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. The caller is expected to have handled
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);
5933 =for apidoc sv_get_backrefs
5935 If the sv is the target of a weakrefence then return
5936 the backrefs structure associated with the sv, otherwise
5939 When returning a non-null result the type of the return
5940 is relevant. If it is an AV then the contents of the AV
5941 are the weakrefs which point at this item. If it is any
5942 other type then the item itself is the weakref.
5944 See also Perl_sv_add_backref(), Perl_sv_del_backref(),
5945 Perl_sv_kill_backrefs()
5951 Perl_sv_get_backrefs(pTHX_ SV *const sv)
5955 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5957 /* find slot to store array or singleton backref */
5959 if (SvTYPE(sv) == SVt_PVHV) {
5961 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5962 backrefs = (SV *)iter->xhv_backreferences;
5964 } else if (SvMAGICAL(sv)) {
5965 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5967 backrefs = mg->mg_obj;
5972 /* Give tsv backref magic if it hasn't already got it, then push a
5973 * back-reference to sv onto the array associated with the backref magic.
5975 * As an optimisation, if there's only one backref and it's not an AV,
5976 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5977 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5981 /* A discussion about the backreferences array and its refcount:
5983 * The AV holding the backreferences is pointed to either as the mg_obj of
5984 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5985 * xhv_backreferences field. The array is created with a refcount
5986 * of 2. This means that if during global destruction the array gets
5987 * picked on before its parent to have its refcount decremented by the
5988 * random zapper, it won't actually be freed, meaning it's still there for
5989 * when its parent gets freed.
5991 * When the parent SV is freed, the extra ref is killed by
5992 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5993 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5995 * When a single backref SV is stored directly, it is not reference
6000 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6006 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6008 /* find slot to store array or singleton backref */
6010 if (SvTYPE(tsv) == SVt_PVHV) {
6011 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6014 mg = mg_find(tsv, PERL_MAGIC_backref);
6016 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6017 svp = &(mg->mg_obj);
6020 /* create or retrieve the array */
6022 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6023 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6027 mg->mg_flags |= MGf_REFCOUNTED;
6030 SvREFCNT_inc_simple_void_NN(av);
6031 /* av now has a refcnt of 2; see discussion above */
6032 av_extend(av, *svp ? 2 : 1);
6034 /* move single existing backref to the array */
6035 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6040 av = MUTABLE_AV(*svp);
6042 /* optimisation: store single backref directly in HvAUX or mg_obj */
6046 assert(SvTYPE(av) == SVt_PVAV);
6047 if (AvFILLp(av) >= AvMAX(av)) {
6048 av_extend(av, AvFILLp(av)+1);
6051 /* push new backref */
6052 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6055 /* delete a back-reference to ourselves from the backref magic associated
6056 * with the SV we point to.
6060 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6064 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6066 if (SvTYPE(tsv) == SVt_PVHV) {
6068 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6070 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6071 /* It's possible for the the last (strong) reference to tsv to have
6072 become freed *before* the last thing holding a weak reference.
6073 If both survive longer than the backreferences array, then when
6074 the referent's reference count drops to 0 and it is freed, it's
6075 not able to chase the backreferences, so they aren't NULLed.
6077 For example, a CV holds a weak reference to its stash. If both the
6078 CV and the stash survive longer than the backreferences array,
6079 and the CV gets picked for the SvBREAK() treatment first,
6080 *and* it turns out that the stash is only being kept alive because
6081 of an our variable in the pad of the CV, then midway during CV
6082 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6083 It ends up pointing to the freed HV. Hence it's chased in here, and
6084 if this block wasn't here, it would hit the !svp panic just below.
6086 I don't believe that "better" destruction ordering is going to help
6087 here - during global destruction there's always going to be the
6088 chance that something goes out of order. We've tried to make it
6089 foolproof before, and it only resulted in evolutionary pressure on
6090 fools. Which made us look foolish for our hubris. :-(
6096 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6097 svp = mg ? &(mg->mg_obj) : NULL;
6101 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6103 /* It's possible that sv is being freed recursively part way through the
6104 freeing of tsv. If this happens, the backreferences array of tsv has
6105 already been freed, and so svp will be NULL. If this is the case,
6106 we should not panic. Instead, nothing needs doing, so return. */
6107 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6109 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6110 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6113 if (SvTYPE(*svp) == SVt_PVAV) {
6117 AV * const av = (AV*)*svp;
6119 assert(!SvIS_FREED(av));
6123 /* for an SV with N weak references to it, if all those
6124 * weak refs are deleted, then sv_del_backref will be called
6125 * N times and O(N^2) compares will be done within the backref
6126 * array. To ameliorate this potential slowness, we:
6127 * 1) make sure this code is as tight as possible;
6128 * 2) when looking for SV, look for it at both the head and tail of the
6129 * array first before searching the rest, since some create/destroy
6130 * patterns will cause the backrefs to be freed in order.
6137 SV **p = &svp[fill];
6138 SV *const topsv = *p;
6145 /* We weren't the last entry.
6146 An unordered list has this property that you
6147 can take the last element off the end to fill
6148 the hole, and it's still an unordered list :-)
6154 break; /* should only be one */
6161 AvFILLp(av) = fill-1;
6163 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6164 /* freed AV; skip */
6167 /* optimisation: only a single backref, stored directly */
6169 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6170 (void*)*svp, (void*)sv);
6177 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6183 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6188 /* after multiple passes through Perl_sv_clean_all() for a thingy
6189 * that has badly leaked, the backref array may have gotten freed,
6190 * since we only protect it against 1 round of cleanup */
6191 if (SvIS_FREED(av)) {
6192 if (PL_in_clean_all) /* All is fair */
6195 "panic: magic_killbackrefs (freed backref AV/SV)");
6199 is_array = (SvTYPE(av) == SVt_PVAV);
6201 assert(!SvIS_FREED(av));
6204 last = svp + AvFILLp(av);
6207 /* optimisation: only a single backref, stored directly */
6213 while (svp <= last) {
6215 SV *const referrer = *svp;
6216 if (SvWEAKREF(referrer)) {
6217 /* XXX Should we check that it hasn't changed? */
6218 assert(SvROK(referrer));
6219 SvRV_set(referrer, 0);
6221 SvWEAKREF_off(referrer);
6222 SvSETMAGIC(referrer);
6223 } else if (SvTYPE(referrer) == SVt_PVGV ||
6224 SvTYPE(referrer) == SVt_PVLV) {
6225 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6226 /* You lookin' at me? */
6227 assert(GvSTASH(referrer));
6228 assert(GvSTASH(referrer) == (const HV *)sv);
6229 GvSTASH(referrer) = 0;
6230 } else if (SvTYPE(referrer) == SVt_PVCV ||
6231 SvTYPE(referrer) == SVt_PVFM) {
6232 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6233 /* You lookin' at me? */
6234 assert(CvSTASH(referrer));
6235 assert(CvSTASH(referrer) == (const HV *)sv);
6236 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6239 assert(SvTYPE(sv) == SVt_PVGV);
6240 /* You lookin' at me? */
6241 assert(CvGV(referrer));
6242 assert(CvGV(referrer) == (const GV *)sv);
6243 anonymise_cv_maybe(MUTABLE_GV(sv),
6244 MUTABLE_CV(referrer));
6249 "panic: magic_killbackrefs (flags=%"UVxf")",
6250 (UV)SvFLAGS(referrer));
6261 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6267 =for apidoc sv_insert
6269 Inserts a string at the specified offset/length within the SV. Similar to
6270 the Perl substr() function. Handles get magic.
6272 =for apidoc sv_insert_flags
6274 Same as C<sv_insert>, but the extra C<flags> are passed to the
6275 C<SvPV_force_flags> that applies to C<bigstr>.
6281 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6287 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6290 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6293 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6294 SvPV_force_flags(bigstr, curlen, flags);
6295 (void)SvPOK_only_UTF8(bigstr);
6296 if (offset + len > curlen) {
6297 SvGROW(bigstr, offset+len+1);
6298 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6299 SvCUR_set(bigstr, offset+len);
6303 i = littlelen - len;
6304 if (i > 0) { /* string might grow */
6305 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6306 mid = big + offset + len;
6307 midend = bigend = big + SvCUR(bigstr);
6310 while (midend > mid) /* shove everything down */
6311 *--bigend = *--midend;
6312 Move(little,big+offset,littlelen,char);
6313 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6318 Move(little,SvPVX(bigstr)+offset,len,char);
6323 big = SvPVX(bigstr);
6326 bigend = big + SvCUR(bigstr);
6328 if (midend > bigend)
6329 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6332 if (mid - big > bigend - midend) { /* faster to shorten from end */
6334 Move(little, mid, littlelen,char);
6337 i = bigend - midend;
6339 Move(midend, mid, i,char);
6343 SvCUR_set(bigstr, mid - big);
6345 else if ((i = mid - big)) { /* faster from front */
6346 midend -= littlelen;
6348 Move(big, midend - i, i, char);
6349 sv_chop(bigstr,midend-i);
6351 Move(little, mid, littlelen,char);
6353 else if (littlelen) {
6354 midend -= littlelen;
6355 sv_chop(bigstr,midend);
6356 Move(little,midend,littlelen,char);
6359 sv_chop(bigstr,midend);
6365 =for apidoc sv_replace
6367 Make the first argument a copy of the second, then delete the original.
6368 The target SV physically takes over ownership of the body of the source SV
6369 and inherits its flags; however, the target keeps any magic it owns,
6370 and any magic in the source is discarded.
6371 Note that this is a rather specialist SV copying operation; most of the
6372 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6378 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6380 const U32 refcnt = SvREFCNT(sv);
6382 PERL_ARGS_ASSERT_SV_REPLACE;
6384 SV_CHECK_THINKFIRST_COW_DROP(sv);
6385 if (SvREFCNT(nsv) != 1) {
6386 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6387 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6389 if (SvMAGICAL(sv)) {
6393 sv_upgrade(nsv, SVt_PVMG);
6394 SvMAGIC_set(nsv, SvMAGIC(sv));
6395 SvFLAGS(nsv) |= SvMAGICAL(sv);
6397 SvMAGIC_set(sv, NULL);
6401 assert(!SvREFCNT(sv));
6402 #ifdef DEBUG_LEAKING_SCALARS
6403 sv->sv_flags = nsv->sv_flags;
6404 sv->sv_any = nsv->sv_any;
6405 sv->sv_refcnt = nsv->sv_refcnt;
6406 sv->sv_u = nsv->sv_u;
6408 StructCopy(nsv,sv,SV);
6410 if(SvTYPE(sv) == SVt_IV) {
6411 SET_SVANY_FOR_BODYLESS_IV(sv);
6415 #ifdef PERL_OLD_COPY_ON_WRITE
6416 if (SvIsCOW_normal(nsv)) {
6417 /* We need to follow the pointers around the loop to make the
6418 previous SV point to sv, rather than nsv. */
6421 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6424 assert(SvPVX_const(current) == SvPVX_const(nsv));
6426 /* Make the SV before us point to the SV after us. */
6428 PerlIO_printf(Perl_debug_log, "previous is\n");
6430 PerlIO_printf(Perl_debug_log,
6431 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6432 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6434 SV_COW_NEXT_SV_SET(current, sv);
6437 SvREFCNT(sv) = refcnt;
6438 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6443 /* We're about to free a GV which has a CV that refers back to us.
6444 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6448 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6453 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6456 assert(SvREFCNT(gv) == 0);
6457 assert(isGV(gv) && isGV_with_GP(gv));
6459 assert(!CvANON(cv));
6460 assert(CvGV(cv) == gv);
6461 assert(!CvNAMED(cv));
6463 /* will the CV shortly be freed by gp_free() ? */
6464 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6465 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6469 /* if not, anonymise: */
6470 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6471 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6472 : newSVpvn_flags( "__ANON__", 8, 0 );
6473 sv_catpvs(gvname, "::__ANON__");
6474 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6475 SvREFCNT_dec_NN(gvname);
6479 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6484 =for apidoc sv_clear
6486 Clear an SV: call any destructors, free up any memory used by the body,
6487 and free the body itself. The SV's head is I<not> freed, although
6488 its type is set to all 1's so that it won't inadvertently be assumed
6489 to be live during global destruction etc.
6490 This function should only be called when REFCNT is zero. Most of the time
6491 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6498 Perl_sv_clear(pTHX_ SV *const orig_sv)
6503 const struct body_details *sv_type_details;
6509 PERL_ARGS_ASSERT_SV_CLEAR;
6511 /* within this loop, sv is the SV currently being freed, and
6512 * iter_sv is the most recent AV or whatever that's being iterated
6513 * over to provide more SVs */
6519 assert(SvREFCNT(sv) == 0);
6520 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6522 if (type <= SVt_IV) {
6523 /* See the comment in sv.h about the collusion between this
6524 * early return and the overloading of the NULL slots in the
6528 SvFLAGS(sv) &= SVf_BREAK;
6529 SvFLAGS(sv) |= SVTYPEMASK;
6533 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6534 for another purpose */
6535 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6537 if (type >= SVt_PVMG) {
6539 if (!curse(sv, 1)) goto get_next_sv;
6540 type = SvTYPE(sv); /* destructor may have changed it */
6542 /* Free back-references before magic, in case the magic calls
6543 * Perl code that has weak references to sv. */
6544 if (type == SVt_PVHV) {
6545 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6549 else if (SvMAGIC(sv)) {
6550 /* Free back-references before other types of magic. */
6551 sv_unmagic(sv, PERL_MAGIC_backref);
6557 /* case SVt_INVLIST: */
6560 IoIFP(sv) != PerlIO_stdin() &&
6561 IoIFP(sv) != PerlIO_stdout() &&
6562 IoIFP(sv) != PerlIO_stderr() &&
6563 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6565 io_close(MUTABLE_IO(sv), NULL, FALSE,
6566 (IoTYPE(sv) == IoTYPE_WRONLY ||
6567 IoTYPE(sv) == IoTYPE_RDWR ||
6568 IoTYPE(sv) == IoTYPE_APPEND));
6570 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6571 PerlDir_close(IoDIRP(sv));
6572 IoDIRP(sv) = (DIR*)NULL;
6573 Safefree(IoTOP_NAME(sv));
6574 Safefree(IoFMT_NAME(sv));
6575 Safefree(IoBOTTOM_NAME(sv));
6576 if ((const GV *)sv == PL_statgv)
6580 /* FIXME for plugins */
6582 pregfree2((REGEXP*) sv);
6586 cv_undef(MUTABLE_CV(sv));
6587 /* If we're in a stash, we don't own a reference to it.
6588 * However it does have a back reference to us, which needs to
6590 if ((stash = CvSTASH(sv)))
6591 sv_del_backref(MUTABLE_SV(stash), sv);
6594 if (PL_last_swash_hv == (const HV *)sv) {
6595 PL_last_swash_hv = NULL;
6597 if (HvTOTALKEYS((HV*)sv) > 0) {
6599 /* this statement should match the one at the beginning of
6600 * hv_undef_flags() */
6601 if ( PL_phase != PERL_PHASE_DESTRUCT
6602 && (name = HvNAME((HV*)sv)))
6604 if (PL_stashcache) {
6605 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6607 (void)hv_deletehek(PL_stashcache,
6608 HvNAME_HEK((HV*)sv), G_DISCARD);
6610 hv_name_set((HV*)sv, NULL, 0, 0);
6613 /* save old iter_sv in unused SvSTASH field */
6614 assert(!SvOBJECT(sv));
6615 SvSTASH(sv) = (HV*)iter_sv;
6618 /* save old hash_index in unused SvMAGIC field */
6619 assert(!SvMAGICAL(sv));
6620 assert(!SvMAGIC(sv));
6621 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6624 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6625 goto get_next_sv; /* process this new sv */
6627 /* free empty hash */
6628 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6629 assert(!HvARRAY((HV*)sv));
6633 AV* av = MUTABLE_AV(sv);
6634 if (PL_comppad == av) {
6638 if (AvREAL(av) && AvFILLp(av) > -1) {
6639 next_sv = AvARRAY(av)[AvFILLp(av)--];
6640 /* save old iter_sv in top-most slot of AV,
6641 * and pray that it doesn't get wiped in the meantime */
6642 AvARRAY(av)[AvMAX(av)] = iter_sv;
6644 goto get_next_sv; /* process this new sv */
6646 Safefree(AvALLOC(av));
6651 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6652 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6653 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6654 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6656 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6657 SvREFCNT_dec(LvTARG(sv));
6658 if (isREGEXP(sv)) goto freeregexp;
6660 if (isGV_with_GP(sv)) {
6661 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6662 && HvENAME_get(stash))
6663 mro_method_changed_in(stash);
6664 gp_free(MUTABLE_GV(sv));
6666 unshare_hek(GvNAME_HEK(sv));
6667 /* If we're in a stash, we don't own a reference to it.
6668 * However it does have a back reference to us, which
6669 * needs to be cleared. */
6670 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6671 sv_del_backref(MUTABLE_SV(stash), sv);
6673 /* FIXME. There are probably more unreferenced pointers to SVs
6674 * in the interpreter struct that we should check and tidy in
6675 * a similar fashion to this: */
6676 /* See also S_sv_unglob, which does the same thing. */
6677 if ((const GV *)sv == PL_last_in_gv)
6678 PL_last_in_gv = NULL;
6679 else if ((const GV *)sv == PL_statgv)
6681 else if ((const GV *)sv == PL_stderrgv)
6689 /* Don't bother with SvOOK_off(sv); as we're only going to
6693 SvOOK_offset(sv, offset);
6694 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6695 /* Don't even bother with turning off the OOK flag. */
6700 SV * const target = SvRV(sv);
6702 sv_del_backref(target, sv);
6708 else if (SvPVX_const(sv)
6709 && !(SvTYPE(sv) == SVt_PVIO
6710 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6714 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6718 # ifdef PERL_OLD_COPY_ON_WRITE
6719 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6721 if (CowREFCNT(sv)) {
6729 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6733 # ifdef PERL_OLD_COPY_ON_WRITE
6737 Safefree(SvPVX_mutable(sv));
6741 else if (SvPVX_const(sv) && SvLEN(sv)
6742 && !(SvTYPE(sv) == SVt_PVIO
6743 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6744 Safefree(SvPVX_mutable(sv));
6745 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6746 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6756 SvFLAGS(sv) &= SVf_BREAK;
6757 SvFLAGS(sv) |= SVTYPEMASK;
6759 sv_type_details = bodies_by_type + type;
6760 if (sv_type_details->arena) {
6761 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6762 &PL_body_roots[type]);
6764 else if (sv_type_details->body_size) {
6765 safefree(SvANY(sv));
6769 /* caller is responsible for freeing the head of the original sv */
6770 if (sv != orig_sv && !SvREFCNT(sv))
6773 /* grab and free next sv, if any */
6781 else if (!iter_sv) {
6783 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6784 AV *const av = (AV*)iter_sv;
6785 if (AvFILLp(av) > -1) {
6786 sv = AvARRAY(av)[AvFILLp(av)--];
6788 else { /* no more elements of current AV to free */
6791 /* restore previous value, squirrelled away */
6792 iter_sv = AvARRAY(av)[AvMAX(av)];
6793 Safefree(AvALLOC(av));
6796 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6797 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6798 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6799 /* no more elements of current HV to free */
6802 /* Restore previous values of iter_sv and hash_index,
6803 * squirrelled away */
6804 assert(!SvOBJECT(sv));
6805 iter_sv = (SV*)SvSTASH(sv);
6806 assert(!SvMAGICAL(sv));
6807 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6809 /* perl -DA does not like rubbish in SvMAGIC. */
6813 /* free any remaining detritus from the hash struct */
6814 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6815 assert(!HvARRAY((HV*)sv));
6820 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6824 if (!SvREFCNT(sv)) {
6828 if (--(SvREFCNT(sv)))
6832 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6833 "Attempt to free temp prematurely: SV 0x%"UVxf
6834 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6838 if (SvIMMORTAL(sv)) {
6839 /* make sure SvREFCNT(sv)==0 happens very seldom */
6840 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6849 /* This routine curses the sv itself, not the object referenced by sv. So
6850 sv does not have to be ROK. */
6853 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6854 PERL_ARGS_ASSERT_CURSE;
6855 assert(SvOBJECT(sv));
6857 if (PL_defstash && /* Still have a symbol table? */
6863 stash = SvSTASH(sv);
6864 assert(SvTYPE(stash) == SVt_PVHV);
6865 if (HvNAME(stash)) {
6866 CV* destructor = NULL;
6867 assert (SvOOK(stash));
6868 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6869 if (!destructor || HvMROMETA(stash)->destroy_gen
6870 != PL_sub_generation)
6873 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6874 if (gv) destructor = GvCV(gv);
6875 if (!SvOBJECT(stash))
6878 destructor ? (HV *)destructor : ((HV *)0)+1;
6879 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6883 assert(!destructor || destructor == ((CV *)0)+1
6884 || SvTYPE(destructor) == SVt_PVCV);
6885 if (destructor && destructor != ((CV *)0)+1
6886 /* A constant subroutine can have no side effects, so
6887 don't bother calling it. */
6888 && !CvCONST(destructor)
6889 /* Don't bother calling an empty destructor or one that
6890 returns immediately. */
6891 && (CvISXSUB(destructor)
6892 || (CvSTART(destructor)
6893 && (CvSTART(destructor)->op_next->op_type
6895 && (CvSTART(destructor)->op_next->op_type
6897 || CvSTART(destructor)->op_next->op_next->op_type
6903 SV* const tmpref = newRV(sv);
6904 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6906 PUSHSTACKi(PERLSI_DESTROY);
6911 call_sv(MUTABLE_SV(destructor),
6912 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6916 if(SvREFCNT(tmpref) < 2) {
6917 /* tmpref is not kept alive! */
6919 SvRV_set(tmpref, NULL);
6922 SvREFCNT_dec_NN(tmpref);
6925 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6928 if (check_refcnt && SvREFCNT(sv)) {
6929 if (PL_in_clean_objs)
6931 "DESTROY created new reference to dead object '%"HEKf"'",
6932 HEKfARG(HvNAME_HEK(stash)));
6933 /* DESTROY gave object new lease on life */
6939 HV * const stash = SvSTASH(sv);
6940 /* Curse before freeing the stash, as freeing the stash could cause
6941 a recursive call into S_curse. */
6942 SvOBJECT_off(sv); /* Curse the object. */
6943 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6944 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6950 =for apidoc sv_newref
6952 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6959 Perl_sv_newref(pTHX_ SV *const sv)
6961 PERL_UNUSED_CONTEXT;
6970 Decrement an SV's reference count, and if it drops to zero, call
6971 C<sv_clear> to invoke destructors and free up any memory used by
6972 the body; finally, deallocate the SV's head itself.
6973 Normally called via a wrapper macro C<SvREFCNT_dec>.
6979 Perl_sv_free(pTHX_ SV *const sv)
6985 /* Private helper function for SvREFCNT_dec().
6986 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6989 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6993 PERL_ARGS_ASSERT_SV_FREE2;
6995 if (LIKELY( rc == 1 )) {
7001 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7002 "Attempt to free temp prematurely: SV 0x%"UVxf
7003 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7007 if (SvIMMORTAL(sv)) {
7008 /* make sure SvREFCNT(sv)==0 happens very seldom */
7009 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7013 if (! SvREFCNT(sv)) /* may have have been resurrected */
7018 /* handle exceptional cases */
7022 if (SvFLAGS(sv) & SVf_BREAK)
7023 /* this SV's refcnt has been artificially decremented to
7024 * trigger cleanup */
7026 if (PL_in_clean_all) /* All is fair */
7028 if (SvIMMORTAL(sv)) {
7029 /* make sure SvREFCNT(sv)==0 happens very seldom */
7030 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7033 if (ckWARN_d(WARN_INTERNAL)) {
7034 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7035 Perl_dump_sv_child(aTHX_ sv);
7037 #ifdef DEBUG_LEAKING_SCALARS
7040 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7041 if (PL_warnhook == PERL_WARNHOOK_FATAL
7042 || ckDEAD(packWARN(WARN_INTERNAL))) {
7043 /* Don't let Perl_warner cause us to escape our fate: */
7047 /* This may not return: */
7048 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7049 "Attempt to free unreferenced scalar: SV 0x%"UVxf
7050 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7053 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7063 Returns the length of the string in the SV. Handles magic and type
7064 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
7065 gives raw access to the xpv_cur slot.
7071 Perl_sv_len(pTHX_ SV *const sv)
7078 (void)SvPV_const(sv, len);
7083 =for apidoc sv_len_utf8
7085 Returns the number of characters in the string in an SV, counting wide
7086 UTF-8 bytes as a single character. Handles magic and type coercion.
7092 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7093 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7094 * (Note that the mg_len is not the length of the mg_ptr field.
7095 * This allows the cache to store the character length of the string without
7096 * needing to malloc() extra storage to attach to the mg_ptr.)
7101 Perl_sv_len_utf8(pTHX_ SV *const sv)
7107 return sv_len_utf8_nomg(sv);
7111 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7114 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7116 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7118 if (PL_utf8cache && SvUTF8(sv)) {
7120 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7122 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7123 if (mg->mg_len != -1)
7126 /* We can use the offset cache for a headstart.
7127 The longer value is stored in the first pair. */
7128 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7130 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7134 if (PL_utf8cache < 0) {
7135 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7136 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7140 ulen = Perl_utf8_length(aTHX_ s, s + len);
7141 utf8_mg_len_cache_update(sv, &mg, ulen);
7145 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7148 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7151 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7152 STRLEN *const uoffset_p, bool *const at_end)
7154 const U8 *s = start;
7155 STRLEN uoffset = *uoffset_p;
7157 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7159 while (s < send && uoffset) {
7166 else if (s > send) {
7168 /* This is the existing behaviour. Possibly it should be a croak, as
7169 it's actually a bounds error */
7172 *uoffset_p -= uoffset;
7176 /* Given the length of the string in both bytes and UTF-8 characters, decide
7177 whether to walk forwards or backwards to find the byte corresponding to
7178 the passed in UTF-8 offset. */
7180 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7181 STRLEN uoffset, const STRLEN uend)
7183 STRLEN backw = uend - uoffset;
7185 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7187 if (uoffset < 2 * backw) {
7188 /* The assumption is that going forwards is twice the speed of going
7189 forward (that's where the 2 * backw comes from).
7190 (The real figure of course depends on the UTF-8 data.) */
7191 const U8 *s = start;
7193 while (s < send && uoffset--)
7203 while (UTF8_IS_CONTINUATION(*send))
7206 return send - start;
7209 /* For the string representation of the given scalar, find the byte
7210 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7211 give another position in the string, *before* the sought offset, which
7212 (which is always true, as 0, 0 is a valid pair of positions), which should
7213 help reduce the amount of linear searching.
7214 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7215 will be used to reduce the amount of linear searching. The cache will be
7216 created if necessary, and the found value offered to it for update. */
7218 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7219 const U8 *const send, STRLEN uoffset,
7220 STRLEN uoffset0, STRLEN boffset0)
7222 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7224 bool at_end = FALSE;
7226 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7228 assert (uoffset >= uoffset0);
7233 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7235 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7236 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7237 if ((*mgp)->mg_ptr) {
7238 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7239 if (cache[0] == uoffset) {
7240 /* An exact match. */
7243 if (cache[2] == uoffset) {
7244 /* An exact match. */
7248 if (cache[0] < uoffset) {
7249 /* The cache already knows part of the way. */
7250 if (cache[0] > uoffset0) {
7251 /* The cache knows more than the passed in pair */
7252 uoffset0 = cache[0];
7253 boffset0 = cache[1];
7255 if ((*mgp)->mg_len != -1) {
7256 /* And we know the end too. */
7258 + sv_pos_u2b_midway(start + boffset0, send,
7260 (*mgp)->mg_len - uoffset0);
7262 uoffset -= uoffset0;
7264 + sv_pos_u2b_forwards(start + boffset0,
7265 send, &uoffset, &at_end);
7266 uoffset += uoffset0;
7269 else if (cache[2] < uoffset) {
7270 /* We're between the two cache entries. */
7271 if (cache[2] > uoffset0) {
7272 /* and the cache knows more than the passed in pair */
7273 uoffset0 = cache[2];
7274 boffset0 = cache[3];
7278 + sv_pos_u2b_midway(start + boffset0,
7281 cache[0] - uoffset0);
7284 + sv_pos_u2b_midway(start + boffset0,
7287 cache[2] - uoffset0);
7291 else if ((*mgp)->mg_len != -1) {
7292 /* If we can take advantage of a passed in offset, do so. */
7293 /* In fact, offset0 is either 0, or less than offset, so don't
7294 need to worry about the other possibility. */
7296 + sv_pos_u2b_midway(start + boffset0, send,
7298 (*mgp)->mg_len - uoffset0);
7303 if (!found || PL_utf8cache < 0) {
7304 STRLEN real_boffset;
7305 uoffset -= uoffset0;
7306 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7307 send, &uoffset, &at_end);
7308 uoffset += uoffset0;
7310 if (found && PL_utf8cache < 0)
7311 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7313 boffset = real_boffset;
7316 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7318 utf8_mg_len_cache_update(sv, mgp, uoffset);
7320 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7327 =for apidoc sv_pos_u2b_flags
7329 Converts the offset from a count of UTF-8 chars from
7330 the start of the string, to a count of the equivalent number of bytes; if
7331 lenp is non-zero, it does the same to lenp, but this time starting from
7332 the offset, rather than from the start
7333 of the string. Handles type coercion.
7334 I<flags> is passed to C<SvPV_flags>, and usually should be
7335 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7341 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7342 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7343 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7348 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7355 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7357 start = (U8*)SvPV_flags(sv, len, flags);
7359 const U8 * const send = start + len;
7361 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7364 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7365 is 0, and *lenp is already set to that. */) {
7366 /* Convert the relative offset to absolute. */
7367 const STRLEN uoffset2 = uoffset + *lenp;
7368 const STRLEN boffset2
7369 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7370 uoffset, boffset) - boffset;
7384 =for apidoc sv_pos_u2b
7386 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7387 the start of the string, to a count of the equivalent number of bytes; if
7388 lenp is non-zero, it does the same to lenp, but this time starting from
7389 the offset, rather than from the start of the string. Handles magic and
7392 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7399 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7400 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7401 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7405 /* This function is subject to size and sign problems */
7408 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7410 PERL_ARGS_ASSERT_SV_POS_U2B;
7413 STRLEN ulen = (STRLEN)*lenp;
7414 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7415 SV_GMAGIC|SV_CONST_RETURN);
7418 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7419 SV_GMAGIC|SV_CONST_RETURN);
7424 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7427 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7428 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7431 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7432 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7433 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7437 (*mgp)->mg_len = ulen;
7440 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7441 byte length pairing. The (byte) length of the total SV is passed in too,
7442 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7443 may not have updated SvCUR, so we can't rely on reading it directly.
7445 The proffered utf8/byte length pairing isn't used if the cache already has
7446 two pairs, and swapping either for the proffered pair would increase the
7447 RMS of the intervals between known byte offsets.
7449 The cache itself consists of 4 STRLEN values
7450 0: larger UTF-8 offset
7451 1: corresponding byte offset
7452 2: smaller UTF-8 offset
7453 3: corresponding byte offset
7455 Unused cache pairs have the value 0, 0.
7456 Keeping the cache "backwards" means that the invariant of
7457 cache[0] >= cache[2] is maintained even with empty slots, which means that
7458 the code that uses it doesn't need to worry if only 1 entry has actually
7459 been set to non-zero. It also makes the "position beyond the end of the
7460 cache" logic much simpler, as the first slot is always the one to start
7464 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7465 const STRLEN utf8, const STRLEN blen)
7469 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7474 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7475 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7476 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7478 (*mgp)->mg_len = -1;
7482 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7483 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7484 (*mgp)->mg_ptr = (char *) cache;
7488 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7489 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7490 a pointer. Note that we no longer cache utf8 offsets on refer-
7491 ences, but this check is still a good idea, for robustness. */
7492 const U8 *start = (const U8 *) SvPVX_const(sv);
7493 const STRLEN realutf8 = utf8_length(start, start + byte);
7495 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7499 /* Cache is held with the later position first, to simplify the code
7500 that deals with unbounded ends. */
7502 ASSERT_UTF8_CACHE(cache);
7503 if (cache[1] == 0) {
7504 /* Cache is totally empty */
7507 } else if (cache[3] == 0) {
7508 if (byte > cache[1]) {
7509 /* New one is larger, so goes first. */
7510 cache[2] = cache[0];
7511 cache[3] = cache[1];
7519 /* float casts necessary? XXX */
7520 #define THREEWAY_SQUARE(a,b,c,d) \
7521 ((float)((d) - (c))) * ((float)((d) - (c))) \
7522 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7523 + ((float)((b) - (a))) * ((float)((b) - (a)))
7525 /* Cache has 2 slots in use, and we know three potential pairs.
7526 Keep the two that give the lowest RMS distance. Do the
7527 calculation in bytes simply because we always know the byte
7528 length. squareroot has the same ordering as the positive value,
7529 so don't bother with the actual square root. */
7530 if (byte > cache[1]) {
7531 /* New position is after the existing pair of pairs. */
7532 const float keep_earlier
7533 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7534 const float keep_later
7535 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7537 if (keep_later < keep_earlier) {
7538 cache[2] = cache[0];
7539 cache[3] = cache[1];
7545 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7546 float b, c, keep_earlier;
7547 if (byte > cache[3]) {
7548 /* New position is between the existing pair of pairs. */
7549 b = (float)cache[3];
7552 /* New position is before the existing pair of pairs. */
7554 c = (float)cache[3];
7556 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7557 if (byte > cache[3]) {
7558 if (keep_later < keep_earlier) {
7568 if (! (keep_later < keep_earlier)) {
7569 cache[0] = cache[2];
7570 cache[1] = cache[3];
7577 ASSERT_UTF8_CACHE(cache);
7580 /* We already know all of the way, now we may be able to walk back. The same
7581 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7582 backward is half the speed of walking forward. */
7584 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7585 const U8 *end, STRLEN endu)
7587 const STRLEN forw = target - s;
7588 STRLEN backw = end - target;
7590 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7592 if (forw < 2 * backw) {
7593 return utf8_length(s, target);
7596 while (end > target) {
7598 while (UTF8_IS_CONTINUATION(*end)) {
7607 =for apidoc sv_pos_b2u_flags
7609 Converts the offset from a count of bytes from the start of the string, to
7610 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7611 I<flags> is passed to C<SvPV_flags>, and usually should be
7612 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7618 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7619 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7624 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7627 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7633 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7635 s = (const U8*)SvPV_flags(sv, blen, flags);
7638 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7639 ", byte=%"UVuf, (UV)blen, (UV)offset);
7645 && SvTYPE(sv) >= SVt_PVMG
7646 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7649 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7650 if (cache[1] == offset) {
7651 /* An exact match. */
7654 if (cache[3] == offset) {
7655 /* An exact match. */
7659 if (cache[1] < offset) {
7660 /* We already know part of the way. */
7661 if (mg->mg_len != -1) {
7662 /* Actually, we know the end too. */
7664 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7665 s + blen, mg->mg_len - cache[0]);
7667 len = cache[0] + utf8_length(s + cache[1], send);
7670 else if (cache[3] < offset) {
7671 /* We're between the two cached pairs, so we do the calculation
7672 offset by the byte/utf-8 positions for the earlier pair,
7673 then add the utf-8 characters from the string start to
7675 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7676 s + cache[1], cache[0] - cache[2])
7680 else { /* cache[3] > offset */
7681 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7685 ASSERT_UTF8_CACHE(cache);
7687 } else if (mg->mg_len != -1) {
7688 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7692 if (!found || PL_utf8cache < 0) {
7693 const STRLEN real_len = utf8_length(s, send);
7695 if (found && PL_utf8cache < 0)
7696 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7702 utf8_mg_len_cache_update(sv, &mg, len);
7704 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7711 =for apidoc sv_pos_b2u
7713 Converts the value pointed to by offsetp from a count of bytes from the
7714 start of the string, to a count of the equivalent number of UTF-8 chars.
7715 Handles magic and type coercion.
7717 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7724 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7725 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7730 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7732 PERL_ARGS_ASSERT_SV_POS_B2U;
7737 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7738 SV_GMAGIC|SV_CONST_RETURN);
7742 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7743 STRLEN real, SV *const sv)
7745 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7747 /* As this is debugging only code, save space by keeping this test here,
7748 rather than inlining it in all the callers. */
7749 if (from_cache == real)
7752 /* Need to turn the assertions off otherwise we may recurse infinitely
7753 while printing error messages. */
7754 SAVEI8(PL_utf8cache);
7756 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7757 func, (UV) from_cache, (UV) real, SVfARG(sv));
7763 Returns a boolean indicating whether the strings in the two SVs are
7764 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7765 coerce its args to strings if necessary.
7767 =for apidoc sv_eq_flags
7769 Returns a boolean indicating whether the strings in the two SVs are
7770 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7771 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7777 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7784 SV* svrecode = NULL;
7791 /* if pv1 and pv2 are the same, second SvPV_const call may
7792 * invalidate pv1 (if we are handling magic), so we may need to
7794 if (sv1 == sv2 && flags & SV_GMAGIC
7795 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7796 pv1 = SvPV_const(sv1, cur1);
7797 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7799 pv1 = SvPV_flags_const(sv1, cur1, flags);
7807 pv2 = SvPV_flags_const(sv2, cur2, flags);
7809 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7810 /* Differing utf8ness.
7811 * Do not UTF8size the comparands as a side-effect. */
7814 svrecode = newSVpvn(pv2, cur2);
7815 sv_recode_to_utf8(svrecode, _get_encoding());
7816 pv2 = SvPV_const(svrecode, cur2);
7819 svrecode = newSVpvn(pv1, cur1);
7820 sv_recode_to_utf8(svrecode, _get_encoding());
7821 pv1 = SvPV_const(svrecode, cur1);
7823 /* Now both are in UTF-8. */
7825 SvREFCNT_dec_NN(svrecode);
7831 /* sv1 is the UTF-8 one */
7832 return bytes_cmp_utf8((const U8*)pv2, cur2,
7833 (const U8*)pv1, cur1) == 0;
7836 /* sv2 is the UTF-8 one */
7837 return bytes_cmp_utf8((const U8*)pv1, cur1,
7838 (const U8*)pv2, cur2) == 0;
7844 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7846 SvREFCNT_dec(svrecode);
7854 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7855 string in C<sv1> is less than, equal to, or greater than the string in
7856 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7857 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7859 =for apidoc sv_cmp_flags
7861 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7862 string in C<sv1> is less than, equal to, or greater than the string in
7863 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7864 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7865 also C<sv_cmp_locale_flags>.
7871 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7873 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7877 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7881 const char *pv1, *pv2;
7883 SV *svrecode = NULL;
7890 pv1 = SvPV_flags_const(sv1, cur1, flags);
7897 pv2 = SvPV_flags_const(sv2, cur2, flags);
7899 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7900 /* Differing utf8ness.
7901 * Do not UTF8size the comparands as a side-effect. */
7904 svrecode = newSVpvn(pv2, cur2);
7905 sv_recode_to_utf8(svrecode, _get_encoding());
7906 pv2 = SvPV_const(svrecode, cur2);
7909 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7910 (const U8*)pv1, cur1);
7911 return retval ? retval < 0 ? -1 : +1 : 0;
7916 svrecode = newSVpvn(pv1, cur1);
7917 sv_recode_to_utf8(svrecode, _get_encoding());
7918 pv1 = SvPV_const(svrecode, cur1);
7921 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7922 (const U8*)pv2, cur2);
7923 return retval ? retval < 0 ? -1 : +1 : 0;
7929 cmp = cur2 ? -1 : 0;
7933 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7936 cmp = retval < 0 ? -1 : 1;
7937 } else if (cur1 == cur2) {
7940 cmp = cur1 < cur2 ? -1 : 1;
7944 SvREFCNT_dec(svrecode);
7950 =for apidoc sv_cmp_locale
7952 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7953 'use bytes' aware, handles get magic, and will coerce its args to strings
7954 if necessary. See also C<sv_cmp>.
7956 =for apidoc sv_cmp_locale_flags
7958 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7959 'use bytes' aware and will coerce its args to strings if necessary. If the
7960 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7966 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7968 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7972 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7975 #ifdef USE_LOCALE_COLLATE
7981 if (PL_collation_standard)
7985 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7987 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7989 if (!pv1 || !len1) {
8000 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8003 return retval < 0 ? -1 : 1;
8006 * When the result of collation is equality, that doesn't mean
8007 * that there are no differences -- some locales exclude some
8008 * characters from consideration. So to avoid false equalities,
8009 * we use the raw string as a tiebreaker.
8016 PERL_UNUSED_ARG(flags);
8017 #endif /* USE_LOCALE_COLLATE */
8019 return sv_cmp(sv1, sv2);
8023 #ifdef USE_LOCALE_COLLATE
8026 =for apidoc sv_collxfrm
8028 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8029 C<sv_collxfrm_flags>.
8031 =for apidoc sv_collxfrm_flags
8033 Add Collate Transform magic to an SV if it doesn't already have it. If the
8034 flags contain SV_GMAGIC, it handles get-magic.
8036 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
8037 scalar data of the variable, but transformed to such a format that a normal
8038 memory comparison can be used to compare the data according to the locale
8045 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8049 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8051 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8052 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8058 Safefree(mg->mg_ptr);
8059 s = SvPV_flags_const(sv, len, flags);
8060 if ((xf = mem_collxfrm(s, len, &xlen))) {
8062 #ifdef PERL_OLD_COPY_ON_WRITE
8064 sv_force_normal_flags(sv, 0);
8066 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8080 if (mg && mg->mg_ptr) {
8082 return mg->mg_ptr + sizeof(PL_collation_ix);
8090 #endif /* USE_LOCALE_COLLATE */
8093 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8095 SV * const tsv = newSV(0);
8098 sv_gets(tsv, fp, 0);
8099 sv_utf8_upgrade_nomg(tsv);
8100 SvCUR_set(sv,append);
8103 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8107 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8110 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8111 /* Grab the size of the record we're getting */
8112 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8119 /* With a true, record-oriented file on VMS, we need to use read directly
8120 * to ensure that we respect RMS record boundaries. The user is responsible
8121 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8122 * record size) field. N.B. This is likely to produce invalid results on
8123 * varying-width character data when a record ends mid-character.
8125 fd = PerlIO_fileno(fp);
8127 && PerlLIO_fstat(fd, &st) == 0
8128 && (st.st_fab_rfm == FAB$C_VAR
8129 || st.st_fab_rfm == FAB$C_VFC
8130 || st.st_fab_rfm == FAB$C_FIX)) {
8132 bytesread = PerlLIO_read(fd, buffer, recsize);
8134 else /* in-memory file from PerlIO::Scalar
8135 * or not a record-oriented file
8139 bytesread = PerlIO_read(fp, buffer, recsize);
8141 /* At this point, the logic in sv_get() means that sv will
8142 be treated as utf-8 if the handle is utf8.
8144 if (PerlIO_isutf8(fp) && bytesread > 0) {
8145 char *bend = buffer + bytesread;
8146 char *bufp = buffer;
8147 size_t charcount = 0;
8148 bool charstart = TRUE;
8151 while (charcount < recsize) {
8152 /* count accumulated characters */
8153 while (bufp < bend) {
8155 skip = UTF8SKIP(bufp);
8157 if (bufp + skip > bend) {
8158 /* partial at the end */
8169 if (charcount < recsize) {
8171 STRLEN bufp_offset = bufp - buffer;
8172 SSize_t morebytesread;
8174 /* originally I read enough to fill any incomplete
8175 character and the first byte of the next
8176 character if needed, but if there's many
8177 multi-byte encoded characters we're going to be
8178 making a read call for every character beyond
8179 the original read size.
8181 So instead, read the rest of the character if
8182 any, and enough bytes to match at least the
8183 start bytes for each character we're going to
8187 readsize = recsize - charcount;
8189 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8190 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8191 bend = buffer + bytesread;
8192 morebytesread = PerlIO_read(fp, bend, readsize);
8193 if (morebytesread <= 0) {
8194 /* we're done, if we still have incomplete
8195 characters the check code in sv_gets() will
8198 I'd originally considered doing
8199 PerlIO_ungetc() on all but the lead
8200 character of the incomplete character, but
8201 read() doesn't do that, so I don't.
8206 /* prepare to scan some more */
8207 bytesread += morebytesread;
8208 bend = buffer + bytesread;
8209 bufp = buffer + bufp_offset;
8217 SvCUR_set(sv, bytesread + append);
8218 buffer[bytesread] = '\0';
8219 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8225 Get a line from the filehandle and store it into the SV, optionally
8226 appending to the currently-stored string. If C<append> is not 0, the
8227 line is appended to the SV instead of overwriting it. C<append> should
8228 be set to the byte offset that the appended string should start at
8229 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8235 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8245 PERL_ARGS_ASSERT_SV_GETS;
8247 if (SvTHINKFIRST(sv))
8248 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8249 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8251 However, perlbench says it's slower, because the existing swipe code
8252 is faster than copy on write.
8253 Swings and roundabouts. */
8254 SvUPGRADE(sv, SVt_PV);
8257 /* line is going to be appended to the existing buffer in the sv */
8258 if (PerlIO_isutf8(fp)) {
8260 sv_utf8_upgrade_nomg(sv);
8261 sv_pos_u2b(sv,&append,0);
8263 } else if (SvUTF8(sv)) {
8264 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8270 /* not appending - "clear" the string by setting SvCUR to 0,
8271 * the pv is still avaiable. */
8274 if (PerlIO_isutf8(fp))
8277 if (IN_PERL_COMPILETIME) {
8278 /* we always read code in line mode */
8282 else if (RsSNARF(PL_rs)) {
8283 /* If it is a regular disk file use size from stat() as estimate
8284 of amount we are going to read -- may result in mallocing
8285 more memory than we really need if the layers below reduce
8286 the size we read (e.g. CRLF or a gzip layer).
8289 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8290 const Off_t offset = PerlIO_tell(fp);
8291 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8292 #ifdef PERL_NEW_COPY_ON_WRITE
8293 /* Add an extra byte for the sake of copy-on-write's
8294 * buffer reference count. */
8295 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8297 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8304 else if (RsRECORD(PL_rs)) {
8305 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8307 else if (RsPARA(PL_rs)) {
8313 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8314 if (PerlIO_isutf8(fp)) {
8315 rsptr = SvPVutf8(PL_rs, rslen);
8318 if (SvUTF8(PL_rs)) {
8319 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8320 Perl_croak(aTHX_ "Wide character in $/");
8323 /* extract the raw pointer to the record separator */
8324 rsptr = SvPV_const(PL_rs, rslen);
8328 /* rslast is the last character in the record separator
8329 * note we don't use rslast except when rslen is true, so the
8330 * null assign is a placeholder. */
8331 rslast = rslen ? rsptr[rslen - 1] : '\0';
8333 if (rspara) { /* have to do this both before and after */
8334 do { /* to make sure file boundaries work right */
8337 i = PerlIO_getc(fp);
8341 PerlIO_ungetc(fp,i);
8347 /* See if we know enough about I/O mechanism to cheat it ! */
8349 /* This used to be #ifdef test - it is made run-time test for ease
8350 of abstracting out stdio interface. One call should be cheap
8351 enough here - and may even be a macro allowing compile
8355 if (PerlIO_fast_gets(fp)) {
8357 * We can do buffer based IO operations on this filehandle.
8359 * This means we can bypass a lot of subcalls and process
8360 * the buffer directly, it also means we know the upper bound
8361 * on the amount of data we might read of the current buffer
8362 * into our sv. Knowing this allows us to preallocate the pv
8363 * to be able to hold that maximum, which allows us to simplify
8364 * a lot of logic. */
8367 * We're going to steal some values from the stdio struct
8368 * and put EVERYTHING in the innermost loop into registers.
8370 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8371 STRLEN bpx; /* length of the data in the target sv
8372 used to fix pointers after a SvGROW */
8373 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8374 of data left in the read-ahead buffer.
8375 If 0 then the pv buffer can hold the full
8376 amount left, otherwise this is the amount it
8379 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8380 /* An ungetc()d char is handled separately from the regular
8381 * buffer, so we getc() it back out and stuff it in the buffer.
8383 i = PerlIO_getc(fp);
8384 if (i == EOF) return 0;
8385 *(--((*fp)->_ptr)) = (unsigned char) i;
8389 /* Here is some breathtakingly efficient cheating */
8391 /* When you read the following logic resist the urge to think
8392 * of record separators that are 1 byte long. They are an
8393 * uninteresting special (simple) case.
8395 * Instead think of record separators which are at least 2 bytes
8396 * long, and keep in mind that we need to deal with such
8397 * separators when they cross a read-ahead buffer boundary.
8399 * Also consider that we need to gracefully deal with separators
8400 * that may be longer than a single read ahead buffer.
8402 * Lastly do not forget we want to copy the delimiter as well. We
8403 * are copying all data in the file _up_to_and_including_ the separator
8406 * Now that you have all that in mind here is what is happening below:
8408 * 1. When we first enter the loop we do some memory book keeping to see
8409 * how much free space there is in the target SV. (This sub assumes that
8410 * it is operating on the same SV most of the time via $_ and that it is
8411 * going to be able to reuse the same pv buffer each call.) If there is
8412 * "enough" room then we set "shortbuffered" to how much space there is
8413 * and start reading forward.
8415 * 2. When we scan forward we copy from the read-ahead buffer to the target
8416 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8417 * and the end of the of pv, as well as for the "rslast", which is the last
8418 * char of the separator.
8420 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8421 * (which has a "complete" record up to the point we saw rslast) and check
8422 * it to see if it matches the separator. If it does we are done. If it doesn't
8423 * we continue on with the scan/copy.
8425 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8426 * the IO system to read the next buffer. We do this by doing a getc(), which
8427 * returns a single char read (or EOF), and prefills the buffer, and also
8428 * allows us to find out how full the buffer is. We use this information to
8429 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8430 * the returned single char into the target sv, and then go back into scan
8433 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8434 * remaining space in the read-buffer.
8436 * Note that this code despite its twisty-turny nature is pretty darn slick.
8437 * It manages single byte separators, multi-byte cross boundary separators,
8438 * and cross-read-buffer separators cleanly and efficiently at the cost
8439 * of potentially greatly overallocating the target SV.
8445 /* get the number of bytes remaining in the read-ahead buffer
8446 * on first call on a given fp this will return 0.*/
8447 cnt = PerlIO_get_cnt(fp);
8449 /* make sure we have the room */
8450 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8451 /* Not room for all of it
8452 if we are looking for a separator and room for some
8454 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8455 /* just process what we have room for */
8456 shortbuffered = cnt - SvLEN(sv) + append + 1;
8457 cnt -= shortbuffered;
8460 /* ensure that the target sv has enough room to hold
8461 * the rest of the read-ahead buffer */
8463 /* remember that cnt can be negative */
8464 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8468 /* we have enough room to hold the full buffer, lets scream */
8472 /* extract the pointer to sv's string buffer, offset by append as necessary */
8473 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8474 /* extract the point to the read-ahead buffer */
8475 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8477 /* some trace debug output */
8478 DEBUG_P(PerlIO_printf(Perl_debug_log,
8479 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8480 DEBUG_P(PerlIO_printf(Perl_debug_log,
8481 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8483 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8484 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8488 /* if there is stuff left in the read-ahead buffer */
8490 /* if there is a separator */
8492 /* loop until we hit the end of the read-ahead buffer */
8493 while (cnt > 0) { /* this | eat */
8494 /* scan forward copying and searching for rslast as we go */
8496 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8497 goto thats_all_folks; /* screams | sed :-) */
8501 /* no separator, slurp the full buffer */
8502 Copy(ptr, bp, cnt, char); /* this | eat */
8503 bp += cnt; /* screams | dust */
8504 ptr += cnt; /* louder | sed :-) */
8506 assert (!shortbuffered);
8507 goto cannot_be_shortbuffered;
8511 if (shortbuffered) { /* oh well, must extend */
8512 /* we didnt have enough room to fit the line into the target buffer
8513 * so we must extend the target buffer and keep going */
8514 cnt = shortbuffered;
8516 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8518 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8519 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8520 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8524 cannot_be_shortbuffered:
8525 /* we need to refill the read-ahead buffer if possible */
8527 DEBUG_P(PerlIO_printf(Perl_debug_log,
8528 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8529 PTR2UV(ptr),(IV)cnt));
8530 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8532 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8533 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8534 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8535 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8538 call PerlIO_getc() to let it prefill the lookahead buffer
8540 This used to call 'filbuf' in stdio form, but as that behaves like
8541 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8542 another abstraction.
8544 Note we have to deal with the char in 'i' if we are not at EOF
8546 i = PerlIO_getc(fp); /* get more characters */
8548 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8549 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8550 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8551 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8553 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8554 cnt = PerlIO_get_cnt(fp);
8555 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8556 DEBUG_P(PerlIO_printf(Perl_debug_log,
8557 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8558 PTR2UV(ptr),(IV)cnt));
8560 if (i == EOF) /* all done for ever? */
8561 goto thats_really_all_folks;
8563 /* make sure we have enough space in the target sv */
8564 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8566 SvGROW(sv, bpx + cnt + 2);
8567 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8569 /* copy of the char we got from getc() */
8570 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8572 /* make sure we deal with the i being the last character of a separator */
8573 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8574 goto thats_all_folks;
8578 /* check if we have actually found the separator - only really applies
8580 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8581 memNE((char*)bp - rslen, rsptr, rslen))
8582 goto screamer; /* go back to the fray */
8583 thats_really_all_folks:
8585 cnt += shortbuffered;
8586 DEBUG_P(PerlIO_printf(Perl_debug_log,
8587 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8588 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8589 DEBUG_P(PerlIO_printf(Perl_debug_log,
8590 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8592 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8593 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8595 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8596 DEBUG_P(PerlIO_printf(Perl_debug_log,
8597 "Screamer: done, len=%ld, string=|%.*s|\n",
8598 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8602 /*The big, slow, and stupid way. */
8603 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8604 STDCHAR *buf = NULL;
8605 Newx(buf, 8192, STDCHAR);
8613 const STDCHAR * const bpe = buf + sizeof(buf);
8615 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8616 ; /* keep reading */
8620 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8621 /* Accommodate broken VAXC compiler, which applies U8 cast to
8622 * both args of ?: operator, causing EOF to change into 255
8625 i = (U8)buf[cnt - 1];
8631 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8633 sv_catpvn_nomg(sv, (char *) buf, cnt);
8635 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8637 if (i != EOF && /* joy */
8639 SvCUR(sv) < rslen ||
8640 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8644 * If we're reading from a TTY and we get a short read,
8645 * indicating that the user hit his EOF character, we need
8646 * to notice it now, because if we try to read from the TTY
8647 * again, the EOF condition will disappear.
8649 * The comparison of cnt to sizeof(buf) is an optimization
8650 * that prevents unnecessary calls to feof().
8654 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8658 #ifdef USE_HEAP_INSTEAD_OF_STACK
8663 if (rspara) { /* have to do this both before and after */
8664 while (i != EOF) { /* to make sure file boundaries work right */
8665 i = PerlIO_getc(fp);
8667 PerlIO_ungetc(fp,i);
8673 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8679 Auto-increment of the value in the SV, doing string to numeric conversion
8680 if necessary. Handles 'get' magic and operator overloading.
8686 Perl_sv_inc(pTHX_ SV *const sv)
8695 =for apidoc sv_inc_nomg
8697 Auto-increment of the value in the SV, doing string to numeric conversion
8698 if necessary. Handles operator overloading. Skips handling 'get' magic.
8704 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8711 if (SvTHINKFIRST(sv)) {
8712 if (SvREADONLY(sv)) {
8713 Perl_croak_no_modify();
8717 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8719 i = PTR2IV(SvRV(sv));
8723 else sv_force_normal_flags(sv, 0);
8725 flags = SvFLAGS(sv);
8726 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8727 /* It's (privately or publicly) a float, but not tested as an
8728 integer, so test it to see. */
8730 flags = SvFLAGS(sv);
8732 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8733 /* It's publicly an integer, or privately an integer-not-float */
8734 #ifdef PERL_PRESERVE_IVUV
8738 if (SvUVX(sv) == UV_MAX)
8739 sv_setnv(sv, UV_MAX_P1);
8741 (void)SvIOK_only_UV(sv);
8742 SvUV_set(sv, SvUVX(sv) + 1);
8744 if (SvIVX(sv) == IV_MAX)
8745 sv_setuv(sv, (UV)IV_MAX + 1);
8747 (void)SvIOK_only(sv);
8748 SvIV_set(sv, SvIVX(sv) + 1);
8753 if (flags & SVp_NOK) {
8754 const NV was = SvNVX(sv);
8755 if (LIKELY(!Perl_isinfnan(was)) &&
8756 NV_OVERFLOWS_INTEGERS_AT &&
8757 was >= NV_OVERFLOWS_INTEGERS_AT) {
8758 /* diag_listed_as: Lost precision when %s %f by 1 */
8759 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8760 "Lost precision when incrementing %" NVff " by 1",
8763 (void)SvNOK_only(sv);
8764 SvNV_set(sv, was + 1.0);
8768 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8769 if ((flags & SVTYPEMASK) < SVt_PVIV)
8770 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8771 (void)SvIOK_only(sv);
8776 while (isALPHA(*d)) d++;
8777 while (isDIGIT(*d)) d++;
8778 if (d < SvEND(sv)) {
8779 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8780 #ifdef PERL_PRESERVE_IVUV
8781 /* Got to punt this as an integer if needs be, but we don't issue
8782 warnings. Probably ought to make the sv_iv_please() that does
8783 the conversion if possible, and silently. */
8784 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8785 /* Need to try really hard to see if it's an integer.
8786 9.22337203685478e+18 is an integer.
8787 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8788 so $a="9.22337203685478e+18"; $a+0; $a++
8789 needs to be the same as $a="9.22337203685478e+18"; $a++
8796 /* sv_2iv *should* have made this an NV */
8797 if (flags & SVp_NOK) {
8798 (void)SvNOK_only(sv);
8799 SvNV_set(sv, SvNVX(sv) + 1.0);
8802 /* I don't think we can get here. Maybe I should assert this
8803 And if we do get here I suspect that sv_setnv will croak. NWC
8805 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8806 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8808 #endif /* PERL_PRESERVE_IVUV */
8809 if (!numtype && ckWARN(WARN_NUMERIC))
8810 not_incrementable(sv);
8811 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8815 while (d >= SvPVX_const(sv)) {
8823 /* MKS: The original code here died if letters weren't consecutive.
8824 * at least it didn't have to worry about non-C locales. The
8825 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8826 * arranged in order (although not consecutively) and that only
8827 * [A-Za-z] are accepted by isALPHA in the C locale.
8829 if (isALPHA_FOLD_NE(*d, 'z')) {
8830 do { ++*d; } while (!isALPHA(*d));
8833 *(d--) -= 'z' - 'a';
8838 *(d--) -= 'z' - 'a' + 1;
8842 /* oh,oh, the number grew */
8843 SvGROW(sv, SvCUR(sv) + 2);
8844 SvCUR_set(sv, SvCUR(sv) + 1);
8845 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8856 Auto-decrement of the value in the SV, doing string to numeric conversion
8857 if necessary. Handles 'get' magic and operator overloading.
8863 Perl_sv_dec(pTHX_ SV *const sv)
8872 =for apidoc sv_dec_nomg
8874 Auto-decrement of the value in the SV, doing string to numeric conversion
8875 if necessary. Handles operator overloading. Skips handling 'get' magic.
8881 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8887 if (SvTHINKFIRST(sv)) {
8888 if (SvREADONLY(sv)) {
8889 Perl_croak_no_modify();
8893 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8895 i = PTR2IV(SvRV(sv));
8899 else sv_force_normal_flags(sv, 0);
8901 /* Unlike sv_inc we don't have to worry about string-never-numbers
8902 and keeping them magic. But we mustn't warn on punting */
8903 flags = SvFLAGS(sv);
8904 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8905 /* It's publicly an integer, or privately an integer-not-float */
8906 #ifdef PERL_PRESERVE_IVUV
8910 if (SvUVX(sv) == 0) {
8911 (void)SvIOK_only(sv);
8915 (void)SvIOK_only_UV(sv);
8916 SvUV_set(sv, SvUVX(sv) - 1);
8919 if (SvIVX(sv) == IV_MIN) {
8920 sv_setnv(sv, (NV)IV_MIN);
8924 (void)SvIOK_only(sv);
8925 SvIV_set(sv, SvIVX(sv) - 1);
8930 if (flags & SVp_NOK) {
8933 const NV was = SvNVX(sv);
8934 if (LIKELY(!Perl_isinfnan(was)) &&
8935 NV_OVERFLOWS_INTEGERS_AT &&
8936 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8937 /* diag_listed_as: Lost precision when %s %f by 1 */
8938 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8939 "Lost precision when decrementing %" NVff " by 1",
8942 (void)SvNOK_only(sv);
8943 SvNV_set(sv, was - 1.0);
8947 if (!(flags & SVp_POK)) {
8948 if ((flags & SVTYPEMASK) < SVt_PVIV)
8949 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8951 (void)SvIOK_only(sv);
8954 #ifdef PERL_PRESERVE_IVUV
8956 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8957 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8958 /* Need to try really hard to see if it's an integer.
8959 9.22337203685478e+18 is an integer.
8960 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8961 so $a="9.22337203685478e+18"; $a+0; $a--
8962 needs to be the same as $a="9.22337203685478e+18"; $a--
8969 /* sv_2iv *should* have made this an NV */
8970 if (flags & SVp_NOK) {
8971 (void)SvNOK_only(sv);
8972 SvNV_set(sv, SvNVX(sv) - 1.0);
8975 /* I don't think we can get here. Maybe I should assert this
8976 And if we do get here I suspect that sv_setnv will croak. NWC
8978 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8979 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8982 #endif /* PERL_PRESERVE_IVUV */
8983 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8986 /* this define is used to eliminate a chunk of duplicated but shared logic
8987 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8988 * used anywhere but here - yves
8990 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8992 SSize_t ix = ++PL_tmps_ix; \
8993 if (UNLIKELY(ix >= PL_tmps_max)) \
8994 ix = tmps_grow_p(ix); \
8995 PL_tmps_stack[ix] = (AnSv); \
8999 =for apidoc sv_mortalcopy
9001 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9002 The new SV is marked as mortal. It will be destroyed "soon", either by an
9003 explicit call to FREETMPS, or by an implicit call at places such as
9004 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
9009 /* Make a string that will exist for the duration of the expression
9010 * evaluation. Actually, it may have to last longer than that, but
9011 * hopefully we won't free it until it has been assigned to a
9012 * permanent location. */
9015 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9019 if (flags & SV_GMAGIC)
9020 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9022 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9023 PUSH_EXTEND_MORTAL__SV_C(sv);
9029 =for apidoc sv_newmortal
9031 Creates a new null SV which is mortal. The reference count of the SV is
9032 set to 1. It will be destroyed "soon", either by an explicit call to
9033 FREETMPS, or by an implicit call at places such as statement boundaries.
9034 See also C<sv_mortalcopy> and C<sv_2mortal>.
9040 Perl_sv_newmortal(pTHX)
9045 SvFLAGS(sv) = SVs_TEMP;
9046 PUSH_EXTEND_MORTAL__SV_C(sv);
9052 =for apidoc newSVpvn_flags
9054 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9055 characters) into it. The reference count for the
9056 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9057 string. You are responsible for ensuring that the source string is at least
9058 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9059 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9060 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9061 returning. If C<SVf_UTF8> is set, C<s>
9062 is considered to be in UTF-8 and the
9063 C<SVf_UTF8> flag will be set on the new SV.
9064 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9066 #define newSVpvn_utf8(s, len, u) \
9067 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9073 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9077 /* All the flags we don't support must be zero.
9078 And we're new code so I'm going to assert this from the start. */
9079 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9081 sv_setpvn(sv,s,len);
9083 /* This code used to do a sv_2mortal(), however we now unroll the call to
9084 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9085 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9086 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9087 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9088 * means that we eliminate quite a few steps than it looks - Yves
9089 * (explaining patch by gfx) */
9091 SvFLAGS(sv) |= flags;
9093 if(flags & SVs_TEMP){
9094 PUSH_EXTEND_MORTAL__SV_C(sv);
9101 =for apidoc sv_2mortal
9103 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9104 by an explicit call to FREETMPS, or by an implicit call at places such as
9105 statement boundaries. SvTEMP() is turned on which means that the SV's
9106 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9107 and C<sv_mortalcopy>.
9113 Perl_sv_2mortal(pTHX_ SV *const sv)
9120 PUSH_EXTEND_MORTAL__SV_C(sv);
9128 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9129 characters) into it. The reference count for the
9130 SV is set to 1. If C<len> is zero, Perl will compute the length using
9131 strlen(), (which means if you use this option, that C<s> can't have embedded
9132 C<NUL> characters and has to have a terminating C<NUL> byte).
9134 For efficiency, consider using C<newSVpvn> instead.
9140 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9145 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9150 =for apidoc newSVpvn
9152 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9153 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9154 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9155 are responsible for ensuring that the source buffer is at least
9156 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9163 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9167 sv_setpvn(sv,buffer,len);
9172 =for apidoc newSVhek
9174 Creates a new SV from the hash key structure. It will generate scalars that
9175 point to the shared string table where possible. Returns a new (undefined)
9176 SV if the hek is NULL.
9182 Perl_newSVhek(pTHX_ const HEK *const hek)
9191 if (HEK_LEN(hek) == HEf_SVKEY) {
9192 return newSVsv(*(SV**)HEK_KEY(hek));
9194 const int flags = HEK_FLAGS(hek);
9195 if (flags & HVhek_WASUTF8) {
9197 Andreas would like keys he put in as utf8 to come back as utf8
9199 STRLEN utf8_len = HEK_LEN(hek);
9200 SV * const sv = newSV_type(SVt_PV);
9201 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9202 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9203 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9206 } else if (flags & HVhek_UNSHARED) {
9207 /* A hash that isn't using shared hash keys has to have
9208 the flag in every key so that we know not to try to call
9209 share_hek_hek on it. */
9211 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9216 /* This will be overwhelminly the most common case. */
9218 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9219 more efficient than sharepvn(). */
9223 sv_upgrade(sv, SVt_PV);
9224 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9225 SvCUR_set(sv, HEK_LEN(hek));
9237 =for apidoc newSVpvn_share
9239 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9240 table. If the string does not already exist in the table, it is
9241 created first. Turns on the SvIsCOW flag (or READONLY
9242 and FAKE in 5.16 and earlier). If the C<hash> parameter
9243 is non-zero, that value is used; otherwise the hash is computed.
9244 The string's hash can later be retrieved from the SV
9245 with the C<SvSHARED_HASH()> macro. The idea here is
9246 that as the string table is used for shared hash keys these strings will have
9247 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9253 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9257 bool is_utf8 = FALSE;
9258 const char *const orig_src = src;
9261 STRLEN tmplen = -len;
9263 /* See the note in hv.c:hv_fetch() --jhi */
9264 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9268 PERL_HASH(hash, src, len);
9270 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9271 changes here, update it there too. */
9272 sv_upgrade(sv, SVt_PV);
9273 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9280 if (src != orig_src)
9286 =for apidoc newSVpv_share
9288 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9295 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9297 return newSVpvn_share(src, strlen(src), hash);
9300 #if defined(PERL_IMPLICIT_CONTEXT)
9302 /* pTHX_ magic can't cope with varargs, so this is a no-context
9303 * version of the main function, (which may itself be aliased to us).
9304 * Don't access this version directly.
9308 Perl_newSVpvf_nocontext(const char *const pat, ...)
9314 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9316 va_start(args, pat);
9317 sv = vnewSVpvf(pat, &args);
9324 =for apidoc newSVpvf
9326 Creates a new SV and initializes it with the string formatted like
9333 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9338 PERL_ARGS_ASSERT_NEWSVPVF;
9340 va_start(args, pat);
9341 sv = vnewSVpvf(pat, &args);
9346 /* backend for newSVpvf() and newSVpvf_nocontext() */
9349 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9353 PERL_ARGS_ASSERT_VNEWSVPVF;
9356 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9363 Creates a new SV and copies a floating point value into it.
9364 The reference count for the SV is set to 1.
9370 Perl_newSVnv(pTHX_ const NV n)
9382 Creates a new SV and copies an integer into it. The reference count for the
9389 Perl_newSViv(pTHX_ const IV i)
9395 /* Inlining ONLY the small relevant subset of sv_setiv here
9396 * for performance. Makes a significant difference. */
9398 /* We're starting from SVt_FIRST, so provided that's
9399 * actual 0, we don't have to unset any SV type flags
9400 * to promote to SVt_IV. */
9401 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9403 SET_SVANY_FOR_BODYLESS_IV(sv);
9404 SvFLAGS(sv) |= SVt_IV;
9416 Creates a new SV and copies an unsigned integer into it.
9417 The reference count for the SV is set to 1.
9423 Perl_newSVuv(pTHX_ const UV u)
9427 /* Inlining ONLY the small relevant subset of sv_setuv here
9428 * for performance. Makes a significant difference. */
9430 /* Using ivs is more efficient than using uvs - see sv_setuv */
9431 if (u <= (UV)IV_MAX) {
9432 return newSViv((IV)u);
9437 /* We're starting from SVt_FIRST, so provided that's
9438 * actual 0, we don't have to unset any SV type flags
9439 * to promote to SVt_IV. */
9440 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9442 SET_SVANY_FOR_BODYLESS_IV(sv);
9443 SvFLAGS(sv) |= SVt_IV;
9445 (void)SvIsUV_on(sv);
9454 =for apidoc newSV_type
9456 Creates a new SV, of the type specified. The reference count for the new SV
9463 Perl_newSV_type(pTHX_ const svtype type)
9468 ASSUME(SvTYPE(sv) == SVt_FIRST);
9469 if(type != SVt_FIRST)
9470 sv_upgrade(sv, type);
9475 =for apidoc newRV_noinc
9477 Creates an RV wrapper for an SV. The reference count for the original
9478 SV is B<not> incremented.
9484 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9488 PERL_ARGS_ASSERT_NEWRV_NOINC;
9492 /* We're starting from SVt_FIRST, so provided that's
9493 * actual 0, we don't have to unset any SV type flags
9494 * to promote to SVt_IV. */
9495 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9497 SET_SVANY_FOR_BODYLESS_IV(sv);
9498 SvFLAGS(sv) |= SVt_IV;
9503 SvRV_set(sv, tmpRef);
9508 /* newRV_inc is the official function name to use now.
9509 * newRV_inc is in fact #defined to newRV in sv.h
9513 Perl_newRV(pTHX_ SV *const sv)
9515 PERL_ARGS_ASSERT_NEWRV;
9517 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9523 Creates a new SV which is an exact duplicate of the original SV.
9530 Perl_newSVsv(pTHX_ SV *const old)
9536 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9537 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9540 /* Do this here, otherwise we leak the new SV if this croaks. */
9543 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9544 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9545 sv_setsv_flags(sv, old, SV_NOSTEAL);
9550 =for apidoc sv_reset
9552 Underlying implementation for the C<reset> Perl function.
9553 Note that the perl-level function is vaguely deprecated.
9559 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9561 PERL_ARGS_ASSERT_SV_RESET;
9563 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9567 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9569 char todo[PERL_UCHAR_MAX+1];
9572 if (!stash || SvTYPE(stash) != SVt_PVHV)
9575 if (!s) { /* reset ?? searches */
9576 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9578 const U32 count = mg->mg_len / sizeof(PMOP**);
9579 PMOP **pmp = (PMOP**) mg->mg_ptr;
9580 PMOP *const *const end = pmp + count;
9584 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9586 (*pmp)->op_pmflags &= ~PMf_USED;
9594 /* reset variables */
9596 if (!HvARRAY(stash))
9599 Zero(todo, 256, char);
9603 I32 i = (unsigned char)*s;
9607 max = (unsigned char)*s++;
9608 for ( ; i <= max; i++) {
9611 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9613 for (entry = HvARRAY(stash)[i];
9615 entry = HeNEXT(entry))
9620 if (!todo[(U8)*HeKEY(entry)])
9622 gv = MUTABLE_GV(HeVAL(entry));
9624 if (sv && !SvREADONLY(sv)) {
9625 SV_CHECK_THINKFIRST_COW_DROP(sv);
9626 if (!isGV(sv)) SvOK_off(sv);
9631 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9642 Using various gambits, try to get an IO from an SV: the IO slot if its a
9643 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9644 named after the PV if we're a string.
9646 'Get' magic is ignored on the sv passed in, but will be called on
9647 C<SvRV(sv)> if sv is an RV.
9653 Perl_sv_2io(pTHX_ SV *const sv)
9658 PERL_ARGS_ASSERT_SV_2IO;
9660 switch (SvTYPE(sv)) {
9662 io = MUTABLE_IO(sv);
9666 if (isGV_with_GP(sv)) {
9667 gv = MUTABLE_GV(sv);
9670 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9671 HEKfARG(GvNAME_HEK(gv)));
9677 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9679 SvGETMAGIC(SvRV(sv));
9680 return sv_2io(SvRV(sv));
9682 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9689 if (SvGMAGICAL(sv)) {
9690 newsv = sv_newmortal();
9691 sv_setsv_nomg(newsv, sv);
9693 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9703 Using various gambits, try to get a CV from an SV; in addition, try if
9704 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9705 The flags in C<lref> are passed to gv_fetchsv.
9711 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9716 PERL_ARGS_ASSERT_SV_2CV;
9723 switch (SvTYPE(sv)) {
9727 return MUTABLE_CV(sv);
9737 sv = amagic_deref_call(sv, to_cv_amg);
9740 if (SvTYPE(sv) == SVt_PVCV) {
9741 cv = MUTABLE_CV(sv);
9746 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9747 gv = MUTABLE_GV(sv);
9749 Perl_croak(aTHX_ "Not a subroutine reference");
9751 else if (isGV_with_GP(sv)) {
9752 gv = MUTABLE_GV(sv);
9755 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9762 /* Some flags to gv_fetchsv mean don't really create the GV */
9763 if (!isGV_with_GP(gv)) {
9768 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9769 /* XXX this is probably not what they think they're getting.
9770 * It has the same effect as "sub name;", i.e. just a forward
9781 Returns true if the SV has a true value by Perl's rules.
9782 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9783 instead use an in-line version.
9789 Perl_sv_true(pTHX_ SV *const sv)
9794 const XPV* const tXpv = (XPV*)SvANY(sv);
9796 (tXpv->xpv_cur > 1 ||
9797 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9804 return SvIVX(sv) != 0;
9807 return SvNVX(sv) != 0.0;
9809 return sv_2bool(sv);
9815 =for apidoc sv_pvn_force
9817 Get a sensible string out of the SV somehow.
9818 A private implementation of the C<SvPV_force> macro for compilers which
9819 can't cope with complex macro expressions. Always use the macro instead.
9821 =for apidoc sv_pvn_force_flags
9823 Get a sensible string out of the SV somehow.
9824 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9825 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9826 implemented in terms of this function.
9827 You normally want to use the various wrapper macros instead: see
9828 C<SvPV_force> and C<SvPV_force_nomg>
9834 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9836 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9838 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9839 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9840 sv_force_normal_flags(sv, 0);
9850 if (SvTYPE(sv) > SVt_PVLV
9851 || isGV_with_GP(sv))
9852 /* diag_listed_as: Can't coerce %s to %s in %s */
9853 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9855 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9862 if (SvTYPE(sv) < SVt_PV ||
9863 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9866 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9867 SvGROW(sv, len + 1);
9868 Move(s,SvPVX(sv),len,char);
9870 SvPVX(sv)[len] = '\0';
9873 SvPOK_on(sv); /* validate pointer */
9875 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9876 PTR2UV(sv),SvPVX_const(sv)));
9879 (void)SvPOK_only_UTF8(sv);
9880 return SvPVX_mutable(sv);
9884 =for apidoc sv_pvbyten_force
9886 The backend for the C<SvPVbytex_force> macro. Always use the macro
9893 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9895 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9897 sv_pvn_force(sv,lp);
9898 sv_utf8_downgrade(sv,0);
9904 =for apidoc sv_pvutf8n_force
9906 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9913 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9915 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9918 sv_utf8_upgrade_nomg(sv);
9924 =for apidoc sv_reftype
9926 Returns a string describing what the SV is a reference to.
9932 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9934 PERL_ARGS_ASSERT_SV_REFTYPE;
9935 if (ob && SvOBJECT(sv)) {
9936 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9939 /* WARNING - There is code, for instance in mg.c, that assumes that
9940 * the only reason that sv_reftype(sv,0) would return a string starting
9941 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9942 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9943 * this routine inside other subs, and it saves time.
9944 * Do not change this assumption without searching for "dodgy type check" in
9947 switch (SvTYPE(sv)) {
9962 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9963 /* tied lvalues should appear to be
9964 * scalars for backwards compatibility */
9965 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9966 ? "SCALAR" : "LVALUE");
9967 case SVt_PVAV: return "ARRAY";
9968 case SVt_PVHV: return "HASH";
9969 case SVt_PVCV: return "CODE";
9970 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9971 ? "GLOB" : "SCALAR");
9972 case SVt_PVFM: return "FORMAT";
9973 case SVt_PVIO: return "IO";
9974 case SVt_INVLIST: return "INVLIST";
9975 case SVt_REGEXP: return "REGEXP";
9976 default: return "UNKNOWN";
9984 Returns a SV describing what the SV passed in is a reference to.
9990 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9992 PERL_ARGS_ASSERT_SV_REF;
9995 dst = sv_newmortal();
9997 if (ob && SvOBJECT(sv)) {
9998 HvNAME_get(SvSTASH(sv))
9999 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10000 : sv_setpvn(dst, "__ANON__", 8);
10003 const char * reftype = sv_reftype(sv, 0);
10004 sv_setpv(dst, reftype);
10010 =for apidoc sv_isobject
10012 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10013 object. If the SV is not an RV, or if the object is not blessed, then this
10020 Perl_sv_isobject(pTHX_ SV *sv)
10036 Returns a boolean indicating whether the SV is blessed into the specified
10037 class. This does not check for subtypes; use C<sv_derived_from> to verify
10038 an inheritance relationship.
10044 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10046 const char *hvname;
10048 PERL_ARGS_ASSERT_SV_ISA;
10058 hvname = HvNAME_get(SvSTASH(sv));
10062 return strEQ(hvname, name);
10066 =for apidoc newSVrv
10068 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10069 RV then it will be upgraded to one. If C<classname> is non-null then the new
10070 SV will be blessed in the specified package. The new SV is returned and its
10071 reference count is 1. The reference count 1 is owned by C<rv>.
10077 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10081 PERL_ARGS_ASSERT_NEWSVRV;
10085 SV_CHECK_THINKFIRST_COW_DROP(rv);
10087 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10088 const U32 refcnt = SvREFCNT(rv);
10092 SvREFCNT(rv) = refcnt;
10094 sv_upgrade(rv, SVt_IV);
10095 } else if (SvROK(rv)) {
10096 SvREFCNT_dec(SvRV(rv));
10098 prepare_SV_for_RV(rv);
10106 HV* const stash = gv_stashpv(classname, GV_ADD);
10107 (void)sv_bless(rv, stash);
10113 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10115 SV * const lv = newSV_type(SVt_PVLV);
10116 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10118 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10119 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10120 LvSTARGOFF(lv) = ix;
10121 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10126 =for apidoc sv_setref_pv
10128 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10129 argument will be upgraded to an RV. That RV will be modified to point to
10130 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10131 into the SV. The C<classname> argument indicates the package for the
10132 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10133 will have a reference count of 1, and the RV will be returned.
10135 Do not use with other Perl types such as HV, AV, SV, CV, because those
10136 objects will become corrupted by the pointer copy process.
10138 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10144 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10146 PERL_ARGS_ASSERT_SV_SETREF_PV;
10149 sv_setsv(rv, &PL_sv_undef);
10153 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10158 =for apidoc sv_setref_iv
10160 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10161 argument will be upgraded to an RV. That RV will be modified to point to
10162 the new SV. The C<classname> argument indicates the package for the
10163 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10164 will have a reference count of 1, and the RV will be returned.
10170 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10172 PERL_ARGS_ASSERT_SV_SETREF_IV;
10174 sv_setiv(newSVrv(rv,classname), iv);
10179 =for apidoc sv_setref_uv
10181 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10182 argument will be upgraded to an RV. That RV will be modified to point to
10183 the new SV. The C<classname> argument indicates the package for the
10184 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10185 will have a reference count of 1, and the RV will be returned.
10191 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10193 PERL_ARGS_ASSERT_SV_SETREF_UV;
10195 sv_setuv(newSVrv(rv,classname), uv);
10200 =for apidoc sv_setref_nv
10202 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10203 argument will be upgraded to an RV. That RV will be modified to point to
10204 the new SV. The C<classname> argument indicates the package for the
10205 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10206 will have a reference count of 1, and the RV will be returned.
10212 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10214 PERL_ARGS_ASSERT_SV_SETREF_NV;
10216 sv_setnv(newSVrv(rv,classname), nv);
10221 =for apidoc sv_setref_pvn
10223 Copies a string into a new SV, optionally blessing the SV. The length of the
10224 string must be specified with C<n>. The C<rv> argument will be upgraded to
10225 an RV. That RV will be modified to point to the new SV. The C<classname>
10226 argument indicates the package for the blessing. Set C<classname> to
10227 C<NULL> to avoid the blessing. The new SV will have a reference count
10228 of 1, and the RV will be returned.
10230 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10236 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10237 const char *const pv, const STRLEN n)
10239 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10241 sv_setpvn(newSVrv(rv,classname), pv, n);
10246 =for apidoc sv_bless
10248 Blesses an SV into a specified package. The SV must be an RV. The package
10249 must be designated by its stash (see C<gv_stashpv()>). The reference count
10250 of the SV is unaffected.
10256 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10259 HV *oldstash = NULL;
10261 PERL_ARGS_ASSERT_SV_BLESS;
10265 Perl_croak(aTHX_ "Can't bless non-reference value");
10267 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10268 if (SvREADONLY(tmpRef))
10269 Perl_croak_no_modify();
10270 if (SvOBJECT(tmpRef)) {
10271 oldstash = SvSTASH(tmpRef);
10274 SvOBJECT_on(tmpRef);
10275 SvUPGRADE(tmpRef, SVt_PVMG);
10276 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10277 SvREFCNT_dec(oldstash);
10279 if(SvSMAGICAL(tmpRef))
10280 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10288 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10289 * as it is after unglobbing it.
10292 PERL_STATIC_INLINE void
10293 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10297 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10299 PERL_ARGS_ASSERT_SV_UNGLOB;
10301 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10303 if (!(flags & SV_COW_DROP_PV))
10304 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10306 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10308 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10309 && HvNAME_get(stash))
10310 mro_method_changed_in(stash);
10311 gp_free(MUTABLE_GV(sv));
10314 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10315 GvSTASH(sv) = NULL;
10318 if (GvNAME_HEK(sv)) {
10319 unshare_hek(GvNAME_HEK(sv));
10321 isGV_with_GP_off(sv);
10323 if(SvTYPE(sv) == SVt_PVGV) {
10324 /* need to keep SvANY(sv) in the right arena */
10325 xpvmg = new_XPVMG();
10326 StructCopy(SvANY(sv), xpvmg, XPVMG);
10327 del_XPVGV(SvANY(sv));
10330 SvFLAGS(sv) &= ~SVTYPEMASK;
10331 SvFLAGS(sv) |= SVt_PVMG;
10334 /* Intentionally not calling any local SET magic, as this isn't so much a
10335 set operation as merely an internal storage change. */
10336 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10337 else sv_setsv_flags(sv, temp, 0);
10339 if ((const GV *)sv == PL_last_in_gv)
10340 PL_last_in_gv = NULL;
10341 else if ((const GV *)sv == PL_statgv)
10346 =for apidoc sv_unref_flags
10348 Unsets the RV status of the SV, and decrements the reference count of
10349 whatever was being referenced by the RV. This can almost be thought of
10350 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10351 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10352 (otherwise the decrementing is conditional on the reference count being
10353 different from one or the reference being a readonly SV).
10360 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10362 SV* const target = SvRV(ref);
10364 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10366 if (SvWEAKREF(ref)) {
10367 sv_del_backref(target, ref);
10368 SvWEAKREF_off(ref);
10369 SvRV_set(ref, NULL);
10372 SvRV_set(ref, NULL);
10374 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10375 assigned to as BEGIN {$a = \"Foo"} will fail. */
10376 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10377 SvREFCNT_dec_NN(target);
10378 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10379 sv_2mortal(target); /* Schedule for freeing later */
10383 =for apidoc sv_untaint
10385 Untaint an SV. Use C<SvTAINTED_off> instead.
10391 Perl_sv_untaint(pTHX_ SV *const sv)
10393 PERL_ARGS_ASSERT_SV_UNTAINT;
10394 PERL_UNUSED_CONTEXT;
10396 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10397 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10404 =for apidoc sv_tainted
10406 Test an SV for taintedness. Use C<SvTAINTED> instead.
10412 Perl_sv_tainted(pTHX_ SV *const sv)
10414 PERL_ARGS_ASSERT_SV_TAINTED;
10415 PERL_UNUSED_CONTEXT;
10417 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10418 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10419 if (mg && (mg->mg_len & 1) )
10426 =for apidoc sv_setpviv
10428 Copies an integer into the given SV, also updating its string value.
10429 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10435 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10437 char buf[TYPE_CHARS(UV)];
10439 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10441 PERL_ARGS_ASSERT_SV_SETPVIV;
10443 sv_setpvn(sv, ptr, ebuf - ptr);
10447 =for apidoc sv_setpviv_mg
10449 Like C<sv_setpviv>, but also handles 'set' magic.
10455 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10457 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10459 sv_setpviv(sv, iv);
10463 #if defined(PERL_IMPLICIT_CONTEXT)
10465 /* pTHX_ magic can't cope with varargs, so this is a no-context
10466 * version of the main function, (which may itself be aliased to us).
10467 * Don't access this version directly.
10471 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10476 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10478 va_start(args, pat);
10479 sv_vsetpvf(sv, pat, &args);
10483 /* pTHX_ magic can't cope with varargs, so this is a no-context
10484 * version of the main function, (which may itself be aliased to us).
10485 * Don't access this version directly.
10489 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10494 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10496 va_start(args, pat);
10497 sv_vsetpvf_mg(sv, pat, &args);
10503 =for apidoc sv_setpvf
10505 Works like C<sv_catpvf> but copies the text into the SV instead of
10506 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10512 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10516 PERL_ARGS_ASSERT_SV_SETPVF;
10518 va_start(args, pat);
10519 sv_vsetpvf(sv, pat, &args);
10524 =for apidoc sv_vsetpvf
10526 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10527 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10529 Usually used via its frontend C<sv_setpvf>.
10535 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10537 PERL_ARGS_ASSERT_SV_VSETPVF;
10539 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10543 =for apidoc sv_setpvf_mg
10545 Like C<sv_setpvf>, but also handles 'set' magic.
10551 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10555 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10557 va_start(args, pat);
10558 sv_vsetpvf_mg(sv, pat, &args);
10563 =for apidoc sv_vsetpvf_mg
10565 Like C<sv_vsetpvf>, but also handles 'set' magic.
10567 Usually used via its frontend C<sv_setpvf_mg>.
10573 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10575 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10577 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10581 #if defined(PERL_IMPLICIT_CONTEXT)
10583 /* pTHX_ magic can't cope with varargs, so this is a no-context
10584 * version of the main function, (which may itself be aliased to us).
10585 * Don't access this version directly.
10589 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10594 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10596 va_start(args, pat);
10597 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10601 /* pTHX_ magic can't cope with varargs, so this is a no-context
10602 * version of the main function, (which may itself be aliased to us).
10603 * Don't access this version directly.
10607 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10612 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10614 va_start(args, pat);
10615 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10622 =for apidoc sv_catpvf
10624 Processes its arguments like C<sprintf> and appends the formatted
10625 output to an SV. If the appended data contains "wide" characters
10626 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10627 and characters >255 formatted with %c), the original SV might get
10628 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10629 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10630 valid UTF-8; if the original SV was bytes, the pattern should be too.
10635 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10639 PERL_ARGS_ASSERT_SV_CATPVF;
10641 va_start(args, pat);
10642 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10647 =for apidoc sv_vcatpvf
10649 Processes its arguments like C<vsprintf> and appends the formatted output
10650 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10652 Usually used via its frontend C<sv_catpvf>.
10658 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10660 PERL_ARGS_ASSERT_SV_VCATPVF;
10662 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10666 =for apidoc sv_catpvf_mg
10668 Like C<sv_catpvf>, but also handles 'set' magic.
10674 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10678 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10680 va_start(args, pat);
10681 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10687 =for apidoc sv_vcatpvf_mg
10689 Like C<sv_vcatpvf>, but also handles 'set' magic.
10691 Usually used via its frontend C<sv_catpvf_mg>.
10697 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10699 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10701 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10706 =for apidoc sv_vsetpvfn
10708 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10711 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10717 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10718 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10720 PERL_ARGS_ASSERT_SV_VSETPVFN;
10723 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10728 * Warn of missing argument to sprintf, and then return a defined value
10729 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10732 S_vcatpvfn_missing_argument(pTHX) {
10733 if (ckWARN(WARN_MISSING)) {
10734 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10735 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10742 S_expect_number(pTHX_ char **const pattern)
10746 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10748 switch (**pattern) {
10749 case '1': case '2': case '3':
10750 case '4': case '5': case '6':
10751 case '7': case '8': case '9':
10752 var = *(*pattern)++ - '0';
10753 while (isDIGIT(**pattern)) {
10754 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10756 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10764 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10766 const int neg = nv < 0;
10769 PERL_ARGS_ASSERT_F0CONVERT;
10771 if (UNLIKELY(Perl_isinfnan(nv))) {
10772 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len);
10782 if (uv & 1 && uv == nv)
10783 uv--; /* Round to even */
10785 const unsigned dig = uv % 10;
10787 } while (uv /= 10);
10798 =for apidoc sv_vcatpvfn
10800 =for apidoc sv_vcatpvfn_flags
10802 Processes its arguments like C<vsprintf> and appends the formatted output
10803 to an SV. Uses an array of SVs if the C style variable argument list is
10804 missing (NULL). When running with taint checks enabled, indicates via
10805 C<maybe_tainted> if results are untrustworthy (often due to the use of
10808 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10810 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10815 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10816 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10817 vec_utf8 = DO_UTF8(vecsv);
10819 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10822 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10823 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10825 PERL_ARGS_ASSERT_SV_VCATPVFN;
10827 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10830 #if DOUBLEKIND == DOUBLE_IS_IEEE_754_32_BIT_LITTLE_ENDIAN || \
10831 DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_LITTLE_ENDIAN || \
10832 DOUBLEKIND == DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10833 # define DOUBLE_LITTLE_ENDIAN
10836 #ifdef HAS_LONG_DOUBLEKIND
10838 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10839 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10840 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10841 # define LONGDOUBLE_LITTLE_ENDIAN
10844 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN || \
10845 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN || \
10846 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10847 # define LONGDOUBLE_BIG_ENDIAN
10850 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10851 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10852 # define LONGDOUBLE_X86_80_BIT
10855 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
10856 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10857 # define LONGDOUBLE_DOUBLEDOUBLE
10858 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10859 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10860 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10861 * after the first 1023 zero bits.
10863 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10864 * of dynamically growing buffer might be better, start at just 16 bytes
10865 * (for example) and grow only when necessary. Or maybe just by looking
10866 * at the exponents of the two doubles? */
10867 # define DOUBLEDOUBLE_MAXBITS 2098
10870 #endif /* HAS_LONG_DOUBLE */
10872 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10873 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10874 * per xdigit. For the double-double case, this can be rather many.
10875 * The non-double-double-long-double overshoots since all bits of NV
10876 * are not mantissa bits, there are also exponent bits. */
10877 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10878 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10880 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10883 /* If we do not have a known long double format, (including not using
10884 * long doubles, or long doubles being equal to doubles) then we will
10885 * fall back to the ldexp/frexp route, with which we can retrieve at
10886 * most as many bits as our widest unsigned integer type is. We try
10887 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10889 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10890 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10892 #if defined(HAS_QUAD) && defined(Uquad_t)
10893 # define MANTISSATYPE Uquad_t
10894 # define MANTISSASIZE 8
10896 # define MANTISSATYPE UV
10897 # define MANTISSASIZE UVSIZE
10900 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10901 # define HEXTRACT_LITTLE_ENDIAN
10902 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10903 # define HEXTRACT_BIG_ENDIAN
10905 # define HEXTRACT_MIX_ENDIAN
10908 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10909 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10910 * are being extracted from (either directly from the long double in-memory
10911 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10912 * is used to update the exponent. vhex is the pointer to the beginning
10913 * of the output buffer (of VHEX_SIZE).
10915 * The tricky part is that S_hextract() needs to be called twice:
10916 * the first time with vend as NULL, and the second time with vend as
10917 * the pointer returned by the first call. What happens is that on
10918 * the first round the output size is computed, and the intended
10919 * extraction sanity checked. On the second round the actual output
10920 * (the extraction of the hexadecimal values) takes place.
10921 * Sanity failures cause fatal failures during both rounds. */
10923 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10927 int ixmin = 0, ixmax = 0;
10929 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10930 * and elsewhere. */
10932 /* These macros are just to reduce typos, they have multiple
10933 * repetitions below, but usually only one (or sometimes two)
10934 * of them is really being used. */
10935 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10936 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10937 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10938 #define HEXTRACT_OUTPUT(ix) \
10940 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10942 #define HEXTRACT_COUNT(ix, c) \
10944 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10946 #define HEXTRACT_BYTE(ix) \
10948 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10950 #define HEXTRACT_LO_NYBBLE(ix) \
10952 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10954 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10955 * to make it look less odd when the top bits of a NV
10956 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10957 * order bits can be in the "low nybble" of a byte. */
10958 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10959 #define HEXTRACT_BYTES_LE(a, b) \
10960 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10961 #define HEXTRACT_BYTES_BE(a, b) \
10962 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10963 #define HEXTRACT_IMPLICIT_BIT(nv) \
10965 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10968 /* Most formats do. Those which don't should undef this. */
10969 #define HEXTRACT_HAS_IMPLICIT_BIT
10970 /* Many formats do. Those which don't should undef this. */
10971 #define HEXTRACT_HAS_TOP_NYBBLE
10973 /* HEXTRACTSIZE is the maximum number of xdigits. */
10974 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10975 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/4)
10977 # define HEXTRACTSIZE 2 * NVSIZE
10980 const U8* vmaxend = vhex + HEXTRACTSIZE;
10981 PERL_UNUSED_VAR(ix); /* might happen */
10982 (void)Perl_frexp(PERL_ABS(nv), exponent);
10983 if (vend && (vend <= vhex || vend > vmaxend))
10984 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10986 /* First check if using long doubles. */
10987 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10988 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10989 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10990 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10991 /* The bytes 13..0 are the mantissa/fraction,
10992 * the 15,14 are the sign+exponent. */
10993 const U8* nvp = (const U8*)(&nv);
10994 HEXTRACT_IMPLICIT_BIT(nv);
10995 # undef HEXTRACT_HAS_TOP_NYBBLE
10996 HEXTRACT_BYTES_LE(13, 0);
10997 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10998 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10999 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11000 /* The bytes 2..15 are the mantissa/fraction,
11001 * the 0,1 are the sign+exponent. */
11002 const U8* nvp = (const U8*)(&nv);
11003 HEXTRACT_IMPLICIT_BIT(nv);
11004 # undef HEXTRACT_HAS_TOP_NYBBLE
11005 HEXTRACT_BYTES_BE(2, 15);
11006 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11007 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11008 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
11009 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
11010 * meaning that 2 or 6 bytes are empty padding. */
11011 /* The bytes 7..0 are the mantissa/fraction */
11012 const U8* nvp = (const U8*)(&nv);
11013 # undef HEXTRACT_HAS_IMPLICIT_BIT
11014 # undef HEXTRACT_HAS_TOP_NYBBLE
11015 HEXTRACT_BYTES_LE(7, 0);
11016 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11017 /* Does this format ever happen? (Wikipedia says the Motorola
11018 * 6888x math coprocessors used format _like_ this but padded
11019 * to 96 bits with 16 unused bits between the exponent and the
11021 const U8* nvp = (const U8*)(&nv);
11022 # undef HEXTRACT_HAS_IMPLICIT_BIT
11023 # undef HEXTRACT_HAS_TOP_NYBBLE
11024 HEXTRACT_BYTES_BE(0, 7);
11026 # define HEXTRACT_FALLBACK
11027 /* Double-double format: two doubles next to each other.
11028 * The first double is the high-order one, exactly like
11029 * it would be for a "lone" double. The second double
11030 * is shifted down using the exponent so that that there
11031 * are no common bits. The tricky part is that the value
11032 * of the double-double is the SUM of the two doubles and
11033 * the second one can be also NEGATIVE.
11035 * Because of this tricky construction the bytewise extraction we
11036 * use for the other long double formats doesn't work, we must
11037 * extract the values bit by bit.
11039 * The little-endian double-double is used .. somewhere?
11041 * The big endian double-double is used in e.g. PPC/Power (AIX)
11044 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11045 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11046 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11049 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11050 /* Using normal doubles, not long doubles.
11052 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11053 * bytes, since we might need to handle printf precision, and
11054 * also need to insert the radix. */
11056 # ifdef HEXTRACT_LITTLE_ENDIAN
11057 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11058 const U8* nvp = (const U8*)(&nv);
11059 HEXTRACT_IMPLICIT_BIT(nv);
11060 HEXTRACT_TOP_NYBBLE(6);
11061 HEXTRACT_BYTES_LE(5, 0);
11062 # elif defined(HEXTRACT_BIG_ENDIAN)
11063 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11064 const U8* nvp = (const U8*)(&nv);
11065 HEXTRACT_IMPLICIT_BIT(nv);
11066 HEXTRACT_TOP_NYBBLE(1);
11067 HEXTRACT_BYTES_BE(2, 7);
11068 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11069 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11070 const U8* nvp = (const U8*)(&nv);
11071 HEXTRACT_IMPLICIT_BIT(nv);
11072 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11073 HEXTRACT_BYTE(1); /* 5 */
11074 HEXTRACT_BYTE(0); /* 4 */
11075 HEXTRACT_BYTE(7); /* 3 */
11076 HEXTRACT_BYTE(6); /* 2 */
11077 HEXTRACT_BYTE(5); /* 1 */
11078 HEXTRACT_BYTE(4); /* 0 */
11079 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11080 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11081 const U8* nvp = (const U8*)(&nv);
11082 HEXTRACT_IMPLICIT_BIT(nv);
11083 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11084 HEXTRACT_BYTE(6); /* 5 */
11085 HEXTRACT_BYTE(7); /* 4 */
11086 HEXTRACT_BYTE(0); /* 3 */
11087 HEXTRACT_BYTE(1); /* 2 */
11088 HEXTRACT_BYTE(2); /* 1 */
11089 HEXTRACT_BYTE(3); /* 0 */
11091 # define HEXTRACT_FALLBACK
11094 # define HEXTRACT_FALLBACK
11096 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11097 # ifdef HEXTRACT_FALLBACK
11098 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11099 /* The fallback is used for the double-double format, and
11100 * for unknown long double formats, and for unknown double
11101 * formats, or in general unknown NV formats. */
11102 if (nv == (NV)0.0) {
11110 NV d = nv < 0 ? -nv : nv;
11112 U8 ha = 0x0; /* hexvalue accumulator */
11113 U8 hd = 0x8; /* hexvalue digit */
11115 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11116 * this is essentially manual frexp(). Multiplying by 0.5 and
11117 * doubling should be lossless in binary floating point. */
11127 while (d >= e + e) {
11131 /* Now e <= d < 2*e */
11133 /* First extract the leading hexdigit (the implicit bit). */
11149 /* Then extract the remaining hexdigits. */
11150 while (d > (NV)0.0) {
11156 /* Output or count in groups of four bits,
11157 * that is, when the hexdigit is down to one. */
11162 /* Reset the hexvalue. */
11171 /* Flush possible pending hexvalue. */
11181 /* Croak for various reasons: if the output pointer escaped the
11182 * output buffer, if the extraction index escaped the extraction
11183 * buffer, or if the ending output pointer didn't match the
11184 * previously computed value. */
11185 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11186 /* For double-double the ixmin and ixmax stay at zero,
11187 * which is convenient since the HEXTRACTSIZE is tricky
11188 * for double-double. */
11189 ixmin < 0 || ixmax >= NVSIZE ||
11190 (vend && v != vend))
11191 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11196 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11197 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11202 const char *patend;
11205 static const char nullstr[] = "(null)";
11207 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11208 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11210 /* Times 4: a decimal digit takes more than 3 binary digits.
11211 * NV_DIG: mantissa takes than many decimal digits.
11212 * Plus 32: Playing safe. */
11213 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11214 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11215 bool hexfp = FALSE; /* hexadecimal floating point? */
11217 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
11219 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11220 PERL_UNUSED_ARG(maybe_tainted);
11222 if (flags & SV_GMAGIC)
11225 /* no matter what, this is a string now */
11226 (void)SvPV_force_nomg(sv, origlen);
11228 /* special-case "", "%s", and "%-p" (SVf - see below) */
11230 if (svmax && ckWARN(WARN_REDUNDANT))
11231 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11232 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11235 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11236 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11237 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11238 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11241 const char * const s = va_arg(*args, char*);
11242 sv_catpv_nomg(sv, s ? s : nullstr);
11244 else if (svix < svmax) {
11245 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11246 SvGETMAGIC(*svargs);
11247 sv_catsv_nomg(sv, *svargs);
11250 S_vcatpvfn_missing_argument(aTHX);
11253 if (args && patlen == 3 && pat[0] == '%' &&
11254 pat[1] == '-' && pat[2] == 'p') {
11255 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11256 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11257 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11258 argsv = MUTABLE_SV(va_arg(*args, void*));
11259 sv_catsv_nomg(sv, argsv);
11263 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11264 /* special-case "%.<number>[gf]" */
11265 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11266 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11267 unsigned digits = 0;
11271 while (*pp >= '0' && *pp <= '9')
11272 digits = 10 * digits + (*pp++ - '0');
11274 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11275 format the first argument and WARN_REDUNDANT if svmax > 1?
11276 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11277 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11278 const NV nv = SvNV(*svargs);
11279 if (LIKELY(!Perl_isinfnan(nv))) {
11281 /* Add check for digits != 0 because it seems that some
11282 gconverts are buggy in this case, and we don't yet have
11283 a Configure test for this. */
11284 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11285 /* 0, point, slack */
11286 STORE_LC_NUMERIC_SET_TO_NEEDED();
11287 SNPRINTF_G(nv, ebuf, size, digits);
11288 sv_catpv_nomg(sv, ebuf);
11289 if (*ebuf) /* May return an empty string for digits==0 */
11292 } else if (!digits) {
11295 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11296 sv_catpvn_nomg(sv, p, l);
11303 #endif /* !USE_LONG_DOUBLE */
11305 if (!args && svix < svmax && DO_UTF8(*svargs))
11308 patend = (char*)pat + patlen;
11309 for (p = (char*)pat; p < patend; p = q) {
11312 bool vectorize = FALSE;
11313 bool vectorarg = FALSE;
11314 bool vec_utf8 = FALSE;
11320 bool has_precis = FALSE;
11322 const I32 osvix = svix;
11323 bool is_utf8 = FALSE; /* is this item utf8? */
11324 #ifdef HAS_LDBL_SPRINTF_BUG
11325 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11326 with sfio - Allen <allens@cpan.org> */
11327 bool fix_ldbl_sprintf_bug = FALSE;
11331 U8 utf8buf[UTF8_MAXBYTES+1];
11332 STRLEN esignlen = 0;
11334 const char *eptr = NULL;
11335 const char *fmtstart;
11338 const U8 *vecstr = NULL;
11345 /* We need a long double target in case HAS_LONG_DOUBLE,
11346 * even without USE_LONG_DOUBLE, so that we can printf with
11347 * long double formats, even without NV being long double.
11348 * But we call the target 'fv' instead of 'nv', since most of
11349 * the time it is not (most compilers these days recognize
11350 * "long double", even if only as a synonym for "double").
11352 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11353 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11355 # ifdef Perl_isfinitel
11356 # define FV_ISFINITE(x) Perl_isfinitel(x)
11358 # define FV_GF PERL_PRIgldbl
11359 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11360 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11361 # define NV_TO_FV(nv,fv) STMT_START { \
11363 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11366 # define NV_TO_FV(nv,fv) (fv)=(nv)
11370 # define FV_GF NVgf
11371 # define NV_TO_FV(nv,fv) (fv)=(nv)
11373 #ifndef FV_ISFINITE
11374 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11379 const char *dotstr = ".";
11380 STRLEN dotstrlen = 1;
11381 I32 efix = 0; /* explicit format parameter index */
11382 I32 ewix = 0; /* explicit width index */
11383 I32 epix = 0; /* explicit precision index */
11384 I32 evix = 0; /* explicit vector index */
11385 bool asterisk = FALSE;
11386 bool infnan = FALSE;
11388 /* echo everything up to the next format specification */
11389 for (q = p; q < patend && *q != '%'; ++q) ;
11391 if (has_utf8 && !pat_utf8)
11392 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11394 sv_catpvn_nomg(sv, p, q - p);
11403 We allow format specification elements in this order:
11404 \d+\$ explicit format parameter index
11406 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11407 0 flag (as above): repeated to allow "v02"
11408 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11409 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11411 [%bcdefginopsuxDFOUX] format (mandatory)
11416 As of perl5.9.3, printf format checking is on by default.
11417 Internally, perl uses %p formats to provide an escape to
11418 some extended formatting. This block deals with those
11419 extensions: if it does not match, (char*)q is reset and
11420 the normal format processing code is used.
11422 Currently defined extensions are:
11423 %p include pointer address (standard)
11424 %-p (SVf) include an SV (previously %_)
11425 %-<num>p include an SV with precision <num>
11427 %3p include a HEK with precision of 256
11428 %4p char* preceded by utf8 flag and length
11429 %<num>p (where num is 1 or > 4) reserved for future
11432 Robin Barker 2005-07-14 (but modified since)
11434 %1p (VDf) removed. RMB 2007-10-19
11441 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11442 /* The argument has already gone through cBOOL, so the cast
11444 is_utf8 = (bool)va_arg(*args, int);
11445 elen = va_arg(*args, UV);
11446 if ((IV)elen < 0) {
11447 /* check if utf8 length is larger than 0 when cast to IV */
11448 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11449 elen= 0; /* otherwise we want to treat this as an empty string */
11451 eptr = va_arg(*args, char *);
11452 q += sizeof(UTF8f)-1;
11455 n = expect_number(&q);
11457 if (sv) { /* SVf */
11462 argsv = MUTABLE_SV(va_arg(*args, void*));
11463 eptr = SvPV_const(argsv, elen);
11464 if (DO_UTF8(argsv))
11468 else if (n==2 || n==3) { /* HEKf */
11469 HEK * const hek = va_arg(*args, HEK *);
11470 eptr = HEK_KEY(hek);
11471 elen = HEK_LEN(hek);
11472 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11473 if (n==3) precis = 256, has_precis = TRUE;
11477 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11478 "internal %%<num>p might conflict with future printf extensions");
11484 if ( (width = expect_number(&q)) ) {
11488 if (!no_redundant_warning)
11489 /* I've forgotten if it's a better
11490 micro-optimization to always set this or to
11491 only set it if it's unset */
11492 no_redundant_warning = TRUE;
11504 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11533 if ( (ewix = expect_number(&q)) )
11542 if ((vectorarg = asterisk)) {
11555 width = expect_number(&q);
11558 if (vectorize && vectorarg) {
11559 /* vectorizing, but not with the default "." */
11561 vecsv = va_arg(*args, SV*);
11563 vecsv = (evix > 0 && evix <= svmax)
11564 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11566 vecsv = svix < svmax
11567 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11569 dotstr = SvPV_const(vecsv, dotstrlen);
11570 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11571 bad with tied or overloaded values that return UTF8. */
11572 if (DO_UTF8(vecsv))
11574 else if (has_utf8) {
11575 vecsv = sv_mortalcopy(vecsv);
11576 sv_utf8_upgrade(vecsv);
11577 dotstr = SvPV_const(vecsv, dotstrlen);
11584 i = va_arg(*args, int);
11586 i = (ewix ? ewix <= svmax : svix < svmax) ?
11587 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11589 width = (i < 0) ? -i : i;
11599 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11601 /* XXX: todo, support specified precision parameter */
11605 i = va_arg(*args, int);
11607 i = (ewix ? ewix <= svmax : svix < svmax)
11608 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11610 has_precis = !(i < 0);
11614 while (isDIGIT(*q))
11615 precis = precis * 10 + (*q++ - '0');
11624 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11625 vecsv = svargs[efix ? efix-1 : svix++];
11626 vecstr = (U8*)SvPV_const(vecsv,veclen);
11627 vec_utf8 = DO_UTF8(vecsv);
11629 /* if this is a version object, we need to convert
11630 * back into v-string notation and then let the
11631 * vectorize happen normally
11633 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11634 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11635 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11636 "vector argument not supported with alpha versions");
11639 vecsv = sv_newmortal();
11640 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11642 vecstr = (U8*)SvPV_const(vecsv, veclen);
11643 vec_utf8 = DO_UTF8(vecsv);
11657 case 'I': /* Ix, I32x, and I64x */
11658 # ifdef USE_64_BIT_INT
11659 if (q[1] == '6' && q[2] == '4') {
11665 if (q[1] == '3' && q[2] == '2') {
11669 # ifdef USE_64_BIT_INT
11675 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11676 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11679 # ifdef USE_QUADMATH
11692 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11693 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11694 if (*q == 'l') { /* lld, llf */
11703 if (*++q == 'h') { /* hhd, hhu */
11732 if (!vectorize && !args) {
11734 const I32 i = efix-1;
11735 argsv = (i >= 0 && i < svmax)
11736 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11738 argsv = (svix >= 0 && svix < svmax)
11739 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11743 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11744 /* XXX va_arg(*args) case? need peek, use va_copy? */
11746 if (UNLIKELY(SvAMAGIC(argsv)))
11747 argsv = sv_2num(argsv);
11748 infnan = UNLIKELY(isinfnansv(argsv));
11751 switch (c = *q++) {
11759 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11760 /* no va_arg() case */
11761 SvNV_nomg(argsv), (int)c);
11762 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11764 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11766 eptr = (char*)utf8buf;
11767 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11781 eptr = va_arg(*args, char*);
11783 elen = strlen(eptr);
11785 eptr = (char *)nullstr;
11786 elen = sizeof nullstr - 1;
11790 eptr = SvPV_const(argsv, elen);
11791 if (DO_UTF8(argsv)) {
11792 STRLEN old_precis = precis;
11793 if (has_precis && precis < elen) {
11794 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11795 STRLEN p = precis > ulen ? ulen : precis;
11796 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11797 /* sticks at end */
11799 if (width) { /* fudge width (can't fudge elen) */
11800 if (has_precis && precis < elen)
11801 width += precis - old_precis;
11804 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11811 if (has_precis && precis < elen)
11819 goto floating_point;
11821 if (alt || vectorize)
11823 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11837 goto floating_point;
11844 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11853 esignbuf[esignlen++] = plus;
11857 case 'c': iv = (char)va_arg(*args, int); break;
11858 case 'h': iv = (short)va_arg(*args, int); break;
11859 case 'l': iv = va_arg(*args, long); break;
11860 case 'V': iv = va_arg(*args, IV); break;
11861 case 'z': iv = va_arg(*args, SSize_t); break;
11862 #ifdef HAS_PTRDIFF_T
11863 case 't': iv = va_arg(*args, ptrdiff_t); break;
11865 default: iv = va_arg(*args, int); break;
11867 case 'j': iv = va_arg(*args, intmax_t); break;
11871 iv = va_arg(*args, Quad_t); break;
11878 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11880 case 'c': iv = (char)tiv; break;
11881 case 'h': iv = (short)tiv; break;
11882 case 'l': iv = (long)tiv; break;
11884 default: iv = tiv; break;
11887 iv = (Quad_t)tiv; break;
11893 if ( !vectorize ) /* we already set uv above */
11898 esignbuf[esignlen++] = plus;
11902 esignbuf[esignlen++] = '-';
11941 goto floating_point;
11949 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11960 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11961 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11962 case 'l': uv = va_arg(*args, unsigned long); break;
11963 case 'V': uv = va_arg(*args, UV); break;
11964 case 'z': uv = va_arg(*args, Size_t); break;
11965 #ifdef HAS_PTRDIFF_T
11966 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11969 case 'j': uv = va_arg(*args, uintmax_t); break;
11971 default: uv = va_arg(*args, unsigned); break;
11974 uv = va_arg(*args, Uquad_t); break;
11981 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11983 case 'c': uv = (unsigned char)tuv; break;
11984 case 'h': uv = (unsigned short)tuv; break;
11985 case 'l': uv = (unsigned long)tuv; break;
11987 default: uv = tuv; break;
11990 uv = (Uquad_t)tuv; break;
11999 char *ptr = ebuf + sizeof ebuf;
12000 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12006 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12010 } while (uv >>= 4);
12012 esignbuf[esignlen++] = '0';
12013 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12019 *--ptr = '0' + dig;
12020 } while (uv >>= 3);
12021 if (alt && *ptr != '0')
12027 *--ptr = '0' + dig;
12028 } while (uv >>= 1);
12030 esignbuf[esignlen++] = '0';
12031 esignbuf[esignlen++] = c;
12034 default: /* it had better be ten or less */
12037 *--ptr = '0' + dig;
12038 } while (uv /= base);
12041 elen = (ebuf + sizeof ebuf) - ptr;
12045 zeros = precis - elen;
12046 else if (precis == 0 && elen == 1 && *eptr == '0'
12047 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12050 /* a precision nullifies the 0 flag. */
12057 /* FLOATING POINT */
12062 c = 'f'; /* maybe %F isn't supported here */
12064 case 'e': case 'E':
12066 case 'g': case 'G':
12067 case 'a': case 'A':
12071 /* This is evil, but floating point is even more evil */
12073 /* for SV-style calling, we can only get NV
12074 for C-style calling, we assume %f is double;
12075 for simplicity we allow any of %Lf, %llf, %qf for long double
12079 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12083 /* [perl #20339] - we should accept and ignore %lf rather than die */
12087 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12088 intsize = args ? 0 : 'q';
12092 #if defined(HAS_LONG_DOUBLE)
12105 /* Now we need (long double) if intsize == 'q', else (double). */
12107 /* Note: do not pull NVs off the va_list with va_arg()
12108 * (pull doubles instead) because if you have a build
12109 * with long doubles, you would always be pulling long
12110 * doubles, which would badly break anyone using only
12111 * doubles (i.e. the majority of builds). In other
12112 * words, you cannot mix doubles and long doubles.
12113 * The only case where you can pull off long doubles
12114 * is when the format specifier explicitly asks so with
12116 #ifdef USE_QUADMATH
12117 fv = intsize == 'q' ?
12118 va_arg(*args, NV) : va_arg(*args, double);
12119 #elif LONG_DOUBLESIZE > DOUBLESIZE
12120 if (intsize == 'q')
12121 fv = va_arg(*args, long double);
12123 NV_TO_FV(va_arg(*args, double), fv);
12125 fv = va_arg(*args, double);
12130 if (!infnan) SvGETMAGIC(argsv);
12131 NV_TO_FV(SvNV_nomg(argsv), fv);
12135 /* frexp() (or frexpl) has some unspecified behaviour for
12136 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12137 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12139 (void)Perl_frexp((NV)fv, &i);
12140 if (i == PERL_INT_MIN)
12141 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12142 /* Do not set hexfp earlier since we want to printf
12143 * Inf/NaN for Inf/NaN, not their hexfp. */
12144 hexfp = isALPHA_FOLD_EQ(c, 'a');
12145 if (UNLIKELY(hexfp)) {
12146 /* This seriously overshoots in most cases, but
12147 * better the undershooting. Firstly, all bytes
12148 * of the NV are not mantissa, some of them are
12149 * exponent. Secondly, for the reasonably common
12150 * long doubles case, the "80-bit extended", two
12151 * or six bytes of the NV are unused. */
12153 (fv < 0) ? 1 : 0 + /* possible unary minus */
12155 1 + /* the very unlikely carry */
12158 2 * NVSIZE + /* 2 hexdigits for each byte */
12160 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12162 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12163 /* However, for the "double double", we need more.
12164 * Since each double has their own exponent, the
12165 * doubles may float (haha) rather far from each
12166 * other, and the number of required bits is much
12167 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12168 * See the definition of DOUBLEDOUBLE_MAXBITS.
12170 * Need 2 hexdigits for each byte. */
12171 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12172 /* the size for the exponent already added */
12174 #ifdef USE_LOCALE_NUMERIC
12175 STORE_LC_NUMERIC_SET_TO_NEEDED();
12176 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12177 need += SvLEN(PL_numeric_radix_sv);
12178 RESTORE_LC_NUMERIC();
12182 need = BIT_DIGITS(i);
12183 } /* if i < 0, the number of digits is hard to predict. */
12185 need += has_precis ? precis : 6; /* known default */
12190 #ifdef HAS_LDBL_SPRINTF_BUG
12191 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12192 with sfio - Allen <allens@cpan.org> */
12195 # define MY_DBL_MAX DBL_MAX
12196 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12197 # if DOUBLESIZE >= 8
12198 # define MY_DBL_MAX 1.7976931348623157E+308L
12200 # define MY_DBL_MAX 3.40282347E+38L
12204 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12205 # define MY_DBL_MAX_BUG 1L
12207 # define MY_DBL_MAX_BUG MY_DBL_MAX
12211 # define MY_DBL_MIN DBL_MIN
12212 # else /* XXX guessing! -Allen */
12213 # if DOUBLESIZE >= 8
12214 # define MY_DBL_MIN 2.2250738585072014E-308L
12216 # define MY_DBL_MIN 1.17549435E-38L
12220 if ((intsize == 'q') && (c == 'f') &&
12221 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12222 (need < DBL_DIG)) {
12223 /* it's going to be short enough that
12224 * long double precision is not needed */
12226 if ((fv <= 0L) && (fv >= -0L))
12227 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12229 /* would use Perl_fp_class as a double-check but not
12230 * functional on IRIX - see perl.h comments */
12232 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12233 /* It's within the range that a double can represent */
12234 #if defined(DBL_MAX) && !defined(DBL_MIN)
12235 if ((fv >= ((long double)1/DBL_MAX)) ||
12236 (fv <= (-(long double)1/DBL_MAX)))
12238 fix_ldbl_sprintf_bug = TRUE;
12241 if (fix_ldbl_sprintf_bug == TRUE) {
12251 # undef MY_DBL_MAX_BUG
12254 #endif /* HAS_LDBL_SPRINTF_BUG */
12256 need += 20; /* fudge factor */
12257 if (PL_efloatsize < need) {
12258 Safefree(PL_efloatbuf);
12259 PL_efloatsize = need + 20; /* more fudge */
12260 Newx(PL_efloatbuf, PL_efloatsize, char);
12261 PL_efloatbuf[0] = '\0';
12264 if ( !(width || left || plus || alt) && fill != '0'
12265 && has_precis && intsize != 'q' /* Shortcuts */
12266 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12267 /* See earlier comment about buggy Gconvert when digits,
12269 if ( c == 'g' && precis ) {
12270 STORE_LC_NUMERIC_SET_TO_NEEDED();
12271 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12272 /* May return an empty string for digits==0 */
12273 if (*PL_efloatbuf) {
12274 elen = strlen(PL_efloatbuf);
12275 goto float_converted;
12277 } else if ( c == 'f' && !precis ) {
12278 if ((eptr = F0convert(fv, ebuf + sizeof ebuf, &elen)))
12283 if (UNLIKELY(hexfp)) {
12284 /* Hexadecimal floating point. */
12285 char* p = PL_efloatbuf;
12286 U8 vhex[VHEX_SIZE];
12287 U8* v = vhex; /* working pointer to vhex */
12288 U8* vend; /* pointer to one beyond last digit of vhex */
12289 U8* vfnz = NULL; /* first non-zero */
12290 const bool lower = (c == 'a');
12291 /* At output the values of vhex (up to vend) will
12292 * be mapped through the xdig to get the actual
12293 * human-readable xdigits. */
12294 const char* xdig = PL_hexdigit;
12295 int zerotail = 0; /* how many extra zeros to append */
12296 int exponent = 0; /* exponent of the floating point input */
12298 /* XXX: denormals, NaN, Inf.
12300 * For example with denormals, (assuming the vanilla
12301 * 64-bit double): the exponent is zero. 1xp-1074 is
12302 * the smallest denormal and the smallest double, it
12303 * should be output as 0x0.0000000000001p-1022 to
12304 * match its internal structure. */
12306 /* Note: fv can be (and often is) long double.
12307 * Here it is explicitly cast to NV. */
12308 vend = S_hextract(aTHX_ (NV)fv, &exponent, vhex, NULL);
12309 S_hextract(aTHX_ (NV)fv, &exponent, vhex, vend);
12311 #if NVSIZE > DOUBLESIZE
12312 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12313 /* In this case there is an implicit bit,
12314 * and therefore the exponent is shifted shift by one. */
12317 /* In this case there is no implicit bit,
12318 * and the exponent is shifted by the first xdigit. */
12333 xdig += 16; /* Use uppercase hex. */
12336 /* Find the first non-zero xdigit. */
12337 for (v = vhex; v < vend; v++) {
12345 U8* vlnz = NULL; /* The last non-zero. */
12347 /* Find the last non-zero xdigit. */
12348 for (v = vend - 1; v >= vhex; v--) {
12355 #if NVSIZE == DOUBLESIZE
12361 v = vhex + precis + 1;
12363 /* Round away from zero: if the tail
12364 * beyond the precis xdigits is equal to
12365 * or greater than 0x8000... */
12366 bool round = *v > 0x8;
12367 if (!round && *v == 0x8) {
12368 for (v++; v < vend; v++) {
12376 for (v = vhex + precis; v >= vhex; v--) {
12383 /* If the carry goes all the way to
12384 * the front, we need to output
12385 * a single '1'. This goes against
12386 * the "xdigit and then radix"
12387 * but since this is "cannot happen"
12388 * category, that is probably good. */
12393 /* The new effective "last non zero". */
12394 vlnz = vhex + precis;
12397 zerotail = precis - (vlnz - vhex);
12404 /* The radix is always output after the first
12405 * non-zero xdigit, or if alt. */
12406 if (vfnz < vlnz || alt) {
12407 #ifndef USE_LOCALE_NUMERIC
12410 STORE_LC_NUMERIC_SET_TO_NEEDED();
12411 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12413 const char* r = SvPV(PL_numeric_radix_sv, n);
12414 Copy(r, p, n, char);
12420 RESTORE_LC_NUMERIC();
12435 elen = p - PL_efloatbuf;
12436 elen += my_snprintf(p, PL_efloatsize - elen,
12437 "%c%+d", lower ? 'p' : 'P',
12440 if (elen < width) {
12442 /* Pad the back with spaces. */
12443 memset(PL_efloatbuf + elen, ' ', width - elen);
12445 else if (fill == '0') {
12446 /* Insert the zeros between the "0x" and
12447 * the digits, otherwise we end up with
12449 STRLEN nzero = width - elen;
12450 char* zerox = PL_efloatbuf + 2;
12451 Move(zerox, zerox + nzero, elen - 2, char);
12452 memset(zerox, fill, nzero);
12455 /* Move it to the right. */
12456 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12458 /* Pad the front with spaces. */
12459 memset(PL_efloatbuf, ' ', width - elen);
12465 elen = S_infnan_2pv(fv, PL_efloatbuf, PL_efloatsize);
12468 char *ptr = ebuf + sizeof ebuf;
12471 #if defined(USE_QUADMATH)
12472 if (intsize == 'q') {
12476 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12477 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12478 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12479 * not USE_LONG_DOUBLE and NVff. In other words,
12480 * this needs to work without USE_LONG_DOUBLE. */
12481 if (intsize == 'q') {
12482 /* Copy the one or more characters in a long double
12483 * format before the 'base' ([efgEFG]) character to
12484 * the format string. */
12485 static char const ldblf[] = PERL_PRIfldbl;
12486 char const *p = ldblf + sizeof(ldblf) - 3;
12487 while (p >= ldblf) { *--ptr = *p--; }
12492 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12497 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12509 /* No taint. Otherwise we are in the strange situation
12510 * where printf() taints but print($float) doesn't.
12513 STORE_LC_NUMERIC_SET_TO_NEEDED();
12515 /* hopefully the above makes ptr a very constrained format
12516 * that is safe to use, even though it's not literal */
12517 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12518 #ifdef USE_QUADMATH
12520 const char* qfmt = quadmath_format_single(ptr);
12522 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12523 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12525 if ((IV)elen == -1)
12526 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12530 #elif defined(HAS_LONG_DOUBLE)
12531 elen = ((intsize == 'q')
12532 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12533 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12535 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12541 eptr = PL_efloatbuf;
12542 assert((IV)elen > 0); /* here zero elen is bad */
12544 #ifdef USE_LOCALE_NUMERIC
12545 /* If the decimal point character in the string is UTF-8, make the
12547 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12548 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12561 i = SvCUR(sv) - origlen;
12564 case 'c': *(va_arg(*args, char*)) = i; break;
12565 case 'h': *(va_arg(*args, short*)) = i; break;
12566 default: *(va_arg(*args, int*)) = i; break;
12567 case 'l': *(va_arg(*args, long*)) = i; break;
12568 case 'V': *(va_arg(*args, IV*)) = i; break;
12569 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12570 #ifdef HAS_PTRDIFF_T
12571 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12574 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12578 *(va_arg(*args, Quad_t*)) = i; break;
12585 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12586 continue; /* not "break" */
12593 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12594 && ckWARN(WARN_PRINTF))
12596 SV * const msg = sv_newmortal();
12597 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12598 (PL_op->op_type == OP_PRTF) ? "" : "s");
12599 if (fmtstart < patend) {
12600 const char * const fmtend = q < patend ? q : patend;
12602 sv_catpvs(msg, "\"%");
12603 for (f = fmtstart; f < fmtend; f++) {
12605 sv_catpvn_nomg(msg, f, 1);
12607 Perl_sv_catpvf(aTHX_ msg,
12608 "\\%03"UVof, (UV)*f & 0xFF);
12611 sv_catpvs(msg, "\"");
12613 sv_catpvs(msg, "end of string");
12615 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12618 /* output mangled stuff ... */
12624 /* ... right here, because formatting flags should not apply */
12625 SvGROW(sv, SvCUR(sv) + elen + 1);
12627 Copy(eptr, p, elen, char);
12630 SvCUR_set(sv, p - SvPVX_const(sv));
12632 continue; /* not "break" */
12635 if (is_utf8 != has_utf8) {
12638 sv_utf8_upgrade(sv);
12641 const STRLEN old_elen = elen;
12642 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12643 sv_utf8_upgrade(nsv);
12644 eptr = SvPVX_const(nsv);
12647 if (width) { /* fudge width (can't fudge elen) */
12648 width += elen - old_elen;
12654 assert((IV)elen >= 0); /* here zero elen is fine */
12655 have = esignlen + zeros + elen;
12657 croak_memory_wrap();
12659 need = (have > width ? have : width);
12662 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12663 croak_memory_wrap();
12664 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12666 if (esignlen && fill == '0') {
12668 for (i = 0; i < (int)esignlen; i++)
12669 *p++ = esignbuf[i];
12671 if (gap && !left) {
12672 memset(p, fill, gap);
12675 if (esignlen && fill != '0') {
12677 for (i = 0; i < (int)esignlen; i++)
12678 *p++ = esignbuf[i];
12682 for (i = zeros; i; i--)
12686 Copy(eptr, p, elen, char);
12690 memset(p, ' ', gap);
12695 Copy(dotstr, p, dotstrlen, char);
12699 vectorize = FALSE; /* done iterating over vecstr */
12706 SvCUR_set(sv, p - SvPVX_const(sv));
12713 /* Now that we've consumed all our printf format arguments (svix)
12714 * do we have things left on the stack that we didn't use?
12716 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12717 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12718 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12723 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12727 /* =========================================================================
12729 =head1 Cloning an interpreter
12733 All the macros and functions in this section are for the private use of
12734 the main function, perl_clone().
12736 The foo_dup() functions make an exact copy of an existing foo thingy.
12737 During the course of a cloning, a hash table is used to map old addresses
12738 to new addresses. The table is created and manipulated with the
12739 ptr_table_* functions.
12741 * =========================================================================*/
12744 #if defined(USE_ITHREADS)
12746 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12747 #ifndef GpREFCNT_inc
12748 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12752 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12753 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12754 If this changes, please unmerge ss_dup.
12755 Likewise, sv_dup_inc_multiple() relies on this fact. */
12756 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12757 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12758 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12759 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12760 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12761 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12762 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12763 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12764 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12765 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12766 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12767 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12768 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12770 /* clone a parser */
12773 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12777 PERL_ARGS_ASSERT_PARSER_DUP;
12782 /* look for it in the table first */
12783 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12787 /* create anew and remember what it is */
12788 Newxz(parser, 1, yy_parser);
12789 ptr_table_store(PL_ptr_table, proto, parser);
12791 /* XXX these not yet duped */
12792 parser->old_parser = NULL;
12793 parser->stack = NULL;
12795 parser->stack_size = 0;
12796 /* XXX parser->stack->state = 0; */
12798 /* XXX eventually, just Copy() most of the parser struct ? */
12800 parser->lex_brackets = proto->lex_brackets;
12801 parser->lex_casemods = proto->lex_casemods;
12802 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12803 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12804 parser->lex_casestack = savepvn(proto->lex_casestack,
12805 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12806 parser->lex_defer = proto->lex_defer;
12807 parser->lex_dojoin = proto->lex_dojoin;
12808 parser->lex_formbrack = proto->lex_formbrack;
12809 parser->lex_inpat = proto->lex_inpat;
12810 parser->lex_inwhat = proto->lex_inwhat;
12811 parser->lex_op = proto->lex_op;
12812 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12813 parser->lex_starts = proto->lex_starts;
12814 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12815 parser->multi_close = proto->multi_close;
12816 parser->multi_open = proto->multi_open;
12817 parser->multi_start = proto->multi_start;
12818 parser->multi_end = proto->multi_end;
12819 parser->preambled = proto->preambled;
12820 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12821 parser->linestr = sv_dup_inc(proto->linestr, param);
12822 parser->expect = proto->expect;
12823 parser->copline = proto->copline;
12824 parser->last_lop_op = proto->last_lop_op;
12825 parser->lex_state = proto->lex_state;
12826 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12827 /* rsfp_filters entries have fake IoDIRP() */
12828 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12829 parser->in_my = proto->in_my;
12830 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12831 parser->error_count = proto->error_count;
12834 parser->linestr = sv_dup_inc(proto->linestr, param);
12837 char * const ols = SvPVX(proto->linestr);
12838 char * const ls = SvPVX(parser->linestr);
12840 parser->bufptr = ls + (proto->bufptr >= ols ?
12841 proto->bufptr - ols : 0);
12842 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12843 proto->oldbufptr - ols : 0);
12844 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12845 proto->oldoldbufptr - ols : 0);
12846 parser->linestart = ls + (proto->linestart >= ols ?
12847 proto->linestart - ols : 0);
12848 parser->last_uni = ls + (proto->last_uni >= ols ?
12849 proto->last_uni - ols : 0);
12850 parser->last_lop = ls + (proto->last_lop >= ols ?
12851 proto->last_lop - ols : 0);
12853 parser->bufend = ls + SvCUR(parser->linestr);
12856 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12859 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12860 Copy(proto->nexttype, parser->nexttype, 5, I32);
12861 parser->nexttoke = proto->nexttoke;
12863 /* XXX should clone saved_curcop here, but we aren't passed
12864 * proto_perl; so do it in perl_clone_using instead */
12870 /* duplicate a file handle */
12873 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12877 PERL_ARGS_ASSERT_FP_DUP;
12878 PERL_UNUSED_ARG(type);
12881 return (PerlIO*)NULL;
12883 /* look for it in the table first */
12884 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12888 /* create anew and remember what it is */
12889 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12890 ptr_table_store(PL_ptr_table, fp, ret);
12894 /* duplicate a directory handle */
12897 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12901 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12903 const Direntry_t *dirent;
12904 char smallbuf[256];
12910 PERL_UNUSED_CONTEXT;
12911 PERL_ARGS_ASSERT_DIRP_DUP;
12916 /* look for it in the table first */
12917 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12921 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12923 PERL_UNUSED_ARG(param);
12927 /* open the current directory (so we can switch back) */
12928 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12930 /* chdir to our dir handle and open the present working directory */
12931 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12932 PerlDir_close(pwd);
12933 return (DIR *)NULL;
12935 /* Now we should have two dir handles pointing to the same dir. */
12937 /* Be nice to the calling code and chdir back to where we were. */
12938 /* XXX If this fails, then what? */
12939 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12941 /* We have no need of the pwd handle any more. */
12942 PerlDir_close(pwd);
12945 # define d_namlen(d) (d)->d_namlen
12947 # define d_namlen(d) strlen((d)->d_name)
12949 /* Iterate once through dp, to get the file name at the current posi-
12950 tion. Then step back. */
12951 pos = PerlDir_tell(dp);
12952 if ((dirent = PerlDir_read(dp))) {
12953 len = d_namlen(dirent);
12954 if (len <= sizeof smallbuf) name = smallbuf;
12955 else Newx(name, len, char);
12956 Move(dirent->d_name, name, len, char);
12958 PerlDir_seek(dp, pos);
12960 /* Iterate through the new dir handle, till we find a file with the
12962 if (!dirent) /* just before the end */
12964 pos = PerlDir_tell(ret);
12965 if (PerlDir_read(ret)) continue; /* not there yet */
12966 PerlDir_seek(ret, pos); /* step back */
12970 const long pos0 = PerlDir_tell(ret);
12972 pos = PerlDir_tell(ret);
12973 if ((dirent = PerlDir_read(ret))) {
12974 if (len == (STRLEN)d_namlen(dirent)
12975 && memEQ(name, dirent->d_name, len)) {
12977 PerlDir_seek(ret, pos); /* step back */
12980 /* else we are not there yet; keep iterating */
12982 else { /* This is not meant to happen. The best we can do is
12983 reset the iterator to the beginning. */
12984 PerlDir_seek(ret, pos0);
12991 if (name && name != smallbuf)
12996 ret = win32_dirp_dup(dp, param);
12999 /* pop it in the pointer table */
13001 ptr_table_store(PL_ptr_table, dp, ret);
13006 /* duplicate a typeglob */
13009 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13013 PERL_ARGS_ASSERT_GP_DUP;
13017 /* look for it in the table first */
13018 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13022 /* create anew and remember what it is */
13024 ptr_table_store(PL_ptr_table, gp, ret);
13027 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13028 on Newxz() to do this for us. */
13029 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13030 ret->gp_io = io_dup_inc(gp->gp_io, param);
13031 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13032 ret->gp_av = av_dup_inc(gp->gp_av, param);
13033 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13034 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13035 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13036 ret->gp_cvgen = gp->gp_cvgen;
13037 ret->gp_line = gp->gp_line;
13038 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13042 /* duplicate a chain of magic */
13045 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13047 MAGIC *mgret = NULL;
13048 MAGIC **mgprev_p = &mgret;
13050 PERL_ARGS_ASSERT_MG_DUP;
13052 for (; mg; mg = mg->mg_moremagic) {
13055 if ((param->flags & CLONEf_JOIN_IN)
13056 && mg->mg_type == PERL_MAGIC_backref)
13057 /* when joining, we let the individual SVs add themselves to
13058 * backref as needed. */
13061 Newx(nmg, 1, MAGIC);
13063 mgprev_p = &(nmg->mg_moremagic);
13065 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13066 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13067 from the original commit adding Perl_mg_dup() - revision 4538.
13068 Similarly there is the annotation "XXX random ptr?" next to the
13069 assignment to nmg->mg_ptr. */
13072 /* FIXME for plugins
13073 if (nmg->mg_type == PERL_MAGIC_qr) {
13074 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13078 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13079 ? nmg->mg_type == PERL_MAGIC_backref
13080 /* The backref AV has its reference
13081 * count deliberately bumped by 1 */
13082 ? SvREFCNT_inc(av_dup_inc((const AV *)
13083 nmg->mg_obj, param))
13084 : sv_dup_inc(nmg->mg_obj, param)
13085 : sv_dup(nmg->mg_obj, param);
13087 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13088 if (nmg->mg_len > 0) {
13089 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13090 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13091 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13093 AMT * const namtp = (AMT*)nmg->mg_ptr;
13094 sv_dup_inc_multiple((SV**)(namtp->table),
13095 (SV**)(namtp->table), NofAMmeth, param);
13098 else if (nmg->mg_len == HEf_SVKEY)
13099 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13101 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13102 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13108 #endif /* USE_ITHREADS */
13110 struct ptr_tbl_arena {
13111 struct ptr_tbl_arena *next;
13112 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13115 /* create a new pointer-mapping table */
13118 Perl_ptr_table_new(pTHX)
13121 PERL_UNUSED_CONTEXT;
13123 Newx(tbl, 1, PTR_TBL_t);
13124 tbl->tbl_max = 511;
13125 tbl->tbl_items = 0;
13126 tbl->tbl_arena = NULL;
13127 tbl->tbl_arena_next = NULL;
13128 tbl->tbl_arena_end = NULL;
13129 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13133 #define PTR_TABLE_HASH(ptr) \
13134 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13136 /* map an existing pointer using a table */
13138 STATIC PTR_TBL_ENT_t *
13139 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13141 PTR_TBL_ENT_t *tblent;
13142 const UV hash = PTR_TABLE_HASH(sv);
13144 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13146 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13147 for (; tblent; tblent = tblent->next) {
13148 if (tblent->oldval == sv)
13155 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13157 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13159 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13160 PERL_UNUSED_CONTEXT;
13162 return tblent ? tblent->newval : NULL;
13165 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13166 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13167 * the core's typical use of ptr_tables in thread cloning. */
13170 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13172 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13174 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13175 PERL_UNUSED_CONTEXT;
13178 tblent->newval = newsv;
13180 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13182 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13183 struct ptr_tbl_arena *new_arena;
13185 Newx(new_arena, 1, struct ptr_tbl_arena);
13186 new_arena->next = tbl->tbl_arena;
13187 tbl->tbl_arena = new_arena;
13188 tbl->tbl_arena_next = new_arena->array;
13189 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13192 tblent = tbl->tbl_arena_next++;
13194 tblent->oldval = oldsv;
13195 tblent->newval = newsv;
13196 tblent->next = tbl->tbl_ary[entry];
13197 tbl->tbl_ary[entry] = tblent;
13199 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13200 ptr_table_split(tbl);
13204 /* double the hash bucket size of an existing ptr table */
13207 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13209 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13210 const UV oldsize = tbl->tbl_max + 1;
13211 UV newsize = oldsize * 2;
13214 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13215 PERL_UNUSED_CONTEXT;
13217 Renew(ary, newsize, PTR_TBL_ENT_t*);
13218 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13219 tbl->tbl_max = --newsize;
13220 tbl->tbl_ary = ary;
13221 for (i=0; i < oldsize; i++, ary++) {
13222 PTR_TBL_ENT_t **entp = ary;
13223 PTR_TBL_ENT_t *ent = *ary;
13224 PTR_TBL_ENT_t **curentp;
13227 curentp = ary + oldsize;
13229 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13231 ent->next = *curentp;
13241 /* remove all the entries from a ptr table */
13242 /* Deprecated - will be removed post 5.14 */
13245 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13247 PERL_UNUSED_CONTEXT;
13248 if (tbl && tbl->tbl_items) {
13249 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13251 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
13254 struct ptr_tbl_arena *next = arena->next;
13260 tbl->tbl_items = 0;
13261 tbl->tbl_arena = NULL;
13262 tbl->tbl_arena_next = NULL;
13263 tbl->tbl_arena_end = NULL;
13267 /* clear and free a ptr table */
13270 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13272 struct ptr_tbl_arena *arena;
13274 PERL_UNUSED_CONTEXT;
13280 arena = tbl->tbl_arena;
13283 struct ptr_tbl_arena *next = arena->next;
13289 Safefree(tbl->tbl_ary);
13293 #if defined(USE_ITHREADS)
13296 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13298 PERL_ARGS_ASSERT_RVPV_DUP;
13300 assert(!isREGEXP(sstr));
13302 if (SvWEAKREF(sstr)) {
13303 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13304 if (param->flags & CLONEf_JOIN_IN) {
13305 /* if joining, we add any back references individually rather
13306 * than copying the whole backref array */
13307 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13311 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13313 else if (SvPVX_const(sstr)) {
13314 /* Has something there */
13316 /* Normal PV - clone whole allocated space */
13317 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13318 /* sstr may not be that normal, but actually copy on write.
13319 But we are a true, independent SV, so: */
13323 /* Special case - not normally malloced for some reason */
13324 if (isGV_with_GP(sstr)) {
13325 /* Don't need to do anything here. */
13327 else if ((SvIsCOW(sstr))) {
13328 /* A "shared" PV - clone it as "shared" PV */
13330 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13334 /* Some other special case - random pointer */
13335 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13340 /* Copy the NULL */
13341 SvPV_set(dstr, NULL);
13345 /* duplicate a list of SVs. source and dest may point to the same memory. */
13347 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13348 SSize_t items, CLONE_PARAMS *const param)
13350 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13352 while (items-- > 0) {
13353 *dest++ = sv_dup_inc(*source++, param);
13359 /* duplicate an SV of any type (including AV, HV etc) */
13362 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13367 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13369 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13370 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13375 /* look for it in the table first */
13376 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13380 if(param->flags & CLONEf_JOIN_IN) {
13381 /** We are joining here so we don't want do clone
13382 something that is bad **/
13383 if (SvTYPE(sstr) == SVt_PVHV) {
13384 const HEK * const hvname = HvNAME_HEK(sstr);
13386 /** don't clone stashes if they already exist **/
13387 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13388 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13389 ptr_table_store(PL_ptr_table, sstr, dstr);
13393 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13394 HV *stash = GvSTASH(sstr);
13395 const HEK * hvname;
13396 if (stash && (hvname = HvNAME_HEK(stash))) {
13397 /** don't clone GVs if they already exist **/
13399 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13400 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13402 stash, GvNAME(sstr),
13408 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13409 ptr_table_store(PL_ptr_table, sstr, *svp);
13416 /* create anew and remember what it is */
13419 #ifdef DEBUG_LEAKING_SCALARS
13420 dstr->sv_debug_optype = sstr->sv_debug_optype;
13421 dstr->sv_debug_line = sstr->sv_debug_line;
13422 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13423 dstr->sv_debug_parent = (SV*)sstr;
13424 FREE_SV_DEBUG_FILE(dstr);
13425 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13428 ptr_table_store(PL_ptr_table, sstr, dstr);
13431 SvFLAGS(dstr) = SvFLAGS(sstr);
13432 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13433 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13436 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13437 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13438 (void*)PL_watch_pvx, SvPVX_const(sstr));
13441 /* don't clone objects whose class has asked us not to */
13443 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13449 switch (SvTYPE(sstr)) {
13451 SvANY(dstr) = NULL;
13454 SET_SVANY_FOR_BODYLESS_IV(dstr);
13456 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13458 SvIV_set(dstr, SvIVX(sstr));
13462 #if NVSIZE <= IVSIZE
13463 SET_SVANY_FOR_BODYLESS_NV(dstr);
13465 SvANY(dstr) = new_XNV();
13467 SvNV_set(dstr, SvNVX(sstr));
13471 /* These are all the types that need complex bodies allocating. */
13473 const svtype sv_type = SvTYPE(sstr);
13474 const struct body_details *const sv_type_details
13475 = bodies_by_type + sv_type;
13479 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13495 assert(sv_type_details->body_size);
13496 if (sv_type_details->arena) {
13497 new_body_inline(new_body, sv_type);
13499 = (void*)((char*)new_body - sv_type_details->offset);
13501 new_body = new_NOARENA(sv_type_details);
13505 SvANY(dstr) = new_body;
13508 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13509 ((char*)SvANY(dstr)) + sv_type_details->offset,
13510 sv_type_details->copy, char);
13512 Copy(((char*)SvANY(sstr)),
13513 ((char*)SvANY(dstr)),
13514 sv_type_details->body_size + sv_type_details->offset, char);
13517 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13518 && !isGV_with_GP(dstr)
13520 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13521 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13523 /* The Copy above means that all the source (unduplicated) pointers
13524 are now in the destination. We can check the flags and the
13525 pointers in either, but it's possible that there's less cache
13526 missing by always going for the destination.
13527 FIXME - instrument and check that assumption */
13528 if (sv_type >= SVt_PVMG) {
13530 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13531 if (SvOBJECT(dstr) && SvSTASH(dstr))
13532 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13533 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13536 /* The cast silences a GCC warning about unhandled types. */
13537 switch ((int)sv_type) {
13548 /* FIXME for plugins */
13549 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13550 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13553 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13554 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13555 LvTARG(dstr) = dstr;
13556 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13557 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13559 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13560 if (isREGEXP(sstr)) goto duprex;
13562 /* non-GP case already handled above */
13563 if(isGV_with_GP(sstr)) {
13564 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13565 /* Don't call sv_add_backref here as it's going to be
13566 created as part of the magic cloning of the symbol
13567 table--unless this is during a join and the stash
13568 is not actually being cloned. */
13569 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13570 at the point of this comment. */
13571 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13572 if (param->flags & CLONEf_JOIN_IN)
13573 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13574 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13575 (void)GpREFCNT_inc(GvGP(dstr));
13579 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13580 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13581 /* I have no idea why fake dirp (rsfps)
13582 should be treated differently but otherwise
13583 we end up with leaks -- sky*/
13584 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13585 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13586 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13588 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13589 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13590 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13591 if (IoDIRP(dstr)) {
13592 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13595 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13597 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13599 if (IoOFP(dstr) == IoIFP(sstr))
13600 IoOFP(dstr) = IoIFP(dstr);
13602 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13603 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13604 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13605 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13608 /* avoid cloning an empty array */
13609 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13610 SV **dst_ary, **src_ary;
13611 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13613 src_ary = AvARRAY((const AV *)sstr);
13614 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13615 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13616 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13617 AvALLOC((const AV *)dstr) = dst_ary;
13618 if (AvREAL((const AV *)sstr)) {
13619 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13623 while (items-- > 0)
13624 *dst_ary++ = sv_dup(*src_ary++, param);
13626 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13627 while (items-- > 0) {
13628 *dst_ary++ = &PL_sv_undef;
13632 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13633 AvALLOC((const AV *)dstr) = (SV**)NULL;
13634 AvMAX( (const AV *)dstr) = -1;
13635 AvFILLp((const AV *)dstr) = -1;
13639 if (HvARRAY((const HV *)sstr)) {
13641 const bool sharekeys = !!HvSHAREKEYS(sstr);
13642 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13643 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13645 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13646 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13648 HvARRAY(dstr) = (HE**)darray;
13649 while (i <= sxhv->xhv_max) {
13650 const HE * const source = HvARRAY(sstr)[i];
13651 HvARRAY(dstr)[i] = source
13652 ? he_dup(source, sharekeys, param) : 0;
13656 const struct xpvhv_aux * const saux = HvAUX(sstr);
13657 struct xpvhv_aux * const daux = HvAUX(dstr);
13658 /* This flag isn't copied. */
13661 if (saux->xhv_name_count) {
13662 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13664 = saux->xhv_name_count < 0
13665 ? -saux->xhv_name_count
13666 : saux->xhv_name_count;
13667 HEK **shekp = sname + count;
13669 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13670 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13671 while (shekp-- > sname) {
13673 *dhekp = hek_dup(*shekp, param);
13677 daux->xhv_name_u.xhvnameu_name
13678 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13681 daux->xhv_name_count = saux->xhv_name_count;
13683 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13684 daux->xhv_aux_flags = saux->xhv_aux_flags;
13685 #ifdef PERL_HASH_RANDOMIZE_KEYS
13686 daux->xhv_rand = saux->xhv_rand;
13687 daux->xhv_last_rand = saux->xhv_last_rand;
13689 daux->xhv_riter = saux->xhv_riter;
13690 daux->xhv_eiter = saux->xhv_eiter
13691 ? he_dup(saux->xhv_eiter,
13692 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13693 /* backref array needs refcnt=2; see sv_add_backref */
13694 daux->xhv_backreferences =
13695 (param->flags & CLONEf_JOIN_IN)
13696 /* when joining, we let the individual GVs and
13697 * CVs add themselves to backref as
13698 * needed. This avoids pulling in stuff
13699 * that isn't required, and simplifies the
13700 * case where stashes aren't cloned back
13701 * if they already exist in the parent
13704 : saux->xhv_backreferences
13705 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13706 ? MUTABLE_AV(SvREFCNT_inc(
13707 sv_dup_inc((const SV *)
13708 saux->xhv_backreferences, param)))
13709 : MUTABLE_AV(sv_dup((const SV *)
13710 saux->xhv_backreferences, param))
13713 daux->xhv_mro_meta = saux->xhv_mro_meta
13714 ? mro_meta_dup(saux->xhv_mro_meta, param)
13717 /* Record stashes for possible cloning in Perl_clone(). */
13719 av_push(param->stashes, dstr);
13723 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13726 if (!(param->flags & CLONEf_COPY_STACKS)) {
13731 /* NOTE: not refcounted */
13732 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13733 hv_dup(CvSTASH(dstr), param);
13734 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13735 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13736 if (!CvISXSUB(dstr)) {
13738 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13740 CvSLABBED_off(dstr);
13741 } else if (CvCONST(dstr)) {
13742 CvXSUBANY(dstr).any_ptr =
13743 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13745 assert(!CvSLABBED(dstr));
13746 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13748 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13749 hek_dup(CvNAME_HEK((CV *)sstr), param);
13750 /* don't dup if copying back - CvGV isn't refcounted, so the
13751 * duped GV may never be freed. A bit of a hack! DAPM */
13753 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13755 ? gv_dup_inc(CvGV(sstr), param)
13756 : (param->flags & CLONEf_JOIN_IN)
13758 : gv_dup(CvGV(sstr), param);
13760 if (!CvISXSUB(sstr)) {
13761 PADLIST * padlist = CvPADLIST(sstr);
13763 padlist = padlist_dup(padlist, param);
13764 CvPADLIST_set(dstr, padlist);
13766 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13767 PoisonPADLIST(dstr);
13770 CvWEAKOUTSIDE(sstr)
13771 ? cv_dup( CvOUTSIDE(dstr), param)
13772 : cv_dup_inc(CvOUTSIDE(dstr), param);
13782 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13784 PERL_ARGS_ASSERT_SV_DUP_INC;
13785 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13789 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13791 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13792 PERL_ARGS_ASSERT_SV_DUP;
13794 /* Track every SV that (at least initially) had a reference count of 0.
13795 We need to do this by holding an actual reference to it in this array.
13796 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13797 (akin to the stashes hash, and the perl stack), we come unstuck if
13798 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13799 thread) is manipulated in a CLONE method, because CLONE runs before the
13800 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13801 (and fix things up by giving each a reference via the temps stack).
13802 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13803 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13804 before the walk of unreferenced happens and a reference to that is SV
13805 added to the temps stack. At which point we have the same SV considered
13806 to be in use, and free to be re-used. Not good.
13808 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13809 assert(param->unreferenced);
13810 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13816 /* duplicate a context */
13819 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13821 PERL_CONTEXT *ncxs;
13823 PERL_ARGS_ASSERT_CX_DUP;
13826 return (PERL_CONTEXT*)NULL;
13828 /* look for it in the table first */
13829 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13833 /* create anew and remember what it is */
13834 Newx(ncxs, max + 1, PERL_CONTEXT);
13835 ptr_table_store(PL_ptr_table, cxs, ncxs);
13836 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13839 PERL_CONTEXT * const ncx = &ncxs[ix];
13840 if (CxTYPE(ncx) == CXt_SUBST) {
13841 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13844 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13845 switch (CxTYPE(ncx)) {
13847 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13848 ? cv_dup_inc(ncx->blk_sub.cv, param)
13849 : cv_dup(ncx->blk_sub.cv,param));
13850 if(CxHASARGS(ncx)){
13851 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13852 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13854 ncx->blk_sub.argarray = NULL;
13855 ncx->blk_sub.savearray = NULL;
13857 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13858 ncx->blk_sub.oldcomppad);
13861 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13863 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13864 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13866 case CXt_LOOP_LAZYSV:
13867 ncx->blk_loop.state_u.lazysv.end
13868 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13869 /* We are taking advantage of av_dup_inc and sv_dup_inc
13870 actually being the same function, and order equivalence of
13872 We can assert the later [but only at run time :-(] */
13873 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13874 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13876 ncx->blk_loop.state_u.ary.ary
13877 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13878 case CXt_LOOP_LAZYIV:
13879 case CXt_LOOP_PLAIN:
13880 if (CxPADLOOP(ncx)) {
13881 ncx->blk_loop.itervar_u.oldcomppad
13882 = (PAD*)ptr_table_fetch(PL_ptr_table,
13883 ncx->blk_loop.itervar_u.oldcomppad);
13885 ncx->blk_loop.itervar_u.gv
13886 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13891 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13892 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13893 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13908 /* duplicate a stack info structure */
13911 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13915 PERL_ARGS_ASSERT_SI_DUP;
13918 return (PERL_SI*)NULL;
13920 /* look for it in the table first */
13921 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13925 /* create anew and remember what it is */
13926 Newxz(nsi, 1, PERL_SI);
13927 ptr_table_store(PL_ptr_table, si, nsi);
13929 nsi->si_stack = av_dup_inc(si->si_stack, param);
13930 nsi->si_cxix = si->si_cxix;
13931 nsi->si_cxmax = si->si_cxmax;
13932 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13933 nsi->si_type = si->si_type;
13934 nsi->si_prev = si_dup(si->si_prev, param);
13935 nsi->si_next = si_dup(si->si_next, param);
13936 nsi->si_markoff = si->si_markoff;
13941 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13942 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13943 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13944 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13945 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13946 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13947 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13948 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13949 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13950 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13951 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13952 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13953 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13954 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13955 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13956 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13959 #define pv_dup_inc(p) SAVEPV(p)
13960 #define pv_dup(p) SAVEPV(p)
13961 #define svp_dup_inc(p,pp) any_dup(p,pp)
13963 /* map any object to the new equivent - either something in the
13964 * ptr table, or something in the interpreter structure
13968 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13972 PERL_ARGS_ASSERT_ANY_DUP;
13975 return (void*)NULL;
13977 /* look for it in the table first */
13978 ret = ptr_table_fetch(PL_ptr_table, v);
13982 /* see if it is part of the interpreter structure */
13983 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13984 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13992 /* duplicate the save stack */
13995 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13998 ANY * const ss = proto_perl->Isavestack;
13999 const I32 max = proto_perl->Isavestack_max;
14000 I32 ix = proto_perl->Isavestack_ix;
14013 void (*dptr) (void*);
14014 void (*dxptr) (pTHX_ void*);
14016 PERL_ARGS_ASSERT_SS_DUP;
14018 Newxz(nss, max, ANY);
14021 const UV uv = POPUV(ss,ix);
14022 const U8 type = (U8)uv & SAVE_MASK;
14024 TOPUV(nss,ix) = uv;
14026 case SAVEt_CLEARSV:
14027 case SAVEt_CLEARPADRANGE:
14029 case SAVEt_HELEM: /* hash element */
14030 case SAVEt_SV: /* scalar reference */
14031 sv = (const SV *)POPPTR(ss,ix);
14032 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14034 case SAVEt_ITEM: /* normal string */
14035 case SAVEt_GVSV: /* scalar slot in GV */
14036 sv = (const SV *)POPPTR(ss,ix);
14037 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14038 if (type == SAVEt_SV)
14042 case SAVEt_MORTALIZESV:
14043 case SAVEt_READONLY_OFF:
14044 sv = (const SV *)POPPTR(ss,ix);
14045 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14047 case SAVEt_FREEPADNAME:
14048 ptr = POPPTR(ss,ix);
14049 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14050 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14052 case SAVEt_SHARED_PVREF: /* char* in shared space */
14053 c = (char*)POPPTR(ss,ix);
14054 TOPPTR(nss,ix) = savesharedpv(c);
14055 ptr = POPPTR(ss,ix);
14056 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14058 case SAVEt_GENERIC_SVREF: /* generic sv */
14059 case SAVEt_SVREF: /* scalar reference */
14060 sv = (const SV *)POPPTR(ss,ix);
14061 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14062 if (type == SAVEt_SVREF)
14063 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14064 ptr = POPPTR(ss,ix);
14065 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14067 case SAVEt_GVSLOT: /* any slot in GV */
14068 sv = (const SV *)POPPTR(ss,ix);
14069 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14070 ptr = POPPTR(ss,ix);
14071 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14072 sv = (const SV *)POPPTR(ss,ix);
14073 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14075 case SAVEt_HV: /* hash reference */
14076 case SAVEt_AV: /* array reference */
14077 sv = (const SV *) POPPTR(ss,ix);
14078 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14080 case SAVEt_COMPPAD:
14082 sv = (const SV *) POPPTR(ss,ix);
14083 TOPPTR(nss,ix) = sv_dup(sv, param);
14085 case SAVEt_INT: /* int reference */
14086 ptr = POPPTR(ss,ix);
14087 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14088 intval = (int)POPINT(ss,ix);
14089 TOPINT(nss,ix) = intval;
14091 case SAVEt_LONG: /* long reference */
14092 ptr = POPPTR(ss,ix);
14093 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14094 longval = (long)POPLONG(ss,ix);
14095 TOPLONG(nss,ix) = longval;
14097 case SAVEt_I32: /* I32 reference */
14098 ptr = POPPTR(ss,ix);
14099 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14101 TOPINT(nss,ix) = i;
14103 case SAVEt_IV: /* IV reference */
14104 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14105 ptr = POPPTR(ss,ix);
14106 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14108 TOPIV(nss,ix) = iv;
14110 case SAVEt_HPTR: /* HV* reference */
14111 case SAVEt_APTR: /* AV* reference */
14112 case SAVEt_SPTR: /* SV* reference */
14113 ptr = POPPTR(ss,ix);
14114 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14115 sv = (const SV *)POPPTR(ss,ix);
14116 TOPPTR(nss,ix) = sv_dup(sv, param);
14118 case SAVEt_VPTR: /* random* reference */
14119 ptr = POPPTR(ss,ix);
14120 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14122 case SAVEt_INT_SMALL:
14123 case SAVEt_I32_SMALL:
14124 case SAVEt_I16: /* I16 reference */
14125 case SAVEt_I8: /* I8 reference */
14127 ptr = POPPTR(ss,ix);
14128 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14130 case SAVEt_GENERIC_PVREF: /* generic char* */
14131 case SAVEt_PPTR: /* char* reference */
14132 ptr = POPPTR(ss,ix);
14133 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14134 c = (char*)POPPTR(ss,ix);
14135 TOPPTR(nss,ix) = pv_dup(c);
14137 case SAVEt_GP: /* scalar reference */
14138 gp = (GP*)POPPTR(ss,ix);
14139 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14140 (void)GpREFCNT_inc(gp);
14141 gv = (const GV *)POPPTR(ss,ix);
14142 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14145 ptr = POPPTR(ss,ix);
14146 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14147 /* these are assumed to be refcounted properly */
14149 switch (((OP*)ptr)->op_type) {
14151 case OP_LEAVESUBLV:
14155 case OP_LEAVEWRITE:
14156 TOPPTR(nss,ix) = ptr;
14159 (void) OpREFCNT_inc(o);
14163 TOPPTR(nss,ix) = NULL;
14168 TOPPTR(nss,ix) = NULL;
14170 case SAVEt_FREECOPHH:
14171 ptr = POPPTR(ss,ix);
14172 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14174 case SAVEt_ADELETE:
14175 av = (const AV *)POPPTR(ss,ix);
14176 TOPPTR(nss,ix) = av_dup_inc(av, param);
14178 TOPINT(nss,ix) = i;
14181 hv = (const HV *)POPPTR(ss,ix);
14182 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14184 TOPINT(nss,ix) = i;
14187 c = (char*)POPPTR(ss,ix);
14188 TOPPTR(nss,ix) = pv_dup_inc(c);
14190 case SAVEt_STACK_POS: /* Position on Perl stack */
14192 TOPINT(nss,ix) = i;
14194 case SAVEt_DESTRUCTOR:
14195 ptr = POPPTR(ss,ix);
14196 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14197 dptr = POPDPTR(ss,ix);
14198 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14199 any_dup(FPTR2DPTR(void *, dptr),
14202 case SAVEt_DESTRUCTOR_X:
14203 ptr = POPPTR(ss,ix);
14204 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14205 dxptr = POPDXPTR(ss,ix);
14206 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14207 any_dup(FPTR2DPTR(void *, dxptr),
14210 case SAVEt_REGCONTEXT:
14212 ix -= uv >> SAVE_TIGHT_SHIFT;
14214 case SAVEt_AELEM: /* array element */
14215 sv = (const SV *)POPPTR(ss,ix);
14216 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14218 TOPINT(nss,ix) = i;
14219 av = (const AV *)POPPTR(ss,ix);
14220 TOPPTR(nss,ix) = av_dup_inc(av, param);
14223 ptr = POPPTR(ss,ix);
14224 TOPPTR(nss,ix) = ptr;
14227 ptr = POPPTR(ss,ix);
14228 ptr = cophh_copy((COPHH*)ptr);
14229 TOPPTR(nss,ix) = ptr;
14231 TOPINT(nss,ix) = i;
14232 if (i & HINT_LOCALIZE_HH) {
14233 hv = (const HV *)POPPTR(ss,ix);
14234 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14237 case SAVEt_PADSV_AND_MORTALIZE:
14238 longval = (long)POPLONG(ss,ix);
14239 TOPLONG(nss,ix) = longval;
14240 ptr = POPPTR(ss,ix);
14241 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14242 sv = (const SV *)POPPTR(ss,ix);
14243 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14245 case SAVEt_SET_SVFLAGS:
14247 TOPINT(nss,ix) = i;
14249 TOPINT(nss,ix) = i;
14250 sv = (const SV *)POPPTR(ss,ix);
14251 TOPPTR(nss,ix) = sv_dup(sv, param);
14253 case SAVEt_COMPILE_WARNINGS:
14254 ptr = POPPTR(ss,ix);
14255 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14258 ptr = POPPTR(ss,ix);
14259 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14261 case SAVEt_GP_ALIASED_SV: {
14262 GP * gp_ptr = (GP *)POPPTR(ss,ix);
14263 GP * new_gp_ptr = gp_dup(gp_ptr, param);
14264 TOPPTR(nss,ix) = new_gp_ptr;
14265 new_gp_ptr->gp_refcnt++;
14270 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14278 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14279 * flag to the result. This is done for each stash before cloning starts,
14280 * so we know which stashes want their objects cloned */
14283 do_mark_cloneable_stash(pTHX_ SV *const sv)
14285 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14287 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14288 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14289 if (cloner && GvCV(cloner)) {
14296 mXPUSHs(newSVhek(hvname));
14298 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14305 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14313 =for apidoc perl_clone
14315 Create and return a new interpreter by cloning the current one.
14317 perl_clone takes these flags as parameters:
14319 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14320 without it we only clone the data and zero the stacks,
14321 with it we copy the stacks and the new perl interpreter is
14322 ready to run at the exact same point as the previous one.
14323 The pseudo-fork code uses COPY_STACKS while the
14324 threads->create doesn't.
14326 CLONEf_KEEP_PTR_TABLE -
14327 perl_clone keeps a ptr_table with the pointer of the old
14328 variable as a key and the new variable as a value,
14329 this allows it to check if something has been cloned and not
14330 clone it again but rather just use the value and increase the
14331 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14332 the ptr_table using the function
14333 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14334 reason to keep it around is if you want to dup some of your own
14335 variable who are outside the graph perl scans, example of this
14336 code is in threads.xs create.
14338 CLONEf_CLONE_HOST -
14339 This is a win32 thing, it is ignored on unix, it tells perls
14340 win32host code (which is c++) to clone itself, this is needed on
14341 win32 if you want to run two threads at the same time,
14342 if you just want to do some stuff in a separate perl interpreter
14343 and then throw it away and return to the original one,
14344 you don't need to do anything.
14349 /* XXX the above needs expanding by someone who actually understands it ! */
14350 EXTERN_C PerlInterpreter *
14351 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14354 perl_clone(PerlInterpreter *proto_perl, UV flags)
14357 #ifdef PERL_IMPLICIT_SYS
14359 PERL_ARGS_ASSERT_PERL_CLONE;
14361 /* perlhost.h so we need to call into it
14362 to clone the host, CPerlHost should have a c interface, sky */
14364 if (flags & CLONEf_CLONE_HOST) {
14365 return perl_clone_host(proto_perl,flags);
14367 return perl_clone_using(proto_perl, flags,
14369 proto_perl->IMemShared,
14370 proto_perl->IMemParse,
14372 proto_perl->IStdIO,
14376 proto_perl->IProc);
14380 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14381 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14382 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14383 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14384 struct IPerlDir* ipD, struct IPerlSock* ipS,
14385 struct IPerlProc* ipP)
14387 /* XXX many of the string copies here can be optimized if they're
14388 * constants; they need to be allocated as common memory and just
14389 * their pointers copied. */
14392 CLONE_PARAMS clone_params;
14393 CLONE_PARAMS* const param = &clone_params;
14395 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14397 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14398 #else /* !PERL_IMPLICIT_SYS */
14400 CLONE_PARAMS clone_params;
14401 CLONE_PARAMS* param = &clone_params;
14402 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14404 PERL_ARGS_ASSERT_PERL_CLONE;
14405 #endif /* PERL_IMPLICIT_SYS */
14407 /* for each stash, determine whether its objects should be cloned */
14408 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14409 PERL_SET_THX(my_perl);
14412 PoisonNew(my_perl, 1, PerlInterpreter);
14415 PL_defstash = NULL; /* may be used by perl malloc() */
14418 PL_scopestack_name = 0;
14420 PL_savestack_ix = 0;
14421 PL_savestack_max = -1;
14422 PL_sig_pending = 0;
14424 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14425 Zero(&PL_padname_undef, 1, PADNAME);
14426 Zero(&PL_padname_const, 1, PADNAME);
14427 # ifdef DEBUG_LEAKING_SCALARS
14428 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14430 #else /* !DEBUGGING */
14431 Zero(my_perl, 1, PerlInterpreter);
14432 #endif /* DEBUGGING */
14434 #ifdef PERL_IMPLICIT_SYS
14435 /* host pointers */
14437 PL_MemShared = ipMS;
14438 PL_MemParse = ipMP;
14445 #endif /* PERL_IMPLICIT_SYS */
14448 param->flags = flags;
14449 /* Nothing in the core code uses this, but we make it available to
14450 extensions (using mg_dup). */
14451 param->proto_perl = proto_perl;
14452 /* Likely nothing will use this, but it is initialised to be consistent
14453 with Perl_clone_params_new(). */
14454 param->new_perl = my_perl;
14455 param->unreferenced = NULL;
14458 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14460 PL_body_arenas = NULL;
14461 Zero(&PL_body_roots, 1, PL_body_roots);
14465 PL_sv_arenaroot = NULL;
14467 PL_debug = proto_perl->Idebug;
14469 /* dbargs array probably holds garbage */
14472 PL_compiling = proto_perl->Icompiling;
14474 /* pseudo environmental stuff */
14475 PL_origargc = proto_perl->Iorigargc;
14476 PL_origargv = proto_perl->Iorigargv;
14478 #ifndef NO_TAINT_SUPPORT
14479 /* Set tainting stuff before PerlIO_debug can possibly get called */
14480 PL_tainting = proto_perl->Itainting;
14481 PL_taint_warn = proto_perl->Itaint_warn;
14483 PL_tainting = FALSE;
14484 PL_taint_warn = FALSE;
14487 PL_minus_c = proto_perl->Iminus_c;
14489 PL_localpatches = proto_perl->Ilocalpatches;
14490 PL_splitstr = proto_perl->Isplitstr;
14491 PL_minus_n = proto_perl->Iminus_n;
14492 PL_minus_p = proto_perl->Iminus_p;
14493 PL_minus_l = proto_perl->Iminus_l;
14494 PL_minus_a = proto_perl->Iminus_a;
14495 PL_minus_E = proto_perl->Iminus_E;
14496 PL_minus_F = proto_perl->Iminus_F;
14497 PL_doswitches = proto_perl->Idoswitches;
14498 PL_dowarn = proto_perl->Idowarn;
14499 PL_sawalias = proto_perl->Isawalias;
14500 #ifdef PERL_SAWAMPERSAND
14501 PL_sawampersand = proto_perl->Isawampersand;
14503 PL_unsafe = proto_perl->Iunsafe;
14504 PL_perldb = proto_perl->Iperldb;
14505 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14506 PL_exit_flags = proto_perl->Iexit_flags;
14508 /* XXX time(&PL_basetime) when asked for? */
14509 PL_basetime = proto_perl->Ibasetime;
14511 PL_maxsysfd = proto_perl->Imaxsysfd;
14512 PL_statusvalue = proto_perl->Istatusvalue;
14514 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14516 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14519 /* RE engine related */
14520 PL_regmatch_slab = NULL;
14521 PL_reg_curpm = NULL;
14523 PL_sub_generation = proto_perl->Isub_generation;
14525 /* funky return mechanisms */
14526 PL_forkprocess = proto_perl->Iforkprocess;
14528 /* internal state */
14529 PL_maxo = proto_perl->Imaxo;
14531 PL_main_start = proto_perl->Imain_start;
14532 PL_eval_root = proto_perl->Ieval_root;
14533 PL_eval_start = proto_perl->Ieval_start;
14535 PL_filemode = proto_perl->Ifilemode;
14536 PL_lastfd = proto_perl->Ilastfd;
14537 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14540 PL_gensym = proto_perl->Igensym;
14542 PL_laststatval = proto_perl->Ilaststatval;
14543 PL_laststype = proto_perl->Ilaststype;
14546 PL_profiledata = NULL;
14548 PL_generation = proto_perl->Igeneration;
14550 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14551 PL_in_clean_all = proto_perl->Iin_clean_all;
14553 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14554 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14555 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14556 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14557 PL_nomemok = proto_perl->Inomemok;
14558 PL_an = proto_perl->Ian;
14559 PL_evalseq = proto_perl->Ievalseq;
14560 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14561 PL_origalen = proto_perl->Iorigalen;
14563 PL_sighandlerp = proto_perl->Isighandlerp;
14565 PL_runops = proto_perl->Irunops;
14567 PL_subline = proto_perl->Isubline;
14570 PL_cryptseen = proto_perl->Icryptseen;
14573 #ifdef USE_LOCALE_COLLATE
14574 PL_collation_ix = proto_perl->Icollation_ix;
14575 PL_collation_standard = proto_perl->Icollation_standard;
14576 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14577 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14578 #endif /* USE_LOCALE_COLLATE */
14580 #ifdef USE_LOCALE_NUMERIC
14581 PL_numeric_standard = proto_perl->Inumeric_standard;
14582 PL_numeric_local = proto_perl->Inumeric_local;
14583 #endif /* !USE_LOCALE_NUMERIC */
14585 /* Did the locale setup indicate UTF-8? */
14586 PL_utf8locale = proto_perl->Iutf8locale;
14587 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14588 /* Unicode features (see perlrun/-C) */
14589 PL_unicode = proto_perl->Iunicode;
14591 /* Pre-5.8 signals control */
14592 PL_signals = proto_perl->Isignals;
14594 /* times() ticks per second */
14595 PL_clocktick = proto_perl->Iclocktick;
14597 /* Recursion stopper for PerlIO_find_layer */
14598 PL_in_load_module = proto_perl->Iin_load_module;
14600 /* sort() routine */
14601 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14603 /* Not really needed/useful since the reenrant_retint is "volatile",
14604 * but do it for consistency's sake. */
14605 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14607 /* Hooks to shared SVs and locks. */
14608 PL_sharehook = proto_perl->Isharehook;
14609 PL_lockhook = proto_perl->Ilockhook;
14610 PL_unlockhook = proto_perl->Iunlockhook;
14611 PL_threadhook = proto_perl->Ithreadhook;
14612 PL_destroyhook = proto_perl->Idestroyhook;
14613 PL_signalhook = proto_perl->Isignalhook;
14615 PL_globhook = proto_perl->Iglobhook;
14618 PL_last_swash_hv = NULL; /* reinits on demand */
14619 PL_last_swash_klen = 0;
14620 PL_last_swash_key[0]= '\0';
14621 PL_last_swash_tmps = (U8*)NULL;
14622 PL_last_swash_slen = 0;
14624 PL_srand_called = proto_perl->Isrand_called;
14625 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14627 if (flags & CLONEf_COPY_STACKS) {
14628 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14629 PL_tmps_ix = proto_perl->Itmps_ix;
14630 PL_tmps_max = proto_perl->Itmps_max;
14631 PL_tmps_floor = proto_perl->Itmps_floor;
14633 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14634 * NOTE: unlike the others! */
14635 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14636 PL_scopestack_max = proto_perl->Iscopestack_max;
14638 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14639 * NOTE: unlike the others! */
14640 PL_savestack_ix = proto_perl->Isavestack_ix;
14641 PL_savestack_max = proto_perl->Isavestack_max;
14644 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14645 PL_top_env = &PL_start_env;
14647 PL_op = proto_perl->Iop;
14650 PL_Xpv = (XPV*)NULL;
14651 my_perl->Ina = proto_perl->Ina;
14653 PL_statbuf = proto_perl->Istatbuf;
14654 PL_statcache = proto_perl->Istatcache;
14656 #ifndef NO_TAINT_SUPPORT
14657 PL_tainted = proto_perl->Itainted;
14659 PL_tainted = FALSE;
14661 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14663 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14665 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14666 PL_restartop = proto_perl->Irestartop;
14667 PL_in_eval = proto_perl->Iin_eval;
14668 PL_delaymagic = proto_perl->Idelaymagic;
14669 PL_phase = proto_perl->Iphase;
14670 PL_localizing = proto_perl->Ilocalizing;
14672 PL_hv_fetch_ent_mh = NULL;
14673 PL_modcount = proto_perl->Imodcount;
14674 PL_lastgotoprobe = NULL;
14675 PL_dumpindent = proto_perl->Idumpindent;
14677 PL_efloatbuf = NULL; /* reinits on demand */
14678 PL_efloatsize = 0; /* reinits on demand */
14682 PL_colorset = 0; /* reinits PL_colors[] */
14683 /*PL_colors[6] = {0,0,0,0,0,0};*/
14685 /* Pluggable optimizer */
14686 PL_peepp = proto_perl->Ipeepp;
14687 PL_rpeepp = proto_perl->Irpeepp;
14688 /* op_free() hook */
14689 PL_opfreehook = proto_perl->Iopfreehook;
14691 #ifdef USE_REENTRANT_API
14692 /* XXX: things like -Dm will segfault here in perlio, but doing
14693 * PERL_SET_CONTEXT(proto_perl);
14694 * breaks too many other things
14696 Perl_reentrant_init(aTHX);
14699 /* create SV map for pointer relocation */
14700 PL_ptr_table = ptr_table_new();
14702 /* initialize these special pointers as early as possible */
14704 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14705 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14706 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14707 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14708 &PL_padname_const);
14710 /* create (a non-shared!) shared string table */
14711 PL_strtab = newHV();
14712 HvSHAREKEYS_off(PL_strtab);
14713 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14714 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14716 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14718 /* This PV will be free'd special way so must set it same way op.c does */
14719 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14720 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14722 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14723 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14724 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14725 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14727 param->stashes = newAV(); /* Setup array of objects to call clone on */
14728 /* This makes no difference to the implementation, as it always pushes
14729 and shifts pointers to other SVs without changing their reference
14730 count, with the array becoming empty before it is freed. However, it
14731 makes it conceptually clear what is going on, and will avoid some
14732 work inside av.c, filling slots between AvFILL() and AvMAX() with
14733 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14734 AvREAL_off(param->stashes);
14736 if (!(flags & CLONEf_COPY_STACKS)) {
14737 param->unreferenced = newAV();
14740 #ifdef PERLIO_LAYERS
14741 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14742 PerlIO_clone(aTHX_ proto_perl, param);
14745 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14746 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14747 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14748 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14749 PL_xsubfilename = proto_perl->Ixsubfilename;
14750 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14751 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14754 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14755 PL_inplace = SAVEPV(proto_perl->Iinplace);
14756 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14758 /* magical thingies */
14760 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14761 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14763 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14764 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14765 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14768 /* Clone the regex array */
14769 /* ORANGE FIXME for plugins, probably in the SV dup code.
14770 newSViv(PTR2IV(CALLREGDUPE(
14771 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14773 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14774 PL_regex_pad = AvARRAY(PL_regex_padav);
14776 PL_stashpadmax = proto_perl->Istashpadmax;
14777 PL_stashpadix = proto_perl->Istashpadix ;
14778 Newx(PL_stashpad, PL_stashpadmax, HV *);
14781 for (; o < PL_stashpadmax; ++o)
14782 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14785 /* shortcuts to various I/O objects */
14786 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14787 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14788 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14789 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14790 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14791 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14792 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14794 /* shortcuts to regexp stuff */
14795 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14797 /* shortcuts to misc objects */
14798 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14800 /* shortcuts to debugging objects */
14801 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14802 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14803 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14804 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14805 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14806 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14807 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14809 /* symbol tables */
14810 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14811 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14812 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14813 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14814 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14816 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14817 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14818 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14819 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14820 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14821 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14822 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14823 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14825 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14827 /* subprocess state */
14828 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14830 if (proto_perl->Iop_mask)
14831 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14834 /* PL_asserting = proto_perl->Iasserting; */
14836 /* current interpreter roots */
14837 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14839 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14842 /* runtime control stuff */
14843 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14845 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14847 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14849 /* interpreter atexit processing */
14850 PL_exitlistlen = proto_perl->Iexitlistlen;
14851 if (PL_exitlistlen) {
14852 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14853 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14856 PL_exitlist = (PerlExitListEntry*)NULL;
14858 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14859 if (PL_my_cxt_size) {
14860 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14861 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14862 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14863 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14864 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14868 PL_my_cxt_list = (void**)NULL;
14869 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14870 PL_my_cxt_keys = (const char**)NULL;
14873 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14874 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14875 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14876 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14878 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14880 PAD_CLONE_VARS(proto_perl, param);
14882 #ifdef HAVE_INTERP_INTERN
14883 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14886 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14888 #ifdef PERL_USES_PL_PIDSTATUS
14889 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14891 PL_osname = SAVEPV(proto_perl->Iosname);
14892 PL_parser = parser_dup(proto_perl->Iparser, param);
14894 /* XXX this only works if the saved cop has already been cloned */
14895 if (proto_perl->Iparser) {
14896 PL_parser->saved_curcop = (COP*)any_dup(
14897 proto_perl->Iparser->saved_curcop,
14901 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14903 #ifdef USE_LOCALE_COLLATE
14904 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14905 #endif /* USE_LOCALE_COLLATE */
14907 #ifdef USE_LOCALE_NUMERIC
14908 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14909 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14910 #endif /* !USE_LOCALE_NUMERIC */
14912 /* Unicode inversion lists */
14913 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14914 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14915 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14916 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14918 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14919 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14921 /* utf8 character class swashes */
14922 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14923 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14925 for (i = 0; i < POSIX_CC_COUNT; i++) {
14926 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14928 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14929 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14930 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14931 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14932 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14933 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14934 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14935 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14936 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14937 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14938 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14939 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14940 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14941 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14942 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14943 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14945 if (proto_perl->Ipsig_pend) {
14946 Newxz(PL_psig_pend, SIG_SIZE, int);
14949 PL_psig_pend = (int*)NULL;
14952 if (proto_perl->Ipsig_name) {
14953 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14954 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14956 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14959 PL_psig_ptr = (SV**)NULL;
14960 PL_psig_name = (SV**)NULL;
14963 if (flags & CLONEf_COPY_STACKS) {
14964 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14965 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14966 PL_tmps_ix+1, param);
14968 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14969 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14970 Newxz(PL_markstack, i, I32);
14971 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14972 - proto_perl->Imarkstack);
14973 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14974 - proto_perl->Imarkstack);
14975 Copy(proto_perl->Imarkstack, PL_markstack,
14976 PL_markstack_ptr - PL_markstack + 1, I32);
14978 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14979 * NOTE: unlike the others! */
14980 Newxz(PL_scopestack, PL_scopestack_max, I32);
14981 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14984 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14985 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14987 /* reset stack AV to correct length before its duped via
14988 * PL_curstackinfo */
14989 AvFILLp(proto_perl->Icurstack) =
14990 proto_perl->Istack_sp - proto_perl->Istack_base;
14992 /* NOTE: si_dup() looks at PL_markstack */
14993 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14995 /* PL_curstack = PL_curstackinfo->si_stack; */
14996 PL_curstack = av_dup(proto_perl->Icurstack, param);
14997 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14999 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15000 PL_stack_base = AvARRAY(PL_curstack);
15001 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15002 - proto_perl->Istack_base);
15003 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15005 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15006 PL_savestack = ss_dup(proto_perl, param);
15010 ENTER; /* perl_destruct() wants to LEAVE; */
15013 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15014 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15016 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15017 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15018 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15019 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15020 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15021 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15023 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15025 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15026 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15027 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15029 PL_stashcache = newHV();
15031 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15032 proto_perl->Iwatchaddr);
15033 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15034 if (PL_debug && PL_watchaddr) {
15035 PerlIO_printf(Perl_debug_log,
15036 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15037 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15038 PTR2UV(PL_watchok));
15041 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15042 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15043 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15045 /* Call the ->CLONE method, if it exists, for each of the stashes
15046 identified by sv_dup() above.
15048 while(av_tindex(param->stashes) != -1) {
15049 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15050 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15051 if (cloner && GvCV(cloner)) {
15056 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15058 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15064 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15065 ptr_table_free(PL_ptr_table);
15066 PL_ptr_table = NULL;
15069 if (!(flags & CLONEf_COPY_STACKS)) {
15070 unreferenced_to_tmp_stack(param->unreferenced);
15073 SvREFCNT_dec(param->stashes);
15075 /* orphaned? eg threads->new inside BEGIN or use */
15076 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15077 SvREFCNT_inc_simple_void(PL_compcv);
15078 SAVEFREESV(PL_compcv);
15085 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15087 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15089 if (AvFILLp(unreferenced) > -1) {
15090 SV **svp = AvARRAY(unreferenced);
15091 SV **const last = svp + AvFILLp(unreferenced);
15095 if (SvREFCNT(*svp) == 1)
15097 } while (++svp <= last);
15099 EXTEND_MORTAL(count);
15100 svp = AvARRAY(unreferenced);
15103 if (SvREFCNT(*svp) == 1) {
15104 /* Our reference is the only one to this SV. This means that
15105 in this thread, the scalar effectively has a 0 reference.
15106 That doesn't work (cleanup never happens), so donate our
15107 reference to it onto the save stack. */
15108 PL_tmps_stack[++PL_tmps_ix] = *svp;
15110 /* As an optimisation, because we are already walking the
15111 entire array, instead of above doing either
15112 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15113 release our reference to the scalar, so that at the end of
15114 the array owns zero references to the scalars it happens to
15115 point to. We are effectively converting the array from
15116 AvREAL() on to AvREAL() off. This saves the av_clear()
15117 (triggered by the SvREFCNT_dec(unreferenced) below) from
15118 walking the array a second time. */
15119 SvREFCNT_dec(*svp);
15122 } while (++svp <= last);
15123 AvREAL_off(unreferenced);
15125 SvREFCNT_dec_NN(unreferenced);
15129 Perl_clone_params_del(CLONE_PARAMS *param)
15131 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15133 PerlInterpreter *const to = param->new_perl;
15135 PerlInterpreter *const was = PERL_GET_THX;
15137 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15143 SvREFCNT_dec(param->stashes);
15144 if (param->unreferenced)
15145 unreferenced_to_tmp_stack(param->unreferenced);
15155 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15158 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15159 does a dTHX; to get the context from thread local storage.
15160 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15161 a version that passes in my_perl. */
15162 PerlInterpreter *const was = PERL_GET_THX;
15163 CLONE_PARAMS *param;
15165 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15171 /* Given that we've set the context, we can do this unshared. */
15172 Newx(param, 1, CLONE_PARAMS);
15175 param->proto_perl = from;
15176 param->new_perl = to;
15177 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15178 AvREAL_off(param->stashes);
15179 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15187 #endif /* USE_ITHREADS */
15190 Perl_init_constants(pTHX)
15192 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15193 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15194 SvANY(&PL_sv_undef) = NULL;
15196 SvANY(&PL_sv_no) = new_XPVNV();
15197 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15198 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15199 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15202 SvANY(&PL_sv_yes) = new_XPVNV();
15203 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15204 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15205 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15208 SvPV_set(&PL_sv_no, (char*)PL_No);
15209 SvCUR_set(&PL_sv_no, 0);
15210 SvLEN_set(&PL_sv_no, 0);
15211 SvIV_set(&PL_sv_no, 0);
15212 SvNV_set(&PL_sv_no, 0);
15214 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15215 SvCUR_set(&PL_sv_yes, 1);
15216 SvLEN_set(&PL_sv_yes, 0);
15217 SvIV_set(&PL_sv_yes, 1);
15218 SvNV_set(&PL_sv_yes, 1);
15220 PadnamePV(&PL_padname_const) = (char *)PL_No;
15224 =head1 Unicode Support
15226 =for apidoc sv_recode_to_utf8
15228 The encoding is assumed to be an Encode object, on entry the PV
15229 of the sv is assumed to be octets in that encoding, and the sv
15230 will be converted into Unicode (and UTF-8).
15232 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15233 is not a reference, nothing is done to the sv. If the encoding is not
15234 an C<Encode::XS> Encoding object, bad things will happen.
15235 (See F<lib/encoding.pm> and L<Encode>.)
15237 The PV of the sv is returned.
15242 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15244 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15246 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15255 if (SvPADTMP(nsv)) {
15256 nsv = sv_newmortal();
15257 SvSetSV_nosteal(nsv, sv);
15265 Passing sv_yes is wrong - it needs to be or'ed set of constants
15266 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15267 remove converted chars from source.
15269 Both will default the value - let them.
15271 XPUSHs(&PL_sv_yes);
15274 call_method("decode", G_SCALAR);
15278 s = SvPV_const(uni, len);
15279 if (s != SvPVX_const(sv)) {
15280 SvGROW(sv, len + 1);
15281 Move(s, SvPVX(sv), len + 1, char);
15282 SvCUR_set(sv, len);
15287 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15288 /* clear pos and any utf8 cache */
15289 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15292 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15293 magic_setutf8(sv,mg); /* clear UTF8 cache */
15298 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15302 =for apidoc sv_cat_decode
15304 The encoding is assumed to be an Encode object, the PV of the ssv is
15305 assumed to be octets in that encoding and decoding the input starts
15306 from the position which (PV + *offset) pointed to. The dsv will be
15307 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
15308 when the string tstr appears in decoding output or the input ends on
15309 the PV of the ssv. The value which the offset points will be modified
15310 to the last input position on the ssv.
15312 Returns TRUE if the terminator was found, else returns FALSE.
15317 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15318 SV *ssv, int *offset, char *tstr, int tlen)
15322 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15324 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
15334 offsv = newSViv(*offset);
15336 mPUSHp(tstr, tlen);
15338 call_method("cat_decode", G_SCALAR);
15340 ret = SvTRUE(TOPs);
15341 *offset = SvIV(offsv);
15347 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15352 /* ---------------------------------------------------------------------
15354 * support functions for report_uninit()
15357 /* the maxiumum size of array or hash where we will scan looking
15358 * for the undefined element that triggered the warning */
15360 #define FUV_MAX_SEARCH_SIZE 1000
15362 /* Look for an entry in the hash whose value has the same SV as val;
15363 * If so, return a mortal copy of the key. */
15366 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15372 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15374 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15375 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15378 array = HvARRAY(hv);
15380 for (i=HvMAX(hv); i>=0; i--) {
15382 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15383 if (HeVAL(entry) != val)
15385 if ( HeVAL(entry) == &PL_sv_undef ||
15386 HeVAL(entry) == &PL_sv_placeholder)
15390 if (HeKLEN(entry) == HEf_SVKEY)
15391 return sv_mortalcopy(HeKEY_sv(entry));
15392 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15398 /* Look for an entry in the array whose value has the same SV as val;
15399 * If so, return the index, otherwise return -1. */
15402 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15404 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15406 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15407 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15410 if (val != &PL_sv_undef) {
15411 SV ** const svp = AvARRAY(av);
15414 for (i=AvFILLp(av); i>=0; i--)
15421 /* varname(): return the name of a variable, optionally with a subscript.
15422 * If gv is non-zero, use the name of that global, along with gvtype (one
15423 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15424 * targ. Depending on the value of the subscript_type flag, return:
15427 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15428 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15429 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15430 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15433 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15434 const SV *const keyname, I32 aindex, int subscript_type)
15437 SV * const name = sv_newmortal();
15438 if (gv && isGV(gv)) {
15440 buffer[0] = gvtype;
15443 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15445 gv_fullname4(name, gv, buffer, 0);
15447 if ((unsigned int)SvPVX(name)[1] <= 26) {
15449 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15451 /* Swap the 1 unprintable control character for the 2 byte pretty
15452 version - ie substr($name, 1, 1) = $buffer; */
15453 sv_insert(name, 1, 1, buffer, 2);
15457 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15460 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15462 if (!cv || !CvPADLIST(cv))
15464 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15465 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15469 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15470 SV * const sv = newSV(0);
15471 *SvPVX(name) = '$';
15472 Perl_sv_catpvf(aTHX_ name, "{%s}",
15473 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15474 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15475 SvREFCNT_dec_NN(sv);
15477 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15478 *SvPVX(name) = '$';
15479 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15481 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15482 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15483 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15491 =for apidoc find_uninit_var
15493 Find the name of the undefined variable (if any) that caused the operator
15494 to issue a "Use of uninitialized value" warning.
15495 If match is true, only return a name if its value matches uninit_sv.
15496 So roughly speaking, if a unary operator (such as OP_COS) generates a
15497 warning, then following the direct child of the op may yield an
15498 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15499 other hand, with OP_ADD there are two branches to follow, so we only print
15500 the variable name if we get an exact match.
15501 desc_p points to a string pointer holding the description of the op.
15502 This may be updated if needed.
15504 The name is returned as a mortal SV.
15506 Assumes that PL_op is the op that originally triggered the error, and that
15507 PL_comppad/PL_curpad points to the currently executing pad.
15513 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15514 bool match, const char **desc_p)
15519 const OP *o, *o2, *kid;
15521 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15523 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15524 uninit_sv == &PL_sv_placeholder)))
15527 switch (obase->op_type) {
15534 const bool pad = ( obase->op_type == OP_PADAV
15535 || obase->op_type == OP_PADHV
15536 || obase->op_type == OP_PADRANGE
15539 const bool hash = ( obase->op_type == OP_PADHV
15540 || obase->op_type == OP_RV2HV
15541 || (obase->op_type == OP_PADRANGE
15542 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15546 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15548 if (pad) { /* @lex, %lex */
15549 sv = PAD_SVl(obase->op_targ);
15553 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15554 /* @global, %global */
15555 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15558 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15560 else if (obase == PL_op) /* @{expr}, %{expr} */
15561 return find_uninit_var(cUNOPx(obase)->op_first,
15562 uninit_sv, match, desc_p);
15563 else /* @{expr}, %{expr} as a sub-expression */
15567 /* attempt to find a match within the aggregate */
15569 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15571 subscript_type = FUV_SUBSCRIPT_HASH;
15574 index = find_array_subscript((const AV *)sv, uninit_sv);
15576 subscript_type = FUV_SUBSCRIPT_ARRAY;
15579 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15582 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15583 keysv, index, subscript_type);
15587 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15589 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15590 if (!gv || !GvSTASH(gv))
15592 if (match && (GvSV(gv) != uninit_sv))
15594 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15597 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15600 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15602 return varname(NULL, '$', obase->op_targ,
15603 NULL, 0, FUV_SUBSCRIPT_NONE);
15606 gv = cGVOPx_gv(obase);
15607 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15609 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15611 case OP_AELEMFAST_LEX:
15614 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15615 if (!av || SvRMAGICAL(av))
15617 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15618 if (!svp || *svp != uninit_sv)
15621 return varname(NULL, '$', obase->op_targ,
15622 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15625 gv = cGVOPx_gv(obase);
15630 AV *const av = GvAV(gv);
15631 if (!av || SvRMAGICAL(av))
15633 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15634 if (!svp || *svp != uninit_sv)
15637 return varname(gv, '$', 0,
15638 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15640 NOT_REACHED; /* NOTREACHED */
15643 o = cUNOPx(obase)->op_first;
15644 if (!o || o->op_type != OP_NULL ||
15645 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15647 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15652 bool negate = FALSE;
15654 if (PL_op == obase)
15655 /* $a[uninit_expr] or $h{uninit_expr} */
15656 return find_uninit_var(cBINOPx(obase)->op_last,
15657 uninit_sv, match, desc_p);
15660 o = cBINOPx(obase)->op_first;
15661 kid = cBINOPx(obase)->op_last;
15663 /* get the av or hv, and optionally the gv */
15665 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15666 sv = PAD_SV(o->op_targ);
15668 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15669 && cUNOPo->op_first->op_type == OP_GV)
15671 gv = cGVOPx_gv(cUNOPo->op_first);
15675 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15680 if (kid && kid->op_type == OP_NEGATE) {
15682 kid = cUNOPx(kid)->op_first;
15685 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15686 /* index is constant */
15689 kidsv = newSVpvs_flags("-", SVs_TEMP);
15690 sv_catsv(kidsv, cSVOPx_sv(kid));
15693 kidsv = cSVOPx_sv(kid);
15697 if (obase->op_type == OP_HELEM) {
15698 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15699 if (!he || HeVAL(he) != uninit_sv)
15703 SV * const opsv = cSVOPx_sv(kid);
15704 const IV opsviv = SvIV(opsv);
15705 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15706 negate ? - opsviv : opsviv,
15708 if (!svp || *svp != uninit_sv)
15712 if (obase->op_type == OP_HELEM)
15713 return varname(gv, '%', o->op_targ,
15714 kidsv, 0, FUV_SUBSCRIPT_HASH);
15716 return varname(gv, '@', o->op_targ, NULL,
15717 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15718 FUV_SUBSCRIPT_ARRAY);
15721 /* index is an expression;
15722 * attempt to find a match within the aggregate */
15723 if (obase->op_type == OP_HELEM) {
15724 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15726 return varname(gv, '%', o->op_targ,
15727 keysv, 0, FUV_SUBSCRIPT_HASH);
15731 = find_array_subscript((const AV *)sv, uninit_sv);
15733 return varname(gv, '@', o->op_targ,
15734 NULL, index, FUV_SUBSCRIPT_ARRAY);
15739 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15741 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15743 NOT_REACHED; /* NOTREACHED */
15746 case OP_MULTIDEREF: {
15747 /* If we were executing OP_MULTIDEREF when the undef warning
15748 * triggered, then it must be one of the index values within
15749 * that triggered it. If not, then the only possibility is that
15750 * the value retrieved by the last aggregate lookup might be the
15751 * culprit. For the former, we set PL_multideref_pc each time before
15752 * using an index, so work though the item list until we reach
15753 * that point. For the latter, just work through the entire item
15754 * list; the last aggregate retrieved will be the candidate.
15757 /* the named aggregate, if any */
15758 PADOFFSET agg_targ = 0;
15760 /* the last-seen index */
15762 PADOFFSET index_targ;
15764 IV index_const_iv = 0; /* init for spurious compiler warn */
15765 SV *index_const_sv;
15766 int depth = 0; /* how many array/hash lookups we've done */
15768 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15769 UNOP_AUX_item *last = NULL;
15770 UV actions = items->uv;
15773 if (PL_op == obase) {
15774 last = PL_multideref_pc;
15775 assert(last >= items && last <= items + items[-1].uv);
15782 switch (actions & MDEREF_ACTION_MASK) {
15784 case MDEREF_reload:
15785 actions = (++items)->uv;
15788 case MDEREF_HV_padhv_helem: /* $lex{...} */
15791 case MDEREF_AV_padav_aelem: /* $lex[...] */
15792 agg_targ = (++items)->pad_offset;
15796 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15799 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15801 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15802 assert(isGV_with_GP(agg_gv));
15805 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15806 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15809 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15810 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15816 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15817 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15820 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15821 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15828 index_const_sv = NULL;
15830 index_type = (actions & MDEREF_INDEX_MASK);
15831 switch (index_type) {
15832 case MDEREF_INDEX_none:
15834 case MDEREF_INDEX_const:
15836 index_const_sv = UNOP_AUX_item_sv(++items)
15838 index_const_iv = (++items)->iv;
15840 case MDEREF_INDEX_padsv:
15841 index_targ = (++items)->pad_offset;
15843 case MDEREF_INDEX_gvsv:
15844 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15845 assert(isGV_with_GP(index_gv));
15849 if (index_type != MDEREF_INDEX_none)
15852 if ( index_type == MDEREF_INDEX_none
15853 || (actions & MDEREF_FLAG_last)
15854 || (last && items == last)
15858 actions >>= MDEREF_SHIFT;
15861 if (PL_op == obase) {
15862 /* index was undef */
15864 *desc_p = ( (actions & MDEREF_FLAG_last)
15865 && (obase->op_private
15866 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15868 (obase->op_private & OPpMULTIDEREF_EXISTS)
15871 : is_hv ? "hash element" : "array element";
15872 assert(index_type != MDEREF_INDEX_none);
15874 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15876 return varname(NULL, '$', index_targ,
15877 NULL, 0, FUV_SUBSCRIPT_NONE);
15878 assert(is_hv); /* AV index is an IV and can't be undef */
15879 /* can a const HV index ever be undef? */
15883 /* the SV returned by pp_multideref() was undef, if anything was */
15889 sv = PAD_SV(agg_targ);
15891 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15895 if (index_type == MDEREF_INDEX_const) {
15900 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15901 if (!he || HeVAL(he) != uninit_sv)
15905 SV * const * const svp =
15906 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15907 if (!svp || *svp != uninit_sv)
15912 ? varname(agg_gv, '%', agg_targ,
15913 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15914 : varname(agg_gv, '@', agg_targ,
15915 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15918 /* index is an var */
15920 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15922 return varname(agg_gv, '%', agg_targ,
15923 keysv, 0, FUV_SUBSCRIPT_HASH);
15927 = find_array_subscript((const AV *)sv, uninit_sv);
15929 return varname(agg_gv, '@', agg_targ,
15930 NULL, index, FUV_SUBSCRIPT_ARRAY);
15934 return varname(agg_gv,
15936 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15938 NOT_REACHED; /* NOTREACHED */
15942 /* only examine RHS */
15943 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
15947 o = cUNOPx(obase)->op_first;
15948 if ( o->op_type == OP_PUSHMARK
15949 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15953 if (!OpHAS_SIBLING(o)) {
15954 /* one-arg version of open is highly magical */
15956 if (o->op_type == OP_GV) { /* open FOO; */
15958 if (match && GvSV(gv) != uninit_sv)
15960 return varname(gv, '$', 0,
15961 NULL, 0, FUV_SUBSCRIPT_NONE);
15963 /* other possibilities not handled are:
15964 * open $x; or open my $x; should return '${*$x}'
15965 * open expr; should return '$'.expr ideally
15971 /* ops where $_ may be an implicit arg */
15976 if ( !(obase->op_flags & OPf_STACKED)) {
15977 if (uninit_sv == DEFSV)
15978 return newSVpvs_flags("$_", SVs_TEMP);
15979 else if (obase->op_targ
15980 && uninit_sv == PAD_SVl(obase->op_targ))
15981 return varname(NULL, '$', obase->op_targ, NULL, 0,
15982 FUV_SUBSCRIPT_NONE);
15989 match = 1; /* print etc can return undef on defined args */
15990 /* skip filehandle as it can't produce 'undef' warning */
15991 o = cUNOPx(obase)->op_first;
15992 if ((obase->op_flags & OPf_STACKED)
15994 ( o->op_type == OP_PUSHMARK
15995 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15996 o = OpSIBLING(OpSIBLING(o));
16000 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16001 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16003 /* the following ops are capable of returning PL_sv_undef even for
16004 * defined arg(s) */
16023 case OP_GETPEERNAME:
16071 case OP_SMARTMATCH:
16080 /* XXX tmp hack: these two may call an XS sub, and currently
16081 XS subs don't have a SUB entry on the context stack, so CV and
16082 pad determination goes wrong, and BAD things happen. So, just
16083 don't try to determine the value under those circumstances.
16084 Need a better fix at dome point. DAPM 11/2007 */
16090 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16091 if (gv && GvSV(gv) == uninit_sv)
16092 return newSVpvs_flags("$.", SVs_TEMP);
16097 /* def-ness of rval pos() is independent of the def-ness of its arg */
16098 if ( !(obase->op_flags & OPf_MOD))
16103 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16104 return newSVpvs_flags("${$/}", SVs_TEMP);
16109 if (!(obase->op_flags & OPf_KIDS))
16111 o = cUNOPx(obase)->op_first;
16117 /* This loop checks all the kid ops, skipping any that cannot pos-
16118 * sibly be responsible for the uninitialized value; i.e., defined
16119 * constants and ops that return nothing. If there is only one op
16120 * left that is not skipped, then we *know* it is responsible for
16121 * the uninitialized value. If there is more than one op left, we
16122 * have to look for an exact match in the while() loop below.
16123 * Note that we skip padrange, because the individual pad ops that
16124 * it replaced are still in the tree, so we work on them instead.
16127 for (kid=o; kid; kid = OpSIBLING(kid)) {
16128 const OPCODE type = kid->op_type;
16129 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16130 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16131 || (type == OP_PUSHMARK)
16132 || (type == OP_PADRANGE)
16136 if (o2) { /* more than one found */
16143 return find_uninit_var(o2, uninit_sv, match, desc_p);
16145 /* scan all args */
16147 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16159 =for apidoc report_uninit
16161 Print appropriate "Use of uninitialized variable" warning.
16167 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16170 SV* varname = NULL;
16173 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16176 if (uninit_sv && PL_curpad) {
16177 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16179 sv_insert(varname, 0, 0, " ", 1);
16181 /* PL_warn_uninit_sv is constant */
16182 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16183 /* diag_listed_as: Use of uninitialized value%s */
16184 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16185 SVfARG(varname ? varname : &PL_sv_no),
16190 /* PL_warn_uninit is constant */
16191 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16192 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16200 * c-indentation-style: bsd
16201 * c-basic-offset: 4
16202 * indent-tabs-mode: nil
16205 * ex: set ts=8 sts=4 sw=4 et: