3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%" UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referent = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referent = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referent will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referent;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1529 /* forward declaration */
1530 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1536 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1537 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1538 Use the C<SvGROW> wrapper instead.
1545 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1549 PERL_ARGS_ASSERT_SV_GROW;
1553 if (SvTYPE(sv) < SVt_PV) {
1554 sv_upgrade(sv, SVt_PV);
1555 s = SvPVX_mutable(sv);
1557 else if (SvOOK(sv)) { /* pv is offset? */
1559 s = SvPVX_mutable(sv);
1560 if (newlen > SvLEN(sv))
1561 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1565 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1566 s = SvPVX_mutable(sv);
1569 #ifdef PERL_COPY_ON_WRITE
1570 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1571 * to store the COW count. So in general, allocate one more byte than
1572 * asked for, to make it likely this byte is always spare: and thus
1573 * make more strings COW-able.
1575 * Only increment if the allocation isn't MEM_SIZE_MAX,
1576 * otherwise it will wrap to 0.
1578 if ( newlen != MEM_SIZE_MAX )
1582 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1583 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586 if (newlen > SvLEN(sv)) { /* need more room? */
1587 STRLEN minlen = SvCUR(sv);
1588 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1589 if (newlen < minlen)
1591 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1593 /* Don't round up on the first allocation, as odds are pretty good that
1594 * the initial request is accurate as to what is really needed */
1596 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1597 if (rounded > newlen)
1601 if (SvLEN(sv) && s) {
1602 s = (char*)saferealloc(s, newlen);
1605 s = (char*)safemalloc(newlen);
1606 if (SvPVX_const(sv) && SvCUR(sv)) {
1607 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1611 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Do this here, do it once, do it right, and then we will never get
1613 called back into sv_grow() unless there really is some growing
1615 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1617 SvLEN_set(sv, newlen);
1624 =for apidoc sv_setiv
1626 Copies an integer into the given SV, upgrading first if necessary.
1627 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1633 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1635 PERL_ARGS_ASSERT_SV_SETIV;
1637 SV_CHECK_THINKFIRST_COW_DROP(sv);
1638 switch (SvTYPE(sv)) {
1641 sv_upgrade(sv, SVt_IV);
1644 sv_upgrade(sv, SVt_PVIV);
1648 if (!isGV_with_GP(sv))
1655 /* diag_listed_as: Can't coerce %s to %s in %s */
1656 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1658 NOT_REACHED; /* NOTREACHED */
1662 (void)SvIOK_only(sv); /* validate number */
1668 =for apidoc sv_setiv_mg
1670 Like C<sv_setiv>, but also handles 'set' magic.
1676 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1678 PERL_ARGS_ASSERT_SV_SETIV_MG;
1685 =for apidoc sv_setuv
1687 Copies an unsigned integer into the given SV, upgrading first if necessary.
1688 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1694 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1696 PERL_ARGS_ASSERT_SV_SETUV;
1698 /* With the if statement to ensure that integers are stored as IVs whenever
1700 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1703 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1705 If you wish to remove the following if statement, so that this routine
1706 (and its callers) always return UVs, please benchmark to see what the
1707 effect is. Modern CPUs may be different. Or may not :-)
1709 if (u <= (UV)IV_MAX) {
1710 sv_setiv(sv, (IV)u);
1719 =for apidoc sv_setuv_mg
1721 Like C<sv_setuv>, but also handles 'set' magic.
1727 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1729 PERL_ARGS_ASSERT_SV_SETUV_MG;
1736 =for apidoc sv_setnv
1738 Copies a double into the given SV, upgrading first if necessary.
1739 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1745 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1747 PERL_ARGS_ASSERT_SV_SETNV;
1749 SV_CHECK_THINKFIRST_COW_DROP(sv);
1750 switch (SvTYPE(sv)) {
1753 sv_upgrade(sv, SVt_NV);
1757 sv_upgrade(sv, SVt_PVNV);
1761 if (!isGV_with_GP(sv))
1768 /* diag_listed_as: Can't coerce %s to %s in %s */
1769 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1771 NOT_REACHED; /* NOTREACHED */
1776 (void)SvNOK_only(sv); /* validate number */
1781 =for apidoc sv_setnv_mg
1783 Like C<sv_setnv>, but also handles 'set' magic.
1789 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1791 PERL_ARGS_ASSERT_SV_SETNV_MG;
1797 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1798 * not incrementable warning display.
1799 * Originally part of S_not_a_number().
1800 * The return value may be != tmpbuf.
1804 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1807 PERL_ARGS_ASSERT_SV_DISPLAY;
1810 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1811 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1814 const char * const limit = tmpbuf + tmpbuf_size - 8;
1815 /* each *s can expand to 4 chars + "...\0",
1816 i.e. need room for 8 chars */
1818 const char *s = SvPVX_const(sv);
1819 const char * const end = s + SvCUR(sv);
1820 for ( ; s < end && d < limit; s++ ) {
1822 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1826 /* Map to ASCII "equivalent" of Latin1 */
1827 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1833 else if (ch == '\r') {
1837 else if (ch == '\f') {
1841 else if (ch == '\\') {
1845 else if (ch == '\0') {
1849 else if (isPRINT_LC(ch))
1868 /* Print an "isn't numeric" warning, using a cleaned-up,
1869 * printable version of the offending string
1873 S_not_a_number(pTHX_ SV *const sv)
1878 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1880 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1883 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1884 /* diag_listed_as: Argument "%s" isn't numeric%s */
1885 "Argument \"%s\" isn't numeric in %s", pv,
1888 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1889 /* diag_listed_as: Argument "%s" isn't numeric%s */
1890 "Argument \"%s\" isn't numeric", pv);
1894 S_not_incrementable(pTHX_ SV *const sv) {
1898 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1900 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1902 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1903 "Argument \"%s\" treated as 0 in increment (++)", pv);
1907 =for apidoc looks_like_number
1909 Test if the content of an SV looks like a number (or is a number).
1910 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1911 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1918 Perl_looks_like_number(pTHX_ SV *const sv)
1924 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1926 if (SvPOK(sv) || SvPOKp(sv)) {
1927 sbegin = SvPV_nomg_const(sv, len);
1930 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1931 numtype = grok_number(sbegin, len, NULL);
1932 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1936 S_glob_2number(pTHX_ GV * const gv)
1938 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1940 /* We know that all GVs stringify to something that is not-a-number,
1941 so no need to test that. */
1942 if (ckWARN(WARN_NUMERIC))
1944 SV *const buffer = sv_newmortal();
1945 gv_efullname3(buffer, gv, "*");
1946 not_a_number(buffer);
1948 /* We just want something true to return, so that S_sv_2iuv_common
1949 can tail call us and return true. */
1953 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1954 until proven guilty, assume that things are not that bad... */
1959 As 64 bit platforms often have an NV that doesn't preserve all bits of
1960 an IV (an assumption perl has been based on to date) it becomes necessary
1961 to remove the assumption that the NV always carries enough precision to
1962 recreate the IV whenever needed, and that the NV is the canonical form.
1963 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1964 precision as a side effect of conversion (which would lead to insanity
1965 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1966 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1967 where precision was lost, and IV/UV/NV slots that have a valid conversion
1968 which has lost no precision
1969 2) to ensure that if a numeric conversion to one form is requested that
1970 would lose precision, the precise conversion (or differently
1971 imprecise conversion) is also performed and cached, to prevent
1972 requests for different numeric formats on the same SV causing
1973 lossy conversion chains. (lossless conversion chains are perfectly
1978 SvIOKp is true if the IV slot contains a valid value
1979 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1980 SvNOKp is true if the NV slot contains a valid value
1981 SvNOK is true only if the NV value is accurate
1984 while converting from PV to NV, check to see if converting that NV to an
1985 IV(or UV) would lose accuracy over a direct conversion from PV to
1986 IV(or UV). If it would, cache both conversions, return NV, but mark
1987 SV as IOK NOKp (ie not NOK).
1989 While converting from PV to IV, check to see if converting that IV to an
1990 NV would lose accuracy over a direct conversion from PV to NV. If it
1991 would, cache both conversions, flag similarly.
1993 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1994 correctly because if IV & NV were set NV *always* overruled.
1995 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1996 changes - now IV and NV together means that the two are interchangeable:
1997 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1999 The benefit of this is that operations such as pp_add know that if
2000 SvIOK is true for both left and right operands, then integer addition
2001 can be used instead of floating point (for cases where the result won't
2002 overflow). Before, floating point was always used, which could lead to
2003 loss of precision compared with integer addition.
2005 * making IV and NV equal status should make maths accurate on 64 bit
2007 * may speed up maths somewhat if pp_add and friends start to use
2008 integers when possible instead of fp. (Hopefully the overhead in
2009 looking for SvIOK and checking for overflow will not outweigh the
2010 fp to integer speedup)
2011 * will slow down integer operations (callers of SvIV) on "inaccurate"
2012 values, as the change from SvIOK to SvIOKp will cause a call into
2013 sv_2iv each time rather than a macro access direct to the IV slot
2014 * should speed up number->string conversion on integers as IV is
2015 favoured when IV and NV are equally accurate
2017 ####################################################################
2018 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2019 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2020 On the other hand, SvUOK is true iff UV.
2021 ####################################################################
2023 Your mileage will vary depending your CPU's relative fp to integer
2027 #ifndef NV_PRESERVES_UV
2028 # define IS_NUMBER_UNDERFLOW_IV 1
2029 # define IS_NUMBER_UNDERFLOW_UV 2
2030 # define IS_NUMBER_IV_AND_UV 2
2031 # define IS_NUMBER_OVERFLOW_IV 4
2032 # define IS_NUMBER_OVERFLOW_UV 5
2034 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2036 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2038 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2044 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2045 PERL_UNUSED_CONTEXT;
2047 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));
2048 if (SvNVX(sv) < (NV)IV_MIN) {
2049 (void)SvIOKp_on(sv);
2051 SvIV_set(sv, IV_MIN);
2052 return IS_NUMBER_UNDERFLOW_IV;
2054 if (SvNVX(sv) > (NV)UV_MAX) {
2055 (void)SvIOKp_on(sv);
2058 SvUV_set(sv, UV_MAX);
2059 return IS_NUMBER_OVERFLOW_UV;
2061 (void)SvIOKp_on(sv);
2063 /* Can't use strtol etc to convert this string. (See truth table in
2065 if (SvNVX(sv) <= (UV)IV_MAX) {
2066 SvIV_set(sv, I_V(SvNVX(sv)));
2067 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2068 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2070 /* Integer is imprecise. NOK, IOKp */
2072 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2075 SvUV_set(sv, U_V(SvNVX(sv)));
2076 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2077 if (SvUVX(sv) == UV_MAX) {
2078 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2079 possibly be preserved by NV. Hence, it must be overflow.
2081 return IS_NUMBER_OVERFLOW_UV;
2083 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2085 /* Integer is imprecise. NOK, IOKp */
2087 return IS_NUMBER_OVERFLOW_IV;
2089 #endif /* !NV_PRESERVES_UV*/
2091 /* If numtype is infnan, set the NV of the sv accordingly.
2092 * If numtype is anything else, try setting the NV using Atof(PV). */
2094 # pragma warning(push)
2095 # pragma warning(disable:4756;disable:4056)
2098 S_sv_setnv(pTHX_ SV* sv, int numtype)
2100 bool pok = cBOOL(SvPOK(sv));
2103 if ((numtype & IS_NUMBER_INFINITY)) {
2104 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2109 if ((numtype & IS_NUMBER_NAN)) {
2110 SvNV_set(sv, NV_NAN);
2115 SvNV_set(sv, Atof(SvPVX_const(sv)));
2116 /* Purposefully no true nok here, since we don't want to blow
2117 * away the possible IOK/UV of an existing sv. */
2120 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2122 SvPOK_on(sv); /* PV is okay, though. */
2126 # pragma warning(pop)
2130 S_sv_2iuv_common(pTHX_ SV *const sv)
2132 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2135 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2136 * without also getting a cached IV/UV from it at the same time
2137 * (ie PV->NV conversion should detect loss of accuracy and cache
2138 * IV or UV at same time to avoid this. */
2139 /* IV-over-UV optimisation - choose to cache IV if possible */
2141 if (SvTYPE(sv) == SVt_NV)
2142 sv_upgrade(sv, SVt_PVNV);
2144 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2145 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2146 certainly cast into the IV range at IV_MAX, whereas the correct
2147 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2149 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2150 if (Perl_isnan(SvNVX(sv))) {
2156 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2157 SvIV_set(sv, I_V(SvNVX(sv)));
2158 if (SvNVX(sv) == (NV) SvIVX(sv)
2159 #ifndef NV_PRESERVES_UV
2160 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2161 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2162 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2163 /* Don't flag it as "accurately an integer" if the number
2164 came from a (by definition imprecise) NV operation, and
2165 we're outside the range of NV integer precision */
2169 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2171 /* scalar has trailing garbage, eg "42a" */
2173 DEBUG_c(PerlIO_printf(Perl_debug_log,
2174 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2180 /* IV not precise. No need to convert from PV, as NV
2181 conversion would already have cached IV if it detected
2182 that PV->IV would be better than PV->NV->IV
2183 flags already correct - don't set public IOK. */
2184 DEBUG_c(PerlIO_printf(Perl_debug_log,
2185 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2190 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2191 but the cast (NV)IV_MIN rounds to a the value less (more
2192 negative) than IV_MIN which happens to be equal to SvNVX ??
2193 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2194 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2195 (NV)UVX == NVX are both true, but the values differ. :-(
2196 Hopefully for 2s complement IV_MIN is something like
2197 0x8000000000000000 which will be exact. NWC */
2200 SvUV_set(sv, U_V(SvNVX(sv)));
2202 (SvNVX(sv) == (NV) SvUVX(sv))
2203 #ifndef NV_PRESERVES_UV
2204 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2205 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2206 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2207 /* Don't flag it as "accurately an integer" if the number
2208 came from a (by definition imprecise) NV operation, and
2209 we're outside the range of NV integer precision */
2215 DEBUG_c(PerlIO_printf(Perl_debug_log,
2216 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2222 else if (SvPOKp(sv)) {
2225 const char *s = SvPVX_const(sv);
2226 const STRLEN cur = SvCUR(sv);
2228 /* short-cut for a single digit string like "1" */
2233 if (SvTYPE(sv) < SVt_PVIV)
2234 sv_upgrade(sv, SVt_PVIV);
2236 SvIV_set(sv, (IV)(c - '0'));
2241 numtype = grok_number(s, cur, &value);
2242 /* We want to avoid a possible problem when we cache an IV/ a UV which
2243 may be later translated to an NV, and the resulting NV is not
2244 the same as the direct translation of the initial string
2245 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2246 be careful to ensure that the value with the .456 is around if the
2247 NV value is requested in the future).
2249 This means that if we cache such an IV/a UV, we need to cache the
2250 NV as well. Moreover, we trade speed for space, and do not
2251 cache the NV if we are sure it's not needed.
2254 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2255 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2256 == IS_NUMBER_IN_UV) {
2257 /* It's definitely an integer, only upgrade to PVIV */
2258 if (SvTYPE(sv) < SVt_PVIV)
2259 sv_upgrade(sv, SVt_PVIV);
2261 } else if (SvTYPE(sv) < SVt_PVNV)
2262 sv_upgrade(sv, SVt_PVNV);
2264 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2265 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2267 S_sv_setnv(aTHX_ sv, numtype);
2271 /* If NVs preserve UVs then we only use the UV value if we know that
2272 we aren't going to call atof() below. If NVs don't preserve UVs
2273 then the value returned may have more precision than atof() will
2274 return, even though value isn't perfectly accurate. */
2275 if ((numtype & (IS_NUMBER_IN_UV
2276 #ifdef NV_PRESERVES_UV
2279 )) == IS_NUMBER_IN_UV) {
2280 /* This won't turn off the public IOK flag if it was set above */
2281 (void)SvIOKp_on(sv);
2283 if (!(numtype & IS_NUMBER_NEG)) {
2285 if (value <= (UV)IV_MAX) {
2286 SvIV_set(sv, (IV)value);
2288 /* it didn't overflow, and it was positive. */
2289 SvUV_set(sv, value);
2293 /* 2s complement assumption */
2294 if (value <= (UV)IV_MIN) {
2295 SvIV_set(sv, value == (UV)IV_MIN
2296 ? IV_MIN : -(IV)value);
2298 /* Too negative for an IV. This is a double upgrade, but
2299 I'm assuming it will be rare. */
2300 if (SvTYPE(sv) < SVt_PVNV)
2301 sv_upgrade(sv, SVt_PVNV);
2305 SvNV_set(sv, -(NV)value);
2306 SvIV_set(sv, IV_MIN);
2310 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2311 will be in the previous block to set the IV slot, and the next
2312 block to set the NV slot. So no else here. */
2314 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2315 != IS_NUMBER_IN_UV) {
2316 /* It wasn't an (integer that doesn't overflow the UV). */
2317 S_sv_setnv(aTHX_ sv, numtype);
2319 if (! numtype && ckWARN(WARN_NUMERIC))
2322 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2323 PTR2UV(sv), SvNVX(sv)));
2325 #ifdef NV_PRESERVES_UV
2326 (void)SvIOKp_on(sv);
2328 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2329 if (Perl_isnan(SvNVX(sv))) {
2335 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2336 SvIV_set(sv, I_V(SvNVX(sv)));
2337 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2340 NOOP; /* Integer is imprecise. NOK, IOKp */
2342 /* UV will not work better than IV */
2344 if (SvNVX(sv) > (NV)UV_MAX) {
2346 /* Integer is inaccurate. NOK, IOKp, is UV */
2347 SvUV_set(sv, UV_MAX);
2349 SvUV_set(sv, U_V(SvNVX(sv)));
2350 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2351 NV preservse UV so can do correct comparison. */
2352 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2355 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2360 #else /* NV_PRESERVES_UV */
2361 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2362 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2363 /* The IV/UV slot will have been set from value returned by
2364 grok_number above. The NV slot has just been set using
2367 assert (SvIOKp(sv));
2369 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2370 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2371 /* Small enough to preserve all bits. */
2372 (void)SvIOKp_on(sv);
2374 SvIV_set(sv, I_V(SvNVX(sv)));
2375 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2377 /* Assumption: first non-preserved integer is < IV_MAX,
2378 this NV is in the preserved range, therefore: */
2379 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2381 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);
2385 0 0 already failed to read UV.
2386 0 1 already failed to read UV.
2387 1 0 you won't get here in this case. IV/UV
2388 slot set, public IOK, Atof() unneeded.
2389 1 1 already read UV.
2390 so there's no point in sv_2iuv_non_preserve() attempting
2391 to use atol, strtol, strtoul etc. */
2393 sv_2iuv_non_preserve (sv, numtype);
2395 sv_2iuv_non_preserve (sv);
2399 #endif /* NV_PRESERVES_UV */
2400 /* It might be more code efficient to go through the entire logic above
2401 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2402 gets complex and potentially buggy, so more programmer efficient
2403 to do it this way, by turning off the public flags: */
2405 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2409 if (isGV_with_GP(sv))
2410 return glob_2number(MUTABLE_GV(sv));
2412 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2414 if (SvTYPE(sv) < SVt_IV)
2415 /* Typically the caller expects that sv_any is not NULL now. */
2416 sv_upgrade(sv, SVt_IV);
2417 /* Return 0 from the caller. */
2424 =for apidoc sv_2iv_flags
2426 Return the integer value of an SV, doing any necessary string
2427 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2428 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2434 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2436 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2438 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2439 && SvTYPE(sv) != SVt_PVFM);
2441 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2447 if (flags & SV_SKIP_OVERLOAD)
2449 tmpstr = AMG_CALLunary(sv, numer_amg);
2450 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2451 return SvIV(tmpstr);
2454 return PTR2IV(SvRV(sv));
2457 if (SvVALID(sv) || isREGEXP(sv)) {
2458 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2459 must not let them cache IVs.
2460 In practice they are extremely unlikely to actually get anywhere
2461 accessible by user Perl code - the only way that I'm aware of is when
2462 a constant subroutine which is used as the second argument to index.
2464 Regexps have no SvIVX and SvNVX fields.
2466 assert(isREGEXP(sv) || SvPOKp(sv));
2469 const char * const ptr =
2470 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2472 = grok_number(ptr, SvCUR(sv), &value);
2474 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2475 == IS_NUMBER_IN_UV) {
2476 /* It's definitely an integer */
2477 if (numtype & IS_NUMBER_NEG) {
2478 if (value < (UV)IV_MIN)
2481 if (value < (UV)IV_MAX)
2486 /* Quite wrong but no good choices. */
2487 if ((numtype & IS_NUMBER_INFINITY)) {
2488 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2489 } else if ((numtype & IS_NUMBER_NAN)) {
2490 return 0; /* So wrong. */
2494 if (ckWARN(WARN_NUMERIC))
2497 return I_V(Atof(ptr));
2501 if (SvTHINKFIRST(sv)) {
2502 if (SvREADONLY(sv) && !SvOK(sv)) {
2503 if (ckWARN(WARN_UNINITIALIZED))
2510 if (S_sv_2iuv_common(aTHX_ sv))
2514 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2515 PTR2UV(sv),SvIVX(sv)));
2516 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2520 =for apidoc sv_2uv_flags
2522 Return the unsigned integer value of an SV, doing any necessary string
2523 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2524 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2530 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2532 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2534 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2540 if (flags & SV_SKIP_OVERLOAD)
2542 tmpstr = AMG_CALLunary(sv, numer_amg);
2543 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2544 return SvUV(tmpstr);
2547 return PTR2UV(SvRV(sv));
2550 if (SvVALID(sv) || isREGEXP(sv)) {
2551 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2552 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2553 Regexps have no SvIVX and SvNVX fields. */
2554 assert(isREGEXP(sv) || SvPOKp(sv));
2557 const char * const ptr =
2558 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2560 = grok_number(ptr, SvCUR(sv), &value);
2562 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2563 == IS_NUMBER_IN_UV) {
2564 /* It's definitely an integer */
2565 if (!(numtype & IS_NUMBER_NEG))
2569 /* Quite wrong but no good choices. */
2570 if ((numtype & IS_NUMBER_INFINITY)) {
2571 return UV_MAX; /* So wrong. */
2572 } else if ((numtype & IS_NUMBER_NAN)) {
2573 return 0; /* So wrong. */
2577 if (ckWARN(WARN_NUMERIC))
2580 return U_V(Atof(ptr));
2584 if (SvTHINKFIRST(sv)) {
2585 if (SvREADONLY(sv) && !SvOK(sv)) {
2586 if (ckWARN(WARN_UNINITIALIZED))
2593 if (S_sv_2iuv_common(aTHX_ sv))
2597 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2598 PTR2UV(sv),SvUVX(sv)));
2599 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2603 =for apidoc sv_2nv_flags
2605 Return the num value of an SV, doing any necessary string or integer
2606 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2607 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2613 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2615 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2617 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2618 && SvTYPE(sv) != SVt_PVFM);
2619 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2620 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2621 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2622 Regexps have no SvIVX and SvNVX fields. */
2624 if (flags & SV_GMAGIC)
2628 if (SvPOKp(sv) && !SvIOKp(sv)) {
2629 ptr = SvPVX_const(sv);
2631 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2632 !grok_number(ptr, SvCUR(sv), NULL))
2638 return (NV)SvUVX(sv);
2640 return (NV)SvIVX(sv);
2646 ptr = RX_WRAPPED((REGEXP *)sv);
2649 assert(SvTYPE(sv) >= SVt_PVMG);
2650 /* This falls through to the report_uninit near the end of the
2652 } else if (SvTHINKFIRST(sv)) {
2657 if (flags & SV_SKIP_OVERLOAD)
2659 tmpstr = AMG_CALLunary(sv, numer_amg);
2660 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2661 return SvNV(tmpstr);
2664 return PTR2NV(SvRV(sv));
2666 if (SvREADONLY(sv) && !SvOK(sv)) {
2667 if (ckWARN(WARN_UNINITIALIZED))
2672 if (SvTYPE(sv) < SVt_NV) {
2673 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2674 sv_upgrade(sv, SVt_NV);
2676 STORE_NUMERIC_LOCAL_SET_STANDARD();
2677 PerlIO_printf(Perl_debug_log,
2678 "0x%" UVxf " num(%" NVgf ")\n",
2679 PTR2UV(sv), SvNVX(sv));
2680 RESTORE_NUMERIC_LOCAL();
2683 else if (SvTYPE(sv) < SVt_PVNV)
2684 sv_upgrade(sv, SVt_PVNV);
2689 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2690 #ifdef NV_PRESERVES_UV
2696 /* Only set the public NV OK flag if this NV preserves the IV */
2697 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2699 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2700 : (SvIVX(sv) == I_V(SvNVX(sv))))
2706 else if (SvPOKp(sv)) {
2708 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2709 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2711 #ifdef NV_PRESERVES_UV
2712 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2713 == IS_NUMBER_IN_UV) {
2714 /* It's definitely an integer */
2715 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2717 S_sv_setnv(aTHX_ sv, numtype);
2724 SvNV_set(sv, Atof(SvPVX_const(sv)));
2725 /* Only set the public NV OK flag if this NV preserves the value in
2726 the PV at least as well as an IV/UV would.
2727 Not sure how to do this 100% reliably. */
2728 /* if that shift count is out of range then Configure's test is
2729 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2731 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2732 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2733 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2734 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2735 /* Can't use strtol etc to convert this string, so don't try.
2736 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2739 /* value has been set. It may not be precise. */
2740 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2741 /* 2s complement assumption for (UV)IV_MIN */
2742 SvNOK_on(sv); /* Integer is too negative. */
2747 if (numtype & IS_NUMBER_NEG) {
2748 /* -IV_MIN is undefined, but we should never reach
2749 * this point with both IS_NUMBER_NEG and value ==
2751 assert(value != (UV)IV_MIN);
2752 SvIV_set(sv, -(IV)value);
2753 } else if (value <= (UV)IV_MAX) {
2754 SvIV_set(sv, (IV)value);
2756 SvUV_set(sv, value);
2760 if (numtype & IS_NUMBER_NOT_INT) {
2761 /* I believe that even if the original PV had decimals,
2762 they are lost beyond the limit of the FP precision.
2763 However, neither is canonical, so both only get p
2764 flags. NWC, 2000/11/25 */
2765 /* Both already have p flags, so do nothing */
2767 const NV nv = SvNVX(sv);
2768 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2769 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2770 if (SvIVX(sv) == I_V(nv)) {
2773 /* It had no "." so it must be integer. */
2777 /* between IV_MAX and NV(UV_MAX).
2778 Could be slightly > UV_MAX */
2780 if (numtype & IS_NUMBER_NOT_INT) {
2781 /* UV and NV both imprecise. */
2783 const UV nv_as_uv = U_V(nv);
2785 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2794 /* It might be more code efficient to go through the entire logic above
2795 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2796 gets complex and potentially buggy, so more programmer efficient
2797 to do it this way, by turning off the public flags: */
2799 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2800 #endif /* NV_PRESERVES_UV */
2803 if (isGV_with_GP(sv)) {
2804 glob_2number(MUTABLE_GV(sv));
2808 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2810 assert (SvTYPE(sv) >= SVt_NV);
2811 /* Typically the caller expects that sv_any is not NULL now. */
2812 /* XXX Ilya implies that this is a bug in callers that assume this
2813 and ideally should be fixed. */
2817 STORE_NUMERIC_LOCAL_SET_STANDARD();
2818 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2819 PTR2UV(sv), SvNVX(sv));
2820 RESTORE_NUMERIC_LOCAL();
2828 Return an SV with the numeric value of the source SV, doing any necessary
2829 reference or overload conversion. The caller is expected to have handled
2836 Perl_sv_2num(pTHX_ SV *const sv)
2838 PERL_ARGS_ASSERT_SV_2NUM;
2843 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2844 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2845 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2846 return sv_2num(tmpsv);
2848 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2851 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2852 * UV as a string towards the end of buf, and return pointers to start and
2855 * We assume that buf is at least TYPE_CHARS(UV) long.
2859 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2861 char *ptr = buf + TYPE_CHARS(UV);
2862 char * const ebuf = ptr;
2865 PERL_ARGS_ASSERT_UIV_2BUF;
2873 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2877 *--ptr = '0' + (char)(uv % 10);
2885 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2886 * infinity or a not-a-number, writes the appropriate strings to the
2887 * buffer, including a zero byte. On success returns the written length,
2888 * excluding the zero byte, on failure (not an infinity, not a nan)
2889 * returns zero, assert-fails on maxlen being too short.
2891 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2892 * shared string constants we point to, instead of generating a new
2893 * string for each instance. */
2895 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2897 assert(maxlen >= 4);
2898 if (Perl_isinf(nv)) {
2900 if (maxlen < 5) /* "-Inf\0" */
2910 else if (Perl_isnan(nv)) {
2914 /* XXX optionally output the payload mantissa bits as
2915 * "(unsigned)" (to match the nan("...") C99 function,
2916 * or maybe as "(0xhhh...)" would make more sense...
2917 * provide a format string so that the user can decide?
2918 * NOTE: would affect the maxlen and assert() logic.*/
2923 assert((s == buffer + 3) || (s == buffer + 4));
2929 =for apidoc sv_2pv_flags
2931 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2932 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2933 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2934 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2940 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2944 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2946 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2947 && SvTYPE(sv) != SVt_PVFM);
2948 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2953 if (flags & SV_SKIP_OVERLOAD)
2955 tmpstr = AMG_CALLunary(sv, string_amg);
2956 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2957 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2959 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2963 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2964 if (flags & SV_CONST_RETURN) {
2965 pv = (char *) SvPVX_const(tmpstr);
2967 pv = (flags & SV_MUTABLE_RETURN)
2968 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2971 *lp = SvCUR(tmpstr);
2973 pv = sv_2pv_flags(tmpstr, lp, flags);
2986 SV *const referent = SvRV(sv);
2990 retval = buffer = savepvn("NULLREF", len);
2991 } else if (SvTYPE(referent) == SVt_REGEXP &&
2992 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2993 amagic_is_enabled(string_amg))) {
2994 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2998 /* If the regex is UTF-8 we want the containing scalar to
2999 have an UTF-8 flag too */
3006 *lp = RX_WRAPLEN(re);
3008 return RX_WRAPPED(re);
3010 const char *const typestr = sv_reftype(referent, 0);
3011 const STRLEN typelen = strlen(typestr);
3012 UV addr = PTR2UV(referent);
3013 const char *stashname = NULL;
3014 STRLEN stashnamelen = 0; /* hush, gcc */
3015 const char *buffer_end;
3017 if (SvOBJECT(referent)) {
3018 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3021 stashname = HEK_KEY(name);
3022 stashnamelen = HEK_LEN(name);
3024 if (HEK_UTF8(name)) {
3030 stashname = "__ANON__";
3033 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3034 + 2 * sizeof(UV) + 2 /* )\0 */;
3036 len = typelen + 3 /* (0x */
3037 + 2 * sizeof(UV) + 2 /* )\0 */;
3040 Newx(buffer, len, char);
3041 buffer_end = retval = buffer + len;
3043 /* Working backwards */
3047 *--retval = PL_hexdigit[addr & 15];
3048 } while (addr >>= 4);
3054 memcpy(retval, typestr, typelen);
3058 retval -= stashnamelen;
3059 memcpy(retval, stashname, stashnamelen);
3061 /* retval may not necessarily have reached the start of the
3063 assert (retval >= buffer);
3065 len = buffer_end - retval - 1; /* -1 for that \0 */
3077 if (flags & SV_MUTABLE_RETURN)
3078 return SvPVX_mutable(sv);
3079 if (flags & SV_CONST_RETURN)
3080 return (char *)SvPVX_const(sv);
3085 /* I'm assuming that if both IV and NV are equally valid then
3086 converting the IV is going to be more efficient */
3087 const U32 isUIOK = SvIsUV(sv);
3088 char buf[TYPE_CHARS(UV)];
3092 if (SvTYPE(sv) < SVt_PVIV)
3093 sv_upgrade(sv, SVt_PVIV);
3094 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3096 /* inlined from sv_setpvn */
3097 s = SvGROW_mutable(sv, len + 1);
3098 Move(ptr, s, len, char);
3103 else if (SvNOK(sv)) {
3104 if (SvTYPE(sv) < SVt_PVNV)
3105 sv_upgrade(sv, SVt_PVNV);
3106 if (SvNVX(sv) == 0.0
3107 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3108 && !Perl_isnan(SvNVX(sv))
3111 s = SvGROW_mutable(sv, 2);
3116 STRLEN size = 5; /* "-Inf\0" */
3118 s = SvGROW_mutable(sv, size);
3119 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3125 /* some Xenix systems wipe out errno here */
3134 5 + /* exponent digits */
3138 s = SvGROW_mutable(sv, size);
3139 #ifndef USE_LOCALE_NUMERIC
3140 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3146 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3147 STORE_LC_NUMERIC_SET_TO_NEEDED();
3149 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3150 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3151 size += SvCUR(PL_numeric_radix_sv) - 1;
3152 s = SvGROW_mutable(sv, size);
3155 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3157 /* If the radix character is UTF-8, and actually is in the
3158 * output, turn on the UTF-8 flag for the scalar */
3160 && SvUTF8(PL_numeric_radix_sv)
3161 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3166 RESTORE_LC_NUMERIC();
3169 /* We don't call SvPOK_on(), because it may come to
3170 * pass that the locale changes so that the
3171 * stringification we just did is no longer correct. We
3172 * will have to re-stringify every time it is needed */
3179 else if (isGV_with_GP(sv)) {
3180 GV *const gv = MUTABLE_GV(sv);
3181 SV *const buffer = sv_newmortal();
3183 gv_efullname3(buffer, gv, "*");
3185 assert(SvPOK(buffer));
3191 *lp = SvCUR(buffer);
3192 return SvPVX(buffer);
3194 else if (isREGEXP(sv)) {
3195 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3196 return RX_WRAPPED((REGEXP *)sv);
3201 if (flags & SV_UNDEF_RETURNS_NULL)
3203 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3205 /* Typically the caller expects that sv_any is not NULL now. */
3206 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3207 sv_upgrade(sv, SVt_PV);
3212 const STRLEN len = s - SvPVX_const(sv);
3217 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3218 PTR2UV(sv),SvPVX_const(sv)));
3219 if (flags & SV_CONST_RETURN)
3220 return (char *)SvPVX_const(sv);
3221 if (flags & SV_MUTABLE_RETURN)
3222 return SvPVX_mutable(sv);
3227 =for apidoc sv_copypv
3229 Copies a stringified representation of the source SV into the
3230 destination SV. Automatically performs any necessary C<mg_get> and
3231 coercion of numeric values into strings. Guaranteed to preserve
3232 C<UTF8> flag even from overloaded objects. Similar in nature to
3233 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3234 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3235 would lose the UTF-8'ness of the PV.
3237 =for apidoc sv_copypv_nomg
3239 Like C<sv_copypv>, but doesn't invoke get magic first.
3241 =for apidoc sv_copypv_flags
3243 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3244 has the C<SV_GMAGIC> bit set.
3250 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3255 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3257 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3258 sv_setpvn(dsv,s,len);
3266 =for apidoc sv_2pvbyte
3268 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3269 to its length. May cause the SV to be downgraded from UTF-8 as a
3272 Usually accessed via the C<SvPVbyte> macro.
3278 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3280 PERL_ARGS_ASSERT_SV_2PVBYTE;
3283 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3284 || isGV_with_GP(sv) || SvROK(sv)) {
3285 SV *sv2 = sv_newmortal();
3286 sv_copypv_nomg(sv2,sv);
3289 sv_utf8_downgrade(sv,0);
3290 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3294 =for apidoc sv_2pvutf8
3296 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3297 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3299 Usually accessed via the C<SvPVutf8> macro.
3305 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3307 PERL_ARGS_ASSERT_SV_2PVUTF8;
3309 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3310 || isGV_with_GP(sv) || SvROK(sv))
3311 sv = sv_mortalcopy(sv);
3314 sv_utf8_upgrade_nomg(sv);
3315 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3320 =for apidoc sv_2bool
3322 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3323 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3324 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3326 =for apidoc sv_2bool_flags
3328 This function is only used by C<sv_true()> and friends, and only if
3329 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3330 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3337 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3339 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3342 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3348 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3349 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3352 if(SvGMAGICAL(sv)) {
3354 goto restart; /* call sv_2bool */
3356 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3357 else if(!SvOK(sv)) {
3360 else if(SvPOK(sv)) {
3361 svb = SvPVXtrue(sv);
3363 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3364 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3365 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3369 goto restart; /* call sv_2bool_nomg */
3374 return SvRV(sv) != 0;
3378 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3379 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3383 =for apidoc sv_utf8_upgrade
3385 Converts the PV of an SV to its UTF-8-encoded form.
3386 Forces the SV to string form if it is not already.
3387 Will C<mg_get> on C<sv> if appropriate.
3388 Always sets the C<SvUTF8> flag to avoid future validity checks even
3389 if the whole string is the same in UTF-8 as not.
3390 Returns the number of bytes in the converted string
3392 This is not a general purpose byte encoding to Unicode interface:
3393 use the Encode extension for that.
3395 =for apidoc sv_utf8_upgrade_nomg
3397 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3399 =for apidoc sv_utf8_upgrade_flags
3401 Converts the PV of an SV to its UTF-8-encoded form.
3402 Forces the SV to string form if it is not already.
3403 Always sets the SvUTF8 flag to avoid future validity checks even
3404 if all the bytes are invariant in UTF-8.
3405 If C<flags> has C<SV_GMAGIC> bit set,
3406 will C<mg_get> on C<sv> if appropriate, else not.
3408 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3409 will expand when converted to UTF-8, and skips the extra work of checking for
3410 that. Typically this flag is used by a routine that has already parsed the
3411 string and found such characters, and passes this information on so that the
3412 work doesn't have to be repeated.
3414 Returns the number of bytes in the converted string.
3416 This is not a general purpose byte encoding to Unicode interface:
3417 use the Encode extension for that.
3419 =for apidoc sv_utf8_upgrade_flags_grow
3421 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3422 the number of unused bytes the string of C<sv> is guaranteed to have free after
3423 it upon return. This allows the caller to reserve extra space that it intends
3424 to fill, to avoid extra grows.
3426 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3427 are implemented in terms of this function.
3429 Returns the number of bytes in the converted string (not including the spares).
3433 (One might think that the calling routine could pass in the position of the
3434 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3435 have to be found again. But that is not the case, because typically when the
3436 caller is likely to use this flag, it won't be calling this routine unless it
3437 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3438 and just use bytes. But some things that do fit into a byte are variants in
3439 utf8, and the caller may not have been keeping track of these.)
3441 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3442 C<NUL> isn't guaranteed due to having other routines do the work in some input
3443 cases, or if the input is already flagged as being in utf8.
3445 The speed of this could perhaps be improved for many cases if someone wanted to
3446 write a fast function that counts the number of variant characters in a string,
3447 especially if it could return the position of the first one.
3452 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3454 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3456 if (sv == &PL_sv_undef)
3458 if (!SvPOK_nog(sv)) {
3460 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3461 (void) sv_2pv_flags(sv,&len, flags);
3463 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3467 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3472 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3477 S_sv_uncow(aTHX_ sv, 0);
3480 if (SvCUR(sv) == 0) {
3481 if (extra) SvGROW(sv, extra);
3482 } else { /* Assume Latin-1/EBCDIC */
3483 /* This function could be much more efficient if we
3484 * had a FLAG in SVs to signal if there are any variant
3485 * chars in the PV. Given that there isn't such a flag
3486 * make the loop as fast as possible (although there are certainly ways
3487 * to speed this up, eg. through vectorization) */
3488 U8 * s = (U8 *) SvPVX_const(sv);
3489 U8 * e = (U8 *) SvEND(sv);
3491 STRLEN two_byte_count;
3493 if (flags & SV_FORCE_UTF8_UPGRADE) {
3497 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3499 /* utf8 conversion not needed because all are invariants. Mark
3500 * as UTF-8 even if no variant - saves scanning loop */
3502 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3506 /* Here, there is at least one variant, and t points to the first
3511 /* Note that the incoming SV may not have a trailing '\0', as certain
3512 * code in pp_formline can send us partially built SVs.
3514 * Here, the string should be converted to utf8, either because of an
3515 * input flag (which causes two_byte_count to be set to 0), or because
3516 * a character that requires 2 bytes was found (two_byte_count = 1). t
3517 * points either to the beginning of the string (if we didn't examine
3518 * anything), or to the first variant. In either case, everything from
3519 * s to t - 1 will occupy only 1 byte each on output.
3521 * There are two main ways to convert. One is to create a new string
3522 * and go through the input starting from the beginning, appending each
3523 * converted value onto the new string as we go along. It's probably
3524 * best to allocate enough space in the string for the worst possible
3525 * case rather than possibly running out of space and having to
3526 * reallocate and then copy what we've done so far. Since everything
3527 * from s to t - 1 is invariant, the destination can be initialized
3528 * with these using a fast memory copy
3530 * The other way is to figure out exactly how big the string should be,
3531 * by parsing the entire input. Then you don't have to make it big
3532 * enough to handle the worst possible case, and more importantly, if
3533 * the string you already have is large enough, you don't have to
3534 * allocate a new string, you can copy the last character in the input
3535 * string to the final position(s) that will be occupied by the
3536 * converted string and go backwards, stopping at t, since everything
3537 * before that is invariant.
3539 * There are advantages and disadvantages to each method.
3541 * In the first method, we can allocate a new string, do the memory
3542 * copy from the s to t - 1, and then proceed through the rest of the
3543 * string byte-by-byte.
3545 * In the second method, we proceed through the rest of the input
3546 * string just calculating how big the converted string will be. Then
3547 * there are two cases:
3548 * 1) if the string has enough extra space to handle the converted
3549 * value. We go backwards through the string, converting until we
3550 * get to the position we are at now, and then stop. If this
3551 * position is far enough along in the string, this method is
3552 * faster than the first method above. If the memory copy were
3553 * the same speed as the byte-by-byte loop, that position would be
3554 * about half-way, as at the half-way mark, parsing to the end and
3555 * back is one complete string's parse, the same amount as
3556 * starting over and going all the way through. Actually, it
3557 * would be somewhat less than half-way, as it's faster to just
3558 * count bytes than to also copy, and we don't have the overhead
3559 * of allocating a new string, changing the scalar to use it, and
3560 * freeing the existing one. But if the memory copy is fast, the
3561 * break-even point is somewhere after half way. The counting
3562 * loop could be sped up by vectorization, etc, to move the
3563 * break-even point further towards the beginning.
3564 * 2) if the string doesn't have enough space to handle the converted
3565 * value. A new string will have to be allocated, and one might
3566 * as well, given that, start from the beginning doing the first
3567 * method. We've spent extra time parsing the string and in
3568 * exchange all we've gotten is that we know precisely how big to
3569 * make the new one. Perl is more optimized for time than space,
3570 * so this case is a loser.
3571 * So what I've decided to do is not use the 2nd method unless it is
3572 * guaranteed that a new string won't have to be allocated, assuming
3573 * the worst case. I also decided not to put any more conditions on it
3574 * than this, for now. It seems likely that, since the worst case is
3575 * twice as big as the unknown portion of the string (plus 1), we won't
3576 * be guaranteed enough space, causing us to go to the first method,
3577 * unless the string is short, or the first variant character is near
3578 * the end of it. In either of these cases, it seems best to use the
3579 * 2nd method. The only circumstance I can think of where this would
3580 * be really slower is if the string had once had much more data in it
3581 * than it does now, but there is still a substantial amount in it */
3584 STRLEN invariant_head = t - s;
3585 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3586 if (SvLEN(sv) < size) {
3588 /* Here, have decided to allocate a new string */
3593 Newx(dst, size, U8);
3595 /* If no known invariants at the beginning of the input string,
3596 * set so starts from there. Otherwise, can use memory copy to
3597 * get up to where we are now, and then start from here */
3599 if (invariant_head == 0) {
3602 Copy(s, dst, invariant_head, char);
3603 d = dst + invariant_head;
3607 append_utf8_from_native_byte(*t, &d);
3611 SvPV_free(sv); /* No longer using pre-existing string */
3612 SvPV_set(sv, (char*)dst);
3613 SvCUR_set(sv, d - dst);
3614 SvLEN_set(sv, size);
3617 /* Here, have decided to get the exact size of the string.
3618 * Currently this happens only when we know that there is
3619 * guaranteed enough space to fit the converted string, so
3620 * don't have to worry about growing. If two_byte_count is 0,
3621 * then t points to the first byte of the string which hasn't
3622 * been examined yet. Otherwise two_byte_count is 1, and t
3623 * points to the first byte in the string that will expand to
3624 * two. Depending on this, start examining at t or 1 after t.
3627 U8 *d = t + two_byte_count;
3630 /* Count up the remaining bytes that expand to two */
3633 const U8 chr = *d++;
3634 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3637 /* The string will expand by just the number of bytes that
3638 * occupy two positions. But we are one afterwards because of
3639 * the increment just above. This is the place to put the
3640 * trailing NUL, and to set the length before we decrement */
3642 d += two_byte_count;
3643 SvCUR_set(sv, d - s);
3647 /* Having decremented d, it points to the position to put the
3648 * very last byte of the expanded string. Go backwards through
3649 * the string, copying and expanding as we go, stopping when we
3650 * get to the part that is invariant the rest of the way down */
3654 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3657 *d-- = UTF8_EIGHT_BIT_LO(*e);
3658 *d-- = UTF8_EIGHT_BIT_HI(*e);
3664 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3665 /* Update pos. We do it at the end rather than during
3666 * the upgrade, to avoid slowing down the common case
3667 * (upgrade without pos).
3668 * pos can be stored as either bytes or characters. Since
3669 * this was previously a byte string we can just turn off
3670 * the bytes flag. */
3671 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3673 mg->mg_flags &= ~MGf_BYTES;
3675 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3676 magic_setutf8(sv,mg); /* clear UTF8 cache */
3681 /* Mark as UTF-8 even if no variant - saves scanning loop */
3687 =for apidoc sv_utf8_downgrade
3689 Attempts to convert the PV of an SV from characters to bytes.
3690 If the PV contains a character that cannot fit
3691 in a byte, this conversion will fail;
3692 in this case, either returns false or, if C<fail_ok> is not
3695 This is not a general purpose Unicode to byte encoding interface:
3696 use the C<Encode> extension for that.
3702 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3704 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3706 if (SvPOKp(sv) && SvUTF8(sv)) {
3710 int mg_flags = SV_GMAGIC;
3713 S_sv_uncow(aTHX_ sv, 0);
3715 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3717 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3718 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3719 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3720 SV_GMAGIC|SV_CONST_RETURN);
3721 mg_flags = 0; /* sv_pos_b2u does get magic */
3723 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3724 magic_setutf8(sv,mg); /* clear UTF8 cache */
3727 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3729 if (!utf8_to_bytes(s, &len)) {
3734 Perl_croak(aTHX_ "Wide character in %s",
3737 Perl_croak(aTHX_ "Wide character");
3748 =for apidoc sv_utf8_encode
3750 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3751 flag off so that it looks like octets again.
3757 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3759 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3761 if (SvREADONLY(sv)) {
3762 sv_force_normal_flags(sv, 0);
3764 (void) sv_utf8_upgrade(sv);
3769 =for apidoc sv_utf8_decode
3771 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3772 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3773 so that it looks like a character. If the PV contains only single-byte
3774 characters, the C<SvUTF8> flag stays off.
3775 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3781 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3783 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3786 const U8 *start, *c;
3788 /* The octets may have got themselves encoded - get them back as
3791 if (!sv_utf8_downgrade(sv, TRUE))
3794 /* it is actually just a matter of turning the utf8 flag on, but
3795 * we want to make sure everything inside is valid utf8 first.
3797 c = start = (const U8 *) SvPVX_const(sv);
3798 if (!is_utf8_string(c, SvCUR(sv)))
3800 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3803 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3804 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3805 after this, clearing pos. Does anything on CPAN
3807 /* adjust pos to the start of a UTF8 char sequence */
3808 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3810 I32 pos = mg->mg_len;
3812 for (c = start + pos; c > start; c--) {
3813 if (UTF8_IS_START(*c))
3816 mg->mg_len = c - start;
3819 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3820 magic_setutf8(sv,mg); /* clear UTF8 cache */
3827 =for apidoc sv_setsv
3829 Copies the contents of the source SV C<ssv> into the destination SV
3830 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3831 function if the source SV needs to be reused. Does not handle 'set' magic on
3832 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3833 performs a copy-by-value, obliterating any previous content of the
3836 You probably want to use one of the assortment of wrappers, such as
3837 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3838 C<SvSetMagicSV_nosteal>.
3840 =for apidoc sv_setsv_flags
3842 Copies the contents of the source SV C<ssv> into the destination SV
3843 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3844 function if the source SV needs to be reused. Does not handle 'set' magic.
3845 Loosely speaking, it performs a copy-by-value, obliterating any previous
3846 content of the destination.
3847 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3848 C<ssv> if appropriate, else not. If the C<flags>
3849 parameter has the C<SV_NOSTEAL> bit set then the
3850 buffers of temps will not be stolen. C<sv_setsv>
3851 and C<sv_setsv_nomg> are implemented in terms of this function.
3853 You probably want to use one of the assortment of wrappers, such as
3854 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3855 C<SvSetMagicSV_nosteal>.
3857 This is the primary function for copying scalars, and most other
3858 copy-ish functions and macros use this underneath.
3864 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3866 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3867 HV *old_stash = NULL;
3869 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3871 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3872 const char * const name = GvNAME(sstr);
3873 const STRLEN len = GvNAMELEN(sstr);
3875 if (dtype >= SVt_PV) {
3881 SvUPGRADE(dstr, SVt_PVGV);
3882 (void)SvOK_off(dstr);
3883 isGV_with_GP_on(dstr);
3885 GvSTASH(dstr) = GvSTASH(sstr);
3887 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3888 gv_name_set(MUTABLE_GV(dstr), name, len,
3889 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3890 SvFAKE_on(dstr); /* can coerce to non-glob */
3893 if(GvGP(MUTABLE_GV(sstr))) {
3894 /* If source has method cache entry, clear it */
3896 SvREFCNT_dec(GvCV(sstr));
3897 GvCV_set(sstr, NULL);
3900 /* If source has a real method, then a method is
3903 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3909 /* If dest already had a real method, that's a change as well */
3911 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3912 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3917 /* We don't need to check the name of the destination if it was not a
3918 glob to begin with. */
3919 if(dtype == SVt_PVGV) {
3920 const char * const name = GvNAME((const GV *)dstr);
3923 /* The stash may have been detached from the symbol table, so
3925 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3929 const STRLEN len = GvNAMELEN(dstr);
3930 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3931 || (len == 1 && name[0] == ':')) {
3934 /* Set aside the old stash, so we can reset isa caches on
3936 if((old_stash = GvHV(dstr)))
3937 /* Make sure we do not lose it early. */
3938 SvREFCNT_inc_simple_void_NN(
3939 sv_2mortal((SV *)old_stash)
3944 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3947 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3948 * so temporarily protect it */
3950 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3951 gp_free(MUTABLE_GV(dstr));
3952 GvINTRO_off(dstr); /* one-shot flag */
3953 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3956 if (SvTAINTED(sstr))
3958 if (GvIMPORTED(dstr) != GVf_IMPORTED
3959 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3961 GvIMPORTED_on(dstr);
3964 if(mro_changes == 2) {
3965 if (GvAV((const GV *)sstr)) {
3967 SV * const sref = (SV *)GvAV((const GV *)dstr);
3968 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3969 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3970 AV * const ary = newAV();
3971 av_push(ary, mg->mg_obj); /* takes the refcount */
3972 mg->mg_obj = (SV *)ary;
3974 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3976 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3978 mro_isa_changed_in(GvSTASH(dstr));
3980 else if(mro_changes == 3) {
3981 HV * const stash = GvHV(dstr);
3982 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3988 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3989 if (GvIO(dstr) && dtype == SVt_PVGV) {
3990 DEBUG_o(Perl_deb(aTHX_
3991 "glob_assign_glob clearing PL_stashcache\n"));
3992 /* It's a cache. It will rebuild itself quite happily.
3993 It's a lot of effort to work out exactly which key (or keys)
3994 might be invalidated by the creation of the this file handle.
3996 hv_clear(PL_stashcache);
4002 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4004 SV * const sref = SvRV(sstr);
4006 const int intro = GvINTRO(dstr);
4009 const U32 stype = SvTYPE(sref);
4011 PERL_ARGS_ASSERT_GV_SETREF;
4014 GvINTRO_off(dstr); /* one-shot flag */
4015 GvLINE(dstr) = CopLINE(PL_curcop);
4016 GvEGV(dstr) = MUTABLE_GV(dstr);
4021 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4022 import_flag = GVf_IMPORTED_CV;
4025 location = (SV **) &GvHV(dstr);
4026 import_flag = GVf_IMPORTED_HV;
4029 location = (SV **) &GvAV(dstr);
4030 import_flag = GVf_IMPORTED_AV;
4033 location = (SV **) &GvIOp(dstr);
4036 location = (SV **) &GvFORM(dstr);
4039 location = &GvSV(dstr);
4040 import_flag = GVf_IMPORTED_SV;
4043 if (stype == SVt_PVCV) {
4044 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4045 if (GvCVGEN(dstr)) {
4046 SvREFCNT_dec(GvCV(dstr));
4047 GvCV_set(dstr, NULL);
4048 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4051 /* SAVEt_GVSLOT takes more room on the savestack and has more
4052 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4053 leave_scope needs access to the GV so it can reset method
4054 caches. We must use SAVEt_GVSLOT whenever the type is
4055 SVt_PVCV, even if the stash is anonymous, as the stash may
4056 gain a name somehow before leave_scope. */
4057 if (stype == SVt_PVCV) {
4058 /* There is no save_pushptrptrptr. Creating it for this
4059 one call site would be overkill. So inline the ss add
4063 SS_ADD_PTR(location);
4064 SS_ADD_PTR(SvREFCNT_inc(*location));
4065 SS_ADD_UV(SAVEt_GVSLOT);
4068 else SAVEGENERICSV(*location);
4071 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4072 CV* const cv = MUTABLE_CV(*location);
4074 if (!GvCVGEN((const GV *)dstr) &&
4075 (CvROOT(cv) || CvXSUB(cv)) &&
4076 /* redundant check that avoids creating the extra SV
4077 most of the time: */
4078 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4080 SV * const new_const_sv =
4081 CvCONST((const CV *)sref)
4082 ? cv_const_sv((const CV *)sref)
4084 HV * const stash = GvSTASH((const GV *)dstr);
4085 report_redefined_cv(
4088 ? Perl_newSVpvf(aTHX_
4089 "%" HEKf "::%" HEKf,
4090 HEKfARG(HvNAME_HEK(stash)),
4091 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4092 : Perl_newSVpvf(aTHX_
4094 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4097 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4101 cv_ckproto_len_flags(cv, (const GV *)dstr,
4102 SvPOK(sref) ? CvPROTO(sref) : NULL,
4103 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4104 SvPOK(sref) ? SvUTF8(sref) : 0);
4106 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4107 GvASSUMECV_on(dstr);
4108 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4109 if (intro && GvREFCNT(dstr) > 1) {
4110 /* temporary remove extra savestack's ref */
4112 gv_method_changed(dstr);
4115 else gv_method_changed(dstr);
4118 *location = SvREFCNT_inc_simple_NN(sref);
4119 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4120 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4121 GvFLAGS(dstr) |= import_flag;
4124 if (stype == SVt_PVHV) {
4125 const char * const name = GvNAME((GV*)dstr);
4126 const STRLEN len = GvNAMELEN(dstr);
4129 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4130 || (len == 1 && name[0] == ':')
4132 && (!dref || HvENAME_get(dref))
4135 (HV *)sref, (HV *)dref,
4141 stype == SVt_PVAV && sref != dref
4142 && strEQ(GvNAME((GV*)dstr), "ISA")
4143 /* The stash may have been detached from the symbol table, so
4144 check its name before doing anything. */
4145 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4148 MAGIC * const omg = dref && SvSMAGICAL(dref)
4149 ? mg_find(dref, PERL_MAGIC_isa)
4151 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4152 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4153 AV * const ary = newAV();
4154 av_push(ary, mg->mg_obj); /* takes the refcount */
4155 mg->mg_obj = (SV *)ary;
4158 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4159 SV **svp = AvARRAY((AV *)omg->mg_obj);
4160 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4164 SvREFCNT_inc_simple_NN(*svp++)
4170 SvREFCNT_inc_simple_NN(omg->mg_obj)
4174 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4180 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4182 for (i = 0; i <= AvFILL(sref); ++i) {
4183 SV **elem = av_fetch ((AV*)sref, i, 0);
4186 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4190 mg = mg_find(sref, PERL_MAGIC_isa);
4192 /* Since the *ISA assignment could have affected more than
4193 one stash, don't call mro_isa_changed_in directly, but let
4194 magic_clearisa do it for us, as it already has the logic for
4195 dealing with globs vs arrays of globs. */
4197 Perl_magic_clearisa(aTHX_ NULL, mg);
4199 else if (stype == SVt_PVIO) {
4200 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4201 /* It's a cache. It will rebuild itself quite happily.
4202 It's a lot of effort to work out exactly which key (or keys)
4203 might be invalidated by the creation of the this file handle.
4205 hv_clear(PL_stashcache);
4209 if (!intro) SvREFCNT_dec(dref);
4210 if (SvTAINTED(sstr))
4218 #ifdef PERL_DEBUG_READONLY_COW
4219 # include <sys/mman.h>
4221 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4222 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4226 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4228 struct perl_memory_debug_header * const header =
4229 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4230 const MEM_SIZE len = header->size;
4231 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4232 # ifdef PERL_TRACK_MEMPOOL
4233 if (!header->readonly) header->readonly = 1;
4235 if (mprotect(header, len, PROT_READ))
4236 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4237 header, len, errno);
4241 S_sv_buf_to_rw(pTHX_ SV *sv)
4243 struct perl_memory_debug_header * const header =
4244 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4245 const MEM_SIZE len = header->size;
4246 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4247 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4248 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4249 header, len, errno);
4250 # ifdef PERL_TRACK_MEMPOOL
4251 header->readonly = 0;
4256 # define sv_buf_to_ro(sv) NOOP
4257 # define sv_buf_to_rw(sv) NOOP
4261 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4266 unsigned int both_type;
4268 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4270 if (UNLIKELY( sstr == dstr ))
4273 if (UNLIKELY( !sstr ))
4274 sstr = &PL_sv_undef;
4276 stype = SvTYPE(sstr);
4277 dtype = SvTYPE(dstr);
4278 both_type = (stype | dtype);
4280 /* with these values, we can check that both SVs are NULL/IV (and not
4281 * freed) just by testing the or'ed types */
4282 STATIC_ASSERT_STMT(SVt_NULL == 0);
4283 STATIC_ASSERT_STMT(SVt_IV == 1);
4284 if (both_type <= 1) {
4285 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4291 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4292 if (SvREADONLY(dstr))
4293 Perl_croak_no_modify();
4295 if (SvWEAKREF(dstr))
4296 sv_unref_flags(dstr, 0);
4298 old_rv = SvRV(dstr);
4301 assert(!SvGMAGICAL(sstr));
4302 assert(!SvGMAGICAL(dstr));
4304 sflags = SvFLAGS(sstr);
4305 if (sflags & (SVf_IOK|SVf_ROK)) {
4306 SET_SVANY_FOR_BODYLESS_IV(dstr);
4307 new_dflags = SVt_IV;
4309 if (sflags & SVf_ROK) {
4310 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4311 new_dflags |= SVf_ROK;
4314 /* both src and dst are <= SVt_IV, so sv_any points to the
4315 * head; so access the head directly
4317 assert( &(sstr->sv_u.svu_iv)
4318 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4319 assert( &(dstr->sv_u.svu_iv)
4320 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4321 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4322 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4326 new_dflags = dtype; /* turn off everything except the type */
4328 SvFLAGS(dstr) = new_dflags;
4329 SvREFCNT_dec(old_rv);
4334 if (UNLIKELY(both_type == SVTYPEMASK)) {
4335 if (SvIS_FREED(dstr)) {
4336 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4337 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4339 if (SvIS_FREED(sstr)) {
4340 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4341 (void*)sstr, (void*)dstr);
4347 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4348 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4350 /* There's a lot of redundancy below but we're going for speed here */
4355 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4356 (void)SvOK_off(dstr);
4364 /* For performance, we inline promoting to type SVt_IV. */
4365 /* We're starting from SVt_NULL, so provided that define is
4366 * actual 0, we don't have to unset any SV type flags
4367 * to promote to SVt_IV. */
4368 STATIC_ASSERT_STMT(SVt_NULL == 0);
4369 SET_SVANY_FOR_BODYLESS_IV(dstr);
4370 SvFLAGS(dstr) |= SVt_IV;
4374 sv_upgrade(dstr, SVt_PVIV);
4378 goto end_of_first_switch;
4380 (void)SvIOK_only(dstr);
4381 SvIV_set(dstr, SvIVX(sstr));
4384 /* SvTAINTED can only be true if the SV has taint magic, which in
4385 turn means that the SV type is PVMG (or greater). This is the
4386 case statement for SVt_IV, so this cannot be true (whatever gcov
4388 assert(!SvTAINTED(sstr));
4393 if (dtype < SVt_PV && dtype != SVt_IV)
4394 sv_upgrade(dstr, SVt_IV);
4398 if (LIKELY( SvNOK(sstr) )) {
4402 sv_upgrade(dstr, SVt_NV);
4406 sv_upgrade(dstr, SVt_PVNV);
4410 goto end_of_first_switch;
4412 SvNV_set(dstr, SvNVX(sstr));
4413 (void)SvNOK_only(dstr);
4414 /* SvTAINTED can only be true if the SV has taint magic, which in
4415 turn means that the SV type is PVMG (or greater). This is the
4416 case statement for SVt_NV, so this cannot be true (whatever gcov
4418 assert(!SvTAINTED(sstr));
4425 sv_upgrade(dstr, SVt_PV);
4428 if (dtype < SVt_PVIV)
4429 sv_upgrade(dstr, SVt_PVIV);
4432 if (dtype < SVt_PVNV)
4433 sv_upgrade(dstr, SVt_PVNV);
4437 const char * const type = sv_reftype(sstr,0);
4439 /* diag_listed_as: Bizarre copy of %s */
4440 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4442 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4444 NOT_REACHED; /* NOTREACHED */
4448 if (dtype < SVt_REGEXP)
4450 if (dtype >= SVt_PV) {
4456 sv_upgrade(dstr, SVt_REGEXP);
4464 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4466 if (SvTYPE(sstr) != stype)
4467 stype = SvTYPE(sstr);
4469 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4470 glob_assign_glob(dstr, sstr, dtype);
4473 if (stype == SVt_PVLV)
4475 if (isREGEXP(sstr)) goto upgregexp;
4476 SvUPGRADE(dstr, SVt_PVNV);
4479 SvUPGRADE(dstr, (svtype)stype);
4481 end_of_first_switch:
4483 /* dstr may have been upgraded. */
4484 dtype = SvTYPE(dstr);
4485 sflags = SvFLAGS(sstr);
4487 if (UNLIKELY( dtype == SVt_PVCV )) {
4488 /* Assigning to a subroutine sets the prototype. */
4491 const char *const ptr = SvPV_const(sstr, len);
4493 SvGROW(dstr, len + 1);
4494 Copy(ptr, SvPVX(dstr), len + 1, char);
4495 SvCUR_set(dstr, len);
4497 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4498 CvAUTOLOAD_off(dstr);
4503 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4504 || dtype == SVt_PVFM))
4506 const char * const type = sv_reftype(dstr,0);
4508 /* diag_listed_as: Cannot copy to %s */
4509 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4511 Perl_croak(aTHX_ "Cannot copy to %s", type);
4512 } else if (sflags & SVf_ROK) {
4513 if (isGV_with_GP(dstr)
4514 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4517 if (GvIMPORTED(dstr) != GVf_IMPORTED
4518 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4520 GvIMPORTED_on(dstr);
4525 glob_assign_glob(dstr, sstr, dtype);
4529 if (dtype >= SVt_PV) {
4530 if (isGV_with_GP(dstr)) {
4531 gv_setref(dstr, sstr);
4534 if (SvPVX_const(dstr)) {
4540 (void)SvOK_off(dstr);
4541 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4542 SvFLAGS(dstr) |= sflags & SVf_ROK;
4543 assert(!(sflags & SVp_NOK));
4544 assert(!(sflags & SVp_IOK));
4545 assert(!(sflags & SVf_NOK));
4546 assert(!(sflags & SVf_IOK));
4548 else if (isGV_with_GP(dstr)) {
4549 if (!(sflags & SVf_OK)) {
4550 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4551 "Undefined value assigned to typeglob");
4554 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4555 if (dstr != (const SV *)gv) {
4556 const char * const name = GvNAME((const GV *)dstr);
4557 const STRLEN len = GvNAMELEN(dstr);
4558 HV *old_stash = NULL;
4559 bool reset_isa = FALSE;
4560 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4561 || (len == 1 && name[0] == ':')) {
4562 /* Set aside the old stash, so we can reset isa caches
4563 on its subclasses. */
4564 if((old_stash = GvHV(dstr))) {
4565 /* Make sure we do not lose it early. */
4566 SvREFCNT_inc_simple_void_NN(
4567 sv_2mortal((SV *)old_stash)
4574 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4575 gp_free(MUTABLE_GV(dstr));
4577 GvGP_set(dstr, gp_ref(GvGP(gv)));
4580 HV * const stash = GvHV(dstr);
4582 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4592 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4593 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4594 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4596 else if (sflags & SVp_POK) {
4597 const STRLEN cur = SvCUR(sstr);
4598 const STRLEN len = SvLEN(sstr);
4601 * We have three basic ways to copy the string:
4607 * Which we choose is based on various factors. The following
4608 * things are listed in order of speed, fastest to slowest:
4610 * - Copying a short string
4611 * - Copy-on-write bookkeeping
4613 * - Copying a long string
4615 * We swipe the string (steal the string buffer) if the SV on the
4616 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4617 * big win on long strings. It should be a win on short strings if
4618 * SvPVX_const(dstr) has to be allocated. If not, it should not
4619 * slow things down, as SvPVX_const(sstr) would have been freed
4622 * We also steal the buffer from a PADTMP (operator target) if it
4623 * is ‘long enough’. For short strings, a swipe does not help
4624 * here, as it causes more malloc calls the next time the target
4625 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4626 * be allocated it is still not worth swiping PADTMPs for short
4627 * strings, as the savings here are small.
4629 * If swiping is not an option, then we see whether it is
4630 * worth using copy-on-write. If the lhs already has a buf-
4631 * fer big enough and the string is short, we skip it and fall back
4632 * to method 3, since memcpy is faster for short strings than the
4633 * later bookkeeping overhead that copy-on-write entails.
4635 * If the rhs is not a copy-on-write string yet, then we also
4636 * consider whether the buffer is too large relative to the string
4637 * it holds. Some operations such as readline allocate a large
4638 * buffer in the expectation of reusing it. But turning such into
4639 * a COW buffer is counter-productive because it increases memory
4640 * usage by making readline allocate a new large buffer the sec-
4641 * ond time round. So, if the buffer is too large, again, we use
4644 * Finally, if there is no buffer on the left, or the buffer is too
4645 * small, then we use copy-on-write and make both SVs share the
4650 /* Whichever path we take through the next code, we want this true,
4651 and doing it now facilitates the COW check. */
4652 (void)SvPOK_only(dstr);
4656 /* slated for free anyway (and not COW)? */
4657 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4658 /* or a swipable TARG */
4660 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4662 /* whose buffer is worth stealing */
4663 && CHECK_COWBUF_THRESHOLD(cur,len)
4666 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4667 (!(flags & SV_NOSTEAL)) &&
4668 /* and we're allowed to steal temps */
4669 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4670 len) /* and really is a string */
4671 { /* Passes the swipe test. */
4672 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4674 SvPV_set(dstr, SvPVX_mutable(sstr));
4675 SvLEN_set(dstr, SvLEN(sstr));
4676 SvCUR_set(dstr, SvCUR(sstr));
4679 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4680 SvPV_set(sstr, NULL);
4685 else if (flags & SV_COW_SHARED_HASH_KEYS
4687 #ifdef PERL_COPY_ON_WRITE
4690 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4691 /* If this is a regular (non-hek) COW, only so
4692 many COW "copies" are possible. */
4693 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4694 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4695 && !(SvFLAGS(dstr) & SVf_BREAK)
4696 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4697 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4701 && !(SvFLAGS(dstr) & SVf_BREAK)
4704 /* Either it's a shared hash key, or it's suitable for
4707 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4712 if (!(sflags & SVf_IsCOW)) {
4714 CowREFCNT(sstr) = 0;
4717 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4723 if (sflags & SVf_IsCOW) {
4727 SvPV_set(dstr, SvPVX_mutable(sstr));
4732 /* SvIsCOW_shared_hash */
4733 DEBUG_C(PerlIO_printf(Perl_debug_log,
4734 "Copy on write: Sharing hash\n"));
4736 assert (SvTYPE(dstr) >= SVt_PV);
4738 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4740 SvLEN_set(dstr, len);
4741 SvCUR_set(dstr, cur);
4744 /* Failed the swipe test, and we cannot do copy-on-write either.
4745 Have to copy the string. */
4746 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4747 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4748 SvCUR_set(dstr, cur);
4749 *SvEND(dstr) = '\0';
4751 if (sflags & SVp_NOK) {
4752 SvNV_set(dstr, SvNVX(sstr));
4754 if (sflags & SVp_IOK) {
4755 SvIV_set(dstr, SvIVX(sstr));
4756 if (sflags & SVf_IVisUV)
4759 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4761 const MAGIC * const smg = SvVSTRING_mg(sstr);
4763 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4764 smg->mg_ptr, smg->mg_len);
4765 SvRMAGICAL_on(dstr);
4769 else if (sflags & (SVp_IOK|SVp_NOK)) {
4770 (void)SvOK_off(dstr);
4771 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4772 if (sflags & SVp_IOK) {
4773 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4774 SvIV_set(dstr, SvIVX(sstr));
4776 if (sflags & SVp_NOK) {
4777 SvNV_set(dstr, SvNVX(sstr));
4781 if (isGV_with_GP(sstr)) {
4782 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4785 (void)SvOK_off(dstr);
4787 if (SvTAINTED(sstr))
4793 =for apidoc sv_set_undef
4795 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4796 Doesn't handle set magic.
4798 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4799 buffer, unlike C<undef $sv>.
4801 Introduced in perl 5.25.12.
4807 Perl_sv_set_undef(pTHX_ SV *sv)
4809 U32 type = SvTYPE(sv);
4811 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4813 /* shortcut, NULL, IV, RV */
4815 if (type <= SVt_IV) {
4816 assert(!SvGMAGICAL(sv));
4817 if (SvREADONLY(sv)) {
4818 /* does undeffing PL_sv_undef count as modifying a read-only
4819 * variable? Some XS code does this */
4820 if (sv == &PL_sv_undef)
4822 Perl_croak_no_modify();
4827 sv_unref_flags(sv, 0);
4830 SvFLAGS(sv) = type; /* quickly turn off all flags */
4831 SvREFCNT_dec_NN(rv);
4835 SvFLAGS(sv) = type; /* quickly turn off all flags */
4840 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4843 SV_CHECK_THINKFIRST_COW_DROP(sv);
4845 if (isGV_with_GP(sv))
4846 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4847 "Undefined value assigned to typeglob");
4855 =for apidoc sv_setsv_mg
4857 Like C<sv_setsv>, but also handles 'set' magic.
4863 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4865 PERL_ARGS_ASSERT_SV_SETSV_MG;
4867 sv_setsv(dstr,sstr);
4872 # define SVt_COW SVt_PV
4874 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4876 STRLEN cur = SvCUR(sstr);
4877 STRLEN len = SvLEN(sstr);
4879 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4880 const bool already = cBOOL(SvIsCOW(sstr));
4883 PERL_ARGS_ASSERT_SV_SETSV_COW;
4886 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4887 (void*)sstr, (void*)dstr);
4894 if (SvTHINKFIRST(dstr))
4895 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4896 else if (SvPVX_const(dstr))
4897 Safefree(SvPVX_mutable(dstr));
4901 SvUPGRADE(dstr, SVt_COW);
4903 assert (SvPOK(sstr));
4904 assert (SvPOKp(sstr));
4906 if (SvIsCOW(sstr)) {
4908 if (SvLEN(sstr) == 0) {
4909 /* source is a COW shared hash key. */
4910 DEBUG_C(PerlIO_printf(Perl_debug_log,
4911 "Fast copy on write: Sharing hash\n"));
4912 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4915 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4916 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4918 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4919 SvUPGRADE(sstr, SVt_COW);
4921 DEBUG_C(PerlIO_printf(Perl_debug_log,
4922 "Fast copy on write: Converting sstr to COW\n"));
4923 CowREFCNT(sstr) = 0;
4925 # ifdef PERL_DEBUG_READONLY_COW
4926 if (already) sv_buf_to_rw(sstr);
4929 new_pv = SvPVX_mutable(sstr);
4933 SvPV_set(dstr, new_pv);
4934 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4937 SvLEN_set(dstr, len);
4938 SvCUR_set(dstr, cur);
4947 =for apidoc sv_setpv_bufsize
4949 Sets the SV to be a string of cur bytes length, with at least
4950 len bytes available. Ensures that there is a null byte at SvEND.
4951 Returns a char * pointer to the SvPV buffer.
4957 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4961 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4963 SV_CHECK_THINKFIRST_COW_DROP(sv);
4964 SvUPGRADE(sv, SVt_PV);
4965 pv = SvGROW(sv, len + 1);
4968 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4971 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4976 =for apidoc sv_setpvn
4978 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4979 The C<len> parameter indicates the number of
4980 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4981 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4987 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4991 PERL_ARGS_ASSERT_SV_SETPVN;
4993 SV_CHECK_THINKFIRST_COW_DROP(sv);
4994 if (isGV_with_GP(sv))
4995 Perl_croak_no_modify();
5001 /* len is STRLEN which is unsigned, need to copy to signed */
5004 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
5007 SvUPGRADE(sv, SVt_PV);
5009 dptr = SvGROW(sv, len + 1);
5010 Move(ptr,dptr,len,char);
5013 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5015 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5019 =for apidoc sv_setpvn_mg
5021 Like C<sv_setpvn>, but also handles 'set' magic.
5027 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5029 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5031 sv_setpvn(sv,ptr,len);
5036 =for apidoc sv_setpv
5038 Copies a string into an SV. The string must be terminated with a C<NUL>
5039 character, and not contain embeded C<NUL>'s.
5040 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5046 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5050 PERL_ARGS_ASSERT_SV_SETPV;
5052 SV_CHECK_THINKFIRST_COW_DROP(sv);
5058 SvUPGRADE(sv, SVt_PV);
5060 SvGROW(sv, len + 1);
5061 Move(ptr,SvPVX(sv),len+1,char);
5063 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5065 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5069 =for apidoc sv_setpv_mg
5071 Like C<sv_setpv>, but also handles 'set' magic.
5077 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5079 PERL_ARGS_ASSERT_SV_SETPV_MG;
5086 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5088 PERL_ARGS_ASSERT_SV_SETHEK;
5094 if (HEK_LEN(hek) == HEf_SVKEY) {
5095 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5098 const int flags = HEK_FLAGS(hek);
5099 if (flags & HVhek_WASUTF8) {
5100 STRLEN utf8_len = HEK_LEN(hek);
5101 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5102 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5105 } else if (flags & HVhek_UNSHARED) {
5106 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5109 else SvUTF8_off(sv);
5113 SV_CHECK_THINKFIRST_COW_DROP(sv);
5114 SvUPGRADE(sv, SVt_PV);
5116 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5117 SvCUR_set(sv, HEK_LEN(hek));
5123 else SvUTF8_off(sv);
5131 =for apidoc sv_usepvn_flags
5133 Tells an SV to use C<ptr> to find its string value. Normally the
5134 string is stored inside the SV, but sv_usepvn allows the SV to use an
5135 outside string. C<ptr> should point to memory that was allocated
5136 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5137 the start of a C<Newx>-ed block of memory, and not a pointer to the
5138 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5139 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5140 string length, C<len>, must be supplied. By default this function
5141 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5142 so that pointer should not be freed or used by the programmer after
5143 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5144 that pointer (e.g. ptr + 1) be used.
5146 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5147 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5149 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5150 C<len>, and already meets the requirements for storing in C<SvPVX>).
5156 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5160 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5162 SV_CHECK_THINKFIRST_COW_DROP(sv);
5163 SvUPGRADE(sv, SVt_PV);
5166 if (flags & SV_SMAGIC)
5170 if (SvPVX_const(sv))
5174 if (flags & SV_HAS_TRAILING_NUL)
5175 assert(ptr[len] == '\0');
5178 allocate = (flags & SV_HAS_TRAILING_NUL)
5180 #ifdef Perl_safesysmalloc_size
5183 PERL_STRLEN_ROUNDUP(len + 1);
5185 if (flags & SV_HAS_TRAILING_NUL) {
5186 /* It's long enough - do nothing.
5187 Specifically Perl_newCONSTSUB is relying on this. */
5190 /* Force a move to shake out bugs in callers. */
5191 char *new_ptr = (char*)safemalloc(allocate);
5192 Copy(ptr, new_ptr, len, char);
5193 PoisonFree(ptr,len,char);
5197 ptr = (char*) saferealloc (ptr, allocate);
5200 #ifdef Perl_safesysmalloc_size
5201 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5203 SvLEN_set(sv, allocate);
5207 if (!(flags & SV_HAS_TRAILING_NUL)) {
5210 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5212 if (flags & SV_SMAGIC)
5218 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5220 assert(SvIsCOW(sv));
5223 const char * const pvx = SvPVX_const(sv);
5224 const STRLEN len = SvLEN(sv);
5225 const STRLEN cur = SvCUR(sv);
5228 PerlIO_printf(Perl_debug_log,
5229 "Copy on write: Force normal %ld\n",
5234 # ifdef PERL_COPY_ON_WRITE
5236 /* Must do this first, since the CowREFCNT uses SvPVX and
5237 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5238 the only owner left of the buffer. */
5239 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5241 U8 cowrefcnt = CowREFCNT(sv);
5242 if(cowrefcnt != 0) {
5244 CowREFCNT(sv) = cowrefcnt;
5249 /* Else we are the only owner of the buffer. */
5254 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5259 if (flags & SV_COW_DROP_PV) {
5260 /* OK, so we don't need to copy our buffer. */
5263 SvGROW(sv, cur + 1);
5264 Move(pvx,SvPVX(sv),cur,char);
5270 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5277 const char * const pvx = SvPVX_const(sv);
5278 const STRLEN len = SvCUR(sv);
5282 if (flags & SV_COW_DROP_PV) {
5283 /* OK, so we don't need to copy our buffer. */
5286 SvGROW(sv, len + 1);
5287 Move(pvx,SvPVX(sv),len,char);
5290 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5297 =for apidoc sv_force_normal_flags
5299 Undo various types of fakery on an SV, where fakery means
5300 "more than" a string: if the PV is a shared string, make
5301 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5302 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5303 we do the copy, and is also used locally; if this is a
5304 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5305 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5306 C<SvPOK_off> rather than making a copy. (Used where this
5307 scalar is about to be set to some other value.) In addition,
5308 the C<flags> parameter gets passed to C<sv_unref_flags()>
5309 when unreffing. C<sv_force_normal> calls this function
5310 with flags set to 0.
5312 This function is expected to be used to signal to perl that this SV is
5313 about to be written to, and any extra book-keeping needs to be taken care
5314 of. Hence, it croaks on read-only values.
5320 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5322 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5325 Perl_croak_no_modify();
5326 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5327 S_sv_uncow(aTHX_ sv, flags);
5329 sv_unref_flags(sv, flags);
5330 else if (SvFAKE(sv) && isGV_with_GP(sv))
5331 sv_unglob(sv, flags);
5332 else if (SvFAKE(sv) && isREGEXP(sv)) {
5333 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5334 to sv_unglob. We only need it here, so inline it. */
5335 const bool islv = SvTYPE(sv) == SVt_PVLV;
5336 const svtype new_type =
5337 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5338 SV *const temp = newSV_type(new_type);
5339 regexp *const temp_p = ReANY((REGEXP *)sv);
5341 if (new_type == SVt_PVMG) {
5342 SvMAGIC_set(temp, SvMAGIC(sv));
5343 SvMAGIC_set(sv, NULL);
5344 SvSTASH_set(temp, SvSTASH(sv));
5345 SvSTASH_set(sv, NULL);
5347 if (!islv) SvCUR_set(temp, SvCUR(sv));
5348 /* Remember that SvPVX is in the head, not the body. But
5349 RX_WRAPPED is in the body. */
5350 assert(ReANY((REGEXP *)sv)->mother_re);
5351 /* Their buffer is already owned by someone else. */
5352 if (flags & SV_COW_DROP_PV) {
5353 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5354 zeroed body. For SVt_PVLV, it should have been set to 0
5355 before turning into a regexp. */
5356 assert(!SvLEN(islv ? sv : temp));
5357 sv->sv_u.svu_pv = 0;
5360 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5361 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5365 /* Now swap the rest of the bodies. */
5369 SvFLAGS(sv) &= ~SVTYPEMASK;
5370 SvFLAGS(sv) |= new_type;
5371 SvANY(sv) = SvANY(temp);
5374 SvFLAGS(temp) &= ~(SVTYPEMASK);
5375 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5376 SvANY(temp) = temp_p;
5377 temp->sv_u.svu_rx = (regexp *)temp_p;
5379 SvREFCNT_dec_NN(temp);
5381 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5387 Efficient removal of characters from the beginning of the string buffer.
5388 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5389 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5390 character of the adjusted string. Uses the C<OOK> hack. On return, only
5391 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5393 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5394 refer to the same chunk of data.
5396 The unfortunate similarity of this function's name to that of Perl's C<chop>
5397 operator is strictly coincidental. This function works from the left;
5398 C<chop> works from the right.
5404 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5415 PERL_ARGS_ASSERT_SV_CHOP;
5417 if (!ptr || !SvPOKp(sv))
5419 delta = ptr - SvPVX_const(sv);
5421 /* Nothing to do. */
5424 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5425 if (delta > max_delta)
5426 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5427 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5428 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5429 SV_CHECK_THINKFIRST(sv);
5430 SvPOK_only_UTF8(sv);
5433 if (!SvLEN(sv)) { /* make copy of shared string */
5434 const char *pvx = SvPVX_const(sv);
5435 const STRLEN len = SvCUR(sv);
5436 SvGROW(sv, len + 1);
5437 Move(pvx,SvPVX(sv),len,char);
5443 SvOOK_offset(sv, old_delta);
5445 SvLEN_set(sv, SvLEN(sv) - delta);
5446 SvCUR_set(sv, SvCUR(sv) - delta);
5447 SvPV_set(sv, SvPVX(sv) + delta);
5449 p = (U8 *)SvPVX_const(sv);
5452 /* how many bytes were evacuated? we will fill them with sentinel
5453 bytes, except for the part holding the new offset of course. */
5456 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5458 assert(evacn <= delta + old_delta);
5462 /* This sets 'delta' to the accumulated value of all deltas so far */
5466 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5467 * the string; otherwise store a 0 byte there and store 'delta' just prior
5468 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5469 * portion of the chopped part of the string */
5470 if (delta < 0x100) {
5474 p -= sizeof(STRLEN);
5475 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5479 /* Fill the preceding buffer with sentinals to verify that no-one is
5489 =for apidoc sv_catpvn
5491 Concatenates the string onto the end of the string which is in the SV.
5492 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5493 status set, then the bytes appended should be valid UTF-8.
5494 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5496 =for apidoc sv_catpvn_flags
5498 Concatenates the string onto the end of the string which is in the SV. The
5499 C<len> indicates number of bytes to copy.
5501 By default, the string appended is assumed to be valid UTF-8 if the SV has
5502 the UTF-8 status set, and a string of bytes otherwise. One can force the
5503 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5504 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5505 string appended will be upgraded to UTF-8 if necessary.
5507 If C<flags> has the C<SV_SMAGIC> bit set, will
5508 C<mg_set> on C<dsv> afterwards if appropriate.
5509 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5510 in terms of this function.
5516 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5519 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5521 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5522 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5524 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5525 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5526 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5529 else SvGROW(dsv, dlen + slen + 3);
5531 sstr = SvPVX_const(dsv);
5532 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5533 SvCUR_set(dsv, SvCUR(dsv) + slen);
5536 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5537 const char * const send = sstr + slen;
5540 /* Something this code does not account for, which I think is
5541 impossible; it would require the same pv to be treated as
5542 bytes *and* utf8, which would indicate a bug elsewhere. */
5543 assert(sstr != dstr);
5545 SvGROW(dsv, dlen + slen * 2 + 3);
5546 d = (U8 *)SvPVX(dsv) + dlen;
5548 while (sstr < send) {
5549 append_utf8_from_native_byte(*sstr, &d);
5552 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5555 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5557 if (flags & SV_SMAGIC)
5562 =for apidoc sv_catsv
5564 Concatenates the string from SV C<ssv> onto the end of the string in SV
5565 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5566 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5567 and C<L</sv_catsv_nomg>>.
5569 =for apidoc sv_catsv_flags
5571 Concatenates the string from SV C<ssv> onto the end of the string in SV
5572 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5573 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5574 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5575 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5576 and C<sv_catsv_mg> are implemented in terms of this function.
5581 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5583 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5587 const char *spv = SvPV_flags_const(ssv, slen, flags);
5588 if (flags & SV_GMAGIC)
5590 sv_catpvn_flags(dsv, spv, slen,
5591 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5592 if (flags & SV_SMAGIC)
5598 =for apidoc sv_catpv
5600 Concatenates the C<NUL>-terminated string onto the end of the string which is
5602 If the SV has the UTF-8 status set, then the bytes appended should be
5603 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5609 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5615 PERL_ARGS_ASSERT_SV_CATPV;
5619 junk = SvPV_force(sv, tlen);
5621 SvGROW(sv, tlen + len + 1);
5623 ptr = SvPVX_const(sv);
5624 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5625 SvCUR_set(sv, SvCUR(sv) + len);
5626 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5631 =for apidoc sv_catpv_flags
5633 Concatenates the C<NUL>-terminated string onto the end of the string which is
5635 If the SV has the UTF-8 status set, then the bytes appended should
5636 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5637 on the modified SV if appropriate.
5643 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5645 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5646 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5650 =for apidoc sv_catpv_mg
5652 Like C<sv_catpv>, but also handles 'set' magic.
5658 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5660 PERL_ARGS_ASSERT_SV_CATPV_MG;
5669 Creates a new SV. A non-zero C<len> parameter indicates the number of
5670 bytes of preallocated string space the SV should have. An extra byte for a
5671 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5672 space is allocated.) The reference count for the new SV is set to 1.
5674 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5675 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5676 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5677 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5678 modules supporting older perls.
5684 Perl_newSV(pTHX_ const STRLEN len)
5690 sv_grow(sv, len + 1);
5695 =for apidoc sv_magicext
5697 Adds magic to an SV, upgrading it if necessary. Applies the
5698 supplied C<vtable> and returns a pointer to the magic added.
5700 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5701 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5702 one instance of the same C<how>.
5704 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5705 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5706 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5707 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5709 (This is now used as a subroutine by C<sv_magic>.)
5714 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5715 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5719 PERL_ARGS_ASSERT_SV_MAGICEXT;
5721 SvUPGRADE(sv, SVt_PVMG);
5722 Newxz(mg, 1, MAGIC);
5723 mg->mg_moremagic = SvMAGIC(sv);
5724 SvMAGIC_set(sv, mg);
5726 /* Sometimes a magic contains a reference loop, where the sv and
5727 object refer to each other. To prevent a reference loop that
5728 would prevent such objects being freed, we look for such loops
5729 and if we find one we avoid incrementing the object refcount.
5731 Note we cannot do this to avoid self-tie loops as intervening RV must
5732 have its REFCNT incremented to keep it in existence.
5735 if (!obj || obj == sv ||
5736 how == PERL_MAGIC_arylen ||
5737 how == PERL_MAGIC_regdata ||
5738 how == PERL_MAGIC_regdatum ||
5739 how == PERL_MAGIC_symtab ||
5740 (SvTYPE(obj) == SVt_PVGV &&
5741 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5742 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5743 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5748 mg->mg_obj = SvREFCNT_inc_simple(obj);
5749 mg->mg_flags |= MGf_REFCOUNTED;
5752 /* Normal self-ties simply pass a null object, and instead of
5753 using mg_obj directly, use the SvTIED_obj macro to produce a
5754 new RV as needed. For glob "self-ties", we are tieing the PVIO
5755 with an RV obj pointing to the glob containing the PVIO. In
5756 this case, to avoid a reference loop, we need to weaken the
5760 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5761 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5767 mg->mg_len = namlen;
5770 mg->mg_ptr = savepvn(name, namlen);
5771 else if (namlen == HEf_SVKEY) {
5772 /* Yes, this is casting away const. This is only for the case of
5773 HEf_SVKEY. I think we need to document this aberation of the
5774 constness of the API, rather than making name non-const, as
5775 that change propagating outwards a long way. */
5776 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5778 mg->mg_ptr = (char *) name;
5780 mg->mg_virtual = (MGVTBL *) vtable;
5787 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5789 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5790 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5791 /* This sv is only a delegate. //g magic must be attached to
5796 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5797 &PL_vtbl_mglob, 0, 0);
5801 =for apidoc sv_magic
5803 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5804 necessary, then adds a new magic item of type C<how> to the head of the
5807 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5808 handling of the C<name> and C<namlen> arguments.
5810 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5811 to add more than one instance of the same C<how>.
5817 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5818 const char *const name, const I32 namlen)
5820 const MGVTBL *vtable;
5823 unsigned int vtable_index;
5825 PERL_ARGS_ASSERT_SV_MAGIC;
5827 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5828 || ((flags = PL_magic_data[how]),
5829 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5830 > magic_vtable_max))
5831 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5833 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5834 Useful for attaching extension internal data to perl vars.
5835 Note that multiple extensions may clash if magical scalars
5836 etc holding private data from one are passed to another. */
5838 vtable = (vtable_index == magic_vtable_max)
5839 ? NULL : PL_magic_vtables + vtable_index;
5841 if (SvREADONLY(sv)) {
5843 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5846 Perl_croak_no_modify();
5849 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5850 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5851 /* sv_magic() refuses to add a magic of the same 'how' as an
5854 if (how == PERL_MAGIC_taint)
5860 /* Force pos to be stored as characters, not bytes. */
5861 if (SvMAGICAL(sv) && DO_UTF8(sv)
5862 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5864 && mg->mg_flags & MGf_BYTES) {
5865 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5867 mg->mg_flags &= ~MGf_BYTES;
5870 /* Rest of work is done else where */
5871 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5874 case PERL_MAGIC_taint:
5877 case PERL_MAGIC_ext:
5878 case PERL_MAGIC_dbfile:
5885 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5892 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5894 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5895 for (mg = *mgp; mg; mg = *mgp) {
5896 const MGVTBL* const virt = mg->mg_virtual;
5897 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5898 *mgp = mg->mg_moremagic;
5899 if (virt && virt->svt_free)
5900 virt->svt_free(aTHX_ sv, mg);
5901 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5903 Safefree(mg->mg_ptr);
5904 else if (mg->mg_len == HEf_SVKEY)
5905 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5906 else if (mg->mg_type == PERL_MAGIC_utf8)
5907 Safefree(mg->mg_ptr);
5909 if (mg->mg_flags & MGf_REFCOUNTED)
5910 SvREFCNT_dec(mg->mg_obj);
5914 mgp = &mg->mg_moremagic;
5917 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5918 mg_magical(sv); /* else fix the flags now */
5927 =for apidoc sv_unmagic
5929 Removes all magic of type C<type> from an SV.
5935 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5937 PERL_ARGS_ASSERT_SV_UNMAGIC;
5938 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5942 =for apidoc sv_unmagicext
5944 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5950 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5952 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5953 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5957 =for apidoc sv_rvweaken
5959 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5960 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5961 push a back-reference to this RV onto the array of backreferences
5962 associated with that magic. If the RV is magical, set magic will be
5963 called after the RV is cleared.
5969 Perl_sv_rvweaken(pTHX_ SV *const sv)
5973 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5975 if (!SvOK(sv)) /* let undefs pass */
5978 Perl_croak(aTHX_ "Can't weaken a nonreference");
5979 else if (SvWEAKREF(sv)) {
5980 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5983 else if (SvREADONLY(sv)) croak_no_modify();
5985 Perl_sv_add_backref(aTHX_ tsv, sv);
5987 SvREFCNT_dec_NN(tsv);
5992 =for apidoc sv_get_backrefs
5994 If C<sv> is the target of a weak reference then it returns the back
5995 references structure associated with the sv; otherwise return C<NULL>.
5997 When returning a non-null result the type of the return is relevant. If it
5998 is an AV then the elements of the AV are the weak reference RVs which
5999 point at this item. If it is any other type then the item itself is the
6002 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6003 C<Perl_sv_kill_backrefs()>
6009 Perl_sv_get_backrefs(SV *const sv)
6013 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6015 /* find slot to store array or singleton backref */
6017 if (SvTYPE(sv) == SVt_PVHV) {
6019 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6020 backrefs = (SV *)iter->xhv_backreferences;
6022 } else if (SvMAGICAL(sv)) {
6023 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6025 backrefs = mg->mg_obj;
6030 /* Give tsv backref magic if it hasn't already got it, then push a
6031 * back-reference to sv onto the array associated with the backref magic.
6033 * As an optimisation, if there's only one backref and it's not an AV,
6034 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6035 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6039 /* A discussion about the backreferences array and its refcount:
6041 * The AV holding the backreferences is pointed to either as the mg_obj of
6042 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6043 * xhv_backreferences field. The array is created with a refcount
6044 * of 2. This means that if during global destruction the array gets
6045 * picked on before its parent to have its refcount decremented by the
6046 * random zapper, it won't actually be freed, meaning it's still there for
6047 * when its parent gets freed.
6049 * When the parent SV is freed, the extra ref is killed by
6050 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6051 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6053 * When a single backref SV is stored directly, it is not reference
6058 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6064 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6066 /* find slot to store array or singleton backref */
6068 if (SvTYPE(tsv) == SVt_PVHV) {
6069 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6072 mg = mg_find(tsv, PERL_MAGIC_backref);
6074 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6075 svp = &(mg->mg_obj);
6078 /* create or retrieve the array */
6080 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6081 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6085 mg->mg_flags |= MGf_REFCOUNTED;
6088 SvREFCNT_inc_simple_void_NN(av);
6089 /* av now has a refcnt of 2; see discussion above */
6090 av_extend(av, *svp ? 2 : 1);
6092 /* move single existing backref to the array */
6093 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6098 av = MUTABLE_AV(*svp);
6100 /* optimisation: store single backref directly in HvAUX or mg_obj */
6104 assert(SvTYPE(av) == SVt_PVAV);
6105 if (AvFILLp(av) >= AvMAX(av)) {
6106 av_extend(av, AvFILLp(av)+1);
6109 /* push new backref */
6110 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6113 /* delete a back-reference to ourselves from the backref magic associated
6114 * with the SV we point to.
6118 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6122 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6124 if (SvTYPE(tsv) == SVt_PVHV) {
6126 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6128 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6129 /* It's possible for the the last (strong) reference to tsv to have
6130 become freed *before* the last thing holding a weak reference.
6131 If both survive longer than the backreferences array, then when
6132 the referent's reference count drops to 0 and it is freed, it's
6133 not able to chase the backreferences, so they aren't NULLed.
6135 For example, a CV holds a weak reference to its stash. If both the
6136 CV and the stash survive longer than the backreferences array,
6137 and the CV gets picked for the SvBREAK() treatment first,
6138 *and* it turns out that the stash is only being kept alive because
6139 of an our variable in the pad of the CV, then midway during CV
6140 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6141 It ends up pointing to the freed HV. Hence it's chased in here, and
6142 if this block wasn't here, it would hit the !svp panic just below.
6144 I don't believe that "better" destruction ordering is going to help
6145 here - during global destruction there's always going to be the
6146 chance that something goes out of order. We've tried to make it
6147 foolproof before, and it only resulted in evolutionary pressure on
6148 fools. Which made us look foolish for our hubris. :-(
6154 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6155 svp = mg ? &(mg->mg_obj) : NULL;
6159 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6161 /* It's possible that sv is being freed recursively part way through the
6162 freeing of tsv. If this happens, the backreferences array of tsv has
6163 already been freed, and so svp will be NULL. If this is the case,
6164 we should not panic. Instead, nothing needs doing, so return. */
6165 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6167 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6168 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6171 if (SvTYPE(*svp) == SVt_PVAV) {
6175 AV * const av = (AV*)*svp;
6177 assert(!SvIS_FREED(av));
6181 /* for an SV with N weak references to it, if all those
6182 * weak refs are deleted, then sv_del_backref will be called
6183 * N times and O(N^2) compares will be done within the backref
6184 * array. To ameliorate this potential slowness, we:
6185 * 1) make sure this code is as tight as possible;
6186 * 2) when looking for SV, look for it at both the head and tail of the
6187 * array first before searching the rest, since some create/destroy
6188 * patterns will cause the backrefs to be freed in order.
6195 SV **p = &svp[fill];
6196 SV *const topsv = *p;
6203 /* We weren't the last entry.
6204 An unordered list has this property that you
6205 can take the last element off the end to fill
6206 the hole, and it's still an unordered list :-)
6212 break; /* should only be one */
6219 AvFILLp(av) = fill-1;
6221 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6222 /* freed AV; skip */
6225 /* optimisation: only a single backref, stored directly */
6227 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6228 (void*)*svp, (void*)sv);
6235 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6241 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6246 /* after multiple passes through Perl_sv_clean_all() for a thingy
6247 * that has badly leaked, the backref array may have gotten freed,
6248 * since we only protect it against 1 round of cleanup */
6249 if (SvIS_FREED(av)) {
6250 if (PL_in_clean_all) /* All is fair */
6253 "panic: magic_killbackrefs (freed backref AV/SV)");
6257 is_array = (SvTYPE(av) == SVt_PVAV);
6259 assert(!SvIS_FREED(av));
6262 last = svp + AvFILLp(av);
6265 /* optimisation: only a single backref, stored directly */
6271 while (svp <= last) {
6273 SV *const referrer = *svp;
6274 if (SvWEAKREF(referrer)) {
6275 /* XXX Should we check that it hasn't changed? */
6276 assert(SvROK(referrer));
6277 SvRV_set(referrer, 0);
6279 SvWEAKREF_off(referrer);
6280 SvSETMAGIC(referrer);
6281 } else if (SvTYPE(referrer) == SVt_PVGV ||
6282 SvTYPE(referrer) == SVt_PVLV) {
6283 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6284 /* You lookin' at me? */
6285 assert(GvSTASH(referrer));
6286 assert(GvSTASH(referrer) == (const HV *)sv);
6287 GvSTASH(referrer) = 0;
6288 } else if (SvTYPE(referrer) == SVt_PVCV ||
6289 SvTYPE(referrer) == SVt_PVFM) {
6290 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6291 /* You lookin' at me? */
6292 assert(CvSTASH(referrer));
6293 assert(CvSTASH(referrer) == (const HV *)sv);
6294 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6297 assert(SvTYPE(sv) == SVt_PVGV);
6298 /* You lookin' at me? */
6299 assert(CvGV(referrer));
6300 assert(CvGV(referrer) == (const GV *)sv);
6301 anonymise_cv_maybe(MUTABLE_GV(sv),
6302 MUTABLE_CV(referrer));
6307 "panic: magic_killbackrefs (flags=%" UVxf ")",
6308 (UV)SvFLAGS(referrer));
6319 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6325 =for apidoc sv_insert
6327 Inserts a string at the specified offset/length within the SV. Similar to
6328 the Perl C<substr()> function. Handles get magic.
6330 =for apidoc sv_insert_flags
6332 Same as C<sv_insert>, but the extra C<flags> are passed to the
6333 C<SvPV_force_flags> that applies to C<bigstr>.
6339 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6345 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6348 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6350 SvPV_force_flags(bigstr, curlen, flags);
6351 (void)SvPOK_only_UTF8(bigstr);
6353 if (little >= SvPVX(bigstr) &&
6354 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6355 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6356 or little...little+littlelen might overlap offset...offset+len we make a copy
6358 little = savepvn(little, littlelen);
6362 if (offset + len > curlen) {
6363 SvGROW(bigstr, offset+len+1);
6364 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6365 SvCUR_set(bigstr, offset+len);
6369 i = littlelen - len;
6370 if (i > 0) { /* string might grow */
6371 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6372 mid = big + offset + len;
6373 midend = bigend = big + SvCUR(bigstr);
6376 while (midend > mid) /* shove everything down */
6377 *--bigend = *--midend;
6378 Move(little,big+offset,littlelen,char);
6379 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6384 Move(little,SvPVX(bigstr)+offset,len,char);
6389 big = SvPVX(bigstr);
6392 bigend = big + SvCUR(bigstr);
6394 if (midend > bigend)
6395 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6398 if (mid - big > bigend - midend) { /* faster to shorten from end */
6400 Move(little, mid, littlelen,char);
6403 i = bigend - midend;
6405 Move(midend, mid, i,char);
6409 SvCUR_set(bigstr, mid - big);
6411 else if ((i = mid - big)) { /* faster from front */
6412 midend -= littlelen;
6414 Move(big, midend - i, i, char);
6415 sv_chop(bigstr,midend-i);
6417 Move(little, mid, littlelen,char);
6419 else if (littlelen) {
6420 midend -= littlelen;
6421 sv_chop(bigstr,midend);
6422 Move(little,midend,littlelen,char);
6425 sv_chop(bigstr,midend);
6431 =for apidoc sv_replace
6433 Make the first argument a copy of the second, then delete the original.
6434 The target SV physically takes over ownership of the body of the source SV
6435 and inherits its flags; however, the target keeps any magic it owns,
6436 and any magic in the source is discarded.
6437 Note that this is a rather specialist SV copying operation; most of the
6438 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6444 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6446 const U32 refcnt = SvREFCNT(sv);
6448 PERL_ARGS_ASSERT_SV_REPLACE;
6450 SV_CHECK_THINKFIRST_COW_DROP(sv);
6451 if (SvREFCNT(nsv) != 1) {
6452 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6453 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6455 if (SvMAGICAL(sv)) {
6459 sv_upgrade(nsv, SVt_PVMG);
6460 SvMAGIC_set(nsv, SvMAGIC(sv));
6461 SvFLAGS(nsv) |= SvMAGICAL(sv);
6463 SvMAGIC_set(sv, NULL);
6467 assert(!SvREFCNT(sv));
6468 #ifdef DEBUG_LEAKING_SCALARS
6469 sv->sv_flags = nsv->sv_flags;
6470 sv->sv_any = nsv->sv_any;
6471 sv->sv_refcnt = nsv->sv_refcnt;
6472 sv->sv_u = nsv->sv_u;
6474 StructCopy(nsv,sv,SV);
6476 if(SvTYPE(sv) == SVt_IV) {
6477 SET_SVANY_FOR_BODYLESS_IV(sv);
6481 SvREFCNT(sv) = refcnt;
6482 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6487 /* We're about to free a GV which has a CV that refers back to us.
6488 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6492 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6497 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6500 assert(SvREFCNT(gv) == 0);
6501 assert(isGV(gv) && isGV_with_GP(gv));
6503 assert(!CvANON(cv));
6504 assert(CvGV(cv) == gv);
6505 assert(!CvNAMED(cv));
6507 /* will the CV shortly be freed by gp_free() ? */
6508 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6509 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6513 /* if not, anonymise: */
6514 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6515 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6516 : newSVpvn_flags( "__ANON__", 8, 0 );
6517 sv_catpvs(gvname, "::__ANON__");
6518 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6519 SvREFCNT_dec_NN(gvname);
6523 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6528 =for apidoc sv_clear
6530 Clear an SV: call any destructors, free up any memory used by the body,
6531 and free the body itself. The SV's head is I<not> freed, although
6532 its type is set to all 1's so that it won't inadvertently be assumed
6533 to be live during global destruction etc.
6534 This function should only be called when C<REFCNT> is zero. Most of the time
6535 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6542 Perl_sv_clear(pTHX_ SV *const orig_sv)
6547 const struct body_details *sv_type_details;
6551 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6552 Not strictly necessary */
6554 PERL_ARGS_ASSERT_SV_CLEAR;
6556 /* within this loop, sv is the SV currently being freed, and
6557 * iter_sv is the most recent AV or whatever that's being iterated
6558 * over to provide more SVs */
6564 assert(SvREFCNT(sv) == 0);
6565 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6567 if (type <= SVt_IV) {
6568 /* See the comment in sv.h about the collusion between this
6569 * early return and the overloading of the NULL slots in the
6573 SvFLAGS(sv) &= SVf_BREAK;
6574 SvFLAGS(sv) |= SVTYPEMASK;
6578 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6579 for another purpose */
6580 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6582 if (type >= SVt_PVMG) {
6584 if (!curse(sv, 1)) goto get_next_sv;
6585 type = SvTYPE(sv); /* destructor may have changed it */
6587 /* Free back-references before magic, in case the magic calls
6588 * Perl code that has weak references to sv. */
6589 if (type == SVt_PVHV) {
6590 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6594 else if (SvMAGIC(sv)) {
6595 /* Free back-references before other types of magic. */
6596 sv_unmagic(sv, PERL_MAGIC_backref);
6602 /* case SVt_INVLIST: */
6605 IoIFP(sv) != PerlIO_stdin() &&
6606 IoIFP(sv) != PerlIO_stdout() &&
6607 IoIFP(sv) != PerlIO_stderr() &&
6608 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6610 io_close(MUTABLE_IO(sv), NULL, FALSE,
6611 (IoTYPE(sv) == IoTYPE_WRONLY ||
6612 IoTYPE(sv) == IoTYPE_RDWR ||
6613 IoTYPE(sv) == IoTYPE_APPEND));
6615 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6616 PerlDir_close(IoDIRP(sv));
6617 IoDIRP(sv) = (DIR*)NULL;
6618 Safefree(IoTOP_NAME(sv));
6619 Safefree(IoFMT_NAME(sv));
6620 Safefree(IoBOTTOM_NAME(sv));
6621 if ((const GV *)sv == PL_statgv)
6625 /* FIXME for plugins */
6627 pregfree2((REGEXP*) sv);
6631 cv_undef(MUTABLE_CV(sv));
6632 /* If we're in a stash, we don't own a reference to it.
6633 * However it does have a back reference to us, which needs to
6635 if ((stash = CvSTASH(sv)))
6636 sv_del_backref(MUTABLE_SV(stash), sv);
6639 if (PL_last_swash_hv == (const HV *)sv) {
6640 PL_last_swash_hv = NULL;
6642 if (HvTOTALKEYS((HV*)sv) > 0) {
6644 /* this statement should match the one at the beginning of
6645 * hv_undef_flags() */
6646 if ( PL_phase != PERL_PHASE_DESTRUCT
6647 && (hek = HvNAME_HEK((HV*)sv)))
6649 if (PL_stashcache) {
6650 DEBUG_o(Perl_deb(aTHX_
6651 "sv_clear clearing PL_stashcache for '%" HEKf
6654 (void)hv_deletehek(PL_stashcache,
6657 hv_name_set((HV*)sv, NULL, 0, 0);
6660 /* save old iter_sv in unused SvSTASH field */
6661 assert(!SvOBJECT(sv));
6662 SvSTASH(sv) = (HV*)iter_sv;
6665 /* save old hash_index in unused SvMAGIC field */
6666 assert(!SvMAGICAL(sv));
6667 assert(!SvMAGIC(sv));
6668 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6671 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6672 goto get_next_sv; /* process this new sv */
6674 /* free empty hash */
6675 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6676 assert(!HvARRAY((HV*)sv));
6680 AV* av = MUTABLE_AV(sv);
6681 if (PL_comppad == av) {
6685 if (AvREAL(av) && AvFILLp(av) > -1) {
6686 next_sv = AvARRAY(av)[AvFILLp(av)--];
6687 /* save old iter_sv in top-most slot of AV,
6688 * and pray that it doesn't get wiped in the meantime */
6689 AvARRAY(av)[AvMAX(av)] = iter_sv;
6691 goto get_next_sv; /* process this new sv */
6693 Safefree(AvALLOC(av));
6698 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6699 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6700 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6701 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6703 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6704 SvREFCNT_dec(LvTARG(sv));
6705 if (isREGEXP(sv)) goto freeregexp;
6708 if (isGV_with_GP(sv)) {
6709 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6710 && HvENAME_get(stash))
6711 mro_method_changed_in(stash);
6712 gp_free(MUTABLE_GV(sv));
6714 unshare_hek(GvNAME_HEK(sv));
6715 /* If we're in a stash, we don't own a reference to it.
6716 * However it does have a back reference to us, which
6717 * needs to be cleared. */
6718 if ((stash = GvSTASH(sv)))
6719 sv_del_backref(MUTABLE_SV(stash), sv);
6721 /* FIXME. There are probably more unreferenced pointers to SVs
6722 * in the interpreter struct that we should check and tidy in
6723 * a similar fashion to this: */
6724 /* See also S_sv_unglob, which does the same thing. */
6725 if ((const GV *)sv == PL_last_in_gv)
6726 PL_last_in_gv = NULL;
6727 else if ((const GV *)sv == PL_statgv)
6729 else if ((const GV *)sv == PL_stderrgv)
6738 /* Don't bother with SvOOK_off(sv); as we're only going to
6742 SvOOK_offset(sv, offset);
6743 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6744 /* Don't even bother with turning off the OOK flag. */
6749 SV * const target = SvRV(sv);
6751 sv_del_backref(target, sv);
6757 else if (SvPVX_const(sv)
6758 && !(SvTYPE(sv) == SVt_PVIO
6759 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6763 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6767 if (CowREFCNT(sv)) {
6774 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6779 Safefree(SvPVX_mutable(sv));
6783 else if (SvPVX_const(sv) && SvLEN(sv)
6784 && !(SvTYPE(sv) == SVt_PVIO
6785 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6786 Safefree(SvPVX_mutable(sv));
6787 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6788 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6798 SvFLAGS(sv) &= SVf_BREAK;
6799 SvFLAGS(sv) |= SVTYPEMASK;
6801 sv_type_details = bodies_by_type + type;
6802 if (sv_type_details->arena) {
6803 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6804 &PL_body_roots[type]);
6806 else if (sv_type_details->body_size) {
6807 safefree(SvANY(sv));
6811 /* caller is responsible for freeing the head of the original sv */
6812 if (sv != orig_sv && !SvREFCNT(sv))
6815 /* grab and free next sv, if any */
6823 else if (!iter_sv) {
6825 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6826 AV *const av = (AV*)iter_sv;
6827 if (AvFILLp(av) > -1) {
6828 sv = AvARRAY(av)[AvFILLp(av)--];
6830 else { /* no more elements of current AV to free */
6833 /* restore previous value, squirrelled away */
6834 iter_sv = AvARRAY(av)[AvMAX(av)];
6835 Safefree(AvALLOC(av));
6838 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6839 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6840 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6841 /* no more elements of current HV to free */
6844 /* Restore previous values of iter_sv and hash_index,
6845 * squirrelled away */
6846 assert(!SvOBJECT(sv));
6847 iter_sv = (SV*)SvSTASH(sv);
6848 assert(!SvMAGICAL(sv));
6849 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6851 /* perl -DA does not like rubbish in SvMAGIC. */
6855 /* free any remaining detritus from the hash struct */
6856 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6857 assert(!HvARRAY((HV*)sv));
6862 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6866 if (!SvREFCNT(sv)) {
6870 if (--(SvREFCNT(sv)))
6874 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6875 "Attempt to free temp prematurely: SV 0x%" UVxf
6876 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6880 if (SvIMMORTAL(sv)) {
6881 /* make sure SvREFCNT(sv)==0 happens very seldom */
6882 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6891 /* This routine curses the sv itself, not the object referenced by sv. So
6892 sv does not have to be ROK. */
6895 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6896 PERL_ARGS_ASSERT_CURSE;
6897 assert(SvOBJECT(sv));
6899 if (PL_defstash && /* Still have a symbol table? */
6905 stash = SvSTASH(sv);
6906 assert(SvTYPE(stash) == SVt_PVHV);
6907 if (HvNAME(stash)) {
6908 CV* destructor = NULL;
6909 struct mro_meta *meta;
6911 assert (SvOOK(stash));
6913 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6916 /* don't make this an initialization above the assert, since it needs
6918 meta = HvMROMETA(stash);
6919 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6920 destructor = meta->destroy;
6921 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6922 (void *)destructor, HvNAME(stash)) );
6925 bool autoload = FALSE;
6927 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6929 destructor = GvCV(gv);
6931 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6932 GV_AUTOLOAD_ISMETHOD);
6934 destructor = GvCV(gv);
6938 /* we don't cache AUTOLOAD for DESTROY, since this code
6939 would then need to set $__PACKAGE__::AUTOLOAD, or the
6940 equivalent for XS AUTOLOADs */
6942 meta->destroy_gen = PL_sub_generation;
6943 meta->destroy = destructor;
6945 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6946 (void *)destructor, HvNAME(stash)) );
6949 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6953 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6955 /* A constant subroutine can have no side effects, so
6956 don't bother calling it. */
6957 && !CvCONST(destructor)
6958 /* Don't bother calling an empty destructor or one that
6959 returns immediately. */
6960 && (CvISXSUB(destructor)
6961 || (CvSTART(destructor)
6962 && (CvSTART(destructor)->op_next->op_type
6964 && (CvSTART(destructor)->op_next->op_type
6966 || CvSTART(destructor)->op_next->op_next->op_type
6972 SV* const tmpref = newRV(sv);
6973 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6975 PUSHSTACKi(PERLSI_DESTROY);
6980 call_sv(MUTABLE_SV(destructor),
6981 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6985 if(SvREFCNT(tmpref) < 2) {
6986 /* tmpref is not kept alive! */
6988 SvRV_set(tmpref, NULL);
6991 SvREFCNT_dec_NN(tmpref);
6994 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6997 if (check_refcnt && SvREFCNT(sv)) {
6998 if (PL_in_clean_objs)
7000 "DESTROY created new reference to dead object '%" HEKf "'",
7001 HEKfARG(HvNAME_HEK(stash)));
7002 /* DESTROY gave object new lease on life */
7008 HV * const stash = SvSTASH(sv);
7009 /* Curse before freeing the stash, as freeing the stash could cause
7010 a recursive call into S_curse. */
7011 SvOBJECT_off(sv); /* Curse the object. */
7012 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7013 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7019 =for apidoc sv_newref
7021 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7028 Perl_sv_newref(pTHX_ SV *const sv)
7030 PERL_UNUSED_CONTEXT;
7039 Decrement an SV's reference count, and if it drops to zero, call
7040 C<sv_clear> to invoke destructors and free up any memory used by
7041 the body; finally, deallocating the SV's head itself.
7042 Normally called via a wrapper macro C<SvREFCNT_dec>.
7048 Perl_sv_free(pTHX_ SV *const sv)
7054 /* Private helper function for SvREFCNT_dec().
7055 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7058 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7062 PERL_ARGS_ASSERT_SV_FREE2;
7064 if (LIKELY( rc == 1 )) {
7070 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7071 "Attempt to free temp prematurely: SV 0x%" UVxf
7072 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7076 if (SvIMMORTAL(sv)) {
7077 /* make sure SvREFCNT(sv)==0 happens very seldom */
7078 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7082 if (! SvREFCNT(sv)) /* may have have been resurrected */
7087 /* handle exceptional cases */
7091 if (SvFLAGS(sv) & SVf_BREAK)
7092 /* this SV's refcnt has been artificially decremented to
7093 * trigger cleanup */
7095 if (PL_in_clean_all) /* All is fair */
7097 if (SvIMMORTAL(sv)) {
7098 /* make sure SvREFCNT(sv)==0 happens very seldom */
7099 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7102 if (ckWARN_d(WARN_INTERNAL)) {
7103 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7104 Perl_dump_sv_child(aTHX_ sv);
7106 #ifdef DEBUG_LEAKING_SCALARS
7109 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7110 if (PL_warnhook == PERL_WARNHOOK_FATAL
7111 || ckDEAD(packWARN(WARN_INTERNAL))) {
7112 /* Don't let Perl_warner cause us to escape our fate: */
7116 /* This may not return: */
7117 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7118 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7119 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7122 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7132 Returns the length of the string in the SV. Handles magic and type
7133 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7134 gives raw access to the C<xpv_cur> slot.
7140 Perl_sv_len(pTHX_ SV *const sv)
7147 (void)SvPV_const(sv, len);
7152 =for apidoc sv_len_utf8
7154 Returns the number of characters in the string in an SV, counting wide
7155 UTF-8 bytes as a single character. Handles magic and type coercion.
7161 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7162 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7163 * (Note that the mg_len is not the length of the mg_ptr field.
7164 * This allows the cache to store the character length of the string without
7165 * needing to malloc() extra storage to attach to the mg_ptr.)
7170 Perl_sv_len_utf8(pTHX_ SV *const sv)
7176 return sv_len_utf8_nomg(sv);
7180 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7183 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7185 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7187 if (PL_utf8cache && SvUTF8(sv)) {
7189 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7191 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7192 if (mg->mg_len != -1)
7195 /* We can use the offset cache for a headstart.
7196 The longer value is stored in the first pair. */
7197 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7199 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7203 if (PL_utf8cache < 0) {
7204 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7205 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7209 ulen = Perl_utf8_length(aTHX_ s, s + len);
7210 utf8_mg_len_cache_update(sv, &mg, ulen);
7214 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7217 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7220 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7221 STRLEN *const uoffset_p, bool *const at_end)
7223 const U8 *s = start;
7224 STRLEN uoffset = *uoffset_p;
7226 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7228 while (s < send && uoffset) {
7235 else if (s > send) {
7237 /* This is the existing behaviour. Possibly it should be a croak, as
7238 it's actually a bounds error */
7241 *uoffset_p -= uoffset;
7245 /* Given the length of the string in both bytes and UTF-8 characters, decide
7246 whether to walk forwards or backwards to find the byte corresponding to
7247 the passed in UTF-8 offset. */
7249 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7250 STRLEN uoffset, const STRLEN uend)
7252 STRLEN backw = uend - uoffset;
7254 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7256 if (uoffset < 2 * backw) {
7257 /* The assumption is that going forwards is twice the speed of going
7258 forward (that's where the 2 * backw comes from).
7259 (The real figure of course depends on the UTF-8 data.) */
7260 const U8 *s = start;
7262 while (s < send && uoffset--)
7272 while (UTF8_IS_CONTINUATION(*send))
7275 return send - start;
7278 /* For the string representation of the given scalar, find the byte
7279 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7280 give another position in the string, *before* the sought offset, which
7281 (which is always true, as 0, 0 is a valid pair of positions), which should
7282 help reduce the amount of linear searching.
7283 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7284 will be used to reduce the amount of linear searching. The cache will be
7285 created if necessary, and the found value offered to it for update. */
7287 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7288 const U8 *const send, STRLEN uoffset,
7289 STRLEN uoffset0, STRLEN boffset0)
7291 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7293 bool at_end = FALSE;
7295 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7297 assert (uoffset >= uoffset0);
7302 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7304 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7305 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7306 if ((*mgp)->mg_ptr) {
7307 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7308 if (cache[0] == uoffset) {
7309 /* An exact match. */
7312 if (cache[2] == uoffset) {
7313 /* An exact match. */
7317 if (cache[0] < uoffset) {
7318 /* The cache already knows part of the way. */
7319 if (cache[0] > uoffset0) {
7320 /* The cache knows more than the passed in pair */
7321 uoffset0 = cache[0];
7322 boffset0 = cache[1];
7324 if ((*mgp)->mg_len != -1) {
7325 /* And we know the end too. */
7327 + sv_pos_u2b_midway(start + boffset0, send,
7329 (*mgp)->mg_len - uoffset0);
7331 uoffset -= uoffset0;
7333 + sv_pos_u2b_forwards(start + boffset0,
7334 send, &uoffset, &at_end);
7335 uoffset += uoffset0;
7338 else if (cache[2] < uoffset) {
7339 /* We're between the two cache entries. */
7340 if (cache[2] > uoffset0) {
7341 /* and the cache knows more than the passed in pair */
7342 uoffset0 = cache[2];
7343 boffset0 = cache[3];
7347 + sv_pos_u2b_midway(start + boffset0,
7350 cache[0] - uoffset0);
7353 + sv_pos_u2b_midway(start + boffset0,
7356 cache[2] - uoffset0);
7360 else if ((*mgp)->mg_len != -1) {
7361 /* If we can take advantage of a passed in offset, do so. */
7362 /* In fact, offset0 is either 0, or less than offset, so don't
7363 need to worry about the other possibility. */
7365 + sv_pos_u2b_midway(start + boffset0, send,
7367 (*mgp)->mg_len - uoffset0);
7372 if (!found || PL_utf8cache < 0) {
7373 STRLEN real_boffset;
7374 uoffset -= uoffset0;
7375 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7376 send, &uoffset, &at_end);
7377 uoffset += uoffset0;
7379 if (found && PL_utf8cache < 0)
7380 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7382 boffset = real_boffset;
7385 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7387 utf8_mg_len_cache_update(sv, mgp, uoffset);
7389 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7396 =for apidoc sv_pos_u2b_flags
7398 Converts the offset from a count of UTF-8 chars from
7399 the start of the string, to a count of the equivalent number of bytes; if
7400 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7401 C<offset>, rather than from the start
7402 of the string. Handles type coercion.
7403 C<flags> is passed to C<SvPV_flags>, and usually should be
7404 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7410 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7411 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7412 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7417 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7424 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7426 start = (U8*)SvPV_flags(sv, len, flags);
7428 const U8 * const send = start + len;
7430 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7433 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7434 is 0, and *lenp is already set to that. */) {
7435 /* Convert the relative offset to absolute. */
7436 const STRLEN uoffset2 = uoffset + *lenp;
7437 const STRLEN boffset2
7438 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7439 uoffset, boffset) - boffset;
7453 =for apidoc sv_pos_u2b
7455 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7456 the start of the string, to a count of the equivalent number of bytes; if
7457 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7458 the offset, rather than from the start of the string. Handles magic and
7461 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7468 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7469 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7470 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7474 /* This function is subject to size and sign problems */
7477 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7479 PERL_ARGS_ASSERT_SV_POS_U2B;
7482 STRLEN ulen = (STRLEN)*lenp;
7483 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7484 SV_GMAGIC|SV_CONST_RETURN);
7487 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7488 SV_GMAGIC|SV_CONST_RETURN);
7493 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7496 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7497 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7500 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7501 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7502 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7506 (*mgp)->mg_len = ulen;
7509 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7510 byte length pairing. The (byte) length of the total SV is passed in too,
7511 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7512 may not have updated SvCUR, so we can't rely on reading it directly.
7514 The proffered utf8/byte length pairing isn't used if the cache already has
7515 two pairs, and swapping either for the proffered pair would increase the
7516 RMS of the intervals between known byte offsets.
7518 The cache itself consists of 4 STRLEN values
7519 0: larger UTF-8 offset
7520 1: corresponding byte offset
7521 2: smaller UTF-8 offset
7522 3: corresponding byte offset
7524 Unused cache pairs have the value 0, 0.
7525 Keeping the cache "backwards" means that the invariant of
7526 cache[0] >= cache[2] is maintained even with empty slots, which means that
7527 the code that uses it doesn't need to worry if only 1 entry has actually
7528 been set to non-zero. It also makes the "position beyond the end of the
7529 cache" logic much simpler, as the first slot is always the one to start
7533 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7534 const STRLEN utf8, const STRLEN blen)
7538 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7543 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7544 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7545 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7547 (*mgp)->mg_len = -1;
7551 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7552 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7553 (*mgp)->mg_ptr = (char *) cache;
7557 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7558 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7559 a pointer. Note that we no longer cache utf8 offsets on refer-
7560 ences, but this check is still a good idea, for robustness. */
7561 const U8 *start = (const U8 *) SvPVX_const(sv);
7562 const STRLEN realutf8 = utf8_length(start, start + byte);
7564 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7568 /* Cache is held with the later position first, to simplify the code
7569 that deals with unbounded ends. */
7571 ASSERT_UTF8_CACHE(cache);
7572 if (cache[1] == 0) {
7573 /* Cache is totally empty */
7576 } else if (cache[3] == 0) {
7577 if (byte > cache[1]) {
7578 /* New one is larger, so goes first. */
7579 cache[2] = cache[0];
7580 cache[3] = cache[1];
7588 /* float casts necessary? XXX */
7589 #define THREEWAY_SQUARE(a,b,c,d) \
7590 ((float)((d) - (c))) * ((float)((d) - (c))) \
7591 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7592 + ((float)((b) - (a))) * ((float)((b) - (a)))
7594 /* Cache has 2 slots in use, and we know three potential pairs.
7595 Keep the two that give the lowest RMS distance. Do the
7596 calculation in bytes simply because we always know the byte
7597 length. squareroot has the same ordering as the positive value,
7598 so don't bother with the actual square root. */
7599 if (byte > cache[1]) {
7600 /* New position is after the existing pair of pairs. */
7601 const float keep_earlier
7602 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7603 const float keep_later
7604 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7606 if (keep_later < keep_earlier) {
7607 cache[2] = cache[0];
7608 cache[3] = cache[1];
7614 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7615 float b, c, keep_earlier;
7616 if (byte > cache[3]) {
7617 /* New position is between the existing pair of pairs. */
7618 b = (float)cache[3];
7621 /* New position is before the existing pair of pairs. */
7623 c = (float)cache[3];
7625 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7626 if (byte > cache[3]) {
7627 if (keep_later < keep_earlier) {
7637 if (! (keep_later < keep_earlier)) {
7638 cache[0] = cache[2];
7639 cache[1] = cache[3];
7646 ASSERT_UTF8_CACHE(cache);
7649 /* We already know all of the way, now we may be able to walk back. The same
7650 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7651 backward is half the speed of walking forward. */
7653 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7654 const U8 *end, STRLEN endu)
7656 const STRLEN forw = target - s;
7657 STRLEN backw = end - target;
7659 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7661 if (forw < 2 * backw) {
7662 return utf8_length(s, target);
7665 while (end > target) {
7667 while (UTF8_IS_CONTINUATION(*end)) {
7676 =for apidoc sv_pos_b2u_flags
7678 Converts C<offset> from a count of bytes from the start of the string, to
7679 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7680 C<flags> is passed to C<SvPV_flags>, and usually should be
7681 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7687 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7688 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7693 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7696 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7702 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7704 s = (const U8*)SvPV_flags(sv, blen, flags);
7707 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7708 ", byte=%" UVuf, (UV)blen, (UV)offset);
7714 && SvTYPE(sv) >= SVt_PVMG
7715 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7718 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7719 if (cache[1] == offset) {
7720 /* An exact match. */
7723 if (cache[3] == offset) {
7724 /* An exact match. */
7728 if (cache[1] < offset) {
7729 /* We already know part of the way. */
7730 if (mg->mg_len != -1) {
7731 /* Actually, we know the end too. */
7733 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7734 s + blen, mg->mg_len - cache[0]);
7736 len = cache[0] + utf8_length(s + cache[1], send);
7739 else if (cache[3] < offset) {
7740 /* We're between the two cached pairs, so we do the calculation
7741 offset by the byte/utf-8 positions for the earlier pair,
7742 then add the utf-8 characters from the string start to
7744 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7745 s + cache[1], cache[0] - cache[2])
7749 else { /* cache[3] > offset */
7750 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7754 ASSERT_UTF8_CACHE(cache);
7756 } else if (mg->mg_len != -1) {
7757 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7761 if (!found || PL_utf8cache < 0) {
7762 const STRLEN real_len = utf8_length(s, send);
7764 if (found && PL_utf8cache < 0)
7765 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7771 utf8_mg_len_cache_update(sv, &mg, len);
7773 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7780 =for apidoc sv_pos_b2u
7782 Converts the value pointed to by C<offsetp> from a count of bytes from the
7783 start of the string, to a count of the equivalent number of UTF-8 chars.
7784 Handles magic and type coercion.
7786 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7793 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7794 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7799 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7801 PERL_ARGS_ASSERT_SV_POS_B2U;
7806 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7807 SV_GMAGIC|SV_CONST_RETURN);
7811 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7812 STRLEN real, SV *const sv)
7814 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7816 /* As this is debugging only code, save space by keeping this test here,
7817 rather than inlining it in all the callers. */
7818 if (from_cache == real)
7821 /* Need to turn the assertions off otherwise we may recurse infinitely
7822 while printing error messages. */
7823 SAVEI8(PL_utf8cache);
7825 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7826 func, (UV) from_cache, (UV) real, SVfARG(sv));
7832 Returns a boolean indicating whether the strings in the two SVs are
7833 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7834 coerce its args to strings if necessary.
7836 =for apidoc sv_eq_flags
7838 Returns a boolean indicating whether the strings in the two SVs are
7839 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7840 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7846 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7853 SV* svrecode = NULL;
7860 /* if pv1 and pv2 are the same, second SvPV_const call may
7861 * invalidate pv1 (if we are handling magic), so we may need to
7863 if (sv1 == sv2 && flags & SV_GMAGIC
7864 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7865 pv1 = SvPV_const(sv1, cur1);
7866 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7868 pv1 = SvPV_flags_const(sv1, cur1, flags);
7876 pv2 = SvPV_flags_const(sv2, cur2, flags);
7878 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7879 /* Differing utf8ness. */
7881 /* sv1 is the UTF-8 one */
7882 return bytes_cmp_utf8((const U8*)pv2, cur2,
7883 (const U8*)pv1, cur1) == 0;
7886 /* sv2 is the UTF-8 one */
7887 return bytes_cmp_utf8((const U8*)pv1, cur1,
7888 (const U8*)pv2, cur2) == 0;
7893 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7895 SvREFCNT_dec(svrecode);
7903 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7904 string in C<sv1> is less than, equal to, or greater than the string in
7905 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7906 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7908 =for apidoc sv_cmp_flags
7910 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7911 string in C<sv1> is less than, equal to, or greater than the string in
7912 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7913 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7914 also C<L</sv_cmp_locale_flags>>.
7920 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7922 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7926 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7930 const char *pv1, *pv2;
7932 SV *svrecode = NULL;
7939 pv1 = SvPV_flags_const(sv1, cur1, flags);
7946 pv2 = SvPV_flags_const(sv2, cur2, flags);
7948 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7949 /* Differing utf8ness. */
7951 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7952 (const U8*)pv1, cur1);
7953 return retval ? retval < 0 ? -1 : +1 : 0;
7956 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7957 (const U8*)pv2, cur2);
7958 return retval ? retval < 0 ? -1 : +1 : 0;
7962 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7965 cmp = cur2 ? -1 : 0;
7969 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7972 if (! DO_UTF8(sv1)) {
7974 const I32 retval = memcmp((const void*)pv1,
7978 cmp = retval < 0 ? -1 : 1;
7979 } else if (cur1 == cur2) {
7982 cmp = cur1 < cur2 ? -1 : 1;
7986 else { /* Both are to be treated as UTF-EBCDIC */
7988 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7989 * which remaps code points 0-255. We therefore generally have to
7990 * unmap back to the original values to get an accurate comparison.
7991 * But we don't have to do that for UTF-8 invariants, as by
7992 * definition, they aren't remapped, nor do we have to do it for
7993 * above-latin1 code points, as they also aren't remapped. (This
7994 * code also works on ASCII platforms, but the memcmp() above is
7997 const char *e = pv1 + shortest_len;
7999 /* Find the first bytes that differ between the two strings */
8000 while (pv1 < e && *pv1 == *pv2) {
8006 if (pv1 == e) { /* Are the same all the way to the end */
8010 cmp = cur1 < cur2 ? -1 : 1;
8013 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8014 * in the strings were. The current bytes may or may not be
8015 * at the beginning of a character. But neither or both are
8016 * (or else earlier bytes would have been different). And
8017 * if we are in the middle of a character, the two
8018 * characters are comprised of the same number of bytes
8019 * (because in this case the start bytes are the same, and
8020 * the start bytes encode the character's length). */
8021 if (UTF8_IS_INVARIANT(*pv1))
8023 /* If both are invariants; can just compare directly */
8024 if (UTF8_IS_INVARIANT(*pv2)) {
8025 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8027 else /* Since *pv1 is invariant, it is the whole character,
8028 which means it is at the beginning of a character.
8029 That means pv2 is also at the beginning of a
8030 character (see earlier comment). Since it isn't
8031 invariant, it must be a start byte. If it starts a
8032 character whose code point is above 255, that
8033 character is greater than any single-byte char, which
8035 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8040 /* Here, pv2 points to a character composed of 2 bytes
8041 * whose code point is < 256. Get its code point and
8042 * compare with *pv1 */
8043 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8048 else /* The code point starting at pv1 isn't a single byte */
8049 if (UTF8_IS_INVARIANT(*pv2))
8051 /* But here, the code point starting at *pv2 is a single byte,
8052 * and so *pv1 must begin a character, hence is a start byte.
8053 * If that character is above 255, it is larger than any
8054 * single-byte char, which *pv2 is */
8055 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8059 /* Here, pv1 points to a character composed of 2 bytes
8060 * whose code point is < 256. Get its code point and
8061 * compare with the single byte character *pv2 */
8062 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8067 else /* Here, we've ruled out either *pv1 and *pv2 being
8068 invariant. That means both are part of variants, but not
8069 necessarily at the start of a character */
8070 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8071 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8073 /* Here, at least one is the start of a character, which means
8074 * the other is also a start byte. And the code point of at
8075 * least one of the characters is above 255. It is a
8076 * characteristic of UTF-EBCDIC that all start bytes for
8077 * above-latin1 code points are well behaved as far as code
8078 * point comparisons go, and all are larger than all other
8079 * start bytes, so the comparison with those is also well
8081 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8084 /* Here both *pv1 and *pv2 are part of variant characters.
8085 * They could be both continuations, or both start characters.
8086 * (One or both could even be an illegal start character (for
8087 * an overlong) which for the purposes of sorting we treat as
8089 if (UTF8_IS_CONTINUATION(*pv1)) {
8091 /* If they are continuations for code points above 255,
8092 * then comparing the current byte is sufficient, as there
8093 * is no remapping of these and so the comparison is
8094 * well-behaved. We determine if they are such
8095 * continuations by looking at the preceding byte. It
8096 * could be a start byte, from which we can tell if it is
8097 * for an above 255 code point. Or it could be a
8098 * continuation, which means the character occupies at
8099 * least 3 bytes, so must be above 255. */
8100 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8101 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8103 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8107 /* Here, the continuations are for code points below 256;
8108 * back up one to get to the start byte */
8113 /* We need to get the actual native code point of each of these
8114 * variants in order to compare them */
8115 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8116 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8125 SvREFCNT_dec(svrecode);
8131 =for apidoc sv_cmp_locale
8133 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8134 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8135 if necessary. See also C<L</sv_cmp>>.
8137 =for apidoc sv_cmp_locale_flags
8139 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8140 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8141 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8142 C<L</sv_cmp_flags>>.
8148 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8150 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8154 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8157 #ifdef USE_LOCALE_COLLATE
8163 if (PL_collation_standard)
8168 /* Revert to using raw compare if both operands exist, but either one
8169 * doesn't transform properly for collation */
8171 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8175 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8181 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8182 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8185 if (!pv1 || !len1) {
8196 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8199 return retval < 0 ? -1 : 1;
8202 * When the result of collation is equality, that doesn't mean
8203 * that there are no differences -- some locales exclude some
8204 * characters from consideration. So to avoid false equalities,
8205 * we use the raw string as a tiebreaker.
8212 PERL_UNUSED_ARG(flags);
8213 #endif /* USE_LOCALE_COLLATE */
8215 return sv_cmp(sv1, sv2);
8219 #ifdef USE_LOCALE_COLLATE
8222 =for apidoc sv_collxfrm
8224 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8225 C<L</sv_collxfrm_flags>>.
8227 =for apidoc sv_collxfrm_flags
8229 Add Collate Transform magic to an SV if it doesn't already have it. If the
8230 flags contain C<SV_GMAGIC>, it handles get-magic.
8232 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8233 scalar data of the variable, but transformed to such a format that a normal
8234 memory comparison can be used to compare the data according to the locale
8241 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8245 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8247 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8249 /* If we don't have collation magic on 'sv', or the locale has changed
8250 * since the last time we calculated it, get it and save it now */
8251 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8256 /* Free the old space */
8258 Safefree(mg->mg_ptr);
8260 s = SvPV_flags_const(sv, len, flags);
8261 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8263 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8278 if (mg && mg->mg_ptr) {
8280 return mg->mg_ptr + sizeof(PL_collation_ix);
8288 #endif /* USE_LOCALE_COLLATE */
8291 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8293 SV * const tsv = newSV(0);
8296 sv_gets(tsv, fp, 0);
8297 sv_utf8_upgrade_nomg(tsv);
8298 SvCUR_set(sv,append);
8301 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8305 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8308 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8309 /* Grab the size of the record we're getting */
8310 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8317 /* With a true, record-oriented file on VMS, we need to use read directly
8318 * to ensure that we respect RMS record boundaries. The user is responsible
8319 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8320 * record size) field. N.B. This is likely to produce invalid results on
8321 * varying-width character data when a record ends mid-character.
8323 fd = PerlIO_fileno(fp);
8325 && PerlLIO_fstat(fd, &st) == 0
8326 && (st.st_fab_rfm == FAB$C_VAR
8327 || st.st_fab_rfm == FAB$C_VFC
8328 || st.st_fab_rfm == FAB$C_FIX)) {
8330 bytesread = PerlLIO_read(fd, buffer, recsize);
8332 else /* in-memory file from PerlIO::Scalar
8333 * or not a record-oriented file
8337 bytesread = PerlIO_read(fp, buffer, recsize);
8339 /* At this point, the logic in sv_get() means that sv will
8340 be treated as utf-8 if the handle is utf8.
8342 if (PerlIO_isutf8(fp) && bytesread > 0) {
8343 char *bend = buffer + bytesread;
8344 char *bufp = buffer;
8345 size_t charcount = 0;
8346 bool charstart = TRUE;
8349 while (charcount < recsize) {
8350 /* count accumulated characters */
8351 while (bufp < bend) {
8353 skip = UTF8SKIP(bufp);
8355 if (bufp + skip > bend) {
8356 /* partial at the end */
8367 if (charcount < recsize) {
8369 STRLEN bufp_offset = bufp - buffer;
8370 SSize_t morebytesread;
8372 /* originally I read enough to fill any incomplete
8373 character and the first byte of the next
8374 character if needed, but if there's many
8375 multi-byte encoded characters we're going to be
8376 making a read call for every character beyond
8377 the original read size.
8379 So instead, read the rest of the character if
8380 any, and enough bytes to match at least the
8381 start bytes for each character we're going to
8385 readsize = recsize - charcount;
8387 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8388 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8389 bend = buffer + bytesread;
8390 morebytesread = PerlIO_read(fp, bend, readsize);
8391 if (morebytesread <= 0) {
8392 /* we're done, if we still have incomplete
8393 characters the check code in sv_gets() will
8396 I'd originally considered doing
8397 PerlIO_ungetc() on all but the lead
8398 character of the incomplete character, but
8399 read() doesn't do that, so I don't.
8404 /* prepare to scan some more */
8405 bytesread += morebytesread;
8406 bend = buffer + bytesread;
8407 bufp = buffer + bufp_offset;
8415 SvCUR_set(sv, bytesread + append);
8416 buffer[bytesread] = '\0';
8417 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8423 Get a line from the filehandle and store it into the SV, optionally
8424 appending to the currently-stored string. If C<append> is not 0, the
8425 line is appended to the SV instead of overwriting it. C<append> should
8426 be set to the byte offset that the appended string should start at
8427 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8433 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8443 PERL_ARGS_ASSERT_SV_GETS;
8445 if (SvTHINKFIRST(sv))
8446 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8447 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8449 However, perlbench says it's slower, because the existing swipe code
8450 is faster than copy on write.
8451 Swings and roundabouts. */
8452 SvUPGRADE(sv, SVt_PV);
8455 /* line is going to be appended to the existing buffer in the sv */
8456 if (PerlIO_isutf8(fp)) {
8458 sv_utf8_upgrade_nomg(sv);
8459 sv_pos_u2b(sv,&append,0);
8461 } else if (SvUTF8(sv)) {
8462 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8468 /* not appending - "clear" the string by setting SvCUR to 0,
8469 * the pv is still avaiable. */
8472 if (PerlIO_isutf8(fp))
8475 if (IN_PERL_COMPILETIME) {
8476 /* we always read code in line mode */
8480 else if (RsSNARF(PL_rs)) {
8481 /* If it is a regular disk file use size from stat() as estimate
8482 of amount we are going to read -- may result in mallocing
8483 more memory than we really need if the layers below reduce
8484 the size we read (e.g. CRLF or a gzip layer).
8487 int fd = PerlIO_fileno(fp);
8488 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8489 const Off_t offset = PerlIO_tell(fp);
8490 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8491 #ifdef PERL_COPY_ON_WRITE
8492 /* Add an extra byte for the sake of copy-on-write's
8493 * buffer reference count. */
8494 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8496 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8503 else if (RsRECORD(PL_rs)) {
8504 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8506 else if (RsPARA(PL_rs)) {
8512 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8513 if (PerlIO_isutf8(fp)) {
8514 rsptr = SvPVutf8(PL_rs, rslen);
8517 if (SvUTF8(PL_rs)) {
8518 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8519 Perl_croak(aTHX_ "Wide character in $/");
8522 /* extract the raw pointer to the record separator */
8523 rsptr = SvPV_const(PL_rs, rslen);
8527 /* rslast is the last character in the record separator
8528 * note we don't use rslast except when rslen is true, so the
8529 * null assign is a placeholder. */
8530 rslast = rslen ? rsptr[rslen - 1] : '\0';
8532 if (rspara) { /* have to do this both before and after */
8533 do { /* to make sure file boundaries work right */
8536 i = PerlIO_getc(fp);
8540 PerlIO_ungetc(fp,i);
8546 /* See if we know enough about I/O mechanism to cheat it ! */
8548 /* This used to be #ifdef test - it is made run-time test for ease
8549 of abstracting out stdio interface. One call should be cheap
8550 enough here - and may even be a macro allowing compile
8554 if (PerlIO_fast_gets(fp)) {
8556 * We can do buffer based IO operations on this filehandle.
8558 * This means we can bypass a lot of subcalls and process
8559 * the buffer directly, it also means we know the upper bound
8560 * on the amount of data we might read of the current buffer
8561 * into our sv. Knowing this allows us to preallocate the pv
8562 * to be able to hold that maximum, which allows us to simplify
8563 * a lot of logic. */
8566 * We're going to steal some values from the stdio struct
8567 * and put EVERYTHING in the innermost loop into registers.
8569 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8570 STRLEN bpx; /* length of the data in the target sv
8571 used to fix pointers after a SvGROW */
8572 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8573 of data left in the read-ahead buffer.
8574 If 0 then the pv buffer can hold the full
8575 amount left, otherwise this is the amount it
8578 /* Here is some breathtakingly efficient cheating */
8580 /* When you read the following logic resist the urge to think
8581 * of record separators that are 1 byte long. They are an
8582 * uninteresting special (simple) case.
8584 * Instead think of record separators which are at least 2 bytes
8585 * long, and keep in mind that we need to deal with such
8586 * separators when they cross a read-ahead buffer boundary.
8588 * Also consider that we need to gracefully deal with separators
8589 * that may be longer than a single read ahead buffer.
8591 * Lastly do not forget we want to copy the delimiter as well. We
8592 * are copying all data in the file _up_to_and_including_ the separator
8595 * Now that you have all that in mind here is what is happening below:
8597 * 1. When we first enter the loop we do some memory book keeping to see
8598 * how much free space there is in the target SV. (This sub assumes that
8599 * it is operating on the same SV most of the time via $_ and that it is
8600 * going to be able to reuse the same pv buffer each call.) If there is
8601 * "enough" room then we set "shortbuffered" to how much space there is
8602 * and start reading forward.
8604 * 2. When we scan forward we copy from the read-ahead buffer to the target
8605 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8606 * and the end of the of pv, as well as for the "rslast", which is the last
8607 * char of the separator.
8609 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8610 * (which has a "complete" record up to the point we saw rslast) and check
8611 * it to see if it matches the separator. If it does we are done. If it doesn't
8612 * we continue on with the scan/copy.
8614 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8615 * the IO system to read the next buffer. We do this by doing a getc(), which
8616 * returns a single char read (or EOF), and prefills the buffer, and also
8617 * allows us to find out how full the buffer is. We use this information to
8618 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8619 * the returned single char into the target sv, and then go back into scan
8622 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8623 * remaining space in the read-buffer.
8625 * Note that this code despite its twisty-turny nature is pretty darn slick.
8626 * It manages single byte separators, multi-byte cross boundary separators,
8627 * and cross-read-buffer separators cleanly and efficiently at the cost
8628 * of potentially greatly overallocating the target SV.
8634 /* get the number of bytes remaining in the read-ahead buffer
8635 * on first call on a given fp this will return 0.*/
8636 cnt = PerlIO_get_cnt(fp);
8638 /* make sure we have the room */
8639 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8640 /* Not room for all of it
8641 if we are looking for a separator and room for some
8643 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8644 /* just process what we have room for */
8645 shortbuffered = cnt - SvLEN(sv) + append + 1;
8646 cnt -= shortbuffered;
8649 /* ensure that the target sv has enough room to hold
8650 * the rest of the read-ahead buffer */
8652 /* remember that cnt can be negative */
8653 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8657 /* we have enough room to hold the full buffer, lets scream */
8661 /* extract the pointer to sv's string buffer, offset by append as necessary */
8662 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8663 /* extract the point to the read-ahead buffer */
8664 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8666 /* some trace debug output */
8667 DEBUG_P(PerlIO_printf(Perl_debug_log,
8668 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8669 DEBUG_P(PerlIO_printf(Perl_debug_log,
8670 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8672 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8673 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8677 /* if there is stuff left in the read-ahead buffer */
8679 /* if there is a separator */
8681 /* find next rslast */
8684 /* shortcut common case of blank line */
8686 if ((*bp++ = *ptr++) == rslast)
8687 goto thats_all_folks;
8689 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8691 SSize_t got = p - ptr + 1;
8692 Copy(ptr, bp, got, STDCHAR);
8696 goto thats_all_folks;
8698 Copy(ptr, bp, cnt, STDCHAR);
8704 /* no separator, slurp the full buffer */
8705 Copy(ptr, bp, cnt, char); /* this | eat */
8706 bp += cnt; /* screams | dust */
8707 ptr += cnt; /* louder | sed :-) */
8709 assert (!shortbuffered);
8710 goto cannot_be_shortbuffered;
8714 if (shortbuffered) { /* oh well, must extend */
8715 /* we didnt have enough room to fit the line into the target buffer
8716 * so we must extend the target buffer and keep going */
8717 cnt = shortbuffered;
8719 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8721 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8722 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8723 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8727 cannot_be_shortbuffered:
8728 /* we need to refill the read-ahead buffer if possible */
8730 DEBUG_P(PerlIO_printf(Perl_debug_log,
8731 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8732 PTR2UV(ptr),(IV)cnt));
8733 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8735 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8736 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8737 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8738 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8741 call PerlIO_getc() to let it prefill the lookahead buffer
8743 This used to call 'filbuf' in stdio form, but as that behaves like
8744 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8745 another abstraction.
8747 Note we have to deal with the char in 'i' if we are not at EOF
8749 i = PerlIO_getc(fp); /* get more characters */
8751 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8752 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8753 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8754 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8756 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8757 cnt = PerlIO_get_cnt(fp);
8758 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8759 DEBUG_P(PerlIO_printf(Perl_debug_log,
8760 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8761 PTR2UV(ptr),(IV)cnt));
8763 if (i == EOF) /* all done for ever? */
8764 goto thats_really_all_folks;
8766 /* make sure we have enough space in the target sv */
8767 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8769 SvGROW(sv, bpx + cnt + 2);
8770 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8772 /* copy of the char we got from getc() */
8773 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8775 /* make sure we deal with the i being the last character of a separator */
8776 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8777 goto thats_all_folks;
8781 /* check if we have actually found the separator - only really applies
8783 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8784 memNE((char*)bp - rslen, rsptr, rslen))
8785 goto screamer; /* go back to the fray */
8786 thats_really_all_folks:
8788 cnt += shortbuffered;
8789 DEBUG_P(PerlIO_printf(Perl_debug_log,
8790 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8791 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8792 DEBUG_P(PerlIO_printf(Perl_debug_log,
8793 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8795 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8796 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8798 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8799 DEBUG_P(PerlIO_printf(Perl_debug_log,
8800 "Screamer: done, len=%ld, string=|%.*s|\n",
8801 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8805 /*The big, slow, and stupid way. */
8806 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8807 STDCHAR *buf = NULL;
8808 Newx(buf, 8192, STDCHAR);
8816 const STDCHAR * const bpe = buf + sizeof(buf);
8818 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8819 ; /* keep reading */
8823 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8824 /* Accommodate broken VAXC compiler, which applies U8 cast to
8825 * both args of ?: operator, causing EOF to change into 255
8828 i = (U8)buf[cnt - 1];
8834 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8836 sv_catpvn_nomg(sv, (char *) buf, cnt);
8838 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8840 if (i != EOF && /* joy */
8842 SvCUR(sv) < rslen ||
8843 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8847 * If we're reading from a TTY and we get a short read,
8848 * indicating that the user hit his EOF character, we need
8849 * to notice it now, because if we try to read from the TTY
8850 * again, the EOF condition will disappear.
8852 * The comparison of cnt to sizeof(buf) is an optimization
8853 * that prevents unnecessary calls to feof().
8857 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8861 #ifdef USE_HEAP_INSTEAD_OF_STACK
8866 if (rspara) { /* have to do this both before and after */
8867 while (i != EOF) { /* to make sure file boundaries work right */
8868 i = PerlIO_getc(fp);
8870 PerlIO_ungetc(fp,i);
8876 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8882 Auto-increment of the value in the SV, doing string to numeric conversion
8883 if necessary. Handles 'get' magic and operator overloading.
8889 Perl_sv_inc(pTHX_ SV *const sv)
8898 =for apidoc sv_inc_nomg
8900 Auto-increment of the value in the SV, doing string to numeric conversion
8901 if necessary. Handles operator overloading. Skips handling 'get' magic.
8907 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8914 if (SvTHINKFIRST(sv)) {
8915 if (SvREADONLY(sv)) {
8916 Perl_croak_no_modify();
8920 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8922 i = PTR2IV(SvRV(sv));
8926 else sv_force_normal_flags(sv, 0);
8928 flags = SvFLAGS(sv);
8929 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8930 /* It's (privately or publicly) a float, but not tested as an
8931 integer, so test it to see. */
8933 flags = SvFLAGS(sv);
8935 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8936 /* It's publicly an integer, or privately an integer-not-float */
8937 #ifdef PERL_PRESERVE_IVUV
8941 if (SvUVX(sv) == UV_MAX)
8942 sv_setnv(sv, UV_MAX_P1);
8944 (void)SvIOK_only_UV(sv);
8945 SvUV_set(sv, SvUVX(sv) + 1);
8947 if (SvIVX(sv) == IV_MAX)
8948 sv_setuv(sv, (UV)IV_MAX + 1);
8950 (void)SvIOK_only(sv);
8951 SvIV_set(sv, SvIVX(sv) + 1);
8956 if (flags & SVp_NOK) {
8957 const NV was = SvNVX(sv);
8958 if (LIKELY(!Perl_isinfnan(was)) &&
8959 NV_OVERFLOWS_INTEGERS_AT &&
8960 was >= NV_OVERFLOWS_INTEGERS_AT) {
8961 /* diag_listed_as: Lost precision when %s %f by 1 */
8962 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8963 "Lost precision when incrementing %" NVff " by 1",
8966 (void)SvNOK_only(sv);
8967 SvNV_set(sv, was + 1.0);
8971 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8972 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8973 Perl_croak_no_modify();
8975 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8976 if ((flags & SVTYPEMASK) < SVt_PVIV)
8977 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8978 (void)SvIOK_only(sv);
8983 while (isALPHA(*d)) d++;
8984 while (isDIGIT(*d)) d++;
8985 if (d < SvEND(sv)) {
8986 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8987 #ifdef PERL_PRESERVE_IVUV
8988 /* Got to punt this as an integer if needs be, but we don't issue
8989 warnings. Probably ought to make the sv_iv_please() that does
8990 the conversion if possible, and silently. */
8991 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8992 /* Need to try really hard to see if it's an integer.
8993 9.22337203685478e+18 is an integer.
8994 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8995 so $a="9.22337203685478e+18"; $a+0; $a++
8996 needs to be the same as $a="9.22337203685478e+18"; $a++
9003 /* sv_2iv *should* have made this an NV */
9004 if (flags & SVp_NOK) {
9005 (void)SvNOK_only(sv);
9006 SvNV_set(sv, SvNVX(sv) + 1.0);
9009 /* I don't think we can get here. Maybe I should assert this
9010 And if we do get here I suspect that sv_setnv will croak. NWC
9012 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9013 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9015 #endif /* PERL_PRESERVE_IVUV */
9016 if (!numtype && ckWARN(WARN_NUMERIC))
9017 not_incrementable(sv);
9018 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9022 while (d >= SvPVX_const(sv)) {
9030 /* MKS: The original code here died if letters weren't consecutive.
9031 * at least it didn't have to worry about non-C locales. The
9032 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9033 * arranged in order (although not consecutively) and that only
9034 * [A-Za-z] are accepted by isALPHA in the C locale.
9036 if (isALPHA_FOLD_NE(*d, 'z')) {
9037 do { ++*d; } while (!isALPHA(*d));
9040 *(d--) -= 'z' - 'a';
9045 *(d--) -= 'z' - 'a' + 1;
9049 /* oh,oh, the number grew */
9050 SvGROW(sv, SvCUR(sv) + 2);
9051 SvCUR_set(sv, SvCUR(sv) + 1);
9052 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9063 Auto-decrement of the value in the SV, doing string to numeric conversion
9064 if necessary. Handles 'get' magic and operator overloading.
9070 Perl_sv_dec(pTHX_ SV *const sv)
9079 =for apidoc sv_dec_nomg
9081 Auto-decrement of the value in the SV, doing string to numeric conversion
9082 if necessary. Handles operator overloading. Skips handling 'get' magic.
9088 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9094 if (SvTHINKFIRST(sv)) {
9095 if (SvREADONLY(sv)) {
9096 Perl_croak_no_modify();
9100 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9102 i = PTR2IV(SvRV(sv));
9106 else sv_force_normal_flags(sv, 0);
9108 /* Unlike sv_inc we don't have to worry about string-never-numbers
9109 and keeping them magic. But we mustn't warn on punting */
9110 flags = SvFLAGS(sv);
9111 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9112 /* It's publicly an integer, or privately an integer-not-float */
9113 #ifdef PERL_PRESERVE_IVUV
9117 if (SvUVX(sv) == 0) {
9118 (void)SvIOK_only(sv);
9122 (void)SvIOK_only_UV(sv);
9123 SvUV_set(sv, SvUVX(sv) - 1);
9126 if (SvIVX(sv) == IV_MIN) {
9127 sv_setnv(sv, (NV)IV_MIN);
9131 (void)SvIOK_only(sv);
9132 SvIV_set(sv, SvIVX(sv) - 1);
9137 if (flags & SVp_NOK) {
9140 const NV was = SvNVX(sv);
9141 if (LIKELY(!Perl_isinfnan(was)) &&
9142 NV_OVERFLOWS_INTEGERS_AT &&
9143 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9144 /* diag_listed_as: Lost precision when %s %f by 1 */
9145 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9146 "Lost precision when decrementing %" NVff " by 1",
9149 (void)SvNOK_only(sv);
9150 SvNV_set(sv, was - 1.0);
9155 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9156 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9157 Perl_croak_no_modify();
9159 if (!(flags & SVp_POK)) {
9160 if ((flags & SVTYPEMASK) < SVt_PVIV)
9161 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9163 (void)SvIOK_only(sv);
9166 #ifdef PERL_PRESERVE_IVUV
9168 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9169 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9170 /* Need to try really hard to see if it's an integer.
9171 9.22337203685478e+18 is an integer.
9172 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9173 so $a="9.22337203685478e+18"; $a+0; $a--
9174 needs to be the same as $a="9.22337203685478e+18"; $a--
9181 /* sv_2iv *should* have made this an NV */
9182 if (flags & SVp_NOK) {
9183 (void)SvNOK_only(sv);
9184 SvNV_set(sv, SvNVX(sv) - 1.0);
9187 /* I don't think we can get here. Maybe I should assert this
9188 And if we do get here I suspect that sv_setnv will croak. NWC
9190 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9191 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9194 #endif /* PERL_PRESERVE_IVUV */
9195 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9198 /* this define is used to eliminate a chunk of duplicated but shared logic
9199 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9200 * used anywhere but here - yves
9202 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9204 SSize_t ix = ++PL_tmps_ix; \
9205 if (UNLIKELY(ix >= PL_tmps_max)) \
9206 ix = tmps_grow_p(ix); \
9207 PL_tmps_stack[ix] = (AnSv); \
9211 =for apidoc sv_mortalcopy
9213 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9214 The new SV is marked as mortal. It will be destroyed "soon", either by an
9215 explicit call to C<FREETMPS>, or by an implicit call at places such as
9216 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9221 /* Make a string that will exist for the duration of the expression
9222 * evaluation. Actually, it may have to last longer than that, but
9223 * hopefully we won't free it until it has been assigned to a
9224 * permanent location. */
9227 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9231 if (flags & SV_GMAGIC)
9232 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9234 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9235 PUSH_EXTEND_MORTAL__SV_C(sv);
9241 =for apidoc sv_newmortal
9243 Creates a new null SV which is mortal. The reference count of the SV is
9244 set to 1. It will be destroyed "soon", either by an explicit call to
9245 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9246 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9252 Perl_sv_newmortal(pTHX)
9257 SvFLAGS(sv) = SVs_TEMP;
9258 PUSH_EXTEND_MORTAL__SV_C(sv);
9264 =for apidoc newSVpvn_flags
9266 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9267 characters) into it. The reference count for the
9268 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9269 string. You are responsible for ensuring that the source string is at least
9270 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9271 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9272 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9273 returning. If C<SVf_UTF8> is set, C<s>
9274 is considered to be in UTF-8 and the
9275 C<SVf_UTF8> flag will be set on the new SV.
9276 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9278 #define newSVpvn_utf8(s, len, u) \
9279 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9285 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9289 /* All the flags we don't support must be zero.
9290 And we're new code so I'm going to assert this from the start. */
9291 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9293 sv_setpvn(sv,s,len);
9295 /* This code used to do a sv_2mortal(), however we now unroll the call to
9296 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9297 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9298 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9299 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9300 * means that we eliminate quite a few steps than it looks - Yves
9301 * (explaining patch by gfx) */
9303 SvFLAGS(sv) |= flags;
9305 if(flags & SVs_TEMP){
9306 PUSH_EXTEND_MORTAL__SV_C(sv);
9313 =for apidoc sv_2mortal
9315 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9316 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9317 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9318 string buffer can be "stolen" if this SV is copied. See also
9319 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9325 Perl_sv_2mortal(pTHX_ SV *const sv)
9332 PUSH_EXTEND_MORTAL__SV_C(sv);
9340 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9341 characters) into it. The reference count for the
9342 SV is set to 1. If C<len> is zero, Perl will compute the length using
9343 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9344 C<NUL> characters and has to have a terminating C<NUL> byte).
9346 This function can cause reliability issues if you are likely to pass in
9347 empty strings that are not null terminated, because it will run
9348 strlen on the string and potentially run past valid memory.
9350 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9351 For string literals use L</newSVpvs> instead. This function will work fine for
9352 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9353 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9359 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9364 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9369 =for apidoc newSVpvn
9371 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9372 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9373 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9374 are responsible for ensuring that the source buffer is at least
9375 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9382 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9386 sv_setpvn(sv,buffer,len);
9391 =for apidoc newSVhek
9393 Creates a new SV from the hash key structure. It will generate scalars that
9394 point to the shared string table where possible. Returns a new (undefined)
9395 SV if C<hek> is NULL.
9401 Perl_newSVhek(pTHX_ const HEK *const hek)
9410 if (HEK_LEN(hek) == HEf_SVKEY) {
9411 return newSVsv(*(SV**)HEK_KEY(hek));
9413 const int flags = HEK_FLAGS(hek);
9414 if (flags & HVhek_WASUTF8) {
9416 Andreas would like keys he put in as utf8 to come back as utf8
9418 STRLEN utf8_len = HEK_LEN(hek);
9419 SV * const sv = newSV_type(SVt_PV);
9420 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9421 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9422 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9425 } else if (flags & HVhek_UNSHARED) {
9426 /* A hash that isn't using shared hash keys has to have
9427 the flag in every key so that we know not to try to call
9428 share_hek_hek on it. */
9430 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9435 /* This will be overwhelminly the most common case. */
9437 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9438 more efficient than sharepvn(). */
9442 sv_upgrade(sv, SVt_PV);
9443 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9444 SvCUR_set(sv, HEK_LEN(hek));
9456 =for apidoc newSVpvn_share
9458 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9459 table. If the string does not already exist in the table, it is
9460 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9461 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9462 is non-zero, that value is used; otherwise the hash is computed.
9463 The string's hash can later be retrieved from the SV
9464 with the C<SvSHARED_HASH()> macro. The idea here is
9465 that as the string table is used for shared hash keys these strings will have
9466 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9472 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9476 bool is_utf8 = FALSE;
9477 const char *const orig_src = src;
9480 STRLEN tmplen = -len;
9482 /* See the note in hv.c:hv_fetch() --jhi */
9483 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9487 PERL_HASH(hash, src, len);
9489 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9490 changes here, update it there too. */
9491 sv_upgrade(sv, SVt_PV);
9492 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9499 if (src != orig_src)
9505 =for apidoc newSVpv_share
9507 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9514 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9516 return newSVpvn_share(src, strlen(src), hash);
9519 #if defined(PERL_IMPLICIT_CONTEXT)
9521 /* pTHX_ magic can't cope with varargs, so this is a no-context
9522 * version of the main function, (which may itself be aliased to us).
9523 * Don't access this version directly.
9527 Perl_newSVpvf_nocontext(const char *const pat, ...)
9533 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9535 va_start(args, pat);
9536 sv = vnewSVpvf(pat, &args);
9543 =for apidoc newSVpvf
9545 Creates a new SV and initializes it with the string formatted like
9552 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9557 PERL_ARGS_ASSERT_NEWSVPVF;
9559 va_start(args, pat);
9560 sv = vnewSVpvf(pat, &args);
9565 /* backend for newSVpvf() and newSVpvf_nocontext() */
9568 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9572 PERL_ARGS_ASSERT_VNEWSVPVF;
9575 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9582 Creates a new SV and copies a floating point value into it.
9583 The reference count for the SV is set to 1.
9589 Perl_newSVnv(pTHX_ const NV n)
9601 Creates a new SV and copies an integer into it. The reference count for the
9608 Perl_newSViv(pTHX_ const IV i)
9614 /* Inlining ONLY the small relevant subset of sv_setiv here
9615 * for performance. Makes a significant difference. */
9617 /* We're starting from SVt_FIRST, so provided that's
9618 * actual 0, we don't have to unset any SV type flags
9619 * to promote to SVt_IV. */
9620 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9622 SET_SVANY_FOR_BODYLESS_IV(sv);
9623 SvFLAGS(sv) |= SVt_IV;
9635 Creates a new SV and copies an unsigned integer into it.
9636 The reference count for the SV is set to 1.
9642 Perl_newSVuv(pTHX_ const UV u)
9646 /* Inlining ONLY the small relevant subset of sv_setuv here
9647 * for performance. Makes a significant difference. */
9649 /* Using ivs is more efficient than using uvs - see sv_setuv */
9650 if (u <= (UV)IV_MAX) {
9651 return newSViv((IV)u);
9656 /* We're starting from SVt_FIRST, so provided that's
9657 * actual 0, we don't have to unset any SV type flags
9658 * to promote to SVt_IV. */
9659 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9661 SET_SVANY_FOR_BODYLESS_IV(sv);
9662 SvFLAGS(sv) |= SVt_IV;
9664 (void)SvIsUV_on(sv);
9673 =for apidoc newSV_type
9675 Creates a new SV, of the type specified. The reference count for the new SV
9682 Perl_newSV_type(pTHX_ const svtype type)
9687 ASSUME(SvTYPE(sv) == SVt_FIRST);
9688 if(type != SVt_FIRST)
9689 sv_upgrade(sv, type);
9694 =for apidoc newRV_noinc
9696 Creates an RV wrapper for an SV. The reference count for the original
9697 SV is B<not> incremented.
9703 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9707 PERL_ARGS_ASSERT_NEWRV_NOINC;
9711 /* We're starting from SVt_FIRST, so provided that's
9712 * actual 0, we don't have to unset any SV type flags
9713 * to promote to SVt_IV. */
9714 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9716 SET_SVANY_FOR_BODYLESS_IV(sv);
9717 SvFLAGS(sv) |= SVt_IV;
9722 SvRV_set(sv, tmpRef);
9727 /* newRV_inc is the official function name to use now.
9728 * newRV_inc is in fact #defined to newRV in sv.h
9732 Perl_newRV(pTHX_ SV *const sv)
9734 PERL_ARGS_ASSERT_NEWRV;
9736 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9742 Creates a new SV which is an exact duplicate of the original SV.
9749 Perl_newSVsv(pTHX_ SV *const old)
9755 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9756 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9759 /* Do this here, otherwise we leak the new SV if this croaks. */
9762 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9763 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9764 sv_setsv_flags(sv, old, SV_NOSTEAL);
9769 =for apidoc sv_reset
9771 Underlying implementation for the C<reset> Perl function.
9772 Note that the perl-level function is vaguely deprecated.
9778 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9780 PERL_ARGS_ASSERT_SV_RESET;
9782 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9786 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9788 char todo[PERL_UCHAR_MAX+1];
9791 if (!stash || SvTYPE(stash) != SVt_PVHV)
9794 if (!s) { /* reset ?? searches */
9795 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9797 const U32 count = mg->mg_len / sizeof(PMOP**);
9798 PMOP **pmp = (PMOP**) mg->mg_ptr;
9799 PMOP *const *const end = pmp + count;
9803 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9805 (*pmp)->op_pmflags &= ~PMf_USED;
9813 /* reset variables */
9815 if (!HvARRAY(stash))
9818 Zero(todo, 256, char);
9822 I32 i = (unsigned char)*s;
9826 max = (unsigned char)*s++;
9827 for ( ; i <= max; i++) {
9830 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9832 for (entry = HvARRAY(stash)[i];
9834 entry = HeNEXT(entry))
9839 if (!todo[(U8)*HeKEY(entry)])
9841 gv = MUTABLE_GV(HeVAL(entry));
9845 if (sv && !SvREADONLY(sv)) {
9846 SV_CHECK_THINKFIRST_COW_DROP(sv);
9847 if (!isGV(sv)) SvOK_off(sv);
9852 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9863 Using various gambits, try to get an IO from an SV: the IO slot if its a
9864 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9865 named after the PV if we're a string.
9867 'Get' magic is ignored on the C<sv> passed in, but will be called on
9868 C<SvRV(sv)> if C<sv> is an RV.
9874 Perl_sv_2io(pTHX_ SV *const sv)
9879 PERL_ARGS_ASSERT_SV_2IO;
9881 switch (SvTYPE(sv)) {
9883 io = MUTABLE_IO(sv);
9887 if (isGV_with_GP(sv)) {
9888 gv = MUTABLE_GV(sv);
9891 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9892 HEKfARG(GvNAME_HEK(gv)));
9898 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9900 SvGETMAGIC(SvRV(sv));
9901 return sv_2io(SvRV(sv));
9903 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9910 if (SvGMAGICAL(sv)) {
9911 newsv = sv_newmortal();
9912 sv_setsv_nomg(newsv, sv);
9914 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9924 Using various gambits, try to get a CV from an SV; in addition, try if
9925 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9926 The flags in C<lref> are passed to C<gv_fetchsv>.
9932 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9937 PERL_ARGS_ASSERT_SV_2CV;
9944 switch (SvTYPE(sv)) {
9948 return MUTABLE_CV(sv);
9958 sv = amagic_deref_call(sv, to_cv_amg);
9961 if (SvTYPE(sv) == SVt_PVCV) {
9962 cv = MUTABLE_CV(sv);
9967 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9968 gv = MUTABLE_GV(sv);
9970 Perl_croak(aTHX_ "Not a subroutine reference");
9972 else if (isGV_with_GP(sv)) {
9973 gv = MUTABLE_GV(sv);
9976 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9983 /* Some flags to gv_fetchsv mean don't really create the GV */
9984 if (!isGV_with_GP(gv)) {
9989 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9990 /* XXX this is probably not what they think they're getting.
9991 * It has the same effect as "sub name;", i.e. just a forward
10000 =for apidoc sv_true
10002 Returns true if the SV has a true value by Perl's rules.
10003 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10004 instead use an in-line version.
10010 Perl_sv_true(pTHX_ SV *const sv)
10015 const XPV* const tXpv = (XPV*)SvANY(sv);
10017 (tXpv->xpv_cur > 1 ||
10018 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10025 return SvIVX(sv) != 0;
10028 return SvNVX(sv) != 0.0;
10030 return sv_2bool(sv);
10036 =for apidoc sv_pvn_force
10038 Get a sensible string out of the SV somehow.
10039 A private implementation of the C<SvPV_force> macro for compilers which
10040 can't cope with complex macro expressions. Always use the macro instead.
10042 =for apidoc sv_pvn_force_flags
10044 Get a sensible string out of the SV somehow.
10045 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10046 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10047 implemented in terms of this function.
10048 You normally want to use the various wrapper macros instead: see
10049 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10055 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10057 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10059 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10060 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10061 sv_force_normal_flags(sv, 0);
10071 if (SvTYPE(sv) > SVt_PVLV
10072 || isGV_with_GP(sv))
10073 /* diag_listed_as: Can't coerce %s to %s in %s */
10074 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10076 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10083 if (SvTYPE(sv) < SVt_PV ||
10084 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10087 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10088 SvGROW(sv, len + 1);
10089 Move(s,SvPVX(sv),len,char);
10090 SvCUR_set(sv, len);
10091 SvPVX(sv)[len] = '\0';
10094 SvPOK_on(sv); /* validate pointer */
10096 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10097 PTR2UV(sv),SvPVX_const(sv)));
10100 (void)SvPOK_only_UTF8(sv);
10101 return SvPVX_mutable(sv);
10105 =for apidoc sv_pvbyten_force
10107 The backend for the C<SvPVbytex_force> macro. Always use the macro
10114 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10116 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10118 sv_pvn_force(sv,lp);
10119 sv_utf8_downgrade(sv,0);
10125 =for apidoc sv_pvutf8n_force
10127 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10134 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10136 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10138 sv_pvn_force(sv,0);
10139 sv_utf8_upgrade_nomg(sv);
10145 =for apidoc sv_reftype
10147 Returns a string describing what the SV is a reference to.
10149 If ob is true and the SV is blessed, the string is the class name,
10150 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10156 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10158 PERL_ARGS_ASSERT_SV_REFTYPE;
10159 if (ob && SvOBJECT(sv)) {
10160 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10163 /* WARNING - There is code, for instance in mg.c, that assumes that
10164 * the only reason that sv_reftype(sv,0) would return a string starting
10165 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10166 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10167 * this routine inside other subs, and it saves time.
10168 * Do not change this assumption without searching for "dodgy type check" in
10171 switch (SvTYPE(sv)) {
10186 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10187 /* tied lvalues should appear to be
10188 * scalars for backwards compatibility */
10189 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10190 ? "SCALAR" : "LVALUE");
10191 case SVt_PVAV: return "ARRAY";
10192 case SVt_PVHV: return "HASH";
10193 case SVt_PVCV: return "CODE";
10194 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10195 ? "GLOB" : "SCALAR");
10196 case SVt_PVFM: return "FORMAT";
10197 case SVt_PVIO: return "IO";
10198 case SVt_INVLIST: return "INVLIST";
10199 case SVt_REGEXP: return "REGEXP";
10200 default: return "UNKNOWN";
10208 Returns a SV describing what the SV passed in is a reference to.
10210 dst can be a SV to be set to the description or NULL, in which case a
10211 mortal SV is returned.
10213 If ob is true and the SV is blessed, the description is the class
10214 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10220 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10222 PERL_ARGS_ASSERT_SV_REF;
10225 dst = sv_newmortal();
10227 if (ob && SvOBJECT(sv)) {
10228 HvNAME_get(SvSTASH(sv))
10229 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10230 : sv_setpvs(dst, "__ANON__");
10233 const char * reftype = sv_reftype(sv, 0);
10234 sv_setpv(dst, reftype);
10240 =for apidoc sv_isobject
10242 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10243 object. If the SV is not an RV, or if the object is not blessed, then this
10250 Perl_sv_isobject(pTHX_ SV *sv)
10266 Returns a boolean indicating whether the SV is blessed into the specified
10267 class. This does not check for subtypes; use C<sv_derived_from> to verify
10268 an inheritance relationship.
10274 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10276 const char *hvname;
10278 PERL_ARGS_ASSERT_SV_ISA;
10288 hvname = HvNAME_get(SvSTASH(sv));
10292 return strEQ(hvname, name);
10296 =for apidoc newSVrv
10298 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10299 RV then it will be upgraded to one. If C<classname> is non-null then the new
10300 SV will be blessed in the specified package. The new SV is returned and its
10301 reference count is 1. The reference count 1 is owned by C<rv>.
10307 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10311 PERL_ARGS_ASSERT_NEWSVRV;
10315 SV_CHECK_THINKFIRST_COW_DROP(rv);
10317 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10318 const U32 refcnt = SvREFCNT(rv);
10322 SvREFCNT(rv) = refcnt;
10324 sv_upgrade(rv, SVt_IV);
10325 } else if (SvROK(rv)) {
10326 SvREFCNT_dec(SvRV(rv));
10328 prepare_SV_for_RV(rv);
10336 HV* const stash = gv_stashpv(classname, GV_ADD);
10337 (void)sv_bless(rv, stash);
10343 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10345 SV * const lv = newSV_type(SVt_PVLV);
10346 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10348 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10349 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10350 LvSTARGOFF(lv) = ix;
10351 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10356 =for apidoc sv_setref_pv
10358 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10359 argument will be upgraded to an RV. That RV will be modified to point to
10360 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10361 into the SV. The C<classname> argument indicates the package for the
10362 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10363 will have a reference count of 1, and the RV will be returned.
10365 Do not use with other Perl types such as HV, AV, SV, CV, because those
10366 objects will become corrupted by the pointer copy process.
10368 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10374 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10376 PERL_ARGS_ASSERT_SV_SETREF_PV;
10383 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10388 =for apidoc sv_setref_iv
10390 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10391 argument will be upgraded to an RV. That RV will be modified to point to
10392 the new SV. The C<classname> argument indicates the package for the
10393 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10394 will have a reference count of 1, and the RV will be returned.
10400 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10402 PERL_ARGS_ASSERT_SV_SETREF_IV;
10404 sv_setiv(newSVrv(rv,classname), iv);
10409 =for apidoc sv_setref_uv
10411 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10412 argument will be upgraded to an RV. That RV will be modified to point to
10413 the new SV. The C<classname> argument indicates the package for the
10414 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10415 will have a reference count of 1, and the RV will be returned.
10421 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10423 PERL_ARGS_ASSERT_SV_SETREF_UV;
10425 sv_setuv(newSVrv(rv,classname), uv);
10430 =for apidoc sv_setref_nv
10432 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10433 argument will be upgraded to an RV. That RV will be modified to point to
10434 the new SV. The C<classname> argument indicates the package for the
10435 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10436 will have a reference count of 1, and the RV will be returned.
10442 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10444 PERL_ARGS_ASSERT_SV_SETREF_NV;
10446 sv_setnv(newSVrv(rv,classname), nv);
10451 =for apidoc sv_setref_pvn
10453 Copies a string into a new SV, optionally blessing the SV. The length of the
10454 string must be specified with C<n>. The C<rv> argument will be upgraded to
10455 an RV. That RV will be modified to point to the new SV. The C<classname>
10456 argument indicates the package for the blessing. Set C<classname> to
10457 C<NULL> to avoid the blessing. The new SV will have a reference count
10458 of 1, and the RV will be returned.
10460 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10466 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10467 const char *const pv, const STRLEN n)
10469 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10471 sv_setpvn(newSVrv(rv,classname), pv, n);
10476 =for apidoc sv_bless
10478 Blesses an SV into a specified package. The SV must be an RV. The package
10479 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10480 of the SV is unaffected.
10486 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10489 HV *oldstash = NULL;
10491 PERL_ARGS_ASSERT_SV_BLESS;
10495 Perl_croak(aTHX_ "Can't bless non-reference value");
10497 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10498 if (SvREADONLY(tmpRef))
10499 Perl_croak_no_modify();
10500 if (SvOBJECT(tmpRef)) {
10501 oldstash = SvSTASH(tmpRef);
10504 SvOBJECT_on(tmpRef);
10505 SvUPGRADE(tmpRef, SVt_PVMG);
10506 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10507 SvREFCNT_dec(oldstash);
10509 if(SvSMAGICAL(tmpRef))
10510 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10518 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10519 * as it is after unglobbing it.
10522 PERL_STATIC_INLINE void
10523 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10527 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10529 PERL_ARGS_ASSERT_SV_UNGLOB;
10531 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10533 if (!(flags & SV_COW_DROP_PV))
10534 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10536 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10538 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10539 && HvNAME_get(stash))
10540 mro_method_changed_in(stash);
10541 gp_free(MUTABLE_GV(sv));
10544 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10545 GvSTASH(sv) = NULL;
10548 if (GvNAME_HEK(sv)) {
10549 unshare_hek(GvNAME_HEK(sv));
10551 isGV_with_GP_off(sv);
10553 if(SvTYPE(sv) == SVt_PVGV) {
10554 /* need to keep SvANY(sv) in the right arena */
10555 xpvmg = new_XPVMG();
10556 StructCopy(SvANY(sv), xpvmg, XPVMG);
10557 del_XPVGV(SvANY(sv));
10560 SvFLAGS(sv) &= ~SVTYPEMASK;
10561 SvFLAGS(sv) |= SVt_PVMG;
10564 /* Intentionally not calling any local SET magic, as this isn't so much a
10565 set operation as merely an internal storage change. */
10566 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10567 else sv_setsv_flags(sv, temp, 0);
10569 if ((const GV *)sv == PL_last_in_gv)
10570 PL_last_in_gv = NULL;
10571 else if ((const GV *)sv == PL_statgv)
10576 =for apidoc sv_unref_flags
10578 Unsets the RV status of the SV, and decrements the reference count of
10579 whatever was being referenced by the RV. This can almost be thought of
10580 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10581 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10582 (otherwise the decrementing is conditional on the reference count being
10583 different from one or the reference being a readonly SV).
10584 See C<L</SvROK_off>>.
10590 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10592 SV* const target = SvRV(ref);
10594 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10596 if (SvWEAKREF(ref)) {
10597 sv_del_backref(target, ref);
10598 SvWEAKREF_off(ref);
10599 SvRV_set(ref, NULL);
10602 SvRV_set(ref, NULL);
10604 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10605 assigned to as BEGIN {$a = \"Foo"} will fail. */
10606 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10607 SvREFCNT_dec_NN(target);
10608 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10609 sv_2mortal(target); /* Schedule for freeing later */
10613 =for apidoc sv_untaint
10615 Untaint an SV. Use C<SvTAINTED_off> instead.
10621 Perl_sv_untaint(pTHX_ SV *const sv)
10623 PERL_ARGS_ASSERT_SV_UNTAINT;
10624 PERL_UNUSED_CONTEXT;
10626 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10627 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10634 =for apidoc sv_tainted
10636 Test an SV for taintedness. Use C<SvTAINTED> instead.
10642 Perl_sv_tainted(pTHX_ SV *const sv)
10644 PERL_ARGS_ASSERT_SV_TAINTED;
10645 PERL_UNUSED_CONTEXT;
10647 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10648 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10649 if (mg && (mg->mg_len & 1) )
10655 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10656 private to this file */
10659 =for apidoc sv_setpviv
10661 Copies an integer into the given SV, also updating its string value.
10662 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10668 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10670 char buf[TYPE_CHARS(UV)];
10672 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10674 PERL_ARGS_ASSERT_SV_SETPVIV;
10676 sv_setpvn(sv, ptr, ebuf - ptr);
10680 =for apidoc sv_setpviv_mg
10682 Like C<sv_setpviv>, but also handles 'set' magic.
10688 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10690 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10692 sv_setpviv(sv, iv);
10696 #endif /* NO_MATHOMS */
10698 #if defined(PERL_IMPLICIT_CONTEXT)
10700 /* pTHX_ magic can't cope with varargs, so this is a no-context
10701 * version of the main function, (which may itself be aliased to us).
10702 * Don't access this version directly.
10706 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10711 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10713 va_start(args, pat);
10714 sv_vsetpvf(sv, pat, &args);
10718 /* pTHX_ magic can't cope with varargs, so this is a no-context
10719 * version of the main function, (which may itself be aliased to us).
10720 * Don't access this version directly.
10724 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10729 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10731 va_start(args, pat);
10732 sv_vsetpvf_mg(sv, pat, &args);
10738 =for apidoc sv_setpvf
10740 Works like C<sv_catpvf> but copies the text into the SV instead of
10741 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10747 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10751 PERL_ARGS_ASSERT_SV_SETPVF;
10753 va_start(args, pat);
10754 sv_vsetpvf(sv, pat, &args);
10759 =for apidoc sv_vsetpvf
10761 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10762 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10764 Usually used via its frontend C<sv_setpvf>.
10770 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10772 PERL_ARGS_ASSERT_SV_VSETPVF;
10774 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10778 =for apidoc sv_setpvf_mg
10780 Like C<sv_setpvf>, but also handles 'set' magic.
10786 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10790 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10792 va_start(args, pat);
10793 sv_vsetpvf_mg(sv, pat, &args);
10798 =for apidoc sv_vsetpvf_mg
10800 Like C<sv_vsetpvf>, but also handles 'set' magic.
10802 Usually used via its frontend C<sv_setpvf_mg>.
10808 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10810 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10812 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10816 #if defined(PERL_IMPLICIT_CONTEXT)
10818 /* pTHX_ magic can't cope with varargs, so this is a no-context
10819 * version of the main function, (which may itself be aliased to us).
10820 * Don't access this version directly.
10824 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10829 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10831 va_start(args, pat);
10832 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10836 /* pTHX_ magic can't cope with varargs, so this is a no-context
10837 * version of the main function, (which may itself be aliased to us).
10838 * Don't access this version directly.
10842 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10847 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10849 va_start(args, pat);
10850 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10857 =for apidoc sv_catpvf
10859 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10860 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10861 variable argument list, argument reordering is not supported.
10862 If the appended data contains "wide" characters
10863 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10864 and characters >255 formatted with C<%c>), the original SV might get
10865 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10866 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10867 valid UTF-8; if the original SV was bytes, the pattern should be too.
10872 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10876 PERL_ARGS_ASSERT_SV_CATPVF;
10878 va_start(args, pat);
10879 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10884 =for apidoc sv_vcatpvf
10886 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10887 variable argument list, and appends the formatted output
10888 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10890 Usually used via its frontend C<sv_catpvf>.
10896 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10898 PERL_ARGS_ASSERT_SV_VCATPVF;
10900 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10904 =for apidoc sv_catpvf_mg
10906 Like C<sv_catpvf>, but also handles 'set' magic.
10912 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10916 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10918 va_start(args, pat);
10919 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10925 =for apidoc sv_vcatpvf_mg
10927 Like C<sv_vcatpvf>, but also handles 'set' magic.
10929 Usually used via its frontend C<sv_catpvf_mg>.
10935 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10937 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10939 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10944 =for apidoc sv_vsetpvfn
10946 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10949 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10955 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10956 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10958 PERL_ARGS_ASSERT_SV_VSETPVFN;
10961 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10965 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10967 PERL_STATIC_INLINE void
10968 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10970 STRLEN const need = len + SvCUR(sv) + 1;
10973 /* can't wrap as both len and SvCUR() are allocated in
10974 * memory and together can't consume all the address space
10976 assert(need > len);
10981 Copy(buf, end, len, char);
10984 SvCUR_set(sv, need - 1);
10989 * Warn of missing argument to sprintf. The value used in place of such
10990 * arguments should be &PL_sv_no; an undefined value would yield
10991 * inappropriate "use of uninit" warnings [perl #71000].
10994 S_warn_vcatpvfn_missing_argument(pTHX) {
10995 if (ckWARN(WARN_MISSING)) {
10996 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10997 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11006 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11007 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11011 /* Given an int i from the next arg (if args is true) or an sv from an arg
11012 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11013 * with overflow checking.
11014 * Sets *neg to true if the value was negative (untouched otherwise.
11015 * Returns the absolute value.
11016 * As an extra margin of safety, it croaks if the returned value would
11017 * exceed the maximum value of a STRLEN / 4.
11021 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11035 if (UNLIKELY(SvIsUV(sv))) {
11036 UV uv = SvUV_nomg(sv);
11038 S_croak_overflow();
11042 iv = SvIV_nomg(sv);
11046 S_croak_overflow();
11052 if (iv > (IV)(((STRLEN)~0) / 4))
11053 S_croak_overflow();
11059 /* Returns true if c is in the range '1'..'9'
11060 * Written with the cast so it only needs one conditional test
11062 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11064 /* Read in and return a number. Updates *pattern to point to the char
11065 * following the number. Expects the first char to 1..9.
11066 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11067 * This is a belt-and-braces safety measure to complement any
11068 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11069 * It means that e.g. on a 32-bit system the width/precision can't be more
11070 * than 1G, which seems reasonable.
11074 S_expect_number(pTHX_ const char **const pattern)
11078 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11080 assert(IS_1_TO_9(**pattern));
11082 var = *(*pattern)++ - '0';
11083 while (isDIGIT(**pattern)) {
11084 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11085 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11086 S_croak_overflow();
11087 var = var * 10 + (*(*pattern)++ - '0');
11092 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11093 * ensures it's big enough), back fill it with the rounded integer part of
11094 * nv. Returns ptr to start of string, and sets *len to its length.
11095 * Returns NULL if not convertible.
11099 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11101 const int neg = nv < 0;
11104 PERL_ARGS_ASSERT_F0CONVERT;
11106 assert(!Perl_isinfnan(nv));
11113 if (uv & 1 && uv == nv)
11114 uv--; /* Round to even */
11116 const unsigned dig = uv % 10;
11118 } while (uv /= 10);
11128 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11131 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11132 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11134 PERL_ARGS_ASSERT_SV_VCATPVFN;
11136 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11140 /* For the vcatpvfn code, we need a long double target in case
11141 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11142 * with long double formats, even without NV being long double. But we
11143 * call the target 'fv' instead of 'nv', since most of the time it is not
11144 * (most compilers these days recognize "long double", even if only as a
11145 * synonym for "double").
11147 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11148 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11149 # define VCATPVFN_FV_GF PERL_PRIgldbl
11150 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11151 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11152 # define VCATPVFN_NV_TO_FV(nv,fv) \
11155 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11158 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11160 typedef long double vcatpvfn_long_double_t;
11162 # define VCATPVFN_FV_GF NVgf
11163 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11164 typedef NV vcatpvfn_long_double_t;
11167 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11168 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11169 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11170 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11171 * after the first 1023 zero bits.
11173 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11174 * of dynamically growing buffer might be better, start at just 16 bytes
11175 * (for example) and grow only when necessary. Or maybe just by looking
11176 * at the exponents of the two doubles? */
11177 # define DOUBLEDOUBLE_MAXBITS 2098
11180 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11181 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11182 * per xdigit. For the double-double case, this can be rather many.
11183 * The non-double-double-long-double overshoots since all bits of NV
11184 * are not mantissa bits, there are also exponent bits. */
11185 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11186 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11188 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11191 /* If we do not have a known long double format, (including not using
11192 * long doubles, or long doubles being equal to doubles) then we will
11193 * fall back to the ldexp/frexp route, with which we can retrieve at
11194 * most as many bits as our widest unsigned integer type is. We try
11195 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11197 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11198 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11200 #if defined(HAS_QUAD) && defined(Uquad_t)
11201 # define MANTISSATYPE Uquad_t
11202 # define MANTISSASIZE 8
11204 # define MANTISSATYPE UV
11205 # define MANTISSASIZE UVSIZE
11208 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11209 # define HEXTRACT_LITTLE_ENDIAN
11210 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11211 # define HEXTRACT_BIG_ENDIAN
11213 # define HEXTRACT_MIX_ENDIAN
11216 /* S_hextract() is a helper for S_format_hexfp, for extracting
11217 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11218 * are being extracted from (either directly from the long double in-memory
11219 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11220 * is used to update the exponent. The subnormal is set to true
11221 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11222 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11224 * The tricky part is that S_hextract() needs to be called twice:
11225 * the first time with vend as NULL, and the second time with vend as
11226 * the pointer returned by the first call. What happens is that on
11227 * the first round the output size is computed, and the intended
11228 * extraction sanity checked. On the second round the actual output
11229 * (the extraction of the hexadecimal values) takes place.
11230 * Sanity failures cause fatal failures during both rounds. */
11232 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11233 U8* vhex, U8* vend)
11237 int ixmin = 0, ixmax = 0;
11239 /* XXX Inf/NaN are not handled here, since it is
11240 * assumed they are to be output as "Inf" and "NaN". */
11242 /* These macros are just to reduce typos, they have multiple
11243 * repetitions below, but usually only one (or sometimes two)
11244 * of them is really being used. */
11245 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11246 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11247 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11248 #define HEXTRACT_OUTPUT(ix) \
11250 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11252 #define HEXTRACT_COUNT(ix, c) \
11254 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11256 #define HEXTRACT_BYTE(ix) \
11258 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11260 #define HEXTRACT_LO_NYBBLE(ix) \
11262 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11264 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11265 * to make it look less odd when the top bits of a NV
11266 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11267 * order bits can be in the "low nybble" of a byte. */
11268 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11269 #define HEXTRACT_BYTES_LE(a, b) \
11270 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11271 #define HEXTRACT_BYTES_BE(a, b) \
11272 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11273 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11274 #define HEXTRACT_IMPLICIT_BIT(nv) \
11276 if (!*subnormal) { \
11277 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11281 /* Most formats do. Those which don't should undef this.
11283 * But also note that IEEE 754 subnormals do not have it, or,
11284 * expressed alternatively, their implicit bit is zero. */
11285 #define HEXTRACT_HAS_IMPLICIT_BIT
11287 /* Many formats do. Those which don't should undef this. */
11288 #define HEXTRACT_HAS_TOP_NYBBLE
11290 /* HEXTRACTSIZE is the maximum number of xdigits. */
11291 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11292 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11294 # define HEXTRACTSIZE 2 * NVSIZE
11297 const U8* vmaxend = vhex + HEXTRACTSIZE;
11299 assert(HEXTRACTSIZE <= VHEX_SIZE);
11301 PERL_UNUSED_VAR(ix); /* might happen */
11302 (void)Perl_frexp(PERL_ABS(nv), exponent);
11303 *subnormal = FALSE;
11304 if (vend && (vend <= vhex || vend > vmaxend)) {
11305 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11306 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11309 /* First check if using long doubles. */
11310 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11311 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11312 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11313 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11314 /* The bytes 13..0 are the mantissa/fraction,
11315 * the 15,14 are the sign+exponent. */
11316 const U8* nvp = (const U8*)(&nv);
11317 HEXTRACT_GET_SUBNORMAL(nv);
11318 HEXTRACT_IMPLICIT_BIT(nv);
11319 # undef HEXTRACT_HAS_TOP_NYBBLE
11320 HEXTRACT_BYTES_LE(13, 0);
11321 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11322 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11323 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11324 /* The bytes 2..15 are the mantissa/fraction,
11325 * the 0,1 are the sign+exponent. */
11326 const U8* nvp = (const U8*)(&nv);
11327 HEXTRACT_GET_SUBNORMAL(nv);
11328 HEXTRACT_IMPLICIT_BIT(nv);
11329 # undef HEXTRACT_HAS_TOP_NYBBLE
11330 HEXTRACT_BYTES_BE(2, 15);
11331 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11332 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11333 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11334 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11335 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11336 /* The bytes 0..1 are the sign+exponent,
11337 * the bytes 2..9 are the mantissa/fraction. */
11338 const U8* nvp = (const U8*)(&nv);
11339 # undef HEXTRACT_HAS_IMPLICIT_BIT
11340 # undef HEXTRACT_HAS_TOP_NYBBLE
11341 HEXTRACT_GET_SUBNORMAL(nv);
11342 HEXTRACT_BYTES_LE(7, 0);
11343 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11344 /* Does this format ever happen? (Wikipedia says the Motorola
11345 * 6888x math coprocessors used format _like_ this but padded
11346 * to 96 bits with 16 unused bits between the exponent and the
11348 const U8* nvp = (const U8*)(&nv);
11349 # undef HEXTRACT_HAS_IMPLICIT_BIT
11350 # undef HEXTRACT_HAS_TOP_NYBBLE
11351 HEXTRACT_GET_SUBNORMAL(nv);
11352 HEXTRACT_BYTES_BE(0, 7);
11354 # define HEXTRACT_FALLBACK
11355 /* Double-double format: two doubles next to each other.
11356 * The first double is the high-order one, exactly like
11357 * it would be for a "lone" double. The second double
11358 * is shifted down using the exponent so that that there
11359 * are no common bits. The tricky part is that the value
11360 * of the double-double is the SUM of the two doubles and
11361 * the second one can be also NEGATIVE.
11363 * Because of this tricky construction the bytewise extraction we
11364 * use for the other long double formats doesn't work, we must
11365 * extract the values bit by bit.
11367 * The little-endian double-double is used .. somewhere?
11369 * The big endian double-double is used in e.g. PPC/Power (AIX)
11372 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11373 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11374 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11377 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11378 /* Using normal doubles, not long doubles.
11380 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11381 * bytes, since we might need to handle printf precision, and
11382 * also need to insert the radix. */
11384 # ifdef HEXTRACT_LITTLE_ENDIAN
11385 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11386 const U8* nvp = (const U8*)(&nv);
11387 HEXTRACT_GET_SUBNORMAL(nv);
11388 HEXTRACT_IMPLICIT_BIT(nv);
11389 HEXTRACT_TOP_NYBBLE(6);
11390 HEXTRACT_BYTES_LE(5, 0);
11391 # elif defined(HEXTRACT_BIG_ENDIAN)
11392 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11393 const U8* nvp = (const U8*)(&nv);
11394 HEXTRACT_GET_SUBNORMAL(nv);
11395 HEXTRACT_IMPLICIT_BIT(nv);
11396 HEXTRACT_TOP_NYBBLE(1);
11397 HEXTRACT_BYTES_BE(2, 7);
11398 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11399 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11400 const U8* nvp = (const U8*)(&nv);
11401 HEXTRACT_GET_SUBNORMAL(nv);
11402 HEXTRACT_IMPLICIT_BIT(nv);
11403 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11404 HEXTRACT_BYTE(1); /* 5 */
11405 HEXTRACT_BYTE(0); /* 4 */
11406 HEXTRACT_BYTE(7); /* 3 */
11407 HEXTRACT_BYTE(6); /* 2 */
11408 HEXTRACT_BYTE(5); /* 1 */
11409 HEXTRACT_BYTE(4); /* 0 */
11410 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11411 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11412 const U8* nvp = (const U8*)(&nv);
11413 HEXTRACT_GET_SUBNORMAL(nv);
11414 HEXTRACT_IMPLICIT_BIT(nv);
11415 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11416 HEXTRACT_BYTE(6); /* 5 */
11417 HEXTRACT_BYTE(7); /* 4 */
11418 HEXTRACT_BYTE(0); /* 3 */
11419 HEXTRACT_BYTE(1); /* 2 */
11420 HEXTRACT_BYTE(2); /* 1 */
11421 HEXTRACT_BYTE(3); /* 0 */
11423 # define HEXTRACT_FALLBACK
11426 # define HEXTRACT_FALLBACK
11428 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11430 #ifdef HEXTRACT_FALLBACK
11431 HEXTRACT_GET_SUBNORMAL(nv);
11432 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11433 /* The fallback is used for the double-double format, and
11434 * for unknown long double formats, and for unknown double
11435 * formats, or in general unknown NV formats. */
11436 if (nv == (NV)0.0) {
11444 NV d = nv < 0 ? -nv : nv;
11446 U8 ha = 0x0; /* hexvalue accumulator */
11447 U8 hd = 0x8; /* hexvalue digit */
11449 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11450 * this is essentially manual frexp(). Multiplying by 0.5 and
11451 * doubling should be lossless in binary floating point. */
11461 while (d >= e + e) {
11465 /* Now e <= d < 2*e */
11467 /* First extract the leading hexdigit (the implicit bit). */
11483 /* Then extract the remaining hexdigits. */
11484 while (d > (NV)0.0) {
11490 /* Output or count in groups of four bits,
11491 * that is, when the hexdigit is down to one. */
11496 /* Reset the hexvalue. */
11505 /* Flush possible pending hexvalue. */
11515 /* Croak for various reasons: if the output pointer escaped the
11516 * output buffer, if the extraction index escaped the extraction
11517 * buffer, or if the ending output pointer didn't match the
11518 * previously computed value. */
11519 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11520 /* For double-double the ixmin and ixmax stay at zero,
11521 * which is convenient since the HEXTRACTSIZE is tricky
11522 * for double-double. */
11523 ixmin < 0 || ixmax >= NVSIZE ||
11524 (vend && v != vend)) {
11525 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11526 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11532 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11534 * Processes the %a/%A hexadecimal floating-point format, since the
11535 * built-in snprintf()s which are used for most of the f/p formats, don't
11536 * universally handle %a/%A.
11537 * Populates buf of length bufsize, and returns the length of the created
11539 * The rest of the args have the same meaning as the local vars of the
11540 * same name within Perl_sv_vcatpvfn_flags().
11542 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11544 * It requires the caller to make buf large enough.
11548 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11549 const NV nv, const vcatpvfn_long_double_t fv,
11550 bool has_precis, STRLEN precis, STRLEN width,
11551 bool alt, char plus, bool left, bool fill)
11553 /* Hexadecimal floating point. */
11555 U8 vhex[VHEX_SIZE];
11556 U8* v = vhex; /* working pointer to vhex */
11557 U8* vend; /* pointer to one beyond last digit of vhex */
11558 U8* vfnz = NULL; /* first non-zero */
11559 U8* vlnz = NULL; /* last non-zero */
11560 U8* v0 = NULL; /* first output */
11561 const bool lower = (c == 'a');
11562 /* At output the values of vhex (up to vend) will
11563 * be mapped through the xdig to get the actual
11564 * human-readable xdigits. */
11565 const char* xdig = PL_hexdigit;
11566 STRLEN zerotail = 0; /* how many extra zeros to append */
11567 int exponent = 0; /* exponent of the floating point input */
11568 bool hexradix = FALSE; /* should we output the radix */
11569 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11570 bool negative = FALSE;
11573 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11575 * For example with denormals, (assuming the vanilla
11576 * 64-bit double): the exponent is zero. 1xp-1074 is
11577 * the smallest denormal and the smallest double, it
11578 * could be output also as 0x0.0000000000001p-1022 to
11579 * match its internal structure. */
11581 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11582 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11584 #if NVSIZE > DOUBLESIZE
11585 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11586 /* In this case there is an implicit bit,
11587 * and therefore the exponent is shifted by one. */
11590 # ifdef NV_X86_80_BIT
11592 /* The subnormals of the x86-80 have a base exponent of -16382,
11593 * (while the physical exponent bits are zero) but the frexp()
11594 * returned the scientific-style floating exponent. We want
11595 * to map the last one as:
11596 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11597 * -16835..-16388 -> -16384
11598 * since we want to keep the first hexdigit
11599 * as one of the [8421]. */
11600 exponent = -4 * ( (exponent + 1) / -4) - 2;
11605 /* TBD: other non-implicit-bit platforms than the x86-80. */
11609 negative = fv < 0 || Perl_signbit(nv);
11620 xdig += 16; /* Use uppercase hex. */
11623 /* Find the first non-zero xdigit. */
11624 for (v = vhex; v < vend; v++) {
11632 /* Find the last non-zero xdigit. */
11633 for (v = vend - 1; v >= vhex; v--) {
11640 #if NVSIZE == DOUBLESIZE
11646 #ifndef NV_X86_80_BIT
11648 /* IEEE 754 subnormals (but not the x86 80-bit):
11649 * we want "normalize" the subnormal,
11650 * so we need to right shift the hex nybbles
11651 * so that the output of the subnormal starts
11652 * from the first true bit. (Another, equally
11653 * valid, policy would be to dump the subnormal
11654 * nybbles as-is, to display the "physical" layout.) */
11657 /* Find the ceil(log2(v[0])) of
11658 * the top non-zero nybble. */
11659 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11662 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11663 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11677 U8* ve = (subnormal ? vlnz + 1 : vend);
11678 SSize_t vn = ve - v0;
11680 if (precis < (Size_t)(vn - 1)) {
11681 bool overflow = FALSE;
11682 if (v0[precis + 1] < 0x8) {
11683 /* Round down, nothing to do. */
11684 } else if (v0[precis + 1] > 0x8) {
11687 overflow = v0[precis] > 0xF;
11689 } else { /* v0[precis] == 0x8 */
11690 /* Half-point: round towards the one
11691 * with the even least-significant digit:
11699 * 78 -> 8 f8 -> 10 */
11700 if ((v0[precis] & 0x1)) {
11703 overflow = v0[precis] > 0xF;
11708 for (v = v0 + precis - 1; v >= v0; v--) {
11710 overflow = *v > 0xF;
11716 if (v == v0 - 1 && overflow) {
11717 /* If the overflow goes all the
11718 * way to the front, we need to
11719 * insert 0x1 in front, and adjust
11721 Move(v0, v0 + 1, vn - 1, char);
11727 /* The new effective "last non zero". */
11728 vlnz = v0 + precis;
11732 subnormal ? precis - vn + 1 :
11733 precis - (vlnz - vhex);
11740 /* If there are non-zero xdigits, the radix
11741 * is output after the first one. */
11752 /* The radix is always output if precis, or if alt. */
11753 if (precis > 0 || alt) {
11758 #ifndef USE_LOCALE_NUMERIC
11761 if (PL_numeric_radix_sv) {
11763 const char* r = SvPV(PL_numeric_radix_sv, n);
11764 assert(IN_LC(LC_NUMERIC));
11765 Copy(r, p, n, char);
11779 if (zerotail > 0) {
11780 while (zerotail--) {
11787 /* sanity checks */
11788 if (elen >= bufsize || width >= bufsize)
11789 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11790 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11792 elen += my_snprintf(p, bufsize - elen,
11793 "%c%+d", lower ? 'p' : 'P',
11796 if (elen < width) {
11797 STRLEN gap = (STRLEN)(width - elen);
11799 /* Pad the back with spaces. */
11800 memset(buf + elen, ' ', gap);
11803 /* Insert the zeros after the "0x" and the
11804 * the potential sign, but before the digits,
11805 * otherwise we end up with "0000xH.HHH...",
11806 * when we want "0x000H.HHH..." */
11807 STRLEN nzero = gap;
11808 char* zerox = buf + 2;
11809 STRLEN nmove = elen - 2;
11810 if (negative || plus) {
11814 Move(zerox, zerox + nzero, nmove, char);
11815 memset(zerox, fill ? '0' : ' ', nzero);
11818 /* Move it to the right. */
11819 Move(buf, buf + gap,
11821 /* Pad the front with spaces. */
11822 memset(buf, ' ', gap);
11831 =for apidoc sv_vcatpvfn
11833 =for apidoc sv_vcatpvfn_flags
11835 Processes its arguments like C<vsprintf> and appends the formatted output
11836 to an SV. Uses an array of SVs if the C-style variable argument list is
11837 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11838 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11839 C<va_list> argument list with a format string that uses argument reordering
11840 will yield an exception.
11842 When running with taint checks enabled, indicates via
11843 C<maybe_tainted> if results are untrustworthy (often due to the use of
11846 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11848 It assumes that pat has the same utf8-ness as sv. It's the caller's
11849 responsibility to ensure that this is so.
11851 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11858 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11859 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11862 const char *fmtstart; /* character following the current '%' */
11863 const char *q; /* current position within format */
11864 const char *patend;
11867 static const char nullstr[] = "(null)";
11869 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11870 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11871 /* Times 4: a decimal digit takes more than 3 binary digits.
11872 * NV_DIG: mantissa takes than many decimal digits.
11873 * Plus 32: Playing safe. */
11874 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11875 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11876 #ifdef USE_LOCALE_NUMERIC
11877 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11878 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11881 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11882 PERL_UNUSED_ARG(maybe_tainted);
11884 if (flags & SV_GMAGIC)
11887 /* no matter what, this is a string now */
11888 (void)SvPV_force_nomg(sv, origlen);
11890 /* the code that scans for flags etc following a % relies on
11891 * a '\0' being present to avoid falling off the end. Ideally that
11892 * should be fixed */
11893 assert(pat[patlen] == '\0');
11896 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11897 * In each case, if there isn't the correct number of args, instead
11898 * fall through to the main code to handle the issuing of any
11902 if (patlen == 0 && (args || sv_count == 0))
11905 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11908 if (patlen == 2 && pat[1] == 's') {
11910 const char * const s = va_arg(*args, char*);
11911 sv_catpv_nomg(sv, s ? s : nullstr);
11914 /* we want get magic on the source but not the target.
11915 * sv_catsv can't do that, though */
11916 SvGETMAGIC(*svargs);
11917 sv_catsv_nomg(sv, *svargs);
11924 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11925 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11926 sv_catsv_nomg(sv, asv);
11930 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11931 /* special-case "%.0f" */
11932 else if ( patlen == 4
11933 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11935 const NV nv = SvNV(*svargs);
11936 if (LIKELY(!Perl_isinfnan(nv))) {
11940 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11941 sv_catpvn_nomg(sv, p, l);
11946 #endif /* !USE_LONG_DOUBLE */
11950 patend = (char*)pat + patlen;
11951 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11952 char intsize = 0; /* size qualifier in "%hi..." etc */
11953 bool alt = FALSE; /* has "%#..." */
11954 bool left = FALSE; /* has "%-..." */
11955 bool fill = FALSE; /* has "%0..." */
11956 char plus = 0; /* has "%+..." */
11957 STRLEN width = 0; /* value of "%NNN..." */
11958 bool has_precis = FALSE; /* has "%.NNN..." */
11959 STRLEN precis = 0; /* value of "%.NNN..." */
11960 int base = 0; /* base to print in, e.g. 8 for %o */
11961 UV uv = 0; /* the value to print of int-ish args */
11963 bool vectorize = FALSE; /* has "%v..." */
11964 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11965 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11966 STRLEN veclen = 0; /* SvCUR(vec arg) */
11967 const char *dotstr = NULL; /* separator string for %v */
11968 STRLEN dotstrlen; /* length of separator string for %v */
11970 Size_t efix = 0; /* explicit format parameter index */
11971 const Size_t osvix = svix; /* original index in case of bad fmt */
11973 bool is_utf8 = FALSE; /* is this item utf8? */
11974 bool arg_missing = FALSE; /* give "Missing argument" warning */
11975 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11976 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11977 STRLEN zeros = 0; /* how many '0' to prepend */
11979 const char *eptr = NULL; /* the address of the element string */
11980 STRLEN elen = 0; /* the length of the element string */
11982 char c; /* the actual format ('d', s' etc) */
11985 /* echo everything up to the next format specification */
11986 for (q = fmtstart; q < patend && *q != '%'; ++q)
11989 if (q > fmtstart) {
11990 if (has_utf8 && !pat_utf8) {
11991 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
11995 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
11997 for (p = fmtstart; p < q; p++)
11998 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12003 for (p = fmtstart; p < q; p++)
12004 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12006 SvCUR_set(sv, need - 1);
12009 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12014 fmtstart = q; /* fmtstart is char following the '%' */
12017 We allow format specification elements in this order:
12018 \d+\$ explicit format parameter index
12020 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12021 0 flag (as above): repeated to allow "v02"
12022 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12023 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12025 [%bcdefginopsuxDFOUX] format (mandatory)
12028 if (IS_1_TO_9(*q)) {
12029 width = expect_number(&q);
12032 Perl_croak_nocontext(
12033 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12035 efix = (Size_t)width;
12037 no_redundant_warning = TRUE;
12049 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12076 /* at this point we can expect one of:
12078 * 123 an explicit width
12079 * * width taken from next arg
12080 * *12$ width taken from 12th arg
12083 * But any width specification may be preceded by a v, in one of its
12088 * So an asterisk may be either a width specifier or a vector
12089 * separator arg specifier, and we don't know which initially
12094 STRLEN ix; /* explicit width/vector separator index */
12096 if (IS_1_TO_9(*q)) {
12097 ix = expect_number(&q);
12100 Perl_croak_nocontext(
12101 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12102 no_redundant_warning = TRUE;
12111 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12112 * with the default "." */
12117 vecsv = va_arg(*args, SV*);
12119 ix = ix ? ix - 1 : svix++;
12120 vecsv = ix < sv_count ? svargs[ix]
12121 : (arg_missing = TRUE, &PL_sv_no);
12123 dotstr = SvPV_const(vecsv, dotstrlen);
12124 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12125 bad with tied or overloaded values that return UTF8. */
12126 if (DO_UTF8(vecsv))
12128 else if (has_utf8) {
12129 vecsv = sv_mortalcopy(vecsv);
12130 sv_utf8_upgrade(vecsv);
12131 dotstr = SvPV_const(vecsv, dotstrlen);
12138 /* the asterisk specified a width */
12143 i = va_arg(*args, int);
12145 ix = ix ? ix - 1 : svix++;
12146 sv = (ix < sv_count) ? svargs[ix]
12147 : (arg_missing = TRUE, (SV*)NULL);
12149 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12152 else if (*q == 'v') {
12163 /* explicit width? */
12169 width = expect_number(&q);
12179 STRLEN ix; /* explicit precision index */
12181 if (IS_1_TO_9(*q)) {
12182 ix = expect_number(&q);
12185 Perl_croak_nocontext(
12186 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12187 no_redundant_warning = TRUE;
12200 i = va_arg(*args, int);
12202 ix = ix ? ix - 1 : svix++;
12203 sv = (ix < sv_count) ? svargs[ix]
12204 : (arg_missing = TRUE, (SV*)NULL);
12206 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12211 /* although it doesn't seem documented, this code has long
12213 * no digits following the '.' is treated like '.0'
12214 * the number may be preceded by any number of zeroes,
12215 * e.g. "%.0001f", which is the same as "%.1f"
12216 * so I've kept that behaviour. DAPM May 2017
12220 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12229 case 'I': /* Ix, I32x, and I64x */
12230 # ifdef USE_64_BIT_INT
12231 if (q[1] == '6' && q[2] == '4') {
12237 if (q[1] == '3' && q[2] == '2') {
12241 # ifdef USE_64_BIT_INT
12247 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12248 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12251 # ifdef USE_QUADMATH
12264 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12265 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12266 if (*q == 'l') { /* lld, llf */
12275 if (*++q == 'h') { /* hhd, hhu */
12294 c = *q++; /* c now holds the conversion type */
12296 /* '%' doesn't have an arg, so skip arg processing */
12305 if (vectorize && !strchr("BbDdiOouUXx", c))
12308 /* get next arg (individual branches do their own va_arg()
12309 * handling for the args case) */
12312 efix = efix ? efix - 1 : svix++;
12313 argsv = efix < sv_count ? svargs[efix]
12314 : (arg_missing = TRUE, &PL_sv_no);
12324 eptr = va_arg(*args, char*);
12326 elen = strlen(eptr);
12328 eptr = (char *)nullstr;
12329 elen = sizeof nullstr - 1;
12333 eptr = SvPV_const(argsv, elen);
12334 if (DO_UTF8(argsv)) {
12335 STRLEN old_precis = precis;
12336 if (has_precis && precis < elen) {
12337 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12338 STRLEN p = precis > ulen ? ulen : precis;
12339 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12340 /* sticks at end */
12342 if (width) { /* fudge width (can't fudge elen) */
12343 if (has_precis && precis < elen)
12344 width += precis - old_precis;
12347 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12354 if (has_precis && precis < elen)
12366 * "%...p" is normally treated like "%...x", except that the
12367 * number to print is the SV's address (or a pointer address
12368 * for C-ish sprintf).
12370 * However, the C-ish sprintf variant allows a few special
12371 * extensions. These are currently:
12373 * %-p (SVf) Like %s, but gets the string from an SV*
12374 * arg rather than a char* arg.
12375 * (This was previously %_).
12377 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12379 * %2p (HEKf) Like %s, but using the key string in a HEK
12381 * %3p (HEKf256) Ditto but like %.256s
12383 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12384 * (cBOOL(utf8), len, string_buf).
12385 * It's handled by the "case 'd'" branch
12386 * rather than here.
12388 * %<num>p where num is 1 or > 4: reserved for future
12389 * extensions. Warns, but then is treated as a
12390 * general %p (print hex address) format.
12398 /* not %*p or %*1$p - any width was explicit */
12402 if (left) { /* %-p (SVf), %-NNNp */
12407 argsv = MUTABLE_SV(va_arg(*args, void*));
12408 eptr = SvPV_const(argsv, elen);
12409 if (DO_UTF8(argsv))
12414 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12415 HEK * const hek = va_arg(*args, HEK *);
12416 eptr = HEK_KEY(hek);
12417 elen = HEK_LEN(hek);
12428 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12429 "internal %%<num>p might conflict with future printf extensions");
12433 /* treat as normal %...p */
12435 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12440 /* Ignore any size specifiers, since they're not documented as
12441 * being allowed for %c (ideally we should warn on e.g. '%hc').
12442 * Setting a default intsize, along with a positive
12443 * (which signals unsigned) base, causes, for C-ish use, the
12444 * va_arg to be interpreted as as unsigned int, when it's
12445 * actually signed, which will convert -ve values to high +ve
12446 * values. Note that unlike the libc %c, values > 255 will
12447 * convert to high unicode points rather than being truncated
12448 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12449 * will again convert -ve args to high -ve values.
12452 base = 1; /* special value that indicates we're doing a 'c' */
12453 goto get_int_arg_val;
12462 goto get_int_arg_val;
12465 /* probably just a plain %d, but it might be the start of the
12466 * special UTF8f format, which usually looks something like
12467 * "%d%lu%4p" (the lu may vary by platform)
12469 assert((UTF8f)[0] == 'd');
12470 assert((UTF8f)[1] == '%');
12472 if ( args /* UTF8f only valid for C-ish sprintf */
12473 && q == fmtstart + 1 /* plain %d, not %....d */
12474 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12476 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12478 /* The argument has already gone through cBOOL, so the cast
12480 is_utf8 = (bool)va_arg(*args, int);
12481 elen = va_arg(*args, UV);
12482 /* if utf8 length is larger than 0x7ffff..., then it might
12483 * have been a signed value that wrapped */
12484 if (elen > ((~(STRLEN)0) >> 1)) {
12485 assert(0); /* in DEBUGGING build we want to crash */
12486 elen = 0; /* otherwise we want to treat this as an empty string */
12488 eptr = va_arg(*args, char *);
12489 q += sizeof(UTF8f) - 2;
12496 goto get_int_arg_val;
12507 goto get_int_arg_val;
12512 goto get_int_arg_val;
12523 goto get_int_arg_val;
12538 esignbuf[esignlen++] = plus;
12541 /* initialise the vector string to iterate over */
12543 vecsv = args ? va_arg(*args, SV*) : argsv;
12545 /* if this is a version object, we need to convert
12546 * back into v-string notation and then let the
12547 * vectorize happen normally
12549 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12550 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12551 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12552 "vector argument not supported with alpha versions");
12556 vecstr = (U8*)SvPV_const(vecsv,veclen);
12557 vecsv = sv_newmortal();
12558 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12562 vecstr = (U8*)SvPV_const(vecsv, veclen);
12563 vec_utf8 = DO_UTF8(vecsv);
12565 /* This is the re-entry point for when we're iterating
12566 * over the individual characters of a vector arg */
12569 goto done_valid_conversion;
12571 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12581 /* test arg for inf/nan. This can trigger an unwanted
12582 * 'str' overload, so manually force 'num' overload first
12586 if (UNLIKELY(SvAMAGIC(argsv)))
12587 argsv = sv_2num(argsv);
12588 if (UNLIKELY(isinfnansv(argsv)))
12589 goto handle_infnan_argsv;
12593 /* signed int type */
12598 case 'c': iv = (char)va_arg(*args, int); break;
12599 case 'h': iv = (short)va_arg(*args, int); break;
12600 case 'l': iv = va_arg(*args, long); break;
12601 case 'V': iv = va_arg(*args, IV); break;
12602 case 'z': iv = va_arg(*args, SSize_t); break;
12603 #ifdef HAS_PTRDIFF_T
12604 case 't': iv = va_arg(*args, ptrdiff_t); break;
12606 default: iv = va_arg(*args, int); break;
12608 case 'j': iv = va_arg(*args, intmax_t); break;
12612 iv = va_arg(*args, Quad_t); break;
12619 /* assign to tiv then cast to iv to work around
12620 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12621 IV tiv = SvIV_nomg(argsv);
12623 case 'c': iv = (char)tiv; break;
12624 case 'h': iv = (short)tiv; break;
12625 case 'l': iv = (long)tiv; break;
12627 default: iv = tiv; break;
12630 iv = (Quad_t)tiv; break;
12637 /* now convert iv to uv */
12641 esignbuf[esignlen++] = plus;
12644 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12645 esignbuf[esignlen++] = '-';
12649 /* unsigned int type */
12652 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12654 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12656 case 'l': uv = va_arg(*args, unsigned long); break;
12657 case 'V': uv = va_arg(*args, UV); break;
12658 case 'z': uv = va_arg(*args, Size_t); break;
12659 #ifdef HAS_PTRDIFF_T
12660 /* will sign extend, but there is no
12661 * uptrdiff_t, so oh well */
12662 case 't': uv = va_arg(*args, ptrdiff_t); break;
12665 case 'j': uv = va_arg(*args, uintmax_t); break;
12667 default: uv = va_arg(*args, unsigned); break;
12670 uv = va_arg(*args, Uquad_t); break;
12677 /* assign to tiv then cast to iv to work around
12678 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12679 UV tuv = SvUV_nomg(argsv);
12681 case 'c': uv = (unsigned char)tuv; break;
12682 case 'h': uv = (unsigned short)tuv; break;
12683 case 'l': uv = (unsigned long)tuv; break;
12685 default: uv = tuv; break;
12688 uv = (Uquad_t)tuv; break;
12699 char *ptr = ebuf + sizeof ebuf;
12706 const char * const p =
12707 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12712 } while (uv >>= 4);
12713 if (alt && *ptr != '0') {
12714 esignbuf[esignlen++] = '0';
12715 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12722 *--ptr = '0' + dig;
12723 } while (uv >>= 3);
12724 if (alt && *ptr != '0')
12730 *--ptr = '0' + dig;
12731 } while (uv >>= 1);
12732 if (alt && *ptr != '0') {
12733 esignbuf[esignlen++] = '0';
12734 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12739 /* special-case: base 1 indicates a 'c' format:
12740 * we use the common code for extracting a uv,
12741 * but handle that value differently here than
12742 * all the other int types */
12744 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12747 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12749 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12754 ebuf[0] = (char)uv;
12759 default: /* it had better be ten or less */
12762 *--ptr = '0' + dig;
12763 } while (uv /= base);
12766 elen = (ebuf + sizeof ebuf) - ptr;
12770 zeros = precis - elen;
12771 else if (precis == 0 && elen == 1 && *eptr == '0'
12772 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12775 /* a precision nullifies the 0 flag. */
12781 /* FLOATING POINT */
12784 c = 'f'; /* maybe %F isn't supported here */
12786 case 'e': case 'E':
12788 case 'g': case 'G':
12789 case 'a': case 'A':
12792 STRLEN float_need; /* what PL_efloatsize needs to become */
12793 bool hexfp; /* hexadecimal floating point? */
12795 vcatpvfn_long_double_t fv;
12798 /* This is evil, but floating point is even more evil */
12800 /* for SV-style calling, we can only get NV
12801 for C-style calling, we assume %f is double;
12802 for simplicity we allow any of %Lf, %llf, %qf for long double
12806 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12810 /* [perl #20339] - we should accept and ignore %lf rather than die */
12814 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12815 intsize = args ? 0 : 'q';
12819 #if defined(HAS_LONG_DOUBLE)
12832 /* Now we need (long double) if intsize == 'q', else (double). */
12834 /* Note: do not pull NVs off the va_list with va_arg()
12835 * (pull doubles instead) because if you have a build
12836 * with long doubles, you would always be pulling long
12837 * doubles, which would badly break anyone using only
12838 * doubles (i.e. the majority of builds). In other
12839 * words, you cannot mix doubles and long doubles.
12840 * The only case where you can pull off long doubles
12841 * is when the format specifier explicitly asks so with
12843 #ifdef USE_QUADMATH
12844 fv = intsize == 'q' ?
12845 va_arg(*args, NV) : va_arg(*args, double);
12847 #elif LONG_DOUBLESIZE > DOUBLESIZE
12848 if (intsize == 'q') {
12849 fv = va_arg(*args, long double);
12852 nv = va_arg(*args, double);
12853 VCATPVFN_NV_TO_FV(nv, fv);
12856 nv = va_arg(*args, double);
12863 /* we jump here if an int-ish format encountered an
12864 * infinite/Nan argsv. After setting nv/fv, it falls
12865 * into the isinfnan block which follows */
12866 handle_infnan_argsv:
12867 nv = SvNV_nomg(argsv);
12868 VCATPVFN_NV_TO_FV(nv, fv);
12871 if (Perl_isinfnan(nv)) {
12873 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12874 SvNV_nomg(argsv), (int)c);
12876 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12885 /* special-case "%.0f" */
12889 && !(width || left || plus || alt)
12892 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12896 /* Determine the buffer size needed for the various
12897 * floating-point formats.
12899 * The basic possibilities are:
12902 * %f 1111111.123456789
12903 * %e 1.111111123e+06
12904 * %a 0x1.0f4471f9bp+20
12906 * %g 1.11111112e+15
12908 * where P is the value of the precision in the format, or 6
12909 * if not specified. Note the two possible output formats of
12910 * %g; in both cases the number of significant digits is <=
12913 * For most of the format types the maximum buffer size needed
12914 * is precision, plus: any leading 1 or 0x1, the radix
12915 * point, and an exponent. The difficult one is %f: for a
12916 * large positive exponent it can have many leading digits,
12917 * which needs to be calculated specially. Also %a is slightly
12918 * different in that in the absence of a specified precision,
12919 * it uses as many digits as necessary to distinguish
12920 * different values.
12922 * First, here are the constant bits. For ease of calculation
12923 * we over-estimate the needed buffer size, for example by
12924 * assuming all formats have an exponent and a leading 0x1.
12926 * Also for production use, add a little extra overhead for
12927 * safety's sake. Under debugging don't, as it means we're
12928 * more likely to quickly spot issues during development.
12931 float_need = 1 /* possible unary minus */
12932 + 4 /* "0x1" plus very unlikely carry */
12933 + 1 /* default radix point '.' */
12934 + 2 /* "e-", "p+" etc */
12935 + 6 /* exponent: up to 16383 (quad fp) */
12937 + 20 /* safety net */
12942 /* determine the radix point len, e.g. length(".") in "1.2" */
12943 #ifdef USE_LOCALE_NUMERIC
12944 /* note that we may either explicitly use PL_numeric_radix_sv
12945 * below, or implicitly, via an snprintf() variant.
12946 * Note also things like ps_AF.utf8 which has
12947 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12948 if (!lc_numeric_set) {
12949 /* only set once and reuse in-locale value on subsequent
12951 * XXX what happens if we die in an eval?
12953 STORE_LC_NUMERIC_SET_TO_NEEDED();
12954 lc_numeric_set = TRUE;
12957 if (PL_numeric_radix_sv) {
12958 assert(IN_LC(LC_NUMERIC));
12959 /* this can't wrap unless PL_numeric_radix_sv is a string
12960 * consuming virtually all the 32-bit or 64-bit address
12963 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12965 /* floating-point formats only get utf8 if the radix point
12966 * is utf8. All other characters in the string are < 128
12967 * and so can be safely appended to both a non-utf8 and utf8
12969 * Note that this will convert the output to utf8 even if
12970 * the radix point didn't get output.
12972 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12973 sv_utf8_upgrade(sv);
12981 if (isALPHA_FOLD_EQ(c, 'f')) {
12982 /* Determine how many digits before the radix point
12983 * might be emitted. frexp() (or frexpl) has some
12984 * unspecified behaviour for nan/inf/-inf, so lucky we've
12985 * already handled them above */
12987 int i = PERL_INT_MIN;
12988 (void)Perl_frexp((NV)fv, &i);
12989 if (i == PERL_INT_MIN)
12990 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
12993 digits = BIT_DIGITS(i);
12994 /* this can't overflow. 'digits' will only be a few
12995 * thousand even for the largest floating-point types.
12996 * And up until now float_need is just some small
12997 * constants plus radix len, which can't be in
12998 * overflow territory unless the radix SV is consuming
12999 * over 1/2 the address space */
13000 assert(float_need < ((STRLEN)~0) - digits);
13001 float_need += digits;
13004 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13007 /* %a in the absence of precision may print as many
13008 * digits as needed to represent the entire mantissa
13010 * This estimate seriously overshoots in most cases,
13011 * but better the undershooting. Firstly, all bytes
13012 * of the NV are not mantissa, some of them are
13013 * exponent. Secondly, for the reasonably common
13014 * long doubles case, the "80-bit extended", two
13015 * or six bytes of the NV are unused. Also, we'll
13016 * still pick up an extra +6 from the default
13017 * precision calculation below. */
13019 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13020 /* For the "double double", we need more.
13021 * Since each double has their own exponent, the
13022 * doubles may float (haha) rather far from each
13023 * other, and the number of required bits is much
13024 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13025 * See the definition of DOUBLEDOUBLE_MAXBITS.
13027 * Need 2 hexdigits for each byte. */
13028 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13030 NVSIZE * 2; /* 2 hexdigits for each byte */
13032 /* see "this can't overflow" comment above */
13033 assert(float_need < ((STRLEN)~0) - digits);
13034 float_need += digits;
13037 /* special-case "%.<number>g" if it will fit in ebuf */
13039 && precis /* See earlier comment about buggy Gconvert
13040 when digits, aka precis, is 0 */
13042 /* check, in manner not involving wrapping, that it will
13044 && float_need < sizeof(ebuf)
13045 && sizeof(ebuf) - float_need > precis
13046 && !(width || left || plus || alt)
13050 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13051 elen = strlen(ebuf);
13058 STRLEN pr = has_precis ? precis : 6; /* known default */
13059 /* this probably can't wrap, since precis is limited
13060 * to 1/4 address space size, but better safe than sorry
13062 if (float_need >= ((STRLEN)~0) - pr)
13063 croak_memory_wrap();
13067 if (float_need < width)
13068 float_need = width;
13070 if (PL_efloatsize <= float_need) {
13071 /* PL_efloatbuf should be at least 1 greater than
13072 * float_need to allow a trailing \0 to be returned by
13073 * snprintf(). If we need to grow, overgrow for the
13074 * benefit of future generations */
13075 const STRLEN extra = 0x20;
13076 if (float_need >= ((STRLEN)~0) - extra)
13077 croak_memory_wrap();
13078 float_need += extra;
13079 Safefree(PL_efloatbuf);
13080 PL_efloatsize = float_need;
13081 Newx(PL_efloatbuf, PL_efloatsize, char);
13082 PL_efloatbuf[0] = '\0';
13085 if (UNLIKELY(hexfp)) {
13086 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13087 nv, fv, has_precis, precis, width,
13088 alt, plus, left, fill);
13091 char *ptr = ebuf + sizeof ebuf;
13094 #if defined(USE_QUADMATH)
13095 if (intsize == 'q') {
13099 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13100 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13101 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13102 * not USE_LONG_DOUBLE and NVff. In other words,
13103 * this needs to work without USE_LONG_DOUBLE. */
13104 if (intsize == 'q') {
13105 /* Copy the one or more characters in a long double
13106 * format before the 'base' ([efgEFG]) character to
13107 * the format string. */
13108 static char const ldblf[] = PERL_PRIfldbl;
13109 char const *p = ldblf + sizeof(ldblf) - 3;
13110 while (p >= ldblf) { *--ptr = *p--; }
13115 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13120 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13132 /* No taint. Otherwise we are in the strange situation
13133 * where printf() taints but print($float) doesn't.
13136 /* hopefully the above makes ptr a very constrained format
13137 * that is safe to use, even though it's not literal */
13138 GCC_DIAG_IGNORE(-Wformat-nonliteral);
13139 #ifdef USE_QUADMATH
13141 const char* qfmt = quadmath_format_single(ptr);
13143 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13144 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13146 if ((IV)elen == -1) {
13149 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13154 #elif defined(HAS_LONG_DOUBLE)
13155 elen = ((intsize == 'q')
13156 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13157 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13159 elen = my_sprintf(PL_efloatbuf, ptr, fv);
13164 eptr = PL_efloatbuf;
13168 /* Since floating-point formats do their own formatting and
13169 * padding, we skip the main block of code at the end of this
13170 * loop which handles appending eptr to sv, and do our own
13171 * stripped-down version */
13176 assert(elen >= width);
13178 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13180 goto done_valid_conversion;
13188 /* XXX ideally we should warn if any flags etc have been
13189 * set, e.g. "%-4.5n" */
13190 /* XXX if sv was originally non-utf8 with a char in the
13191 * range 0x80-0xff, then if it got upgraded, we should
13192 * calculate char len rather than byte len here */
13193 len = SvCUR(sv) - origlen;
13195 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13198 case 'c': *(va_arg(*args, char*)) = i; break;
13199 case 'h': *(va_arg(*args, short*)) = i; break;
13200 default: *(va_arg(*args, int*)) = i; break;
13201 case 'l': *(va_arg(*args, long*)) = i; break;
13202 case 'V': *(va_arg(*args, IV*)) = i; break;
13203 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13204 #ifdef HAS_PTRDIFF_T
13205 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13208 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
13212 *(va_arg(*args, Quad_t*)) = i; break;
13220 Perl_croak_nocontext(
13221 "Missing argument for %%n in %s",
13222 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13223 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13225 goto done_valid_conversion;
13233 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13234 && ckWARN(WARN_PRINTF))
13236 SV * const msg = sv_newmortal();
13237 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13238 (PL_op->op_type == OP_PRTF) ? "" : "s");
13239 if (fmtstart < patend) {
13240 const char * const fmtend = q < patend ? q : patend;
13242 sv_catpvs(msg, "\"%");
13243 for (f = fmtstart; f < fmtend; f++) {
13245 sv_catpvn_nomg(msg, f, 1);
13247 Perl_sv_catpvf(aTHX_ msg,
13248 "\\%03" UVof, (UV)*f & 0xFF);
13251 sv_catpvs(msg, "\"");
13253 sv_catpvs(msg, "end of string");
13255 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13258 /* mangled format: output the '%', then continue from the
13259 * character following that */
13260 sv_catpvn_nomg(sv, fmtstart-1, 1);
13263 /* Any "redundant arg" warning from now onwards will probably
13264 * just be misleading, so don't bother. */
13265 no_redundant_warning = TRUE;
13266 continue; /* not "break" */
13269 if (is_utf8 != has_utf8) {
13272 sv_utf8_upgrade(sv);
13275 const STRLEN old_elen = elen;
13276 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13277 sv_utf8_upgrade(nsv);
13278 eptr = SvPVX_const(nsv);
13281 if (width) { /* fudge width (can't fudge elen) */
13282 width += elen - old_elen;
13289 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13292 STRLEN need, have, gap;
13296 /* signed value that's wrapped? */
13297 assert(elen <= ((~(STRLEN)0) >> 1));
13299 /* if zeros is non-zero, then it represents filler between
13300 * elen and precis. So adding elen and zeros together will
13301 * always be <= precis, and the addition can never wrap */
13302 assert(!zeros || (precis > elen && precis - elen == zeros));
13303 have = elen + zeros;
13305 if (have >= (((STRLEN)~0) - esignlen))
13306 croak_memory_wrap();
13309 need = (have > width ? have : width);
13312 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13313 croak_memory_wrap();
13314 need += (SvCUR(sv) + 1);
13321 for (i = 0; i < esignlen; i++)
13322 *s++ = esignbuf[i];
13323 for (i = zeros; i; i--)
13325 Copy(eptr, s, elen, char);
13327 for (i = gap; i; i--)
13332 for (i = 0; i < esignlen; i++)
13333 *s++ = esignbuf[i];
13338 for (i = gap; i; i--)
13340 for (i = 0; i < esignlen; i++)
13341 *s++ = esignbuf[i];
13344 for (i = zeros; i; i--)
13346 Copy(eptr, s, elen, char);
13351 SvCUR_set(sv, s - SvPVX_const(sv));
13359 if (vectorize && veclen) {
13360 /* we append the vector separator separately since %v isn't
13361 * very common: don't slow down the general case by adding
13362 * dotstrlen to need etc */
13363 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13365 goto vector; /* do next iteration */
13368 done_valid_conversion:
13371 S_warn_vcatpvfn_missing_argument(aTHX);
13374 /* Now that we've consumed all our printf format arguments (svix)
13375 * do we have things left on the stack that we didn't use?
13377 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13378 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13379 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13384 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13388 /* =========================================================================
13390 =head1 Cloning an interpreter
13394 All the macros and functions in this section are for the private use of
13395 the main function, perl_clone().
13397 The foo_dup() functions make an exact copy of an existing foo thingy.
13398 During the course of a cloning, a hash table is used to map old addresses
13399 to new addresses. The table is created and manipulated with the
13400 ptr_table_* functions.
13402 * =========================================================================*/
13405 #if defined(USE_ITHREADS)
13407 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13408 #ifndef GpREFCNT_inc
13409 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13413 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13414 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13415 If this changes, please unmerge ss_dup.
13416 Likewise, sv_dup_inc_multiple() relies on this fact. */
13417 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13418 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13419 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13420 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13421 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13422 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13423 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13424 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13425 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13426 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13427 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13428 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13429 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13431 /* clone a parser */
13434 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13438 PERL_ARGS_ASSERT_PARSER_DUP;
13443 /* look for it in the table first */
13444 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13448 /* create anew and remember what it is */
13449 Newxz(parser, 1, yy_parser);
13450 ptr_table_store(PL_ptr_table, proto, parser);
13452 /* XXX these not yet duped */
13453 parser->old_parser = NULL;
13454 parser->stack = NULL;
13456 parser->stack_max1 = 0;
13457 /* XXX parser->stack->state = 0; */
13459 /* XXX eventually, just Copy() most of the parser struct ? */
13461 parser->lex_brackets = proto->lex_brackets;
13462 parser->lex_casemods = proto->lex_casemods;
13463 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13464 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13465 parser->lex_casestack = savepvn(proto->lex_casestack,
13466 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13467 parser->lex_defer = proto->lex_defer;
13468 parser->lex_dojoin = proto->lex_dojoin;
13469 parser->lex_formbrack = proto->lex_formbrack;
13470 parser->lex_inpat = proto->lex_inpat;
13471 parser->lex_inwhat = proto->lex_inwhat;
13472 parser->lex_op = proto->lex_op;
13473 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13474 parser->lex_starts = proto->lex_starts;
13475 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13476 parser->multi_close = proto->multi_close;
13477 parser->multi_open = proto->multi_open;
13478 parser->multi_start = proto->multi_start;
13479 parser->multi_end = proto->multi_end;
13480 parser->preambled = proto->preambled;
13481 parser->lex_super_state = proto->lex_super_state;
13482 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13483 parser->lex_sub_op = proto->lex_sub_op;
13484 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13485 parser->linestr = sv_dup_inc(proto->linestr, param);
13486 parser->expect = proto->expect;
13487 parser->copline = proto->copline;
13488 parser->last_lop_op = proto->last_lop_op;
13489 parser->lex_state = proto->lex_state;
13490 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13491 /* rsfp_filters entries have fake IoDIRP() */
13492 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13493 parser->in_my = proto->in_my;
13494 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13495 parser->error_count = proto->error_count;
13496 parser->sig_elems = proto->sig_elems;
13497 parser->sig_optelems= proto->sig_optelems;
13498 parser->sig_slurpy = proto->sig_slurpy;
13499 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13500 parser->linestr = sv_dup_inc(proto->linestr, param);
13503 char * const ols = SvPVX(proto->linestr);
13504 char * const ls = SvPVX(parser->linestr);
13506 parser->bufptr = ls + (proto->bufptr >= ols ?
13507 proto->bufptr - ols : 0);
13508 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13509 proto->oldbufptr - ols : 0);
13510 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13511 proto->oldoldbufptr - ols : 0);
13512 parser->linestart = ls + (proto->linestart >= ols ?
13513 proto->linestart - ols : 0);
13514 parser->last_uni = ls + (proto->last_uni >= ols ?
13515 proto->last_uni - ols : 0);
13516 parser->last_lop = ls + (proto->last_lop >= ols ?
13517 proto->last_lop - ols : 0);
13519 parser->bufend = ls + SvCUR(parser->linestr);
13522 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13525 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13526 Copy(proto->nexttype, parser->nexttype, 5, I32);
13527 parser->nexttoke = proto->nexttoke;
13529 /* XXX should clone saved_curcop here, but we aren't passed
13530 * proto_perl; so do it in perl_clone_using instead */
13536 /* duplicate a file handle */
13539 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13543 PERL_ARGS_ASSERT_FP_DUP;
13544 PERL_UNUSED_ARG(type);
13547 return (PerlIO*)NULL;
13549 /* look for it in the table first */
13550 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13554 /* create anew and remember what it is */
13555 #ifdef __amigaos4__
13556 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13558 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13560 ptr_table_store(PL_ptr_table, fp, ret);
13564 /* duplicate a directory handle */
13567 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13571 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13573 const Direntry_t *dirent;
13574 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13580 PERL_UNUSED_CONTEXT;
13581 PERL_ARGS_ASSERT_DIRP_DUP;
13586 /* look for it in the table first */
13587 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13591 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13593 PERL_UNUSED_ARG(param);
13597 /* open the current directory (so we can switch back) */
13598 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13600 /* chdir to our dir handle and open the present working directory */
13601 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13602 PerlDir_close(pwd);
13603 return (DIR *)NULL;
13605 /* Now we should have two dir handles pointing to the same dir. */
13607 /* Be nice to the calling code and chdir back to where we were. */
13608 /* XXX If this fails, then what? */
13609 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13611 /* We have no need of the pwd handle any more. */
13612 PerlDir_close(pwd);
13615 # define d_namlen(d) (d)->d_namlen
13617 # define d_namlen(d) strlen((d)->d_name)
13619 /* Iterate once through dp, to get the file name at the current posi-
13620 tion. Then step back. */
13621 pos = PerlDir_tell(dp);
13622 if ((dirent = PerlDir_read(dp))) {
13623 len = d_namlen(dirent);
13624 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13625 /* If the len is somehow magically longer than the
13626 * maximum length of the directory entry, even though
13627 * we could fit it in a buffer, we could not copy it
13628 * from the dirent. Bail out. */
13629 PerlDir_close(ret);
13632 if (len <= sizeof smallbuf) name = smallbuf;
13633 else Newx(name, len, char);
13634 Move(dirent->d_name, name, len, char);
13636 PerlDir_seek(dp, pos);
13638 /* Iterate through the new dir handle, till we find a file with the
13640 if (!dirent) /* just before the end */
13642 pos = PerlDir_tell(ret);
13643 if (PerlDir_read(ret)) continue; /* not there yet */
13644 PerlDir_seek(ret, pos); /* step back */
13648 const long pos0 = PerlDir_tell(ret);
13650 pos = PerlDir_tell(ret);
13651 if ((dirent = PerlDir_read(ret))) {
13652 if (len == (STRLEN)d_namlen(dirent)
13653 && memEQ(name, dirent->d_name, len)) {
13655 PerlDir_seek(ret, pos); /* step back */
13658 /* else we are not there yet; keep iterating */
13660 else { /* This is not meant to happen. The best we can do is
13661 reset the iterator to the beginning. */
13662 PerlDir_seek(ret, pos0);
13669 if (name && name != smallbuf)
13674 ret = win32_dirp_dup(dp, param);
13677 /* pop it in the pointer table */
13679 ptr_table_store(PL_ptr_table, dp, ret);
13684 /* duplicate a typeglob */
13687 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13691 PERL_ARGS_ASSERT_GP_DUP;
13695 /* look for it in the table first */
13696 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13700 /* create anew and remember what it is */
13702 ptr_table_store(PL_ptr_table, gp, ret);
13705 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13706 on Newxz() to do this for us. */
13707 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13708 ret->gp_io = io_dup_inc(gp->gp_io, param);
13709 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13710 ret->gp_av = av_dup_inc(gp->gp_av, param);
13711 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13712 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13713 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13714 ret->gp_cvgen = gp->gp_cvgen;
13715 ret->gp_line = gp->gp_line;
13716 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13720 /* duplicate a chain of magic */
13723 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13725 MAGIC *mgret = NULL;
13726 MAGIC **mgprev_p = &mgret;
13728 PERL_ARGS_ASSERT_MG_DUP;
13730 for (; mg; mg = mg->mg_moremagic) {
13733 if ((param->flags & CLONEf_JOIN_IN)
13734 && mg->mg_type == PERL_MAGIC_backref)
13735 /* when joining, we let the individual SVs add themselves to
13736 * backref as needed. */
13739 Newx(nmg, 1, MAGIC);
13741 mgprev_p = &(nmg->mg_moremagic);
13743 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13744 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13745 from the original commit adding Perl_mg_dup() - revision 4538.
13746 Similarly there is the annotation "XXX random ptr?" next to the
13747 assignment to nmg->mg_ptr. */
13750 /* FIXME for plugins
13751 if (nmg->mg_type == PERL_MAGIC_qr) {
13752 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13756 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13757 ? nmg->mg_type == PERL_MAGIC_backref
13758 /* The backref AV has its reference
13759 * count deliberately bumped by 1 */
13760 ? SvREFCNT_inc(av_dup_inc((const AV *)
13761 nmg->mg_obj, param))
13762 : sv_dup_inc(nmg->mg_obj, param)
13763 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13764 nmg->mg_type == PERL_MAGIC_regdata)
13766 : sv_dup(nmg->mg_obj, param);
13768 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13769 if (nmg->mg_len > 0) {
13770 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13771 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13772 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13774 AMT * const namtp = (AMT*)nmg->mg_ptr;
13775 sv_dup_inc_multiple((SV**)(namtp->table),
13776 (SV**)(namtp->table), NofAMmeth, param);
13779 else if (nmg->mg_len == HEf_SVKEY)
13780 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13782 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13783 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13789 #endif /* USE_ITHREADS */
13791 struct ptr_tbl_arena {
13792 struct ptr_tbl_arena *next;
13793 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13796 /* create a new pointer-mapping table */
13799 Perl_ptr_table_new(pTHX)
13802 PERL_UNUSED_CONTEXT;
13804 Newx(tbl, 1, PTR_TBL_t);
13805 tbl->tbl_max = 511;
13806 tbl->tbl_items = 0;
13807 tbl->tbl_arena = NULL;
13808 tbl->tbl_arena_next = NULL;
13809 tbl->tbl_arena_end = NULL;
13810 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13814 #define PTR_TABLE_HASH(ptr) \
13815 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13817 /* map an existing pointer using a table */
13819 STATIC PTR_TBL_ENT_t *
13820 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13822 PTR_TBL_ENT_t *tblent;
13823 const UV hash = PTR_TABLE_HASH(sv);
13825 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13827 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13828 for (; tblent; tblent = tblent->next) {
13829 if (tblent->oldval == sv)
13836 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13838 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13840 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13841 PERL_UNUSED_CONTEXT;
13843 return tblent ? tblent->newval : NULL;
13846 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13847 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13848 * the core's typical use of ptr_tables in thread cloning. */
13851 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13853 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13855 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13856 PERL_UNUSED_CONTEXT;
13859 tblent->newval = newsv;
13861 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13863 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13864 struct ptr_tbl_arena *new_arena;
13866 Newx(new_arena, 1, struct ptr_tbl_arena);
13867 new_arena->next = tbl->tbl_arena;
13868 tbl->tbl_arena = new_arena;
13869 tbl->tbl_arena_next = new_arena->array;
13870 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13873 tblent = tbl->tbl_arena_next++;
13875 tblent->oldval = oldsv;
13876 tblent->newval = newsv;
13877 tblent->next = tbl->tbl_ary[entry];
13878 tbl->tbl_ary[entry] = tblent;
13880 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13881 ptr_table_split(tbl);
13885 /* double the hash bucket size of an existing ptr table */
13888 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13890 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13891 const UV oldsize = tbl->tbl_max + 1;
13892 UV newsize = oldsize * 2;
13895 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13896 PERL_UNUSED_CONTEXT;
13898 Renew(ary, newsize, PTR_TBL_ENT_t*);
13899 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13900 tbl->tbl_max = --newsize;
13901 tbl->tbl_ary = ary;
13902 for (i=0; i < oldsize; i++, ary++) {
13903 PTR_TBL_ENT_t **entp = ary;
13904 PTR_TBL_ENT_t *ent = *ary;
13905 PTR_TBL_ENT_t **curentp;
13908 curentp = ary + oldsize;
13910 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13912 ent->next = *curentp;
13922 /* remove all the entries from a ptr table */
13923 /* Deprecated - will be removed post 5.14 */
13926 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13928 PERL_UNUSED_CONTEXT;
13929 if (tbl && tbl->tbl_items) {
13930 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13932 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13935 struct ptr_tbl_arena *next = arena->next;
13941 tbl->tbl_items = 0;
13942 tbl->tbl_arena = NULL;
13943 tbl->tbl_arena_next = NULL;
13944 tbl->tbl_arena_end = NULL;
13948 /* clear and free a ptr table */
13951 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13953 struct ptr_tbl_arena *arena;
13955 PERL_UNUSED_CONTEXT;
13961 arena = tbl->tbl_arena;
13964 struct ptr_tbl_arena *next = arena->next;
13970 Safefree(tbl->tbl_ary);
13974 #if defined(USE_ITHREADS)
13977 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13979 PERL_ARGS_ASSERT_RVPV_DUP;
13981 assert(!isREGEXP(sstr));
13983 if (SvWEAKREF(sstr)) {
13984 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13985 if (param->flags & CLONEf_JOIN_IN) {
13986 /* if joining, we add any back references individually rather
13987 * than copying the whole backref array */
13988 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13992 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13994 else if (SvPVX_const(sstr)) {
13995 /* Has something there */
13997 /* Normal PV - clone whole allocated space */
13998 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13999 /* sstr may not be that normal, but actually copy on write.
14000 But we are a true, independent SV, so: */
14004 /* Special case - not normally malloced for some reason */
14005 if (isGV_with_GP(sstr)) {
14006 /* Don't need to do anything here. */
14008 else if ((SvIsCOW(sstr))) {
14009 /* A "shared" PV - clone it as "shared" PV */
14011 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14015 /* Some other special case - random pointer */
14016 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14021 /* Copy the NULL */
14022 SvPV_set(dstr, NULL);
14026 /* duplicate a list of SVs. source and dest may point to the same memory. */
14028 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14029 SSize_t items, CLONE_PARAMS *const param)
14031 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14033 while (items-- > 0) {
14034 *dest++ = sv_dup_inc(*source++, param);
14040 /* duplicate an SV of any type (including AV, HV etc) */
14043 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14048 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14050 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14051 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14056 /* look for it in the table first */
14057 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14061 if(param->flags & CLONEf_JOIN_IN) {
14062 /** We are joining here so we don't want do clone
14063 something that is bad **/
14064 if (SvTYPE(sstr) == SVt_PVHV) {
14065 const HEK * const hvname = HvNAME_HEK(sstr);
14067 /** don't clone stashes if they already exist **/
14068 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14069 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14070 ptr_table_store(PL_ptr_table, sstr, dstr);
14074 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14075 HV *stash = GvSTASH(sstr);
14076 const HEK * hvname;
14077 if (stash && (hvname = HvNAME_HEK(stash))) {
14078 /** don't clone GVs if they already exist **/
14080 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14081 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14083 stash, GvNAME(sstr),
14089 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14090 ptr_table_store(PL_ptr_table, sstr, *svp);
14097 /* create anew and remember what it is */
14100 #ifdef DEBUG_LEAKING_SCALARS
14101 dstr->sv_debug_optype = sstr->sv_debug_optype;
14102 dstr->sv_debug_line = sstr->sv_debug_line;
14103 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14104 dstr->sv_debug_parent = (SV*)sstr;
14105 FREE_SV_DEBUG_FILE(dstr);
14106 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14109 ptr_table_store(PL_ptr_table, sstr, dstr);
14112 SvFLAGS(dstr) = SvFLAGS(sstr);
14113 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14114 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14117 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14118 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14119 (void*)PL_watch_pvx, SvPVX_const(sstr));
14122 /* don't clone objects whose class has asked us not to */
14124 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14130 switch (SvTYPE(sstr)) {
14132 SvANY(dstr) = NULL;
14135 SET_SVANY_FOR_BODYLESS_IV(dstr);
14137 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14139 SvIV_set(dstr, SvIVX(sstr));
14143 #if NVSIZE <= IVSIZE
14144 SET_SVANY_FOR_BODYLESS_NV(dstr);
14146 SvANY(dstr) = new_XNV();
14148 SvNV_set(dstr, SvNVX(sstr));
14152 /* These are all the types that need complex bodies allocating. */
14154 const svtype sv_type = SvTYPE(sstr);
14155 const struct body_details *const sv_type_details
14156 = bodies_by_type + sv_type;
14160 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14161 NOT_REACHED; /* NOTREACHED */
14177 assert(sv_type_details->body_size);
14178 if (sv_type_details->arena) {
14179 new_body_inline(new_body, sv_type);
14181 = (void*)((char*)new_body - sv_type_details->offset);
14183 new_body = new_NOARENA(sv_type_details);
14187 SvANY(dstr) = new_body;
14190 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14191 ((char*)SvANY(dstr)) + sv_type_details->offset,
14192 sv_type_details->copy, char);
14194 Copy(((char*)SvANY(sstr)),
14195 ((char*)SvANY(dstr)),
14196 sv_type_details->body_size + sv_type_details->offset, char);
14199 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14200 && !isGV_with_GP(dstr)
14202 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14203 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14205 /* The Copy above means that all the source (unduplicated) pointers
14206 are now in the destination. We can check the flags and the
14207 pointers in either, but it's possible that there's less cache
14208 missing by always going for the destination.
14209 FIXME - instrument and check that assumption */
14210 if (sv_type >= SVt_PVMG) {
14212 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14213 if (SvOBJECT(dstr) && SvSTASH(dstr))
14214 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14215 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14218 /* The cast silences a GCC warning about unhandled types. */
14219 switch ((int)sv_type) {
14230 /* FIXME for plugins */
14231 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
14232 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14235 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14236 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14237 LvTARG(dstr) = dstr;
14238 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14239 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14241 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14242 if (isREGEXP(sstr)) goto duprex;
14244 /* non-GP case already handled above */
14245 if(isGV_with_GP(sstr)) {
14246 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14247 /* Don't call sv_add_backref here as it's going to be
14248 created as part of the magic cloning of the symbol
14249 table--unless this is during a join and the stash
14250 is not actually being cloned. */
14251 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14252 at the point of this comment. */
14253 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14254 if (param->flags & CLONEf_JOIN_IN)
14255 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14256 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14257 (void)GpREFCNT_inc(GvGP(dstr));
14261 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14262 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14263 /* I have no idea why fake dirp (rsfps)
14264 should be treated differently but otherwise
14265 we end up with leaks -- sky*/
14266 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14267 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14268 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14270 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14271 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14272 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14273 if (IoDIRP(dstr)) {
14274 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14277 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14279 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14281 if (IoOFP(dstr) == IoIFP(sstr))
14282 IoOFP(dstr) = IoIFP(dstr);
14284 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14285 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14286 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14287 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14290 /* avoid cloning an empty array */
14291 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14292 SV **dst_ary, **src_ary;
14293 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14295 src_ary = AvARRAY((const AV *)sstr);
14296 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14297 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14298 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14299 AvALLOC((const AV *)dstr) = dst_ary;
14300 if (AvREAL((const AV *)sstr)) {
14301 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14305 while (items-- > 0)
14306 *dst_ary++ = sv_dup(*src_ary++, param);
14308 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14309 while (items-- > 0) {
14314 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14315 AvALLOC((const AV *)dstr) = (SV**)NULL;
14316 AvMAX( (const AV *)dstr) = -1;
14317 AvFILLp((const AV *)dstr) = -1;
14321 if (HvARRAY((const HV *)sstr)) {
14323 const bool sharekeys = !!HvSHAREKEYS(sstr);
14324 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14325 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14327 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14328 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14330 HvARRAY(dstr) = (HE**)darray;
14331 while (i <= sxhv->xhv_max) {
14332 const HE * const source = HvARRAY(sstr)[i];
14333 HvARRAY(dstr)[i] = source
14334 ? he_dup(source, sharekeys, param) : 0;
14338 const struct xpvhv_aux * const saux = HvAUX(sstr);
14339 struct xpvhv_aux * const daux = HvAUX(dstr);
14340 /* This flag isn't copied. */
14343 if (saux->xhv_name_count) {
14344 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14346 = saux->xhv_name_count < 0
14347 ? -saux->xhv_name_count
14348 : saux->xhv_name_count;
14349 HEK **shekp = sname + count;
14351 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14352 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14353 while (shekp-- > sname) {
14355 *dhekp = hek_dup(*shekp, param);
14359 daux->xhv_name_u.xhvnameu_name
14360 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14363 daux->xhv_name_count = saux->xhv_name_count;
14365 daux->xhv_aux_flags = saux->xhv_aux_flags;
14366 #ifdef PERL_HASH_RANDOMIZE_KEYS
14367 daux->xhv_rand = saux->xhv_rand;
14368 daux->xhv_last_rand = saux->xhv_last_rand;
14370 daux->xhv_riter = saux->xhv_riter;
14371 daux->xhv_eiter = saux->xhv_eiter
14372 ? he_dup(saux->xhv_eiter,
14373 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14374 /* backref array needs refcnt=2; see sv_add_backref */
14375 daux->xhv_backreferences =
14376 (param->flags & CLONEf_JOIN_IN)
14377 /* when joining, we let the individual GVs and
14378 * CVs add themselves to backref as
14379 * needed. This avoids pulling in stuff
14380 * that isn't required, and simplifies the
14381 * case where stashes aren't cloned back
14382 * if they already exist in the parent
14385 : saux->xhv_backreferences
14386 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14387 ? MUTABLE_AV(SvREFCNT_inc(
14388 sv_dup_inc((const SV *)
14389 saux->xhv_backreferences, param)))
14390 : MUTABLE_AV(sv_dup((const SV *)
14391 saux->xhv_backreferences, param))
14394 daux->xhv_mro_meta = saux->xhv_mro_meta
14395 ? mro_meta_dup(saux->xhv_mro_meta, param)
14398 /* Record stashes for possible cloning in Perl_clone(). */
14400 av_push(param->stashes, dstr);
14404 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14407 if (!(param->flags & CLONEf_COPY_STACKS)) {
14412 /* NOTE: not refcounted */
14413 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14414 hv_dup(CvSTASH(dstr), param);
14415 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14416 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14417 if (!CvISXSUB(dstr)) {
14419 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14421 CvSLABBED_off(dstr);
14422 } else if (CvCONST(dstr)) {
14423 CvXSUBANY(dstr).any_ptr =
14424 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14426 assert(!CvSLABBED(dstr));
14427 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14429 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14430 hek_dup(CvNAME_HEK((CV *)sstr), param);
14431 /* don't dup if copying back - CvGV isn't refcounted, so the
14432 * duped GV may never be freed. A bit of a hack! DAPM */
14434 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14436 ? gv_dup_inc(CvGV(sstr), param)
14437 : (param->flags & CLONEf_JOIN_IN)
14439 : gv_dup(CvGV(sstr), param);
14441 if (!CvISXSUB(sstr)) {
14442 PADLIST * padlist = CvPADLIST(sstr);
14444 padlist = padlist_dup(padlist, param);
14445 CvPADLIST_set(dstr, padlist);
14447 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14448 PoisonPADLIST(dstr);
14451 CvWEAKOUTSIDE(sstr)
14452 ? cv_dup( CvOUTSIDE(dstr), param)
14453 : cv_dup_inc(CvOUTSIDE(dstr), param);
14463 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14465 PERL_ARGS_ASSERT_SV_DUP_INC;
14466 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14470 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14472 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14473 PERL_ARGS_ASSERT_SV_DUP;
14475 /* Track every SV that (at least initially) had a reference count of 0.
14476 We need to do this by holding an actual reference to it in this array.
14477 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14478 (akin to the stashes hash, and the perl stack), we come unstuck if
14479 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14480 thread) is manipulated in a CLONE method, because CLONE runs before the
14481 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14482 (and fix things up by giving each a reference via the temps stack).
14483 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14484 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14485 before the walk of unreferenced happens and a reference to that is SV
14486 added to the temps stack. At which point we have the same SV considered
14487 to be in use, and free to be re-used. Not good.
14489 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14490 assert(param->unreferenced);
14491 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14497 /* duplicate a context */
14500 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14502 PERL_CONTEXT *ncxs;
14504 PERL_ARGS_ASSERT_CX_DUP;
14507 return (PERL_CONTEXT*)NULL;
14509 /* look for it in the table first */
14510 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14514 /* create anew and remember what it is */
14515 Newx(ncxs, max + 1, PERL_CONTEXT);
14516 ptr_table_store(PL_ptr_table, cxs, ncxs);
14517 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14520 PERL_CONTEXT * const ncx = &ncxs[ix];
14521 if (CxTYPE(ncx) == CXt_SUBST) {
14522 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14525 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14526 switch (CxTYPE(ncx)) {
14528 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14529 if(CxHASARGS(ncx)){
14530 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14532 ncx->blk_sub.savearray = NULL;
14534 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14535 ncx->blk_sub.prevcomppad);
14538 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14540 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14541 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14542 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14543 /* XXX what do do with cur_top_env ???? */
14545 case CXt_LOOP_LAZYSV:
14546 ncx->blk_loop.state_u.lazysv.end
14547 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14548 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14549 duplication code instead.
14550 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14551 actually being the same function, and (2) order
14552 equivalence of the two unions.
14553 We can assert the later [but only at run time :-(] */
14554 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14555 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14558 ncx->blk_loop.state_u.ary.ary
14559 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14561 case CXt_LOOP_LIST:
14562 case CXt_LOOP_LAZYIV:
14563 /* code common to all 'for' CXt_LOOP_* types */
14564 ncx->blk_loop.itersave =
14565 sv_dup_inc(ncx->blk_loop.itersave, param);
14566 if (CxPADLOOP(ncx)) {
14567 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14568 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14569 ncx->blk_loop.oldcomppad =
14570 (PAD*)ptr_table_fetch(PL_ptr_table,
14571 ncx->blk_loop.oldcomppad);
14572 ncx->blk_loop.itervar_u.svp =
14573 &CX_CURPAD_SV(ncx->blk_loop, off);
14576 /* this copies the GV if CXp_FOR_GV, or the SV for an
14577 * alias (for \$x (...)) - relies on gv_dup being the
14578 * same as sv_dup */
14579 ncx->blk_loop.itervar_u.gv
14580 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14584 case CXt_LOOP_PLAIN:
14587 ncx->blk_format.prevcomppad =
14588 (PAD*)ptr_table_fetch(PL_ptr_table,
14589 ncx->blk_format.prevcomppad);
14590 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14591 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14592 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14596 ncx->blk_givwhen.defsv_save =
14597 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14610 /* duplicate a stack info structure */
14613 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14617 PERL_ARGS_ASSERT_SI_DUP;
14620 return (PERL_SI*)NULL;
14622 /* look for it in the table first */
14623 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14627 /* create anew and remember what it is */
14628 Newxz(nsi, 1, PERL_SI);
14629 ptr_table_store(PL_ptr_table, si, nsi);
14631 nsi->si_stack = av_dup_inc(si->si_stack, param);
14632 nsi->si_cxix = si->si_cxix;
14633 nsi->si_cxmax = si->si_cxmax;
14634 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14635 nsi->si_type = si->si_type;
14636 nsi->si_prev = si_dup(si->si_prev, param);
14637 nsi->si_next = si_dup(si->si_next, param);
14638 nsi->si_markoff = si->si_markoff;
14643 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14644 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14645 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14646 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14647 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14648 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14649 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14650 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14651 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14652 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14653 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14654 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14655 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14656 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14657 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14658 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14661 #define pv_dup_inc(p) SAVEPV(p)
14662 #define pv_dup(p) SAVEPV(p)
14663 #define svp_dup_inc(p,pp) any_dup(p,pp)
14665 /* map any object to the new equivent - either something in the
14666 * ptr table, or something in the interpreter structure
14670 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14674 PERL_ARGS_ASSERT_ANY_DUP;
14677 return (void*)NULL;
14679 /* look for it in the table first */
14680 ret = ptr_table_fetch(PL_ptr_table, v);
14684 /* see if it is part of the interpreter structure */
14685 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14686 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14694 /* duplicate the save stack */
14697 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14700 ANY * const ss = proto_perl->Isavestack;
14701 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14702 I32 ix = proto_perl->Isavestack_ix;
14715 void (*dptr) (void*);
14716 void (*dxptr) (pTHX_ void*);
14718 PERL_ARGS_ASSERT_SS_DUP;
14720 Newxz(nss, max, ANY);
14723 const UV uv = POPUV(ss,ix);
14724 const U8 type = (U8)uv & SAVE_MASK;
14726 TOPUV(nss,ix) = uv;
14728 case SAVEt_CLEARSV:
14729 case SAVEt_CLEARPADRANGE:
14731 case SAVEt_HELEM: /* hash element */
14732 case SAVEt_SV: /* scalar reference */
14733 sv = (const SV *)POPPTR(ss,ix);
14734 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14736 case SAVEt_ITEM: /* normal string */
14737 case SAVEt_GVSV: /* scalar slot in GV */
14738 sv = (const SV *)POPPTR(ss,ix);
14739 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14740 if (type == SAVEt_SV)
14744 case SAVEt_MORTALIZESV:
14745 case SAVEt_READONLY_OFF:
14746 sv = (const SV *)POPPTR(ss,ix);
14747 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14749 case SAVEt_FREEPADNAME:
14750 ptr = POPPTR(ss,ix);
14751 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14752 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14754 case SAVEt_SHARED_PVREF: /* char* in shared space */
14755 c = (char*)POPPTR(ss,ix);
14756 TOPPTR(nss,ix) = savesharedpv(c);
14757 ptr = POPPTR(ss,ix);
14758 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14760 case SAVEt_GENERIC_SVREF: /* generic sv */
14761 case SAVEt_SVREF: /* scalar reference */
14762 sv = (const SV *)POPPTR(ss,ix);
14763 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14764 if (type == SAVEt_SVREF)
14765 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14766 ptr = POPPTR(ss,ix);
14767 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14769 case SAVEt_GVSLOT: /* any slot in GV */
14770 sv = (const SV *)POPPTR(ss,ix);
14771 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14772 ptr = POPPTR(ss,ix);
14773 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14774 sv = (const SV *)POPPTR(ss,ix);
14775 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14777 case SAVEt_HV: /* hash reference */
14778 case SAVEt_AV: /* array reference */
14779 sv = (const SV *) POPPTR(ss,ix);
14780 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14782 case SAVEt_COMPPAD:
14784 sv = (const SV *) POPPTR(ss,ix);
14785 TOPPTR(nss,ix) = sv_dup(sv, param);
14787 case SAVEt_INT: /* int reference */
14788 ptr = POPPTR(ss,ix);
14789 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14790 intval = (int)POPINT(ss,ix);
14791 TOPINT(nss,ix) = intval;
14793 case SAVEt_LONG: /* long reference */
14794 ptr = POPPTR(ss,ix);
14795 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14796 longval = (long)POPLONG(ss,ix);
14797 TOPLONG(nss,ix) = longval;
14799 case SAVEt_I32: /* I32 reference */
14800 ptr = POPPTR(ss,ix);
14801 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14803 TOPINT(nss,ix) = i;
14805 case SAVEt_IV: /* IV reference */
14806 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14807 ptr = POPPTR(ss,ix);
14808 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14810 TOPIV(nss,ix) = iv;
14812 case SAVEt_TMPSFLOOR:
14814 TOPIV(nss,ix) = iv;
14816 case SAVEt_HPTR: /* HV* reference */
14817 case SAVEt_APTR: /* AV* reference */
14818 case SAVEt_SPTR: /* SV* reference */
14819 ptr = POPPTR(ss,ix);
14820 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14821 sv = (const SV *)POPPTR(ss,ix);
14822 TOPPTR(nss,ix) = sv_dup(sv, param);
14824 case SAVEt_VPTR: /* random* reference */
14825 ptr = POPPTR(ss,ix);
14826 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14828 case SAVEt_INT_SMALL:
14829 case SAVEt_I32_SMALL:
14830 case SAVEt_I16: /* I16 reference */
14831 case SAVEt_I8: /* I8 reference */
14833 ptr = POPPTR(ss,ix);
14834 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14836 case SAVEt_GENERIC_PVREF: /* generic char* */
14837 case SAVEt_PPTR: /* char* reference */
14838 ptr = POPPTR(ss,ix);
14839 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14840 c = (char*)POPPTR(ss,ix);
14841 TOPPTR(nss,ix) = pv_dup(c);
14843 case SAVEt_GP: /* scalar reference */
14844 gp = (GP*)POPPTR(ss,ix);
14845 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14846 (void)GpREFCNT_inc(gp);
14847 gv = (const GV *)POPPTR(ss,ix);
14848 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14851 ptr = POPPTR(ss,ix);
14852 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14853 /* these are assumed to be refcounted properly */
14855 switch (((OP*)ptr)->op_type) {
14857 case OP_LEAVESUBLV:
14861 case OP_LEAVEWRITE:
14862 TOPPTR(nss,ix) = ptr;
14865 (void) OpREFCNT_inc(o);
14869 TOPPTR(nss,ix) = NULL;
14874 TOPPTR(nss,ix) = NULL;
14876 case SAVEt_FREECOPHH:
14877 ptr = POPPTR(ss,ix);
14878 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14880 case SAVEt_ADELETE:
14881 av = (const AV *)POPPTR(ss,ix);
14882 TOPPTR(nss,ix) = av_dup_inc(av, param);
14884 TOPINT(nss,ix) = i;
14887 hv = (const HV *)POPPTR(ss,ix);
14888 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14890 TOPINT(nss,ix) = i;
14893 c = (char*)POPPTR(ss,ix);
14894 TOPPTR(nss,ix) = pv_dup_inc(c);
14896 case SAVEt_STACK_POS: /* Position on Perl stack */
14898 TOPINT(nss,ix) = i;
14900 case SAVEt_DESTRUCTOR:
14901 ptr = POPPTR(ss,ix);
14902 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14903 dptr = POPDPTR(ss,ix);
14904 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14905 any_dup(FPTR2DPTR(void *, dptr),
14908 case SAVEt_DESTRUCTOR_X:
14909 ptr = POPPTR(ss,ix);
14910 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14911 dxptr = POPDXPTR(ss,ix);
14912 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14913 any_dup(FPTR2DPTR(void *, dxptr),
14916 case SAVEt_REGCONTEXT:
14918 ix -= uv >> SAVE_TIGHT_SHIFT;
14920 case SAVEt_AELEM: /* array element */
14921 sv = (const SV *)POPPTR(ss,ix);
14922 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14924 TOPINT(nss,ix) = i;
14925 av = (const AV *)POPPTR(ss,ix);
14926 TOPPTR(nss,ix) = av_dup_inc(av, param);
14929 ptr = POPPTR(ss,ix);
14930 TOPPTR(nss,ix) = ptr;
14933 ptr = POPPTR(ss,ix);
14934 ptr = cophh_copy((COPHH*)ptr);
14935 TOPPTR(nss,ix) = ptr;
14937 TOPINT(nss,ix) = i;
14938 if (i & HINT_LOCALIZE_HH) {
14939 hv = (const HV *)POPPTR(ss,ix);
14940 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14943 case SAVEt_PADSV_AND_MORTALIZE:
14944 longval = (long)POPLONG(ss,ix);
14945 TOPLONG(nss,ix) = longval;
14946 ptr = POPPTR(ss,ix);
14947 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14948 sv = (const SV *)POPPTR(ss,ix);
14949 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14951 case SAVEt_SET_SVFLAGS:
14953 TOPINT(nss,ix) = i;
14955 TOPINT(nss,ix) = i;
14956 sv = (const SV *)POPPTR(ss,ix);
14957 TOPPTR(nss,ix) = sv_dup(sv, param);
14959 case SAVEt_COMPILE_WARNINGS:
14960 ptr = POPPTR(ss,ix);
14961 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14964 ptr = POPPTR(ss,ix);
14965 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14969 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14977 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14978 * flag to the result. This is done for each stash before cloning starts,
14979 * so we know which stashes want their objects cloned */
14982 do_mark_cloneable_stash(pTHX_ SV *const sv)
14984 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14986 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14987 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14988 if (cloner && GvCV(cloner)) {
14995 mXPUSHs(newSVhek(hvname));
14997 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15004 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15012 =for apidoc perl_clone
15014 Create and return a new interpreter by cloning the current one.
15016 C<perl_clone> takes these flags as parameters:
15018 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15019 without it we only clone the data and zero the stacks,
15020 with it we copy the stacks and the new perl interpreter is
15021 ready to run at the exact same point as the previous one.
15022 The pseudo-fork code uses C<COPY_STACKS> while the
15023 threads->create doesn't.
15025 C<CLONEf_KEEP_PTR_TABLE> -
15026 C<perl_clone> keeps a ptr_table with the pointer of the old
15027 variable as a key and the new variable as a value,
15028 this allows it to check if something has been cloned and not
15029 clone it again but rather just use the value and increase the
15030 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15031 the ptr_table using the function
15032 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15033 reason to keep it around is if you want to dup some of your own
15034 variable who are outside the graph perl scans, an example of this
15035 code is in F<threads.xs> create.
15037 C<CLONEf_CLONE_HOST> -
15038 This is a win32 thing, it is ignored on unix, it tells perls
15039 win32host code (which is c++) to clone itself, this is needed on
15040 win32 if you want to run two threads at the same time,
15041 if you just want to do some stuff in a separate perl interpreter
15042 and then throw it away and return to the original one,
15043 you don't need to do anything.
15048 /* XXX the above needs expanding by someone who actually understands it ! */
15049 EXTERN_C PerlInterpreter *
15050 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15053 perl_clone(PerlInterpreter *proto_perl, UV flags)
15056 #ifdef PERL_IMPLICIT_SYS
15058 PERL_ARGS_ASSERT_PERL_CLONE;
15060 /* perlhost.h so we need to call into it
15061 to clone the host, CPerlHost should have a c interface, sky */
15063 #ifndef __amigaos4__
15064 if (flags & CLONEf_CLONE_HOST) {
15065 return perl_clone_host(proto_perl,flags);
15068 return perl_clone_using(proto_perl, flags,
15070 proto_perl->IMemShared,
15071 proto_perl->IMemParse,
15073 proto_perl->IStdIO,
15077 proto_perl->IProc);
15081 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15082 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15083 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15084 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15085 struct IPerlDir* ipD, struct IPerlSock* ipS,
15086 struct IPerlProc* ipP)
15088 /* XXX many of the string copies here can be optimized if they're
15089 * constants; they need to be allocated as common memory and just
15090 * their pointers copied. */
15093 CLONE_PARAMS clone_params;
15094 CLONE_PARAMS* const param = &clone_params;
15096 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15098 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15099 #else /* !PERL_IMPLICIT_SYS */
15101 CLONE_PARAMS clone_params;
15102 CLONE_PARAMS* param = &clone_params;
15103 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15105 PERL_ARGS_ASSERT_PERL_CLONE;
15106 #endif /* PERL_IMPLICIT_SYS */
15108 /* for each stash, determine whether its objects should be cloned */
15109 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15110 PERL_SET_THX(my_perl);
15113 PoisonNew(my_perl, 1, PerlInterpreter);
15116 PL_defstash = NULL; /* may be used by perl malloc() */
15119 PL_scopestack_name = 0;
15121 PL_savestack_ix = 0;
15122 PL_savestack_max = -1;
15123 PL_sig_pending = 0;
15125 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15126 Zero(&PL_padname_undef, 1, PADNAME);
15127 Zero(&PL_padname_const, 1, PADNAME);
15128 # ifdef DEBUG_LEAKING_SCALARS
15129 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15131 # ifdef PERL_TRACE_OPS
15132 Zero(PL_op_exec_cnt, OP_max+2, UV);
15134 #else /* !DEBUGGING */
15135 Zero(my_perl, 1, PerlInterpreter);
15136 #endif /* DEBUGGING */
15138 #ifdef PERL_IMPLICIT_SYS
15139 /* host pointers */
15141 PL_MemShared = ipMS;
15142 PL_MemParse = ipMP;
15149 #endif /* PERL_IMPLICIT_SYS */
15152 param->flags = flags;
15153 /* Nothing in the core code uses this, but we make it available to
15154 extensions (using mg_dup). */
15155 param->proto_perl = proto_perl;
15156 /* Likely nothing will use this, but it is initialised to be consistent
15157 with Perl_clone_params_new(). */
15158 param->new_perl = my_perl;
15159 param->unreferenced = NULL;
15162 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15164 PL_body_arenas = NULL;
15165 Zero(&PL_body_roots, 1, PL_body_roots);
15169 PL_sv_arenaroot = NULL;
15171 PL_debug = proto_perl->Idebug;
15173 /* dbargs array probably holds garbage */
15176 PL_compiling = proto_perl->Icompiling;
15178 /* pseudo environmental stuff */
15179 PL_origargc = proto_perl->Iorigargc;
15180 PL_origargv = proto_perl->Iorigargv;
15182 #ifndef NO_TAINT_SUPPORT
15183 /* Set tainting stuff before PerlIO_debug can possibly get called */
15184 PL_tainting = proto_perl->Itainting;
15185 PL_taint_warn = proto_perl->Itaint_warn;
15187 PL_tainting = FALSE;
15188 PL_taint_warn = FALSE;
15191 PL_minus_c = proto_perl->Iminus_c;
15193 PL_localpatches = proto_perl->Ilocalpatches;
15194 PL_splitstr = proto_perl->Isplitstr;
15195 PL_minus_n = proto_perl->Iminus_n;
15196 PL_minus_p = proto_perl->Iminus_p;
15197 PL_minus_l = proto_perl->Iminus_l;
15198 PL_minus_a = proto_perl->Iminus_a;
15199 PL_minus_E = proto_perl->Iminus_E;
15200 PL_minus_F = proto_perl->Iminus_F;
15201 PL_doswitches = proto_perl->Idoswitches;
15202 PL_dowarn = proto_perl->Idowarn;
15203 #ifdef PERL_SAWAMPERSAND
15204 PL_sawampersand = proto_perl->Isawampersand;
15206 PL_unsafe = proto_perl->Iunsafe;
15207 PL_perldb = proto_perl->Iperldb;
15208 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15209 PL_exit_flags = proto_perl->Iexit_flags;
15211 /* XXX time(&PL_basetime) when asked for? */
15212 PL_basetime = proto_perl->Ibasetime;
15214 PL_maxsysfd = proto_perl->Imaxsysfd;
15215 PL_statusvalue = proto_perl->Istatusvalue;
15217 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15219 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15222 /* RE engine related */
15223 PL_regmatch_slab = NULL;
15224 PL_reg_curpm = NULL;
15226 PL_sub_generation = proto_perl->Isub_generation;
15228 /* funky return mechanisms */
15229 PL_forkprocess = proto_perl->Iforkprocess;
15231 /* internal state */
15232 PL_main_start = proto_perl->Imain_start;
15233 PL_eval_root = proto_perl->Ieval_root;
15234 PL_eval_start = proto_perl->Ieval_start;
15236 PL_filemode = proto_perl->Ifilemode;
15237 PL_lastfd = proto_perl->Ilastfd;
15238 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15241 PL_gensym = proto_perl->Igensym;
15243 PL_laststatval = proto_perl->Ilaststatval;
15244 PL_laststype = proto_perl->Ilaststype;
15247 PL_profiledata = NULL;
15249 PL_generation = proto_perl->Igeneration;
15251 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15252 PL_in_clean_all = proto_perl->Iin_clean_all;
15254 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15255 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15256 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15257 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15258 PL_nomemok = proto_perl->Inomemok;
15259 PL_an = proto_perl->Ian;
15260 PL_evalseq = proto_perl->Ievalseq;
15261 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15262 PL_origalen = proto_perl->Iorigalen;
15264 PL_sighandlerp = proto_perl->Isighandlerp;
15266 PL_runops = proto_perl->Irunops;
15268 PL_subline = proto_perl->Isubline;
15270 PL_cv_has_eval = proto_perl->Icv_has_eval;
15273 PL_cryptseen = proto_perl->Icryptseen;
15276 #ifdef USE_LOCALE_COLLATE
15277 PL_collation_ix = proto_perl->Icollation_ix;
15278 PL_collation_standard = proto_perl->Icollation_standard;
15279 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15280 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15281 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15282 #endif /* USE_LOCALE_COLLATE */
15284 #ifdef USE_LOCALE_NUMERIC
15285 PL_numeric_standard = proto_perl->Inumeric_standard;
15286 PL_numeric_local = proto_perl->Inumeric_local;
15287 #endif /* !USE_LOCALE_NUMERIC */
15289 /* Did the locale setup indicate UTF-8? */
15290 PL_utf8locale = proto_perl->Iutf8locale;
15291 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15292 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15293 /* Unicode features (see perlrun/-C) */
15294 PL_unicode = proto_perl->Iunicode;
15296 /* Pre-5.8 signals control */
15297 PL_signals = proto_perl->Isignals;
15299 /* times() ticks per second */
15300 PL_clocktick = proto_perl->Iclocktick;
15302 /* Recursion stopper for PerlIO_find_layer */
15303 PL_in_load_module = proto_perl->Iin_load_module;
15305 /* sort() routine */
15306 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15308 /* Not really needed/useful since the reenrant_retint is "volatile",
15309 * but do it for consistency's sake. */
15310 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15312 /* Hooks to shared SVs and locks. */
15313 PL_sharehook = proto_perl->Isharehook;
15314 PL_lockhook = proto_perl->Ilockhook;
15315 PL_unlockhook = proto_perl->Iunlockhook;
15316 PL_threadhook = proto_perl->Ithreadhook;
15317 PL_destroyhook = proto_perl->Idestroyhook;
15318 PL_signalhook = proto_perl->Isignalhook;
15320 PL_globhook = proto_perl->Iglobhook;
15323 PL_last_swash_hv = NULL; /* reinits on demand */
15324 PL_last_swash_klen = 0;
15325 PL_last_swash_key[0]= '\0';
15326 PL_last_swash_tmps = (U8*)NULL;
15327 PL_last_swash_slen = 0;
15329 PL_srand_called = proto_perl->Isrand_called;
15330 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15332 if (flags & CLONEf_COPY_STACKS) {
15333 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15334 PL_tmps_ix = proto_perl->Itmps_ix;
15335 PL_tmps_max = proto_perl->Itmps_max;
15336 PL_tmps_floor = proto_perl->Itmps_floor;
15338 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15339 * NOTE: unlike the others! */
15340 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15341 PL_scopestack_max = proto_perl->Iscopestack_max;
15343 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15344 * NOTE: unlike the others! */
15345 PL_savestack_ix = proto_perl->Isavestack_ix;
15346 PL_savestack_max = proto_perl->Isavestack_max;
15349 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15350 PL_top_env = &PL_start_env;
15352 PL_op = proto_perl->Iop;
15355 PL_Xpv = (XPV*)NULL;
15356 my_perl->Ina = proto_perl->Ina;
15358 PL_statcache = proto_perl->Istatcache;
15360 #ifndef NO_TAINT_SUPPORT
15361 PL_tainted = proto_perl->Itainted;
15363 PL_tainted = FALSE;
15365 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15367 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15369 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15370 PL_restartop = proto_perl->Irestartop;
15371 PL_in_eval = proto_perl->Iin_eval;
15372 PL_delaymagic = proto_perl->Idelaymagic;
15373 PL_phase = proto_perl->Iphase;
15374 PL_localizing = proto_perl->Ilocalizing;
15376 PL_hv_fetch_ent_mh = NULL;
15377 PL_modcount = proto_perl->Imodcount;
15378 PL_lastgotoprobe = NULL;
15379 PL_dumpindent = proto_perl->Idumpindent;
15381 PL_efloatbuf = NULL; /* reinits on demand */
15382 PL_efloatsize = 0; /* reinits on demand */
15386 PL_colorset = 0; /* reinits PL_colors[] */
15387 /*PL_colors[6] = {0,0,0,0,0,0};*/
15389 /* Pluggable optimizer */
15390 PL_peepp = proto_perl->Ipeepp;
15391 PL_rpeepp = proto_perl->Irpeepp;
15392 /* op_free() hook */
15393 PL_opfreehook = proto_perl->Iopfreehook;
15395 #ifdef USE_REENTRANT_API
15396 /* XXX: things like -Dm will segfault here in perlio, but doing
15397 * PERL_SET_CONTEXT(proto_perl);
15398 * breaks too many other things
15400 Perl_reentrant_init(aTHX);
15403 /* create SV map for pointer relocation */
15404 PL_ptr_table = ptr_table_new();
15406 /* initialize these special pointers as early as possible */
15408 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15409 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15410 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15411 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15412 &PL_padname_const);
15414 /* create (a non-shared!) shared string table */
15415 PL_strtab = newHV();
15416 HvSHAREKEYS_off(PL_strtab);
15417 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15418 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15420 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15422 /* This PV will be free'd special way so must set it same way op.c does */
15423 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15424 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15426 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15427 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15428 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15429 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15431 param->stashes = newAV(); /* Setup array of objects to call clone on */
15432 /* This makes no difference to the implementation, as it always pushes
15433 and shifts pointers to other SVs without changing their reference
15434 count, with the array becoming empty before it is freed. However, it
15435 makes it conceptually clear what is going on, and will avoid some
15436 work inside av.c, filling slots between AvFILL() and AvMAX() with
15437 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15438 AvREAL_off(param->stashes);
15440 if (!(flags & CLONEf_COPY_STACKS)) {
15441 param->unreferenced = newAV();
15444 #ifdef PERLIO_LAYERS
15445 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15446 PerlIO_clone(aTHX_ proto_perl, param);
15449 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15450 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15451 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15452 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15453 PL_xsubfilename = proto_perl->Ixsubfilename;
15454 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15455 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15458 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15459 PL_inplace = SAVEPV(proto_perl->Iinplace);
15460 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15462 /* magical thingies */
15464 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15465 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15466 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15469 /* Clone the regex array */
15470 /* ORANGE FIXME for plugins, probably in the SV dup code.
15471 newSViv(PTR2IV(CALLREGDUPE(
15472 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15474 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15475 PL_regex_pad = AvARRAY(PL_regex_padav);
15477 PL_stashpadmax = proto_perl->Istashpadmax;
15478 PL_stashpadix = proto_perl->Istashpadix ;
15479 Newx(PL_stashpad, PL_stashpadmax, HV *);
15482 for (; o < PL_stashpadmax; ++o)
15483 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15486 /* shortcuts to various I/O objects */
15487 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15488 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15489 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15490 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15491 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15492 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15493 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15495 /* shortcuts to regexp stuff */
15496 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15498 /* shortcuts to misc objects */
15499 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15501 /* shortcuts to debugging objects */
15502 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15503 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15504 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15505 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15506 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15507 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15508 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15510 /* symbol tables */
15511 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15512 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15513 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15514 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15515 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15517 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15518 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15519 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15520 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15521 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15522 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15523 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15524 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15525 PL_savebegin = proto_perl->Isavebegin;
15527 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15529 /* subprocess state */
15530 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15532 if (proto_perl->Iop_mask)
15533 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15536 /* PL_asserting = proto_perl->Iasserting; */
15538 /* current interpreter roots */
15539 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15541 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15544 /* runtime control stuff */
15545 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15547 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15549 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15551 /* interpreter atexit processing */
15552 PL_exitlistlen = proto_perl->Iexitlistlen;
15553 if (PL_exitlistlen) {
15554 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15555 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15558 PL_exitlist = (PerlExitListEntry*)NULL;
15560 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15561 if (PL_my_cxt_size) {
15562 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15563 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15564 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15565 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15566 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15570 PL_my_cxt_list = (void**)NULL;
15571 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15572 PL_my_cxt_keys = (const char**)NULL;
15575 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15576 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15577 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15578 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15580 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15582 PAD_CLONE_VARS(proto_perl, param);
15584 #ifdef HAVE_INTERP_INTERN
15585 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15588 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15590 #ifdef PERL_USES_PL_PIDSTATUS
15591 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15593 PL_osname = SAVEPV(proto_perl->Iosname);
15594 PL_parser = parser_dup(proto_perl->Iparser, param);
15596 /* XXX this only works if the saved cop has already been cloned */
15597 if (proto_perl->Iparser) {
15598 PL_parser->saved_curcop = (COP*)any_dup(
15599 proto_perl->Iparser->saved_curcop,
15603 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15605 #ifdef USE_LOCALE_CTYPE
15606 /* Should we warn if uses locale? */
15607 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15610 #ifdef USE_LOCALE_COLLATE
15611 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15612 #endif /* USE_LOCALE_COLLATE */
15614 #ifdef USE_LOCALE_NUMERIC
15615 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15616 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15617 #endif /* !USE_LOCALE_NUMERIC */
15619 /* Unicode inversion lists */
15620 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15621 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15622 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15623 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15625 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15626 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15628 /* utf8 character class swashes */
15629 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15630 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15632 for (i = 0; i < POSIX_CC_COUNT; i++) {
15633 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15635 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15636 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15637 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15638 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15639 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15640 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15641 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15642 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15643 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15644 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15645 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15646 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15647 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15648 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15649 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15650 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15651 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15652 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15654 if (proto_perl->Ipsig_pend) {
15655 Newxz(PL_psig_pend, SIG_SIZE, int);
15658 PL_psig_pend = (int*)NULL;
15661 if (proto_perl->Ipsig_name) {
15662 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15663 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15665 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15668 PL_psig_ptr = (SV**)NULL;
15669 PL_psig_name = (SV**)NULL;
15672 if (flags & CLONEf_COPY_STACKS) {
15673 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15674 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15675 PL_tmps_ix+1, param);
15677 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15678 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15679 Newxz(PL_markstack, i, I32);
15680 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15681 - proto_perl->Imarkstack);
15682 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15683 - proto_perl->Imarkstack);
15684 Copy(proto_perl->Imarkstack, PL_markstack,
15685 PL_markstack_ptr - PL_markstack + 1, I32);
15687 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15688 * NOTE: unlike the others! */
15689 Newxz(PL_scopestack, PL_scopestack_max, I32);
15690 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15693 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15694 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15696 /* reset stack AV to correct length before its duped via
15697 * PL_curstackinfo */
15698 AvFILLp(proto_perl->Icurstack) =
15699 proto_perl->Istack_sp - proto_perl->Istack_base;
15701 /* NOTE: si_dup() looks at PL_markstack */
15702 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15704 /* PL_curstack = PL_curstackinfo->si_stack; */
15705 PL_curstack = av_dup(proto_perl->Icurstack, param);
15706 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15708 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15709 PL_stack_base = AvARRAY(PL_curstack);
15710 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15711 - proto_perl->Istack_base);
15712 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15714 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15715 PL_savestack = ss_dup(proto_perl, param);
15719 ENTER; /* perl_destruct() wants to LEAVE; */
15722 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15723 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15725 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15726 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15727 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15728 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15729 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15730 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15732 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15734 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15735 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15736 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15738 PL_stashcache = newHV();
15740 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15741 proto_perl->Iwatchaddr);
15742 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15743 if (PL_debug && PL_watchaddr) {
15744 PerlIO_printf(Perl_debug_log,
15745 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15746 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15747 PTR2UV(PL_watchok));
15750 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15751 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15752 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15754 /* Call the ->CLONE method, if it exists, for each of the stashes
15755 identified by sv_dup() above.
15757 while(av_tindex(param->stashes) != -1) {
15758 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15759 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15760 if (cloner && GvCV(cloner)) {
15765 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15767 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15773 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15774 ptr_table_free(PL_ptr_table);
15775 PL_ptr_table = NULL;
15778 if (!(flags & CLONEf_COPY_STACKS)) {
15779 unreferenced_to_tmp_stack(param->unreferenced);
15782 SvREFCNT_dec(param->stashes);
15784 /* orphaned? eg threads->new inside BEGIN or use */
15785 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15786 SvREFCNT_inc_simple_void(PL_compcv);
15787 SAVEFREESV(PL_compcv);
15794 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15796 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15798 if (AvFILLp(unreferenced) > -1) {
15799 SV **svp = AvARRAY(unreferenced);
15800 SV **const last = svp + AvFILLp(unreferenced);
15804 if (SvREFCNT(*svp) == 1)
15806 } while (++svp <= last);
15808 EXTEND_MORTAL(count);
15809 svp = AvARRAY(unreferenced);
15812 if (SvREFCNT(*svp) == 1) {
15813 /* Our reference is the only one to this SV. This means that
15814 in this thread, the scalar effectively has a 0 reference.
15815 That doesn't work (cleanup never happens), so donate our
15816 reference to it onto the save stack. */
15817 PL_tmps_stack[++PL_tmps_ix] = *svp;
15819 /* As an optimisation, because we are already walking the
15820 entire array, instead of above doing either
15821 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15822 release our reference to the scalar, so that at the end of
15823 the array owns zero references to the scalars it happens to
15824 point to. We are effectively converting the array from
15825 AvREAL() on to AvREAL() off. This saves the av_clear()
15826 (triggered by the SvREFCNT_dec(unreferenced) below) from
15827 walking the array a second time. */
15828 SvREFCNT_dec(*svp);
15831 } while (++svp <= last);
15832 AvREAL_off(unreferenced);
15834 SvREFCNT_dec_NN(unreferenced);
15838 Perl_clone_params_del(CLONE_PARAMS *param)
15840 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15842 PerlInterpreter *const to = param->new_perl;
15844 PerlInterpreter *const was = PERL_GET_THX;
15846 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15852 SvREFCNT_dec(param->stashes);
15853 if (param->unreferenced)
15854 unreferenced_to_tmp_stack(param->unreferenced);
15864 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15867 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15868 does a dTHX; to get the context from thread local storage.
15869 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15870 a version that passes in my_perl. */
15871 PerlInterpreter *const was = PERL_GET_THX;
15872 CLONE_PARAMS *param;
15874 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15880 /* Given that we've set the context, we can do this unshared. */
15881 Newx(param, 1, CLONE_PARAMS);
15884 param->proto_perl = from;
15885 param->new_perl = to;
15886 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15887 AvREAL_off(param->stashes);
15888 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15896 #endif /* USE_ITHREADS */
15899 Perl_init_constants(pTHX)
15901 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15902 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15903 SvANY(&PL_sv_undef) = NULL;
15905 SvANY(&PL_sv_no) = new_XPVNV();
15906 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15907 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15908 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15911 SvANY(&PL_sv_yes) = new_XPVNV();
15912 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15913 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15914 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15917 SvPV_set(&PL_sv_no, (char*)PL_No);
15918 SvCUR_set(&PL_sv_no, 0);
15919 SvLEN_set(&PL_sv_no, 0);
15920 SvIV_set(&PL_sv_no, 0);
15921 SvNV_set(&PL_sv_no, 0);
15923 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15924 SvCUR_set(&PL_sv_yes, 1);
15925 SvLEN_set(&PL_sv_yes, 0);
15926 SvIV_set(&PL_sv_yes, 1);
15927 SvNV_set(&PL_sv_yes, 1);
15929 PadnamePV(&PL_padname_const) = (char *)PL_No;
15933 =head1 Unicode Support
15935 =for apidoc sv_recode_to_utf8
15937 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15938 of C<sv> is assumed to be octets in that encoding, and C<sv>
15939 will be converted into Unicode (and UTF-8).
15941 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15942 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15943 an C<Encode::XS> Encoding object, bad things will happen.
15944 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15946 The PV of C<sv> is returned.
15951 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15953 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15955 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15964 if (SvPADTMP(nsv)) {
15965 nsv = sv_newmortal();
15966 SvSetSV_nosteal(nsv, sv);
15975 Passing sv_yes is wrong - it needs to be or'ed set of constants
15976 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15977 remove converted chars from source.
15979 Both will default the value - let them.
15981 XPUSHs(&PL_sv_yes);
15984 call_method("decode", G_SCALAR);
15988 s = SvPV_const(uni, len);
15989 if (s != SvPVX_const(sv)) {
15990 SvGROW(sv, len + 1);
15991 Move(s, SvPVX(sv), len + 1, char);
15992 SvCUR_set(sv, len);
15997 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15998 /* clear pos and any utf8 cache */
15999 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16002 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16003 magic_setutf8(sv,mg); /* clear UTF8 cache */
16008 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16012 =for apidoc sv_cat_decode
16014 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16015 assumed to be octets in that encoding and decoding the input starts
16016 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16017 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16018 when the string C<tstr> appears in decoding output or the input ends on
16019 the PV of C<ssv>. The value which C<offset> points will be modified
16020 to the last input position on C<ssv>.
16022 Returns TRUE if the terminator was found, else returns FALSE.
16027 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16028 SV *ssv, int *offset, char *tstr, int tlen)
16032 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16034 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16045 offsv = newSViv(*offset);
16047 mPUSHp(tstr, tlen);
16049 call_method("cat_decode", G_SCALAR);
16051 ret = SvTRUE(TOPs);
16052 *offset = SvIV(offsv);
16058 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16063 /* ---------------------------------------------------------------------
16065 * support functions for report_uninit()
16068 /* the maxiumum size of array or hash where we will scan looking
16069 * for the undefined element that triggered the warning */
16071 #define FUV_MAX_SEARCH_SIZE 1000
16073 /* Look for an entry in the hash whose value has the same SV as val;
16074 * If so, return a mortal copy of the key. */
16077 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16083 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16085 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16086 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16089 array = HvARRAY(hv);
16091 for (i=HvMAX(hv); i>=0; i--) {
16093 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16094 if (HeVAL(entry) != val)
16096 if ( HeVAL(entry) == &PL_sv_undef ||
16097 HeVAL(entry) == &PL_sv_placeholder)
16101 if (HeKLEN(entry) == HEf_SVKEY)
16102 return sv_mortalcopy(HeKEY_sv(entry));
16103 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16109 /* Look for an entry in the array whose value has the same SV as val;
16110 * If so, return the index, otherwise return -1. */
16113 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16115 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16117 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16118 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16121 if (val != &PL_sv_undef) {
16122 SV ** const svp = AvARRAY(av);
16125 for (i=AvFILLp(av); i>=0; i--)
16132 /* varname(): return the name of a variable, optionally with a subscript.
16133 * If gv is non-zero, use the name of that global, along with gvtype (one
16134 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16135 * targ. Depending on the value of the subscript_type flag, return:
16138 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16139 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16140 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16141 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16144 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16145 const SV *const keyname, SSize_t aindex, int subscript_type)
16148 SV * const name = sv_newmortal();
16149 if (gv && isGV(gv)) {
16151 buffer[0] = gvtype;
16154 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16156 gv_fullname4(name, gv, buffer, 0);
16158 if ((unsigned int)SvPVX(name)[1] <= 26) {
16160 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16162 /* Swap the 1 unprintable control character for the 2 byte pretty
16163 version - ie substr($name, 1, 1) = $buffer; */
16164 sv_insert(name, 1, 1, buffer, 2);
16168 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16171 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16173 if (!cv || !CvPADLIST(cv))
16175 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16176 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16180 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16181 SV * const sv = newSV(0);
16183 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16185 *SvPVX(name) = '$';
16186 Perl_sv_catpvf(aTHX_ name, "{%s}",
16187 pv_pretty(sv, pv, len, 32, NULL, NULL,
16188 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16189 SvREFCNT_dec_NN(sv);
16191 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16192 *SvPVX(name) = '$';
16193 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16195 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16196 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16197 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16205 =for apidoc find_uninit_var
16207 Find the name of the undefined variable (if any) that caused the operator
16208 to issue a "Use of uninitialized value" warning.
16209 If match is true, only return a name if its value matches C<uninit_sv>.
16210 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16211 warning, then following the direct child of the op may yield an
16212 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16213 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16214 the variable name if we get an exact match.
16215 C<desc_p> points to a string pointer holding the description of the op.
16216 This may be updated if needed.
16218 The name is returned as a mortal SV.
16220 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16221 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16227 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16228 bool match, const char **desc_p)
16233 const OP *o, *o2, *kid;
16235 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16237 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16238 uninit_sv == &PL_sv_placeholder)))
16241 switch (obase->op_type) {
16244 /* undef should care if its args are undef - any warnings
16245 * will be from tied/magic vars */
16253 const bool pad = ( obase->op_type == OP_PADAV
16254 || obase->op_type == OP_PADHV
16255 || obase->op_type == OP_PADRANGE
16258 const bool hash = ( obase->op_type == OP_PADHV
16259 || obase->op_type == OP_RV2HV
16260 || (obase->op_type == OP_PADRANGE
16261 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16265 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16267 if (pad) { /* @lex, %lex */
16268 sv = PAD_SVl(obase->op_targ);
16272 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16273 /* @global, %global */
16274 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16277 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16279 else if (obase == PL_op) /* @{expr}, %{expr} */
16280 return find_uninit_var(cUNOPx(obase)->op_first,
16281 uninit_sv, match, desc_p);
16282 else /* @{expr}, %{expr} as a sub-expression */
16286 /* attempt to find a match within the aggregate */
16288 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16290 subscript_type = FUV_SUBSCRIPT_HASH;
16293 index = find_array_subscript((const AV *)sv, uninit_sv);
16295 subscript_type = FUV_SUBSCRIPT_ARRAY;
16298 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16301 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16302 keysv, index, subscript_type);
16306 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16308 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16309 if (!gv || !GvSTASH(gv))
16311 if (match && (GvSV(gv) != uninit_sv))
16313 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16316 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16319 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16321 return varname(NULL, '$', obase->op_targ,
16322 NULL, 0, FUV_SUBSCRIPT_NONE);
16325 gv = cGVOPx_gv(obase);
16326 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16328 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16330 case OP_AELEMFAST_LEX:
16333 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16334 if (!av || SvRMAGICAL(av))
16336 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16337 if (!svp || *svp != uninit_sv)
16340 return varname(NULL, '$', obase->op_targ,
16341 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16344 gv = cGVOPx_gv(obase);
16349 AV *const av = GvAV(gv);
16350 if (!av || SvRMAGICAL(av))
16352 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16353 if (!svp || *svp != uninit_sv)
16356 return varname(gv, '$', 0,
16357 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16359 NOT_REACHED; /* NOTREACHED */
16362 o = cUNOPx(obase)->op_first;
16363 if (!o || o->op_type != OP_NULL ||
16364 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16366 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16371 bool negate = FALSE;
16373 if (PL_op == obase)
16374 /* $a[uninit_expr] or $h{uninit_expr} */
16375 return find_uninit_var(cBINOPx(obase)->op_last,
16376 uninit_sv, match, desc_p);
16379 o = cBINOPx(obase)->op_first;
16380 kid = cBINOPx(obase)->op_last;
16382 /* get the av or hv, and optionally the gv */
16384 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16385 sv = PAD_SV(o->op_targ);
16387 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16388 && cUNOPo->op_first->op_type == OP_GV)
16390 gv = cGVOPx_gv(cUNOPo->op_first);
16394 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16399 if (kid && kid->op_type == OP_NEGATE) {
16401 kid = cUNOPx(kid)->op_first;
16404 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16405 /* index is constant */
16408 kidsv = newSVpvs_flags("-", SVs_TEMP);
16409 sv_catsv(kidsv, cSVOPx_sv(kid));
16412 kidsv = cSVOPx_sv(kid);
16416 if (obase->op_type == OP_HELEM) {
16417 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16418 if (!he || HeVAL(he) != uninit_sv)
16422 SV * const opsv = cSVOPx_sv(kid);
16423 const IV opsviv = SvIV(opsv);
16424 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16425 negate ? - opsviv : opsviv,
16427 if (!svp || *svp != uninit_sv)
16431 if (obase->op_type == OP_HELEM)
16432 return varname(gv, '%', o->op_targ,
16433 kidsv, 0, FUV_SUBSCRIPT_HASH);
16435 return varname(gv, '@', o->op_targ, NULL,
16436 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16437 FUV_SUBSCRIPT_ARRAY);
16440 /* index is an expression;
16441 * attempt to find a match within the aggregate */
16442 if (obase->op_type == OP_HELEM) {
16443 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16445 return varname(gv, '%', o->op_targ,
16446 keysv, 0, FUV_SUBSCRIPT_HASH);
16449 const SSize_t index
16450 = find_array_subscript((const AV *)sv, uninit_sv);
16452 return varname(gv, '@', o->op_targ,
16453 NULL, index, FUV_SUBSCRIPT_ARRAY);
16458 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16460 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16462 NOT_REACHED; /* NOTREACHED */
16465 case OP_MULTIDEREF: {
16466 /* If we were executing OP_MULTIDEREF when the undef warning
16467 * triggered, then it must be one of the index values within
16468 * that triggered it. If not, then the only possibility is that
16469 * the value retrieved by the last aggregate index might be the
16470 * culprit. For the former, we set PL_multideref_pc each time before
16471 * using an index, so work though the item list until we reach
16472 * that point. For the latter, just work through the entire item
16473 * list; the last aggregate retrieved will be the candidate.
16474 * There is a third rare possibility: something triggered
16475 * magic while fetching an array/hash element. Just display
16476 * nothing in this case.
16479 /* the named aggregate, if any */
16480 PADOFFSET agg_targ = 0;
16482 /* the last-seen index */
16484 PADOFFSET index_targ;
16486 IV index_const_iv = 0; /* init for spurious compiler warn */
16487 SV *index_const_sv;
16488 int depth = 0; /* how many array/hash lookups we've done */
16490 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16491 UNOP_AUX_item *last = NULL;
16492 UV actions = items->uv;
16495 if (PL_op == obase) {
16496 last = PL_multideref_pc;
16497 assert(last >= items && last <= items + items[-1].uv);
16504 switch (actions & MDEREF_ACTION_MASK) {
16506 case MDEREF_reload:
16507 actions = (++items)->uv;
16510 case MDEREF_HV_padhv_helem: /* $lex{...} */
16513 case MDEREF_AV_padav_aelem: /* $lex[...] */
16514 agg_targ = (++items)->pad_offset;
16518 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16521 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16523 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16524 assert(isGV_with_GP(agg_gv));
16527 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16528 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16531 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16532 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16538 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16539 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16542 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16543 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16550 index_const_sv = NULL;
16552 index_type = (actions & MDEREF_INDEX_MASK);
16553 switch (index_type) {
16554 case MDEREF_INDEX_none:
16556 case MDEREF_INDEX_const:
16558 index_const_sv = UNOP_AUX_item_sv(++items)
16560 index_const_iv = (++items)->iv;
16562 case MDEREF_INDEX_padsv:
16563 index_targ = (++items)->pad_offset;
16565 case MDEREF_INDEX_gvsv:
16566 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16567 assert(isGV_with_GP(index_gv));
16571 if (index_type != MDEREF_INDEX_none)
16574 if ( index_type == MDEREF_INDEX_none
16575 || (actions & MDEREF_FLAG_last)
16576 || (last && items >= last)
16580 actions >>= MDEREF_SHIFT;
16583 if (PL_op == obase) {
16584 /* most likely index was undef */
16586 *desc_p = ( (actions & MDEREF_FLAG_last)
16587 && (obase->op_private
16588 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16590 (obase->op_private & OPpMULTIDEREF_EXISTS)
16593 : is_hv ? "hash element" : "array element";
16594 assert(index_type != MDEREF_INDEX_none);
16596 if (GvSV(index_gv) == uninit_sv)
16597 return varname(index_gv, '$', 0, NULL, 0,
16598 FUV_SUBSCRIPT_NONE);
16603 if (PL_curpad[index_targ] == uninit_sv)
16604 return varname(NULL, '$', index_targ,
16605 NULL, 0, FUV_SUBSCRIPT_NONE);
16609 /* If we got to this point it was undef on a const subscript,
16610 * so magic probably involved, e.g. $ISA[0]. Give up. */
16614 /* the SV returned by pp_multideref() was undef, if anything was */
16620 sv = PAD_SV(agg_targ);
16622 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16626 if (index_type == MDEREF_INDEX_const) {
16631 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16632 if (!he || HeVAL(he) != uninit_sv)
16636 SV * const * const svp =
16637 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16638 if (!svp || *svp != uninit_sv)
16643 ? varname(agg_gv, '%', agg_targ,
16644 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16645 : varname(agg_gv, '@', agg_targ,
16646 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16649 /* index is an var */
16651 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16653 return varname(agg_gv, '%', agg_targ,
16654 keysv, 0, FUV_SUBSCRIPT_HASH);
16657 const SSize_t index
16658 = find_array_subscript((const AV *)sv, uninit_sv);
16660 return varname(agg_gv, '@', agg_targ,
16661 NULL, index, FUV_SUBSCRIPT_ARRAY);
16665 return varname(agg_gv,
16667 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16669 NOT_REACHED; /* NOTREACHED */
16673 /* only examine RHS */
16674 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16678 o = cUNOPx(obase)->op_first;
16679 if ( o->op_type == OP_PUSHMARK
16680 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16684 if (!OpHAS_SIBLING(o)) {
16685 /* one-arg version of open is highly magical */
16687 if (o->op_type == OP_GV) { /* open FOO; */
16689 if (match && GvSV(gv) != uninit_sv)
16691 return varname(gv, '$', 0,
16692 NULL, 0, FUV_SUBSCRIPT_NONE);
16694 /* other possibilities not handled are:
16695 * open $x; or open my $x; should return '${*$x}'
16696 * open expr; should return '$'.expr ideally
16703 /* ops where $_ may be an implicit arg */
16708 if ( !(obase->op_flags & OPf_STACKED)) {
16709 if (uninit_sv == DEFSV)
16710 return newSVpvs_flags("$_", SVs_TEMP);
16711 else if (obase->op_targ
16712 && uninit_sv == PAD_SVl(obase->op_targ))
16713 return varname(NULL, '$', obase->op_targ, NULL, 0,
16714 FUV_SUBSCRIPT_NONE);
16721 match = 1; /* print etc can return undef on defined args */
16722 /* skip filehandle as it can't produce 'undef' warning */
16723 o = cUNOPx(obase)->op_first;
16724 if ((obase->op_flags & OPf_STACKED)
16726 ( o->op_type == OP_PUSHMARK
16727 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16728 o = OpSIBLING(OpSIBLING(o));
16732 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16733 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16735 /* the following ops are capable of returning PL_sv_undef even for
16736 * defined arg(s) */
16755 case OP_GETPEERNAME:
16802 case OP_SMARTMATCH:
16811 /* XXX tmp hack: these two may call an XS sub, and currently
16812 XS subs don't have a SUB entry on the context stack, so CV and
16813 pad determination goes wrong, and BAD things happen. So, just
16814 don't try to determine the value under those circumstances.
16815 Need a better fix at dome point. DAPM 11/2007 */
16821 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16822 if (gv && GvSV(gv) == uninit_sv)
16823 return newSVpvs_flags("$.", SVs_TEMP);
16828 /* def-ness of rval pos() is independent of the def-ness of its arg */
16829 if ( !(obase->op_flags & OPf_MOD))
16834 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16835 return newSVpvs_flags("${$/}", SVs_TEMP);
16840 if (!(obase->op_flags & OPf_KIDS))
16842 o = cUNOPx(obase)->op_first;
16848 /* This loop checks all the kid ops, skipping any that cannot pos-
16849 * sibly be responsible for the uninitialized value; i.e., defined
16850 * constants and ops that return nothing. If there is only one op
16851 * left that is not skipped, then we *know* it is responsible for
16852 * the uninitialized value. If there is more than one op left, we
16853 * have to look for an exact match in the while() loop below.
16854 * Note that we skip padrange, because the individual pad ops that
16855 * it replaced are still in the tree, so we work on them instead.
16858 for (kid=o; kid; kid = OpSIBLING(kid)) {
16859 const OPCODE type = kid->op_type;
16860 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16861 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16862 || (type == OP_PUSHMARK)
16863 || (type == OP_PADRANGE)
16867 if (o2) { /* more than one found */
16874 return find_uninit_var(o2, uninit_sv, match, desc_p);
16876 /* scan all args */
16878 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16890 =for apidoc report_uninit
16892 Print appropriate "Use of uninitialized variable" warning.
16898 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16900 const char *desc = NULL;
16901 SV* varname = NULL;
16904 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16907 if (uninit_sv && PL_curpad) {
16908 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16910 sv_insert(varname, 0, 0, " ", 1);
16913 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16914 /* we've reached the end of a sort block or sub,
16915 * and the uninit value is probably what that code returned */
16918 /* PL_warn_uninit_sv is constant */
16919 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16921 /* diag_listed_as: Use of uninitialized value%s */
16922 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16923 SVfARG(varname ? varname : &PL_sv_no),
16926 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16932 * ex: set ts=8 sts=4 sw=4 et: