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 *);
43 #ifdef PERL_NEW_COPY_ON_WRITE
44 # ifndef SV_COW_THRESHOLD
45 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
47 # ifndef SV_COWBUF_THRESHOLD
48 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
50 # ifndef SV_COW_MAX_WASTE_THRESHOLD
51 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
53 # ifndef SV_COWBUF_WASTE_THRESHOLD
54 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
56 # ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
57 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
59 # ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
60 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
63 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
66 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
68 # define GE_COW_THRESHOLD(cur) 1
70 #if SV_COWBUF_THRESHOLD
71 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
73 # define GE_COWBUF_THRESHOLD(cur) 1
75 #if SV_COW_MAX_WASTE_THRESHOLD
76 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
78 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
80 #if SV_COWBUF_WASTE_THRESHOLD
81 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
83 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
85 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
86 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
88 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
90 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
91 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
93 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #define CHECK_COW_THRESHOLD(cur,len) (\
97 GE_COW_THRESHOLD((cur)) && \
98 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
99 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
101 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
102 GE_COWBUF_THRESHOLD((cur)) && \
103 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
104 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #ifdef PERL_UTF8_CACHE_ASSERT
108 /* if adding more checks watch out for the following tests:
109 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
110 * lib/utf8.t lib/Unicode/Collate/t/index.t
113 # define ASSERT_UTF8_CACHE(cache) \
114 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
115 assert((cache)[2] <= (cache)[3]); \
116 assert((cache)[3] <= (cache)[1]);} \
119 # define ASSERT_UTF8_CACHE(cache) NOOP
122 #ifdef PERL_OLD_COPY_ON_WRITE
123 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
124 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
127 /* ============================================================================
129 =head1 Allocation and deallocation of SVs.
130 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
131 sv, av, hv...) contains type and reference count information, and for
132 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
133 contains fields specific to each type. Some types store all they need
134 in the head, so don't have a body.
136 In all but the most memory-paranoid configurations (ex: PURIFY), heads
137 and bodies are allocated out of arenas, which by default are
138 approximately 4K chunks of memory parcelled up into N heads or bodies.
139 Sv-bodies are allocated by their sv-type, guaranteeing size
140 consistency needed to allocate safely from arrays.
142 For SV-heads, the first slot in each arena is reserved, and holds a
143 link to the next arena, some flags, and a note of the number of slots.
144 Snaked through each arena chain is a linked list of free items; when
145 this becomes empty, an extra arena is allocated and divided up into N
146 items which are threaded into the free list.
148 SV-bodies are similar, but they use arena-sets by default, which
149 separate the link and info from the arena itself, and reclaim the 1st
150 slot in the arena. SV-bodies are further described later.
152 The following global variables are associated with arenas:
154 PL_sv_arenaroot pointer to list of SV arenas
155 PL_sv_root pointer to list of free SV structures
157 PL_body_arenas head of linked-list of body arenas
158 PL_body_roots[] array of pointers to list of free bodies of svtype
159 arrays are indexed by the svtype needed
161 A few special SV heads are not allocated from an arena, but are
162 instead directly created in the interpreter structure, eg PL_sv_undef.
163 The size of arenas can be changed from the default by setting
164 PERL_ARENA_SIZE appropriately at compile time.
166 The SV arena serves the secondary purpose of allowing still-live SVs
167 to be located and destroyed during final cleanup.
169 At the lowest level, the macros new_SV() and del_SV() grab and free
170 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
171 to return the SV to the free list with error checking.) new_SV() calls
172 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
173 SVs in the free list have their SvTYPE field set to all ones.
175 At the time of very final cleanup, sv_free_arenas() is called from
176 perl_destruct() to physically free all the arenas allocated since the
177 start of the interpreter.
179 The function visit() scans the SV arenas list, and calls a specified
180 function for each SV it finds which is still live - ie which has an SvTYPE
181 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
182 following functions (specified as [function that calls visit()] / [function
183 called by visit() for each SV]):
185 sv_report_used() / do_report_used()
186 dump all remaining SVs (debugging aid)
188 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
189 do_clean_named_io_objs(),do_curse()
190 Attempt to free all objects pointed to by RVs,
191 try to do the same for all objects indir-
192 ectly referenced by typeglobs too, and
193 then do a final sweep, cursing any
194 objects that remain. Called once from
195 perl_destruct(), prior to calling sv_clean_all()
198 sv_clean_all() / do_clean_all()
199 SvREFCNT_dec(sv) each remaining SV, possibly
200 triggering an sv_free(). It also sets the
201 SVf_BREAK flag on the SV to indicate that the
202 refcnt has been artificially lowered, and thus
203 stopping sv_free() from giving spurious warnings
204 about SVs which unexpectedly have a refcnt
205 of zero. called repeatedly from perl_destruct()
206 until there are no SVs left.
208 =head2 Arena allocator API Summary
210 Private API to rest of sv.c
214 new_XPVNV(), del_XPVGV(),
219 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
223 * ========================================================================= */
226 * "A time to plant, and a time to uproot what was planted..."
230 # define MEM_LOG_NEW_SV(sv, file, line, func) \
231 Perl_mem_log_new_sv(sv, file, line, func)
232 # define MEM_LOG_DEL_SV(sv, file, line, func) \
233 Perl_mem_log_del_sv(sv, file, line, func)
235 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
236 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
239 #ifdef DEBUG_LEAKING_SCALARS
240 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
241 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
243 # define DEBUG_SV_SERIAL(sv) \
244 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
245 PTR2UV(sv), (long)(sv)->sv_debug_serial))
247 # define FREE_SV_DEBUG_FILE(sv)
248 # define DEBUG_SV_SERIAL(sv) NOOP
252 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
253 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
254 /* Whilst I'd love to do this, it seems that things like to check on
256 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
258 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
259 PoisonNew(&SvREFCNT(sv), 1, U32)
261 # define SvARENA_CHAIN(sv) SvANY(sv)
262 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
263 # define POSION_SV_HEAD(sv)
266 /* Mark an SV head as unused, and add to free list.
268 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
269 * its refcount artificially decremented during global destruction, so
270 * there may be dangling pointers to it. The last thing we want in that
271 * case is for it to be reused. */
273 #define plant_SV(p) \
275 const U32 old_flags = SvFLAGS(p); \
276 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
277 DEBUG_SV_SERIAL(p); \
278 FREE_SV_DEBUG_FILE(p); \
280 SvFLAGS(p) = SVTYPEMASK; \
281 if (!(old_flags & SVf_BREAK)) { \
282 SvARENA_CHAIN_SET(p, PL_sv_root); \
288 #define uproot_SV(p) \
291 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
296 /* make some more SVs by adding another arena */
302 char *chunk; /* must use New here to match call to */
303 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
304 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
309 /* new_SV(): return a new, empty SV head */
311 #ifdef DEBUG_LEAKING_SCALARS
312 /* provide a real function for a debugger to play with */
314 S_new_SV(pTHX_ const char *file, int line, const char *func)
321 sv = S_more_sv(aTHX);
325 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
326 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
332 sv->sv_debug_inpad = 0;
333 sv->sv_debug_parent = NULL;
334 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
336 sv->sv_debug_serial = PL_sv_serial++;
338 MEM_LOG_NEW_SV(sv, file, line, func);
339 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
340 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
344 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
352 (p) = S_more_sv(aTHX); \
356 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
361 /* del_SV(): return an empty SV head to the free list */
374 S_del_sv(pTHX_ SV *p)
376 PERL_ARGS_ASSERT_DEL_SV;
381 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
382 const SV * const sv = sva + 1;
383 const SV * const svend = &sva[SvREFCNT(sva)];
384 if (p >= sv && p < svend) {
390 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
391 "Attempt to free non-arena SV: 0x%"UVxf
392 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
399 #else /* ! DEBUGGING */
401 #define del_SV(p) plant_SV(p)
403 #endif /* DEBUGGING */
407 =head1 SV Manipulation Functions
409 =for apidoc sv_add_arena
411 Given a chunk of memory, link it to the head of the list of arenas,
412 and split it into a list of free SVs.
418 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
420 SV *const sva = MUTABLE_SV(ptr);
424 PERL_ARGS_ASSERT_SV_ADD_ARENA;
426 /* The first SV in an arena isn't an SV. */
427 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
428 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
429 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
431 PL_sv_arenaroot = sva;
432 PL_sv_root = sva + 1;
434 svend = &sva[SvREFCNT(sva) - 1];
437 SvARENA_CHAIN_SET(sv, (sv + 1));
441 /* Must always set typemask because it's always checked in on cleanup
442 when the arenas are walked looking for objects. */
443 SvFLAGS(sv) = SVTYPEMASK;
446 SvARENA_CHAIN_SET(sv, 0);
450 SvFLAGS(sv) = SVTYPEMASK;
453 /* visit(): call the named function for each non-free SV in the arenas
454 * whose flags field matches the flags/mask args. */
457 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
462 PERL_ARGS_ASSERT_VISIT;
464 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
465 const SV * const svend = &sva[SvREFCNT(sva)];
467 for (sv = sva + 1; sv < svend; ++sv) {
468 if (SvTYPE(sv) != (svtype)SVTYPEMASK
469 && (sv->sv_flags & mask) == flags
482 /* called by sv_report_used() for each live SV */
485 do_report_used(pTHX_ SV *const sv)
487 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
488 PerlIO_printf(Perl_debug_log, "****\n");
495 =for apidoc sv_report_used
497 Dump the contents of all SVs not yet freed (debugging aid).
503 Perl_sv_report_used(pTHX)
506 visit(do_report_used, 0, 0);
512 /* called by sv_clean_objs() for each live SV */
515 do_clean_objs(pTHX_ SV *const ref)
519 SV * const target = SvRV(ref);
520 if (SvOBJECT(target)) {
521 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
522 if (SvWEAKREF(ref)) {
523 sv_del_backref(target, ref);
529 SvREFCNT_dec_NN(target);
536 /* clear any slots in a GV which hold objects - except IO;
537 * called by sv_clean_objs() for each live GV */
540 do_clean_named_objs(pTHX_ SV *const sv)
543 assert(SvTYPE(sv) == SVt_PVGV);
544 assert(isGV_with_GP(sv));
548 /* freeing GP entries may indirectly free the current GV;
549 * hold onto it while we mess with the GP slots */
552 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
553 DEBUG_D((PerlIO_printf(Perl_debug_log,
554 "Cleaning named glob SV object:\n "), sv_dump(obj)));
556 SvREFCNT_dec_NN(obj);
558 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
559 DEBUG_D((PerlIO_printf(Perl_debug_log,
560 "Cleaning named glob AV object:\n "), sv_dump(obj)));
562 SvREFCNT_dec_NN(obj);
564 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
565 DEBUG_D((PerlIO_printf(Perl_debug_log,
566 "Cleaning named glob HV object:\n "), sv_dump(obj)));
568 SvREFCNT_dec_NN(obj);
570 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
571 DEBUG_D((PerlIO_printf(Perl_debug_log,
572 "Cleaning named glob CV object:\n "), sv_dump(obj)));
574 SvREFCNT_dec_NN(obj);
576 SvREFCNT_dec_NN(sv); /* undo the inc above */
579 /* clear any IO slots in a GV which hold objects (except stderr, defout);
580 * called by sv_clean_objs() for each live GV */
583 do_clean_named_io_objs(pTHX_ SV *const sv)
586 assert(SvTYPE(sv) == SVt_PVGV);
587 assert(isGV_with_GP(sv));
588 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
592 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
593 DEBUG_D((PerlIO_printf(Perl_debug_log,
594 "Cleaning named glob IO object:\n "), sv_dump(obj)));
596 SvREFCNT_dec_NN(obj);
598 SvREFCNT_dec_NN(sv); /* undo the inc above */
601 /* Void wrapper to pass to visit() */
603 do_curse(pTHX_ SV * const sv) {
604 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
605 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
611 =for apidoc sv_clean_objs
613 Attempt to destroy all objects not yet freed.
619 Perl_sv_clean_objs(pTHX)
622 PL_in_clean_objs = TRUE;
623 visit(do_clean_objs, SVf_ROK, SVf_ROK);
624 /* Some barnacles may yet remain, clinging to typeglobs.
625 * Run the non-IO destructors first: they may want to output
626 * error messages, close files etc */
627 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
628 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
629 /* And if there are some very tenacious barnacles clinging to arrays,
630 closures, or what have you.... */
631 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
632 olddef = PL_defoutgv;
633 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
634 if (olddef && isGV_with_GP(olddef))
635 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
636 olderr = PL_stderrgv;
637 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
638 if (olderr && isGV_with_GP(olderr))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
640 SvREFCNT_dec(olddef);
641 PL_in_clean_objs = FALSE;
644 /* called by sv_clean_all() for each live SV */
647 do_clean_all(pTHX_ SV *const sv)
649 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
650 /* don't clean pid table and strtab */
653 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
654 SvFLAGS(sv) |= SVf_BREAK;
659 =for apidoc sv_clean_all
661 Decrement the refcnt of each remaining SV, possibly triggering a
662 cleanup. This function may have to be called multiple times to free
663 SVs which are in complex self-referential hierarchies.
669 Perl_sv_clean_all(pTHX)
672 PL_in_clean_all = TRUE;
673 cleaned = visit(do_clean_all, 0,0);
678 ARENASETS: a meta-arena implementation which separates arena-info
679 into struct arena_set, which contains an array of struct
680 arena_descs, each holding info for a single arena. By separating
681 the meta-info from the arena, we recover the 1st slot, formerly
682 borrowed for list management. The arena_set is about the size of an
683 arena, avoiding the needless malloc overhead of a naive linked-list.
685 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
686 memory in the last arena-set (1/2 on average). In trade, we get
687 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
688 smaller types). The recovery of the wasted space allows use of
689 small arenas for large, rare body types, by changing array* fields
690 in body_details_by_type[] below.
693 char *arena; /* the raw storage, allocated aligned */
694 size_t size; /* its size ~4k typ */
695 svtype utype; /* bodytype stored in arena */
700 /* Get the maximum number of elements in set[] such that struct arena_set
701 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
702 therefore likely to be 1 aligned memory page. */
704 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
705 - 2 * sizeof(int)) / sizeof (struct arena_desc))
708 struct arena_set* next;
709 unsigned int set_size; /* ie ARENAS_PER_SET */
710 unsigned int curr; /* index of next available arena-desc */
711 struct arena_desc set[ARENAS_PER_SET];
715 =for apidoc sv_free_arenas
717 Deallocate the memory used by all arenas. Note that all the individual SV
718 heads and bodies within the arenas must already have been freed.
724 Perl_sv_free_arenas(pTHX)
730 /* Free arenas here, but be careful about fake ones. (We assume
731 contiguity of the fake ones with the corresponding real ones.) */
733 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
734 svanext = MUTABLE_SV(SvANY(sva));
735 while (svanext && SvFAKE(svanext))
736 svanext = MUTABLE_SV(SvANY(svanext));
743 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
746 struct arena_set *current = aroot;
749 assert(aroot->set[i].arena);
750 Safefree(aroot->set[i].arena);
758 i = PERL_ARENA_ROOTS_SIZE;
760 PL_body_roots[i] = 0;
767 Here are mid-level routines that manage the allocation of bodies out
768 of the various arenas. There are 5 kinds of arenas:
770 1. SV-head arenas, which are discussed and handled above
771 2. regular body arenas
772 3. arenas for reduced-size bodies
775 Arena types 2 & 3 are chained by body-type off an array of
776 arena-root pointers, which is indexed by svtype. Some of the
777 larger/less used body types are malloced singly, since a large
778 unused block of them is wasteful. Also, several svtypes dont have
779 bodies; the data fits into the sv-head itself. The arena-root
780 pointer thus has a few unused root-pointers (which may be hijacked
781 later for arena types 4,5)
783 3 differs from 2 as an optimization; some body types have several
784 unused fields in the front of the structure (which are kept in-place
785 for consistency). These bodies can be allocated in smaller chunks,
786 because the leading fields arent accessed. Pointers to such bodies
787 are decremented to point at the unused 'ghost' memory, knowing that
788 the pointers are used with offsets to the real memory.
791 =head1 SV-Body Allocation
795 Allocation of SV-bodies is similar to SV-heads, differing as follows;
796 the allocation mechanism is used for many body types, so is somewhat
797 more complicated, it uses arena-sets, and has no need for still-live
800 At the outermost level, (new|del)_X*V macros return bodies of the
801 appropriate type. These macros call either (new|del)_body_type or
802 (new|del)_body_allocated macro pairs, depending on specifics of the
803 type. Most body types use the former pair, the latter pair is used to
804 allocate body types with "ghost fields".
806 "ghost fields" are fields that are unused in certain types, and
807 consequently don't need to actually exist. They are declared because
808 they're part of a "base type", which allows use of functions as
809 methods. The simplest examples are AVs and HVs, 2 aggregate types
810 which don't use the fields which support SCALAR semantics.
812 For these types, the arenas are carved up into appropriately sized
813 chunks, we thus avoid wasted memory for those unaccessed members.
814 When bodies are allocated, we adjust the pointer back in memory by the
815 size of the part not allocated, so it's as if we allocated the full
816 structure. (But things will all go boom if you write to the part that
817 is "not there", because you'll be overwriting the last members of the
818 preceding structure in memory.)
820 We calculate the correction using the STRUCT_OFFSET macro on the first
821 member present. If the allocated structure is smaller (no initial NV
822 actually allocated) then the net effect is to subtract the size of the NV
823 from the pointer, to return a new pointer as if an initial NV were actually
824 allocated. (We were using structures named *_allocated for this, but
825 this turned out to be a subtle bug, because a structure without an NV
826 could have a lower alignment constraint, but the compiler is allowed to
827 optimised accesses based on the alignment constraint of the actual pointer
828 to the full structure, for example, using a single 64 bit load instruction
829 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
831 This is the same trick as was used for NV and IV bodies. Ironically it
832 doesn't need to be used for NV bodies any more, because NV is now at
833 the start of the structure. IV bodies don't need it either, because
834 they are no longer allocated.
836 In turn, the new_body_* allocators call S_new_body(), which invokes
837 new_body_inline macro, which takes a lock, and takes a body off the
838 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
839 necessary to refresh an empty list. Then the lock is released, and
840 the body is returned.
842 Perl_more_bodies allocates a new arena, and carves it up into an array of N
843 bodies, which it strings into a linked list. It looks up arena-size
844 and body-size from the body_details table described below, thus
845 supporting the multiple body-types.
847 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848 the (new|del)_X*V macros are mapped directly to malloc/free.
850 For each sv-type, struct body_details bodies_by_type[] carries
851 parameters which control these aspects of SV handling:
853 Arena_size determines whether arenas are used for this body type, and if
854 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855 zero, forcing individual mallocs and frees.
857 Body_size determines how big a body is, and therefore how many fit into
858 each arena. Offset carries the body-pointer adjustment needed for
859 "ghost fields", and is used in *_allocated macros.
861 But its main purpose is to parameterize info needed in
862 Perl_sv_upgrade(). The info here dramatically simplifies the function
863 vs the implementation in 5.8.8, making it table-driven. All fields
864 are used for this, except for arena_size.
866 For the sv-types that have no bodies, arenas are not used, so those
867 PL_body_roots[sv_type] are unused, and can be overloaded. In
868 something of a special case, SVt_NULL is borrowed for HE arenas;
869 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
870 bodies_by_type[SVt_NULL] slot is not used, as the table is not
875 struct body_details {
876 U8 body_size; /* Size to allocate */
877 U8 copy; /* Size of structure to copy (may be shorter) */
879 unsigned int type : 4; /* We have space for a sanity check. */
880 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
881 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
882 unsigned int arena : 1; /* Allocated from an arena */
883 size_t arena_size; /* Size of arena to allocate */
891 /* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893 #define HASARENA FALSE
895 #define HASARENA TRUE
897 #define NOARENA FALSE
899 /* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
907 #define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909 #define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913 #define FIT_ARENA(count,body_size) \
915 ? FIT_ARENAn (count, body_size) \
916 : FIT_ARENA0 (body_size)
918 /* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
922 #define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
926 static const struct body_details bodies_by_type[] = {
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
930 /* IVs are in the head, so the allocation size is 0. */
932 sizeof(IV), /* This is used to copy out the IV body. */
933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
934 NOARENA /* IVS don't need an arena */, 0
937 { sizeof(NV), sizeof(NV),
938 STRUCT_OFFSET(XPVNV, xnv_u),
939 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
941 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
942 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
943 + STRUCT_OFFSET(XPV, xpv_cur),
944 SVt_PV, FALSE, NONV, HASARENA,
945 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
947 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_INVLIST, TRUE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
953 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_PVIV, FALSE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVNV, FALSE, HADNV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
966 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
971 SVt_REGEXP, TRUE, NONV, HASARENA,
972 FIT_ARENA(0, sizeof(regexp))
975 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
978 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
979 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
982 copy_length(XPVAV, xav_alloc),
984 SVt_PVAV, TRUE, NONV, HASARENA,
985 FIT_ARENA(0, sizeof(XPVAV)) },
988 copy_length(XPVHV, xhv_max),
990 SVt_PVHV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVHV)) },
996 SVt_PVCV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVCV)) },
1002 SVt_PVFM, TRUE, NONV, NOARENA,
1003 FIT_ARENA(20, sizeof(XPVFM)) },
1008 SVt_PVIO, TRUE, NONV, HASARENA,
1009 FIT_ARENA(24, sizeof(XPVIO)) },
1012 #define new_body_allocated(sv_type) \
1013 (void *)((char *)S_new_body(aTHX_ sv_type) \
1014 - bodies_by_type[sv_type].offset)
1016 /* return a thing to the free list */
1018 #define del_body(thing, root) \
1020 void ** const thing_copy = (void **)thing; \
1021 *thing_copy = *root; \
1022 *root = (void*)thing_copy; \
1027 #define new_XNV() safemalloc(sizeof(XPVNV))
1028 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1029 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1031 #define del_XPVGV(p) safefree(p)
1035 #define new_XNV() new_body_allocated(SVt_NV)
1036 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1037 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1039 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1040 &PL_body_roots[SVt_PVGV])
1044 /* no arena for you! */
1046 #define new_NOARENA(details) \
1047 safemalloc((details)->body_size + (details)->offset)
1048 #define new_NOARENAZ(details) \
1049 safecalloc((details)->body_size + (details)->offset, 1)
1052 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1053 const size_t arena_size)
1055 void ** const root = &PL_body_roots[sv_type];
1056 struct arena_desc *adesc;
1057 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1061 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1062 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1065 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1066 static bool done_sanity_check;
1068 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1069 * variables like done_sanity_check. */
1070 if (!done_sanity_check) {
1071 unsigned int i = SVt_LAST;
1073 done_sanity_check = TRUE;
1076 assert (bodies_by_type[i].type == i);
1082 /* may need new arena-set to hold new arena */
1083 if (!aroot || aroot->curr >= aroot->set_size) {
1084 struct arena_set *newroot;
1085 Newxz(newroot, 1, struct arena_set);
1086 newroot->set_size = ARENAS_PER_SET;
1087 newroot->next = aroot;
1089 PL_body_arenas = (void *) newroot;
1090 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1093 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1094 curr = aroot->curr++;
1095 adesc = &(aroot->set[curr]);
1096 assert(!adesc->arena);
1098 Newx(adesc->arena, good_arena_size, char);
1099 adesc->size = good_arena_size;
1100 adesc->utype = sv_type;
1101 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1102 curr, (void*)adesc->arena, (UV)good_arena_size));
1104 start = (char *) adesc->arena;
1106 /* Get the address of the byte after the end of the last body we can fit.
1107 Remember, this is integer division: */
1108 end = start + good_arena_size / body_size * body_size;
1110 /* computed count doesn't reflect the 1st slot reservation */
1111 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1112 DEBUG_m(PerlIO_printf(Perl_debug_log,
1113 "arena %p end %p arena-size %d (from %d) type %d "
1115 (void*)start, (void*)end, (int)good_arena_size,
1116 (int)arena_size, sv_type, (int)body_size,
1117 (int)good_arena_size / (int)body_size));
1119 DEBUG_m(PerlIO_printf(Perl_debug_log,
1120 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1121 (void*)start, (void*)end,
1122 (int)arena_size, sv_type, (int)body_size,
1123 (int)good_arena_size / (int)body_size));
1125 *root = (void *)start;
1128 /* Where the next body would start: */
1129 char * const next = start + body_size;
1132 /* This is the last body: */
1133 assert(next == end);
1135 *(void **)start = 0;
1139 *(void**) start = (void *)next;
1144 /* grab a new thing from the free list, allocating more if necessary.
1145 The inline version is used for speed in hot routines, and the
1146 function using it serves the rest (unless PURIFY).
1148 #define new_body_inline(xpv, sv_type) \
1150 void ** const r3wt = &PL_body_roots[sv_type]; \
1151 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1152 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1153 bodies_by_type[sv_type].body_size,\
1154 bodies_by_type[sv_type].arena_size)); \
1155 *(r3wt) = *(void**)(xpv); \
1161 S_new_body(pTHX_ const svtype sv_type)
1164 new_body_inline(xpv, sv_type);
1170 static const struct body_details fake_rv =
1171 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1174 =for apidoc sv_upgrade
1176 Upgrade an SV to a more complex form. Generally adds a new body type to the
1177 SV, then copies across as much information as possible from the old body.
1178 It croaks if the SV is already in a more complex form than requested. You
1179 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1180 before calling C<sv_upgrade>, and hence does not croak. See also
1187 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1191 const svtype old_type = SvTYPE(sv);
1192 const struct body_details *new_type_details;
1193 const struct body_details *old_type_details
1194 = bodies_by_type + old_type;
1195 SV *referant = NULL;
1197 PERL_ARGS_ASSERT_SV_UPGRADE;
1199 if (old_type == new_type)
1202 /* This clause was purposefully added ahead of the early return above to
1203 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1204 inference by Nick I-S that it would fix other troublesome cases. See
1205 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1207 Given that shared hash key scalars are no longer PVIV, but PV, there is
1208 no longer need to unshare so as to free up the IVX slot for its proper
1209 purpose. So it's safe to move the early return earlier. */
1211 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1212 sv_force_normal_flags(sv, 0);
1215 old_body = SvANY(sv);
1217 /* Copying structures onto other structures that have been neatly zeroed
1218 has a subtle gotcha. Consider XPVMG
1220 +------+------+------+------+------+-------+-------+
1221 | NV | CUR | LEN | IV | MAGIC | STASH |
1222 +------+------+------+------+------+-------+-------+
1223 0 4 8 12 16 20 24 28
1225 where NVs are aligned to 8 bytes, so that sizeof that structure is
1226 actually 32 bytes long, with 4 bytes of padding at the end:
1228 +------+------+------+------+------+-------+-------+------+
1229 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1230 +------+------+------+------+------+-------+-------+------+
1231 0 4 8 12 16 20 24 28 32
1233 so what happens if you allocate memory for this structure:
1235 +------+------+------+------+------+-------+-------+------+------+...
1236 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1237 +------+------+------+------+------+-------+-------+------+------+...
1238 0 4 8 12 16 20 24 28 32 36
1240 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1241 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1242 started out as zero once, but it's quite possible that it isn't. So now,
1243 rather than a nicely zeroed GP, you have it pointing somewhere random.
1246 (In fact, GP ends up pointing at a previous GP structure, because the
1247 principle cause of the padding in XPVMG getting garbage is a copy of
1248 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1249 this happens to be moot because XPVGV has been re-ordered, with GP
1250 no longer after STASH)
1252 So we are careful and work out the size of used parts of all the
1260 referant = SvRV(sv);
1261 old_type_details = &fake_rv;
1262 if (new_type == SVt_NV)
1263 new_type = SVt_PVNV;
1265 if (new_type < SVt_PVIV) {
1266 new_type = (new_type == SVt_NV)
1267 ? SVt_PVNV : SVt_PVIV;
1272 if (new_type < SVt_PVNV) {
1273 new_type = SVt_PVNV;
1277 assert(new_type > SVt_PV);
1278 assert(SVt_IV < SVt_PV);
1279 assert(SVt_NV < SVt_PV);
1286 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1287 there's no way that it can be safely upgraded, because perl.c
1288 expects to Safefree(SvANY(PL_mess_sv)) */
1289 assert(sv != PL_mess_sv);
1290 /* This flag bit is used to mean other things in other scalar types.
1291 Given that it only has meaning inside the pad, it shouldn't be set
1292 on anything that can get upgraded. */
1293 assert(!SvPAD_TYPED(sv));
1296 if (UNLIKELY(old_type_details->cant_upgrade))
1297 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1298 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1301 if (UNLIKELY(old_type > new_type))
1302 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1303 (int)old_type, (int)new_type);
1305 new_type_details = bodies_by_type + new_type;
1307 SvFLAGS(sv) &= ~SVTYPEMASK;
1308 SvFLAGS(sv) |= new_type;
1310 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1311 the return statements above will have triggered. */
1312 assert (new_type != SVt_NULL);
1315 assert(old_type == SVt_NULL);
1316 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1320 assert(old_type == SVt_NULL);
1321 SvANY(sv) = new_XNV();
1326 assert(new_type_details->body_size);
1329 assert(new_type_details->arena);
1330 assert(new_type_details->arena_size);
1331 /* This points to the start of the allocated area. */
1332 new_body_inline(new_body, new_type);
1333 Zero(new_body, new_type_details->body_size, char);
1334 new_body = ((char *)new_body) - new_type_details->offset;
1336 /* We always allocated the full length item with PURIFY. To do this
1337 we fake things so that arena is false for all 16 types.. */
1338 new_body = new_NOARENAZ(new_type_details);
1340 SvANY(sv) = new_body;
1341 if (new_type == SVt_PVAV) {
1345 if (old_type_details->body_size) {
1348 /* It will have been zeroed when the new body was allocated.
1349 Lets not write to it, in case it confuses a write-back
1355 #ifndef NODEFAULT_SHAREKEYS
1356 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1358 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1359 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1362 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1363 The target created by newSVrv also is, and it can have magic.
1364 However, it never has SvPVX set.
1366 if (old_type == SVt_IV) {
1368 } else if (old_type >= SVt_PV) {
1369 assert(SvPVX_const(sv) == 0);
1372 if (old_type >= SVt_PVMG) {
1373 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1374 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1376 sv->sv_u.svu_array = NULL; /* or svu_hash */
1381 /* XXX Is this still needed? Was it ever needed? Surely as there is
1382 no route from NV to PVIV, NOK can never be true */
1383 assert(!SvNOKp(sv));
1396 assert(new_type_details->body_size);
1397 /* We always allocated the full length item with PURIFY. To do this
1398 we fake things so that arena is false for all 16 types.. */
1399 if(new_type_details->arena) {
1400 /* This points to the start of the allocated area. */
1401 new_body_inline(new_body, new_type);
1402 Zero(new_body, new_type_details->body_size, char);
1403 new_body = ((char *)new_body) - new_type_details->offset;
1405 new_body = new_NOARENAZ(new_type_details);
1407 SvANY(sv) = new_body;
1409 if (old_type_details->copy) {
1410 /* There is now the potential for an upgrade from something without
1411 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1412 int offset = old_type_details->offset;
1413 int length = old_type_details->copy;
1415 if (new_type_details->offset > old_type_details->offset) {
1416 const int difference
1417 = new_type_details->offset - old_type_details->offset;
1418 offset += difference;
1419 length -= difference;
1421 assert (length >= 0);
1423 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1427 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1428 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1429 * correct 0.0 for us. Otherwise, if the old body didn't have an
1430 * NV slot, but the new one does, then we need to initialise the
1431 * freshly created NV slot with whatever the correct bit pattern is
1433 if (old_type_details->zero_nv && !new_type_details->zero_nv
1434 && !isGV_with_GP(sv))
1438 if (UNLIKELY(new_type == SVt_PVIO)) {
1439 IO * const io = MUTABLE_IO(sv);
1440 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1443 /* Clear the stashcache because a new IO could overrule a package
1445 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1446 hv_clear(PL_stashcache);
1448 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1449 IoPAGE_LEN(sv) = 60;
1451 if (UNLIKELY(new_type == SVt_REGEXP))
1452 sv->sv_u.svu_rx = (regexp *)new_body;
1453 else if (old_type < SVt_PV) {
1454 /* referant will be NULL unless the old type was SVt_IV emulating
1456 sv->sv_u.svu_rv = referant;
1460 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1461 (unsigned long)new_type);
1464 if (old_type > SVt_IV) {
1468 /* Note that there is an assumption that all bodies of types that
1469 can be upgraded came from arenas. Only the more complex non-
1470 upgradable types are allowed to be directly malloc()ed. */
1471 assert(old_type_details->arena);
1472 del_body((void*)((char*)old_body + old_type_details->offset),
1473 &PL_body_roots[old_type]);
1479 =for apidoc sv_backoff
1481 Remove any string offset. You should normally use the C<SvOOK_off> macro
1488 Perl_sv_backoff(SV *const sv)
1491 const char * const s = SvPVX_const(sv);
1493 PERL_ARGS_ASSERT_SV_BACKOFF;
1496 assert(SvTYPE(sv) != SVt_PVHV);
1497 assert(SvTYPE(sv) != SVt_PVAV);
1499 SvOOK_offset(sv, delta);
1501 SvLEN_set(sv, SvLEN(sv) + delta);
1502 SvPV_set(sv, SvPVX(sv) - delta);
1503 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1504 SvFLAGS(sv) &= ~SVf_OOK;
1511 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1512 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1513 Use the C<SvGROW> wrapper instead.
1518 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1521 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1525 PERL_ARGS_ASSERT_SV_GROW;
1529 if (SvTYPE(sv) < SVt_PV) {
1530 sv_upgrade(sv, SVt_PV);
1531 s = SvPVX_mutable(sv);
1533 else if (SvOOK(sv)) { /* pv is offset? */
1535 s = SvPVX_mutable(sv);
1536 if (newlen > SvLEN(sv))
1537 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1541 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1542 s = SvPVX_mutable(sv);
1545 #ifdef PERL_NEW_COPY_ON_WRITE
1546 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1547 * to store the COW count. So in general, allocate one more byte than
1548 * asked for, to make it likely this byte is always spare: and thus
1549 * make more strings COW-able.
1550 * If the new size is a big power of two, don't bother: we assume the
1551 * caller wanted a nice 2^N sized block and will be annoyed at getting
1557 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1558 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1561 if (newlen > SvLEN(sv)) { /* need more room? */
1562 STRLEN minlen = SvCUR(sv);
1563 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1564 if (newlen < minlen)
1566 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1568 /* Don't round up on the first allocation, as odds are pretty good that
1569 * the initial request is accurate as to what is really needed */
1571 newlen = PERL_STRLEN_ROUNDUP(newlen);
1574 if (SvLEN(sv) && s) {
1575 s = (char*)saferealloc(s, newlen);
1578 s = (char*)safemalloc(newlen);
1579 if (SvPVX_const(sv) && SvCUR(sv)) {
1580 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1584 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1585 /* Do this here, do it once, do it right, and then we will never get
1586 called back into sv_grow() unless there really is some growing
1588 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1590 SvLEN_set(sv, newlen);
1597 =for apidoc sv_setiv
1599 Copies an integer into the given SV, upgrading first if necessary.
1600 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1606 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1608 PERL_ARGS_ASSERT_SV_SETIV;
1610 SV_CHECK_THINKFIRST_COW_DROP(sv);
1611 switch (SvTYPE(sv)) {
1614 sv_upgrade(sv, SVt_IV);
1617 sv_upgrade(sv, SVt_PVIV);
1621 if (!isGV_with_GP(sv))
1628 /* diag_listed_as: Can't coerce %s to %s in %s */
1629 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1633 (void)SvIOK_only(sv); /* validate number */
1639 =for apidoc sv_setiv_mg
1641 Like C<sv_setiv>, but also handles 'set' magic.
1647 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1649 PERL_ARGS_ASSERT_SV_SETIV_MG;
1656 =for apidoc sv_setuv
1658 Copies an unsigned integer into the given SV, upgrading first if necessary.
1659 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1665 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1667 PERL_ARGS_ASSERT_SV_SETUV;
1669 /* With the if statement to ensure that integers are stored as IVs whenever
1671 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1674 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1676 If you wish to remove the following if statement, so that this routine
1677 (and its callers) always return UVs, please benchmark to see what the
1678 effect is. Modern CPUs may be different. Or may not :-)
1680 if (u <= (UV)IV_MAX) {
1681 sv_setiv(sv, (IV)u);
1690 =for apidoc sv_setuv_mg
1692 Like C<sv_setuv>, but also handles 'set' magic.
1698 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1700 PERL_ARGS_ASSERT_SV_SETUV_MG;
1707 =for apidoc sv_setnv
1709 Copies a double into the given SV, upgrading first if necessary.
1710 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1716 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1718 PERL_ARGS_ASSERT_SV_SETNV;
1720 SV_CHECK_THINKFIRST_COW_DROP(sv);
1721 switch (SvTYPE(sv)) {
1724 sv_upgrade(sv, SVt_NV);
1728 sv_upgrade(sv, SVt_PVNV);
1732 if (!isGV_with_GP(sv))
1739 /* diag_listed_as: Can't coerce %s to %s in %s */
1740 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1745 (void)SvNOK_only(sv); /* validate number */
1750 =for apidoc sv_setnv_mg
1752 Like C<sv_setnv>, but also handles 'set' magic.
1758 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1760 PERL_ARGS_ASSERT_SV_SETNV_MG;
1766 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1767 * not incrementable warning display.
1768 * Originally part of S_not_a_number().
1769 * The return value may be != tmpbuf.
1773 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1776 PERL_ARGS_ASSERT_SV_DISPLAY;
1779 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1780 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1783 const char * const limit = tmpbuf + tmpbuf_size - 8;
1784 /* each *s can expand to 4 chars + "...\0",
1785 i.e. need room for 8 chars */
1787 const char *s = SvPVX_const(sv);
1788 const char * const end = s + SvCUR(sv);
1789 for ( ; s < end && d < limit; s++ ) {
1791 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1795 /* Map to ASCII "equivalent" of Latin1 */
1796 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1802 else if (ch == '\r') {
1806 else if (ch == '\f') {
1810 else if (ch == '\\') {
1814 else if (ch == '\0') {
1818 else if (isPRINT_LC(ch))
1837 /* Print an "isn't numeric" warning, using a cleaned-up,
1838 * printable version of the offending string
1842 S_not_a_number(pTHX_ SV *const sv)
1847 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1849 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1852 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1853 /* diag_listed_as: Argument "%s" isn't numeric%s */
1854 "Argument \"%s\" isn't numeric in %s", pv,
1857 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1858 /* diag_listed_as: Argument "%s" isn't numeric%s */
1859 "Argument \"%s\" isn't numeric", pv);
1863 S_not_incrementable(pTHX_ SV *const sv) {
1867 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1869 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1871 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1872 "Argument \"%s\" treated as 0 in increment (++)", pv);
1876 =for apidoc looks_like_number
1878 Test if the content of an SV looks like a number (or is a number).
1879 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1880 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1887 Perl_looks_like_number(pTHX_ SV *const sv)
1892 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1894 if (SvPOK(sv) || SvPOKp(sv)) {
1895 sbegin = SvPV_nomg_const(sv, len);
1898 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1899 return grok_number(sbegin, len, NULL);
1903 S_glob_2number(pTHX_ GV * const gv)
1905 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1907 /* We know that all GVs stringify to something that is not-a-number,
1908 so no need to test that. */
1909 if (ckWARN(WARN_NUMERIC))
1911 SV *const buffer = sv_newmortal();
1912 gv_efullname3(buffer, gv, "*");
1913 not_a_number(buffer);
1915 /* We just want something true to return, so that S_sv_2iuv_common
1916 can tail call us and return true. */
1920 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1921 until proven guilty, assume that things are not that bad... */
1926 As 64 bit platforms often have an NV that doesn't preserve all bits of
1927 an IV (an assumption perl has been based on to date) it becomes necessary
1928 to remove the assumption that the NV always carries enough precision to
1929 recreate the IV whenever needed, and that the NV is the canonical form.
1930 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1931 precision as a side effect of conversion (which would lead to insanity
1932 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1933 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1934 where precision was lost, and IV/UV/NV slots that have a valid conversion
1935 which has lost no precision
1936 2) to ensure that if a numeric conversion to one form is requested that
1937 would lose precision, the precise conversion (or differently
1938 imprecise conversion) is also performed and cached, to prevent
1939 requests for different numeric formats on the same SV causing
1940 lossy conversion chains. (lossless conversion chains are perfectly
1945 SvIOKp is true if the IV slot contains a valid value
1946 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1947 SvNOKp is true if the NV slot contains a valid value
1948 SvNOK is true only if the NV value is accurate
1951 while converting from PV to NV, check to see if converting that NV to an
1952 IV(or UV) would lose accuracy over a direct conversion from PV to
1953 IV(or UV). If it would, cache both conversions, return NV, but mark
1954 SV as IOK NOKp (ie not NOK).
1956 While converting from PV to IV, check to see if converting that IV to an
1957 NV would lose accuracy over a direct conversion from PV to NV. If it
1958 would, cache both conversions, flag similarly.
1960 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1961 correctly because if IV & NV were set NV *always* overruled.
1962 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1963 changes - now IV and NV together means that the two are interchangeable:
1964 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1966 The benefit of this is that operations such as pp_add know that if
1967 SvIOK is true for both left and right operands, then integer addition
1968 can be used instead of floating point (for cases where the result won't
1969 overflow). Before, floating point was always used, which could lead to
1970 loss of precision compared with integer addition.
1972 * making IV and NV equal status should make maths accurate on 64 bit
1974 * may speed up maths somewhat if pp_add and friends start to use
1975 integers when possible instead of fp. (Hopefully the overhead in
1976 looking for SvIOK and checking for overflow will not outweigh the
1977 fp to integer speedup)
1978 * will slow down integer operations (callers of SvIV) on "inaccurate"
1979 values, as the change from SvIOK to SvIOKp will cause a call into
1980 sv_2iv each time rather than a macro access direct to the IV slot
1981 * should speed up number->string conversion on integers as IV is
1982 favoured when IV and NV are equally accurate
1984 ####################################################################
1985 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1986 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1987 On the other hand, SvUOK is true iff UV.
1988 ####################################################################
1990 Your mileage will vary depending your CPU's relative fp to integer
1994 #ifndef NV_PRESERVES_UV
1995 # define IS_NUMBER_UNDERFLOW_IV 1
1996 # define IS_NUMBER_UNDERFLOW_UV 2
1997 # define IS_NUMBER_IV_AND_UV 2
1998 # define IS_NUMBER_OVERFLOW_IV 4
1999 # define IS_NUMBER_OVERFLOW_UV 5
2001 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2003 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2005 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2011 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2012 PERL_UNUSED_CONTEXT;
2014 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%"UVxf" NV=%"NVgf" inttype=%"UVXf"\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2015 if (SvNVX(sv) < (NV)IV_MIN) {
2016 (void)SvIOKp_on(sv);
2018 SvIV_set(sv, IV_MIN);
2019 return IS_NUMBER_UNDERFLOW_IV;
2021 if (SvNVX(sv) > (NV)UV_MAX) {
2022 (void)SvIOKp_on(sv);
2025 SvUV_set(sv, UV_MAX);
2026 return IS_NUMBER_OVERFLOW_UV;
2028 (void)SvIOKp_on(sv);
2030 /* Can't use strtol etc to convert this string. (See truth table in
2032 if (SvNVX(sv) <= (UV)IV_MAX) {
2033 SvIV_set(sv, I_V(SvNVX(sv)));
2034 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2035 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2037 /* Integer is imprecise. NOK, IOKp */
2039 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2042 SvUV_set(sv, U_V(SvNVX(sv)));
2043 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2044 if (SvUVX(sv) == UV_MAX) {
2045 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2046 possibly be preserved by NV. Hence, it must be overflow.
2048 return IS_NUMBER_OVERFLOW_UV;
2050 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2052 /* Integer is imprecise. NOK, IOKp */
2054 return IS_NUMBER_OVERFLOW_IV;
2056 #endif /* !NV_PRESERVES_UV*/
2059 S_sv_2iuv_common(pTHX_ SV *const sv)
2061 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2064 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2065 * without also getting a cached IV/UV from it at the same time
2066 * (ie PV->NV conversion should detect loss of accuracy and cache
2067 * IV or UV at same time to avoid this. */
2068 /* IV-over-UV optimisation - choose to cache IV if possible */
2070 if (SvTYPE(sv) == SVt_NV)
2071 sv_upgrade(sv, SVt_PVNV);
2073 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2074 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2075 certainly cast into the IV range at IV_MAX, whereas the correct
2076 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2078 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2079 if (Perl_isnan(SvNVX(sv))) {
2085 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2086 SvIV_set(sv, I_V(SvNVX(sv)));
2087 if (SvNVX(sv) == (NV) SvIVX(sv)
2088 #ifndef NV_PRESERVES_UV
2089 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2090 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2091 /* Don't flag it as "accurately an integer" if the number
2092 came from a (by definition imprecise) NV operation, and
2093 we're outside the range of NV integer precision */
2097 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2099 /* scalar has trailing garbage, eg "42a" */
2101 DEBUG_c(PerlIO_printf(Perl_debug_log,
2102 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2108 /* IV not precise. No need to convert from PV, as NV
2109 conversion would already have cached IV if it detected
2110 that PV->IV would be better than PV->NV->IV
2111 flags already correct - don't set public IOK. */
2112 DEBUG_c(PerlIO_printf(Perl_debug_log,
2113 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2118 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2119 but the cast (NV)IV_MIN rounds to a the value less (more
2120 negative) than IV_MIN which happens to be equal to SvNVX ??
2121 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2122 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2123 (NV)UVX == NVX are both true, but the values differ. :-(
2124 Hopefully for 2s complement IV_MIN is something like
2125 0x8000000000000000 which will be exact. NWC */
2128 SvUV_set(sv, U_V(SvNVX(sv)));
2130 (SvNVX(sv) == (NV) SvUVX(sv))
2131 #ifndef NV_PRESERVES_UV
2132 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2133 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2134 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2135 /* Don't flag it as "accurately an integer" if the number
2136 came from a (by definition imprecise) NV operation, and
2137 we're outside the range of NV integer precision */
2143 DEBUG_c(PerlIO_printf(Perl_debug_log,
2144 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2150 else if (SvPOKp(sv)) {
2152 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2153 /* We want to avoid a possible problem when we cache an IV/ a UV which
2154 may be later translated to an NV, and the resulting NV is not
2155 the same as the direct translation of the initial string
2156 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2157 be careful to ensure that the value with the .456 is around if the
2158 NV value is requested in the future).
2160 This means that if we cache such an IV/a UV, we need to cache the
2161 NV as well. Moreover, we trade speed for space, and do not
2162 cache the NV if we are sure it's not needed.
2165 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2166 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2167 == IS_NUMBER_IN_UV) {
2168 /* It's definitely an integer, only upgrade to PVIV */
2169 if (SvTYPE(sv) < SVt_PVIV)
2170 sv_upgrade(sv, SVt_PVIV);
2172 } else if (SvTYPE(sv) < SVt_PVNV)
2173 sv_upgrade(sv, SVt_PVNV);
2175 /* If NVs preserve UVs then we only use the UV value if we know that
2176 we aren't going to call atof() below. If NVs don't preserve UVs
2177 then the value returned may have more precision than atof() will
2178 return, even though value isn't perfectly accurate. */
2179 if ((numtype & (IS_NUMBER_IN_UV
2180 #ifdef NV_PRESERVES_UV
2183 )) == IS_NUMBER_IN_UV) {
2184 /* This won't turn off the public IOK flag if it was set above */
2185 (void)SvIOKp_on(sv);
2187 if (!(numtype & IS_NUMBER_NEG)) {
2189 if (value <= (UV)IV_MAX) {
2190 SvIV_set(sv, (IV)value);
2192 /* it didn't overflow, and it was positive. */
2193 SvUV_set(sv, value);
2197 /* 2s complement assumption */
2198 if (value <= (UV)IV_MIN) {
2199 SvIV_set(sv, -(IV)value);
2201 /* Too negative for an IV. This is a double upgrade, but
2202 I'm assuming it will be rare. */
2203 if (SvTYPE(sv) < SVt_PVNV)
2204 sv_upgrade(sv, SVt_PVNV);
2208 SvNV_set(sv, -(NV)value);
2209 SvIV_set(sv, IV_MIN);
2213 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2214 will be in the previous block to set the IV slot, and the next
2215 block to set the NV slot. So no else here. */
2217 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2218 != IS_NUMBER_IN_UV) {
2219 /* It wasn't an (integer that doesn't overflow the UV). */
2220 SvNV_set(sv, Atof(SvPVX_const(sv)));
2222 if (! numtype && ckWARN(WARN_NUMERIC))
2225 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2226 PTR2UV(sv), SvNVX(sv)));
2228 #ifdef NV_PRESERVES_UV
2229 (void)SvIOKp_on(sv);
2231 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2232 if (Perl_isnan(SvNVX(sv))) {
2238 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2239 SvIV_set(sv, I_V(SvNVX(sv)));
2240 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2243 NOOP; /* Integer is imprecise. NOK, IOKp */
2245 /* UV will not work better than IV */
2247 if (SvNVX(sv) > (NV)UV_MAX) {
2249 /* Integer is inaccurate. NOK, IOKp, is UV */
2250 SvUV_set(sv, UV_MAX);
2252 SvUV_set(sv, U_V(SvNVX(sv)));
2253 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2254 NV preservse UV so can do correct comparison. */
2255 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2258 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2263 #else /* NV_PRESERVES_UV */
2264 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2265 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2266 /* The IV/UV slot will have been set from value returned by
2267 grok_number above. The NV slot has just been set using
2270 assert (SvIOKp(sv));
2272 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2273 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2274 /* Small enough to preserve all bits. */
2275 (void)SvIOKp_on(sv);
2277 SvIV_set(sv, I_V(SvNVX(sv)));
2278 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2280 /* Assumption: first non-preserved integer is < IV_MAX,
2281 this NV is in the preserved range, therefore: */
2282 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2284 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2288 0 0 already failed to read UV.
2289 0 1 already failed to read UV.
2290 1 0 you won't get here in this case. IV/UV
2291 slot set, public IOK, Atof() unneeded.
2292 1 1 already read UV.
2293 so there's no point in sv_2iuv_non_preserve() attempting
2294 to use atol, strtol, strtoul etc. */
2296 sv_2iuv_non_preserve (sv, numtype);
2298 sv_2iuv_non_preserve (sv);
2302 #endif /* NV_PRESERVES_UV */
2303 /* It might be more code efficient to go through the entire logic above
2304 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2305 gets complex and potentially buggy, so more programmer efficient
2306 to do it this way, by turning off the public flags: */
2308 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2312 if (isGV_with_GP(sv))
2313 return glob_2number(MUTABLE_GV(sv));
2315 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2317 if (SvTYPE(sv) < SVt_IV)
2318 /* Typically the caller expects that sv_any is not NULL now. */
2319 sv_upgrade(sv, SVt_IV);
2320 /* Return 0 from the caller. */
2327 =for apidoc sv_2iv_flags
2329 Return the integer value of an SV, doing any necessary string
2330 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2331 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2337 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2339 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2341 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2342 && SvTYPE(sv) != SVt_PVFM);
2344 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2350 if (flags & SV_SKIP_OVERLOAD)
2352 tmpstr = AMG_CALLunary(sv, numer_amg);
2353 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2354 return SvIV(tmpstr);
2357 return PTR2IV(SvRV(sv));
2360 if (SvVALID(sv) || isREGEXP(sv)) {
2361 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2362 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2363 In practice they are extremely unlikely to actually get anywhere
2364 accessible by user Perl code - the only way that I'm aware of is when
2365 a constant subroutine which is used as the second argument to index.
2367 Regexps have no SvIVX and SvNVX fields.
2369 assert(isREGEXP(sv) || SvPOKp(sv));
2372 const char * const ptr =
2373 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2375 = grok_number(ptr, SvCUR(sv), &value);
2377 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2378 == IS_NUMBER_IN_UV) {
2379 /* It's definitely an integer */
2380 if (numtype & IS_NUMBER_NEG) {
2381 if (value < (UV)IV_MIN)
2384 if (value < (UV)IV_MAX)
2389 /* Quite wrong but no good choices. */
2390 if ((numtype & IS_NUMBER_INFINITY)) {
2391 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2392 } else if ((numtype & IS_NUMBER_NAN)) {
2393 return 0; /* So wrong. */
2397 if (ckWARN(WARN_NUMERIC))
2400 return I_V(Atof(ptr));
2404 if (SvTHINKFIRST(sv)) {
2405 #ifdef PERL_OLD_COPY_ON_WRITE
2407 sv_force_normal_flags(sv, 0);
2410 if (SvREADONLY(sv) && !SvOK(sv)) {
2411 if (ckWARN(WARN_UNINITIALIZED))
2418 if (S_sv_2iuv_common(aTHX_ sv))
2422 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2423 PTR2UV(sv),SvIVX(sv)));
2424 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2428 =for apidoc sv_2uv_flags
2430 Return the unsigned integer value of an SV, doing any necessary string
2431 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2432 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2438 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2440 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2442 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2448 if (flags & SV_SKIP_OVERLOAD)
2450 tmpstr = AMG_CALLunary(sv, numer_amg);
2451 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2452 return SvUV(tmpstr);
2455 return PTR2UV(SvRV(sv));
2458 if (SvVALID(sv) || isREGEXP(sv)) {
2459 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2460 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2461 Regexps have no SvIVX and SvNVX fields. */
2462 assert(isREGEXP(sv) || SvPOKp(sv));
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2468 = grok_number(ptr, SvCUR(sv), &value);
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (!(numtype & IS_NUMBER_NEG))
2477 /* Quite wrong but no good choices. */
2478 if ((numtype & IS_NUMBER_INFINITY)) {
2479 return UV_MAX; /* So wrong. */
2480 } else if ((numtype & IS_NUMBER_NAN)) {
2481 return 0; /* So wrong. */
2485 if (ckWARN(WARN_NUMERIC))
2488 return U_V(Atof(ptr));
2492 if (SvTHINKFIRST(sv)) {
2493 #ifdef PERL_OLD_COPY_ON_WRITE
2495 sv_force_normal_flags(sv, 0);
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2506 if (S_sv_2iuv_common(aTHX_ sv))
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2511 PTR2UV(sv),SvUVX(sv)));
2512 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2516 =for apidoc sv_2nv_flags
2518 Return the num value of an SV, doing any necessary string or integer
2519 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2520 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2526 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2528 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2530 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2531 && SvTYPE(sv) != SVt_PVFM);
2532 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2533 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2534 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2535 Regexps have no SvIVX and SvNVX fields. */
2537 if (flags & SV_GMAGIC)
2541 if (SvPOKp(sv) && !SvIOKp(sv)) {
2542 ptr = SvPVX_const(sv);
2544 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2545 !grok_number(ptr, SvCUR(sv), NULL))
2551 return (NV)SvUVX(sv);
2553 return (NV)SvIVX(sv);
2559 ptr = RX_WRAPPED((REGEXP *)sv);
2562 assert(SvTYPE(sv) >= SVt_PVMG);
2563 /* This falls through to the report_uninit near the end of the
2565 } else if (SvTHINKFIRST(sv)) {
2570 if (flags & SV_SKIP_OVERLOAD)
2572 tmpstr = AMG_CALLunary(sv, numer_amg);
2573 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2574 return SvNV(tmpstr);
2577 return PTR2NV(SvRV(sv));
2579 #ifdef PERL_OLD_COPY_ON_WRITE
2581 sv_force_normal_flags(sv, 0);
2584 if (SvREADONLY(sv) && !SvOK(sv)) {
2585 if (ckWARN(WARN_UNINITIALIZED))
2590 if (SvTYPE(sv) < SVt_NV) {
2591 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2592 sv_upgrade(sv, SVt_NV);
2594 STORE_NUMERIC_LOCAL_SET_STANDARD();
2595 PerlIO_printf(Perl_debug_log,
2596 "0x%"UVxf" num(%" NVgf ")\n",
2597 PTR2UV(sv), SvNVX(sv));
2598 RESTORE_NUMERIC_LOCAL();
2601 else if (SvTYPE(sv) < SVt_PVNV)
2602 sv_upgrade(sv, SVt_PVNV);
2607 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2608 #ifdef NV_PRESERVES_UV
2614 /* Only set the public NV OK flag if this NV preserves the IV */
2615 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2617 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2618 : (SvIVX(sv) == I_V(SvNVX(sv))))
2624 else if (SvPOKp(sv)) {
2626 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2627 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2629 #ifdef NV_PRESERVES_UV
2630 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2631 == IS_NUMBER_IN_UV) {
2632 /* It's definitely an integer */
2633 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2635 if ((numtype & IS_NUMBER_INFINITY)) {
2636 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2637 } else if ((numtype & IS_NUMBER_NAN)) {
2638 SvNV_set(sv, NV_NAN);
2640 SvNV_set(sv, Atof(SvPVX_const(sv)));
2647 SvNV_set(sv, Atof(SvPVX_const(sv)));
2648 /* Only set the public NV OK flag if this NV preserves the value in
2649 the PV at least as well as an IV/UV would.
2650 Not sure how to do this 100% reliably. */
2651 /* if that shift count is out of range then Configure's test is
2652 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2654 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2655 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2656 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2657 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2658 /* Can't use strtol etc to convert this string, so don't try.
2659 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2662 /* value has been set. It may not be precise. */
2663 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2664 /* 2s complement assumption for (UV)IV_MIN */
2665 SvNOK_on(sv); /* Integer is too negative. */
2670 if (numtype & IS_NUMBER_NEG) {
2671 SvIV_set(sv, -(IV)value);
2672 } else if (value <= (UV)IV_MAX) {
2673 SvIV_set(sv, (IV)value);
2675 SvUV_set(sv, value);
2679 if (numtype & IS_NUMBER_NOT_INT) {
2680 /* I believe that even if the original PV had decimals,
2681 they are lost beyond the limit of the FP precision.
2682 However, neither is canonical, so both only get p
2683 flags. NWC, 2000/11/25 */
2684 /* Both already have p flags, so do nothing */
2686 const NV nv = SvNVX(sv);
2687 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2688 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2689 if (SvIVX(sv) == I_V(nv)) {
2692 /* It had no "." so it must be integer. */
2696 /* between IV_MAX and NV(UV_MAX).
2697 Could be slightly > UV_MAX */
2699 if (numtype & IS_NUMBER_NOT_INT) {
2700 /* UV and NV both imprecise. */
2702 const UV nv_as_uv = U_V(nv);
2704 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2713 /* It might be more code efficient to go through the entire logic above
2714 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2715 gets complex and potentially buggy, so more programmer efficient
2716 to do it this way, by turning off the public flags: */
2718 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2719 #endif /* NV_PRESERVES_UV */
2722 if (isGV_with_GP(sv)) {
2723 glob_2number(MUTABLE_GV(sv));
2727 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2729 assert (SvTYPE(sv) >= SVt_NV);
2730 /* Typically the caller expects that sv_any is not NULL now. */
2731 /* XXX Ilya implies that this is a bug in callers that assume this
2732 and ideally should be fixed. */
2736 STORE_NUMERIC_LOCAL_SET_STANDARD();
2737 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2738 PTR2UV(sv), SvNVX(sv));
2739 RESTORE_NUMERIC_LOCAL();
2747 Return an SV with the numeric value of the source SV, doing any necessary
2748 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2749 access this function.
2755 Perl_sv_2num(pTHX_ SV *const sv)
2757 PERL_ARGS_ASSERT_SV_2NUM;
2762 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2763 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2764 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2765 return sv_2num(tmpsv);
2767 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2770 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2771 * UV as a string towards the end of buf, and return pointers to start and
2774 * We assume that buf is at least TYPE_CHARS(UV) long.
2778 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2780 char *ptr = buf + TYPE_CHARS(UV);
2781 char * const ebuf = ptr;
2784 PERL_ARGS_ASSERT_UIV_2BUF;
2796 *--ptr = '0' + (char)(uv % 10);
2804 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2805 * infinity or a not-a-number, writes the appropriate strings to the
2806 * buffer, including a zero byte. On success returns the written length,
2807 * excluding the zero byte, on failure (not an infinity, not a nan, or the
2808 * maxlen too small) returns zero. */
2810 S_infnan_2pv(NV nv, char* buffer, size_t maxlen) {
2811 /* XXX this should be an assert */
2812 if (maxlen < 4) /* "Inf\0", "NaN\0" */
2816 /* isnan must be first due to NAN_COMPARE_BROKEN builds, since NAN might
2817 use the broken for NAN >/< ops in the inf check, and then the inf
2818 check returns true for NAN on NAN_COMPARE_BROKEN compilers */
2819 if (Perl_isnan(nv)) {
2823 /* XXX optionally output the payload mantissa bits as
2824 * "(unsigned)" (to match the nan("...") C99 function,
2825 * or maybe as "(0xhhh...)" would make more sense...
2826 * provide a format string so that the user can decide?
2827 * NOTE: would affect the maxlen and assert() logic.*/
2829 else if (Perl_isinf(nv)) {
2831 if (maxlen < 5) /* "-Inf\0" */
2842 assert((s == buffer + 3) || (s == buffer + 4));
2844 return s - buffer - 1; /* -1: excluding the zero byte */
2849 =for apidoc sv_2pv_flags
2851 Returns a pointer to the string value of an SV, and sets *lp to its length.
2852 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2853 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2854 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2860 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2864 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2866 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2867 && SvTYPE(sv) != SVt_PVFM);
2868 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2873 if (flags & SV_SKIP_OVERLOAD)
2875 tmpstr = AMG_CALLunary(sv, string_amg);
2876 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2877 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2879 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2883 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2884 if (flags & SV_CONST_RETURN) {
2885 pv = (char *) SvPVX_const(tmpstr);
2887 pv = (flags & SV_MUTABLE_RETURN)
2888 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2891 *lp = SvCUR(tmpstr);
2893 pv = sv_2pv_flags(tmpstr, lp, flags);
2906 SV *const referent = SvRV(sv);
2910 retval = buffer = savepvn("NULLREF", len);
2911 } else if (SvTYPE(referent) == SVt_REGEXP &&
2912 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2913 amagic_is_enabled(string_amg))) {
2914 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2918 /* If the regex is UTF-8 we want the containing scalar to
2919 have an UTF-8 flag too */
2926 *lp = RX_WRAPLEN(re);
2928 return RX_WRAPPED(re);
2930 const char *const typestr = sv_reftype(referent, 0);
2931 const STRLEN typelen = strlen(typestr);
2932 UV addr = PTR2UV(referent);
2933 const char *stashname = NULL;
2934 STRLEN stashnamelen = 0; /* hush, gcc */
2935 const char *buffer_end;
2937 if (SvOBJECT(referent)) {
2938 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2941 stashname = HEK_KEY(name);
2942 stashnamelen = HEK_LEN(name);
2944 if (HEK_UTF8(name)) {
2950 stashname = "__ANON__";
2953 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2954 + 2 * sizeof(UV) + 2 /* )\0 */;
2956 len = typelen + 3 /* (0x */
2957 + 2 * sizeof(UV) + 2 /* )\0 */;
2960 Newx(buffer, len, char);
2961 buffer_end = retval = buffer + len;
2963 /* Working backwards */
2967 *--retval = PL_hexdigit[addr & 15];
2968 } while (addr >>= 4);
2974 memcpy(retval, typestr, typelen);
2978 retval -= stashnamelen;
2979 memcpy(retval, stashname, stashnamelen);
2981 /* retval may not necessarily have reached the start of the
2983 assert (retval >= buffer);
2985 len = buffer_end - retval - 1; /* -1 for that \0 */
2997 if (flags & SV_MUTABLE_RETURN)
2998 return SvPVX_mutable(sv);
2999 if (flags & SV_CONST_RETURN)
3000 return (char *)SvPVX_const(sv);
3005 /* I'm assuming that if both IV and NV are equally valid then
3006 converting the IV is going to be more efficient */
3007 const U32 isUIOK = SvIsUV(sv);
3008 char buf[TYPE_CHARS(UV)];
3012 if (SvTYPE(sv) < SVt_PVIV)
3013 sv_upgrade(sv, SVt_PVIV);
3014 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3016 /* inlined from sv_setpvn */
3017 s = SvGROW_mutable(sv, len + 1);
3018 Move(ptr, s, len, char);
3023 else if (SvNOK(sv)) {
3024 if (SvTYPE(sv) < SVt_PVNV)
3025 sv_upgrade(sv, SVt_PVNV);
3026 if (SvNVX(sv) == 0.0
3027 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3028 && !Perl_isnan(SvNVX(sv))
3031 s = SvGROW_mutable(sv, 2);
3035 /* The +20 is pure guesswork. Configure test needed. --jhi */
3036 STRLEN size = NV_DIG + 20;
3038 s = SvGROW_mutable(sv, size);
3040 len = S_infnan_2pv(SvNVX(sv), s, size);
3045 /* some Xenix systems wipe out errno here */
3047 #ifndef USE_LOCALE_NUMERIC
3048 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3052 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3053 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3055 /* If the radix character is UTF-8, and actually is in the
3056 * output, turn on the UTF-8 flag for the scalar */
3057 if (PL_numeric_local
3058 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
3059 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3063 RESTORE_LC_NUMERIC();
3066 /* We don't call SvPOK_on(), because it may come to
3067 * pass that the locale changes so that the
3068 * stringification we just did is no longer correct. We
3069 * will have to re-stringify every time it is needed */
3076 else if (isGV_with_GP(sv)) {
3077 GV *const gv = MUTABLE_GV(sv);
3078 SV *const buffer = sv_newmortal();
3080 gv_efullname3(buffer, gv, "*");
3082 assert(SvPOK(buffer));
3086 *lp = SvCUR(buffer);
3087 return SvPVX(buffer);
3089 else if (isREGEXP(sv)) {
3090 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3091 return RX_WRAPPED((REGEXP *)sv);
3096 if (flags & SV_UNDEF_RETURNS_NULL)
3098 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3100 /* Typically the caller expects that sv_any is not NULL now. */
3101 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3102 sv_upgrade(sv, SVt_PV);
3107 const STRLEN len = s - SvPVX_const(sv);
3112 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3113 PTR2UV(sv),SvPVX_const(sv)));
3114 if (flags & SV_CONST_RETURN)
3115 return (char *)SvPVX_const(sv);
3116 if (flags & SV_MUTABLE_RETURN)
3117 return SvPVX_mutable(sv);
3122 =for apidoc sv_copypv
3124 Copies a stringified representation of the source SV into the
3125 destination SV. Automatically performs any necessary mg_get and
3126 coercion of numeric values into strings. Guaranteed to preserve
3127 UTF8 flag even from overloaded objects. Similar in nature to
3128 sv_2pv[_flags] but operates directly on an SV instead of just the
3129 string. Mostly uses sv_2pv_flags to do its work, except when that
3130 would lose the UTF-8'ness of the PV.
3132 =for apidoc sv_copypv_nomg
3134 Like sv_copypv, but doesn't invoke get magic first.
3136 =for apidoc sv_copypv_flags
3138 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3145 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3147 PERL_ARGS_ASSERT_SV_COPYPV;
3149 sv_copypv_flags(dsv, ssv, 0);
3153 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3158 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3160 s = SvPV_flags_const(ssv,len,flags & SV_GMAGIC);
3161 sv_setpvn(dsv,s,len);
3169 =for apidoc sv_2pvbyte
3171 Return a pointer to the byte-encoded representation of the SV, and set *lp
3172 to its length. May cause the SV to be downgraded from UTF-8 as a
3175 Usually accessed via the C<SvPVbyte> macro.
3181 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3183 PERL_ARGS_ASSERT_SV_2PVBYTE;
3186 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3187 || isGV_with_GP(sv) || SvROK(sv)) {
3188 SV *sv2 = sv_newmortal();
3189 sv_copypv_nomg(sv2,sv);
3192 sv_utf8_downgrade(sv,0);
3193 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3197 =for apidoc sv_2pvutf8
3199 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3200 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3202 Usually accessed via the C<SvPVutf8> macro.
3208 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3210 PERL_ARGS_ASSERT_SV_2PVUTF8;
3212 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3213 || isGV_with_GP(sv) || SvROK(sv))
3214 sv = sv_mortalcopy(sv);
3217 sv_utf8_upgrade_nomg(sv);
3218 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3223 =for apidoc sv_2bool
3225 This macro is only used by sv_true() or its macro equivalent, and only if
3226 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3227 It calls sv_2bool_flags with the SV_GMAGIC flag.
3229 =for apidoc sv_2bool_flags
3231 This function is only used by sv_true() and friends, and only if
3232 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3233 contain SV_GMAGIC, then it does an mg_get() first.
3240 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3242 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3245 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3251 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3252 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3255 if(SvGMAGICAL(sv)) {
3257 goto restart; /* call sv_2bool */
3259 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3260 else if(!SvOK(sv)) {
3263 else if(SvPOK(sv)) {
3264 svb = SvPVXtrue(sv);
3266 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3267 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3268 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3272 goto restart; /* call sv_2bool_nomg */
3277 return SvRV(sv) != 0;
3281 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3282 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3286 =for apidoc sv_utf8_upgrade
3288 Converts the PV of an SV to its UTF-8-encoded form.
3289 Forces the SV to string form if it is not already.
3290 Will C<mg_get> on C<sv> if appropriate.
3291 Always sets the SvUTF8 flag to avoid future validity checks even
3292 if the whole string is the same in UTF-8 as not.
3293 Returns the number of bytes in the converted string
3295 This is not a general purpose byte encoding to Unicode interface:
3296 use the Encode extension for that.
3298 =for apidoc sv_utf8_upgrade_nomg
3300 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3302 =for apidoc sv_utf8_upgrade_flags
3304 Converts the PV of an SV to its UTF-8-encoded form.
3305 Forces the SV to string form if it is not already.
3306 Always sets the SvUTF8 flag to avoid future validity checks even
3307 if all the bytes are invariant in UTF-8.
3308 If C<flags> has C<SV_GMAGIC> bit set,
3309 will C<mg_get> on C<sv> if appropriate, else not.
3311 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3312 will expand when converted to UTF-8, and skips the extra work of checking for
3313 that. Typically this flag is used by a routine that has already parsed the
3314 string and found such characters, and passes this information on so that the
3315 work doesn't have to be repeated.
3317 Returns the number of bytes in the converted string.
3319 This is not a general purpose byte encoding to Unicode interface:
3320 use the Encode extension for that.
3322 =for apidoc sv_utf8_upgrade_flags_grow
3324 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3325 the number of unused bytes the string of 'sv' is guaranteed to have free after
3326 it upon return. This allows the caller to reserve extra space that it intends
3327 to fill, to avoid extra grows.
3329 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3330 are implemented in terms of this function.
3332 Returns the number of bytes in the converted string (not including the spares).
3336 (One might think that the calling routine could pass in the position of the
3337 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3338 have to be found again. But that is not the case, because typically when the
3339 caller is likely to use this flag, it won't be calling this routine unless it
3340 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3341 and just use bytes. But some things that do fit into a byte are variants in
3342 utf8, and the caller may not have been keeping track of these.)
3344 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3345 C<NUL> isn't guaranteed due to having other routines do the work in some input
3346 cases, or if the input is already flagged as being in utf8.
3348 The speed of this could perhaps be improved for many cases if someone wanted to
3349 write a fast function that counts the number of variant characters in a string,
3350 especially if it could return the position of the first one.
3355 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3357 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3359 if (sv == &PL_sv_undef)
3361 if (!SvPOK_nog(sv)) {
3363 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3364 (void) sv_2pv_flags(sv,&len, flags);
3366 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3370 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3375 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3380 S_sv_uncow(aTHX_ sv, 0);
3383 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3384 sv_recode_to_utf8(sv, PL_encoding);
3385 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3389 if (SvCUR(sv) == 0) {
3390 if (extra) SvGROW(sv, extra);
3391 } else { /* Assume Latin-1/EBCDIC */
3392 /* This function could be much more efficient if we
3393 * had a FLAG in SVs to signal if there are any variant
3394 * chars in the PV. Given that there isn't such a flag
3395 * make the loop as fast as possible (although there are certainly ways
3396 * to speed this up, eg. through vectorization) */
3397 U8 * s = (U8 *) SvPVX_const(sv);
3398 U8 * e = (U8 *) SvEND(sv);
3400 STRLEN two_byte_count = 0;
3402 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3404 /* See if really will need to convert to utf8. We mustn't rely on our
3405 * incoming SV being well formed and having a trailing '\0', as certain
3406 * code in pp_formline can send us partially built SVs. */
3410 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3412 t--; /* t already incremented; re-point to first variant */
3417 /* utf8 conversion not needed because all are invariants. Mark as
3418 * UTF-8 even if no variant - saves scanning loop */
3420 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3425 /* Here, the string should be converted to utf8, either because of an
3426 * input flag (two_byte_count = 0), or because a character that
3427 * requires 2 bytes was found (two_byte_count = 1). t points either to
3428 * the beginning of the string (if we didn't examine anything), or to
3429 * the first variant. In either case, everything from s to t - 1 will
3430 * occupy only 1 byte each on output.
3432 * There are two main ways to convert. One is to create a new string
3433 * and go through the input starting from the beginning, appending each
3434 * converted value onto the new string as we go along. It's probably
3435 * best to allocate enough space in the string for the worst possible
3436 * case rather than possibly running out of space and having to
3437 * reallocate and then copy what we've done so far. Since everything
3438 * from s to t - 1 is invariant, the destination can be initialized
3439 * with these using a fast memory copy
3441 * The other way is to figure out exactly how big the string should be
3442 * by parsing the entire input. Then you don't have to make it big
3443 * enough to handle the worst possible case, and more importantly, if
3444 * the string you already have is large enough, you don't have to
3445 * allocate a new string, you can copy the last character in the input
3446 * string to the final position(s) that will be occupied by the
3447 * converted string and go backwards, stopping at t, since everything
3448 * before that is invariant.
3450 * There are advantages and disadvantages to each method.
3452 * In the first method, we can allocate a new string, do the memory
3453 * copy from the s to t - 1, and then proceed through the rest of the
3454 * string byte-by-byte.
3456 * In the second method, we proceed through the rest of the input
3457 * string just calculating how big the converted string will be. Then
3458 * there are two cases:
3459 * 1) if the string has enough extra space to handle the converted
3460 * value. We go backwards through the string, converting until we
3461 * get to the position we are at now, and then stop. If this
3462 * position is far enough along in the string, this method is
3463 * faster than the other method. If the memory copy were the same
3464 * speed as the byte-by-byte loop, that position would be about
3465 * half-way, as at the half-way mark, parsing to the end and back
3466 * is one complete string's parse, the same amount as starting
3467 * over and going all the way through. Actually, it would be
3468 * somewhat less than half-way, as it's faster to just count bytes
3469 * than to also copy, and we don't have the overhead of allocating
3470 * a new string, changing the scalar to use it, and freeing the
3471 * existing one. But if the memory copy is fast, the break-even
3472 * point is somewhere after half way. The counting loop could be
3473 * sped up by vectorization, etc, to move the break-even point
3474 * further towards the beginning.
3475 * 2) if the string doesn't have enough space to handle the converted
3476 * value. A new string will have to be allocated, and one might
3477 * as well, given that, start from the beginning doing the first
3478 * method. We've spent extra time parsing the string and in
3479 * exchange all we've gotten is that we know precisely how big to
3480 * make the new one. Perl is more optimized for time than space,
3481 * so this case is a loser.
3482 * So what I've decided to do is not use the 2nd method unless it is
3483 * guaranteed that a new string won't have to be allocated, assuming
3484 * the worst case. I also decided not to put any more conditions on it
3485 * than this, for now. It seems likely that, since the worst case is
3486 * twice as big as the unknown portion of the string (plus 1), we won't
3487 * be guaranteed enough space, causing us to go to the first method,
3488 * unless the string is short, or the first variant character is near
3489 * the end of it. In either of these cases, it seems best to use the
3490 * 2nd method. The only circumstance I can think of where this would
3491 * be really slower is if the string had once had much more data in it
3492 * than it does now, but there is still a substantial amount in it */
3495 STRLEN invariant_head = t - s;
3496 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3497 if (SvLEN(sv) < size) {
3499 /* Here, have decided to allocate a new string */
3504 Newx(dst, size, U8);
3506 /* If no known invariants at the beginning of the input string,
3507 * set so starts from there. Otherwise, can use memory copy to
3508 * get up to where we are now, and then start from here */
3510 if (invariant_head == 0) {
3513 Copy(s, dst, invariant_head, char);
3514 d = dst + invariant_head;
3518 append_utf8_from_native_byte(*t, &d);
3522 SvPV_free(sv); /* No longer using pre-existing string */
3523 SvPV_set(sv, (char*)dst);
3524 SvCUR_set(sv, d - dst);
3525 SvLEN_set(sv, size);
3528 /* Here, have decided to get the exact size of the string.
3529 * Currently this happens only when we know that there is
3530 * guaranteed enough space to fit the converted string, so
3531 * don't have to worry about growing. If two_byte_count is 0,
3532 * then t points to the first byte of the string which hasn't
3533 * been examined yet. Otherwise two_byte_count is 1, and t
3534 * points to the first byte in the string that will expand to
3535 * two. Depending on this, start examining at t or 1 after t.
3538 U8 *d = t + two_byte_count;
3541 /* Count up the remaining bytes that expand to two */
3544 const U8 chr = *d++;
3545 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3548 /* The string will expand by just the number of bytes that
3549 * occupy two positions. But we are one afterwards because of
3550 * the increment just above. This is the place to put the
3551 * trailing NUL, and to set the length before we decrement */
3553 d += two_byte_count;
3554 SvCUR_set(sv, d - s);
3558 /* Having decremented d, it points to the position to put the
3559 * very last byte of the expanded string. Go backwards through
3560 * the string, copying and expanding as we go, stopping when we
3561 * get to the part that is invariant the rest of the way down */
3565 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3568 *d-- = UTF8_EIGHT_BIT_LO(*e);
3569 *d-- = UTF8_EIGHT_BIT_HI(*e);
3575 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3576 /* Update pos. We do it at the end rather than during
3577 * the upgrade, to avoid slowing down the common case
3578 * (upgrade without pos).
3579 * pos can be stored as either bytes or characters. Since
3580 * this was previously a byte string we can just turn off
3581 * the bytes flag. */
3582 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3584 mg->mg_flags &= ~MGf_BYTES;
3586 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3587 magic_setutf8(sv,mg); /* clear UTF8 cache */
3592 /* Mark as UTF-8 even if no variant - saves scanning loop */
3598 =for apidoc sv_utf8_downgrade
3600 Attempts to convert the PV of an SV from characters to bytes.
3601 If the PV contains a character that cannot fit
3602 in a byte, this conversion will fail;
3603 in this case, either returns false or, if C<fail_ok> is not
3606 This is not a general purpose Unicode to byte encoding interface:
3607 use the Encode extension for that.
3613 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3615 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3617 if (SvPOKp(sv) && SvUTF8(sv)) {
3621 int mg_flags = SV_GMAGIC;
3624 S_sv_uncow(aTHX_ sv, 0);
3626 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3628 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3629 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3630 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3631 SV_GMAGIC|SV_CONST_RETURN);
3632 mg_flags = 0; /* sv_pos_b2u does get magic */
3634 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3635 magic_setutf8(sv,mg); /* clear UTF8 cache */
3638 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3640 if (!utf8_to_bytes(s, &len)) {
3645 Perl_croak(aTHX_ "Wide character in %s",
3648 Perl_croak(aTHX_ "Wide character");
3659 =for apidoc sv_utf8_encode
3661 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3662 flag off so that it looks like octets again.
3668 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3670 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3672 if (SvREADONLY(sv)) {
3673 sv_force_normal_flags(sv, 0);
3675 (void) sv_utf8_upgrade(sv);
3680 =for apidoc sv_utf8_decode
3682 If the PV of the SV is an octet sequence in UTF-8
3683 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3684 so that it looks like a character. If the PV contains only single-byte
3685 characters, the C<SvUTF8> flag stays off.
3686 Scans PV for validity and returns false if the PV is invalid UTF-8.
3692 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3694 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3697 const U8 *start, *c;
3700 /* The octets may have got themselves encoded - get them back as
3703 if (!sv_utf8_downgrade(sv, TRUE))
3706 /* it is actually just a matter of turning the utf8 flag on, but
3707 * we want to make sure everything inside is valid utf8 first.
3709 c = start = (const U8 *) SvPVX_const(sv);
3710 if (!is_utf8_string(c, SvCUR(sv)))
3712 e = (const U8 *) SvEND(sv);
3715 if (!UTF8_IS_INVARIANT(ch)) {
3720 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3721 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3722 after this, clearing pos. Does anything on CPAN
3724 /* adjust pos to the start of a UTF8 char sequence */
3725 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3727 I32 pos = mg->mg_len;
3729 for (c = start + pos; c > start; c--) {
3730 if (UTF8_IS_START(*c))
3733 mg->mg_len = c - start;
3736 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3737 magic_setutf8(sv,mg); /* clear UTF8 cache */
3744 =for apidoc sv_setsv
3746 Copies the contents of the source SV C<ssv> into the destination SV
3747 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3748 function if the source SV needs to be reused. Does not handle 'set' magic on
3749 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3750 performs a copy-by-value, obliterating any previous content of the
3753 You probably want to use one of the assortment of wrappers, such as
3754 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3755 C<SvSetMagicSV_nosteal>.
3757 =for apidoc sv_setsv_flags
3759 Copies the contents of the source SV C<ssv> into the destination SV
3760 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3761 function if the source SV needs to be reused. Does not handle 'set' magic.
3762 Loosely speaking, it performs a copy-by-value, obliterating any previous
3763 content of the destination.
3764 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3765 C<ssv> if appropriate, else not. If the C<flags>
3766 parameter has the C<SV_NOSTEAL> bit set then the
3767 buffers of temps will not be stolen. <sv_setsv>
3768 and C<sv_setsv_nomg> are implemented in terms of this function.
3770 You probably want to use one of the assortment of wrappers, such as
3771 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3772 C<SvSetMagicSV_nosteal>.
3774 This is the primary function for copying scalars, and most other
3775 copy-ish functions and macros use this underneath.
3781 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3783 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3784 HV *old_stash = NULL;
3786 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3788 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3789 const char * const name = GvNAME(sstr);
3790 const STRLEN len = GvNAMELEN(sstr);
3792 if (dtype >= SVt_PV) {
3798 SvUPGRADE(dstr, SVt_PVGV);
3799 (void)SvOK_off(dstr);
3800 isGV_with_GP_on(dstr);
3802 GvSTASH(dstr) = GvSTASH(sstr);
3804 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3805 gv_name_set(MUTABLE_GV(dstr), name, len,
3806 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3807 SvFAKE_on(dstr); /* can coerce to non-glob */
3810 if(GvGP(MUTABLE_GV(sstr))) {
3811 /* If source has method cache entry, clear it */
3813 SvREFCNT_dec(GvCV(sstr));
3814 GvCV_set(sstr, NULL);
3817 /* If source has a real method, then a method is
3820 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3826 /* If dest already had a real method, that's a change as well */
3828 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3829 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3834 /* We don't need to check the name of the destination if it was not a
3835 glob to begin with. */
3836 if(dtype == SVt_PVGV) {
3837 const char * const name = GvNAME((const GV *)dstr);
3840 /* The stash may have been detached from the symbol table, so
3842 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3846 const STRLEN len = GvNAMELEN(dstr);
3847 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3848 || (len == 1 && name[0] == ':')) {
3851 /* Set aside the old stash, so we can reset isa caches on
3853 if((old_stash = GvHV(dstr)))
3854 /* Make sure we do not lose it early. */
3855 SvREFCNT_inc_simple_void_NN(
3856 sv_2mortal((SV *)old_stash)
3861 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3864 gp_free(MUTABLE_GV(dstr));
3865 GvINTRO_off(dstr); /* one-shot flag */
3866 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3867 if (SvTAINTED(sstr))
3869 if (GvIMPORTED(dstr) != GVf_IMPORTED
3870 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3872 GvIMPORTED_on(dstr);
3875 if(mro_changes == 2) {
3876 if (GvAV((const GV *)sstr)) {
3878 SV * const sref = (SV *)GvAV((const GV *)dstr);
3879 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3880 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3881 AV * const ary = newAV();
3882 av_push(ary, mg->mg_obj); /* takes the refcount */
3883 mg->mg_obj = (SV *)ary;
3885 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3887 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3889 mro_isa_changed_in(GvSTASH(dstr));
3891 else if(mro_changes == 3) {
3892 HV * const stash = GvHV(dstr);
3893 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3899 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3900 if (GvIO(dstr) && dtype == SVt_PVGV) {
3901 DEBUG_o(Perl_deb(aTHX_
3902 "glob_assign_glob clearing PL_stashcache\n"));
3903 /* It's a cache. It will rebuild itself quite happily.
3904 It's a lot of effort to work out exactly which key (or keys)
3905 might be invalidated by the creation of the this file handle.
3907 hv_clear(PL_stashcache);
3913 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3915 SV * const sref = SvRV(sstr);
3917 const int intro = GvINTRO(dstr);
3920 const U32 stype = SvTYPE(sref);
3922 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3925 GvINTRO_off(dstr); /* one-shot flag */
3926 GvLINE(dstr) = CopLINE(PL_curcop);
3927 GvEGV(dstr) = MUTABLE_GV(dstr);
3932 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3933 import_flag = GVf_IMPORTED_CV;
3936 location = (SV **) &GvHV(dstr);
3937 import_flag = GVf_IMPORTED_HV;
3940 location = (SV **) &GvAV(dstr);
3941 import_flag = GVf_IMPORTED_AV;
3944 location = (SV **) &GvIOp(dstr);
3947 location = (SV **) &GvFORM(dstr);
3950 location = &GvSV(dstr);
3951 import_flag = GVf_IMPORTED_SV;
3954 if (stype == SVt_PVCV) {
3955 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3956 if (GvCVGEN(dstr)) {
3957 SvREFCNT_dec(GvCV(dstr));
3958 GvCV_set(dstr, NULL);
3959 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3962 /* SAVEt_GVSLOT takes more room on the savestack and has more
3963 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3964 leave_scope needs access to the GV so it can reset method
3965 caches. We must use SAVEt_GVSLOT whenever the type is
3966 SVt_PVCV, even if the stash is anonymous, as the stash may
3967 gain a name somehow before leave_scope. */
3968 if (stype == SVt_PVCV) {
3969 /* There is no save_pushptrptrptr. Creating it for this
3970 one call site would be overkill. So inline the ss add
3974 SS_ADD_PTR(location);
3975 SS_ADD_PTR(SvREFCNT_inc(*location));
3976 SS_ADD_UV(SAVEt_GVSLOT);
3979 else SAVEGENERICSV(*location);
3982 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3983 CV* const cv = MUTABLE_CV(*location);
3985 if (!GvCVGEN((const GV *)dstr) &&
3986 (CvROOT(cv) || CvXSUB(cv)) &&
3987 /* redundant check that avoids creating the extra SV
3988 most of the time: */
3989 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3991 SV * const new_const_sv =
3992 CvCONST((const CV *)sref)
3993 ? cv_const_sv((const CV *)sref)
3995 report_redefined_cv(
3996 sv_2mortal(Perl_newSVpvf(aTHX_
3999 HvNAME_HEK(GvSTASH((const GV *)dstr))
4001 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4004 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4008 cv_ckproto_len_flags(cv, (const GV *)dstr,
4009 SvPOK(sref) ? CvPROTO(sref) : NULL,
4010 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4011 SvPOK(sref) ? SvUTF8(sref) : 0);
4013 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4014 GvASSUMECV_on(dstr);
4015 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4016 if (intro && GvREFCNT(dstr) > 1) {
4017 /* temporary remove extra savestack's ref */
4019 gv_method_changed(dstr);
4022 else gv_method_changed(dstr);
4025 *location = SvREFCNT_inc_simple_NN(sref);
4026 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4027 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4028 GvFLAGS(dstr) |= import_flag;
4030 if (import_flag == GVf_IMPORTED_SV) {
4033 SS_ADD_PTR(gp_ref(GvGP(dstr)));
4034 SS_ADD_UV(SAVEt_GP_ALIASED_SV
4035 | cBOOL(GvALIASED_SV(dstr)) << 8);
4038 /* Turn off the flag if sref is not referenced elsewhere,
4039 even by weak refs. (SvRMAGICAL is a pessimistic check for
4041 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4042 GvALIASED_SV_off(dstr);
4044 GvALIASED_SV_on(dstr);
4046 if (stype == SVt_PVHV) {
4047 const char * const name = GvNAME((GV*)dstr);
4048 const STRLEN len = GvNAMELEN(dstr);
4051 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4052 || (len == 1 && name[0] == ':')
4054 && (!dref || HvENAME_get(dref))
4057 (HV *)sref, (HV *)dref,
4063 stype == SVt_PVAV && sref != dref
4064 && strEQ(GvNAME((GV*)dstr), "ISA")
4065 /* The stash may have been detached from the symbol table, so
4066 check its name before doing anything. */
4067 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4070 MAGIC * const omg = dref && SvSMAGICAL(dref)
4071 ? mg_find(dref, PERL_MAGIC_isa)
4073 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4074 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4075 AV * const ary = newAV();
4076 av_push(ary, mg->mg_obj); /* takes the refcount */
4077 mg->mg_obj = (SV *)ary;
4080 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4081 SV **svp = AvARRAY((AV *)omg->mg_obj);
4082 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4086 SvREFCNT_inc_simple_NN(*svp++)
4092 SvREFCNT_inc_simple_NN(omg->mg_obj)
4096 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4101 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4103 mg = mg_find(sref, PERL_MAGIC_isa);
4105 /* Since the *ISA assignment could have affected more than
4106 one stash, don't call mro_isa_changed_in directly, but let
4107 magic_clearisa do it for us, as it already has the logic for
4108 dealing with globs vs arrays of globs. */
4110 Perl_magic_clearisa(aTHX_ NULL, mg);
4112 else if (stype == SVt_PVIO) {
4113 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4114 /* It's a cache. It will rebuild itself quite happily.
4115 It's a lot of effort to work out exactly which key (or keys)
4116 might be invalidated by the creation of the this file handle.
4118 hv_clear(PL_stashcache);
4122 if (!intro) SvREFCNT_dec(dref);
4123 if (SvTAINTED(sstr))
4131 #ifdef PERL_DEBUG_READONLY_COW
4132 # include <sys/mman.h>
4134 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4135 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4139 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4141 struct perl_memory_debug_header * const header =
4142 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4143 const MEM_SIZE len = header->size;
4144 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4145 # ifdef PERL_TRACK_MEMPOOL
4146 if (!header->readonly) header->readonly = 1;
4148 if (mprotect(header, len, PROT_READ))
4149 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4150 header, len, errno);
4154 S_sv_buf_to_rw(pTHX_ SV *sv)
4156 struct perl_memory_debug_header * const header =
4157 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4158 const MEM_SIZE len = header->size;
4159 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4160 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4161 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4162 header, len, errno);
4163 # ifdef PERL_TRACK_MEMPOOL
4164 header->readonly = 0;
4169 # define sv_buf_to_ro(sv) NOOP
4170 # define sv_buf_to_rw(sv) NOOP
4174 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4180 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4185 if (SvIS_FREED(dstr)) {
4186 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4187 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4189 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4191 sstr = &PL_sv_undef;
4192 if (SvIS_FREED(sstr)) {
4193 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4194 (void*)sstr, (void*)dstr);
4196 stype = SvTYPE(sstr);
4197 dtype = SvTYPE(dstr);
4199 /* There's a lot of redundancy below but we're going for speed here */
4204 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4205 (void)SvOK_off(dstr);
4213 sv_upgrade(dstr, SVt_IV);
4217 sv_upgrade(dstr, SVt_PVIV);
4221 goto end_of_first_switch;
4223 (void)SvIOK_only(dstr);
4224 SvIV_set(dstr, SvIVX(sstr));
4227 /* SvTAINTED can only be true if the SV has taint magic, which in
4228 turn means that the SV type is PVMG (or greater). This is the
4229 case statement for SVt_IV, so this cannot be true (whatever gcov
4231 assert(!SvTAINTED(sstr));
4236 if (dtype < SVt_PV && dtype != SVt_IV)
4237 sv_upgrade(dstr, SVt_IV);
4245 sv_upgrade(dstr, SVt_NV);
4249 sv_upgrade(dstr, SVt_PVNV);
4253 goto end_of_first_switch;
4255 SvNV_set(dstr, SvNVX(sstr));
4256 (void)SvNOK_only(dstr);
4257 /* SvTAINTED can only be true if the SV has taint magic, which in
4258 turn means that the SV type is PVMG (or greater). This is the
4259 case statement for SVt_NV, so this cannot be true (whatever gcov
4261 assert(!SvTAINTED(sstr));
4268 sv_upgrade(dstr, SVt_PV);
4271 if (dtype < SVt_PVIV)
4272 sv_upgrade(dstr, SVt_PVIV);
4275 if (dtype < SVt_PVNV)
4276 sv_upgrade(dstr, SVt_PVNV);
4280 const char * const type = sv_reftype(sstr,0);
4282 /* diag_listed_as: Bizarre copy of %s */
4283 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4285 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4287 NOT_REACHED; /* NOTREACHED */
4291 if (dtype < SVt_REGEXP)
4293 if (dtype >= SVt_PV) {
4299 sv_upgrade(dstr, SVt_REGEXP);
4307 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4309 if (SvTYPE(sstr) != stype)
4310 stype = SvTYPE(sstr);
4312 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4313 glob_assign_glob(dstr, sstr, dtype);
4316 if (stype == SVt_PVLV)
4318 if (isREGEXP(sstr)) goto upgregexp;
4319 SvUPGRADE(dstr, SVt_PVNV);
4322 SvUPGRADE(dstr, (svtype)stype);
4324 end_of_first_switch:
4326 /* dstr may have been upgraded. */
4327 dtype = SvTYPE(dstr);
4328 sflags = SvFLAGS(sstr);
4330 if (dtype == SVt_PVCV) {
4331 /* Assigning to a subroutine sets the prototype. */
4334 const char *const ptr = SvPV_const(sstr, len);
4336 SvGROW(dstr, len + 1);
4337 Copy(ptr, SvPVX(dstr), len + 1, char);
4338 SvCUR_set(dstr, len);
4340 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4341 CvAUTOLOAD_off(dstr);
4346 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4347 const char * const type = sv_reftype(dstr,0);
4349 /* diag_listed_as: Cannot copy to %s */
4350 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4352 Perl_croak(aTHX_ "Cannot copy to %s", type);
4353 } else if (sflags & SVf_ROK) {
4354 if (isGV_with_GP(dstr)
4355 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4358 if (GvIMPORTED(dstr) != GVf_IMPORTED
4359 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4361 GvIMPORTED_on(dstr);
4366 glob_assign_glob(dstr, sstr, dtype);
4370 if (dtype >= SVt_PV) {
4371 if (isGV_with_GP(dstr)) {
4372 glob_assign_ref(dstr, sstr);
4375 if (SvPVX_const(dstr)) {
4381 (void)SvOK_off(dstr);
4382 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4383 SvFLAGS(dstr) |= sflags & SVf_ROK;
4384 assert(!(sflags & SVp_NOK));
4385 assert(!(sflags & SVp_IOK));
4386 assert(!(sflags & SVf_NOK));
4387 assert(!(sflags & SVf_IOK));
4389 else if (isGV_with_GP(dstr)) {
4390 if (!(sflags & SVf_OK)) {
4391 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4392 "Undefined value assigned to typeglob");
4395 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4396 if (dstr != (const SV *)gv) {
4397 const char * const name = GvNAME((const GV *)dstr);
4398 const STRLEN len = GvNAMELEN(dstr);
4399 HV *old_stash = NULL;
4400 bool reset_isa = FALSE;
4401 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4402 || (len == 1 && name[0] == ':')) {
4403 /* Set aside the old stash, so we can reset isa caches
4404 on its subclasses. */
4405 if((old_stash = GvHV(dstr))) {
4406 /* Make sure we do not lose it early. */
4407 SvREFCNT_inc_simple_void_NN(
4408 sv_2mortal((SV *)old_stash)
4415 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4416 gp_free(MUTABLE_GV(dstr));
4418 GvGP_set(dstr, gp_ref(GvGP(gv)));
4421 HV * const stash = GvHV(dstr);
4423 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4433 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4434 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4435 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4437 else if (sflags & SVp_POK) {
4438 const STRLEN cur = SvCUR(sstr);
4439 const STRLEN len = SvLEN(sstr);
4442 * We have three basic ways to copy the string:
4448 * Which we choose is based on various factors. The following
4449 * things are listed in order of speed, fastest to slowest:
4451 * - Copying a short string
4452 * - Copy-on-write bookkeeping
4454 * - Copying a long string
4456 * We swipe the string (steal the string buffer) if the SV on the
4457 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4458 * big win on long strings. It should be a win on short strings if
4459 * SvPVX_const(dstr) has to be allocated. If not, it should not
4460 * slow things down, as SvPVX_const(sstr) would have been freed
4463 * We also steal the buffer from a PADTMP (operator target) if it
4464 * is ‘long enough’. For short strings, a swipe does not help
4465 * here, as it causes more malloc calls the next time the target
4466 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4467 * be allocated it is still not worth swiping PADTMPs for short
4468 * strings, as the savings here are small.
4470 * If the rhs is already flagged as a copy-on-write string and COW
4471 * is possible here, we use copy-on-write and make both SVs share
4472 * the string buffer.
4474 * If the rhs is not flagged as copy-on-write, then we see whether
4475 * it is worth upgrading it to such. If the lhs already has a buf-
4476 * fer big enough and the string is short, we skip it and fall back
4477 * to method 3, since memcpy is faster for short strings than the
4478 * later bookkeeping overhead that copy-on-write entails.
4480 * If there is no buffer on the left, or the buffer is too small,
4481 * then we use copy-on-write.
4484 /* Whichever path we take through the next code, we want this true,
4485 and doing it now facilitates the COW check. */
4486 (void)SvPOK_only(dstr);
4490 /* slated for free anyway (and not COW)? */
4491 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4492 /* or a swipable TARG */
4493 || ((sflags & (SVs_PADTMP|SVf_READONLY|SVf_IsCOW))
4495 /* whose buffer is worth stealing */
4496 && CHECK_COWBUF_THRESHOLD(cur,len)
4499 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4500 (!(flags & SV_NOSTEAL)) &&
4501 /* and we're allowed to steal temps */
4502 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4503 len) /* and really is a string */
4504 { /* Passes the swipe test. */
4505 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4507 SvPV_set(dstr, SvPVX_mutable(sstr));
4508 SvLEN_set(dstr, SvLEN(sstr));
4509 SvCUR_set(dstr, SvCUR(sstr));
4512 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4513 SvPV_set(sstr, NULL);
4518 else if (flags & SV_COW_SHARED_HASH_KEYS
4520 #ifdef PERL_OLD_COPY_ON_WRITE
4521 ( sflags & SVf_IsCOW
4522 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4523 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4524 && SvTYPE(sstr) >= SVt_PVIV && len
4527 #elif defined(PERL_NEW_COPY_ON_WRITE)
4530 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4531 /* If this is a regular (non-hek) COW, only so
4532 many COW "copies" are possible. */
4533 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4534 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4535 && !(SvFLAGS(dstr) & SVf_BREAK)
4536 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4537 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4541 && !(SvFLAGS(dstr) & SVf_BREAK)
4544 /* Either it's a shared hash key, or it's suitable for
4547 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4552 if (!(sflags & SVf_IsCOW)) {
4554 # ifdef PERL_OLD_COPY_ON_WRITE
4555 /* Make the source SV into a loop of 1.
4556 (about to become 2) */
4557 SV_COW_NEXT_SV_SET(sstr, sstr);
4559 CowREFCNT(sstr) = 0;
4563 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4569 # ifdef PERL_OLD_COPY_ON_WRITE
4570 assert (SvTYPE(dstr) >= SVt_PVIV);
4571 /* SvIsCOW_normal */
4572 /* splice us in between source and next-after-source. */
4573 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4574 SV_COW_NEXT_SV_SET(sstr, dstr);
4576 if (sflags & SVf_IsCOW) {
4581 SvPV_set(dstr, SvPVX_mutable(sstr));
4586 /* SvIsCOW_shared_hash */
4587 DEBUG_C(PerlIO_printf(Perl_debug_log,
4588 "Copy on write: Sharing hash\n"));
4590 assert (SvTYPE(dstr) >= SVt_PV);
4592 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4594 SvLEN_set(dstr, len);
4595 SvCUR_set(dstr, cur);
4598 /* Failed the swipe test, and we cannot do copy-on-write either.
4599 Have to copy the string. */
4600 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4601 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4602 SvCUR_set(dstr, cur);
4603 *SvEND(dstr) = '\0';
4605 if (sflags & SVp_NOK) {
4606 SvNV_set(dstr, SvNVX(sstr));
4608 if (sflags & SVp_IOK) {
4609 SvIV_set(dstr, SvIVX(sstr));
4610 /* Must do this otherwise some other overloaded use of 0x80000000
4611 gets confused. I guess SVpbm_VALID */
4612 if (sflags & SVf_IVisUV)
4615 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4617 const MAGIC * const smg = SvVSTRING_mg(sstr);
4619 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4620 smg->mg_ptr, smg->mg_len);
4621 SvRMAGICAL_on(dstr);
4625 else if (sflags & (SVp_IOK|SVp_NOK)) {
4626 (void)SvOK_off(dstr);
4627 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4628 if (sflags & SVp_IOK) {
4629 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4630 SvIV_set(dstr, SvIVX(sstr));
4632 if (sflags & SVp_NOK) {
4633 SvNV_set(dstr, SvNVX(sstr));
4637 if (isGV_with_GP(sstr)) {
4638 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4641 (void)SvOK_off(dstr);
4643 if (SvTAINTED(sstr))
4648 =for apidoc sv_setsv_mg
4650 Like C<sv_setsv>, but also handles 'set' magic.
4656 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4658 PERL_ARGS_ASSERT_SV_SETSV_MG;
4660 sv_setsv(dstr,sstr);
4665 # ifdef PERL_OLD_COPY_ON_WRITE
4666 # define SVt_COW SVt_PVIV
4668 # define SVt_COW SVt_PV
4671 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4673 STRLEN cur = SvCUR(sstr);
4674 STRLEN len = SvLEN(sstr);
4676 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4677 const bool already = cBOOL(SvIsCOW(sstr));
4680 PERL_ARGS_ASSERT_SV_SETSV_COW;
4683 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4684 (void*)sstr, (void*)dstr);
4691 if (SvTHINKFIRST(dstr))
4692 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4693 else if (SvPVX_const(dstr))
4694 Safefree(SvPVX_mutable(dstr));
4698 SvUPGRADE(dstr, SVt_COW);
4700 assert (SvPOK(sstr));
4701 assert (SvPOKp(sstr));
4702 # ifdef PERL_OLD_COPY_ON_WRITE
4703 assert (!SvIOK(sstr));
4704 assert (!SvIOKp(sstr));
4705 assert (!SvNOK(sstr));
4706 assert (!SvNOKp(sstr));
4709 if (SvIsCOW(sstr)) {
4711 if (SvLEN(sstr) == 0) {
4712 /* source is a COW shared hash key. */
4713 DEBUG_C(PerlIO_printf(Perl_debug_log,
4714 "Fast copy on write: Sharing hash\n"));
4715 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4718 # ifdef PERL_OLD_COPY_ON_WRITE
4719 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4721 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4722 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4725 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4726 SvUPGRADE(sstr, SVt_COW);
4728 DEBUG_C(PerlIO_printf(Perl_debug_log,
4729 "Fast copy on write: Converting sstr to COW\n"));
4730 # ifdef PERL_OLD_COPY_ON_WRITE
4731 SV_COW_NEXT_SV_SET(dstr, sstr);
4733 CowREFCNT(sstr) = 0;
4736 # ifdef PERL_OLD_COPY_ON_WRITE
4737 SV_COW_NEXT_SV_SET(sstr, dstr);
4739 # ifdef PERL_DEBUG_READONLY_COW
4740 if (already) sv_buf_to_rw(sstr);
4744 new_pv = SvPVX_mutable(sstr);
4748 SvPV_set(dstr, new_pv);
4749 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4752 SvLEN_set(dstr, len);
4753 SvCUR_set(dstr, cur);
4762 =for apidoc sv_setpvn
4764 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4765 The C<len> parameter indicates the number of
4766 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4767 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4773 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4777 PERL_ARGS_ASSERT_SV_SETPVN;
4779 SV_CHECK_THINKFIRST_COW_DROP(sv);
4785 /* len is STRLEN which is unsigned, need to copy to signed */
4788 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4791 SvUPGRADE(sv, SVt_PV);
4793 dptr = SvGROW(sv, len + 1);
4794 Move(ptr,dptr,len,char);
4797 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4799 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4803 =for apidoc sv_setpvn_mg
4805 Like C<sv_setpvn>, but also handles 'set' magic.
4811 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4813 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4815 sv_setpvn(sv,ptr,len);
4820 =for apidoc sv_setpv
4822 Copies a string into an SV. The string must be terminated with a C<NUL>
4824 Does not handle 'set' magic. See C<sv_setpv_mg>.
4830 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4834 PERL_ARGS_ASSERT_SV_SETPV;
4836 SV_CHECK_THINKFIRST_COW_DROP(sv);
4842 SvUPGRADE(sv, SVt_PV);
4844 SvGROW(sv, len + 1);
4845 Move(ptr,SvPVX(sv),len+1,char);
4847 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4849 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4853 =for apidoc sv_setpv_mg
4855 Like C<sv_setpv>, but also handles 'set' magic.
4861 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4863 PERL_ARGS_ASSERT_SV_SETPV_MG;
4870 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4872 PERL_ARGS_ASSERT_SV_SETHEK;
4878 if (HEK_LEN(hek) == HEf_SVKEY) {
4879 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4882 const int flags = HEK_FLAGS(hek);
4883 if (flags & HVhek_WASUTF8) {
4884 STRLEN utf8_len = HEK_LEN(hek);
4885 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4886 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4889 } else if (flags & HVhek_UNSHARED) {
4890 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4893 else SvUTF8_off(sv);
4897 SV_CHECK_THINKFIRST_COW_DROP(sv);
4898 SvUPGRADE(sv, SVt_PV);
4900 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4901 SvCUR_set(sv, HEK_LEN(hek));
4907 else SvUTF8_off(sv);
4915 =for apidoc sv_usepvn_flags
4917 Tells an SV to use C<ptr> to find its string value. Normally the
4918 string is stored inside the SV, but sv_usepvn allows the SV to use an
4919 outside string. The C<ptr> should point to memory that was allocated
4920 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4921 the start of a Newx-ed block of memory, and not a pointer to the
4922 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4923 and not be from a non-Newx memory allocator like C<malloc>. The
4924 string length, C<len>, must be supplied. By default this function
4925 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4926 so that pointer should not be freed or used by the programmer after
4927 giving it to sv_usepvn, and neither should any pointers from "behind"
4928 that pointer (e.g. ptr + 1) be used.
4930 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4931 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
4932 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4933 C<len>, and already meets the requirements for storing in C<SvPVX>).
4939 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4943 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4945 SV_CHECK_THINKFIRST_COW_DROP(sv);
4946 SvUPGRADE(sv, SVt_PV);
4949 if (flags & SV_SMAGIC)
4953 if (SvPVX_const(sv))
4957 if (flags & SV_HAS_TRAILING_NUL)
4958 assert(ptr[len] == '\0');
4961 allocate = (flags & SV_HAS_TRAILING_NUL)
4963 #ifdef Perl_safesysmalloc_size
4966 PERL_STRLEN_ROUNDUP(len + 1);
4968 if (flags & SV_HAS_TRAILING_NUL) {
4969 /* It's long enough - do nothing.
4970 Specifically Perl_newCONSTSUB is relying on this. */
4973 /* Force a move to shake out bugs in callers. */
4974 char *new_ptr = (char*)safemalloc(allocate);
4975 Copy(ptr, new_ptr, len, char);
4976 PoisonFree(ptr,len,char);
4980 ptr = (char*) saferealloc (ptr, allocate);
4983 #ifdef Perl_safesysmalloc_size
4984 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4986 SvLEN_set(sv, allocate);
4990 if (!(flags & SV_HAS_TRAILING_NUL)) {
4993 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4995 if (flags & SV_SMAGIC)
4999 #ifdef PERL_OLD_COPY_ON_WRITE
5000 /* Need to do this *after* making the SV normal, as we need the buffer
5001 pointer to remain valid until after we've copied it. If we let go too early,
5002 another thread could invalidate it by unsharing last of the same hash key
5003 (which it can do by means other than releasing copy-on-write Svs)
5004 or by changing the other copy-on-write SVs in the loop. */
5006 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5008 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5010 { /* this SV was SvIsCOW_normal(sv) */
5011 /* we need to find the SV pointing to us. */
5012 SV *current = SV_COW_NEXT_SV(after);
5014 if (current == sv) {
5015 /* The SV we point to points back to us (there were only two of us
5017 Hence other SV is no longer copy on write either. */
5019 sv_buf_to_rw(after);
5021 /* We need to follow the pointers around the loop. */
5023 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5026 /* don't loop forever if the structure is bust, and we have
5027 a pointer into a closed loop. */
5028 assert (current != after);
5029 assert (SvPVX_const(current) == pvx);
5031 /* Make the SV before us point to the SV after us. */
5032 SV_COW_NEXT_SV_SET(current, after);
5038 =for apidoc sv_force_normal_flags
5040 Undo various types of fakery on an SV, where fakery means
5041 "more than" a string: if the PV is a shared string, make
5042 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5043 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5044 we do the copy, and is also used locally; if this is a
5045 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5046 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5047 SvPOK_off rather than making a copy. (Used where this
5048 scalar is about to be set to some other value.) In addition,
5049 the C<flags> parameter gets passed to C<sv_unref_flags()>
5050 when unreffing. C<sv_force_normal> calls this function
5051 with flags set to 0.
5053 This function is expected to be used to signal to perl that this SV is
5054 about to be written to, and any extra book-keeping needs to be taken care
5055 of. Hence, it croaks on read-only values.
5061 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5063 assert(SvIsCOW(sv));
5066 const char * const pvx = SvPVX_const(sv);
5067 const STRLEN len = SvLEN(sv);
5068 const STRLEN cur = SvCUR(sv);
5069 # ifdef PERL_OLD_COPY_ON_WRITE
5070 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5071 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5072 we'll fail an assertion. */
5073 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5077 PerlIO_printf(Perl_debug_log,
5078 "Copy on write: Force normal %ld\n",
5083 # ifdef PERL_NEW_COPY_ON_WRITE
5084 if (len && CowREFCNT(sv) == 0)
5085 /* We own the buffer ourselves. */
5091 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5092 # ifdef PERL_NEW_COPY_ON_WRITE
5093 /* Must do this first, since the macro uses SvPVX. */
5103 if (flags & SV_COW_DROP_PV) {
5104 /* OK, so we don't need to copy our buffer. */
5107 SvGROW(sv, cur + 1);
5108 Move(pvx,SvPVX(sv),cur,char);
5113 # ifdef PERL_OLD_COPY_ON_WRITE
5114 sv_release_COW(sv, pvx, next);
5117 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5124 const char * const pvx = SvPVX_const(sv);
5125 const STRLEN len = SvCUR(sv);
5129 if (flags & SV_COW_DROP_PV) {
5130 /* OK, so we don't need to copy our buffer. */
5133 SvGROW(sv, len + 1);
5134 Move(pvx,SvPVX(sv),len,char);
5137 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5143 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5145 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5148 Perl_croak_no_modify();
5149 else if (SvIsCOW(sv))
5150 S_sv_uncow(aTHX_ sv, flags);
5152 sv_unref_flags(sv, flags);
5153 else if (SvFAKE(sv) && isGV_with_GP(sv))
5154 sv_unglob(sv, flags);
5155 else if (SvFAKE(sv) && isREGEXP(sv)) {
5156 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5157 to sv_unglob. We only need it here, so inline it. */
5158 const bool islv = SvTYPE(sv) == SVt_PVLV;
5159 const svtype new_type =
5160 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5161 SV *const temp = newSV_type(new_type);
5162 regexp *const temp_p = ReANY((REGEXP *)sv);
5164 if (new_type == SVt_PVMG) {
5165 SvMAGIC_set(temp, SvMAGIC(sv));
5166 SvMAGIC_set(sv, NULL);
5167 SvSTASH_set(temp, SvSTASH(sv));
5168 SvSTASH_set(sv, NULL);
5170 if (!islv) SvCUR_set(temp, SvCUR(sv));
5171 /* Remember that SvPVX is in the head, not the body. But
5172 RX_WRAPPED is in the body. */
5173 assert(ReANY((REGEXP *)sv)->mother_re);
5174 /* Their buffer is already owned by someone else. */
5175 if (flags & SV_COW_DROP_PV) {
5176 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5177 zeroed body. For SVt_PVLV, it should have been set to 0
5178 before turning into a regexp. */
5179 assert(!SvLEN(islv ? sv : temp));
5180 sv->sv_u.svu_pv = 0;
5183 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5184 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5188 /* Now swap the rest of the bodies. */
5192 SvFLAGS(sv) &= ~SVTYPEMASK;
5193 SvFLAGS(sv) |= new_type;
5194 SvANY(sv) = SvANY(temp);
5197 SvFLAGS(temp) &= ~(SVTYPEMASK);
5198 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5199 SvANY(temp) = temp_p;
5200 temp->sv_u.svu_rx = (regexp *)temp_p;
5202 SvREFCNT_dec_NN(temp);
5204 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5210 Efficient removal of characters from the beginning of the string buffer.
5211 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5212 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5213 character of the adjusted string. Uses the "OOK hack". On return, only
5214 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5216 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5217 refer to the same chunk of data.
5219 The unfortunate similarity of this function's name to that of Perl's C<chop>
5220 operator is strictly coincidental. This function works from the left;
5221 C<chop> works from the right.
5227 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5238 PERL_ARGS_ASSERT_SV_CHOP;
5240 if (!ptr || !SvPOKp(sv))
5242 delta = ptr - SvPVX_const(sv);
5244 /* Nothing to do. */
5247 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5248 if (delta > max_delta)
5249 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5250 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5251 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5252 SV_CHECK_THINKFIRST(sv);
5253 SvPOK_only_UTF8(sv);
5256 if (!SvLEN(sv)) { /* make copy of shared string */
5257 const char *pvx = SvPVX_const(sv);
5258 const STRLEN len = SvCUR(sv);
5259 SvGROW(sv, len + 1);
5260 Move(pvx,SvPVX(sv),len,char);
5266 SvOOK_offset(sv, old_delta);
5268 SvLEN_set(sv, SvLEN(sv) - delta);
5269 SvCUR_set(sv, SvCUR(sv) - delta);
5270 SvPV_set(sv, SvPVX(sv) + delta);
5272 p = (U8 *)SvPVX_const(sv);
5275 /* how many bytes were evacuated? we will fill them with sentinel
5276 bytes, except for the part holding the new offset of course. */
5279 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5281 assert(evacn <= delta + old_delta);
5285 /* This sets 'delta' to the accumulated value of all deltas so far */
5289 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5290 * the string; otherwise store a 0 byte there and store 'delta' just prior
5291 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5292 * portion of the chopped part of the string */
5293 if (delta < 0x100) {
5297 p -= sizeof(STRLEN);
5298 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5302 /* Fill the preceding buffer with sentinals to verify that no-one is
5312 =for apidoc sv_catpvn
5314 Concatenates the string onto the end of the string which is in the SV. The
5315 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5316 status set, then the bytes appended should be valid UTF-8.
5317 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5319 =for apidoc sv_catpvn_flags
5321 Concatenates the string onto the end of the string which is in the SV. The
5322 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5323 status set, then the bytes appended should be valid UTF-8.
5324 If C<flags> has the C<SV_SMAGIC> bit set, will
5325 C<mg_set> on C<dsv> afterwards if appropriate.
5326 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5327 in terms of this function.
5333 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5336 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5338 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5339 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5341 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5342 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5343 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5346 else SvGROW(dsv, dlen + slen + 1);
5348 sstr = SvPVX_const(dsv);
5349 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5350 SvCUR_set(dsv, SvCUR(dsv) + slen);
5353 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5354 const char * const send = sstr + slen;
5357 /* Something this code does not account for, which I think is
5358 impossible; it would require the same pv to be treated as
5359 bytes *and* utf8, which would indicate a bug elsewhere. */
5360 assert(sstr != dstr);
5362 SvGROW(dsv, dlen + slen * 2 + 1);
5363 d = (U8 *)SvPVX(dsv) + dlen;
5365 while (sstr < send) {
5366 append_utf8_from_native_byte(*sstr, &d);
5369 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5372 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5374 if (flags & SV_SMAGIC)
5379 =for apidoc sv_catsv
5381 Concatenates the string from SV C<ssv> onto the end of the string in SV
5382 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5383 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5386 =for apidoc sv_catsv_flags
5388 Concatenates the string from SV C<ssv> onto the end of the string in SV
5389 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5390 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5391 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5392 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5393 and C<sv_catsv_mg> are implemented in terms of this function.
5398 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5400 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5404 const char *spv = SvPV_flags_const(ssv, slen, flags);
5406 if (flags & SV_GMAGIC)
5408 sv_catpvn_flags(dsv, spv, slen,
5409 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5410 if (flags & SV_SMAGIC)
5417 =for apidoc sv_catpv
5419 Concatenates the C<NUL>-terminated string onto the end of the string which is
5421 If the SV has the UTF-8 status set, then the bytes appended should be
5422 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5427 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5433 PERL_ARGS_ASSERT_SV_CATPV;
5437 junk = SvPV_force(sv, tlen);
5439 SvGROW(sv, tlen + len + 1);
5441 ptr = SvPVX_const(sv);
5442 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5443 SvCUR_set(sv, SvCUR(sv) + len);
5444 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5449 =for apidoc sv_catpv_flags
5451 Concatenates the C<NUL>-terminated string onto the end of the string which is
5453 If the SV has the UTF-8 status set, then the bytes appended should
5454 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5455 on the modified SV if appropriate.
5461 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5463 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5464 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5468 =for apidoc sv_catpv_mg
5470 Like C<sv_catpv>, but also handles 'set' magic.
5476 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5478 PERL_ARGS_ASSERT_SV_CATPV_MG;
5487 Creates a new SV. A non-zero C<len> parameter indicates the number of
5488 bytes of preallocated string space the SV should have. An extra byte for a
5489 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5490 space is allocated.) The reference count for the new SV is set to 1.
5492 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5493 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5494 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5495 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5496 modules supporting older perls.
5502 Perl_newSV(pTHX_ const STRLEN len)
5508 sv_upgrade(sv, SVt_PV);
5509 SvGROW(sv, len + 1);
5514 =for apidoc sv_magicext
5516 Adds magic to an SV, upgrading it if necessary. Applies the
5517 supplied vtable and returns a pointer to the magic added.
5519 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5520 In particular, you can add magic to SvREADONLY SVs, and add more than
5521 one instance of the same 'how'.
5523 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5524 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5525 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5526 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5528 (This is now used as a subroutine by C<sv_magic>.)
5533 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5534 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5538 PERL_ARGS_ASSERT_SV_MAGICEXT;
5540 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5542 SvUPGRADE(sv, SVt_PVMG);
5543 Newxz(mg, 1, MAGIC);
5544 mg->mg_moremagic = SvMAGIC(sv);
5545 SvMAGIC_set(sv, mg);
5547 /* Sometimes a magic contains a reference loop, where the sv and
5548 object refer to each other. To prevent a reference loop that
5549 would prevent such objects being freed, we look for such loops
5550 and if we find one we avoid incrementing the object refcount.
5552 Note we cannot do this to avoid self-tie loops as intervening RV must
5553 have its REFCNT incremented to keep it in existence.
5556 if (!obj || obj == sv ||
5557 how == PERL_MAGIC_arylen ||
5558 how == PERL_MAGIC_symtab ||
5559 (SvTYPE(obj) == SVt_PVGV &&
5560 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5561 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5562 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5567 mg->mg_obj = SvREFCNT_inc_simple(obj);
5568 mg->mg_flags |= MGf_REFCOUNTED;
5571 /* Normal self-ties simply pass a null object, and instead of
5572 using mg_obj directly, use the SvTIED_obj macro to produce a
5573 new RV as needed. For glob "self-ties", we are tieing the PVIO
5574 with an RV obj pointing to the glob containing the PVIO. In
5575 this case, to avoid a reference loop, we need to weaken the
5579 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5580 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5586 mg->mg_len = namlen;
5589 mg->mg_ptr = savepvn(name, namlen);
5590 else if (namlen == HEf_SVKEY) {
5591 /* Yes, this is casting away const. This is only for the case of
5592 HEf_SVKEY. I think we need to document this aberation of the
5593 constness of the API, rather than making name non-const, as
5594 that change propagating outwards a long way. */
5595 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5597 mg->mg_ptr = (char *) name;
5599 mg->mg_virtual = (MGVTBL *) vtable;
5606 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5608 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5609 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5610 /* This sv is only a delegate. //g magic must be attached to
5615 #ifdef PERL_OLD_COPY_ON_WRITE
5617 sv_force_normal_flags(sv, 0);
5619 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5620 &PL_vtbl_mglob, 0, 0);
5624 =for apidoc sv_magic
5626 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5627 necessary, then adds a new magic item of type C<how> to the head of the
5630 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5631 handling of the C<name> and C<namlen> arguments.
5633 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5634 to add more than one instance of the same 'how'.
5640 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5641 const char *const name, const I32 namlen)
5643 const MGVTBL *vtable;
5646 unsigned int vtable_index;
5648 PERL_ARGS_ASSERT_SV_MAGIC;
5650 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5651 || ((flags = PL_magic_data[how]),
5652 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5653 > magic_vtable_max))
5654 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5656 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5657 Useful for attaching extension internal data to perl vars.
5658 Note that multiple extensions may clash if magical scalars
5659 etc holding private data from one are passed to another. */
5661 vtable = (vtable_index == magic_vtable_max)
5662 ? NULL : PL_magic_vtables + vtable_index;
5664 #ifdef PERL_OLD_COPY_ON_WRITE
5666 sv_force_normal_flags(sv, 0);
5668 if (SvREADONLY(sv)) {
5670 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5673 Perl_croak_no_modify();
5676 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5677 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5678 /* sv_magic() refuses to add a magic of the same 'how' as an
5681 if (how == PERL_MAGIC_taint)
5687 /* Force pos to be stored as characters, not bytes. */
5688 if (SvMAGICAL(sv) && DO_UTF8(sv)
5689 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5691 && mg->mg_flags & MGf_BYTES) {
5692 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5694 mg->mg_flags &= ~MGf_BYTES;
5697 /* Rest of work is done else where */
5698 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5701 case PERL_MAGIC_taint:
5704 case PERL_MAGIC_ext:
5705 case PERL_MAGIC_dbfile:
5712 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5719 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5721 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5722 for (mg = *mgp; mg; mg = *mgp) {
5723 const MGVTBL* const virt = mg->mg_virtual;
5724 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5725 *mgp = mg->mg_moremagic;
5726 if (virt && virt->svt_free)
5727 virt->svt_free(aTHX_ sv, mg);
5728 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5730 Safefree(mg->mg_ptr);
5731 else if (mg->mg_len == HEf_SVKEY)
5732 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5733 else if (mg->mg_type == PERL_MAGIC_utf8)
5734 Safefree(mg->mg_ptr);
5736 if (mg->mg_flags & MGf_REFCOUNTED)
5737 SvREFCNT_dec(mg->mg_obj);
5741 mgp = &mg->mg_moremagic;
5744 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5745 mg_magical(sv); /* else fix the flags now */
5749 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5755 =for apidoc sv_unmagic
5757 Removes all magic of type C<type> from an SV.
5763 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5765 PERL_ARGS_ASSERT_SV_UNMAGIC;
5766 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5770 =for apidoc sv_unmagicext
5772 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5778 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5780 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5781 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5785 =for apidoc sv_rvweaken
5787 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5788 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5789 push a back-reference to this RV onto the array of backreferences
5790 associated with that magic. If the RV is magical, set magic will be
5791 called after the RV is cleared.
5797 Perl_sv_rvweaken(pTHX_ SV *const sv)
5801 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5803 if (!SvOK(sv)) /* let undefs pass */
5806 Perl_croak(aTHX_ "Can't weaken a nonreference");
5807 else if (SvWEAKREF(sv)) {
5808 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5811 else if (SvREADONLY(sv)) croak_no_modify();
5813 Perl_sv_add_backref(aTHX_ tsv, sv);
5815 SvREFCNT_dec_NN(tsv);
5819 /* Give tsv backref magic if it hasn't already got it, then push a
5820 * back-reference to sv onto the array associated with the backref magic.
5822 * As an optimisation, if there's only one backref and it's not an AV,
5823 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5824 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5828 /* A discussion about the backreferences array and its refcount:
5830 * The AV holding the backreferences is pointed to either as the mg_obj of
5831 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5832 * xhv_backreferences field. The array is created with a refcount
5833 * of 2. This means that if during global destruction the array gets
5834 * picked on before its parent to have its refcount decremented by the
5835 * random zapper, it won't actually be freed, meaning it's still there for
5836 * when its parent gets freed.
5838 * When the parent SV is freed, the extra ref is killed by
5839 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5840 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5842 * When a single backref SV is stored directly, it is not reference
5847 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5853 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5855 /* find slot to store array or singleton backref */
5857 if (SvTYPE(tsv) == SVt_PVHV) {
5858 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5861 mg = mg_find(tsv, PERL_MAGIC_backref);
5863 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5864 svp = &(mg->mg_obj);
5867 /* create or retrieve the array */
5869 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5870 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5874 mg->mg_flags |= MGf_REFCOUNTED;
5877 SvREFCNT_inc_simple_void_NN(av);
5878 /* av now has a refcnt of 2; see discussion above */
5879 av_extend(av, *svp ? 2 : 1);
5881 /* move single existing backref to the array */
5882 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5887 av = MUTABLE_AV(*svp);
5889 /* optimisation: store single backref directly in HvAUX or mg_obj */
5893 assert(SvTYPE(av) == SVt_PVAV);
5894 if (AvFILLp(av) >= AvMAX(av)) {
5895 av_extend(av, AvFILLp(av)+1);
5898 /* push new backref */
5899 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5902 /* delete a back-reference to ourselves from the backref magic associated
5903 * with the SV we point to.
5907 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5911 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5913 if (SvTYPE(tsv) == SVt_PVHV) {
5915 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5917 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5918 /* It's possible for the the last (strong) reference to tsv to have
5919 become freed *before* the last thing holding a weak reference.
5920 If both survive longer than the backreferences array, then when
5921 the referent's reference count drops to 0 and it is freed, it's
5922 not able to chase the backreferences, so they aren't NULLed.
5924 For example, a CV holds a weak reference to its stash. If both the
5925 CV and the stash survive longer than the backreferences array,
5926 and the CV gets picked for the SvBREAK() treatment first,
5927 *and* it turns out that the stash is only being kept alive because
5928 of an our variable in the pad of the CV, then midway during CV
5929 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5930 It ends up pointing to the freed HV. Hence it's chased in here, and
5931 if this block wasn't here, it would hit the !svp panic just below.
5933 I don't believe that "better" destruction ordering is going to help
5934 here - during global destruction there's always going to be the
5935 chance that something goes out of order. We've tried to make it
5936 foolproof before, and it only resulted in evolutionary pressure on
5937 fools. Which made us look foolish for our hubris. :-(
5943 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5944 svp = mg ? &(mg->mg_obj) : NULL;
5948 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5950 /* It's possible that sv is being freed recursively part way through the
5951 freeing of tsv. If this happens, the backreferences array of tsv has
5952 already been freed, and so svp will be NULL. If this is the case,
5953 we should not panic. Instead, nothing needs doing, so return. */
5954 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
5956 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
5957 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
5960 if (SvTYPE(*svp) == SVt_PVAV) {
5964 AV * const av = (AV*)*svp;
5966 assert(!SvIS_FREED(av));
5970 /* for an SV with N weak references to it, if all those
5971 * weak refs are deleted, then sv_del_backref will be called
5972 * N times and O(N^2) compares will be done within the backref
5973 * array. To ameliorate this potential slowness, we:
5974 * 1) make sure this code is as tight as possible;
5975 * 2) when looking for SV, look for it at both the head and tail of the
5976 * array first before searching the rest, since some create/destroy
5977 * patterns will cause the backrefs to be freed in order.
5984 SV **p = &svp[fill];
5985 SV *const topsv = *p;
5992 /* We weren't the last entry.
5993 An unordered list has this property that you
5994 can take the last element off the end to fill
5995 the hole, and it's still an unordered list :-)
6001 break; /* should only be one */
6008 AvFILLp(av) = fill-1;
6010 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6011 /* freed AV; skip */
6014 /* optimisation: only a single backref, stored directly */
6016 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6017 (void*)*svp, (void*)sv);
6024 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6030 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6035 /* after multiple passes through Perl_sv_clean_all() for a thingy
6036 * that has badly leaked, the backref array may have gotten freed,
6037 * since we only protect it against 1 round of cleanup */
6038 if (SvIS_FREED(av)) {
6039 if (PL_in_clean_all) /* All is fair */
6042 "panic: magic_killbackrefs (freed backref AV/SV)");
6046 is_array = (SvTYPE(av) == SVt_PVAV);
6048 assert(!SvIS_FREED(av));
6051 last = svp + AvFILLp(av);
6054 /* optimisation: only a single backref, stored directly */
6060 while (svp <= last) {
6062 SV *const referrer = *svp;
6063 if (SvWEAKREF(referrer)) {
6064 /* XXX Should we check that it hasn't changed? */
6065 assert(SvROK(referrer));
6066 SvRV_set(referrer, 0);
6068 SvWEAKREF_off(referrer);
6069 SvSETMAGIC(referrer);
6070 } else if (SvTYPE(referrer) == SVt_PVGV ||
6071 SvTYPE(referrer) == SVt_PVLV) {
6072 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6073 /* You lookin' at me? */
6074 assert(GvSTASH(referrer));
6075 assert(GvSTASH(referrer) == (const HV *)sv);
6076 GvSTASH(referrer) = 0;
6077 } else if (SvTYPE(referrer) == SVt_PVCV ||
6078 SvTYPE(referrer) == SVt_PVFM) {
6079 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6080 /* You lookin' at me? */
6081 assert(CvSTASH(referrer));
6082 assert(CvSTASH(referrer) == (const HV *)sv);
6083 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6086 assert(SvTYPE(sv) == SVt_PVGV);
6087 /* You lookin' at me? */
6088 assert(CvGV(referrer));
6089 assert(CvGV(referrer) == (const GV *)sv);
6090 anonymise_cv_maybe(MUTABLE_GV(sv),
6091 MUTABLE_CV(referrer));
6096 "panic: magic_killbackrefs (flags=%"UVxf")",
6097 (UV)SvFLAGS(referrer));
6108 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6114 =for apidoc sv_insert
6116 Inserts a string at the specified offset/length within the SV. Similar to
6117 the Perl substr() function. Handles get magic.
6119 =for apidoc sv_insert_flags
6121 Same as C<sv_insert>, but the extra C<flags> are passed to the
6122 C<SvPV_force_flags> that applies to C<bigstr>.
6128 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6134 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6137 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6140 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6141 SvPV_force_flags(bigstr, curlen, flags);
6142 (void)SvPOK_only_UTF8(bigstr);
6143 if (offset + len > curlen) {
6144 SvGROW(bigstr, offset+len+1);
6145 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6146 SvCUR_set(bigstr, offset+len);
6150 i = littlelen - len;
6151 if (i > 0) { /* string might grow */
6152 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6153 mid = big + offset + len;
6154 midend = bigend = big + SvCUR(bigstr);
6157 while (midend > mid) /* shove everything down */
6158 *--bigend = *--midend;
6159 Move(little,big+offset,littlelen,char);
6160 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6165 Move(little,SvPVX(bigstr)+offset,len,char);
6170 big = SvPVX(bigstr);
6173 bigend = big + SvCUR(bigstr);
6175 if (midend > bigend)
6176 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6179 if (mid - big > bigend - midend) { /* faster to shorten from end */
6181 Move(little, mid, littlelen,char);
6184 i = bigend - midend;
6186 Move(midend, mid, i,char);
6190 SvCUR_set(bigstr, mid - big);
6192 else if ((i = mid - big)) { /* faster from front */
6193 midend -= littlelen;
6195 Move(big, midend - i, i, char);
6196 sv_chop(bigstr,midend-i);
6198 Move(little, mid, littlelen,char);
6200 else if (littlelen) {
6201 midend -= littlelen;
6202 sv_chop(bigstr,midend);
6203 Move(little,midend,littlelen,char);
6206 sv_chop(bigstr,midend);
6212 =for apidoc sv_replace
6214 Make the first argument a copy of the second, then delete the original.
6215 The target SV physically takes over ownership of the body of the source SV
6216 and inherits its flags; however, the target keeps any magic it owns,
6217 and any magic in the source is discarded.
6218 Note that this is a rather specialist SV copying operation; most of the
6219 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6225 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6227 const U32 refcnt = SvREFCNT(sv);
6229 PERL_ARGS_ASSERT_SV_REPLACE;
6231 SV_CHECK_THINKFIRST_COW_DROP(sv);
6232 if (SvREFCNT(nsv) != 1) {
6233 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6234 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6236 if (SvMAGICAL(sv)) {
6240 sv_upgrade(nsv, SVt_PVMG);
6241 SvMAGIC_set(nsv, SvMAGIC(sv));
6242 SvFLAGS(nsv) |= SvMAGICAL(sv);
6244 SvMAGIC_set(sv, NULL);
6248 assert(!SvREFCNT(sv));
6249 #ifdef DEBUG_LEAKING_SCALARS
6250 sv->sv_flags = nsv->sv_flags;
6251 sv->sv_any = nsv->sv_any;
6252 sv->sv_refcnt = nsv->sv_refcnt;
6253 sv->sv_u = nsv->sv_u;
6255 StructCopy(nsv,sv,SV);
6257 if(SvTYPE(sv) == SVt_IV) {
6259 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6263 #ifdef PERL_OLD_COPY_ON_WRITE
6264 if (SvIsCOW_normal(nsv)) {
6265 /* We need to follow the pointers around the loop to make the
6266 previous SV point to sv, rather than nsv. */
6269 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6272 assert(SvPVX_const(current) == SvPVX_const(nsv));
6274 /* Make the SV before us point to the SV after us. */
6276 PerlIO_printf(Perl_debug_log, "previous is\n");
6278 PerlIO_printf(Perl_debug_log,
6279 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6280 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6282 SV_COW_NEXT_SV_SET(current, sv);
6285 SvREFCNT(sv) = refcnt;
6286 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6291 /* We're about to free a GV which has a CV that refers back to us.
6292 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6296 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6301 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6304 assert(SvREFCNT(gv) == 0);
6305 assert(isGV(gv) && isGV_with_GP(gv));
6307 assert(!CvANON(cv));
6308 assert(CvGV(cv) == gv);
6309 assert(!CvNAMED(cv));
6311 /* will the CV shortly be freed by gp_free() ? */
6312 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6313 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6317 /* if not, anonymise: */
6318 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6319 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6320 : newSVpvn_flags( "__ANON__", 8, 0 );
6321 sv_catpvs(gvname, "::__ANON__");
6322 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6323 SvREFCNT_dec_NN(gvname);
6327 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6332 =for apidoc sv_clear
6334 Clear an SV: call any destructors, free up any memory used by the body,
6335 and free the body itself. The SV's head is I<not> freed, although
6336 its type is set to all 1's so that it won't inadvertently be assumed
6337 to be live during global destruction etc.
6338 This function should only be called when REFCNT is zero. Most of the time
6339 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6346 Perl_sv_clear(pTHX_ SV *const orig_sv)
6351 const struct body_details *sv_type_details;
6357 PERL_ARGS_ASSERT_SV_CLEAR;
6359 /* within this loop, sv is the SV currently being freed, and
6360 * iter_sv is the most recent AV or whatever that's being iterated
6361 * over to provide more SVs */
6367 assert(SvREFCNT(sv) == 0);
6368 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6370 if (type <= SVt_IV) {
6371 /* See the comment in sv.h about the collusion between this
6372 * early return and the overloading of the NULL slots in the
6376 SvFLAGS(sv) &= SVf_BREAK;
6377 SvFLAGS(sv) |= SVTYPEMASK;
6381 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6383 if (type >= SVt_PVMG) {
6385 if (!curse(sv, 1)) goto get_next_sv;
6386 type = SvTYPE(sv); /* destructor may have changed it */
6388 /* Free back-references before magic, in case the magic calls
6389 * Perl code that has weak references to sv. */
6390 if (type == SVt_PVHV) {
6391 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6395 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6396 SvREFCNT_dec(SvOURSTASH(sv));
6398 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6399 assert(!SvMAGICAL(sv));
6400 } else if (SvMAGIC(sv)) {
6401 /* Free back-references before other types of magic. */
6402 sv_unmagic(sv, PERL_MAGIC_backref);
6406 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6407 SvREFCNT_dec(SvSTASH(sv));
6410 /* case SVt_INVLIST: */
6413 IoIFP(sv) != PerlIO_stdin() &&
6414 IoIFP(sv) != PerlIO_stdout() &&
6415 IoIFP(sv) != PerlIO_stderr() &&
6416 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6418 io_close(MUTABLE_IO(sv), FALSE);
6420 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6421 PerlDir_close(IoDIRP(sv));
6422 IoDIRP(sv) = (DIR*)NULL;
6423 Safefree(IoTOP_NAME(sv));
6424 Safefree(IoFMT_NAME(sv));
6425 Safefree(IoBOTTOM_NAME(sv));
6426 if ((const GV *)sv == PL_statgv)
6430 /* FIXME for plugins */
6432 pregfree2((REGEXP*) sv);
6436 cv_undef(MUTABLE_CV(sv));
6437 /* If we're in a stash, we don't own a reference to it.
6438 * However it does have a back reference to us, which needs to
6440 if ((stash = CvSTASH(sv)))
6441 sv_del_backref(MUTABLE_SV(stash), sv);
6444 if (PL_last_swash_hv == (const HV *)sv) {
6445 PL_last_swash_hv = NULL;
6447 if (HvTOTALKEYS((HV*)sv) > 0) {
6449 /* this statement should match the one at the beginning of
6450 * hv_undef_flags() */
6451 if ( PL_phase != PERL_PHASE_DESTRUCT
6452 && (name = HvNAME((HV*)sv)))
6454 if (PL_stashcache) {
6455 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6457 (void)hv_deletehek(PL_stashcache,
6458 HvNAME_HEK((HV*)sv), G_DISCARD);
6460 hv_name_set((HV*)sv, NULL, 0, 0);
6463 /* save old iter_sv in unused SvSTASH field */
6464 assert(!SvOBJECT(sv));
6465 SvSTASH(sv) = (HV*)iter_sv;
6468 /* save old hash_index in unused SvMAGIC field */
6469 assert(!SvMAGICAL(sv));
6470 assert(!SvMAGIC(sv));
6471 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6474 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6475 goto get_next_sv; /* process this new sv */
6477 /* free empty hash */
6478 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6479 assert(!HvARRAY((HV*)sv));
6483 AV* av = MUTABLE_AV(sv);
6484 if (PL_comppad == av) {
6488 if (AvREAL(av) && AvFILLp(av) > -1) {
6489 next_sv = AvARRAY(av)[AvFILLp(av)--];
6490 /* save old iter_sv in top-most slot of AV,
6491 * and pray that it doesn't get wiped in the meantime */
6492 AvARRAY(av)[AvMAX(av)] = iter_sv;
6494 goto get_next_sv; /* process this new sv */
6496 Safefree(AvALLOC(av));
6501 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6502 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6503 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6504 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6506 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6507 SvREFCNT_dec(LvTARG(sv));
6508 if (isREGEXP(sv)) goto freeregexp;
6510 if (isGV_with_GP(sv)) {
6511 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6512 && HvENAME_get(stash))
6513 mro_method_changed_in(stash);
6514 gp_free(MUTABLE_GV(sv));
6516 unshare_hek(GvNAME_HEK(sv));
6517 /* If we're in a stash, we don't own a reference to it.
6518 * However it does have a back reference to us, which
6519 * needs to be cleared. */
6520 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6521 sv_del_backref(MUTABLE_SV(stash), sv);
6523 /* FIXME. There are probably more unreferenced pointers to SVs
6524 * in the interpreter struct that we should check and tidy in
6525 * a similar fashion to this: */
6526 /* See also S_sv_unglob, which does the same thing. */
6527 if ((const GV *)sv == PL_last_in_gv)
6528 PL_last_in_gv = NULL;
6529 else if ((const GV *)sv == PL_statgv)
6531 else if ((const GV *)sv == PL_stderrgv)
6539 /* Don't bother with SvOOK_off(sv); as we're only going to
6543 SvOOK_offset(sv, offset);
6544 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6545 /* Don't even bother with turning off the OOK flag. */
6550 SV * const target = SvRV(sv);
6552 sv_del_backref(target, sv);
6558 else if (SvPVX_const(sv)
6559 && !(SvTYPE(sv) == SVt_PVIO
6560 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6564 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6568 # ifdef PERL_OLD_COPY_ON_WRITE
6569 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6571 if (CowREFCNT(sv)) {
6579 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6583 # ifdef PERL_OLD_COPY_ON_WRITE
6587 Safefree(SvPVX_mutable(sv));
6591 else if (SvPVX_const(sv) && SvLEN(sv)
6592 && !(SvTYPE(sv) == SVt_PVIO
6593 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6594 Safefree(SvPVX_mutable(sv));
6595 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6596 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6606 SvFLAGS(sv) &= SVf_BREAK;
6607 SvFLAGS(sv) |= SVTYPEMASK;
6609 sv_type_details = bodies_by_type + type;
6610 if (sv_type_details->arena) {
6611 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6612 &PL_body_roots[type]);
6614 else if (sv_type_details->body_size) {
6615 safefree(SvANY(sv));
6619 /* caller is responsible for freeing the head of the original sv */
6620 if (sv != orig_sv && !SvREFCNT(sv))
6623 /* grab and free next sv, if any */
6631 else if (!iter_sv) {
6633 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6634 AV *const av = (AV*)iter_sv;
6635 if (AvFILLp(av) > -1) {
6636 sv = AvARRAY(av)[AvFILLp(av)--];
6638 else { /* no more elements of current AV to free */
6641 /* restore previous value, squirrelled away */
6642 iter_sv = AvARRAY(av)[AvMAX(av)];
6643 Safefree(AvALLOC(av));
6646 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6647 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6648 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6649 /* no more elements of current HV to free */
6652 /* Restore previous values of iter_sv and hash_index,
6653 * squirrelled away */
6654 assert(!SvOBJECT(sv));
6655 iter_sv = (SV*)SvSTASH(sv);
6656 assert(!SvMAGICAL(sv));
6657 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6659 /* perl -DA does not like rubbish in SvMAGIC. */
6663 /* free any remaining detritus from the hash struct */
6664 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6665 assert(!HvARRAY((HV*)sv));
6670 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6674 if (!SvREFCNT(sv)) {
6678 if (--(SvREFCNT(sv)))
6682 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6683 "Attempt to free temp prematurely: SV 0x%"UVxf
6684 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6688 if (SvIMMORTAL(sv)) {
6689 /* make sure SvREFCNT(sv)==0 happens very seldom */
6690 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6699 /* This routine curses the sv itself, not the object referenced by sv. So
6700 sv does not have to be ROK. */
6703 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6704 PERL_ARGS_ASSERT_CURSE;
6705 assert(SvOBJECT(sv));
6707 if (PL_defstash && /* Still have a symbol table? */
6713 stash = SvSTASH(sv);
6714 assert(SvTYPE(stash) == SVt_PVHV);
6715 if (HvNAME(stash)) {
6716 CV* destructor = NULL;
6717 assert (SvOOK(stash));
6718 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6719 if (!destructor || HvMROMETA(stash)->destroy_gen
6720 != PL_sub_generation)
6723 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6724 if (gv) destructor = GvCV(gv);
6725 if (!SvOBJECT(stash))
6728 destructor ? (HV *)destructor : ((HV *)0)+1;
6729 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6733 assert(!destructor || destructor == ((CV *)0)+1
6734 || SvTYPE(destructor) == SVt_PVCV);
6735 if (destructor && destructor != ((CV *)0)+1
6736 /* A constant subroutine can have no side effects, so
6737 don't bother calling it. */
6738 && !CvCONST(destructor)
6739 /* Don't bother calling an empty destructor or one that
6740 returns immediately. */
6741 && (CvISXSUB(destructor)
6742 || (CvSTART(destructor)
6743 && (CvSTART(destructor)->op_next->op_type
6745 && (CvSTART(destructor)->op_next->op_type
6747 || CvSTART(destructor)->op_next->op_next->op_type
6753 SV* const tmpref = newRV(sv);
6754 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6756 PUSHSTACKi(PERLSI_DESTROY);
6761 call_sv(MUTABLE_SV(destructor),
6762 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6766 if(SvREFCNT(tmpref) < 2) {
6767 /* tmpref is not kept alive! */
6769 SvRV_set(tmpref, NULL);
6772 SvREFCNT_dec_NN(tmpref);
6775 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6778 if (check_refcnt && SvREFCNT(sv)) {
6779 if (PL_in_clean_objs)
6781 "DESTROY created new reference to dead object '%"HEKf"'",
6782 HEKfARG(HvNAME_HEK(stash)));
6783 /* DESTROY gave object new lease on life */
6789 HV * const stash = SvSTASH(sv);
6790 /* Curse before freeing the stash, as freeing the stash could cause
6791 a recursive call into S_curse. */
6792 SvOBJECT_off(sv); /* Curse the object. */
6793 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6794 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6800 =for apidoc sv_newref
6802 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6809 Perl_sv_newref(pTHX_ SV *const sv)
6811 PERL_UNUSED_CONTEXT;
6820 Decrement an SV's reference count, and if it drops to zero, call
6821 C<sv_clear> to invoke destructors and free up any memory used by
6822 the body; finally, deallocate the SV's head itself.
6823 Normally called via a wrapper macro C<SvREFCNT_dec>.
6829 Perl_sv_free(pTHX_ SV *const sv)
6835 /* Private helper function for SvREFCNT_dec().
6836 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6839 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6843 PERL_ARGS_ASSERT_SV_FREE2;
6845 if (LIKELY( rc == 1 )) {
6851 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6852 "Attempt to free temp prematurely: SV 0x%"UVxf
6853 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6857 if (SvIMMORTAL(sv)) {
6858 /* make sure SvREFCNT(sv)==0 happens very seldom */
6859 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6863 if (! SvREFCNT(sv)) /* may have have been resurrected */
6868 /* handle exceptional cases */
6872 if (SvFLAGS(sv) & SVf_BREAK)
6873 /* this SV's refcnt has been artificially decremented to
6874 * trigger cleanup */
6876 if (PL_in_clean_all) /* All is fair */
6878 if (SvIMMORTAL(sv)) {
6879 /* make sure SvREFCNT(sv)==0 happens very seldom */
6880 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6883 if (ckWARN_d(WARN_INTERNAL)) {
6884 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6885 Perl_dump_sv_child(aTHX_ sv);
6887 #ifdef DEBUG_LEAKING_SCALARS
6890 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6891 if (PL_warnhook == PERL_WARNHOOK_FATAL
6892 || ckDEAD(packWARN(WARN_INTERNAL))) {
6893 /* Don't let Perl_warner cause us to escape our fate: */
6897 /* This may not return: */
6898 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6899 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6900 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6903 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6913 Returns the length of the string in the SV. Handles magic and type
6914 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6915 gives raw access to the xpv_cur slot.
6921 Perl_sv_len(pTHX_ SV *const sv)
6928 (void)SvPV_const(sv, len);
6933 =for apidoc sv_len_utf8
6935 Returns the number of characters in the string in an SV, counting wide
6936 UTF-8 bytes as a single character. Handles magic and type coercion.
6942 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6943 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6944 * (Note that the mg_len is not the length of the mg_ptr field.
6945 * This allows the cache to store the character length of the string without
6946 * needing to malloc() extra storage to attach to the mg_ptr.)
6951 Perl_sv_len_utf8(pTHX_ SV *const sv)
6957 return sv_len_utf8_nomg(sv);
6961 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6964 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6966 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6968 if (PL_utf8cache && SvUTF8(sv)) {
6970 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6972 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6973 if (mg->mg_len != -1)
6976 /* We can use the offset cache for a headstart.
6977 The longer value is stored in the first pair. */
6978 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6980 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
6984 if (PL_utf8cache < 0) {
6985 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
6986 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
6990 ulen = Perl_utf8_length(aTHX_ s, s + len);
6991 utf8_mg_len_cache_update(sv, &mg, ulen);
6995 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
6998 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7001 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7002 STRLEN *const uoffset_p, bool *const at_end)
7004 const U8 *s = start;
7005 STRLEN uoffset = *uoffset_p;
7007 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7009 while (s < send && uoffset) {
7016 else if (s > send) {
7018 /* This is the existing behaviour. Possibly it should be a croak, as
7019 it's actually a bounds error */
7022 *uoffset_p -= uoffset;
7026 /* Given the length of the string in both bytes and UTF-8 characters, decide
7027 whether to walk forwards or backwards to find the byte corresponding to
7028 the passed in UTF-8 offset. */
7030 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7031 STRLEN uoffset, const STRLEN uend)
7033 STRLEN backw = uend - uoffset;
7035 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7037 if (uoffset < 2 * backw) {
7038 /* The assumption is that going forwards is twice the speed of going
7039 forward (that's where the 2 * backw comes from).
7040 (The real figure of course depends on the UTF-8 data.) */
7041 const U8 *s = start;
7043 while (s < send && uoffset--)
7053 while (UTF8_IS_CONTINUATION(*send))
7056 return send - start;
7059 /* For the string representation of the given scalar, find the byte
7060 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7061 give another position in the string, *before* the sought offset, which
7062 (which is always true, as 0, 0 is a valid pair of positions), which should
7063 help reduce the amount of linear searching.
7064 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7065 will be used to reduce the amount of linear searching. The cache will be
7066 created if necessary, and the found value offered to it for update. */
7068 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7069 const U8 *const send, STRLEN uoffset,
7070 STRLEN uoffset0, STRLEN boffset0)
7072 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7074 bool at_end = FALSE;
7076 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7078 assert (uoffset >= uoffset0);
7083 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7085 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7086 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7087 if ((*mgp)->mg_ptr) {
7088 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7089 if (cache[0] == uoffset) {
7090 /* An exact match. */
7093 if (cache[2] == uoffset) {
7094 /* An exact match. */
7098 if (cache[0] < uoffset) {
7099 /* The cache already knows part of the way. */
7100 if (cache[0] > uoffset0) {
7101 /* The cache knows more than the passed in pair */
7102 uoffset0 = cache[0];
7103 boffset0 = cache[1];
7105 if ((*mgp)->mg_len != -1) {
7106 /* And we know the end too. */
7108 + sv_pos_u2b_midway(start + boffset0, send,
7110 (*mgp)->mg_len - uoffset0);
7112 uoffset -= uoffset0;
7114 + sv_pos_u2b_forwards(start + boffset0,
7115 send, &uoffset, &at_end);
7116 uoffset += uoffset0;
7119 else if (cache[2] < uoffset) {
7120 /* We're between the two cache entries. */
7121 if (cache[2] > uoffset0) {
7122 /* and the cache knows more than the passed in pair */
7123 uoffset0 = cache[2];
7124 boffset0 = cache[3];
7128 + sv_pos_u2b_midway(start + boffset0,
7131 cache[0] - uoffset0);
7134 + sv_pos_u2b_midway(start + boffset0,
7137 cache[2] - uoffset0);
7141 else if ((*mgp)->mg_len != -1) {
7142 /* If we can take advantage of a passed in offset, do so. */
7143 /* In fact, offset0 is either 0, or less than offset, so don't
7144 need to worry about the other possibility. */
7146 + sv_pos_u2b_midway(start + boffset0, send,
7148 (*mgp)->mg_len - uoffset0);
7153 if (!found || PL_utf8cache < 0) {
7154 STRLEN real_boffset;
7155 uoffset -= uoffset0;
7156 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7157 send, &uoffset, &at_end);
7158 uoffset += uoffset0;
7160 if (found && PL_utf8cache < 0)
7161 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7163 boffset = real_boffset;
7166 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7168 utf8_mg_len_cache_update(sv, mgp, uoffset);
7170 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7177 =for apidoc sv_pos_u2b_flags
7179 Converts the offset from a count of UTF-8 chars from
7180 the start of the string, to a count of the equivalent number of bytes; if
7181 lenp is non-zero, it does the same to lenp, but this time starting from
7182 the offset, rather than from the start
7183 of the string. Handles type coercion.
7184 I<flags> is passed to C<SvPV_flags>, and usually should be
7185 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7191 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7192 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7193 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7198 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7205 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7207 start = (U8*)SvPV_flags(sv, len, flags);
7209 const U8 * const send = start + len;
7211 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7214 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7215 is 0, and *lenp is already set to that. */) {
7216 /* Convert the relative offset to absolute. */
7217 const STRLEN uoffset2 = uoffset + *lenp;
7218 const STRLEN boffset2
7219 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7220 uoffset, boffset) - boffset;
7234 =for apidoc sv_pos_u2b
7236 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7237 the start of the string, to a count of the equivalent number of bytes; if
7238 lenp is non-zero, it does the same to lenp, but this time starting from
7239 the offset, rather than from the start of the string. Handles magic and
7242 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7249 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7250 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7251 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7255 /* This function is subject to size and sign problems */
7258 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7260 PERL_ARGS_ASSERT_SV_POS_U2B;
7263 STRLEN ulen = (STRLEN)*lenp;
7264 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7265 SV_GMAGIC|SV_CONST_RETURN);
7268 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7269 SV_GMAGIC|SV_CONST_RETURN);
7274 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7277 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7278 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7281 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7282 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7283 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7287 (*mgp)->mg_len = ulen;
7290 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7291 byte length pairing. The (byte) length of the total SV is passed in too,
7292 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7293 may not have updated SvCUR, so we can't rely on reading it directly.
7295 The proffered utf8/byte length pairing isn't used if the cache already has
7296 two pairs, and swapping either for the proffered pair would increase the
7297 RMS of the intervals between known byte offsets.
7299 The cache itself consists of 4 STRLEN values
7300 0: larger UTF-8 offset
7301 1: corresponding byte offset
7302 2: smaller UTF-8 offset
7303 3: corresponding byte offset
7305 Unused cache pairs have the value 0, 0.
7306 Keeping the cache "backwards" means that the invariant of
7307 cache[0] >= cache[2] is maintained even with empty slots, which means that
7308 the code that uses it doesn't need to worry if only 1 entry has actually
7309 been set to non-zero. It also makes the "position beyond the end of the
7310 cache" logic much simpler, as the first slot is always the one to start
7314 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7315 const STRLEN utf8, const STRLEN blen)
7319 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7324 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7325 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7326 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7328 (*mgp)->mg_len = -1;
7332 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7333 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7334 (*mgp)->mg_ptr = (char *) cache;
7338 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7339 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7340 a pointer. Note that we no longer cache utf8 offsets on refer-
7341 ences, but this check is still a good idea, for robustness. */
7342 const U8 *start = (const U8 *) SvPVX_const(sv);
7343 const STRLEN realutf8 = utf8_length(start, start + byte);
7345 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7349 /* Cache is held with the later position first, to simplify the code
7350 that deals with unbounded ends. */
7352 ASSERT_UTF8_CACHE(cache);
7353 if (cache[1] == 0) {
7354 /* Cache is totally empty */
7357 } else if (cache[3] == 0) {
7358 if (byte > cache[1]) {
7359 /* New one is larger, so goes first. */
7360 cache[2] = cache[0];
7361 cache[3] = cache[1];
7369 #define THREEWAY_SQUARE(a,b,c,d) \
7370 ((float)((d) - (c))) * ((float)((d) - (c))) \
7371 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7372 + ((float)((b) - (a))) * ((float)((b) - (a)))
7374 /* Cache has 2 slots in use, and we know three potential pairs.
7375 Keep the two that give the lowest RMS distance. Do the
7376 calculation in bytes simply because we always know the byte
7377 length. squareroot has the same ordering as the positive value,
7378 so don't bother with the actual square root. */
7379 if (byte > cache[1]) {
7380 /* New position is after the existing pair of pairs. */
7381 const float keep_earlier
7382 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7383 const float keep_later
7384 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7386 if (keep_later < keep_earlier) {
7387 cache[2] = cache[0];
7388 cache[3] = cache[1];
7397 else if (byte > cache[3]) {
7398 /* New position is between the existing pair of pairs. */
7399 const float keep_earlier
7400 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7401 const float keep_later
7402 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7404 if (keep_later < keep_earlier) {
7414 /* New position is before the existing pair of pairs. */
7415 const float keep_earlier
7416 = THREEWAY_SQUARE(0, byte, cache[3], blen);
7417 const float keep_later
7418 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7420 if (keep_later < keep_earlier) {
7425 cache[0] = cache[2];
7426 cache[1] = cache[3];
7432 ASSERT_UTF8_CACHE(cache);
7435 /* We already know all of the way, now we may be able to walk back. The same
7436 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7437 backward is half the speed of walking forward. */
7439 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7440 const U8 *end, STRLEN endu)
7442 const STRLEN forw = target - s;
7443 STRLEN backw = end - target;
7445 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7447 if (forw < 2 * backw) {
7448 return utf8_length(s, target);
7451 while (end > target) {
7453 while (UTF8_IS_CONTINUATION(*end)) {
7462 =for apidoc sv_pos_b2u_flags
7464 Converts the offset from a count of bytes from the start of the string, to
7465 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7466 I<flags> is passed to C<SvPV_flags>, and usually should be
7467 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7473 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7474 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7479 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7482 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7488 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7490 s = (const U8*)SvPV_flags(sv, blen, flags);
7493 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7494 ", byte=%"UVuf, (UV)blen, (UV)offset);
7500 && SvTYPE(sv) >= SVt_PVMG
7501 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7504 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7505 if (cache[1] == offset) {
7506 /* An exact match. */
7509 if (cache[3] == offset) {
7510 /* An exact match. */
7514 if (cache[1] < offset) {
7515 /* We already know part of the way. */
7516 if (mg->mg_len != -1) {
7517 /* Actually, we know the end too. */
7519 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7520 s + blen, mg->mg_len - cache[0]);
7522 len = cache[0] + utf8_length(s + cache[1], send);
7525 else if (cache[3] < offset) {
7526 /* We're between the two cached pairs, so we do the calculation
7527 offset by the byte/utf-8 positions for the earlier pair,
7528 then add the utf-8 characters from the string start to
7530 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7531 s + cache[1], cache[0] - cache[2])
7535 else { /* cache[3] > offset */
7536 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7540 ASSERT_UTF8_CACHE(cache);
7542 } else if (mg->mg_len != -1) {
7543 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7547 if (!found || PL_utf8cache < 0) {
7548 const STRLEN real_len = utf8_length(s, send);
7550 if (found && PL_utf8cache < 0)
7551 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7557 utf8_mg_len_cache_update(sv, &mg, len);
7559 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7566 =for apidoc sv_pos_b2u
7568 Converts the value pointed to by offsetp from a count of bytes from the
7569 start of the string, to a count of the equivalent number of UTF-8 chars.
7570 Handles magic and type coercion.
7572 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7579 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7580 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7585 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7587 PERL_ARGS_ASSERT_SV_POS_B2U;
7592 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7593 SV_GMAGIC|SV_CONST_RETURN);
7597 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7598 STRLEN real, SV *const sv)
7600 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7602 /* As this is debugging only code, save space by keeping this test here,
7603 rather than inlining it in all the callers. */
7604 if (from_cache == real)
7607 /* Need to turn the assertions off otherwise we may recurse infinitely
7608 while printing error messages. */
7609 SAVEI8(PL_utf8cache);
7611 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7612 func, (UV) from_cache, (UV) real, SVfARG(sv));
7618 Returns a boolean indicating whether the strings in the two SVs are
7619 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7620 coerce its args to strings if necessary.
7622 =for apidoc sv_eq_flags
7624 Returns a boolean indicating whether the strings in the two SVs are
7625 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7626 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7632 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7639 SV* svrecode = NULL;
7646 /* if pv1 and pv2 are the same, second SvPV_const call may
7647 * invalidate pv1 (if we are handling magic), so we may need to
7649 if (sv1 == sv2 && flags & SV_GMAGIC
7650 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7651 pv1 = SvPV_const(sv1, cur1);
7652 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7654 pv1 = SvPV_flags_const(sv1, cur1, flags);
7662 pv2 = SvPV_flags_const(sv2, cur2, flags);
7664 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7665 /* Differing utf8ness.
7666 * Do not UTF8size the comparands as a side-effect. */
7669 svrecode = newSVpvn(pv2, cur2);
7670 sv_recode_to_utf8(svrecode, PL_encoding);
7671 pv2 = SvPV_const(svrecode, cur2);
7674 svrecode = newSVpvn(pv1, cur1);
7675 sv_recode_to_utf8(svrecode, PL_encoding);
7676 pv1 = SvPV_const(svrecode, cur1);
7678 /* Now both are in UTF-8. */
7680 SvREFCNT_dec_NN(svrecode);
7686 /* sv1 is the UTF-8 one */
7687 return bytes_cmp_utf8((const U8*)pv2, cur2,
7688 (const U8*)pv1, cur1) == 0;
7691 /* sv2 is the UTF-8 one */
7692 return bytes_cmp_utf8((const U8*)pv1, cur1,
7693 (const U8*)pv2, cur2) == 0;
7699 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7701 SvREFCNT_dec(svrecode);
7709 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7710 string in C<sv1> is less than, equal to, or greater than the string in
7711 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7712 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7714 =for apidoc sv_cmp_flags
7716 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7717 string in C<sv1> is less than, equal to, or greater than the string in
7718 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7719 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7720 also C<sv_cmp_locale_flags>.
7726 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7728 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7732 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7736 const char *pv1, *pv2;
7738 SV *svrecode = NULL;
7745 pv1 = SvPV_flags_const(sv1, cur1, flags);
7752 pv2 = SvPV_flags_const(sv2, cur2, flags);
7754 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7755 /* Differing utf8ness.
7756 * Do not UTF8size the comparands as a side-effect. */
7759 svrecode = newSVpvn(pv2, cur2);
7760 sv_recode_to_utf8(svrecode, PL_encoding);
7761 pv2 = SvPV_const(svrecode, cur2);
7764 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7765 (const U8*)pv1, cur1);
7766 return retval ? retval < 0 ? -1 : +1 : 0;
7771 svrecode = newSVpvn(pv1, cur1);
7772 sv_recode_to_utf8(svrecode, PL_encoding);
7773 pv1 = SvPV_const(svrecode, cur1);
7776 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7777 (const U8*)pv2, cur2);
7778 return retval ? retval < 0 ? -1 : +1 : 0;
7784 cmp = cur2 ? -1 : 0;
7788 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7791 cmp = retval < 0 ? -1 : 1;
7792 } else if (cur1 == cur2) {
7795 cmp = cur1 < cur2 ? -1 : 1;
7799 SvREFCNT_dec(svrecode);
7805 =for apidoc sv_cmp_locale
7807 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7808 'use bytes' aware, handles get magic, and will coerce its args to strings
7809 if necessary. See also C<sv_cmp>.
7811 =for apidoc sv_cmp_locale_flags
7813 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7814 'use bytes' aware and will coerce its args to strings if necessary. If the
7815 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7821 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7823 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7827 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7830 #ifdef USE_LOCALE_COLLATE
7836 if (PL_collation_standard)
7840 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7842 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7844 if (!pv1 || !len1) {
7855 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7858 return retval < 0 ? -1 : 1;
7861 * When the result of collation is equality, that doesn't mean
7862 * that there are no differences -- some locales exclude some
7863 * characters from consideration. So to avoid false equalities,
7864 * we use the raw string as a tiebreaker.
7871 PERL_UNUSED_ARG(flags);
7872 #endif /* USE_LOCALE_COLLATE */
7874 return sv_cmp(sv1, sv2);
7878 #ifdef USE_LOCALE_COLLATE
7881 =for apidoc sv_collxfrm
7883 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7884 C<sv_collxfrm_flags>.
7886 =for apidoc sv_collxfrm_flags
7888 Add Collate Transform magic to an SV if it doesn't already have it. If the
7889 flags contain SV_GMAGIC, it handles get-magic.
7891 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7892 scalar data of the variable, but transformed to such a format that a normal
7893 memory comparison can be used to compare the data according to the locale
7900 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7904 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7906 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7907 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7913 Safefree(mg->mg_ptr);
7914 s = SvPV_flags_const(sv, len, flags);
7915 if ((xf = mem_collxfrm(s, len, &xlen))) {
7917 #ifdef PERL_OLD_COPY_ON_WRITE
7919 sv_force_normal_flags(sv, 0);
7921 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
7935 if (mg && mg->mg_ptr) {
7937 return mg->mg_ptr + sizeof(PL_collation_ix);
7945 #endif /* USE_LOCALE_COLLATE */
7948 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7950 SV * const tsv = newSV(0);
7953 sv_gets(tsv, fp, 0);
7954 sv_utf8_upgrade_nomg(tsv);
7955 SvCUR_set(sv,append);
7958 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
7962 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7965 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
7966 /* Grab the size of the record we're getting */
7967 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
7974 /* With a true, record-oriented file on VMS, we need to use read directly
7975 * to ensure that we respect RMS record boundaries. The user is responsible
7976 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
7977 * record size) field. N.B. This is likely to produce invalid results on
7978 * varying-width character data when a record ends mid-character.
7980 fd = PerlIO_fileno(fp);
7982 && PerlLIO_fstat(fd, &st) == 0
7983 && (st.st_fab_rfm == FAB$C_VAR
7984 || st.st_fab_rfm == FAB$C_VFC
7985 || st.st_fab_rfm == FAB$C_FIX)) {
7987 bytesread = PerlLIO_read(fd, buffer, recsize);
7989 else /* in-memory file from PerlIO::Scalar
7990 * or not a record-oriented file
7994 bytesread = PerlIO_read(fp, buffer, recsize);
7996 /* At this point, the logic in sv_get() means that sv will
7997 be treated as utf-8 if the handle is utf8.
7999 if (PerlIO_isutf8(fp) && bytesread > 0) {
8000 char *bend = buffer + bytesread;
8001 char *bufp = buffer;
8002 size_t charcount = 0;
8003 bool charstart = TRUE;
8006 while (charcount < recsize) {
8007 /* count accumulated characters */
8008 while (bufp < bend) {
8010 skip = UTF8SKIP(bufp);
8012 if (bufp + skip > bend) {
8013 /* partial at the end */
8024 if (charcount < recsize) {
8026 STRLEN bufp_offset = bufp - buffer;
8027 SSize_t morebytesread;
8029 /* originally I read enough to fill any incomplete
8030 character and the first byte of the next
8031 character if needed, but if there's many
8032 multi-byte encoded characters we're going to be
8033 making a read call for every character beyond
8034 the original read size.
8036 So instead, read the rest of the character if
8037 any, and enough bytes to match at least the
8038 start bytes for each character we're going to
8042 readsize = recsize - charcount;
8044 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8045 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8046 bend = buffer + bytesread;
8047 morebytesread = PerlIO_read(fp, bend, readsize);
8048 if (morebytesread <= 0) {
8049 /* we're done, if we still have incomplete
8050 characters the check code in sv_gets() will
8053 I'd originally considered doing
8054 PerlIO_ungetc() on all but the lead
8055 character of the incomplete character, but
8056 read() doesn't do that, so I don't.
8061 /* prepare to scan some more */
8062 bytesread += morebytesread;
8063 bend = buffer + bytesread;
8064 bufp = buffer + bufp_offset;
8072 SvCUR_set(sv, bytesread + append);
8073 buffer[bytesread] = '\0';
8074 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8080 Get a line from the filehandle and store it into the SV, optionally
8081 appending to the currently-stored string. If C<append> is not 0, the
8082 line is appended to the SV instead of overwriting it. C<append> should
8083 be set to the byte offset that the appended string should start at
8084 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8090 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8100 PERL_ARGS_ASSERT_SV_GETS;
8102 if (SvTHINKFIRST(sv))
8103 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8104 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8106 However, perlbench says it's slower, because the existing swipe code
8107 is faster than copy on write.
8108 Swings and roundabouts. */
8109 SvUPGRADE(sv, SVt_PV);
8112 /* line is going to be appended to the existing buffer in the sv */
8113 if (PerlIO_isutf8(fp)) {
8115 sv_utf8_upgrade_nomg(sv);
8116 sv_pos_u2b(sv,&append,0);
8118 } else if (SvUTF8(sv)) {
8119 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8125 /* not appending - "clear" the string by setting SvCUR to 0,
8126 * the pv is still avaiable. */
8129 if (PerlIO_isutf8(fp))
8132 if (IN_PERL_COMPILETIME) {
8133 /* we always read code in line mode */
8137 else if (RsSNARF(PL_rs)) {
8138 /* If it is a regular disk file use size from stat() as estimate
8139 of amount we are going to read -- may result in mallocing
8140 more memory than we really need if the layers below reduce
8141 the size we read (e.g. CRLF or a gzip layer).
8144 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8145 const Off_t offset = PerlIO_tell(fp);
8146 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8147 #ifdef PERL_NEW_COPY_ON_WRITE
8148 /* Add an extra byte for the sake of copy-on-write's
8149 * buffer reference count. */
8150 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8152 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8159 else if (RsRECORD(PL_rs)) {
8160 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8162 else if (RsPARA(PL_rs)) {
8168 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8169 if (PerlIO_isutf8(fp)) {
8170 rsptr = SvPVutf8(PL_rs, rslen);
8173 if (SvUTF8(PL_rs)) {
8174 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8175 Perl_croak(aTHX_ "Wide character in $/");
8178 /* extract the raw pointer to the record separator */
8179 rsptr = SvPV_const(PL_rs, rslen);
8183 /* rslast is the last character in the record separator
8184 * note we don't use rslast except when rslen is true, so the
8185 * null assign is a placeholder. */
8186 rslast = rslen ? rsptr[rslen - 1] : '\0';
8188 if (rspara) { /* have to do this both before and after */
8189 do { /* to make sure file boundaries work right */
8192 i = PerlIO_getc(fp);
8196 PerlIO_ungetc(fp,i);
8202 /* See if we know enough about I/O mechanism to cheat it ! */
8204 /* This used to be #ifdef test - it is made run-time test for ease
8205 of abstracting out stdio interface. One call should be cheap
8206 enough here - and may even be a macro allowing compile
8210 if (PerlIO_fast_gets(fp)) {
8212 * We can do buffer based IO operations on this filehandle.
8214 * This means we can bypass a lot of subcalls and process
8215 * the buffer directly, it also means we know the upper bound
8216 * on the amount of data we might read of the current buffer
8217 * into our sv. Knowing this allows us to preallocate the pv
8218 * to be able to hold that maximum, which allows us to simplify
8219 * a lot of logic. */
8222 * We're going to steal some values from the stdio struct
8223 * and put EVERYTHING in the innermost loop into registers.
8225 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8226 STRLEN bpx; /* length of the data in the target sv
8227 used to fix pointers after a SvGROW */
8228 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8229 of data left in the read-ahead buffer.
8230 If 0 then the pv buffer can hold the full
8231 amount left, otherwise this is the amount it
8234 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8235 /* An ungetc()d char is handled separately from the regular
8236 * buffer, so we getc() it back out and stuff it in the buffer.
8238 i = PerlIO_getc(fp);
8239 if (i == EOF) return 0;
8240 *(--((*fp)->_ptr)) = (unsigned char) i;
8244 /* Here is some breathtakingly efficient cheating */
8246 /* When you read the following logic resist the urge to think
8247 * of record separators that are 1 byte long. They are an
8248 * uninteresting special (simple) case.
8250 * Instead think of record separators which are at least 2 bytes
8251 * long, and keep in mind that we need to deal with such
8252 * separators when they cross a read-ahead buffer boundary.
8254 * Also consider that we need to gracefully deal with separators
8255 * that may be longer than a single read ahead buffer.
8257 * Lastly do not forget we want to copy the delimiter as well. We
8258 * are copying all data in the file _up_to_and_including_ the separator
8261 * Now that you have all that in mind here is what is happening below:
8263 * 1. When we first enter the loop we do some memory book keeping to see
8264 * how much free space there is in the target SV. (This sub assumes that
8265 * it is operating on the same SV most of the time via $_ and that it is
8266 * going to be able to reuse the same pv buffer each call.) If there is
8267 * "enough" room then we set "shortbuffered" to how much space there is
8268 * and start reading forward.
8270 * 2. When we scan forward we copy from the read-ahead buffer to the target
8271 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8272 * and the end of the of pv, as well as for the "rslast", which is the last
8273 * char of the separator.
8275 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8276 * (which has a "complete" record up to the point we saw rslast) and check
8277 * it to see if it matches the separator. If it does we are done. If it doesn't
8278 * we continue on with the scan/copy.
8280 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8281 * the IO system to read the next buffer. We do this by doing a getc(), which
8282 * returns a single char read (or EOF), and prefills the buffer, and also
8283 * allows us to find out how full the buffer is. We use this information to
8284 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8285 * the returned single char into the target sv, and then go back into scan
8288 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8289 * remaining space in the read-buffer.
8291 * Note that this code despite its twisty-turny nature is pretty darn slick.
8292 * It manages single byte separators, multi-byte cross boundary separators,
8293 * and cross-read-buffer separators cleanly and efficiently at the cost
8294 * of potentially greatly overallocating the target SV.
8300 /* get the number of bytes remaining in the read-ahead buffer
8301 * on first call on a given fp this will return 0.*/
8302 cnt = PerlIO_get_cnt(fp);
8304 /* make sure we have the room */
8305 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8306 /* Not room for all of it
8307 if we are looking for a separator and room for some
8309 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8310 /* just process what we have room for */
8311 shortbuffered = cnt - SvLEN(sv) + append + 1;
8312 cnt -= shortbuffered;
8315 /* ensure that the target sv has enough room to hold
8316 * the rest of the read-ahead buffer */
8318 /* remember that cnt can be negative */
8319 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8323 /* we have enough room to hold the full buffer, lets scream */
8327 /* extract the pointer to sv's string buffer, offset by append as necessary */
8328 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8329 /* extract the point to the read-ahead buffer */
8330 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8332 /* some trace debug output */
8333 DEBUG_P(PerlIO_printf(Perl_debug_log,
8334 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8335 DEBUG_P(PerlIO_printf(Perl_debug_log,
8336 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8338 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8339 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8343 /* if there is stuff left in the read-ahead buffer */
8345 /* if there is a separator */
8347 /* loop until we hit the end of the read-ahead buffer */
8348 while (cnt > 0) { /* this | eat */
8349 /* scan forward copying and searching for rslast as we go */
8351 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8352 goto thats_all_folks; /* screams | sed :-) */
8356 /* no separator, slurp the full buffer */
8357 Copy(ptr, bp, cnt, char); /* this | eat */
8358 bp += cnt; /* screams | dust */
8359 ptr += cnt; /* louder | sed :-) */
8361 assert (!shortbuffered);
8362 goto cannot_be_shortbuffered;
8366 if (shortbuffered) { /* oh well, must extend */
8367 /* we didnt have enough room to fit the line into the target buffer
8368 * so we must extend the target buffer and keep going */
8369 cnt = shortbuffered;
8371 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8373 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8374 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8375 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8379 cannot_be_shortbuffered:
8380 /* we need to refill the read-ahead buffer if possible */
8382 DEBUG_P(PerlIO_printf(Perl_debug_log,
8383 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8384 PTR2UV(ptr),(IV)cnt));
8385 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8387 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8388 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8389 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8390 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8393 call PerlIO_getc() to let it prefill the lookahead buffer
8395 This used to call 'filbuf' in stdio form, but as that behaves like
8396 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8397 another abstraction.
8399 Note we have to deal with the char in 'i' if we are not at EOF
8401 i = PerlIO_getc(fp); /* get more characters */
8403 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8404 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8405 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8406 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8408 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8409 cnt = PerlIO_get_cnt(fp);
8410 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8411 DEBUG_P(PerlIO_printf(Perl_debug_log,
8412 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8413 PTR2UV(ptr),(IV)cnt));
8415 if (i == EOF) /* all done for ever? */
8416 goto thats_really_all_folks;
8418 /* make sure we have enough space in the target sv */
8419 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8421 SvGROW(sv, bpx + cnt + 2);
8422 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8424 /* copy of the char we got from getc() */
8425 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8427 /* make sure we deal with the i being the last character of a separator */
8428 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8429 goto thats_all_folks;
8433 /* check if we have actually found the separator - only really applies
8435 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8436 memNE((char*)bp - rslen, rsptr, rslen))
8437 goto screamer; /* go back to the fray */
8438 thats_really_all_folks:
8440 cnt += shortbuffered;
8441 DEBUG_P(PerlIO_printf(Perl_debug_log,
8442 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8443 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8444 DEBUG_P(PerlIO_printf(Perl_debug_log,
8445 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8447 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8448 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8450 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8451 DEBUG_P(PerlIO_printf(Perl_debug_log,
8452 "Screamer: done, len=%ld, string=|%.*s|\n",
8453 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8457 /*The big, slow, and stupid way. */
8458 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8459 STDCHAR *buf = NULL;
8460 Newx(buf, 8192, STDCHAR);
8468 const STDCHAR * const bpe = buf + sizeof(buf);
8470 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8471 ; /* keep reading */
8475 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8476 /* Accommodate broken VAXC compiler, which applies U8 cast to
8477 * both args of ?: operator, causing EOF to change into 255
8480 i = (U8)buf[cnt - 1];
8486 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8488 sv_catpvn_nomg(sv, (char *) buf, cnt);
8490 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8492 if (i != EOF && /* joy */
8494 SvCUR(sv) < rslen ||
8495 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8499 * If we're reading from a TTY and we get a short read,
8500 * indicating that the user hit his EOF character, we need
8501 * to notice it now, because if we try to read from the TTY
8502 * again, the EOF condition will disappear.
8504 * The comparison of cnt to sizeof(buf) is an optimization
8505 * that prevents unnecessary calls to feof().
8509 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8513 #ifdef USE_HEAP_INSTEAD_OF_STACK
8518 if (rspara) { /* have to do this both before and after */
8519 while (i != EOF) { /* to make sure file boundaries work right */
8520 i = PerlIO_getc(fp);
8522 PerlIO_ungetc(fp,i);
8528 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8534 Auto-increment of the value in the SV, doing string to numeric conversion
8535 if necessary. Handles 'get' magic and operator overloading.
8541 Perl_sv_inc(pTHX_ SV *const sv)
8550 =for apidoc sv_inc_nomg
8552 Auto-increment of the value in the SV, doing string to numeric conversion
8553 if necessary. Handles operator overloading. Skips handling 'get' magic.
8559 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8566 if (SvTHINKFIRST(sv)) {
8567 if (SvREADONLY(sv)) {
8568 Perl_croak_no_modify();
8572 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8574 i = PTR2IV(SvRV(sv));
8578 else sv_force_normal_flags(sv, 0);
8580 flags = SvFLAGS(sv);
8581 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8582 /* It's (privately or publicly) a float, but not tested as an
8583 integer, so test it to see. */
8585 flags = SvFLAGS(sv);
8587 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8588 /* It's publicly an integer, or privately an integer-not-float */
8589 #ifdef PERL_PRESERVE_IVUV
8593 if (SvUVX(sv) == UV_MAX)
8594 sv_setnv(sv, UV_MAX_P1);
8596 (void)SvIOK_only_UV(sv);
8597 SvUV_set(sv, SvUVX(sv) + 1);
8599 if (SvIVX(sv) == IV_MAX)
8600 sv_setuv(sv, (UV)IV_MAX + 1);
8602 (void)SvIOK_only(sv);
8603 SvIV_set(sv, SvIVX(sv) + 1);
8608 if (flags & SVp_NOK) {
8609 const NV was = SvNVX(sv);
8610 if (!Perl_isinfnan(was) &&
8611 NV_OVERFLOWS_INTEGERS_AT &&
8612 was >= NV_OVERFLOWS_INTEGERS_AT) {
8613 /* diag_listed_as: Lost precision when %s %f by 1 */
8614 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8615 "Lost precision when incrementing %" NVff " by 1",
8618 (void)SvNOK_only(sv);
8619 SvNV_set(sv, was + 1.0);
8623 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8624 if ((flags & SVTYPEMASK) < SVt_PVIV)
8625 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8626 (void)SvIOK_only(sv);
8631 while (isALPHA(*d)) d++;
8632 while (isDIGIT(*d)) d++;
8633 if (d < SvEND(sv)) {
8634 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8635 #ifdef PERL_PRESERVE_IVUV
8636 /* Got to punt this as an integer if needs be, but we don't issue
8637 warnings. Probably ought to make the sv_iv_please() that does
8638 the conversion if possible, and silently. */
8639 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8640 /* Need to try really hard to see if it's an integer.
8641 9.22337203685478e+18 is an integer.
8642 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8643 so $a="9.22337203685478e+18"; $a+0; $a++
8644 needs to be the same as $a="9.22337203685478e+18"; $a++
8651 /* sv_2iv *should* have made this an NV */
8652 if (flags & SVp_NOK) {
8653 (void)SvNOK_only(sv);
8654 SvNV_set(sv, SvNVX(sv) + 1.0);
8657 /* I don't think we can get here. Maybe I should assert this
8658 And if we do get here I suspect that sv_setnv will croak. NWC
8660 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8661 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8663 #endif /* PERL_PRESERVE_IVUV */
8664 if (!numtype && ckWARN(WARN_NUMERIC))
8665 not_incrementable(sv);
8666 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8670 while (d >= SvPVX_const(sv)) {
8678 /* MKS: The original code here died if letters weren't consecutive.
8679 * at least it didn't have to worry about non-C locales. The
8680 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8681 * arranged in order (although not consecutively) and that only
8682 * [A-Za-z] are accepted by isALPHA in the C locale.
8684 if (isALPHA_FOLD_NE(*d, 'z')) {
8685 do { ++*d; } while (!isALPHA(*d));
8688 *(d--) -= 'z' - 'a';
8693 *(d--) -= 'z' - 'a' + 1;
8697 /* oh,oh, the number grew */
8698 SvGROW(sv, SvCUR(sv) + 2);
8699 SvCUR_set(sv, SvCUR(sv) + 1);
8700 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8711 Auto-decrement of the value in the SV, doing string to numeric conversion
8712 if necessary. Handles 'get' magic and operator overloading.
8718 Perl_sv_dec(pTHX_ SV *const sv)
8727 =for apidoc sv_dec_nomg
8729 Auto-decrement of the value in the SV, doing string to numeric conversion
8730 if necessary. Handles operator overloading. Skips handling 'get' magic.
8736 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8742 if (SvTHINKFIRST(sv)) {
8743 if (SvREADONLY(sv)) {
8744 Perl_croak_no_modify();
8748 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8750 i = PTR2IV(SvRV(sv));
8754 else sv_force_normal_flags(sv, 0);
8756 /* Unlike sv_inc we don't have to worry about string-never-numbers
8757 and keeping them magic. But we mustn't warn on punting */
8758 flags = SvFLAGS(sv);
8759 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8760 /* It's publicly an integer, or privately an integer-not-float */
8761 #ifdef PERL_PRESERVE_IVUV
8765 if (SvUVX(sv) == 0) {
8766 (void)SvIOK_only(sv);
8770 (void)SvIOK_only_UV(sv);
8771 SvUV_set(sv, SvUVX(sv) - 1);
8774 if (SvIVX(sv) == IV_MIN) {
8775 sv_setnv(sv, (NV)IV_MIN);
8779 (void)SvIOK_only(sv);
8780 SvIV_set(sv, SvIVX(sv) - 1);
8785 if (flags & SVp_NOK) {
8788 const NV was = SvNVX(sv);
8789 if (!Perl_isinfnan(was) &&
8790 NV_OVERFLOWS_INTEGERS_AT &&
8791 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8792 /* diag_listed_as: Lost precision when %s %f by 1 */
8793 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8794 "Lost precision when decrementing %" NVff " by 1",
8797 (void)SvNOK_only(sv);
8798 SvNV_set(sv, was - 1.0);
8802 if (!(flags & SVp_POK)) {
8803 if ((flags & SVTYPEMASK) < SVt_PVIV)
8804 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8806 (void)SvIOK_only(sv);
8809 #ifdef PERL_PRESERVE_IVUV
8811 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8812 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8813 /* Need to try really hard to see if it's an integer.
8814 9.22337203685478e+18 is an integer.
8815 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8816 so $a="9.22337203685478e+18"; $a+0; $a--
8817 needs to be the same as $a="9.22337203685478e+18"; $a--
8824 /* sv_2iv *should* have made this an NV */
8825 if (flags & SVp_NOK) {
8826 (void)SvNOK_only(sv);
8827 SvNV_set(sv, SvNVX(sv) - 1.0);
8830 /* I don't think we can get here. Maybe I should assert this
8831 And if we do get here I suspect that sv_setnv will croak. NWC
8833 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8834 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8837 #endif /* PERL_PRESERVE_IVUV */
8838 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8841 /* this define is used to eliminate a chunk of duplicated but shared logic
8842 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8843 * used anywhere but here - yves
8845 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8848 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8852 =for apidoc sv_mortalcopy
8854 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8855 The new SV is marked as mortal. It will be destroyed "soon", either by an
8856 explicit call to FREETMPS, or by an implicit call at places such as
8857 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8862 /* Make a string that will exist for the duration of the expression
8863 * evaluation. Actually, it may have to last longer than that, but
8864 * hopefully we won't free it until it has been assigned to a
8865 * permanent location. */
8868 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8872 if (flags & SV_GMAGIC)
8873 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8875 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8876 PUSH_EXTEND_MORTAL__SV_C(sv);
8882 =for apidoc sv_newmortal
8884 Creates a new null SV which is mortal. The reference count of the SV is
8885 set to 1. It will be destroyed "soon", either by an explicit call to
8886 FREETMPS, or by an implicit call at places such as statement boundaries.
8887 See also C<sv_mortalcopy> and C<sv_2mortal>.
8893 Perl_sv_newmortal(pTHX)
8898 SvFLAGS(sv) = SVs_TEMP;
8899 PUSH_EXTEND_MORTAL__SV_C(sv);
8905 =for apidoc newSVpvn_flags
8907 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8908 characters) into it. The reference count for the
8909 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8910 string. You are responsible for ensuring that the source string is at least
8911 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8912 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8913 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8914 returning. If C<SVf_UTF8> is set, C<s>
8915 is considered to be in UTF-8 and the
8916 C<SVf_UTF8> flag will be set on the new SV.
8917 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8919 #define newSVpvn_utf8(s, len, u) \
8920 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8926 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8930 /* All the flags we don't support must be zero.
8931 And we're new code so I'm going to assert this from the start. */
8932 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
8934 sv_setpvn(sv,s,len);
8936 /* This code used to do a sv_2mortal(), however we now unroll the call to
8937 * sv_2mortal() and do what it does ourselves here. Since we have asserted
8938 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
8939 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
8940 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
8941 * means that we eliminate quite a few steps than it looks - Yves
8942 * (explaining patch by gfx) */
8944 SvFLAGS(sv) |= flags;
8946 if(flags & SVs_TEMP){
8947 PUSH_EXTEND_MORTAL__SV_C(sv);
8954 =for apidoc sv_2mortal
8956 Marks an existing SV as mortal. The SV will be destroyed "soon", either
8957 by an explicit call to FREETMPS, or by an implicit call at places such as
8958 statement boundaries. SvTEMP() is turned on which means that the SV's
8959 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
8960 and C<sv_mortalcopy>.
8966 Perl_sv_2mortal(pTHX_ SV *const sv)
8973 PUSH_EXTEND_MORTAL__SV_C(sv);
8981 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8982 characters) into it. The reference count for the
8983 SV is set to 1. If C<len> is zero, Perl will compute the length using
8984 strlen(), (which means if you use this option, that C<s> can't have embedded
8985 C<NUL> characters and has to have a terminating C<NUL> byte).
8987 For efficiency, consider using C<newSVpvn> instead.
8993 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
8998 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9003 =for apidoc newSVpvn
9005 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9006 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9007 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9008 are responsible for ensuring that the source buffer is at least
9009 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9016 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9020 sv_setpvn(sv,buffer,len);
9025 =for apidoc newSVhek
9027 Creates a new SV from the hash key structure. It will generate scalars that
9028 point to the shared string table where possible. Returns a new (undefined)
9029 SV if the hek is NULL.
9035 Perl_newSVhek(pTHX_ const HEK *const hek)
9044 if (HEK_LEN(hek) == HEf_SVKEY) {
9045 return newSVsv(*(SV**)HEK_KEY(hek));
9047 const int flags = HEK_FLAGS(hek);
9048 if (flags & HVhek_WASUTF8) {
9050 Andreas would like keys he put in as utf8 to come back as utf8
9052 STRLEN utf8_len = HEK_LEN(hek);
9053 SV * const sv = newSV_type(SVt_PV);
9054 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9055 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9056 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9059 } else if (flags & HVhek_UNSHARED) {
9060 /* A hash that isn't using shared hash keys has to have
9061 the flag in every key so that we know not to try to call
9062 share_hek_hek on it. */
9064 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9069 /* This will be overwhelminly the most common case. */
9071 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9072 more efficient than sharepvn(). */
9076 sv_upgrade(sv, SVt_PV);
9077 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9078 SvCUR_set(sv, HEK_LEN(hek));
9090 =for apidoc newSVpvn_share
9092 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9093 table. If the string does not already exist in the table, it is
9094 created first. Turns on the SvIsCOW flag (or READONLY
9095 and FAKE in 5.16 and earlier). If the C<hash> parameter
9096 is non-zero, that value is used; otherwise the hash is computed.
9097 The string's hash can later be retrieved from the SV
9098 with the C<SvSHARED_HASH()> macro. The idea here is
9099 that as the string table is used for shared hash keys these strings will have
9100 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9106 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9110 bool is_utf8 = FALSE;
9111 const char *const orig_src = src;
9114 STRLEN tmplen = -len;
9116 /* See the note in hv.c:hv_fetch() --jhi */
9117 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9121 PERL_HASH(hash, src, len);
9123 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9124 changes here, update it there too. */
9125 sv_upgrade(sv, SVt_PV);
9126 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9133 if (src != orig_src)
9139 =for apidoc newSVpv_share
9141 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9148 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9150 return newSVpvn_share(src, strlen(src), hash);
9153 #if defined(PERL_IMPLICIT_CONTEXT)
9155 /* pTHX_ magic can't cope with varargs, so this is a no-context
9156 * version of the main function, (which may itself be aliased to us).
9157 * Don't access this version directly.
9161 Perl_newSVpvf_nocontext(const char *const pat, ...)
9167 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9169 va_start(args, pat);
9170 sv = vnewSVpvf(pat, &args);
9177 =for apidoc newSVpvf
9179 Creates a new SV and initializes it with the string formatted like
9186 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9191 PERL_ARGS_ASSERT_NEWSVPVF;
9193 va_start(args, pat);
9194 sv = vnewSVpvf(pat, &args);
9199 /* backend for newSVpvf() and newSVpvf_nocontext() */
9202 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9206 PERL_ARGS_ASSERT_VNEWSVPVF;
9209 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9216 Creates a new SV and copies a floating point value into it.
9217 The reference count for the SV is set to 1.
9223 Perl_newSVnv(pTHX_ const NV n)
9235 Creates a new SV and copies an integer into it. The reference count for the
9242 Perl_newSViv(pTHX_ const IV i)
9254 Creates a new SV and copies an unsigned integer into it.
9255 The reference count for the SV is set to 1.
9261 Perl_newSVuv(pTHX_ const UV u)
9271 =for apidoc newSV_type
9273 Creates a new SV, of the type specified. The reference count for the new SV
9280 Perl_newSV_type(pTHX_ const svtype type)
9285 sv_upgrade(sv, type);
9290 =for apidoc newRV_noinc
9292 Creates an RV wrapper for an SV. The reference count for the original
9293 SV is B<not> incremented.
9299 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9301 SV *sv = newSV_type(SVt_IV);
9303 PERL_ARGS_ASSERT_NEWRV_NOINC;
9306 SvRV_set(sv, tmpRef);
9311 /* newRV_inc is the official function name to use now.
9312 * newRV_inc is in fact #defined to newRV in sv.h
9316 Perl_newRV(pTHX_ SV *const sv)
9318 PERL_ARGS_ASSERT_NEWRV;
9320 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9326 Creates a new SV which is an exact duplicate of the original SV.
9333 Perl_newSVsv(pTHX_ SV *const old)
9339 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9340 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9343 /* Do this here, otherwise we leak the new SV if this croaks. */
9346 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9347 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9348 sv_setsv_flags(sv, old, SV_NOSTEAL);
9353 =for apidoc sv_reset
9355 Underlying implementation for the C<reset> Perl function.
9356 Note that the perl-level function is vaguely deprecated.
9362 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9364 PERL_ARGS_ASSERT_SV_RESET;
9366 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9370 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9372 char todo[PERL_UCHAR_MAX+1];
9375 if (!stash || SvTYPE(stash) != SVt_PVHV)
9378 if (!s) { /* reset ?? searches */
9379 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9381 const U32 count = mg->mg_len / sizeof(PMOP**);
9382 PMOP **pmp = (PMOP**) mg->mg_ptr;
9383 PMOP *const *const end = pmp + count;
9387 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9389 (*pmp)->op_pmflags &= ~PMf_USED;
9397 /* reset variables */
9399 if (!HvARRAY(stash))
9402 Zero(todo, 256, char);
9406 I32 i = (unsigned char)*s;
9410 max = (unsigned char)*s++;
9411 for ( ; i <= max; i++) {
9414 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9416 for (entry = HvARRAY(stash)[i];
9418 entry = HeNEXT(entry))
9423 if (!todo[(U8)*HeKEY(entry)])
9425 gv = MUTABLE_GV(HeVAL(entry));
9427 if (sv && !SvREADONLY(sv)) {
9428 SV_CHECK_THINKFIRST_COW_DROP(sv);
9429 if (!isGV(sv)) SvOK_off(sv);
9434 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9445 Using various gambits, try to get an IO from an SV: the IO slot if its a
9446 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9447 named after the PV if we're a string.
9449 'Get' magic is ignored on the sv passed in, but will be called on
9450 C<SvRV(sv)> if sv is an RV.
9456 Perl_sv_2io(pTHX_ SV *const sv)
9461 PERL_ARGS_ASSERT_SV_2IO;
9463 switch (SvTYPE(sv)) {
9465 io = MUTABLE_IO(sv);
9469 if (isGV_with_GP(sv)) {
9470 gv = MUTABLE_GV(sv);
9473 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9474 HEKfARG(GvNAME_HEK(gv)));
9480 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9482 SvGETMAGIC(SvRV(sv));
9483 return sv_2io(SvRV(sv));
9485 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9492 if (SvGMAGICAL(sv)) {
9493 newsv = sv_newmortal();
9494 sv_setsv_nomg(newsv, sv);
9496 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9506 Using various gambits, try to get a CV from an SV; in addition, try if
9507 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9508 The flags in C<lref> are passed to gv_fetchsv.
9514 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9519 PERL_ARGS_ASSERT_SV_2CV;
9526 switch (SvTYPE(sv)) {
9530 return MUTABLE_CV(sv);
9540 sv = amagic_deref_call(sv, to_cv_amg);
9543 if (SvTYPE(sv) == SVt_PVCV) {
9544 cv = MUTABLE_CV(sv);
9549 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9550 gv = MUTABLE_GV(sv);
9552 Perl_croak(aTHX_ "Not a subroutine reference");
9554 else if (isGV_with_GP(sv)) {
9555 gv = MUTABLE_GV(sv);
9558 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9565 /* Some flags to gv_fetchsv mean don't really create the GV */
9566 if (!isGV_with_GP(gv)) {
9571 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9572 /* XXX this is probably not what they think they're getting.
9573 * It has the same effect as "sub name;", i.e. just a forward
9584 Returns true if the SV has a true value by Perl's rules.
9585 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9586 instead use an in-line version.
9592 Perl_sv_true(pTHX_ SV *const sv)
9597 const XPV* const tXpv = (XPV*)SvANY(sv);
9599 (tXpv->xpv_cur > 1 ||
9600 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9607 return SvIVX(sv) != 0;
9610 return SvNVX(sv) != 0.0;
9612 return sv_2bool(sv);
9618 =for apidoc sv_pvn_force
9620 Get a sensible string out of the SV somehow.
9621 A private implementation of the C<SvPV_force> macro for compilers which
9622 can't cope with complex macro expressions. Always use the macro instead.
9624 =for apidoc sv_pvn_force_flags
9626 Get a sensible string out of the SV somehow.
9627 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9628 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9629 implemented in terms of this function.
9630 You normally want to use the various wrapper macros instead: see
9631 C<SvPV_force> and C<SvPV_force_nomg>
9637 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9639 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9641 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9642 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9643 sv_force_normal_flags(sv, 0);
9653 if (SvTYPE(sv) > SVt_PVLV
9654 || isGV_with_GP(sv))
9655 /* diag_listed_as: Can't coerce %s to %s in %s */
9656 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9658 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9665 if (SvTYPE(sv) < SVt_PV ||
9666 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9669 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9670 SvGROW(sv, len + 1);
9671 Move(s,SvPVX(sv),len,char);
9673 SvPVX(sv)[len] = '\0';
9676 SvPOK_on(sv); /* validate pointer */
9678 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9679 PTR2UV(sv),SvPVX_const(sv)));
9682 (void)SvPOK_only_UTF8(sv);
9683 return SvPVX_mutable(sv);
9687 =for apidoc sv_pvbyten_force
9689 The backend for the C<SvPVbytex_force> macro. Always use the macro
9696 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9698 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9700 sv_pvn_force(sv,lp);
9701 sv_utf8_downgrade(sv,0);
9707 =for apidoc sv_pvutf8n_force
9709 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9716 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9718 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9721 sv_utf8_upgrade_nomg(sv);
9727 =for apidoc sv_reftype
9729 Returns a string describing what the SV is a reference to.
9735 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9737 PERL_ARGS_ASSERT_SV_REFTYPE;
9738 if (ob && SvOBJECT(sv)) {
9739 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9742 /* WARNING - There is code, for instance in mg.c, that assumes that
9743 * the only reason that sv_reftype(sv,0) would return a string starting
9744 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9745 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9746 * this routine inside other subs, and it saves time.
9747 * Do not change this assumption without searching for "dodgy type check" in
9750 switch (SvTYPE(sv)) {
9765 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9766 /* tied lvalues should appear to be
9767 * scalars for backwards compatibility */
9768 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9769 ? "SCALAR" : "LVALUE");
9770 case SVt_PVAV: return "ARRAY";
9771 case SVt_PVHV: return "HASH";
9772 case SVt_PVCV: return "CODE";
9773 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9774 ? "GLOB" : "SCALAR");
9775 case SVt_PVFM: return "FORMAT";
9776 case SVt_PVIO: return "IO";
9777 case SVt_INVLIST: return "INVLIST";
9778 case SVt_REGEXP: return "REGEXP";
9779 default: return "UNKNOWN";
9787 Returns a SV describing what the SV passed in is a reference to.
9793 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9795 PERL_ARGS_ASSERT_SV_REF;
9798 dst = sv_newmortal();
9800 if (ob && SvOBJECT(sv)) {
9801 HvNAME_get(SvSTASH(sv))
9802 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9803 : sv_setpvn(dst, "__ANON__", 8);
9806 const char * reftype = sv_reftype(sv, 0);
9807 sv_setpv(dst, reftype);
9813 =for apidoc sv_isobject
9815 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9816 object. If the SV is not an RV, or if the object is not blessed, then this
9823 Perl_sv_isobject(pTHX_ SV *sv)
9839 Returns a boolean indicating whether the SV is blessed into the specified
9840 class. This does not check for subtypes; use C<sv_derived_from> to verify
9841 an inheritance relationship.
9847 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9851 PERL_ARGS_ASSERT_SV_ISA;
9861 hvname = HvNAME_get(SvSTASH(sv));
9865 return strEQ(hvname, name);
9871 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9872 RV then it will be upgraded to one. If C<classname> is non-null then the new
9873 SV will be blessed in the specified package. The new SV is returned and its
9874 reference count is 1. The reference count 1 is owned by C<rv>.
9880 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9884 PERL_ARGS_ASSERT_NEWSVRV;
9888 SV_CHECK_THINKFIRST_COW_DROP(rv);
9890 if (SvTYPE(rv) >= SVt_PVMG) {
9891 const U32 refcnt = SvREFCNT(rv);
9895 SvREFCNT(rv) = refcnt;
9897 sv_upgrade(rv, SVt_IV);
9898 } else if (SvROK(rv)) {
9899 SvREFCNT_dec(SvRV(rv));
9901 prepare_SV_for_RV(rv);
9909 HV* const stash = gv_stashpv(classname, GV_ADD);
9910 (void)sv_bless(rv, stash);
9916 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9918 SV * const lv = newSV_type(SVt_PVLV);
9919 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9921 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9922 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9923 LvSTARGOFF(lv) = ix;
9924 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9929 =for apidoc sv_setref_pv
9931 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9932 argument will be upgraded to an RV. That RV will be modified to point to
9933 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
9934 into the SV. The C<classname> argument indicates the package for the
9935 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9936 will have a reference count of 1, and the RV will be returned.
9938 Do not use with other Perl types such as HV, AV, SV, CV, because those
9939 objects will become corrupted by the pointer copy process.
9941 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
9947 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
9949 PERL_ARGS_ASSERT_SV_SETREF_PV;
9952 sv_setsv(rv, &PL_sv_undef);
9956 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
9961 =for apidoc sv_setref_iv
9963 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
9964 argument will be upgraded to an RV. That RV will be modified to point to
9965 the new SV. The C<classname> argument indicates the package for the
9966 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9967 will have a reference count of 1, and the RV will be returned.
9973 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
9975 PERL_ARGS_ASSERT_SV_SETREF_IV;
9977 sv_setiv(newSVrv(rv,classname), iv);
9982 =for apidoc sv_setref_uv
9984 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
9985 argument will be upgraded to an RV. That RV will be modified to point to
9986 the new SV. The C<classname> argument indicates the package for the
9987 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9988 will have a reference count of 1, and the RV will be returned.
9994 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
9996 PERL_ARGS_ASSERT_SV_SETREF_UV;
9998 sv_setuv(newSVrv(rv,classname), uv);
10003 =for apidoc sv_setref_nv
10005 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10006 argument will be upgraded to an RV. That RV will be modified to point to
10007 the new SV. The C<classname> argument indicates the package for the
10008 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10009 will have a reference count of 1, and the RV will be returned.
10015 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10017 PERL_ARGS_ASSERT_SV_SETREF_NV;
10019 sv_setnv(newSVrv(rv,classname), nv);
10024 =for apidoc sv_setref_pvn
10026 Copies a string into a new SV, optionally blessing the SV. The length of the
10027 string must be specified with C<n>. The C<rv> argument will be upgraded to
10028 an RV. That RV will be modified to point to the new SV. The C<classname>
10029 argument indicates the package for the blessing. Set C<classname> to
10030 C<NULL> to avoid the blessing. The new SV will have a reference count
10031 of 1, and the RV will be returned.
10033 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10039 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10040 const char *const pv, const STRLEN n)
10042 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10044 sv_setpvn(newSVrv(rv,classname), pv, n);
10049 =for apidoc sv_bless
10051 Blesses an SV into a specified package. The SV must be an RV. The package
10052 must be designated by its stash (see C<gv_stashpv()>). The reference count
10053 of the SV is unaffected.
10059 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10062 HV *oldstash = NULL;
10064 PERL_ARGS_ASSERT_SV_BLESS;
10068 Perl_croak(aTHX_ "Can't bless non-reference value");
10070 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY)) {
10071 if (SvREADONLY(tmpRef))
10072 Perl_croak_no_modify();
10073 if (SvOBJECT(tmpRef)) {
10074 oldstash = SvSTASH(tmpRef);
10077 SvOBJECT_on(tmpRef);
10078 SvUPGRADE(tmpRef, SVt_PVMG);
10079 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10080 SvREFCNT_dec(oldstash);
10082 if(SvSMAGICAL(tmpRef))
10083 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10091 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10092 * as it is after unglobbing it.
10095 PERL_STATIC_INLINE void
10096 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10100 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10102 PERL_ARGS_ASSERT_SV_UNGLOB;
10104 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10106 if (!(flags & SV_COW_DROP_PV))
10107 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10109 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10111 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10112 && HvNAME_get(stash))
10113 mro_method_changed_in(stash);
10114 gp_free(MUTABLE_GV(sv));
10117 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10118 GvSTASH(sv) = NULL;
10121 if (GvNAME_HEK(sv)) {
10122 unshare_hek(GvNAME_HEK(sv));
10124 isGV_with_GP_off(sv);
10126 if(SvTYPE(sv) == SVt_PVGV) {
10127 /* need to keep SvANY(sv) in the right arena */
10128 xpvmg = new_XPVMG();
10129 StructCopy(SvANY(sv), xpvmg, XPVMG);
10130 del_XPVGV(SvANY(sv));
10133 SvFLAGS(sv) &= ~SVTYPEMASK;
10134 SvFLAGS(sv) |= SVt_PVMG;
10137 /* Intentionally not calling any local SET magic, as this isn't so much a
10138 set operation as merely an internal storage change. */
10139 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10140 else sv_setsv_flags(sv, temp, 0);
10142 if ((const GV *)sv == PL_last_in_gv)
10143 PL_last_in_gv = NULL;
10144 else if ((const GV *)sv == PL_statgv)
10149 =for apidoc sv_unref_flags
10151 Unsets the RV status of the SV, and decrements the reference count of
10152 whatever was being referenced by the RV. This can almost be thought of
10153 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10154 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10155 (otherwise the decrementing is conditional on the reference count being
10156 different from one or the reference being a readonly SV).
10163 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10165 SV* const target = SvRV(ref);
10167 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10169 if (SvWEAKREF(ref)) {
10170 sv_del_backref(target, ref);
10171 SvWEAKREF_off(ref);
10172 SvRV_set(ref, NULL);
10175 SvRV_set(ref, NULL);
10177 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10178 assigned to as BEGIN {$a = \"Foo"} will fail. */
10179 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10180 SvREFCNT_dec_NN(target);
10181 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10182 sv_2mortal(target); /* Schedule for freeing later */
10186 =for apidoc sv_untaint
10188 Untaint an SV. Use C<SvTAINTED_off> instead.
10194 Perl_sv_untaint(pTHX_ SV *const sv)
10196 PERL_ARGS_ASSERT_SV_UNTAINT;
10197 PERL_UNUSED_CONTEXT;
10199 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10200 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10207 =for apidoc sv_tainted
10209 Test an SV for taintedness. Use C<SvTAINTED> instead.
10215 Perl_sv_tainted(pTHX_ SV *const sv)
10217 PERL_ARGS_ASSERT_SV_TAINTED;
10218 PERL_UNUSED_CONTEXT;
10220 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10221 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10222 if (mg && (mg->mg_len & 1) )
10229 =for apidoc sv_setpviv
10231 Copies an integer into the given SV, also updating its string value.
10232 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10238 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10240 char buf[TYPE_CHARS(UV)];
10242 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10244 PERL_ARGS_ASSERT_SV_SETPVIV;
10246 sv_setpvn(sv, ptr, ebuf - ptr);
10250 =for apidoc sv_setpviv_mg
10252 Like C<sv_setpviv>, but also handles 'set' magic.
10258 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10260 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10262 sv_setpviv(sv, iv);
10266 #if defined(PERL_IMPLICIT_CONTEXT)
10268 /* pTHX_ magic can't cope with varargs, so this is a no-context
10269 * version of the main function, (which may itself be aliased to us).
10270 * Don't access this version directly.
10274 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10279 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10281 va_start(args, pat);
10282 sv_vsetpvf(sv, pat, &args);
10286 /* pTHX_ magic can't cope with varargs, so this is a no-context
10287 * version of the main function, (which may itself be aliased to us).
10288 * Don't access this version directly.
10292 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10297 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10299 va_start(args, pat);
10300 sv_vsetpvf_mg(sv, pat, &args);
10306 =for apidoc sv_setpvf
10308 Works like C<sv_catpvf> but copies the text into the SV instead of
10309 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10315 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10319 PERL_ARGS_ASSERT_SV_SETPVF;
10321 va_start(args, pat);
10322 sv_vsetpvf(sv, pat, &args);
10327 =for apidoc sv_vsetpvf
10329 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10330 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10332 Usually used via its frontend C<sv_setpvf>.
10338 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10340 PERL_ARGS_ASSERT_SV_VSETPVF;
10342 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10346 =for apidoc sv_setpvf_mg
10348 Like C<sv_setpvf>, but also handles 'set' magic.
10354 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10358 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10360 va_start(args, pat);
10361 sv_vsetpvf_mg(sv, pat, &args);
10366 =for apidoc sv_vsetpvf_mg
10368 Like C<sv_vsetpvf>, but also handles 'set' magic.
10370 Usually used via its frontend C<sv_setpvf_mg>.
10376 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10378 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10380 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10384 #if defined(PERL_IMPLICIT_CONTEXT)
10386 /* pTHX_ magic can't cope with varargs, so this is a no-context
10387 * version of the main function, (which may itself be aliased to us).
10388 * Don't access this version directly.
10392 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10397 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10399 va_start(args, pat);
10400 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10404 /* pTHX_ magic can't cope with varargs, so this is a no-context
10405 * version of the main function, (which may itself be aliased to us).
10406 * Don't access this version directly.
10410 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10415 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10417 va_start(args, pat);
10418 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10425 =for apidoc sv_catpvf
10427 Processes its arguments like C<sprintf> and appends the formatted
10428 output to an SV. If the appended data contains "wide" characters
10429 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10430 and characters >255 formatted with %c), the original SV might get
10431 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10432 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10433 valid UTF-8; if the original SV was bytes, the pattern should be too.
10438 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10442 PERL_ARGS_ASSERT_SV_CATPVF;
10444 va_start(args, pat);
10445 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10450 =for apidoc sv_vcatpvf
10452 Processes its arguments like C<vsprintf> and appends the formatted output
10453 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10455 Usually used via its frontend C<sv_catpvf>.
10461 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10463 PERL_ARGS_ASSERT_SV_VCATPVF;
10465 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10469 =for apidoc sv_catpvf_mg
10471 Like C<sv_catpvf>, but also handles 'set' magic.
10477 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10481 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10483 va_start(args, pat);
10484 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10490 =for apidoc sv_vcatpvf_mg
10492 Like C<sv_vcatpvf>, but also handles 'set' magic.
10494 Usually used via its frontend C<sv_catpvf_mg>.
10500 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10502 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10504 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10509 =for apidoc sv_vsetpvfn
10511 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10514 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10520 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10521 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10523 PERL_ARGS_ASSERT_SV_VSETPVFN;
10526 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10531 * Warn of missing argument to sprintf, and then return a defined value
10532 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10535 S_vcatpvfn_missing_argument(pTHX) {
10536 if (ckWARN(WARN_MISSING)) {
10537 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10538 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10545 S_expect_number(pTHX_ char **const pattern)
10549 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10551 switch (**pattern) {
10552 case '1': case '2': case '3':
10553 case '4': case '5': case '6':
10554 case '7': case '8': case '9':
10555 var = *(*pattern)++ - '0';
10556 while (isDIGIT(**pattern)) {
10557 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10559 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10567 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10569 const int neg = nv < 0;
10572 PERL_ARGS_ASSERT_F0CONVERT;
10574 if (Perl_isinfnan(nv)) {
10575 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len);
10585 if (uv & 1 && uv == nv)
10586 uv--; /* Round to even */
10588 const unsigned dig = uv % 10;
10590 } while (uv /= 10);
10601 =for apidoc sv_vcatpvfn
10603 =for apidoc sv_vcatpvfn_flags
10605 Processes its arguments like C<vsprintf> and appends the formatted output
10606 to an SV. Uses an array of SVs if the C style variable argument list is
10607 missing (NULL). When running with taint checks enabled, indicates via
10608 C<maybe_tainted> if results are untrustworthy (often due to the use of
10611 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10613 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10618 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10619 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10620 vec_utf8 = DO_UTF8(vecsv);
10622 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10625 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10626 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10628 PERL_ARGS_ASSERT_SV_VCATPVFN;
10630 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10633 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10634 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10635 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10636 # define LONGDOUBLE_LITTLE_ENDIAN
10639 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN || \
10640 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN || \
10641 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10642 # define LONGDOUBLE_BIG_ENDIAN
10645 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10646 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10647 # define LONGDOUBLE_X86_80_BIT
10650 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
10651 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10652 # define LONGDOUBLE_DOUBLEDOUBLE
10653 # define DOUBLEDOUBLE_MAXBITS 1028
10656 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10657 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10659 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10660 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10662 # define VHEX_SIZE (1+128/4)
10665 /* If we do not have a known long double format, (including not using
10666 * long doubles, or long doubles being equal to doubles) then we will
10667 * fall back to the ldexp/frexp route, with which we can retrieve at
10668 * most as many bits as our widest unsigned integer type is. We try
10669 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10671 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10672 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10674 #if defined(HAS_QUAD) && defined(Uquad_t)
10675 # define MANTISSATYPE Uquad_t
10676 # define MANTISSASIZE 8
10678 # define MANTISSATYPE UV
10679 # define MANTISSASIZE UVSIZE
10682 /* We make here the wild assumption that the endianness of doubles
10683 * is similar to the endianness of integers, and that there is no
10684 * middle-endianness. This may come back to haunt us (the rumor
10685 * has it that ARM can be quite haunted). */
10686 #if BYTEORDER == 0x12345678 || BYTEORDER == 0x1234 || \
10687 defined(DOUBLEKIND_LITTLE_ENDIAN)
10688 # define HEXTRACT_LITTLE_ENDIAN
10690 # define HEXTRACT_BIG_ENDIAN
10693 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10694 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10695 * are being extracted from (either directly from the long double in-memory
10696 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10697 * is used to update the exponent. vhex is the pointer to the beginning
10698 * of the output buffer (of VHEX_SIZE).
10700 * The tricky part is that S_hextract() needs to be called twice:
10701 * the first time with vend as NULL, and the second time with vend as
10702 * the pointer returned by the first call. What happens is that on
10703 * the first round the output size is computed, and the intended
10704 * extraction sanity checked. On the second round the actual output
10705 * (the extraction of the hexadecimal values) takes place.
10706 * Sanity failures cause fatal failures during both rounds. */
10708 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10712 int ixmin = 0, ixmax = 0;
10714 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10715 * and elsewhere. */
10717 /* These macros are just to reduce typos, they have multiple
10718 * repetitions below, but usually only one (or sometimes two)
10719 * of them is really being used. */
10720 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10721 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10722 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10723 #define HEXTRACT_OUTPUT(ix) \
10725 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10727 #define HEXTRACT_COUNT(ix, c) \
10729 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10731 #define HEXTRACT_BYTE(ix) \
10733 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10735 #define HEXTRACT_LO_NYBBLE(ix) \
10737 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10739 # define HEXTRACT_IMPLICIT_BIT(nv) \
10741 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10744 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10745 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/8)
10747 # define HEXTRACTSIZE NVSIZE
10750 const U8* nvp = (const U8*)(&nv);
10751 const U8* vmaxend = vhex + 2 * HEXTRACTSIZE + 1;
10752 (void)Perl_frexp(PERL_ABS(nv), exponent);
10753 if (vend && (vend <= vhex || vend > vmaxend))
10754 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10756 /* First check if using long doubles. */
10757 #if NVSIZE > DOUBLESIZE
10758 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10759 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10760 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10761 /* The bytes 13..0 are the mantissa/fraction,
10762 * the 15,14 are the sign+exponent. */
10763 HEXTRACT_IMPLICIT_BIT(nv);
10764 for (ix = 13; ix >= 0; ix--) {
10767 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10768 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10769 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10770 /* The bytes 2..15 are the mantissa/fraction,
10771 * the 0,1 are the sign+exponent. */
10772 HEXTRACT_IMPLICIT_BIT(nv);
10773 for (ix = 2; ix <= 15; ix++) {
10776 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10777 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10778 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10779 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10780 * meaning that 2 or 6 bytes are empty padding. */
10781 /* The bytes 7..0 are the mantissa/fraction */
10783 /* Intentionally NO HEXTRACT_IMPLICIT_BIT here. */
10784 for (ix = 7; ix >= 0; ix--) {
10787 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10788 /* Does this format ever happen? (Wikipedia says the Motorola
10789 * 6888x math coprocessors used format _like_ this but padded
10790 * to 96 bits with 16 unused bits between the exponent and the
10793 /* Intentionally NO HEXTRACT_IMPLICIT_BIT here. */
10794 for (ix = 0; ix < 8; ix++) {
10797 # elif defined(LONGDOUBLE_DOUBLEDOUBLE)
10798 /* Double-double format: two doubles next to each other.
10799 * The first double is the high-order one, exactly like
10800 * it would be for a "lone" double. The second double
10801 * is shifted down using the exponent so that that there
10802 * are no common bits. The tricky part is that the value
10803 * of the double-double is the SUM of the two doubles and
10804 * the second one can be also NEGATIVE.
10806 * Because of this tricky construction the bytewise extraction we
10807 * use for the other long double formats doesn't work, we must
10808 * extract the values bit by bit.
10810 * The little-endian double-double is used .. somewhere?
10812 * The big endian double-double is used in e.g. PPC/Power (AIX)
10815 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
10816 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
10817 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
10820 if (nv == (NV)0.0) {
10828 NV d = nv < 0 ? -nv : nv;
10830 U8 ha = 0x0; /* hexvalue accumulator */
10831 U8 hd = 0x8; /* hexvalue digit */
10833 /* Shift d and e (and update exponent) so that e <= d < 2*e,
10834 * this is essentially manual frexp(). Multiplying by 0.5 and
10835 * doubling should be lossless in binary floating point. */
10845 while (d >= e + e) {
10849 /* Now e <= d < 2*e */
10851 /* First extract the leading hexdigit (the implicit bit). */
10867 /* Then extract the remaining hexdigits. */
10868 while (d > (NV)0.0) {
10874 /* Output or count in groups of four bits,
10875 * that is, when the hexdigit is down to one. */
10880 /* Reset the hexvalue. */
10889 /* Flush possible pending hexvalue. */
10899 "Hexadecimal float: unsupported long double format");
10902 /* Using normal doubles, not long doubles.
10904 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
10905 * bytes, since we might need to handle printf precision, and
10906 * also need to insert the radix. */
10907 HEXTRACT_IMPLICIT_BIT(nv);
10908 # ifdef HEXTRACT_LITTLE_ENDIAN
10909 HEXTRACT_LO_NYBBLE(6);
10910 for (ix = 5; ix >= 0; ix--) {
10914 HEXTRACT_LO_NYBBLE(1);
10915 for (ix = 2; ix < HEXTRACTSIZE; ix++) {
10920 /* Croak for various reasons: if the output pointer escaped the
10921 * output buffer, if the extraction index escaped the extraction
10922 * buffer, or if the ending output pointer didn't match the
10923 * previously computed value. */
10924 if (v <= vhex || v - vhex >= VHEX_SIZE ||
10925 /* For double-double the ixmin and ixmax stay at zero,
10926 * which is convenient since the HEXTRACTSIZE is tricky
10927 * for double-double. */
10928 ixmin < 0 || ixmax >= HEXTRACTSIZE ||
10929 (vend && v != vend))
10930 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10935 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10936 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
10941 const char *patend;
10944 static const char nullstr[] = "(null)";
10946 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
10947 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
10949 /* Times 4: a decimal digit takes more than 3 binary digits.
10950 * NV_DIG: mantissa takes than many decimal digits.
10951 * Plus 32: Playing safe. */
10952 char ebuf[IV_DIG * 4 + NV_DIG + 32];
10953 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
10954 bool hexfp = FALSE; /* hexadecimal floating point? */
10956 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
10958 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
10959 PERL_UNUSED_ARG(maybe_tainted);
10961 if (flags & SV_GMAGIC)
10964 /* no matter what, this is a string now */
10965 (void)SvPV_force_nomg(sv, origlen);
10967 /* special-case "", "%s", and "%-p" (SVf - see below) */
10969 if (svmax && ckWARN(WARN_REDUNDANT))
10970 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10971 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10974 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
10975 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10976 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10977 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10980 const char * const s = va_arg(*args, char*);
10981 sv_catpv_nomg(sv, s ? s : nullstr);
10983 else if (svix < svmax) {
10984 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
10985 SvGETMAGIC(*svargs);
10986 sv_catsv_nomg(sv, *svargs);
10989 S_vcatpvfn_missing_argument(aTHX);
10992 if (args && patlen == 3 && pat[0] == '%' &&
10993 pat[1] == '-' && pat[2] == 'p') {
10994 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10995 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10996 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10997 argsv = MUTABLE_SV(va_arg(*args, void*));
10998 sv_catsv_nomg(sv, argsv);
11002 #ifndef USE_LONG_DOUBLE
11003 /* special-case "%.<number>[gf]" */
11004 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11005 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11006 unsigned digits = 0;
11010 while (*pp >= '0' && *pp <= '9')
11011 digits = 10 * digits + (*pp++ - '0');
11013 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11014 format the first argument and WARN_REDUNDANT if svmax > 1?
11015 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11016 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11017 const NV nv = SvNV(*svargs);
11018 if (LIKELY(!Perl_isinfnan(nv))) {
11020 /* Add check for digits != 0 because it seems that some
11021 gconverts are buggy in this case, and we don't yet have
11022 a Configure test for this. */
11023 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11024 /* 0, point, slack */
11025 STORE_LC_NUMERIC_SET_TO_NEEDED();
11026 PERL_UNUSED_RESULT(Gconvert(nv, (int)digits, 0, ebuf));
11027 sv_catpv_nomg(sv, ebuf);
11028 if (*ebuf) /* May return an empty string for digits==0 */
11031 } else if (!digits) {
11034 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11035 sv_catpvn_nomg(sv, p, l);
11042 #endif /* !USE_LONG_DOUBLE */
11044 if (!args && svix < svmax && DO_UTF8(*svargs))
11047 patend = (char*)pat + patlen;
11048 for (p = (char*)pat; p < patend; p = q) {
11051 bool vectorize = FALSE;
11052 bool vectorarg = FALSE;
11053 bool vec_utf8 = FALSE;
11059 bool has_precis = FALSE;
11061 const I32 osvix = svix;
11062 bool is_utf8 = FALSE; /* is this item utf8? */
11063 #ifdef HAS_LDBL_SPRINTF_BUG
11064 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11065 with sfio - Allen <allens@cpan.org> */
11066 bool fix_ldbl_sprintf_bug = FALSE;
11070 U8 utf8buf[UTF8_MAXBYTES+1];
11071 STRLEN esignlen = 0;
11073 const char *eptr = NULL;
11074 const char *fmtstart;
11077 const U8 *vecstr = NULL;
11084 /* We need a long double target in case HAS_LONG_DOUBLE,
11085 * even without USE_LONG_DOUBLE, so that we can printf with
11086 * long double formats, even without NV being long double.
11087 * But we call the target 'fv' instead of 'nv', since most of
11088 * the time it is not (most compilers these days recognize
11089 * "long double", even if only as a synonym for "double").
11091 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && defined(PERL_PRIgldbl)
11093 # define FV_ISFINITE(x) Perl_isfinitel(x)
11094 # define FV_GF PERL_PRIgldbl
11097 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11098 # define FV_GF NVgf
11103 const char *dotstr = ".";
11104 STRLEN dotstrlen = 1;
11105 I32 efix = 0; /* explicit format parameter index */
11106 I32 ewix = 0; /* explicit width index */
11107 I32 epix = 0; /* explicit precision index */
11108 I32 evix = 0; /* explicit vector index */
11109 bool asterisk = FALSE;
11110 bool infnan = FALSE;
11112 /* echo everything up to the next format specification */
11113 for (q = p; q < patend && *q != '%'; ++q) ;
11115 if (has_utf8 && !pat_utf8)
11116 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11118 sv_catpvn_nomg(sv, p, q - p);
11127 We allow format specification elements in this order:
11128 \d+\$ explicit format parameter index
11130 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11131 0 flag (as above): repeated to allow "v02"
11132 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11133 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11135 [%bcdefginopsuxDFOUX] format (mandatory)
11140 As of perl5.9.3, printf format checking is on by default.
11141 Internally, perl uses %p formats to provide an escape to
11142 some extended formatting. This block deals with those
11143 extensions: if it does not match, (char*)q is reset and
11144 the normal format processing code is used.
11146 Currently defined extensions are:
11147 %p include pointer address (standard)
11148 %-p (SVf) include an SV (previously %_)
11149 %-<num>p include an SV with precision <num>
11151 %3p include a HEK with precision of 256
11152 %4p char* preceded by utf8 flag and length
11153 %<num>p (where num is 1 or > 4) reserved for future
11156 Robin Barker 2005-07-14 (but modified since)
11158 %1p (VDf) removed. RMB 2007-10-19
11165 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11166 /* The argument has already gone through cBOOL, so the cast
11168 is_utf8 = (bool)va_arg(*args, int);
11169 elen = va_arg(*args, UV);
11170 eptr = va_arg(*args, char *);
11171 q += sizeof(UTF8f)-1;
11174 n = expect_number(&q);
11176 if (sv) { /* SVf */
11181 argsv = MUTABLE_SV(va_arg(*args, void*));
11182 eptr = SvPV_const(argsv, elen);
11183 if (DO_UTF8(argsv))
11187 else if (n==2 || n==3) { /* HEKf */
11188 HEK * const hek = va_arg(*args, HEK *);
11189 eptr = HEK_KEY(hek);
11190 elen = HEK_LEN(hek);
11191 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11192 if (n==3) precis = 256, has_precis = TRUE;
11196 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11197 "internal %%<num>p might conflict with future printf extensions");
11203 if ( (width = expect_number(&q)) ) {
11207 if (!no_redundant_warning)
11208 /* I've forgotten if it's a better
11209 micro-optimization to always set this or to
11210 only set it if it's unset */
11211 no_redundant_warning = TRUE;
11223 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11252 if ( (ewix = expect_number(&q)) )
11261 if ((vectorarg = asterisk)) {
11274 width = expect_number(&q);
11277 if (vectorize && vectorarg) {
11278 /* vectorizing, but not with the default "." */
11280 vecsv = va_arg(*args, SV*);
11282 vecsv = (evix > 0 && evix <= svmax)
11283 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11285 vecsv = svix < svmax
11286 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11288 dotstr = SvPV_const(vecsv, dotstrlen);
11289 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11290 bad with tied or overloaded values that return UTF8. */
11291 if (DO_UTF8(vecsv))
11293 else if (has_utf8) {
11294 vecsv = sv_mortalcopy(vecsv);
11295 sv_utf8_upgrade(vecsv);
11296 dotstr = SvPV_const(vecsv, dotstrlen);
11303 i = va_arg(*args, int);
11305 i = (ewix ? ewix <= svmax : svix < svmax) ?
11306 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11308 width = (i < 0) ? -i : i;
11318 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11320 /* XXX: todo, support specified precision parameter */
11324 i = va_arg(*args, int);
11326 i = (ewix ? ewix <= svmax : svix < svmax)
11327 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11329 has_precis = !(i < 0);
11333 while (isDIGIT(*q))
11334 precis = precis * 10 + (*q++ - '0');
11343 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11344 vecsv = svargs[efix ? efix-1 : svix++];
11345 vecstr = (U8*)SvPV_const(vecsv,veclen);
11346 vec_utf8 = DO_UTF8(vecsv);
11348 /* if this is a version object, we need to convert
11349 * back into v-string notation and then let the
11350 * vectorize happen normally
11352 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11353 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11354 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11355 "vector argument not supported with alpha versions");
11358 vecsv = sv_newmortal();
11359 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11361 vecstr = (U8*)SvPV_const(vecsv, veclen);
11362 vec_utf8 = DO_UTF8(vecsv);
11376 case 'I': /* Ix, I32x, and I64x */
11377 # ifdef USE_64_BIT_INT
11378 if (q[1] == '6' && q[2] == '4') {
11384 if (q[1] == '3' && q[2] == '2') {
11388 # ifdef USE_64_BIT_INT
11394 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11406 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11407 if (*q == 'l') { /* lld, llf */
11416 if (*++q == 'h') { /* hhd, hhu */
11445 if (!vectorize && !args) {
11447 const I32 i = efix-1;
11448 argsv = (i >= 0 && i < svmax)
11449 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11451 argsv = (svix >= 0 && svix < svmax)
11452 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11456 if (argsv && SvNOK(argsv)) {
11457 /* XXX va_arg(*args) case? */
11458 infnan = Perl_isinfnan(SvNV(argsv));
11461 switch (c = *q++) {
11468 uv = (args) ? va_arg(*args, int) :
11469 infnan ? UNICODE_REPLACEMENT : SvIV(argsv);
11471 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11473 eptr = (char*)utf8buf;
11474 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11488 eptr = va_arg(*args, char*);
11490 elen = strlen(eptr);
11492 eptr = (char *)nullstr;
11493 elen = sizeof nullstr - 1;
11497 eptr = SvPV_const(argsv, elen);
11498 if (DO_UTF8(argsv)) {
11499 STRLEN old_precis = precis;
11500 if (has_precis && precis < elen) {
11501 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11502 STRLEN p = precis > ulen ? ulen : precis;
11503 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11504 /* sticks at end */
11506 if (width) { /* fudge width (can't fudge elen) */
11507 if (has_precis && precis < elen)
11508 width += precis - old_precis;
11511 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11518 if (has_precis && precis < elen)
11527 goto floating_point;
11529 if (alt || vectorize)
11531 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11546 goto floating_point;
11553 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11562 esignbuf[esignlen++] = plus;
11566 case 'c': iv = (char)va_arg(*args, int); break;
11567 case 'h': iv = (short)va_arg(*args, int); break;
11568 case 'l': iv = va_arg(*args, long); break;
11569 case 'V': iv = va_arg(*args, IV); break;
11570 case 'z': iv = va_arg(*args, SSize_t); break;
11571 #ifdef HAS_PTRDIFF_T
11572 case 't': iv = va_arg(*args, ptrdiff_t); break;
11574 default: iv = va_arg(*args, int); break;
11576 case 'j': iv = va_arg(*args, intmax_t); break;
11580 iv = va_arg(*args, Quad_t); break;
11587 IV tiv = SvIV(argsv); /* work around GCC bug #13488 */
11589 case 'c': iv = (char)tiv; break;
11590 case 'h': iv = (short)tiv; break;
11591 case 'l': iv = (long)tiv; break;
11593 default: iv = tiv; break;
11596 iv = (Quad_t)tiv; break;
11602 if ( !vectorize ) /* we already set uv above */
11607 esignbuf[esignlen++] = plus;
11611 esignbuf[esignlen++] = '-';
11651 goto floating_point;
11659 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11670 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11671 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11672 case 'l': uv = va_arg(*args, unsigned long); break;
11673 case 'V': uv = va_arg(*args, UV); break;
11674 case 'z': uv = va_arg(*args, Size_t); break;
11675 #ifdef HAS_PTRDIFF_T
11676 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11679 case 'j': uv = va_arg(*args, uintmax_t); break;
11681 default: uv = va_arg(*args, unsigned); break;
11684 uv = va_arg(*args, Uquad_t); break;
11691 UV tuv = SvUV(argsv); /* work around GCC bug #13488 */
11693 case 'c': uv = (unsigned char)tuv; break;
11694 case 'h': uv = (unsigned short)tuv; break;
11695 case 'l': uv = (unsigned long)tuv; break;
11697 default: uv = tuv; break;
11700 uv = (Uquad_t)tuv; break;
11709 char *ptr = ebuf + sizeof ebuf;
11710 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11716 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11720 } while (uv >>= 4);
11722 esignbuf[esignlen++] = '0';
11723 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11729 *--ptr = '0' + dig;
11730 } while (uv >>= 3);
11731 if (alt && *ptr != '0')
11737 *--ptr = '0' + dig;
11738 } while (uv >>= 1);
11740 esignbuf[esignlen++] = '0';
11741 esignbuf[esignlen++] = c;
11744 default: /* it had better be ten or less */
11747 *--ptr = '0' + dig;
11748 } while (uv /= base);
11751 elen = (ebuf + sizeof ebuf) - ptr;
11755 zeros = precis - elen;
11756 else if (precis == 0 && elen == 1 && *eptr == '0'
11757 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11760 /* a precision nullifies the 0 flag. */
11767 /* FLOATING POINT */
11772 c = 'f'; /* maybe %F isn't supported here */
11774 case 'e': case 'E':
11776 case 'g': case 'G':
11777 case 'a': case 'A':
11781 /* This is evil, but floating point is even more evil */
11783 /* for SV-style calling, we can only get NV
11784 for C-style calling, we assume %f is double;
11785 for simplicity we allow any of %Lf, %llf, %qf for long double
11789 #if defined(USE_LONG_DOUBLE)
11793 /* [perl #20339] - we should accept and ignore %lf rather than die */
11797 #if defined(USE_LONG_DOUBLE)
11798 intsize = args ? 0 : 'q';
11802 #if defined(HAS_LONG_DOUBLE)
11815 /* Now we need (long double) if intsize == 'q', else (double). */
11817 /* Note: do not pull NVs off the va_list with va_arg()
11818 * (pull doubles instead) because if you have a build
11819 * with long doubles, you would always be pulling long
11820 * doubles, which would badly break anyone using only
11821 * doubles (i.e. the majority of builds). In other
11822 * words, you cannot mix doubles and long doubles.
11823 * The only case where you can pull off long doubles
11824 * is when the format specifier explicitly asks so with
11826 #if LONG_DOUBLESIZE > DOUBLESIZE
11827 fv = intsize == 'q' ?
11828 va_arg(*args, long double) : va_arg(*args, double);
11830 fv = va_arg(*args, double);
11837 /* frexp() (or frexpl) has some unspecified behaviour for
11838 * nan/inf/-inf, so let's avoid calling that on non-finites. */
11839 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
11841 (void)Perl_frexp((NV)fv, &i);
11842 if (i == PERL_INT_MIN)
11843 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
11844 /* Do not set hexfp earlier since we want to printf
11845 * Inf/NaN for Inf/NaN, not their hexfp. */
11846 hexfp = isALPHA_FOLD_EQ(c, 'a');
11847 if (UNLIKELY(hexfp)) {
11848 /* This seriously overshoots in most cases, but
11849 * better the undershooting. Firstly, all bytes
11850 * of the NV are not mantissa, some of them are
11851 * exponent. Secondly, for the reasonably common
11852 * long doubles case, the "80-bit extended", two
11853 * or six bytes of the NV are unused. */
11855 (fv < 0) ? 1 : 0 + /* possible unary minus */
11857 1 + /* the very unlikely carry */
11860 2 * NVSIZE + /* 2 hexdigits for each byte */
11862 6 + /* exponent: sign, plus up to 16383 (quad fp) */
11864 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11865 /* However, for the "double double", we need more.
11866 * Since each double has their own exponent, the
11867 * doubles may float (haha) rather far from each
11868 * other, and the number of required bits is much
11869 * larger, up to total of 1028 bits. (NOTE: this
11870 * is not actually implemented properly yet,
11871 * we are using just the first double, see
11872 * S_hextract() for details. But let's prepare
11873 * for the future.) */
11875 /* 2 hexdigits for each byte. */
11876 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
11877 /* the size for the exponent already added */
11879 #ifdef USE_LOCALE_NUMERIC
11880 STORE_LC_NUMERIC_SET_TO_NEEDED();
11881 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
11882 need += SvLEN(PL_numeric_radix_sv);
11883 RESTORE_LC_NUMERIC();
11887 need = BIT_DIGITS(i);
11888 } /* if i < 0, the number of digits is hard to predict. */
11890 need += has_precis ? precis : 6; /* known default */
11895 #ifdef HAS_LDBL_SPRINTF_BUG
11896 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11897 with sfio - Allen <allens@cpan.org> */
11900 # define MY_DBL_MAX DBL_MAX
11901 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11902 # if DOUBLESIZE >= 8
11903 # define MY_DBL_MAX 1.7976931348623157E+308L
11905 # define MY_DBL_MAX 3.40282347E+38L
11909 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
11910 # define MY_DBL_MAX_BUG 1L
11912 # define MY_DBL_MAX_BUG MY_DBL_MAX
11916 # define MY_DBL_MIN DBL_MIN
11917 # else /* XXX guessing! -Allen */
11918 # if DOUBLESIZE >= 8
11919 # define MY_DBL_MIN 2.2250738585072014E-308L
11921 # define MY_DBL_MIN 1.17549435E-38L
11925 if ((intsize == 'q') && (c == 'f') &&
11926 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
11927 (need < DBL_DIG)) {
11928 /* it's going to be short enough that
11929 * long double precision is not needed */
11931 if ((fv <= 0L) && (fv >= -0L))
11932 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
11934 /* would use Perl_fp_class as a double-check but not
11935 * functional on IRIX - see perl.h comments */
11937 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
11938 /* It's within the range that a double can represent */
11939 #if defined(DBL_MAX) && !defined(DBL_MIN)
11940 if ((fv >= ((long double)1/DBL_MAX)) ||
11941 (fv <= (-(long double)1/DBL_MAX)))
11943 fix_ldbl_sprintf_bug = TRUE;
11946 if (fix_ldbl_sprintf_bug == TRUE) {
11956 # undef MY_DBL_MAX_BUG
11959 #endif /* HAS_LDBL_SPRINTF_BUG */
11961 need += 20; /* fudge factor */
11962 if (PL_efloatsize < need) {
11963 Safefree(PL_efloatbuf);
11964 PL_efloatsize = need + 20; /* more fudge */
11965 Newx(PL_efloatbuf, PL_efloatsize, char);
11966 PL_efloatbuf[0] = '\0';
11969 if ( !(width || left || plus || alt) && fill != '0'
11970 && has_precis && intsize != 'q' /* Shortcuts */
11971 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
11972 /* See earlier comment about buggy Gconvert when digits,
11974 if ( c == 'g' && precis ) {
11975 STORE_LC_NUMERIC_SET_TO_NEEDED();
11976 PERL_UNUSED_RESULT(Gconvert((NV)fv, (int)precis, 0, PL_efloatbuf));
11977 /* May return an empty string for digits==0 */
11978 if (*PL_efloatbuf) {
11979 elen = strlen(PL_efloatbuf);
11980 goto float_converted;
11982 } else if ( c == 'f' && !precis ) {
11983 if ((eptr = F0convert(fv, ebuf + sizeof ebuf, &elen)))
11988 if (UNLIKELY(hexfp)) {
11989 /* Hexadecimal floating point. */
11990 char* p = PL_efloatbuf;
11991 U8 vhex[VHEX_SIZE];
11992 U8* v = vhex; /* working pointer to vhex */
11993 U8* vend; /* pointer to one beyond last digit of vhex */
11994 U8* vfnz = NULL; /* first non-zero */
11995 const bool lower = (c == 'a');
11996 /* At output the values of vhex (up to vend) will
11997 * be mapped through the xdig to get the actual
11998 * human-readable xdigits. */
11999 const char* xdig = PL_hexdigit;
12000 int zerotail = 0; /* how many extra zeros to append */
12001 int exponent = 0; /* exponent of the floating point input */
12003 /* XXX: denormals, NaN, Inf.
12005 * For example with denormals, (assuming the vanilla
12006 * 64-bit double): the exponent is zero. 1xp-1074 is
12007 * the smallest denormal and the smallest double, it
12008 * should be output as 0x0.0000000000001p-1022 to
12009 * match its internal structure. */
12011 /* Note: fv can be (and often is) long double.
12012 * Here it is explicitly cast to NV. */
12013 vend = S_hextract(aTHX_ (NV)fv, &exponent, vhex, NULL);
12014 S_hextract(aTHX_ (NV)fv, &exponent, vhex, vend);
12016 #if NVSIZE > DOUBLESIZE
12017 # ifdef LONGDOUBLE_X86_80_BIT
12034 xdig += 16; /* Use uppercase hex. */
12037 /* Find the first non-zero xdigit. */
12038 for (v = vhex; v < vend; v++) {
12046 U8* vlnz = NULL; /* The last non-zero. */
12048 /* Find the last non-zero xdigit. */
12049 for (v = vend - 1; v >= vhex; v--) {
12056 #if NVSIZE == DOUBLESIZE
12062 v = vhex + precis + 1;
12064 /* Round away from zero: if the tail
12065 * beyond the precis xdigits is equal to
12066 * or greater than 0x8000... */
12067 bool round = *v > 0x8;
12068 if (!round && *v == 0x8) {
12069 for (v++; v < vend; v++) {
12077 for (v = vhex + precis; v >= vhex; v--) {
12084 /* If the carry goes all the way to
12085 * the front, we need to output
12086 * a single '1'. This goes against
12087 * the "xdigit and then radix"
12088 * but since this is "cannot happen"
12089 * category, that is probably good. */
12094 /* The new effective "last non zero". */
12095 vlnz = vhex + precis;
12098 zerotail = precis - (vlnz - vhex);
12105 /* The radix is always output after the first
12106 * non-zero xdigit, or if alt. */
12107 if (vfnz < vlnz || alt) {
12108 #ifndef USE_LOCALE_NUMERIC
12111 STORE_LC_NUMERIC_SET_TO_NEEDED();
12112 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12114 const char* r = SvPV(PL_numeric_radix_sv, n);
12115 Copy(r, p, n, char);
12121 RESTORE_LC_NUMERIC();
12136 elen = p - PL_efloatbuf;
12137 elen += my_snprintf(p, PL_efloatsize - elen,
12138 "%c%+d", lower ? 'p' : 'P',
12141 if (elen < width) {
12143 /* Pad the back with spaces. */
12144 memset(PL_efloatbuf + elen, ' ', width - elen);
12146 else if (fill == '0') {
12147 /* Insert the zeros between the "0x" and
12148 * the digits, otherwise we end up with
12150 STRLEN nzero = width - elen;
12151 char* zerox = PL_efloatbuf + 2;
12152 Move(zerox, zerox + nzero, elen - 2, char);
12153 memset(zerox, fill, nzero);
12156 /* Move it to the right. */
12157 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12159 /* Pad the front with spaces. */
12160 memset(PL_efloatbuf, ' ', width - elen);
12166 elen = S_infnan_2pv(fv, PL_efloatbuf, PL_efloatsize);
12169 char *ptr = ebuf + sizeof ebuf;
12172 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12173 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12174 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12175 * not USE_LONG_DOUBLE and NVff. In other words,
12176 * this needs to work without USE_LONG_DOUBLE. */
12177 if (intsize == 'q') {
12178 /* Copy the one or more characters in a long double
12179 * format before the 'base' ([efgEFG]) character to
12180 * the format string. */
12181 static char const ldblf[] = PERL_PRIfldbl;
12182 char const *p = ldblf + sizeof(ldblf) - 3;
12183 while (p >= ldblf) { *--ptr = *p--; }
12188 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12193 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12205 /* No taint. Otherwise we are in the strange situation
12206 * where printf() taints but print($float) doesn't.
12209 STORE_LC_NUMERIC_SET_TO_NEEDED();
12211 /* hopefully the above makes ptr a very constrained format
12212 * that is safe to use, even though it's not literal */
12213 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12214 #if defined(HAS_LONG_DOUBLE)
12215 elen = ((intsize == 'q')
12216 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12217 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12219 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12225 eptr = PL_efloatbuf;
12226 assert((IV)elen > 0); /* here zero elen is bad */
12228 #ifdef USE_LOCALE_NUMERIC
12229 /* If the decimal point character in the string is UTF-8, make the
12231 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12232 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12245 i = SvCUR(sv) - origlen;
12248 case 'c': *(va_arg(*args, char*)) = i; break;
12249 case 'h': *(va_arg(*args, short*)) = i; break;
12250 default: *(va_arg(*args, int*)) = i; break;
12251 case 'l': *(va_arg(*args, long*)) = i; break;
12252 case 'V': *(va_arg(*args, IV*)) = i; break;
12253 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12254 #ifdef HAS_PTRDIFF_T
12255 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12258 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12262 *(va_arg(*args, Quad_t*)) = i; break;
12269 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12270 continue; /* not "break" */
12277 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12278 && ckWARN(WARN_PRINTF))
12280 SV * const msg = sv_newmortal();
12281 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12282 (PL_op->op_type == OP_PRTF) ? "" : "s");
12283 if (fmtstart < patend) {
12284 const char * const fmtend = q < patend ? q : patend;
12286 sv_catpvs(msg, "\"%");
12287 for (f = fmtstart; f < fmtend; f++) {
12289 sv_catpvn_nomg(msg, f, 1);
12291 Perl_sv_catpvf(aTHX_ msg,
12292 "\\%03"UVof, (UV)*f & 0xFF);
12295 sv_catpvs(msg, "\"");
12297 sv_catpvs(msg, "end of string");
12299 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12302 /* output mangled stuff ... */
12308 /* ... right here, because formatting flags should not apply */
12309 SvGROW(sv, SvCUR(sv) + elen + 1);
12311 Copy(eptr, p, elen, char);
12314 SvCUR_set(sv, p - SvPVX_const(sv));
12316 continue; /* not "break" */
12319 if (is_utf8 != has_utf8) {
12322 sv_utf8_upgrade(sv);
12325 const STRLEN old_elen = elen;
12326 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12327 sv_utf8_upgrade(nsv);
12328 eptr = SvPVX_const(nsv);
12331 if (width) { /* fudge width (can't fudge elen) */
12332 width += elen - old_elen;
12338 assert((IV)elen >= 0); /* here zero elen is fine */
12339 have = esignlen + zeros + elen;
12341 croak_memory_wrap();
12343 need = (have > width ? have : width);
12346 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12347 croak_memory_wrap();
12348 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12350 if (esignlen && fill == '0') {
12352 for (i = 0; i < (int)esignlen; i++)
12353 *p++ = esignbuf[i];
12355 if (gap && !left) {
12356 memset(p, fill, gap);
12359 if (esignlen && fill != '0') {
12361 for (i = 0; i < (int)esignlen; i++)
12362 *p++ = esignbuf[i];
12366 for (i = zeros; i; i--)
12370 Copy(eptr, p, elen, char);
12374 memset(p, ' ', gap);
12379 Copy(dotstr, p, dotstrlen, char);
12383 vectorize = FALSE; /* done iterating over vecstr */
12390 SvCUR_set(sv, p - SvPVX_const(sv));
12397 /* Now that we've consumed all our printf format arguments (svix)
12398 * do we have things left on the stack that we didn't use?
12400 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12401 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12402 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12407 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12411 /* =========================================================================
12413 =head1 Cloning an interpreter
12417 All the macros and functions in this section are for the private use of
12418 the main function, perl_clone().
12420 The foo_dup() functions make an exact copy of an existing foo thingy.
12421 During the course of a cloning, a hash table is used to map old addresses
12422 to new addresses. The table is created and manipulated with the
12423 ptr_table_* functions.
12425 * =========================================================================*/
12428 #if defined(USE_ITHREADS)
12430 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12431 #ifndef GpREFCNT_inc
12432 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12436 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12437 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12438 If this changes, please unmerge ss_dup.
12439 Likewise, sv_dup_inc_multiple() relies on this fact. */
12440 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12441 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12442 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12443 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12444 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12445 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12446 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12447 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12448 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12449 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12450 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12451 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12452 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12454 /* clone a parser */
12457 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12461 PERL_ARGS_ASSERT_PARSER_DUP;
12466 /* look for it in the table first */
12467 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12471 /* create anew and remember what it is */
12472 Newxz(parser, 1, yy_parser);
12473 ptr_table_store(PL_ptr_table, proto, parser);
12475 /* XXX these not yet duped */
12476 parser->old_parser = NULL;
12477 parser->stack = NULL;
12479 parser->stack_size = 0;
12480 /* XXX parser->stack->state = 0; */
12482 /* XXX eventually, just Copy() most of the parser struct ? */
12484 parser->lex_brackets = proto->lex_brackets;
12485 parser->lex_casemods = proto->lex_casemods;
12486 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12487 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12488 parser->lex_casestack = savepvn(proto->lex_casestack,
12489 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12490 parser->lex_defer = proto->lex_defer;
12491 parser->lex_dojoin = proto->lex_dojoin;
12492 parser->lex_formbrack = proto->lex_formbrack;
12493 parser->lex_inpat = proto->lex_inpat;
12494 parser->lex_inwhat = proto->lex_inwhat;
12495 parser->lex_op = proto->lex_op;
12496 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12497 parser->lex_starts = proto->lex_starts;
12498 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12499 parser->multi_close = proto->multi_close;
12500 parser->multi_open = proto->multi_open;
12501 parser->multi_start = proto->multi_start;
12502 parser->multi_end = proto->multi_end;
12503 parser->preambled = proto->preambled;
12504 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12505 parser->linestr = sv_dup_inc(proto->linestr, param);
12506 parser->expect = proto->expect;
12507 parser->copline = proto->copline;
12508 parser->last_lop_op = proto->last_lop_op;
12509 parser->lex_state = proto->lex_state;
12510 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12511 /* rsfp_filters entries have fake IoDIRP() */
12512 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12513 parser->in_my = proto->in_my;
12514 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12515 parser->error_count = proto->error_count;
12518 parser->linestr = sv_dup_inc(proto->linestr, param);
12521 char * const ols = SvPVX(proto->linestr);
12522 char * const ls = SvPVX(parser->linestr);
12524 parser->bufptr = ls + (proto->bufptr >= ols ?
12525 proto->bufptr - ols : 0);
12526 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12527 proto->oldbufptr - ols : 0);
12528 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12529 proto->oldoldbufptr - ols : 0);
12530 parser->linestart = ls + (proto->linestart >= ols ?
12531 proto->linestart - ols : 0);
12532 parser->last_uni = ls + (proto->last_uni >= ols ?
12533 proto->last_uni - ols : 0);
12534 parser->last_lop = ls + (proto->last_lop >= ols ?
12535 proto->last_lop - ols : 0);
12537 parser->bufend = ls + SvCUR(parser->linestr);
12540 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12543 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12544 Copy(proto->nexttype, parser->nexttype, 5, I32);
12545 parser->nexttoke = proto->nexttoke;
12547 /* XXX should clone saved_curcop here, but we aren't passed
12548 * proto_perl; so do it in perl_clone_using instead */
12554 /* duplicate a file handle */
12557 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12561 PERL_ARGS_ASSERT_FP_DUP;
12562 PERL_UNUSED_ARG(type);
12565 return (PerlIO*)NULL;
12567 /* look for it in the table first */
12568 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12572 /* create anew and remember what it is */
12573 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12574 ptr_table_store(PL_ptr_table, fp, ret);
12578 /* duplicate a directory handle */
12581 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12585 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12587 const Direntry_t *dirent;
12588 char smallbuf[256];
12594 PERL_UNUSED_CONTEXT;
12595 PERL_ARGS_ASSERT_DIRP_DUP;
12600 /* look for it in the table first */
12601 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12605 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12607 PERL_UNUSED_ARG(param);
12611 /* open the current directory (so we can switch back) */
12612 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12614 /* chdir to our dir handle and open the present working directory */
12615 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12616 PerlDir_close(pwd);
12617 return (DIR *)NULL;
12619 /* Now we should have two dir handles pointing to the same dir. */
12621 /* Be nice to the calling code and chdir back to where we were. */
12622 /* XXX If this fails, then what? */
12623 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12625 /* We have no need of the pwd handle any more. */
12626 PerlDir_close(pwd);
12629 # define d_namlen(d) (d)->d_namlen
12631 # define d_namlen(d) strlen((d)->d_name)
12633 /* Iterate once through dp, to get the file name at the current posi-
12634 tion. Then step back. */
12635 pos = PerlDir_tell(dp);
12636 if ((dirent = PerlDir_read(dp))) {
12637 len = d_namlen(dirent);
12638 if (len <= sizeof smallbuf) name = smallbuf;
12639 else Newx(name, len, char);
12640 Move(dirent->d_name, name, len, char);
12642 PerlDir_seek(dp, pos);
12644 /* Iterate through the new dir handle, till we find a file with the
12646 if (!dirent) /* just before the end */
12648 pos = PerlDir_tell(ret);
12649 if (PerlDir_read(ret)) continue; /* not there yet */
12650 PerlDir_seek(ret, pos); /* step back */
12654 const long pos0 = PerlDir_tell(ret);
12656 pos = PerlDir_tell(ret);
12657 if ((dirent = PerlDir_read(ret))) {
12658 if (len == (STRLEN)d_namlen(dirent)
12659 && memEQ(name, dirent->d_name, len)) {
12661 PerlDir_seek(ret, pos); /* step back */
12664 /* else we are not there yet; keep iterating */
12666 else { /* This is not meant to happen. The best we can do is
12667 reset the iterator to the beginning. */
12668 PerlDir_seek(ret, pos0);
12675 if (name && name != smallbuf)
12680 ret = win32_dirp_dup(dp, param);
12683 /* pop it in the pointer table */
12685 ptr_table_store(PL_ptr_table, dp, ret);
12690 /* duplicate a typeglob */
12693 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12697 PERL_ARGS_ASSERT_GP_DUP;
12701 /* look for it in the table first */
12702 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12706 /* create anew and remember what it is */
12708 ptr_table_store(PL_ptr_table, gp, ret);
12711 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12712 on Newxz() to do this for us. */
12713 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12714 ret->gp_io = io_dup_inc(gp->gp_io, param);
12715 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12716 ret->gp_av = av_dup_inc(gp->gp_av, param);
12717 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12718 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12719 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12720 ret->gp_cvgen = gp->gp_cvgen;
12721 ret->gp_line = gp->gp_line;
12722 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12726 /* duplicate a chain of magic */
12729 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12731 MAGIC *mgret = NULL;
12732 MAGIC **mgprev_p = &mgret;
12734 PERL_ARGS_ASSERT_MG_DUP;
12736 for (; mg; mg = mg->mg_moremagic) {
12739 if ((param->flags & CLONEf_JOIN_IN)
12740 && mg->mg_type == PERL_MAGIC_backref)
12741 /* when joining, we let the individual SVs add themselves to
12742 * backref as needed. */
12745 Newx(nmg, 1, MAGIC);
12747 mgprev_p = &(nmg->mg_moremagic);
12749 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12750 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12751 from the original commit adding Perl_mg_dup() - revision 4538.
12752 Similarly there is the annotation "XXX random ptr?" next to the
12753 assignment to nmg->mg_ptr. */
12756 /* FIXME for plugins
12757 if (nmg->mg_type == PERL_MAGIC_qr) {
12758 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12762 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12763 ? nmg->mg_type == PERL_MAGIC_backref
12764 /* The backref AV has its reference
12765 * count deliberately bumped by 1 */
12766 ? SvREFCNT_inc(av_dup_inc((const AV *)
12767 nmg->mg_obj, param))
12768 : sv_dup_inc(nmg->mg_obj, param)
12769 : sv_dup(nmg->mg_obj, param);
12771 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12772 if (nmg->mg_len > 0) {
12773 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12774 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12775 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12777 AMT * const namtp = (AMT*)nmg->mg_ptr;
12778 sv_dup_inc_multiple((SV**)(namtp->table),
12779 (SV**)(namtp->table), NofAMmeth, param);
12782 else if (nmg->mg_len == HEf_SVKEY)
12783 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12785 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12786 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12792 #endif /* USE_ITHREADS */
12794 struct ptr_tbl_arena {
12795 struct ptr_tbl_arena *next;
12796 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12799 /* create a new pointer-mapping table */
12802 Perl_ptr_table_new(pTHX)
12805 PERL_UNUSED_CONTEXT;
12807 Newx(tbl, 1, PTR_TBL_t);
12808 tbl->tbl_max = 511;
12809 tbl->tbl_items = 0;
12810 tbl->tbl_arena = NULL;
12811 tbl->tbl_arena_next = NULL;
12812 tbl->tbl_arena_end = NULL;
12813 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12817 #define PTR_TABLE_HASH(ptr) \
12818 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12820 /* map an existing pointer using a table */
12822 STATIC PTR_TBL_ENT_t *
12823 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12825 PTR_TBL_ENT_t *tblent;
12826 const UV hash = PTR_TABLE_HASH(sv);
12828 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12830 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12831 for (; tblent; tblent = tblent->next) {
12832 if (tblent->oldval == sv)
12839 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12841 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12843 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12844 PERL_UNUSED_CONTEXT;
12846 return tblent ? tblent->newval : NULL;
12849 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12850 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12851 * the core's typical use of ptr_tables in thread cloning. */
12854 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12856 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12858 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12859 PERL_UNUSED_CONTEXT;
12862 tblent->newval = newsv;
12864 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12866 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12867 struct ptr_tbl_arena *new_arena;
12869 Newx(new_arena, 1, struct ptr_tbl_arena);
12870 new_arena->next = tbl->tbl_arena;
12871 tbl->tbl_arena = new_arena;
12872 tbl->tbl_arena_next = new_arena->array;
12873 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
12876 tblent = tbl->tbl_arena_next++;
12878 tblent->oldval = oldsv;
12879 tblent->newval = newsv;
12880 tblent->next = tbl->tbl_ary[entry];
12881 tbl->tbl_ary[entry] = tblent;
12883 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12884 ptr_table_split(tbl);
12888 /* double the hash bucket size of an existing ptr table */
12891 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
12893 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
12894 const UV oldsize = tbl->tbl_max + 1;
12895 UV newsize = oldsize * 2;
12898 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
12899 PERL_UNUSED_CONTEXT;
12901 Renew(ary, newsize, PTR_TBL_ENT_t*);
12902 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
12903 tbl->tbl_max = --newsize;
12904 tbl->tbl_ary = ary;
12905 for (i=0; i < oldsize; i++, ary++) {
12906 PTR_TBL_ENT_t **entp = ary;
12907 PTR_TBL_ENT_t *ent = *ary;
12908 PTR_TBL_ENT_t **curentp;
12911 curentp = ary + oldsize;
12913 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
12915 ent->next = *curentp;
12925 /* remove all the entries from a ptr table */
12926 /* Deprecated - will be removed post 5.14 */
12929 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
12931 PERL_UNUSED_CONTEXT;
12932 if (tbl && tbl->tbl_items) {
12933 struct ptr_tbl_arena *arena = tbl->tbl_arena;
12935 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
12938 struct ptr_tbl_arena *next = arena->next;
12944 tbl->tbl_items = 0;
12945 tbl->tbl_arena = NULL;
12946 tbl->tbl_arena_next = NULL;
12947 tbl->tbl_arena_end = NULL;
12951 /* clear and free a ptr table */
12954 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
12956 struct ptr_tbl_arena *arena;
12958 PERL_UNUSED_CONTEXT;
12964 arena = tbl->tbl_arena;
12967 struct ptr_tbl_arena *next = arena->next;
12973 Safefree(tbl->tbl_ary);
12977 #if defined(USE_ITHREADS)
12980 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
12982 PERL_ARGS_ASSERT_RVPV_DUP;
12984 assert(!isREGEXP(sstr));
12986 if (SvWEAKREF(sstr)) {
12987 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
12988 if (param->flags & CLONEf_JOIN_IN) {
12989 /* if joining, we add any back references individually rather
12990 * than copying the whole backref array */
12991 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
12995 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
12997 else if (SvPVX_const(sstr)) {
12998 /* Has something there */
13000 /* Normal PV - clone whole allocated space */
13001 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13002 /* sstr may not be that normal, but actually copy on write.
13003 But we are a true, independent SV, so: */
13007 /* Special case - not normally malloced for some reason */
13008 if (isGV_with_GP(sstr)) {
13009 /* Don't need to do anything here. */
13011 else if ((SvIsCOW(sstr))) {
13012 /* A "shared" PV - clone it as "shared" PV */
13014 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13018 /* Some other special case - random pointer */
13019 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13024 /* Copy the NULL */
13025 SvPV_set(dstr, NULL);
13029 /* duplicate a list of SVs. source and dest may point to the same memory. */
13031 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13032 SSize_t items, CLONE_PARAMS *const param)
13034 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13036 while (items-- > 0) {
13037 *dest++ = sv_dup_inc(*source++, param);
13043 /* duplicate an SV of any type (including AV, HV etc) */
13046 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13051 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13053 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13054 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13059 /* look for it in the table first */
13060 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13064 if(param->flags & CLONEf_JOIN_IN) {
13065 /** We are joining here so we don't want do clone
13066 something that is bad **/
13067 if (SvTYPE(sstr) == SVt_PVHV) {
13068 const HEK * const hvname = HvNAME_HEK(sstr);
13070 /** don't clone stashes if they already exist **/
13071 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13072 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13073 ptr_table_store(PL_ptr_table, sstr, dstr);
13077 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13078 HV *stash = GvSTASH(sstr);
13079 const HEK * hvname;
13080 if (stash && (hvname = HvNAME_HEK(stash))) {
13081 /** don't clone GVs if they already exist **/
13083 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13084 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13086 stash, GvNAME(sstr),
13092 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13093 ptr_table_store(PL_ptr_table, sstr, *svp);
13100 /* create anew and remember what it is */
13103 #ifdef DEBUG_LEAKING_SCALARS
13104 dstr->sv_debug_optype = sstr->sv_debug_optype;
13105 dstr->sv_debug_line = sstr->sv_debug_line;
13106 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13107 dstr->sv_debug_parent = (SV*)sstr;
13108 FREE_SV_DEBUG_FILE(dstr);
13109 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13112 ptr_table_store(PL_ptr_table, sstr, dstr);
13115 SvFLAGS(dstr) = SvFLAGS(sstr);
13116 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13117 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13120 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13121 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13122 (void*)PL_watch_pvx, SvPVX_const(sstr));
13125 /* don't clone objects whose class has asked us not to */
13126 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
13131 switch (SvTYPE(sstr)) {
13133 SvANY(dstr) = NULL;
13136 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
13138 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13140 SvIV_set(dstr, SvIVX(sstr));
13144 SvANY(dstr) = new_XNV();
13145 SvNV_set(dstr, SvNVX(sstr));
13149 /* These are all the types that need complex bodies allocating. */
13151 const svtype sv_type = SvTYPE(sstr);
13152 const struct body_details *const sv_type_details
13153 = bodies_by_type + sv_type;
13157 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13173 assert(sv_type_details->body_size);
13174 if (sv_type_details->arena) {
13175 new_body_inline(new_body, sv_type);
13177 = (void*)((char*)new_body - sv_type_details->offset);
13179 new_body = new_NOARENA(sv_type_details);
13183 SvANY(dstr) = new_body;
13186 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13187 ((char*)SvANY(dstr)) + sv_type_details->offset,
13188 sv_type_details->copy, char);
13190 Copy(((char*)SvANY(sstr)),
13191 ((char*)SvANY(dstr)),
13192 sv_type_details->body_size + sv_type_details->offset, char);
13195 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13196 && !isGV_with_GP(dstr)
13198 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13199 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13201 /* The Copy above means that all the source (unduplicated) pointers
13202 are now in the destination. We can check the flags and the
13203 pointers in either, but it's possible that there's less cache
13204 missing by always going for the destination.
13205 FIXME - instrument and check that assumption */
13206 if (sv_type >= SVt_PVMG) {
13207 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
13208 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
13209 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
13211 } else if (SvMAGIC(dstr))
13212 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13213 if (SvOBJECT(dstr) && SvSTASH(dstr))
13214 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13215 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13218 /* The cast silences a GCC warning about unhandled types. */
13219 switch ((int)sv_type) {
13230 /* FIXME for plugins */
13231 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13232 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13235 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13236 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13237 LvTARG(dstr) = dstr;
13238 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13239 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13241 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13242 if (isREGEXP(sstr)) goto duprex;
13244 /* non-GP case already handled above */
13245 if(isGV_with_GP(sstr)) {
13246 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13247 /* Don't call sv_add_backref here as it's going to be
13248 created as part of the magic cloning of the symbol
13249 table--unless this is during a join and the stash
13250 is not actually being cloned. */
13251 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13252 at the point of this comment. */
13253 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13254 if (param->flags & CLONEf_JOIN_IN)
13255 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13256 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13257 (void)GpREFCNT_inc(GvGP(dstr));
13261 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13262 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13263 /* I have no idea why fake dirp (rsfps)
13264 should be treated differently but otherwise
13265 we end up with leaks -- sky*/
13266 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13267 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13268 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13270 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13271 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13272 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13273 if (IoDIRP(dstr)) {
13274 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13277 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13279 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13281 if (IoOFP(dstr) == IoIFP(sstr))
13282 IoOFP(dstr) = IoIFP(dstr);
13284 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13285 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13286 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13287 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13290 /* avoid cloning an empty array */
13291 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13292 SV **dst_ary, **src_ary;
13293 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13295 src_ary = AvARRAY((const AV *)sstr);
13296 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13297 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13298 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13299 AvALLOC((const AV *)dstr) = dst_ary;
13300 if (AvREAL((const AV *)sstr)) {
13301 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13305 while (items-- > 0)
13306 *dst_ary++ = sv_dup(*src_ary++, param);
13308 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13309 while (items-- > 0) {
13310 *dst_ary++ = &PL_sv_undef;
13314 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13315 AvALLOC((const AV *)dstr) = (SV**)NULL;
13316 AvMAX( (const AV *)dstr) = -1;
13317 AvFILLp((const AV *)dstr) = -1;
13321 if (HvARRAY((const HV *)sstr)) {
13323 const bool sharekeys = !!HvSHAREKEYS(sstr);
13324 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13325 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13327 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13328 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13330 HvARRAY(dstr) = (HE**)darray;
13331 while (i <= sxhv->xhv_max) {
13332 const HE * const source = HvARRAY(sstr)[i];
13333 HvARRAY(dstr)[i] = source
13334 ? he_dup(source, sharekeys, param) : 0;
13338 const struct xpvhv_aux * const saux = HvAUX(sstr);
13339 struct xpvhv_aux * const daux = HvAUX(dstr);
13340 /* This flag isn't copied. */
13343 if (saux->xhv_name_count) {
13344 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13346 = saux->xhv_name_count < 0
13347 ? -saux->xhv_name_count
13348 : saux->xhv_name_count;
13349 HEK **shekp = sname + count;
13351 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13352 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13353 while (shekp-- > sname) {
13355 *dhekp = hek_dup(*shekp, param);
13359 daux->xhv_name_u.xhvnameu_name
13360 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13363 daux->xhv_name_count = saux->xhv_name_count;
13365 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13366 daux->xhv_aux_flags = saux->xhv_aux_flags;
13367 #ifdef PERL_HASH_RANDOMIZE_KEYS
13368 daux->xhv_rand = saux->xhv_rand;
13369 daux->xhv_last_rand = saux->xhv_last_rand;
13371 daux->xhv_riter = saux->xhv_riter;
13372 daux->xhv_eiter = saux->xhv_eiter
13373 ? he_dup(saux->xhv_eiter,
13374 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13375 /* backref array needs refcnt=2; see sv_add_backref */
13376 daux->xhv_backreferences =
13377 (param->flags & CLONEf_JOIN_IN)
13378 /* when joining, we let the individual GVs and
13379 * CVs add themselves to backref as
13380 * needed. This avoids pulling in stuff
13381 * that isn't required, and simplifies the
13382 * case where stashes aren't cloned back
13383 * if they already exist in the parent
13386 : saux->xhv_backreferences
13387 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13388 ? MUTABLE_AV(SvREFCNT_inc(
13389 sv_dup_inc((const SV *)
13390 saux->xhv_backreferences, param)))
13391 : MUTABLE_AV(sv_dup((const SV *)
13392 saux->xhv_backreferences, param))
13395 daux->xhv_mro_meta = saux->xhv_mro_meta
13396 ? mro_meta_dup(saux->xhv_mro_meta, param)
13399 /* Record stashes for possible cloning in Perl_clone(). */
13401 av_push(param->stashes, dstr);
13405 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13408 if (!(param->flags & CLONEf_COPY_STACKS)) {
13413 /* NOTE: not refcounted */
13414 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13415 hv_dup(CvSTASH(dstr), param);
13416 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13417 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13418 if (!CvISXSUB(dstr)) {
13420 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13422 CvSLABBED_off(dstr);
13423 } else if (CvCONST(dstr)) {
13424 CvXSUBANY(dstr).any_ptr =
13425 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13427 assert(!CvSLABBED(dstr));
13428 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13430 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13431 hek_dup(CvNAME_HEK((CV *)sstr), param);
13432 /* don't dup if copying back - CvGV isn't refcounted, so the
13433 * duped GV may never be freed. A bit of a hack! DAPM */
13435 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13437 ? gv_dup_inc(CvGV(sstr), param)
13438 : (param->flags & CLONEf_JOIN_IN)
13440 : gv_dup(CvGV(sstr), param);
13442 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
13444 CvWEAKOUTSIDE(sstr)
13445 ? cv_dup( CvOUTSIDE(dstr), param)
13446 : cv_dup_inc(CvOUTSIDE(dstr), param);
13456 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13458 PERL_ARGS_ASSERT_SV_DUP_INC;
13459 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13463 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13465 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13466 PERL_ARGS_ASSERT_SV_DUP;
13468 /* Track every SV that (at least initially) had a reference count of 0.
13469 We need to do this by holding an actual reference to it in this array.
13470 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13471 (akin to the stashes hash, and the perl stack), we come unstuck if
13472 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13473 thread) is manipulated in a CLONE method, because CLONE runs before the
13474 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13475 (and fix things up by giving each a reference via the temps stack).
13476 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13477 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13478 before the walk of unreferenced happens and a reference to that is SV
13479 added to the temps stack. At which point we have the same SV considered
13480 to be in use, and free to be re-used. Not good.
13482 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13483 assert(param->unreferenced);
13484 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13490 /* duplicate a context */
13493 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13495 PERL_CONTEXT *ncxs;
13497 PERL_ARGS_ASSERT_CX_DUP;
13500 return (PERL_CONTEXT*)NULL;
13502 /* look for it in the table first */
13503 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13507 /* create anew and remember what it is */
13508 Newx(ncxs, max + 1, PERL_CONTEXT);
13509 ptr_table_store(PL_ptr_table, cxs, ncxs);
13510 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13513 PERL_CONTEXT * const ncx = &ncxs[ix];
13514 if (CxTYPE(ncx) == CXt_SUBST) {
13515 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13518 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13519 switch (CxTYPE(ncx)) {
13521 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13522 ? cv_dup_inc(ncx->blk_sub.cv, param)
13523 : cv_dup(ncx->blk_sub.cv,param));
13524 if(CxHASARGS(ncx)){
13525 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13526 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13528 ncx->blk_sub.argarray = NULL;
13529 ncx->blk_sub.savearray = NULL;
13531 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13532 ncx->blk_sub.oldcomppad);
13535 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13537 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13538 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13540 case CXt_LOOP_LAZYSV:
13541 ncx->blk_loop.state_u.lazysv.end
13542 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13543 /* We are taking advantage of av_dup_inc and sv_dup_inc
13544 actually being the same function, and order equivalence of
13546 We can assert the later [but only at run time :-(] */
13547 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13548 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13550 ncx->blk_loop.state_u.ary.ary
13551 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13552 case CXt_LOOP_LAZYIV:
13553 case CXt_LOOP_PLAIN:
13554 if (CxPADLOOP(ncx)) {
13555 ncx->blk_loop.itervar_u.oldcomppad
13556 = (PAD*)ptr_table_fetch(PL_ptr_table,
13557 ncx->blk_loop.itervar_u.oldcomppad);
13559 ncx->blk_loop.itervar_u.gv
13560 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13565 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13566 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13567 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13582 /* duplicate a stack info structure */
13585 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13589 PERL_ARGS_ASSERT_SI_DUP;
13592 return (PERL_SI*)NULL;
13594 /* look for it in the table first */
13595 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13599 /* create anew and remember what it is */
13600 Newxz(nsi, 1, PERL_SI);
13601 ptr_table_store(PL_ptr_table, si, nsi);
13603 nsi->si_stack = av_dup_inc(si->si_stack, param);
13604 nsi->si_cxix = si->si_cxix;
13605 nsi->si_cxmax = si->si_cxmax;
13606 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13607 nsi->si_type = si->si_type;
13608 nsi->si_prev = si_dup(si->si_prev, param);
13609 nsi->si_next = si_dup(si->si_next, param);
13610 nsi->si_markoff = si->si_markoff;
13615 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13616 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13617 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13618 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13619 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13620 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13621 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13622 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13623 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13624 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13625 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13626 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13627 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13628 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13629 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13630 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13633 #define pv_dup_inc(p) SAVEPV(p)
13634 #define pv_dup(p) SAVEPV(p)
13635 #define svp_dup_inc(p,pp) any_dup(p,pp)
13637 /* map any object to the new equivent - either something in the
13638 * ptr table, or something in the interpreter structure
13642 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13646 PERL_ARGS_ASSERT_ANY_DUP;
13649 return (void*)NULL;
13651 /* look for it in the table first */
13652 ret = ptr_table_fetch(PL_ptr_table, v);
13656 /* see if it is part of the interpreter structure */
13657 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13658 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13666 /* duplicate the save stack */
13669 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13672 ANY * const ss = proto_perl->Isavestack;
13673 const I32 max = proto_perl->Isavestack_max;
13674 I32 ix = proto_perl->Isavestack_ix;
13687 void (*dptr) (void*);
13688 void (*dxptr) (pTHX_ void*);
13690 PERL_ARGS_ASSERT_SS_DUP;
13692 Newxz(nss, max, ANY);
13695 const UV uv = POPUV(ss,ix);
13696 const U8 type = (U8)uv & SAVE_MASK;
13698 TOPUV(nss,ix) = uv;
13700 case SAVEt_CLEARSV:
13701 case SAVEt_CLEARPADRANGE:
13703 case SAVEt_HELEM: /* hash element */
13704 sv = (const SV *)POPPTR(ss,ix);
13705 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13707 case SAVEt_ITEM: /* normal string */
13708 case SAVEt_GVSV: /* scalar slot in GV */
13709 case SAVEt_SV: /* scalar reference */
13710 sv = (const SV *)POPPTR(ss,ix);
13711 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13714 case SAVEt_MORTALIZESV:
13715 case SAVEt_READONLY_OFF:
13716 sv = (const SV *)POPPTR(ss,ix);
13717 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13719 case SAVEt_SHARED_PVREF: /* char* in shared space */
13720 c = (char*)POPPTR(ss,ix);
13721 TOPPTR(nss,ix) = savesharedpv(c);
13722 ptr = POPPTR(ss,ix);
13723 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13725 case SAVEt_GENERIC_SVREF: /* generic sv */
13726 case SAVEt_SVREF: /* scalar reference */
13727 sv = (const SV *)POPPTR(ss,ix);
13728 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13729 ptr = POPPTR(ss,ix);
13730 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13732 case SAVEt_GVSLOT: /* any slot in GV */
13733 sv = (const SV *)POPPTR(ss,ix);
13734 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13735 ptr = POPPTR(ss,ix);
13736 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13737 sv = (const SV *)POPPTR(ss,ix);
13738 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13740 case SAVEt_HV: /* hash reference */
13741 case SAVEt_AV: /* array reference */
13742 sv = (const SV *) POPPTR(ss,ix);
13743 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13745 case SAVEt_COMPPAD:
13747 sv = (const SV *) POPPTR(ss,ix);
13748 TOPPTR(nss,ix) = sv_dup(sv, param);
13750 case SAVEt_INT: /* int reference */
13751 ptr = POPPTR(ss,ix);
13752 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13753 intval = (int)POPINT(ss,ix);
13754 TOPINT(nss,ix) = intval;
13756 case SAVEt_LONG: /* long reference */
13757 ptr = POPPTR(ss,ix);
13758 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13759 longval = (long)POPLONG(ss,ix);
13760 TOPLONG(nss,ix) = longval;
13762 case SAVEt_I32: /* I32 reference */
13763 ptr = POPPTR(ss,ix);
13764 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13766 TOPINT(nss,ix) = i;
13768 case SAVEt_IV: /* IV reference */
13769 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13770 ptr = POPPTR(ss,ix);
13771 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13773 TOPIV(nss,ix) = iv;
13775 case SAVEt_HPTR: /* HV* reference */
13776 case SAVEt_APTR: /* AV* reference */
13777 case SAVEt_SPTR: /* SV* reference */
13778 ptr = POPPTR(ss,ix);
13779 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13780 sv = (const SV *)POPPTR(ss,ix);
13781 TOPPTR(nss,ix) = sv_dup(sv, param);
13783 case SAVEt_VPTR: /* random* reference */
13784 ptr = POPPTR(ss,ix);
13785 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13787 case SAVEt_INT_SMALL:
13788 case SAVEt_I32_SMALL:
13789 case SAVEt_I16: /* I16 reference */
13790 case SAVEt_I8: /* I8 reference */
13792 ptr = POPPTR(ss,ix);
13793 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13795 case SAVEt_GENERIC_PVREF: /* generic char* */
13796 case SAVEt_PPTR: /* char* reference */
13797 ptr = POPPTR(ss,ix);
13798 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13799 c = (char*)POPPTR(ss,ix);
13800 TOPPTR(nss,ix) = pv_dup(c);
13802 case SAVEt_GP: /* scalar reference */
13803 gp = (GP*)POPPTR(ss,ix);
13804 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13805 (void)GpREFCNT_inc(gp);
13806 gv = (const GV *)POPPTR(ss,ix);
13807 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13810 ptr = POPPTR(ss,ix);
13811 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13812 /* these are assumed to be refcounted properly */
13814 switch (((OP*)ptr)->op_type) {
13816 case OP_LEAVESUBLV:
13820 case OP_LEAVEWRITE:
13821 TOPPTR(nss,ix) = ptr;
13824 (void) OpREFCNT_inc(o);
13828 TOPPTR(nss,ix) = NULL;
13833 TOPPTR(nss,ix) = NULL;
13835 case SAVEt_FREECOPHH:
13836 ptr = POPPTR(ss,ix);
13837 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13839 case SAVEt_ADELETE:
13840 av = (const AV *)POPPTR(ss,ix);
13841 TOPPTR(nss,ix) = av_dup_inc(av, param);
13843 TOPINT(nss,ix) = i;
13846 hv = (const HV *)POPPTR(ss,ix);
13847 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13849 TOPINT(nss,ix) = i;
13852 c = (char*)POPPTR(ss,ix);
13853 TOPPTR(nss,ix) = pv_dup_inc(c);
13855 case SAVEt_STACK_POS: /* Position on Perl stack */
13857 TOPINT(nss,ix) = i;
13859 case SAVEt_DESTRUCTOR:
13860 ptr = POPPTR(ss,ix);
13861 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13862 dptr = POPDPTR(ss,ix);
13863 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13864 any_dup(FPTR2DPTR(void *, dptr),
13867 case SAVEt_DESTRUCTOR_X:
13868 ptr = POPPTR(ss,ix);
13869 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13870 dxptr = POPDXPTR(ss,ix);
13871 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13872 any_dup(FPTR2DPTR(void *, dxptr),
13875 case SAVEt_REGCONTEXT:
13877 ix -= uv >> SAVE_TIGHT_SHIFT;
13879 case SAVEt_AELEM: /* array element */
13880 sv = (const SV *)POPPTR(ss,ix);
13881 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13883 TOPINT(nss,ix) = i;
13884 av = (const AV *)POPPTR(ss,ix);
13885 TOPPTR(nss,ix) = av_dup_inc(av, param);
13888 ptr = POPPTR(ss,ix);
13889 TOPPTR(nss,ix) = ptr;
13892 ptr = POPPTR(ss,ix);
13893 ptr = cophh_copy((COPHH*)ptr);
13894 TOPPTR(nss,ix) = ptr;
13896 TOPINT(nss,ix) = i;
13897 if (i & HINT_LOCALIZE_HH) {
13898 hv = (const HV *)POPPTR(ss,ix);
13899 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13902 case SAVEt_PADSV_AND_MORTALIZE:
13903 longval = (long)POPLONG(ss,ix);
13904 TOPLONG(nss,ix) = longval;
13905 ptr = POPPTR(ss,ix);
13906 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13907 sv = (const SV *)POPPTR(ss,ix);
13908 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13910 case SAVEt_SET_SVFLAGS:
13912 TOPINT(nss,ix) = i;
13914 TOPINT(nss,ix) = i;
13915 sv = (const SV *)POPPTR(ss,ix);
13916 TOPPTR(nss,ix) = sv_dup(sv, param);
13918 case SAVEt_COMPILE_WARNINGS:
13919 ptr = POPPTR(ss,ix);
13920 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
13923 ptr = POPPTR(ss,ix);
13924 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
13926 case SAVEt_GP_ALIASED_SV:
13927 ptr = POPPTR(ss,ix);
13928 TOPPTR(nss,ix) = gp_dup((GP *)ptr, param);
13929 ((GP *)ptr)->gp_refcnt++;
13933 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
13941 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
13942 * flag to the result. This is done for each stash before cloning starts,
13943 * so we know which stashes want their objects cloned */
13946 do_mark_cloneable_stash(pTHX_ SV *const sv)
13948 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
13950 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
13951 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
13952 if (cloner && GvCV(cloner)) {
13959 mXPUSHs(newSVhek(hvname));
13961 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
13968 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
13976 =for apidoc perl_clone
13978 Create and return a new interpreter by cloning the current one.
13980 perl_clone takes these flags as parameters:
13982 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
13983 without it we only clone the data and zero the stacks,
13984 with it we copy the stacks and the new perl interpreter is
13985 ready to run at the exact same point as the previous one.
13986 The pseudo-fork code uses COPY_STACKS while the
13987 threads->create doesn't.
13989 CLONEf_KEEP_PTR_TABLE -
13990 perl_clone keeps a ptr_table with the pointer of the old
13991 variable as a key and the new variable as a value,
13992 this allows it to check if something has been cloned and not
13993 clone it again but rather just use the value and increase the
13994 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
13995 the ptr_table using the function
13996 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
13997 reason to keep it around is if you want to dup some of your own
13998 variable who are outside the graph perl scans, example of this
13999 code is in threads.xs create.
14001 CLONEf_CLONE_HOST -
14002 This is a win32 thing, it is ignored on unix, it tells perls
14003 win32host code (which is c++) to clone itself, this is needed on
14004 win32 if you want to run two threads at the same time,
14005 if you just want to do some stuff in a separate perl interpreter
14006 and then throw it away and return to the original one,
14007 you don't need to do anything.
14012 /* XXX the above needs expanding by someone who actually understands it ! */
14013 EXTERN_C PerlInterpreter *
14014 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14017 perl_clone(PerlInterpreter *proto_perl, UV flags)
14020 #ifdef PERL_IMPLICIT_SYS
14022 PERL_ARGS_ASSERT_PERL_CLONE;
14024 /* perlhost.h so we need to call into it
14025 to clone the host, CPerlHost should have a c interface, sky */
14027 if (flags & CLONEf_CLONE_HOST) {
14028 return perl_clone_host(proto_perl,flags);
14030 return perl_clone_using(proto_perl, flags,
14032 proto_perl->IMemShared,
14033 proto_perl->IMemParse,
14035 proto_perl->IStdIO,
14039 proto_perl->IProc);
14043 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14044 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14045 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14046 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14047 struct IPerlDir* ipD, struct IPerlSock* ipS,
14048 struct IPerlProc* ipP)
14050 /* XXX many of the string copies here can be optimized if they're
14051 * constants; they need to be allocated as common memory and just
14052 * their pointers copied. */
14055 CLONE_PARAMS clone_params;
14056 CLONE_PARAMS* const param = &clone_params;
14058 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14060 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14061 #else /* !PERL_IMPLICIT_SYS */
14063 CLONE_PARAMS clone_params;
14064 CLONE_PARAMS* param = &clone_params;
14065 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14067 PERL_ARGS_ASSERT_PERL_CLONE;
14068 #endif /* PERL_IMPLICIT_SYS */
14070 /* for each stash, determine whether its objects should be cloned */
14071 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14072 PERL_SET_THX(my_perl);
14075 PoisonNew(my_perl, 1, PerlInterpreter);
14078 PL_defstash = NULL; /* may be used by perl malloc() */
14081 PL_scopestack_name = 0;
14083 PL_savestack_ix = 0;
14084 PL_savestack_max = -1;
14085 PL_sig_pending = 0;
14087 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14088 # ifdef DEBUG_LEAKING_SCALARS
14089 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14091 #else /* !DEBUGGING */
14092 Zero(my_perl, 1, PerlInterpreter);
14093 #endif /* DEBUGGING */
14095 #ifdef PERL_IMPLICIT_SYS
14096 /* host pointers */
14098 PL_MemShared = ipMS;
14099 PL_MemParse = ipMP;
14106 #endif /* PERL_IMPLICIT_SYS */
14109 param->flags = flags;
14110 /* Nothing in the core code uses this, but we make it available to
14111 extensions (using mg_dup). */
14112 param->proto_perl = proto_perl;
14113 /* Likely nothing will use this, but it is initialised to be consistent
14114 with Perl_clone_params_new(). */
14115 param->new_perl = my_perl;
14116 param->unreferenced = NULL;
14119 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14121 PL_body_arenas = NULL;
14122 Zero(&PL_body_roots, 1, PL_body_roots);
14126 PL_sv_arenaroot = NULL;
14128 PL_debug = proto_perl->Idebug;
14130 /* dbargs array probably holds garbage */
14133 PL_compiling = proto_perl->Icompiling;
14135 /* pseudo environmental stuff */
14136 PL_origargc = proto_perl->Iorigargc;
14137 PL_origargv = proto_perl->Iorigargv;
14139 #ifndef NO_TAINT_SUPPORT
14140 /* Set tainting stuff before PerlIO_debug can possibly get called */
14141 PL_tainting = proto_perl->Itainting;
14142 PL_taint_warn = proto_perl->Itaint_warn;
14144 PL_tainting = FALSE;
14145 PL_taint_warn = FALSE;
14148 PL_minus_c = proto_perl->Iminus_c;
14150 PL_localpatches = proto_perl->Ilocalpatches;
14151 PL_splitstr = proto_perl->Isplitstr;
14152 PL_minus_n = proto_perl->Iminus_n;
14153 PL_minus_p = proto_perl->Iminus_p;
14154 PL_minus_l = proto_perl->Iminus_l;
14155 PL_minus_a = proto_perl->Iminus_a;
14156 PL_minus_E = proto_perl->Iminus_E;
14157 PL_minus_F = proto_perl->Iminus_F;
14158 PL_doswitches = proto_perl->Idoswitches;
14159 PL_dowarn = proto_perl->Idowarn;
14160 PL_sawalias = proto_perl->Isawalias;
14161 #ifdef PERL_SAWAMPERSAND
14162 PL_sawampersand = proto_perl->Isawampersand;
14164 PL_unsafe = proto_perl->Iunsafe;
14165 PL_perldb = proto_perl->Iperldb;
14166 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14167 PL_exit_flags = proto_perl->Iexit_flags;
14169 /* XXX time(&PL_basetime) when asked for? */
14170 PL_basetime = proto_perl->Ibasetime;
14172 PL_maxsysfd = proto_perl->Imaxsysfd;
14173 PL_statusvalue = proto_perl->Istatusvalue;
14175 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14177 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14180 /* RE engine related */
14181 PL_regmatch_slab = NULL;
14182 PL_reg_curpm = NULL;
14184 PL_sub_generation = proto_perl->Isub_generation;
14186 /* funky return mechanisms */
14187 PL_forkprocess = proto_perl->Iforkprocess;
14189 /* internal state */
14190 PL_maxo = proto_perl->Imaxo;
14192 PL_main_start = proto_perl->Imain_start;
14193 PL_eval_root = proto_perl->Ieval_root;
14194 PL_eval_start = proto_perl->Ieval_start;
14196 PL_filemode = proto_perl->Ifilemode;
14197 PL_lastfd = proto_perl->Ilastfd;
14198 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14201 PL_gensym = proto_perl->Igensym;
14203 PL_laststatval = proto_perl->Ilaststatval;
14204 PL_laststype = proto_perl->Ilaststype;
14207 PL_profiledata = NULL;
14209 PL_generation = proto_perl->Igeneration;
14211 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14212 PL_in_clean_all = proto_perl->Iin_clean_all;
14214 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14215 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14216 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14217 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14218 PL_nomemok = proto_perl->Inomemok;
14219 PL_an = proto_perl->Ian;
14220 PL_evalseq = proto_perl->Ievalseq;
14221 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14222 PL_origalen = proto_perl->Iorigalen;
14224 PL_sighandlerp = proto_perl->Isighandlerp;
14226 PL_runops = proto_perl->Irunops;
14228 PL_subline = proto_perl->Isubline;
14231 PL_cryptseen = proto_perl->Icryptseen;
14234 #ifdef USE_LOCALE_COLLATE
14235 PL_collation_ix = proto_perl->Icollation_ix;
14236 PL_collation_standard = proto_perl->Icollation_standard;
14237 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14238 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14239 #endif /* USE_LOCALE_COLLATE */
14241 #ifdef USE_LOCALE_NUMERIC
14242 PL_numeric_standard = proto_perl->Inumeric_standard;
14243 PL_numeric_local = proto_perl->Inumeric_local;
14244 #endif /* !USE_LOCALE_NUMERIC */
14246 /* Did the locale setup indicate UTF-8? */
14247 PL_utf8locale = proto_perl->Iutf8locale;
14248 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14249 /* Unicode features (see perlrun/-C) */
14250 PL_unicode = proto_perl->Iunicode;
14252 /* Pre-5.8 signals control */
14253 PL_signals = proto_perl->Isignals;
14255 /* times() ticks per second */
14256 PL_clocktick = proto_perl->Iclocktick;
14258 /* Recursion stopper for PerlIO_find_layer */
14259 PL_in_load_module = proto_perl->Iin_load_module;
14261 /* sort() routine */
14262 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14264 /* Not really needed/useful since the reenrant_retint is "volatile",
14265 * but do it for consistency's sake. */
14266 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14268 /* Hooks to shared SVs and locks. */
14269 PL_sharehook = proto_perl->Isharehook;
14270 PL_lockhook = proto_perl->Ilockhook;
14271 PL_unlockhook = proto_perl->Iunlockhook;
14272 PL_threadhook = proto_perl->Ithreadhook;
14273 PL_destroyhook = proto_perl->Idestroyhook;
14274 PL_signalhook = proto_perl->Isignalhook;
14276 PL_globhook = proto_perl->Iglobhook;
14279 PL_last_swash_hv = NULL; /* reinits on demand */
14280 PL_last_swash_klen = 0;
14281 PL_last_swash_key[0]= '\0';
14282 PL_last_swash_tmps = (U8*)NULL;
14283 PL_last_swash_slen = 0;
14285 PL_srand_called = proto_perl->Isrand_called;
14286 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14288 if (flags & CLONEf_COPY_STACKS) {
14289 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14290 PL_tmps_ix = proto_perl->Itmps_ix;
14291 PL_tmps_max = proto_perl->Itmps_max;
14292 PL_tmps_floor = proto_perl->Itmps_floor;
14294 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14295 * NOTE: unlike the others! */
14296 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14297 PL_scopestack_max = proto_perl->Iscopestack_max;
14299 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14300 * NOTE: unlike the others! */
14301 PL_savestack_ix = proto_perl->Isavestack_ix;
14302 PL_savestack_max = proto_perl->Isavestack_max;
14305 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14306 PL_top_env = &PL_start_env;
14308 PL_op = proto_perl->Iop;
14311 PL_Xpv = (XPV*)NULL;
14312 my_perl->Ina = proto_perl->Ina;
14314 PL_statbuf = proto_perl->Istatbuf;
14315 PL_statcache = proto_perl->Istatcache;
14317 #ifndef NO_TAINT_SUPPORT
14318 PL_tainted = proto_perl->Itainted;
14320 PL_tainted = FALSE;
14322 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14324 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14326 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14327 PL_restartop = proto_perl->Irestartop;
14328 PL_in_eval = proto_perl->Iin_eval;
14329 PL_delaymagic = proto_perl->Idelaymagic;
14330 PL_phase = proto_perl->Iphase;
14331 PL_localizing = proto_perl->Ilocalizing;
14333 PL_hv_fetch_ent_mh = NULL;
14334 PL_modcount = proto_perl->Imodcount;
14335 PL_lastgotoprobe = NULL;
14336 PL_dumpindent = proto_perl->Idumpindent;
14338 PL_efloatbuf = NULL; /* reinits on demand */
14339 PL_efloatsize = 0; /* reinits on demand */
14343 PL_colorset = 0; /* reinits PL_colors[] */
14344 /*PL_colors[6] = {0,0,0,0,0,0};*/
14346 /* Pluggable optimizer */
14347 PL_peepp = proto_perl->Ipeepp;
14348 PL_rpeepp = proto_perl->Irpeepp;
14349 /* op_free() hook */
14350 PL_opfreehook = proto_perl->Iopfreehook;
14352 #ifdef USE_REENTRANT_API
14353 /* XXX: things like -Dm will segfault here in perlio, but doing
14354 * PERL_SET_CONTEXT(proto_perl);
14355 * breaks too many other things
14357 Perl_reentrant_init(aTHX);
14360 /* create SV map for pointer relocation */
14361 PL_ptr_table = ptr_table_new();
14363 /* initialize these special pointers as early as possible */
14365 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14366 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14367 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14369 /* create (a non-shared!) shared string table */
14370 PL_strtab = newHV();
14371 HvSHAREKEYS_off(PL_strtab);
14372 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14373 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14375 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14377 /* This PV will be free'd special way so must set it same way op.c does */
14378 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14379 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14381 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14382 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14383 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14384 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14386 param->stashes = newAV(); /* Setup array of objects to call clone on */
14387 /* This makes no difference to the implementation, as it always pushes
14388 and shifts pointers to other SVs without changing their reference
14389 count, with the array becoming empty before it is freed. However, it
14390 makes it conceptually clear what is going on, and will avoid some
14391 work inside av.c, filling slots between AvFILL() and AvMAX() with
14392 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14393 AvREAL_off(param->stashes);
14395 if (!(flags & CLONEf_COPY_STACKS)) {
14396 param->unreferenced = newAV();
14399 #ifdef PERLIO_LAYERS
14400 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14401 PerlIO_clone(aTHX_ proto_perl, param);
14404 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14405 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14406 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14407 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14408 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14409 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14412 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14413 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
14414 PL_inplace = SAVEPV(proto_perl->Iinplace);
14415 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14417 /* magical thingies */
14419 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14421 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14422 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14423 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14426 /* Clone the regex array */
14427 /* ORANGE FIXME for plugins, probably in the SV dup code.
14428 newSViv(PTR2IV(CALLREGDUPE(
14429 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14431 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14432 PL_regex_pad = AvARRAY(PL_regex_padav);
14434 PL_stashpadmax = proto_perl->Istashpadmax;
14435 PL_stashpadix = proto_perl->Istashpadix ;
14436 Newx(PL_stashpad, PL_stashpadmax, HV *);
14439 for (; o < PL_stashpadmax; ++o)
14440 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14443 /* shortcuts to various I/O objects */
14444 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14445 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14446 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14447 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14448 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14449 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14450 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14452 /* shortcuts to regexp stuff */
14453 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14455 /* shortcuts to misc objects */
14456 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14458 /* shortcuts to debugging objects */
14459 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14460 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14461 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14462 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14463 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14464 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14466 /* symbol tables */
14467 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14468 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14469 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14470 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14471 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14473 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14474 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14475 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14476 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14477 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14478 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14479 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14480 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14482 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14484 /* subprocess state */
14485 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14487 if (proto_perl->Iop_mask)
14488 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14491 /* PL_asserting = proto_perl->Iasserting; */
14493 /* current interpreter roots */
14494 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14496 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14499 /* runtime control stuff */
14500 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14502 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14504 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14506 /* interpreter atexit processing */
14507 PL_exitlistlen = proto_perl->Iexitlistlen;
14508 if (PL_exitlistlen) {
14509 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14510 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14513 PL_exitlist = (PerlExitListEntry*)NULL;
14515 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14516 if (PL_my_cxt_size) {
14517 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14518 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14519 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14520 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14521 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14525 PL_my_cxt_list = (void**)NULL;
14526 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14527 PL_my_cxt_keys = (const char**)NULL;
14530 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14531 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14532 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14533 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14535 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14537 PAD_CLONE_VARS(proto_perl, param);
14539 #ifdef HAVE_INTERP_INTERN
14540 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14543 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14545 #ifdef PERL_USES_PL_PIDSTATUS
14546 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14548 PL_osname = SAVEPV(proto_perl->Iosname);
14549 PL_parser = parser_dup(proto_perl->Iparser, param);
14551 /* XXX this only works if the saved cop has already been cloned */
14552 if (proto_perl->Iparser) {
14553 PL_parser->saved_curcop = (COP*)any_dup(
14554 proto_perl->Iparser->saved_curcop,
14558 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14560 #ifdef USE_LOCALE_COLLATE
14561 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14562 #endif /* USE_LOCALE_COLLATE */
14564 #ifdef USE_LOCALE_NUMERIC
14565 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14566 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14567 #endif /* !USE_LOCALE_NUMERIC */
14569 /* Unicode inversion lists */
14570 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14571 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14572 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14573 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14575 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14576 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14578 /* utf8 character class swashes */
14579 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14580 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14582 for (i = 0; i < POSIX_CC_COUNT; i++) {
14583 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14585 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14586 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14587 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14588 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14589 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14590 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14591 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14592 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14593 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14594 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14595 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14596 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14597 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14598 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14599 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14600 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14602 if (proto_perl->Ipsig_pend) {
14603 Newxz(PL_psig_pend, SIG_SIZE, int);
14606 PL_psig_pend = (int*)NULL;
14609 if (proto_perl->Ipsig_name) {
14610 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14611 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14613 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14616 PL_psig_ptr = (SV**)NULL;
14617 PL_psig_name = (SV**)NULL;
14620 if (flags & CLONEf_COPY_STACKS) {
14621 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14622 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14623 PL_tmps_ix+1, param);
14625 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14626 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14627 Newxz(PL_markstack, i, I32);
14628 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14629 - proto_perl->Imarkstack);
14630 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14631 - proto_perl->Imarkstack);
14632 Copy(proto_perl->Imarkstack, PL_markstack,
14633 PL_markstack_ptr - PL_markstack + 1, I32);
14635 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14636 * NOTE: unlike the others! */
14637 Newxz(PL_scopestack, PL_scopestack_max, I32);
14638 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14641 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14642 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14644 /* reset stack AV to correct length before its duped via
14645 * PL_curstackinfo */
14646 AvFILLp(proto_perl->Icurstack) =
14647 proto_perl->Istack_sp - proto_perl->Istack_base;
14649 /* NOTE: si_dup() looks at PL_markstack */
14650 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14652 /* PL_curstack = PL_curstackinfo->si_stack; */
14653 PL_curstack = av_dup(proto_perl->Icurstack, param);
14654 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14656 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14657 PL_stack_base = AvARRAY(PL_curstack);
14658 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14659 - proto_perl->Istack_base);
14660 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14662 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14663 PL_savestack = ss_dup(proto_perl, param);
14667 ENTER; /* perl_destruct() wants to LEAVE; */
14670 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14671 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14673 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14674 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14675 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14676 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14677 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14678 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14680 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14682 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14683 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14684 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14686 PL_stashcache = newHV();
14688 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14689 proto_perl->Iwatchaddr);
14690 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14691 if (PL_debug && PL_watchaddr) {
14692 PerlIO_printf(Perl_debug_log,
14693 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14694 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14695 PTR2UV(PL_watchok));
14698 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14699 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14700 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14702 /* Call the ->CLONE method, if it exists, for each of the stashes
14703 identified by sv_dup() above.
14705 while(av_tindex(param->stashes) != -1) {
14706 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14707 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14708 if (cloner && GvCV(cloner)) {
14713 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14715 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14721 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14722 ptr_table_free(PL_ptr_table);
14723 PL_ptr_table = NULL;
14726 if (!(flags & CLONEf_COPY_STACKS)) {
14727 unreferenced_to_tmp_stack(param->unreferenced);
14730 SvREFCNT_dec(param->stashes);
14732 /* orphaned? eg threads->new inside BEGIN or use */
14733 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14734 SvREFCNT_inc_simple_void(PL_compcv);
14735 SAVEFREESV(PL_compcv);
14742 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14744 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14746 if (AvFILLp(unreferenced) > -1) {
14747 SV **svp = AvARRAY(unreferenced);
14748 SV **const last = svp + AvFILLp(unreferenced);
14752 if (SvREFCNT(*svp) == 1)
14754 } while (++svp <= last);
14756 EXTEND_MORTAL(count);
14757 svp = AvARRAY(unreferenced);
14760 if (SvREFCNT(*svp) == 1) {
14761 /* Our reference is the only one to this SV. This means that
14762 in this thread, the scalar effectively has a 0 reference.
14763 That doesn't work (cleanup never happens), so donate our
14764 reference to it onto the save stack. */
14765 PL_tmps_stack[++PL_tmps_ix] = *svp;
14767 /* As an optimisation, because we are already walking the
14768 entire array, instead of above doing either
14769 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14770 release our reference to the scalar, so that at the end of
14771 the array owns zero references to the scalars it happens to
14772 point to. We are effectively converting the array from
14773 AvREAL() on to AvREAL() off. This saves the av_clear()
14774 (triggered by the SvREFCNT_dec(unreferenced) below) from
14775 walking the array a second time. */
14776 SvREFCNT_dec(*svp);
14779 } while (++svp <= last);
14780 AvREAL_off(unreferenced);
14782 SvREFCNT_dec_NN(unreferenced);
14786 Perl_clone_params_del(CLONE_PARAMS *param)
14788 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14790 PerlInterpreter *const to = param->new_perl;
14792 PerlInterpreter *const was = PERL_GET_THX;
14794 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14800 SvREFCNT_dec(param->stashes);
14801 if (param->unreferenced)
14802 unreferenced_to_tmp_stack(param->unreferenced);
14812 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14815 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14816 does a dTHX; to get the context from thread local storage.
14817 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14818 a version that passes in my_perl. */
14819 PerlInterpreter *const was = PERL_GET_THX;
14820 CLONE_PARAMS *param;
14822 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14828 /* Given that we've set the context, we can do this unshared. */
14829 Newx(param, 1, CLONE_PARAMS);
14832 param->proto_perl = from;
14833 param->new_perl = to;
14834 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14835 AvREAL_off(param->stashes);
14836 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14844 #endif /* USE_ITHREADS */
14847 Perl_init_constants(pTHX)
14849 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14850 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVt_NULL;
14851 SvANY(&PL_sv_undef) = NULL;
14853 SvANY(&PL_sv_no) = new_XPVNV();
14854 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14855 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY
14856 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14859 SvANY(&PL_sv_yes) = new_XPVNV();
14860 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14861 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY
14862 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14865 SvPV_set(&PL_sv_no, (char*)PL_No);
14866 SvCUR_set(&PL_sv_no, 0);
14867 SvLEN_set(&PL_sv_no, 0);
14868 SvIV_set(&PL_sv_no, 0);
14869 SvNV_set(&PL_sv_no, 0);
14871 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14872 SvCUR_set(&PL_sv_yes, 1);
14873 SvLEN_set(&PL_sv_yes, 0);
14874 SvIV_set(&PL_sv_yes, 1);
14875 SvNV_set(&PL_sv_yes, 1);
14879 =head1 Unicode Support
14881 =for apidoc sv_recode_to_utf8
14883 The encoding is assumed to be an Encode object, on entry the PV
14884 of the sv is assumed to be octets in that encoding, and the sv
14885 will be converted into Unicode (and UTF-8).
14887 If the sv already is UTF-8 (or if it is not POK), or if the encoding
14888 is not a reference, nothing is done to the sv. If the encoding is not
14889 an C<Encode::XS> Encoding object, bad things will happen.
14890 (See F<lib/encoding.pm> and L<Encode>.)
14892 The PV of the sv is returned.
14897 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
14899 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
14901 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
14910 if (SvPADTMP(nsv)) {
14911 nsv = sv_newmortal();
14912 SvSetSV_nosteal(nsv, sv);
14920 Passing sv_yes is wrong - it needs to be or'ed set of constants
14921 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
14922 remove converted chars from source.
14924 Both will default the value - let them.
14926 XPUSHs(&PL_sv_yes);
14929 call_method("decode", G_SCALAR);
14933 s = SvPV_const(uni, len);
14934 if (s != SvPVX_const(sv)) {
14935 SvGROW(sv, len + 1);
14936 Move(s, SvPVX(sv), len + 1, char);
14937 SvCUR_set(sv, len);
14942 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
14943 /* clear pos and any utf8 cache */
14944 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
14947 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
14948 magic_setutf8(sv,mg); /* clear UTF8 cache */
14953 return SvPOKp(sv) ? SvPVX(sv) : NULL;
14957 =for apidoc sv_cat_decode
14959 The encoding is assumed to be an Encode object, the PV of the ssv is
14960 assumed to be octets in that encoding and decoding the input starts
14961 from the position which (PV + *offset) pointed to. The dsv will be
14962 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
14963 when the string tstr appears in decoding output or the input ends on
14964 the PV of the ssv. The value which the offset points will be modified
14965 to the last input position on the ssv.
14967 Returns TRUE if the terminator was found, else returns FALSE.
14972 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
14973 SV *ssv, int *offset, char *tstr, int tlen)
14977 PERL_ARGS_ASSERT_SV_CAT_DECODE;
14979 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
14989 offsv = newSViv(*offset);
14991 mPUSHp(tstr, tlen);
14993 call_method("cat_decode", G_SCALAR);
14995 ret = SvTRUE(TOPs);
14996 *offset = SvIV(offsv);
15002 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15007 /* ---------------------------------------------------------------------
15009 * support functions for report_uninit()
15012 /* the maxiumum size of array or hash where we will scan looking
15013 * for the undefined element that triggered the warning */
15015 #define FUV_MAX_SEARCH_SIZE 1000
15017 /* Look for an entry in the hash whose value has the same SV as val;
15018 * If so, return a mortal copy of the key. */
15021 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15027 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15029 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15030 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15033 array = HvARRAY(hv);
15035 for (i=HvMAX(hv); i>=0; i--) {
15037 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15038 if (HeVAL(entry) != val)
15040 if ( HeVAL(entry) == &PL_sv_undef ||
15041 HeVAL(entry) == &PL_sv_placeholder)
15045 if (HeKLEN(entry) == HEf_SVKEY)
15046 return sv_mortalcopy(HeKEY_sv(entry));
15047 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15053 /* Look for an entry in the array whose value has the same SV as val;
15054 * If so, return the index, otherwise return -1. */
15057 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15059 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15061 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15062 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15065 if (val != &PL_sv_undef) {
15066 SV ** const svp = AvARRAY(av);
15069 for (i=AvFILLp(av); i>=0; i--)
15076 /* varname(): return the name of a variable, optionally with a subscript.
15077 * If gv is non-zero, use the name of that global, along with gvtype (one
15078 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15079 * targ. Depending on the value of the subscript_type flag, return:
15082 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15083 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15084 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15085 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15088 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15089 const SV *const keyname, I32 aindex, int subscript_type)
15092 SV * const name = sv_newmortal();
15093 if (gv && isGV(gv)) {
15095 buffer[0] = gvtype;
15098 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15100 gv_fullname4(name, gv, buffer, 0);
15102 if ((unsigned int)SvPVX(name)[1] <= 26) {
15104 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15106 /* Swap the 1 unprintable control character for the 2 byte pretty
15107 version - ie substr($name, 1, 1) = $buffer; */
15108 sv_insert(name, 1, 1, buffer, 2);
15112 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15116 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15118 if (!cv || !CvPADLIST(cv))
15120 av = *PadlistARRAY(CvPADLIST(cv));
15121 sv = *av_fetch(av, targ, FALSE);
15122 sv_setsv_flags(name, sv, 0);
15125 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15126 SV * const sv = newSV(0);
15127 *SvPVX(name) = '$';
15128 Perl_sv_catpvf(aTHX_ name, "{%s}",
15129 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15130 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15131 SvREFCNT_dec_NN(sv);
15133 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15134 *SvPVX(name) = '$';
15135 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15137 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15138 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15139 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15147 =for apidoc find_uninit_var
15149 Find the name of the undefined variable (if any) that caused the operator
15150 to issue a "Use of uninitialized value" warning.
15151 If match is true, only return a name if its value matches uninit_sv.
15152 So roughly speaking, if a unary operator (such as OP_COS) generates a
15153 warning, then following the direct child of the op may yield an
15154 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15155 other hand, with OP_ADD there are two branches to follow, so we only print
15156 the variable name if we get an exact match.
15158 The name is returned as a mortal SV.
15160 Assumes that PL_op is the op that originally triggered the error, and that
15161 PL_comppad/PL_curpad points to the currently executing pad.
15167 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15173 const OP *o, *o2, *kid;
15175 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15176 uninit_sv == &PL_sv_placeholder)))
15179 switch (obase->op_type) {
15186 const bool pad = ( obase->op_type == OP_PADAV
15187 || obase->op_type == OP_PADHV
15188 || obase->op_type == OP_PADRANGE
15191 const bool hash = ( obase->op_type == OP_PADHV
15192 || obase->op_type == OP_RV2HV
15193 || (obase->op_type == OP_PADRANGE
15194 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15198 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15200 if (pad) { /* @lex, %lex */
15201 sv = PAD_SVl(obase->op_targ);
15205 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15206 /* @global, %global */
15207 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15210 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15212 else if (obase == PL_op) /* @{expr}, %{expr} */
15213 return find_uninit_var(cUNOPx(obase)->op_first,
15215 else /* @{expr}, %{expr} as a sub-expression */
15219 /* attempt to find a match within the aggregate */
15221 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15223 subscript_type = FUV_SUBSCRIPT_HASH;
15226 index = find_array_subscript((const AV *)sv, uninit_sv);
15228 subscript_type = FUV_SUBSCRIPT_ARRAY;
15231 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15234 return varname(gv, hash ? '%' : '@', obase->op_targ,
15235 keysv, index, subscript_type);
15239 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15241 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15242 if (!gv || !GvSTASH(gv))
15244 if (match && (GvSV(gv) != uninit_sv))
15246 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15249 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
15252 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15254 return varname(NULL, '$', obase->op_targ,
15255 NULL, 0, FUV_SUBSCRIPT_NONE);
15258 gv = cGVOPx_gv(obase);
15259 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15261 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15263 case OP_AELEMFAST_LEX:
15266 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15267 if (!av || SvRMAGICAL(av))
15269 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15270 if (!svp || *svp != uninit_sv)
15273 return varname(NULL, '$', obase->op_targ,
15274 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15277 gv = cGVOPx_gv(obase);
15282 AV *const av = GvAV(gv);
15283 if (!av || SvRMAGICAL(av))
15285 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15286 if (!svp || *svp != uninit_sv)
15289 return varname(gv, '$', 0,
15290 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15292 NOT_REACHED; /* NOTREACHED */
15295 o = cUNOPx(obase)->op_first;
15296 if (!o || o->op_type != OP_NULL ||
15297 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15299 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
15304 bool negate = FALSE;
15306 if (PL_op == obase)
15307 /* $a[uninit_expr] or $h{uninit_expr} */
15308 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
15311 o = cBINOPx(obase)->op_first;
15312 kid = cBINOPx(obase)->op_last;
15314 /* get the av or hv, and optionally the gv */
15316 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15317 sv = PAD_SV(o->op_targ);
15319 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15320 && cUNOPo->op_first->op_type == OP_GV)
15322 gv = cGVOPx_gv(cUNOPo->op_first);
15326 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15331 if (kid && kid->op_type == OP_NEGATE) {
15333 kid = cUNOPx(kid)->op_first;
15336 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15337 /* index is constant */
15340 kidsv = sv_2mortal(newSVpvs("-"));
15341 sv_catsv(kidsv, cSVOPx_sv(kid));
15344 kidsv = cSVOPx_sv(kid);
15348 if (obase->op_type == OP_HELEM) {
15349 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15350 if (!he || HeVAL(he) != uninit_sv)
15354 SV * const opsv = cSVOPx_sv(kid);
15355 const IV opsviv = SvIV(opsv);
15356 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15357 negate ? - opsviv : opsviv,
15359 if (!svp || *svp != uninit_sv)
15363 if (obase->op_type == OP_HELEM)
15364 return varname(gv, '%', o->op_targ,
15365 kidsv, 0, FUV_SUBSCRIPT_HASH);
15367 return varname(gv, '@', o->op_targ, NULL,
15368 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15369 FUV_SUBSCRIPT_ARRAY);
15372 /* index is an expression;
15373 * attempt to find a match within the aggregate */
15374 if (obase->op_type == OP_HELEM) {
15375 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15377 return varname(gv, '%', o->op_targ,
15378 keysv, 0, FUV_SUBSCRIPT_HASH);
15382 = find_array_subscript((const AV *)sv, uninit_sv);
15384 return varname(gv, '@', o->op_targ,
15385 NULL, index, FUV_SUBSCRIPT_ARRAY);
15390 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15392 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15394 NOT_REACHED; /* NOTREACHED */
15398 /* only examine RHS */
15399 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
15402 o = cUNOPx(obase)->op_first;
15403 if ( o->op_type == OP_PUSHMARK
15404 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15408 if (!OP_HAS_SIBLING(o)) {
15409 /* one-arg version of open is highly magical */
15411 if (o->op_type == OP_GV) { /* open FOO; */
15413 if (match && GvSV(gv) != uninit_sv)
15415 return varname(gv, '$', 0,
15416 NULL, 0, FUV_SUBSCRIPT_NONE);
15418 /* other possibilities not handled are:
15419 * open $x; or open my $x; should return '${*$x}'
15420 * open expr; should return '$'.expr ideally
15426 /* ops where $_ may be an implicit arg */
15431 if ( !(obase->op_flags & OPf_STACKED)) {
15432 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
15433 ? PAD_SVl(obase->op_targ)
15436 sv = sv_newmortal();
15437 sv_setpvs(sv, "$_");
15446 match = 1; /* print etc can return undef on defined args */
15447 /* skip filehandle as it can't produce 'undef' warning */
15448 o = cUNOPx(obase)->op_first;
15449 if ((obase->op_flags & OPf_STACKED)
15451 ( o->op_type == OP_PUSHMARK
15452 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15453 o = OP_SIBLING(OP_SIBLING(o));
15457 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
15458 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
15460 /* the following ops are capable of returning PL_sv_undef even for
15461 * defined arg(s) */
15480 case OP_GETPEERNAME:
15528 case OP_SMARTMATCH:
15537 /* XXX tmp hack: these two may call an XS sub, and currently
15538 XS subs don't have a SUB entry on the context stack, so CV and
15539 pad determination goes wrong, and BAD things happen. So, just
15540 don't try to determine the value under those circumstances.
15541 Need a better fix at dome point. DAPM 11/2007 */
15547 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
15548 if (gv && GvSV(gv) == uninit_sv)
15549 return newSVpvs_flags("$.", SVs_TEMP);
15554 /* def-ness of rval pos() is independent of the def-ness of its arg */
15555 if ( !(obase->op_flags & OPf_MOD))
15560 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
15561 return newSVpvs_flags("${$/}", SVs_TEMP);
15566 if (!(obase->op_flags & OPf_KIDS))
15568 o = cUNOPx(obase)->op_first;
15574 /* This loop checks all the kid ops, skipping any that cannot pos-
15575 * sibly be responsible for the uninitialized value; i.e., defined
15576 * constants and ops that return nothing. If there is only one op
15577 * left that is not skipped, then we *know* it is responsible for
15578 * the uninitialized value. If there is more than one op left, we
15579 * have to look for an exact match in the while() loop below.
15580 * Note that we skip padrange, because the individual pad ops that
15581 * it replaced are still in the tree, so we work on them instead.
15584 for (kid=o; kid; kid = OP_SIBLING(kid)) {
15585 const OPCODE type = kid->op_type;
15586 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
15587 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
15588 || (type == OP_PUSHMARK)
15589 || (type == OP_PADRANGE)
15593 if (o2) { /* more than one found */
15600 return find_uninit_var(o2, uninit_sv, match);
15602 /* scan all args */
15604 sv = find_uninit_var(o, uninit_sv, 1);
15616 =for apidoc report_uninit
15618 Print appropriate "Use of uninitialized variable" warning.
15624 Perl_report_uninit(pTHX_ const SV *uninit_sv)
15627 SV* varname = NULL;
15628 if (uninit_sv && PL_curpad) {
15629 varname = find_uninit_var(PL_op, uninit_sv,0);
15631 sv_insert(varname, 0, 0, " ", 1);
15633 /* PL_warn_uninit_sv is constant */
15634 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15635 /* diag_listed_as: Use of uninitialized value%s */
15636 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
15637 SVfARG(varname ? varname : &PL_sv_no),
15638 " in ", OP_DESC(PL_op));
15642 /* PL_warn_uninit is constant */
15643 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15644 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15652 * c-indentation-style: bsd
15653 * c-basic-offset: 4
15654 * indent-tabs-mode: nil
15657 * ex: set ts=8 sts=4 sw=4 et: