3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 #ifdef PERL_OLD_COPY_ON_WRITE
129 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
130 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
133 /* ============================================================================
135 =head1 Allocation and deallocation of SVs.
136 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
137 sv, av, hv...) contains type and reference count information, and for
138 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
139 contains fields specific to each type. Some types store all they need
140 in the head, so don't have a body.
142 In all but the most memory-paranoid configurations (ex: PURIFY), heads
143 and bodies are allocated out of arenas, which by default are
144 approximately 4K chunks of memory parcelled up into N heads or bodies.
145 Sv-bodies are allocated by their sv-type, guaranteeing size
146 consistency needed to allocate safely from arrays.
148 For SV-heads, the first slot in each arena is reserved, and holds a
149 link to the next arena, some flags, and a note of the number of slots.
150 Snaked through each arena chain is a linked list of free items; when
151 this becomes empty, an extra arena is allocated and divided up into N
152 items which are threaded into the free list.
154 SV-bodies are similar, but they use arena-sets by default, which
155 separate the link and info from the arena itself, and reclaim the 1st
156 slot in the arena. SV-bodies are further described later.
158 The following global variables are associated with arenas:
160 PL_sv_arenaroot pointer to list of SV arenas
161 PL_sv_root pointer to list of free SV structures
163 PL_body_arenas head of linked-list of body arenas
164 PL_body_roots[] array of pointers to list of free bodies of svtype
165 arrays are indexed by the svtype needed
167 A few special SV heads are not allocated from an arena, but are
168 instead directly created in the interpreter structure, eg PL_sv_undef.
169 The size of arenas can be changed from the default by setting
170 PERL_ARENA_SIZE appropriately at compile time.
172 The SV arena serves the secondary purpose of allowing still-live SVs
173 to be located and destroyed during final cleanup.
175 At the lowest level, the macros new_SV() and del_SV() grab and free
176 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
177 to return the SV to the free list with error checking.) new_SV() calls
178 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
179 SVs in the free list have their SvTYPE field set to all ones.
181 At the time of very final cleanup, sv_free_arenas() is called from
182 perl_destruct() to physically free all the arenas allocated since the
183 start of the interpreter.
185 The function visit() scans the SV arenas list, and calls a specified
186 function for each SV it finds which is still live - ie which has an SvTYPE
187 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
188 following functions (specified as [function that calls visit()] / [function
189 called by visit() for each SV]):
191 sv_report_used() / do_report_used()
192 dump all remaining SVs (debugging aid)
194 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
195 do_clean_named_io_objs(),do_curse()
196 Attempt to free all objects pointed to by RVs,
197 try to do the same for all objects indir-
198 ectly referenced by typeglobs too, and
199 then do a final sweep, cursing any
200 objects that remain. Called once from
201 perl_destruct(), prior to calling sv_clean_all()
204 sv_clean_all() / do_clean_all()
205 SvREFCNT_dec(sv) each remaining SV, possibly
206 triggering an sv_free(). It also sets the
207 SVf_BREAK flag on the SV to indicate that the
208 refcnt has been artificially lowered, and thus
209 stopping sv_free() from giving spurious warnings
210 about SVs which unexpectedly have a refcnt
211 of zero. called repeatedly from perl_destruct()
212 until there are no SVs left.
214 =head2 Arena allocator API Summary
216 Private API to rest of sv.c
220 new_XPVNV(), del_XPVGV(),
225 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
229 * ========================================================================= */
232 * "A time to plant, and a time to uproot what was planted..."
236 # define MEM_LOG_NEW_SV(sv, file, line, func) \
237 Perl_mem_log_new_sv(sv, file, line, func)
238 # define MEM_LOG_DEL_SV(sv, file, line, func) \
239 Perl_mem_log_del_sv(sv, file, line, func)
241 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
242 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
245 #ifdef DEBUG_LEAKING_SCALARS
246 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
247 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
249 # define DEBUG_SV_SERIAL(sv) \
250 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
251 PTR2UV(sv), (long)(sv)->sv_debug_serial))
253 # define FREE_SV_DEBUG_FILE(sv)
254 # define DEBUG_SV_SERIAL(sv) NOOP
258 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
259 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
260 /* Whilst I'd love to do this, it seems that things like to check on
262 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
264 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
265 PoisonNew(&SvREFCNT(sv), 1, U32)
267 # define SvARENA_CHAIN(sv) SvANY(sv)
268 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
269 # define POISON_SV_HEAD(sv)
272 /* Mark an SV head as unused, and add to free list.
274 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
275 * its refcount artificially decremented during global destruction, so
276 * there may be dangling pointers to it. The last thing we want in that
277 * case is for it to be reused. */
279 #define plant_SV(p) \
281 const U32 old_flags = SvFLAGS(p); \
282 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
283 DEBUG_SV_SERIAL(p); \
284 FREE_SV_DEBUG_FILE(p); \
286 SvFLAGS(p) = SVTYPEMASK; \
287 if (!(old_flags & SVf_BREAK)) { \
288 SvARENA_CHAIN_SET(p, PL_sv_root); \
294 #define uproot_SV(p) \
297 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
302 /* make some more SVs by adding another arena */
308 char *chunk; /* must use New here to match call to */
309 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
310 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
315 /* new_SV(): return a new, empty SV head */
317 #ifdef DEBUG_LEAKING_SCALARS
318 /* provide a real function for a debugger to play with */
320 S_new_SV(pTHX_ const char *file, int line, const char *func)
327 sv = S_more_sv(aTHX);
331 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
332 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
338 sv->sv_debug_inpad = 0;
339 sv->sv_debug_parent = NULL;
340 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
342 sv->sv_debug_serial = PL_sv_serial++;
344 MEM_LOG_NEW_SV(sv, file, line, func);
345 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
346 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
350 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
358 (p) = S_more_sv(aTHX); \
362 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
367 /* del_SV(): return an empty SV head to the free list */
380 S_del_sv(pTHX_ SV *p)
382 PERL_ARGS_ASSERT_DEL_SV;
387 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
388 const SV * const sv = sva + 1;
389 const SV * const svend = &sva[SvREFCNT(sva)];
390 if (p >= sv && p < svend) {
396 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
397 "Attempt to free non-arena SV: 0x%"UVxf
398 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
405 #else /* ! DEBUGGING */
407 #define del_SV(p) plant_SV(p)
409 #endif /* DEBUGGING */
412 * Bodyless IVs and NVs!
414 * Since 5.9.2, we can avoid allocating a body for SVt_IV-type SVs.
415 * Since the larger IV-holding variants of SVs store their integer
416 * values in their respective bodies, the family of SvIV() accessor
417 * macros would naively have to branch on the SV type to find the
418 * integer value either in the HEAD or BODY. In order to avoid this
419 * expensive branch, a clever soul has deployed a great hack:
420 * We set up the SvANY pointer such that instead of pointing to a
421 * real body, it points into the memory before the location of the
422 * head. We compute this pointer such that the location of
423 * the integer member of the hypothetical body struct happens to
424 * be the same as the location of the integer member of the bodyless
425 * SV head. This now means that the SvIV() family of accessors can
426 * always read from the (hypothetical or real) body via SvANY.
428 * Since the 5.21 dev series, we employ the same trick for NVs
429 * if the architecture can support it (NVSIZE <= IVSIZE).
432 /* The following two macros compute the necessary offsets for the above
433 * trick and store them in SvANY for SvIV() (and friends) to use. */
434 #define SET_SVANY_FOR_BODYLESS_IV(sv) \
435 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv))
437 #define SET_SVANY_FOR_BODYLESS_NV(sv) \
438 SvANY(sv) = (XPVNV*)((char*)&(sv->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv))
441 =head1 SV Manipulation Functions
443 =for apidoc sv_add_arena
445 Given a chunk of memory, link it to the head of the list of arenas,
446 and split it into a list of free SVs.
452 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
454 SV *const sva = MUTABLE_SV(ptr);
458 PERL_ARGS_ASSERT_SV_ADD_ARENA;
460 /* The first SV in an arena isn't an SV. */
461 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
462 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
463 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
465 PL_sv_arenaroot = sva;
466 PL_sv_root = sva + 1;
468 svend = &sva[SvREFCNT(sva) - 1];
471 SvARENA_CHAIN_SET(sv, (sv + 1));
475 /* Must always set typemask because it's always checked in on cleanup
476 when the arenas are walked looking for objects. */
477 SvFLAGS(sv) = SVTYPEMASK;
480 SvARENA_CHAIN_SET(sv, 0);
484 SvFLAGS(sv) = SVTYPEMASK;
487 /* visit(): call the named function for each non-free SV in the arenas
488 * whose flags field matches the flags/mask args. */
491 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
496 PERL_ARGS_ASSERT_VISIT;
498 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
499 const SV * const svend = &sva[SvREFCNT(sva)];
501 for (sv = sva + 1; sv < svend; ++sv) {
502 if (SvTYPE(sv) != (svtype)SVTYPEMASK
503 && (sv->sv_flags & mask) == flags
516 /* called by sv_report_used() for each live SV */
519 do_report_used(pTHX_ SV *const sv)
521 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
522 PerlIO_printf(Perl_debug_log, "****\n");
529 =for apidoc sv_report_used
531 Dump the contents of all SVs not yet freed (debugging aid).
537 Perl_sv_report_used(pTHX)
540 visit(do_report_used, 0, 0);
546 /* called by sv_clean_objs() for each live SV */
549 do_clean_objs(pTHX_ SV *const ref)
553 SV * const target = SvRV(ref);
554 if (SvOBJECT(target)) {
555 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
556 if (SvWEAKREF(ref)) {
557 sv_del_backref(target, ref);
563 SvREFCNT_dec_NN(target);
570 /* clear any slots in a GV which hold objects - except IO;
571 * called by sv_clean_objs() for each live GV */
574 do_clean_named_objs(pTHX_ SV *const sv)
577 assert(SvTYPE(sv) == SVt_PVGV);
578 assert(isGV_with_GP(sv));
582 /* freeing GP entries may indirectly free the current GV;
583 * hold onto it while we mess with the GP slots */
586 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
587 DEBUG_D((PerlIO_printf(Perl_debug_log,
588 "Cleaning named glob SV object:\n "), sv_dump(obj)));
590 SvREFCNT_dec_NN(obj);
592 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
593 DEBUG_D((PerlIO_printf(Perl_debug_log,
594 "Cleaning named glob AV object:\n "), sv_dump(obj)));
596 SvREFCNT_dec_NN(obj);
598 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
599 DEBUG_D((PerlIO_printf(Perl_debug_log,
600 "Cleaning named glob HV object:\n "), sv_dump(obj)));
602 SvREFCNT_dec_NN(obj);
604 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
605 DEBUG_D((PerlIO_printf(Perl_debug_log,
606 "Cleaning named glob CV object:\n "), sv_dump(obj)));
608 SvREFCNT_dec_NN(obj);
610 SvREFCNT_dec_NN(sv); /* undo the inc above */
613 /* clear any IO slots in a GV which hold objects (except stderr, defout);
614 * called by sv_clean_objs() for each live GV */
617 do_clean_named_io_objs(pTHX_ SV *const sv)
620 assert(SvTYPE(sv) == SVt_PVGV);
621 assert(isGV_with_GP(sv));
622 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
626 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
627 DEBUG_D((PerlIO_printf(Perl_debug_log,
628 "Cleaning named glob IO object:\n "), sv_dump(obj)));
630 SvREFCNT_dec_NN(obj);
632 SvREFCNT_dec_NN(sv); /* undo the inc above */
635 /* Void wrapper to pass to visit() */
637 do_curse(pTHX_ SV * const sv) {
638 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
639 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
645 =for apidoc sv_clean_objs
647 Attempt to destroy all objects not yet freed.
653 Perl_sv_clean_objs(pTHX)
656 PL_in_clean_objs = TRUE;
657 visit(do_clean_objs, SVf_ROK, SVf_ROK);
658 /* Some barnacles may yet remain, clinging to typeglobs.
659 * Run the non-IO destructors first: they may want to output
660 * error messages, close files etc */
661 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
662 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
663 /* And if there are some very tenacious barnacles clinging to arrays,
664 closures, or what have you.... */
665 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
666 olddef = PL_defoutgv;
667 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
668 if (olddef && isGV_with_GP(olddef))
669 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
670 olderr = PL_stderrgv;
671 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
672 if (olderr && isGV_with_GP(olderr))
673 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
674 SvREFCNT_dec(olddef);
675 PL_in_clean_objs = FALSE;
678 /* called by sv_clean_all() for each live SV */
681 do_clean_all(pTHX_ SV *const sv)
683 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
684 /* don't clean pid table and strtab */
687 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
688 SvFLAGS(sv) |= SVf_BREAK;
693 =for apidoc sv_clean_all
695 Decrement the refcnt of each remaining SV, possibly triggering a
696 cleanup. This function may have to be called multiple times to free
697 SVs which are in complex self-referential hierarchies.
703 Perl_sv_clean_all(pTHX)
706 PL_in_clean_all = TRUE;
707 cleaned = visit(do_clean_all, 0,0);
712 ARENASETS: a meta-arena implementation which separates arena-info
713 into struct arena_set, which contains an array of struct
714 arena_descs, each holding info for a single arena. By separating
715 the meta-info from the arena, we recover the 1st slot, formerly
716 borrowed for list management. The arena_set is about the size of an
717 arena, avoiding the needless malloc overhead of a naive linked-list.
719 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
720 memory in the last arena-set (1/2 on average). In trade, we get
721 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
722 smaller types). The recovery of the wasted space allows use of
723 small arenas for large, rare body types, by changing array* fields
724 in body_details_by_type[] below.
727 char *arena; /* the raw storage, allocated aligned */
728 size_t size; /* its size ~4k typ */
729 svtype utype; /* bodytype stored in arena */
734 /* Get the maximum number of elements in set[] such that struct arena_set
735 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
736 therefore likely to be 1 aligned memory page. */
738 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
739 - 2 * sizeof(int)) / sizeof (struct arena_desc))
742 struct arena_set* next;
743 unsigned int set_size; /* ie ARENAS_PER_SET */
744 unsigned int curr; /* index of next available arena-desc */
745 struct arena_desc set[ARENAS_PER_SET];
749 =for apidoc sv_free_arenas
751 Deallocate the memory used by all arenas. Note that all the individual SV
752 heads and bodies within the arenas must already have been freed.
758 Perl_sv_free_arenas(pTHX)
764 /* Free arenas here, but be careful about fake ones. (We assume
765 contiguity of the fake ones with the corresponding real ones.) */
767 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
768 svanext = MUTABLE_SV(SvANY(sva));
769 while (svanext && SvFAKE(svanext))
770 svanext = MUTABLE_SV(SvANY(svanext));
777 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
780 struct arena_set *current = aroot;
783 assert(aroot->set[i].arena);
784 Safefree(aroot->set[i].arena);
792 i = PERL_ARENA_ROOTS_SIZE;
794 PL_body_roots[i] = 0;
801 Here are mid-level routines that manage the allocation of bodies out
802 of the various arenas. There are 5 kinds of arenas:
804 1. SV-head arenas, which are discussed and handled above
805 2. regular body arenas
806 3. arenas for reduced-size bodies
809 Arena types 2 & 3 are chained by body-type off an array of
810 arena-root pointers, which is indexed by svtype. Some of the
811 larger/less used body types are malloced singly, since a large
812 unused block of them is wasteful. Also, several svtypes dont have
813 bodies; the data fits into the sv-head itself. The arena-root
814 pointer thus has a few unused root-pointers (which may be hijacked
815 later for arena types 4,5)
817 3 differs from 2 as an optimization; some body types have several
818 unused fields in the front of the structure (which are kept in-place
819 for consistency). These bodies can be allocated in smaller chunks,
820 because the leading fields arent accessed. Pointers to such bodies
821 are decremented to point at the unused 'ghost' memory, knowing that
822 the pointers are used with offsets to the real memory.
825 =head1 SV-Body Allocation
829 Allocation of SV-bodies is similar to SV-heads, differing as follows;
830 the allocation mechanism is used for many body types, so is somewhat
831 more complicated, it uses arena-sets, and has no need for still-live
834 At the outermost level, (new|del)_X*V macros return bodies of the
835 appropriate type. These macros call either (new|del)_body_type or
836 (new|del)_body_allocated macro pairs, depending on specifics of the
837 type. Most body types use the former pair, the latter pair is used to
838 allocate body types with "ghost fields".
840 "ghost fields" are fields that are unused in certain types, and
841 consequently don't need to actually exist. They are declared because
842 they're part of a "base type", which allows use of functions as
843 methods. The simplest examples are AVs and HVs, 2 aggregate types
844 which don't use the fields which support SCALAR semantics.
846 For these types, the arenas are carved up into appropriately sized
847 chunks, we thus avoid wasted memory for those unaccessed members.
848 When bodies are allocated, we adjust the pointer back in memory by the
849 size of the part not allocated, so it's as if we allocated the full
850 structure. (But things will all go boom if you write to the part that
851 is "not there", because you'll be overwriting the last members of the
852 preceding structure in memory.)
854 We calculate the correction using the STRUCT_OFFSET macro on the first
855 member present. If the allocated structure is smaller (no initial NV
856 actually allocated) then the net effect is to subtract the size of the NV
857 from the pointer, to return a new pointer as if an initial NV were actually
858 allocated. (We were using structures named *_allocated for this, but
859 this turned out to be a subtle bug, because a structure without an NV
860 could have a lower alignment constraint, but the compiler is allowed to
861 optimised accesses based on the alignment constraint of the actual pointer
862 to the full structure, for example, using a single 64 bit load instruction
863 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
865 This is the same trick as was used for NV and IV bodies. Ironically it
866 doesn't need to be used for NV bodies any more, because NV is now at
867 the start of the structure. IV bodies, and also in some builds NV bodies,
868 don't need it either, because they are no longer allocated.
870 In turn, the new_body_* allocators call S_new_body(), which invokes
871 new_body_inline macro, which takes a lock, and takes a body off the
872 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
873 necessary to refresh an empty list. Then the lock is released, and
874 the body is returned.
876 Perl_more_bodies allocates a new arena, and carves it up into an array of N
877 bodies, which it strings into a linked list. It looks up arena-size
878 and body-size from the body_details table described below, thus
879 supporting the multiple body-types.
881 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
882 the (new|del)_X*V macros are mapped directly to malloc/free.
884 For each sv-type, struct body_details bodies_by_type[] carries
885 parameters which control these aspects of SV handling:
887 Arena_size determines whether arenas are used for this body type, and if
888 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
889 zero, forcing individual mallocs and frees.
891 Body_size determines how big a body is, and therefore how many fit into
892 each arena. Offset carries the body-pointer adjustment needed for
893 "ghost fields", and is used in *_allocated macros.
895 But its main purpose is to parameterize info needed in
896 Perl_sv_upgrade(). The info here dramatically simplifies the function
897 vs the implementation in 5.8.8, making it table-driven. All fields
898 are used for this, except for arena_size.
900 For the sv-types that have no bodies, arenas are not used, so those
901 PL_body_roots[sv_type] are unused, and can be overloaded. In
902 something of a special case, SVt_NULL is borrowed for HE arenas;
903 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
904 bodies_by_type[SVt_NULL] slot is not used, as the table is not
909 struct body_details {
910 U8 body_size; /* Size to allocate */
911 U8 copy; /* Size of structure to copy (may be shorter) */
912 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
913 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
914 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
915 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
916 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
917 U32 arena_size; /* Size of arena to allocate */
925 /* With -DPURFIY we allocate everything directly, and don't use arenas.
926 This seems a rather elegant way to simplify some of the code below. */
927 #define HASARENA FALSE
929 #define HASARENA TRUE
931 #define NOARENA FALSE
933 /* Size the arenas to exactly fit a given number of bodies. A count
934 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
935 simplifying the default. If count > 0, the arena is sized to fit
936 only that many bodies, allowing arenas to be used for large, rare
937 bodies (XPVFM, XPVIO) without undue waste. The arena size is
938 limited by PERL_ARENA_SIZE, so we can safely oversize the
941 #define FIT_ARENA0(body_size) \
942 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
943 #define FIT_ARENAn(count,body_size) \
944 ( count * body_size <= PERL_ARENA_SIZE) \
945 ? count * body_size \
946 : FIT_ARENA0 (body_size)
947 #define FIT_ARENA(count,body_size) \
949 ? FIT_ARENAn (count, body_size) \
950 : FIT_ARENA0 (body_size))
952 /* Calculate the length to copy. Specifically work out the length less any
953 final padding the compiler needed to add. See the comment in sv_upgrade
954 for why copying the padding proved to be a bug. */
956 #define copy_length(type, last_member) \
957 STRUCT_OFFSET(type, last_member) \
958 + sizeof (((type*)SvANY((const SV *)0))->last_member)
960 static const struct body_details bodies_by_type[] = {
961 /* HEs use this offset for their arena. */
962 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
964 /* IVs are in the head, so the allocation size is 0. */
966 sizeof(IV), /* This is used to copy out the IV body. */
967 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
968 NOARENA /* IVS don't need an arena */, 0
973 STRUCT_OFFSET(XPVNV, xnv_u),
974 SVt_NV, FALSE, HADNV, NOARENA, 0 },
976 { sizeof(NV), sizeof(NV),
977 STRUCT_OFFSET(XPVNV, xnv_u),
978 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
981 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
982 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
983 + STRUCT_OFFSET(XPV, xpv_cur),
984 SVt_PV, FALSE, NONV, HASARENA,
985 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
987 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
988 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
989 + STRUCT_OFFSET(XPV, xpv_cur),
990 SVt_INVLIST, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
993 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
994 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
995 + STRUCT_OFFSET(XPV, xpv_cur),
996 SVt_PVIV, FALSE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
999 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
1000 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
1001 + STRUCT_OFFSET(XPV, xpv_cur),
1002 SVt_PVNV, FALSE, HADNV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
1005 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
1006 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
1011 SVt_REGEXP, TRUE, NONV, HASARENA,
1012 FIT_ARENA(0, sizeof(regexp))
1015 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
1016 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
1018 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1019 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
1022 copy_length(XPVAV, xav_alloc),
1024 SVt_PVAV, TRUE, NONV, HASARENA,
1025 FIT_ARENA(0, sizeof(XPVAV)) },
1028 copy_length(XPVHV, xhv_max),
1030 SVt_PVHV, TRUE, NONV, HASARENA,
1031 FIT_ARENA(0, sizeof(XPVHV)) },
1036 SVt_PVCV, TRUE, NONV, HASARENA,
1037 FIT_ARENA(0, sizeof(XPVCV)) },
1042 SVt_PVFM, TRUE, NONV, NOARENA,
1043 FIT_ARENA(20, sizeof(XPVFM)) },
1048 SVt_PVIO, TRUE, NONV, HASARENA,
1049 FIT_ARENA(24, sizeof(XPVIO)) },
1052 #define new_body_allocated(sv_type) \
1053 (void *)((char *)S_new_body(aTHX_ sv_type) \
1054 - bodies_by_type[sv_type].offset)
1056 /* return a thing to the free list */
1058 #define del_body(thing, root) \
1060 void ** const thing_copy = (void **)thing; \
1061 *thing_copy = *root; \
1062 *root = (void*)thing_copy; \
1066 #if !(NVSIZE <= IVSIZE)
1067 # define new_XNV() safemalloc(sizeof(XPVNV))
1069 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1070 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1072 #define del_XPVGV(p) safefree(p)
1076 #if !(NVSIZE <= IVSIZE)
1077 # define new_XNV() new_body_allocated(SVt_NV)
1079 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1080 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1082 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1083 &PL_body_roots[SVt_PVGV])
1087 /* no arena for you! */
1089 #define new_NOARENA(details) \
1090 safemalloc((details)->body_size + (details)->offset)
1091 #define new_NOARENAZ(details) \
1092 safecalloc((details)->body_size + (details)->offset, 1)
1095 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1096 const size_t arena_size)
1098 void ** const root = &PL_body_roots[sv_type];
1099 struct arena_desc *adesc;
1100 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1104 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1105 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1108 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1109 static bool done_sanity_check;
1111 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1112 * variables like done_sanity_check. */
1113 if (!done_sanity_check) {
1114 unsigned int i = SVt_LAST;
1116 done_sanity_check = TRUE;
1119 assert (bodies_by_type[i].type == i);
1125 /* may need new arena-set to hold new arena */
1126 if (!aroot || aroot->curr >= aroot->set_size) {
1127 struct arena_set *newroot;
1128 Newxz(newroot, 1, struct arena_set);
1129 newroot->set_size = ARENAS_PER_SET;
1130 newroot->next = aroot;
1132 PL_body_arenas = (void *) newroot;
1133 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1136 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1137 curr = aroot->curr++;
1138 adesc = &(aroot->set[curr]);
1139 assert(!adesc->arena);
1141 Newx(adesc->arena, good_arena_size, char);
1142 adesc->size = good_arena_size;
1143 adesc->utype = sv_type;
1144 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1145 curr, (void*)adesc->arena, (UV)good_arena_size));
1147 start = (char *) adesc->arena;
1149 /* Get the address of the byte after the end of the last body we can fit.
1150 Remember, this is integer division: */
1151 end = start + good_arena_size / body_size * body_size;
1153 /* computed count doesn't reflect the 1st slot reservation */
1154 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1155 DEBUG_m(PerlIO_printf(Perl_debug_log,
1156 "arena %p end %p arena-size %d (from %d) type %d "
1158 (void*)start, (void*)end, (int)good_arena_size,
1159 (int)arena_size, sv_type, (int)body_size,
1160 (int)good_arena_size / (int)body_size));
1162 DEBUG_m(PerlIO_printf(Perl_debug_log,
1163 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1164 (void*)start, (void*)end,
1165 (int)arena_size, sv_type, (int)body_size,
1166 (int)good_arena_size / (int)body_size));
1168 *root = (void *)start;
1171 /* Where the next body would start: */
1172 char * const next = start + body_size;
1175 /* This is the last body: */
1176 assert(next == end);
1178 *(void **)start = 0;
1182 *(void**) start = (void *)next;
1187 /* grab a new thing from the free list, allocating more if necessary.
1188 The inline version is used for speed in hot routines, and the
1189 function using it serves the rest (unless PURIFY).
1191 #define new_body_inline(xpv, sv_type) \
1193 void ** const r3wt = &PL_body_roots[sv_type]; \
1194 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1195 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1196 bodies_by_type[sv_type].body_size,\
1197 bodies_by_type[sv_type].arena_size)); \
1198 *(r3wt) = *(void**)(xpv); \
1204 S_new_body(pTHX_ const svtype sv_type)
1207 new_body_inline(xpv, sv_type);
1213 static const struct body_details fake_rv =
1214 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1217 =for apidoc sv_upgrade
1219 Upgrade an SV to a more complex form. Generally adds a new body type to the
1220 SV, then copies across as much information as possible from the old body.
1221 It croaks if the SV is already in a more complex form than requested. You
1222 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1223 before calling C<sv_upgrade>, and hence does not croak. See also
1230 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1234 const svtype old_type = SvTYPE(sv);
1235 const struct body_details *new_type_details;
1236 const struct body_details *old_type_details
1237 = bodies_by_type + old_type;
1238 SV *referant = NULL;
1240 PERL_ARGS_ASSERT_SV_UPGRADE;
1242 if (old_type == new_type)
1245 /* This clause was purposefully added ahead of the early return above to
1246 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1247 inference by Nick I-S that it would fix other troublesome cases. See
1248 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1250 Given that shared hash key scalars are no longer PVIV, but PV, there is
1251 no longer need to unshare so as to free up the IVX slot for its proper
1252 purpose. So it's safe to move the early return earlier. */
1254 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1255 sv_force_normal_flags(sv, 0);
1258 old_body = SvANY(sv);
1260 /* Copying structures onto other structures that have been neatly zeroed
1261 has a subtle gotcha. Consider XPVMG
1263 +------+------+------+------+------+-------+-------+
1264 | NV | CUR | LEN | IV | MAGIC | STASH |
1265 +------+------+------+------+------+-------+-------+
1266 0 4 8 12 16 20 24 28
1268 where NVs are aligned to 8 bytes, so that sizeof that structure is
1269 actually 32 bytes long, with 4 bytes of padding at the end:
1271 +------+------+------+------+------+-------+-------+------+
1272 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1273 +------+------+------+------+------+-------+-------+------+
1274 0 4 8 12 16 20 24 28 32
1276 so what happens if you allocate memory for this structure:
1278 +------+------+------+------+------+-------+-------+------+------+...
1279 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1280 +------+------+------+------+------+-------+-------+------+------+...
1281 0 4 8 12 16 20 24 28 32 36
1283 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1284 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1285 started out as zero once, but it's quite possible that it isn't. So now,
1286 rather than a nicely zeroed GP, you have it pointing somewhere random.
1289 (In fact, GP ends up pointing at a previous GP structure, because the
1290 principle cause of the padding in XPVMG getting garbage is a copy of
1291 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1292 this happens to be moot because XPVGV has been re-ordered, with GP
1293 no longer after STASH)
1295 So we are careful and work out the size of used parts of all the
1303 referant = SvRV(sv);
1304 old_type_details = &fake_rv;
1305 if (new_type == SVt_NV)
1306 new_type = SVt_PVNV;
1308 if (new_type < SVt_PVIV) {
1309 new_type = (new_type == SVt_NV)
1310 ? SVt_PVNV : SVt_PVIV;
1315 if (new_type < SVt_PVNV) {
1316 new_type = SVt_PVNV;
1320 assert(new_type > SVt_PV);
1321 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1322 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1329 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1330 there's no way that it can be safely upgraded, because perl.c
1331 expects to Safefree(SvANY(PL_mess_sv)) */
1332 assert(sv != PL_mess_sv);
1335 if (UNLIKELY(old_type_details->cant_upgrade))
1336 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1337 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1340 if (UNLIKELY(old_type > new_type))
1341 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1342 (int)old_type, (int)new_type);
1344 new_type_details = bodies_by_type + new_type;
1346 SvFLAGS(sv) &= ~SVTYPEMASK;
1347 SvFLAGS(sv) |= new_type;
1349 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1350 the return statements above will have triggered. */
1351 assert (new_type != SVt_NULL);
1354 assert(old_type == SVt_NULL);
1355 SET_SVANY_FOR_BODYLESS_IV(sv);
1359 assert(old_type == SVt_NULL);
1360 #if NVSIZE <= IVSIZE
1361 SET_SVANY_FOR_BODYLESS_NV(sv);
1363 SvANY(sv) = new_XNV();
1369 assert(new_type_details->body_size);
1372 assert(new_type_details->arena);
1373 assert(new_type_details->arena_size);
1374 /* This points to the start of the allocated area. */
1375 new_body_inline(new_body, new_type);
1376 Zero(new_body, new_type_details->body_size, char);
1377 new_body = ((char *)new_body) - new_type_details->offset;
1379 /* We always allocated the full length item with PURIFY. To do this
1380 we fake things so that arena is false for all 16 types.. */
1381 new_body = new_NOARENAZ(new_type_details);
1383 SvANY(sv) = new_body;
1384 if (new_type == SVt_PVAV) {
1388 if (old_type_details->body_size) {
1391 /* It will have been zeroed when the new body was allocated.
1392 Lets not write to it, in case it confuses a write-back
1398 #ifndef NODEFAULT_SHAREKEYS
1399 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1401 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1402 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1405 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1406 The target created by newSVrv also is, and it can have magic.
1407 However, it never has SvPVX set.
1409 if (old_type == SVt_IV) {
1411 } else if (old_type >= SVt_PV) {
1412 assert(SvPVX_const(sv) == 0);
1415 if (old_type >= SVt_PVMG) {
1416 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1417 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1419 sv->sv_u.svu_array = NULL; /* or svu_hash */
1424 /* XXX Is this still needed? Was it ever needed? Surely as there is
1425 no route from NV to PVIV, NOK can never be true */
1426 assert(!SvNOKp(sv));
1439 assert(new_type_details->body_size);
1440 /* We always allocated the full length item with PURIFY. To do this
1441 we fake things so that arena is false for all 16 types.. */
1442 if(new_type_details->arena) {
1443 /* This points to the start of the allocated area. */
1444 new_body_inline(new_body, new_type);
1445 Zero(new_body, new_type_details->body_size, char);
1446 new_body = ((char *)new_body) - new_type_details->offset;
1448 new_body = new_NOARENAZ(new_type_details);
1450 SvANY(sv) = new_body;
1452 if (old_type_details->copy) {
1453 /* There is now the potential for an upgrade from something without
1454 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1455 int offset = old_type_details->offset;
1456 int length = old_type_details->copy;
1458 if (new_type_details->offset > old_type_details->offset) {
1459 const int difference
1460 = new_type_details->offset - old_type_details->offset;
1461 offset += difference;
1462 length -= difference;
1464 assert (length >= 0);
1466 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1470 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1471 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1472 * correct 0.0 for us. Otherwise, if the old body didn't have an
1473 * NV slot, but the new one does, then we need to initialise the
1474 * freshly created NV slot with whatever the correct bit pattern is
1476 if (old_type_details->zero_nv && !new_type_details->zero_nv
1477 && !isGV_with_GP(sv))
1481 if (UNLIKELY(new_type == SVt_PVIO)) {
1482 IO * const io = MUTABLE_IO(sv);
1483 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1486 /* Clear the stashcache because a new IO could overrule a package
1488 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1489 hv_clear(PL_stashcache);
1491 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1492 IoPAGE_LEN(sv) = 60;
1494 if (UNLIKELY(new_type == SVt_REGEXP))
1495 sv->sv_u.svu_rx = (regexp *)new_body;
1496 else if (old_type < SVt_PV) {
1497 /* referant will be NULL unless the old type was SVt_IV emulating
1499 sv->sv_u.svu_rv = referant;
1503 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1504 (unsigned long)new_type);
1507 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1508 and sometimes SVt_NV */
1509 if (old_type_details->body_size) {
1513 /* Note that there is an assumption that all bodies of types that
1514 can be upgraded came from arenas. Only the more complex non-
1515 upgradable types are allowed to be directly malloc()ed. */
1516 assert(old_type_details->arena);
1517 del_body((void*)((char*)old_body + old_type_details->offset),
1518 &PL_body_roots[old_type]);
1524 =for apidoc sv_backoff
1526 Remove any string offset. You should normally use the C<SvOOK_off> macro
1533 Perl_sv_backoff(SV *const sv)
1536 const char * const s = SvPVX_const(sv);
1538 PERL_ARGS_ASSERT_SV_BACKOFF;
1541 assert(SvTYPE(sv) != SVt_PVHV);
1542 assert(SvTYPE(sv) != SVt_PVAV);
1544 SvOOK_offset(sv, delta);
1546 SvLEN_set(sv, SvLEN(sv) + delta);
1547 SvPV_set(sv, SvPVX(sv) - delta);
1548 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1549 SvFLAGS(sv) &= ~SVf_OOK;
1556 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1557 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1558 Use the C<SvGROW> wrapper instead.
1563 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1566 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1570 PERL_ARGS_ASSERT_SV_GROW;
1574 if (SvTYPE(sv) < SVt_PV) {
1575 sv_upgrade(sv, SVt_PV);
1576 s = SvPVX_mutable(sv);
1578 else if (SvOOK(sv)) { /* pv is offset? */
1580 s = SvPVX_mutable(sv);
1581 if (newlen > SvLEN(sv))
1582 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1586 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1587 s = SvPVX_mutable(sv);
1590 #ifdef PERL_NEW_COPY_ON_WRITE
1591 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1592 * to store the COW count. So in general, allocate one more byte than
1593 * asked for, to make it likely this byte is always spare: and thus
1594 * make more strings COW-able.
1595 * If the new size is a big power of two, don't bother: we assume the
1596 * caller wanted a nice 2^N sized block and will be annoyed at getting
1602 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1603 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1606 if (newlen > SvLEN(sv)) { /* need more room? */
1607 STRLEN minlen = SvCUR(sv);
1608 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1609 if (newlen < minlen)
1611 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1613 /* Don't round up on the first allocation, as odds are pretty good that
1614 * the initial request is accurate as to what is really needed */
1616 newlen = PERL_STRLEN_ROUNDUP(newlen);
1619 if (SvLEN(sv) && s) {
1620 s = (char*)saferealloc(s, newlen);
1623 s = (char*)safemalloc(newlen);
1624 if (SvPVX_const(sv) && SvCUR(sv)) {
1625 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1629 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1630 /* Do this here, do it once, do it right, and then we will never get
1631 called back into sv_grow() unless there really is some growing
1633 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1635 SvLEN_set(sv, newlen);
1642 =for apidoc sv_setiv
1644 Copies an integer into the given SV, upgrading first if necessary.
1645 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1651 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1653 PERL_ARGS_ASSERT_SV_SETIV;
1655 SV_CHECK_THINKFIRST_COW_DROP(sv);
1656 switch (SvTYPE(sv)) {
1659 sv_upgrade(sv, SVt_IV);
1662 sv_upgrade(sv, SVt_PVIV);
1666 if (!isGV_with_GP(sv))
1673 /* diag_listed_as: Can't coerce %s to %s in %s */
1674 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1678 (void)SvIOK_only(sv); /* validate number */
1684 =for apidoc sv_setiv_mg
1686 Like C<sv_setiv>, but also handles 'set' magic.
1692 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1694 PERL_ARGS_ASSERT_SV_SETIV_MG;
1701 =for apidoc sv_setuv
1703 Copies an unsigned integer into the given SV, upgrading first if necessary.
1704 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1710 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1712 PERL_ARGS_ASSERT_SV_SETUV;
1714 /* With the if statement to ensure that integers are stored as IVs whenever
1716 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1719 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1721 If you wish to remove the following if statement, so that this routine
1722 (and its callers) always return UVs, please benchmark to see what the
1723 effect is. Modern CPUs may be different. Or may not :-)
1725 if (u <= (UV)IV_MAX) {
1726 sv_setiv(sv, (IV)u);
1735 =for apidoc sv_setuv_mg
1737 Like C<sv_setuv>, but also handles 'set' magic.
1743 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1745 PERL_ARGS_ASSERT_SV_SETUV_MG;
1752 =for apidoc sv_setnv
1754 Copies a double into the given SV, upgrading first if necessary.
1755 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1761 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1763 PERL_ARGS_ASSERT_SV_SETNV;
1765 SV_CHECK_THINKFIRST_COW_DROP(sv);
1766 switch (SvTYPE(sv)) {
1769 sv_upgrade(sv, SVt_NV);
1773 sv_upgrade(sv, SVt_PVNV);
1777 if (!isGV_with_GP(sv))
1784 /* diag_listed_as: Can't coerce %s to %s in %s */
1785 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1790 (void)SvNOK_only(sv); /* validate number */
1795 =for apidoc sv_setnv_mg
1797 Like C<sv_setnv>, but also handles 'set' magic.
1803 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1805 PERL_ARGS_ASSERT_SV_SETNV_MG;
1811 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1812 * not incrementable warning display.
1813 * Originally part of S_not_a_number().
1814 * The return value may be != tmpbuf.
1818 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1821 PERL_ARGS_ASSERT_SV_DISPLAY;
1824 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1825 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1828 const char * const limit = tmpbuf + tmpbuf_size - 8;
1829 /* each *s can expand to 4 chars + "...\0",
1830 i.e. need room for 8 chars */
1832 const char *s = SvPVX_const(sv);
1833 const char * const end = s + SvCUR(sv);
1834 for ( ; s < end && d < limit; s++ ) {
1836 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1840 /* Map to ASCII "equivalent" of Latin1 */
1841 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1847 else if (ch == '\r') {
1851 else if (ch == '\f') {
1855 else if (ch == '\\') {
1859 else if (ch == '\0') {
1863 else if (isPRINT_LC(ch))
1882 /* Print an "isn't numeric" warning, using a cleaned-up,
1883 * printable version of the offending string
1887 S_not_a_number(pTHX_ SV *const sv)
1892 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1894 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1897 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1898 /* diag_listed_as: Argument "%s" isn't numeric%s */
1899 "Argument \"%s\" isn't numeric in %s", pv,
1902 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1903 /* diag_listed_as: Argument "%s" isn't numeric%s */
1904 "Argument \"%s\" isn't numeric", pv);
1908 S_not_incrementable(pTHX_ SV *const sv) {
1912 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1914 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1916 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1917 "Argument \"%s\" treated as 0 in increment (++)", pv);
1921 =for apidoc looks_like_number
1923 Test if the content of an SV looks like a number (or is a number).
1924 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1925 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1932 Perl_looks_like_number(pTHX_ SV *const sv)
1937 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1939 if (SvPOK(sv) || SvPOKp(sv)) {
1940 sbegin = SvPV_nomg_const(sv, len);
1943 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1944 return grok_number(sbegin, len, NULL);
1948 S_glob_2number(pTHX_ GV * const gv)
1950 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1952 /* We know that all GVs stringify to something that is not-a-number,
1953 so no need to test that. */
1954 if (ckWARN(WARN_NUMERIC))
1956 SV *const buffer = sv_newmortal();
1957 gv_efullname3(buffer, gv, "*");
1958 not_a_number(buffer);
1960 /* We just want something true to return, so that S_sv_2iuv_common
1961 can tail call us and return true. */
1965 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1966 until proven guilty, assume that things are not that bad... */
1971 As 64 bit platforms often have an NV that doesn't preserve all bits of
1972 an IV (an assumption perl has been based on to date) it becomes necessary
1973 to remove the assumption that the NV always carries enough precision to
1974 recreate the IV whenever needed, and that the NV is the canonical form.
1975 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1976 precision as a side effect of conversion (which would lead to insanity
1977 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1978 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1979 where precision was lost, and IV/UV/NV slots that have a valid conversion
1980 which has lost no precision
1981 2) to ensure that if a numeric conversion to one form is requested that
1982 would lose precision, the precise conversion (or differently
1983 imprecise conversion) is also performed and cached, to prevent
1984 requests for different numeric formats on the same SV causing
1985 lossy conversion chains. (lossless conversion chains are perfectly
1990 SvIOKp is true if the IV slot contains a valid value
1991 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1992 SvNOKp is true if the NV slot contains a valid value
1993 SvNOK is true only if the NV value is accurate
1996 while converting from PV to NV, check to see if converting that NV to an
1997 IV(or UV) would lose accuracy over a direct conversion from PV to
1998 IV(or UV). If it would, cache both conversions, return NV, but mark
1999 SV as IOK NOKp (ie not NOK).
2001 While converting from PV to IV, check to see if converting that IV to an
2002 NV would lose accuracy over a direct conversion from PV to NV. If it
2003 would, cache both conversions, flag similarly.
2005 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2006 correctly because if IV & NV were set NV *always* overruled.
2007 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2008 changes - now IV and NV together means that the two are interchangeable:
2009 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
2011 The benefit of this is that operations such as pp_add know that if
2012 SvIOK is true for both left and right operands, then integer addition
2013 can be used instead of floating point (for cases where the result won't
2014 overflow). Before, floating point was always used, which could lead to
2015 loss of precision compared with integer addition.
2017 * making IV and NV equal status should make maths accurate on 64 bit
2019 * may speed up maths somewhat if pp_add and friends start to use
2020 integers when possible instead of fp. (Hopefully the overhead in
2021 looking for SvIOK and checking for overflow will not outweigh the
2022 fp to integer speedup)
2023 * will slow down integer operations (callers of SvIV) on "inaccurate"
2024 values, as the change from SvIOK to SvIOKp will cause a call into
2025 sv_2iv each time rather than a macro access direct to the IV slot
2026 * should speed up number->string conversion on integers as IV is
2027 favoured when IV and NV are equally accurate
2029 ####################################################################
2030 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2031 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2032 On the other hand, SvUOK is true iff UV.
2033 ####################################################################
2035 Your mileage will vary depending your CPU's relative fp to integer
2039 #ifndef NV_PRESERVES_UV
2040 # define IS_NUMBER_UNDERFLOW_IV 1
2041 # define IS_NUMBER_UNDERFLOW_UV 2
2042 # define IS_NUMBER_IV_AND_UV 2
2043 # define IS_NUMBER_OVERFLOW_IV 4
2044 # define IS_NUMBER_OVERFLOW_UV 5
2046 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2048 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2050 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2056 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2057 PERL_UNUSED_CONTEXT;
2059 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));
2060 if (SvNVX(sv) < (NV)IV_MIN) {
2061 (void)SvIOKp_on(sv);
2063 SvIV_set(sv, IV_MIN);
2064 return IS_NUMBER_UNDERFLOW_IV;
2066 if (SvNVX(sv) > (NV)UV_MAX) {
2067 (void)SvIOKp_on(sv);
2070 SvUV_set(sv, UV_MAX);
2071 return IS_NUMBER_OVERFLOW_UV;
2073 (void)SvIOKp_on(sv);
2075 /* Can't use strtol etc to convert this string. (See truth table in
2077 if (SvNVX(sv) <= (UV)IV_MAX) {
2078 SvIV_set(sv, I_V(SvNVX(sv)));
2079 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2080 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2082 /* Integer is imprecise. NOK, IOKp */
2084 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2087 SvUV_set(sv, U_V(SvNVX(sv)));
2088 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2089 if (SvUVX(sv) == UV_MAX) {
2090 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2091 possibly be preserved by NV. Hence, it must be overflow.
2093 return IS_NUMBER_OVERFLOW_UV;
2095 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2097 /* Integer is imprecise. NOK, IOKp */
2099 return IS_NUMBER_OVERFLOW_IV;
2101 #endif /* !NV_PRESERVES_UV*/
2103 /* If numtype is infnan, set the NV of the sv accordingly.
2104 * If numtype is anything else, try setting the NV using Atof(PV). */
2106 S_sv_setnv(pTHX_ SV* sv, int numtype)
2108 bool pok = cBOOL(SvPOK(sv));
2110 if ((numtype & IS_NUMBER_INFINITY)) {
2111 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2114 else if ((numtype & IS_NUMBER_NAN)) {
2115 SvNV_set(sv, NV_NAN);
2119 SvNV_set(sv, Atof(SvPVX_const(sv)));
2120 /* Purposefully no true nok here, since we don't want to blow
2121 * away the possible IOK/UV of an existing sv. */
2124 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2126 SvPOK_on(sv); /* PV is okay, though. */
2131 S_sv_2iuv_common(pTHX_ SV *const sv)
2133 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2136 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2137 * without also getting a cached IV/UV from it at the same time
2138 * (ie PV->NV conversion should detect loss of accuracy and cache
2139 * IV or UV at same time to avoid this. */
2140 /* IV-over-UV optimisation - choose to cache IV if possible */
2142 if (SvTYPE(sv) == SVt_NV)
2143 sv_upgrade(sv, SVt_PVNV);
2145 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2146 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2147 certainly cast into the IV range at IV_MAX, whereas the correct
2148 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2150 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2151 if (Perl_isnan(SvNVX(sv))) {
2157 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2158 SvIV_set(sv, I_V(SvNVX(sv)));
2159 if (SvNVX(sv) == (NV) SvIVX(sv)
2160 #ifndef NV_PRESERVES_UV
2161 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2162 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2163 /* Don't flag it as "accurately an integer" if the number
2164 came from a (by definition imprecise) NV operation, and
2165 we're outside the range of NV integer precision */
2169 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2171 /* scalar has trailing garbage, eg "42a" */
2173 DEBUG_c(PerlIO_printf(Perl_debug_log,
2174 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2180 /* IV not precise. No need to convert from PV, as NV
2181 conversion would already have cached IV if it detected
2182 that PV->IV would be better than PV->NV->IV
2183 flags already correct - don't set public IOK. */
2184 DEBUG_c(PerlIO_printf(Perl_debug_log,
2185 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2190 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2191 but the cast (NV)IV_MIN rounds to a the value less (more
2192 negative) than IV_MIN which happens to be equal to SvNVX ??
2193 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2194 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2195 (NV)UVX == NVX are both true, but the values differ. :-(
2196 Hopefully for 2s complement IV_MIN is something like
2197 0x8000000000000000 which will be exact. NWC */
2200 SvUV_set(sv, U_V(SvNVX(sv)));
2202 (SvNVX(sv) == (NV) SvUVX(sv))
2203 #ifndef NV_PRESERVES_UV
2204 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2205 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2206 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2207 /* Don't flag it as "accurately an integer" if the number
2208 came from a (by definition imprecise) NV operation, and
2209 we're outside the range of NV integer precision */
2215 DEBUG_c(PerlIO_printf(Perl_debug_log,
2216 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2222 else if (SvPOKp(sv)) {
2224 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2225 /* We want to avoid a possible problem when we cache an IV/ a UV which
2226 may be later translated to an NV, and the resulting NV is not
2227 the same as the direct translation of the initial string
2228 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2229 be careful to ensure that the value with the .456 is around if the
2230 NV value is requested in the future).
2232 This means that if we cache such an IV/a UV, we need to cache the
2233 NV as well. Moreover, we trade speed for space, and do not
2234 cache the NV if we are sure it's not needed.
2237 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2238 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2239 == IS_NUMBER_IN_UV) {
2240 /* It's definitely an integer, only upgrade to PVIV */
2241 if (SvTYPE(sv) < SVt_PVIV)
2242 sv_upgrade(sv, SVt_PVIV);
2244 } else if (SvTYPE(sv) < SVt_PVNV)
2245 sv_upgrade(sv, SVt_PVNV);
2247 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2248 S_sv_setnv(aTHX_ sv, numtype);
2252 /* If NVs preserve UVs then we only use the UV value if we know that
2253 we aren't going to call atof() below. If NVs don't preserve UVs
2254 then the value returned may have more precision than atof() will
2255 return, even though value isn't perfectly accurate. */
2256 if ((numtype & (IS_NUMBER_IN_UV
2257 #ifdef NV_PRESERVES_UV
2260 )) == IS_NUMBER_IN_UV) {
2261 /* This won't turn off the public IOK flag if it was set above */
2262 (void)SvIOKp_on(sv);
2264 if (!(numtype & IS_NUMBER_NEG)) {
2266 if (value <= (UV)IV_MAX) {
2267 SvIV_set(sv, (IV)value);
2269 /* it didn't overflow, and it was positive. */
2270 SvUV_set(sv, value);
2274 /* 2s complement assumption */
2275 if (value <= (UV)IV_MIN) {
2276 SvIV_set(sv, -(IV)value);
2278 /* Too negative for an IV. This is a double upgrade, but
2279 I'm assuming it will be rare. */
2280 if (SvTYPE(sv) < SVt_PVNV)
2281 sv_upgrade(sv, SVt_PVNV);
2285 SvNV_set(sv, -(NV)value);
2286 SvIV_set(sv, IV_MIN);
2290 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2291 will be in the previous block to set the IV slot, and the next
2292 block to set the NV slot. So no else here. */
2294 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2295 != IS_NUMBER_IN_UV) {
2296 /* It wasn't an (integer that doesn't overflow the UV). */
2297 S_sv_setnv(aTHX_ sv, numtype);
2299 if (! numtype && ckWARN(WARN_NUMERIC))
2302 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2303 PTR2UV(sv), SvNVX(sv)));
2305 #ifdef NV_PRESERVES_UV
2306 (void)SvIOKp_on(sv);
2308 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2309 if (Perl_isnan(SvNVX(sv))) {
2315 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2316 SvIV_set(sv, I_V(SvNVX(sv)));
2317 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2320 NOOP; /* Integer is imprecise. NOK, IOKp */
2322 /* UV will not work better than IV */
2324 if (SvNVX(sv) > (NV)UV_MAX) {
2326 /* Integer is inaccurate. NOK, IOKp, is UV */
2327 SvUV_set(sv, UV_MAX);
2329 SvUV_set(sv, U_V(SvNVX(sv)));
2330 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2331 NV preservse UV so can do correct comparison. */
2332 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2335 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2340 #else /* NV_PRESERVES_UV */
2341 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2342 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2343 /* The IV/UV slot will have been set from value returned by
2344 grok_number above. The NV slot has just been set using
2347 assert (SvIOKp(sv));
2349 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2350 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2351 /* Small enough to preserve all bits. */
2352 (void)SvIOKp_on(sv);
2354 SvIV_set(sv, I_V(SvNVX(sv)));
2355 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2357 /* Assumption: first non-preserved integer is < IV_MAX,
2358 this NV is in the preserved range, therefore: */
2359 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2361 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);
2365 0 0 already failed to read UV.
2366 0 1 already failed to read UV.
2367 1 0 you won't get here in this case. IV/UV
2368 slot set, public IOK, Atof() unneeded.
2369 1 1 already read UV.
2370 so there's no point in sv_2iuv_non_preserve() attempting
2371 to use atol, strtol, strtoul etc. */
2373 sv_2iuv_non_preserve (sv, numtype);
2375 sv_2iuv_non_preserve (sv);
2379 #endif /* NV_PRESERVES_UV */
2380 /* It might be more code efficient to go through the entire logic above
2381 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2382 gets complex and potentially buggy, so more programmer efficient
2383 to do it this way, by turning off the public flags: */
2385 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2389 if (isGV_with_GP(sv))
2390 return glob_2number(MUTABLE_GV(sv));
2392 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2394 if (SvTYPE(sv) < SVt_IV)
2395 /* Typically the caller expects that sv_any is not NULL now. */
2396 sv_upgrade(sv, SVt_IV);
2397 /* Return 0 from the caller. */
2404 =for apidoc sv_2iv_flags
2406 Return the integer value of an SV, doing any necessary string
2407 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2408 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2414 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2416 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2418 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2419 && SvTYPE(sv) != SVt_PVFM);
2421 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2427 if (flags & SV_SKIP_OVERLOAD)
2429 tmpstr = AMG_CALLunary(sv, numer_amg);
2430 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2431 return SvIV(tmpstr);
2434 return PTR2IV(SvRV(sv));
2437 if (SvVALID(sv) || isREGEXP(sv)) {
2438 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2439 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2440 In practice they are extremely unlikely to actually get anywhere
2441 accessible by user Perl code - the only way that I'm aware of is when
2442 a constant subroutine which is used as the second argument to index.
2444 Regexps have no SvIVX and SvNVX fields.
2446 assert(isREGEXP(sv) || SvPOKp(sv));
2449 const char * const ptr =
2450 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2452 = grok_number(ptr, SvCUR(sv), &value);
2454 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2455 == IS_NUMBER_IN_UV) {
2456 /* It's definitely an integer */
2457 if (numtype & IS_NUMBER_NEG) {
2458 if (value < (UV)IV_MIN)
2461 if (value < (UV)IV_MAX)
2466 /* Quite wrong but no good choices. */
2467 if ((numtype & IS_NUMBER_INFINITY)) {
2468 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2469 } else if ((numtype & IS_NUMBER_NAN)) {
2470 return 0; /* So wrong. */
2474 if (ckWARN(WARN_NUMERIC))
2477 return I_V(Atof(ptr));
2481 if (SvTHINKFIRST(sv)) {
2482 #ifdef PERL_OLD_COPY_ON_WRITE
2484 sv_force_normal_flags(sv, 0);
2487 if (SvREADONLY(sv) && !SvOK(sv)) {
2488 if (ckWARN(WARN_UNINITIALIZED))
2495 if (S_sv_2iuv_common(aTHX_ sv))
2499 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2500 PTR2UV(sv),SvIVX(sv)));
2501 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2505 =for apidoc sv_2uv_flags
2507 Return the unsigned integer value of an SV, doing any necessary string
2508 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2509 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2515 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2517 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2519 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2525 if (flags & SV_SKIP_OVERLOAD)
2527 tmpstr = AMG_CALLunary(sv, numer_amg);
2528 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2529 return SvUV(tmpstr);
2532 return PTR2UV(SvRV(sv));
2535 if (SvVALID(sv) || isREGEXP(sv)) {
2536 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2537 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2538 Regexps have no SvIVX and SvNVX fields. */
2539 assert(isREGEXP(sv) || SvPOKp(sv));
2542 const char * const ptr =
2543 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2545 = grok_number(ptr, SvCUR(sv), &value);
2547 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2548 == IS_NUMBER_IN_UV) {
2549 /* It's definitely an integer */
2550 if (!(numtype & IS_NUMBER_NEG))
2554 /* Quite wrong but no good choices. */
2555 if ((numtype & IS_NUMBER_INFINITY)) {
2556 return UV_MAX; /* So wrong. */
2557 } else if ((numtype & IS_NUMBER_NAN)) {
2558 return 0; /* So wrong. */
2562 if (ckWARN(WARN_NUMERIC))
2565 return U_V(Atof(ptr));
2569 if (SvTHINKFIRST(sv)) {
2570 #ifdef PERL_OLD_COPY_ON_WRITE
2572 sv_force_normal_flags(sv, 0);
2575 if (SvREADONLY(sv) && !SvOK(sv)) {
2576 if (ckWARN(WARN_UNINITIALIZED))
2583 if (S_sv_2iuv_common(aTHX_ sv))
2587 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2588 PTR2UV(sv),SvUVX(sv)));
2589 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2593 =for apidoc sv_2nv_flags
2595 Return the num value of an SV, doing any necessary string or integer
2596 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2597 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2603 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2605 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2607 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2608 && SvTYPE(sv) != SVt_PVFM);
2609 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2610 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2611 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2612 Regexps have no SvIVX and SvNVX fields. */
2614 if (flags & SV_GMAGIC)
2618 if (SvPOKp(sv) && !SvIOKp(sv)) {
2619 ptr = SvPVX_const(sv);
2621 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2622 !grok_number(ptr, SvCUR(sv), NULL))
2628 return (NV)SvUVX(sv);
2630 return (NV)SvIVX(sv);
2636 ptr = RX_WRAPPED((REGEXP *)sv);
2639 assert(SvTYPE(sv) >= SVt_PVMG);
2640 /* This falls through to the report_uninit near the end of the
2642 } else if (SvTHINKFIRST(sv)) {
2647 if (flags & SV_SKIP_OVERLOAD)
2649 tmpstr = AMG_CALLunary(sv, numer_amg);
2650 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2651 return SvNV(tmpstr);
2654 return PTR2NV(SvRV(sv));
2656 #ifdef PERL_OLD_COPY_ON_WRITE
2658 sv_force_normal_flags(sv, 0);
2661 if (SvREADONLY(sv) && !SvOK(sv)) {
2662 if (ckWARN(WARN_UNINITIALIZED))
2667 if (SvTYPE(sv) < SVt_NV) {
2668 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2669 sv_upgrade(sv, SVt_NV);
2671 STORE_NUMERIC_LOCAL_SET_STANDARD();
2672 PerlIO_printf(Perl_debug_log,
2673 "0x%"UVxf" num(%" NVgf ")\n",
2674 PTR2UV(sv), SvNVX(sv));
2675 RESTORE_NUMERIC_LOCAL();
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2685 #ifdef NV_PRESERVES_UV
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2701 else if (SvPOKp(sv)) {
2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2706 #ifdef NV_PRESERVES_UV
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
2709 /* It's definitely an integer */
2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2712 S_sv_setnv(aTHX_ sv, numtype);
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2734 /* value has been set. It may not be precise. */
2735 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
2742 if (numtype & IS_NUMBER_NEG) {
2743 SvIV_set(sv, -(IV)value);
2744 } else if (value <= (UV)IV_MAX) {
2745 SvIV_set(sv, (IV)value);
2747 SvUV_set(sv, value);
2751 if (numtype & IS_NUMBER_NOT_INT) {
2752 /* I believe that even if the original PV had decimals,
2753 they are lost beyond the limit of the FP precision.
2754 However, neither is canonical, so both only get p
2755 flags. NWC, 2000/11/25 */
2756 /* Both already have p flags, so do nothing */
2758 const NV nv = SvNVX(sv);
2759 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2760 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2761 if (SvIVX(sv) == I_V(nv)) {
2764 /* It had no "." so it must be integer. */
2768 /* between IV_MAX and NV(UV_MAX).
2769 Could be slightly > UV_MAX */
2771 if (numtype & IS_NUMBER_NOT_INT) {
2772 /* UV and NV both imprecise. */
2774 const UV nv_as_uv = U_V(nv);
2776 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2785 /* It might be more code efficient to go through the entire logic above
2786 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2787 gets complex and potentially buggy, so more programmer efficient
2788 to do it this way, by turning off the public flags: */
2790 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2791 #endif /* NV_PRESERVES_UV */
2794 if (isGV_with_GP(sv)) {
2795 glob_2number(MUTABLE_GV(sv));
2799 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2801 assert (SvTYPE(sv) >= SVt_NV);
2802 /* Typically the caller expects that sv_any is not NULL now. */
2803 /* XXX Ilya implies that this is a bug in callers that assume this
2804 and ideally should be fixed. */
2808 STORE_NUMERIC_LOCAL_SET_STANDARD();
2809 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2810 PTR2UV(sv), SvNVX(sv));
2811 RESTORE_NUMERIC_LOCAL();
2819 Return an SV with the numeric value of the source SV, doing any necessary
2820 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2821 access this function.
2827 Perl_sv_2num(pTHX_ SV *const sv)
2829 PERL_ARGS_ASSERT_SV_2NUM;
2834 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2835 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2836 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2837 return sv_2num(tmpsv);
2839 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2842 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2843 * UV as a string towards the end of buf, and return pointers to start and
2846 * We assume that buf is at least TYPE_CHARS(UV) long.
2850 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2852 char *ptr = buf + TYPE_CHARS(UV);
2853 char * const ebuf = ptr;
2856 PERL_ARGS_ASSERT_UIV_2BUF;
2868 *--ptr = '0' + (char)(uv % 10);
2876 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2877 * infinity or a not-a-number, writes the appropriate strings to the
2878 * buffer, including a zero byte. On success returns the written length,
2879 * excluding the zero byte, on failure (not an infinity, not a nan, or the
2880 * maxlen too small) returns zero.
2882 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2883 * shared string constants we point to, instead of generating a new
2884 * string for each instance. */
2886 S_infnan_2pv(NV nv, char* buffer, size_t maxlen) {
2887 assert(maxlen >= 4);
2888 if (maxlen < 4) /* "Inf\0", "NaN\0" */
2892 if (Perl_isinf(nv)) {
2894 if (maxlen < 5) /* "-Inf\0" */
2901 } else if (Perl_isnan(nv)) {
2905 /* XXX optionally output the payload mantissa bits as
2906 * "(unsigned)" (to match the nan("...") C99 function,
2907 * or maybe as "(0xhhh...)" would make more sense...
2908 * provide a format string so that the user can decide?
2909 * NOTE: would affect the maxlen and assert() logic.*/
2914 assert((s == buffer + 3) || (s == buffer + 4));
2916 return s - buffer - 1; /* -1: excluding the zero byte */
2921 =for apidoc sv_2pv_flags
2923 Returns a pointer to the string value of an SV, and sets *lp to its length.
2924 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2925 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2926 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2932 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2936 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2938 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2939 && SvTYPE(sv) != SVt_PVFM);
2940 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2945 if (flags & SV_SKIP_OVERLOAD)
2947 tmpstr = AMG_CALLunary(sv, string_amg);
2948 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2949 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2951 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2955 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2956 if (flags & SV_CONST_RETURN) {
2957 pv = (char *) SvPVX_const(tmpstr);
2959 pv = (flags & SV_MUTABLE_RETURN)
2960 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2963 *lp = SvCUR(tmpstr);
2965 pv = sv_2pv_flags(tmpstr, lp, flags);
2978 SV *const referent = SvRV(sv);
2982 retval = buffer = savepvn("NULLREF", len);
2983 } else if (SvTYPE(referent) == SVt_REGEXP &&
2984 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2985 amagic_is_enabled(string_amg))) {
2986 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2990 /* If the regex is UTF-8 we want the containing scalar to
2991 have an UTF-8 flag too */
2998 *lp = RX_WRAPLEN(re);
3000 return RX_WRAPPED(re);
3002 const char *const typestr = sv_reftype(referent, 0);
3003 const STRLEN typelen = strlen(typestr);
3004 UV addr = PTR2UV(referent);
3005 const char *stashname = NULL;
3006 STRLEN stashnamelen = 0; /* hush, gcc */
3007 const char *buffer_end;
3009 if (SvOBJECT(referent)) {
3010 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3013 stashname = HEK_KEY(name);
3014 stashnamelen = HEK_LEN(name);
3016 if (HEK_UTF8(name)) {
3022 stashname = "__ANON__";
3025 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3026 + 2 * sizeof(UV) + 2 /* )\0 */;
3028 len = typelen + 3 /* (0x */
3029 + 2 * sizeof(UV) + 2 /* )\0 */;
3032 Newx(buffer, len, char);
3033 buffer_end = retval = buffer + len;
3035 /* Working backwards */
3039 *--retval = PL_hexdigit[addr & 15];
3040 } while (addr >>= 4);
3046 memcpy(retval, typestr, typelen);
3050 retval -= stashnamelen;
3051 memcpy(retval, stashname, stashnamelen);
3053 /* retval may not necessarily have reached the start of the
3055 assert (retval >= buffer);
3057 len = buffer_end - retval - 1; /* -1 for that \0 */
3069 if (flags & SV_MUTABLE_RETURN)
3070 return SvPVX_mutable(sv);
3071 if (flags & SV_CONST_RETURN)
3072 return (char *)SvPVX_const(sv);
3077 /* I'm assuming that if both IV and NV are equally valid then
3078 converting the IV is going to be more efficient */
3079 const U32 isUIOK = SvIsUV(sv);
3080 char buf[TYPE_CHARS(UV)];
3084 if (SvTYPE(sv) < SVt_PVIV)
3085 sv_upgrade(sv, SVt_PVIV);
3086 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3088 /* inlined from sv_setpvn */
3089 s = SvGROW_mutable(sv, len + 1);
3090 Move(ptr, s, len, char);
3095 else if (SvNOK(sv)) {
3096 if (SvTYPE(sv) < SVt_PVNV)
3097 sv_upgrade(sv, SVt_PVNV);
3098 if (SvNVX(sv) == 0.0
3099 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3100 && !Perl_isnan(SvNVX(sv))
3103 s = SvGROW_mutable(sv, 2);
3108 STRLEN size = 5; /* "-Inf\0" */
3110 s = SvGROW_mutable(sv, size);
3111 len = S_infnan_2pv(SvNVX(sv), s, size);
3117 /* some Xenix systems wipe out errno here */
3126 5 + /* exponent digits */
3130 s = SvGROW_mutable(sv, size);
3131 #ifndef USE_LOCALE_NUMERIC
3132 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3138 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3142 PL_numeric_radix_sv &&
3143 SvUTF8(PL_numeric_radix_sv);
3144 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3145 size += SvLEN(PL_numeric_radix_sv) - 1;
3146 s = SvGROW_mutable(sv, size);
3149 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3151 /* If the radix character is UTF-8, and actually is in the
3152 * output, turn on the UTF-8 flag for the scalar */
3154 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3158 RESTORE_LC_NUMERIC();
3161 /* We don't call SvPOK_on(), because it may come to
3162 * pass that the locale changes so that the
3163 * stringification we just did is no longer correct. We
3164 * will have to re-stringify every time it is needed */
3171 else if (isGV_with_GP(sv)) {
3172 GV *const gv = MUTABLE_GV(sv);
3173 SV *const buffer = sv_newmortal();
3175 gv_efullname3(buffer, gv, "*");
3177 assert(SvPOK(buffer));
3181 *lp = SvCUR(buffer);
3182 return SvPVX(buffer);
3184 else if (isREGEXP(sv)) {
3185 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3186 return RX_WRAPPED((REGEXP *)sv);
3191 if (flags & SV_UNDEF_RETURNS_NULL)
3193 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3195 /* Typically the caller expects that sv_any is not NULL now. */
3196 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3197 sv_upgrade(sv, SVt_PV);
3202 const STRLEN len = s - SvPVX_const(sv);
3207 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3208 PTR2UV(sv),SvPVX_const(sv)));
3209 if (flags & SV_CONST_RETURN)
3210 return (char *)SvPVX_const(sv);
3211 if (flags & SV_MUTABLE_RETURN)
3212 return SvPVX_mutable(sv);
3217 =for apidoc sv_copypv
3219 Copies a stringified representation of the source SV into the
3220 destination SV. Automatically performs any necessary mg_get and
3221 coercion of numeric values into strings. Guaranteed to preserve
3222 UTF8 flag even from overloaded objects. Similar in nature to
3223 sv_2pv[_flags] but operates directly on an SV instead of just the
3224 string. Mostly uses sv_2pv_flags to do its work, except when that
3225 would lose the UTF-8'ness of the PV.
3227 =for apidoc sv_copypv_nomg
3229 Like sv_copypv, but doesn't invoke get magic first.
3231 =for apidoc sv_copypv_flags
3233 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3240 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3242 PERL_ARGS_ASSERT_SV_COPYPV;
3244 sv_copypv_flags(dsv, ssv, 0);
3248 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3253 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3255 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3256 sv_setpvn(dsv,s,len);
3264 =for apidoc sv_2pvbyte
3266 Return a pointer to the byte-encoded representation of the SV, and set *lp
3267 to its length. May cause the SV to be downgraded from UTF-8 as a
3270 Usually accessed via the C<SvPVbyte> macro.
3276 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3278 PERL_ARGS_ASSERT_SV_2PVBYTE;
3281 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3282 || isGV_with_GP(sv) || SvROK(sv)) {
3283 SV *sv2 = sv_newmortal();
3284 sv_copypv_nomg(sv2,sv);
3287 sv_utf8_downgrade(sv,0);
3288 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3292 =for apidoc sv_2pvutf8
3294 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3295 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3297 Usually accessed via the C<SvPVutf8> macro.
3303 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3305 PERL_ARGS_ASSERT_SV_2PVUTF8;
3307 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3308 || isGV_with_GP(sv) || SvROK(sv))
3309 sv = sv_mortalcopy(sv);
3312 sv_utf8_upgrade_nomg(sv);
3313 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3318 =for apidoc sv_2bool
3320 This macro is only used by sv_true() or its macro equivalent, and only if
3321 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3322 It calls sv_2bool_flags with the SV_GMAGIC flag.
3324 =for apidoc sv_2bool_flags
3326 This function is only used by sv_true() and friends, and only if
3327 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3328 contain SV_GMAGIC, then it does an mg_get() first.
3335 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3337 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3340 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3346 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3347 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3350 if(SvGMAGICAL(sv)) {
3352 goto restart; /* call sv_2bool */
3354 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3355 else if(!SvOK(sv)) {
3358 else if(SvPOK(sv)) {
3359 svb = SvPVXtrue(sv);
3361 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3362 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3363 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3367 goto restart; /* call sv_2bool_nomg */
3372 return SvRV(sv) != 0;
3376 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3377 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3381 =for apidoc sv_utf8_upgrade
3383 Converts the PV of an SV to its UTF-8-encoded form.
3384 Forces the SV to string form if it is not already.
3385 Will C<mg_get> on C<sv> if appropriate.
3386 Always sets the SvUTF8 flag to avoid future validity checks even
3387 if the whole string is the same in UTF-8 as not.
3388 Returns the number of bytes in the converted string
3390 This is not a general purpose byte encoding to Unicode interface:
3391 use the Encode extension for that.
3393 =for apidoc sv_utf8_upgrade_nomg
3395 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3397 =for apidoc sv_utf8_upgrade_flags
3399 Converts the PV of an SV to its UTF-8-encoded form.
3400 Forces the SV to string form if it is not already.
3401 Always sets the SvUTF8 flag to avoid future validity checks even
3402 if all the bytes are invariant in UTF-8.
3403 If C<flags> has C<SV_GMAGIC> bit set,
3404 will C<mg_get> on C<sv> if appropriate, else not.
3406 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3407 will expand when converted to UTF-8, and skips the extra work of checking for
3408 that. Typically this flag is used by a routine that has already parsed the
3409 string and found such characters, and passes this information on so that the
3410 work doesn't have to be repeated.
3412 Returns the number of bytes in the converted string.
3414 This is not a general purpose byte encoding to Unicode interface:
3415 use the Encode extension for that.
3417 =for apidoc sv_utf8_upgrade_flags_grow
3419 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3420 the number of unused bytes the string of 'sv' is guaranteed to have free after
3421 it upon return. This allows the caller to reserve extra space that it intends
3422 to fill, to avoid extra grows.
3424 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3425 are implemented in terms of this function.
3427 Returns the number of bytes in the converted string (not including the spares).
3431 (One might think that the calling routine could pass in the position of the
3432 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3433 have to be found again. But that is not the case, because typically when the
3434 caller is likely to use this flag, it won't be calling this routine unless it
3435 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3436 and just use bytes. But some things that do fit into a byte are variants in
3437 utf8, and the caller may not have been keeping track of these.)
3439 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3440 C<NUL> isn't guaranteed due to having other routines do the work in some input
3441 cases, or if the input is already flagged as being in utf8.
3443 The speed of this could perhaps be improved for many cases if someone wanted to
3444 write a fast function that counts the number of variant characters in a string,
3445 especially if it could return the position of the first one.
3450 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3452 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3454 if (sv == &PL_sv_undef)
3456 if (!SvPOK_nog(sv)) {
3458 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3459 (void) sv_2pv_flags(sv,&len, flags);
3461 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3465 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3470 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3475 S_sv_uncow(aTHX_ sv, 0);
3478 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3479 sv_recode_to_utf8(sv, _get_encoding());
3480 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3484 if (SvCUR(sv) == 0) {
3485 if (extra) SvGROW(sv, extra);
3486 } else { /* Assume Latin-1/EBCDIC */
3487 /* This function could be much more efficient if we
3488 * had a FLAG in SVs to signal if there are any variant
3489 * chars in the PV. Given that there isn't such a flag
3490 * make the loop as fast as possible (although there are certainly ways
3491 * to speed this up, eg. through vectorization) */
3492 U8 * s = (U8 *) SvPVX_const(sv);
3493 U8 * e = (U8 *) SvEND(sv);
3495 STRLEN two_byte_count = 0;
3497 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3499 /* See if really will need to convert to utf8. We mustn't rely on our
3500 * incoming SV being well formed and having a trailing '\0', as certain
3501 * code in pp_formline can send us partially built SVs. */
3505 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3507 t--; /* t already incremented; re-point to first variant */
3512 /* utf8 conversion not needed because all are invariants. Mark as
3513 * UTF-8 even if no variant - saves scanning loop */
3515 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3520 /* Here, the string should be converted to utf8, either because of an
3521 * input flag (two_byte_count = 0), or because a character that
3522 * requires 2 bytes was found (two_byte_count = 1). t points either to
3523 * the beginning of the string (if we didn't examine anything), or to
3524 * the first variant. In either case, everything from s to t - 1 will
3525 * occupy only 1 byte each on output.
3527 * There are two main ways to convert. One is to create a new string
3528 * and go through the input starting from the beginning, appending each
3529 * converted value onto the new string as we go along. It's probably
3530 * best to allocate enough space in the string for the worst possible
3531 * case rather than possibly running out of space and having to
3532 * reallocate and then copy what we've done so far. Since everything
3533 * from s to t - 1 is invariant, the destination can be initialized
3534 * with these using a fast memory copy
3536 * The other way is to figure out exactly how big the string should be
3537 * by parsing the entire input. Then you don't have to make it big
3538 * enough to handle the worst possible case, and more importantly, if
3539 * the string you already have is large enough, you don't have to
3540 * allocate a new string, you can copy the last character in the input
3541 * string to the final position(s) that will be occupied by the
3542 * converted string and go backwards, stopping at t, since everything
3543 * before that is invariant.
3545 * There are advantages and disadvantages to each method.
3547 * In the first method, we can allocate a new string, do the memory
3548 * copy from the s to t - 1, and then proceed through the rest of the
3549 * string byte-by-byte.
3551 * In the second method, we proceed through the rest of the input
3552 * string just calculating how big the converted string will be. Then
3553 * there are two cases:
3554 * 1) if the string has enough extra space to handle the converted
3555 * value. We go backwards through the string, converting until we
3556 * get to the position we are at now, and then stop. If this
3557 * position is far enough along in the string, this method is
3558 * faster than the other method. If the memory copy were the same
3559 * speed as the byte-by-byte loop, that position would be about
3560 * half-way, as at the half-way mark, parsing to the end and back
3561 * is one complete string's parse, the same amount as starting
3562 * over and going all the way through. Actually, it would be
3563 * somewhat less than half-way, as it's faster to just count bytes
3564 * than to also copy, and we don't have the overhead of allocating
3565 * a new string, changing the scalar to use it, and freeing the
3566 * existing one. But if the memory copy is fast, the break-even
3567 * point is somewhere after half way. The counting loop could be
3568 * sped up by vectorization, etc, to move the break-even point
3569 * further towards the beginning.
3570 * 2) if the string doesn't have enough space to handle the converted
3571 * value. A new string will have to be allocated, and one might
3572 * as well, given that, start from the beginning doing the first
3573 * method. We've spent extra time parsing the string and in
3574 * exchange all we've gotten is that we know precisely how big to
3575 * make the new one. Perl is more optimized for time than space,
3576 * so this case is a loser.
3577 * So what I've decided to do is not use the 2nd method unless it is
3578 * guaranteed that a new string won't have to be allocated, assuming
3579 * the worst case. I also decided not to put any more conditions on it
3580 * than this, for now. It seems likely that, since the worst case is
3581 * twice as big as the unknown portion of the string (plus 1), we won't
3582 * be guaranteed enough space, causing us to go to the first method,
3583 * unless the string is short, or the first variant character is near
3584 * the end of it. In either of these cases, it seems best to use the
3585 * 2nd method. The only circumstance I can think of where this would
3586 * be really slower is if the string had once had much more data in it
3587 * than it does now, but there is still a substantial amount in it */
3590 STRLEN invariant_head = t - s;
3591 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3592 if (SvLEN(sv) < size) {
3594 /* Here, have decided to allocate a new string */
3599 Newx(dst, size, U8);
3601 /* If no known invariants at the beginning of the input string,
3602 * set so starts from there. Otherwise, can use memory copy to
3603 * get up to where we are now, and then start from here */
3605 if (invariant_head == 0) {
3608 Copy(s, dst, invariant_head, char);
3609 d = dst + invariant_head;
3613 append_utf8_from_native_byte(*t, &d);
3617 SvPV_free(sv); /* No longer using pre-existing string */
3618 SvPV_set(sv, (char*)dst);
3619 SvCUR_set(sv, d - dst);
3620 SvLEN_set(sv, size);
3623 /* Here, have decided to get the exact size of the string.
3624 * Currently this happens only when we know that there is
3625 * guaranteed enough space to fit the converted string, so
3626 * don't have to worry about growing. If two_byte_count is 0,
3627 * then t points to the first byte of the string which hasn't
3628 * been examined yet. Otherwise two_byte_count is 1, and t
3629 * points to the first byte in the string that will expand to
3630 * two. Depending on this, start examining at t or 1 after t.
3633 U8 *d = t + two_byte_count;
3636 /* Count up the remaining bytes that expand to two */
3639 const U8 chr = *d++;
3640 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3643 /* The string will expand by just the number of bytes that
3644 * occupy two positions. But we are one afterwards because of
3645 * the increment just above. This is the place to put the
3646 * trailing NUL, and to set the length before we decrement */
3648 d += two_byte_count;
3649 SvCUR_set(sv, d - s);
3653 /* Having decremented d, it points to the position to put the
3654 * very last byte of the expanded string. Go backwards through
3655 * the string, copying and expanding as we go, stopping when we
3656 * get to the part that is invariant the rest of the way down */
3660 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3663 *d-- = UTF8_EIGHT_BIT_LO(*e);
3664 *d-- = UTF8_EIGHT_BIT_HI(*e);
3670 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3671 /* Update pos. We do it at the end rather than during
3672 * the upgrade, to avoid slowing down the common case
3673 * (upgrade without pos).
3674 * pos can be stored as either bytes or characters. Since
3675 * this was previously a byte string we can just turn off
3676 * the bytes flag. */
3677 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3679 mg->mg_flags &= ~MGf_BYTES;
3681 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3682 magic_setutf8(sv,mg); /* clear UTF8 cache */
3687 /* Mark as UTF-8 even if no variant - saves scanning loop */
3693 =for apidoc sv_utf8_downgrade
3695 Attempts to convert the PV of an SV from characters to bytes.
3696 If the PV contains a character that cannot fit
3697 in a byte, this conversion will fail;
3698 in this case, either returns false or, if C<fail_ok> is not
3701 This is not a general purpose Unicode to byte encoding interface:
3702 use the Encode extension for that.
3708 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3710 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3712 if (SvPOKp(sv) && SvUTF8(sv)) {
3716 int mg_flags = SV_GMAGIC;
3719 S_sv_uncow(aTHX_ sv, 0);
3721 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3723 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3724 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3725 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3726 SV_GMAGIC|SV_CONST_RETURN);
3727 mg_flags = 0; /* sv_pos_b2u does get magic */
3729 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3730 magic_setutf8(sv,mg); /* clear UTF8 cache */
3733 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3735 if (!utf8_to_bytes(s, &len)) {
3740 Perl_croak(aTHX_ "Wide character in %s",
3743 Perl_croak(aTHX_ "Wide character");
3754 =for apidoc sv_utf8_encode
3756 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3757 flag off so that it looks like octets again.
3763 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3765 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3767 if (SvREADONLY(sv)) {
3768 sv_force_normal_flags(sv, 0);
3770 (void) sv_utf8_upgrade(sv);
3775 =for apidoc sv_utf8_decode
3777 If the PV of the SV is an octet sequence in UTF-8
3778 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3779 so that it looks like a character. If the PV contains only single-byte
3780 characters, the C<SvUTF8> flag stays off.
3781 Scans PV for validity and returns false if the PV is invalid UTF-8.
3787 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3789 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3792 const U8 *start, *c;
3795 /* The octets may have got themselves encoded - get them back as
3798 if (!sv_utf8_downgrade(sv, TRUE))
3801 /* it is actually just a matter of turning the utf8 flag on, but
3802 * we want to make sure everything inside is valid utf8 first.
3804 c = start = (const U8 *) SvPVX_const(sv);
3805 if (!is_utf8_string(c, SvCUR(sv)))
3807 e = (const U8 *) SvEND(sv);
3810 if (!UTF8_IS_INVARIANT(ch)) {
3815 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3816 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3817 after this, clearing pos. Does anything on CPAN
3819 /* adjust pos to the start of a UTF8 char sequence */
3820 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3822 I32 pos = mg->mg_len;
3824 for (c = start + pos; c > start; c--) {
3825 if (UTF8_IS_START(*c))
3828 mg->mg_len = c - start;
3831 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3832 magic_setutf8(sv,mg); /* clear UTF8 cache */
3839 =for apidoc sv_setsv
3841 Copies the contents of the source SV C<ssv> into the destination SV
3842 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3843 function if the source SV needs to be reused. Does not handle 'set' magic on
3844 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3845 performs a copy-by-value, obliterating any previous content of the
3848 You probably want to use one of the assortment of wrappers, such as
3849 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3850 C<SvSetMagicSV_nosteal>.
3852 =for apidoc sv_setsv_flags
3854 Copies the contents of the source SV C<ssv> into the destination SV
3855 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3856 function if the source SV needs to be reused. Does not handle 'set' magic.
3857 Loosely speaking, it performs a copy-by-value, obliterating any previous
3858 content of the destination.
3859 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3860 C<ssv> if appropriate, else not. If the C<flags>
3861 parameter has the C<SV_NOSTEAL> bit set then the
3862 buffers of temps will not be stolen. <sv_setsv>
3863 and C<sv_setsv_nomg> are implemented in terms of this function.
3865 You probably want to use one of the assortment of wrappers, such as
3866 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3867 C<SvSetMagicSV_nosteal>.
3869 This is the primary function for copying scalars, and most other
3870 copy-ish functions and macros use this underneath.
3876 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3878 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3879 HV *old_stash = NULL;
3881 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3883 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3884 const char * const name = GvNAME(sstr);
3885 const STRLEN len = GvNAMELEN(sstr);
3887 if (dtype >= SVt_PV) {
3893 SvUPGRADE(dstr, SVt_PVGV);
3894 (void)SvOK_off(dstr);
3895 isGV_with_GP_on(dstr);
3897 GvSTASH(dstr) = GvSTASH(sstr);
3899 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3900 gv_name_set(MUTABLE_GV(dstr), name, len,
3901 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3902 SvFAKE_on(dstr); /* can coerce to non-glob */
3905 if(GvGP(MUTABLE_GV(sstr))) {
3906 /* If source has method cache entry, clear it */
3908 SvREFCNT_dec(GvCV(sstr));
3909 GvCV_set(sstr, NULL);
3912 /* If source has a real method, then a method is
3915 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3921 /* If dest already had a real method, that's a change as well */
3923 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3924 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3929 /* We don't need to check the name of the destination if it was not a
3930 glob to begin with. */
3931 if(dtype == SVt_PVGV) {
3932 const char * const name = GvNAME((const GV *)dstr);
3935 /* The stash may have been detached from the symbol table, so
3937 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3941 const STRLEN len = GvNAMELEN(dstr);
3942 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3943 || (len == 1 && name[0] == ':')) {
3946 /* Set aside the old stash, so we can reset isa caches on
3948 if((old_stash = GvHV(dstr)))
3949 /* Make sure we do not lose it early. */
3950 SvREFCNT_inc_simple_void_NN(
3951 sv_2mortal((SV *)old_stash)
3956 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3959 gp_free(MUTABLE_GV(dstr));
3960 GvINTRO_off(dstr); /* one-shot flag */
3961 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3962 if (SvTAINTED(sstr))
3964 if (GvIMPORTED(dstr) != GVf_IMPORTED
3965 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3967 GvIMPORTED_on(dstr);
3970 if(mro_changes == 2) {
3971 if (GvAV((const GV *)sstr)) {
3973 SV * const sref = (SV *)GvAV((const GV *)dstr);
3974 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3975 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3976 AV * const ary = newAV();
3977 av_push(ary, mg->mg_obj); /* takes the refcount */
3978 mg->mg_obj = (SV *)ary;
3980 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3982 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3984 mro_isa_changed_in(GvSTASH(dstr));
3986 else if(mro_changes == 3) {
3987 HV * const stash = GvHV(dstr);
3988 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3994 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3995 if (GvIO(dstr) && dtype == SVt_PVGV) {
3996 DEBUG_o(Perl_deb(aTHX_
3997 "glob_assign_glob clearing PL_stashcache\n"));
3998 /* It's a cache. It will rebuild itself quite happily.
3999 It's a lot of effort to work out exactly which key (or keys)
4000 might be invalidated by the creation of the this file handle.
4002 hv_clear(PL_stashcache);
4008 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4010 SV * const sref = SvRV(sstr);
4012 const int intro = GvINTRO(dstr);
4015 const U32 stype = SvTYPE(sref);
4017 PERL_ARGS_ASSERT_GV_SETREF;
4020 GvINTRO_off(dstr); /* one-shot flag */
4021 GvLINE(dstr) = CopLINE(PL_curcop);
4022 GvEGV(dstr) = MUTABLE_GV(dstr);
4027 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4028 import_flag = GVf_IMPORTED_CV;
4031 location = (SV **) &GvHV(dstr);
4032 import_flag = GVf_IMPORTED_HV;
4035 location = (SV **) &GvAV(dstr);
4036 import_flag = GVf_IMPORTED_AV;
4039 location = (SV **) &GvIOp(dstr);
4042 location = (SV **) &GvFORM(dstr);
4045 location = &GvSV(dstr);
4046 import_flag = GVf_IMPORTED_SV;
4049 if (stype == SVt_PVCV) {
4050 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4051 if (GvCVGEN(dstr)) {
4052 SvREFCNT_dec(GvCV(dstr));
4053 GvCV_set(dstr, NULL);
4054 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4057 /* SAVEt_GVSLOT takes more room on the savestack and has more
4058 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4059 leave_scope needs access to the GV so it can reset method
4060 caches. We must use SAVEt_GVSLOT whenever the type is
4061 SVt_PVCV, even if the stash is anonymous, as the stash may
4062 gain a name somehow before leave_scope. */
4063 if (stype == SVt_PVCV) {
4064 /* There is no save_pushptrptrptr. Creating it for this
4065 one call site would be overkill. So inline the ss add
4069 SS_ADD_PTR(location);
4070 SS_ADD_PTR(SvREFCNT_inc(*location));
4071 SS_ADD_UV(SAVEt_GVSLOT);
4074 else SAVEGENERICSV(*location);
4077 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4078 CV* const cv = MUTABLE_CV(*location);
4080 if (!GvCVGEN((const GV *)dstr) &&
4081 (CvROOT(cv) || CvXSUB(cv)) &&
4082 /* redundant check that avoids creating the extra SV
4083 most of the time: */
4084 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4086 SV * const new_const_sv =
4087 CvCONST((const CV *)sref)
4088 ? cv_const_sv((const CV *)sref)
4090 report_redefined_cv(
4091 sv_2mortal(Perl_newSVpvf(aTHX_
4094 HvNAME_HEK(GvSTASH((const GV *)dstr))
4096 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4099 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4103 cv_ckproto_len_flags(cv, (const GV *)dstr,
4104 SvPOK(sref) ? CvPROTO(sref) : NULL,
4105 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4106 SvPOK(sref) ? SvUTF8(sref) : 0);
4108 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4109 GvASSUMECV_on(dstr);
4110 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4111 if (intro && GvREFCNT(dstr) > 1) {
4112 /* temporary remove extra savestack's ref */
4114 gv_method_changed(dstr);
4117 else gv_method_changed(dstr);
4120 *location = SvREFCNT_inc_simple_NN(sref);
4121 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4122 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4123 GvFLAGS(dstr) |= import_flag;
4125 if (import_flag == GVf_IMPORTED_SV) {
4127 save_aliased_sv((GV *)dstr);
4129 /* Turn off the flag if sref is not referenced elsewhere,
4130 even by weak refs. (SvRMAGICAL is a pessimistic check for
4132 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4133 GvALIASED_SV_off(dstr);
4135 GvALIASED_SV_on(dstr);
4137 if (stype == SVt_PVHV) {
4138 const char * const name = GvNAME((GV*)dstr);
4139 const STRLEN len = GvNAMELEN(dstr);
4142 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4143 || (len == 1 && name[0] == ':')
4145 && (!dref || HvENAME_get(dref))
4148 (HV *)sref, (HV *)dref,
4154 stype == SVt_PVAV && sref != dref
4155 && strEQ(GvNAME((GV*)dstr), "ISA")
4156 /* The stash may have been detached from the symbol table, so
4157 check its name before doing anything. */
4158 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4161 MAGIC * const omg = dref && SvSMAGICAL(dref)
4162 ? mg_find(dref, PERL_MAGIC_isa)
4164 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4165 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4166 AV * const ary = newAV();
4167 av_push(ary, mg->mg_obj); /* takes the refcount */
4168 mg->mg_obj = (SV *)ary;
4171 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4172 SV **svp = AvARRAY((AV *)omg->mg_obj);
4173 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4177 SvREFCNT_inc_simple_NN(*svp++)
4183 SvREFCNT_inc_simple_NN(omg->mg_obj)
4187 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4192 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4194 mg = mg_find(sref, PERL_MAGIC_isa);
4196 /* Since the *ISA assignment could have affected more than
4197 one stash, don't call mro_isa_changed_in directly, but let
4198 magic_clearisa do it for us, as it already has the logic for
4199 dealing with globs vs arrays of globs. */
4201 Perl_magic_clearisa(aTHX_ NULL, mg);
4203 else if (stype == SVt_PVIO) {
4204 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4205 /* It's a cache. It will rebuild itself quite happily.
4206 It's a lot of effort to work out exactly which key (or keys)
4207 might be invalidated by the creation of the this file handle.
4209 hv_clear(PL_stashcache);
4213 if (!intro) SvREFCNT_dec(dref);
4214 if (SvTAINTED(sstr))
4222 #ifdef PERL_DEBUG_READONLY_COW
4223 # include <sys/mman.h>
4225 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4226 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4230 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4232 struct perl_memory_debug_header * const header =
4233 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4234 const MEM_SIZE len = header->size;
4235 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4236 # ifdef PERL_TRACK_MEMPOOL
4237 if (!header->readonly) header->readonly = 1;
4239 if (mprotect(header, len, PROT_READ))
4240 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4241 header, len, errno);
4245 S_sv_buf_to_rw(pTHX_ SV *sv)
4247 struct perl_memory_debug_header * const header =
4248 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4249 const MEM_SIZE len = header->size;
4250 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4251 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4252 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4253 header, len, errno);
4254 # ifdef PERL_TRACK_MEMPOOL
4255 header->readonly = 0;
4260 # define sv_buf_to_ro(sv) NOOP
4261 # define sv_buf_to_rw(sv) NOOP
4265 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4271 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4273 if (UNLIKELY( sstr == dstr ))
4276 if (SvIS_FREED(dstr)) {
4277 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4278 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4280 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4281 if (UNLIKELY( !sstr ))
4282 sstr = &PL_sv_undef;
4283 if (SvIS_FREED(sstr)) {
4284 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4285 (void*)sstr, (void*)dstr);
4287 stype = SvTYPE(sstr);
4288 dtype = SvTYPE(dstr);
4290 /* There's a lot of redundancy below but we're going for speed here */
4295 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4296 (void)SvOK_off(dstr);
4304 /* For performance, we inline promoting to type SVt_IV. */
4305 /* We're starting from SVt_NULL, so provided that define is
4306 * actual 0, we don't have to unset any SV type flags
4307 * to promote to SVt_IV. */
4308 STATIC_ASSERT_STMT(SVt_NULL == 0);
4309 SET_SVANY_FOR_BODYLESS_IV(dstr);
4310 SvFLAGS(dstr) |= SVt_IV;
4314 sv_upgrade(dstr, SVt_PVIV);
4318 goto end_of_first_switch;
4320 (void)SvIOK_only(dstr);
4321 SvIV_set(dstr, SvIVX(sstr));
4324 /* SvTAINTED can only be true if the SV has taint magic, which in
4325 turn means that the SV type is PVMG (or greater). This is the
4326 case statement for SVt_IV, so this cannot be true (whatever gcov
4328 assert(!SvTAINTED(sstr));
4333 if (dtype < SVt_PV && dtype != SVt_IV)
4334 sv_upgrade(dstr, SVt_IV);
4338 if (LIKELY( SvNOK(sstr) )) {
4342 sv_upgrade(dstr, SVt_NV);
4346 sv_upgrade(dstr, SVt_PVNV);
4350 goto end_of_first_switch;
4352 SvNV_set(dstr, SvNVX(sstr));
4353 (void)SvNOK_only(dstr);
4354 /* SvTAINTED can only be true if the SV has taint magic, which in
4355 turn means that the SV type is PVMG (or greater). This is the
4356 case statement for SVt_NV, so this cannot be true (whatever gcov
4358 assert(!SvTAINTED(sstr));
4365 sv_upgrade(dstr, SVt_PV);
4368 if (dtype < SVt_PVIV)
4369 sv_upgrade(dstr, SVt_PVIV);
4372 if (dtype < SVt_PVNV)
4373 sv_upgrade(dstr, SVt_PVNV);
4377 const char * const type = sv_reftype(sstr,0);
4379 /* diag_listed_as: Bizarre copy of %s */
4380 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4382 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4384 NOT_REACHED; /* NOTREACHED */
4388 if (dtype < SVt_REGEXP)
4390 if (dtype >= SVt_PV) {
4396 sv_upgrade(dstr, SVt_REGEXP);
4404 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4406 if (SvTYPE(sstr) != stype)
4407 stype = SvTYPE(sstr);
4409 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4410 glob_assign_glob(dstr, sstr, dtype);
4413 if (stype == SVt_PVLV)
4415 if (isREGEXP(sstr)) goto upgregexp;
4416 SvUPGRADE(dstr, SVt_PVNV);
4419 SvUPGRADE(dstr, (svtype)stype);
4421 end_of_first_switch:
4423 /* dstr may have been upgraded. */
4424 dtype = SvTYPE(dstr);
4425 sflags = SvFLAGS(sstr);
4427 if (UNLIKELY( dtype == SVt_PVCV )) {
4428 /* Assigning to a subroutine sets the prototype. */
4431 const char *const ptr = SvPV_const(sstr, len);
4433 SvGROW(dstr, len + 1);
4434 Copy(ptr, SvPVX(dstr), len + 1, char);
4435 SvCUR_set(dstr, len);
4437 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4438 CvAUTOLOAD_off(dstr);
4443 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4444 || dtype == SVt_PVFM))
4446 const char * const type = sv_reftype(dstr,0);
4448 /* diag_listed_as: Cannot copy to %s */
4449 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4451 Perl_croak(aTHX_ "Cannot copy to %s", type);
4452 } else if (sflags & SVf_ROK) {
4453 if (isGV_with_GP(dstr)
4454 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4457 if (GvIMPORTED(dstr) != GVf_IMPORTED
4458 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4460 GvIMPORTED_on(dstr);
4465 glob_assign_glob(dstr, sstr, dtype);
4469 if (dtype >= SVt_PV) {
4470 if (isGV_with_GP(dstr)) {
4471 gv_setref(dstr, sstr);
4474 if (SvPVX_const(dstr)) {
4480 (void)SvOK_off(dstr);
4481 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4482 SvFLAGS(dstr) |= sflags & SVf_ROK;
4483 assert(!(sflags & SVp_NOK));
4484 assert(!(sflags & SVp_IOK));
4485 assert(!(sflags & SVf_NOK));
4486 assert(!(sflags & SVf_IOK));
4488 else if (isGV_with_GP(dstr)) {
4489 if (!(sflags & SVf_OK)) {
4490 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4491 "Undefined value assigned to typeglob");
4494 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4495 if (dstr != (const SV *)gv) {
4496 const char * const name = GvNAME((const GV *)dstr);
4497 const STRLEN len = GvNAMELEN(dstr);
4498 HV *old_stash = NULL;
4499 bool reset_isa = FALSE;
4500 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4501 || (len == 1 && name[0] == ':')) {
4502 /* Set aside the old stash, so we can reset isa caches
4503 on its subclasses. */
4504 if((old_stash = GvHV(dstr))) {
4505 /* Make sure we do not lose it early. */
4506 SvREFCNT_inc_simple_void_NN(
4507 sv_2mortal((SV *)old_stash)
4514 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4515 gp_free(MUTABLE_GV(dstr));
4517 GvGP_set(dstr, gp_ref(GvGP(gv)));
4520 HV * const stash = GvHV(dstr);
4522 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4532 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4533 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4534 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4536 else if (sflags & SVp_POK) {
4537 const STRLEN cur = SvCUR(sstr);
4538 const STRLEN len = SvLEN(sstr);
4541 * We have three basic ways to copy the string:
4547 * Which we choose is based on various factors. The following
4548 * things are listed in order of speed, fastest to slowest:
4550 * - Copying a short string
4551 * - Copy-on-write bookkeeping
4553 * - Copying a long string
4555 * We swipe the string (steal the string buffer) if the SV on the
4556 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4557 * big win on long strings. It should be a win on short strings if
4558 * SvPVX_const(dstr) has to be allocated. If not, it should not
4559 * slow things down, as SvPVX_const(sstr) would have been freed
4562 * We also steal the buffer from a PADTMP (operator target) if it
4563 * is ‘long enough’. For short strings, a swipe does not help
4564 * here, as it causes more malloc calls the next time the target
4565 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4566 * be allocated it is still not worth swiping PADTMPs for short
4567 * strings, as the savings here are small.
4569 * If swiping is not an option, then we see whether it is
4570 * worth using copy-on-write. If the lhs already has a buf-
4571 * fer big enough and the string is short, we skip it and fall back
4572 * to method 3, since memcpy is faster for short strings than the
4573 * later bookkeeping overhead that copy-on-write entails.
4575 * If the rhs is not a copy-on-write string yet, then we also
4576 * consider whether the buffer is too large relative to the string
4577 * it holds. Some operations such as readline allocate a large
4578 * buffer in the expectation of reusing it. But turning such into
4579 * a COW buffer is counter-productive because it increases memory
4580 * usage by making readline allocate a new large buffer the sec-
4581 * ond time round. So, if the buffer is too large, again, we use
4584 * Finally, if there is no buffer on the left, or the buffer is too
4585 * small, then we use copy-on-write and make both SVs share the
4590 /* Whichever path we take through the next code, we want this true,
4591 and doing it now facilitates the COW check. */
4592 (void)SvPOK_only(dstr);
4596 /* slated for free anyway (and not COW)? */
4597 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4598 /* or a swipable TARG */
4600 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4602 /* whose buffer is worth stealing */
4603 && CHECK_COWBUF_THRESHOLD(cur,len)
4606 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4607 (!(flags & SV_NOSTEAL)) &&
4608 /* and we're allowed to steal temps */
4609 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4610 len) /* and really is a string */
4611 { /* Passes the swipe test. */
4612 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4614 SvPV_set(dstr, SvPVX_mutable(sstr));
4615 SvLEN_set(dstr, SvLEN(sstr));
4616 SvCUR_set(dstr, SvCUR(sstr));
4619 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4620 SvPV_set(sstr, NULL);
4625 else if (flags & SV_COW_SHARED_HASH_KEYS
4627 #ifdef PERL_OLD_COPY_ON_WRITE
4628 ( sflags & SVf_IsCOW
4629 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4630 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4631 && SvTYPE(sstr) >= SVt_PVIV && len
4634 #elif defined(PERL_NEW_COPY_ON_WRITE)
4637 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4638 /* If this is a regular (non-hek) COW, only so
4639 many COW "copies" are possible. */
4640 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4641 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4642 && !(SvFLAGS(dstr) & SVf_BREAK)
4643 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4644 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4648 && !(SvFLAGS(dstr) & SVf_BREAK)
4651 /* Either it's a shared hash key, or it's suitable for
4654 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4659 if (!(sflags & SVf_IsCOW)) {
4661 # ifdef PERL_OLD_COPY_ON_WRITE
4662 /* Make the source SV into a loop of 1.
4663 (about to become 2) */
4664 SV_COW_NEXT_SV_SET(sstr, sstr);
4666 CowREFCNT(sstr) = 0;
4670 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4676 # ifdef PERL_OLD_COPY_ON_WRITE
4677 assert (SvTYPE(dstr) >= SVt_PVIV);
4678 /* SvIsCOW_normal */
4679 /* splice us in between source and next-after-source. */
4680 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4681 SV_COW_NEXT_SV_SET(sstr, dstr);
4683 if (sflags & SVf_IsCOW) {
4688 SvPV_set(dstr, SvPVX_mutable(sstr));
4693 /* SvIsCOW_shared_hash */
4694 DEBUG_C(PerlIO_printf(Perl_debug_log,
4695 "Copy on write: Sharing hash\n"));
4697 assert (SvTYPE(dstr) >= SVt_PV);
4699 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4701 SvLEN_set(dstr, len);
4702 SvCUR_set(dstr, cur);
4705 /* Failed the swipe test, and we cannot do copy-on-write either.
4706 Have to copy the string. */
4707 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4708 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4709 SvCUR_set(dstr, cur);
4710 *SvEND(dstr) = '\0';
4712 if (sflags & SVp_NOK) {
4713 SvNV_set(dstr, SvNVX(sstr));
4715 if (sflags & SVp_IOK) {
4716 SvIV_set(dstr, SvIVX(sstr));
4717 /* Must do this otherwise some other overloaded use of 0x80000000
4718 gets confused. I guess SVpbm_VALID */
4719 if (sflags & SVf_IVisUV)
4722 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4724 const MAGIC * const smg = SvVSTRING_mg(sstr);
4726 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4727 smg->mg_ptr, smg->mg_len);
4728 SvRMAGICAL_on(dstr);
4732 else if (sflags & (SVp_IOK|SVp_NOK)) {
4733 (void)SvOK_off(dstr);
4734 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4735 if (sflags & SVp_IOK) {
4736 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4737 SvIV_set(dstr, SvIVX(sstr));
4739 if (sflags & SVp_NOK) {
4740 SvNV_set(dstr, SvNVX(sstr));
4744 if (isGV_with_GP(sstr)) {
4745 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4748 (void)SvOK_off(dstr);
4750 if (SvTAINTED(sstr))
4755 =for apidoc sv_setsv_mg
4757 Like C<sv_setsv>, but also handles 'set' magic.
4763 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4765 PERL_ARGS_ASSERT_SV_SETSV_MG;
4767 sv_setsv(dstr,sstr);
4772 # ifdef PERL_OLD_COPY_ON_WRITE
4773 # define SVt_COW SVt_PVIV
4775 # define SVt_COW SVt_PV
4778 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4780 STRLEN cur = SvCUR(sstr);
4781 STRLEN len = SvLEN(sstr);
4783 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4784 const bool already = cBOOL(SvIsCOW(sstr));
4787 PERL_ARGS_ASSERT_SV_SETSV_COW;
4790 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4791 (void*)sstr, (void*)dstr);
4798 if (SvTHINKFIRST(dstr))
4799 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4800 else if (SvPVX_const(dstr))
4801 Safefree(SvPVX_mutable(dstr));
4805 SvUPGRADE(dstr, SVt_COW);
4807 assert (SvPOK(sstr));
4808 assert (SvPOKp(sstr));
4809 # ifdef PERL_OLD_COPY_ON_WRITE
4810 assert (!SvIOK(sstr));
4811 assert (!SvIOKp(sstr));
4812 assert (!SvNOK(sstr));
4813 assert (!SvNOKp(sstr));
4816 if (SvIsCOW(sstr)) {
4818 if (SvLEN(sstr) == 0) {
4819 /* source is a COW shared hash key. */
4820 DEBUG_C(PerlIO_printf(Perl_debug_log,
4821 "Fast copy on write: Sharing hash\n"));
4822 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4825 # ifdef PERL_OLD_COPY_ON_WRITE
4826 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4828 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4829 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4832 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4833 SvUPGRADE(sstr, SVt_COW);
4835 DEBUG_C(PerlIO_printf(Perl_debug_log,
4836 "Fast copy on write: Converting sstr to COW\n"));
4837 # ifdef PERL_OLD_COPY_ON_WRITE
4838 SV_COW_NEXT_SV_SET(dstr, sstr);
4840 CowREFCNT(sstr) = 0;
4843 # ifdef PERL_OLD_COPY_ON_WRITE
4844 SV_COW_NEXT_SV_SET(sstr, dstr);
4846 # ifdef PERL_DEBUG_READONLY_COW
4847 if (already) sv_buf_to_rw(sstr);
4851 new_pv = SvPVX_mutable(sstr);
4855 SvPV_set(dstr, new_pv);
4856 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4859 SvLEN_set(dstr, len);
4860 SvCUR_set(dstr, cur);
4869 =for apidoc sv_setpvn
4871 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4872 The C<len> parameter indicates the number of
4873 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4874 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4880 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4884 PERL_ARGS_ASSERT_SV_SETPVN;
4886 SV_CHECK_THINKFIRST_COW_DROP(sv);
4892 /* len is STRLEN which is unsigned, need to copy to signed */
4895 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4898 SvUPGRADE(sv, SVt_PV);
4900 dptr = SvGROW(sv, len + 1);
4901 Move(ptr,dptr,len,char);
4904 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4906 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4910 =for apidoc sv_setpvn_mg
4912 Like C<sv_setpvn>, but also handles 'set' magic.
4918 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4920 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4922 sv_setpvn(sv,ptr,len);
4927 =for apidoc sv_setpv
4929 Copies a string into an SV. The string must be terminated with a C<NUL>
4931 Does not handle 'set' magic. See C<sv_setpv_mg>.
4937 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4941 PERL_ARGS_ASSERT_SV_SETPV;
4943 SV_CHECK_THINKFIRST_COW_DROP(sv);
4949 SvUPGRADE(sv, SVt_PV);
4951 SvGROW(sv, len + 1);
4952 Move(ptr,SvPVX(sv),len+1,char);
4954 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4956 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4960 =for apidoc sv_setpv_mg
4962 Like C<sv_setpv>, but also handles 'set' magic.
4968 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4970 PERL_ARGS_ASSERT_SV_SETPV_MG;
4977 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4979 PERL_ARGS_ASSERT_SV_SETHEK;
4985 if (HEK_LEN(hek) == HEf_SVKEY) {
4986 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4989 const int flags = HEK_FLAGS(hek);
4990 if (flags & HVhek_WASUTF8) {
4991 STRLEN utf8_len = HEK_LEN(hek);
4992 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4993 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4996 } else if (flags & HVhek_UNSHARED) {
4997 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5000 else SvUTF8_off(sv);
5004 SV_CHECK_THINKFIRST_COW_DROP(sv);
5005 SvUPGRADE(sv, SVt_PV);
5007 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5008 SvCUR_set(sv, HEK_LEN(hek));
5014 else SvUTF8_off(sv);
5022 =for apidoc sv_usepvn_flags
5024 Tells an SV to use C<ptr> to find its string value. Normally the
5025 string is stored inside the SV, but sv_usepvn allows the SV to use an
5026 outside string. The C<ptr> should point to memory that was allocated
5027 by L<Newx|perlclib/Memory Management and String Handling>. It must be
5028 the start of a Newx-ed block of memory, and not a pointer to the
5029 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
5030 and not be from a non-Newx memory allocator like C<malloc>. The
5031 string length, C<len>, must be supplied. By default this function
5032 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5033 so that pointer should not be freed or used by the programmer after
5034 giving it to sv_usepvn, and neither should any pointers from "behind"
5035 that pointer (e.g. ptr + 1) be used.
5037 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
5038 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
5039 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5040 C<len>, and already meets the requirements for storing in C<SvPVX>).
5046 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5050 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5052 SV_CHECK_THINKFIRST_COW_DROP(sv);
5053 SvUPGRADE(sv, SVt_PV);
5056 if (flags & SV_SMAGIC)
5060 if (SvPVX_const(sv))
5064 if (flags & SV_HAS_TRAILING_NUL)
5065 assert(ptr[len] == '\0');
5068 allocate = (flags & SV_HAS_TRAILING_NUL)
5070 #ifdef Perl_safesysmalloc_size
5073 PERL_STRLEN_ROUNDUP(len + 1);
5075 if (flags & SV_HAS_TRAILING_NUL) {
5076 /* It's long enough - do nothing.
5077 Specifically Perl_newCONSTSUB is relying on this. */
5080 /* Force a move to shake out bugs in callers. */
5081 char *new_ptr = (char*)safemalloc(allocate);
5082 Copy(ptr, new_ptr, len, char);
5083 PoisonFree(ptr,len,char);
5087 ptr = (char*) saferealloc (ptr, allocate);
5090 #ifdef Perl_safesysmalloc_size
5091 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5093 SvLEN_set(sv, allocate);
5097 if (!(flags & SV_HAS_TRAILING_NUL)) {
5100 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5102 if (flags & SV_SMAGIC)
5106 #ifdef PERL_OLD_COPY_ON_WRITE
5107 /* Need to do this *after* making the SV normal, as we need the buffer
5108 pointer to remain valid until after we've copied it. If we let go too early,
5109 another thread could invalidate it by unsharing last of the same hash key
5110 (which it can do by means other than releasing copy-on-write Svs)
5111 or by changing the other copy-on-write SVs in the loop. */
5113 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5115 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5117 { /* this SV was SvIsCOW_normal(sv) */
5118 /* we need to find the SV pointing to us. */
5119 SV *current = SV_COW_NEXT_SV(after);
5121 if (current == sv) {
5122 /* The SV we point to points back to us (there were only two of us
5124 Hence other SV is no longer copy on write either. */
5126 sv_buf_to_rw(after);
5128 /* We need to follow the pointers around the loop. */
5130 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5133 /* don't loop forever if the structure is bust, and we have
5134 a pointer into a closed loop. */
5135 assert (current != after);
5136 assert (SvPVX_const(current) == pvx);
5138 /* Make the SV before us point to the SV after us. */
5139 SV_COW_NEXT_SV_SET(current, after);
5145 =for apidoc sv_force_normal_flags
5147 Undo various types of fakery on an SV, where fakery means
5148 "more than" a string: if the PV is a shared string, make
5149 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5150 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5151 we do the copy, and is also used locally; if this is a
5152 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5153 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5154 SvPOK_off rather than making a copy. (Used where this
5155 scalar is about to be set to some other value.) In addition,
5156 the C<flags> parameter gets passed to C<sv_unref_flags()>
5157 when unreffing. C<sv_force_normal> calls this function
5158 with flags set to 0.
5160 This function is expected to be used to signal to perl that this SV is
5161 about to be written to, and any extra book-keeping needs to be taken care
5162 of. Hence, it croaks on read-only values.
5168 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5170 assert(SvIsCOW(sv));
5173 const char * const pvx = SvPVX_const(sv);
5174 const STRLEN len = SvLEN(sv);
5175 const STRLEN cur = SvCUR(sv);
5176 # ifdef PERL_OLD_COPY_ON_WRITE
5177 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5178 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5179 we'll fail an assertion. */
5180 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5184 PerlIO_printf(Perl_debug_log,
5185 "Copy on write: Force normal %ld\n",
5190 # ifdef PERL_NEW_COPY_ON_WRITE
5192 /* Must do this first, since the CowREFCNT uses SvPVX and
5193 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5194 the only owner left of the buffer. */
5195 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5197 U8 cowrefcnt = CowREFCNT(sv);
5198 if(cowrefcnt != 0) {
5200 CowREFCNT(sv) = cowrefcnt;
5205 /* Else we are the only owner of the buffer. */
5210 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5215 if (flags & SV_COW_DROP_PV) {
5216 /* OK, so we don't need to copy our buffer. */
5219 SvGROW(sv, cur + 1);
5220 Move(pvx,SvPVX(sv),cur,char);
5225 # ifdef PERL_OLD_COPY_ON_WRITE
5226 sv_release_COW(sv, pvx, next);
5229 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5236 const char * const pvx = SvPVX_const(sv);
5237 const STRLEN len = SvCUR(sv);
5241 if (flags & SV_COW_DROP_PV) {
5242 /* OK, so we don't need to copy our buffer. */
5245 SvGROW(sv, len + 1);
5246 Move(pvx,SvPVX(sv),len,char);
5249 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5255 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5257 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5260 Perl_croak_no_modify();
5261 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5262 S_sv_uncow(aTHX_ sv, flags);
5264 sv_unref_flags(sv, flags);
5265 else if (SvFAKE(sv) && isGV_with_GP(sv))
5266 sv_unglob(sv, flags);
5267 else if (SvFAKE(sv) && isREGEXP(sv)) {
5268 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5269 to sv_unglob. We only need it here, so inline it. */
5270 const bool islv = SvTYPE(sv) == SVt_PVLV;
5271 const svtype new_type =
5272 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5273 SV *const temp = newSV_type(new_type);
5274 regexp *const temp_p = ReANY((REGEXP *)sv);
5276 if (new_type == SVt_PVMG) {
5277 SvMAGIC_set(temp, SvMAGIC(sv));
5278 SvMAGIC_set(sv, NULL);
5279 SvSTASH_set(temp, SvSTASH(sv));
5280 SvSTASH_set(sv, NULL);
5282 if (!islv) SvCUR_set(temp, SvCUR(sv));
5283 /* Remember that SvPVX is in the head, not the body. But
5284 RX_WRAPPED is in the body. */
5285 assert(ReANY((REGEXP *)sv)->mother_re);
5286 /* Their buffer is already owned by someone else. */
5287 if (flags & SV_COW_DROP_PV) {
5288 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5289 zeroed body. For SVt_PVLV, it should have been set to 0
5290 before turning into a regexp. */
5291 assert(!SvLEN(islv ? sv : temp));
5292 sv->sv_u.svu_pv = 0;
5295 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5296 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5300 /* Now swap the rest of the bodies. */
5304 SvFLAGS(sv) &= ~SVTYPEMASK;
5305 SvFLAGS(sv) |= new_type;
5306 SvANY(sv) = SvANY(temp);
5309 SvFLAGS(temp) &= ~(SVTYPEMASK);
5310 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5311 SvANY(temp) = temp_p;
5312 temp->sv_u.svu_rx = (regexp *)temp_p;
5314 SvREFCNT_dec_NN(temp);
5316 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5322 Efficient removal of characters from the beginning of the string buffer.
5323 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5324 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5325 character of the adjusted string. Uses the "OOK hack". On return, only
5326 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5328 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5329 refer to the same chunk of data.
5331 The unfortunate similarity of this function's name to that of Perl's C<chop>
5332 operator is strictly coincidental. This function works from the left;
5333 C<chop> works from the right.
5339 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5350 PERL_ARGS_ASSERT_SV_CHOP;
5352 if (!ptr || !SvPOKp(sv))
5354 delta = ptr - SvPVX_const(sv);
5356 /* Nothing to do. */
5359 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5360 if (delta > max_delta)
5361 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5362 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5363 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5364 SV_CHECK_THINKFIRST(sv);
5365 SvPOK_only_UTF8(sv);
5368 if (!SvLEN(sv)) { /* make copy of shared string */
5369 const char *pvx = SvPVX_const(sv);
5370 const STRLEN len = SvCUR(sv);
5371 SvGROW(sv, len + 1);
5372 Move(pvx,SvPVX(sv),len,char);
5378 SvOOK_offset(sv, old_delta);
5380 SvLEN_set(sv, SvLEN(sv) - delta);
5381 SvCUR_set(sv, SvCUR(sv) - delta);
5382 SvPV_set(sv, SvPVX(sv) + delta);
5384 p = (U8 *)SvPVX_const(sv);
5387 /* how many bytes were evacuated? we will fill them with sentinel
5388 bytes, except for the part holding the new offset of course. */
5391 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5393 assert(evacn <= delta + old_delta);
5397 /* This sets 'delta' to the accumulated value of all deltas so far */
5401 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5402 * the string; otherwise store a 0 byte there and store 'delta' just prior
5403 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5404 * portion of the chopped part of the string */
5405 if (delta < 0x100) {
5409 p -= sizeof(STRLEN);
5410 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5414 /* Fill the preceding buffer with sentinals to verify that no-one is
5424 =for apidoc sv_catpvn
5426 Concatenates the string onto the end of the string which is in the SV. The
5427 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5428 status set, then the bytes appended should be valid UTF-8.
5429 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5431 =for apidoc sv_catpvn_flags
5433 Concatenates the string onto the end of the string which is in the SV. The
5434 C<len> indicates number of bytes to copy.
5436 By default, the string appended is assumed to be valid UTF-8 if the SV has
5437 the UTF-8 status set, and a string of bytes otherwise. One can force the
5438 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5439 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5440 string appended will be upgraded to UTF-8 if necessary.
5442 If C<flags> has the C<SV_SMAGIC> bit set, will
5443 C<mg_set> on C<dsv> afterwards if appropriate.
5444 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5445 in terms of this function.
5451 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5454 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5456 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5457 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5459 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5460 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5461 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5464 else SvGROW(dsv, dlen + slen + 1);
5466 sstr = SvPVX_const(dsv);
5467 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5468 SvCUR_set(dsv, SvCUR(dsv) + slen);
5471 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5472 const char * const send = sstr + slen;
5475 /* Something this code does not account for, which I think is
5476 impossible; it would require the same pv to be treated as
5477 bytes *and* utf8, which would indicate a bug elsewhere. */
5478 assert(sstr != dstr);
5480 SvGROW(dsv, dlen + slen * 2 + 1);
5481 d = (U8 *)SvPVX(dsv) + dlen;
5483 while (sstr < send) {
5484 append_utf8_from_native_byte(*sstr, &d);
5487 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5490 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5492 if (flags & SV_SMAGIC)
5497 =for apidoc sv_catsv
5499 Concatenates the string from SV C<ssv> onto the end of the string in SV
5500 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5501 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5504 =for apidoc sv_catsv_flags
5506 Concatenates the string from SV C<ssv> onto the end of the string in SV
5507 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5508 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5509 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5510 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5511 and C<sv_catsv_mg> are implemented in terms of this function.
5516 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5518 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5522 const char *spv = SvPV_flags_const(ssv, slen, flags);
5523 if (flags & SV_GMAGIC)
5525 sv_catpvn_flags(dsv, spv, slen,
5526 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5527 if (flags & SV_SMAGIC)
5533 =for apidoc sv_catpv
5535 Concatenates the C<NUL>-terminated string onto the end of the string which is
5537 If the SV has the UTF-8 status set, then the bytes appended should be
5538 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5543 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5549 PERL_ARGS_ASSERT_SV_CATPV;
5553 junk = SvPV_force(sv, tlen);
5555 SvGROW(sv, tlen + len + 1);
5557 ptr = SvPVX_const(sv);
5558 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5559 SvCUR_set(sv, SvCUR(sv) + len);
5560 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5565 =for apidoc sv_catpv_flags
5567 Concatenates the C<NUL>-terminated string onto the end of the string which is
5569 If the SV has the UTF-8 status set, then the bytes appended should
5570 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5571 on the modified SV if appropriate.
5577 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5579 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5580 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5584 =for apidoc sv_catpv_mg
5586 Like C<sv_catpv>, but also handles 'set' magic.
5592 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5594 PERL_ARGS_ASSERT_SV_CATPV_MG;
5603 Creates a new SV. A non-zero C<len> parameter indicates the number of
5604 bytes of preallocated string space the SV should have. An extra byte for a
5605 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5606 space is allocated.) The reference count for the new SV is set to 1.
5608 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5609 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5610 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5611 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5612 modules supporting older perls.
5618 Perl_newSV(pTHX_ const STRLEN len)
5624 sv_grow(sv, len + 1);
5629 =for apidoc sv_magicext
5631 Adds magic to an SV, upgrading it if necessary. Applies the
5632 supplied vtable and returns a pointer to the magic added.
5634 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5635 In particular, you can add magic to SvREADONLY SVs, and add more than
5636 one instance of the same 'how'.
5638 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5639 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5640 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5641 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5643 (This is now used as a subroutine by C<sv_magic>.)
5648 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5649 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5653 PERL_ARGS_ASSERT_SV_MAGICEXT;
5655 SvUPGRADE(sv, SVt_PVMG);
5656 Newxz(mg, 1, MAGIC);
5657 mg->mg_moremagic = SvMAGIC(sv);
5658 SvMAGIC_set(sv, mg);
5660 /* Sometimes a magic contains a reference loop, where the sv and
5661 object refer to each other. To prevent a reference loop that
5662 would prevent such objects being freed, we look for such loops
5663 and if we find one we avoid incrementing the object refcount.
5665 Note we cannot do this to avoid self-tie loops as intervening RV must
5666 have its REFCNT incremented to keep it in existence.
5669 if (!obj || obj == sv ||
5670 how == PERL_MAGIC_arylen ||
5671 how == PERL_MAGIC_symtab ||
5672 (SvTYPE(obj) == SVt_PVGV &&
5673 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5674 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5675 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5680 mg->mg_obj = SvREFCNT_inc_simple(obj);
5681 mg->mg_flags |= MGf_REFCOUNTED;
5684 /* Normal self-ties simply pass a null object, and instead of
5685 using mg_obj directly, use the SvTIED_obj macro to produce a
5686 new RV as needed. For glob "self-ties", we are tieing the PVIO
5687 with an RV obj pointing to the glob containing the PVIO. In
5688 this case, to avoid a reference loop, we need to weaken the
5692 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5693 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5699 mg->mg_len = namlen;
5702 mg->mg_ptr = savepvn(name, namlen);
5703 else if (namlen == HEf_SVKEY) {
5704 /* Yes, this is casting away const. This is only for the case of
5705 HEf_SVKEY. I think we need to document this aberation of the
5706 constness of the API, rather than making name non-const, as
5707 that change propagating outwards a long way. */
5708 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5710 mg->mg_ptr = (char *) name;
5712 mg->mg_virtual = (MGVTBL *) vtable;
5719 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5721 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5722 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5723 /* This sv is only a delegate. //g magic must be attached to
5728 #ifdef PERL_OLD_COPY_ON_WRITE
5730 sv_force_normal_flags(sv, 0);
5732 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5733 &PL_vtbl_mglob, 0, 0);
5737 =for apidoc sv_magic
5739 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5740 necessary, then adds a new magic item of type C<how> to the head of the
5743 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5744 handling of the C<name> and C<namlen> arguments.
5746 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5747 to add more than one instance of the same 'how'.
5753 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5754 const char *const name, const I32 namlen)
5756 const MGVTBL *vtable;
5759 unsigned int vtable_index;
5761 PERL_ARGS_ASSERT_SV_MAGIC;
5763 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5764 || ((flags = PL_magic_data[how]),
5765 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5766 > magic_vtable_max))
5767 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5769 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5770 Useful for attaching extension internal data to perl vars.
5771 Note that multiple extensions may clash if magical scalars
5772 etc holding private data from one are passed to another. */
5774 vtable = (vtable_index == magic_vtable_max)
5775 ? NULL : PL_magic_vtables + vtable_index;
5777 #ifdef PERL_OLD_COPY_ON_WRITE
5779 sv_force_normal_flags(sv, 0);
5781 if (SvREADONLY(sv)) {
5783 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5786 Perl_croak_no_modify();
5789 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5790 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5791 /* sv_magic() refuses to add a magic of the same 'how' as an
5794 if (how == PERL_MAGIC_taint)
5800 /* Force pos to be stored as characters, not bytes. */
5801 if (SvMAGICAL(sv) && DO_UTF8(sv)
5802 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5804 && mg->mg_flags & MGf_BYTES) {
5805 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5807 mg->mg_flags &= ~MGf_BYTES;
5810 /* Rest of work is done else where */
5811 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5814 case PERL_MAGIC_taint:
5817 case PERL_MAGIC_ext:
5818 case PERL_MAGIC_dbfile:
5825 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5832 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5834 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5835 for (mg = *mgp; mg; mg = *mgp) {
5836 const MGVTBL* const virt = mg->mg_virtual;
5837 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5838 *mgp = mg->mg_moremagic;
5839 if (virt && virt->svt_free)
5840 virt->svt_free(aTHX_ sv, mg);
5841 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5843 Safefree(mg->mg_ptr);
5844 else if (mg->mg_len == HEf_SVKEY)
5845 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5846 else if (mg->mg_type == PERL_MAGIC_utf8)
5847 Safefree(mg->mg_ptr);
5849 if (mg->mg_flags & MGf_REFCOUNTED)
5850 SvREFCNT_dec(mg->mg_obj);
5854 mgp = &mg->mg_moremagic;
5857 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5858 mg_magical(sv); /* else fix the flags now */
5862 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5868 =for apidoc sv_unmagic
5870 Removes all magic of type C<type> from an SV.
5876 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5878 PERL_ARGS_ASSERT_SV_UNMAGIC;
5879 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5883 =for apidoc sv_unmagicext
5885 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5891 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5893 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5894 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5898 =for apidoc sv_rvweaken
5900 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5901 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5902 push a back-reference to this RV onto the array of backreferences
5903 associated with that magic. If the RV is magical, set magic will be
5904 called after the RV is cleared.
5910 Perl_sv_rvweaken(pTHX_ SV *const sv)
5914 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5916 if (!SvOK(sv)) /* let undefs pass */
5919 Perl_croak(aTHX_ "Can't weaken a nonreference");
5920 else if (SvWEAKREF(sv)) {
5921 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5924 else if (SvREADONLY(sv)) croak_no_modify();
5926 Perl_sv_add_backref(aTHX_ tsv, sv);
5928 SvREFCNT_dec_NN(tsv);
5932 /* Give tsv backref magic if it hasn't already got it, then push a
5933 * back-reference to sv onto the array associated with the backref magic.
5935 * As an optimisation, if there's only one backref and it's not an AV,
5936 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5937 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5941 /* A discussion about the backreferences array and its refcount:
5943 * The AV holding the backreferences is pointed to either as the mg_obj of
5944 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5945 * xhv_backreferences field. The array is created with a refcount
5946 * of 2. This means that if during global destruction the array gets
5947 * picked on before its parent to have its refcount decremented by the
5948 * random zapper, it won't actually be freed, meaning it's still there for
5949 * when its parent gets freed.
5951 * When the parent SV is freed, the extra ref is killed by
5952 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5953 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5955 * When a single backref SV is stored directly, it is not reference
5960 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5966 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5968 /* find slot to store array or singleton backref */
5970 if (SvTYPE(tsv) == SVt_PVHV) {
5971 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5974 mg = mg_find(tsv, PERL_MAGIC_backref);
5976 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5977 svp = &(mg->mg_obj);
5980 /* create or retrieve the array */
5982 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5983 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5987 mg->mg_flags |= MGf_REFCOUNTED;
5990 SvREFCNT_inc_simple_void_NN(av);
5991 /* av now has a refcnt of 2; see discussion above */
5992 av_extend(av, *svp ? 2 : 1);
5994 /* move single existing backref to the array */
5995 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6000 av = MUTABLE_AV(*svp);
6002 /* optimisation: store single backref directly in HvAUX or mg_obj */
6006 assert(SvTYPE(av) == SVt_PVAV);
6007 if (AvFILLp(av) >= AvMAX(av)) {
6008 av_extend(av, AvFILLp(av)+1);
6011 /* push new backref */
6012 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6015 /* delete a back-reference to ourselves from the backref magic associated
6016 * with the SV we point to.
6020 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6024 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6026 if (SvTYPE(tsv) == SVt_PVHV) {
6028 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6030 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6031 /* It's possible for the the last (strong) reference to tsv to have
6032 become freed *before* the last thing holding a weak reference.
6033 If both survive longer than the backreferences array, then when
6034 the referent's reference count drops to 0 and it is freed, it's
6035 not able to chase the backreferences, so they aren't NULLed.
6037 For example, a CV holds a weak reference to its stash. If both the
6038 CV and the stash survive longer than the backreferences array,
6039 and the CV gets picked for the SvBREAK() treatment first,
6040 *and* it turns out that the stash is only being kept alive because
6041 of an our variable in the pad of the CV, then midway during CV
6042 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6043 It ends up pointing to the freed HV. Hence it's chased in here, and
6044 if this block wasn't here, it would hit the !svp panic just below.
6046 I don't believe that "better" destruction ordering is going to help
6047 here - during global destruction there's always going to be the
6048 chance that something goes out of order. We've tried to make it
6049 foolproof before, and it only resulted in evolutionary pressure on
6050 fools. Which made us look foolish for our hubris. :-(
6056 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6057 svp = mg ? &(mg->mg_obj) : NULL;
6061 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6063 /* It's possible that sv is being freed recursively part way through the
6064 freeing of tsv. If this happens, the backreferences array of tsv has
6065 already been freed, and so svp will be NULL. If this is the case,
6066 we should not panic. Instead, nothing needs doing, so return. */
6067 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6069 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6070 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6073 if (SvTYPE(*svp) == SVt_PVAV) {
6077 AV * const av = (AV*)*svp;
6079 assert(!SvIS_FREED(av));
6083 /* for an SV with N weak references to it, if all those
6084 * weak refs are deleted, then sv_del_backref will be called
6085 * N times and O(N^2) compares will be done within the backref
6086 * array. To ameliorate this potential slowness, we:
6087 * 1) make sure this code is as tight as possible;
6088 * 2) when looking for SV, look for it at both the head and tail of the
6089 * array first before searching the rest, since some create/destroy
6090 * patterns will cause the backrefs to be freed in order.
6097 SV **p = &svp[fill];
6098 SV *const topsv = *p;
6105 /* We weren't the last entry.
6106 An unordered list has this property that you
6107 can take the last element off the end to fill
6108 the hole, and it's still an unordered list :-)
6114 break; /* should only be one */
6121 AvFILLp(av) = fill-1;
6123 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6124 /* freed AV; skip */
6127 /* optimisation: only a single backref, stored directly */
6129 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6130 (void*)*svp, (void*)sv);
6137 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6143 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6148 /* after multiple passes through Perl_sv_clean_all() for a thingy
6149 * that has badly leaked, the backref array may have gotten freed,
6150 * since we only protect it against 1 round of cleanup */
6151 if (SvIS_FREED(av)) {
6152 if (PL_in_clean_all) /* All is fair */
6155 "panic: magic_killbackrefs (freed backref AV/SV)");
6159 is_array = (SvTYPE(av) == SVt_PVAV);
6161 assert(!SvIS_FREED(av));
6164 last = svp + AvFILLp(av);
6167 /* optimisation: only a single backref, stored directly */
6173 while (svp <= last) {
6175 SV *const referrer = *svp;
6176 if (SvWEAKREF(referrer)) {
6177 /* XXX Should we check that it hasn't changed? */
6178 assert(SvROK(referrer));
6179 SvRV_set(referrer, 0);
6181 SvWEAKREF_off(referrer);
6182 SvSETMAGIC(referrer);
6183 } else if (SvTYPE(referrer) == SVt_PVGV ||
6184 SvTYPE(referrer) == SVt_PVLV) {
6185 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6186 /* You lookin' at me? */
6187 assert(GvSTASH(referrer));
6188 assert(GvSTASH(referrer) == (const HV *)sv);
6189 GvSTASH(referrer) = 0;
6190 } else if (SvTYPE(referrer) == SVt_PVCV ||
6191 SvTYPE(referrer) == SVt_PVFM) {
6192 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6193 /* You lookin' at me? */
6194 assert(CvSTASH(referrer));
6195 assert(CvSTASH(referrer) == (const HV *)sv);
6196 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6199 assert(SvTYPE(sv) == SVt_PVGV);
6200 /* You lookin' at me? */
6201 assert(CvGV(referrer));
6202 assert(CvGV(referrer) == (const GV *)sv);
6203 anonymise_cv_maybe(MUTABLE_GV(sv),
6204 MUTABLE_CV(referrer));
6209 "panic: magic_killbackrefs (flags=%"UVxf")",
6210 (UV)SvFLAGS(referrer));
6221 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6227 =for apidoc sv_insert
6229 Inserts a string at the specified offset/length within the SV. Similar to
6230 the Perl substr() function. Handles get magic.
6232 =for apidoc sv_insert_flags
6234 Same as C<sv_insert>, but the extra C<flags> are passed to the
6235 C<SvPV_force_flags> that applies to C<bigstr>.
6241 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6247 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6250 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6253 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6254 SvPV_force_flags(bigstr, curlen, flags);
6255 (void)SvPOK_only_UTF8(bigstr);
6256 if (offset + len > curlen) {
6257 SvGROW(bigstr, offset+len+1);
6258 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6259 SvCUR_set(bigstr, offset+len);
6263 i = littlelen - len;
6264 if (i > 0) { /* string might grow */
6265 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6266 mid = big + offset + len;
6267 midend = bigend = big + SvCUR(bigstr);
6270 while (midend > mid) /* shove everything down */
6271 *--bigend = *--midend;
6272 Move(little,big+offset,littlelen,char);
6273 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6278 Move(little,SvPVX(bigstr)+offset,len,char);
6283 big = SvPVX(bigstr);
6286 bigend = big + SvCUR(bigstr);
6288 if (midend > bigend)
6289 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6292 if (mid - big > bigend - midend) { /* faster to shorten from end */
6294 Move(little, mid, littlelen,char);
6297 i = bigend - midend;
6299 Move(midend, mid, i,char);
6303 SvCUR_set(bigstr, mid - big);
6305 else if ((i = mid - big)) { /* faster from front */
6306 midend -= littlelen;
6308 Move(big, midend - i, i, char);
6309 sv_chop(bigstr,midend-i);
6311 Move(little, mid, littlelen,char);
6313 else if (littlelen) {
6314 midend -= littlelen;
6315 sv_chop(bigstr,midend);
6316 Move(little,midend,littlelen,char);
6319 sv_chop(bigstr,midend);
6325 =for apidoc sv_replace
6327 Make the first argument a copy of the second, then delete the original.
6328 The target SV physically takes over ownership of the body of the source SV
6329 and inherits its flags; however, the target keeps any magic it owns,
6330 and any magic in the source is discarded.
6331 Note that this is a rather specialist SV copying operation; most of the
6332 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6338 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6340 const U32 refcnt = SvREFCNT(sv);
6342 PERL_ARGS_ASSERT_SV_REPLACE;
6344 SV_CHECK_THINKFIRST_COW_DROP(sv);
6345 if (SvREFCNT(nsv) != 1) {
6346 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6347 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6349 if (SvMAGICAL(sv)) {
6353 sv_upgrade(nsv, SVt_PVMG);
6354 SvMAGIC_set(nsv, SvMAGIC(sv));
6355 SvFLAGS(nsv) |= SvMAGICAL(sv);
6357 SvMAGIC_set(sv, NULL);
6361 assert(!SvREFCNT(sv));
6362 #ifdef DEBUG_LEAKING_SCALARS
6363 sv->sv_flags = nsv->sv_flags;
6364 sv->sv_any = nsv->sv_any;
6365 sv->sv_refcnt = nsv->sv_refcnt;
6366 sv->sv_u = nsv->sv_u;
6368 StructCopy(nsv,sv,SV);
6370 if(SvTYPE(sv) == SVt_IV) {
6371 SET_SVANY_FOR_BODYLESS_IV(sv);
6375 #ifdef PERL_OLD_COPY_ON_WRITE
6376 if (SvIsCOW_normal(nsv)) {
6377 /* We need to follow the pointers around the loop to make the
6378 previous SV point to sv, rather than nsv. */
6381 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6384 assert(SvPVX_const(current) == SvPVX_const(nsv));
6386 /* Make the SV before us point to the SV after us. */
6388 PerlIO_printf(Perl_debug_log, "previous is\n");
6390 PerlIO_printf(Perl_debug_log,
6391 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6392 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6394 SV_COW_NEXT_SV_SET(current, sv);
6397 SvREFCNT(sv) = refcnt;
6398 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6403 /* We're about to free a GV which has a CV that refers back to us.
6404 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6408 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6413 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6416 assert(SvREFCNT(gv) == 0);
6417 assert(isGV(gv) && isGV_with_GP(gv));
6419 assert(!CvANON(cv));
6420 assert(CvGV(cv) == gv);
6421 assert(!CvNAMED(cv));
6423 /* will the CV shortly be freed by gp_free() ? */
6424 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6425 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6429 /* if not, anonymise: */
6430 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6431 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6432 : newSVpvn_flags( "__ANON__", 8, 0 );
6433 sv_catpvs(gvname, "::__ANON__");
6434 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6435 SvREFCNT_dec_NN(gvname);
6439 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6444 =for apidoc sv_clear
6446 Clear an SV: call any destructors, free up any memory used by the body,
6447 and free the body itself. The SV's head is I<not> freed, although
6448 its type is set to all 1's so that it won't inadvertently be assumed
6449 to be live during global destruction etc.
6450 This function should only be called when REFCNT is zero. Most of the time
6451 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6458 Perl_sv_clear(pTHX_ SV *const orig_sv)
6463 const struct body_details *sv_type_details;
6469 PERL_ARGS_ASSERT_SV_CLEAR;
6471 /* within this loop, sv is the SV currently being freed, and
6472 * iter_sv is the most recent AV or whatever that's being iterated
6473 * over to provide more SVs */
6479 assert(SvREFCNT(sv) == 0);
6480 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6482 if (type <= SVt_IV) {
6483 /* See the comment in sv.h about the collusion between this
6484 * early return and the overloading of the NULL slots in the
6488 SvFLAGS(sv) &= SVf_BREAK;
6489 SvFLAGS(sv) |= SVTYPEMASK;
6493 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6494 for another purpose */
6495 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6497 if (type >= SVt_PVMG) {
6499 if (!curse(sv, 1)) goto get_next_sv;
6500 type = SvTYPE(sv); /* destructor may have changed it */
6502 /* Free back-references before magic, in case the magic calls
6503 * Perl code that has weak references to sv. */
6504 if (type == SVt_PVHV) {
6505 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6509 else if (SvMAGIC(sv)) {
6510 /* Free back-references before other types of magic. */
6511 sv_unmagic(sv, PERL_MAGIC_backref);
6517 /* case SVt_INVLIST: */
6520 IoIFP(sv) != PerlIO_stdin() &&
6521 IoIFP(sv) != PerlIO_stdout() &&
6522 IoIFP(sv) != PerlIO_stderr() &&
6523 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6525 io_close(MUTABLE_IO(sv), NULL, FALSE,
6526 (IoTYPE(sv) == IoTYPE_WRONLY ||
6527 IoTYPE(sv) == IoTYPE_RDWR ||
6528 IoTYPE(sv) == IoTYPE_APPEND));
6530 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6531 PerlDir_close(IoDIRP(sv));
6532 IoDIRP(sv) = (DIR*)NULL;
6533 Safefree(IoTOP_NAME(sv));
6534 Safefree(IoFMT_NAME(sv));
6535 Safefree(IoBOTTOM_NAME(sv));
6536 if ((const GV *)sv == PL_statgv)
6540 /* FIXME for plugins */
6542 pregfree2((REGEXP*) sv);
6546 cv_undef(MUTABLE_CV(sv));
6547 /* If we're in a stash, we don't own a reference to it.
6548 * However it does have a back reference to us, which needs to
6550 if ((stash = CvSTASH(sv)))
6551 sv_del_backref(MUTABLE_SV(stash), sv);
6554 if (PL_last_swash_hv == (const HV *)sv) {
6555 PL_last_swash_hv = NULL;
6557 if (HvTOTALKEYS((HV*)sv) > 0) {
6559 /* this statement should match the one at the beginning of
6560 * hv_undef_flags() */
6561 if ( PL_phase != PERL_PHASE_DESTRUCT
6562 && (name = HvNAME((HV*)sv)))
6564 if (PL_stashcache) {
6565 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6567 (void)hv_deletehek(PL_stashcache,
6568 HvNAME_HEK((HV*)sv), G_DISCARD);
6570 hv_name_set((HV*)sv, NULL, 0, 0);
6573 /* save old iter_sv in unused SvSTASH field */
6574 assert(!SvOBJECT(sv));
6575 SvSTASH(sv) = (HV*)iter_sv;
6578 /* save old hash_index in unused SvMAGIC field */
6579 assert(!SvMAGICAL(sv));
6580 assert(!SvMAGIC(sv));
6581 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6584 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6585 goto get_next_sv; /* process this new sv */
6587 /* free empty hash */
6588 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6589 assert(!HvARRAY((HV*)sv));
6593 AV* av = MUTABLE_AV(sv);
6594 if (PL_comppad == av) {
6598 if (AvREAL(av) && AvFILLp(av) > -1) {
6599 next_sv = AvARRAY(av)[AvFILLp(av)--];
6600 /* save old iter_sv in top-most slot of AV,
6601 * and pray that it doesn't get wiped in the meantime */
6602 AvARRAY(av)[AvMAX(av)] = iter_sv;
6604 goto get_next_sv; /* process this new sv */
6606 Safefree(AvALLOC(av));
6611 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6612 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6613 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6614 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6616 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6617 SvREFCNT_dec(LvTARG(sv));
6618 if (isREGEXP(sv)) goto freeregexp;
6620 if (isGV_with_GP(sv)) {
6621 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6622 && HvENAME_get(stash))
6623 mro_method_changed_in(stash);
6624 gp_free(MUTABLE_GV(sv));
6626 unshare_hek(GvNAME_HEK(sv));
6627 /* If we're in a stash, we don't own a reference to it.
6628 * However it does have a back reference to us, which
6629 * needs to be cleared. */
6630 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6631 sv_del_backref(MUTABLE_SV(stash), sv);
6633 /* FIXME. There are probably more unreferenced pointers to SVs
6634 * in the interpreter struct that we should check and tidy in
6635 * a similar fashion to this: */
6636 /* See also S_sv_unglob, which does the same thing. */
6637 if ((const GV *)sv == PL_last_in_gv)
6638 PL_last_in_gv = NULL;
6639 else if ((const GV *)sv == PL_statgv)
6641 else if ((const GV *)sv == PL_stderrgv)
6649 /* Don't bother with SvOOK_off(sv); as we're only going to
6653 SvOOK_offset(sv, offset);
6654 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6655 /* Don't even bother with turning off the OOK flag. */
6660 SV * const target = SvRV(sv);
6662 sv_del_backref(target, sv);
6668 else if (SvPVX_const(sv)
6669 && !(SvTYPE(sv) == SVt_PVIO
6670 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6674 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6678 # ifdef PERL_OLD_COPY_ON_WRITE
6679 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6681 if (CowREFCNT(sv)) {
6689 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6693 # ifdef PERL_OLD_COPY_ON_WRITE
6697 Safefree(SvPVX_mutable(sv));
6701 else if (SvPVX_const(sv) && SvLEN(sv)
6702 && !(SvTYPE(sv) == SVt_PVIO
6703 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6704 Safefree(SvPVX_mutable(sv));
6705 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6706 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6716 SvFLAGS(sv) &= SVf_BREAK;
6717 SvFLAGS(sv) |= SVTYPEMASK;
6719 sv_type_details = bodies_by_type + type;
6720 if (sv_type_details->arena) {
6721 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6722 &PL_body_roots[type]);
6724 else if (sv_type_details->body_size) {
6725 safefree(SvANY(sv));
6729 /* caller is responsible for freeing the head of the original sv */
6730 if (sv != orig_sv && !SvREFCNT(sv))
6733 /* grab and free next sv, if any */
6741 else if (!iter_sv) {
6743 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6744 AV *const av = (AV*)iter_sv;
6745 if (AvFILLp(av) > -1) {
6746 sv = AvARRAY(av)[AvFILLp(av)--];
6748 else { /* no more elements of current AV to free */
6751 /* restore previous value, squirrelled away */
6752 iter_sv = AvARRAY(av)[AvMAX(av)];
6753 Safefree(AvALLOC(av));
6756 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6757 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6758 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6759 /* no more elements of current HV to free */
6762 /* Restore previous values of iter_sv and hash_index,
6763 * squirrelled away */
6764 assert(!SvOBJECT(sv));
6765 iter_sv = (SV*)SvSTASH(sv);
6766 assert(!SvMAGICAL(sv));
6767 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6769 /* perl -DA does not like rubbish in SvMAGIC. */
6773 /* free any remaining detritus from the hash struct */
6774 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6775 assert(!HvARRAY((HV*)sv));
6780 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6784 if (!SvREFCNT(sv)) {
6788 if (--(SvREFCNT(sv)))
6792 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6793 "Attempt to free temp prematurely: SV 0x%"UVxf
6794 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6798 if (SvIMMORTAL(sv)) {
6799 /* make sure SvREFCNT(sv)==0 happens very seldom */
6800 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6809 /* This routine curses the sv itself, not the object referenced by sv. So
6810 sv does not have to be ROK. */
6813 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6814 PERL_ARGS_ASSERT_CURSE;
6815 assert(SvOBJECT(sv));
6817 if (PL_defstash && /* Still have a symbol table? */
6823 stash = SvSTASH(sv);
6824 assert(SvTYPE(stash) == SVt_PVHV);
6825 if (HvNAME(stash)) {
6826 CV* destructor = NULL;
6827 assert (SvOOK(stash));
6828 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6829 if (!destructor || HvMROMETA(stash)->destroy_gen
6830 != PL_sub_generation)
6833 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6834 if (gv) destructor = GvCV(gv);
6835 if (!SvOBJECT(stash))
6838 destructor ? (HV *)destructor : ((HV *)0)+1;
6839 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6843 assert(!destructor || destructor == ((CV *)0)+1
6844 || SvTYPE(destructor) == SVt_PVCV);
6845 if (destructor && destructor != ((CV *)0)+1
6846 /* A constant subroutine can have no side effects, so
6847 don't bother calling it. */
6848 && !CvCONST(destructor)
6849 /* Don't bother calling an empty destructor or one that
6850 returns immediately. */
6851 && (CvISXSUB(destructor)
6852 || (CvSTART(destructor)
6853 && (CvSTART(destructor)->op_next->op_type
6855 && (CvSTART(destructor)->op_next->op_type
6857 || CvSTART(destructor)->op_next->op_next->op_type
6863 SV* const tmpref = newRV(sv);
6864 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6866 PUSHSTACKi(PERLSI_DESTROY);
6871 call_sv(MUTABLE_SV(destructor),
6872 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6876 if(SvREFCNT(tmpref) < 2) {
6877 /* tmpref is not kept alive! */
6879 SvRV_set(tmpref, NULL);
6882 SvREFCNT_dec_NN(tmpref);
6885 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6888 if (check_refcnt && SvREFCNT(sv)) {
6889 if (PL_in_clean_objs)
6891 "DESTROY created new reference to dead object '%"HEKf"'",
6892 HEKfARG(HvNAME_HEK(stash)));
6893 /* DESTROY gave object new lease on life */
6899 HV * const stash = SvSTASH(sv);
6900 /* Curse before freeing the stash, as freeing the stash could cause
6901 a recursive call into S_curse. */
6902 SvOBJECT_off(sv); /* Curse the object. */
6903 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6904 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6910 =for apidoc sv_newref
6912 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6919 Perl_sv_newref(pTHX_ SV *const sv)
6921 PERL_UNUSED_CONTEXT;
6930 Decrement an SV's reference count, and if it drops to zero, call
6931 C<sv_clear> to invoke destructors and free up any memory used by
6932 the body; finally, deallocate the SV's head itself.
6933 Normally called via a wrapper macro C<SvREFCNT_dec>.
6939 Perl_sv_free(pTHX_ SV *const sv)
6945 /* Private helper function for SvREFCNT_dec().
6946 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6949 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6953 PERL_ARGS_ASSERT_SV_FREE2;
6955 if (LIKELY( rc == 1 )) {
6961 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6962 "Attempt to free temp prematurely: SV 0x%"UVxf
6963 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6967 if (SvIMMORTAL(sv)) {
6968 /* make sure SvREFCNT(sv)==0 happens very seldom */
6969 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6973 if (! SvREFCNT(sv)) /* may have have been resurrected */
6978 /* handle exceptional cases */
6982 if (SvFLAGS(sv) & SVf_BREAK)
6983 /* this SV's refcnt has been artificially decremented to
6984 * trigger cleanup */
6986 if (PL_in_clean_all) /* All is fair */
6988 if (SvIMMORTAL(sv)) {
6989 /* make sure SvREFCNT(sv)==0 happens very seldom */
6990 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6993 if (ckWARN_d(WARN_INTERNAL)) {
6994 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6995 Perl_dump_sv_child(aTHX_ sv);
6997 #ifdef DEBUG_LEAKING_SCALARS
7000 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7001 if (PL_warnhook == PERL_WARNHOOK_FATAL
7002 || ckDEAD(packWARN(WARN_INTERNAL))) {
7003 /* Don't let Perl_warner cause us to escape our fate: */
7007 /* This may not return: */
7008 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7009 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7010 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7013 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7023 Returns the length of the string in the SV. Handles magic and type
7024 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
7025 gives raw access to the xpv_cur slot.
7031 Perl_sv_len(pTHX_ SV *const sv)
7038 (void)SvPV_const(sv, len);
7043 =for apidoc sv_len_utf8
7045 Returns the number of characters in the string in an SV, counting wide
7046 UTF-8 bytes as a single character. Handles magic and type coercion.
7052 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7053 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7054 * (Note that the mg_len is not the length of the mg_ptr field.
7055 * This allows the cache to store the character length of the string without
7056 * needing to malloc() extra storage to attach to the mg_ptr.)
7061 Perl_sv_len_utf8(pTHX_ SV *const sv)
7067 return sv_len_utf8_nomg(sv);
7071 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7074 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7076 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7078 if (PL_utf8cache && SvUTF8(sv)) {
7080 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7082 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7083 if (mg->mg_len != -1)
7086 /* We can use the offset cache for a headstart.
7087 The longer value is stored in the first pair. */
7088 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7090 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7094 if (PL_utf8cache < 0) {
7095 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7096 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7100 ulen = Perl_utf8_length(aTHX_ s, s + len);
7101 utf8_mg_len_cache_update(sv, &mg, ulen);
7105 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7108 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7111 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7112 STRLEN *const uoffset_p, bool *const at_end)
7114 const U8 *s = start;
7115 STRLEN uoffset = *uoffset_p;
7117 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7119 while (s < send && uoffset) {
7126 else if (s > send) {
7128 /* This is the existing behaviour. Possibly it should be a croak, as
7129 it's actually a bounds error */
7132 *uoffset_p -= uoffset;
7136 /* Given the length of the string in both bytes and UTF-8 characters, decide
7137 whether to walk forwards or backwards to find the byte corresponding to
7138 the passed in UTF-8 offset. */
7140 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7141 STRLEN uoffset, const STRLEN uend)
7143 STRLEN backw = uend - uoffset;
7145 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7147 if (uoffset < 2 * backw) {
7148 /* The assumption is that going forwards is twice the speed of going
7149 forward (that's where the 2 * backw comes from).
7150 (The real figure of course depends on the UTF-8 data.) */
7151 const U8 *s = start;
7153 while (s < send && uoffset--)
7163 while (UTF8_IS_CONTINUATION(*send))
7166 return send - start;
7169 /* For the string representation of the given scalar, find the byte
7170 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7171 give another position in the string, *before* the sought offset, which
7172 (which is always true, as 0, 0 is a valid pair of positions), which should
7173 help reduce the amount of linear searching.
7174 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7175 will be used to reduce the amount of linear searching. The cache will be
7176 created if necessary, and the found value offered to it for update. */
7178 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7179 const U8 *const send, STRLEN uoffset,
7180 STRLEN uoffset0, STRLEN boffset0)
7182 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7184 bool at_end = FALSE;
7186 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7188 assert (uoffset >= uoffset0);
7193 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7195 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7196 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7197 if ((*mgp)->mg_ptr) {
7198 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7199 if (cache[0] == uoffset) {
7200 /* An exact match. */
7203 if (cache[2] == uoffset) {
7204 /* An exact match. */
7208 if (cache[0] < uoffset) {
7209 /* The cache already knows part of the way. */
7210 if (cache[0] > uoffset0) {
7211 /* The cache knows more than the passed in pair */
7212 uoffset0 = cache[0];
7213 boffset0 = cache[1];
7215 if ((*mgp)->mg_len != -1) {
7216 /* And we know the end too. */
7218 + sv_pos_u2b_midway(start + boffset0, send,
7220 (*mgp)->mg_len - uoffset0);
7222 uoffset -= uoffset0;
7224 + sv_pos_u2b_forwards(start + boffset0,
7225 send, &uoffset, &at_end);
7226 uoffset += uoffset0;
7229 else if (cache[2] < uoffset) {
7230 /* We're between the two cache entries. */
7231 if (cache[2] > uoffset0) {
7232 /* and the cache knows more than the passed in pair */
7233 uoffset0 = cache[2];
7234 boffset0 = cache[3];
7238 + sv_pos_u2b_midway(start + boffset0,
7241 cache[0] - uoffset0);
7244 + sv_pos_u2b_midway(start + boffset0,
7247 cache[2] - uoffset0);
7251 else if ((*mgp)->mg_len != -1) {
7252 /* If we can take advantage of a passed in offset, do so. */
7253 /* In fact, offset0 is either 0, or less than offset, so don't
7254 need to worry about the other possibility. */
7256 + sv_pos_u2b_midway(start + boffset0, send,
7258 (*mgp)->mg_len - uoffset0);
7263 if (!found || PL_utf8cache < 0) {
7264 STRLEN real_boffset;
7265 uoffset -= uoffset0;
7266 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7267 send, &uoffset, &at_end);
7268 uoffset += uoffset0;
7270 if (found && PL_utf8cache < 0)
7271 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7273 boffset = real_boffset;
7276 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7278 utf8_mg_len_cache_update(sv, mgp, uoffset);
7280 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7287 =for apidoc sv_pos_u2b_flags
7289 Converts the offset from a count of UTF-8 chars from
7290 the start of the string, to a count of the equivalent number of bytes; if
7291 lenp is non-zero, it does the same to lenp, but this time starting from
7292 the offset, rather than from the start
7293 of the string. Handles type coercion.
7294 I<flags> is passed to C<SvPV_flags>, and usually should be
7295 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7301 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7302 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7303 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7308 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7315 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7317 start = (U8*)SvPV_flags(sv, len, flags);
7319 const U8 * const send = start + len;
7321 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7324 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7325 is 0, and *lenp is already set to that. */) {
7326 /* Convert the relative offset to absolute. */
7327 const STRLEN uoffset2 = uoffset + *lenp;
7328 const STRLEN boffset2
7329 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7330 uoffset, boffset) - boffset;
7344 =for apidoc sv_pos_u2b
7346 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7347 the start of the string, to a count of the equivalent number of bytes; if
7348 lenp is non-zero, it does the same to lenp, but this time starting from
7349 the offset, rather than from the start of the string. Handles magic and
7352 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7359 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7360 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7361 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7365 /* This function is subject to size and sign problems */
7368 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7370 PERL_ARGS_ASSERT_SV_POS_U2B;
7373 STRLEN ulen = (STRLEN)*lenp;
7374 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7375 SV_GMAGIC|SV_CONST_RETURN);
7378 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7379 SV_GMAGIC|SV_CONST_RETURN);
7384 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7387 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7388 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7391 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7392 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7393 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7397 (*mgp)->mg_len = ulen;
7400 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7401 byte length pairing. The (byte) length of the total SV is passed in too,
7402 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7403 may not have updated SvCUR, so we can't rely on reading it directly.
7405 The proffered utf8/byte length pairing isn't used if the cache already has
7406 two pairs, and swapping either for the proffered pair would increase the
7407 RMS of the intervals between known byte offsets.
7409 The cache itself consists of 4 STRLEN values
7410 0: larger UTF-8 offset
7411 1: corresponding byte offset
7412 2: smaller UTF-8 offset
7413 3: corresponding byte offset
7415 Unused cache pairs have the value 0, 0.
7416 Keeping the cache "backwards" means that the invariant of
7417 cache[0] >= cache[2] is maintained even with empty slots, which means that
7418 the code that uses it doesn't need to worry if only 1 entry has actually
7419 been set to non-zero. It also makes the "position beyond the end of the
7420 cache" logic much simpler, as the first slot is always the one to start
7424 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7425 const STRLEN utf8, const STRLEN blen)
7429 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7434 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7435 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7436 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7438 (*mgp)->mg_len = -1;
7442 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7443 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7444 (*mgp)->mg_ptr = (char *) cache;
7448 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7449 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7450 a pointer. Note that we no longer cache utf8 offsets on refer-
7451 ences, but this check is still a good idea, for robustness. */
7452 const U8 *start = (const U8 *) SvPVX_const(sv);
7453 const STRLEN realutf8 = utf8_length(start, start + byte);
7455 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7459 /* Cache is held with the later position first, to simplify the code
7460 that deals with unbounded ends. */
7462 ASSERT_UTF8_CACHE(cache);
7463 if (cache[1] == 0) {
7464 /* Cache is totally empty */
7467 } else if (cache[3] == 0) {
7468 if (byte > cache[1]) {
7469 /* New one is larger, so goes first. */
7470 cache[2] = cache[0];
7471 cache[3] = cache[1];
7479 /* float casts necessary? XXX */
7480 #define THREEWAY_SQUARE(a,b,c,d) \
7481 ((float)((d) - (c))) * ((float)((d) - (c))) \
7482 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7483 + ((float)((b) - (a))) * ((float)((b) - (a)))
7485 /* Cache has 2 slots in use, and we know three potential pairs.
7486 Keep the two that give the lowest RMS distance. Do the
7487 calculation in bytes simply because we always know the byte
7488 length. squareroot has the same ordering as the positive value,
7489 so don't bother with the actual square root. */
7490 if (byte > cache[1]) {
7491 /* New position is after the existing pair of pairs. */
7492 const float keep_earlier
7493 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7494 const float keep_later
7495 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7497 if (keep_later < keep_earlier) {
7498 cache[2] = cache[0];
7499 cache[3] = cache[1];
7505 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7506 float b, c, keep_earlier;
7507 if (byte > cache[3]) {
7508 /* New position is between the existing pair of pairs. */
7509 b = (float)cache[3];
7512 /* New position is before the existing pair of pairs. */
7514 c = (float)cache[3];
7516 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7517 if (byte > cache[3]) {
7518 if (keep_later < keep_earlier) {
7528 if (! (keep_later < keep_earlier)) {
7529 cache[0] = cache[2];
7530 cache[1] = cache[3];
7537 ASSERT_UTF8_CACHE(cache);
7540 /* We already know all of the way, now we may be able to walk back. The same
7541 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7542 backward is half the speed of walking forward. */
7544 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7545 const U8 *end, STRLEN endu)
7547 const STRLEN forw = target - s;
7548 STRLEN backw = end - target;
7550 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7552 if (forw < 2 * backw) {
7553 return utf8_length(s, target);
7556 while (end > target) {
7558 while (UTF8_IS_CONTINUATION(*end)) {
7567 =for apidoc sv_pos_b2u_flags
7569 Converts the offset from a count of bytes from the start of the string, to
7570 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7571 I<flags> is passed to C<SvPV_flags>, and usually should be
7572 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7578 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7579 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7584 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7587 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7593 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7595 s = (const U8*)SvPV_flags(sv, blen, flags);
7598 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7599 ", byte=%"UVuf, (UV)blen, (UV)offset);
7605 && SvTYPE(sv) >= SVt_PVMG
7606 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7609 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7610 if (cache[1] == offset) {
7611 /* An exact match. */
7614 if (cache[3] == offset) {
7615 /* An exact match. */
7619 if (cache[1] < offset) {
7620 /* We already know part of the way. */
7621 if (mg->mg_len != -1) {
7622 /* Actually, we know the end too. */
7624 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7625 s + blen, mg->mg_len - cache[0]);
7627 len = cache[0] + utf8_length(s + cache[1], send);
7630 else if (cache[3] < offset) {
7631 /* We're between the two cached pairs, so we do the calculation
7632 offset by the byte/utf-8 positions for the earlier pair,
7633 then add the utf-8 characters from the string start to
7635 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7636 s + cache[1], cache[0] - cache[2])
7640 else { /* cache[3] > offset */
7641 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7645 ASSERT_UTF8_CACHE(cache);
7647 } else if (mg->mg_len != -1) {
7648 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7652 if (!found || PL_utf8cache < 0) {
7653 const STRLEN real_len = utf8_length(s, send);
7655 if (found && PL_utf8cache < 0)
7656 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7662 utf8_mg_len_cache_update(sv, &mg, len);
7664 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7671 =for apidoc sv_pos_b2u
7673 Converts the value pointed to by offsetp from a count of bytes from the
7674 start of the string, to a count of the equivalent number of UTF-8 chars.
7675 Handles magic and type coercion.
7677 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7684 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7685 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7690 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7692 PERL_ARGS_ASSERT_SV_POS_B2U;
7697 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7698 SV_GMAGIC|SV_CONST_RETURN);
7702 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7703 STRLEN real, SV *const sv)
7705 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7707 /* As this is debugging only code, save space by keeping this test here,
7708 rather than inlining it in all the callers. */
7709 if (from_cache == real)
7712 /* Need to turn the assertions off otherwise we may recurse infinitely
7713 while printing error messages. */
7714 SAVEI8(PL_utf8cache);
7716 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7717 func, (UV) from_cache, (UV) real, SVfARG(sv));
7723 Returns a boolean indicating whether the strings in the two SVs are
7724 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7725 coerce its args to strings if necessary.
7727 =for apidoc sv_eq_flags
7729 Returns a boolean indicating whether the strings in the two SVs are
7730 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7731 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7737 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7744 SV* svrecode = NULL;
7751 /* if pv1 and pv2 are the same, second SvPV_const call may
7752 * invalidate pv1 (if we are handling magic), so we may need to
7754 if (sv1 == sv2 && flags & SV_GMAGIC
7755 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7756 pv1 = SvPV_const(sv1, cur1);
7757 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7759 pv1 = SvPV_flags_const(sv1, cur1, flags);
7767 pv2 = SvPV_flags_const(sv2, cur2, flags);
7769 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7770 /* Differing utf8ness.
7771 * Do not UTF8size the comparands as a side-effect. */
7774 svrecode = newSVpvn(pv2, cur2);
7775 sv_recode_to_utf8(svrecode, _get_encoding());
7776 pv2 = SvPV_const(svrecode, cur2);
7779 svrecode = newSVpvn(pv1, cur1);
7780 sv_recode_to_utf8(svrecode, _get_encoding());
7781 pv1 = SvPV_const(svrecode, cur1);
7783 /* Now both are in UTF-8. */
7785 SvREFCNT_dec_NN(svrecode);
7791 /* sv1 is the UTF-8 one */
7792 return bytes_cmp_utf8((const U8*)pv2, cur2,
7793 (const U8*)pv1, cur1) == 0;
7796 /* sv2 is the UTF-8 one */
7797 return bytes_cmp_utf8((const U8*)pv1, cur1,
7798 (const U8*)pv2, cur2) == 0;
7804 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7806 SvREFCNT_dec(svrecode);
7814 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7815 string in C<sv1> is less than, equal to, or greater than the string in
7816 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7817 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7819 =for apidoc sv_cmp_flags
7821 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7822 string in C<sv1> is less than, equal to, or greater than the string in
7823 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7824 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7825 also C<sv_cmp_locale_flags>.
7831 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7833 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7837 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7841 const char *pv1, *pv2;
7843 SV *svrecode = NULL;
7850 pv1 = SvPV_flags_const(sv1, cur1, flags);
7857 pv2 = SvPV_flags_const(sv2, cur2, flags);
7859 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7860 /* Differing utf8ness.
7861 * Do not UTF8size the comparands as a side-effect. */
7864 svrecode = newSVpvn(pv2, cur2);
7865 sv_recode_to_utf8(svrecode, _get_encoding());
7866 pv2 = SvPV_const(svrecode, cur2);
7869 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7870 (const U8*)pv1, cur1);
7871 return retval ? retval < 0 ? -1 : +1 : 0;
7876 svrecode = newSVpvn(pv1, cur1);
7877 sv_recode_to_utf8(svrecode, _get_encoding());
7878 pv1 = SvPV_const(svrecode, cur1);
7881 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7882 (const U8*)pv2, cur2);
7883 return retval ? retval < 0 ? -1 : +1 : 0;
7889 cmp = cur2 ? -1 : 0;
7893 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7896 cmp = retval < 0 ? -1 : 1;
7897 } else if (cur1 == cur2) {
7900 cmp = cur1 < cur2 ? -1 : 1;
7904 SvREFCNT_dec(svrecode);
7910 =for apidoc sv_cmp_locale
7912 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7913 'use bytes' aware, handles get magic, and will coerce its args to strings
7914 if necessary. See also C<sv_cmp>.
7916 =for apidoc sv_cmp_locale_flags
7918 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7919 'use bytes' aware and will coerce its args to strings if necessary. If the
7920 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7926 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7928 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7932 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7935 #ifdef USE_LOCALE_COLLATE
7941 if (PL_collation_standard)
7945 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7947 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7949 if (!pv1 || !len1) {
7960 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7963 return retval < 0 ? -1 : 1;
7966 * When the result of collation is equality, that doesn't mean
7967 * that there are no differences -- some locales exclude some
7968 * characters from consideration. So to avoid false equalities,
7969 * we use the raw string as a tiebreaker.
7976 PERL_UNUSED_ARG(flags);
7977 #endif /* USE_LOCALE_COLLATE */
7979 return sv_cmp(sv1, sv2);
7983 #ifdef USE_LOCALE_COLLATE
7986 =for apidoc sv_collxfrm
7988 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7989 C<sv_collxfrm_flags>.
7991 =for apidoc sv_collxfrm_flags
7993 Add Collate Transform magic to an SV if it doesn't already have it. If the
7994 flags contain SV_GMAGIC, it handles get-magic.
7996 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7997 scalar data of the variable, but transformed to such a format that a normal
7998 memory comparison can be used to compare the data according to the locale
8005 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8009 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8011 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8012 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8018 Safefree(mg->mg_ptr);
8019 s = SvPV_flags_const(sv, len, flags);
8020 if ((xf = mem_collxfrm(s, len, &xlen))) {
8022 #ifdef PERL_OLD_COPY_ON_WRITE
8024 sv_force_normal_flags(sv, 0);
8026 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8040 if (mg && mg->mg_ptr) {
8042 return mg->mg_ptr + sizeof(PL_collation_ix);
8050 #endif /* USE_LOCALE_COLLATE */
8053 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8055 SV * const tsv = newSV(0);
8058 sv_gets(tsv, fp, 0);
8059 sv_utf8_upgrade_nomg(tsv);
8060 SvCUR_set(sv,append);
8063 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8067 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8070 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8071 /* Grab the size of the record we're getting */
8072 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8079 /* With a true, record-oriented file on VMS, we need to use read directly
8080 * to ensure that we respect RMS record boundaries. The user is responsible
8081 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8082 * record size) field. N.B. This is likely to produce invalid results on
8083 * varying-width character data when a record ends mid-character.
8085 fd = PerlIO_fileno(fp);
8087 && PerlLIO_fstat(fd, &st) == 0
8088 && (st.st_fab_rfm == FAB$C_VAR
8089 || st.st_fab_rfm == FAB$C_VFC
8090 || st.st_fab_rfm == FAB$C_FIX)) {
8092 bytesread = PerlLIO_read(fd, buffer, recsize);
8094 else /* in-memory file from PerlIO::Scalar
8095 * or not a record-oriented file
8099 bytesread = PerlIO_read(fp, buffer, recsize);
8101 /* At this point, the logic in sv_get() means that sv will
8102 be treated as utf-8 if the handle is utf8.
8104 if (PerlIO_isutf8(fp) && bytesread > 0) {
8105 char *bend = buffer + bytesread;
8106 char *bufp = buffer;
8107 size_t charcount = 0;
8108 bool charstart = TRUE;
8111 while (charcount < recsize) {
8112 /* count accumulated characters */
8113 while (bufp < bend) {
8115 skip = UTF8SKIP(bufp);
8117 if (bufp + skip > bend) {
8118 /* partial at the end */
8129 if (charcount < recsize) {
8131 STRLEN bufp_offset = bufp - buffer;
8132 SSize_t morebytesread;
8134 /* originally I read enough to fill any incomplete
8135 character and the first byte of the next
8136 character if needed, but if there's many
8137 multi-byte encoded characters we're going to be
8138 making a read call for every character beyond
8139 the original read size.
8141 So instead, read the rest of the character if
8142 any, and enough bytes to match at least the
8143 start bytes for each character we're going to
8147 readsize = recsize - charcount;
8149 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8150 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8151 bend = buffer + bytesread;
8152 morebytesread = PerlIO_read(fp, bend, readsize);
8153 if (morebytesread <= 0) {
8154 /* we're done, if we still have incomplete
8155 characters the check code in sv_gets() will
8158 I'd originally considered doing
8159 PerlIO_ungetc() on all but the lead
8160 character of the incomplete character, but
8161 read() doesn't do that, so I don't.
8166 /* prepare to scan some more */
8167 bytesread += morebytesread;
8168 bend = buffer + bytesread;
8169 bufp = buffer + bufp_offset;
8177 SvCUR_set(sv, bytesread + append);
8178 buffer[bytesread] = '\0';
8179 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8185 Get a line from the filehandle and store it into the SV, optionally
8186 appending to the currently-stored string. If C<append> is not 0, the
8187 line is appended to the SV instead of overwriting it. C<append> should
8188 be set to the byte offset that the appended string should start at
8189 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8195 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8205 PERL_ARGS_ASSERT_SV_GETS;
8207 if (SvTHINKFIRST(sv))
8208 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8209 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8211 However, perlbench says it's slower, because the existing swipe code
8212 is faster than copy on write.
8213 Swings and roundabouts. */
8214 SvUPGRADE(sv, SVt_PV);
8217 /* line is going to be appended to the existing buffer in the sv */
8218 if (PerlIO_isutf8(fp)) {
8220 sv_utf8_upgrade_nomg(sv);
8221 sv_pos_u2b(sv,&append,0);
8223 } else if (SvUTF8(sv)) {
8224 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8230 /* not appending - "clear" the string by setting SvCUR to 0,
8231 * the pv is still avaiable. */
8234 if (PerlIO_isutf8(fp))
8237 if (IN_PERL_COMPILETIME) {
8238 /* we always read code in line mode */
8242 else if (RsSNARF(PL_rs)) {
8243 /* If it is a regular disk file use size from stat() as estimate
8244 of amount we are going to read -- may result in mallocing
8245 more memory than we really need if the layers below reduce
8246 the size we read (e.g. CRLF or a gzip layer).
8249 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8250 const Off_t offset = PerlIO_tell(fp);
8251 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8252 #ifdef PERL_NEW_COPY_ON_WRITE
8253 /* Add an extra byte for the sake of copy-on-write's
8254 * buffer reference count. */
8255 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8257 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8264 else if (RsRECORD(PL_rs)) {
8265 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8267 else if (RsPARA(PL_rs)) {
8273 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8274 if (PerlIO_isutf8(fp)) {
8275 rsptr = SvPVutf8(PL_rs, rslen);
8278 if (SvUTF8(PL_rs)) {
8279 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8280 Perl_croak(aTHX_ "Wide character in $/");
8283 /* extract the raw pointer to the record separator */
8284 rsptr = SvPV_const(PL_rs, rslen);
8288 /* rslast is the last character in the record separator
8289 * note we don't use rslast except when rslen is true, so the
8290 * null assign is a placeholder. */
8291 rslast = rslen ? rsptr[rslen - 1] : '\0';
8293 if (rspara) { /* have to do this both before and after */
8294 do { /* to make sure file boundaries work right */
8297 i = PerlIO_getc(fp);
8301 PerlIO_ungetc(fp,i);
8307 /* See if we know enough about I/O mechanism to cheat it ! */
8309 /* This used to be #ifdef test - it is made run-time test for ease
8310 of abstracting out stdio interface. One call should be cheap
8311 enough here - and may even be a macro allowing compile
8315 if (PerlIO_fast_gets(fp)) {
8317 * We can do buffer based IO operations on this filehandle.
8319 * This means we can bypass a lot of subcalls and process
8320 * the buffer directly, it also means we know the upper bound
8321 * on the amount of data we might read of the current buffer
8322 * into our sv. Knowing this allows us to preallocate the pv
8323 * to be able to hold that maximum, which allows us to simplify
8324 * a lot of logic. */
8327 * We're going to steal some values from the stdio struct
8328 * and put EVERYTHING in the innermost loop into registers.
8330 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8331 STRLEN bpx; /* length of the data in the target sv
8332 used to fix pointers after a SvGROW */
8333 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8334 of data left in the read-ahead buffer.
8335 If 0 then the pv buffer can hold the full
8336 amount left, otherwise this is the amount it
8339 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8340 /* An ungetc()d char is handled separately from the regular
8341 * buffer, so we getc() it back out and stuff it in the buffer.
8343 i = PerlIO_getc(fp);
8344 if (i == EOF) return 0;
8345 *(--((*fp)->_ptr)) = (unsigned char) i;
8349 /* Here is some breathtakingly efficient cheating */
8351 /* When you read the following logic resist the urge to think
8352 * of record separators that are 1 byte long. They are an
8353 * uninteresting special (simple) case.
8355 * Instead think of record separators which are at least 2 bytes
8356 * long, and keep in mind that we need to deal with such
8357 * separators when they cross a read-ahead buffer boundary.
8359 * Also consider that we need to gracefully deal with separators
8360 * that may be longer than a single read ahead buffer.
8362 * Lastly do not forget we want to copy the delimiter as well. We
8363 * are copying all data in the file _up_to_and_including_ the separator
8366 * Now that you have all that in mind here is what is happening below:
8368 * 1. When we first enter the loop we do some memory book keeping to see
8369 * how much free space there is in the target SV. (This sub assumes that
8370 * it is operating on the same SV most of the time via $_ and that it is
8371 * going to be able to reuse the same pv buffer each call.) If there is
8372 * "enough" room then we set "shortbuffered" to how much space there is
8373 * and start reading forward.
8375 * 2. When we scan forward we copy from the read-ahead buffer to the target
8376 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8377 * and the end of the of pv, as well as for the "rslast", which is the last
8378 * char of the separator.
8380 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8381 * (which has a "complete" record up to the point we saw rslast) and check
8382 * it to see if it matches the separator. If it does we are done. If it doesn't
8383 * we continue on with the scan/copy.
8385 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8386 * the IO system to read the next buffer. We do this by doing a getc(), which
8387 * returns a single char read (or EOF), and prefills the buffer, and also
8388 * allows us to find out how full the buffer is. We use this information to
8389 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8390 * the returned single char into the target sv, and then go back into scan
8393 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8394 * remaining space in the read-buffer.
8396 * Note that this code despite its twisty-turny nature is pretty darn slick.
8397 * It manages single byte separators, multi-byte cross boundary separators,
8398 * and cross-read-buffer separators cleanly and efficiently at the cost
8399 * of potentially greatly overallocating the target SV.
8405 /* get the number of bytes remaining in the read-ahead buffer
8406 * on first call on a given fp this will return 0.*/
8407 cnt = PerlIO_get_cnt(fp);
8409 /* make sure we have the room */
8410 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8411 /* Not room for all of it
8412 if we are looking for a separator and room for some
8414 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8415 /* just process what we have room for */
8416 shortbuffered = cnt - SvLEN(sv) + append + 1;
8417 cnt -= shortbuffered;
8420 /* ensure that the target sv has enough room to hold
8421 * the rest of the read-ahead buffer */
8423 /* remember that cnt can be negative */
8424 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8428 /* we have enough room to hold the full buffer, lets scream */
8432 /* extract the pointer to sv's string buffer, offset by append as necessary */
8433 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8434 /* extract the point to the read-ahead buffer */
8435 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8437 /* some trace debug output */
8438 DEBUG_P(PerlIO_printf(Perl_debug_log,
8439 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8440 DEBUG_P(PerlIO_printf(Perl_debug_log,
8441 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8443 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8444 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8448 /* if there is stuff left in the read-ahead buffer */
8450 /* if there is a separator */
8452 /* loop until we hit the end of the read-ahead buffer */
8453 while (cnt > 0) { /* this | eat */
8454 /* scan forward copying and searching for rslast as we go */
8456 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8457 goto thats_all_folks; /* screams | sed :-) */
8461 /* no separator, slurp the full buffer */
8462 Copy(ptr, bp, cnt, char); /* this | eat */
8463 bp += cnt; /* screams | dust */
8464 ptr += cnt; /* louder | sed :-) */
8466 assert (!shortbuffered);
8467 goto cannot_be_shortbuffered;
8471 if (shortbuffered) { /* oh well, must extend */
8472 /* we didnt have enough room to fit the line into the target buffer
8473 * so we must extend the target buffer and keep going */
8474 cnt = shortbuffered;
8476 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8478 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8479 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8480 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8484 cannot_be_shortbuffered:
8485 /* we need to refill the read-ahead buffer if possible */
8487 DEBUG_P(PerlIO_printf(Perl_debug_log,
8488 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8489 PTR2UV(ptr),(IV)cnt));
8490 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8492 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8493 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8494 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8495 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8498 call PerlIO_getc() to let it prefill the lookahead buffer
8500 This used to call 'filbuf' in stdio form, but as that behaves like
8501 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8502 another abstraction.
8504 Note we have to deal with the char in 'i' if we are not at EOF
8506 i = PerlIO_getc(fp); /* get more characters */
8508 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8509 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8510 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8511 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8513 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8514 cnt = PerlIO_get_cnt(fp);
8515 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8516 DEBUG_P(PerlIO_printf(Perl_debug_log,
8517 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8518 PTR2UV(ptr),(IV)cnt));
8520 if (i == EOF) /* all done for ever? */
8521 goto thats_really_all_folks;
8523 /* make sure we have enough space in the target sv */
8524 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8526 SvGROW(sv, bpx + cnt + 2);
8527 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8529 /* copy of the char we got from getc() */
8530 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8532 /* make sure we deal with the i being the last character of a separator */
8533 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8534 goto thats_all_folks;
8538 /* check if we have actually found the separator - only really applies
8540 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8541 memNE((char*)bp - rslen, rsptr, rslen))
8542 goto screamer; /* go back to the fray */
8543 thats_really_all_folks:
8545 cnt += shortbuffered;
8546 DEBUG_P(PerlIO_printf(Perl_debug_log,
8547 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8548 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8549 DEBUG_P(PerlIO_printf(Perl_debug_log,
8550 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8552 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8553 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8555 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8556 DEBUG_P(PerlIO_printf(Perl_debug_log,
8557 "Screamer: done, len=%ld, string=|%.*s|\n",
8558 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8562 /*The big, slow, and stupid way. */
8563 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8564 STDCHAR *buf = NULL;
8565 Newx(buf, 8192, STDCHAR);
8573 const STDCHAR * const bpe = buf + sizeof(buf);
8575 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8576 ; /* keep reading */
8580 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8581 /* Accommodate broken VAXC compiler, which applies U8 cast to
8582 * both args of ?: operator, causing EOF to change into 255
8585 i = (U8)buf[cnt - 1];
8591 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8593 sv_catpvn_nomg(sv, (char *) buf, cnt);
8595 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8597 if (i != EOF && /* joy */
8599 SvCUR(sv) < rslen ||
8600 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8604 * If we're reading from a TTY and we get a short read,
8605 * indicating that the user hit his EOF character, we need
8606 * to notice it now, because if we try to read from the TTY
8607 * again, the EOF condition will disappear.
8609 * The comparison of cnt to sizeof(buf) is an optimization
8610 * that prevents unnecessary calls to feof().
8614 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8618 #ifdef USE_HEAP_INSTEAD_OF_STACK
8623 if (rspara) { /* have to do this both before and after */
8624 while (i != EOF) { /* to make sure file boundaries work right */
8625 i = PerlIO_getc(fp);
8627 PerlIO_ungetc(fp,i);
8633 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8639 Auto-increment of the value in the SV, doing string to numeric conversion
8640 if necessary. Handles 'get' magic and operator overloading.
8646 Perl_sv_inc(pTHX_ SV *const sv)
8655 =for apidoc sv_inc_nomg
8657 Auto-increment of the value in the SV, doing string to numeric conversion
8658 if necessary. Handles operator overloading. Skips handling 'get' magic.
8664 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8671 if (SvTHINKFIRST(sv)) {
8672 if (SvREADONLY(sv)) {
8673 Perl_croak_no_modify();
8677 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8679 i = PTR2IV(SvRV(sv));
8683 else sv_force_normal_flags(sv, 0);
8685 flags = SvFLAGS(sv);
8686 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8687 /* It's (privately or publicly) a float, but not tested as an
8688 integer, so test it to see. */
8690 flags = SvFLAGS(sv);
8692 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8693 /* It's publicly an integer, or privately an integer-not-float */
8694 #ifdef PERL_PRESERVE_IVUV
8698 if (SvUVX(sv) == UV_MAX)
8699 sv_setnv(sv, UV_MAX_P1);
8701 (void)SvIOK_only_UV(sv);
8702 SvUV_set(sv, SvUVX(sv) + 1);
8704 if (SvIVX(sv) == IV_MAX)
8705 sv_setuv(sv, (UV)IV_MAX + 1);
8707 (void)SvIOK_only(sv);
8708 SvIV_set(sv, SvIVX(sv) + 1);
8713 if (flags & SVp_NOK) {
8714 const NV was = SvNVX(sv);
8715 if (LIKELY(!Perl_isinfnan(was)) &&
8716 NV_OVERFLOWS_INTEGERS_AT &&
8717 was >= NV_OVERFLOWS_INTEGERS_AT) {
8718 /* diag_listed_as: Lost precision when %s %f by 1 */
8719 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8720 "Lost precision when incrementing %" NVff " by 1",
8723 (void)SvNOK_only(sv);
8724 SvNV_set(sv, was + 1.0);
8728 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8729 if ((flags & SVTYPEMASK) < SVt_PVIV)
8730 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8731 (void)SvIOK_only(sv);
8736 while (isALPHA(*d)) d++;
8737 while (isDIGIT(*d)) d++;
8738 if (d < SvEND(sv)) {
8739 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8740 #ifdef PERL_PRESERVE_IVUV
8741 /* Got to punt this as an integer if needs be, but we don't issue
8742 warnings. Probably ought to make the sv_iv_please() that does
8743 the conversion if possible, and silently. */
8744 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8745 /* Need to try really hard to see if it's an integer.
8746 9.22337203685478e+18 is an integer.
8747 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8748 so $a="9.22337203685478e+18"; $a+0; $a++
8749 needs to be the same as $a="9.22337203685478e+18"; $a++
8756 /* sv_2iv *should* have made this an NV */
8757 if (flags & SVp_NOK) {
8758 (void)SvNOK_only(sv);
8759 SvNV_set(sv, SvNVX(sv) + 1.0);
8762 /* I don't think we can get here. Maybe I should assert this
8763 And if we do get here I suspect that sv_setnv will croak. NWC
8765 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8766 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8768 #endif /* PERL_PRESERVE_IVUV */
8769 if (!numtype && ckWARN(WARN_NUMERIC))
8770 not_incrementable(sv);
8771 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8775 while (d >= SvPVX_const(sv)) {
8783 /* MKS: The original code here died if letters weren't consecutive.
8784 * at least it didn't have to worry about non-C locales. The
8785 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8786 * arranged in order (although not consecutively) and that only
8787 * [A-Za-z] are accepted by isALPHA in the C locale.
8789 if (isALPHA_FOLD_NE(*d, 'z')) {
8790 do { ++*d; } while (!isALPHA(*d));
8793 *(d--) -= 'z' - 'a';
8798 *(d--) -= 'z' - 'a' + 1;
8802 /* oh,oh, the number grew */
8803 SvGROW(sv, SvCUR(sv) + 2);
8804 SvCUR_set(sv, SvCUR(sv) + 1);
8805 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8816 Auto-decrement of the value in the SV, doing string to numeric conversion
8817 if necessary. Handles 'get' magic and operator overloading.
8823 Perl_sv_dec(pTHX_ SV *const sv)
8832 =for apidoc sv_dec_nomg
8834 Auto-decrement of the value in the SV, doing string to numeric conversion
8835 if necessary. Handles operator overloading. Skips handling 'get' magic.
8841 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8847 if (SvTHINKFIRST(sv)) {
8848 if (SvREADONLY(sv)) {
8849 Perl_croak_no_modify();
8853 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8855 i = PTR2IV(SvRV(sv));
8859 else sv_force_normal_flags(sv, 0);
8861 /* Unlike sv_inc we don't have to worry about string-never-numbers
8862 and keeping them magic. But we mustn't warn on punting */
8863 flags = SvFLAGS(sv);
8864 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8865 /* It's publicly an integer, or privately an integer-not-float */
8866 #ifdef PERL_PRESERVE_IVUV
8870 if (SvUVX(sv) == 0) {
8871 (void)SvIOK_only(sv);
8875 (void)SvIOK_only_UV(sv);
8876 SvUV_set(sv, SvUVX(sv) - 1);
8879 if (SvIVX(sv) == IV_MIN) {
8880 sv_setnv(sv, (NV)IV_MIN);
8884 (void)SvIOK_only(sv);
8885 SvIV_set(sv, SvIVX(sv) - 1);
8890 if (flags & SVp_NOK) {
8893 const NV was = SvNVX(sv);
8894 if (LIKELY(!Perl_isinfnan(was)) &&
8895 NV_OVERFLOWS_INTEGERS_AT &&
8896 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8897 /* diag_listed_as: Lost precision when %s %f by 1 */
8898 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8899 "Lost precision when decrementing %" NVff " by 1",
8902 (void)SvNOK_only(sv);
8903 SvNV_set(sv, was - 1.0);
8907 if (!(flags & SVp_POK)) {
8908 if ((flags & SVTYPEMASK) < SVt_PVIV)
8909 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8911 (void)SvIOK_only(sv);
8914 #ifdef PERL_PRESERVE_IVUV
8916 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8917 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8918 /* Need to try really hard to see if it's an integer.
8919 9.22337203685478e+18 is an integer.
8920 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8921 so $a="9.22337203685478e+18"; $a+0; $a--
8922 needs to be the same as $a="9.22337203685478e+18"; $a--
8929 /* sv_2iv *should* have made this an NV */
8930 if (flags & SVp_NOK) {
8931 (void)SvNOK_only(sv);
8932 SvNV_set(sv, SvNVX(sv) - 1.0);
8935 /* I don't think we can get here. Maybe I should assert this
8936 And if we do get here I suspect that sv_setnv will croak. NWC
8938 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8939 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8942 #endif /* PERL_PRESERVE_IVUV */
8943 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8946 /* this define is used to eliminate a chunk of duplicated but shared logic
8947 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8948 * used anywhere but here - yves
8950 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8952 SSize_t ix = ++PL_tmps_ix; \
8953 if (UNLIKELY(ix >= PL_tmps_max)) \
8954 ix = tmps_grow_p(ix); \
8955 PL_tmps_stack[ix] = (AnSv); \
8959 =for apidoc sv_mortalcopy
8961 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8962 The new SV is marked as mortal. It will be destroyed "soon", either by an
8963 explicit call to FREETMPS, or by an implicit call at places such as
8964 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8969 /* Make a string that will exist for the duration of the expression
8970 * evaluation. Actually, it may have to last longer than that, but
8971 * hopefully we won't free it until it has been assigned to a
8972 * permanent location. */
8975 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8979 if (flags & SV_GMAGIC)
8980 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8982 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8983 PUSH_EXTEND_MORTAL__SV_C(sv);
8989 =for apidoc sv_newmortal
8991 Creates a new null SV which is mortal. The reference count of the SV is
8992 set to 1. It will be destroyed "soon", either by an explicit call to
8993 FREETMPS, or by an implicit call at places such as statement boundaries.
8994 See also C<sv_mortalcopy> and C<sv_2mortal>.
9000 Perl_sv_newmortal(pTHX)
9005 SvFLAGS(sv) = SVs_TEMP;
9006 PUSH_EXTEND_MORTAL__SV_C(sv);
9012 =for apidoc newSVpvn_flags
9014 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9015 characters) into it. The reference count for the
9016 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9017 string. You are responsible for ensuring that the source string is at least
9018 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9019 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9020 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9021 returning. If C<SVf_UTF8> is set, C<s>
9022 is considered to be in UTF-8 and the
9023 C<SVf_UTF8> flag will be set on the new SV.
9024 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9026 #define newSVpvn_utf8(s, len, u) \
9027 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9033 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9037 /* All the flags we don't support must be zero.
9038 And we're new code so I'm going to assert this from the start. */
9039 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9041 sv_setpvn(sv,s,len);
9043 /* This code used to do a sv_2mortal(), however we now unroll the call to
9044 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9045 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9046 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9047 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9048 * means that we eliminate quite a few steps than it looks - Yves
9049 * (explaining patch by gfx) */
9051 SvFLAGS(sv) |= flags;
9053 if(flags & SVs_TEMP){
9054 PUSH_EXTEND_MORTAL__SV_C(sv);
9061 =for apidoc sv_2mortal
9063 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9064 by an explicit call to FREETMPS, or by an implicit call at places such as
9065 statement boundaries. SvTEMP() is turned on which means that the SV's
9066 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9067 and C<sv_mortalcopy>.
9073 Perl_sv_2mortal(pTHX_ SV *const sv)
9080 PUSH_EXTEND_MORTAL__SV_C(sv);
9088 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9089 characters) into it. The reference count for the
9090 SV is set to 1. If C<len> is zero, Perl will compute the length using
9091 strlen(), (which means if you use this option, that C<s> can't have embedded
9092 C<NUL> characters and has to have a terminating C<NUL> byte).
9094 For efficiency, consider using C<newSVpvn> instead.
9100 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9105 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9110 =for apidoc newSVpvn
9112 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9113 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9114 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9115 are responsible for ensuring that the source buffer is at least
9116 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9123 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9127 sv_setpvn(sv,buffer,len);
9132 =for apidoc newSVhek
9134 Creates a new SV from the hash key structure. It will generate scalars that
9135 point to the shared string table where possible. Returns a new (undefined)
9136 SV if the hek is NULL.
9142 Perl_newSVhek(pTHX_ const HEK *const hek)
9151 if (HEK_LEN(hek) == HEf_SVKEY) {
9152 return newSVsv(*(SV**)HEK_KEY(hek));
9154 const int flags = HEK_FLAGS(hek);
9155 if (flags & HVhek_WASUTF8) {
9157 Andreas would like keys he put in as utf8 to come back as utf8
9159 STRLEN utf8_len = HEK_LEN(hek);
9160 SV * const sv = newSV_type(SVt_PV);
9161 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9162 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9163 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9166 } else if (flags & HVhek_UNSHARED) {
9167 /* A hash that isn't using shared hash keys has to have
9168 the flag in every key so that we know not to try to call
9169 share_hek_hek on it. */
9171 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9176 /* This will be overwhelminly the most common case. */
9178 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9179 more efficient than sharepvn(). */
9183 sv_upgrade(sv, SVt_PV);
9184 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9185 SvCUR_set(sv, HEK_LEN(hek));
9197 =for apidoc newSVpvn_share
9199 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9200 table. If the string does not already exist in the table, it is
9201 created first. Turns on the SvIsCOW flag (or READONLY
9202 and FAKE in 5.16 and earlier). If the C<hash> parameter
9203 is non-zero, that value is used; otherwise the hash is computed.
9204 The string's hash can later be retrieved from the SV
9205 with the C<SvSHARED_HASH()> macro. The idea here is
9206 that as the string table is used for shared hash keys these strings will have
9207 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9213 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9217 bool is_utf8 = FALSE;
9218 const char *const orig_src = src;
9221 STRLEN tmplen = -len;
9223 /* See the note in hv.c:hv_fetch() --jhi */
9224 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9228 PERL_HASH(hash, src, len);
9230 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9231 changes here, update it there too. */
9232 sv_upgrade(sv, SVt_PV);
9233 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9240 if (src != orig_src)
9246 =for apidoc newSVpv_share
9248 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9255 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9257 return newSVpvn_share(src, strlen(src), hash);
9260 #if defined(PERL_IMPLICIT_CONTEXT)
9262 /* pTHX_ magic can't cope with varargs, so this is a no-context
9263 * version of the main function, (which may itself be aliased to us).
9264 * Don't access this version directly.
9268 Perl_newSVpvf_nocontext(const char *const pat, ...)
9274 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9276 va_start(args, pat);
9277 sv = vnewSVpvf(pat, &args);
9284 =for apidoc newSVpvf
9286 Creates a new SV and initializes it with the string formatted like
9293 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9298 PERL_ARGS_ASSERT_NEWSVPVF;
9300 va_start(args, pat);
9301 sv = vnewSVpvf(pat, &args);
9306 /* backend for newSVpvf() and newSVpvf_nocontext() */
9309 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9313 PERL_ARGS_ASSERT_VNEWSVPVF;
9316 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9323 Creates a new SV and copies a floating point value into it.
9324 The reference count for the SV is set to 1.
9330 Perl_newSVnv(pTHX_ const NV n)
9342 Creates a new SV and copies an integer into it. The reference count for the
9349 Perl_newSViv(pTHX_ const IV i)
9355 /* Inlining ONLY the small relevant subset of sv_setiv here
9356 * for performance. Makes a significant difference. */
9358 /* We're starting from SVt_FIRST, so provided that's
9359 * actual 0, we don't have to unset any SV type flags
9360 * to promote to SVt_IV. */
9361 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9363 SET_SVANY_FOR_BODYLESS_IV(sv);
9364 SvFLAGS(sv) |= SVt_IV;
9376 Creates a new SV and copies an unsigned integer into it.
9377 The reference count for the SV is set to 1.
9383 Perl_newSVuv(pTHX_ const UV u)
9387 /* Inlining ONLY the small relevant subset of sv_setuv here
9388 * for performance. Makes a significant difference. */
9390 /* Using ivs is more efficient than using uvs - see sv_setuv */
9391 if (u <= (UV)IV_MAX) {
9392 return newSViv((IV)u);
9397 /* We're starting from SVt_FIRST, so provided that's
9398 * actual 0, we don't have to unset any SV type flags
9399 * to promote to SVt_IV. */
9400 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9402 SET_SVANY_FOR_BODYLESS_IV(sv);
9403 SvFLAGS(sv) |= SVt_IV;
9405 (void)SvIsUV_on(sv);
9414 =for apidoc newSV_type
9416 Creates a new SV, of the type specified. The reference count for the new SV
9423 Perl_newSV_type(pTHX_ const svtype type)
9428 ASSUME(SvTYPE(sv) == SVt_FIRST);
9429 if(type != SVt_FIRST)
9430 sv_upgrade(sv, type);
9435 =for apidoc newRV_noinc
9437 Creates an RV wrapper for an SV. The reference count for the original
9438 SV is B<not> incremented.
9444 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9448 PERL_ARGS_ASSERT_NEWRV_NOINC;
9452 /* We're starting from SVt_FIRST, so provided that's
9453 * actual 0, we don't have to unset any SV type flags
9454 * to promote to SVt_IV. */
9455 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9457 SET_SVANY_FOR_BODYLESS_IV(sv);
9458 SvFLAGS(sv) |= SVt_IV;
9463 SvRV_set(sv, tmpRef);
9468 /* newRV_inc is the official function name to use now.
9469 * newRV_inc is in fact #defined to newRV in sv.h
9473 Perl_newRV(pTHX_ SV *const sv)
9475 PERL_ARGS_ASSERT_NEWRV;
9477 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9483 Creates a new SV which is an exact duplicate of the original SV.
9490 Perl_newSVsv(pTHX_ SV *const old)
9496 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9497 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9500 /* Do this here, otherwise we leak the new SV if this croaks. */
9503 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9504 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9505 sv_setsv_flags(sv, old, SV_NOSTEAL);
9510 =for apidoc sv_reset
9512 Underlying implementation for the C<reset> Perl function.
9513 Note that the perl-level function is vaguely deprecated.
9519 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9521 PERL_ARGS_ASSERT_SV_RESET;
9523 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9527 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9529 char todo[PERL_UCHAR_MAX+1];
9532 if (!stash || SvTYPE(stash) != SVt_PVHV)
9535 if (!s) { /* reset ?? searches */
9536 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9538 const U32 count = mg->mg_len / sizeof(PMOP**);
9539 PMOP **pmp = (PMOP**) mg->mg_ptr;
9540 PMOP *const *const end = pmp + count;
9544 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9546 (*pmp)->op_pmflags &= ~PMf_USED;
9554 /* reset variables */
9556 if (!HvARRAY(stash))
9559 Zero(todo, 256, char);
9563 I32 i = (unsigned char)*s;
9567 max = (unsigned char)*s++;
9568 for ( ; i <= max; i++) {
9571 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9573 for (entry = HvARRAY(stash)[i];
9575 entry = HeNEXT(entry))
9580 if (!todo[(U8)*HeKEY(entry)])
9582 gv = MUTABLE_GV(HeVAL(entry));
9584 if (sv && !SvREADONLY(sv)) {
9585 SV_CHECK_THINKFIRST_COW_DROP(sv);
9586 if (!isGV(sv)) SvOK_off(sv);
9591 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9602 Using various gambits, try to get an IO from an SV: the IO slot if its a
9603 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9604 named after the PV if we're a string.
9606 'Get' magic is ignored on the sv passed in, but will be called on
9607 C<SvRV(sv)> if sv is an RV.
9613 Perl_sv_2io(pTHX_ SV *const sv)
9618 PERL_ARGS_ASSERT_SV_2IO;
9620 switch (SvTYPE(sv)) {
9622 io = MUTABLE_IO(sv);
9626 if (isGV_with_GP(sv)) {
9627 gv = MUTABLE_GV(sv);
9630 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9631 HEKfARG(GvNAME_HEK(gv)));
9637 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9639 SvGETMAGIC(SvRV(sv));
9640 return sv_2io(SvRV(sv));
9642 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9649 if (SvGMAGICAL(sv)) {
9650 newsv = sv_newmortal();
9651 sv_setsv_nomg(newsv, sv);
9653 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9663 Using various gambits, try to get a CV from an SV; in addition, try if
9664 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9665 The flags in C<lref> are passed to gv_fetchsv.
9671 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9676 PERL_ARGS_ASSERT_SV_2CV;
9683 switch (SvTYPE(sv)) {
9687 return MUTABLE_CV(sv);
9697 sv = amagic_deref_call(sv, to_cv_amg);
9700 if (SvTYPE(sv) == SVt_PVCV) {
9701 cv = MUTABLE_CV(sv);
9706 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9707 gv = MUTABLE_GV(sv);
9709 Perl_croak(aTHX_ "Not a subroutine reference");
9711 else if (isGV_with_GP(sv)) {
9712 gv = MUTABLE_GV(sv);
9715 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9722 /* Some flags to gv_fetchsv mean don't really create the GV */
9723 if (!isGV_with_GP(gv)) {
9728 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9729 /* XXX this is probably not what they think they're getting.
9730 * It has the same effect as "sub name;", i.e. just a forward
9741 Returns true if the SV has a true value by Perl's rules.
9742 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9743 instead use an in-line version.
9749 Perl_sv_true(pTHX_ SV *const sv)
9754 const XPV* const tXpv = (XPV*)SvANY(sv);
9756 (tXpv->xpv_cur > 1 ||
9757 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9764 return SvIVX(sv) != 0;
9767 return SvNVX(sv) != 0.0;
9769 return sv_2bool(sv);
9775 =for apidoc sv_pvn_force
9777 Get a sensible string out of the SV somehow.
9778 A private implementation of the C<SvPV_force> macro for compilers which
9779 can't cope with complex macro expressions. Always use the macro instead.
9781 =for apidoc sv_pvn_force_flags
9783 Get a sensible string out of the SV somehow.
9784 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9785 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9786 implemented in terms of this function.
9787 You normally want to use the various wrapper macros instead: see
9788 C<SvPV_force> and C<SvPV_force_nomg>
9794 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9796 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9798 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9799 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9800 sv_force_normal_flags(sv, 0);
9810 if (SvTYPE(sv) > SVt_PVLV
9811 || isGV_with_GP(sv))
9812 /* diag_listed_as: Can't coerce %s to %s in %s */
9813 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9815 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9822 if (SvTYPE(sv) < SVt_PV ||
9823 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9826 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9827 SvGROW(sv, len + 1);
9828 Move(s,SvPVX(sv),len,char);
9830 SvPVX(sv)[len] = '\0';
9833 SvPOK_on(sv); /* validate pointer */
9835 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9836 PTR2UV(sv),SvPVX_const(sv)));
9839 (void)SvPOK_only_UTF8(sv);
9840 return SvPVX_mutable(sv);
9844 =for apidoc sv_pvbyten_force
9846 The backend for the C<SvPVbytex_force> macro. Always use the macro
9853 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9855 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9857 sv_pvn_force(sv,lp);
9858 sv_utf8_downgrade(sv,0);
9864 =for apidoc sv_pvutf8n_force
9866 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9873 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9875 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9878 sv_utf8_upgrade_nomg(sv);
9884 =for apidoc sv_reftype
9886 Returns a string describing what the SV is a reference to.
9892 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9894 PERL_ARGS_ASSERT_SV_REFTYPE;
9895 if (ob && SvOBJECT(sv)) {
9896 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9899 /* WARNING - There is code, for instance in mg.c, that assumes that
9900 * the only reason that sv_reftype(sv,0) would return a string starting
9901 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9902 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9903 * this routine inside other subs, and it saves time.
9904 * Do not change this assumption without searching for "dodgy type check" in
9907 switch (SvTYPE(sv)) {
9922 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9923 /* tied lvalues should appear to be
9924 * scalars for backwards compatibility */
9925 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9926 ? "SCALAR" : "LVALUE");
9927 case SVt_PVAV: return "ARRAY";
9928 case SVt_PVHV: return "HASH";
9929 case SVt_PVCV: return "CODE";
9930 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9931 ? "GLOB" : "SCALAR");
9932 case SVt_PVFM: return "FORMAT";
9933 case SVt_PVIO: return "IO";
9934 case SVt_INVLIST: return "INVLIST";
9935 case SVt_REGEXP: return "REGEXP";
9936 default: return "UNKNOWN";
9944 Returns a SV describing what the SV passed in is a reference to.
9950 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9952 PERL_ARGS_ASSERT_SV_REF;
9955 dst = sv_newmortal();
9957 if (ob && SvOBJECT(sv)) {
9958 HvNAME_get(SvSTASH(sv))
9959 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9960 : sv_setpvn(dst, "__ANON__", 8);
9963 const char * reftype = sv_reftype(sv, 0);
9964 sv_setpv(dst, reftype);
9970 =for apidoc sv_isobject
9972 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9973 object. If the SV is not an RV, or if the object is not blessed, then this
9980 Perl_sv_isobject(pTHX_ SV *sv)
9996 Returns a boolean indicating whether the SV is blessed into the specified
9997 class. This does not check for subtypes; use C<sv_derived_from> to verify
9998 an inheritance relationship.
10004 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10006 const char *hvname;
10008 PERL_ARGS_ASSERT_SV_ISA;
10018 hvname = HvNAME_get(SvSTASH(sv));
10022 return strEQ(hvname, name);
10026 =for apidoc newSVrv
10028 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10029 RV then it will be upgraded to one. If C<classname> is non-null then the new
10030 SV will be blessed in the specified package. The new SV is returned and its
10031 reference count is 1. The reference count 1 is owned by C<rv>.
10037 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10041 PERL_ARGS_ASSERT_NEWSVRV;
10045 SV_CHECK_THINKFIRST_COW_DROP(rv);
10047 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10048 const U32 refcnt = SvREFCNT(rv);
10052 SvREFCNT(rv) = refcnt;
10054 sv_upgrade(rv, SVt_IV);
10055 } else if (SvROK(rv)) {
10056 SvREFCNT_dec(SvRV(rv));
10058 prepare_SV_for_RV(rv);
10066 HV* const stash = gv_stashpv(classname, GV_ADD);
10067 (void)sv_bless(rv, stash);
10073 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10075 SV * const lv = newSV_type(SVt_PVLV);
10076 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10078 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10079 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10080 LvSTARGOFF(lv) = ix;
10081 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10086 =for apidoc sv_setref_pv
10088 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10089 argument will be upgraded to an RV. That RV will be modified to point to
10090 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10091 into the SV. The C<classname> argument indicates the package for the
10092 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10093 will have a reference count of 1, and the RV will be returned.
10095 Do not use with other Perl types such as HV, AV, SV, CV, because those
10096 objects will become corrupted by the pointer copy process.
10098 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10104 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10106 PERL_ARGS_ASSERT_SV_SETREF_PV;
10109 sv_setsv(rv, &PL_sv_undef);
10113 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10118 =for apidoc sv_setref_iv
10120 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10121 argument will be upgraded to an RV. That RV will be modified to point to
10122 the new SV. The C<classname> argument indicates the package for the
10123 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10124 will have a reference count of 1, and the RV will be returned.
10130 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10132 PERL_ARGS_ASSERT_SV_SETREF_IV;
10134 sv_setiv(newSVrv(rv,classname), iv);
10139 =for apidoc sv_setref_uv
10141 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10142 argument will be upgraded to an RV. That RV will be modified to point to
10143 the new SV. The C<classname> argument indicates the package for the
10144 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10145 will have a reference count of 1, and the RV will be returned.
10151 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10153 PERL_ARGS_ASSERT_SV_SETREF_UV;
10155 sv_setuv(newSVrv(rv,classname), uv);
10160 =for apidoc sv_setref_nv
10162 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10163 argument will be upgraded to an RV. That RV will be modified to point to
10164 the new SV. The C<classname> argument indicates the package for the
10165 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10166 will have a reference count of 1, and the RV will be returned.
10172 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10174 PERL_ARGS_ASSERT_SV_SETREF_NV;
10176 sv_setnv(newSVrv(rv,classname), nv);
10181 =for apidoc sv_setref_pvn
10183 Copies a string into a new SV, optionally blessing the SV. The length of the
10184 string must be specified with C<n>. The C<rv> argument will be upgraded to
10185 an RV. That RV will be modified to point to the new SV. The C<classname>
10186 argument indicates the package for the blessing. Set C<classname> to
10187 C<NULL> to avoid the blessing. The new SV will have a reference count
10188 of 1, and the RV will be returned.
10190 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10196 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10197 const char *const pv, const STRLEN n)
10199 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10201 sv_setpvn(newSVrv(rv,classname), pv, n);
10206 =for apidoc sv_bless
10208 Blesses an SV into a specified package. The SV must be an RV. The package
10209 must be designated by its stash (see C<gv_stashpv()>). The reference count
10210 of the SV is unaffected.
10216 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10219 HV *oldstash = NULL;
10221 PERL_ARGS_ASSERT_SV_BLESS;
10225 Perl_croak(aTHX_ "Can't bless non-reference value");
10227 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10228 if (SvREADONLY(tmpRef))
10229 Perl_croak_no_modify();
10230 if (SvOBJECT(tmpRef)) {
10231 oldstash = SvSTASH(tmpRef);
10234 SvOBJECT_on(tmpRef);
10235 SvUPGRADE(tmpRef, SVt_PVMG);
10236 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10237 SvREFCNT_dec(oldstash);
10239 if(SvSMAGICAL(tmpRef))
10240 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10248 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10249 * as it is after unglobbing it.
10252 PERL_STATIC_INLINE void
10253 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10257 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10259 PERL_ARGS_ASSERT_SV_UNGLOB;
10261 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10263 if (!(flags & SV_COW_DROP_PV))
10264 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10266 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10268 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10269 && HvNAME_get(stash))
10270 mro_method_changed_in(stash);
10271 gp_free(MUTABLE_GV(sv));
10274 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10275 GvSTASH(sv) = NULL;
10278 if (GvNAME_HEK(sv)) {
10279 unshare_hek(GvNAME_HEK(sv));
10281 isGV_with_GP_off(sv);
10283 if(SvTYPE(sv) == SVt_PVGV) {
10284 /* need to keep SvANY(sv) in the right arena */
10285 xpvmg = new_XPVMG();
10286 StructCopy(SvANY(sv), xpvmg, XPVMG);
10287 del_XPVGV(SvANY(sv));
10290 SvFLAGS(sv) &= ~SVTYPEMASK;
10291 SvFLAGS(sv) |= SVt_PVMG;
10294 /* Intentionally not calling any local SET magic, as this isn't so much a
10295 set operation as merely an internal storage change. */
10296 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10297 else sv_setsv_flags(sv, temp, 0);
10299 if ((const GV *)sv == PL_last_in_gv)
10300 PL_last_in_gv = NULL;
10301 else if ((const GV *)sv == PL_statgv)
10306 =for apidoc sv_unref_flags
10308 Unsets the RV status of the SV, and decrements the reference count of
10309 whatever was being referenced by the RV. This can almost be thought of
10310 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10311 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10312 (otherwise the decrementing is conditional on the reference count being
10313 different from one or the reference being a readonly SV).
10320 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10322 SV* const target = SvRV(ref);
10324 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10326 if (SvWEAKREF(ref)) {
10327 sv_del_backref(target, ref);
10328 SvWEAKREF_off(ref);
10329 SvRV_set(ref, NULL);
10332 SvRV_set(ref, NULL);
10334 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10335 assigned to as BEGIN {$a = \"Foo"} will fail. */
10336 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10337 SvREFCNT_dec_NN(target);
10338 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10339 sv_2mortal(target); /* Schedule for freeing later */
10343 =for apidoc sv_untaint
10345 Untaint an SV. Use C<SvTAINTED_off> instead.
10351 Perl_sv_untaint(pTHX_ SV *const sv)
10353 PERL_ARGS_ASSERT_SV_UNTAINT;
10354 PERL_UNUSED_CONTEXT;
10356 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10357 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10364 =for apidoc sv_tainted
10366 Test an SV for taintedness. Use C<SvTAINTED> instead.
10372 Perl_sv_tainted(pTHX_ SV *const sv)
10374 PERL_ARGS_ASSERT_SV_TAINTED;
10375 PERL_UNUSED_CONTEXT;
10377 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10378 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10379 if (mg && (mg->mg_len & 1) )
10386 =for apidoc sv_setpviv
10388 Copies an integer into the given SV, also updating its string value.
10389 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10395 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10397 char buf[TYPE_CHARS(UV)];
10399 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10401 PERL_ARGS_ASSERT_SV_SETPVIV;
10403 sv_setpvn(sv, ptr, ebuf - ptr);
10407 =for apidoc sv_setpviv_mg
10409 Like C<sv_setpviv>, but also handles 'set' magic.
10415 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10417 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10419 sv_setpviv(sv, iv);
10423 #if defined(PERL_IMPLICIT_CONTEXT)
10425 /* pTHX_ magic can't cope with varargs, so this is a no-context
10426 * version of the main function, (which may itself be aliased to us).
10427 * Don't access this version directly.
10431 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10436 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10438 va_start(args, pat);
10439 sv_vsetpvf(sv, pat, &args);
10443 /* pTHX_ magic can't cope with varargs, so this is a no-context
10444 * version of the main function, (which may itself be aliased to us).
10445 * Don't access this version directly.
10449 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10454 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10456 va_start(args, pat);
10457 sv_vsetpvf_mg(sv, pat, &args);
10463 =for apidoc sv_setpvf
10465 Works like C<sv_catpvf> but copies the text into the SV instead of
10466 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10472 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10476 PERL_ARGS_ASSERT_SV_SETPVF;
10478 va_start(args, pat);
10479 sv_vsetpvf(sv, pat, &args);
10484 =for apidoc sv_vsetpvf
10486 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10487 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10489 Usually used via its frontend C<sv_setpvf>.
10495 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10497 PERL_ARGS_ASSERT_SV_VSETPVF;
10499 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10503 =for apidoc sv_setpvf_mg
10505 Like C<sv_setpvf>, but also handles 'set' magic.
10511 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10515 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10517 va_start(args, pat);
10518 sv_vsetpvf_mg(sv, pat, &args);
10523 =for apidoc sv_vsetpvf_mg
10525 Like C<sv_vsetpvf>, but also handles 'set' magic.
10527 Usually used via its frontend C<sv_setpvf_mg>.
10533 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10535 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10537 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10541 #if defined(PERL_IMPLICIT_CONTEXT)
10543 /* pTHX_ magic can't cope with varargs, so this is a no-context
10544 * version of the main function, (which may itself be aliased to us).
10545 * Don't access this version directly.
10549 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10554 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10556 va_start(args, pat);
10557 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10561 /* pTHX_ magic can't cope with varargs, so this is a no-context
10562 * version of the main function, (which may itself be aliased to us).
10563 * Don't access this version directly.
10567 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10572 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10574 va_start(args, pat);
10575 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10582 =for apidoc sv_catpvf
10584 Processes its arguments like C<sprintf> and appends the formatted
10585 output to an SV. If the appended data contains "wide" characters
10586 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10587 and characters >255 formatted with %c), the original SV might get
10588 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10589 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10590 valid UTF-8; if the original SV was bytes, the pattern should be too.
10595 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10599 PERL_ARGS_ASSERT_SV_CATPVF;
10601 va_start(args, pat);
10602 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10607 =for apidoc sv_vcatpvf
10609 Processes its arguments like C<vsprintf> and appends the formatted output
10610 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10612 Usually used via its frontend C<sv_catpvf>.
10618 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10620 PERL_ARGS_ASSERT_SV_VCATPVF;
10622 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10626 =for apidoc sv_catpvf_mg
10628 Like C<sv_catpvf>, but also handles 'set' magic.
10634 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10638 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10640 va_start(args, pat);
10641 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10647 =for apidoc sv_vcatpvf_mg
10649 Like C<sv_vcatpvf>, but also handles 'set' magic.
10651 Usually used via its frontend C<sv_catpvf_mg>.
10657 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10659 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10661 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10666 =for apidoc sv_vsetpvfn
10668 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10671 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10677 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10678 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10680 PERL_ARGS_ASSERT_SV_VSETPVFN;
10683 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10688 * Warn of missing argument to sprintf, and then return a defined value
10689 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10692 S_vcatpvfn_missing_argument(pTHX) {
10693 if (ckWARN(WARN_MISSING)) {
10694 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10695 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10702 S_expect_number(pTHX_ char **const pattern)
10706 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10708 switch (**pattern) {
10709 case '1': case '2': case '3':
10710 case '4': case '5': case '6':
10711 case '7': case '8': case '9':
10712 var = *(*pattern)++ - '0';
10713 while (isDIGIT(**pattern)) {
10714 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10716 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10724 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10726 const int neg = nv < 0;
10729 PERL_ARGS_ASSERT_F0CONVERT;
10731 if (UNLIKELY(Perl_isinfnan(nv))) {
10732 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len);
10742 if (uv & 1 && uv == nv)
10743 uv--; /* Round to even */
10745 const unsigned dig = uv % 10;
10747 } while (uv /= 10);
10758 =for apidoc sv_vcatpvfn
10760 =for apidoc sv_vcatpvfn_flags
10762 Processes its arguments like C<vsprintf> and appends the formatted output
10763 to an SV. Uses an array of SVs if the C style variable argument list is
10764 missing (NULL). When running with taint checks enabled, indicates via
10765 C<maybe_tainted> if results are untrustworthy (often due to the use of
10768 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10770 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10775 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10776 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10777 vec_utf8 = DO_UTF8(vecsv);
10779 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10782 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10783 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10785 PERL_ARGS_ASSERT_SV_VCATPVFN;
10787 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10790 #if DOUBLEKIND == DOUBLE_IS_IEEE_754_32_BIT_LITTLE_ENDIAN || \
10791 DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_LITTLE_ENDIAN || \
10792 DOUBLEKIND == DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10793 # define DOUBLE_LITTLE_ENDIAN
10796 #ifdef HAS_LONG_DOUBLEKIND
10798 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10799 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10800 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10801 # define LONGDOUBLE_LITTLE_ENDIAN
10804 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN || \
10805 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN || \
10806 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10807 # define LONGDOUBLE_BIG_ENDIAN
10810 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10811 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10812 # define LONGDOUBLE_X86_80_BIT
10815 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
10816 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10817 # define LONGDOUBLE_DOUBLEDOUBLE
10818 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10819 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10820 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10821 * after the first 1023 zero bits.
10823 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10824 * of dynamically growing buffer might be better, start at just 16 bytes
10825 * (for example) and grow only when necessary. Or maybe just by looking
10826 * at the exponents of the two doubles? */
10827 # define DOUBLEDOUBLE_MAXBITS 2098
10830 #endif /* HAS_LONG_DOUBLE */
10832 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10833 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10834 * per xdigit. For the double-double case, this can be rather many.
10835 * The non-double-double-long-double overshoots since all bits of NV
10836 * are not mantissa bits, there are also exponent bits. */
10837 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10838 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10840 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10843 /* If we do not have a known long double format, (including not using
10844 * long doubles, or long doubles being equal to doubles) then we will
10845 * fall back to the ldexp/frexp route, with which we can retrieve at
10846 * most as many bits as our widest unsigned integer type is. We try
10847 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10849 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10850 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10852 #if defined(HAS_QUAD) && defined(Uquad_t)
10853 # define MANTISSATYPE Uquad_t
10854 # define MANTISSASIZE 8
10856 # define MANTISSATYPE UV
10857 # define MANTISSASIZE UVSIZE
10860 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10861 # define HEXTRACT_LITTLE_ENDIAN
10862 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10863 # define HEXTRACT_BIG_ENDIAN
10865 # define HEXTRACT_MIX_ENDIAN
10868 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10869 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10870 * are being extracted from (either directly from the long double in-memory
10871 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10872 * is used to update the exponent. vhex is the pointer to the beginning
10873 * of the output buffer (of VHEX_SIZE).
10875 * The tricky part is that S_hextract() needs to be called twice:
10876 * the first time with vend as NULL, and the second time with vend as
10877 * the pointer returned by the first call. What happens is that on
10878 * the first round the output size is computed, and the intended
10879 * extraction sanity checked. On the second round the actual output
10880 * (the extraction of the hexadecimal values) takes place.
10881 * Sanity failures cause fatal failures during both rounds. */
10883 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10887 int ixmin = 0, ixmax = 0;
10889 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10890 * and elsewhere. */
10892 /* These macros are just to reduce typos, they have multiple
10893 * repetitions below, but usually only one (or sometimes two)
10894 * of them is really being used. */
10895 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10896 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10897 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10898 #define HEXTRACT_OUTPUT(ix) \
10900 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10902 #define HEXTRACT_COUNT(ix, c) \
10904 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10906 #define HEXTRACT_BYTE(ix) \
10908 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10910 #define HEXTRACT_LO_NYBBLE(ix) \
10912 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10914 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10915 * to make it look less odd when the top bits of a NV
10916 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10917 * order bits can be in the "low nybble" of a byte. */
10918 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10919 #define HEXTRACT_BYTES_LE(a, b) \
10920 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10921 #define HEXTRACT_BYTES_BE(a, b) \
10922 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10923 #define HEXTRACT_IMPLICIT_BIT(nv) \
10925 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10928 /* Most formats do. Those which don't should undef this. */
10929 #define HEXTRACT_HAS_IMPLICIT_BIT
10930 /* Many formats do. Those which don't should undef this. */
10931 #define HEXTRACT_HAS_TOP_NYBBLE
10933 /* HEXTRACTSIZE is the maximum number of xdigits. */
10934 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10935 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/4)
10937 # define HEXTRACTSIZE 2 * NVSIZE
10940 const U8* vmaxend = vhex + HEXTRACTSIZE;
10941 PERL_UNUSED_VAR(ix); /* might happen */
10942 (void)Perl_frexp(PERL_ABS(nv), exponent);
10943 if (vend && (vend <= vhex || vend > vmaxend))
10944 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10946 /* First check if using long doubles. */
10947 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10948 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10949 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10950 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10951 /* The bytes 13..0 are the mantissa/fraction,
10952 * the 15,14 are the sign+exponent. */
10953 const U8* nvp = (const U8*)(&nv);
10954 HEXTRACT_IMPLICIT_BIT(nv);
10955 # undef HEXTRACT_HAS_TOP_NYBBLE
10956 HEXTRACT_BYTES_LE(13, 0);
10957 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10958 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10959 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10960 /* The bytes 2..15 are the mantissa/fraction,
10961 * the 0,1 are the sign+exponent. */
10962 const U8* nvp = (const U8*)(&nv);
10963 HEXTRACT_IMPLICIT_BIT(nv);
10964 # undef HEXTRACT_HAS_TOP_NYBBLE
10965 HEXTRACT_BYTES_BE(2, 15);
10966 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10967 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10968 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10969 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10970 * meaning that 2 or 6 bytes are empty padding. */
10971 /* The bytes 7..0 are the mantissa/fraction */
10972 const U8* nvp = (const U8*)(&nv);
10973 # undef HEXTRACT_HAS_IMPLICIT_BIT
10974 # undef HEXTRACT_HAS_TOP_NYBBLE
10975 HEXTRACT_BYTES_LE(7, 0);
10976 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10977 /* Does this format ever happen? (Wikipedia says the Motorola
10978 * 6888x math coprocessors used format _like_ this but padded
10979 * to 96 bits with 16 unused bits between the exponent and the
10981 const U8* nvp = (const U8*)(&nv);
10982 # undef HEXTRACT_HAS_IMPLICIT_BIT
10983 # undef HEXTRACT_HAS_TOP_NYBBLE
10984 HEXTRACT_BYTES_BE(0, 7);
10986 # define HEXTRACT_FALLBACK
10987 /* Double-double format: two doubles next to each other.
10988 * The first double is the high-order one, exactly like
10989 * it would be for a "lone" double. The second double
10990 * is shifted down using the exponent so that that there
10991 * are no common bits. The tricky part is that the value
10992 * of the double-double is the SUM of the two doubles and
10993 * the second one can be also NEGATIVE.
10995 * Because of this tricky construction the bytewise extraction we
10996 * use for the other long double formats doesn't work, we must
10997 * extract the values bit by bit.
10999 * The little-endian double-double is used .. somewhere?
11001 * The big endian double-double is used in e.g. PPC/Power (AIX)
11004 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11005 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11006 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11009 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11010 /* Using normal doubles, not long doubles.
11012 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11013 * bytes, since we might need to handle printf precision, and
11014 * also need to insert the radix. */
11016 # ifdef HEXTRACT_LITTLE_ENDIAN
11017 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11018 const U8* nvp = (const U8*)(&nv);
11019 HEXTRACT_IMPLICIT_BIT(nv);
11020 HEXTRACT_TOP_NYBBLE(6);
11021 HEXTRACT_BYTES_LE(5, 0);
11022 # elif defined(HEXTRACT_BIG_ENDIAN)
11023 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11024 const U8* nvp = (const U8*)(&nv);
11025 HEXTRACT_IMPLICIT_BIT(nv);
11026 HEXTRACT_TOP_NYBBLE(1);
11027 HEXTRACT_BYTES_BE(2, 7);
11028 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11029 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11030 const U8* nvp = (const U8*)(&nv);
11031 HEXTRACT_IMPLICIT_BIT(nv);
11032 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11033 HEXTRACT_BYTE(1); /* 5 */
11034 HEXTRACT_BYTE(0); /* 4 */
11035 HEXTRACT_BYTE(7); /* 3 */
11036 HEXTRACT_BYTE(6); /* 2 */
11037 HEXTRACT_BYTE(5); /* 1 */
11038 HEXTRACT_BYTE(4); /* 0 */
11039 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11040 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11041 const U8* nvp = (const U8*)(&nv);
11042 HEXTRACT_IMPLICIT_BIT(nv);
11043 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11044 HEXTRACT_BYTE(6); /* 5 */
11045 HEXTRACT_BYTE(7); /* 4 */
11046 HEXTRACT_BYTE(0); /* 3 */
11047 HEXTRACT_BYTE(1); /* 2 */
11048 HEXTRACT_BYTE(2); /* 1 */
11049 HEXTRACT_BYTE(3); /* 0 */
11051 # define HEXTRACT_FALLBACK
11054 # define HEXTRACT_FALLBACK
11056 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11057 # ifdef HEXTRACT_FALLBACK
11058 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11059 /* The fallback is used for the double-double format, and
11060 * for unknown long double formats, and for unknown double
11061 * formats, or in general unknown NV formats. */
11062 if (nv == (NV)0.0) {
11070 NV d = nv < 0 ? -nv : nv;
11072 U8 ha = 0x0; /* hexvalue accumulator */
11073 U8 hd = 0x8; /* hexvalue digit */
11075 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11076 * this is essentially manual frexp(). Multiplying by 0.5 and
11077 * doubling should be lossless in binary floating point. */
11087 while (d >= e + e) {
11091 /* Now e <= d < 2*e */
11093 /* First extract the leading hexdigit (the implicit bit). */
11109 /* Then extract the remaining hexdigits. */
11110 while (d > (NV)0.0) {
11116 /* Output or count in groups of four bits,
11117 * that is, when the hexdigit is down to one. */
11122 /* Reset the hexvalue. */
11131 /* Flush possible pending hexvalue. */
11141 /* Croak for various reasons: if the output pointer escaped the
11142 * output buffer, if the extraction index escaped the extraction
11143 * buffer, or if the ending output pointer didn't match the
11144 * previously computed value. */
11145 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11146 /* For double-double the ixmin and ixmax stay at zero,
11147 * which is convenient since the HEXTRACTSIZE is tricky
11148 * for double-double. */
11149 ixmin < 0 || ixmax >= NVSIZE ||
11150 (vend && v != vend))
11151 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11156 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11157 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11162 const char *patend;
11165 static const char nullstr[] = "(null)";
11167 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11168 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11170 /* Times 4: a decimal digit takes more than 3 binary digits.
11171 * NV_DIG: mantissa takes than many decimal digits.
11172 * Plus 32: Playing safe. */
11173 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11174 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11175 bool hexfp = FALSE; /* hexadecimal floating point? */
11177 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
11179 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11180 PERL_UNUSED_ARG(maybe_tainted);
11182 if (flags & SV_GMAGIC)
11185 /* no matter what, this is a string now */
11186 (void)SvPV_force_nomg(sv, origlen);
11188 /* special-case "", "%s", and "%-p" (SVf - see below) */
11190 if (svmax && ckWARN(WARN_REDUNDANT))
11191 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11192 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11195 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11196 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11197 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11198 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11201 const char * const s = va_arg(*args, char*);
11202 sv_catpv_nomg(sv, s ? s : nullstr);
11204 else if (svix < svmax) {
11205 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11206 SvGETMAGIC(*svargs);
11207 sv_catsv_nomg(sv, *svargs);
11210 S_vcatpvfn_missing_argument(aTHX);
11213 if (args && patlen == 3 && pat[0] == '%' &&
11214 pat[1] == '-' && pat[2] == 'p') {
11215 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11216 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11217 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11218 argsv = MUTABLE_SV(va_arg(*args, void*));
11219 sv_catsv_nomg(sv, argsv);
11223 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11224 /* special-case "%.<number>[gf]" */
11225 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11226 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11227 unsigned digits = 0;
11231 while (*pp >= '0' && *pp <= '9')
11232 digits = 10 * digits + (*pp++ - '0');
11234 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11235 format the first argument and WARN_REDUNDANT if svmax > 1?
11236 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11237 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11238 const NV nv = SvNV(*svargs);
11239 if (LIKELY(!Perl_isinfnan(nv))) {
11241 /* Add check for digits != 0 because it seems that some
11242 gconverts are buggy in this case, and we don't yet have
11243 a Configure test for this. */
11244 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11245 /* 0, point, slack */
11246 STORE_LC_NUMERIC_SET_TO_NEEDED();
11247 SNPRINTF_G(nv, ebuf, size, digits);
11248 sv_catpv_nomg(sv, ebuf);
11249 if (*ebuf) /* May return an empty string for digits==0 */
11252 } else if (!digits) {
11255 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11256 sv_catpvn_nomg(sv, p, l);
11263 #endif /* !USE_LONG_DOUBLE */
11265 if (!args && svix < svmax && DO_UTF8(*svargs))
11268 patend = (char*)pat + patlen;
11269 for (p = (char*)pat; p < patend; p = q) {
11272 bool vectorize = FALSE;
11273 bool vectorarg = FALSE;
11274 bool vec_utf8 = FALSE;
11280 bool has_precis = FALSE;
11282 const I32 osvix = svix;
11283 bool is_utf8 = FALSE; /* is this item utf8? */
11284 #ifdef HAS_LDBL_SPRINTF_BUG
11285 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11286 with sfio - Allen <allens@cpan.org> */
11287 bool fix_ldbl_sprintf_bug = FALSE;
11291 U8 utf8buf[UTF8_MAXBYTES+1];
11292 STRLEN esignlen = 0;
11294 const char *eptr = NULL;
11295 const char *fmtstart;
11298 const U8 *vecstr = NULL;
11305 /* We need a long double target in case HAS_LONG_DOUBLE,
11306 * even without USE_LONG_DOUBLE, so that we can printf with
11307 * long double formats, even without NV being long double.
11308 * But we call the target 'fv' instead of 'nv', since most of
11309 * the time it is not (most compilers these days recognize
11310 * "long double", even if only as a synonym for "double").
11312 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11313 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11315 # ifdef Perl_isfinitel
11316 # define FV_ISFINITE(x) Perl_isfinitel(x)
11318 # define FV_GF PERL_PRIgldbl
11319 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11320 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11321 # define NV_TO_FV(nv,fv) STMT_START { \
11323 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11326 # define NV_TO_FV(nv,fv) (fv)=(nv)
11330 # define FV_GF NVgf
11331 # define NV_TO_FV(nv,fv) (fv)=(nv)
11333 #ifndef FV_ISFINITE
11334 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11339 const char *dotstr = ".";
11340 STRLEN dotstrlen = 1;
11341 I32 efix = 0; /* explicit format parameter index */
11342 I32 ewix = 0; /* explicit width index */
11343 I32 epix = 0; /* explicit precision index */
11344 I32 evix = 0; /* explicit vector index */
11345 bool asterisk = FALSE;
11346 bool infnan = FALSE;
11348 /* echo everything up to the next format specification */
11349 for (q = p; q < patend && *q != '%'; ++q) ;
11351 if (has_utf8 && !pat_utf8)
11352 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11354 sv_catpvn_nomg(sv, p, q - p);
11363 We allow format specification elements in this order:
11364 \d+\$ explicit format parameter index
11366 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11367 0 flag (as above): repeated to allow "v02"
11368 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11369 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11371 [%bcdefginopsuxDFOUX] format (mandatory)
11376 As of perl5.9.3, printf format checking is on by default.
11377 Internally, perl uses %p formats to provide an escape to
11378 some extended formatting. This block deals with those
11379 extensions: if it does not match, (char*)q is reset and
11380 the normal format processing code is used.
11382 Currently defined extensions are:
11383 %p include pointer address (standard)
11384 %-p (SVf) include an SV (previously %_)
11385 %-<num>p include an SV with precision <num>
11387 %3p include a HEK with precision of 256
11388 %4p char* preceded by utf8 flag and length
11389 %<num>p (where num is 1 or > 4) reserved for future
11392 Robin Barker 2005-07-14 (but modified since)
11394 %1p (VDf) removed. RMB 2007-10-19
11401 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11402 /* The argument has already gone through cBOOL, so the cast
11404 is_utf8 = (bool)va_arg(*args, int);
11405 elen = va_arg(*args, UV);
11406 if ((IV)elen < 0) {
11407 /* check if utf8 length is larger than 0 when cast to IV */
11408 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11409 elen= 0; /* otherwise we want to treat this as an empty string */
11411 eptr = va_arg(*args, char *);
11412 q += sizeof(UTF8f)-1;
11415 n = expect_number(&q);
11417 if (sv) { /* SVf */
11422 argsv = MUTABLE_SV(va_arg(*args, void*));
11423 eptr = SvPV_const(argsv, elen);
11424 if (DO_UTF8(argsv))
11428 else if (n==2 || n==3) { /* HEKf */
11429 HEK * const hek = va_arg(*args, HEK *);
11430 eptr = HEK_KEY(hek);
11431 elen = HEK_LEN(hek);
11432 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11433 if (n==3) precis = 256, has_precis = TRUE;
11437 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11438 "internal %%<num>p might conflict with future printf extensions");
11444 if ( (width = expect_number(&q)) ) {
11448 if (!no_redundant_warning)
11449 /* I've forgotten if it's a better
11450 micro-optimization to always set this or to
11451 only set it if it's unset */
11452 no_redundant_warning = TRUE;
11464 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11493 if ( (ewix = expect_number(&q)) )
11502 if ((vectorarg = asterisk)) {
11515 width = expect_number(&q);
11518 if (vectorize && vectorarg) {
11519 /* vectorizing, but not with the default "." */
11521 vecsv = va_arg(*args, SV*);
11523 vecsv = (evix > 0 && evix <= svmax)
11524 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11526 vecsv = svix < svmax
11527 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11529 dotstr = SvPV_const(vecsv, dotstrlen);
11530 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11531 bad with tied or overloaded values that return UTF8. */
11532 if (DO_UTF8(vecsv))
11534 else if (has_utf8) {
11535 vecsv = sv_mortalcopy(vecsv);
11536 sv_utf8_upgrade(vecsv);
11537 dotstr = SvPV_const(vecsv, dotstrlen);
11544 i = va_arg(*args, int);
11546 i = (ewix ? ewix <= svmax : svix < svmax) ?
11547 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11549 width = (i < 0) ? -i : i;
11559 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11561 /* XXX: todo, support specified precision parameter */
11565 i = va_arg(*args, int);
11567 i = (ewix ? ewix <= svmax : svix < svmax)
11568 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11570 has_precis = !(i < 0);
11574 while (isDIGIT(*q))
11575 precis = precis * 10 + (*q++ - '0');
11584 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11585 vecsv = svargs[efix ? efix-1 : svix++];
11586 vecstr = (U8*)SvPV_const(vecsv,veclen);
11587 vec_utf8 = DO_UTF8(vecsv);
11589 /* if this is a version object, we need to convert
11590 * back into v-string notation and then let the
11591 * vectorize happen normally
11593 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11594 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11595 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11596 "vector argument not supported with alpha versions");
11599 vecsv = sv_newmortal();
11600 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11602 vecstr = (U8*)SvPV_const(vecsv, veclen);
11603 vec_utf8 = DO_UTF8(vecsv);
11617 case 'I': /* Ix, I32x, and I64x */
11618 # ifdef USE_64_BIT_INT
11619 if (q[1] == '6' && q[2] == '4') {
11625 if (q[1] == '3' && q[2] == '2') {
11629 # ifdef USE_64_BIT_INT
11635 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11636 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11639 # ifdef USE_QUADMATH
11652 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11653 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11654 if (*q == 'l') { /* lld, llf */
11663 if (*++q == 'h') { /* hhd, hhu */
11692 if (!vectorize && !args) {
11694 const I32 i = efix-1;
11695 argsv = (i >= 0 && i < svmax)
11696 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11698 argsv = (svix >= 0 && svix < svmax)
11699 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11703 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11704 /* XXX va_arg(*args) case? need peek, use va_copy? */
11706 infnan = UNLIKELY(isinfnansv(argsv));
11709 switch (c = *q++) {
11717 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11718 /* no va_arg() case */
11719 SvNV_nomg(argsv), (int)c);
11720 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11722 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11724 eptr = (char*)utf8buf;
11725 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11739 eptr = va_arg(*args, char*);
11741 elen = strlen(eptr);
11743 eptr = (char *)nullstr;
11744 elen = sizeof nullstr - 1;
11748 eptr = SvPV_const(argsv, elen);
11749 if (DO_UTF8(argsv)) {
11750 STRLEN old_precis = precis;
11751 if (has_precis && precis < elen) {
11752 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11753 STRLEN p = precis > ulen ? ulen : precis;
11754 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11755 /* sticks at end */
11757 if (width) { /* fudge width (can't fudge elen) */
11758 if (has_precis && precis < elen)
11759 width += precis - old_precis;
11762 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11769 if (has_precis && precis < elen)
11777 goto floating_point;
11779 if (alt || vectorize)
11781 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11795 goto floating_point;
11802 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11811 esignbuf[esignlen++] = plus;
11815 case 'c': iv = (char)va_arg(*args, int); break;
11816 case 'h': iv = (short)va_arg(*args, int); break;
11817 case 'l': iv = va_arg(*args, long); break;
11818 case 'V': iv = va_arg(*args, IV); break;
11819 case 'z': iv = va_arg(*args, SSize_t); break;
11820 #ifdef HAS_PTRDIFF_T
11821 case 't': iv = va_arg(*args, ptrdiff_t); break;
11823 default: iv = va_arg(*args, int); break;
11825 case 'j': iv = va_arg(*args, intmax_t); break;
11829 iv = va_arg(*args, Quad_t); break;
11836 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11838 case 'c': iv = (char)tiv; break;
11839 case 'h': iv = (short)tiv; break;
11840 case 'l': iv = (long)tiv; break;
11842 default: iv = tiv; break;
11845 iv = (Quad_t)tiv; break;
11851 if ( !vectorize ) /* we already set uv above */
11856 esignbuf[esignlen++] = plus;
11860 esignbuf[esignlen++] = '-';
11899 goto floating_point;
11907 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11918 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11919 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11920 case 'l': uv = va_arg(*args, unsigned long); break;
11921 case 'V': uv = va_arg(*args, UV); break;
11922 case 'z': uv = va_arg(*args, Size_t); break;
11923 #ifdef HAS_PTRDIFF_T
11924 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11927 case 'j': uv = va_arg(*args, uintmax_t); break;
11929 default: uv = va_arg(*args, unsigned); break;
11932 uv = va_arg(*args, Uquad_t); break;
11939 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11941 case 'c': uv = (unsigned char)tuv; break;
11942 case 'h': uv = (unsigned short)tuv; break;
11943 case 'l': uv = (unsigned long)tuv; break;
11945 default: uv = tuv; break;
11948 uv = (Uquad_t)tuv; break;
11957 char *ptr = ebuf + sizeof ebuf;
11958 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11964 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11968 } while (uv >>= 4);
11970 esignbuf[esignlen++] = '0';
11971 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11977 *--ptr = '0' + dig;
11978 } while (uv >>= 3);
11979 if (alt && *ptr != '0')
11985 *--ptr = '0' + dig;
11986 } while (uv >>= 1);
11988 esignbuf[esignlen++] = '0';
11989 esignbuf[esignlen++] = c;
11992 default: /* it had better be ten or less */
11995 *--ptr = '0' + dig;
11996 } while (uv /= base);
11999 elen = (ebuf + sizeof ebuf) - ptr;
12003 zeros = precis - elen;
12004 else if (precis == 0 && elen == 1 && *eptr == '0'
12005 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12008 /* a precision nullifies the 0 flag. */
12015 /* FLOATING POINT */
12020 c = 'f'; /* maybe %F isn't supported here */
12022 case 'e': case 'E':
12024 case 'g': case 'G':
12025 case 'a': case 'A':
12029 /* This is evil, but floating point is even more evil */
12031 /* for SV-style calling, we can only get NV
12032 for C-style calling, we assume %f is double;
12033 for simplicity we allow any of %Lf, %llf, %qf for long double
12037 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12041 /* [perl #20339] - we should accept and ignore %lf rather than die */
12045 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12046 intsize = args ? 0 : 'q';
12050 #if defined(HAS_LONG_DOUBLE)
12063 /* Now we need (long double) if intsize == 'q', else (double). */
12065 /* Note: do not pull NVs off the va_list with va_arg()
12066 * (pull doubles instead) because if you have a build
12067 * with long doubles, you would always be pulling long
12068 * doubles, which would badly break anyone using only
12069 * doubles (i.e. the majority of builds). In other
12070 * words, you cannot mix doubles and long doubles.
12071 * The only case where you can pull off long doubles
12072 * is when the format specifier explicitly asks so with
12074 #ifdef USE_QUADMATH
12075 fv = intsize == 'q' ?
12076 va_arg(*args, NV) : va_arg(*args, double);
12077 #elif LONG_DOUBLESIZE > DOUBLESIZE
12078 if (intsize == 'q')
12079 fv = va_arg(*args, long double);
12081 NV_TO_FV(va_arg(*args, double), fv);
12083 fv = va_arg(*args, double);
12088 if (!infnan) SvGETMAGIC(argsv);
12089 NV_TO_FV(SvNV_nomg(argsv), fv);
12093 /* frexp() (or frexpl) has some unspecified behaviour for
12094 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12095 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12097 (void)Perl_frexp((NV)fv, &i);
12098 if (i == PERL_INT_MIN)
12099 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12100 /* Do not set hexfp earlier since we want to printf
12101 * Inf/NaN for Inf/NaN, not their hexfp. */
12102 hexfp = isALPHA_FOLD_EQ(c, 'a');
12103 if (UNLIKELY(hexfp)) {
12104 /* This seriously overshoots in most cases, but
12105 * better the undershooting. Firstly, all bytes
12106 * of the NV are not mantissa, some of them are
12107 * exponent. Secondly, for the reasonably common
12108 * long doubles case, the "80-bit extended", two
12109 * or six bytes of the NV are unused. */
12111 (fv < 0) ? 1 : 0 + /* possible unary minus */
12113 1 + /* the very unlikely carry */
12116 2 * NVSIZE + /* 2 hexdigits for each byte */
12118 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12120 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12121 /* However, for the "double double", we need more.
12122 * Since each double has their own exponent, the
12123 * doubles may float (haha) rather far from each
12124 * other, and the number of required bits is much
12125 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12126 * See the definition of DOUBLEDOUBLE_MAXBITS.
12128 * Need 2 hexdigits for each byte. */
12129 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12130 /* the size for the exponent already added */
12132 #ifdef USE_LOCALE_NUMERIC
12133 STORE_LC_NUMERIC_SET_TO_NEEDED();
12134 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12135 need += SvLEN(PL_numeric_radix_sv);
12136 RESTORE_LC_NUMERIC();
12140 need = BIT_DIGITS(i);
12141 } /* if i < 0, the number of digits is hard to predict. */
12143 need += has_precis ? precis : 6; /* known default */
12148 #ifdef HAS_LDBL_SPRINTF_BUG
12149 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12150 with sfio - Allen <allens@cpan.org> */
12153 # define MY_DBL_MAX DBL_MAX
12154 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12155 # if DOUBLESIZE >= 8
12156 # define MY_DBL_MAX 1.7976931348623157E+308L
12158 # define MY_DBL_MAX 3.40282347E+38L
12162 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12163 # define MY_DBL_MAX_BUG 1L
12165 # define MY_DBL_MAX_BUG MY_DBL_MAX
12169 # define MY_DBL_MIN DBL_MIN
12170 # else /* XXX guessing! -Allen */
12171 # if DOUBLESIZE >= 8
12172 # define MY_DBL_MIN 2.2250738585072014E-308L
12174 # define MY_DBL_MIN 1.17549435E-38L
12178 if ((intsize == 'q') && (c == 'f') &&
12179 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12180 (need < DBL_DIG)) {
12181 /* it's going to be short enough that
12182 * long double precision is not needed */
12184 if ((fv <= 0L) && (fv >= -0L))
12185 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12187 /* would use Perl_fp_class as a double-check but not
12188 * functional on IRIX - see perl.h comments */
12190 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12191 /* It's within the range that a double can represent */
12192 #if defined(DBL_MAX) && !defined(DBL_MIN)
12193 if ((fv >= ((long double)1/DBL_MAX)) ||
12194 (fv <= (-(long double)1/DBL_MAX)))
12196 fix_ldbl_sprintf_bug = TRUE;
12199 if (fix_ldbl_sprintf_bug == TRUE) {
12209 # undef MY_DBL_MAX_BUG
12212 #endif /* HAS_LDBL_SPRINTF_BUG */
12214 need += 20; /* fudge factor */
12215 if (PL_efloatsize < need) {
12216 Safefree(PL_efloatbuf);
12217 PL_efloatsize = need + 20; /* more fudge */
12218 Newx(PL_efloatbuf, PL_efloatsize, char);
12219 PL_efloatbuf[0] = '\0';
12222 if ( !(width || left || plus || alt) && fill != '0'
12223 && has_precis && intsize != 'q' /* Shortcuts */
12224 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12225 /* See earlier comment about buggy Gconvert when digits,
12227 if ( c == 'g' && precis ) {
12228 STORE_LC_NUMERIC_SET_TO_NEEDED();
12229 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12230 /* May return an empty string for digits==0 */
12231 if (*PL_efloatbuf) {
12232 elen = strlen(PL_efloatbuf);
12233 goto float_converted;
12235 } else if ( c == 'f' && !precis ) {
12236 if ((eptr = F0convert(fv, ebuf + sizeof ebuf, &elen)))
12241 if (UNLIKELY(hexfp)) {
12242 /* Hexadecimal floating point. */
12243 char* p = PL_efloatbuf;
12244 U8 vhex[VHEX_SIZE];
12245 U8* v = vhex; /* working pointer to vhex */
12246 U8* vend; /* pointer to one beyond last digit of vhex */
12247 U8* vfnz = NULL; /* first non-zero */
12248 const bool lower = (c == 'a');
12249 /* At output the values of vhex (up to vend) will
12250 * be mapped through the xdig to get the actual
12251 * human-readable xdigits. */
12252 const char* xdig = PL_hexdigit;
12253 int zerotail = 0; /* how many extra zeros to append */
12254 int exponent = 0; /* exponent of the floating point input */
12256 /* XXX: denormals, NaN, Inf.
12258 * For example with denormals, (assuming the vanilla
12259 * 64-bit double): the exponent is zero. 1xp-1074 is
12260 * the smallest denormal and the smallest double, it
12261 * should be output as 0x0.0000000000001p-1022 to
12262 * match its internal structure. */
12264 /* Note: fv can be (and often is) long double.
12265 * Here it is explicitly cast to NV. */
12266 vend = S_hextract(aTHX_ (NV)fv, &exponent, vhex, NULL);
12267 S_hextract(aTHX_ (NV)fv, &exponent, vhex, vend);
12269 #if NVSIZE > DOUBLESIZE
12270 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12271 /* In this case there is an implicit bit,
12272 * and therefore the exponent is shifted shift by one. */
12275 /* In this case there is no implicit bit,
12276 * and the exponent is shifted by the first xdigit. */
12291 xdig += 16; /* Use uppercase hex. */
12294 /* Find the first non-zero xdigit. */
12295 for (v = vhex; v < vend; v++) {
12303 U8* vlnz = NULL; /* The last non-zero. */
12305 /* Find the last non-zero xdigit. */
12306 for (v = vend - 1; v >= vhex; v--) {
12313 #if NVSIZE == DOUBLESIZE
12319 v = vhex + precis + 1;
12321 /* Round away from zero: if the tail
12322 * beyond the precis xdigits is equal to
12323 * or greater than 0x8000... */
12324 bool round = *v > 0x8;
12325 if (!round && *v == 0x8) {
12326 for (v++; v < vend; v++) {
12334 for (v = vhex + precis; v >= vhex; v--) {
12341 /* If the carry goes all the way to
12342 * the front, we need to output
12343 * a single '1'. This goes against
12344 * the "xdigit and then radix"
12345 * but since this is "cannot happen"
12346 * category, that is probably good. */
12351 /* The new effective "last non zero". */
12352 vlnz = vhex + precis;
12355 zerotail = precis - (vlnz - vhex);
12362 /* The radix is always output after the first
12363 * non-zero xdigit, or if alt. */
12364 if (vfnz < vlnz || alt) {
12365 #ifndef USE_LOCALE_NUMERIC
12368 STORE_LC_NUMERIC_SET_TO_NEEDED();
12369 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12371 const char* r = SvPV(PL_numeric_radix_sv, n);
12372 Copy(r, p, n, char);
12378 RESTORE_LC_NUMERIC();
12393 elen = p - PL_efloatbuf;
12394 elen += my_snprintf(p, PL_efloatsize - elen,
12395 "%c%+d", lower ? 'p' : 'P',
12398 if (elen < width) {
12400 /* Pad the back with spaces. */
12401 memset(PL_efloatbuf + elen, ' ', width - elen);
12403 else if (fill == '0') {
12404 /* Insert the zeros between the "0x" and
12405 * the digits, otherwise we end up with
12407 STRLEN nzero = width - elen;
12408 char* zerox = PL_efloatbuf + 2;
12409 Move(zerox, zerox + nzero, elen - 2, char);
12410 memset(zerox, fill, nzero);
12413 /* Move it to the right. */
12414 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12416 /* Pad the front with spaces. */
12417 memset(PL_efloatbuf, ' ', width - elen);
12423 elen = S_infnan_2pv(fv, PL_efloatbuf, PL_efloatsize);
12426 char *ptr = ebuf + sizeof ebuf;
12429 #if defined(USE_QUADMATH)
12430 if (intsize == 'q') {
12434 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12435 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12436 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12437 * not USE_LONG_DOUBLE and NVff. In other words,
12438 * this needs to work without USE_LONG_DOUBLE. */
12439 if (intsize == 'q') {
12440 /* Copy the one or more characters in a long double
12441 * format before the 'base' ([efgEFG]) character to
12442 * the format string. */
12443 static char const ldblf[] = PERL_PRIfldbl;
12444 char const *p = ldblf + sizeof(ldblf) - 3;
12445 while (p >= ldblf) { *--ptr = *p--; }
12450 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12455 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12467 /* No taint. Otherwise we are in the strange situation
12468 * where printf() taints but print($float) doesn't.
12471 STORE_LC_NUMERIC_SET_TO_NEEDED();
12473 /* hopefully the above makes ptr a very constrained format
12474 * that is safe to use, even though it's not literal */
12475 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12476 #ifdef USE_QUADMATH
12478 const char* qfmt = quadmath_format_single(ptr);
12480 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12481 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12483 if ((IV)elen == -1)
12484 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12488 #elif defined(HAS_LONG_DOUBLE)
12489 elen = ((intsize == 'q')
12490 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12491 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12493 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12499 eptr = PL_efloatbuf;
12500 assert((IV)elen > 0); /* here zero elen is bad */
12502 #ifdef USE_LOCALE_NUMERIC
12503 /* If the decimal point character in the string is UTF-8, make the
12505 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12506 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12519 i = SvCUR(sv) - origlen;
12522 case 'c': *(va_arg(*args, char*)) = i; break;
12523 case 'h': *(va_arg(*args, short*)) = i; break;
12524 default: *(va_arg(*args, int*)) = i; break;
12525 case 'l': *(va_arg(*args, long*)) = i; break;
12526 case 'V': *(va_arg(*args, IV*)) = i; break;
12527 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12528 #ifdef HAS_PTRDIFF_T
12529 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12532 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12536 *(va_arg(*args, Quad_t*)) = i; break;
12543 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12544 continue; /* not "break" */
12551 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12552 && ckWARN(WARN_PRINTF))
12554 SV * const msg = sv_newmortal();
12555 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12556 (PL_op->op_type == OP_PRTF) ? "" : "s");
12557 if (fmtstart < patend) {
12558 const char * const fmtend = q < patend ? q : patend;
12560 sv_catpvs(msg, "\"%");
12561 for (f = fmtstart; f < fmtend; f++) {
12563 sv_catpvn_nomg(msg, f, 1);
12565 Perl_sv_catpvf(aTHX_ msg,
12566 "\\%03"UVof, (UV)*f & 0xFF);
12569 sv_catpvs(msg, "\"");
12571 sv_catpvs(msg, "end of string");
12573 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12576 /* output mangled stuff ... */
12582 /* ... right here, because formatting flags should not apply */
12583 SvGROW(sv, SvCUR(sv) + elen + 1);
12585 Copy(eptr, p, elen, char);
12588 SvCUR_set(sv, p - SvPVX_const(sv));
12590 continue; /* not "break" */
12593 if (is_utf8 != has_utf8) {
12596 sv_utf8_upgrade(sv);
12599 const STRLEN old_elen = elen;
12600 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12601 sv_utf8_upgrade(nsv);
12602 eptr = SvPVX_const(nsv);
12605 if (width) { /* fudge width (can't fudge elen) */
12606 width += elen - old_elen;
12612 assert((IV)elen >= 0); /* here zero elen is fine */
12613 have = esignlen + zeros + elen;
12615 croak_memory_wrap();
12617 need = (have > width ? have : width);
12620 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12621 croak_memory_wrap();
12622 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12624 if (esignlen && fill == '0') {
12626 for (i = 0; i < (int)esignlen; i++)
12627 *p++ = esignbuf[i];
12629 if (gap && !left) {
12630 memset(p, fill, gap);
12633 if (esignlen && fill != '0') {
12635 for (i = 0; i < (int)esignlen; i++)
12636 *p++ = esignbuf[i];
12640 for (i = zeros; i; i--)
12644 Copy(eptr, p, elen, char);
12648 memset(p, ' ', gap);
12653 Copy(dotstr, p, dotstrlen, char);
12657 vectorize = FALSE; /* done iterating over vecstr */
12664 SvCUR_set(sv, p - SvPVX_const(sv));
12671 /* Now that we've consumed all our printf format arguments (svix)
12672 * do we have things left on the stack that we didn't use?
12674 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12675 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12676 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12681 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12685 /* =========================================================================
12687 =head1 Cloning an interpreter
12691 All the macros and functions in this section are for the private use of
12692 the main function, perl_clone().
12694 The foo_dup() functions make an exact copy of an existing foo thingy.
12695 During the course of a cloning, a hash table is used to map old addresses
12696 to new addresses. The table is created and manipulated with the
12697 ptr_table_* functions.
12699 * =========================================================================*/
12702 #if defined(USE_ITHREADS)
12704 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12705 #ifndef GpREFCNT_inc
12706 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12710 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12711 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12712 If this changes, please unmerge ss_dup.
12713 Likewise, sv_dup_inc_multiple() relies on this fact. */
12714 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12715 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12716 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12717 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12718 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12719 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12720 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12721 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12722 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12723 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12724 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12725 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12726 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12728 /* clone a parser */
12731 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12735 PERL_ARGS_ASSERT_PARSER_DUP;
12740 /* look for it in the table first */
12741 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12745 /* create anew and remember what it is */
12746 Newxz(parser, 1, yy_parser);
12747 ptr_table_store(PL_ptr_table, proto, parser);
12749 /* XXX these not yet duped */
12750 parser->old_parser = NULL;
12751 parser->stack = NULL;
12753 parser->stack_size = 0;
12754 /* XXX parser->stack->state = 0; */
12756 /* XXX eventually, just Copy() most of the parser struct ? */
12758 parser->lex_brackets = proto->lex_brackets;
12759 parser->lex_casemods = proto->lex_casemods;
12760 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12761 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12762 parser->lex_casestack = savepvn(proto->lex_casestack,
12763 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12764 parser->lex_defer = proto->lex_defer;
12765 parser->lex_dojoin = proto->lex_dojoin;
12766 parser->lex_formbrack = proto->lex_formbrack;
12767 parser->lex_inpat = proto->lex_inpat;
12768 parser->lex_inwhat = proto->lex_inwhat;
12769 parser->lex_op = proto->lex_op;
12770 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12771 parser->lex_starts = proto->lex_starts;
12772 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12773 parser->multi_close = proto->multi_close;
12774 parser->multi_open = proto->multi_open;
12775 parser->multi_start = proto->multi_start;
12776 parser->multi_end = proto->multi_end;
12777 parser->preambled = proto->preambled;
12778 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12779 parser->linestr = sv_dup_inc(proto->linestr, param);
12780 parser->expect = proto->expect;
12781 parser->copline = proto->copline;
12782 parser->last_lop_op = proto->last_lop_op;
12783 parser->lex_state = proto->lex_state;
12784 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12785 /* rsfp_filters entries have fake IoDIRP() */
12786 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12787 parser->in_my = proto->in_my;
12788 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12789 parser->error_count = proto->error_count;
12792 parser->linestr = sv_dup_inc(proto->linestr, param);
12795 char * const ols = SvPVX(proto->linestr);
12796 char * const ls = SvPVX(parser->linestr);
12798 parser->bufptr = ls + (proto->bufptr >= ols ?
12799 proto->bufptr - ols : 0);
12800 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12801 proto->oldbufptr - ols : 0);
12802 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12803 proto->oldoldbufptr - ols : 0);
12804 parser->linestart = ls + (proto->linestart >= ols ?
12805 proto->linestart - ols : 0);
12806 parser->last_uni = ls + (proto->last_uni >= ols ?
12807 proto->last_uni - ols : 0);
12808 parser->last_lop = ls + (proto->last_lop >= ols ?
12809 proto->last_lop - ols : 0);
12811 parser->bufend = ls + SvCUR(parser->linestr);
12814 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12817 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12818 Copy(proto->nexttype, parser->nexttype, 5, I32);
12819 parser->nexttoke = proto->nexttoke;
12821 /* XXX should clone saved_curcop here, but we aren't passed
12822 * proto_perl; so do it in perl_clone_using instead */
12828 /* duplicate a file handle */
12831 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12835 PERL_ARGS_ASSERT_FP_DUP;
12836 PERL_UNUSED_ARG(type);
12839 return (PerlIO*)NULL;
12841 /* look for it in the table first */
12842 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12846 /* create anew and remember what it is */
12847 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12848 ptr_table_store(PL_ptr_table, fp, ret);
12852 /* duplicate a directory handle */
12855 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12859 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12861 const Direntry_t *dirent;
12862 char smallbuf[256];
12868 PERL_UNUSED_CONTEXT;
12869 PERL_ARGS_ASSERT_DIRP_DUP;
12874 /* look for it in the table first */
12875 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12879 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12881 PERL_UNUSED_ARG(param);
12885 /* open the current directory (so we can switch back) */
12886 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12888 /* chdir to our dir handle and open the present working directory */
12889 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12890 PerlDir_close(pwd);
12891 return (DIR *)NULL;
12893 /* Now we should have two dir handles pointing to the same dir. */
12895 /* Be nice to the calling code and chdir back to where we were. */
12896 /* XXX If this fails, then what? */
12897 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12899 /* We have no need of the pwd handle any more. */
12900 PerlDir_close(pwd);
12903 # define d_namlen(d) (d)->d_namlen
12905 # define d_namlen(d) strlen((d)->d_name)
12907 /* Iterate once through dp, to get the file name at the current posi-
12908 tion. Then step back. */
12909 pos = PerlDir_tell(dp);
12910 if ((dirent = PerlDir_read(dp))) {
12911 len = d_namlen(dirent);
12912 if (len <= sizeof smallbuf) name = smallbuf;
12913 else Newx(name, len, char);
12914 Move(dirent->d_name, name, len, char);
12916 PerlDir_seek(dp, pos);
12918 /* Iterate through the new dir handle, till we find a file with the
12920 if (!dirent) /* just before the end */
12922 pos = PerlDir_tell(ret);
12923 if (PerlDir_read(ret)) continue; /* not there yet */
12924 PerlDir_seek(ret, pos); /* step back */
12928 const long pos0 = PerlDir_tell(ret);
12930 pos = PerlDir_tell(ret);
12931 if ((dirent = PerlDir_read(ret))) {
12932 if (len == (STRLEN)d_namlen(dirent)
12933 && memEQ(name, dirent->d_name, len)) {
12935 PerlDir_seek(ret, pos); /* step back */
12938 /* else we are not there yet; keep iterating */
12940 else { /* This is not meant to happen. The best we can do is
12941 reset the iterator to the beginning. */
12942 PerlDir_seek(ret, pos0);
12949 if (name && name != smallbuf)
12954 ret = win32_dirp_dup(dp, param);
12957 /* pop it in the pointer table */
12959 ptr_table_store(PL_ptr_table, dp, ret);
12964 /* duplicate a typeglob */
12967 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12971 PERL_ARGS_ASSERT_GP_DUP;
12975 /* look for it in the table first */
12976 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12980 /* create anew and remember what it is */
12982 ptr_table_store(PL_ptr_table, gp, ret);
12985 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12986 on Newxz() to do this for us. */
12987 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12988 ret->gp_io = io_dup_inc(gp->gp_io, param);
12989 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12990 ret->gp_av = av_dup_inc(gp->gp_av, param);
12991 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12992 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12993 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12994 ret->gp_cvgen = gp->gp_cvgen;
12995 ret->gp_line = gp->gp_line;
12996 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13000 /* duplicate a chain of magic */
13003 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13005 MAGIC *mgret = NULL;
13006 MAGIC **mgprev_p = &mgret;
13008 PERL_ARGS_ASSERT_MG_DUP;
13010 for (; mg; mg = mg->mg_moremagic) {
13013 if ((param->flags & CLONEf_JOIN_IN)
13014 && mg->mg_type == PERL_MAGIC_backref)
13015 /* when joining, we let the individual SVs add themselves to
13016 * backref as needed. */
13019 Newx(nmg, 1, MAGIC);
13021 mgprev_p = &(nmg->mg_moremagic);
13023 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13024 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13025 from the original commit adding Perl_mg_dup() - revision 4538.
13026 Similarly there is the annotation "XXX random ptr?" next to the
13027 assignment to nmg->mg_ptr. */
13030 /* FIXME for plugins
13031 if (nmg->mg_type == PERL_MAGIC_qr) {
13032 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13036 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13037 ? nmg->mg_type == PERL_MAGIC_backref
13038 /* The backref AV has its reference
13039 * count deliberately bumped by 1 */
13040 ? SvREFCNT_inc(av_dup_inc((const AV *)
13041 nmg->mg_obj, param))
13042 : sv_dup_inc(nmg->mg_obj, param)
13043 : sv_dup(nmg->mg_obj, param);
13045 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13046 if (nmg->mg_len > 0) {
13047 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13048 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13049 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13051 AMT * const namtp = (AMT*)nmg->mg_ptr;
13052 sv_dup_inc_multiple((SV**)(namtp->table),
13053 (SV**)(namtp->table), NofAMmeth, param);
13056 else if (nmg->mg_len == HEf_SVKEY)
13057 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13059 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13060 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13066 #endif /* USE_ITHREADS */
13068 struct ptr_tbl_arena {
13069 struct ptr_tbl_arena *next;
13070 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13073 /* create a new pointer-mapping table */
13076 Perl_ptr_table_new(pTHX)
13079 PERL_UNUSED_CONTEXT;
13081 Newx(tbl, 1, PTR_TBL_t);
13082 tbl->tbl_max = 511;
13083 tbl->tbl_items = 0;
13084 tbl->tbl_arena = NULL;
13085 tbl->tbl_arena_next = NULL;
13086 tbl->tbl_arena_end = NULL;
13087 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13091 #define PTR_TABLE_HASH(ptr) \
13092 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13094 /* map an existing pointer using a table */
13096 STATIC PTR_TBL_ENT_t *
13097 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13099 PTR_TBL_ENT_t *tblent;
13100 const UV hash = PTR_TABLE_HASH(sv);
13102 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13104 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13105 for (; tblent; tblent = tblent->next) {
13106 if (tblent->oldval == sv)
13113 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13115 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13117 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13118 PERL_UNUSED_CONTEXT;
13120 return tblent ? tblent->newval : NULL;
13123 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13124 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13125 * the core's typical use of ptr_tables in thread cloning. */
13128 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13130 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13132 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13133 PERL_UNUSED_CONTEXT;
13136 tblent->newval = newsv;
13138 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13140 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13141 struct ptr_tbl_arena *new_arena;
13143 Newx(new_arena, 1, struct ptr_tbl_arena);
13144 new_arena->next = tbl->tbl_arena;
13145 tbl->tbl_arena = new_arena;
13146 tbl->tbl_arena_next = new_arena->array;
13147 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13150 tblent = tbl->tbl_arena_next++;
13152 tblent->oldval = oldsv;
13153 tblent->newval = newsv;
13154 tblent->next = tbl->tbl_ary[entry];
13155 tbl->tbl_ary[entry] = tblent;
13157 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13158 ptr_table_split(tbl);
13162 /* double the hash bucket size of an existing ptr table */
13165 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13167 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13168 const UV oldsize = tbl->tbl_max + 1;
13169 UV newsize = oldsize * 2;
13172 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13173 PERL_UNUSED_CONTEXT;
13175 Renew(ary, newsize, PTR_TBL_ENT_t*);
13176 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13177 tbl->tbl_max = --newsize;
13178 tbl->tbl_ary = ary;
13179 for (i=0; i < oldsize; i++, ary++) {
13180 PTR_TBL_ENT_t **entp = ary;
13181 PTR_TBL_ENT_t *ent = *ary;
13182 PTR_TBL_ENT_t **curentp;
13185 curentp = ary + oldsize;
13187 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13189 ent->next = *curentp;
13199 /* remove all the entries from a ptr table */
13200 /* Deprecated - will be removed post 5.14 */
13203 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13205 PERL_UNUSED_CONTEXT;
13206 if (tbl && tbl->tbl_items) {
13207 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13209 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
13212 struct ptr_tbl_arena *next = arena->next;
13218 tbl->tbl_items = 0;
13219 tbl->tbl_arena = NULL;
13220 tbl->tbl_arena_next = NULL;
13221 tbl->tbl_arena_end = NULL;
13225 /* clear and free a ptr table */
13228 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13230 struct ptr_tbl_arena *arena;
13232 PERL_UNUSED_CONTEXT;
13238 arena = tbl->tbl_arena;
13241 struct ptr_tbl_arena *next = arena->next;
13247 Safefree(tbl->tbl_ary);
13251 #if defined(USE_ITHREADS)
13254 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13256 PERL_ARGS_ASSERT_RVPV_DUP;
13258 assert(!isREGEXP(sstr));
13260 if (SvWEAKREF(sstr)) {
13261 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13262 if (param->flags & CLONEf_JOIN_IN) {
13263 /* if joining, we add any back references individually rather
13264 * than copying the whole backref array */
13265 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13269 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13271 else if (SvPVX_const(sstr)) {
13272 /* Has something there */
13274 /* Normal PV - clone whole allocated space */
13275 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13276 /* sstr may not be that normal, but actually copy on write.
13277 But we are a true, independent SV, so: */
13281 /* Special case - not normally malloced for some reason */
13282 if (isGV_with_GP(sstr)) {
13283 /* Don't need to do anything here. */
13285 else if ((SvIsCOW(sstr))) {
13286 /* A "shared" PV - clone it as "shared" PV */
13288 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13292 /* Some other special case - random pointer */
13293 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13298 /* Copy the NULL */
13299 SvPV_set(dstr, NULL);
13303 /* duplicate a list of SVs. source and dest may point to the same memory. */
13305 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13306 SSize_t items, CLONE_PARAMS *const param)
13308 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13310 while (items-- > 0) {
13311 *dest++ = sv_dup_inc(*source++, param);
13317 /* duplicate an SV of any type (including AV, HV etc) */
13320 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13325 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13327 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13328 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13333 /* look for it in the table first */
13334 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13338 if(param->flags & CLONEf_JOIN_IN) {
13339 /** We are joining here so we don't want do clone
13340 something that is bad **/
13341 if (SvTYPE(sstr) == SVt_PVHV) {
13342 const HEK * const hvname = HvNAME_HEK(sstr);
13344 /** don't clone stashes if they already exist **/
13345 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13346 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13347 ptr_table_store(PL_ptr_table, sstr, dstr);
13351 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13352 HV *stash = GvSTASH(sstr);
13353 const HEK * hvname;
13354 if (stash && (hvname = HvNAME_HEK(stash))) {
13355 /** don't clone GVs if they already exist **/
13357 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13358 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13360 stash, GvNAME(sstr),
13366 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13367 ptr_table_store(PL_ptr_table, sstr, *svp);
13374 /* create anew and remember what it is */
13377 #ifdef DEBUG_LEAKING_SCALARS
13378 dstr->sv_debug_optype = sstr->sv_debug_optype;
13379 dstr->sv_debug_line = sstr->sv_debug_line;
13380 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13381 dstr->sv_debug_parent = (SV*)sstr;
13382 FREE_SV_DEBUG_FILE(dstr);
13383 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13386 ptr_table_store(PL_ptr_table, sstr, dstr);
13389 SvFLAGS(dstr) = SvFLAGS(sstr);
13390 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13391 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13394 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13395 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13396 (void*)PL_watch_pvx, SvPVX_const(sstr));
13399 /* don't clone objects whose class has asked us not to */
13401 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13407 switch (SvTYPE(sstr)) {
13409 SvANY(dstr) = NULL;
13412 SET_SVANY_FOR_BODYLESS_IV(dstr);
13414 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13416 SvIV_set(dstr, SvIVX(sstr));
13420 #if NVSIZE <= IVSIZE
13421 SET_SVANY_FOR_BODYLESS_NV(dstr);
13423 SvANY(dstr) = new_XNV();
13425 SvNV_set(dstr, SvNVX(sstr));
13429 /* These are all the types that need complex bodies allocating. */
13431 const svtype sv_type = SvTYPE(sstr);
13432 const struct body_details *const sv_type_details
13433 = bodies_by_type + sv_type;
13437 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13453 assert(sv_type_details->body_size);
13454 if (sv_type_details->arena) {
13455 new_body_inline(new_body, sv_type);
13457 = (void*)((char*)new_body - sv_type_details->offset);
13459 new_body = new_NOARENA(sv_type_details);
13463 SvANY(dstr) = new_body;
13466 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13467 ((char*)SvANY(dstr)) + sv_type_details->offset,
13468 sv_type_details->copy, char);
13470 Copy(((char*)SvANY(sstr)),
13471 ((char*)SvANY(dstr)),
13472 sv_type_details->body_size + sv_type_details->offset, char);
13475 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13476 && !isGV_with_GP(dstr)
13478 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13479 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13481 /* The Copy above means that all the source (unduplicated) pointers
13482 are now in the destination. We can check the flags and the
13483 pointers in either, but it's possible that there's less cache
13484 missing by always going for the destination.
13485 FIXME - instrument and check that assumption */
13486 if (sv_type >= SVt_PVMG) {
13488 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13489 if (SvOBJECT(dstr) && SvSTASH(dstr))
13490 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13491 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13494 /* The cast silences a GCC warning about unhandled types. */
13495 switch ((int)sv_type) {
13506 /* FIXME for plugins */
13507 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13508 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13511 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13512 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13513 LvTARG(dstr) = dstr;
13514 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13515 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13517 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13518 if (isREGEXP(sstr)) goto duprex;
13520 /* non-GP case already handled above */
13521 if(isGV_with_GP(sstr)) {
13522 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13523 /* Don't call sv_add_backref here as it's going to be
13524 created as part of the magic cloning of the symbol
13525 table--unless this is during a join and the stash
13526 is not actually being cloned. */
13527 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13528 at the point of this comment. */
13529 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13530 if (param->flags & CLONEf_JOIN_IN)
13531 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13532 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13533 (void)GpREFCNT_inc(GvGP(dstr));
13537 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13538 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13539 /* I have no idea why fake dirp (rsfps)
13540 should be treated differently but otherwise
13541 we end up with leaks -- sky*/
13542 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13543 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13544 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13546 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13547 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13548 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13549 if (IoDIRP(dstr)) {
13550 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13553 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13555 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13557 if (IoOFP(dstr) == IoIFP(sstr))
13558 IoOFP(dstr) = IoIFP(dstr);
13560 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13561 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13562 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13563 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13566 /* avoid cloning an empty array */
13567 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13568 SV **dst_ary, **src_ary;
13569 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13571 src_ary = AvARRAY((const AV *)sstr);
13572 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13573 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13574 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13575 AvALLOC((const AV *)dstr) = dst_ary;
13576 if (AvREAL((const AV *)sstr)) {
13577 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13581 while (items-- > 0)
13582 *dst_ary++ = sv_dup(*src_ary++, param);
13584 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13585 while (items-- > 0) {
13586 *dst_ary++ = &PL_sv_undef;
13590 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13591 AvALLOC((const AV *)dstr) = (SV**)NULL;
13592 AvMAX( (const AV *)dstr) = -1;
13593 AvFILLp((const AV *)dstr) = -1;
13597 if (HvARRAY((const HV *)sstr)) {
13599 const bool sharekeys = !!HvSHAREKEYS(sstr);
13600 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13601 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13603 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13604 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13606 HvARRAY(dstr) = (HE**)darray;
13607 while (i <= sxhv->xhv_max) {
13608 const HE * const source = HvARRAY(sstr)[i];
13609 HvARRAY(dstr)[i] = source
13610 ? he_dup(source, sharekeys, param) : 0;
13614 const struct xpvhv_aux * const saux = HvAUX(sstr);
13615 struct xpvhv_aux * const daux = HvAUX(dstr);
13616 /* This flag isn't copied. */
13619 if (saux->xhv_name_count) {
13620 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13622 = saux->xhv_name_count < 0
13623 ? -saux->xhv_name_count
13624 : saux->xhv_name_count;
13625 HEK **shekp = sname + count;
13627 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13628 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13629 while (shekp-- > sname) {
13631 *dhekp = hek_dup(*shekp, param);
13635 daux->xhv_name_u.xhvnameu_name
13636 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13639 daux->xhv_name_count = saux->xhv_name_count;
13641 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13642 daux->xhv_aux_flags = saux->xhv_aux_flags;
13643 #ifdef PERL_HASH_RANDOMIZE_KEYS
13644 daux->xhv_rand = saux->xhv_rand;
13645 daux->xhv_last_rand = saux->xhv_last_rand;
13647 daux->xhv_riter = saux->xhv_riter;
13648 daux->xhv_eiter = saux->xhv_eiter
13649 ? he_dup(saux->xhv_eiter,
13650 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13651 /* backref array needs refcnt=2; see sv_add_backref */
13652 daux->xhv_backreferences =
13653 (param->flags & CLONEf_JOIN_IN)
13654 /* when joining, we let the individual GVs and
13655 * CVs add themselves to backref as
13656 * needed. This avoids pulling in stuff
13657 * that isn't required, and simplifies the
13658 * case where stashes aren't cloned back
13659 * if they already exist in the parent
13662 : saux->xhv_backreferences
13663 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13664 ? MUTABLE_AV(SvREFCNT_inc(
13665 sv_dup_inc((const SV *)
13666 saux->xhv_backreferences, param)))
13667 : MUTABLE_AV(sv_dup((const SV *)
13668 saux->xhv_backreferences, param))
13671 daux->xhv_mro_meta = saux->xhv_mro_meta
13672 ? mro_meta_dup(saux->xhv_mro_meta, param)
13675 /* Record stashes for possible cloning in Perl_clone(). */
13677 av_push(param->stashes, dstr);
13681 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13684 if (!(param->flags & CLONEf_COPY_STACKS)) {
13689 /* NOTE: not refcounted */
13690 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13691 hv_dup(CvSTASH(dstr), param);
13692 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13693 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13694 if (!CvISXSUB(dstr)) {
13696 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13698 CvSLABBED_off(dstr);
13699 } else if (CvCONST(dstr)) {
13700 CvXSUBANY(dstr).any_ptr =
13701 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13703 assert(!CvSLABBED(dstr));
13704 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13706 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13707 hek_dup(CvNAME_HEK((CV *)sstr), param);
13708 /* don't dup if copying back - CvGV isn't refcounted, so the
13709 * duped GV may never be freed. A bit of a hack! DAPM */
13711 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13713 ? gv_dup_inc(CvGV(sstr), param)
13714 : (param->flags & CLONEf_JOIN_IN)
13716 : gv_dup(CvGV(sstr), param);
13718 if (!CvISXSUB(sstr)) {
13719 PADLIST * padlist = CvPADLIST(sstr);
13721 padlist = padlist_dup(padlist, param);
13722 CvPADLIST_set(dstr, padlist);
13724 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13725 PoisonPADLIST(dstr);
13728 CvWEAKOUTSIDE(sstr)
13729 ? cv_dup( CvOUTSIDE(dstr), param)
13730 : cv_dup_inc(CvOUTSIDE(dstr), param);
13740 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13742 PERL_ARGS_ASSERT_SV_DUP_INC;
13743 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13747 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13749 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13750 PERL_ARGS_ASSERT_SV_DUP;
13752 /* Track every SV that (at least initially) had a reference count of 0.
13753 We need to do this by holding an actual reference to it in this array.
13754 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13755 (akin to the stashes hash, and the perl stack), we come unstuck if
13756 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13757 thread) is manipulated in a CLONE method, because CLONE runs before the
13758 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13759 (and fix things up by giving each a reference via the temps stack).
13760 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13761 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13762 before the walk of unreferenced happens and a reference to that is SV
13763 added to the temps stack. At which point we have the same SV considered
13764 to be in use, and free to be re-used. Not good.
13766 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13767 assert(param->unreferenced);
13768 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13774 /* duplicate a context */
13777 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13779 PERL_CONTEXT *ncxs;
13781 PERL_ARGS_ASSERT_CX_DUP;
13784 return (PERL_CONTEXT*)NULL;
13786 /* look for it in the table first */
13787 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13791 /* create anew and remember what it is */
13792 Newx(ncxs, max + 1, PERL_CONTEXT);
13793 ptr_table_store(PL_ptr_table, cxs, ncxs);
13794 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13797 PERL_CONTEXT * const ncx = &ncxs[ix];
13798 if (CxTYPE(ncx) == CXt_SUBST) {
13799 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13802 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13803 switch (CxTYPE(ncx)) {
13805 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13806 ? cv_dup_inc(ncx->blk_sub.cv, param)
13807 : cv_dup(ncx->blk_sub.cv,param));
13808 if(CxHASARGS(ncx)){
13809 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13810 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13812 ncx->blk_sub.argarray = NULL;
13813 ncx->blk_sub.savearray = NULL;
13815 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13816 ncx->blk_sub.oldcomppad);
13819 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13821 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13822 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13824 case CXt_LOOP_LAZYSV:
13825 ncx->blk_loop.state_u.lazysv.end
13826 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13827 /* We are taking advantage of av_dup_inc and sv_dup_inc
13828 actually being the same function, and order equivalence of
13830 We can assert the later [but only at run time :-(] */
13831 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13832 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13834 ncx->blk_loop.state_u.ary.ary
13835 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13836 case CXt_LOOP_LAZYIV:
13837 case CXt_LOOP_PLAIN:
13838 if (CxPADLOOP(ncx)) {
13839 ncx->blk_loop.itervar_u.oldcomppad
13840 = (PAD*)ptr_table_fetch(PL_ptr_table,
13841 ncx->blk_loop.itervar_u.oldcomppad);
13843 ncx->blk_loop.itervar_u.gv
13844 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13849 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13850 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13851 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13866 /* duplicate a stack info structure */
13869 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13873 PERL_ARGS_ASSERT_SI_DUP;
13876 return (PERL_SI*)NULL;
13878 /* look for it in the table first */
13879 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13883 /* create anew and remember what it is */
13884 Newxz(nsi, 1, PERL_SI);
13885 ptr_table_store(PL_ptr_table, si, nsi);
13887 nsi->si_stack = av_dup_inc(si->si_stack, param);
13888 nsi->si_cxix = si->si_cxix;
13889 nsi->si_cxmax = si->si_cxmax;
13890 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13891 nsi->si_type = si->si_type;
13892 nsi->si_prev = si_dup(si->si_prev, param);
13893 nsi->si_next = si_dup(si->si_next, param);
13894 nsi->si_markoff = si->si_markoff;
13899 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13900 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13901 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13902 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13903 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13904 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13905 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13906 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13907 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13908 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13909 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13910 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13911 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13912 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13913 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13914 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13917 #define pv_dup_inc(p) SAVEPV(p)
13918 #define pv_dup(p) SAVEPV(p)
13919 #define svp_dup_inc(p,pp) any_dup(p,pp)
13921 /* map any object to the new equivent - either something in the
13922 * ptr table, or something in the interpreter structure
13926 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13930 PERL_ARGS_ASSERT_ANY_DUP;
13933 return (void*)NULL;
13935 /* look for it in the table first */
13936 ret = ptr_table_fetch(PL_ptr_table, v);
13940 /* see if it is part of the interpreter structure */
13941 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13942 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13950 /* duplicate the save stack */
13953 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13956 ANY * const ss = proto_perl->Isavestack;
13957 const I32 max = proto_perl->Isavestack_max;
13958 I32 ix = proto_perl->Isavestack_ix;
13971 void (*dptr) (void*);
13972 void (*dxptr) (pTHX_ void*);
13974 PERL_ARGS_ASSERT_SS_DUP;
13976 Newxz(nss, max, ANY);
13979 const UV uv = POPUV(ss,ix);
13980 const U8 type = (U8)uv & SAVE_MASK;
13982 TOPUV(nss,ix) = uv;
13984 case SAVEt_CLEARSV:
13985 case SAVEt_CLEARPADRANGE:
13987 case SAVEt_HELEM: /* hash element */
13988 case SAVEt_SV: /* scalar reference */
13989 sv = (const SV *)POPPTR(ss,ix);
13990 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
13992 case SAVEt_ITEM: /* normal string */
13993 case SAVEt_GVSV: /* scalar slot in GV */
13994 sv = (const SV *)POPPTR(ss,ix);
13995 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13996 if (type == SAVEt_SV)
14000 case SAVEt_MORTALIZESV:
14001 case SAVEt_READONLY_OFF:
14002 sv = (const SV *)POPPTR(ss,ix);
14003 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14005 case SAVEt_FREEPADNAME:
14006 ptr = POPPTR(ss,ix);
14007 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14008 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14010 case SAVEt_SHARED_PVREF: /* char* in shared space */
14011 c = (char*)POPPTR(ss,ix);
14012 TOPPTR(nss,ix) = savesharedpv(c);
14013 ptr = POPPTR(ss,ix);
14014 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14016 case SAVEt_GENERIC_SVREF: /* generic sv */
14017 case SAVEt_SVREF: /* scalar reference */
14018 sv = (const SV *)POPPTR(ss,ix);
14019 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14020 if (type == SAVEt_SVREF)
14021 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14022 ptr = POPPTR(ss,ix);
14023 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14025 case SAVEt_GVSLOT: /* any slot in GV */
14026 sv = (const SV *)POPPTR(ss,ix);
14027 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14028 ptr = POPPTR(ss,ix);
14029 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14030 sv = (const SV *)POPPTR(ss,ix);
14031 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14033 case SAVEt_HV: /* hash reference */
14034 case SAVEt_AV: /* array reference */
14035 sv = (const SV *) POPPTR(ss,ix);
14036 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14038 case SAVEt_COMPPAD:
14040 sv = (const SV *) POPPTR(ss,ix);
14041 TOPPTR(nss,ix) = sv_dup(sv, param);
14043 case SAVEt_INT: /* int reference */
14044 ptr = POPPTR(ss,ix);
14045 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14046 intval = (int)POPINT(ss,ix);
14047 TOPINT(nss,ix) = intval;
14049 case SAVEt_LONG: /* long reference */
14050 ptr = POPPTR(ss,ix);
14051 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14052 longval = (long)POPLONG(ss,ix);
14053 TOPLONG(nss,ix) = longval;
14055 case SAVEt_I32: /* I32 reference */
14056 ptr = POPPTR(ss,ix);
14057 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14059 TOPINT(nss,ix) = i;
14061 case SAVEt_IV: /* IV reference */
14062 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14063 ptr = POPPTR(ss,ix);
14064 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14066 TOPIV(nss,ix) = iv;
14068 case SAVEt_HPTR: /* HV* reference */
14069 case SAVEt_APTR: /* AV* reference */
14070 case SAVEt_SPTR: /* SV* reference */
14071 ptr = POPPTR(ss,ix);
14072 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14073 sv = (const SV *)POPPTR(ss,ix);
14074 TOPPTR(nss,ix) = sv_dup(sv, param);
14076 case SAVEt_VPTR: /* random* reference */
14077 ptr = POPPTR(ss,ix);
14078 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14080 case SAVEt_INT_SMALL:
14081 case SAVEt_I32_SMALL:
14082 case SAVEt_I16: /* I16 reference */
14083 case SAVEt_I8: /* I8 reference */
14085 ptr = POPPTR(ss,ix);
14086 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14088 case SAVEt_GENERIC_PVREF: /* generic char* */
14089 case SAVEt_PPTR: /* char* reference */
14090 ptr = POPPTR(ss,ix);
14091 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14092 c = (char*)POPPTR(ss,ix);
14093 TOPPTR(nss,ix) = pv_dup(c);
14095 case SAVEt_GP: /* scalar reference */
14096 gp = (GP*)POPPTR(ss,ix);
14097 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14098 (void)GpREFCNT_inc(gp);
14099 gv = (const GV *)POPPTR(ss,ix);
14100 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14103 ptr = POPPTR(ss,ix);
14104 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14105 /* these are assumed to be refcounted properly */
14107 switch (((OP*)ptr)->op_type) {
14109 case OP_LEAVESUBLV:
14113 case OP_LEAVEWRITE:
14114 TOPPTR(nss,ix) = ptr;
14117 (void) OpREFCNT_inc(o);
14121 TOPPTR(nss,ix) = NULL;
14126 TOPPTR(nss,ix) = NULL;
14128 case SAVEt_FREECOPHH:
14129 ptr = POPPTR(ss,ix);
14130 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14132 case SAVEt_ADELETE:
14133 av = (const AV *)POPPTR(ss,ix);
14134 TOPPTR(nss,ix) = av_dup_inc(av, param);
14136 TOPINT(nss,ix) = i;
14139 hv = (const HV *)POPPTR(ss,ix);
14140 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14142 TOPINT(nss,ix) = i;
14145 c = (char*)POPPTR(ss,ix);
14146 TOPPTR(nss,ix) = pv_dup_inc(c);
14148 case SAVEt_STACK_POS: /* Position on Perl stack */
14150 TOPINT(nss,ix) = i;
14152 case SAVEt_DESTRUCTOR:
14153 ptr = POPPTR(ss,ix);
14154 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14155 dptr = POPDPTR(ss,ix);
14156 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14157 any_dup(FPTR2DPTR(void *, dptr),
14160 case SAVEt_DESTRUCTOR_X:
14161 ptr = POPPTR(ss,ix);
14162 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14163 dxptr = POPDXPTR(ss,ix);
14164 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14165 any_dup(FPTR2DPTR(void *, dxptr),
14168 case SAVEt_REGCONTEXT:
14170 ix -= uv >> SAVE_TIGHT_SHIFT;
14172 case SAVEt_AELEM: /* array element */
14173 sv = (const SV *)POPPTR(ss,ix);
14174 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14176 TOPINT(nss,ix) = i;
14177 av = (const AV *)POPPTR(ss,ix);
14178 TOPPTR(nss,ix) = av_dup_inc(av, param);
14181 ptr = POPPTR(ss,ix);
14182 TOPPTR(nss,ix) = ptr;
14185 ptr = POPPTR(ss,ix);
14186 ptr = cophh_copy((COPHH*)ptr);
14187 TOPPTR(nss,ix) = ptr;
14189 TOPINT(nss,ix) = i;
14190 if (i & HINT_LOCALIZE_HH) {
14191 hv = (const HV *)POPPTR(ss,ix);
14192 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14195 case SAVEt_PADSV_AND_MORTALIZE:
14196 longval = (long)POPLONG(ss,ix);
14197 TOPLONG(nss,ix) = longval;
14198 ptr = POPPTR(ss,ix);
14199 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14200 sv = (const SV *)POPPTR(ss,ix);
14201 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14203 case SAVEt_SET_SVFLAGS:
14205 TOPINT(nss,ix) = i;
14207 TOPINT(nss,ix) = i;
14208 sv = (const SV *)POPPTR(ss,ix);
14209 TOPPTR(nss,ix) = sv_dup(sv, param);
14211 case SAVEt_COMPILE_WARNINGS:
14212 ptr = POPPTR(ss,ix);
14213 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14216 ptr = POPPTR(ss,ix);
14217 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14219 case SAVEt_GP_ALIASED_SV: {
14220 GP * gp_ptr = (GP *)POPPTR(ss,ix);
14221 GP * new_gp_ptr = gp_dup(gp_ptr, param);
14222 TOPPTR(nss,ix) = new_gp_ptr;
14223 new_gp_ptr->gp_refcnt++;
14228 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14236 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14237 * flag to the result. This is done for each stash before cloning starts,
14238 * so we know which stashes want their objects cloned */
14241 do_mark_cloneable_stash(pTHX_ SV *const sv)
14243 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14245 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14246 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14247 if (cloner && GvCV(cloner)) {
14254 mXPUSHs(newSVhek(hvname));
14256 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14263 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14271 =for apidoc perl_clone
14273 Create and return a new interpreter by cloning the current one.
14275 perl_clone takes these flags as parameters:
14277 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14278 without it we only clone the data and zero the stacks,
14279 with it we copy the stacks and the new perl interpreter is
14280 ready to run at the exact same point as the previous one.
14281 The pseudo-fork code uses COPY_STACKS while the
14282 threads->create doesn't.
14284 CLONEf_KEEP_PTR_TABLE -
14285 perl_clone keeps a ptr_table with the pointer of the old
14286 variable as a key and the new variable as a value,
14287 this allows it to check if something has been cloned and not
14288 clone it again but rather just use the value and increase the
14289 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14290 the ptr_table using the function
14291 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14292 reason to keep it around is if you want to dup some of your own
14293 variable who are outside the graph perl scans, example of this
14294 code is in threads.xs create.
14296 CLONEf_CLONE_HOST -
14297 This is a win32 thing, it is ignored on unix, it tells perls
14298 win32host code (which is c++) to clone itself, this is needed on
14299 win32 if you want to run two threads at the same time,
14300 if you just want to do some stuff in a separate perl interpreter
14301 and then throw it away and return to the original one,
14302 you don't need to do anything.
14307 /* XXX the above needs expanding by someone who actually understands it ! */
14308 EXTERN_C PerlInterpreter *
14309 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14312 perl_clone(PerlInterpreter *proto_perl, UV flags)
14315 #ifdef PERL_IMPLICIT_SYS
14317 PERL_ARGS_ASSERT_PERL_CLONE;
14319 /* perlhost.h so we need to call into it
14320 to clone the host, CPerlHost should have a c interface, sky */
14322 if (flags & CLONEf_CLONE_HOST) {
14323 return perl_clone_host(proto_perl,flags);
14325 return perl_clone_using(proto_perl, flags,
14327 proto_perl->IMemShared,
14328 proto_perl->IMemParse,
14330 proto_perl->IStdIO,
14334 proto_perl->IProc);
14338 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14339 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14340 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14341 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14342 struct IPerlDir* ipD, struct IPerlSock* ipS,
14343 struct IPerlProc* ipP)
14345 /* XXX many of the string copies here can be optimized if they're
14346 * constants; they need to be allocated as common memory and just
14347 * their pointers copied. */
14350 CLONE_PARAMS clone_params;
14351 CLONE_PARAMS* const param = &clone_params;
14353 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14355 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14356 #else /* !PERL_IMPLICIT_SYS */
14358 CLONE_PARAMS clone_params;
14359 CLONE_PARAMS* param = &clone_params;
14360 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14362 PERL_ARGS_ASSERT_PERL_CLONE;
14363 #endif /* PERL_IMPLICIT_SYS */
14365 /* for each stash, determine whether its objects should be cloned */
14366 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14367 PERL_SET_THX(my_perl);
14370 PoisonNew(my_perl, 1, PerlInterpreter);
14373 PL_defstash = NULL; /* may be used by perl malloc() */
14376 PL_scopestack_name = 0;
14378 PL_savestack_ix = 0;
14379 PL_savestack_max = -1;
14380 PL_sig_pending = 0;
14382 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14383 Zero(&PL_padname_undef, 1, PADNAME);
14384 Zero(&PL_padname_const, 1, PADNAME);
14385 # ifdef DEBUG_LEAKING_SCALARS
14386 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14388 #else /* !DEBUGGING */
14389 Zero(my_perl, 1, PerlInterpreter);
14390 #endif /* DEBUGGING */
14392 #ifdef PERL_IMPLICIT_SYS
14393 /* host pointers */
14395 PL_MemShared = ipMS;
14396 PL_MemParse = ipMP;
14403 #endif /* PERL_IMPLICIT_SYS */
14406 param->flags = flags;
14407 /* Nothing in the core code uses this, but we make it available to
14408 extensions (using mg_dup). */
14409 param->proto_perl = proto_perl;
14410 /* Likely nothing will use this, but it is initialised to be consistent
14411 with Perl_clone_params_new(). */
14412 param->new_perl = my_perl;
14413 param->unreferenced = NULL;
14416 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14418 PL_body_arenas = NULL;
14419 Zero(&PL_body_roots, 1, PL_body_roots);
14423 PL_sv_arenaroot = NULL;
14425 PL_debug = proto_perl->Idebug;
14427 /* dbargs array probably holds garbage */
14430 PL_compiling = proto_perl->Icompiling;
14432 /* pseudo environmental stuff */
14433 PL_origargc = proto_perl->Iorigargc;
14434 PL_origargv = proto_perl->Iorigargv;
14436 #ifndef NO_TAINT_SUPPORT
14437 /* Set tainting stuff before PerlIO_debug can possibly get called */
14438 PL_tainting = proto_perl->Itainting;
14439 PL_taint_warn = proto_perl->Itaint_warn;
14441 PL_tainting = FALSE;
14442 PL_taint_warn = FALSE;
14445 PL_minus_c = proto_perl->Iminus_c;
14447 PL_localpatches = proto_perl->Ilocalpatches;
14448 PL_splitstr = proto_perl->Isplitstr;
14449 PL_minus_n = proto_perl->Iminus_n;
14450 PL_minus_p = proto_perl->Iminus_p;
14451 PL_minus_l = proto_perl->Iminus_l;
14452 PL_minus_a = proto_perl->Iminus_a;
14453 PL_minus_E = proto_perl->Iminus_E;
14454 PL_minus_F = proto_perl->Iminus_F;
14455 PL_doswitches = proto_perl->Idoswitches;
14456 PL_dowarn = proto_perl->Idowarn;
14457 PL_sawalias = proto_perl->Isawalias;
14458 #ifdef PERL_SAWAMPERSAND
14459 PL_sawampersand = proto_perl->Isawampersand;
14461 PL_unsafe = proto_perl->Iunsafe;
14462 PL_perldb = proto_perl->Iperldb;
14463 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14464 PL_exit_flags = proto_perl->Iexit_flags;
14466 /* XXX time(&PL_basetime) when asked for? */
14467 PL_basetime = proto_perl->Ibasetime;
14469 PL_maxsysfd = proto_perl->Imaxsysfd;
14470 PL_statusvalue = proto_perl->Istatusvalue;
14472 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14474 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14477 /* RE engine related */
14478 PL_regmatch_slab = NULL;
14479 PL_reg_curpm = NULL;
14481 PL_sub_generation = proto_perl->Isub_generation;
14483 /* funky return mechanisms */
14484 PL_forkprocess = proto_perl->Iforkprocess;
14486 /* internal state */
14487 PL_maxo = proto_perl->Imaxo;
14489 PL_main_start = proto_perl->Imain_start;
14490 PL_eval_root = proto_perl->Ieval_root;
14491 PL_eval_start = proto_perl->Ieval_start;
14493 PL_filemode = proto_perl->Ifilemode;
14494 PL_lastfd = proto_perl->Ilastfd;
14495 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14498 PL_gensym = proto_perl->Igensym;
14500 PL_laststatval = proto_perl->Ilaststatval;
14501 PL_laststype = proto_perl->Ilaststype;
14504 PL_profiledata = NULL;
14506 PL_generation = proto_perl->Igeneration;
14508 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14509 PL_in_clean_all = proto_perl->Iin_clean_all;
14511 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14512 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14513 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14514 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14515 PL_nomemok = proto_perl->Inomemok;
14516 PL_an = proto_perl->Ian;
14517 PL_evalseq = proto_perl->Ievalseq;
14518 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14519 PL_origalen = proto_perl->Iorigalen;
14521 PL_sighandlerp = proto_perl->Isighandlerp;
14523 PL_runops = proto_perl->Irunops;
14525 PL_subline = proto_perl->Isubline;
14528 PL_cryptseen = proto_perl->Icryptseen;
14531 #ifdef USE_LOCALE_COLLATE
14532 PL_collation_ix = proto_perl->Icollation_ix;
14533 PL_collation_standard = proto_perl->Icollation_standard;
14534 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14535 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14536 #endif /* USE_LOCALE_COLLATE */
14538 #ifdef USE_LOCALE_NUMERIC
14539 PL_numeric_standard = proto_perl->Inumeric_standard;
14540 PL_numeric_local = proto_perl->Inumeric_local;
14541 #endif /* !USE_LOCALE_NUMERIC */
14543 /* Did the locale setup indicate UTF-8? */
14544 PL_utf8locale = proto_perl->Iutf8locale;
14545 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14546 /* Unicode features (see perlrun/-C) */
14547 PL_unicode = proto_perl->Iunicode;
14549 /* Pre-5.8 signals control */
14550 PL_signals = proto_perl->Isignals;
14552 /* times() ticks per second */
14553 PL_clocktick = proto_perl->Iclocktick;
14555 /* Recursion stopper for PerlIO_find_layer */
14556 PL_in_load_module = proto_perl->Iin_load_module;
14558 /* sort() routine */
14559 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14561 /* Not really needed/useful since the reenrant_retint is "volatile",
14562 * but do it for consistency's sake. */
14563 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14565 /* Hooks to shared SVs and locks. */
14566 PL_sharehook = proto_perl->Isharehook;
14567 PL_lockhook = proto_perl->Ilockhook;
14568 PL_unlockhook = proto_perl->Iunlockhook;
14569 PL_threadhook = proto_perl->Ithreadhook;
14570 PL_destroyhook = proto_perl->Idestroyhook;
14571 PL_signalhook = proto_perl->Isignalhook;
14573 PL_globhook = proto_perl->Iglobhook;
14576 PL_last_swash_hv = NULL; /* reinits on demand */
14577 PL_last_swash_klen = 0;
14578 PL_last_swash_key[0]= '\0';
14579 PL_last_swash_tmps = (U8*)NULL;
14580 PL_last_swash_slen = 0;
14582 PL_srand_called = proto_perl->Isrand_called;
14583 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14585 if (flags & CLONEf_COPY_STACKS) {
14586 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14587 PL_tmps_ix = proto_perl->Itmps_ix;
14588 PL_tmps_max = proto_perl->Itmps_max;
14589 PL_tmps_floor = proto_perl->Itmps_floor;
14591 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14592 * NOTE: unlike the others! */
14593 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14594 PL_scopestack_max = proto_perl->Iscopestack_max;
14596 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14597 * NOTE: unlike the others! */
14598 PL_savestack_ix = proto_perl->Isavestack_ix;
14599 PL_savestack_max = proto_perl->Isavestack_max;
14602 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14603 PL_top_env = &PL_start_env;
14605 PL_op = proto_perl->Iop;
14608 PL_Xpv = (XPV*)NULL;
14609 my_perl->Ina = proto_perl->Ina;
14611 PL_statbuf = proto_perl->Istatbuf;
14612 PL_statcache = proto_perl->Istatcache;
14614 #ifndef NO_TAINT_SUPPORT
14615 PL_tainted = proto_perl->Itainted;
14617 PL_tainted = FALSE;
14619 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14621 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14623 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14624 PL_restartop = proto_perl->Irestartop;
14625 PL_in_eval = proto_perl->Iin_eval;
14626 PL_delaymagic = proto_perl->Idelaymagic;
14627 PL_phase = proto_perl->Iphase;
14628 PL_localizing = proto_perl->Ilocalizing;
14630 PL_hv_fetch_ent_mh = NULL;
14631 PL_modcount = proto_perl->Imodcount;
14632 PL_lastgotoprobe = NULL;
14633 PL_dumpindent = proto_perl->Idumpindent;
14635 PL_efloatbuf = NULL; /* reinits on demand */
14636 PL_efloatsize = 0; /* reinits on demand */
14640 PL_colorset = 0; /* reinits PL_colors[] */
14641 /*PL_colors[6] = {0,0,0,0,0,0};*/
14643 /* Pluggable optimizer */
14644 PL_peepp = proto_perl->Ipeepp;
14645 PL_rpeepp = proto_perl->Irpeepp;
14646 /* op_free() hook */
14647 PL_opfreehook = proto_perl->Iopfreehook;
14649 #ifdef USE_REENTRANT_API
14650 /* XXX: things like -Dm will segfault here in perlio, but doing
14651 * PERL_SET_CONTEXT(proto_perl);
14652 * breaks too many other things
14654 Perl_reentrant_init(aTHX);
14657 /* create SV map for pointer relocation */
14658 PL_ptr_table = ptr_table_new();
14660 /* initialize these special pointers as early as possible */
14662 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14663 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14664 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14665 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14666 &PL_padname_const);
14668 /* create (a non-shared!) shared string table */
14669 PL_strtab = newHV();
14670 HvSHAREKEYS_off(PL_strtab);
14671 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14672 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14674 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14676 /* This PV will be free'd special way so must set it same way op.c does */
14677 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14678 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14680 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14681 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14682 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14683 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14685 param->stashes = newAV(); /* Setup array of objects to call clone on */
14686 /* This makes no difference to the implementation, as it always pushes
14687 and shifts pointers to other SVs without changing their reference
14688 count, with the array becoming empty before it is freed. However, it
14689 makes it conceptually clear what is going on, and will avoid some
14690 work inside av.c, filling slots between AvFILL() and AvMAX() with
14691 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14692 AvREAL_off(param->stashes);
14694 if (!(flags & CLONEf_COPY_STACKS)) {
14695 param->unreferenced = newAV();
14698 #ifdef PERLIO_LAYERS
14699 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14700 PerlIO_clone(aTHX_ proto_perl, param);
14703 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14704 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14705 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14706 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14707 PL_xsubfilename = proto_perl->Ixsubfilename;
14708 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14709 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14712 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14713 PL_inplace = SAVEPV(proto_perl->Iinplace);
14714 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14716 /* magical thingies */
14718 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14719 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14721 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14722 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14723 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14726 /* Clone the regex array */
14727 /* ORANGE FIXME for plugins, probably in the SV dup code.
14728 newSViv(PTR2IV(CALLREGDUPE(
14729 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14731 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14732 PL_regex_pad = AvARRAY(PL_regex_padav);
14734 PL_stashpadmax = proto_perl->Istashpadmax;
14735 PL_stashpadix = proto_perl->Istashpadix ;
14736 Newx(PL_stashpad, PL_stashpadmax, HV *);
14739 for (; o < PL_stashpadmax; ++o)
14740 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14743 /* shortcuts to various I/O objects */
14744 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14745 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14746 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14747 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14748 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14749 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14750 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14752 /* shortcuts to regexp stuff */
14753 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14755 /* shortcuts to misc objects */
14756 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14758 /* shortcuts to debugging objects */
14759 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14760 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14761 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14762 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14763 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14764 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14765 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14767 /* symbol tables */
14768 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14769 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14770 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14771 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14772 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14774 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14775 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14776 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14777 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14778 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14779 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14780 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14781 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14783 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14785 /* subprocess state */
14786 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14788 if (proto_perl->Iop_mask)
14789 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14792 /* PL_asserting = proto_perl->Iasserting; */
14794 /* current interpreter roots */
14795 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14797 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14800 /* runtime control stuff */
14801 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14803 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14805 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14807 /* interpreter atexit processing */
14808 PL_exitlistlen = proto_perl->Iexitlistlen;
14809 if (PL_exitlistlen) {
14810 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14811 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14814 PL_exitlist = (PerlExitListEntry*)NULL;
14816 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14817 if (PL_my_cxt_size) {
14818 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14819 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14820 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14821 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14822 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14826 PL_my_cxt_list = (void**)NULL;
14827 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14828 PL_my_cxt_keys = (const char**)NULL;
14831 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14832 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14833 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14834 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14836 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14838 PAD_CLONE_VARS(proto_perl, param);
14840 #ifdef HAVE_INTERP_INTERN
14841 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14844 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14846 #ifdef PERL_USES_PL_PIDSTATUS
14847 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14849 PL_osname = SAVEPV(proto_perl->Iosname);
14850 PL_parser = parser_dup(proto_perl->Iparser, param);
14852 /* XXX this only works if the saved cop has already been cloned */
14853 if (proto_perl->Iparser) {
14854 PL_parser->saved_curcop = (COP*)any_dup(
14855 proto_perl->Iparser->saved_curcop,
14859 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14861 #ifdef USE_LOCALE_COLLATE
14862 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14863 #endif /* USE_LOCALE_COLLATE */
14865 #ifdef USE_LOCALE_NUMERIC
14866 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14867 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14868 #endif /* !USE_LOCALE_NUMERIC */
14870 /* Unicode inversion lists */
14871 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14872 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14873 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14874 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14876 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14877 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14879 /* utf8 character class swashes */
14880 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14881 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14883 for (i = 0; i < POSIX_CC_COUNT; i++) {
14884 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14886 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14887 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14888 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14889 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14890 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14891 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14892 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14893 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14894 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14895 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14896 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14897 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14898 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14899 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14900 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14901 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14903 if (proto_perl->Ipsig_pend) {
14904 Newxz(PL_psig_pend, SIG_SIZE, int);
14907 PL_psig_pend = (int*)NULL;
14910 if (proto_perl->Ipsig_name) {
14911 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14912 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14914 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14917 PL_psig_ptr = (SV**)NULL;
14918 PL_psig_name = (SV**)NULL;
14921 if (flags & CLONEf_COPY_STACKS) {
14922 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14923 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14924 PL_tmps_ix+1, param);
14926 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14927 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14928 Newxz(PL_markstack, i, I32);
14929 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14930 - proto_perl->Imarkstack);
14931 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14932 - proto_perl->Imarkstack);
14933 Copy(proto_perl->Imarkstack, PL_markstack,
14934 PL_markstack_ptr - PL_markstack + 1, I32);
14936 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14937 * NOTE: unlike the others! */
14938 Newxz(PL_scopestack, PL_scopestack_max, I32);
14939 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14942 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14943 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14945 /* reset stack AV to correct length before its duped via
14946 * PL_curstackinfo */
14947 AvFILLp(proto_perl->Icurstack) =
14948 proto_perl->Istack_sp - proto_perl->Istack_base;
14950 /* NOTE: si_dup() looks at PL_markstack */
14951 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14953 /* PL_curstack = PL_curstackinfo->si_stack; */
14954 PL_curstack = av_dup(proto_perl->Icurstack, param);
14955 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14957 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14958 PL_stack_base = AvARRAY(PL_curstack);
14959 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14960 - proto_perl->Istack_base);
14961 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14963 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14964 PL_savestack = ss_dup(proto_perl, param);
14968 ENTER; /* perl_destruct() wants to LEAVE; */
14971 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14972 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14974 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14975 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14976 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14977 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14978 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14979 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14981 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14983 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14984 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14985 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14987 PL_stashcache = newHV();
14989 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14990 proto_perl->Iwatchaddr);
14991 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14992 if (PL_debug && PL_watchaddr) {
14993 PerlIO_printf(Perl_debug_log,
14994 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14995 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14996 PTR2UV(PL_watchok));
14999 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15000 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15001 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15003 /* Call the ->CLONE method, if it exists, for each of the stashes
15004 identified by sv_dup() above.
15006 while(av_tindex(param->stashes) != -1) {
15007 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15008 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15009 if (cloner && GvCV(cloner)) {
15014 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15016 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15022 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15023 ptr_table_free(PL_ptr_table);
15024 PL_ptr_table = NULL;
15027 if (!(flags & CLONEf_COPY_STACKS)) {
15028 unreferenced_to_tmp_stack(param->unreferenced);
15031 SvREFCNT_dec(param->stashes);
15033 /* orphaned? eg threads->new inside BEGIN or use */
15034 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15035 SvREFCNT_inc_simple_void(PL_compcv);
15036 SAVEFREESV(PL_compcv);
15043 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15045 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15047 if (AvFILLp(unreferenced) > -1) {
15048 SV **svp = AvARRAY(unreferenced);
15049 SV **const last = svp + AvFILLp(unreferenced);
15053 if (SvREFCNT(*svp) == 1)
15055 } while (++svp <= last);
15057 EXTEND_MORTAL(count);
15058 svp = AvARRAY(unreferenced);
15061 if (SvREFCNT(*svp) == 1) {
15062 /* Our reference is the only one to this SV. This means that
15063 in this thread, the scalar effectively has a 0 reference.
15064 That doesn't work (cleanup never happens), so donate our
15065 reference to it onto the save stack. */
15066 PL_tmps_stack[++PL_tmps_ix] = *svp;
15068 /* As an optimisation, because we are already walking the
15069 entire array, instead of above doing either
15070 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15071 release our reference to the scalar, so that at the end of
15072 the array owns zero references to the scalars it happens to
15073 point to. We are effectively converting the array from
15074 AvREAL() on to AvREAL() off. This saves the av_clear()
15075 (triggered by the SvREFCNT_dec(unreferenced) below) from
15076 walking the array a second time. */
15077 SvREFCNT_dec(*svp);
15080 } while (++svp <= last);
15081 AvREAL_off(unreferenced);
15083 SvREFCNT_dec_NN(unreferenced);
15087 Perl_clone_params_del(CLONE_PARAMS *param)
15089 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15091 PerlInterpreter *const to = param->new_perl;
15093 PerlInterpreter *const was = PERL_GET_THX;
15095 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15101 SvREFCNT_dec(param->stashes);
15102 if (param->unreferenced)
15103 unreferenced_to_tmp_stack(param->unreferenced);
15113 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15116 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15117 does a dTHX; to get the context from thread local storage.
15118 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15119 a version that passes in my_perl. */
15120 PerlInterpreter *const was = PERL_GET_THX;
15121 CLONE_PARAMS *param;
15123 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15129 /* Given that we've set the context, we can do this unshared. */
15130 Newx(param, 1, CLONE_PARAMS);
15133 param->proto_perl = from;
15134 param->new_perl = to;
15135 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15136 AvREAL_off(param->stashes);
15137 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15145 #endif /* USE_ITHREADS */
15148 Perl_init_constants(pTHX)
15150 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15151 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15152 SvANY(&PL_sv_undef) = NULL;
15154 SvANY(&PL_sv_no) = new_XPVNV();
15155 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15156 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15157 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15160 SvANY(&PL_sv_yes) = new_XPVNV();
15161 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15162 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15163 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15166 SvPV_set(&PL_sv_no, (char*)PL_No);
15167 SvCUR_set(&PL_sv_no, 0);
15168 SvLEN_set(&PL_sv_no, 0);
15169 SvIV_set(&PL_sv_no, 0);
15170 SvNV_set(&PL_sv_no, 0);
15172 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15173 SvCUR_set(&PL_sv_yes, 1);
15174 SvLEN_set(&PL_sv_yes, 0);
15175 SvIV_set(&PL_sv_yes, 1);
15176 SvNV_set(&PL_sv_yes, 1);
15178 PadnamePV(&PL_padname_const) = (char *)PL_No;
15182 =head1 Unicode Support
15184 =for apidoc sv_recode_to_utf8
15186 The encoding is assumed to be an Encode object, on entry the PV
15187 of the sv is assumed to be octets in that encoding, and the sv
15188 will be converted into Unicode (and UTF-8).
15190 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15191 is not a reference, nothing is done to the sv. If the encoding is not
15192 an C<Encode::XS> Encoding object, bad things will happen.
15193 (See F<lib/encoding.pm> and L<Encode>.)
15195 The PV of the sv is returned.
15200 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15202 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15204 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15213 if (SvPADTMP(nsv)) {
15214 nsv = sv_newmortal();
15215 SvSetSV_nosteal(nsv, sv);
15223 Passing sv_yes is wrong - it needs to be or'ed set of constants
15224 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15225 remove converted chars from source.
15227 Both will default the value - let them.
15229 XPUSHs(&PL_sv_yes);
15232 call_method("decode", G_SCALAR);
15236 s = SvPV_const(uni, len);
15237 if (s != SvPVX_const(sv)) {
15238 SvGROW(sv, len + 1);
15239 Move(s, SvPVX(sv), len + 1, char);
15240 SvCUR_set(sv, len);
15245 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15246 /* clear pos and any utf8 cache */
15247 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15250 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15251 magic_setutf8(sv,mg); /* clear UTF8 cache */
15256 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15260 =for apidoc sv_cat_decode
15262 The encoding is assumed to be an Encode object, the PV of the ssv is
15263 assumed to be octets in that encoding and decoding the input starts
15264 from the position which (PV + *offset) pointed to. The dsv will be
15265 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
15266 when the string tstr appears in decoding output or the input ends on
15267 the PV of the ssv. The value which the offset points will be modified
15268 to the last input position on the ssv.
15270 Returns TRUE if the terminator was found, else returns FALSE.
15275 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15276 SV *ssv, int *offset, char *tstr, int tlen)
15280 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15282 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
15292 offsv = newSViv(*offset);
15294 mPUSHp(tstr, tlen);
15296 call_method("cat_decode", G_SCALAR);
15298 ret = SvTRUE(TOPs);
15299 *offset = SvIV(offsv);
15305 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15310 /* ---------------------------------------------------------------------
15312 * support functions for report_uninit()
15315 /* the maxiumum size of array or hash where we will scan looking
15316 * for the undefined element that triggered the warning */
15318 #define FUV_MAX_SEARCH_SIZE 1000
15320 /* Look for an entry in the hash whose value has the same SV as val;
15321 * If so, return a mortal copy of the key. */
15324 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15330 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15332 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15333 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15336 array = HvARRAY(hv);
15338 for (i=HvMAX(hv); i>=0; i--) {
15340 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15341 if (HeVAL(entry) != val)
15343 if ( HeVAL(entry) == &PL_sv_undef ||
15344 HeVAL(entry) == &PL_sv_placeholder)
15348 if (HeKLEN(entry) == HEf_SVKEY)
15349 return sv_mortalcopy(HeKEY_sv(entry));
15350 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15356 /* Look for an entry in the array whose value has the same SV as val;
15357 * If so, return the index, otherwise return -1. */
15360 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15362 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15364 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15365 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15368 if (val != &PL_sv_undef) {
15369 SV ** const svp = AvARRAY(av);
15372 for (i=AvFILLp(av); i>=0; i--)
15379 /* varname(): return the name of a variable, optionally with a subscript.
15380 * If gv is non-zero, use the name of that global, along with gvtype (one
15381 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15382 * targ. Depending on the value of the subscript_type flag, return:
15385 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15386 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15387 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15388 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15391 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15392 const SV *const keyname, I32 aindex, int subscript_type)
15395 SV * const name = sv_newmortal();
15396 if (gv && isGV(gv)) {
15398 buffer[0] = gvtype;
15401 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15403 gv_fullname4(name, gv, buffer, 0);
15405 if ((unsigned int)SvPVX(name)[1] <= 26) {
15407 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15409 /* Swap the 1 unprintable control character for the 2 byte pretty
15410 version - ie substr($name, 1, 1) = $buffer; */
15411 sv_insert(name, 1, 1, buffer, 2);
15415 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15418 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15420 if (!cv || !CvPADLIST(cv))
15422 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15423 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15427 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15428 SV * const sv = newSV(0);
15429 *SvPVX(name) = '$';
15430 Perl_sv_catpvf(aTHX_ name, "{%s}",
15431 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15432 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15433 SvREFCNT_dec_NN(sv);
15435 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15436 *SvPVX(name) = '$';
15437 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15439 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15440 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15441 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15449 =for apidoc find_uninit_var
15451 Find the name of the undefined variable (if any) that caused the operator
15452 to issue a "Use of uninitialized value" warning.
15453 If match is true, only return a name if its value matches uninit_sv.
15454 So roughly speaking, if a unary operator (such as OP_COS) generates a
15455 warning, then following the direct child of the op may yield an
15456 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15457 other hand, with OP_ADD there are two branches to follow, so we only print
15458 the variable name if we get an exact match.
15459 desc_p points to a string pointer holding the description of the op.
15460 This may be updated if needed.
15462 The name is returned as a mortal SV.
15464 Assumes that PL_op is the op that originally triggered the error, and that
15465 PL_comppad/PL_curpad points to the currently executing pad.
15471 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15472 bool match, const char **desc_p)
15477 const OP *o, *o2, *kid;
15479 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15481 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15482 uninit_sv == &PL_sv_placeholder)))
15485 switch (obase->op_type) {
15492 const bool pad = ( obase->op_type == OP_PADAV
15493 || obase->op_type == OP_PADHV
15494 || obase->op_type == OP_PADRANGE
15497 const bool hash = ( obase->op_type == OP_PADHV
15498 || obase->op_type == OP_RV2HV
15499 || (obase->op_type == OP_PADRANGE
15500 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15504 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15506 if (pad) { /* @lex, %lex */
15507 sv = PAD_SVl(obase->op_targ);
15511 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15512 /* @global, %global */
15513 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15516 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15518 else if (obase == PL_op) /* @{expr}, %{expr} */
15519 return find_uninit_var(cUNOPx(obase)->op_first,
15520 uninit_sv, match, desc_p);
15521 else /* @{expr}, %{expr} as a sub-expression */
15525 /* attempt to find a match within the aggregate */
15527 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15529 subscript_type = FUV_SUBSCRIPT_HASH;
15532 index = find_array_subscript((const AV *)sv, uninit_sv);
15534 subscript_type = FUV_SUBSCRIPT_ARRAY;
15537 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15540 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15541 keysv, index, subscript_type);
15545 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15547 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15548 if (!gv || !GvSTASH(gv))
15550 if (match && (GvSV(gv) != uninit_sv))
15552 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15555 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15558 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15560 return varname(NULL, '$', obase->op_targ,
15561 NULL, 0, FUV_SUBSCRIPT_NONE);
15564 gv = cGVOPx_gv(obase);
15565 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15567 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15569 case OP_AELEMFAST_LEX:
15572 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15573 if (!av || SvRMAGICAL(av))
15575 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15576 if (!svp || *svp != uninit_sv)
15579 return varname(NULL, '$', obase->op_targ,
15580 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15583 gv = cGVOPx_gv(obase);
15588 AV *const av = GvAV(gv);
15589 if (!av || SvRMAGICAL(av))
15591 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15592 if (!svp || *svp != uninit_sv)
15595 return varname(gv, '$', 0,
15596 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15598 NOT_REACHED; /* NOTREACHED */
15601 o = cUNOPx(obase)->op_first;
15602 if (!o || o->op_type != OP_NULL ||
15603 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15605 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15610 bool negate = FALSE;
15612 if (PL_op == obase)
15613 /* $a[uninit_expr] or $h{uninit_expr} */
15614 return find_uninit_var(cBINOPx(obase)->op_last,
15615 uninit_sv, match, desc_p);
15618 o = cBINOPx(obase)->op_first;
15619 kid = cBINOPx(obase)->op_last;
15621 /* get the av or hv, and optionally the gv */
15623 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15624 sv = PAD_SV(o->op_targ);
15626 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15627 && cUNOPo->op_first->op_type == OP_GV)
15629 gv = cGVOPx_gv(cUNOPo->op_first);
15633 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15638 if (kid && kid->op_type == OP_NEGATE) {
15640 kid = cUNOPx(kid)->op_first;
15643 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15644 /* index is constant */
15647 kidsv = newSVpvs_flags("-", SVs_TEMP);
15648 sv_catsv(kidsv, cSVOPx_sv(kid));
15651 kidsv = cSVOPx_sv(kid);
15655 if (obase->op_type == OP_HELEM) {
15656 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15657 if (!he || HeVAL(he) != uninit_sv)
15661 SV * const opsv = cSVOPx_sv(kid);
15662 const IV opsviv = SvIV(opsv);
15663 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15664 negate ? - opsviv : opsviv,
15666 if (!svp || *svp != uninit_sv)
15670 if (obase->op_type == OP_HELEM)
15671 return varname(gv, '%', o->op_targ,
15672 kidsv, 0, FUV_SUBSCRIPT_HASH);
15674 return varname(gv, '@', o->op_targ, NULL,
15675 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15676 FUV_SUBSCRIPT_ARRAY);
15679 /* index is an expression;
15680 * attempt to find a match within the aggregate */
15681 if (obase->op_type == OP_HELEM) {
15682 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15684 return varname(gv, '%', o->op_targ,
15685 keysv, 0, FUV_SUBSCRIPT_HASH);
15689 = find_array_subscript((const AV *)sv, uninit_sv);
15691 return varname(gv, '@', o->op_targ,
15692 NULL, index, FUV_SUBSCRIPT_ARRAY);
15697 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15699 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15701 NOT_REACHED; /* NOTREACHED */
15704 case OP_MULTIDEREF: {
15705 /* If we were executing OP_MULTIDEREF when the undef warning
15706 * triggered, then it must be one of the index values within
15707 * that triggered it. If not, then the only possibility is that
15708 * the value retrieved by the last aggregate lookup might be the
15709 * culprit. For the former, we set PL_multideref_pc each time before
15710 * using an index, so work though the item list until we reach
15711 * that point. For the latter, just work through the entire item
15712 * list; the last aggregate retrieved will be the candidate.
15715 /* the named aggregate, if any */
15716 PADOFFSET agg_targ = 0;
15718 /* the last-seen index */
15720 PADOFFSET index_targ;
15722 IV index_const_iv = 0; /* init for spurious compiler warn */
15723 SV *index_const_sv;
15724 int depth = 0; /* how many array/hash lookups we've done */
15726 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15727 UNOP_AUX_item *last = NULL;
15728 UV actions = items->uv;
15731 if (PL_op == obase) {
15732 last = PL_multideref_pc;
15733 assert(last >= items && last <= items + items[-1].uv);
15740 switch (actions & MDEREF_ACTION_MASK) {
15742 case MDEREF_reload:
15743 actions = (++items)->uv;
15746 case MDEREF_HV_padhv_helem: /* $lex{...} */
15749 case MDEREF_AV_padav_aelem: /* $lex[...] */
15750 agg_targ = (++items)->pad_offset;
15754 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15757 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15759 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15760 assert(isGV_with_GP(agg_gv));
15763 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15764 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15767 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15768 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15774 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15775 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15778 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15779 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15786 index_const_sv = NULL;
15788 index_type = (actions & MDEREF_INDEX_MASK);
15789 switch (index_type) {
15790 case MDEREF_INDEX_none:
15792 case MDEREF_INDEX_const:
15794 index_const_sv = UNOP_AUX_item_sv(++items)
15796 index_const_iv = (++items)->iv;
15798 case MDEREF_INDEX_padsv:
15799 index_targ = (++items)->pad_offset;
15801 case MDEREF_INDEX_gvsv:
15802 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15803 assert(isGV_with_GP(index_gv));
15807 if (index_type != MDEREF_INDEX_none)
15810 if ( index_type == MDEREF_INDEX_none
15811 || (actions & MDEREF_FLAG_last)
15812 || (last && items == last)
15816 actions >>= MDEREF_SHIFT;
15819 if (PL_op == obase) {
15820 /* index was undef */
15822 *desc_p = ( (actions & MDEREF_FLAG_last)
15823 && (obase->op_private
15824 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15826 (obase->op_private & OPpMULTIDEREF_EXISTS)
15829 : is_hv ? "hash element" : "array element";
15830 assert(index_type != MDEREF_INDEX_none);
15832 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15834 return varname(NULL, '$', index_targ,
15835 NULL, 0, FUV_SUBSCRIPT_NONE);
15836 assert(is_hv); /* AV index is an IV and can't be undef */
15837 /* can a const HV index ever be undef? */
15841 /* the SV returned by pp_multideref() was undef, if anything was */
15847 sv = PAD_SV(agg_targ);
15849 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15853 if (index_type == MDEREF_INDEX_const) {
15858 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15859 if (!he || HeVAL(he) != uninit_sv)
15863 SV * const * const svp =
15864 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15865 if (!svp || *svp != uninit_sv)
15870 ? varname(agg_gv, '%', agg_targ,
15871 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15872 : varname(agg_gv, '@', agg_targ,
15873 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15876 /* index is an var */
15878 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15880 return varname(agg_gv, '%', agg_targ,
15881 keysv, 0, FUV_SUBSCRIPT_HASH);
15885 = find_array_subscript((const AV *)sv, uninit_sv);
15887 return varname(agg_gv, '@', agg_targ,
15888 NULL, index, FUV_SUBSCRIPT_ARRAY);
15892 return varname(agg_gv,
15894 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15896 NOT_REACHED; /* NOTREACHED */
15900 /* only examine RHS */
15901 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
15905 o = cUNOPx(obase)->op_first;
15906 if ( o->op_type == OP_PUSHMARK
15907 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15911 if (!OpHAS_SIBLING(o)) {
15912 /* one-arg version of open is highly magical */
15914 if (o->op_type == OP_GV) { /* open FOO; */
15916 if (match && GvSV(gv) != uninit_sv)
15918 return varname(gv, '$', 0,
15919 NULL, 0, FUV_SUBSCRIPT_NONE);
15921 /* other possibilities not handled are:
15922 * open $x; or open my $x; should return '${*$x}'
15923 * open expr; should return '$'.expr ideally
15929 /* ops where $_ may be an implicit arg */
15934 if ( !(obase->op_flags & OPf_STACKED)) {
15935 if (uninit_sv == DEFSV)
15936 return newSVpvs_flags("$_", SVs_TEMP);
15937 else if (obase->op_targ
15938 && uninit_sv == PAD_SVl(obase->op_targ))
15939 return varname(NULL, '$', obase->op_targ, NULL, 0,
15940 FUV_SUBSCRIPT_NONE);
15947 match = 1; /* print etc can return undef on defined args */
15948 /* skip filehandle as it can't produce 'undef' warning */
15949 o = cUNOPx(obase)->op_first;
15950 if ((obase->op_flags & OPf_STACKED)
15952 ( o->op_type == OP_PUSHMARK
15953 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15954 o = OpSIBLING(OpSIBLING(o));
15958 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
15959 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
15961 /* the following ops are capable of returning PL_sv_undef even for
15962 * defined arg(s) */
15981 case OP_GETPEERNAME:
16029 case OP_SMARTMATCH:
16038 /* XXX tmp hack: these two may call an XS sub, and currently
16039 XS subs don't have a SUB entry on the context stack, so CV and
16040 pad determination goes wrong, and BAD things happen. So, just
16041 don't try to determine the value under those circumstances.
16042 Need a better fix at dome point. DAPM 11/2007 */
16048 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16049 if (gv && GvSV(gv) == uninit_sv)
16050 return newSVpvs_flags("$.", SVs_TEMP);
16055 /* def-ness of rval pos() is independent of the def-ness of its arg */
16056 if ( !(obase->op_flags & OPf_MOD))
16061 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16062 return newSVpvs_flags("${$/}", SVs_TEMP);
16067 if (!(obase->op_flags & OPf_KIDS))
16069 o = cUNOPx(obase)->op_first;
16075 /* This loop checks all the kid ops, skipping any that cannot pos-
16076 * sibly be responsible for the uninitialized value; i.e., defined
16077 * constants and ops that return nothing. If there is only one op
16078 * left that is not skipped, then we *know* it is responsible for
16079 * the uninitialized value. If there is more than one op left, we
16080 * have to look for an exact match in the while() loop below.
16081 * Note that we skip padrange, because the individual pad ops that
16082 * it replaced are still in the tree, so we work on them instead.
16085 for (kid=o; kid; kid = OpSIBLING(kid)) {
16086 const OPCODE type = kid->op_type;
16087 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16088 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16089 || (type == OP_PUSHMARK)
16090 || (type == OP_PADRANGE)
16094 if (o2) { /* more than one found */
16101 return find_uninit_var(o2, uninit_sv, match, desc_p);
16103 /* scan all args */
16105 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16117 =for apidoc report_uninit
16119 Print appropriate "Use of uninitialized variable" warning.
16125 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16128 SV* varname = NULL;
16131 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16134 if (uninit_sv && PL_curpad) {
16135 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16137 sv_insert(varname, 0, 0, " ", 1);
16139 /* PL_warn_uninit_sv is constant */
16140 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16141 /* diag_listed_as: Use of uninitialized value%s */
16142 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16143 SVfARG(varname ? varname : &PL_sv_no),
16148 /* PL_warn_uninit is constant */
16149 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16150 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16158 * c-indentation-style: bsd
16159 * c-basic-offset: 4
16160 * indent-tabs-mode: nil
16163 * ex: set ts=8 sts=4 sw=4 et: