSVOPx_sv(kid);
14766 if (obase->op_type == OP_HELEM) {
14767 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
14768 if (!he || HeVAL(he) != uninit_sv)
14772 SV * const opsv = cSVOPx_sv(kid);
14773 const IV opsviv = SvIV(opsv);
14774 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
14775 negate ? - opsviv : opsviv,
14777 if (!svp || *svp != uninit_sv)
14781 if (obase->op_type == OP_HELEM)
14782 return varname(gv, '%', o->op_targ,
14783 kidsv, 0, FUV_SUBSCRIPT_HASH);
14785 return varname(gv, '@', o->op_targ, NULL,
14786 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
14787 FUV_SUBSCRIPT_ARRAY);
14790 /* index is an expression;
14791 * attempt to find a match within the aggregate */
14792 if (obase->op_type == OP_HELEM) {
14793 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
14795 return varname(gv, '%', o->op_targ,
14796 keysv, 0, FUV_SUBSCRIPT_HASH);
14800 = find_array_subscript((const AV *)sv, uninit_sv);
14802 return varname(gv, '@', o->op_targ,
14803 NULL, index, FUV_SUBSCRIPT_ARRAY);
14808 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
14810 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
14812 NOT_REACHED; /* NOTREACHED */
14816 /* only examine RHS */
14817 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
14820 o = cUNOPx(obase)->op_first;
14821 if ( o->op_type == OP_PUSHMARK
14822 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
14826 if (!o->op_sibling) {
14827 /* one-arg version of open is highly magical */
14829 if (o->op_type == OP_GV) { /* open FOO; */
14831 if (match && GvSV(gv) != uninit_sv)
14833 return varname(gv, '$', 0,
14834 NULL, 0, FUV_SUBSCRIPT_NONE);
14836 /* other possibilities not handled are:
14837 * open $x; or open my $x; should return '${*$x}'
14838 * open expr; should return '$'.expr ideally
14844 /* ops where $_ may be an implicit arg */
14849 if ( !(obase->op_flags & OPf_STACKED)) {
14850 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
14851 ? PAD_SVl(obase->op_targ)
14854 sv = sv_newmortal();
14855 sv_setpvs(sv, "$_");
14864 match = 1; /* print etc can return undef on defined args */
14865 /* skip filehandle as it can't produce 'undef' warning */
14866 o = cUNOPx(obase)->op_first;
14867 if ((obase->op_flags & OPf_STACKED)
14869 ( o->op_type == OP_PUSHMARK
14870 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
14871 o = o->op_sibling->op_sibling;
14875 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
14876 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
14878 /* the following ops are capable of returning PL_sv_undef even for
14879 * defined arg(s) */
14898 case OP_GETPEERNAME:
14946 case OP_SMARTMATCH:
14955 /* XXX tmp hack: these two may call an XS sub, and currently
14956 XS subs don't have a SUB entry on the context stack, so CV and
14957 pad determination goes wrong, and BAD things happen. So, just
14958 don't try to determine the value under those circumstances.
14959 Need a better fix at dome point. DAPM 11/2007 */
14965 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
14966 if (gv && GvSV(gv) == uninit_sv)
14967 return newSVpvs_flags("$.", SVs_TEMP);
14972 /* def-ness of rval pos() is independent of the def-ness of its arg */
14973 if ( !(obase->op_flags & OPf_MOD))
14978 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
14979 return newSVpvs_flags("${$/}", SVs_TEMP);
14984 if (!(obase->op_flags & OPf_KIDS))
14986 o = cUNOPx(obase)->op_first;
14992 /* This loop checks all the kid ops, skipping any that cannot pos-
14993 * sibly be responsible for the uninitialized value; i.e., defined
14994 * constants and ops that return nothing. If there is only one op
14995 * left that is not skipped, then we *know* it is responsible for
14996 * the uninitialized value. If there is more than one op left, we
14997 * have to look for an exact match in the while() loop below.
14998 * Note that we skip padrange, because the individual pad ops that
14999 * it replaced are still in the tree, so we work on them instead.
15002 for (kid=o; kid; kid = kid->op_sibling) {
15003 const OPCODE type = kid->op_type;
15004 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
15005 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
15006 || (type == OP_PUSHMARK)
15007 || (type == OP_PADRANGE)
15011 if (o2) { /* more than one found */
15018 return find_uninit_var(o2, uninit_sv, match);
15020 /* scan all args */
15022 sv = find_uninit_var(o, uninit_sv, 1);
15034 =for apidoc report_uninit
15036 Print appropriate "Use of uninitialized variable" warning.
15042 Perl_report_uninit(pTHX_ const SV *uninit_sv)
15046 SV* varname = NULL;
15047 if (uninit_sv && PL_curpad) {
15048 varname = find_uninit_var(PL_op, uninit_sv,0);
15050 sv_insert(varname, 0, 0, " ", 1);
15052 /* PL_warn_uninit_sv is constant */
15053 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15054 /* diag_listed_as: Use of uninitialized value%s */
15055 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
15056 SVfARG(varname ? varname : &PL_sv_no),
15057 " in ", OP_DESC(PL_op));
15061 /* PL_warn_uninit is constant */
15062 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15063 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15071 * c-indentation-style: bsd
15072 * c-basic-offset: 4
15073 * indent-tabs-mode: nil
15076 * ex: set ts=8 sts=4 sw=4 et: